Emergence of community behaviors in the gut microbiota upon drug treatment

Pharmaceuticals can directly inhibit the growth of gut bacteria, but the degree such interactions manifest in complex community settings is an open question. Here we compared the effects of 30 drugs on a 32-species synthetic community with their effects on each community member in isolation. While most individual drug–species interactions remained the same in the community context, communal behaviors emerged in 26% of all tested cases. Cross-protection, during which drug-sensitive species became protected in community, was 6-times more frequent than cross-sensitization, the converse phenomenon. Cross-protection decreased and cross-sensitization increased at higher drug concentrations, suggesting that the resilience of microbial communities can collapse when perturbations get stronger. By metabolically profiling drug-treated communities, we showed that both drug biotransformation and bioaccumulation contribute mechanistically to communal protection. As a proof-of-principle, we molecularly dissected a prominent case: species expressing specific nitroreductases degraded niclosamide, thereby protecting both themselves and sensitive community members.

Commonly prescribed therapeutics are associated with changes in the composition and 50 function of the human gut microbiome 1,2 . Hundreds of drugs, including both antibiotics and 51 those targeting human proteins, can directly inhibit the growth of commensal gut bacteria at 52 physiologically-relevant concentrations 3,4 . Reciprocally, drug sequestration and/or 53 biotransformation by gut bacteria can affect the bioavailability, efficacy, mode of action, and 54 adverse effects of pharmaceuticals, thereby contributing to the interpersonal variability of 55 drug responses 5,6 . Further molecular understanding of such drug-gut microbe interactions is 56 crucial to design improved therapies with fewer side effects, including dysbiosis. 57 58 Several studies have used in vitro, ex vivo, and in vivo approaches to probe the impact of a 59 limited set of drugs in diverse communities, showing both that drugs affect the community 60 biomass and structure [7][8][9][10][11] and that communities affect drug activity, via mechanisms that 61 include drug biotransformation 6,8,[12][13][14]  Here, we assembled a synthetic community containing 32 representative species of the 69 healthy human gut microbiota 3 and compared the effect of 30 diverse drugs on 21 species 70 reproducibly detected in the community versus in isolation. We detected at least one species 71 being protected or sensitized in the community setting for all drugs tested, and in total a 72 quarter of all drug-microbe interactions (465/1823 cases) changing in the community setting. 73 Cross-protection was the most frequent scenario, indicating that communities are more 74 resilient to external insults than individual bacteria. However, such communal protective 75 strategies decreased with increasing drug concentrations, while, cases of cross-sensitization 76 increased. Thus, at higher drug concentrations, communities are disturbed the most -not 77 only because more species may be targeted by the drug, but also because communities lose 78 capacity for cross-protection and negative interactions (cross-sensitization) increase. 79 Moreover, we demonstrated that both drug biotransformation and drug bioaccumulation 80 contributed to many cross-protection instances, and mechanistically dissected a case of 81 communal protection, identifying the species protecting the community and the enzymes 82 degrading the drug. Using the knowledge about detoxifying species we could design 83 synthetic communities that would enable growth of otherwise highly sensitive communities, 84 opening the path for future use of such knowledge to optimize community composition to 85 reduce adverse drug affects or increase drug efficacy. Overall, we provide insights into the 86 degree of emerging behaviors upon treatment of microbial communities with drugs, identify 87 some of their underlying mechanisms and map their dependence on drug concentration. 88 89

Results 90
Evaluating the impact of drugs on the composition of a complex synthetic community 91 We assembled a synthetic community of 32 species from 26 genera across 6 phyla (Suppl. 92 Table 1), cultured at 37 °C under anaerobic conditions in Gifu Anaerobic Medium Broth, 93 Modified (mGAM). Species were selected to be representative of the gut microbiome of 94 healthy humans 3 . From over 1,200 drugs previously tested against gut microbes in isolation 95 cases in which we did not obtain reproducible growth data in monocultures and the highest 133 concentration for 3 drugs (chlorpromazine, ciprofloxacin, and doxycycline), in which the 134 community did not grow at all. Three outcomes were identified ( Figure 1C, E & 2A, using 135 methotrexate as example): i) expected outcome-growth was similarly affected (Veillonela 136 parvula) or unaffected (Escherichia coli) in both community and monoculture; ii) cross-137 sensitization (emergent communal behavior) -the species growth was not affected by the 138 drug when alone, but its abundance was reduced in the community (Fusobacterium 139 nucleatum); iii) cross-protection (emergent communal behavior) -the species was inhibited 140 in monoculture but grew normally in community (Erysipelatoclostridium ramosum). To 141 assess the degree of cross-protection and cross-sensitization per drug, we calculated the 142 percentage of species that were either protected or sensitized in the community. For 143 protection we divided by the total number of species that the drug inhibited in monoculture 144 (as only those could be protected in community), whereas for sensitization we divided by the 145 total number of species that grew normally in single-species experiments ( Figure 2C, for the 146 treatment amount closest to the gut concentration; Figure S2 for all treatment conditions). 147 We observed at least one emergent behavior for all drugs probed, and in total 26% of drug-148 microbe interactions changed in the community setting. When taking into account only drug-149 sensitive species, protection in the community amounted to 47% of all cases, whereas 150 community-specific sensitization was observed in 8% of all cases of resistant species. At the 151 drug concentrations closest to the estimated human gut concentrations, these fractions were 152 very similar with 49% and 9%, respectively. Overall, this suggests that numerous 153 community-dependent protection and sensitization events can be expected upon drug 154 treatment of human gut microbiotas, and those can vary between individuals, which harbor 155 different community compositions. 156 157 High drug concentrations overwhelm community resilience 158 For each drug, we tested 3 concentrations. When starting from concentrations at which there 159 was at least one sensitive species in monoculture (to be able to detect cross-protection), we 160 could detect a significant drop in the percentage of protected species within the community 161 across all drugs as the drug concentration increased ( Figure 2D). Vice versa, the percentage 162 of sensitized species significantly increased at higher drug concentrations ( Figure 2E). Since 163 the concentration steps were not always equally spaced, we also verified the concentration 164 dependence in a separate model that uses the actual drug concentrations ( Figure S3A-B). 165 Overall, this means that the community stays relatively unaffected at low drug 166 concentrations, since the perturbation is buffered and sensitive species are protected. In 167 contrast, the impact on composition increases disproportionally at higher drug concentration, 168 as not only the community fails to protect other members, but negative interactions emerge 169 that sensitize otherwise resistant species. 170 171 Bacterial drug biotransformation and bioaccumulation underpin emergent cross-172 protection 173 Many mechanisms could be driving the prevalent emergent behaviors we observed: 174 interspecies interactions, new niches created by reduced growth of some species, altered 175 stress responses in presence of others and/or modifications in the drug availability. Since 176 previous work had showcased the extended ability of gut microbes to transform or 177 intracellularly accumulate drugs 6,12,13 , we decided to assess the degree to which emergent 178 communal phenotypes, and especially cross-protection, could be explained by such 179 phenomena. To this aim, we measured drug concentrations over time using liquid 180 chromatography-coupled mass spectrometry (LC-MS) in the same synthetic community 181 upon treatment with the same panel of drugs as above, typically at the concentration for 182 which we observed the highest percentage of emergent behaviors ( Figure S2). Samples 183 were collected on the second day of treatment, at different timepoints after passage into 184 fresh, drug-exposed medium (0, 1.25, 2.5, 5, 7.5 and 10 h after treatment). We collected two 185 fractions per community: one containing the whole community (WC), i.e. both supernatant 186 and bacteria, and one containing only supernatant (SN), to be able to distinguish between 187 biotransformation and bioaccumulation 6 . In parallel, we used only mGAM and the same time 188 course to assess drug decay in the medium. For each time course, we normalized 189 concentrations to the maximum value among the two first time-points of the time course and 190 calculated AUCs. Overall biological replicates were consistent with mean standard deviation 191 between replicates for all measurements being 8% for both community experiments and 192 media control. From the time courses of each drug, we used AUCs to calculate the extent of 193 biotransformation (media control minus WC), bioaccumulation (WC minus SN) and of both 194 phenomena (media control minus SN) ( Figure 3A and S4A). For example, the antibiotic 195 ciprofloxacin was stable across all conditions; the proton-pump inhibitor lansoprazole 196 decayed on its own, and this process was accelerated by the community; the antiparasitic 197 niclosamide was rapidly metabolized by the community; and the antimalaria drug mefloquine 198 was both bioaccumulated and biotransformed ( Figure 3A). 199

200
We found an overall positive correlation between the fraction of species that were protected 201 by the community and the degree to which the drug was biotransformed by the community 202 showing a significant correlation with protection ( Figure 3B and S4B-C). When combined 206 with biotransformation, bioaccumulation often increased the overall correlation with the 207 fraction of protected species in the community ( Figure 3B and S4B). Overall, this means that 208 the protective community effects can be at least partially explained by the drug being 209 transformed (and to some degree accumulated) by one or more species in the community. 210 Nonetheless, there are other mechanisms at play that remain to be elucidated in the future; 211 for example, dienestrol was neither biotransformed nor bioaccumulated, yet 49% of 212 susceptible species were protected in the community. In such cases, altered metabolism of 213 drug-treated species may create new niches or neutralize the drug effect for the sensitive 214 species. In contrast to communal protection, cross-sensitization did not correlate with 215 biotransformation, and only partially correlated with bioaccumulation ( Figure S4B-C). This 216 could be due to the smaller number of cases identified. Having a better understanding of 217 what drives cross-protection, we decided to study some of the underlying mechanisms in 218 further detail. 219 220 Different species protect against different nitroaromatic drugs in communities 221 All three nitroaromatic drugs used in the screen, the antiparasitic drugs niclosamide and 222 nifurtimox, and a drug to treat Parkinson's disease, entacapone, were rapidly biotransformed 223 by the community (Figure S4A). At the highest tested concentration, these drugs inhibited on 224 average 92% of the tested species in monoculture (Table S1). In line with the observed 225 biotransformation, we found 47% of these sensitive species to be protected by the 226 community ( Figure S2). Since both activation and detoxification of nitroaromatic compounds 227 rely on reduction of the nitro group 15,16 , we wondered whether it is the same set of microbes 228 with potent nitroreductases that efficiently transformed the drugs to an inactive form. 229

230
To explore this further we selected seven species from the community, which covered a 231 wide range of sensitivities to niclosamide in monoculture ( Figure 4A). Using LC-MS, we 232 checked for the ability of these seven species to reduce niclosamide to its amine form. Three 233 out of the four (partially) resistant species in monoculture (Roseburia intestinalis, 234 Coprococcus comes and Fusobacterium nucleatum, Figure 4A) stoichiometrically and 235 rapidly reduced niclosamide to aminoniclosamide ( Figure 4B-C and S5A), which was not 236 toxic to any of the gut bacterial isolates tested ( Figure S5B). Indeed, this biotransformation of 237 the drug to the inactive amino form resulted in protection of niclosamide-sensitive species 238 from niclosamide toxicity, as we showed by growing sensitive species in spent media of 239 bacteria with biotransformation capacity ( Figure 4B). The degree of protection of sensitive 240 species was directly related to the ability of the protecting species to degrade the drug 241 ( Figure 4C). 242

243
In contrast to the other three (partially) resistant species, we found that the most resistant 244 species, E. coli ED1a, could not degrade niclosamide, and the drug was even more stable in 245 the culture than in the medium control ( Figure 4C). We hypothesized that E. coli ED1A 246 bioaccumulates niclosamide, preventing the non-enzymatic reduction of the drug 17 in the 247 medium. Indeed, by using drug-treated cultures and spent media, we showed that a 248 significant amount of niclosamide bioaccumulated in E. coli ED1a (~ 2 µM in 5 h, Fig. 4D) 249 without affecting the fitness of the organism. However, this amount of bioaccumulation in E. Streptococcus parasanguinis ( Figure S5C), only partially metabolized the drug after 7.5 h 257 incubation and hence offer limited protection to sensitive species ( Figure S5D). Overall, this 258 highlights the diversity of drug-microbe interaction mechanisms, even with drugs with the 259 same functional group that is subject to the same type of bacterial transformation. It also 260 highlights that the degree of resistance is not predictive of the ability to protect other species 261 in the community, as resistance mechanisms differ. 262

263
We further wondered whether we could use this knowledge of selective protection to 264 engineer communities that are more robust to drug treatment. To do this, we grew a small 265 community of gut bacteria sensitive to niclosamide, composed of Bifidobacterium longum, 266 Bacteroides uniformis, S. parasanguinis, S. salivarius (Minimal Inhibitory Concentrations ≤ 267 1.25 µM), to which we added or not the detoxifier C. comes ( Figure 4E). C. comes restored 268 the growth of the community, even after 10 µM niclosamide treatment, yielding species 269 relative abundances similar to those of the untreated control ( Figure 4F and diverse family of proteins, mainly present in bacteria, which have been recently 281 reclassified into 14 subgroups according to their sequence similarities 19 . We found that all 282 species in our community encode at least two of these enzymes belonging to different 283 subgroups (Table S2). To further understand the basis of niclosamide reduction, we focused 284 on two species: a) R. intestinalis, moderately resistant to niclosamide and a good protector, 285 encoding only two putative FMN-dependent nitroreductases (other resistant species 286 encoded more nitroreductases), and b) P. vulgatus, sensitive to niclosamide (MIC = 0.625 287 µM), but intriguingly encoding seven putative nitroreductases (Table S2). To test the ability 288 of each of these nitroreductases to degrade niclosamide, we cloned and overexpressed 289 them in the sensitive P. vulgatus. Overexpressing any of the two R. intestinalis 290 nitroreductases conferred 8-fold or more increase in MIC to niclosamide for P. vulgatus 291 ( Figure 5A & S5E), suggesting that these enzymes are responsible for the resistance of R. 292 intestinalis in niclosamide. Indeed, nitroreductase C7GA87 was highly expressed in R. 293 intestinalis ( Figure 5B). Interestingly, overexpression of two out of the seven P. vulgatus 294 nitroreductases conferred also significant resistance to niclosamide ( Figure 5A & S5E). We 295 reasoned that these proteins should be silent or lowly expressed from endogenous locus, 296 and hence P. vulgatus is sensitive to niclosamide. Indeed, both proteins had low abundance, 297 which was not further induced by niclosamide in monoculture ( Figure 5B). Overexpression of 298 nitroreductase Pv2039 led to > 16-fold increase in protein levels (Table S2)  Overall, these results indicate that nitroreductases are specific to the substrate. When we 304 overexpressed the same nitroreductases as the ones above in P. vulgatus, resistance to 305 nifurtimox did not change ( Figure S5E), which was consistent with the sensitivity of both P. 306 vulgatus and R. intestinalis to nifurtimox ( Figure S5C). Overall, our data suggest that that 307 many species may have the capacity to biotransform drugs, but the respective selective 308 enzymes may not be expressed and/or induced upon drug treatment. 309 310 Discussion 311 In recent years, advances in gut microbiota culture techniques have made it possible to 312 systematically determine the direct interactions between hundreds of commonly used drugs 313 and specific members of the gut microbiota 3,7,13 . However, little is known about whether 314 these direct drug-bacterial species interactions are relevant when the same strain/species is 315 part of a bacterial community. Here we show that 74% of a total of 1823 directly determined 316 drug-species interactions remained the same in the context of a community. Nevertheless, 317 communal behaviors were substantial, and were present in every drug we tested. Protection 318 of sensitive species was the most common outcome, and even more prevalent at low drug 319 concentrations. At higher concentrations, closer to that found in the colon, community 320 protection decreased, with more species behaving the same as when growing alone. Hence, 321 despite the occurrence of communal behaviors, single species-drug interactions are 322 relatively good predictors of what will happen to a species when it is part of a community. 323 This is consistent with decades of clinical work where antibiotic sensitivity of enteric 324 pathogens is tested in isolation and not in community settings. 325

326
In other contexts, it has been shown that positive interactions between community members 327 increase with the degree of species dissimilarity 20 . We assembled a diverse community of 328 32 bacterial species belonging to 26 genera, representative of the human intestinal 329 microbiota from healthy individuals 3,21 . It would be interesting to establish whether 330 communities with less phylogenetic diversity, typically found in dysbiosis, maintain the same 331 degree of resilience and cross-protection to drugs as that observed here. 332 333 Based on targeted metabolomics data, we established that bacterial drug biotransformation 334 partially explains community protection phenotypes. This implies that in most cases drug 335 biotransformation yields harmless or less toxic products. In our in vitro setting, drug 336 bioaccumulation only mildly correlated with protective phenotypes. This could be because 337 we added the drug during passaging, meaning that at the beginning there is a relatively low 338 bacterial biomass that could accumulate the drug. Bioaccumulation has only been recently 339 reported for drugs and gut microbes 6 . In addition to the previously reported drugs, we found 340 here that intestinal bacteria bioaccumulate four more drugs: ebselen, mefloquine, 341 simvastatin and tamoxifen. Ebselen may accumulate due to its ability to bind to thiol groups 342 22 . In the case of tamoxifen or mefloquine, bioaccumulation could be a result of their ability to 343 interact with bacterial membranes, as previously shown for Bacillus stearothermophilus and 344 E. coli, respectively 23,24 . The case of simvastatin is interesting, since previous studies 345 suggest a link between microbiome composition and the efficacy of the drug in lowering LDL 346 cholesterol in patients [25][26][27][28][29] . Statins are one of the most prescribed drug groups in western 347 countries 30 , and hence commonly found into wastewater 31 . Due to their high water-octanol 348 partition coefficient, statins tend to bioaccumulate in aquatic animals, causing a serious 349 environmental problem 32,33 . It would interesting investigate in the future if bioaccumulation of 350 statins in bacteria has implications for drug efficacy in patients, and/or may also affect the 351 accumulation of the drug in environmental reservoirs. 352 353 Our data underline the broad ability of the microbiota to transform xenobiotics. Working with 354 a bottom-up assembled synthetic community allowed us to gain insights into the 355 biotransformation mechanisms that lead to communal protection. We focused on 356 nitroaromatic compounds, and observed that the nitroreductases that reduce them are rather 357 specific. For example, both nitroreductases of the niclosamide-resistant R. intestinalis 358 efficiently reduced niclosamide but not nifurtimox. Importantly, P. vulgatus encoded 359 nitroreductases that could render it resistant to niclosamide, but did not express them in 360 relevant amounts even in the presence of the drug. This denotes that gut bacteria likely have 361 an even larger potential to transform drugs than previously appreciated, which would render 362 them able to evolve resistance if exposed to the drug for longer periods. 363

364
Albeit less frequent at low drug concentrations, cross-sensitization became more dominant 365 as emergent behavior at higher drug concentrations. Cross-sensitization may arise from 366 altered biotransformation capacities of the strain accumulating the drug 6 or induction of toxic 367 stress responses by strains targeted by the drug. Although the underlying mechanisms 368 behind such cases remain to be explored, cross-sensitization further disturbs community 369 composition. Together with loss of cross-protection at high drug concentrations, they 370 presumably lead to an accumulative strong impact in community stability. 371

372
In summary, we established that most drug-gut bacteria interactions remain the same in 373 communities, but cross-protection and/or cross-sensitization of specific community members 374 happens for almost every drug. The knowledge of such interactions that change in the 375 community setting could help to build more accurate predictive models of community 376 responses to drug in the future. We further showed that communities have higher resilience 377 than individual species to certain drugs, but up to a certain concentration limit, after which 378 communal protection drops and cross-sensitization and individualized behaviors prevail. This 379 resilience is partially explained by bacterial drug biotransformation activities, which is 380 facilitated by the expanded functional diversity of the community. However, this is not the 381 only mechanism by which communities become resilient. Understanding the remaining 382 underlying mechanisms can facilitate the targeted design of designed communities that are 383 resilient to specific drugs, as we show here for niclosamide. Our work also contributes to 384 expanding the notion that drug bioaccumulation is widespread among gut bacteria 6 . 385 Understanding the drivers of bioaccumulation, as well as its potential to affect drug mode of 386 action, merit both deeper exploration. Overall, we have used a bottom-up approach to 387 assess the degree of communal behaviors that emerge in response to drugs, and to map 388 some of the underlying mechanisms that govern interactions between drugs and gut 389 bacteria. carrying different nitroreductases were treated with DMSO control, 10 µM, or 20 µM 544 niclosamide for 5 h, after which cultures were filter-sterilized. Spent media from these 545 cultures was mixed 1:1 with fresh mGAM, and used to grow niclosamide sensitive species. 546 547 (D) Following the scheme in (C), growth of the indicated strains in spent media from 548 niclosamide treated cultures. Growth of the sensitive strains in spent media of treated 549 cultures was normalized with their growth in spent media from untreated cultures (controls). 550

Bacterial species and growth conditions 552
Species used in this study were purchased from DSMZ, BEI Resources, ATCC and Dupont 553 Health & Nutrition, or were gifts from the Denamur Laboratory (INSERM) and processed as 554 previously described 3 . All species (monoculture or community) were grown in mGAM 555 (HyServe GmbH & Co.KG, Germany) except the monocultures of Veillonella parvula and of 556 Bilophila wadsworthia, which were grown in Todd-Hewitt Broth supplemented with 0.6% 557 sodium lactate and mGAM supplemented with 60 mM sodium formate and 10 mM taurine, 558 respectively. Media was pre-reduced to a minimum of 24 h under anoxic conditions (2% H 2 , 559 12% CO 2 , 86% N 2 ) in an anaerobic chamber (Coy Laboratory Products Inc.). Unless 560 otherwise specified, experiments were performed at 37°C and under anaerobiosis, species 561 were inoculated from frozen stocks into liquid culturing media, and passaged twice overnight 562 to ensure robust growth, and drug stocks in DMSO were pre-reduced overnight in the 563 anaerobic chamber inside an ice-box at 4 °C. 564 A representative core of species in the human gut microbiome were selected as previously 565 described 3,21 . From this core, 32 diverse species, differing more than 3% on their 16S rDNA 566 sequences at the V4 region, were selected for the screening. 567 568

Chemical screening of a bottom-up assembled bacterial community of 32 species 569
Screening plates preparation 570 Drugs were dissolved with the appropriate solvent (Table S2) at a concentration 100-fold 571 higher than the screening concentration, distributed in 96-well V-bottom plates (Greiner, 572 651261), each well containing 11 µl of dissolved compound or vehicle, and stored at -30°C 573 for up to 1 month. One day before the experiment, drug plates were thawed, and each of the 574 11 µl of drug or vehicle per well were added to 539 µl/well of mGAM in 96-deep well plates 575 (Costar 3959) using the Biomek FXP (Beckman Coulter) liquid handling system. These 576 plates were pre-reduced in the anaerobic chamber overnight ("Community Plates Day1"). 577 Inoculation of community passage 1. For community assembly, the optical density (OD) was 578 individually measured at 578 nm for the 32 species. These were added together into 200 ml 579 of mGAM with the volume required to reach a total of 2x the desired initial OD of 0.01: 580 therefore, each species was added at an OD of 0.0006. 550 µl of the assembled community 581 was added into each well of the "Community Plates Day 1" with an epMotion 96 (Bio-Rad) 582 semi-automated electronic 96 channel pipette. Final drug concentrations are described in 583 Table S1 and each well contained 1% DMSO. After inoculation, 100 µl/well were transferred 584 from the "Community Plates Day 1" to U-bottom shallow 96-well plates ("Community Growth 585 Plates Day 1") (Fisher Scientific 168136), sealed with breathable membranes (Breathe-Easy, 586 Sigma), and incubated at 37°C. These plates were used for monitoring community growth 587 after drug treatment, for which OD 578 was measured with a microplate spectrophotometer 588 (EON, Biotek) every hour during 24 h, with shaking only few seconds before OD 589 measurement. The "Community Plates Day 1" were also sealed with a breathable 590 membrane (AeraSeal, Sigma) and incubated during 24 h. The untreated community grew to 591 an average OD 578 of 1 (7 generations). Inoculation of community passage 2. After the initial 592 24 h incubation, the drug-treated community was passaged with a 1:50 dilution (OD 578 of 593 0.02 for the untreated community) into a new drug-containing deep-well plate for an 594 additional 24 h incubation as follows: during the initial 24 h incubation period, new drug 595 plates were thawed, and the 11 µl of drug or vehicle was added to 1067 µl/well of mGAM in 596 96-deep well plates ("Community Plates Day 2"). These were pre-reduced in the anaerobic 597 chamber overnight. Wells in the "Community Plates Day 1" were mixed with the 96 channel 598 epMotion pipette, and from here 22 µl were transferred to the "Community Plates Day 2". Of 599 these, 100 µl were transferred to 96-well U-bottom plates ("Community Growth Plates Day 600 2"), for growth curve acquisition, as described above. The "Community Plates Day 2" were 601 incubated for 24 h anaerobically and at 37°C. The untreated community grew to an average 602 OD 578 of 1 (6 generations). After 24 h of incubation, cell pellets were collected by 603 centrifugation. DNA extraction and 16S rDNA sequencing were performed as described 604 below. These experiments were performed in 3 biological replicates with 2 technical 605 replicates each. 606

Genomic DNA isolation 608
Genomic DNA isolation was done as previously described 4 . Briefly, DNA was extracted in a 609 96-well plate format using a Biomek FXP (Beckman Coulter) liquid handling system or in 610 single tubes depending on the number of samples. Cells were first washed with PBS and 611 resuspended in 281 µl of cell suspension solution (MP GNOME DNA kit). Cell suspension 612 was treated with lysozyme (25 µl; 400.000 U/ml) and incubated for 1 h at 37 °C. Cell 613 suspensions were then further lysed by three freeze/thaw cycles using liquid nitrogen, before 614 the addition of 15.2 µl of cell lysis solution (MP GNOME DNA kit) and 20 µl of RNA mix (MP 615 GNOME DNA kit). A last step of lysis was performed using glass beads (Glasperlen,616 Edmund Bühler) by bead beating twice for 5 min at 30 Hz in a Tissue Lyzer II (QIAGEN). 617 Lysates were then incubated for 30 min at 37 °C with shaking. 12.8 µl of protease mix (MP 618 GNOME DNA kit) was subsequently added and the lysates were incubated for 2 h at 55 °C. The 16S libraries were then prepared for sequencing using a two-step PCR method 632 according to 35 , using the Phire Hot Start II DNA polymerase (Thermo Scientific). Briefly, the 633 V4 region was amplified by a first PCR with the 515F/806R primers (Table S2). The resulting 634 amplicons were subsequently amplified again using barcoded primers that contain Illumina 635 adaptors. These libraries were sequenced in the EMBL GeneCore sequencing facility on an 636 Illumina MiSeq (250 base pairs, paired-end). 637 638

Processing of 16S data and quantification of species abundances 639
To estimate the species abundance, a database of 16S rRNA regions was constructed by 640 manually querying the SILVA rRNA database 36 and extracting the representative sequence 641 from each of our 32 species. Amplicon sequencing reads were then mapped against this 642 database using MAPseq v1.2. 37 . Paired reads were mapped independently and 643 assignments were only considered upon agreement. Abundance estimates were then 644 produced by counting the number of reads mapping to each genome included in the study. 645 Eleven species whose median read count in controls was below 10 were designated as rare 646 species and excluded from the subsequent analysis, as for these species abundance ratios 647 would be unreliable. Relative species abundances were calculated by dividing the number of 648 reads mapping to each species by the total number of reads for a sample. To estimate 649 absolute species abundances, we multiplied the relative abundances by the final normalised 650 OD (which was set to 1 for control wells). To ensure that relying on OD as a proxy for cell 651 counts does not lead to large distortions, we also calculated standardized relative 652 abundances by dividing the relative abundances within each condition by the 75 th percentile 653 of the relative abundances. We chose the 75 th percentile instead of the median, as this is 654 more robust to the case where the growth of many species is inhibited. 655 656

Quantification of treatment effects 657
Control abundances were separately determined for each biological and technical replicate 658 by calculating the mean abundance across six control wells. For each species and treatment 659 condition, we calculated the ratio of treatment and control abundances to estimate the effect 660 of drug treatment on each species independently of its absolute abundance. Species whose 661 abundance was decreased to 50% or less upon drug treatment were designated as being 662 reduced in the community. 663 664 Community metabolomics 665 LC-MS analyses were employed to assess the presence/absence of a drug after incubation 666 with the community of 30 species (compared to the initial community, Ruminococcus bromii, 667 and Ruminococcus torques, which did not grow reliably in the community, were not added as 668 cells did not grow from stocks). Drug plates, community assembly, passages and 669 incubations were performed as described above, with the only exception that the community 670 inoculation in the 2 nd passage was performed in a final volume of 1.5 ml per well. Final drug 671 concentrations are listed in Table S1. Immediately after inoculating the community in the 2 nd 672 passage: i) 100 µl were transferred to 96-well U-bottom plates for growth curve acquisition, For inoculation, the second overnight culture was diluted into fresh medium to an OD 578 of 735 0.02 (2x). 50 µl/well were added to the 96-well U-bottom drug containing plates, to final drug 736 concentrations as indicated in Table S2, 1% DMSO, and starting bacterial cultures at OD 578 737 of 0.01. Plates were sealed with breathable membranes and incubated at 37 °C, with 738 shaking only a few seconds before OD measurement, as described above. These 739 experiments were performed in 3 biological replicates with 2 technical replicates each. 740 741

Single-species growth curves 742
Growth curves for single species were determined for the 21 species that were consistently 743 detected in the in vitro communities (Table S1). As previously described 3 , growth curves 744 were quality-controlled, and truncated at the end of the exponential phase under control 745 conditions. The AUC was calculated for all conditions and divided by the median control 746 AUC within each plate. Across all treatment conditions, the standard deviation between 747 biological and technical replicates had a mean value of 0.04 and 0.03, respectively. In our 748 previous screen 3 , a large number of control wells and wells with inactive drugs made it 749 possible to calculate a distribution of AUCs for normal growth, and to calculate p-values 750 based on this distribution. In this more focused screen, this was not possible and we 751 therefore opted to use an AUC threshold of 0.75 (i.e. 25% reduced growth) to determine 752 whether a species was susceptible to drug treatment. that are protected in the community for a certain treatment condition, we only considered the 802 subset of species that are affected by the drug in the monoculture experiment. For this 803 subset of species, we divided the number of species that grew normally in the community by 804 the number of affected species. Conversely, to determine the fraction of species that are 805 cross-sensitized in the community we divided the number of species that showed reduced 806 growth in the community (but grew normally in monoculture) by the number of species that 807 grew normally in the monoculture experiment. This calculation was done on the level of 808 technical replicates from the community experiment. 809 810

Concentration dependency 811
The concentration dependency of the community effects was calculated both on the 812 concentration steps of the drug (low/middle/high) and the actual numerical concentration. In 813 the first instance, we determined for each drug the concentration steps for which the 814 community effect could be determined. For example, consider a drug which only reduced the 815 growth of any species monoculture at its middle and high concentration but not at its lowest 816 concentrations. For determining the concentration dependency, these would be considered 817 as the low and middle concentration. The distributions of the fractions of affected species 818 were compared using Wilcoxon signed-rank tests. 819 In addition, we fitted separate logistic functions for sensitization and protection for all drugs 820 which were affected at more than one concentration step. Parameters are a common growth 821 rate k and drug-specific offset d i to capture the relation between the fraction of affected 822 species f i,j for the concentration step c i,j : 823 We compared the resulting model with a simplified model that has no concentration 824 dependency, i.e., only one constant value for each drug. We ascertained that the 825 concentration-dependent model provides a better fit to the data using both an ANOVA 826 (analysis of variance) and the BIC (Bayesian information criterion). 827

Spent media experiments 829
Spent media experiments were used to evaluate specie's effects on drugs and whether 830 these protect (potential protectors) other species (drug-sensitive species) from drug toxicity. 831 All species were inoculated from glycerol stocks into liquid medium, passaged for two days, 832 and diluted to final OD 578 of 0.01 to perform the experiments. Potential protectors and mGAM 833 controls were grown, for the indicated times, in the presence of 2x the indicated drug 834 concentrations. After the incubation periods, the potential protectors and mGAM controls 835 were sterilized with 0.22 µm PVDF filters (Millipore). 50% of the filtrate (growth spent media) 836 was diluted with 50% of fresh mGAM containing drug-sensitive species at OD 578  Coprococcus comes at a starting OD 578 nm of 0.01 were kept untreated or treated with 10 847 µM niclosamide for 24 h. At 24 h pellets were collected by centrifugation and DNA was 848 isolated and bacterial community composition was quantified by qPCR. Community growth 849 was monitored in 96-well U-bottom plates as described above. 850 851 Bacterial community composition quantification by qPCR. 852 Primer design. Species-specific primers were designed using the NCBI primer design tool by 853 selecting the following common parameters: product length of 200 bp maximum, 25 bp 854 primer length, Tm 70 °C ± 3 °C, G/C content < 60%. Primers were tested in silico for 855 homology to non-specific sites against nr and RefSeq representative genomes by BLAST,856 and in vitro by PCR against genomic DNA isolated from each individual species included in 857 the community. The 515F/806R primers 35 (Table S2) targeting the V4 hypervariable regions  858 of 16S rDNA were used for the amplification of the 16S rDNA of the community in each 859 sample. Four-point standard curves were prepared from ten-fold serial dilutions of DNA 860 prepared for each individual species, starting at concentration of ~1 ng/µL using 1 µL per 861 well in duplicate reactions (range from ~1 ng -2 pg). The standard curves demonstrated 862 good linearity in four orders of magnitude (R 2 of 0.989-1.000) for all DNAs except for DNA 863 isolated from B. uniformis, where linearity was three orders of magnitude. Species specific 864 primer's efficiencies ranged from 91 to 100 %, 16S rDNA primer's efficiencies were 84.5% ± 865 5.5%. Reaction conditions. 20 µL qPCR reactions containing 10 µL SYBR TM master mix 866 (Applied Biosystems), 1 µL nuclease-free water, 4 µL of 4 µM primer mix (0.2 µM each final), 867 and 1 µL of template DNA (5-10 ng) were run on 96-well plates on a StepOne Plus Real-868 Time PCR system (ThermoFisher Scientific). A melting curve analysis was carried out to 869 confirm the amplification of a single product in each reaction. Data analysis.

Overexpression of nitroreductases 876
Proteomics 877 For proteomics experiments, overnight cultures of the different species were diluted to an 878 OD 578 0.1, and anaerobically grown until OD 578~1 . At this point, cell cultures were treated 879 with vehicle (DMSO) or 20 µM niclosamide for 30 minutes. Cell pellets were collected by 10 880 min centrifugation at 4000 x g, washed with ice-cold PBS, and frozen at -80ºC until further 881 processing. Cells were then resuspended in a lysis buffer (2% SDS, 250 U/ml benzonase, 882 and 1 mM MgCl 2 in PBS) and boiled at 95 ºC for 10 min. After measuring protein 883 plated on mGAM plates supplemented with erythromycin (10 µg/ml) and gentamicin (200  Putative FMN-dependent nitroreductases in the gut bacterial species from this study, 929 differential proteomics analysis of R. intestinalis and P. vulgatus nitroreductases 930 overexpressing strains upon niclosamide treatment, and list of primers used in this study. 931 932