High IRF8 expression correlates with CD8 T cell infiltration and is a predictive biomarker of therapy response in ER-negative breast cancer

To examine IRF8 expression we performed immunohistochemistry on a BC commercial tissue microarray (TMA, US Biomax, USA, breast cancer molecular subtype was not provided). TMA was deparaffinized in xylene and rehydrated using graded alcohols, and antigen retrieval was performed using citrate buffer (30 min, 95 °C), followed by blocking and subsequent incubation with IRF8 antibody 1/1000 (Rabbit Anti-IRF8 monoclonal antibody [EPR20441] #ab207418, Abcam). The UltraView Universal diaminobenzidina tetrachloride (DAB) detection kit was used to detect protein expression. Slides were counterstained with Hematoxylin. IRF8 staining patterns were evaluated and scored based on intensity and percentage of positive cells as previously described [18]. The score is designated as 0 when no tumor cells stain, 1+ when 10–20% of cells stain (weak), 2+ when 20–50% of cells stain (moderate), and 3+ when > 50% of cells stain (strong). The immunohistochemistry scoring was performed by two observers that were blinded and the degree of agreement was good. Tumors were then stratified as IRF8 low (< 2 score) or IRF8 high (≥ 2 score).

For analysis of the immune infiltrates, we performed quantitative multiplex immunohistochemistry as previously described [19] on a formalin-fixed and paraffin embedded (FFPE) section of a homemade TMA, generated using ER-negative breast tumor tissue blocks (24 samples). Tumor samples were stained with anti-CD45 antibody and selection of tumor areas was made by choosing the regions with the highest quantity of CD45⁺ cells. Sections were matched to their corresponding paraffin blocks (donor blocks), and two tumor cores with a diameter of 2 mm were punched from tumor regions of each donor block and precisely arrayed into a new recipient paraffin block (TMA block) using the Galileo Tissue MicroArrayer CK 4500 (Transgenomic). Five-micrometer-thick slices were cut from the TMA FFPE blocks and transferred to glass histology slides which were processed, stained, and analyzed as previously described [19]. Briefly, chromogenic sequential IHC was conducted with 5 mm of FFPE tissue sections. Following deparaffinization, slides were stained by hematoxylin (S3301, Dako) for 1 min, followed by whole tissue scanning using Aperio ImageScope AT (Leica Biosystems). Slides were subjected to endogenous peroxidase blocking followed by heat-mediated antigen retrieval. Then, sequential IHC consisting of iterative cycles of staining, scanning, and antibody and chromogen stripping was performed as described in Tsujikawa et al. [19]. The digital image workflow encompasses three steps: image preprocessing, visualization, and quantitative image analysis. In image preprocessing, iteratively digitized images were coregistered so that cell features overlap down to a single-pixel level, using Matlab software. Visualization was performed via conversion of coregistered images to pseudo-colored single-marker images in ImageJ v.1.48 [20] and ImageScope (Leica Biosystems). In quantitative image assessment, single cell segmentation and quantification of staining intensity were performed using a CellProfiler v.2.1.1 pipeline, CellID_FlowCyt-6.9.15. cpproj (available under GPLv2 at https://​github.​com/​multiplexIHC/​cppipe). All pixel intensity and shape-size measurements were saved to a file format compatible with flow and image cytometry data analysis software, FCS Express 5 Image Cytometry v.5.01.0029 (De Novo Software). Expression of IRF8 was also corroborated in the custom TMA and tumors were stratified as IRF8 low (< 2 score) or IRF8 high (≥ 2 score). Tonsil tissue was used as positive control.

MDA-MD231 (MDA231) and MCF7 cells were grown in DMEM containing 10% heat-inactivated FBS (Biowest), 100 IU/ml penicillin, 100 μg/ml streptomycin, 2.0 mM GlutaMAX (all from Thermo Fisher Scientific). All cell lines were tested as mycoplasma-negative by PCR.

Genomic DNA was purified using DNeasy Tissue kit (Qiagen) according to the manufacturer’s instructions. Sodium bisulfite treatment of genomic DNA was carried out using EZ DNA Methylation-Lightning Kit (Zymo Research) according to the manufacturer’s instructions. MS-PCR was carried out as previously described [10].

The MethHC database tool (http://​methhc.​mbc.​nctu.​edu.​tw/​) [21] was used to analyze the correlation of IRF8 methylation and gene expression in breast tumor samples compared to normal breast tissue samples. The relationships between IRF8 expression and BC patient overall survival and relapse free survival were analyzed using the Kaplan-Meier plotter data base (http://​kmplot.​com/​analysis/​) [22]. To validate IRF8 as predictive biomarker, we used the ROCplot website (http://​www.​rocplot.​org/​) [23]. The patients were assigned to two cohorts (responder and nonresponder) based on their clinical characteristics. Patients with neoadjuvant chemotherapy were classified according to pathological response as published by the authors. In this, instead of four cohorts (progressive disease, stable disease, partial response and complete response), all patients were assigned into two cohorts, including those where no residual histological evidence of the tumor remains after chemotherapy (responders) and all other patients with residual tumor tissue (nonresponders) [23]. Patient samples were grouped based on the expression of IRF8 (Jetset best probeset) using the median cutoff value.

The results reported here are partially based on data generated by TCGA Research Network: https://​www.​cancer.​gov/​tcga. BC expression data was downloaded as RNA-seq rsem raw counts and TMM (https://​genomebiology.​biomedcentral.​com/​articles/​10.​1186/​gb-2010-11-3-r25) normalized. BC survival data was truncated at 10 years follow-up time. Survival analyses cutoff values were roughly optimized by studying quantile cutoffs from 0.2 to 0.8 by 0.1 and choosing the cutoff with lowest p value. MIXTURE deconvolution method was used to estimate tumor immune infiltration from expression data. To study CD8⁺ T cells abundance in relation to IRF8 expression, IRF8 expression was categorized as high or low respective to the median IRF8 expression value. Isobaric tags for relative and absolute quantification (iTRAQ)-based mass spectrometry data from TCGA was used to interrogate protein expression in primary BC samples.

Data handling, analysis, and graphic representation (all shown as mean ± SEM) were performed using Prism 7.0 (GraphPad Software), R statistical software (https://​www.​r-bloggers.​com/​its-easy-to-cite-and-reference-r/​) and Survival package (https://​cran.​r-project.​org/​web/​packages/​survival/​citation.​html). For multiple comparison, two-way analyses of variance (ANOVA) with Sidak’s post-test were performed. For the comparison between two groups, two-tailed unpaired Student T test was performed. A p < 0.05 was considered statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001). Comparisons for the abundance of TCD8⁺ cells between IRF8-high and IRF8-low groups were performed with Wilcoxon signed-rank test.

In order to investigate whether loss of IRF8 expression in BC is an important factor in disease progression and metastasis, we evaluated its expression by immunodetection of the protein in invasive ductal BC of various grades and sentinel lymph node metastases. Although information regarding the ER status of these samples was not available, our own analysis using Kaplan-Meier plotter data base show that most ER-negative tumors fall also into the more aggressive grade 3 phenotype (see Fig. 3a).

Even if IRF8 is mainly expressed in the nuclei of tumor cells, weak IRF8 expression in cytoplasm was also detectable (Fig. 1a). Interestingly, IRF8 expression was robust in grade 1 and 2 tumors (G1 and G2, well- or moderately differentiated respectively) (Fig. 1a, b), showing a score ≥ 2+ in 73% of the samples analyzed (Fig. 1c). In contrast, only 34% of the grade 3 (G3) or poorly differentiated tumor samples exhibited high IRF8 expression (Fig. 1b, c). Furthermore, expression of IRF8 was not detected in metastatic foci in sentinel nodes (Fig. 1b, c). Also, the percentage of samples with high IRF8 expression diminished significantly as the disease worsens (p < 0.0001) and none of the invaded sentinel nodes evidenced expression of IRF8 (Fig. 1c).

Hypermethylation of IRF8 promoter has been reported to underlie IRF8 silencing or downregulation in multiple cancers, including BC [7, 9‐11]. We therefore determined IRF8 promoter methylation status by a methylation-specific PCR test in the BC cell lines MCF7 and MDA231, which have low and high metastatic potential, respectively. As expected, the IRF8 promoter was readily methylated in MDA231 cells, accompanied by no expression of IRF8 (Fig. 2a, b), whereas the less aggressive MCF7 cell line expressed IRF8 and demonstrated simultaneous unmethylated and methylated promoter alleles, potentially indicating absence of homozygous silencing of IRF8 promoter in these cells. These results are in concordance with those observed by immunostaining where IRF8 expression is absent in MDA231 cells [10] and in BC samples with more aggressive phenotype [17]. To further verify the correlation of IRF8 methylation and its expression in BC samples, we took advantage of the MethHC database tool [20]. This exercise revealed that IRF8 promoter methylation correlates inversely with its gene expression (Fig. 2c), indicating that epigenetic changes are a major component of IRF8 downregulation. Altogether, our data indicate a potential prognostic value for IRF8 in BC progression, justifying a deeper characterization of its expression in different BC subtypes.

To determine whether IRF8 could act as a biomarker that accurately stratifies patients for prognosis and potential response to therapies, we investigated whether tumor IRF8 expression correlated with improved outcome in distinct BC molecular subtypes. The online available database Kaplan–Meier plotter [22] was used to identify the relationship between IRF8 expression and overall survival or relapse-free survival. High expression of IRF8 was significantly associated with a longer overall and relapse-free survival in BC, but only in the ER-negative molecular subtypes, HER2+ and TNBC (Fig. 3a, b). Interestingly, most ER-negative subtypes are also grade 3 tumors. Indeed, according to KMplotter data base only 30% of the patients have ER+ grade 3 tumors in contrast to nearly 80% of the ER-negative patients which show grade 3 tumors. This has also been reported by Putti et al., 2005 [24]. Moreover, IRF8 expression predicts a better outcome only in the subgroup of ER-negative patients with grade 3 and 2 tumors (Fig. 3c), suggesting that IRF8 expression could more accurately discriminate the prognosis of ER-negative patients than the mere be histological classification.

To date, expression of the hormone receptor ER stratified patients for use of aromatase inhibitors or anti-estrogen therapy (tamoxifen), whereas expression of HER2+ indicates potential usefulness of trastuzumab. Chemotherapy is widely used as a combination of available choices according to the subtype of BC and clinical staging. The combination of fluorouracil, adriamycin, and cyclophosphamide (FAC) is a chemotherapy regimen sometimes given for localized BC with a relatively high risk for recurrence, whereas cyclophosphamide, methotrexate, and fluorouracil (CMF) is often used for earlier-stage BC that has not spread beyond the breast or lymph nodes. Thus, we determined if there was an association between IRF8 expression and the pathological complete response to these different therapeutic regimes.

As anticipated, there was no significant association between IRF8 expression levels and response to endocrine therapy in patients with ER+ BC (luminal cancers). In contrast, in ER-negative BC (TNBC or HER2+), high expression levels of IRF8 was significantly associated with complete pathological response in patients treated with FAC (p = 0.0003) or trastuzumab (p = 0.027) but not CMF (p = 0.746) (Fig. 4). Interestingly, in TNBC, a higher expression of IRF8 was observed in patients who respond to anthracycline therapy, although not statistically different (p = 0.055). ROC curves confirmed the value of IRF8 as a prognostic factor of therapeutic response (Additional file 1: Figure S1).

In sum, these analyses indicate that IRF8 expression should be considered a prognostic biomarker in ER-negative BC where its expression also predicts complete pathological response to certain therapeutic regimes used for patients harboring these BC subtypes.

Since high IRF8 expression levels are correlated with improved outcome in HER2+ and TNBC subtypes, we evaluated whether tumor cell IRF8 expression correlated with the complexity or contexture of tumor-infiltrating immune cells. To this end, we took advantage of a homemade TMA consisting of 24 ER-negative BC specimens, classified according to either low or high IRF8 cancer cell expression as determined by traditional immunohistochemistry. On the TMA, we performed sequential immunostaining of FFPE slides with two panels of 12 antibodies reporting leukocyte lineage for quantitative evaluation of lymphoid and myeloid lineage cells. Infiltrates were quantified by evaluating the chromogenic intensity of each cell using single-cell segmentation algorithms and image cytometry-based cell population analyses, an analytic process described previously [18]. A lineage gating strategy was used to identify each cell subpopulation (Fig. 5a, b). Chromogenic signals for lymphoid and myeloid biomarkers were then pseudo-colored and visualized in the tissue context (Fig. 5c, d). This analysis revealed a significant increase in the frequency of CD3⁺CD8⁺ T cells in tumors that also exhibited high neoplastic cell IRF8 expression (p < 0.05), whereas no differences were observed in the percentages of other lymphoid populations including Th0, Th1, Th2, Th17 T cells, regulatory T cells (Treg), CD20⁺ B cells, or natural killer (NK) cells (Fig. 6a, b). Interestingly, when we evaluated the differentiation and/or functional status of CD3⁺CD8⁺T cells, a higher percentage of these in tumors co-expressed granzyme B (GrzmB) correlating with high expression of IRF8 as compared to those with low expression of IRF8 (Fig. 6c), although it was not significantly different. In addition, the same tendency was seen in CD8⁺GrzmB⁺ cells, which also exhibited a more proliferative phenotype (Ki67-positivity) in tumors also expressing high levels of IRF8 (Fig. 6d), revealing that higher CD8⁺T cell infiltration was associated with a more activated phenotype. These data were also corroborated using iTRAQ-based mass spectrometry data from TCGA; this analysis allowed interrogation of protein expression in primary BC samples obtained from TCGA. Indeed, we determined a significant positive correlation between IRF8 protein expression and molecules associated with antitumor immune responses (e.g., CD8A, GZMB, PRF1) in ER-negative tumors (Fig. 6e). Moreover, when we used TCGA data combined with MIXTURE, an immune tumor microenvironment estimation method based on gene expression data, we validated again that IRF8 expression was associated with a relatively more abundant CD8⁺ T cell infiltration in ER-negative, HER2+, and TNBC subtypes (Additional file 2: Figure S2).

With regard to myeloid cell infiltration of tumors, we did not observe significant differences in myeloid subsets correlating with IRF8 status (Fig. 7a), although a higher percentage of HLA-DR-expressing CD45⁺ cells were present in tumors with higher expression of IRF8 (p < 0.05) (Fig. 7b). Again, this result was corroborated using iTRAQ analysis, where we observed a positive correlation between IRF8 and HLA-DR expression (Fig. 7c). Interestingly, a non-significant decrease in immature dendritic cells, identified by the presence of the marker DC-sign, and IL-10-expressing macrophages (known to inhibit DC maturation via IL-10-dependent mechanisms [25]), was observed within tumors with high expression of IRF8 (Fig. 7d, e). Altogether, these results indicate that the presence of IRF8 in ER-negative BC is associated with an immune infiltration consistent with a more robust antitumor immune response.