HAND1 and BARX1 Act as Transcriptional and Anatomic Determinants of Malignancy in Gastrointestinal Stromal Tumor

Gastrointestinal stromal tumor (GIST) is the most common soft-tissue sarcoma, with population-based annual incidence rates of approximately 1 per 100,000 (1, 2). GIST arises from transformation of interstitial cells of Cajal (ICC) or their cellular progenitors (3), which normally act as pacemaker cells regulating gut motility. In at least 85% of cases, GIST lesions harbor activating mutations in one of the related type III receptor tyrosine kinases (RTK) KIT or PDGFRA, with ligand-independent kinase signaling mediating oncogenesis (4). Other less common genetic variants can also give rise to GIST, such as inactivating mutations in genes encoding components of the succinate dehydrogenase (SDH) complex or neurofibromatosis type 1 (NF1), which have characteristically unique disease presentations and a more indolent clinical course (5, 6).

Following the identification of KIT mutations as drivers of GIST (7), the tyrosine kinase inhibitor (TKI) imatinib was found to inhibit KIT signaling and effectively treat patients with GIST (8, 9). With the advent of imatinib, durable disease control measured in years can be achieved in nearly 80% of patients with metastatic GIST (10). The success of imatinib has led to its regulatory approval for use as adjuvant therapy for primary resected GIST at high risk of recurrence, as well as in neoadjuvant (preoperative) therapy to shrink tumors for which surgery would lead to unacceptable risk or morbidity (11–14). In primary resected GIST, risk of recurrence is defined by characteristics such as anatomic site of origin, size, and mitotic rate, but more reliable predictive markers would be of clinical benefit (15). Aside from GIST mutational subtype, no molecular markers are clinically available to identify those patients most likely to benefit from receiving adjuvant imatinib. Particularly in cases in which tumors are labeled as intermediate risk, when there are borderline assessments of mitotic index, or when neoadjuvant imatinib is administered and confounds mitotic analyses of the surgical specimen, novel molecular markers would be very useful in risk stratification. Furthermore, there are few prognostic tools to guide clinical management in the setting of metastatic disease and to support the rational development of new therapeutics. The lack of clinically useful molecular markers is related to a limited understanding of the biological factors that influence GIST oncogenesis and interindividual variation in the malignant phenotype.

Core transcription factor (TF) circuits establish and maintain normal lineage-restricted cellular development and function throughout biological systems (16, 17). Cancer exploits the core TF networks of precursor cells, either through amplifying or modifying lineage-specific transcriptional programs (18–20). We have previously mapped transcriptionally active chromatin and TF DNA occupancy in GIST and characterized the enhancer and transcriptional landscape of GIST tissue samples and cell lines (21). Using this information, we identified a TF network supporting the GIST gene expression program. Furthermore, through comparative analysis of localized and metastatic tumors, we identified two accessory TFs, HAND1 and BARX1, which are expressed in a pattern specific to disease state. Here, we characterize the transcriptional programs driven by these two accessory TFs, describe clinical, mutational, and functional phenotypes in which they are engaged, use high-dimensional tissue imaging to identify high-risk cellular populations, and explore the anatomic and developmental origins of the programs that characterize GIST pathogenesis. Our findings suggest a new molecular subclassification of GIST that justifies further study to inform basic, translational, and clinical studies.

All cell lines tested negative for Mycoplasma infection on routine surveillance (MycoAlert, Lonza Bioscience). Human embryonic kidney (HEK) 293FT (Thermo Fisher Scientific, catalog No. R70007, RRID: CVCL_6911) and the GIST cell line GIST-T1 (Cosmo Bio, catalog No. PMC-GIST01-COS, RRID:CVCL_4976; KIT mutation in exon 11 Δ560-578) were cultured in DMEM containing 10% FBS, 2 mmol/L l-glutamine, 100 mg/mL penicillin, and 100 mg/mL streptomycin. GIST-T1 cell line identity was confirmed by sequencing KIT exons to confirm the expected mutation. Cell lines were thawed from original or derived stocks and used in the described experiments within approximately 3 months. Transfections were performed with X-tremeGene (Roche). Lentiviral production was performed as described previously (22). Briefly, 293FT cells were cotransfected with pMD2.G (Addgene No. 12259), psPAX2 (Addgene No. 12260), and the lentiviral expression plasmid. Viral supernatant was collected at approximately 72 hours and debris removed by centrifugation at 1,000 × g for 5 minutes. Cells were transduced with viral supernatant and polybrene at 8 μg/mL by spinoculation at 680 × g for 60 minutes. For growth over time assays, 15 × 10³ cells were dispensed per well in a 96-well plate and cell count performed approximately twice per week on a Guava easyCyte Flow Cytometer (EMD Millipore).

Cell lines stably expressing a human codon-optimized Streptococcus pyogenes Cas9 (Addgene No. 73310) were generated by viral transduction. CRISPR single-guide RNAs (sgRNA) targeting HAND1, ETV1, or control constructs have been described previously (21). The BARX1 lentiviral expression vector was synthesized with codon optimization (Twist Bioscience). The Dependency Map (DepMap) portal data was accessed through depmap.org (23), using the CRISPR (Avana) Public 20Q3 release.

GIST-T1 cells were transduced with the indicated sgRNAs and incubated for 5 days before collection. Total RNA was isolated using an RNeasy Plus Kit (Qiagen), and concentration measured by Nanodrop (Thermo Fisher Scientific) and quality by Bioanalyzer (Agilent). Library preparation was performed using the NEBNext Ultra II nonstranded library prep kit (New England Biolabs). Paired-end 150-bp sequencing was performed on a NovaSeq 6000 (Illumina). All GIST tumor sequencing data have been published previously (21, 24–26). RNA sequencing (RNA-seq) data were aligned to hg19 using STAR (27) with expression quantification using Cufflinks (RRID:SCR_014597; ref. 28) to generate gene expression values in fragments per kilobase of transcript per million mapped reads (FPKM) units. EdgeR (RRID:SCR_012802) was used for differential expression analysis of tumor sequencing data (29). Merging of chromatin immunoprecipitation and differential expression data was performed using binding and expression target analysis (BETA, RRID:SCR_005396; ref. 30). Gene set enrichment analysis (GSEA, RRID:SCR_003199; ref. 31) was performed using Hallmark gene lists in the Molecular Signatures Database (software.broadinstitute.org/gsea/).

All chromatin immunoprecipitation with sequencing (ChIP-seq) and assay for transposase-accessible chromatin using sequencing (ATAC-seq) data were aligned to the human reference genome assembly hg19 using Bowtie2 (RRID:SCR_005476; ref. 32). Normalized read density was calculated using Bamliquidator (version 1.0) read density calculator. Aligned reads were extended by 200 bp and the density of reads per bp was calculated. In each region, the density of reads was normalized to the total number of million mapped reads, generating read density in units of reads per million mapped reads per bp (rpm/bp). Individual ChIP-seq track displays were generated using bamplot (github.com/linlabbcm).

Formalin-fixed and paraffin-embedded (FFPE) archival GIST samples were obtained under an institutional review board (IRB)-approved protocol at Brigham and Women's Hospital/Dana-Farber Cancer Institute (Boston, MA) or Stanford University (Stanford, CA). The majority of tumor samples were evaluated as components of a tissue microarray. Where available, pathologic and molecular features of tumors and clinical outcomes were compiled. For relapse-free survival assessments, all included patients received care before the regulatory approval of imatinib as a standard adjuvant therapy option. Normal gastrointestinal tissues were retrieved from the archives of Brigham and Women's Hospital (Boston, MA) with IRB approval as part of a discarded/excess tissue protocol.

FFPE sections were deparaffinized, dehydrated, and endogenous peroxidase activity blocked. Antigen retrieval was performed in a pressure cooker in citrate buffer (S1699, Dako) at 123°C at 15 PSI for 45 seconds. Slides were incubated with HAND1 antibody (1:300, OriGene, catalog No. TA502671, RRID:AB_11125431) or BARX1 antibody (1:100, Atlas Antibodies catalog No. HPA055858, RRID:AB_2682947) for 45 minutes, washed, then incubated with either Labeled Polymer-HRP anti-mouse secondary antibody (K4007, Dako) for HAND1 or Post Primary (Leica Novolink) followed by Novolink Polymer Detection System (RE7150-K, Leica) for BARX1. Slides were then incubated with Dako DAB+ solution (K3468, Dako) and counterstained with hematoxylin. Additional antibodies used for IHC include PDGFRA (1:200, Cell Signaling Technology, catalog No. 5241, RRID:AB_10692773), DOG-1 (1:50, Leica Biosystems, catalog No. NCL-L-DOG-1, RRID:AB_10555293), SDHB (1:250, Abcam, catalog No. ab14714, RRID:AB_301432), c-KIT (1:250, Agilent, catalog No. A4502, RRID:AB_2335702), PLAGL1 (1:400, Thermo Fisher Scientific, catalog No. MA5-31853, RRID:AB_2787476), and HOXC10 (1:300, Thermo Fisher Scientific, catalog No. PA5-31078, RRID:AB_2548552). For scoring cases, we used a visual scoring system including relative intensity of signal (0–3 scale) and the percent of cells positive (in 5% increments). A positive case was defined as 1+ scoring in 5% or more of tumor cells in two of three cores.

c-KIT (brown)/HAND1 (red) double IHC was performed by incubating sections with c-KIT antibody (1:150, A4502, Dako) for 45 minutes, followed by Labeled Polymer-HRP anti-rabbit secondary antibody (K4011, Dako) and then DAKO DAB+ solution. Antigen retrieval was then performed in a pressure cooker in citrate buffer (S1699, Dako) at 123°C at 15 PSI for 45 seconds followed by incubation with HAND1 at (1:300, TA502671, Origene) and then incubation with AP Polymer anti-mouse secondary antibody (Ultra Vision LP, TL-125-AP, Thermo Fisher Scientific), followed by incubation with Alkaline Phosphatase (Red Substrate Kit, SK-5105, Vector). Slides were counterstained with Hematoxylin. c-KIT (red)/BARX1 (brown) double IHC was performed by incubating sections with c-KIT antibody (1:150, A4512, Dako) for 45 minutes, followed by AP Polymer anti-mouse secondary antibody (Ultra Vision LP, TL-125-AP, Thermo Fisher Scientific) and then incubated with Alkaline Phosphatase (Red Substrate Kit, SK-5105, Vector). Antigen retrieval was then performed in a pressure cooker in citrate buffer (S1699, Dako) at 123°C at 15 PSI for 45 seconds followed by incubation with Novolink Polymer Detection Protein Block (RE7150-K, Leica), BARX1 at 1:100 (HPA055858, Sigma) for 45 minutes, followed by Post Primary and then Novolink Polymer Detection System (RE7150-K, Leica) and developed with Dako DAB+ solution (K3468, Dako) and counterstained with hematoxylin. Differentiation of KIT-positive ICC from mast cells in normal gastrointestinal tissues was supported by cell morphology and tissue localization, with ICC favored in spindled cells with ovoid nuclei and prominent cytoplasmic processes, which are readily identifiable along the myenteric plexus in most specimens, whereas mast cells exhibit round nuclei and cell bodies with no distinct cytoplasmic processes.

Tissue-based cyclic immunofluorescence (t-CyCIF) consisted of iterative cycles of antibody incubation, imaging, and fluorophore inactivation, and was performed on FFPE specimens as described previously (33, 34) using antibodies listed (Supplementary Table S2). c-KIT signal using this protocol was nonspecific and inadequate to detect differences between GIST subsets. Slides were initially prepared as described above for HAND1 and BARX1 IHC. Image acquisition was performed with a RareCyte CyteFinder Slide Scanning Fluorescence Microscope. Photobleaching was performed with a solution of 4.5% H₂O₂ and 20 mmol/L NaOH in 1× PBS and incubation under a light-emitting diode for 2 hours at room temperature. Image processing was performed as previously described to define cell states and identify mesoscale neighborhood information.

Cells were lysed in RIPA buffer containing protease inhibitor cocktail (Roche) and centrifuged at 14,000 × g for 10 minutes to remove genomic DNA and debris. Protein concentrations were determined using a bicinchoninic acid–based assay (Pierce Biotechnology). Protein samples were subjected to SDS-PAGE and Western blotting with the following antibodies: BARX1 (1:500, Thermo Fisher Scientific, catalog No. PA5-68362, RRID:AB_2688478) and ERK (1:2,000; Cell Signaling Technology, catalog No. 9107, RRID:AB_10695739). Western blots were probed with anti-mouse or anti-rabbit secondary antibodies and detected using the Odyssey CLx infrared imaging system (LI-COR Biosciences). Immunoblots shown are representative of at least three independent experiments.

Center values, error bars, P value cutoffs, number of replicates, and statistical tests are identified in the corresponding figure legends. Replicates represent separate tumors where indicated, or in experiments using the same cell line the replicates represent separate treatments. Error bars are shown for all data points with replicates as a measure of variation within each group. Box plot elements include the center line representing the median, the box representing the top and bottom quartiles, and the whiskers plotted using the Tukey method. Samples sizes were not predetermined. The investigator scoring IHC cases was blinded to clinical information and outcome assessment during data acquisition. Kaplan–Meier analysis of relapse-free survival was calculated from the date of initial diagnosis to the date of first metastasis, relapse, last follow-up or death. Progression-free survival was calculated from the date of imatinib initiation as first-line therapy for metastatic GIST until documented disease progression. HRs and P values were calculated using the log-rank test.

Novel RNA-seq data are available through the Gene Expression Omnibus Publication Reference ID GSE151323. Additional RNA-seq and ChIP-seq datasets analyzed in this study include GSE71119 (24), GSE95864 (21), GSE107447 (25), and PRJNA521803 (26).

Previously, we used multiple methods to identify and characterize a core group of TFs in GIST tissue samples and cell lines; these methods included ChIP-seq, ATAC-seq, and RNA-seq. These core TFs cooperatively interact in a regulatory network that is shared across GIST mutational subtypes and clinical phenotypes (21). Accessory TFs HAND1 and BARX1, however, were expressed in disease-state specific patterns predictive of metastasis-free survival. To better understand the distinct transcriptional programs driven by these two accessory TFs and their relationship to clinical factors, we characterized gene expression datasets from 35 GIST clinical samples for which clinical annotation was available. We used differential expression analysis of two clinically distinct subgroups to compare transcription in RTK-driven localized and never-recurrent samples (the “No Recurrence” group) and localized and later-recurrent or metastatic GIST (the “Metastatic/Recurrent” group; Fig. 1A; Supplementary Table S1). We found that the overall gene expression program was highly similar across all tumors regardless of clinical status, with <5% of all expressed genes (n = 10,000) exhibiting differential expression between the two groups. Among all differentially expressed transcripts, only the TFs BARX1 and PLAGL1 were enriched in “No Recurrence” GIST, whereas HAND1 and HOXC10 were the only TFs enriched in “Metastatic/Recurrent” GIST. BARX1 and HAND1 showed the greatest differences in expression between groups, with >100-fold enrichment of each transcript in their respective group. Moreover, HAND1 expression was commonly and exclusively found in KIT-mutant tumors in the “Metastatic/Recurrent” group. In contrast, BARX1 was expressed at high levels in KIT-mutant GIST in the “No Recurrence” localized group as well as in PDGFRA-mutant GIST and SDH-deficient GIST (Fig. 1B).

Recently, an independent cohort of RNA-seq data from 75 GIST specimens was reported (35). Combining this publicly available dataset with our clinically annotated samples, we found a strong anticorrelation between HAND1 and BARX1 expression values, with high levels of expression of one or the other TF observed in all but a subset of samples (total n = 110; n = 15 with FPKM <50 for both TFs; Fig. 1C). Dimensionality reduction using principal component analysis (PCA) of the public RNA-seq data from all samples that express HAND1 (n = 21) or BARX1 (n = 39) revealed that tumor clustering was correlated with accessory TF expression (Fig. 1D). We then used GSEA to identify the gene expression programs associated with these disease state-specific accessory TFs (i.e., analyzing “No Recurrence” vs. “Metastatic/Recurrent” groups in the clinically annotated cohort, and BARX1+ vs. HAND1+ GIST in the independent published cohort). The “Metastatic/Recurrent” and the HAND1+ groups showed significant enrichment in Hallmark signatures including G₂–M-phase checkpoint and MTORC1 signaling gene sets, whereas the “No Recurrence” and BARX1+ groups both showed enrichment in the epithelial–mesenchymal transition (EMT) gene set (Fig. 1E). The G₂–M-phase checkpoint was among the most enriched gene sets in the HAND1-expressing groups, suggestive of higher proliferation and a more aggressive phenotype (Fig. 1F–G). Taken together, data from these independent cohorts indicate that unique gene expression programs are associated with accessory TF expression, and that HAND1-positive tumors engage cell growth-associated signal transduction pathways and proliferation programs.

To explore the mechanism behind the association of HAND1 with more aggressive disease, we used a CRISPR-based system to inactivate HAND1, or the core GIST TF ETV1 (36) as comparator, in the KIT-dependent GIST cell line GIST-T1. Loss of either HAND1 (sgHAND1) or ETV1 (sgETV1) decreased GIST cell proliferation in a 21-day assay as compared with a luciferase guide RNA control (sgLuc; Fig. 2A). We performed RNA-seq on cells collected prior to the onset of proliferative effects (i.e., 5 days of sgRNA exposure). Compared with control (sgLuc), cells treated with sgHAND1 and sgETV1 showed global changes in gene expression (Fig. 2B; Supplementary Fig. S1A). Following acute loss of either HAND1 or ETV1, Hallmark gene sets were enriched for the EMT signature; in comparison, cells transduced with sgLuc were enriched for MTORC1 signaling (Fig. 2C–E). No significant differences in the G₂–M-phase checkpoint gene set were noted at this early time point. To combine data on genes differentially expressed because of loss of HAND1 with known genomic regions of HAND1 binding identified by ChIP-seq, we used BETA (21, 30). BETA showed that HAND1 loss of function was responsible both for significant downregulation or upregulation of HAND1-associated genes (Fig. 2F). A waterfall plot of the significantly altered genes following sgHAND1 treatment showed enrichment for members of the core GIST TF network among both negatively and positively regulated genes, as well as other GIST-associated genes including KIT and negative regulators of KIT signaling (Fig. 2G). BETA analysis of sgETV1 transduced cells revealed similar global dysregulation of ETV1-regulated genes, although changes were less pronounced as compared with sgHAND1 (Supplementary Fig. S1B).

Several genes exhibited significantly different changes in gene expression with sgETV1 or sgHAND1 exposure. As expected, levels of ETV1 mRNA in the sgETV1 condition and HAND1 mRNA in the sgHAND1 condition showed the greatest decreases, which serves as evidence of CRISPR-mediated on-target nonsense-mediated decay (Fig. 2H). We also found that HOXC10, a noncore TF associated with metastatic GIST (Fig. 1A), was decreased only in the sgHAND1 condition, suggesting coregulation of these metastasis-associated TFs. Similarly, FOXF1 (37) and HOXA11 expression levels fell only with HAND1 loss. Both sgHAND1 and sgETV1 exposure led to a significant increase in PITX1 (21) expression, suggesting corepression of this core GIST TF. Notably, the expression of KIT, the primary oncogene in these cells, was significantly decreased by loss of HAND1 but not ETV1, whereas negative regulators of KIT signaling SPRY4 and DUSP6 showed significantly decreased expression with loss of either ETV1 or HAND1 (Fig. 2I). Nonetheless, both TFs showed similar binding at disease relevant loci by ChIP-seq (Fig. 2J; Supplementary Fig. S1C), with HAND1 loss consistently having a greater effect on mRNA levels at target genes (Supplementary Fig. S1D). Forced expression of BARX1 in these cells resulted in no observable proliferative phenotype (Supplementary Fig. S1E and S1F). No other cell lines profiled by the DepMap project (23) exhibit dependencies upon either of these accessory TFs, suggesting their unique relevance in GIST (Supplementary Fig. S1G). Taken together, these data support a role for HAND1 in driving an aggressive disease phenotype by regulating the core GIST TF network, maintaining KIT gene expression and supporting proliferation-associated signal transduction.

To further characterize the expression of HAND1, BARX1, and other disease state–associated TFs (i.e., HOXC10 and PLAGL1) in GIST tissues resected from patients, we developed and optimized IHC-based assays for each of these proteins and characterized their expression in a cohort of 87 tumors. We also characterized these tumors using IHC for DOG1, KIT, PDGFRA, and SDHB. We found that HAND1 and HOXC10 expression were restricted to KIT-mutant GIST (Fig. 3A). BARX1 and PLAGL1 were expressed in KIT-mutant GIST, albeit in a smaller fraction of the tumors, but both were additionally expressed in PDGFRA-mutant and SDH-deficient GIST, which characteristically exhibit a perinuclear Golgi-like staining pattern for PDGFRA (38) and loss of SDHB (39), respectively. Among KIT-mutant GISTs, only the metastatic gastric tumors—as opposed to the localized gastric GIST—expressed HAND1 and HOXC10, whereas the majority of all small bowel–derived GIST (i.e., localized or metastatic) expressed both high-risk markers (Fig. 3B). In contrast, BARX1 was most commonly expressed in localized gastric GIST. Subdividing tumors by activating KIT mutations, both exon 9 and 11 subtypes exhibited expression of all accessory TFs, whereas metastatic tumors bearing exon 13 or compound mutations exhibited enrichment for HAND1 and HOXC10 with an associated loss of BARX1 (Fig. 3C).

As is expected for TFs, HAND1, HOXC10, BARX1, and PLAGL1 localized to the nucleus (Fig. 3D; Supplementary Fig. S2A). There was clear coexpression of HAND1 and HOXC10, and separately of BARX1 and PLAGL1, with mutual exclusion of expression of these accessory TFs between a majority of tumors (Supplementary Fig. S2B). In tumors expressing either HAND1 or BARX1, >80% of all tumor cells expressed either protein (Supplementary Fig. S2C). HAND1 positivity had a strong positive predictive value for the mutational subtype of tumors bearing KIT mutations, with PDGFRA expression also showing predictive value for PDGFRA-mutant tumors and SDHB loss for SDH-deficient GIST, as expected (Fig. 3E). These results demonstrate the feasibility of assessing accessory TF expression in clinical specimens and support the association of accessory TFs with mutational and clinical subtypes of GIST.

To further explore the association of HAND1 and BARX1 with markers of cellular proliferation in this GIST cohort, we performed multiplexed t-CyCIF (33, 34) of 87 tumors to enable marker expression to be compared at a single-cell level. We simultaneously assayed the tumors for the expression of HAND1, BARX1, PDGFRA, SDHB, Ki-67, and proliferating cell nuclear antigen (PCNA) as well as other markers informative of cell type or state (Supplementary Table S2). SDHB signal was lost from SDH-deficient GIST, whereas PDGFRA expression was significantly elevated in PDGFRA-mutant tumors, as expected (Fig. 4A). There was significantly higher HAND1 and BARX1 signal in KIT-mutant tumors that had scored positive for these markers by IHC (cross-validating the two imaging methods), with correspondingly lower levels of HAND1 and higher levels of BARX1 in PDGFRA-mutant and SDH-deficient tumors (Fig. 4B). We also observed a significant correlation between Ki-67 signal and mitotic index assessed by expert review (Fig. 4C), and also between Ki-67 and HAND1 signal (Fig. 4D). Selected markers of signal transduction and proliferation, including phospho-ERK and p21, were not different between GIST mutational subtypes, whereas the mitochondrial protein COXIV was present at significantly higher levels in SDH-deficient GIST (Supplementary Fig. S4A), consistent with the increased abundance of mitochondria in this GIST subtype (40). Analysis of several immune cell cluster of differentiation (CD) markers, including CD163, CD68, and CD11b, revealed that only a small number of these immune subsets infiltrated the tumor microenvironment, and that there were no significant differences between GIST mutational subtypes (Supplementary Fig. S3B). Whereas Ki-67 and PCNA signal were higher in metastatic tumors in comparison with localized tumors, S100A, a previously reported marker of high-risk GIST (41), was expressed only in small bowel–derived GIST, in both localized and metastatic tumors (Supplementary Fig. S3C).

To study associations in the expression of TFs and proliferative markers at a single-cell level, we segmented CyCIF images and recorded signal intensities in each channel on a per-cell basis. We then used Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction of combined single-cell data from all tumors. We focused on cells that expressed HAND1 and/or BARX1, thereby limiting the analysis to neoplastic cells. BARX1-positive cells clustered together with SDHB-deficient and PDGFRA-high cells, whereas HAND1-positive cells exhibited higher levels of PCNA and Ki-67 (Fig. 4E). Cells expressing S100A represented a subset of the HAND1-positive population (Supplementary Fig. S3D). Ki-67– and PCNA-positive cells were significantly more enriched in the HAND1-positive population (Fig. 4F; Supplementary Table S3).

To determine whether a small population of HAND1-positive cells was present in localized gastric GIST, we analyzed single-cell data from the 15 gastric-primary GIST samples within our cohort (Fig. 3B). UMAP showed coclustering of a small population of HAND1-positive cells (3,273 of the 24,859 tumor cells; 13%); these cells expressed higher levels of Ki-67 and PCNA compared with BARX1-positive cells (Fig. 4G–I). By IHC, five KIT-mutant tumors from this cohort expressed both HAND1 and BARX1, with at least 5% of neoplastic cells being positive for either marker in the same tumor. Single-cell analysis of these tumors showed enrichment of Ki-67 and PCNA in HAND1-positive cells in comparison with BARX1-positive cells (Supplementary Fig. S3E and S3F). CyCIF also identified an additional population of cells that expressed both HAND1 and BARX1, and this subgroup had the highest Ki-67 signal intensity of all tumor cells (Supplementary Fig. S3G and H; Supplementary Table S3). Taken together, these data demonstrate that HAND1 expression is correlated with cellular proliferation, and that small populations of HAND1-positive cells may exist in localized tumors which otherwise appear histologically indolent. These cells may represent subpopulations prone to evolve over time to dominate the tumor population and drive recurrence and metastasis.

We next extended our analysis to a larger cohort of tumors from 437 patients, many with detailed information on tumor genotype (305/437), anatomic origin (384/437), and clinical course (130/437; Supplementary Table S4 details a cohort summary). BARX1 and HAND1 IHC showed that the majority of GISTs arising from the esophagus and stomach expressed BARX1 (84%), whereas the majority of GISTs arising from the small or large intestine or rectum expressed HAND1 (75%; Fig. 5A). Stratifying the cohort by mutational subtype, HAND1 was expressed most frequently in KIT-mutant tumors (53%), in agreement with our findings from the primary cohort (Fig. 5B). One of 34 PDGFRA-mutant tumors expressed HAND1, as did both NF1-associated GISTs (Fig. 5B). HAND1 and BARX1 positivity were seen in all subtypes of KIT mutations, although HAND1 was present in nearly all of the exon 9 tumors that commonly originate from the small bowel (42) and in nearly all GIST with compound mutations arising in metastatic and TKI-resistant disease (43) (Fig. 5C). Although primary GISTs expressed BARX1 and HAND1 at similar rates, the majority (63/84, 75%) of metastatic or recurrent GIST cases were positive for HAND1 (Fig. 5D). The majority of all tumors expressed either HAND1 or BARX1 (343/437, 78%), and a minority were positive (28/437, 6%) or negative (66/437, 15%) for both markers, consistent with the gene expression profiling data (Fig. 1C; Fig. 5E). In cases that expressed both HAND1 and BARX1, the majority (22/28, 79%) expressed predominantly one marker over the other or had low expression of both markers (Supplementary Fig. S4A and S4B). As part of this cohort, we also evaluated seven micro-GISTs, defined as gastric GISTs ≤1 cm in size with benign clinical behavior (44); all were positive for BARX1 and negative for HAND1 (Supplementary Fig. S4C and S4D). For GIST negative for both HAND1 and BARX1, cases were overrepresented by tumors arising from the colon or rectum (Supplementary Table S5). Extragastric GIST positive only for BARX1 were overrepresented in tumors arising from the esophagus or gastroesophageal junction (Supplementary Table S6).

For a subset of these tumors (130/437, 30%), annotation was available to assess clinical outcomes. For macroscopically resected primary tumors, standard risk stratification into high-risk and non-high-risk groups using tumor size, anatomic location, and mitotic index showed a trend, but no statistically significant difference, in relapse-free survival (Fig. 5F; P = 0.1171). In contrast, stratification of patients by the expression of HAND1 or BARX1 showed a significant difference in relapse-free survival (P = 0.0092), with an associated HR of 2.62 [95% confidence interval (CI), 1.14–6.00; Fig. 5G]. A subset of these tumors expressed both or neither marker, but the sample size was too small to permit statistical analysis of these groups. Multivariate analysis of HAND1 expression, BARX1 expression, and recurrence risk level using a Cox proportional hazards model of relapse-free survival showed independent but nonsignificant trends toward shorter relapse-free times (i.e., higher risk of relapse) with HAND1 expression or high-risk level, and longer relapse-free times (i.e., lower risk of relapse) with BARX1 expression; similarly, HAND1 expression and extragastric tumors showed independent nonsignificant trends toward shorter relapse-free survival (Supplementary Fig. S5A and S5B). Finally, in patients diagnosed with metastatic GIST and started on first-line imatinib, progression-free survival was significantly shorter in HAND1-expressing tumors (P = 0.0127), with a HR of 2.322 (95% CI, 1.27–4.25; Fig. 5H); only one of these tumors, in the HAND1-negative group, expressed BARX1, and 46 of 49 tumors had oncogenic KIT mutations, whereas the remaining three cases had no mutational testing available. These data demonstrate that HAND1 and BARX1 expression correlates with clinical outcome and may exceed the current standard-of-care tools for predicting relapse-free survival; these new biomarkers may also offer a tool for predicting progression-free survival in the metastatic setting.

To evaluate accessory TF expression in normal ICCs, which provide insights into the cell of origin of GIST, we performed IHC on normal tissues along the gastrointestinal tract. Using two-marker IHC, we identified ICCs with an antibody recognizing KIT and characterized accessory TF expression using antibodies recognizing either HAND1 or BARX1. We assessed several thousand individual ICCs derived from different anatomic regions spanning from the esophagus to the rectum. ICCs from the stomach showed frequent expression of BARX1, but never HAND1; in contrast, the small intestine ICCs showed frequent expression of HAND1 but not BARX1 (Fig. 6A). The majority of ICCs in these disparate anatomic compartments express HAND1 or BARX1, and in a mutually exclusive pattern (Fig. 6B). In contrast, ICCs of the esophagus, appendix, colon, and rectum expressed neither BARX1 nor HAND1. In the small intestine, where intramuscular and myenteric ICCs could be readily distinguished, HAND1 expression was more common in ICCs within the myenteric plexus (Supplementary Fig. S5C), suggesting transcriptional differences in ICCs at different stations within the same organ; this suggests that specific ICC subsets are at higher risk for malignant transformation. In aggregate, these data outline anatomic and spatial differences in accessory TF expression in ICCs, and suggest that anatomic restriction of accessory TF expression underlies the differential clinical behavior of GIST derived from the stomach versus the small intestine.

A core circuit of TFs is essential for establishing and maintaining a normal transcriptional state of cellular homeostasis, and this state can become dysregulated during the process of neoplastic transformation (16–20). Accessory disease state–specific TFs can modulate biological and molecular phenotypes and thereby shape clinical behavior. Here, we show that BARX1 and HAND1 are the principal accessory TFs in GIST that modify the core transcriptional circuitry and show strong associations with clinical features and disease outcomes. In addition, we reveal a striking anatomic restriction of accessory TF expression that reflects the developmental origins of these tumors and the difference in malignant potential between gastric and small intestine GIST.

BARX1 is a member of the Bar subclass of homeobox TFs and has been studied in multiple developmental contexts, including its role in coordinating gastric myogenesis and intestinal rotation (45). HAND1 (Heart And Neural Crest Derivatives Expressed 1) is a basic helix-loop-helix family TF best understood for its role in early embryonic development and cardiac morphogenesis (46). Both BARX1 and HAND1 have previously been studied in cancer contexts, with reports suggesting suppressive or supportive roles in different tumor histologies (47–49). In GIST, there is remarkable homogeneity of the gene expression program between tumors (Fig. 1A), although mutation and clinical subtypes display subtle transcriptional differences as also observed in other studies (35, 50, 51). Early in the course of disease there is typically limited chromosomal instability in GIST (52, 53), including an absence of amplifications involving the KIT locus (54). The lack of chromosomal aberrations, together with a recurrent enhancer landscape across disease states (21), suggests that GISTs rely upon native ICC chromatin and transcriptional environments. It appears that chromosomal stability is selected, at least in part, to support the enhancer-driven expression of KIT in the majority of tumors (55). This dependency places accessory TFs HAND1 and BARX1 in a pivotal position to influence disease biology by modifying the basal transcriptional state of GIST cells.

Active signal transduction and cell-cycle programs, as observed at the transcriptional level in HAND1-expressing tumors, is consistent with the higher mitotic rate and proliferative features seen in high-risk and metastatic GIST (12, 53). The primary influence of HAND1 may be to modify the expression of, or protein interactions within, the core TF network to support the GIST growth program, most notably KIT gene expression. The HAND1-driven program may not only offer prognostic value, but also represent a therapeutic vulnerability either itself or by targeting genes whose expression is regulated by HAND1. For example, HAND1 regulates GPR20 expression (Supplementary Fig. S1D), and an anti-GPR20 antibody–drug conjugate has now entered clinical trials (NCT04276415). Additional mechanistic studies are needed to better define how HAND1 orchestrates a more malignant phenotype, which may reveal additional therapeutic vulnerabilities. We observed recurrent enrichment of an EMT signature in BARX1-expressing GIST, or following loss of HAND1 function in vitro. This may suggest that, in the context of GIST, other biological features of the EMT program are engaged including those involved in organ development, fibrosis or tissue repair (56); this contrasts with epithelial-derived malignancies. Of note, our analysis of GIST transcription focused on biologically and clinically distinct categories of GIST and did not prioritize subtypes by incidence; additionally, full clinical annotation of all samples (i.e., adjuvant therapy use) was unavailable, which represents a weakness of this study.

In our primary tumor cohort, HAND1 expression had a strong positive predictive value for KIT mutation. In all samples with known underlying mutation, HAND1 was expressed in 53% (134/255) of KIT-mutant tumors, 3% (1/34) of PDGFRA mutant tumors, 0% (0/14) of SDH-deficient tumors, and 100% (2/2) of NF1-associated tumors. The lone PDGFRA-mutant GIST that expressed HAND1 also expressed BARX1 and was found to have an uncommon in-frame deletion (p.D842_H845del); at resection it was classified as a low-risk gastric GIST. Because NF1-associated GIST have a different underlying etiology and clinical course than more common GIST subtypes (6), and these tumors are derived from small bowel ICCs that natively express HAND1, the contribution of HAND1 expression to oncogenesis in this tumor type is less clear. Although the majority of GISTs express either HAND1 or BARX1, 15% are negative and 6% positive for both markers. The numbers of these GIST subsets in our cohort were too small to assess outcomes, and further study is needed to assess BARX1 and HAND1 expression in a larger cohort of these rare tumor subtypes and characterize their clinical course. Furthermore, given the retrospective nature of these studies and limited available clinical data, additional studies are needed to better establish these findings.

Multiplexed tissue imaging (CycIF) makes it possible to study correlations in the expression of HAND1, BARX1, and other markers of cellular proliferation at the single-cell level. Consistent with what was observed in bulk tumor RNA-seq data and in vitro with loss of HAND1, GIST cells expressing HAND1 have higher expression of proliferative markers. Rare HAND1-positive cells were also observed in localized gastric GIST, which predominantly express BARX1. Conceivably, because stomach ICCs do not natively express HAND1, as malignant clones of gastric-derived GIST evolve over time they may activate HAND1 expression, giving rise to this double-positive tumor subset. With disease progression, the naturally higher proliferative rates of HAND1-positive cells may cause them to dominate the tumor landscape in the recurrent or metastatic setting, and cells may eventually lose expression of BARX1. Additional study of accessory TFs in case series with paired primary and metastatic samples may further define this evolution. Our observation that BARX1 or HAND1 is absent in ICCs of the esophagus, colon, appendix, and rectum agrees with clinical observations that these anatomic compartments are less competent to support transformation into GIST (57). That the majority (85%) of GISTs express BARX1 and/or HAND1, inclusive of tumors from sites whose ICCs natively lack these TFs, suggests that the presence or elaboration of these accessory TFs is commonly required to support malignancy. From in vitro experiments, ectopic expression of BARX1 was not deleterious to the HAND1-driven proliferative program (Supplementary Fig. S1E and S1F), suggesting that both TFs are permissive of a malignancy program, but that HAND1 is the driver of more aggressive behavior. The prominent anticorrelation of these accessory TFs in ICCs and GIST samples warrants further research on their role in ICC development and mechanisms of cross-regulation.

S100A has previously been reported to be a marker of high-risk GIST (41); however, in our tumor cohort, we only identified S100A within tumors derived from the small intestine. Thus, although HAND1 and S100A expression may both be correlated with small intestine origin, S100A appears to be solely associated with this anatomic site, whereas HAND1 appears to be an acquired driver of aggressive behavior in other regions. CyCIF allowed the interrogation of select immune subsets in a diverse array of GIST samples. We found that myeloid-lineage (CD11b) and macrophage (CD68, CD163) markers were present at low levels in tumors, and there was no significant difference between GIST mutational subtypes.

The anatomic and spatially restricted expression of BARX1 and HAND1 in ICCs, and associated intrinsic differences in clinical behavior based on the anatomic origin of GIST, further argues for the role these TFs play in the pathogenesis of GIST. ICCs in different anatomic compartments have previously been shown to have different densities, distributions, and gene expression profiles (58, 59), although the results presented here are the first to make a mechanistic link between a gene expression program in ICCs and clinical behavior upon transformation into GIST. Additional genetic studies in laboratory models of GIST altering accessory TF expression in ICCs may further clarify their role in directing oncogenesis.

The greatest potential value in assessing HAND1 and BARX1 expression in GIST lies in predicting the eventual course of the disease in patients. Especially for macroscopically resected GIST deemed intermediate risk, or where there is discrepancy, lack of expertise, or an inability to determine mitotic index (i.e., in cases confounded by preoperative administration of imatinib), HAND1 and BARX1 expression may help support clinical decisions on assessing the relative benefits of adjuvant imatinib therapy. As the majority of tumor cells stain positive for either HAND1 or BARX1, IHC is a standard and effective means of assessing the presence or absence of these biomarkers. In the metastatic setting, our data show that HAND1 expression is associated with a shorter progression-free survival, which may inform disease assessment intervals or clinical trial eligibility criteria. Further study is warranted to assess the added clinical value of HAND1 and BARX1 assessment, and whether routine assay of these TFs could enhance the current standard of care and improve risk stratification for patients with GIST.

E. Mazzola reported personal fees from The VeraMedica Institute LLC outside the submitted work. P.K. Sorger reported grants from NCI and personal fees and nonfinancial support from RareCyte Inc during the conduct of the study, as well as personal fees from Applied Biomatch LLC, Glencoe Software Inc., and NanoString Inc. outside the submitted work. S.A. Armstrong reported personal fees from Imago Biosciences, Vitae/Allerga, Cyteir Therapeutics, C4 Therapeutics, OxStem Oncology, Accent Therapeutics, Mana Therapeutics, and Neomorph, Inc and grants from Janssen, Novartis, and AstraZeneca outside the submitted work. G.D. Demetri reported personal fees and nonfinancial support from Bayer; personal fees from Pfizer; other from Novartis; and personal fees and other from Blueprint Medicines outside the submitted work; in addition, G.D. Demetri had a patent for Imatinib for GIST issued, licensed, and with royalties paid from Novartis; and reported scientific consultancy with sponsored research to Dana-Farber with Epizyme, Roche/Genentech, Epizyme, LOXO Oncology, AbbVie, GlaxoSmithKline, Janssen, PharmaMar, Daiichi-Sankyo, and AdaptImmune, scientific consultancy with GlaxoSmithKline, EMD-Serono, Sanofi, ICON plc, MEDSCAPE, Mirati, WCG/Arsenal Capital, Polaris, MJ Hennessey/OncLive, C4 Therapeutics, Synlogic, and McCann Health, consultant/SAB membership with minor equity holding with G1 Therapeutics, Caris Life Sciences, Erasca Pharmaceuticals, RELAY Therapeutics, Bessor Pharmaceuticals, Champions Biotechnology, and Caprion/HistoGeneX, board of director membership and scientific advisory board consultancy with minor equity holding with Blueprint Medicines and Translate BIO, and nonfinancial interests in AACR Science Policy and Government Affairs Committee Chair and Alexandria Real Estate Equities. S. Santagata reported personal fees from RareCyte, Inc outside the submitted work. No disclosures were reported by the other authors.

M.L. Hemming: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. S. Coy: Data curation, formal analysis, investigation, visualization, methodology, writing-review and editing. J.-R. Lin: Software, formal analysis, investigation, visualization, methodology, writing-review and editing. J.L. Andersen: Data curation, investigation, writing-review and editing. J. Przybyl: Data curation, investigation, writing-review and editing. E. Mazzola: Formal analysis, writing-review and editing. A.H. Abdelhamid Ahmed: Data curation, investigation, writing-review and editing. M. van de Rijn: Data curation, project administration, writing-review and editing. P.K. Sorger: Data curation, project administration, writing-review and editing. S.A. Armstrong: Supervision, project administration, writing-review and editing. G.D. Demetri: Resources, supervision, funding acquisition, writing-review and editing. S. Santagata: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing-review and editing.

We thank Dana-Farber/Harvard Cancer Center in Boston, MA, for the use of the Specialized Histopathology Core, which provided histology and tissue microarray services. Dana-Farber/Harvard Cancer Center is supported in part by an NCI Cancer Center Support Grant NIH P30-CA06516. Funding for this study was provided by the Harvard Catalyst Medical Research Investigator Training Program, NIH Award UL 1TR002541 (to M.L. Hemming), NIH Award 1 K08 CA245235-01A1 (to M.L. Hemming), the Spivak Faculty Advancement Fund (to M.L. Hemming), the DFCI Medical Oncology Grant Program (to M.L. Hemming), the Ludwig Center at Harvard (to G.D. Demetri, P.K. Sorger, and S. Santagata), NIH grant U54-CA225088 (to P.K. Sorger and S. Santagata), NIH grant T32 GM007748 (to S. Coy), and the Dubai Harvard Foundation for Medical Research (to A.H. Abdelhamid Ahmed).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.