CAR-T cell therapy-related cytokine release syndrome and therapeutic response is modulated by the gut microbiome in hematologic malignancies

Abstract

Immunotherapy utilizing chimeric antigen receptor T cell (CAR-T) therapy holds promise for hematologic malignancies, however, response rates and associated immune-related adverse effects widely vary among patients. Here we show, by comparing diversity and composition of the gut microbiome during different CAR-T therapeutic phases in the clinical trial ChiCTR1800017404, that the gut flora characteristically differs among patients and according to treatment stages, and might also reflect patient response to therapy in relapsed/refractory multiple myeloma (MM; n = 43), acute lympholastic leukemia (ALL; n = 23) and non-Hodgkin lymphoma (NHL; n = 12). We observe significant temporal differences in diversity and abundance of Bifidobacterium, Prevotella, Sutterella, and Collinsella between MM patients in complete remission (n = 24) and those in partial remission (n = 11). Furthermore, we find that patients with severe cytokine release syndrome present with higher abundance of Bifidobacterium, Leuconostoc, Stenotrophomonas, and Staphylococcus, which is reproducible in an independent cohort of 38 MM patients. This study has important implications for understanding the biological role of the microbiome in CAR-T treatment responsiveness of hematologic malignancy patients, and may guide therapeutic intervention to increase efficacy. The success rate of CAR-T cell therapy is high in blood cancers, yet individual patient characteristics might reduce therapeutic benefit. Here we show that therapeutic response in MM, ALL and NHL, and occurrence of severe cytokine release syndrome in multiple myeloma are associated with specific gut microbiome alterations.

Fig. 1. Trial profile and clinical response in r/r MM patients treated with CAR-T cell infusion.

a Patient enrollment. b Anti–BCMA single-chain variable fragment (scFv), a hinge and transmembrane regions, and 4-1BB costimulatory moiety, and CD3ζ T-cell activation domain. c Blood and fecal sample collection. d Clinical response; CRS grade distribution in 43 r/r MM patients. e Numbers of BCMA CAR-T cell percentages in PB assessed by FACS in different therapy stages after CAR-T cell infusion and serum concentrations of IL-10 and IFN-γ in different therapy stages among the CR (n = 24 biologically independent patients), VGPR (n = 6 biologically independent patients), and PR (n = 11 biologically independent patients) groups. Blue, green, and red colors indicate CR, VGPR, and PR group, respectively. Data are presented as mean values ± SEM. Significance determined by two-sided Kruskal-Wallis test and adjustments were made for multiple comparison. P values for CAR-T percent in PB, serum IL-10 and IFN-γ between CR and PR groups in CRSb stage were 0.004, 0.048, 0.085, respectively. *p < 0.05, **p < 0.01. f Body temperature and serum concentrations of IL-6 and IFN-γ in different therapy stages among CRS grade groups. (Grade 1 CRS group: n = 8 biologically independent patients, Grade 2 CRS group: n = 16 biologically independent patients, and Grade 3 CRS group: n = 19 biologically independent patients). Data are presented as mean values ± SEM. Significance determined by two-sided Kruskal-Wallis test and adjustments were made for multiple comparison. P values for serum IL-6 and IFN-γ between Grade 1 CRS and Grade 3 CRS were 0.002 and 0.006, respectively. *p < 0.05, **p < 0.01. g Representative MM patients with impressive antimyeloma response. Positron emission tomography-computed tomography scans before and 5 months after CAR-T cell treatment showing complete elimination of large number of MM bone metastases. Before receiving CAR-T cell infusion, 43.5% of bone marrow cells of the patient were plasma cells, but after 1.5 months of infusion, dramatic eradication of MM from the bone marrow was observed; and MM cells became undetectable by flow cytometry. The bar indicates a length of 5 μm.

Fig. 2. Changes of microbial composition during CAR-T therapy in MM patients.

a Shannon diversity indices of gut microbiome across CAR-T stages in all myeloma patients. Differential tests by Friedman’s tests and two-tailed Wilcoxon rank-sum tests for 10 pairwise comparisons of the five timepoints (n = 14). Bonferroni correction was applied for multiple testing; *FDR < 0.05, **FDR < 0.01. For FCa versus CRSc, adjusted p = 0.023; FCb versus CRSc, adjusted p = 0.009; CRSa versus CRSc, adjusted p = 0.017. Boxplots indicate the median (thick bar), first and third quartiles (lower and upper bounds of the box, respectively), lowest and highest data value within 1.5 times the interquartile range (lower and upper bounds of the whisker). b Pairwise Spearman correlation of OTU-level bacterial abundance across different timepoints. Rho value for each significant correlation is labeled inside box. c Stacked bar plot of mean phylum-level phylogenetic composition of bacterial taxa in myeloma patients across different therapy stages. d Significant features identified by longitudinal analysis in Qiime2 “feature-volatility” plugin to identify taxonomic features associated with therapy stages. Scatter plot shows importance and average change of each important features by the longitudinal analysis. Genus-level features are labeled in the figure. Genus identified by both longitudinal analysis in Qiime2 and maSigPro are bolded and underlined. e Bar plot in the left shows significantly changed genera across the therapy identified by Friedman’s tests (FDR < 0.05, n = 14). Effect size was estimated by Kendall’s W Test. Heatmap in the right side denotes difference of each genus between two therapy stages. Red represents significant enrichment while blue represents significant depletion of the genus in the posterior stage comprising with the anterior stage. Significant p values were labeled in the boxes. Significances by two-tailed Wilcoxon rank-sum tests with FDR correction.

Fig. 3. Association of compositional differences in gut microbiome with responses to CAR-T therapy in MM patients.

a Shannon diversity indices of gut microbiome differed between CR and PR groups across CAR-T stages. Significances were assessed by two-sided Wilcoxon rank-sum test (n = 35). P values were 0.077, 0.040, 0.036 for FCb, CRSa, and CRSb, respectively. Boxplots indicate the median (thick bar), first and third quartiles (lower and upper bounds of the box, respectively), lowest and highest data value within 1.5 times the interquartile range (lower and upper bounds of the whisker). b Principal coordinate analysis of fecal samples in CRSb stage by response (CR versus PR) using Canberra distance. P value was calculated by PERMANOVA (n = 35). c Summary of number of PR or CR-enriched OTUs in different therapy stages. Difference between CR and PR groups was assessed by two-sided Wilcoxon rank-sum test. P value significant cutoff was 0.05 (n = 35). d Heatmap for abundance of OTUs with significant temporal differences between CR and PR groups identified by maSigPro (FDR < 0.05). Rows denote bacterial OTUs grouped into three sets according to regression coefficients and sorted by mean abundance within each set. Individual fecal samples were organized in columns and grouped by therapy stages. Columns in the blue and red dashed box show abundance and longitudinal changes of these OTUs in CR and PR groups across the five timepoints. Color of the heatmap is proportional to OTU abundance (red indicates higher abundance and blue indicates lower abundance). e Profiles of significant gene clusters correspond to d. Solid lines denote median profile of abundance of OTUs within cluster for each experimental group through time. Fitted curve of each group is displayed as dotted line. f Phylogenetic composition of OTUs within each cluster in d at phylum and order levels.

Fig. 4. Determination of correlated genera with clinical response to CAR-T therapy in MM patients.

a Differentially abundant genera between CR and PR group. Bubble plot in the left represents p values by maSigPro. Bar plots in the middle and right show significances and coefficients by generalized linear-mixed models (GLMMs) before and after CAR-T infusion (n = 35). Blue bars indicate significant enrichment in CR group while red bars indicate significant enrichment in PR group (FDR < 0.05). Red stars marked genera that was identified to be differentially abundant by linear discriminant analysis (p < 0.05 for Kruskal–Wallis H statistic and LDA score >2). P values by linear discriminant analysis for Sutterella, Collinsella, Paraprevotella, Bifidobacterium, Anaerotruncus, Prevotella, and Oscillospira before CAR-T were 0.0017, 0.0014, 0.038, 0.0015, 0.0064, 0.030, and 0.006, respectively; P values by linear discriminant analysis for Sutterella, Collinsella, Paraprevotella, Bifidobacterium, Anaerotruncus, Prevotella, Oscillospira, Faecalibacterium, Gemmiger, Clostridium, Odoribacter, Roseburia, Dialister, Enhydrobacter, Ruminococcus, and Dorea after CAR-T were 0.00012, 0.00076, 0.0060, 0.0.0067, 0.042, 0.0049, 0.011, 0.00017, 0.0035, 0.0058, 0.0073, 0.0013, 0.000038, 0.021, 0.0056, and 0.017, respectively. b Mean bacterial abundance [log2 (percentage + 1)] of CR, VGPR, and PR myeloma patents before and after CAR-T cell infusion (n = 43). Red stars indicate significant difference between CR and PR group by all three methods in panel a. P values for Sutterella by maSigPro were 1.17e-06, by generalized linear-mixed model were 7.86e-12 and 1.51e-14 before and after CAR-T, by linear discriminant analysis were 0.0017 and 0.00012 before and after CAR-T, respectively; P values for Faecalibacterium by maSigPro were 0.0093, by generalized linear-mixed model and linear discriminant analysis were 1.22e-10 and 0.00017 after CAR-T, respectively; P values for Bifidobacterium by maSigPro were 2.19e-06, by generalized linear-mixed model were 5.67e-08 and 1.51e-08 before and after CAR-T, by linear discriminant analysis were 0.0015 and 0.0067 before and after CAR-T, respectively; P values for Ruminococcus by maSigPro were 1.49e-08, by generalized linear-mixed model and linear discriminant analysis were 0.00031 and 0.0056 after CAR-T, respectively. c Relative abundance [log2 (percentage + 1)] of top discriminative signatures at baseline (FCa) timepoint identified by RF feature selection procedure (n = 35). Genera with highest scores of mean decreases in Gini were selected. Importance scores in RF classification model and fold-change levels in log₂ scale are noted below plot for each genus. Blue and red colors indicate CR and PR group, respectively. d Same as panel c for post-chemotherapy (FCb) timepoint (n = 35). Only signatures enriched in CR patents are displayed. Those depleted in CR patents are displayed in Fig. S2C. e Receiver operating characteristic (ROC) curve of RF model using discriminatory genera as predictors for baseline timepoint. f Same as panel e for post-chemotherapy timepoint. g Kaplan–Meier (KM) plot of PFS curves by log-rank test for patients with high (dark blue), median (green), or low (red) abundance of Sutterella. Abundance of genus Sutterella was in terms of median abundance of all timepoints. Boxplots indicate the median (thick bar), first and third quartiles (lower and upper bounds of the box, respectively), lowest and highest data value within 1.5 times the interquartile range (lower and upper bounds of the whisker).

Fig. 5. Compositional differences between subjects with different CRS grades in MM patients.

a Correlation of differentially abundant genera with CRSgrade. Bubble plot in the left shows significant genera between severe and mild CRS groups by maSigPro (n = 27). Bar plots in the middle and right show significances and coefficients by generalized linear-mixed models (GLMMs) before and during CRS. Orange bars indicate positive correlation with CRS. Green bars indicate negative correlation. Red stars marked genera that was identified to be differentially abundant by linear discriminant analysis (p < 0.05 for Kruskal-Wallis H statistic and LDA score >2). P values by linear discriminant analysis for Bifidobacterium and Butyricicoccus before CAR-T were 0.003 and 0.027, respectively; P values by linear discriminant analysis for Leuconostoc, Bifidobacterium, Lactococcus, and Enhydrobacter after CAR-T were 0.016, 0.029, 0.0029, and 0.037, respectively. b Mean bacterial abundance in MM patients with different CRS grades before and during occurrence of CRS (n = 43). Red stars indicate significant difference between Grade 1 CRS and Grade 3 CRS group by all three methods in panel a. P values for Bifidobacterium by maSigPro was 8.9e-08, by generalized linear-mixed model were 9.75e-06 and 1.42e-08 before and after CAR-T, by linear discriminant analysis were 0.003 and 0.029 before and after CAR-T, respectively; P values for Leuconostoc by maSigPro was 1.29e-14, by generalized linear-mixed model and linear discriminant analysis were 3.14e-11 and 0.016 after CAR-T, respectively. Boxplots indicate the median (thick bar), first and third quartiles (lower and upper bounds of the box, respectively), lowest and highest data value within 1.5 times the interquartile range (lower and upper bounds of the whisker). c Network representing correlations between gut microbes (gray nodes), immune cells and inflammatory markers (green nodes) at FDR < 0.05. Correlations were measured by repeated measure correlation analysis (rmcorr). Red edges indicate positive correlations and blue edges negative correlations. Edge width is proportional to correlation coefficient (ρ) calculated by Spearman correlation test. Only genera identified as associated with clinical response and CRS grade were included in correlation analysis. d Top 2 positive and negative correlations in repeated measure correlation analysis. Data are presented as mean ± SEM.