Autocrine CCL5 Effect Mediates Trastuzumab Resistance by ERK Pathway Activation in HER2-Positive Breast Cancer

Breast cancer accounts for 20%–25% of all cancer cases worldwide and is the most prevalent in women (1). Breast cancer comprises many biologically different diseases with distinct pathologic features and clinical implications, thus making accurate grouping of clinically relevant subtypes of importance (2). Among these subtypes, HER2-positive breast cancer, which accounts for 20% of all breast cancers (3), is characterized by gene overexpression, high rates of cell proliferation and metastasis, poor prognosis, low overall survival (OS), and variable chemotherapy response with poor outcome (4). Together with this prognostic value, the HER2 receptor is currently considered part of the standard assessment protocol as a predictor of response to treatment (5).

During the past decade, systemic therapeutic management of breast cancer has undergone a significant transformation, leading to the emergence of targeted therapy. For patients with HER2-positive breast cancer, targeting HER2 has become an attractive therapeutic approach. Trastuzumab (Herceptin), a humanized IgG1 mAb that selectively targets the HER2 receptor, became the first FDA-approved targeted therapy for metastatic breast cancer in 1998 (6). Since then, therapies such as trastuzumab combined with chemotherapy have been considered the standard of care for patients with HER2-positive breast cancer (7). However, about 25% of HER2-positive breast cancers do not respond initially to trastuzumab (8), and 70% of the trastuzumab-responsive metastatic cancers progress to therapy within the first year due to acquisition of trastuzumab resistance (9). Several potential resistance mechanisms to trastuzumab have been reported during the last decade and their details have been described in numerous reviews (10–12).

Nowadays, it is widely accepted that many pathways may be involved in the development of resistance to antibody-based treatments. Specifically, there is a widespread belief that a tumor must be examined in the context of its microenvironment, and therefore, should be considered as an entity with a heterogeneous cellular origin in continuous interaction with the stroma, nontumor cells, and the immune system (13), highly modulated by the inflammatory cells found in the tumor (14). This interaction of the tumor with the microenvironment is mainly regulated through cytokines, which signal the participation of distinct pathways in processes of cell proliferation and differentiation (15).

Many chemokines and their receptors are expressed by tumor cells. The chemokine CCL5/RANTES, is well-recognized for its activities in the immune context, where it induces leukocyte-directed motility. CCL5 has affinity for the G protein–coupled receptors CCR1, CCR3, and, especially, CCR5; and a lesser binding capacity with other receptors. Recently, CCL5/CCR5 have been implicated in proliferation and metastasis in breast cancer (16, 17) and have been recognized as potential therapeutic targets. Moreover, a recent study showed that CCL5 signaling promotes breast cancer recurrence following HER2 inhibition through the recruitment of macrophages (18). In the context of human immunodeficiency virus/acquired immunodeficiency syndrome studies, potent antagonist inhibitors have been developed, which prevent binding of ligands to the receptors; the only CCR5 antagonist currently approved by the FDA and the European Medicines Agency for the treatment of infected patients is maraviroc (Pfizer; refs. 19, 20). The use of this drug in breast cancer could prevent the activation of the CCR5 receptor mediated by the autocrine action of CCL5 (21). Preliminary studies on cell lines and murine models have shown that maraviroc prevents binding of CCL5 to CCR5 by decreasing proliferation and development of metastases (22).

In this study, we aimed to identify, quantify, and functionally evaluate potential biomarkers that might be involved in trastuzumab resistance in breast cancer. We examined differential gene expression in HER2-positive trastuzumab-sensitive and -acquired resistant human breast cancer cell models (23). Differential gene expression analysis indicated an overexpression of 16 genes, highlighting CCL5 as the most overexpressed chemokine with involvement in breast cancer. Because of the emerging role of these proteins as mediators of normal proliferation, migration, and metastasis (16, 24, 25), we explored the CCL5 implication in acquired resistance to trastuzumab by functional studies, which were validated in clinical samples from HER2-positive breast cancer. Notably, our results reveal the previously undescribed involvement of CCL5 in the development of trastuzumab resistance.

BT-474, HCC1569, and HCC1954 cell lines were purchased from ATCC and authenticated (LGC Standards; tracking no: 710259498). JIMT1 cell line was purchased from DSMZ. Trastuzumab-resistant (BT-474.rT) cell line was generated by continuous exposure to trastuzumab, cultured, authenticated, and tested for Mycoplasma as described previously (23). Cell growth and proliferation assays were performed as reported previously (23). Trastuzumab (Genentech) was supplied by the pharmacy of our hospital. Maraviroc and selumetinib were obtained from Selleckchem. Recombinant human CCL5r was from R&D Systems.

Migration assays were performed using 24-well plates with transwell permeable supports of 6.5 mm insert and a polycarbonate membrane with an 8-μm pore size (Costar 3422, Corning). Cells were seeded in the top chamber at 2 × 10⁴ cells/mL in 0.1 mL of serum-free RPMI1640 media. A volume of 0.8 mL of media supplemented with 10% FBS were placed in the bottom well as a chemoattractant. After incubation for 24 hours at 37°C in an atmosphere containing 5% CO₂, migrated cells on the bottom surface were stained using crystal violet and counted under a light microscope.

Experiments were performed in 6-well plates coated with 3 mL of 0.6% soft agarose (Sigma-Aldrich) in medium. A total of 5 × 10³ cells were suspended in 0.3% agarose in medium and plated in triplicates over the precoated wells. Fresh medium was supplied twice a week. After 21 days, colonies were stained with MTT (M-565, Sigma-Aldrich) for 4 hours at 37°C. Then, colonies were fixed by adding DMSO overnight at 37°C. Colony numbers were determined from triplicates and three independent experiments were carried out for each condition and cell line.

Total RNA extracts were isolated using RNeasy Mini Kit (Qiagen), their quality was assessed by NanoDrop determination, and RNA was transcribed to cDNA using Universal Transcriptor cDNA Kit (Roche). qPCR amplification was performed in a LightCycler-480 system (Roche) using assays specific for ATP5E, CCL5, CCR5, CXCL10, CXCL11, IFNλ1, and IFNλ2 (Supplementary Table S1). Relative gene expression was calculated according to the comparative method (26), using ATP5E expression for normalization.

Total protein lysates were prepared with RIPA buffer containing protease and phosphatase inhibitors. Nuclear and cytosolic protein fractions were isolated using the K266-25 Nuclear/Cytosol Fractionation Kit (BioVision). Protein extracts were clarified, denatured, and subjected to SDS-PAGE and Western blotting. The primary antibodies were: AKT, pAKT-Thr308, pAKT-Ser473, ERK1/2, pERK1/2-Thr202/Tyr204, and HER2 (Cell Signaling Technology); CCL5 (R&D Systems); and GAPDH (Sigma-Aldrich). Secondary antibodies were conjugated to alkaline phosphatase (Sigma-Aldrich).

The CCL5/RANTES Immunoassay (R&D Systems) was used in cell protein extracts, cell culture supernatants, or patient sera, in duplicate. Absorbance was measured at 450 nm.

Total RNA from BT-474 and BT-474.rT cells either untreated or treated with 15 μg/mL trastuzumab for 48 hours from independent biological duplicates was used for gene expression profiling with the Genechip Human Gene 2.0 ST (Affymetrix). Data were processed following the methodology described previously (27). Cut-off value was set so that genes with >2-fold change (sensitive/resistant) in expression levels were considered significantly altered. Further identification of prospective biomarkers of resistance to trastuzumab was performed using a filtering process to determine candidates that were differentially regulated between BT-474 and BT-474.rT, both at basal conditions and trastuzumab exposure conditions. Data are available through Gene Expression Omnibus with dataset identifier GSE89216.

Gene set enrichment analysis (GSEA; ref. 28) was applied using annotations from MsigDB, Reactome, Kyoto Encyclopedia of Genes and Genomes, and NCI databases. Genes were ranked on the basis of limma moderated t-statistic. After Kolmogorov–Smirnoff testing, those gene sets showing FDR < 0.05 were considered enriched between classes under comparison.

BT-474.rT cells were transfected with siRNAs targeting CCL5 (Smart-pool of four siRNAs: on-target-plus CCL5 siRNA L-007844-00-0005, 5 nmol) and CCR5 (Smart-pool of four siRNAs: on-target-plus CCR5 siRNA L-004855-00-0005, 5 nmol), and scrambled siRNA as a control (Dharmacon), dissolved in a mixture of Opti-MEM and Lipofectamine 2000 (Invitrogen). After transfection, BT-474.rT cells were treated with 15 μg/mL trastuzumab for 7 days, and cell growth was assessed. Transfection was repeated after 72 hours to maintain silencing. For gene and protein expression analysis, cells were subjected to siRNA silencing for 48 hours, lysed, and subjected to qPCR and Western blot assays as described above.

We developed an in vivo xenograft subcutaneous breast cancer model at the facilities of the Barcelona Biomedical Research Park (PRBB, Barcelona, Spain). All experiments were performed in accordance with the 2010/63/EU Directive on the protection of animals and approved by the Ethical Committee for Animal Research of the PRBB. Six-week-old female mice, with SCID/beige (Charles River Laboratory), were selected for inoculation. A 17β-estradiol pellet, 0.72 mg, 60-day release (Innovative Research of America) was implanted subcutaneously into each mouse 48 hours before cell injection. Twenty mice were subcutaneously inoculated in their right flank with 2.5 × 10⁶ BT-474.rT cells mixed with 1:1 Matrigel (BD Biosciences) in PBS as described previously (29). Tumor diameters were serially measured with digital calipers and tumor volumes were calculated by the equation: volume = (width² × length)/2. When the average volume of tumors reached 100 mm³, mice were randomly allocated into four groups of 5 mice each. For therapeutic studies, the concentration and time of treatments were based on previous reports, and administered as follows: group 1, mice received control treatment with human IgG1ĸ (10 mg/kg); group 2, trastuzumab (10 mg/kg); group 3, maraviroc (10 mg/kg); and group 4, mice received the combination of trastuzumab and maraviroc (10 mg/kg and 10 mg/kg, respectively). All treatments were freshly prepared in PBS, allowing for an injection volume of 100 μL/20 g mouse intraperitoneally every other day for 3 weeks. After 3 weeks, tumor xenografts obtained from BT-474.rT cells were excised and measured.

Formalin-fixed, paraffin-embedded (FFPE) sections (2–3 μm) were obtained from cell pellets of patient samples and human tumor xenografts in mice and IHC was performed as described previously (30). Primary antibodies against pERK, cleaved-caspase3, pHistone3 (Cell Signaling Technology), and CCL5 (R&D Systems) were used, and HER2 status was assessed by HercepTest (Agilent Technologies). A semiquantitative histoscore (H score) was calculated by estimation of the percentage of tumor cells positively stained with low, medium, or high staining intensity, and the results ranged from 0 to 300 (31). All IHC staining was performed on a Dako Autostainer Platform (Agilent Technologies).

We downloaded read count data for 1,097 primary breast tumors from The Cancer Genome Atlas (TCGA; refs. 32, 33; http://cancergenome.nih.gov, January 2015) using the R package in TCGA-Assembler. We obtained clinical data for the breast cancer samples from the original clinical dataset (clinical_patient_piblic_brca.txt) as described in the previous studies.

A total of 146 specimens from primary breast tumors were obtained from the Fundación Jiménez Díaz Biobank. Tumor specimens from FFPE blocks were retrospectively selected from consecutive patients with breast cancer diagnosed between 2000 and 2014, following these criteria: infiltrating carcinomas, operable, and neoadjuvant (N = 64), or adjuvant therapy (N = 82) containing trastuzumab, and enough available tissue and clinical follow-up. In addition, 17 cases for which serum samples were available prior to treatment were selected from the Biobank of the Hospital del Mar. For all cases, clinical data were collected from medical clinical records by oncologists, following written informed consent from the patients. The studies were conducted in accordance with ethical guidelines from the Declaration of Helsinki. The ethical committee and institutional review boards from our hospitals approved the project. Clinical tumor response to primary chemotherapy was evaluated for pathologic response according to the International Union against Cancer Criteria staging system (34).

ROC analysis was used to determine the optimal cut-off point based on relapse endpoint for CCL5 expression as described previously (35). Survival was analyzed by the Kaplan–Meier method using the log-rank test. OS was defined as the time from diagnosis to the date of death from any cause or last follow-up. Disease-free survival (DFS) was defined as the time from diagnosis until the first event, in which relapse at any location, death, or end of follow-up were considered events. Multivariate analyses were carried out using the Cox proportional hazards model. Analysis of experimental conditions was done by paired t test. Statistical significance was analyzed by a two-tailed Student t test (*, P < 0.05; **, P < 0.01; ***, P < 0.001). This work was carried out in accordance with REMARK guidelines (36).

After generation of trastuzumab-resistant lines (23) we decided to search for differences at the gene level between these cells and their corresponding parental sensitive lines, in the absence or presence of 15 μg/mL trastuzumab. Changes in gene expression were considered as significant for −2 ≥ logFC ≥ 2 and a P < 0.05. Principal component analysis revealed separate clusters corresponding to the sensitive BT-474 and the resistant BT-474.rT cells and proved that the differences in their gene expression pattern were independent of trastuzumab exposure in culture. Expression level data were obtained for 10,508 genes, and we identified 25 genes with differential expression between BT-474 and BT-474.rT. Of these, 15 genes were overexpressed in BT-474.rT compared with BT-474 (Supplementary Fig. S1).

GSEA demonstrated that the ERBB2/HER2 gene set was enriched in BT-474.rT, and that the genes with higher overexpression in this line relative to the parental contributed to the enrichment of this gene set. The most significantly contributing genes were IFN-inducing proteins IFI44, IFIT1, and IFI44L, all members of the superfamily of cytokines (Supplementary Fig. S2). Given their involvement in proliferation and metastasis processes in breast cancer, five genes from the cytokine family (CXCL10, CCL5, CXCL11, INFL1, and INFL2) were selected for assessment of their transcript expression profiles by qRT-PCR. All five genes exhibited elevated expression levels in the BT-474.rT line compared with the sensitive BT-474 (Fig. 1A). In addition, their expression levels were assessed in lines with primary resistance to trastuzumab (JIMT11, HCC1569, and HCC1954): CCL5, CXCL10, and CXCL11 were found to be overexpressed in all three lines, although to a lesser degree than BT-474.rT, whereas IFNL1 showed levels of intermediate expression between BT-474.rT and BT-474 (Supplementary Fig. S3).

Immunodetection by Western blot analysis showed a marked increase in CCL5 protein in BT-474.rT relative to its parental line, which had very similar protein levels to the HCC1569 line (Fig. 1B). This protein overexpression was also verified by IHC from cell pellets (Fig. 1C): no expression of CCL5 was detected in BT-474, whereas a heterogeneous CCL5 overexpression was observed in BT-474.rT cells. In the primary resistance line, HCC1569, a homogeneous CCL5 intermediate expression was identified in almost all cells. As cytokines are known for their abilities to induce cellular migration, and CCL5-overexpressing cells are specifically reported to acquire invasion and migration abilities (37), cell invasion assays were also performed to detect the invasive capacity of these breast cancer cells. Interestingly, we observed significantly increased migration in BT-474.rT cells in comparison with BT-474 control cells (Fig. 1D), thereby evidencing that resistance acquisition plays a relevant role in regulating the migration of the cells. In addition, clonogenic assays were performed to analyze whether mechanistic changes because of acquired resistance could alter the malignancy of BT-474.rT cells. We detected that treatment with trastuzumab for 21 days did not significantly impair the colony-forming ability of cells with acquired resistance, in comparison with sensitive BT-474 cells, proving that it was a stable resistance (Supplementary Fig. S4).

Finally, we assessed whether the CCL5 protein produced by these cell lines was secreted into the medium, as happens in physiologic conditions with chemokines. The increase in CCL5 synthesis in BT-474.rT cells was determined by an ELISA assay in the secretion to the medium (Fig. 1E), with a concentration of 7,504 pg/mL CCL5 in BT-474.rT compared with 144 pg/mL in BT-474.

The silencing of CCL5 by siRNA in BT-474.rT reversed trastuzumab resistance significantly (P = 0.001) in a 7-day cell proliferation experiment with respect to the control (54% vs. 82%), with a ΔGR value of 1.3 being sensitive according to the algorithm defined by O'Brien and colleagues (ref. 38; Fig. 2A). However, the HCC1569 line transfected with siCCL5 presents growth in the presence of trastuzumab similar to the control condition (siC), indicating that this line remains resistant to trastuzumab independently of the reduction of CCL5 levels (Fig. 2A). The silencing of its receptor CCR5 and the dual silencing of CCL5/CCR5 reversed trastuzumab resistance even more intensely, adding more evidence to the implication of CCL5 in the acquisition of resistance (Fig. 2B). The efficiency of CCL5 and CCR5 silencing was confirmed by qPCR and ELISA in both lines (Supplementary Fig. S5).

Next, we evaluated whether sensitization to trastuzumab in the BT-474.rT line caused by the silencing of CCL5 was compensated by the addition of exogenous CCL5 (CCL5r; Supplementary Fig. S6). The addition of CCL5r under CCL5 silencing conditions significantly increased cell growth (P = 0.003) in the presence of trastuzumab up to resistance levels (ΔGR = 1.1). These data support the argument that an increase in levels of CCL5 is implicated in acquired resistance to trastuzumab.

Maraviroc is a negative allosteric modulator antagonist of the CCR5 receptor and blocks its activation by preventing the binding of CCL5 to the receptor (22). The cell lines BT-474 and BT-474.rT were exposed to different concentrations of maraviroc (5–200 μmol/L) to assess its effect in cell growth. A count after 7 days revealed that exposure to maraviroc at high concentrations did not result in a significant inhibition of cell proliferation in any of the lines tested. When BT-474.rT cells were treated with 15 μg/mL trastuzumab, either with 10 μmol/L maraviroc or with the combination of both drugs, it was observed that treatments individually did not provoke inhibition of proliferation; however, the combination of trastuzumab plus maraviroc caused a significant decrease in cell proliferation and reversed acquired resistance (47%, P < 0.001; Fig. 2C).

In treatment conditions with trastuzumab, a marked decrease in pERK was observed in BT-474 (70% decrease on average, by densitometric analysis), while ERK phosphorylation levels are less affected by trastuzumab treatment in BT-474.rT (20% decrease). This fact indicates that, on the BT-474.rT cell line, treatment with trastuzumab is not able to block the activation of ERK (Fig. 2D). At the molecular level, the combination of trastuzumab and maraviroc caused a significant decrease in ERK activation levels in the resistant cell line that was not observed in the single treatment with maraviroc or trastuzumab (Fig. 2E). Treatment with maraviroc in combination with trastuzumab causes resensitization to trastuzumab in the resistant line, because of decreased levels of pERK, suggesting that increased CCL5 levels favor activation of ERK (Fig. 2E). In addition, treatment with maraviroc alone or in combination with trastuzumab did not result in a decrease in HER2 levels. Similarly, a decrease in pAKT levels (both in Thr308 and Ser473 levels) was attributed to trastuzumab treatment (alone or in combination) but not to maraviroc alone (Supplementary Fig. S7). The confirmation of our hypothesis that high CCL5 levels cause the activation of ERK in the resistant cells was revealed with the dual treatment of the BT-474.rT cell line with trastuzumab plus selumetinib (Fig. 2F). When the activation of ERK was repressed with the MEK-specific inhibitor selumetinib (Fig. 2G), cell proliferation of resistant cells treated with the drug combination significantly decreased as compared with trastuzumab treatment alone (Fig. 2F), therefore improving its therapeutic response.

The assessment of maraviroc-mediated restoration of sensitivity to trastuzumab was performed with a BT-474.rT cell line–derived xenograft model to investigate the role of CCL5 in tumor growth. After tumors reached a minimum volume of 100 mm³, mice were allocated at random to one of four treatment groups: control group (10 mg/kg IgG1ĸ) and treatment groups (trastuzumab 10 mg/kg, maraviroc 10 mg/kg, and trastuzumab 10 mg/kg plus maraviroc 10 mg/kg in combination). Similar to the in vitro results, mice treated with the combination of trastuzumab and maraviroc displayed significantly less tumor growth (10%, compared with day 0; P = 0.004) than those of the control group treated with IgG ĸ and those of the groups receiving trastuzumab or maraviroc treatments alone (170%, 120% and 60%, respectively, compared with day 0; Fig. 3).

The validation of CCL5 detection was performed according to the Rimm algorithm for IHC validation (39), and the optimal dilution of the anti-CCL5 antibody was determined to be 1:40. The CCL5 overexpression threshold was determined by the ROC curve based on the endpoint of relapse, calculated as the AUC (Supplementary Fig. S8). Samples with values of H score > 150 (sensitivity = 75%, specificity = 100%) were considered as having high CCL5 overexpression. Tumor tissue sections showed a CCL5 cytoplasmic expression in patches or homogeneous staining in neoplastic cells, with variable intensity ranging between weak and strong. In normal mammary epithelial tissue adjacent to the tumor, CCL5 expression could not be observed. Consistently, and as expected, CCL5 expression was detected in lymphocytes and plasma cells present in the tumor stroma and adjacent mammary parenchyma. This expression of CCL5 in mononuclear inflammatory cells was used as positive internal control in the cases evaluated in the different study cohorts (Fig. 4A). For pERK1/2, nuclear staining was required for considering a tumor cell as positive (Fig. 4A). A tumor was scored as positive when any proportion of tumor cells was stained. Endothelial cells and lymphocytes were considered as internal positive and negative controls for each slide.

Clinical assessment of CCL5 overexpression in resistance to trastuzumab was studied in samples from a clinical cohort of 146 cases of early HER2-positive breast carcinoma treated with trastuzumab in different regimens. Sixty-four FFPE samples of neoadjuvant treatment of trastuzumab plus chemotherapy were included in this cohort (Supplementary Table S2). One-third of these cases showed high levels of CCL5 expression in tumor cells. Expression of CCL5 was not significantly correlated with hormonal status, histologic type and grade, hormone receptors, proliferation, or stage. However, CCL5 expression did correlate significantly with complete pathologic response to treatment (Fig. 4B; Supplementary Table S2). Seventy-one percent of the tumors with no evidence of pathologic response (Miller and Payne grade G1–G3) had a high expression of CCL5, while only 14% of tumors with either almost complete tumor response (G4) or complete response (G5) had high expression of CCL5 (P < 0.001). In fact, none of the tumors with complete tumor response (G5) had high expression of CCL5. Considering the degree of lymph node response, CCL5 overexpression was detected in 52% of the cases without evidence of response to treatment (B and C; Fig. 4C). On the other hand, high levels of CCL5 were detected in five cases with negative lymph nodes (A) and five cases with lymph nodes without residual neoplastic infiltration (D; P = 0.013).

Taking together the pathologic response in the breast and in the axilla, 24 cases with complete pathologic response were identified, all with low CCL5 expression levels. Conversely, most of the 40 cases that did not present complete pathologic response showed high levels of CCL5 (P < 0.001; Supplementary Table S2). Finally, CCL5 overexpression correlated with more frequent disease relapse: 73% of cases with increased expression suffered relapse, compared with 27% of those with low expression levels (P = 0.002; Supplementary Table S2). These findings indicated that an increase in CCL5 expression levels in the tumor component predicts a worse response to trastuzumab treatment.

The adjuvant cohort consisted of 82 samples taken prior to therapy from patients with early breast cancer treated with trastuzumab plus chemotherapy (Supplementary Table S3). In this cohort, CCL5 overexpression was detected in 22% of the cases. CCL5 expression was not significantly associated with stage, histology, tumor grade, hormonal receptors, or proliferation. Relapse of the disease (35% of the patients) was significantly correlated with the expression of CCL5 (P < 0.001), with high levels of CCL5 being identified in 72% of cases with relapse (Supplementary Table S3). As we had evidence, from the molecular assays in cell lines, of the potential role of ERK activation as a mediator of the resistance elicited by CCL5, we also determined the expression levels of pERK in this cohort. We found a strong correlation of the overexpression of both markers (P = 0.001; Supplementary Fig. S9).

Finally, we correlated CCL5 expression levels with evolution of patients in the entire cohort of early breast cancer. Increased CCL5 expression in the tumor component was significantly associated with lower DFS (median 29 vs. 42 months; P < 0.01; Fig. 5A), and a lower OS (P < 0.001; Fig. 5B). The DFS analysis separately in neoadjuvant and adjuvant series was also significant (median 54 vs. 104 months, P = 0.001 in neoadjuvant; 44 vs. 93 months, P < 0.001 in adjuvant setting; Supplementary Fig. S10). A multivariate Cox analysis including all the significant clinical–pathologic factors from the univariate study, as well as estrogen receptors and chemotherapy regimen, revealed that CCL5 overexpression behaved as an independent factor of poor prognosis in patients with early HER2-positive breast cancer (HR, 13.6; 95% confidence interval, 3.4–54.8; P < 0.001; Supplementary Table S4).

A small cohort of 14 sera samples from patients with HER2-positive breast cancer collected prior to initiation of trastuzumab treatment in a neoadjuvant regimen was also analyzed. CCL5 concentration in sera was determined by ELISA assay, and the optimum cut-off point was calculated using a ROC curve, which was established at a concentration of CCL5 ≥ 90 ng/mL in serum (sensitivity = 71%, specificity = 86%). Forty-three percent of serum samples displayed high CCL5 concentrations, correlating significantly with the degree of pathologic response: 75% of the patients with low CCL5 concentration in serum presented complete response, whereas only 17% of patients with a high concentration of CCL5 showed this complete response (Supplementary Table S5).

We decided to evaluate the modulation of CCL5 expression in response to trastuzumab exposure in a cohort of 44 patients with paired samples (pre- and posttreatment samples). Of these, 25 pairs were obtained from patients with early HER2-positive breast cancer treated in a neoadjuvant regimen (the pretreatment sample corresponded to the diagnostic biopsy and the posttreatment sample to the surgical specimen). Continued exposure to trastuzumab in those patients caused a significant increase in CCL5 expression: 40% of posttreatment samples showed increased CCL5 expression in their tumor component as compared with their pretreatment samples (P = 0.022), whereas 28% of them exhibited lower CCL5 levels, and the remainder did not vary (Fig. 5C).

Moreover, 19 additional cases were selected from patients in whom progression of the disease had been detected during the conventional treatment of chemotherapy plus trastuzumab, and for whom both a pretreatment diagnostic sample and a metastatic after treatment sample were available. Analysis of CCL5 expression revealed that 58% of the posttreatment samples presented a statistically significant increase in CCL5 expression (P = 0.012), while 16% of samples showed a decrease in CCL5 expression after treatment (and 26% of paired samples did not change their CCL5 expression levels; Fig. 5D).

We analyzed mRNA expression levels of CCL5 in a series of 182 patients with HER2-positive breast cancer from the available data on TCGA and correlated with OS. A tendency toward association of CCL5 overexpression with worse OS was demonstrated, although this was nonsignificant (P = 0.069; Supplementary Fig. S11). On the other hand, expression of CCR5 (the most common receptor for CCL5) showed a nonsignificant tendency to be associated with overall poorer prognosis (P = 0.086). The correlation study with the CCL5 remaining receptors (CCR1, CCR2, CCR3, and CCR4) showed that overexpression of CCR3 correlated significantly with a lower OS in patients with HER2-positive breast cancer (P = 0.038; Supplementary Fig. S11).

HER2-positive tumors account for 20%–25% of all cases of breast cancer. With the introduction of anti-HER2 therapies in clinical practice, however, the prognosis of these tumors is now favorable, and DFS and OS are increasing. However, a high percentage of patients with HER2-positive tumors do not respond to therapy with trastuzumab, mostly due to the presence of other genetic alterations in the tumor that are either unknown or undetermined in clinical practice. These additional alterations may cause the tumor to be refractory to treatment with trastuzumab, promoting tumor proliferation and metastasis.

Our results suggest that continued exposure to trastuzumab in cellular models leads to drug resistance. Differential gene expression analysis identified 16 significantly overexpressed genes in the BT-474.rT line compared with the BT-474 line. Most strikingly, 13 of 16 genes belong to the cytokine superfamily (CXCL10, CCL5, CXCL11, IFNL1, and IFNL2) or are involved in the activation of members of this family (IFIT3, IFI44, IFI6, IFIT1, IFI44L, IFIT2, OAS1, and OASL). Cytokines are the main proteins secreted into the extracellular domain, and although their main role is the recruitment and activation of the immune response, their implication becomes increasingly relevant in neoplastic processes of invasion, metastasis, and immune response evasion (40, 41). Indeed, it has been reported that the cytokine profile secreted from the tumor site varies between different breast cancer subtypes. In the HER2-positive breast cancer subtype, production of cytokines such as IL6 and CCL5 are predominant and implicated in cell proliferation (42). Notoriously, CCL5 was one of the top genes differentially overexpressed in the expression analysis. The main function of the CCL5 chemokine is its chemotactic activity in the immune system; however, it has an important autocrine function in the tumoral component. In addition, reports suggest that CCL5 is frequently overexpressed in basal phenotype breast cancer, HER2-positive, as well as in advanced disease (42). In all these settings, high levels of CCL5 in breast lines are associated with increased proliferation and migration (43), stem phenotype (44), metastasis (17), and immune cell infiltration (18). Although the other chemokines detected in the differential expression array made in resistant and sensitive cells are also described in breast cancer, their role is more controversial, and is mainly associated with luminal type breast tumors.

CCL5 overexpression induction in tumor cells is modulated by the transcription factor AP-1, activated by the binding of the also transcription factor c-Jun, which in turn is activated by the cellular signaling path JNK. The activation of AP-1 for CCL5 synthesis is mainly by NF-κB via AKT, and the MAPKs pathway (44). Notably, the overexpression of CCL5 prompted cell proliferation and migration through the activation of mTOR (25). Recently, CCL5 has been associated with resistance to different treatments in breast cancer: in cellular models of luminal subtype, overexpression of CCL5 caused phosphorylation and activation of STAT3, and it was postulated as a possible mechanism of resistance to tamoxifen (45). Another study showed the activation of an IL6 inflammatory loop including CCL5-mediated trastuzumab resistance in HER2-positive breast cancer cellular models by expanding the cancer stem cell population (46): an IL6-mediated increase resulted in activation of AKT, STAT3, and NF-κB, and the subsequent rise of CCL5 levels, ultimately responsible for the resistance. Recently, a study linked the inhibition of HER2 with an increased CCL5 signaling, which promoted macrophage recruitment and ultimately resulted in tumor recurrence (18). One might speculate that CCL5 plays a role in acquired trastuzumab resistance by attracting macrophages in the stroma, which supply residual cancer cells with collagen and ultimately promote tumor growth. Other reports, however, suggest that overexpression of CCL5 is associated with a better response to treatment with trastuzumab (42). The acquisition of resistance to specific anti-HER2 therapies by autocrine production of ligands that activate compensatory pathways has also been described in HER2-positive breast cancer models. In particular, acquired therapeutic resistance was reported when incomplete inhibition of EGFR by lapatinib resulted in selection of an heregulin-driven feedback that activated the HER3–EGFR–PI3K–PDK1 signaling axis (47). Similar, insulin growth factor-I (IGF-I) activation of receptor IGF-IR signaling has been associated with trastuzumab resistance in HER2-positive breast cancer models (48).

Our functional studies indicate that CCL5 is involved in acquired resistance to trastuzumab. Both, the silencing of CCL5 and its pharmacologic inhibition by maraviroc, provoke a reversal of resistance to trastuzumab in BT-474.rT cells. Our results further suggest that such resistance could be produced by the constitutive activation of ERK: unlike that, which occurs in the sensitive model, treatment with trastuzumab does not affect ERK activation in resistant cells, indicating that this pathway is being activated by other mechanisms. In addition, treatment with maraviroc plus trastuzumab results in a decrease in ERK activation, suggesting that the reversal of resistance to trastuzumab caused by CCL5 blockade is mediated by a reduction in ERK activation levels. These results agree with recent studies describing that the interaction of CCL5 with its receptor CCR5 causes direct activation of ERK, thus promoting an increase in cell migration (49) and providing antiapoptotic signals for cell survival (50). In addition, the interaction of CCL5 with other receptors such as CCR1 in different cancer types also causes an increase in ERK phosphorylation that promotes the expression of MMP2 and MMP9 in taxane-resistant models (51, 52). In addition, the decrease in cell proliferation of resistant cells treated with trastuzumab in combination with selumetinib, a specific MEK repressor, confirmed the role of ERK activation in CCL5-mediated acquisition of trastuzumab resistance. Our results also indicate that the resistance was not mediated by pAKT, because the combined treatment of trastuzumab with maraviroc modulates AKT in a way that resembles trastuzumab monotherapy. This agrees with previous reports indicating that CCR5 chemokines did not induce any significant AKT phosphorylation (53).

Our proposal of this role of CCL5 as a mediator of resistance acquisition to trastuzumab was not applicable to primary resistant cell models. As it was proved for the HCC1569 line, blocking of CCL5 (either by RNA silencing or by treatment with maraviroc) did not significantly decrease cell proliferation. In addition, the evidence of hyperactivation of the PI3K/AKT/mTOR pathway in HCC1569, mediated by the loss of PTEN expression (38), suggested that CCL5 was not responsible of the primary resistance in those cells.

On the other hand, the addition of exogenous CCL5 (CCL5r) to the sensitive line does not lead to an increase in resistance to trastuzumab, suggesting that the acquisition of resistance is a complex process that may require additional alterations to the overexpression of CCL5. Previous studies identified that the interaction between CCL5 and CCR5 was required to observe proliferation and migration processes, which in addition required overexpression of both CCL5 and its receptor (22, 25). In summary, our work suggests that CCL5 overexpression causes resistance to trastuzumab treatment through ERK activation, and that the resistance process could be mediated by different CCL5 receptors. Notably, there are other cytokines overexpressed in our resistance cell model that have not been assessed in this study and might also contribute to acquired resistance to trastuzumab.

Finally, the tumorigenic potential of the BT-474.rT cell line and the trastuzumab-resistant phenotype of the derived tumors were both confirmed in a BT-474.rT xenograft model in mice. Although maraviroc administration as a single treatment was ineffective, it showed significant antitumor effects when it was administered in combination with trastuzumab, in terms of size reduction. This supported our previous in vitro observations, and therefore suggested that acquired trastuzumab resistance may be mediated by CCL5 activity in vivo. Collectively, these results indicate that the combined therapy with verteporfin overcomes acquired trastuzumab resistance in vivo by blocking tumor growth and inducing tumor reversion.

Overexpression of CCL5 has also been described in patients with breast cancer, correlated with an advanced disease and presence of a greater number of metastases, and associated with worse prognosis (17, 54). The increase in plasma CCL5 concentration has also been associated with poor prognosis of the disease (43). In addition, both the basal and the HER2-positive molecular subtypes exhibit a higher CCL5 expression in the tumor component (42). Our analyses demonstrate that CCL5 overexpression is an independent factor of poor prognosis that is significantly associated with lower DFS and OS in early HER2-positive breast cancer, and to lower OS in advanced HER2-positive breast cancer.

In addition, overexpression of this chemokine could be implicated in resistance to trastuzumab in early HER2-positive breast cancer. Patients with high levels of CCL5 have significantly poorer response rates to trastuzumab when treated with neoadjuvant anti-HER2 trastuzumab, and no pathologic complete response has been observed in these cases. The analysis of CCL5 in serum samples prior to treatment showed that a high CCL5 concentration significantly correlated with a poorer pathologic response to neoadjuvant treatment. Furthermore, the analysis of the evolution of CCL5 levels over the course of the disease (by comparing pre- and posttreatment samples) significantly demonstrated that continued exposure to trastuzumab resulted in increased CCL5 expression, which correlated with a higher rate of relapse.

Previous reports have proposed CCL5 as a good predictor of response to neoadjuvant trastuzumab therapy in HER2-positive breast cancer because of its primary function as chemoattractant of lymphocytes and immune cells (55). The CCL5 transcript levels determined in those works corresponded, however, to global expression levels that do not discriminate between expression in the stroma, lymphocytes, or tumor component; they also value the paracrine effect of CCL5, which increases lymphocyte recruitment. Our study, on the other hand, emphasizes the importance of CCL5 expression in the tumor component, revealing a lower pathological complete response (pCR) to trastuzumab when CCL5 expression is increased. Survival analysis from results from the TCGA database showed that patients with CCL5 overexpression exhibited a tendency toward a worse OS. In addition, the correlation analysis with the CCL5 receptors showed that the increase of CCR3 is significantly associated with a lower OS, and that there is a trend in the same direction for increased expression of CCR1 and CCR5. These data support the hypothesis that an increase in CCL5 levels and its implication in resistance to trastuzumab or its incidence in survival must be associated with a corresponding increase of its receptors (CCR1, CCR3, and CCR5). Accordingly, the scientific literature confirms that in HER2-positive breast tumors, CCL5 overexpression correlates with CCR5 overexpression, as well as an increase in CCR1 expression in the basal and HER2-positive subtypes (22, 42).

Our findings in samples of patients with HER2-positive breast cancer treated with trastuzumab indicate that CCL5 overexpression in the infiltrating tumor component behaves as an independent predictive factor of lower response to treatment and is associated with a poorer prognosis of the disease. In addition, we suggest that the activation or increase in the expression rate of some CCL5 receptors might be implicated in the acquired resistance to trastuzumab. In any case, determination of CCL5/CCL5–receptor expression levels in the tumor component in patients with early HER2-positive breast cancer who are candidates for treatment with trastuzumab could be used as a predictor of response to treatment.

A. Lluch has an advisory board relationship with Novartis, Pfizer, Roche/Genentech, Eisai, and Celgene. J. Albanell is a paid consultant and reports receiving a commercial research grant from Roche, reports receiving speakers bureau honoraria from Roche, has ownership interest (including patents) with Roche, has an advisory board relationship and has provided expert testimony with Roche. F. Rojo has an advisory board relationship with Roche, AstraZeneca, BMS, MSD, Novartis, Bayer, and Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Zazo, E. Martín-Aparicio, A. Rovira, J. Albanell, J. Madoz-Gúrpide, F. Rojo

Development of methodology: S. Zazo, P. González-Alonso, E. Martín-Aparicio, C. Chamizo, M. Luque, M. Sanz-Álvarez, P. Mínguez, I. Cristóbal, C. Caramés, A. Rovira, F. Rojo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Zazo, P. González-Alonso, E. Martín-Aparicio, C. Caramés, A. Lluch, O. Arpí, J. Madoz-Gúrpide, F. Rojo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Zazo, P. González-Alonso, E. Martín-Aparicio, P. Mínguez, G. Gómez-López, I. Cristóbal, C. Caramés, P. Eroles, J. Madoz-Gúrpide, F. Rojo

Writing, review, and/or revision of the manuscript: S. Zazo, P. Mínguez, P. Eroles, A. Lluch, A. Rovira, J. Albanell, J. Madoz-Gúrpide, F. Rojo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. González-Alonso, E. Martín-Aparicio, J. Albanell, J. Madoz-Gúrpide, F. Rojo

Study supervision: J. García-Foncillas, J. Albanell, J. Madoz-Gúrpide, F. Rojo

We thank Oliver Shaw for linguistic correction of the article. This work was supported by grants from the Spanish Ministry of Health, Consumer Affairs and Social Welfare (AES Program, grants PI15/00934, PI18/00382, and PI18/00006); the Biomedical Research Networking Centre for Cancer (CIBERONC); the Biobanks Platform, PT13/0010/0012; the Community of Madrid (S2010/BMD-2344); and ProteoRed (PRB2-ISCIII, PT13/0001). P. González-Alonso was supported by a Fundación Conchita Rábago de Jiménez Díaz grant. P. Mínguez was supported by the ISCIII Miguel Servet Program (CP16/00116).

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.