High XIST and Low 53BP1 Expression Predict Poor Outcome after High-Dose Alkylating Chemotherapy in Patients with a BRCA1-like Breast Cancer

Germline mutations in the BRCA1 gene confer a substantially increased breast cancer risk (1, 2). The genomic instability associated with such cancers is driven by defects in DNA repair, because BRCA1 is critical in the homologous recombination (HR) pathway. This pathway is the only known error-free repair mechanism for DNA double-strand breaks (DSB). In the absence of HR, repair of these lesions is error-prone and mainly performed by nonhomologous end-joining (NHEJ; refs. 3–6). In preclinical models, it has been extensively demonstrated that BRCA1-deficient tumor cells are sensitive to specific therapies. These therapies either damage the DNA by introducing DNA DSBs, such as alkylating agents, or through a synthetic lethal approach by PARP1 inhibition (4, 7–10).

Despite the sensitivity of BRCA1-deficient tumors to DNA DSB-inducing chemotherapy and PARP inhibitors, both intrinsic and acquired resistances can occur and thwart effective treatment (11). Recently, two biomarkers associated with resistance were identified in BRCA1-deficient model systems (12, 13). The first biomarker, X inactive–specific transcript (XIST), was found to be differentially expressed in BRCA1-deficient mouse tumors that had early and late relapses after treatment with cisplatin. Specifically, low gene expression of XIST was associated with late relapse, suggesting that this may be a predictive biomarker (12). Subsequently, XIST expression was investigated in HER2-negative patients from a randomized trial comparing high-dose (HD) alkylating chemotherapy and conventional (CONV) chemotherapy. We will use the trial's abbreviation of the arms (HD and CONV), but the observed effects may be due to the alkylating agents, the HD, or a combination. Low expression of XIST correlated with longer recurrence-free survival (RFS) after HD alkylating chemotherapy compared with conventional chemotherapy. Thus, the patients in the clinical study did not represent the BRCA1 deficiency of the mouse model, because many patients were included without features of BRCA1 deficiency (12, 14, 15). Furthermore, because associations between dysfunction of BRCA1 and XIST have been frequently described, it may be that BRCA1 deficiency and low XIST expression identify a similar group of patients (12, 16–21). For these two reasons, it is important to study XIST expression in the presence of BRCA1 deficiency, to recapitulate the findings in a brca1-deficient mouse model as closely as possible in the human situation (12).

The second biomarker associated with resistance of BRCA1-deficient cells was 53BP1. The choice of repair pathway of DNA DSBs is tightly regulated to ensure the proper activation of NHEJ and HR. 53BP1 loss has been shown to partially restore HR in BRCA1-deficient cells. Loss of 53BP1 allowed end resection of breaks ends, which is an important step in preparing the DNA strand for HR rather than NHEJ (13, 22, 23) This partial restoration rendered tumor cells more resistant to cisplatin, mitomycin C, and olaparib (a PARP inhibitor; refs. 13, 24). Subsequently, 53BP1 expression was investigated in two cohorts of patients. In these cohorts, associations with clinical characteristics and survival analysis were found (13). The clinical analyses for both biomarkers did not investigate whether these markers could be used as predictive biomarkers for therapy responses within BRCA1-deficient patients, leaving critical questions on the applicability of the findings.

We previously investigated a BRCA1-like copy number classifier as predictive biomarker for benefit of HD alkylating chemotherapy (25, 26). This test classifies tumors as BRCA1-like or non–BRCA1-like based on the presence of characteristic DNA gains and losses associated with BRCA1-mutated and promoter methylated breast cancers (14, 27). Patients with a BRCA1-like tumor had a 6-fold lower risk of recurrence and death when treated with HD chemotherapy compared with conventional chemotherapy. This benefit was not observed in patients with a non–BRCA1-like tumor (14). We found similar estimates in an independent cohort, in which BRCA1-like patients treated with HD chemotherapy had a 6-fold lower risk of recurrence and death (28).

We assessed whether the expression of XIST and 53BP1 can be used to identify a population of patients with a BRCA1-like breast cancer that do not benefit from HD chemotherapy. Our hypothesis was that matching the targetable defect of the model system to patients with that defect would result in good concordance of the preclinical and clinical findings and thus create opportunities for translation toward patients.

The patients in these studies have been reported before (14, 29) and were treated in a randomized controlled trial comparing HD alkylating to conventional chemotherapy. The conventional regimen consisted of five courses of 500 mg/m² 5-fluorouracil, 90 mg/m² epirubicin, and 500 mg/m² cyclophosphamide (FEC). Patients in the HD treatment cohort were administered four courses of FEC, followed by stem cell harvesting and one course of 1,600 mg/m² carboplatin, 480 mg/m² thiotepa, and 6,000 mg/m² cyclophosphamide. All patients have given written informed consent to be included in the study that was approved by the Institutional Review Committee (29). According to Dutch law, this implied consent for the analysis of residual tissue specimens obtained for diagnostic purposes and anonymized publication of the results (http://www.federa.org/code-goed-gebruik-van-lichaamsmateriaal-2011).

Patients were classified based on their copy number profile to belong to the BRCA1-like class or the non–BRCA1-like class as described before (14, 26). These classifiers assign a probability of being BRCA1-mutated (similar to being BRCA1-mutated, thus “BRCA1-like”) or non–BRCA1-mutated class (not similar to BRCA1-mutated, thus non–BRCA1-like). The BRCA1-like class is enriched for BRCA1 mutation carriers as well as patients with tumors that have BRCA1 promoter methylation. Patients with a BRCA1-like copy number profile benefited from receiving HD chemotherapy (14, 27). Because the preclinical models do not provide evidence for prediction in a BRCA1-proficient system, we did not investigate non–BRCA1-like patients.

We segmented 3.2K bacterial artificial chromosome (BAC) genome-wide aCGH profiles with the cghseg package (30). We then used the segmented log ratio of the BAC clone in which 53BP1 is located and the average of the two probes surrounding the XIST location for further calculations. We used a cutoff of <−0.1 segmented log ratio for identifying regions that may have lost the locus. To calculate the fraction of the chromosome lost, we counted the number of probes with segmented log ratio <−0.1 and divided that by the total number of probes on the chromosome.

Immunohistochemistry for 53BP1 was performed using a standardized protocol on the Ventana Benchmark Ultra system. Antigen retrieval was performed with citrate buffer for 44 minutes followed by antibody incubation for 32 minutes (Bethyl A300-272A dilution 1:2,000). The stainings were scored on a percentage of positive tumor cells by M. Opdam and J. Wesseling, discordant cases were resolved by consensus. A cutoff for loss of expression was based on the histogram of 53BP1 expression of tumors of 523 patients present in the tissue microarray (TMA). The distribution of the marker was bimodal, and we put a cutoff for low expression between the two groups, independent of survival analyses. The histogram of the data and representative pictures of a TMA core without 53BP1 expression and one with 53BP1 expression are shown in Fig. 1.

The gene expression analysis of XIST was performed with an Reverse Transcriptase–Multiplex Ligation-dependent Probe Amplification (RT-MLPA) kit (MRC Holland) as described before (12). We used the same cutoff for aberrant expression as this study.

We investigated the influence of XIST and 53BP1 to predict outcome after HD chemotherapy compared with conventional chemotherapy by dividing them into two groups: (i) “XIST-low-and-53BP1-high,” predicted to be sensitive, and (ii) “XIST-high-or-53BP1-low,” predicted to be resistant. We also report on 53BP1 status as independent predictive biomarker in the whole cohort as this analysis has not been described elsewhere. Furthermore, we investigated independent value of 53BP1 in the triple-negative (TN) subgroup because aberrant expression has been reported to be enriched in this subgroup of patients (13, 31–33).

We used the Kaplan–Meier method to visualize the survival [disease-free survival (DFS), RFS, and overall survival (OS), as defined in ref. 29] curves comparing HD and conventional chemotherapy of all BRCA1-like patients by XIST and 53BP1 status with exact log rank tests. We performed unadjusted Cox proportional hazards regression to estimate the effect of HD chemotherapy in XIST-low–53BP1-high and XIST-high-or-53BP1-low groups. To adjust for confounding, we performed a logistic regression of chemotherapy group on estrogen receptor (ER), grade, tumor size, and lymph node status, and used the fitted propensity score as adjustment factor in the Cox proportional hazards model. This way we could adjust for the characteristics of the four biomarker-treatment groups with only one extra covariate given the sample size. We used a treatment by biomarker interaction to attribute prognostic and/or predictive value of the biomarkers. In the analysis of 53BP1 as independent biomarker, we adjusted for ER, PR, HER2 status (except in the TN analysis), tumor size, lymph node status, and grade. Supplementary Table S1 shows a description of this study according to the reporting recommendations for tumor marker prognostic studies (REMARK) criteria.

We chose to present the patients for which all biomarkers were available. However, we also performed the same analyses on either of the biomarkers separately. Furthermore, we checked whether changing the cutoff for 53BP1 from 30% (main analysis) to 20% or 40% would influence the analysis. We did not change the XIST cutoff from the already published article (12).

All calculations were performed with R version 3.0.2 (34).

Array CGH-based BRCA1-like classification was available for 230 patients of the randomized trial; XIST expression was available for 60 patients that had either a BRCA1-like or TN cancer as described before (12, 14). We stained TMAs of the series with 53BP1 antibodies. We focused on the BRCA1-like patients with XIST and 53BP1 status to investigate the predictions of the preclinical model. Forty-one of the 230 patients had BRCA1-like cancers. We obtained immunohistochemistry of 53BP1 for 37 patients, dropout being caused by missing TMA cores or TMA cores that did not contain tumor cells for reliable assessment. We drew histograms of the data of 413 (after dropout of missing or non–tumor-containing cores) patients, found a bimodal distribution of the immunohistochemical scores, and determined the cutoff to be 30% (Fig. 1).

XIST RNA expression was available in 32 of 41 patients, the dropout being due to low RNA quality. Twenty-eight patients had available BRCA1-like status, XIST, and 53BP1 expression status. Table 1 describes the clinical characteristics of these 28 patients.

As previously described, the cohort of BRCA1-like patients has a good outcome after HD chemotherapy, although some patients have an event (Fig. 2A and D). Figure 2B and C and Fig. 2E and F split the overall BRCA1-like group into a group that according to the preclinical models should be sensitive (XIST-low-and-53BP1-high) and one that is resistant, either XIST-high-or-53BP1-low. This selects a subset of BRCA1-like patients that seem to have equally poor outcome after HD as after conventional therapy.

We further investigated this by calculating the event rates and comparing these for the subset of patients with exact log rank tests (Table 2). XIST-low, 53BP1-high patients have lower rates of events compared with XIST-high-or-53BP1-low patients, irrespective of chemotherapy. Furthermore, among XIST-low, 53BP1-high patients, those treated with HD chemotherapy had significantly lower event rates than those treated with conventional chemotherapy on DFS, RFS, and OS.

We then estimated HRs of HD versus conventional chemotherapy in the BRCA1-like sensitive and resistant group and the interaction HR, adjusting for ER status, tumor size, lymph node status, and grade by means of the propensity score (Table 3). BRCA1-like, XIST-low, 53BP1-high patients have significantly better outcomes after HD than after conventional chemotherapy in event rate analysis as well as Cox models.

Cox regression of RFS and OS showed the same direction of effects, but with unreliable estimates of the effect sizes due to 0 event in the HD-treated BRCA1-like, XIST-low, 53BP1-high patients. We performed sensitivity analysis with similar results when using cutoffs of 20% or 40% for 53BP1 and when analyzing both genes separately (data not shown). In non–BRCA1-like patients, the chemotherapy effect did not differ by XIST-53BP1 status on DFS, RFS, and OS (Supplementary Fig. S1 and Supplementary Table S2).

Although our interest in performing these analyses was driven by preclinical observations, knowing that this would result in a small subgroup analysis, we have data on 53BP1 protein expression for many more patients. Therefore, we calculated whether 53BP1 expression as an independent marker has any prognostic or predictive value in this cohort (Supplementary Table S3). 53BP1 did not have significant prognostic or predictive value (for HD chemotherapy) in the whole cohort (Supplementary Table S4). Patients with a 53BP1-expressing tumor in the TN subgroup (Supplementary Table S5) had better RFS, DFS, and OS on HD chemotherapy than on conventional chemotherapy [RFS hazard rate 0.34; 95% confidence interval (CI), 0.15–0.78; P = 0.01; Supplementary Table S6, similar for DFS and OS). The P values for interaction were not significant (0.65, 0.33, 0.18). We also calculated whether 53BP1 may be predictive for the conventionally dosed regimen. We did not find significant prognostic or predictive value in the whole cohort (Supplementary Table S7). In the TN subgroup, conventionally treated patients have poorer responses than HD-treated patients, but 53BP1 does not seem to be a prognostic or predictive modifier of this effect (Supplementary Table S8).

We investigated whether we could find evidence for genomic loss of XIST and 53BP1 in the aCGH profiles, which could result in, respectively, low RNA and protein expression of these genes. The mechanism of low expression of XIST was previously investigated by overlapping RNA FISH, aCGH, and RT-MLPA in a subset of patients (12). These findings suggested that low XIST expression was related to loss of the inactive X chromosome. In some cases, the active X chromosome was duplicated (12). We plotted the aCGH log ratios and the fraction of the chromosome lost by the XIST RNA and 53BP1 protein expression (Fig. 3). Patients that lost XIST expression had evidence for losing substantial parts of the X chromosome. However, this loss was not found for all patients, which could be obscured by duplication of the X chromosome. Of the 3 patients with 53BP1 loss, 2 patients showed indications that the 53BP1 locus might be lost, but these deletions seemed restricted to only a part of the chromosome (Fig. 4). The analyses on all patients with aCGH and, respectively, XIST and 53BP1 expression yielded similar conclusions (Supplementary Figs. S2 and S3).

In this study, we investigated whether XIST and 53BP1 expression, two potential markers of resistance to HD alkylating chemotherapy in BRCA1-deficient tumors identified in preclinical models (12, 13), could be used as predictive markers in breast cancer patients. We present early and preliminary evidence that high XIST expression or low 53BP1 expression could be used to identify BRCA1-like tumors that have early events and poorer outcome than BRCA1-like tumors that have low XIST and high 53BP1 expression.

The mechanism of loss of XIST expression seems to be through loss of the inactive X chromosome (12). We observed that the copy number profiles of some patients hinted at losing an X chromosome. However, we are unable to tell whether this is the inactive X chromosome and whether those without apparent loss may have copy number neutral DNA loss. These findings fit with the literature in which XIST expression, deregulation, and loss of the inactive X chromosome have been described in basal-like breast cancer over the last few years (16–21). Unfortunately, this does not shed light on how XIST deregulation or the loss of an inactive X chromosome confers a resistance phenotype. A link between BRCA1 and XIST has been investigated intensively (35–39). Furthermore, the deletion of Xist in mice leads to hematologic malignancies (40) and genomic instability of the X chromosome (41). One hypothesis would be that true BRCA1-loss–driven cancers require (a sequence of) events or multiple functions of BRCA1 that results in losing the inactive X chromosome. A tumor that does not lose the inactive X chromosome would not be true BRCA1-loss–driven, but may retain characteristics associated with the dysfunction of a single gene.

The mechanism of action of 53BP1 has been described better, with it being an important factor skewing DNA repair to NHEJ rather than homology-directed DNA repair in the absence of BRCA1. Losing both genes restores this balance and restores some level of HR (42). Genomic loss of the 53BP1 locus may contribute to the low expression of 53BP1, although the size of the aberrations in our series was to be certain that 53BP1 was driving the loss.

Several biomarkers to identify BRCA1 deficiency have been proposed (mutation status, gene expression (classifiers), methylation, copy number classifiers, etc.; ref. 43). All of these should enrich a patient population for BRCA1 defects. Clinical data from retrospective studies and early trials suggest that BRCA1-mutated tumors are sensitive to DNA DSB-inducing therapy, demonstrated by high pathologic complete remission rates, DFS, or OS benefits (44, 45). However, for none of these markers there is conclusive evidence as many studies have (combinations of) caveats, such as nonrandomized designs, lack of control groups to properly assess whether BRCA1 status is a prognostic marker or a predictive marker, the use of mixed chemotherapy regimens or lack of replication in independent studies. In this study, we used the genomic signature of BRCA1-mutated breast cancer, as it seems to have several advantages over other markers. First, it identifies both BRCA1-mutated cancers and tumors with BRCA1 promoter methylation (14, 27). This combination enlarges the group of patients with a potential defect in BRCA1. This is similar to our current finding that XIST and 53BP1 both seem to contribute to part of the poor outcome cases of HD chemotherapy. Secondly, BRCA1-like status is a replicated predictor of benefit of HD chemotherapy in a randomized controlled trial (14, 28). This enables us to investigate extra markers on top of BRCA1 deficiency identified through BRCA1-like classification of copy number profiles. Because separate biomarkers usually differ (hopefully slightly if aiming to identify the same population), it is prudent to separately investigate overlaying XIST and 53BP1 with other genomic scars (46) that could identify cancers with deficient HR repair until a formal comparison has been made. It is uncertain whether the dose, the specific agents, or the combination of drug and dose confer the benefit of BRCA1-like patients for the HD regimen (14, 28, 15, 47). Furthermore, it could be that drugs with a less heavy treatment burden than HD chemotherapy, such as carboplatin or PARP1 inhibitors, may be predicted with similar clinical outcome (48, 49).

Previously, XIST and 53BP1 were not assessed in a preselected background enriching for BRCA1-deficient tumors based on BRCA1-like status. The aim of the study that identified XIST was the identification of biomarkers of response to cisplatin (12). However, this finding was subsequently tested in a more general population (HER2-negative breast cancer) than the initial preclinical setting (BRCA1-deficient mouse tumors). We would argue that the good outcome observed by Rottenberg and colleagues, who assessed XIST as independent marker, can be explained by a strong correlation with BRCA1-like status. In the 53BP1 study, only prognostic value was investigated, despite data that suggest a role as predictive rather than prognostic biomarker (13). Unfortunately, investigating putative biomarkers in relation to chemotherapy outcome within a specific genetic defect automatically limits the number of patients, but may be very valuable in evaluating more personalized treatment options. In this case, for example, the heavy burden of HD chemotherapy could be prevented if the poor outcome could be expected, as predicted by the preclinical data and as our clinical data suggest. A robust preclinical modeling system that is indicative of human cancers could thus be very valuable. On the other hand, one could argue that only high prevalence markers with solid statistical proof in large cohorts will ultimately be clinically important (50). Unfortunately, in these large cohorts, the context of the preclinical modeling system is often forgotten. It could be that many biomarkers derived from such modeling systems get lost in translation when the cohorts are not sufficiently matched to the characteristics of the preclinical system. We investigated 53BP1 in the whole cohort (a reasonable powered situation) and found no association and a minor effect in the TN subgroup, in which those patients with 53BP1-expressing tumors benefit from HD regimen. Although potentially interesting, we are not aware of any other evidence explaining the inverse relationship between therapy responses. This association follows the analysis in the BRCA1-like patients: TN or BRCA1-like patients with a 53BP1-expressing tumor benefit from the HD regimen compared with the conventional regimen. We found similar association of 53BP1 loss in hormone receptor–negative tumors as others did for TN status (although only a trend in our dataset; refs. 13, 33).

As has been discussed for BRCA1 mutation status, it is extremely rare to have a randomized controlled trial and high numbers of patients to investigate the response of BRCA1-associated breast cancer to a specific therapy (51). Even combining all trials that randomized between a HD and conventional therapy regimen would not yield a large number of patients that have the specific combination of biomarkers. Therefore, it seems more feasible to follow these findings up further in prospective studies that investigate therapy that targets a BRCA1 defect.

For concluding, we show that XIST and 53BP1 can be used to identify BRCA1-like breast cancer patients that have higher event rates and poor outcome after HD chemotherapy. These observations have to be confirmed in trials with similar questions that have been conducted in the past, or as exploratory analysis in future randomized trials.

S.C. Linn has ownership interest as co-inventor on a patent application for a BRCAness gene expression classifier. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P.C. Schouten, M.A. Vollebergh, S.C. Linn

Development of methodology: P.C. Schouten, M. Jonkers, M. Loden, S.C. Linn

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.A. Vollebergh, M. Opdam, M. Jonkers, J. Wesseling

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.C. Schouten, M.A. Vollebergh, M. Opdam, M. Hauptmann, S.C. Linn

Writing, review, and/or revision of the manuscript: P.C. Schouten, M.A. Vollebergh, J. Wesseling, M. Hauptmann, S.C. Linn

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Jonkers

Study supervision: J. Wesseling, S.C. Linn

The authors thank Peter Bouwman, Jos Jonkers, and Piet Borst for providing feedback on the article and the reviewers for discussion and suggestions.

This study was supported by funding from the Dutch Cancer Society (NKI 2006-3706) and A Sister's Hope/Pink Ribbon (to S.C. Linn).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.