Defining the molecular pathology of pancreatic body and tail adenocarcinoma

Abstract Background Pancreatic ductal adenocarcinoma (PDAC) remains a dismal disease, with very little improvement in survival over the past 50 years. Recent large-scale genomic studies have improved understanding of the genomic and transcriptomic landscape of the disease, yet very little is known about molecular heterogeneity according to tumour location in the pancreas; body and tail PDACs especially tend to have a significantly worse prognosis. The aim was to investigate the molecular differences between PDAC of the head and those of the body and tail of the pancreas. Methods Detailed correlative analysis of clinicopathological variables, including tumour location, genomic and transcriptomic data, was performed using the Australian Pancreatic Cancer Genome Initiative (APGI) cohort, part of the International Cancer Genome Consortium study. Results Clinicopathological data were available for 518 patients recruited to the APGI, of whom 421 underwent genomic analyses; 179 of these patients underwent whole-genome and 96 RNA sequencing. Patients with tumours of the body and tail had significantly worse survival than those with pancreatic head tumours (12·1 versus 22·0 months; P = 0·001). Location in the body and tail was associated with the squamous subtype of PDAC. Body and tail PDACs enriched for gene programmes involved in tumour invasion and epithelial-to-mesenchymal transition, as well as features of poor antitumour immune response. Whether this is due to a molecular predisposition from the outset, or reflects a later time point on the tumour molecular clock, requires further investigation using well designed prospective studies in pancreatic cancer. Conclusion PDACs of the body and tail demonstrate aggressive tumour biology that may explain worse clinical outcomes.


Introduction
Pancreatic ductal adenocarcinoma (PDAC) is expected to become the third leading cause of cancer-related death in Western societies, and to overtake breast cancer for the first time in 2017 1 . The 5-year survival rate, almost unchanged in 50 years, remains less than 10 per cent 1 . Surgical resection is the only chance of cure, with chemotherapy adding modest benefit to overall survival 2 -4 . Early recurrence remains a major clinical concern for patients undergoing pancreatectomy, and accurate preoperative prognostication is currently a challenge.
Numerous studies 5 -9 have demonstrated meaningful differences in outcome of pancreatic cancers located in the head, compared with those of the body and tail.
Approximately 15 per cent of PDACs occur in the body and tail, and differences in outcome have been largely attributed to late presentation in comparison with tumours of the pancreatic head 5 -7,9 . Tumours of the head and uncinate process often present with jaundice, and are therefore thought to present earlier in the disease process. Body and tail pancreatic cancer usually presents with weight loss and pain, symptoms more in keeping with advanced disease 7 . Yet, previous data suggest that TNM stage at presentation is not significantly different between the two tumour locations 10 .
Recent large-scale sequencing studies 11 -16 have demonstrated that PDAC harbours significant interpatient genomic heterogeneity. Apart from a few well known mutations that are currently not targetable (KRAS, TP53, CDKN2A and the loss of SMAD4), most genetic aberrations occur at low frequency (10 per cent or less). Whole-genome sequencing of 100 resected PDACs demonstrated unique structural variation subtypes based on chromosomal rearrangement numbers and patterns that appear to predict response to platinum-based chemotherapy in a 'synthetic lethality' manner 13 . Additionally, integrated genomic analyses 14 -16 revealed distinct molecular subtypes of PDAC based on transcriptomic profiles that correspond to clinical outcomes. Bailey and colleagues 14 described a poor prognostic 'squamous' subtype that is enriched for histopathological adenosquamous tumours, TP53 mutations 17 , and gene programmes associated with inflammation, hypoxia response, metabolic reprogramming, MYC pathway activation and transforming growth factor β signalling 14 . The squamous subtype was enriched for gene programmes that are common in squamous-like tumours of breast, bladder, lung, and head and neck cancer 18 ; and characterized by hypermethylation and downregulation of genes involved in pancreatic endodermal differentiation (PDX1, MNX1, GATA6, HNF1B) 14 . In addition, the squamous subtype was enriched for mutations and loss of key epigenetic regulators (such as KDM6A) and this may contribute to the loss of endodermal properties of these tumours 14 . Expression patterns of immune cell populations within the tumour microenvironment demonstrated unique differences between transcriptomic subtypes, with evidence of immune avoidance in the squamous subtype 14 .
The molecular pathology of PDAC has recently been studied intensely, yet the exact genetic and molecular differences between PDACs of the head and those of the body and tail have not been fully elucidated. Therefore, the aim of this study was to explore and define the genomic and transcriptomic differences between head and body/tail PDAC based on a large integrated genomic analysis of PDAC 14 .

Methods
Patients were recruited prospectively through the Australian Pancreatic Cancer Genome Initiative (APGI) as part of the International Cancer Genome Consortium (ICGC). Informed consent and human research ethics approvals were obtained from each contributing clinical centre (Appendix S1, supporting information). Contributing patients were restricted to those with resectable, chemotherapy-and radiotherapy-naïve PDAC, who underwent either Whipple's pancreatoduodenectomy, or total or distal pancreatectomy. Following surgical resection, histopathological analysis was performed by a pancreatic pathologist. Specimens with macroscopic evidence of tumour were snap-frozen as a source of tumour DNA, along with samples of duodenum, stomach or spleen as a source of germline DNA. Standard histopathological processing was undertaken, and a diagnosis of PDAC was confirmed independently by two consultant pathologists with a specialist interest in pancreatic cancer. Fresh-frozen tumour samples underwent full-face cryosection to confirm the presence of tumour and to estimate epithelial cellularity. Macrodissection was performed before DNA and RNA extraction to enrich for tumour epithelium. DNA and RNA extraction was carried out using the All-Prep Kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol.

Data analysis
Whole-exome and/or whole-genome and RNA sequencing was performed and analysed as described previously 12 -14 . Briefly, genome sequence data were aligned and mapped using the Genome Reference Consortium GRCh37 assembly, and the Burrows-Wheeler alignment tool 19 . Single-nucleotide substitutions and insertions/deletions, structural variations, copy number analysis, mutational signature and RNA sequencing analysis was carried out 12 -14 . Hierarchical clustering, gene set and pathway enrichment analysis was performed using the R package (R Project for Statistical Computing, Vienna, Austria) 14 .

Statistical analysis
Disease-specific survival was used for all survival analyses. Median survival was estimated by the Kaplan-Meier method, and differences tested using the log rank test 20 . P < 0⋅050 was considered statistically significant. Clinicopathological variables with P < 0⋅100 on log rank testing were entered into a Cox proportional hazards multivariable analysis 21 . Statistical analysis was performed using SPSS ® version 22.0 (IBM, Armonk, New York, USA) and R 3.3.1. Table 1 and Table S1 (supporting information). In total 518 patients with detailed clinical and pathological data were identified; 421 PDACs underwent DNA sequencing, consisting of 179 whole genomes and 242 whole exomes. Ninety-six patients underwent whole-transcriptome RNA sequencing and another 266 underwent transcriptomic characterization based on microarray gene expression analysis owing  to lower tumour epithelial content, of whom 262 had survival data available for analysis (Table S1, supporting information).

Prognosis after pancreatectomy in relation to tumour location
In the APGI cohort, the majority of tumours (426, 82⋅2 per cent) were located in the head of the pancreas and 92 (17⋅8 per cent) presented with body and tail pancreatic cancer. There was no difference in patient demographics between those presenting with head versus body/tail tumours ( Table 1). Body and tail pancreatic cancers were more likely to be of lower pathological T category (T1-2) (21 versus 11⋅7 per cent; P = 0⋅021), yet significantly larger at the time of resection (P = 0⋅007) ( Table S2, supporting information). Although there was no  discernible difference in well known prognostic pathological variables between the two groups ( Table S2, supporting information), survival was significantly worse in patients presenting with body and tail tumours (median survival 12⋅1 versus 22⋅0 months; P = 0⋅001) (Fig. S1A, supporting information). This remained significant in multivariable analysis (hazard ratio 1⋅72, 95 per cent c.i. 1⋅31 to 2⋅26; P < 0⋅001) ( Table S3, supporting information).

Association between squamous subtype, tumour grade and recurrence
In view of the significant association between the poor prognostic squamous subtype and body and tail pancreatic cancer, a detailed clinical and pathological analysis of the squamous subtype was undertaken in 96 patients who underwent whole-transcriptome sequencing as part of the ICGC project. This demonstrated a significant association with poor tumour differentiation and higher tumour grade (P < 0⋅001), but not with pathological T category (P = 0⋅467), node status (P = 0⋅520) or surgical margin status (P = 0⋅615) ( Table 3). These findings were recapitulated in the microarray set (Table S5, supporting information). The squamous subtype was associated significantly with liver recurrence (P = 0⋅002) ( Table 3), and liver recurrence was associated with significantly worse survival compared with local and other distant recurrence patterns (median survival 13⋅6 versus 20⋅0 months respectively; P < 0⋅001)  ADAM17  DUSP14  ANXA1  TPD52L2  PANX1  S100A2  RND3  KRT6A  FSCN1  CEBPB  PLXNA1  GPR87  OSMR  ZBED2  ULBP2  IL1A  EGFR  TMSB10  ANGPTL4  ITGA3  SLC16A3  GSDMC  DRAP1  C16orf74  ARHGAP23  LDHA  PKM  HCAR2  PGAM1  ARPC2  KIAA1609  FZD6  FJX1  YWHAG  TMEM132A  PORCN  BTBD11  ANXA8  P4HA1  MIR205HG  AP2M1  MAP4K4  KIAA1467  BZW1  NAMPT  ARNTL2  SLC7A5  PTPN1  EREG 2 Graphical representation of the association between pancreatic adenocarcinoma location and differential transcriptional networks in the Australian Pancreatic Cancer Genome Initiative cohort. A potential theory of subtype evolution suggests that tumour size increases along the molecular clock, associated with dedifferentiation from pancreatic progenitor-like to squamous-like. EMT, epithelial-to-mesenchymal transition (Fig. S1B, supporting information), regardless of primary tumour location. Furthermore, all 20 patients with squamous tumours that had developed recurrence during follow-up died from distant metastatic disease ( Table 3).

Prognosis in relation to tumour location and squamous subtype
Kaplan-Meier survival analysis demonstrated that patients with the squamous subtype of PDAC had a significantly worse prognosis in the RNAseq set (median survival 13⋅3 months versus 23⋅7 months in those with non-squamous tumours; P = 0⋅010) (Fig. S1C, supporting  information). Resected tumours were segregated into tumour location and transcriptomic subtype, as defined in Bailey et al. 14 , to assess the difference in prognosis in squamous tumours of the head or body and tail. Patients who had squamous tumours of the body and tail had an extremely poor survival compared with the rest of the cohort (median survival 5⋅2 versus 22⋅0 months; P < 0⋅001) (Fig. 1b). These findings were recapitulated in the microarray set (median survival 25⋅0, 18⋅4, 15⋅9 and 11⋅5 months among patients with non-squamous tumours of the head, squamous tumours of the head, non-squamous tumours of the body/tail and squamous tumours of the body/tail respectively; P = 0⋅001) (Fig. S1D, supporting  information). The prognostic value of mutational signatures was assessed against the APGI cohort in 167 patients who underwent whole-genome sequencing and had sufficient clinical data available for analysis. There was no relationship between mutational signatures associated with loss of mismatch repair (P = 0⋅573), signature 8 (unknown aetiology) (P = 0⋅227), signature 17 (oesophageal cancer) (P = 0⋅639), deamination (P = 0⋅716), APOBEC (P = 0⋅899) or BRCA (P = 0⋅575) and survival in patients with resected PDAC.

Association between body and tail pancreatic cancer and molecular features of aggressive disease
An in-depth analysis of tumour location in relation to gene programmes that define the Bailey subtypes was undertaken. Body and tail pancreatic cancer was significantly associated with gene networks involved in epithelial-tomesenchymal transition (EMT), inflammation, hypoxia response, metabolic reprogramming, TP63 expression and squamous differentiation (gene programme 2) (Fig. 1a) 14 . Conversely, head tumours were enriched for gene programmes 6 and 8, which are associated with B cell and CD8-positive T cell signalling respectively 14 (Fig. 1; Table S6, and Figs S2 and S3, supporting information). Body and tail pancreatic cancer exhibited immune signatures corresponding to low dendritic cell infiltrate (P = 0⋅005), low co-stimulation of antigen-presenting cells (P = 0⋅041) and a low type II interferon response (P = 0⋅002) ( Table S6, supporting information). These findings suggest that, relative to pancreatic head cancer, body and tail tumours are associated with more aggressive disease biology and potentially exhibit a dampened antitumour immune response and increased immune avoidance.

Discussion
Clinical outcomes for body and tail pancreatic cancer appear to be significantly worse in both the resectable and advanced disease stages. The present study has demonstrated distinct molecular differences between resectable PDAC from the head and body/tail. Body and tail pancreatic cancer is associated with the squamous subtype of pancreatic cancer and enriched for gene programmes associated with inflammation, EMT and potential immune avoidance mechanisms.
The temporal sequence of genomic and epigenomic events leading to the progression to different PDAC subtypes has yet to be fully elucidated. However, the squamous subtype appears to be more advanced on the molecular clock than other subtypes, and this may reflect an additional level of genomic instability, owing to the accumulation of DNA damage, and molecular events that contribute to the unique transcriptome of these tumours 14 . Previous studies 14 -16 have found that these tumours are enriched for epigenetic events leading to hypermethylation and downregulation of genes involved in pancreatic development, and enriched for molecular drivers of EMT. The present results suggest that body and tail pancreatic cancer is more likely to be of a squamous subtype, suggesting that these are biologically more aggressive at the time of diagnosis, or surgical resection, than cancers of the pancreatic head. The results also suggest that the squamous subtype of PDAC is associated predominantly with liver recurrence; this may in part explain the worse prognosis of patients with liver metastases compared with those with metastases at other distant sites 22 . In this cohort, none of the intraductal papillary mucinous neoplasm-related PDACs were classified as the squamous subtype (Table 3), potentially revealing a different carcinogenesis pathway for the squamous subtype 14 . The squamous subtype correlates with increased size and poor tumour differentiation, which may reflect tumours that present at a later stage, or have an accelerated dedifferentiation pathway (Fig. 2).
Transcriptomic subtyping of PDAC remains a debated topic, and it has recently been suggested that there is significant overlap between the squamous subtype described by Bailey et al. 14 , and the basal 16 and quasi-mesenchymal 15 classifications. The role of stromal factors and immune infiltrate remains crucial to tumour growth and response to therapy, and is likely to remain a crucial aspect of molecular subtyping in PDAC 14,16 . As a result, the implications of tumour cellularity on transcriptomic analysis remain a priority area of research. However, several studies 23 -25 have demonstrated that epithelial cell lines of PDAC recapitulate published transcriptomic subtypes, including transcripts native to pancreatic exocrine and endocrine cells defining an aberrantly differentiated endocrine exocrine subtype (ADEX).
Genes involved in inflammation, EMT and invasion are enriched in body and tail pancreatic cancer; these molecular factors are known to be associated with poor prognosis. Levels of mRNA transcripts of calcium-binding protein S100A2, which accelerates tumour invasion, are greater in body and tail pancreatic cancer than in cancers of the pancreatic head, and this is one of the most differentially expressed genes in gene programme 2 ( Fig. 1). High S100A2 expression has been shown previously to be highly prognostic in PDAC, and forms a key molecular predictor of early recurrence in a preoperative molec-  26,27 . In the RNAseq cohort here, patients with squamous subtype body and tail pancreatic cancer had extremely poor survival (median 5⋅2 months), in comparison with the rest of the resected cohort. This suggests that such patients may be better treated with a neoadjuvant approach as occult metastatic disease may manifest over this period, and avoid futile major surgery. Because of the molecular features of aggressive disease associated with body and tail pancreatic cancer, it could also be argued that, until molecular markers of early recurrence are better defined, all body and tail tumours are better served with a neoadjuvant approach to identify patients whose tumours are likely to recur early.
It has been demonstrated here that head tumours are relatively enriched for B cell signalling (gene programme 6) and this has been shown to be associated with a better prognosis 14 . Similarly, body and tail pancreatic cancer lacks CD8-positive T cell signalling (gene programme 8), suggesting an immunosuppressive tumour microenvironment. This may well reflect an increase in myeloid cell infiltration and tumour-associated macrophage-related immunosuppression and inflammation 28 . In-depth analysis of the tumour microenvironment may reveal potential targets for novel immunotherapy agents in body and tail pancreatic cancer, including the immune checkpoint and myeloid signalling axes 28 .
It remains to be determined whether body and tail pancreatic cancer presents at a later stage of tumour evolution, or whether these tumours are biologically different and more aggressive from the outset. However, some of the present findings seem to reflect the relatively late presentation of body and tail pancreatic cancer at a more advanced stage, both clinically and molecularly (Fig. 2). First, tumours of the body and tail are larger, which may reflect a biologically older tumour. Second, body and tail pancreatic cancers correlate with molecular features that are driven by epigenetic events associated with chromosomal instability and epigenetic events that may drive intratumoral heterogeneity 29,30 . The exact sequence of these events in tumorigenesis and progression has yet to be elucidated, but may be associated with a later stage of disease evolution.
Identifying patients who will benefit from procedures with high morbidity rates such as pancreatectomy is an important task for surgeons. Personalized medicine platforms, such as Precision Panc 31 , will allow clinicians and scientists to correlate molecular profiles with clinicopathological outcomes, as well as define and refine molecular subgroups that respond to personalized treatment regimens, including surgery. Well designed clinical trials, particularly in the operable and neoadjuvant setting, will allow detailed study of the temporal and spatial clonal evolution of PDAC, and may shed light on the relationship between disease progression and the molecular timeline of tumours. This will contribute to an expanding knowledge bank of molecular and clinical data, acquired from multiple initiatives globally, and will further delineate the relationship between tumour location, stage at presentation and the molecular features of PDAC.