Introduction
The use of adjuvant systemic therapy for breast cancer has increased substantially in the last few decades, and is now a mainstay modality in clinics worldwide. Although adjuvant chemotherapy has significantly improved patient management, a fundamental limitation of this approach is that many patients fail to benefit from therapy, and physicians are frequently unable to predict the responses of individual patients to a given regimen. As such, much effort has been directed in recent years toward identifying clinical and biological predictive features to better tailor therapy to the needs of individual patients.
Anthracycline-based chemotherapy (regimens involving anthracyclines such as doxorubicin or epirubicin) has been used clinically for more than two decades and has largely supplanted first-generation regimens such as CMF [
1]. Anthracyclines are thought to exert their effects through a variety of mechanisms including intercalation of DNA, cross-linking of DNA to proteins, and generation of free radicals [
2]. Nevertheless, the precise mechanisms by which anthracyclines exert their therapeutic effects
in vivo remain unclear, nor has significant progress been made in establishing predictive biomarkers to identify patients likely to derive benefit from anthracycline-based therapy [
2,
3]. For example, aberrations of
HER2 and
TOP2A initially showed promise as anthracycline-response predictors, but have failed to demonstrate consistent clinical utility [
2].
Attention has focused recently on the prognostic and predictive potential of antitumor immune responses, detected indirectly via gene expression signatures derived from tumor-infiltrating lymphocytes (TIL) or directly via immunohistochemical TIL staining. In breast cancer, TIL have been shown in several studies to correlate with favorable long-term prognosis, although primarily for hormone receptor-negative, Her2-positive, or high grade/highly proliferative lesions [
4‐
10]. Denkert
et al. [
11] recently reported that TIL were associated with a favorable response to neoadjuvant anthracycline/taxane therapy in two large breast cancer cohorts, providing compelling evidence that TIL could potentially serve as predictive markers for anthracycline-based therapy. However, no such studies have been conducted specifically with the ER-negative breast cancer subset. This is important for two reasons: first, pathologic complete response to neoadjuvant chemotherapy, defined as the complete absence of invasive tumor cells in the breast and lymph nodes after treatment, occurs almost exclusively in tumors negative for ER [
12‐
15]; second, ER-negative tumors typically feature higher levels of TIL than do ER-positive tumors [
6,
11,
16‐
18]. Therefore, our objective in this study was to determine whether TIL correlate with sensitivity to anthracycline-based chemotherapy in ER-negative breast cancer. To do so, we examined two independent cohorts of ER-negative breast cancer cases. One cohort received anthracycline-based therapy in the neoadjuvant setting, permitting the assessment of short-term clinical responses to therapy. The second cohort included patients with long-term follow-up treated with anthracyclines in the adjuvant setting, allowing us to examine overall and disease-free survival rates after systemic therapy.
Materials and methods
Neoadjuvant chemotherapy cohort
To assess the relationship between TIL and short-term response to chemotherapy, we analyzed publically available gene expression data [
19] derived from pretreatment incisional or core tumor biopsies from a cohort of patients with ER-negative invasive ductal breast carcinomas. Note that the original publication reporting analysis of gene signatures and treatment response [
19] has since been retracted because of concerns regarding the statistical methods used to generate predictive models (see
Lancet Oncol 2011, Feb; 12:116 and
Lancet Oncol 2010 Sep; 11:813-814). However, our analysis is based on original data obtained from the Gene Expression Omnibus website (
http://www.ncbi.nlm.nih.gov/geo; accession number GSE6861) and is therefore not affected by this issue. Clinical characteristics of the study cohort are summarized in Table
1. All patients were enrolled in the EORTC 10994/BIG 00-01 clinical trial and received chemotherapy with either FEC (fluorouracil, epirubicin, and cyclophosphamide for six cycles) or TET (three cycles of docetaxel followed by three cycles of docetaxel plus epirubicin) as primary (neoadjuvant) therapy. The primary end point was pathologic complete response (pCR) versus residual disease (RD).
Table 1
Characteristics of all assessed patients in the EORTC and MBTB cohorts
Age at diagnosis | <50 years | 56 | 50% | 85 | 33% |
| ≥50 years | 57 | 50% | 168 | 66% |
Tumor size | ≤2 cm | 3 | 3% | 51 | 20% |
| <2 cm to ≤5 | 68 | 60% | 142 | 56% |
| >5 cm | 42 | 37% | 39 | 15% |
| Unknown | 0 | 0 | 23 | 9% |
Nodal status | Negative | 42 | 37% | 88 | 35% |
| Positive | 71 | 63% | 142 | 56% |
| Unknown | 0 | 0 | 25 | 10% |
Grade | 1-2 | 36 | 32% | 120 | 47% |
| 3 | 65 | 58% | 119 | 47% |
| Unknown | 12 | 11% | 16 | 6% |
ERa | Negative | 113 | 100% | 255 | 100% |
| Positive | 0 | 0 | 0 | 0 |
PRb | Negative | 110 | 97% | 212 | 84% |
| Positive | 3 | 3% | 41 | 16% |
| Unknown | 0 | 0 | 2 | <1% |
Her2c | Negative | 83 | 73% | 157 | 62% |
| Positive | 30 | 27% | 58 | 23% |
| Unknown | 0 | 0 | 40 | 16% |
Analysis of microarray data
To determine whether lymphocyte-associated gene expression correlated with pCR rates, we analyzed microarray data from the neoadjuvant (EORTC) cohort and obtained an eight-gene TIL signature by using the following steps. We first compiled a list of genes with known biological functions that are expressed at high levels by lymphocytes, based on information in the BioGPS gene portal (
http://biogps.gnf.org). Expression data for these genes were compiled and z-normalized (that is, number of standard deviations from the mean). For genes with redundant probe sets, only those probes with the highest interquartile range were selected. To enrich for genes with expression patterns consistent with a lymphocyte cell of origin, expression values of all genes were Spearman correlated with those of
CD247 (CD3-zeta chain) and
MS4A1 (CD20), which were chosen as specific and prototypical T- and B-lymphocyte genes, respectively. Only genes that correlated to
CD247 or
MS4A1 with a Spearman
R value of 0.7 or greater were retained, resulting in a list of 33 genes. With this gene set, cases were grouped by unsupervised hierarchical clustering using the centroid method and euclidean distance as the similarity metric (performed with Cluster 3.0). Heat maps and cluster dendograms were produced using Java TreeView.
We further refined the TIL signature by assessing differential expression of each gene in pCR versus RD cases and selecting genes with a Mann-Whitney t test P value less than 0.05. The final eight-gene signature included CD19, CD3D, CD48, GZMB, LCK, MS4A1, PRF1, and SELL. Data corresponding to this signature were used to cluster the EORTC cases, as described above. Two cases that failed to cluster with any of the primary centroids were considered outliers and discarded from further analysis, leaving a final assessed cohort size of 111 cases. Univariate associations of TIL status and pathologic parameters with pCR rates were calculated using two-sided Fisher's exact tests in Prism 5.0 (GraphPad, La Jolla, CA). Multivariate logistic regression models for prediction of pCR were constructed in SPSS Statistics 14 (SPSS, Chicago, IL).
Adjuvant chemotherapy cohort
To assess the relationship between TIL and long-term outcome after adjuvant chemotherapy, we analyzed TIL in primary tumor biopsies or resections taken before adjuvant therapy from a cohort of patients with ER-negative invasive ductal breast carcinomas registered by the Manitoba Breast Tumor Bank (MBTB). Cases were accrued between the years 1988 and 2000 and had a median follow-up time of 83 months. Clinical characteristics of the study cohort are summarized in Table
1 and Additional file
1 (Supplementary Table S1). All selected cases were determined to be ER-negative by a single central provincial clinical laboratory and subsequently managed as ER-negative cases at a single provincial cancer center, where adjuvant systemic therapy was administered to 61% of the cohort. Of these patients, 58% were treated with CMF (cyclophosphamide, methotrexate, fluorouracil), and 37% were treated with one of several anthracycline-based regimens (for example, AC, CAF, CEF, where A is Adriamycin/doxorubicin, and E is epirubicin). Disease-free survival (DFS) was defined as the time from surgery to the first instance of disease recurrence or disease-specific death, and overall survival as the time to death from all causes. The MBTB has approval from the Research Ethics Board, Faculty of Medicine, University of Manitoba, to collect, store, and distribute anonymous cases from its archive under the Canadian Tri-Council Policy Statement waiver of consent. The current study was conducted with approval from the University of British Columbia/British Columbia Cancer Agency Research Ethics Board.
Tissue microarray construction
An initial cohort of 255 ER-negative cases was selected on the basis of (a) ER-negative status defined by ligand-binding analysis of <10 fmol/mg protein; (b) invasive ductal carcinoma components occupying more than 20% of the tumor section, and (c) no prior therapy. To construct a tissue microarray (TMA), all cases were rereviewed to confirm and select areas for coring of corresponding blocks. Duplicate tissue cores (0.6 mm diameter) were taken from central cellular areas of each tumor with a tissue arrayer instrument (Beecher Instruments, Silver Spring, MD). Prior use and exhaustion of some tissue cores reduced the interpretable cohort size to approximately 160, depending on the TIL marker analyzed.
Immunohistochemistry and TMA scoring
Immunohistochemistry (IHC) was performed for CD3, CD8, CD4, CD20, and TIA-1 on deparaffinized sections from TMAs by using a Ventana Discovery XT autostainer (Ventana, Tucson, AZ). Ventana's standard CC1 protocol was used for antigen retrieval. Primary antibodies used are as follows (clone, supplier, animal source, concentration): CD3 (RM-9107, Lab Vision, rabbit, 1/150); CD8 (RM-9116, Lab Vision, rabbit, 1/100); CD4 (MS-1528, Lab Vision, mouse, 1/10); CD20 (polyclonal, catalogue no. RB-9013, Lab Vision, rabbit, 1/250); TIA-1 (TIA-1, Abcam, mouse, 1/50). TMA sections were incubated with primary antibodies for 60 minutes at room temperature followed by the appropriate cross-adsorbed, biotinylated secondary antibody (Jackson Immunoresearch, West Grove, PA) for 32 minutes. Antibodies were detected using the DABMap kit (Ventana). Slides were counterstained with hematoxylin (Ventana) and coverslipped manually with Cytoseal-60 (Richard Allan, Kalamazoo, MI).
IHC scoring was undertaken by using a microscope eyepiece grid to standardize the assessed area. In brief, duplicate cores of each immunostained tumor were reviewed at low magnification, and the core exhibiting a tumor/stroma ratio closest to 50:50 and the highest density of positive cells was selected. This core was then assessed at higher magnification (×20 objective) with a grid overlaid on the center of the core. Under a ×20 objective magnification, this grid defined an area of 0.56 mm
2. The number of positive intraepithelial lymphocytes was quantified within the area of the grid (intraepithelial localization was defined as lymphocytes within tumor cell nests or in direct contact with tumor cells, consistent with the method used by Denkert
et al. [
11]). To account for variation in epithelial-stromal proportions between different samples, intraepithelial TIL levels were calculated by dividing the number of observed intraepithelial TIL by the fraction of grid area occupied by epithelium.
Statistical analysis of IHC data
All analyses were performed using Prism 5.0. Unless otherwise specified, median values for TIL markers were used as predetermined cut-points to define high versus low cases. Associations of TIL with pathologic features were evaluated with Fisher's exact test. Survival outcomes were assessed via Kaplan-Meier methods and compared using Log-rank tests. All statistical tests were two-sided, with significance established at P values less than 0.05.
Additional validation datasets
We assessed publically available gene-expression data from ER-negative breast cancers within three additional cohorts (accessible via the Gene Expression Omnibus website), the characteristics of which are as follows: GSE21974, a cohort of 32 breast cancer patients (14 ER-negative) treated with neoadjuvant epirubicin, cyclophosphamide, and docetaxel (not previously published); GSE19615 [
20], a cohort of 115 breast cancers, 36 of which were ER-negative and treated with adjuvant anthracycline-based therapy (primarily AC and AC plus taxol); and GSE18864 [
21], a cohort of 28 triple-negative breast cancers (24 with complete data) treated with neoadjuvant cisplatin. We tested the eight-gene TIL signature in these cohorts by extracting expression data for the eight genes and processing it as described above. In each cohort, patients were divided into two groups based on clinical outcome. For GSE21974, this was based on pathologic complete response or residual disease; for GSE19615, patients were divided based on the presence or absence of distant recurrence after 36 months of follow-up; and for GSE18864, patients were divided into good and poor response groups based on Miller-Payne scores of 3, 4, or 5 versus scores of 0, 1, or 2, respectively. For each response group, the average expression level of each gene was calculated and plotted as a box-and-whiskers plot; Mann-Whitney
t tests were used to assess differences in overall TIL-signature expression.
Discussion
We have demonstrated that a subgroup of ER-negative breast cancer defined by high lymphocyte gene expression has a remarkably high rate (74%) of pathologic complete response to neoadjuvant anthracycline-based therapy. We have further shown that the clinical benefit of adjuvant anthracycline-based therapy may be restricted to patients with high numbers of tumor-infiltrating T-cells, especially those of the CD8+ and TIA-1+ subsets. In contrast, TIL show no association with outcome after classic CMF therapy.
Our findings are consistent with several prior reports. In 2008, Hornychová
et al. [
29] observed high levels of intraepithelial CD3
+ TIL in pretreatment biopsies of patients who achieved pCR after neoadjuvant therapy with doxorubicin and paclitaxel. Ladoire
et al. [
30] analyzed pre- and post-treatment samples from 56 patients treated with neoadjuvant anthracycline-based therapy and found that high ratios of CD8
+ to Foxp3
+ TIL (putative regulatory T-cells (Tregs)) in surgical specimens were correlated with a high rate of pCR. Consistent with our own findings, they also reported that, relative to pretreatment levels, TIA-1
+ and GzmB
+ TIL (CTL) were significantly increased in the posttreatment samples of patients who achieved pCR, but not in those who had residual disease. Finally, Denkert
et al. [
11] reported a significant relationship between TIL (identified by a combination of H&E assessment, IHC for CD3 and CD20, and expression analysis of several TIL genes by PCR) and pathologic response to neoadjuvant anthracycline/taxane therapy in a large group of 1,058 patients, approximately one fourth of whom were ER-negative. As such, based on our data and prior reports, TIL are consistently associated with improved outcomes after anthracycline-based therapy. Intriguingly, in 2001, Demaria
et al. [
31] observed in a small group of breast cancer patients (
n = 25) treated with neoadjuvant paclitaxel therapy that levels of CD3
+ TIL increased in the tumor bed after treatment, and that this correlated with therapeutic response. Although our data imply that TIL are not related to therapeutic responses in patients treated with cisplatin or CMF, the Demaria study nevertheless suggests that the correlation of TIL with treatment response may not be restricted entirely to anthracycline-based therapy.
It is important to note that TIL have been correlated with improved survival in several breast cancer studies to date, the association being observed primarily in high-grade, ER-negative, or Her2-positive lesions [
4‐
10]. Given that many of the patients evaluated in these studies were not systemically treated, TIL may also be a feature of breast cancers with a naturally favorable clinical outcome, which could have implications for the interpretation of the results in our study. This may also be the case for a variety of other cancer types, including malignancies of the colon [
32], ovary [
33], lung [
34], and bladder [
35]. In the systemically untreated patients of our long-term follow-up cohort, however, those with high levels of CD3
+ or CD8
+ TIL had only a modest, nonsignificant survival advantage over those with low TIL levels. It is therefore unlikely that the inherent association between TIL and improved prognosis was a significant underlying feature of our observations with regard to chemotherapy response.
The mechanism underlying the relationship between TIL and anthracycline sensitivity in humans is not currently understood. It is possible that TIL may not be causally related to therapeutic outcomes, and are instead biomarkers of an unknown feature that burdens tumors with anthracycline sensitivity. However, it is equally plausible that TIL directly mediate treatment responses. Indeed, this is supported in preclinical models, discussed here at length.
Four primary explanations for the relationship between anthracyclines and immunity have arisen based on observations in both the laboratory and the clinic [
36,
37]. One possible scenario is that anthracyclines kill or suppress Tregs in the tumor microenvironment, thus relieving inhibition of the anti-tumor immune response. For example, in the aforementioned study by Ladoire
et al. [
30], Foxp3
+ TIL were significantly reduced in the tumor bed after successful neoadjuvant anthracycline-based therapy, whereas their levels were unchanged in patients with residual disease. However, interpretation of this result is difficult because patients in the Ladoire study received a regimen (FEC) containing cyclophosphamide, a drug known to deplete Tregs selectively at low doses (as well as conventional T-cells at high doses) [
38‐
41]. Similarly, many of the drug combinations used to treat patients in our study also included cyclophosphamide. Further studies designed to elucidate the specific effects of anthracyclines on intratumoral Treg populations will be required to determine the validity of this concept.
A second possible mechanism is derived from observations that transient lymphopenia (induced therapeutically via low-dose total-body irradiation or by myelosuppressive drugs such as cyclophosphamide) can enhance immunotherapy in the context of adoptive transfer or vaccination. Through a variety of mechanisms, therapeutic lymphopenia appears to trigger homeostatic processes that favor the proliferation and functionality of anti-tumor effector T-cells, while also ameliorating tumor-induced immunosuppression [
42]. However, considering the myelosuppressive properties of cyclophosphamide, this idea does not account for our observation that TIL are associated with outcome after anthracycline-based but not CMF chemotherapy.
A third hypothesis is that chemotherapy can sensitize tumor cells to T-cell-mediated cytotoxicity. For example, Yang and Haluska [
43] found that treatment of human melanoma cell lines with either 5-fluorouracil or dacarbazine improved the efficacy of perforin/granzyme-mediated killing by antigen-specific CTL. Similar results have been reported with respect to paclitaxel, cisplatin, and doxorubicin [
44]. Cyclophosphamide has also been shown to sensitize mesothelioma cells to CTL killing, although through a TRAIL (TNF-related apoptosis-inducing ligand)-mediated effect [
45]. Finally, doxorubicin has been shown to induce Fas expression in breast cancer cell lines, leading to increased susceptibility to Fas ligand-mediated apoptosis [
46]. These studies collectively suggest an attractive model involving cytotoxic synergy between chemotherapy and the immune system. Nevertheless, as with the myelosuppression model discussed above, this mechanism would support a relationship between TIL and CMF therapy, which is not observed in our study.
The fourth hypothesis is that anthracyclines, unlike most chemotherapeutic drugs, induce immunogenic tumor cell death and thereby function indirectly as immunostimulatory agents [
42]. Casares
et al. [
47] treated murine CT26 colon carcinoma cells with doxorubicin and inoculated the dying cells into syngeneic hosts, which resulted in protection from concurrent or subsequent challenge with live CT26 cells. Notably, this effect was dependent on dendritic cells and CD8
+ T-cells and was specific for the CT26 tumor line, indicating an immunologic mechanism. In contrast, cells stimulated to undergo apoptosis by using mitomycin-C, a non-anthracycline DNA cross-linking agent, did not elicit protective immunity. Obeid
et al. [
48] expanded on this work by demonstrating that anthracyclines trigger preapoptotic shuttling of calreticulin to the dying cell surface, where it acts as an opsonin to stimulate phagocytosis and processing by dendritic cells. Compared with a variety of apoptosis-inducing drugs with various mechanisms of action (including staurosporine, tunicamycin, mitomycin-C, and etoposide), anthracyclines induced greater calreticulin surface presentation on target cells, which correlated with the ability to elicit protective immunity. Indeed, blockade or knockdown of calreticulin in anthracycline-treated cells abolished their immunogenicity in mice [
48]. Finally, Apetoh
et al. [
49] demonstrated that doxorubicin additionally causes the extracellular release of HMGB1 (high-mobility-group box 1), an endogenous ligand of Toll-like receptor 4 (TLR4). HMGB1 stimulation of dendritic cells was necessary for processing and cross-presentation of antigens from dying tumor cells. The authors further showed that breast cancer patients bearing loss-of-function
TLR4 alleles had significantly worse prognosis when treated with anthracycline-based chemotherapy [
49]. Taken together, mounting evidence indicates that anthracyclines have the rare ability to elicit specific immunity against target cells via stimulation of dendritic cells in a calreticulin and HMGB1/TLR-4-dependent fashion. This may explain the observation that anthracycline-based therapy appears most effective in breast cancer patients with high levels of endogenous tumor immunity, as well as the finding of Ladoire
et al. [
30] that CTL accumulate in the tumor beds of patients treated successfully with neoadjuvant anthracycline-based therapy. One unanswered question is whether anthracycline-based treatments can induce new tumor antigen-specific T-cells (
de novo immunity) or, alternatively, augment the expansion or functional activity of preexisting tumor-reactive T-cells.
An important confounding feature of our study is the heterogeneous and multivalent chemotherapy used in the assessed cohorts. Although we have emphasized anthracyclines as the unifying components of the treatments we assessed, it is difficult to determine the extent to which anthracyclines are specifically responsible for our observations. For example, many of the patients in this study were treated with regimens involving taxanes, which may also have clinically relevant effects on host immunity [
31]. Furthermore, the potentially competing or cooperative effects on tumor immunity that agents such as anthracyclines, cyclophosphamide, and taxanes may possess when administered sequentially or simultaneously are not currently understood. A second caveat worthy of note pertains to the use of gene-expression arrays to infer the presence of TIL. The use of IHC to detect TIL is advantageous for several reasons, including the ability to directly quantify cells that express a given marker, as well as to accurately determine their specific localization within a tissue. This information cannot be acquired by using conventional methods of gene-expression analysis. Thus, although the data presented for the eight-gene TIL signature reflect a relative abundance of TIL, they lack the desirable single-cell resolution that is achievable through IHC.
A further limitation of this study is that the adjuvant cohort did not contain enough anthracycline-treated cases to permit rigorous multivariate analysis (although this was performed in the neoadjuvant cohort). This was largely because ER-negative tumors constitute only ~30% of breast cancer cases, and CMF was the dominant systemic regimen for breast cancer in Canada at the time that this cohort was treated. Our additional validation cohorts are also small for two primary reasons: first, because many publically available datasets are derived from fine-needle aspirates (as opposed to core or open-biopsy specimens), and so may not accurately reflect overall lymphocyte infiltration; and second, because most cohorts contain both ER-positive and ER-negative tumors, with relatively few cases of the latter. Our observations thus warrant validation in large, independent cohorts with long-term follow-up. Ultimately, prospective trials of therapeutic outcome prediction will be required to fully assess the clinical utility of the concepts presented herein.
Our findings indicate that TIL, which are easily detectable with standard IHC, may prove useful for identifying patients who are likely to benefit from anthracycline-based therapy. Moreover, our findings suggest that patients with low TIL levels may benefit most from alternative regimens such as CMF. In addition, our study provides support for the idea of combining anthracycline-based chemotherapy with immunotherapeutic strategies such as cancer vaccines or adoptive T-cell therapy to fully engage the host immune response against breast cancer.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
NRW participated in study design, performed analyses, and drafted/revised the manuscript. KM performed the immunohistochemistry. PTT, NM, and BHN contributed to study design, data interpretation, and manuscript revision. PHW conceived the study, scored IHC tissue sections, and participated in data interpretation and manuscript revision. All authors read and approved the final manuscript.