Gut microbiota as a potential predictor of therapeutic response in adults with Crohn’s disease: A systematic review

Abstract

BACKGROUND

Various therapeutic options are available for the treatment of Crohn’s disease (CD). About 30%-40% patients experience primary non-response, and 20%-30% secondary loss of response to biological therapy. Predicting therapeutic response is challenging and an area of active research. Gut microbiota has emerged as an important player in the pathogenesis of CD and also appears to be a promising biomarker for predicting therapeutic response.

AIM

To systematically review the literature on the current status of gut microbiota as a tool to predict response to treatment in adults with CD.

METHODS

We searched the literature database (PubMed, Scopus, and Cochrane database) from inception to August 2025. We screened for studies reporting on adult patients with CD receiving biologic or immunomodulator therapies, with baseline microbiome analyses performed prior to treatment. Papers reporting on baseline gut microbiota as a predictor of therapeutic response were finally included. The utility of bacterial diversity, microbial community structure, and the role of specific operational taxonomic units as biomarkers of therapeutic response was reviewed. The results were grouped based on the bacterial parameters studied and presented in separate tables. The quality of the included studies was assessed using the MINORS criteria. The review was registered prospectively in PROSPERO.

RESULTS

After applying the selection criteria, sixteen studies were included in this systematic review. The majority of the papers were from Europe and the United States. All except two papers assessed gut bacterial population using 16S rRNA gene sequencing. Ten of the sixteen studies were of high quality. Among the sixteen studies included, most identified an association between microbial taxa and treatment response, while the relation with alpha-diversity was inconsistent. The functional characteristics were reported in only four studies and were found to be useful. The best prediction was achieved when microbial characteristics were combined with clinical and other parameters, with area under the curve values up to 0.96.

CONCLUSION

The overall results suggest good performance of microbial parameters as a novel biomarker of therapeutic response. However, there are variations across individual studies, probably related to the methodology of assessing microbial communities and the therapeutic agent used. Future multicenter studies integrating microbial, clinical, and metabolomic data are warranted to develop predictive models for personalized therapy in CD.

Key Words: Gut microbiota; Crohn’s disease; Therapy; Predictor; Diversity; Bacteria

Core Tip: Crohn’s disease is a chronic inflammatory bowel condition requiring lifelong therapy. However, about one-third of patients do not respond to any particular treatment, and a change of therapy is needed. There is a need for identifying predictors of therapy before treatment to avoid non-response, but such markers are uncommon. Accumulating evidence suggests that the gut bacterial population can serve as a potentially useful prediction tool. In this review, we have presented the current status of the role of gut microbial community as a biomarker of response. The results are encouraging but further work is needed before they can be used in clinical practice.

INTRODUCTION

Crohn’s disease (CD) is a chronic inflammatory condition of the gastrointestinal tract, and its exact pathogenesis remains to be elucidated[1]. Globally, inflammatory bowel disease (IBD) is estimated to affect 4.9 million people with an age-standardized prevalence of 59.25 per 100000 people[2]. The incidence and prevalence of IBD, including CD, vary globally, with higher prevalence in Europe and North America[3]. The data on the prevalence of CD from Asian countries (up to 25 per 100000 persons) is limited, while those in Europe and North America generally range from 50-135 per 100000 persons[3]. At the onset of the disease, most patients are in their second to fourth decade of life, and in the absence of a medical cure, its impact on physical and social functioning may be quite significant[1]. The therapeutic armamentarium has expanded at a good pace in the past 20-25 years, mainly driven by better insights into the disease pathogenesis. Yet, sustained response to most therapies is capped at about 50%-60%[4-6]. Reasons for non-response include genetic variability, differences in disease phenotype and expression of drug targets, and modulation of immune response by gut microbiota[7]. Non-response to therapy results not only in a delay in achieving disease control but also in exposure to the side effects of the drug and cost. Hence, there is a pressing need to identify predictors of response to therapy to have a patient-centric rather than disease-centric approach[4].

There are several categories of predictors of response to therapy. These include demographic variables, disease characteristics (extent, behaviour, severity, presence of extra-intestinal manifestations, etc.), biochemical parameters (C-reactive protein, fecal calprotectin, etc.) and genetic factors[4,8]. The use of clinical data and inflammatory markers generally have an accuracy of about 60%-65% which is not ideal for clinical practice[9,10]. More recently, with the growing recognition of gut microbiota as an important player in the pathogenesis of IBD, microbial-based predictors have been investigated[11,12]. Gut microbiota has been directly linked to the pathogenesis of CD, with the majority of studies reporting a reduced bacterial diversity and shift in bacterial composition/function towards a pro-inflammatory state[13,14]. Microbial-based predictors have already found application in the field of oncology, and it is hoped that they are useful for IBD as well[15]. IBD is a heterogeneous entity, and the two main types are CD and ulcerative colitis. In this systematic review, we discuss the role of the gut bacterial population as a predictor of response to therapy. Previous reviews have focused on both ulcerative colitis and CD, and the last review had data up to July 2022[12,16]. As microbiome research is rapidly expanding, with more publications being added over time, it is essential to conduct a fresh systematic review. Pediatric-onset CD has differences compared to adult-onset disease. We have provided an up-to-date review and focused on adult patients with CD to keep the group more homogeneous and make appropriate inferences. We have reviewed various attributes of the gut bacterial population, including their diversity, composition, function, and metabolic profile.

MATERIALS AND METHODS

Literature search

We searched articles in the National Library of Medicine database (PubMed), Scopus database, and Cochrane Database from inception up to August 2025. The search words included CD, therapy, treatment, Infliximab, Adalimumab, Vedolizumab, Ustekinumab, Azathioprine, biological therapy, methotrexate, mercaptopurine, immunomodulator, steroid, response to therapy, gut microbiota, predictor, association, determinant, metagenomics, pharmacomicrobiomics, precision medicine, microbial diversity, microbial function and microbial composition. MesH terms were used where available. The detailed search strategy is mentioned in the Supplementary material.

Inclusion criteria, exclusion criteria, and data extraction

Inclusion criteria: Adults with active CD. Microbial predictors evaluated in samples collected before starting therapy. Follow-up interval clearly stated. Treatment response defined.

Exclusion criteria: Case series and case reports. Narrative reviews. Animal studies. Studies in the pediatric population. Manuscripts in non-English language. Assessment of gut microbial population by predetermined selection of bacterial taxa using microarray, qPCR, etc. Evaluating non-pharmacological therapy such as surgery, fecal microbiota transplantation, etc. CD in remission at baseline microbiota sample collection.

Data extraction: The titles, abstracts, and full texts were independently reviewed by two authors for inclusion, and any disagreement was resolved by consensus. The data from the included studies were extracted on a predesigned standard proforma. The information extracted included title and year of publication, study design, study period, Country of study, details of study participants (number, age, gender), sample size, method of evaluating gut microbiota, therapy used, criteria for response to therapy, and other relevant information. The outcomes assessed were the roles of microbial diversity, structure, and function in predicting response to therapy. The response to therapy can be assessed in several ways, including resolution of clinical symptoms or reduction in disease activity scores, improvement in biochemical markers such as C reactive protein (CRP) and fecal calprotectin, endoscopic healing, or histopathological improvement. We noted the details of assessing response to therapy for each individual study.

Assessment of study quality

The studies in this review do not have a comparison group at baseline, unlike a cohort or case-control study, as all patients received therapy for CD. There are limited quality assessment tools for a single-arm observational study. Hence, we assessed the quality of included articles by MINORS criteria, in which nine out of 12 criteria are suitable for a single-arm study[17]. We assessed the studies based on nine parameters, including statement of aim, method of recruitment of patients and data collection, estimation of sample size, appropriate duration of follow-up, adequate and properly assessed endpoints, and adequate statistical analyses. The study quality was considered high for scores ranging from 7-9, good for scores 5 or 6, and poor if it was below 5. We did not assess the interrater reliability.

Data synthesis

As there was significant heterogeneity in the types of outcomes and the methods of assessing predictors as well as outcomes, only qualitative synthesis was performed, and meta-analysis was not feasible. The systematic review was prospectively registered in PROSPERO (PROSPERO 2025 CRD420251083427).

RESULTS

Search outcomes

Figure 1 shows the PRISMA literature search diagram. The search from the three databases identified 1266 titles. After removing 113 duplicates, 1153 titles were identified. After initial screening of the title and abstract, 1099 were excluded for various reasons mentioned in Figure 1. The remaining 54 papers were reviewed in detail, and 38 were excluded after full-text review. In twelve of these, no predictions were made based on the baseline microbial composition[18-29]. Seven papers evaluated non-pharmacological interventions such as fecal microbiota transplantation, exclusive enteral nutrition, etc.[30-36]. The bacterial population studied was predetermined in four papers, and five did not evaluate direct therapeutic outcomes[37-45]. In three publications on IBD, separate results for CD were not provided, and four reports were from the pediatric population[46-52]. Seong et al[53] studied patients in remission at baseline and Buffet-Bataillon et al[54] evaluated predictors of relapse. One paper studied data from external source[55]. The remaining sixteen papers were included for the final systematic review[9,10,56-69].

Figure 1  PRISMA flow diagram for the selection of studies.

Characteristics of the included studies

Sixteen studies were included. The characteristics of the included studies are shown in Table 1. The majority of the papers were from Europe and the United States. Out of the four studies from Asia, three were from China. The data collection was done prospectively in fourteen studies and retrospectively in two. The number of participants was quite small in many of these studies, and only five had more than fifty subjects. The microbial population was usually evaluated in the stool sample using 16S rRNA gene sequencing. Two studies used deeper sequencing in stool samples instead of 16S rRNA[57,63]. Both these studies included patients from the prospective IBD cohort at Massachusetts General Hospital, United States. Sanchis-Artero used 16S rRNA for identifying bacterial taxa in stool and, in addition, used Droplet Digital PCR system for absolute quantification of Faecalibacterium prausnitzii and Escherichia coli[66]. Only two papers investigated the mucosal microbial population[60,68]. In addition to bacteria, two papers reported on the fungal community in stool samples using the internal transcribed spacer gene sequencing[62,67].

Table 1 Characteristics of the studies included in the systematic review.

Ref.	Study design patient details	Sample and microbiota assessment technique	Therapy given, definition of response, duration of follow-up
Studies using TNF inhibitors
Chen et al[58], 2022, China	Retrospective study. n = 8. Age > 18 years	Stool sample. 16S rRNA gene sequencing	Adalimumab. Remission: CDAI < 150. 12 weeks
Park et al[64], 2022, Korea	Prospective study. n = 10. Mean age 31 years	Stool and saliva sample. 16S rRNA gene sequencing	Infliximab. Adalimumab. Response: Decrease in CDAI > 70 points or Remission: CDAI < 150 points. 3 months
Sanchis-Artero et al[66], 2021, Spain	Prospective multicenter study. n = 27. Mean age 41.4 ± 17.4 years. Men 55.6%	Stool sample. 16S rRNA gene sequencing	Infliximab. Adalimumab. Response assessment by Clinical, laboratory, stool tests. 3 months and 6 months
Yilmaz et al[68], 2019, Switzerland	Retrospective study. n = 270	Mucosal tissue sample. 16S rRNA gene sequencing	Anti TNF drugs. Response: Clinical and biochemical. Up to 9 years
Aden et al[56], 2019, Germany	Prospective study. n = 18	Stool sample. 16S rRNA gene sequencing	Anti TNF drugs. Response: HBI decrease ≥ 2. Remission: HBI ≤ 4. 2, 6, 14, and 30 weeks
Ding et al[59], 2020, United Kingdom	Prospective study. n = 76. Mean age 38.49 ± 14.6 years. Males 47%	Stool sample. 16S rRNA gene sequencing	Infliximab. Adalimumab. Response: Clinical indices and objective markers including FC, CRP, and radiological or endoscopic improvement. 3 monthly follow-up to 16 months
Guo et al[62], 2024, China	Prospective study. n = 28	Stool sample. 16S rRNA gene and ITS2 gene sequencing	Infliximab. Remission based on symptoms and HBI. 14 weeks
Dovrolis et al[60], 2020, Greece	Prospective study. n = 10	Mucosal tissue sample. 16S rRNA gene sequencing	Infliximab. Complete responder: HBI < 4, normal CRP and mucosal healing. 12-20 weeks
Ribaldone et al[65], 2019, Italy	Prospective study. n = 20. Median age 52.5 (26–69) years. Males: 12 (60%)	Stool sample. 16S rRNA gene sequencing	Adalimumab. Response: Decrease in the HBI score > 2 or HBI 4, 6 months
Ventin-Holmberg et al[67], 2021, Finland	Prospective study. CD-25	Stool sample. 16S rRNA gene and ITS1 gene sequencing	Infliximab. 12 weeks
Zhou et al[10], 2018, China	Prospective study. CD-16	Stool sample. 16S rRNA gene sequencing	Infliximab. Clinical response: ≥ 70-point reduction in CDAI. Clinical remission: CDAI < 150. 30 weeks
Study using integrin inhibitors
Ananthakrishnan et al[57], 2017, United States	Prospective. n = 47	Stool sample. Shotgun metagenome sequencing	Vedolizumab. Response: HBI < 4 or reduction in HBI by ≥ 3 points. 14 weeks
Study using interleukin inhibitor
Doherty et al[9], 2018, United States	Prospective study. n = 232. Mean age 38 ± 13 years. Males 36.6%	Stool sample. 16S rRNA gene sequencing	Ustekinumab. Response: Decrease in CDAI > 100 points or Remission: CDAI < 150 points, 6 weeks
Study using drugs from multiple class
Effenberger et al[61], 2021, Austria and Germany	Prospective study. n = 43	Stool sample. 16S rRNA gene sequencing	Azathioprine. Anti-TNF inhibitors. Remission: CDAI < 150. 12 and 30 weeks
Lee et al[63], 2021, United States	Prospective study. n = 108	Stool sample. Metagenomic sequencing	Anti cytokine (Anti TNF, Ustekinumab). Anti integrin (Vedolizumab). Response: Clinical and endoscopic, 14 weeks and 52 weeks
Caenepeel et al[69], 2024, Belgium	Prospective study. CD-203	Stool sample. 16S rRNA gene sequencing	Infliximab, Adalimumab, Golimumab, Vedolizumab, Ustekinumab. Response: Endoscopic remission SES CD < 3 or absence of ulceration, 24 weeks

TNF: Tumour necrosis factor; CD: Crohn’s disease; CDAI: Crohn’s disease activity index; HBI: Harvey Bradshaw index; FC: Fecal calprotectin; CRP: C-reactive protein; ITS: Internal transcribed spacer; SES: Simple endoscopic score.

All patients were treated with biological agents, commonly anti-tumor necrosis factor-alpha (TNF-α), except for one study where both azathioprine and infliximab were used[59]. Biological drugs included infliximab, adalimumab, vedolizumab, golimumab and ustekinumab. The follow-up interval mostly ranged from 3 months to 6 months, although some papers reported longer follow-up. Doherty et al[9] assessed treatment response at 6 weeks after therapy. The response to therapy was assessed by various means, including reduction in CD activity index (CDAI), reduction in Harvey Bradshaw Index (HBI), improvement in biochemical parameters, improvement in simple endoscopic score, endoscopic mucosal healing, and radiological improvement. Among these, CDAI and HBI were most commonly used.

Study quality

The details of study quality using MINORS criteria are shown in the Supplementary Table 1. The formal estimation of the sample size of patients in whom gut microbiota at baseline was assessed was not stated in any of the studies. The endpoints were fairly well defined in the included studies and were estimated with suitable statistical methods. Overall, ten of the sixteen studies were of high quality[56,57,59,61-66,69]. The remaining six studies were of moderate quality[9,10,58,60,67,68].

Bacterial diversity as a predictor of therapeutic response

Alpha diversity refers to the richness and abundance of bacterial taxa in an individual patient. There are different methods to assess alpha diversity. The utility of bacterial alpha and beta diversity at baseline is shown in Table 2. Ananthakrishnan et al[57] found a higher alpha diversity (Fisher’s alpha) at baseline in patients who subsequently responded to vedolizumab therapy. Lee et al[63] found the responders to anti-cytokine therapy to have significantly higher indices of microbial richness at baseline compared to non-responders. They both used a deep sequencing method on the stool sample. The other four studies reporting on the utility of baseline alpha diversity used the 16S rRNA gene sequencing method in stool, and none found any difference between responders and non-responders[9,61,62,68]. Overall, alpha diversity was not very useful as a predictor of response to therapy when assessed by 16S rRNA gene sequencing.

Table 2 Microbial diversity at baseline and prediction of response to therapy.

Ref.	Alpha diversity1	Beta diversity1
Ananthakrishnan et al[57], 2017	Alpha diversity (Fisher’s alpha) higher in those achieving remission	Lower beta diversity in remission group (Bray Curtis dissimilarity)
Lee et al[63], 2021	Responders to anti-cytokine therapy had significantly higher indices of microbial richness at baseline	NA
Doherty et al[9], 2018	No significant difference in alpha diversity (Inverse Simpson index)	Baseline beta diversity was significantly different by response (P = 0.018)
Effenberger et al[61], 2021	No significant difference in alpha diversity (Shannon index)	No significant difference in beta diversity (paired Bray Curtis distance)
Yilmaz et al[68], 2019	No significant difference in alpha diversity	NA
Guo et al[62], 2024	No significant difference in alpha diversity (microbial richness and evenness)	NA
Dovrolis et al[60], 2020	NA	Distinct enterotype of bacteria noted in the two groups

¹Comparison between responders and non-responders to therapy.

NA: Not available.

Four papers reported on the baseline beta diversity between responders and non-responders (Table 2)[9,57,60,61]. Two studies found differences in beta diversity at baseline. Dovrolis et al[60] showed qualitative differences in the enterotypes of the responders and non-responders. In contrast, no difference was noted on beta diversity estimates at baseline for both azathioprine and anti-TNF inhibitors in the paper by Effenberger[61].

Bacterial taxonomic composition as a predictor of therapeutic response

The gut bacterial population is made up of a large number of genera and species. The papers report taxonomy at various levels of classification from phylum to genus and species. 16S rRNA is able to predict composition up to genus level with good accuracy, while identification of species requires greater depth of sequencing by methods such as shotgun and Sanger sequencing. Hence, only two papers that used a greater depth of sequencing method have reported on the predictive capacity at the species level[57,63]. Burkholderiales species and Roseburia inulinivorans, Clostridium citronae, Agathobaculum butyriciproducens, B. stercoris, B. caccae, and B. ovatus were identified as predictors of improvement after treatment. Some of these organisms were predictors specific to the type of therapy used[63].

The bacterial population in stool and tissue has differences in structure and function. Dovrolis et al[60] and Yilmaz et al[68] reported the bacterial community on mucosal surface as biomarker of response to therapy. Patients received anti-TNF agents in both of these studies. Higher abundance of Bifidobacterium, Collinsella, Lachnospira, Lachnospiraceae, Roseburia, Eggerthella, Parvimonas, Hungatella, Ruminococcus, and Stenotrophomonas was noted in those responding to therapy. Phascolarctobacterium, Negativibacillus, Faecalibacterium, Eubacterium_hallii_group, Blautia and Ruminococcus_gnavus_group were more common in non-responders.

Among the remaining papers (all evaluating 16S rRNA in stool), three found no difference in fecal bacterial taxa between responders and non-responders[59,61,65]. The determinants of treatment outcome in the study by Park included a higher level of Actinobacteria, Dorea, Agathobaculum, and Blautia, and a lower level of Proteobacteria, Enterobacteriaceae, Odoribacter, and Ruminococcus gnavus in treatment responders[64]. Sanchis-Artero et al[66] found a higher ratio of Faecalibacterium prausnitzii/Escherichia coli in responders compared to non-responders (12 vs 6, P = 0.006). The recent study from Belgium grouped the fecal bacterial population into enterotypes[69]. The enterotype Bacteroides2 (Bact2) was negatively associated with remission when using Vedolizumab but not Ustekinumab or anti-TNF agents.

Only one paper reported on the usefulness of the fungal community in predicting therapeutic response[62]. A lower abundance of Asterotremella and Wallemia was noted in the non-responder group compared to the responders. Table 3 provides a summary of the baseline microbial taxonomy as a determinant of response to therapy. Overall, nine of the twelve papers evaluating taxonomy found an association with therapeutic response. Table 4 shows the microbial taxonomic association with different class of drugs. Interestingly, only five bacterial taxa were present in more than one paper, and two of them showed contrasting effects.

Table 3 Bacterial taxonomy at baseline before starting treatment and prediction of response to therapy.

Ref.	Bacterial taxonomy
Ananthakrishnan et al[57], 2017	Treatment responders had higher abundance of Burkholderiales species and Roseburia inulinivorans and butyrate producers than non-responders. Microbial dysbiosis index showed no significant difference between the two groups
Doherty et al[9], 2018	Treatment responders had higher abundance of Bacteroides (P = 0.022) and Faecalibacterium (P = 0.003) than non-responders
Park et al[64], 2022	Stool samples: Actinobacteria, Dorea, Agathobaculum, and Blautia levels were higher and Proteobacteria, Enterobacteriaceae, Odoribacter and Ruminococcus gnavus were lower in anti-TNF-α responders. Saliva samples: Levels of Abiotrophia defective-species, and FJ976422_s-species were higher and Ralstonia were lower in responders than in non-responders
Yilmaz et al[68], 2019	Higher abundance of Bifidobacterium, Collinsella, Lachnospira, Lachnospiraceae, Roseburia, Eggerthella taxa and lower abundance of Phascolarctobacterium noted in treatment responders than non-responders to anti TNF therapy. No differences noted between the responders and non-responders to steroid treatment
Guo et al[62], 2024	Lower abundance of Ruminococcus, Lachnoclostridium, Akkermansia in bacterial community and lower abundance of Asterotremella and Wallemia in fungal community noted in non-responder group compared to responders
Dovrolis et al[60], 2020	Parvimonas, Hungatella, Roseburia, Ruminococcus and Stenotrophomonas were associated with responders, and Negativibacillus, Faecalibacterium, Eubacterium_hallii_group, Blautia, Ruminococcus_gnavus_group were associated with non responders
Lee et al[63], 2021	Clostridium citronae, Agathobaculum butyriciproducens associated with remission after anti-cytokine therapy, B stercoris with anti-integrin therapy and B. caccae and B. ovatus with both types of therapies
Caenepeel et al[69], 2024	Enterotype Bacteroides 2 was negatively associated with remission after treatment with Vedolizumab. No association was found with Ustekinumab or anti-TNF agents
Sanchis-Artero et al[66], 2021	Significant difference in Faecalibacterium prausnitzii/Escherichia coli (F/E) ratio noted between responders and non-responders (P < 0.01)
Effenberger et al[61], 2021	No significant difference at genus level between responders and non-responders
Ding et al[59], 2020	No significant difference in micorbial profiles of responders and non-responders
Ribaldone et al[65], 2019	No significant difference in concentration of all taxa between responder and non-responder

TNF: Tumour necrosis factor.

Table 4 Microbial taxonomic associations with response to therapy in studies using a single class of drug to treat Crohn’s disease.

Microbial taxa	Park et al[64], 2022, CDAI	Guo et al[62], 2024, HBI	Dovrolis et al[60], 2020, HBI	Yilmaz et al[68], 2019, C and B	Ananthakrishnan et al[57], 2017, HBI	Doherty et al[9], 2018, CDAI
	TNF inhibitor				Integrin inhibitor	Interleukin inhibitor
Actinobacteria	P
Dorea	P
Agathobaculum	P
Blautia	P		N
Proteobacteria	N
Enterobacteriaceae	N
Odoribacter	N
Ruminococcus gnavus	N		N
Lachnoclostridium		P
Akkermansia		P
Ruminococcus		P	P
Parvimonas			P
Hungatella			P
Roseburia			P	P
Stenotrophomonas			P
Negativibacillus			N
Faecalibacterium			N			P
Eubacterium_hallii_group			N
Bifidobacterium				P
Collinsella				P
Lachnospiraceae				P
Eggerthella				P
Phascolarctobacterium				N
Bacteroides						P
Burkholderiales					P
Roseburia inulinivorans					P

P: Positive association; N: Negative association; TNF: Tumour necrosis factor; CDAI: Crohn’s disease activity index; HBI: Harvey Bradshaw index; C and B: Clinical and biochemical.

Baseline functional profile of bacteria and response to therapy

The functional potential of the microbial community appears to be an important driver of physiological and pathological processes locally as well as systemically. Accurate estimation of functional capacity requires sequencing of large or entire bacterial genomes, although this is less frequently done due to resource constraints. In this regard, the information on functional characteristics of gut bacteria from the study by Ananthakrishnan et al[57] and Lee et al[63] is more robust as they both sequenced a large portion of the bacterial genome (Table 4). They both identified a number of bacterial metabolic pathways that were differentially abundant between responders and non-responders. This included pathways involved in biosynthesis of L-citrulline, L-isoleucine, arginine, and polyamine[57].

Lee et al[63] found 120 enzyme pathways from six enzyme classes to be differentially expressed between the two groups. Of these, eight pathways were contributed predominantly by Egerthella lenta, several of which are involved in secondary bile acid synthesis. The remaining 112 pathways were enriched in non-responders, and 40% were mainly associated with hydrolase activity associated with the breakdown of mucin-derived glycans.

Data on functional capacity of mucosal microbiota as predictors or response was lacking. Higher butyrate production and higher intercellular exchange of butyrate were linked to positive outcomes after therapy (Table 5)[56,61].

Table 5 Bacterial function characteristics and baseline predictive models for predicting response to therapy.

Ref.	Bacterial function and predictive models
Studies evaluating combined predictive models
Ananthakrishnan et al[57], 2017	Baseline enrichment of 13 microbial pathways noted in treatment responsive CD patients. Neural network model (Vedonet): Combination of clinical, taxonomic and metabolic pathway data predicted clinical remission at week 14
Doherty et al[9], 2018	Combination of clinical and baseline fecal microbiome data had an AUC of 0.844 for predicting response to therapy
Lee et al[63], 2021	120 bacterial enzyme pathways were mostly differentially abundant at baseline in responders to anti-cytokine therapy. Combination of clinical, metagenomic, metabolomic and proteomic markers had AUC of 0.96 in predicting response to anti-cytokine therapy
Zhou et al[10], 2018	Gut microbiota alone predicted responses to therapy with to 86.5% accuracy. Combination of gut microbiota, fecal calprotectin and CDAI improve the accuracy of prediction to 93.8%
Caenepeel et al[69], 2024	Combination of clinical data, microbial load in stool, bacterial enterotype Bacteroides2 in stool, fecal moisture, and fecal calprotectin level predicted response to therapy with AUC of 0.74
Studies evaluating microbial function (without a combined model)
Effenberger et al[61], 2021	Higher butyrate production capacity was noted in responders compared to non-responders
Aden et al[56], 2019	Intercellular exchange of butyrate was significantly reduced in non-responders compared to responders

CD: Crohn’s disease; AUC: Area under the curve; CDAI: Crohn’s disease activity index.

Combination of features to predict response to therapy

Another approach to identifying a suitable drug for a specific patient is by combining various types of data into an integrated prediction tool. Ananthakrishnan et al[57] used a neural network model (Vedonet) which combined clinical, taxonomic, and metabolic pathways to predict remission at 14 weeks after starting vedolizumab [area under the curve (AUC) 0.776]. However, a model with 40 microbiome variable performed better than the combined model (AUC 0.872). Lee et al[63] who recruited patients from the same center, found a combination of clinical, metagenomic, metabolomic, and proteomic markers to have an impressive AUC of 0.96 in predicting response to anti-cytokine therapy.

Doherty et al[9] combined clinical and microbiome data before treatment to predict remission at 6 weeks after starting ustekinumab therapy. The clinical data included components of CDAI, patient metadata, and inflammatory markers. When clinical data were used alone, AUC was only 0.616. The combined clinical and microbiome data had an AUC of 0.844 for predicting response to therapy.

The study from Belgium used a combination of the clinical data of patients, microbial load in stool, bacterial enterotype Bact2 carrier status in stool, fecal moisture, and fecal calprotectin level to predict response to therapy[69]. The sensitivity was 67.5%, specificity 67.6% with AUC of 0.74.

Zhou et al[10] used CDAI, fecal calprotectin, and gut microbial data individually and in combination to assess their performance in predicting successful therapy with infliximab at 30 weeks. When used individually, CDAI had a prediction accuracy of 58.7%, fecal calpeptin 62.5% and gut microbiota 86.5%. When all three were combined, accuracy improved to 93.8%. This study supports the superiority of microbial prediction over clinical and biochemical markers and the benefit of combining all parameters together. In summary, combined prediction tools performed better than individual parameters with an AUC of up to 0.96.

DISCUSSION

This review has highlighted the potential of gut microbiota as a novel biomarker of response to therapy in CD. This may take us one step closer to the goal of precision medicine for patients with CD. Different attributes of the bacterial community, including their diversity, composition, and function, were found to be useful. While bacterial alpha-diversity performed less consistently, differences in functional potential (e.g., short-chain fatty acid metabolism) showed a more consistent trend. It was generally found to be superior to clinical and biochemical markers as a prediction tool.

Alpha diversity is influenced by several factors-site of sample (mucosa vs stool), depth of sequencing, and method of estimation. Hence, direct comparison may not be feasible. Also, it appears a bit simplistic as a given genus may have species with contrasting functions, which do not get captured by this measure. A better alpha diversity is a sign of a healthy bacterial community, and in the two studies where the sequencing was deeper to allow for species-level resolution, a better diversity was a predictor of response[57,63]. This effect was not noted when sequencing was done using 16S rRNA. Comparison of diversity between responders and non-responders (beta diversity) showed differences at baseline in three out of four papers. The diversity of the microbial community may influence outcomes but translating this into clinical predictor model remains a challenge.

Of the twelve studies evaluating baseline microbial taxonomic composition, nine of them found various bacterial taxa to be associated with therapeutic outcome. These included higher abundance of Burkholderiales species, Roseburia inulinivorans, Bacteroides, Faecalibacterium, Actinobacteria, Dorea, Agathobaculum, Blautia, Bifidobacterium, Collinsella, Lachnospiraceae, Roseburia, Eggerthella, Parvimonas, Hungatella, Ruminococcus and Stenotrophomonas in responders. In addition, a lower abundance of Proteobacteria, Enterobacteriaceae, Odoribacter, Ruminococcus gnavus, Phascolarctobacterium, etc., was noted in responders. Several of these bacteria are associated with short-chain fatty acid metabolism. Short-chain fatty acid is an important bacterial metabolite that influences gut immune function. Most of the observations in one paper were not replicated in another, and this is an important barrier when considering their use as biomarkers of response to treatment. Filtering out important and consistent predictors from among these based on the functional potential may help in this regard, as functional signatures appear to be more reproducible. Further, predictor data need validation at other centers to ascertain their utility.

The data from the bacterial metabolic and functional characteristics appear quite promising. Microbial communities with higher butyrate production capacity increased the chance of successful therapy[56,61]. Reports with in-depth sequencing of bacterial genes to assess functional potential showed several microbial pathways to be differentially abundant between responders and non-responders[57,63]. While microbial composition may vary among individuals, the genetic pool may be more stable, suggesting functional redundancy[70]. As the functional potential of microbial communities may be more strongly linked to the interactions with the host, it appears to be a more robust determinant of outcome than individual bacterial taxa. This may also help reconcile/explain the significant differences noted between studies at the taxonomic level. Studying functional redundancy is more challenging than taxonomic composition and includes the assessment of gene evolution and interaction networks[70]. However, only some papers have looked into this, perhaps due to cost and bioinformatic constraints.

The most exciting data comes from prediction tools that have used a combination of clinical, laboratory, and microbial parameters. Individually, microbial characteristics outperformed clinical and biochemical features. Combining these parameters led to further improvement in the accuracy of prediction. With a multitude of factors affecting the response to therapy, it seems logical that a combined prediction model would be more appropriate. Integrating microbiome predictors could reduce ineffective therapy, improve cost-effectiveness, and advance precision medicine. While clinical and biochemical parameters are comparable across populations, inherent variations in microbial community populations in different regions may pose a challenge in identifying microbial factors that would be uniformly applicable.

Readiness for clinical implementation

Five studies described multivariate prediction models[9,10,57,63,69]. External validation of these models is essential for clinical application. This is especially important considering the microbial component present in these models, as there is marked variation in taxonomic predictors among studies. However, the current prediction models generally lack external validation, which limits their clinical implementation. There are tools and checklists, such as Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis and Developmental and Exploratory Clinical Investigations of Decision support systems driven by Artificial Intelligence, which help to improve the quality of reporting of such models, including those using artificial intelligence[71,72]. Use of these tools may help in further improving the quality of such models in the future.

Limitations

The findings on microbial predictors are at times conflicting, with the same organism showing opposing behaviour in different studies[60,64]. Studies focus on deriving singular species that are enriched or depleted in responders, usually by employing machine learning algorithms like random forest classification, and fail to account for microbial interactions. Analyses that take into consideration the interaction of species, such as network analysis and ecological co-occurrence modeling, can help understand these functional interactions in diseased states. Most of the included studies had a small sample size, and there was heterogeneity with regard to the therapy used and the definition of response to treatment. There were only a few studies from outside Europe and the United States, and as the microbial population is affected by geographical location, the results may not be easy to apply to different populations. A large intercontinental study with IBD patients from Ireland and Canada revealed geographical location to be the most important factor accounting for variance in microbial population[73]. Therefore, the results from microbial predictors as well as composite predictors require external validation in different geographical regions before being considered for clinical use. Finally, there is limited information on the mechanistic implications of microbial determinants, but these are needed to differentiate association from causality.

CONCLUSION

In conclusion, the microbiome’s predictive capacity represents a transformative advance in the management of CD, offering clinicians the ability to personalize treatment selection based on individual microbial signatures (Figure 2). The functional microbial pathways (e.g., short-chain fatty acid metabolism) are more consistent predictors of therapeutic response than alpha-diversity or single taxa. Future predictive models in CD will likely need to combine microbial functional signatures with clinical and biochemical markers to achieve clinical utility. The predictive biomarkers can guide therapy choice, reduce costs, and improve patient outcomes. Integrating microbiome functional signatures with clinical and metabolomic data represents the most promising path toward precision therapy in CD.

Figure 2  Potential parameters for predicting response to therapy in Crohn’s disease.