Encapsulated faecal microbiota transfer in young women with anorexia nervosa: an open-label feasibility pilot trial

Abstract

Perturbations of the gut microbiome have been associated with anorexia nervosa (AN) suggesting microbiome-modulation treatments, like faecal microbiota transfer (FMT), may offer therapeutic benefits. This open-label feasibility pilot trial evaluated the tolerability and microbiological impact of encapsulated, multi-donor FMT in 18 young women with AN (Registration: ACTRN12621001504808). Participants completed clinical and microbiome assessments at enrolment (3 weeks pre-treatment), baseline, and 3, 6, and 12 weeks post-treatment. Fifteen participants completed FMT, and 11 completed the final follow-up. The primary outcome was the change in gut microbiome composition from baseline to 3 weeks compared with natural variation between enrolment and baseline. FMT produced a significantly greater shift post-treatment (mean ± SD Bray–Curtis dissimilarity 0.36 ± 0.11; p = 0.0007), with participants gaining 38 ± 16 new species. Donor-derived strains comprised 41 ± 12% of the microbiome at 3 weeks, with engraftment persisting at 6 and 12 weeks. FMT was generally well tolerated; adverse events were mostly mild to moderate and overlapped with typical AN symptomatology. Monitoring of clinical outcomes supported the safety profile and suggested potential improvements in anxiety and metabolic parameters; however, the small sample and absence of a control arm preclude safety and efficacy inference. Overall, these findings warrant further investigation through randomised controlled trials in AN.

Here, in an open-label pilot trial, the authors show that multi-donor fecal microbiota transfer in young women with anorexia nervosa was well tolerated and led to lasting changes in gut bacteria, supporting further research on microbiome-based treatments.

Introduction

Anorexia nervosa (AN) is a complex and challenging, multi-faceted eating disorder that primarily affects young women and results in considerable weight loss, body dysmorphia, and a host of life-threatening physical and mental health issues¹. Current treatment options for AN, such as cognitive behavioural therapy and nutritional rehabilitation, have limited success, particularly in people with longstanding illness². Recent research suggests that microbiome perturbations may contribute to the pathophysiology of AN^3–6. For example, the gut microbiome in people with AN differs from that observed in healthy individuals⁷, and can induce weight loss and anxiety-like behaviours when transplanted into germ-free mice^8,9. This has led to a growing interest in the use of microbiome-targeted therapies, such as Faecal Microbiota Transfer (FMT), which aim to restore a healthy gut microbial ecosystem and improve both mental and physical health outcomes in individuals with AN.

While FMT has shown promise in treating conditions such as Clostridioides difficile infection, inflammatory bowel disease^10,11, and metabolic disorders^12,13, its application in modulating the gut microbiome in psychiatric and eating disorders remains largely unexplored¹⁴. Studies in depression and autism spectrum disorders indicate that FMT may modulate gut-brain interactions and improve symptoms, suggesting it could be beneficial for individuals with AN^15–18. However, it is unknown whether FMT would be a feasible treatment for AN, particularly given the complex interplay of biological, psychological, and social factors that contribute to poor treatment engagement and response¹⁹. Moreover, while FMT is generally considered a safe treatment for gastrointestinal conditions (e.g. C. difficile infection) when performed under established guidelines^20–22, its safety in AN must be carefully determined due to the vulnerable and complex health needs of these individuals. To date, only two case reports with mixed findings have examined FMT treatment in AN, with one reporting weight gain and increased microbial diversity²³, and the other showing microbial shifts without significant changes in eating behaviour or weight²⁴, highlighting the need for further investigation.

We hypothesised that FMT administered orally via capsules could shift the gut microbiome in individuals with AN to a healthier state, and potentially aid recovery by altering the condition’s pathophysiology²⁵. However, before progressing to a large randomised controlled trial, it is essential to assess whether FMT is feasible, tolerated by participants, and can be administered without notable adverse effects in this population. To address these questions, we conducted an open-label pilot trial in a small cohort of young women with AN.

Here, we show that encapsulated, multi-donor FMT is feasible and generally well tolerated in young women with AN, and produces a substantial shift in gut microbiome composition, including sustained donor-strain engraftment. This pilot study establishes proof of microbiological impact and procedural feasibility, providing essential groundwork for future controlled trials.

Results

Study population

Over the 18-month recruitment period, 99 expressions of interest were received (Fig. 1). A pre-screening phone call was conducted to assess initial eligibility before inviting participants to a screening and enrolment clinical visit. Thirty-two individuals were ineligible, primarily due to not meeting the age or body mass index (BMI) criteria or living outside of Auckland (Fig. 1). Nine individuals declined participation, citing an inability to commit to the study, and 35 either could not be contacted or lost interest after being sent the participant information sheet (Fig. 1).

Fig. 1. CONSORT flow diagram.

The diagram outlines participant flow from initial expressions of interest through screening, enrolment, FMT treatment, and follow-up visits. Numbers lost to follow-up, ineligible at each stage, or withdrawn after enrolment are shown, along with reasons for exclusion where applicable.

While most eligibility criteria were assessed during the initial phone pre-screening, final eligibility could only be confirmed at the Screening and Enrolment clinic visit after verifying BMI requirements. Eighteen of the 23 individuals who attended the clinic visit met the inclusion criteria and were enroled in the trial, which was slightly below our recruitment target of 20 participants.

Baseline assessment and treatment visits were scheduled 3 weeks after enrolment to enable quantification of the natural temporal variation in each participant’s gut microbiome over a 3-week period, prior to FMT. Two participants withdrew during this pre-intervention phase; one due to clinical deterioration, the other due to FMT-related concerns expressed by their routine healthcare provider.

The remaining 16 participants were aged 21.7 ± 3.3 years (range 16–28 years) and predominantly self-identified as New Zealand European or Other European ethnicity (14/16). The mean BMI was 17.9 ± 1.3 kg/m², reflecting inclusion of participants meeting DSM-5 criteria for both typical and atypical AN. A formal AN diagnosis had been made, on average, 5.1 ± 2.9 years earlier. The majority of participants had mild eating disorder symptomatology (11/16), while depressive and anxiety symptoms tended to fall in the moderate-to-severe range, with 11/16 scoring above the threshold for mild depression and 13/16 (81%) scoring above the threshold for mild anxiety.

Treatment feasibility and tolerability

Sixteen participants attended the baseline assessment visit, where they were given the option to take their 20 FMT capsules over one, two, or four consecutive days. One participant withdrew as she was unable to swallow any capsules (6%; 1/16), two (13%) opted to take all 20 on the same day, and the remaining 13 (81%) split the dose over two days. Notably, none chose to take their capsules over four consecutive days.

There were no adverse reactions observed during the one-hour monitoring period following capsule administration. On the final day of treatment, participants completed a tolerability questionnaire (Supplementary Table 1). Half of participants (8/16) reported it was ‘not difficult at all’ to swallow the treatment capsules, seven (44%) found it ‘somewhat difficult,’ and one described it as ‘impossible’ and withdrew from the study.

Regarding immediate side effects or unpleasant tastes after capsule intake, four participants reported mild side effects (27%; 4/15) such as burping, feeling cold, and slight abdominal pain, while two participants (13%) noted a mild ‘powdery’ or ‘weird’ taste (Supplementary Table 1). Encouragingly, excluding the participant who could not swallow any capsules, most treated recipients (87%; 13/15) would be willing to undergo FMT again if shown to yield health benefits, with only two ‘not sure’ about it (Supplementary Table 1).

FMT alters gut microbiome composition (primary outcome)

The primary outcome was to determine whether FMT could shift participants’ gut microbiome composition. The gut microbiome composition naturally drifts within an individual over time²⁶ and can be quantified using the Bray-Curtis (BC) dissimilarity index, which ranges from 0 (i.e. the two profiles are completely identical) to 1 (the two samples are completely distinct). To quantify the natural microbiome drift before treatment, we calculated the BC dissimilarity between each participant’s enrolment and baseline samples, collected 3 weeks apart (n = 15, mean 0.22 ± 0.11, Fig. 2A). We then calculated the BC dissimilarity between each participant’s baseline and post-FMT samples, collected 3, 6, and 12 weeks after treatment. The shift in composition between baseline and 3-weeks post-FMT (n = 11, mean 0.36 ± 0.11) was significantly greater than the prior natural drift observed over the same timeframe, suggesting that FMT led to a substantial change in gut microbiome composition (paired t-test, p = 0.0007, Fig. 2A). This shift persisted at weeks 6 and 12 post-FMT, indicating a sustained impact of the intervention (Fig. 2A).

Fig. 2. Augmentation of the gut microbiome following FMT treatment.

A Bray-Curtis dissimilarity index, ranging from 0 (identical composition) to 1 (completely dissimilar), for the recipients’ gut microbiome samples at enrolment (EN, 3 weeks before baseline) and post-FMT compared to baseline; ***p = 0.0007 from a two-sided paired t-test. No statistical testing was performed at week 6 or 12, as there was no equivalent pre-intervention time interval for comparison. B Species richness (number of unique species detected) in donor and recipient gut microbiome samples. Each ‘Individual capsule’ represents the processed donation from a single donor. The ‘Multi-donor FMT batch’ represents the combined microbiota profiles of the four donors that comprised the FMT capsule batch administered to recipients. Differences in donor samples before and after stool processing were assessed by a two-sided paired t-test. Changes in recipients’ richness compared to baseline were evaluated using two-sided Wilcoxon signed rank tests (*p < 0.05, **p < 0.01, ***p < 0.001). C Top 30 species that were differentially abundant in recipients' gut microbiomes post-FMT. The strength of association was calculated as ‘-log(qval)*abs(coef)’, incorporating both the false discovery rate-adjusted p value (qval) and the model coefficient (coef). Species highlighted in bold represent taxa with evidence of donor strain engraftment. D Proportion of donor-matching strains in recipients’ gut microbiomes before and after FMT treatment. E Relative contribution of each donor to engrafted strains across the three FMT treatment batches. F Proportion of strains matching either the recipient’s baseline strains or one of the four contributing donor strains. Strains not matching any of these were categorised as ‘other’. In A, B, D, the boxes represent the interquartile range (IQR) divided by the median, with whiskers expanding up to 1.5 x IQR. Individual points overlaid on the boxes represent measurements from a single participant unless otherwise specified.

FMT increases species richness

We reasoned that species richness should increase following FMT as organisms fill available ecological niches^27,28. Prior to FMT treatment, participants’ gut microbiomes exhibited similar species richness to the individual healthy donor samples (t-test, p = 0.34), and considerably lower than the combined multi-donor profile (i.e. representative of the FMT batch they received) (Fig. 2B). Following treatment, participants’ species richness increased, gaining an average of 38 ± 16 distinct species by week 3 (Wilcoxon signed-rank test, p = 0.001). Notably, species richness remained elevated (compared to baseline) throughout the follow-up period, indicating stable retention of acquired species (Fig. 2B).

While species richness increased in response to FMT, the Shannon diversity index — a metric which incorporates both species richness and evenness — did not change (Supplementary Fig. 1). This suggests that FMT led to an uneven distribution of species abundances, where the increase in richness was accompanied by dominance of certain species that likely filled unoccupied niches within the recipient communities. This was supported by differential abundance testing, which identified 118 (44%) species that were significantly altered in the post-FMT samples relative to baseline: 87 species increased and 31 species decreased in abundance (Fig. 2C and Supplementary Data 1). Comparatively, only 2 (0.7%) species were differentially abundant in the 3 weeks between enrolment and baseline. The top 10 species with the greatest increase in abundance post-FMT were: Prevotella sp. CAG-604, Holdemanella biformis, Prevotella copri, Dialister sp. CAG-486, Desulfovibrio piger, Eggerthella sp. CAG-209, Prevotella sp. AM42-24, Bacteroides coprophilus, Coprobacter secundus, and Ruminococcus lactaris (Fig. 2C).

At the genus level, Desulfovibrio, Holdemanella, and Prevotella were the most significantly enriched genera post-treatment (Supplementary Data 1). Notably, Prevotella showed a marked increase in relative abundance (enrichment) from 2.1 ± 4.6% at baseline to 14 ± 11% at week 3 (Supplementary Fig. 2C).

Stable engraftment of donor bacterial strains

To confirm that the species enrichments we observed were due to donor strain engraftment, we compared the genetic similarity of strains in donor and recipient samples before and after FMT. Less than 1% of recipients’ strains matched donor strains prior to FMT and were subsequently excluded from post-treatment engraftment estimates (Fig. 2D). Following FMT, an average of 40.8 ± 11.5% of recipient strains genetically matched donor strains (Fig. 2D). This proportion remained stable at subsequent follow-up visits (Week 6, 41.1 ± 8.6%; Week 12, 43.4 ± 10.5%) indicating long-term retention of donor-derived strains (Fig. 2D).

Consistent with findings from our previous multi-donor FMT trial²⁷, participants’ microbiomes did not uniformly shift toward the combined multi-donor FMT batch profile (Supplementary Fig. 2A). Instead, recipient microbiomes exhibited variable changes, in some cases becoming more similar to select individual donor profiles (e.g. DN-F19 and DN-F21) (Supplementary Fig. 2B). Because some donors were used in more than one treatment batch, we were able to monitor strain engraftment performance in the context of different donor pools. Notably, strains from donor DN-F21 engrafted well in both batches 1 and 3, whereas strains from donor DN-F22 engrafted effectively in batch 2 but not batch 3 (Fig. 2E). These findings suggest potential host effects and/or competitive interactions between donor strains that may influence engraftment outcomes.

In addition to the engraftment of donor’s microbial strains, we also observed a high degree of dominant strain turnover from ‘other’ sources (Fig. 2F). The contribution from ‘other’ sources post-FMT (41 ± 10%) was significantly higher than pre-FMT (12 ± 16%, paired t-test, p < 0.001), and similar to the proportion of ‘capsule’ matching strains (42 ± 10%, paired t-test, p = 0.77) (Fig. 2F). However, because only the dominant strain of a given species was profiled, it is not possible to determine whether these ‘other‘ strains represent secondary strains belonging to one of the donors or the recipient at baseline, or, alternatively, newly acquired strains picked up from their environment.

Safety monitoring and adverse events

A total of 189 adverse events (AEs) were reported across the 12-week study period. The majority were mild in severity (143/189; 76%) and included symptoms commonly associated with AN (e.g. bloating, fatigue, mild abdominal discomfort). These are summarised in Supplementary Table 2 and are not considered clinically concerning.

This section focuses on the moderate (43 events, 23%) and severe AEs (3 events, 1.6%) which are more clinically relevant (Fig. 3). The three severe AEs comprise two instances of fatigue reported 48 h and 1 week after treatment, both from the same participant, and one case of vomiting by another participant at 3 weeks. All three events were reviewed by an independent safety monitoring committee and were not considered FMT-related. Among the moderate AEs, fatigue was the most frequently reported AE (10 events, 5 participants), followed by abdominal pain (6 events, 4 participants), bloating (5 events, 3 participants), and agitation (5 events, 3 participants) (Fig. 3).

Fig. 3. Reported adverse events following FMT.

Adverse events (AEs) were graded in accordance with the Common Terminology Criteria for Adverse Events (CTCAE) v.4.03 and v.5.0 (Supplementary Tables 9, 10). Only AEs graded ≥2 (i.e. moderate or severe) are shown. A comprehensive list of all reported AEs is provided in Supplementary Table 2, and individual participant-level data are presented in Supplementary Fig. 3.

AE frequency appeared consistent across the three batches of treatment capsules (Supplementary Fig. 3). However, batch 2 capsules underwent additional pathogen testing after one participant developed abdominal pain, nausea, and loose stools 24 hours after treatment  (AN-F11; Supplementary Fig. 3). Laboratory testing identified enteropathogenic Escherichia coli (EPEC) in the participant’s post-treatment stool, which was absent at baseline, suggesting a possible treatment-related infection. Although all four donors’ stools tested negative for pathogens at the time of donation, subsequent re-testing detected EPEC in the capsules from one donor (DN-F16). Two additional participants previously treated with capsules from the same batch did not report any symptoms indicative of infection (AN-F09 & AN-F10; Supplementary Fig. 3). In response to this event, a fresh batch of treatment capsules (FMT batch 3) was prepared, and donor screening protocols were modified to require negative pathogen results for both the unprocessed stool donation and processed capsule contents before release.

Clinical outcomes

Monitoring of clinical parameters and symptom scores over 12 weeks post-FMT did not raise any safety concerns. Below, we describe changes from baseline that were observed during the follow-up period; however, no causal inferences should be made given the lack of a control group.

Serum serotonin levels, which are typically elevated in AN, dropped by 11.3% at week 6, and 16.6% by week 12 (Fig. 4A and Table 1). Serum creatinine was also reduced at week 12 (Fig. 4B and Table 1). Additionally, a slight increase in the liver enzyme aspartate transaminase (AST) was noted at week 12, though this was statistically significant only in the paired analysis and was still within the normal range (Supplementary Table 3).

Fig. 4. Clinical outcomes with change from baseline in participants with complete data.

A Circulating serotonin (log scale), B circulating creatinine, and C anxiety symptoms as measured by General Anxiety Disorder 7-item scale (GAD-7) total score throughout the trial period, including only those participants who had measurements for all visits. Differences in clinical outcomes between visits were assessed using repeated measures linear mixed models. P values are two-sided and were not adjusted for multiple comparisons (*p < 0.05, **p < 0.01). In A, B, C the boxes represent the interquartile range (IQR) divided by the median, with whiskers expanding up to 1.5 x IQR. Individual points overlaid on the boxes represent measurements from a single participant.

Table 1.

Analysis of blood parameters throughout the trial using all available data

Variable	Mean ± SD or Median [Q1, Q3]			Baseline vs Week 6		Baseline vs Week 12
	Baseline (n = 15)	Week 6 (n = 9)	Week 12 (n = 11)	aMD [95% CI]	p	aMD [95% CI]	p
Electrolytes
Sodium (mmol/L) Ln	139 [136, 140]	140 [138, 140]	139 [138, 140]	1.2% [−0.3%, 2.8%]	0.13	0.7% [−0.7%, 2.2%]	0.35
Potassium (mmol/L) Ln	4.1 [3.9, 4.3]	4.1 [3.9, 4.1]	4.0 [4.0, 4.3]	−0.8% [−6.6%, 5.4%]	0.80	−0.9% [−6.3%, 4.9%]	0.77
Chloride (mmol/L)	102.3 ± 2.4	103.3 ± 2.3	102.8 ± 1.1	1.1 [−0.6, 2.7]	0.22	0.5 [−1.0, 2.1]	0.50
Nutritional status
Albumin (g/L)	45.4 ± 2.4	45.2 ± 3.2	44.9 ± 1.5	−0.1 [−1.7, 1.4]	0.88	−0.3 [−1.8, 1.1]	0.66
Total protein (g/L) Ln	72.6 [71.2, 74.4]	72.2 [71.0, 75.1]	73.4 [72.4, 75.2]	−0.3% [−3.4%, 3.0%]	0.87	0.0% [−3.0%, 3.1%]	0.99
Ferritin (ug/L) Ln	39.4 [21.7, 53.8]	27.8 [21.1, 37.8]	25.7 [20.1, 43.0]	−21.4% [−38.9%, 1.2%]	0.08	−21.3% [−37.6%, −0.6%]	0.06
Vitamin B12 (pg/ml)	521 ± 200	497 ± 149	423 ± 149	−29 [−79, 22]	0.28	−36 [−83, 11]	0.15
Folate (ng/ml)	12.8 ± 6.4	11.2 ± 4.8	10.8 ± 5.5	0.3 [−2.1, 2.7]	0.82	0.1 [−2.1, 2.4]	0.90
Liver, thyroid and kidney function
ALP (U/L)	63.5 ± 15.7	63.8 ± 13.9	58.6 ± 12.6	−0.7 [−5.1, 3.7]	0.75	−1.3 [−5.4, 2.8]	0.54
ALT (U/L) Ln	14.5 [13.4, 17.8]	13.4 [11.5, 15.9]	14.8 [11.9, 19.7]	−15.8% [−28.7%, −0.5%]	0.06	8.2% [−7.2%, 26.2%]	0.33
AST (U/L) Ln	18.8 [16.1, 21.4]	17.6 [16.4, 19.4]	18.4 [16.7, 20.8]	−10.2% [−21.6%, 2.9%]	0.14	5.2% [−7.3%, 19.3%]	0.44
GGT (U/L) Ln	15.0 [12.0, 16.0]	14.0 [11.0, 17.0]	15.0 [13.0, 17.0]	−3.1% [−15.6%, 11.3%]	0.66	12.0% [−1.5%, 27.3%]	0.10
Cholinesterase (U/L)Ln	6647 [6003, 7504]	6520 [6347, 7152]	6687 [6098, 7657]	1.3% [−4.5%, 7.6%]	0.67	1.5% [−4.0%, 7.2%]	0.61
TSH (uIU/ml)Ln	1.74 [1.41, 2.51]	1.22 [0.99, 2.35]	1.25 [1.06, 2.18]	−14.4% [−29.9%, 4.4%]	0.14	−8.7% [−24.0%, 9.8%]	0.35
FT4 (pmol/ml)	14.4 ± 2.6	13.5 ± 2.2	15.4 ± 2.5	−0.5 [−1.7, 0.7]	0.39	0.6 [−0.6, 1.7]	0.34
Creatinine (umol/L)	64.4 ± 7.8	64.0 ± 8.0	59.3 ± 6.4	−1.5 [−5.2, 2.3]	0.44	−5.1 [−8.6, −1.6]	0.0095
Stress, inflammation and mood
Cortisol (nmol/L)Ln	434 [349, 490]	410 [326, 589]	359 [210, 638]	−2.2% [−25.2%, 27.9%]	0.87	−13.8% [−32.8%, 10.5%]	0.26
hsCRP (mg/L)Ln	0.38 [0.19, 0.93]	0.30 [0.12, 1.10]	0.84 [0.13, 1.53]	−41.4% [−72.2%, 23.5%]	0.18	−5.5% [−52.7%, 88.8%]	0.87
Serotonin (ng/ml)Ln	291 [168, 395]	285 [207, 302]	250 [183, 325]	−11.3% [−19.9%, −1.8%]	0.033	−16.6% [−24.1%, −8.3%]	0.0097

Differences in blood measures between visits were assessed using repeated measures linear mixed models (fixed effect = Visit, random effect = Participant) using all available data.

P values are two-sided and were not adjusted for multiple comparisons. Statistically significant p values (p < 0.05) are highlighted in bold.

^LnVariables log-transformed prior to analyses; therefore, their back-transformed adjusted mean difference (aMD) and 95% confidence interval (CI) are reported here as proportional differences compared to baseline.

ALP Alkaline Phosphatase, ALT Alanine Transaminase, AST Aspartate Transaminase, FT4 free thyroxine, GGT Gamma-Glutamyl Transferase, hsCRP high-sensitivity C-reactive protein, TSH thyroid-stimulating hormone.

No changes were observed in anthropometry (weight and BMI) or body composition (e.g. total body fat percentage, total lean mass percentage, or bone mineral density) over the course of the study in either ITT (Table 2) or paired analyses (Supplementary Table 4). Similarly, there was no detectable change from baseline in PHQ-9 total scores (depression symptoms), EDEQ total scores (eating disorder symptoms), and EDEQ sub-domain scores (restraint, eating concern, shape concern, weight concern) (Table 3 and Supplementary Table 5). However, in both the ITT and paired analyses, we observed a reduction in anxiety symptoms, as indicated by a 3-point drop in GAD-7 total scores, at 12 weeks post-treatment (Fig. 4C and Table 3). A similar reduction in GAD-7 total scores was also observed at 6 weeks in the paired analysis (Supplementary Table 5). These changes reflected an overall shift from moderate to mild anxiety symptoms.

Table 2.

Analysis of anthropometry and body composition measures at enrolment and week 12 using all available data

Variable	Mean ± SD		Enrolment vs Week 12
	Enrolment (n = 18)	Week 12 (n = 11)	aMD [95% CI]	p
Weight (kg)	48.7 ± 4.3	51.3 ± 4.7	0.71 [−0.19, 1.60]	0.15
BMI (kg/m2)	17.77 ± 1.27	18.50 ± 0.96	0.22 [−0.09, 0.53]	0.19
Total body fat (%)	24.8 ± 6.2	26.7 ± 3.1	0.3 [−0.9, 1.5]	0.65
Total lean mass (%)	72.0 ± 5.7	70.3 ± 2.9	−0.3 [−1.4, 0.9]	0.66
BMD (g/cm2)	1.029 ± 0.072	1.024 ± 0.072	0.001 [−0.012, 0.013]	0.93
BMD Z-score	−0.01 ± 0.79	0.04 ± 0.80	−0.02 [−0.13, 0.10]	0.79

Intention-to-treat analysis was used. Adjusted mean difference (aMD), 95% confidence intervals (CI), and p values derived from linear mixed-effect models. P values are two-sided and were not adjusted for multiple comparisons.

BMI body mass index, BMD bone mineral density.

Table 3.

Analysis of eating disorder, depression and anxiety symptoms across the trial using all available data

Variable	Mean ± SD			Baseline vs Week 6		Baseline vs Week 12
	Baseline (n = 16)	Week 6 (n = 13)	Week 12 (n = 11)	aMD [95% CI]	p	aMD [95% CI]	p
Eating Disorder
EDEQ–Total Score	2.5 ± 1.5	2.3 ± 1.5	2.1 ± 1.2	−0.3 [−0.9, 0.3]	0.37	−0.4 [−1.0, 0.3]	0.25
EDEQ–Restraint	2.3 ± 2.1	1.9 ± 2.1	1.5 ± 1.6	−0.3 [−1.2, 0.5]	0.46	−0.6 [−1.6, 0.3]	0.18
EDEQ–Eating concern	1.9 ± 1.2	1.6 ± 1.5	1.2 ± 1.2	−0.2 [−0.8, 0.4]	0.55	−0.5 [−1.2, 0.1]	0.13
EDEQ–Shape concern	3.2 ± 1.5	3.0 ± 1.6	3.0 ± 1.2	−0.3 [−0.9, 0.4]	0.40	−0.2 [−0.9, 0.5]	0.52
EDEQ–Weight concern	2.8 ± 1.6	2.5 ± 1.7	2.6 ± 1.5	−0.3 [−0.9, 0.3]	0.30	−0.2 [−0.8, 0.5]	0.65
Depression
PHQ-9–Total Score	8.6 ± 6.3	8.1 ± 6.4	6.1 ± 4.9	−0.1 [−2.3, 2.1]	0.93	−1.1 [−3.5, 1.2]	0.36
Anxiety
GAD-7–Total Score	10.6 ± 5.8	8.1 ± 5.3	7.3 ± 4.1	−2.7 [−5.5, 0.0]	0.065	−3.2 [−6.1, −0.3]	0.043

Self-completed questionnaires administered at baseline (BL), week 6 (wk6) and week 12 (wk12). Intention-to-treat analysis was used. Adjusted mean difference (aMD), 95% confidence intervals (CI), and p values derived from linear mixed-effect models.

GAD-7 General Anxiety Disorder 7-item scale⁵⁹, EDEQ Eating Disorder Examination Questionnaire V.6.0⁵⁷, PHQ-9 Patient Health Questionnaire 9⁵⁸.

P values are two-sided and were not adjusted for multiple comparisons. Statistically significant p values (<0.05) are highlighted in bold.

To explore potential microbial contributors to these changes, we conducted a post-hoc analysis examining associations between microbiome composition and GAD-7 total scores. Fourteen bacterial species were associated with GAD-7 scores after adjusting for antidepressant class; 10 negatively and 4 positively (Supplementary Data 1). Among these, Parabacteroides goldsteinii, Blautia wexlerae, Ruminococcus lactaris, and Desulfovibrio piger showed the strongest negative association.

Discussion

This one-arm, open-label pilot study evaluated the feasibility, tolerability, and efficacy of FMT for altering the gut microbiome in individuals with AN. FMT was well-tolerated and resulted in a substantial shift in gut microbiome composition via the stable engraftment of donor strains. While safety considerations are addressed in further detail below, clinical monitoring throughout the 12-week trial suggested that encapsulated FMT can be administered in this population, provided robust donor screening is in place.

Our findings are consistent with microbiome restoration trials in other conditions^29–31, demonstrating that FMT successfully altered the gut microbiome in individuals with AN, with participants acquiring an average of 38 new species. Notably, this study achieved similar levels of donor strain engraftment to our previous trial²⁷, despite omitting a bowel cleanse prior to FMT administration due to safety concerns. This finding suggests that a bowel cleanse may not be necessary to promote bacterial engraftment in AN, which has important clinical implications. Eliminating the bowel cleanse simplifies the procedure, reduces potential complications (e.g. electrolyte imbalances), and enhances tolerability for participants in future studies^32,33.

Individuals with AN often experience a degree of ambivalence towards recovery or do not believe their condition warrants treatment³⁴. Accordingly, recruitment proved challenging, taking 18 months to enrol 18 participants, of whom only 15 completed the treatment and 11 attended the final 12-week follow-up. These recruitment challenges likely reflect, at least in part, the limited acceptability of stool-derived therapies among this population, in addition to the perceived ‘yuck factor’ and safety concerns^35,36. Encouragingly, most participants who completed the study expressed willingness to undergo FMT again if proven effective, suggesting that, with appropriate support and information, this novel therapeutic approach may be acceptable to young women with AN.

While this pilot feasibility study suggests that FMT was generally well-tolerated by participants, they frequently reported a variety of mild-to-moderate adverse events, most commonly fatigue, abdominal pain, bloating, and agitation. However, interpreting these events is challenging, as many overlap with symptoms commonly experienced in AN and in the absence of a control group, it is difficult to distinguish treatment-related effects from background symptomatology^37,38. Consequently, firm conclusions about the safety of FMT in this population cannot be drawn. It will therefore be important for future randomised controlled trials, alongside efficacy assessments, to include measures that more reliably differentiate the complex symptomatology inherent in AN from adverse events potentially associated with FMT.

Nonetheless, there was one adverse event of particular concern. FMT resulted in the transmission of enteropathogenic EPEC to a recipient due to the false-negative screening of a donor’s stool sample. In response, our entire FMT protocol (from donor screening to FMT administration) was thoroughly reviewed, and the donor screening protocols were consequently revised. Our updated protocol now requires that both the raw stool sample and the post-processed capsule contents test negative before release. This incident highlights that potentially pathogenic bacteria occurring at nearly undetectable levels in asymptomatic donors may cause potential harm when transferred to another individual³⁹, emphasising the need for rigorous screening processes, particularly when administering FMT to vulnerable recipients.

While this study was not designed to assess clinical outcomes, monitoring a variety of blood parameters provided no evidence of metabolic harm, with limited data suggestive of potential benefits to anxiety symptoms. Specifically, we observed a reduction in GAD-7 scores over the course of the study, which coincided with a modest decline in circulating serotonin levels. Although these findings cannot be attributed to FMT in the absence of a control group, these preliminary results align with emerging literature on the microbiota-gut-brain axis, which proposes that alterations in gut microbial communities may influence central nervous system function, in part via microbial metabolism and regulation of neuroactive compounds, such as serotonin^40–43. Given that up to 95% of the body’s serotonin is produced in the gastrointestinal tract⁴⁴, it is biologically plausible that changes in microbial composition could modulate serotonergic signalling. However, the degree to which serotonin produced in the periphery can affect central nervous system function is uncertain, and it remains possible that no direct communication occurs via this pathway. This uncertainty may be particularly relevant for individuals with AN, given reports of a hyperserotonergic phenotype in some cases, which may contribute to heightened anxiety^45–47. The observed temporal association between reduced anxiety symptoms and lower serotonin levels in our study is hypothesis-generating and warrants further investigation in larger, controlled trials.

FMT had a prominent impact on the relative abundance of multiple species, several of which were negatively correlated with anxiety severity (i.e. GAD-7 score). Some of these taxa, particularly Ruminococcus spp., are known to possess tryptophan decarboxylases, a key enzyme influencing serotonin biosynthesis in the gut⁴⁸. Blautia abundance has previously been shown to be negatively associated with anxiety phenotypes in an autism mouse model⁴⁹. Similarly, hydrogen sulfide, predominantly produced by Desulfovibrio piger, has been shown to exert neuromodulatory effects with emerging links to cognitive and mood processes^50,51, including anxiety resilience in animal models⁵². It is, however, important to emphasise that the relationships we observed in our study are purely correlative. Further, whether potential therapeutic benefits may derive from a few key species or from broader shifts in microbial community composition and function remains undetermined and warrants further investigation.

Beyond those already noted, several key limitations of this pilot feasibility study should be acknowledged. The small sample size, absence of a control group, and substantial attrition rate mean we cannot ascertain the clinical efficacy of FMT to ameliorate AN symptomatology. Without a control arm, we cannot rule out the influence of confounding factors such as the natural course of illness, regression to the mean, or non-specific effects of study participation (i.e. the Hawthorne effect). Eligibility was restricted to individuals with AN who were medically stable at enrolment; consequently, our findings may not be generalisable to those with more severe clinical presentations. Furthermore, the recruitment challenges we experienced underscore the difficulties in engaging this population. These factors highlight the need for improved recruitment strategies in future AN trials, potentially including broader eligibility criteria, multi-site collaborations, and enhanced participant support to improve retention and engagement.

Despite these limitations, this pilot study represents a critical first step in evaluating the feasibility, tolerability, and microbiological impact of FMT in young women with AN. The study provided evidence of a successful gut microbiome shift, characterised by an increase in species richness via the stable engraftment of donors’ bacterial strains. These findings provide important proof-of-concept evidence supporting the therapeutic potential of microbiome-targeted interventions in AN. With appropriate safety protocols in place, encapsulated FMT appears to be a feasible and tolerable intervention. Future research should prioritise placebo-controlled randomised trials with larger sample sizes and mechanistic endpoints to better understand the clinical impact, safety, and underlying causal pathways of FMT. These trials must also incorporate a robust design to differentiate the complex AN symptomatology from potential FMT-related side effects. Collectively, our results lay the groundwork for future studies exploring how microbiome-targeted interventions might complement existing treatments by harnessing the gut–brain axis in the management of AN.

Methods

Study design

The Gut Bugs in AN trial was a one-arm, open-label pilot trial investigating the feasibility, tolerability, and practicability of FMT to alter the gut microbiome among young females with AN. The trial was conducted in Auckland (New Zealand) from September 2022 to February 2024 (i.e.18 months). All clinical assessments, stool collection and processing, and treatment administration were carried out at the Liggins Institute Clinical Research Unit (University of Auckland).

Participation involved four clinic visits:

1. Screening and enrolment,

2. Baseline assessments and treatment (3 weeks after enrolment),

3. 6-week follow-up (6 weeks after treatment)

4. 12-week follow-up (12 weeks after treatment).

A detailed description of the study design and methodology is available in the published trial protocol²⁵. The full study schedule has also been reproduced in the Supplementary Table 6.

FMT recipients

Participants were recruited through local eating disorder clinics, Eating Disorders Association of New Zealand (EDANZ) newsletters, and social media advertisements. Full inclusion and exclusion criteria have been published previously²⁵ and are also provided in the Supplementary Table 7. In brief, eligible participants were biological females aged ≥16 to <33 years with a BMI between 13–19 kg/m² who met DSM-5 criteria for AN. Recruitment was restricted to medically stable individuals, following consultation with study clinicians and EDANZ, to ensure participation did not pose undue clinical risk or stress for those requiring acute stabilisation. Because eligibility included individuals with BMI up to 19 kg/m², the cohort reflects both typical and atypical presentations of AN as defined by DSM-5 (i.e. those with low BMI as well as those meeting all other criteria for AN but above the underweight threshold).

Stool donors and FMT encapsulation

We recruited eight stool donors using a comprehensive donor screening protocol²⁵. Donor eligibility criteria and pathogen screening information are also provided, with permission, in Supplementary Tables 7, 8. Donors were healthy biological females, aged 27 ± 3 years, with a BMI of 21.5 ± 1.6 kg/m² (mean ± SD). Fresh stool donations were individually processed using validated methods as described in the trial protocol²⁵. In brief, stool samples were homogenised with saline, sieved to remove particulate matter, and the microbiota fraction concentrated via a series of differential centrifugation steps. Individual donors’ gut microbiota were subsequently resuspended in a cryoprotective saline solution and double encapsulated in acid-resistant delayed release capsules (Size 0, and 00, DRcaps, Capsugel, Sydney, Australia). Each capsule contained 0.5 grams of concentrated microbiota from a single donor. Capsules were stored at −80 °C for up to 6 months. Given the 18-month study duration, three separate FMT treatment batches were prepared with different combinations of donor microbiota.

Treatment administration

Unlike our previous FMT trials^12,54, no bowel cleanse was administered prior to treatment in this cohort. This change was made for perceived safety reasons following discussions with patient support groups and healthcare specialists. In total, participants received 20 FMT capsules containing gut microbiota from four donors (i.e. five capsules per donor), corresponding to a total dose of 10 g of concentrated gut microbiota.

Depending on the participant’s preference, FMT capsule administration was offered over one (20 capsules/day), two (10 capsules/day), or four consecutive days (5 capsules/day). Participants fasted overnight and were supervised by a research nurse or clinician while swallowing the capsules with water. Breakfast was delayed by one hour after capsule ingestion to minimise exposure to stomach acid and speed up their passage to the small intestine.

Adverse events

Participants were closely monitored in the clinic for potential adverse events during the first hour immediately after FMT administration. Participants were also interviewed about adverse events, either over the phone or during their clinic visit, 24 and 48 hours after their first dose of capsules, and after 1, 3, 6, and 12 weeks.

Participants were asked to seek immediate medical attention if they experienced any severe adverse reactions after FMT. The adverse events that participants were asked to report are described in the Supplementary Tables 9, 10, which were defined and graded based on the Common Terminology Criteria for Adverse Events (CTCAE) v4.03⁵⁵ and v5.0⁵⁶. An independent data monitoring committee reviewed clinical and adverse event data quarterly or within 24 hours of a severe adverse event being reported.

Blood parameters

Blood samples were collected at baseline, 6 weeks, and 12 weeks and stored at –80 °C. The following parameters were assessed using standard Roche kit protocols on Cobas C311 and Cobas e601 analyzers (Roche Diagnostics, Basel, Switzerland): electrolytes (sodium, potassium, chloride), creatinine, ferritin, total protein, albumin, liver enzymes [Alkaline Phosphatase (ALP), Alanine Transaminase (ALT), AST, and Gamma-Glutamyl Transferase (GGT)], high-sensitivity C-reactive protein (hsCRP), cholinesterase, folate, vitamin B12, cortisol, free thyroxine (FT4), and thyroid-stimulating hormone (TSH). Serotonin levels were determined using the Serotonin/5-Hydroxytryptamine Competitive ELISA Kit (#EEL006, Invitrogen, Waltham, Massachusetts, U.S.) according to the manufacturer’s protocol, with 50 µl of neat sample used as input.

Body composition

Whole-body dual-energy X-ray absorptiometry (DXA) scans were performed at enrolment and the 12-week follow-up visit to assess body weight and composition (GE Healthcare Lunar i-DXA, encore version 17). DXA-derived body weights were combined with heights measured using a wall-mounted stadiometer to calculate BMI.

Questionnaires

Participants completed three questionnaires at baseline, 6 weeks, and 12 weeks post-treatment:

• Eating Disorder Examination Questionnaire (EDEQ V.6.0)⁵⁷

• Patient Health Questionnaire 9 (PHQ-9) for symptoms of depression⁵⁸

• General Anxiety Disorder 7-item scale (GAD-7)⁵⁹.

For each of these questionnaires, total scores were calculated, with EDEQ also producing subscale scores across four domains: restraint, eating concern, shape concern, and weight concern. For interpretability, EDEQ scores ≤2.8 were considered mild eating disorder symptomatology⁶⁰; PHQ-9 scores were categorised as: ≤5 = mild, 6–10 = moderate, 11–15 = moderately severe, and ≥16 = severe depression; and GAD-7 scores as: ≤5 = mild, 6–10 = moderate, 11–15 = moderately severe, and ≥16 = severe anxiety.

Additionally, participants completed a brief questionnaire on their final day of treatment to gather their initial views and experience of taking the treatment (Supplementary Table 1).

Gut microbiome profiling

Stool samples were collected at five time points: at enrolment (3 weeks before treatment), baseline, and 3, 6, and 12 weeks post-treatment. Samples were preserved in room-temperature-stable DNA/RNA Shield Fecal Collection Tubes (R1101, Zymo Research, Irvine, California, U.S.) before transfer to –80 °C for long-term storage. For donors, pre- and post-processing stool samples were collected at each donation and stored at –80 °C in DNA/RNA Shield.

DNA was extracted from 750 µl aliquots using the ZymoBIOMICS 96 MagBead DNA Kit (D4308, Zymo Research, Irvine, California, U.S.). Homogenisation involved two consecutive rounds of bead bashing (2 × 5 min at 1500 rpm) on a SPEX Sample Prep Mini G. Each extraction batch included a blank extraction control (DNA/RNA shield), the ZymoBIOMICS Gut Microbiome Standard (D6331), and the ZYMOBIOMICS Microbial Community Standard (D6300). Extracted DNA was sequenced by a commercial provider (Novogene, Beijing, China) using Illumina’s 150 bp paired-end sequencing platform to an average depth of 6 GB/sample.

Raw sequencing data were processed with tools from the bioBakery3 suite⁶¹:

• KneadData (v0.10.0)—to trim and remove poor-quality reads and those mapping to the human genome;

• MetaPhlan3 (v3.1)—for taxonomic profiling; and

• StrainPhlAn3 (v3.1)—for SNP haplotyping of the dominant strain for a given species within a given sample.

Downstream microbiome analyses were run in R v.4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). Shannon diversity and Bray-Curtis (BC) dissimilarities were computed using the vegan package⁶², while strain DNA distances were calculated using ape and phangorn packages⁶³. Dominant strain matching was performed as previously described²⁷. In brief, pairwise DNA distances between conspecific strains were calculated using the Jukes and Cantor (JC69) model implemented in the phangorn R package. To account for differences in strain diversity across species, DNA distances were normalised by the median pairwise distance within each species. Strains were considered a match if the normalised DNA distance was ≤0.2. Recipient strains that matched a strain from any of the four contributing donors, but not their own pre-treatment strain (if present), were classified as donor-strain engraftment events. In cases where a recipient’s post-treatment strain matched multiple donor strains, the source was assigned based on the lowest normalised DNA distance. Strains present in recipients at baseline that matched donor strains were considered ambiguous and excluded from engraftment analyses.

Statistical analyses: gut microbiome

The primary outcome was the BC dissimilarity in gut microbiome composition between each participant’s baseline and 3-week post-FMT sample. BC dissimilarity ranges from 0 (completely identical profiles) to 1 (completely distinct). To account for natural temporal variation in microbiome composition, post-FMT dissimilarities were compared to those between each participant’s enrolment and baseline sample, which were also collected 3 weeks apart. Differences between these two indices were assessed using a paired t-test. No statistical testing was performed for BC dissimilarities at week 6 or 12, as there was no equivalent pre-intervention time interval for comparison.

For alpha diversity metrics, paired t-tests or Wilcoxon signed-rank tests were used, as appropriate, to assess changes in species richness and Shannon diversity, including comparisons between donor stools before and after processing, and changes in recipients relative to baseline.

Multivariate Association with Linear Models (MaAsLin2⁶⁴) were used to identify microbial taxa associated with (1) post-treatment visits and (2) GAD-7 anxiety total scores with adjustment for antidepressant medication classes. For all MaAsLin2 models, participant ID was added as a random effect to account for repeated measures, and taxa relative abundances were log-transformed to approximate a normal distribution. A minimum prevalence threshold of 10% was applied, and the default statistical significance threshold of q < 0.25 was used [False Discovery Rate correction using Benjamini-Hochberg Procedure].

Statistical analyses: clinical outcomes

Clinical outcomes were analysed using an intention-to-treat (ITT) approach. To account for missing data, we also conducted a secondary per-protocol analysis, which included only participants who had complete data for all time points for each respective outcome measure. This aimed to address potential bias introduced by differential dropout, as participants who withdrew after treatment tended to be those with the highest severity of eating disorder symptoms and anxiety scores at baseline.

For clinical outcomes, linear mixed-effects models were used to evaluate changes from baseline (i.e. fixed effect = visit, random effect = participant ID). Following visual inspection, variables with a skewed distribution were log-transformed to approximate a normal distribution prior to analyses. All statistical tests for clinical outcomes were 2-sided with significance set at p < 0.05 without adjustment for multiple comparisons. Where appropriate, data in the manuscript text are reported as mean ± standard deviation.

Ethics

Ethics approval for the study was granted by the Central Health and Disability Ethics Committee (reference number: 21/CEN/212). All participants provided written informed consent before enroling in the study and were free to withdraw at any time. This study was performed in accordance with all appropriate institutional, national, and international guidelines and regulations for medical research, in line with the principles of the Declaration of Helsinki⁵³.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Funding acquisition: W.S.C. and J.M.O. Study design and consultation: B.C.W., R.Y.T.-C., M.D., C.C., B.B.A., J.G.B.D., H.T., W.S.C., and J.M.O. Trial management: T.E., J.M.O., and W.S.C. Clinical assessments: R.Y.T.-C., M.D., C.C., B.B.A., W.S.C., and H.T. Sample processing: B.C.W. and S.G. Data analysis and statistics: B.C.W. and D.H. Manuscript drafting: B.C.W and J.G.B.D. Manuscript revision: B.C.W., R.Y.T.-C., B.B.A., J.G.B.D., D.H., M.D., C.C., T.E., S.G., H.T., W.S.C., and J.M.O.

Peer review

Peer review information

Nature Communications thanks Nadia Andreani and the other anonymous reviewer for their contribution to the peer review of this work. A peer review file is available.

Data availability

Metagenomic sequencing data have been deposited in the NCBI Sequence Read Archive under BioProject PRJNA1268178. For access to clinical data, readers will need to request permission for the de-identified clinical data file from the corresponding author. Requestors would need to provide a methodologically sound proposal, obtain appropriate ethics approval, and sign a Data Access Agreement. The Data Access Agreement would need to include a commitment to using the data only for the specified proposal, not to attempt to identify any individual participants, to securely store and use the data, and to destroy or return the data after completion of the project.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-67267-6.