Capecitabine Efficacy Is Correlated with TYMP and RB1 Expression in PDX Established from Triple-Negative Breast Cancers

Triple-negative breast carcinomas (TNBC), which are negative for estrogen/progesterone (ER/PR) receptor expression and lack HER2 overexpression, account for 15% to 20% of breast cancers. TNBC differ from other subtypes of breast cancer in terms of axillary lymph node involvement, locoregional recurrence, time to metastasis (these patients have a high risk of early relapse and much lower survival in the first three to 5 years after treatment) and patterns of distant recurrence (1). TNBC is emerging as a highly heterogeneous disease in terms of histologic features, clinical behavior and sensitivity to systemic treatments (1). It constitutes a major clinical challenge, as there have been no major improvements in the treatment of this subgroup of tumors, apart from chemotherapy (anthracycline- and taxane-based chemotherapy regimens), which has reduced mortality by about 30% (2).

Cytotoxic chemotherapy is currently the only treatment option for TNBC. For TNBC patients, a pathologic complete response (pCR) to neoadjuvant chemotherapy (NAC) is associated with prolonged overall and event-free survival times (3–5). Although the rate of pCR following standard NAC is significantly higher for TNBC as than for the other breast cancer subtypes, more than 50% of TNBC patients treated in the neoadjuvant setting have residual disease and are associated with an increased risk of distant recurrence, particularly during the first 3 years after treatment, and a poor outcome. This paradox may be explained by the presence of residual disease in more than half the patients (>60%). At present, there is no standard of care for adjuvant treatment in TNBC patients with residual disease after NAC, and chemotherapy in patients with metastatic disease is of limited efficacy.

Preclinical studies focusing on residual tumors after NAC could provide an important tool to identify efficient therapies and predictive biomarkers in tumors showing resistance to NAC (6). In this study, we established patient-derived xenografts (PDX) from residual breast tumors after NAC and found that capecitabine was an effective chemotherapy agent in 60% of TNBC models. We also identified the expression of thymidylate phosphorylase (TYMP) and RB1 as biomarkers predictive of the response to capecitabine.

Female Swiss nude mice were purchased from Charles River Laboratories and maintained under specific pathogen-free conditions. Their care and housing were in accordance with institutional guidelines and the rules of the French Ethics Committee (project authorization no. 02163.02). PDX were established from primary breast cancer patients with residual disease after neoadjuvant chemotherapy, with informed consent from the patient, in accordance with published protocols (7, 8). For 2 patients (#90 and #95), xenografts were established from a local breast cancer recurrence treated by NAC. These two patients were excluded from the disease-free survival analysis.

Histology and IHC status (ER, PR, and HER2) was determined for the PDX and compared with that of the patient tumors, as described elsewhere (7, 9). The PDX derived from primary tumors have been described elsewhere (7, 8, 10).

Adriamycin (DOX, doxorubicin, Teva Pharmaceuticals), docetaxel (Sanofi Aventis), cisplatin (CDDP, Teva), and cyclophosphamide (Endoxan, Baxter) were administered by the intraperitoneal (i.p.) route, at doses of 2, 20, 6, and 100 mg/kg, respectively, every 3 weeks. Capecitabine (Xeloda, Roche Laboratories) was administered per os at a dose of 540 mg/kg/day, q5d × 6 weeks. Paclitaxel (Fresenius Kabi) was administered weekly, at a dose of 20 mg/kg (i.p.), and bevacizumab (Roche) was administered at a dose of 10 mg/kg (i.p.) every 3 weeks. Trastuzumab (Roche) was administered weekly, at a dose of 10 mg/kg. Fluorouracil (5-FU; Fresenius Kabi) was administered at a dose of 50 mg/kg (i.p.) and navelbine (Pierre Fabre Laboratories) was administered at a dose of 3 mg/kg (i.v.) every 2 weeks. The doses for drug combinations were determined in tolerance studies, based on human equivalences (11) and clinical schedules.

Relative tumor volumes (RTV) and the percent tumor growth inhibition (TGI) were calculated as previously described (7). Kaplan–Meier survival analysis was performed on HBCx-95 xenografts treated with various chemotherapy regimens, with GraphPad Prism software. An event was defined as a quadrupling of tumor volume from the initial tumor volume (12).

Percent change in tumor volume was calculated for each tumor as (Vf − V0/V0) × 100, where V0 is the initial volume (at the beginning of treatment) and Vf is the final volume (at the end of treatment). Tumor regression (R) was defined as a decrease in tumor volume of at least 50% taking as reference the baseline tumor volume; at least a 35% increase in tumor volume identified progressive disease (PD) and responses that were between +35% and −50% were considered as stable disease (SD; ref. 13).

Categoric variables were analyzed with the Fisher exact test. The statistical significance of TGI was determined in Mann–Whitney U tests.

We analyzed the clinical parameters of 12 patients in the residual disease group (2 patients receiving neoadjuvant chemotherapy for local breast cancer relapse were excluded) and 30 patients in the primary tumor group.

The clinical data for the patients are expressed as frequencies for qualitative variables or medians and associated range for quantitative variables. They were compared in χ² or Fisher exact tests, as appropriate. Disease-free probabilities were estimated by the Kaplan–Meier method, and survival curves were compared in log-rank tests. Analyses were performed with survival and ggplot2 R libraries.

DNA extraction and targeted next-generation sequencing of TP53, BRCA1, and BRCA2 were performed by the ICGex NGS platform of Institut Curie, as previously described (14). Germline BRCA1 and BRCA2 mutations (BRCA1: p.E849X; BRCA2: p.T1282fs) were present in both the initial tumors and the PDX.

RNA extraction and RT-qPCR were performed as previously described (10, 15). For gene normalization, we used the human TATA box-binding protein (TBP, GeBbank accession no. NM_003194). We used protocols for cDNA synthesis and PCR amplification described in detail elsewhere (16, 17). The results are expressed as N-fold differences in target gene expression relative to the TBP gene. In the pharmacodynamics analysis, these N-fold differences values for each xenograft were subsequently normalized to yield a median value of 1 for the control xenograft group.

GeneChip Human 1.1 ST arrays were hybridized according to Affymetrix recommendations, using the Ambion WT Expression Kit protocol (Life Technologies) and Affymetrix labeling and hybridization kits. All analyses were performed with R software (version 3.3; R Core Team 2017) and Bioconductor packages (version 3.4; ref. 18). Arrays were normalized according to the RMA normalization procedure from the oligo package (19). Differential expression between PDX responding and resistant to capecitabine was assessed by a linear model method, with the Limma package (20). We used the false discovery rate (FDR) procedure to adjust for multiple testing. We considered log fold changes in expression > 1.5, and P values < 0.05 to be statistically significant.

The HBCc-12A cell line was established from the residual TNBC PDX HBCx-12A by mechanical dissociation and removal of cellular debris and erythrocytes. The HBCc-12A cell line and the HBCx-12A PDX were genotyped with the GenePrint 10 System (#B9510, Promega) that allows coamplification and three-color detection of nine human loci. The cell line was grown at 37°C in DMEM Glutamax medium supplemented with 10% fetal bovine serum, 1% penicillin–streptomycine, 1% Hepes buffer, 1% sodium pyruvate, and 10 μg/mL insulin (Sigma).

HBCc-12A cells were transfected with 20 nmol/L siRNA against RB1 (SI00007091, QIAGEN) and AllStars siRNA as negative control (SI03650318, QIAGEN) using INTERFERin reagent (POL409-50, Ozyme) in Opti-MEM medium (Life Technologies), according to the manufacturers' instructions.

Cells were plated in duplicate into 96-well culture plates at a density of 7 × 10³ cells per well and left overnight. siRNA and 5-FU were added 24 and 30 hours after seeding, and on day 5, cell proliferation was determined by the MTT assay (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide; MTT; M-2128, Sigma).

For Western blot analyses, 115 × 10³ HBCc-12A cells were plated into 6-well culture plates. siRNA and 5-FU were added at the same concentrations and timelines of MTT proliferation assays. Proteins were extracted as described previously (8). Lysates were resolved on 4% to 12% TGX gels (Bio-Rad), transferred into nitrocellulose membranes (Bio-Rad), and immunoblotted with rabbit antibodies against GAPDH (#2118, Cell Signaling Technology), RB1 (#9309, Cell Signaling Technology), p-RB1 (#9308, Cell Signaling Technology). After washes, membranes were incubated with the appropriate secondary antibodies horseradish peroxidase–conjugated affinity-purified goat anti-rabbit (111-035-045, The Jackson ImmunoResearch Laboratories, Inc.).

Xenografted tumors were fixed in 10% neutral-buffered formalin, embedded in paraffin and stained with hematoxylin and eosin. Immunostaining was performed on a Discovery XT Platform (Ventana Medical System, part of Roche Diagnostics; ref. 17). The slides were incubated with a polyclonal rabbit antibody against TYMP (#HPA001072, Sigma), a monoclonal mouse antibody against RB1 (#9309, clone 4H1, Cell Signaling Technology), and a monoclonal rabbit antibody against TYMS (#9045, Ozyme). Slides immunostained with mouse and rabbit IgG were used as negative controls. Slides were incubated with anti-rabbit/mouse secondary antibodies (horseradish peroxidase complex) and 3,3′-diaminobenzidine tetrahydrochloride (DAB) as the substrate for color development (ChromoMap Kit with Anti-rabbit OmniMap, Ventana Medical System). TYMP immunostaining was assessed by determining the intensity and distribution of stained cancer cells, as previously described (21). Immunostaining intensity was scored as follows: 0, negative staining; 1, weak staining; 2, moderate staining; and 3, strong staining. Immunostaining intensity scores were grouped into three classes, as follows: (i) ≤33% of the tumor cells stained; (ii), 33% to 66% of the tumor cells stained; and (iii) 66% to 100% of the tumor cells stained. Summed scores were also obtained by adding the scores for two parameters (21).

We established 22 PDX by transplanting 53 breast tumors from patients with residual disease after NAC. The take rates for the ER⁺, HER2⁺, and triple-negative subtypes were 23%, 22%, and 78%, respectively (Supplementary Table S1). When compared with PDX derived from treatment-naïve primary tumors, previously reported (7, 9), take rates of residual tumors were higher for ER⁺ and triple-negative subtypes (Fig. 1A; Supplementary Table S1). In the TNBC group, median PDX latency after engraftment was 67 days for residual tumors, versus 160 days for treatment-naïve primary tumors (Fig. 1B). With a high tumor take and a short latency, residual TNBC PDX may provide an opportunity for preclinical testing of anticancer agents prior to relapse in patients. We therefore retrospectively compared the metastasis-free survival (MFS) of the patients with the latency periods of the 14 PDX established from the residual tumors (Fig. 1C). Ten of these patients presented a recurrence of metastatic disease, 3 remain metastasis free and 1 died 5 months after mastectomy (HBC-12A, an 89-year-old patient). PDX latencies were considerably shorter than MFS for 8 patients (P = 0.021, paired t test).

The clinical characteristics of patients in the residual disease and primary tumors groups are summarized in Supplementary Table S2; Supplementary Fig. S1. Prognosis was worse in the residual disease group of patients, for whom survival was 28% at 25 months, versus 75% for the patients in the primary tumor group (P = 0.048, log-rank test; Fig. 1D).

These results highlight the aggressive phenotype of residual tumors, with high tumor take rates, a short tumor latency during PDX establishment, and low survival of the corresponding patients.

We determined the response to chemotherapy in 10 PDX derived from residual tumors and compared this response with that of PDX derived from primary tumors (7). The chemotherapy regimens used were AC (adriamycin plus cyclophosphamide), docetaxel, cisplatin, and capecitabine. Only 10%, 20%, and 0% of residual TNBC PDX responded to AC, cisplatin, and docetaxel, with a TGI of at least 90%, whereas 50%, 52%, and 10% of treatment-naïve TNBC PDX responded to these treatments (Fig. 1E). Conversely, residual TNBC PDX displayed a higher frequency of response to capecitabine (60%) as compared with treatment-naïve PDX (36%).

Table 1 summarizes the NAC administered to patients, tumor histology, status for BRCA1/2 and TP53 mutations, and response to chemotherapy, as assessed by TGI and adapted RECIST criteria (13), for 10 residual TNBC PDX.

Only one residual PDX responded to AC with a TGI≥ 90%, none responded to docetaxel and two (one with a BRCA1 mutation and the other with a BRCA2 mutation) responded to cisplatin (with tumor regression observed in only one of these cases; Table 1 and Supplementary Fig. S2A).

Six of the 10 residual PDX responded to capecitabine with a TGI ≥ 90: 4 (HBCx-12A, HBCx-63, HBCx-92, and HBCx106) with SD and 2 models with tumor regression (HBCx-39 and HBCx-95; Table 1). Three of the four capecitabine-resistant PDX corresponded to metaplastic breast cancer. The drug response profiles of 3 capecitabine-responder PDX are illustrated in Fig. 1F. Response to capecitabine was homogeneous in the different mice (Fig. 1G). HBCx-95 capecitabine-treated xenografts showed complete responses at day 40: the xenografts of this model were monitored for 10 months after the end of treatment. A Kaplan–Meier survival curve (Supplementary Fig. S2B) showed that event-free survival was greater for animals treated with capecitabine than for animals receiving other types of chemotherapy (P = 0.0004, log-rank test).

In the advanced setting, capecitabine is usually administered as a second-line or third-line treatment. We hypothesized that prior treatment of tumors with a different chemotherapy regimen might affect the response to capecitabine. We tested this hypothesis in one residual TNBC PDX, through the sequential administration of chemotherapies administered to the patient in the adjuvant and metastatic settings. Patient #63 received FEC and docetaxel as NAC, followed by 5-FU and navelbine as an adjuvant treatment, then paclitaxel + bevacizumab (first line) and capecitabine (second line) in the metastatic setting (Fig. 2A). We administered the same treatments sequentially to the HBCx-63 PDX, established from the residual tumor, and compared it with the efficacy of capecitabine in non pretreated tumor. Results are shown in Fig. 2B: capecitabine given in xenografts pretreated by 5-FU + navelbine or by paclitaxel + bevacizumab resulted in PD in 4 of 5 and 6 of 6 xenografts (mean of 144% and 130%, respectively). By contrast, 7 of 9 xenografts treated with capecitabine without prior treatment displayed SD, 1 PD and 1 tumor regression (mean = 14%). Thus, prior treatment by 5-FU and navelbine decreased capecitabine efficacy.

We next analyzed the expression of two enzymes whose expression may influence capecitabine response: thymidylate synthase (TYMS), the target enzyme for 5-FU, and TYMP, an enzyme involved in the metabolism of 5-FU and capecitabine (22). TYMS expression was increased after treatment by 5-FU and navelbine, and this increase was maintained in xenografts treated by paclitaxel and bevaciuzumab (Fig. 2C). By contrast, TYMP expression was unchanged by the different treatments (data not shown).

For the identification of markers predictive of capecitabine response, we combined the capecitabine response data for 10 residual TNBC PDX with those obtained from 22 primary tumor–derived TNBC PDX previously determined (unpublished data). The responses to capecitabine treatment of the 32 TNBC PDX are represented as waterfall plots in Fig. 3A: in 15 PDX, capecitabine treatment resulted in SD or tumor regression, whereas in the other 17 PDX, it resulted in PD. To identify genes associated with the capecitabine response, we performed a differential analysis of gene expression microarray datasets available for 25 TNBC PDX (15 resistant and 10 sensitive). Hierarchical clustering based on 37 genes (17 upregulated and 20 downregulated in sensitive PDX) grouped 100% of sensitive PDX together (Fig. 3B). The 37 genes concerned are listed in Supplementary Table S3. Three of these genes encode proteins involved in cell-cycle regulation: RB1 and CCND1, which were upregulated in responder models, and CDKN2A, which was upregulated in resistant models. Among genes implicated in capecitabine metabolism (TYMP, TYMS, TK1, and DPYD), only TYMP expression was higher in responder models, although the fold change did not pass the threshold for global differential analysis (Supplementary Fig. S3).

To validate RB1 at the protein level, we analyzed its expression by IHC in the 32 TNBC PDX. RB1 protein was detected in all 15 (100%) responder PDX and in 6 of 17 resistant (35%) PDX (Fig. 3C and D). Positive and negative predictive values were 71% and 100%, respectively (P = 0.0001, Fisher exact test).

We then quantified TYMP protein expression in the 32 TNBC PDX by determining the IHC H-score (21). Examples of negative and positive TYMP IHC results are shown in Fig. 3E. Thymidine phosphorylase IHC H-score was significantly higher in responder tumors (Fig. 3F). Median TYMP IHC H-score was 4.25 (range, 0–6): 80% of the responding PDX had TYMP expression scores ≥4.25 and 76% of resistant PDX had scores < 4.25. Positive and negative predictive values were 75% and 81%, respectively (P = 0.0038; Fisher exact test). Among the 6 RB1-positive PDX that did not respond to capecitabine, 5 showed no or low expression of TYMP (data not shown). The combination of RB1 positivity and a TYMP score> 4.25 had a positive predictive value of 92% and a negative predictive value of 100%, accurately predicting response in 12 of 15 xenografts and resistance in 8 of 17.

Finally, we assessed capecitabine efficacy in one PDX derived from a patient with HER2-positive breast cancer treated with trastuzumab in the preoperative setting (HBCx-73). The xenograft was RB1-positive and expressed high levels of TYMP (Supplementary Fig. S4A). Capecitabine had strong antitumor activity against this xenograft (TGI = 99%), whereas trastuzumab did not arrest tumor growth (Supplementary Fig. S4B).

RB1 plays a critical role in coordinating the expression of genes required for cell-cycle progression through interaction with the transcription factor E2F. We therefore looked at the correlation between capecitabine's efficacy and the expression changes of 8 genes included in the “RB1 gene expression signature” and involved in cell proliferation (CCNA2, CCNB1, TOP2A, TTK, PLK1, and AURKA) and nucleotides metabolism (TYMS and TK1; ref. 23). We analyzed control and posttreatment xenografts of 7 PDX, including both capecitabine-resistant and capecitabine-sensitive PDX (Fig. 4A). The expression levels of the eight genes, determined by RT-PCR, are shown in Supplementary Fig. S5. A heatmap representing the fold-change difference between capecitabine-treated and control xenografts is shown in Fig. 4B. All genes were strongly downregulated in the three RB1-positive PDX with the highest levels of TGI (HBCx-63, HBCx-12A, and HBCx-39), whereas no change was found in the three PDX in which capecitabine had little or no effect. The correlation between fold-change differences in gene expression and TGI by capecitabine is shown in Fig. 4C and was statistically significant for all genes.

We tested whether RB1 inhibition could have an impact on capecitabine response in the HBCc-12A cell line, established from the RB1-positive HBCx-12A PDX (residual tumor-derived PDX). This PDX responds to capecitabine with a TGI of 96% (SD with RECIST criteria; Fig. 5A). We performed a cell viability assay to determine the in vitro response of the cell line to 5-FU: the cell line showed an IC₅₀ of 33 μmol/L (Fig. 5B). The cell line was then transfected with 2 different siRNA directed against RB1 or control (scramble siRNA) and exposed to 40 μmol/L of 5-FU for 120 hours. Western blot analysis demonstrated downregulation of RB1 protein in cell exposed to siRB1 and siRB2, with an increased inhibition for siRB1, in both untreated and 5-FU–treated cells (Fig. 5C). Phosphorylation of RB1 protein was inhibited by 5-FU. Inhibition of RB1 decreased sensitivity to 5-FU of 40% as assessed by the MTT viability assay (the increased viability was greater for siRB1 as compared with siRB2; Fig. 5D). In summary, these results show that RB1 has a role in determining the cellular response to 5-FU treatment.

We established a cohort of breast cancer PDX models by engrafting residual tumors after NAC. Residual TNBC PDX had a higher tumor take rate and a shorter latency than TNBC PDX derived from treatment-naïve breast cancers. Both these characteristics are suggestive of an aggressive phenotype. These results are similar to those reported by McAuliffe and colleagues who showed that take rates were higher for breast cancers treated in the neoadjuvant setting (24). The aggressiveness of these tumors was confirmed by the poor outcomes of the corresponding patients. In more than half the TNBC patients, PDX latency was shorter than MFS. The residual tumor PDX therefore becomes available for in vivo drug testing over a time frame compatible with clinical decision-making concerning first-line chemotherapy. Furthermore, residual TNBC PDX were resistant to anthracyclines- and docetaxel-based chemotherapy. As these drugs were given to patients in the neoadjuvant setting, in which they were of limited efficacy, the resistance phenotype of the patients' tumors is maintained and reproduced in xenograft models. Residual tumor PDX were also resistant to cisplatin, with the exception of PDX established from BRCA1/2-mutated breast cancers. Despite the small size of this series of xenograft models, this finding indicates that residual tumors display a multidrug resistance phenotype extending to different chemotherapy classes, including anthracyclines, cyclophosphamide, docetaxel, and cisplatin. It also suggests that patients with BRCA1- and BRCA2-mutated tumors may still respond to platin-based chemotherapies. The patient with the BRCA1-mutated tumor has not yet relapsed, and the patient carrying the BRCA2-mutated tumor displayed a partial response to six cycles of first-line treatment with paclitaxel and carboplatin. However, given the small number of cases, these results require confirmation in a larger cohort.

Responses to capecitabine were observed in 6 of the 10 residual tumor-derived PDX displaying cross-resistance to anthracyclines, docetaxel, and cisplatin. Three of the four capecitabine-resistant PDX were established from metaplastic breast cancers, a histologic subtype of TNBC known to be particularly aggressive and resistant to chemotherapy (25).

These findings are of particular interest for several reasons. First, they are consistent with the results of a recent clinical trial (CREATE-X) demonstrating the efficacy of adjuvant capecitabine treatment in TNBC patients with residual invasive disease after neoadjuvant chemotherapy (26). Second, capecitabine treatment is well tolerated and may be compatible with adjuvant treatment in patients with a high risk of tumor recurrence. The FinXX trial also reported beneficial effects on survival in TNBC patients treated with a capecitabine-containing regimen in the adjuvant setting (27).

In the CREATE-X trial, TNBC disease-free survival was 69.8% in the capecitabine group, versus 56.1% in the control group (26). Thus, a large proportion of patients remained that did not benefit from adjuvant capecitabine treatment. Predictive biomarkers are crucial for identifying the patients most likely to benefit from treatment.

Based on the analysis of transcriptomic data, we identified and validated RB1 as a biomarker predictive of the response to capecitabine in TNBC PDX. RB1 is a tumor suppressor protein that regulates cell-cycle progression by preventing entry into the mitotic phase (28). It may be inactivated by hyperphosphorylation or protein loss, both of which disrupt its association with the transcription factor E2F, facilitating the expression of a transcriptional program leading to progression through S-phase and mitosis (29). RB1 is also a key regulator of cell-cycle arrest upon DNA damage (30). The finding that capecitabine efficacy is correlated with RB1 expression may be explained by the role of this protein in cell-cycle control and the checkpoint response to chemotherapy. Consistent with this hypothesis, RB1 reconstitution in RB1-deficient lung cancer cell lines reestablishes the regulation of RB1/E2F target genes and restores the G₁ checkpoint response to chemotherapy regimens, including 5-FU (31). Similarly, inactivation of the RB1 gene by siRNA in a cell line derived from a capecitabine-responder PDX decreased sensitivity to 5-FU, further supporting a role of RB1 in determining the response to 5-FU/capecitabine.

In addition to its role in canonical cell cycle–dependent regulation, RB1 also inhibits cell proliferation by regulating nucleotide metabolism by inhibiting enzymes involved in dNTP synthesis. The absence of functional RB1 leads to a deregulation of the expression of dNTP enzymes (such as TYMS) and antimetabolite resistance (32). Li and colleagues demonstrated that a lack of functional RB1 increases resistance to antimetabolite drugs in human sarcoma cells, by increasing the expression of the DHFR and TYMS genes, both of which are under the control of the E2F factor and targeted by antimetabolite chemotherapy (33). The expression of genes encoding proteins involved in nucleotide metabolism (TYMS) and the RB1-dependent S-phase DNA damage response (CCNA2; ref. 34) was strongly inhibited in RB1-positive models treated with capecitabine, suggesting that the canonical and metabolic regulation of cell proliferation may together account for the observed antitumor effects of capecitabine.

From the analysis of transcriptomic data, RB1 and CDKN2A expression appeared inversely correlated, a finding that is in line with previous reports (35). CDKN2A gene encodes the p16(INK4a) protein that plays an important role in cell cycle through the regulation of the cyclin-dependent kinase (CDK) 4/6 and cyclin D complexes. Loss of p16(INK4a) can occur by genetic or epigenetic mechanisms and results in RB1 protein activation, leading to unchecked cellular proliferation (36). Evaluation of p16(INK4a) expression by IHC analyses in TNBC PDX will be necessary in order to validate it as potential biomarker of capecitabine resistance.

Finally, an association has been found between the RB1 pathway and chemotherapy response in breast cancer patients (28). In metastatic breast cancer, mutations of the RB1 gene are associated with a poor response to anthracyclines (37, 38). RB1 pathway inactivation is predictive of resistance to DNA damage–based treatments in locally advanced breast cancer (39).

Capecitabine is converted to 5-FU in cancer cells by the TYMP enzyme (40). Thymidine phosphorylase also activates 5-FU, by converting it to fluorodeoxyuridine (FUDR), a precursor of FdUMP that inhibits the TYMS enzyme, responsible for de novo thymidylate synthesis (22). Accordingly, capecitabine-based chemotherapies have been reported to be more effective in tumors with high levels of TYMP (41–44). Expression of TYMP, as assessed by IHC, was significantly correlated with the response to capecitabine in the cohort of 32 TNBC PDX (including PDX derived from both residual and treatment-naïve tumors). These results are consistent with those of various different clinical studies showing a correlation between TYMP expression and response to capecitabine in metastatic breast cancer patients previously treated with anthracyclines and taxanes (42, 45).

The evaluation of TYMP and RB1 as predictive biomarkers could be translated into clinical practice. Immunohistochemical analyses of one of these two biomarkers on residual tumors at surgery could identify those patients likely to benefit from adjuvant capecitabine treatment. Retrospective validations of these biomarkers in clinical trials evaluating capecitabine against controls (like the CREATE-X study) could also be used to provide support for our results.

Our results may also have implications for the advanced cancer setting, as capecitabine is usually given as a second-line or third-line treatment in patients with recurrences after NAC (46, 47). The efficacy of capecitabine was decreased after pretreatment of tumors with 5-FU, and this was associated with increased expression of TYMS. The upregulation of TYMS in 5-FU–treated tumors is consistent with previous finding in cell lines and tumors and can be explained by the inhibition of a negative feedback mechanism in which TYMS protein inhibits the translation of TYMS mRNA (48). When bound to the active metabolite of 5-FU (FdUMP), TYMS is not able to bind to its mRNA, which results in increased protein expression (49). This constitutes a mechanism of acquired resistance to 5-FU (50, 51). This case suggests that capecitabine might be more effective if introduced earlier, as adjuvant or first-line treatment, and raises the question of giving 5-FU prior to capecitabine.

In conclusion, we have developed a panel of PDX models from breast cancer patients with residual disease after neoadjuvant treatment. These chemoresistant PDX provided us with an opportunity to identify effective treatment strategies that might improve the survival of patients with a high risk of relapse and poor prognosis. We showed that capecitabine was effective against 60% of TNBC PDX derived from tumors previously treated with anthracyclines and taxanes, and we identified TYMP and RB1 expression as putative biomarkers predictive of the response to capecitabine.

No potential conflicts of interest were disclosed.

Conception and design: E. Marangoni, S. Roman-Roman, F. Reyal

Development of methodology: S. Château-Joubert, J.-L. Servely, P. Foidart, N.E. Sounni, F. Reyal

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Coussy, R. El-Botty, S. Château-Joubert, J.-L. Servely, F. Assayag, A. Dahmani, E. Montaudon, F. Nemati, J. Fleury, D. Gentien, A. Rapinat, P. Foidart, A. Noel, A. Vincent-Salomon, M. Lae, I. Bièche, F. Reyal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Marangoni, C. Laurent, F. Coussy, S. Château-Joubert, J.-L. Servely, S. Vacher, P. Foidart, S. Roman-Roman, I. Bièche, F. Reyal

Writing, review, and/or revision of the manuscript: E. Marangoni, F. Coussy, R. El-Botty, N.E. Sounni, A. Noel, M. Lae, S. Roman-Roman, M. Piccart, F. Reyal

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. El-Botty, L. de Plater, F. Assayag, A. Dahmani, E. Montaudon, D. Decaudin, I. Bièche, F. Reyal

Study supervision: E. Marangoni, F. Assayag, F. Nemati, F. Reyal

This work was supported by grants from the Association pour la Recherche Contre le Cancer (ARC) and the Parisian Alliance of Cancer Research Institutes (PACRI). High-throughput sequencing was performed at the ICGex NGS platform of the Institut Curie, which is supported by the ANR-10-EQPX-03 (Equipx) and ANR-10-INBS-09-08 (France Génomique Consortium) grants from the Agence Nationale de la Recherche (“Investissements d'Avenir" program), Canceropole Ile-de-France and the SiRIC-Curie program—SiRIC Grant “INCa-DGOS-4654.” We thank the animal platform of the Institut Curie (Isabelle Grandjean) and Julie Sappa of Alex Edelman and Associates for her help in correcting the English version of the article.

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.