Clinical implications of prospective genomic profiling of metastatic breast cancer patients

Breast cancer remains the most common malignancy and cause of cancer-associated death amongst women worldwide [1, 2]. For patients with metastatic breast cancer (MBC) the disease is almost always fatal, with 5-year survival rates of approximately 26% [3, 4]. In early-stage breast cancer, treatment decisions are guided by the clinical subtypes of oestrogen receptor positive (ER+HER2−), human epidermal growth factor receptor 2 amplified (HER2+) and triple-negative breast cancer (TNBC), which are defined by the presence or absence of the oestrogen receptor and progesterone receptor and HER2 overexpression or gene amplification. This broad classification does not account for significant tumour evolution during disease progression, driven by selective pressures [1, 5]. Additionally, germline mutation testing for breast cancer patients, which could identify targetable alterations across all patients, is only common in the cases of young women, male breast cancer, or individuals with a strong family history [6].

There are an increasing number of agents that target specific genomic alterations, especially mutated protein kinases [7, 8]. Identification of suitable patients and gaining access to the corresponding targeted therapy has been challenging due to the cost of sequencing technology, the requirement in the past for non-formalin-fixed material, the inherent intra-tumoural heterogeneity of breast cancer and the lack of frequently occurring genomic alterations that can be targeted. In MBC, few genes are mutated at a frequency above 10%, leaving a wide array of infrequently altered potential targets [8]. Due to the rarity of these targets, clinical trials of targeted therapies are challenging to conduct.

In the era of precision oncology, few large-scale genomic studies have focused on metastatic breast cancer, and hence, our understanding of the breast cancer genome has relied on large-scale studies in primary tumours [7, 9, 10]. However, it is now well established that early breast cancer and MBC may have differing genomic profiles even in the same patient. More data surrounding the genomic landscape of MBC will be useful to identify oncogenic drivers specific for advanced disease with potential therapeutic implications, such as ESR1 and resistance to endocrine therapies [10, 11].

Here we describe such an initiative, from an academic tertiary referral cancer centre, where individuals with MBC had their tumour samples prospectively profiled using a targeted next generation sequencing (NGS) assay. The aim of this study was to explore the feasibility and clinical relevance of this approach.

Between February 2014 and May 2019, 322 patients with MBC were recruited. The majority of these patients were ER+HER2− (68%, n = 219), with HER2+ and TNBC representing 12% (n = 38) and 20% (n = 65) respectively. The median age of these patients was 46 years (23–74) at primary diagnosis and 52 years (24–78) at metastatic diagnosis (Table 2).

Across the cohort, 748 tumour samples were requested, with an average of 2 samples per patient (Fig. 1a). In total, 82% (n = 614) of the requested samples were received; however, the time between request and receipt varied considerably with an average of 29 days for samples stored by the parent institution and 49 days for samples retrieved from external pathology providers.

Amongst the received samples, 35% (n = 215) had insufficient DNA quantity for sequencing, and the remaining 65% (n = 399) underwent sequencing with a success rate of 90% (n = 357). The failure rate of 10% of samples (n = 42) was largely attributed to poor quality DNA due to FFPE fixation, as neither the age of the sample nor its location were significant predictors of failure. The majority of the failed sequencing samples came from primary lesions (70%, n = 29). Metastatic samples, from sites such as lymph node (12%, n = 5), liver (5%, n = 2) and lung (5%, n = 2), accounted for less than a quarter of sequencing failures. A single specimen from the bone, ovary, brain and skin also failed sequencing (2% each). Ultimately, this led to 72% (n = 234) of recruited patients being successfully sequenced. The majority of samples sequenced successfully were primary tumours (53%, n = 189). Samples from metastatic lymph nodes (10%, n = 37) and the liver (9%, n = 31) accounted for the majority of metastatic samples (n = 168, 47%). Other metastatic sites included bone (7%, n = 26), brain (5%, n = 19), lung (3%, n = 12) and ovarian (2%, n = 6). The remaining 6 metastatic samples each came from a unique site and accounted for overall 2% (n = 6). Paired primary and metastatic tissue was available for 35% of patients (n = 81).

Across the sequenced samples, there was a median of 3 non-synonymous somatic mutations per sample (range 0–10) and 4 focal CN alterations (range 1–28). Level 1 alterations were found in 49% (n = 114) of all patients sequenced and 25% (n = 57) of all patients were found to carry a level 2 change (Fig. 1b). This amounts to 74% (n = 171) of patients harbouring a potentially actionable alteration. The majority of these samples were of the ER+HER2− subtype (77%, n = 131), with HER2+ and TNBC subtypes accounting for 6% (n = 11) and 17% (n = 29) of the total cohort respectively. Fifty-seven percent of ER+HER2− cases (n = 93/162), 24% of HER2+ (n = 6/25) and 32% of TNBC cases (n = 15/47) had level 1 alterations.

As shown in Fig. 1c, PIK3CA was the most common level 1 actionable alteration, being found in 30% of sequenced patients and accounting for 52% of patients with level 1 actionable mutations (n = 71). Of the cohort with actionable alterations (n = 171), 43% (n = 74) were referred onto a molecularly targeted clinical trial. The majority of these patients had PIK3CA mutations (89%, n = 68), followed by 2 patients with PTEN mutations, and 4 patients with ERBB2 mutations. A further 2 patients with FGFR1 amplifications were referred for FGFR inhibitor trials, although this gene is not considered actionable according to ESMO Clinical Actionability guidelines [13]. Of ER+HER2− patients, 10% (n = 24) were found to carry an ESR1 mutation in their metastatic tumour sample. Nearly all of these patients had been treated with prior endocrine therapy (Table 2).

A total of 21 patients were found to harbour BRCA1 or BRCA2 mutations. Of these, 36% (n = 8) were previously known pathogenic germline BRCA1 or BRCA2 carriers. The remaining 64% (n = 13) were referred on to germline testing. Of these patients, 2 were confirmed to carry germline variants of unknown significance, 5 confirmed wild type and 6 patients did not attend germline testing. A previously known pathogenic germline mutation in CHEK2 was also identified in a single patient.

Figure 2a shows the genomic landscape of the sequenced metastatic samples (n = 167 samples). Amongst these samples, TP53 was the most commonly altered at 40%, followed by PIK3CA, CCND1, NCOR1 and FGFR1. In order to identify any gene that may be enriched in the metastatic cohort, the mutational frequencies were compared to those in the TCGA breast cancer dataset comprised of primary tumours [31]. The genes AKT1, BRCA2, CHEK2, ESR1, FGFR4, KMT2C, NCOR1, PIK3CA and TSC2 were significantly enriched in our metastatic study population compared to the TCGA cohort (Supplementary Table 1).

We performed an analysis to identify alterations that may be significantly co-occurring or mutually exclusive. Each subtype was analysed individually using the 25 most frequently altered genes within each cohort. In ER+HER2− metastatic samples (n = 112) (Fig. 2b and Supplementary Table 2), 43 gene-pairs were found to significantly co-occur. When this analysis was performed in the TNBC subtype (n = 40 samples), only 2 gene pairs were found to be co-occurring: RB1 and FGFR1 (p = 0.01), as well as CHEK2 and CDH1 (p = 0.025). In the ER+HER2− group, RB1 mutations were found to co-occur with 8 other genes, suggesting a significant role within this subtype (Supplementary Table 2). Overall, in our cohort TNBC metastatic samples have less significantly co-occurring genes than ER+HER2− breast cancers.

The concordance of key driver alterations was investigated between primary and metastatic samples (Fig. 3) across the 81 patients (35%) of the cohort with paired samples. Of note, PIK3CA was largely concordant with a score of 69% (n = 20/29 patients). In the 9 patients who displayed discordance, the PIK3CA mutation was found exclusively in the metastatic lesion in 5 patients and exclusively in the primary lesion in 4 patients. The only gene to show 100% concordance was BRAF (n = 4 patients); however, TP53, FGFR1 amplification, ERBB2 amplification and IFG1R showed concordance scores greater than 75%. FOXA1, STK11, TSC1 and TSC2 were found to have no concordance between tumours and were largely restricted to metastases (Fig. 3). When tumour concordance was evaluated within each patient, only 18% (n = 15 patients) were found to have 100% concordance between their primary and metastatic lesion, whilst 7% of patients (n = 6) had no concordant findings, highlighting the genomic heterogeneity of metastatic breast cancer within individual patients.

We investigated factors associated with survival, and unsurprisingly, the clinical subtype (Fig. 4a) had a significant impact on overall survival (OS). When the time between primary disease and metastatic disease diagnosis was compared amongst the patients, those with a longer disease-free interval had a significantly better OS (hazard ratio (HR) = 0.61, confidence interval (CI) = 0.39–0.94, p = 0.02). Amongst the patients with sequenced samples, we found that higher mutation number was associated with significantly worse survival (HR = 1.1 per mutation, CI = 1.0–1.2, p = 0.01, Fig. 4b). This trend was observed in both the ER+HER2− subtype (HR 1.1, CI = 1.0–1.2, p = 0.04; Fig. 4c) and the HER2+ subtype (HR = 2.5, CI = 1.1–5.3, p = 0.02), but not in the TNBC subtype (HR = 1.1, CI = 0.86–1.4, p = 0.4; Fig. 4d). The same subtype level analysis was performed using copy number; however, no significant results were observed.

Although breast cancer is not commonly considered to be immunogenic, the presence of TILs has been known to impact survival in primary breast cancer [32]. In our dataset, TILs in the metastatic lesion were not significantly associated with OS for any subtype, as calculated from time of metastatic diagnosis (Supplementary Table 3). As anticipated, amongst the primary samples, TNBC patients with a higher number of TILs had a significantly longer DFI (Supplementary Table 3). This was also evident using the TNBC metastatic samples, where those with higher TILs had a significantly longer disease-free interval than those with lower TILs (HR 4.07, CI 1.2–13.6, p value: 0.02). Consistent with previous data some metastatic sites had significantly greater TIL infiltrate (Fig. 4e) [33].

In this study, 49% (n = 114) of patients were found to harbour a level 1 actionable alteration, which corresponds to the ESCAT LOE I/II genes. Level 1 actionable alterations were of high clinical relevance with 65% (n = 74/111) of patients with alterations in this category referred for clinical trial. Level 2 actionable alterations were identified in 25% (n = 57) of patients and did not impact clinical decision making due to the unavailability of a matched therapy. This amounts to almost three-quarters (74%, n = 171) of the cohort being found to harbour at least one actionable alteration, albeit with variable clinical importance. Of ER+HER2− patients, 10% (n = 24) were found to carry ESR1 alterations, and 64% (n = 13) of patients with a BRCA1 or BRCA2 mutation with a high variant allele fraction were recommended for germline testing. Additionally, we found several genes that were enriched in the metastatic setting across the whole cohort, as well as discordant findings in patients with both primary and metastatic samples, highlighting the need to profile metastatic tissue where possible. Finally, we found that higher mutation number was also associated with worse survival. These findings were obtained using a targeted panel sequencing far fewer genes (around one-third) than some which are commercially available. The outcomes of this study suggest that a panel featuring only ESCAT LOE I/II genes, equivalent to level 1 in this study, would be sufficient for the majority of patients who could benefit from genomic interrogation. These smaller and less expensive assays may make genomic profiling more broadly accessible to breast cancer patients and clinicians.

The distribution and frequency of metastatic genomic alterations in this study are similar to the findings of other studies, including enrichment of ESR1 mutations believed to be the result of therapeutic selective pressure [7, 10, 11, 34]. Mutations in ESR1 are almost exclusively found in metastatic lesions of ER+HER2− patients, although some studies have shown they can exist prior to the onset of metastases in patients treated with endocrine therapies [35, 36]. This was observed in 2 patients, shown in Fig. 3, where the primary samples were collected post-endocrine therapy and had already developed an ESR1 alteration. ESR1 mutations arise due to prolonged treatment with tamoxifen and aromatase inhibitors; however, retrospective analysis of the PALOMA-2 and PALOMA-3 studies found that patients treated with a selective oestrogen degrader either alone or in combination with CDK4/6 inhibitors also showed development of ESR1 mutations [37‐39].

As expected, TP53 alterations were the most prevalent across all samples. Of all the level 1 actionable genes, PIK3CA was found to be the most commonly altered, consistent with other studies [34]. Recently, the α-subunit specific PIK3CA inhibitor alpelisib was approved for use in ER+HER2− MBC patients with hotspot PIK3CA mutations, providing a matched therapy for this most common alteration [34, 40, 41]. Conversely, the finding that 25% of patients carried a level 2 alteration which lack proven therapeutic options highlights the difficulties in matching the majority of patients with targeted therapies. It is uncertain how many of these level 2 alterations may be clinically relevant in the near future. Nevertheless, the high prevalence of PIK3CA alterations alone means that a nearly a quarter of MBC patients could be treated with a matched targeted therapy, even accounting for those that could not have a metastatic sample sequenced as there was a high concordance rate (69% concordance) with the primary tumour.

Like other studies, we found a proportion of patients (26%) did not carry a potentially actionable alteration. Patients where no alteration was found may benefit from more comprehensive genomic analysis such as whole-genome sequencing, where mutational signatures such as homologous recombination repair deficiency or more rarely, microsatellite instability could be identified and utilised for therapeutic decisions [42, 43].

Our concordance analysis in 81 patients with metachronous primary and metastatic samples allows observation of how key driver genes change in metastatic disease. Mutations in both TSC2 and STK11 were found exclusively in metastatic samples and are possibly associated with the development of metastatic disease or resistance to therapy. The finding that TSC2 was also enriched in our metastatic cohort compared to the TCGA primary data set also supports this hypothesis [11]. The high frequency of mutations in the genes RB1, TSC1 and FOXA1 in the metastatic samples with few or no concordant cases also suggests these genes are similarly relevant to the biology of metastatic disease. Co-mutation analysis in both the ER+HER2− and TNBC subgroups suggested that RB1 may also play a key role in the emergence of metastatic and treatment-resistant subclonal elements. This finding is supported by other studies that have linked alterations in RB1 with resistance to endocrine therapies and CDK4 inhibitors [10, 44, 45].

Our study has several limitations. Firstly, the somatic landscape of the sequenced tumour may not be indicative of the whole metastatic disease burden. Novel methods such as circulating tumour DNA sequencing could be utilised to avoid sampling bias and increase the number of patients for whom genomic information is available. Nevertheless there are still cases of discordance between circulating tumour DNA and tumour DNA from sequenced samples, and plasma sampling has a higher rate of false negatives [1, 46]. Secondly, a panel of the size employed in this study cannot accurately estimate the true total mutation burden, and the mutation number per sample reported here cannot be considered a surrogate measure. Thirdly, the cohort of patients in our study may not be representative of the general metastatic breast cancer population as there was likely to be some bias towards enrolling younger and fitter patients. Fourthly, the rate at which patients receive therapies matched to their genomic profile is subject to the availability of these therapies and clinical trials, which will vary greatly across clinical contexts, and the timeframe in which the genomic profile can be delivered. The latter was heavily influenced by the time required to receive tissue samples. Reducing this further will require closer integration of pathology services with the treatment team and greater awareness of the urgency of tissue requests. Fifthly, consistent with other studies, we found that a relatively high rate of patients could not be delivered a genomic profile due to the difficulty in obtaining a suitable sample as 35% of samples had insufficient DNA. Due to the inherent difficulties in obtaining metastatic biopsies from MBC patients, these failure rates for tumour sequencing are unlikely to improve.

This study has found that prospective genomic sequencing for the management of MBC is both feasible and of utility as an adjunct to standard management with 49% of sequenced patients carrying a level 1 actionable alteration. PIK3CA was the most common actionable mutation, found in 52% (n = 71) of patients with a level 1 actionable alteration. Of patients referred for a clinical trial with a matched therapy, 89% (n = 68) had a PIK3CA mutation. We found that PIK3CA mutation status, and genomic profiles in general, demonstrates within-patient heterogeneity which is clinically relevant, although PIK3CA was usually concordant across primary and metastatic samples. Nevertheless, the presence of genomic heterogeneity indicates sequencing of metastatic lesions is preferred if possible. Due to the inherent difficulties in obtaining biopsies of metastatic disease, the combination of tumour and circulating tumour DNA sequencing is likely to be the optimal strategy to improve the rate of successful genomic profiling. It is expected the clinical utility of genomic profiling will increase over time with greater availability of targeted therapies, as has already transpired with the recent approval of the PIK3CA inhibitor alpelisib. We conclude that where possible, patients with MBC should undergo molecular profiling.