Accounting for population variation in targeted proteomics

Abstract

Individual proteomes typically differ from the reference human proteome at ~10,000 single amino acid variants. When viewed at the population scale, this individual variation results in a wide variety of protein sequences. In targeted proteomics experiments, such variability can confound accurate protein quantification. To assist researchers in identifying target peptides with high variability within the human population we have created the Population Variation plug-in for Skyline, which provides easy access to the polymorphisms stored in dbSNP. Given a set of peptides, the tool reports minor allele frequency for common polymorphisms. We highlight the importance of considering genetic variation by applying the tool to public datasets.

Introduction

In the era of personalized genomics and precision medicine, tens of thousands of human genomes are being sequenced to elucidate the genetic basis for diversity and disease [1]. Compared to the reference human genome, individuals often differ at millions of nucleotides, including both small single nucleotide polymorphisms (SNPs) and larger variations. For SNPs, individual genomes typically show ~10,000 non-synonymous variants that change protein sequence [2], [3], [4]. A second category of SNPs, stop-gain or indels, have a more pervasive effect and alter all subsequent amino acids. Several large-scale sequencing efforts aim to categorize genomic diversity of the human population as a whole. The HapMap consortium initially obtained information for 1 million SNPs from 269 individuals [5]. More recently, the 1000 Genome Project performed whole genome sequencing to discover SNPs as well as larger sequence variants [6]. Such projects continue to expand their sampling and add to the knowledge of human genetic variation. One benefit of population studies is that they are able to estimate the frequency of variants for the entire human population or specific sub-populations.

Targeted proteomics measurements are a high throughput method to accurately quantify protein abundances. The reliability of the method lends it to use in biomarker development studies that require a large number of samples. For example, Whiteaker and colleagues utilized targeted proteomics to quantify proteins in 80 mouse plasma samples [7]. Targeted studies in humans often use cell lines, however recent work by the Carr group studied 13 human cardiac patients and 52 exercising controls to identify biomarkers for myocardial infarction [8].

The diversity in human protein sequences poses a computational challenge for targeted proteomics workflows. As peptide sequences are the quantified surrogate for protein abundance, studies need to account for possible sequence variation across the cohort. Individuals with a variant amino acid within the peptide region would have a null or noise value from a targeted assay. Selecting the best peptide to represent a protein, or assay design, is a crucial aspect of any targeted proteomics experiment [9]. Considerations for peptide selection typically include fragmentation intensity, potential for chemical modification, and interference from the background matrix; many software tools have been created to address these factors [10], [11], [12], [13], [14]. However, there is currently no tool to aiding researchers in identifying peptides that have high variability within the human population. We present the Population Variation tool which uses data from dbSNP to identify the minor allele frequency of peptide targets for MRM/SRM experiments. The tool is available as a plug-in from the Skyline store.

Methods

Database Setup

The human subset of dbSNP build 137 was downloaded in November 2013 from ftp://ftp.ncbi.nih.gov/snp/organisms/human_9606/. Our goal was to obtain a database containing SNPs of a known minor allele frequency. We limited our results using the following criteria: SNPs kept must have a minor allele frequency > 0.01; SNPs kept must have a non-null protein accession; SNPs kept must be of type missence, stop-gain, or frameshift. With these constraints, only three tables were relevant: SNPContigLocusID, Allele, and SNPAlleleFreq_TGP. We simultaneously filter, merge the tables and removed most columns, keeping: prot_acc, residue, aa_pos, snp_id, fxn_code, minorAlleleFreq. This produces a 9 MB database, whereas the original dbSNP download was >15 GB. The resulting data is stored in a SQLite database and distributed with the plug-in.

Database Access

PopulationVariation is programmed in C# using .NET 4.0 framework. It is available as an external tool for Skyline [13], or as a zip from http://omics.pnl.gov/software/PopulationVariation.php Protein accession and peptide sequences are obtained from Skyline via a custom data form. Protein objects in Skyline must contain proper accessions as this is the key into the SNP database. Accepted accessions are NCBI RefSeq (ftp://ftp.ncbi.nlm.nih.gov/genomes/H_sapiens/protein/protein.fa.gz), or Uniprot (ftp://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/proteomes/HUMAN.fasta.gz). dbSNP uses RefSeq accessions as the key, therefore Uniprot accessions are mapped using ftp://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/idmapping/.

Results

In the context of clinical or population studies using targeted proteomics, a peptide with a high natural variation is problematic. Subjects with the variant allele have a different amino acid sequence. Therefore, targeted MRM/SRM approaches which isolate a specific m/z would register a null or noise value for the peptide target, confounding downstream analysis. To assist researchers in determining whether their target peptides have such variability we have created the Population Variation plug-in for the Skyline software program [13]. Three kinds of mutations that alter protein coding sequences are reported: non-synonymous variants that change a single amino acid, and frame-shift and stop-gain mutations that alter all downstream amino acids.

The Population Variation tool draws upon dbSNP as the source of information for genetic variation [15]. Among the many projects that contribute to dbSNP, the 1000 Genome Project uses its broad sampling to report not only the location and category of the polymorphism, but also a global population estimate for minor allele frequency (MAF) [6]. In its first phase, the 1000 Genome project reported 125,204 variants in 20,283 proteins with a minor allele frequency greater than 1%, and 62,418 variants in 13,792 proteins with greater than 5% MAF. There are also a large number of highly common variants; 22,740 variants in 6811 proteins are present with a MAF > 25% (Figure 1).

Figure 1.

Effects of SNPs on MRM assays. In the top graph, the number of SNPs is plotted by their minor allele frequency. For example, 46,018 SNPs have a minor allele frequency greater than or equal to 0.10 (10%). The bottom panel show dbSNP entry rs7190823, a missense mutation to the franconi anemia complementation group A protein. This SNP changes the sequence and therefore the mass of the tryptic peptide.

To highlight the utility of the Population Variation plug-in, we have chosen to re-analyze publicly available SRM studies to show peptides that are impacted by these variations. The plug-in accesses a local version of dbSNP using the protein accession listed in the Skyline file (see Methods). Using the 96 human transcription factors studied by Stergachis et al [16], we searched for common variants which affect protein coding sequences. Twelve peptides were discovered to harbor variants with greater than 5% MAF. In zinc finger protein 530 (NP_065931), peptide DILQMIELQASPCGQK has a 24% minor allele frequency and was one of only three high quality peptides for the protein. This peptide also highlights a subtle problem when using cell lines or other banked biomaterials. The original assay design by Stergachis derived protein sequences from the cDNA clone library used for protein overexpression. Curiously, for zinc finger protein 530 (HsCD00301131) this clone harbored the minor allele. Thus, the peptide identified in the original paper is not present in the majority of the population.

In identifying targets for >1000 prominent cancer proteins, the Aebersold group designed ~5500 SRM assays [17]. We further applied the plug-in on this large-scale dataset. Our result showed that 72 peptides contain variants with MAF > 5% and 30 peptides with a variant allele >20% MAF. We further checked the status of 34 potential candidate biomarkers mentioned in Aebersold’s paper. Five proteins out of 34 were detected with MAF > 1%, with one peptide from complement protein 7 harboring a with 8% minor allele frequency.

Conclusions

As proteomics research moves towards more clinical applications, accounting for natural genetic variation will become increasingly important. Unlike research in model organisms where individual subjects are typically inbred, human studies contain patients with diverse genetic backgrounds. In targeted proteomics, if the desire is to consistently measure the abundance of a single protein, peptide targets should be universal within the cohort. An alternate way of using the tool is to mimic a SNP microarray, and design peptide targets for both alleles. In this manner, researchers could examine the relative abundance of alleles. The Population Variation plug-in for Skyline assists researchers by exposing to them human sequence variation within the peptides and proteins in their experiment. This tool will be regularly updated as new data from 1000 Genome Project are stored in dbSNP, or other projects which estimate population level variation.