Alterations of the composition and spatial organization of the microenvironment following non-dysplastic Barrett's esophagus through progression to cancer

Abstract

Barrett's esophagus (BE), a metaplastic condition that is the only known precursor for esophageal adenocarcinoma (EAC), is relatively common, but progression to cancer is infrequent. BE is inflamed but the contribution of the immune system to the carcinogenic process is unknown. To this end, we contrasted non-dysplastic metaplasia of BE patients, captured when they did not progress (non-progressors), did subsequently, but had not yet progressed (pre-progressors) or had already progressed to EAC (progressors). Using spatial multiplexed 56-protein analysis, serial laser capture microdissection (LCM) RNAseq and shallow whole genome sequencing, we identified prooncogenic immune neighbourhoods and dysregulated immune cell populations predictive of subsequent progression to EAC. Indeed, spatial analysis revealed that M1 macrophages, regulatory natural killer (NK) cells, neutrophils and altered ratios of intraepithelial CD4⁺ and CD8⁺ lymphocytes typify tumor microenvironmental (TME) changes associated with cancer initiation. Spatially derived cell-to-cell interactions revealed progression-specific immune cell interaction signatures predominantly involving M1 macrophages NK cells and plasma cells. Furthermore, LCM RNAseq analysis identified gene expression 'hot' signatures enriched in pre-progression and progression samples. Notably, we also observed a correlation between immune cells and copy number alterations in progressor metaplasia. By exposing coordinated changes in the immune cell landscape in patients at high risk of developing EAC, this multi-omic dataset provides novel diagnostic and therapeutic opportunities.

Figure 1.. The metaplastic Barrett’s immune landscape before and during progression.

(A)The workflow for CODEX analysis. Non-dysplastic Barrett’s esophagus (NDBE) samples are taken from non-progressor (NP), pre-progressor (PP) and progressor (P) patients and subjected to multiplex CODEX staining. Cells are mapped, quantified and their spatial relationship analysed on a per core basis. Created using Biorender. (B) UMAP dimensionality reduction illustrates all immune cell types present in all samples. (C) Dot plot of cell markers and their relative frequency and cell specificity from the CODEX panel. (D) Quantification for all identified immune cells as a percentage of all cells in NP, PP and P, respectively. (E) Shannon index of diversity of all immune cells in NP, PP and P core samples. (F) Representative CODEX images for NK cells (i-iv)(CD11b, red: AnnexinA1, blue. i, iii, v x100 magnification, ii, iv, vi x200 magnification. Green dotted box represents magnified area), neutrophils (vii-xii) (CD11b, red: Arginase1, green. vii, ix, xi x100 magnification, viii, x, xii x200 magnification. Green dotted box represents magnified area) in non-progressors, pre-progressors and progressors respectively. (*P<0.05, **P<0.01, ***P<0.001, ****P<0.0001, Wilcoxon rank sum test).

Figure 2.. IEI cell and LP cell landscape in metaplastic progression.

(A) H&E of a representative core (B) Map processed by Spatstat (cross=IEI cells, dot= LP cells) revealing the epithelial boundaries of (A). (C) 5-plex image of Arginase1, CD68, CD8, CD79a and CD4. (D) High power image of a high power glands revealing a CD8⁺ T IEI cell (red, arrow) and a sub-epithelial macrophage (yellow, arrow). (E) Quantification of IEI cells per 100 epithelial cells across NP, P and PP cores. (F) Quantification of lamina propria cells as a percentage of all segmented lamina propria cells. (G) Fraction of CD4⁺ and CD8⁺ T cells as percentage of IEI cells. (H) Ratio of CD8⁺:CD4⁺ T cell IEI cells. (I) Fraction of CD4⁺ and CD8⁺ LPLs as a percentage of all cells. (J) Ratio of CD8⁺:CD4⁺ T cell LP cells. ((*P<0.05, **P<0.01, ***P<0.001, ****P<0.0001, Wilcoxon rank sum test).

Figure 3.. Cell to cell interactions in metaplastic progression.

(A) Illustration of the Ripley K function analysis of attraction and repulsion (arrows represent distance) on pairwise cell comparisons (Green and red hypothetical cells). Concentrations of red cells are quantified at increasing radii from each green cell. (B) Representative core H&E. (C) Map highlighting center pixel of all immune cells within the representative core, epithelial cells are excluded. (D) An example of cell-to-cell attraction (Plasma cell: CD4 T cell). (E) An example of cell-to-cell repulsion CD4 Treg: Dendritic cell. (F) TOP: map of plasma cell and M1 macrophages in a non-progressor core. BOTTOM: with Ripley L values plotted against complete spatial randomness (CSR) showing a strong attraction. (G) In a pre-progressor showing weak distant attraction. (H) In a progressor showing strong attraction. (I) The number of cores exhibiting attraction, repulsion or neutral (random) distribution in each condition. (J) Illustration of differential mean interaction scores between non- and pre-progressors. The Size of each circle represents the frequency of cell pairs deviating from CSR. Highlighted are M1 macrophage (green) and NK cell (blue) interaction score comparisons. (K) Interaction scores for non- and progressor samples.. (L) Interaction scores for pre- to progressor samples. (M) Summary dot plot of interactions scores at 100 distance units. Size of circles represent mean interaction scores (*P<0.05, Fisher’s exact test).

Figure 4.. Cell neighborhoods in metaplastic progression.

(A) Heatmap of nearest neighbor clustering across all immune cells and all samples. Neighborhoods named after dominant cell type(s). (B-D) All cells, neighborhoods and individual immune cell types as per x, y coordinates. (E-G) Mapped onto the respective H&E image taken after CODEX completion. (H) TOP: Cell composition of CNs (% of total immune cells) in order of non-progressor (NP), pre-progressor (PP) and progressor (P) samples. BOTTOM: Percentage of cells representing each neighborhood in NP, PP and P samples. Pairwise neighborhood-to-neighborhood in (I) Non- to preprogressors, (J) Non- to progressor and (K) Pre- to progressor samples. (M) Summary of overall neighborhood interaction scores across all samples. (*P<0.05, **P<0.01, ***P<0.001, ****P<0.0001, Fisher’s exact test).

Figure 5.. Gene expression analysis of metaplastic progression.

(A) RNAseq protocol showing stromal (green) and epithelial (purple) regions of interest. NGS LCM slides are cut between two CODEX slides (top and bottom). (B) H&E image of the studied core. (C) NGS LCM slides are cut between two CODEX slides (top and bottom). (D) Area of interest for laser capture microdissection: glandular epithelium (black), stroma (green). (E&F) t-distributed Stochastic Neighbor Embedding (t-SNE) demonstrating K means clusters of gene expression. (G) Percentage of non-progressor (NP), pre-progressor (PP) and progressor (P) samples associated with each K means cluster. (E) Heatmap of the 500 most differentially expressed genes supervised for K means cluster. Condition = NP, PP and P. (I) GSVA analysis of differentially expressed pathways supervised for K means. (J) Bagaev et al., devised a pan-cancer microenvironment gene expression signature that was successful in predicting immunotherapy outcome based on 4 expression signatures of anti-tumor, pro-tumor, angiogenesis/fibroblasts and EMT/proliferation. Based on this report, we subjected our LMC RNAseq dataset to approach an unsupervised analysis in NP, PP and P and K means clusters. (H) WCGNA analysis for Kmeans, NP, PP, P, pathology histology at progression and time (TPLS, time to progression or time to last known surveillance). (L-Q) Volcano plots demonstrating >log2 fold expression and adj P value<0.05 for (L) NP vs PP.), (M) NP vs P, (N) P vs P. Comparison of Kmeans clusters (O) 1 vs 2, (P) 1 vs 3 and (Q) 2 vs 3.

Figure 6.. Fibroblast interactions and gene expression alters through progression.

(A) WCGNA yellow module, altered in pre-progressors is enriched for extracellular matrix genes. (B) Enrichr analysis of cell lineage from yellow module. (C) Dot plot of cell specific CODEX panel markers for non-immune cells. CODEX staining images for PDPN and Arginase1 for (D) non-progressor, (E) pre-progressor and (F) progressor metaplasia. Pairwise, cell interaction scores at 100 arbitrary distance units for (G) non-progressor: pre-progressor, (H) pre-progressor: progressor and (I) non-progressor: progressor samples. (J) A summary dot-plot for all conditions showing the strength of attraction and repulsion interactions. (*P<0.05, Fischer’s exact test).

Figure 7.. Immune cell correlations with copy number variations.

(A) Unsupervised heat map of immune cell expression correlation with non-progressors or progressors (NP or P) with copy number changes (Orange= all CNVs except in chromosome 18 and 9, Green=Chr18 amplification and Chr9 loss, Red= Chr18 amplification only, Purple= Chr9 deletion only, Grey= diploid and Gradient= P53 expression level). (B) Correlation with NDBE progression status and CNVs supervised with concentration of each individual immune cell type. (C) Frequency of segment sizes of identified CNVs within non-progressor(NP) and progressor (P) NDBE.