Domain- and Lineage-Specific Heterogeneity in Selective Constraint on Gliomedin (GLDN), a Node of Ranvier Gene Restricted to Jawed Vertebrates

<title>Abstract</title> <p>Almost every protein-coding gene sits under some degree of purifying selection, so a single gene-wide dN/dS value tells us little on its own. The more useful question is where the constraint sits: in which domain, on which lineages, and over what phylogenetic depth the gene can be traced at all. We took gliomedin (GLDN) as a test case. GLDN is the Schwann cell-derived glial ligand for the axonal cell adhesion molecules neurofascin-186 and NrCAM, and it is required for the molecular assembly of the nodes of Ranvier. Part of our motivation came from an earlier genome-wide association study by our group, in which the GLDN locus gave a suggestive but sub-genome-wide-significant signal for bladder cancer risk alongside reduced nuclear GLDN expression in more aggressive tumors. We assembled an orthology-verified set of 539 jawed vertebrate GLDN coding sequences and put three questions to it. The first concerned the two structurally distinct domains of the protein. Maximum-likelihood codon models (PAML), cross-validated against an independent framework (HyPhy), gave ω = 0.165 for the ligand-binding olfactomedin (OLF) domain against ω = 0.365 for the adjacent repetitive collagen-like domain, a difference of more than twofold. The second concerned lineages: mammalian branches proved significantly more constrained (ω = 0.130) than non-mammalian branches (ω = 0.170). The third concerned the age of the gene. Systematic sequence and phylogenetic searches recovered no GLDN ortholog in jawless vertebrates (Petromyzontidae, Myxini), and every detectable homolog belonged to a paralogous olfactomedin-family gene. GLDN thus appears confined to the jawed vertebrates, the same group that has myelin. One site-selection method flagged a candidate positively selected residue; in the solved crystal structure of the human GLDN OLF domain that residue falls at the disordered C-terminal boundary of the domain. Independent site-selection tests did not reproduce the signal, and we read the position instead as one of relaxed structural constraint. None of these three patterns would be visible in a single gene-wide ω.</p>