CoChat
Sign Up
Cited by 165Gut microbiota and healthCancer Immunotherapy and BiomarkersProbiotics and Fermented Foods

A gut microbial signature for combination immune checkpoint blockade across cancer types

Open access
Gunjur, Ashray, Shao, Yan, Rozday, Timothy, Klein, Oliver, Mu, Andre, Haak, Bastiaan W., Markman, Ben, Kee, Damien +10 more·2024·Nature Medicine
DOI 10.1038/s41591-024-02823-zPMC PMC10957475
Open this paper in CoChat — free accountOpen PDF

Abstract

Immune checkpoint blockade (ICB) targeting programmed cell death protein 1 (PD-1) and cytotoxic T lymphocyte protein 4 (CTLA-4) can induce remarkable, yet unpredictable, responses across a variety of cancers. Studies suggest that there is a relationship between a cancer patient’s gut microbiota composition and clinical response to ICB; however, defining microbiome-based biomarkers that generalize across cohorts has been challenging. This may relate to previous efforts quantifying microbiota to species (or higher taxonomic rank) abundances, whereas microbial functions are often strain specific. Here, we performed deep shotgun metagenomic sequencing of baseline fecal samples from a unique, richly annotated phase 2 trial cohort of patients with diverse rare cancers treated with combination ICB ( n = 106 discovery cohort). We demonstrate that strain-resolved microbial abundances improve machine learning predictions of ICB response and 12-month progression-free survival relative to models built using species-rank quantifications or comprehensive pretreatment clinical factors. Through a meta-analysis of gut metagenomes from a further six comparable studies ( n = 364 validation cohort), we found cross-cancer (and cross-country) validity of strain–response signatures, but only when the training and test cohorts used concordant ICB regimens (anti-PD-1 monotherapy or combination anti-PD-1 plus anti-CTLA-4). This suggests that future development of gut microbiome diagnostics or therapeutics should be tailored according to ICB treatment regimen rather than according to cancer type.

AI snapshot

Written from the paper’s full text.

TL;DR

In a phase 2 trial of combination anti-PD-1 plus anti-CTLA-4 in diverse rare cancers (n = 106), strain-resolved baseline gut microbial abundances predicted response and 12-month PFS better than species-level abundances or clinical factors. A meta-analysis of six external cohorts (n = 364) found cross-cancer and cross-country validity only when training and test cohorts used concordant ICB regimens, suggesting microbiome signatures are regimen specific.

Key findings

  • Strain-resolved abundances gave the best predictive performance: AUC = 0.73 for response versus progression (RvsP) and AUC = 0.70 for 12-month PFS, versus clinical factors alone at AUC = 0.56 for RvsP and AUC = 0.65 for PFS12; performance increased monotonically with taxonomic resolution (Kendall P = 1.1 × 10⁻¹¹ for RvsP, P = 7.1 × 10⁻¹⁵ for PFS12).
  • Leave-one-histology-out cross-validation across GYN, NEN and UGB cancer cohorts yielded an overall mean AUC = 0.75, supporting tumor-agnostic potential within the trial.
  • The top 22 strains were mostly gram-positive (20/22) and Firmicutes (18/20); positive strains clustered in Ruminococcaceae including four Faecalibacterium strains, while negative strains included Bifidobacterium dentium, Akkermansia muciniphila B and Spyradocola merdavium. The acetyl-CoA butyrate pathway was complete in 5/5 positive top-22 Ruminococcaceae and 0/2 negative top-22 Ruminococcaceae, but strain-agnostic but + buk abundance was not significantly enriched in responders.
  • External validation showed modest generalizability of the CA209-538 strain-RvsP signature to CICB cohorts (mean AUC = 0.65) but no generalizability to anti-PD-1 cohorts (mean AUC = 0.51); cross-study predictions were better for concordant than discordant ICB regimens (Mann-Whitney U test P = 2.8 × 10⁻⁷).

Methods

  • Discovery cohort: CA209-538 phase 2 trial (NCT02923934) of nivolumab plus ipilimumab in 120 enrolled patients with advanced rare cancers; 106 baseline fecal samples underwent deep shotgun metagenomic sequencing (median 20.4 million paired-end reads). A custom genome-resolved reference database of 1,397 strain genomes covering 904 known species plus 34 new strains was used, with Bowtie 2 and inStrain for mapping and quality control.
  • Supervised machine learning: random forest classifiers with 20-times repeated fivefold cross-validation and a 1,000-iteration random hyperparameter search; endpoints were RvsP (CR/PR versus PD/cPD, excluding SD n = 29) and PFS12; SHAP TreeExplainer was used for feature importance.
  • Meta-analysis: six external shotgun metagenomic studies (n = 364; total n = 470, n = 383 after excluding SD) were reanalyzed uniformly and split by ICB regimen (CICB versus anti-PD-1 monotherapy); pairwise train/test comparisons across cohorts and PERMANOVA for metadata variance were performed.

Limitations

  • Stated: small individual and total sample sizes limit statistical power; metagenome-assembled genomes may contain errors such as fragmentation, assembly breaks and contamination; lack of fecal strain culturing due to the collection medium prevented in vitro and in vivo mechanistic testing and causality confirmation; external signature performance fell short of clinical use.
  • Stated/evident: cross-cohort heterogeneity in fecal collection and DNA extraction methods was present, and the authors did not batch-correct beyond CLR transformation, so technical and geographic confounding may remain.
  • Evident: the discovery cohort comprised Australian patients with rare cancers while external cohorts were mostly melanoma; the primary endpoint excluded SD patients, and leave-one-histology-out analyses had small groups, so generalizability and the utility of predictions within SD remain uncertain.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41591-024-02823-z

Cites

97 of 97 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 29 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    A PERMANOVA of baseline microbial variance versus a moving PFS threshold revealed a peak association at <4 months.

    “A PERMANOVA of baseline microbial variance versus a moving PFS threshold revealed a peak association at <4 months (Extended Data Fig. 1e)”
    Quote in the paperNot a substantial claim
  • Findingown data
    § A gut microbial signature for combination immune checkpoint blockade across cancer types

    There is a negative monotonic association between neutrophil-to-lymphocyte ratio (NLR) and BOR (Kendall P = 0.0033).

    “and a negative monotonic association between neutrophil-to-lymphocyte ratio (NLR) and BOR (Kendall P = 0.0033) (Extended Data Fig. 1c).”
    Quote in the paperNot a substantial claim
  • Findingown data
    § A gut microbial signature for combination immune checkpoint blockade across cancer types

    There is a strong positive monotonic association between albumin and BOR (Kendall P = 0.0056).

    “we found a strong positive monotonic association between albumin and BOR (Kendall P = 0.0056)”
    Quote in the paperNot a substantial claim
  • Findingown data

    These top 22 strains were neither rare (<5% prevalent) nor core (>50% prevalent) taxa within the cohort.

    “Interestingly, these strains were neither rare (<5% prevalent) nor core (>50% prevalent) taxa within our cohort (Extended Data Fig. 3b).”
    Quote in the paperNot a substantial claim
  • Findingown data

    There are gross microbial compositional differences by BOR group (PERMANOVA P = 0.0319).

    “We then assessed intersample beta diversity using the Aitchison distance and also found gross microbial compositional differences by BOR group (permutational multivariate analysis of variance (PERMANOVA) P = 0.0319) (Fig. 1f).”
    Quote in the paperSubstantial claim
  • May combine two claims. 24
  • Classified as probably not a substantial claim. 27

  • May combine two claims.

    Findingown data

    Most of the 15 negative strains were Firmicutes, with 6, 3 and 2 (of the 15) strains belonging to the Lachnospiraceae, Oscillospiraceae and Ruminococcaceae families, respectively.

    “As before, most were Firmicutes, with 6, 3 and 2 (of the 15) strains belonging to the Lachnospiraceae, Oscillospiraceae and Ruminococcaceae families, respectively.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § A gut microbial signature for combination immune checkpoint blockade across cancer types

    Overall survival significantly differed by histology in the CA209-538 cohort.

    “As expected, overall survival (OS) significantly differed by histology (Extended Data Fig. 1a)”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The most 'beneficial' strains (higher strain abundances shifted predictions toward 'response') clustered in one clade of the Ruminococcaceae family, with four being strains within the Faecalibacterium genus.

    “The most 'beneficial' strains (that is, higher strain abundances shifted predictions toward 'response') clustered in one clade of the Ruminococcaceae family, with four being strains within the Faecalibacterium genus.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 22 top strains identified by SHAP are biologically meaningful and not artifacts of the model.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Eight of the negative strains belonged to thus far uncultivated (and thus unnamed) species.

    “Notably, eight of these strains belonged to thus far uncultivated (and thus unnamed) species.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Strain-resolved abundances provided the best predictive performance (AUC = 0.73 for RvsP, AUC = 0.70 for PFS12), significantly outperforming species-level abundances.

    “In particular, strain-resolved abundances provided the best predictive performance (AUC = 0.73 for RvsP, AUC = 0.70 for PFS12), significantly outperforming the more common species-level abundances.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Twenty of the 22 top strains were gram positives, with most (18 of 20) belonging to the Firmicutes (Bacillota) phylum.

    “This demonstrated that 20 of the 22 strains were gram positives, with most (18 of 20) belonging to the Firmicutes (Bacillota) phylum.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    At the 98% genomic identity threshold, the custom strain reference library included n = 35 distinct Faecalibacterium strains (from n = 13 distinct species), with the most important (and prevalent) clustering near the F. prausnitzii D phylogenetic clade.

    “In keeping with this, at the 98% genomic identity threshold, our custom strain reference library included n = 35 distinct Faecalibacterium strains (from n = 13 distinct species), with the most important (and prevalent) clustering near the F. prausnitzii D phylogenetic clade (Supplementary Fig. 2).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Despite being trained on binary RvsP, the predicted probabilities of patients were correctly ranked by their actual BOR category (Kendall P < 2.2 × 10 -16), including central predictions for the SD group that were 'unseen' during model training (n = 29).

    “Notably, despite being trained on binary RvsP, the predicted probabilities of patients were correctly ranked by their actual BOR category (Kendall P < 2.2 × 10 -16 ), including (on average) central predictions for the SD group that were 'unseen' during model training (n = 29) (Extended Data Fig. 2c).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § A gut microbial signature for combination immune checkpoint blockade across cancer types

    Patients with rapid clinical progression (cPD) had significantly lower albumin and higher NLR.

    “This was particularly driven by patients with rapid clinical progression (clinical PD (cPD)) having significantly lower albumin and higher NLR”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Strain-level response signatures show external generalizability across cancer histology types and countries, both within the trial and externally to metastatic melanoma cohorts from other industrialized countries.

    “Furthermore, we show the external generalizability of strain-level response signatures across cancer histology types and countries, both within the trial (comparing across the predetermined histology cohorts) and externally (to metastatic melanoma cohorts from other industrialized countries).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingreasoning

    Six studies could be included in the meta-analysis.

    “Finally, of the remaining ten studies, four were excluded as they did not report response, yielding six studies that could be included in our meta-analysis”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Of the 28 unique studies, 19 used shotgun metagenomics, and 13 made raw data available.

    “Of these, 19 studies used shotgun metagenomics, and 13 studies made these raw data available.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Among the top 22 strains, some have positive and some negative associations with response.

    “Top 22 strains coloured (blue and red for positive and negative associations with response, respectively), with importance threshold depicted (red dashed line).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The top 22 strains harbored no virulence factors, suggesting that even the negative strains are not prototypical 'pathogens'.

    “We first evaluated them for virulence factors and found that they harbored none, suggesting that even the negative strains are not prototypical 'pathogens'.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The 22 top strains identified by SHAP are biologically meaningful and not artifacts of the model.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    A total of 28 unique studies met eligibility criterion 1.

    “Titles and abstracts were manually reviewed to identify a total of 28 unique studies meeting eligibility criterion 1.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    There was a marked difference in the performance of the CA209-538 strain-RvsP signature between CICB and anti-PD-1 cohorts, with overall modest external generalizability to CICB cohorts (mean AUC = 0.65) but no generalizability to anti-PD-1 cohorts (mean AUC = 0.51).

    “Notably, there was a marked difference in the performance of the CA209-538 strain-RvsP signature between these groups, with overall modest external generalizability to CICB cohorts (mean AUC = 0.65; Fig. 4b) but no generalizability to anti-PD-1 cohorts (mean AUC = 0.51; Fig. 4c).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Despite a wide range in the number of quality-controlled paired-end reads per sample, in general, all were deeply sequenced.

    “Despite a wide range in the number of quality-controlled paired-end reads per sample, in general, all were deeply sequenced (Table 2).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The PubMed search yielded 1,181 records up to December 31, 2022.

    “In total, this search yielded 1,181 records up to December 31, 2022.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Strain-resolution gut microbial signatures outperform clinical predictors and cross-validate across tumor histology types.

    “Fig. 2 | Fig. 2 | Strain-resolution gut microbial signatures outperform clinical predictors and cross-validate across tumor histology types.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    There are clinical and gut microbiome compositional differences between responders and nonresponders.

    “Fig. 1 | Fig. 1 | Clinical and gut microbiome compositional differences between responders and nonresponders.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Responses to CICB are unpredictable, seen in only 20-60% of patients across cancer indications.

    “However, this success is tempered by the unpredictable nature of responses (seen in only 20-60% of patients across these cancer indications 7 )”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [7] Morad, G., Helmink, B. A., Sharma, P. & Wargo, J. A. Hallmarks of response, resistance, and toxicity to immune checkpoint blockade. Cell 184, 5309–5337 (2021). source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data

    The Bowtie 2 alignment rates to the tailored strain reference library were high (median 88.4%), with a median of 10.2 million mapped paired-end reads (50%) passing stringent quality control.

    “The Bowtie 2 alignment rates to our tailored strain reference library were high (median 88.4%), with a median of 10.2 million mapped paired-end reads (50%) passing stringent quality control and used for precise strain quantification (Supplementary Fig. 1 and Methods).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § A gut microbial signature for combination immune checkpoint blockade across cancer types

    Albumin and NLR are responses to inflammation shown to be strongly prognostic across cancer types and treatment settings.

    “both responses to inflammation shown to be strongly prognostic across cancer types and treatment settings 27,28 .”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [27] Gupta, D. & Lis, C. G. Pretreatment serum albumin as a predictor of cancer survival: a systematic review of the epidemiological literature. Nutr. J. 9, 69 (2010). source

      Not checked in the free run.

    • cites [28] Cupp, M. A. et al. Neutrophil to lymphocyte ratio and cancer prognosis: an umbrella review of systematic reviews and meta-analyses of observational studies. BMC Med. 18, 360 (2020). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A. muciniphila B is a distinct species from A. muciniphila.

    “It should be noted that A. muciniphila B is a distinct species from A. muciniphila; although the latter was positively implicated in anti-PD-1 efficacy in NSCLC 13 (also positive in our study but not within the top 22 strains), recent analyses have revealed that it is phylogenetically and phenotypically distinct from A. muciniphila B (known as Akkermansia SGB9228 by MetaPhlAn4 taxonomy) 21 .”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [13] Derosa, L. et al. Intestinal Akkermansia muciniphila predicts clinical response to PD-1 blockade in patients with advanced non-small-cell lung cancer. Nat. Med. 28, 315–324 (2022). source

      Not checked in the free run.

    • cites [21] Karcher, N. et al. Genomic diversity and ecology of human-associated Akkermansia species in the gut microbiome revealed by extensive metagenomic assembly. Genome Biol. 22, 209 (2021). source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Currently approved tumor-agnostic biomarkers for PD-1 blockade include tumor mutational burden and mismatch repair deficiency.

    “Currently approved tumor-agnostic biomarkers for PD-1 blockade include tumor mutational burden and mismatch repair deficiency 9”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [9] Wang, Y. et al. FDA-approved and emerging next generation predictive biomarkers for immune checkpoint inhibitors in cancer patients. Front. Oncol. 11, 683419 (2021). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Strain-level gut microbial predictors of CICB response may be relatively robust across diverse cancer types and are superior to ML predictors built using routine clinically available data.

    “Our ML analysis of our discovery cohort demonstrates that strain-level gut microbial predictors of CICB response may be relatively robust across diverse cancer types and are superior to ML predictors built using routine clinically available data.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Severe immune-related adverse effects are more frequent with CICB when compared to anti-PD-1 or anti-PD-L1 monotherapy.

    “and the more frequent severe immune-related adverse effects experienced with CICB when compared to anti-PD-1 or anti-PD-L1 monotherapy 8 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [8] de Ávila Machado, M. A. et al. Real-world analyses of therapy discontinuation of checkpoint inhibitors in metastatic melanoma patients. Sci. Rep. 10, 14607 (2020). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Predictions trained on binary RvsP appear to capture the RECIST BOR biologically and may have utility for predicting the durability of SD.

    “Furthermore, predictions trained on binary RvsP appear to capture the RECIST BOR biologically and may have utility for predicting the durability of SD.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Studies of clinical cohorts have found associations between baseline Akkermansia muciniphila (lung cancer) and Faecalibacterium prausnitzii (melanoma) fecal abundances and tumor responses among anti-PD-1 recipients.

    “with studies of clinical cohorts finding associations between baseline Akkermansia muciniphila (lung cancer) [10][11][12][13] and Faecalibacterium prausnitzii (melanoma) [14][15][16] fecal abundances and tumor responses among anti-PD-1 recipients.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [10][11][12][13]

      Not checked in the free run.

    • cites [14][15][16]

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation
    § A gut microbial signature for combination immune checkpoint blockade across cancer types

    Combination ICB targeting both PD-1 and CTLA-4 has demonstrated synergistic antitumor activity preclinically.

    “Combination ICB (CICB) targeting both PD-1 and CTLA-4 has demonstrated synergistic antitumor activity preclinically 1”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [1] Curran, M. A., Montalvo, W., Yagita, H. & Allison, J. P. PD-1 and CTLA-4 combination blockade expands infiltrating T cells and reduces regulatory T and myeloid cells within B16 melanoma tumors. Proc. Natl Acad. Sci. USA 107, 4275–4280 (2010). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    The supervised ML pipeline's training and testing procedures were repeated 100 times, and predictions were averaged to account for stochasticity.

    “Whenever evaluating model performance, training and testing procedures were repeated 100 times, and the resultant predictions were averaged to account for the stochasticity of our RF pipeline.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    One strain was identified as a potential contaminant in over 10% of samples from the discovery cohort and was removed.

    “Based on the 'frequency' method (inverse correlation between the abundance of strains and the DNA concentration of submitted samples), one strain was identified as a potential contaminant in over 10% of samples from our discovery cohort (CA209-538 cohort) and was thus removed (Pseudomonas E sp002874965; Supplementary Fig. 4).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    OMR-200 kits are designed to stabilize DNA and have been shown to enhance DNA quantifications and stability across storage temperatures relative to nonpreservative alternatives.

    “OMR-200 kits are designed to stabilize DNA and have been shown to enhance DNA quantifications and stability across storage temperatures relative to nonpreservative alternatives 64 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [64] Maghini, D. G. et al. Quantifying bias introduced by sample collection in relative and absolute microbiome measurements. Nat. Biotechnol. https://doi.org/10.1038/s41587-023-01754-3 (2023). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    No bacteria were identified in the negative control sample for the discovery cohort.

    “Reassuringly, after the above steps, no bacteria were identified in our negative control sample for the discovery cohort.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Feature importance computation was repeated 1,000 times to robustly account for the RF pipeline's stochasticity.

    “We repeated this procedure 1,000 times to account robustly for the RF pipeline's stochasticity.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    The addition of anti-CTLA-4 has a profound effect on gut barrier permeability, potentially changing the influence of the gut microbiome on ICB response.

    “Furthermore, the addition of anti-CTLA-4 has a profound effect on gut barrier permeability 50,51 , potentially changing the influence of the gut microbiome on ICB response.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [50] Coutzac, C. et al. Systemic short chain fatty acids limit antitumor effect of CTLA-4 blockade in hosts with cancer. Nat. Commun. 11, 2168 (2020). source

      Not checked in the free run.

    • cites [51] Mager, L. F. et al. Microbiome-derived inosine modulates response to checkpoint inhibitor immunotherapy. Science 369, 1481–1489 (2020). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    Patient-specific culturing is necessary to perform in vitro and in vivo testing of microbial strains or consortia to derive precise mechanistic insights into their associations with response or nonresponse to ICB and to confirm the direction of causality.

    “Finally, such patient-specific culturing is necessary to perform in vitro and in vivo testing of microbial strains or consortia to derive precise mechanistic insights into their associations with response or nonresponse to ICB and to confirm the direction of causality.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    There are more discrepancies in the pretreatment tumor immunotranscriptomic landscape of anti-PD-1 and CICB nonresponders compared to responders.

    “This is also conceptually consistent with the observation of more discrepancies in the pretreatment tumor immunotranscriptomic landscape of anti-PD-1 and CICB nonresponders compared to responders 49 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [49] Gide, T. N. et al. Distinct immune cell populations define response to anti-PD-1 monotherapy and anti-PD-1/anti-CTLA-4 combined therapy. Cancer Cell 35, 238–255 (2019). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationown data

    Strain-aware approaches are needed to develop context-specific functional hypotheses.

    “highlighting the need for strain-aware approaches to develop context-specific functional hypotheses.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The acetyl-CoA butyrate pathway completeness indicates butyrate production potential.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The MAGs generated in this study potentially harbor errors such as fragmentation, assembly breaks and contamination.

    “Moreover, although we used a state-of-the-art bioinformatics pipeline to generate and quality control MAGs to represent study-specific strains (many of which are new or uncultivated), they still potentially harbor errors (such as fragmentation, assembly breaks and contamination) 53 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [53] Chen, L.-X., Anantharaman, K., Shaiber, A., Eren, A. M. & Banfield, J. F. Accurate and complete genomes from metagenomes. Genome Res. 30, 315–333 (2020). source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Pseudoprogression occurs in a minority of ICB recipients.

    “Given the accumulating evidence of 'pseudoprogression' in a minority of ICB recipients 62”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [62] Park, H. J. et al. Incidence of pseudoprogression during immune checkpoint inhibitor therapy for solid tumors: a systematic review and meta-analysis. Radiology 297, 87–96 (2020). source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    CLR abundances may better account for the inherent compositionality of microbial abundance data.

    “As CLR abundances may better account for the inherent compositionality of microbial abundance data 89 , CLR-transformed feature abundances were exclusively used for the supervised ML analyses.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [89] Gloor, G. B., Macklaim, J. M., Pawlowsky-Glahn, V. & Egozcue, J. J. Microbiome datasets are compositional: and this is not optional. Front. Microbiol. 8, 2224 (2017). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Anti-CTLA-4 appears to modify gut barrier integrity and could feasibly change microbial compositions.

    “for example, anti-CTLA-4 appears to modify gut barrier integrity 51 and thus could feasibly change microbial compositions”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [51] Mager, L. F. et al. Microbiome-derived inosine modulates response to checkpoint inhibitor immunotherapy. Science 369, 1481–1489 (2020). source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    CICB has a distinct mechanism of action compared to anti-PD-1 monotherapy.

    “This is biologically plausible, given that we know that CICB has a distinct mechanism of action compared to anti-PD-1 monotherapy 48 and distinct baseline tumor immune microenvironment signatures 49 .”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [48] Wei, S. C. et al. Combination anti-CTLA-4 plus anti-PD-1 checkpoint blockade utilizes cellular mechanisms partially distinct from monotherapies. Proc. Natl Acad. Sci. USA 116, 22699–22709 (2019). source

      Not checked in the free run.

    • cites [49] Gide, T. N. et al. Distinct immune cell populations define response to anti-PD-1 monotherapy and anti-PD-1/anti-CTLA-4 combined therapy. Cancer Cell 35, 238–255 (2019). source

      Not checked in the free run.

44 of 44

  • The paper relies on this without support.

    The baseline fecal microbiome composition is a relevant determinant of ICB response.

    “A promising 'tumor-extrinsic' avenue for predicting ICB response and/or toxicity a priori is assessing a patient's baseline gut microbiome composition”
    “This suggests that future development of gut microbiome diagnostics or therapeutics should be tailored according to ICB treatment regimen rather than according to cancer type.”

  • The paper relies on this without support.

    The 22 top strains identified by SHAP are biologically meaningful and not artifacts of the model.

    “Twenty-two strains were within half as impactful as the most important strain (a strain of Faecalibacterium sp900539885, an uncultured species), which we opted to focus on subsequently.”
    “We next aimed to interrogate the genomes of the top 22 strains to understand functional potentials that may underpin their strong (positive and negative) response associations.”

  • The paper relies on this without support.

    The acetyl-CoA butyrate pathway completeness indicates butyrate production potential.

    “Indeed, the acetyl-CoA butyrate pathway (which dominates among Firmicutes bacteria) was complete in all (five of five) positive but no negative (none of two) top 22 Ruminococacceae (Fig. 3b).”
    “One metabolite of particular interest was butyrate, given that it has been implicated in anticancer cytotoxic T cell activation preclinically [39][40][41] , and fecal butyrate has been positively associated with ICB efficacy in clinical cohorts 42,43 .”
3 of 3

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Deep shotgun metagenomics in 2 cohorts of rare cancers confirmed cross-cancer validity of strain-response signatures dependent on ICI treatment rather than cancer type.

    “Another study performed deep shotgun metagenomics sequencing in 2 cohorts of patients with diverse rare cancers treated with combination ICI.16 Comforted by a meta-analysis of gut metagenomes from 6 additional comparable studies, they confirmed the cross-cancer (and cross-country) validity of strain – response signatures which were dependent on the ICI treatment rather than on the cancer type.16”

    From <i>Faecalibacterium prausnitzii</i> strain EXL01 boosts efficacy of immune checkpoint inhibitors (2024) · cites this paper as [16]

  • No single microbial species can reliably predict treatment response across all contexts, suggesting a complex interplay among microbiome-immunotherapy interactions.

    “Indeed, multiple studies have highlighted that no single microbial species can reliably predict treatment response across all contexts, suggesting a complex interplay among microbiome–immunotherapy interactions (Gunjur et al., 2024; Lee et al., 2022).”

    From Microbiota-centered interventions to boost immune checkpoint blockade therapies (2025) · cites this paper as Gunjur et al., 2024

  • Challenges remain regarding the reproducibility and generalizability of microbiome-derived signatures across patient cohorts and immunotherapy regimens.

    “Although ML algorithms have improved the predictive accuracy of treatment outcomes, challenges remain regarding the reproducibility and generalizability of microbiome-derived signatures across patient cohorts and immunotherapy regimens (Gunjur et al., 2024; Lee et al., 2022).”

    From Microbiota-centered interventions to boost immune checkpoint blockade therapies (2025) · cites this paper as Gunjur et al., 2024

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Extended Data Figure 1a is assigned two different survival outcomes

    Minor

    The figure caption calls panel 1a a progression-free-survival curve, while the results assign overall survival to panel 1a and progression-free survival to panel 1b.

  • NovaSeq read length differs between methods and Table 2

    Minor

    The sequencing methods describe 150-base paired-end reads, whereas Table 2 lists the CA209-538 NovaSeq reads as 151 bases per read.

  • Figure 3 undercounts archaeal genomes

    Minor

    The methods identify six known archaeal species clusters plus two additional archaeal genomes, implying at least eight archaeal genomes in the reference library. Figure 3 says it excludes only six archaea from that library.

3 of 3

Cites

Flagged first, then by citations

Research works printed but not linked: 3

  1. Pedregosa, F. et al. Scikit-learn: machine learning in Python. J. Mach. Learn. Res. 12, 2825–2830 (2011).
  2. Lemaître, G., Nogueira, F. & Aridas, C. K. Imbalanced-learn: a Python toolbox to tackle the curse of imbalanced datasets in machine learning. J. Mach. Learn. Res. 18, 559–563 (2017).
  3. Therneau, T. M. & Grambsch, P. M. Modeling Survival Data: Extending the Cox Model (Springer Science & Business Media, 2013).

Cited by

Sorted by citations

Authors
Gunjur, A., Shao, Y., Rozday, T., Klein, O., Mu, A., Haak, B. W., Markman, B., Kee, D., Carlino, M. S., Underhill, C., Frentzas, S., Michael, M., Gao, B., Palmer, J., Cebon, J., Behren, A., Adams, D. J., & Lawley, T. D.
Affiliations
  • Wellcome Sanger Institute (GB)
  • La Trobe University (AU)
  • European Bioinformatics Institute (GB)
  • The Alfred Hospital (AU)
  • Walter and Eliza Hall Institute of Medical Research (AU)
  • The University of Sydney (AU)
  • UNSW Sydney (AU)
  • Monash Health (AU)
  • The University of Melbourne (AU)
  • Westmead Hospital (AU)
Corresponding author
  • Gunjur, Ashray
  • Lawley, Trevor D.
ORCID
Venue
Nature Medicine
Publisher
Springer Science and Business Media LLC
Published
2024-03
Volume
30
Issue
3
Pages
797-809
ISSN
  • 1078-8956 (ISSN-L)
  • 1546-170X
DOI
10.1038/s41591-024-02823-z
PMC
PMC10957475
OpenAlex
W4392356761
Cited by
165
References
97
Citations by year
2026: 75 · 2025: 70 · 2024: 19 · 2022: 1
FWCI
22.26
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Medicine (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Pure Amsterdam UMC (submittedVersion, cc-by)
  • Figshare (submittedVersion, cc-by)
  • Minerva Access (University of Melbourne) (submittedVersion, cc-by)
  • La Trobe University (cc-by)
Funders
  • Cancer Research Institute
  • Bristol-Myers Squibb
  • Wellcome Trust (220540/Z/20/A, Z/17/Z, 206194/Z17/Z, 220540, 206194)
  • Australian Government
  • Cancer Research UK (A25117, 21717, PHDSTU-Hist\100121, [C9685/A25117], C20510/A21717, 25117)
  • Peter MacCallum Cancer Centre
  • Hudson Institute of Medical Research
  • Tour de Cure
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gut microbiota and health
  • Health Sciences › Medicine › Oncology › Cancer Immunotherapy and Biomarkers
  • Life Sciences › Agricultural and Biological Sciences › Food Science › Probiotics and Fermented Foods
Keywords
gut microbiota, immune checkpoint blockade, PD-1, CTLA-4, strain-resolved metagenomics, microbiome biomarkers, treatment response prediction, rare tumor, progression-free survival, anti-PD-1 monotherapy
MeSH terms
Gastrointestinal Microbiome, Immune Checkpoint Inhibitors, Humans, Neoplasms

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-05.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-05.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review