The Longitudinal Transcriptional Response to Neoadjuvant Chemotherapy with and without Bevacizumab in Breast Cancer Free

VEGF stimulates angiogenesis by influencing vessel formation through regulation of proliferation, migration, and survival of endothelial cells (1–3). Blocking VEGF by bevacizumab, an mAb, has been proposed to cause inhibition of neovascularization, regression of existing immature microvessels, and normalization of abnormal vasculature (4, 5); this has consequences for the flux of oxygen, nutrients, metabolites, and therapeutic agents, ultimately preventing tumor growth and resulting in tumor shrinkage. However, addition of bevacizumab to standard chemotherapy in unselected breast cancer patients has resulted only in a modest increase in response rate and progression-free survival (6–11). A more in-depth exploration of the effects of angiogenesis-blocking agents is important for future efforts to improve patient survival. The effect on the molecular level of bevacizumab administered in combination with chemotherapy has been less studied, and there is a lack of biomarkers for prediction of tumor responsiveness to bevacizumab (12). Breast cancer is a heterogeneous disease with at least five subtypes according to the PAM50 classification, each with distinct biology and clinical outcome (13–15). The frequency and prognostic impact of pathologic complete response (pCR) are known to vary between PAM50 subtypes (16–19). Subtype-specific responses to bevacizumab have not yet been systematically studied. Stratification of patients by molecular subtype may be important to identify patient subpopulations that will benefit from such treatment.

In this study, the effect of bevacizumab in HER2-negative breast carcinomas treated with a neoadjuvant chemotherapy regimen is investigated by comparing gene expression profiles of responding and nonresponding tumors. Tumors were also compared before, during, and after treatment in a subtype- and treatment-specific manner, to reveal molecular changes influenced by the therapy.

Patients were recruited at two sites in Norway (Oslo University Hospital, Oslo and St. Olav's Hospital, Trondheim), between November 2008 and July 2012. Written informed consent was obtained from all patients prior to inclusion. The study was approved by the Institutional Protocol Review Board, the regional ethics committee, the Norwegian Medicines Agency, and carried out in accordance with the Declaration of Helsinki, International Conference on Harmony/Good Clinical practice. The study is registered in the http://www.ClinicalTrials.gov/ database with the identifier NCT00773695.

Patients with HER2-negative mammary carcinomas with size ≥2.5 cm previously untreated for breast cancer were eligible. Other key inclusion criteria were WHO performance status ≤2, adequate hematologic and biochemical parameters, and no sign of metastatic disease. Additional prerequisites were normal organ function in general and normal left ventricular ejection fraction. Concomitant medications with anticoagulants, other than low-dose acetylsalicylic acid (160 mg or lower) were not allowed. A block randomization procedure was used, and the randomization was performed by the centralized research support facility at Oslo University Hospital (Oslo, Norway). The randomization list was not known to the personnel responsible for providing information or treatment to the patients. The patients were stratified on the basis of their tumor size (2.5 ≤ T ≤ 5 cm, T > 5 cm) and hormone receptor status [positive for estrogen (ER), progesterone, or both], and randomized 1:1 to receive bevacizumab and chemotherapy (combination therapy arm) or chemotherapy alone (chemotherapy arm). Of the 150 patients enrolled, 138 were assigned to treatment with chemotherapy, and 12 (independently randomized; not reported herein) received endocrine therapy as determined by the responsible oncologist. Of the 138 patients treated with chemotherapy, 66 in each group were included in the primary efficacy analysis (Fig. 1).

The chemotherapy regimen consisted of four cycles of FEC100 (5-fluorouracil 600 mg/m², epirubicin 100 mg/m², and cyclophosphamide 600 mg/m²) every 3 weeks, followed by docetaxel 100 mg/m² every 3 weeks or 12 weekly infusions of paclitaxel 80 mg/m². Bevacizumab was administered intravenously at a dose of 15 mg/kg every third week or 10 mg/kg every other week in patients receiving docetaxel or paclitaxel, respectively.

Hematologic parameters were evaluated before each chemotherapy administration, whereas biochemical parameters were evaluated every third week. The local pathologist, blinded to the treatment assignment, performed histopathologic examination of the breast using study-specific guidelines.

Samples were sequentially collected before treatment [core needle biopsies; termed “week 0 samples”; n = 132), during treatment (core needle biopsies 12 weeks into treatment and minimum 3 weeks after the last FEC dose; termed “week 12 samples”; n = 115), and after treatment (at surgery, minimum 3 weeks after the last taxane dose; termed “week 25 samples”; n = 112). Altogether, 358 specimens were available for further analyses. Matched tumor samples from all three time points were available from 96 patients.

pCR, herein defined as complete eradication of all invasive cancer cells in both breast and axillary lymph nodes, was the primary endpoint.

Gene expression profiling was performed using 40 ng total RNA and one color SurePrint G3 Human GE 8 × 60 k Microarrays (Agilent Technologies) following the manufacturer's protocol (details in Supplementary Data). Gene expression profiles were successfully obtained for 358 samples: 131 week 0 samples, 115 week 12 samples, and 112 week 25 samples. Microarray data are available in the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-4439.

The PAM50 subtyping algorithm developed by Parker and colleagues, 2009 (14), was used to assign a subtype label to each sample (details in Supplementary Data). A proliferation score was derived for each sample by computing mean expression values of 11 proliferation-related PAM50 genes: CCNB1, UBE2C, BIRC5, KNTC2, CDC20, PTTG1,RRM2, MKI67, TYMS, CEP55, and CDCA1 (14). TP53 mutation status was determined by sequencing the entire coding region (exons 2–11), including splice junctions (details in Supplementary Data).

All statistical analyses were performed in R version 3.0.3 (20). Associations between variables were assessed with Fisher's exact tests, t-tests and Kruskal–Wallis tests as appropriate. Significance Analysis of Microarrays (SAM) was used to find differentially expressed genes (DEGs) between two groups using the R package samr (21, 22) and reporting genes called significant at False Discovery Rate (FDR) 5%. Database for Annotation, Visualization, and Integrated Discovery (DAVID) was used for enrichment analysis on the DEGs. A measure of KEGG pathways activity was obtained using the R package qusage (23).

The study overview is presented in Fig. 1. Demographic characteristics were balanced between the treatment arms (Table 1). No significant skewness in distribution of important clinical and molecular parameters, such as tumor size, grade, lymph node status, hormone receptor status, TP53 mutation status, or PAM50 subtypes (Fisher exact test and χ² test), was observed, neither were any genes differentially expressed in the pretreatment biopsies between the treatment arms (two-class unpaired SAM).

Febrile neutropenia, proteinuria, and hypertension were the most frequent adverse events observed in both treatment arms (Table 2). In the combination therapy arm, significantly higher frequencies of bleeding disorders and hypertension were observed (Fisher exact test; P < 0.001). Serious adverse events, mainly febrile neutropenia and infection, were also significantly more frequent in this arm (Fisher exact test; P < 0.001 and P < 0.05, respectively). One death occurred after 12 weeks in the combination therapy arm, but despite autopsy, a specific cause could not be established.

A total of 23 of the 132 patients (17%) achieved pCR. The overall pCR rates were significantly higher in ER-negative versus ER-positive tumors, in TP53-mutated versus TP53 wild-type tumors, and in Basal-like subtype versus other PAM50 subtypes (Fisher exact test and χ² test as appropriate; P < 0.01). The frequency of pCR was 23% (n = 15) in the combination therapy arm compared with 12% (n = 8) in the chemotherapy arm (Fisher exact test, P = 0.17). For ER-positive tumors, bevacizumab-treated patients had a pCR rate of 20% (n = 11), whereas only 5% (n = 3) achieved pCR in the chemotherapy alone group (Fisher exact test, P = 0.02). Patients with ER-negative tumors did not seem to benefit from the addition of bevacizumab to chemotherapy in this patient cohort (Fig. 2).

Comparing tumor gene expression profiles before treatment in the combination therapy arm in patients that achieved pCR (n = 15) with those that did not (n = 51) identified 720 differentially expressed genes (DEG; two-class unpaired SAM; Supplementary Table S1). Functional annotation analyses of these genes using DAVID (24) showed that the gene list included known cell surface antigens (e.g., CD3D, CD8B, CD38), chemokines (e.g., CCL5, CXCL9, CXCL10), interleukins and interleukin receptors (e.g., IL12B, IL18, IL21), Toll-like receptors (e.g., TLR1, TLR2, TLR6), and tumor necrosis factors (e.g., TNF, TNFRSF13B, TNFRSF21). The gene list was significantly enriched for genes involved in immune response–related processes and included Gene Ontology (GO) terms, such as leucocyte cell–cell adhesion, regulation of lymphocytes, T-cell activation and proliferation, and positive regulation of lymphocytic proliferation (Supplementary Table S2).

Similar analyses in the chemotherapy alone arm identified 1,243 DEGs between the complete responders (n = 8) and the noncomplete responders (n = 57; two-class unpaired SAM; Supplementary Table S3). The gene list was modestly enriched for genes involved in cell cycle–related processes and included GO terms such as chromatin, nucleus and chromosomal part, cell division, and nuclear chromosome segregation (DAVID; Supplementary Table S4). A total of 173 DEGs (24% of DEGs in the combination therapy arm) between the complete responders and noncomplete responders overlapped among the two treatment arms and were enriched for genes involved in DNA damage and repair, and cell-cycle processes.

The gene expression of the DEGs was further correlated with the continuous response ratio (ratio of tumor size at start of treatment to size at surgery). In the ER-positive subset of the combination therapy arm, 345 of 720 DEGs remained after correction for multiple testing (Spearman correlation test, FDR < 0.05); the gene list was highly enriched for genes involved in immune processes. In the ER-positive subset of the chemotherapy arm, only one of 1,243 DEGs remained significant after multiple testing corrections (FDR < 0.05).

We next evaluated whether gene expression profiles were differently affected by the two treatment regimes. The week 12 samples in the combination therapy arm showed significant lower expression of 42 genes compared with the chemotherapy arm (two-class unpaired SAM; Supplementary Table S5). The gene list was highly enriched for genes involved in cell cycle–related and cellular components maintaining pathways (DAVID; Supplementary Table S6). Thus, at week 12, cell cycle–related processes were significantly more suppressed in the combination therapy arm. An overall decrease in PAM50 proliferation score (14) was observed in both treatment arms over time (Kruskal–Wallis test, P < 0.001; Supplementary Fig. S1). Notably at week 12, the proliferation score was significantly lower in the combination therapy arm compared with the chemotherapy arm (Kruskal–Wallis test, P = 0.006; Fig. 3A). Stratification by ER status showed a similar, but more pronounced, downregulation of cell-cycle genes at week 12 in the ER-positive subset of the combination therapy arm, with correspondingly larger difference in proliferation score between treatment arms (Kruskal–Wallis test, P = 0.001). In the ER-negative subset, no significant differences in gene expression profiles were detected between the treatment arms at any time point.

Following treatment, higher proportion of samples were categorized as luminal-A and normal-like, indicating a shift toward PAM50 expression profiles associated with better prognosis in response to treatment (Fig. 3B and C).

Net changes in gene expression and pathway activities in response to anthracycline ± bevacizumab (from week 0 to week 12) and taxane ± bevacizumab (from week 12 to week 25) were assessed for matched tumor pairs from individual patients. Analyses were performed separately for the PAM50 subtypes to uncover subtype-specific changes in response to treatment. HER2-enriched and normal-like subgroups were excluded due to limited numbers.

In the combination therapy arm, 2,524, 9,053, and 4,248 DEGs were identified between week 0 and week 12 samples in the basal-like, luminal-B, and luminal-A subtypes, respectively (Fig. 4; Supplementary Tables S7–S9). In the chemotherapy arm, 4,210, 2,346, and 1,082 genes showed differential expression in basal-like, luminal-B, and luminal-A subgroups respectively, with an overlap of 57%, 22%, and 20% with the corresponding gene lists in the combination therapy arm (Fig. 4A and Supplementary Table S10; Fig. 4B and Supplementary Table S11; Fig. 4C and Supplementary Table S12).

Pathway activity analyses performed using 186 pathways in the KEGG database revealed that corresponding to the highest numbers of genes affected in luminal-B tumors, several pathways showed significantly lower activity in the combination arm compared with the chemotherapy arm. Pathways involved in DNA replication and repair, cell cycle, RNA degradation, purine, and pyrimidine metabolism were among the top hits (Fig. 4D; Supplementary Table S13). For luminal-A and basal-like subgroups, fewer pathways showed significantly differential activity between the two treatment arms.

Changes in gene expression were less substantial from week 12 to week 25 and involved fewer DEGs compared with changes from week 0 to week 12. In the combination therapy arm, one, seven, and 2,241 DEGs were identified in the basal-like, luminal-B, and luminal-A subtypes, respectively; the corresponding numbers in the chemotherapy arm were none, 439, and 2,545 (Fig. 4A–C).

Interestingly, for the luminal-A subgroup in the chemotherapy arm, the number of DEGs after taxane treatment (between week 12 and week 25 tumors) greatly exceeded the corresponding number after anthracycline treatment (between week 0 and week 12 tumors; n = 2,545 vs. n = 1,082; Fig. 4C). A subsequent pathway activity analysis showed significant differences in pathway activity between the two treatment regimens for around 100 pathways, including retinol metabolism, PPAR signaling, ribosome and spliceosome synthesis, and cell-cycle and DNA repair pathways (Supplementary Fig. S2; Supplementary Table S14).

Corresponding to few/none DEGs in basal-like and luminal-B subgroup after taxane treatment, no significant differences in pathway activities were detected between week 12 and week 25 samples.

In this study, transcriptional response to treatment with standard chemotherapy, with and without bevacizumab, was studied in serial biopsies from 132 patients. More patients achieved pCR in the combination therapy arm compared with the chemotherapy-alone arm, in line with reports from previous studies (6, 7, 10, 11). A significant benefit of adding bevacizumab was observed in the ER-positive subgroup consistent with the larger NSAPB-B40 trial (11), in contradiction to the GeparQuinto trial, where the ER-negative tumors were found to benefit from bevacizumab (10), and the CALGB40603 study, including patients with a low ER expression, or triple-negative tumors (6). Different chemotherapy and bevacizumab regimens, pCR definitions, and threshold to define ER-positive/negative tumors make cross-study comparisons difficult. However, our results indicate possibilities for use of gene expression profiling for selection of patients who might benefit from bevacizumab.

Interestingly, immune response–related pathways were significantly upregulated in the complete responders in the combination therapy arm. This is the first unsupervised study to demonstrate a link between activated immune response pathways and pCR under treatment with bevacizumab. Previously, supervised studies have shown association between immune gene modules and tumor-infiltrating lymphocytes with standard neoadjuvant treatment response (25–30). Mechanisms such as promotion of immunogenic tumor cell death (31) by stimulating tumor antigen release, stimulation of dendritic cell maturation, proliferation of tumor-specific CD8⁺ cells, and sensitizing tumor cells to CD8⁺ T cell–mediated apoptosis (32–34) are suggested. The prominent association of activated immune response pathways and response in the combination therapy arm indicates that the immunogenic cell death promoted by immunomodulatory effects of chemotherapy may have been enhanced by the addition of bevacizumab, ultimately resulting in a higher incidence of pCR in tumors receiving combination therapy. Verification of this in independent cohorts is necessary to understand whether this can be clinically useful for selection of patients likely to benefit from such therapy. The association with activated immune response pathways is particularly interesting, in relation to the growing interest for testing immunologic checkpoint inhibition in breast cancer (35). It should be further investigated whether the responder population identified in our study has common features with responding patients treated with CTLA-4 or PD-1 checkpoint inhibitors.

Overall pCR rates were higher in tumors with unfavorable prognostic features such as ER negativity, presence of TP53 mutations, and basal-like gene expression profile. As ER-negative tumors tend to more often be TP53 mutated and basal like, it is challenging to evaluate the individual predictive power of these features. Differential response to preoperative chemotherapy in subtypes of breast cancer has been repeatedly reported (10, 18, 36, 37). High efficacy of cytotoxic agents resulting in death and elimination of highly proliferating tumor cell population is likely the reason behind the major shift in transcriptional programming observed in response to treatment; it potentially explains the overall shift in molecular profiles toward normal-like/luminal-A subtypes with low proliferating phenotypes as well as higher pCR rates in tumors with proliferating phenotype (14, 15).

Following treatment, the proliferation scores were reduced, likely corresponding to elimination of the rapidly proliferating tumor cells. Importantly, the magnitude of reduction in proliferation scores in the combination therapy arm was significantly higher compared with those in the chemotherapy arm. This accelerated reduction of proliferation scores also indicates that bevacizumab enhances the effects of chemotherapy. It is, however, difficult to infer whether this enhancement is a result of improved drug delivery due to changes in vascular permeability induced by bevacizumab.

Substantial changes in gene expression were observed in luminal-A tumors both from week 0 to 12 and from week 12 to 25 (Fig. 4C). In the chemotherapy alone arm, the luminal-A tumors showed suppressed ribosome and spliceosome synthesis pathways in response to anthracycline treatment, but significantly higher activity in response to taxane treatment (Supplementary Fig. S2). Upregulation of ribosome and spliceosome synthesis pathways may be related to treatment resistance in luminal-A tumors as efficacy of chemotherapeutic agents has been suggested to involve inhibition of ribosome biogenesis (38).

Differential gene expression and pathway activity between the two treatment arms were most evident in luminal-B tumors, exhibiting significantly lower activity of pathways involved in DNA repair and cell cycle (Fig. 4D). Elimination of almost all samples of luminal-B subtype in the combination therapy arm after treatment with anthracycline (Fig. 3B) may be related to this radical change in molecular activity. In the basal-like subtype, very few additional genes were affected by combination therapy, concordant with the lack of additional clinical response seen in this patient group.

The clinical toxicity increased with bevacizumab treatment, as more patients experienced febrile neutropenia in addition to hypertension and episodes of bleeding. Increases in febrile neutropenia and hypertension have also been seen in similar studies (10, 11). The clinical practice guidelines at the time of the study were not recommending the use of G-CSF for all patients, only those having experienced febrile neutropenia, and this may have increased the frequency of this adverse event.

In this study, immune response was found to be a strong predictor of response in patients treated with bevacizumab in addition to the standard chemotherapy. The effect of addition of bevacizumab at gene expression level was most evident in the ER-positive luminal-B tumors. In conclusion, the study provides valuable insights into changes in tumor behavior after neoadjuvant chemotherapy with and without bevacizumab, which may aid in identifying patients more likely to benefit from the addition of bevacizumab to the standard chemotherapy.

H.K.M. Vollan is an employee of F. Hoffmann—la Roche AG. O. Engebraaten reports receiving other commercial research support from Roche Norway. E.A. Wist reports receiving lecture fees from Novartis, Pierre Fabre, and Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E. Schlitchting, E. Wist, B. Naume, A.-L. Børresen-Dale, O. Engebraaten

Development of methodology: O.C. Lingjærde, A.-L. Børresen-Dale, O. Engebraaten

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. von der Lippe Gythfeldt, M. Krohn, T. Olsen, P. Vu, Ø. Garred, H. Skjerven, A. Fangberget, M.M. Holmen, E. Schlitchting, E. Wille, M.N. Stokke, S. Lundgren, E. Wist, B. Naume, A.-L. Børresen-Dale, O. Engebraaten

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Silwal-Pandit, S. Nord, E.K. Møller, E. Rødland, M.M. Holmen, E. Schlitchting, H.K.M. Vollan, V. Kristensen, A. Langerød, E. Wist, B. Naume, O.C. Lingjærde, A.-L. Børresen-Dale, O. Engebraaten

Writing, review, and/or revision of the manuscript: L. Silwal-Pandit, S. Nord, H. von der Lippe Gythfeldt, E.K. Møller, T. Fleischer, Ø. Garred, M.M. Holmen, E. Schlitchting, E. Wille, H.K.M. Vollan, V. Kristensen, A. Langerød, S. Lundgren, E. Wist, B. Naume, O.C. Lingjærde, A.-L. Børresen-Dale, O. Engebraaten

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Silwal-Pandit, E. Borgen, M.N. Stokke, O. Engebraaten

Study supervision: M.N. Stokke, O.C. Lingjærde, A.-L. Børresen-Dale, O. Engebraaten

The contribution to the study from all the participating patients is greatly acknowledged.

The study was funded in part by generous grants from the Pink Ribbon Movement and Norwegian Breast Cancer Society (project no. 11003001), and the Norwegian Research Council (project no. 191436/V50). In addition, K. G. Jebsen Center for Breast Cancer Research and South-Eastern Norway Regional Health Authority supported the project. The study was co-sponsored by Roche Norway and Sanofi-Aventis Norway.

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.