Intramolecular allosteric communication in dopamine D2 receptor revealed by evolutionary amino acid covariation

Significance

Characterizing relationships among protein structure, function, and evolution requires understanding the evolutionary constraints on each constituent residue of a protein. Previous studies have shown that structural information can be retrieved from evolutionary residue covariation in protein families. However, whether the evolutionary history in protein sequences informs on functional interactions between nonadjacent residues has been unclear. Here, we developed a method that uses evolutionary amino acid covariation to infer functionally coupled residue pairs in the dopamine D2 receptor. We discovered functional coupling between residue pairs that have coevolved mainly to control responses to dopamine and maintain them at wild-type levels. Our findings demonstrate the possibility of extracting the networks of intramolecular allosteric communication from evolutionary residue covariation patterns.

Abstract

The structural basis of allosteric signaling in G protein-coupled receptors (GPCRs) is important in guiding design of therapeutics and understanding phenotypic consequences of genetic variation. The Evolutionary Trace (ET) algorithm previously proved effective in redesigning receptors to mimic the ligand specificities of functionally distinct homologs. We now expand ET to consider mutual information, with validation in GPCR structure and dopamine D2 receptor (D2R) function. The new algorithm, called ET-MIp, identifies evolutionarily relevant patterns of amino acid covariations. The improved predictions of structural proximity and D2R mutagenesis demonstrate that ET-MIp predicts functional interactions between residue pairs, particularly potency and efficacy of activation by dopamine. Remarkably, although most of the residue pairs chosen for mutagenesis are neither in the binding pocket nor in contact with each other, many exhibited functional interactions, implying at-a-distance coupling. The functional interaction between the coupled pairs correlated best with the evolutionary coupling potential derived from dopamine receptor sequences rather than with broader sets of GPCR sequences. These data suggest that the allosteric communication responsible for dopamine responses is resolved by ET-MIp and best discerned within a short evolutionary distance. Most double mutants restored dopamine response to wild-type levels, also suggesting that tight regulation of the response to dopamine drove the coevolution and intramolecular communications between coupled residues. Our approach provides a general tool to identify evolutionary covariation patterns in small sets of close sequence homologs and to translate them into functional linkages between residues.

Identifying residues that coevolved to maintain or acquire fitness properties is critical for understanding protein structure, function, and evolution (1). Previous studies have shown that covarying residue pairs, those that exhibit correlated amino acid changes in large multiple sequence alignments, tend to form structural contacts (2–7), enhancing predictions of protein 3D structures (8–11). Covariation can also involve distal residues, but the function of these at-a-distance couplings is elusive and has been attributed to background noise, alternative protein conformations, or subunit interactions of protein homooligomers (5, 7, 12). Alternately, distal covarying residue pairs could indicate allosteric couplings (6, 13–18).

The possibility of capturing intramolecular allosteric communication by amino acid covariation analysis of protein family sequences has not been extensively explored. Nonproximal thermodynamic coupling between correlated residue pairs was noted in 274 PDZ domains (14), but the relationship to allostery is still debated (19, 20). It may be that distinctive allosteric mechanisms, even among close homologs, limit the extraction of allosteric couplings from sequences (13). Our previous identification of residues important for allosteric signaling within G protein-coupled receptors (GPCRs) using Evolutionary Trace (ET) (21–24) and strong conservation of some of the residues implicated led us to ask whether ET could also uncover couplings among protein sequence positions not in direct contact.

ET estimates the relative functional sensitivity of a protein to variations at each residue position using phylogenetic distances to account for the functional divergence among sequence homologs (25, 26). Similar ideas can be applied to pairs of sequence positions to recompute ET as the average importance of the couplings between a residue and its direct structural neighbors (27). To measure the evolutionary coupling information between residue pairs, we present a new algorithm, ET-MIp, that integrates the mutual information metric (MIp) (5) to the ET framework. We used dopamine D2 receptor (D2R), a target of drugs for neurological and psychiatric diseases (28), to test whether ET-MIp could elucidate the allosteric functional communications from amino acid covariation patterns and resolve the evolutionary distance at which the allosteric pathways of D2R homologs are sufficiently conserved to detect residue−residue coupling signatures. D2R is expressed in the central nervous system and responds to dopamine, the major catecholamine neurotransmitter. Canonical D2R signaling is effected by G_i/o class G proteins, which regulate ion channels (29, 30), MAPK kinases (31), phospholipase C (32), and inhibition of adenylyl cyclase (33). D1 class receptors (D1R and D5R) have lower affinities for dopamine (34–36) and activate adenylyl cyclase through G_s class G proteins. To characterize allosteric communication between covarying pairs of residues ranked as important by ET (ET residue pairs), we examined functional coupling for ligand binding affinities and downstream G_i activation induced by agonist-stimulated D2R.

Results

ET-MIp Identified Pairs of Residue Positions with Evolutionary Covariation Patterns.

We hypothesized that accounting for species divergence would improve the detection of functionally coupled residues over covariation analyses that ignore phylogenetic information. ET-MIp adds the mutual information metric (5) to ET to keep track of the phylogenetic distance at which a pair of residues vary (Fig. 1A; see Materials and Methods and Supporting Information for details). In ∼2,500 Class A GPCR transmembrane (TM) domain sequences, we found that residue pairs with high ET-MIp scores were more enriched for direct contacts in a reference structure (PDB 2RH1) compared with results obtained with leading alternative methods (5, 37, 38) (Fig. 1B), showing that GPCR phylogenetic information improves the coupling signal. Preliminary analysis of other protein families suggests that this result may be fairly general. This opens the possibility that ET-MIp also detects functionally relevant covariation among structurally distant yet coupled residue pairs.

Fig. 1.

ET-MIp decodes evolutionary correlations between residue positions. (A) Examples of residue positions displaying covariation patterns in the context of evolutionary trees. For simplicity, only residues from human sequences are shown. Covarying residue pairs were mapped onto the D3R structure (PDB 3PBL). Some have structural contacts (blue spheres), whereas others are distant in the structure (red spheres). These positions are hypothesized to be functionally coupled during evolution. (B) Receiver operating characteristic curves comparing the performance of ET-MIp and other mutual information (MI)-based methods, MIp (5), normalized MI (nMI) (38), and DCA algorithms, FN and DI (37), in identifying residues in contact (within 6 Å) in the structure of the β₂ adrenergic receptor (PDB 2RH1). Each method was applied to aligned sequences of ∼2,500 Class A GPCRs. The areas under the curves, which indicate the accuracy of the prediction, are 0.75 for ET-MIp, 0.68 for DCA–FN, 0.65 for DCA–DI, 0.65 for MIp, and 0.56 for nMI.

Residue and Sequence Selection.

To test predicted couplings experimentally, we selected 10 covarying ET residue pairs in D2R in which one or both of the residues were involved in allosteric pathways of D2R ligand responses (23), plus two more pairs. Most of these are predicted to be functionally important by ET and alter function upon mutation (23). These pairs cover a range of ET-MIp coupling scores and involve structurally distant residues (except T205M^5.54/L379F^6.41; see Table S1) whose couplings, if any, would be allosteric. To probe the role of functional coupling in discriminating between dopamine and serotonin during evolution, ET residues in D2R were mutated to the corresponding residues in the closely related 5-HT_2A serotonin receptor (5-HT_2AR), so only sites at which 5-HT_2AR and D2R differ were included. As single mutations, these substitutions still allow for a functional receptor (23). In addition to mutation selection, we considered the choice of sequences used for calculating scores, because this can strongly impact ET-MIp couplings. For structural contacts, ET-MIp can be applied to Class A GPCR sequences because they are all structurally similar. For functional allostery in D2R, tuned to a specific ligand and signaling bias, a more restricted alignment may be best. Accordingly, multiple alignments were tested (Class A, bioamine, dopamine, and D2Rs) and yielded distinct coupling scores (Table S2).

Table S1.

Distance between the covarying ET residue pairs

Covarying ET residue pair (Ballesteros–Weinstein number)	Distance between the two residues, Å
M117 (3.35)/L387 (6.49)	13.4
F202 (5.51)/Y213 (5.62)	13.2
V152 (4.42)/L171 (4.61)	23.3
I105 (3.23)/I195 (5.44)*	23.4
V191 (5.40)*/S409 (7.36)	18.7
N124 (3.42)/T205 (5.54)	7.0
N124 (3.42)/L379 (6.41)	8.2
T205 (5.54)/L379 (6.41)	3.6
M117 (3.35)/Y199 (5.48)	13.1
I48 (1.46)/F110 (3.28)	13.0
V83 (2.53)/V91 (2.61)	6.8
S193 (5.42)/C385 (6.47)	14.2

The distance between the covarying ET residue pairs was measured by mapping each covarying ET residue onto the structure of human D3R (PDB 3PBL).

^*The covarying ET residues in D2R are the same as the corresponding residues in D3R, except I195 (5.44) and V191 (5.40), which are valine and isoleucine, respectively, in D3R.

Table S2.

Evolutionary coupling potential for each covarying ET residue pair based on different input sequence sets

Mutation (Ballesteros–Weinstein number)	Above 50% sequence identity with D2R	Above 42% sequence identity with D2R	Above 35% sequence identity with D2R	Bioamine sequence set	Class A sequence set	Average evolutionary coupling potential*
M117F (3.35)/L387C (6.49)	43.8	69.3	57.6	99.6	34.0	60.9
F202L (5.51)/Y213I (5.62)	93.6	93.6	77.9	5.8	81.3	70.4
V152A (4.42)/L171P (4.61)	88.7	81.0	77.6	3.7	87.8	67.8
I105K (3.23)/I195F (5.44)	73.2	79.8	40.2	5.5	95.6	58.9
V191L (5.40)/S409N (7.36)	43.9	31.3	17.7	0.1	56.1	29.8
N124H (3.42)/T205M (5.54)	48.9	89.0	74.9	76.9	66.4	71.2
N124H (3.42)/L379F (6.41)	68.7	71.4	71.2	81.9	41.6	67.0
T205M (5.54)/L379F (6.41)	87.3	96.9	94.2	90.2	51.8	84.1
M117F (3.35)/Y199F (5.48)	83.8	81.9	94.6	96.1	41.3	79.5
I48T (1.46)/F110W (3.28)	94.7	97.5	91.3	97.7	85.8	93.4
V83L (2.53)/V91S (2.61)	93.7	98.0	97.2	95.8	60.7	89.1
S193G (5.42)/C385M (6.47)	49.9	88.8	91.4	97.3	88.8	83.2

Evolutionary coupling potential (normalized score) for each covarying ET residue pair was derived from the sequences sharing above 50%, 42%, or 35% identity with D2R, 402 bioamine receptor sequences, or 2,500 Class A GPCR sequences.

^*Average of all evolutionary coupling potentials predicted for each covarying ET residue pair. Scores range from 0 to 100, with higher scores indicating greater evolutionary coupling potential.

Functional Interactions Between Covarying ET Residue Pairs Maintained Dopamine Response at WT Level.

To test whether the selected ET residue pairs were functionally coupled, we first compared the effects of single and double mutations on dopamine efficacy using a fluorescence-based assay to study G_i activation induced by agonist-stimulated D2R (Fig. 2; see Materials and Methods and Supporting Information for details). For five pairs (V83L^2.53/V91S^2.61, M117F^3.35/Y199F^5.48, I48T^1.46/F110W^3.28, V152A^4.42/L171P^4.61, and N124H^3.42/T205M^5.54), activation of G_i in response to dopamine was unexpectedly decreased or restored to a near-wild-type (WT) level in the double mutants, even though one or both of the constituent single mutants showed significantly enhanced response (Fig. 2 A and C). These results indicate that covarying ET residue pairs help maintain dopamine responses at the WT level. A trend was that functional coupling was more apparent in pairs with high evolutionary coupling potential when calculated using sequences only from dopamine receptors (Fig. 2C). For example, the loss-of-function mutation L379F^6.41 is rescued by T205M^5.54, with which it has a strong evolutionary coupling potential, but not by N124H^3.42, with which the coupling potential is weaker (Fig. 2 B and C). To estimate the epistatic effect of double mutations, we used four standard models (product, logarithmic, minimal, and additive interaction models) (39, 40). Except for the minimal model, the five residue pairs that were functionally coupled to maintain WT dopamine responses yielded higher epistasis scores, and the epistasis scores correlated better with the evolutionary coupling scores when calculated using input sequences made up only of dopamine receptors (Fig. 2C and Figs. S1–S4). This suggests that the allosteric communication responsible for dopamine response is a unique evolutionary signature of dopamine receptors. Compared with direct-coupling analysis (DCA), Frobenius norm (FN), and direct information (DI) algorithms (37) (Tables S3 and S4), these ET-MIp results correlated better with experimental epistasis scores (Table S5).

Fig. 2.

Functional coupling between covarying ET residue pairs was observed at G_i activation by dopamine. (A and B) Examples of D2R activation curves for covarying ET residue pairs. Membrane potential changes induced by G_i activation in response to dopamine stimulation of D2Rs were detected with the membrane potential assay. HEK293 cells stably expressing TRPC4β were transiently transfected with negative control [pcDNA3.1(+)], WT, or mutant D2R plasmids, loaded with potential-sensing dye, and stimulated with 10 μM dopamine after baseline fluorescence was read for 30 s. Each trace was normalized by receptor surface expression after subtraction of the signal of negative control cells. (C) Maximal activation of mutant D2Rs, normalized to WT (bars indicate mean ± SEM, n = 3–8). Mutants were compared with WT using one-sample t test against WT $\equiv$ 1 (asterisks directly above bars), and mutants within a covarying group were compared with each other using one-way ANOVA followed by Bonferroni's multiple comparison test (asterisks with brackets) (**P < 0.001; *P < 0.05). Bars are color-coded according to the evolutionary coupling potential predicted using amino acid sequences sharing >35% identity with D2R; see Table S2 for values. (D and E) Dopamine dose–response curves for G_i activation were generated with the membrane potential assay as in A. Examples of nonadditive and additive effects of double mutations on the potency of dopamine are shown in D and E, respectively. (F) Dopamine EC₅₀ values for WT and mutant D2Rs were determined by dose–response curves (details in Table S6) and used for the log-additive analysis. Results represent mean ± SEM (n = 3–7; **P < 0.001; *P < 0.05; independent two-tailed Student’s t tests). The color scale is as in C. (G) Evolutionary coupling potential, predicted as described for C, is plotted against deviation from additivity $\left|\ln \left(E{C}_{50mutantA}/E{C}_{50WT}\right)+\ln \left(E{C}_{50mutantB}/E{C}_{50WT}\right)-\ln \left(E{C}_{50mutantAB}/E{C}_{50WT}\right)\right|$ using EC₅₀ data shown in F. The two were highly correlated (Pearson’s R = 0.762, P = 0.017). (H) Positions of the covarying ET residues, shown as spheres (Cα atoms), mapped onto the structure of D3R (PDB 3PBL). Different colors indicate different groups of covarying ET residue pairs.

Fig. S1.

Relationships between the evolutionary coupling potential and epistasis scores calculated using the additive interaction model. Epistasis scores calculated using the additive interaction model were computed from the equation ε^add = M_ab − (M_a + M_b − 1), where M_a and M_b are the functional scores of the single mutants and M_ab is the functional score of the double mutant. The functional score of each mutant was derived from the dopamine efficacy data, which were normalized to WT as shown in Fig. 2C. For each covarying ET residue pair, the epistasis score is plotted on the y axis, and the evolutionary coupling potential predicted based on the sequence sets of Class A GPCRs (A), bioamine family (B), and above 35% (C), 42% (D), and 50% (E) sequence identity with D2R is plotted on the x axis. (F) The average of all evolutionary coupling potentials predicted for each covarying ET residue pair is plotted against epistasis scores.

Fig. S4.

Relationships between the evolutionary coupling potential and epistasis scores calculated using the minimal interaction model. Epistasis scores calculated using the minimal interaction model were computed from the equation ε^min = M_ab − min(M_a, M_b), where M_a and M_b are the functional scores of the single mutants and M_ab is the functional score of the double mutant. The functional score of each mutant was derived from the dopamine efficacy data, which were normalized to WT as shown in Fig. 2C. For each covarying ET residue pair, the epistasis score is plotted on the y axis, and the evolutionary coupling potential predicted based on sequence sets of Class A GPCRs (A), bioamine family (B), and above 35% (C), 42% (D), and 50% (E) sequence identity with D2R is plotted on the x axis. (F) The average of all evolutionary coupling potentials predicted for each covarying ET residue pair is plotted against epistasis scores.

Table S3.

Ranking scores of each covarying ET residue pair computed by DCA–DI (37) based on different input sequence sets

Mutation (Ballesteros–Weinstein number)	Above 50% sequence identity with D2R	Above 42% sequence identity with D2R	Above 35% sequence identity with D2R	Bioamine sequence set	Class A sequence set	Average evolutionary coupling potential*
M117F (3.35)/L387C (6.49)	61.5	41.0	49.0	70.9	9.8	46.5
F202L (5.51)/Y213I (5.62)	92.0	88.9	95.5	98.7	67.9	88.6
V152A (4.42)/L171P (4.61)	89.4	70.6	97.4	44.3	60.9	72.5
I105K (3.23)/I195F (5.44)	3.6	98.8	90.9	11.9	38.3	48.7
V191L (5.40)/S409N (7.36)	97.5	90.9	43.7	27.1	96.6	71.2
N124H (3.42)/T205M (5.54)	56.9	62.9	61.5	45.2	83.4	62.0
N124H (3.42)/L379F (6.41)	70.2	65.8	75.2	71.1	7.7	58.0
T205M (5.54)/L379F (6.41)	81.4	98.0	96.8	76.4	88.3	88.2
M117F (3.35)/Y199F (5.48)	79.5	83.6	98.5	5.3	4.4	54.3
I48T (1.46)/F110W (3.28)	92.9	95.1	85.2	15.5	44.6	66.7
V83L (2.53)/V91S (2.61)	90.4	97.3	98.3	94.9	91.3	94.4
S193G (5.42)/C385M (6.47)	77.9	93.3	91.2	62.7	46.4	74.3

The ranking scores (normalized scores) for each covarying ET residue pair were derived from the sequences sharing above 50%, 42%, or 35% identity with D2R, 402 bioamine receptor sequences, or 2,500 Class A GPCR sequences.

^*Average of all ranking scores predicted for each covarying ET residue pair. Scores range from 0 to 100, with higher scores indicating greater coupling potential.

Table S4.

Ranking scores of each covarying ET residue pair computed by DCA–FN (37) based on different input sequence sets

Mutation (Ballesteros–Weinstein number)	Above 50% sequence identity with D2R	Above 42% sequence identity with D2R	Above 35% sequence identity with D2R	Bioamine sequence set	Class A sequence set	Average evolutionary coupling potential*
M117F (3.35)/L387C (6.49)	56.0	28.3	22.8	75.4	11.8	38.9
F202L (5.51)/Y213I (5.62)	95.4	91.2	95.4	99.8	65.1	89.4
V152A (4.42)/L171P (4.61)	88.8	75.5	96.5	41.0	64.2	73.2
I105K (3.23)/I195F (5.44)	22.2	97.0	82.5	5.8	23.6	46.2
V191L (5.40)/S409N (7.36)	94.7	74.9	26.0	32.9	96.8	65.1
N124H (3.42)/T205M (5.54)	52.5	70.2	71.3	38.5	84.2	63.3
N124H (3.42)/L379F (6.41)	67.0	72.2	69.8	64.2	21.1	58.9
T205M (5.54)/L379F (6.41)	82.3	97.9	94.7	93.9	89.3	91.6
M117F (3.35)/Y199F (5.48)	76.0	81.7	99.0	3.9	0.7	52.2
I48T (1.46)/F110W (3.28)	93.6	95.3	79.1	23.7	38.4	66.0
V83L (2.53)/V91S (2.61)	93.3	97.6	97.2	95.6	90.8	94.9
S193G (5.42)/C385M (6.47)	75.9	91.8	89.9	65.3	44.6	73.5

The ranking scores (normalized scores) for each covarying ET residue pair were derived from the sequences sharing above 50%, 42%, or 35% identity with D2R, 402 bioamine receptor sequences, or 2,500 Class A GPCR sequences.

^*Average of all ranking scores predicted for each covarying ET residue pair. Scores range from 0 to 100 with higher scores indicating greater coupling potential.

Table S5.

Correlation between the extent of functional coupling at dopamine efficacy and coupling potential predicted by ET-MIp, DCA–DI (37), and DCA–FN (37)

Epistasis model	Input sequence set	ET-MIp		DCA–DI		DCA–FN
		R	P value	R	P value	R	P value
Additive interaction	Class A	0.109	0.736	−0.152	0.637	−0.188	0.558
	Bioamine	0.215	0.501	−0.488	0.107	−0.465	0.128
	>35% SI w/D2R	0.478	0.116	0.350	0.264	0.377	0.227
	>42% SI w/D2R	0.279	0.380	0.115	0.723	0.172	0.593
	>50% SI w/D2R	0.471	0.123	0.398	0.199	0.396	0.203
Product interaction	Class A	0.230	0.472	−0.086	0.789	−0.132	0.682
	Bioamine	0.231	0.471	−0.478	0.116	−0.413	0.182
	>35% SI w/D2R	0.428	0.165	0.296	0.350	0.290	0.361
	>42% SI w/D2R	0.336	0.285	0.200	0.533	0.247	0.440
	>50% SI w/D2R	0.532	0.075	0.363	0.246	0.392	0.207
Log interaction	Class A	0.101	0.756	−0.095	0.770	−0.131	0.684
	Bioamine	0.246	0.440	−0.476	0.118	−0.426	0.167
	>35% SI w/D2R	0.512	0.089	0.384	0.217	0.400	0.197
	>42% SI w/D2R	0.332	0.291	0.170	0.598	0.228	0.476
	>50% SI w/D2R	0.543	0.068	0.406	0.190	0.415	0.180
Minimal interaction	Class A	0.411	0.185	0.144	0.654	0.059	0.855
	Bioamine	−0.051	0.874	0.010	0.975	0.131	0.684
	>35% SI w/D2R	0.318	0.314	0.502	0.096	0.430	0.162
	>42% SI w/D2R	0.510	0.091	0.575	0.051	0.585	0.046
	>50% SI w/D2R	0.612	0.034	0.115	0.722	0.245	0.444

The extent of functional coupling is estimated by epistasis models as described in SI Materials and Methods. Coupling potential was derived from the sequences sharing above 50%, 42%, or 35% identity with D2R (SI w/D2R), 402 bioamine receptor sequences (Bioamine), or 2,500 Class A GPCR sequences (Class A). R, Pearson's R; SI, sequence identity.

Fig. S2.

Relationships between the evolutionary coupling potential and epistasis scores calculated using the product interaction model. Epistasis scores calculated using the product interaction model were computed from the equation ε^pro = M_ab − M_a∙M_b, where M_a and M_b are the functional scores of the single mutants and M_ab is the functional score of the double mutant. The functional score of each mutant was derived from the dopamine efficacy data, which were normalized to WT as shown in Fig. 2C. For each covarying ET residue pair, the epistasis score is plotted on the y axis, and the evolutionary coupling potential predicted based on the sequence sets of Class A GPCRs (A), bioamine family (B), and above 35% (C), 42% (D), and 50% (E) sequence identity with D2R is plotted on the x axis. (F) The average of all evolutionary coupling potentials predicted for each covarying ET residue pair is plotted against epistasis scores.

Fig. S3.

Relationships between the evolutionary coupling potential and epistasis scores calculated using the logarithmic interaction model. Epistasis scores calculated using the logarithmic interaction model were computed from the equation ε^log = M_ab − log₂[(2^Ma − 1)·(2^Mb − 1) + 1], where M_a and M_b are the functional scores of the single mutants and M_ab is the functional score of the double mutant. The functional score of each mutant was derived from the dopamine efficacy data, which were normalized to WT as shown in Fig. 2C. For each covarying ET residue pair, the epistasis score is plotted on the y axis, and the evolutionary coupling potential predicted based on sequence sets of Class A GPCRs (A), bioamine family (B), and above 35% (C), 42% (D), and 50% (E) sequence identity with D2R is plotted on the x axis. (F) The average of all evolutionary coupling potentials predicted for each covarying ET residue pair was plotted against epistasis scores.

To assess the effect of functional coupling on dopamine potency, dose–response curves were generated to derive EC₅₀ values for WT and mutant D2Rs (Fig. 2 D and E and Table S6). We used relative ln(EC₅₀) values, defined as the difference between ln(EC_{50 mutant}) and ln(EC_{50 WT}), to approximate free energy changes, because potency reflects differences in ligand binding affinity and/or activation kinetics, either of which must be determined by a free energy term. To assess nonadditivity, we compared the sum of relative ln(EC₅₀) values of the single mutants with the relative ln(EC₅₀) value of the corresponding double mutants (Fig. 2F), except when mutants showed almost no response to dopamine. Unlike the three residue pairs with the lowest evolutionary coupling potential, the six pairs with the highest evolutionary coupling potential were nonadditive—a functional coupling that presumably fine-tunes the sensitivity to dopamine (Fig. 2F). Here again, deviation from additivity correlated most highly with the evolutionary coupling potential derived from the dopamine receptor sequence set (Fig. 2G and Fig. S5), and, overall, ET-MIp correlated better with nonadditivity than DCA–FN and DCA–DI (Table S7).

Table S6.

Potency of dopamine or serotonin in G_i activation for D2R mutants

D2R mutants	Dopamine		Serotonin
	pEC50 ± SEM	(EC50, nM)	pEC50 ± SEM	(EC50, µM)
WT D2R	8.01 ± 0.05	(9.7)	4.57 ± 0.09	(26.8)
I48T	8.74 ± 0.06	(1.8)	5.46 ± 0.02	(3.5)
F110W	7.21 ± 0.08	(61.4)	N.D.
I48T/F110W	8.06 ± 0.06	(8.8)	4.93 ± 0.07	(11.7)
V83L	7.68 ± 0.17	(20.8)	4.29 ± 0.20	(51.2)
V91S	7.24 ± 0.16	(57.0)	N.D.
V83L/V91S	7.67 ± 0.14	(21.4)	N.D.
I105K	7.99 ± 0.09	(10.2)	4.77 ± 0.08	(17.0)
I195F	7.55 ± 0.04	(28.4)	N.D.
I105K/I195F	7.39 ± 0.13	(40.9)	N.D.
M117F	7.97 ± 0.09	(10.8)	4.66 ± 0.10	(21.8)
Y199F	6.78 ± 0.23	(167.5)	N.D.
M117F/Y199F	7.30 ± 0.14	(50.5)	N.D.
L387C	8.14 ± 0.07	(7.2)	4.99 ± 0.19	(10.3)
M117F/L387C	8.14 ± 0.13	(7.3)	5.13 ± 0.17	(7.5)
N124H	6.26 ± 0.12	(554.6)	N.D.
T205M	8.60 ± 0.14	(2.5)	5.60 ± 0.17	(2.5)
L379F	N.D.		N.D.
N124H/T205M	7.58 ± 0.17	(26.4)	4.89 ± 0.08	(13.0)
N124H/L379F	N.D.		N.D.
T205M/L379F	7.83 ± 0.14	(14.8)	N.D.
V152A	6.83 ± 0.11	(148.3)	N.D.
L171P	7.23 ± 0.12	(59.2)	N.D.
V152A/L171P	6.23 ± 0.12	(594.3)	N.D.
S193G	N.D.		N.D.
C385M	6.10 ± 0.12	(799.8)	N.D.
S193G/C385M	N.D.		N.D.
V191L	7.57 ± 0.19	(26.7)	4.83 ± 0.11	(15.0)
S409N	7.78 ± 0.13	(16.6)	4.77 ± 0.16	(17.2)
V191L/S409N	7.39 ± 0.10	(40.9)	4.52 ± 0.17	(30.2)
F202L	6.53 ± 0.08	(293.8)	N.D.
Y213I	8.48 ± 0.13	(3.3)	5.86 ± 0.12	(1.4)
F202L/Y213I	7.23 ± 0.16	(58.3)	N.D.

Membrane potential assays for detecting G_i activation and nonlinear regression curve-fitting analysis were done as described in SI Materials and Methods. Results represent mean ± SEM of three to eight experiments. Some D2R mutants showed almost no response to dopamine, precluding determination of accurate EC₅₀ values. N.D., not determined.

Fig. S5.

Relationships between the degree of nonadditivity and predicted evolutionary coupling potential. Deviation from additivity = $\left|\text{ln}\left({EC}_{\mathrm{50}mutantA}/{EC}_{50WT}\right)+\text{ln}\left({EC}_{50mutantB}/{EC}_{50WT}\right)-\text{ln}\left({EC}_{50mutantAB}/{EC}_{50WT}\right)\right|$ with the dopamine EC₅₀ data in Fig. 2F. For each covarying ET residue pair, deviation from additivity is plotted on the y axis, and the evolutionary coupling potential predicted based on the sequence sets of Class A GPCRs (A), bioamine family (B), and above 35% (C), 42% (D), and 50% (E) sequence identity with D2R is plotted on the x axis. The degree of nonadditivity correlated most highly with the evolutionary coupling potential derived from the sequences sharing above 35% identity with D2R. (F) The average of all evolutionary coupling potentials predicted for each covarying ET residue pair is plotted against deviation from additivity.

Table S7.

Correlation between the extent of functional coupling at dopamine potency and coupling potential predicted by ET-MIp, DCA–DI (37), and DCA–FN (37)

Input sequence set	ET-MIp		DCA–DI		DCA–FN
	R	P value	R	P value	R	P value
Class A	−0.302	0.430	−0.004	0.992	0.019	0.961
Bioamine	0.607	0.083	0.058	0.883	0.038	0.923
>35% SI w/D2R	0.762	0.017	0.374	0.321	0.524	0.147
>42% SI w/D2R	0.521	0.151	0.130	0.740	0.281	0.464
>50% SI w/D2R	0.348	0.358	0.336	0.377	0.285	0.457

^aThe extent of functional coupling is the deviation from additivity as described in Fig. 2G. Coupling potential was derived from the sequences sharing above 50%, 42%, or 35% identity with D2R (SI w/D2R), 402 bioamine receptor sequences (Bioamine), or 2,500 Class A GPCR sequences (Class A). R, Pearson's R; SI, sequence identity.

Rescue Effect on Serotonin Activation Was Observed for Some D2R Double Mutants.

Some of the studied ET residues were involved in discriminating against G₁₆ activation induced by serotonin stimulation of D2R (23). To test whether the covarying ET residue pairs are functionally coupled with respect to regulating agonist specificity, we compared the effects of single and double mutations on serotonin responses. The single mutants I48T^1.46, Y213I^5.62, T205M^5.54, L387C^6.49, and I105K^3.23 had enhanced responses to serotonin, indicating that these positions in D2R participate in discriminating against G_i activation induced by serotonin (Fig. 3C). A rescue effect was observed at I48T^1.46/F110W^3.28 and N124H^3.42/T205M^5.54, even though F110W^3.28 and N124H^3.42 alone abolished activation by serotonin, suggesting functional coupling in controlling the specificity for D2R activation in these cases (Fig. 3 A−C). Given the small number of functional coupling cases found, we infer that discriminating against serotonin is not the main functional role of the covarying ET residue pairs studied here. Moreover, this finding suggests that the allosteric communication responsible for specificity of receptor activation may vary across a protein family, making it difficult for ET-MIp to extract such a pattern from a broad sequence input.

Fig. 3.

A rescue effect on serotonin responses was observed in some of the covarying ET residue pairs. (A and B) Membrane potential changes induced by G_i activation in response to serotonin stimulation of D2Rs were measured and analyzed as in Fig. 2 A and B. (C) Maximal activation by serotonin of mutant D2Rs was normalized to that of WT, which was defined as 1. Bars, statistics, and color-coding are as described for Fig. 2C.

Nonadditivity of Free Energy Changes upon Ligand Binding Was Observed at Some D2R Double Mutants.

To investigate whether covarying ET residue pairs interact to control ligand affinity, spiperone competition binding experiments determined K_i for both dopamine and serotonin and measured the energetic perturbations by calculating the Gibbs free energy change (∆∆G⁰) (Table S8). In contrast to EC₅₀, these measurements on whole cell membranes reflect low-affinity non-G protein-coupled binding. Significant differences between (∆∆G⁰_A + ∆∆G⁰_B) and ∆∆G⁰_AB observed for F202L^5.51/Y213I^5.62 and T205M^5.54/L379F^6.41 revealed nonadditive effects on both dopamine (Fig. 4 A and C) and serotonin (Fig. 4 B and D) binding, indicating functional coupling in regulating receptor−ligand binding affinity. Strikingly, F202L^5.51/Y213I^5.62 and T205M^5.54/L379F^6.41 are outside the ligand binding pocket, and yet alter receptor−ligand interaction (Fig. 2H), indicating their involvement in allosteric pathways linking ligand binding and receptor conformational change. Further, the free energy change demonstrated that ET-MIp was able to detect energetic coupling for ligand binding at both proximal (3.6 Å) and long-distance (13.2 Å) residue pairs, even when located far away from the ligand binding site. Because the double mutations F202L^5.51/Y213I^5.62 and T205M^5.54/L379F^6.41 rendered both dopamine and serotonin binding energetically more favorable (Fig. 4 C and D), we did not find evidence for functional coupling responsible for the specificity of ligand binding.

Table S8.

Binding affinity for dopamine and serotonin at D2R mutants

D2R mutants	Dopamine		Serotonin
	pKi (Ki, μM)	∆∆G0, kJ/mol	pKi (Ki, μM)	∆∆G0, kJ/mol
WT D2R	4.51 ± 0.03 (30.9)		3.72 ± 0.04 (191.9)
I48T	4.49 ± 0.04 (32.6)	−0.33 ± 0.29	3.49 ± 0.06 (322.1)	1.06 ± 0.10
F110W	4.62 ± 0.06 (24.2)	−1.08 ± 0.31	3.78 ± 0.12 (167.9)	−0.56 ± 0.66
I48T/F110W	4.48 ± 0.06 (33.1)	−0.30 ± 0.47	3.47 ± 0.12 (335.7)	1.16 ± 0.62
I105K	4.54 ± 0.10 (28.6)	0.77 ± 0.56	3.25 ± 0.03 (558.5)	1.38 ± 0.35
I195F	4.58 ± 0.04 (26.2)	0.55 ± 0.39	3.20 ± 0.08 (633.9)	1.69 ± 0.78
I105K/I195F	4.57 ± 0.04 (27.2)	0.65 ± 0.38	3.41 ± 0.06 (390.8)	0.50 ± 0.66
M117F	4.26 ± 0.15 (55.0)	0.86 ± 0.54	3.53 ± 0.05 (295.1)	1.56 ± 0.38
Y199F	4.26 ± 0.07 (55.5)	0.89 ± 0.17	3.81 ± 0.05 (153.5)	−0.07 ± 0.38
M117F/Y199F	4.18 ± 0.03 (66.2)	1.32 ± 0.26	3.26 ± 0.04 (553.4)	3.12 ± 0.30
M117F	4.42 ± 0.08 (38.2)	0.93 ± 0.28	3.16 ± 0.06 (690.2)	1.97 ± 0.37
L387C	4.71 ± 0.01 (19.5)	−0.73 ± 0.28	3.34 ± 0.09 (460.3)	0.97 ± 0.31
M117F/L387C	4.39 ± 0.03 (41.0)	1.11 ± 0.34	3.15 ± 0.08 (706.3)	2.03 ± 0.46
N124H	4.57 ± 0.15 (26.7)	−0.32 ± 0.62	3.73 ± 0.08 (186.2)	−0.27 ± 0.23
T205M	5.47 ± 0.07 (3.4)	−5.45 ± 0.21	3.81 ± 0.03 (156.0)	−0.71 ± 0.38
N124H/T205M	5.44 ± 0.10 (3.6)	−5.29 ± 0.38	3.91 ± 0.03 (122.2)	−1.32 ± 0.55
N124H	4.68 ± 0.03 (20.9)	−0.76 ± 0.36	3.68 ± 0.05 (207.0)	−0.69 ± 0.27
L379F	4.37 ± 0.01 (43.1)	1.04 ± 0.15	3.42 ± 0.11 (383.7)	0.84 ± 0.52
N124H/L379F	4.28 ± 0.09 (52.2)	1.52 ± 0.68	3.49 ± 0.06 (320.6)	0.39 ± 0.51
T205M	5.39 ± 0.04 (4.1)	−5.15 ± 0.52	3.80 ± 0.05 (157.4)	−0.48 ± 0.09
L379F	4.12 ± 0.03 (75.7)	2.10 ± 0.51	3.50 ± 0.04 (317.0)	1.25 ± 0.38
T205M/L379F	5.64 ± 0.08 (2.3)	−6.59 ± 0.57	4.13 ± 0.05 (74.8)	−2.33 ± 0.36
V152A	4.13 ± 0.09 (74.1)	1.31 ± 0.62	3.65 ± 0.12 (224.9)	1.80 ± 0.87
L171P	3.91 ± 0.09 (122.5)	2.56 ± 0.50	3.30 ± 0.06 (505.8)	3.81 ± 1.03
V152A/L171P	3.51 ± 0.05 (310.5)	4.86 ± 0.26	3.00 ± 0.17 (997.7)	5.49 ± 0.35
V191L	4.38 ± 0.02 (41.6)	1.09 ± 0.36	3.51 ± 0.23 (306.9)	0.99 ± 0.21
S409N	4.71 ± 0.05 (19.5)	−0.78 ± 0.65	3.66 ± 0.27 (218.3)	0.14 ± 0.49
V191L/S409N	4.67 ± 0.06 (21.2)	−0.58 ± 0.42	3.75 ± 0.33 (176.6)	−0.39 ± 0.45
F202L	4.46 ± 0.10 (35.1)	1.43 ± 0.74	3.52 ± 0.05 (303.4)	3.23 ± 0.89
Y213I	4.66 ± 0.08 (21.7)	0.24 ± 0.34	3.33 ± 0.04 (465.6)	4.29 ± 1.22
F202L/Y213I	4.81 ± 0.06 (15.6)	−0.58 ± 0.42	3.87 ± 0.15 (134.0)	1.21 ± 0.39

K_i for dopamine or serotonin was determined by [³H]spiperone competition binding experiments. Change in free energy upon dopamine or serotonin binding due to the mutations was calculated as follows: ∆∆G⁰_mutant = ∆G⁰_mutant − ∆G⁰_wild-type = RT ln(K_{i mutant}/K_{i wild-type}), where R = 8.314 J K⁻¹⋅mol⁻¹ and T = 298.15 K. Data represent mean ± SEM of three to six experiments. No [³H]spiperone binding to D2R mutants V91S and S193G/C385M was detected, so the residue pairs V83L/V91S and S193G/C385M were excluded from the measurement of ∆∆G⁰.

Fig. 4.

Nonadditive effects on Gibbs free energy change upon ligand binding. (A and B) The [³H]spiperone binding to WT or mutant D2Rs was detected in the presence of various concentrations of competing dopamine or serotonin. Examples of competition binding curves for dopamine and serotonin are shown in A and B, respectively. Nonlinear regression analysis was performed with the one-site model to determine the IC₅₀, which was then used in the K_i. (C and D) Values for free energy change (∆G⁰) upon dopamine or serotonin binding were derived from ∆G⁰ = RT lnK_i, where R = 8.314 J K⁻¹⋅mol⁻¹ and T = 298.15 K. The mutational effects on free energy change upon binding to dopamine (C) or serotonin (D) were examined by calculating ∆∆G⁰ of each D2R mutant as ∆∆G⁰_mutant = ∆G⁰_mutant − ∆G⁰_WT. Results of the log-additive analysis are expressed as mean ± SEM (n = 3–6; **P < 0.001; *P < 0.05; independent two-tailed Student’s t tests). Color-coding of the bars is as described for Fig. 2C.

G Protein-Specific Effects.

Our new assay for G_i activation let us compare mutational effects on D2R activation of G_i vs. G₁₆ (23), and revealed several striking contrasts. For serotonin activation, N124H^3.42 greatly decreased maximal activation of G_i (Fig. 3C) but significantly increased maximal activation of G₁₆. For maximal dopamine activation, T205M^5.54, F110W^3.28, I48T^1.46, and V91S^2.61 all displayed activation of G₁₆ close to that of WT (23), but activated the physiological partner of D2R, G_i, at levels ranging from 2.5-fold to nearly fourfold higher than WT (Fig. 2C), implying somewhat different coupling mechanisms for these two G proteins.

Discussion

ET-MIp predicts functionally coupled residue pairs by weighing mutual variations by the phylogenetic depth of the associated evolutionary divergences. Experiments in D2R suggest the method can reveal allosteric communication between distant and evolutionarily important residue positions.

The most striking observations are that covarying ET residue pairs work together to modulate efficacy and potency of dopamine in D2R, and that the main function encoded by the allosteric communication is to maintain the dopamine response at the WT level and fine-tune sensitivity to dopamine. ET-MIp identified functional couplings that regulate receptor−ligand interactions through residue pairs outside the ligand binding pocket, underscoring the allosteric nature of the pathways extending from the ligand binding site. Covarying ET residue pairs control dopamine responsiveness tightly, possibly reflecting an evolutionary need to maintain distinct dopamine affinities among dopamine receptors (D1, D2, D3, D4, and D5) (34–36), each activating diverse downstream signaling pathways. Coupling scores calculated at varying evolutionary depth indicate that coevolution coincides with functional separation of the subfamilies fairly late in evolution. The results of cosubstitution suggest that deviation in either direction from this highly tuned response conferred selective evolutionary disadvantage.

There have been few studies of the functional relationships between covarying residues (14, 16–18, 41). Our results reveal functional coupling of the covarying ET residues in D2R, and the synergistic or antagonistic features of the coupling. For example, F202L^5.51 and Y213I^5.62 have a synergistic effect on dopamine efficacy, whereas the effects on dopamine potency are antagonistic (Fig. 2 C and F), suggesting that different interactions govern efficacy and potency. The observation that covarying ET residue pairs mediate ligand-specific functional interactions without contacting the ligand directly supports the previous proposal of a conformational filter in D2R (23).

Previous studies have proposed models of molecular switches associated with receptor activation, such as the ionic lock (D/E)RY (42–47), transmission switch (48), and tyrosine toggle switch (NPxxY) (44–47, 49). The discovery of functionally coupled covarying ET residue pairs provides insight into the allosteric pathways connecting ligand binding, molecular switches, and G protein coupling in D2R. Based on the positions of the covarying ET residue pairs mapped onto the structure of D3R (50), we can classify the coupling mechanisms into three categories. (i) The covarying ET residues are far from each other (Table S1), with one at or near the ligand binding site and the other close to the molecular switches, e.g., I48T^1.46/F110W^3.28. F110^3.28 is at the orthosteric binding pocket (50, 51), and I48^1.46 is in direct contact with D80^2.50, which interacts with the transmission switch and NPxxY motif through water molecules in molecular dynamics simulations of β₁ and β₂ adrenergic receptors (52), suggesting a role for this pair in coupling ligand binding to switch residues. (ii) Both of the covarying ET residues are at or near the ligand binding site, with one close to the 3–7 lock switch (D114^3.32–Y416^7.43), which forms a link between TM3 and TM7 that breaks upon receptor activation in rhodopsin (47, 50, 53). These include V83L^2.53/V91S^2.61 and M117F^3.35/Y199F^5.48. A side-chain rotamer of V83L^2.53 can interact with Y416^7.43 (Fig. S6A) and an inward-facing side-chain rotamer of M117F^3.35 could interact with D114^3.32, if TM2 and TM3 move apart from each other (Fig. S6B). Their coupling may contribute to propagation of the dopamine binding signal to the 3–7 lock. (iii) Both of the covarying ET residues are far from the ligand binding pocket but close to the molecular switches or G protein-binding region. These include F202L^5.51/Y213I^5.62, N124H^3.42/T205M^5.54, and T205M^5.54/L379F^6.41. F202^5.51 is part of the transmission switch, and the side chain of the mutant F202L^5.51 can contact the side chain of I122^3.40, which is part of the transmission switch (48) (Fig. S6C). Y213^5.62 interacts with V215^5.64, which was found to participate in receptor−G protein interactions (44, 46) (Fig. S6C). N124H^3.42, T205M^5.54, and L379F^6.41 may interact with the residues making up the hydrophobic barrier in TM2, TM3, and TM6 (47, 54). In addition, T205^5.54 and L379^6.41 are in direct contact with P201^5.50 and F382^6.44, respectively, of the transmission switch(48) (Fig. S6D). A backbone hydrogen bond links L379F^6.41 and the G protein interacting residue L375^6.37 (V331^6.37 in D3R) (46) (Fig. S6E). Thus, T205M^5.54 and L379F^6.41 form the lynchpin in a continuous network of contacts leading from the transmission switch to the G protein. The observation that T205M^5.54/L379F^6.41 are coupled in ligand binding is consistent with bidirectional communication between ligand binding and G protein activation. The idea that coupled residues are links in a chain of allosteric interactions is supported by the observation that intervening residues contacting them tend to have high coupling scores (Fig. S7 and Table S9). However, these residues are identical in D2R and 5-HT_2AR and could not be tested in our paradigm.

Fig. S6.

The functionally coupled covarying ET residue pairs are proximal to the molecular switches or potential G protein-binding region. The functionally coupled covarying ET residue pairs in D2R were mapped onto the corresponding positions in the structure of D3R with the bound ligand eticlopride (shown in light gray) (PDB 3PBL). (A) V83L^2.53 (orange) forms the hydrophobic contact with Y416^7.43 (blue) of the 3–7 lock switch. (B) One of the side-chain rotamers of M117F^3.35 (orange) interacts with D114^3.32 (blue) of the 3–7 lock switch via the hydrogen bonds and hydrophobic contacts, while another rotamer points outward (green). For both cases, TM2 and TM3 need to move apart from each other to accommodate the mutation. (C) The cytoplasmic view shows that F202L^5.51 (orange) interacts with I122^3.40 (blue) of the transmission switch via the hydrophobic contact. The backbone of Y213^5.62 (green) forms hydrogen bonds with the backbone of V215^5.64 (green), which is involved in the receptor−G protein interactions in the β₂ adrenergic receptor and rhodopsin. (D) T205^5.54 and L379^6.41 interact with P201^5.50 and F382^6.44 of the transmission switch, respectively, via hydrogen bonds. (E) After mutation, the interaction between the side chains of T205M^5.54 (orange) and L379F^6.41 (orange) is potentially enhanced by forming a hydrophobic stacking interaction. The side chain of L379F^6.41 forms hydrogen bonds with the backbone of L375^6.37 (V331^6.37 in D3R, blue), which is a G protein interacting residue in the β₂ adrenergic receptor.

Fig. S7.

Residues at the intermediate positions of networks connecting the functionally coupled residue pairs. The covarying ET residue pairs in D2R were mapped onto the corresponding positions in the structure of D3R with the bound ligand eticlopride (shown in light gray) (PDB 3PBL). Residues that were predicted by ET-MIp to couple to both of the constituent residues of the functionally coupled pairs, F202^5.51/Y213^5.62, N124^3.42/T205^5.54, T205^5.54/L379^6.41, I48^1.46/F110^3.28, and V83^2.53/V91^2.61, are shown in red, orange, green, yellow, and blue, respectively. Except for W90^2.60, L206^5.55, L207^5.56, and V208^5.57, the intermediate residues are identical between D2R and 5-HT_2AR; therefore, these were not tested experimentally using the substitution mutagenesis. The evolutionary coupling potential is indicated in Table S9.

Table S9.

Evolutionary coupling potential of covarying ET residue pairs at intermediate positions of pathways connecting the allosterically coupled residues

Residue pair (Ballesteros–Weinstein number)	Evolutionary coupling potential
I48 (1.46)/V87 (2.57)	94.1
V87 (2.57)/F110 (3.28)	76.1
V83 (2.53)/L86 (2.56)	94.7
L86 (2.56)/V91 (2.61)	87.4
V83 (2.53)/V87 (2.57)	97.1
V87 (2.57)/V91 (2.61)	95.2
V83 (2.53)/M88 (2.58)	99.2
M88 (2.58)/V91 (2.61)	95.2
V83 (2.53)/P89 (2.59)	92.2
P89 (2.59)/V91 (2.61)	93.8
V83 (2.53)/W90 (2.60)	98.9
W90 (2.60)/V91 (2.61)	94.4
N124 (3.42)/L125 (3.43)	90.6
L125 (3.43)/T205 (5.54)	71.7
T205 (5.54)/V378 (6.40)	76.8
V378 (6.40)/L379 (6.41)	82.2
F202 (5.51)/L206 (5.55)	75.0
L206 (5.55)/Y213 (5.62)	87.5
F202 (5.51)/L207 (5.56)	97.8
L207 (5.56)/Y213 (5.62)	96.8
F202 (5.51)/V208 (5.57)	96.5
V208 (5.57)/Y213 (5.62)	87.0
F202 (5.51)/Y209 (5.58)	75.1
Y209 (5.58)/Y213 (5.62)	94.9

Evolutionary coupling potential (normalized score) for each covarying ET residue pair was predicted by ET-MIp using the input sequences sharing above 35% identity with D2R. Scores range from 0 to 100, with higher scores indicating greater evolutionary coupling potential.

The side chain of M117F^3.35 points toward the putative cholesterol binding site (55, 56) and may enhance cholesterol interactions; Y199F^5.48 points toward the putative receptor−dimerization interface (57, 58). V152^4.42 contacts cholesterol-contacting position I156^4.46 (56), and L171^4.61 points toward the dimerization interface. The side chains of both F202L^5.51 and Y213I^5.62 may lie at the same interface. The functional coupling observed may reflect effects on modulation by cholesterol (59) and GPCR dimerization (57, 60). The equivalent residue to Y213^5.62 in activated β₂ adrenergic receptor (46) stabilizes the “outward” movement of TM6 through a van der Waals contact, consistent with effects of Y213I^5.62 on G protein coupling.

Overall, the results support the idea that covariation patterns are signatures preserved in protein sequences during evolution and reflect functional interactions important for fitness-conferring properties. They open the possibility of improving structure-based drug design by accounting for intramolecular allosteric communication. The involvement of covarying ET residue pairs in allosteric pathways linking ligand binding, molecular switches, and G protein coupling offers the potential to reengineer allosteric pathways of receptor activation. The divergent effects on G_i and G₁₆ (23) activation observed in some mutants suggest that covarying ET residue pairs play a role in governing receptor preference for downstream effectors. Thus, this work also serves as a starting point for studies on bias in activation of effectors controlled by covarying ET residue pairs, and on the interpretation of genome variations.

Materials and Methods

The key methods are briefly described here. For full details, please see Supporting Information.

Evolutionary Trace and MI Analysis.

To identify evolutionarily coupled residue pairs, we integrated MI into the ET framework as follows:

$$ET\left(MIp\left(i,j\right)\right)={\displaystyle \sum }_{n=1}^N\frac{1}{n}{\displaystyle \sum }_{g=1}^{n}MI{p}^g(i,j).$$

The alignment is broken up into subalignments g according to the phylogenetic tree with N nodes. Our measure of mutual information, MIp (5), is computed for all possible residue pairs i and j for each subalignment selected by the phylogenetic tree. The performance of ET-MIp in contact prediction was compared with other methods as described in Supporting Information, using an alignment of 2,500 Class A GPCRs to predict interresidue contacts in the structure of the β₂ adrenergic receptor (PDB 2RH1). To identify covarying residues in D2R, BLAST (Basic Local Alignment Search Tool) analysis of D2R was first performed against the Uniref90 sequence database (61). To identify homologs, protein sequences were filtered by protein length and sequence identity (>35%, >42%, >50%), where each alignment was respectively made up of all dopamine receptors, dopamine D2 and D3 receptors, and only D2 receptors. The bioamine and Class A GPCR alignments were described previously (23).

Membrane Potential Assay.

G_i activation induced by agonist-stimulated D2R triggers the opening of TRPC4β channels in HEK293 cells, leading to membrane potential changes (62). Details are given in Supporting Information.

SI Materials and Methods

ET and MI.

To identify evolutionarily coupled residue pairs that are compensatory or synergistic for transferring function when comutated, we integrated MI into the ET framework as shown,

$$ET\left(MIp\left(i,j\right)\right)={\displaystyle \sum }_{n=1}^N\frac{1}{n}{\displaystyle \sum }_{g=1}^{n}MI{p}^g(i,j).$$

In this formula, the total alignment is broken up into subalignments g according to the phylogenetic tree with N nodes. Our chosen measure of mutual information MIp (described below) is computed for all possible residue pairs i and j for each subalignment selected by the phylogenetic tree.

MI is a standard measure for the shared information between two discrete sets of events. In this study, the sets are defined by the amino acids found at positions i and j within the protein sequence alignment. MI is defined as

$$\text{MI}(i,j)=E\left(i\right)+E\left(j\right)-E(i,j),$$

where E(i) and E(j) are the marginal entropies and are written

$$E\left(i\right)=-{\displaystyle \sum }_x^{N}p\left(x\right)\hspace{0.17em}log\hspace{0.17em}p\left(x\right),$$

where x represents the possible events (i.e., amino acid types) and p(x) represents the frequency of these events. The joint entropy E(i, j) is written

$$E(i,j)=-{\displaystyle \sum }_x^N{\displaystyle \sum }_y^{N}p(x,y)\hspace{0.17em}log\hspace{0.17em}p(x,y),$$

where p(x, y) represents the frequency that events x and y occur together. MI can be normalized by dividing by the total entropy of the individual events

$$\text{nMI}(i,j)=\frac{\text{MI}(i,j)}{E\left(i\right)+E\left(j\right)}.$$

In this study, we used the alternative normalization measure introduced by Dunn et al. (5), which approximates the background MI shared by positions i and j. This is called the average product correction (APC) and is computed

$$\text{APC}(i,j)=\frac{\text{MI}(i,\overline{\text{x}})\hspace{0.17em}\cdot \hspace{0.17em}\text{MI}(j,\overline{\text{x}})}{\overline{\text{MI}}},$$

where the mean mutual information of column i or j is written

$$\text{MI}(i,\overline{\text{x}})=\frac{1}{m}{\displaystyle \sum }_x^m\text{MI}(i,x),$$

and the overall mean mutual information is defined as

$$\overline{\text{MI}}=\frac{2}{nm}{\displaystyle \sum }_i^m{\displaystyle \sum }_j^n\text{MI}(i,j),$$

where m and n represent the number of columns in the alignment. Then normalized mutual information MIp is the difference between MI and the APC

$$\text{MIp}(i,j)=\text{MI}(i,j)-\text{APC}(i,j),$$

where MIp was the measure integrated into the ET framework to identify evolutionarily coupled residue pairs. Note that the raw score was normalized to a percentile as shown in Table S2 for easy comparison.

Multiple Sequence Alignments.

To identify covarying residues in D2R, the D2R sequence was first blasted against the Uniref90 sequence database (61). To identify homologs, the protein sequences were filtered by protein length and sequence identity (>35%, >42%, >50%), where each alignment was respectively made up of all dopamine receptors, dopamine D2 and D3 receptors, and only D2Rs. The bioamine and Class A GPCR alignments were described previously (23).

Structure Contact Analysis.

A common testing metric for sequence mutual information methods is to test their ability to predict structural contacts. To test whether adding phylogeny improved predictions, we applied ET-MIp to the alignment made up of ∼2,500 Class A GPCR TM domain sequences and tested our ability to predict β₂ adrenergic receptor (PDB 2RH1) structure contacts. In this experiment, we defined structural contacts (positive set) as amino acids within 6 Å (ignoring hydrogen atoms). Fig. 1B shows the improved receiver operating characteristic curves with the new algorithm, ET-MIp, compared with MIp (5), DCA algorithms, FN, and DI (37), and a standard measure for normalized mutual information (nMI) (38). We found that including phylogeny dramatically improved the structure prediction.

Epistasis Models.

To estimate the epistatic effect of double mutations, we used previously described epistasis models, typically referred to as product, logarithmic, minimal, and additive interaction models (39, 40), to calculate epistasis scores. These models are used to study interactions between genes (39) but have also been used to study residue interactions within the WW domain (40). Each model suggests that multiple perturbations should combine in a specific manner, where M_a, M_b, and M_ab are experimental values for the individual mutations a and b and the double mutation a + b, respectively. WT represents the experimental value of the WT D2R. In our case, we normalized the data to WT (WT = 1).

$${\mathit{\epsilon }}_{\mathit{ab}}^{\mathit{product}}={\mathit{M}}_{\mathit{ab}}-{\mathit{M}}_{\mathit{a}}\cdot {\mathit{M}}_{\mathit{b}},$$

$${\mathit{\epsilon }}_{\mathit{ab}}^{\mathit{additive}}=({\mathit{M}}_{\mathit{ab}}+\mathit{WT})-({\mathit{M}}_{\mathit{a}}+{\mathit{M}}_{\mathit{b}}),$$

$${\mathit{\epsilon }}_{\mathit{ab}}^{\mathit{log}}={\mathit{M}}_{\mathit{ab}}-{\log }_2((2^{{\mathit{M}}_{\mathit{a}}}-\mathit{WT})\cdot (2^{{\mathit{M}}_{\mathit{b}}}-\mathit{WT})+\mathit{WT}),$$

$${\mathit{\epsilon }}_{\mathit{ab}}^{\mathit{min}}={\mathit{M}}_{\mathit{ab}}-\mathit{min}({\mathit{M}}_{\mathit{a}},{\mathit{M}}_{\mathit{b}}).$$

To compare prediction by algorithms to phenotype information, each theoretical “coupling” value was compared with epistatic effects estimated by each model to determine how well the data are explained by the model. We then measured Pearson’s R to indicate how well data fit a model.

DCA Comparison.

To compare algorithms, the code for DCA (37) was downloaded and applied to the same five alignments discussed in Materials and Methods. The code can perform two types of analysis, the original DI score and FN score (37). The default parameters were used for both. For comparison, we normalized these scores to a percentile as described for ET-MIp. The normalized scores for each algorithm are shown in Tables S3 and S4. These scores were then compared with phenotypes in the same fashion as ET-MIp (Tables S5 and S7).

Cell Culture.

HEK293 cells and HEK293 cells stably expressing TRPC4β channels, which were a generous gift from Dr. Michael Xi Zhu (The University of Texas Health Science Center at Houston, Houston, TX) were grown in Dulbecco's Modified Eagle’s Medium (Corning) containing 4.5 g/L glucose and l-glutamine supplemented with 10% (vol/vol) FBS (HyClone), 50 unit/mL penicillin, 50 µg/mL streptomycin (Lonza), and 500 mg/L G418 (for TRPC4β-expressing cells only) at 37 °C in a humidified atmosphere with 5% CO₂.

D2R Mutagenesis and Transfection.

To generate D2R mutants, the cDNA clone of hemagglutinin (HA)-tagged human D2R (long isoform) (DRD020TN00; Missouri S&T cDNA Resource Center) was used as a template for site-directed mutagenesis using the QuikChange method. The mutant plasmids were sequenced to confirm the desired mutations and then transfected into TRPC4β-expressing HEK293 cells using Lipofectamine 2000 (Invitrogen) according to the manufacturer’s protocol. For membrane potential assay and cell-ELISA, cells were reverse transfected in 96-well plates coated with 25 µg/mL poly-d-lysine.

Membrane Potential Assay.

Twenty-four hours after transfection of negative control [pcDNA3.1(+)], WT, or mutant D2Rs, the original culture medium was discarded. The transfected TRPC4β-expressing HEK293 cells were then incubated with FLIPR membrane potential dye and quencher combination (FMP2; Molecular Devices), which was diluted in Krebs−Ringer−Hepes (KRH) buffer (10 mM Hepes, 4.7 mM KCl, 2.2 mM CaCl₂, 1.2 mM KH₂PO₄, 1.2 mM MgSO₄, 120 mM NaCl, pH 7.4) supplemented with 1.8 g/L glucose at room temperature (RT) for 40 min before the assay. Assays were performed in a Flexstation 3 microplate reader (Molecular Devices) that recorded fluorescence (excitation at 530 nm, emission at 565 nm) at 1.5-s intervals for a total of 180 s at 32 °C. Agonists (dopamine or serotonin) in KRH buffer were added automatically by the Flexstation 3 apparatus to each well after taking a baseline of background fluorescence for 30 s. G_i activation induced by agonist-stimulated D2R triggers the opening of TRPC4β channels, leading to changes in membrane potential, which were reported by enhanced fluorescence from the membrane potential dye. To determine the efficacy of G_i activation induced by dopamine-stimulated D2R, maximal fluorescence values induced by a saturating amount (10 μM) of dopamine were subtracted by the negative control background and then normalized to the receptor surface expression, which was determined by cell-ELISA. For generation of dose–response curves to determine EC₅₀, cells were treated with various concentrations of agonists, and the data were collected as described above. Maximal fluorescence values induced by agonists were used as points in the dose–response curves. The dose–response data were then analyzed by nonlinear regression curve-fitting analysis with the Hill coefficient set at 1 using GraphPad Prism v. 5.01 software to obtain EC₅₀. For both efficacy and potency measurements in intact cells, the amount of DNA for transfection was determined by titration to yield a response whose amplitude was linear with DNA amount, avoiding complications due to excessive receptor levels.

Cell-ELISA.

At 24 h posttransfection of negative control [pcDNA3.1(+)], WT, or mutant D2Rs, the transfected TRPC4β-expressing HEK293 cells were washed with phosphate-buffered saline (PBS) and fixed with 4% (vol/vol) paraformaldehyde in PBS for 15 min at RT. The cells were then washed three times with PBS and incubated with 2% (wt/vol) BSA in PBS for 1 h followed by 1-h incubation with the mouse monoclonal HA-probe antibody (clone F7; Santa Cruz Biotechnology) and then goat anti-mouse IgG secondary antibody conjugated with HRP (Thermo Scientific) at RT for 1 h. The antibodies used were diluted in PBS with 1% (wt/vol) BSA. PBS washes were performed after each incubation of the antibody. To detect the cell surface expression level of D2Rs, SuperSignal ELISA Femto Maximum Sensitivity Substrate (Thermo Scientific) was added, and the luminescence was detected with a Flexstation 3 microplate reader. The obtained luminescence values were then subtracted by the negative control background. For the measurement of total D2R expression, the cells were permeabilized with 0.4% (vol/vol) Triton X-100 in PBS before incubation with antibodies. The cell surface to total expression ratios of mutant D2Rs were normalized to that of WT as shown in Fig. S8.

Fig. S8.

Quantification of cell surface and total expression levels of D2R. TRPC4β-expressing HEK293 cells were transiently transfected with the WT or mutant D2Rs for cell-ELISA to determine the cell surface and total expression levels of D2R. D2R expression was detected using an HA-probe antibody followed by secondary antibody conjugated with HRP. For the detection of total D2R expression, cells were permeabilized by 0.4% (vol/vol) Triton X-100 before antibody labeling. The ratio of cell surface to total D2R expression was calculated, and normalized to that of WT D2R, which was defined as 1. Each bar represents mean ± SEM (n = 3–7; **P < 0.001; one-sample t test against 1). Bars are color-coded according to the evolutionary coupling potential predicted using amino acid sequences sharing >35% identity with D2R.

Membrane Preparation from HEK293 Cells.

HEK293 cells transfected with WT or mutant D2Rs were scraped and rinsed with PBS. Cell pellets were resuspended in buffer (1 mM Tris⋅HCl, 10 mM EDTA, pH 7.4) containing Complete protease inhibitor mixture (Roche). Cells were homogenized with a glass/teflon Potter−Elvehjem homogenizer and then further lysed by repeated passage through a 26-gauge needle. To remove nuclei and unbroken cells, the homogenates were centrifuged at 260 × g for 10 min at 4 °C. The supernatant was collected and centrifuged at 110,000 × g for 1 h at 4 °C. The resulting membrane pellets were resuspended in reaction buffer (20 mM Tris⋅HCl, 150 mM NaCl, 1 mM MgCl₂, 10 mM EDTA, pH 7.4) supplemented with Complete protease inhibitor mixture (Roche) and stored at −80 °C until use.

Radioligand Binding Assays.

The radioligand binding assays were carried out with washed cell membranes from cells transfected and expressing much higher levels of D2R than cells for membrane potential assays. Based on immunofluorescence, most of the overexpressed D2R is in internal membranes, and most is likely not coupled to G proteins because of both the location and the large excess of receptor. Although this type of assay does not reflect the affinity of the receptor at the cell surface under physiological conditions, where there would be two (or possibly more) effective affinities, it allows us to make comparisons among receptor variants independent of their efficiency of G protein coupling, with results that are well fit with a single IC₅₀. The membrane suspension (5−25 µg of protein) in reaction buffer was incubated with [³H]spiperone at a final concentration ranging from 1 nM to 8 nM in the saturation binding experiments. Competition binding experiments were performed with a fixed concentration of [³H]spiperone (0.3 nM for serotonin competition, 0.4 nM for dopamine competition) and various concentrations of cold ligand competitors dopamine or serotonin. The nonspecific binding was defined as binding of [³H]spiperone to the membranes of negative control [pcDNA3.1(+)] HEK293 cells. To determine the specific [³H]spiperone binding under each condition, nonspecific binding was subtracted from the total binding of [³H]spiperone. Each binding reaction was performed in triplicate and incubated at RT for 1.5 h in a final volume of 525 µL. The reaction was terminated by filtration through glass microfiber filters (Whatman GF/A) using a 12-well sampling vacuum manifold (Millipore) followed by three washes of 2 mL ice-cold TBS buffer (50 mM Tris⋅HCl, 150 mM NaCl, pH 7.4). The filter-bound radioactivity, which reported on the amount of [³H]spiperone binding to the cell membranes, was quantified using a liquid scintillation counter (Beckman LS6500). Data from saturation and competition binding were analyzed by nonlinear regression analysis using GraphPad Prism v. 5.01 as described in ref. 23 to obtain K_d and K_i.

Calculation of Gibbs Free Energy Change upon Receptor−Ligand Binding.

Values of the free energy change upon receptor−ligand binding were calculated from the equation ∆G⁰ = RT lnK_i, where R = 8.314 J K⁻¹⋅mol⁻¹ and T = 298.15 K. K_i for dopamine or serotonin was determined by competition binding experiments. To assess the mutational effects on binding free energy, ∆∆G⁰ of each mutant was calculated by the equation ∆∆G⁰_mutant = ∆G⁰_mutant − ∆G⁰_wt. To determine whether the effect of double mutations is additive or not, independent two-tailed Student’s t tests were applied to the comparison of (∆∆G⁰_A + ∆∆G⁰_B) versus ∆∆G⁰_AB, where A is single mutant A, B is single mutant B, and AB is double mutant.