Cohesive modules of engraftment in fecal microbiota transplantation

Summary

While single-strain probiotics fail to address community-level microbiota injuries in dysbiosis-related conditions and fecal microbiota transplantation (FMT) produces unpredictable communities, a middle-ground approach has emerged. This approach involves using small consortia of species, combining the precision of single-strain probiotics and the holistic approach intrinsic to FMT. The species selection in this oligomicrobial strategy is typically proprietary or based on studies linking single species to disease or health. To advance this approach, we developed the concept of cohesive modules of engraftment (CME) and a workflow for their identification from FMT trials. CMEs represent small donor microbiota subsets that engraft as units (modularity), while maintaining their original composition (cohesiveness). In benchmarking, we identified >200 highly cohesive CMEs (2–5 species) in 5 FMT trials and found evidence for cross feeding as a mechanism for CME integrity. Due to their predictable post-treatment compositions, CMEs deserve investigation as potential ingredients of future therapeutic microbial consortia.

Graphical abstract

Highlights

• •Lack of clear guidelines for species/strain selection in therapeutic microbial consortia
• •Cohesive modules of engraftment (CME) yield predictable post-treatment compositions
• •Benchmarked workflow for CME identification in fecal microbiota transplant trials
• •Cross feeding as a mechanism driving CME integrity

immunology; precision medicine; microbiome; bioinformatics; systems biology.

Introduction

Gut microbiota disruptions have been associated with a broad spectrum of pathologies, ranging from the prototype Clostridioides difficile (C. difficile) infection¹ to diabetes,² depression,³ and cancer,⁴ motivating creative attempts to restore a healthy microbiota and thereby prevent or treat dysbiosis-related diseases. Two extreme forms of live microbiota therapeutics are single-strain probiotics and fecal microbiota transplantation (FMT). While FMT involves the transfer of a non-selective sample of the entire community of colonic microbiota, single-strain probiotics contain a specific strain of a bacterial species thought to be beneficial for the host. FMT is based on the premise that a community may perform functions that are more than a sum of the functions performed by its individual components.⁵ In a cross-feeding example, species A produces metabolite a, which cannot be utilized by the host unless converted by species B to metabolite b. A probiotic containing species A or B (but not both) will not generate metabolite b, whereas providing both species will generate metabolite b for the host. This simple 2-species example can be generalized to more complex scenarios where many species are needed for a specific beneficial effect.^6,7,8 The association of microbiota diversity loss with several diseases supports a community approach to microbiota restoration.

An alternative intermediate approach, microbial consortium therapy,⁹ has recently gained popularity. A key concept in this approach is functional redundancy,¹⁰ that is, a subset of the microbiota may provide the same functionality provided by the entire community. Compared to conventional probiotics, a rationally designed microbial consortium has the advantage of utilizing complexity to enable higher-order interactions between different species and to yield novel metabolic functions. A microbial consortium is also more precise than FMT, which is currently guided only by safety criteria.^11,12 In addition, since a substantial fraction of donor microbiota does not engraft after FMT,^13,14,15 consortium therapy can potentially decrease the debatable futility of transferring non-engrafting species. Finally, although mandatory screening has reduced pathogen transmission risks associated with FMT,¹⁶ any remaining risks related to unclassified/unknown pathogens are minimized when only a small number of well-known species are used in the product.

The main challenge in the design of an oligomicrobial consortium is the choice of its components. As microbiota inherently form a well-connected network,¹⁷ the function of specific subsets cannot be easily predicted. Limited help is provided by data from controlled animal experiments (e.g., antibiotic-treated mice inoculated with specific microbial species), or from observational human studies showing associations between certain species and disease. This is partly because if species A and B are both associated with a certain condition, their combined association with that condition is not guaranteed. A particularly problematic but common scenario occurs when associations of species A and B with the condition of interest are derived from different studies. The identity and proportion of different species, and the rationale for their inclusion in microbiota consortia, are often proprietary pharmaceutical information. Finally, the absence of defined guidelines for microbiota consortium design has hindered progress in this field.

Here, we developed a transparent operational framework to inform the design of evidence-based microbiota consortia. The only prerequisite is shotgun metagenomes from one or more FMT trials, including donor microbiota and pre- and post-FMT patient microbiota. In our “top-down” approach, the “top” is FMT and the “down” is a module, a precisely defined subset of donor microbiota that maintains its original composition (in the donor) after engraftment. This compositional cohesiveness makes post-treatment communities more predictable, enabling precision therapy. The cohesive modules of engraftment (CME) can be readily identified from FMT trials, facilitating the development of therapeutic microbial consortia.

Results

We used data from 60 (donor: patient) pairs in five FMT trials conducted in the US (three trials), Canada (one trial), and France (one trial). The underlying disease included hematologic malignancies (one trial), C. difficile infection (two trials), Crohn disease (one trial), and melanoma (one trial). A total of 29 unique donors were used across these trials.

A small fraction of donor species engraft with moderately cohesive co-engraftment

We first generated a list of all (species:patient:donor) triads, consisting of a patient, their FMT donor, and an engraftable species in this FMT procedure. These 11,108 triads included a total of 1,069 unique species. The contribution of each trial to these triads and species is shown in Table 1. Engraftment occurred in 1,484 triads, indicating an overall engraftment rate of 13.3%. The two C. difficile trials had the highest engraftment rates (Figure 2A). A total of 294 unique species engrafted in at least one triad; the most frequently engrafting taxa were an unclassified Oscillospiraceae species (23 triads), Anaerobutyricum hallii (A. hallii) (20 triads), and Dorea formicigenerans (D. formicigenerans) (20 triads) (Figure 2B).

Table 1.

Summary of previous FMT trials included in the present analysis

	Trial #1	Trial #2	Trial #3	Trial #4	Trial #5
First author	Reddi	Watson	Verma	Sokol	Davar
Country	US	Canada	US	France	US
Disease	blood cancers	C. difficile	C. difficile	Crohn	melanoma
Main endpoint	engraftment	clinical response	clinical response	engraftment	immunotherapy response
Bowel prep	-	+	+	+	+
Antibiotics before FMT	multiple typesa	PO vancomycin	multiple typesd	-	-
FMTs/patient, N	1b	1	1	1	1
FMT route	capsule	capsule or lower endoscopy	upper/lower endoscopy	lower endoscopy	lower endoscopy
FMT donors included, N	3	2	12	5	7e
Patients included, N	17	9	12	8	14
Samples included	post-Abx pre-FMT ∼1 m post-FMT	post-Abx pre-FMT ∼1 m post-FMT	pre-FMT 1–2 m post-FMT	pre-FMT 6 w post-FMT	pre-FMT 3–4 w post-FMT
(Species:patient:donor) triads includedc, N	4,166	2,074	2,205	1,412	1,251
Unique species in triads, N	431	505	778	407	345

^aAll patients were exposed to antibiotics before FMT as part of standard care. Antibiotics were not used as FMT conditioning.

^bTreatment was administered over 7 days.

^cTriads with an engraftable species.

^dAll patients were exposed to antibiotics before FMT as part of treatment for their disease. Antibiotics were not used as FMT conditioning.

^eDonors were patients with melanoma with long-term response to immunotherapy. The sum of the last row is greater than the total number of unique species in the combined dataset (1,069 triads) due to the presence of some species in triads from multiple trials. m: month; w: week.

Figure 2.

Engraftment and its cohesiveness

(A) Engraftment of donor microbiota in five FMT trials. Engraftment is defined as the presence of an engraftable species in post-FMT patient microbiota and confirmed at the strain level. To calculate engraftment, all (species:patient:donor) triads with an engraftable species were considered. The radius of each circle indicates engraftment rate. Each row shows the result for one trial and the bottom row shows the overall engraftment rate across all five trials.

(B) The most common engrafting species. Species engrafting in at least 10 cases are shown, with the three most frequent species named. The list of all species shown can be found in Table S1.

(C) Relationship between species abundances in donor microbiota and engraftment. Boxes indicate medians and interquartile ranges. Whiskers extend to 5th and 95th quantiles. The p value is from a mixed model, with subject ID nested under study ID as a random effect. ∗∗∗p < 0.001.

(D) Relationship between donor vs. post-FMT patient abundance of species in all (patient:donor:species) triads with an engrafted species. Each point is a species. The color coding indicates different trials and their corresponding regression lines from mixed models, with subject ID as a random effect. The black dashed line is the mixed model regression line for all trials combined, with subject ID nested under study ID as a random effect.

(E) Study-level overall cohesiveness of engraftment in five FMT trials. Cohesiveness was defined as the cosine similarity between donor and post-FMT vectors of species relative abundances. To determine overall cohesiveness in each trial, all engrafted species in all (patient:donor) pairs were considered together. The five trials shown in (A), (D), and (E) are Davar et al.,¹⁸ Sokol et al.,¹⁹ Verma et al.,²⁰ Watson et al.,²¹ and Reddi et al.²² FMT, fecal microbiota transplantation.

To examine the added value of our meta-analytic approach combining triads from all five trials, we examined the 33 species that engrafted in 10 or more triads and recorded the trial of origin for the triad showing engraftment. Four of these species came from triads in all five trials, 21 from triads in four trials, and eight from triads in three trials. There were no species on this list that came from triads in only one or two trials. As single species showing engraftment are the building blocks for CMEs, this observation indicates that combining trials will increase the likelihood of finding CMEs. Discovering the same CME in multiple trials supports inherent, trial-independent CME characteristics (i.e., cohesiveness and modularity).

Compared to the species that did not engraft, engrafting species had a significantly higher abundance in the donor (p < 0.001 from a mixed model with subject identifier (ID) nested in study ID as a random effect; Figure 2C). Furthermore, when all engrafted species were considered together, there was a strong correlation between species abundances in post-FMT vs. donor microbiota (p < 0.001 from a mixed model with subject ID nested in study ID as a random effect; Figure 2D). The slope of the regression line was 0.38, indicating higher abundances in donor vs. post-FMT microbiota and suggesting that the engrafted species had a higher fitness in their native donor microbiota than in the new, post-dysbiotic patient microbiota. When considering all engrafted species in each (patient:donor) pair, an overall moderate degree of cohesiveness was present (median cosine similarity 0.60, range 0.06–0.99). The overall cohesiveness per trial (range 0.47–0.66) is shown in Figure 2E.

Numerous CMEs are identifiable across FMT trials

In the next step, we examined the list of individual engrafting species to identify multi-species CMEs with a high level of cohesiveness. We investigated all combinations of N species (module size = N) that co-engrafted in at least P patients. To maximize the cohesiveness of the selected CMEs, we considered a threshold of 0.90 for median cohesiveness across triads showing co-engraftment. This procedure was iteratively performed for N = 2, 3, … to yield CMEs of different sizes.

In our search for CMEs, we fixed P at 8, requiring co-engraftment in eight or more patients. We first created a list of species meeting this criterion at the single-species level. By definition, this list includes members of all CMEs. Therefore, instead of searching 2-member combinations of all species to identify 2-species CMEs, we only considered our smaller list to reduce computation time. The smaller list included 55 species (Table S1), yielding 122 two-species CMEs (Figure 3A; Table S2), 70 three-species CMEs (Figure 3B; Table S3), 24 four-species CMEs (Figure 3C; Table S4), and 2 five-species CMEs (Figure 3D; Table S5). There were no CMEs with six or more species. CMEs identified using thresholds of 0.85 and 0.95 for C are listed in Tables S6, S7, S8, S9, S10, S11, and S12, respectively. The number of CMEs of different sizes using different cohesiveness thresholds is plotted in Figure 3E.

Figure 3.

Identification of cohesive modules of engraftment in five FMT trials

Each image shows CMEs of a given size sorted along the x axis based on the number of cases in which the CME engrafted. The color gradient indicates median cohesiveness of different CMEs among cases with engraftment. Only CMEs with a median cohesiveness >0.90 and engrafted in at least eight cases are shown. Two-, three-, four-, and five-species CMEs are shown in (A), (B), (C), and (D), respectively. Members of each CME are listed in Tables S1, S2, S3, S4, and S5. The number of CMEs for different combinations of module size (N) and cohesiveness threshold (C) are shown in (E). CME, cohesive module of engraftment

Examination of CME compositions suggests complex interactions among species and indicates emergent properties of larger CMEs

Next, we asked whether larger CMEs are simply assemblages of smaller CMEs, or whether some larger CMEs may be novelties that cannot be created by combining smaller CMEs. To perform this exploratory analysis, we first identified the five most frequent species of 2-, 3-, and 4-species CMEs (Table 2). Of note, we found CME novelties as larger CMEs were formed. An unclassified Oscillospiraceae species, A. hallii and D. formicigenerans were frequent members of both 2-species and 3-species CMEs. In addition, all 2-species combinations of these three species were among the most frequent 2-species CMEs but these three species did not form a 3-species CME together. Intriguingly, with the addition of Blautia wexlerae (B. wexlerae), they formed a frequent 4-species CME. Overall, these findings suggest that the first three bacteria do form 2-species CMEs but a presumably antagonistic relationship prevents the formation of a 3-species CME consisting of all three species, while B. wexlerae mitigates this antagonism, enabling the formation of a larger 4-species CMEl.

Table 2.

The most frequent members of 2-, 3-, and 4-species CMEs

	2-species CMEs	3-species CMEs	4-species CMEs
Species 1, N	SGB14861, 24	SGB4532, 36	SGB4532, 18
Species 2, N	SGB4532, 23	SGB14861, 21	SGB4826, 12
Species 3, N	SGB4575, 18	SGB4933, 21	SGB4575, 11
Species 4, N	SGB4826, 18	SGB4826, 20	SGB4837, 10
Species 5, N	SGB4874, 18	SGB4837, 18	SGB14861, 9

Each column lists the most frequent species present in CMEs of the mentioned size and the number of such CMEs. Species names can be found in Table S1.

Vitamin B biosynthesis and quorum sensing are enriched in CMEs

A. hallii, D. formicigenerans, B. wexlerae, and B. massiliensis were the most frequent taxa characterized at the species level among CMEs. To gain insight into the possible mechanisms governing the formation of cohesive CMEs from these species, we investigated the metabolic function of each species and whether complementary features such as cross feeding might drive CME formation. In pathway enrichment analysis using Kyoto Encyclopedia of Genes and Genomes (KEGG) terms corresponding to genes in each species (Figure 4) pantothenate (vitamin B5) and thiamine (vitamin B1) biosynthesis pathways were enriched in A. hallii and D. formicigenerans, respectively. Similarly, the “biosynthesis of cofactors” pathway responsible for biosynthesis of various vitamin B molecules was enriched in B. massiliensis. Quorum sensing was enriched in B. wexlerae.

Figure 4.

Pathway enrichment analysis

Pathway over-representation analysis in four species based on KEGG IDs corresponding to genes. The y axis indicates the pathway name, and the x axis indicates gene ratio defined as the percentage of the number of differentially abundant genes annotated in a pathway relative to the total number of genes annotated in that pathway. A higher gene ratio indicates a greater degree of enrichment. The bubble size indicates the number of genes. The significance of the enriched terms was evaluated by a hypergeometric test and a threshold of 0.05 for adjusted p values.

Engraftment of specific CMEs may influence clinical outcomes

While the present work was a methods development study with the objective of finding units of engraftment with desirable properties for inclusion in oligomicrobial therapeutics, we considered the possibility that these properties may not necessarily imply clinical efficacy. Different CMEs could be clinically beneficial, neutral, or even detrimental, and this may well depend on the disease and other clinical contexts. Given the relatively small number of patients in each study and different diseases, we opted to focus on one study and rank the CMEs according to their engraftment rate in patients with vs. without a relevant clinical outcome. For this purpose, we selected our own trial (study #1), which was the largest in the dataset and the one for which we had direct access to clinical outcomes. We chose relapse-free survival (RFS) at 1 year as the outcome. The events counting toward this outcome are relapse of the underlying hematologic disorder or death. Large previous microbiome analyses in allogeneic hematopoietic cell transplantation (alloHCT) recipients have indicated a significant association between dysbiosis and both relapse²³ and death.²⁴ We classified patients into those with favorable (RFS at 1 year; N = 13) vs. unfavorable (relapse or death by 1 year; N = 7) outcome. We then calculated the engraftment rate of each CME in each group defined as the number of patients in that group in whom the CME engrafted divided by the number of patients in that group in whom the CME was engraftable. Finally, we ranked the CMEs according to their differential engraftment between the two groups. Differential engraftment would suggest the potential for clinical efficacy, whether beneficial (i.e., greater engraftment in patients with favorable outcome) or detrimental (greater engraftment in patients with unfavorable outcome). The results of this analysis (Figure S1) identified a CME consisting of Blautia massiliensis and Lachnospira eligens as the module with the largest differential engraftment in favor of better outcomes.

Discussion

In the present work, we (1) defined CMEs, (2) developed a simple workflow to identify CMEs from FMT trials, (3) benchmarked our methods on five published FMT trials in different clinical settings, (4) identified CMEs with up to five species, and (5) found preliminary evidence for functional pathways that might drive the formation and integrity of CMEs. The list of CMEs can be screened for differential engraftment among patients with favorable vs. unfavorable outcomes to generate a subset of CMEs that are most likely to be clinically effective. Two features make CMEs desirable constituents of therapeutic consortia. First, the components of a CME engraft as a whole. Species that have high engraftment rates in isolation do not necessarily have a high co-engraftment rate (e.g., due to nutrient competition). Second, CMEs maintain their original composition in donor microbiota after engraftment. This feature increases the predictability of post-treatment communities, which is particularly important when a specific ratio of species is needed for clinical efficacy. Together, these features enhance the ecological stability of a CME. Our methods and workflow can be robustly applied across a wide range of thresholds and module sizes of interest. The cross-cohort approach to benchmarking enhances generalizability and enables potential application to a large array of clinical settings in different populations.

General frameworks have been proposed to aid in the development of microbial consortia therapeutics, using both top-down and bottom-up approaches.^9,25,26 In the top-down approach, relevant species are selected from the stool of treated and/or untreated host. These species are then further studied in animal models, functionally characterized, and eventually considered for inclusion in a consortium.²⁷ The bottom-up approach starts with assessing the function of individual species. This step is followed by combining species of designed functions to achieve community-level emergent functions.²⁸ Functional interactions between species can guide the construction of synthetic communities. Naturally prevalent metabolic interaction motifs²⁹ may be identified by combining metagenomic and species co-occurrence data, and using methods from graph theory applied to metabolic network models of each species. Alternatively, metabolite exchange models can guide the identification of metabolic interaction motifs.^30,31 Once identified, metabolic interaction motifs can be converted into dynamical systems using coupled ordinary differential equations (e.g., Lotka-Volterra-like growth models). Simulation of these systems enables understanding of species dynamics, as well as metabolic and functional outputs of the system.^32,33 Genome-scale metabolic models (GEMs) represent another powerful method to predict metabolism in larger communities resembling a natural setting using known information about individual species.^34,35 Dynamic flux balance analysis is often used for the simulation of GEMs. In this method, in each timestep, species growth rates are calculated from external metabolites from which species densities and metabolite levels are updated.^36,37,38 The general top-down idea of using data from FMT trials to discover putative components of defined microbiota consortia as therapeutic products is not novel. As an example, engraftment rate has been proposed as a selection criterion for individual species and strains.³⁹ We advanced the field by developing the concept of cohesive modularity for engraftment. Microbial modules characterized based on co-occurrence have been extensively studied in relation to health and disease.⁴⁰ CMEs are distinguished from co-occurrence modules in two aspects. First, they are generally derived from healthy individuals (stool donors used for FMT). Second, they maintain their original composition in the donor microbiota even after being transferred to the patient. These features make CMEs a distinct advance over existing concepts of microbiota modules.

Several microbiota consortium therapeutics have been developed and tested in trials, with varying levels of success.^41,42 The most successful of these products is VE303, a defined consortium of eight strains of commensal Clostridia developed for prevention of C. difficile recurrence. Results from a phase 2 randomized placebo-controlled trial were positive association engraftment with clinical response.⁴³ Metabolic effects of the product relevant for recurrence prevention included increases in short-chain fatty acids, secondary bile acids, and bile salt hydrolase genes.⁴² The percentage of each strain that was included in the product, and whether this composition was maintained in the recipient after engraftment, was not reported.

When the same strain of a species is present in the donor and pre-FMT patient microbiota, engraftment cannot be defined. As strain-level overlap between unrelated individuals is rare, we expect that most FMT trials using unrelated donors will yield some CMEs. This was clearly demonstrated in our meta-analytic approach, identifying CMEs spanning multiple trials. In previous analyses, species with a high engraftment rate and persistence post-FMT were usually commensals of the gut microbiota,^21,44,45 suggesting that at least some CMEs may be part of a shared core of healthy gut microbiota among healthy donors. The three most frequent species in our CMEs were obligate anaerobes of the Clostridia class and commensal species that are abundant in healthy adults.

Consistent with a previous meta-analysis,¹³ our analyses showed that an overall small fraction (13.3%) of donor strains engrafted. This finding does not contradict the repeatedly observed replacement of a much larger part of the post-FMT microbiota by donor-derived taxa⁴⁶ as the latter implies that the engrafting species and strains are relatively rare but highly successful in occupying recipient niches. Determinants of engraftment are incompletely understood but include both donor and patient factors.^{13,14,15,47,48} Consistent with previous analyses,^13,48 we found higher donor abundance to be a significant predictor of engraftment, reminiscent of propagule pressure in invasion ecology.⁴⁹ Therefore, when using CMEs as the backbone of potential microbiome therapeutics, increasing the microbiota dose may enhance co-engraftment of CME species and the entire module.

Determinants of co-engraftment likely include factors in addition to those regulating the engraftment of individual species. For example, cooperative nutrient utilization (e.g., cross feeding) and other functional relationships between CME species may increase co-engraftment. An elegant prior analysis suggested that FMT serves as an environmental filter, favoring populations with metabolic autonomy.²¹ With complementary functional capacities, species of such populations encode complete metabolic modules that synthesize key cellular building blocks, such as cofactors and vitamins required for cellular maintenance and growth.²¹ Some CMEs may represent metabolically autonomous modules. In contrast, antagonistic relationships would reduce not only the likelihood of co-engraftment but also the formation of a CME in the first place. The complex interactions within CMEs, and the importance of understanding the ecosystem properties of CMEs when developing oligomicrobial therapeutics, is highlighted by our finding that three species formed all possible 2-species CMEs but not a 3-species CME unless a fourth species was added. Simple ecological principles such as cooperative nutrient utilization and metabolic interactions could explain higher-order interactions regulating the formation and stability of CMEs. As an example, consider species A, B, and C, each of which can use nutrients i or j for growth and survival. Any combination of two species could form a stable CME, in which the two members would not have to compete for the same nutrient (i.e., one would use i and the other, j). However, with all three species together, at least two would compete for the same nutrition which could be prohibitive of stable co-existence. If species D provides an alternative nutrient for at least one of these species, then all four species could co-exist in a CME. B. wexlerae (species D in our identified 4-species CME) has perfect metabolic characteristics to play a “linker” role. This species is known to cross feed many members of the commensal gut microbiota by providing them with substrates such as lactate and acetate.⁵⁰ On the other hand, A. hallii (species A in our 4-species CME) ferments lactate and acetate into butyrate.^46,51 Therefore, B. wexlerae could make A. hallii independent of nutrients i and j used by species B and C.

Our functional analysis further supported a key role of cross feeding in the formation and integrity of some of the most frequently identified CMEs. Specifically, vitamin B biosynthetic pathways enriched in species such as A. hallii, D. formicigenerans, and B. massiliensis enable them to provide important cofactors (e.g., thiamine and pantothenate) for various critical metabolic pathways (e.g., butyrate production) to auxotrophic species vis cross feeding.^52,53,54,55 Vitamin auxotrophy is a driver of cooperation in microbial communities.⁵⁶ B. wexlerae, another frequent member of our CMEs, harbors quorum sensing pathways, suggesting its regulatory role in group-level synchrony and cooperative behavior, resulting in increased overall fitness of the CME.⁵⁷ Although confirmatory mechanistic studies are needed, these preliminary findings indicate cooperative nutrient utilization and quorum sensing as two drivers of CME formation and integrity. Metabolic dependencies underlying cross feeding interactions are known to drive species co-occurrence in microbial communities, especially under nutrient-poor conditions. The most frequently exchanged metabolites are amino acids and sugars.⁵⁸ As co-occurrence is a prerequisite for cohesiveness, cross feeding is a plausible mechanism regulating CME formation. Colonization resistance of the host microbiota is largely dependent on nutrient blocking,⁵⁹ with host microbes occupying nutrient niches of the gut and, through priority effects, making them less readily available to new species. Cross feeding and other cooperative ecological interactions within CMEs enable them to more efficiently use the limited resources in the face of competition with host microbiota. The concept of cohesiveness that we have developed in the setting of FMT and microbiota engraftment is somewhat similar to guilds defined in the non-FMT setting and using correlation-based co-occurrence.^60,61 Members of a guild grow and diminish together, due to their similar exploitations of the same class of environmental resources. Guilds tend to specialize in unique metabolic functions.⁶² The relationship between guilds and CMEs is an interesting topic for future research.

In conclusion, we provide a transparent operational framework to identify putative members of oligomicrobial consortia from shotgun metagenomic data of FMT trials. We successfully benchmarked the proposed methods on five published FMT trials, identifying >100 CMEs. We expect this work to accelerate progress in the field and pave the path for novel precision microbiota therapeutics. Future work should examine functional and ecological properties that make certain species capable of forming a CME, determinants of co-engraftment, and modular functionalities required for a CME to be an effective therapeutic for a given pathology.

Limitations of the study

This study has some limitations. First, although modularity and cohesiveness are desirable properties of CMEs as potential building blocks of microbiota consortium therapeutics, we consider the translational potential of the CMEs preliminary and the observed clinical efficacy signals exploratory and hypothesis-generating. Specifically, possible incorporation of the CMEs into precision oligomicrobial therapeutics requires further mechanistic research in the following important areas: (1) biological interactions among members of the CME, both in donor and post-treatment microbiota, (2) metabolic output of the CME, and (3) CME interactions with the host (e.g., immune system, gut barrier). Since the only source of data in present study was publicly available metagenomes from previous FMT trials, functional/mechanistic information was indirect and limited. Similarly, a microbiota consortium may consist of a single CME or several CMEs with complementary biology. The analysis of higher order interactions among CMEs is beyond the scope of the present study but is likely important for mechanistic understanding of the function of more complex consortia. Second, CMEs might have some level of disease specificity as the patient’s gut microenvironment and immune system could influence engraftment and relative fitness of different modules. Disease heterogeneity and modest sample size of previous trials limited the analysis of clinical efficacy in the present analysis. Third, for strain-level analysis, we used SameStr, which was originally developed and benchmarked for application in FMT trials and engraftment analysis.^13,44 Our algorithm and definitions do not require characterization of the specific strains. Nonetheless, it is possible that different methods yield somewhat different CMEs due to different algorithms used to define strain overlap. We recommend methods with a track record in the analysis of FMT trials.^14,15,48 Finally, diet, climate, lifestyle, ethnicity, and urbanization can affect the ecology of the human gut microbiota,^63,64,65 potentially influencing the specific CMEs pertinent to different diseases in different populations. We did not find an eligible trial in a non-Western patient population for inclusion in our benchmarking dataset. Although we believe the workflow is generalizable, a focused application of our methods to a future Asian trial would be informative to the generalizability of the discovered CMEs.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Armin Rashidi (arashidi@fredhutch.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •This paper analyzes existing, publicly available data, accessible at the NCBI Sequence Read Archive, BioProject IDs PRJNA1209112 [trial #1], PRJNA701961 [trial #2], PRJNA705895 [trial #3], PRJNA625520 [trial #4], and PRJNA672867 [trial #5].
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Acknowledgments

We acknowledge Dr. Alexander Khoruts for providing critical intellectual input and Dr. Deborah Banker for language editing. This work was supported by a grant from the Leukemia & Lymphoma Society (Academic Clinical Trials award #ACT9016-24) to A.R. and the National Institutes of Health (NIH) award P30 CA015704 to the Fred Hutch/University of Washington/Seattle Children’s Cancer Consortium, and NIH awards S10-OD-020069 and S10-OD-028685 to Fred Hutch Scientific Computing. The funders did not have any role in data collection, interpretation, or reporting. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Author contributions

A.R., investigation, visualization, data curation, formal analysis, and writing – original draft; S.S.M., and D.P., formal analysis; G.R.H. and S.J.L., writing – review and editing.

Declaration of interests

A.R. is an inventor on a patent application by Fred Hutchinson Cancer Center related to the methods and results presented in this work entitled “Cohesive modules of engraftment in FMT.”

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Raw sequencing files for trial #1	NCBI Sequence Read Archive	BioProject ID PRJNA1209112
Raw sequencing files for trial #2	NCBI Sequence Read Archive	BioProject ID PRJNA701961
Raw sequencing files for trial #3	NCBI Sequence Read Archive	BioProject ID PRJNA705895
Raw sequencing files for trial #4	NCBI Sequence Read Archive	BioProject ID PRJNA625520
Raw sequencing files for trial #5	NCBI Sequence Read Archive	BioProject ID PRJNA672867
Software and algorithms
R Statistical Software, version 4.2.0	R Core Team (2025). https://www.R-project.org/	RRID:SCR_001905
KneadData, version 0.12.0	Huttenhower Lab	https://huttenhower.sph.harvard.edu/kneaddata/
MetaPhlAn 4	Huttenhower Lab	RRID:SCR_004915
SameStr, version 3	Podlesny et al., 2022	https://github.com/danielpodlesny/SameStr
clusterProfiler v4.14.3	Yu et al., 2012 (https://bioconductor.org/packages/clusterProfiler)	RRID:SCR_016884
nlme R package, version 3.1-169	Pinheiro et al., 2026 (https://CRAN.R-project.org/package=nlme)	RRID:SCR_015655

Method details

Definitions

• iEngraftable species: A microorganism present in donor microbiota but absent in pre-FMT patient microbiota. To ascertain absence in pre-FMT patient microbiota, we use strain-level data. As most shared species among unrelated individuals represent different strains,⁶⁶ most species shared between an unrelated donor and the recipient represent different strains and are thus engraftable. Using this definition, we can identify all (patient:donor) pairs in which a given species is engraftable in an FMT trial (Figure 1).
• iiEngraftment, co-engraftment, and modules of engraftment: Engraftment is defined as the presence of an engraftable species in post-FMT patient microbiota and confirmed at the strain level. Co-engraftment is defined as the engraftment of multiple engraftable species in the same patient. A module of engraftment is defined as a consortium of co-engrafted species.
• iiiModular composition: We define the composition of a module of engraftment as the relative ratios of its species abundances in the microbiota. As an example, consider a module consisting of 3 species with relative abundances r1, r2, and r3 in the microbiota. The modular composition of this module can be defined by 2 non-redundant terms r1/r2 and r1/r3. This is a familiar concept because it forms the basis of centered log-ratio transformation in compositional data analysis.⁶⁷
• ivCohesiveness: The similarity between an engrafted module’s composition in post-FMT patient microbiota and donor microbiota. We measure cohesiveness using cosine similarity, which is the cosine of the angle between two N-dimensional vectors. Here, N is the number of species in the module. The coordinates of the 2 vectors, one representing the donor and the other one representing the patient, are the relative abundances of the N species in donor and post-FMT patient microbiota, respectively. Cosine similarity, ranging between −1 and 1, shows how much the vectors point in the same direction, regardless of their magnitude. As relative abundances are non-negative numbers, the minimum cosine similarly in our workflow is 0. A cosine similarity of 1 indicates that the species of the module maintain their original abundance ratios in the donor after engraftment.
• vCohesive module of engraftment (CME): A module of engraftment with cohesiveness greater than a user-defined threshold, C. Larger values of C enhance the predictability of the composition of the engrafted module post-treatment.

Figure 1.

Definitions

(A) Species types in the FMT product based on their outcomes after FMT.

(B) A module of engraftment with three species.

(C) Cohesiveness. Two modules of engraftment with different levels of cohesiveness are shown. Cohesiveness is defined as the cosine of the angle between the vectors of relative abundances of the module’s species in the donor vs. post-FMT patient microbiota. FMT, fecal microbiota transplantation.

CME search algorithm

Using these definitions, a simple search algorithm can identify all CMEs in an FMT trial. Two parameters will be used: (i) cohesiveness threshold, C, with cohesiveness defined as above. This parameter determines the minimum acceptable cohesiveness for a module to be defined as a CME. (ii) module size, N. This parameter indicates the number of species within a CME. The user chooses the value of C according to their desired cohesiveness threshold, keeping in mind that larger values will yield more cohesive CMEs and more predictable post-treatment communities. For demonstration purposes, we used a high threshold (0.90) for C which would be biologically justifiable. As a point of comparison, a less strict threshold of 0.85 would accept CMEs that intuitively appear non-cohesive as exemplified by the following case. The cohesiveness of a 2-species module with species abundance of r and 4 × r in the donor vs. r and r in the patient post-treatment would be ∼0.86; thus this module would be accepted as a CME if one used a threshold of 0.85 for C.

We first construct a list of all (species:patient:donor) triads consisting of an engraftable species for a specific patient and their FMT donor. A search of the 2D grid for N and C, starting with N = 2 and a reasonably high initial value for C (e.g., C = 0.90), should then be performed. The number of CMEs of a given size quickly declines and reaches zero with increasing values of C. Also, there is a larger number of smaller CMEs than larger ones for a given value of C. Therefore, only a small fraction of the 2D space will need to be searched. This simple CME search algorithm can be applied to a collection of FMT procedures in a single trial or multiple trials.

FMT trials used for benchmarking

We applied the methods described above to 5 published FMT trials (Table 1). The trials were chosen based on the availability of raw shotgun metagenomes from donor microbiota and pre- and post-FMT patient microbiota, and clear labeling of donor and patient samples, pre-vs. post-FMT timepoints, and (patient:donor) links. Post-FMT samples were selected from 3 to 8 weeks post-FMT during which microbiota engraftment typically reaches a plateau. Trials that administered FMT at several timepoints but did not include samples 3–8 weeks after the first FMT were not included. While different clinical settings might reveal different CMEs, we included trials in a wide range of diseases as we considered it advantageous for benchmarking of our workflow at this methods development stage. In addition, similar to the case of single species where disease-independent factors (e.g., abundance in donor, pre-FMT alpha diversity) regulate engraftment,^13,14,15 we considered that at least some determinants of cohesive modular engraftment might be disease-invariant.

Trial #1, FMT to prevent acute graft-versus-host disease (aGVHD) following allogeneic hematopoietic cell transplantation (alloHCT)²²: This work, conducted in the US, was the recently completed single-arm run-in phase of our randomized placebo-controlled trial of FMT to prevent aGVHD following alloHCT. Patients in the run-in phase were adults undergoing alloHCT for the treatment of hematologic disorders (all but 1 were cancer). Due to their profound immunosuppression early after alloHCT, all patients were exposed to antibacterial antibiotics (up to 5 different classes) for prevention and treatment of post-transplant infections before initiating FMT. No additional antibiotics or bowel regimen were used for FMT conditioning. FMT was administered as oral capsules for 7 consecutive days. Three healthy stool donors were used; product from each donor was administered to 4–8 consecutive patients (total of 20 patients). Stool samples were collected at baseline (before the initiation of alloHCT conditioning), pre-FMT (shortly before the first dose of FMT), and post-FMT (4 ± 1 week after the last dose of FMT). The primary endpoint was donor microbiota engraftment. Data from the 3 donors and all 17 patients who provided both pre- and post-FMT stool samples and were not exposed to antibacterial antibiotics between pre-FMT and post-FMT samples were included in the present analysis.

Trial #2, FMT to treat recurrent C. difficile infection^21,68: In this trial conducted in Canada, 10 patients with recurrent C. difficile infection received FMT, each from one of the two healthy donors. A single treatment was administered in the capsule form or colonoscopically after a 10-day course of oral vancomycin conditioning and a bowel regimen. Stool samples were collected at baseline (post-antibiotic, pre-FMT) and post-FMT (1 week, 4 weeks, 3 months, 6 months, and 1 year). The primary endpoint was clinical response. Data from both donors and the 9 patients who provided both baseline and 4-week post-FMT samples were included in the present analysis.

Trial #3: FMT to treat recurrent C. difficile infection²⁰: In this trial conducted in the US, 22 adults with recurrent C. difficile infection received a single administration of healthy donor FMT partially via upper and partially lower endoscopy, after a bowel regimen. All patients were exposed to antibiotics before FMT for the treatment of their infection. No additional antibiotics were used for FMT conditioning. Pre-FMT samples and serial post-FMT samples until ∼1 year post-FMT were collected. The primary endpoint was clinical response. Data from 12 donors and all 12 patients who provided both a pre-FMT sample and a sample 1–2 months after FMT were included in the present analysis.

Trial #4, FMT to prevent flare in Crohn’s disease^19,69: In this trial conducted in France, adults with Crohn’s disease achieving a remission using corticosteroids received a single colonoscopic administration of FMT after a bowel regimen, with no antibiotic conditioning. Stool samples were collected before FMT and at weeks 2, 6, 10, 14, 18, and 24 post-FMT. The primary endpoint was engraftment of donor microbiota at 6 weeks. Data from 5 donors and all 8 patients who provided both pre-FMT and 6-week post-FMT samples were included in the present analysis.

Trial #5, FMT to improve immunotherapy response in refractory melanoma¹⁸: In this trial conducted in the US, patients with PD-1-refractory metastatic melanoma received an anti-PD-1 immunotherapeutic agent (pembrolizumab) plus a single administration of FMT from melanoma patients with long-term response to immunotherapy. FMT was administered colonoscopically and after a bowel regimen, with no antibiotic conditioning. The primary endpoint was clinical response. Data from 7 donors and all 14 patients who provided both a pre-FMT sample and a 3- or 4-week post-FMT sample were included in the present analysis.

Analysis of sequencing data

Unprocessed fastq files from all trials were downloaded (NCBI BioProject IDs PRJNA1209112 [trial #1], PRJNA701961 [trial #2], PRJNA705895 [trial #3], PRJNA625520 [trial #4], and PRJNA672867 [trial #5]). Raw reads were first quality-processed and host-decontaminated with KneadData version 0.12.0, then inputted to MetaPhlAn 4 for species assignment.⁷⁰ For strain-level analysis, we used SameStr⁴⁴ which leverages MetaPhlAn’s clade-specific markers to resolve within-species phylogenetic sequence variations. SameStr operates in 3 steps: First, MetaPhlAn marker alignments are converted to single nucleotide variant profiles. These profiles are then filtered, merged, and compared between samples based on the maximum variant profile similarity (MVS) to detect strains that were shared between samples. Shared strains are called if species alignments between samples overlapped by ≥ 5 kb and with an MVS of ≥99.9% (default values in SameStr).

Pathway enrichment analysis

To understand unique or enriched metabolic functions of the species forming a CME, we performed pathway enrichment analysis in clusterProfiler v4.14.3 (enricher function) in R.⁷¹ Kyoto Encyclopedia of Genes and Genomes (KEGG)⁷² corresponding to the genes of the CME species, Benjamini-Hochberg adjustment of the p values for multiple testing, and a threshold of 0.05 for adjusted p values were used in over-representation analysis. This method uses a default hypergeometric test to identify significantly enriched pathways in the dataset and estimate their corresponding p values.

Quantification and statistical analysis

Continuous variables are presented as medians (ranges). The correlation between species abundances in donor vs. post-FMT patient microbiota was determined using mixed models (nlme package version 3.1-169), with subject ID nested under study ID as a random effect and the MLE-based inference method to estimate the p values. A similar model was used to determine the association of engraftment with species abundances in the donor. All analyses were conducted using R version 4.2.0 with statistical significance thresholds set at ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001.