A Quantitative Model to Estimate Drug Resistance in Pathogens

Frazier N. Baker, Melanie T. Cushion, Aleksey PorolloJournal of Fungi 2016 · DOI

TL;DR

This paper introduces a novel quantitative model that predicts how mutations in pathogen proteins affect their resistance to drugs. By leveraging evolutionary information and amino acid co-evolution, the model accurately forecasts drug resistance, showing promise for evaluating new pathogen variants and guiding therapeutic strategies.

What Problem Does This Paper Address?

Pathogenic microorganisms, particularly those infecting immunocompromised individuals like Pneumocystis jirovecii (which causes Pneumocystis pneumonia or PCP), are increasingly developing resistance to existing treatments. The primary treatment for PCP, trimethoprim-sulfamethoxazole (TMP-SMX), targets two key enzymes in the folate biosynthesis pathway. However, mutations in these enzymes are leading to reduced drug efficacy. While mutations conferring resistance are being identified, predicting their precise impact on drug binding and overall resistance remains a challenge. Existing computational methods have limitations, such as only considering single mutations or being computationally intensive. This paper aims to bridge this gap by developing a more comprehensive and practical quantitative model to predict drug resistance based on protein mutations.

Methods

The authors developed a new quantitative model that builds upon previous work in predicting protein function changes due to mutations. Their approach incorporates two main components:

  • Individual Position Contributions ($\Delta\Phi_{WIND}$): This component considers the effect of mutations at individual amino acid positions.
  • Pairwise Epistatic Interactions ($\Delta\Phi_{W EPI}$): This component accounts for how mutations at one position can influence the effect of mutations at another position, reflecting "epistatic coupling."

Crucially, their model differs from prior methods by:

  • Simultaneous Mutation Analysis: It can evaluate the combined effect of multiple mutations within a protein simultaneously, rather than just single mutations.
  • Weighted Contributions: It assigns weights to individual positions and pairwise interactions, acknowledging that some sites or couplings might be more critical for protein function or drug binding than others.
  • Co-variance Metrics: It utilizes co-variance metrics (Mutual Information, Chi-squared, and Pearson correlation) derived from multiple sequence alignments (MSAs) to capture these epistatic interactions. They specifically highlight the use of Mutual Information adjusted with Average Product Correction (APC) as a key component.
  • Normalization: The model's output is normalized using a logistic sigmoid function to ensure predictions fall within a biologically relevant range.

The final drug resistance score ($\Delta\Phi_{WDR}$) is a product of the normalized individual and pairwise interaction components.

Datasets

The study utilized several experimental datasets for training and evaluating their model:

  • Training Data: Kinetic data (Ki) for Pneumocystis jirovecii DHFR (dihydrofolate reductase) variants inhibited by Trimethoprim (TMP). This dataset included 13 single variants and 6 double mutants.
  • Evaluation Data:
    • Kinetic data for P. jirovecii DHFR variants with a different inhibitor, OAAG324.
  • Kinetic data for P. jirovecii DHPS (dihydropteroate synthase) variants inhibited by Sulfamethoxazole (SMX).
  • Kinetic data for Staphylococcus aureus DHFR variants inhibited by TMP.

These datasets represent different inhibitors, drug targets within the same organism, and drug targets in different organisms, allowing for a comprehensive evaluation of the model's generalization capabilities.

Key Insights and Ideas

  • Epistasis Matters: The model's success hinges on capturing not just individual amino acid changes but also the complex interplay (epistasis) between mutations at different sites. This suggests that drug resistance is often a multi-factorial phenomenon.
  • Beyond Single Mutations: The ability to model multiple simultaneous mutations is a significant advantage, as resistance often arises from a combination of genetic changes.
  • Transferability is Key: The model's ability to perform well across different inhibitors, drug targets, and even different species (fungal vs. bacterial) highlights the underlying universality of the principles it captures.
  • Evolutionary Information as a Predictor: The strong correlation between evolutionary patterns (co-evolution) and drug resistance underscores the power of using evolutionary history to predict functional outcomes.

Key Results

The developed quantitative model, $\Delta\Phi_{WDR}$, demonstrated strong performance:

  • Training Performance: When trained on P. jirovecii DHFR/TMP data, the model achieved a Pearson correlation coefficient (r) of 0.72 with experimental kinetic data.
  • Generalization:
    • For a different inhibitor (OAAG324) targeting the same enzyme, the correlation dropped slightly to r = 0.60.
  • For a different drug target (PjDHPS/SMX), the model achieved a strong correlation of r = 0.79.
  • For a different organism (S. aureus DHFR/TMP), the model showed excellent performance with r = 0.91.
  • Comparison to SIFT: The model significantly outperformed SIFT, a commonly used tool for predicting the functional impact of amino acid substitutions, particularly for multiple mutations. SIFT was often unable to provide meaningful predictions or showed poor correlation.

Limitations and Future Work

The authors acknowledge several limitations:

  • Inability to Quantify Insertions/Deletions: The model is currently designed for missense mutations (substitutions) and cannot directly quantify the effects of insertions or deletions.
  • Limited Scope of Resistance Mechanisms: The model focuses on mutations within the drug target protein. It cannot account for other resistance mechanisms, such as gene amplification, altered drug permeability, or efflux pumps. This means its correlation with minimal inhibitory concentrations (MICs), which reflect overall resistance, might be limited.
  • Drug Target Specificity: While transferable, the model's performance might be less robust for inhibitors with significantly different binding modes compared to the training data.

Future work could involve extending the model to include insertions and deletions, integrating it with other resistance mechanisms, and applying it to a wider range of drug targets and pathogens. The authors envision its use in quantitatively informing researchers about resistance levels in sequenced strains and predicting future resistance-conferring mutations.

Summary updated

Abstract

Pneumocystis pneumonia (PCP) is an opportunistic infection that occurs in humans and other mammals with debilitated immune systems. These infections are caused by fungi in the genus Pneumocystis, which are not susceptible to standard antifungal agents. Despite decades of research and drug development, the primary treatment and prophylaxis for PCP remains a combination of trimethoprim (TMP) and sulfamethoxazole (SMX) that targets two enzymes in folic acid biosynthesis, dihydrofolate reductase (DHFR) and dihydropteroate synthase (DHPS), respectively. There is growing evidence of emerging resistance by <i>Pneumocystis jirovecii</i> (the species that infects humans) to TMP-SMX associated with mutations in the targeted enzymes. In the present study, we report the development of an accurate quantitative model to predict changes in the binding affinity of inhibitors (<i>K</i><sub>i</sub>, <i>IC</i><sub>50</sub>) to the mutated proteins. The model is based on evolutionary information and amino acid covariance analysis. Predicted changes in binding affinity upon mutations highly correlate with the experimentally measured data. While trained on <i>Pneumocystis jirovecii</i> DHFR/TMP data, the model shows similar or better performance when evaluated on the resistance data for a different inhibitor of PjDFHR, another drug/target pair (PjDHPS/SMX) and another organism (<i>Staphylococcus aureus</i> DHFR/TMP). Therefore, we anticipate that the developed prediction model will be useful in the evaluation of possible resistance of the newly sequenced variants of the pathogen and can be extended to other drug targets and organisms.
BibTeX
@article{baker2016quantitative,
  author = {Frazier N. Baker and Melanie T. Cushion and Aleksey Porollo},
  title = {{A Quantitative Model to Estimate Drug Resistance in Pathogens}},
  journal = {Journal of Fungi},
  year = {2016},
  doi = {10.3390/jof2040030},
}
Figure