Fig 1.
Generation and validation of mouse mammary squamous tumor signature.
(A)Tumor histologies observed in a study MMTV-PyMT tumors[54].(B) Venn diagram illustrating the number of genes identified in each comparison using significance analysis of microarrays, 184 genes were commonly identified and proposed as signature genes. (C) Heatmap representation of unsupervised hierarchical clustering of MMTV-PyMT tumors limited to squamous signature genes shows performance of the signature on the training dataset. Levels of RMA normalized median centered expression values are shown according the colorbar. Genes expression data is deposited on GEO datasets GSE104397. (D) Heatmap representation of unsupervised hierarchical clustering of MMTV-Myc tumors limited to squamous signature genes shows performance of the signature on the validation dataset. Levels of RMA normalized median centered expression values are shown according the colorbar. (E) Gene set enrichment analysis testing for enrichment of the proposed squamous signature genes shows significant enrichment in MMTV-Myc squamous tumors (normalized enrichment score, NES = 1.48, nominal p-value = 0.0, FDR q-value = 0.029).
Table 1.
Cellular organization and features associated with tumor histological types.
Table 2.
Known histological associations for single genes within signatures.
Fig 2.
Testing mouse mammary tumor histology signatures across a gene expression database of mouse mammary tumor models.
Above the heatmap black bars depict the position of each tumor from a given major oncogenic model of mammary tumorigenesis. Samples marked with red bars are mouse cells lines that are labeled according to their differentiation as being mesenchymal, epithelial, or basal. Next, blue bars depict available histological annotations for individual tumors in the dataset. Beside the middle heatmap, the black bars indicate the position of signature genes row by row. The middle heatmap displays the RMA-normalized median centered expression level of each gene in a given signature across samples; expression levels are depicted by the color bar on the right hand side. The bottom heatmap displays enrichment scores for each sample and signature from single sample gene set enrichment analysis (ssGSEA). Each sample was tested for the signature (consisting of ‘up’ genes) that it scored maximum for. Samples were grouped according to their highest scoring signature and all samples and heatmaps were sorted high to low within each ssGSEA identified group.
Table 3.
Gene expression signatures of histology predict known histological associations in mouse models.
Fig 3.
ssGSEA testing of signatures suggests key phenotypic traits of histological tumor classes.
(A) Boxplot of enrichment scores pertaining to normal mammary cell differentiation states. (B) Boxplot of enrichment scores pertaining to molecular subtypes of breast cancer, the gene-sets used for this analysis were derived from a prior publication[5]. (C) Boxplot of enrichment scores pertaining to key features of tumor progression and metastasis. * = Significant Enrichment detected in standard GSEA(nominal p-value<0.05); see additional S30 File for additional details and statistics.
Fig 4.
ssGSEA testing of signatures shows key molecular traits histological tumor classes.
(A) Boxplot of enrichment scores pertaining to key cell signaling pathways. (B) Boxplot of enrichment scores pertaining to gene sets for gene targets of specific transcription factors. (C) Boxplot of enrichment scores pertaining to gene sets for gene targets of specific miRNAs. * = Significant Enrichment detected in standard GSEA (nominal p-value<0.05); see additional S30 File for additional details and statistics.
Fig 5.
Histology signature analysis across intrinsic subtypes of breast cancer.
(A)Heatmap depicting expression of mouse mammary tumor derived histology signatures in the METABRIC dataset featuring each intrinsic subtype of breast cancer. Genes are illustrated my median centered, log2 normalized expression level. Samples were clustered within each intrinsic and then genes were clustered within each signature across tumors. (B) Analysis of the EMT signature (highly expressed in EMT) genes in basal-like human breast cancers shows that high expression of this signature is significantly associated with earlier incidence of tumor relapse. pval = 0.0108, HR = 2.05, N = 142. (C) Analysis of the adenomyoepithelial signature genes in basal-like human breast cancers shows that high expression of this signature is significantly associated with earlier incidence of tumor relapse. pval = 7e−04, HR = 2.63, N = 142. (D) Analysis of the adenomyoepithelial signature genes in luminal B human breast cancers shows that high expression of this signature is significantly associated with earlier incidence of tumor relapse. pval = 7e−04, HR = 2.63, N = 447. (E) Analysis of the solid signature genes (highly expressed in solid tumors) in luminal B human breast cancers shows that high expression of this signature is significantly associated with delayed incidence of tumor relapse. pval = 0.0159, HR = 0.721, N = 447. Kaplan-Meier analysis was done using the http://geneanalytics.duhs.duke.edu/Surv_sig.html tool. Samples were assigned to groups based on being above or below the median population value.
Fig 6.
Histology signature analysis across human cancers.
(A)Unsupervised hierarchical clustering of human tumors on the basis of mouse mammary tumor histology signatures. Above the heatmap, blue depict the position of individual tumors annotated by cancer type in the dendrogram above and heatmap below. The red bars provide histological and additional information for each sample. The black bars beside the heatmap annotate the position of each signature gene in the heatmap. The green cluster highlights the cluster featuring the majority of squamous tumors while the blue cluster in the dendrogram highlights a mesenchymal cluster largely composed of melanoma. (B) Geneset enrichment analysis shows significant enrichment of mouse derived squamous histology in tumors from the green squamous cluster. NES = 1.92, nominal p-value 0.0, FDR q-value = 0.003 (C) Geneset enrichment analysis shows significant enrichment for high expression expression of mouse derived EMT histology signature (genes highly expressed expressed in EMT tumors) in samples belonging to the blue cluster in A. NES = 1.93, nominal p-value 0.0, FDR q-value = 6.97 E -4.
Fig 7.
A summary of the detected similarities between mouse mammary tumors and human tumors across cancer types.
Mouse mammary tumor types and human cancers and are ordered to reflect similarities in differentiation and pathway activation. The far left window depicts mouse mammary EMT tumor models that parallel human claudin low breast and melanoma. As illustrated, these tumors show Kras pathway activity and stem-cell like gene expression profiles (as marked by pink bars). The middle window highlights the murine models that feature squamous tumors; these tumors show similarities to human pan-cancer squamous tumors for basal cell like gene expression (as marked by red bars) and to both human squamous cancers and basal-like breast cancers for Hras pathway activity. As indicated by green bars, human basal tumors had luminal progenitor-like gene expression profiles. The far right window highlights murine solid tumor models and microacinar models, which show more luminal-like features (marked by blue bars). Like human luminal tumors, these murine models feature high expression of ER-target genes and tend to show gene expression patterns that suggest they are more differentiated than their other species-specific tumor types.