The heterogeneity of the GAIN domain necessitates structure-based residue numbering
A comprehensive analysis of GAIN domains by means of multiple sequence alignments fails due to their low sequence identity and variable number of segments (α-helices and β-strands)8,9,35,36. Thus, to enable a comprehensive description of the GAIN domain, we opted for a structure-based approach, for which we generated a set of 14,435 GAIN domain models encompassing orthologs of the 33 mammalian aGPCR and 916 non-ortholog proteins with ColabFold/AlphaFold 246,47.
In order to assess the composition of both GAIN subdomains, we used structural alignments with GESAMT50 for indexing segments. Using this approach, the segment position in space determines its index instead of their sequence-based order, allowing the assignment of equivalent positions for GRN indexing in the context of variable domain composition. Based on the complete set of GAIN domain models, we asserted that subdomains A and B are composed of two to six helices and 12 to 14 strands, respectively, which we indexed using the identifiers H1-6 and S1-14 (Fig. 2a). Subdomain B exhibits generally high segment conservation, with only strand 4 specific to subfamilies A and G (Fig. 2b). The composition of subdomain A is more variable. While the A, B, C, F and L subfamilies all have six helices, the D and G subfamilies show heterogeneity ranging from two to six helices (Fig. 2b). Structures reflecting subdomain A variability are for example the rat ADGRL1 with a six-helix bundle (Fig. 1c)8, and the human ADGRG1 GAIN domain with only Helix 4 and 636, the two most conserved helices in the dataset.
When looking at individual residue positions (to which GRN labels are assigned, Fig. 2c), notably the center regions of helix 3, 4 and 6 are more frequently occupied (occupancy referring to the fraction of models containing a segment in the dataset) than the extreme positions, highlighting varying helix lengths in the model dataset, with especially the L subfamily exhibiting longer helices. Aside from the residues in the less conserved strands 4 and 7, there is high occupancy in subdomain B. The unindexed GAIN domain loops connecting the structured elements show very different lengths, frequently exceeding 50 residues (Supplementary Fig. 6). Notably, a total of 84 homologs of ADGRA1 in 47 species have a GAIN domain, which is not found in human ADGRA18, whereas 78 GAIN domains were identified for ADGRE4, which is a pseudogene in human (Supplementary Table 1).
Generic residue numbering denotes corresponding GAIN domain amino acids across receptors
Based on the GAIN-GRN indexing of all 14,435 structural models, we created comprehensive alignments of the GAIN domain in structure and in sequence (Supplementary Fig. 1). These provide a novel utility to map data across all adhesion GPCRs and cross-map sequence-position specific data between homologs. A schematic of the GRN assignment process is outlined in Fig. 3. Each GRN consists of the segment identifier (e. g. Helix 6 = “H6”) and the respective index relative to the most conserved residue in the segment, separated by a dot (e.g. “H6.50”). In this example, representative GAIN domains from each aGPCR subfamily are structurally aligned, with the Cα-atoms corresponding to GRN positions (Fig. 3a). Using the backbone alignment positions as the basis for a sequence alignment, the most conserved position is identified here as the acidic E/D and gets assigned the “center” .50 index (Fig. 3b). A residue table of the aligned segments highlights variation in segment lengths and reveals positions with similar physicochemical properties, e.g. the H6.50 as acidic, H6.45, H6.46 and H6.52 as aliphatic, despite low sequence identity (Fig. 3c).
GPCRdb resources aiding use of GAIN GRNs
As part of the GPCRdb integration, we assigned the 6 helices (H1-6), 14 strands (S1-14), and 21 loops connecting segments (h1h2, h2h3, h3h4, h4h5, h5h6, h6s1, s1s2, s2s3, s3s4, s4s5, s5s6, s6s7, s7s8, s8s9, s9s10, s10s11, s11s12, s12s13, s13gps, gpss14, s14tm1) to all Class B2 (Adhesion) sequences. In addition, the GPS motif has a separate assigned segment with the conventional GPS-2, GPS-1 and GPS + 1 notation for the three residues closest to the catalytic site. With the introduction of these new segments, researchers can apply the GPCRdb toolkit to the whole, or selected parts of the GAIN domain. We updated the snake plots of the Class B2 (Adhesion) GPCRs to contain the segments of the GAIN domain (Fig. 4b). The snake plots can be found on the Receptor page (https://gpcrdb.org/protein/) with custom coloring options51. Along the already provided data, GAIN domain data is also accessible programmatically via REST API.
Consensus contacts stabilizing the GAIN domain fold
Addressing of GAIN domain positions via GRN enables mapping any GRN-label-dependent information across aGPCR homologs. Corroborating the analysis of tertiary structures, we exploit the GRN indexing to consolidate pairwise residue-residue contacts - particular to each structure - into unified consensus contacts. The entirety of GRN-label pairs occurring over the complete dataset with a given frequency yields the GAIN domain contactome, shown in Supplementary Fig. 7 as a flareplot. This plot represents a contact matrix, individually resolving consensus contacts at the residue-level while highlighting contact relationships between the different GAIN segments. The importance of H6 as a “hub” connecting the subdomains A and B is clearly seen with highly conserved contacts to primarily S6 and also S2, S8, S10, and S14. Furthermore, H4 partially tethers subdomains A and B via highly conserved contacts to S1 and S2. In Supplementary Tables 3 and 4, the most frequent inter-domain and GPS contacts are listed individually, respectively. Additionally, we coarse-grained the contactome into the GAIN segments (Supplementary Fig. 8) reproducing the tethering structure of the segments regardless of individual contacts, further highlighting H4 and H6 as the segments mediating contacts between both GAIN subdomains.
A map of cancer-enriched mutations in the GAIN domain
The GAIN domain, present in 31 of 32 human aGPCRs, is a mutational hotspot affected in various pathologies6,8,52,53. To find cancer-enriched positions and differentiate them from variance-enriched positions, we adopted the cancer-enrichment score from Wright et al.54 for all 31 human aGPCR GAIN domains (Fig. 5a-b) indexed by the GAIN-GRN. While generally, enriched positions of both types are distributed throughout the GAIN domain (Fig. 5c), eight of the ten most cancer-enriched GRN positions are found in subdomain B carrying the TA (Fig. 5a, Supplementary Fig. 2). We identify a “VWWL” motif consisting of four conserved top-ten cancer-enriched residues S7.50, S10.50, S11.50 and S14.50. This buried motif is located in direct vicinity of the GPS cleavage site (Fig. 5d-e, Supplementary Fig. 1), with mutations known to affect GAIN domain autoproteolysis and TA function: the conserved leucine at the S14.50 position (Supplementary Fig. 1) is a TA residue deeply buried into the orthosteric binding site of the 7TM domain in active aGPCR-7TM structures19,30–34,55, and its mutation led to altered receptor activity10,22,56,57. Strikingly, the mutation of any tryptophane within the “VWWL” motif causes loss-of-function in rat ADGRL18. With a comprehensive analysis of human aGPCR GAIN domains, we find a total of 46 cancer-enriched positions (Supplementary Fig. 2). By using the GAIN-GRN, homologous cancer enriched residue positions can now be assigned to any GAIN domain. This allows the transferal of positional information between GAIN domains in different species, particularly from human to model organisms, such as D. melanogaster, C. elegans or D. rerio 58–61. The functional analyses in these model systems can now provide valuable insights into the molecular causes of cancer mutations in human in future studies.
GAIN domains of PKD1/PKD1-like proteins possess an extended topology
The only other protein family known to contain GAIN domains are PKD (polycystic kidney disease)1/PKD1-like proteins (in short here PKD1; also referred to as polycystin-1[PC1])8. Mutations in PKD1 are responsible for the majority of autosomal dominant kidney PKD, a devastating disorder that entails the development of cysts in the kidney and other organs leading to their eventual failure62. PKD1 GAIN domains display similar molecular properties as aGPCR GAIN domains with autoproteolytic cleavage resulting in a bipartite NTF-CTF protein layout after proteolysis63. Enabling the comparison and transferal of experimental and mutational knowledge between aGPCR and PKD1 GAIN folds is the basis for the understanding of similarities and differences between the two, and can offer valuable insights into the cell biological and physiological consequences of GAIN domain functions. However, thus far such transfer has been obstructed by the lack of clear homology assignments of primary and secondary structural positions between aGPCR and PKD1 GAIN domains. Thus, we next employed the GAIN-GRN scheme to allocate positional labels in PKD1 GAIN domains and compare them to those of aGPCR GAIN folds.
Since no experimental structure of PKD1 GAIN domains is available yet, we prepared 2,738 structural models analogously to the aGPCR dataset63–65. We applied the GAIN-GRN scheme to the models, which on average resulted in four subdomain A α-helices and twelve subdomain B β-strands recognized by the GAIN-GRN method (Fig. 2b), thus structural elements homologous to aGPCR GAIN domains. Interestingly, we also observed differences to aGPCR GAIN domain layouts as the PKD1 GAIN domains showed an additional β-sandwich fold, which contains an extension of S10 and C-terminally elongated TA. Finally, we also observed up to a total of eight subdomain A helices with additional, unindexed helices (Supplementary Fig. 3).
In sum, the GAIN-GRN scheme provides a robust strategy to allocate structural homologies at the primary and secondary levels also to GAIN folds of PKD1 molecules, which now renders positions in both GAIN domain types comparable to one another.