Characterization, immune cell phenotyping and differential gene expression analysis of brain metastases and primary breast cancer samples.

Introduction
Metastatic breast cancer (MBC) is generally an incurable disease. Effective antitumor therapies are available, but intrinsic or acquired resistance leads to tumor progression despite treatment. Tumor heterogeneity in breast cancer is today widely recognized. Based on gene-expression assays breast cancer can be divided into different subgroups with distinct biological, clinical, and prognostic behavior allowing more individualized treatment strategies. Most of the available knowledge, however, derives from the investigation of primary breast tumors, while metastatic lesions are much less studied. Available studies have shown differences in hormone receptor and HER2 expression between primary tumors and metastatic lesions1, and emerging data reveal differences in gene mutations and copy number alterations as well2–4. Furthermore, quantification of PD-L1 and tumor-infiltrating lymphocytes (TILs) has shown marked alterations in paired primary and metastatic tissues5.
Fifteen to thirty% of patients with MBC develop BM during the course of their disease6,7, making it the second most common cause of BM after lung cancer. BM are not only associated with poor prognosis but also with neurological impairments leading to a major limitation of life expectancy and quality of life for many patients with MBC8. The incidence of BM in MBC is rising9, probably because many patients survive longer due to improvements of systemic therapies and better extracranial disease control. Recently, effective systemic therapies for breast cancer, that cross also the blood-brain-barrier (BBB), have entered the clinical practice. These therapies are, however, mainly restricted to the HER2-positive subtype. The currently available immunotherapy with PD-1/L1 directed checkpoint inhibitors, in contrast, is only effective in a subgroup of triple-negative MBC10–12 and has no sufficient clinical evidence for efficacy against BM13. Therefore, locoregional treatment strategies such as surgery, focal radiotherapy and whole brain radiotherapy (WBRT) are frequently used to achieve local disease control. Consequently, there is a high medical need for improved systemic treatment options for breast cancer BM including new immunotherapeutic strategies.
Integrated analyses of immune cell phenotyping, characterization, and gene expression analyses in metastases from breast cancer are still scarce and this is especially true for BM, for which tumor samples are uncommonly available. Such analyses are, however, fundamental in order to understand mechanisms of BM formation, progression, and immune evasion. A deeper comprehension of these mechanisms could pave the way for the development of innovative treatment approaches in the future, such as novel immunological treatments.
Here, we present an integrated analysis of breast cancer BM by using different methods including immunohistochemistry, in-situ hybridization (ISH), quantification of tumor-infiltrating lymphocytes (TILs), and bulk RNA-sequencing analysis to capture comprehensively their biology and differences to the corresponding primary tumor. A special focus was set on the characterization of the immune infiltration in the two different compartments.

Patients and methods
All experiments and analyses were performed in accordance with regulations by the Ethics Committee of the province Salzburg, the principles of Good Clinical Practice and the Declaration of Helsinki. All patients who underwent neurosurgical resection of BM at the University Hospital Salzburg between 10/2001 and 10/2016 with available follow-up data were identified. Patients with enough tumor material for isolation of at least 300 ng of RNA both from the BM and the primary tumor were included. A tumor cell content of at least 30% in the selected tumor samples was required. Microdissection was performed if indicated.
In all samples from BM and corresponding primary tumors the following analyses were performed: (1) Immunohistochemical (IHC) staining of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). (2) In situ hybridization of HER2 in case of IHC 2+. (3) Definition of the histologic subtype and histologic grade (4) Quantification of intratumoral and stromal tumor infiltrating lymphocytes (TILs) according to the algorithm of the TIL working group14, and (5) whole transcriptome RNA-sequencing (RNAseq).
For the RNAseq-analysis, RNA was extracted from ten consecutive 10 μm sections from FFPE tissue using the RNeasy FFPE Kit (QIAGEN, Hilden, Germany) following the manufacturer’s protocol. RNA concentration and purity were assessed using a Tapestation instrument (Agilent, Santa Clara, USA), and quality control was performed with MiSeq sequencing (Illumina, San Diego, USA). Due to a high level of degraded RNA an amplification step was required. Therefore, RNA libraries were generated through exome target enrichment with All Prep Kit and Exome Panel (Illumina, San Diego, USA). Finally, libraries were sequenced on a NovaSeq platform (Illumina, San Diego, USA). Fastq files were quality trimmed using trimmomatic (v0.39)15 and aligned to the human hg38 genome using STAR aligner (v2.7.9a)16. Gene expression quantification was performed by activating the GeneCounts mode during STAR alignment. Based on RNA expression levels, the intrinsic breast cancer subtypes according to PAM50 classification (subtypes luminal A, luminal B, HER2-enriched and basal-like) were calculated17,18. Immune cells infiltrating the tumor microenvironment were deconvoluted and quantified using Newman’s 22 leukocyte signature matrix (LM22)19 and the R packages „ADAPTS“20 and „DeconRNASeq“21. Digital cell quantification (DCQ)22 was chosen as appropriate method. Differential Expression (DE) analysis between BM and primary tumor was performed using the „edgeR“ R package23. The most deregulated genes and gene sets were analyzed and visualized using the „EnhancedVolcano“24, „enrichplot“25, „gsva“26 and „clusterProfiler“27 R packages. Heatmaps were created using the „pheatmap“ R package.

Results
A total of ten patients with 21 tumor samples (one patient had bilateral primary breast tumors) were identified with sufficient amount and quality of RNA. Given the random selection and the small sample size, patients’ characteristics were very heterogeneous regarding age (range 37–62 years), disease-free survival from diagnosis of primary tumor until diagnosis of MBC (range 10.8–49.7 months), brain-metastases-free survival (calculated from diagnosis of MBC; range 0–27.1 months), treatment and prognosis (Table 1). Therefore, no correlations of clinicopathological characteristics or survival data with the translational research results were performed.
More than 60% of the primary tumors and almost all BM were of high histologic grade (90%). Only one patient (10%) had lobular breast cancer, all others were classified as no special type (NST; 90%; Table 2). Concerning receptor status, four patients (36.4%) had triple-negative, two patients (18.2%) HER2-positive, and four patients (36.4%) luminal primary breast tumors (five patients in total, when both primary tumors of patient n°9 were counted). In four patients (40%) the receptor status changed between primary tumor and BM: in patient n°2, the BM lost ER expression compared to the corresponding primary tumor (ER 30%) under the selection pressure of two lines of endocrine therapy, while in patient n°7 the BM gained an ER expression of 5% from a completely ER-negative primary tumor. In two other patients the HER2 status changed from negative to positive in the BM (patients n°9 and n°10; Table 2, Supplementary Figure S1). This resulted in 20% triple-negative, 40% HER2-positive, and 40% luminal BM. Interestingly, the PAM50 classification did not change in the four patients with changing receptor expression. However, the PAM50 subtype changed in three different patients (30%): patient n°3 had a HER2-enriched primary tumor and a Luminal B BM (both primary and BM were HER2 IHC 3+; Table 2; Supplementary Figure S2). When examined in greater detail, the subcategories, that collectively define the final PAM50 classification, revealed that only a minor change in the genes responsible for subtype classification triggered this outcome, while the overall biology remained largely unchanged (Fig. 1). In patient n°8, the brain metastasis was classified as Luminal B, whereas the primary tumor was classified as Luminal A, indicating tumor evolution during the metastatic process (Table 2, Supplementary Figure S2). In patient n°4, in contrast, tumor biology changed completely between primary tumor and BM: despite the fact that both tumors were of lobular subtype and immunohistochemically HR+/HER2-, the primary tumor was classified as Luminal A and the BM as HER2-enriched (Fig. 1, Supplementary Figure S2).
A total of 251 genes were differentially expressed (DE) between primary tumors and BM with a false discovery rate (FDR) of < 0.05. Among these, 199 genes were downregulated, and 35 genes were upregulated in BM with an FDR of < 0.05 and a log2 fold-change (log2 FC) of <−1 or > 1, respectively. Stratification for IHC subtype in the primary tumor showed an aggregation of DE genes in HER2 negative tumors (Supplementary Figure S3), however, the limited patient number (TNBC: 4, HER2-pos: 2, HER2-neg: 4) for each subgroup does not warrant definitive conclusions. We performed gene set enrichment analysis (GSEA) based on the DE data in order to identify dysregulated gene sets or pathways in five different gene set databases (GO:BP, KEGG28, MSigDB, Reactome, WikiPathway). The top enriched pathways in primary samples included predominantly immune and inflammatory signaling and lymphocyte differentiation, while genes relevant to mitochondrial electron transfer chain (ETC) and oxidative phosphorylation (OXPHOS) were enriched in the metastatic samples. (Fig. 2 and Supplementary Table T1).
Hierarchical unsupervised clustering of deconvoluted and quantified immune cells from RNA sequencing revealed two main clusters strongly correlating with sample origin: in Cluster 1, all but one sample originated from primary tumors (Supplementary Figure S4). This classification was primarily driven by differences in B-cell (particularly plasma cell) and CD8 + T-cell infiltration, which were strongly associated with primary tumors and almost absent in BM. In contrast, M2 macrophages and follicular helper T-cells were more frequently detected in BM (Fig. 3; Supplementary Figures S4 and S5). These distinct differences in lymphocyte infiltration were not reflected by the count of stromal or intratumoral TILs conducted visually by a pathologist: in five patients (50%) the percentage of TILs was higher in the primary tumor and in five patients (50%) the percentage was higher in the BM (Table 2, Supplementary Figure S6). The median TIL percentage was lower within the tumor of primary lesions (2% vs. 3%; IQR 6.5 vs. 4.0) but higher in the stromal compartment (12% vs. 10%; IQR 17.5 vs. 11.3), however neither difference was statistically significant (p = 0.55; p = 0.33, respectively; Supplementary Figure S6).

Discussion
Through this integrated analysis of paired BM and primary breast tumors, we observed considerable intercompartmental variability. Differences were noted not only in receptor status (ER, PR or HER2) and PAM50 subtypes, but also in TILs levels. At the gene expression level, a number of genes appeared downregulated in BM compared to the corresponding primary breast tumor, including genes involved in immune-related pathways such as lymphocyte differentiation and adaptive immune responses. Furthermore, RNA-based analyses suggested a lower relative abundance of B-cells (particularly plasma cells) and CD8 + T-cells in BM, whereas M2 macrophages and follicular helper T-cells appeared more frequent.
These observations are well in line with those from others showing that breast cancer metastases had a significantly lower abundance of total immune cells, including CD8 + T-cells, regulatory T-cells and dendritic cells compared to the corresponding primary tumors. In contrast, M2-like anti-inflammatory, protumorigenic macrophages were found to be higher in metastases across different organ sites including BM29,30.
Previous studies have demonstrated that brain cancers and BM from various tumors contain a high number of tumor-associated macrophages31–33. A key challenge in studying BM is distinguishing brain-resident microglia from other tumor-associated macrophages. Recent single-cell RNA sequencing research successfully identified microglia as a distinct population with a pro-inflammatory and tumor-suppressive role in breast cancer BM34. In genetic and humanized mouse models, the absence of microglia led to increased metastasis, reduced survival, and diminished natural killer and T-cell responses, highlighting their crucial role in promoting anti-tumor immunity to suppress BM from breast cancer34.
The AURORA US network performed a profound integrated analysis of 55 females with metastatic breast cancer (51 primary cancers and 102 metastases including eleven BM). Similar to our results that study described lower immune and stromal-related signatures in brain metastases as compared to primary tumors35.
Consistent with previous reports, we observed a lower number of histologically assessed TILs in BM compared with primary tumors34; however, this pattern did not hold at the individual level: in 50% of patients, the BM actually contained higher TIL counts than the corresponding primary tumor. Notably, RNA sequencing–based immune cell deconvolution suggested a more consistent reduction in B cells and CD8⁺ T cells in BM, with the potential advantage of distinguishing between specific immune cell subsets and, to some extent, functional states. This discrepancy may, however, also be explained by methodological differences: histologic TIL assessment primarily quantifies overall stromal lymphocytic infiltration, while quantification of intratumoral TILs is less reproducible14. In contrast, bulk RNA sequencing averages gene expression signals across different tissue regions. While this approach is widely used, it remains indirect and sensitive to RNA quality, tumor purity, and reference signatures - particularly in FFPE-derived samples with RNA degradation. Given the small and clinically heterogeneous cohort, variability at the individual level is likely amplified. Together, these considerations highlight that histologic and transcriptomic immune assessments are complementary but not interchangeable and should be interpreted with caution in exploratory analyses such as ours.
A related study characterized the genomic and immune landscape of BM from a cohort of 42 breast cancer patients using whole-exome sequencing (WES) and total RNA libraries and compared findings with extracranial metastases36. In contrast to our cohort, only a limited number of primary tumors were available. Consistent to our observations, the authors reported a 60% concordance of PAM50 subtypes between intracranial and extracranial metastases. They further identified enrichment of MYC targets, DNA damage repair, cholesterol homeostasis, and oxidative phosphorylation pathways in BM, whereas immune-related pathways were downregulated. Notably, and in contrast to our findings, they observed an increased abundance of activated dendritic cells in BM; we did not detect activated dendritic cells in any of the BM samples analyzed36.
Another study analyzed 25 TNBC BM and 9 matched primary tumors using WES and RNAseq. Again, the authors reported an overall reduction in immune gene signature expression in BM, however an elevation of a fibroblast-associated wound-healing signature was reported. In addition, metrics of T and B cell receptor diversity were decreased in BM37.
In melanoma, combined immunotherapy with the anti-CTLA-4 antibody ipilimumab and the anti-PD-1 antibody nivolumab has demonstrated substantial intracranial activity in patients with active BM (intracranial response rate of 56%)38. In breast cancer BM, however, data supporting efficacy of immune therapy are still scarce. The registration trial for pembrolizumab plus chemotherapy as first-line therapy for triple-negative breast cancer (KEYNOTE-355) included a small number of patients with stable and treated BM (3% of the intention-to-treat population), however, CNS-specific outcomes have not yet been reported39. IMpassion130, the registration trial for atezolizumab, included 61 patients (6.8%) with asymptomatic or previously treated BM. Unfortunately, in this subgroup both PFS and OS were worse in patients receiving atezolizumab plus nab-paclitaxel than in patients who received nab-paclitaxel alone, irrespective of PD-L1 expression. The low number of patients (only 26 patients were classified as PD-L1 positive), however, does not allow definitive conclusions about the efficacy of immune checkpoint inhibitors in breast cancer BM40. In a single arm phase II trial 58 patients with untreated or progressive BM of diverse histologies, including 34 patients with breast cancer, were treated with pembrolizumab monotherapy41. Even if the study reported an intracranial benefit rate of 42.1% (37.1% in breast cancer), the intracranial progression-free survival was short (median 1.6 months; 90% CI 1.4 to 2.9 months) and responses defined as complete (CR) or partial response (PR) according to the RANO criteria42 were infrequent (8.8%). None of the included patients with breast cancer experienced a radiologic response41. In a single-arm, multicenter, phase II trial the anti-PD-L1 antibody atezolizumab was added to pertuzumab, and high-dose trastuzumab in patients with HER2-positive breast cancer BM. The CNS overall response rate of 10.5% did not meet the prespecified efficacy threshold and the study was terminated early43.
It is therefore reasonable to postulate that in low-immunogenic tumors like breast cancer, other types of immune therapy are needed to overcome the immune escape mechanisms present in BM in order to yield clinical response and long-term tumor control.
The primary limitation of our study is the small sample size and descriptive nature of the analyses, which prevented subgroup analyses and limits the generalizability of our findings. However, expanding biobanks for breast cancer BM remains challenging, as most patients undergo primary radiotherapy without neurosurgical intervention or histologic assessment. Another limitation is that we did not perform functional assays to confirm the metabolic reprogramming suggested by the pathway analyses. Therefore, our findings should be interpreted as hypothesis-generating rather than definitive evidence of consistent biological differences between primary tumors and BM.

Conclusion
An integrated analysis of paired BM and primary breast tumors revealed changes in receptor status and/or PAM50 breast cancer subtypes, as well as differences in immune cell infiltration. In particular, B- and CD8 + T-lymphocytes were less frequently detected in BM, whereas M2 macrophages and follicular helper T-cells appeared more prevalent. Despite the limited sample size, our results suggest a possible shift toward a more immunosuppressive microenvironment in BM, potentially contributing to immune escape and limited efficacy of immune checkpoint inhibitors; however, these observations require validation in larger cohorts. Improved immunotherapeutic approaches are urgently needed as well as trials sufficiently including patients with BM to change the natural course of CNS involvement of breast cancer.

Supplementary Information
Below is the link to the electronic supplementary material.