Characterization of the Immune Microenvironment in Inflammatory Breast Cancer Using Multiplex Immunofluorescence

The immune tumour microenvironment is one of the key factors in the development and progression of various types of cancer including breast cancer (BC) [1]. It is the result of dynamic interactions between cancer cells and the immune system of the host. The importance of the immune cell infiltration in the tumour microenvironment has been shown to halt the metastatic progression and predict responses to various treatment regimens in BC (reviewed in Tower et al. [2]). Several studies demonstrated that the increased accumulation of T-lymphocytes, specifically CD8+ TILs, in triple-negative breast cancer (TNBC) and HER2+ BC positively correlates with a good prognosis and better response to chemotherapy [3‒6]. On the other hand, while a higher number of FOXP3+ TILs were associated with poor overall survival (OS) in the estrogen receptor (ER+) disease, increased accumulation of these cells in TNBC patients correlated with a favourable prognosis [7].

While these earlier studies laid a solid foundation for dissecting a molecular network of interactions underlying the formation of the immune microenvironment in BC, it has become apparent that a more detailed characterization of the spatial distribution of different immune cell subtypes within a tumour relative to one another would help to improve understanding of relevant pathways of intercellular communications and therapeutic options. Multiplex immunofluorescence (mIF) has recently been developed as the method of choice for mapping the immune microenvironment in cancer [8]. Furthermore, mIF was shown to be a reliable method in predicting responses to programmed cell death receptor 1/programmed cell death ligand 1 (PD-L1)-based anticancer therapies [9].

Inflammatory breast cancer (IBC) is a rare and highly aggressive form of BC characterized by a rapid progressive clinical course with a high risk of lymph node involvement and distant metastases. It is responsible for almost 10% of all BC-related deaths [10]. Although neoadjuvant chemotherapy (NACT), surgery, and radiation therapy have improved the survival of IBC patients [11, 12], they still exhibit poorer survival than stage matched non-IBC subjects with a 5-year survival of up to 61% [13‒16]. We and others have explored predictors of pathological response to NACT in BC epithelial cells [17]. However, understanding the complex interaction between the epithelial component and stromal microenvironment is urgently required to address this unmet clinical need.

The immune microenvironment in IBC is poorly characterized. It has been reported that the increased infiltration of lymphocytes, specifically CD8+ TILs, correlated with better patient response to standard IBC NACT regimens [18]. Interestingly, this study also demonstrated that patients characterized by a close proximity of mast cells to CD8+ cells or CD163+ macrophages were less likely to respond to the treatment. In another study, increased accumulation of B cells correlated with a better prognosis [19]. We and others have recently reported that increased infiltration of CD68+ cells correlated with better responses to the treatment [20]. Furthermore, a more pronounced infiltration of CD14+ monocytes/macrophages was identified in post-treatment IBC carcinoma compared with non-IBC tumours [21]. In this study, we used mIF to characterize the expression and spatial distribution of various types of TILs and CD68+ cells in a large, multicentre, well-characterized series of IBC samples and their relation to tumour response to NACT and patient outcome.

Patients diagnosed clinically with IBC as per the published international guidelines [13] were included. All experiments were approved by the West Midlands – Black Country NRES Committee (07/Q2702/24), Ethics Committee of the Juntendo University Hospital, Japan (14–144), and Leeds Research Ethical Committee (06/Q1206/180). Clinical details of the cohort are shown in Table 1.

A large international IBC cohort (n = 93) was used for the simultaneous assessment of five components of the tumour microenvironment including CD68+, CD20+, CD4+, CD8+, and FOXP3+ cells. All patients were clinically diagnosed with IBC following the established guidelines [13]. They had received NACT followed by breast surgery in the period between 1997 and 2015 in seven UK centres and one Japanese institution. The median time between the diagnosis of core biopsy and the surgical resection was 6 months; patients who did not undergo surgery were excluded. The regimens of NACT that were given to all patients included combinations of taxanes, anthracyclines, and/or cyclophosphamide.

Formalin-fixed, paraffin-embedded diagnostic core biopsy blocks were collected from the following UK centres: Queen Elizabeth Hospital Birmingham, Birmingham City Hospital, Heart of England Hospital, Birmingham, the Royal Wolverhampton Hospital, Wolverhampton, the Leeds Teaching Hospitals NHS Trust, Leeds, and Bart’s Cancer Institute, Queen Mary University of London, London and from the Department of Breast Oncology, Juntendo University, Japan. Clinico-pathological data, where available, were collected on both pre-treatment core biopsy samples and residual post-NACT tumours, which included tumour type, grade, ER, progesterone receptor (PR), HER2 status, pathological response to NACT, and patient OS.

The histological grading of the tumours was performed in accordance with the Nottingham grading system. The pathological response was assessed in the surgical specimens using the published criteria [22]. Pathological response was classified into pathological complete response (pCR) (no residual invasive carcinoma with negative axillary nodes; ypTis or ypT0), pathological partial response (residual invasive carcinoma with histological evidence of tumour response), and no response (no histological evidence of tumour response).

All samples underwent multiplex immunostaining. The core biopsies of treatment-naïve patients were stained for CD68, CD20, CD4, CD8 and FOXP3 (relevant antibodies are listed in online suppl. Table 1; see www.karger.com/doi/10.1159/000524549 for all online suppl. material). Individual antibodies were optimized using recommended positive controls as per manufacturer’s recommendation; for mIF, non-BC sections were used as positive controls. Negative controls, in which the primary antibody was omitted had been used in each batch of staining. Epitope retrieval was performed at pH 6.0 (Leica Bond Epitope Retrieval Solution 1 [ER1]) and heated to 100°C for 20 min. Each primary antibody was followed by an HRP-conjugated rabbit anti-mouse secondary antibody (Dako). Following tyramide signal amplification, each subsequent staining cycle was preceded by an antibody complex elution cycle with ER1 at 100°C, followed by a peroxide blocking step (hydrogen peroxide) to minimize quench any HRP that was not eluted.

For multiplex scanning, whole slides were scanned at ×10 with one band captured for each filter cube (DAPI, FITC, CY3, TEXAS RED, and CY5). DAPI was used for tissue detection and autofocus. Exposure times were set according to representative fields on a control tissue section. Single-stained slides were prepared for each OPAL and were scanned as above, including an autofluorescence slide for subsequent spectral library construction in Inform 2.2.1.

Phenochart 1.0.12 was used to manually annotate both relevant intratumoural and stromal fields performed by two pathologists (N.M.B. and A.M.S.). Areas of tissue necrosis, blood vessels, and carcinoma in situ were excluded and regions of interest were selected to cover the whole tissue cores for each case as recommended by the International Immuno-Oncology Biomarker Working Group on BC [23]. Annotations representing the field of view on the Vectra 3.0 (682 μm × 510 μm) were placed to cover the tissue section. Slides were rescanned to image the annotations using ×20 magnification and multiple emission wave lengths in each filter cube (37 planes in total), referred to as multispectral images.

Scanned images were digitally analysed by a pathologist (N.M.B.), overseen by a specialist breast pathologist (A.M.S.) using Inform 2.2.1 (Akoya Biosciences). Images were subjected to an analysis pipeline, which included correction for autofluorescence and spectral deconvolution, automated tissue segmentation, cell segmentation (membrane and nuclear), and cell phenotyping (Fig. 1). Spectral deconvolution was employed to reduce the 37 imaged planes to 7 planes representing the signal belonging to each of the 6 fluorophores and the tissue autofluorescence.

Inform employs an algorithm which requires additional training to fit to each dataset. For tissue category segmentation, the software was trained to divide the IBC tissue into three regions of tumour, stromal, and empty categories. Cell segmentation was performed using the aid of membrane and nuclear stains. Subsequently, cell phenotyping was applied. A minimum of five cells per cell phenotype was chosen to train the classifier to differentiate between various immune phenotypes based on the thresholds associated with the various combination of antibodies. All phenotypes included DAPI. Except for FOXP3+ cells, which were classified based on both CD4 and FOXP3 positivity, all other phenotypes were named after the single marker they utilized.

These algorithms were trained using a representative subset of fields. Finally, all slides were processed with the trained protocol and all data for individual patients were exported (data files and images) into individual directories.

For cell density analysis, cell segmentation summary data were used and tabulated in Excel. Statistical analysis was done using the IBM SPSS package (version 26). Analysis for pre- and post-treatment categorical variables including tumour type, grade, ER/PR Allred score, and HER2 expression was done using χ² test. Receptor status was also dichotomized into negative and positive using a cut-off value of Allred score 2/8 for ER/PR and 3+ or 2+ fluorescence in situ hybridization positive for HER2 to define positivity. Cases were categorized into ER-positive BC (ER+, PR+/−, HER2−), TNBC (ER−, PR−, HER2−), and HER2-positive BC (ER+/−, PR+/−, HER2+).

Mann-Whitney and Kruskal-Wallis tests were used to compare phenotype densities between IBC patients in terms of clinical and pathological data. Two-tailed p values <0.05 were considered statistically significant. The Kaplan-Meier method and multivariate cox regression were used for survival analysis. OS was calculated as the duration in months between the date of diagnosis and the date of last follow-up or death. Comparing the pairwise infiltration, Spearman correlation coefficients from all possible immune cell combinations were used to assess various immune cell co-accumulations.

A total of 93 IBC patients were included in this study. The pre-treatment clinico-pathological characteristics of the patients are summarized in Table 1. On pre-treatment biopsies, 50.5% of the tumours were grade 3 differentiation. Of all analysed IBC samples, 36.6% of tumours were ER positive, 28.8% of the cases were triple negative, and 37.6% were HER2+ BC. pCR was achieved in 17.2% patients. Overall patient survival ranged from 9 to 138 months with a median of 34 months.

pCR correlated with better OS in the whole cohort (p = 0.02). HER2+ patients had better OS (p = 0.003) compared to those of other molecular types with TNBC patients having the worst prognosis (online suppl. Fig. 1). Multivariate Cox regression analysis revealed that response to therapy was the only independent factor affecting patient OS (p = 0.01).

To assess the composition and spatial organization of the immune microenvironment in IBC in detail, we carried out quantitative mIF staining using a panel of five antibodies to detect CD68+, CD20+, CD4+, CD8+, and FOXP3+ cells. Figure 1 shows a representative example of the steps of mIF analysis of an IBC sample. For each sample, we performed a separate quantitative analysis of the immune cell accumulation in the tumour and in stroma. Of the five types of immune cells analysed in this study, CD68+ cells were the most abundant in IBC tissues followed by CD4+ cells in all molecular subtypes. Similarly, CD68+ cells were the most abundant immune cells when analysed separately within the epithelial component and in the stroma.

When correlated with tumour grade, increased accumulation of CD68+ cells was observed in the stroma of grade 2 and 3 carcinomas when compared to grade 1 (p = 0.031) tumours (Fig. 2a). Conversely, grade 1 tumours displayed an increased number of intratumoural CD20+ and FOXP3+ cells when compared to grades 2 and 3 cancers (Fig. 2b, c, p = 0.047). A higher number of intratumoural CD4+ cells were observed in grade 2 IBC samples (Fig. 2d, p = 0.04). The accumulation of CD8+ cells either in the tumours or in the peri-tumoural stroma did not correlate with the tumour grade (online suppl. Fig. 2). No significant differences in the number of CD4+ and FOXP3+ cells among the various molecular subtypes in IBC were found (online suppl. Fig. 3). In contrast, although not statistically significant, there was a trend for increased accumulation of CD20+ cells and decreased accumulation of CD68+ cells in HER2+ tumours when compared to the TNBC samples (p = 0.08 and p = 0.075, respectively, Fig. 3a, b). The number of intratumoural CD8+ cells was higher in the HER2-positive tumours when compared to the ER-positive group (p = 0.048, Fig. 3c).

Next, we examined spatial location and co-accumulation of different types of immune cells in IBC tissues. When analysed in relation to the tumour grade, we found co-accumulation of CD20+ cells and FOXP3+ cells in all samples (Table 2). Furthermore, there was a direct correlation between accumulation of CD20+ cells and CD4+ cells in grades 2 and 3 tumours. In grade 3 IBC tumours, a higher number of CD20+ cells correlated with the increased accumulation of both CD8+ cells and CD68+ cells (Table 2). Grade 3 tumours were also characterized by co-accumulation of CD8+ with CD4+ and FOXP3+ cells (Table 2).

When the immune landscape was analysed in relation to the molecular subtypes of BC, the ER+ tumours were characterized by co-accumulation of FOXP3+ cells with CD20+ and CD8+ cells (Table 3). Co-accumulation of CD20+ cells with CD68+, FOXP3+, and CD4+ cells was observed in TNBCs (Table 3). These tumours were also characterized by co-accumulation of CD68+ cells with CD4+ and FOXP3+ cells. Finally, in the HER2+ subtype, we observed co-accumulation of CD8+ cells with CD68+ cells (Table 3).

IBC patients who responded to NACT were characterized by a higher number of CD20+ cells (p = 0.037, Fig. 4a). Better responses were also seen in patients with high CD8+ cell infiltration within the tumour (p = 0.003, Fig. 4f). There was no correlation between patient response to therapy and accumulation of other types of immune cells when the assessment was made for the total number of cells in cancer tissues (Fig. 4b–e) or when intratumoural or stromal populations were evaluated separately.

In a pairwise analysis, we show that co-infiltration of CD20+ cells with CD8+, CD4+, and FOXP3+ cells and co-existence of CD68+ cells with FOXP3+ cells correlated with tumours’ better responses to the treatment (Table 4). Conversely, tumours that were resistant to the treatment were characterized by co-accumulation of CD4+ cells with CD8+, FOXP3+, and CD68+ cells and co-presence of CD68+ cells with CD20+ cells (Table 4).

Tumour microenvironment plays an established role in BC development and progression. While a number of recent studies have extensively characterized the immune microenvironment in different subtypes of non-IBC, the immune composition of IBC tissues remains poorly investigated. Research into IBC has long been hampered by the paucity of large well-characterized patient cohort due to the rarity of the disease and the standard treatment by NACT. Through an international collaborative IBC initiative, we collected a large number of IBC treatment naïve tissue samples for a detailed analysis of the tumour microenvironment.

Recent data from an international group of breast pathologists manually analysing the DCIS tumour microenvironment revealed high interobserver variability [24]. Here, for the first time, we performed mIF staining to profile the immune microenvironment in a large cohort of IBC patients using quantitative digital multiplex profiling. The technique is of particular value in lesions where tissue availability may be limited. It allows simultaneous quantitative analysis of markers of interest in one tissue section. The quantitative objective analysis avoids the drawbacks of manual assessment. Taken together, our findings suggest that the immune microenvironment in IBC and non-IBC patients is likely to be distinct and governed by different mechanisms.

In contrast to previous data on non-IBC patients [25], we observed a clear trend to a decreased accumulation of CD68+ cells in HER2+ IBC patients. Furthermore, while a higher number of macrophages were associated with poor clinico-pathological features such as a higher tumour grade, we found that in contrast to non-IBC patients, this did not correlate with survival of IBC patients.

We found no correlation between the abundance of CD4+, CD8+, and FOXP3+ cells and histological grade in the IBC cohort. These results contrast with published data on non-IBC patients describing an increased accumulation of T cells in higher grade tumours [26]. Furthermore, we show that only HER2+ IBC tumours (but not TNBC as in the case of non-IBC patients) have a higher number of CD8+ cells. High density of CD8+ cells was associated with favourable survival in HER2+-invasive BC in previous studies [27]. Based on our findings and previously published data, increased accumulation of CD4+ cells correlated with better survival in patients with either IBC or non-IBC diagnosis. These data indicate that increased proportions of CD4+ cells could be an important factor in predicting favourable survival of BC patients and assessment of this immunophenotype may be useful in clinical routine practice.

For the purpose of this work, we interpreted staining with anti-CD4 and anti-CD8 antibodies as an indication of accumulation of T-lymphocytes in cancerous tissues. We acknowledge that weak expression of these proteins can be detected on other cell types. However, both markers have been extensively used in similar types of studies to characterize accumulation of T cells in cancerous tissues. It is also important to note that both markers are widely used in the diagnostic clinical setting to mark T cells. Further experiments involving additional panels of antibodies will be necessary to define specificities of CD4+ and CD8+ cells in tumour tissues.

While increased expression of tumour infiltrating CD20+ cells in a higher tumour grade was seen in non-IBC patients [28], we observed that an increase in CD20+ cell density inversely correlated with tumour grade in the IBC cohort. Taken together, these results suggest that cytokine- and chemokine-based networks that control communications between different cell types in IBC and non-IBC tumours are likely to be distinct. Indeed, previous studies have shown that tumour-infiltrating macrophages in IBC release pro-invasive and proangiogenic factors such as TNF, IL-6, IL-8, and IL-10 [29], and the expression of those cytokines was significantly higher among IBC in comparison to non-IBC patients [21]. In other studies, the expression of CCL21, CXCL12, and CXCL1/GROα chemokines as well as IL-15 and CSF-1 cytokines was significantly higher in IBC samples [30, 31]. Future experiments will be necessary to uncover the contribution of each of these factors both individually and in combination to the formation of the IBC-specific immune landscape.

We used mIF to start unravelling the network of communication between the major immune cell subtypes in IBC tissues. We observed molecular subtype-specific co-accumulation of immune cells, thus demonstrating that expression signatures of cancer cells play a pivotal role in defining the immune landscape in IBC. Indeed, co-accumulation of CD8+ and FOXP3+ cells was only observed in triple-negative IBC. By contrast, the correlation between the numbers of CD20+ cells and FOXP3+ cells was seen only in ER-positive IBC patients. While molecular networks underlying communication between FOXP3+ cells and other types of immune cells in IBC remain unknown, Treg-derived IL-10, TGFβ, IL-21, and IL-35 seemed to affect proliferation and differentiation of CD8+ cells and CD20+ lymphocytes [32, 33].

Lu et al. [9] demonstrated that an mIHC/mIF technique could improve performance in predicting the response to programmed cell death ligand 1/programmed cell death receptor 1 treatment in different solid tumour types when compared with PD-L1 immunohistochemistry, tumour mutational burden, or gene expression profiling alone [34]. In our study, we found that while the accumulation of a particular immune cell subtypes by itself does not predict a favourable patient response to the treatment, it becomes informative when evaluated in combination with infiltration of another type of immune cells (i.e., FOXP3+ cells alone vs. FOXP3+ and CD68+ cells). These results reflect the complex nature of communication between immune cells in IBC microenvironment, which could be an important factor in patient responses to a specific type of therapy.

In conclusion, our results strongly suggest that the composition of the immune microenvironment in IBC is governed by the disease-specific networks of communication between various types of immune cells and cancer cells. A further, more detailed multispectral analysis of the tumour microenvironment in a larger cohort of IBC patients and extensive profiling of IBC-associated signalling pathways will be required for better understanding molecular mechanisms, which define the immune landscape in IBC and open the scope for novel immunotherapy drugs for treating this aggressive cancer.

The authors are grateful for excellent technical help of Dr. G. Baldwin. The authors wish to acknowledge the role of the Breast Cancer Now Tissue Bank in collecting and making available the samples used in the generation of this publication, and the patients who donated to the Bank.

All experiments were approved by the West Midlands – Black Country NRES Committee (07/Q2702/24), Ethics Committee of the Juntendo University Hospital (14-144), and Leeds Research Ethical Committee (06/Q1206/180). All donors provided written informed consent for the collection of blood samples and subsequent analysis.

The authors declare no conflict of interest. A.M. Shaaban is a member of the Editorial Board of Pathobiology.

Nahla M. Badr was funded by the Egyptian Mission Sector, The Ministry of Higher Education and Scientific Research, Egypt. Jack L. McMurray was funded by the CRUK studentship. Abeer M. Shaaban is funded by Birmingham Cancer Research UK Centre (C17422/A25154). The work was funded by the Inflammatory Breast Cancer Network UK.

Abeer M. Shaaban and Fedor Berditchevski conceived the original idea and experimental design. Nahla M. Badr, Jack L. McMurray, Abeer M. Shaaban, and Fedor Berditchevski analysed data. Nahla M. Badr, Irini Danial, and Steven Hayward carried out experiments. Nahla M. Badr, Jack L. McMurray, Irini Danial, Steven Hayward, Nancy Y. Asaad, Moshira M. Abd El-Wahed, Asmaa G. Abdou, Marwa M. Serag El-Dien, Nisha Sharma, Yoshiya Horimoto, Tapan Sircar, Raghavan Vidya, Fiona Hoar, Daniel Rea, J. Louise Jones, Andrea Stevens, David Spooner, Reena Merard, Paul Lewis, Kelly John Hunter, Fedor Berditchevski, and Abeer M. Shaaban were involved in writing the paper and had final approval of the submitted and published versions.

The datasets used and/or analysed during the current study are available from the corresponding authors on reasonable request.