Introduction
Women with locally advanced or high-risk early-stage breast cancers are eligible for neoadjuvant chemotherapy (NAC), an approach whereby patients receive systemic chemotherapy before surgical removal of the tumor. NAC can downstage tumors to allow breast-conserving surgery to be performed [
1,
2] and permits the evaluation of individual tumor response to monitor the effectiveness of standard and/or investigational systemic therapy [
3‐
6]. Recent clinical studies involving NAC have provided an opportunity to evaluate prognosis based on the presence of residual disease and to elucidate predictors of response to different types of chemotherapeutics [
7‐
11]. These studies have consistently shown that cancers in women who respond to NAC with a pathological complete response (pCR) are much less likely to recur than those in women with residual disease [
5]. Unfortunately, only a subset of patients achieve a pCR to neoadjuvant treatment [
12‐
14]. The results of a large meta-analysis were recently reported [
15]. To date, there are no genomic markers that can predict response to NAC [
16,
17].
Why many patients fail to respond to NAC is poorly understood. Questions remain about which genes and pathways in breast tumors are perturbed in response to treatment [
8,
11] and how these molecular signals differ between responders and nonresponders, as well as in those whose cancer recurs early vs. those whose cancer does not. For example, early changes in the proliferation marker Ki-67 have been found to correlate positively with pathological response [
18]. Compared with pretreatment levels [
19], post-NAC Ki-67 levels appeared to show a stronger relationship with recurrence-free survival (RFS) [
20]. In addition to single-gene studies, genomic approaches are needed to maximize the discovery of useful classifiers and druggable targets in the NAC setting.
We performed an exploratory study to investigate the dynamics of tumor gene expression in early high-risk breast cancer patients receiving NAC. We analyzed serial cDNA microarray expression data obtained from breast tumor biopsies before treatment (T1), at 24–96 hours after the first dose of NAC (T2) and in residual tumors at the time of surgery (TS). We assessed differentially expressed genes between time points (T1 vs. T2 and T1 vs. TS) as well as changes in pathways and molecular subtypes. We also explored associations between early gene expression changes (T2 − T1) and response to chemotherapy, as well as gene expression in residual tumors (TS and TS − T1) and recurrence.
Discussion
The impact of NAC on the biology of breast cancers is not well understood. Recent clinical studies have included correlative gene expression analyses to understand the effects of NAC on the breast tumor transcriptome [
8,
11,
26‐
29]. For example, Hannemann and colleagues [
11] compared gene expression before and after NAC and observed that sensitive tumors showed significant changes in their gene expression, whereas resistant tumors did not. Gonzalez-Angulo and colleagues [
8] found that different pathways were preferentially perturbed in basal-like vs. non-basal-like breast cancers. In prior gene expression profiling studies, researchers have also attempted to identify candidate predictive markers for chemotherapeutic response using pretreatment biopsies [
30‐
32]. However, this approach does not account for chemotherapy-induced perturbations that may be gleaned from serial gene expression analysis of tumors in patients undergoing neoadjuvant treatment. Most importantly, examination of serial changes in tumor gene expression may identify predictors for favorable outcome with better sensitivity than baseline signatures alone [
33,
34].
In the present study, we examined changes in gene expression in serial breast tumor biopsies and the surgical specimen treated with NAC, as well as their association with clinical outcomes. We performed serial cDNA microarray profiling before, during and after treatment. We compared gene expression and molecular subtype assignment of matched tumors collected at different time points. We also performed exploratory analyses to evaluate associations between early changes in gene expression and response as well as expression in residual tumors and recurrence. To our knowledge, this is the first study of changes in tumor gene expression over three different time points in the neoadjuvant setting.
Transcriptomic analysis in breast cancer tumors revealed genes and pathways that were significantly perturbed after initiation of first cycle of NAC. Early effects of NAC on tumor gene expression include a substantial downregulation of genes and pathways involved in proliferation and immune function. These results must be interpreted with caution, however, as tissue biopsies were collected over a 3-day period. Therefore, confounding factors, including treatment-associated changes in the stroma and immune cell infiltrates, may potentially influence gene expression at this time point (T2).
The expression of the proliferation marker Ki-67 was evaluated by immunohistochemistry in a small subset of paired samples. Ki-67 expression decreased early during therapy, which is consistent with the observed downregulation of proliferation genes as determined by microarray analysis.
Residual tumors after NAC also showed significant modulation of genes and pathways involved in immune response pathways. These observations are consistent with the known immune-suppressive and cytotoxic effects of chemotherapy and the steroids that are often coadministered [
2,
35]. Genes and pathways significantly altered in residual tumors may represent candidate markers for chemoresistance [
36]. For example, perturbations in the expression of genes involved in IGF-1 signaling (e.g., the upregulation of
CYR61) have been observed in previous studies in which gene expression in pre- and post-NAC biopsies was compared [
11,
36]. Interestingly, the gene
CYR61 (
IGFBP10) has been shown to be associated with breast cancer progression [
37,
38] and with resistance to apoptosis [
39] and can be a potential target for therapy [
40].
We observed changes in molecular subtypes of tumors in a subset of patients during the course of NAC. Discordance in subtype assignment was higher when pretreatment tumors were compared with matched residual tumors at the time of surgery (T1 vs. TS) than with tumors obtained early in treatment (T1 vs. T2). The most frequent change in molecular subtype that we observed after cycle 1 of NAC was from luminal B to luminal A. Korde and colleagues reported similar findings [
31]. The switch from luminal B to luminal A may reflect the selective killing of highly proliferative cells that are chemotherapy-sensitive, leaving behind tumor cells that are more hormone-sensitive and less responsive to chemotherapy [
31]. Gonzales-Angulo et al. [
8] reported a similar concordance rate of subtypes (62 %) of paired pre- and post-NAC samples (n = 21) (Additional file
12). The lower concordance for T1 vs. TS than for T1 vs. T2 may reflect the much larger effect of the full regimen of NAC on the cellularity and composition of tumor tissue. It is unclear, however, whether the interconversion between non-basal-like subtypes is a result of chemotherapy exposure or is due to the relative instability in these subtypes as compared with the basal-like subtypes [
8]. Finally, our results regarding changes in molecular subtype assignment must be interpreted with caution, owing to the small sample size and the lack of consensus for molecular subtype assignment. Furthermore, changes in subtype can be attributed to reduced proliferation and changes in cellularity in both the tumor and the stroma. It is currently impossible to distinguish among these possibilities, and, even so, the clinical utility of the observation remains unclear.
Exploratory analysis comparing expression profiles of responders vs. nonresponders revealed differentially expressed genes involved in amino acid metabolism and cell proliferation. The relationship between amino acid metabolism and response to chemotherapy is currently unclear. The results of a recent preclinical study suggested, however, that the activation of amino acid metabolic pathways might be important in acquiring resistance to chemotherapy [
41]. We have shown in our previous work that high Ki-67 expression (n = 166 patients) at T1 was associated with favorable response to chemotherapy [
5]. In this study involving a small subset of patients with paired expression data, we did not observe a significant relationship between Ki-67 scores and response (
P = 0.18). Interestingly, we found that, early during chemotherapy, decreased expression of cell cycle inhibitors rather than increased expression of positive regulators of cell cycle (e.g., Ki-67 [
18]) was associated with poor response.
Molecular analysis of residual tumors may provide prognostic and predictive information and may facilitate the development of biomarkers, along with efficacious single-agent and combination therapies, to prevent or delay recurrence. Currently, there are no genomic predictors to determine which patients will experience an early recurrence [
7]. We found that increased interferon signaling over the course of chemotherapy among nonresponding patients was associated with shorter RFS, and we speculate that it may represent an immune tolerance response in aggressive disease. This seemingly contradictory relationship between interferon signaling and poor outcome vs. the more typically reported associations between T cell–B cell immune system signals and good outcome has been well documented [
42]. For example, a recent study has shown that activated interferon signaling is associated with increased risk of distant metastasis among luminal subtype tumors [
43]. Interferon signaling has also been associated with resistance to chemotherapy and radiation treatment [
44], and it has been identified as a coexpression module independent of other immune signaling in breast and other types of cancers [
45,
46]. Taken together, our results suggest that treatment-induced changes in interferon pathway signaling may be an important component in assessing risk factors for breast cancer recurrence and may hint at the potential utility of immune modulating therapy in nonresponding patients.
Examining changes in expression (e.g., T2 − T1 and TS − T1) may provide nonredundant information regarding response and recurrence beyond those obtained from static time points alone. Integrating gene expression data from different time points, including the changes observed between them, may facilitate the development of improved predictors for poor response and early recurrence.
Molecular markers of response and survival can vary across breast cancer subtypes. Therefore, molecular subtypes need to be considered when evaluating associations between tumor expression and clinical outcomes. In this study, the sample size was too small to perform subset analysis, but we found similar results in models adjusted for hormone receptor (HR) status. For example, analyses with or without adjustment for HR status revealed a significant association between RFS and interferon signaling genes (e.g., IFIT2, IFIH1 and IFI44L). Likewise, the gene expression changes that were most highly associated with response (e.g., CDKN2B, EIF4EBP1) were similar in both univariate and HR adjusted models.
A limitation of the present study was the relatively small sample size. Therefore, we consider our analysis to be exploratory, and larger studies are warranted to validate our findings. A similar study incorporating the approaches described here will be performed within the I-SPY 2 TRIAL [
47]. This neoadjuvant clinical trial is particularly well designed to test and develop predictive biomarkers because participating patients undergo serial tumor biopsies before treatment, after three cycles of neoadjuvant treatment and at the time of surgery. The present study did not include mutation profiling or characterization of immune infiltrates. We plan to perform these types of analyses in the I-SPY 2 TRIAL, which will potentially have greater power to validate specific associations between molecular data and clinical outcomes.
Competing interests
LJVV is a cofounder of Agendia. The rest of the authors declare that they have no competing interests.
Authors’ contributions
MJM, DMW and CY performed bioinformatics and statistical analysis and wrote the manuscript. JC and SED carried out cDNA microarray experiments. AA and CL performed pathologic review of samples. The I-SPY 1 Trial Investigators provided samples and clinical data for the study. CMH supervised and designed cDNA microarray analysis and participated in the review of the manuscript. LJVV, LE and HSR participated in study design and coordination. LJVV and JWP helped to draft the manuscript. All authors read and approved the final manuscript.