Liquid biopsy of peripheral blood using mass spectrometry detects primary extramedullary disease in multiple myeloma patients.

Multiple myeloma (MM) is the second most prevalent hematological malignancy, characterized by infiltration of the bone marrow by malignant plasma cells. Extramedullary disease (EMD) represents a more aggressive condition involving the migration of a subclone of plasma cells to paraskeletal or extraskeletal sites. Liquid biopsies could improve and speed diagnosis, as they can better capture the disease heterogeneity while lowering patients' discomfort due to minimal invasiveness. Recent studies have confirmed alterations in the proteome across various malignancies, suggesting specific changes in protein classes. In this study, we show that MALDI-TOF mass spectrometry fingerprinting of peripheral blood can differentiate between MM and primary EMD patients. We constructed a predictive model using a supervised learning method, partial least squares-discriminant analysis (PLS-DA) and evaluated its generalization performance on a test dataset. The outcome of this analysis is a method that predicts specifically primary EMD with high sensitivity (86.4%), accuracy (78.4%), and specificity (72.4%). Given the simplicity of this approach and its minimally invasive character, this method provides rapid identification of primary EMD and could prove helpful in clinical practice.

Intact cell mass spectrometry coupled with machine learning reveals minute changes induced by single gene silencing.

Intact (whole) cell MALDI TOF mass spectrometry is a commonly used tool in clinical microbiology for several decades. Recently it was introduced to analysis of eukaryotic cells, including cancer and stem cells. Besides targeted metabolomic and proteomic applications, the intact cell MALDI TOF mass spectrometry provides a sufficient sensitivity and specificity to discriminate cell types, isogenous cell lines or even the metabolic states. This makes the intact cell MALDI TOF mass spectrometry a promising tool for quality control in advanced cell cultures with a potential to reveal batch-to-batch variation, aberrant clones, or unwanted shifts in cell phenotype. However, cellular alterations induced by change in expression of a single gene has not been addressed by intact cell mass spectrometry yet. In this work we used a well-characterized human ovarian cancer cell line SKOV3 with silenced expression of a tumor suppressor candidate 3 gene (TUSC3). TUSC3 is involved in co-translational N-glycosylation of proteins with well-known global impact on cell phenotype. Altogether, this experimental design represents a highly suitable model for optimization of intact cell mass spectrometry and analysis of spectral data. Here we investigated five machine learning algorithms (k-nearest neighbors, decision tree, random forest, partial least squares discrimination, and artificial neural network) and optimized their performance either in pure populations or in two-component mixtures composed of cells with normal or silenced expression of TUSC3. All five algorithms reached accuracy over 90 % and were able to reveal even subtle changes in mass spectra corresponding to alterations of TUSC3 expression. In summary, we demonstrate that spectral fingerprints generated by intact cell MALDI-TOF mass spectrometry coupled to a machine learning classifier can reveal minute changes induced by alteration of a single gene, and therefore contribute to the portfolio of quality control applications in routine cell and tissue cultures.

Improved Screening of Monoclonal Gammopathy Patients by MALDI-TOF Mass Spectrometry.

Monoclonal gammopathies are a group of blood diseases characterized by presence of abnormal immunoglobulins in peripheral blood and/or urine of patients. Multiple myeloma and plasma cell leukemia are monoclonal gammopathies with unclear etiology, caused by malignant transformation of bone marrow plasma cells. Mass spectrometry with matrix-assisted laser desorption/ionization and time-of-flight detection is commonly used for investigation of the peptidome and small proteome of blood plasma with high accuracy, robustness, and cost-effectivity. In addition, mass spectrometry coupled with advanced statistics can be used for molecular profiling, classification, and diagnosis of liquid biopsies and tissue specimens in various malignancies. Despite the fact there have been fully optimized protocols for mass spectrometry of normal blood plasma available for decades, in monoclonal gammopathy patients, the massive alterations of biophysical and biochemical parameters of peripheral blood plasma often limit the mass spectrometry measurements. In this paper, we present a new two-step extraction protocol and demonstrated the enhanced resolution and intensity (>50×) of mass spectra obtained from extracts of peripheral blood plasma from monoclonal gammopathy patients. When coupled with advanced statistics and machine learning, the mass spectra profiles enabled the direct identification, classification, and discrimination of multiple myeloma and plasma cell leukemia patients with high accuracy and precision. A model based on PLS-DA achieved the best performance with 71.5% accuracy (95% confidence interval, CI = 57.1-83.3%) when the 10× repeated 5-fold CV was performed. In summary, the two-step extraction protocol improved the analysis of monoclonal gammopathy peripheral blood plasma samples by mass spectrometry and provided a tool for addressing the complex molecular etiology of monoclonal gammopathies.