Spatial Transcriptomics Suggests That Alterations Occur in the Preneoplastic Breast Microenvironment of BRCA1/2 Mutation Carriers Free

This article is featured in Selected Articles from This Issue, p. 111

Breast cancer is the most common cancer in females, affecting one in every eight women and accounting for the majority of cancer-related deaths in women worldwide (1). Germline predisposition, reproductive history, and mammographic density are the best-known predictors of breast cancer risk. Parous women have a decreased risk of breast cancer than nulliparous women. Furthermore, younger parous women at the time of their first pregnancy have a lower risk than elderly parous women at the time of their first pregnancy. African American women had more births and a significantly younger mean age at first birth than white women, which can impact their pregnancy-modified risk of breast cancer (2). Mammographic density is a strong risk factor for breast cancer in white, African-American, and Asian-American women (3). Body mass index (BMI) is generally higher among African American and Hispanic women, which is associated with lower breast density (4). A major group of women who are at high risk for breast cancer are those with germline mutations in the BRCA1 or BRCA2 genes. Cumulative breast cancer risk is 72% for BRCA1 mutation carriers and 69% for BRCA2 mutation carriers by the age of 80 (5). Importantly, breast cancers arising in BRCA1 and BRCA2 mutation carriers differ in their cancer subtype classification. BRCA1-mutated breast cancers have a higher prevalence of triple-negative breast cancer (TNBC), which lacks estrogen receptor (ER) and progesterone receptor (PR) expression as well as HER2 gene amplification, whereas BRCA2-mutated breast cancers are histologically and immunophenotypically more similar to sporadic cancers (6, 7). Furthermore, the proportion of TNBC decreases with age at diagnosis in BRCA1 mutation carriers but increases with age at diagnosis of BRCA2 mutation carriers (8). The majority of BRCA1-mutated tumors exhibit a basal/myoepithelial phenotype by expressing several basal markers including cyto-keratins such as CK5/CK6 and CK14, whereas most BRCA2-mutated tumors show a luminal phenotype by overexpressing ER, PR, and cyto-keratins CK8 and CK18 (9). Basal-like breast cancers are proliferative, high-grade, and associated with a poor prognosis. In these cancers, therapeutic options include chemotherapy, PARP inhibitors, and immunotherapy (10, 11). On the other hand, luminal-like breast cancers are often low-grade, differentiated tumors expressing hormone receptors. They can be treated with hormonal therapy and generally display a relatively favorable survival (12). Therefore, the etiology and treatment of breast cancer varies significantly between BRCA1 and BRCA2 germline mutation carriers. However, the underlying biological mechanism remains largely unknown.

The human mammary gland comprises of epithelial and stromal cells communicating with each other via extracellular matrix (ECM; ref. 13). The mammary epithelium consists of two differentiated cell types organized into two cell layers, an inner layer of luminal epithelial cells and an outer layer of basal (myoepithelial) cells in direct contact with the basement membrane (14). Prior transcriptomic analyses of patient specimens have revealed loss of lineage fidelity in preneoplastic, that is, heterozygous but ostensibly normal, mammary epithelial cells of germline BRCA1/2 mutation carriers (15–19). However, cell fate decisions are driven by changes in environmental signals (20). The role of the microenvironment in mammary epithelial cell fate determination in BRCA1/2 mutation carriers has not been investigated. Here, we used NanoString GeoMx Digital Spatial Profiling (DSP) technique to allow spatial characterization of the epithelia and microenvironment in normal and preneoplastic BRCA1/2-mutant (BRCA1/2^mut/+) human breast tissues. We discovered spatially defined receptor–ligand interactions for autocrine and paracrine signaling, enabling comprehensive understanding of communication processes between mammary cells. We discovered that while BRCA2 deficiency activates |${\beta }1$|-integrin mediated mechano-transduction in preneoplastic human mammary epithelial cells (HMEC), BRCA1 deficiency does not, implying that BRCA1- and BRCA2-deficient cells communicate with the ECM in fundamentally different ways. On the other hand, BRCA1- and BRCA2-deficient preneoplastic human breast tissues are comparable in terms of paracrine signaling between epithelial cells and surrounding stromal cells, with BRCA1/2^mut/+ breast tissues having higher paracrine signaling than control tissues. These findings pave the way for developing novel breast cancer chemoprevention strategies in high-risk patients.

Fixed human breast tissues were obtained from the Cleveland Clinic with approval by the local Institutional Review Board (protocol 19–1257). Details of patient characteristics are provided in Table 1. The samples consisted of either normal breast tissues from reduction mammoplasties or noncancerous breast tissues from prophylactic mastectomies of known BRCA1 or BRCA2 mutation carriers. BRCA1/2 mutation carrier status was determined by clinical germline genetic testing before tissue collection.

Primary antibodies used for IHC were 1:1,000 KRT14 (BioLegend, 905301, RRID: AB_2565048), 1:500 KRT18 (abcam, ab24561, RRID: AB_736383), 1:500 KRT8/18 (Invitrogen, MA5–14088, RRID: AB_10984885), 1:300 active ITGβ1 (Sigma-Aldrich, MAB2079Z, RRID: AB_2233964), and 1:200 phospho-p44/42 MAPK (Erk1/2; Thr202/Tyr204; Cell Signaling Technology, 9101S, RRID: AB_331646). Primary antibodies used for IHC of FFPE tissues were 1:50 COL6 (Rockland, 600–401–108–0.1, RRID: AB_217576) and 1:500 ITGA6 (abcam, ab181551, RRID: AB_2927695). Secondary antibodies used for both methods, all diluted 1:1,000, were Alexa Fluor 488 Goat anti-Rabbit IgG Highly Cross-Adsorbed (Invitrogen, A-11034, RRID: AB_2576217), Alexa Fluor 647 Goat anti-Mouse IgG Cross-Adsorbed (Invitrogen, A-21235, RRID: AB_2535804), and Alexa Fluor 546 Goat anti-Rabbit IgG Cross-Adsorbed (Invitrogen, A-11010, RRID: AB_2534077).

Serially sectioned FFPE sections (5 μmol/L) of specimens were prepared to generate consecutive sections that were processed for H&E and RNA hybridization. After drying overnight, the slides were baked at 42°C for 3.0 hours and again baked at 60°C for 3.0 hours just prior to deparaffinization. RNA assay slides were deparaffinized and subjected to antigen retrieval (Tris-EDTA, pH 9.0; 15 minutes; >99°C), and enzymatic exposure of RNA targets (0.5 μg/mL Proteinase K; 15 minutes; 37°C). Slides were then hybridized with human whole transcriptome overnight (nanoString, #GMX-RNA-NGS-HuWTA-4) at 37°C. The next morning slides underwent stringent washes and were prepared for morphology marker staining. Slides were treated with blocking buffer for 30 minutes at room temperature in a humidified chamber, then co-incubated with fluorescently labeled visualization markers; PanCK [pan-cytokeratin (CK; Novus, Clone: AE1+AE3, Catalog No. NBP2–33200, RRID: AB_2924722], and with SYTO 13 (Thermo Fisher Scientific, Catalog No. S7575) for nuclei detection for 1 hour at room temperature in a humidified chamber. Slides were washed twice then loaded into the GeoMx for imaging. Once the samples were imaged, and regions of interest (ROI) were selected, and any segmentation made. During collection, each capture area (whole ROI or segmented areas) was exposed to UV light that cleaved the linker and released the barcoded oligos for capture by microfluidics. The released barcodes were collected in 96-well plates and were used in the NGS readout library preparation procedure protocol. The resulted libraries were sequenced by the Illumina NextSeq 6000 platform using reversed, 2 × 27 base paired reads. The resulting GeoMx DSP data were then coupled to next-generation sequencing (NGS) readout.

Nanostring DSP quantification DCC files were analyzed with R version 4.1.3 (21). R packages GeoMxWorkflows (22), GeomxTools (23), and NanoStringNCTools (24) were used to perform data normalization and quality control. Default values were used unless otherwise indicated. DCC counts and the PKC file (Whole Transcriptome Atlas, https://nanostring.com/wp-content/uploads/Hs_R_NGS_WTA_v1.0.pkc_.zip) were imported and segments identified as containing primarily ECM were excluded. All 0 counts were imputed to a value of 1. Candidate segment filter criteria were 1,000 minimum reads, 80% trimming, 80% stitching, 80% alignment, 50% sequencing saturation, 10 reads as a minimum negative control count, 1,000 reads as a maximum observed in NCT wells, and an area of 5,000. Geometric means for negative and endogenous control probes were calculated. One segment was removed due to low negative and endogenous probe geometric means, which also had a sequencing saturation under 30%.

A global (across all segments) probe filter criteria was applied of a minimum probe ratio of 0.1 (geometric mean of a given probe's counts across all segments, divided by geometric mean of all probes for the corresponding gene target across all segments) and if found to be an outlier via Grubb test in at least 20% of segments. In addition, probes were filtered for a given segment if found to be an outlier according to Grubb test in that segment. Probe counts were aggregated to gene counts. A limit of quantification (LOQ) per segment was calculated as the product of the negative probes’ geometric mean and the square of negative probes’ SD for each segment, or a value of 2, whichever was larger. Segments with less than 1% of genes above the LOQ were considered for exclusion. Genes were excluded if having a LOQ across segments <1%, leaving 11,833 genes for downstream analysis. 75th percentile (Q3) normalization was applied to the gene counts, which were then log2-transformed.

Visualization of segments before and after batch (scan) correction involved correcting log2-normalized counts with ComBat (25), using scan as the batch indicator and genotype as the group indicator. Principal components were calculated for corrected and uncorrected counts and plotted with ggplot2 (RRID: SCR_014601; ref. 26). Plots of Q3 normalization and genes detected above LOQ were created with ggplot2.

Differential gene expression between genotype pairs was performed using linear mixed-effects models, which included genotype as a fixed effect, and a random intercept at the patient level. Testing was performed on ComBat-adjusted log2-normalized values within each segment (stroma and epithelium). The results were adjusted for multiplicity at an FDR threshold of 0.05 across each gene and genotype pairing. Genes with nominal significance (unadjusted P < 0.05) were considered significant.

Known ligand–receptor pairs identified by Ramilowski 2015 (27) were downloaded from FANTOM (https://fantom.gsc.riken.jp/5/suppl/Ramilowski_et_al_2015/). Genes with average CAGE expression in the FANTOM mammary epithelial tissues were included for pairwise correlation analysis. Log2-normalized values were adjusted for batch by regressing out scan effect with linear regression and paired genes were compared between genotype pairs, both within each segment (Pearson correlation) and across segments (Spearman correlation) using a two-sized test. Multiple testing correction was performed with the Benjamini and Hochberg method, and adjusted P values less than 0.05 were considered significant.

Breast tissues were collected and processed as previously described with minor modifications (19). Briefly, the tissues were collected on ice, washed with PBS containing 50 U/mL penicillin–streptomycin, finely minced with scalpel, resuspended in 1 to 2 mg/mL collagenase solution (Sigma-Aldrich, C9722–50MG) in HBSS with Ca²⁺ and Mg²⁺ (Gibco, 24020) for 1 hour at 37°C. The tissue was then dissociated into a single cell suspension, and filtered through 40 μm filter. Single cells were seeded onto collagen (Advanced BioMatrix, 5074)—coated plates in complete media: DMEM/F12 medium (Thermo Fisher Scientific, 11330057) with 5% FBS [Cleveland Clinic Foundation (CCF) Cell Culture Core], 5 μg/mL insulin (Sigma-Aldrich, I1882–100MG), 2 mmol/L L-glutamine (Gibco, 25030–081), 10 ng/mL human recombinant epidermal growth factor (PeproTech, AF-100–15), 500 ng/mL hydrocortisone (Sigma-Aldrich, H0888–5G), 1× antibiotic/antimycotic (CCF Cell Culture Core), and 50 U/mL pen strep (CCF Cell Culture Core). Dead and floating cells were removed by replacing media daily. The cultures were initially obtained as mixed co-cultures of epithelial and stromal cells. Using differential trypsinization, the stroma was separated from the epithelia. Basal and luminal cytokeratin markers, KRT14 and KRT8/18, were co-immunostained on epithelial cultures to demonstrate the preservation of both basal and luminal mammary epithelial cell lineages. Cells are regularly tested for Mycoplasma using the MycoAlert Mycoplasma Detection Kit (Lonza, LT07–218).

siRNA transfections were carried out using Lipofectamine RNAiMAX (Thermo Fisher Scientific, 13778075) according to the manufacturer's protocol for “reverse transfection.” The siRNAs were provided as a “SMARTpool” of four separate siRNAs from Horizon Discovery (Table 2).

Total RNA was isolated using TRizol (Invitrogen, 15596026) according to manufacturer's protocol. Complementary DNA synthesis was performed using PrimeScript RT Reagent Kit (Takara RR037A). The qPCR reactions were carried out using Power SYBR Green PCR Master Mix (Applied Biosystems 4367659) run on a 7500 Real-Time PCR System (Applied Biosystems). Β-Actin expression levels were used to normalize expression between samples and 2^ -ΔCT values were used for quantifying relative expression of genes, with siCTRL values set to a mean of 1. The primer sequences were listed on Table 2.

Cells were seeded on 8-well chamber slides (Ibidi, 80841) for immunofluorescence staining. Briefly, 50 μg/mL of poly-D-Lysine (Sigma Aldrich, P6407–5MG) in cell culture grade water (Invitrogen) was added to the wells of the chamber slides and allowed to incubate for 1 hour at room temperature to aid in cell attachment. This solution was then aspirated, followed by three washes with cell culture grade water. Slides were then allowed to dry at room temperature for one hour if being used immediately or wrapped in Parafilm (USA Scientific, 3023–4526) and stored at 4°C for later use. Slides were subsequently coated with collagen I solution as previously described in “Primary cell culture” section. Cells were then reverse transfected onto chamber slides, cultured for three days, and washed twice with PBS before being fixed in chilled 100% methanol for 10 minutes. After that, cells were washed twice with PBS containing 10% normal goat serum (Cell Signaling Technology, 5425S) before being permeabilized for 5 minutes with PBS containing 0.1% Triton X-100 (Sigma Aldrich, T8787–50ML). Cells were then washed and blocked for 1 hour at room temperature with PBS + 10% goat serum and 0.1% Triton X-100, then incubated with the appropriate primary antibodies overnight at 4°C. Cells were then washed and incubated with fluorescently labeled secondary antibodies for 1 hour at room temperature. After washing, a drop of ProLong Diamond Antifade Mountant (Invitrogen, P36962) was added to each chamber and mounted with a glass coverslip (Electron Microscopy Sciences, 72200–40). Slides were then sealed with nail polish and imaged on EVOS M5000 microscope (Thermo Fisher Scientific, A40486). Images [at least 10–20 fields of view (FOV) per siRNA condition] were analyzed using ImageJ software. The integrated density of the fluorescent channel corresponding to the protein of interest was measured after using the “color threshold” tool to set the same minimum brightness threshold to all samples. To quantify dual positive integrated density, the color threshold of the merged image was adjusted to filter out non-yellow regions within each FOV. These integrated density values were then divided by the number of nuclei in the same FOV, which was obtained by auto-thresholding the DAPI channel and applying the “watershed” and “analyze particles” tools in ImageJ, resulting in a mean fluorescence value for each FOV.

Slides of sectioned FFPE patient tissues were baked at 62°C for 1 hour, deparaffinized with three washes in xylenes for five minutes each, and rehydrated in sequentially decreasing concentrations of ethanol for ten minutes each. Antigen retrieval was performed in 10 mmol/L sodium citrate buffer pH 6.0 (for COL6) or 10 mmol/L Tris / 1 mmol/L EDTA buffer pH 9.0 (for ITGA6) at 110°C for 17 minutes in a Decloaking Chamber NxGen from BioCare Medical. After cooling to room temperature, slides were blocked in PBS + 10% goat serum for 1 hour and incubated with primary antbodies overnight at 4°C in PBS + 0.5% BSA. Cells were then washed three times with PBS and incubated with secondary antibodies for one hour at room temperature in PBS + 0.5% BSA. After 4 washes with PBS, slides were mounted with ProLong Diamond Antifade Mountant, coverslipped, and sealed with nail polish. Imaging occurred on an EVOS M5000. Images were then analyzed with ImageJ software (RRID: SCR_00370). To quantify epithelial ITGA6 or stromal COL6, epithelial regions were outlined based on morphology of DAPI images. The selected region was then transferred to the channel of the protein of interest, and either the “clear outside” or “fill” tool was applied to block out the stromal or epithelial regions, respectively. “Color threshold” was then applied to set the same minimum brightness to all samples, and the mean intensity value was recorded. After subtracting the mean background obtained from no primary antibody control slides using this same method, mean intensity values for ITGA6 and COL6 were normalized separately by dividing each value by the average of its genotype within an experiment.

The raw and processed RNA-sequencing data generated in this study is available in Gene Expression Omnibus (RRID: SCR_005012; GSE234017). The remaining data are available within the Article and Supplementary Information.

Code for gene expression quality control and downstream analyses is available on github: https://github.com/bazelep/MolCancerRes_Caputo_10_23.

The study cohort included a total of 12 patients (Fig. 1A; Table 1). Noncancerous human breast tissues were collected from germline BRCA1/2 mutation carriers who underwent prophylactic mastectomy and from nonmutation carriers who underwent elective breast reduction mammoplasty (wild-type, WT). Because genetic breast cancers have a younger age onset, we collected samples from patients who were younger than 50 years of age. Three formalin-fixed paraffin-embedded (FFPE) tissue sections (each from one WT, one BRCA1 and one BRCA2 mutation carrier patient) were placed on one slide for performing RNA sequencing using the NanoString GeoMx DSP platform. After quality control filtering, expression data for 142 regions-of-interest (ROI) across 11,833 genes were selected for downstream analysis (Supplementary Fig. S1A). The raw sequencing counts were normalized using the third quartile of expression (Q3 normalization) and log2-transformed for all subsequent analyses (Supplementary Figs. S1B and S1C). Principal component analysis (PCA) showed batch effects which were corrected for in the downstream analyses (Supplementary Fig. S1D). Key findings were validated experimentally to ensure quality.

To decipher expression programs in spatially organized epithelial and surrounding stromal cells, histologic analyses following hematoxylin and eosin staining and immunostaining with the epithelial cell marker pan-cytokeratin (PanCK) and SYTO13 to mark the cell nuclei were used. Cell-type segments were created on the basis of PanCK staining (epithelia segment is positive for PanCK and SYTO13, whereas stroma segment is negative for PanCK and positive for SYTO13; Supplementary Fig. S1E). Our sequencing data contained 40% epithelial cells and 60% of stromal cells, according to segmentation masks that were utilized to aggregate the percent of total segment area occupied by each compartment across all ROIs (Supplementary Fig. S1F). We identified more nominally differentially expressed genes (DEG) in epithelial cells compared with stromal cells in pairwise comparisons across genotypes (Fig. 1B). A total of 819 genes were found to be nominally differentially expressed in epithelial cells between the WT and BRCA1^mut/+ groups, 485 genes between the WT and BRCA2^mut/+ groups and 652 genes between the BRCA1^mut/+ and BRCA2^mut/+ groups (unadjusted P < 0.05). A total of 474 genes were found to be nominally differentially expressed in stromal cells between WT and BRCA1^mut/+ groups, 349 genes between the WT and BRCA2^mut/+ groups and 494 genes between the BRCA1^mut/+ and BRCA2^mut/+ groups (unadjusted P < 0.05). Intriguingly, DEGs identified in WT-BRCA1^mut/+ comparison did not overlap with those identified in the WT-BRCA2^mut/+ comparison (∼80% genes were unique), suggesting that the expression patterns of preneoplastic mammary cells from BRCA1 and BRCA2 mutation carriers are significantly different (Fig. 1C). Pathway analyses of DEGs in BRCA1^mut/+-BRCA2^mut/+ groups showed that integrin and growth factor signaling is differentially regulated between BRCA1 and BRCA2 mutation carrier patients (Fig. 1D; Supplementary Fig. S1G).

To uncover autocrine signaling in epithelia and stroma compartments of BRCA1/2-deficient human breast tissues, we identified spatially defined receptor–ligand pairs co-expressed in either epithelial or stromal cells (28). Integrins constituted the majority of the receptor family that was found to be differentially correlated in genotype comparisons (Fig. 2A). β1-integrin (ITGB1) represents the predominantly expressed integrin in mammary cells and it is crucial for mammary gland development and differentiation (29, 30). |${\beta }$|1-integrin expression was reduced in BRCA1^mut/+ epithelial and stromal cells when compared with control and BRCA2^mut/+ cells (Supplementary Fig. S2A). Therefore, we explored |${\beta}$|1-integrin-ligand pairs in control and BRCA1/2-deficient human breast epithelial and stromal cells. |${\beta }$|1-integrin–ligand pairs were more prevalent in mammary epithelial cells than in stromal cells, and they were the most differentially correlated in the BRCA1^mut/+–BRCA2^mut/+ comparison (Fig. 2B and C). |${\beta }$|1-integrin expression correlated with the expression of matrix degrading enzymes (ADAM17), angiogenesis genes (THBS1), or growth factors/chemokines (CSF2, CXCL12) in BRCA1^mut/+ cells. On the other hand, |${\beta }$|1-integrin expression correlated with the expression of type I collagen (COL1A1) and laminin genes (LAMA4, NID1) in BRCA2^mut/+ cells. Integrins are heterodimeric transmembrane receptors made up of |${\alpha}$| and |${\beta }$| subunits (31). |${\beta }$|1 integrin becomes a collagen receptor when heterodimerizes with the |${\alpha }11$| subunit (32). Expression of |${\alpha }11$| subunit gene ITGA11 also correlated with the expression of type 1 collagen genes (COL1A1 and COL1A2) in BRCA2^mut/+ mammary cells.

Enzyme-linked receptors constituted the second most abundant receptor family that was found to be differentially correlated in genotype comparisons (Supplementary Fig. S2B). TGF |${\beta }$| pathway receptors were most significantly altered among the enzyme-linked receptors in genotype comparisons. Importantly, TGFBR3–TGFB2, BMPR2–BMP2, and BMPR2–RGMB pairs were shown to be differentially correlated in BRCA2^mut/+ mammary epithelial cells compared with BRCA1^mut/+ cells. TGF |$\beta $| signaling is a potent inducer of collagen production by both epithelial and stromal cells in culture (33). Integrin binding to collagen generates a mechanical force in the ECM, which is linked to the cytoskeletal tension and activates mechano-signaling pathways (34, 35). Interestingly, several actin/myosin genes were upregulated in BRCA2^mut/+ mammary epithelial cells compared with BRCA1^mut/+ epithelial cells (Supplementary Fig. S2C). On the other hand, expression of the matrix metalloproteinases MMP3 and MMP8, which are collagen degrading enzymes, were increased in BRCA1^mut/+ epithelial cells (Supplementary Fig. S2A). These findings suggest that ECM remodeling mechanisms differ significantly between BRCA1^mut/+ and BRCA2^mut/+ preneoplastic human mammary cells, with BRCA1^mut/+ cells favoring ECM breakdown and BRCA2^mut/+ cells favoring force-mediated ECM modification.

Cells can detect and interpret the biological information in the ECM by adhering to it via integrin-based adhesions and eliciting a series of dynamic signaling events known as mechano-signaling. Integrins are transmembrane ECM receptors that function as mechano-transducers and integrin clustering activates growth factor-ERK-dependent cell growth (36, 37). In this context, we asked whether |${\beta }$|1 integrin-mediated mechano-signaling was differentially regulated in response to BRCA1 or BRCA2 loss. Our data showed that expression of activated |${\beta }$|1 integrin and phosphorylated ERK was decreased with BRCA1 loss-of-function compared with BRCA2 loss-of-function in primary HMECs (Fig. 3; Supplementary Fig. S3). We replicated these findings using ATCC's HMEC cells (Supplementary Fig. S4).

Paracrine signaling mechanisms are essential for both normal mammary gland development and breast cancer (38). To reveal the role of paracrine signaling in mammary epithelial cell fate determination in BRCA1/2^mut/+ human breast tissues, we identified spatially defined receptor–ligand pairs co-expressed across ROIs in epithelial and stromal cells. In other words, we assessed the expression of a receptor in one compartment (epithelium or stroma) and its cognate ligand expression in the other compartment, or vice versa. Similar to autocrine signaling, integrins and enzyme-linked receptors comprised the majority of the receptor family in paracrine signaling that were found to be differentially correlated in pairwise genotype comparisons. As opposed to autocrine signaling, syndecans were a significant receptor family in paracrine signaling between epithelial and stromal cells (compare Figs. 2A and 4A). We discovered that BRCA1/2^mut/+ human breast tissues exhibit greater integrin-dependent paracrine signaling than control tissues (Fig. 4B and C). Interestingly, in BRCA1/2^mut/+ breast tissues, integrin and syndecans function more as stromal receptors than epithelial receptors, in striking contrast to control tissues. The ITGA6–COL6A1 was one such receptor–ligand pair that was found to be differentially correlated between control and BRCA1/2^mut/+ human breast tissues. In control tissues, epithelial ITGA6 expression was positively correlated with stromal COL6A1 expression; however, in BRCA1/2^mut/+ breast tissues, this correlation disappeared. Consistently, stromal ITGA6 expression was negatively correlated with epithelial COL6A1 expression in control tissues; however, this correlation was absent in BRCA1/2^mut/+ breast tissues (Supplementary Fig. S5A). We quantified ITGA6 immunostaining in mammary epithelial cells and type VI collagen immunostaining in the adjacent stroma using indirect immunofluorescence analysis, allowing us to validate our spatial transcriptomics findings (Fig. 5A) at the protein level (Fig 5B and C). Although comparable results were observed for enzyme-linked receptors, fewer receptor–ligand pairs with differential correlations were identified compared with integrin receptors in BRCA1/2^mut/+ breast tissues (Supplementary Fig. S5B). FGFR2–FGF2 was one such receptor–ligand pair that functioned more as stromal receptors than epithelial receptors in BRCA1/2^mut/+ breast tissues. Similar correlation patterns were observed between the expression of FGFR2–FGF2 receptor–ligand and ITGA6–COL6A1 (Supplementary Fig. S5C). Together, these data indicate that stromal cells in BRCA1/2^mut/+ human breast tissues may have more active signaling with the microenvironment compared with stromal cells in control human breast tissues.

Cell fate decisions are regulated by the microenvironment. Using whole-transcriptome digital spatial profiling of human breast tissues from control and BRCA1/2 mutation carriers, we show for the first time that BRCA1^mut/+ and BRCA2^mut/+ preneoplastic epithelial cells communicate differently with their surrounding microenvironment. Our findings reveal both distinct and shared aspects of microenvironmental interactions in BRCA1^mut/+ and BRCA2^mut/+ breast tissues. We investigated autocrine and paracrine cell signaling through the identification of spatially defined receptor–ligand interactions. We found that integrin-promoted autocrine signaling in epithelial cells distinguished BRCA2^mut/+ epithelial cells from BRCA1^mut/+ epithelial cells. We discovered that expression of ITGB1 and ITGA11 correlated with the expression of type I collagen genes (COL1A1, COL1A2) in BRCA2^mut/+ cells, but not in BRCA1^mut/+ cells. Furthermore, we showed that in contrast to BRCA1^mut/+ mammary epithelial cells, expression of multiple receptor–ligand pairs of TGF |${\beta }$| pathway, which is a major pathway required for collagen production, are differentially correlated in BRCA2^mut/+ mammary epithelial cells. On the other hand, expression of ITGB1 correlated with the expression of ECM degrading enzymes, growth factors, or angiogenesis genes in BRCA1^mut/+ epithelial cells compared with BRCA2^mut/+ epithelial cells. The enhanced protease activity may cause subsequent ECM degradation that releases the ECM-bound growth factors and thereby increase their bioavailability in BRCA1^mut/+ epithelial cells. In addition, we found that expression of collagen degrading enzymes MMP3 and MMP8 are increased in BRCA1^mut/+ epithelial cells (Supplementary Fig. S2A). Interestingly, a recent study discovered pre-cancer-associated fibroblasts that produce protumorigenic MMP3, which promotes BRCA1-driven tumorigenesis in vivo (39). Dysregulation of ECM can lead to several severe human conditions, including cancer. In some cases, the ECM becomes mechanically or enzymatically degraded. In other cases, a net accumulation of ECM is part of the pathologic event (40, 41). Our data suggest that there is more collagen deposition in BRCA2^mut/+ breast tissues, whereas there is more collagen degradation in BRCA1^mut/+ breast tissues. Our findings may shed light on the question of why BRCA1 and BRCA2 mutation carriers develop breast tumors with distinct pathologic characteristics. Regarding the paracrine signaling between epithelial cells and surrounding stromal cells, BRCA1^mut/+ and BRCA2^mut/+ preneoplastic breast tissues were comparable. In contrast to normal tissues, more integrin–ligand pairs were differentially correlated in BRCA1/2^mut/+ breast tissues, with stromal cells expressing more integrin receptors. These findings suggest that communication from epithelia to surrounding stroma is increased in BRCA1/2^mut/+ breast tissues. We hypothesize that epithelial cells may alter ECM composition to influence and educate stromal cells in BRCA1/2 mutation carriers to acquire a premalignant phenotype.

Breast cancer is a heterogenous disease. Comprehensive gene expression profiling studies identified five major molecular subtypes: basal-like, luminal A, luminal B, HER2-positive, and claudin-low (42, 43). These subtypes are associated with different clinical outcomes and treatment responses (44, 45). Our study investigated a critical question: why do germline BRCA1 and BRCA2 mutation carriers develop distinct subtypes of breast cancer? It is hypothesized that mammary stem cells and progenitors are the origin cells of breast cancer (46). Both BRCA1 and BRCA2 mutation carriers demonstrate abnormal mammary epithelial cell differentiation (17–19). Given the disparities in clinical features of BRCA1 and BRCA2 mutation carriers’ breast cancers, we suggest that BRCA1 and BRCA2 loss do not have the same effect on mammary epithelial cell differentiation. Clearly, microenvironmental cues play a crucial role in cell fate regulation and tissue-specific development. Here, we showed that in contrast to BRCA1 deficiency, BRCA2 deficiency increased the expression of activated |${\beta }$|1 integrin and phosphorylated ERK in HMECs, reflecting an increase in force-mediated cellular mechano-signaling. High mechano-signaling is associated with the stiffness of the ECM and aggressiveness of the breast cancer. Furthermore, a stiff ECM may facilitate detrimental tumor characteristics by promoting stemness and a mesenchymal-like phenotype (47–49). Among breast cancer subtypes, the invasive front of the most aggressive subtypes (basal-like and HER-2) was shown to be the stiffest and most heterogeneous, containing cells with the highest mechano-signaling, as compared with the less aggressive subtypes (luminal A and Luminal B; ref. 50). Because we found an evidence of increased mechano-signaling in BRCA2 mutation carriers in preneoplastic human breast cells and increased mechano-signaling is associated with aggressive basal type breast cancers, our findings may seem contradictory. However, it is well known that ECM is extensively remodeled at the advanced stages of cancer, and less is known about the role of ECM in tumor initiation or in driving early phenotypic changes in high-risk breast tissues. We suggest that ECM evolves throughout cancer progression. ECM changes that occur during the early stages of tumorigenesis may not be similar to ECM changes that occur in established, advanced tumors. Our findings suggest that BRCA1 or BRCA2 deficiency may affect integrin-mediated mechano-signaling differently in preneoplastic human breast epithelial cells, shedding insight on BRCA1/2 mutation carriers’ susceptibility to breast cancers with distinct clinical characteristics. Here, although we expanded on prior studies by dissecting epithelia-microenvironment communication in BRCA1/2 mutation carriers, we were limited by the sample size. Nevertheless, we complemented our spatial transcriptomics-based findings with patient-derived cell culture models.

In conclusion, our work provides a framework for understanding the spatial interactions between mammary epithelial cells and the microenvironment directing epithelial cell fate in BRCA1/2 mutation carriers and suggests markers for future translational studies aimed at improving risk prediction and breast cancer prevention in high-risk women.

R.A. Keri reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

A. Caputo: Data curation, formal analysis, validation, investigation, methodology. K. Vipparthi: Validation, investigation, methodology. P. Bazeley: Data curation, software, formal analysis, investigation, methodology. E. Downs-Kelly: Resources, methodology. P. McIntire: Resources. L.A. Duckworth: Resources. Y. Ni: Methodology. B. Hu: Formal analysis, methodology. R.A. Keri: Conceptualization, funding acquisition. M. Karaayvaz: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

We thank the Advanced Genomics Core at the University of Michigan for library preparation and next-generation sequencing. We are very grateful to Cleveland Clinic Central Biorepository for providing us human breast tissues. We also thank pathologists Drs. Xiaoyan Cui and John Van Arnam for their help with patient sample access. All of the figures were created by BioRender. This work was supported by seed funds from the Cleveland Clinic's Lerner Research Institute (to M. Karaayvaz), R01CA213843 (to R.A. Keri), R01CA257502 (to R.A. Keri), and VeloSano for the Cure (to R.A. Keri).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.