DNA Damage Repair Deficiency

Around half of all high-grade serous ovarian cancers have defects in genes related to the normal repair of DNA,1 leaving them susceptible to the DNA cross-linkage caused by platinum agents or to synthetic lethality via poly(ADP-ribose) polymerase (PARP) inhibition. To appreciate how these gene aberrations impact on high-grade serous ovarian cancer, it is first important to appreciate the normal functioning of the individual pathways.

Normal DNA Damage Repair

As shown in Figure 1, single-strand breaks and double-stand breaks to the DNA double helix are repaired via a series of distinct yet interconnected mechanisms known as the DNA damage response pathways. Single-strand breaks are normally repaired by base excision repair, nucleotide excision repair, or mismatch repair. Base excision repair primarily repairs small chemical alterations of bases and is therefore particularly important in the prevention of mutagenesis. The damaged bases are excised from the double helix and replaced with newly synthesized DNA.2 PARP1 is a nuclear enzyme crucial within the base excision repair pathway; after binding to single-strand breaks, it converts nicotinamide adenosine diphosphate (NAD+) into adenosine diphosphate (ADP)-ribose polymers, allowing the recruitment of DNA repair and scaffolding proteins.3

Figure 1

DNA damage repair pathways. Single-strand breaks to DNA are repaired by base excision repair, nucleotide excision repair, or mismatch repair. Base excision repair primarily repairs small chemical alterations of bases, nucleotide excision repair eliminates helix-distorting damage that affects one of the two DNA strands, and mismatch repair repairs nucleotide mutations, substitutions, deletions, and insertions. Double-stand breaks are repaired through the processes of homologous recombination, non-homologous end joining, and microhomology-mediated end joining. In homologous recombination, a section of the DNA sequence around the double-strand break is resected and new DNA is synthesized using the homologous sister chromatid as the template; non-homologous end joining mediates repair by directly ligating the ends of a double-strand break together, and in microhomology-mediated end joining there is alignment of microhomolgous sequences internal to the broken ends before joining.

Nucleotide excision repair eliminates helix-distorting damage that affects one of the two DNA strands, such as that caused by platinum agents. It is a multistep ‘cut-and-patch’ reaction involving more than 30 proteins, whereby the damaged sequence is excised and filled in using the intact complementary strand as a template.4

Mismatch repair targets nucleotide mutations, substitutions, deletions, and insertions that occur either during DNA replication or as a consequence of oxidative DNA damage. The first step in this pathway is the recognition of mismatched DNA by MutS, which binds to the affected region and forms a homodimer which varies depending on the nature of the damage; the MutSα-complex (formed with either MSH2/MSH6) recognizes single base mismatches and short insertion-deletion loops while the MutSβ-complex (formed with either MSH2/MSH3) recognizes larger loops. These homodimers subsequently recruit MutLα (formed by MLH1 and PMH2) which mediates the excision and re-synthesis of the damaged strand.5

Double-strand breaks are repaired through the processes of homologous recombination, non-homologous end joining, and microhomology-mediated end joining. In homologous recombination, a section of the DNA sequence around the double-strand break is resected and new DNA is synthesized using the homologous sister chromatid as the template. Non-homologous end joining mediates repair by directly ligating the ends of a double-strand break together and can often cause deletion or mutation of DNA sequences near the double-strand break.2 In microhomology-mediated end joining, there is alignment of microhomolgous sequences internal to the broken ends before joining; this pathway is error prone and associated with deletions.6

PARP Inhibition

Homologous recombination is a high-fidelity pathway and in the absence of its normal function, there is an increased reliance on the error-prone non-homologous end joining and microhomology-mediated end joining pathways. Homologous recombination deficient (HRD) cancer cells are particularly sensitive to PARP inhibition as by blocking the normal function of PARP1 and PARP2 there is a failure of the base excision repair pathway, leading to an accumulation of persistent single-strand breaks that ultimately progress to double-strand breaks which the cells are unable to repair, resulting in loss of genomic integrity. As shown in Table 1, a number of PARP inhibitors are now licenced for the treatment of epithelial ovarian cancer, with choice of agent guided by genomic biomarkers.

Table 1

Key studies exploring the use of PARP inhibitors in the treatment of epithelial ovarian cancer

Identification of Homologous Recombination Deficient CancersPlatinum Sensitivity

Platinum sensitivity is in itself a strong biomarker for HRD as evidenced in the initial PARP inhibitor studies such as Study 19, where responses were observed in patients with platinum-sensitive disease regardless of their BRCA or HRD status.7 This is a relatively crude method and it is worth noting that PARP and platinum sensitivities do not always overlap.8 9 The current approach to identifying patients most likely to benefit from PARP inhibition involve either recognizing potential causes of HRD or quantifying the effects of HRD on the genome.

Homologous Recombination Gene Tests

BRCA1 and BRCA2 are essential for homologous recombination, but not non-homologous end joining or microhomology-mediated end joining, and if either are deficient, double-strand breaks are repaired through the more mutagenic mechanisms.10 BRCA1/2 mutations are found in approximately 20% of high-grade serous ovarian cancers, with the majority being due to germline loss, although an additional 5–7% of high-grade serous ovarian cancers harbor a somatic mutation.1 BRCA1/2 mutations are highly predictive of response to PARP inhibition, as has been repeatedly demonstrated across multiple clinical trials since the initial Study 1911 (Table 1), with the recent 7-year data from SOLO1 study confirming that patients with BRCA1/2 mutations have an improved survival with maintenance olaparib versus placebo in the first-line setting.12

In addition to BRCA1/2, mutations in other genes involved in homologous recombination are observed in around 30% of high-grade serous ovarian cancers.1 These include RAD51C/D, BRIP1, PALB2, CHEK1, CHEK2, and several Fanconi anemia genes.2 However, not all of these genes are as essential to the successful execution of homologous recombination, and the contribution of individual mutations to an HRD phenotype is not well understood at present. In the exploratory analysis of ARIEL2 parts 1 and 2, RAD51C/D were the only non-BRCA genes associated with a response to PARP inhibition, with comparable response rates to rucaparib (71.4%; 29% to 96.3%) to that of BRCA1 (39.3%; 29.1% to 50.3%), and BRCA2 (57.1%; 42.2% to 71.2%),13 although caution is required with interpretation of this dataset as the numbers involved were small (n=7).

Outside BRCA1/2 and perhaps RAD51C/D, single gene mutations are of limited use in predicting response to PARP inhibition. Attempts to find a group of genes, which together may result in HRD, have been unsuccessful to date. As part of the translational analysis of the PAOLA-1 trial, panels of homologous recombination genes (which excluded BRCA) were not predictive of benefit to maintenance olaparib with bevacizumab.14

In addition to germline and somatic mutations, gene function can be influenced by epigenetic silencing. In epithelial ovarian cancer, gene promotor hypermethylation has been observed in RAD51C (2%)1 and BRCA1 (15%) with loss of protein expression,15 resulting in higher levels of genomic scarring associated with HRD.16 Evidence regarding the accuracy of HRD gene promoter methylation as a predictive biomarker for PARP inhibitor (and platinum) response in high-grade serous ovarian cancer has been inconsistent and unreliable.17–19 However, more recently it has become apparent that earlier studies were affected by technical factors and that the zygosity of BRCA1 methylation is a key factor in predicting PARP inhibitor sensitivity. Both copies of BRCA1 must be methylated for response to PARP inhibitors, and loss of a single copy can restore homologous recombination leading to PARP inhibitor resistance.20 As yet, it is unknown whether the same requirements for methylation zygosity applies to RAD51C methylated cases. Caution is required when assigning methylation status of homologous recombination genes, and gene copy number is critical for accurate HRD assessment. Furthermore, our current assays using next generation sequencing do not detect epigenetic modification.21

Genomic Scar Assays

While a gene signature that can predict HRD in BRCA wildtype high-grade serous ovarian cancer remains elusive, there are a number of approaches in quantifying the consequences of HRD. When homologous recombination is impaired, there is an accumulation of somatic mutations and somatic copy number alterations, which result in genomic scarring. This is a permanent footprint that reflects cumulative DNA repair deficiency, regardless of the underlying etiology.22 Genomic scarring can be quantified using a number of metrics, including loss of heterozygosity (LOH),16 telomeric allelic imbalance (TAI),23 and large-scale state transition (LST),24 all three of which, when quantified, correlate with HRD status and to each other.25

To date, two commercial assays have been prospectively validated in clinical trials exploring the role of PARP inhibition in epithelial ovarian cancer: FoundationOne CDx Assay, and MyChoice CDx Myriad Assay (Table 1). The FoundationOne CDx assay quantifies the percentage of the genome affected by LOH using copy number profiling and minor allele frequencies across 3500 single nucleotide polymorphisms. This assay was used in ARIEL2, with the pre-defined cut-off between LOH low and high being 14%, which was based on TGCA microarray data and post-platinum survival.26 MyChoice CDx Myriad Assay quantifies LOH, TAI, and LST separately before combining them to generate a gene instability score (GIS), with a score ≥42 being deemed HRD.27 Although both these assays do not demonstrate full equivalence, there is significant concordance between them.28

There are, however, a number of issues with identifying HRD patients via genomic scarring. First, the use of a pre-defined cut-off for identifying HRD results in some patients being misclassified. For example, 5% of BRCA mutated tumors have a GIS <42,27 a score that on its own would infer homologous recombination proficiency. To account for this, the VELIA study adopted an unvalidated GIS score of ≥33 to define HRD tumors following retrospective analysis of previous studies. Using this lower cut-off, this study failed to show a clinical benefit in the BRCA wildtype/HRD subgroup analysis, with median progression-free survival for veliparib versus placebo of 22.9 months versus 19.8 months (0.74; 0.52 to 1.06)29; however, the absence of platinum sensitivity guiding patient selection may also have contributed to this outcome. Second, genomic scar assays provide a static snapshot of a metric that is continuously changing. As observed in the ARIEL2 study, patients who were previously homologous recombination proficient can become deficient overtime; however, once a tumor is classed as HRD the genomic scars never revert.26 Given that homologous recombination proficiency is restored in the presence of BRCA1/2 reversion mutations (which are present in up to 22% BRCA1 and 30.7% BRCA230 patients who develop PARP resistance), the power of these assays decreases with time and subsequent lines of treatment. Thirdly, current genomic scarring assays have failed to reliably detect which patients will fail to benefit from PARP inhibition. Clinical benefit is still derived in homologous recombination proficient patients, although to a lesser extent than in HRD cases as demonstrated in trials such as PRIMA and ATHENA (Table 1). In SOLO1, an exploratory analysis of HRD status revealed no difference in degree of benefit to olaparib in those defined as LOH high (0.29; 0.15 to 0.58) and LOH low (0.29; 0.20 to 0.43).31 The only exception to this observation is PAOLA-1, where homologous recombination proficient patients derived no significant benefit (1.00; 0.75 to 1.35); however, the addition of bevacizumab within this study may be a compounding factor. Finally, these assays are expensive and require central analysis, two factors which are driving the development of academic assays such as AmoyDx HRD Focus Panel32 and Leuven HRD test.33 Alongside these institutional efforts, there is an international academic effort underway to identify more reliable, cost-effective HRD assays using samples from the PAOLA-1 trial.

Functional Assays

As mentioned above, genomic scars are permanent and do not reflect tumor evolution and changes in homologous recombination status over time. Functional homologous recombination assays aim to overcome this limitation by providing a dynamic evaluation of the current homologous recombination status. They aim to do this by quantifying a downstream event that would infer either the successful or unsuccessful execution of all upstream events. One such assay currently in development focuses on RAD51, which typically forms foci at sites of DNA damage. By quantifying the ability or inability to accumulate RAD51 in these regions, the assay generates a read-out which is predictive of PARP and platinum sensitivity.34–36 When compared with MyChoice CDx, one RAD51 functional assay had a failure rate of 30%, but did detect additional HRD patients37 not picked up by GIS scoring. While this technology is promising, it still requires additional validation in prospective studies. Furthermore, these assays are often reliant on fresh tumor samples and can be technically difficult to perform, and therefore are not yet ready for widespread clinical use, although they hold great promise.

Mismatch Repair Deficiency

Hereditary non-polyposis syndrome (HNPCC or Lynch syndrome) is an autosomal dominant cancer family syndrome involving defective mismatch repair genes, accounting for around 2% of all epithelial ovarian cancers.38 As with BRCA, loss of mismatch repair gene function can be due to both somatic mutation or epigenetic silencing.39 Although this represents a small population of patients with high-grade serous ovarian cancer, mismatch repair deficient tumors have a unique immune microenvironment, characterized by increased CD3+ and CD8+ tumor infiltrating lymphocytes and heightened programmed death-ligand 1 (PD-L1) expression,40 and may benefit from treatment with the anti-PD1 drug pembrolizumab, which has tissue agnostic approval for patients with mismatch repair deficient tumors. Mismatch repair status is not routinely screened for in high-grade serous ovarian cancer; however, testing should be considered in the recurrent setting, particularly in younger patients with strong family histories given the potential therapeutic implications.

Nucleotide Excision Repair Deficiency

Alterations in the nucleotide excision repair pathway is found in around 8% of high-grade serous ovarian cancer, with a subset of these patients (such as those with GTF2H5 downregulation) displaying a platinum-sensitive phenotype similar to that of BRCA mutated tumors, with an improved overall and progression-free survival.9 41 At the moment, nucleotide excision repair is not routinely evaluated when considering treatment choice for patients with high-grade serous ovarian cancer; however, with the development of drugs which target the pathway, such as trabectedin,42 identifying genomic biomarkers within it to better guide patient selection will become increasingly important.

Microhomology-mediated End Joining

There is increasing interest in the microhomology-mediated end joining pathway as part of the treatment paradigm of high-grade serous ovarian cancer. A large proportion of post-PARP or platinum BRCA reversion mutations are felt to be mediated by microhomology-mediated end joining43 and are associated with decreased clinical benefit from PARP inhibition,44 with circulating tumor DNA being used to track the evolution of these in real-time. By inhibiting the action of POLθ (DNA polymerase theta), a DNA polymerase within the microhomology-mediated end joining pathway, there is an increase in death in cells with HRD45 suggesting that POLθ inhibition may play a role in enhancing PARP response and delaying or preventing resistance.

Copy Number Variations

While early and late-stage high-grade serous ovarian cancer have similar patterns of mutation and somatic copy number alterations, differences are observed in copy number signature analysis, with late-stage disease having changes consistent with whole exome duplication.46 In late-stage disease, these signatures are predictive in terms of survival and treatment response, with signatures 3 (BRCA related HRD) and 7 (non-BRCA related HRD) having a good overall survival whereas signature 1 (RAS/MAPK signaling) is predictive of platinum-resistant relapse and poor overall survival.47 Copy number signatures have the potential to guide a personalized framework that could help guide treatment choices towards multiple mutational processes; however, they are only detectable with whole exome sequencing,

Long Non-coding RNA

Long non-coding RNAs (lncRNAs) have the potential to augment our understanding of cancer biology and assist in cancer detection and prognosticating as exemplified by lncRNA PCA3, which has been approved as a diagnostic biomarker in prostate cancer.48 In high-grade serous ovarian cancer, the lncRNAs DNM30S, MEG3, and MIAT are critical in epithelial-to-mesenchymal transition of ovarian cancer, with DNM30S overexpression being associated with a worse overall survival with links to cell migration and invasion.49