Summary Accurate classification of breast tumors is vital for patient management decisions and enables more precise cancer treatment. Here, we present a quantitative proteotyping approach based on sequential windowed acquisition of all theoretical fragment ion spectra (SWATH) mass spectrometry and establish key proteins for breast tumor classification. The study is based on 96 tissue samples representing five conventional breast cancer subtypes. SWATH proteotype patterns largely recapitulate these subtypes; however, they also reveal varying heterogeneity within the conventional subtypes, with triple negative tumors being the most heterogeneous. Proteins that contribute most strongly to the proteotype-based classification include INPP4B, CDK1, and ERBB2 and are associated with estrogen receptor (ER) status, tumor grade status, and HER2 status. Although these three key proteins exhibit high levels of correlation with transcript levels (R > 0.67), general correlation did not exceed R = 0.29, indicating the value of protein-level measurements of disease-regulated genes. Overall, this study highlights how cancer tissue proteotyping can lead to more accurate patient stratification.

A B S T R A C TInterferon-induced transmembrane proteins IFITM1 and IFITM3 (IFITM1/3) play a role in both RNA viral restriction and in human cancer progression. Using immunohistochemical staining of FFPE tissue, we identified subgroups of cervical cancer patients where IFITM1/3 protein expression is inversely related to metastasis. Guide RNA-CAS9 methods were used to develop an isogenic IFITM1/IFITM3 double null cervical cancer model in order to define dominant pathways triggered by presence or absence of IFITM1/3 signalling. A pulse SILAC methodology identified IRF1, HLA-B, and ISG15 as the most dominating IFNγ inducible proteins whose synthesis was attenuated in the IFITM1/IFITM3 double-null cells. Conversely, SWATH-IP mass spectrometry of ectopically expressed SBP-tagged IFITM1 identified ISG15 and HLA-B as dominant co-associated proteins. ISG15ylation was attenuated in IFNγ treated IFITM1/IFITM3 double-null cells. Proximity ligation assays indicated that HLA-B can interact with IFITM1/3 proteins in parental SiHa cells. Cell surface expression of HLA-B was attenuated in IFNγ treated IFITM1/IFITM3 double-null cells. SWATH-MS proteomic screens in cells treated with IFITM1-targeted siRNA cells resulted in the attenuation of an interferon regulated protein subpopulation including MHC Class I molecules as well as IFITM3, STAT1, B2M, and ISG15. These data have implications for the function of IFITM1/3 in mediating IFNγ stimulated protein synthesis including ISG15ylation and MHC Class I production in cancer cells. The data together suggest that pro-metastatic growth associated with IFITM1/3 negative cervical cancers relates to attenuated expression of MHC Class I molecules that would support tumor immune escape.Abbreviations: B2M, beta-2-microglobulin; FASP, filter-aided sample preparation; FA, formic acid; HLA, human leucocyte antigen; IFN, interferon; IFITM1/3, interferon-induced transmembrane receptors 1 and 3; ISG15, interferon-stimulated gene 15; IRF1, interferon regulatory factor 1; MHC, major histocompatibility complex; MS, mass spectrometry; NMWCO, nominal molecular weight cut-off; PBS, phosphate-buffered saline; PLA, proximity ligation assay; RT, room temperature; SWATH-IP, SWATH immunoprecipitation; SBP, twin streptavidin binding protein; UPLC, ultra performance liquid chromatography ⁎ Corresponding authors at

Accurate breast cancer classification is vital for patient management decisions, and better tumour classification is expected to enable more precise and eventually personalized treatment to improve patient outcomes. Here, we present a novel quantitative proteotyping approach based on SWATH mass spectrometry and establish key proteins for breast tumour classification derived from proteotype data. The study was based on 96 tissue samples representing five breast cancer subtypes according to conventional classification. Correlation of SWATH proteotype patterns indicated groups that largely recapitulate these subtypes. However, 2 the proteotype-based classification also revealed varying degrees of heterogeneity within the conventional subtypes, with triple negative tumours being the most heterogeneous. Proteins that contributed most strongly to the proteotype-based classification include INPP4B, CDK1, and ERBB2, which are associated with oestrogen receptor status, tumour grade, and HER2 status, respectively. While these three key proteins exhibited high levels of correlation between protein and transcript levels (R>0.67), general correlation did not exceed R=0.29, indicating the value of protein-level measurements of biomarkers and disease-regulated genes. Overall, our data shows how large-scale protein-level measurements by next-generation proteomics can lead to improved patient stratification for precision medicine.

Targeted mass spectrometry-based proteomics approaches enable the simultaneous and reproducible quantification of multiple protein analytes across numerous conditions in biology and clinical studies. These approaches involve e.g. selected reaction monitoring (SRM) typically conducted on a triple quadrupole mass spectrometer, its high-resolution variant named pseudo-SRM (p-SRM), carried out in a quadrupole coupled with an TOF analyzer (qTOF), and "sequential window acquisition of all theoretical spectra" (SWATH). Here we compared these methods in terms of signal-to-noise ratio (S/N), coefficient of variance (CV), fold change (FC), limit of detection and quantitation (LOD, LOQ). We have shown the highest S/N for p-SRM mode, followed by SRM and SWATH, demonstrating a trade-off between sensitivity and level of multiplexing for SRM, p-SRM, and SWATH. SRM was more sensitive than p-SRM based on determining their LOD and LOQ. Although SWATH has the worst S/N, it enables peptide multiplexing with post-acquisition definition of the targets, leading to better proteome coverage. FC between breast tumors of different clinical-pathological characteristics were highly correlated (R >0.97) across three methods and consistent with the previous study on 96 tumor tissues. Our technical note presented here, therefore, confirmed that outputs of all the three methods were biologically relevant and highly applicable to cancer research.