Clinical Proteomics Reveals Vulnerabilities in Noninvasive Breast Ductal Carcinoma and Drives Personalized Treatment Strategies

Abstract

Abstract: Ductal carcinoma in situ (DCIS) is the most common type (80%) of noninvasive breast lesions in women. The lack of validated prognostic markers, limited patient numbers, and variable tissue quality have a significant impact on the diagnosis, risk stratification, patient enrollment, and results of clinical studies. In this study, we performed label-free quantitative proteomics on 50 clinical formalin-fixed, paraffin-embedded biopsies, validating 22 putative biomarkers from independent genetic studies. Our comprehensive proteomic phenotyping reveals more than 380 differentially expressed proteins and metabolic vulnerabilities, which can inform new therapeutic strategies for DCIS and invasive ductal carcinoma. Due to the readily druggable nature of proteins and metabolic enzymes or metabolism inhibitors, this study is of high interest for clinical research and the pharmaceutical industry. To further evaluate our findings, and to promote the clinical translation of our study, we developed a highly multiplexed targeted proteomics assay for 90 proteins associated with cancer metabolism, RNA regulation, and signature cancer pathways, such as PI3K/AKT/mTOR and EGFR/RAS/RAF.

Significance: This study provides real-world evidence for DCIS, a disease for which currently no molecular tools or biomarkers exist, and gives an unbiased, comprehensive, and deep proteomic profile, identifying >380 actionable targets.

Figure 1

Experimental design. LFQ proteomics was performed in a cohort of carefully curated patients treated with DCIS and IDC (n = 51) to investigate changes in the protein expression. Protein extraction of FFPE tissue cores (1 mm diameter, ∼0.8 mm³ tissue volume) used an optimized FFPE proteomics protocol published in this study (11). The samples were analyzed on a “plug-and-play” platform built for standardization in clinical proteomics, and the data were processed using state-of-the-art data analysis tools, including machine learning/artificial intelligence–driven algorithms for improved and higher confidence mechanistic insights.

Figure 2

Data quality and evaluation of variability. A, Dynamic range of ∼2,860 proteins quantified in ductal breast cancer, at a 1% FDR. All −log₁₀ values were based on NSAF values, which were used to normalize the spectral count (25). High NSAF values represent a high level of expression. Six orders of magnitude of the DCIS/IDC proteome are covered using ∼1% of the total sample and a standard data-dependent acquisition method without fractionation. B, Sparse partial least squares regression for discrimination analysis showing good clustering of the two study groups. The oval shape represents 95% confidence intervals. IQR and Robust regression and outlier removal methods identified no outlier samples. C and D, Loading plots of the sparse partial least squares regression for discrimination analysis showing proteins/genes that drive the variability and clustering between DCIS and IDC. The right x-axis shows expression levels of these drivers in the DCIS/IDC samples. NSAF, normalized spectral abundance factor.

Figure 3

Differential expression analysis reflects loss of basal membrane stability, inflammatory processes, and epithelial–mesenchymal transition as key events toward DCIS to IDC progression. A, Volcano plot of the proteome of IDC compared with DCIS lesions showing 388 DEPs (unpaired t test with post hoc Benjamini–Krieger analysis P < 0.01, absolute log₂ FC >2). B, Volcano plot of the proteome of paired IDC lesions compared with paired DCIS lesions showing 10 DEPs (unpaired t test with post hoc Benjamini–Krieger analysis P < 0.05, absolute log₂ FC >2). C, Molecular networks representing up/downregulated pathways in IDC compared with DCIS lesions. adj, adjusted; EMT, epithelial-to-mesenchymal transition; UPR, unfolded protein response.

Figure 4

Dysregulation of central energy metabolism is a key event in the DCIS tumor phenotype. A, Graphical representation of hallmarks of cancer [modified from (82)] characteristic for proteomic tumor profiling of DCIS and IDC tumors. B, Top 5 canonical pathways from Ingenuity Pathway Analysis on DEPs in 42 DCIS and 31 IDC tumors. C, Signature proteins potentially driving DCIS progression through glycolysis, hypoxia (or “pseudohypoxia”), and PI3K/AKT/mTOR pathway, identified by GSEA. D, STRING network showing the protein expression profile of signature proteins associated with cancer metabolism, RNA regulation, and major cancer pathways, such as PI3K/AKT/mTOR and EGFR/RAS/RAF. Absolute concentration of the proteins was determined by PRM. The color of the nodes represents q values from multiple hypothesis testing using unpaired t tests with post hoc correction using the Benjamini–Krieger FDR method (1% FDR). The node size represents the FC. Gray nodes were not quantified, either because no SIS/NAT was available or because there were more than 60% missing values. Edges represent physical and/or functional interaction partners based on the STRING database.