Plasma Proteomics at Scale Enabling Lung Cancer, Alzheimer’s Disease and Proteogenomics Studies with the ProteographTM Product Suite
Asim Siddiqui*, John E. Blume, Sangtae Kim, Margaret K. R. Donovan, Theodore L. Platt, Ryan W. Benz, Martin Goldberg, Serafim Batzoglou, Juan C. Cuevas, Philip Ma, and Omid C. Farokhzad
Proteograph Platform Delivers Unbiased, Deep and Rapid Proteomics at Scale
Proteograph Study of Non-small Cell Lung Cancer (NSCLC) and Alzheimer’s Disease (AD)
Proteogenomic Studies at Scale Require Unbiased, Deep and Rapid Methods
The ~20,000 genes in the human genome encode over one million protein variants, given alternative splice forms, allelic variation and protein modifications. Though large-scale genomics studies have expanded our understanding of biology, similarly-scaled unbiased, deep proteomics studies of biofluids have remained impractical due to complexity of workflows. Here we show how unbiased, deep proteomics enabled by the Proteograph Product Suite1 enables large-scale proteomics and proteogenomic studies.
Genome
Static indicator of risk
800
600
400
200 0
Proteome
Dynamic indicator of status
High utility but low accessibility of content
~695M genetic variants catalogued
1,000
High accessibility of content but low utility
Utility of Information
10M+ human exomes 1M+ genomes
< 0.2% Core Technology Proprietary Engineered Nanoparticles Plasma Proteins Nanoparticles Protein Coronas Nanoparticles form specific and reproducible protein coronas based on their physicochemical properties Proteograph Product Suite of genetic variants fully characterized Sample is ready to be analyzed on most LC-MS instruments • • NSCLC Study1 141 samples, NSCLC lung cancer (n=61) and control (n=80) plasma samples Total experiment time of ~2 weeks on 3 MS instruments AD Study • 200 AD samples (n=50), mild cognitive impairment (MCI; n=50) and control (n=100) plasma samples • Total experiment time of ~3 weeks on 2 MS instruments Proteograph Enables Deep and Unbiased Plasma Proteomics Protein Groups Detected Across 141 Subjects in Control vs. Early NSCLC Study Protein Groups Detected Across 200 Subjects in AD/MCI Study 2499 protein groups are found across all subjects and 1,992 in 25% of the subjects. 21,959 peptides were detected in total with a median 8 peptides per protein across NSCLC Study. Figure 1. Number of protein groups detected across a percentage of the subjects in these two Proteograph studies. Figure 2. Number of peptides for each protein group in each study. 2391 protein groups are found across all subjects and 2,085 in 25% of all the subjects. 26,264 peptides were detected in total with a median 9 peptides per protein across AD Study. 0.91 AUC Healthy vs. Early NSCLC 0.91+/− 0.0093 1.00 0.90 0.75 0.50 0.25 0.10 0.00 Robust and Efficient Biomarker Discovery Workflow with Proteograph Technology 1.00 0.90 0.75 0.50 0.25 0.10 0.00 0.91 AUC Healthy vs. AD 0. 91+/− 0.0054 Figure 3. ROC plot of the 10x10 cross validation of the results in each study. Further verification studies are required to validate the model. False positive fraction False positive fraction To investigate the relative abundance of protein variants and isoforms across different biological states, there are three necessary things to consider; 1) abundance measurement of the variants present, which necessitates multiple measurements across the protein (i.e. multiple peptides), 2) Sufficient scale to measure across many samples and 3) Unbiased interrogation, so measurements are not restricted to the most frequent alleles as the rare alleles are known to play an important role in complex disease. To test the application of the Proteograph solution to analyze protein variants, we exon sequenced 29 individuals from the NSCLC study, creating personalized mass spectrometry search libraries which were used to identify 464 amino acid variants in these individuals. Preliminary investigation of proteins containing these variants suggests putative allele specific presence in at least 178 genes. We observed an increased prevalence of low frequency alleles in the imbalanced variants (Figure 5) which is suggestive of functional implications of these differences. Further investigation is required to rule out technical artifacts of MS processing that may cause loss of these peptides. Together, these results demonstrate how the Proteograph Product Suite can support unbiased, deep proteogenomic studies at scale. AF Across 464 Protein Variant From 29 Subjects vs 1K Genome 100000000 1000000 10000 100 1 0.00 0.25 0.50 0.75 1.00 Alternate Allele Frequency ASE: allele specific expression AF: allele frequency Figure 5. Figure 4. Allele frequency analysis of protein variants. The allele frequency of the variants found in the 29 individuals against the background of allele frequencies in the 1000 genomes project2 shows the distributions are similar, demonstrating the unbiased nature of the Proteograph solution. Variants with ASE AF vs. Protein Variant AF = variants showing ASE = all 464 protein variants Density plot of alleles with allele specific translation/expression as a function of AF. We define putative ASE where we observe one or more peptides mapping to only the reference or only the alternative allele, but not both, in all subjects with that genotype (i.e., monoallelic translation). Kernal smoothing used on density plot. The Proteograph Product Suite has the throughput required for measurements of individual protein variants across the proteome in an unbiased manner, enabling unbiased proteogenomics at scale. References: 1. Blume et al. Nat. Comm. (2020) 2. 1000 Genomes (2015) Seer, Inc., Redwood City, CA - *asiddiqui@seer.bio 0.00 0.10 0.25 0.50 0.75 0.90 1.00 0.00 0.10 0.25 0.50 0.75 0.90 1.00 True positive fraction True positive fraction Number of Variants (log10) Cataloged variants (mm)