Fecal microbiota transplantation plus immunotherapy in metastatic renal cell carcinoma: the phase 1 PERFORM trial

Study participants

The PERFORM trial (NCT04163289) is an open-label, single-center, phase 1 clinical trial designed to evaluate the safety, tolerability and preliminary efficacy of LND101—a healthy donor-derived microbiome transplantation product—in combination with doublet immunotherapy or immunotherapy plus targeted therapy in patients with mRCC. The study was conducted at the London Health Sciences Centre after approval by the Western University Health Sciences Research Ethics Board, in accordance with applicable national and international regulatory guidelines. These included the Canadian Food and Drug Regulations (C.05.001), the United States Code of Federal Regulations (21 CFR Part 56), the International Conference on Harmonization Good Clinical Practice guidelines and the Declaration of Helsinki.

Administrative oversight was provided by the Clinical Research Unit at the Verspeeten Family Cancer Centre, which also established a data safety monitoring committee (DSMC) to provide independent oversight to ensure participant safety and protocol adherence.

Participant characteristics and consent

All 20 enrolled participants were adults with mRCC. Sixteen (80%) were male and four (20%) were female. The median age was 60 years (range, 47−71). Written informed consent was obtained from all participants in accordance with institutional research ethics board approval and Good Clinical Practice guidelines. Participants did not receive financial compensation for enrollment or participation in the study.

Sex and gender considerations

Sex was recorded for all participants based on self-report at enrollment. Gender identity was not separately collected, consistent with institutional clinical trial standards during the study period. Sex was not used as an inclusion or exclusion criterion, and the study was not powered to evaluate sex-based differences in safety, clinical outcomes, microbiome engraftment or immune correlates. Therefore, no sex-stratified analyses were performed. The unequal distribution of males and females (16 males and four females) further limited the feasibility of such analyses. However, sex distribution is reported transparently, and future larger studies will be required to assess potential sex-related differences.

Patient eligibility criteria

Eligible patients were aged 18 years or older with histologically confirmed advanced (not amenable to curative surgery or radiation) or metastatic RCC (American Joint Committee on Cancer stage IV), classified as favorable, intermediate-risk or poor-risk disease according to IMDC criteria, which consider time from diagnosis to treatment <1 year, hemoglobin below the lower limit of normal, corrected calcium >10 mg dl⁻¹, elevated neutrophil or platelet counts and Karnofsky performance status <80%. Additional requirements were investigator-assessed evaluable disease, ability to ingest oral capsules, provision of written informed consent, acknowledgment of potential non-infectious risks of FMT and recovery to baseline or ≤grade 1 toxicity (per CTCAE v5.0) from prior treatments unless adverse events were clinically insignificant. Adequate organ and marrow function was required, defined by absolute neutrophil count ≥1,500/mm³ without growth factor support, white blood cell count ≥2,000/mm³, platelets ≥100,000/mm³ without transfusion, hemoglobin ≥9 g dl⁻¹ without transfusion, alanine aminotransferase and aspartate aminotransferase ≤3× the upper limit of normal (ULN), total bilirubin ≤1.5× ULN (≤3× ULN in patients with Gilbert syndrome) and serum creatinine ≤1.5× ULN or creatinine clearance ≥40 ml min⁻¹ calculated by the Cockcroft−Gault equation.

Patients were excluded if they had received prior systemic therapy for unresectable, locally advanced or metastatic RCC, including investigational agents, or if they had received radiation therapy for bone metastases within 2 weeks or any other radiotherapy within 4 weeks of study entry. Exclusions also applied to patients with clinically relevant ongoing complications from prior radiotherapy, those who were pregnant or breastfeeding or those planning to conceive during the study or within 120 days after the last immunotherapy dose. Additional exclusions included known immunodeficiency (for example, HIV infection or prior transplantation), systemic corticosteroid use >10 mg of prednisone equivalent daily or other immunosuppressive therapy (with allowances for adrenal replacement, short-term prophylaxis or inhaled/topical formulations) and systemic antibiotic use within 2 weeks of the first FMT. Patients with chronic intestinal disease (including celiac disease, inflammatory bowel disease, malabsorption or colonic tumors) or absolute contraindications to FMT (such as toxic megacolon or severe dietary allergies including shellfish, nuts or seafood) were not eligible. Use of concurrent systemic or local antineoplastic therapy was prohibited, although concurrent bisphosphonate or denosumab for bone metastases was permitted.

Other exclusions included active central nervous system metastases or leptomeningeal disease unless stable (≥4 weeks after radiotherapy or ≥8 weeks after major surgery), active autoimmune disease requiring systemic steroids or immunosuppressants (except vitiligo, type 1 diabetes or resolved childhood asthma/atopy), history of non-infectious pneumonitis requiring steroids or current pneumonitis, serious uncontrolled comorbidities (for example, uncontrolled cardiovascular disease, ischemia, arrhythmia, bleeding disorders, severe pulmonary disease, inflammatory bowel disorders or active systemic infection, including hepatitis B or C) and active infection requiring systemic therapy. Patients who had received live attenuated vaccines within 4 weeks of treatment initiation were excluded, although inactivated injectable influenza vaccines were permitted while intranasal live attenuated vaccines were prohibited. Additional exclusions included psychiatric or substance use disorders likely to interfere with study compliance and concurrent use of probiotic-containing food products during immunotherapy.

Trial sample size

The planned enrollment for the trial was 20 patients, determined prior to the first participant’s inclusion. The rationale for the sample size was that, at the time the study was initiated, there were no existing publications assessing the use of encapsulated FMT (LND101) in combination with doublet immunotherapy or immunotherapy plus targeted therapy in this patient population.

Given the novelty of the approach and the fact that this was the first study of encapsulated healthy donor FMT in combination with ICI-based regimens in mRCC, no formal statistical calculation could be applied to determine sample size. Instead, the rationale was safety-driven: a cohort of 20 patients was considered adequate to detect an unforeseen safety signal yet small enough to avoid and limit unnecessary exposure should the intervention prove unsafe or detrimental. Thus, 20 patients were determined to provide an appropriate balance between patient safety and the ability to generate meaningful preliminary safety data to inform the design of future randomized studies. The trial was overseen by an institutional DSMC composed of five members, including at least two independent oncologists not involved in the conduct of the study. The DSMC reviewed safety data at 6-month intervals. As an additional safety precaution, the first three patients were monitored for 3 months to assess for any new adverse event or those that occurred at a higher grade than those previously reported in clinical trials, according to the National Cancer Institute’s CTCAE v5.0. In the event of such findings, the DSMC was empowered to recommend protocol modification or cessation of the trial. This phase 1 study was not statistically powered to evaluate efficacy.

FMT product LND101

Feces collection and encapsulation were conducted under the supervision of the Division of Infectious Diseases at St. Joseph’s Hospital in London, Ontario, Canada. Each set of capsules (one full dose and two half-doses) contains material from a single donor only. Patients were required to consume 36−40 capsules for the initial full dose, followed by two half-doses of 20−25 capsules each, under supervision and within 2 hours of capsule defrosting, followed by a 30-minute period of observation.

Healthy donor recruitment

Public-facing campaigns (hospital press releases, advertisements, flyers and social media) were launched to recruit healthy individuals who were interested in being stool donors. Interested individuals were prescreened using a self-administered email screening questionnaire (age, location, occupation and basic health). Individuals who self-reported eligibility on the questionnaire were subsequently contacted over the phone by study coordinators to review eligibility (questionnaire and inclusion and exclusion criteria). After the screening, the individual’s health was assessed by laboratory tests, and the medical history was recorded. Final eligibility was confirmed after a comprehensive physician assessment.

Healthy stool donor inclusion and exclusion criteria

A healthy donor who has a body mass index of 18.5−30 kg m⁻² and who satisfies the following criteria was selected for donation^22,53.

Donors were excluded for any underlying metabolic disease, including hypertension, hyperlipidemia, diabetes, insulin resistance or atherosclerosis. Positive stool test or nasal swab results for any multidrug-resistant organisms (MDROs) (extended spectrum β-lactamase (ESBL)-producing enterobacteriaceae, vancomycin-resistant enterococci (VRE), carbapenem-resistant enterobacteriaceae (CRE/CPE), SARS-CoV-2 or methicillin-resistant Staphylococcus aureus (MRSA)). Any of the following symptoms 14 days before any stool collection: fever, cough, loss of taste, anosmia, fatigue/malaise, myalgia, sore throat, nausea, abdominal pain, diarrhea, dyspnea, chest pain, rash, conjunctivitis or headache. Any symptoms suggestive of SARS-CoV-2 or known SARS-CoV-2 infection within the last 3 months. New rash or spots indicative of monkeypox (Mpox). Any symptoms suggestive of the Mpox virus or may have been recently infected or exposed to the Mpox virus within the last 3 months. Donors were also excluded if they were men who have sex with men or women with a male partner who has sex with men. A history of any gastrointestinal or liver disorders or cancers, including, but not limited to, gastroesophageal reflux, peptic ulcer disease, celiac disease, inflammatory bowel disease (Crohn’s disease or ulcerative colitis), microscopic colitis, motility disorders (including gastroparesis and irritable bowel syndrome) or diverticular disease. Previous surgery to the intestine, liver or gallbladder (except remote appendectomy). History of any malignancy. Use within the last 3 months of any antibiotics. Hospitalization within the last 3 months. Recent travel to a developing country (within the last 3 months). New sexual partner (within last 3 months). Street drug use, family history of diabetes, coronary disease or gastrointestinal or liver disease, colon cancer or familial malignancy. Psychiatric history (major affective disorder, psychotic illness or ongoing use of any psychiatric medications). Any positive laboratory results for a transmissible pathogen. Alcohol intake with a cutoff value of <10 g per day in women and <20 g per day in men. Urine test for chlamydia and gonorrhea. Serology for HIV 1/2, HTLV 1/2, hepatitis A IgM, hepatitis BsAg, hepatitis BcAb, hepatitis C, cytomegalovirus, Epstein−Barr virus, Helicobacter pylori, syphilis, strongyloidiasis, schistosomiasis, amebiasis, anti-TTG antibody and, if travel history, trypanosomiasis. Throat swab for chlamydia and gonorrhea. Rectal swab for chlamydia, gonorrhea, MRSA and Mpox. Nasal swab for MRSA, SARS-CoV-2 and Mpox.

Capsule preparation

FMT capsules (LND-101) were prepared according to an established protocol as described in a Canadian study²⁹. In brief, donor stool was processed aerobically under biosafety level 2 conditions. The material was then encapsulated, flash frozen at −55 °C on dry ice and stored at −80 °C for up to 2 months before use. Capsules were thawed at room temperature approximately 30 minutes prior to ingestion. Donations of 80−100 g (for a full dose) and 50−60 g (for a half-dose) were processed individually without pooling by mixing in 0.9% normal saline and glycerol and were filtered using a stomacher bag. The filtrate was centrifuged, and the final sediment was mixed to incorporate residual liquid, allowing it to be pipetted into capsules. Size 1 gelatin capsules (PCCA 30-1005) were manually filled and over-encapsulated twice with size 0 (PCCA 30-1126) and size 00 (PCCA 30-3533) capsules. Finally, the capsules were flash frozen at −55 °C on dry ice and stored at −80 °C for up to 2 months until 30 minutes prior to ingestion. One donation of 80−100 g produced approximately 40 capsules (full dose), and 50−60 g produced 20−25 capsules (half-dose).

Frozen capsules were shipped on dry ice to the London Health Sciences Centre, where patients received their follow-up half-doses after the first full FMT treatment at St. Joseph’s Hospital, London. Capsule temperatures were continuously monitored using an EL-GFX-TC Thermocouple Data Logger to ensure stability during shipment. Upon arrival, capsules were stored at −80 °C and used for FMT within 2 months of manufacturing. The capsules are triple-encapsulated with hard gelatin shells to delay dissolution and protect against gastric acid, ensuring release in the distal small intestine and colon. SARS-CoV-2 testing was conducted according to our previously published protocol²². In brief, donors received a questionnaire before each stool sample dropoff, which included COVID-19 symptoms, positive COVID-19 tests, any infectious symptoms, including Mpox, and recent travel outside Canada. If the donors passed the questionnaire, they were allowed to drop off their stool samples. During each visit, donors were also required to provide a nasal swab for COVID-19 testing (polymerase chain reaction with reverse transcription (RT−PCR)), and a portion of the stool sample was also tested for SARS-CoV-2 RT−PCR and MDROs. Stool donations were processed and quarantined (−80 °C) until PCR and MDRO results became available.

Treatment regimen and assessments

Patients underwent a standard bowel preparation with 250−4,000 ml of polyethylene glycol with electrolytes (PEG3350) solution (as tolerated) the evening before the first FMT administration with LND101. In brief, after an early light supper the evening prior to the first full-dose FMT, patients underwent a standard polyethylene glycol electrolyte bowel preparation identical to that used for colonoscopy (up to 4 l, consumed in 200-ml aliquots every 15 minutes until clear watery stools were achieved). Water or clear liquids were permitted until 2 hours before capsule ingestion, which occurred the following morning over approximately 30 minutes under supervision. Patients received LND101 orally on a defined schedule, initiated prior to standard immunotherapy (for example, nivolumab/ipilimumab) or immunotherapy combined with a vascular endothelial growth factor (VEGF)-targeted agent (for example, lenvatinib and axitinib). Two half-doses (20−25 capsules, equivalent to 50−60 g of feces) were administered at weeks 3 and 6 without bowel preparation. The specific systemic therapy was selected by the treating physician in accordance with Canadian guidelines.

Patients received either dual ICIs or anti-PD-1 therapy combined with a VEGF inhibitor. Publicly funded regimens followed standard protocols: nivolumab (3 mg kg⁻¹ every 3 weeks) plus ipilimumab (1 mg kg⁻¹ every 3 weeks) up to four cycles, followed by maintenance nivolumab (240 mg every 2 weeks or 480 mg every 4 weeks) or a combination of pembrolizumab with a VEGF-targeted agent (for example, lenvatinib and axitinib).

To allow microbiota engraftment, the first immunotherapy dose was given at least 7 days after the initial FMT (Fig. 1b). Treatment continued until disease progression, unacceptable toxicity, completion of therapy or death, at the physician’s discretion.

Radiographic assessments were performed every 12 weeks for the first 2 years and then at least annually for up to 5 years or more often if clinically indicated. Imaging modalities included computed tomography or magnetic resonance imaging, with efforts made to use the same modality throughout follow-up.

Tumor response was evaluated according to RECIST v1.1, with immune-related RECIST (iRECIST) applied when relevant. CR was defined as the disappearance of all targeted lesions; PR as ≥30% reduction in the sum of target lesion diameters from baseline; PD as a ≥20% increase in target lesion size or the appearance of new lesions; and SD as the absence of CR, PR or PD for ≥3 months. ORR was defined as the proportion of patients with CR or PR, and clinical benefit was defined as the proportion of patients with CR, PR or SD lasting ≥6 months. For correlative analyses, patients achieving CR or PR were categorized as responders, and those with SD or PD were classified as non-responders.

Sample and data collection

Clinical data, including demographics, imaging results and adverse events, were prospectively collected at each visit and entered into a secure REDCap electronic database (version 12.4.17; Vanderbilt University, 2022). Access was restricted to authorized clinical research staff and the principal investigator. Each participant was assigned a unique study ID with the master linking log securely to preserve confidentiality.

Biological samples, including stool and blood, were collected at five predefined timepoints: (1) at baseline (prior to any intervention), (2) 1 week after the first full-dose FMT and just before initiating systemic therapy, (3) 4 weeks after the first FMT and prior to the second immunotherapy dose, (4) 7 weeks after the first FMT and prior to the third immunotherapy dose and (5) approximately 10 weeks after the first FMT and prior to the fourth immunotherapy dose. Samples 3 and 4 were obtained 1−3 days after the first and second half-dose FMTs, respectively.

At each timepoint, clinical assessments were performed alongside routine laboratory evaluations, including renal and liver function tests, complete blood count, electrolytes and lactate dehydrogenase. Organ function was documented at baseline and monitored throughout follow-up. Participants were free to withdraw consent at any time without affecting their clinical care. Samples collected prior to withdrawal were retained for analysis.

Adverse event monitoring and management

Adverse events were graded using CTCAE v5.0 (ref. ⁵⁴). Toxicities occurring after the first FMT but before initiation of immunotherapy-based combination were attributed to FMT. Adverse events occurring within 90 days of starting immunotherapy were considered related to the combination of FMT and immunotherapy-based combination, whereas those occurring beyond 90 days were attributed to immunotherapy-based combination alone. All adverse events were managed in accordance with standard clinical practice guidelines. Immunotherapy and/or VEGF-TKI were temporarily held or permanently discontinued in cases of severe toxicity, at the discretion of the treating oncologist. All events were reviewed by the trial principal investigator to ensure standardized attribution and grading.

Dose modification guidelines for treatment-related adverse events

The Cancer Care Ontario (CCO) Immune Checkpoint Inhibitor Side Effect Toolkit was used to guide dose modifications. The full guidelines are publicly available on the CCO website: https://www.cancercareontario.ca/en/guidelines-advice/modality/immunotherapy/immune-therapy-toolkit.

Study endpoints

Primary endpoint

The primary endpoint of the PERFORM trial was safety and tolerability of LND101 when administered in combination with immunotherapy-based regimens. Safety was assessed by the incidence and severity of irAEs graded according to CTCAE v5.0 (ref. ⁵⁴). Attribution categories were unrelated, unlikely, possible, probable or definite. irAEs were defined as those with attribution of possible/probable/definite. For the primary analysis, we report irAEs of any grade using a uniform threshold of ≥possible. In a prespecified sensitivity analysis, to reduce potential noise from non-specific low-grade events, we restricted grade 1−2 treatment-related adverse events (TRAEs) to probable/definite while retaining possible/probable/definite for grade 3 or higher. The combination was considered safe if irAEs did not occur more frequently or with greater severity than expected from the relevant product monograph of each backbone regimen. Given that patients received different standard regimens (for example, nivolumab/ipilimumab, pembrolizumab/axitinib and pembrolizumab/lenvatinib), the incidence and spectrum of observed irAEs were interpreted relative to the established toxicity profile of each respective regimen. This framework enabled assessment of whether the addition of LND101 introduced excess or unexpected immune-related toxicity beyond the known safety spectrum. As prespecified in the protocol, all safety analyses were restricted to irAEs, which represent the most clinically relevant toxicities in the context of microbiome−immunity interactions.

Secondary endpoints

Secondary endpoints included the incidence of grade 3 or higher irAEs occurring from treatment initiation to 120 days after the last immunotherapy dose; clinical response assessment by RECIST v1.1 (ref. ²⁴); longitudinal changes in gut microbiome composition from fecal samples; immune activation markers (for example, circulating cytokines and immune cell subsets); and health-related quality of life, assessed using the EQ-5D-5L questionnaire.

Exploratory endpoints

Exploratory endpoints included characterization of the tumor microenvironment and immune profile and assessment of PFS and overall survival using the Kaplan−Meier method. In accordance with the trial protocol, adverse events focused on immune-related events, which represent the most clinically relevant toxicities in the context of microbiome−immune interactions. This prespecified focus also guided downstream translational analyses for the exploratory endpoints, ensuring alignment between clinical and mechanistic endpoints. By focusing on irAEs, the study was able to interrogate mechanistic links between the microbiome and immune-related toxicity without confounding from regimen-specific non-immune adverse events.

Statistical analyses

Safety analyses were conducted in the ITT population, defined as all patients who received at least one dose of LND101 in combination with immunotherapy-based therapy (n = 20). Overall survival and correlative analyses for toxicity (microbiome, immune and metabolomic) were also performed in the ITT population.

Efficacy analyses and correlative analyses based on clinical response were conducted in the per-protocol population, defined as all patients with measurable disease at baseline according to investigator assessment (n = 18). The ORR was calculated as the proportion of patients achieving CR or PR per RECIST v1.1. The clinical benefit rate was defined as CR or PR or SD lasting ≥6 months. PFS was analyzed in the per-protocol population (n = 18). Kaplan−Meier methods were used to estimate PFS and overall survival, with medians and 95% confidence intervals reported.

Patient-reported quality of life was analyzed in all patients with available EQ-5D-5L assessments. Baseline distributions across the five domains and global VAS scores were summarized descriptively. Longitudinal changes from baseline were evaluated through median score differences and distribution shifts across cycles, with particular attention to cycle 4 to align with early clinical response and safety assessment windows. Exploratory subgroup analyses were performed stratified by the occurrence of grade 3 or higher irAEs.

Descriptive statistics were used to summarize patient characteristics, safety events and correlative endpoints. No formal hypothesis testing or adjustment for multiple comparisons was planned, consistent with the exploratory nature of this phase 1 study.

Metagenomics gut microbiome analysis

DNA extraction and sequencing

Sequencing libraries were prepared using the Illumina DNA LP Tagmentation kit, according to the manufacturer’s instructions. Libraries were dual-indexed and sequenced as paired-end 300-bp reads on the Illumina NovaSeq X Plus platform. In total, 3,914.1 Gb of raw sequencing data were generated, corresponding to an average of approximately 57.3 million paired-end reads per sample (150-bp average read length) across 113 samples prior to quality control and preprocessing.

Quality control and preprocessing

Initial preprocessing involved trimming sequencing primers, removing reads shorter than 75 bp and discarding low-quality reads (quality score < Q20) using Trimmomatic version 0.39 (ref. ⁵⁵). Reads containing two or more ambiguous nucleotides were also filtered out. Contaminant sequences were removed by aligning reads to the human reference genome (hg19) and the phiX174 control genome using Bowtie 2 version 2.5.4 (ref. ⁵⁶) with the very-sensitive-local preset to conservatively exclude host genome reads. After preprocessing, an average of 55.8 million paired-end reads per sample remained and were retained for downstream analyses.

Taxonomic profiling

Taxonomic composition was profiled from shotgun metagenomic data to the species level using MetaPhlAn version 4.1.1 (ref. ⁵⁷) with the mpa_vJune23 reference database. This profiling identified a total of 1,494 microbial species and 1,591 species-level genome bins (SGBs). The unclassified_estimation parameter was set to ensure accurate estimation of unclassified. Bowtie 2 alignment and SAM output files were retained for downstream analyses. Relative abundance tables were merged using MetaPhlAn’s merge_metaphlan_tables.py utility and normalized to CPM for subsequent statistical analysis.

Functional profiling

Functional profiling was conducted using the HMP Unified Metabolic Analysis Network (HUMAnN) version 3.9 (ref. ⁵⁸). For nucleotide alignment, the ChocoPhlAn database (mpa_vJun23) consistent with MetaPhlAn was used, and UniRef90 served as the protein database. The HUMAnN pipeline was executed with DIAMOND version 2.1.8 (ref. ⁵⁹). The output included gene family abundances, pathway abundance and pathway coverage tables. These tables were merged using the humann_join_tables utility. Gene families were regrouped to EC numbers using the humann_regroup_table function with the uniref90_level4ec mapping. All tables were CPM normalized for downstream statistical analysis.

Data integration and processing in R

After functional profiling, all resulting tables were transferred from the HPC cloud environment to a local workstation for subsequent analysis in R version 4.5.0. A unified list structure was constructed in R, incorporating all relevant data layers, including metadata, species-level abundance, ECs, gene families, pathway abundance, pathway coverage, genus-level profiles and SGBs. Sample identifiers were standardized across all tables to ensure consistency, allowing for accurate linkage between metadata and metagenomic data. MetaPhlAn annotates features hierarchically from Kingdom to SGB, and filtering was performed to isolate the species and SGB levels for downstream analysis. The dplyr R package version 1.1.4 and tidyr R package version 1.3.1 were used for data manipulation, and the ggplot2 version 3.5.2 package was used for all visualizations.

α-Diversity

α-Diversity was calculated at the species level using Shannon and inverse Simpson indices via the vegan R package (version 2.7-1)⁶⁰. Shannon index accounts for both richness (number of species) and evenness (distribution of abundances), making it sensitive to rare species. By contrast, the inverse Simpson index emphasizes dominant species and provides a measure of how evenly distributed the microbial community is. Together, these indices offer complementary views of within-sample diversity.

Statistical analyses

To assess whether α-diversity differed between clinical subgroups, we performed one-way ANOVA separately at each timepoint, using the diversity metric as the outcome and clinical subgroup as the predictor. To correct for multiple comparisons and identify specific differences between group pairs, Tukey’s honest significant difference (HSD) post hoc test was applied after each ANOVA. This allowed for pair-wise comparisons while controlling the family-wise error rate and minimizing type I error. Timepoints with fewer than two group levels were excluded from testing. Statistically significant pair-wise comparisons (adjusted P < 0.05) were annotated in visualizations.

Species-level β-diversity analysis

β-Diversity was first assessed at the species level using Bray−Curtis dissimilarity, which captures differences in community composition by accounting for shared species and their relative abundances. This measure emphasizes differences in dominant taxa, making it particularly useful for identifying compositional shifts in microbial abundance profiles. To visualize these compositional changes over time, Bray−Curtis dissimilarity matrices were subjected to principal coordinate analysis (PCoA).

Statistical analyses

Group-level differences were statistically assessed using PERMANOVA (adonis2 function, vegan package, 999 permutations) at baseline (timepoint 1) and 10 weeks after FMT (timepoint 5) to determine whether microbial communities diverged based on clinical subgroups.

Ordination and discriminatory species identification

We further examined which microbial species were most strongly associated with the observed ordination axes using environmental vector fitting (envfit function, vegan). This analysis identified species whose relative abundance patterns aligned significantly with the PCoA structure.

Statistical analyses

The significance of each species was tested through 999 permutations, and only those with P < 0.05 were retained. From these, the top 12 species with the highest vector magnitude (calculated from their contributions to PCoA axes 1 and 2) were selected and plotted as arrows to highlight their directional influence on community variation.

Longitudinal patient−donor microbiome similarity

To quantify the similarity between each patient’s microbiome and their respective donor’s profile over time, two additional dissimilarity measures along with Bray−Curtis—Jaccard and Aitchison distances—were calculated.

Jaccard dissimilarity, used on species-level data, is based on presence/absence and evaluates overlap in community membership, regardless of abundance. This allowed us to track whether patients were acquiring donor-like species, even in small amounts.

Aitchison distance, used on EC profiles, applies a centered log-ratio (CLR) transformation, making it more appropriate for compositional functional data. It enabled us to assess convergence or divergence in functional potential between patients and their donors over time.

For species (Jaccard and Bray−Curtis) and enzyme (Aitchison) distances, we calculated the dissimilarity between each patient and their respective donor across timepoints. These values were normalized by each patient’s baseline (pre-FMT) distance to generate fold changes. This normalization allowed for patient-specific tracking of relative shifts in similarity to the donor.

Statistical analyses

Two types of statistical comparisons were conducted on the fold change data: (1) within-group comparisons across timepoints, to assess whether each clinical group showed a significant shift in donor similarity over time, and (2) between-group comparisons at timepoints 4 and 5 only, to assess the similarity of samples to the donor at 7 weeks and 10 weeks after FMT. For each test, the Shapiro−Wilk test was first applied to determine the distribution of the data. If both baseline and current values followed a normal distribution, a two-sample t-test was used; otherwise, a Wilcoxon rank-sum test was applied. To control for multiple testing in the within-group comparisons, P values were adjusted using the Benjamini−Hochberg false discovery rate (FDR) method in each test.

Species-level differential abundance analysis

Differential abundance analysis of species was conducted using the lefser function from the lefser package (version 1.18.0)⁶¹. Patients were grouped based on grade 3 toxicity status for this analysis. Linear discriminant analysis effect size (LEfSe) first applies a non-parametric Kruskal−Wallis test to identify features with significantly different abundances between the two groups. It then uses linear discriminant analysis (LDA) to estimate the effect size of each discriminatory feature. A heatmap was generated using the pheatmap R package (version 1.0.13) for species with LDA > 2 to visualize their abundance patterns across all timepoints and both clinical groups, highlighting the longitudinal dynamics of discriminatory taxa.

Statistical analyses

Features with LDA > 2 were retained for downstream analysis. The LDA score reflects both statistical significance and effect size; features passing this threshold also meet the default significance cutoff of P < 0.05 from the Kruskal−Wallis test used within the LEfSe method. Therefore, no additional statistical testing was applied to features selected based on this LDA threshold.

Enzyme-level differential abundance analysis

Differential analysis of EC numbers was performed using MaAsLin3 (ref. ²⁸) on CPM-normalized HUMAnN output. MaAsLin3 fits generalized linear models for each microbial feature to test for associations with metadata while adjusting for covariates. It supports multivariable modeling, normalizes and transforms input data and applies FDR correction to control for multiple testing. For each feature, MaAsLin3 reports effect sizes, P values and FDR-adjusted q values, enabling robust identification of differentially abundant features.

In this analysis, two covariates were included: toxicity status (whether the recipient developed grade 3 toxicity) and timepoint, with donor samples assigned to T0. Donor samples were grouped according to the toxicity status of their respective recipients, thereby assessing whether features in the donor microbiota were associated with future toxicity outcomes in patients. This combined selection approach was implemented to ensure that the identified enzyme features were both significantly different at baseline between donors (based on recipient outcome) and retained in recipients at 10 weeks after FMT (timepoint 5), suggesting potential functional persistence.

Statistical analyses

Candidate enzyme features were selected based on a joint P < 0.05, a joint q < 0.1 and an absolute model coefficient > 0.1, indicating both statistical significance and a meaningful effect size. Only enzymes that passed these thresholds at both the donor (T0) and patient (T5) timepoints were retained as candidate features.

Compositional taxonomic profiling of top taxa

Species were ranked by their total cumulative abundance (CPM) across all collected trial samples. Taxa names were cleaned by removing prefixes and truncating them to species-level labels. Separate compositional visualizations were created for donors and patients. For donors, the mean CPM values of each top taxon were calculated across all donations of a donor and displayed per donor. For patients, the same top taxa were profiled over time (T1 to T5). The patient’s label was modified to reflect their experience of grade 3 toxicity.

Strain profiling

Strain profiling was performed using StrainPhlAn version 4.0 (ref. ⁶²) on SGBs identified by MetaPhlAn. To improve the sensitivity and resolution of strain-level engraftment detection, we incorporated publicly available metagenomic samples from three previously published FMT trials: Baruch et al.¹⁴, Davar et al.¹⁵ and Routy et al. (MIMIC study)²⁹. Including these datasets enhanced marker selection quality and increased the phylogenetic signal for strain tracking across samples.

For each sample, consensus marker genes were extracted using the sample2markers.py and extract_markers.py utility on the MetaPhlAn SAM output, and the list of relevant SGBs was generated by enabling the print_clades_only option in StrainPhlAn. Marker filtering parameters were set to include rare but valid SGBs by setting the minimum percentage thresholds on per-sample marker coverage to 0, the minimum number of samples sharing a marker to 20 and requiring that each marker be present in at least 10% of the samples (–sample_with_n_markers_perc 0, –sample_with_n_markers 20, –marker_in_n_samples_perc 10).

Phylogenetic trees were reconstructed for each SGB using marker gene alignments, and normalized pair-wise phylogenetic distances were calculated. To define same-strain pairs, SGB-specific thresholds were optimized using Youden’s index, which maximizes the separation between longitudinal (same-host) and unrelated (inter-host) sample pairs.

Strain engraftment was inferred when the post-FMT recipient shared a strain with their donor for a given SGB and this strain was absent in the recipient’s pre-FMT sample. The engraftment rate was defined as the number of engrafted strains divided by the number of strains that could engraft, where engraftable strains are those detected in both the donor and the recipient after FMT sample but not present in the recipient at baseline. This approach allowed high-resolution detection of donor-derived strain persistence after FMT.

Statistical analyses

Two types of statistical comparisons were conducted on the strain engraftment rate data: (1) within-group comparisons, to assess whether each clinical group showed significant changes in engraftment relative to baseline, and (2) between-group comparisons at timepoints 2, 3, 4 and 5, to evaluate differences in engraftment rates based on clinical outcomes. A Wilcoxon rank-sum test was applied for each comparison. Multiple testing correction using the Benjamini−Hochberg FDR method was performed separately for each of the within-group comparisons and the between-group comparisons. This ensured that P values were adjusted independently for each analysis context.

S. copri comparison across FMT trials

Metagenomic samples from three published FMT trials (Baruch et al.¹⁴, Davar et al.¹⁵ and Routy et al.²⁹) were downloaded and processed as described above. Taxonomic profiling was performed using MetaPhlAn version 4.1.1 with the mpa_vJune23 database. S. copri abundance (CPM) was extracted from each trial’s merged abundance table and averaged per patient at the pre-FMT and post-FMT timepoints, if multiple samples were collected. Significant differences in S. copri levels were assessed using the Wilcoxon rank-sum test, with P values adjusted using the Benjamini−Hochberg FDR method.

Blood sample collection and processing

Blood samples were collected in EDTA anticoagulant vacutainers. For plasma collection, blood was centrifuged at 200g for 20 minutes with no brake. Plasma was collected and frozen at −80 °C for metabolomics and cytokine and chemokine quantification. For PBMC isolation, blood was diluted with PBS, and PBMCs were isolated using Lymphoprep density gradient medium (STEMCELL Technologies, 07861) in SepMate tubes (STEMCELL Technologies, 85460). Blood was layered on Lymphoprep medium and centrifuged at 1,200g for 15 minutes at room temperature with brake. PBMCs were collected, and red blood cells were removed using ACK lysing buffer and washed. PBMCs were resuspended at a concentration of 5 × 10⁶ cells per milliliter in freezing medium (RPMI 1640 + 12.5% human serum albumin + 10% DMSO). Samples were frozen in a Mr. Frosty container for 24−48 hours at −80 °C and stored at −150 °C for long-term storage.

Multiparameter flow cytometry and analysis of PBMCs

Cryopreserved PBMCs were thawed and resuspended in RPMI 1640 media supplemented with 10% heat-inactivated FBS and incubated for 1 hour at 37 °C. One million cells were then stained with Zombie Aqua for 20 minutes at room temperature. Fc receptors were blocked with TruStain FcX for 10 minutes at room temperature, followed by labeling with a cocktail of extracellular antibodies (Supplementary Table 4) in the presence of Brilliant Stain buffer and Monocyte Blocker for 20 minutes on ice. Cells were subsequently fixed with 2% paraformaldehyde and resuspended in staining buffer (5% heat-inactivated FBS in PBS) for acquisition. For FOXP3 intracellular staining, cells were fixed with the eBioscience Foxp3 / Transcription Factor Staining Buffer Set for 50 minutes at room temperature. Cells were then permeabilized, and Fc receptors were blocked again with TruStain FcX for 10 minutes at room temperature. Anti-FOXP3 staining was performed in permeabilization buffer for 30 minutes at room temperature, followed by fixation with 2%. paraformaldehyde. Data were acquired on a BD FACSymphony A1 cytometer, followed by analysis using FlowJo version 10.6.2 and GraphPad Prism 9 software.

Data were first normalized using CytoNorm to remove batch effects, followed by supervised gating analysis using fluorescence-minus-one controls (Supplementary Figs. 1 and 2). For unsupervised clustering analysis, T cells (CD3⁺) and monocytes (CD14⁺CD16^+/−) were first gated on live CD45⁺ cells. Using DownSample, each population was downsampled to 10,000 events for T cells and monocytes. Sample populations were concatenated, and unsupervised dimensionality reduction was conducted using t-SNE, followed by clustering analysis using FlowSOM and ClusterExplorer. Statistical analysis was conducted using a two-sided unpaired Wilcoxon rank-sum test for comparisons at T5 or a two-sided Wilcoxon matched-pairs signed-rank test with Benjamini−Hochberg correction for multiple within-group comparisons

Cytokine and chemokine analysis

Plasma cytokine levels were quantified by Eve Technologies using the Human Cytokine 48-Plex Discovery Assay (HD48). The multiplex assay was performed using a Luminex xMAP technology platform, allowing simultaneous quantification of 48 human cytokines, chemokines and growth factors. Each specimen on each panel was run in singlet on the Luminex xMAP technology, and data were analyzed by Eve Technologies using standard curve interpolation. Statistical analysis was conducted using a two-sided unpaired Wilcoxon rank-sum test for comparisons at T5 or a two-sided Wilcoxon matched-pairs signed-rank test with Benjamini−Hochberg correction for multiple within-group comparisons

Metabolomics

Metabolomic analysis of patient plasma was conducted according to a previously published targeted metabolomics method⁶³. In brief, 20 μl of plasma was added to 80 μl of high-performance liquid chromatography (HPLC)-grade methanol, and 5 μl of diluted internal standard (diluted 1:1 with HPLC-grade methanol; Cambridge Isotope Laboratories, MSK-A2-1.2) was added to each sample, which was then vortexed, incubated for 30 minutes at −80 °C to precipitate proteins and then centrifuged at 15,000g to clarify. Then, 10 μl of supernatant was diluted in 990 μl of buffer containing 95% acetonitrile and 5% 20 mM ammonium carbonate (pH 9.8). Quality control samples were prepared by pooling 10 μl of each sample. All samples, including quality control samples, were then analyzed by hydrophilic interaction or reversed-phase liquid chromatography and selected reaction monitoring (SRM) with a SCIEX QTRAP 5500 triple quadrupole linear ion trap tandem mass spectrometer.

Data were captured using Analyst version 1.6.2 software (SCIEX); peak integration was performed using Skyline version 24.1. An in-house R script was used for data normalization against pooled quality control samples and removal of metabolites with low signal or high quality control variance (version 4.4.2, http://www.r-project.org, accessed on 15 July 2021). Scaling, heatmaps and t-tests and multiple hypothesis correction (Benjamini−Hochberg) were performed using MetaboAnalystR (version 4.0). The survival graphs were generated using the ITT population. Survival plots were generated using the R package survminer (version 0.5)⁶⁴.

Summary of trial protocol amendments

Across successive protocol amendments for the PERFORM trial, updates primarily focused on expanding eligibility criteria, clarifying safety endpoints and broadening translational objectives.

The protocol title and scope were revised to include patients treated with either doublet immunotherapy or anti-PD-1 plus VEGFR-TKI combinations, reflecting evolving standards of care in mRCC. Therefore, corresponding modifications were made throughout the Background, Inclusion Criteria and Treatment Plan sections to incorporate regimens such as pembrolizumab/axitinib, pembrolizumab/lenvatinib and nivolumab/cabozantinib. The primary endpoint remained safety—irAEs assessed by CTCAE v5.0—and secondary endpoints included clinical response (RECIST v1.1), gut microbiome, immune correlates and patient-reported quality of life (EQ-5D-5L).

Methodological refinements included mandating stool sample 3 for longitudinal microbiome analysis and adding new collaborators for biospecimen analyses to enable multiomic integration (metagenomics, metabolomics, immune profiling and tumor epigenomics). Shipping instructions, appendix references and consent documents were updated accordingly to ensure consistency and transparency in sample handling and data sharing. Administrative corrections included reference formatting, version control and inclusion of new institutional contact information. Overall, these amendments maintained the original study intent—evaluating safety, tolerability and mechanistic correlates of healthy donor encapsulated FMT combined with standard ICI-based regimens—while broadening translational depth and improving operational clarity to support future multisite expansion.

Reporting summary

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