The Impact of Molecular Subtyping on Pathological Staging of Pancreatic Cancer

Background: The long-term outcomes following surgical resection for pancreatic ductal adenocarcinoma (PDAC) remains poor, with only 20% of patients surviving 5 years after pancreatectomy. Patient selection for surgery remains suboptimal largely due to the absence of consideration of aggressive tumor biology. Objective: The aim of this study was to evaluate traditional staging criteria for PDAC in the setting of molecular subtypes. Methods: Clinicopathological data were obtained for 5 independent cohorts of consecutive unselected patients, totaling n = 1298, including n = 442 that underwent molecular subtyping. The main outcome measure was disease-specific survival following surgical resection for PDAC stratified according to the American Joint Commission for Cancer (TNM) staging criteria, margin status, and molecular subtype. Results: TNM staging criteria and margin status confers prognostic value only in tumors with classical pancreatic subtype. Patients with tumors that are of squamous subtype, have a poor outcome irrespective of favorable traditional pathological staging [hazard ratio (HR) 1.54, 95% confidence interval (CI) 1.04–2.28, P = 0.032]. Margin status has no impact on survival in the squamous subtype (16.0 vs 12.1 months, P = 0.374). There were no differences in molecular subtype or gene expression of tumors with positive resection margin status. Conclusions: Aggressive tumor biology as measured by molecular subtype predicts poor outcome following pancreatectomy for PDAC and should be utilized to inform patient selection for surgery.

Transcriptomic molecular subtyping of PDAC has consistently identified a subgroup, termed squamous (also known as basal) subtype, characterized by epigenetic changes that drive immune evasion and epithelial-to-mesenchymal transition as compared to the classical pancreatic subtype. [3][4][5] The distinct molecular features underpinning the squamous subtype is associated with aggressive tumor biology and a poorer outcome. Although these features may potentially provide novel targets for subtype-specific vulnerabilities and ultimately direct therapy for PDAC, they are not yet utilized clinically to inform prognosis. 3,6 Pathological staging of resected PDAC has been enhanced through modifying the American Joint Commission for Cancer (AJCC) staging system (8 th Edition), 7 and resection margin involve-ment is established as a powerful predictor of poor outcome. [8][9][10][11] Although the prognostic value of traditional pathological staging criteria including margin status 9 has been validated in several studies, 7,12 these traditional histopathological features have not been rigorously assessed according to molecular subtype.
Incorporating molecular characterization of PDAC is vital for individualized outcome prediction, therapy allocation, clinical trial eligibility, and to facilitate result comparison across studies and institutions. In this study, using 5 independent highly annotated multi-institutional cohorts, we sought to validate the AJCC 8 th edition staging criteria, and determine the value of resection margin involvement in the context of established PDAC molecular subtypes, in patients following resection for pancreatic cancer.

Patient Cohort Description
Clinicopathological and complete outcome data were obtained from prospectively maintained independent cohorts of patients with resected PDAC. Patients were accrued prospectively for the Australian Pancreatic Cancer Genome Initiative (APGI) cohort (www.pancreaticcancer.net.au) as part of the International Cancer Genome Consortium (ICGC; www. icgc.org). 13

Neoadjuvant and Adjuvant therapy regimens
For patients that received neoadjuvant therapy (NAT), a modified FOLFIRINOX regimen was administered to patients with good performance status (PS) (0). For those patients with poorer PS (1), gemcitabine either as monotherapy or in combination was administered. Dose reductions or delays were instituted at the discretion of the medical oncology team. When given, chemoradia-tion (CRT) with 50.4 Gy and GemCAP were administered. 14 Following resection, adjuvant therapy was administered if patient performance status allowed it and the regimen was left to the discretion of the treating oncologist based on local and international guidelines. Some patients were enrolled in previous randomized trials (ESPAC trials) and included in previous studies. 5,[15][16][17][18][19][20][21][22][23]

Pathology Assessment
Patients with resected PDAC were staged according to the AJCC 8 th staging criteria, with T-stage based on maximum tumor diameter, and N-stage determined by the number of positive lymph nodes har-vested. 7,12 Margin involvement (R1) was defined according to the Royal College of Pathologists criteria as the presence of tumor at or < 1mm (R1 < 1mm) of a margin or surface when assessed by microscopy of a hematoxylin and eosin stained slide. This criterion has proven capable of discriminating outcome following resection. [9][10][11] For the APGI and Verona cohorts, margin status was originally defined as evidence of tumor at any margin or surface (R10mm). Where possible, these were re-staged according to R1 < 1mm with a separate analysis performed for each margin criteria (R1 0mm and R1 < 1mm ) for these cohorts.

Transcriptomic Profiling
The molecular subtyping criteria was generated as part of the ICGC landmark study of PDAC. 5 RNA was extracted from bulk tumor and profiled using RNA sequencing (RNAseq) and gene expression microarrays as previously described. 5 Selecting patient samples to undergo sequencing was based on a number of factors, including cost, tissue quality, and tumor cellularity. All samples were fresh frozen upon collection. Tumors with cellularity < 40% (n = 249) underwent gene expression micro-array analysis, whereas RNA sequencing was performed in those tumor specimens with cellularity > 40% (n =193) since there is strong evidence to suggest that bulk tumor RNA profiling technologies are comparable. 24 Individual tumors were classified as either squamous or classical pancreatic subtypes. The classical pancreatic subtype encompassed the pancreatic progenitor, aberrantly differentiated endocrine exocrine (ADEX), and immunogenic subclasses described by Bailey et al. 3 Tumors underwent molecular subtyping from 4 independent cohorts. The APGI (n = 90) and Verona (n = 103) cohorts underwent RNA sequencing (total n = 193). The remaining patients (n = 249) from the APGI (n = 174), Glasgow (n = 47), and Milan (n = 28) cohorts underwent validated targeted RNA expression and micro-array gene expression analysis 5 . Differential gene expression was performed using the standard pipeline from the Bioconductor [Bioconductor.org] package "limma"in groups with positive and negative resection margins. Expression counts were processed and normalized as previously described 5 . S100A2 and S100A4 protein expression were used as surrogate immunohistochem-ical (IHC) biomarkers of the squamous subtype as previously described. 25 Briefly, high S100A2 expression was defined as cytoplasmic staining with intensity 3+ in > 30% of cells and positive S100A4 expression was defined as either nuclear and/or cytoplasmic staining of any intensity in > 1% of cells. 25

Statistical Analysis
Categorical variables were compared using the chi-square test. The Mann-Whitney U test was used to compare continuous variables. The principal outcome measure was length of disease specific survival (DSS) as measured from the time of original surgery, or commencement of NAT. Patients alive at the time of follow-up point were censored. The last follow-up period for patients still alive was October 2020. KaplanMeier survival analysis was used to analyze the DSS. To compare the length of survival between curves, a log-rank test was performed. A Cox proportional hazards model was used for univariate analysis to adjust for competing risk factors, and the hazard ratio (HR) with 95% confidence intervals (CIs) was reported as an estimate of the risk of DSS. Variables found to be significant on univariate analysis at P < 0.10 were included in multivariate analysis in a backwards stepwise fashion. Statistical analysis was performed using SPSS (Version 25.0; IBM SPSS Statistics, IBM Corporation, Armonk, NY) and R 3.4.0 (The R Project for Statistical Computing, Vienna, Austria).

Predicting Outcome According to Molecular Subtype
The squamous subtype was associated with significantly worse DSS (median survival 14.9 vs 26.5 months, P < 0.001) (Fig. 1)   The impact of S100A2 and S100A4 protein expression on pathological staging was investigated in patients from the molecular subtype cohort that had immunohistochemistry data available to assess the utility of simple IHC biomarkers of the squamous sub-type 25 . S100A2and S100A4 expression did not have the same impact on margin and nodal status as full molecular subtyping using gene expression (supplementary figure 6, http://links.lww.com/SLA/ D251), with both margin and lymph node status remaining prognostic despite expression of both biomarkers. associated with molecular subtype (Fig. 3). Differential gene expression analysis from the APGI molecular subtype cohort demonstrated no discernible difference between tumors defined as resection margin positive or negative (Fig. 3). RNAseq and gene expression microarray analysis demonstrated no significant differences in gene expression between the groups, even when R1 0mm status was used.

DISCUSSION
Comprehensive characterization of pancreatic cancer resection specimens should include evaluation of molecular profile. In 1298 resected PDACs we validated the prognostic value of the AJCC 8 th staging system incorporating detailed margin status annotation. Yet stratification according to molecular subtype (aggressive disease biology), confounded the FIGURE 1. A, Transcriptomic profiling strategy of the molecular subtype cohort. Transcriptomic analysis was performed using either RNA sequencing (RNAseq) or gene expression microarray based on cellularity and adequate RNA quality of the sample in a selection (n = 442) of PDACs. Tumor cellularity > 40% allowed whole genome sequencing, whilst RNAseq was performed in tumors with sufficient cellularity ( > 40%) and quality RNA. Kaplan-Meier survival curves for all patients stratified by AJCC 8 th edition staging criteria for B) T-stage, C) N-stage, and D) margin status and E) molecular subtype.
prognostic value of this pathological staging system, and negated the impact of resection margin involvement. This suggests that biological factors including molecular subtype convey significant prognostic value with potential to impact the personalization of surgical management algorithms.
The AJCC 8 th staging criteria has been validated in multiple unselected cohorts of resected PDAC, significantly improving prognostication for patients. 7,12 This is particularly pronounced with the updated N-stage criteria discriminating according to lymph node metastases burden, however,  modifications have already been proposed to further enhance outcome prediction. 27 These studies remain limited, as there has been a failure to account for heterogeneity driven by tumor biology and molecular determinants of disease outcome.
Transcriptomic subtyping has transformed our understanding of the molecular taxonomy for most cancers. 3 The existence of two distinct PDAC subtypes has been demonstrated and validated in numerous classifiers, suggesting the concept of opposing lineages is robust. 3,5,[28][29][30][31][32][33] Previous studies have demonstrated the squamous molecular subtype to be associated with poor prognosis, whereas a simplified protein-based expression of a squamous biomarker predicts poor outcome following upfront pancreatectomy. 4,5 No previous study has investigated the impact of molecular subtype on pathological staging criteria such as margin or lymph node status, which makes the results presented in this study novel.
Patients with the squamous molecular subtype had a significantly reduced median DSS compared to patients with the classical subtype. We have demonstrated that molecular transcriptomic analysis using a variety of techniques can robustly subtype patients from independent institutions following resection of PDAC. Resection margin status, particularly R1 < 1mm, is a critical independent predictor of outcome following pancreatectomy for PDAC, including in prospective clinical trial analysis 9,10,34 . Our results suggest that for patients with the aggressive squamous molecular phenotype, both lymph node involvement and R1 status fail to impact prognosis. Presumed metastatic dissemination, occurring early in pancreatic carcinogenesis, driven by particular molecular features in the squa-mous subtype may explain this finding. Although the mechanisms require elucidation, these results have significant implication for clinical trial design and interpretation particularly if R1 status is employed as a surrogate endpoint. The current results confer support for the squamous molecular subtype being regarded as a disease entity distinct from the classical pancreatic subtype, particularly in combination with preliminary evidence that suggests chemothera-peutic response differs according to transcriptomic subtype. 35 It appears that the classical pancreatic subtype is the default molecular lineage with evolution into squamous subtype occurring in some patients. 32,36 Previous work from our group suggests that the squamous molecular subtype is more frequent in PDAC originating within the pancreatic body and tail. 26 Whether this is determined early in carcinogenesis or if the squamous subtype simply reflects molecular evolution requires elucidation. Interestingly, the classical pancreatic subtype had a higher frequency of perineural invasion (88% vs 81%, P = 0.049). The reason for this is not clear from this study but may be that classical tumors are more likely to invade locally and cause local recurrence. Whereas squamous tumors are more likely to cause early distant, hepatic recurrence and may explain its association with lymphovascular invasion (59% vs 47%). In this study, we used bulk tumor samples for gene expression analysis and subtyping. Recently, several studies have demonstrated that molecular subtype can exist on a spectrum within the same tumor. 32,36 Bulk tumor transcriptomic sequencing likely classify tumors based on the dominant subtype, yet further study is required to investigate the extent of transcriptomic subtype intratumoral heterogeneity and how this impacts clinical outcome. 36 Our results demonstrated no difference in the gene expression of tumors deemed margin positive versus those that were margin negative. Resection margin status is often viewed as a surrogate of aggressive tumor biology; however, in this cohort, there were no significant transcriptomic differences. This may result from margin status being influenced by other biological factors, for example, tumor microenvironment composition or immune infiltration. Spatial transcriptomic characterization has the potential to elucidate subtype heterogeneity and impact of the microenvironment on margin status. 36,37 Ultimately, margin status, may simply reflect anatomical location of the tumor, rather than a surrogate for tumor biology. This is particularly apparent for the squamous subtype, where local control by clear margins appears to be less crucial due to early systemic dissemination in these patients. There is some evidence that margin involvement in proximity to vasculature or neural plexus is associated with local recurrence. 10 Prospective studies, with comprehensive molecular and pathological characterization, are necessary to further elucidate this.
Based on results presented here, and by others 4,5 , we support the concept that comprehensive staging for a patient diagnosed with PDAC, particularly in the potentially operable setting [resectable, borderline resectable (BR) and locally advanced (LA)], should integrate biological predictors of disease prognosis. 38 To date, attempts have been made to identify aggressive tumor biology utilizing tumor markers (CA19-9) or tumor stage, without accounting for the molecular features that make up and drive each individual tumor. 39 Molecular characterization of the patient and tumor, both at the transcriptomic and genomic levels, at the time of diagnostic biopsy may enable better selection of patients for resection, optimizing high-risk surgical management strategies for patients with BR and LA PDAC. 4 The natural progression of health care, and cancer treatment, trends toward a precision medicine strategy where therapy selection aligns with individual and tumor features. Despite development of novel personalized molecular and histological tools that evaluate tumor biology, 4 these metrics have as yet failed to integrate into clinical practice. For PDAC, the evolution toward a precision oncology strategy is driven by global initiatives including PRECISION-Panc in the United Kingdom, which aim to harness molecular variation to guide therapy. This study suggests that using only pathological staging as a predictor of postoperative outcome fails to encompass the biological attributes of the tumor. This, in turn will impact the utility of clinical and pathological staging, particularly in the setting of clinical trials where survival or margin status is often used as study endpoints. We propose that molecular subtyping is determined in these settings to accurately compare new treatment regimens and avoid unaccounted biases in clinical studies and prospective trials.
This study is limited firstly as the majority of molecularly subtyped tumors were not treated with NAT, an increasingly common strategy for managing PDAC, and thus the impact of response to NAT is not considered. A previous study has suggested that neoadjuvant FOLFIRINOX may induce evolution from the classical to the squa-mous subtype, 40 yet we were unable to investigate this here. There is growing evidence that the squamous (or basal) subtype is less likely to respond to current chemotherapeutic regimes than Classical Pancreatic subtype. 35 In this cohort, the squamous subtype were less likely to receive adjuvant chemotherapy which the authors believe demonstrates likely significant micro metastatic disease burden leading to poor performance status post pancreatectomy, which in turn impedes adjuvant therapy allocation. Circulating biomarkers such as CA19-9 and circulating tumor cells (CTCs) were not available for a large proportion and thus could not be assessed. The relationship between molecular features, therapeutic response and prognosis is complex and remains to be determined.

The
PRIMUS-002 neoadjuvant trial (ISRCTN34129115), part of the PRECISION-Panc platform (ISRCTN14879538), is currently recruiting, and aims to delineate the interaction of genomic and transcriptomic molecular subtype, circulating biomarkers (CA19-9, CTCs and cell free DNA), therapeutic response and prognosis in the setting of nonmetastatic PDAC. 6 This prospective trial will provide molecular subtype characterization based on preoperative biopsies and resection specimens, which will allow investigation of the impact of neoadjuvant therapy on molecular subtype evolution and heterogeneity.

CONCLUSIONS
These results demonstrate that biological characteristics as determined by transcriptomic subtype are a strong predictor of outcome following pancreatectomy forPDAC. Standard pathological staging criteria, particularly margin status, in PDAC failed to predict outcome in patients with tumors of the squamous molecular subtype. This indicates that tumor biology should be accounted for when staging patients following surgical resection of PDAC, particularly in the setting of clinical trials, as these features have potential to personalize both treatment allocation and prognosis. We envisage soon that transcriptomic subtype, in addition to genomic characterization, will be determined preoperatively and facilitate patient centered algorithms to improve outcomes for patients with PDAC.