Homologous recombination deficiency and host anti-tumor immunity in triple-negative breast cancer

Reported in 2005, BRCA1- and BRCA2-deficient cells were found to be markedly sensitive to inhibition of PARP in contrast to those cells that were wild-type or heterozygous for BRCA1/2, implicating the potential for synthetic lethality [10, 11]. Based on these observations, proof-of-concept studies tested PARP inhibition in advanced BRCA1 and BRCA2 mutation-associated breast and ovarian cancer. Among 27 heavily pretreated BRCA1/2 mutant breast cancer subjects who received olaparib 400 mg orally twice daily, Tutt and colleagues reported an objective response rate of 41% and a clinical benefit rate of 85% [12]. Responses were seen in both BRCA1 and BRCA2 mutation carriers and were irrespective of breast cancer subtype.

Around the same time, a small proof-of-concept trial explored the activity of single-agent cisplatin 75 mg/m² IV every 3 weeks for 4 cycles as neoadjuvant therapy for BRCA1 mutation-associated breast cancer and demonstrated 61% pathologic complete response [13]. However, in the neoadjuvant GeparSixto clinical trial of weekly paclitaxel, weekly non-pegylated liposomal doxorubicin and bevacizumab with or without weekly carboplatin, BRCA1/2 mutation carriers did not have higher pCR rates with the addition of carboplatin [14]. Response rates to standard chemotherapy in BRCA1 and BRCA2 mutation carriers are not well known. A retrospective analysis of BRCA1/2 mutation carriers at a single center reported a pCR rate of 46% in BRCA1 mutant versus 31% BRCA1/2 wild-type TNBC patients treated with anthracyclines with or without taxanes [15]. Another retrospective series reported a pCR rate of 67% in BRCA1/2 mutant versus 37% in wild-type TNBC patients [16]. In the frontline metastatic TNBC setting, the TNT trial compared single-agent carboplatin versus single-agent docetaxel for 6 cycles followed by cross-over upon progression [17]. While both agents resulted in similar rates of response in the overall study, objective response rates among BRCA1/2 mutation carriers were 68% with carboplatin compared to 33% with docetaxel (absolute difference 34.7%; 95% CI 6.3–63.1; p = 0.03).

More recently, we have seen Phase III results comparing the PARP inhibitor olaparib or talazoparib to non-DNA-damaging chemotherapy treatment of physician’s choice in BRCA1 and BRCA2 mutation carriers with advanced HER2-negative breast cancer. Both studies showed improvement of median PFS with the PARP inhibitor of ~ 3 months and greater tolerability, with no overall survival benefit evident as yet. Among patients with measurable disease, the objective response rates were approximately 60% with the PARP inhibitor compared to approximately 30% in physician choice (need to check talazaparib) [18, 19]. The FDA approved olaparib for the treatment of germline BRCA1/2 metastatic breast cancer in January 2018.

Of the genes implicated in familial breast cancer beyond BRCA1 and BRCA2, many moderate penetrance genes are likewise involved in the HR DNA repair pathway [20]. In women testing negative for BRCA1 and BRCA2 mutations, multi-gene germline sequencing identifies up to another 10% of patients with pathogenic mutations [21]. Recently, a large study of multi-gene germline testing in over 1,800 unselected patients with TNBC demonstrated that overall mutation rate was 14.6% with 8.5% having mutations in BRCA1, 2.7% in BRCA2 and an additional 3.7% with mutations in other genes such as PALB2 (1.2%), BRIP1, BARD1, and RAD51C among others [22]. Cancers arising in these genetic backgrounds are hypothesized to have HR DNA repair defects and therefore may have similar chemosensitivity to DNA-damaging therapies as BRCA1/2 mutation carriers [4, 23]. Clinical trials are currently underway to assess this hypothesis [24].

An important outstanding question is whether evidence of HR deficiency in TNBC patients without germline alterations is also a biomarker for treatment. The Homologous Recombination Deficiency (‘HRD’) assay measures levels of genomic instability or ‘scarring’ caused by any number of alterations in DNA repair capacity. At present, the HRD assay incorporates three measures of genomic instability: telomeric allelic imbalance (TAI; the number of regions with allelic imbalance that extend to the subtelomere, but do not cross the centromere), loss of heterozygosity (LOH; the number of regions > 15 Mb and less than one chromosome lost across the genome), and large-scale state transitions (LST; the number of chromosomal breaks between adjacent genomic regions longer than 10 Mb after filtering regions < 3 Mb) [25‐27]. The HRD score is currently calculated by adding the LOH, TAI, and LST scores and is a continuous score from 0 to 100 with a score of < 41 (previously < 10) as HR proficient and ≥ 42 (previously ≥ 10) as HR deficient. Using this cutoff, it is estimated that approximately 50% of TNBC patients will be classified as HR deficient. Tumors with BRCA1/2 mutations are classified as HR deficient regardless of HRD score.

Recently, multiple groups have reported that the HRD assay can be used to identify both BRCA1/2 mutant and wild-type TNBC patients more likely to achieve a favorable response to platinum-based neoadjuvant chemotherapy [27]. In the neoadjuvant PrECOG 0105 trial, responders to this platinum-based therapy had significantly higher mean HRD-LOH scores compared to non-responders; this was true for both BRCA1/2 wild-type and mutant responders [28]. Overall, 66% of patients with an HRD-LOH score of ≥ 10 or BRCA1/2 mutation responded compared with 8% of patients with an HRD-LOH score of < 10 and intact BRCA1/2.

The combined HRD score has been assessed in a number of clinical trials of platinum-based therapy. In a pooled analysis of six phase 2 neoadjuvant platinum-based TNBC trials (n = 267), the adjusted odds ratio for pathological response in HR deficient compared to non-deficient tumors was 4.64 (95% CI 2.32–9.27; p = < 0.0001) [29]. In addition to assessments in multiple single-arm studies of neoadjuvant platinum, in the randomized phase 2 GeparSixto trial, patients with high HRD score or tumor BRCA mutation were more likely to achieve pCR (55.9% vs. 29.8%), odds ratio 2.51 (p = 0.009) in multivariate analyses [30]. The addition of carboplatin numerically improved pCR rates with a non-significant interaction between HR deficiency and carboplatin benefit [31]. These data suggest that the HRD assay is promising in concept but whether it can be used to identify germline BRCA1 and BRCA2 wild-type patients who may benefit from platinum-based therapy remains to be seen.

While the results in the neoadjuvant setting have been relatively consistent, results in the metastatic setting are more difficult to interpret. In the metastatic TBCRC009 clinical trial of platinum in the 1st and 2nd line setting, mean HRD scores were significantly higher in patients achieving an objective response [32]. In the larger TNT phase 3 study of docetaxel versus carboplatin, however, platinum sensitivity was not associated with higher HRD scores [17]. Subsequent analyses suggested greater numerical (but not significant) response rates to docetaxel than carboplatin for both BRCA1 promoter methylation and silencing [33]. It is hypothesized that a genomic scar once induced in a tumor will persist, but secondary events may lead to a restoration of HR repair capacity. While genomic scars certainly have significant potential to be clinically useful particularly in the newly diagnosed setting, questions surrounding the specificity of the test and whether the assay predicts sensitivity to chemotherapy in general or specifically platinum and other DNA repair-targeted therapies require additional clarification. Ultimately, prospective validation of the predictive value of HRD in both platinum and PARP inhibitor benefit is needed.

An alternative approach to the genomic ‘scar’ phenotype detected by the HRD assay is to use mutational signatures derived from exome- or genome-level sequencing. Polak and colleagues investigated presence of mutational signature 3 [34] in whole exome sequencing data from nearly 1000 breast cancer patients [35]. They found that the presence of this signature could accurately classify BRCA1/2 alterations, while also demonstrating that epigenetic silencing of RAD51C and BRCA1 and germline variants in PALB2 were associated with signature 3 [35]. A second approach defined six new mutational signatures from whole genome sequencing of breast cancer patients that were predictive of BRCA1/2 deficiency. They integrated these in a weighted model, termed ‘HRDetect,’ which identifies BRCA1/BRCA2-deficient tumors with high (98.7%) sensitivity and also identifies tumors with somatic loss or functional BRCA1/BRCA2 deficiency [36, 37]. HRDetect was validated in an independent dataset and evaluation for clinical utility is ongoing [37].

Non-genomic approaches to characterize ‘BRCAness’ have also been explored. The ‘DDR deficiency assay’ is a 44-gene expression signature derived to identify loss of the Fanconi anemia/BRCA DNA repair pathways [38]. In two independent datasets of patients treated with (neo)adjuvant fluorouracil, anthracycline and cyclophosphamide, the DDR deficiency assay was significantly associated with pCR and relapse-free survival. In the neoadjuvant I-SPY2 trial evaluating the addition of carboplatin and PARP inhibitor veliparib to standard anthracycline and taxane-based therapy, a 77-gene BRCAness signature was associated with response to veliparib/carboplatin relative to control [39]. Methods to functionally characterize HR deficiency in fresh tumor samples offer a ‘gold standard’ for assessing HR, but are technically difficult and at this time not feasible to scale in a clinical setting. Powell and colleagues demonstrated that lack of RAD51 foci after ex vivo ionizing radiation on fresh breast cancer samples was significantly associated with genomic scars (LOH, LST, TAI) and biallelic inactivation of DNA repair genes [40]. In a prospective series, 16/56 (29%) primary breast tumors revealed defective RAD51 recruitment following irradiation [41]. Both gene expression signatures and functional characterization of HR require further validation in prospective studies yet could help to distinguish between functional HR capacity in the context of a genomic scar.

Tumor cells evade host anti-tumor immune responses through a variety of mechanisms, including tumor cell-intrinsic avoidance of immune detection (loss of tumor antigens or MHC proteins), elaboration of immunosuppressive cytokines, immunosuppressive cell recruitment, and immunoediting—the process by which tumors evolve or are selected to escape host anti-tumor immunity [51‐53]. Failure of TILs to kill adjacent tumor cells offers both a challenge and potential avenue for therapeutic intervention. TILs are primarily T-cells that express checkpoints such as PD-1 and CTLA-4, implying chronic antigen exposure but progressive loss of T-cell functions, elevated inhibitory receptors, and impaired cytokine production resulting in the inability to target tumor cells even in close proximity [46, 54, 55].

The ‘exhausted’ T-cell phenotype could reflect the end stage of immunoediting. Immunoediting is thought to begin early in cancer development as highly immunogenic tumor subclones are eliminated or adapt a less immunogenic profile in response to anti-tumor T-cells [55, 56]. (Fig. 2). As the tumor develops, induction of repressive molecules and decrease in co-stimulatory molecules in the infiltrating T-cells, among other processes, leads to the exhaustion phenotype [57]. Perturbations to this balance between host anti-tumor immunity and tumor immunosuppressive signals—such as surgery, chemotherapy, or radiation—may temporarily reinvigorate memory T-cells until immunogenic tumor cells are eliminated or another round of immunoediting occurs [55]. In this way, the subset of tumors with substantial numbers of TILs locked in this ‘stalemate’ may be primed for reactivation by immune checkpoint inhibitors.

Programmed death receptor 1 (PD-1) is an inhibitory immune checkpoint receptor expressed on activated T-cells and other immune cells and plays a key role in regulating the activity of effector T-cells. The PD-1 ligand, PD-L1, is expressed by multiple cancer types and upon binding to PD-1 exerts an immunosuppressive signal. The expression of PD-L1 has emerged as an important mechanism by which multiple cancer types are able to evade destruction by anti-tumor effector cells. Monoclonal antibodies targeting PD-1 (pembrolizumab, nivolumab) or PD-L1 (atezolizumab, avelumab, durvalumab) have garnered FDA approvals for the treatment of multiple malignancies. In TNBC, multiple studies are exploring the activity of these agents alone and in combination with cytotoxic chemotherapy, targeted therapy or radiation. As yet, none of these drugs have been FDA approved in breast cancer.

To date in metastatic TNBC, responses to single-agent anti-PD-1/anti-PD-L1 monoclonal antibodies have been modest and biomarkers of response have been actively pursued. Nanda and colleagues reported initial results from the KEYNOTE 0–12 Phase Ib study of pembrolizumab in patients with metastatic triple-negative breast cancer with PD-L1-positive tumor status (expression in stroma or in ≥ 1% tumor cells). Among 27 evaluable patients with a median of 2 prior therapies for metastatic disease, the objective response rate was 18.5% [58]. The subsequent KEYNOTE-086 Phase II trial evaluated pembrolizumab in metastatic TNBC patients that were previously treated with any level of PD-L1 expression (Cohort A) and as 1st line treatment in patients with PD-L1+ status (Cohort B) [59]. Among 170 patients enrolled in cohort A, 105 (61.8%) had PD-L1+ status and 43.5% had received 3 or more prior lines of therapy. Objective responses were seen in 4.7% with no significant difference in ORR based on PD-L1 expression status. Patients enrolled in Cohort B (n = 52) had an ORR of 23.1% suggesting greater benefit in patients with less heavily pretreated disease.

A Phase Ia study of atezolizumab enrolled 115 patients with metastatic TNBC. Patients had heavily pretreated disease with a median of 7 prior lines of therapy and 63% of patients were PD-L1+, 33% PD-L1- and 4% had unknown PD-L1 status [60]. The overall ORR was 10% with a 13% ORR in PD-L1+ patients and 5% ORR in PD-L1− patients. Among 1st line patients, the ORR was higher at 26%. Avelumab was evaluated in the Phase I JAVELIN study and among 58 patients with TNBC, an ORR of 8.6% was observed [61]. Patients with PD-L1+ status (n = 9) had an ORR of 44.4% and those with PD-L1− status had an ORR of 2.4%. Taken together, these early phase results in metastatic TNBC suggest higher activity in 1st line treatment and in those with PD-L1+ status. Multiple phase II and III trials of PD-1/PD-L1 checkpoint inhibitors in metastatic TNBC are ongoing and are evaluating monotherapy as well as combinations with chemotherapy, other immunotherapeutic agents and targeted therapies (Table 1).

In addition, the role of PD-1/PD-L1 blockade in TNBC is also being actively assessed in the adjuvant and neoadjuvant settings. Recent data from the I-SPY 2 trial demonstrated that adding pembrolizumab to neoadjuvant anthracycline and taxane-based chemotherapy results in an estimated pCR rate of 60% versus 20% in TNBC patients [62]. Multiple phase II and II studies are currently ongoing in the curable setting assessing these agents in combination with standard chemotherapy and also as monotherapy following completion of standard combination chemotherapy in high-risk patients.

Many of the strategies to enhance tumor immunogenicity focus broadly on the induction of DNA damage and neoantigens, for example through radiation or chemotherapy. However, certain—but not all—chemotherapeutics (taxanes, anthracyclines, cyclophosphamide) appear to induce ‘immunogenic cell death’ by activating type I IFN-receptor ligand systems resulting in recruitment of T-cells, among other mechanisms [67, 68]. Radiation enhances the diversity of the T-cell receptor repertoire of intratumoral T-cells but optimal dose, schedule, and immunotherapy combination is not known [69]. Monoclonal antibodies trigger antibody-dependent cellular cytotoxicity that subsequently activates antigen-presenting cells, a phenomenon observed with trastuzumab in HER2-positive breast cancers [70]. Certain oncogenic drivers, such as the Ras/MAPK pathway, may specifically downregulate MHC proteins and TILs, implying that inhibition may synergize with checkpoint inhibition [49]. A variety of other approaches to augment T-cell infiltration, including standard or personalized/neoantigen vaccines, chemotherapy or radiotherapy, immune agonists targeting a variety of molecules including STING, adoptive T-cell transfer, and others are under investigation.

Recent clinical trials demonstrate that adding PARP inhibitors may enhance activity of immunotherapy in breast cancer [71, 72]. Several recent pre-clinical studies suggest that PARP inhibition enhances immunogenicity in models of ovarian and breast cancer and that combining the two results in synergistic effects [73‐75]. TOPACIO, a multicenter, open-label phase 1/2 study evaluated the PARP inhibitor niraparib plus pembrolizumab in metastatic TNBC and advanced ovarian cancer and during the dose-finding portion of the trial, none of the eight evaluable patients progressed: 4/8 had objective response and 4/8 had stable disease [71]. The MEDIOLA trial of olaparib plus durvalumab included a cohort of BRCA1/2 mutant metastatic TNBCs and showed an objective response rate of 52% [72]. Collectively, these pre-clinical and early phase clinical data suggest that targeting or enhancing HR defects may enhance tumor immunogenicity and potentially sensitize to immune checkpoint inhibitors.