Structure Reveals Hidden GPCR-Like Proteins in Dark Proteome
When the Tools We Trust Go Blind
Sequence similarity has long been the workhorse of protein annotation. If a new protein looks like a known one, we assume it does the same job. That assumption has served biology well for decades. But it has also left blind spots — proteins whose sequences resemble nothing familiar, yet whose three-dimensional shapes tell a different story. A team mining millions of AlphaFold2 models went looking in exactly those blind spots. AlphaFold2 is a computational tool that predicts a protein’s 3D shape from its amino acid sequence. What the team found were two human families of what they call “superdark” seven-transmembrane proteins — TM184 and PRRT — that share structural homology with G-protein-coupled receptors, or GPCRs. GPCRs are the largest family of receptors in the human body, sitting in cell membranes and relaying signals from outside to inside. The superdark proteins had never been annotated as GPCRs because their sequences gave no hint. Their shapes did.
The researchers did not stop at structural resemblance. They tested whether these proteins actually behave like GPCRs. They do. The proteins recruit beta-arrestin — a protein that normally shuts down GPCR signaling — and they undergo phosphorylation by GPCR kinases, or GRKs, enzymes that mark activated receptors. These are hallmark GPCR activities, measurable and specific. The finding matters because it means structure-based discovery can reveal functions that sequence-based methods miss entirely. The dark proteome — the vast set of proteins with no known function — just got a little smaller.
Among the two families, one protein stood out. TM184C is the most broadly expressed and the most evolutionarily conserved of the superdark GPCR-like proteins. Broadly expressed means it shows up in many different cell types. Evolutionarily conserved means it has changed very little across species, which usually signals an important job. But TM184C does something unusual for a GPCR-like protein: it does not sit on the plasma membrane, the outer surface of the cell. Instead, it localizes to highly dynamic intracellular vesicles — small bubble-like compartments inside the cell that move around. That localization alone sets it apart from every classical GPCR.
Two Lenses, One Protein
The vesicles carrying TM184C move along microtubules, the cell’s internal railway system. They accumulate in cell projections — outward extensions of the cell surface. And they promote the formation of tunnelling nanotubes and tumour microtubule-like intercellular connections. Tunnelling nanotubes are thin bridges that connect one cell to another. Through these bridges, cells share organelles — the functional compartments inside a cell. This is not a passive process. It requires the TM184C C-terminal tail, the protein’s trailing end, and a specific sequence called an arrestin code motif. That motif links the protein’s GPCR-like regulation — beta-arrestin and GRK activity — directly to vesicle function and intercellular connectivity. In other words, the same molecular machinery that controls GPCR signaling also controls how cells build bridges to one another and exchange material.
A second function emerged from the same protein. TM184C constrains autophagic flux. Autophagy is the cell’s recycling system: it wraps up damaged components in a membrane, forms a structure called an autophagosome, and delivers the contents for breakdown. TM184C limits this process by restricting LC3B lipidation — the chemical attachment of a lipid to the LC3B protein — and by limiting autophagosome accumulation. Fewer autophagosomes form when TM184C is active. This role is deeply conserved. Human TM184C can restore autophagic body homeostasis in yeast that lack their own homologue, Hfl1. Yeast are single-celled organisms separated from humans by more than a billion years of evolution. If a human protein can do the job of a yeast protein, the function is ancient and fundamental.
The research points to a broader principle: exploring the understudied human proteome pays off. Independent groups are converging on the same idea — the Human Protein Atlas consortium and the AlphaFold Protein Structure Database team at EMBL-EBI are both cataloguing unannotated human proteins by predicted structure rather than sequence, building the same kind of map from different institutions. The superdark proteins were invisible to sequence-based annotation. Structure prediction made them visible. Functional assays confirmed they are real GPCR-like signaling proteins with roles in autophagy, intercellular connectivity, and material exchange. TM184C is now identified as an ancient GPCR-like regulator of all three. The work shows that combining large-scale structural modeling with targeted experiments can illuminate parts of the dark proteome that have resisted every other approach. The protein was always there. The tools to see it were not.
Sources
2. AlphaFold2