Beyond the biosynthetic gene cluster paradigm: Genome-wide coexpression networks connect clustered and unclustered transcription factors to secondary metabolic pathways [original]

Beyond the Biosynthetic Gene Cluster Paradigm: Genome-Wide

Coexpression Networks Connect Clustered and Unclustered

Transcription Factors to Secondary Metabolic Pathways

Min Jin Kwon,

Charlotte Steiniger,

Timothy C. Cairns,

Jennifer H. Wisecaver,

Abigail L. Lind,

Carsten Pohl,

Carmen Regner,

Antonis Rokas,

Vera Meyer

Chair of Applied and Molecular Microbiology, Institute of Biotechnology, Technische Universität Berlin, Berlin, Germany

Department of Biochemistry, Center for Plant Biology, Purdue University, West Lafayette, Indiana, USA

Department of Biological Sciences, Vanderbilt University, Nashville, Tennessee, USA

Gladstone Institute for Data Science and Biotechnology, San Francisco, California, USA

Department of Biomedical Informatics, Vanderbilt University School of Medicine, Nashville, Tennessee, USA

ABSTRACT Fungal secondary metabolites are widely used as therapeutics and are vital

components of drug discovery programs. A major challenge hindering discovery of novel

secondary metabolites is that the underlying pathways involved in their biosynthesis are

transcriptionally silent under typical laboratory growth conditions, making it difﬁcult to

identify the transcriptional networks that they are embedded in. Furthermore, while the

genes participating in secondary metabolic pathways are typically found in contiguous

clusters on the genome, known as biosynthetic gene clusters (BGCs), this is not always

the case, especially for global and pathway-speciﬁc regulators of pathways’activities. To

address these challenges, we used 283 genome-wide gene expression data sets of the

ascomycete cell factory Aspergillus niger generated during growth under 155 different

conditions to construct two gene coexpression networks based on Spearman’scorrela-

tion coefﬁcients (SCCs) and on mutual rank-transformed Pearson’s correlation coefﬁ-

cients (MR-PCCs). By mining these networks, we predicted six transcription factors,

namedMjkAtoMjkF,toregulatesecondarymetabolisminA. niger. Overexpression

of each transcription factor using the Tet-On cassette modulated the production of

multiple secondary metabolites. We found that the SCC and MR-PCC approaches

complemented each other, enabling the delineation of putative global (SCC) and

pathway-speciﬁc (MR-PCC) transcription factors. These results highlight the

potential of coexpression network approaches to identify and activate fungal sec-

ondary metabolic pathways and their products. More broadly, we argue that drug

discovery programs in fungi should move beyond the BGC paradigm and focus

on understanding the global regulatory networks in which secondary metabolic

pathways are embedded.

IMPORTANCE There is an urgent need for novel bioactive molecules in both agricul-

ture and medicine. The genomes of fungi are thought to contain vast numbers of met-

abolic pathways involved in the biosynthesis of secondary metabolites with diverse bio-

activities. Because these metabolites are biosynthesized only under speciﬁc conditions,

the vast majority of the fungal pharmacopeia awaits discovery. To discover the genetic

networks that regulate the activity of secondary metabolites, we examined the genome-

wide proﬁles of gene activity of the cell factory Aspergillus niger across hundreds of con-

ditions. By constructing global networks that link genes with similar activities across con-

ditions, we identiﬁed six putative global and pathway-speciﬁc regulators of secondary

metabolite biosynthesis. Our study shows that elucidating the behavior of the genetic

networks of fungi under diverse conditions harbors enormous promise for understand-

ing fungal secondary metabolism, which ultimately may lead to novel drug candidates.

Citation Kwon MJ, Steiniger C, Cairns TC,

Wisecaver JH, Lind AL, Pohl C, Regner C, Rokas

A, Meyer V. 2021. Beyond the biosynthetic

gene cluster paradigm: genome-wide

coexpression networks connect clustered and

unclustered transcription factors to secondary

metabolic pathways. Microbiol Spectr 9:

e00898-21. https://doi.org/10.1128/Spectrum

.00898-21.

Editor Gustavo H. Goldman, Universidade de

Sao Paulo

access article distributed under the terms of

the Creative Commons Attribution 4.0

International license.

Address correspondence to Vera Meyer,

[email protected].

Received 29 July 2021

Accepted 30 July 2021

Published 15 September 2021

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 1

RESEARCH ARTICLE

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

KEYWORDS ﬁlamentous fungi, Aspergillus niger, secondary metabolite gene clusters,

gene coexpression, correlation network, natural product, specialized metabolism,

genetic network, gene regulation

Fungal secondary metabolites (SMs) are bioactive, usually low-molecular-weight com-

pounds that have restricted taxonomic distribution and are produced at speciﬁcstagesof

growth and development (1). The most well-known clinical applications of these molecules

include their use as antibiotics, cholesterol-lowering agents, and immunosuppressants (e.g.,

penicillin, statins, and cyclosporins, respectively) (2). However, they also play an important

role in drug discovery programs, with recently marketed therapeutics consisting of either

fungal SMs or their semisynthetic derivatives (3). In contrast to these contributions to human

welfare, fungal SMs also include potent carcinogenic crop contaminants (4), and the myco-

toxin-producing capacity of commonly used fungal cell factories in food or biotechnological

processes is often either unknown (5) or underestimated (6). Moreover, plant-infecting fungi

deploy numerous SMs as virulence factors that facilitate successful infection (7), ultimately

destroying enough food for 10% of the human population per year (8). Improved under-

standing of the genetic, molecular, and biochemical aspects of fungal secondary metabolism

thus promises to drive novel medical breakthroughs, while also ensuring improvements in

global food safety and security (9).

A common feature of SM-producing fungi is that the genes required for producing

a single secondary metabolite are often found in contiguous clusters on the genome,

which may facilitate both horizontal gene transfer of SMs and epigenetic regulation via

chromatin remodeling (1, 10). Biosynthetic gene clusters (BGCs) typically include a

gene encoding a core biosynthetic enzyme, most commonly a nonribosomal peptide

synthetase (NRPS), polyketide synthase (PKS), or terpene cyclase, that is responsible for

the ﬁrst metabolic step in product synthesis (11). Additionally, BGCs include genes

encoding so-called “tailoring”enzymes, such as P450 monooxygenases or methyltrans-

ferases, that modify the molecule produced by the core enzyme (11, 12). Moreover,

many BGCs contain either genes encoding putative membrane transporters, which are

required for metabolite efﬂux from the cell in some (13) but not all (14) cases, or addi-

tional so-called “resistance”genes, which are necessary for detoxiﬁcation/self-protec-

tion against the molecules produced (15).

Most BGCs are transcriptionally silent under standard laboratory and industrial culti-

vation conditions, which is a major challenge to the discovery of their cognate mole-

cules (16). Interestingly, many BGCs also contain transcription factor (TF)-encoding

genes that regulate their activity (11, 12, 17). In several instances, these TF-encoding

genes have been overexpressed to activate transcription of the respective BGC, ulti-

mately leading to the discovery of novel SMs (13, 18–21). However, this strategy cannot

be used for the approximately 40% of fungal BGCs that lack a resident TF (17).

An alternative approach to engineering SM-overproducing isolates has been to

identify and genetically target global regulators of multiple BGCs. These include epige-

netic regulators, notably components of the heterotrimeric velvet complex, which links

development, light responses, and SM production in ascomycetes (22). Alternatively,

globally acting TFs that coordinate SM biosynthesis with differentiation (e.g., BrlA/

StuA) and responses to environmental stimuli, such as pH (PacC) or nitrogen availabil-

ity (AreA), can be activated using molecular approaches for elevated natural product

biosynthesis (1, 17, 23). A limitation to these strategies, however, is that all global regu-

lators discovered to date activate only a fraction of the predicted BGCs in a single ge-

nome. For example, deletion of genes predicted to encode the methyltransferase

LaeA, which is thought to silence BGC expression by the formation of transcriptionally

silent heterochromatin, increased the expression of 7 of 17 BGCs in the biomass-

degrading fungus Trichoderma reesei and 13 of 22 BGCs analyzed in the human-infect-

ing mould Aspergillus fumigatus (24, 25).

Aﬁnal confounding factor in understanding and functionally analyzing fungal BGCs

and their products is that there is considerable variation in the degrees to which core,

Kwon et al.

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 2

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

tailoring, transport, and regulatory genes are contiguously clustered in fungal genomes (10).

This includes so-called “partial”clusters in which some genes encoding biosynthetic

enzymes and transporters are not physically linked with other clustered genes (26, 27),

“superclusters”in which two or more NRPS-/PKS-encoding genes reside in close

physical proximity (28, 29), and SM-biosynthetic genes that are not contiguously

clustered (30).

Consequently, innovative strategies are required both to discover novel transcrip-

tional activators of BGCs and to accurately delineate their boundaries. Over the past

several years, an approach that has gained considerable interest has been the utiliza-

tion of coexpression networks to analyze BGCs, for example, during laboratory culture

of industrial isolates (29, 31) or during infectious growth of plant-infecting fungi (32). A

limitation to these studies, however, was the relatively small number of conditions

tested (up to several dozen), which resulted in the inability to detect the transcriptional

activity of numerous BGCs. To overcome this limitation, we recently conducted a meta-

analysis of 283 microarray data sets covering 155 different cultivation conditions for

the biotechnologically exploited cell factory Aspergillus niger. This data collection cov-

ers a diverse range of environmental conditions and genetic perturbations and was

used to construct a global gene coexpression network based on Spearman’s correla-

tion coefﬁcient (SCC) (33). We found that 53 of the 81 predicted BGC core genes in A.

niger are expressed under at least 1 of the 155 conditions, and we were able to delin-

eate the boundaries of numerous BGCs, including, for example, the partial cluster

required for biosynthesis of the siderophore triacetylfusarinine C.

Our analysis also suggested that only a minority of BGCs are coexpressed with their

resident TF; speciﬁcally, from the 25 of the 53 expressed BGCs that contained a TF, only

8 BGCs were coexpressed with their respective TF. However, we were able to use this

network to successfully predict TFs that, independent of their physical location on the

genome, regulate multiple BGCs. This relied on the so-called “guilt-by-association”

principle, whereby genes that are part of similar (or the same) biosynthetic pathways

or genetic networks tend to have highly comparable patterns of gene expression. We

functionally analyzed two of these coexpressed TFs (MjkA and MjkB) (Table 1) by gen-

erating loss-of-function and gain-of-function A. niger mutants and could indeed dem-

onstrate that their overexpression modulated (either indirectly or directly) the tran-

scriptional activity of 45 (MjkA) and 43 (MjkB) BGC core genes (33).

Despite the utility of coexpression network analyses, there are several possible limi-

tations to the construction of transcriptional networks based on correlation coefﬁcients

TABLE 1 Selected list of transcription factors analyzed in this study that are coexpressed with BGCs in A. niger

Transcription

factor ORF Protein domain

No. of coexpressed

BGC core genes

based on: Clustered in

a BGC

Tet-On-based

overexpression

phenotype on solid

growth mediumSCC MR-PCC

MjkA An07g07370 Myb-like DNA-binding

domain

14 No Red pigment formation,

reduced growth,

irregular sclerotia

formation

MjkB An12g07690 Fungal Zn

-Cys

binuclear

cluster domain

13 No Red pigment formation

MjkC An01g14020 Fungal Zn

-Cys

binuclear

cluster domain

17 No Yellow pigment formation,

reduced growth

MjkD An07g02880 Fungal-speciﬁc

transcription factor

domain

10 No Yellow pigment formation

MjkE An08g11000 Fungal Zn

-Cys

binuclear

cluster domain

13 1 Yes (BGC 34) Brown pigment formation

MjkF An08g10880 Fungal Zn

-Cys

binuclear

cluster domain

15 1 Yes (BGC 34) Reduced growth, frequent

reversions

Beyond the Cluster Paradigm

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 3

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

like Spearman’s or Pearson’s. In these networks, correlation coefﬁcients are used as

weighted edges to connect genes (nodes). One major challenge when constructing

these networks is determining the edge weight threshold below which correlation

coefﬁcients are excluded from the network, with the goal being to remove nonbiologi-

cally relevant gene associations. We have previously used in silico data randomization

experiments to test the likely threshold of biologically meaningful coexpression based

on Spearman (33); however, it is still likely that for many BGCs, the correlation coefﬁ-

cient cutoff chosen (

$j0.5j) may be unnecessarily stringent, resulting in false-nega-

tive coexpression relationships for BGCs. Additionally, average correlation coefﬁcients

can vary by gene function and input data (34). Importantly, in the case of BGC genes

that are only expressed under a few or only one speciﬁc environmental condition, it is

likely that the expression vector for a given BGC gene will be sparse and, therefore,

more likely to artiﬁcially correlate with other rarely expressed genes rather than with

genes with a functional link.

To overcome these challenges, in this study, we reanalyzed the existing A. niger

transcriptome data set with a speciﬁc focus on A. niger BGCs. First, we generated gene

expression modules based on a mutual rank approach, which can capture functional

relationships for rarely expressed secondary metabolism genes (34, 35), as we have

previously shown in analyses of secondary metabolism in plants (36). We compared

this mutual rank strategy with our existing Spearman coexpression data sets and, by

integrating both approaches, generated a short list of six TF-encoding genes (including

mjkA and mjkB) that we hypothesized may regulate multiple BGCs. Functional analyses

of these genes by overexpression using the Tet-On gene switch revealed that they play

multiple roles in the growth, development, and pigment formation of A. niger as assayed

by standard growth tests on agar plates and in shake ﬂasks. Moreover, metabolomic proﬁl-

ing revealed a change in the metabolite patterns of the overexpression strains analyzed.

Finally, by in silico analysis, we generated a list of predicted molecules and associated them

with putative BGCs. The methods and resources developed in this study will thus enable

the efﬁcient activation of fungal SMs for novel drug discovery programs and other studies.

More broadly, our general approach holds potential for deciphering the global regulatory

network governing BGCs and secondary metabolic pathways in fungi.

RESULTS

Mining coexpression networks to identify biosynthetic and regulatory modules.

Using the SCC approach, we previously estimated the global transcriptional activities

of A. niger BGCs among the 283 microarray experiments by assessing the gene expression of

the predicted core enzymes (33). These data highlighted that BGC expression varies consider-

ably, with genes encoding core enzymes transcriptionally deployed frequently (several dozen

experiments), rarely (.5 experiments), or not at all (which was the case for 28 core genes)

(33). We reasoned that this microarray meta-analysis was also a promising resource for further

interrogation of BGCs using a mutual rank-transformed Pearson’s correlation coefﬁcient (MR-

PCC) approach (34, 36). Transforming PCCs into mutual ranks has been shown to improve the

recovery of known pathways as discrete subgraphs (i.e., modules) in global coexpression net-

works (35). We constructed three MR-PCC networks (NET25, NET10, and NET05), each using a

different coexpression threshold for assigning edge weights (i.e., associations) between nodes

(i.e., genes) in the network. We then used the graph-clustering method ClusterONE (37) to dis-

cover modules of coexpressed genes within the global networks. ClusterONE is unusual

among graph clustering methods, e.g., Markov clustering (MCL) (38), due to its ability to assign

genes to multiple overlapping modules, which is more reﬂective of complex biological net-

works. Networks were ordered on size (i.e., total number of edges between nodes) such that

NET25, the most relaxed coexpression threshold, represents the largest network with the larg-

est modules, whereas NET05, the most stringent threshold, represents the smallest network

with the smallest modules.

In total, 2,041 modules were recovered from the NET25 network, 2,944 modules from the

NET10 network, and 2,999 modules from the NET05 network (Table S1A in the supplemental

Kwon et al.

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 4

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

material). The median module sizes for the NET25, NET10, and NET05 networks were 11, 7,

and 5 genes, respectively. Of the 78 predicted BGCs comprising, in total, 81 core genes in the

A. niger genome, 43 predicted BGCs had one or more genes recovered within a single mod-

ule(TableS1B).These43BGCshadvariouslevelsofcoexpression,whichwedeﬁne here as

being assigned to at least one shared module. For some BGCs, such as the fumonisin-produc-

ing BGC, most genes in the gene cluster are coexpressed at high levels (Fig. 1A). For others,

FIG 1 Heatmaps depicting the Pearson’s correlations of coexpression of genes within three canonical BGCs. Across panels A-C,

gene ids within the canonical cluster are bolded in the heatmap and the corresponding gene arrow is colored red in the

accompanying depiction of the chromosome segment. Two ﬂanking genes are included on either side and corresponding

arrows are colored gray. Gene ids have been abbreviated. (A) A signiﬁcant fraction of genes within the fumonisin metabolic

gene cluster are coexpressed. (B) Coexpression of predicted BGC 34, which contains two transcription factors. Both gene ids

are colored green in the heatmap, and other clustered gene ids recovered in the metamodule are colored pink. (C) A small

fraction of genes within predicted BGC 38 are coexpressed. Genes ids are color coded in the heatmap as in panel A; gene ids

recovered in a metamodule are colored orange. (D) Network map of transcription factor metamodule containing all genes

coexpressed with both transcription factors across all three network analyses. Nodes in the map represent genes, and edges

connecting two genes represent the weight (transformed MR score) for the association. Transcription factors are colored

green. Other genes present in BGC 34 are colored pink. Genes present in BGC 38 are colored orange. All other genes are

colored gray.

Beyond the Cluster Paradigm

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 5

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

either a small subset of the genes in the BGC were not coexpressed (e.g., BGC 34) (Fig. 1B) or

only a small fraction of genes was coexpressed (e.g., BGC 38, where only 6 of 22 genes in the

BGC were coexpressed) (Fig. 1C). During our analysis, we noticed that the two putative TF

genes in BGC 34 were assigned to multiple overlapping modules due to the nature of the

ClusterONE approach. Many of these modules also included genes from BGC 38. We col-

lapsed all modules that contained these two predicted TF-encoding genes into one nonover-

lapping metamodule that, notably, contained 7 genes from BGC 38 and 10 genes from BGC

34 (Fig. 1D). This metamodule consisted of 50 genes in total, including 1 core gene (fatty acid

synthase [FAS]), 2 TFs from BGC 34 and 2 core genes (PKS and NRPS) from BGC 38.

Concordantly, we could also identify coexpression between BGC 34 and BGC 38 cluster mem-

bers via the SCC approach. Notably, MultiGeneBlast showed that BGC 34 and 38 are con-

served in black aspergilli (Fig. S1). Both clusters belong to a large SCC subnetwork comprised

of 1,804 genes (Fig. 2), which is the largest gene coexpression subnetwork with BGC genes

based on the Spearman rank coefﬁcient

$j0.5j. This subnetwork included many TFs that

are not physically located inside BGCs or are coexpressed with nonresident BGC genes.

It has been speculated over recent decades that BGC-resident TFs may coregulate

gene expression at more than one BGC (1, 17). Both coexpression network approaches

supported this hypothesis for A. niger, as evidenced by the coexpression of two TFs

residing in BGC 34 (open reading frames [ORFs] An08g11000 and An08g10880, chro-

mosome 1) with multiple genes in BGC 38 (chromosome 8), including the predicted

NRPS (Fig. 3). This was especially interesting given that (i) BGC 38 does not contain a

predicted TF, (ii) both these BGCs are present in 22 (BGC 34) or 24 (BGC 38) of 83 ana-

lyzed genomes of the genus Aspergillus, and (iii) BGC 38 is in close proximity to the

functionally characterized BGC 39 that is necessary for azanigerone production (39).

Interestingly, our analysis demonstrates that the SCC approach primarily carves out

coexpression of frequently expressed genes, whereas the strength of the MR-PCC

approach is the identiﬁcation of coexpression relationships among rarely expressed

genes. We thus decided to study the impact of six putative TF-encoding genes on A. ni-

ger secondary metabolism in more depth. Four were predicted by the SCC approach to

be coexpressed with at least 10 BGC core genes and are unclustered (MjkA to MjkD),

whereas the remaining 2 were predicted by the MR-PCC approach to be coexpressed

FIG 2 The largest Spearman subnetwork containing predicted BGC core and tailoring genes

(highlighted in pink), as well as transcription factors (highlighted in blue). The six transcription factors

studied by molecular analyses in this study (MjkA to -F) are indicated in green.

Kwon et al.

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 6

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

with both BGC 34 and BGC 38 and are clustered with BGC 34 (MjkE and MjkF) (Table 1

and Fig. 3).

Overexpression of predicted transcription factors MjkA to -F modulates A. niger

pigmentation and development. Prior to conducting gene functional analysis experi-

ments, we assessed the gene expression proﬁles for mjkA to mjkF across our 155 cultivation

conditions. While both mjkE and mjkF, which reside in BGC 34, were rarely expressed, the

four genes mjkA to mjkD encoding unclustered TFs were transcribed under numerous condi-

tions, with mjkA notably expressed to 90% of the level of A. niger actin under several condi-

tions (Fig. 4).

To assess the role of these TFs in modulating BGC expression, we generated condi-

tional expression isolates in which a Tet-On gene switch was placed upstream from the

open reading frame as previously described for the genes mjkA and mjkB (33) (Fig. S2

presents cloning information and Southern blot conﬁrmation). The Tet-On gene switch

has undetectable levels of basal expression in the absence of induction, and the addi-

tion of 10

g/ml doxycycline (DOX) enables expression above that of the A. niger glu-

coamylase gene, whose promoter is often used for overexpression studies (33, 40, 41).

Conditional expression isolates previously constructed for genes mjkA and mjkB were

also analyzed in this study to further assess their role in A. niger secondary metabolism

and development (Table S1C).

Standard growth assays on solid and in liquid media clearly identiﬁed differences in

medium pigmentation in overexpression isolates compared to the progenitor control (Fig. 5

and Fig. S3), suggesting a role of these genes in A. niger development and/or secondary me-

tabolism. The MjkA, MjkD, and MjkF conditional expression strains also displayed reduced

growth on solid agar under overexpression conditions (Fig. S3). Intriguingly, mjkA overex-

pression resulted in the infrequent formation of sclerotia (Fig. 5 and Fig. S3), which are an

important prerequisite for sexual development in Aspergillus (42, 43). The observed putative

sclerotia were highly similar in size, structure, and color to those recently characterized

from A. niger (44). However, A. niger sensu stricto has not been reported to have a sexual

cycle. Additionally, A. niger rarely produces sclerotia under speciﬁc growth conditions, which

is paralleled by the production of many secondary metabolites, including indolterpenes of

the aﬂavinine type (42). We thus reanalyzed transcriptomic data that were available for this

isolate and for the MjkB overexpression strain from bioreactor cultivation (33) to screen for

differential expression of developmental regulators following conditional MjkA and/or MjkB

FIG 3 Schematic representation of BGC 34 and BGC 38 as predicted by antiSMASH. Based on sequence similarity and

gene functional prediction, BGC 34 corresponds to the alkyl citrate-producing cluster identiﬁed in parallel to this study

in A. niger NRRL3 (54). BGC 38 is positioned next to the azanigerone cluster (39).

Beyond the Cluster Paradigm

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 7

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

expression. Strikingly, the expression of 36 and 27 regulators and TFs was affected when

mjkA or mjkB, respectively, was up- or downregulated (Fig. 6). Notably, the overexpression

of mjkA resulted in downregulation of genes encoding transcription factors known to con-

trol primary metabolism (creA,areB,xlnR,amyR,prtT,pacC,crzA,hapX,farA,farB,andacuB)

(45) and asexual development (brlA,abaA,stuA,ﬂbA,ﬂbB,andﬂbC)(45),aswellaschroma-

tin structure (laeA,velB,vipC,mtfA,andhdaA)(45),inAspergillus (Fig. 6). Deletion of mjkA

caused strong upregulation of the regulator-encoding genes areA,cpcA,msnA,csnE,ﬂbD,

and vosA (Table S1D), with functions in primary metabolism and development (45), imply-

ing that MjkA is a regulator of multiple A. niger BGCs, differentiation, and development.

Note that the MjkA-encoding gene can be found in 61 of 83 sequenced Aspergillus

genomes as identiﬁed by BLAST analyses (Table S2).

We further analyzed strains overexpressing genes mjkA to -Fin shake ﬂask cultures

following induction with DOX by quantitative PCR (qPCR) (Table S1E). All genes dis-

played elevated transcription following the addition of DOX to the growth medium,

with the exception of the mjkD overexpression strain, which we hypothesize could

potentially be due to an autoregulation phenomenon (Table S1E) (46). Notably, qPCR

analysis revealed that the overexpression of several putative TFs modiﬁed the expres-

sion of various known or predicted secondary metabolite-regulating genes, including

brlA,pacC, and mjkC in the isolate overexpressing mjkA, and mjkE and mjkF in the iso-

late overexpressing mjkC (Table S1E). Furthermore, overexpression of mjkF caused

marked upregulation of predicted core genes in cluster 34 (An08g10930 and An08g10860)

and cluster 38 (An09g01690), suggesting that the mjkF-encoded protein may indeed regu-

late the expression of these clusters. Taken together, qPCR experimentation supported

microarray-based coexpression analyses suggesting that genes mjkA,-B,-C,-E,and-Fmay

regulate the expression of secondary metabolite gene expression.

Overexpression of predicted transcription factors MjkA to -F modulates the

secondary metabolite proﬁle of A. niger.To understand the effects of the MjkA to -F

TFs on the secondary metabolite proﬁle of A. niger, we next conducted untargeted

metabolome analysis of the progenitor strain and mjkA to mjkF conditional expression

strains after 2, 4, or 10 days of incubation on minimal agar plates supplemented with

g/ml DOX (Fig. S3). For each overexpression strain, a single time point was

selected for metabolome analysis. Time points were chosen when the greatest devia-

tion in either medium pigmentation or growth relative to that of the control strain was

observed (Fig. S3). Since culture samples were harvested at both the center and the

outer edges of the growing colonies and pooled for analysis, the results obtained com-

prise metabolites from both old and young mycelia. This analysis detected a total of

FIG 4 Expression levels for all 6 TFs under 155 expression conditions. Note the different scales. mjkE (An08g11000) and mjkF

(An08g10880) expression levels are notably elevated during maltose-limited bioreactor growth in a DﬂbA mutant (77).

Kwon et al.

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 8

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

2,063 compounds, of which 1,835 were annotated. Metabolic pathway visualization of

the identiﬁed metabolites using iPATH showed that intermediates from various biosyn-

thetic routes toward SMs (Fig. S5, S6, and S7) were covered (47–49). Statistical analysis

(ttest) identiﬁed numerous metabolites that were signiﬁcantly different (P#0.05 and

log

ratio greater than 1 or 21) for the genotypes and time points compared (Fig. 7A).

Generally, overexpression of mjkC and mjkF (2 days) and of mjkA and mjkD (4 days)

each affected more than 140 metabolites compared to the levels in the control strain

at the respective time point (Fig. 7B). Interestingly, only overexpression of mjkC led to

an upregulation of more than half of the affected metabolites, whereas overexpression

of mjkA,mjkD, and mjkF led to downregulation relative to the levels in the control

(Fig. 7B). In comparison, overexpression of mjkB and mjkE (10 days) apparently affected

fewer metabolites (66 and 43, respectively), which might also be due to a reduced

overall metabolic activity of the cultures after prolonged cultivation.

Among the signiﬁcantly affected metabolites, several known SMs of A. niger and related

species (50) could be putatively identiﬁed by means of liquid chromatography-quadrupole

time of ﬂight high-resolution mass spectrometry (LC-QTOF-HRMS), based on mass and

retention time (Fig. 7C and Fig. S5 to S7). These compounds comprise naphtho-

-pyrones

(aurasperones, isonigerone, fonsecin, and carbonarins), bicoumarins (bicoumanigrin, kota-

nin, desmethylkotanin, and funalenone), and fumonisins. Moreover, overexpression of the

putative TFs affected meroterpenoids (1-hydroxyyanuthone A) and benzoquinone-type

pigments (atromentin and cycloleucomelone), as well as different types of alkaloids, such

as pyranonigrins, pyrophens (aspernigrin A, carbonarone A, and nygerone A), nigragillins

(nigragillin and nigerazine B), and tensidols. Not found among the signiﬁcantly affected

compounds were some known SMs of A. niger that have already been linked to their corre-

sponding BGCs, such as azanigerone (39), TAN-1612 (51), and ochratoxin (52).

Notably, the list of previously identiﬁed SMs of A. niger is almost exclusively comprised

of polyketide products (Fig. S6). Thus, even though the peptide-forming NRPS from BGC

38 (An09g01690) is present in a mutual rank metamodule with MjkE and MjkF, the biosyn-

thetic product of BGC 38 is unlikely to be one of the compounds identiﬁed in the current

study. Based on an in silico assembly line prediction using antiSMASH, An09g01690 enco-

des a bimodular NRPS that cannot be classiﬁed yet into a linear or iterative assembly type,

and its product is thus not predictable. Since it is coexpressed with two putative fatty acid

synthase-encoding genes (An09g01740 and An09g01750) in BGC 38 (Fig. 1 and 3), the

FIG 5 Tet-On-based overexpression of mjkA modiﬁes A. niger development. Overexpression of mjkA

induced by the addition of 10

g/ml doxycycline leads to irregular formation of putative sclerotia on

agar plates, an example of which is shown. Strains were grown on MM or CM for 144 h at 30°C in

the dark. Colony sectoring observed in this isolate is not due to formation of unstable heterokaryons,

as evidenced by PCR and Southern blot conﬁrmation of homokaryotic strains. Estimated scale bar

indicates approximately 1 mm.

Beyond the Cluster Paradigm

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 9

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

encoded peptide presumably features a fatty acid moiety of various lengths based on the

available fatty acid pool of A. niger. Similar patterns have been observed for other nonribo-

somally synthesized lipopeptides, such as daptomycin (53).

In parallel to this study, BGC 34 (Fig. 3) was recently demonstrated to be responsible

for alkyl citrate production in A. niger strain NRRL3 (54). For this SM class, a range of

bioactivities has been reported, including antiparasitic (55), antifungal (56), antibacte-

rial (57), and plant root growth promotion effects (58). Other complex alkyl citrates

(zaragozic acids, also called squalestatins) have been shown to be among the most

potent natural squalene synthase inhibitors (59, 60). Notably, the metabolome analysis

in this study showed that several alkyl citrates, such as hexylaconitic acid A, hexylita-

conic acid J, and tensyuic acids C and E, were also differentially produced at different

time points upon TF overexpression (Fig. 7C and Fig. S5 to S7).

DISCUSSION

This study has demonstrated that gene coexpression analysis enables the identiﬁca-

tion of fungal transcriptional networks in which secondary metabolite genes are em-

bedded. By comparing mutual rank and Spearman-derived coexpression networks, we

have identiﬁed, respectively, both BGC-resident and, additionally, unclustered TFs, a

ﬁnding that is broadly consistent with the existence of SM regulatory genes that reside

outside predicted BGC loci (17). However, there is a growing body of evidence to sug-

gest that, at least in some instances, there has been an overreliance on physical cluster-

ing for the prediction of SM pathway genes and their cognate transporters/regulators.

Indeed, with several notable exceptions (61, 62), it is still relatively rare that genes

required for the biosynthesis of an entire fungal SM are, ﬁrst, experimentally veriﬁed

and, second, fully contiguously clustered. Thus, the true extent of SM pathway gene

clustering in fungi remains unclear. This is further complicated by divergence in the

degrees to which the BGCs are intact across fungal genomes, which is even true for

FIG 6 Differential gene expression of transcription factors following overexpression of mjkA and mjkB

genes during controlled bioreactor batch cultivations of A. niger performed in our previous study

(33). Note that overexpression of MjkA strongly affects expression of predicted regulators during both

growth phases, whereas the effect of MjkB is limited to the post-exponential growth phase. ORF

names are given.

Kwon et al.

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 10

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

“gold standard”BGCs, such as those necessary for epipolythiodioxopiperazine synthe-

sis (e.g., gliotoxin/sirodesmin) (61). Hence, experimental approaches to activate and

functionally analyze the full fungal SM repertoire cannot exclusively rely on in silico

genomics approaches.

Given that coexpression approaches have only recently been applied to deﬁne fun-

gal BGC boundaries and their transcriptional networks (29, 31, 33, 63), in this study, we

examined the potential utility of two different approaches for constructing coexpres-

sion networks, namely, mutual rank and Spearman approaches. Our results suggest

that both approaches enable the delineation and reﬁnement of contiguous BGC boun-

daries. However, whereas the Spearman approach was better suited for the identiﬁca-

tion of global TFs, the mutual rank approach was better suited for the identiﬁcation of

pathway-speciﬁc TFs. This work should therefore guide future coexpression analyses of

other fungal transcriptional data sets based on the requirements of the end user (i.e.,

global or pathway-speciﬁc studies). Regarding the minimal number of transcriptomic

data sets necessary to generate useful coexpression resources, we argue that, given

FIG 7 Overexpression of mjkA to mjkF genes affects numerous metabolites in A. niger. (A) Annotated metabolites were

plotted by signiﬁcance (Pvalue) versus fold change (log

ratio). Metabolites reaching a Pvalue of ,0.05 are marked

orange. Metabolites with a Pvalue of ,0.05 and a log

ratio greater than 1 or 21 were considered signiﬁcant. (B)

Numbers of signiﬁcantly affected metabolites (Pvalue of ,0.05 and log

ratio greater than 1 or 21) in comparison to

their expression in the control strain. (C) Exemplary visualization of tensyuic acid C (alkyl citrate) and fumonisin B4

abundances during cultivation of overexpression and control strains of A. niger on agar plates at different time points

(biological duplicates).

Beyond the Cluster Paradigm

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 11

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

the variation between BGC expression levels between fungi, such estimations are not

currently possible. However, we advocate that, ideally, interested users should (i) utilize a

maximum amount of transcriptomic data from high-quality public databases, (ii) select

and generate transcriptomic data from conditions known or predicted to activate BGC

expression for the BGC of interest, or (iii) use a combination of these approaches.

Overexpression of six TF-encoding genes (mjkA to -F) predicted from coexpression

networks to be involved in A. niger SM regulation enabled the modiﬁcation of A. niger

secondary metabolite proﬁles, which included the production of SMs that were not

detected in the progenitor control (Fig. S7). Thus, wholesale modulation of fungal SMs in

standard laboratory culture is possible using hypotheses derived from both Spearman and

mutual rank network approaches. The simplicity of the culture conditions is an attractive

aspect of the discovery pipeline in this work, which may be preferable to more complex

experimental setups, such as cocultivation experiments or isolation of novel metabolites

from the complex fungal niche (e.g., soil) or marine environments (64).

From a methodological perspective, our data support the notion that TF overex-

pression using an inducible gene switch is an effective strategy for SM activation and is

probably preferable to conventional gene deletion approaches (33). It should be noted,

however, that this study was clearly not able to activate all A. niger SMs, as we only an-

alyzed SM proﬁles from a single growth stage/time point for each mutant. Therefore,

we speculate that activation of other metabolites will be observed under different cul-

ture conditions or at different growth phases. Consequently, the full exploration of the

SM repertoire of A. niger MjkA to -F isolates will be conducted in follow-up studies. It

should be noted that it is currently unclear whether the mjkA to -F genes will enable

the activation of BGCs at levels comparable to the levels induced by existing regula-

tors, such as laeA,pacC,areA,creA,stuA,orbrlA. Where conservation of MjkA to -F is

observed in other fungal genomes, the functional analysis (i.e., overexpression) of such

orthologues to activate and discover other SM molecules appears feasible.

An exciting observation during this study was the irregular formation of putative

sclerotia due to overexpression of MjkA, which can be viewed as a preliminary (and

tentative) step toward laboratory-controlled sex, opening up the possibility of classical

genetics in this species (65). Such developmental jackpots may be viewed as an addi-

tional beneﬁt to wholesale analysis of fungal SMs using coexpression networks.

In this work, we also conducted signiﬁcant in silico and mass spectrometry-based

characterization of differential SM production proﬁles and attempted to link empiri-

cally observed SMs to speciﬁc BGCs. Despite recent advances in publicly available tools

for such experiments, including the prediction of putative SM structures based on the

analysis of PKS/NRPS domains (66), coupling BGCs to their products is still challenging.

In this respect, linking BGCs among multiple differentially produced SMs between con-

trol and experimental cohorts remains a signiﬁcant bottleneck in discovery pipelines

and requires experimental validation of putative BGC metabolite candidates, e.g., by

means of core gene knockout or overexpression.

In summary, this study has generated novel coexpression resources and methods

for the microbial cell factory A. niger. Overexpression strains MjkA to -F are promising

tools for metabolite discovery and will be used in future to reverse engineer the tran-

scriptional networks to which they belong. Our data clearly support the well-estab-

lished prevalence of BGCs in ﬁlamentous fungal genomes but suggest a reﬁnement to

this paradigm, whereby for activation and functional analysis experiments of SMs, it

may be safer to consider that the necessary genes for a fungal SM of interest (including

core genes, tailoring genes, transporters, detoxiﬁers, and regulators) may be unclus-

tered but can be identiﬁed by means of both SCC and MR-PCC coexpression analyses.

Such shifts in experimental thinking may help facilitate the full exploitation and com-

prehensive understanding of SMs among the fungal kingdom.

MATERIALS AND METHODS

Calculating mutual rank for microarray experiments. A. niger microarray data sets obtained across

a range of experimental conditions and genetic backgrounds (33) were analyzed in R using the affy,

Kwon et al.

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 12

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

simpleaffy, and makecdfenv packages (67–69). Raw data from each of the 283 individual microarrays

were normalized using the robust multiarray average (RMA) expression method as implemented in the

affy package (67). To enable cross-experiment comparisons, expression values were normalized by scal-

ing to the cross-experiment trimmed mean (excluding the top and bottom 5% of expression values).

Pearson’s correlation coefﬁcient was calculated between every pair of genes across all conditions. An or-

dered list of all genes from most to least correlated was generated for each gene. For every pair of

genes, the mutual rank was calculated by taking the geometric mean of the rank of each gene in the

other gene’s ordered list. The mutual rank (MR) of two genes A and B is the geometric mean of each

gene’s correlation rank and is given by the following formula:

MutualRankA;B¼ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

RankAðBÞRankBðAÞ

where Rank

A(B)

is the rank of gene B in an ordered list of the correlation coefﬁcients of all genes with

respect to gene A ranked from most to least correlated (34). MR scores were transformed to network edge

weights using the exponential decay function e

2(MR 21/x)

; three different networks were constructed with x

set to 5, 10, and 25, respectively. Edges with a Pearson’s correlation coefﬁcient of ,0.3 or an edge weight

of ,0.1 were excluded from the global network, which was then visualized in Cytoscape (70). Modules of

coexpressed genes were inferred using ClusterONE with default parameters (37). Modules were analyzed

for the presence of transcription factors and for SM backbone genes based on protein domains found

within these genes and from gene annotations predicted by antiSMASH (71). For two transcription factor

genes (mjkE and mjkF), the results from all coexpression networks were combined by collapsing all mod-

ules containing these genes of interest into a metamodule of nonoverlapping genes. For identiﬁcation of

shared clusters in Aspergillus species (Table S2), MultiGeneBlast (72) was used with 83 available representa-

tive genome assemblies available on NCBI Assembly as the search database (Table S1F).

Strains and molecular techniques. The A. niger strains used in this study are summarized in Table

S1C. Medium compositions and the methods for transformation of A. niger, strain puriﬁcation, and fun-

gal chromosomal DNA isolation were as previously described (73). Standard PCR and cloning procedures

were used for the generation of all constructs (74), and all cloned fragments were conﬁrmed by DNA

sequencing. Correct integrations of constructs in A. niger were veriﬁed by Southern blot analysis (74).

For overexpressing mjkC,mjkD,MjkE, and mjkF, the respective open reading frames were cloned into the

Tet-On vector pVG2.2 (40) and the resulting plasmids integrated as single or multiple copies at the pyrG

locus of strain MA169.4. The expression of the respective Tet-On-controlled gene was measured 24 h af-

ter DOX induction in all overexpression mutants, using qPCR (Table S1E). Details on cloning protocols,

primers used, and Southern blot results are available upon request from the authors.

Growth assays. All A. niger isolates were routinely cultured in the dark at 30°C in either minimal me-

dium (MM) (75) or complete medium (CM), which consisted of MM supplemented with 0.5% Casamino

Acids and 1% yeast extract as described previously (75). Doxycycline (DOX) was added to either solid or

liquid medium where indicated to a ﬁnal concentration of 10

g/ml. For growth assays on solid medium,

spores were inoculated on CM or MM with or without DOX and grown for up to 144 h. For A. niger

cultivations in liquid medium, spores were inoculated into 50 ml of MM at a concentration of 10

/ml.

Cultures were incubated at 30°C and 200 rpm. DOX was added 16 h after inoculation, which approxi-

mates the exponential growth phase. DOX was then added every 24 h until a maximum period of 92 h.

Strain MJK17.25 served as the control strain for all growth assays (Table S1C). For qPCR, 50 ml Aspergillus

minimal medium supplemented with 0.1% yeast extract to enhance spore germination was inoculated

at a concentration of 5 10

spores/ml into 250-ml shake ﬂasks. Technical duplicates were prepared for

all overexpression strains, and a quadruplicate cultivation was performed for the parental strain. Flasks

were shaken at 250 rpm and 30°C for 12 h to allow spore germination before the addition of 20

g/ml

doxycycline to induce the expression of genes under the control of the Tet-On system. The parental

strain was treated identically. Samples for qPCR were taken 24 h after induction by separating broth and

mycelium via vacuum ﬁltration, brieﬂy washing with Milli-Q water, and sampling 100 to 200 mg myce-

lium into screw-cap tubes with sterile glass beads and 1 ml TRIzol reagent. Until total RNA extraction,

samples were stored at 280°C.

RNA isolation and qPCR. For isolation of total RNA, samples were thawed and homogenized using

a FastPrep 120 (Thermo Fisher Savant) and processed further using a Direct-zol RNA miniprep kit (Zymo

Research). Total RNA was quantiﬁed using a BioSpectrometer (Eppendorf), and 2

g of total RNA was

used for reverse transcription in a 20-

l reaction mixture with random hexameric primers following the

instructions of the RevertAid H minus ﬁrst-strand cDNA synthesis kit (Thermo Fisher). Reverse transcrip-

tion reaction mixtures were diluted 12-fold with nuclease-free water, and 2

l of the dilution was used

as the input for a 10-

l SYBR-based qPCR (Blue S9Green qPCR kit; Biozym Scientiﬁc) on an AriaMx real-

time PCR system (Agilent). Primers are listed in Table S1E. For each sample, technical duplicates for each

target gene were measured. Raw threshold cycle (C

) data were exported to MS Excel and the D(C

) val-

ues from technical replicates (shake ﬂasks and reaction replicates) were averaged before calculating the

DD(C

) values according to the method in reference 76, assuming an ampliﬁcation efﬁciency of one.

Actin (An15g00560) was used as the reference gene.

Metabolome proﬁling. Metabolites were extracted from colonies of A. niger MJK17.25 grown on

agar plates (independent biological duplicates) by Metabolon (Potsdam, Germany). In brief, three agar

plugs (outer edge to plate, center of colony, and outer edge adjacent to next colony) were collected at

different time points from a colony cultivated for 2 to 10 days on minimal agar medium and pooled in

one reaction tube. Each sample was extracted in a concentration of 0.5 g/ml with isopropanol:ethyl

Beyond the Cluster Paradigm

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 13

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

acetate (1:3, vol/vol) by ultrasound for 60 min and centrifuged at 4°C at 13,500 rpm for 20 min. The su-

pernatant was sterile ﬁltrated (0.22

m; Carl Roth) and transferred into a new Eppendorf tube. All subse-

quent steps were carried out at Metabolon (Potsdam, Germany). Metabolites were identiﬁed in compari-

son to Metabolon’s database entries of authentic standards. The LC separation was performed using

hydrophilic interaction chromatography with an iHILIC-Fusion, 150- by 2.1-mm, 5-

m, 200-Å column

(Hilicon, Umeå Sweden), operated by an Agilent 1290 ultraperformance liquid chromatography (UPLC)

system (Agilent, Santa Clara, CA, USA).

The LC mobile phase A was 10 mM ammonium acetate (Sigma-Aldrich, USA) in water (Thermo

Fisher, USA) with 95% acetonitrile (pH 6; Thermo Fisher, USA), and mobile phase B was acetonitrile with

5% 10 mM ammonium acetate in 95% water. The LC mobile phase was a linear gradient from 95% to

65% acetonitrile over 8.5 min, followed by a linear gradient from 65% to 5% acetonitrile over 1 min and

then a 2.5-min wash with 5% and a 3-min reequilibration with 95% acetonitrile (ﬂow rate, 400

l/min).

Mass spectrometry was performed using a high-resolution 6540 QTOF/MS detector (Agilent, Santa Clara,

CA, USA). Spectra were recorded in a mass range from 50 m/z to 1,700 m/z in positive and negative ioni-

zation mode. The measured metabolite concentration was normalized to the internal standard.

Signiﬁcant concentration changes of metabolites in different samples were analyzed by appropriate sta-

tistical test procedures (Student test, Welch test, and Mann-Whitney test). A Pvalue of ,0.05 was consid-

ered signiﬁcant.

SUPPLEMENTAL MATERIAL

Supplemental material is available online only.

SUPPLEMENTAL FILE 1, PDF ﬁle, 1.7 MB.

SUPPLEMENTAL FILE 2, XLSX ﬁle, 0.6 MB.

SUPPLEMENTAL FILE 3, XLSX ﬁle, 0.1 MB.

ACKNOWLEDGMENTS

We thank the European Commission for support (Marie Curie International Training

Network QuantFung, FP7-People-2013-ITN, grant no. 607332 to V.M.), the German

Research Foundation (project 404295023 to V.M.), and the National Science Foundation

(http://www.nsf.gov) for funding grants number IOS-1401682 and DEB-1831493 to

J.H.W. We acknowledge support by the German Research Foundation and the Open

Access Publication Funds of TU Berlin.

This work was conducted in part using computational resources provided by the

Advanced Computing Center for Research and Education at Vanderbilt University and

Information Technology at Purdue.

REFERENCES

1. Keller NP. 2019. Fungal secondary metabolism: regulation, function and

drug discovery. Nat Rev Microbiol 17:167–180. https://doi.org/10.1038/

s41579-018-0121-1.

2. Keller NP, Turner G, Bennett JW. 2005. Fungal secondary metabolism:

from biochemistry to genomics. Nat Rev Microbiol 3:937–947. https://doi

.org/10.1038/nrmicro1286.

3. Newman DJ, Cragg GM. 2016. Natural products as sources of new drugs

from 1981 to 2014. J Nat Prod 79:629–661. https://doi.org/10.1021/acs

.jnatprod.5b01055.

4. Liu Y, Wu F. 2010. Global burden of aﬂatoxin-induced hepatocellular car-

cinoma: a risk assessment. Environ Health Perspect 118:818–824. https://

doi.org/10.1289/ehp.0901388.

5. Meyer V, Andersen MR, Brakhage AA, Braus GH, Caddick MX, Cairns TC, de

Vries RP, Haarmann T, Hansen K, Hertz-Fowler C, Krappmann S, Mortensen

UH, Peñalva MA, Ram AFJ, Head RM. 2016. Current challenges of research on

ﬁlamentous fungi in relation to human welfare and a sustainable bio-econ-

omy: a white paper. Fungal Biol Biotechnol 3:6. https://doi.org/10.1186/

s40694-016-0024-8.

6. Frisvad JC, Larsen T, Thrane U, Meijer M, Varga J, Samson RA, Nielsen KF.

2011. Fumonisin and ochratoxin production in industrial Aspergillus niger

strains. PLoS One 6:e23496. https://doi.org/10.1371/journal.pone.0023496.

7. Pusztahelyi T, Holb IJ, Pócsi I. 2015. Secondary metabolites in fungus-plant

interactions. Front Plant Sci 6:573. https://doi.org/10.3389/fpls.2015.00573.

8. Fisher MC, Henk DA, Briggs CJ, Brownstein JS, Madoff LC, McCraw SL, Gurr

SJ. 2012. Emerging fungal threats to animal, plant and ecosystem health.

Nature 484:186–194. https://doi.org/10.1038/nature10947.

9. Meyer V, Basenko EY, Benz JP, Braus GH, Caddick MX, Csukai M, de Vries

RP, Endy D, Frisvad JC, Gunde-Cimerman N, Haarmann T, Hadar Y, Hansen

K, Johnson RI, Keller NP, Kraševec N, Mortensen UH, Perez R, Ram AFJ,

Record E, Ross P, Shapaval V, Steiniger C, van den Brink H, van Munster J,

Yarden O, Wösten HAB. 2020. Growing a circular economy with fungal

biotechnology: a white paper. Fungal Biol Biotechnol 7:5. https://doi.org/

10.1186/s40694-020-00095-z.

10. Rokas A, Wisecaver JH, Lind AL. 2018. The birth, evolution and death of

metabolic gene clusters in fungi. Nat Rev Microbiol 16:731–744. https://

doi.org/10.1038/s41579-018-0075-3.

11. Khaldi N, Seifuddin FT, Turner G, Haft D, Nierman WC, Wolfe KH, Fedorova

ND. 2010. SMURF: genomic mapping of fungal secondary metabolite clusters.

Fungal Genet Biol 47:736–741. https://doi.org/10.1016/j.fgb.2010.06.003.

12. Weber T, Blin K, Duddela S, Krug D, Kim HU, Bruccoleri R, Lee SY,

Fischbach MA, Müller R, Wohlleben W, Breitling R, Takano E, Medema MH.

2015. antiSMASH 3.0-a comprehensive resource for the genome mining

of biosynthetic gene clusters. Nucleic Acids Res 43:W237–W243. https://

doi.org/10.1093/nar/gkv437.

13. Wang D-N, Toyotome T, Muraosa Y, Watanabe A, Wuren T, Bunsupa S,

Aoyagi K, Yamazaki M, Takino M, Kamei K. 2014. GliA in Aspergillus fumiga-

tus is required for its tolerance to gliotoxin and affects the amount of

extracellular and intracellular gliotoxin. Med Mycol 52:506–518. https://

doi.org/10.1093/mmy/myu007.

14. Chang PK, Yu J, Yu JH. 2004. aﬂT, a MFS transporter-encoding gene

located in the aﬂatoxin gene cluster, does not have a signiﬁcant role in

aﬂatoxin secretion. Fungal Genet Biol 41:911–920. https://doi.org/10

.1016/j.fgb.2004.06.007.

15. Schrettl M, Carberry S, Kavanagh K, Haas H, Jones GW, O’Brien J, Nolan A,

Stephens J, Fenelon O, Doyle S. 2010. Self-protection against gliotoxin—a

component of the gliotoxin biosynthetic cluster, GliT, completely protects

Kwon et al.

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 14

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

Aspergillus fumigatus against exogenous gliotoxin. PLoS Pathog 6:e1000952.

https://doi.org/10.1371/journal.ppat.1000952.

16. Macheleidt J, Mattern DJ, Fischer J, Netzker T, Weber J, Schroeckh V, Valiante

V, Brakhage AA. 2016. Regulation and role of fungal secondary metabolites.

Annu Rev Genet 50:371–392. https://doi.org/10.1146/annurev-genet-120215

-035203.

17. Brakhage AA. 2013. Regulation of fungal secondary metabolism. Nat Rev

Microbiol 11:21–32. https://doi.org/10.1038/nrmicro2916.

18. Bergmann S, Schümann J, Scherlach K, Lange C, Brakhage AA, Hertweck

C. 2007. Genomics-driven discovery of PKS-NRPS hybrid metabolites from

Aspergillus nidulans. Nat Chem Biol 3:213–217. https://doi.org/10.1038/

nchembio869.

19. Bok JW, Chung D, Balajee SA, Marr KA, Andes D, Nielsen KF, Frisvad JC,

Kirby KA, Keller NP. 2006. GliZ, a transcriptional regulator of gliotoxin bio-

synthesis, contributes to Aspergillus fumigatus virulence. Infect Immun 74:

6761–6768. https://doi.org/10.1128/IAI.00780-06.

20. Marui J, Yamane N, Ohashi-Kunihiro S, Ando T, Terabayashi Y, Sano M,

Ohashi S, Ohshima E, Tachibana K, Higa Y, Nishimura M, Koike H, Machida M.

2011. Kojic acid biosynthesis in Aspergillus oryzae is regulated by a Zn(II)(2)

Cys(6) transcriptional activator and induced by kojic acid at the transcrip-

tional level. J Biosci Bioeng 112:40–43. https://doi.org/10.1016/j.jbiosc.2011

.03.010.

21. NiehausE-M,JanevskaS,vonBargenKW,SieberCMK,HarrerH,HumpfH-U,

Tudzynski B. 2014. Apicidin F: characterization and genetic manipulation of a

new secondary metabolite gene cluster in the rice pathogen Fusarium fuji-

kuroi. PLoS One 24:e103336. https://doi.org/10.1371/journal.pone.0103336.

22. Bayram O, Krappmann S, Ni M, Bok JW, Helmstaedt K, Valerius O, Braus-

Stromeyer S, Kwon N-J, Keller NP, Yu J-H, Braus GH. 2008. VelB/VeA/LaeA

complex coordinates light signal with fungal development and second-

ary metabolism. Science 320:1504–1506. https://doi.org/10.1126/science

.1155888.

23. Sigl C, Haas H, Specht T, Pfaller K, Kürnsteiner H, Zadra I. 2011. Among de-

velopmental regulators, StuA but not BrlA is essential for penicillin V pro-

duction in Penicillium chrysogenum. Appl Environ Microbiol 77:972–982.

https://doi.org/10.1128/AEM.01557-10.

24. Karimi-Aghcheh R, Bok JW, Phatale PA, Smith KM, Baker SE, Lichius A,

Omann M, Zeilinger S, Seiboth B, Rhee C, Keller NP, Freitag M, Kubicek CP.

2013. Functional analyses of Trichoderma reesei LAE1 reveal conserved

and contrasting roles of this regulator. G3 (Bethesda) 3:369–378. https://

doi.org/10.1534/g3.112.005140.

25. Perrin RM, Fedorova ND, Bok JW, Cramer RA, Wortman JR, Kim HS,

Nierman WC, Keller NP. 2007. Transcriptional regulation of chemical di-

versity in Aspergillus fumigatus by LaeA. PLoS Pathog 3:508–517. https://

doi.org/10.1371/journal.ppat.0030050.

26. Haas H. 2014. Fungal siderophore metabolism with a focus on Aspergil-

lus fumigatus. Nat Prod Rep 31:1266–1276. https://doi.org/10.1039/

c4np00071d.

27. Zhang S, Schwelm A, Jin H, Collins LJ, Bradshaw RE. 2007. A fragmented

aﬂatoxin-like gene cluster in the forest pathogen Dothistroma septospo-

rum. Fungal Genet Biol. https://doi.org/10.1016/j.fgb.2007.06.005.

28. Wiemann P, Guo C-J, Palmer JM, Sekonyela R, Wang CCC, Keller NP. 2013.

Prototype of an intertwined secondary-metabolite supercluster. Proc Natl

Acad Sci U S A 110:17065–17070. https://doi.org/10.1073/pnas.1313258110.

29. Andersen MR, Nielsen JB, Klitgaard A, Petersen LM, Zachariasen M, Hansen

TJ,BlicherLH,GotfredsenCH,LarsenTO,NielsenKF,MortensenUH.2013.

Accurate prediction of secondary metabolite gene clusters in ﬁlamentous

fungi. Proc Natl Acad Sci U S A 110:E99–E107. https://doi.org/10.1073/pnas

.1205532110.

30. Tang M-C, Lin H-C, Li D, Zou Y, Li J, Xu W, Cacho RA, Hillenmeyer ME, Garg

NK, Tang Y. 2015. Discovery of unclustered fungal indole diterpene bio-

synthetic pathways through combinatorial pathway reassembly in engi-

neered yeast. J Am Chem Soc 137:13724–13727. https://doi.org/10.1021/

jacs.5b06108.

31. Vesth TC, Brandl J, Andersen MR. 2016. FunGeneClusterS: predicting fun-

gal gene clusters from genome and transcriptome data. Synth Syst Bio-

technol 1:122–129. https://doi.org/10.1016/j.synbio.2016.01.002.

32. Cairns T, Meyer V. 2017. In silico prediction and characterization of secondary

metabolite biosynthetic gene clusters in the wheat pathogen Zymoseptoria

tritici.BMCGenomics18:631.https://doi.org/10.1186/s12864-017-3969-y.

33. Schäpe P, Kwon M, Baumann B, Gutschmann B, Jung S, Lenz S, Nitsche B,

Paege N, Schütze T, Cairns TC, Meyer V. 2019. Updating genome annota-

tion for the microbial cell factory Aspergillus niger using gene co-expres-

sion networks. Nucleic Acids Res 47:559–569. https://doi.org/10.1093/

nar/gky1183.

34. Obayashi T, Kinoshita K. 2009. Rank of correlation coefﬁcient as a compa-

rable measure for biological signiﬁcance of gene coexpression. DNA Res

16:249–260. https://doi.org/10.1093/dnares/dsp016.

35. Liesecke F, Daudu D, Dugé de Bernonville R, Besseau S, Clastre M,

Courdavault V, de Craene J-O, Crèche J, Giglioli-Guivarc’h N, Glévarec G,

Pichon O, Dugé de Bernonville T. 2018. Ranking genome-wide correlation

measurements improves microarray and RNA-seq based global and tar-

geted co-expression networks. Sci Rep 8:10885. https://doi.org/10.1038/

s41598-018-29077-3.

36. Wisecaver JH, Borowsky AT, Tzin V, Jander G, Kliebenstein DJ, Rokas A.

2017. A global coexpression network approach for connecting genes to

specialized metabolic pathways in plants. Plant Cell 29:944–959. https://

doi.org/10.1105/tpc.17.00009.

37. Nepusz T, Yu H, Paccanaro A. 2012. Detecting overlapping protein com-

plexes in protein-protein interaction networks. Nat Methods 9:471–472.

https://doi.org/10.1038/nmeth.1938.

38. Van Dongen S, Abreu-Goodger C. 2012. Using MCL to extract clusters

from networks. Methods Mol Biol 804:281–295. https://doi.org/10.1007/

978-1-61779-361-5_15.

39. Zabala AO, Xu W, Chooi Y-H, Tang Y. 2012. Characterization of a silent aza-

philone gene cluster from Aspergillus niger ATCC 1015 reveals a hydroxy-

lation-mediated pyran-ring formation. Chem Biol 19:1049–1059. https://

doi.org/10.1016/j.chembiol.2012.07.004.

40. Meyer V, Wanka F, van Gent J, Arentshorst M, van den Hondel CAMJJ, Ram

AFJ. 2011. Fungal gene expression on demand: an inducible, tunable, and

metabolism-independent expression system for Aspergillus niger.ApplEnvi-

ron Microbiol 77:2975–2983. https://doi.org/10.1128/AEM.02740-10.

41. Wanka F, Cairns T, Boecker S, Berens C, Happel A, Zheng X, Sun J,

Krappmann S, Meyer V. 2016. Tet-On, or Tet-Off, that is the question:

advanced conditional gene expression in Aspergillus. Fungal Genet Biol

89:72–83. https://doi.org/10.1016/j.fgb.2015.11.003.

42. Frisvad JC, Petersen LM, Lyhne EK, Larsen TO. 2014. Formation of sclerotia

and production of indoloterpenes by Aspergillus niger and other species

in section Nigri. PLoS One 9:e94857. https://doi.org/10.1371/journal.pone

.0094857.

43. Jørgensen TR, Burggraaf A-M, Arentshorst M, Schutze T, Lamers G, Niu J,

Kwon MJ, Park J, Frisvad JC, Nielsen KF, Meyer V, van den Hondel CAMJJ,

Dyer PS, Ram AFJ. 2020. Identiﬁcation of SclB, a Zn(II)

Cys

transcription

factor involved in sclerotium formation in Aspergillus niger. Fungal Genet

Biol 139:103377. https://doi.org/10.1016/j.fgb.2020.103377.

44. Ellena V, Bucchieri D, Arcalis E, Sauer M, Steiger MG. 2021. Sclerotia formed

by citric acid producing strains of Aspergillus niger: induction and morpho-

logical analysis. Fungal Biol 125:485–494. https://doi.org/10.1016/j.funbio

.2021.01.008.

45. Cerqueira GC, Arnaud MB, Inglis DO, Skrzypek MS, Binkley G, Simison M,

Miyasato SR, Binkley J, Orvis J, Shah P, Wymore F, Sherlock G, Wortman

JR. 2014. The Aspergillus Genome Database: multispecies curation and

incorporation of RNA-Seq data to improve structural gene annotations.

Nucleic Acids Res 42:D705–D710. https://doi.org/10.1093/nar/gkt1029.

46. Trieu M, Ma A, Eng SR, Fedtsova N, Turner EE. 2003. Direct autoregulation

and gene dosage compensation by POU-domain transcription factor

Brn3a. Development. https://doi.org/10.1242/dev.00194.

47. Hossain AH, Hendrikx A, Punt PJ. 2019. Identiﬁcation of novel citramalate

biosynthesis pathways in Aspergillus niger. Fungal Biol Biotechnol 6:19.

https://doi.org/10.1186/s40694-019-0084-7.

48. Hossain AH, van Gerven R, Overkamp KM, Lübeck PS, Taspınar H, Türker M,

Punt PJ. 2019. Metabolic engineering with ATP-citrate lyase and nitrogen

source supplementation improves itaconic acid production in Aspergillus ni-

ger. Biotechnol Biofuels 12:233. https://doi.org/10.1186/s13068-019-1577-6.

49. Hossain AH, Li A, Brickwedde A, Wilms L, Caspers M, Overkamp K, Punt PJ.

2016. Rewiring a secondary metabolite pathway towards itaconic acid

production in Aspergillus niger. Microb Cell Fact 15:130. https://doi.org/10

.1186/s12934-016-0527-2.

50. Nielsen KF, Mogensen JM, Johansen M, Larsen TO, Frisvad JC. 2009.

Review of secondary metabolites and mycotoxins from the Aspergillus ni-

ger group. Anal Bioanal Chem 395:1225–1242. https://doi.org/10.1007/

s00216-009-3081-5.

51. Li Y, Chooi YH, Sheng Y, Valentine JS, Tang Y. 2011. Comparative characteri-

zation of fungal anthracenone and naphthacenedione biosynthetic path-

ways reveals an

-hydroxylation-dependent claisen-like cyclization catalyzed

by a dimanganese thioesterase. J Am Chem Soc 133:15773–15785. https://

doi.org/10.1021/ja206906d.

52. GalloA,BrunoKS,SolfrizzoM,PerroneG,MulèG,ViscontiA,BakerSE.2012.

New insight into the ochratoxin a biosynthetic pathway through deletion of

Beyond the Cluster Paradigm

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 15

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.

a nonribosomal peptide synthetase gene in Aspergillus carbonarius.ApplEn-

viron Microbiol 78:8208–8218. https://doi.org/10.1128/AEM.02508-12.

53. Miao V, Coëffet-LeGal M-F, Brian P, Brost R, Penn J, Whiting A, Martin S,

Ford R, Parr I, Bouchard M, Silva CJ, Wrigley SK, Baltz RH. 2005. Daptomy-

cin biosynthesis in Streptomyces roseosporus: cloning and analysis of the

gene cluster and revision of peptide stereochemistry. Microbiology

(Reading) 151(Pt 5):1507–1523. https://doi.org/10.1099/mic.0.27757-0.

54. Palys S, Pham TTM, Tsang A. 2020. Biosynthesis of alkylcitric acids in As-

pergillus niger involves both co-localized and unlinked genes. Front

Microbiol 11:1378. https://doi.org/10.3389/fmicb.2020.01378.

55. Matsumaru T, Sunazuka T, Hirose T, Ishiyama A, Namatame M, Fukuda T,

Tomoda H, Otoguro K, Ōmura S. 2008. Synthesis and biological properties of

tensyuic acids B, C, and E, and investigation of the optical purity of natural ten-

syuic acid B. Tetrahedron 64:7369–7377. https://doi.org/10.1016/j.tet.2008.05

.035.

56. Koch L, Lodin A, Herold I, Ilan M, Carmeli S, Yarden O. 2014. Sensitivity of

Neurospora crassa to a marine-derived Aspergillus tubingensis anhydride

exhibiting antifungal activity that is mediated by the MAS1 protein. Mar

Drugs 12:4713–4731. https://doi.org/10.3390/md12094713.

57. Hasegawa Y, Fukuda T, Hagimori K, Tomoda H, Ōmura S. 2007. Tensyuic

acids, new antibiotics produced by Aspergillus niger FKI-2342. Chem

Pharm Bull (Tokyo) 55:1338–1341. https://doi.org/10.1248/cpb.55.1338.

58. Akira I, Washizu M, Kondo K, Murakoshi S, Suzuki A. 1984. Isolation and iden-

tiﬁcation of (1)-hexylitaconic acid as a plant growth regulator. Agric Biol

Chem 48:2607–2609. https://doi.org/10.1080/00021369.1984.10866557.

59. Wilson KE, Burk RM, Biftu T, Ball RG, Hoogsteen K. 1992. Zaragozic acid A, a

potent inhibitor of squalene synthase: initial chemistry and absolute stereo-

chemistry. J Org Chem 57:7151–7158. https://doi.org/10.1021/jo00052a032.

60. Dawson MJ, Farthing JE, Marshall PS, Middleton RF, O’Neill MJ,

Shuttleworth A, Stylli C, Tait RM, Taylor PM, Wildman HG. 1992. The squa-

lestatins, novel inhibitors of squalene synthase produced by a species of

Phoma. I. Taxonomy, fermentation, isolation, physico-chemical properties

and biological activity. J Antibiot (Tokyo) 45:639–647. https://doi.org/10

.7164/antibiotics.45.639.

61. Patron NJ, Waller RF, Cozijnsen AJ, Straney DC, Gardiner DM, Nierman

WC, Howlett BJ. 2007. Origin and distribution of epipolythiodioxopipera-

zine (ETP) gene clusters in ﬁlamentous ascomycetes. BMC Evol Biol 7:174.

https://doi.org/10.1186/1471-2148-7-174.

62. Tsai HF, Wheeler MH, Chang YC, Kwon-Chung KJ. 1999. A developmen-

tally regulated gene cluster involved in conidial pigment biosynthesis in

Aspergillus fumigatus. J Bacteriol 181:6469–6477. https://doi.org/10.1128/

JB.181.20.6469-6477.1999.

63. de Vries RP, Riley R, Wiebenga A, Aguilar-Osorio G, Amillis S, Uchima CA,

Anderluh G, Asadollahi M, Askin M, Barry K, Battaglia E, Bayram Ö, Benocci T,

Braus-StromeyerSA,CaldanaC,CánovasD,CerqueiraGC,ChenF,ChenW,

ChoiC,ClumA,dosSantosRAC,deLimaDamásioA,DiallinasG,EmriT,

Fekete E, Flipphi M, Freyberg S, Gallo A, Gournas C, Habgood R, Hainaut M,

Harispe ML, Henrissat B, Hildén KS, Hope R, Hossain A, Karabika E, Karaffa L,

Karányi Z, Kraševec N, Kuo A, Kusch H, LaButti K, Lagendijk EL, Lapidus A,

Levasseur A, Lindquist E, Lipzen A, Logrieco AF, et al. 2017. Comparative

genomics reveals high biological diversity and speciﬁc adaptations in the

industrially and medically important fungal genus Aspergillus. Genome Biol

18:28. https://doi.org/10.1186/s13059-017-1151-0.

64. Nai C, Meyer V. 2018. From axenic to mixed cultures: technological advan-

ces accelerating a paradigm shift in microbiology. Trends Microbiol 26:

538–554. https://doi.org/10.1016/j.tim.2017.11.004.

65. Todd RB, Davis MA, Hynes MJ. 2007. Genetic manipulation of Aspergillus

nidulans: meiotic progeny for genetic analysis and strain construction.

Nat Protoc 2:811–821. https://doi.org/10.1038/nprot.2007.112.

66. Blin K, Shaw S, Steinke K, Villebro R, Ziemert N, Lee SY, Medema MH,

Weber T. 2019. antiSMASH 5.0: updates to the secondary metabolite ge-

nome mining pipeline. Nucleic Acids Res 47:W81–W87. https://doi.org/10

.1093/nar/gkz310.

67. Gautier L, Cope L, Bolstad BM, Irizarry RA. 2004. affy—analysis of Affyme-

trix GeneChip data at the probe level. Bioinformatics 20:307–315. https://

doi.org/10.1093/bioinformatics/btg405.

68. Irizarry RA, Gautier L, Huber W, Bolstad B. 2018. makecdfenv: CDF environ-

ment maker. R package version 1.58.0.

69. Miller CJ. 2018. simpleaffy: very simple high level analysis of Affymetrix

data. http://www.bioconductor.org,http://bioinformatics.picr.man.ac.uk/

simpleaffy/.

70. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N,

Schwikowski B, Ideker T. 2003. Cytoscape: a software environment for

integrated models of biomolecular interaction networks. Genome Res 13:

2498–2504. https://doi.org/10.1101/gr.1239303.

71. Medema MH, Blin K, Cimermancic P, de Jager V, Zakrzewski P, Fischbach

MA, Weber T, Takano E, Breitling R. 2011. antiSMASH: rapid identiﬁcation,

annotation and analysis of secondary metabolite biosynthesis gene clus-

ters in bacterial and fungal genome sequences. Nucleic Acids Res 39:

W339–W346. https://doi.org/10.1093/nar/gkr466.

72. Medema MH, Takano E, Breitling R. 2013. Detecting sequence homology

at the gene cluster level with multigeneblast. Mol Biol Evol 30:1218–1223.

https://doi.org/10.1093/molbev/mst025.

73. Meyer V, Ram AFJ, Punt PJ. 2010. Genetics, genetic manipulation, and

approaches to strain improvement of ﬁlamentous fungi. In Bull AT,

Junker B, Katz L, Lynd LR, Masurekar P, Reeves CD, Zhao H (ed), Manual of

industrial microbiology and biotechnology, 3rd ed. ASM Press, Washing-

ton, DC. https://doi.org/10.1128/9781555816827.ch22.

74. Green MR, Sambrook J. 2012. Molecular cloning: a laboratory manual, 4th

ed. Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY.

75. Arentshorst M, Ram AFJ, Meyer V. 2012. Using non-homologous end-join-

ing-deﬁcient strains for functional gene analyses in ﬁlamentous fungi.

Methods Mol Biol 835:133–150. https://doi.org/10.1007/978-1-61779-501

-5_9.

76. Livak KJ, Schmittgen TD. 2001. Analysis of relative gene expression data

using real-time quantitative PCR and the 2-DDCT method. Methods 25:

402–408. https://doi.org/10.1006/meth.2001.1262.

77. van Munster JM, Nitsche BM, Akeroyd M, Dijkhuizen L, van der Maarel

MJEC, Ram AFJ. 2015. Systems approaches to predict the functions of gly-

coside hydrolases during the life cycle of Aspergillus niger using develop-

mental mutants DbrlA and DﬂbA. PLoS One 10:e0116269. https://doi.org/

10.1371/journal.pone.0116269.

Kwon et al.

Volume 9 Issue 2 e00898-21 MicrobiolSpectrum.asm.org 16

Downloaded from https://journals.asm.org/journal/spectrum on 11 January 2022 by 92.195.8.55.