Spatial molecular profiling of mixed invasive ductal and lobular breast cancers reveals heterogeneity in intrinsic molecular subtypes, oncogenic signatures, and mutations

Significance

Mixed invasive ductal and lobular carcinoma (MDLC) displays both ductal and lobular tumor regions. Our multiomic profiling approach revealed that these morphologically distinct tumor regions harbor distinct intrinsic subtypes and oncogenic features that may cause prognostic uncertainty and therapeutic dilemma. Thus, histopathological/molecular profiling of individual tumor regions may guide clinical decision-making and benefit patients with MDLC, particularly in the advanced setting where there is increased reliance on next-generation sequencing.

Abstract

Mixed invasive ductal and lobular carcinoma (MDLC) is a rare histologic subtype of breast cancer displaying both E-cadherin positive ductal and E-cadherin negative lobular morphologies within the same tumor, posing challenges with regard to anticipated clinical management. It remains unclear whether these distinct morphologies also have distinct biology and risk of recurrence. Our spatially resolved transcriptomic, genomic, and single-cell profiling revealed clinically significant differences between ductal and lobular tumor regions including distinct intrinsic subtype heterogeneity – e.g., MDLC with triple-negative breast cancer (TNBC) or basal ductal and estrogen receptor positive (ER+) luminal lobular regions, distinct enrichment of cell cycle arrest/senescence and oncogenic (ER and MYC) signatures, genetic and epigenetic CDH1 inactivation in lobular but not ductal regions, and single-cell ductal and lobular subpopulations with unique oncogenic signatures further highlighting intraregional heterogeneity. Altogether, we demonstrated that the intratumoral morphological/histological heterogeneity within MDLC is underpinned by intrinsic subtype and oncogenic heterogeneity which may result in prognostic uncertainty and therapeutic dilemma.

Breast cancer is a prevalent disease worldwide (1) and is the most frequently diagnosed cancer in women in the United States, with approximately 300,000 cases per year (2). The majority (approximately 75%) of cases are classified histologically as invasive breast carcinoma of no special type (IBC-NST), previously described as invasive ductal cancer (IDC). Invasive lobular carcinoma (ILC) is the most common special type accounting for approximately 10 to 15% of all invasive breast cancer diagnoses (3, 4). Apart from the well-defined histologic and tissue histological differences, there are numerous notable molecular distinctions between ILC and IBC-NST. ILC exhibits enrichment of CDH1, FOXA1, TBX3, ERBB2, and PTEN alterations, while IBC-NST shows a higher frequency of GATA3, MAP3K1, TP53, and MYC alterations (5). As a result of CDH1 alterations, ILC lacks expression of E-cadherin, an important cell adhesion molecule (4), and displays cytoplasmic p120 staining while IBC-NST retains membranous expression for both molecules (6). Clinically, ILC is more common in elderly women, presents as larger tumors, and is frequently multicentric and multifocal compared to IBC-NST (7, 8). ILC is usually detected later and leads to higher rates of mastectomy (7–11). ILC is also associated with more frequent metastases to urogenital and gastrointestinal tracts as well as to serosal surfaces than IBC-NST (12, 13). Despite presenting with markers of favorable prognosis, such as low grade, low proliferation index, estrogen receptor positive (ER+), and human epidermal growth factor receptor (HER2) low/negative, patients with ILC have a higher rate of late recurrences and worse long-term prognosis than those with IBC-NST (8, 14, 15). Given its distinct prognosis and outcome compared to IBC-NST, ILC is increasingly being recognized as a distinct disease, yet it is still treated similarly to IBC-NST in the clinic due to a lack of specific treatment guidelines (8, 15, 16). Ongoing clinical trials focused on patients with ILC may lead to the development of more tailored and effective treatment options in the future (15).

Apart from ILC, there are numerous other special types of breast cancers such as mixed invasive breast carcinoma. This is an elusive group of breast cancers which the World Health Organization (WHO) defines as tumors presenting with areas of 10 to 90% of special histology (such as ILC) admixed with at least 10% of IBC-NST (17). Importantly, this current WHO definition of mixed invasive breast cancers MDLC differs considerably from its previous definition, i.e., tumors presenting with areas of 10 to 49% of the special subtype and at least 50% IBC-NST (18), indicating the fluid interpretation of this disease. Mixed tumors composed of both ILC and IBC-NST are the most common, accounting for ~5 to 10% of all breast cancer cases (5, 19–23), and have different terminologies in the literature including mixed invasive ductal and lobular breast carcinoma (MDLC) (23) (used throughout this study), mixed invasive ductolobular breast cancer (MIDLC) (24), invasive ductolobular carcinomas (IDLC) (25), among others. Despite increasing appreciation of MDLC disease (26, 27), it remains understudied, leading to diagnostic and prognostic uncertainty and clinical management challenges for the patients with this elusive disease. Although there are limited reports focused on MDLC, a few seminal studies have provided insights into its complex pathobiology. Immunohistochemical (IHC) studies have confirmed while that ductal regions of MDLC retain membranous E-cadherin, while lobular regions either lose or display aberrant E-cadherin IHC staining (28–30). Notably, the current MDLC diagnosis guidelines from WHO only consider the presence of mixed histological patterns and lack recommendations for support from E-cadherin IHC. However, a recent study suggests that the use of E-cadherin IHC can improve diagnostic accuracy of lobular tumor components in ILC and MDLC tumors (31). Clinical studies have often yielded contradictory findings (20–22, 27, 28, 30, 32, 33), but overall, they suggest that MDLC is clinically more similar to pure ILC than IBC-NST (23, 34, 35). These discrepancies may be due to the nonstandardized and fluid definition of MDLC disease, limited consideration of its underlying tumor regions and their distinct features and proportions, and the vast intratumoral heterogeneity of this disease (23, 27, 31, 36). Findings from previous reports point to several subtypes of MDLC tumors based on their histological and E-cadherin staining patterns: 1) intermixed/intermingled subtype where the E-cadherin positive ductal and E-cadherin negative/aberrant lobular tumor cells appear closely intermixed/intermingled together (25, 37), 2) collision subtype where the E-cadherin positive ductal and E-cadherin negative/aberrant lobular tumor regions appear in spatially distinct regions which may collide into each other forming intermixed regions (25, 30), and 3) lobular-like IBC-NST subtype where there are histologically distinct ductal and lobular tumor regions but the overall tumor is predominately E-cadherin positive (38).

Moreover, molecular studies with profiling of individual tumor regions have been limited. The largest molecular study on MDLC tumors was from the TCGA group (5), where 88 MDLC were classified into lobular-like and ductal-like groups based on transcriptomic and mutational similarity to ILC and IBC-NST tumors. The lobular-like group comprised ~25% of MDLC cases and showed enrichment of mutations in and lower expression of CDH1 gene. However, these findings may not have any biological relevance as these samples lacked comprehensive annotation and may have been incorrectly defined as MDLC as well as the study lacked individual tumor region analysis. Additional molecular studies have largely focused on analysis of somatic genetic alterations (27, 30, 37). While these have provided valuable insights into the tumorigenesis pathways and distinct mutational landscape of ductal and lobular tumor regions (25, 30), comprehensive studies delineating transcriptomic differences between these distinct regions have not been performed.

To fill this knowledge gap, we employed spatially resolved transcriptomic, genomic, and single-cell profiling approaches to comprehensively investigate MDLC and highlight its intratumoral heterogeneity and complexity. We delineate the biological and intrinsic subtype differences between its underlying ductal and lobular tumor regions. Notably, our findings highlight that in some MDLC cases, distinct tumor regions with distinct biology may pose distinct progression risk. Hence, histopathological and molecular profiling of MDLC and its individual tumor regions may inform treatment decisions.

Results

Patient Specimens and Their Clinicopathologic Characteristics.

A panel of three molecular pathologists identified invasive breast cancer cases diagnosed as “mixed invasive ductal and lobular carcinoma” and shortlisted three cases for this pilot study. All three cases were reported as ER+, HER2 low/negative and their ages ranged from 46 to 63 y (Fig. 1A and Dataset S1A). Assessment of H&E and E-cadherin/p120 dual staining confirmed that each case exhibited histomorphic features in line with the latest MDLC definition i.e., admix of E-cadherin negative lobular tumor regions with at least 10 to 90% of E-cadherin positive ductal tumor regions (17) (Dataset S1B). Importantly, overall, these cases exhibited collision of spatially distinct ductal and lobular tumor regions (25, 30) along with some regions of intermixing and hence could be broadly defined as MDLC of collision type. These cases were particularly amenable for downstream multiomic profiling compared to those that exhibited more intermingling or intermixing with close proximity of histologic tumor regions. Clinical diagnostic and treatment timelines of these cases, source of their tumor tissue used in this study, and their histomorphic features are illustrated in Fig. 1A. Both MDLC-2 and MDLC-3 eventually developed recurrent disease resulting in patient death in the case of MDLC-2. As of the last follow-up in 2022, in both MDLC-3 and MDLC-1 cases, the patients were alive with tumors, and in the case of MDLC-1, the patient was free of recurrent disease.

Fig. 1.

Digital spatial profiling of MDLC ductal and lobular tumor regions. (A) The Top panel shows clinical timelines of the studied cases including the time of core biopsies, surgical interventions, treatments, metastasis, and last known vital status. The source of the tumor tissue used in this study is highlighted with *in red color. The Bottom panels show high (40×) magnification H&E and Ecad/p120 IHC images highlighting histomorphic and E-cadherin features of ductal and lobular regions across each case. ER: Estrogen receptor, PR: Progesterone receptor, and HER2: Human epidermal growth factor receptor 2. (B) E-cadherin (Ecad) and Pan-CK (CK) immunofluorescence staining of a representative case (MDLC-3). Selected regions of interest (ROIs) are shown with dotted blue (for ductal) and red (for lobular). (C) Segmentation of ductal and lobular cells using Ecad and CK staining masks. (D) Segmented areas containing only ductal or lobular tumor cells defined using ductal and lobular masks illustrated in panel C. (E) E-cadherin staining intensity in representative ductal and lobular regions profiled across cases.

Whole-Transcriptome Spatial Profiling of Ductal and Lobular Regions.

To better understand the biological heterogeneity and transcriptomic features of ductal and lobular regions, we performed GeoMx whole-transcriptome digital spatial profiling (DSP) of these three MDLC cases. In total, we analyzed 26 ROIs with at least 6 ROIs per case. Selected ROIs from a representative case (MDLC-3) are shown in Fig. 1B. We segmented ductal and lobular areas using E-cadherin and Pan-CK staining (Fig. 1C). This enabled profiling of pure ductal and lobular areas from heterogeneous regions that were predominantly of one histology or a mixture of both (Fig. 1D). E-cadherin expression in segmented ductal and lobular regions showed expected results (Fig. 1E). All profiled regions had high RNA sequencing saturation (>94%) and low noise (Q3 count > Negative probe count). However, three regions had low read count (<0.3 million) and were excluded from further analysis (SI Appendix, Fig. S1 A–C). Our final DSP dataset is composed of 13 ductal and 13 lobular regions.

Transcriptomic Clusters, Intrinsic Molecular Subtypes, and Differential Transcriptomic Features.

Exploration using top variable genes revealed that regions from each tumor clustered based upon their histology (Fig. 2A and SI Appendix, Fig. S1D). MDLC-3 ductal regions showed the most distinct mRNA expression compared to all other regions. PAM50 signature based molecular subtyping analysis showed both inter- and intra-tumor heterogeneity of prognostically relevant intrinsic molecular subtypes (IMS). MDLC-2 ductal regions were a mix of Luminal A and HER2-Enriched subtypes while lobular regions were HER2-Enriched (but stained negative for HER2 IHC—Dataset S1A). MDLC-1 ductal and lobular regions were both of luminal IMS, with the former being Luminal A and the latter being Luminal B. Finally, in line with mRNA expression patterns, MDLC-3 ductal and lobular regions showed stark IMS differences (former Luminal A and later Basal-Like). MDLC-3 ductal region was the only triple-negative breast cancer (TNBC) or Basal-Like tumor (and had low ESR1, PGR, and ERBB2 mRNA expression—SI Appendix, Fig. S1E). We further confirmed the TNBC status of MDLC-3 ductal regions using ER and HER2 IHC (Dataset S1A). Overall, our transcriptomic clustering and IMS analysis suggests that ductal and lobular tumor regions within MDLC may have distinct transcriptomic features and molecular subtypes.

Fig. 2.

Molecular subtypes and differentially expressed genes (DEGs) in ductal and lobular tumor regions. (A) Profiled ductal and lobular tumor regions separated by consensus clusters. All regions are annotated by cases, histology, and consensus clusters. Heatmap shows expression of top variable genes and followed by PAM50 signature scores for each subtype across ductal and lobular regions. (B) DEGs between ER+ ductal and lobular regions. The Top panel shows the DEG heatmap, while the Bottom panel shows the volcano plot along with the number of significantly up-regulated (N = 168) and down-regulated (N = 198) DEGs in ductal vs. lobular regions.

To investigate global transcriptomic differences between ductal and lobular regions across all cases, we performed differential gene expression analysis in a pan-patient manner. To avoid biases caused by fundamental differences between ER+ and ER− disease, we excluded MDLC-3 ER−/TNBC ductal regions from this analysis and later downstream analyses (SI Appendix, Fig. S2C and Dataset S5). In total, we identified 168 significantly up-regulated and 198 significantly down-regulated DEGs in the lobular compared to ductal regions across all cases (Fig. 2B and Dataset S5). These pan-patient DEGs showed strong overlap with patient-specific DEGs (SI Appendix, Fig. S2 A–C) indicating that our pan-patient analysis captured most if not all strong global differences between ductal and lobular tumor regions in each case.

To assess the transcriptomic similarity of MDLC ductal and lobular regions to pure counterparts, i.e., IBC-NST (IDC) and ILC, we compared MDLC ductal vs. lobular region DEGs with those from the comparison of TCGA ER+ IBC-NST (N = 327) vs. ILC (N = 113), respectively (SI Appendix, Fig. S2D). We found KLK11, KLK10, NRAP, SHROOM1, BTG2, and PDK4 were up-regulated, while CDH1, DCD, SERPINA1, and CPB1 were down-regulated in both comparisons (i.e., MDLC ductal vs. lobular tumor regions and IBC-NST vs. ILC tumors). This overlap confirmed the well-known downregulation of CDH1 as a marker of the lobular phenotype as well as other potential markers.

Differential Enrichment of ER and MYC Signatures and Recurring Biological Themes.

To understand the distinct biology of ductal and lobular regions, we scored each DSP region for hallmark signatures (39). Twenty signatures were differentially enriched between ductal and lobular regions with P-value ≤ 0.05 (Fig. 3A and Dataset S6). Of note, many of these pathways can also be found differentially enriched between ER+ IBC-NST and ER+ ILC in TCGA – e.g., Kras and TNFA signaling as well as Hypoxia are among significantly enriched pathways in ER+ ILC TCGA samples, whereas Myc targets are enriched in ER+ IBC-NST TCGA samples (Fig. 3B). In line with differential enrichment of Estrogen Response and Myc Targets signatures, our transcription factor (TF) signature analysis revealed enrichment of ESR1/FOXA TF signatures in lobular and MYC TF signatures in ductal regions (SI Appendix, Fig. S3).

Fig. 3.

Distinct hallmark and biological signatures in ductal and lobular tumor regions. (A) Differentially enriched cancer hallmark signatures. The heatmap shows GSVA scores of each differentially enriched pathway in ductal vs. lobular tumor regions. Pathways with bold red and blue names are the ones that also were significant in the scRNAseq dataset (SI Appendix, Fig. S8A). Enrichment score (ES) for a pathway shows mean differences of GSVA score for that pathway in lobular—ductal regions and the −log10 of FDR-adjusted P-value (P-val) shows the statistical significance of enrichment based on the T test. (P-val score definitions: 4 = P ≤ 0.0001, 3 = P ≤ 0.001, 2 = P ≤ 0.01, and 1 = P ≤ 0.05). (B) The heatmap shows GSVA scores of differentially enriched signatures between TCGA ER+ IBC-NST vs. ILC that overlap with MDLC ductal vs. lobular tumor region hallmark signatures in panel A. (C) Recurrent biological themes in ductal and lobular tumor regions. The Left panel shows the Jaccard similarity between all signatures. Signatures are grouped based on their Jaccard similarity into signature clusters. Each cluster is associated with a broad biological theme (shown in the Middle). The Right panel shows the mean GSVA score of each signature cluster in ductal and lobular regions. (D) E-cadherin (ECad) and pan-cytokeratin (pan-CK26) staining in representative image of adjacent ductal and lobular tumor regions. (E) Representative immunofluorescence images (Right panel) and single-cell quantification (Left panel) of cell cycle and senescence biomarkers from all ER+ ductal and lobular regions. Statistical significance is determined using Student’s t test (****P ≤ 0.0001). ECM: extracellular matrix, DSB: double-stranded break, HR: homologous recombination.

To determine recurring biological themes, we expanded our differential enrichment analysis to include all GO (40), REACTOME, (41) and KEGG (42) genesets. Signatures differentially enriched between ductal and lobular regions with P-value ≤ 0.05 were further assessed for robustness against background permutations. Robust signatures were grouped into signature clusters based on the similarity of overlapping genes in ductal and lobular regions. All signature clusters were further grouped into broad biological themes based on the underlying biological terms (Fig. 3 C, Left panel). The mean score of all signature clusters in ductal and lobular regions is shown in Fig. 3 C, Right panel. TP53 & senescence, collagen & extracellular matrix (ECM), MAPK/ERK targets (in line with KRAS signatures shown in Fig. 3A) were enriched in Lobular regions, while double-stranded break & homologous repair (DSB/HR), mTOR signaling (in line with Mtorc1/Mtor signatures shown in Fig. 3A) were enriched in ductal regions among other signatures (Dataset S7).

Notably, several genes including MYC, G3BP1, and SNRPD1 overlapped between Myc Targets V1 and G2m Checkpoint signatures (both enriched in ductal regions). Both of these signatures had no overlapping genes with DNA damage and repair (D1: DBS/HR) signatures (SI Appendix, Fig. S4A) indicating that Myc-related signatures in ductal regions are associated with proliferation. On the other hand, several genes associated with senescence and cell death, including CDKN1A/2B/2C, TP53, and BCL6, were found to be shared between the L1: Cell cycle and L2: TP53 & Senescence signatures (SI Appendix, Fig. S4B). This indicates negative regulation of cell cycle in lobular regions via TP53 signaling and cell cycle inhibitors. Key genes associated with these notable and other biological signatures are highlighted in SI Appendix, Fig. S4C. In summary, we identified differential enrichment of various hallmark oncogenic and biological signatures between ductal vs. lobular regions including cell cycle (Myc and G2m genes in ductal, and TP53 signaling and senescence or cell cycle arrest genes in lobular), immune, collagen & ECM and metabolism among others.

To further investigate some of the notable distinctions in gene expression signatures between ductal and lobular tumor regions, we performed multiplexed immunofluorescence to obtain single-cell proteomic measurements of key cell cycle and senescence effectors in ductal and lobular tumor cells, identified by staining for pan-cytokeratin 26 and E-cadherin (Fig. 3D and Dataset S12). Individual cells within ductal regions exhibited significantly higher expression of proliferative markers including cMYC, cyclin E2, and cyclin A1, while lobular regions had significantly higher expression of the senescence biomarker p16 (Fig. 3E). These findings confirmed the activation of Myc signaling and proliferative pathways in ductal and activation of cell cycle arrest/senescence in lobular tumor regions.

Genomic Landscape and Clinically Relevant Genomic Alterations.

To explore the genomic landscape of ductal and lobular regions, we employed the FDA-approved targeted MSK-IMPACT panel. We microdissected the ductal and lobular tumor regions from FFPE blocks of each case and sequenced them at a median depth of 416× (range 287 to 563) (SI Appendix, Fig. S5A). In all cases, at least one pathogenic mutation was detected in known breast cancer driver genes, including CDH1, PIK3CA, TP53, GATA3, KMT2C, MAPK31, ARID1A, and CBFB, in both ductal and lobular regions, as depicted in Fig. 4A. Furthermore, all pathogenic mutations had a high cancer cell fraction (SI Appendix, Fig. S5B).

Fig. 4.

Overview of genomic alterations in ductal and lobular tumor regions. (A) Somatic mutations identified using the MSK-IMPACT panel in ductal and lobular components across samples. Gene names are followed by amino acid change due to mutation. Genes in black font had somatic mutations in breast cancer driver genes while those in gray color had passenger/nondriver mutations. (B) Somatic copy number (CN) changes in ductal and lobular components across samples. For each case, genome-wide ductal and lobular CN profiles are shown. The black line shows the total CN, the red line shows the minor allele CN, and off-white regions show aneuploidy (change in minor allele CN including LOH). (C) Mutational signatures in MDLC-2 and MDCL-3 ductal and lobular tumor regions. Pie charts show the percentage similarity of mutational patterns of each tumor component with established mutational patterns (associated with different means/definition). SBS: Single Base Substitution. (D) Predicted disease evolution. Tumors with overlapping mutations were linked to common root node indicating common progenitor or precursor neoplasm (N). Those without common root node likely indicate evolution from independent precursors. Key mutations and events (CDH1 promoter methylation and amplifications) are shown on the branches.

Loss of E-cadherin is an established driver of lobular tumorigenesis (43). Typically, it results from biallelic CDH1 inactivation, through a combination of truncating mutation and loss of heterozygosity (LOH) (5). All lobular regions exhibited loss of E-cadherin, with both MDLC-1 and MDLC-3 harboring CDH1 truncating mutations and LOH (Figs. 1E and 4B and SI Appendix, Fig. S1E). The lobular region of MDLC-2 did not have CDH1 genomic alterations but showed epigenetic inactivation via methylation of its promoter. A putative passenger mutation in CDH1 was also observed in the MDLC-3 ductal region, which had no effect on mRNA or E-cadherin levels (Fig. 1E and SI Appendix, Fig. S1E). The loss of E-cadherin is consistent with the downstream downregulation of Adherens Junction signatures in all lobular regions (Fig. 3A). Hotspot PIK3CA mutations, which are early oncogenic driver events frequently observed in ER+ BC (44), were present in all cases except MDLC-3. The MDLC-1 ductal region had mutations in GATA3 – a critical transcriptional binding partner of ER-alpha (45) and CBFB – which forms complex with RUNX1 to suppress NOTCH signaling (46). The MDLC-3 ductal region had loss of TP53, an event associated with more aggressive types of breast cancer such as TNBC (47) and consistent with its basal-like intrinsic subtype. MDLC-2 ductal region had truncating mutation in KMT2C, an important regulator of ER-alpha activity (48). The MDLC-3 lobular region had mutations in ATRX and ARID1A, two chromatin remodeling and DNA repair genes (49). In addition, while MDLC-1 and MDLC-3 ductal and lobular regions did not share any somatic clonal mutations, MDLC-2 ductal and lobular regions shared several somatic clonal mutations in PIK3CA and MAPK31 genes indicating shared tumor origin.

To investigate the CN landscape of ductal and lobular regions, we performed CN analysis using FACETS (50) (Fig. 4B). All regions exhibited breast cancer characteristic chromosome 1q gains and 16q losses (47). Overall, ductal tumor regions showed more frequent aneuploidy events than lobular tumor regions. MDLC-3 ductal region showed the most aberrant CN profile consistent with its basal/TNBC etiology (47). Moreover, chromosome 22 aneuploidy was only observed in ductal tumor regions. Notably, both histologic regions of MDLC-3 and ductal region of MDLC-2 had NOTCH2 gains or amplifications (SI Appendix, Fig. S5C).

Mutational Processes Driving Tumor Evolution.

To determine the mutational processes underlying the evolution of ductal and lobular tumor regions within MDLC, we performed mutational signature analysis of all cases (Fig. 4C). Both histologic regions in MDLC-2 were associated with an “aging” signature. In MDLC-3, the ductal region was associated with homologous repair deficiency (HRD) while the lobular regions with APOBEC signature. Tumors with APOBEC signatures have a higher frequency of mutations in chromatin remodeling and DNA repair genes (49), as seen in the MDLC-3 lobular region. We did not identify any signatures for MDLC-1 ductal and lobular regions. The putative evolution of each MDLC case and its histologic region based upon the mutations, CDH1 alterations, and mutational signatures is shown in Fig. 4D. Notably, the MDLC-2 ductal and lobular regions not only shared clonal pathogenic (PIK3CA) and passenger (MST1R and MAP3K1) mutations but also the aging mutational signature. Loss of function CDH1 alterations were the hallmark of all lobular regions across cases.

Single-Cell RNA Sequencing and Identification of Ductal and Lobular Populations.

To explore the heterogeneity of MDLC at a single-cell level, we performed scRNAseq sequencing of a fresh tumor specimen from MDLC-1. The resulting scRNAseq comprised of 4,671 cells with high sequencing saturation (78.4%) and mean reads per cell (~90,000) (SI Appendix, Fig. S7A). Cell-type annotation based on canonical markers (see details in SI Appendix, Methods) revealed various tumor and non-tumor cell types (SI Appendix, Fig. S7B). scRNAseq inferred CN revealed that only the epithelial tumor cells had aneuploid CN status further confirming their nature (SI Appendix, Fig. S7C). For the downstream analyses, we only focused on the tumor cell cluster. Clustering of this revealed two transcriptionally distinct tumor cell populations with similar cell cycle profiles; however, we were unable to annotate them as either ductal or lobular tumors using CDH1 mRNA expression (Fig. 5A). Notably, while the MDLC-1 lobular regions showed loss of E-cadherin protein (Fig. 1E), the CDH1 p.Q23* truncating mutation in these regions (Fig. 4A) didn’t alter CDH1 mRNA expression (SI Appendix, Fig. S1E). Previous studies suggest that some protein-truncating mutations can escape nonsense-mediated decay and thereby maintain gene expression levels but lack a functional protein (51–53). Indeed, similar effects of Q23 mutations were seen in the TCGA cohort, where CDH1 mRNA levels in Q23* mutant tumors were similar to those in CDH1 WT tumors, while the E-cadherin protein levels in Q23* mutant tumors were similar to the tumors with all other CDH1 truncating mutations (SI Appendix, Fig. S6) (51–53).

Fig. 5.

Single-cell heterogeneity of ductal and lobular tumor populations. (A) Tumor populations in the MDLC-1 scRNAseq dataset (Left) and assessment of CDH1 expression in the MDLC-1 scRNAseq dataset (Middle and Right plots). Statistical significance was evaluated using the T test. (B) Deconvolution of scRNAseq using DSP-derived ductal and lobular signatures. GSVA scores for ductal and lobular signatures are shown on the Right panel. (C) Subclustering and identification of ductal and lobular subpopulations (Left) with unique marker genes (Right). (D) Scoring spatial regions using GSVA of subcluster marker genes. (E) Mapping of single-cell populations to spatial regions based on GSVA scores in panel D.

To identify ductal and lobular single populations, we derived ductal and lobular transcriptome signatures (Dataset S5) from differential gene expression analysis of the MDLC-1 DSP ductal vs. lobular tumor regions and used these signatures to score each single cell in MDLC-1 scRNAseq dataset. Using this deconvolution approach, we successfully annotated ductal and lobular tumor populations in the scRNAseq dataset (Fig. 5B). scRNAseq inferred CN profiles of ductal and lobular cells revealed various breast cancer–specific aberrant/aneuploid events in chromosomes 1, 8, 11, 16, 18 among others (SI Appendix, Fig. S7E). Overall, the CN profiles of ductal tumor cells were more aberrant than those of lobular tumor cells. Ductal and lobular tumor regions DEGs in scRNAseq and DSP datasets showed significant overlap (SI Appendix, Fig. S7 D and F). Differential enrichment of several hallmark signatures and recurring biological themes in ductal and lobular regions from DSP data was recapitulated in our scRNAseq data (SI Appendix, Fig. S8 A and C). Notably, we saw enrichment of Estrogen and senescence signatures in lobular and Myc signature in ductal tumor cells in both DSP and scRNAseq datasets.

Single-Cell Heterogeneity and Mapping of Single Cells to Spatial Regions.

To investigate single-cell heterogeneity of ductal and lobular populations, we increased the resolution of clustering to identify subclusters. Each subcluster had unique marker genes (Fig. 5C) and several of these also showed differential enrichment of hallmark signatures and recurring biological theme clusters (SI Appendix, Fig. S8 B and D), indicating intratumor heterogeneity within ductal and lobular tumor populations. To investigate whether single-cell transcriptomic heterogeneity was associated with spatial heterogeneity, we mapped scRNAseq subclusters to DSP regions. To assess the similarity between each region and subcluster, we scored all regions for subcluster marker genesets using GSVA (Fig. 5D). Each region was mapped to a subcluster for which it had most similarity (i.e., highest GSVA score) (Fig. 5E). The top five marker genes of each scRNAseq subcluster (SI Appendix, Fig. S7G) are shown beside each subcluster and spatial region link. These findings perhaps suggest that the single cells in each subcluster came from a spatially distinct niche in the tumor.

Discussion

Despite an increasing incidence of MDLC (26, 27), this special subtype of breast cancer remains poorly understood, with limited knowledge about its overall biology and that of its underlying ductal and lobular tumor regions, as well as their clinical implications.

Spatially resolved and single-cell profiling techniques have revolutionized our understanding of physiological and pathological processes and tissue architecture (54). Efforts such as Human Cell Atlas (55) have successfully uncovered the major cell types and their spatial organization in both normal breast tissue (56, 57) and breast cancer (58). Spatial and single-cell profiling of primary breast cancer subtypes and metastatic disease have been instrumental in improving our understanding of key molecular features and their prognostic value (59–63). However, to date, genome-wide spatial and single-cell investigation of MDLC has not been reported. To address this gap, we employed spatially resolved transcriptomics, genomics, and single-cell profiling to investigate the biology of individual tumor regions within MDLC of collision type. Our study confirmed various previously reported observations of genetic inactivation of CDH1 and concomitant loss of E-cadherin in lobular regions (25, 28–30) and the presence of distinct driver alterations in individual tumor regions (25, 30). We also uncovered findings that highlight the vast heterogeneity within MDLC, several of which may be indicative of distinct prognosis, hence highlighting the significance of reporting this heterogeneity in clinical diagnosis. Ductal and lobular tumor regions had distinct biological signatures, in particular cell cycle arrest/senescence, ER and MYC signaling, and IMS. The IMS distinctions were most prominent in the case of MDLC-3, which was clinically reported as ER+ tumor but molecular profiling revealed that the underlying ductal region was ER−/TNBC with basal-like IMS while the lobular region was ER+ with luminal-A IMS. Notably, both ER− status and basal-like IMS are associated with worse prognosis and outcome than ER+ and luminal IMS (64, 65). In this case, the patient had brain and lung metastases, both of which were also ER−, suggesting progression of the primary ER− ductal tumor. Another notable finding was enrichment of cell cycle arrest/senescence and ER signaling signatures in lobular while MYC signatures in ductal tumor regions. Multiplex immunofluorescence staining of various effector proteins confirmed that Myc and Cell proliferation signaling were active in ductal tumor cells, while cell cycle arrest/senescence signaling was active in lobular regions. Further analysis at the single-cell RNA level revealed that ER signatures were enriched in only one lobular subcluster (L2), highlighting molecular heterogeneity even within individual histologically distinct tumor regions. The majority of ILC are strongly ER+ and respond well to endocrine therapies, while they are mostly less proliferative, responding poorly to chemotherapy (5, 8, 15, 66, 67). Whether this is also true for lobular regions within MDLC is an important question and must be further investigated. Perhaps, future trials can study whether tumors with predominantly ductal vs. lobular MDLC respond differently to endocrine and chemotherapies.

Development of robust mutational signature analysis tools such as signature multivariate analysis (SigMa) has allowed accurate identification of mutational signatures associated with such HRD and APOBEC defects from targeted sequencing panels such as MSK-IMPACT (68–72). Utilization of the SigMa algorithm in our study revealed that individual tumor regions with distinct alterations may share common (Aging in MDLC-2) or distinct (APOBEC or HRD in MDLC-3) mutational signatures. Collectively, our results indicate ductal and lobular regions in MDLC-2 have a shared origin, while MDLC-3 is a clear collision of two independent tumors. The etiology of MDLC-1 is questionable – while there are no shared mutations, we note that this interpretation is based upon a small panel of genes measured by the MSK-IMPACT panel with strict filtering. We cannot exclude the possibility of shared mutations in other cancer-related genes not covered by the MSK-IMPACT panel. Larger efforts using whole genome profiling of more MDLC tumors are needed to comprehensively investigate the mutational signatures and origins of ductal and lobular tumor regions in mixed tumors.

The presence of distinct clinically actionable alterations in individual tumor regions within MDLC is of clinical relevance and has been previously reported (30). Tumor regions with distinct alterations can complicate clinical management. Alterations in components of ER signaling (such as GATA3, KMT2C) as well as NOTCH signaling (CBFB, NOTCH2) can lead to endocrine resistance (73–75), while alterations in DNA repair response genes (ARID1A, ATRX) may confer increased sensitivity to chemotherapy and other DNA repair pathway targeting drugs including PARP inhibitors (76, 77). Future studies will need to address the important question whether treatment decisions for patients with MDLC benefit from considering the distinct alterations in individual tumor regions.

For a long time, it was unclear whether ductal and lobular regions within MDLC share any molecular similarity with their pure counterparts (i.e., ILC and IBC-NST tumors). To fill in the gap, in our study, we compared the molecular features of individual tumor regions within MDLC with their pure counterparts (i.e., ILC and IBC-NST tumors) in the TCGA cohort (5) and identified a set of shared markers of lobular phenotype. There was some but limited overlap of the DEGs between ductal and lobular tumor regions in MDLC cases with differential gene expression in ER+ ILC vs. ER+ IBC-NST in the TCGA cohort. This could be due to technical reasons such as bulk sequencing of frozen tumor samples (in TCGA) vs. spatial sequencing of epithelial tumor regions from FFPE tissues (in this study). Furthermore, this also could also be reflective of true biology i.e., unique gene expression patterns in MDLC tumor regions compared to pure ILC and IBC-NST tumors. As expected, one of overlapping genes that was down-regulated in both MDLC lobular tumor regions and ILC tumors was CDH1, an important driver of lobular tumorigenesis (43). This primarily occurs via genetic inactivation of CDH1 in ILC tumors (5), similar to what was observed in the case of lobular regions in MDLC-1 and MDLC-3. However, in the lobular region of MDLC-2, CDH1 promoter methylation was observed. CDH1 promoter methylation is also associated with epithelial-to-mesenchymal transition (EMT), however, EMT is not a typical feature of lobular tumors and is primarily associated with more complex subtypes of breast cancer such as metaplastic and claudin-low subtypes (78–81). EMT markers (78) were not consistently up-regulated in MDLC-2 lobular region, hence ruling out EMT-associated CDH1 promoter methylation (SI Appendix, Fig. S1F). This is an intriguing observation, suggesting that the pathogenesis of lobular regions within MDLC may occur via both genetic and epigenetic inactivation of CDH1, unlike ILC where genetic inactivation of CDH1 is predominant (4, 25, 82–84). A recent study revealed the synthetic lethality in E-cadherin deficient breast cancers with inhibition of ROS1, a tyrosine kinase (85), and trials are ongoing testing efficacy of ROS1 inhibitors in ER+ ILC disease. Future studies could assess ROS1 inhibitors or other drugs in treating patients with predominantly lobular MDLC disease. Additional studies are also warranted to investigate whether E-cadherin deficient tumors with distinct etiology, such as CDH1 epigenetic inactivation, have distinct biology potentially associated with differential treatment response.

Our study focused primarily on profiling MDLC of collision type as these were the most amenable for our molecular profiling approaches i.e., we could clearly annotate ductal and lobular tumor regions based on histological and E-cadherin staining patterns and cleanly capture RNA/DNA from these individual regions. Notably, we did attempt to profile ductal and lobular tumor cells separately from intermixed regions identified in MDLC-2; however, the resulting data were of poor quality due to low tumor cellularity. Although our study presents a unique analysis of MDLC of collision type, it has some limitations, including a small sample size and lack of analysis of other subtypes [such as lobular-like IBC-NST (38) and intermixed/intermingled MDLC (25, 37)] and tumor microenvironment of individual tumor regions within MDLC. Future studies, using rapidly improving spatial sequencing technologies with increased single-cell capabilities will be necessary to validate and expand our observations in larger and more diverse cohorts of MDLC. Generating a comprehensive MDLC cohort for future studies may not be an easy task, given the challenges associated with the diagnosis of this mixed entity. However, the use of E-cadherin IHC in histological assessments and evaluation by a panel of pathologists can ensure the generation of a high-confidence and well-annotated MDLC cohort for future studies. Moreover, the application of digital pathology and AI will make this more feasible (86). These tools will not only improve diagnostic accuracy (87) but will also allow scalable annotation of ductal and lobular tumor regions across heterogenous tumor samples (88).

In summary, our study provides deeper insights into the molecular heterogeneity of MDLC and its underlying ductal and lobular tumor regions using spatially resolved transcriptomic, genomic, and single-cell profiling. We identified distinct biological signatures and IMS between these regions that may have important implications for patient prognosis and clinical management. Our findings also highlight the need for comprehensive histopathological and molecular profiling of individual tumor regions within MDLC to better understand its progression and response to cytotoxic, endocrine, and other targeted therapies.

Methods

Sample Collection.

Informed patient consent was obtained for tissue collection, and institutional review board approval was obtained from the University of Pittsburgh prior to initiation of the study. Patients diagnosed with “mixed invasive ductal and lobular” were identified from the UPMC breast cancer registry and formalin-fixed paraffin-embedded (FFPE) blocks requested from Pitt Biospecimen Core. Four-um-thick FFPE sections were used for H&E, ER-alpha, and E-cadherin IHC staining. A panel of pathologists reviewed the H&E and E-cadherin staining to confirm MDLC diagnosis in line with recent WHO guidelines (17) (10 to 90% of ILC and at least 10% IBC-NST) and shortlisted MDLC cases amenable for multiomic profiling, i.e., those with good separation of ductal and lobular tumor regions to avoid spill over and contamination of individual ductal and lobular tumor regions by neighboring lobular and ductal tumor cells, respectively. Hence, we selected three collision-type MDLC cases that showed both clear separation of ductal and lobular tumor regions along with some intermixed/intermingled regions. These included cases MDLC-1, MDLC-2, and MDLC-3 (clinical features and pathology review notes are shown in Datasets S1A and S1B). All cases underwent NanoString digital spatial, MSK-IMPACT mutation, and CDH1 promoter methylation profiling. Fresh tumor was only available for MDLC-1 and was utilized for single-cell RNAseq. See details in SI Appendix, Methods.

Multiomic Profiling.

DSP was performed using the NanoString GeoMX whole-transcriptome atlas platform. Mutation profiling was performed using the MSK-IMPACT targeted NGS panel. CDH1 promoter methylation was assessed using the ddPCR system. Finally, single-cell RNA sequencing was performed using the 10× genomics platform based on 3′end chemistry v3. Please see SI Appendix, Methods for details.

Multiplexed Immunofluorescence.

Tissue sections were prepared and iteratively stained as described previously (89). Briefly, tissue sections were baked overnight at 60 °C, deparaffinized with xylene, and rehydrated by decreasing ethanol washes, followed by proprietary antigen retrieval optimized for multiple antigens (Leica Biosystems). Tissues were incubated in 0.3% Triton X-100 in PBS for 10 min followed by blocking in 10% (w/v) donkey serum and 3% (w/v) bovine serum albumin (BSA) in PBS for 1 h at room temperature. Primary antibodies were diluted in blocking solution and incubated for 1 h at room temperature (anti-E-cadherin, Cell Signaling Technology, 96743; anti-pan-cytokeratin, Millipore Sigma, C5992; anti-c-Myc Alexa Fluor 647, Abcam, ab190560; anti-p16 Alexa Fluor 647, Abcam, ab192054; anti-cyclin E2 Alexa Fluor 488, Abcam, ab207336; anti-cyclin A1, R&D Systems, MAB7046; anti-p21, R&D Systems, AF1047; anti-phospho-H2AX Alexa Fluor 555, Abcam, ab206900). In-house primary antibody fluorescent conjugates were generated using the Alexa Fluor 647 labeling kit (Invitrogen, A20186) or the Alexa Fluor 555 labeling kit (Invitrogen, A20187). Nuclear labeling was performed using Hoechst (2 μg/mL, Invitrogen, H3570) for 10 min at room temperature. Imaging was performed in 50% glycerol (v/v) in PBS. After imaging, dye inactivation was performed in an alkali solution containing H₂O₂ for 15 min with agitation followed by a PBS wash. Samples were reimaged following dye inactivation to measure residual fluorescence. Additional rounds of antibody staining and imaging were performed as described above, starting with primary antibody incubation. Image acquisition, flat-field correction, autofluorescence removal, and registration were performed using the Cell DIVE Imager (Leica Biosystems). Single-cell segmentation was performed using Cellpose (90) and intensity measurements were extracted using scikit-image (91). Measurements greater than 1.5× the interquartile range were identified as outliers and removed from analysis.

Figures and Illustrations.

DSP spatial images were exported from GeoMx® Data Analysis Suite (DSPDA) and used directly in the manuscript without graphical modifications. Most bioinformatic analysis visualizations were generated using R. These visualizations were aesthetically improved in Microsoft PowerPoint to produce the final figures. Illustrations were prepared using BioRender.com.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Dataset S06 (XLSX)

Dataset S07 (XLSX)

Dataset S08 (XLSX)

Dataset S09 (XLSX)

Dataset S10 (XLSX)

Dataset S11 (XLSX)

Dataset S12 (XLSX)

Data, Materials, and Software Availability

Data and code generated in this study are shared in Datasets S1–S12, Mendeley (https://data.mendeley.com/datasets/btv7g7n9ys/1) (92), and GitHub (https://github.com/osamashiraz/MDLC_Spatial_Analysis_2023) (93). All other data are included in the manuscript and/or supporting information.