菌群失调、胆汁酸代谢与肿瘤免疫治疗反应

Figure 1. Gut dysbiosis perturbs BA composition (A) Experimental settings at two different centers; D, day. (B) Supervised hierarchical clustering of BAs significantly different between ABX and water-fed tumor bearers assessed by metabolomics. (C and D) Kinetics of the percentages of distinct circulating BAs among the total BA. (E) Id. in the plasma of naive mice at GRCC and orally gavaged with E. clostridioformis or water controls. (F and G) Metabolite levels in tumor beds after 15 days of ABX. Volcano plot comparing ABX vs. drinking water (F), red dots indicate metabolites significantly increased after ABX. Dot plots depicting the percentages of distinct BA (G). Error bars indicate mean ± SEM (G). One representative experiment containing 8 mice/group (B, C, D, F, and G) and 6 and 7 mice/group (E) is depicted. Comparisons between groups were performed using non-parametric Mann-Whitney U test (2 groups), followed by multiple-comparisons test controlling the false discovery rate (FDR) (B, F, and G), or non-parametric Friedman test (>2 paired groups for longitudinal metabolomics analysis), followed by two stage set-up method of Benjamini, Krieger, and Yekutieli (C, D, and E). See also Figures S1–S3 and Table S1.

Figure 2. GDCA rescues gut dysbiosis-associated resistance to immunotherapy (A) Experimental setting at GRCC. (B–F) Effect of E. clostridioformis with or without GDCA on the response to PD-1 blockade (B and C) or PD-1+CTLA-4 blockade (D and E) against MCA205 fi- brosarcomas (B and C) or RET (D and E). Tumor size at day 13–14 (B and D, left) or tumor growth kinetics (B right and E left) in three (C) or two (D) concatenated experiments comprising 25–29 (B), 9–10 (C), and 12–14 or 5 mice/group (E) and one representative experiment (E). (F) MCA205 tumor growth kinetics in mice treated with ABX (2 weeks) and orally gavaged twice with E. clostridioformis and BA (TDCA and DCA) treated with α-PD-1. Two pooled independent experiments of 4–5 animals/group. Comparisons between groups were performed using non-parametric Kruskal-Wallis H test (>2 groups), followed by multiple-comparisons test controlling the false discovery rate (FDR). p values were calculated at each time point comparing each group with E. clostridioformis + corn oil (red curves) with the Mann-Whitney U test (2 groups) (C, E right, and F). The tumor growth kinetics were analyzed using the online tool ‘‘TumGrowth’’ (see STAR Methods) (B right and E left). Mean ± SEM (B, D, and E left) or individual tumor growth kinetics (C, E four right panels, and F) are depicted. (G) Heatmaps representing the log2FC of BA levels between before (at D-1) and after oral BA at day 7 (left) and day 12 (right) in 5 mice/group. p value (*, p < 0.05; **, p < 0.01) on the log2FC between oral BA versus corn oil (Mann-Whitney U test, 2 groups). (H) Volcano plot contrasting plasma metabolites after GDCA versus LCA in dysbiotic mice at day 7 (n = 5 mice/group). Green and red dots are significant metabolites (Mann-Whitney U test, 2 groups). See also Figures S4 and S5 and Tables S1 and S2.

Figure 3. Modulations of tauro-conjugated primary BAs and efficacy of PD-1 blockade (A) Surgical bile duct ligation (BDL) in MCA205 tumor-bearing mice treated with anti-PD-1 Abs at Centre de Recherche des Cordeliers. (B) Livers were stained with hematoxylin and eosin (top) as well as Sirius Red (bottom). One representative photograph of each liver in each group of 8–10 mice is shown. (C and D) Percentages of distinct BAs influenced by BDL. (E) Mean tumor growth curves in both groups. The tumor growth kinetics were analyzed using the online tool ‘‘TumGrowth’’ (see STAR Methods). (F) Experimental setting of ABX + E. clostridioformis-induced gut dysbiosis compensated by fexaramine in MCA205 tumor bearers at GRCC. (G) Tumor growth kinetics were analyzed as in (E). A representative experiment comprising 5 mice/group is shown, out of two yielding similar results. (H) Flow cytometric analysis of TILs in mice receiving ABX and/or fexaramine and treated with α-PD-1 mAb. One representative experiment comprising 10 mice/ group is depicted. One dot represents one tumor. (I and J) Longitudinal plasma metabolomics analyzing various BAs (in percentages of the total BA) over time. Each dot represents one mouse housed at GRCC animal facility. (K) Ex vivo stimulation of CD3+ T splenocytes exposed to various concentrations of BAs (left), followed by flow cytometry analysis of the co-culture for cell viability and proliferation. The absolute number of live CD3+ T cells or percentage of Ki67+ in CD8+ T cells recovered at 24 h of BA exposure is shown (middle and right, respectively). Also refer to Figures S6F and S6G for CD4+ T cells and fold change calculations. Each dot represents one well, and each experiment was performed in triplicate wells and performed between three to five times (n = 9–15 wells), except for T-β-MCA condition that was performed once (n = 3 wells). n = 9–10 mice/group (B–E and H–J). Error bars indicate mean ± SEM (C, D, E, G, H, and K). Comparisons between groups were analyzed using non-parametric Mann-Whitney U test (2 groups) or Kruskal-Wallis H test (>2 groups), followed by multiple-comparisons test controlling the false discovery rate (FDR) (C, D, H, and K), or non-parametric Friedman test (>2 paired groups for longitudinal metabolomics analysis), followed by two stage set-up method of Benjamini, Krieger, and Yekutieli (I and J). The tumor growth kinetics (E and G) were analyzed using the online tool ‘‘TumGrowth’’ (see STAR Methods). See also Figure S6 and Table S1.

Figure 4. GDCA acts at sites distal from the liver to reprogram the TME (A–D) Intratumoral abundance of orally gavaged BAs determined by mass-spectrometry-based metabolomics in E. clostridioformis dysbiotic mice treated with α-PD-1 on day 12 post MCA205 injection, comparing BA vs. water + corn oil controls using Z scores of AreaQCcorr. Shown is the concatenation of 2 experiments including 3–5 mice/group. (legend continued on next page) (E–G) Flow cytometric analysis of LAG-3 expression on various CD4+ T subsets infiltrating the tumor of dysbiotic mice receiving BA and α-PD-1 mAb. One representative experiment comprising 5–10 mice/group is depicted. (H–K) Violin plots showing the expression levels of selected genes encoding the exhaustion signature in all CD8+ TILs in each treatment group on day 11. See also Figure S7. (L and M) In vitro screening of several BAs at 100 μM or cilofexor at 50 μM on Madcam1 relative expression by RT-qPCR in TSECs exposed to TNF-α (vs. negative control [Ctrl-]) (L) or TSECs genetically modified by means of CRISPR-Cas9 gene editing (M). Results from 3–4 independent experiments were pooled, and each dot represents one well; n = 18–42 (L) and 24 (M). (N) In vivo effects of orally gavaged BA on Madcam1 expression in the ileum of mice treated with ABX. Results from 2 independent experiments are pooled using Z score, and each dot represents one mouse; n = 10. (O–Q) Experimental setting (O) and flow cytometric determination of the percentage of CFSE+ T cell subsets reaching the tdLN in day 12 tumor-bearing mice at 24 h after CFSE injection into mLN (P and Q). Results from 2 independent experiments are pooled, and each dot represents one mouse; n = 7–11. (R and S) Differential effects of corn oil (R) versus GDCA (S) in WT vs. Madcam-1−/−MCA205-bearing mice treated with PD-1 blockade. Two pooled experiments comprising n = 9–10 mice/group, showing individual kinetics and mean + SEM tumor growths. p values were calculated at each time point, comparing the two groups. Error bars indicate mean ± SEM (A–N and P–S). Comparisons between groups were performed using non-parametric Kruskal-Wallis H test (>2 groups), followed by multiple-comparisons test controlling the false discovery rate (FDR), except written otherwise (A–G, L–N, P, and Q), or significance was calculated with Wilcoxon rank-sum test and adjusted for multiple testing using Benjamini-Hochberg correction (H–K). The tumor growth kinetics (R left and S left) were analyzed using the online tool ‘‘TumGrowth’’ (see STAR Methods). For R (right) and S (right), p values were calculated at each time point comparing the two groups together using the Mann-Whitney U test. See also Figures S4–S8 and Tables S1 and S3.

Figure 5. Gut dysbiosis alters BA composition in cancer patients (A–D) Percentages of each BA among the total BAs in the plasma, assessed using metabolomics in 133 cancer patients (n = 138 samples) from the ONCOBIOTICS cohort who took ABX (n = 16 + [0–30 days] (A–D); n = 9 +[>30 days] and n = 13 + [>60 days] (B–D)) or did not take ABX (n = 100) (A–D) within different time windows post-ABX cessation as well as for plasma ratio of glyco- over tauro-conjugated BAs (D, bottom). Error bars indicate mean ± SD. Comparisons between groups were performed using non-parametric Mann-Whitney U test (2 groups) in (A) and non-parametric Kruskal-Wallis H test (>2 groups) in (B)–(D). (E) Clustering of Spearman’s correlation coefficients of plasma BAs in patients who took ABX (n = 13 +[0–30 days], left) or did not take ABX (n = 85, right). (F) Spearman’s correlations between distinct plasma BAs (in percentages of total BA) and the paired stool S-score,36 calculated in 61 ONCOBIOTICS patients before and until 15 days after immunotherapy initiation. See also Figures S9–S11 and Tables S4, S5, and S6..

Figure 6. Cancer patients can be stratified based upon plasma BA composition (A) Non-supervised hierarchical clustering of the percentages of plasma BAs in 111 ONCOBIOTICS patients before and until 42 days post ICB initiation. (B–E) Dot plots of the percentages of BAs or conjugated BAs and calprotectin (S100A9) measured in the plasma according to the three clusters illustrated in (A). Also refer to Table S4 and Figures S9 and S10. Each dot represents plasma levels in one patient (n = 111). Error bars indicate mean ± SD. Comparisons between groups were performed using non-parametric Kruskal-Wallis H test (>2 groups), followed by multiple-comparisons test controlling the false discovery rate (FDR). (F) Kaplan-Meier survival curves and univariate Cox regression to assess the prognostic value of each BA cluster (Cox-Mantel test). See also Figures S9–S11 and Tables S4, S5, and S6

Figure 7. Subclinical cholestasis reduces cancer patient survival (A) Patient cohorts amenable to immunotherapy (ICB), +/− chemotherapy (CT), radiotherapy (RT), or tyrosine kinase inhibitors (TKI) and randomized in 2 arms (AtezoTRIBE) for whom γGT and alkaline phosphatase (ALP) levels were assessed in the plasma. (B–G) Univariate analyses for overall survival (OS) (B, F, and G) and progression-free survival (PFS) (C–G), using the Kaplan-Meier estimator and Cox regression with indicated hazard ratios (HRs) and log rank (Mantel Cox) test p values to assess the prognostic value of γGT in each cohort aligned in (A). Also refer to Tables S5 and S6 for univariate and multivariate analyses, respectively. (H) Longitudinal evolution of the mean γGT levels in U/L (scale in log10) for 111 patients who progressed (PFS < 6, orange) or not (PFS ≥6, blue) within 6 months of ICB initiation in the ONCOBIOTICS cohort, as well as for different periods of time. Comparisons between time points (0–1, 1–6, and 6–12 months) were performed using non-parametric Wilcoxon matched-paired test for each PFS group. Comparisons between groups (PFS <6 or ≥6) were performed using non-parametric Kruskal-Wallis H test (>2 groups), followed by multiple-comparisons test controlling the false discovery rate (FDR). (I) Clustering of Spearman’s correlation coefficients between BAs (%), liver parameters, and sMAdCAM-1 (in ng/mL) in the plasma of patients from ONCOBIOTICS, FMT-LUMINATE, MIMIC, and AtezoTRIBE cohorts (p values: *, <0.05; **, <0.01; ***, <0.001). High γGT levels were defined as ≥2ULN (ULN: male patients ≥45U/L; female patients ≥35U/L). See also Figures S9–S11 and Tables S4, S5, and S6.

Figure 8. Subclinical cholestasis is a surrogate marker of gut dysbiosis (A) Absolute number and percentages of SIG1 bacteria among the 37 SIG136 in patient feces according to plasma γGT < or ≥2 ULN (n = 73 and 8 samples, respectively). (B) Plasma γGT levels (in U/L) in patients treated with ABX according to the window of blood sampling before ABX (n = 85 in no ABX and 18 samples in ABX cessation less than 60 days, respectively). (C) Spearman’s correlation between percentages of tauro-conjugated BAs and SIG1 bacteria species among 37 SIG1 at baseline.36 (D) Increase in neutrophil blood counts and plasma soluble ST2 levels in patients with γGT ≥2ULN versus those with γGT < 2ULN (n = 16 and 95, respectively). (E) Spearman’s correlation between γGT (U/L) level and sMAdCAM-1 (ng/mL) in ONCOBIOTICS, FMT-LUMINATE, and MIMIC pooled cohorts and PREMIS. (F) Trajectory of TCDCA (in mean and CI95% Log2AreaQCcorr) in the plasma (n = 32 patients) or in the stools (n = 34 patients) of responder vs. non-responder patients before and during ICB in pooled MIMIC and FMT-LUMINATE cohorts. Comparison between responders and non-responders were performed using a longitudinal slope estimation approach. Left: the significance of temporal changes of plasma TCDCA levels within responders and non-responders was assessed from the estimated time effect in each group, and p values are annotated (related to STAR Methods). Right: Spearman’s correlations between fecal TCDCA levels and days after FMT. (G) Boxplots of raw values of TCDCA (AreaQCcorr) at three time points in TACITO (n = 45). The box shows the interquartile range (Q1–Q3), with the median marked inside. Whiskers extend up to 1.5 × IQR beyond Q1 and Q3. Values outside the whiskers are plotted separately as outliers. (H and I) Idem as in (F) right in the 2 arms of TACITO for TCDCA (H) and TCA (I), depicted in trajectories. Also refer to Figure S11I to visualize the differential evolution of all BAs in the two treatment arms. The significance of temporal changes of plasma TCDCA (H) and TCA (I) levels within placebo and FMT arms after the treatment start was assessed using Spearman’s correlations, and p values and coefficients correlation (Rho) are annotated. Comparisons between groups were performed using non-parametric Mann-Whitney U test (2 groups) (A, B, and D) or Kruskal-Wallis H test (>2 groups) (G), followed by multiple-comparisons test, controlling the false discovery rate (FDR). Error bars represent mean ± SD in (A)–(D). See also Figures S9–S11 and Tables S4, S5, and S6.