Characterization of four subtypes in morphologically normal tissue excised proximal and distal to breast cancer

Abstract

Widespread mammographic screening programs and improved self-monitoring allow for breast cancer to be detected earlier than ever before. Breast-conserving surgery is a successful treatment for select women. However, up to 40% of women develop local recurrence after surgery despite apparently tumor-free margins. This suggests that morphologically normal breast may harbor early alterations that contribute to increased risk of cancer recurrence. We conducted a comprehensive transcriptomic and proteomic analysis to characterize 57 fresh-frozen tissues from breast cancers and matched histologically normal tissues resected proximal to (<2 cm) and distant from (5-10 cm) the primary tumor, using tissues from cosmetic reduction mammoplasties as baseline. Four distinct transcriptomic subtypes are identified within matched normal tissues: metabolic; immune; matrisome/epithelial-mesenchymal transition, and non-coding enriched. Key components of the subtypes are supported by proteomic and tissue composition analyses. We find that the metabolic subtype is associated with poor prognosis (p < 0.001, HR6.1). Examination of genes representing the metabolic signature identifies several genes able to prognosticate outcome from histologically normal tissues. A subset of these have been reported for their predictive ability in cancer but, to the best of our knowledge, these have not been reported altered in matched normal tissues. This study takes an important first step toward characterizing matched normal tissues resected at pre-defined margins from the primary tumor. Unlocking the predictive potential of unexcised tissue could prove key to driving the realization of personalized medicine for breast cancer patients, allowing for more biologically-driven analyses of tissue margins than morphology alone.

Keywords: Breast cancer; Tumour biomarkers.

Fig. 1. Representative immunohistochemical images and locations of sample types.

Immunohistochemical images representing tissues collected from tumor, TP, TD, RR, and RM. Images shown at ×20 magnification, with scale bars at 50 μm.

Fig. 2. Four subtypes are identified in matched HN tissues.

a NMF on data derived from morphologically normal tissues resolves into four clusters (cophenetic coefficient 0.9991). b Heatmap of normalized abundance for genes defined as highly differentially abundant between the classes in a SAMseq multiclass analysis reveals the distinct transcriptomic patterns between the subtypes.

Fig. 3. The phosphoproteomic profiles of tissues in each transcriptomic group exhibit altered phosphorylation and kinase activities in proteins associated with distinct biochemical and signaling processes.

a Diverse substrate groups differentially enriched between the subtypes. b Heatmap of the top kinase pathways enriched in the subtypes. A complete list of pathways is available from Supplementary Data Set 4.

Fig. 4. PIT detects translational patterns in the data.

The rows in the UpSet represent the spatial location or subtype being assessed and the columns represent the intersections and aggregates of conditions. The bar plot displays the number of translated elements identified for each intersection or aggregate. a The number of TGEs decreases with distance from the primary tumor, with translated elements being reported common between both HN subtypes and tumor. b The translational profiles of the transcriptomics-derived subtypes differ. While they exhibit distinct characteristics, all of the HN profiles also show commonalities with each other and with the tumor profile. c The nc-enriched subtype is enriched for TGE linked to alternative splicing events.

Fig. 5. Boxplots showing the non-negative expression values of three well-known ncRNAs associated with alternative splicing across all HN subtypes and primary breast cancer.

The central line in the boxes represent the median expression value, the boundaries of the boxes represent the interquartile range and the ends of the whiskers represent the minimum and maximum values in the data. The expression of NEAT1 a, XIST b, and HOTAIRM1 c are significantly higher in the nc-enriched subtype relative to the remaining HN subtypes and in primary tumors.

Fig. 6. Deconvolution analysis of the transcriptional subtypes.

Each subtype has unique features of cellular composition.

Fig. 7. Kaplan–Meier plot showing the relationship between the metabolic subtype and outcome.

Cancer-adjacent samples from the TCGA BRCA cohort was used to estimate the prognostic value of our findings. Each cancer-adjacent sample from the TCGA was assigned to a transcriptomic subtype. Patients allocated to the metabolic risk group have a worse prognosis relative to patients in all other risk groups (log-rank p < 0.001, hazard ratio 5.8). The hazard ratio was estimated using a Cox proportional hazards model, and curves were compared using a log-rank test.