Targetable fibroblast phenotypes and EMT malignant cell states cooperate to promote tumor progression in esophageal adenocarcinoma

doi:10.21203/rs.3.rs-2572575/v1

Download PDF

Research Article

Targetable fibroblast phenotypes and EMT malignant cell states cooperate to promote tumor progression in esophageal adenocarcinoma

https://doi.org/10.21203/rs.3.rs-2572575/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: Esophageal adenocarcinoma (EAC) is usually resistant to cytotoxic therapies and immunotherapies have gained little traction. Cancer-associated fibroblasts (CAF) are a major stromal cell population in the EAC tumor microenvironment associated with prognosis and treatment outcomes. Recent evidence suggests that cancer cell phenotypes related to EMT may determine CAF heterogeneity, but the molecular and cellular biology that underlies myofibroblast fate in EAC is not well understood. Methods: To obtain the most comprehensive profile of CAF heterogeneity in EAC, we performed histopathologic, single-cell RNA sequencing and transcriptomic analyses on 28 samples from 26 patients and prognostic validation using two EAC cohorts from genomic consortia. Results: Combining histologic and molecular profiles revealed five CAF phenotypes, including three myofibroblast phenotypes, associated with EMT-related signatures in EAC cells and cellular interactions that promote tumor progression and metastasis. We identified a specific myofibroblast subtype (CAF5) in close proximity to cancer cells and tumor vasculature with exclusive expression of TRPA1, offering a potential therapeutic vulnerability. We reconstructed CAF differentiation trajectories from esophagus-resident universal fibroblasts to identify candidate genes central to the CAF phenotype, and used this knowledge to construct a combined EMT-myofibroblast four gene signature (GSN, ATF1, ZEB2 and POSTN) that was highly prognostic in EAC and several other solid tumors. Conclusions: This study stratifies EAC patients into two prognostic groups and reports important data supporting a cancer – CAF signalling axis, promoting myofibroblastic differentiation, that may inform more effective treatment strategies.

Esophageal adenocarcinoma

microenvironment

fibroblasts

scRNA-seq

Esophageal adenocarcinoma (EAC) is resistant to current therapies and precision treatments have been disappointing, hampered by a scarcity of recurrent, targetable driver events (1) and cellular heterogeneity within and between primary tumors and metastases (2). Many cell types of the tumor microenvironment (TME) have been shown to be important determinants of cancer treatment and prognosis. Cancer-associated fibroblasts (CAFs) are recognized across cancers and are prognostic in EAC (3). We have shown that the tumor-promoting and chemoprotective properties of CAFs are associated with an activated myofibroblast (a-Smooth Muscle Actin/ACTA2⁺) phenotype that can be reversed with Phosphodiesterase type 5 inhibitors (PDE5i) (4). Recently, our colleagues provided a new mechanism for generating fibroblast heterogeneity, via the differential extracellular-vesicle (EV) transfer of miRNA, dependent on the epithelial or mesenchymal phenotype of the cancer, where mesenchymal colorectal cancer EVs (depleted for miR-200) did not suppress TGF-b-driven myofibroblast differentiation (5). The actin cytoskeleton promotes myofibroblast function during fibrogenesis and in tumor stroma remodeling (6). Numerous actin binding proteins are linked to cancer, leading to abnormal cytoskeleton architecture, instability and metastasis (Reviewed in (7)). One such protein is Gelsolin (GSN), a multifunctional regulator of cell structure that is a tumor suppressor in many cancers (8). GSN downregulation, in a glycoprotein panel, is a diagnostic biomarker for Barrett’s esophagus requiring clinical intervention (9). It is unknown whether this is cancer cell intrinsic and/or related to myofibroblast-derived changes during disease progression. Reduction of GSN in epithelial cells induces EMT and increases motility and invasiveness (10, 11) and in GSN null mice, dermal fibroblasts have excessive actin stress fibers and have increased contractility in vitro (12). Single-cell RNA sequencing (scRNA-seq) was performed on human tumors to identify the malignant and fibroblast cell states present and the cellular crosstalk across the EAC TME. We reconstructed CAF differentiation trajectories from esophagus-resident universal fibroblasts to identify candidate genes central to the cancer-associated fibroblast phenotype. We performed cellular assays to investigate the consequences of TGF-b exposure on GSN expression in cultured fibroblasts (13) and immunohistochemistry to gain spatial context in tumor tissue. Finally, we used pan-cancer genome datasets to assess the utility of GSN as part of a prognostic model.

Patient cohort

Human tissue samples were obtained from patients through University Hospitals Southampton NHS trust after informed consent. Twenty-two patients with EAC undergoing treatment with curative intent were recruited. Samples of their tumor were taken at either a staging investigation (n = 4) or resection (n = 20). Samples from staging investigation consisted of six to eight 2x2 mm endoscopic biopsies and samples from resection consisted of a single 8 mm punch biopsy.

Immunohistochemistry

Immunohistochemical staining was performed on 4um thick formalin-fixed paraffin embedded tissue sections. Appropriate heat-induced epitope retrieval for each stain was performed on a Dako PT Link instrument. Staining was performed individually for CD3 (IS503, Dako), EpCAM (Dako), POSTN (ab14041, abcam), α-SMA (M085129-2, Dako), and GSN (12953S, Cell Signaling Technologies) using a Dako Autostainer 48S machine according to the manufacturer’s instructions. A separate adjacent section was taken to perform routine hematoxylin & eosin staining to identify regions containing tumor and normal esophageal and gastric epithelium.

Drop-seq

Tissue was promptly disaggregated to a single cell suspension (14) and analyzed using Drop-seq (15), using PDMS microfluidic devices (https://dropletkitchen.github.io/). 1000 STAMPs per sample were then selected for PCR amplification (15 cycles), library preparation (Nextera XT, Illumina) and Illumina sequencing using a custom read 1 primer on the NextSeq-500 platform. Raw reads were processed and counted using Drop-seq-tools-v2.1, details provided in the Online supplement.

Unsupervised Clustering And Downstream Analysis

A Seurat (R package version 4.1.1) object was filtered for cell barcodes with 150-4,000 genes detected, less than 25% mitochondrial genes and less than 10% threshold for tissue dissociation signature genes (van den Brink et al., 2017). Counts were normalized and variable genes selected using SCTransform with latent variables to control for technical variation, prior to PCA, community detection (Louvain) and visualization using the RunUMAP function. Cluster-specific genes were detected using the Wilcoxon rank sum test (p < 1 × 10^− 3 was considered significant). Cell identity for each cluster was defined by the expression of canonical markers and a priori knowledge (Supplementary Data). Treatment-naïve sample data was used as reference atlas for subsequent automated cell identification in the chemotherapy and CRT-treated datasets using canonical correlation analysis, cells with a prediction score < 0.8 were excluded from further analysis. After manual inspection of lineage markers of each dataset, further low-quality or doublet cells were removed, as well as patients with low cellularity (n = 5). Module scoring was applied to the cancer cells using The MSigDB Hallmark collection (Liberzon et al., 2015; Subramanian et al., 2005) and Seurat’s “AddModuleScore” function. Quiescent cells were not analyzed further due to the low number of cells present (n = 256). For each separate cell lineage dataset, we iteratively pruned cells where we were uncertain of their cell identity (by marker log-normalized expression < 1). The pruned datasets were then re-clustered using the steps described above prior and as described in the manuscript.

Cell Culture And Analysis

Human Fetal Foreskin Fibroblast 2 (HFFF2) cells (ECACC Cat-no: 86031405) were maintained in Dulbecco’s modified Eagle’s medium (DMEM, Invitrogen) supplemented with 10% (v/v) fetal calf serum (Autogen Bioclear), 2 mM L-glutamine and 1% penicillin/streptomycin (Invitrogen) at 37°C and 5% CO2. TGF-β was used to induce myofibroblast differentiation and was manufactured in-house by Dr Patrick Duriez. HFFF2 were grown under cell culture conditions for 6h. After the cells adhered, media was then changed to low serum (1%) DMEM for 24h before the addition of 2 ng/mL TGF-β or vehicle (4 mM HCl, 0.4 mg/mL BSA) and grown for 72h. cDNA synthesis was performed using the High-Capacity Reverse Transcription Kit (Applied Biosystems). The TaqMan assays to POSTN, ACTA2, ZEB1, ZEB2, ATF1 and GSN were run in biological triplicate on the Real-time PCR system (ThermoScientific) with a VIC-TAMRA-labelled β-actin assay (ThermoScientific) as an internal control. Quantities of RNA per well were interpolated from a standard curve, normalized to the internal control samples as indicated. Protein expression was carried out by Western Blotting in biological triplicates. Protein was extracted using cell lysis buffer (Cell Signaling) and resolved on 3–8% Bis-Tris acrylamide gels and transferred onto PVDF membranes using the X-Cell II and iBlot systems (Invitrogen). Blocking and antibody incubations were done in 3% low-fat milk in PBS-0.025% Tween 20, and washes were in PBS–0.1% Tween 20. Images were collected using a CCD camera.

Statistical analysis

Statistical analysis was performed using R version 4.1.1. Enrichment of cell types and subtypes was calculated using 2x2 contingency table X² tests. Categorical variables were compared using the Chi-Squared test. A two-tailed Student’s t test was used to analyze parametric data and non-parametric data was analyzed by a Mann-Whitney test (unpaired data) or Wilcoxon test (paired data). To assess survival differences, Kaplan-Meier curves were produced and analyzed by Log-rank testing. p values < 0.05 were regarded as statistically significant. Unless stated otherwise, p values are represented as follows, *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001.

The cellular microenvironment of EAC tumors

Normal (n = 16) and tumor samples (n = 28) were obtained from 26 patients, of which 13 from treatment-naïve patients, 6 from patients treated with neoadjuvant chemoradiotherapy (CRT) and 9 from patients treated with neoadjuvant chemotherapy, representative of contemporary cohorts (Table 1/S1 & Figure S1A). Using the Mandard Tumor Regression Grade (TRG), patients were classified into responders (TRG 1–2; N = 4) and non-responders (TRG 4–5; N = 6); TRG3’s (N = 5) were considered separately due to their prognostically heterogenous behavior (16, 17). After data QC, we retained a total of 42,388 cells (Figure S1B), 13,368 from tumors and 29,020 cells from normal esophagus and we classified cells into 10 cell lineages (Supplementary Data 1 & Fig. 1B). We observed differences in cell lineage proportions between the treatment groups (Fig. 1C) and when comparing normal-derived cell lineages (Fig. 1D), there was an increase in plasma, macrophage, and T-cells in tumors. Expert pathological assessment of the available diagnostic tumor blocks (n = 19) was conducted and supplemented with immunohistochemistry (IHC) of CD3 (T cells), EpCAM (epithelial cancer cells), smooth muscle actin a-SMA (ACTA2) and Periostin (POSTN) (fibroblasts/CAFs) and our candidate gene, Gelsolin (GSN). Analysis of tumor areas separated patients into two clusters, where cluster-1 exhibited high staining for POSTN/a-SMA (CAF high) and cluster-2 high staining for CD3/EPCAM and low staining for POSTN/a-SMA (CAF low) (Figure S2A). There was a positive correlation between the number of T cells and cancer cells (r = .74, p < 0.0001; not shown) and GSN expression was inversely correlated with POSTN/a-SMA (Figure S2B). By visual assessment, lower GSN staining was observed in the stromal rich regions (a-SMA/POSTN) adjacent to areas of tumor staining for EpCAM (Figure S2B). There were no significant differences between gender, age, tumor location, treatment status, TRG and clinical staging between cluster-1 and − 2 patients (Table S1). However, cluster-1 patients (CAF high) were of higher pathological tumor (pT) stage (8 vs. 3 cases for pT stage ≥ 3) and higher pathological nodal (pN) stage (5 vs. 1 case for pN stage 3; p = 0.043) compared to cluster-2. Within pT stage 3, the most common stage treated with curative intent in clinical practice, 66% of those in cluster-2 had a pN stage of 1 or more in comparison to cluster-1, where 88% had pN stage of 1 or more. pN stage is the most important determinant of overall survival following surgery for EAC (18); and our data suggests a link between a-SMA/POSTN + ve stromal content in the primary tumor and local lymph node metastasis. Using scRNA-seq data we assigned patients into three ‘TME subtypes’: Fibroblast-high, Fibroblast-low, and Cancer-cell-high (Fig. 1E). In an unpaired analysis smaller numbers of T cells and cancer cells were seen in post-treatment samples when compared to pre-treatment samples (Fig. 1F). Comparison of IHC data across TME subtypes, highlighted a lower amount of T cells (~ 2-fold less) in the Fibroblast-high versus the Cancer-cell-high subtype (Fig. 1G), in keeping with CAFs impeding immune cell infiltration (19). There was not 100% concordance between our IHC and scRNA-seq groupings (Fig. 1H); these disparities may be attributable to sampling differences. We observed no significant correlations between TME subtypes and clinical parameters; but did observe two trends. Gastro-esophageal junction tumors were more prevalent in the Cancer-cell-high subtype (p = 0.06) and responders were more prevalent in the Fibroblast-high subtype (p = 0.08). The three compartments (tumor, immune and stroma) were analyzed separately to better characterize the cell types in the scRNA-seq dataset (Figure S3). The malignant cells primarily clustered by patient, reflecting the interpatient transcriptomic and genomic heterogeneity shown in previous studies (Fig. 2A, S3A), whereas cells from the immune and stromal compartments clustered by cell type (Figure S3).

Cancer cell EMT signatures associate with the Fibroblast-high subtype

To confirm the identity of the 4,452 malignant cells from tumors, marked by EAC genes (EPCAM, AGR2, CEACAM6 and KRT 8, 18 and 19) we characterized patterns of acquired copy number alterations by InferCNV analysis (20) (Figure S4). Patients with < 30 cells were removed, prior to re-clustering where the malignant cells formed 27 clusters and generally formed communities (N = 21/27 clusters) based on their patient origin (> 75% cells in a cluster), in keeping other tumor types (Fig. 2A, B). Next, we investigated whether cancer cell transcriptomes shared co-expression patterns enriched for MSigDB hallmark genesets (21). We aggregated the top 50 differentially expressed genes (DEGs) for each cluster into 9 co-expression modules using hierarchical clustering (Fig. 2C). We observed different levels of module activity across the TME subtypes where module 7 activity was significantly higher and module 1 and 5 lower in the Fibroblast-high subtype (Fig. 2D). We identified hallmark gene set enrichment (Fig. 2C) and annotated the modules accordingly (by EnrichR significance or a priori knowledge). We further measured the activity of these programs by aggregating the program scores for each patient (Fig. 2E) and inferred the interactions among the expression programs by correlation analysis. We observed that the EMT2 (module 7) signature co-occurred with the EMT1 (module 6) and Epi (module 9) signatures (Fig. 2F). Consistent with this, further hierarchical clustering of the averaged module scores for each patient resulted in three distinct groups of cases, one group was defined by activity of the EMT1-EMT2-Epi programs and was over-represented within Fibroblast-high tumors (Fig. 2G). In keeping with the mechanism described by Bhome et al (5), ZEB1 and ZEB2 (inducers of EMT; ZEB2 not shown), Vimentin (Vim), Cadherin 11 (CHD11) and a-SMA (ACTA2) were preferentially expressed in a small subset of cancer cell transcriptomes from patients defined as having an active EMT1-EMT2-Epi expression program (Fig. 2H). Furthermore, Fibroblast-high tumors scored higher for well-known mesenchymal marker genes relative to known epithelial genes (Fig. 2I) (22). miR-200C host gene transcripts (MIR200CHG) were detected 1.7–3.5 times less in the cancer cells from the Fibroblast-high, compared to the other TME subtypes (Fig. 2J). Together, the EMT1-EMT2-Epi program appeared to be more activated and prevalent in a fibroblast rich microenvironment and may represent sub-populations of malignant cells undergoing EMT.

Steady state and activated fibroblast phenotypes in EAC

We identified 6,415 fibroblasts (DCN + and/or VIM + cells), clustering into four normal-associated fibroblast (NOFs, 3,445 cells) and five cancer-associated fibroblast groups (CAFs, 2,970) (Fig. 3A). NOF1-4 clusters were marked by expression of PTGDS/APOD, PCOLCE2/SLPI, MT1X/MT2A and C7/APOD and CAF1-5 clusters by expression of COL1A1/COL3A1, IGFBP3/CXCL8, SFRP4/PI16, PLA2G2A/SFRP2 and TRPA1/F3, respectively (Fig. 3B). Focusing on the CAF clusters, CAF1, 2, and 5 overexpressed ACTA2 and POSTN in comparison to CAF3 and 4 and the NOF clusters (Fig. 3C-D, Supplementary data 3). These three POSTN + CAF clusters may represent myofibroblastic phenotypes (myoCAF) that have both prognostic value and can promote tumor cell invasion (3). The CAF1 contained 1,324 cells from 21 patients, with most cells (912/1,324; 68.9%) derived from just two patients, both responders to treatment. In total, 96% of these cells came from treated samples, suggesting they can persist after tumor regression. In contrast, CAF2 consisted of 520 cells from 22 different patients, and CAF5 consisted of 287 cells from 18 patients, where ~ 43 and ~ 58% of cells were from treatment naïve patients, respectively. Altogether, CAF1, 2 and 5 cells were detected in 24/26 patients and after excluding samples with low total CAF counts (n < 50), CAF1, 2 and 5 cells are on average 76% (95% CI: 55–96) of all CAF types. Excluding CAF1 cells, CAF2 and 5 on their own equate to ~ 36% (95% CI: 10–62) of all CAFs present (Supplementary Data 2). Suggesting CAF2 and 5 are ubiquitous and present in most patients before treatment. CAF1 markers, COL1A1 and COL3A1 encode components of the extracellular matrix with an important role in cell adhesion and differentiation (23); whereas CAF2 expressed IGFBP3 and IL8 that are known to regulate IGF signaling and tumor angiogenesis (24). CAF5 express F3 (CD142, tissue factor) and Transient Receptor Potential Cation Channel Subfamily A Member 1 (TRPA1); interestingly CD142 is normally only expressed in subendothelial cells (Reviewed in (25)). TRPA1, also known as the Wasabi receptor, is a Ca²⁺ dependent membrane bound stress sensor (26). CAF3 consisted of 516 cells from 14 cases, where ~ 90% were obtained from tumors following chemotherapy and shared many overexpressed genes with the inflammatory CAFs (iCAFs) previously identified in pancreatic ductal adenocarcinoma (PDAC) (27, 28). CAF3 also overexpressed the genes PI16, and DCN associated with a detox-iCAF population identified by Kieffer et al (29); while the presence of overexpression of Decorin (DCN) suggests that CAF3 may be related to desmoplastic CAFs, a phenotype that can overlap with iCAFs and present in normal and tumor tissues (30). CAF4 (n = 323 cells from 15 patients) was predominantly enriched from patients treated with CRT (82%, 265/323 cells from four patients). CAF4 shares its top marker gene PLA2G2A (phospholipase A2 group IIA) with metabolic CAFs previously identified in PDAC (31). However, they do not overexpress LDHB and PGK1 (not shown), genes associated with a high metabolic rate. Furthermore, they clustered with NOFs and next to CAF3 when considering CAF-associated gene-sets curated by Qian et al (32) (Fig. 3C). Together, this suggests that CAF3 and 4 may represent overlapping phenotypes present after treatment and may be closely related to fibroblasts seen in the normal esophagus, but distinct from the myoCAFs (CAF1, 2 and 5) (Fig. 3E). Focusing on the NOF clusters, NOF2 and 4 expressed PI16, DPT and CD34 (Fig. 3D). Fibroblasts with these characteristics have been described as adventitial or universal fibroblasts (33). It is thought that all fibroblasts, including CAFs, may originate from these cells. Interestingly, CAF3 shares many similarities with NOF2 and 4 (Fig. 3C-D) and had a high universal fibroblast score concomitant with the lowest myofibroblastic score (Fig. 3E).

Trajectory analysis reveals co-expression and gene-regulatory networks associated with myoCAF phenotypes

To decipher the relationship between universal fibroblasts and our cancer-associated fibroblasts we performed pseudotime transcriptional trajectory (Monocle3) and cellular trajectory reconstruction analysis (CytoTRACE). For Monocle3, we rooted the analysis in the UMAP space occupied by NOF2, marked by the universal fibroblast genes. This resulted in a looped trajectory, terminating with the largest pseudotime values in the part of the UMAP with the highest expression of myofibroblast gene expression (ACTA2, POSTN, PDGFRA) (Figure S5). CytoTRACE predicted the differentiation ordering of all fibroblast clusters ending in CAF2 and 5 with the highest scores (Fig. 3F). The top genes associated with myofibroblast differentiation based on their positive correlation with this score (Fig. 3G), are consistent with the myoCAF associated genes observed in Fig. 3C. In our IHC work, gelsolin was inversely correlated with a-SMA and periostin levels in tumor tissue. Gelsolin (GSN) was negatively correlated with the CytoTRACE score and its mRNA expression was significantly reduced in CAF2 and 5 (Fig. 3G). While GSN expression was lower in CAF1 and CAF4; this was not the case for CAF3. These observations indicate a relationship with GSN down-regulation in most myoCAF phenotypes. This is further supported by gene co-expression analysis using COTAN (34). Where, GSN was situated in a cluster of 114 genes (including PI16, DPT, and CFD) (Fig. 3H). The most enriched gene signature for this cluster was a lung adventitial fibroblast signature (P < 0.0001, 58 of 410 genes), these cells express genes involved in canonical fibroblast functions, but are distinct from lung myoCAFs (35) (Supplementary Data 4, 5). POSTN and ACTA2 are part of a cluster of 262 genes (Fig. 3H) that are involved in extracellular matrix, actin cytoskeleton organization, cell migration and possessed an enriched signature resembling that of pancreas mesenchymal stromal cells (P < 0.0001, 169 of 1,282 genes) (36). This suggests that genes co-expressed with POSTN/ACTA2 are involved in tissue remodeling and subsequent myofibroblast formation, while the opposite is true for genes co-expressed with GSN. In myoCAF phenotypes from scRNA-seq data in colorectal, gastric, breast, head and neck, esophageal squamous, and pancreatic cancers, significantly lower GSN counts were observed (Fig. 3I). In TCGA EAC cases, we found that lower GSN mRNA expression is associated with shorter overall survival (OS) (HR = 1.92; p = 0.042; https://kmplot.com, accessed Jan2023). To determine possible regulators of this potential pan-cancer phenomenon we applied SCENIC to the data (37). Single-cell regulatory network inference revealed gene regulatory networks for each fibroblast cluster and investigation of the predicted high-confidence regulons for CAF2 and 5 indicated MYLK and STAT1 as regulators of ACTA2 and POSTN expression (Fig. 3J). Probable regulators of GSN expression, included HLF and BCLAF1 a known transcriptional repressor upregulated in all myoCAFs (Fig. 3J). Scenic annotates 32 BCLAF1 binding motifs in GSN (+/- 10kb TSS) in comparison to only 6 for HLF; and high BCLAF1 expression in TCGA was associated with shorter OS (Figure S6A). Thus, our data identifies CAF2 and CAF5 as major POSTN + myoCAF populations associated with GSN downregulation and a distinct transcription factor network.

myoCAFs are important in the EAC cellular interaction network

To further understand the importance of myoCAFs in the EAC TME, we reconstructed ligand-receptor (LR) interactions by CellPhoneDB and utilized network-based analyses (CrossTalkR) to prioritize changes in fibroblast crosstalk to the cancer, immune and stroma compartments (38, 39). CAF5 appears to have the most influence on the CAF interactome with the most interactions across all three TME components (Fig. 4A). In the CAF Secretome, both CAF2 and 5 are distinguished by the highest number of interactions with non-malignant cells and CAF5 is stratified further, with the largest number of secreted factors targeting cancer cells (Fig. 4B & Supplementary Data 6 & 7). Conversely, CAF3 had the lowest number of interactions. Enrichment analysis highlighted a core interactome enriched for extracellular matrix terms for all CAFs (Fig. 4C). CAF2 was enriched for IL8 receptor binding terms, CAF4 was enriched for immunosuppression and CAF5 for anaplasia. Only the CAF2 and 5 shared interactions enriched for tumor angiogenesis (Supplementary Data 8). Several myofibroblast interactions involved cancer cell crosstalk with CAF2 and 5 and notably included Annexin 2 (PLAT/ANXA2) known to promote invasion and metastasis (40) and Neuregulin 1 (ERBB3/NRG1) known to activate the HER2-HER3 signaling axis (41). Generally, CAF5 was distinguished from CAF2-cancer crosstalk through epithelial growth factor signaling interactions (Fig. 4D). They express epiregulin (EREG) and were predicted to interact with EGFR on cancer cells. Fibroblast-derived epiregulin can promote tumor growth, via the JAK2-STAT3 pathway (42). CD44 is a known oncogenic regulator, CAF5 were predicted to interact with cancer cells through secretion of HGF. c-MET (MET) expression was weakly correlated with CD44 expression in the presence (Spearman’s R² = 0.091; FDR = 0.016), but not in absence of CAF5 cells (R² = 0.023; FDR = 0.191). These molecules are mechanistically implicated in metastasis (43). Each CAF cluster had more than 100 cell contact interactions (membrane bound ligand-receptor pair) with the endothelium originating from a CAF cell (Fig. 4E, top); we found interactions through angiogenic stimulatory signals from CAFs (including CAF2 and 5) via VEGF and its receptors FLT1/VEGFR1 and KDR/VEGFR2 along with growth factor signaling via PDGFB-PDGFR interactions. The cell-pairs, directionality and strength of signaling (scaled mean expression of LR pairs) for some of these interactions are in Fig. 4F. Notch signaling in fibroblast activation is well established (44). However, direct heterotypic interactions are not well described in EAC. We observed several contact interactions between CAF2 and 5, via the DLL1 ligand to NOTCH receptors expressed on endothelium, pericytes and smooth muscle populations (Fig. 4F). This suggests myoCAFs may be involved in juxtracrine signaling to the vasculature. Next, we investigated which immune subpopulations (Figure S7) communicate with CAFs. Immunomodulatory signaling was a prominent feature, the CAF interactome was dominated by interactions with T cells and antigen presenting cells (Fig. 4E, top); in the CAF secretome, antigen presenting cell type partners were more dominant (Fig. 4E, bottom). Many known CAF-immune interactions were predicted and are shown in Fig. 4F. Conversely, the predicted PLAU/PLAUR interaction exclusive to CAF2 and 5 and mono/macrophages/cDC2s, suggests they may be important for innate immune cell adhesion and retention (45, 46). In keeping with distinct functions for myoCAF subtypes, CAF5 were not predicted to signal via CXCL12-CXCL4 (Fig. 4F, bottom plot). Interestingly, CAF5 is marked by high levels of CXCL14 and TRPA1 expression that was consistently expressed in cells from either treated or naïve samples (Fig. 4G). In our scenic analysis, HIF1A was annotated as a regulator of TRPA1 expression and CXCL14 can be induced in hypoxia (47). When we correlated the abundance of immune cells with fibroblasts, we observed a general negative correlation (for CD8+/Cytotoxic T cells) for CAFs in comparison to NOFs (Fig. 4H). With CAF2 and 5 we found positive correlations with CD14+/CD16 + monocytes, M2-like macrophages, Tregs, NK, MDSCs and DC abundance in comparison to other CAF subtypes. Utilizing CrossTalkR, we found that POSTN + CAFs had the second highest Pagerank (Fig. 4I) indicating that they are important signalers and the most significantly deregulated KEGG-pathways associated with Pagerank in the tumor network, included Calcium signaling (Figure S8); the genes enriching these pathways are widely expressed in CAF2 and 5 and are responsive to hypoxia.

Stromal cell populations with reduced gelsolin expression surround tumor nests

Our single cell data and compelling evidence from other cancers (Fig. 3I) highlighted a negative relationship between GSN expression and myoCAF differentiation. To validate this, we investigated GSN in HFFF2 cells following TGF-β exposure. Upon stimulation with TGF-β, HFFF2 fibroblasts showed a flattened shape, indicative of a myofibroblastic phenotype after 24h (Figure S6B). mRNA and protein expression of ACTA2, POSTN; our candidate gene GSN and its regulator, ATF1 (48) were measured from 6h to 72h. As expected, POSTN, ACTA2 (α-SMA) increased and plateaued at 6h and 24h, respectively (Figure S6C); but TGF-β did not induce downregulation of GSN (Figure S6D) or modulation of ATF1 levels (Figure S6D). Protein expression of α-SMA was increased at 48h, but there was no difference in GSN (Figure S6E). Re-analysis of our previously published scRNA-seq data (4) from ex vivo patient-derived CAFs co-cultured with MFD1 EAC cells, demonstrated lower expression of GSN and overexpression of ACTA2 and other myoCAF markers and MYLK, the predicted regulator of POSTN and ACTA2 expression (Fig. 5A). To better understand the role of Gelsolin in the wider TME, we revisited the IHC data, we observed that expression is localized to intratumoral fibroblast and lymphocyte populations. Staining of individual cells localized to the basal and suprabasal layers within the normal squamous epithelium of these tumor specimens is also apparent. It is also expressed within lymphoid aggregates and immune infiltrate with a heterogeneous membranous and cytoplasmic expression pattern (Figure S9). While tumor epithelium heterogeneously and weakly stained in several of our cases (n = 5, Supplementary Data 9), most tumor cells in our cohort do not express GSN (74%, n = 14/19), in keeping with its role as a tumor suppressor. We compared GSN expression to adjacent sections of tumor stained for POSTN, α-SMA, CD3 and EpCAM (Fig. 5B). Surrounding tumor nests, we found distinct peritumoral stromal regions positive for POSTN/α-SMA (i.e., myoCAFs) that did not appear to express GSN. Furthermore, intratumoral EpCAM expression was reduced among IHC cluster-1 (CAF-high) patients (Figure S2). This result is suggestive of a loss of epithelial cell status. IHC cluster-1 also had less CD3 + lymphocyte infiltration than cluster-2 patients, this difference was unlikely to be due to T-cell exclusion to the periphery (only 21% immune-excluded, n = 4/19, Supplementary Data 9), nor due to an absence of T-cells (only 11% immune-desert, n = 2/19).

A combined EMT and myofibroblast signature is prognostic across cancers

We discovered that a weighted expression model (See methodology) including GSN, Gelsolin and EMT regulators and myofibroblast genes (i.e., inverted-GSN, ATF1, ZEB2 and POSTN) was associated with a reduction in overall survival (Fig. 6A). Where strikingly, TCGA EAC cases with low values did not reach median OS after 46 months of follow-up; furthermore, this expression model retained independence in a multivariant analysis, including age, gender and tumor stage; with a c-index of 0.7. When considering the other pan-cancer datasets from the TGCA study, our expression model was statistically significant (P ≤ 0.05) for OS in 11 out of 21 cancers (Table S4) and generally associated with poorer OS outcomes (n = 9/11). We then applied our prognostic model to the independent OCCAMs cohort of EAC cases analyzed by bulk tumor RNA sequencing (n = 139 cases) (49). In this second patient cohort, we validated our model using both univariate and multivariate analyses; where cases with a low or high score achieved a median OS of 44 and 19 months, respectively (Fig. 6A). We used the scRNA-seq-derived cell signatures in a Gene Set Variation Analysis to deconvolute the bulk RNA-seq data into myoCAF, CD8 + T cell, Treg, NK, M2-like macrophage (TAMs) and DC associated gene sets and stratified OCCAMs cases by their expression model score. Interestingly, patients with high values (associated with shorter OS) clustered into distinct groups with clear enrichment for myoCAF and/or immune signatures (Fig. 6B).

Currently, the molecular and cellular biology that underlies myofibroblast fate in EAC is not well understood. Our initial hypothesis was that cancer cells undergoing EMT promote myofibroblast differentiation (5). Our data suggests that there may be cooperation between EAC cancer cells and cancer-associated fibroblast subtypes in specific cellular neighborhoods. The myofibroblast CAF5 subtype is likely located close to epithelial tumor cells and the endothelium, whilst CAF1 and CAF2 subtypes are more distant from cancer cells in the tumor tissue. The myoCAF subtypes were positive for POSTN expression, suggesting a shared origin (50) and these populations may represent the cycling of different transitional cell states from early activated myoCAFs (CAF5) through to more terminal differentiated myoCAFs (CAF2) to a population of less contractile and more ECM/ pro-collagen producing cells (CAF1). The proximity of CAF5 to cancer cells and the endothelium makes this myoCAF subtype of particular interest. CAF5 are exclusive in their expression of TRPA1 mRNA. TRPA1 is known to regulate calcium influx and can control fibrosis via fibroblast activation (51). Our data suggests that TRPA1 expression is activated by HIF1A (52) and therefore, CAF5 cells may reside in a hypoxic area of the tumor. Restricted TRPA1 expression is observed in the Podoplanin (PDPN+) selected and adipogenic (PTGDS+) EAC-derived CAF populations recently described by Croft et al (53), concomitant with the growth factor expression profile identified in our CAF5 phenotype (Figure S10). Interestingly, most of our myoCAFs did not express PDPN or PTGDS mRNA, suggesting that these important functional phenotypes are not restricted to a particular cell of origin and current CAF markers may not fully capture the molecular and physiological characteristics of these populations. A study in prostate cancer CAFs has identified that TRPA1 activation by resveratrol leads to strong Ca² + cytosolic influx and secretion of VEGF and HGF (54). In the same study, activation of TRPA1 in a co-culture model reduced prostate cancer cell death by 40%. CAF5 cells also expressed VEGF, VEGFR and HGF along with other growth factors that had importance in our predicted EAC interaction network, suggesting that CAF5 may function similarly in EAC. Contrastingly, AITC-induced TRPA1 activation has been shown to augment ERK1/2 phosphorylation in cultured lung fibroblasts, which in turn may inhibit TGF- β receptor signaling (26). TRPA1 targeting agents should be employed to investigate myoCAF function further. We identified expression of the chemokine CXCL14 by CAF5 irrespective of tumor treatment. CXCL14 certainly influences leukocyte migration, but its precise role in cancer has not been clearly defined. The effects of CXCL14 are likely to be cell and context specific and dependent on the overall cytokine milieu of the tumor. CXCL14 is an angiostatic factor that can be induced by hypoxia and is known to stimulate the autocrine growth of CAFs, cause tumor infiltration by monocytes and immature DCs and act as chemoattractant for macrophages and NK cells, but not T cells (55). Work in colorectal cancer and pancreatic cancer syngeneic mouse models identified a chemokine switch in CAFs, induced by TGFB1 inhibition using G9a and EZH2 inhibitors, from CXCL14 expression towards CXCL9/10 production and subsequent antigen-specific T cell recruitment. Importantly, the suppression of chemoimmunotherapy efficacy was negated and led to improved survival (56). Licensed EZH2 inhibitors could target this chemokine switch in the human tumor context and the myoCAF phenotype responsible for CXCL14 production in EAC might be a good target for immunomodulatory therapies. Linking cancer EMT activity and myoCAF differentiation, we observed down-regulation of GSN in myoCAFs and demonstrated that this is not solely driven by TGF-β exposure in vitro. Suggesting that other factors may regulate gelsolin, and loss of gelsolin expression may not be an absolute requirement for production of the myoCAF phenotypes but could alter their morphology and motility in vivo (57, 58). Our ex vivo data suggests that intimate cancer/CAF interactions, and specific tissue states, are required for the full induction of these myofibroblast phenotypes.

Our findings suggest that cancer cells undergoing EMT in specific tissue niches drive the differentiation of myoCAFs via pathways that may not involve TGF-β, such as Ca²⁺ signaling via TRP receptors (59). This is intriguing as it is known that loss of mitochondrial Ca²⁺ uptake promotes myofibroblast differentiation by epigenetic reprogramming (60). Calcium signaling has been implicated in cancer-associated fibroblast-induced drug resistance in ovarian cancer, but the key players of calcium signaling in EAC are unclear (61). In this context, CAF5 associate with both cancer cells and the peritumoral endothelium promoting angio-genesis or –stasis, secrete specific immunosuppressive chemokines (e.g., CXCL14) and participate in growth factor signaling that would amplify both fibrosis and cancer progression (Fig. 6C). In summary, this study is a comprehensive compendium of cell phenotypes that provides insights into CAF biology and cancer progression in EAC; myoCAFs represent a potential therapeutic axis to repurpose agents to enhance current therapies and immunotherapy.

EAC

esophageal adenocarcinoma

CAF

cancer associated fibroblast

EMT

epithelial to mesenchymal Transition

TME

tumor microenvironment

extracellular-vesicle

PDMS

polydimethylsiloxane

STAMPs

single-cell transcriptomes attached to microparticles

PCA

principal component analysis

CRT

chemoradiotherapy

quality control

IGF

insulin growth factor

UMAP

Uniform Manifold Approximation and Projection

ligand-receptor

FDR

false discovery rate.

Ethics approval and consent to participate

This study was approved by The Proportionate Review Sub-Committee of the North East - Newcastle & North Tyneside 1 Research Ethics Committee. (REC No: 18/NE/0234).

Consent for publication

All authors give full consent for publication

Availability of data and materials

The scRNA-seq datasets supporting the conclusions of this article are available in the NCBI GEO repository, Accession number: GSE173950.

Competing interests

The authors declare no competing interests.

Funding

TJU is supported by a Royal College of Surgeons of England and Cancer Research UK Advanced Clinician Scientist Fellowship (A23924). RCW is supported by a Cancer Research UK Clinical Research Training Fellowship (A25162). MJJRZ is supported by a John Goldman Fellowship (2016/JGF/0003) and the Southampton CRUK Centre Development Fund. MS is supported by a UKRI Future Leaders Fellowship (MR/T042184/1). This work was supported by a grant from the Chan Zuckerberg Initiative (Human Cell Atlas Pilot project: START: Standardization of Single-Cell and Single-Nucleus RNA-Seq Protocols for Tumors) to JJW, TJU and MJJRZ. Results shown here are in part based upon data generated by the TCGA Research Network (https://www.cancer.gov/tcga). Development of Drop-seq technology in Southampton was funded by MRC grant (MC_PC_15078).

Authors' contributions

Conceptualization, R.C.W., S.P.B., M.J.J.R-Z, T.J.U; Methodology, R.C.W., M.J.J.R-Z., S.P.B., T.J.U.; Formal analysis, S.P.B., M.J.J.R-Z., I.R., M.S.; Investigation, S.P.B., R.C.W., J.H., B.G., M.J.J.R-Z. Resources, J.J.W.; Data Curation, S.P.B., R.C.W., B.G., J.H.; Writing – Original Draft, S.P.B., R.C.W., M.J.J.R-Z.; Writing – Review & Editing, M.J.J.R-Z., S.P.B., T.J.U.; Visualization, S.P.B., B.P.S., M.S.; Supervision, T.J.U., M.J.J.R-Z.; Project Administration, R.C.W., S.P.B., M.J.J.R-Z., T.J.U.; Funding Acquisition, R.C.W., S.P.B., M.J.J.R-Z., T.J.U.

Acknowledgements

The authors thank University Hospital Southampton patients for their participation in this study and the Faculty of Medicine Tissue Bank at the University of Southampton. Dr Patrick Duriez, Protein Core Laboratory, Centre for Cancer Immunology, University of Southampton. WISH laboratory, Faculty of Medicine, University of Southampton.

Oesophageal Cancer Clinical and Molecular Stratification (OCCAMS) Consortium:

Rebecca C. Fitzgerald¹, Paul A.W. Edwards^1,2, Nicola Grehan¹, Barbara Nutzinger¹, Caitriona Hughes¹, Elwira Fidziukiewicz¹, Shona MacRae¹,Alex Northrop¹, Xiaodun Li¹,Annalise Katz-Summercorn¹,Sujath Abbas¹,Maria O’Donovan^1,3, Ahmad Miremadi^1,3, Shalini Malhotra^1,3, Monika Tripathi^1,3, EC Smyth³⁸, Simon Tavaré², Andy G. Lynch², Matthew Eldridge², Maria Secrier², Ginny Devonshire², Sriganesh Jammula², Aisling M Redmond¹, Sarah Killcoyne¹ Amber Grantham¹, Adrienn Blasco¹, Jim Davies⁵, Charles Crichton⁵, Nick Carroll⁶, Peter Safranek⁶, Andrew Hindmarsh⁶, Vijayendran Sujendran⁶, J. Robert O’Neill⁶, Stephen J. Hayes^7,14, Yeng Ang^7,8,29, Andrew Sharrocks²⁹, Shaun R. Preston⁹, Sarah Oakes⁹, Izhar Bagwan⁹, Vicki Save¹⁰, Richard J.E. Skipworth¹⁰, Ted R. Hupp²³, Olga Tucker^11,33, Andrew Beggs^11,28, Philippe Taniere¹¹, Sonia Puig¹¹, Gianmarco Contino¹¹,Timothy J. Underwood^12,13, Robert C. Walker^12,13, Ben L. Grace¹², Hugh Barr¹⁵,Neil Shepherd¹⁵, Oliver Old¹⁵, Jesper Lagergren^16,25, James Gossage^16,24, Andrew Davies^16,24, Fuju Chang^16,24, Janine Zylstra^16,24, Ula Mahadeva¹⁶, Vicky Goh²⁴, Francesca D. Ciccarelli²⁴, Grant Sanders¹⁷, Richard Berrisford¹⁷, Catherine Harden¹⁷, Mike Lewis¹⁸, Ed Cheong¹⁸, Bhaskar Kumar¹⁸,Simon L Parsons¹⁹, Irshad Soomro¹⁹, Philip Kaye¹⁹, John Saunders¹⁹, Laurence Lovat²⁰, Rehan Haidry²⁰, Michael Scott²², Sharmila Sothi²⁶, Sari Suortamo²⁶, Suzy Lishman²⁷, George B. Hanna³¹, Christopher J. Peters³¹,Krishna Moorthy³¹, Anna Grabowska³², Richard Turkington³⁴, Damian McManus³⁴, Helen Coleman³⁴, David Khoo³⁵, Will Fickling³⁵, Tom DL.Crosby³⁶ , Russell D Petty³⁷

¹Medical Research Council Cancer Unit, Hutchison/Medical Research Council Research Centre, University of Cambridge, Cambridge, UK

²Cancer Research UK Cambridge Institute, University of Cambridge, Cambridge, UK

³Department of Histopathology, Addenbrooke’s Hospital, Cambridge, UK

⁴Oxford ComLab, University of Oxford, UK, OX1 2JD

⁵Department of Computer Science, University of Oxford, UK, OX1 3QD

⁶Cambridge University Hospitals NHS Foundation Trust, Cambridge, UK, CB2 0QQ

⁷Salford Royal NHS Foundation Trust, Salford, UK, M6 8HD

⁸Wigan and Leigh NHS Foundation Trust, Wigan, Manchester, UK, WN1 2NN

⁹Royal Surrey County Hospital NHS Foundation Trust, Guildford, UK, GU2 7XX

¹⁰Edinburgh Royal Infirmary, Edinburgh, UK, EH16 4SA

¹¹University Hospitals Birmingham NHS Foundation Trust, Birmingham, UK, B15 2GW

¹²University Hospital Southampton NHS Foundation Trust, Southampton, UK, SO16 6YD

¹³Cancer Sciences Division, University of Southampton, Southampton, UK, SO17 1BJ

¹⁴Faculty of Medical and Human Sciences, University of Manchester, UK, M13 9PL

¹⁵Gloucester Royal Hospital, Gloucester, UK, GL1 3NN

¹⁶Guy’s and St Thomas’s NHS Foundation Trust, London, UK, SE1 7EH

¹⁷Plymouth Hospitals NHS Trust, Plymouth, UK, PL6 8DH

¹⁸Norfolk and Norwich University Hospital NHS Foundation Trust, Norwich, UK, NR4 7UY

¹⁹Nottingham University Hospitals NHS Trust, Nottingham, UK, NG7 2UH

²⁰University College London, London, UK, WC1E 6BT

²¹Norfolk and Waveney Cellular Pathology Network, Norwich, UK, NR4 7UY

²²Wythenshawe Hospital, Manchester, UK, M23 9LT

²³Edinburgh University, Edinburgh, UK, EH8 9YL

²⁴King’s College London, London, UK, WC2R 2LS

²⁵Karolinska Institute, Stockholm, Sweden, SE-171 77

²⁶University Hospitals Coventry and Warwickshire NHS, Trust, Coventry, UK, CV2 2DX

²⁷Peterborough Hospitals NHS Trust, Peterborough City Hospital, Peterborough, UK, PE3 9GZ

²⁸Institute of Cancer and Genomic sciences, University of Birmingham, B15 2TT

²⁹GI science centre, University of Manchester, UK, M13 9PL.

³⁰Queen’s Medical Centre, University of Nottingham, Nottingham, UK, NG7 2UH

³¹Department of Surgery and Cancer, Imperial College, London, UK, W2 1NY

³²Queen’s Medical Centre, University of Nottingham, Nottingham, UK

³³Heart of England NHS Foundation Trust, Birmingham, UK, B9 5SS.

³⁴Centre for Cancer Research and Cell Biology, Queen’s University Belfast, NI, UK, BT7 1NN

³⁵Queen’s Hospital, Romford, RM7 0AG

³⁶Velindre University NHS Trust, Cardiff, Wales, UK, CF15 7QZ

³⁷Medical Oncology, Division of Molecular and Clinical Medicine, Ninewells Hospital and Medical School, University of Dundee, James Arnott Drive, Dundee, Scotland ,DD6 8JN.

³⁸Department of Oncology, Cambridge University Hospitals NHS Foundation Trust, Cambridge, UK, CB2 0QQ

Kim J, Bowlby R, Mungall AJ, Robertson AG, Odze RD, Cherniack AD, et al. Integrated genomic characterization of oesophageal carcinoma. Nature. 2017;541(7636):169–74. date = 2017-01-04.
Pectasides E, Stachler MD, Derks S, Liu Y, Maron S, Islam M, et al. Genomic heterogeneity as a barrier to precision medicine in gastroesophageal adenocarcinoma. Cancer Discov. 2018;8(1, pmid––28978556):37–48. date = 2018-01-01.
Underwood TJ, Hayden AL, Derouet M, Garcia E, Noble F, White MJ, et al. Cancer-associated fibroblasts predict poor outcome and promote periostin-dependent invasion in oesophageal adenocarcinoma. J Pathol. 2015;235(3):466–77.
Sharpe BP, Hayden A, Manousopoulou A, Cowie A, Walker RC, Harrington J, et al. Phosphodiesterase type 5 inhibitors enhance chemotherapy in preclinical models of esophageal adenocarcinoma by targeting cancer-associated fibroblasts. Cell Rep Med. 2022;3(6):100541.
Bhome R, Emaduddin M, James V, House LM, Thirdborough SM, Mellone M, et al. Epithelial to mesenchymal transition influences fibroblast phenotype in colorectal cancer by altering miR-200 levels in extracellular vesicles. J Extracell Vesicles. 2022;11(5):e12226.
Hanley CJ, Noble F, Ward M, Bullock M, Drifka C, Mellone M, et al. A subset of myofibroblastic cancer-associated fibroblasts regulate collagen fiber elongation, which is prognostic in multiple cancers. Oncotarget. 2016;7(5):6159–74.
Aseervatham J. Cytoskeletal Remodeling in Cancer. Biology (Basel). 2020;9(11).
Feldt J, Schicht M, Garreis F, Welss J, Schneider UW, Paulsen F. Structure, regulation and related diseases of the actin-binding protein gelsolin. Expert Rev Mol Med. 2019;20:e7.
Shah AK, Hartel G, Brown I, Winterford C, Na R, Cao KL, et al. Evaluation of Serum Glycoprotein Biomarker Candidates for Detection of Esophageal Adenocarcinoma and Surveillance of Barrett's Esophagus. Mol Cell Proteomics. 2018;17(12):2324–34.
Chen ZY, Wang PW, Shieh DB, Chiu KY, Liou YM. Involvement of gelsolin in TGF-beta 1 induced epithelial to mesenchymal transition in breast cancer cells. J Biomed Sci. 2015;22:90.
Van den Abbeele A, De Corte V, Van Impe K, Bruyneel E, Boucherie C, Bracke M, et al. Downregulation of gelsolin family proteins counteracts cancer cell invasion in vitro. Cancer Lett. 2007;255(1):57–70.
Witke W, Sharpe AH, Hartwig JH, Azuma T, Stossel TP, Kwiatkowski DJ. Hemostatic, inflammatory, and fibroblast responses are blunted in mice lacking gelsolin. Cell. 1995;81(1):41–51.
Miyazaki Y, Oda T, Inagaki Y, Kushige H, Saito Y, Mori N, et al. Adipose-derived mesenchymal stem cells differentiate into heterogeneous cancer-associated fibroblasts in a stroma-rich xenograft model. Sci Rep. 2021;11(1):4690.
Waise S, Parker R, Rose-Zerilli MJJ, Layfield DM, Wood O, West J, et al. An optimised tissue disaggregation and data processing pipeline for characterising fibroblast phenotypes using single-cell RNA sequencing. Sci Rep. 2019;9(1):9580. date = 2019-12-03.
Macosko EZ, Basu A, Satija R, Nemesh J, Shekhar K, Goldman M, et al. Highly Parallel Genome-wide Expression Profiling of Individual Cells Using Nanoliter Droplets. Cell. 2015;161(5):1202–14.
Mancini R, Pattaro G, Diodoro MG, Sperduti I, Garufi C, Stigliano V, et al. Tumor Regression Grade After Neoadjuvant Chemoradiation and Surgery for Low Rectal Cancer Evaluated by Multiple Correspondence Analysis: Ten Years as Minimum Follow-up. Clin Colorectal Cancer. 2018;17(1):e13–e9.
Noble F, Lloyd MA, Turkington R, Griffiths E, O'Donovan M, O'Neill JR, et al. Multicentre cohort study to define and validate pathological assessment of response to neoadjuvant therapy in oesophagogastric adenocarcinoma. Br J Surg. 2017;104(13):1816–28.
Rahman SA, Walker RC, Maynard N, Trudgill N, Crosby T, Cromwell DA et al. The AUGIS Survival Predictor: Prediction of Long-term and Conditional Survival after Esophagectomy Using Random Survival Forests. Ann Surg. 2021.
Belhabib I, Zaghdoudi S, Lac C, Bousquet C, Jean C. Extracellular Matrices and Cancer-Associated Fibroblasts: Targets for Cancer Diagnosis and Therapy? Cancers. (Basel). 2021;13:14.
Tirosh I, Izar B, Prakadan SM, Wadsworth MH 2nd, Treacy D, Trombetta JJ, et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science. 2016;352(6282):189–96.
Liberzon A, Birger C, Thorvaldsdottir H, Ghandi M, Mesirov JP, Tamayo P. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst. 2015;1(6):417–25.
Salt MB, Bandyopadhyay S, McCormick F. Epithelial-to-mesenchymal transition rewires the molecular path to PI3K-dependent proliferation. Cancer Discov. 2014;4(2):186–99.
Kauppila S, Stenback F, Risteli J, Jukkola A, Risteli L. Aberrant type I and type III collagen gene expression in human breast cancer in vivo. J Pathol. 1998;186(3):262–8.
Pausch TM, Aue E, Wirsik NM, Freire Valls A, Shen Y, Radhakrishnan P, et al. Metastasis-associated fibroblasts promote angiogenesis in metastasized pancreatic cancer via the CXCL8 and the CCL2 axes. Sci Rep. 2020;10(1):5420.
Ruf W, Disse J, Carneiro-Lobo TC, Yokota N, Schaffner F. Tissue factor and cell signalling in cancer progression and thrombosis. J Thromb Haemost. 2011;9(Suppl 1):306–15.
Geiger F, Zeitlmayr S, Staab-Weijnitz CA, Rajan S, Breit A, Gudermann T et al. An Inhibitory Function of TRPA1 Channels in TGF-beta1-driven Fibroblast to Myofibroblast Differentiation.Am J Respir Cell Mol Biol. 2022.
Elyada E, Bolisetty M, Laise P, Flynn WF, Courtois ET, Burkhart RA, et al. Cross-Species Single-Cell Analysis of Pancreatic Ductal Adenocarcinoma Reveals Antigen-Presenting Cancer-Associated Fibroblasts. Cancer Discov. 2019;9(8):1102–23.
Orth M, Metzger P, Gerum S, Mayerle J, Schneider G, Belka C, et al. Pancreatic ductal adenocarcinoma: biological hallmarks, current status, and future perspectives of combined modality treatment approaches. Radiat Oncol. 2019;14(1):141.
Kieffer Y, Hocine HR, Gentric G, Pelon F, Bernard C, Bourachot B, et al. Single-Cell Analysis Reveals Fibroblast Clusters Linked to Immunotherapy Resistance in Cancer. Cancer Discov. 2020;10(9):1330–51. date = 2020-09.
Simon T, Salhia B. Cancer-Associated Fibroblast Subpopulations With Diverse and Dynamic Roles in the Tumor Microenvironment. Mol Cancer Res. 2022;20(2):183–92.
Wang Y, Liang Y, Xu H, Zhang X, Mao T, Cui J, et al. Single-cell analysis of pancreatic ductal adenocarcinoma identifies a novel fibroblast subtype associated with poor prognosis but better immunotherapy response. Cell Discov. 2021;7(1):36.
Qian J, Olbrecht S, Boeckx B, Vos H, Laoui D, Etlioglu E, et al. A pan-cancer blueprint of the heterogeneous tumor microenvironment revealed by single-cell profiling. Cell Res. 2020;30(9):745–62.
Buechler MB, Pradhan RN, Krishnamurty AT, Cox C, Calviello AK, Wang AW, et al. Cross-tissue organization of the fibroblast lineage. Nature. 2021;593(7860):575–9.
Galfre SG, Morandin F, Pietrosanto M, Cremisi F, Helmer-Citterich M. COTAN: scRNA-seq data analysis based on gene co-expression. NAR Genom Bioinform. 2021;3(3):lqab072.
Travaglini KJ, Nabhan AN, Penland L, Sinha R, Gillich A, Sit RV, et al. A molecular cell atlas of the human lung from single-cell RNA sequencing. Nature. 2020;587(7835):619–25.
Muraro MJ, Dharmadhikari G, Grun D, Groen N, Dielen T, Jansen E, et al. A Single-Cell Transcriptome Atlas of the Human Pancreas. Cell Syst. 2016;3(4):385–94. e3.
Aibar S, Gonzalez-Blas CB, Moerman T, Huynh-Thu VA, Imrichova H, Hulselmans G, et al. SCENIC: single-cell regulatory network inference and clustering. Nat Methods. 2017;14(11):1083–6.
Efremova M, Vento-Tormo M, Teichmann SA, Vento-Tormo R. CellPhoneDB: inferring cell-cell communication from combined expression of multi-subunit ligand-receptor complexes. Nat Protoc. 2020;15(4):1484–506.
Nagai JS, Leimkuhler NB, Schaub MT, Schneider RK, Costa IG. CrossTalkeR: analysis and visualization of ligand-receptorne tworks. Bioinformatics. 2021;37(22):4263–5.
Ma S, Lu C-C, Yang L-Y, Wang J-J, Wang B-S, Cai H-Q et al. ANXA2 promotes esophageal cancer progression by activating MYC-HIF1A-VEGF axis. Journal of experimental & clinical cancer research: CR. 2018;37(1):183, date = 2018-08.
Ebbing EA, Steins A, Fessler E, Stathi P, Lesterhuis WJ, Krishnadath KK, et al. Esophageal Adenocarcinoma Cells and Xenograft Tumors Exposed to Erb-b2 Receptor Tyrosine Kinase 2 and 3 Inhibitors Activate Transforming Growth Factor Beta Signaling, Which Induces Epithelial to Mesenchymal Transition. Gastroenterology. 2017;153(1):63–76. .e14, date = 2017-07.
Wang Y, Jing Y, Ding L, Zhang X, Song Y, Chen S, et al. Epiregulin reprograms cancer-associated fibroblasts and facilitates oral squamous cell carcinoma invasion via JAK2-STAT3 pathway. J Exp Clin Cancer Res. 2019;38(1):274.
Orian-Rousseau V, Chen L, Sleeman JP, Herrlich P, Ponta H. CD44 is required for two consecutive steps in HGF/c-Met signaling. Genes Dev. 2002;16(23):3074–86.
Meurette O, Mehlen P. Notch Signaling in the Tumor Microenvironment. Cancer Cell. 2018;34(4):536–48. date = 2018-10.
Fleetwood AJ, Achuthan A, Schultz H, Nansen A, Almholt K, Usher P, et al. Urokinase plasminogen activator is a central regulator of macrophage three-dimensional invasion, matrix degradation, and adhesion. J Immunol. 2014;192(8):3540–7.
Mahmood N, Mihalcioiu C, Rabbani SA. Multifaceted Role of the Urokinase-Type Plasminogen Activator (uPA) and Its Receptor (uPAR): Diagnostic, Prognostic, and Therapeutic Applications. Front Oncol. 2018;8:24.
Lee HT, Liu SP, Lin CH, Lee SW, Hsu CY, Sytwu HK, et al. A Crucial Role of CXCL14 for Promoting Regulatory T Cells Activation in Stroke. Theranostics. 2017;7(4):855–75.
Li GH, Arora PD, Chen Y, McCulloch CA, Liu P. Multifunctional roles of gelsolin in health and diseases. Med Res Rev. 2012;32(5):999–1025.
Frankell AM, Jammula S, Li X, Contino G, Killcoyne S, Abbas S, et al. The landscape of selection in 551 esophageal adenocarcinomas defines genomic biomarkers for the clinic. Nat Genet. 2019;51(3):506–16.
Kanisicak O, Khalil H, Ivey MJ, Karch J, Maliken BD, Correll RN, et al. Genetic lineage tracing defines myofibroblast origin and function in the injured heart. Nat Commun. 2016;7:12260.
Wei Y, Cai J, Zhu R, Xu K, Li H, Li J. Function and therapeutic potential of transient receptor potential ankyrin 1 in fibrosis. Front Pharmacol. 2022;13:1014041.
Hatano N, Itoh Y, Suzuki H, Muraki Y, Hayashi H, Onozaki K, et al. Hypoxia-inducible factor-1alpha (HIF1alpha) switches on transient receptor potential ankyrin repeat 1 (TRPA1) gene expression via a hypoxia response element-like motif to modulate cytokine release. J Biol Chem. 2012;287(38):31962–72.
Croft W, Evans RPT, Pearce H, Elshafie M, Griffiths EA, Moss P. The single cell transcriptional landscape of esophageal adenocarcinoma and its modulation by neoadjuvant chemotherapy. Mol Cancer. 2022;21(1):200.
Vancauwenberghe E, Noyer L, Derouiche S, Lemonnier L, Gosset P, Sadofsky LR, et al. Activation of mutated TRPA1 ion channel by resveratrol in human prostate cancer associated fibroblasts (CAF). Mol Carcinog. 2017;56(8):1851–67.
Liu T, Zhou L, Li D, Andl T, Zhang Y. Cancer-Associated Fibroblasts Build and Secure the Tumor Microenvironment. Front Cell Dev Biol. 2019;7:60.
Vienot A, Pallandre JR, Renaude E, Viot J, Bouard A, Spehner L, et al. Chemokine switch regulated by TGF-beta1 in cancer-associated fibroblast subsets determines the efficacy of chemo-immunotherapy. Oncoimmunology. 2022;11(1):2144669.
Azuma T, Witke W, Stossel TP, Hartwig JH, Kwiatkowski DJ. Gelsolin is a downstream effector of rac for fibroblast motility. EMBO J. 1998;17(5):1362–70.
Chan MW, El Sayegh TY, Arora PD, Laschinger CA, Overall CM, Morrison C, et al. Regulation of intercellular adhesion strength in fibroblasts. J Biol Chem. 2004;279(39):41047–57.
Davis J, Burr AR, Davis GF, Birnbaumer L, Molkentin JD. A TRPC6-dependent pathway for myofibroblast transdifferentiation and wound healing in vivo. Dev Cell. 2012;23(4):705–15.
Lombardi AA, Gibb AA, Arif E, Kolmetzky DW, Tomar D, Luongo TS, et al. Mitochondrial calcium exchange links metabolism with the epigenome to control cellular differentiation. Nat Commun. 2019;10(1):4509.
Leung CS, Yeung TL, Yip KP, Wong KK, Ho SY, Mangala LS, et al. Cancer-associated fibroblasts regulate endothelial adhesion protein LPP to promote ovarian cancer chemoresistance. J Clin Invest. 2018;128(2):589–606.

Table 1 is available in the Supplementary Files section

Supplementary Figure Legends

Figure S1. Study overview (A) Overview of EAC patient cohort across the three treatment groups (B) Overview of quality control methods during bioinformatics analysis

Figure S2. Major TME components in EAC patient’s tissue. (A) Heatmap of quantified IHC data for GSN, CD3+ tumoral (T-cells), CD3+ stromal (T-cells), EPCAM (epithelial cancer cells), aSMA and POSTN staining (fibroblasts). Patient cases (n=19) are stratified by CAF low expression (low staining for POSTN/aSMA, cluster 1) and CAF high expression (high staining for POSTN/aSMA, cluster 2). Two grey squares depict missing cases for GSN staining. (B) Examples of IHC staining patterns for these markers and a corresponding H&E in a representative patient from cluster 1 and cluster 2. The corresponding digital pathology annotated image to the right of each IHC stain shows the overlay for stain segmentation and quantification on that region of tissue, with positive tissue areas (or positive cells, for CD3 staining) shown in red, and negative areas (or negative cells, for CD3 staining) shown in blue or uncolored. H&E images are annotated with white dashed lines to label major tumor regions (T) and stroma (S). Scale bars – 500um.

Figure S3. UMAP projections for the broad TME groups (tumor, immune and stroma). (A) UMAP of cancer cell subset by patient origin (B) UMAP of immune cell subset patient origin (C) UMAP of stromal cell subset by patient origin.

Figure S4. Cancer cells exhibit the greatest transcriptomic and genomic complexity. Inferred chromosomal landscape of copy number aberrations (CNAs) in cells from the cancer cell clusters compared to the NDCs as reference by inferCNV analysis in the chemo-naïve (A) chemoradiotherapy treated (B) and chemotherapy treated (C) cohorts.

Figure S5. Trajectory analysis of the fibroblast clusters. (A) Looped trajectory with starting point at fibroblasts with universal phenotype (NOF2) and end point at myofibroblast by Monocle 3. Each dot indicates single cells colored by their cluster (B) Feature plot of universal fibroblast (PI16 and CD34) and myofibroblast gene expression (ACTA2, POSTN, PDGFRA).

Figure S6. Validation of prognostic genes in EAC. (A) Kaplan-Meier survival curve for BCLAF1 gene in TCGA EAC cases (B) HFFF2 fibroblasts treated with TGF-β for 72h show CAF differentiation compared to vehicle (C) Relative expression of mRNA of myofibroblast genes ACTA2 and POSTN following CAF differentiation by TaqMan assay at 6h, 24h, 48h and 72h compared to vehicle (n=3 replicate experiments) (D) Relative expression of mRNA of ZEB1 and ZEB2 (EMT regulation), ATF1 and GSN (cytoskeleton factors) following CAF differentiation by TaqMan assay at 6h, 24h, 48h and 72h compared to vehicle (n=3 replicate experiments) (E) Protein expression of GSN, αSMA and GAPDH a in TGF-β treated HFFF2 cells and vehicle quantified by Western Blotting and densitometry in biological triplicates (unpaired two-tailed t-test).

Figure S7. Immune subpopulations in EAC. (A) UMAP of T cell populations and Heatmap of selected lineage markers (B) UMAP of B cell populations and Heatmap of selected lineage markers (C) UMAP of macrophage populations and Heatmap of selected lineage markers (D) UMAP of Mast cell populations and Heatmap of selected lineage markers.

Figure S8. Topology of the EAC CAF communication network. (A) Communication network pot showing interactions between CAF clusters and other cell populations, thickness of line corresponds to the number and significance of interactions by Pagerank Log Ratio (B) Heatmap illustrating role of CAFs (Influencer or Mediator) in the tumor-derived communication network on KEGG-pathways (C) Table of most significantly deregulated KEGG-pathways associated with Pagerank in the tumor-derived communication network.

Figure S9. Gelsolin expression patterns across EAC tissue specimens. (A) Histologically normal esophageal stratified squamous epithelium with scattered GSN+ cells in the basal and suprabasal epithelial layers (B) EAC with heterogeneous membranous and cytoplasmic positive GSN expression in tumor epithelium (C) Lymphoid aggregate within EAC showing membranous and cytoplasmic positive GSN expression in scattered cells. (D) EAC with negative GSN expression in tumor epithelium, and positive GSN expression in stromal fibroblasts surrounding it (E) EAC with negative GSN expression in tumor epithelium, positive GSN expression in stromal fibroblasts, and GSN positive cells infiltrating within tumor epithelium (F) EAC with negative GSN expression in tumor epithelium and positive GSN expression in cells localized to intraepithelial lumina. Scale bars - 250um.

Figure S10. Comparision to fibroblast phenotypes in Croft et al. (A) Violin plots of CAF1-5 marker genes and selected marker genes from the Croft et al study, applied across both datasets. (B) Dot plot of CAF5-associated growth factor genes, applied across both datasets. Where color shows average expression and size of dot indicate % of cells expressing. Boxes highlight marker gene conservation and differences across studies.

Supplementary Table Titles

Table S1. Clinical characteristics for each EAC patient.

Table S2: Composition of each cell population by sample type across tumor and normal derived cells.

Table S3: Composition of each cell population by sample type.

Table S4: Prognostic significance of the weighted gene expression model in TCGA pan-cancer datasets.

No competing interests reported.

SupplementaryData.zip

Download PDF

Version 1

posted

You are reading this latest preprint version

Targetable fibroblast phenotypes and EMT malignant cell states cooperate to promote tumor progression in esophageal adenocarcinoma

Status:

Version 1

Abstract

Figures

Background

Methods

Patient cohort

Immunohistochemistry

Drop-seq

Unsupervised Clustering And Downstream Analysis

Cell Culture And Analysis

Statistical analysis

Results

The cellular microenvironment of EAC tumors

Cancer cell EMT signatures associate with the Fibroblast-high subtype

Steady state and activated fibroblast phenotypes in EAC

Trajectory analysis reveals co-expression and gene-regulatory networks associated with myoCAF phenotypes

myoCAFs are important in the EAC cellular interaction network

Stromal cell populations with reduced gelsolin expression surround tumor nests

A combined EMT and myofibroblast signature is prognostic across cancers

Discussion

Conclusions

Abbreviations

Declarations

References

Table 1

Supplemental Information Titles And Legends

Additional Declarations

Supplementary Files

Status:

Version 1