HNF4A Defines Molecular Subtypes and Vulnerability to TGFβ-pathway Targeted Therapies in Cancers of the Distal Esophagus

The transforming growth factor beta (TGFβ) pathway exhibits pro-tumorigenic effects across multiple malignancies, via its impact on tumor cells, microenvironment, and/or the immune system^{1, 2}. There is therefore a growing impetus to target the TGFβ signaling axis across malignancies^{1, 2}. While early-phase/dose-defining trials with TGFβR1 (ALK5) inhibitor in glioma³ and TGFβ2 oligo therapy in anaplastic astrocytoma⁴ do show improved outcomes, albeit in a small subset of patients, the molecular determinants that dictate response to TGFβ-pathway targeted therapies however remain elusive; underscoring a broader unmet need in precision oncology. Furthermore, lack of such robust response biomarkers will continue to pose challenges, undermining ongoing/future clinical trials with TGFβ-pathway targeted therapies². Identifying pertinent molecular- and tumor-contexts where such treatments could be most beneficial will not only improve treatment paradigms and clinical outcomes, but also minimize unwarranted therapy-associated burdens.

We recently reported that esophageal adenocarcinomas (EAC) are characterized by hyperactivation of the TGFβ pathway early on in their development, unlike other gastrointestinal (GI) malignancies⁵. These distal-esophageal tumors in particular are alarmingly increasing in incidence, even among younger individuals under 50 years of age⁶. Current treatment options (targeted or otherwise) for EAC are extremely limited and/or ineffective, with vast majority becoming refractory to standard chemoradiation therapies; overall leading to a dismal 5-year survival⁷. Consequently, we posited whether TGFβ-pathway targeted therapies could be beneficial in EACs, and accordingly evaluated the anti-tumor efficacy of a new generation ALK5 inhibitor (Vactosertib)⁸ in multiple in-vivo preclinical models of EAC.

Of the EAC models tested, we observed ALK5i to effectively suppress or even abrogate the growth of a subset of EAC tumors (Figure 1A). ALK5i treatment nonetheless effectively abrogated Smad2/3 phosphorylation (p-Smad), a critical mediator of TGFβ signaling, across all the EAC tumor models in-vivo (Figure 1A); suggesting that sensitivity to TGFβ pathway inhibition is not likely due to differences in ALK5i-pharmacodynamics or in TGFβ receptor activities amongst the EAC models.

Figure 1: HNF4A mediates ALK5i resistance in EAC and molecular subtypes of EAC.

(A) Tumor xenografts from indicated EAC cell lines were established subcutaneously in immune-compromised mice, and randomized to Vactosertib (ALK5i;100 mg/kg) or vehicle treatment via oral gavage, once daily for 5 consecutive days and 2-day off-treatment. Y-axis in the line graph depicts the average fold-change (FC) in tumor volume over time (x-axis) in respective treatment groups, normalized to the first day of treatment (time point 0 on x-axis). Data are plotted as mean ± SEM, obtained from at least 8 tumors per ALK5i or vehicle arm. *** (P<0.0005) indicates significant differences in tumor volume at the final time point between ALK5i- vs vehicle-treatment groups, estimated using a Student’s t-test. Western blot images depict the levels of total and p-Smad proteins in representative tumor xenografts, obtained 1-week after treatment initiation, from respective EAC models. β-actin was used as an endogenous loading control in the Western blots. EACs (EsoAd1, FLO-1) that were highly sensitive to ALK5i-mediated growth inhibition were shown in the top row, along with photographs of excised tumors from ALK5i and vehicle treated mice at the end of the study. SKGT4 showed modest sensitivity to ALK5i, whereas CW1474, OE19, and Eso26 were resistant to ALK5i (bottom row). Average body weights (B. wt) of mice at the start and end of treatments were provided for the sensitive EsoAd1 and FLO-1 xenografts to demonstrate no apparent adverse effects from ALK5i treatment in-vivo. (B) Western blot analyses demonstrating relative HNF4A protein expression, and the green bars above depict the degree of sensitivity (on an arbitrary 1–5 scale) to ALK5i-induced growth suppression across EAC models. Note the striking correlation of HNF4A RNA/protein expression with ALK5i sensitivity. (C) Tumor xenografts from EsoAd1 EAC cells, stably-reconstituted with HNF4A (+HNF4A) or isogenic stuffer control (st-Control), were established in immune-deficient mice and treated with Vactosertib (ALK5i) or vehicle, and growth kinetics measured over time as in Figure 1A. Data are plotted as mean ± SEM, obtained from at least 8 tumors per ALK5i or vehicle arm. *** (P<0.0005) indicates significant differences in tumor volume at the final time point between ALK5i- vs vehicle-treatment groups, estimated using a Student’s t-test. Also shown are photographs of excised xenograft tumors obtained at the end of the study. Note that restoration of HNF4A completely abrogates sensitivity to ALK5i in EsoAd1 cells. (Bottom) representative images from IHC analysis depicting HNF4A protein levels in respective xenograft tumors, obtained 1-week after treatment initiation. Scale bar, 75 μm. Western blot images to the right show protein levels of HNF4A in control (st-Con) and in HNF4A-reconstituted (+HNF4A) EsoAd1 tumor-xenografts derived from immune-deficient mice treated with ALK5i or vehicle. Bar graphs below show relative fold-change (FC) in p-Smad protein levels, normalized to vehicle control treatments, in respective control and +HNF4A groups. (D) (Top) HNF4A-reconstituted EsoAd1 cells showing significant increase in HNF4A transcription-factor (TF) activity, and (bottom) enrichment of HNF4A motif and footprint in accessible chromatin regions (bp; basepair), when compared to isogenic control (st-Con) EsoAd1 cells. (E) Pie charts demonstrating that 25% of HNF4A-induced genes (≥2-fold increased expression in HNF4A-reconstituted vs. st-Con EsoAd1 cells) are associated with regions exhibiting enhanced chromatin accessibility upon HNF4A induction; significantly different from the 2% increase in chromatin accessibility for genes unaltered by HNF4A. (F) RNAseq-based analysis showing genes that are significantly and positively correlated with HNF4A expression in primary EAC tumors. (G) (Top row) Western blot analysis of representative HNF4A-induced genes (VIL1, EPS8L3) showing significant induction of respective proteins in HNF4A-reconstituted EsoAd1, and in ALK5i-resistant parental EAC cells (CW1474, OE19, ESO26). β-actin was included as a loading control in the Western blots. (Bottom row) ATACseq-based Tag density plots (UCSC Genome Browser) show enhanced peaks in respective VIL1 and EPS8L3 gene promoters (green boxes), upon HNF4A induction. (H) Heatmaps from unsupervised hierarchical clustering of HNF4A gene-targets in primary EACs, derived from in-house (CWRU) and TCGA RNAseq datasets, revealing two major molecular subclasses/clusters of EAC. Note the significant (***; P<0.0005) differences in HNF4A gene expression (Y-axis; log2 FPKM values) between the two EAC clusters in CWRU and TCGA RNAseq datasets (see Methods). (I) (Top) IHC images demonstrating nuclear HNF4A protein (brown staining) in representative HNF4A-negative, -low/moderate, and -high EACs. Shown below are the number (percentage) of EAC samples with negative, low-moderate, and high protein expression of HNF4A by IHC. (Below) IHC analysis depicting protein expression of HNF4A-targets, VIL1 and EPS8L3, in the respective tumor sections from above. Scale bar, 275 μm.

Our further assessments of aberrations in specific EAC-associated driver genes, such as P53, SMAD4, ERBB2, or our RNAseq-based assessments of TGFβ response signatures showed no apparent associations with ALK5i responses in EACs (Supplementary Figure 1). We therefore used an agnostic approach to identify molecular factors associated with ALK5i response. Intriguingly, HNF4A, a GI lineage defining transcription factor implicated in BE and EAC^{9, 10}, emerged as the top candidate with markedly increased expression in ALK5i-resistant compared to ALK5i-sensitive EACs (Figure 1B, Supplementary Figure 1). We next asked whether HNF4A could also functionally regulate ALK5i response in EACs. Since cancer cells with high baseline HNF4A expression are highly sensitive to HNF4A depletion⁹ (and our own observations), we conversely assessed whether induction of HNF4A would impart resistance to ALK5i treatment in otherwise HNF4A-low/null (ALK5i-sensitive) EAC xenografts. Indeed, stable reconstitution of HNF4A completely abrogated the anti-tumor efficacy of ALK5i in-vivo, again with no differences in p-Smad levels between control and HNF4A-reconstituted EAC tumors (Figure 1C). These findings provocatively suggest HNF4A as playing a functional role in conferring inherent resistance to ALK5i therapy.

Although HNF4A is a well-recognized GI lineage transcription factor, its molecular effects in the EAC context remains unknown. We therefore sought to understand if/how HNF4A impacts the molecular programs in EAC, as this could provide insights into its function in this malignancy. We performed integrative RNAseq and ATACseq analyses in HNF4A-low/negative EAC (EsoAd1) cells, with or without stable reconstitution of HNF4A. Ensuring that ectopically reconstituted HNF4A is functionally active, we observed a significant enrichment in HNF4A transcriptional activity and genomic footprint in EAC cells (Figure 1D). Notably, we found HNF4A-induced genes were much more likely to be within HNF4A-induced open chromatin regions, compared to genes that were unaltered by HNF4A (P=0.0001) (Figure 1E). Gene Ontology analysis of HNF4A-induced genes revealed a significant enrichment of digestion-associated biologic process (P=8.65E-08), consistent with its role in promoting GI cell fate. Furthermore, ~40% of HNF4A-induced genes showed significant positive correlation with HNF4A expression in primary EAC tumors (Figure 1F, and Supplementary Figure 2). Subsequent Western blot analyses of representative HNF4A gene-targets, VIL1 and EPS8L3 (intestinal epithelial/differentiation markers), confirmed a marked increase in protein expression along with enriched ATAC peaks in respective gene-promoters in EAC cells (Figure 1G). In line with this, we found both VIL1 and EPS8L3 to be expressed selectively in HNF4A-high (ALK5i-resistant) parental EAC lines (Figure 1G). Collectively, these findings strongly suggest that baseline HNF4A expression confers a digestive/intestinal-type molecular phenotype in EACs with potential implications in disease pathogenesis.

To further understand how HNF4A could be impacting ALK5i response on a molecular level, we performed RNAseq in HNF4A-reconstituted vs. isogenic control EAC cells treated with ALK5i. Pathway and protein-protein interaction enrichment analyses revealed significant (P<0.05) alterations in interleukin-18/matrix-metalloprotein/inflammatory sub-network between HNF4A-positive vs. -negative isogenic EAC cells treated with ALK5i (Supplementary Figure 2). Orthogonal qPCR-based assessment of a representative gene of this network, TNFAIP3 (a TNF alpha induced protein and an upstream regulator of IL-18), showed HNF4A to abrogate ALK5i-mediated suppression of TNFAIP3 in isogenic, as well as in HNF4A-high parental EAC cells (Supplementary Figure 2). These studies besides revealing ALK5i response pathways, also offer potential mechanisms by which HNF4A impedes response to ALK5i therapy.

As our findings above implicate HNF4A as a candidate biomarker of ALK5i response in EACs, it is imperative to understand the extent to which primary EAC tumors differ in HNF4A status. Unsupervised clustering using an in-house RNAseq dataset of treatment-naïve EACs (n=48)⁵ revealed two broad and distinct clusters of EACs stratified by differences in the expression of HNF4A target-genes (Figure 1H). Subsequent supervised analyses showed a significant (P<0.0005) difference in HNF4A expression between the respective sample clusters (Figure 1H); findings that were similarly observed in an independent TCGA EAC RNAseq dataset (n=72) (Figure 1H). Additional evaluation of HNF4A protein status using IHC in a third independent in-house cohort of treatment-naïve EAC tumor biopsy tissues (n=44) showed markedly distinct HNF4A protein expression, as well as HNF4A targets (VIL1, EPS8L3), in EAC tumors (Figure 1I). Taken together the RNAseq and protein analyses, we estimate the majority of primary EAC tumors (60%−80%) to exhibit low/moderate or high HNF4A expression, with ~20% of tumors lacking HNF4A. Overall survival probability based on HNF4A expression status however did not significantly differ in any of the three patient cohorts (data not shown). Nonetheless, intriguingly, majority (70%) of HNF4A-negative tumors (based on histologically-defined IHC cohort) were poorly differentiated, as opposed to only 5% of HNF4A-high tumors (Fisher’s Exact P=0.0027), suggesting that HNF4A may further define a histologic subset of EACs. Collectively, these findings underscore the inherent heterogeneity of EAC tumors, defined by HNF4A, and their associated implications for TGFβ-pathway targeted therapies in this disease.

In summary, we identify TGFβ pathway as a therapeutic vulnerability in a highly aggressive GI malignancy. Our findings are timely given the unmet need in identifying effective targeted therapies for these highly-refractory upper GI malignancies¹, and the growing efforts in exploiting this pathway for cancer treatment^{1, 2}. Also, our identification of HNF4A as an intrinsic modulator of ALK5i response has broader clinical implications as a predictive biomarker that can be evaluated in ongoing/future clinical trials, such that evidence-based strategies can be implemented to aid in selecting appropriate tumor contexts and patient cohorts who may derive most benefit from TGFβ-pathway targeted therapies.

Data Transparency Statement:

Pertinent high-throughput sequencing and integrative analytics in cell line models, where applicable, will be deposited in SRA and made freely available to general public following acceptance of the manuscript.

Abbreviations:

EAC

Esophageal adenocarcinoma

ALK5i

ALK5 inhibitor

IHC

Immunohistochemistry

ATACseq

Assay for Transposase-Accessible Chromatin using sequencing

GI

gastrointestinal