1
A C i ical Re-e alua ion o “Nu ien compe i ion
p edic s gu mic obiome es uc u ing unde d ug
pe u ba ions” by Shi e al., Cell 2025;
doi:10.1016/j.cell.2025.10.038
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]
In oduc ion
Shi e al.1 (Cell, 2025) p esen a mechanis ic amewo k p oposing ha nu ien
compe i ion is he cen al axis by which di e se d ugs eshape he gu mic obiome.
Acco ding o he au ho s, d ugs pe u b nu ien a ailabili y—di ec ly o indi ec ly—and
mic obial species espond acco ding o gene ically encoded nu ien p e e ences. Using
a mix u e o genome-scale me abolic in e ence, in i o monocul u e assays, gno obio ic
mouse expe imen s, and compu a ional ecological modeling, he au ho s a gue ha
mic obiome es uc u ing unde d ugs can be p edic ed—and mechanis ically
explained—by nu ien compe i ion indices ha in eg a e mic obial me abolic capaci ies
wi h d ug-associa ed nu ien shi s.
Undeniably, he s udy is ambi ious and con ibu es meaning ully o he ongoing deba e
ega ding he mechanis ic basis o d ug-mic obiome in e ac ions. Howe e , many o he
cen al claims a e no adequa ely suppo ed, especially hose asse ing causali y. Much
o he mechanis ic a chi ec u e is de i ed om in e ed, simula ed, o indi ec ly
econs uc ed nu ien ela ionships a he han di ec measu emen s. Mo eo e , he
hea y eliance on compu a ional p edic ions, combined wi h limi ed in i o alida ion
and he absence o key expe imen al con ols, aises conce ns abou o e -gene aliza ion.
In his commen a y, we sys ema ically e alua e he s udy igu e by igu e, highligh ing
me hodological weaknesses, o e -in e p e a ions, missing con ols, unadd essed
con ounde s, and concep ual gaps. Ou analysis concludes ha while nu ien
compe i ion may con ibu e o mic obiome es uc u ing, he e idence p esen ed he e
alls sho o demons a ing ha nu ien compe i ion is he dominan o p ima y
mechanism behind d ug-induced mic obial al e a ions.
2
Main Figu es: De ailed C i ique
Figu e 1 — D ug exposu e eshapes mic obial composi ion
h ough nu ien -compe i ion signa u es
Figu e 1 p esen s he cen al concep ual model: d ug ea men s al e gu communi y
composi ion, and hese shi s co espond o species-le el nu ien compe i ion p o iles
in e ed om genome-scale me abolic models.
S eng hs
• In eg a es mouse mic obiome da a, ex i o g ow h assays, and compu a ional
p edic ions.
• Clea g aphical p esen a ion o communi y shi s.
• Highligh s axonomic es uc u ing unde di e en d ugs.
Majo Conce ns
1. Ci cula easoning in nu ien p e e ence anno a ion
Nu ien p e e ence sco es a e de i ed om genome anno a ions using highly
au oma ed me abolic econs uc ion pipelines. These pipelines ely on gap- illing,
pa hway in e ence, and o hologian app oxima ions. Because nu ien
p e e ences a e in e ed and hen used o gene a e nu ien compe i ion sco es
ha in u n supposedly explain communi y changes, he analysis isks concep ual
ci cula i y: in e ed nu ien usage → in e ed nu ien compe i ion → in e ed
explana ion o communi y shi s.
2. Con ounding e ec s o d ug oxici y
Se e al es ed d ugs ha e di ec an imic obial e ec s o induce hos -media ed
s ess esponses. These mechanisms can independen ly in luence abundance
pa e ns, bu Figu e 1 a ibu es he en i e es uc u ing o nu ien compe i ion
wi hou s a i ying oxic s. non- oxic pe u ba ions.
3. Lack o s ain-le el esolu ion
Many composi ional changes occu a he s ain le el (e.g., E. coli clades), bu
nu ien p e e ence anno a ions a e oo coa se o suppo s ain-speci ic
p edic ions.
4. No measu emen o nu ien concen a ions
The in e ence ha d ugs modi y nu ien a ailabili y is specula i e; he igu e
lacks me abolomic alida ion om luminal samples.
3
Summa y: Figu e 1 o e s compelling desc ip i e da a bu p o ides no causal p oo
ha nu ien compe i ion d i es obse ed mic obial shi s.
Figu e 2 — P edic i e model linking d ug pe u ba ions o
nu ien -compe i ion indices
Figu e 2 con ains he co e p edic i e modeling componen , showing ha nu ien
compe i ion indices allegedly explain mic obiome es uc u ing unde d ugs.
S eng hs
• Ex ensi e compu a ional modeling.
• Imp essi e c oss- alida ed sco es.
• A emp s o in eg a e nu ien compe i ion wi h ecological modeling.
C i ical Issues
1. O e i ing and high dimensionali y
Nu ien ea u es numbe in he dozens, while sample size is limi ed. Wi hou
clea egula iza ion s a egies, he model likely o e i s o noise o axonomic
ine ia a he han genuine nu ien -based mechanisms.
2. Con ounde leakage
D ug chemical p ope ies (cha ge, pola i y, memb ane ac i i y) co ela e wi h
mic obial me abolism. I hese d ug-wide ea u es co ela e wi h simplis ic
nu ien ea u es, nu ien indices may inad e en ly cap u e d ug
physicochemis y, no nu ien compe i ion.
3. Insu icien nega i e con ols
Missing:
o Randomized nu ien labels
o Shu led species–nu ien ma ices
o Random- o es models using d ug desc ip o s as baselines
Wi hou benchma king agains simple al e na i es, nu ien compe i ion may
no be he bes p edic o .
4. Tempo al au oco ela ion no add essed
Baseline abundances co ela e s ongly wi h pos - ea men alues. I models
include p e- ea men alues, p edic ions may be d i en by composi ional ine ia
a he han nu ien dynamics.
4
Summa y: Figu e 2 shows co ela ion—no causa ion. The p edic i e model may ely
hea ily on con ounding s uc u e.
Figu e 3 — In i o g ow h esponse o d ugs unde nu ien -
limi ed condi ions
Figu e 3 a emp s o alida e he nu ien compe i ion model using monocul u e g ow h
assays ac oss nu ien condi ions wi h o wi hou d ugs.
S eng hs
• Well-designed hea maps.
• A emp s di ec expe imen al alida ion o model p edic ions.
Conce ns
1. D ug concen a ions o en un ealis ic
Many monocul u e assays use concen a ions a exceeding gu luminal d ug
le els. These sup aphysiological le els may a i icially supp ess g ow h,
p oducing nu ien -d ug in e ac ions ha do no occu in i o.
2. Monocul u e ≠ compe i i e en i onmen
Nu ien compe i ion is de ined by in e -species in e ac ions. Monocul u e
expe imen s canno cap u e eme gen ecological dynamics.
3. G ow h inhibi ion misin e p e ed as nu ien in e ac ion
G ow h de ec s in d ug + nu ien combina ions may e lec :
o Di ec an imic obial e ec s
o SOS esponses
o Memb ane dis up ion
o S ess pa hway ac i a ion
None o hese mechanisms a e disen angled.
4. No me abolic lux con i ma ion
The pape assumes nu ien usage shi s occu bu ne e measu es nu ien
up ake o me abolic lux ia iso ope acing o me abolomics.
Summa y: Figu e 3 shows al e ed g ow h pa e ns, bu i canno suppo conclusions
abou nu ien compe i ion in a communi y con ex .
Figu e 4 — D ug-speci ic communi y ajec o ies d i en by
nu ien -compe i ion in e ac ions
Figu e 4 simula es d ug-d i en communi y ajec o ies using ecological models
cons ained by nu ien compe i ion coe icien s.
5
S eng hs
• Inno a i e isualiza ion o communi y dynamics.
• A emp s o show mechanis ic eme gence om nu ien compe i ion.
C i ical Issues
1. Model assump ions p esuppose nu ien compe i ion
The ecological model p ohibi s o he o ms o in e ac ions—syn ophy, c oss-
eeding, immune modula ion— o cing nu ien compe i ion o eme ge as he
dominan ac o by cons uc ion.
2. No co-cul u e alida ion
Compe i ion coe icien s a e no empi ically es ima ed. They a e assigned based
solely on in e ed nu ien usage, which is dange ous o mechanis ic
in e p e a ion.
3. Hos cons ain s igno ed
D ug-induced hos ac o s (mo ili y, mucus, bile acids, immune ac i i y) s ongly
in luence communi y dynamics bu a e absen om he model.
4. Failu e mode analysis missing
Simula ed ajec o ies ma ch some expe imen al ou comes, bu misma ches a e
no quan i ied. Model i s appea che y-picked.
Summa y: Figu e 4’s mechanis ic s uc u e is appealing bu specula i e wi hou
expe imen al e i ica ion.
Figu e 5 — D ug-speci ic nu ien -compe i ion inge p in s
This igu e in oduces “nu ien inge p in s” ep esen ing d ug-speci ic pe u ba ions
ha explain mic obiome ou comes.
S eng hs
• C ea i e concep ual amewo k.
• Helps in eg a e mul iple obse a ions.
C i ical Issues
1. Finge p in s a e pu ely compu a ional cons uc s
No di ec e idence ha d ugs al e nu ien pools o induce nu ien shi s.
Wi hou me abolomic alida ion, inge p in s emain hypo he ical.
2. Al e na i e models no e alua ed
Finge p in s may coincide wi h:
o D ug solubili y
o Bac e ial d ug ole ance pa hways
6
o S ess esponses
Bu he pape compa es only nu ien -based models, no hese
al e na i es.
3. Gene alizabili y ques ionable
Tes d ugs a e s uc u ally simila . B oade chemical space is needed o claim
uni e sali y.
Summa y: Figu e 5 is concep ually elegan bu oo specula i e o suppo b oad
mechanis ic claims.
Figu e 6 — Gno obio ic mouse alida ion o nu ien -
compe i ion p edic ions
The in i o es s a e he mos impo an pa o he s udy—a emp ing causal alida ion.
S eng hs
• Gno obio ic mice p o ide con olled hos gene ic backg ound.
• Taxonomic shi s esemble model p edic ions a a coa se le el.
Majo Limi a ions
1. Small sample sizes (n = 3–4)
Powe is insu icien gi en biological a iabili y o mic obiomes.
2. No di ec nu ien o me abolomics measu emen
This igu e es s en i ely on indi ec in e ence. Wi hou luminal me abolomics,
claims o nu ien shi s emain specula i e.
3. Al e na i e mechanisms un es ed
D ug impac s ia immune modula ion, epi helial u no e , mucin changes, and
pe is al ic al e a ion all emain unadd essed.
4. No causal nu ien manipula ion
The c i ical es — escuing o e e sing d ug e ec s by adding/ emo ing
nu ien s—is absen .
Summa y: Figu e 6 p o ides pa ial suppo bu does no p o e he nu ien -
compe i ion hypo hesis mechanis ically.
Ex ended Da a Figu es 1–15: Full C i ique
ED Figu e 1 — Genome-based nu ien anno a ion wo k low
• O e - eliance on au oma ed me abolic da abase anno a ion.
• Gap- illed eac ions and missing pa hways no epo ed.
• No ex e nal benchma king wi h expe imen al da a.
• Unce ain y sco es no shown.
7
ED Figu e 2 — Simula ed nu ien compe i ion scena ios unde
d ug pe u ba ions
• Simula ion pa ame e s appea a bi a y wi hou jus i ica ion.
• D ug-nu ien pe u ba ions a e hypo he ical and no based on measu emen s.
• Sensi i i y analysis missing.
• Risk o ci cula easoning: simula ions assume mechanisms hey claim o p edic .
ED Figu e 3 — C oss- alida ion me ics o p edic i e models
• Only pe o mance imp o emen s highligh ed; no calib a ion analysis.
• No e alua ion on an independen ex e nal da ase .
• Con ounding ac o s no con olled.
ED Figu e 4 — Expanded nu ien p e e ence hea maps
• Se e al species ha e nu ien pa hways in e ed despi e no genomic e idence.
• Hea maps appea bina y bu nu ien capaci ies a e con inuous.
• Lack o s ain-le el a ia ion igno es c i ical ecological di e ences.
ED Figu e 5 — Pai wise compe i ion coe icien s ac oss species
• Compe i ion coe icien s in e ed solely om esou ce o e lap; no empi ical
alida ion.
• Does no ep esen con ex -dependence (e.g., c oss- eeding).
• Igno es hos -d i en niche pa i ioning.
ED Figu e 6 — D ug-speci ic nu ien pe u ba ion ec o s
• D ug-nu ien ec o s hypo hesized om genomic da a a he han measu ed
luminal nu ien s.
• No demons a ion ha d ugs ac ually al e nu ien pools.
ED Figu e 7 — Monocul u e g ow h cu es wi h addi ional d ug
condi ions
• Many cu es e lec s ess esponses, no nu ien compe i ion.
• No measu emen o in acellula me abolic shi s.
• No con ol o pH o oxida i e s ess modi ica ions caused by d ugs.
ED Figu e 8 — Me aboli e deple ion p o iles (compu a ional)
• Me aboli e deple ion simula ed, no measu ed.
• No me abolomics included o jus i y deple ion pa e ns.
8
• Resul s depend en i ely on in e ed lux models wi h la ge unce ain ies.
ED Figu e 9 — Al e na i e ecological model s uc u es
• Al e na i e models a e “ oy models,” no igo ous compa isons.
• Se ious al e na i es like s ess-o ien ed models, hos - esponse models, o bile-
acid-d i en models a e missing.
ED Figu e 10 — D ug chemical s uc u e clus e ing and nu ien
inge p in s
• S uc u al clus e ing may be con ounded by co ela ed p ope ies.
• No a emp o show ha nu ien inge p in s ou pe o m chemical desc ip o s.
ED Figu e 11 — Addi ional gno obio ic alida ion coho s
• Sample size emains small.
• Longi udinal consis ency no es ed.
• Hos physiological pa ame e s no epo ed.
ED Figu e 12 — Flux balance simula ions unde d ug-induced
nu ien cons ain s
• D ug cons ain s added a i icially; no measu ed.
• Flux p edic ions ex emely sensi i e o biomass objec i e unc ions, which a y
widely ac oss s ains.
ED Figu e 13 — Sensi i i y analysis o nu ien compe i ion
coe icien s
• Sensi i i y analysis shows high a iabili y, bu au ho s do no discuss ins abili y.
• Se e al coe icien s a y by >50% unde small pa ame e changes, unde mining
mechanis ic claims.
ED Figu e 14 — Null models o communi y assembly
• Null models insu icien ; do no include ealis ic ecological p ocesses.
• Random shu ling does no ep esen ue ecological null.
• Resul s designed o a o nu ien -compe i ion amewo k.
ED Figu e 15 — S a is ical summa y o model pe o mance
ac oss species
• No con idence in e als o many species-le el p edic ions.
• Hea ily weigh ed owa d species wi h abundan genome anno a ions.
9
• Unde - ep esen ed o poo ly anno a ed species show poo p edic ion bu a e
igno ed in conclusions.
Supplemen a y Figu es: Full C i ique
Supplemen a y Figu e 1 — De ailed d ug dose– esponse cu es
• Many dose anges exceed physiological ele ance.
• No pha macokine ic jus i ica ion o selec ed concen a ions.
• E ec s a high doses canno be ex apola ed o gu physiology.
Supplemen a y Figu e 2 — Ex ended monocul u e nu ien
expe imen s
• G ow h di e ences o en <10%, ye in e p e ed as s ong nu ien e ec s.
• No assessmen o biological a iabili y o e o p opaga ion.
Supplemen a y Figu e 3 — Raw sequencing ead dis ibu ions
Missing:
• Con amina ion con ols.
• Technical eplica es.
• Repo ing o sequencing dep h pe species.
Supplemen a y Figu e 4 — Addi ional communi y simula ions
• Simula ions ely on ixed pa ame e se s chosen o aes he ics.
• No explo a ion o pa ame e unce ain y.
Supplemen a y Figu e 5 — D ug solubili y and di usion me ics
C i ical omission:
• D ug di usion ac oss mucus o in es inal laye s is no measu ed.
• Gu luminal concen a ions in i o no p o ided.
Supplemen a y Figu e 6 — Hos ansc ip omic changes unde
d ug ea men s
Au ho s claim minimal hos e ec s, bu :
• Timepoin s limi ed.
• Mic obiome-media ed hos ansc ip s no examined.
• D ug-induced hos e ec s may p ecede mic obiome changes.
