mura et al Schmidlin et al These studies The three
mura et al., 2016; Schmidlin et al., 2016). These studies
The three key Voriconazole identified by our decision tree are strongly
include our recent comparison of S/MRM, pseudo-SRM/
associated with important clinical parameters, namely ER status
MRMHR, and SWATH-MS analytical parameters in selected
(INPP4B), tumor grade (CDK1), and HER2 status (ERBB2; Fig-
samples from the same breast cancer tissue collection (Faktor
ures 3B–3D). The receptor tyrosine protein kinase ERBB2 is
et al., 2017). Based on the above data, it has been well demon-
the protein product of the HER-2/NEU gene and is routinely be-
strated both experimentally in our breast tumor sample set
ing used for breast cancer classification into HER2+ or HER2
and in the literature that SWATH-MS provides data highly
phenotypes. It also is the target of anti-HER2 therapy via US
Figure 6. Correlation between Protein and Transcript Levels
Plots show Spearman’s correlation of log2 fold changes (log2FC) between our SWATH-MS protein dataset (96 patients) and five independent transcriptomics datasets DHHCC, DFHCC2, PNC, IRB, and SUPERTAM_HGU_133_PLUS2 (Haibe-Kains et al., 2012; 883 patients; see Data S4 for dataset details) for (A) all protein-transcript pairs and (B) the three key proteins (versus transcripts) selected by the decision tree, INPP4B, CDK1, and ERBB2.
Food and Drug Administration (FDA)-approved humanized monoclonal antibody trastuzumab. The strong association of the ERBB2 protein with HER2 status in our dataset internally val-idates our proteomics data and design of the study. Of note, higher levels of ERBB2 protein observed here in ER /HER2+ versus ER+/HER2+ tumors are also consistent with the better response to therapy of ER /HER2+ versus ER+/HER2+ tumors (Bhargava et al., 2011).
INPP4B is known to dephosphorylate phosphatidylinositol 3,4-bisphosphate in the PI3K pathway, which co-activates cell growth and movement via Akt kinases (Malek et al., 2017). Hence, it serves as a tumor suppressor, and our earlier observa-tions that it is significantly associated with ER+ tumors (Fedele et al., 2010) suggest that it should be explored as a candidate therapeutic target for ER+ breast cancer. The second of our key proteins, mitotic kinase CDK1, is known to accelerate critical processes required for mitosis (Enserink and Kolodner, 2010) and correlates with tumor grade (Chae et al., 2011). Moreover, inhibitors of the family members CDK4/6 have been FDA-approved for the treatment of metastatic breast cancer in a first-line setting (Bilgin et al., 2017). In conclusion, although this is a pilot discovery study and follow-ups with larger patient co-horts are required to further train and validate our classifier, our findings suggest that both INPP4B and CDK1 are promising alternative targets for anti-cancer therapy, as they exhibit a
similar level of association with ER status and tumor grade as ERBB2 with HER2 status, which is already successfully targeted to treat HER2+ breast cancer patients. We would like to note that subsequent validation studies with S/MRM can now be set up easily as the information required for the acquisition methods can be obtained directly from the SWATH assay library. A small panel of validated protein biomarkers could be subsequently im-plemented as part of an IHC panel or assessed with other tech-niques used in the clinic.
Molecular Features Available in Proteotype and Not in Conventional Breast Cancer Subtypes
Although there was a high concordance (84%) between classi-fication based on proteotypes and conventional subtypes, some samples with identical conventional subtype showed distinct proteotypes. We find such proteotype heterogeneity, for example, in triple-negative tumors, a genetically heteroge-neous group that can indeed be sub-divided further. For example, Lehmann et al. (2011) suggested six subtypes based on gene expression profiling: basal-like 1; basal-like 2; immuno-modulatory; mesenchymal; mesenchymal-stem-like; and luminal androgen receptor subtype; others suggested a similar division (Palma et al., 2015). We also found that some HER2-en-riched tumors were more similar to luminal B HER2+ tumors than reflected in current subtyping; this is evident also in data from
Brozkova and colleagues (Brozkova et al., 2008). All these data indicate that proteotypes have the potential of enabling finer stratification of a patient population than conventional subtyp-ing. Current clinical practice shows that treatment based on conventional subtypes is far from optimal with respect to patient response, and proteotyping can potentially provide a more accu-rate picture of the actual molecular state of a cancerous tissue and could thereby enable more precise or even personalized treatment.