Clinicopathologic and Molecular Features of Pancreatic Ductal Adenocarcinomas Harboring Alterations in COMPASS-like Complex Genes

With University of Michigan (Ann Arbor, Michigan) Institutional Review Board approval (HUM 00098128, 4/12/2022), a retrospective search was performed to identify patients receiving oncologic care at our institution for PDACs that have undergone comprehensive molecular profiling either using large commercially available panels (Tempus, Foundation Medicine) or internally through the Michigan Oncology Sequencing Center. The molecular sequencing reports were reviewed to identify tumors with alterations in genes comprising the COMPASS-like complex, including KMT2A, KMT2B, KMT2C, KMT2D, and KDM6A. Alterations considered to be likely benign (polymorphisms) or variants of unknown significance, which comprised nearly all missense mutations, were excluded from further analysis, as the functional significance of these alterations could not be verified.

Demographic data (age and sex) and associated risk factors (smoking and alcohol consumption), serum CA19-9 levels, radiographic features, treatment data, disease recurrence, and patient survival were obtained from the electronic medical record. Regarding outcome data, disease-free survival (DFS) for patients who underwent primary resection was defined as the time from the date of surgery to the time of disease recurrence or progression. Progression-free survival for those with unresectable, advanced disease was defined as the time from treatment initiation to disease progression or death. Overall survival was defined as the time from date of diagnosis to the date of death from any cause.

All available hematoxylin-eosin (H&E)–stained slides were from biopsies and/or resection specimens from primary sites or metastases that underwent molecular profiling were reviewed. In cases of metastatic or recurrent disease, prior resections were also reviewed when available. Histologic subtype and grade were categorized according to the current World Health Organization criteria.¹⁰  Other features, including lymphovascular invasion, perineural invasion, and lymph nodes metastasis, were documented. Primary resections were assigned pathologic stage according to the American Joint Committee on Cancer 8th edition criteria.¹¹  For unresectable cases, the clinical staging was documented according to radiologic findings and clinical impressions reported in the electronic medical record.

For cases with available tissue, a p63 immunohistochemical stain was performed on a BenchMark Ultra autostainer (Ventana, Tucson, Arizona). Formalin-fixed, paraffin-embedded tissue was cut at 4 μm and deparaffinized. Heat-induced epitope retrieval was performed using cell-conditioning 1 buffer (Ventana) at 95°C for 36 minutes. Slides were then incubated with the p63 antibody (clone 4A4, prediluted, Ventana) for 32 minutes at 36°C, and immunoreactivity was detected using the Ultraview universal DAB detection kit (Ventana). The slides were then counterstained with hematoxylin for 8 minutes, followed by a bluing step for 4 minutes.

Due to the limited number of cases identified in our internal cohort, 2 databases were also interrogated for PDACs with mutations in KMT2 or KDM6A genes through cBioPortal, including the Cancer Genome Atlas (TCGA), PanCancer Atlas data set (whole-exome sequencing and copy-number profiling of 184 samples), and the Australian Pancreatic Cancer Genome Initiative database (APGI) as part of the International Cancer Genome Consortium (whole-exome sequencing of 456 pancreatic carcinoma samples) (http://cbioportal.org).^7,12,13  PDACs harboring likely pathogenic deleterious alterations in the abovementioned genes were further evaluated for clinicopathologic features and concurrent molecular alterations and compared with wild-type tumors.

The χ² and Fisher exact tests were used to assess the association between categorical clinicopathologic features and the mutation status, and analysis of variance models were used to compare continuous clinicopathologic features between different groups. For the survival data (progression-free or overall survival), the Kaplan-Meier method was used to estimate the survival functions for different groups, and the log-rank test was used for the comparison. Statistical analysis was performed using SAS (version 9.4). Statistical significance was defined as P < .05.

Altogether, 103 patients with PDAC treated at our institution, whose tumors had undergone comprehensive molecular profiling, were identified between 2015 and 2021. Of those, 13 tumors harbored variants of COMPASS-like complex genes that were predicted to have a deleterious effect due to premature protein truncation or significant structural alteration. Twenty-one patients had missense mutations reported as variants of unknown significance in COMPASS-like complex genes, several of which had at least some evidence to suggest they may represent benign polymorphisms, and these patients were excluded from further analysis. A detailed summary of each of the 13 cases within the study cohort, including clinicopathologic and molecular characteristics as well as treatment-related information and outcomes, are presented in the supplemental digital content at https://meridian.allenpress.com/aplm in the September 2023 table of contents. The male to female ratio was 9:4 with a mean age of 66 years (range, 51–75 years). The COMPASS-like complex gene mutations predominantly consisted of nonsense/truncating and frameshift mutations in KMT2D (n = 7) and KDM6A (n = 4) (Figures 1 and 2). One tumor harbored frameshift/truncating mutations in both KMT2C and KMT2D, and 1 tumor harbored 2 separate frameshift mutations in KMT2D. Several of the cohort cases also harbored additional missense mutations of unknown significance in COMPASS-like complex genes. These tumors often harbored concurrent alterations in known driver gene mutations in PDAC including KRAS (13/13, 100%), TP53 (8/13, 62%), SMAD4 (3/13, 23%), and CDKN2A (6/13, 46%). The variant allele fraction (VAF) information was available for 5 tumors (cases 5 and 8–11). Interestingly, the VAF of the COMPASS-like complex genes are comparable with the KRAS/TP53/SMAD4/CDKN2A VAFs, suggesting they may represent driver mutations. Mismatch repair status, as determined by next-generation sequencing, was available for 10 tumors. One tumor (case 2) was found to be mismatch repair–deficient as well as tumor mutation burden (TMB)–high. This tumor and 1 additional case (case 13) also harbored a frameshift mutation in BRCA2.

Four tissues tested were from the primary site, whereas the remaining 9 tissue samples were from metastatic sites. Eight tumors were located in the pancreatic head/uncinate, whereas 4 were in the body and 1 was in the tail. The median tumor size was 2.5 cm (range, 1.5–3.5 cm). Seven (7/13) patients were originally found to have localized disease and underwent curative intent resection, of which 5 specimens had H&E-stained slides available for re-review, whereas the remaining 6 patients presented with locally advanced (unresectable) (n = 3) or metastatic disease (n = 3). Three patients who underwent resection received neoadjuvant chemotherapy. Tumor grade was initially classified as moderately differentiated in all primary resections. However, a poorly differentiated component was at least focally present in 4 of the 5 (80%) resection specimens on re-reviewing the material (Figure 3, A and B). Lymphovascular and perineural invasion were present in 5/7 (71%) and 6/7 (86%) primary resections, respectively. Lymph node metastases were identified in 5/7 (71%) cases. Of the 3 patients treated with neoadjuvant chemotherapy, 2 tumors showed partial response to treatment (College of American Pathologists response grade 2), while there was extensive residual cancer without evident tumor regression in 1 tumor (College of American Pathologists response grade 3).

H&E-stained slides were also evaluated from 7 of 9 metastatic samples that underwent sequencing and were available for review. Six of 7 (86%) reviewed cases showed poorly differentiated features. Three of 5 (60%) metastatic cases with tissue available for immunohistochemical staining showed squamous differentiation, as confirmed with patchy nuclear p63 positivity (Figure 4, A through D). However, 4 primary resections with available tissue were all diffusely negative for p63 (Figure 4, E and F).

To assess whether the clinicopathologic features of these COMPASS-like complex gene–mutated PDACs are different from traditional PDACs, 45 patients with PDACs lacking alterations in COMPASS-like complex genes who had adequate clinical documentation and tissue available for histologic review were evaluated as a control group. The patient demographics, risk factors, tumor histology, clinical stage, and pertinent molecular alterations for both the study cohort and control cases are presented in Table 1. Tumor size was smaller in the study cohort compared with the control cases (2.5 versus 3.2 cm, respectively, P = .04). Tumors in the study cohort more frequently contained poorly differentiated foci than in the control group (8/10, 80% versus 10/32, 31%, respectively, P = .01). The tumors in the study cohort also tended to have more squamous differentiation (3/10, 30% versus 3/32, 9%) (Table 1). However, there was no significant difference in patient demographics, smoking and alcohol use history, CA19-9, histologic subtype, overall tumor grade, or stage at presentation between the study cohort and controls. Treatment regimens, including the number of patients who underwent primary resection or received neoadjuvant or adjuvant chemotherapy, also did not differ. Lastly, the alteration rate of mutations in known driver genes (KRAS, TP53, SMAD4, and CDKN2A) did not significantly differ between the 2 groups nor did it differ from what has been reported in the literature for PDAC in general.

Clinical follow-up was available in all patients, with a median follow-up interval of 19 and 24 months (range, 4–97 months and 2–110 months) in the study and control cohorts, respectively (Table 1). Approximately half of the patients in each group had localized resectable disease at the time of initial presentation. Of these, all patients (7/7, 100%) in the study cohort eventually developed disease recurrence or progression, with a median DFS of 8 months (range, 3–40 months), compared to 20 of 22 (91%) patients in the control group with localized disease. The control group also had a prolonged median DFS of 18 months (range, 2–41 months). Overall, metastatic disease, either at the time of presentation or with subsequent progression, was slightly more frequent in the study population versus controls (85% versus 72%, respectively), though this did not reach statistical significance. For patients with unresectable disease in the study and control groups, all were started on FOLFIRINOX (3/6 study group patients and 13/23 control patients) or gemcitabine-based (2/6 study group patients and 10/23 control patients) chemotherapy regimens, except 1 patient who died before systemic therapy could be initiated. Disease progression occurred in all patients (5/5, 100%) in the study cohort and 21 of 23 (90%) of patients in the control cohort, with a median progression-free survival of 5 months (range, 2–8 months) in the study group compared to 6 months (range, 1–15 months) in the control group. At the end of the follow-up, 10 of 13 (77%) patients in the study cohort were deceased, and 3 of 13 (23%) were alive with disease, similar to the control group where 35 of 45 (78%) patients were deceased and 9 of 45 (20%) were alive with disease. Overall survival from diagnosis to death was a median of 21 months (range, 4–97 months) in the study cohort compared to a median of 28 months (range, 7–110 months) in the control group. Overall, no statistically significant differences in therapeutic response or overall prognosis were identified between the study cohort and control cases.

Due to the limited patient number in our study cohort, we decided to survey the TCGA and APGI data available through the cBioPortal to identify additional cases of PDACs harboring likely pathogenic deleterious mutations in COMPASS-like complex–related genes. Notably, these cohorts were comprised predominantly of low-stage, treatment-naïve primary tumors that underwent surgical resection, with tissue from the primary site being sequenced. The initial search revealed variable mutation types, including missense, truncation, frameshift, and splice-site mutations. However, similar to those identified in our internal cohort, missense mutations were excluded due to a lack of functional characterization. Our interrogation revealed 47 of 640 (7%) PDACs harboring alterations predicted to have a deleterious effect due to splice-site mutations, nonsense or truncating mutations, or frameshift alterations in genes associated with COMPASS-like complex: the clinicopathologic and molecular features of these cases are summarized in Table 2 and Figure 5. The average age was 66 years (range, 34–90 years), and the male to female ratio was 24:23. Most tumors were located in the head of the pancreas (70%, 33/47), and most were low stage based on the 7th edition of the American Joint Committee on Cancer staging system (9%, 4/47 stage I, 83%, 39/47 stage II, and 9%, 4/47 stage III).¹⁴  Four cases (9%, 4/47) were noted to have squamous differentiation per reports, and 4/47 (9%) were considered undifferentiated tumors. Altogether, histologic tumor grading was considered poorly differentiated by the initial pathology report in 26/47 (55%) of cases. Forty-three of 47 (91%) and 39/47 (83%) cases harbored concomitant KRAS mutations and TP53 mutations, respectively. Mutations or loss of SMAD4 and CDKN2A were seen in 17/47 (36%) and 14/47 (30%) cases. A comparison of the clinicopathologic features in tumors with alterations in COMPASS-like complex genes between the in-house cohort and the population database cases are summarized in Supplemental Table 2.

The histopathologic and molecular features of the 47 TCGA and APGI study cases were compared with available data on PDACs harboring wild-type COMPASS-like complex genes in these 2 cohorts, the findings of which are summarized in Table 3. There were no significant differences in age, sex, or stage at the time of tissue acquisition between the 2 groups. However, tumors with COMPASS-like complex gene mutations, though most often located in the head overall, did show increased frequency within the body or tail of the pancreas compared with the control group (32%, 15/47 versus 15%, 79/515, P = .005) and were more likely to show poor histologic differentiation or undifferentiated carcinoma (55%, 26/47 versus 33%, 166/499, P = .008). In addition, these tumors more often harbored TP53 mutations (83%, 39/47 versus 61%, 290/479, P = .002) and showed a trend toward more frequent concurrent KRAS mutations (91%, 43/47 versus 80%, 385/479, P = .06). Detailed outcome data were available for the TCGA cohort, and overall patient survival was slightly decreased in the COMPASS-like complex gene–mutated tumors compared with wild-type PDACs (log-rank test P value = .06) (Figure 6), with a median survival of 17.49 months in the mutated tumors versus 21 months in the wild-type tumors. For both datasets, patient status at the time of the last follow-up was available. Alterations in COMPASS-like complex genes were associated with increased death rates, with 37/47 deaths (79%) in patients with mutated tumors compared to 306/511 deaths (60%) in patients with wild-type tumors (odd ratio: 2.48, P = .007).

Herein, we present the largest case series of PDACs with alterations in COMPASS-like complex genes. Using an internal cohort of patients along with publicly available data from the TCGA and APGI cohorts, we found that alterations in genes associated with the COMPASS-like complex are identified in just less than 10% of pancreatic carcinomas, even when using stringent inclusion criteria due to the lack of functional data available on the majority of missense mutations detected in these genes. While most clinicopathologic features did not significantly differ between this subset of tumors and controls, we did find at least focal areas of poor differentiation to be a significantly more common histologic finding, and squamous differentiation was not infrequently seen. Interestingly, the latter was present exclusively in metastatic samples but not with primary resections, the significance of which requires further exploration. Additionally, despite the smaller tumor size at the time of presentation, we found a slight trend toward decreased DFS and overall survival in the study group compared to the control group, though response to therapy did not significantly differ between these groups.

As comprehensive, large-scale studies that go beyond simple DNA-based mutational analyses expand our understanding of the complexity of PDAC pathogenesis, it is becoming increasingly clear that specific molecular subtypes of PDAC do exist, which may have significant therapeutic and prognostic significance. In fact, several studies have proposed various subclassifications based on molecular profiling data. Initial clustering techniques using gene expression microarray analysis revealed a 3-group classification scheme including classical, quasi-mesenchymal, or exocrine-like tumors.¹⁵  Subsequent analyses using gene expression profiling with microdissection by Moffitt et al¹⁶  identified 2 main groups, with classical and basal-like tumor signatures. An integrated analysis by APGI using an RNA sequencing-based approach revealed 4 subtypes, including squamous, pancreatic progenitor, immunogenic, and aberrantly differentiated endocrine exocrine.¹⁷  Later, the TCGA consortium applied similar clustering techniques from the previously mentioned studies and found that when using high tumor-content samples, 2 main groups were identified with overlap from the previously mentioned subclassification schemes: the classical/progenitor group and basal-like/squamous group, which has been further corroborated by more recent studies.^13,18,19  This subclassification of pancreatic carcinomas has potential clinical relevance, as multiple studies have found that the basal-like/squamous group is associated with poorly differentiated tumors and decreased overall survival.^16,17,20,21  Interestingly, the TCGA study and others found that the distinction between basal-like and classical subtypes of pancreatic carcinoma could be made based on methylation patterns and distinct chromatin states, suggesting that these 2 phenotypes may be related to epigenetic modifications rather than mutations in driver genes.^22,23 

In support of epigenetic modification driving distinct molecular subgroups, several specific pathways involved in epigenetic regulation via histone modification have been shown to be altered in pancreas cancer,^17,24  which is not surprising given that posttranslational modifications in histone subunits including acetylation and methylation are critical components of pancreatic cancer tumorigenesis.^25,26  Alterations in histone modification enzymes (ie, KDM6A, SETD2, MLL2/3, CBP/p300), resulting in upregulation or suppression of gene expression, have been identified in up to 24% of pancreatic tumors.^17,24  In particular, the methylation of lysine residues is an important component of this regulation, and lysine-specific histone demethylases have been shown to play a pertinent role in human cancer.^27,28  The histone methylation is controlled in part by the COMPASS-like complex, comprised of KDM6A and KMT2 enzyme families (histone H3K4-specific methyltransferases) and related core proteins involved in the methylation of lysine residues to establish active enhancers or superenhancers (clusters of enhancers that activate gene expression). Alterations in genes associated with the COMPASS-like complex are frequently identified in a wide variety of solid tumors and hematologic malignancies.^29–32  Studies, including ours, have proposed that mutations in these genes favor the development of poorly differentiated, aggressive tumors.^9,33  A multiplatform study of 12 cancer types by Hoadley et al²⁹  found that COMPASS-like complex members (KMT2C, KMT2D, and KDM6A) were the third most mutated (37.1% mutation rate) subnetwork in the squamous-like cancer subtype and conferred a poor prognosis. A recent study evaluating molecular alterations in metaplastic or sarcomatoid carcinoma identified frequent mutations in KMT2D often co-occurring with TP53 mutations.³⁴  Mutations were predominantly nonsense and frameshift insertions/deletions, and 2 or more mutations in one patient (ie, a double hit) were seen in multiple cases. They found that KMT2D mutations were associated with large tumor size and unfavorable prognosis. Further evidence of functional significance was shown in this study by correlating mutation status with a low-level expression of KMT2D by immunohistochemistry.

In large cohorts of pancreatic cancer, sequencing studies have found similar trends. A study by Singhi et al²⁴  found that mutations in MLL2 (KMT2D) were more likely seen in metastatic disease. A whole-genome and copy-number variation analysis on pancreatic carcinomas identified KDM6A alterations in up to 18% of pancreatic tumors,⁶  and additional evidence has suggested that these alterations are associated with the pancreatic squamous molecular subtype.¹⁷  Studies have shown that loss of KDM6A disrupts the formation of a COMPASS-like complex at specific superenhancer sites, ultimately resulting in overexpression of genes linked to squamous differentiation such as TP63, MYC, and RUNX3. While KDM6A shows strong expression in well-differentiated pancreatic carcinomas, expression is absent in poorly differentiated components or metastatic tissue sites.^9,17  Andricovich et al,⁸  using genomic databases and cancer cell lines, found that loss of KDM6A can induce squamous differentiation through upregulation of these transcription factors, and similar to previous reports, loss of function of KDM6A correlated with shorter patient survival. Furthermore, through screening of cell lines, they found that KDM6A-mutant cells were sensitive to bromodomain and extraterminal inhibitors⁸ . Mechanistically, our group has found that loss of KDM6A upregulates the expression of activin A, a member of the TGF-β superfamily, and activates a p38 MAPK-dependent noncanonical activin A pathway to promote epithelial-mesenchymal transition, invasion, and metastasis.⁹  Another study found that tumors with mutations in KMT2D may be amendable to checkpoint blockade immunotherapy, with increased infiltration of CD4 and CD8 T cells, which may offer additional implications for future therapeutic opportunities.³³ 

While some of our findings are in keeping with the abovementioned data, our study has several limitations, and additional investigation into this unique molecular subgroup is warranted. While we did not observe a statistically significant difference in progression-free or overall survival between the study cohort and a control group of wild-type tumors, the power to detect such a difference was likely limited by our small study sample size. Another limitation of the current study is the lack of solid evidence to classify missense mutations as likely pathogenic or pathogenic, which may have excluded patients with COMPASS-like complex functional alterations. Alternatively, while we presumed truncating, frameshift, and splice-site variants to be likely pathogenic deleterious alterations, these alterations may not always result in nonsense-mediated mRNA decay, and thus, may not significantly affect protein function. Demonstration of loss of COMPASS-like complex function with immunohistochemical analysis or expressional data would further strengthen the variant classification and stratify additional patients into those affected and those with wild-type expression.

This study comprehensively evaluates the clinicopathologic and molecular characteristics of a subgroup of PDACs harboring alterations in the COMPASS-like complex genes. Correlating with evidence proposed in recent literature, we found that these tumors, despite being smaller in size in our internal cohort, often harbor poorly differentiated histologic features, and a subset show evidence of squamous differentiation. While the former may seem contradictory to the reported aggressive behavior of tumors harboring these alterations, evidence from prior studies has suggested that alterations in COMPASS-like complex promote mesenchymal differentiation and increased invasion and metastasis rather than directly inducing increased tumor growth.⁹  In external data sets, they were also associated with increased TP53 mutations, body or tail localization, and poor differentiation or undifferentiated carcinoma. Patients with PDACs carrying COMPASS-like complex gene alterations have a higher death rate and tend to have a shorter disease-free and overall survival compared to control population, albeit without statistical significance in the latter due to small sample size. Additional investigation with functional characterization of mutations and their impact on epigenetic regulation as well as evaluation of larger study sets will further expand our understanding of the pertinent role that histone modification may have on pancreatic carcinoma. Deciphering this role in the context of specific driver gene mutations may have pertinent treatment-related and prognostic implications that could allow for a more personalized therapeutic approach to managing patients with such a relentless and aggressive disease.