Quantifying Homologous Proteins and Proteoforms

Dmitry Malioutov, Tianchi Chen, Edoardo M. Airoldi, Jacob D. Jaffe, Bogdan Budnik, Nikolai SlavovMolecular 2018 · DOI

TL;DR

This paper introduces HIquant, a novel first-principles model that quantifies protein and proteoform stoichiometries from mass spectrometry data, overcoming limitations of existing methods that are prone to peptide-specific biases. HIquant achieves accurate quantification without relying on external standards, enabling a deeper understanding of complex biological systems.

What Problem Does This Paper Address?

Understanding the diverse functions of proteins within a cell often requires distinguishing between different "proteoforms." These proteoforms can arise from alternative splicing of mRNA, post-translational modifications (PTMs) like phosphorylation or methylation, or from genes with similar sequences (paralogs). While these proteoforms can have distinct, even opposing, biological roles, accurately quantifying their relative abundances using standard bottom-up mass spectrometry (MS) is challenging. This is because the measured signal for a peptide (its precursor ion area) is influenced by many factors beyond the protein's actual abundance, such as how efficiently it's digested by enzymes, how well it ionizes in the mass spectrometer, and interference from other co-eluting peptides. These "peptide-specific biases" can lead to inaccurate estimations of proteoform ratios, hindering our ability to study complex biological processes.

Methods

The core of this paper is the development and application of HIquant, a first-principles model designed to infer proteoform stoichiometries. HIquant treats the measured peptide levels across different experimental conditions as a superposition of the underlying protein levels, each scaled by an unknown, peptide-specific bias (nuisance factor).

  • Superposition Model: HIquant models the relationship between measured peptide intensities and protein abundances using a system of linear equations. Shared peptides, which are present in multiple proteoforms, act as crucial "internal standards" that link the measurements of different proteoforms together.
  • Optimal Inference Algorithm: The model is solved using convex optimization techniques. This approach allows HIquant to infer protein levels independently of the nuisance factors, effectively correcting for peptide-specific biases.
  • Condition-Based Inference: A key innovation is that HIquant requires measurements across multiple conditions (N > 1). This allows for a unique or near-unique solution to be determined, even when peptides are shared, which is not possible with data from a single condition.

Datasets

The paper describes experiments using two main types of datasets for validation:

  • Spiked-in Proteoform Mixtures: A "gold standard" was created using the dynamic universal proteomics standard (UPS2). Cysteines in UPS2 proteins were modified with either iodoacetamide or vinylpyridine, creating distinct proteoforms. These were then mixed in known ratios and spiked into yeast lysate, followed by MS quantification.
  • Endogenous Histone H3 PTM Proteoforms: Data from histone H3 peptides were used, quantified by selective reaction monitoring (SRM) across multiple experimental perturbations. These were compared against estimates derived from pre-established "MasterMix" external standards with known concentrations.

Key Insights and Ideas

  • Shared Peptides as Internal Standards: The paper highlights the power of using peptides that are common to multiple proteoforms as implicit internal standards. This is a fundamental concept that allows HIquant to decouple protein abundance from peptide-specific biases.
  • Multi-Condition Data is Crucial: The necessity of analyzing data across multiple conditions (N > 1) is emphasized. This is what provides the statistical power to resolve individual proteoform abundances and overcome the underdetermined nature of single-condition measurements.
  • Systematic Biases Can Be Absorbed: HIquant is designed to absorb systematic biases (like consistent incomplete digestion or co-isolation interference across samples) into its nuisance parameters, making the inferred protein levels robust to these issues.
  • Beyond Unique Peptides: While unique peptides are useful, HIquant's strength lies in its ability to leverage shared peptides, making it particularly valuable for quantifying proteoforms that may not have many unique peptides.

Key Results

  • Accurate Quantification of Simple Proteoforms: For alkylated UPS2 proteoforms with known mixing ratios, HIquant achieved a median error below 10% when inferring ratios. This was significantly better than relying solely on precursor ion areas of unique peptides.
  • Accurate Quantification of Complex PTM Proteoforms: For endogenous histone H3 PTMs, HIquant inferred fractional site occupancies with relative errors of 5-15% for simple proteoforms and 20-30% for more complex ones, when compared to estimates from external standards.
  • Robustness to Bias: The model demonstrated robustness to systematic biases, such as incomplete peptide cleavage, as long as the bias was consistent across conditions.
  • Confidence Estimation: HIquant evaluates the reliability of its inferred abundances by assigning confidence levels, using metrics like explained variance and noise sensitivity.

Limitations and Future Work

  • Dependence on Relative Quantification Accuracy: HIquant's accuracy is fundamentally dependent on the accuracy of relative quantification of corresponding ions. While technological advancements are improving this, significant differences in abundance (3-6 orders of magnitude) might still be better inferred from unique peptides.
  • Abundances Below Detection Limits: Proteoforms with extremely low abundances (over 1000-fold lower than main proteoforms) and their corresponding peptides were omitted from inference due to the unrealistically high accuracy required for their quantification.
  • Future Directions: The authors suggest that HIquant's general framework is not limited to proteoforms and can be applied to any set of proteins sharing peptides. This opens doors for broader applications in quantitative proteomics, especially for systems with homologous proteins. The implementation of a HIquant server is also a significant step towards making this method accessible.

Summary updated

Abstract

Many proteoforms-arising from alternative splicing, post-translational modifications (PTM), or paralogous genes-have distinct biological functions, such as histone PTM proteoforms. However, their quantification by existing bottom-up mass-spectrometry (MS) methods is undermined by peptide-specific biases. To avoid these biases, we developed and implemented a first-principles model (HI<i>quant</i>) for quantifying proteoform stoichiometries. We characterized when MS data allow inferring proteoform stoichiometries by HI<i>quant</i> and derived an algorithm for optimal inference. We applied this algorithm to infer proteoform stoichiometries in two experimental systems that supported rigorous bench-marking: alkylated proteoforms spiked-in at known ratios and endogenous histone 3 PTM proteoforms quantified relative to internal heavy standards. When compared with the benchmarks, the proteoform stoichiometries interfered by HI<i>quant</i> without using external standards had relative error of 5-15% for simple proteoforms and 20-30% for complex proteoforms. A HI<i>quant</i> server is implemented at: https://web.northeastern.edu/slavov/2014HIquant/.
BibTeX
@inproceedings{malioutov2018quantifying,
  author = {Dmitry Malioutov and Tianchi Chen and Edoardo M. Airoldi and Jacob D. Jaffe and Bogdan Budnik and Nikolai Slavov},
  title = {{Quantifying Homologous Proteins and Proteoforms}},
  booktitle = {Molecular},
  year = {2018},
  doi = {10.1074/mcp.tir118.000947},
}
Figure