Endogenous Versus Tumor-Specific Host Response to Breast Carcinoma: A Study of Stromal Response in Synchronous Breast Primaries and Biopsy Site Changes

Although prognostication of malignant neoplasms has been classically focused on tumor histopathology, emerging studies have highlighted an interplay between tumor epithelium and stromal response, the results of which have significant implications for tumor progression and clinical outcome (1–15). Olumi and colleagues demonstrated that transplanted fibroblasts previously associated with prostatic carcinomas stimulated tumor progression compared to fibroblasts adjacent to normal prostatic epithelium (1). Ayala and colleagues showed that specific stromal markers in prostatic carcinoma were significant predictors of recurrence-free survival (2). Furthermore, the amount of reactive stroma to prostate carcinomas was shown to be an independent predictor of cancer recurrence (3). In the area of breast carcinoma, Orimo and colleagues showed that implanted carcinoma-associated fibroblasts promoted growth of carcinomatous epithelium more so than fibroblasts associated with normal epithelium (4).

Our previous studies of tenosynovial giant cell tumor and desmoid-type fibromatosis showed 2 distinct patterns of gene overexpression. Tenosynovial giant cell tumors and pigmented villonodular synovitis showed a macrophage-associated pattern of gene overexpression, which we refer to as the “CSF1 response.” Desmoid-type fibromatosis and solitary fibrous tumors express a fibroblast-associated pattern of gene overexpression, which we refer to as the “DTF response.” We subsequently applied these 2 patterns of gene expression to the study of stromal response in breast cancer. In doing so, we have characterized 2 sets of stromal response in breast cancer which are associated with tumor features and patient survival: the CSF1 macrophage stromal response signature was associated with higher tumor grade and decreased hormonal receptor expression (5), whereas the DTF fibroblast (fibromatosis-like) stromal signature was associated with lower grade tumors, estrogen receptor positivity, and longer survival (6). In subsequent studies, we showed that both CSF1 and DTF signatures could be identified in in situ lesions (16). Although the 2 stromal profiles have been characterized with respect to survival and tumor grade, it remains unclear whether these stromal responses are generated from a baseline host response to all malignant growth, or whether the tumor itself induces a specific set of stromal responses.

In this study, we address the role of endogenous host response versus tumor-specific response by studying stromal signatures in patients with dual breast primaries. We then evaluated matched patient samples of normal stroma, biopsy site stroma, and stroma adjacent to carcinoma to distinguish tumor-specific response from generalized injury response.

HIPAA compliant Stanford University Medical Center and University of Washington Medical Center (UWMC) institutional review board approval was obtained for this study. The UWMC pathology database was searched to identify patients with previously excised synchronous, independent primary breast cancers. Synchronous, independent primaries were defined as invasive carcinomas that either presented in separate breasts or as 2 clinically distinct carcinomas within the same breast. Twenty-six patients with archival formalin fixed paraffin embedded tissue were identified with synchronous independent primary breast cancers for a total of 52 cases (TA228, Table 1). To specifically study the stromal responses between normal, malignant, and biopsy site changes, a second tissue microarray (TMA 242) was constructed using the same material as that used in the breast scar gene expression profiling studies. Areas representing a range of lesions were taken, including normal breast tissue (27 spots), granulomatous mastitis (2), previous biopsy sites (27), ductal carcinoma in situ, DCIS (1), infiltrating ductal carcinoma, IDC (19), and infiltrating lobular carcinoma, ILC (2) for a total of 78 tissues.

Tissue sections were deparaffinized followed by blockade of endogeneous peroxidases and antigen retrieval using Antigen Unmasking Solution (Vector). Estrogen receptor (ER) clone 1D5 using a dilution of 1:1000, following a 15 minute pretreatment in citrate buffer, pH = 6.0; progesterone receptor (PR) (BioGenex) clone PR88 using a dilution of 1:100 following an 18 minute pretreatment in citrate buffer pH = 6.0; and Her-2/neu (Dako) using a dilution of 1:800, following a 15 minute pretreatment in citrate buffer, pH = 6.0. The slides were then counterstained in hematoxylin, dehydrated, and mounted. Positive and negative controls were performed. HER2 was defined as positive if IHC was 3+ (strong circumferential membranous staining). It was considered negative in those tumors scoring 0–1+ (no to weak, noncircumferential staining). For those tumors with a 2+ score on IHC, gene amplification using FISH was used to determine HER2 status. HER2 was considered positive if the ratio of the copies of chromosome 17 to the number of HER2 gene copies was >2. Cellular proliferation was assessed by measurement of Ki-67 antigen by MIB-1 antibody (Dako), scoring the percentage of positive cells. Ki-67 was stratified into 4 levels depending on percentage of nuclear labeling: score = 0, <5% labeling; score = 1, 5–10% labeling; score = 2, 10–25% labeling; score = 3, >25% labeling. Nottingham grade was scored using the modified Scarff–Bloom–Richardson system.

Tissue microarrays of synchronous breast cancers (TA228) and breast scar (TA242) were constructed and immunohistochemical (IHC) stains were performed. Serial sections of 4 μm were cut from the TMA blocks, deparaffinized in xylene, and hydrated in a graded series of alcohol. The slides were pretreated with citrate buffer and a microwave step. The expression of 5 fibroblast response markers (DTF signature) was evaluated (cadherin 11, MMP 11, CSPG2, CD138, osteonectin). The expression of 4 macrophage response markers (CSF1 signature) was evaluated as well (cathepsin L, CD163, CD16, CD32). The primary antibodies used were osteonectin (Zymed, 1:1000 dilution, ENVISION citrate), CSPG2 (Santa Cruz Biotechnology incorporated, 1:150 dilution, VECTOR citrate), cadherin 11 (Invitogen, Catalog no. 32-1700, 1:10 dilution, ENVISIOn citrate), SDC1 (CD138, Serotec, Catalog no. MCA681H, 1:30 dilution, Ventana Benchmak Autostainer ventana mild CC1) and MMP11 (Calbiochem, Catalog no. IM86, 1:200 dilution, VECTOR citrate), Cathepsin L (CTSL1, MCA2374, mouse monoclonal; 1:25 dilution, AbD Serotec, ENVISION no pretreatment), CD163 (NCL-CD163, mouse monoclonal; Novocastra 1:100 dilution, Ventana Benchmak Autostainer ventana mild CC1), CD16 (FCGR3a; MCA1816, mouse monocloncal; AbD Serotec, 1:40 dilution, ENVISION citrate), and CD32 (FCGR2a; AB45143, rabbit monoclonal; Abcam, 1:200 dilution, ENVISION Citrate). The IHC reactions were visualized using mouse and rabbit versions of the EnVision+ system (Dako) and Vector Vectastain Elite ABC mouse and rabbit systems (Vector Labs) using diaminobenzidine. CD163 and CD138 were done with Ventana Benchmark autostainer.

The IHC studies were interpreted by histopathologic evaluation by the primary author (J.W.) with input from surgical pathologists (M.v.R. and R.B.W.) using previously published scoring criteria (5, 16).

The stromal signature score was determined by averaging the scores of the evaluable markers. Because of core loss from TMA sections, not all stains were evaluable for each tumor set. If fewer than 4 of the 5 markers were evaluable from the DTF series, or only 3 (or fewer) of the 4 markers were evaluable from the CSF1 series, the signature score was not included in the paired analysis. The stromal signature scores were calculated by averaging the scores for the evaluable markers. A total of 17 cases of paired breast carcinomas were analyzed for concordance in stromal signatures. To assess for concordance in patient matched samples, a random permutation analysis was used to determine statistical significance. Cases were divided into random pairs and the average absolute difference in stromal signature score for each set of random pairs was computed. 200,000 iterations of this procedure were performed and P values were estimated by computing the proportion of iterations resulting in less average absolute difference than was observed in the actual paired data. The entire data set containing 52 cases (26 pairs) was used to evaluate associations of histologic subtype (ductal vs. lobular) and tumor location (right vs. left) between the 2 breast primaries. For the breast scar TMA (TA242), to analyze conservation of signatures within each patient compared to other patients, the mean standard deviation across normal, scar, and cancer observed in the patients is taken and then compared to the mean standard deviation following permutation of the samples. A total of 150 iterations were performed. The P value is computed as the proportion of permuted mean standard deviations that are less or equal to the true mean standard deviation.

In addition, matched patient data were analyzed for intrapatient variations in stromal expression among normal, scar, and invasive carcinoma. To compare differences among normal, scar, and cancer stromal signatures from paired sets of samples, a Friedman test was performed. A total of 6 complete sets were analyzed for DTF and 8 complete sets were analyzed for CSF1. To examine differences between subcategories (i.e. normal stroma versus biopsy site, biopsy site versus carcinoma, normal stroma versus carcinoma) across all samples a Wilcoxon signed rank sum test was used. Comparisons were performed for DTF response profile between normal stroma versus biopsy site stroma (11 sets) and between normal stroma versus stroma surrounding carcinoma (6 sets). Similarly, comparisons were performed for CSF1 response profile between normal stroma and biopsy site stroma (12 sets) and between normal stroma and stroma surrounding carcinoma (8 sets). Finally, Spearman correlation between DTF and CSF1 marker expression and days following biopsy was computed to assess for an association of marker expression with days following biopsy.

To study the overall pattern of CSF1 and DTF marker expression in the context of breast scar tissue, expression profiling with gene microarrays were performed using material taken from archival material from a separate set of 16 patients seen at Stanford University Medical Center. Total RNA was isolated by using the RecoverAll Total Nucleic Acid isolation kit (Ambion Inc., Catalog no. 1975) according to the manufacturer's protocol with some modifications. The paraffin-embedded tissue blocks were cut with a disposable microtome blade into 5–6 × 20 μm sections; and placed in RNAse-free Eppendorf tubes. The tissues were deparaffinized by 100% xylene and heat at 50C for 3 minutes and washed by 100% ETOH. After deparaffinization and centrifugation, the pellets were dried with Speed-vac for 5–10 minutes. Protease digestion was carried out by incubation at 50°C for 3 hours and 85C for 2 minutes with gentle agitation until complete digestion. DNase treatment and final purification were performed using a filter cartridge. The total RNA was eluted in 60 μL of RNase-free water, quantified, aliquoted, and stored at −80°C until use. Total RNA was amplified by using the Sense AMP RNA amplification kit (Genisphere Inc.). Two hundred and fifty nanograms of total RNA were heated at 80ºC for 10 minutes with dT and random primers. The Genisphere SenseAmp linear mRNA amplification method produces sense-strand amplified mRNA by incorporating a double stranded T7 promoter into the 3′ end of the first strand cDNA, driving transcription of an amplified RNA with the same strandedness as mRNA. RNA amplification was carried out according to the manufacturers' instructions. The amplified Sense RNA (aRNA) was purified using the Rneasy minElute Cleanup Kit (Qiagen) and the eluate was collected in 15 μL RNase-free water. Amplified Sense RNA purity was determined by using a spectrophotometer to evaluate the A260/280 ratio. The fold-amplifications were evaluated by assuming that mRNA constitutes 1% to 5% of a total RNA population and only 20% of Sense RNA obtained from amplification was estimated to be true mRNA, whereas the rest could be amplified ribosomal RNA (http://www.genisphere.com/rna_amp_faqs.html and Sense AMP RNA amplification kit data sheet). The Human Exonic Evidence Based Oligonucleotide microarrays (HEEBO, Stanford) which contain 44,544 70-mer probes that were designed using a transcriptome-based annotation of exonic structure for genomic loci were used for expression profiling. The 5 to 10 μg of amplified RNA was reverse-transcribed into cDNA using a mixture of oligo dT and random hexamer primers with incorporation of amino allyl-dUTP. Cy3 and Cy5 dyes were used for indirect labeling of the cDNA from amplified reference RNA (universal human reference RNA) and cDNA from amplified RNA of FFPE specimens. Microarray hybridization and washing was performed using standard procedures. Microarrays were scanned on a GenePix 4000 microarray scanner (Molecular Devices).

In prior studies, we found that subsets of breast cancers are associated with discrete stromal gene signatures (5, 6, 13). To gain insight as to whether these signatures are initiated by the individual tumors or are specific to the host, we looked for intrapatient variation of both the DTF fibroblast and the CSF1 macrophage response signatures in 26 pairs of breast cancer samples from patients with dual breast primaries. The 52 breast cancer samples were assessed for these 2 signatures by immunohistochemistry using previously defined sets of 5 biomarkers for the DTF fibroblast signature (cadherin 11, MMP 11, CSPG2, CD138, and osteonectin) and 4 biomarkers for the CSF1 macrophage response signature (cathepsin L, CD163, CD16, CD32; refs. 6, 13, and 16).

The mean level of expression of DTF fibroblast response markers showed significantly less intrapatient discordance, that is more similarity, in matched primaries compared with the discordance observed in pairs of tumors from different patients (mean discordance = 0.78 in matched pairs vs. 1.21 in random pairs; P = 0.007, Table 2). In contrast, the expression of CSF1 macrophage response markers did not show significantly more agreement in matched pairs from a single patient compared with unmatched pairs from separate patients (mean discordance = 0.71 in matched pairs vs. 0.80 in random pairs; P = 0.22, Table 2). When examined individually, all DTF markers but only 2 of the 4 CSF1 markers show a trend for more concordance in matched primaries compared with random pairs (all P < 0.10, Table 2).

We also sought to assess the conservation of a set of additional biomarkers (ER, PR, HER2, Ki-67) and tumor characteristics (histologic subtype, tumor grade, size, laterality) in paired breast primaries. ER, PR and HER2 could not be examined due to the lack of variation among the specimens. For the rest of the markers and tumor characteristics evaluated, only Ki-67 showed significant correlation between the 2 primaries (P = 0.014, Table 2). There was no difference between the amount of stromal component in the breast primaries.

To further characterize these stromal signatures in the overall setting of tissue injury response, we compared CSF1 macrophage and DTF fibroblast response patterns in the setting of normal breast, prior biopsy site, and invasive carcinoma by gene expression profiling. Cluster analysis for both the CSF1 macrophage and the DTF fibroblast core gene lists showed variation of signature profiles among patients in biopsy site stroma (Fig. 1). This finding suggests that the CSF1 macrophage and DTF fibroblast response patterns show variable patterns of expression in biopsy site stroma, a process that we have previously characterized in breast cancer stroma (5, 6, 13).

To analyze these variations in detail, a tissue microarray of normal, biopsy site, and carcinoma-related stroma was constructed and analyzed with immunohistochemistry using the CSF1 macrophage and DTF fibroblast markers (Fig. 2). The DTF fibroblast signature showed more concordance across normal, cancer, and biopsy site samples from within a patient, than across normal, cancer, and biopsy site samples from a random group of patients (P = 0.04, Table 3). In contrast, the CSF1 macrophage signature did not show a significant concordance in lesions from a single patient compared to lesions from different patients (P = 0.32). This result is consistent with the synchronous breast cancer portion of this study, where patients mounted consistent DTF fibroblast responses to their dual breast primaries.

The previous analysis determined the degree to which the stromal response was maintained within a patient across several tissue types (normal, cancer, biopsy site). We then determined the association of stromal response with sample type (normal, biopsy site, and carcinoma) using Friedman's test to detect nonparametric differences across the 3 sample types from matched patient samples. There was no significant difference across all 3 categories in CSF1 macrophage response marker expression (P = 0.30), whereas DTF fibroblast response markers approached significance (P = 0.083). We next performed a Wilcoxon signed rank test to assess for an association between marker expression and tissue type across all patients. In contrast to Friedman's test, the Wilcoxon test was not limited to patients with matched data from multiple tissue types. In the DTF signature group, there was significantly increased expression in biopsy site change versus normal (P = 0.0075), and a trend for increased expression in carcinoma versus normal (P = 0.0892), with no significant difference between biopsy site change and carcinoma (P = 0.1681). Therefore, the stromal expression of DTF markers varied most significantly when normal tissue is disrupted (normal vs. biopsy site changes, normal vs. carcinoma), and showed little difference between biopsy site changes and carcinoma. The CSF1 macrophage response showed a trend toward increased expression in biopsy site changes versus normal (P = 0.0646), and no significant difference between biopsy site changes versus carcinoma (P = 0.7974) or normal versus carcinoma (P = 0.3829). These data suggest that the CSF1 response is not generated as a generalized response to tissue disruption.

To assess temporal alterations in stromal marker expression, Spearman correlation was computed between DTF/CSF1 scores and number of days following initial core biopsy (Table 5). Of the individual markers, only MMP11 showed a marginally significant correlation of expression level with days following biopsy (Rho = 0.63; P = 0.03, uncorrected for multiple hypothesis testing). There was no significant association observed between time following biopsy and marker expression for any of the other individual markers, the CSF1 or DTF stromal signatures, or the markers averaged together (all P > 0.05).

We have previously identified 2 stromal signatures that are present in a subset of breast cancers that correspond to clinical–pathologic features: the DTF fibroblast response signature, associated with lower tumor grade and increased survival (6, 13), and the CSF1 macrophage response signature, associated with decreased survival among low-grade tumors (5). We have found that these signatures are also present in preinvasive cancer, that is ductal carcinoma in situ (16). A number of recent studies have identified stromal signatures that predict response to chemotherapy (15–17). We attempted to further understand the biology of stromal response by asking the following question: is the stromal response unique to the tumor, or is the stromal response an inherent reaction of the host to tissue alteration?

We approached this question by studying the variability in the stromal responses in clinically distinct tumors from the same patient. If the stromal response is dependent primarily on the tumor, we would not expect significant stromal response concordance between 2 separate tumors from the same patient. Therefore, in the first part of our study, we compared stromal responses to synchronous breast primaries in the same patient to determine whether stromal response to cancer is patient-dependent or tumor-provoked. If the stromal response is primarily dependent on the host, we would see a significant concordance in stromal responses between the 2 cancers from the same patient. In contrast, if the stromal response is largely dependent on the inciting tumor, we would not expect to see significant concordance in stromal responses from cancers derived from the same patient. In the second part of our study, we proceeded to ask whether the stromal response is a generalized response to tissue injury, or a specific response generated in reaction to the inciting tumor. We addressed this question by examining stromal signatures in normal breast, biopsy site changes, and carcinoma. If the stromal response is a tumor-specific reaction, we would anticipate significant differences in stromal signature response between biopsy site changes and carcinoma. If the stromal response is a more generalized response to tissue injury, we would anticipate significant differences in stromal signature response between normal tissue and biopsy site changes, but negligible differences between biopsy site changes and carcinoma. We then studied stromal response to biopsy site changes and carcinoma to delineate tumor-specific response from a more generalized tissue injury response. In doing so, we found that the 2 stromal signatures have divergent associations.

Our studies indicate that the DTF fibroblast response signature is significantly conserved in patients with paired breast primaries. In examining stroma taken from normal breast and biopsy site changes, we find that the response to biopsy site changes is similar to the response to carcinoma, and conclude that DTF fibroblast response is a more generalized response to tissue disruption and injury that appears to be specific to individual patients. This suggests that the DTF fibroblast response is a host-specific response and that the ability to generate this response varies between patients.

In contrast to the DTF fibroblast response, which showed significant concordance within a patient, the CSF1 macrophage response showed no significant concordance within patients with paired breast primaries. However, we cannot exclude that in a larger patient cohort, the CSF1 macrophage response signature may or may not show significant correlation as a couple of the markers within the signature show correlation within a patient. Because the response is increased in association with higher tumor grade, it may be that the CSF1 macrophage response is at least in part invoked by tumor.

These findings are in keeping with a separate study in our group on colon and breast cancer metastases, which demonstrated that DTF stromal expression is dependent on the tissue type (lymph node vs. colon), whereas CSF1 stromal expression is conserved within the same tumor regardless of tissue background (17). This finding complements our conclusions from studying breast cancer, in that DTF appears to be not only patient specific, but tissue specific, whereas CSF1 response may be dependent on the individual tumor characteristics.

We have demonstrated 2 divergent stromal responses to breast cancer by studying marker expression in patients with synchronous breast primaries. A potential confounding factor would be if a patient developed 2 biologically similar tumors that would then induce the same stromal response. To evaluate whether epithelial markers were more conserved in paired primaries than would be expected by chance, we analyzed the paired breast primaries for conventional tumor biomarkers (ER, PR, HER2, and Ki-67) and tumor characteristics (histologic subtype, tumor grade, size, and laterality). Of these features, the only one that attained statistical significance for concordance in paired primaries was Ki67 (P = 0.014), suggesting that proliferative index may be conserved in dual primaries. Because of the relatively small sample size and small amount of total variation in ER, PR and HER2 across all the samples, our study was underpowered for identifying statistically significant conservation of these markers among paired primaries. We did not find any evidence of significant conservation of these makers, but we note that this hypothesis needs to be evaluated in a larger study with higher proportions of hormone receptor negative and HER2 positive patients. Taken together, our findings are consistent with evidence in the literature suggesting that dual breast primaries show distinct patterns of molecular alterations, based on molecular studies (18, 19), histologic subtype, and hormonal receptor status (20–22), and by cytogenetics (23).

The interaction of cancer and stroma is reminiscent of the seed and soil theory proposed by Stephen Paget in 1889 to explain the preferential metastases of breast cancer to sites such as bone (24). In the case of cancer treatment, therapy has predominantly relied on attacking the “seed,” that is the epithelial component. Study of the “soil,” or the stroma, allows for understanding of the tumor microenvironment and opens up possibilities for a different direction of therapy. The study of the milieu in which breast cancer is situated is challenging, as the stromal response may be a product of tumor provocation as well as the patient's innate healing mechanism. In this paper, we have attempted a better understanding of this “soil,” in hopes that knowledge of the stromal response can in the future lead to novel treatments which create less fertile grounds for tumor growth.

No potential conflicts of interest were disclosed.

Supported by the National Institutes of Health (CA129927) and the California Breast Cancer Research Program (15NB-0156).

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.