High-throughput and deep plasma proteomics allows the classification of cancer types

Plasma proteomics is challenging with the 22 most abundant proteins constituting more than 99% of the total protein content, limiting the detection of less abundant proteins using discovery mass spectrometry–based proteomics. To address this limitation, workflows based on protein corona formation have emerged as a strong method to dig deeper into the proteome. Plasma proteins upon exposure to particles forms a layer around it, known as protein corona, leading to a reduction in dynamic range and enabling of detection of lesser abundant proteins. Here, we introduce an enrichment workflow that allows for robust, quantitative and deep enrichment of low-abundant plasma proteins, termed P2 Plasma Enrichment System. We applied this workflow to a cohort of 62 plasma samples from 4 cancers and 36 healthy patients to discover classification biomarkers.

We developed and optimized the P2 Plasma Enrichment System starting from 100µL of plasma followed by tryptic digestion. All samples were analyzed using a 17min analytical gradient and data independent acquisition on a timsTOF HT (Bruker). Data analysis was performed using Spectronaut 19 (Biognosys) using directDIA. From a reference pool plasma, we achieved a proteome depth of 6324 proteins. The enrichment factor is up to 10x the one of neat plasma and 2x compared to an optimized enrichment based on MagReSyn SAX-beads (Mag-Net protocol). Our experimental conditions allow for robust protein corona formation leading to high workflow reproducibility with coefficient of variation on protein level of 6.7% in median, 5498 proteins with a CV <20%. To evaluate quantitative performance of our workflow we performed a controlled quantitative experiment (CQE) based on human plasma spiked with known amounts of chicken plasma (1:9). The mixture of plasma of different species shows the conservation of relative protein quantities, demonstrates its suitable for quantitative plasma proteomics.

We applied the enrichment workflow to 62 plasma samples from 4 cancers type (Breast n= 15; Colon Cancer n=19, Lung n=13 and Prostate n=15) and 36 healthy patients. We identified cumulatively 6979 protein groups in the study and 5498 protein groups on average per sample with a gradient of 17min.

We probed the 4 cancer cohort for classification of the plasma specimen into the cancer types and healthy. Therefore, an analysis pipeline was developed consisting of an ANOVA analysis with filtering by adjusted p-value <0.05. Next, correlating features were removed (R2 < 0.7), finally a s-PLSDA model was trained on 70% of the data (with 10-fold cross validation). The model was tested for classification accuracy on the 30% hold out data and was found to be 92% on average for 10 repeats of the analysis pipeline.

Taken together, the optimized plasma proteomic workflow showcased a powerful way to profile plasma biomarkers. The ease of automating and scaling up such an approach could enable a broader application to other indications and biofluids.