Imaging-proteomic analysis for prediction of neoadjuvant chemotherapy responses in patients with breast cancer

Abstract

Background: Optimizing patient selection for neoadjuvant chemotherapy in patients with breast cancer remains an unmet clinical need. Quantitative features from medical imaging were reported to be predictive of treatment responses. However, the biologic meaning of these latent features is poorly understood, preventing the clinical use of such noninvasive imaging markers. The study aimed to develop a deep learning signature (DLS) from pretreatment magnetic resonance imaging (MRI) for predicting responses to neoadjuvant chemotherapy in patients with breast cancer and to further investigate the biologic meaning of the DLS by identifying its underlying pathways using paired MRI and proteomic sequencing data.

Methods: MRI-based DLS was constructed (radiogenomic training dataset, n = 105) and validated (radiogenomic validation dataset, n = 26) for the prediction of pathologic complete response (pCR) to neoadjuvant chemotherapy. Proteomic sequencing revealed biological functions facilitating pCR (n = 139). Their associations with DLS were uncovered by radiogenomic analysis.

Results: The DLS achieved a prediction accuracy of 0.923 with an AUC of 0.958, higher than the performance of the model trained by transfer learning. Cellular membrane formation, endocytosis, insulin-like growth factor binding, protein localization to membranes, and cytoskeleton-dependent trafficking were differentially regulated in patients showing pCR. Oncogenic signaling pathways, features correlated with human phenotypes, and features correlated with general biological processes were significantly correlated with DLS in both training and validation dataset (p.adj < 0.05).

Conclusions: Our study offers a biologically interpretable DLS for the prediction of pCR to neoadjuvant chemotherapy in patients with breast cancer, which may guide personalized medication.

Keywords: Radiogenomics; breast cancer; deep learning; neoadjuvant chemotherapy; pathologic complete response.

FIGURE 1

The overview of the study design. The major steps include MR image acquisition, establishment of deep learning signature (DLS) for the prediction of pathological complete response (pCR) to neoadjuvant chemotherapy, proteomic sequencing for identifying deferentially expressed proteins (DEPs) and protein–protein interaction (PPI) network building, and imaging‐proteomic analysis for revealing the biological meanings of the DLS.

FIGURE 2

Patient enrollment process of the study cohort. The patient enrollment process and the exclusion criteria were shown in the flow chart.

FIGURE 3

Deep learning signature (DLS) predicts pathologic complete response (pCR) to neoadjuvant chemotherapy. (A) Heatmap showing the quantitative values of 512 radiomic features forming the DLS. The 512 radiomic features were visualized in the radiogenomic training and radiogenomic validation dataset separately. Patients in the pCR and non‐pCR group were color‐coded. (B) Receiver operating characteristic curves (ROC) and progesterone receptor (PR) curves demonstrated the predictive power of the DLS in the radiogenomic training dataset (dark blue) and the radiogenomic validation dataset (red). (C) As in (B), but showing the ROC and PR curves for DLS trained by transfer learning.

FIGURE 4

Deferentially expressed proteins (DEPs) and biological pathways upregulated in patients showing pathologic complete response (pCR) to neoadjuvant chemotherapy. (A) Protein–protein interaction (PPI) network of DEPs constructed using the STRING database. Connected nodes were visualized in Cytoscape, and node sizes were associated to the log fold change of the DEPs. Pink nodes were DEPs upregulated in the pCR group, and blue nodes denoted DEPs downregulated in the pCR group. (B) Enrichment analysis of the DEPs. The –log10 (p) values were visualized and color‐coded by the category of biological functions; light blue for mitochondrial energy metabolism, green for vesicle budding and vascular transportation, orange for leukocyte apoptotic process, purple for glucose metabolism, brick for protein localization to membranes, steel blue for messenger ribonucleic acid processing, and pink for cytoskeleton‐dependent trafficking. (C) Single‐sample GSEA scores of immune response and immune suppressive markers were shown by bloxplots. Patients showing PCR were indicated by gray boxes, whereas nonresponders were indicated by light blue boxes.

FIGURE 5

Deep learning signature (DLS) significantly correlated biological functions underlying pathologic complete response (pCR) to neoadjuvant chemotherapy. (A) Pretreatment MR images of tumors from the pCR (upper panel) and non‐pCR (lower panel) group. (B) Heatmap showing the gene set variation analysis scores of biological pathways significantly associated with DLS. (C) Heatmap showing correlations between 512 deep features and biological functions. Adjusted p values were indicated by color, p.adj < 0.05 was shown in red, and p.adj ≥ 0.05 was shown in blue. (D) Protein expression levels were compared between the high DLS (red) and low DLS (gray) groups in the radiogenomic training (n = 100) and validation (n = 25) datasets. *indicates p < 0.05, **indicates p < 0.01, ***indicates p < 0.001. Statistical difference was compared by Mann–Whitney test.