J Cancer 2020; 11(20):6081-6089. doi:10.7150/jca.47698 This issue
1. Institute of Pathology, Heidelberg University, Heidelberg, Germany.
2. Proteopath Trier, Trier, Germany.
3. Institute of Pathology, University Hospital Erlangen-Nürnberg, Erlangen, Germany.
4. Institute of Pathology and Molecular Pathology, University Hospital Zurich, Switzerland.
5. Department of Urology and Pediatric Urology, University of Saarland, Homburg/Saar, Germany.
6. Institute of Pathology, TU Munich, Munich, Germany.
7. Bruker Daltonik, Bremen, Germany.
8. Institute of Pathology, University of Mainz, Germany.
9. Centre for Histology, Cytology and molecular Diagnostics Trier, Trier, Germany.
10. Danube Private University, Krems, Austria.
11. Department Hematology, Oncology and Rheumatology, Heidelberg University, Heidelberg, Germany.
Background: While subtyping of the majority of malignant chromophobe renal cell carcinoma (cRCC) and benign renal oncocytoma (rO) is possible on morphology alone, additional histochemical, immunohistochemical or molecular investigations are required in a subset of cases. As currently used histochemical and immunohistological stains as well as genetic aberrations show considerable overlap in both tumors, additional techniques are required for differential diagnostics. Mass spectrometry imaging (MSI) combining the detection of multiple peptides with information about their localization in tissue may be a suitable technology to overcome this diagnostic challenge.
Patients and Methods: Formalin-fixed paraffin embedded (FFPE) tissue specimens from cRCC (n=71) and rO (n=64) were analyzed by MSI. Data were classified by linear discriminant analysis (LDA), classification and regression trees (CART), k-nearest neighbors (KNN), support vector machine (SVM), and random forest (RF) algorithm with internal cross validation and visualized by t-distributed stochastic neighbor embedding (t-SNE). Most important variables for classification were identified and the classification algorithm was optimized.
Results: Applying different machine learning algorithms on all m/z peaks, classification accuracy between cRCC and rO was 85%, 82%, 84%, 77% and 64% for RF, SVM, KNN, CART and LDA. Under the assumption that a reduction of m/z peaks would lead to improved classification accuracy, m/z peaks were ranked based on their variable importance. Reduction to six most important m/z peaks resulted in improved accuracy of 89%, 85%, 85% and 85% for RF, SVM, KNN, and LDA and remained at the level of 77% for CART. t-SNE showed clear separation of cRCC and rO after algorithm improvement.
Conclusion: In summary, we acquired MSI data on FFPE tissue specimens of cRCC and rO, performed classification and detected most relevant biomarkers for the differential diagnosis of both diseases. MSI data might be a useful adjunct method in the differential diagnosis of cRCC and rO.
Keywords: Oncocytic renal tumors, chromophobe renal cell carcinoma, renal oncocytoma, mass spectrometry imaging, proteomics