Improving ulcerative colitis prospects through fecal microbiota transfer: atypical donor microbiota can boost success rate

ABSTRACT

Fecal microbiota transfer (FMT) has been used with variable success in the experimental treatment of ulcerative colitis (UC), and efforts to improve its efficacy very much remain a matter of trial and error. We recently predicted that atypical donor microbiota could improve results. Here, we provide experimental support for this prediction in a rat model where we induced a transition of the intestinal ecosystem to an alternative state characterized by chronic low-grade inflammation and dysbiosis. While autologous FMT did barely or not enhance the restoration of a healthy microbiota compared to a control group without FMT, the atypical allogenic microbiota from one of two donor rat strains proved remarkably successful in the restoration of a healthy microbiota, in some cases accompanied by a healthy distal colon histology. These results allow the rationalization of research efforts towards improvement of FMT efficacy in humans, and indicate that (initial) success of FMT should be monitored at the microbiota level as much as at the level of clinical symptoms. More importantly, they provide further support for our earlier published, clinical-data-based, conceptual model of the intestinal ecosystem which suggests promising opportunities for therapeutic innovation in UC treatment. This model notably predicts that, and explains why, symbio-therapy, acting on both microbiota and inflammation, may be more efficient than conventional inflammation-directed therapies, and can be used to guide and monitor treatments.

Introduction

Reciprocal microbiota–host influences constitute the basis for the establishment of a relatively stable state of equilibrium in the intestinal ecosystem.^1,2 Under given external conditions, this equilibrium can take different forms representing so-called alternative stable states (Figure 1A,B; as opposed to condition-dependent states, Figure 1C), associated with health or (pre-)disease.^3,4 There are strong indications that the stability of disease-associated microbiota states over a range of intestinal immune-mediated inflammation conditions, including when strongly reducing inflammation under conventional treatment, hinders or prohibits the recovery of intestinal health in (pediatric) ulcerative colitis (UC) patients,³ thus perpetuating disease and the risk of relapse through mutually reinforcing deleterious microbiota–host interactions (Figure 1D).

Figure 1.

Alternative stable states of the intestinal ecosystem and study design. (A) Alternative stable states of an ecosystem as beads in a stability landscape. The dashed line represents the boundary between two basins of attraction (transition fold). (B) Alternative states (solid lines) can both exist under a range of identical conditions (bi-stable range). Dashed line, see panel A. Width and shape of the basins of attraction (top), and thereby the stability (and relative frequency) of the alternative states and likelihood of transitions across the transition fold (due to stochastic movements; red bidirectional arrow), change with changing conditions, as illustrated by the changing distances between solid lines and the dashed line.⁵ When changing conditions push the ecosystem beyond a tipping point (black dashed arrow), where the basin of attraction of its present state disappears, it rapidly transits to an alternative state (red arrow). (C) Condition-dependent states. Assuming that the original ecosystem state is represented by the green dots, the models from panels B and C both predict a change in the ecosystem state when the external conditions change from c1 to c2. When the conditions change back to c1, the model in panel B predicts that the system will remain in the alternative state (red dot), while the model in panel C predicts that the system returns to its original state. In model B, conditions would have to reach the tipping point to the left of c1 to force a return to the original ecosystem state. (D) Mutually reinforcing microbiota–host interactions may propel the intestinal ecosystem to an alternative state when a tipping point is reached.¹ (EFG) Conceptional model of the intestinal ecosystem with alternative microbiota states, alternative host (inflammation) states and holobiont states.³ (E) Conventional inflammation-directed treatment (*) of UC will not systematically restore a healthy microbiota state (upper solid blue line) as the tipping point for microbiota restoration (a virtual point situated at a negative inflammation value (bottom left; cf Figure 1B)) cannot be reached.³ (F) FMT (*) will not systematically reduce inflammation to a non-pathologic level (healthy state represented by the left-hand solid brown line) unless the tipping point for inflammation reduction (top right) can be overcome to force a transition to the healthy state (cf Figure 1B). (G) Alternative stable states of the intestinal ecosystem (holobiont states) at intersections of stable microbiota and host states: “healthy” and “UC”. Open circles indicate less-stable states (closer to tipping points and transition folds). (H) Study design. A–D indicate experimental groups (10 mice per group), E indicates allogenic donors (one donor for each recipient in group D). See the main text for subgroups of D and E. Allogenic and autologous donor microbiota were collected at days 48 and 54, respectively, and were administered on days 96–98 (dashed blue lines). (I) Experimental summary: DSS-induced inflammation and antibiotics treatments are used to profoundly perturb the intestinal ecosystem (red star, bottom right). The perturbed system is then left alone or subjected to FMT to determine whether it evolves to the normal state of an age-matched control group (green) or an alternative state (red). (J) Alpha diversity of microbiota samples. The colors indicate the groups as shown in Figure H. The blue arrows indicate the microbiota at the time of FMT donor microbiota collection. See also Supplemental Figure 1A for the Inv Simpson and Pielou diversity indices. (K) Fecal calprotectin levels at days 87 and 95 were used as a marker of DSS-induced intestinal inflammation.⁶ Samples are indicated as experimental group (Figure H), followed by rat number. Mean and SEM for three technical replicates (independent extractions); two replicates for group C. Fecal material for this analysis was available for only some of the animals. Note that while for rat 19 no calprotectin induction was observed at day 87, this rat showed a higher-than-normal calprotectin level, similar to that found in all other tested DSS-treated rats at day 95, as well as one of the highest distal colon histologic inflammation scores at the end of the experiment (Supplemental Table 4). The former result may be due to prolonged sample storage before measurement or an unidentified technical problem.

We recently proposed a conceptual model picturing the interdependent behavior of intestinal microbiota and host immune (inflammation) status to illustrate this concept and its consequences for the treatment of disease (Figure 1E−G).³ Our model predicts that conventional, inflammation-directed, treatment of UC will never lead to the systematic recovery of a healthy intestinal ecosystem and stable remission in all patients as the tipping point for microbiota restoration cannot be reached (Figure 1E, bottom left). In the pediatric UC study⁷ at the basis of our analyses, only 36% of patients reached a non-pathologic inflammation level after 1 y of treatment, which is in line with generally observed success rates in clinical UC practice,⁸ and of these, only approximately half restored a healthy microbiota.³ In accordance, the initial microbiota status proved to be highly determining for remission after 1 y of conventional treatment. A second prediction of the model is that an alternative, still largely experimental, treatment, fecal microbiota transfer (FMT) using the microbiota from an average healthy donor, will never be 100% successful either. To achieve 100% efficacy, a “healthier than healthy” or atypical donor microbiota allowing to surpass the tipping point for inflammation reduction (Figure 1F, top right) would be needed. Even if a goal of 100% efficacy may be utopic, the model predicts that microbiota “compositional quality” (and associated functionality) will be an important factor influencing the success rate as a microbiota that brings the system closer to the inflammation tipping point (Figure 1F, top right) augments the chances of transition to the healthy state (no inflammation) as a result of stochastic movements (cf Figure 1B). These predictions are in agreement with the variable success of FMT,^9,10 and our model provides a rationale for much of current FMT research seeking to define and identify “high-quality” donor microbiota in order to improve success rates.¹¹

The present study aims to verify these theoretical predictions regarding FMT in an experimental setup in rats, where we compare the efficacy of autologous and allogenic FMT. Autologous FMT, using the subject’s own fecal microbiota collected before the onset of disease (material generally not available in human patients), may be expected to ideally match the subject’s needs, but would according to our model not be sufficient to guarantee success in the suppression of inflammation (Figure 1F). Allogenic FMT, using the microbiota from a healthy donor, may perform better or worse, depending on the “quality” of the microbiota. We treated this as an open question, considering that either result would be of interest. Another reason for this open approach is that, today, no one truly knows what characterizes a healthy¹² or efficient FMT donor microbiota (although there appears to be a consensus on a requirement of high diversity¹¹), let alone what could constitute a “healthier than healthy” microbiota evoked above. Our results indicate that FMT with an atypical allogenic donor microbiota can outperform autologous FMT in the recovery of a highly perturbed intestinal ecosystem, at least for the microbiota component of the system. The data suggests that this in turn improves the odds of suppressing colon inflammation. These results are in agreement with and provide further support for our holistic model, which may, beyond FMT, lead to important therapeutic innovation in the treatment of UC.

Materials and methods

Study design

The study design is schematically represented in Figure 1H, and the objectives are summarized in Figure 1I. Detailed explanations are provided in the first paragraphs of the Results section. Forty Fischer 344 rats were obtained from Charles River Laboratories and assigned to four experimental groups (A, B, C, D; 10 rats per group (numbers based on previous experience in a comparable experiment without FMT⁴)), randomizing affiliation (litter) and cage occupancy at the provider to minimize the confounding effects that these factors might have on microbiota composition. The body weight distribution was checked to be similar in all groups. In addition, five Lewis rats were obtained from Envigo (group E1), and five Dark Agouti rats were obtained from Janvier Labs (group E2) (Supplemental Table 1), to serve as allogenic FMT donors. All the rats (male, 5 weeks old) were housed in individual cages to prevent inter-rat microbiota transfer by coprophagy, with autoclaved wood chips bedding and paper towel enrichment, in a conventional facility at INRAE (IERP lab). They initially received a γ-irradiated standard chow diet (Supplemental Table 2, diet 1). From day 18 onward, the chow diet was replaced with a diet without crude fiber, with slightly different macronutrient composition (Supplemental Table 2, diet 2). Groups B, C, and D received 3% w/v of dextran sodium sulfate (DSS; MP Biomedicals, MW 36,000–50,000) in autoclaved drinking water over three periods of 3 d between days 55 and 85 (Figure 1H) and a cocktail of three antibiotics (vancomycin (0.5 g/L), neomycin sulfate (1 g/L) and ampicillin (1 g/L)) in autoclaved drinking water from days 87 to 95. Group C was subjected to autologous FMT by intragastric gavage on days 96, 97, and 98, with fecal samples that had been collected on day 54 and stored at −80 °C in a maltodextrin-trehalose mixture (MD) supplemented with ascorbic acid and cysteine.¹³ Subgroups D1 and D2 (5 rats each) were subjected to allogenic FMT with fecal samples that had been collected from groups E1 and E2, respectively, on day 48 (a separate donor for each recipient), and group B was subjected to placebo treatment with water. At the end of the experiment, the rats were euthanized by deep gas anesthesia with isoflurane followed by exsanguination by cardiac puncture.

Study endpoints

At various time points during the study, we examined how the intestinal microbiota of individual rats evolved under the different treatments (principal coordinates analysis (PCoA)). At the end of the experiment, we examined whether alternative microbiota states could be distinguished and characterized (PCoA analysis, Dirichlet-multinomial mixture models (DMM), differential taxa composition, genus correlation networks), and evaluated the effect of autologous or allogenic FMT vs control on microbiota fate (i.e. alternative microbiota state association) after DSS- and antibiotics-induced perturbation. At the host level, we evaluated the existence of alternative states through histological evaluation of the distal colon to identify signs of low-level inflammation. The microbiota and host results were integrated to identify alternative holobiont states. Based on the results of these analyses, efficient donor microbiota, capable of restoring a healthy microbiota or holobiont state were identified.

All the rats were included in the analyses. Individual samples are sometimes missing owing to technical problems (insufficient amounts of fecal material).

Ethics approval

The experimental protocol received the authorization from the French Ministère de l’Éducation Nationale, de l’Enseignement Supérieur et de la Recherche (APAFIS # 2734-201510281818208 v5). This study adheres to the ARRIVE guidelines for preclinical studies.

Fecal microbiota 16S rRNA gene profiling

Fresh fecal samples were collected at several time points and stored at −80 °C. Total DNA was extracted according to Costea et al.¹⁴ (protocol #1). DNA integrity was assessed using a Fragment Analyzer (Agilent Technologies), and the DNA concentration was determined by Qubit (Invitrogen) and Nanodrop (Thermo Scientific). For each sample, the microbiota composition was assessed by Miseq sequencing of the V3–V4 region of the bacterial 16S rRNA gene. Samples were prepared according to the Illumina protocol, using the forward primer TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG and reverse primer GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC for amplicon PCR (Biofortis).

Host inflammatory status

Induction of inflammation by DSS treatments was verified through the measurement of fecal calprotectin levels by ELISA (Novus Biologicals kit NBP3-06957) according to the instructions of the supplier at days 87 (after the last DSS treatment) and 95. At sacrifice, Swiss rolls were prepared from the distal part of the colon for histological evaluation of host tissue inflammatory status. Hematoxylin eosin saffron (HES) and periodic acid schiff (PAS) stained slides were prepared at the histology facility @BRIDGe (GABI lab, INRAE). Inflammatory status was evaluated by a veterinary histopathologist in a blinded procedure, using the criteria provided in Supplemental Table 3. Scores for individual criteria (Supplemental Table 4) were totaled to obtain an overall score.

Bioinformatics and statistical analysis

Raw 16S rRNA sequence reads were trimmed to remove remaining adapters and primer sequences, and checked for quality (≥30) and length (≥250 bp) using cutadapt.¹⁵ Reads were further corrected for known sequencing errors using SPAdes¹⁶ and then merged using PEAR.¹⁷ Vsearch¹⁸ was used to dereplicate (--derep_ --prefix–minuquesize 2) and cluster (--cluster_fast --id 0.97) the merged reads into operational taxonomic units (OTUs), as well as check for chimeras (--uchime3_denovo). Representative sequences for each OTU were taxonomically assigned using the RDP tools suite and database.¹⁹ The OTU table was rarefied hundred times to a total read count of 45,000 per sample (R-Vegan) and the average rarefied count table was retained for downstream analysis. The OTU table was then aggregated at the genus level (R-PhyloSeq).

Diversity and richness were estimated at the OTU level (Vegan). Principal coordinate analysis was computed on Bray‒Curtis dissimilarity matrices obtained from the genus abundance table (Vegan). Dirichlet-multinomial mixture models (DMM)²⁰ were applied to the genus count table of samples collected at day 130. The optimal number of clusters was assessed using the Laplace index. The differential taxa composition between groups was determined using linear models for differential abundance analysis (Linda (R-MicrobiomeStat)). The DMM model fit estimate was used to select genera contributing the most to DMM CI/II classification (n = 43 genera). Recursive feature elimination (R-Caret and R-Randomforest) using the random forest classifier algorithm was iteratively applied a hundred times to select an optimal features list (n = 18 genera). The final classification model was trained on the abundance of the 18 genera selected above at D130 using the random forest algorithm. The performance of the machine learning model was evaluated using the R-MLeval package with ROC and AUC values used as descriptors. The DMM CI or CII classification of samples between days 54 and 130 was predicted using the final classification model.

Genus correlation networks were assessed using FastSpar²¹ and displayed using ggnet2 (R-GGally). The shortest paths between the main genera (i.e. Phocaeicola_A_858004, Ruminococcus_E, Akkermansia, Ruminococcaceae) appearing in the correlation graphs were calculated as the minimal distance (number of edges) to connect vertices (genera).

Pairwise Wilcoxon rank sum tests were used to evaluate differences in alpha diversity between autologous and allogenic donors and differences between allogenic and autologous donors or DSS and antibiotics treatment effects on the main PCoA axes. Envfit (R-Vegan) was used to fit genus vectors to PCoA plots on the basis of the Pearson correlation coefficient and p-value criteria, to establish a first characterization of successful allogenic donor microbiota.

Results

FMT in a strongly perturbed intestinal ecosystem

In order to test the efficacy of autologous and allogenic FMT, we first sought to profoundly perturb the rat intestinal ecosystem. To this end, the rats were successively subjected to (a) an alimentary shift to a fiber-free (FF) diet (Supplemental Table 2), which was administered until the end of the experiment, thus setting the context to an extreme form of the fiber reduction associated with a modern western diet and known to compromise the microbiota (Figure 1H, all groups), (b) repeated DSS treatments to induce chronic colitis, a model of UC²² (groups B, C, D), and (c) a broad-spectrum antibiotics cocktail (groups B, C, D) (prior to clinical FMT, antibiotics are often administered to “clear” the GI tract and putatively favor engraftment). The control group (group A) received no particular treatment other than the diet shift. As expected, diet shift and more so antibiotics treatment strongly reduced microbiota diversity (Figure 1J, Supplemental Figure 1A). In the FF diet context, the DSS treatments reduced microbiota richness (number of OTUs) but did hardly or not affect Shannon, Inv Simpson, or Pielou diversity. On the host side, the intended induction of low-grade intestinal inflammation in groups B, C, and D was confirmed by an increase in the fecal inflammation marker calprotectin at day 87 (after the last DSS treatment) (Figure 1K). At day 95, after the antibiotics treatment and just before FMT, the calprotectin levels in groups B, C, and D were still higher than in control group A.

After these treatments, the rats were either left alone (group B) or subjected to autologous (group C) or allogenic (group D) FMT. Fecal samples for autologous FMT were collected while the animals were fed the FF diet, before the first DSS treatment (Figure 1H, day 54). Allogenic fecal samples were collected from two rat strains (groups E1/E2) belonging to different major histocompatibility complex (MHC) classes distinct from that of the receiving animals (Supplemental Table 1) and obtained from two providers other than the provider for all other rats to maximize the likelihood that the allogenic microbiota would be different from the autologous microbiota.^23,24 These rats also received the FF diet during the month before sample collection at day 48 (Figure 1H) to mirror the recruitment of allogenic donors from a population with dietary habits similar to those of the recipients. The allogenic group E1 donor microbiota showed a diversity similar (Shannon index) to or slightly lower (9% less OTUs) than that of the autologous donors (Supplemental Figure 1B). The allogenic group E2 microbiota showed considerably less diversity by both measures (43% less OTUs).

After the antibiotics treatment, natural recovery (group B) led to a significant improvement in OTU richness (268 ± 28 (mean ± SD) OTUs), which remained much lower than that observed in the control group (A: 401 ± 30 OTUs), however (Figure 1J, day 130; p(A vs B) = 2.2e-5, pairwise Wilcoxon rank sum test). FMT did not significantly improve this result (group C: 286 ± 14 OTUs; group D1: 284 ± 24 OTUs; group D2: 248 ± 65 OTUs; p(B vs C) = 0.09; p(B vs D1) = 0.30; p(B vs D2) = 0.63).

Alternative microbiota states

Principal coordinates analyses (PCoA) of the data from days 7, 14, 48, and 54 revealed an important effect of diet change (at day 18), as expected, which had stabilized at day 48 (not shown). Further analyses were performed on samples from day 48 onward only to focus on the effects of DSS and antibiotics treatments and subsequent recovery, with or without FMT, in the FF diet context (Figure 1I). The results of these analyses (Figure 2, detailed in Supplemental Figure 2) show that (1) the microbiota used in allogenic FMT (groups E1 and E2) clearly stand apart from the microbiota in the control group (A; day 48, Axis 3 (PCoA3, p = 0.00056 for E1 and p = 0.002 for E2; pairwise Wilcoxon rank sum tests) and for group E1, also PCoA2 (p = 0.0014)). (2) The DSS treatments caused a shift on PCoA3 (day 86, p = 0.00055), in the opposite direction of the aforementioned deviation of the allogenic donor samples, and a minor effect on PCoA1 (p = 0.00094). (3) Antibiotics treatment had a major impact on PCoA1 (day 96, p = 9.4e−09 (DSS + antibiotics vs control)) and in addition minimized variability on all axes, yielding a compact cluster comprising the microbiota of nearly all treated rats.

Figure 2.

Principal coordinates analysis. Principal coordinates analysis (PCoA) (based on Bray–Curtis distance, OTU data aggregated at the genus level) between microbiota samples from days 48, 54, 86, 96, 123, and 130. Each 3D plot shows a selection (1 d) from the complete analysis. Days 48 and 54, before DSS treatment. Day 86, after three DSS treatments. Day 96, after antibiotics treatment. Day 130, one month after FMT. Colors, experimental groups as indicated: A, control; B, DSS + antibiotics treatment; C, as B, followed by autologous FMT; D1/D2, as B, followed by allogenic FMT; E1/E2, allogenic donors. The semi-transparent grey screen illustrates that lasting separation is induced on axis 1: on days 48, 54, and 86, nearly all the microbiota are situated behind this screen (pale colors), while on days 123 and 130 part of the microbiota are situated in front of the screen (bright colors), in addition to being separated on axis 2. Rectangles I and II (day 130): projection of data points on the axis2 vs axis1 plane reveals two clearly separated clusters (alternative microbiota states; cf red rectangles in Figure 3A top right).

At days 123 and 130, about 1 month after FMT, two clusters become visible that can be regarded as alternative microbiota states (as external conditions, including diet, were the same for all rats at this stage; cf Figure 1B). Clustering was especially clear at day 130 in the PCoA1 vs. PCoA2 plot (Figures 2 and 3A top right panel: rectangles I and II). This two-dimensional PCoA-based separation into two clusters was confirmed with only minor differences by Dirichlet multinomial mixture (DMM) analysis²⁰ of the data from day 130, which also classified the two rats that were not clearly assigned to either cluster in the PCoA plot (Figure 3A, bottom right panel compared to top right panel; Figure 4A,B, Supplemental Figure 3). Hereafter, we will consider the DMM clustering, as it takes more dimensions of the microbiota into account than the two PCoA axes and results from an analysis of day 130 data only. DMM cluster I, which we will refer to as the healthy state I microbiota (be it with reduced diversity due to the fiber-free diet context), contains the microbiota from the control group (A). DMM cluster II, which we refer to as the alternative or (pre-)disease state II microbiota, as it is practically always associated with low-grade intestinal inflammation (see paragraph “alternative host states” below), contains the microbiota from the DSS and antibiotics-treated rats without FMT (group B) (Figure 4B). This cluster shows considerably lower diversity (OTU richness) than cluster I (Figure 4C). Together, these results show that a strong perturbation of the intestinal ecosystem by transient inflammation-inducing DSS and subsequent antibiotics treatments led to a microbiota state transition.

Figure 3.

Proof of concept: atypical donor microbiota can boost FMT results. (A) Partial results of the PCoA analysis shown in Figure 2. Left-hand panels, microbiota at day 96 (just before FMT). The black rectangle in the top left panel contains the untreated controls (group A, day 96) as well as the allogenic (groups E1/E2) and autologous (group C) donor microbiota (data from days 48 or 54, respectively). Right-hand panels, day 130 (one month after FMT). Top right, red rectangles indicate microbiota states I (containing untreated controls) and II defined on PCoA axes 1 and 2. Bottom right, microbiota clusters I and II as defined by Dirichlet (DMM) analysis of data from day 130. Top panels, dashed lines connect the donor microbiota, the recipient microbiota at day 96 and the recipient state at day 130 when state I is reached, only the latter two (in black) when state II is reached. Black * indicate cases where donor microbiota data are not available (green line) or where no FMT was applied (red lines). Curved plain black arrows in the top right panel indicate samples that received cluster attribution after DMM analysis (which in addition re-assigned two samples (group B, red) from cluster I to cluster II). Bottom panels, dashed lines connect the donor microbiota, recipient microbiota at day 96 and recipient microbiota at day 130 for the three DSS-treated rats with healthy colon histology at day 130 (after FMT). In the bottom left panel, the red arrow indicates the proximity of one E2 donor (yellow) to the E1 donor group (purple oval); the red rectangle indicates the position of most of the control group A microbiota; the black circle indicates the position of the strongly disturbed recipient microbiota. All panels, data point colors represent experimental groups as in Figure 2. The symbol indicates donor‒recipient combinations. The dashed lines only connect points as indicated and hold no information on microbiota trajectories in ordination space. (B) Fitting of genus abundance vectors (gradients) to the ordination plots from the bottom panels in Figure A using the envfit function from the vegan package in R. Vector length represents the correlation coefficient r. Only genera that were identified in both panels with Pearson correlation p-value < 0.01, r² ≥ 0.3 and relative abundance ≥ 0.1% in at least one sample are shown. For more genera, see Supplemental Figure 4.

Figure 4.

Alternative microbiota states. (A) Dirichlet (DMM) analysis of the day 130 data identified two clusters (two microbiota states) based on the Laplace index. (B) Distribution of experimental groups (Figure 1H) over DMM clusters CI and CII at day 130: grey, untreated control group A; red, group B (DSS + antibiotics, no FMT); green, group C (autologous FMT); blue, group D1 (allogenic FMT); orange, group D2 (allogenic FMT). The numbers indicate the number of rats. (C) Diversity of DMM clusters. (D) Differential genus abundance analysis of the two DMM clusters. Green, overrepresented in cluster I; purple, overrepresented in cluster II (>5-fold difference; q < 0.05). Log2(fold change) was defined by a linear model for differential abundance analysis (linda—MicrobiomeStat). See Supplemental Figure 6 for the corresponding heatmap. (E) Retrospective cluster assignment of microbiota from different experimental groups and time points using a classification model. The numbers indicate the number of rats. Panels B and E: Group totals can be lower than 10 (groups A, B, C) or 5 (groups D1, D2) owing to technical problems (insufficient amounts of fecal material and hence missing sequence data).

Allogenic FMT can rescue the microbiota from a (pre-)disease state destination

How did FMT affect this transition from a healthy state to a (pre-)disease-associated microbiota state? Most of the rats that underwent autologous FMT (group C) finished in the (pre-)disease microbiota state II (Figures 2 and 3A right-hand panels: state II, Figure 4B), as did most of the recipients (group D2) of allogenic microbiota from the donor group E2. For two rats in cluster I, the role of FMT, if any, in microbiota recovery is not clear, as they appeared less affected by the antibiotics treatments before FMT (Figure 3A, top left panel: rats 27 (group C) and 39 (group D2)). Thus, autologous FMT or allogenic FMT using group E2 microbiota did barely or not improve the restoration of a healthy microbiota. Allogenic FMT using the microbiota from group E1, in contrast, was remarkably successful: in five out of five recipients (group D1), a microbiota similar to that of the untreated controls (group A) was restored (Figures 2 and 3A right-hand panels: state I, Figure 4B). The PCoA analysis suggests that, at the time of FMT (day 96), the successful group E1 donor microbiota overcompensated for the difference between the healthy microbiota (control group A) and the severely altered recipient microbiota (Figure 3A, bottom left), and can, in that sense, be regarded as “healthier than healthy” donor microbiota. At the taxonomic level, this notably corresponds to gradients of the genera Phocaeicola (P. dorei) and Ruminococcus (R. bromii) (Figure 3B, Supplemental Figure 4, lefthand panels) that are also found between healthy and (pre)disease-associated microbiota states one month after FMT (day 130; Figure 3B, Supplemental Figure 4, righthand panels).

Phocaeicola was (much) more abundant in the allogenic E1 donor group (9508 ± 1900 reads (mean ± SD) on a total of 45000 reads per sample) than in allogenic donor group E2 (3623 ± 2458 reads; p(E1 vs E2) = 0.01, pairwise Wilcoxon rank sum test), autologous donor group C (22 ± 43 reads; p(E1 vs C) = 0.003), or control group A (7526 ± 1634 reads; p(E1 vs A) = 0.06) at day 96 (just before FMT; Supplemental Figure 5A). Similarly, Ruminococcus was (much) more abundant in the E1 donor group (5527 ± 2249 reads) than in donor group C (34 ± 46 reads; p(E1 vs C) = 0.003), or control group A at day 96 (2700 ± 1303 reads; p(E1 vs A) = 0.03) (Supplemental Figure 5B). A comparison with donor group E2 (3936 ± 2665 reads) showed the same tendency, but the difference was not statistically significant because of high intra-group variability (p(E1 vs E2) = 0.4).

At day 130, one month after FMT, Rumincoccus was much less abundant in the autologous FMT group C (1410 ± 2669 reads) than in group D1 (recipients of the allogenic E1 microbiota; 4551 ± 2503 reads; p(C vs D1) = 0.02) or A (control group: 3306 ± 1978 reads; p(C vs A) = 0.038). A similar tendency was observed in the comparison to group D2 (recipients of the allogenic E2 microbiota; 3359 ± 2711 reads), although the difference was not statistically significant (p(C vs D2) = 0.445). (Supplemental Figure 5B). Phocaeicola showed similar abundances in groups D1 (4158 ± 2722 reads) and control group A (5725 ± 1219 reads; p(D1 vs A) = 0.35), which were higher than in groups C (2754 ± 3509 reads) and D2 (1614 ± 3220 reads) ((Supplemental Figure 5A) although only the difference with group C was statistically significant (p(C vs A = 0.011; p(D2 vs A = 0.105).

Together, Phocaeicola and Ruminococcus made up about 20% (average) of the microbiota in the D1 FMT group (recipients of the E1 donor microbiota) and control group A at day 130, compared to about 10% in the other groups with (C, D2) or without (B) FMT.

Characterization of microbiota states

To further characterize the difference between the DMM-based microbiota states I and II, we performed a differential abundance analysis. The results summarized in Figure 4D and Supplemental Figure 6 confirm the overrepresentation of the Phocaeicola and Ruminococcus genera in the healthy state (I), together with the low-abundance genus Butyricimonas. Akkermansia (A. muciniphila), Hungatella and Phocea were among the most overrepresented genera in state II.

We then performed genus-level network analyses of the two microbiota states at day 130. A first analysis without a priori assumptions (Supplemental Figure 7) revealed a densely interconnected structure in the healthy microbiota state (I) (Pearson r ≥ 0.7, p ≤ 0.001) with a central role for an unknown genus of the Ruminococcaceae family, and very few connections in the (pre-)disease microbiota state (II). An analysis centered on the genera Phocaeicola and Ruminococcus that putatively contributed to the success of the E1 donor microbiota, Akkermansia (underrepresented in microbiota state I (Figure 4D) and nearly undetectable in E1 donors (Supplemental Figures 8 and 9)) and the abovementioned Ruminococcaceae genus (Supplemental Figure 10), revealed a strongly interconnected network structure in the healthy microbiota state (I) (Figure 5). In this network (r ≥ 0.5, p ≤ 0.05), Phocaeicola, Ruminococus and the Ruminococcaceae genus at the one hand and Akkermansia at the other make part of two opposing poles (i.e. with negative correlations). Using the same criteria, no connections between these genera were detected in the (pre-)disease microbiota state (II).

Figure 5.

Genus correlation networks. Networks centered on Phocaeicola, Ruminococcus, Akkermansia and an unknown genus of the Ruminococcaceae family in microbiota states I and II, respectively, at day 130 (Pearson r ≥ 0.5, p ≤ 0.05). Red lines, positive correlations; black lines, negative correlations. Thick lines, shortest paths.

In order to examine the evolution of the two microbiota states over time, we used the results of the DMM analysis to build a highly accurate classification model comprising 18 genera, assigning microbiota from day 130 to either state I or state II (AUC-ROC = 0.98; Supplemental Figure 11). We then used this model to retrospectively classify the microbiota from the different experimental groups at days 54, 96, 106, 116, 123, and 130 (Figure 4E). At day 54, all the microbiota were classified as state I (healthy), and the control group A microbiota remained in this state for all the following time points. At day 96, all the microbiota other than those in group A, i.e. all the microbiota from the rats treated with DSS and antibiotics, were classified as state II, while within this state occupying an extreme position according to PCoA analysis (Figure 2, Axis 1 position at day 96 compared to Axis 1 position of state II at day 130). The group B microbiota (no FMT) and most of the group C (autologous FMT) and group D2 (FMT with donors from group E2) microbiota subsequently remained in this state, while all of the group D1 microbiota returned to a state I classification, after FMT with microbiota from the allogenic E1 donors. Together, these results are coherent with the scenarios proposed in Figure 1I, showing that a transition to a (pre-)disease state is initiated by the DSS and antibiotics treatments, and corrected by FMT using allogenic E1 microbiota.

Alternative host states

The condition of the host was evaluated through histological examination of the distal colon at day 133, five weeks after FMT. Host inflammation intensity showed a bimodal distribution under identical external conditions, diet and age (Supplemental Figure 12A), indicating the existence of alternative host states. Chronic low-grade inflammation was observed in all the rats that had undergone the DSS and antibiotics treatments, followed by FMT (groups C, D) or not (group B), with three exceptions (Supplemental Figure 12B): two of the five rats in group D1 (rats 31 and 35) and one in group D2 (rat 36), all recipients of allogenic FMT, consistently showed untreated control-(group A)-like scores for each of the three discriminating histologic criteria of mononuclear cell infiltration, edema and epithelial atrophy (Supplemental Figure 12C−E), suggesting that FMT had led to the cessation of inflammation and the healing or prevention of tissue damage (note that although one rat (nr 29) in group C obtained a low overall score, this rat showed higher than normal scores for the critical criteria of cell infiltration and edema). For rats 31 and 36 the prior induction of inflammation was confirmed by the fecal calprotectin assay after the last DSS treatment (Figure 1K). For rat 35, there was no fecal material available for this analysis, but there is no reason to doubt the induction of inflammation, as higher than control calprotectin levels were detected in all the DSS-treated animals that could be tested. In rats 31 and 35, the healthy colon histology at day 133 coincided with the abovementioned recovery of a healthy microbiota (Figure 3A, bottom right: state I). Intriguingly, the microbiota of rat 36 appeared to be in the (pre-)disease state (state II), suggesting either spontaneous recovery of a non-inflammatory phenotype, independent of the microbiota (something that did not happen in any of the group B rats without FMT), or a recovery due to a transient improvement of the microbiota (the latter scenario would be coherent with the observation that the donor microbiota used closely resembled the successful E1 donors in the PCoA analysis (Figure 3A, bottom left, red arrow)).

Alternative states of the intestinal ecosystem

Above, we defined alternative microbiota states at day 130 as distinct microbiota clusters under identical external conditions (including diet and age). Alternative microbiota states can also be defined as occurring under identical host intestinal inflammatory status (i.e., only using intrinsic parameters of the intestinal ecosystem).^3,4 Figure 6A shows that there is a range of host inflammation conditions where both microbiota states can exist. The healthy microbiota state is more stable, and therefore more frequent, at low inflammation levels, and the (pre-)disease microbiota state at higher inflammation levels. These properties, schematically represented by an interrupted Z-figure where the dashed line represents the boundary between the respective basins of attraction, are characteristic of alternative stable states (cf Figure 1B). Likewise, two host states can exist under a range of microbiota conditions, and their properties are schematically described by an inverted Z-figure (Figure 6A). Together, the microbiota and host state determine the alternative states of the intestinal ecosystem (Figure 6B; cf Figure 1G). Figure 6B shows that nearly all the rats that received DSS and antibiotic treatments without FMT (group B, red) or with autologous FMT (group C, green) reached a (pre-)disease ecosystem state (in or near the red circle) by the end of our experiment. Two rats (31 and 35, both in group D1, blue) reached a healthy intestinal ecosystem state, comparable to that of the untreated controls (green circle), after allogenic FMT with donors from group E1. Some of the rats that underwent allogenic or autologous FMT were found to be in a state combining a healthy microbiota with relatively high inflammation levels (black circle, top right) or combining a deteriorated microbiota with low levels of inflammation (black circle, bottom left; two cases discussed above). Owing to their proximity to tipping points and transition folds, the latter two states are less stable (and therefore less frequent). Rats in these states are susceptible to evolve to either of the two first, more stable states after relatively small (stochastic) changes in their microbiota or inflammation status (cf Figure 1B). Collectively, these data are consistent with our conceptual model of the intestinal ecosystem, where alternative stable state properties are expected to affect the outcome of conventional (inflammation-directed) or FMT strategies in the treatment of UC (Figures 1G and 6C,D).³

Figure 6.

Intestinal ecosystem states and FMT: a model. Conceptual model describing alternative stable states of the intestinal ecosystem and implications for the clinical treatment of UC. (A) Alternative microbiota states (solid black lines, representing the median PCoA2 coordinates for the microbiota in each of two Dirichlet clusters (Figure 3A, I) (“healthy”) and II (“(pre-)disease”), at day 130) can both exist over a range of identical distal colon inflammation intensities (at day 133 (Supplemental Table 4)). The dashed black line represents the transition fold between the two basins of attraction (cf Figure 1A,B) and is arbitrarily drawn as a straight line (approximate position). Alternative host inflammation states (solid red lines, representing the median distal colon inflammation scores at day 133 of two groups in the bimodal inflammation score distribution (Supplemental Figure 12A)) can both exist over a range of identical microbiota PCoA2 values. The dashed red line represents the transition fold. (B) As in (A), but colored by experimental group as in Figure 2. The green (“healthy”) and red (“(pre-)disease”) circles represent alternative stable states of the intestinal ecosystem (holobiont states), combining healthy microbiota and host inflammation states, or alternative microbiota and host states, respectively. The basin of attraction of the healthy state extends above and to the left of the dashed lines, which represent the approximate position of transition folds between basins of attraction. The black circles represent less-stable states (closer to tipping points and transition folds), combining a healthy microbiota with a deteriorated inflammation state or a deteriorated microbiota with a healthy inflammation state, respectively. (C) Hypothetical trajectories of the intestinal ecosystem after initial perturbation and FMT. The solid blue and orange lines represent alternative stable states (attractors) of the microbiota and host, respectively. The dashed blue and orange lines represent transition folds (deflectors). The yellow dot indicates a strongly perturbed intestinal ecosystem. Yellow arrow, natural recovery towards the red attraction point (stable, chronic (pre-)disease state). Dashed black arrows, suboptimal FMT may lead to the red attraction point (alternative stable state) or the red circle (a less-stable state). From this circle, the system may move across the transition folds or the tipping points due to stochastic variations, to reach the green (healthy state) or red attraction points. The solid black arrow, optimal FMT (using “healthier than healthy” donor microbiota) may circumvent the inflammation state tipping point and directly lead to the green (healthy state) attraction point. (D) Once a reference map of microbiota vs contemporaneously collected inflammation data is available (schematically presented in Figure 6B; see figure 2C in reference³ for a real-life example in the context of pediatric UC), patients could be positioned on this map to guide treatments (emphasizing microbiota management, inflammation-directed treatments, or both, depending on the position on the map). Note that the microbiota tipping point at the bottom left of the figure is inaccessible, as it is situated at a negative inflammation value. This feature predicts that inflammation suppression alone will not always lead to the restoration of a healthy microbiota state, in agreement with clinical observations.³ Symbio-therapy (red arrows), concerted action on the host and microbiota, is predicted to require less effort on each of these parameters to ensure transition to a healthy state, without the need to reach the respective tipping points, than action on only one parameter.³ Symbio-therapy would allow to reach and pass the transition folds separating the basins of attractions of the alternative microbiota and host states, respectively (cf Figure 1A,B). From there, microbiota and inflammation level are predicted to spontaneously evolve into the basins of attraction of the respective stable states (green arrows), towards a healthy equilibrium.

Discussion

FMT has been used with variable success in the treatment of UC,⁹ where inflammation-directed care remains the approved, if imperfect,⁸ standard. Efforts to improve the efficacy of FMT, notably pursuing the search for high-performing donor microbiota, remain a matter of trial and error, with the occasional a posteriori identification of a “super-donor”.¹¹

We recently proposed a conceptual model of intestinal ecosystem behavior which predicts that even the successful engraftment of a healthy (autologous or allogenic) donor microbiota will not systematically restore a healthy intestinal ecosystem with little or no inflammation in every patient. Rather, an equilibrium is expected between patients who will regain a healthy holobiont state and those who, in spite of a healthy microbiota, will continue to show an inflammation phenotype (Figure 1G) with a risk of relapse to a stable state of mutually sustaining inflammation and altered microbiota (Figure 1D). This is exactly what we observed in our results (Figure 6B). In studies where each donor microbiota is tested on only one or few recipients, chance will therefore play an important role in the observed results of FMT when judged at the level of endoscopic remission, as usually practiced¹¹: donor microbiota that were (initially) successful in microbiota restoration may be classified as “unsuccessful”, thereby introducing a bias in the comparison of “efficient” and “non-efficient” (transfer of) donor microbiota. An evaluation at the level of the recipient's microbiota therefore is essential to advance our understanding of what constitutes a high-performing donor microbiota.

According to our model, a higher success rate in the restoration of a healthy intestinal ecosystem may be expected when using a “healthier than healthy” or atypical donor microbiota, which could in addition compensate for less than perfect microbiota transfer and engraftment due to partial (selective) loss of viability caused by aerobic exposition, freezing and thawing, or other technical factors and/or ecological challenges to colonization for strains entering an established ecosystem. The results of our present study support this prediction in a striking manner. Transient DSS and antibiotics treatments led to a transition to an alternative microbiota state. With the exception of one or two cases out of eight, autologous FMT, as a model for the use of a purportedly “ideal” subject-matched healthy microbiota, failed to change this result. In contrast, one of two allogenic donor rat strains (Figure 3A, group E1) was remarkably successful (5/5) in the restoration of a healthy microbiota (i.e., resembling the microbiota of the control group at day 130, according to DMM analysis). In agreement with our model, the successful E1 donor microbiota appeared to overcompensate for the difference between the strongly perturbed recipient microbiota and the untreated control microbiota at the time of FMT (day 96, Figure 3A, bottom left). Moreover, in two of these five cases, this led to the restoration of a healthy distal colon histology or prevention of its deterioration, thus restoring a healthy holobiont state (Figure 6B). The prediction of our model is that additional measures in the form of an anti-inflammatory treatment may further improve this result.³

The successful donor microbiota (group E1), the microbiota of their recipients (group D1) one month after FMT (day 130), and the microbiota of the control group (A) at the same time point shared elevated levels of the genera Phocaeicola and Ruminococcus, together amounting to around 20% of the microbiota at day 130 (in group D1 and control group A). More importantly, this FMT allowed the return to a healthy microbiota state characterized by a dense network of apparent inter-genus connections (as judged by covariance) that is not observed in the (pre-)disease microbiota state. This network is organized into two opposing poles, associated with Phocaeicola and Ruminococcus or with Akkermansia, respectively. While Ruminococcus bromii has been linked to various beneficial effects in humans, including through supporting the growth of other health-related taxa,²⁵ and Phocaeicola dorei improved inflammation and pathological damage of acute DSS-induced colitis in mice,²⁶ Akkermansia muciniphila can have a detrimental, inflammation promoting, effect in the fiber-free environment in which our experiment was set, through degradation of the intestinal barrier.²⁷ The (near) absence of Akkermansia in the successful donor microbiota (E1) may therefore be as important as the high abundance of Phocaeicola and Ruminococcus in shifting the microbiota equilibrium toward the Phocaeicola-Ruminococcus pole, thus creating conditions that permit the restoration of intestinal barrier integrity. Of note, Akkermansia could not be detected in the two rats (31 and 35) where a healthy holobiont state combining a healthy microbiota and non-pathological low or undetectable inflammation levels was restored (day 130; Figure 6B), while it was detected (albeit at the limit of detection) in the other FMT recipients in group D1, which still showed histological signs of inflammation.

In the present study, we used allogenic donor microbiota from rats that received the same FF diet as the autologous donors and FMT recipients. One reason, in addition to what was mentioned before, was that the establishment and maintenance of more diverse donor microbiota obtained from rats on a complete diet would likely require a complete diet after FMT, and therefore substantially complicate the experimental design to dissociate FMT and diet effects. In final-result-oriented clinical applications (as opposed to preclinical mechanistic studies), this would be less of a problem (on the contrary: a healthy diet is recommended), and including diet information in donor selection may constitute a means of obtaining higher-quality donor microbiota.

Our results provide important indications to guide and rationalize research efforts towards improvement of FMT efficacy in humans. These findings indicate the necessity to characterize the microbiota of patients and potential donors and compare these to the microbiota of a relevant reference dataset of healthy subjects, and show that, maybe counterintuitively, atypical allogenic donor microbiota can outperform autologous microbiota in the restoration of a healthy holobiont state and curation of disease. They further indicate that (initial) success of FMT should be monitored at the level of the microbiota as much as at the level of clinical symptoms.

Beyond FMT, the present results complement our earlier reported analyses of clinical UC data regarding the impact of conventional, inflammation-directed, treatment of UC. Both studies support a holistic model for disease management which can aid clinicians in the advanced diagnosis of UC and decision making about treatment strategies through simultaneous evaluation of intestinal inflammation (which, in humans can be done through the measurement of fecal calprotectin) and intestinal microbiota composition, and comparison with relevant reference data that allow interpretation of the observations. This model predicts that a combination of FMT or other forms of microbiota management and conventional therapy, symbio-therapy, would lower the requirements for each and be more effective than either treatment strategy on its own (Figure 6D).³

Supplementary Material

Van_de_Guchte_GutMicrobes_rev2_suppl_unmarked.docx

Van_de_Guchte_GutMicrobes_rev2_suppl_unmarked.docx

Van_de_Guchte_GutMicrobes_rev2_SupplFigs.pptx

Van_de_Guchte_GutMicrobes_rev2_SupplFigs.pptx

Funding Statement

This project was partly funded by the European Commission under ERC-2017-AdG n. 788191 – Homo.symbiosus, and MaaT Pharma (Lyon, France). ERC-2017-AdG n. 788191.

Supplemental material

Supplemental data for this article can be accessed at https://doi.org/10.1080/19490976.2025.2609457.

Disclosure of potential conflicts of interest

JD is co-founder and scientific advisor of MaaT Pharma. The other authors have no conflicts of interest to declare.

Data availability statement

16S rRNA gene sequence data are available in the NCBI SRA repository (accession number: PRJNA986321). All other data are included in this article and its supplementary data files.

Abbreviations

FMT

fecal microbiota transfer

UC

ulcerative colitis

DSS

dextran sodium sulfate

MHC

major histocompatibility complex