Identifying survival associated morphological features of triple negative breast cancer using multiple datasets

Abstract

Background and objective: Biomarkers for subtyping triple negative breast cancer (TNBC) are needed given the absence of responsive therapy and relatively poor prediction of survival. Morphology of cancer tissues is widely used in clinical practice for stratifying cancer patients, while genomic data are highly effective to classify cancer patients into subgroups. Thus integration of both morphological and genomic data is a promising approach in discovering new biomarkers for cancer outcome prediction. Here we propose a workflow for analyzing histopathological images and integrate them with genomic data for discovering biomarkers for TNBC.

Materials and methods: We developed an image analysis workflow for extracting a large collection of morphological features and deployed the same on histological images from The Cancer Genome Atlas (TCGA) TNBC samples during the discovery phase (n=44). Strong correlations between salient morphological features and gene expression profiles from the same patients were identified. We then evaluated the same morphological features in predicting survival using a local TNBC cohort (n=143). We further tested the predictive power on patient prognosis of correlated gene clusters using two other public gene expression datasets.

Results and conclusion: Using TCGA data, we identified 48 pairs of significantly correlated morphological features and gene clusters; four morphological features were able to separate the local cohort with significantly different survival outcomes. Gene clusters correlated with these four morphological features further proved to be effective in predicting patient survival using multiple public gene expression datasets. These results suggest the efficacy of our workflow and demonstrate that integrative analysis holds promise for discovering biomarkers of complex diseases.

Keywords: Biomarker Identification; Cancer Survival; Computational Biology; Image Analysis; The Cancer Genome Atlas; Triple Negative Breast Cancer.

Figure 1

The workflow of cross-datasets feature discovery and validation. (A) Steps for discovering correlations between morphological features and expression profiles of gene clusters using The Cancer Genome Atlas (TCGA) data. (B) Survival-related morphological features are discovered using the Ohio State University (OSU) triple negative breast cancer (TNBC) cohort. (C) Gene clusters strongly correlating with the survival-related morphological features are tested for survival using public breast cancer datasets. TMA, tissue microarray.

Figure 2

The workflow of the histopathological image analysis. First, after the removal of background and noise, each tissue slide (or tissue microarray, TMA) image was segmented into ‘superpixels’ delineating the tumor and the stromal compartments of the tissue (green lines mark the boundary of the superpixels in the slide). Then, each superpixel is represented by a series of quantitative morphological features. OSU, Ohio State University; TCGA, The Cancer Genome Atlas; TNBC, triple negative breast cancer.

Figure 3

Nuclei segmentation within each region of interest. (A) An example of the original superpixel. (B) The result after Ostu cellular segmentation. Some of the nuclei overlap and form large clumps. (C) Boundary of the cell nuclei. (D) Final segmentation of the cell nuclei after using edge cut (example indicated by red arrow).

Figure 4

Pairwise correlation heat map between metagene expression and morphology of tissue in The Cancer Genome Atlas discover set. (A) Continuous correlation without threshold. The blue color demonstrates negative correlation; the red color demonstrates the positive correlation. (B) Thresholded correlation (|PCC|>0.5).

Figure 5

Kaplan–Meier survival curves of prognostic model in Ohio State University triple negative breast cancer tissue microarray. Higher values of the image features are plotted as blue lines, and lower values are plotted as red lines. (A) Survival on the two groups with distinct values of ‘Rel_Area_Cell_Nuclei’. (B) Survival of feature ‘Density_Cell_Nuclei_stddev’. (C) Survival of feature ‘Contrast_To_Neighbor_Layer_3’. (D) Survival of feature ‘Area_Cell_Nuclei_stddev’.

Figure 6

Kaplan–Meier curves for the metagene that shows the highest correlation with the image features. All time is represented in months. (A) Survival on the NKI ER-negative patient subset. A higher risk group was revealed by this metagene. (B) Survival on the Perou dataset, ER-negative patient subset. This list of genes can separate the patients into groups with very significantly different outcomes.