Data analysis strategies for proteome-wide
crosslinking mass spectrometry
vorgelegt von
M. Sc.
Swantje Lenz
ORCID: 0000-0002-8839-5371
an der Fakultät III - Prozesswissenschaften
der Technischen Universität Berlin
zur Erlangung des akademischen Grades
Doktorin der Naturwissenschaften
-Dr. rer. nat. -
genehmigte Dissertation
Promotionsausschuss:
Vorsitzende: Sina Bartfeld
Gutachter: Bernhard Renard
Gutachter: Juri Rappsilber
Tag der wissenschaftlichen Aussprache: 24. Mai 2022
Berlin 2023
3
Declaration of Authorship
I, Swantje Lenz, declare that this thesis titled, “Data analysis foundation of proteome-wide
crosslinking mass spectrometry” and the work presented in it are my own.
I confirm that:
•This work was done wholly or mainly while in candidature for a research degree at
this University.
•Where any part of this thesis has previously been submitted for a degree or any other
qualification at this University or any other institution, this has been clearly stated.
•Where I have consulted the published work of others, this is always clearly attributed.
•Where I have quoted from the work of others, the source is always given. Apart from
such quotations, this thesis is entirely my own work.
•I have acknowledged all main sources of help.
•Where the thesis is based on work done by myself jointly with others, I have made
clear exactly what was done by others and what I have contributed myself.
Signature
Date
4
Table of contents
Abstract ..................................................................................................................................... 5
Zusammenfassung ................................................................................................................... 6
Introduction .............................................................................................................................. 7
Crosslinking MS as a technology to investigate protein-protein interactions ........... 7
The crosslinking MS workflow ................................................................................. 8
Crosslinked peptide identification ........................................................................... 11
Error estimation in crosslinking MS ........................................................................ 13
Contributions and Main Findings ........................................................................................ 17
Manuscript 1: In-Search Assignment of Monoisotopic Peaks Improves the Identification
of Cross-Linked Peptides. ..................................................................................................... 19
Manuscript 2: Improved peptide backbone fragmentation is the primary advantage of MS-
cleavable crosslinkers. ........................................................................................................... 31
Manuscript 3: Reliable identification of protein-protein interactions by crosslinking mass
spectrometry. ......................................................................................................................... 39
Outlook ................................................................................................................................... 53
Acknowledgements ................................................................................................................ 55
References ............................................................................................................................... 56
Supplement ............................................................................................................................. 60
5
Abstract
As enzymes, building blocks or messengers, proteins are essential molecules in life. They often
function together with other proteins in complexes or in transient interactions. A crucial part
of understanding their function is knowing their structure and their interaction partners.
Crosslinking mass spectrometry (crosslinking MS) has by now been established as a
method to study protein interactions and structures by delivering medium-resolution inter-
residue distances. Initially, crosslinking MS studies were limited to single proteins or
complexes, but the technology has the potential to be used in more complex samples, with the
goal to detect protein-protein interactions at a proteome-wide scale. To realise this, the
technology requires development and optimization of multiple steps of the workflow.
In this thesis, I focus on the data analysis of crosslinking MS data at different points of
the workflow. The work of this thesis increased the number of identifications during database
search and provides the groundwork for further optimisation of the crosslinking MS workflow.
It demonstrates data-driven evaluation of experimental tests and provides a reliable procedure
for error estimation.
First, I show that for crosslinked peptides, due to their low abundance and large size,
the monoisotopic precursor mass is often misassigned by the mass spectrometer software. We
implemented a solution into our database search, where multiple masses are searched. This
increased the number of crosslinked identifications significantly.
Another important factor in MS acquisition is the fragmentation of crosslinked
peptides. I therefore analysed the fragmentation behaviour of the MS-cleavable crosslinker
DSSO, which is commonly used for large-scale crosslinking MS studies. We analyse
commonly used workflows regarding the peptide fragmentation and utilisation of the
characteristic peaks during database search. This showed that the advantage of MS-cleavable
crosslinkers lies in the improved fragmentation and showed that some workflows are
suboptimal in their speed.
Finally, we use a controlled sample of E. coli lysate to demonstrate a reliable procedure
to estimate the error of crosslinked PPIs. The study was set up to allow for an experimental
control of the error. With this and three other controls we show that for a reliable error
estimation in crosslinked PPIs, the FDR needs to be calculated separately for self and
heteromeric matches and on the PPI-level. This error estimation was applied to our E. coli
lysate and provided a reliable network of protein-protein interaction. Here, we found an
unknown binder to RNA polymerase which we map to its binding site with use of the structural
information of the crosslinks.
Overall, the results of this work allowed us to use crosslinking MS on the scale of
proteome-wide, in-cell studies. The next challenge will be increasing the depth to allow
detection of low abundant proteins, which will require further optimisation of crosslinking MS.
6
Zusammenfassung
In Form von Enzymen, Bausteinen oder Botenstoffen agieren Proteine als wesentliche
Moleküle des Lebens. Gemeinsam mit anderen Proteinen wirken sie oft als Protein-Komplexe
oder in kurzlebigen Interaktionen. Dabei ist die Kenntnis ihrer Struktur und Interaktionspartner
für das Verständnis ihrer Funktion entscheidend.
Als Methode zur Untersuchung von Proteininteraktionen und -strukturen hat sich die
Crosslinking-Massenspektrometrie (Crosslinking MS) etabliert, da sie Distanzen zwischen
Aminosäureresten in mittlerer Auflösung liefert. Ursprünglich waren Crosslinking MS Studien
auf einzelne Proteine oder Komplexe beschränkt, aber die Technologie hat das Potenzial, in
komplexeren Proben eingesetzt zu werden und Protein-Protein-Interaktionen auf einem
proteomweiten Maßstab zu erkennen. Um dieses Potenzial auszuschöpfen, muss die
Technologie in mehreren Schritten des Arbeitsablaufs weiterentwickelt und optimiert werden.
In dieser Arbeit konzentriere ich mich auf die Datenanalyse von Crosslinking-MS-
Daten an verschiedenen Stellen des Arbeitsablaufs. Die Arbeit hat die Zahl der
Identifizierungen der Datenbanksuche erhöht und liefert die Grundlage für die weitere
Optimierung des Arbeitsablaufs bei der Crosslinking MS. Sie demonstriert die datengesteuerte
Auswertung experimenteller Tests und liefert ein zuverlässiges Verfahren zur
Fehlerabschätzung.
Zunächst zeige ich, dass bei vernetzten Peptiden aufgrund ihrer Größe und geringen
Abundanz die monoisotopische Vorläufermasse von der Software des Massenspektrometers
oft falsch zugeordnet wird. Wir haben eine Lösung in unsere Datenbanksuche implementiert,
bei der nach mehreren Massen gesucht wird. Dadurch kann die Zahl der vernetzten
Identifizierungen erheblich gesteigert werden.
Ein weiterer wichtiger Faktor bei der MS Akquisition ist die Fragmentierung von
vernetzten Peptiden. Daher haben wir das Fragmentierungsverhalten des MS-spaltbaren
Crosslinkers DSSO analysiert, der üblicherweise für groß angelegte Crosslinking MS Studien
verwendet wird. Wir analysierten gängige Arbeitsabläufe hinsichtlich der
Peptidfragmentierung und der Nutzung der charakteristischen Peaks bei der Datenbanksuche.
Dabei zeigte sich, dass der Vorteil von MS-spaltbaren Crosslinkern in der verbesserten
Fragmentierung liegt und dass einige Arbeitsabläufe in ihrer Geschwindigkeit suboptimal sind.
Schließlich demonstrieren wir anhand einer kontrollierten Probe von E. coli Lysat ein
zuverlässiges Verfahren zur Abschätzung des Fehlers von vernetzten PPIs. Die Studie
ermöglicht uns eine experimentelle Kontrolle des Fehlers. Mit dieser und drei weiteren
Kontrollen zeigen wir, dass für eine zuverlässige Fehlerabschätzung für vernetzte PPIs die FDR
separat für Selbst- und Heteromere PPIs und auf PPI-Ebene berechnet werden muss. Diese
Fehlerabschätzung wird auf unser E. coli Lysat angewandt und liefert ein zuverlässiges PPI-
Netzwerk. Wir haben einen unbekannten Binder für die RNA-Polymerase gefunden, den wir
mit Hilfe der Strukturinformationen der Crosslinks auf seine Bindungsstelle abbilden.
Insgesamt ermöglichen uns die Ergebnisse dieser Arbeit, Crosslinking MS auf die
Ebene von proteomweiten, zellinternen Studien zu bringen. Die nächste Herausforderung
besteht darin, die Tiefe der Analyse zu erweitern, um auch Proteine mit geringer Abundanz
nachzuweisen, wofür weitere Optimierungen der Crosslinking MS erforderlich sind.
7
Introduction
Crosslinking MS as a technology to investigate protein-protein interactions
Proteins are one of the most important molecules in biology and are therefore a focus of
molecular studies. Among the wide range of functions they can perform, they can work as
catalysts, information messengers and structural units. They often do not function alone, but as
part of multi-protein complexes with specific structures and dynamics fine-tuned by evolution.
Identifying the members of these complexes and their three-dimensional architecture is
important to understand their function. Until recently, this had to be done by purification and
reconstitution outside of the cellular context, but recent technological developments are now
allowing for in situ studies (Böhning & Bharat, 2021).
Traditional methods to study these protein-protein interactions (PPIs) are for example
two-hybrid techniques (Parrish et al., 2006) or, using proteomics, affinity-purification MS
(Rigaut et al., 1999) or co-fractionation MS (Kristensen et al., 2012). These techniques require
genetic tagging of the protein coding genes, lysis of the cells, or purification of the proteins
and therefore take the proteins outside of their natural context prior to analysis. This can
introduce experimental biases into the results, for example losing transient or weak interactions
by the lysis or washing steps.
In the past years, crosslinking MS has emerged as an alternative tool to study PPIs
(O’Reilly & Rappsilber, 2018). Crosslinking introduces covalent bonds between amino acid
residues of the proteins. Followed up by mass spectrometric analysis, the linked peptides, and
thereby the linkage positions, can be determined. Since the crosslinker reagent has a known
length, this leads to a known distance restraint between the two residues. In contrast to the
aforementioned methods, crosslinking MS can be performed in situ and therefore can generate
information on the protein interactions inside intact cells, including transient or weak
interactions that are often lost during cell disruption.
Crosslinks between two proteins inform that they were close in space during the
crosslinking reaction and therefore are very likely to be directly interacting. Since crosslinking
delivers information on residue-level, it also provides structural information about binding sites
of these interactions. These can be used as distance restraints for further structural studies, for
example in integrative modelling in combination with electron-microscopy densities
(Robinson et al., 2015; von Appen et al., 2015). Therefore, crosslinking MS is a versatile tool
that can be used in both focused structural biology studies and also systems-wide studies to
map PPIs and their topologies.
Prior to the work of this thesis, crosslinking MS experiments tended to be performed
on single proteins or complexes as proof-of-concept studies or providing extra information to
structural studies (Z. A. Chen et al., 2010; Lasker et al., 2012). In recent years, crosslinking
MS has developed into a technology that can handle the complexity of proteome-wide protein-
protein interaction networks (Chavez et al., 2018; Liu et al., 2018). To analyse these very
complex protein mixtures, multiple adjustments and optimization steps in the workflow were
needed. The focus of my PhD was to develop data analysis approaches to allow crosslinking
MS to handle the complexity of a whole crosslinked cell. The advances that I describe here
8
have underpinned several biological discoveries and will underpin many more in the future as
this technology becomes mainstream.
The crosslinking MS workflow
During bottom-up proteomics, whole proteins are digested to peptides, which are subsequently
identified by mass spectrometry. Crosslinked peptides have certain properties that make their
identification by mass spectrometry more challenging than for linear peptides. For one, they
are always much less abundant than the background of linear peptides in digestion mixtures,
which hinders their detection in the MS. In addition to this experimental challenge, identifying
the crosslinked peptide is more complex since two peptides have to be matched successfully
from the same fragmentation spectrum. Both of these problems are intensified as the sample
complexity is growing to larger scales. During the past years, all steps of the workflow were
further optimised to meet those challenges. Briefly, the experimental workflow has five distinct
steps that have each undergone significant optimisations by many labs over the years.
Crosslinking reaction. As a first step, the crosslinker is added to the sample, for
example a purified protein complex or, for in situ experiments, to cells. The reagents consist
of two reactive groups, which react with amino acids, separated by a spacer (Belsom &
Rappsilber, 2021). Commonly NHS-ester based crosslinker reagents are used, which react
primarily with lysines or n-termini, as well as serine, threonine or tyrosine in a side reaction.
These include for example disuccinimidyl suberate (DSS) or bis(sulfosuccinimidyl)suberate
(BS3), crosslinkers with two NHS-ester groups. Another popular crosslinker succinimidyl 4,4-
azipentanoate (SDA) consists of one NHS-group and a diazirine group as the second reactive
group. The diazirine is activated by UV-light, resulting in a carbene intermediate which can
rapidly react with any amino acid. In addition, crosslinker spacers have been modified to be
cleavable in the MS, for example disuccinimidyl sulfoxide (DSSO) (Kao et al., 2011) or DSBU
(Müller et al., 2010). Other crosslinkers contain an enrichment group on the spacer
(Steigenberger et al., 2019; Tang & Bruce, 2010) to thereby allow targeted enrichment of
crosslinked peptides.
The choice of crosslinker depends on the purpose and scale of the experiment. For
example, SDA as a crosslinker resulting in a high density of crosslinks was shown to be
beneficial in structural studies (Lee et al., 2020), but results in a highly complex set of peptide
pairs that makes identification in large-scale samples challenging. Its fast reaction time makes
it especially suitable for cleanly capturing conformational changes (Belsom & Rappsilber,
2021). For large-scale studies, cleavable crosslinkers like DSSO have been widely used, as it
has been thought that cleavage in the mass spectrometers aids identification by leaving
characteristic features in the spectrum. There remained questions on the essentiality of
cleavable crosslinkers and their implementation, which we addressed with the work in this
thesis.
Crosslink enrichment. After digestion of the proteins to peptides, crosslinked peptides
are very low abundant in comparison to linear peptides. Since crosslinked peptides are larger
and higher charged than linear peptides, these properties can be used to separate them both in
the mass spectrometer, and by chromatographic methods prior to MS acquisition.
Chromatographic fractionation is used to enrich crosslinks by separating the sample based on
9
size and or charge. Methods previously used are for example size exclusion chromatography,
strong cation exchange or hydrophilic strong anion exchange (Z. A. Chen et al., 2010; Fritzsche
et al., 2012; Leitner et al., 2012; O’Reilly et al., 2020). Enrichment becomes crucial for large-
scale samples such as proteome-wide studies, as their complexity is much higher than that of
simpler complexes. In these cases, multiple fractionations performed subsequently can be
performed to improve the results (Lenz et al., 2021; O’Reilly et al., 2020).
Data Acquisition and MS data. Fractionated peptides are further separated by liquid-
chromatography, typically reversed-phase, and injected in the mass spectrometer via
electrospray ionisation. This derives m/z peak positions and their intensities and the charge of
the peptides coming off the column in an MS1 scan. In addition to this, metadata for each scan
is collected, for example the retention time, measurement settings or fill times. During
acquisition, MS1 scans are continuously recorded. They provide information about the
unfragmented peptides, which, for example, can be used for their quantification (Cox & Mann,
2008). During the MS1 scan, specific precursors are selected for further fragmentation. A
subsequent MS2 scan is the representation of the resulting fragments and is used for peptide
identification. In the most common fragmentation methods in proteomics, collision-induced
dissociation (CID) and higher energy collision dissociation (HCD), b- and y-ions are the most
commonly seen fragments.
Typically in crosslinking MS workflows, the precursor selection is dependent on the
charge state to reduce sampling of lower charged linear peptides. The fragmentation methods
have been optimised for crosslinked peptides (Kolbowski et al., 2017; Liu et al., 2017), and
most commonly use HCD or CID fragmentation. In some crosslinking workflows performed
with cleavable crosslinkers, MS3 fragmentation of the single peptides is done in addition to the
MS2 (Liu et al., 2017).
Because crosslinked peptides consist of two covalently bound peptides instead of a
single continuous peptide, their MS2 spectra are slightly different. For one, they present more
complex, chimeric spectra with fragment peaks coming from both peptides. In addition,
crosslinked spectra contain not only linear fragments (continuous chain of amino acids from a
single peptide), but also fragments that include the crosslinker and second peptide (Figure 1).
These crosslinking site containing fragments are of special interest because they give extra
confidence of the peptides being crosslinked and the position of the link site.
These steps are followed by the computational parts of the workflow, i.e. database search and
error estimation. Besides the data analysis of experimental data, these two steps are the focus
of this thesis, therefore I will describe them in detail below.
10
Figure 1. Crosslinking MS workflow. Crosslinker reagent is added to the sample (e.g. protein
complexes or cells), proteins are digested and the sample is enriched for crosslinked peptides.
After MS acquisition, MS2 scans are searched against a sequence database and the error is
estimated and a cutoff applied. Adapted from (O’Reilly & Rappsilber, 2018). The spectrum is
a representative spectrum of a BS3-crosslinked peptide with fragments containing the second
peptide marked in bold.
11
Crosslinked peptide identification
MS acquisitions generate a large amount of raw data, from which the peptides need to be
identified. In this section I describe how the spectra are searched during database search, with
a focus on crosslinking MS.
Database search concepts. To identify the peptides from the MS2 spectra, database
search algorithms can be employed. A user defined sequence database is in silico digested to
create peptides which are likely present in the sample. During search, the algorithms are
matching generated theoretical spectra of candidate peptides against the experimentally
observed spectra. The candidates are then scored to represent the quality of the match. The
scoring procedure and algorithm depend on the software used, but is typically a measure of the
agreement between theoretical and experimental spectra in terms of matching of predicted and
observed fragments and their intensities (Cox et al., 2011; Kong et al., 2017).
Fragment spectra of crosslinked peptides tend to be more complex than the spectra of
linear peptides, as they are chimeric spectra of two peptides. This makes it more difficult to
successfully identify the correct match. In addition, combinations of two peptides have to be
considered during database search, which leads to a quadratic expansion of the search space,
commonly referred to as the n-square problem. A larger search space tends to not only increase
search time, but also adds noise, and therefore increases the probability for random matches
during search. Several different search approaches exist specifically designed for crosslink
identification (Z.-L. Chen et al., 2019; Götze et al., 2019; Hoopmann et al., 2015; Liu et al.,
2017; Mendes et al., 2019). Typically, one of the peptides fragments less well and is harder to
identify (Trnka et al., 2014). The better fragmenting peptide is utilised by some approaches,
which simplify the search by first identifying one of the peptides in an open-modification-like
search (Z.-L. Chen et al., 2019; Hoopmann et al., 2015).
One workflow commonly used in large-scale studies is based on MS-cleavable
crosslinkers. Upon cleavage in the MS, they can produce signature peptide doublet peaks.
These can be used to calculate the masses of the individual peptides, which in turn can be used
to simplify the search (Götze et al., 2019). In addition, MS3 fragmentation can be triggered on
the peptide doublets, which results in a linear spectrum of a crosslinked peptide. This simplifies
the crosslinking search to that of two linear peptides (Liu et al., 2017). In my thesis work, I
performed the first systematic evaluation of these workflows, providing conclusive evidence
that the gains in identification do not stem from utilising the peptide masses, but instead from
a better fragment coverage of the peptides.
xiSEARCH algorithm. Here, I will focus on describing xiSEARCH, the search
software developed by the Rappsilber group that has been featured in a number of publications
(Mendes et al., 2019; O’Reilly et al., 2020; Ryl et al., 2020) and used throughout this thesis.
To circumvent the n-square search space problem, xiSEARCH first matches one of the
crosslinked peptides as a linear peptide with an unknown modification. First, the MS2 spectrum
is de-charged (all peaks are shifted to an m/z value as if they were singly charged) and
linearized (crosslinked fragments get shifted to the m/z value of linear complement fragment)
(Giese et al., 2016). Then, the most abundant peaks of the spectrum are used to identify
candidate peptides (alpha candidates). The theoretical mass of the second peptide is calculated
by subtracting the crosslinker and the alpha candidate mass from the experimental precursor
12
mass. This theoretical mass of the peptide has to match the experimental mass in a user defined
mass tolerance, which is dependent on the resolution of the MS1 scan. Suitable beta candidates
of a matching mass are derived for a defined number of the alpha candidates.
Both peptides are then scored together against the spectrum. xiSEARCH calculates
multiple subscores which are combined to a final match score. These subscores represent
different properties influencing the confidence of matches. For one, the number of fragments
matched to one or both of the peptides is an important factor, as is also the coverage of the
matched peptide sequence. In addition there are scores for how many and with which intensity
the peaks in the spectrum are explained by the matched peptide. MS1 and MS2 errors of the
match also influence the score. Other scores are based on the ranks during candidate selection.
Considerations for large-scale data. As the field is moving towards studies on
proteome-wide scales, searches need to be adjusted to account for the increased complexity of
the sample and the size of the database. In contrast to linear samples or even simple crosslinked
samples, the chance of random matching is highly increased. For one, this is due to the
quadratic increase of the search space itself, which makes it more likely to match another
peptide by chance. In addition, the complexity of the sample decreases the spectral quality due
to an increase of interfering species (Houel et al., 2010) and lower abundance of crosslinked
peptides.
Search parameters have to be set carefully as the gain in true matches can be outweighed
by their contribution to random matches. For example, the increase of the search space by an
excess of modifications or proteins is usually tolerated in linear or simple crosslinked searches,
but can reduce identifications in large-scale searches. An approach used in this thesis work is
to only search the most abundant proteins (Lenz et al., 2021; Linden et al., 2020; Ryl et al.,
2020) and to solely search common modifications. In addition, the tightening of error
tolerances during search can be beneficial. Simple crosslinking data have been acquired with a
high MS resolution already, and during the work of this thesis, we have increased the resolution
further, which allowed to lower the error tolerance even more.
Recently developed computational approaches are being adapted to crosslinking search
algorithms and await evaluation of their benefit in real world conditions. Retention time
prediction has been previously established for linear proteomics (Klammer et al., 2007) and
recently for crosslinking MS (Giese et al., 2021). In addition, rescoring the search results of
linear searches via machine learning approaches has been established (The et al., 2016). For
crosslinking, the search engine pLink2 has implemented a support vector machine to improve
their results after search (Z.-L. Chen et al., 2019). However, machine learning approaches have
to be performed with care to avoid overtraining on the decoy matches, which would lead to
their removal, but not that of the false positive matches. This is especially the case in datasets
with smaller numbers of identifications, which is the typical case in crosslinking MS.
13
Error estimation in crosslinking MS
Matches coming from the search engine will also contain random and therefore wrong matches,
which need to be filtered out. Since scores can be dependent on the search algorithm, search
database and search settings, a score cutoff generally can not be transferred between
experiments. Instead, the error inside the data needs to be estimated reliably. This allows the
user to apply a cutoff at the desired error, usually between 1 and 5%. The method for error
estimation most common in proteomics is a false discovery rate (FDR) estimation using a
target-decoy approach (Elias & Gygi, 2007). Here, random matches are modelled based on
known wrong (decoy) sequences which are added to the database before search. This approach
was first established in linear proteomics and has been transferred to crosslinking MS
successfully (Fischer & Rappsilber, 2017; Walzthoeni et al., 2012).
Decoy generation and random space. The target-decoy approach for FDR calculation
is based on the assumption that a random match will be matched to targets with the same
likelihood as to decoys, i.e. that matches to the decoy database model the noise in matches to
the target database. The proportion of decoys in the targets is then used to calculate the error
rate depending on the score cutoff.
Decoy sequences are generated from the target database defined by the user. Multiple
approaches have been proposed to create the decoys (Elias & Gygi, 2007; Wang et al., 2009),
and typically the protein sequences are reversed or shuffled. The number of targets and decoys
in the database should be equal and decoys should be similar to the targets in size and amino
acid composition, however not contain the same peptides as in the targets. Target and decoy
peptides need to be treated the same and be matched to spectra under the same scoring function,
while competing for an identification.
FDR estimation in crosslinking MS is slightly different to linear approaches due to the
combination of two peptides. Here, not only the simple case of correct or random needs to be
considered, but also a match consisting of one correct and one random peptide. Therefore, not
only target and decoy matches exist, but also their combinations, i.e. target-target, target-decoy
(which is the same as decoy-target for a non-directional crosslinker) and decoy-decoy. For the
typical case of a non-directional crosslinker, the random space for target-decoy matches is
approximately double the size of target-target and decoy-decoy space (i.e. 1:2:1). A formula
taking this difference in random spaces into account was previously derived (Fischer &
Rappsilber, 2017). Apart from the FDR formula, two other considerations need to be taken into
account for crosslinking MS, about which I will go into detail below.
Self and heteromeric matches. The two peptides of the crosslink can either stem from
the same (self or intra crosslink) or from two different proteins (heteromeric or inter crosslink).
While the first case also includes homodimers, the latter is the focus of most interaction studies.
The chance of random matching is inherently different for both types of matches: Assuming
protein A and protein B in the search, the crosslink A1-A2 is the same as A2-A1. In contrast,
A1-B2 is different from B1-A2. This makes the number of possible random combinations
different for self and heteromeric matches.
The estimated FDR is only valid on the total set of CSMs that was used for calculation.
If the data are further subsetted after FDR estimation to heteromeric matches only (e.g. for
reporting PPIs) this subset will be enriched in false positive identifications. In these
14
heteromeric matches, the error will then be higher than what was initially estimated. Therefore,
FDR needs to be calculated separately for self and heteromeric matches. This splitting of the
two data types has been discussed early on (Fischer & Rappsilber, 2017; Walzthoeni et al.,
2012), but is not generally implemented in the field. In the work of this thesis, we demonstrate
the relevance of the FDR separation for the derivation of crosslinked PPIs.
A common misconception about the splitting of the FDR estimation is that this is done
to increase the number of identifications. While self matches indeed tend to increase in number
as random heteromeric matches are removed, typically the number of heteromeric matches
decreases as the error is estimated correctly. A separate FDR calculation is therefore not a way
of increasing numbers of heteromeric matches, but ensures a reliable error estimation for them.
Result levels. Similar to linear proteomics, crosslinking data can be evaluated on
different result levels (Fischer & Rappsilber, 2017) (Figure 2A). The initial match of peptide
to spectrum is referred to as peptide spectrum match, or in the case of crosslinking MS,
crosslinked spectrum match (CSM). These can exist in multiple copies if the same precursor
was triggered for fragmentation multiple times, so often only the highest scoring unique CSM
is considered, so as to not bias the FDR estimation. CSMs of different charge states are
aggregated into peptide pairs. Multiple different peptide sequences can represent the same
residues being crosslinked, for example due to missed cleavages or modifications. Therefore
peptide pairs are further aggregated into residue pairs, only including the position of the
crosslink on the protein(s). Finally, the residue pairs can merge further into PPIs, which are the
main focus of proteome-wide interaction studies. Merging to higher levels is not commonly
done, and so comparisons on the approaches of merging results to higher levels are rare and
only existing for linear proteomics (Audain et al., 2017). xiFDR calculates the new score as
𝑆𝑐𝑜𝑟𝑒!"#!$% =
(∑
𝑆𝑐𝑜𝑟𝑒&'($% )
and estimates the FDR based on that score. This has the
implicit assumption of modelling error propagation from multiple matches as independent
events.
For increasing result levels, true matches will corroborate and aggregate with each
other. Random matches however will do so less often. Therefore, the error grows with
increasing result levels, making it crucial that the FDR is calculated and the cutoff applied on
the level of interest. The result level of choice depends on the purpose of the study. For analysis
of search parameters or performance, it is usually sufficient and recommended to analyse
(unique) CSMs, as these are the results directly stemming from the search. Structural studies
of single proteins or protein complexes are usually interested in residue pairs as distance
restraints, therefore the FDR should be estimated on residue pair level. For large-scale
interaction studies the FDR needs to be estimated on PPI-level, as shown by work included in
this thesis.
Pre-filtering of data. A common practice of increasing the confidence of results is the
application of (pre-)filters to the data. With increasing confidence, the number of matches
passing the FDR cutoff will also grow. Pre-filters are applied to target matches as well as decoy
matches. A common filtering parameter is the delta score, as this describes the score difference
to the next match and can therefore be related to the confidence of the match. To increase
numbers of matches on higher result levels, it has been shown that a prefilter on the lower level
results can be beneficial (Fischer & Rappsilber, 2017). Our in-house software xiFDR
15
implements a multistep grid search (“boosting”) that is testing different cutoffs numerically to
find the best pre-filter settings.
Of note, some studies employ a filter on the data after FDR estimation (Bartolec et al.,
2020; Yugandhar, Wang, Leung, et al., 2020), since they notice an unexpected amount of noise
in their results. A properly working FDR estimation should filter out noise in the data and result
only in a reduction in number of matches. Instead of a sign for noisy data, the observations in
these studies are more likely to stem from an unreliable FDR estimation. For a sensible
statement about the error in the data, any pre-filters should be applied first with FDR estimated
after that.
Validation. In the beginning, crosslinking results were commonly validated on
structures. Based on the known distance the crosslinker can cover, a maximum plausible
distance is calculated. The corresponding distances of the resulting residue pairs are calculated
on solved structures and the percentage of violation compared to the maximum crosslinker
distance is used as an estimate for false positive results. A downside of this approach is that
flexibility of the proteins and conformational changes are ignored.
In addition, while this might have been a sensible approach for single proteins or
complexes, on a larger scale this leads to an underestimation of the error (Yugandhar, Wang,
Wierbowski, et al., 2020). The selection of certain complexes only allows the validation of a
subset of the PPI network (Figure 2B) that in addition has already been validated by other
methods (i.e. solving the structure experimentally). Therefore there is a bias in pre-selection
for true matches. In addition, this can lead to a selection of complexes that are also highly
abundant in the cell, leading to high abundant and therefore high quality links.
Another validation approach targeted at search softwares was proposed based on a
synthetic library of crosslinked peptides (Beveridge et al., 2020). Based on different poolings
of the synthetic peptides, certain peptide combinations were known to be wrong and could be
used to validate the FDR. However, the sample size was relatively small to simulate proteome-
wide studies and fails to simulate the complexity and noise of large-scale spectra. In the work
of this thesis we have established a similar control based on experimentally impossible
crosslinks, however on a proteome-wide scale.
A more suitable and universal control of search softwares and their FDR is the
employment of an entrapment database, as demonstrated by work in this thesis. Here, in
addition to the sequences expected in the sample, sequences known to be wrong are added to
the database. It is important that the entrapment space is large enough to be matched at random
sufficiently. The principle is highly similar to the target-decoy approach, however here the
identification process is truly blind to the entrapment proteins. These matches can be used as a
decoy independent control of the search results and FDR, and is particularly useful to check
for wrong decoy generation or overtraining during machine-learning.
The work of this thesis includes improvements on the identification of crosslinks and
demonstrates procedures to reliably estimate the error on large-scale data. Together, these
advances have allowed for crosslinking MS to become a method suitable for proteome-wide
analyses.
16
Figure 2. Error estimation and validation for crosslinked PPIs. A) Aggregation of matches
for different result levels. B) Validation of PPI networks by mapping crosslinks to structures is
unsuitable. It only considers a small subset of the network, while there is no information on the
crosslinks not contained in the protein structure. Protein complex image adapted from
(Yugandhar, Wang, Wierbowski, et al., 2020).
17
Contributions and Main Findings
The main focus of my work was data analysis of crosslink identifications at different points in
the workflow. The overall goal was to increase the depth of identifications and their reliability
to bring crosslinking MS from analysing complexes to proteome-wide analyses. In this
cumulative thesis, I combine the manuscripts that are the result of this work. Manuscript 1 and
3 are accepted in a peer-reviewed journal. Manuscript 2 is included here as the preprint which
was submitted to Analytical Chemistry, but is now accepted.
Manuscript 1 describes a problem arising specifically for crosslinked peptides during
data acquisition, for which we implemented a solution into the database search (Lenz et al.,
2018). The second manuscript is an evaluation of the cleavable crosslinker DSSO in its
fragmentation behaviour as well as its effect on the database search. As cleavable crosslinkers
are largely used and proposed to be beneficial for large-scale analyses, a better understanding
is necessary to drive further optimisation. As more proteome-wide studies with a focus on PPIs
are being performed, the error of this data needs to be evaluated. In manuscript 3, we therefore
demonstrate a reliable procedure to estimate the error for crosslinked PPIs (Lenz et al., 2021).
The first manuscript “In-Search Assignment of Monoisotopic Peaks Improves the
Identification of Cross-Linked Peptides.” is a study into the monoisotopic peak detection
and correction for crosslinked peptides. Due to the low abundance and large size of crosslinked
peptides, the monoisotopic peak of the precursor isotope cluster is often small and therefore
wrongly assigned by the vendor software. Other correction software, which detects precursor
features on MS1 level, can correct the precursor mass only partially. We implement a solution
into xiSEARCH by enabling it to search multiple monoisotopic masses. This led to an increase
in the number of crosslinked identifications at constant FDR.
For this project, I performed the database searches, analysed the data and created the
figures. Lutz Fischer implemented the algorithm into xiSEARCH. The manuscript was written
by me and my co-authors.
The second manuscript “Improved peptide backbone fragmentation is the primary
advantage of MS-cleavable crosslinkers.” is a study into the fragmentation behaviour of the
MS-cleavable crosslinker DSSO. Many of the proteome-wide crosslinking studies have used
MS-cleavable crosslinkers and it was suggested that successful crosslinking of complex
samples depends on them. We therefore used previously published DSSO datasets to derive
statistical evidence of the suggested advantages. We found that almost all CSMs had the
peptide doublet signature peaks, from which peptide masses could be calculated. However,
knowing the peptide masses turned out not to be the main advantage, but instead an improved
backbone fragmentation through cleavability of the crosslinker. In addition, we found that the
commonly used MS3 approaches lack sensitivity and specificity and are therefore surpassed
by steppedHCD methods. Our analysis of current methods demonstrates the importance of
thorough data analysis in experimental optimisation and will allow for further improvements
in crosslinking reagent development.
I performed the re-searches of the datasets and the analysis of their results, as well as
the analysis of peak ranks, sequence coverages and the BS3-DSSO comparison. The
18
manuscript was written and the figures created in close collaboration with the second co-author
Lars Kolbowski.
In the third manuscript “Reliable identification of protein-protein interactions by
crosslinking mass spectrometry.”, we use a controlled large-scale sample of E. coli lysate to
demonstrate the error estimation for crosslinking MS derived PPIs. While FDR for crosslinking
MS has been discussed, no consensus has emerged, especially for PPIs. Using proteome
fractionation and crosslinking in separate fractions, we are able to define non-crosslinkable
PPIs, and therefore pinpoint those that were falsely identified. We establish this as an
experimental control to validate the PPI error in our data. With this and three additional
computational controls, we demonstrate that other, previously used, FDR approaches
drastically underestimate the error on PPIs. Only FDR estimation performed separately on self
and heteromeric crosslinks, and estimated on the PPI level, can correctly estimate the error for
PPIs. Using this heteromeric PPI-FDR, we derive an E. coli lysate interaction network of 590
PPIs at 1% FDR. In this network, we found a previously unknown interactor of RNA
polymerase. With the structural information contained in the crosslinks, we map its binding
site to the DNA exit channel of RNA polymerase.
I performed database searches and FDR estimations, established and implemented the
controls of the target-decoy FDR and performed the binding site analysis of YacL. The
manuscript was written with the two other co-authors, Francis O’Reilly and Ludwig Sinn. This
work is also included in the thesis of Ludwig Sinn.
In addition I co-authored the following papers:
O’Reilly, F. J., Xue, L., Graziadei, A., Sinn, L., Lenz, S., Tegunov, D., Blötz, C.,
Singh, N., Hagen, W. J. H., Cramer, P., Stülke, J., Mahamid, J., & Rappsilber, J. (2020). In-
cell architecture of an actively transcribing-translating expressome. Science, 369(6503), 554–
557. https://doi.org/10.1126/science.abb3758
Schäpe, P., Kwon, M. J., Baumann, B., Gutschmann, B., Jung, S., Lenz, S., Nitsche,
B., Paege, N., Schütze, T., Cairns, T. C., & Meyer, V. (2019). Updating genome annotation
for the microbial cell factory Aspergillus niger using gene co-expression networks. Nucleic
Acids Research, 47(2), 559–569. https://doi.org/10.1093/nar/gky1183
Ryl, P. S. J., Bohlke-Schneider, M., Lenz, S., Fischer, L., Budzinski, L., Stuiver, M.,
Mendes, M. M. L., Sinn, L., O’Reilly, F. J., & Rappsilber, J. (2020). In Situ Structural
Restraints from Cross-Linking Mass Spectrometry in Human Mitochondria. Journal of
Proteome Research, 19(1), 327–336. https://doi.org/10.1021/acs.jproteome.9b00541
In-Search Assignment of Monoisotopic Peaks Improves the
Identification of Cross-Linked Peptides
Swantje Lenz,
†
Sven H. Giese,
†
Lutz Fischer,
‡
and Juri Rappsilber*
,†,‡
†
Bioanalytics, Institute of Biotechnology, Technische Universita t Berlin, 13355 Berlin, Germany
‡
Wellcome Centre for Cell Biology, University of Edinburgh, Edinburgh EH9 3BF, United Kingdom
*
SSupporting Information
ABSTRACT: Cross-linking/mass spectrometry has undergone
a maturation process akin to standard proteomics by adapting
key methods such as false discovery rate control and
quantification. A poorly evaluated search setting in proteomics
is the consideration of multiple (lighter) alternative values for
the monoisotopic precursor mass to compensate for possible
misassignments of the monoisotopic peak. Here, we show that
monoisotopic peak assignment is a major weakness of current
data handling approaches in cross-linking. Cross-linked
peptides often have high precursor masses, which reduces the
presence of the monoisotopic peak in the isotope envelope.
Paired with generally low peak intensity, this generates a challenge that may not be completely solvable by precursor mass
assignment routines. We therefore took an alternative route by ‘”in-search assignment of the monoisotopic peak”in the cross-
link database search tool Xi (Xi-MPA), which considers multiple precursor masses during database search. We compare and
evaluate the performance of established preprocessing workflows that partly correct the monoisotopic peak and Xi-MPA on
three publicly available data sets. Xi-MPA always delivered the highest number of identifications with ∼2 to 4-fold increase of
PSMs without compromising identification accuracy as determined by FDR estimation and comparison to crystallographic
models.
KEYWORDS: cross-linking, mass spectrometry, data processing, proteomics, software, structural proteomics, BS3, SDA, peptides,
monoisotopic mass
■INTRODUCTION
Several approaches have been utilized to increase the numbers
of identified cross-links, for example enriching for cross-linked
peptides,
1−4
using different proteases
1,5,6
or optimizing
fragmentation methods.
7,8
In parallel with experimental
developments, data analysis has also progressed to extract
even more cross-links to be used as distance restraints for
modeling of proteins and their complexes.
9,10
Search software
has been designed for the identification of cross-linked
peptides, for example Kojak,
11
xQuest,
12
pLink,
13
XlinkX,
14
or Xi.
5
In addition, cross-linking workflows can make use of
preprocessing methods to improve data quality and reduce file
sizes,
15
as well as postprocessing methods to filter out false
identifications
11,16
and custom-tailored false discovery rate
(FDR) estimation.
17−19
Preprocessing can improve peptide
identification by correcting the MS1 precursor ion m/zand
simplifying MS2 fragment spectra. Established proteomics
software perform such preprocessing, including MaxQuant
20,21
and OpenMS.
22,23
For example, MaxQuant performs a variety
of preprocessing steps: it corrects the precursor m/zby an
intensity-weighted average if a suitable peptide feature is found,
reassigns the monoisotopic peak and contains options for
intensity filtering of MS2 peaks. Despite such correction of the
precursor mass, many linear search engines have integrated the
possibility of considering multiple monoisotopic peaks during
search.
24−26
However, the benefits of this search feature are
currently unclear. It seems that the assignment of mono-
isotopic mass for tryptic peptides is already achieved
adequately either during acquisition or as part of preprocess-
ing.
Cross-linked peptides have characteristics that may render
MS1 monoisotopic precursor mass assignment as used for
linear peptides nonoptimal: high-charge states, large masses,
and low abundances. Several cross-link search engines include
MS1 correction in their pipeline: pLink corrects monoisotopic
peaks based on previous work with linear peptides,
27
however
does not include a parameter for searching multiple precursor
masses. Kojak averages precursor ion signals of neighboring
scans to create a composite spectrum and infer the true
monoisotopic mass of the precursor. If this step fails, precursor
masses up to −2 Da lighter are searched.
11
For previous
searches in Xi, MaxQuant was used to perform preprocessing.
Neither xQuest nor XLinkX describe precursor correction in
their workflow documentation and there is no option for
additionally searched masses available in the respective search
Received: August 4, 2018
Published: October 8, 2018
Article
pubs.acs.org/jpr
Cite This: J. Proteome Res. 2018, 17, 3923−3931
© 2018 American Chemical Society 3923 DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
This is an open access article published under a Creative Commons Attribution (CC-BY)
License, which permits unrestricted use, distribution and reproduction in any medium,
provided the author and source are cited.
Downloaded via 88.130.155.49 on February 14, 2022 at 10:19:59 (UTC).
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
parameters. We are not aware of a detailed evaluation of the
impact of different preprocessing techniques for cross-link
identification, independent of the search software. Correcting
the monoisotopic mass of precursors, although acknowledged
as an issue,
11,28
awaits systematic evaluation.
In this study, we show that errors in assigning monoisotopic
peaks during data acquisition are frequent for cross-linked
peptides because of their size and generally low abundance.
This adversely affects their identification. We show that
standard software suites, MaxQuant and OpenMS correct
monoisotopic precursor masses of cross-linked peptides with
variable success. We then implement an option in Xi to
consider multiple precursor masses during search, to minimize
the impact of false monoisotopic precursor mass assignment
on the identification of cross-links.
■METHODS
Data Sets
In this study, we used three publicly available data sets (Table
1). The three data sets were chosen to reflect a range of
applications of cross-linking mass spectrometry as well as a
range of data complexity: the first data set is Human Serum
Albumin (HSA) cross-linked with succinimidyl 4,4-azipenta-
noate (SDA) and fragmented using five different methods
(PXD003737).
29
The second data set is a pooled pseudocom-
plex sample with seven separately cross-linked proteins with
bis(sulfosuccinimidyl) suberate (BS3) (PXD006131).
7
This
data set includes data from four different fragmentation
methods. The third data set is the most complex sample,
composed of 15 size exclusion chromatography fractions of
Chaetomium thermophilum lysate cross-linked with BS3 and
fragmented only with HCD (PXD006626).
30
The first and last
size exclusion fractions were used to optimize the search
parameters for this data set. All samples were analyzed on an
Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo
Fisher Scientific, San Jose, CA) using Xcalibur (version 2.0 and
2.1).
Preprocessing
Raw files were preprocessed independently using MaxQuant
(1.5.5.30), OpenMS (2.0.1) and the ProteoWizard
31
tool
msconvert (3.0.9576) for comparison. Scripts automating the
preprocessing, search and evaluation were written in Python
(2.7).
The essential steps during the preprocessing can be divided
into two parts: (1) correction of the m/zor charge of the
precursor peak for MS2 spectra and (2) denoising of MS2
spectra. MaxQuant and OpenMS both try to correct the
precursor information via additional feature finding steps, i.e.
identifying a peptide feature from the retention time, m/zand
intensity domain of the LC-MS run. Additionally, denoising of
the MS2 spectra is performed by simply filtering the most
intense peaks in defined m/zwindows. The preprocessing is by
default enabled in MaxQuant and was run using the partial
processing option (steps 1−5) with default settings except for
inactivated “deisotoping”and “top peaks per 100 Da”, which
was set to 20. The OpenMS preprocessing workflow includes
centroiding, feature finding,
32
precursor correction (mass and
charge) using the identified features and MS2 denoising as
described above (Supporting Information (SI) Figure S1).
Msconvert was used to convert the raw files to mgf files
without any correction. These peak files were denoted as
“uncorrected”and used as our baseline to quantify improve-
ments in the subsequent database search. For the “in-search
assignment of the monoisotopic peak”in Xi (Xi-MPA), we
used msconvert to convert raw files to mgf files and included a
MS2 peak filter for the 20 most intense peaks in a 100 m/z
window.
Data Analysis
Peak files were searched separately in Xi (1.6.731) with the
following settings: MS accuracy 3 ppm, MS/MS accuracy 10
ppm, oxidation of methionine as variable modification, tryptic
digestion, two missed cleavages. For samples cross-linked with
SDA, linkage sites were allowed on lysine, serine, tyrosine,
threonine, and protein n-terminus on one end and all amino
acids on the other end of the cross-linker. Variable
modifications were monolink SDA (110.048 Da), SDA loop-
links (82.0419 Da), SDA hydrolyzed (100.0524 Da), SDA
oxidized (98.0368 Da)
31
as well as carbamidomethylation on
cysteine. For searches with BS3, linkage sites were lysine,
serine, threonine, tyrosine, and the protein n-terminus.
Carbamidomethylation on cysteine was set as fixed mod-
ification. Allowed variable modifications of the cross-linker
were aminated BS3 (155.0946 Da), hydrolyzed BS3 (156.0786
Da) and loop-linked BS3 (138.0681 Da). For collision-induced
dissociation (CID) and beam-type CID, also referred to as
higher-energy C-trap dissociation (HCD), b- and y-ions were
searched for, whereas for electron transfer dissociation (ETD)
c- and z-ions were allowed. For ETciD and EThcD, b-, c-, z-,
and y-ions were allowed. The HSA and pseudocomplex data
sets were searched against the known proteins in the sample.
For each protein fraction of the C. thermophilum data set, the
databases of the original publication were used, where a
database was created for each fraction by taking the most
abundant proteins (iBAQ value above 106). For searches
employing Xi-MPA, the parameter "missing_isotope_peaks"
was set to the respective mass range searched. Data sets 1 and
2 were searched with a reversed decoy database, whereas data
set 3 was searched with a shuffled decoy database due to
palindromic sequences. For the reversed decoy database,
lysines and arginines were swapped with the preceding amino
acid before peptide generation.
17,20
For cross-linking, there are different information levels:
PSMs, peptide pairs, residue pairs (links) and protein pairs.
The false discovery rate (FDR) can be calculated on each one
of these levels and should be reported for the level at which the
information is given.
12
The FDR was calculated as described in
Fischer et al.
17
using xiFDR (1.0.14.34) according to the
following equation: =
−
FDR TD DD
TT . A 5% PSM level cutoff
was imposed. The setting “uniquePSMs”was enabled and the
FDR was calculated separately on self-and between links.
Minimal peptide length was set to 6. In data set 2, identified
cross-linked residues were mapped to the crystal structure of
the respective protein and the Euclidian distance between the
alpha-carbons was calculated. Structures were downloaded
Table 1. Overview of Datasets Used
data set sample database size
a
reference
1 HSA 1 Giese et al. 2016
2 pseudocomplex 7 Kolbowski et al. 2017
3C. thermophilum 198−400
b
Kastritis et al. 2017
a
Database size refers to the number of proteins in the database.
b
Multiple size exclusion chromatography fractions (n= 15).
Journal of Proteome Research Article
DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
3924
from the PDB (IDs: 1AO6, 5GKN, 2CRK, 3NBS, 1OVT,
2FRJ). Kojak (1.5.5) was run via the Trans-Proteomic Pipeline
(5.1.0)
33
using default settings except: MS1 resolution 120
000, BS3 allowed on lysine, serine, threonine, tyrosine, and the
protein n-terminus, aminated BS3 (155.0946 Da) as variable
modification of the cross-linker, 3 ppm mass tolerance on MS1
level. For the uncorrected search, the isotope error was set to 0
and precursor refinement was disabled. PSMs were validated
using PeptideProphet
34
and FDR calculated as described above
on the resulting probability.
The mass spectrometry data have been deposited to the
ProteomeXchange Consortium via the PRIDE
35
partner
repository with the data set identifier PXD011121. For
transparency, python scripts are available on GitHub under:
https://github.com/Rappsilber-Laboratory/Xi-MPA_scripts.
■RESULTS AND DISCUSSION
We evaluated the impact on cross-link identification in Xi of
changing the precursor monoisotopic mass that was initially
assigned during data acquisition (“uncorrected”). In this
analysis, MaxQuant and OpenMS were used as preprocessing
tools. We used three different data sets that differ in complexity
and fragmentation regimes. To measure the improvements
from using the preprocessing tools, a simple conversion from
raw files to mgf format was done with msconvert and used as a
baseline. Note that in the spectrum header, there are two m/z
values: the trigger mass of the MS2 and the assigned
monoisotopic peak of the isotope cluster. Msconvert extracts
the assigned monoisotopic mass. Processed data were searched
separately in Xi and evaluated on PSM level or link (= unique
residue pair) level, with a 5% FDR. Finally, the newly
implemented in-search assignment of monoisotopic peaks in Xi
was compared to the elaborate preprocessing pipelines in
OpenMS and MaxQuant.
Preprocessing Increases the Number of Cross-Link PSMs
by Finding the Correct Monoisotopic Peak
The data sets were preprocessed in MaxQuant and OpenMS
and numbers of identified PSMs were compared to those
obtained using uncorrected data. Data sets 1 (HSA) and 2
(pseudocomplex) were acquired with different acquisition
methods. For comparability to data set 3 (complex mixture),
we focused on the HCD acquired data. Cross-links between
proteins were excluded, either because they were experimen-
tally not possible (data set 2) or observed in too low numbers
for reliable FDR calculation (data set 3).
For uncorrected data, 672, 354, and 2157 cross-link PSMs
resulted for the HSA data set (data set 1), pseudocomplex
(data set 2), and first and last fractions of C. thermophilum
respectively (data set 3). Both preprocessing approaches
improved numbers of identified PSMs for all data sets:
Preprocessing in MaxQuant led to 1127 (68% increase), 966
(173% increase), and 2966 (38% increase), while for OpenMS,
1044 (55% increase), 598 (69% increase) and 2394 (11%
increase) PSMs were identified (Figure 1A).
We assessed the gains in identified PSMs of preprocessed
data compared to uncorrected data (focusing on data set 2)
regarding three forms of precursor correction: (1) correction
of the monoisotopic mass, (2) charge state correction, and (3)
small corrections of the m/zvalue based on averaging the m/z
values across the peptide feature (Figure 1B). Precursor mass
and charge state of spectra identified solely in MaxQuant-Xi
were compared to their counterparts when searching
uncorrected data in Xi. Of the 756 newly identified spectra,
686 (91%) had a different monoisotopic precursor mass.
Precursors were primarily corrected to lighter masses by
MaxQuant, that is, the monoisotopic peak correction by −1
(208 spectra), −2 (215 spectra), −3 (149 spectra), and −4Da
(62 spectra). Greater shifts (−5to−7 Da) only occurred 30
times, and corrections to heavier masses were observed 22
times. Only 30 spectra (4%) were corrected in their charge
state. For the 60 spectra (8%) without correction in charge
state or monoisotopic peak, we only identified nine spectra
that had a higher error than 3 ppm before preprocessing,
indicating a small correction of the initial precursor m/z(by
averaging of peptide feature peaks). The main proportion of
these identifications is likely a product of noise removal in MS2
spectra or small changes in the score distribution. Similarly, for
OpenMS-Xi, the monoisotopic peak correction had the
greatest impact: Of the 314 spectra that OpenMS added
over uncorrected data, 139 were precursor corrected by −1Da
and 108 to −2 Da. In contrast to MaxQuant, corrections to −3
or lighter were not observed, which might explain the higher
number of identifications obtained with MaxQuant-Xi.
Figure 1. Correction of the monoisotopic peak is crucial in cross-link identification. (A) The data sets were preprocessed using MaxQuant and
OpenMS, leading to more identified PSMs in all cases. Fold changes from uncorrected data (msconvert conversion of Xcalibur data) were
calculated for each file separately and the mean plotted. Error bars represent the standard error of the mean between different acquisitions (HSA: n
= 3, pseudocomplex: n=3,C. thermophilum:n= 8). (B) The majority of additional identifications after preprocessing are due to correction of the
precursor mass to lighter monoisotopic masses. Spectra that are unique to MaxQuant preprocessed searches of HCD acquisitions from data set 2
were evaluated in terms of precursor correction. The main proportion of the gain was corrected to lighter masses of up to −3 Da, while charge state
correction or correction to heavier masses rarely occurred. (C) Isotope cluster of a corrected precursor of m/z992.71 (z=5,m= 4958.6 Da) was
solely identified in MaxQuant preprocessed results. In OpenMS preprocessed and uncorrected data, the wrong monoisotopic mass was selected for
unknown reasons.
Journal of Proteome Research Article
DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
3925
For data set 1 and 3, the gains of preprocessing are smaller
than for data set 2. The median peptide mass of data set 1
(3368 Da) is smaller than the median mass of data set 2 (3946
Da) and we later show that this is a major factor in precursor
mass assignment. This reflects in the distribution of lighter
masses assigned: 46%, 33%, and 21% were corrected by −1 Da,
−2 Da, and to even lighter peaks, respectively. Data set 3 was
acquired with a different version of Xcalibur, for which we saw
better mass assignment than for earlier versions (data not
shown). As an implication of this, already 67% of the lighter
corrected masses are shifted by −1 Da, 21% by −2 Da, 12% to
even lighter masses. However, we were not able to follow up
on this to our content, since the source code of the vendor
software is not available.
In summary, preprocessing, especially monoisotopic peak
correction, leads to a notable increase in identifications. Using
the 3-dimensional peptide feature is advantageous compared to
on-the-fly detection of the monoisotopic peak. If the preceding
MS1 spectrum was acquired during the beginning (or end) of
the elution profile of a peptide, the intensity will be low. Thus,
the monoisotopic peak might not even be detectable at the
time of fragmentation. For large (cross-linked) peptides, this
effect might be exacerbated by the monoisotopic peak usually
being less intense than other isotope peaks. Therefore, using
the additional information from the retention time domain will
be beneficial. The same feature information can also be used to
determine or validate the assigned charge state of the
precursor. However, the instrument software almost always
assigned the same charge state as MaxQuant or OpenMS.
Thus, the main advantage for identifying cross-linked peptides
arises from monoisotopic peak correction.
Interestingly, OpenMS and MaxQuant did not always agree
on or find the same monoisotopic peak (Figure 1C). Of the
total MaxQuant-corrected spectra with a different mono-
isotopic mass, 81% were not corrected and 6% corrected
differently with OpenMS. Vice versa, 15% of the monoisotopic
peaks corrected by OpenMS were not corrected by MaxQuant
and 25% were corrected differently. Both MaxQuant and
OpenMS have their own implementations for precursor
correction - therefore, there might be instances where
MaxQuant is able to find a corresponding peptide feature
where OpenMS does not and vice versa. Although OpenMS
did not lead to the same improvements in the number of
identifications as MaxQuant, it did correct some precursors
that the latter did not. We therefore suspect that there are also
precursors with a falsely assigned monoisotopic peak that were
corrected with neither algorithm. Furthermore, 3-dimensional
detection of peptide features is challenging for low intensity
peptides. In conclusion, there likely remain falsely assigned
monoisotopic peaks in the data, ultimately leading to missed or
false identifications.
In-Search Monoisotopic Peak Assignment Increases the
Number of Identifications
We observed multiple cases where MaxQuant and OpenMS
disagreed in their monoisotopic peak choice, indicating that
the problem of monoisotopic peak assignment (MPA) cannot
be solved easily at MS1 level. Indeed, we found instances
where the monoisotopic peak is simply not distinguishable
from noise, so a feature-based correction would not be feasible.
Nevertheless, the associated MS2 spectra could be matched to
a cross-linked peptide when considering multiple different
monoisotopic masses during search. This shows that the extra
information on obtaining a peptide-spectrum match is
advantageous to MPA over considering MS1 information
alone. Therefore, we implemented a monoisotopic peak
assignment in Xi: for each MS2 spectrum, multiple precursor
masses are considered during a single search and the highest
scoring peptide-pair assigns the precursor mass. Note that this
is different from simply searching with a wide mass error for
MS1. The mass accuracy of MS1 is minimally compromised as
multiple candidates for the monoisotopic mass are taken and
considered with the original mass accuracy of the measure-
ment.
To find a good trade-offbetween increased search space and
sensitivity, we tested different mass range settings on the data
sets. For data set 2 (HCD subset), the number of PSMs
increased with ranges up to −5 Da on the considered
monoisotopic masses (Figure 2A). However, the increase in
identifications from −4to−5 Da was only 3% and considering
the increase in search time, we continued with a maximal
correction to −4 Da as the optimal setting for this data set. Xi-
MPA yielded 1508 PSMs, which is a 326% increase compared
to searching uncorrected data and a 56% increase compared to
MaxQuant-Xi. Similar improvements are observed for the
other fragmentation methods in this data set (SI Figure S2).
Additionally, we corrected up to −7 Da to test if a large
increase in search space increases random spectra matches as
measured by the target-decoy approach. The number of
identifications at 5% FDR decreased only slightly compared to
−5Da(−1%), but still led to more identifications than up to
−4 Da (3%). In the HSA data set, Xi-MPA with up to −4Da
increased the number of identified PSMs by 170% compared
to uncorrected data (Figure S3).
As a final evaluation of in-search monoisotopic peak
assignment, we searched the complete data set of C.
thermophilum. We used 0 to −3 Da as the range of Xi-MPA,
since an initial analysis of the first and last fraction of the C.
thermophilum data set returned a similar number of
identifications when running Xi-MPA up to −4Daor−3
Da (Figure S4). As a comparison, we took the original peak
files obtained from PRIDE. The FDR was calculated separately
on self-and between links, enabled boosting (automatic
prefiltering on PSM and peptide pair level
17
), with a minimum
of three fragments per peptide and a minimal delta score of 0.5.
For the original peak files, which were preprocessed in
MaxQuant, we identified 3848 PSMs, 2594 peptide pairs and
1653 cross-links, with a 5% FDR on each respective level
(Figure 2B). Xi-MPA resulted in 4952 PSMs (29% increase),
3566 peptide pairs (37% increase), and 2273 cross-links (38%
increase).
Next, we looked into the complementarity of search results
with the different approaches, using data set 2 at 5% link-FDR.
Preprocessing via MaxQuant and OpenMS led to 172 and 158
links, respectively, while Xi-MPA resulted in 243 links. While
the overlap between links of OpenMS-Xi and MaxQuant-Xi is
only 50%, Xi-MPA identifications cover 76% of both searches
(Figure 2C). Nineteen and 23 links are uniquely found in
MaxQuant and OpenMS preprocessed data, respectively.
However, there are five decoy links as well in each unique
set (resulting in a link-FDR of 26% and 22%). For Xi-MPA,
there are 75 unique target links with 12% link-FDR.
Identification-based monoisotopic peak assignment as
employed by Xi-MPA results in more identifications than the
feature-based assignment algorithms of OpenMS and Max-
Quant. Neither OpenMS nor MaxQuant correct all precursor
Journal of Proteome Research Article
DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
3926
masses that are incorrectly assigned during data acquisition. In
Xi-MPA, spectra are searched with multiple monoisotopic
masses, thereby relying less on the MS1 information. The
quality of the precursor isotope cluster does not contribute to
the decision of monoisotopic mass and spectra for which
correction failed will be identifiable. One could hypothesize
that increasing the search space by considering multiple masses
will lead to more false positives, thereby reducing the number
of true identifications. This is not the case, as we match
substantially more PSMs at constant FDR by considering
alternative monoisotopic masses. As a second plausible caveat,
this approach increases the search time. However, the use of
relatively cheap computational time appears balanced by the
notable increase in identified cross-links. The optimal range of
additional monoisotopic peaks to search will however be
dependent on complexity and quality of MS1 acquisition and
the instrument software. To reduce the mass range considered
in Xi-MPA, we developed a MS1 level-based approach. For
each precursor, we search lighter isotope peaks in MS1 and use
this to narrow the search space (explained in detail in the SI).
This led to an average of 24% less values to be considered,
while only reducing the number of identifications by 3%. We
hope that our observation of the monoisotopic peak detection
challenge in cross-linking together with our publicly available
data sets will lead to further improvements in monoisotopic
peak-assignment algorithms in the future, possibly tailored to
cross-link data.
The cross-link search engine Kojak employs a precursor
correction in its pipeline.
11
As we could not find a detailed
evaluation of the impact of precursor correction in Kojak, we
searched the HCD data of the pseudocomplex data set without
correction as well as with their default correction settings. We
focused on FDR 10% data as there were too few identifications
for a reliable calculation of FDR 5%. Just 171 cross-linked
PSMs passed for the unprocessed data, whereas for the default
search, 1088 PSMs passed (536% increase). Of those, 862
(79%) were corrected in their monoisotopic precursor peak.
These results support our observations with Xi.
In-Search Monoisotopic Peak Assignment Does Not
Compromise Search Accuracy
Changing the search could lead to several problems. We
already excluded that the increased search space leads to high-
scoring decoy matches that in turn reduce the number of
identifications at a given FDR cutoff. As an additional
validation, we assessed our results against known PDB
structures using the HCD data from the pseudocomplex data
set (data set 2), at 5% link-FDR. Assuming a crystal structure is
correct, a cross-link can be unexpectedly long either because
the link is false or because of in-solution structural dynamics.
If, however, the proportion of long-distance links in results of
two approaches is identical, then at least the two results have
equal quality.
We first tested the results of all three approaches against
crystal structures. Residue pairs were mapped to PDB
structures and the distance between the two alpha-carbons
was calculated (see Methods). Thirty Å was set as the maximal
distance for BS3, links with a greater distance were classified as
long-distance. In this evaluation, we excluded the protein C3B
because its flexible regions make it unsuitable for this analysis.
For MaxQuant and OpenMS preprocessed results, 11.8% and
6.1% long-distance cross-links were identified, respectively. In
Xi-MPA, 8.1% long distance cross-links were identified (Figure
3A). Of the links uniquely identified through Xi-MPA, only
5.3% were long distance links. Therefore, Xi-MPA as such does
not lead to an enrichment in long-distance cross-links.
However, it could be that mass-corrected precursors tend to
have a higher proportion of long-distance links. We therefore
split the Xi-MPA results into five groups corresponding to the
monoisotopic mass change (0, −1, −2, −3, −4 Da) and looked
at their match to crystal structures. If a link originated from
PSMs with different mass corrections, all of those were
considered. We conducted a “nonparametric ANOVA”
(Kruskal−Wallis test) to detect any significant changes in the
distance distributions of Xi-MPA identifications with different
Figure 2. In-search monoisotopic peak assignment outperforms
preprocessing. (A) Performance of Xi-MPA on data set 2. HCD data
from the pseudocomplex data set were searched assuming different
ranges of missing monoisotopic peaks. With increased ranges, the
number of identified PSMs also increases. (B) Performance of Xi-
MPA on the complete C. thermophilum data set. All 15 fractions were
searched with the original preprocessed data as well as with Xi-MPA.
(C) Overlap of identified residue pairs of MaxQuant-Xi and OpenMS-
Xi to residue pairs gained from Xi-MPA (data set 2). Numbers in
brackets are the proportion of decoys in the respective regions.
Journal of Proteome Research Article
DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
3927
shifts and decoy distribution. However, we fail to reject the
null hypothesis at the predetermined significance level of α=
0.05 (p-value: 0.13), indicating that the distance distributions
for all subsets are similar. This matches the visual inspection of
distance distributions (Figure 3B). Furthermore, all individual
distance distributions were significantly smaller than the
derived reference distribution (one-sided Wilcoxon test, see
SI Table S5). In conclusion, we do not see any evidence of in-
search monoisotopic mass assignment leading to increased
conflicts with crystal structures. We then evaluated the effect of
in-search monoisotopic mass assignment on PSM quality as
assessed by the search score. First, we compared the scores of
PSMs with a mass shift (Xi-MPA identifications) to the scores
of the same spectrum without a mass shift (uncorrected data).
While scores with shifted mass have a median of 6.7, the
median score is 2.3 when using the uncorrected masses (Figure
3C). As one would expect from an increased search space, the
scores of decoy hits also improve, albeit only marginally. We
find that the score difference of target PSMs is significantly
larger than of decoy PSMs (one-sided Wilcoxon test, p-value:
<2.2 ×10−16). We then turned to a “decoy mass search”for
which we not only searched the range from 0 Da to −4 Da, but
also +1 Da to +4 Da. Assuming the monoisotopic peak in the
uncorrected data is rarely lighter than the true monoisotopic
peak, the new identifications should score like decoy
identifications. Indeed, the resulting score distributions for
targets with a positive mass shift follow the decoy distribution
(Figure 3D). In contrast, identifications with a negative shift
are distributed like the identifications without mass shift. In
conclusion, in-search monoisotopic mass change leads to
significantly improved scores with a distribution that resembles
that of precursors that did not see a mass change (0 Da).
Importantly, these improvements are not random since an
equally large search space increase (+1 Da to +4 Da) results in
a completely different score distribution that resembles the
decoy distribution but not the distribution of identifications
without a mass shift.
Heavy and Low Intensity Peptides Are Corrected More
Frequently
One would especially expect to observe shifted mass
assignment for peptides of high mass and low abundance.
For large peptides (approximately >2000 Da), the mono-
isotopic peak will not be the most intense peak in the isotope
cluster. If the peptide is of low abundance, the monoisotopic
Figure 3. Matches with and without in-search mass shift show similar
quality metrics. (A) Evaluation of Xi-MPA derived links on crystal
structures (data set 2). Distances between αcarbon atoms of
identified cross-linked residues in the crystal structure of the proteins
are shown in light gray while a reference distribution of all possible
Figure 3. continued
pairwise C-alpha distances of cross-linkable residues is shown in dark
gray. Thirty Å is set as a limit, above which links are defined as long
distance. (B) Distance distribution of identifications with different
mass corrections. There was no significant difference between the
different mass shifts, while all had a significant difference to the decoy
distribution. (C) PSM scores of spectra identified with a mass shift are
significantly higher than the corresponding score in uncorrected data.
Shown are the score distributions of uncorrected and Xi-MPA results,
as well as the corresponding decoy distribution. (D) Score
distribution of PSM matches of the “decoy mass search”.
Identifications with a positive mass shift generally follow the decoy
distribution (note that there are correct identifications with a positive
mass shift, albeit few, see Figure 1B) while identifications with a
negative shift resemble unshifted identifications. The scores of
negative-shifted PSMs are significantly higher than those of positive-
shifted PSMs (one-sided Wilcoxon test, p-value: <2.2×10−16).
Journal of Proteome Research Article
DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
3928
peak may be of too low intensity to be detected. We therefore
analyzed the monoisotopic peak assignment in Xi-MPA
regarding the precursor mass and intensity. Indeed, precursors
with higher masses are more often corrected to lighter
monoisotopic peaks (Figure 4A). While the median precursor
mass for uncorrected matches is 2952 Da, for matches
corrected by −2 Da it is already 4062 Da and for −4Dait
is 4684 Da. Of the identifications with a mass above 3000 Da,
88% were identified with a lighter mass. For precursors lighter
than 3000 Da, the proportion was 42%. Like for mass
dependency, there is a trend toward larger correction ranges
for lower intensity peptides (SI Figure S5). However, this is
less strong than it is for precursor mass.
When evaluating the newly matched precursors of Xi-MPA,
the advantage of not having to rely on MS1 identification is
evident. Matches not made through any of the preprocessing
methods are generally much less intense (Figure 4B) and larger
(SI Figure S6) than matches that are common to all
approaches. Manual analysis of isotope clusters of corrected
precursors from data set 2 revealed many cases where the
monoisotopic peak was present in the MS1 spectrum but was
not recognized during acquisition. For some, this might be due
to the peak being of low intensity and discarded as noise, or
because of other interfering peaks (Figure 4C). However, there
are also cases where the cluster is well resolved (Figure 1C).
Without details of how the instrument software determines the
monoisotopic peak, a full evaluation is difficult. For a complete
list of precursor m/zfor Xi-MPA identifications and
corresponding m/zof uncorrected, MaxQuant and OpenMS
data, see SI Table S1−S3.
Note that in many acquisition methods, the machine only
fragments peaks where it can successfully identify a full isotope
cluster. Therefore, there might be instances of cross-linked
peptides not being fragmented because of insufficient isotopic
cluster quality, leading to lost identifications.
■CONCLUSION
The size and low abundance of cross-linked peptides leads to
frequent misassignment of the monoisotopic mass by instru-
ment software, which in some instances even escapes
correction by sophisticated correction approaches employed
by MaxQuant and OpenMS. Considering multiple mono-
isotopic masses during search increases the number of cross-
link PSMs 1.8−4.2-fold, without compromising search
accuracy as judged by multiple assessment strategies including
comparison of the gains against solved protein structures. The
problem of wrongly assigned monoisotopic peaks will have an
impact on most cross-link search engines since these all rely in
some part on the precursor mass. The extent of the
misassignment will however be sample and software-depend-
ent. Even with improved acquisition or correction software,
there will remain instances where the monoisotopic peak
cannot be determined correctly before searching due to low
intensity. Our search-assisted monoisotopic peak assignment
provides a general solution to this problem by relying on MS2
identification in addition to precursor information.
■ASSOCIATED CONTENT
*
SSupporting Information
The Supporting Information is available free of charge on the
ACS Publications website at DOI: 10.1021/acs.jproteo-
me.8b00600.
Table S4: Xi-MPA mass range reduction. Figure S1:
OpenMS preprocessing workflow. Figure S2: Perform-
ance of Xi-MPA on EThcD, CID, and ETciD
acquisitions of the pseudocomplex data set. Table S5:
Summary for conducted significance tests for Figure 3B
in the main text. Figure S3: Performance of Xi-MPA on
HCD data of the HSA data set. Figure S4: Performance
of Xi-MPA on the test fractions of the C. thermophi-
lumdata set. Figure S5: Dependency of the monoisotopic
mass correction on precursor intensity. Figure S6:
Dependency of identifications in Xi-MPA on mass
(PDF)
Table S1: Precursor m/zof different processing methods
of data set 1. Table S2: Precursor m/zof different
processing methods of data set 2. Table S3: Precursor
m/zof different processing methods of data set 3.
Supporting Information: MS1 based mass range
reduction (XLS)
Figure 4. Correction is dependent on precursor mass and intensity. (A) Box plot of the precursor mass and monoisotopic mass correction of
identified PSMs after Xi-MPA. PSMs with higher mass more often require monoisotopic mass correction to lighter masses. Whiskers show the 5
and 95% quantiles of the data. Asterisks denote the significance calculated by a one tailed ttest (****:p-value: <0.0001). (B) Precursors of cross-
linked PSMs identified in all three approaches, MaxQuant-Xi, OpenMS-Xi, and Xi-MPA (“common”), are more intense than precursors of PSMs
that are only identified in Xi-MPA. In other words, successful correction happens more often for abundant precursors, whereas Xi-MPA identifies
precursors of lower intensity. (C) MS1 isotope cluster of a cross-linked peptide. The monoisotopic peak of m/z758.16 (z=4,m= 3028.6 Da) was
falsely assigned during acquisition and not corrected in any preprocessing approach. Xi-MPA identifies a PSM for a precursor with a mass that is 3
Da lighter.
Journal of Proteome Research Article
DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
3929
■AUTHOR INFORMATION
Corresponding Author
*E-mail: [email protected].
ORCID
Swantje Lenz: 0000-0002-8839-5371
Sven H. Giese: 0000-0002-9886-2447
Lutz Fischer: 0000-0003-4978-0864
Juri Rappsilber: 0000-0001-5999-1310
Notes
The authors declare no competing financial interest.
■ACKNOWLEDGMENTS
We thank Dr Francis O’Reilly for comments and helpful
discussions. This work was supported by the Einstein
Foundation, the DFG [RA 2365/4-1], and the Wellcome
Trust through a Senior Research Fellowship to JR [103139]
and a multiuser equipment grant [108504]. The Wellcome
Centre for Cell Biology is supported by core funding from the
Wellcome Trust [203149].
■REFERENCES
(1) Leitner, A.; Reischl, R.; Walzthoeni, T.; Herzog, F.; Bohn, S.;
Fo rster, F.; Aebersold, R. Expanding the chemical cross-linking
toolbox by the use of multiple proteases and enrichment by size
exclusion chromatography. Mol. Cell. Proteomics 2012,11,
M111.014126.
(2) Chen, Z. A.; Jawhari, A.; Fischer, L.; Buchen, C.; Tahir, S.;
Kamenski, T.; Rasmussen, M.; Lariviere, L.; Bukowski-Wills, J.-C.;
Nilges, M.; et al. Architecture of the RNA polymerase II-TFIIF
complex revealed by cross-linking and mass spectrometry. EMBO J.
2010,29, 717−726.
(3) Tan, D.; Li, Q.; Zhang, M.-J.; Liu, C.; Ma, C.; Zhang, P.; Ding,
Y.-H.; Fan, S.-B.; Tao, L.; Yang, B. et al. Trifunctional cross-linker for
mapping protein-protein interaction networks and comparing protein
conformational states. Elife 2016,5,DOI: DOI: 10.7554/
eLife.12509.
(4) Rampler, E.; Stranzl, T.; Orban-Nemeth, Z.; Hollenstein, D. M.;
Hudecz, O.; Schlo gelhofer, P.; Mechtler, K. Comprehensive Cross-
Linking Mass Spectrometry Reveals Parallel Orientation and Flexible
Conformations of Plant HOP2-MND1. J. Proteome Res. 2015,14,
5048−5062.
(5) Mendes, M. L.; Fischer, L.; Chen, Z. A.; Barbon, M.; O’Reilly, F.
J.; Bohlke-Schneider, M.; Belsom, A.; Dau, T.; Combe, C. W.;
Graham, M. et al. An integrated workflow for cross-linking/mass
spectrometry. 2018.
(6) Belsom, A.; Schneider, M.; Fischer, L.; Mabrouk, M.; Stahl, K.;
Brock, O.; Rappsilber, J. Blind testing cross-linking/mass spectrom-
etry under the auspices of the 11thcritical assessment of methods of
protein structure prediction (CASP11). Wellcome open research 2016,
1, 24.
(7) Kolbowski, L.; Mendes, M. L.; Rappsilber, J. Optimizing the
Parameters Governing the Fragmentation of Cross-Linked Peptides in
a Tribrid Mass Spectrometer. Anal. Chem. 2017,89, 5311−5318.
(8) Liu, F.; Lo ssl, P.; Scheltema, R.; Viner, R.; Heck, A. J. R.
Optimized fragmentation schemes and data analysis strategies for
proteome-wide cross-link identification. Nat. Commun. 2017,8,
15473.
(9) Orba
n-Ne
meth, Z.; Beveridge, R.; Hollenstein, D. M.; Rampler,
E.; Stranzl, T.; Hudecz, O.; Doblmann, J.; Schlo gelhofer, P.; Mechtler,
K. Structural prediction of protein models using distance restraints
derived from cross-linking mass spectrometry data. Nat. Protoc. 2018,
13, 478−494.
(10) Schneider, M.; Belsom, A.; Rappsilber, J. Protein Tertiary
Structure by Crosslinking/Mass Spectrometry. Trends Biochem. Sci.
2018,43, 157−169.
(11) Hoopmann, M. R.; Zelter, A.; Johnson, R. S.; Riffle, M.;
MacCoss, M. J.; Davis, T. N.; Moritz, R. L. Kojak: efficient analysis of
chemically cross-linked protein complexes. J. Proteome Res. 2015,14,
2190−2198.
(12) Leitner, A.; Walzthoeni, T.; Aebersold, R. Lysine-specific
chemical cross-linking of protein complexes and identification of
cross-linking sites using LC-MS/MS and the xQuest/xProphet
software pipeline. Nat. Protoc. 2014,9, 120−137.
(13) Yang, B.; Wu, Y.-J.; Zhu, M.; Fan, S.-B.; Lin, J.; Zhang, K.; Li,
S.; Chi, H.; Li, Y.-X.; Chen, H.-F.; et al. Identification of cross-linked
peptides from complex samples. Nat. Methods 2012,9, 904−906.
(14) Liu, F.; Rijkers, D. T. S.; Post, H.; Heck, A. J. R. Proteome-wide
profiling of protein assemblies by cross-linking mass spectrometry.
Nat. Methods 2015,12, 1179−1184.
(15) Renard, B. Y.; Kirchner, M.; Monigatti, F.; Ivanov, A. R.;
Rappsilber, J.; Winter, D.; Steen, J. A. J.; Hamprecht, F. A.; Steen, H.
When less can yield more - Computational preprocessing of MS/MS
spectra for peptide identification. Proteomics 2009,9, 4978−4984.
(16) Ka ll, L.; Canterbury, J. D.; Weston, J.; Noble, W. S.; MacCoss,
M. J. Semi-supervised learning for peptide identification from shotgun
proteomics datasets. Nat. Methods 2007,4, 923−925.
(17) Fischer, L.; Rappsilber, J. Quirks of Error Estimation in Cross-
Linking/Mass Spectrometry. Anal. Chem. 2017,89, 3829−3833.
(18) Walzthoeni, T.; Claassen, M.; Leitner, A.; Herzog, F.; Bohn, S.;
Fo rster, F.; Beck, M.; Aebersold, R. False discovery rate estimation for
cross-linked peptides identified by mass spectrometry. Nat. Methods
2012,9, 901−903.
(19) Maiolica, A.; Cittaro, D.; Borsotti, D.; Sennels, L.; Ciferri, C.;
Tarricone, C.; Musacchio, A.; Rappsilber, J. Structural analysis of
multiprotein complexes by cross-linking, mass spectrometry, and
database searching. Mol. Cell. Proteomics 2007,6, 2200−2211.
(20) Cox, J.; Mann, M. MaxQuant enables high peptide
identification rates, individualized p.p.b.-range mass accuracies and
proteome-wide protein quantification. Nat. Biotechnol. 2008,26,
1367−1372.
(21) Tyanova, S.; Temu, T.; Cox, J. The MaxQuant computational
platform for mass spectrometry-based shotgun proteomics. Nat.
Protoc. 2016,11, 2301−2319.
(22) Sturm, M.; Bertsch, A.; Gro pl, C.; Hildebrandt, A.; Hussong,
R.; Lange, E.; Pfeifer, N.; Schulz-Trieglaff, O.; Zerck, A.; Reinert, K.;
et al. OpenMS - an open-source software framework for mass
spectrometry. BMC Bioinf. 2008,9, 163.
(23) Ro st, H. L.; Sachsenberg, T.; Aiche, S.; Bielow, C.; Weisser, H.;
Aicheler, F.; Andreotti, S.; Ehrlich, H.-C.; Gutenbrunner, P.; Kenar,
E.; et al. OpenMS: a flexible open-source software platform for mass
spectrometry data analysis. Nat. Methods 2016,13, 741−748.
(24) Craig, R.; Beavis, R. C. TANDEM: matching proteins with
tandem mass spectra. Bioinformatics 2004,20, 1466−1467.
(25) Eng, J. K.; Jahan, T. A.; Hoopmann, M. R. Comet: an open-
source MS/MS sequence database search tool. Proteomics 2013,13,
22−24.
(26) Geer, L. Y.; Markey, S. P.; Kowalak, J. A.; Wagner, L.; Xu, M.;
Maynard, D. M.; Yang, X.; Shi, W.; Bryant, S. H. Open mass
spectrometry search algorithm. J. Proteome Res. 2004,3, 958−964.
(27) Yuan, Z.-F.; Liu, C.; Wang, H.-P.; Sun, R.-X.; Fu, Y.; Zhang, J.-
F.; Wang, L.-H.; Chi, H.; Li, Y.; Xiu, L.-Y.; et al. pParse: a method for
accurate determination of monoisotopic peaks in high-resolution mass
spectra. Proteomics 2012,12, 226−235.
(28) Iacobucci, C.; Sinz, A. To Be or Not to Be? Five Guidelines to
Avoid Misassignments in Cross-Linking/Mass Spectrometry. Anal.
Chem. 2017,89, 7832−7835.
(29) Giese, S. H.; Belsom, A.; Rappsilber, J. Optimized
Fragmentation Regime for Diazirine Photo-Cross-Linked Peptides.
Anal. Chem. 2016,88, 8239−8247.
(30) Kastritis, P. L.; O’Reilly, F. J.; Bock, T.; Li, Y.; Rogon, M. Z.;
Buczak, K.; Romanov, N.; Betts, M. J.; Bui, K. H.; Hagen, W. J.; et al.
Capturing protein communities by structural proteomics in a
thermophilic eukaryote. Mol. Syst. Biol. 2017,13,936.
Journal of Proteome Research Article
DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
3930
(31) Kessner, D.; Chambers, M.; Burke, R.; Agus, D.; Mallick, P.
ProteoWizard: open source software for rapid proteomics tools
development. Bioinformatics 2008,24, 2534−2536.
(32) Weisser, H.; Nahnsen, S.; Grossmann, J.; Nilse, L.; Quandt, A.;
Brauer, H.; Sturm, M.; Kenar, E.; Kohlbacher, O.; Aebersold, R.; et al.
An automated pipeline for high-throughput label-free quantitative
proteomics. J. Proteome Res. 2013,12, 1628−1644.
(33) Deutsch, E. W.; Mendoza, L.; Shteynberg, D.; Slagel, J.; Sun, Z.;
Moritz,R.L.Trans-ProteomicPipeline,astandardizeddata
processing pipeline for large-scale reproducible proteomics infor-
matics. Proteomics: Clin. Appl. 2015,9, 745−754.
(34) Keller, A.; Nesvizhskii, A. I.; Kolker, E.; Aebersold, R. Empirical
Statistical Model To Estimate the Accuracy of Peptide Identifications
Made by MS/MS and Database Search. Anal. Chem. 2002,74, 5383−
5392.
(35) Vizcaíno, J. A.; Csordas, A.; del-Toro, N.; Dianes, J. A.; Griss, J.;
Lavidas, I.; Mayer, G.; Perez-Riverol, Y.; Reisinger, F.; Ternent, T.;
et al. 2016 update of the PRIDE database and its related tools. Nucleic
Acids Res. 2016,44, D447−56.
Journal of Proteome Research Article
DOI: 10.1021/acs.jproteome.8b00600
J. Proteome Res. 2018, 17, 3923−3931
3931
Improved peptide backbone fragmentation is the primary advantage
of MS-cleavable crosslinkers
Lars Kolbowski*,1, Swantje Lenz*,1, Lutz Fischer1, Ludwig R Sinn1, Francis J O’Reilly1, Juri Rappsil-
ber1,2
* These authors contributed equally to this work
1 Technische Universität Berlin, Chair of Bioanalytics, 10623 Berlin, Germany
2 University of Edinburgh, Wellcome Centre for Cell Biology, Edinburgh EH9 3BF, UK
Proteome-wide crosslinking mass spectrometry studies have coincided with the advent of MS-cleavable crosslinkers that can reveal
the individual masses of the two crosslinked peptides. However, recently such studies have also been published with non-cleavable
crosslinkers suggesting that MS-cleavability is not essential. We therefore examined in detail the advantages and disadvantages of
using the most popular MS-cleavable crosslinker, DSSO. Indeed, DSSO gave rise to signature peptide fragments with a distinct
mass difference (doublet) for nearly all identified crosslinked peptides. Surprisingly, we could show that it was not these peptide
masses that proved the main advantage of MS-cleavability of the crosslinker, but improved peptide backbone fragmentation which
reduces ambiguity of peptide identifications. We also show that the more intricate MS3-based data acquisition approaches lack
sensitivity and specificity, causing them to be outperformed by the simpler and faster stepped HCD method. This understanding
will guide future developments and applications of proteome-wide crosslinking mass spectrometry.
INTRODUCTION
Crosslinking combined with mass spectrometry (Crosslink-
ing MS) is a powerful tool for detecting protein-protein inter-
actions and the structural characterization of proteins. Many
key advances have been made in recent years to expand the
complexity of the samples that can be analysed with this tech-
nology. These include the database search software1–3, FDR
estimation4 and the enrichment of crosslinked peptides5–7. One
of the key problems when identifying crosslinked peptides is
that one must in principle identify two peptides from the same
MS1 signal. The search space is therefore initially very large,
comprising every pairwise combination of the peptides that are
in the database, i.e. (n2+n)/2 crosslinked peptides. This large
search space can be reduced experimentally by separating the
crosslinked peptides during the measurement by help of an
MS-cleavable crosslinker such as disuccinimidyl sulfoxide
(DSSO)8 or any of its alternatives9.
The conceptual advantage of MS-cleavable crosslinkers is
evident. The crosslinker readily cleaves upon activation in the
mass spectrometer, releasing the individual peptides and
thereby enabling the measurement of their individual masses.
In the case of the most popular MS-cleavable crosslinker
DSSO, the crosslinker cleaves preferentially at two different
sites, leading to different crosslinker remnants (also called
stubs) for each peptide (Fig. 1A). The asymmetric cleavage of
this crosslinker produces a pair of alkene (A) and sulfenic acid
(S) stub fragments8. The S stub fragment commonly loses
water forming the unsaturated thiol (T). The two most fre-
quently observed stub peaks per peptide, the A and the T
fragment, form a signature doublet signal with a distinct mass
difference, allowing their detection and subsequent calculation
of peptide masses10.
Knowing the individual peptide masses simplifies the data-
base search, as it reduces the search space to pairwise combi-
nations of peptides with these masses. With the individual
peptides released in the mass spectrometer, one can also de-
sign more intricate data acquisition approaches. The two pep-
tides can be fragmented individually using MS3, which pro-
vides separate fragment information of the two - now linear -
peptides. For this, generally the crosslinked peptide is frag-
mented with a low-energy CID fragmentation first, to prefer-
entially cleave the crosslinker instead of the peptide backbone.
Then signature doublets are selected for MS3. This approach
is routinely employed by studies that use the PIR crosslinker7
and DSSO, while some others using DSSO supplement this
with a complementary ETD MS2 spectrum11.
In an alternative acquisition method, stepped HCD (sHCD),
only a single MS2 spectrum is recorded for each crosslinked
peptide pair. The peptide is subjected to multiple different
collision energies and the fragments are recorded in a single
MS2 spectrum. This spectrum should contain the signature
doublet (from lower fragmentation energies) as well as addi-
tional backbone fragments (from higher fragmentation ener-
gies). These spectra can be searched in most crosslinking
search tools, with optional filtering for spectra containing
cleaved signature peaks during2 or after12 search.
Despite the clean crosslinker cleavage producing dominant
signature peaks in proof-of-concept data of either approach,
there is a lack of statistical data of how often this happens in
general. It is unclear how many crosslinked peptides give rise
to doublets, how prominent these doublets are, and how suc-
cessful doublet selection is at covering the peptides. It is there-
fore unknown how many crosslinked spectra are left unidenti-
fied when relying on these doublets. sHCD compared
Table 1. Overview of analysed datasets.
sample crosslinker acquisition method variable modifications used in re-analysis ref
E. coli lysate DSSO stepped HCD (sHCD) oxidation (M), methylation (D, E), deamidation (N, Q),
BS3/DSSO -OH; -NH2 (K, nterm)
4
BS3
M. musculus
synaptosomes DSSO CID-MS2-MS3+ETD-MS2 DSSO -OH; -NH2 (K, nterm) 15
E. coli 70S ribo-
some DSSO stepped HCD (sHCD) DSSO -OH; -NH2 (K, nterm) 13
CID-MS2-HCD-MS3
favourably to CID methods in the number of crosslinks
identified13, but a methodical analysis comparing the infor-
mation contained in their fragmentation spectra is missing and
yet is crucial for future design of crosslinkers and acquisition
methods.
MS-cleavable crosslinkers have been the tool of choice in
many proteome-wide crosslinking MS studies, and it has been
suggested that large-scale crosslinking MS depends on MS-
cleavable crosslinkers14. While conceptually appealing, these
advantages and potential limitations of MS-cleavable cross-
linkers have yet to be analysed in detail in 'real world' scenari-
os - some comparisons exist, but usually only comparing a few
crosslink spectrum matches (CSMs). We systematically inves-
tigated the influence of the popular MS-cleavable crosslinker
DSSO on the fragmentation of crosslinked peptides. We
achieve this by using crosslinker search software that does not
rely on the cleaved stubs for identification. This allowed us to
clarify how wide-spread the cleavage of DSSO actually is, and
to probe the gain of knowing the individual peptide masses for
identifying crosslinks.
METHODS
Database search and FDR filtering
Mass spectrometry raw data were processed using MScon-
vert23 (v3.0.11729) to convert to mgf-file format. A linear
peptide search was employed to determine median precursor
and fragment mass errors. Peak list files were then re-
calibrated to account for mass shifts during measurement prior
to analysis using xiSEARCH3 1.7.6.1 with the following set-
tings: MS1 error tolerances of 3 ppm; MS2 error tolerance of 5
ppm for the E. coli lysate dataset and 15 ppm for the others; up
to two missing precursor isotope peaks; tryptic digestion spec-
ificity with up to two missed cleavages; modifications: car-
bamidomethylation (Cys, +57.021464 Da) as fixed and oxida-
tion (Met, +15.994915 Da), deamidation (Asn and Gln,
+0.984016 Da), methylation (Glu and Asp, +14.015650 Da),
amidated crosslinker (Lys and protein N-terminus, DSSO-
NH2: +175.03031 Da; BS3-NH2: 155.09463 Da) and hydro-
lysed crosslinker (Lys and protein N-terminus, DSSO-OH:
+176.01433 Da; BS3-OH: +156.07864 Da) as variable modifi-
cations; Maximum number of variable modifications per pep-
tide: 1; losses: –CH3SOH, –H2O, –NH3 and additionally
masses for crosslinker-containing ions were defined account-
ing for its cleavability (A: 54.01056 Da, S: 103.99320 Da, T:
85.98264). Crosslink sites for both reagents were allowed for
side chains of Lys, Tyr, Ser, Thr and the protein N-terminus.
Note that we included a “non-covalent” crosslinker with a
mass of zero to flag spectra potentially arising from gas-phase
associated peptides24. These spectra were removed prior to
false-discovery-rate (FDR) estimation. Results were filtered
prior to FDR to matches having a minimum of three matched
fragments per peptide, a delta score of > 15% of the match
score and a peptide length of at least six amino acids. Addi-
tionally, identifications of peptide sequences that are found in
two or more proteins were removed. FDR was estimated using
xiFDR16 (v2.1.2) on a unique CSM level to 5% grouped by
self- and heteromeric matches.
Data evaluation
CSMs passing FDR were re-annotated with pyXiAnnotator
(https://github.com/Rappsilber-Laboratory/pyXiAnnotator/)
with peptide, b-, and y-type ions using MS2 tolerances as de-
scribed above. The resulting matched fragments were used to
check for the occurrence of DSSO A-T doublets and to calcu-
late fragment sequence coverages. We calculated the sequence
coverage for our CSMs conservatively, as the ratio of matched
N-terminal and C-terminal sequence fragments to the number
of theoretically possible sequence fragments (i.e., 100% se-
quence coverage would mean the detection of at least one
fragment from the N-terminal and one from the C-terminal
series between all amino acid residues of a peptide). To evalu-
ate the MS3 triggering behaviour the MS3 precursor m/z was
extracted from the scan header and compared with the frag-
ment annotation result of the corresponding MS2 CSM. If the
MS3 precursor matched a crosslinked peptide stub fragment
with 20 ppm error tolerance it was counted as correctly trig-
gered.
RESULTS AND DISCUSSION
Prevalence of peptide doublets in fragmentation spectra
of DSSO crosslinked peptides
We analysed three publicly available datasets of DSSO
crosslinking experiments coming from three different labs,
differing in acquisition method and sample complexity (Table
1). The dataset of crosslinked E. coli lysate was acquired using
sHCD with a low, medium, and high normalized collision
energy for each MS24. sHCD is also one of two acquisition
methods used to record a dataset of crosslinked, purified 70S
ribosomes13. In addition to this, Stieger et. al also employed a
CID-MS2-HCD-MS3 approach. For this, first a low-energy
CID-MS2 was acquired. Then MS3 was triggered when dou-
blets of the correct mass difference (32 Da for A-T) were de-
tected (Fig. 1A). Finally, the third dataset called here “Syn-
apse dataset” covered crosslinked mouse synaptosomes and
was acquired with a CID-MS2-MS3+ETD-MS2 approach15.
As in the Ribosome dataset, a low-energy CID-MS2 was ac-
quired for doublet detection. Then, MS3 was acquired as de-
scribed above, supplemented by an additional ETD-MS2 on
the same MS1 precursor.
To assess the prevalence of doublets in the fragmentation
spectra of crosslinked peptides, we re-searched the datasets
using a search algorithm that does not rely on peptide doublets
for crosslink identification. After database search and filtering
to 5% heteromeric (inter protein) CSM-level FDR16, we
looked for signature A and T stub fragment doublet peaks of
the identified peptides and the intensity rank of these doublets
in each spectrum (Fig. 1A).
Even though we did not require doublets to identify cross-
linked peptides, they were very common features in our
CSMs. We found doublets frequently for at least one peptide,
independent of dataset and acquisition method (90 - 98%)
(Fig. 1B). The CID acquisitions displayed a higher proportion
of CSMs with both peptide doublets detected compared to the
sHCD datasets. If one looks at only the common identifica-
tions of CID and HCD to make up for the difference in num-
ber of identifications, the amount of doublets detected for both
peptides increases noticeably for sHCD (71%), making the
difference to CID (81%) less pronounced (Fig. 1B) as does
considering only single stub peaks (Fig. S1).
We next looked at the intensity of the doublet peaks across
these datasets, as this is important for their use during acquisi-
tion and data analysis (Fig. 1C). In the majority of the spectra,
the more abundant doublet is among the most intense peaks,
independent of the fragmentation method used. In fact, a dou-
blet peak is frequently the most abundant peak (34 - 53% of
the doublet containing spectra). Almost all (94 - 98%) doublet-
containing spectra have a peak of the more intense peptide
doublet among the 20 most intense peaks.
Spectra typically displayed in publication figures suggest
that also the less intense doublet is seen prominently in CID
spectra. However, this was only the case for 10% (Ribosome)
or 20% (Synapse) of the doublet containing CID spectra of our
investigated data. Nevertheless, it is seen among the top 20
peaks in 78% (Ribosome) or 91% (Synapse) of the doublet
containing CID spectra. For the sHCD data, the doublet ranks
are lower, yet still approximately 70% of spectra have them
among the 20 most intense peaks (Fig. 1C).
In conclusion, the first doublet is among the most intense
peaks for the majority of CSMs independent of the fragmenta-
tion method. While the second doublet increases confidence in
doublet calling, only one peptide doublet is necessary for de-
riving both peptide masses given that we know the precursor
mass. The visibility of the second peptide doublet is crucial,
however, for the successful selection of both peptides for
MS3. We therefore investigated how successful selecting dou-
blets from CID-MS2 spectra for MS3 was at covering one or
both crosslinked peptides, and if this more complex approach
produces more confident identifications than HCD-MS2.
Speed of HCD outperforms higher sequence coverage of
CID+MS3
The ratio of identified doublets and their intensity ranks are
important criteria for selecting peptides for MS3 fragmenta-
tion. However, absolute numbers of crosslink identifications
may also be influenced by other aspects, such as backbone
fragmentation and acquisition speed. We used the Ribosome
dataset to compare these aspects, as it uses both methods on
the same sample. Here, sHCD leads to 1.4 times more residue
pairs identified than CID-MS317.
When comparing the common CSMs between CID and sHCD,
the overall sequence coverage in sHCD is higher compared to
low-energy CID (Fig. 2A). This comes as no surprise, as low-
energy CID is primarily applied to separate the crosslinked
peptides and not for peptide backbone fragmentation. It is
intentionally combined with MS3 scans and ETD fragmenta-
tion to provide additional sequence information. When we
include the corresponding MS3 scans, the sequence coverage
increases noticeably compared to that of low-energy CID
alone. The overall coverage from combining fragments from
CID and MS3 surpasses the sHCD coverage. Therefore, the
backbone fragmentation does not explain the higher number of
CSMs for sHCD.
Figure 1. Statistics on frequency and intensity of peptide dou-
blet peaks. (a) Illustration of DSSO cleavage and the resulting
signature peptide doublets with the distinct mass difference Δm.
Numbers annotate the intensity rank of the peaks, with the rank of
the more intense of the doublet peaks being the rank of the whole
doublet. (b) Ratio of identified target-target (TT) CSMs that con-
tain one (lighter colour) or both (darker colour) peptide doublets
in each dataset (5% CSM-level FDR). Datasets using sHCD are
shown in orange-red while CID-MS3 based methods are in blue
colours. (c) Percentage of detected doublets passing each intensity
rank cut-off. Shown is the cumulative proportion of CSMs con-
taining doublets. Datasets are coloured as in (b). Synapse (Syn);
Ribosome (Ribo).
Figure 2. Speed of HCD outperforms higher sequence coverage of CID+MS3 in the Ribosome dataset. (a) Sequence coverage of
common CSMs (n=776) identified in both sHCD and CID. Additionally, sequence coverage of CID spectra combined with their respective
MS3 scans is shown. Boxplots depict the median (middle line), upper and lower quartiles (boxes), and 1.5 times the interquartile range
(whiskers). Asterisks indicate significance calculated by a two-sided Wilcoxon signed-rank (p-value > 0.05: n.s., p-value < 0.0001: ****).
(b) Number of acquired MS scans per fragmentation method. Error bars show the 0.95 confidence interval (n=7). (c) Number of triggered
MS3 scans per MS2 scan, for CSMs, linear peptide spectrum matches, and crosslinker modified linear peptide spectrum matches, respec-
tively. (d) Proportion of common CID CSMs having no doublets, only one, or both peptide doublets correctly triggered for MS3.
MS3 acquisition schemes require multiple scan and frag-
mentation events, while sHCD only acquires a single MS2
scan. This difference in complexity and, more importantly,
acquisition speed is reflected in the number of total MS2 scans
acquired, which on average is almost 3 times lower for the
CID-MS3 method, because a lot of acquisition time is spent on
acquiring the additional MS3 scans (Fig. 2B). The drastically
lower sampling of precursors for fragmentation will conse-
quently lead to the reduced detection of crosslinked peptides,
which subsequently results in a lower number of crosslink
identifications. This is exacerbated by many MS3 spectra be-
ing acquired for crosslinker-modified and even for unmodified
linear peptides (Fig. 2C). Despite this excessive MS3 trigger-
ing, for only 41% of the CSMs, MS3 was triggered correctly
on both peptide doublets (Fig. 2D). This is also reflected in the
wider spread of sequence coverage for the worse fragmented
peptide, which is crucial for unambiguous identification of
both linked peptides (Fig. 2A). Note also that for this peptide
the sequence coverage is not significantly increased in
CID+MS3 over sHCD.
In this dataset, the speed of sHCD compensates for its
slightly lower sequence coverage. sHCD also shows a more
symmetric fragmentation of both peptides, as the MS3 ap-
proach is limited by its dependency on triggering on the cor-
rect doublets. Further development of MS3 approaches should
focus on a more sensitive and selective MS3 selection, which
in part is governed by the yield of the crosslinker cleavage.
Peptide doublets for quality control
While some database search algorithms have been built
around peptide masses from doublets, others have been built
without relying on them. Unarguably, peptide masses are use-
ful information. In an attempt to quantify their value, we in-
vestigated the target-decoy CSMs (as representation of the
random matches) for the occurrence of peptide doublets. Be-
cause heteromeric CSMs are the focus of most biological re-
search questions and are also more challenging to identify, we
focused on those for the analysis.
A substantial fraction of random matches have matching
peptide doublets (>47% of heteromeric target-decoy CSMs,
Fig. 3A). However, their extent varies considerably between
the datasets. The highest proportion of doublets among target-
decoy CSMs is found in the Ribosome dataset (88% or 92%
for sHCD and CID, respectively). The E. coli dataset contains
at least one doublet in 66% of the target-decoy CSMs, while
this proportion decreases to 47% for the Synapse dataset. The
amount of identified doublets present in target-decoy matches
seems less dependent on the acquisition method, and more on
the sample and database.
Although heteromeric target-decoy CSMs contain peptide
doublets, they do so less often than the heteromeric target-
target matches (Fig. S2). Based on this difference, we investi-
gated the effect of using this metric as a quality filter. We pre-
filtered the search results to those spectra that contain at least
one peptide with a detected doublet and then re-estimated 5%
CSM-level FDR. The gains using this approach are very much
dependent on the complexity of the dataset (Fig. 3B). Unsur-
prisingly, the Synapse dataset, which had the least target-
decoys containing a matching doublet, shows the largest gains
using this approach (19%). However, the E. coli dataset only
gains 5% in heteromeric CSMs, even though there is a large
difference in the proportion of peptide doublets between target
and false matches (97% vs. 66%; Fig. S2, 3A). This led us to
investigate the score distribution of doublet containing match-
es in more detail (Fig. 3C).
The vast majority of high-scoring target-target CSMs con-
tain at least one doublet and are therefore not removed, while
targets without a matched peptide doublet tend to have lower
scores. In this lower scoring region, there is a steep increase in
target-decoy matches, which is only slightly reduced by pre-
filtering for a doublet. The effect becomes more apparent
when looking at the FDR at different score thresholds. While
the increase in error is not as steep for the filtered matches as
for the unfiltered, it still grows exponentially (Fig. 3D). This
holds true also for the Ribosome datasets and to a lesser extent
for the Synapse dataset (Fig. S3-5).
Figure 3. Peptide doublets as a quality control metric for heteromeric identifications. (a) Percentage of heteromeric target-decoy
CSMs that contain one or two peptide doublets across datasets at 5% CSM-level FDR. (b) Proportion of heteromeric CSMs at 5% CSM-
level FDR when filtering spectra to contain a peptide doublet compared to unfiltered data. (c) Score distribution of heteromeric matches in
the E. coli dataset. Shown is the distribution of targets and target-decoy matches with and without filtering for peptide doublets. Dashed
lines show the resulting score cutoffs at 5% FDR. (d) FDR (interpolated values for visualization) of unfiltered and peptide doublet filtered
E. coli data. Synapse (Syn); Ribosome (Ribo).
The moderate gains of using doublets for post-search filter-
ing also suggests that using them during search will offer only
moderate gains. Presumably, spectra of high quality, which
contain doublets, also tend to contain sufficient peptide frag-
ment peaks so that identification is possible without relying on
peptide mass information.
Comparison of a cleavable to a non-cleavable crosslinker
Non-cleavable crosslinkers are widely believed to be un-
suitable for complex samples14,18,19. This bases on the assump-
tion that not knowing the individual peptide masses before the
search results in the need for exhaustive combination of all
peptides in the database and thus an explosion of the search
space. However, there are multiple large-scale studies that
have successfully employed a non-cleavable crosslinker de-
spite these assumptions4,12,20,21. These are based on a detailed
understanding of how crosslinked peptides fragment22, that
offered a computational solution to knowing the individual
peptide masses which was then implemented in the search
algorithm xiSEARCH3. In light of successful usages of both
types of crosslinkers, we decided to compare their spectral
information to understand any costs and benefits. In addition
to DSSO, the published E. coli dataset also contains data from
the non-cleavable crosslinker BS3. As the data for both cross-
linkers were prepared and acquired in a very comparable man-
ner, this dataset offers an opportunity to directly compare the
effects of BS3 to DSSO on a complex mixture analysis. Im-
portantly, because of its size and the high number of CSMs
identified, the dataset is well suited for statistical evaluation.
A manual side-by-side comparison of CSMs identified in
both datasets suggests DSSO to have richer spectra with more
fragments. Especially, fragments containing the crosslinking
site appear to be more present, mostly as fragments containing
an A/S/T stub of DSSO (Fig. 4A). We then performed a statis-
tical evaluation of this observation over common CSMs of the
two crosslinkers (Fig. 4B). This confirmed that DSSO led
indeed to a significantly higher sequence coverage than BS3.
While the coverage of linear fragments is very similar between
the two crosslinkers, the coverage of link site-containing
fragments is significantly higher for DSSO. Link site-
containing fragments contain the full second peptide (+P) or,
additionally for cleavable crosslinkers, just a cleaved cross-
linker stub. Indeed, A/S/T stub fragments are the major source
of link site-containing fragments for DSSO, while +P cover-
age is lower than that of BS3. This means that the increased
sequence coverage for DSSO stems exclusively from cleaved
crosslinker fragments. Crosslinker cleavage appears to pro-
mote the cleavage of peptide backbone sites and/or their detec-
tion.
The better sequence coverage of DSSO-linked peptides im-
proves the separation of true from false CSMs (Fig. 4C). For
heteromeric matches, DSSO has a larger area under the curve,
and especially more high scoring targets, effectively leading to
an increase in heteromeric CSMs. While for BS3 3308 heter-
omeric CSMs were identified, the DSSO dataset resulted in
more than twice as many (7316, +121%) (Fig. 4D). For self-
CSMs, only 29% more CSMs were identified with DSSO than
with BS3 (Fig. S6), indicating that self-CSMs are approaching
exhaustive coverage at the given experimental detection limit.
Similar results were seen when including retention time data
of heteromeric and self-CSMs20.
To investigate the effect of the cleaved crosslinker frag-
ments on the overall crosslink search performance, we per-
formed another search in which the DSSO crosslinker was
treated as non-cleavable. In this search, only 1866 heteromeric
CSMs were identified (-74%). Filtering these results for dou-
blet containing results, as described before, increased identifi-
cations to 3064. This is, however, still a loss of 58% of CSMs
compared to the search considering DSSO as cleavable. Col-
lectively, these observations demonstrate that A/S/T stub
fragments play a central role in the success of DSSO for cross-
linking mass spectrometry, especially for more complex sam-
ples.
CONCLUSIONS
Our work finds a surprisingly limited value of doublet in-
formation stemming from crosslinker cleavage for the identifi-
cation of crosslinks. Nonetheless, we find cleavable crosslink-
ers to lead to the identification of substantially more heter-
omeric CSMs. We pinpoint improved sequence coverage as
the major contributor to this. This has implications for how to
conduct crosslinking studies and the future development of the
methodology. Firstly, as many suspected but possibly not for
the right reasons, cleavable crosslinkers are preferable for
crosslink mixture analyses.
Figure 4. Comparison of non-cleavable crosslinker BS3 to the MS-cleavable crosslinker DSSO. (a) Example MS2 spectrum of a high
scoring CSM identified in both datasets. Upper panel shows the CSM from the BS3 dataset. Lower panel shows the same peptide m/z-
species identified in the DSSO dataset. Unique fragments are highlighted in bold. (b) Sequence coverage of all, linear and link site-
containing fragments (CSMs: n=1437). For DSSO, link site-containing fragments are additionally separated into fragments containing the
full second peptide (+P) or only the cleaved crosslinker stub (A/S/T). Boxplots depict the median (middle line), upper and lower quartiles
(boxes), and 1.5 times the interquartile range (whiskers). Asterisks indicate significance calculated by a two-sided Wilcoxon signed-rank
(p-value < 0.0001: ****). (c) Target-target and target-decoy score distributions of heteromeric CSMs for BS3 and DSSO. Scores were
normalized to their respective score cut-off at 10% FDR. (d) Number of heteromeric CSMs passing 5% CSM-level FDR for BS3 and
DSSO. As a control, DSSO was additionally searched as a non-cleavable crosslinker and also filtered for the presence of peptide doublets.
Secondly, sHCD is the recommended acquisition method as it
achieves almost the same sequence coverage as CID-MS3, but
is much faster. CID-MS3 currently lacks speed, specificity and
sensitivity. Consequently, future developments of crosslinkers
and acquisition methods should focus primarily on sequence
information, without compromising acquisition speed. Current
choices governing acquisition schemes rely on experimental
comparisons, to which we add a methodological understanding
of the key parameters that govern crosslink identification.
With this, we hope to pave the way for simplified, cost-
effective, and standardised workflows that a wider number of
labs can use.
ASSOCIATED CONTENT
AUTHOR INFORMATION
Corresponding Author
Juri Rappsilber. Email: juri.rappsilber@tu-berlin.de
Author Contributions
S.L., L.K., L.S., F.J.O., and J.R. designed the experiments. S.L.,
L.K., and L.F. processed Crosslinking MS data; S.L., L.K., F.J.O.,
and J.R. prepared figures and wrote the manuscript with input
from all authors.
ACKNOWLEDGMENT
The work was funded by the Deutsche Forschungsgemeinschaft
(DFG, German Research Foundation) under Germany's Excel-
lence Strategy – EXC 2008 – 390540038 – UniSysCat and grant
no. 426290502 and by the Wellcome Trust through a Senior Re-
search Fellowship to JR (103139). The Wellcome Centre for Cell
Biology is supported by core funding from the Wellcome Trust
(203149).
REFERENCES
1. Chen, Z.-L. et al. A high-speed search engine pLink 2 with
systematic evaluation for proteome-scale identification of cross-
linked peptides. Nat. Commun. 10, 3404 (2019).
2. Götze, M., Iacobucci, C., Ihling, C. H. & Sinz, A. A Simple
Cross-Linking/Mass Spectrometry Workflow for Studying System-
wide Protein Interactions. Anal. Chem. 91, 10236–10244 (2019).
3. Mendes, M. L. et al. An integrated workflow for crosslink-
ing mass spectrometry. Mol. Syst. Biol. 15, e8994 (2019).
4. Lenz, S. et al. Reliable identification of protein-protein in-
teractions by crosslinking mass spectrometry. Nat. Commun. 12, 3564
(2021).
5. Belsom, A. & Rappsilber, J. Anatomy of a crosslinker.
Curr. Opin. Chem. Biol. 60, 39–46 (2021).
6. Steigenberger, B., Pieters, R. J., Heck, A. J. R. & Schel-
tema, R. A. PhoX: An IMAC-Enrichable Cross-Linking Reagent.
ACS Cent Sci 5, 1514–1522 (2019).
7. Zhang, H. et al. Identification of protein-protein interac-
tions and topologies in living cells with chemical cross-linking and
mass spectrometry. Mol. Cell. Proteomics 8, 409–420 (2009).
8. Randall, A., Baldi, P., Rychnovsky, S. D. & Huang, L. De-
velopment of a novel cross-linking strategy for fast and accurate iden-
tification of cross-linked peptides of protein complexes. Molecular &
Cellular (2011).
9. Matzinger, M. & Mechtler, K. Cleavable Cross-Linkers and
Mass Spectrometry for the Ultimate Task of Profiling Protein-Protein
Interaction Networks in Vivo. J. Proteome Res. 20, 78–93 (2021).
10. Sinz, A. Divide and conquer: cleavable cross-linkers to
study protein conformation and protein-protein interactions. Anal.
Bioanal. Chem. 409, 33–44 (2017).
11. Liu, F., Rijkers, D. T. S., Post, H. & Heck, A. J. R. Prote-
ome-wide profiling of protein assemblies by cross-linking mass spec-
trometry. Nat. Methods 12, 1179–1184 (2015).
12. O’Reilly, F. J. et al. In-cell architecture of an actively tran-
scribing-translating expressome. Science 369, 554–557 (2020).
13. Stieger, C. E., Doppler, P. & Mechtler, K. Optimized
Fragmentation Improves the Identification of Peptides Cross-Linked
by MS-Cleavable Reagents. J. Proteome Res. 18, 1363–1370 (2019).
14. Piersimoni, L. & Sinz, A. Cross-linking/mass spectrometry
at the crossroads. Anal. Bioanal. Chem. 412, 5981–5987 (2020).
15. Gonzalez-Lozano, M. A. et al. Stitching the synapse:
Cross-linking mass spectrometry into resolving synaptic protein inter-
actions. Sci Adv 6, eaax5783 (2020).
16. Fischer, L. & Rappsilber, J. Quirks of Error Estimation in
Cross-Linking/Mass Spectrometry. Anal. Chem. 89, 3829–3833
(2017).
17. Stieger, C. E., Doppler, P. & Mechtler, K. Optimized
Fragmentation Improves the Identification of Peptides Cross-Linked
by MS-Cleavable Reagents. J. Proteome Res. 18, 1363–1370 (2019).
18. Steigenberger, B., Albanese, P., Heck, A. J. R. & Schel-
tema, R. A. To Cleave or Not To Cleave in XL-MS? J. Am. Soc.
Mass Spectrom. 31, 196–206 (2020).
19. Yu, C. & Huang, L. Cross-Linking Mass Spectrometry: An
Emerging Technology for Interactomics and Structural Biology. Anal.
Chem. 90, 144–165 (2018).
20. Giese, S. H., Sinn, L. R., Wegner, F. & Rappsilber, J. Re-
tention time prediction using neural networks increases identifications
in crosslinking mass spectrometry. Nat. Commun. 12, 3237 (2021).
21. Linden, A. et al. A cross-linking mass spectrometry ap-
proach defines protein interactions in yeast mitochondria. Mol. Cell.
Proteomics (2020) doi:10.1074/mcp.RA120.002028.
22. Giese, S. H., Fischer, L. & Rappsilber, J. A Study into the
Collision-induced Dissociation (CID) Behavior of Cross-Linked Pep-
tides. Mol. Cell. Proteomics 15, 1094–1104 (2016).
23. Holman, J. D., Tabb, D. L. & Mallick, P. Employing Prote-
oWizard to Convert Raw Mass Spectrometry Data. Curr. Protoc.
Bioinformatics 46, 13.24.1–9 (2014).
24. Giese, S. H., Belsom, A., Sinn, L., Fischer, L. & Rappsil-
ber, J. Noncovalently Associated Peptides Observed during Liquid
Chromatography-Mass Spectrometry and Their Effect on Cross-Link
Analyses. Anal. Chem. 91, 2678–2685 (2019).
ARTICLE
Reliable identification of protein-protein
interactions by crosslinking mass spectrometry
Swantje Lenz 1,3, Ludwig R. Sinn 1,3, Francis J. O’Reilly 1,3, Lutz Fischer1, Fritz Wegner1&
Juri Rappsilber 1,2✉
Protein-protein interactions govern most cellular pathways and processes, and multiple
technologies have emerged to systematically map them. Assessing the error of interaction
networks has been a challenge. Crosslinking mass spectrometry is currently widening its
scope from structural analyses of purified multi-protein complexes towards systems-wide
analyses of protein-protein interactions (PPIs). Using a carefully controlled large-scale ana-
lysis of Escherichia coli cell lysate, we demonstrate that false-discovery rates (FDR) for PPIs
identified by crosslinking mass spectrometry can be reliably estimated. We present an
interaction network comprising 590 PPIs at 1% decoy-based PPI-FDR. The structural infor-
mation included in this network localises the binding site of the hitherto uncharacterised
protein YacL to near the DNA exit tunnel on the RNA polymerase.
https://doi.org/10.1038/s41467-021-23666-z OPEN
1Bioanalytics, Institute of Biotechnology, Technische Universität Berlin, Berlin, Germany. 2Wellcome Centre for Cell Biology, University of Edinburgh,
Edinburgh, UK.
3
These authors contributed equally: Swantje Lenz, Ludwig R. Sinn, Francis J. O’Reilly. ✉email: Juri.Rappsil[email protected]
NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications 1
1234567890():,;
Crosslinking mass spectrometry (Crosslinking MS) has
become a key technology for understanding the archi-
tecture of multi-protein complexes by providing distance
restraints between protein residues1. These studies are typically
performed on purified complexes, but in recent years pioneering
studies have used Crosslinking MS to study the topology of PPIs
in more complex systems, such as cell lysates, organelles or whole
cells2–13. Crosslinking MS is therefore emerging as a technique
for mapping PPIs alongside existing tools, such as two-hybrid
screens, affinity purification, proximity labelling techniques and
co-fractionation studies. Importantly, Crosslinking MS studies do
not require tagging of proteins and can fixate interactions inside
cells prior to cell disruption. Crosslinking MS can therefore detect
otherwise difficult to observe PPIs, including weak or transient
interactions and interactions involving proteins that are not easily
solubilised. Unlike other large-scale PPI mapping technologies,
the interactions are detected between individual residues and
therefore also provide information on protein complex topology.
As for any technology for mapping PPIs, the reported inter-
actions must be reliable to be useful. Large numbers of spurious
PPIs are avoided by correctly estimating FDRs and then trimming
the list of reported PPIs to the desired error rate. The standard
method for error estimation in classical LC-MS-based proteomics
is the target-decoy approach, where a decoy database of spurious
peptide sequences is included to model random identifications.
This approach assumes that the rate of matches to the decoy
database is an estimator of false positives (type I error rate). This
target-decoy approach has been adapted for Crosslinking
MS14–18. Recently, however, concerns have emerged regarding
current FDR methods12,19,20 and the need for improvements is
recognised widely across the Crosslinking MS field21.
Matches in Crosslinking MS are different from those in classical
proteomics because two peptides are combined to make one
match. This leaves two potential opportunities for a false match,
which requires additional considerations when applying the
target-decoy approach, such as a crosslink-specific equation for
calculating FDR15,16. Two additional considerations have been
suggested for correctly estimating errors in crosslinking-based PPI
screens. The first, whether to consider crosslinks between peptides
within one protein sequence (self-links, including homomeric
crosslinks) separately from crosslinks between distinct protein
sequences (heteromeric crosslink)4,15. The second, how to handle
propagation of error between the different levels of information,
i.e. from crosslinked spectrum matches (CSMs), to peptide pairs,
to residue pairs and finally to PPIs16. However, both considera-
tions have not been systematically tested and therefore they have
remained controversial with no consensus emerging for if and
how they should be implemented (Supplementary Table 1).
In this work, we tested different approaches for FDR estima-
tion and demonstrated how incorrect handling of the error esti-
mation can have huge effects on the reliability of the reported
PPIs. For this, we designed a carefully controlled large-scale
crosslinking study of the model organism E. coli by fractionating
lysate via size exclusion chromatography (SEC), crosslinking
within the individual fractions, and then pooling all fractions.
Proteins that did not share the same SEC fraction could not be
crosslinked and therefore reveal false PPIs, without needing to
rely on decoys. We used this sample to demonstrate that self-links
and heteromeric crosslinks must always be separated for FDR and
that data must be merged into PPIs before correct estimation of
error in crosslinking-based PPI investigations.
Results
Theoretical considerations on FDR estimation in crosslinking
MS. Naively, FDR is estimated based on a score distribution of
CSMs to the target and decoy databases, using the decoy matches
as a model of random and hence false target matches. However,
the size of the search space, and therefore the chance of random
matching, is inherently different for heteromeric crosslinks and
self-links (Fig. 1a). In our database of 4350 proteins, the chance of
matching a decoy crosslink (random) within the heteromeric
crosslinks is 10.6 times higher than within the self-links (Fig. 1b).
Controlling FDR in the total set of CSMs, and then selecting only
heteromeric matches thus enriches for false positives. This leads
to a large underestimation of the error within heteromeric CSMs,
which describe PPIs (Supplementary Fig. 1). Consequently, het-
eromeric crosslinks must be considered separately from self-links
during FDR estimation.
A second consideration is that a naïve FDR for CSMs may not
reflect the error among reported PPIs. When merging data from
CSMs to PPIs false and true matches may behave differently and
thus the relative error will change. CSMs merge into peptide
pairs, peptide pairs into residue pairs and residue pairs into PPIs
(see Methods). False PPIs are the result of random CSMs and
thus less likely than true PPIs to be supported by multiple CSMs.
Multiple true CSMs are therefore much more likely to merge into
a single PPI. This leads to a change in ratio between true and false
matches as one merges CSMs into PPIs. Consequently, CSMs
must be merged into PPIs before FDR estimation of PPIs
(Fig. 1c).
These considerations apply universally, as they are indepen-
dent of crosslinker chemistry and data analysis workflow.
Construction of a test system to investigate methods of FDR
estimation in crosslinking MS. To test different approaches for
FDR estimation we produced a sample for which we experi-
mentally know a large number of the potential false PPIs. We
prepared simplified cellular fractions enriched in protein com-
plexes by separating E. coli lysate by size exclusion chromato-
graphy (Fig. 1d). The resulting 44 fractions span the molecular
weight range from ~3 MDa to 150 kDa. A portion of each fraction
was analysed by label‐free quantitative proteomics to generate
elution profiles of each protein across all 44 fractions. We iden-
tified 1926 E. coli proteins in these fractions combined. Conse-
quently, the complexity of our sample approximates that of whole
E. coli cells22. The abundance of the detected proteins spans six
orders of magnitude, producing a challenging sample for
detecting crosslinks.
The remainder of each fraction was split equally and
crosslinked with BS3 or DSSO, respectively. For each crosslinker,
the crosslinked fractions were then pooled and digested. The
crosslinked peptides were first enriched by strong cation exchange
chromatography (SCX) to enrich crosslinked peptides in nine
high-salt fractions. Each high-salt SCX fraction was subsequently
fractionated in a second chromatographic dimension by hydro-
philic strong anion exchange chromatography (hSAX) into ten
fractions. Following this extensive fractionation, the crosslinked
peptides were acquired by LC-MS (2 × 90 fractions, 32.5 days of
mass-spectrometric acquisition per crosslinker) to generate a
substantial dataset for testing FDR methods.
Proteins eluting in the same size exclusion fraction may be
crosslinked in this analysis. In contrast, proteins that were not in
the same fraction cannot be crosslinked together, i.e. are ‘non-
crosslinkable’pairs (either because they were not identified at all
or below an abundance threshold (Supplementary Fig. 1). If such
a non-crosslinkable protein pair is identified during data analysis,
it is a false match. This experimental assessment of PPI error is
independent of the target-decoy approaches and therefore offers
an opportunity to benchmark target-decoy-based PPI-FDR
methods.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z
2NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications
Impact of CSM-FDR estimation on the reliability of identified
PPIs.Wefirst searched against a database comprising all (4350) E.
coli proteins, including those not detected in our sample. Crosslinks
of protein pairs defined as non-crosslinkable above were defined as
false. At a naïve 5% decoy-based CSM-level FDR (not distin-
guishing self and heteromeric crosslinks), we identified 20,833 (5655
heteromeric) unique CSMs for BS3 and 22,296 (6923 heteromeric)
unique CSMs for DSSO. We chose 5% FDR to have sufficient false
identifications for precise FDR estimation at all information levels.
In close agreement, our experimental control revealed that 4% of
these CSMs are false (Fig. 2a, b). Note that CSMs in this manuscript
refer to unique CSMs; using redundant CSMs will produce spurious
FDR estimations (Supplementary Fig. 2).
However, naïve CSM-level FDR leads to many false PPIs, as
our experimental control reveals. For this, the heteromeric CSMs
of naïve 5% decoy-based CSM-FDR were merged into PPIs.
Counting our non-crosslinkable PPIs then revealed that 36% of
the reported PPIs were false in this DSSO dataset (Fig. 2a). In the
BS3 dataset the results were very similar with naïve 5% decoy-
based CSM-FDR leading to 35% false PPIs (Fig. 2b).
Given this large deviation between naïve decoy-based CSM-
FDR and experimentally determined PPI error we sought further
controls at the level of data analysis. As additional independent
controls of decoy-based FDR we therefore used three entrapment
database searches. First, we searched our spectra against E. coli
sequences supplemented with the same number of human protein
sequences. Second, we added the full S. cerevisiae proteome to the
E. coli sequences and, finally, both databases were combined into
an even larger entrapment database. Here we know any identified
PPI that includes a human or yeast protein is false. According to
these entrapment controls, at a naïve 5% decoy-based CSM-FDR,
the PPI error reached an average of 45% (Supplementary Fig. 2).
Although this corroborated the notion that naïve decoy-based
CSM-FDR leads to a gross underestimation of the PPI error we
devised two additional controls of decoy-based FDRs. For one, we
performed searches using a fictional (wrong mass) crosslinker in
addition to BS3 or DSSO, respectively. Any CSM involving this
fictional crosslinker is a known false positive. While one of the
two matched peptides in such a CSM might be correct, the other
must be false to compensate for the false crosslinker mass when
making up to the precursor mass. As a last control we searched
previously high-confidence matched scans with shifted precursor
masses to generate a set of false crosslinked peptides. One of the
peptides constituting the original precursor could still be matched
correctly. However, as the precursor mass was shifted, again, the
second peptide cannot be matched correctly. So, any peptide pair
match to these spectra constitutes a false positive. These controls
reported a PPI error of 50 and 60%, respectively, at naïve 5%
decoy-based CSM-FDR (Supplementary Fig. 2). In summary, not
only the experimental but also the three entrapment and the two
wrong mass controls revealed naïve decoy-based CSM-FDR to be
inadequate for estimating PPI error.
Protein A Protein B Protein C
Protein AProtein C Protein B
123 11
322
1
23 11
2 23
Residue pairs
10% error
PPIs
25% error
Negative control
Positive control
FDR method
testing
High molecular
weight E. coli lysate
crosslinked in
44 fractions
Protein coelution analysis
Crosslinked PPI
identification
m/z
m/z
I
m/z
d
c
b
a
self hetero-
0%
20%
40%
60%
80%
Fraction of decoys
meric
6%
67%
Fig. 1 Considerations for crosslinked PPI-FDR and experimental workflow. a For matches within the same protein sequence (with a non-directional
crosslinker17), a crosslink from A1 to A2 is indistinguishable from A2 to A1 (theoretically possible search space shown as purple triangles). In contrast,
heteromeric matches are not symmetrical, and therefore occupy a larger random space (green squares). bFraction of decoys in random picks of 100 self
and 100 heteromeric CSMs from the search output before any FDR filtering (random picks, n=20, i.e. ten per crosslinker dataset). Error bars show
standard deviation from the mean. Source data are provided as a Source Data file. cSchematic showing error increase when merging crosslinked residue
pairs to PPIs. Proteins are indicated as circles; blue and red lines represent true and false linkages, respectively. dExperimental workflow. E. coli lysate was
separated and crosslinked in individual high molecular weight fractions, pooled again to simulate a complex mixture and analyzed by mass spectrometry.
Quantitative proteomics of uncrosslinked fractions provided protein coelution data.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z ARTICLE
NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications 3
Of note, in our experimental control 87% of false PPIs involved
proteins that were seen only with heteromeric crosslinks, i.e. that
lacked self-links (Fig. 2c). In the entrapment control this number
increased to 100% (Fig. 2c). If observed at all, heteromeric-only
proteins had a lower median abundance than all proteins in the
sample suggesting that they are enriched in random matches
(Fig. 2d). In contrast, the median abundance of proteins detected
with both, self and heteromeric crosslinks, was 14.8-fold higher
than the median of all identified proteins (significantly higher
abundance than all identified proteins, p< 0.0001 using a one-
sided Kolmogorov–Smirnov test) (Fig. 2d). The proportion of
PPIs involving heteromeric-only proteins may thus be an
indicator of reliability when evaluating published Crosslinking
MS data.
Comparative analysis of different methods of FDR estimation
in crosslinking MS. To address this inflated error of the naïve
decoy-based CSM-FDR we returned to our initial theoretical
considerations. Indeed, assessing heteromeric matches separately
from self-matches decreased false PPIs substantially (35 to 16%
and 36 to 15%, for BS3 and DSSO, respectively) (Fig. 2e). How-
ever, the error remained three times higher than the targeted 5%.
We therefore also considered error propagation between infor-
mation levels. As predicted, CSMs rarely corroborated each other
in false PPIs while plausible PPIs were supported by multiple
CSMs (1.2 versus 4.6 for BS3, 1.3 versus 5.2 for DSSO), irre-
spective of the crosslinker (Fig. 2a, b). This effect was most
pronounced when merging unique residue pairs into PPIs. Error
control at lower information levels therefore leads to large pro-
portions of reported PPIs being false (Fig. 2e). This also holds
true for all other reporting levels (i.e. CSMs, peptide pairs and
residue pairs) (Supplementary Fig. 3).
In contrast, first merging CSMs for each PPI and then assessing
the FDR gave more reliable results: 6.6% and 4.9% false PPIs
when applying 5% decoy-based PPI-FDR (Fig. 2e) for BS3 and
DSSO, respectively. This is also supported by the other controls,
which indicated an actual error close to 5% (4.8% for BS3 and
4.9% for DSSO) when applying 5% decoy-based PPI-FDR
(Supplementary Fig. 2).
As a positive control, we evaluated the proportion of PPIs that
were supported by correlation of protein coelution profiles (Fig. 2f,
Supplementary Fig. 4). The fraction of supported PPIs was highest
when using heteromeric PPI-FDR and the proportion decreased
when raising the FDR threshold, as expected. The same trends are
true for the alternative positive control of using interaction
evidence from the STRING database (Supplementary Fig. 4).
High-quality PPIs in E. coli lysate. An FDR threshold should
be chosen to meet the stringency required by the study (Sup-
plementary Fig. 5). At a heteromeric decoy-based PPI-FDR of 5%
0%
10%
20%
30%
Estimated PPI error
BS3
DSSO
CSM Pep.
Pair
CSM Pep.
Pair
Res.
Pair
PPI
FDR method at 5%
heteromeric methodsnaïve
5%
CSMs PPIs
Res.
Pairs
Pep.
Pairs
0
2000
4000
CSMs
20833
35%4%
12%
15% 17% 1.2 fold
less
false
4.6 fold
less
plausible
all
matches
heteromeric matches only
n
BS3
f
1% 5% 10%
FDR threshold
0%
20%
40%
60%
80%
Coeluting PPIs
(H) heteromeric
(N) naïve
CSM (N)
Peptide Pair (N)
Peptide Pair (H)
Residue Pair (H)
CSM (H)
PPI (H)
bca
d
9
8
7
6
5
4
log(average iBAQ)
self and
heteromeric
heteromeric
only
self
only
median abundance
all identifications
525365584725933013
DSSO
BS3
e
CSMs PPIs
Res.
Pairs
Pep.
Pairs
0
2000
4000
36%
CSMs
22296
4%
12%
15% 18%
1.3 fold
less
false
5.2 fold
less
plausible
6000
all
matches
heteromeric matches only
DSSO
n
0%
20%
40%
60%
80%
100%
Proportion of false proteins
non-crosslinkable
human proteins
0
htiwciremoreteh
self only
n =
Fig. 2 Comparative analysis of different methods of FDR estimation in crosslinking MS. a,bFalse identifications as a function of merging heteromeric
CSMs passing a naïve decoy-based CSM-FDR of 5% for aDSSO and bBS3, respectively. When merging crosslink data from CSMs to PPIs, the number of
identifications decreases and the fraction of false identifications increases. CSMs rarely corroborated each other in false PPIs while plausible PPIs were
supported by multiple CSMs. Heteromeric CSMs are indicted by circles connected with a straight line; self-CSMs by a circle with curved line. cProportion
of proteins involved in false PPIs with self-links or with only heteromeric crosslinks, of non-crosslinkable E. coli proteins and the entrapment database.
dProteins found exclusively in heteromeric PPIs had a lower abundance than all identified proteins and thus a low chance to be detected. Boxplots depict
the median (middle line), upper and lower quartiles (boxes), 1.5 times of the interquartile range (whiskers) as well as outliers (single points). ePPI error
resulting from a 5% FDR threshold of FDR approaches performed in other studies (Supplementary Table 1). Each bar is from a separate FDR calculation.
Diamond denotes the method leading to the PPI error closest to 5%. fFraction of protein pairs with similar elution profiles (correlation coefficient > 0.5)
among the PPIs passing a given FDR threshold, applying different published FDR methods (Supplementary Table 1). Averages of BS3 and DSSO
data are shown (Also presented sepearately in Supplementary Fig. 4). Source data for panels a,band dare provided as a Source Data file.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z
4NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications
applied on our data, 756 PPIs are reported, with 38 expected to be
false. To focus on a high-quality subset of PPIs in the E. coli
lysate, we applied a 1% heteromeric PPI-FDR cut-off, yielding 590
PPIs involving 308 proteins (Fig. 3a, Supplementary Data 1),
connected with a total of 2539 residues pairs.
Three hundred sixty-six (62%) of these PPIs are connected by
more than one residue pair (Supplementary Fig. 5). Eleven
percent of the proteins found in PPIs had no self-links, but most
had abundances higher than the sample median. These proteins
tend to be small and thus produce few peptides, so can be difficult
to observe by mass spectrometry (Supplementary Fig. 5). We
found 63% (370) of the detected PPIs in the STRING database
(Fig. 3b). Ninety eight percent (576) were found to be eluting in a
fraction together and 68% had similar elution profiles (correlation
coefficient > 0.5), suggesting that they form stable complexes
(Supplementary Data 2). Ribosomal proteins had a complex
elution pattern, presumably due to the presence of assembly
intermediates, although many of the proteins that were found
crosslinked to the ribosome are known interactors (26 of 53).
The crosslink-based PPI network included 289 protein pairs
with highly similar coelution (correlation coefficient > 0.8). The
majority of these were known interactions including complexes
like ATP synthase, pyruvate dehydrogenase, MukBEF or DNA
gyrase. The data confirmed binding of acyl carrier protein to
MukBEF, and of YacG to DNA gyrase (Supplementary Fig. 6 and
7). In addition, 130 PPIs with highly similar coelution were not
yet experimentally confirmed for E. coli K12, though 55 of these
had a STRING entry based on other evidence. Novel interactions
included those between the small ribosomal regulators ElaB,
YgaM and YqjD, the periplasmic endoproteases DegP and DegQ,
the ubiquinone biosynthesis accessory factors UbiK and UbiJ, as
well as GroEL and potential substrates (Supplementary Fig. 8).
RNA polymerase (RNAP) crosslinked to 23 proteins (Fig. 4a).
Previous interaction evidence was available for 20 of these,
including the transcription factors RpoD and GreB, and the
transcriptional regulators NusG, NusA and RapA; all crosslinks
are in agreement with previously suggested binding sites
(Supplementary Fig. 9). YacL, a protein of unknown function
that was found to be associated with RNAP in pull-down
experiments23, crosslinked to the beta and beta’subunit of RNAP
(four residue pairs), as well as to NusG (two residue pairs). It also
coeluted with RNAP (correlation of 0.988) with an abundance
comparable to NusG (Fig. 4b).
To confirm the interaction, we performed pulldowns using
K12 strains with endogenously tagged ORFs of YacL, NusG and
RpoB to carry an affinity-tag. YacL affinity-enriched RNAP and
NusG (Fig. 4c) and, conversely, NusG and RpoB enriched YacL,
thus confirming the association of YacL with NusG and RNAP
(Supplementary Fig. 10). To further constrain the binding site of
YacL on the RNAP and confirm the interaction site by use of a
different crosslinker we crosslinked these affinity-enriched
complexes using the photoactivatable crosslinker sulfo-SDA,
which can provide a higher density of crosslinks than DSSO or
BS324. Sulfo-SDA crosslinking either of the three affinity-
enriched proteins confirmed the direct binding of YacL to the
RNAP and NusG with a combined 14 unique residue pairs
(Supplementary Fig. 10 and Supplementary Data 3-5). The total
of 20 residue pairs from DSSO, BS3 and sulfo-SDA thus constrain
the binding site of YacL to RNAP and NusG. Sixteen of these
residue pairs were between our I-TASSER model of YacL and
regions of RNAP-NusG included in the solved structure (PDB
6C6U [https://doi.org/10.2210/pdb6c6u/pdb]). These were used
in DisVis to calculate the accessible interaction space and localize
YacL on RNAP next to NusG at the DNA exit site (Fig. 4d).
Discussion
There have been several recent advances in enrichment and
detection of crosslinked peptides by Crosslinking MS that suggest
it will soon be able to map large portions of the cellular inter-
actome in a single experiment3,13,25. This will open the door to
detecting changes in these interactomes in different cellular states
by quantification of the abundances of the detected crosslinks2,26.
All of these advances require correctly controlled FDR to produce
results that can be relied upon.
In previous studies, the quality of identified crosslinks was
assessed by measuring inter-residue distances in known protein
structures. However, for proteome-wide crosslinking studies, this
approach is inherently biased towards true interactions as they
are likely to be enriched in known complexes27. The majority of
random PPIs are neglected by this FDR evaluation method,
making this approach completely inadequate for reliable PPI
error estimation.
In this work, we experimentally demonstrated that Cross-
linking MS can reliably identify PPIs using the target-decoy
approach as a quantitative error metric. Decoys are only a
model of false positives with a number of underlying
abCrosslink
PPI network
Correlating
PPIs
PPIs in
STRING
All PPIs
in network
24%
12%
51%
Random pairs of
identified proteins
22%
3%
7%
All
randomized
PPIs
30S Ribosome
ATP
synthase
elaB-yqjD-ygaM
MukBEF
degP-degQ
Pyruvate
dehydrogenase
DNA gyrase
GroEL
ubiJ-ubiK
RNA
polymerase
50S Ribosome
yacL
nusG
1
2
3+
Res. Pairs
Fig. 3 Heteromeric PPI-FDR leads to high fidelity PPI network in E. coli lysate. a Crosslinking MS-derived PPI network of soluble high molecular weight E.
coli proteome. Selected proteins (circles) and protein complexes are highlighted. The proteins AceA and TnaA were removed for clarity. bCharacterization
of the obtained PPI network in comparison to random PPIs from proteins identified in coelution data. Shown are the overlaps with STRING database and
coelution data (correlation coefficient > 0.5).
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z ARTICLE
NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications 5
assumptions16 and they cannot model false positives that do
not arise from spectral matching, such as peptides non-
covalently associating during LC-MS28.Consideringthese
caveats, it is reassuring that our four different controls closely
agree with the outcome of the target-decoy approach. This
negates the need for any additional heuristics suggested by
others27. We showed that the target-decoy approach requires
separating self and heteromeric crosslinks and that error should
be estimated for the information level that is being reported.
For example, when reporting residue pairs for structural ana-
lyses of individual protein complexes, residue pair-level FDR
should be applied. However, when reporting PPIs, CSMs need
to be merged to PPI level prior to FDR estimation. Other ways
of merging CSM scores into PPI scores from the one we use
here are possible. However, for accurate PPI error estimation,
these methods would need to adhere to the two fundamental
considerations. These concepts were implemented in our open-
source FDR estimation software tool, xiFDR v2.0, which is
crosslink search software independent. The large dataset pre-
sented here, with its internal controls, will allow testing of other
aspects of the Crosslinking MS workflow in the future.
Correctly controlled error is an important element of any
discovery-based technology. This remains a challenge even in
well-established PPI mapping technologies including two-hybrid
and affinity purification studies. Crosslinking MS for mapping
PPIs now has a reliable FDR estimation procedure. This is an
essential prerequisite for this technology to bridge the gap
between structural studies and systems biology by reliably
revealing topologies of PPIs in their native environments.
Methods
Materials. Unless otherwise stated, reagents were purchased in the highest quality
available from Sigma (now Merck, Darmstadt, Germany). Empore 3M C18-
Material for LC-MS sample cleanup was from Sigma (St. Louis, MO, USA), glycerol
from Carl Roth (Karlsruhe, Germany). The BS3 (bis (sulfosuccinimidyl) suberate)
and sulfo-SDA (sulfosuccinimidyl 4,4’-azipentanoate) crosslinkers were supplied
by Pierce Biotechnology (Thermo Fisher Scientific, Waltham, MA, USA) and the
DSSO (disuccinimidyl sulfoxide) crosslinker from Cayman Chemical (Ann Arbor,
MI, USA).
Biomass production. A single clone of Escherichia coli K12 strain (BW25113
purchased from DSMZ, Germany; https://www.dsmz.de/) grown on Agar plates
was selected for inoculation of lysogeny broth (LB)-media. A preculture aliquot was
used to start fermentation in a Biostat A plus bioreactor (Sartorius, Göttingen,
a
msyB
dksA
rpoZ
rapA
maeB
nusA
rpoD
hupB
hupA
rpsO
rplB
yacL
greB
rpoA
ihfA
rpoB
groL
nusG
rpoE
rpoH
yifE
rpoN
rpsN
rpsL
rho
entF
rpoC
c
b
10 20 30 40
SEC Fraction
0.0
1.6
Abundance (x109)
rpoD
(0.98)
RNAP
0.8 nusA (0.98)
yacL (0.98)
nusG (0.98)
d
yacL
nusG
RNAP
DNA
entry site
DNA
exit site
rapA
yacL-SPA
nusA
nusG
−15 −7.5 0 7.5 15
log2-fold change
0
2
4
6
8
10
-log(P-value)
rpoC
rpoB
rpoA
rpoZ
Fig. 4 The uncharacterised protein YacL binds to RNA polymerase. a PPI subnetwork of RNAP. Line thickness between proteins (circles) increases with
frequency of observed crosslinks (i.e. one, two, three and more crosslinked residue pairs). Colour scheme for RNAP binders: light grey =coelution
correlation > 0.5, dark grey =STRING database combined score ≥150, black =both of the previous categories. bElution traces of RNAP (average
abundance of its subunits) and selected RNAP binders with their minimal elution correlation coefficient to any RNAP constituent. cVolcano plot showing
the affinity enrichment of SPA-tagged YacL, which co-enriches RNAP (orange) and a number of proteins found crosslinked to the RNAP in the YacL
affinity-enrichment experiment (violet). dRNAP with bound NusG (PDB 6C6U [10.2210/pdb6c6u/pdb]) and the region of Crosslinking MS-defined
accessible interaction space with 14 satisfied restraints for YacL (I-TASSER model, placed for visualisation purposes) highlighted. Crosslinks between YacL
and NusG or RNAP are highlighted in blue.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z
6NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications
Germany) in LB medium with 0.5% (w/v) glucose and at 37 °C. The pH and
dissolved oxygen were monitored and adjusted by the addition of sodium hydro-
xide/phosphoric acid or stir speed control, respectively. Overall growth was
monitored by optical density measurements at 600 nm. When the culture reached
an optical density of 10, the fermentation was stopped and the culture rapidly
cooled in stirred ice water followed by harvesting the biomass by centrifugation at
5000 × g, 4 °C for 15 min. Cell pellets were stored at −80 °C after washing with PBS
and snap-freezing in liquid nitrogen.
For pull-down experiments, Escherichia coli K12 strains with endogenously C-
terminal SPA-tagged rpoB, nusG and yacL (purchased from Horizon, Cambridge,
United Kingdom, https://horizondiscovery.com/) were plated according to
distributor’s instructions. A single clone of each strain was selected for genetic
validation and subsequent starter cultures. Gene sequences were validated by PCR
using primers hybridizing upstream of each open reading frame of interest and
within the SPA-tag sequence (Supplementary Data 6). With the exception of the
yacL-ORF which had a non-silent point mutation (Q118L), all protein- and tag-
coding sequences were correct. Production cultures were inoculated into terrific
broth medium and cultivated at 32 °C in baffled flasks until late log-phase. Biomass
was harvested and stored after snap-freezing as described above.
Cell lysis and high-molecular-weight proteome fractionation by size exclusion
chromatography (SEC). Cell pellets were suspended at 0.2 g wet-mass per ml in
ice-cold lysis buffer (50 mM Hepes pH 7.2 at RT, 50 mM KCl, 10 mM NaCl, 1.5
mM MgCl
2
, 5% (v/v) glycerol, 1 mM dithiothreitol (DTT), spatula tip of chicken
egg white lysozyme (Sigma, St. Louis, MO, USA)). Cells were lysed by sonication
on ice. Prior to sonication cOmplete EDTA-free protease-inhibitor cocktail (Roche,
Basel, Switzerland) was added according to the manufacturer’s instructions. After
sonication, 125 units of Benzonase (Merck, Darmstadt, Germany) were added.
Subsequently, the lysate was cleared of cellular debris by centrifugation for 15 min
at 4 °C and 15,000 × g. DTT was added again to 2 mM. This cleared lysate was
subjected to ultracentrifugation using a 70 Ti fixed-angle rotor for 1 h at 106,000 ×
gand 4 °C. Then, the supernatant was concentrated using ultrafiltration with
Amicon spin filters (15 kDa molecular weight cut-off) to reach a total protein
concentration of 10 mg/ml, as judged by microBCA assay (Thermo Fisher Scien-
tific, Waltham, MA, USA). Aggregates were removed by centrifugation for 5 min at
16,900 × gand 4 °C. Two milligrams of soluble high molecular weight proteome
was loaded onto a BioSep SEC-S4000 column (600 × 7.8 mm, pore size 500 Å,
particle size 5 µm, Phenomenex, CA, USA) and fractionated at 200 µl/min flow rate
and 4 °C while collecting fractions of 200 µl over the separation range from ~3
MDa to 150 kDa (as judged by Gel filtration calibration kit (HMW), GE
Healthcare).
Affinity-pulldowns of RNA polymerase constituents and binders. Cells were
lysed by sonication identically to the protocol described above. The supernatants
from centrifugation for 1 h at 4 °C and 20,000 × gwere incubated with washed
Anti-FLAG M2 agarose beads (Sigma, St. Louis, MO, USA) on a vertical rotator for
2 h at 4 °C, according to the specifications of the manufacturer. Supernatants after
incubation were discarded and beads washed twice with wash buffer (10 mM
Tris*HCl pH 7.4 at RT, 100 mM NaCl, 10% (v/v) glycerol) and once with modified
lysis buffer (50 mM Hepes pH 7.2 at RT, 50 mM KCl, 10 mM NaCl, 1.5 mM MgCl
2
,
5% (v/v) glycerol). M2 beads from replica pulldowns were pooled in a single tube
and again resuspended in modified lysis buffer. TEV protease (Sigma, St. Louis,
MO, USA) was added >0.5 U/µl M2 beads and the protein complexes of interest
eluted over 1 h at 16 °C with gentle agitation. Aliquots of cleared supernatants and
eluates from TEV cleavage were collected and processed as described below.
Sample preparation for LC-MS protein identification with non-crosslinked
samples. For protein identification from SEC fractionation, aliquots (40 µl) of each
fraction were precipitated by adding four volumes of cold acetone followed by an
incubation at −20 °C overnight. Pellets were collected by centrifugation and
supernatants discarded. Protein pellets were air-dried and subsequently solubilized
using 6 M urea, 2 M thiourea, 100 mM ABC (ammonium bicarbonate). Derivati-
zation was accomplished by incubating for 30 min at RT with 10 mM DTT fol-
lowed by 20 mM IAA (iodoacetamide) for 30 min in the dark at RT, respectively.
Proteases were added to the samples: LysC (1:100 (m/m)) for 4.5 h at 37 °C, fol-
lowed by diluting 1:5 with 100 mM ABC and continued with trypsin (1:25 (m/m))
at 37 °C for 16 h. The reactions were stopped by adding TFA (trifluoroacetic acid)
to a pH of 2–3. Subsequently, sample cleanup following the Stage-tip protocol was
performed and samples were stored at −20 °C until LC-MS acquisition. Samples
from pulldowns were processed similarly with the following changes: use of 8 M
urea, 100 mM ammonium bicarbonate for solubilization; blocking of reduced
cysteines with 30 mM IAA; digestion with trypsin (ca. 1:50 (m/m)).
LC-MS protein identification with non-crosslinked samples. Protein identifi-
cations in SEC fractions and from pull-down experiments via LC-MS were con-
ducted using a Q Exactive HF mass spectrometer (Thermo Fisher Scientific,
Bremen, Germany) coupled to an Ultimate 3000 RSLC nano system (Dionex,
Thermo Fisher Scientific, Sunnyvale, USA), operated under Tune 2.9, SII for
Xcalibur 1.4 and Xcalibur 4.1. 0.1% (v/v) formic acid and 80% (v/v) acetonitrile,
0.1% (v/v) formic acid served as mobile phases A and B, respectively. Samples were
loaded in 1.6% acetonitrile, 0.1% formic acid on an Easy-Spray column (C18, 50
cm, 75 µm ID, 2 µm particle size, 100 Å pore size) operated at 45 °C and running
with 300 nl/min flow. Peptides were eluted with the following gradient: 2 to 6%
buffer B in 1 min, 6 to 10% B in 2 min, 10 to 30%B in 37 min, 30 to 35% in 5 min
followed by 35 to 45%B in 2 min. Then, the column was set to washing conditions
within 1.5 min to 90% buffer B and flushed for another 5 min. For the mass
spectrometer the following settings were used: MS1 scans resolution 120,000, AGC
(automatic gain control) target 3 × 106, maximum injection time 50 ms, scan range
from 350 to 1600 m/z. The ten most abundant precursor ions with z =2–6, passing
the peptide match filter (“preferred”) were selected for HCD (higher-energy col-
lisional dissociation) fragmentation employing stepped normalized collision
energies (29 ± 2). The quadrupole isolation window was set to 1.6 m/z. Minimum
AGC target was 2.5 × 104, maximum injection time was 80 ms. Fragment ion scans
were recorded with a resolution of 15,000, AGC target set to 1 × 105, scanning with
afixed first mass of 100 m/z. Dynamic exclusion was enabled for 30 s after a single
count and included isotopes. Each LC-MS acquisition took 75 min.
Quantitative proteomics database search. Raw data from bottoms-up pro-
teomics experiments were processed using MaxQuant29 version 1.6.0.16 operated
under default settings (fully tryptic digestion with two missed cleavages maximum;
up to five variable modifications per peptide (oxidised methionine and acetylated
protein N-termini), MS1 match tolerance 20 ppm (first search)/4.5 ppm (main
search), MS/MS match tolerance 20 ppm); carbamidomethylation of cysteine set as
fixed modification; 1% PSM and protein group FDR). Each SEC fraction or pull-
down replica injection was treated as an individual experiment. Quantitation by
iBAQ30 requiring a minimum of two peptides (unique +razor) and matching
between runs were enabled. For data from pull-down experiments, label-free
quantitation was enabled with default settings (LFQ minimum ratio count of 2,
Fast LFQ enabled, minimum number/average of neighbour 2/6, stabilize large LFQ
ratios and requirement for MS2 for LFQ comparisons enabled). Supernatant
samples from cell lysis were included to increase absolute protein identifications via
the matching between runs feature. The database used was the Uniprot curated
reference proteome UP000000625 with two unreviewed entries removed summing
to a total 4350 proteins (retrieved on 04/08/2019).
Protein enrichment from pull-down experiments was assessed using Perseus31
version 1.5.6.0. Proteins identified by site only, reverse hits and contaminants were
filtered out. LFQ protein quantitation data was log2-transformed and filtered to
contain three valid values in at least one experiment (e.g. in any TEV eluate).
Missing values were imputed on the total matrix with default settings (width: 0.3,
downshift 1.8). Volcano plots comparing TEV eluates of targeted affinity
enrichment with K12 wildtype mock enrichment were created using a two-sided,
two-sample t-test with 1% FDR and an artificial variance S0 of 2. For high-
resolution figures, the matrix and the cut-off curve were exported to reproduce the
plots in python 3.7 with pandas 0.24.2 using the seaborn 0.9.0 package.
Protein crosslinking, digestion and sample cleanup of SEC fractions. The
remaining parts of the SEC fractions (160 µl, see above) were split into two 75 µl
aliquots, for the two crosslinking reactions, and adjusted to 97.5 µl with 1× SEC-
Running buffer. Crosslinker stock solutions were prepared freshly at 30 mM in
water free DMF. Crosslinking of the fractions was initiated by quickly mixing each
sample with 2.5 µl crosslinker stock to a final concentration of 0.75 mM crosslinker.
The crosslinking reaction was incubated for 2 h on ice before quenching with ABC
at 50 mM and further incubation for 30 min on ice. The crosslinked samples were
acetone-precipitated at −20 °C overnight (see above). Protein was solubilized in 6
M Urea, 2 M Thiourea, 100 mM ABC. For sample reduction and alkylation, 10 mM
DTT for 30 min at RT and 20 mM IAA for 30 min at RT in the dark were
employed. For sample proteolysis, LysC was added at 1:100 (m/m) ratio and
incubated for 4 h at 37 °C. Upon 1:5 dilution with 100 mM ABC, Trypsin was
added to the sample (1:25 (m/m)) and digestion continued for 16 h at 37 °C fol-
lowed by stopping via addition of TFA to a pH of 2–3. The digests were desalted
using SPE cartridges following the manufacturer’s instruction and eluates dried,
aliquoted and stored at −20 °C until further use.
Multidimensional offline fractionation of crosslinked peptide samples. All
crosslinked peptide pools were fractionated using an Äkta pure system (GE
Healthcare, Chicago, IL, USA) employing a PolySulfoethyl A SCX column (100 ×
2.1 mm, 300 Å, 3 µm) equipped with a guard column of identical stationary phase
(10 × 2.0 mm) (PolyLC, Columbia, MD, USA) running at 0.2 ml/min for the first
separation dimension. Here, mobile phase A consisted of 10 mM KH
2
PO
4
pH 3.0,
30% ACN while mobile phase B contained 1 M KCl in addition. The system was
kept at 21 °C throughout the fractionation. Dried digestion aliquots of 400 ug
peptides were dissolved in mobile phase A. Upon injection, peptides were eluted
isocratically for 2 min followed by an exponential gradient up to 700 mM KCl with
following steps: 12 min to 12.7%, followed by 1-min steps to 14.5, 16.3, 18.8, 23.0,
30.0, 40.0, 70.0% B. Fractions of 200 µl size were collected over the elution range.
The same nine high-salt fractions from five replica SCX runs were pooled for
desalting using Stage-tips. Dried Stage-tip eluates of each individual SCX fraction
were then subjected to the second dimension offline fractionation by hSAX
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z ARTICLE
NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications 7
chromatography. Here, a Dionex IonPac AS-24 hSAX column (250 × 2.0 mm) with
an AG-24 guard column (Thermo Fisher Scientific, Dreieich, Germany) were used
on Äkta pure system (see above). Mobile phase A consisted of 20 mM Tris*HCl pH
8.0 with mobile phase B containing 1 M NaCl in addition. The system was kept at
15 °C for these experiments. Samples were eluted from Stage-tips, dried and
resuspended in mobile phase A. Again, peptides were loaded under isocratic
conditions for 3 min, and then eluted by an exponential gradient with the following
steps 1.8, 3.5, 5.3, 7.1, 9.1, 11.2, 13.5, 16.3, 19.7, 24.1, 30.2, 38.8, 51.5, 70.6, 100% B
lasting for one minute each. Fractions of 150 µl size were collected throughout the
elution phase. Adjacent fractions were pooled to give ten pools in total (fractions
3–6/7–14/15–17/18–19/20–21/22–23/24–25/26–27/29–29/30–35), that were desal-
ted using Stage-tips.
Protein crosslinking, digestion and crosslink enrichment of pull-down eluates.
The remaining pull-down eluate fractions (minus aliquots for protein identifica-
tion, see above) were split into five fractions. The heterobifunctional photo-
activatable crosslinker sulfo-SDA was dissolved in modified lysis buffer (50 mM
Hepes pH 7.2 at RT, 50 mM KCl, 10 mM NaCl, 1.5 mM MgCl
2
, 5% (v/v) glycerol)
and immediately added to the samples at 50, 100, 250, 500 and 1000 µM. The
crosslinking reaction proceeded in the dark for 2 h on ice. UV-crosslinking was
achieved by irradiation with a high-power UV-A LED laser (LuxiGenTM) at 365
nm for 15 s at one Ampere32. Samples were frozen and stored at −20 °C. Next, the
samples were denatured by adding solid urea to give an 8 M solution, reduced using
DTT at 10 mM following incubation at RT for 30 min and derivatized at 30 mM
IAA over 20 min at RT and in the dark. LysC protease was added (protease:protein
ratio ca. 1:100 (m/m)) and the samples digested for 4 h at 37 °C. Then, the samples
were diluted 1:5 with 100 mM ABC and trypsin was added at a ratio of ~1:50 (m/
m). Digestion progressed for 16 h at 37 °C until stopping with TFA. Digests were
cleaned up using C18 StageTips.
Eluted peptides were fractionated using a Superdex Peptide 3.2/300 column (GE
Healthcare, Chicago, IL, USA) at a flow rate of 10 µl min−1 using 30% (v/v)
acetonitrile and 0.1 % (v/v) trifluoroacetic acid as mobile phase33. Early 50-µl
fractions were collected, dried and stored at −20 °C prior to LC-MS analysis.
LC-MS for crosslink identification. LC-MS analysis of crosslinked peptides
derived from the SEC-separated E. coli proteome and multidimensional fractio-
nation was performed using a Q Exactive HF mass spectrometer (Thermo Fisher
Scientific, Bremen, Germany) coupled to an Ultimate 3000 RSLC nano system
(Dionex, Thermo Fisher Scientific, Sunnyvale, USA), operated under Tune 2.11, SII
for Xcalibur 1.5 and Xcalibur 4.2. Mobile phases A and B consisted of 0.1% (v/v)
formic acid and 80% (v/v) acetonitrile, 0.1% (v/v) formic acid, respectively. Samples
were loaded in 1.6% acetonitrile, 0.1% formic acid on an Easy-Spray column (C18,
50 cm, 75 µm ID, 2 µm particle size, 100 Å pore size) running at 300 nl/min flow
and kept at 45 °C. Analytes were eluted with the following gradient: 2 to 7.5%
buffer B in 5 min, followed by a linear 80-min gradient of 7.5 to 42.5% and an
increase to 50% B over 2.5 min. Then, the column was set to washing conditions
within 2.5 min to 95% buffer B and flushed for another 5 min. The mass-
spectrometric settings for MS1 scans used were: resolution set to 120,000, AGC of
3×10
6, maximum injection time of 50 ms, scanning from 400–1450 m/z in profile
mode. The ten most intense precursor ions that passed the peptide match filter
(“preferred”) and with z =3–6 were isolated using a 1.4 m/z window and frag-
mented by HCD using in-house optimized stepped normalized collision energies
(BS3: 30 ± 6; DSSO: 24 ± 6). Fragment ion scans were acquired at a resolution of
60,000, AGC of 5 × 104, maximum injection time of 120 ms scanning from
200–2000 m/z, underfill ratio set to 1%. Dynamic exclusion was enabled for 30 s
(including isotopes). In-source-CID was enabled at 15 eV to minimize gas-phase
associated peptides28. Each LC-MS run took 120 min.
For LC-MS/MS analysis of sulfo-SDA crosslinked samples, we used an Orbitrap
Fusion Lumos Tribrid mass spectrometer (Thermo Fisher Scientific, Germany)
connected to an Ultimate 3000 RSLCnano system (Dionex, Thermo Fisher
Scientific, Germany), which were operated under Tune 3.4, SII for Xcalibur 1.6 and
Xcalibur 4.4. Fractions from SEC were resuspended in 1.6% acetonitrile 0.1%
formic acid and loaded onto an EASY-Spray column of 50 cm length (Thermo
Scientific) running at 300 nl/min. Gradient elution using water with 0.1% formic
acid and 80% acetonitrile with 0.1% formic acid was accomplished using optimised
gradients for each SEC fraction (from 2–18% mobile phase B to 37.5-46.5% over
90 min, followed by a linear increase to 45–55 and 95% over 2.5 min each). Each
fraction was analysed in duplicate. The settings of the mass spectrometer were as
follows: Data-dependent mode with 2.5s-Top-speed setting; MS1 scan in the
Orbitrap at 120,000 resolution over 400 to 1500 m/z with 250% normalized AGC
target; MS2 scan trigger only on precursors with z =3–7+, AGC target set to
“standard”, maximum injection time set to “dynamic”; fragmentation by HCD
employing a decision tree logic with optimised collision energies34,35; MS2 scan in
the Orbitrap at resolution of 60,000; dynamic exclusion was enabled upon a single
observation for 60 s.
Crosslink database search for BS3 and DSSO. Raw data from mass spectrometry
were processed using msConvert (version 3.0.11729)36 including denoising (top 20
peaks in 100 m/z bins) and conversion to mgf-file format. Precursor masses were
re-calibrated to account for mass shifts during measurement. Obtained peak files
were analysed using xiSEARCH 1.6.7465with the following settings: MS1/MS2
error tolerances 3 and 5 ppm, allowing up to two missing isotope peaks37, tryptic
digestion specificity with up to two missed cleavages, carbamidomethylation on
cysteine as fixed and oxidation on methionine as variable modification, losses:
–CH
3
SOH/–H
2
O/–NH
3
, crosslinker BS3 (138.06807 Da linkage mass) or DSSO
(158.0037648 Da linkage mass) with variable crosslinker modifications on linear
peptides (“BS3-NH2”155.09463 Da, “BS3-OH”156.07864 Da, “DSSO-NH2”
175.03031 Da, “DSSO-OH”176.01433 Da). xiSEARCH algorithms are identical for
both crosslinkers (BS3 and DSSO). For samples crosslinked with DSSO, additional
loss masses for crosslinker-containing ions were defined accounting for its clea-
vability (“A”54.01056 Da, “S”103.99320 Da, “T”85.98264). Matches were not
filtered for having DSSO-specific signature peaks. Crosslink sites for both reagents
were allowed for side chains of Lys, Tyr, Ser, Thr and the protein N-terminus. Note
that we included a non-covalent crosslinker with a mass of zero to flag spectra
potentially arising from gas-phase associated peptides. These spectra were removed
prior to false-discovery-rate (FDR) estimation28.
As for the non-crosslinked samples, the full E. coli proteome of 4350 proteins was
used. For the entrapment database control the database was extended by three different
entrapment databases (see below). For the final PPI network, the search database was
reduced to only proteins identified in our 44 SEC fractions, to reduce noise in the
database. Decoys were generated for all searches, including the entrapment database.
For this, protein sequences were reversed and for each decoy protein the enzyme
specific amino acids were swapped with their preceding amino acid29.
FDR calculation for BS3 and DSSO datasets. Results were filtered prior to FDR
to crosslinked peptide matches having a minimum of three matched fragments per
peptide, a delta score of 15% of the match score and a peptide length of at least six
amino acids. Additionally, identifications ambiguously matching to two proteins or
more were removed. FDR was calculated based on decoy matches by xiFDR
(version 2.0dev) using Eq. (1):16
FDR ¼TD DD
TT ð1Þ
Depending on the experiment, FDR was employed on different result levels
(CSM, peptide pair, residue pair or protein pair) with defined thresholds. Scores of
higher levels were calculated as described by Fischer and Rappsilber16 using Eq. (2):
Scorehigher level ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ΣðScorelower levelÞ2
qð2Þ
FDR was solely calculated based on that score, no further improvement by other
information was done at this point. To account for the improvement of the
identification by prefiltering on lower levels16, the same threshold was employed on
each of the lower levels. Self- and heteromeric crosslinks were handled together or
separately by enabling/disabling the grouping option.
For the final PPI network, BS3 and DSSO PPIs were separately filtered to 1%
heteromeric PPI-FDR. As the score cut-offs differed between the two datasets, the scores
from each dataset were first normalized (i.e. the local FDR was used as a normalized
score) to range between 0 and 1. Subsequently, the two tables were concatenated and
the FDR calculated again as described above and filtered to 1% FDR.
Non-crosslinkable control. Due to the high sensitivity of mass spectrometry we
identified a long list of proteins in each SEC fraction. Theoretically, all of the
proteins in a fraction could be crosslinked. In practice, however, even if this were
the case, we could not detect all these crosslinks because many would be below our
detection limit. We therefore set out to heuristically determine the detection limit
in our analysis. For this, the iBAQ values were determined across all SEC fractions
of all proteins that were part of an identified heteromeric peptide pair, at a gen-
erous 10% heteromeric peptide-pair FDR. For each identified pair of proteins, the
respective iBAQ pairs were determined across all SEC fractions (note that iBAQ
could be zero if a protein was not identified in a given fraction). Looking into each
fraction the lower of the two iBAQ values was kept. The maximum of this dis-
tribution over all fractions was then taken, called “best lower iBAQ”of a protein
pair. We assumed this to be the appropriate abundance estimate for a protein pair
and therefore the best estimate of the chance for this pair to be observed in our
experiment as crosslinked. The question now is what abundance is sufficient. As a
heuristic, we removed the lower 5% iBAQ values (iBAQ of 4.3E6). This removed
very few (169, 7%) of our identified protein pairs (n.b. at a very loose FDR
threshold), i.e. did not change much the outcome of our identification data by
generating false negatives.
For any identified protein pair to be considered plausible, both proteins had to be
found in at least one SEC fraction together with individual iBAQ values above our
iBAQ threshold (iBAQ of 4.3E6). Otherwise, they were defined as non-crosslinkable.
544,274 (6% of all theoretically possible PPIs in the E. coli proteome of 4350
proteins) PPIs are defined as plausible (Supplementary Data 7), while 8,914,801
(94%) PPIs are non-crosslinkable. Note that unlike an error control using an
entrapment database during search, only proteins that could make up the sample
were considered.
Additionally, the difference in the sizes of the false and plausible search spaces
needs to be taken into account for error estimation, i.e. the different number of
possible tryptic peptides. While all matches in the false search space are false by
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z
8NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications
definition, some matches in the plausible search space will also be random. To
account for these, the Lysine/Arginine content of proteins in the respective groups
was used as an estimate for the number of possible peptides and the observed error
is calculated with Eq. (3):
ErrorPPI;noncrosslinkable ¼nfalse
nplausible þnfalse
KRplausible þKRfalse
KRfalse
ð3Þ
¼nfalse
nplausible þnfalse
1:09
where n
false
is the number of PPIs defined as “non-crosslinkable”,n
plausible
the
number of PPIs that are plausible, both after passing the respective FDR
calculation. KR
plausible
and KR
false
are the sums of Lysines and Arginines in the
proteins of the plausible or false interactions, respectively. Therefore, the Lysine
and Arginine normalisation factor, here 1.09, is database specific.
Entrapment database control calculation. As a second control, the error of
matched PPIs was estimated based on known wrong matches to three entrapment
databases of different sizes. For one, the same number (4350) of human proteins of
similar size was added by sampling a human protein similar in Lysine and Arginine
content for each E. coli protein. As a second entrapment database, the full S.
cerevisiae proteome was added. Finally, both databases were combined for a third
entrapment database.
PPIs were defined as false if one or more proteins in the PPI was a human or
yeast protein. Additionally, the difference in entrapment and possible search space
has to be taken into account, similar to the approach for the non-crosslinkable
control, following Eq. (4):
ErrorPPI;entrapment ¼nentrapment
nE:coli þnentrapment
KRE:coli þKRentrapment
KRentrapment
ð4Þ
As expected from doubling the original database size by adding an entrapment
database of equal size, the search space normalisation approximates to 2 (1.998) for
the human entrapment database. For the yeast proteome and human proteins and
yeast proteome databases the Arginine and Lysine normalization factors are 1.19
and 1.16, respectively.
Wrong crosslinker control. As a third control we performed searches using a
wrong mass crosslinker in addition to BS3 or DSSO, respectively. Both crosslinker
masses were reduced by 28.031 Dalton. Note that for these searches, DSSO was
treated as non-cleavable. Wrong mass matches were treated separately, i.e. the
same PPI matched to correct and wrong crosslinker appeared twice in the results.
The error was normalized by a factor of 2 (see above).
Wrong precursor mass control. Spectra passing a 1% heteromeric CSM-FDR
were extracted. For these spectra, the precursor mass was downshifted by 28.031
and 42.047 Dalton (corresponding to the mass of two or three methylations).
Correct and shifted spectra were searched together, every match to a known wrong
spectrum counted as wrong. Here, the error was not corrected as we assume the
unknown wrong matches in the correct mass spectra to be only at 1% based on the
FDR employed before.
Correlation of protein elution profiles. Proteins were quantified in each SEC
fraction as described above. iBAQ values for each protein were normalized by the
maximum of the respective protein over the course of fractionation, leading to
normalized abundance values between 1 and 0. For each combination of proteins,
elution peaks were detected via the scipy python package (1.4.1). In an elution
window of 7 or more fractions, the abundances of the proteins were correlated
(Pearson). PPIs with elution profiles with a correlation coefficient >0.5 were
counted as having similar elution profiles. Code was written in python 3.7.
PPI network comparison with STRING database. For all E. coli K12 proteins
identified in quantitative proteomics experiments, interaction evidence from the
STRING database v10.538 was used (scores ranging from 0 to 1000, retrieved from
https://string-db.org on 12/19/18). PPIs were accounted as known if the STRING
combined score was equal or higher than 150. PPIs were defined as lacking
experimental evidence when the STRING experimental score was lower than 150.
Note that STRING defines 150 as the lowest cut-off in favour of an interaction.
Plotting protein elution profiles. For the creation of protein elution profiles,
fraction-wise iBAQ intensities for each protein from the MaxQuant search were
used (see above), hereinafter referred to as abundance. Individual proteins are
represented by their gene names while protein complexes, when shown in the
figures, are labelled with their respective complex name. Abundance values for
protein complexes were averaged for all components as listed in EcoCyc39,
(retrieved from https://ecocyc.org/ on 9/25/18). Plots were created in python 3.7
with pandas 0.24.2 using the seaborn 0.9.0 package.
Protein structural models. Models of protein complexes with mapped residue
pairs (Supplementary Data 8) were prepared with xiVIEW40, python 3.7 with
pandas 0.24.2 and ChimeraX 0.9241.
All structural PPI models were downloaded from the protein data bank (https://
www.rcsb.org/): PDB 5t4O [https://doi.org/10.2210/pdb5T4O/pdb] (ATP
synthase42), PDB 6RKW [https://doi.org/10.2210/pdb6RKW/pdb] (DNA gyrase43),
PDB 4PKO [https://doi.org/10.2210/pdb4PKO/pdb] (GroEL44), PDB 4S20 [https://
doi.org/10.2210/pdb4S20/pdb] (RapA45), PDB 6RIN [https://doi.org/10.2210/
pdb6RIN/pdb] (GreB46), PDB 5MS0 [https://doi.org/10.2210/pdb5MS0/pdb]
(NusG47), PDB 6FLQ [https://doi.org/10.2210/pdb6FLQ/pdb] (NusA48) and PDB
4ZH3 [https://doi.org/10.2210/pdb4ZH3/pdb] (RpoD49).
Database search and FDR calculation for sulfo-SDA crosslinked pulldowns.A
recalibration of the precursor m/z was conducted based on high-confidence linear
peptide identifications37. The re-calibrated peak lists were searched against the
sequences of proteins identified in a given pull-down and with an iBAQ ≥5e6 along
with their reversed sequences (as decoys) using xiSEARCH (v.1.7.6.2) for identi-
fication. MS-cleavability of the sulfo-SDA crosslinker was considered50. Final
crosslink lists were compiled using the identified candidates filtered to 2% FDR on
residue pair-level and 5% on PPI level with xiFDR v.2.1.517.
RNA polymerase binding site of YacL. An I-TASSER41 (v.5.1)51 model for YacL
was generated with default settings based on the Uniprot sequence (see above). DisVis
(v.2.0)52 ran under default settings, with YacL as scanning model and fixed model PDB
6C6U [https://doi.org/10.2210/pdb6c6u/pdb]53 with residue 118–127 of NusG mod-
elled using the Modeller54 plug-in in Chimera55.ResiduepairsofYacLtoRNAPand
NusG were used as restraints with a minimal distance of 2 Å, and a maximal distance
of 30 or 20Å for DSSO/BS3 and sulfo-SDA, respectively. The density displayed in
Fig. 4d corresponds to the accessible interaction space with 14 satisfied restraints. The
I-TASSER model was placed for visualisation purposes only.
Reporting summary. Further information on research design is available in the Nature
Research Reporting Summary linked to this article.
Data availability
Raw data and MaxQuant outputs from quantitative proteomics SEC-MS experiments
were deposited with the ProteomeXchange Consortium partner repository jPOSTrepo
under the accession codes JPST00084356 and PXD019004. Raw data and MaxQuant
outputs from quantitative proteomics AP-MS experiments were deposited with the
ProteomeXchange Consortium partner repository jPOSTrepo under the accession codes
JPST00109056 and PXD024146. All raw data, peak lists and search result files from BS3/
DSSO crosslinking experiments in the SEC fractions and after multidimensional
fractionation were deposited with the ProteomeXchange Consortium partner repository
jPOSTrepo under the accession codes JPST00084556 and PXD019120. All raw data, peak
lists and search result files from affinity-enrichment and crosslinking experiments were
deposited with the ProteomeXchange Consortium partner repository jPOSTrepo under
the accession JPST00109156 and PXD024148. We accessed the STRING database (v10.5)
via https://string-db.org/. The new link for this version is https://version-10-5.string-db.
org/. The used resource can be downloaded using the following link: https://version-10-5.
string-db.org/download/protein.links.detailed.v10.5/511145.protein.links.detailed.v10.5.
txt.gz. Models from the protein data bank (PDB) can be found under the following links:
PDB 5t4O [https://doi.org/10.2210/pdb5T4O/pdb] (ATP synthase42), PDB 6RKW
[https://doi.org/10.2210/pdb6RKW/pdb] (DNA gyrase43), PDB 4PKO [https://doi.org/
10.2210/pdb4PKO/pdb] (GroEL44), PDB 4S20 [https://doi.org/10.2210/pdb4S20/pdb]
(RapA45), PDB 6RIN [https://doi.org/10.2210/pdb6RIN/pdb] (GreB46), PDB 5MS0
[https://doi.org/10.2210/pdb5MS0/pdb] (NusG47), PDB 6FLQ [https://doi.org/10.2210/
pdb6FLQ/pdb] (NusA48), PDB 4ZH3 [https://doi.org/10.2210/pdb4ZH3/pdb] (RpoD49),
PDB 6C6U [https://doi.org/10.2210/pdb6c6u/pdb] (RNAP-NusG53). Source data are
provided with this paper.
Code availability
The xiFDR version57 used in this manuscript (v2.0.dev) is available via Zenodo at https://
doi.org/10.5281/zenodo.4682917. More recent xiFDR versions can be downloaded from
https://github.com/Rappsilber-Laboratory/xiFDR or https://www.rappsilberlab.org/
software/xifdr/.
Received: 21 August 2020; Accepted: 28 April 2021;
References
1. O’Reilly, F. J. & Rappsilber, J. Cross-linking mass spectrometry: methods and
applications in structural, molecular and systems biology. Nat. Struct. Mol.
Biol. 25, 1000–1008 (2018).
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z ARTICLE
NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications 9
2. Chavez, J. D., Keller, A., Zhou, B., Tian, R. & Bruce, J. E. Cellular interactome
dynamics during paclitaxel treatment. Cell Rep. 29, 2371–2383.e5 (2019).
3. Steigenberger, B., Pieters, R. J., Heck, A. J. R. & Scheltema, R. A. PhoX: an
IMAC-enrichable cross-linking reagent. ACS Cent. Sci. 5, 1514–1522 (2019).
4. Chen, Z.-L. et al. A high-speed search engine pLink 2 with systematic
evaluation for proteome-scale identification of cross-linked peptides. Nat.
Commun. 10, 3404 (2019).
5. Mendes, M. L. et al. An integrated workflow for crosslinking mass
spectrometry. Mol. Syst. Biol. 15, e8994 (2019).
6. Ryl, P. S. J. et al. In situ structural restraints from crosslinking mass
spectrometry in human mitochondria. J. Proteome Res.https://doi.org/
10.1021/acs.jproteome.9b00541 (2019).
7. Gonzalez-Lozano, M. A. et al. Stitching the synapse: cross-linking mass
spectrometry into resolving synaptic protein interactions. Sci. Adv. 6, eaax5783
(2020).
8. Makepeace, K. A. T. et al. Improving identification of in-organello protein-
protein interactions using an affinity-enrichable, isotopically coded, and mass
spectrometry-cleavable chemical crosslinker. Mol. Cell. Proteomics 19,
624–639 (2020).
9. Bartolec, T. K. et al. Cross-linking mass spectrometry analysis of the yeast
nucleus reveals extensive protein-protein interactions not detected by
systematic two-hybrid or affinity purification-mass spectrometry. Anal. Chem.
92, 1874–1882 (2020).
10. Linden, A. et al. A cross-linking mass spectrometry approach defines protein
interactions in yeast mitochondria. Mol. Cell. Proteomics https://doi.org/
10.1074/mcp.RA120.002028 (2020).
11. Götze, M., Iacobucci, C., Ihling, C. H. & Sinz, A. A simple cross-linking/mass
spectrometry workflow for studying system-wide protein interactions. Anal.
Chem. 91, 10236–10244 (2019).
12. Yugandhar, K. et al. MaXLinker: proteome-wide cross-link identifications
with high specificity and sensitivity. Mol. Cell. Proteomics 19, 554–568 (2020).
13. O’Reilly, F. J. et al. In-cell architecture of an actively transcribing-translating
expressome. Science 369, 554–557 (2020).
14. Maiolica, A. et al. Structural analysis of multiprotein complexes by cross-
linking, mass spectrometry, and database searching. Mol. Cell. Proteomics 6,
2200–2211 (2007).
15. Walzthoeni, T. et al. False discovery rate estimation for cross-linked peptides
identified by mass spectrometry. Nat. Methods 9, 901–903 (2012).
16. Fischer, L. & Rappsilber, J. Quirks of error estimation in cross-linking/mass
spectrometry. Anal. Chem. 89, 3829–3833 (2017).
17. Fischer, L. & Rappsilber, J. False discovery rate estimation and
heterobifunctional cross-linkers. PLoS ONE 13, e0196672 (2018).
18. Yang, B. et al. Identification of cross-linked peptides from complex samples.
Nat. Methods 9, 904–906 (2012).
19. Keller, A., Chavez, J. D., Felt, K. C. & Bruce, J. E. Prediction of an upper limit
for the fraction of interprotein cross-links in large-scale in vivo cross-linking
studies. J. Proteome Res. 18, 3077–3085 (2019).
20. Beveridge, R., Stadlmann, J., Penninger, J. M. & Mechtler, K. A synthetic
peptide library for benchmarking crosslinking-mass spectrometry search
engines for proteins and protein complexes. Nat. Commun. 11, 742 (2020).
21. Leitner, A. et al. Towards increased reliability, transparency and accessibility
in cross-linking mass spectrometry. Structure 28, 1259–1268 (2020).
22. Schmidt, A. et al. The quantitative and condition-dependent Escherichia coli
proteome. Nat. Biotechnol. 34, 104–110 (2016).
23. Butland, G. et al. Interaction network containing conserved and essential
protein complexes in Escherichia coli. Nature 433, 531–537 (2005).
24. Belsom, A., Schneider, M., Fischer, L., Brock, O. & Rappsilber, J. Serum
albumin domain structures in human blood serum by mass spectrometry and
computational biology. Mol. Cell. Proteomics 15, 1105–1116 (2016).
25. Matzinger, M., Kandioller, W., Doppler, P., Heiss, E. H. & Mechtler, K. Fast
and highly efficient affinity enrichment of Azide-A-DSBSO cross-linked
peptides. J. Proteome Res. 19, 2071–2079 (2020).
26. Chen, Z. A. & Rappsilber, J. Quantitative cross-linking/mass spectrometry to
elucidate structural changes in proteins and their complexes. Nat. Protoc. 14,
171–201 (2019).
27. Yugandhar,K.,Wang,T.-Y.,Wierbowski,S.D.,Shayhidin,E.E.&Yu,H.
Structure-based validation can drastically underestimate error rate in proteome-
wide cross-linking mass spectrometry studies. Nat. Methods 17,985–988 (2020).
28. Giese, S. H., Belsom, A., Sinn, L., Fischer, L. & Rappsilber, J. Noncovalently
associated peptides observed during liquid chromatography-mass spectrometry
and their effect on cross-link analyses. Anal. Chem. 91, 2678–2685 (2019).
29. Cox, J. & Mann, M. MaxQuant enables high peptide identification rates,
individualized p.p.b.-range mass accuracies and proteome-wide protein
quantification. Nat. Biotechnol. 26, 1367–1372 (2008).
30. Schwanhäusser, B. et al. Global quantification of mammalian gene expression
control. Nature 473, 337–342 (2011).
31. Tyanova, S. et al. The Perseus computational platform for comprehensive
analysis of (prote)omics data. Nat. Methods 13, 731–740 (2016).
32. Horne, J. E. et al. Rapid mapping of protein interactions using tag-transfer
photocrosslinkers. Angew. Chem. Int. Ed. 57, 16688–16692 (2018).
33. Leitner, A. et al. Expanding the chemical cross-linking toolbox by the use of
multiple proteases and enrichment by size exclusion chromatography. Mol.
Cell. Proteomics 11, M111.014126 (2012).
34. Giese, S. H., Belsom, A. & Rappsilber, J. Optimized fragmentation regime for
diazirine photo-cross-linked peptides. Anal. Chem. 88, 8239–8247 (2016).
35. Kolbowski, L., Mendes, M. L. & Rappsilber, J. Optimizing the parameters
governing the fragmentation of cross-linked peptides in a tribrid mass
spectrometer. Anal. Chem. 89, 5311–5318 (2017).
36. Chambers, M. C. et al. A cross-platform toolkit for mass spectrometry and
proteomics. Nat. Biotechnol. 30, 918–920 (2012).
37. Lenz, S., Giese, S. H., Fischer, L. & Rappsilber, J. In-search assignment of
monoisotopic peaks improves the identification of cross-linked peptides. J.
Proteome Res. 17, 3923–3931 (2018).
38. Szklarczyk, D. et al. STRING v11: protein-protein association networks with
increased coverage, supporting functional discovery in genome-wide
experimental datasets. Nucleic Acids Res. 47, D607–D613 (2019).
39. Keseler, I. M. et al. The EcoCyc database: reflecting new knowledge about
Escherichia coli K-12. Nucleic Acids Res. 45, D543–D550 (2017).
40. Combe, C. W., Fischer, L. & Rappsilber, J. xiNET: cross-link network maps
with residue resolution. Mol. Cell. Proteomics 14, 1137–1147 (2015).
41. Goddard, T. D. et al. UCSF ChimeraX: Meeting modern challenges in
visualization and analysis. Protein Sci. 27,14–25 (2018).
42. Sobti, M. et al. Cryo-EM structures of the autoinhibited E. coli ATP synthase
in three rotational states. elife 5, e21598 (2016).
43. Vanden Broeck, A., Lotz, C., Ortiz, J. & Lamour, V. Cryo-EM structure of the
complete E. coli DNA gyrase nucleoprotein complex. Nat. Commun. 10, 4935
(2019).
44. Fei, X., Ye, X., LaRonde, N. A. & Lorimer, G. H. Formation and structures of
GroEL:GroES2 chaperonin footballs, the protein-folding functional form.
Proc. Natl Acad. Sci. USA 111, 12775–12780 (2014).
45. Liu, B., Zuo, Y. & Steitz, T. A. Structural basis for transcription reactivation by
RapA. Proc. Natl Acad. Sci. USA 112, 2006–2010 (2015).
46. Abdelkareem, M. ’men et al. Structural basis of transcription: RNA
polymerase backtracking and its reactivation. Mol. Cell 75, 298–309.e4 (2019).
47. Said, N. et al. Structural basis for λN-dependent processive transcription
antitermination. Nat. Microbiol 2, 17062 (2017).
48. Guo, X. et al. Structural basis for NusA stabilized transcriptional pausing. Mol.
Cell 69, 816–827.e4 (2018).
49. Feng, Y. et al. Structural basis of transcription inhibition by CBR
hydroxamidines and CBR pyrazoles. Structure 23, 1470–1481 (2015).
50. Iacobucci, C. et al. Carboxyl-photo-reactive MS-cleavable cross-linkers: unveiling a
hidden aspect of diazirine-based reagents. Anal. Chem. 90,2805–2809 (2018).
51. Yang, J. & Zhang, Y. I-TASSER server: new development for protein structure
and function predictions. Nucleic Acids Res.43, W174–W181 (2015).
52. van Zundert, G. C. P. & Bonvin, A. M. J. J. DisVis: quantifying and visualizing
accessible interaction space of distance-restrained biomolecular complexes.
Bioinformatics 31, 3222–3224 (2015).
53. Kang, J. Y. et al. Structural basis for transcript elongation control by NusG
family universal regulators. Cell 173, 1650–1662.e14 (2018).
54. Sali, A. & Blundell, T. L. Comparative protein modelling by satisfaction of
spatial restraints. J. Mol. Biol. 234, 779–815 (1993).
55. Pettersen, E. F. et al. UCSF Chimera–a visualization system for exploratory
research and analysis. J. Comput. Chem. 25, 1605–1612 (2004).
56. Okuda, S. et al. jPOSTrepo: an international standard data repository for
proteomes. Nucleic Acids Res.45, D1107–D1111 (2017).
57. Lenz, S. et al. Reliable identification of protein-protein interactions by
crosslinking mass spectrometry. Zenodo https://doi.org/10.5281/
zenodo.4682917 (2021).
Acknowledgements
We would like to thank Richard Scheltema, Alexander Leitner, Andrea Sinz, Michael
Hoopmann, Marc Wilkins, Fan Liu, Henning Urlaub, James Bruce, Si-Min He, Meng-
Qiu Dong and Dermot Harnett for comments on the manuscript. We thank Tabea
Schütze for fermenting E. coli. The work was funded by the Deutsche For-
schungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence
Strategy—EXC 2008—390540038—UniSysCat, by grant no. 392923329/GRK2473, grant
no. 426290502 and by the Wellcome Trust through a Senior Research Fellowship to J.R.
(103139). The Wellcome Centre for Cell Biology is supported by core funding from the
Wellcome Trust (203149).
Author contributions
F.O., L.S., S.L. and J.R. designed the experiments; L.S., F.W. and F.O. prepared the
samples. L.S. did the affinity-enrichment experiments. L.S., S.L. and L.F. collected and
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z
10 NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications
processed Crosslinking MS data; L.F. designed and implemented xiFDR software; S.L.,
L.S., F.O. and J.R. prepared figures and wrote the manuscript with input from all authors.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information The online version contains supplementary material
available at https://doi.org/10.1038/s41467-021-23666-z.
Correspondence and requests for materials should be addressed to J.R.
Peer review information Nature Communications thanks Si-Min He, Nir Kalisman and
the other, anonymous, reviewer for their contribution to the peer review of this work.
Peer reviewer reports are available.
Reprints and permission information is available at http://www.nature.com/reprints
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the article’s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
article’s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this license, visit http://creativecommons.org/
licenses/by/4.0/.
© The Author(s) 2021
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23666-z ARTICLE
NATURE COMMUNICATIONS | (2021)12:3564 | https://doi.org/10.1038/s41467-021-23666-z | www.nature.com/naturecommunications 11
53
Outlook
Crosslinking MS is a powerful tool producing valuable information to aid structural
biology. Its application to purified complexes can by now be seen as a standardised and easy
workflow and is routinely applied in in vitro structural studies. As a complementary method to
high-resolution structural biology techniques like electron microscopy in integrative structural
biology approaches, it aids in solving lower resolution data or regions (O’Reilly et al., 2020).
In addition, crosslinking MS is becoming a useful tool for systems biology studies. A set of
successful large-scale, and in situ, studies has been published, delivering PPIs and structural
information at the same time (Chavez et al., 2019; Gonzalez-Lozano et al., 2020; Linden et al.,
2020). However, the experimental effort required for large-scale crosslinking is still
considerable and often only covers the most abundant parts of the proteome. More
developments will be needed to increase depth and feasibility of in-cell crosslinking MS.
In this thesis I have demonstrated developments in the crosslinking MS pipeline that
yielded large improvements in data acquisition and reliability. Any further optimization, wet
or dry lab, needs thorough analysis of the data. For example, to further improve database search
results, implementation of machine learning in the crosslinking search engines seems like an
obvious next step as it has been shown to be beneficial in standard proteomics search engines
(The et al., 2016), although it will require careful implementation to avoid overtraining.
One sometimes overlooked optimization aspect is the spectral quality. Many people
have tried to optimise for fast acquisitions to gain as many spectra as possible. However, as
spectra are the basis for any identification, to further improve crosslinking analyses it is
required that their quality improves further. While it is possible some gains will be made by
invention of more sensitive mass spectrometers, another way to improve spectral quality is by
increasing the abundance of the crosslinked peptides. Besides chromatographic enrichment,
enrichment of the crosslinker itself has been developed, but is yet to be widely employed.
Recently, mass spectrometer vendors have made the instrument application programming
interface available to users. This allows real-time analysis of the data and on the fly decision
making during acquisition and has been used in linear proteomics already (Wichmann et al.,
2019). Adapting this to crosslinking MS workflows offers multiple routes to increase spectral
quality and depth, e.g. by optimising precursor selection (currently hindered by the dynamic
range problem on MS1) or choice of fragmentation schemas.
Another focus should also be the reliability of the crosslinking MS results. This requires
an established error control, which we demonstrated in this work for large-scale crosslinking
data and their PPIs. We provide a dataset to test search and FDR approaches and demonstrate
multiple controls transferable to other samples, which allows adaptation of these validations
by other crosslinking MS labs. As crosslinking PPIs are starting to be fed into public databases
used by others (Schweppe et al., 2016), their reliability is getting more important. Improved
quality of crosslinking data will also lead to growing credibility outside of the field.
While the work of this thesis has focused on NHS-ester crosslinkers, photo-crosslinking
with the UV-activatable crosslinker SDA on a larger scale is more difficult. Because it can link
to any amino acid on one side, the crosslinked peptide mixture is more complex and each
54
crosslink less abundant. Besides experimental optimisation, large scale photo-crosslinking will
require more computational efforts, i.e. further optimisation of the database search.
In the next years, proteome-wide crosslinking will become more feasible, enabling its
application with less effort and by a wider user community. Increased depth of the analysis will
lead to more detailed interaction networks that also include lower abundant proteins. Taken
together the field is racing towards a future where the protein interaction maps can be generated
in cells and in tissues without gene-tagging or cell disruption. The quantitative approaches that
are possible with proteomics will allow comparison of these cellular interaction maps with
disease states.
55
Acknowledgements
A big thank you goes to my family and friends, who have accompanied me to where I am today.
To my parents, who I knew I could always turn to if I needed something. To my sisters, who
would always be just a phone call or message away if there was something to talk about. And
to my brother, who was always happy to be here and who I hope will keep being curious.
Theresa, you have been a big emotional support throughout the years and I’m happy to have
such a good and long lasting friend. And to Sevgi - we have made the best out of a difficult
start and I want to thank you for the open ear and support over the past months.
Doing this thesis required a bit more extra help than for others. Anne, Ralf, Petra and many
others, you often have done more for me than what was required and had to share my ups and
downs – thank you.
My time in the Rappsilber lab involved a lot of work, but at the same time a great deal of fun.
First, I want to thank Juri for creating the environment which I very much enjoyed working in
the past years and for giving me the opportunity to do my PhD there. I have learned many
things besides crosslinking from you and could always ask for valuable advice if needed. I have
appreciated the many scientific discussions we had, which I hope will be continued in the future
in one way or the other.
Francis, besides handling my many outbursts gracefully, you have taught me your approach to
science and gave me confidence in me and my work over the past years. Andrea, you had a lot
of knowledge to share and I enjoyed learning from you so much. You both have been massive
influences on me and I’m happy to call you my friends.
Sven, I am thankful to you for introducing me to crosslinking when I started out in the group.
Lutz, you have always been there for any (technical) support or detailed discussions. Lars, you
have made working on our fastest project as enjoyable as possible. Ludwig, you have been a
great colleague - and friend - to share drinks, banter, or extensive discussions. Fabi, besides the
almost never ending supply of coffee, it was a pleasure to share an office with you.
I don’t think I would be where I am today without the influences the team has had on me.
Everyone in the lab has supported me in one way or the other over the years. Thank you all!
56
References
Audain, E., Uszkoreit, J., Sachsenberg, T., Pfeuffer, J., Liang, X., Hermjakob, H., Sanchez, A.,
Eisenacher, M., Reinert, K., Tabb, D. L., Kohlbacher, O., & Perez-Riverol, Y. (2017). In-depth
analysis of protein inference algorithms using multiple search engines and well-defined metrics.
Journal of Proteomics, 150, 170–182. https://doi.org/10.1016/j.jprot.2016.08.002
Bartolec, T. K., Smith, D.-L., Pang, C. N. I., Xu, Y. D., Hamey, J. J., & Wilkins, M. R. (2020). Cross-
linking Mass Spectrometry Analysis of the Yeast Nucleus Reveals Extensive Protein-Protein
Interactions Not Detected by Systematic Two-Hybrid or Affinity Purification-Mass
Spectrometry. Analytical Chemistry, 92(2), 1874–1882.
https://doi.org/10.1021/acs.analchem.9b03975
Belsom, A., & Rappsilber, J. (2021). Anatomy of a crosslinker. Current Opinion in Chemical Biology,
60, 39–46. https://doi.org/10.1016/j.cbpa.2020.07.008
Beveridge, R., Stadlmann, J., Penninger, J. M., & Mechtler, K. (2020). A synthetic peptide library for
benchmarking crosslinking-mass spectrometry search engines for proteins and protein
complexes. Nature Communications, 11(1), 742. https://doi.org/10.1038/s41467-020-14608-2
Böhning, J., & Bharat, T. A. M. (2021). Towards high-throughput in situ structural biology using
electron cryotomography. Progress in Biophysics and Molecular Biology, 160, 97–103.
https://doi.org/10.1016/j.pbiomolbio.2020.05.010
Chavez, J. D., Keller, A., Zhou, B., Tian, R., & Bruce, J. E. (2019). Cellular Interactome Dynamics
during Paclitaxel Treatment. Cell Reports, 29(8), 2371–2383.e5.
https://doi.org/10.1016/j.celrep.2019.10.063
Chavez, J. D., Lee, C. F., Caudal, A., Keller, A., Tian, R., & Bruce, J. E. (2018). Chemical
Crosslinking Mass Spectrometry Analysis of Protein Conformations and Supercomplexes in
Heart Tissue. Cell Systems, 6(1), 136–141.e5. https://doi.org/10.1016/j.cels.2017.10.017
Chen, Z. A., Jawhari, A., Fischer, L., Buchen, C., Tahir, S., Kamenski, T., Rasmussen, M., Lariviere,
L., Bukowski-Wills, J.-C., Nilges, M., Cramer, P., & Rappsilber, J. (2010). Architecture of the
RNA polymerase II-TFIIF complex revealed by cross-linking and mass spectrometry. The
EMBO Journal, 29(4), 717–726. https://doi.org/10.1038/emboj.2009.401
Chen, Z.-L., Meng, J.-M., Cao, Y., Yin, J.-L., Fang, R.-Q., Fan, S.-B., Liu, C., Zeng, W.-F., Ding, Y.-
H., Tan, D., Wu, L., Zhou, W.-J., Chi, H., Sun, R.-X., Dong, M.-Q., & He, S.-M. (2019). A high-
speed search engine pLink 2 with systematic evaluation for proteome-scale identification of
cross-linked peptides. Nature Communications, 10(1), 3404. https://doi.org/10.1038/s41467-019-
11337-z
Cox, J., & Mann, M. (2008). MaxQuant enables high peptide identification rates, individualized
p.p.b.-range mass accuracies and proteome-wide protein quantification. Nature Biotechnology,
26(12), 1367–1372. https://doi.org/10.1038/nbt.1511
Cox, J., Neuhauser, N., Michalski, A., Scheltema, R. A., Olsen, J. V., & Mann, M. (2011).
Andromeda: a peptide search engine integrated into the MaxQuant environment. Journal of
Proteome Research, 10(4), 1794–1805. https://doi.org/10.1021/pr101065j
Elias, J. E., & Gygi, S. P. (2007). Target-decoy search strategy for increased confidence in large-scale
protein identifications by mass spectrometry. Nature Methods, 4(3), 207–214.
https://doi.org/10.1038/nmeth1019
Fischer, L., & Rappsilber, J. (2017). Quirks of Error Estimation in Cross-Linking/Mass Spectrometry.
Analytical Chemistry, 89(7), 3829–3833. https://doi.org/10.1021/acs.analchem.6b03745
Fritzsche, R., Ihling, C. H., Götze, M., & Sinz, A. (2012). Optimizing the enrichment of cross-linked
products for mass spectrometric protein analysis. Rapid Communications in Mass Spectrometry:
57
RCM, 26(6), 653–658. https://doi.org/10.1002/rcm.6150
Giese, S. H., Fischer, L., & Rappsilber, J. (2016). A Study into the Collision-induced Dissociation
(CID) Behavior of Cross-Linked Peptides*. Molecular & Cellular Proteomics: MCP, 15(3),
1094–1104. https://doi.org/10.1074/mcp.M115.049296
Giese, S. H., Sinn, L. R., Wegner, F., & Rappsilber, J. (2021). Retention time prediction using neural
networks increases identifications in crosslinking mass spectrometry. Nature Communications,
12(1), 3237. https://doi.org/10.1038/s41467-021-23441-0
Gonzalez-Lozano, M. A., Koopmans, F., Sullivan, P. F., Protze, J., Krause, G., Verhage, M., Li, K.
W., Liu, F., & Smit, A. B. (2020). Stitching the synapse: Cross-linking mass spectrometry into
resolving synaptic protein interactions. Science Advances, 6(8), eaax5783.
https://doi.org/10.1126/sciadv.aax5783
Götze, M., Iacobucci, C., Ihling, C. H., & Sinz, A. (2019). A Simple Cross-Linking/Mass
Spectrometry Workflow for Studying System-wide Protein Interactions. Analytical Chemistry,
91(15), 10236–10244. https://doi.org/10.1021/acs.analchem.9b02372
Hoopmann, M. R., Zelter, A., Johnson, R. S., Riffle, M., MacCoss, M. J., Davis, T. N., & Moritz, R.
L. (2015). Kojak: efficient analysis of chemically cross-linked protein complexes. Journal of
Proteome Research, 14(5), 2190–2198. https://doi.org/10.1021/pr501321h
Houel, S., Abernathy, R., Renganathan, K., Meyer-Arendt, K., Ahn, N. G., & Old, W. M. (2010).
Quantifying the impact of chimera MS/MS spectra on peptide identification in large-scale
proteomics studies. Journal of Proteome Research, 9(8), 4152–4160.
https://doi.org/10.1021/pr1003856
Kao, A., Chiu, C.-L., Vellucci, D., Yang, Y., Patel, V. R., Guan, S., Randall, A., Baldi, P.,
Rychnovsky, S. D., & Huang, L. (2011). Development of a novel cross-linking strategy for fast
and accurate identification of cross-linked peptides of protein complexes. Molecular & Cellular
Proteomics: MCP, 10(1), M110.002212. https://doi.org/10.1074/mcp.M110.002212
Klammer, A. A., Yi, X., MacCoss, M. J., & Noble, W. S. (2007). Improving tandem mass spectrum
identification using peptide retention time prediction across diverse chromatography conditions.
Analytical Chemistry, 79(16), 6111–6118. https://doi.org/10.1021/ac070262k
Kolbowski, L., Mendes, M. L., & Rappsilber, J. (2017). Optimizing the Parameters Governing the
Fragmentation of Cross-Linked Peptides in a Tribrid Mass Spectrometer. Analytical Chemistry,
89(10), 5311–5318. https://doi.org/10.1021/acs.analchem.6b04935
Kong, A. T., Leprevost, F. V., Avtonomov, D. M., Mellacheruvu, D., & Nesvizhskii, A. I. (2017).
MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry-based
proteomics. Nature Methods, 14(5), 513–520. https://doi.org/10.1038/nmeth.4256
Kristensen, A. R., Gsponer, J., & Foster, L. J. (2012). A high-throughput approach for measuring
temporal changes in the interactome. Nature Methods, 9(9), 907–909.
https://doi.org/10.1038/nmeth.2131
Lasker, K., Förster, F., Bohn, S., Walzthoeni, T., Villa, E., Unverdorben, P., Beck, F., Aebersold, R.,
Sali, A., & Baumeister, W. (2012). Molecular architecture of the 26S proteasome holocomplex
determined by an integrative approach. Proceedings of the National Academy of Sciences of the
United States of America, 109(5), 1380–1387. https://doi.org/10.1073/pnas.1120559109
Lee, B.-G., Merkel, F., Allegretti, M., Hassler, M., Cawood, C., Lecomte, L., O’Reilly, F. J., Sinn, L.
R., Gutierrez-Escribano, P., Kschonsak, M., Bravo, S., Nakane, T., Rappsilber, J., Aragon, L.,
Beck, M., Löwe, J., & Haering, C. H. (2020). Cryo-EM structures of holo condensin reveal a
subunit flip-flop mechanism. Nature Structural & Molecular Biology, 27(8), 743–751.
https://doi.org/10.1038/s41594-020-0457-x
Leitner, A., Reischl, R., Walzthoeni, T., Herzog, F., Bohn, S., Förster, F., & Aebersold, R. (2012).
Expanding the chemical cross-linking toolbox by the use of multiple proteases and enrichment
58
by size exclusion chromatography. Molecular & Cellular Proteomics: MCP, 11(3),
M111.014126. https://doi.org/10.1074/mcp.M111.014126
Lenz, S., Giese, S. H., Fischer, L., & Rappsilber, J. (2018). In-Search Assignment of Monoisotopic
Peaks Improves the Identification of Cross-Linked Peptides. Journal of Proteome Research,
17(11), 3923–3931. https://doi.org/10.1021/acs.jproteome.8b00600
Lenz, S., Sinn, L. R., O’Reilly, F. J., Fischer, L., Wegner, F., & Rappsilber, J. (2021). Reliable
identification of protein-protein interactions by crosslinking mass spectrometry. Nature
Communications, 12(1), 3564. https://doi.org/10.1038/s41467-021-23666-z
Linden, A., Deckers, M., Parfentev, I., Pflanz, R., Homberg, B., Neumann, P., Ficner, R., Rehling, P.,
& Urlaub, H. (2020). A cross-linking mass spectrometry approach defines protein interactions in
yeast mitochondria. Molecular & Cellular Proteomics: MCP.
https://doi.org/10.1074/mcp.RA120.002028
Liu, F., Lössl, P., Rabbitts, B. M., Balaban, R. S., & Heck, A. J. R. (2018). The interactome of intact
mitochondria by cross-linking mass spectrometry provides evidence for coexisting respiratory
supercomplexes. Molecular & Cellular Proteomics: MCP, 17(2), 216–232.
https://doi.org/10.1074/mcp.RA117.000470
Liu, F., Lössl, P., Scheltema, R., Viner, R., & Heck, A. J. R. (2017). Optimized fragmentation
schemes and data analysis strategies for proteome-wide cross-link identification. Nature
Communications, 8, 15473. https://doi.org/10.1038/ncomms15473
Mendes, M. L., Fischer, L., Chen, Z. A., Barbon, M., O’Reilly, F. J., Giese, S. H., Bohlke-Schneider,
M., Belsom, A., Dau, T., Combe, C. W., Graham, M., Eisele, M. R., Baumeister, W., Speck, C.,
& Rappsilber, J. (2019). An integrated workflow for crosslinking mass spectrometry. Molecular
Systems Biology, 15(9), e8994. https://doi.org/10.15252/msb.20198994
Müller, M. Q., Dreiocker, F., Ihling, C. H., Schäfer, M., & Sinz, A. (2010). Cleavable cross-linker for
protein structure analysis: reliable identification of cross-linking products by tandem MS.
Analytical Chemistry, 82(16), 6958–6968. https://doi.org/10.1021/ac101241t
O’Reilly, F. J., & Rappsilber, J. (2018). Cross-linking mass spectrometry: methods and applications in
structural, molecular and systems biology. Nature Structural & Molecular Biology, 25(11),
1000–1008. https://doi.org/10.1038/s41594-018-0147-0
O’Reilly, F. J., Xue, L., Graziadei, A., Sinn, L., Lenz, S., Tegunov, D., Blötz, C., Singh, N., Hagen,
W. J. H., Cramer, P., Stülke, J., Mahamid, J., & Rappsilber, J. (2020). In-cell architecture of an
actively transcribing-translating expressome. Science, 369(6503), 554–557.
https://doi.org/10.1126/science.abb3758
Parrish, J. R., Gulyas, K. D., & Finley, R. L., Jr. (2006). Yeast two-hybrid contributions to
interactome mapping. Current Opinion in Biotechnology, 17(4), 387–393.
https://doi.org/10.1016/j.copbio.2006.06.006
Rigaut, G., Shevchenko, A., Rutz, B., Wilm, M., Mann, M., & Séraphin, B. (1999). A generic protein
purification method for protein complex characterization and proteome exploration. Nature
Biotechnology, 17(10), 1030–1032. https://doi.org/10.1038/13732
Robinson, P. J., Trnka, M. J., Pellarin, R., Greenberg, C. H., Bushnell, D. A., Davis, R., Burlingame,
A. L., Sali, A., & Kornberg, R. D. (2015). Molecular architecture of the yeast Mediator complex.
eLife, 4. https://doi.org/10.7554/eLife.08719
Ryl, P. S. J., Bohlke-Schneider, M., Lenz, S., Fischer, L., Budzinski, L., Stuiver, M., Mendes, M. M.
L., Sinn, L., O’Reilly, F. J., & Rappsilber, J. (2020). In Situ Structural Restraints from Cross-
Linking Mass Spectrometry in Human Mitochondria. Journal of Proteome Research, 19(1), 327–
336. https://doi.org/10.1021/acs.jproteome.9b00541
Schäpe, P., Kwon, M. J., Baumann, B., Gutschmann, B., Jung, S., Lenz, S., Nitsche, B., Paege, N.,
Schütze, T., Cairns, T. C., & Meyer, V. (2019). Updating genome annotation for the microbial
59
cell factory Aspergillus niger using gene co-expression networks. Nucleic Acids Research, 47(2),
559–569. https://doi.org/10.1093/nar/gky1183
Schweppe, D. K., Zheng, C., Chavez, J. D., Navare, A. T., Wu, X., Eng, J. K., & Bruce, J. E. (2016).
XLinkDB 2.0: integrated, large-scale structural analysis of protein crosslinking data.
Bioinformatics , 32(17), 2716–2718. https://doi.org/10.1093/bioinformatics/btw232
Steigenberger, B., Pieters, R. J., Heck, A. J. R., & Scheltema, R. A. (2019). PhoX: An IMAC-
Enrichable Cross-Linking Reagent. ACS Central Science, 5(9), 1514–1522.
https://doi.org/10.1021/acscentsci.9b00416
Tang, X., & Bruce, J. E. (2010). A new cross-linking strategy: protein interaction reporter (PIR)
technology for protein-protein interaction studies. Molecular bioSystems, 6(6), 939–947.
https://doi.org/10.1039/b920876c
The, M., MacCoss, M. J., Noble, W. S., & Käll, L. (2016). Fast and Accurate Protein False Discovery
Rates on Large-Scale Proteomics Data Sets with Percolator 3.0. Journal of the American Society
for Mass Spectrometry, 27(11), 1719–1727. https://doi.org/10.1007/s13361-016-1460-7
Trnka, M. J., Baker, P. R., Robinson, P. J. J., Burlingame, A. L., & Chalkley, R. J. (2014). Matching
Cross-linked Peptide Spectra: Only as Good as the Worse Identification *. Molecular & Cellular
Proteomics: MCP, 13(2), 420–434. https://doi.org/10.1074/mcp.M113.034009
von Appen, A., Kosinski, J., Sparks, L., Ori, A., DiGuilio, A. L., Vollmer, B., Mackmull, M.-T.,
Banterle, N., Parca, L., Kastritis, P., Buczak, K., Mosalaganti, S., Hagen, W., Andres-Pons, A.,
Lemke, E. A., Bork, P., Antonin, W., Glavy, J. S., Bui, K. H., & Beck, M. (2015). In situ
structural analysis of the human nuclear pore complex. Nature, 526(7571), 140–143.
https://doi.org/10.1038/nature15381
Walzthoeni, T., Claassen, M., Leitner, A., Herzog, F., Bohn, S., Förster, F., Beck, M., & Aebersold,
R. (2012). False discovery rate estimation for cross-linked peptides identified by mass
spectrometry. Nature Methods, 9(9), 901–903. https://doi.org/10.1038/nmeth.2103
Wang, G., Wu, W. W., Zhang, Z., Masilamani, S., & Shen, R.-F. (2009). Decoy methods for assessing
false positives and false discovery rates in shotgun proteomics. Analytical Chemistry, 81(1),
146–159. https://doi.org/10.1021/ac801664q
Wichmann, C., Meier, F., Virreira Winter, S., Brunner, A.-D., Cox, J., & Mann, M. (2019).
MaxQuant.Live Enables Global Targeting of More Than 25,000 Peptides. Molecular & Cellular
Proteomics: MCP, 18(5), 982–994. https://doi.org/10.1074/mcp.TIR118.001131
Yugandhar, K., Wang, T.-Y., Leung, A. K.-Y., Lanz, M. C., Motorykin, I., Liang, J., Shayhidin, E. E.,
Smolka, M. B., Zhang, S., & Yu, H. (2020). MaXLinker: Proteome-wide Cross-link
Identifications with High Specificity and Sensitivity. Molecular & Cellular Proteomics: MCP,
19(3), 554–568. https://doi.org/10.1074/mcp.TIR119.001847
Yugandhar, K., Wang, T.-Y., Wierbowski, S. D., Shayhidin, E. E., & Yu, H. (2020). Structure-based
validation can drastically underestimate error rate in proteome-wide cross-linking mass
spectrometry studies. Nature Methods, 17(10), 985–988. https://doi.org/10.1038/s41592-020-
0959-9
60
Supplement
Supporting Information:
In-Search Assignment of Monoisotopic Peaks Improves the
Identification of Cross-Linked Peptides
Swantje Lenz1, Sven H. Giese1, Lutz Fischer2, and Juri Rappsilber∗1, 2
1Bioanalytics, Institute of Biotechnology, Technische Universit¨at Berlin, 13355
Berlin, Germany
2Wellcome Centre for Cell Biology, School of Biological Sciences, University of
Edinburgh, Edinburgh EH9 3BF, United Kingdom
List of Supporting Information
Table S1. Precursor m/z of different processing methods of dataset 1
Table S2. Precursor m/z of different processing methods of dataset 2
Table S3. Precursor m/z of different processing methods of dataset 3
Supporting Information: MS1 based mass range reduction
Table S4. Xi-MPA mass range reduction.
Figure S1: OpenMS preprocessing workflow
Figure S2: Performance of Xi-MPA on EThcD, CID, and ETciD acquisitions of the pseudo-complex
dataset
Table S5: Summary for conducted significance tests for Figure 3B in the main text
Figure S3: Performance of Xi-MPA on HCD data of the HSA dataset
Figure S4: Performance of Xi-MPA on the test fractions of the C. thermophilum dataset
Figure S5: Dependency of the monoisotopic mass correction on precursor intensity
Figure S6: Dependency of identifications in Xi-MPA on mass
∗juri.rappsilber@tu-berlin.de
S-1
MS1 based mass range reduction
Since considering multiple precursor masses increases search time, we developed an approach to
reduce the range searched in Xi-MPA. For each MS2 spectrum, the precursor peak is identified in
the corresponding MS1 spectrum. Then, the most abundant occurrence of this peak is searched
in a retention time window of 20 seconds (±10 seconds) and the corresponding MS1 spectrum is
extracted. Assuming that the assigned charge state was correct, the newly extracted MS1 spectrum
is searched for the true monoisotopic peak, i.e. lighter isotope peaks than the one that was reported
in the MS2 spectrum. According to Table S4 each MS2 spectrum gets assigned an individual range
of possible precursor masses to be used during Xi-MPA.
Table S4: Xi-MPA mass range reduction.
Lighter Peaks Present Search Range
none mass range without lightest mass*
continuous (without gaps) lightest two peaks found
single peaks (with gaps) mass range up to lightest peak found
*In case the range this approach is done is only up to -2 Da, -1 Da
will still be searched here.
This approach was evaluated on the first and last fractions of dataset 3 with a mass range of
up to -4 Da. On average, the masses searched in Xi-MPA were reduced by 24% per file, while the
number of within PSMs is 97% of the search without range reduction.
Note that this approach increases the time of the preprocessing before search to some extent.
Therefore, it is only worthwhile undertaking for searches with a large database, for which the time
of the search itself is long.
In previous approaches we tried to incorporate a mass and / or intensity cutoff or dependency
when selecting the considered mass range. However, the applied heuristics resulted in significant
losses in PSM numbers, presumably because a clear cut in the distributions is missing (Figure 4A
and S5).
Python scripts were written using the pyopenms package [1] and are available under
https://github.com/Rappsilber-Laboratory/Xi-MPA scripts.
S-2
Figure S1: OpenMS preprocessing workflow. PeakPickerHiRes was used with the following set-
tings: ’ms levels’ was set to 1 and ’signal to noise’ set to 0 (disabled). For the tool FeatureFinder-
Centroided the following settings were changed: ’feature:min score’ - 0.6, ’mass trace:min spectra’
- 7, ’isotopic pattern:charge low’ - 3, and ’isotopic pattern:charge high’ - 7. In HighResPrecur-
sorMassCorrector ’feature:rt tolerance’ was changed to 15. SpectraFilterWindowMower was used
with ’movetype’ - jump, ’windowsize’ - 100, and ’peakcount’ 20.
test statistic p-value H1Data significant
35196.5 2.54E-29 less 0 vs. ref True
36626.5 4.68E-27 less -1 vs. ref True
38792.5 6.30E-32 less -2 vs. ref True
39451.0 4.66E-25 less -3 vs. ref True
28239.5 4.78E-19 less -4 vs. ref True
Table S5: Summary for conducted significance tests for Fig. 3B
in the main text. The significance level αwas set to 0.05 before
the statistical analysis. The Wilcoxon rank sum test with con-
tinuity correction was used in R. Abbreviations: ref - reference
distance distribution derived from all cross-linkable residues. 0,
-1, -2, -3 and -4 denote the subsets of PSMs with the correspond-
ing mass shift of the precursor.
S-3
Figure S2: Performance of Xi-MPA on EThcd, CID, and ETciD acquired data of the pseudo-
complex dataset. Different ranges in Xi-MPA were tested and evaluated on the number of PSMs.
Shown is the mean fold change of the respective setting to the number of PSMs from unprocessed
data. For all fragmentation methods, the number of identifications increases compared to the
unprocessed data. While for EThcD 251 PSMs were identified for the unprocessed data, 434 PSMs
resulted from MaxQuant-Xi and 542 PSMs from Xi-MPA with up to -4 Da. Numbers of identified
PSMs for CID data are: 265 PSMs for unprocessed, 552 for MaxQuant-Xi, and 753 for Xi-MPA
(-4 Da). Finally, 197 PSMs are identified in unprocessed data for ETciD, while 340 resulted from
MaxQuant-Xi and 502 from Xi-MPA. Although the increase of Xi-MPA is smaller for EThcD and
ETciD data than for CID and HCD data, it is the approach with the most identifications.
S-4
Figure S3: Performance of Xi-MPA on HCD acquisitions of the HSA dataset. The dashed line
equals a fold change of 1, meaning the same number of PSMs as in the unprocessed data was
identified. Different ranges of Xi-MPA were tested and compared to MaxQuant-Xi results. While
the latter led to 1127 PSMs, Xi-MPA with up to -4 Da resulted in 1816 identifications.
Figure S4: Performance of Xi-MPA on the first and last fraction of the C. thermophilum dataset.
As for the other two datasets, different ranges of Xi-MPA were compared to MaxQuant-Xi results.
MaxQuant led to 2966 identifications, while Xi-MPA with -4 Da identified 4013 PSMs. Considering
masses up to -3 Da led to a similar number of PSMs (3945) than up to -4 Da.
S-5
Figure S5: Dependency of the monoisotopic mass correction on precursor intensity. Scans of
the pseudo-complex dataset identified in the Xi-MPA search were evaluated regarding their mass
correction. Corrections to lighter masses occur more often for precursors with lower intensity.
Significance is denoted by asterisks (ns: p-value>0.05, *: p-value<0.05, ***: p-value<0.001).
Figure S6: Correction of the monoisotopic mass is more successful for lighter peptides, while Xi-
MPA identifies larger peptides more often. Precursor masses of scans identified in all approaches
(preprocessing in MaxQuant and OpenMS and Xi-MPA) are compared to scans solely identified
in Xi-MPA. (****: p-value<0.0001)
S-6
References
[1] Hannes L. R¨ost, Uwe Schmitt, Ruedi Aebersold, and Lars Malmstr¨om. pyopenms: A python-
based interface to the openms mass-spectrometry algorithm library. PROTEOMICS, 14(1):74–
77, 2014.
S-7
Supplementary Materials for
Improved peptide backbone fragmentation is the primary advantage of
MS-cleavable crosslinkers
Lars Kolbowski*,1, Swantje Lenz*,1, Lutz Fischer1, Ludwig R Sinn1, Francis J O’Reilly1, Juri
Rappsilber†,1,2
1Bioanalytics, Institute of Biotechnology, Technische Universität Berlin, 13355 Berlin, Germany
2Wellcome Centre for Cell Biology, University of Edinburgh, Edinburgh EH9 3BF, UK
*These authors contributed equally to this work.
†Corresponding author email address: [email protected]
This file includes:
Supplementary Figures 1-6
1
Figure S1. Ratio of identified target-target (TT) CSMs that contain at least one peptide stub
peak for one (lighter colour) or both (darker colour) crosslinked peptides across datasets (5%
CSM-level FDR).
Figures S2. Ratio of identified heteromeric target-target (TT) CSMs that contain one (lighter
colour) or both (darker colour) peptide doublets across datasets (5% CSM-level FDR).
2
Figure S3. Score distribution of heteromeric matches in the Ribosome HCD dataset. Shown is
the distribution of targets and target-decoy matches with and without filtering for peptide
doublets. Arrows show the resulting score cutoffs at 5% FDR.
Figure S4. Score distribution of heteromeric matches in the Ribosome CID dataset. Shown is
the distribution of targets and target-decoy matches with and without filtering for peptide
doublets. Arrows show the resulting score cutoffs at 5% FDR.
3
Figure S5. Score distribution of heteromeric matches in the Synapse dataset. Shown is the
distribution of targets and target-decoy matches with and without filtering for peptide doublets.
Arrows show the resulting score cutoffs at 5% FDR.
Figure S6. Number of self-CSMs passing 5% CSM-level FDR for BS3 and DSSO. DSSO was
additionally searched as a non-cleavable crosslinker and filtered for the presence of peptide
doublets.
4
1
Supplementary Materials for
Reliable identification of protein-protein interactions by crosslinking mass
spectrometry
Swantje Lenz1*, Ludwig R. Sinn1*, Francis O’Reilly1*, Lutz Fischer1, Fritz Wegner1, Juri
Rappsilber1,2
1 Bioanalytics, Institute of Biotechnology, Technische Universität Berlin, 13355 Berlin, Germany
2 Wellcome Centre for Cell Biology, University of Edinburgh, Edinburgh EH9 3BF, UK
*These authors contributed equally.
Correspondence to: Juri.[email protected]
This file includes:
Supplementary Table 1
Supplementary Figures 1-10
2
Supplementary Table 1: Recent large scale crosslinking mass spectrometry studies and
aspects of their FDR method
Sample
Search / FDR
Software
Self- / heteromeric
crosslinks separated
for FDR
FDR level
threshold
reference
Murine synaptosomes
XlinkX / PD
no
CSM
2%
1
Saccharomyces
cerevisiae nucleus
XlinkX / PD
no
CSM
1%
2
Saccharomyces
cerevisiae
mitochondria
pLink1
no
CSM
1%
3
Several previously
published datasets
pLink2
yes
CSM
5%
4
Drosophila
melanogaster embryo
lysate
MeroX
yes
CSM
1%
5
Human cells
Comet / XLinkProphet
yes
Peptide pair
1%
6
Human cell lysate
MaxLinker
no
Peptide pair
1%
7
Human cell lysate
XlinkX / PD
no
Peptide pair
1%
8
Saccharomyces
cerevisiae
mitochondria
Kojak
yes
Peptide pair
2%
9
Human cell lysate
xiSEARCH / xiFDR
yes
Residue pair
5%
10
Human mitochondria
xiSEARCH / xiFDR
yes
Residue pair
5%
11
Mycoplasma
pneumoniae cells
xiSEARCH / xiFDR
yes
Residue pair & PPI
5%
12
3
Supplementary Figure 1: Non-crosslinkable control definition.
a Illustration of the distribution of decoy matches for self and heteromeric crosslinks. If
considered together for FDR calculation, heteromeric decoy matches will be matched more
frequently and make up most of the summed decoy matches. If subsequently heteromeric
matches are evaluated separately (e.g. for reporting PPIs), their error will be larger than the
previously calculated FDR (which would only be correct for the data as a whole). b Abundance
distribution of proteins identified with heteromeric crosslinks at a generous 10% heteromeric
peptide pair FDR (best lower iBAQ, see Methods). Protein pairs are accepted as plausible if both
proteins reach the 5th percentile in the same fraction, otherwise the pair is defined as ‘non-
crosslinkable’. 544,274 (6% of all possible PPIs in the E. coli proteome of 4350 proteins) PPIs
are defined as plausible (Supplementary data 7), while 8,914,801 (94%) PPIs are non-
crosslinkable. The dashed line represents the chosen 5th-percentile cutoff of 4.3E6. c Example
of proteins defined either as plausible to crosslink to NusA (GreB, dark grey) or “non-
crosslinkable” (AtpF, red). Although AtpF is present in some fractions together with NusA, it is
too low abundant in those, therefore the proteins are defined as "non-crosslinkable". In contrast,
GreB reaches the cutoff in the same fractions with NusA, therefore the two are crosslinkable.
Source data are provided as a Source Data file.
4
Supplementary Figure 2: Observed PPI-level error using a 5% FDR cutoff at different
information levels applied to entrapment, wrong crosslinker, wrong precursor mass and
non-crosslinkable controls.
Each bar represents a separate FDR calculation performed at different information levels on
each of the controls. The y-axis displays the observed PPI-level errors of the respective controls
employing a 5% a naive or b heteromeric FDR thresholds on different information levels. Bars
represent the mean error, with individual data points shown on top (for each crosslinker dataset
or entrapment database). Source data are provided as a Source Data file.
5
Supplementary Figure 3: Observed errors using the non-crosslinkable control on
information levels lower than PPI-level.
Each bar represents a separate FDR calculation performed at different information levels while
considering all CSMs (naive) or heteromeric crosslinks only (heteromeric). The y-axis displays
the observed errors of the respective information level at 5% FDR. Observed error in a identified
CSMs calculated on redundant or (unique) CSM-level, b identified peptide pairs calculated on
redundant CSM-level, CSM-level and peptide pair-level, c identified residue pairs calculated on
redundant CSM-level, CSM-level, peptide pair-level and residue pair-level, respectively.
Individual values from DSSO and BS3 datasets are depicted by dots; the dashed line indicates
the respective 5% FDR threshold. Source data are provided as a Source Data file.
6
Supplementary Figure 4. Results of positive controls.
a, b Fraction of coeluting PPIs (correlation coefficient > 0.5) among the heteromeric PPIs passing
a given FDR threshold, applying different published FDR approaches separated for a BS3 and
b DSSO. c, d Fraction of PPIs present in the STRING database (STRING combined score >=
150) among the PPIs passing a given FDR threshold, applying different published FDR
approaches (Supplementary Table 1) for c BS3 and d DSSO. Source data are provided as a
Source Data file.
7
Supplementary Figure 5: Properties of final crosslink PPI network.
a The overall heteromeric PPI counts and the respective fraction of false PPIs are shown for
varying FDR thresholds. Expected false PPIs were calculated based on decoy matches. For this
dataset increasing FDR thresholds up to 4% add more true than false matches. b Distribution of
residue pairs per PPI. c Abundance of identified crosslinked proteins in the respective categories.
Heteromeric-only proteins are significantly more abundant than all identified proteins (p = 0.044
using a one-sided Kolmogorov–Smirnov test). d Heteromeric-only proteins tend to be smaller
than the median of proteins in the database, and are therefore less likely to produce self-links.
Boxplots in c and d depict the median (middle line), upper and lower quartiles (boxes), 1.5 times
of the interquartile range (whiskers) as well as outliers (single points). Source data for panels c
and d are provided as a Source Data file.
8
Supplementary Figure 6: ATP synthase and Pyruvate dehydrogenase Crosslinking MS
subnetworks.
a PPI subnetwork of ATP synthase in xiNET13. Green shade on protein sequences illustrate
sequence areas covered by the PDB model (PDB 5T4O)14. b SEC-Coelution traces of ATP
synthase subunits. Note that all subunits also coelute in a very early fraction, probably containing
lipid vesicles. c Structural model of ATP synthase with mapped heteromeric crosslinks (PDB
5T4O). All protein chains are colored in grey. Heteromeric links are below 35 Å and are colored
in blue. d PPI subnetwork of pyruvate dehydrogenase and 2-oxoglutarate complexes with
collapsed protein nodes and e with proteins shown as bars in xiNET13. f Coelution of pyruvate
dehydrogenase / 2-oxoglutarate complex components showing lpdA eluting with both.
9
Supplementary Figure 7: MukBEF complex and DNA gyrase Crosslinking MS
subnetworks.
a PPI subnetwork of MukBEF complex with proteins shown as bars in xiNET13. b Coelution traces
of MukBEF complex with its binder AcpP (acpP). The MukBEF complex (fraction 13) dissociated
into another MukBEF assembly with lower occupancy for MukB (fraction 19) as judged by
coelution. The interactor AcpP is a known binder of MukBEF that could be confirmed by
Crosslinking MS and coelution. c Crosslink distribution on MukBEF protein sequences compared
to a recent in-vitro study15. Crosslinks are colored in blue if shared between the studies, in grey
for in-vitro data only and in green if unique to this study. d PPI subnetwork of DNA gyrase in
xiNET13. Green shade on protein sequences illustrate sequence areas covered by the PDB
model (PDB 6RKW) 16. Heteromeric links are colored in blue if below 35 Å, red when exceeding
and grey when absent from the model used. e Coelution of DNA gyrase complex components
with its inhibitor YacG and phosphate acetyltransferase Pta. The abundances for GyrA, GyrB
and YacG were magnified 10-fold. f Structural model of DNA gyrase with mapped heteromeric
crosslinks (PDB 6RKW) 16. GyrA is shown in dark green and GyrB in light green. DNA is shown
in grey. For crosslink coloration see panel d. Of note, the DNA gyrase inhibitor YacG was
suggested to associate with proteins involved in coenzyme A metabolism such as Pta17, which
is supported by this analysis.
10
Supplementary Figure 8: Novel identified PPIs.
a, c, e, h Selected heteromeric PPI networks with proteins shown as bars or g in a network
diagram13 b, d, f, i with their corresponding SEC elution traces. Coloration in panels a, c, e
represent: transmembrane domains for elaB network; PDZ domains for degP and degQ (pink);
coiled-coil domain (green) for ubiK and SCP2 domain for ubiJ (pink). The abundance for ubiJ in
11
panel f was magnified 5-fold; abundances of GroEL interactors in panel i were magnified 10-fold.
High-confidence binders of GroEL, displayed with darker nodes in panel g, were selected for the
panels h & i. h Selected binders of GroEL shown as bars13. Green areas mark GroEL’s aligned
region with a structural model (PDB 4PKO)18. Lysines on the inside of GroEL are highlighted with
an orange circle, the ones on the outside with a circle in light yellow. i SEC coelution trace of
GroEL and selected binders. j GroEL structural model (PDB 4PKO) 18 with crosslinked residues
highlighted. One ring of the GroEL barrel assembly is shown with one subunit of GroEL colored
in pink. Crosslinked lysine residues are indicated as spheres and colored as in panel h.
12
13
Supplementary Figure 9: Selected RNA polymerase binders.
a, d, g, j, m Collected experimental support for heteromeric PPIs between RNAP and selected
binders - rapA, greB, nusG, nusA and rpoD - is shown from SEC elution profiles, b, e, h, k, n
Crosslinking MS-based PPI screening in xiNET13 and c, f, i, l, o matching to existing structural
PPI models. The abundance of greB was magnified 50-fold. Green shade on protein sequence
bars illustrate areas covered by PDB models that were used to map heteromeric links onto
structures. In protein crosslink diagrams and protein structural models, heteromeric crosslinks
satisfying euclidean distance thresholds of 35 Å are shown in blue while longer links are colored
in red. The structural models used were: rapA (PDB 4S20)19, greB (PDB 6RIN) 20, nusG (PDB
5MS0) 21, nusA (PDB 6FLQ) 22 and rpoD (PDB 4ZH3) 23. If present in the model, DNA is colored
in grey and RNA in black. Labeled spheres in panel c mark lysine residues crosslinked to C-
termini of RNAP’s α-subunits (absent from this model).
14
15
Supplementary Figure 10: AP-MS and sulfo-SDA crosslinking.
Enrichment analysis from affinity-enriched a YacL-SPA, c RpoB-SPA and e NusG-SPA with b,
d, f corresponding crosslink subnetworks from sulfo-SDA crosslinked eluates from these
enrichments. In the volcano plots, proteins of interest are labelled as follows: components of
RNA polymerase in orange; proteins found crosslinked to RNA polymerase from lysate SEC
fractionation in red; proteins found crosslinked to RNA polymerase from a given affinity-
enrichment experiment in blue; overlapping proteins between the two sample workflows in red
and blue; all other proteins are indicated as grey crosses. In the crosslink subnetworks generated
with xiNET13, RNA polymerase constituents are labelled in orange and SPA-tagged proteins are
highlighted in blue. YacL is shown in expanded sequence view. The thickness of the lines
represents the number of unique residue pairs between the proteins.
16
Supplementary References
1. Gonzalez-Lozano, M. A. et al. Stitching the synapse: Cross-linking mass spectrometry
into resolving synaptic protein interactions. Sci Adv 6, eaax5783 (2020).
2. Bartolec, T. K. et al. Cross-linking Mass Spectrometry Analysis of the Yeast Nucleus
Reveals Extensive Protein-Protein Interactions Not Detected by Systematic Two-Hybrid
or Affinity Purification-Mass Spectrometry. Anal. Chem. 92, 1874–1882 (2020).
3. Linden, A. et al. A cross-linking mass spectrometry approach defines protein
interactions in yeast mitochondria. Mol. Cell. Proteomics (2020)
doi:10.1074/mcp.RA120.002028.
4. Chen, Z.-L. et al. A high-speed search engine pLink 2 with systematic evaluation for
proteome scale identification of cross-linked peptides. Nat. Commun. 10, 3404 (2019).
5. Götze, M., Iacobucci, C., Ihling, C. H. & Sinz, A. A Simple Cross-Linking/Mass
Spectrometry Workflow for Studying System-wide Protein Interactions. Anal. Chem. 91,
10236–10244 (2019).
6. Chavez, J. D., Keller, A., Zhou, B., Tian, R. & Bruce, J. E. Cellular Interactome
Dynamics during Paclitaxel Treatment. Cell Rep. 29, 2371–2383.e5 (2019).
7. Yugandhar, K. et al. MaXLinker: Proteome-wide Cross-link Identifications with High
Specificity and Sensitivity. Mol. Cell. Proteomics 19, 554–568 (2020).
8. Steigenberger, B., Pieters, R. J., Heck, A. J. R. & Scheltema, R. A. PhoX: An IMAC-
Enrichable Cross-Linking Reagent. ACS Cent Sci 5, 1514–1522 (2019).
9. Makepeace, K. A. T. et al. Improving Identification of In-organello Protein-Protein
Interactions Using an Affinity-enrichable, Isotopically Coded, and Mass Spectrometry-
cleavable Chemical Crosslinker. Mol. Cell. Proteomics 19, 624–639 (2020).
10. Mendes, M. L. et al. An integrated workflow for crosslinking mass spectrometry. Mol.
Syst. Biol. 15, e8994 (2019).
11. Ryl, P. S. J. et al. In Situ Structural Restraints from Crosslinking Mass Spectrometry in
17
Human Mitochondria. J. Proteome Res. (2019) doi:10.1021/acs.jproteome.9b00541.
12. O’Reilly, F. J. et al. In-cell architecture of an actively transcribing-translating
expressome. Science 369, 554–557 (2020).
13. Combe, C. W., Fischer, L. & Rappsilber, J. xiNET: cross-link network maps with residue
resolution. Mol. Cell. Proteomics 14, 1137–1147 (2015).
14. Sobti, M. et al. Cryo-EM structures of the autoinhibited E. coli ATP synthase in three
rotational states. Elife 5, (2016).
15. Bürmann, F. et al. A folded conformation of MukBEF and cohesin. Nat. Struct. Mol. Biol.
26, 227–236 (2019).
16. Vanden Broeck, A., Lotz, C., Ortiz, J. & Lamour, V. Cryo-EM structure of the complete
E. coli DNA gyrase nucleoprotein complex. Nat. Commun. 10, 4935 (2019).
17. Vos, S. M. et al. Direct control of type IIA topoisomerase activity by a chromosomally
encoded regulatory protein. Genes Dev. 28, 1485-1497 (2014).
18. Fei, X., Ye, X., LaRonde, N. A. & Lorimer, G. H. Formation and structures of
GroEL:GroES2 chaperonin footballs, the protein-folding functional form. Proc. Natl.
Acad. Sci. U. S. A. 111, 12775–12780 (2014).
19. Liu, B., Zuo, Y. & Steitz, T. A. Structural basis for transcription reactivation by RapA.
Proc. Natl. Acad. Sci. U. S. A. 112, 2006–2010 (2015).Asda
20. Abdelkareem, M. ’men et al. Structural Basis of Transcription: RNA Polymerase
Backtracking and Its Reactivation. Mol. Cell 75, 298–309.e4 (2019).
21. Said, N. et al. Structural basis for λN-dependent processive transcription
antitermination. Nat Microbiol 2, 17062 (2017).
22. Guo, X. et al. Structural Basis for NusA Stabilized Transcriptional Pausing. Mol. Cell 69,
816–827.e4 (2018).
23. Feng, Y. et al. Structural Basis of Transcription Inhibition by CBR Hydroxamidines and
CBR Pyrazoles. Structure 23, 1470–1481 (2015).
