Inside every human cell, thousands of proteins must find their way to the correct address — the nucleus, the mitochondria, the cell membrane, or the tangled vesicles of the endosomal system. A new study published in Nature Structural & Molecular Biology reveals that this molecular postal service is controlled, in part, by some of the smallest players in the genome: microexons, stretches of DNA less than 30 nucleotides long that encode just a handful of amino acids. Researchers led by Peter Kjer-Hansen and Robert J. Weatheritt at the University of New South Wales and the Garvan Institute of Medical Research have shown that alternative splicing — the process by which a single gene can produce many different protein variants — fine-tunes the interfaces between proteins in ways that dramatically alter where they end up in the cell. The findings suggest that these tiny insertions, often shorter than five amino acids, punch far above their weight in organizing the proteome.
Alternative splicing occurs in nearly all multiexon human genes, allowing cells to include or skip particular coding segments and thereby generate multiple protein isoforms from a single gene. While the sheer scale of this proteome diversity has long been debated, recent mass spectrometry efforts have confirmed that most well-expressed alternative transcripts do give rise to detectable proteins. What has remained murky is how this diversity translates into function — specifically, how splicing decisions reshape the web of protein-protein interactions and determine subcellular localization. The new study tackles this question by combining network modeling, high-content imaging, machine learning, structural bioinformatics and classical biochemistry into a single integrated framework.
The team began by constructing an interface-resolved map of the human interactome, drawing on curated interaction data from resources including UniProt, DOMINO, IntAct, 3DID and the Eukaryotic Linear Motif database. In this network, every edge represents not just a protein-protein interaction but a specific, experimentally mapped contact surface. By integrating differential splicing data across 75 tissues and cell types, the researchers identified frame-preserving alternative splicing events whose encoded protein regions overlapped at least one mapped interface residue. Roughly 6.5 percent of all annotated interfaces turned out to be subject to splicing regulation, defining a proteome-wide landscape of interface-level variation concentrated at highly connected modular hubs.
Percolation analysis — a technique borrowed from network science in which edges are progressively removed to see how the network fragments — revealed a striking division of labor. The most tissue-specific exons preferentially rewired interactions locally, within functional modules such as individual signaling pathways, without breaking the overall connectivity of the interactome. In contrast, more tissue-wide splicing events, and microexons in particular, were significantly enriched among interfaces that connect distinct functional communities. In other words, microexons tend to sit at the junctions between pathways and complexes, occupying network positions that modulate crosstalk between different cellular processes. Regression modeling confirmed that this intercommunity effect depended on both the degree of tissue regulation and microexon status, suggesting that these tiny exons represent a specialized class of tissue-specific events with a disproportionate influence on how the proteome is organized.
The analysis also showed that tissue-specific splicing preferentially targets multidomain, multifunctional proteins — those built from many distinct binding modules and capable of interacting with proteins, DNA, RNA and lipids. Such proteins appear to tolerate local changes in their binding surfaces without losing their core activities. Crucially, the researchers found that proteins regulated by tissue-specific splicing also showed greater subcellular localization diversity, being found in more cellular compartments than their less-spliced counterparts. This association was strongest among highly modular proteins, hinting that splicing and spatial deployment are mechanistically linked. A computational screen using the protein language model ProtGPS predicted that nearly 28 percent of tested exons would redistribute their host proteins across subcellular compartments — a figure compelling enough to justify direct experimental testing.
That test came in the form of a high-content imaging screen. The team cloned 68 protein isoforms across 32 genes involved in endocytosis and cytoskeletal organization, each pair differing only by the inclusion or exclusion of a single alternative exon, and fused them to the fluorescent marker mCherry. After transfecting the constructs into HeLa cells and imaging them by automated confocal microscopy, the researchers extracted roughly 200 features from each cell’s fluorescence signal and trained machine-learning classifiers to distinguish between isoforms based on localization alone. The result: 38 percent of the 42 isoform pairs tested showed significantly different subcellular distributions, with microexons and longer exons contributing at nearly identical rates. Examples included the membrane-associated proteins ANK2, NUMB and OCRL, whose isoforms displayed graded differences in membrane association and granularity rather than wholesale relocalization.
The standout hit was sorting nexin 2, or SNX2, a multidomain protein involved in retrograde transport and recycling of internalized membrane proteins. Its two isoforms differ by a four-residue microexon encoding the sequence EIEE, inserted directly into the BAR domain — a coiled-coil structure that mediates lipid binding, dimerization and assembly into higher-order protein coats. AlphaFold2 modeling predicted that the insertion adds a single helical turn to the domain’s α3 helix, preserving the overall fold while making the surface more negatively charged and repositioning flanking charged residues, particularly arginine 451 and glutamate 452. The isoform without the microexon formed punctate, vesicular structures and aggregates at high expression, while the microexon-containing isoform showed a more diffuse cytoplasmic distribution — a difference that persisted even when untagged isoforms were expressed at near-endogenous levels in engineered rescue cell lines.
The most surprising finding came from mutagenesis. The team generated 18 SNX2 mutants and found that the presence or absence of the four-residue insertion, rather than the precise identity of its amino acids, was the main determinant of localization behavior. Even mutants with drastically altered exon sequences clustered with the microexon-containing isoform. Conversely, progressively mutating the flanking arginine 451 from positively charged to neutral to negative gradually shifted the exon-less construct toward the microexon-containing phenotype. Interaction assays reinforced the story: both isoforms formed canonical SNX-BAR complexes, but the microexon-containing isoform bound the insulin receptor roughly twice as strongly, and this enhanced binding followed the insertion rather than the exon’s composition. Together, these results demonstrate that a microexon can act as a structural spacer, physically displacing pre-existing charged residues and retuning the electrostatic landscape of an interaction interface without destroying the protein fold.
The broader implications are considerable. The study establishes a dual-mode model of splicing regulation: longer exons tend to act as digital switches that add or remove entire interaction modules within pathways, while microexons operate as analog fine-tuners at intercommunity interfaces, subtly adjusting electrostatics and residue positioning to modulate both localization and partner specificity. Because splicing-driven localization changes appear to be graded rather than binary, single exons can dial protein deployment up or down without wholesale relocalization — a mechanism with echoes in SRC kinase, where a microexon in the SH3 domain switches its peptide-binding preference, and in the gastric epithelium-specific exon of MLCK1, whose isoform-specific interactions are relevant to inflammatory bowel disease. The authors caution that their analysis is limited to frame-preserving events overlapping annotated interfaces, and that imaging assays relied primarily on tagged overexpression, though the SNX2 rescue experiments mitigated this concern. Even so, the message is clear: to understand where a protein lives and whom it talks to, scientists may need to look not just at the gene, but at the tiny splicing decisions that sculpt its interfaces one residue at a time.
Subject of Research: How alternative splicing and microexons reshape protein interaction interfaces and subcellular localization
Article Title: Alternative splicing fine-tunes modular protein interfaces to control protein localization
Article References: Alternative splicing fine-tunes modular protein interfaces to control protein localization. (n.d.). https://doi.org/10.1038/s41594-026-01875-8
Image Credits: AI Generated
DOI: 10.1038/s41594-026-01875-8
Keywords: alternative splicing, microexons, protein localization, protein-protein interactions, proteome organization, SNX2, BAR domain, high-content imaging, AlphaFold2, electrostatics, interactome, RNA splicing
