Fitness advantage of Bacteroides thetaiotaomicron capsular polysaccharide in the mouse gut depends on the resident microbiota

  1. Daniel Hoces
  2. Giorgia Greter
  3. Markus Arnoldini
  4. Melanie L Stäubli
  5. Claudia Moresi
  6. Anna Sintsova
  7. Sara Berent
  8. Isabel Kolinko
  9. Florence Bansept
  10. Aurore Woller
  11. Janine Häfliger
  12. Eric Martens
  13. Wolf-Dietrich Hardt
  14. Shinichi Sunagawa
  15. Claude Loverdo
  16. Emma Slack  Is a corresponding author
  1. Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Switzerland
  2. Institute of Microbiology, Department of Biology, ETH Zurich, Switzerland
  3. Sorbonne Université, CNRS, Institut de Biologie Paris-Seine (IBPS), Laboratoire Jean Perrin (LJP), France
  4. Department of Microbiology and Immunology, University of Michigan Medical School, United States

Abstract

Many microbiota-based therapeutics rely on our ability to introduce a microbe of choice into an already-colonized intestine. In this study, we used genetically barcoded Bacteroides thetaiotaomicron (B. theta) strains to quantify population bottlenecks experienced by a B. theta population during colonization of the mouse gut. As expected, this reveals an inverse relationship between microbiota complexity and the probability that an individual wildtype B. theta clone will colonize the gut. The polysaccharide capsule of B. theta is important for resistance against attacks from other bacteria, phage, and the host immune system, and correspondingly acapsular B. theta loses in competitive colonization against the wildtype strain. Surprisingly, the acapsular strain did not show a colonization defect in mice with a low-complexity microbiota, as we found that acapsular strains have an indistinguishable colonization probability to the wildtype strain on single-strain colonization. This discrepancy could be resolved by tracking in vivo growth dynamics of both strains: acapsular B.theta shows a longer lag phase in the gut lumen as well as a slightly slower net growth rate. Therefore, as long as there is no niche competitor for the acapsular strain, this has only a small influence on colonization probability. However, the presence of a strong niche competitor (i.e., wildtype B. theta, SPF microbiota) rapidly excludes the acapsular strain during competitive colonization. Correspondingly, the acapsular strain shows a similarly low colonization probability in the context of a co-colonization with the wildtype strain or a complete microbiota. In summary, neutral tagging and detailed analysis of bacterial growth kinetics can therefore quantify the mechanisms of colonization resistance in differently-colonized animals.

Editor's evaluation

The authors have developed an innovative approach to analyze microbial population dynamics within a host and used this technique to address an important question – how the composition of the microbiota influences the intestinal colonization of an encapsulated vs unencapsulated version of an important commensal organism. The revisions add evidence to support their claims and make the approach more accessible.

https://doi.org/10.7554/eLife.81212.sa0

Introduction

From the moment that we first contact microbes at birth, we continuously encounter environmental and food-borne microbes. Whether such encounters are transient or will lead to long-term colonization is influenced by complex ecological interactions between the invading species and the existing consortium, as well as the host’s dietary habits and the physiology of the intestine (David et al., 2014; Wotzka et al., 2019). A better understanding of the factors determining colonization efficiency is crucial in the development of microbiota engineering strategies (Donia, 2015; Pham et al., 2017; Sheth et al., 2016) and in the use of bacterial species as biosensors to probe microbiota function and stability (Goodman et al., 2009).

One established way of studying ecological processes within hosts is genetic barcode tagging of otherwise isogenic microbes. This has previously been used to study population dynamics of pathogens such as Vibrio cholerae or Salmonella Typhimurium within the infected host (Abel et al., 2015; Vlazaki et al., 2019). Based on barcode recovery and mathematical modeling, it has been possible to infer parameters such as growth, clearance, and migration rates (Dybowski et al., 2015; Grant et al., 2008; Kaiser et al., 2014; Kaiser et al., 2013), as well as the size of population bottlenecks imposed during colonization (Abel et al., 2015; Li et al., 2013), antibiotic treatment (Vlazaki et al., 2020), or immunity (Coward et al., 2014; Hausmann et al., 2020; Lim et al., 2014; Maier et al., 2014; Moor et al., 2017). Furthermore, combining neutral genetic barcodes with targeted mutant strains, this experimental tool can be used to mechanistically analyze the fitness effect of individual genes that regulate successful gut colonization or tissue invasion (Di Martino et al., 2019; Nguyen et al., 2020).

To study the dynamics of invasion of a novel microbiota member, we chose to use Bacteroides thetaiotaomicron (B. theta) as a model microbe. B. theta is a common commensal member of the human intestinal microbiota, and the availability of precise tools for genetically engineering it (Lim et al., 2017; Mimee et al., 2015; Whitaker et al., 2017) makes it a strong candidate for introducing novel functions into microbiomes. A common feature of Bacteroides species is the ability to use phase variation to modulate the expression of 3–10 capsular polysaccharide (CPS) operons, leading to the production of distinct capsule structures (Porter and Martens, 2017). B. theta strains lacking a capsule have been shown to engraft poorly in an existing microbiota when competing with CPS-expressing strains (Martens et al., 2009; Porter et al., 2017). However, the deletion of all capsule gene clusters is not expected to negatively influence the growth rate of B. theta per se (Rogers et al., 2013). Rather it can affect its survival on exposure to noxious stimuli, like bile acids, stomach acid, antimicrobial peptides, or phage (Porter et al., 2017; Porter et al., 2020). These characteristics of different B. theta strains make them a good model to test our ability to quantify population dynamics in vivo.

To quantify the processes determining success of B. theta colonization in the presence of different resident microbiota, we generated genetically barcoded B. theta strains able to produce capsular polysaccharides (wild type [WT]) or not (acapsular, with deletion of all eight capsular polysaccharide synthesis loci; Porter et al., 2017). The genetic barcodes were linked to an erythromycin resistance cassette to allow amplification of our barcodes by cultivation. Frequencies of barcoded strains in a sample can be easily determined by plating for CFU determination, recovering all colonies growing in the presence of the relevant antibiotic and quantifying relative barcode frequencies by quantitative PCR (qPCR). Combining these values gives CFU of each barcoded strain (Grant et al., 2008; Lim et al., 2014), which can be combined with simple mathematical models to estimate the probability of individual clones to colonize under different conditions. This revealed similar colonization success (i.e., encounter of similar population bottlenecks) between acapsular and WT B. theta, during colonization of mice carrying low-complexity microbiota (OligoMM12 and LCM microbiota). However, the probability of colonization dropped approximately 10-fold for WT strains and 100-fold for acapsular strains colonizing the gut of mice with a complex microbiota (specific pathogen free [SPF]). Despite similar fitness on single-colonization of low-complexity microbiota mice, the acapsular strain competed poorly against the WT strain in competitive co-colonization in the same setting. This apparent discrepancy between colonization probability and competitive fitness could be explained by a longer lag phase and very slightly reduced net growth rate of the acapsular B. theta. This gives WT B. theta a head-start to occupy most of the available niche, excluding acapsular strains. The same barcoding system can also be used to quantify the bottlenecks experienced by a steady-state B. theta population in the gut subjected to an acute inflammatory reaction. Therefore, neutral tagging and simple mathematical modeling can infer quantitative insights into the behavior of B. theta during gut colonization.

Results

Genetically barcoded B. theta strains to study within-host population dynamics

B. theta VPI-5482 can phase vary the expression of eight different capsular polysaccharides (WT). B. theta Δcps strain that cannot produce capsule was generated previously by sequentially deleting all CPS gene clusters (acapsular) (Porter et al., 2017). We first established neutrally tagged clones of these strains by inserting a genetic barcode linked to an antibiotic resistance cassette and a fluorescent protein gene into the genome using the previously described pNBU2 integration plasmid. Six barcode sequences, previously developed and validated for Salmonella (wild-type isogenic tagged strains [WITS]; Grant et al., 2008; Maier et al., 2014) were inserted adjacent to an erythromycin-resistance cassette, ermG (Cheng et al., 2000), and a GFP or mCherry fluorescent protein gene under a strong constitutive promoter. Untagged strains were generated by inserting pNBU2 carrying the tetQb tetracycline resistance cassette (Nikolich et al., 1992) and a GFP gene under the control of a strong constitutive promoter into the same integration site (Figure 1A, Key Resources Table). Fluorescent proteins (GFP or mCherry), expressed from a phage promoter with an optimized ribosome binding site, were included to permit later visualization of clones (Wang et al., 2000; Whitaker et al., 2017).

Figure 1 with 2 supplements see all
Barcoded B. theta strains have similar fitness for growing in vitro and in vivo.

(A) Schematic representation of insertions in B. theta genome. The barcoded strains carried the barcode, erythromycin resistance cassette, and fluorescent protein in the genome. The untagged B. theta carried a tetracycline resistance cassette together with a fluorescent protein inserted at the same position in the genome. (B) Workflow for barcoded strain enrichment and quantification. (C) Growth curves of B. theta WT and acapsular barcoded strains in BHIS media (n = 3, replicative cultures per strain) and growth rates (1/hr, mean and 95% confidence interval) per barcode. (D) Experimental design of in vivo competitions to confirm equal fitness of the barcoded strains. Each strain (B. theta untagged and six B. theta barcoded) were mixed in an equal ratio (inoculum: 106 CFU per strain). (E) Barcode distribution during colonization among six B. theta barcoded strains either WT or acapsular. Plots show distribution of barcodes in the inoculum, in cecal content of individual mice after 48 hr of colonization and fold increase of each barcode per mouse compared to the inoculum (n = 5 mice colonized). (F) Competitive index of tetracycline-resistant untagged strain (Tet) over the erythromycin-resistant barcoded strain (Ery) in vivo after 48 hr of colonization of B. theta WT or acapsular in cecal content and in feces (n = 2 replicative cultures in inoculum; n = 5 mice colonized). Points represent individual mice, and the horizontal line is the mean. p-Values were obtained by one-way ANOVA followed by Tukey’s honest significance test. Data are included in Figure 1—source data 1.

Figure 1—source data 1

Barcoded B. theta strains have similar fitness for growing in vitro and in vivo.

https://cdn.elifesciences.org/articles/81212/elife-81212-fig1-data1-v2.xlsx

We validated a system to enrich barcoded strains from overall very low frequencies (Figure 1B) Samples were plated on BHIS agar with gentamycin to determine the total B. theta CFU and on BHIS agar containing the appropriate antibiotic (erythromycin or tetracycline) to determine the total barcoded B. theta CFU. Subsequently, all B. theta was washed from the plates, and genomic DNA extracted. The relative frequency of each barcode among the recovered colonies was obtained by qPCR using primer sets specific for each barcode. CFU of each individual barcoded strain can then be determined by simple multiplication. Serial dilution and recovery of barcoded strains in in vitro systems shows excellent resolution over five orders of magnitude (Figure 1—figure supplement 1).

Genetic barcode tags do not strongly affect fitness of B. theta strains

A critical assumption of any analysis using genetically barcoded strains is that the chromosomal insertions, as well as the construction process, have not altered the fitness of the strains compared to the WT strain. Anaerobic growth in BHIS media was near-identical in all the barcoded B. theta strains (median doubling time ranged from 78 to 90 min) (Figure 1C). Correspondingly, the barcoded strains maintained their relative abundances, as evaluated by qPCR, when all barcode strains were mixed and grown overnight (Figure 1—figure supplement 2A and B). Whole-genome sequencing of the barcoded strains revealed a few synonymous and non-synonymous mutations, as would be expected for the construction process of individual strains (Supplementary file 1). However, none of the identified mutations is expected to have a major fitness effect, consistent with our observed data.

To test whether the tags confer a fitness effect upon colonization of a host, we colonized germ-free (GF) mice with a uniform mixture of 106 CFUs of the untagged strain and each barcoded strain (Figure 1D). At this barcode abundance, stochastic loss of tags is highly unlikely. We compared the relative abundance of each barcode in the inoculum to that in the cecal content and feces after 48 hr of colonization. These experiments revealed small, random deviations in the distribution, consistent with uniform fitness (Figure 1E). Finally, as erythromycin- and tetracycline resistance were used to distinguish the barcoded and untagged B. theta strains, we also tested whether the antibiotic resistant cassettes alter competitive fitness. In culture, we found a minor competitive advantage of the tetracycline expressing strains over erythromycin only in the acapsular strain (Figure 1—figure supplement 2C). However, there was no significant competitive advantage of any antibiotic cassette after 2 days of colonization in GF mice (Figure 1F). As all barcoded strains carried the erythromycin resistance cassette, small fitness effects associated with the cassette will not affect competition between the barcoded strains. In competitive colonization, tetracycline and erythromycin resistances were reversed in two sets of experiments with similar results, and a simple model based on this data suggests a maximum error due to the antibiotic fitness effect of twofold. This is small compared to the relative competitive fitness of the acapsular and wildtype B. theta strains. Therefore, while absolute equivalent fitness is near-impossible to achieve in these systems, the error due to unintended fitness effects is within an acceptable range.

Determining inoculum size of barcoded strains that yields maximal information upon B. theta colonization

We then applied the neutrally barcoded B. theta strains to estimate colonization probabilities. As invasion probabilities depend on the interaction with the resident microbiota (Kurkjian et al., 2021), we probed B. theta colonization in mice carrying three different communities: low-complexity microbiota (LCM) Stecher et al., 2010; Oligo Mouse Microbiota (OligoMM12) (Brugiroux et al., 2016), and SPF microbiota. These include two low-complexity microbiota models with a reduced set of strains (LCM: 8 strains and OligoMM12: 12 strains) and the closest model for complete microbiota in laboratory mice (SPF: 12 families that include several species) (Figure 2—figure supplement 1). We evaluate the distribution of barcoded B. theta cells in cecum content at 48 hr after initial colonization, a time point shortly after the B. theta population in the cecum reaches carrying capacity (Figure 2—figure supplement 2).

Assuming that the change in relative abundance of tags before and after the colonization process is due to stochastic loss of B. theta, we formulated a simple ‘initial’ model that allows us to infer a per-cell colonization probability for B. theta. The model assumes that B. theta cells undergo random killing during their transition through the stomach and small intestine, that is, the population experiences an initial bottleneck event. Surviving cells arriving in the cecum start growing and the clonal progeny of these cells can be quantified at 48 hr post-colonization via a combination of plating and qPCR (Figure 2A). The number of cells of an individual barcoded clone in the inoculum, n0, is low in our experiments. Correspondingly, the distribution of barcoded bacteria introduced into the stomach of each animal is better approximated by a Poisson distribution of mean n0 than by a uniform distribution. The probability of losing a barcoded clone can be considered equivalent to the fraction of barcoded clones lost across all animals. Considering this early loss of clones as a binomial sampling process, we can express this probability as e-βn0 , where β is the colonization probability of an individual clone from the inoculum. β can therefore be simply computed for animals all receiving an identical inoculum. To increase the power of our observations, we have also pooled data across multiple experiments with small deviations in n0 by maximizing the likelihood of the experimental observations (Figure 2—figure supplements 3 and 4). A more complex calculation can be carried out using the variance of the barcoded population rather than defined loss/retention, a method that can take more information into account, although it is more complicated to execute (see ‘Mathematical modeling’).

Figure 2 with 8 supplements see all
Colonization probability of B. theta strains in low-complexity microbiota (LCM), Oligo Mouse Microbiota (Oligo-MM12), and specific pathogen free (SPF) mice.

(A) Schematic representation of experimental estimation of colonization probability. The untagged strain was tetracycline-resistant, and all barcoded strains were erythromycin-resistant. (B) Pielou’s evenness vs. n0. Pielou’s evenness was estimated with a maximum possible value of Hmax = ln(6) for all data points (six barcoded strains). Each dot represents the evenness calculated per mouse. Values of n0 for which all barcodes were lost, and therefore no evenness could be estimated, were marked with ‘Extinct.’ Total inoculum size was maintained at 107 CFU. The exact inocula compositions are shown in Figures 3 and 4. (C, E) Total B. theta population in the cecum at 48 hr post-colonization for (C) WT and (E) acapsular strains. Points represent individual mice and boxplot quartiles provide summary statistics. p-Values were obtained by one-way ANOVA followed by Tukey’s honest significance test. (D, F) Probability of colonization (β) in (D) WT and (F) acapsular in the cecum after 48 hr of colonization using the loss method. Circles represent the best estimate and vertical line the higher and lower bound of the 95% confidence interval. Estimation based on six barcodes times the number of mice (LCM = 17, OligoMM12 = 10, SPF = 11). See ‘Methods’ for parameter estimations. Data are included in Figure 2—source data 1.

Figure 2—source data 1

Colonization probability of B. theta strains in low-complexity microbiota (LCM), Oligo Mouse Microbiota (Oligo-MM12), and specific pathogen free (SPF) mice.

https://cdn.elifesciences.org/articles/81212/elife-81212-fig2-data1-v2.xlsx

It is important to note that if all tags are lost, or if all tags are recovered, only upper or lower bounds for β can be estimated. Correspondingly, experiments where some, but not all, tags are lost from the final population yield maximum information. To find the optimal n0 that leads to partial barcode loss in vivo, we titrated the barcoded strains into an untagged B. theta population to give n0 values ranging from 18 to 26,666 CFU for each barcoded strain (Figure 2—figure supplements 3 and 4). This was carried out in the context of three different microbiota communities, and for both WT and acapsular B. theta. Recovery of total B. theta and CFUg–1 of barcoded strains from the cecum was determined at 48 hr post-colonization. We used the Pielou evenness (Pielou, 1967) as a summary representation of the distributions of barcoded population (Figure 2B, Figure 2—figure supplements 3 and 4). The resulting β estimates are most robust at an n0 which results in approximately half of the tags being lost (see Appendix 1 ‘Supplementary methods’). In the case of LCM and OligoMM12 low-complexity microbiota mice, an n0 of between 10 and 100 was optimal for both wild-type and acapsular B. theta, that is, 10–100 CFU of each barcoded strain was spiked into an inoculum of 107 untagged B. theta. In the SPF mice, carrying a complex microbiota, an n0 of around 500 was informative for WT, but 5000 CFU were needed of each acapsular barcoded strain.

Finally, we challenged the assumption that loss of barcoded clones was due to stochastic loss. Selective sweeps of a clone, or clones, that have acquired a beneficial mutation would also explain barcode loss. We therefore reisolated abundant barcoded B. theta strains from the 48 hr time point from previous experiments. These were used to assemble an inoculum in which some barcodes were represented by re-isolated strains and others by original ancestral strains. These were mixed at a high n0 (approximately 2 × 106 CFU of each clone per inoculum) and used to colonize SPF mice. There was no consistent advantage of re-isolated strains over ancestral strains in colonizing the cecum at 48 hr post-colonization (Figure 2—figure supplement 5A), consistent with the absence of strongly beneficial mutations in surviving clones. To further confirm the assumption of equal stochastic loss of each barcode, we re-calculated β for all experiments excluding each individual barcode in turn. Excluding data from any one barcode has no statistically significant effect on the estimates of β, as calculated using barcode loss (Figure 2—figure supplement 5B) or variance (Figure 2—figure supplement 5C).

B. theta colonization probability in LCM, OligoMM12, and SPF mice

The resident microbiota composition in the mammalian gut is one of the main factors constraining the colonization of newly arriving species. This can happen through various mechanisms such as competition for nutrients (Bäumler and Sperandio, 2016; Maier et al., 2013), modification of the intestinal environment (Cremer et al., 2017), via direct suppression of the invaders by phages (Almeida et al., 2019; Barr et al., 2013) or type VI secretion systems (Chatzidaki-Livanis et al., 2016; Wexler et al., 2016). Additionally, the microbiota stimulates host mucosal immunity and influences intestinal physiology (Zheng et al., 2020). B. theta loads in the cecal content at 48 hr post-inoculation were similar in LCM and OligoMM12 mice, but significantly lower in SPF mice (Figure 2C). The colonization probability, β, of barcoded B. theta WT strains, calculated using loss or variance methods, was also lower in SPF mice (Log10(β, colonization probability) ± 2 standard deviations = –2.35 ± 0.14) compared to the two LCM (–1.50 ± 0.10; and Oligo, –1.54 ± 0.13) (Figure 2D, Figure 2—figure supplement 6A). Of note, while the relative size of the final population is 100-fold lower in SPF mice, and the relative colonization probability is only 10-fold lower than in animals with a low-complexity microbiota. This indicates that size of the open niche does not linearly translate into colonization probability, that is, the neutral tagging approach reveals information that cannot be simply gleaned from standard CFU determination.

As CPS are thought to play a crucial role in phage evasion/infection (Porter et al., 2020), immune evasion (Fanning et al., 2012; Hsieh et al., 2020; Porter et al., 2017), and protection from other environmental stressors, we expected to see a decreased colonization probability for acapsular B. theta strains in all microbiota backgrounds. Surprisingly, in LCM mice, the total population size of acapsular B. theta (Figure 2E) and the probability to colonize (Log10β: LCM, –1.43 ± 0.13; and OligoMM12, –1.49 ± 0.14) were not significantly different to the WT strain (Figure 2F, Figure 2—figure supplement 6B). There was therefore no measurable fitness benefit of CPS in gut colonization up to 48 hr post-inoculation in these settings. However, we observed a different scenario when we inoculated acapsular B. theta into mice carrying a SPF microbiota. Both the total population size of acapsular B. theta (Figure 2E) and the colonization probability (Log10β: SPF, –3.65 ± 0.13; Figure 2F) were tenfold lower compared to the WT strain, indicating a strong fitness benefit of CPS in the context of a more diverse microbiota. We could not definitively tie this increased clearance to any particular host or microbial mechanism: SPF mice do not have measurable IgA titers specific for acapsular B. theta (Figure 2—figure supplement 7A), nor could we identify lytic spots produced by phage specific for acapsular B. theta from the cecum content of SPF mice (Figure 2—figure supplement 7B). As microbiota-driven restriction of acapsular B. theta colonization is expected to establish very rapidly on recolonization of a GF mouse, but host-driven mechanisms such as upregulation of antibody responses may take several days to weeks, we compared the acapsular B. theta colonization probability in ex-GF mice that had been recolonized by rehousing with SPF mice for 2 days or for 2 weeks. Although short-term recolonization resulted in a larger B. theta population in the cecum than long-term re-colonization with an SPF microbiota (Figure 2—figure supplement 8A), the colonization probability was near-identical between the two groups (Figure 2—figure supplement 8B). Therefore, mechanisms restricting colonization of acapsular B. theta in SPF mice are either direct microbial competition (e.g., via type VI secretion) and/or are very rapidly induced host mechanisms, or a combination of both.

Competitive colonization with acapsular and WT B. theta reveals a role of CPS in OligoMM12-colonized mice

The absence of a decreased colonization probability for acapsular B. theta in LCM mice apparently conflicts with previous studies showing a competitive fitness defect of this strain (Coyne et al., 2008; Porter et al., 2017). We therefore carried out competitive colonizations with B. theta WT and acapsular strains in all microbiota backgrounds. Starting at a 1:1 ratio, we inoculated mice with erythromycin-resistant WT and tetracycline-resistant acapsular B. theta and quantified the cecal bacterial load 48 hr post-inoculation (Figure 3—figure supplement 1A). This reveals a gradient of competition with the WT winning over the acapsular strain, obtaining a competitive index (abundance of WT over acapsular B. theta at the end of the experiment) of approximately 20 in GF mice, 100 in LCM mice, and 104 in SPF mice (Figure 3A). Therefore, the competitive fitness benefit of CPS increased in proportion to microbiota complexity, despite the fact that no difference on colonization probability of the WT and acapsular B. theta could be detected on single colonization.

Figure 3 with 5 supplements see all
Competitive colonization with acapsular and WT strains.

(A) Competitive index (ratio between WT over acapsular) in the cecum after 48 hr of colonization starting at a 1:1 ratio (inoculum: approximately 106 CFU of each: erythromycin-resistant WT and tetracycline-resistant acapsular; germ-free [GF] = 3, low-complexity microbiota [LCM] = 4, Oligo Mouse Microbiota [OligoMM12] = 4, specific pathogen free [SPF] = 4). Points represent individual mice, and the horizontal line is the mean. p-Values were obtained by one-way ANOVA followed by Tukey’s honest significance test. (B) (Left) Probability of colonization by the acapsular strain during co-colonization with the WT strain. Circles represent the best estimate and vertical line the higher and lower bound of the 95% confidence interval. Barcoded erythromycin-resistant acapsular strains were spiked into a WT untagged strain inoculum (inoculum: 107 of untagged tetracycline-resistant B. theta WT and n0 CFU of erythromycin-resistant barcoded B. theta acapsular adjusted to each microbiota background: n0GF ~ 10, n0LCM ~ 103, n0Oligo ~ 103, n0SPF ~ 106; number of mice per group: GF = 7, LCM = 13, OligoMM12=12, SPF = 16). (Right) Probability of colonization by the acapsular strain when colonizing alone. Graph generated using the same data as Figure 2F. (C, D) Estimation of (C) net growth rate and (D) and delay to start exponential growth (see Figure 3—figure supplement 2 and Appendix 1 for fitting function; n = 7 for WT and n = 9 for acapsular). Points represent estimations of individual mice, and the horizontal line is the mean. p-Values were obtained by Welch t-test. (E) Schematic representation of competitive advantage of the WT over the acapsular B. theta having a similar initial probability of colonization of the cecum: difference in lag phase (mean time to growth commencement in acapsular (τa) and WT (τw)) and difference in net growth rate (growth rate in acapsular (ra) and WT (rw); clearance rate in acapsular (ca) and WT (cw)). Clearance can be due to both flow/loss in the fecal stream and death. (F, G) Estimation of the (F) competitive index and (G) colonization probability of the acapsular strain assuming a mean 4.5 hr difference in lag phase and the estimated difference in net growth rate between the WT and acapsular strains. Circles represent experimental data from (B) and Figure 1C. Triangles represent the best estimate and vertical line the higher and lower bound of the 95% confidence interval. See 'Methods' for parameter estimations. Data are included in Figure 3—source data 1.

Figure 3—source data 1

Competitive colonization with acapsular and WT strains.

https://cdn.elifesciences.org/articles/81212/elife-81212-fig3-data1-v2.xlsx

To better understand the mechanisms generating a competitive disadvantage for the acapsular strains, we performed a competition experiment in the same microbiota backgrounds, but this time, using tetracycline-resistant B. theta WT and barcoded erythromycin-resistant acapsular B. theta strains. WT strain density (Figure 3—figure supplement 1B) and the average increase in the WT relative to the acapsular strains (after adjusting for the initial ratio in the inoculum, Figure 3—figure supplement 1C) were similar between the two experiments. Interestingly, the colonization probability β was lower for the acapsular B. theta strain when co-colonizing with the WT strain than when colonizing alone (Figure 3B). This indicates that the competition with WT B. theta results in both a lower total population size and increased clonal extinction in the acapsular strains.

Longer lag phase and higher clearance rate explains fitness defect of acapsular B. theta in competitive colonization

To understand how acapsular B. theta can have an indistinguishable colonization probability when inoculated alone, but a major fitness defect in competition with B. theta WT, we carried out time courses of feces collection to estimate the net growth rates (i.e., growth minus clearance) of both strains colonized individually in OligoMM12 mice (Figure 3—figure supplement 2). Using peak-to-trough analysis, we observed that growth rates of B. theta in feces and cecum is similar, and higher than all other OligoMM12 strains at 8 hr post-inoculation, as would be predicted for active growth (Figure 3—figure supplement 3). Longitudinal feces collection was therefore used as a proxy for large-intestinal colonization. Tracking B. theta CFUg–1 in feces over 48 hr demonstrated that the net growth rate of acapsular B. theta is lower than that of WT (Figure 3C, WT: 0.50/hr and acapsular: 0.40/h, p<0.01). Interestingly, detectable exponential growth of acapsular strain starts around 4.5 hr later that for the WT (Figure 3D, WT: 3.2 hr and acapsular: 7.7 hr, p=0.02). As there is an inherent detection limit for CFU, as well as an intrinsic time delay due flow through the gastrointestinal tract this delay could be explained by (1) a classic lag phase (i.e., period of adaption before growth begins), (2) strongly increased killing of the acapsular B. theta during stomach and small intestinal transit, or (3) retention of acapsular B. theta in the non-growth-permissive small intestine. We could exclude differential retention of acapsular B. theta in the small intestine. Analysis of B. theta distribution in the small and large intestine at 8 hr post-colonization indicated that most B. theta had already arrived in the large intestine at this time point. There was no evidence of differential retention of acapsular and WT B. theta in the small intestine (Figure 3—figure supplement 4). We can also largely exclude killing prior to reaching the cecum as B. theta WT and acapsular clones have a very similar probability of colonization in single colonizations of OligoMM12 mice (Figure 2 and figure supplements). For early killing to explain out-competition of the acapsular strain, it would be necessary that the presence of WT B. theta increases acapsular B. theta killing in the stomach or small intestine. As the bacteria are at low density during early infection, it is unlikely that they affect either each other or the host prior to arrival in the large intestine. In contrast, the delay in detectable growth of the acapsular strain is observed both competitive and single colonization of OligoMM12 mice, consistent with a classical lag phase. The magnitude of the delay to detection is similar on single and competitive colonization with WT B. theta (Figure 3—figure supplement 2E and F), consistent with this being an intrinsic feature of the acapsular B. theta strain. We therefore propose that acapsular B. theta exhibits a classical extended lag phase in vivo, likely due to a longer adaption period for the acapsular B. theta to growth in the gut environment.

To further explore this hypothesis, we extended our one-step colonization model to include both a difference in lag phase after arrival in the cecum and/or a difference in net growth rates (Figure 3E). Combining these additional variables generated a model that quite well recapitulates the expected competitive fitness (Figure 3F) and colonization probabilities (Figure 3G) (see 'Methods' for a brief description of the model and Appendix 1 'Supplementary methods' for the description of all parameters used). Running the same model based only on identical growth rates, but different lag phase produces a worse prediction of the competitive index, while omitting the lag-phase difference produces a similarly good fit (Figure 3—figure supplement 5). Therefore, the competitive fitness defect of the acapsular B. theta strains can be explained by a slightly slowed in vivo net growth rate, with a small contribution from an extended in vivo lag phase.

Acute challenges modify B. theta population dynamics in vivo

Finally, as a proof of concept for neutral tagging in the study of established microbiota, we used our system to probe clonal extinction when an established B. theta population in the gut is challenged by two major environmental perturbations: (1) shifting from standard chow to a high-fat no-fiber diet (HFD) (David et al., 2014; Wotzka et al., 2019) and (2) infectious inflammation driven by Salmonella Typhimurium (Stm) (Maier et al., 2014). To exclude microbe–microbe interactions from the possible observed mechanisms, we monocolonized GF mice with a mixture of untagged and barcoded B. theta WT strains such that all tags were present with a roughly uniform distribution prior to intervention, that is, minimum loss, in the gut lumen prior to challenge (Figure 4A, Figure 4—figure supplement 1).

Figure 4 with 2 supplements see all
Acute challenges imposed a population bottleneck in the resident B. theta population.

(A) Schematic representation of experimental estimation of colonization survival probability after acute challenges. (B, C) Germ-free (GF) mice were colonized with an inoculum of ~109 CFU untagged tetracycline-resistant B. theta WT spiked with barcoded erythromycin-resistant WT strains. The number of spiked CFU of each individual barcode per experiment (n0) is described in Figure 4—figure supplement 1. (B) Population of B. theta in monocolonized ex-GF mice kept under standard chow (control) and during acute challenge with high-fat diet (HFD), infection with attenuated Salmonella (Stm) (ΔssaV) or avirulent (ΔssaVΔinvG) Salmonella. Points represent individual mice and boxplot quartiles provide summary statistics. p-Values were obtained by one-way ANOVA followed by Tukey’s honest significance test. (C) Probability of surviving in the cecum after 3 days of the acute challenge. Estimation based on six barcodes timer the total number of mice (Control = 12, HFD = 13, Stm attenuated = 14, Stm avirulent = 5). Circle represent the best estimate and vertical line the higher and lower bound of the 95% confidence interval. Data are included in Figure 3—source data 1.

Figure 4—source data 1

Acute challenges imposed a population bottleneck in the resident B. theta population.

https://cdn.elifesciences.org/articles/81212/elife-81212-fig4-data1-v2.xlsx

In the first set of challenges, after 4 days of colonization, we exposed mice to oral infection with 106–107 CFUs of a Stm strain either attenuated (SL1344 ΔssaV, no SPI-2) or fully avirulent (SL1344 ΔinvGΔsseD, no SPI-1 or SPI-2). Despite similar Stm loads (Figure 4—figure supplement 2A), the attenuated strain induces moderate intestinal inflammation in GF mice while the fully avirulent strain does not induce visible gut inflammation (Hapfelmeier et al., 2005; Stecher et al., 2005; Figure 4—figure supplement 2B). B. theta populations were monitored in feces before the challenge was administered and 3 days after the challenge in feces and cecum. In line with the models presented above, cecum content values were used for inference of the bottleneck size.

After 3 days of infection with attenuated Stm, the total B. theta population was reduced approximately 100-fold in feces (Figure 4B). In addition, the probability of an established clone to survive this acute inflammation challenge was approximately 1 in 2000 in cecum (Figure 4C). This means that the initial estimated population of 1010 CFUg–1B. theta in the cecum is reduced to an effective population of between 106–107 clones, while maintaining a total population size of between 109–1010 CFUg–1 (Figure 4B). Challenge with avirulent Stm has a limited effect on total population density of B. theta (Figure 4B); however, it still induces a bottleneck of approximately 1 in 250 in the cecum, potentially due to subclinical inflammation induced by colonization with a noninvasive Enterobacteriaceae (Figure 4C). Of note, we cannot formally exclude direct toxicity of Salmonella against B. theta based on these data. Rather around 90% of the B. theta clearance observed in virulent Salmonella infections can be attributed to the strong inflammatory response induced (Figure 4—figure supplement 2B).

In a second set of experiments, B. theta monocolonized mice were fed an HFD. Despite setting the initial barcoded population size in the cecum at between 100 and 1000 CFUg–1 (Figure 4—figure supplement 1), HFD feeding did not significantly change the cecum population size (Figure 4B), nor did it increase the loss of barcoded B. theta (Figure 4C). To calculate our limit of sensitivity with this system, we estimated that detecting loss of one barcode in one of the tested mice would occur with a bottleneck population size of between 109 and 1010 clones, that is., a population contraction of up to tenfold would be within the experimental noise of our measurements. Published reports indicate that B. theta is sensitive to bile acids (Wotzka et al., 2019), which should be abundantly induced by HFD feeding. The conclusion of the data is therefore rather than the current neutral tagging system is not sufficiently precise to detect the magnitude of population dynamics changes induced by HFD consumption in GF mice.

Discussion

Understanding the different mechanistic factors determining a new species can invade into a resident gut microbiota is of considerable importance for combating mucosal infections (Kreuzer and Hardt, 2020; Stecher, 2021), but also for rationally introducing new functions into existing communities (Cubillos-Ruiz et al., 2021). These factors can include, among others, nutrient/energy availability, environmental factors such as pH and flow/dilution rate (Arnoldini et al., 2018), and the presence of directly toxic or aggressive activities derived from the host (Cullen et al., 2015) or from other microbiota species (García-Bayona and Comstock, 2018), all of which can affect different microbes in different ways. Here, we condense all these mechanisms into three processes: factors affecting immigration rate (i.e., arrival into growth-permissive sites in the gut), factors affecting growth, and factors affecting clearance/death (Hoces et al., 2020). Combining in vivo experiments using genetically barcoded strains with mathematical modeling, we were able to empirically (e.g., net growth rates through plating) or deductively (e.g., probability of colonization) estimate the relative contribution of these three processes for colonization of B. theta under different conditions. In addition, we were able to quantify how fitness-relevant genetic changes in B. theta (production of capsular polysaccharide) affect colonization success.

When analyzing the effect of capsular polysaccharide expression on the process of colonizing mice with different resident microbiota, we found that the previously shown fitness disadvantage of acapsular strains (Coyne et al., 2008; Porter et al., 2017) depends on the microbiota context, rather than on host effects. The WT and the acapsular strains colonize mice with a relatively simple microbiota (LCM, Oligo) similarly well. However, in mice with more complex microbiota (SPF), B. theta can be outcompeted by the resident microbiota (Lee et al., 2013), and the acapsular strain engrafted significantly less well than the WT. While a possible explanation includes more robust intestinal immunity in fully colonized SPF mice (Cullen et al., 2015; Hsieh et al., 2020), it is also probable that expression of CPS is important for interaction with or protection against other microbes. Possible microbe-inflicted processes against which CPS can protect include microbe-on-microbe killing (Chatzidaki-Livanis et al., 2016; Wexler et al., 2016) and susceptibility to phages (De Sordi et al., 2019; Porter et al., 2020). Intriguingly, co-colonization experiments with mixed B. theta inoculums consisting of WT and acapsular strains recapitulate similar colonization probabilities for the acapsular strain as that observed in a complete SPF microbiota. As direct toxicity between the acapsular and WT strains can be largely excluded, this prompted us to examine growth kinetics of B. theta in the gut. Longitudinal analysis of fecal CFU densities demonstrated both a lower net growth rate in vivo (likely explained by a higher death rate rather than a lower replication rate) and a longer lag phase before commencing growing in the gut. Based on the data for competitive colonization in LCM mice, as well as the outcomes of our mathematical models, we can conclude that a small difference in net growth rate, combined with an extended lag phase, is sufficient to numerically explain the competitive fitness loss observed in numbers and colonization probability of acapsular B. theta.

The effect of two important challenges with known effects on resident commensals – high-fat feeding (Wotzka et al., 2019) and inflammation (Maier et al., 2014) – seems to depend on the ecological context in different ways. Feeding mice that are stably monocolonized with B. theta a fiber-less HFD imposes a bottleneck that must represent less than a tenfold reduction in effective population size of the B. theta at 3 days post-treatment, even though this intervention has been shown to increase bile acid concentrations to levels that inhibit B. theta growth (Wotzka et al., 2019). It has been shown that B. theta rapidly evolves to adapt to dietary challenges in the context of a resident microbiota (Dapa et al., 2022). Therefore, the observed mild population bottleneck imposed by HFD feeding might only manifest if B. theta competes against other gut microbiota members, for example, that are more resistant to bile salts (e.g., Escherichia coli, see Wotzka et al., 2019). When infecting mice that are stably monocolonized with B. theta with Salmonella, we observed a larger decrease in B. theta clonal survival probability, which is consistent with the sensitivity of commensal species to gut inflammation (Stecher et al., 2007). Consistent with inflammation driving the main part of this phenomenon, we see less of a bottleneck when infected mice with a SPI1/2 double-mutant avirulent Salmonella that does not cause clinically overt inflammation in the gut. Interestingly, inflammation also causes a population bottleneck in the infecting Salmonella population (Maier et al., 2014), but this is less pronounced than the one we observe for B. theta, and the total population size of Salmonella rapidly recovers after this bottleneck. Therefore, the rapid killing/clearance of gut luminal B. theta seems to be representative of microbiota suppression that underlies the loss of colonization resistance in Salmonella-induced colitis.

Our mathematical models are based on certain assumptions that are useful to simplify calculations but always risk introducing bias. Most notably, we have made estimates for a single population of bacteria that has a constant growth and clearance rate. Necessarily the reality is more complex than this – the nutrient profile and motility of the intestine will vary with circadian rhythm. Also, previous work has demonstrated that particular CPS-expressing clones may have an advantage in colonizing the dense mucus layers of the distal colon (Donaldson et al., 2018) although LCM studies suggest that at the population level B. theta grows at a similar rate in the mucus layer and lumen (Li et al., 2015). Nevertheless, recognizing these limitations, our estimates of colonization probability, growth, and clearance rates still give a good overview of the harsh processes with strong effects on the total intestinal B. theta population. It is also interesting to compare the neutral tagging approach to sequencing-based methods for growth rate estimation. It should be noted that these techniques give different information: peak-to-trough replication rate analysis reveals the growth rate at the time of measurement, whereas neutral tagging typically reveals population dynamics parameters averaged over much larger time spans. The approaches also have different limitations. While sequencing requires that the strain of interest is >0.1% of the total microbiota in order to generate sufficient confidence in the reads with reasonable sequencing runs, neutral tags can be used to examine very small populations. However, far more factors influence the interpretation of neutral tagging experiments (bottlenecks, clearance rates, etc.) than peak-to-trough ratios and therefore mathematical modeling is needed to understand the results. An interesting future direction of the field will be to include individual-based models that can evaluate the impact of bacterial clones with different distributions of growth/clearance rates, as well as working with experimental models (e.g., microfluidics) that would allow experimental investigation of the impact of single-cell level variation on total population behavior.

By combining mathematical modeling with direct quantification of bacterial population dynamics, we can gain insight into the major phenomena influencing colonization efficiency. Not only do these results help us understand the different steps of B. theta colonization, but they also serve as a proof of concept for studying other complex, multistep biological processes using the set of experimental and data-analysis tools we are describing here. This raises the possibility to optimize colonization conditions in order to promote the efficient engraftment of beneficial species into target microbiota or to better understand the processes of invasion of pathogens and the functional basis of colonization resistance.

Methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Genetic reagent (Bacteroides thetaiotaomicron)B. theta WTPorter et al., 2017Not applicabletdk-
Parent strain of B. theta VPI-5482 (ATCC 29148). Used to generate wild-type CPS mutants in this study.
Genetic reagent (B. thetaiotaomicron)B. theta acapsularPorter et al., 2017Not applicabletdk- Δcps1-8
Acapsular B. theta with deletion of capsular polysaccharide locus. Used to generate acapsular mutants in this study.
Genetic reagent (B. thetaiotaomicron)B. theta WT barcode 1This studyPRJEB57876 (ERP142888)tdk-:: pNBU2-cat-ermG-GFP-WITS1
B. theta WT strain isogenic barcode 1; GFP tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta WT barcode 2This studyPRJEB57876 (ERP142888)tdk-:: pNBU2-cat-ermG-GFP-WITS2
B. theta WT strain isogenic barcode 2; GFP tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta WT barcode 11This studyPRJEB57876 (ERP142888)tdk-:: pNBU2-cat-ermG-GFP-WITS11
B. theta WT strain isogenic barcode 11; GFP tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta WT barcode 17This studyPRJEB57876 (ERP142888)tdk-:: pNBU2-cat-ermG-mCherry-WITS17
B. theta WT strain isogenic barcode 17; mCherry tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta WT barcode 19This studyPRJEB57876 (ERP142888)tdk-:: pNBU2-cat-ermG-mCherry-WITS19
B. theta WT strain isogenic barcode 19; mCherry tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta WT barcode 21This studyPRJEB57876 (ERP142888)tdk-:: pNBU2-cat-ermG-mCherry-WITS21
B. theta WT strain isogenic barcode 21; mCherry tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta acapsular barcode 1This studyPRJEB57876 (ERP142888)tdk- Δcps1-8:: pNBU2-cat-ermG-GFP-WITS1
B. theta acapsular strain with isogenic barcode 1; GFP tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta acapsular barcode 2This studyPRJEB57876 (ERP142888)tdk- Δcps1-8:: pNBU2-cat-ermG-GFP-WITS2
B. theta acapsular strain with isogenic barcode 1; GFP tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta acapsular barcode 11This studyPRJEB57876 (ERP142888)tdk- Δcps1-8:: pNBU2-cat-ermG-GFP-WITS11
B. theta acapsular strain with isogenic barcode 1; GFP tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta acapsular barcode 17This studyPRJEB57876 (ERP142888)tdk- Δcps1-8:: pNBU2-cat-ermG-mCherry-WITS17
B. theta acapsular strain with isogenic barcode 17; mCherry tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta acapsular barcode 19This studyPRJEB57876 (ERP142888)tdk- Δcps1-8:: pNBU2-cat-ermG-mCherry-WITS19
B. theta acapsular strain with isogenic barcode 19; mCherry tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta acapsular barcode 21This studyPRJEB57876 (ERP142888)tdk- Δcps1-8:: pNBU2-cat-ermG-mCherry-WITS21
B. theta acapsular strain with isogenic barcode 21; mCherry tag; erythromycin resistant.
Genetic reagent (B. thetaiotaomicron)B. theta WT untaggedThis studyPRJEB57876 (ERP142888)tdk-:: pNBU2-bla-tetQb
Untagged strain. B. theta WT strain with GFP insert; tetracycline resistant.
Genetic reagent (B. thetaiotaomicron)B. theta acapsular untaggedThis studyPRJEB57876 (ERP142888)tdk- Δcps1-8:: pNBU2-bla-tetQb
Untagged strain. B. theta acapsular strain with GFP insert; tetracycline resistant.
Genetic reagent (Salmonella enterica)Stm attenuated (M3103)Diard et al., 2017Not applicableΔssaV
Salmonella enterica serovar Typhimurium (SL1344), attenuated (SPI-2 KO)
Genetic reagent (S. enterica)Stm avirulent (M2702)Diard et al., 2017Not applicableΔinvG ΔssaV
Salmonella enterica serovar Typhimurium (SL1344), avirulent (SPI-1 KO and SPI-2 KO).
Sequence-based reagentWITS01_FMaier et al., 2014Forward primer barcoded strain 1acgacaccactccacaccta
Sequence-based reagentWITS02_FMaier et al., 2014Forward primer barcoded strain 2acccgcaataccaacaactc
Sequence-based reagentWITS11_FMaier et al., 2014Forward primer barcoded strain 11atcccacacactcgatctca
Sequence-based reagentWITS17_FMaier et al., 2014Forward primer barcoded strain 17tcaccagcccaccccctca
Sequence-based reagentWITS19_FMaier et al., 2014Forward primer barcoded strain 19gcactatccagccccataac
Sequence-based reagentWITS21_FMaier et al., 2014Forward primer barcoded strain 21acaaccaccgatcactctcc
Sequence-based reagentydgA_RMaier et al., 2014Common reverse primer for all tagged strainggctgtccgcaatgggtc
Sequence-based reagentBTt70-CHFJacobson, 2017pNBU2 vector genome integration testTTCAAATTGCTCGGTAAAGCTC
Sequence-based reagentBTt70-CHRJacobson, 2017pNBU2 vector genome integration testAAAACCTTGATTTTACGGGAC
Sequence-based reagentBTt71-CHF3Jacobson, 2017pNBU2 vector genome integration testTTCGAGGAATGAAGCATCTCCGTA
Sequence-based reagentBTt71-CHR3Jacobson, 2017pNBU2 vector genome integration testACCGTTCCGATTCAATTTCGT
Sequence-based reagentIntN2BTt71-CHF3Jacobson, 2017pNBU2 vector genome integration testTTTCCGGCTCTCCAATGCAA

Bacterial strains and cultures

B. theta strains were grown anaerobically (5% H2, 10% CO2, rest N2) at 37°C, overnight in brain heart infusion (BHI) supplemented media (BHIS: 37 g/L BHI [Sigma]; 1 g/L-cysteine [Sigma]; 1 mg/L Hemin [Sigma]). For enrichment cultures in plates, we used BHI-blood agar plates (37 g/L BHI [Sigma]; 1 g/L-cysteine [Sigma]; 10% v/v defribinated sheep blood [Sigma]). Antibiotics were added to liquid cultures or plates as required for strain selection: erythromycin 25 µg/L or tetracycline 2 µg/L. In the case of BHI-blood agar plates used for cloning or gut content enrichment, we additionally added gentamycin 200 µg/L to all plates to prevent growth of other microbiota species. Plates were incubated for 48–72 hr at 37°C in anaerobic conditions. For a complete list of the bacterial strains used in this study, see Key Resources Table.

Isogenic barcode construction and integration

Genetic tags, fluorescent proteins, and antibiotic resistance were introduced by using the mobilizable Bacteroides element NBU2, which integrates into the Bacteroides genomes at a conserved location at either BTt70 or BTt71 (Wang et al., 2000). Gene fragments containing a unique 40 bp sequence (biding site for forward primer) and a 609 bp sequence with the ydgA pseudogene (common binding site for reverse primer) were synthesized (gBlocks, Integrated DNA Technologies) and cloned by Gibson Assembly Master Mix (NEB) into an NBU2 plasmid carrying the erythromycin-resistant cassette ermG (barcoded B. theta strains) and a fluorescent GFP or mCherry protein (see Key Resources Table for specific combination of fluorescent protein and tag). A similar NBU2 plasmid carrying the tetracycline-resistant cassette tetQb and the GFP protein was used to construct the untagged strains (B. theta untagged). All fluorescent protein genes have high-expression promoter and an optimized RBS (Whitaker et al., 2017). For both, we used 10 µL of desalted assembly reaction products to transform competent E. coli S17-1 cells (mid-log cells, washed three times in deionized ice-cold water) by electroporation (V = 1.8 kV; MicroPulser, Bio-Rad). After 1 hr recovery at 37°C in 1 mL of LB, cells were plated on LB plates with chloramphenicol (12 µg/mL) and grown overnight. Plasmid-carrying E. coli S17-1 and B. theta strains were cultured overnight in 5 mL of liquid media. E. coli S17-1 and B. theta were washed with PBS, pooled in 1 mL of PBS, plated BHI-blood agar plates without antibiotics, and grown aerobically at 37°C for at least 16 hr. The lawn of E. coli S17-1 and B. theta was collected in 5 mL of PBS, homogenized by vortex, and 100 µL were plated in BHI-blood agar plates supplemented with erythromycin 25 µg/L and gentamycin 200 µg/L. After 48 hr, single colonies were streaked in fresh BHI-blood agar plates with antibiotics to avoid potential contamination with WT strains. Successful insertion in the BTt70 or BTt71 sites was evaluated by PCR (Key Resources Table). To minimize potential variation, we used strains with a single insertion in BTt70. In summary, the barcoded strains carried the barcode, erythromycin resistance cassette and fluorescent protein (GFP or mCherry) in the genome. The untagged B. theta carried a tetracycline resistance cassette together with a GFP protein inserted at the same position in the genome.

Quantification of barcoded B. theta from mixed samples

One investigator blinded to sample metadata (i.e., microbiota, bacteria strain used, acute challenge used) performed the sample processing and qPCR quantification of barcodes. Samples were serially diluted and plated on appropriate selective BHIS agar (gentamycin 200 µg/L plus either erythromycin 25 µg/L or tetracycline 2 µg/L) and cultured in 5% H2, 10% CO2, rest N2, at 37°C for 48 hr (Coy Anaerobic tent). CFU determination was carried out by counting, then colonies were washed from the plates (all plates with at least 30 colonies were included), pooled in 5 mL of PBS and homogenized by vortex. Genomic DNA was isolated with the QIAamp DNA Mini Kit (QIAGEN). qPCR was performed using with FastStart Universal SYBR Green Master Mix (Roche, Cat# 4385610). Primers (Key Resources Table) were mixed to a final concentration of 1 µM. Between 160 and 200 ng of DNA was amplified in duplicates using StepOne Plus or QuantStudio 7 Flex instruments (Applied Biosystems) using the following protocol: initial denaturation at 95°C for 14 min followed by 40 cycles of 94°C for 15 s, 61°C for 30 s, and 72°C for 20 s as described previously. As qPCR reactions for these barcodes have identical efficiencies, genomic DNA extracted from a single barcoded strain was used as an internal standard and CT values were backcalculated to this standard curve to generate a relative frequency of each barcode in the pooled colonies. These relative frequencies were then multiplied by the total CFU/g of barcoded B. theta to obtain the CFU/g of each barcoded strain.

In vitro growth curves and competition

Individual B. theta strains were grown overnight on BHIS. Stationary-phase cultures were washed with PBS, and O.D. was quantified and adjusted to 0.05 in 200 µL of fresh BHIS in a round 96-well tissue culture plates. Plates were transferred into the anaerobic tent and growth was quantified at 37°C with shaking using a plate reader (Infinite PRO 200, Tecan).

For competition experiments, stationary-phase cultures B. theta WT or acapsular strains were washed with PBS, O.D. was quantified and adjusted to approximately 5 × 106 CFU/mL per strain (one B. theta untagged and six B. theta barcoded strains) in 10 mL of fresh BHIS. An aliquot of this mix was serially diluted and plated in BHI-blood agar plates with the respective antibiotics for CFU quantification. Cultures were kept overnight, with shaking (800 rpm) at 37°C in the anaerobic tent. Afterward, an aliquot was plated as described before. For assessing the competition between B. theta barcoded strains, we isolated DNA from one of the dilutions used for quantification and assessed the relative distribution of the tags by qPCR (see ‘Quantitative PCR’ section). For assessing the competition among strains with different antibiotic resistances, we calculated the competition index by dividing the CFU/mL of the untagged B. theta untagged strain (tetracycline-resistant) by the adjusted number of B. theta barcoded strains (erythromycin-resistant; CFU/mL divided by six, as all the barcoded strains were present in the culture).

Mice

All animal experiments were performed with approval from the Zürich Cantonal Authority under license number ZH120/19 and ZH009/21. In all experiments, we used mice with C57BL/6J genetic background, between 12 and 15 weeks old and of variable gender. C57BL/6J GF and gnotobiotic mouse lines (LCM [Stecher et al., 2010]; OligoMM12 [Brugiroux et al., 2016]) were raised in surgical isolators under high hygiene standards at the ETH Phenomic Center and were regularly tested for contamination by aerobic and anaerobic cultivation, culture-independent assessment of intestinal bacterial densities and serology/PCR for common viruses and eukaryotic pathogens. Note that all LCM mouse lines were bred for 1 year (OligoMM12) or more than 10 years (LCM) with their gnotobiotic microbiota. C57BL/6J SPF mouse line was raised in IVC cages in a different barrier unit of the same facility. Mice were transferred to our experimental facility in sterile, tight closed cages and house into the IsoCage P- Bioexclusion System (Tecniplast) for 24–48 hr before the experiment to adapt to new housing conditions. In all experiments, standard chow and water was prepared under strict aseptic conditions to avoid any potential contaminations. Although mice themselves were not randomized on each experiment, cages containing appropriate mouse numbers were randomly assigned to each inoculum/treatment and both genders are represented in all groups.

In vivo growth curves and competition

B. theta WT WITS 01 or B. theta acapsular WITS 01 strain was grown overnight in BHIS. Stationary-phase cultures were washed with PBS, and an inoculum of ~5 × 107 CFUs/100 µL dose was prepared. C57BL/6J mice carrying the described microbiota composition (LCM, OligoMM12, SPF, see figure legends for specific group numbers) were gavaged with the inoculum (either B. theta WT or B. theta acapsular). Fecal pellets were collected approximately every 4 hr post-inoculation. Fecal pellets were weighted and homogenized in 500 µL of PBS with steel ball by mixing (25 Hz, 2.5 min) in a TissueLyser (QIAGEN). Serial dilutions were plated for quantification on BHI-blood agar plates supplemented with gentamycin 200 µg/L and erythromycin 25 µg/L and cultured in anaerobic conditions. Note that zero colonies grew from the feces of LCM, OligoMM12, and SPF mice on BHIS plates with gentamycin and erythromycin.

For competition experiments, B. theta strains were grown overnight in 8 mL of BHIS with corresponding antibiotics (erythromycin 25 µg/L or tetracycline 2 µg/L). Each culture was spun down at 3000 × g for 20 min and resuspended in 10 mL of PBS and individual O.D. was measured (cell number estimation 1 O.D. = ~4 × 108 cells/mL). Each strain was adjusted to approximately 5 × 106 CFU/100 µL dose per strain in the inoculum mix (one B. theta untagged and six B. theta barcoded strains). GF mice were gavaged with 100 µL of the inoculum. After 48 hr, fecal and cecal contents were collected. Fecal content was homogenized and plated for quantification as described before. Cecal content was resuspended in 1 mL of PBS and homogenized with steel ball by mixing with the same protocol (25 Hz, 2.5 min). Serial dilutions were prepared and plated in BHI-blood agar plates supplemented with gentamycin plus either erythromycin or tetracycline for CFU quantification of each strain. Like the in vitro competition experiment, we isolated DNA from one of the dilutions used for quantification to assess the competition between B. theta barcoded strains. Relative distribution of the tags was obtained by qPCR. For calculating the competition index among strains with different antibiotic resistances, we divided the bacteria density of the untagged B. theta untagged strain by a sixth of the bacteria density of the total B. theta barcoded strains (as all six barcoded strains were present in the culture).

Colonization experiments

Stationary-phase B. theta cultures were prepared overnight as described before. Each culture was washed with PBS to remove residual antibiotics and adjusted in the inoculum based on its O.D.600nm. Unless otherwise stated, the untagged strain (B. theta untagged) was present at ~5 × 107 CFUs/100 µL dose of the inoculum. For the B. theta barcoded strains, we prepared an initial 1:1:1:1:1:1 mix of all six barcoded strains in 50 mL of PBS at a concentration of 105 CFU/mL of each strain. After mixing by vortex for 1 min, the required amount of B. theta barcoded strains was prepared by serial dilution and spiked into the inoculum (between 30 and 5 × 104 CFUs depending on the experiment). LCM (Oligo) and SPF C57BL/6J mice were gavaged with the 100 µL inoculum. To check composition, the inoculum was serially diluted and plated for quantification of CFUs in BHI-blood agar plates supplemented with gentamycin 200 µg/L plus either erythromycin 25 µg/L or tetracycline 2 µg/L. In addition, three whole doses (100 µL) were directly plated on three BHI-blood agar plates with gentamycin plus erythromycin or tetracycline (as appropriate) to address initial distribution of B. theta barcoded strains in the inoculum by quantitative PCR (qPCR). Unless otherwise stated, 2 days after colonization, mice were euthanized and cecal content was collected in 2 mL Eppendorf tubes and weighted. Cecal content was homogenized as described before. Serial dilutions were prepared and plated in BHI-blood agar plates supplemented with gentamycin plus either erythromycin or tetracycline for CFU quantification. In addition, 100 µL of homogenized content was plated directly in BHI-blood agar plates with gentamycin plus erythromycin to generate biomass of the assessment of the distribution of B. theta barcoded strains by qPCR.

In vivo competition of post-colonization versus original strains

To discard potential increased colonization fitness in the barcoded strains that were present in the cecum content after 48 hr, we isolated single B. theta WT barcoded strains that were present in the cecal content of SPF mice during a colonization experiment. Single colonies were expanded in liquid media, and the presence of a single strain was confirmed by qPCR analysis of the barcodes present. We randomly selected three of the B. theta WT barcoded strains isolated from the cecal content. We prepared an inoculum as described before for the in vivo competition experiments with approximately 5 × 106 CFU/100 µL of each strain in the inoculum mix. We complemented the inoculum with the remaining three B. theta WT barcoded strains coming from the original stock. SPF mice were inoculated by gavage and cecal content was collected 48 hr later. Cecal content was processed as described before for CFU quantification and relative barcode distribution by qPCR.

Diet modification and infection challenge experiments

In accordance with what we described before, B. theta WT strains were grown overnight in BHIS with corresponding antibiotics. As the untagged strain B. theta untagged was used in higher concentrations, we prepared between 50 and 100 mL of liquid culture depending on the number of mice to colonize. Inoculum was prepared as previously described with a concentration of 108–109 CFUs/100 µL dose of untagged B. theta untagged, spiked with approximately 30 CFU of each B. theta barcoded strains. GF mice were gavaged with 100 µL of the inoculum. Mice were maintained on standard chow diet (Kliba Nafag, 3537; autoclaved; per weight: 4.5% fat, 18.5% protein, ~50% carbohydrates, 4.5% fiber) for 4 days. Afterward, mice were housed on fresh IsoCages, and challenges were applied as follows: (1) control group (continuation of standard chow diet); (2) Western-type diet without fiber (BioServ, S3282; 60% kcal fat; irradiated; per weight: 36% fat, 20.5% protein, 35.7% carbohydrates, 0% fiber); or (3) infection with 5 × 107 CFU of attenuated Salmonella Typhimurium (Stm SL1344ΔSPI-2). Fecal pellets were collected pre-challenge (day 0) and during the following 3 days. On day 3, mice were euthanized and cecal content was collected. Fecal pellets were weighted and homogenized in 500 µL of PBS as described before. Serial dilutions were prepared and plated in BHI-blood agar plates supplemented with corresponding antibiotics for CFU quantification. In addition, 100–300 µL of homogenized content was plated directly for further assessment of the distribution of B. theta barcoded strains by qPCR. Cecal content was processed as previously described.

DNA extraction for community composition analysis and growth estimates

To assess microbial community composition, fecal pellets from LCM mice and cecum content from SPF mice were obtained and flash frozen. To generate growth estimates of B. theta in an OligoMM12 background, both flash-frozen fecal pellets and cecum content were used. For enzymatic lysis, half a fecal pellet or roughly 30 mg of flash-frozen cecum content per sample were incubated in 100 μL of 1× TE buffer (30 mM Tris-HCl and 1 mM EDTA) supplemented with 30 mg/mL Lysozyme (Sigma-Aldrich), 1.6 U/mL Proteinase K (New England Biolabs), 10 U/mL Mutanolysin (Sigma-Aldrich), and 1 U/μL SUPERase•In RNase Inhibitor (Invitrogen) at room temperature for 10 min. To aid disruption, one 2 mm metal bead was added, and the samples were vortexed every 2 min. Subsequently, the samples were mixed with 550 μL RLT Plus buffer (QIAGEN) complemented with 5.5 μL 2-beta-mercaptoethanol (Sigma-Aldrich) and prefilled tubes with 100 μm Zirconium beads (OPS Diagnostics LLC). The samples were disrupted twice at 30 Hz for 3 min using the mixer mill Retsch MM400 with 5 min incubation at room temperature between each disruption. DNA was extracted from all samples with the DNA/RNA Mini kit (QIAGEN) following the standard protocol and eluting the DNA in 100 μL elution buffer (EB). For the LCM samples, three negative extraction controls and three negative PCR controls were included. For the SPF samples, one water sample was used as negative extraction control and subsequently split into three negative library controls undergoing the same library preparation as all samples. The integrity and quality of the extracted DNA was assessed on a Qubit and Fragment Analyzer, respectively. The DNA was purified by overnight ethanol precipitation at −20°C in 275 μL ice-cold Ethanol (Sigma-Aldrich), 10 μL 3 M sodium acetate (Invitrogen), and 1 μL 20 mg/mL glycogen (Invitrogen) with subsequent centrifugation at 4°C for 30 min and two wash steps in 500 μL ice-cold 75% ethanol with centrifugation at 4°C for 10 min each time. The DNA purity was assessed on a Nanodrop.

16S sequencing for LCM, OligoMM12, and SPF community composition analysis

16S amplicon libraries were generated from 50 ng input DNA with the Illumina primer set 515F (Parada et al., 2016) and 806R (Apprill et al., 2015), 12 cycles in PCR 1 and 13 cycles in PCR 2. Three positive controls containing 11 ng input DNA of ZymoBIOMICS Microbial Community DNA Standard II (Zymo Research) were used. Illumina Unique Dual Indexing Primers (UDP) were used for library multiplexing. A 12 pM library pool spiked with 20% PhiX was sequenced at the Functional Genomics Center Zurich (FGCZ) using the MiSeq platform and 2 × 300 bp PE-reads with a target fragment size of 450 bp, resulting in approximately 60,000 and 400,000 reads per sample for the LCM and OligoMM12/SPF sequencing runs, respectively. Raw sequencing data from LCM and SPF mice can be accessed on ENA (https://www.ebi.ac.uk/ena/browser/home) under Project ID PRJEB57876. The OligoMM12 was previously published and can be accessed on ENA under the Project ID PRJEB53981 (Hoces et al., 2022).

Metagenomic sequencing for B. theta growth estimates in OligoMM12 background

Genomic DNA was sheared to a target fragment size of 350 bp length with the ultrasonicator Covaris LE220 following a standard protocol (30 μL volume, 220 W peak incident power, 89 s treatment time). Metagenomic libraries were prepared from 10 ng sheared DNA with the NebNext Ultra II DNA Library Prep Kit for Illumina. Sample-specific adaptations included tenfold adapter dilution, no size selection by adding 1 volume (89 μL) of Cytiva Sera-Mag Select beads in the first cleanup and eight PCR cycles in the amplification step. Nebnext Multiplex Oligos for Illumina (Dual Index Primer Set 1) were used for library multiplexing. The final cleanup was done with a left side size selection by adding 0.7 volumes (35 μL) of Cytiva Sera-Mag Select beads. A 1 nM library pool spiked with 3% PhiX was sequenced at the FGCZ using the NextSeq2000 platform and 2 × 150 bp PE-reads with a target fragment size of 500 bp, resulting in approximately 30,000,000 reads per sample.

Data analysis

16S community composition analysis

Raw sequencing reads from all samples and 3–6 positive/negative controls served as input for the inference of ASVs using dada2 v1.22 (Callahan et al., 2016). Primer sequences (515F, 806R) were removed using cutadapt v2.8 (Martin, 2011), and only inserts that contained both primers and were at least 75 bases were kept for downstream analysis. Next, reads were quality filtered using the filterAndTrim function of the dada2 R package (maxEE = 2, truncQ = 3, trimRight = (40, 40)). The learnErrors and dada functions were used to calculate sample inference using pool = pseudo as parameter. Reads were merged using the mergePairs function and bimeras were removed with the removeBimeraDenovo (method = pooled). Remaining ASVs were then taxonomically annotated using the IDTAXA classifier (Murali et al., 2018) in combination with the Silva v138 database (Quast et al., 2013) available at http://www2.decipher.codes/Downloads.html. The resulting ASV table was used to check for contaminations with the decontam R package (Davis et al., 2018) using both frequency-based and prevalence-based classification with a single probability threshold of 0.05 computed by combining both probabilities with Fisher’s method (method = combined). ASVs classified as contaminants as well as the positive/negative controls were excluded from downstream analyses. The remaining ASV abundance table was downsampled to a common sequencing depth (28,000 reads per sample for LCM and 190,000 reads per sample for Oligo/SPF) to correct for differences in sequencing depth between samples using the rrarefy function of the vegan R package. Relative abundance plots at different taxonomic levels were generated (LCM at species level, OligoMM12 at strain level, SPF at family level).

For assessing the LCM composition, ASVs were clustered at 97% sequence identity with VSEARCH (usearch_global) (Rognes et al., 2016), which resulted in eight distinct ASVs with a maximum sequence identity of 96% between the two most similar ASVs. These ASVs were annotated at species level by alignment to 16S sequences of known community members from the original Schaedler flora (ASV01), from the OMM12 community (ASV03, ASV04, ASV08) and by alignment against the Silva v138 database. Due to annotation inconsistencies, ASV05 could only be annotated at the family level. The OligoMM12 strains were identified using the package bio for rRNA sequence extraction from the GenBank accessions described earlier (Hoces et al., 2022) and the tool VSEARCH (search_exact) (Rognes et al., 2016) for sequence alignment to the 16S sequences from the detected ASVs. ASVs with a mean relative abundance below 0.05% across all samples were grouped into ‘Other.’ Megasphaera was detected at a mean relative abundance of 0.06% but was also grouped into the category ‘Other’ since it was not knowingly part of the original OligoMM12 community. The category ‘Other’ in total amounted to roughly 0.11% of the total relative abundances, thus the oligo strains represented at least 99.8% of the detected ASV abundances. For the SPF community composition, ASVs were clustered at family level. ASVs with mean relative abundance below 1% or without taxonomic annotation at family level were grouped into the category ‘Other.’ The category ‘Other’ in total amounted to roughly 5% of the total relative abundances.

Metagenomic analysis for growth rate estimation

Sequencing reads from all metagenomic samples of B. theta in OligoMM12 background from feces and cecum were quality filtered using BBMap (v.38.71; Bushnell, 2014) by removing sequencing adapters from the reads, removing reads that mapped to quality control sequences (PhiX genome) and discarding low-quality reads using the parameters trimq = 14, maq = 20, maxns = 0, and minlength = 45. The in situ growth rate prediction tool CoPTR (Joseph et al., 2022) was used to compute growth rate estimates from the quality-controlled metagenomic reads by aligning them against a database containing all 12 OligoMM12 genomes available under Bioproject PRJNA317592 and the B. theta genome available under GenBank accession number CP092641.1.

Identification of genomic variants among B. theta strains

Genomic DNA from all B. theta strains was isolated from overnight cultures using the QIAamp DNA Mini Kit (QIAGEN). Samples were sent for whole genome sequencing at Novogene. Data preprocessing pipeline for adapter trimming and contaminant filtering is described at https://methods-in-microbiomics.readthedocs.io/en/latest/preprocessing/preprocessing.html. The raw reads for each strain were trimmed and filtered using BBMap v. 38.18 (Bushnell, 2014). The reads were mapped against Bacteroides thetaiotaomicron strain VPI 5482 genome (CP092641.1) using bwa v. 0.7.17 (Li and Durbin, 2009). The resulting bam files were sorted according to the coordinates and indexed using samtools-1.9 (Danecek et al., 2021). The duplicated reads were removed using gatk v.4.2 MarkDuplicates (McKenna et al., 2010). Variant calling and filtering (bcftools filter -Ov -sLowQual -g5 -G10 -e 'QUAL <10 || DP4[2]<10 || DP4[3]<10 ||(DP4[2]+DP4[3])/sum(DP4)<0.9 || MQ <50') was performed using Bcftools v1.133. The variant annotation was done using snpEff (Cingolani et al., 2012).

Mathematical modeling overview (see Appendix 1 ‘Supplementary methods’ for more detailed description)

Estimation of colonization probability based on lost tags

Let us denote C the bacterial concentration in the prepared solution. If we have volume V of this solution, then there are N=CV bacteria. Therefore, the probability to have taken n0 starting bacteria into an inoculum of volume v0 is

(1) p(n0)=Binomial distribution(N,n0V)=(v0V)n(1v0V)Nn0N!(Nn0)!n0!

when N=cV is large and v0V,

(2) p(n0) Poisson distribution(Nv0V)=(Nv0V)n0expexp (Nv0V) n0!

We define β as the probability for each bacterium to get to the cecum alive, and then have its lineage survive until measurement. Logically, the probability for a barcoded B. theta strain not to be present at measurement time is the zero of the Poisson distribution of average βn0, and thus

(3) ploss=expexp -βn0

n0 is estimated via the concentration and volume of the inoculum, and ploss is estimated via the number of tags lost divided by the total number of tags. Therefore, β is estimated as

(4) βloglog (nloss tagsntags) n0

To consider the fact that not all tags have the same n0 when we pool data from multiple experiments, β is actually estimated by maximizing the probability of the experimental observations:

(5) LLL=i=1ωlog((exp(βni))li(1exp(βni))1li)

This expression is also used for calculating the confidence interval, as detailed in Appendix 1.

Estimation of colonization probability based on variance

The variance on the proportions is

(6) var(p)=1h1(pi1h)2=1h1(ninj1h)2

In the limit where the initial number of bacteria are of the same order of magnitude, we find

(7) var(p)var(p0)1hnj,0var1m12

with varp0 the variance in proportions in the inoculum, nj,0 the total number of tags in the inoculum, and var1/m12 the relative variance starting from one bacterium. We find (see Appendix 1 ‘Supplementary methods’) that var1/m12 is 2/(colonization probability). var1/m12 can be estimated for each mouse using equation (7), and the average variance is used to estimate var1/m12 . The standard error on var1/m12 is used to obtain the confidence interval for the colonization probability.

Estimation of clearance rate due to flow

We examined the expected magnitude of the effect of an extended lag phase in the cecum on colonization probability to determine whether this is consistent with our observed neutral tagging data. It should be noted that the cecum is a dynamic environment with pulsatile arrival of material from the small intestine and loss of material to the feces. This generates a clearance rate due to flow on top of any clearance rate due to bacterial death. Assuming that the main site of growth of B. theta is the cecum/upper colon, the parameter for clearance due to flow can be estimated by quantifying the volume of cecum content lost per day. This can be empirically estimated by measuring (1) fecal dry mass produced per day, and (2) the water content of cecum content. Assuming minimal change in dry mass during colon transit in the mice, this infers a dilution rate of cecal content in the order of 0.12 volumes/hr in a GF mouse and 0.18 volumes/hr in an SPF mouse; LCM mice will have a value in between these two. Bacterial clearance due to killing will contribute over and above these values. Of note, bacteria with a long lag phase after introduction into the cecum will be cleared by the flow before growth can start, that is, during the early phase of colonization this will be a determinant of colonization probability.

Estimation of cecum turnover rates
Water fraction in cecal mass (%)Dry fecal excretion (g/day)Estimated wet cecal mass excretion (g/day)Wet cecal mass (g)Estimated cecum turnover rate (volume/day)Estimated cecum turnover rate (volume/hr)
Germ-free80.9 (0.4)1.55 (0.27)8.12 (1.42)2.83 (0.59)2.87 (0.78)0.12 (0.03)
Specific pathogen free76.2 (1.2)0.81 (0.09)3.40 (0.42)0.77 (0.32)4.42 (1.91)0.18 (0.08)
Estimation of the competitive index

We assume that bacteria have first a probability of survival qi (with i = w for the WT strain, and i = a for the acapsular strain). Then once the cecum is reached, they have a loss rate ci.

During an initial lag-phase τi , B. theta does not grow. On exit from lag phase, each bacterial strain grows logistically, initially at a rate ri , which saturates when approaching carrying capacity K with a factor (1-(A+W)/K). A and W denote the population density of acapsular and wildtype B. theta, respectively, when carrying capacity is reached, the total number of bacteria remains constant until the end of the experiment at time ttot , with both loss and replication ongoing and compensating each other. Given the growth rates for WT and acapsular B. theta are similar in vitro, we assume rw=ra also in vivo. Correspondingly, the difference in the initial net growth rates (ri-ci) considered to originate from differences in the in vivo clearance rates ca>cw . In the competition setting at carrying capacity, the majority of the population is composed of WT B. theta, such that the global population size is cleared with rate approximately cw . By definition, to maintain the total population size the effective replication rate of both strains of B. theta must exactly compensate this loss rate, cw . However, as the acapsular strain implicitly has a higher clearance rate ca, their net growth rate at carrying capacity becomes negative, that is, although the total population size remains constant, the acapsular B. theta population size will continuously decrease in frequency over time.

With this model (see detailed calculations in Appendix 1 ‘Supplementary methods’), we find that the relative ratio between WT and acapsular is

(8) qωqaexp exp ((netw+cω)(τaτw))(exp(netwneta)ttot)

For all the microbiota except SPF, the colonization probabilities (q) were similar for the WT and acapsular, as determined for single colonizations of LCM and OligoMM12 mice. Therefore, we assume qa = qw. For SPF, we use the ratio of qw /qa from the single-colonization experiments, adjusted for the fact that the full colonization probability also includes steps after the initial death before reaching the cecum. All the parameters used were determined from single-colonization experiments in the relevant microbiota backgrounds.

Estimation of colonization probability during competition

The overall survival probability for acapsular B. theta in the competition experiment is the colonization probability from single-colonization experiments, multiplied by a factor considering later loss (when the carrying capacity is reached by the WT and the acapsular decreases). The complete expression can be found in the corresponding section of Appendix 1 ‘Supplementary methods’.

Estimation of survival probability after challenge

In these experiments, at the time of the start of the challenge, the bacterial population is at carrying capacity, so the net growth rate is zero (i.e., the growth rate [likely limited by availability of nutrients] is the same as the loss rate due to flow/clearance). We also assume that the population size is known at the start of the challenge. The challenge may have different effects: it may impose a temporary bottleneck in the population (loss becomes higher than reproduction) or it may increase the loss rate (with the reproduction rate increasing enough to compensate), and thus the turnover of the population. In any case, we can calculate β as the probability that a bacteria present at day = 0 of the challenge has its lineage still alive and detectable in cecum content at day 3 via either mechanism. If there are n0 bacteria carrying a given barcode at day = 0, then

(9) ploss=(1β)n0

To estimate the total population size in the cecum before the challenge (n0), we assume that (1) all the animals are colonized at steady state at day = 0, and (2) the cecum mass and bacteria concentration is the same on day = 0 in all mice as it is on day = 3 in the control group. As there are small but relatively consistent differences in total CFU between feces and cecum, we used the average relationship between feces and cecum CFU in control mice on day = 3 to estimate the bacterial concentration in the cecum at day = 0. We assumed cecum barcoded strain abundance based on qPCR/plating and concentration correction of fecal samples on day = 0. In addition, ploss was estimated via the number of barcodes lost from cecum content on day = 3 divided by the total number of barcoded strains across all mice analyzed. β can then be estimated based on equation (6).

Statistical analysis

Sample size was determined based on previous experiments (Maier et al., 2014; Porter et al., 2017; Wotzka et al., 2019) using at least five mice per group where large effect size was expected. All group sizes are described in the figure legends.

Where errors are expected to be log-normal distributed (e.g., CFU density comparisons), all statistical tests were carried out on log-normalized data. One-way ANOVA followed by Tukey’s honest significance test was used for comparison of three or more groups. For model-inferred parameters, we compared mean and standard deviation calculated as described in the previous section. No data points were omitted from statistical analysis or for the estimation of parameters. Statistical analysis was performed with RStudio v1.2 and R v3.6.

Resource availability

Lead contact

Any further communication, including those related to resource sharing, may be directed to, and fulfilled by, the lead contact Emma Slack (emma.slack@hest.ethz.ch).

Materials availability

All strain and material generated in this study are available upon request to the corresponding author.

Appendix 1

1. Colonization

1.1 Initial bacteria number

Let us denote C0 the bacterial concentration in the prepared solution. Then if there is a volume V of solution, there are N=C0V bacteria. Then, the probability to have taken n0 bacteria in a volume v0 is

(A1) p(n0)=BinomialDistribution(N,v0/V)=(v0V)n0(1v0V)Nn0N!(Nn0)!n0!

In the limit of N=cV large and v0V,

(A2) p(n0)PoissonDistribution(Nv0/V)=(Nv0V)n0exp(Nv0V)n0!

1.2 Colonization probability via the loss: Base theory

Let us define β as the probability for each bacterium to get to the cecum alive, and then have its lineage survive until measurement. There is a priori no interaction early on between incoming bacteria as their concentration is initially low enough to limit the competition between them. Then, if started with an average of n0 bacteria (Poisson distributed), the probability for a given barcode to be absent at measurement time is the zero of the Poisson distribution of average βn0 , and thus

(A3) ploss=exp(βn0)

Then as n0 is estimated via the concentration and volume of the inoculum, and ploss is best estimated via the number of barcodes lost divided by the total number of barcodes, β is estimated as

(A4) βlog(nlosttags/ntags)n0

1.3 Colonization probability via the loss: Handling different starting inoculum sizes and calculating best estimate

The barcoded bacteria are not necessarily in equal concentrations in the inoculum and the data from several experiments with different inoculums need to be combined. Let us define w the number of barcoded multiplied by the number of mice. For each of these w, there were ni barcoded bacteria in the inoculum, and we define li as 1 of the barcodes was lost, 0 otherwise. Then the estimate of β is β, which maximizes probal1,....,lw the likelihood to observe {l1,l2,...lw}; and it is the same as maximizing the log likelihood. As for each barcoded bacterium the process will be considered as independent, it will then be simply the maximization of

(A5) LL=i=1wlog((exp(βni))li(1exp(βni))1li)

This is the same as maximizing the following expression:

(A6) LL=i=1w(liβni+(1li)log(1exp(βni)))

Let us note x=β.

(A7) dLLdx=i=1wni(li(1li)exp(xni)1exp(xni))

The value of x that makes this expression zero is found numerically and enables to infer β, while combining data with different initial number of bacteria in the inoculum coming from different mice.

1.4 Colonization probability via the loss: Handling different starting inoculum sizes and calculating confidence interval

To calculate a confidence interval, it is useful to obtain an estimate of the probability that a given β is the true value knowing the given observations, which can be denoted pβ|observations . Therefore, we used a Bayesian approach, pβ|observations=pobservations|βpβ/pobservations . In this expression, pobservations|β is the probability to observe {l1,l2,...lw} for a given β, so it is actually expLL . Then the prior on pβ has to be chosen. As it may be frequent to have a very low probability of survival, a flat prior for plogβ may be better than a flat prior for pβ . Then a similar reasoning can be done, plogβ|observations=pobservations|logβplogβ/pobservations , and with a flat prior for plogβ , plogβ|observationspobservations|logβ . Then exp(LL(log(β))) renormalized by its integral for logβ gives an estimate of the distribution of probability of inference of logβ . The exponential of the average of logβ on this distribution is very close to the value of β maximizing LL`. This distribution is very close to a Gaussian distribution (see Appendix 1—figure 1). Then the mean logβ ± twice the standard deviation of logβ on this distribution gives a 95% confidence interval.

Appendix 1—figure 1
Distribution of the probability estimation of log10β .

The colors represent experiments: low-complexity microbiota (LCM) WT (red), Oligo WT (green), specific pathogen free (SPF) WT (blue), LCM acapsular (pink), Oligo acapsular (light green), and SPF acapsular (light blue). Solid lines represent the numerical result for the probability distribution of β using renormalized equation (6), the dashed lines represent the normal distribution with the same mean log10β and same variance than the numerical distribution.

Note that if we had chosen a flat prior on pβ rather than plogβ , the result would have been very similar, though with a mean slightly further away from β maximizing LL`, and the distribution would be less similar to a normal distribution.

1.5 Colonization probability via the loss: Results

The results here are the values of β maximizing the log likelihood, with the confidence interval as explained in previous section:

Low-complexity microbiotaOligoSpecific pathogen free
WT10-1.50-1.60:-1.4010-1.54-1.67:-1.4210-2.35-2.50:-2.21
Acapsular10-1.43-1.56:-1.3110-1.49-1.63:-1.3510-3.65-3.79:-3.52

In principle, as we make multiple comparisons, the confidence interval should be adjusted, but here, even without correction, the LCM/oligo mice for both WT and acapsular are not significantly different (adjusting for multiple comparisons would make them even less distinguishable). In the case of the SPF mice, the resulting log10β is more than 10 standard deviations away for the WT bacteria, and more than 30 standard deviations for the acapsular bacteria, relative to LCM/Oligo mice. Within the SFP mice, the difference between WT and acapsular bacteria is more than 18 standard deviations. Because the differences are very large, even when adjusting for multiple comparisons, they would remain significant.

1.6 Colonization probability via the loss: Interpretation of the apparent loss probability (initial loss and subsequent loss)

1.6.1 Principle

What is obtained here is the probability for a bacterium in the inoculum to have seeded a lineage still present at measurement. For the lineage to be still present, it needs to

  • make it alive to the cecum. Let us denote this probability q.

  • escape stochastic fluctuations, mainly during the initial rounds of reproduction.

If there is little subsequent loss, then βq.

1.6.2 Minimal model

Here, we assume that each bacterium has a probability q of establishing, and then its lineage grows at a rate r and is lost at a rate c. After some time, carrying capacity is reached. Here, we assume the population is large, and thus stochastic loss will be negligible at this point. Therefore, as the stochastic effects occur when the population size is small, then it is legitimate to focus on this step and neglect saturation.

If a bacterium survives the first step (probability q), and its lineage reaches size n at time time t with probability p1n,t , given bacteria replicate at a rate r and are cleared at a rate c,

(A8) dp1(n,t)dt=n(c+r)p1(n,t)+c(n+1)p1(n+1,t)+r(n1)p1(n1,t).

Multiplying this equation by zn , and summing all equations from n=0 (with p1-1,t=0) to infinity, and defining the generating function:

(A9) g1(z,t)=n=0p1(n,t)zn,
(A10) g1(z,t)t=(1z)(crz)g1(z,t)z.

Using the method of the characteristics, and with the condition that g1z,t=0=z, as by definition p1n=1,t=0=1 and p1n1,t=0=0, it can be calculated that

(A11) g1(z,t)=(rzc)exp((rc)t)+c(1z)(rzc)exp((rc)t)+r(1z)

Then, defining pn,t the probability to observe a lineage of size n at time t; starting with 1 bacterium, before the bottleneck, and gz,t the associated generating function, then p0,t=1-q+p10,t , and p(n>0,t)=qp1(n,t) , which leads to gz,t=1-q+qg1z,t , and finally

(A12) g(z,t)=1q(rc)(1z)(rzc)exp((rc)t)+r(1z)

The resulting loss probability is

(A13) ploss=g(0,t)=1qrcrcexp((rc)t)(rc)t11q(1cr)
1.6.3 Model with delay

As shown in Figure 3—figure supplement 2, we observed a delay in the start of growth in the mouse gut, during which time bacteria will continue to be lost due to flow but will not replicate. Therefore, in a model with a fixed delay, it would be the same, except with qexp-cτ instead of q.

1.7 Colonization probability via the loss: Optimal n0

Here β is the probability for the lineage of one bacterium, present in the inoculum, to be found at measurement time. It is a combination of making it alive to the cecum, and not being lost afterwards. It is the quantity to estimate.

The mean number of bacteria n0 of a given barcode in the inoculum can be controlled, and thus the question is what is the optimal n0 to use, which will give the most accurate β estimate. For instance, one can perform preliminary experiments to assess the order of magnitude of the colonization probability and then tune n0 for better accuracy.

w is the total number of different barcodes measured (the number of different bacterial barcodes in each mouse multiplied by the number of mice). We note nl the number of barcodes among them that are lost in the experiment, that is, that are not detected at the end of the experiment.

As seen in Section 1.2 at equation (4), the simplest β estimate is -lognl/w/n0 .

If in the inoculum there are an average of n0 bacteria of each barcode, Poisson distributed (see Section 1.1), the probability that nl lineages are lost is

(A14) pnl=exp-βn0nl1-exp-βn0w-nlw!w-nl!nl!
Probability of all lost and none lost

One issue is that there may be experiments with nl=0 (in which case the previous expression is not well defined) or nl=w (in which case the estimate is that β=0, i.e., no bacteria survive).

The probability to observe the loss of all barcodes is

(A15) p(all lost)=exp(βn0w)

The probability to observe no barcode loss is

(A16) p(no loss)=(1exp(βn0))w

The sum of these two probabilities is minimized for exp-βn0=1/2, i.e., βn0=log20.69.

In the case for which we do not estimate β when either all or no barcodes are lost, the expected error in the estimate can be written as

(A17) (βestimatedβtrue)2βtrue2=nl=1w1(log(nl/w)βn01)2exp(βn0)nl(1exp(βn0))wnlw!(wnl)!nl!1(1exp(βn0))wexp(βn0w)
Numerical results

As shown in Appendix 1—figure 2, the maximum probability of obtaining an estimate is for βn0=log20.69, as expected. For this value, the expected error in the estimate is relatively small. The error slowly increases for decreasing βn0 , is minimal for βn0 somewhat larger than 0.69, but then with increasing βn0 , the probability to obtain no estimate quickly increases.

Appendix 1—figure 2
Mean relative square error in the estimate of β, as a function of n0β.

The solid-colored lines are the expression (17) for w= 12 (blue), 18 (red), 24 (purple), 40 (orange), and 120 (brown). The dotted lines are the probability that no estimate is given, either because all the tags are lost, or no tag is lost. They are the sum of equations (15) and (16).

Conclusion

The value of n0 leading to the best β estimates (likely possibility of obtaining an estimate and minimized relative difference between the estimate and the true value) is such that βn0 is of the order of 1.

1.8 Colonization probability using the variance in the barcoded population sizes: Based theory

1.8.1 Simple bottleneck

Here, we assume that each bacterium has a probability β of establishing and then grows deterministically. If the growth rate is net and the total population size remains below the carrying capacity at t, bacteria grow by a factor of expnet×t after the bottleneck.

Then

(A18) ploss=1-β
(A19) var=(β(1β)2+(1β)2β)exp(2nett)=β(1β)exp(2nett)

The relative variance, which is not sensitive to the fact that different mice have different final carrying capacity, is

(A20) varrel=varβ2exp(2nett)=1βββ11β

Here varrel=1/1-ploss.

1.8.2 Bottleneck, replication rate, and loss rate

Similar to the previously described model, we assume that each bacterium has a probability q of establishing, and then grows at a rate r and is lost at a rate c, stochastically. As seen in the previous section, the corresponding generating function for the population size at time t, starting from one bacterium before the bottleneck, is

(A21) g(z,t)=1q(rc)(1z)(rzc)exp((rc)t)+r(1z)

and the resulting loss probability is

(A22) ploss=g(0,t)(rc)t11q(1cr)

The resulting variance on the population size is

(A23) var=(2gz2+gz(gz)2)z=1
(A24) var=q2e2(rc)t(1q(1cr)(2(1+cr)e(rc)t)1cr1+cr)
(A25) var(rc)t1q2e2(rc)t(2q(1cr)1cr1+cr)
(A26) varrel=var(qexp((rc)t))2=2q(1c/r)1c/r1+c/rq12q(1c/r)

Here varrel=2/(1ploss).

1.8.3 Different initial number of bacteria

In the limit where the initial number of bacteria of the different barcodes are of the same order of magnitude, and with h the number of different barcodes in one mouse, the variance on the proportions is

(A27) varp=1h-1pi-1/h2=1h-1ninj-1h2

After some calculations, we find

(A28) varp-varp01hnj,0var1m12

with varp0 the variance in proportions in the inoculum, nj,0 the total number of different barcodes in the inoculum, and var1/m12 the relative variance starting from one bacterium, which is approximately 2q1-c/r and thus expected to be approximately equal to 2/β, with β the apparent survival probability.

Then

(A29) var1m12hnj,0varp-varp0
1.8.4 Procedure

The procedure is then to estimate var1m12 for each mouse using this equation, and then average the results to obtain var1m12, and

(A30) β2var1/m12.

There is one mouse (for the acapsular in SFP) for which all the barcodes were lost. We remove this mouse from the analysis but keep the 14 others so that the bias is likely minimal.

Then twice the standard error (the standard error is the standard deviation divided by the square root of the number of mice for each condition, reflecting that the more mice, the better the average is determined) is used for the confidence interval around var1m12, which bounds are then used to define the bounds of the confidence interval for the estimate of β via the variance.

1.9 Colonization probability using the variance in the sizes of barcoded populations: Results and comparison with barcode loss

Low-complexity microbiotaOligoSpecific pathogen free
WT10-1.66-1.78:-1.5110-1.85-2.05:-1.5110-2.42-2.62:-2.02
Acapsular10-1.90-2.07:-1.6310-1.61-1.72:-1.4710-3.85-3.99:-3.66

Appendix 1—figure 3 compares the loss and the variance method. Overall:

  • Both methods give the same orders of magnitude.

  • Both methods show that in this dataset, there is no significant difference for all the experiments in LCM and oligo mice; while WT in SPF mice has a tighter bottleneck (though not significant in the variance method), and acapsular in SPF mice even more.

  • Overall, the variance method results in smaller β estimates. The variance method is more sensitive to the model used, and approximations made in the minimal model may explain this systematic difference.

Appendix 1—figure 3
Comparison between the lineage survival probability estimated via the loss method and via the variance method.

The colors represent experiments: low-complexity microbiota (LCM) WT (red), Oligo WT (green), specific pathogen free (SPF) WT (blue), LCM acapsular (pink), Oligo acapsular (light green), and SPF acapsular (light blue). Circles represent best estimation and bars the error. The error bars are given via the direct study of the estimation probability for the loss method via the standard error between mice for the variance method.

Note that if there is a fixed delay τ before growing, then the equations are the same, except for q which is replaced by qexp-cτ . Then the relation between the β estimate via the loss or the variance remains the same, and β is an apparent survival probability, considering delay additionally to the initial bottleneck.

1.10 Experimental noise

There are incertitudes on measurements:

  • To estimate overall bacterial concentration in the solution from which the inoculum is taken from, a number of colonies n is counted, and so we expect a typical relative error in the concentration of 1/n .

  • The relative proportion of the different barcoded strains are analyzed by qPCR. In an experiment, three replicates were measured for each of the barcoded strains, and the standard deviation in the counts was 0.22. Note that 20.22100.066 , thus if the incertitude in the counts is ±0.22, then if the number/concentration estimated using the qPCR is expressed as a power of 10, the incertitude will be ±0.066 in the exponent.

For the loss method, the end point measurement is the absence/presence of a barcoded strain, which is not sensitive to the qPCR incertitudes (at least at the small level of qPCR noise of the experiments). However, the calculations actually estimate βni , thus errors in the estimate of ni , the initial expected number of bacteria for each barcoded strain, will affect the estimate of β.

For instance, the typical number of colonies counted for checking the initial concentration from which n0 is calculated is of the order of 40, resulting in a typical relative error of 1/40 , that is, about 15%. Then overestimating n0 by 15% will result in underestimating β by about 15%. The relative error is not biased, it will just increase the incertitude. ±15% will corresponds to approximately ±0.06 in the log10β . This is if all data was from the same inoculum. Actually, for each condition, the data is pooled from different experiments, with different starting inoculums, with uncorrelated incertitudes on the initial number of barcoded bacteria (2–5 different starting inoculums depending on the condition). Then this incertitude is smaller than the incertitude as calculated previously. As a consequence, while it would somewhat increase the confidence interval, taking into account this source of incertitude has a small impact, and it is not included in our main results for simplicity.

The impact of the qPCR incertitude on the loss method is smaller. Indeed, the incertitude due to qPCR counts for one barcoded strain is of the same order of magnitude as the incertitude on the total n0 discussed in previous paragraph; but then for each inoculum there are six barcoded strains, and the total number of barcoded bacteria is fixed, thus the overestimates and underestimates will almost compensate, and the resulting incertitude will be small compared to the incertitude on the total n0 .

For the variance method, calculations show that the expressions can be modified to consider the incertitude on n0 , and with σ the standard deviation in noise for the number of counts,

(A31) varpvarp0+var1m121h2n0+1h2n0+2log(2)2σ2h2.

With the experimental values, this correction is very small and thus is not included in the main results for simplicity.

1.11 No tag bias in the colonization results

Bacteria with different tags may grow at different rates and this may bias the results on colonization probabilities. We checked that in vitro, the growth rates between the different barcoded bacteria are not significantly different. In this section, we show that if we were analyzing only a subset of the barcoded bacteria, the results on the colonization probabilities would be very similar, showing the robustness of the results.

In fact, in the case of the loss method, as the readout (absent or present) is sensitive only to the early dynamics, and not later growth (as soon as the number of bacteria becomes large enough), we do not expect any effect of tag fitness.

Indeed, as shown in Figure 2—figure supplement 5B, while the results using only one tag are noisier, they are in line with the estimate using all the data, and there is no sign of a tag having a consistent bias.

Several tags for each mouse are necessary for the variance method. Removing a tag with a substantially different growth rate than the others would decrease the variance, and thus increase the β estimate. In Figure 2—figure supplement 5C, removing one barcode has very little effect on the estimates.

2. Growth curves

2.1 Net growth rate

If the bacteria grow at rate r and are cleared at rate c, the slope of the growth curve, with the log of the bacterial concentration represented as function of time, will be the net growth rate net=r-c. We remove the first points (which may also contain bacteria from the inoculum that are passively carried and had not really settled in the cecum); and the last points, at which the bacterial concentration is close to carrying capacity, and thus the population is not growing any more.

2.2 Linear growth fits

See Figure 3—figure supplement 2 for the fits for either WT or acapsular, growing in mice with OligoMM12 microbiota, and Figure 3C for the values. We find a significant difference (p<0.01), at 0.50/hr (WT) vs. 0.40/hr (acapsular).

In the data of Figure 2—figure supplement 2 for the WT in Oligo and LCM mice, there were no significant differences between the growth rates in these microbiotas, which was expected. A linear fit on WT bacteria in three SPF mice gives a net growth rate of 0.26, 0.44, and 0.40/hr, that is, and average of 0.37 ± 0.09/hr. The net growth rate lower by 0.13/hr is, with the error bars, actually consistent with a higher turnover in SPF mice, 0.23/hr instead of 0.13/hr.

2.3 Delay to start exponential growth: Base theory

As observed in the growth curves in Figure 3—figure supplement 2, there seem to be a delay in growth. To quantify this, we analyzed the growth curves as shown in Figure 3—figure supplement 2A.

With a survival probability q, a delay τ with clearance rate c, and then regrowth at rate net, the equation for the part of the curve with growth is n0qexp-cτexpnett-τ . For the colonization, qapp=qexp-cτ1-c/r . Then the equation for the curve is n0qappexpnett-τ/1-c/r . When fitting the curve of the growth part, the intercept I (=concentration value when t=0) is then n0qappexpnet-τ/1-c/r . Then τ can be estimated as

(A32) τ=log((n0qapp)/(I(1c/r)))/net

I and net comes from the fit, n0 is the inoculum size, qapp can be taken from the colonization experiments.

2.4 Delay to start exponential growth: Bacterial concentration in feces vs. absolute numbers in cecum

Note that the reasoning of the previous subsection is for the absolute number of bacteria in the cecum, whereas the growth curves are obtained from bacterial concentration in feces. This is not an issue if we only measure net, but to estimate the delay, we need to convert n0 of the inoculum in effective c0 for the feces. To do so, we need both the cecum mass, and the feces to cecum concentration factor.

If the bacterial concentration in the feces c is f times higher than in the cecum (mostly due to water absorption), and that the cecum has mass m, then the absolute number in the cecum n is such that c=f×n/m. Then if we look at the time course in the feces concentration, to convert the inoculum absolute n0 in equivalent concentration in feces, c0=f×n0/m. Thus, we need to determine m and f.

In experiments where the total cecum mass in oligo mice was measured (Hoces et al., 2022), the mean mass was 1.5±0.2g. Note that an error of 50% in the mass for a net growth rate of 0.5/hr would be around log1.5/0.50.8h.

Now let us estimate f. We expect a higher concentration in feces relative to the cecum due to water absorption. However, in experimental data, which is noisy, we find that the average ratio of concentration of feces relative to the cecum very close to 1.

2.5 Fitting the delay

Process

c0 is the effective n0 multiplied by f/m0.7. qapp is taken as 10-1.5 from the colonization experiment. We estimate with the expression (32), taking c=0.13 for the WT, and c=0.23 for the acapsular (considering that net is smaller for the acapsular, likely because of a higher clearance rate).

Results

See Figure 3—figure supplement 2 for looking at all the fits, and Figure 3D for the resulting delays. We find a significant difference between WT and acapsular, with an average delay of 3.2 hr for the WT and 7.7 hr for the acapsular (p=0.02).

Note that the delay for the WT could come just from the fact that it takes time for bacteria to physically go from the oesophagus to the ceum. The delay for the acapsular could come from a combination of such a transport delay, +a delay to resume growth.

These 4.5 hr of delay in acapsular relative to the WT is at the limit to keep coherence with the colonization probabilities. With such a delay, and for the higher loss rate of acapsular, that we estimate at 0.23/h, then the expected difference in apparent colonization probability (assuming the same initial survival probability) is 100.5 , which is almost the double than what is coherent with the values found and their incertitudes. Note that several factors could influence this result: if we underestimate f, we overestimate the relative delay between WT and acapsular. If we overestimate the clearance difference between WT and acapsular, we also overestimate the relative delay between WT and acapsular.

2.6 Growth dynamics with competition

We propose here to fit the growth data of an experiment in which WT and acapsular were given in a 1:1 ratio.

As the WT is quite more concentrated in the feces relative to the acapsular when approaching carrying capacity, it is fair to assume its growth is independent of the acapsular one, and use logistic growth,

(A33) dndt=r1-nKmaxnt-cnt

with Kmax the maximum carrying capacity, r the maximal growth rate, and c the loss rate. Then, after some calculations, and with K=Kmax1-c/r the effective carrying capacity, and rnet=r-c the effective growth rate,

(A34) n(t)=K1+(Kn01)exp(rnett).

For the acapsular, the dynamics for the acapsular, as acapsular and WT are likely to be limited by the same nutrients:

(A35) dAdt=ra1-A+WKmaxA-caA.

In this experiment, either (A+W)/Kmax1, or AW, thus

(A36) dAdtra1-WKmaxA-caA,

which with equation (A34), leads to

(A37) A(t)=a0exp(netat)(Kw0exp(netwt)1+Kw0)rr

with K=Kmax1-cw/rw , a0 the initial concentration in acapsular, w0 the initial concentration in wild type, neta=ra-ca the net maximal growth rate of the acapsular, and rr=ra/rw the ratio of the maximal growth rates.

Figure 3—figure supplement 2E shows the resulting fits of competitive colonization starting with WT and acapsular strains at a 1:1 ratio. Because of the start at a higher concentration than in Figure 3—figure supplement 2B and C, the effect of the carrying capacity is felt early and the fit underestimates the net growth rate in the beginning, but there is clearly a lower net growth rate for the acapsular relative to the WT. The second part of the dynamics show a decrease of the acapsular relative to the WT. Fitting the curve gives an estimate of rr=ra/rw , but because there are only a few points in this second part, the estimates are noisy. They are found to be between 0.5 and 1.1, compatible with the expected value of 1.

3. Competition

3.1 General dynamics

Appendix 1—figure 4 depicts the general dynamics of the competition. The WT and acapsular are thought to interact only through competition for food. After a first bottleneck of survival probability qi for each initial bacteria, their dynamics consists of

  • Fixed loss rate of ci (with i standing for either WT or acapsular; and the loss rate at least equal to the cecum turnover rate)

  • With a maximal shared carrying capacity Kmax , a growth rate of ri1-W+A/Kmax , with ri the maximal growth rate, W the WT abundance, and A the acapsular abundance.

Appendix 1—figure 4
Schematic of the general dynamics in the minimal model.

The WT (black) and acapsular (blue) population are in competition. After a first bottleneck (at t=0), as the acapsular has a larger loss rate, its net growth rate is smaller. As the bacteria only interact through food, they follow their own dynamics until carrying capacity is reached, then WT is with a null net growth rate, and the acapsular, with its higher loss rate, has a negative net growth rate (we usually assume the absolute growth rate are equal, i.e., rr=ra/rw=1). A barcoded population may be lost at the first bottleneck; or in the initial dynamics at low numbers; or towards the end of the experiment when the acapsular population decreases.

The effective carrying capacity for the WT is K=Kmax1-cw/rw .

Early on, there are few WT and acapsular bacteria, so they grow at their maximal rate. We will approximate the more realistic logistic growth by an exponential growth until reaching the effective carrying capacity. The overall dynamics is very similar to the logistic growth, while being much easier to handle analytically. Then in the experiments we analyze, given that WT and acapsular start in similar abundances, and that the acapsular has a smaller net growth rate, the WT is much more abundant than the acapsular when the effective carrying capacity is reached. Thus, the time to reach carrying capacity is approximately tw , such that K=n0wqwexprw-cwtw . After tw , the net growth rate of the WT is zero, as growth compensate loss, and because WA, W+AWK, thus the net growth rate of the acapsular is

(A38) neta=ra(1K/Kmax)ca=racw/rw+netara=netara(1cw/rw)=netarrnetw.

If rr=ra/rw is close to 1, and because neta=raca<rwcw=netw , neta` is predicted to be smaller than 0, and this is in agreement with experimental data showing that the acapsular abundance in competition decreases once the WT has hit carrying capacity.

The generating function for the size of a lineage starting from for an acapsular bacterium is 1-qa+qagupgplatz with gup the generating function for the exponential growth when the population size goes up

(A39) gup(z)=1neta(1z)exp(netatw)razca+(1z)raexp(netatw),

and gplat is the generating function for the acapsular for the phase with a plateau in the WT population size:

(A40) gplat(z)=1neta(1z)exp(neta(ttottw))racwrwzca+(1z)racwrwexp(neta(ttottw))

and thus, the overall survival probability is qa1-gupgplat0 .

When bacteria are not in competition, they reach a large carrying capacity, so later loss is negligible, resulting in a total survival probability for one bacterium of

(A41) qi,app=qi1-ci/ri

3.2 Relative ratio

The relative ratio is nf,WTn0,a/n0,WTnf,a , with f for the final population size, 0 for the number of bacteria in the inoculum, WT for the wild type, and a for the acapsular. We start by its general expression; then show that for rr=ra/rw=1, the expression can be simplified and expressed as a function of the parameters measured in other experiments, thus allowing for a prediction; and in the case of rr1 we discuss different expressions.

The expected relative ratio between the WT and the acapsular is

(A42) R=qwexp(netwtw)qaexp(netatw)exp(neta(ttottw))=Kn0wqaexp(netatw)exp(neta(ttottw))
(A43) R=Kn0wqaexp-netatw-net`attot-tw

Using the expression for net`a from (38),

(A44) R=Kn0wqaexp-rrnetwtw-neta-rrnetwttot

With n0w the initial number of WT bacteria, n0wqwexptwnetw=K, leading to

(A45) R=qwrrqaKn0w1-rrexprrnetw-netattot

For the Oligo and LCM microbiota, the colonization experiments show that the colonization probability is very similar for the WT and the acapsular, so that it can be assumed that qa=qw in these cases. There is no reason a priori for this to be different for GF mice, so this assumption also extends to GF mice. For SPF mice, the apparent colonization probability was quite different, so let us define qq=qa/qw . For the colonization process, as the B. theta population grows to large sizes, qi,1,app the apparent colonization probability of type i (WT or acapsular) in the colonization process (1 stands for colonization with a single strain type), qi,1,appqi1-ci/ri . Then

(A46) qqSPF=qa,1,apprwnetaqw,1,appranetw

When rr=ra/rw=1:

Given that the WT and the acapsular grow at the same rate in vitro, the null assumption is to take them also growing at the same rate in vivo, but with different clearance rates, leading to different net growth rates. With this assumption,

(A47) Rrr=1=qwqaexpnetw-netattot=1qqexpnetw-netattot

=qa,1,appnetaqw,1,appnetw for GF, Oligo and LCM, and is for SPF.

This prediction mixes results from the growth experiments (netw and neta), and results from the colonization experiment (qq=1 in most cases, qa,1,app and qw,1,app to calculate qq in the SPF case).

When rr1

Predicted relative ratio

Reformulating (45):

(A48) R=qwrr-1qqKn0w1-rrexprrnetw-netattot.

For GF, Oligo, and LCM mice, qq is taken as = 1, and for SPF, expression (46) is used; for qw=qw,1,app/1-cw/rw=qw,1,appnetw+cw/netw . In summary, qq and qw are taken from the colonization experiments, netw and neta from the growth experiments, n0w and ttot are controlled experimental parameters, and K is experimentally measured. As explained earlier, there are indication that rr should be = 1, but in the confidence interval calculations we explore variations around 1.

Relative ratio given the loss

Reformulating (45):

(A49) R=qarr-1qqrrKn0w1-rrexprrnetw-netattot

As before, qq is taken as 1, except for SPF mice, for which expression (46) is used. The difference with the previous expression is that now qa is taken from the competition experiment. There are two expressions linking qa to the apparent survival probability qa,app .

  • In the limit when the population of acapsular remains high enough at the end of the experiment despite competition with the WT, then the same approximation as in the colonization experiments can be made, and qaqa,app/1-ca/ra=qa,apprrnetw+cw/neta . Note that here one additional assumption is made, on the value of cw . It is generally taken as the cecum turnover rate, which is its minimal value, but higher values are explored for the confidence intervals.

  • The full expression is qa=qa,app/1-gupgplat0

3.3 Predicted survival probability

The expression for the survival probability is qa1-gupgplat0

The estimate for qa is taken from the colonization experiment, with the approximation qa,1,app=qa1-ca/ra , thus

(A50) psurv=qa,1,app(1gup(gplat(0)))1ca/ra.

As seen earlier, gup depends on neta , tw , ra , ca . Also, gplat depends on net`a , ttot , tw , racw/rw=rrcw , ca . We note that neta`=neta-rrnetw , that ra=rrrw=rrnetw+cw and that ca=ra-neta=rrnetw+cw-neta . As n0wqwexptwnetw=K, and qwqw,1,app/1-cw/rw=qw,1,appnetw+cw/netw , tw=logKnetw/n0wqw,1,appnetw+cw/netw .

Overall, the survival probability ends up being dependent on qa,1,app , qw,1,app , (both measured in the colonization experiment), neta , netw (both measured in the growth experiment), ttot , n0w (both known controlled parameters of the experiment), K (measured), cw (in general taken as its lower bound, the cecum turn over), and rr (generally taken as 1).

3.4 Effect of a fixed delay in growth

If there was a delay τi for strain i before starting growth after ingestion, then

(A51) qi,1,app=qiexp-ciτi1-ci/ri,

K=n0,wqwexp-cwτwexpnetwtw-τw (note that here tw is the time at which carrying capacity is reached, with duration τw of population decrease, and tw-τw of net growth netw of the wildtype), thus

(A52) exp(netw(twτw))=Kexp(cwτw)n0,wqw=Knetwn0,wqw,1,app(netw+cw)

As in the model with no delay, we aim to predict the relative ratio and the survival probability, with parameter values taken from other experiments.

The expected relative ratio between the WT and the acapsular is

(A53) R=qwexp(cwτw)exp(netw(twτw))qaexp(caτa)exp(neta(twτa))exp(neta(ttottw))

As neta`=neta-rrnetw , replacing expnetwtw-τw by previous expression, and after some calculations,

(A54) R=1qqKexpcwτwn0wqwexpnetwτw1-rrexpnetw+cwrrτa-τwexprrnetw-netattot

If rr=1, then

(A55) R=1qqexpnetw+cwτa-τwexpnetw-netattot

As before, the assumption is qq=1, except for the SPF, when

qq=qa,1,app1-cw/rwexp-cwτw/qw,1,app1-ca/raexp-caτa

If rr different from 1, in the case of a prediction from the rest of the experiments,

(A56) R=1qqKexpcwτwn0wqwexpnetwτw1-rrexpnetw+cwrrτa-τwexprrnetw-netattot
(A57) R=1qqKnetwn0wqw,1,appnetw+cwexpnetwτw1-rrexpnetw+cwrrτa-τwexprrnetw-netattot

If rr is different from 1, in the case of checking for internal consistency of the competition experiment,

(A58) R=1qqrrKexpcwτwn0wqaexpnetwτw1-rrexpnetw+cwrrτa-τwexprrnetw-netattot

with qa is such that qa,app=qaexp(caτa)(1gup(gplat(0))) (note that gup is for tup=tw-τw , and gplat for tplat=ttot-tw).

And the expected survival probability is qaexp-caτa1-gupgplat0 , with qaqa,1,appexpcaτa/1-ca/ra , and thus

(A59) surv=qa,1,app1-gupgplat0/1-ca/ra

3.5 List of parameters and values

SymbolMeaning (more info)How determined/values taken (germ-free [GF], Oligo, low-complexity microbiota [LCM], specific pathogen free [SPF])
qqqa/qw(1)1, 1, 1, Depends on model
n0wNumber of WT bacteria in the inoculum (2)Controlled and measured
107.63±0.1 , 107.54±0.17 , 107.31±0.49 , 107.56±0.20 , , ,
KEffective carrying capacity (3)Measured
1010.86±0.23 , 1010.17±0.47 , 109.49±0.33 , 107.35±0.53 , , ,
rrRatio of the growth rates, ra/rw (4)1 [0.5–1.2] (for all microbiota)
netwWT net growth rate, rw-cw (5)Measured in the growth experiments in Oligo and SPF
0.50/h[0.350.66] , 0.50/h[0.450.56] , 0.50/h[0.450.56] , 0.37/h[0.260.44] , , ,
netaAcapsular net growth rate, ra-ca (5)Measured in the growth experiments in Oligo
0.45/h[0.300.50] , 0.40/h[0.350.44] , 0.40/h[0.350.44] , 0.27/h[0.060.44] , , ,
qa,appApparent survival probability,Measured in the competition experiments
acapsular in competition (6)101.27[1.35,1.19],103.4[3.47,3.33],102.63[2.75,2.51],106.59[6.64,6.54]
qa,1,appApparent survival probability (7)Measured in the colonization experiments (except GF)
when only acapsular100.7[1.3,0.1],101.48[1.65,1.31],101.42[1.58,1.26],103.60[3.77,3.43]
qw,1,appApparent survival probability (7)Measured in the colonization experiments (except GF)
when only WT100.7[1.3,0.1],101.54[1.71,1.37],101.49[1.64,1.34],102.34[2.51,2.17]
totTotal experimental time (8)Fixed at 44 hr (taken 44 hr [40,48])
cwLoss rate of the WT (9)Minimum is the cecum turnover rate
0.13/h[0.130.23] , 0.13/h0.13-0.23 , 0.13/h[0.130.23] , 0.23/h[0.230.33] , , ,
mwMean growth delay for the WT (10)3.2 hr [0–5.4]
Delay model only
maMean growth delay
for the acapsular (10)
7.7 hr [4.0–11.6]
Delay model only
rwGrowth rate of the WTrw=netw+cw
raGrowth rate of the acapsularra=rr(netw+cw)
caLoss rate of the acapsularca=raneta=rr(netw+cw)neta
qiSurvival probability initial bottleneck i (w or a) (11)Estimated using qi,1,app or qa,app
twTime for the WT to reach carrying capacityEstimated from the other parameters
Detailed notes
  1. As the colonization experiment gives very similar apparent probability for WT and acapsular, except for SPF, the assumption is that qq=1 for all except SPF. For SPF, qq=qa/qw , and in the simple model, taking the approximation that in the colonization experiments, qapp=q1-c/r , then qq=qa,app1-cw/rw/qw,app1-ca/ra , which can also be written as =qa,1,appnetwrr/qw,1,appneta .

  2. Measured in the inoculum, averaged over the inoculum used for the given microbiota, ± the standard deviation between values for the different used inoculum (average and sd calculated on the log values). For GF, only one inoculum is used, so there is no standard deviation, the error is estimated ±0.1 of the log10 of the concentration.

  3. Actually, what is measured is the number of bacteria per g of cecum content, whereas K is the absolute value. The assumption is that the cecum is about 1 g. In reality, it is often a bit smaller, but the difference is small compared to the differences in observed final bacterial concentrations.

  4. The rationale is that in vitro, WT and acapsular have the same growth rate, so there is no specific reason to believe they are different. So, this ratio is usually taken as 1. This ratio could in principle be determined from the last part of the dynamics of the acapsular in competition with the WT in the growth experiment. Though there are only a few data points, 1 is in the range of fit values, and the minimal and maximal values fit values are used for the confidence interval.

  5. We estimated the net growth rate for the acapsular strain in Oligo: neta=0.40/h0.35-0.44 (± standard deviation). For the WT strain in Oligo, netw=0.50/h0.45-0.56 , and in SPF 0.37/h (0.26–0.44, here the lowest and higher fit values). In the experiment shown in Figure 3—figure supplement 2, the growth rate of bacteria in Oligo and LCM was similar, so the same values are taken for Oligo and LCM for both acapsular and WT strains. GF mice seem closer in terms of turnover to the LCM and Oligo mice, so the same growth rate is taken for GF for WT strains, increasing the window of uncertainty by ±0.1/hr to consider the data limitation. For the acapsular strain, we assume less clearance based on the lack of competing microbiota, therefore we used an average neta=0.45. For SPF, it is observed for the WT that the net growth rate is decreased by 0.13/hr, which is consistent with a higher turnover in the cecum of SPF mice (higher by 0.1/hr compared to GF mice), which points towards an increase in c rather than a decrease in r. For neta in SPF, in the absence of direct data, the assumption is thus to take the same decrease relative to the Oligo mice, with a higher uncertainty range to consider the absence of direct data. Note that with this the lowest boundary for the net growth rate of acapsular in SPF is small, and this is consistent with the observations that in some experiments, acapsular in SPF barely grows.

  6. Measured from the barcode loss in the competition experiment.

  7. Measured in the colonization experiments, the value taken is the estimate via the barcode loss. For GF, a lower bound is the apparent survival probability in the competition experiment (10-1.2710-1.3), as it is smaller than in the colonization experiments. An upper bound is considering that in the simple model, subsequent survival is at most equal to 1-c/r=net/net+c , and with a net growth rate of the order of 0.53/hr (WT) and 0.40/hr(acapsular) (both measured in Oligo but expected to be similar in GF), and the c at least equal to 0.13/hr, the survival is at most 10-0.1 . The main value is taken as the middle (in log scale) between these two boundaries.

  8. The initial survival probability may represent very early processes, in the stomach and small intestine, so the time spent in the cecum is actually smaller than 48 hr (the total time between inoculation and the end of the experiment), so it is why in exploring the parameter space the choice is made to take 40–48 hr as the confidence interval.

  9. The clearance rate is at least equal to the cecum turnover rate. It is 0.13/hr for the GF, and 0.23/hr for the SPF. LCM and Oligo are thought to be closer to the GF. It would be higher in the presence of killing. Without evidence of killing, the main value is this minimal value. The upper bound is taken as this value, +0.1/hr to represent the impact of potential killing.

  10. The delay model explores the possibility that the acapsular takes a longer time to recover and grow back. In the growth curves, all are consistent with exponential growth from 12 hr onward, with points beforehand that suggest some delay, explored in Section 2.5.

  11. In the colonization experiments, the bacterial lineages reach and remain at a large population size, so later loss is negligible, thus qi,appqi1-ci/ri for the simpler model. See text for other models.

3.6 Computation

The code used for computation is uploaded in the code repository. The results are first computed for the main values of the parameters, for all the models. Then, for a chosen number of iterations (in general 1000), a new set of parameters is taken at random. For each parameter, with probability 0.5 it is chosen randomly uniformly between the lower bound and the main value; and with probability 0.5, it is chosen uniformly between the main value and the upper bound of the confidence interval. It is checked that these parameters combined with qi,1,app lead to qa and qw below 1. For the simple model, if the resulting qi>1, all the parameters are chosen again at random. For the other models, first, for at most 300 iterations, only the parameters linked to the delay are changed, and then if the result gives still survival greater than 1, a full set of parameters is again chosen randomly. This procedure helps avoid bias in the choice of the parameters value.

The results are computed for each of these sets of parameters, and the confidence interval is given as the mean ± the standard deviation on all these iterations.

4. Acute challenges

4.1 Estimate of n0

We need to estimate the number of barcoded bacteria in the cecum at the start of the challenge from the intermediate data collected in feces, which are a subsampling of the cecum, with potential bias.

4.1.1 Cecum mass

A first question is the mass of the cecum as the feces gives only access to a concentration. For all cases, the conditions are like the control conditions up to day 0, thus reasonably at day 0 the cecum is expected to weight like the cecum in the control case at day 3 at the end of the experiment. The average mass of the cecum in GF mice is 2.95 g, standard deviation 0.62 g (Hoces et al., 2022).

4.1.2 Concentration in feces vs. cecum

At day 3, the data consists of both feces and cecum, thus concentrations can be compared; see Appendix 1—figure 5. Using only the data from the untagged strain (less bias because no missing points, but overall there is no large difference between dashed and dotted lines), feces are 100.67+-0.23 more concentrated than the cecum for the control (± represent twice the standard deviation for a 95% confidence interval), feces are slightly less concentrated than the cecum for HFD (high-fat diet challenge), and feces are less concentrated than the cecum in Stm (Salmonella challenge).

Appendix 1—figure 5
Concentration of B. theta in the cecum and the feces.

Concentration of untagged (triangle) and barcoded (cross) on day 3 after acute challenge (see color legend for each challenge). The dashed lines represent the average ratio considering only the untagged strain data, whereas the dotted lines represent the average of both untagged and barcoded strains (except the case of total loss of barcoded).

4.1.3 Factor to convert concentration in feces at day 0 to absolute number in cecum at day 0

To estimate n0 , the total number of barcoded bacteria in a mouse at day 0 is taken as the number of erythromycin-resistant colonies in feces, divided by the weight of the sample to obtain the concentration, then multiplied by the mean cecum mass. Finally, the concentration factor for the control case is applied. Then for a given tag, the counts are used to get the proportion of bacteria with that tag in the total number of barcoded bacteria.

4.1.4 Tags unseen at day 0

In most cases, tags are both seen in feces at day 0, and in cecum at day 3.

In a fraction of cases, tags are seen in feces at day 0, but not in cecum at day 3. In this case, as the sample is a significant portion of the cecum, it is unlikely that the tag is actually still present, so here we count this situation as a loss.

In the counts of barcoded bacteria, there are a few cases where some tags are not seen at day 0, but still are seen at day 3, due to the feces sample weighting in average less than 10% of the cecum and these barcoded bacteria being in very low numbers.

This also points to the possibility that in the cases where the tags are not seen neither at day 0 nor at day 3, they may have been actually present at day 0 and lost before day 3.

For the loss, we will analyze the data:

  • Excluding all cases in which the tags were not seen at day 0.

  • Excluding all cases in which the tags were not seen neither at day 0 nor day 3, and for tags not seen at day 0 and seen at day 3, n0 for day 0 is taken as the average estimated total number of barcoded bacteria in that mouse at day 0, divided by 6 (total number of tags).

  • For all cases with no tags seen at day 0, n0 for day 0 is taken as the average estimated total number of barcoded subpopulations in that mouse at day 0, divided by 6 (total number of tags).

In practice, the differences in the results are negligible.

4.2 Method of the loss

4.2.1 Survival probability of a bacterial lineage

We use the same method as in previous section to estimate the colonization probability, except that n0 is estimated from the feces at day 0 instead of the inoculum.

There was no loss for neither the control nor HFD (there is one loss in HFD, but it is also a loss at day = 0, for a large total number of barcoded bacteria in the feces at day 0 in that mouse, so it is likely that this tag was actually lost at day 0). Then to obtain a lower bound for the survival probability, the case with the smallest n0 is taken as lost and the survival probability is calculated from it.

4.2.2 Results
  • There is no loss for the control case, so the best estimate of β is 1, with lower bound 10-1.5 .

  • There is no loss for the HFD case, so the best estimate of β is 1, with lower bound 10-1.75 .

  • For Stm, β=10-3.28±0.10 (for the incertitude: for the log10 of the concentration, the standard error is 0.21, but there are 14 different mice for Stm, thus 0.06 overall expected on the combined data if the ratio of the concentrations are independent in the different mice; for the log10 of the survival probability using a Bayesian approach, the standard deviation is found to be around 0.08; so overall the standard deviation is expected to be around 0.10, then taking 2 standard deviations we find the result).

  • For the avirulent Stm, β=10-2.35±0.3 (similar reasoning)

Thus attenuated Stm imposes a relatively strong bottleneck on B. theta. The bottleneck from the avirulent Salmonella is less stringent. HFD does not have a strong enough effect to be detected in the experimental conditions.

4.2.3 Interpretation

This survival probability gives the probability that a bacteria present at day = 0 has still a lineage in the cecum at day 3.

It could either come from a temporary bottleneck or from constant loss during the challenge. If there is a constant turnover rate c, then the survival after a time t is 1/1+ct . For c in GF mice of 0.13/hr, and for 3 × 24 hr, we find a survival of about 0.1 (10-1), which is compatible with the results for the Control and HFD.

HFD could be causing a larger c, but as no loss is observed, we cannot quantify it.

For Stm, the decrease in total population size is at most a factor of 10, whereas survival probability is well below 0.1, pointing to a more stringent bottleneck followed by re-growth (and possibly additional continuous loss).

Data availability

Relevant numerical source data for figures and figure supplements has been provided. Raw sequencing data accessed on ENA (https://www.ebi.ac.uk/ena/browser/home) under Project ID PRJEB57876 and PRJEB53981. Raw data and code used for generating all figures in this publication are made available in a curated data archive at ETH Zurich (https://www.research-collection.ethz.ch/) under the https://doi.org/10.3929/ethz-b-000557179.

The following data sets were generated
    1. Slack E
    (2022) European Nucleotide Archive
    ID PRJEB57876. Population dynamics of Bacteroides thetaiotaomicron during the early colonization of the murine gut.
    1. Burga H
    2. Alexander D
    (2022) ETH Zurich Research Collection
    Fitness advantage of Bacteroides thetaiotaomicron capsular polysaccharide is dependent on the resident microbiota.
    https://doi.org/10.3929/ethz-b-000557179
The following previously published data sets were used
    1. ETH ZURICH
    (2022) European Nucleotide Archive
    ID PRJEB53981. Metabolic reconstitution by a gnotobiotic microbiota varies over the circadian cycle.

References

  1. Book
    1. Bushnell B
    (2014)
    BBMap: A Fast, Accurate
    Splice-Aware Aligner.

Decision letter

  1. Matthew K Waldor
    Reviewing Editor; Brigham and Women's Hospital, United States
  2. Wendy S Garrett
    Senior Editor; Harvard T.H. Chan School of Public Health, United States

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

[Editors’ note: the authors submitted for reconsideration following the decision after peer review. What follows is the decision letter after the first round of review.]

Thank you for submitting the paper "Fitness advantage of Bacteroides thetaiotaomicron capsular polysaccharide is dependent on the resident microbiota" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by a Senior Editor. The reviewers have opted to remain anonymous.

Comments to the Authors:

We are sorry to say that, after consultation with the reviewers, we have decided that this work will not be considered further for publication by eLife given the extensive amount of experiments and data that would be required to address the major concerns. In general, eLife only allows for one revision round with a target of no more than 8 weeks for resubmission. We hope you find the comments and suggestions helpful.

Reviewer #1 (Recommendations for the authors):

a. The generation of gnotobiotic mice is not described.

b. Figure 1D cartoon, greys should be 1/7th of population? Better description of inoculum including non-WITS population would be helpful throughout.

c. Figure 1F, the type of average should be expressed in the legend. Similar statements should be made in other figures as well.

d. Figure 1F, Ery appears to have an advantage.

e. Figure 2A cartoon over-simplifies the biology, and suggests all death occurs prior to reaching the site of colonization, when in fact clones can be eliminated in the cecum or flow through without establishing colonization.

f. Figure 3 clearance is misspelled.

g. Figure 4b the challenges appear to independently alter populations in the cecum and feces, suggesting that the feces is not a reliable indicator of cecal populations.

Reviewer #2 (Recommendations for the authors):

Line 61 – The use of the term "gnotobiotic" is incorrect, as it does not refer to any specific level of diversity. Gnotobiotics just means that the composition is known, which can be 0, 1 or 100s of microbes.

Need to be careful not to use "CFU" to refer to qPCR data, which measures DNA levels not viable cells. A common technical term is "genome equivalents".

The growth curve format and analysis was unclear. For example, Figure 1b would be aided by overlapping all the lines on the same graph, removing the log scaling on the y axis, and quantifying carrying capacity and lag phase parameters. Error bars are also missing/not visible.

Figure 1e – I cannot see the error bars, only 1 dot/tag is visible.

The correct abbreviation for overnight is "O/N" not "ON". Better yet, just spell out the words to avoid confusion.

Line 112 – this seems like a big assumption – how do we know selection can't occur in 48 hours?

Line 117 – waiting to grow in the cecum also seems like a big assumption to me.

Figure 2c – this panel is really unclear, need to indicate which tags are which by color.

Figures 2d-g – it would help to include the wt and mutant strains on the same graph, to enable comparisons between strains and appropriate statistics.

Figure 2b – multiple typos: "extintino", "optimun"

Figure 3e,f – these curves raised a lot of questions. The first timepoint has high levels, especially for panel f. A time zero needs to be included to rule out background amplification. Given the 3 OOM difference between the two panels in starting levels I'm not sure it is fair to compare the curves. The n is very low here, need more mice and a replicate experiment. Controls should be included with the individual members of the OligoMM community to ensure that there is no background amplification.

Line 257 – this is interesting but seems very speculative. It would really help to formulate a testable hypothesis.

Line 260 – this section did not follow from the rest of the paper and might be better to move it to a follow-up study. The experiments are also very preliminary and conflict with prior literature. For example, the diet experiment would need to be repeated with a matched control diet, 16S profiling to ensure that the rest of the gut microbiota is altered, bile acid quantification, and a longer time series (3 days is quite short).

Figure 4b – the difference in the control group is worrisome, although perhaps that's due to comparing cecal contents to feces. It would probably be safer to just compare the same sample type across groups.

Line 314 – this conclusion is not well supported given all the potential host differences.

Reviewer #3 (Recommendations for the authors):

My comments below are mainly focused towards the interpretation of their data, in particular the spatial localization question I bring up below.

– Line 97-100: they state that the antibiotic cassette isn't really altering fitness, which I found surprising – there is generally some fitness cost to antibiotic cassettes, at least in in vitro experiments. Upon examination of Figure 1F, it actually looks like for the wild-type, the Ery cassette wins out in competition. This should at least be tested statistically, as a half-log-fold after 2 days is fairly important biologically.

– Looks like there are consistent biases for certain barcodes in Figure 1E? e.g., pink/purple is consistently lowest in WT, albeit not acapsular. Again, a half-log-fold after 2 days is not small here and would accumulate over time. Can they rule out fitness differences in any other way? I realize that they test for mutation in selection by reinoculating isolated strains, but this does not test for pre-existing mutation.

– Line 171-174: they conclude that size of open niche does not translate linearly into colonization probability, but this seems like a given, since the barcodes are competing with themselves, and so if one barcode happens to have a delay, then the niche that it experiences is now smaller since the other barcodes have expanded to fill part of the overall niche. As a result, I don't think I agree with their conclusion that "net growth during the first 48h post-inoculation has to be lower than in the gnotobiotic mice."

– Figure 3E: It would be helpful to have more information on how they estimated growth rate; I don't see in the plot such a large difference (25%) between 0.53 and 0.4.

– For their conclusions related to increased clearance rate, how do they balance this conclusion/model with one related to variable growth rate in the gut due to nutrient availability and/or spatial localization. I think in particular since they are dealing with colonization, the spatial question is v. important.

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Fitness advantage of Bacteroides thetaiotaomicron capsular polysaccharide is dependent on the resident microbiota" for further consideration by eLife. Your revised article has been evaluated by Wendy Garrett (Senior Editor) and a Reviewing Editor.

The authors have made major efforts revising the manuscript, substantially addressing the reviewers concerns through a combination of additional experiments and analyses.

A few issues remain to be addressed:

1) There have been significant improvements to the lag phase analysis, and we accept the authors conclusions given their careful language, consideration of alternate hypothesis, and the good fit of their models. However, there continues to be issues in what has become Figure 3 -supplement 2. There appears to be data points in figure supplement 2B that are missing from figure supplement 2D. Perhaps these points are obscured, as occurred previously, but the source data for this supplement has less animals than reported in the legend.

2) Addition of challenge with avirulent Stm is a good experiment to demonstrate that the bottleneck following Stm infection occurs due to inflammation. However, the presentation of the Stm avirulent data in Figure 4C and its discussion in the text needs clarification. In Figure 4 —figure supplement 1 at n0=100 (the only dose used for Stm avirulent), the data for Stm avirulent appear nearly identical to the matched control. Yet, in Figure 4C the bottleneck caused by Stm avirulent appears substantially different than Control and in the text the authors claim that Stm avirulent creates a bottleneck. The authors should clarify why the difference between the two groups is so apparent in Figure 4C, but not as much in Figure 4—figure supplement 1.

3) The authors present three potential scenarios to explain why there is a bottleneck for the acapsular strain in mixed infection: 1) competition prior to reaching the cecum, 2) competition in the cecum prior to growth, and 3) difference in growth kinetics. The authors then carefully explain why they believe the bottleneck is attributable to growth kinetics. However, in the Results section (line 317), the authors propose the scenarios as 1) lag phase, 2) killing in the stomach, and 3) retention in the small intestine. We encourage the authors to include the explanation of why they exclude hypothesis 2, which they mention in the rebuttal, in the main text. Perhaps, a figure showing "a similar delay in CFU arrival in the feces for acapsular B. theta on single colonization as in competition", as they mention in the rebuttal, would help clarify this point.

https://doi.org/10.7554/eLife.81212.sa1

Author response

[Editors’ note: The authors appealed the original decision. What follows is the authors’ response to the first round of review.]

Reviewer #1 (Recommendations for the authors):

a. The generation of gnotobiotic mice is not described.

All gnotobiotic colonies were bred with the listed microbiota. This is now clearly described in the methods section and 16S sequencing data of all microbiotas has been added (Figure 2 —figure supplement 1).

b. Figure 1D cartoon, greys should be 1/7th of population? Better description of inoculum including non-WITS population would be helpful throughout.

We agree there was lack of clarity in description of our inocula throughout the manuscript. In fact, the diagram is correct: The six WITS strains were pooled at equal proportions, then the WITS pool was mixed 1:1 with the untagged population. To address this, we have added the actual measured composition of all inocula either to the main figures or as supplementary figures where several inocula contributed to one graph. We have also improved the methods section describing B. theta growth and preparation.

c. Figure 1F, the type of average should be expressed in the legend. Similar statements should be made in other figures as well.

Better information on the averages used and statistics has been added to all figure legends. Apologies for this omission.

d. Figure 1F, Ery appears to have an advantage.

Small differences due to antibiotic resistance cassettes are unavoidable. This statement has been corrected in the text. Additionally, we have run a simulation of the expected effect of fitness contributions due to the antibiotic resistance cassettes on the outcome of the competition experiments and this is expected to be less than 2-fold. As the CI between WT and acapsular B. theta range from 20- 106-fold, this is a relatively small error. Additionally, competition experiments were carried out with the antibiotic resistances exchanges with very similar results.

e. Figure 2A cartoon over-simplifies the biology, and suggests all death occurs prior to reaching the site of colonization, when in fact clones can be eliminated in the cecum or flow through without establishing colonization.

Indeed, we are aware that this is an oversimplification of a complex dynamic process, but the cartoon means to exemplify the simplified mathematical model which in the first instance used in Figure 2 assumes no further loss of tags after cecum arrival. This model was initially applied to calculate bottleneck size. This is later complemented with a more complex model depicted in Figure 3. This is now referred to as our “initial” model to avoid further confusion.

f. Figure 3 clearance is misspelled.

Corrected.

g. Figure 4b the challenges appear to independently alter populations in the cecum and feces, suggesting that the feces is not a reliable indicator of cecal populations.

This is a broad issue in microbiome research – by definition feces are always a sub-sample of the end-point of what is going on in the proximal large intestine and has been investigated both in mice and humans. However, it is not ethically justified in this case to sequentially sample the cecum of one mouse, making feces a useful if imperfect read-out for longitudinal sampling. Based on this, we relied on fecal population only to report a pre-challenge state (for which most mice showed all tag strains present). We provided the post-challenge fecal population as a point of comparison to pre-challenge, but all survival probabilities were calculated on based on the lost of tags in the cecal population. This has been clarified in the text and methods (highlighted in the text).

Reviewer #2 (Recommendations for the authors):

Line 61 – The use of the term "gnotobiotic" is incorrect, as it does not refer to any specific level of diversity. Gnotobiotics just means that the composition is known, which can be 0, 1 or 100s of microbes.

A good point – we have modified this to “low-complexity” or to simply naming the consortia.

Need to be careful not to use "CFU" to refer to qPCR data, which measures DNA levels not viable cells. A common technical term is "genome equivalents".

Note that we are actually reporting CFU. This was poorly described in the first version of the manuscript and has been much better reported here. qPCR is used to generate relative frequencies of barcoded strains, whilst the total CFU/g of all barcoded strains is determined by plating. The CFU/g of each barcoded strain is then calculated by simple multiplication.

The growth curve format and analysis was unclear. For example, Figure 1b would be aided by overlapping all the lines on the same graph, removing the log scaling on the y axis, and quantifying carrying capacity and lag phase parameters. Error bars are also missing/not visible.

Figure 1e – I cannot see the error bars, only 1 dot/tag is visible.

This is an excellent suggestion and has been carried out as suggested. Error bars are present but very narrow.

The correct abbreviation for overnight is "O/N" not "ON". Better yet, just spell out the words to avoid confusion.

Point taken! This has been corrected.

Line 112 – this seems like a big assumption – how do we know selection can't occur in 48 hours?

This is very valid criticism – rather the probability of non-random selection increases over time. This statement has been adapted to be more accurate. By re-isolating clones from the 48h-colonized mice and using these to produce a new barcoded inoculum we could experimentally exclude major increases in fitness over this timeframe in our models (Data in Figure 2 —figure supplement 6).

Line 117 – waiting to grow in the cecum also seems like a big assumption to me.

This statement is based on data that there is little to no colonization of the small intestine by B. theta. We have now added data where OligoMM12 were colonized for 8h with a large inoculum to allow consistent detection and we can demonstrate very low colonization of the upper or lower small intestine, consistent with this environment not being permissive for B. theta growth (Figure 3 —figure supplement 4).

Figure 2c – this panel is really unclear, need to indicate which tags are which by color.

This is a misunderstanding – each dot is one mouse not one tag (the evenness is based on the complete tag distribution within each animal), now better indicated in the figure legend. For observing the distribution of individual tags on each experiment we have added Figure 2 —figure supplement 3 and 4 including all raw data.

Figures 2d-g – it would help to include the wt and mutant strains on the same graph, to enable comparisons between strains and appropriate statistics.

We consider that separation between the two graphs helps to deliver the message about comparisons for a particular strain (WT or acapsular) among the 3 different microbiotas. For easiness of comparison between strains we have maintain the same y axis range.

Figure 2b – multiple typos: "extintino", "optimun"

Corrected.

Figure 3e,f – these curves raised a lot of questions. The first timepoint has high levels, especially for panel f. A time zero needs to be included to rule out background amplification. Given the 3 OOM difference between the two panels in starting levels I'm not sure it is fair to compare the curves. The n is very low here, need more mice and a replicate experiment. Controls should be included with the individual members of the OligoMM community to ensure that there is no background amplification.

N has been increased to at least 7 mice for these growth curves. Note that there is no qPCR in this panel – the data is from plating and no colonies are detected on the selective antibiotics used from the OligoMM12 colony at time zero – this control is explicitly stated in the methods section referring to this figure. These graphs and their analysis have been updated and are now shown in figure 3 —figure supplement 2

Line 257 – this is interesting but seems very speculative. It would really help to formulate a testable hypothesis.

Based on improved estimates for in vivo growth curves, this section has been considerably modified and I hope is now clearer.

Line 260 – this section did not follow from the rest of the paper and might be better to move it to a follow-up study. The experiments are also very preliminary and conflict with prior literature. For example, the diet experiment would need to be repeated with a matched control diet, 16S profiling to ensure that the rest of the gut microbiota is altered, bile acid quantification, and a longer time series (3 days is quite short).

This is a very fair criticism, and these data were included to give the research community some insight into the limits and possibilities of this approach, rather than to reveal deep biological insight. Given that the phenotype in high-fat diet is below detection limit, and the evenness cannot increase, we would not suggest using a control diet here. This section has been party re-written. We have also included an addition control with colonization with “avirulent” Salmonella to clarify the mechanism driving a tight bottleneck in the attenuated Salmonella infection. Data on the extent of inflammation is also now included.

Figure 4b – the difference in the control group is worrisome, although perhaps that's due to comparing cecal contents to feces. It would probably be safer to just compare the same sample type across groups.

We completely agree with this observation and note that a large number of publications in this field rely entirely on feces analysis. Owing to this, we have done all the quantitative analyses on cecum content between groups (clarified also in the text) and include the feces analysis only for absolute transparency, and for comparison to published feces-based studies. Specifically, the data points in Figure 4C are based on cecum content.

Line 314 – this conclusion is not well supported given all the potential host differences.

We agree that currently we do not present sufficient data to completely delineate host and microbiota effects. Additional data has been added allowing us to exclude effects of intestinal IgA and of lytic phage, as well as to determine that SPF-microbiota-driven mechanisms are already present 24h post re-colonization of germ-free mice. Again, this text has been extensively re-written.

Reviewer #3 (Recommendations for the authors):

My comments below are mainly focused towards the interpretation of their data, in particular the spatial localization question I bring up below.

– Line 97-100: they state that the antibiotic cassette isn't really altering fitness, which I found surprising – there is generally some fitness cost to antibiotic cassettes, at least in in vitro experiments. Upon examination of Figure 1F, it actually looks like for the wild-type, the Ery cassette wins out in competition. This should at least be tested statistically, as a half-log-fold after 2 days is fairly important biologically.

We agree that there are inevitable and variable costs of antibiotic cassette production, and these may have a numerical effect on our calculations. To address this, we have carried out a simulation of the effect of a slight fitness advantage/disadvantage of our strains in the competition experiments and come to the conclusion that only a 2-fold difference in the competitive index would result. Given that we have competitive indexes in the rage of 20-106, this is not going to change our conclusions. Nevertheless, this now carefully discussed in the Results section. For our barcode analysis the effect is uniform over all barcodes and therefore does not alter the competition between barcoded strains

– Looks like there are consistent biases for certain barcodes in Figure 1E? e.g., pink/purple is consistently lowest in WT, albeit not acapsular. Again, a half-log-fold after 2 days is not small here and would accumulate over time. Can they rule out fitness differences in any other way? I realize that they test for mutation in selection by reinoculating isolated strains, but this does not test for pre-existing mutation.

“Neutrality” is absolutely the most challenging part of neutral tagging approaches, particularly where strain construction is a multi-step process. To address this, we have:

1) Whole-genome sequenced all of our barcoded strains. This reveals random SNPs in some of the barcoded strains. However, none of these are in genes known to have a major fitness effect in vivo.

2) We have re-done our estimates of β sequentially leaving out one of the barcoded strains from the analysis. This evaluation is now shown in Figure 2 supplement 6. Removing any individual barcode had no significant effect on the estimated values of β.

– Line 171-174: they conclude that size of open niche does not translate linearly into colonization probability, but this seems like a given, since the barcodes are competing with themselves, and so if one barcode happens to have a delay, then the niche that it experiences is now smaller since the other barcodes have expanded to fill part of the overall niche. As a result, I don't think I agree with their conclusion that "net growth during the first 48h post-inoculation has to be lower than in the gnotobiotic mice."

This is an interesting point. In the SPF mice, the expansion of the population relative to the inoculum is 100-fold less than in low-complexity microbiota mice, but the probability of an individual clone to colonize is only 10-fold lower. Our logic was as follows: Β multiplied by the total population size of B. theta in the inoculum can be interpreted that the number of clones from the inoculum represented in the 48h cecal population. This is 10-fold smaller in SPF mice than in an OligoMM12-colonized mouse. However, the total population size is 100-fold smaller, indicating that the net expansion of each clone required to generate the final population is less. We are using “net growth rate” to mean growth rate minus clearance rate, i.e., equivalent to the rate of expansion of the population.

As the barcoded strains are massively outnumbered by the untagged B. theta (between 1000- and 1’000’000-fold depending on the experiment) then we assume that competition mainly occurs between a tagged clone and untagged clones, and this will be the case in both the SPF microbiota background and the low-complexity microbiota. Large stochastic differences in lag phase exist between barcoded clones can have interesting effects on the estimate of β, mainly by increasing the variation in situations where uniform recovery is expected and we did investigate this in the context of some of the models, but the effect on the final estimates of β are small unless there are consistent differences in lag phase distribution between the strains (see the comparison for lag-phase effects for acapsular and wildtype B. theta).

– Figure 3E: It would be helpful to have more information on how they estimated growth rate; I don't see in the plot such a large difference (25%) between 0.53 and 0.4.

This data is now added in Figure 3 —figure supplement 2. Indeed, this was a point of considerable discussion between the mathematical modelers and biologists on this project and we have added more mice in order to improve these estimates.

– For their conclusions related to increased clearance rate, how do they balance this conclusion/model with one related to variable growth rate in the gut due to nutrient availability and/or spatial localization. I think in particular since they are dealing with colonization, the spatial question is v. important.

We agree there are major hypotheses relating to differences in growth niches and distributions of growth rates for Bacteroides strains in the gut (Lee et al., 2013). The effect of subpopulations with different growth rates was investigated in a previous modeling project looking at Salmonella gut colonization (Maier et al., 2014) but at the resolution we are working at, this generally fails to explain the distribution of barcodes recovered. A discussion will be added to the modelling supplement, which will also be re-worded to increase clarity. Of note the cecum does not contain structured mucus layers – these are only present in the mid-distal colon after the point of sampling for β estimates in our work.

[Editors’ note: what follows is the authors’ response to the second round of review.]

The authors have made major efforts revising the manuscript, substantially addressing the reviewers concerns through a combination of additional experiments and analyses.

A few issues remain to be addressed:

1) There have been significant improvements to the lag phase analysis, and we accept the authors conclusions given their careful language, consideration of alternate hypothesis, and the good fit of their models. However, there continues to be issues in what has become Figure 3 -supplement 2. There appears to be data points in figure supplement 2B that are missing from figure supplement 2D. Perhaps these points are obscured, as occurred previously, but the source data for this supplement has less animals than reported in the legend.

Yikes! Thank you for spotting this error! Figure 3 —figure supplement 2 has been corrected accordingly. We discovered an error in the code that generated Figure 3 – Supplement 2D which had resulted in some data points failing to appear on the graph and source data file. This has been resolved for the source data file and Figure 3- Supplement 2D has been updated. It now includes all points from Figure 3 —figure supplement 2B and C. Additionally, points have been jittered in the x and y-axes in all figures to allow better visualization of overlayed data. Figure legend have also been updated to include describe changes. This does not change the interpretation of the data.

2) Addition of challenge with avirulent Stm is a good experiment to demonstrate that the bottleneck following Stm infection occurs due to inflammation. However, the presentation of the Stm avirulent data in Figure 4C and its discussion in the text needs clarification. In Figure 4 —figure supplement 1 at n0=100 (the only dose used for Stm avirulent), the data for Stm avirulent appear nearly identical to the matched control. Yet, in Figure 4C the bottleneck caused by Stm avirulent appears substantially different than Control and in the text the authors claim that Stm avirulent creates a bottleneck. The authors should clarify why the difference between the two groups is so apparent in Figure 4C, but not as much in Figure 4—figure supplement 1.

We absolutely agree that this is confusing. The reason for the discrepancy is that the qPCR data here was carried out on enrichments of 50% of the cecum, while CFU determination was carried out with much less material so the detection limit for CFU is higher than that for qPCR detection. Figure 4 —figure supplement 1 now shows raw CT values rather than calculated CFUs of each tagged strain: These are the data used in calculation of the model. Here it is clear that the CT values are slightly lower (i.e. effectively all barcodes are present) in the control mice at day 3, whilst some barcodes are below detection (CT>30) in the avirulent Salmonella-infected mice. Noticeably, the avirulent Salmonella has much less of a bottleneck effect than the attenuated-Salmonella infected mice, which had measurable intestinal inflammation: here there is very little barcode recovery (i.e., except in one mouse, the majority of CT values are >30). We hope this is now clearer.

3) The authors present three potential scenarios to explain why there is a bottleneck for the acapsular strain in mixed infection: 1) competition prior to reaching the cecum, 2) competition in the cecum prior to growth, and 3) difference in growth kinetics. The authors then carefully explain why they believe the bottleneck is attributable to growth kinetics. However, in the Results section (line 317), the authors propose the scenarios as 1) lag phase, 2) killing in the stomach, and 3) retention in the small intestine. We encourage the authors to include the explanation of why they exclude hypothesis 2, which they mention in the rebuttal, in the main text. Perhaps, a figure showing "a similar delay in CFU arrival in the feces for acapsular B. theta on single colonization as in competition", as they mention in the rebuttal, would help clarify this point.

This is an excellent point. We have modified the text around line 317 to include the very clear argumentation used in the rebuttal letter. Additionally, one graphs showing the competitive colonization dynamics of both strains have been moved from the mathematical modeling appendix into Figure 3 —figure supplement 2E. Using this data, a new graph has been generated (Figure 3 —figure supplement 2F) in order to better support these claims by showing no difference in the delay during single colonization as in competition.

https://doi.org/10.7554/eLife.81212.sa2

Article and author information

Author details

  1. Daniel Hoces

    Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Zürich, Switzerland
    Contribution
    Conceptualization, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1451-5166
  2. Giorgia Greter

    Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Zürich, Switzerland
    Contribution
    Investigation, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  3. Markus Arnoldini

    Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Zürich, Switzerland
    Contribution
    Conceptualization, Formal analysis, Visualization, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  4. Melanie L Stäubli

    Institute of Microbiology, Department of Biology, ETH Zurich, Zurich, Switzerland
    Contribution
    Formal analysis, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  5. Claudia Moresi

    Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Zürich, Switzerland
    Contribution
    Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
  6. Anna Sintsova

    Institute of Microbiology, Department of Biology, ETH Zurich, Zurich, Switzerland
    Contribution
    Formal analysis, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  7. Sara Berent

    Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Zürich, Switzerland
    Contribution
    Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
  8. Isabel Kolinko

    Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Zürich, Switzerland
    Contribution
    Resources, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  9. Florence Bansept

    Sorbonne Université, CNRS, Institut de Biologie Paris-Seine (IBPS), Laboratoire Jean Perrin (LJP), Paris, France
    Present address
    Department for Evolutionary Theory, Max Planck Institute for Evolutionary Biology, Plön, Germany
    Contribution
    Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0562-9222
  10. Aurore Woller

    Sorbonne Université, CNRS, Institut de Biologie Paris-Seine (IBPS), Laboratoire Jean Perrin (LJP), Paris, France
    Present address
    Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot, Israel
    Contribution
    Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  11. Janine Häfliger

    Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Zürich, Switzerland
    Present address
    Klinik für Gastroenterologie und Hepatologie, University Hospital Zurich, Zurich, Switzerland
    Contribution
    Resources, Writing – review and editing
    Competing interests
    No competing interests declared
  12. Eric Martens

    Department of Microbiology and Immunology, University of Michigan Medical School, Ann Arbor, United States
    Contribution
    Resources, Writing – review and editing
    Competing interests
    No competing interests declared
  13. Wolf-Dietrich Hardt

    Institute of Microbiology, Department of Biology, ETH Zurich, Zurich, Switzerland
    Contribution
    Resources, Funding acquisition, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9892-6420
  14. Shinichi Sunagawa

    Institute of Microbiology, Department of Biology, ETH Zurich, Zurich, Switzerland
    Contribution
    Supervision, Funding acquisition, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3065-0314
  15. Claude Loverdo

    Sorbonne Université, CNRS, Institut de Biologie Paris-Seine (IBPS), Laboratoire Jean Perrin (LJP), Paris, France
    Contribution
    Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0888-1717
  16. Emma Slack

    Institute of Food, Nutrition and Health, Department of Health Sciences and Technology, ETH Zurich, Zürich, Switzerland
    Contribution
    Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Investigation, Methodology, Writing – original draft, Writing – review and editing
    For correspondence
    emma.slack@hest.ethz.ch
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2473-1145

Funding

Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (NCCR Microbiome)

  • Wolf-Dietrich Hardt
  • Shinichi Sunagawa
  • Emma Slack

Gebert Rüf Stiftung (GR073_17)

  • Claude Loverdo
  • Emma Slack

Botnar Research Centre for Child Health, University of Basel (BRCCH_MIP)

  • Shinichi Sunagawa
  • Emma Slack

Agence Nationale de la Recherche (ANR-21-CE45-0015)

  • Claude Loverdo

Centre National de la Recherche Scientifique (MITI CNRS AAP adaptation du vivant à son environnement)

  • Claude Loverdo

Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (40B2-0_180953)

  • Emma Slack

Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (310030_185128)

  • Emma Slack

Agence Nationale de la Recherche (ANR-20-CE30-0001)

  • Claude Loverdo

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We express our gratitude to Sven Nowok and Dominik Bacovcin for their great support in maintaining the gnotobiotic mouse facility and to the whole team in EPIC and RCHCI. Also, we thank Dr. Annika Hausmann and Verena Lentsch for the discussions about this project. We acknowledge Prof. Bärbel Stecher for provision of the OligoMM12 bacterial strains used to generate the mouse colony used in these experiments.

This work was funded by NCCR Microbiomes, a research consortium financed by the Swiss National Science Foundation (ES, SS, W-DH); Swiss National Science Foundation (40B2-0_180953, 310030_185128) (ES), European Research Council Consolidator Grant (no. 865730-SNUGly) (ES), Gebert Rüf Microbials (GR073_17) (ES, CL); Botnar Research Centre for Child Health Multi-Invesitigator Project 2020 (BRCCH_MIP: Microbiota Engineering for Child Health) (ES, SS), Agence Nationale de la Recherche (ANR-21-CE45-0015, ANR-20-CE30-0001) (CL), and MITI CNRS AAP adaptation du vivant à son environnement (CL). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Ethics

All animal experiments were performed with approval from the Zürich Cantonal Authority under license number ZH120/19.

Senior Editor

  1. Wendy S Garrett, Harvard T.H. Chan School of Public Health, United States

Reviewing Editor

  1. Matthew K Waldor, Brigham and Women's Hospital, United States

Version history

  1. Preprint posted: June 19, 2022 (view preprint)
  2. Received: June 20, 2022
  3. Accepted: February 8, 2023
  4. Accepted Manuscript published: February 9, 2023 (version 1)
  5. Version of Record published: March 14, 2023 (version 2)

Copyright

© 2023, Hoces et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,611
    Page views
  • 274
    Downloads
  • 3
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Daniel Hoces
  2. Giorgia Greter
  3. Markus Arnoldini
  4. Melanie L Stäubli
  5. Claudia Moresi
  6. Anna Sintsova
  7. Sara Berent
  8. Isabel Kolinko
  9. Florence Bansept
  10. Aurore Woller
  11. Janine Häfliger
  12. Eric Martens
  13. Wolf-Dietrich Hardt
  14. Shinichi Sunagawa
  15. Claude Loverdo
  16. Emma Slack
(2023)
Fitness advantage of Bacteroides thetaiotaomicron capsular polysaccharide in the mouse gut depends on the resident microbiota
eLife 12:e81212.
https://doi.org/10.7554/eLife.81212

Share this article

https://doi.org/10.7554/eLife.81212

Further reading

    1. Microbiology and Infectious Disease
    Hui Han, Rong-Hua Luo ... Cheng-Gang Zou
    Research Article

    Angiotensin-converting enzyme 2 (ACE2) is a major cell entry receptor for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). The induction of ACE2 expression may serve as a strategy by SARS-CoV-2 to facilitate its propagation. However, the regulatory mechanisms of ACE2 expression after viral infection remain largely unknown. Using 45 different luciferase reporters, the transcription factors SP1 and HNF4α were found to positively and negatively regulate ACE2 expression, respectively, at the transcriptional level in human lung epithelial cells (HPAEpiCs). SARS-CoV-2 infection increased the transcriptional activity of SP1 while inhibiting that of HNF4α. The PI3K/AKT signaling pathway, activated by SARS-CoV-2 infection, served as a crucial regulatory node, inducing ACE2 expression by enhancing SP1 phosphorylation—a marker of its activity—and reducing the nuclear localization of HNF4α. However, colchicine treatment inhibited the PI3K/AKT signaling pathway, thereby suppressing ACE2 expression. In Syrian hamsters (Mesocricetus auratus) infected with SARS-CoV-2, inhibition of SP1 by either mithramycin A or colchicine resulted in reduced viral replication and tissue injury. In summary, our study uncovers a novel function of SP1 in the regulation of ACE2 expression and identifies SP1 as a potential target to reduce SARS-CoV-2 infection.

    1. Microbiology and Infectious Disease
    Sabrina Tetzlaff, Arne Hillebrand ... Christian Schmitz-Linneweber
    Research Article

    The mitochondrial genomes of apicomplexans comprise merely three protein-coding genes, alongside a set of thirty to forty genes encoding small RNAs (sRNAs), many of which exhibit homologies to rRNA from E. coli. The expression status and integration of these short RNAs into ribosomes remains unclear and direct evidence for active ribosomes within apicomplexan mitochondria is still lacking. In this study, we conducted small RNA sequencing on the apicomplexan Toxoplasma gondii to investigate the occurrence and function of mitochondrial sRNAs. To enhance the analysis of sRNA sequencing outcomes, we also re-sequenced the T. gondii mitochondrial genome using an improved organelle enrichment protocol and Nanopore sequencing. It has been established previously that the T. gondii genome comprises 21 sequence blocks that undergo recombination among themselves but that their order is not entirely random. The enhanced coverage of the mitochondrial genome allowed us to characterize block combinations at increased resolution. Employing this refined genome for sRNA mapping, we find that many small RNAs originated from the junction sites between protein-coding blocks and rRNA sequence blocks. Surprisingly, such block border sRNAs were incorporated into polysomes together with canonical rRNA fragments and mRNAs. In conclusion, apicomplexan ribosomes are active within polysomes and are indeed assembled through the integration of sRNAs, including previously undetected sRNAs with merged mRNA-rRNA sequences. Our findings lead to the hypothesis that T. gondii's block-based genome organization enables the dual utilization of mitochondrial sequences as both messenger RNAs and ribosomal RNAs, potentially establishing a link between the regulation of rRNA and mRNA expression.