The molecular biology of hepatopancreatobiliary cancer

The PDAC genome

Clinically relevant subtypes of pancreatic cancer have been identified; international collaborative studies have now sequenced a large number of sizeable patient cohorts and a signature of recurrent mutated genes in pancreatic ductal adenocarcinoma (PDAC) has emerged. Alterations in the proto-oncogene KRAS are seen in > 95% patients, whilst 50% will exhibit mutations in the tumour suppressor TP53, SMAD4 (a transcription factor mediating TGFβ signalling) and the cell cycle regulator CDKN2A. ^, A larger subset of mutations is expressed at 5–10% prevalence, including genes involved in chromatin remodelling ( ARID1A , MLL3 ), DNA damage repair ( KDM6A ) and Wnt signalling ( RNF43 ). By far, the majority of alterations occur at < 1% prevalence, however, often described as the ‘long tail’ of infrequent mutations.

Metastatic pancreatic cancer is particularly refractory to systemic anticancer chemotherapy. A small proportion of patients have germline BRCA1 or BRCA2 mutations. Inhibitors of PARP (poly [adenosine diphosphate-ribose] polymerase), such as olaparib, have demonstrated activity in patients with breast and ovarian cancers with these germline mutations. In the POLO trial, maintenance olaparib provided a significant progression-free benefit to patients with a germline BRCA mutation and metastatic pancreatic cancer who had not progressed during platinum-based chemotherapy, raising the tantalising possibility that molecular targeted therapy can enhance outcome in this hitherto refractory disease.

Recurrently observed mutations have been aggregated into 10–12 core pathways including Ras, TGFβ, Wnt, Notch, ROBO/SLIT signalling, SW1-SNF, chromatin modification, DNA repair and RNA processing. This has led to a suggested ‘molecular taxonomy,’ a proposed model with which we can better understand the mechanisms underpinning carcinogenesis in PDAC. ^{, ,} However, mutations between these mechanistic groups are not mutually exclusive and therefore at the time of writing such classifications have not yet led to clinical utility or a basis for trial design ( Fig. 4.1 ).

The use of whole genome sequencing (WGS) to analyse somatic structural chromosomal rearrangements such as insertions, deletions, translocations and amplifications in a cohort of 100 samples has similarly led to a suggested classification of PDAC into four subtypes.

• ‘Stable’ tumours (20%) exhibit < 50 structural variation events and aneuploidy suggesting defective mitosis.

• ‘Scattered’ tumours (36%) show a moderate range of non-random chromosomal damage and < 200 ‘structural events’.

• ‘Unstable’ tumours (14%) display a large number (200–558) of structural events suggesting a defect in DNA maintenance (and perhaps therefore sensitivity to DNA damaging therapeutics).

• ‘Locally rearranged’ tumours (30%) are characterised by a significant focal event on one or two chromosomes (8).

Such a classification might be exploited clinically as a surrogate marker for therapeutic sensitivity. For example, cancers with an ‘unstable’ phenotype frequently exhibit mutations in BRCA1 , BRCA2 or a described mutational signature associated with the BRCA phenotype and therefore a defective DNA damage response (DDR) which might imply susceptibility to agents targeting DDR deficiency ( Fig. 4.2 ).

Mutations in KRAS have been demonstrated to be associated with poor clinical outcomes and specifically the KRAS G12D mutation with the worst survival (a single point mutation in codon 12 resulting in substitution of the amino acid glycine for aspartate and constitutive activation of downstream signalling). ^, In a cohort of 356 patients, median overall survival (OS) was 20.3 months (IQR 11.3–38.3) in the presence of KRAS mutation compared with 38.6 months (IQR 16.6–63.1) in patients with wild-type tumours. The accumulation of alterations in the other four principal driver genes ( KRAS , TP53 , SMAD4 , CDKN2A ) negatively affects clinical outcomes further. The use of multivariable Cox proportional hazard regression to analyse the effects of combinatorial gene alterations with survival concluded that inferior disease-free survival and OS is observed in patients with three or four mutations versus those with one or two alterations (after adjustment for age, sex, nodal status, tumour grade, lymphovascular invasion, receipt of perioperative treatment, resection margin status and institution).

For many years RAS was considered ‘undruggable’, but there has been a co-ordinated effort in recent times to develop novel approaches to RAS blockade including mutant-specific covalent inhibitors, ^, the first of which (AMG-510) has now entered clinical trials in non-small cell lung cancer ( https://clinicaltrials.gov/ct2/show/NCT04625647 ). Pre-clinical studies however have demonstrated immense redundancy in RAS signalling systems and experiments using pancreatic cancer cells with KRAS inhibition using gene editing with CRISPR/Cas9 retain active downstream signalling driving cell proliferation via the PI3K/MAPK pathway. To effectively treat RAS-driven cancers, the MAPK pathway must be completely supressed and combination of RAS inhibitors with additional kinase inhibitors may be required as in other cancer types, e.g. melanoma. Finally, RAS may also prove useful clinically as a biomarker, such as in the detection of early metastatic disease or recurrence. Studies assessing the utility of ‘liquid biopsy’ (the detection of circulating tumour DNA [ctDNA] in the blood) have demonstrated that detectable KRAS G12D in plasma was associated with inferior survival.

The large number of low prevalence drivers in PDAC has led to concerns that previous trials testing targeted inhibitors in unselected patient populations might have failed to detect an existing underlying benefit ‘signal’. This has led to the development of novel research platforms whereby NGS can be used to characterise tumour genomes and detect actionable targets prior to study enrolment. PRECISION-Panc in the UK and PRECISION-Promise in the USA aim to perform personalised molecular analysis and drug response studies on individual samples to inform trial recruitment i.e. ‘the right trial for the right patient’. This ‘precision oncology’ model of patient selection to enter trials according to personalised tumour sequencing has also now been adopted in other cancers including hepatocellular cancer (HCC) and biliary tract cancer (BTC). ^,

The PDAC transcriptome

In addition to characterisation of the PDAC genome, transcriptomic analyses have been conducted to assess how genetic alterations are transcribed to the RNA level and define the networks of gene expression associated with tumorigenesis. ^{, ,} Studies have examined both whole tumour and microdissected samples and observed enrichment of gene programmes that cluster into subtypes which may correlate with clinical features including histopathology and survival. These have been named in one study of 456 PDACs as ‘Squamous’, ‘Progenitor’, ‘Immunogenic’ and ‘Aberrantly Differentiated Endocrine Exocrine’ (ADEX). Other studies have identified overlapping subtypes in independent cohorts, using both actual and virtual microdissection to minimise contamination from stromal or normal pancreas tissue. ^, Further work is ongoing to corroborate the applicability of such classifications to wider cohorts and identify the potential clinical utility of gene expression analyses ( Fig. 4.3 ).

The PDAC epigenome

We now know that alterations in DNA molecules themselves (mutations) account for only a proportion of the tumour heterogeneity observed in solid cancers. Epigenetics has been defined as the study of inherited phenotypes through mechanisms that do not involve the coding capacity of the DNA sequence. Such events involve the covalent addition of methyl groups to DNA or modification to histones (highly alkaline proteins that associate with DNA to form the nucleosome and regulate gene expression), including phosphorylation, acetylation, ubiquitylation, and sumoylation. These modifications are observed on genes involved in a wide range of cellular activities including cell growth and differentiation. Some histone modifications are observed at actively transcribed gene promoters (H3K4me3, H3K27Ac), whereas others are thought to be repressive for transcription (H3K27me3, H3K9me3). In healthy development and adult tissue homeostasis, epigenetic regulation ensures that the correct gene is expressed at the correct time and in the correct context to give rise to a particular phenotype. It is therefore easy to appreciate how dysregulation of such modifications plays a role in cancer via effects on proliferation, apoptosis, senescence and invasion. Indeed, the epigenetic landscape has been shown to be profoundly altered in neoplastic cells.

Our understanding of the complexities of how the epigenome is regulated is still evolving and the mechanisms by which epigenetic changes contribute to cancer are still to be elucidated. Mutations in oncogenes, e.g. KRAS, can alter histone and DNA modifications via direct regulation of histone proteins and modifying enzymes. Somatic mutations in genes responsible for nucleosome remodelling, such as ARID1A , MLL2 and KDM6A , have been observed in PDAC. Furthermore, dysregulation of non-coding RNAs (ncRNAs; RNA species directly able to regulate the epigenome) is similarly described and some specific species (e.g. LINC00673 ) may be associated with clinical outcomes.

In PDAC, analysis of DNA methylation patterns can distinguish between transcriptional subtypes of the disease, including the ‘classical’ and ‘basal’ classification described by The Cancer Genome Atlas. Specific ‘super enhancers’ (regions of the genome comprising multiple enhancers or lengths of DNA that bind transcription factors) have been demonstrated to be driven by distinct upstream regulators resulting in altered gene expression networks between subtypes. ^, For example, high levels of the transcription factor GATA6 is associated with the classical subtype, whereas super-enhancers in PDAC exhibiting the basal subtype appear to be regulated by the hepatocyte growth factor receptor (MET). ^, Indeed, some have argued that as these phenotypes cannot be distinguished based on their somatic variants (mutations), precision medicine platforms should also incorporate RNA-seq analyses. Furthermore, epigenetic markers are able to distinguish tumours that appear identical at the genomic level, e.g., metastatic deposits and the primary tumour. Others have similarly characterised the epigenetic remodelling that occurs at different phases of disease progression; the formation of metastases is associated with global enrichment of H3K27ac chromatin. Large-scale mapping projects aim to further characterise the methylation, histone modifications and other epigenetic hallmarks of PDAC; these will require integration with genomic and phenotypic data before we fully understand how the epigenome contributes to the clinical behaviour of patients with pancreatic cancer and how best to use therapies, such as inhibitors of histone deacetylases or of DNA methylation.

Microsatellite instability and immunotherapy

Microsatellite instability (MSI) refers to a state of ‘hypermutability’ or heightened susceptibility to mutation, arising as a result of defective DNA mismatch repair (MMR). Loss of function of one or more genes encoding the MMR proteins ( MSH1, PMS2, MLH1, MSH6 ) leads to an inability to correct mistakes that occur during DNA replication, resulting in accumulation of thousands of errors and generation of microsatellite fragments (short,1-6 base pair repetitive sequences of DNA, frequently GT or CA repeats) that recur throughout the genome. For example, 15% patients with colorectal cancer have a deficiency of DNA MMR, most commonly as a result of sporadic hypermethylation of the MSH1 promoter, or due to germline mutation of the MLH1 or MSH2 genes as occurs in Lynch syndrome. Also, 1–2% of surgically resected PDACs exhibit MSI or defective DNA MMR. Lynch syndrome is associated with an 8.6-fold (C95% CI 4.7–15.7) increased risk of PDAC compared with the general population, and these cancers frequently exhibit a medullary/colloid phenotype with lymphocytic infiltrate and unlike conventional PDAC, are typically wild type for KRAS / TP53. ^, The potential benefits of surveillance for PDAC in Lynch syndrome remain unknown and since the absolute risk is thought to be relatively low, it is not advocated by current guidelines.

Much interest has been generated by the success of immune checkpoint inhibitors in colorectal cancers exhibiting MSI. These agents include humanised monoclonal antibodies targeted to the immune checkpoint receptors programmed death-1 (PD-1; e.g. pembrolizumab) or cytotoxic T-lymphocyte-associated protein 4 (CTLA-4; e.g. ipilimumab). Blockade of these receptors promotes CD8 ⁺ cell-mediated killing of cancer cells and since tumours exhibiting MSI have a high mutational burden, they are considered to have increased ‘neoantigen load’. They are therefore thought to be more susceptible to immune surveillance and hence are predicted to exhibit an enhanced response to immune checkpoint inhibition. Initial trials of these agents as monotherapy in PDAC have produced mixed results, ^, although the more recent multicentre KEYNOTE-158 study which included exclusively patients with MSI high tumours reported a RECIST overall response rate (ORR) of 18.2% (95% C.I. 5.2–40.3) in advanced PDAC. Further studies investigating checkpoint inhibitors in combination with other immunomodulators are proving to offer more promise. The COMBAT trial was a phase IIa, dual cohort study evaluating the safety and efficacy of the checkpoint inhibitor pembrolizumab in combination with an antagonist of the chemokine receptor 4 (CXCR4), BL-8040. Patients were given the above combination either in isolation or with chemotherapy, observing an ORR of 32% when combined with cytotoxic chemotherapy. It should be noted that subjects in this trial had microsatellite stable disease, and further randomised trials are needed to fully evaluate the role for immunotherapy in PDAC ( Fig. 4.4 ).

Pancreatic intraepithelial neoplasias

Many epithelial adenocarcinomas are known to arise from non-invasive precursor lesions which accumulate genetic and epigenetic changes. PDAC most commonly arises from precursor epithelial proliferations known as pancreatic intraepithelial neoplasias (PanINs), which have been classified as ‘low grade’ (PanIN-1, PanIN-2) or ‘high grade’ (PanIN-3), exhibiting increasing features of cytological atypia, and dissection of pancreata harbouring PDAC will very commonly reveal multiple advanced PanINs. ^, The majority of PanINs harbour KRAS mutations and as lesions progress into invasive carcinoma, other driver alterations are accumulated later ( TP53, SMAD4, CDKN2A ). Genetic studies sequencing PanINs and PDAC from the same pancreas have shown that there can be clonal expansion of a single ancestral cell that gives rise to more than one lesion which then accumulate additional driver mutations to form an invasive cancer.

Less commonly PDAC arises from cystic lesions of the pancreas: intraductal papillary mucinous neoplasms (IPMNs) or mucinous cystic neoplasms (MCNs) (see Chapter 18 ). As is seen with PanINs, IPMNs can be multifocal and can involve the entire pancreatic duct. The majority follow an indolent course, with a 1-, 5- and 10-year survival rate of 100%, 100% and 94.2%, respectively. The role of cancer susceptibility genes in IPMN and MCN remains poorly understood.

Biomarkers

Genetic analyses of metastatic PDAC deposits compared with their primary tumour indicate that there is considerable latency between the appearance of the primary malignant clone and seeding of the index metastasis (6.8 years on average in one study). This suggests significant opportunity for the early detection of pancreatic cancer; however, unfortunately a sensitive and specific serum assay for the diagnosis and prevention of early PDAC has not yet been developed.

Serum carbohydrate antigen (CA) 19-9 is a sialylated Lewis blood group antigen produced by PDAC cells and is the most commonly used biomarker worldwide. Approximately 6% of the White population and 22% of the Black population in the USA do not generate the specific sialyl antigen and do not produce CA19-9. Furthermore, CA19-9 can be raised in biliary obstruction, pancreatitis and extra-pancreatic malignancies and does not have sufficient specificity or sensitivity to act as a screening tool for early detection in asymptomatic populations. ^, A meta-analysis of 11 studies encompassing 2316 individuals estimated a pooled sensitivity of 0.80 (95% CI 0.77–0.82), specificity of 0.80 (95% CI 0.77–0.82) and area under the curve of 0.87 for the diagnostic accuracy of CA19-9 in patients with PDAC. CA19-9 is therefore used rather for disease monitoring and levels have been shown to predict tumour stage, resectability, OS and response to therapy.

Carcinoembryonic antigen (CEA), a glycoprotein widely used for disease monitoring in colorectal cancer, is also used as a biomarker for pancreatic cancer. CEA has an inferior sensitivity/specificity profile, cited by one meta-analysis to be 0.43 (95% CI 0.39–0.47) and 0.82 (95% CI 0.79–0.84), respectively, and consequently CA19-9 remains the only approved serum diagnostic biomarker. CEA is however frequently measured as part of fluid analysis as part of endoscopic evaluation of pancreatic cysts. A threshold of 192 ng/dL is widely used as a cut-off for diagnosis of mucinous pancreatic cystic lesions; however, CEA levels cannot distinguish benign from malignant cystic lesions. ^,

Research into novel biomarkers has moved towards assessing the utility of ctDNA and panels of multiple blood-based markers as diagnostic ‘signatures’ to boost test sensitivity, as well as the role of microRNAs and genomic or transcriptomic data from biopsies to predict response to therapy.