Background

Neoadjuvant chemotherapy (NCT) improves clinical outcome as patients whose tumors respond to NCT have superior disease-free and overall survival than patients whose tumors do not (Rastogi et al. 2008). Thus, predicting the chance of response before starting NCT is crucial to identify patients who will benefit from NCT and help to select for the appropriate drugs. Up to now, several markers such as tumor size, histological type, hormone receptor status, and HER2 expression are used to predict efficiency of NCT. However, despite the consideration of these tumor characteristics, heterogeneity in therapy efficacy still subsists and accurate predictive markers for NCT are lacking.

Genetic variations such as single nucleotide polymorphisms (SNPs) are present across individuals and might affect pharmacokinetics and pharmacodynamics of drugs. SNPs in genes encoding drug transporters, drug metabolizing enzymes and target genes can directly affect the drug uptake, its activation and its excretion, accessibility of the target and also related pathways that can potentiate its action. Consequently, SNPs might lead to therapeutic failures and/or adverse drug reactions. Currently, UDP glucuronosyltransferase 1A1polymorphism UGT1A1*28 is used to predict toxicity of irinotecan in colorectal cancer treatment. Genotyping allows individuals homozygous for the variant allele of UGT1A1 to be treated with reduced doses of irinotecan and to avoid severe toxicity.

The aim of our study was to identify SNPs as predictive markers for breast cancer NCT. In our institution, breast cancer women treated with NCT receive 5-FU, cyclophosphamide, anthracyclines and/or taxanes. Thus, we selected SNPs previously reported in the literature to be associated with response to these chemotherapeutic agents. Thirty-seven SNPs, located in genes encoding influx and efflux transporters such as SLCO1B3, MDR1 and ABCC4, in genes belonging to drug metabolism (CYP1B1, CYP2B6, CYP3A4, MTHFR and GSTP1…) and also in genes of DNA repair pathway such as XRCC1, ERCC1, BRCA1 and p53 (Additional file 1: Table S1) were investigated in breast cancer patients receiving NCT.

Patients and methods

Patients

191 women with histological proven breast adenocarcinoma were enrolled in the study between November 2007 and January 2012 (ClinicalTrials.gov identifier: NCT00959556). Patients were treated with anthracyclines and/or taxanes based NCT in our institution (Centre Oscar Lambret, Lille, France). Among them, 118 received FEC100-Taxotere, 46 patients with HER2+ tumors were treated with FEC100-Taxotere-Herceptin, 16 patients were treated with FEC100 and 11 patients received diverse chemotherapy regimens.

Patient characteristics are summarized in Table 1 for the two main groups. Median age at diagnosis was 46.5 years for patients treated with FEC100-taxotere and 45.5 years for patients treated with FEC100-taxotere-herceptin. The median size of the tumors was 30 mm for patients treated with FEC100-taxotere and 30.5 mm for patients treated with FEC100-taxotere-herceptin. The histoprognostic grading was established according to Scarff Bloom and Richardson modified by Elston and Ellis (1998). FEC100-Taxotere was mostly administrated during 6 cycles (84.7%) and the majority of patients (82.2%) received 3 cycles of FEC100 followed by 3 cycles of Taxotere. The median number of treatment cycles of Herceptin per patient was 18 (3–18).

Table 1 Characteristics of patients treated with FEC100-taxotere and FEC100-taxotere-herceptin
Full size table

Our study only focused on patients treated with FEC100-Taxotere (n = 118), in order to identify potential predictive polymorphisms in a homogeneous group according to treatment.

Clinical response assessment

Clinical evaluation of response to therapy was assessed by measuring tumor size (physical examination and ultrasonography) before NCT, and after 3 and 6 cycles. Change in tumor size was determined by comparing the tumor size before and after NCT according to RECIST guidelines, version 1.1 (Eisenhauer et al. 2009).

Pathological response assessment

Tumor response was assessed by a pathologist and graded according Sataloff et al. (1995). Pathological tumor response was used as the gold standard to evaluate treatment response. On the basis of patients’ pathological response in primary breast site and/or in the lymph nodes, patients were designated having a pCR when surgical samples showed total or near-total effect and absence of nodal involvement (TA and NA or NB).

Hormone receptors and HER2 assessment

Immunohistochemistry (IHC) was used for evaluation of oestrogen receptors (ER) (1D5 Dako clone, then SP1 Ventana clone) and progesterone receptors (PR) (1A6 Dako clone, then PR88 Biogenex clone); to be positive, more than 10% of the tumor nucleus cells had to be stained. Evaluation of HER2 was performed with IHC (CB11 Biogenex clone, then HER 485 Dako clone); only tumors classified 3+ were considered positive. All tumors HER2 2+ with IHC were tested by FISH. Triple negative tumors were defined as ER and PR negative and HER2 negative by immune histochemistry (IHC) or by FISH in case of HER2 2+.

DNA extraction

Genomic DNA was extracted from EDTA-treated blood samples (300 µl) using the MagNA Pure Compact instrument (Roche Diagnostics, Meylan, France) and the MagNA Pure Compact Nucleic Acid Isolation Kit according to the manufacturer’s instructions.

Genotyping

SNP genotyping was assessed using allelic discrimination with SNPType assays ordered from Fluidigm. When SNPType assays were not available, TaqMan assays (Life Technologies) were used instead. Because DNA concentration available was below supplier recommendation, a Specific Target Amplification was performed to enrich targeted SNP sequences according to the supplier’s instructions.

Instrumentation and nanofluidic chips

48.48 Dynamic Array used in the present study are nanofluidic chips able to analyze 48 samples with 48 SNP assays on the BioMark platform (Fluidigm). The BioMark system is used to thermal cycle these nanofluidic chips and image the data in real time.

SNP genotyping using SNPType assays

3 µL of SNPType ASP1/ASP2 (allele specific primer 1 and 2) and 8 µL of SNPtype LSP were premixed and dilute with 29 µL of DNA suspension buffer to prepare SNPType assay mix. Each assay (5 µL) comprised 1 µL SNPType assay mix, 2.5µL assay loading reagent 2X (fluidigm) and 1.5 µL DNA-free water. Each sample (6 µL) comprised 3 µL Biotium fast probe master mix 2X (Biotium), 0.3 µL sample loading reagent 20X (fluidigm), 0.1 µL SNPType reagent 60X (Fluidigm), 0.036 µL ROX 50X (Invitrogen), 0.064 µL DNA-free water and 2.5µL amplified genomic DNA. Each of the assays (4 µL) and samples (5 µL) was pipetted into separate inlets in the chip. Amplification was carried out under the following conditions: 95°C for 5 min, 38 cycles of 95°C for 15 s, 60°C for 45 s, 72°C for 15 s followed by 30 s at 20°C for fluorescence measurement.

SNP genotyping using TaqMan assays

Each assay (5 µL) comprised 2.5 µL assay loading reagent 2X (fluidigm), 0.25 µL ROX 50X (Invitrogen), 1.25 µL SNP Genotyping assay mix 40X (Applied Biosystems) and 1 µL DNA-free water. Each sample (6 µL) comprised 3 µL TaqMan universal PCR master mix 2X (Applied Biosystems), 0.3 µL sample loading reagent 20X (fluidigm), 0.3 U AmpliTaq gold polymerase (Applied Biosystems), 0.12 µL DNA-free water and 2.52 µL amplified genomic DNA. Each of the assays (4 µL) and samples (5 µL) was pipetted into separate inlets in the chip. Amplification was carried out under the following conditions: 50°C for 2 min, 98°C for 10 min, 40 cycles of 95°C for 15 s, 60°C for 60 s. Fluorescence was measured at each slice.

Software

The data were analyzed using the BioMark SNP Genotyping Analysis software version 3.1.2 to obtain genotype. Briefly, the software calculates the FAM, VIC or HEX fluorescence intensities relative to ROX fluorescence background, and then automatically classifies the samples into three possible genotypes.

Statistical analysis

Clinical and histopathological characteristics were presented as frequencies and percentages for categorical variables and as medians and range for continuous variables. Associations between pCR and clinicopathological characteristics were assessed using Khi-2 test for qualitative variables or Fisher exact test in the case of small counts. Associations between pCR and continuous variables were performed using Wilcoxon Mann–Whitney test.

After ensuring that Hardy–Weinberg equilibrium was respected, the search of SNP correlated to pCR was performed by univariate logistic regression (R SNPassoc package). For each SNP, analyses were done considering genotypes separately or grouped to compare each homozygous or heterozygous genotype to other ones (only significant results are shown). Univariate analyses were repeated on subgroups of the population according to ER tumor status.

Multivariate analysis was performed on whole population combining significant SNP in a stepwise multivariate logistic regression selecting variables according to the Akaike Information Criterion (AIC). The model was validated internally on 1,000 random samples with replacement on the whole dataset. The percentage of times each variable was selected was extracted. Only those variables which were selected in >80% of models were retained. The final model was adjusted for ER status.

Results of statistical tests were considered significant at the 5% level. Analyses were performed using v11.2 Stata software (StataCorp. 2009. Stata Statistical Software: Release 11. College Station, TX, USA) and v2.15.2 R software (R Core Team 2012).

Results

Association between pCR and clinicopathological characteristics

25 patients (21.2%, CI 95%: 14.2–29.7%) reached complete pathologic response according to Sataloff criteria. Median age at diagnosis was 46 years for responder patients, 47 years for non-responder patients and was not significantly correlated to pCR.

The associations between pCR and clinicopathological features are summarized in Table 2. No statistically significant correlations were observed between pCR and tumor size, nodal status or KI67. Patients with higher tumor grade were significantly more likely to achieve a pCR than those with lower tumor grade (p = 0.009). We also observed different pCR rates according to Hormone Receptors (HR) status: patients whose tumors were positive for at least one HR had a pCR rate of 15.6% whereas patients who were HR negative had a pCR rate of 33.3% (p = 0.003). Contrary to PR negativity, ER negativity is a significant marker of pCR (p = 0.005). pCR rate is higher for ER− tumors (35.7%) than for ER+ tumors (13.5%; p = 0.005). Tumors with triple negative phenotype achieved significantly higher pathologic response rate than those with non-triple negative phenotype (40.6 vs 16.0%, p = 0.006).

Table 2 Association between clinicopathological features and pCR in FEC-Taxotere treated group (N = 118)
Full size table

Association between pCR and the 37 selected SNPs

Among the 37 SNPs evaluated, genotyping data were available for 114 of the 118 patients enrolled in the study. The distribution of the genotypes and allelotypes indicated that all frequencies were in Hardy–Weinberg equilibrium and were consistent with previously reported Caucasian populations (data not shown).

Univariate analysis revealed that 4 SNPs are significantly associated with pCR (Table 3).

Table 3 pCR is related to 4 SNPs (N = 114)
Full size table

Patients carrying at least one variant allele for BRCA1 (T), ERCC1 (C) and SLCO1B3 (G) have about a threefold higher likelihood to achieve a pCR to NCT than other patients. Patients homozygous for CYP1B1 Leu432Val showed increased pCR rate compared to heterozygous patients (respectively pCR = 29% and pCR = 11.1%, p = 0.020).

Subsequently, assessment of associations between combinations of these SNPs and pCR showed that a model containing only ERCC1 and CYP1B1 polymorphisms performed as well as more complex models. Stepwise logistic regression combining the four significant SNPs identified the model containing CYP1B1 and ERCC1 as the best: these SNPs were validated respectively in 81.0 and 80.8% of the models whereas BRCA1 and SLCO1B3 were selected in 69.2 and 72.3% of the models. The ERCC1 CT genotype associated with CYP1B1 CC genotype had the highest OR of 8.5 (p = 0.013; Table 4). pCR rate reached 50% for patients harboring these genotypes whereas the pCR rate is 21.2% in the whole population.

Table 4 pCR is related to combination of CYP1B1 and ERCC1 polymorphisms
Full size table

Association between pCR and SNPs considering ER status

In breast cancer, tumor subtypes are associated with different responses to neoadjuvant therapies. For example, highly proliferative ER− breast tumors are more sensitive to NCT than ER+ breast tumors: pCR rates are respectively 28–32 and 2–10% (von Minckwitz 2013; Colleoni et al. 2004; Kaufmann et al. 2012).

As pCR can be achieved only in a minority of patients with ER+ breast tumors, it is of particular interest to identify patients who can avoid neoadjuvant chemotherapy or who are candidates for a very high probability of pCR after chemotherapy. Therefore, among the four SNPs previously described as significantly associated to pCR, we searched for potential predictive marker for ER+ breast tumors. pCR is significantly increased for patients carrying variant alleles of BRCA1, SLCO1B3 and CYP1B1 (Table 5, genotyping data obtained for 71 patients).

Table 5 BRCA1, SLCO1B3 and CYP1B1 are significantly associated with pCR in ER + tumors (N = 71)
Full size table

Considering patients with ER− tumors, only ERCC1 (Asn118Asn) is significantly associated with response to NCT (Table 6, genotyping data obtained for 41 patients). Patients carrying at least one variant allele for ERCC1 (C) have a fourfold higher likelihood to achieve a pCR after NCT than patients with other genotypes.

Table 6 ERCC1 is significantly associated with pCR in ER- tumors (N = 41)
Full size table

Moreover, combined ERCC1-CT and CYP1B1-CC genotypes remain significant when adjusted on ER status (p = 0.005; Table 7). In consequence, pCR rate reached 41.2% for patients with ER+ tumors harboring these genotypes whereas the pCR rate is only 13.5% for unselected patients with ER+ tumors. pCR rate is twice higher for patients with ER- tumors harboring combined ERCC1-CT and CYP1B1-CC genotypes than for unselected patients (80 vs 37.5%).

Table 7 pCR is related to combination of CYP1B1 and ERCC1 polymorphisms when adjusted on ER status
Full size table

Discussion

In order to identify breast cancer patients who will benefit from anthracyclines and/or taxanes based NCT, we analyzed the relationship between 37 SNPs and the response to these chemotherapeutic agents.

First overall, our data confirmed that response to NCT is related to clinicopathological features such as histoprognostic tumor grade and hormone receptor status. According to previous published data (von Minckwitz 2013), we observed that pCR rate is increased in SBRIII and HR negative tumors and that ER- tumors respond better to NCT than ER+ tumors. Our data also indicated that patients with triple negative tumors achieve significantly higher pathologic complete response rate than those with non-triple negative tumors. Unexpectedly, pCR was not significantly associated to KI67 expression in our study. Nevertheless, this observation may be explained in part by the fact that data were missing for more than half of patients.

Secondly, considering the predictive value of the 37 studied SNPs, we demonstrated using univariate analyses that BRCA1 (Pro871Leu) and ERCC1 (Asn118Asn), two polymorphisms located in DNA repair genes, and also CYP1B1 (Leu432Val) and SLCO1B3 (IVS12-5676), two polymorphisms in genes involved in drug metabolism, are linked to response to taxanes and anthracyclines. Multivariate analysis revealed that ERCC1 (Asn118Asn) combined to CYP1B1 (Leu432Val) might be of particular interest as predictive markers for NCT in breast cancer. Indeed, we observed that patients exhibiting ERCC1-CT genotype combined to CYP1B1-CC genotype presented a doubling of pCR rate compared with unselected patients (50 vs 21.2%).

Our results are in agreement with previous published data. The BRCA1 variant allele (T) is associated with better overall survival and longer progression-free survival compared to the reference allele in patients with advanced gastric cancer treated with taxanes and cisplatin (Shim et al. 2010).

Although considered as neutral, the functional significance of Pro871Leu (C > T) polymorphism of BRCA1 is still unknown. Pro871Leu might affect expression, activity or interaction of BRCA1 with its partners. As a consequence, Pro871Leu might impair DNA repairing function of BRCA1 and might contribute to potentiate efficacy of DNA damaging agents.

ERCC1 CC genotype was reported to confer longer survival and time to progression in patients with NSCLC treated with cisplatin plus docetaxel (Isla et al. 2004). Any other study described a link between ERCC1 polymorphism and response to taxanes/anthracyclines, probably because ERCC1 polymorphism has extensively been studied in response to platinium compounds, ERCC1 C genotype being related to both a better (Isla et al. 2004; Chang et al. 2009; Kalikaki et al. 2009; Warnecke-Eberz et al. 2009; Metzger et al. 2012) and a worse response (Viguier et al. 2005; Kamikozuru et al. 2008; Ren et al. 2012). The consequence of the silent polymorphism Asn118Asn (T > C) on ERCC1 transcription and expression is not elucidated and remains still conflicting. Gao & co had demonstrated no difference in ERCC1 transcription nor expression between cells stably expressing ERCC1-T allele or ERCC1-C allele (Gao et al. 2011). However, data obtained on lymphocytes of prostate cancer patients revealed that carriers of CC genotype showed lower ERCC1 mRNA levels (Woelfelschneider et al. 2008).

Thus, by decreasing ERCC1 mRNA levels, C genotype might decrease ERCC1 mediated DNA repair and might contribute to potentiate efficacy of DNA damaging drugs. This hypothesis is in agreement with clinical data which present that low levels of ERCC1 are favorable for sensitivity to platinium compounds (Li et al. 2000; Shirota et al. 2001; Lord et al. 2002; Wang et al. 2008).

CYP1B1 Val genotype is associated with lower response rate, shorter progression-free-survival and decrease overall survival in breast and prostate cancer patients treated with taxanes based chemotherapy (Marsh et al. 2007; Sissung et al. 2008; Pastina et al. 2010). However, in our study, only Leu/Val genotype provided the best data fit for the SNP Leu432Val. The hypothesis of an overdominant model, previously described in bladder cancer (Salinas-Sanchez et al. 2012), might be explained by a higher enzyme activity in heterozygous than in homozygous. CYP1B1 binds and sequestrates docetaxel reducing its available concentration and influencing its activity (McFadyen et al. 2001). CYP1B1 also produces oestrogen metabolites that limit docetaxel efficacy by inhibition of tubulin polymerization (Sissung et al. 2008). Thus, by increasing CYP1B1 mRNA expression and catalytic activity (Shimada et al. 1999; Hanna et al. 2000; Li et al. 2000; Landi et al. 2005), Leu432Val substitution appears to modulate response to taxanes.

SLCO1B3 is an influx transporter involved in the uptake of a broad range of drug substrates including docetaxel (Konig et al. 2000; Smith et al. 2005). The functional impact of rs11045585 polymorphism is yet unknown. However, asian nasopharyngeal cancer patients homozygous for the variant allele had higher area under curve and less plasma clearance of docetaxel compared to patients carrying at least one reference allele (Chew et al. 2011). It is to be noted that rs11045585 polymorphism is proposed to exert its functional effect by being in haplotypic combination with 3 other polymorphism in the SLCO1B3 gene (Chew et al. 2012).

Finally, as pCR is rarely achieved by NCT in ER+ breast tumors, we searched if BRCA1 (Pro871Leu), ERCC1 (Asn118Asn), CYP1B1 (Leu432Val) and SLCO1B3 (IVS12-5676) could stand for predictive markers of NCT for patients with this subtype of tumor. Our results indicated that BRCA1 (Pro871Leu), CYP1B1 (Leu432Val) and SLCO1B3 are significantly associated with response to NCT in ER+ tumors. ERCC1/CYP1B1 combination remains significant for ER+ tumors and might be useful to help indication of NCT in this particular type of tumors as patients exhibiting ERCC1-CT genotype combined to CYP1B1-CC genotype presented a threefold higher pCR rate compared with unselected patients: 41.2 vs 13.5%.

Our data also indicated that ERCC1 Asn118Asn might be a potential predictive marker for NCT in ER- tumors, as pCR is increased to 50% for patients carrying at least one variant allele (C).

In conclusion, BRCA1 (Pro871Leu), ERCC1 (Asn118Asn), CYP1B1 (Leu432Val), and SLCO1B3 (rs11045585) are associated with response to NCT in our cohort of breast cancer patients. To our knowledge, this study is the first to report ERCC1, BRCA1 and SLCO1B3 as markers of taxanes and/or anthracyclines response in breast cancer NCT.

Moreover ERCC1/CYPB1 combination appeared to be a potential predictive marker to guide NCT indication in breast cancer and particularly for ER+ breast tumors.