At the intersection of proteomics and pathology : application of mass spectrometry-based protein quantification to histopathology and antibody validation

Abstract

The paradigm of Precision Medicine relies on the molecular stratification of patients to select “the right patient for the right drug”. The genomic approach has tried to answer this question, but large genomic studies have raised more questions than given answers. It has become fundamental to investigate the biological consequences of genetic mutations and how they influence the proteome. The recent development of LC-MS/MS has helped to address this task, giving the possibility of classifying tumors based on their molecular phenotype and the impact of the genome on the proteome. Among this background, the relationship between tumor histology and the proteomic (or molecular classification) is a recently recognized field that has seen its development in the last 6 years. Thanks to the availability of in-situ multiplexing techniques and deep learning tools for histology, it is now easier to connect a molecular aspect to a histological pattern. One of the main causes for scarce reproducibility in biomedicine is the problem of insufficient antibody validation, especially for FFPE IHC. Many techniques are available for assessing specificity and selectivity of antibodies, showing different advantages and disadvantages, but only orthogonal approaches can give an antibody-free measurement of a specific protein.

The main aims of this thesis have been 1) to investigate the relationship between histology and molecular classification in breast and lung tumors; 2) Develop a proteomic orthogonal validation technique to be used with FFPE IF, based on a dataset of 18 cell lines. The paper I investigated a cohort of 45 breast carcinomas through proteomics, CNA, SNP, mRNA microarray and metabolomics1. The comparison of a proteomic-based classification with the mRNA-based PAM50 subtyping showed good agreement between the two approaches. A consensus clustering performed on a subset of highly variant proteins divided the cohort into 6 different clusters, dividing basal-like tumors in 2 groups and fusing some luminal-B with Her2-enriched tumors. A correlation matrix performed to identify the coexpression of drug target showed how MET and EGFR were present together in basal-like and normal-like tumors. An immunohistochemical study confirmed the co-expression of the two proteins in both basal- and normal-like, but with important differences. While the basallike expressed the two molecules in the invasive component, the normal-like showed the coexpression confined to the DCIS component. A super-resolution microscopy study of both subtypes showed differences in subcellular localization and colocalization between the two.

The paper II described the phenotype of NSCLC in terms of molecular classification, proteogenomics, immunology of cancer and differences between different subtypes2. A cohort of 141 NSCLC was sampled for LC-MS proteomics, DNA panel sequencing, DNA methylation and RNA-seq. The proteomic classification of the tumors divided the lung cohort in 6 subtypes, largely correlated with the histological aspect (subtypes 1-4 were mostly LUAD, the subtype 5 was mostly composed of LNELCC and the subtype 6 almost exclusively of SqCC). A network analysis of the MS data showed that subtypes 2 and 3 were mostly enriched in immune-related proteins and there were important differences between them (expression of CD3, CD8 and PD-L1 in the subtype 2, enriched in CD20 the subtype 3 and suggestive for TLSs). A histopathological examination confirmed the presence of TLSs in subtype 3 and IHC analysis showed high levels of PD-L1 in the subtype 2. Moreover, such immunological differences were matched with differences in histological growth pattern (solid for subtype 2, mixed for subtype 3).

The paper III illustrated a new method for antibody validation on FFPE IF, based on MSbased orthogonal validation3. An 18 cell lines dataset was created that could cover an ample part of the human proteome (diversity cell line set), and the dataset was submitted for LC-MS labelled DDA and for FFPE cell block microarray, with 3 biological replicates for each cell line. As an output of the analysis, we evaluated the reproducibility of both MS and IF data, the correlation between proteomic and AB-based signals. Additionally, we correlated the IF intensities of the 18 cell lines with the entire proteome and then ranked the distribution of correlation coefficients in relation to the target protein, as an estimate measure of specificity and selectivity of the antibody. The analysis of 45 different antibodies showed that the ranking position and the corrispective correlation coefficient were highly related to each other; majority of clones used for IVD showed higher levels of MS-AB correlation if compared to RUO, reflecting a better selection of antibodies used for diagnostic purposes.

The results described above show how a better integration of the two approaches could be useful for improving the diagnostic stratification of patients with cancer, first. The availability of molecular classification, together with a morphological evaluation, will expand the pathologist’s toolbox for diagnostics. Secondary, MS-based proteomics can be used for providing a high-throughput way to validate antibodies for IHC diagnostic use and help to develop new biomarkers, in increased demand for PM.