Species interactions drive the spread of ampicillin resistance in human-associated gut microbiota

Abstract Background and objectives Slowing the spread of antimicrobial resistance is urgent if we are to continue treating infectious diseases successfully. There is increasing evidence microbial interactions between and within species are significant drivers of resistance. On one hand, cross-protection by resistant genotypes can shelter susceptible microbes from the adverse effects of antibiotics, reducing the advantage of resistance. On the other hand, antibiotic-mediated killing of susceptible genotypes can alleviate competition and allow resistant strains to thrive (competitive release). Here, by observing interactions both within and between species in microbial communities sampled from humans, we investigate the potential role for cross-protection and competitive release in driving the spread of ampicillin resistance in the ubiquitous gut commensal and opportunistic pathogen Escherichia coli. Methodology Using anaerobic gut microcosms comprising E.coli embedded within gut microbiota sampled from humans, we tested for cross-protection and competitive release both within and between species in response to the clinically important beta-lactam antibiotic ampicillin. Results While cross-protection gave an advantage to antibiotic-susceptible E.coli in standard laboratory conditions (well-mixed LB medium), competitive release instead drove the spread of antibiotic-resistant E.coli in gut microcosms (ampicillin boosted growth of resistant bacteria in the presence of susceptible strains). Conclusions and implications Competition between resistant strains and other members of the gut microbiota can restrict the spread of ampicillin resistance. If antibiotic therapy alleviates competition with resident microbes by killing susceptible strains, as here, microbiota-based interventions that restore competition could be a key for slowing the spread of resistance. Lay Summary Slowing the spread of global antibiotic resistance is an urgent task. In this paper, we ask how interactions between microbial species drive the spread of resistance. We show that antibiotic killing of susceptible microbes can free up resources for resistant microbes and allow them to thrive. Therefore, we should consider microbes in light of their social interactions to understand the spread of resistance.

killing of susceptible microbes can free up resources for resistant microbes and allow them to thrive. Therefore, we should consider microbes in light of their social interactions to understand the spread of resistance.
K E Y W O R D S : evolution; ecology; microbiology; antibiotic resistance; competitive release; microbiome BACKGROUND Overuse and misuse of antibiotics has led to the rapid emergence of resistance. The spread of resistance depends on the rate at which resistance mechanisms arise [1] and population growth of resistant bacteria in the presence [2] and absence [3] of antibiotics. However, it is increasingly clear that the spread of resistant bacteria also depends on how they interact with other strains or species. For example, bacteria can interact competitively (e.g. for shared resources [4]) or cooperatively (e.g. cross-feeding [5]) and we would expect resistant genotypes to spread faster when antibiotics eliminate competing rather than cooperating strains. There is accumulating evidence that interactions within species can alter the rate and trajectory of antibiotic resistance evolution [6][7][8][9][10]. However, the effects of antibiotic resistance mechanisms are not solely directed at conspecifics, suggesting that inter-specific interactions are also important for driving resistance [11][12][13][14]. Nevertheless, our understanding of how antibiotic resistance is shaped by intraand inter-species interactions in diverse microbial communities (such as the human gastrointestinal tract) remains limited [15,16]. Given that antibiotic-resistant pathogens exist in complex communities in nature, progress here will facilitate the prediction and management of resistance.
Some types of intra-and inter-specific interactions are likely to slow the spread of resistant genotypes upon exposure to antibiotics. For example, if growth of resistant bacteria reduces the effective antibiotic concentration experienced by susceptible strains or species (cross-protection), the relative fitness advantage of resistance is reduced (compared to in the absence of such cross-protection). This effect has been observed for b-lactam antibiotics degraded by b-lactamase-producing microbes, benefiting not just the producer but nearby susceptible cells. This allows susceptible strains to act as social cheats and gain a frequency-dependent fitness advantage [6,7,9,15,17,18]. However, cross-protection is sensitive to factors such as spatial structure [18], which reduces population mixing and opportunities for exploitation. Moreover, in complex communities, such as the mammalian gastrointestinal tract, the potential for crossprotection will depend on which resistance mechanisms are circulating and whether resident strains or species can detoxify the environment in other ways (e.g. inoculum effects [19]). Therefore, while results from simplified systems indicate crossprotection can occur in some scenarios, it is not yet clear whether cross-protection plays a role in protecting susceptible genotypes in complex communities. One recent study [15] found evidence of kanamycin cross-protection in a pig gut microbiota, albeit only at intermediate antibiotic concentrations Conversely, other types of microbial interactions could promote the spread of resistance. For example, when competitive interactions dominate microbial communities, antibiotic therapy can kill sensitive competitors, freeing up resources for resistant strains and driving the spread of resistance through a community or population (competitive release [20]). Competitive release is a driving force behind Clostridium difficile infection of the gut, where treatment with broad-spectrum antibiotics kills protective microbiota, opening up niche space for the invasion of C.difficile [21]. A similar phenomenon is observed in the cystic fibrosis lung, where loss of microbial community diversity in response to antibiotic treatment in early childhood precedes invasion by the highly antibiotic-resistant pathogen Pseudomonas aeruginosa [22][23][24]. There is also evidence for competitive release driving within-species population dynamics, primarily chemotherapy of acute infections of the rodent malaria Plasmodium chabaudi in laboratory mice. In these studies, an expansion in the numbers of resistant parasites is observed following drug administration [25][26][27][28].
The relative importance of cross-protection and competitive release in the human gastrointestinal tract is not well understood, even though this is a key battleground in the antibiotic resistance crisis [29]. In part, this knowledge gap reflects the difficulty of quantifying the net effect of interactions with other strains or species in communities sampled from human gastrointestinal tracts. Here, we overcome this challenge using an anaerobic human gut microcosm system [16]. We chose Escherichia coli as our focal strain because it is a ubiquitous gut commensal [30][31][32] and key opportunistic pathogen with rising antibiotic resistance [33]. We focus on the beta-lactam antibiotic ampicillin, because it is widely used and resistance is a key problem in E.coli [30], including via mechanisms that also apply to other species and antibiotics [31]. We inoculated each microcosm with susceptible and/or resistant genotypes of a focal E.coli strain, before tracking their population growth with/without ampicillin, and with/without the resident microbiota. By inoculating resistant and sensitive genotypes of our focal strain both in monocultures (where the two genotypes cannot directly compete or engage in cross-protection) and in co-cultures (allowing possibilities for competition and cross-protection), we simultaneously observed opportunities for cross-protection and competitive release, both within and between species. We hypothesized that if cross-protection is in play, the susceptible focal E.coli strain would experience weaker ampicillin inhibition in the presence-versus-absence of (i) the resistant focal E.coli strain, and/or (ii) the resident microbiota (if the microbiota contains microbes that reduce the effective antibiotic concentration in the microcosm). Conversely, competitive release driven by ampicillin inhibition of sensitive bacteria would result in the resistant focal E.coli strain growing better in the presence-versus-absence of (i) the resident microbiota (if the microbiota contains ampicillin-susceptible competitors) and/or (ii) sensitive focal strain. Our results show that while crossprotection increased the relative fitness of antibiotic-susceptible E.coli in standard laboratory conditions [well-mixed lysogeny broth (LB)], competitive release of antibiotic-resistant E.coli drove the spread of resistance in human gut microcosms.

Bacterial strains
We used E.coli K-12 MG1655 with a chromosomal fluorescent dTomato tag and a chloramphenicol resistance cassette as our ampicillin-susceptible strain (K-12 susc ). For our ampicillinresistant strain (K-12 res ), we inserted a non-conjugative bla TEM plasmid conferring ampicillin resistance into K-12 susc [34] (Supplementary methods S1). We confirmed the stability of the plasmid by verifying that the number of K-12 res colonies did not differ between LB plates either supplemented with or without 100 lg/ml ampicillin after 24h growth in LB media (mean counts in LB medium: 84.3 6 6.8; mean CFU counts LB supplemented with ampicillin: 88.7 6 9.1). We note E.coli K-12 possesses an inducible ampC beta-lactamase [35], which may influence its susceptibility to ampicillin; there was nevertheless a clear difference in susceptibility between our susceptible and resistant focal strains here. Resistance to chloramphenicol ensured both strains could be isolated from the microbiota when plated on LB agar supplemented with 25 lg/ml chloramphenicol.

Human microbiome samples
We collected stool samples from three human donors on 15 May 2018. Samples were stored at À80 C until this experiment was conducted in July 2019 (Supplementary methods S1). The full sampling regime is outlined in Reference [16] and approved by the ETH Zürich Ethics Commission (EK 2016-N-55). For our competition experiment in anaerobic gut microcosms, we combined all three samples and tested how the fitness of K-12 res and K-12 susc depended on the presence or absence of this combined gut microbiota. We used frozen stool samples to make faecal slurry, consistent with our aim of including microbial communities sampled from human gastrointestinal tracts (but not necessarily reproducing entire communities or all physiological conditions, which is unrealistic ex vivo). Past work indicates frozen samples are taxonomically similar to fresh samples and have similar effects in downstream experiments in anaerobic fermenters [36]. We chose to pool the microbiota samples from the three donors for three reasons. First, this allowed us to test the effects of multiple factors (microbiota, ampicillin and culture conditions) with multiple replicates in each treatment and extending the generality of our results beyond a single human donor sample, but with a feasible experimental scale. Second, pooled samples have been used successfully in experiments with anaerobic fermenters filled with human gut slurry [37]. Third, previous amplicon sequencing showed that taxonomic composition of these three sampled communities was similar (dominated by Ruminococcaceae, Lachnospiraceae and Bacteroidaceae), suggesting that pooling them results in an aggregate microbiota, rather than an entirely novel kind of community [16]. The final combined slurry was plated on LB agar supplemented with 25 lg/ml chloramphenicol to ensure the culturable component of the resident microbiota was susceptible to this antibiotic. This facilitated the isolation of our chloramphenicol-resistant focal E.coli strains in our competition experiments. Finally, we confirmed that the culturable component of our microbiota was inhibited by ampicillin (total colony counts were approximately halved when plated on LB agar supplemented with 100 lg/ml ampicillin compared to ampicillinfree plates). This indicated the presence of both ampicillinresistant and susceptible microbes within the microbiota, and therefore potential opportunities for microbiota-mediated cross-protection or competitive release, respectively. This was further confirmed using flow cytometry in our anaerobic gut microcosm experiment (below).

Competition assays in LB media
We first established whether ampicillin resistance in our focal strain could confer a protective benefit to susceptible cells in well-mixed, nutrient-rich, standard laboratory media, exposed to sublethal ampicillin concentrations (where we expected crossprotection to be relatively likely [38]). To test this, we grew K-12 res and K-12 susc alone (monoculture) and together (co-culture) in the presence and absence of 7.2 lg/ml ampicillin ($90% of the minimum inhibitory concentration, of the sensitive strain [16]). For monocultures, we added $10 5 cfu/ml each strain to 5 ml LB (Sigma-Aldrich). For co-cultures, we added $5 Â 10 4 cfu/ml each strain, so the total density was similar to monocultures. We made six replicates for each set of conditions (36 cultures in total). We then incubated each culture at 37 C, shaking at 180 rpm for 24 h, before serially diluting and plating on (i) LB agar (Sigma-Aldrich) and (ii) LB agar supplemented with ampicillin (100 lg/ml). This allowed us to enumerate total bacterial density and that of K-12 res only. K-12 susc densities were estimated by subtracting K-12 res from total bacterial densities.
We estimated the total change in population density for each strain in each microcosm as the Malthusian growth parameter (m): ln(final density/start density) [39]. Starting densities were quantified by plating overnight cultures of each strain before inoculation. We then tested whether ampicillin (with vs without) and/or culture conditions (monoculture vs co-culture) affected the growth (m) of each strain (K-12 res and K-12 susc ) using analysis of variance, fitted separately for each strain and including the ampicillinÂculture conditions interaction. We used Box-Cox transformation to improve the normality of the data, after first adding a constant value (1.5) to all m values (accounting for four cases, where m 0; further details are given below). Thus, in both the presence and absence of antibiotics, we tested whether growth of each strain (K-12 susc and K-12 res ) was higher or lower when grown in isolation (monoculture) or together (co-culture), following earlier work on social interactions among closely related strains [40,41]. We used population growth (m), rather than final population density (cfu/ml), as our response variable here because this accounts for variation in starting densities of each strain (e.g. in co-cultures relative to monocultures). Thus, if each strain grows equally well (same number of replications, and similar values of m) in monoculture versus coculture, this suggests cells of each type replicate similarly well when surrounded by clonemates (monoculture) as when surrounded by a mixture of clonemates and cells of the other type (co-culture). We also provide the final population densities (cfu/ml) from these experiments ( Supplementary Fig. S1), which supported similar qualitative conclusions. Data were analysed using R version 3.2.4.
We inoculated each tube with (i) 10 5 cfu/ml K-12 res (ii) 10 5 cfu/ml K-12 susc or (iii) 5 Â 10 4 cfu/ml of each, as above. Half of the tubes (n ¼ 18) were exposed to 7.2 lg/ml ampicillin and the remaining half with an equivalent volume of sterilized water. Finally, half of our tubes (n ¼ 18) were inoculated with the resident microbiota (350 ll of 'fresh' gut slurry plus 500 ll of sterilized slurry) and the remaining half with a microbiota-free control (850 ll sterilized gut slurry), giving a total volume in each tube of 8 ml. Each treatment combination was replicated thrice, giving a fully factorial experimental design (Fig. 1).
Tubes were grown without shaking at 37 C for 24 h, before diluting and plating on (i) LB agar supplemented with 25 lg/ml chloramphenicol (to distinguish the focal strain from resident microbiota) and (ii) LB agar supplemented with 25 lg/ml chloramphenicol and 100 lg/ml ampicillin (to distinguish K-12 res from K-12 susc ). Under this plating regime, the detection limit for the final abundance of K-12 susc is high when K-12 susc is very rare relative to K-12 res . In these cases (where the count from the ampicillin plate was equal to or greater than the count from the ampicillin-free plate), we assigned a final K-12 susc density equal to its starting density in the same microcosm (that is, assuming zero net population growth in these microcosms, n ¼ 3; these cases have m ¼ 0 in Fig. 3). We verified that colonies on our selective plates were derived from the focal strain (and not resident E.coli) by using a fluorescence-capable stereoscope, confirming that all plated colonies had the dTomato marker. Finally, to quantify ampicillin inhibition of the microbial community in each microcosm, we measured total microbial abundance (cells/ml) using flow cytometry (benchtop flow cytometer Novocyte 2000 R, ACEA Biosciences Inc.), described in Reference [16]. This procedure separates cells from background noise in our system, as elsewhere [42] although we acknowledge flow cytometry can have other limitations (e.g. detecting cells in aggregates or clumps).
As above in LB, we estimated the total change in population density for each variant of the focal strain (K-12 susc and K-12 res ) in each microcosm as the Malthusian growth parameter (m). We then tested whether ampicillin (with vs without), culture conditions (monoculture vs co-culture) and resident microbiota (with vs without) affected the growth (m) of each strain (K-12 res and K-12 susc ) using analysis of variance, fitted separately for K-12 sus and K-12 res , and including interaction terms. Thus, interactions between K-12 res and K-12 susc were tested as above in LB medium, but interactions of each of these strains with the resident microbiota were tested in a slightly different way. Specifically, by this experimental design, the total initial focal strain population density (in both mono-and co-cultures) was the same in microcosms incubated with versus without the resident microbiota. This allowed us to test whether addition of the resident microbiota competitively suppressed the focal strain, by comparing focal strain growth with versus without resident microbiota [43,44]. As with our LB experiment, we also provide final densities (cfu/ml) from these experiments, which support the same qualitative conclusions ( Supplementary Figs S2 and S3).

Supernatant addition experiment
Some key mechanisms of competition involve changes to the local abiotic environment (e.g. resource depletion or accumulation of toxins). We therefore tested whether population growth of focal resistant E.coli varied upon exposure to supernatants extracted from cultures including (i) focal E.coli, (ii) the resident microbiota or (iii) no bacteria. Our goal here was to observe overall differences among these three main classes of supernatants, but to be consistent with the community treatments used in our main anaerobic microcosm experiment (above), we included multiple subtypes within these categories (e.g. resident microbiota alone as well as microbiota in which a focal E.coli strain was embedded). Supernatant was  of susceptible (yellow) and resistant (green) genotypes of our focal E.coli strain grown in the presence and absence of ampicillin in monoculture (left panel) or co-culture (right panel) conditions. There was a significant interaction between ampicillin and culture conditions, so that ampicillin inhibition of K-12susc was weaker in co-culture versus monoculture-consistent with cross-protection (see main text for statistics). Black horizontal bars show mean values; n¼6 obtained by inoculating microcosms of basal medium containing 7.2 lg/ml ampicillin with: (i) 10 5 cfu/ml K-12 res , (ii) 10 5 cfu/ ml K-12 susc , (iii) 350 ll faecal slurry, (iv) 350 ll faecal slurry þ 10 5 cfu/ml K-12 res or (v) 350 ll faecal slurry þ 10 5 cfu/ml K-12 susc . We also included two control tubes containing sterile basal medium. One control tube was treated with ampicillin, the second was not. Cultures were grown for 24 h static at 37 C under anaerobic conditions, after which they were transferred to 15 ml falcon tubes and centrifuged for 5 min at 4000 rpm. The supernatant was removed, and filter sterilized (0.22 filter). We then tested for population growth of K-12 res in each supernatant, by inoculating 10 5 cfu/ml of K-12 res into replicate microplate wells containing 198 ll of supernatant (four replicates per treatment). We incubated the microplate at 37 C static for 24 h under aerobic conditions, after which cell densities were quantified using flow cytometry [16].

Ampicillin resistance is a shareable public good in wellmixed LB media
We first examined the costs and benefits of ampicillin resistance by measuring the change in population density (Malthusian growth parameter, m) for our focal E.coli susceptible and resistant strains (K-12 susc and K-12 res ), growing under mono-and co-culture conditions, in the presence and absence of ampicillin, in well-mixed LB media.
We found that when both strains grew in separate microcosms, K-12 susc grew better than K-12 res in the absence of antibiotics (monocultures; Fig. 2; bootstrapped t-test, T 10 ¼ 7.14, P < 0.001), indicating a resistance cost in this context (equivalent to a reduction in final population density of $40%; Supplementary Fig. S1). Exposure to ampicillin reversed this effect: K-12 res grew better than K-12 susc, indicating a benefit to resistance in the presence of ampicillin (monocultures; Fig. 2; bootstrapped t-test, t ¼ 5.65, df ¼ 5.3, P < 0.03). By contrast, when both strains were in the same microcosm (co-culture), addition of ampicillin had a much weaker inhibitory effect on K-12 susc compared to the inhibition observed in monoculture [linear model with Box-Cox transformation (k ¼ 2); culture condi-tionÂampicillin interaction, F 1,20 ¼ 26.58, P < 0.0001] (Fig. 2). In other words, K-12 susc gained a protective benefit from the presence of the resistant strain, consistent with this strain detoxifying the local environment and conferring crossprotection.
No evidence for cooperative ampicillin resistance in anaerobic gut microcosms In anaerobic gut microcosms (see Methodology), K-12 susc growth (m) was unaffected by the presence of the resistant strain or the presence of the resident microbiota [linear model with Box-Cox transformation (k ¼ 2); effect of microbiota, F 1,20 ¼ 0.55, P ¼ 0.5; effect of culture condition, F 1,20 ¼ 0.45, P ¼ 0.51] (Fig. 3) over the timescale of our experiment. The strongest effect, we observed on K-12 susc growth was for addition of ampicillin, which significantly reduced growth of K-12 susc in both mono-and co-cultures and in the presence and absence of resident microbiota [linear model with Box-Cox transformation (k ¼ 2), effect of ampicillin; F 1,20 ¼ 40.98, P < 0.0001] (Fig. 3). Importantly, we found no evidence that the extent of ampicillin inhibition depended on culture conditions or microbiota treatment [linear model with Box-Cox transformation (k ¼ 2); culture condi-tionÂampicillin interaction, F 1,17 ¼ 0.0004, P ¼ 0.98; micro-biotaÂampicillin interaction, F 1,17 ¼ 0.04, P ¼ 0.85] (Fig. 3). Hence, extracellular detoxification of ampicillin by K-12 res or by resident microbiota could not rescue the poor growth of K-12 susc in anaerobic gut microcosms.
Evidence for competitive release in anaerobic gut microcosms Unlike for the susceptible K-12 susc strain, population growth (m) of the resistant K-12 res strain was increased by addition of ampicillin, but only in the presence of the resident microbiota [linear model with Box-Cox transformation (k ¼ 4); microbiotaÂampicillin interaction: F 1,17 ¼ 24.42, P ¼ 0.0001; post hoc Tukey HSD: effect of ampicillin in treatment groups with microbiota: P < 0.05/without microbiota: P > 0.05] (Fig. 4). Growth of K-12 res was also higher in co-culture with K-12 susc compared to monoculture, and this difference was amplified in the presence of ampicillin [linear model with Box-Cox transformation (k ¼ 4); culture conditionÂampicillin . Effects of an antibiotic, an isogenic sensitive strain and a natural microbial community on growth of antibiotic-resistant E.coli. Population growth (m) for K-12res is shown after growth in the presence and absence of ampicillin (x-axis), in the presence and absence of the sensitive K-12sus strain (upper vs lower row of panels) and in the presence and absence of the natural gut microbial community (right vs left columns of panels). Growth (m) of K-12res was increased by addition of ampicillin, but only in the presence of the resident microbial community or K-12susc (see main text for statistics). Black horizontal bars show mean values, n¼3. Asterisks denote P<0.05 based on Tukey HSD post hoc tests interaction: F 1,17 ¼ 8.99, P < 0.01]. Together, this is consistent with ampicillin-mediated killing of susceptible microbes (either our K-12 susc strain or susceptible members of the microbiota), that releases K-12 res from competition with susceptibles and enhances the spread of ampicillin resistance.

Competition between resistant bacteria and resident microbiota
If the growth increase observed for resistant bacteria (upon addition of ampicillin in the presence of resident microbiota) were due to competitive release, we would expect K-12 res to reach lower densities in the presence versus absence of the resident microbiota without ampicillin (consistent with competitive suppression, from which they can be released by the addition of antibiotics). This is supported by K-12 res reaching lower final cell densities in cultures with versus without resident microbiota in the absence of ampicillin (and without the sensitive strain) (linear model, effect of microbiota on final population density: F 1,4 ¼ 10.27, P < 0.05). When we take population growth as the Malthusian parameter (instead of final cell densities), we see a similar pattern (lower in every replicate with resident microbiota compared to corresponding replicates without resident microbiota in the absence of ampicillin and without the sensitive strain), although here the effect is not significant on average (Tukey test P > 0.05) (Fig. 4). This lack of statistical significance may reflect limited power of this test, dictated by our sample size and detection limit: there was one replicate in the þampicillin/þmicrobiota/monoculture treatment where the resistant focal strain was below the detection limit imposed by the plating scheme we used, increasing the within-group variance here. Thus, these data are consistent with competition between K-12 res and resident microbes, although they do not demonstrate it conclusively.
As a second test for competition between resident microbiota and the resistant strain, we analysed population growth (m) of K-12 res in supernatants extracted from cultures with versus without the microbiota (Fig. 5). Different types of supernatants varied in their ability to support K-12 res growth (F 6,21 ¼ 476.99, P < 0.0001), with supernatants from microbiota treatments consistently supporting less growth than supernatants from cultures containing only a focal E.coli strain (Fig. 5). This is consistent with the resident microbiota changing the local abiotic conditions in a way that reduces population growth of K-12 res (e.g. via nutrient depletion). Supernatant originating from cultures of focal E.coli also supported less growth than supernatant from sterile control tubes (Dunnett's test; control vs K-12 res supernatant, P < 0.0001, control vs K-12 susc supernatant, P < 0.0001).

Susceptibility of resident microbiota to antibiotic inhibition
A further requirement for competitive release is that addition of ampicillin inhibits the resident microbiota (thereby relieving the negative effect of the microbiota on the resistant strain). Consistent with this, flow cytometric measurements indicated that addition of ampicillin to the resident microbiota had a negative effect on total abundance [linear model with Box-Cox transformation (k ¼ 2), F 1,14 ¼ 43.29, P < 0.0001, Supplementary Fig. S4]. In particular, in microcosms containing K-12 res plus resident microbiota (where we previously observed no inhibition of K-12 res ), the decrease in total abundance here must reflect inhibition of the resident microbiota ( Supplementary Fig. S4). Thus, our data support competition between the resistant strain and the resident microbiota, which was alleviated by ampicillin because this inhibited the resident microbiota but not K-12 res.

CONCLUSIONS AND IMPLICATIONS
Microbes live in complex communities, where species interact both antagonistically (e.g. resource competition or direct killing) as well as cooperatively (e.g. cross-protection or crossfeeding). Understanding how microbes within complex communities, such as the gut microbiota respond to antibiotic treatment is therefore challenging, yet a task that is crucial to predict the spread of antibiotic-resistant pathogens. Here, we provide evidence that in microbial communities sampled from healthy humans, exposure to ampicillin can release a focal resistant E.coli strain (K-12 res ) from competition with susceptible bacteria. Growth assays in supernatant indicate this arises from competition between K-12 res and the resident microbiota, which is alleviated when ampicillin-susceptible members of the microbiota are killed. Surprisingly, we found no evidence of crossprotection (where susceptible cells benefit from antibioticdegrading activity of resistant cells) in our gut microcosms, in contrast to results from well-mixed LB medium (this study) and previous work in anaerobic fermenters [15] and in vivo [8]. Hence, although cross-protective effects can occur between resistant and susceptible genotypes, they are highly context dependent.
In our experimental setup, competitive release of K-12 res was observed in the presence of the resident microbiota upon antibiotic exposure. This is important because it goes beyond past work with other types of pathogens and in non-human microbiota [15,28] to demonstrate directly that competitive release can contribute to the spread of antibiotic-resistant bacteria in human-associated microbiota. By contrast, while we saw some evidence of competitive release of K-12 res in the presence of K-12 susc only (i.e. in co-culture, without the microbiota), this effect was much weaker than in the presence of the microbiota. We pose two mutually non-exclusive explanations. First, competitive inhibition of our resistant focal strain by the sensitive focal strain was weaker than inhibition by the resident microbial community. This is supported by our supernatant experiment, where supernatant originating from microcosms containing the resident microbial community supported less growth than supernatant originating from a focal E.coli strain only. Possible drivers of this relatively strong suppression by the resident microbiota include resource competition (effective scavenging by a diverse community, or closely related strains that are strong competitors against K-12, as we have seen previously [16]) and/or direct antagonistic interactions, such as toxin production [45]. Second, our experimental design ensured total inoculant densities of our focal E.coli strain were consistent between mono-and co-culture conditions. In contrast, the effect of the microbiota was tested by the addition of microbiota to a fixed number of E.coli cells. Hence, microbiota-mediated competition for nutrients was directly imposed, whereas competition for nutrients (at least initially) did not differ between co-cultures and monocultures. Note, supernatant from microbiota-free focal strain cultures supported less growth than control supernatant, indicating intra-species competition, albeit with a smaller effect than in microbiota treatments. One caveat is that in order to use flow cytometry to enumerate K-12 res cells, supernatant effects were not tested under the same anaerobic conditions established in our gut microcosms (although the supernatant itself was anaerobically produced).
Differences between supernatant subtypes in supporting K-12 res growth are harder explain-such as our finding that supernatant from communities in which a susceptible or resistant E.coli strain is already embedded differ in their ability to facilitate K-12 res growth. One possibility is that killing of K-12 susc provides recyclable nutrients that can promote K-12 res growth [50]. However, it was beyond the scope of this study to infer what was driving these differences and comparisons between subtypes are supported by low statistical power.
A key insight from our results is that the impact of interactions (both within and between species) on the spread of antibiotic resistance can differ greatly between simple, two-strain experiments and more complex communities. Our experiment in well-mixed LB medium showed that under a specific set of conditions that favour cross-protection [38,40,41] ampicillin resistance can be a cooperative public good, increasing the fitness of non-detoxifying 'cheaters' that benefit from crossprotection. However, in our gut microcosms, any opportunity for cooperation was trounced by a net competitive interaction between strains. One explanation for these differing dynamics is the relatively high levels of spatial structure in the latter. Spatial structure keeps producers and products closer together, limiting opportunities for exploitation and keeping relatedness between interacting partners high [46]. Note, that localized cross-protection can nevertheless emerge in spatially structured populations, such as those on agar surfaces [47]. Ultimately, the relative importance of cross-protection and competitive release in natural systems will be governed by the balance between resource competition between and within species as well as environmental constraints on cooperation.
Our experiment has some important limitations. Although ampicillin and E.coli are undoubtedly of high real-world relevance, generalizing our results to other species/antibiotics should be done with caution. A recent study by Letten et al. [48] with a similar setup, but different antibiotics, samples and focal strain, found little evidence that antibiotics released an invading strain from competition with resident microbiota. This difference may be explained by different antibiotics having different effects on resident microbiota, variation among focal resistant strains, or other differences between experiments (e.g. baseline competition with resident microbiota was relatively strong in Letten et al. [48]). Conversely, a past experiment also using this study system showed an ampicillin-resistance plasmid conferred a relatively large fitness benefit to its host when embedded in a gut microbiota and exposed to ampicillin [16], consistent with our results here. Together, these findings are in line with the above evidence that competitive release and crossprotection can be important, but identifying the conditions that give rise to them is a key challenge for translation to real-world