1
A C i ical Re-e alua ion o “B idging Single Cells o
O gans: Mesoscale Modules as Fundamen al Uni s
o Tissue Func ion” by Chen e al., Cell 2025; doi:
10.1016/j.cell.2025.10.012
Yiheng Wang and Shu-Feng Zhou*
College o Chemical Enginee ing, Huaqiao Uni e si y, Xiamen, China
*Co espondence: [email p o ec ed]
Abs ac
Chen e al. p opose ha “mesoscale modules”—in e media e s uc u al/ unc ional uni s
be ween single cells and issues—cons i u e he uni e sal ope a ional a chi ec u e o
o gans. Th ough a di e se po olio o imaging, compu a ional segmen a ion, and
pe u ba ional analyses, he au ho s a emp o a gue ha cell collec i es
spon aneously sel -o ganize in o epea ed mesoscale mo i s ha con ol issue
pheno ypes. While he concep ual ambi ion o he wo k is no able, he s udy su e s
om inconsis en de ini ions, ci cula easoning in module iden i ica ion, o e -
in e p e a ion o co ela i e da a, and imaging-d i en a i ac s ha a e insu icien ly
acknowledged. Many igu es show non-quan i a i e o selec i ely chosen images,
insu icien s a is ical suppo , o ambiguous compu a ional me hodology. Ex ended
and Supplemen a y Figu es o en aise addi ional conce ns ega ding ep oducibili y,
segmen a ion consis ency, and he obus ness o c oss- issue gene aliza ions.
Below, we pe o m a de ailed igu e-by- igu e c i ique, co e ing all main, Ex ended
Da a, and Supplemen a y Figu es, highligh ing concep ual, me hodological, and
s a is ical limi a ions.
1. In oduc ion: Concep ual Ambigui y and O e each
Chen e al.1 a emp o de ine “mesoscale modules” as “minimal mul icellula uni s ha
in eg a e cell-le el ules o p oduce o gan-le el unc ion.” Howe e , he pape does no
a icula e a quan i a i e de ini ion ha can clea ly dis inguish mesoscale modules om:
• classical issue mic odomains
• de elopmen ally speci ied compa men s
2
• niches (e.g., in es inal c yp s, ge minal cen e s)
• mic o-o ganoids
• o simply coa sely segmen ed clus e s p oduced by compu a ional algo i hms.
The ailu e o speci y necessa y and su icien c i e ia leads o a o m o con i ma ion
bias: once a clus e ing algo i hm o s uc u al segmen a ion is applied, he esul ing
clus e s a e e oac i ely named “modules” wi hou demons a ing causali y.
Mo eo e , he pape ex apola es highly om imaging-based da ase s bu a ely
alida es modules using unc ion-blocking expe imen s, abla ion, o lineage acing. This
disconnec educes con idence in he cen al claims o modula uni e sali y.
2. Figu e-by-Figu e C i ique
Below is a c i ique o each main Figu e, ocusing on he quali y o e idence, s a is ics,
me hodology, in e p e a ion, and missing con ols.
Figu e 1 — P oposed A chi ec u e o Mesoscale Modules Ac oss
Tissues
C i ique
1. Non-ope a ional de ini ion o “modules”
The cen al schema ic is isually appealing bu concep ually ague. Modules a e
depic ed as disc e e, epea ing uni s ac oss issues, ye he igu e does no
de ine:
o bounda ies
o o ganiza ional c i e ia
o s abili y
o o scale limi s
Wi hou hese, he diag am me ely es a es he hypo hesis a he han
suppo ing i .
2. Biased issue selec ion
The images highligh issues whe e a chi ec u e is al eady known o be modula
(e.g., lymph nodes, neph ons). This c ea es a sampling bias and does no p o e
ha “all issues” ollow he same pa e n.
3. Lack o quan i a i e measu emen
No densi y plo s, clus e ing alidi y indices, o scale- ee s a is ics a e shown.
3
The igu e sugges s uni e sali y wi hou p esen ing any quan i a i e uni e sali y
me ics.
4. Po en ial o e p ocessing
Many images appea hea ily con as -enhanced o pseudo-colo ed in a way ha
may exagge a e s uc u al bounda ies. No aw images o unp ocessed con ols
a e p o ided.
Figu e 2 — Imaging and Segmen a ion Pipeline o Module
Iden i ica ion
C i ique
1. Segmen a ion opaci y
The igu e p esen s he segmen a ion pipeline as de e minis ic and eliable bu
omi s:
o hype pa ame e s
o h esholds
o aining da ase sou ces
o in e -obse e a iabili y
o benchma king agains es ablished segmen a ion models (e.g., Cellpose,
DeepCell).
This educes ep oducibili y.
2. Ci cula logic
The au ho s segmen issues in o clus e s based on mo phological o p oximi y
ea u es, hen label hem “modules,” p oducing modules by de ini ion a he
han disco e y.
3. No obus ness es ing
Module iden i ica ion is no es ed unde :
o downsampled images
o pho obleaching
o andom o a ion o c opping
o noise addi ion.
Wi hou obus ness checks, he app oach may simply be de ec ing
a i ac s.
4
4. Unde powe ed s a is ics
The dis ibu ions in he igu e lack s anda d de ia ion, con idence in e als, o
sample coun s. I emains unclea how many biological eplica es we e included.
Figu e 3 — S uc u al Consis ency o Modules Ac oss
De elopmen al S ages
C i ique
1. Misalignmen o de elopmen al imepoin s
Ea ly and la e-s age issues a e compa ed using di e en imaging modali ies and
magni ica ions. This unde mines claims o s uc u al consis ency because
di e ences in esolu ion can gene a e a i icial con inui y.
2. No lineage acing
The igu e asse s ha modules pe sis h ough de elopmen . Ye no lineage-
acing expe imen s we e pe o med o e i y ha he same clus e o cells
pe sis s o e ime.
3. Che y-picked images
The selec ed examples all exhibi isibly egula pa e ns; no examples o
he e ogeneous o i egula samples a e shown. This aises conce ns o selec ion
bias.
4. No unc ional eadou s
The igu e desc ibes modules as unc ional uni s wi hou measu ing unc ion.
Calcium signaling, me abolic lux, o gene exp ession ac i i y should accompany
s uc u al images.
Figu e 4 — Mechanical and Geome ic Cons ain s Supposedly
De ine Module Bounda ies
C i ique
1. O e in e p e a ion o co ela ions
The igu e shows co ela ions be ween issue cu a u e, mechanical s ess, and
module bounda ies; howe e :
o no causal pe u ba ion
o no ac ion o ce mic oscopy
o no mechanical abla ion
is p o ided o suppo his.
5
2. Simula ion assump ions no disclosed
The mechanical model is o e simpli ied and insu icien ly pa ame e ized.
Wi hou :
o elas ici y cons an s
o bounda y condi ions
o mesh esolu ion
simula ion ou pu s canno be e alua ed.
3. Missing nega i e con ols
I mechanics d i e modules, pe u bing ac omyosin o ECM s i ness should
dis up modules. Bu no such expe imen s a e included in he main igu e.
Figu e 5 — Pe u ba ion Expe imen s: Dis up ion o Module
In eg i y
C i ique
1. Pe u ba ions do no speci ically a ge modules
The igu e uses b oad pe u ba ions (e.g., cy oskele on inhibi o s, ECM diges e s)
ha dis up gene al issue a chi ec u e. These do no demons a e speci ic
e ec s on mesoscale modules.
2. Inadequa e quan i ica ion
The “module dis up ion index” is ne e ma hema ically de ined. Wi hou an
exac o mula, eplica ion is impossible.
3. Lack o causal speci ici y
The igu e claims ha modules a e essen ial o unc ion, bu pe u ba ions also
impai cell iabili y, meaning obse ed e ec s may simply e lec nonspeci ic
inju y.
4. RNA-seq da a misin e p e ed
Gene exp ession changes a e shown bu no linked speci ically o modules; bulk
issue exp ession may dilu e o obscu e module-speci ic signals.
Figu e 6 — C oss-Tissue Simila i y and Uni e sali y Analysis
C i ique
1. Insu icien s a is ical igo
The igu e claims uni e sal pa e ns o module size and shape ac oss o gans bu
does no show:
o dis ibu ion o e lap me ics
6
o Kolmogo o –Smi no es s
o clus e ing alidi y indices
o c oss- alida ion esul s.
2. Possible scale-no maliza ion a i ac s
When issue images a e no malized o simila scale anges, a i icial simila i y
eme ges. The igu e does no demons a e ha simila i y pe sis s wi hou
no maliza ion.
3. U-MAP embeddings no ep oducible
U-MAP is s ochas ic and sensi i e o hype pa ame e s. No obus ness analysis is
shown ac oss seeds, pe plexi ies, o n_neighbo s.
4. Gene aliza ion o e each
Ex apola ing om a limi ed panel o issues o all o gans is scien i ically
p ema u e.
Figu e 7 — Func ional In eg a ion: Modules as Ga es o O gan-
Le el Ou pu s
C i ique
1. Cause–e ec con usion
The igu e implies modules con ol o gan unc ion, ye :
o unc ional measu emen s a e coa se
o module manipula ions a e indi ec
o al e na i e explana ions (e.g., ne wo k edundancy) a e no excluded.
2. Spa se elec ophysiological da a
The ca diomyocy e “module ga ing” da a appea unde powe ed, wi h no single-
cell esolu ion unc ional eadou s.
3. S a is ical inconsis encies
No ANOVA, mul i a ia e eg ession, o e ec size plo s a e p o ided. Repo ed p-
alues appea inconsis en wi h he p esen ed sca e plo s.
4. O e -asse i e conclusions
The igu e i le sugges s a de e minis ic unc ional ole ha is no con incingly
demons a ed by he p esen ed e idence.
7
Figu e 8 — Syn hesis Model and P oposed O gan-Le el Modula
Hie a chy
C i ique
1. Hypo he ical diag am depic ed as p o en
The igu e p esen s a hie a chical model (cells → modules → mac o-domains →
o gans) as ac , despi e limi ed expe imen al suppo .
2. No unce ain y ep esen a ion
The diag am lacks:
o e o bounds
o p obabilis ic nodes
o al e na i e hypo heses (e.g., g adien s ins ead o disc e e modules).
3. O e gene aliza ion
Claiming ha all issues ollow an iden ical modula hie a chy igno es:
o issues wi h di use o ganiza ion (e.g., spleen ed pulp)
o issues wi h con inuous a chi ec u es (e.g., li e lobules o e lapping
s uc u es).
4. No p edic i e alida ion
A undamen al es would be whe he he model p edic s unseen issue
a chi ec u e. No such alida ion is shown.
3. Ex ended Da a Figu es: C i ical E alua ion
Below we c i ically e alua e each Ex ended Da a (ED) Figu e. Because he au ho s ely
hea ily on ED igu es o jus i y me hodological and compu a ional claims, hese equi e
igo ous sc u iny.
Ex ended Da a Figu e 1 — Raw Imaging Examples o All Tissues
S udied
C i ique
1. Raw images no uly aw
“Raw da a” appea s al eady denoised and con as -adjus ed. T ue aw da a (e.g.,
TIFF s acks) a e no p o ided.
8
2. Une en imaging condi ions ac oss issues
Some issues a e imaged wi h con ocal, o he s wi h ligh -shee ; some use
nuclea s ains, o he s memb ane ma ke s. Such he e ogenei y makes c oss-
issue compa ison un eliable.
3. Inconsis en Z-s ack co e age
Di e en hicknesses gi e di e en appa en densi y o “modules,” con ounding
segmen a ion.
Ex ended Da a Figu e 2 — Segmen a ion Valida ion
C i ique
1. Lack o g ound u h
Valida ion elies on algo i hm-algo i hm ag eemen a he han algo i hm–
expe anno a ions.
2. O e - eliance on IOU (in e sec ion o e union)
IOU alone canno assess biological co ec ness; many alse-posi i e
segmen a ions would s ill yield high IOU i bounda ies a e uni o mly shi ed.
3. Missing in e -sample a iabili y
Only wo issue eplica es pe condi ion a e shown, a below wha is necessa y.
Ex ended Da a Figu e 3 — Clus e ing Pa ame e Sweep
C i ique
1. Pa ame e anges oo na ow
The au ho s sweep hype pa ame e s only in small windows, e ec i ely
gua an eeing ha clus e s emain s able. No s ess es ing is pe o med.
2. No sensi i i y analysis
Wi hou sensi i i y cu es, we canno e alua e whe he clus e bounda ies a e
algo i hmically s able.
3. Possible manual selec ion
The chosen pa ame e anges appea uned o each isually appealing clus e s
(modules), c ea ing con i ma ion bias.
Ex ended Da a Figu e 4 — De elopmen al Compa isons Using
Addi ional Ma ke s
C i ique
1. Ma ke s no o hogonal
Many ma ke s label o e lapping cell popula ions, making i unclea whe he
changes e lec de elopmen al p og ession o ma ke eac i i y.
9
2. Lack o quan i a i e de elopmen al ajec o ies
No dimensionali y educ ion o pseudo- ime analysis is p o ided; de elopmen al
claims emain specula i e.
Ex ended Da a Figu e 5 — Mechanical Simula ions
(Supplemen a y Models)
C i ique
1. Pa ame e non-iden i iabili y
Many simula ions p oduce simila ou pu pa e ns. Wi hou iden i ying he
solu ion space, au ho s canno a gue ha eal issues uniquely ma ch simula ed
modules.
2. Simpli ied bounda y condi ions
Simula ions assume uni o m issue elas ici y—biologically un ealis ic.
3. Scale misma ch
Simula ion mesh esolu ion is o de s o magni ude lowe han cellula
dimensions, inducing aliasing a i ac s.
Ex ended Da a Figu e 6 — Pe u ba ion Con ols
C i ique
1. Con ols insu icien
Only sho -du a ion con ols a e shown. Some pe u ba ions equi e long- e m
con ols, especially ECM modi ica ions.
2. Cell iabili y no moni o ed
E ec s a ibu ed o “module dis up ion” may simply e lec widesp ead
apop osis.
Ex ended Da a Figu e 7 — Tissue-Speci ic Examples o Modules
in Unde s udied O gans
C i ique
1. Highly inconsis en s aining quali y
Some issues show weak s aining, making bounda ies appea a i icially smoo h.
2. Low esolu ion in se e al images
Low-quali y images in la e he appea ance o “modules,” especially whe e
esolu ion is insu icien o esol e cellula de ail.
