Prognostic Significance of Interferon-γ and Its Signaling Pathway in Early Breast Cancer Depends on the Molecular Subtypes

Interferons are crucial for adaptive immunity and play an important role in the immune landscape of breast cancer. Using microarray-based gene expression analysis, we examined the subtype-specific prognostic significance of interferon-γ (IFN-γ) as a single gene as well as an IFN-γ signature covering the signaling pathway in 461 breast cancer patients. Prognostic significance of IFN-γ, as well as the IFN-γ signature for metastasis-free survival (MFS), were examined using Kaplan–Meier as well as univariate and multivariate Cox regression analyses in the whole cohort and in different molecular subtypes. The independent prognostic significance of IFN-γ as a single gene was limited to basal-like breast cancer (hazard ratio (HR) 2.779, 95% confidence interval (95% CI) 1.117–6.919, p = 0.028). In contrast, the IFN-γ-associated gene signature was an independent prognostic factor in the whole cohort (HR 2.287, 95% CI 1.410–3.633, p < 0.001) as well as in the basal-like (HR 3.458, 95% CI 1.154–10.359, p = 0.027) and luminal B (HR 2.690, 95% CI 1.416–5.112, p = 0.003) molecular subtypes. These results underline the subtype-dependent prognostic influence of the immune system in early breast cancer.


Introduction
The immune system plays a decisive, but also ambivalent role in tumorigenesis as well as in tumor elimination of solid tumors [1]. Immune infiltrates, which are composed of different types of immune cells, are known to have a prognostic and predictive impact in breast cancer [2][3][4]. Dependent on the predominantly existing immune cell type and the microenvironment, this immune reaction can result in tumor rejection, but also in tumor progression [5]. The dynamic interaction between the immune To obtain a complete picture of the prognostic role of IFN-γ in the immune landscape of breast cancer, we defined an IFN-γ signature that includes the abovementioned genes. This prompted us to evaluate the prognostic significance of IFN-γ and an IFN-γ-associated gene signature in a well-characterized cohort of 461 patients with early breast cancer.

Expression of IFN-γ and the IFN-γ Signature in Different Molecular Subtypes
In a cohort of 461 breast cancer patients, we evaluated the prognostic significance of IFN-γ as a single gene as well as an IFN-γ gene expression signature composed of IFN-γ and genes downstream in the IFN-γ pathway: IFN-γ receptor 1, IFN-γ receptor 2, the tyrosine kinases JAK1 and JAK2, as well as the transcription factors STAT1 and IRF1 [7] (Table 1). Table 1. Genes and probesets belonging to the IFN-γ signature.

Gene Gene Symbol Probeset
Interferon-γ IFNG 210354_at Interferon-γ receptor 1 IFNGR1 202727_s_at 211676_s_at Interferon-γ receptor 2 IFNGR2 201642_at Interferon regulatory factor 1 IRF1 202531_at Janus kinase 1 JAK1 201648_at Janus kinase 2 JAK2 205841_at 205842_s_at Signal transducer and activator of transcription 1 STAT1 200887_s_at 209969_s_at The boxplot diagrams in Figure 1 illustrate the distribution of gene expression of IFN-γ as a single gene (Figure 1a To obtain a complete picture of the prognostic role of IFN-γ in the immune landscape of breast cancer, we defined an IFN-γ signature that includes the abovementioned genes. This prompted us to evaluate the prognostic significance of IFN-γ and an IFN-γ-associated gene signature in a wellcharacterized cohort of 461 patients with early breast cancer.

Expression of IFN-γ and the IFN-γ Signature in Different Molecular Subtypes
In a cohort of 461 breast cancer patients, we evaluated the prognostic significance of IFN-γ as a single gene as well as an IFN-γ gene expression signature composed of IFN-γ and genes downstream in the IFN-γ pathway: IFN-γ receptor 1, IFN-γ receptor 2, the tyrosine kinases JAK1 and JAK2, as well as the transcription factors STAT1 and IRF1 [7] (Table 1). The boxplot diagrams in Figure 1 illustrate the distribution of gene expression of IFN-γ as a single gene (Figure 1a) and of the IFN-γ gene signature (Figure 1b) in the different molecular subtypes (i.e., luminal A, luminal B, HER2-positive, basal-like). The boxplot diagrams show that the expression of both IFN-γ and the IFN-γ signature is lowest in luminal A and highest in basal-like breast cancer (p < 0.001).

Prognostic Impact of IFN-γ as a Single Gene on MFS
In the whole cohort, IFN-γ had no significant impact on MFS (p = 0.698, log rank; Figure 2). In univariate Cox regression analysis, IFN-γ as a single gene also failed to show a prognostic impact on the whole cohort (HR 1.084, 95% CI: 0.769-1.527, p = 0.646; Table 2); the prognostic impact of IFN-γ as a single gene was confined to the basal-like subtype (p = 0.033, log rank; Figure 3d).    High tumor IFN-γ content was significantly associated with favorable MFS in basal-like breast cancer (HR 2.459, 95% CI 1.040-5.815, p = 0.040), unlike in the rest of the molecular subtypes. Tumor IFN-γ content also had an independent influence on MFS in a multivariate analysis adjusted for age, tumor size, axillary nodal status, and histological grade of differentiation (HR 2.779, 95% CI 1.117-6.919; p = 0.028; Table 3). High tumor IFN-γ content was significantly associated with favorable MFS in basal-like breast cancer (HR 2.459, 95% CI 1.040-5.815, p = 0.040), unlike in the rest of the molecular subtypes. Tumor IFN-γ content also had an independent influence on MFS in a multivariate analysis adjusted for age, tumor size, axillary nodal status, and histological grade of differentiation (HR 2.779, 95% CI 1.117-6.919; p = 0.028; Table 3).

Impact of an IFN-γ Signature on MFS
Kaplan-Meier curves demonstrated the prognostic impact of the IFN-γ signature: higher expression of the IFN-γ signature was associated with a significantly longer MFS in the whole cohort (p = 0.012, log rank; Figure 4).
In subtype analysis, the prognostic effect of the IFN-γ signature was particularly pronounced in the luminal B (p = 0.007, log rank), HER2-positive (p = 0.033, log rank), and basal-like molecular subtype (p = 0.050, log rank), but not in the Luminal A breast cancer samples ( Figure 5).   Table 2).
In subtype analysis, the prognostic effect of the IFN-γ signature was particularly pronounced in the luminal B (p = 0.007, log rank), HER2-positive (p = 0.033, log rank), and basal-like molecular subtype (p = 0.050, log rank), but not in the Luminal A breast cancer samples ( Figure 5).

Impact of an IFN-γ Signature on MFS
Kaplan-Meier curves demonstrated the prognostic impact of the IFN-γ signature: higher expression of the IFN-γ signature was associated with a significantly longer MFS in the whole cohort (p = 0.012, log rank; Figure 4).  Table 2).

Discussion
We showed in this retrospective gene expression study that IFN-γ as a single gene had significant prognostic influence only in basal-like breast cancer. However, an IFN-γ signature covering the IFN-γ pathway had a prognostic impact on the entire cohort of 461 breast cancer patients with long-term follow-up: higher expression of the IFN-γ-signature was associated with better prognosis. Since IFN-γ is more implicated in antitumor activity than the type I interferons, we decided to focus on it. While IFN-γ is certainly an important member in the respective pathway, adding information from other pathway components improves the prognostic power. Apparently, a clinically relevant pathway is better captured by a multicomponent signature than by a single gene alone. Indeed, the IFN-γ signature retained its prognostic impact in multivariate analysis. The prognostic impact of the IFN-γ signature was particularly pronounced in the luminal B, HER2-positive, and basal-like subtypes, suggesting that the IFN-γ pathway plays an important role, particularly in highly proliferating tumors. This is in line with our results of previous gene expression studies evaluating the prognostic impact of different immune-related gene signatures. The B-cell metagene was correlated with improved outcomes in highly proliferating node-negative breast cancer patients regardless of ER or HER2 status [9]. In a further study evaluating CXCL13 mRNA expression in the FinHER trial, CXCL13 as a marker of the humoral immune system was associated with favorable prognosis, particularly in triple-negative breast cancer (TNBC) [11]. One possible explanation for the finding that immune-related gene signatures have prognostic and predictive effects, particularly in the case of strongly proliferating tumors, especially TNBC or basal-like breast cancer, is the occurrence of an increased mutation load and neoepitopes that can induce or enhance an antitumor immune response [18,19]. In fact, TNBC, which is not synonymous with basal-like breast cancer but largely overlaps with it, has a higher level of tumor-infiltrating lymphocytes [2,3]. By triggering protective antitumor immune reactions, e.g., via IFN-γ as effector cytokine, this might lead to an improved prognosis.
Interestingly, Callari et al. described, in a retrospective study using gene expression analyses, a significant impact of an interferon-induced metagene depending on ER and HER2 status. In ER-positive HER2-negative breast cancer, a significant correlation was found between a worse prognosis and higher expression of an IFN-correlated metagene, while in patients with HER2-positive breast cancer, a higher expression of the IFN metagene was associated with a better clinical outcome [20].
Our study has some strengths and limitations. A potential weakness is that our study was retrospective and performed at a single certified breast cancer center. Another limitation is that we have not been able to validate our microarray results with real-time polymerase chain reaction, as we report on patients treated more than 2 or 3 decades ago. Unfortunately, not enough tumor tissue is available for additional investigations. Nevertheless, we have previously reported a good correlation of both methods, which has led to the development of the prognostic breast cancer biomarker EndoPredict™ [21].
However, a major strength of our study is the consecutive inclusion of all patients with (i) an adequate amount of fresh-frozen tissue available for successful DNA microarray analysis with (ii) long-term follow-up and (iii) well-defined adjuvant treatment strategies. In this cohort, we show not only the prognostic significance of IFN-γ as a single gene in basal-like breast cancer but also an independent association of a signature covering the IFN-γ pathway in the entire cohort of 461 breast cancer patients with long-term follow-up. This finding underlines the prognostic role of the IFN-γ pathway in early breast cancer across different molecular subtypes. This favorable prognostic effect was particularly pronounced in rapidly proliferating molecular subtypes like luminal B and basal-like. In contrast, no prognostic effect was found for luminal A breast cancer.
In summary, we show, in a comprehensive overview of interferon-γ in the immune landscape of early breast cancer, the subtype-dependent prognostic role of interferon-γ and its signaling pathway. In particular, the results in the luminal B molecular subtype could pave the way for studies in which immune therapies like immune checkpoint inhibitors or personalized vaccination strategies can be used beyond estrogen-receptor-negative breast cancer.

Patient Characteristics and Tissue Specimens
Briefly, 461 patients with early breast cancer, who received surgery between 1986 to 2000 at the Department of Gynecology and Obstetrics of the University Medical Center Mainz, entered the study (Table 6). We included all consecutive patients with an adequate amount of fresh-frozen tumor tissue available for successful Affymetrix (Santa Clara, CA, USA) microarray analysis. The entire cohort consisted of three subgroups with different systemic treatments: (i) "N0 cohort": 200 node-negative early breast cancer patients with no further adjuvant therapy after surgery and irradiation.
(iii) "chemotherapy cohort": 96 patients treated with either cyclophosphamide, methotrexate, fluorouracil (CMF; n = 34) or epirubicin, cyclophosphamide (EC; n = 62) without endocrine therapy in the adjuvant setting.  The established prognostic factors (histologic grade, tumor size, nodal status, age at diagnosis, ER, PR, HER2, and Ki-67) were collected from the pathology reports and the breast cancer database of our department. For all tumors, samples were snap-frozen and stored at −80 • C. Tumor cell content exceeded 40% in all samples. Approximately 50 mg of frozen breast tumor tissue were crushed in liquid nitrogen. RLT buffer was added, and the homogenate was centrifuged through a QIAshredder column (Qiagen Hilden, Germany). From the eluate, total RNA was isolated with the RNeasy Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. RNA yield was determined by UV absorbance, and RNA quality was assessed by analysis of rRNA band integrity on an Agilent 2100 Bioanalyzer RNA 6000 LabChip kit (Agilent Technologies, Santa Clara, CA, USA), as previously described [9]. The study was approved by the ethical review board of the medical association of Rhineland-Palatinate.
The median age of the patients at diagnosis was 62 years. The mean follow-up time was 12 years; 132 patients (28.6%) developed distant metastases from breast cancer ( Table 6).

Gene Expression Analysis
Fresh-frozen tumors (n = 461) obtained from the Department of Obstetrics and Gynecology of the University Medical Center Mainz were profiled on HG-U133A arrays (Affymetrix, Santa Clara, CA, USA) to quantify the relative transcript abundance in the breast cancer tissue, as previously described [9I. Briefly, starting from 5 µg total RNA, labeled cRNA was prepared using the Roche Microarray cDNA Synthesis, Microarray RNA Target Synthesis (T7), and Microarray Target Purification kits (Roche Applied Science, Mannheim, Germany), according to the manufacturer's instructions. Raw expression data (CEL files) were normalized using frozen robust multiarray analysis (fRMA). In the analysis settings, the global scaling procedure was chosen, which multiplied the output signal intensities of each array to a mean target intensity of 500. Samples with suboptimal average signal intensities (i.e., scaling factors >25) or glyceraldehyde-3-phosphate dehydrogenase 3 /5 ratios >5 were relabeled and rehybridized on new arrays. Throughout the text below, all expression values and respective thresholds from fresh frozen material measured by HG-U133A arrays refer to TGT500 scaling. The majority of the samples had already been deposited at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) under the accession numbers GSE11121 and GSE26971. Since these datasets were submitted in 2008 and 2011, we decided to file the complete record of 461 samples used in the current study with updated follow-up at the NCBI in the GEO database under accession number GSE158309.

Molecular Subtypes
The determination of intrinsic subtypes was performed according to Haibe-Kains and coworkers [22], who postulated a three-gene model with the estrogen receptor gene (ESR1), HER2, and aurora kinase A (AURKA). Briefly, ER and HER2 status were derived from the bimodally distributed mRNA levels of the corresponding genes (probesets: ESR1 205225_at and ERBB2 216836_s_at) based on fRMA normalized expression values. The cut-off for ESR1 was determined by model-based clustering. The cut-off for ERBB2 was selected by the upper quartile plus interquartile range of the mRNA level. For AURKA, the median of the mRNA expression of the corresponding probe set (208079_s_at) was used as a cut-off, as previously described [12].
This procedure resulted in the following molecular subtypes: -ESR1-positive, HER2 negative, low proliferation (AURKA low) → luminal A-like -ESR1-positive, HER2 negative, high proliferation (AURKA high) → luminal B-like -HER2-positive -ESR1 negative, HER2-negative → basal-like Table 6 shows the absolute and relative frequencies of the molecular subtypes (which were determined based on gene expression data) within the investigated cohort of patients.

Statistical Analysis
Statistical analyses were performed using the SPSS statistical software program, version 23.0 (SPSS Inc, Chicago, IL, USA) and R. Patients' characteristics were given in absolute and relative numbers (Table 3). Differences in the expression of IFN-γ and the IFN-γ signature between molecular subtypes were assessed using analysis of variance (ANOVA) and posthoc Tukey's test. The prognostic significance of IFN-γ as a single gene and an IFN-γ signature for MFS was examined by Kaplan-Meier survival analysis (≤median vs. >median) as well as univariate and multivariate Cox regression analysis. This was adjusted for pT stage (T1, -2 vs. T3, -4), histological grade (GI + GII vs. GIII), and immunohistochemically determined ER (negative vs. positive), PR (negative vs. positive), HER2 (negative vs. positive), and Ki-67 (≤20% vs. >20%). The significance of Kaplan-Meier survival analysis was assessed by the p-value of the log-rank test. All p-values are two-sided. As no correction for multiple testing was carried out, these are descriptive measures.

Conflicts of Interest:
The authors declare no conflict of interest. The funder had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.