Tumor monocyte content predicts immunochemotherapy outcomes in esophageal adenocarcinoma

Abstract

For inoperable esophageal adenocarcinoma (EAC), identifying patients likely to benefit from recently approved immunochemotherapy (ICI+CTX) treatments remains a key challenge. We address this using a uniquely designed window-of-opportunity trial (LUD2015-005), in which 35 inoperable EAC patients received first-line immune checkpoint inhibitors for four weeks (ICI-4W), followed by ICI+CTX. Comprehensive biomarker profiling, including generation of a 65,000-cell single-cell RNA-sequencing atlas of esophageal cancer, as well as multi-timepoint transcriptomic profiling of EAC during ICI-4W, reveals a novel T cell inflammation signature (INCITE) whose upregulation correlates with ICI-induced tumor shrinkage. Deconvolution of pre-treatment gastro-esophageal cancer transcriptomes using our single-cell atlas identifies high tumor monocyte content (TMC) as an unexpected ICI+CTX-specific predictor of greater overall survival (OS) in LUD2015-005 patients and of ICI response in prevalent gastric cancer subtypes from independent cohorts. Tumor mutational burden is an additional independent and additive predictor of LUD2015-005 OS. TMC can improve patient selection for emerging ICI+CTX therapies in gastro-esophageal cancer.

Keywords: cell type deconvolution; esophageal cancer; immune checkpoint inhibitors; immune priming; immunochemotherapy; molecular profiling; predictive biomarkers; single-cell RNA-sequencing atlas; tumor associated monocytes; tumor mutational burden.

Graphical abstract

Figure 1

LUD2015-005 design and clinical outcomes (see also Figure S1) (A) LUD2015-005 timeline of treatment, sampling, and CT response-assessment events. On-treatment CT scans are named according to their timing relative to ICI+CTX (C1D8 = cycle 1, day 8). (B and C) Kaplan-Meier curves for (B) OS and (C) PFS for all LUD2015-005 inoperable patients. Risk table (below) shows number of patients with ongoing survival at each timepoint. (D) Maximal target lesion shrinkage (sum of diameters; not including new/non-measurable lesions) attained at any CT scan during the study. Bars are colored by irRECIST BOR. Three patients (1 irCR, 2 SD) not shown due to absence of measurable target lesions. Three patients who passed away from clear clinical progression prior to any on-treatment CT (clinical PD) also not shown. ^∗Patients with irSD of target lesions but unequivocal progression due to new lesions (overall irRECIST response of irPD). (E) Spider plot showing CT-assessed change in target lesion size from the pre-treatment scan for each patient throughout the study. Certain patients highlight the difficulty in summarizing outcomes using response: ↓ = lesions grew during ICI-4W, but eventually attained irPR during immunochemotherapy; ^∗ = irCR but average PFS (progressed at 9.3 months); ˆ = PFS>12 months but no irRECIST response (two additional cases had non-assessable target lesion sizes).

Figure 2

Molecular features of response and resistance to ICI (see also Figures S2 and S3) (A) Top: Endoscopic sampling sites for biomarker analyses. Esophageal and gastric biopsies were taken ≥2cm away from tumor and gastro-esophageal junction. Bottom: Summary of bulk RNA-sequencing dataset. (B) Differential expression between PreTx and ICI-4W for patients with biopsies at both timepoints (n = 28), controlling for patient-specific effects (see STAR Methods). Moderated log fold change and FDR are shown; the top 70 significant genes are labeled. All significant DEGs (FDR<0.1) are in red. Inset: Mean variance stabilization transformed (VST) expression of INCITE genes (top 70 DEGs), Z score normalized and displayed with CB and NCB facets. Lines connect values for the same patient across timepoints. (C) Mean VST expression change of INCITE genes (scaled, without centering) compared to percentage change of tumor size (target lesions) at the C1D8 scan from PreTx. Points are colored by the timing of first report of irRECIST response, and the MSI tumor is labeled. Pearson correlation statistics are displayed. (D–F) Fast gene set enrichment analysis (FGSEA) results showing the most significantly enriched pathways. Bar length represents normalized enrichment score (NES); color reflects the adjusted p value (labeled). For (D), the FGSEA test statistic was the correlation coefficient between changes in VST-normalized gene expression (ICI-4W–PreTx) and changes in tumor size during ICI-4W. For (E) (PreTx) and (F) (ICI-4W), the test statistic was moderated log fold change calculated by DESeq2 at each timepoint using scaled ICI-4W tumor size changes (continuous variable) in the design formula.

Figure 3

Myeloid phenotype in EAC predicts CB on ICI+CTX (see also Figures S4 and S5) (A) FGSEA results showing significantly enriched PreTx pathways in CB and NCB (FDR<0.1). Bar length represents NES; color reflects adjusted p value (labeled). Test statistic is DESeq2 moderated log fold changes (CB vs. NCB). (B) Heatmap of Z-score-normalized logTPM expression for markers of general immune infiltration (PTPRC/CD45), T/NK cells, B cells, dendritic cells (DC), other myeloid cells, and a panel of myeloid-targeted cytokines and chemokines, across different PreTx tissues. Genes significantly (FDR<0.1) up or down in EAC compared to other tissues are labeled. (C) Preprocessing summary for the LUD2015-005 atlas showing cell barcodes remaining after each filtration step (see STAR Methods). Total cell barcodes are the true cell barcodes called by Cell Ranger (filtered feature-barcode matrix). (D) Sankey plot illustrating the contribution of each patient-tissue type combination to the four broad cellular compartments and their constituent cell types. (E) Uniform manifold approximation and projection (UMAP) dimensionality reduction of all cells surviving quality control (QC) and filtering in the LUD2015-005 atlas, colored by cellular compartments. Batch effects due to dissociation method were first removed using FastMNN. (F) Dot plot of significant PreTx CB-associated DEGs in innate immune-related gene sets (from Figure 3A). For each gene, dot color represents average expression in each cell type (scaled and log-normalized), while size reflects the percentage of cells with detectable expression in each cell type.

Figure 4

Increased TMC is an ICI+CTX-specific predictor of improved outcomes (see also Figures S7 and S8) (A) Deconvolution-assessed levels of phagocytic immune cell types in LUD2015-005 PreTx EAC biopsies were log10-transformed and scaled by column prior to hierarchical clustering (ward.D linkage). Results are shown as a clustered heatmap, with cells colored according to the scaled deconvolution estimates, and row annotation according to clinical outcomes. CB rates are shown for the monocyte-high cluster and other samples as indicated by the vertical lines. p value (for difference between these rates) by two-proportions z-test. (B) Kaplan-Meier plots showing OS of TMC-high and -low groups from LUD2015-005 EAC patients, as split by the cohort median. p value by log rank test. (C) TMC Kaplan-Meier plots as in (B), but for TCGA (left) and ICGC (right) stage III and IV EACs, used as reference cohorts for non-ICI management. (D) Kaplan-Meier plots comparing TMC-high and TMC-low patients for αPD-L1 (left) and αPD-L1+αCTLA-4 (right) LUD2015-005 treatment subgroups, both split by subgroup median TMC. p values by log rank test. (E) TMC assessment in an independent pooled cohort of ICI-treated GC.^, Deconvolution-assessed TMC from LUD2015–005 and GC cohorts are both shown in EBV-/MSS and EBV+/MSI facets, grouped by response. LUD2015-005 response criteria (irRECIST) differed from the GC cohort (RECIST/RECIST v1.1); however binary response calls were the same for irRECIST and RECIST v1.1 criteria in all LUD2015-005 patients shown. p values by Mann-Whitney U-test.

Figure 5

Pre-treatment TMB and TMC are complementary predictive biomarkers for ICI+CTX (see also Figure S8) (A) OncoPrint showing genomic alterations of cancer driver genes in PreTx EAC biopsies. Top barplot indicates the fraction of mutations assigned to predefined single base substitution (SBS) mutational signatures. TMB (coding mutations/Mb) and total SNV and indel numbers are shown. Eight patients had WGS available from multiple PreTx biopsies; for these, TMB, SNVs, and indels represent the average across biopsies, while OncoPrint and SBS signatures were calculated using the union of calls. (B) PreTx TMB for each EAC patient grouped by CB status. p value between groups was calculated by Mann-Whitney U-test. (C and D) Kaplan-Meier plots of TMB-high and TMB-low groups, defined using the cohort median, for (C) LUD2015-005 and (D) non-ICI-treated reference cohorts (as in Figure 4C), split into TCGA (top) and ICGC (bottom). For TCGA and ICGC, only stage III and IV EAC tumors were assessed. p values by log rank test. (E) Kaplan-Meier plots of four subgroups defined by splitting PreTx TMC and TMB values by their respective cohort medians. p value by log rank test (overall difference between the four groups). (F) Forest plot of multivariable Cox regression for OS (top) and PFS (bottom) with PreTx TMC (log10-transformed) and TMB. Both values scaled and centered before regression. Hazard ratio (HR) and 95% CI are shown (HR < 1: association with longer survival; >1: with reduced survival).

Figure 6

Exploring correlates and potential mechanisms of TMC (see also Table S5) (A) Comparison of INCITE upregulation during ICI-4W and PreTx TMC (log10-transformed). Pearson and Spearman correlation statistics are displayed. (B) Multiple linear regression between PreTx TMC (log10-transformed) and INCITE upregulation with tumor size changes during ICI-4W as dependent variable. Coefficients with p < 0.05 are bolded, representing significant associations with ICI-4W tumor size change. (C) TMC values (log10-transformed) across timepoints, grouped into CB and NCB facets. Each line connects values for the same patient. Crossbars represent median value for that timepoint. (D) Trajectory analysis of non-mast cell phagocytes using Monocle2 (DDRTree). Numbers represent trajectory branch points. Approximate locations of cell types are labeled. (E) Trajectory coordinates from (D) plotted separately for PreTx TMC-high (left, n = 5) and TMC-low tumors (right, n = 3), classified using overall cohort PreTx median. Within each group, the two-dimensional kernel density of cells along the trajectory is shown at PreTx (blue) and ICI-4W (red). Regions enveloped by more contours have more cells present at the specified timepoint. (F) Top: Fraction of immune-phagocyte compartment composed of M1-like macrophages, cDC1, cDC2, and LAMP3-high DCs (phenotypes enriched at ICI-4W in TMC-high tumors in E), assessed by deconvolution. Bottom: as above, but for the fraction of TAMs and M2-like macrophages in immune-phagocyte compartment. p values by Mann-Whitney U-test.

Figure 7

scRNA-seq reveals EAC- and T cell-specific expression patterns predictive of ICI+CTX outcomes (A) Top: EAC-specific PreTx DEGs from pseudobulk differential expression (see STAR Methods) with FDR < 0.1. Dot size represents percentage of cells with any expression; color represents average expression (scaled log-normalized counts). EAC-GDBD was excluded due to insufficient EAC cells. DEGs are sorted by FDR and sign of change (CB leftwards, NCB rightwards). Bottom: as above for T cell-specific DEGs (including EAC-GDBD, having sufficient T cells). Black line: Interferon-stimulated genes (ISGs). (B and C) Violin plot of PreTx expression (log-normalized) of (B) IGFBP2 and (C) OAS1 in single cells from each patient. Plots are faceted by cellular compartments.