Comparative analysis of the molecular subtype landscape in canine and human mammary gland tumors.

Breast cancers in humans belong to one of several intrinsic molecular subtypes each with different tumor biology and different clinical impact. Mammary gland tumors in dogs are proposed as a relevant comparative model for human breast cancer; however, it is still unclear whether the intrinsic molecular subtypes have the same significance in dogs and humans. Using publicly available data, we analyzed gene expression and whole-exome sequencing data from 158 canine mammary gland tumors. We performed molecular subtyping using the PAM50 method followed by subtype-specific comparisons of gene expression characteristics, mutation patterns and copy number profiles between canine tumors and human breast tumors from The Cancer Genome Atlas (TCGA) breast cancer cohort (n = 1097). We found that luminal A canine tumors greatly resemble luminal A human tumors both in gene expression characteristics, mutations and copy number profiles. Also, the basal-like canine and human tumors were relatively similar, with low expression of luminal epithelial markers and high expression of genes involved in cell proliferation. There were, however, distinct differences in immune-related gene expression patterns in basal-like tumors between the two species. Characteristic HER2-enriched and luminal B subtypes were not present in the canine cohort, and we found no tumors with high-level ERBB2 amplifications. Benign and malignant canine tumors displayed similar PAM50 subtype characteristics. Our findings indicate that deeper understanding of the different molecular subtypes in canine mammary gland tumors will further improve the value of canines as comparative models for human breast cancer.

Tailored graphical lasso for data integration in gene network reconstruction.

BACKGROUND: Identifying gene interactions is a topic of great importance in genomics, and approaches based on network models provide a powerful tool for studying these. Assuming a Gaussian graphical model, a gene association network may be estimated from multiomic data based on the non-zero entries of the inverse covariance matrix. Inferring such biological networks is challenging because of the high dimensionality of the problem, making traditional estimators unsuitable. The graphical lasso is constructed for the estimation of sparse inverse covariance matrices in such situations, using [Formula: see text]-penalization on the matrix entries. The weighted graphical lasso is an extension in which prior biological information from other sources is integrated into the model. There are however issues with this approach, as it naïvely forces the prior information into the network estimation, even if it is misleading or does not agree with the data at hand. Further, if an associated network based on other data is used as the prior, the method often fails to utilize the information effectively. RESULTS: We propose a novel graphical lasso approach, the tailored graphical lasso, that aims to handle prior information of unknown accuracy more effectively. We provide an R package implementing the method, tailoredGlasso. Applying the method to both simulated and real multiomic data sets, we find that it outperforms the unweighted and weighted graphical lasso in terms of all performance measures we consider. In fact, the graphical lasso and weighted graphical lasso can be considered special cases of the tailored graphical lasso, and a parameter determined by the data measures the usefulness of the prior information. We also find that among a larger set of methods, the tailored graphical is the most suitable for network inference from high-dimensional data with prior information of unknown accuracy. With our method, mRNA data are demonstrated to provide highly useful prior information for protein-protein interaction networks. CONCLUSIONS: The method we introduce utilizes useful prior information more effectively without involving any risk of loss of accuracy should the prior information be misleading.

GLI1-induced mammary gland tumours are transplantable and maintain major molecular features.

Increased expression of GLI1, the main Hedgehog signalling pathway effector, is related to unfavourable prognosis and progressive disease of certain breast cancer subtypes. We used conditional transgenic mice induced to overexpress GLI1 in the mammary epithelium either alone or in combination with deletion of one Trp53 allele to address the role of elevated GLI1 expression in breast tumour initiation and progression. Induced GLI1 expression facilitates mammary gland tumour formation and this was further increased upon heterozygous deletion of Trp53. The GLI1-induced primary tumours were of different murine molecular subtypes, including Normal-likeEx , Class8Ex , Claudin-LowEx and Erbb2-likeEx . The gene expression profiles of some of the tumours correlated well with the PAM50 subtypes for human breast cancer. Whole-exome sequencing revealed somatic mutation profiles with only little overlap between the primary tumours. Orthotopically serially transplanted GLI1-induced tumours maintained the main morphological characteristics of the primary tumours for ≥10 generations. Independent of Trp53 status and molecular subtype, the serially transplanted GLI1-induced tumours were able to grow both in the absence of transgenic GLI1 expression and in the presence of the GLI1 inhibitor GANT61. These data suggest that elevated GLI1 expression has a determinant role in tumour initiation; however, additional genetic events are required for tumour progression.

A Bayesian two-way latent structure model for genomic data integration reveals few pan-genomic cluster subtypes in a breast cancer cohort.

MOTIVATION: Unsupervised clustering is important in disease subtyping, among having other genomic applications. As genomic data has become more multifaceted, how to cluster across data sources for more precise subtyping is an ever more important area of research. Many of the methods proposed so far, including iCluster and Cluster of Cluster Assignments (COCAs), make an unreasonable assumption of a common clustering across all data sources, and those that do not are fewer and tend to be computationally intensive. RESULTS: We propose a Bayesian parametric model for integrative, unsupervised clustering across data sources. In our two-way latent structure model, samples are clustered in relation to each specific data source, distinguishing it from methods like COCAs and iCluster, but cluster labels have across-dataset meaning, allowing cluster information to be shared between data sources. A common scaling across data sources is not required, and inference is obtained by a Gibbs Sampler, which we improve with a warm start strategy and modified density functions to robustify and speed convergence. Posterior interpretation allows for inference on common clusterings occurring among subsets of data sources. An interesting statistical formulation of the model results in sampling from closed-form posteriors despite incorporation of a complex latent structure. We fit the model with Gaussian and more general densities, which influences the degree of across-dataset cluster label sharing. Uniquely among integrative clustering models, our formulation makes no nestedness assumptions of samples across data sources so that a sample missing data from one genomic source can be clustered according to its existing data sources. We apply our model to a Norwegian breast cancer cohort of ductal carcinoma in situ and invasive tumors, comprised of somatic copy-number alteration, methylation and expression datasets. We find enrichment in the Her2 subtype and ductal carcinoma among those observations exhibiting greater cluster correspondence across expression and CNA data. In general, there are few pan-genomic clusterings, suggesting that models assuming a common clustering across genomic data sources might yield misleading results. AVAILABILITY AND IMPLEMENTATION: The model is implemented in an R package called twl ('two-way latent'), available on CRAN. Data for analysis are available within the R package. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A Longitudinal Study of the Association between Mammographic Density and Gene Expression in Normal Breast Tissue.

High mammographic density (MD) is associated with a 4-6 times increase in breast cancer risk. For post-menopausal women, MD often decreases over time, but little is known about the underlying biological mechanisms. MD reflects breast tissue composition, and may be associated with microenvironment subtypes previously identified in tumor-adjacent normal tissue. Currently, these subtypes have not been explored in normal breast tissue. We obtained biopsies from breasts of healthy women at two different time points several years apart and performed microarray gene expression analysis. At time point 1, 65 samples with both MD and gene expression were available. At time point 2, gene expression and MD data were available from 17 women, of which 11 also had gene expression data available from the first time point. We validated findings from our previous study; negative correlation between RBL1 and MD in post-menopausal women, indicating involvement of the TGFß pathway. We also found that breast tissue samples from women with a large decrease in MD sustained higher expression of genes in the histone family H4. In addition, we explored the previously defined active and inactive microenvironment subtypes and demonstrated that normal breast samples of the active subtype had characteristics similar to the claudin-low breast cancer subtype. Breast biopsies from healthy women are challenging to obtain, but despite a limited sample size, we have identified possible mechanisms relevant for changes in breast biology and MD over time that may be of importance for breast cancer risk and tumor initiation.

Claudin-low-like mouse mammary tumors show distinct transcriptomic patterns uncoupled from genomic drivers.

BACKGROUND: Claudin-low breast cancer is a molecular subtype associated with poor prognosis and without targeted treatment options. The claudin-low subtype is defined by certain biological characteristics, some of which may be clinically actionable, such as high immunogenicity. In mice, the medroxyprogesterone acetate (MPA) and 7,12-dimethylbenzanthracene (DMBA)-induced mammary tumor model yields a heterogeneous set of tumors, a subset of which display claudin-low features. Neither the genomic characteristics of MPA/DMBA-induced claudin-low tumors nor those of human claudin-low breast tumors have been thoroughly explored. METHODS: The transcriptomic characteristics and subtypes of MPA/DMBA-induced mouse mammary tumors were determined using gene expression microarrays. Somatic mutations and copy number aberrations in MPA/DMBA-induced tumors were identified from whole exome sequencing data. A publicly available dataset was queried to explore the genomic characteristics of human claudin-low breast cancer and to validate findings in the murine tumors. RESULTS: Half of MPA/DMBA-induced tumors showed a claudin-low-like subtype. All tumors carried mutations in known driver genes. While the specific genes carrying mutations varied between tumors, there was a consistent mutational signature with an overweight of T>A transversions in TG dinucleotides. Most tumors carried copy number aberrations with a potential oncogenic driver effect. Overall, several genomic events were observed recurrently; however, none accurately delineated claudin-low-like tumors. Human claudin-low breast cancers carried a distinct set of genomic characteristics, in particular a relatively low burden of mutations and copy number aberrations. The gene expression characteristics of claudin-low-like MPA/DMBA-induced tumors accurately reflected those of human claudin-low tumors, including epithelial-mesenchymal transition phenotype, high level of immune activation, and low degree of differentiation. There was an elevated expression of the immunosuppressive genes PTGS2 (encoding COX-2) and CD274 (encoding PD-L1) in human and murine claudin-low tumors. CONCLUSIONS: Our findings show that the claudin-low breast cancer subtype is not demarcated by specific genomic aberrations, but carries potentially targetable characteristics warranting further research.

Loss of Epigenetic Regulation Disrupts Lineage Integrity, Induces Aberrant Alveogenesis, and Promotes Breast Cancer.

Systematically investigating the scores of genes mutated in cancer and discerning disease drivers from inconsequential bystanders is a prerequisite for precision medicine but remains challenging. Here, we developed a somatic CRISPR/Cas9 mutagenesis screen to study 215 recurrent "long-tail" breast cancer genes, which revealed epigenetic regulation as a major tumor-suppressive mechanism. We report that components of the BAP1 and COMPASS-like complexes, including KMT2C/D, KDM6A, BAP1, and ASXL1/2 ("EpiDrivers"), cooperate with PIK3CAH1047R to transform mouse and human breast epithelial cells. Mechanistically, we find that activation of PIK3CAH1047R and concomitant EpiDriver loss triggered an alveolar-like lineage conversion of basal mammary epithelial cells and accelerated formation of luminal-like tumors, suggesting a basal origin for luminal tumors. EpiDriver mutations are found in ∼39% of human breast cancers, and ∼50% of ductal carcinoma in situ express casein, suggesting that lineage infidelity and alveogenic mimicry may significantly contribute to early steps of breast cancer etiology. SIGNIFICANCE: Infrequently mutated genes comprise most of the mutational burden in breast tumors but are poorly understood. In vivo CRISPR screening identified functional tumor suppressors that converged on epigenetic regulation. Loss of epigenetic regulators accelerated tumorigenesis and revealed lineage infidelity and aberrant expression of alveogenesis genes as potential early events in tumorigenesis. This article is highlighted in the In This Issue feature, p. 2711.

Multi-Omics Marker Analysis Enables Early Prediction of Breast Tumor Progression.

Ductal carcinoma in situ (DCIS) is a preinvasive form of breast cancer with a highly variable potential of becoming invasive and affecting mortality of the patients. Due to the lack of accurate markers of disease progression, many women with detected DCIS are currently overtreated. To distinguish those DCIS cases who are likely to require therapy from those who should be left untreated, there is a need for robust and predictive biomarkers extracted from molecular or genetic profiles. We developed a supervised machine learning approach that implements multi-omics feature selection and model regularization for the identification of biomarker combinations that could be used to distinguish low-risk DCIS lesions from those with a higher likelihood of progression. To investigate the genetic heterogeneity of disease progression, we applied this approach to 40 pure DCIS and 259 invasive breast cancer (IBC) samples profiled with genome-wide transcriptomics, DNA methylation, and DNA copy number variation. Feature selection using the multi-omics Lasso-regularized algorithm identified both known genes involved in breast cancer development, as well as novel markers for early detection. Even though the gene expression-based model features led to the highest classification accuracy alone, methylation data provided a complementary source of features and improved especially the sensitivity of correctly classifying DCIS cases. We also identified a number of repeatedly misclassified DCIS cases when using either the expression or methylation markers. A small panel of 10 gene markers was able to distinguish DCIS and IBC cases with high accuracy in nested cross-validation (AU-ROC = 0.99). The marker panel was not specific to any of the established breast cancer subtypes, suggesting that the 10-gene signature may provide a subtype-agnostic and cost-effective approach for breast cancer detection and patient stratification. We further confirmed high accuracy of the 10-gene signature in an external validation cohort (AU-ROC = 0.95), profiled using distinct transcriptomic assay, hence demonstrating robustness of the risk signature.

Best Practices for Spatial Profiling for Breast Cancer Research with the GeoMx® Digital Spatial Profiler.

Breast cancer is a heterogenous disease with variability in tumor cells and in the surrounding tumor microenvironment (TME). Understanding the molecular diversity in breast cancer is critical for improving prediction of therapeutic response and prognostication. High-plex spatial profiling of tumors enables characterization of heterogeneity in the breast TME, which can holistically illuminate the biology of tumor growth, dissemination and, ultimately, response to therapy. The GeoMx Digital Spatial Profiler (DSP) enables researchers to spatially resolve and quantify proteins and RNA transcripts from tissue sections. The platform is compatible with both formalin-fixed paraffin-embedded and frozen tissues. RNA profiling was developed at the whole transcriptome level for human and mouse samples and protein profiling of 100-plex for human samples. Tissue can be optically segmented for analysis of regions of interest or cell populations to study biology-directed tissue characterization. The GeoMx Breast Cancer Consortium (GBCC) is composed of breast cancer researchers who are developing innovative approaches for spatial profiling to accelerate biomarker discovery. Here, the GBCC presents best practices for GeoMx profiling to promote the collection of high-quality data, optimization of data analysis and integration of datasets to advance collaboration and meta-analyses. Although the capabilities of the platform are presented in the context of breast cancer research, they can be generalized to a variety of other tumor types that are characterized by high heterogeneity.

Re-definition of claudin-low as a breast cancer phenotype.

The claudin-low breast cancer subtype is defined by gene expression characteristics and encompasses a remarkably diverse range of breast tumors. Here, we investigate genomic, transcriptomic, and clinical features of claudin-low breast tumors. We show that claudin-low is not simply a subtype analogous to the intrinsic subtypes (basal-like, HER2-enriched, luminal A, luminal B and normal-like) as previously portrayed, but is a complex additional phenotype which may permeate breast tumors of various intrinsic subtypes. Claudin-low tumors are distinguished by low genomic instability, mutational burden and proliferation levels, and high levels of immune and stromal cell infiltration. In other aspects, claudin-low tumors reflect characteristics of their intrinsic subtype. Finally, we explore an alternative method for identifying claudin-low tumors and thereby uncover potential weaknesses in the established claudin-low classifier. In sum, these findings elucidate the heterogeneity in claudin-low breast tumors, and substantiate a re-definition of claudin-low as a cancer phenotype.

Comparable cancer-relevant mutation profiles in synchronous ductal carcinoma in situ and invasive breast cancer.

BACKGROUND: Ductal carcinoma in situ (DCIS) comprises a diverse group of preinvasive lesions in the breast and poses a considerable clinical challenge due to lack of markers of progression. Genomic alterations are to a large extent similar in DCIS and invasive carcinomas, although differences in copy number aberrations, gene expression patterns, and mutations exist. In mixed tumors with synchronous invasive breast cancer (IBC) and DCIS, it is still unclear to what extent invasive tumor cells are directly derived from the DCIS cells. AIM: Our aim was to compare cancer-relevant mutation profiles of different cellular compartments in mixed DCIS/IBC and pure DCIS tumors. METHODS AND RESULTS: We performed targeted sequencing of 50 oncogenes in microdissected tissue from three different epithelial cell compartments (in situ, invasive, and normal adjacent epithelium) from 26 mixed breast carcinomas. In total, 44 tissue samples (19 invasive, 16 in situ, 9 normal) were subjected to sequencing using the Ion Torrent platform and the AmpliSeq Cancer Hotspot Panel v2. For comparison, 10 additional, pure DCIS lesions were sequenced. Across all mixed samples, we detected 23 variants previously described in cancer. The most commonly affected genes were TP53, PIK3CA, and ERBB2. The PIK3CA:p.H1047R variant was found in nine samples from six patients. Most variants detected in invasive compartments were also found in the corresponding in situ cell compartment indicating a clonal relationship between the tumor stages. A lower frequency of variants were observed in pure DCIS lesions. CONCLUSION: Similar mutation profiles between in situ and invasive cell compartments indicate a similar origin of the two tumor stages in mixed breast tumors. The lower number of potential driver variants found in pure DCIS compared with the in situ cell compartments of mixed tumors may imply that pure DCIS is captured earlier in the path of progression to invasive disease.

Contrasting DCIS and invasive breast cancer by subtype suggests basal-like DCIS as distinct lesions.

Ductal carcinoma in situ (DCIS) is a non-invasive type of breast cancer with highly variable potential of becoming invasive and affecting mortality. Currently, many patients with DCIS are overtreated due to the lack of specific biomarkers that distinguish low risk lesions from those with a higher risk of progression. In this study, we analyzed 57 pure DCIS and 313 invasive breast cancers (IBC) from different patients. Three levels of genomic data were obtained; gene expression, DNA methylation, and DNA copy number. We performed subtype stratified analyses and identified key differences between DCIS and IBC that suggest subtype specific progression. Prominent differences were found in tumors of the basal-like subtype: Basal-like DCIS were less proliferative and showed a higher degree of differentiation than basal-like IBC. Also, core basal tumors (characterized by high correlation to the basal-like centroid) were not identified amongst DCIS as opposed to IBC. At the copy number level, basal-like DCIS exhibited fewer copy number aberrations compared with basal-like IBC. An intriguing finding through analysis of the methylome was hypermethylation of multiple protocadherin genes in basal-like IBC compared with basal-like DCIS and normal tissue, possibly caused by long range epigenetic silencing. This points to silencing of cell adhesion-related genes specifically in IBC of the basal-like subtype. Our work confirms that subtype stratification is essential when studying progression from DCIS to IBC, and we provide evidence that basal-like DCIS show less aggressive characteristics and question the assumption that basal-like DCIS is a direct precursor of basal-like invasive breast cancer.

Immune Escape in Breast Cancer During In Situ to Invasive Carcinoma Transition.

To investigate immune escape during breast tumor progression, we analyzed the composition of leukocytes in normal breast tissues, ductal carcinoma in situ (DCIS), and invasive ductal carcinomas (IDC). We found significant tissue and tumor subtype-specific differences in multiple cell types including T cells and neutrophils. Gene expression profiling of CD45+CD3+ T cells demonstrated a decrease in CD8+ signatures in IDCs. Immunofluorescence analysis showed fewer activated GZMB+CD8+ T cells in IDC than in DCIS, including in matched DCIS and recurrent IDC. T-cell receptor clonotype diversity was significantly higher in DCIS than in IDCs. Immune checkpoint protein TIGIT-expressing T cells were more frequent in DCIS, whereas high PD-L1 expression and amplification of CD274 (encoding PD-L1) was only detected in triple-negative IDCs. Coamplification of a 17q12 chemokine cluster with ERBB2 subdivided HER2+ breast tumors into immunologically and clinically distinct subtypes. Our results show coevolution of cancer cells and the immune microenvironment during tumor progression.Significance: The design of effective cancer immunotherapies requires the understanding of mechanisms underlying immune escape during tumor progression. Here we demonstrate a switch to a less active tumor immune environment during the in situ to invasive breast carcinoma transition, and identify immune regulators and genomic alterations that shape tumor evolution. Cancer Discov; 7(10); 1098-115. ©2017 AACR.See related commentary by Speiser and Verdeil, p. 1062This article is highlighted in the In This Issue feature, p. 1047.