TL;DR
This paper tackles a critical blind spot in drug discovery: the impact of protein isoforms generated by alternative splicing. By analyzing drug-target interactions at the isoform level, the researchers found that a significant majority of drugs may miss intended targets or hit unintended ones due to isoform variations, highlighting the need for isoform-aware drug design.
What Problem Does This Paper Address?
The drug discovery process is notoriously complex and expensive, with a high failure rate. A major reason for this is the assumption that a single gene produces a single protein product. However, alternative splicing, a fundamental biological process, allows a single gene to generate multiple protein isoforms, each with potentially different functions and interactions. Current drug discovery efforts often overlook these isoforms, focusing solely on the "canonical" or most common version of a protein. This oversight can lead to drugs that are ineffective because they don't interact with the relevant isoform, or worse, cause unexpected side effects by binding to unintended isoforms. This paper aims to bridge this gap by computationally analyzing how alternative splicing affects drug-target interactions, specifically for small molecule cancer drugs.
Methods
The researchers developed a computational pipeline to analyze drug-target interactions at the isoform level. Key steps included:
- Data Curation: Integrating data from multiple public databases (DGIdb, Ensembl, BioLiP, DrugBank, ChEMBL) to create a comprehensive list of FDA-approved and investigational cancer drugs and their targets, along with known ligand-binding residues.
- Isoform Alignment: Using multiple sequence alignment tools (like msa package with Cluster Omega) to compare the binding pocket regions of different protein isoforms for the same gene. This allowed them to identify differences in amino acid sequences within these critical interaction sites.
- Structure Modeling and Docking: For a subset of interactions, they built 3D structural models of protein isoforms and performed in silico docking simulations using Schrödinger suite (Glide docking tool) to assess how drugs bind to different isoforms at a structural level.
- Expression Analysis: Analyzing gene expression data from TCGA and GTEx to understand how different isoforms are expressed in normal versus cancer tissues. This helped in identifying "onco-isoforms" (upregulated in cancer) and potential tumor suppressors (downregulated in cancer).
Datasets
The study utilized several publicly available datasets:
- Drug Gene Interaction Database (DGIdb): For initial drug-target interactions.
- Ensembl: For gene and protein isoform annotations.
- BioLiP: For ligand-protein interaction data and binding site residues.
- DrugBank and ChEMBL: For drug information and standardization.
- The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) projects: For gene expression data across various cancer types and normal tissues.
- PDB: Protein Data Bank for structural information.
- Pfam: For protein domain annotations.
- FusionGDB: For information on fusion proteins.
The researchers also made their curated isoform-level drug-target interaction data publicly available.
Key Insights and Ideas
- Isoform Diversity is the Norm: The study confirmed that most cancer drug targets are not single proteins but rather a family of isoforms, with an average of 5.22 isoforms per gene among the curated drug targets. This underscores the pervasive nature of alternative splicing.
- Drugs Often Miss the Mark: A striking finding is that 76% of the analyzed drugs either miss a potential target isoform or target isoforms with varied expression patterns in normal tissues. This suggests a significant disconnect between how drugs are designed and the actual protein landscape they encounter.
- Structural Differences Matter: Beyond sequence variations, the study revealed that ligand-binding pockets of different isoforms can differ in size, shape, and electrostatic properties. This structural nuance can lead to differential drug binding affinities and modes, even if the core binding residues are similar.
- Expression Patterns Dictate Relevance: The expression levels of isoforms in cancer versus normal tissues are crucial. Drugs that target isoforms highly expressed in cancer but not in normal tissues are ideal. Conversely, drugs that target isoforms also expressed in normal tissues could lead to off-target effects.
- Redefining Drug Specificity: The paper introduces the concept of "isoform-level specificity," emphasizing that true specificity means targeting the correct isoform(s) in a specific context, not just a gene.
Key Results
- Out of 1,434 drug-target genes analyzed, 618 had 5 or more annotated protein isoforms.
- 76% of the analyzed drugs were found to either miss a potential target isoform or target isoforms with varied expression in normal tissues.
- Structural analysis of EGFR isoforms showed significant differences in ligand-binding pocket architecture, leading to varied drug binding scores for the same drug across different isoforms.
- Case studies on Imatinib, Erlotinib, and PD-0325901 highlighted how isoform expression and structural differences can influence drug efficacy and potential failure. For instance, a highly expressed, non-target isoform of MAP2K1 might have contributed to the failure of PD-0325901.
Limitations and Future Work
The authors acknowledge several limitations:
- Data Availability: The mapping of isoforms between public databases and literature can be inconsistent, and many previously documented isoforms are not found in standard databases like Ensembl. This hinders comprehensive structural and functional annotation.
- Assumptions: The study relies on the assumptions that isoforms beneficial for cancer progression are overexpressed and should be inhibited, and vice versa. These assumptions might not always hold true.
- Protein-Level Expression Data: The analysis primarily uses transcriptomic data. Direct protein-level expression data (e.g., from mass spectrometry) would provide stronger validation.
Future work should focus on:
- Robust Isoform Identification: Developing better methods to identify and annotate all protein isoforms.
- Computational Approaches: Improving structural modeling and prediction of isoforms.
- Biomarker Discovery: Identifying isoform-level cancer biomarkers and associating them with drug sensitivity.
- Database Integration: Enhancing public databases to better integrate isoform-level data and annotations.
- Experimental Validation: Conducting experimental studies to validate computational findings on isoform-specific drug interactions and their clinical implications.
Summary updated