Adjuvant Capecitabine-Containing Chemotherapy Benet and Homologous Recombination Deciency Status Among Early-Stage TNBC Patients in the FinXX trial

Background: Recent data demonstrate that patients with early-stage triple negative breast cancer (TNBC) benet from escalating adjuvant treatment with capecitabine. However, since a substantial proportion of patients does not benet, predictive biomarkers to select those individuals upfront are needed. Over half of all TNBCs have a BRCA1-like DNA copy number signature similar to the prole found in germline BRCA1-mutated breast cancers and indicative for homologous recombination deciency. We evaluate this signature as a predictive biomarker for capecitabine benet in archived specimens of the randomized controlled FinXX trial. Additionally, we compared the concordance of our DNA-based BRCA1-like classier with the RNA-based NanoString BRCAness signature. Methods: Early-stage TNBC patients were randomized between adjuvant capecitabine-containing chemotherapy (TX+CEX: capecitabine plus docetaxel, followed by cyclophosphamide, epirubicin and capecitabine) and conventional adjuvant chemotherapy (T+CEF: docetaxel, followed by cyclophosphamide, epirubicin, and uorouracil). Breast tumor BRCA1-like status was determined on low coverage, whole genome next-generation sequencing data using an established DNA comparative genomic hybridization algorithm. We used interaction analysis in proportional hazards models to evaluate whether benet of adjuvant capecitabine-containing versus conventional chemotherapy differs between BRCA1-like and non-BRCA1-like tumors in early-stage TNBC patients. Results: For 129 (63.9%) of the 202 TNBC patients the BRCA1-like status could be determined. Thirty-ve recurrences and 32 deaths occurred during a median follow-up of 10.7 years. The capecitabine effect on recurrence-free survival did not signicantly differ between the 68 patients (52.7%) with a BRCA1-like tumor (HR 0.66, 95% CI 0.24-1.81) and others (HR 0.23, 95% CI 0.08-0.70, P interaction = 0.17),


Background
Triple negative breast cancer (TNBC) accounts for 10-20% of all breast cancers and is associated with a high risk of early recurrence and poor survival once metastasized [1,2]. Trials evaluating escalation of adjuvant treatment are emerging [3,4], including trials on the addition of capecitabine for early-stage TNBC patients. This prodrug of 5-uoruouracil belongs to the class of antimetabolites and shows cytotoxic activity through the inhibition of thymidylate synthase and the incorporation of its metabolites into DNA and RNA [5]. A recent meta-analysis including 3,854 early-stage TNBC patients showed that adjuvant capecitabine following or added to standard neoadjuvant anthracycline-and taxane-based therapy substantially improved disease-free survival (DFS) and overall survival (OS) [6,7]. Hence, this approach has been incorporated in national and international guidelines [8][9][10]. Although these results are promising, a substantial proportion of TNBC patients does not bene t from the addition of adjuvant capecitabine. This is, for instance, illustrated by the 8.5% absolute overall survival bene t at 5-years for TNBC patients who did not achieve a pathological complete remission on neoadjuvant chemotherapy with subsequently 6-8 courses of capecitabine versus those who did not receive capecitabine (78.8% versus 70.3%, respectively) [7]. To our knowledge, the only study exploring biomarkers that predict capecitabine bene t failed to identify a predictive marker in an 800-gene expression analysis [11]. Therefore, it remains unclear which early-stage TNBC patients have a high chance of bene tting from capecitabine.
Homologous recombination de ciency (HRD) may serve as a putative predictive biomarker to guide decisions on systemic therapy for patients with early-stage TNBC, especially for DNA damaging agents [12,13]. In unselected TNBC patients, approximately 10% of the patients harbor a deleterious BRCA1/2 mutation which results in tumors that are de cient in homologous recombination [14][15][16][17][18]. In TNBC patients without a germline BRCA1/2 mutation, a signi cant number of tumors harbor HRD [12,19,20]. The array comparative genomic hybridization (aCGH) BRCA1-like classi er is an HRD-test that has been developed from the characteristic DNA copy number aberrations of BRCA1-mutated breast cancers [21]. This BRCA1-like classi er showed clinical validity and utility to predict bene t of intensi ed platinumbased chemotherapy for stage III HER2-negative breast cancer patients [22][23][24][25]. However, the predictive value of the BRCA1-like classi er for outcome after (neo)adjuvant treatment with other DNA damaging agent-containing regimens and/or dose-intensities is currently unknown.
In data from two studies contributing to the results of the meta-analysis of Van Mackelenbergh et al. [6], the GAIN trial and the CREATE-X trial, there are indications that patients with BRCA1-like TNBC tumors may bene t from capecitabine-containing treatment. First, Van Rossum et al. observed a trend for improved survival in the BRCA1-like TNBC patients of the GAIN trial treated with capecitabine-containing chemotherapy compared with intensi ed dose-dense chemotherapy [26]. Second, based on the CREATE-X trial, which was limited to patients with residual disease after neoadjuvant treatment, it seems that capecitabine could have a greater role in patients with tumors that are less sensitive or partially resistant to regimens containing anthracyclines and taxanes [7]. We hypothesize that the improved outcome of TNBC patients treated with capecitabine-containing chemotherapy in the CREATE-X and the FinXX trial were driven by patients with BRCA1-like tumors. Contrarily, recent ndings of Asleh et al., using a RNAbased 800-gene panel in exploratory analyses, suggest that the BRCAness signature is not associated with the bene t of adjuvant capecitabine in early-stage TNBC [11]. Therefore, the predictive value of the DNA-based BRCA1-like status for the differential bene t from adjuvant capecitabine in early-stage TNBC needs further evaluation.
Our aim is to evaluate whether BRCA1-like status determines bene t of adjuvant capecitabine-containing systemic treatment in early-stage TNBC patients within the FinXX trial. The FinXX trial is a large phase III, randomized controlled trial comparing adjuvant conventional chemotherapy with adjuvant capecitabinecontaining chemotherapy [27].

Patients
We studied early-stage TNBC patients who were included in the Finland Capecitabine (FinXX) trial; a large, multicenter, randomized controlled clinical trial conducted in Finland and Sweden between 2004 and 2007 [27,28]. Eligibility criteria have been published previously [27]. In summary, the patients were younger than 65 years, had histologically con rmed invasive breast cancer with either regional lymph nodes containing cancer or node negative cancer with primary tumors of ≥ 20 mm diameter and negative progesterone receptor (PR) expression in immunohistochemistry, no distant metastases, and no prior neoadjuvant chemotherapy. TNBC was de ned as estrogen (ER) and progesterone receptor (PR) negativity (< 10%), and no HER2 overexpression (determined either by immunohistochemistry or in situ hybridization). The study was approved by the Ethics Committee of the participating medical institutions and the National Agency for Medicines, Finland. Patients supplied written informed consent to allow the use of their tumor tissue for clinical study related research purposes. The Institutional Review Board at the Helsinki University Hospital, Finland, approved the use of archival tissue for the current translational study.

DNA extraction
Tumor DNA was isolated from two 10 µm whole slides of formalin-xed para n-embedded (FFPE) tissue containing at least 50% tumor cells. A manual microdissection was carried out for slides containing ≤ 50% of representative tumor area to increase the percentage of neoplastic cells. Para n was removed with Qiagen's Depara nization Solution, and tissue was lysed using a mixture of 20 µL Proteinase K (20 mg/ml, included in the QIAsymphony DSP DNA kit) and 200 µL lysis buffer (0.05 M Tris-HCl ph 8.5, 0.04 mM EDTA, 0.5% Tween20) per sample at 56 °C overnight. DNA extraction was performed with QIAsymphony SP instrument using DSP DNA mini kit with 100 µL elution volume (Qiagen, Venlo, The Netherlands).
Low coverage whole genome sequencing and data processing The amount of double-stranded DNA in the genomic DNA samples was quanti ed using the Qubit® dsDNA HS Assay Kit (Invitrogen, cat no Q32851). Up to 500 ng of double-stranded genomic DNA was fragmented using ultrasonicator shearing (Covaris.com, Massachusetts, USA) to obtain fragment sizes of 160-180 bp. Samples were puri ed using 1.8X Agencourt AMPure XP PCR Puri cation beads according to manufacturer's instructions (Beckman Coulter, cat no A63881). DNA library preparation for Illumina sequencing was performed using the KAPA Hyper Prep Kit (KAPA Biosystems, KK8504). During the ligation 144 unique adapter indices, manufactured by IDT (Integrated DNA Technologies IDT, Inc. Coralville, Iowa, USA), were used in a molarity of 15 µM. Six PCR cycles were used during library enrichment to obtain enough yield for sequencing. All DNA libraries were analyzed on the Caliper GX bioanalyzer (PerkinElmer) using the HT DNA High Sensitivity LabChip for determining the molarity. Up to 133 uniquely indexed samples were mixed together by equimolar pooling. The pools were analyzed on the Agilent Technologies 2100 Bioanalyzer and subsequently diluted to 10 nM. Each pool was subjected to sequencing in one lane of a single read 65 bp run, on an Illumina HiSeq2500 machine, according to manufacturer's instructions.
Reads were aligned to the reference genome GRCh38 using BWA-MEM algorithm (version 0.7.17) [29]. Per bin of 20 kb, using BEDTools [30], reads on autosomes were counted. Excluded were sites attracting excessive anomalous read mappings (ENCODE) [31] and bins that had a GRCh38 reference mappability below 0.2. Mappability is the fraction of 65 bp sequences, per bin, that aligns to itself. Local GC effects in samples were tted with a non-linear loess, including a subset of reference mappabilities over 0.8, to correct sample bin counts. A line can be tted through the origin and center of GC corrected counts per mappability density. The slope of this line is used to scale mappabilities to reference counts. Genomic pro les consist of log2 ratios of GC corrected bin counts divided by these scaled reference counts.
The sequencing data discussed in this publication have been deposited in NCBI's Sequence Read Archive (SRA) and are accessible through BioProject number PRJNA647428 [32].

BRCA1 -like classi cation
Genomic pro les were analyzed using a BRCA1-like classi er, which was originally developed using array comparative genomic hybridization (aCGH) data generated from breast cancers that were or were not associated with germline BRCA1 mutations [21]. In brief, the BRCA1-like classi er is a shrunken centroid classi er that assigns a genomic pro le to a BRCA1-like class using a probability score between 0 (non-BRCA1-like) and 1 (BRCA1-like). The threshold for assigning a breast tumor to the BRCA1-like group was set at ≥ 0.63 as obtained and validated in previous studies [22][23][24][25]. The BRCA1-like classi er can be used on genomic copy number variation (CNV) pro les obtained by low coverage whole genome sequencing [26,33]. Recently, we implemented several updates in the processing of CNV sequencing (CNVseq) data and validated the BRCA1-like classi cation obtained with these data. A detailed description is provided in the Supplementary methods (Additional le 1: Supplementary methods). In brief, the BRCA1-like classi cation of copy number pro les can reliably be obtained with the updated CNVseq data with an accuracy of 85-93% when compared to the original BAC aCGH BRCA1-like classi er (which is similar to previously established performance on low coverage, whole genome nextgeneration sequencing [33]).
Quality checks of the CNV pro les of the TNBC FinXX patients were performed blinded for BRCA1-like score and outcome. Samples with low quality were excluded from analyses.
Previously, the identi cation of BRCAness has been explored on the same dataset of early-stage TNBC patients using the RNA-based NanoString BRCAness signature [11]. Signature scores were calculated using prescribed algorithms developed by NanoString technologies [34]. In the present study, we additionally compared the concordance of our DNA-based CNV BRCA1-like classi er with the RNA-based NanoString BRCAness signature.

Statistical analyses
Characteristics of patients were compared by BRCA1-like status using Fisher's exact, Chi-square or linearby-linear tests for categorical variables and Mann-Whitney U tests for continuous variables.
Recurrence-free survival (RFS) was de ned as time from randomization to local or distant invasive breast cancer recurrence, death from any cause, or to the last date of follow-up, whichever occurred rst. Overall survival (OS) was de ned as time from randomization to death from any cause or the last date of followup. Median follow-up was calculated using the reversed Kaplan Meier estimator. Survival curves were computed with the Kaplan-Meier method. To evaluate whether bene t from adjuvant capecitabinecontaining chemotherapy versus adjuvant conventional chemotherapy differs between BRCA1-like and non-BRCA1-like tumors, we applied Cox proportional hazards regressions that included the interaction term between treatment and BRCA1-like status. We estimated and compared interaction coe cients that were unadjusted and adjusted for the following variables: age at randomization, World Health Organization (WHO) performance status (0, 1), type of surgery (breast conserving, mastectomy), axillary surgery (dissection, sentinel node biopsy), T-stage (pT1, pT2, pT3), axillary nodal status (≤ 3 vs > 3 positive lymph nodes), histological type (ductal, lobular, other), and histological grade (1,2,3). Due to the relatively small number of events, interaction coe cients were adjusted for one covariate at a time. The prognostic effects of all covariates were also evaluated in separate models. The proportionality of hazards was checked using Schoenfeld residuals. A two-sided p-value < 0.05 was considered statistically signi cant. Statistical analyses were performed using SPSS version 25 (IBM Corp., Armonk, NY, USA) and Stata 16 (StataCorp., College Station, TX, USA).
To determine the concordance between our DNA-based CNV BRCA1-like classi er and the RNA-based NanoString BRCAness signature, we dichotomized the acquired continuous scores at the percentile of the established cut-off for the BRCA1-like classi er, as there is no prede ned cut-off for the NanoString BRCAness signature.

Patient characteristics
Of 202 TNBC patients included in the FinXX trial, we obtained BRCA1-like status for 129 (63.9%) patients (Fig. 1). Main reasons for failure were lack of available tumor tissue, low tumor percentage and insu cient amounts of isolated DNA. This subgroup did not differ substantially for the variables mentioned in Table 1 from those TNBC FinXX patients not included in the current analyses (Additional le 2: Table S1). Sixty-eight (52.7%) of the 129 tumors had a BRCA1-like pro le ( Table 1). As expected, patients with a BRCA1-like tumor had more frequently poorly differentiated tumors compared to patients with non-BRCA1-like tumors (P = 0.03) and had signi cantly more often less than four positive axillary lymph nodes (P = 0.047). Furthermore, BRCA1-like tumors had a higher T-stage (P = 0.03).  Interaction test between BRCA1-like status and chemotherapy regimen: P = 0.17; P = 0.09. ҂ All Cox proportional hazard analyses shown here were unadjusted for clinic-pathological variables. Similar results were obtained when adjusted for one covariate at the time (due to the relative small number of events).
DNA-based CNV BRCA1-like status versus RNA-based NanoString BRCAness signature Both scores of DNA-based CNV BRCA1-like classi er and the RNA-based NanoString BRCAness signature were available for 103/202 TNBC patients (Fig. 3)

Discussion
In the present study we found no evidence that bene t from adjuvant capecitabine-containing chemotherapy compared to conventional chemotherapy is associated with BRCA1-like status in earlystage TNBC patients. Therefore, it is unlikely that the BRCA1-like classi er will help in selecting patients for adjuvant capecitabine-enriched chemotherapy.

Our hypothesis of differential treatment effects was based on previous observations made by Van
Rossum et al. [26]. Their retrospective analysis of the randomized GAIN study suggested that patients with BRCA1-like tumors have a selective treatment bene t from a conventional dose-dense chemotherapy regimen, containing capecitabine, if compared with non-myeloablative, intensi ed dose-dense chemotherapy. There are three important differences between the GAIN study and the FinXX study. The main difference is that the FinXX study compared two regimens that only differed by the addition of capecitabine (yes/no) and a slightly lower dose of docetaxel in the capecitabine arm [27], while the GAIN study compared miscellaneous regimens that differed in drug dose and drug combinations [35]. Second, the GAIN study accrued only node-positive patients, while the FinXX study also recruited high risk nodenegative patients. And lastly, the capecitabine dose and schedule differed between the studies (10 cycles of capecitabine 1000 mg/m 2 twice daily administered on day 1-14 in a three weekly schedule in the GAIN trial versus 6 cycles of capecitabine 900 mg/m² twice daily on day 1-15 every 3 weeks in the FinXX study). Interestingly, we observed a signi cant capecitabine bene t for the non-BRCA1-like group (RFS and OS P value < 0.01) and less evidence of bene t for the BRCA1-like group (RFS P value 0.42; OS P value 0.59), in contrast to the observations of Van Rossum et al. [26]. Whether or not these two observations represent inconsistent patterns is di cult to determine due to the small sample size in these unplanned subgroup analyses. The patterns are actually reversed, so unless the direction of effects changes with more data, narrower con dence intervals will only make both studies less consistent with each other.
Two biological processes may explain why BRCA1-like status is not predictive for bene t of adjuvant capecitabine treatment. First, although capecitabine causes DNA-damage [5], it may not speci cally result in DNA damage that is dependent on a pro cient homologous recombination machinery resulting in errorfree DNA repair [36]. In fact, capecitabine and its active form 5-uorouracil lead to (1) DNA base pair mismatches which are repaired by the DNA mismatch repair (MMR) pathway [37] and (2) inhibition of DNA replication, leading to abasic sites that are repaired by base excision repair (BER) proteins [38].
Second, there are some data suggesting that capecitabine might result in better outcome in TNBC patients with grade 2 tumors, but not grade 3 [39]. Since BRCA1-like tumors are enriched for high-grade tumors, our ndings of no predictive value of BRCA1-like status for bene t of capecitabine support these observations.
Our results are in line with previous preclinical ndings of Quinn et al who observed no differential doseresponse effect of capecitabine in BRCA1-mutated compared with wild-type BRCA1 human BC cells [40], and with Alli et al who found a 5-fold lower sensitivity to 5-uoro-uracil of BRCA1-de cient compared to wild-type BRCA1 murine mammary epithelial cells [41]. Moreover, our ndings are consistent with the recent observations of Asleh et al. [11]. In the same dataset, they found no signi cant association of improved outcome and signatures associated with DNA damage repair, including HRD and BRCAness, using an 800-gene panel in 111 early-stage TNBC patients treated with adjuvant capecitabine in the FinXX trial. Additionally, we demonstrated 78.6% concordance between our DNA-based CNV BRCA1-like classi er and the RNA-based NanoString BRCAness signature used by Asleh et al. [11]. Taken together, these data indicate that bene t of adjuvant capecitabine-containing chemotherapy could not be predicted by de ciency in homologous recombination. Currently, the NordicTrip (ClinicalTrials.gov Identi er: NCT04335669) is an ongoing translational clinical trial in early-stage TNBC prospectively comparing the effect on pathologic complete response (pCR) rate of adding capecitabine to epirubicin plus cyclophosphamide followed by carboplatin plus paclitaxel neoadjuvant chemotherapy, strati ed for HRD positive versus HRD negative/ HRD-intermediate. Results of this study have to be awaited to further clarify the value of HRD as a predictive biomarker for bene t of capecitabine-containing chemotherapy in early-stage TNBC.
The main strength of our study is the study design, i.e., a prospective, randomized controlled trial with collection of archival material. This prospective-retrospective design is the rst choice to assess a putative predictive biomarker in case a prospective randomized clinical trial is not feasible, because such trials require huge numbers of patients, are costly and take many years to complete [42]. An additional strength, in contrast to the exploratory analyses of Asleh et al. using an RNA 800-gene panel without prede ned cutoff for the BRCAness signature [11], is that we evaluated a single biomarker with a prede ned cutoff based on prior biological and empirical evidence [21,[23][24][25]. This con rmatory objective is required to establish the implementation of a predictive biomarker or to refute it [42].
A limitation of the present study is the small sample size, which is due to the fact that the FinXX trial was powered to evaluate the main effect of capecitabine among patients with any biological type of breast cancer rather than a treatment-marker interaction in the subgroup of TNBC patients. In addition, the number of patients was further reduced by the failure to obtain BRCA1-like status for all TNBC patients for several reasons. However, our patient group accounts for 129 (63.9%) of the 202 accrued TNBC patient in the FinXX trial, which is within the recommended range of sample size for a study to evaluate predictive biomarkers [42]. Furthermore, the included patients did not differ meaningfully from all accrued cycles of docetaxel 80 mg/m² 3-weekly, followed by 3 cycles of cyclophosphamide 600 mg/m², epirubicin 75 mg/m², and uorouracil 600 mg/m², 3-weekly; TX+CEX: 3 cycles of capecitabine 900 mg/m² twice daily plus docetaxel 60 mg/m² 3-weekly, followed by 3 cycles of cyclophosphamide 600 mg/m², epirubicin 75 mg/m² and capecitabine 900 mg/m² twice daily, 3-weekly; WHO: World Health Organization; X: capecitabine Declarations Ethics approval and consent to participate The study was approved by the Ethics Committee of the participating medical institutions and the National Agency for Medicines, Finland. Patients supplied written informed consent to allow the use of their tumor tissue for clinical study related research purposes. The Institutional Review Boards at Helsinki University, Finland, approved the use of archival tissue for the current translational study.

Consent for publication
Not Applicable.

Availability of data and materials
The clinical datasets used and analyzed during the current study are available from Heikki Joensuu on reasonable request. The sequencing datasets generated and analyzed during the current study are available in the SRA repository, [https://www.ncbi.nlm.nih.gov/Traces/study1/?acc=PRJNA647428&].
Competing interests HJ has ownership interest (including patents) at Sartar Therapeutics and is a board member, has a coappointment at Orion Pharma and is employed by Orion Pharma, and has received fees from Neutron Therapeutics. SCL reports grants from ZonMw and A Sister's Hope during the conduct of the study. SCL is an advisory board member for AstraZeneca, Cergentis, IBM, P zer and Roche, and received institutional research grants from Agendia, AstraZeneca, Eurocept-pharmaceuticals and P zer. In addition, SCL received institutional research grants and institutional non-nancial support from Genentech, Novartis, Roche, Tesaro and Immunomedics and other institutional support from AstraZeneca, P zer, Cergentis, Daiichi Sankyo, IBM and Bayer outside the submitted work. MK is an advisory board member for BMS, Roche, MSD and Daiichi and received institutional research support from AstraZeneca, BMS and Roche outside the submitted work.
The other authors declare that they have no competing interests.

Funding
This work was supported by the Dutch Cancer Society (Project Number 10603). The funding source had no role in study design, data collection and analyses, decision to publish, or preparation of the manuscript.