1 2 3 GPCR-G protein selectivity – a unified meta-analysis 4 5

20 Two-thirds of human hormones and one-third of clinical drugs act on membrane receptors that couple to G 21 proteins to achieve appropriate functional responses. While G protein transducers from literature are 22 annotated in the Guide to Pharmacology database, two recent large-scale datasets now expand the receptor23 G protein ‘couplome’. However, these three datasets differ in scope and reported G protein couplings giving 24 different coverage and conclusions on GPCR-G protein signaling. Here, we report a meta-analysis unifying 25 GPCR-G protein coupling, by standardized normalization and consensus support, into a common coupling 26 map. This unravels novel consensus couplings for receptors supported by two independent sources and 27 insights on coupling selectivity of GPCRs and classification of co-coupling G proteins. The coupling 28 protocol, map and selectivity resources will promote advances in receptor research and cellular signaling 29 towards the exploitation of G protein signaling specificity in design of safer drugs. 30


31
G protein-coupled receptors (GPCRs) represent the largest family of proteins involved in signal propagation across biological 32 membranes. They recognize a vast diversity of signals going from photons and odors to neurotransmitters, hormones, and cytokines

33
(1). Their main signaling modality involves the engagement and activation of G proteins. G proteins are heterotrimeric proteins 34 consisting of a a, b and g subunits that dissociate to a and a bg upon activation by a GPCR. G proteins are named by their a subunit 35 (16 in human) and are divided into four families which share homology and downstream signaling pathways: Gs (Gs and Golf), Gi/o 36 (Gi1,Gi2,Gi3,Go,Gz,Gt1,Gt2 and Ggust), Gq/11 (Gq,G11,G14 and G15) and G12/13 (G12 and G13). A GPCR's activation of G proteins 37 can be very selective or promiscuous and change upon ligand-dependent biased signaling that alters its profile on the G protein 38 subtype or family levels. The pleiotropic signaling and ligand-dependent bias of GPCRs pose a grand challenge in human biology 39 to map the differential activation of specific G proteins.

40
The IUPHAR/BPS Guide to Pharmacology (GtP) database contains reference data from expert curation of literature (1). GtP 41 couplings covers 253 GPCRs and the four G protein families. The G protein families have been classified into "primary" and 42 "secondary" transducers without quantitative values. Recently, the Inoue group determined the first large-scale quantitative 43 coupling profiles of 148 GPCRs to the Gq wildtype and 10 G protein chimeras employing a TGF-α shedding assay (2) (NTS1 and 44 TRH1 added herein making it 150 receptors). Those chimeras consist of a Gq backbone in which the six most C-terminal Ga 45 residues -a part of the H5 domain inserting to the intracellular receptor cavity -have been replaced to represent all 16 human G 46 proteins (five of which have identical sequences to other G proteins, see below). In a paper accompanying the present analysis , the 47 Bouvier group quantified the couplings of 100 GPCRs to 12 G proteins: Gs, Gi1, Gi2, Go (GoA and GoB isoforms), Gz, Gq, G11, G14, 48 G15, G12 and G13 but not Golf (couples mainly to olfactory receptors), Gi3 and Gt1-2 (Transducin, couples to rhodopsin (visual) 49 receptors) and Ggust (Gustducin, couples to taste receptors) (3). The authors used novel enhanced bystander bioluminescence 50 resonance energy transfer biosensors that allow to monitor G protein activation (G protein Effector Membrane Translocation assay; 51 GEMTA) without need to modify the G protein subunits (except for Gs) or the receptors.

52
Here, we analyze the GtP, Inoue and Bouvier coupling datasets to determine confident couplings supported by at least two 53 independent sources, including novel couplings discovered jointly by the two latter sources. We establish a scalable protocol to 54 normalize quantitative G protein couplings, combine Emax and EC50 into a common log(Emax/EC50) (4) value and aggregate subtypes 55 to allow comparisons across G protein families. On this basis, we develop a unified map of GPCR-G protein couplings that is also 56 made available in the GPCRdb database hub (5), describe GPCR-G protein selectivity across an unprecedented number of receptors 57 and coupling data points and reveal correlated co-couplings.

Each profiling study doubled the receptors' average number of G protein family couplings 62
To obtain an overview of the coverages of GPCR-G protein coupling sources, we compared all couplings reported by both the 63 Bouvier (3) and Inoue (2) groups and annotated in the Guide to Pharmacology database (GtP (1)) ( Fig. 1a). This shows that the 64 three sources together comprise couplings for 265 (67%) out of the 398 non-olfactory GPCRs and that 70 of these receptors are 65 present in all datasets. The Bouvier and Inoue datasets have collectively quantified individual G protein couplings of 178 receptors 66 using one assay, whereas the remaining 87 receptors have so far only been annotated in GtP on the G protein family level from a 67 multitude of publications and assays (Fig. 1a).

68
To allow the comparison of coupling densities and distributions across datasets, we selected the Emax threshold (1.4 standard 69 deviations above basal) that gives the best agreement between the Bouvier and Inoue dataset. We believe that this is the best 70 possible means to estimate what is correct data (rather than false negative/positive couplings), as large-scale information about 71 what G proteins and GPCRs coupling in nature is not available. This cut-off is more stringent the minimum of 3% signal above 72 basal used in Inoue's original report (2). We also aggregated G protein subtype couplings of the families (see Methods). This 73 reveals that while GtP covers the largest number of receptors they have relatively few couplings -62% of all GPCR-G protein 74 family pairs are 'non-couplers' (or not tested or annotated) compared to 35% in the Bouvier and 25% in the Inoue dataset (average 75 of 1.5, 2.6 and 3.0 G protein families per GPCR, respectively; Fig. 1b

85
we sought to determine to which extent they report the same couplings for the same GPCRs i.e., their 70 common receptors (data 86 in tab 'BIG-QualComp' in Spreadsheet S3). We find that all three sources agree on 39 (55%) of Gs, 45 (64%) of Gi/o, 35 (50%) of 87 Gq/11 and 18 (25%) of G12/13 couplings/non-couplings (mean 34 (49%), Fig. 2a) showing that the agreement varies substantially 88 across the individual G protein families. When instead analyzing the agreement of just the quantitative studies (excluding GtP) and 89 the G protein subtype level (Fig. 2b), it increases to an overall of 68% and ranging from at least 53% (for G15) up to 81% (for GoA).

90
These findings define a sizeable reference set of consensus G protein couplings and show that consistent large-scale profiling 91 studies generate more comparable results than literature.

92
We next identified the 'novel' G protein couplings for which a family annotation is missing in GtP but have high confidence 93 from dual support by the Bouvier and Inoue groups (black in Fig. 2b). This revealed 38 receptors with novel couplings to 101 G 94 proteins distributed across all families: Gs,:4, Gi/o:15, Gq/11:10 and G12/13:21 (Fig. S1). The largest expansions -an increase by 95 three of G protein families -was obtained for the histamine H1 and endothelin ETA receptors which was found to couple to all G 96 protein families but only have Gq/11-coupling in GtP. Whereas it could be expected that GtP would miss couplings, we also analyzed 97 if its expert curation excluded couplings that may be false positives as they are contradicted by both quantitative studies. This 98 uncovered such Gs-coupling to the α2C-adrenoceptor and cannabinoid CB1-2 receptors, G12/13 coupling to the purinergic P2Y2 99 receptor and Gi/o-coupling to the β2-adrenoceptor, which however had weak Gz and GoB coupling in the Bouvier study but did not 100 cross the signal threshold (3). Notably, this is only 2% (5/254) of all GtP's GPCR-G protein family pairs. Taken together, these 101 findings serve to quantify the expansion of the GPCR-G protein 'couplome' while also confirming the outstanding accuracy of the 102 expert annotation in the GtP database. Therefore, the large number of new couplings results mainly from that these couplings were 103 not investigated in literature and not inadequate curation.

105
Comparison across datasets reveal unsupported couplings requiring additional studies 106 While the Bouvier and recent Inoue datasets give an unprecedented mapping of GPCR activation on the single G protein subtype 107 level, the above analyses point to a need to explain their observed differences to establish which couplings to use where datasets 108 do not agree (Fig. 2). To this end, we analyzed the 70 common receptors for 'unsupported' and 'missing' couplings defined as 109 being present or absent in only one of the three datasets (i.e., non-couplings and couplings, respectively, that are proposed by one 110 but not supported by two datasets). For GtP, we consider coupling to a G protein subtype possible if a coupling has been observed 111 from the respective family.

112
For G15, 30% of Bouvier's couplings are unsupported (Fig. 3a). The Bouvier dataset has a large fraction, 37% of G15 couplings 113 that are new compared to GtP whereas the Inoue dataset has no new couplings and is missing 13% G15 reported in GtP. This

114
indicates that G15 couplings are underrepresented in both literature (which is annotated in GtP) as well as in Inoue's data. The few 115 known couplings for G15 is likely explained by its lack of expression in the cells most used for in vitro experiments, such as HEK293 116 cells (3) and by the lack or weak (at very high concentration) effect of the Gq/11 inhibitor tools YM-254890 and FR900359, 117 respectively (9). Furthermore, several of these novel G15 couplings were validated using Ca 2+ assays in the accompanying paper 118 from the Bouvier group (3). Taken together, these findings suggest that most of the unsupported G15 couplings in the Bouvier 119 dataset are real (true positive) unique couplings.

120
For Gs and G12, 27% and 24%, respectively of Inoue's couplings are unsupported (not present in Bouvier or GtP) and 29% of 121 Gs and 49% of G12 couplings are new compared to GtP (Fig. 3b). In contrast, the Bouvier dataset has no and 13% unsupported Gs 122 and G12 couplings, respectively and is missing 7% of Gs and 6% of G12 couplings in GtP. The lower representation of these two G 123 proteins in Bouvier compared to Inoue may be due to the lower assay window (tab 'DataStats' in Spreadsheet S1). Therefore, 124 additional studies would be needed to distinguish the real new Gs couplings in the Inoue dataset from non-couplers in GtP, which 125 are likely many since robust assays and selective inhibitors have long been available for Gs signaling. Furthermore, the Bouvier 126 data contains fewer G12 couplings than the Inoue group (32% relative 50%, overlap is 26%) that are new compared to GtP. Further 127 studies with alternative assays maintaining a high G12 sensitivity would be desirable to gain independent support to these couplings.
To explain the reason to the unsupported couplings, beyond those addressed above for G15, Gs and G12, we investigated their 138 potencies and efficacies (rows 106-110 in tab 'BIG-QualComp' in Spreadsheet S3). We find that unsupported couplings in the 139 Bouvier and Inoue datasets have average pEC50 values that are 0.4 and 1.5 log units -2.5-fold and 32-fold, respectively -lower 140 than the supported couplings. Furthermore, their corresponding average Emax values are 18% and 15% lower. This points to the 141 possibility that part of differences across datasets is due to weak couplings which can be difficult to distinguish from basal levels,

142
although the lack of support by GtP may suggest that some couplings are false positives. Both possibilities may be addressed by 143 future datasets or by optimization or development of the profiling platforms. However, until further data becomes available, we 144 recommend a requirement of two independent studies in support of a GPCR-G protein coupling. This is the criterion used herein 145 for the novel couplings and in our online coupling atlas in the GPCRdb database (5). Together these data highlight the importance 146 of large-scale G protein coupling profiling studies using complementary approaches to establish a robust atlas of GPCR coupling 147 selectivity. The Inoue and Bouvier datasets represent the first steps in this direction and there is no doubt that additional studies 148 will be forthcoming.

150
Coupling map reveals 73% promiscuity within and 49% selectivity across protein families

158
To gain insight into their levels of coupling selectivity, we intersected the G protein profiles of all receptors and counted the 159 number of coupling partners for GPCRs and G proteins (panels a, b and c, respectively in Fig. 5). On the G protein family level 160 (topmost in Fig. 5), our analysis spans 90 receptors with data only in GtP and 166 GPCRs with data from the Bouvier and/or Inoue 161 groups -totaling 256 receptors. We require couplings from the Bouvier and Inoue groups to be supported by a second dataset (GtP 162 couplings are already typically supported by several publications). We find that a GPCR couples to on average 1.7 G protein 163 families distributed as 126 single-, 83 double-, 34 triple-and 13 all-family activating receptors (49%, 32%, 13% and 5%, 164 respectively, Fig. 5b). The share of fully selective (single-family activating) receptors differs largely across G protein families 165 spanning from 6% for G12/13 to 22% for Gq/11, 26% for Gs and up to 40% for Gi/o (tab 'Fig_5' in Spreadsheet S5). Interestingly, all 166 fully promiscuous receptors are class A GPCRs: adenosine A1, adrenergic a1A,2A and β1, bradykinin B2, cannabinoid 1, 167 cholecystokinin CCK1, endothelin ETA, prostanoid FP, GPR4, histamine H1, lysophospholipid LPA4 and orexin OX2 receptors 168 (216 (76%) class A, 11 (61%) class B1 and 5 (23%) class C GPCRs have been profiled, so far). Conversely, a G protein family has 169 supported couplings to on average 112 GPCRs (28% of all receptors) distributed as Gs: 87, Gi/o 176, Gq/11: 134 and G12/13: 49 170 receptors (34%, 69%, 52% and 19%, respectively of all receptors, Fig. 5c). Given that 101 of the GPCR-G protein family pairs 171 tested by the Bouvier and Inoue groups represent novel couplings (above) more couplings are expected to be identified as expanding 172 and confirmatory studies emerge. Hence, whereas the results described here represent the currently known supported couplings, 173 the total 'couplome' will undoubtedly comprise additional yet undetected and unconfirmed couplings, especially among receptors 174 never profiled with a pan-G protein platform.

175
Within each G protein family (rows 2-5 in Fig. 5), on average 73% of GPCRs promiscuously activate all its members (Gq/11: 176 73%, G12/13: 66%, Gi/o: 75% and Gs: 80%). In contrast, activation of only one subtype of a G protein family is only observed for 11 177 Gs, 4 Gz, 1 G14, 10 G15, 5 G12 and 10 G13-coupled receptors (Fig. 4 or Spreadsheet S5). Most other receptors are activated by a 178 subset of G proteins in each family. Strikingly, P2Y1,4 (G15), P2Y14 (Gz) and GPR55 (G13) are fully selective also when considering 179 G proteins from all families i.e., they only couple to a single of the 16 human G proteins (1.4*SD cut-off applied, 6 Gs-coupling 180 receptors are left out, as Golf has so far only been tested by the Inoue group). However, the three purinergic receptors have additional 181 couplings although not supported by a second dataset (P2Y4 and P2Y14) or above the 1.4*SD cut-off (P2Y1 was below) and it is 182 possible that as the characterizations expand even fewer, or no receptors are found to engage only a single G protein.

183
In all, our meta-analysis of GPCR-G protein selectivity reveal a high coupling diversity spanning signaling via a single effector 184 to physiological ligand-receptor systems able to activate all four G protein pathways. Such selective or combined profiles, in 185 interplay with differential spatiotemporal expression (3), can be critical to achieve a specific physiological effect.

Gi/o correlation and G15 atypicality 189
To assess whether the evolutionary classification dividing G proteins into four families also represents their pharmacology, i.e., 190 couplings to receptors, we made pairwise correlations of G proteins using Jaccard indices (% of couplings to the same GPCRs) and 191 average log(Emax/EC50) differences for the receptors that couple both G proteins ( Fig. 6a-b). By combining these two measures and 192 calculating Pearson standard correlation coefficients (Fig. 6c), we find that the strongest correlations are seen for G proteins 193 belonging to the same G protein family (Gs: Gs-Golf: 0.79, G12/13: G12-G13: 0.79, Gio: min: Gi1-Gz: 0.73 and max: Gi1-Gi2: 0.99, and 194 Gq/11: min: Gq-G15: 0.73 and max: Gq-G11: 0.99). Altogether, these findings provided confirmation that the phylogenetic 195 relationships typically used to classify G proteins into families also reflect their pharmacological relationships.

196
G15 shares the largest fraction of receptor couplings with other members of the Gq/11 family but this fraction is relatively low 197 compared to other G protein families and the log(Emax/EC50) differences of G15 vs other Gq pairs are higher than observed for any 198 other intra-family comparison (Fig. 6b). This is reflected in G15 having the lowest mean Pearson standard correlation coefficient to 199 other members of the Gq/11 family (0.76 compared to 0.90-0.99 for the other subtypes) but similar to Gs and Golf (0.79) and to G12 200 and G13 (0.79). G15 also has a larger overlap than all other G proteins in its receptor couplings to another G protein family, the Gi/o 201 family (Fig. 6a). These findings confirm that G15 is a member of the Gq/11 family although with a weaker relationship and suggests 202 that this is due to a combination of differential receptor coupling selectivity and a activities.

203
When aggregating the receptor couplings and log(Emax/EC50) values for the four G protein families (Fig. 6d-

208
Together, this shows that G protein family pairs have evolved differing levels of receptor co-coupling or selectivity, which they 209 have achieved by weakening or abolishing the binding to the same GPCRs.    (Fig. 5b). We found that the highest selectivity exists 229 among Gs-versus Gi/o-coupled GPCRs which most rarely co-couple and, when they do, exhibit the highest average difference in 230 log(Emax/EC50) levels (Fig. 6). Given that Gs stimulates, and Gi/o inhibits production of the same cellular second messenger, cAMP 231 the selective activation of different receptors at different times presents a very plausible mechanism to modulate cellular cAMP 232 levels with temporal selectivity. Differential engagement of the Gi/o family could also have functional consequences through other 233 pathways, as Gi was recently shown to be required for scaffolding β-arrestin binding and signaling for some receptors (25).

234
In contrast, the Gi/o and Gq/11 families most frequently couple to the same receptors and they also have the largest number of 235 overall receptor couplings: 176 and 134, respectively (compared to 87 Gs and 49 G12/13-coupled GPCRs). The abundant coupling 236 of Gi/o-and Gq/11 to many receptors with dual (or more) pathways suggests that these G proteins have more versatile functions. 6 subtypes may be much more frequent than currently appreciated. G proteins that belong to the same family can have diverse 241 functional outcomes pertaining to effector engagement selectivity and kinetic profiles (26, 27), suggesting that bias within a G protein family may have physiological and/or therapeutic implications (28). The observation that many GPCRs can couple to more 243 than one family of G proteins but yet can show selectivity between members of a same family, opens important questions 244 concerning the structural determinants of such multifaceted selectivity profiles.

245
In all, our cross-dataset analysis has established a protocol and reference set aiding GPCR-G protein coupling studies. The 246 selectivity profiles are the most comprehensive to date spanning 256 receptors with a very diverse activation of a single to all G 247 protein pathways, and presented in a dedicated online atlas of G protein couplings integrated in the GPCRdb database hub (5). The 248 analyses and data presented herein will be very valuable to further our understanding of undercharacterised pharmacological 249 phenomena such as constitutive activity (29), pre-coupling of G proteins (16, 30) and ligand-dependent biased G protein signaling 250 (31), and to uncover their underlying determinants. They also present the foundation to integrate more coupling data as future 251 studies expand the characterization of the 'couplome'.

Study design 255
The primary objective of this study was to generate a unified map of G protein couplings (Fig. 4) across the three available large 256 datasets from the Bouvier (3) and Inoue (2) groups, and the Guide to Pharmacology database (1), respectively. To do this, we 257 identified the Emax standard deviation cut-off, quantitative normalization protocol and aggregation of G proteins into families 258 giving the best possible agreement between the GPCR-G protein couplings from the Bouvier and Inoue groups. This analysis also 259 involved assessing the agreement of the Bouvier and Inoue group datasets and determining the number of high-confidence novel 260 couplings supported by these two datasets but not reported in the Guide to Pharmacology database. Furthermore, it included a 261 benchmarking of techniques to determine which biosensor produces the most reproducible qualitative coupling determination 262 (coupling vs. non-coupling) for each G protein family. To this end, we compared the three datasets to pinpoint, for each dataset 263 and G protein family, the fraction of couplings supported by another dataset. The scientific analyses based on the map feature the 264 most comprehensive analysis to date of GPCR-G protein selectivity. This was done by intersecting the 11 G proteins tested by both 265 Bouvier and Inoue, within and across their families, with respect to the common and unique receptors that they couple to. Finally, 266 the receptor profiles were also used to classify G proteins and to determine any co-correlation between different G protein subtypes 267 and families to GPCRs.

269
Coupling datasets 270 Updated Spreadsheets containing the pEC50, Emax, basal signals and standard deviation (SD) values were supplied by Asuka Inoue.

271
Basal signals, spontaneous AP-TGF-a elease (in % of total AP-TGF-a expression) were recalculated from raw data of the previous 272 coupling-profiling campaign (2) and their SD values were computed from independent experiments (n ≥ 3). This file contains the 273 neurotensin 1 (NTS1) and Thyrotropin-releasing hormone (TRH1) receptors not included in the previous publication (2). In the 274 Inoue dataset, the protease-activated receptors PAR3-4 had negative pEC50 values (concentration of mU/ml because the ligand, 275 thrombin, was supplied with its enzymatic activity). For the easiness of integration into the coupling map, we added a value of 10 276 to their pEC50 values. Data qualities (sigmoidal curves) for the individual GPCR-G protein pairs were manually inspected and 277 concentration-response curves that did not converge nor exceed a threshold (typically, 3% AP-TGF-α release) were regarded as no 278 activity.

279
The Bouvier dataset (n ≥ 3) contained some datapoints that were included or excluded based on dedicated analyses. Firstly, we 280 excluded ligand-promoted responses of overexpressed receptors that were equivalent to those of endogenously expressed receptors 281 (yellow fill in tab 'B-EmEC' in Spreadsheet S1). Secondly, couplings with only approximate Emax and pEC50 values because the 282 dose-response curve did not converge were only included if supported, and not contradicted, by the Inoue and/or GtP datasets 283 (orange fill in tab 'B-EmEC' and analyzed separately in tab 'UnconvergedDRV' in Spreadsheet S1). This is because a coupling 284 with an unprecise quantitative value is better than no coupling, especially when making qualitative comparisons (coupling vs. non-285 coupling). Based on these criteria seven couplings were included: 5-HT1D-Gz, BLT1-G11, FFA3-G13, GPR4-Gi1-2, GPR84-GoA and 286 k-G12, and six couplings were excluded: CCR5-G14, CXCR5-Gq/G11/G14, GPR183-G13 and k-G13.

288
Standard deviation cut-off 289 We made a special investigation of couplings that have a full dose-response curve but an Emax less than 2 standard deviations (SDs) 290 from the basal signal (red fill in tab 'B-EmEC' and analyzed separately in tab tabs 'B-<2SDs' and 'I-<2SDs' in Spreadsheet S1).

291
To achieve the best possible separation of putative false and real but weak couplings, we identified the threshold value (a number 292 of SDs from basal signal) that gives the best agreement between the Bouvier and Inoue dataset among the common receptors and 7 G proteins tested by both groups. The obtained cut-off, 1.4 SDs was applied as a filter to exclude all Bouvier and Inoue Emax and 294 pEC50 values for couplings below this cut-off (columns with '>1.4SD' in the heading in tabs 'B-EmEC' and 'I-EmEC' in 295 Spreadsheet S1). As a note, while intra-day measurement error is small for the TGF-α shedding assay (typically, 1-2% AP-TGF-α 296 release), inter-day variability varies widely depending on cell conditions. Since the SD represents inter-day variability, the basal 297 SD cut-off removes more couplings than the SD cut-off of ligand-induced signal or the significance of individual experiments. As 298 a consequence, some of manually annotated couplings in the Inoue dataset are regarded as non-coupling by the basal SD cut-off 299 criteria, including for P2RY2 and P2RY6 that had no couplings above this cut-off.

301
Generating a subset of comparable GPCR-G protein couplings 302 To enable qualitative comparison of corresponding datapoints in the datasets from the Bouvier (32) and Inoue (2) groups, we used 303 the subset of 70 GPCRs present in both datasets and belonging to the same class, A (removed only two receptors from class B1).

304
The quantitative comparisons focused on a smaller subset of 51 GPCRs tested with the same ligand and excluded non-coupling 305 GPCR-G protein pairs, as they could not be represented by 0 values (due to e.g., an underrepresentation of couplings in a given 306 datasets and G proteins, see Results). All analyses herein included the 12 G proteins: Gs, Gi1, Gi2, GoA, GoB, Gz, Gq, G11, G14, G15, 307 G12 and G13. Golf and Gi3 could not be analyzed, as they had not been tested by the Bouvier group (Supplementary Table 1). The

308
Inoue data for the pairs Gi1-2, GoA-B and Gq and G11, respectively, were generated with identical chimera inserting the Ga C-terminal 309 hexamer into a Gq backbone (2). Qualitative analyses compared the presence or absence of each GPCR-G protein coupling while 310 the quantitative analyses were limited to common G protein couplings, i.e., data points in which both the Bouvier and Inoue groups 311 generated a pEC50 and Emax value.

313
Validation of comparability when Emax or reference agonist differ 314 To get an overview of the distribution of data in the two datasets, we determined the pEC50 and Emax mean, median, min, max, span 315 and standard deviation values and plotted box and whiskers plots for each G protein (Spreadsheet S1). This showed that the Emax 316 values vary much more across the different G proteins than pEC50 values. The Emax variation is largest in the Bouvier data wherein 317 G15 ranges across three orders of magnitude (16 to 1,067). The Bouvier dataset has low means and spans for Gs (42 and 47) and 318 G12 (71 and 101) indicating a narrow assay signal window (low signal-to-noise). We find that minimum-maximum normalization 319 (to 100%) of each G protein across receptors gives a more uniform distribution (SDs for Bouvier: [19][20][21][22][23][24][25][26][27][28][29][30][31]rightmost 320 plot pair in Spreadsheet S1).

321
Whereas both studies tested a majority of receptors with their endogenous ligand, surrogate agonists were used for 15 and 4 322 GPCRs in the data from the Inoue and Bouvier groups, respectively. Although those ligands were selected for their reference 323 character with similar pharmacology to the endogenous ligand, they could introduce differences in a receptor's G protein profile 324 due to ligand-dependent signaling bias. Analysis of GPCR-G protein couplers and non-couplers (tab 'BI' in Spreadsheet S2) shows 325 that on average 74% and 71% agreeing qualitative couplings (i.e., coupling vs. non-coupling GPCR-G protein pairs) for receptors 326 tested with the same and different agonists, respectively. This is a rather small difference providing confirmation that receptors 327 tested with different ligands can be compared on the qualitative coupling/non-coupling level and be included in the comparison of 328 the different G protein coupling datasets and determination of novel G protein couplings (Fig. S1).

330 Dataset integration into a unified coupling map -Normalization and log(Emax/EC50) values 331
To enable quantitative correlation of the Bouvier and Inoue couplings, we further filtered the 70 common GPCRs to yield 51 332 common class A GPCRs tested with the same ligand (excluding 29 receptors tested with different ligands and two class B1 GPCRs).

333
To assess the value of normalization, we calculated the average 'similarity' (Bouvier/Inoue ratio) of individual values (tabs ending 334 with '-sim' in Spreadsheet S4) and the 'linear correlation' (r2 value) of each receptor across all G proteins (reported below as 335 averages of individual couplings and G proteins, respectively) (tabs ending with '-r2' in Spreadsheet S4). Linear correlation was 336 only done for receptors with at least three common G proteins/families. Minimum-maximum normalized Emax values were 337 represented as percentage values while decimal values (0-1) were used for the calculation of log(Emax/EC50) values as 338 recommended in (4). For both the Bouvier and Inoue groups, the minimum and maximum represent the signal without (0%) and 339 with (100%) an agonist, respectively, in each experiment replica. The minimum Emax value was therefore set 0 while the maximum 340 was set to the highest value for the given G protein (first, use normalization) or receptor (second, tested but not used normalization).

350 Dataset integration into a unified coupling map -G protein family aggregation 351
Given that G proteins belong to families that are functionally grouped by sharing downstream signaling pathways, we next 352 investigated the best member-to-family aggregation scheme -specifically, whether the most comparable G protein family values 353 are obtained if using the maximum value from any single subtype or the mean of all subtype members. We found that aggregation 354 using max rather than mean values gives a better similarity for Emax (0.69 vs. 0.65), marginally lower similarities for pEC50 (0.89 355 relative 0.90) and identical similarities for log(Emax/EC50) (0.88). However, max performs better overall than mean in the

363
Statistical analysis 364 The aggregated sample size is n=3 or higher for all analyzed GPCR-G protein couplings. The specific sample size for each such    tested by the Inoue group. Inoue used the same chimera to represent the pairs GoA-GoB, Gi1-Gi2 and Gq-G11. (Table S1) second dataset (Fig. 4b) and panel A additionally includes GtP couplings (Fig. 4c). This analysis of the Gs family leaves out 11 514 receptors tested for coupling to Gs but not to Golf, and Golf couplings are only counted if there is a supported Gs coupling. All source 517 518