Introduction

Cell activity is regulated by extracellular signals that are recognized and transduced inside the cell via different classes of plasma membrane receptors [16, 122, 168]. G protein-coupled receptors (GPCRs) constitute the largest superfamily of signal transduction membrane proteins, which play a central role in many essential physiological processes including vision (reviewed in refs [14, 48, 83, 85, 122]). Vision in dim-light conditions is triggered upon photon absorption by the visual pigment rhodopsin, which is made of a protein, the GPCR opsin, and the chromophore 11-cis-retinal, covalently linked via Schiff base (SB) to K296 [35, 70, 111, 115]. As a GPCR, rod opsin is made of seven transmembrane (TM) α-helices (H) holding an up-down bundle architecture. The TM helices are connected by three intracellular (cytosolic) loops (I) and an equal number of extracellular loops (E) (intradiscal in the case of rod opsin) [35, 70, 111, 115]. The helix bundle begins with an intradiscal N-terminus (Nt) and ends up with a cytosolic C-terminus (Ct).

The common mechanism of signal transduction by GPCRs consists in agonist-induced promotion of allosteric interaction between the receptor and a member of α-family heterotrimeric αβγ guanine nucleotide-binding proteins or G proteins, which belong to the Ras GTPase superfamily [165]. These are specialized transducers that broadcast signals to intracellular effectors. Transducin (Gt) is the Gα protein deputed to visual phototransduction, the process in which light captured by a visual pigment molecule generates a detectable electrical response [11, 29, 30, 84].

The first event of vision in dim-light conditions is the absorption of a photon by rhodopsin, which causes the cis-trans isomerization of 11-cis-retinal and a conformational change to its active state (metarhodopsin II, MII state). MII activates heterotrimeric GDP-bound transducin by catalyzing the exchange of GDP for GTP. In that respect, as all GPCRs, MII acts as a guanine nucleotide exchange factor (GEF). Upon dissociation from the βγ dimer, the Gtα-GTP complex stimulates in turn the activation of phosphodiesterase (PDE), thus decreasing the cytoplasmic concentration of cGMP leading to channel closure. Rapid inactivation of MII requires phosphorylation of MII by rhodopsin kinase followed by binding to visual arrestin (arrestin 1 (Arr1)). Deactivation of Gtα requires the hydrolysis of GTP to GDP by the GTPase activity of Gtα, which is catalyzed by a member of the regulators of G protein signaling (RGS) family, acting as a GTPase activating protein (GAP). Complete recovery of the photoresponse requires also restoration of cytoplasmic cGMP to the dark level by the intervention of guanylate cyclase (GC) and calcium-sensing guanylate cyclase-activating proteins (GCAPs) [45, 79].

GPCRs, including rod opsin, undergo dimerization/oligomerization, as an essential component of their life cycle (reviewed in refs [15, 93, 101, 116, 121]).

Another feature of GPCRs is that they can be constitutively active [28, 86], i.e., they can be activated in the absence of an agonist. Constitutively active mutants (CAMs) of GPCRs have been linked to a number of human diseases (reviewed in refs [4, 143, 158, 160]). Diseases can be caused either by activating mutations (“gain-of-function” mutations) or by deactivating mutations (“loss of function” mutations) of GPCRs. Spontaneous mutations in any player of visual phototransduction have been found associated with retinal disease including autosomal congenital stationary night blindness (adCSNB) and autosomal dominant/recessive retinitis pigmentosa (adRP and arRP). While adCSNB is essentially linked to gain-of-function mutations, adRP is mainly linked to loss-of-function mutations. RP comprises a class of hereditary diseases causing the progressive degeneration of the photoreceptor system [54] and visual impairment in 1.5 million patients worldwide [40]. RP affects about 1 in 3000–7000 people, being the most common cause of inherited blindness in developed countries [40, 73, 100]. Predominantly affected are rod cells that aggregate and progressively affect cone cell viability [125]. More than 40 genes have been linked to this hereditary disease. Mutations of the RHO gene represent the most common cause of RP, accounting for 25% of adRP and 8 to 10% of all RP. The majority of the over 140 missense mutations found in RHO are linked to adRP [18].

adRP RHO mutants have been subjected to extensive biochemical and cellular characterizations, leading to a classification into three major groups (I, IIa/II, and IIb/III) (reviewed in [18, 82]), based on rod opsin ability to reconstitute with 11-cis-retinal and to be retained into endoplasmic reticulum (ER) [82]. In detail, class I mutants resemble wild type (WT) in ability to regenerate with 11-cis-retinal and subcellular localization. In contrast, class IIa (or II in [82]) and class IIb (or III in [82]) mutants show altered membrane localization in their opsin state associated with ability or inability, respectively, to regenerate with 11-cis-retinal in vitro.

Another classification of RHO mutants linked to adRP and adCSNB is based on a number of variegate effects on rod opsin function (classes 1–7) [5, 100]. The considered effects that found this alternative classification include post-Golgi trafficking and outer segment targeting (class 1), ER retention and impairment in 11-cis-retinal binding (class 2), vesicular trafficking and endocytosis (class 3), post-translational modifications (class 4), Gt activation (class 5), constitutive activation (class 6), and efficiency to dimerize (class 7) [5, 100].

The majority of the over 140 RHO mutants linked to adRP are likely misfolded and variedly retained into ER (reviewed in [5, 18, 82, 127]). Proteins are monitored by a quality-control system (QCS) in the ER that can retain misfolded structures in the ER for their subsequent degradation (reviewed in [27]). QCS is able to recognize general “errors,” e.g., non-native solvent exposure of hydrophobic regions, defects in secondary structures and their packing interactions, unpaired cysteines, or immature glycans (reviewed in [27]), thus preventing protein aggregation. The etiology of conformational diseases can be due to proteins that are either misfolded immediately after synthesis or undergo post-translational conformational alterations. Intracellular GPCR oligomerization can lead to either cell-surface targeting or misrouting. Misrouted mutant GPCRs, when co-expressed with the WT form, can cause intracellular retention of the WT receptor via oligomerization, thus exerting a dominant-negative effect (reviewed in [27]). ER retention by misfolded rod opsin mutants causes ER stress [91], to which cells respond by activating a signaling network called the unfolded protein response (UPR). As a conformational disease [27, 90, 137], adRP linked to RHO mutations is amenable to treatment with pharmacological chaperones, small molecules that bind specific sites within a protein native or quasi-native structure, thereby shifting the folding equilibrium towards the native state, thus allowing correct routing. adRP RHO mutants that respond to small chaperones like 11-cis-retinal can be seen as structurally less stable than the native state due to the loss of some native intramolecular interactions, which is recovered by the small chaperone. Pharmacological therapies for misfolding diseases are based on either promoting correct folding, inhibiting aggregation, increasing degradation, or protecting from cell death [99, 100]. Thus, pharmacological chaperones for misfolded mutant opsins have to intervene early in protein synthesis and folding to compensate for the observed decrease in opsin stability leading to rod and cone cell death [99].

This review article focuses on the advances into rod opsin structure and their implication in our understanding of the determinants of adCSNB and adRP.

Structural models of the dark state of vertebrate rhodopsin

Rod opsin is the founder of family A GPCRs and the best source of high-resolution information on the homologous receptors.

The first highly resolved structure of bovine rhodopsin, deposited in the protein data bank (PDB, http://www.wwpdb.org/) under the identifier 1F88, showed all major structural features, which had been predicted by years of biochemical, biophysical, and bioinformatics studies [118].

Bovine rod opsin contains 348 amino acids, which fold into seven TM helices, varying in length from 19 to 34 residues, and one cytoplasmic helix, H8. All the seven TM helices, except for H3, hold a number of irregularities, e.g., kinks, twists, and bends (Fig. 1a), due to the presence of glycines, prolines, or threonines. Such irregularities are also present in all rod opsin structures successively released. All helices, except for H4 and H7, are tilted from the membrane normal [118].

Fig. 1
figure 1

Side view, in a direction parallel to the membrane surface, of the superimposed highest resolution dark rhodopsin structures (PDB: 1U19 [113] and 1GZM [87]). On the left side, a cartoon representation of the whole structures is shown, including the 11-cis-retinal, represented by black sticks. On the right side, a stick representation of selected highly conserved amino acids in the seven-helix bundle is shown. The cytosolic side is at the top. H1, H2, H3, H4, H5, H6, and H7 are, respectively, colored in blue, orange, green, pink, yellow, cyan, and violet; Nt and Ct, the latter including H8, are red, I1 and E1 are lime, I2 and E2 are gray, and I3 and E3 are magenta

Full size image

Rhodopsin family GPCRs share a few highly conserved amino acids distributed in all TM helices. They include N55(1.50), N78(2.45), D83(2.50), C110(3.25), E134(3.49), R135(3.50), Y136(3.51), W161(4.50), P215(5.50), Y223(5.58), F261(6.44), W265(6.48), P267(6.50), N302(7.49), P303(7.50), and Y306(7.53) (the numbering in parentheses follows the arbitrary scheme by Ballesteros and Weinstein [8]; the first number identifies the helix number, whereas the two numbers after the dot indicate the position relative to a reference residue among the most conserved amino acids in the helix; such a reference residue is arbitrarily assigned the number 50). The cytosolic extension of H3 holds the highly conserved E/DRY motif. In the inactive state of rhodopsin (i.e., in the dark), R135(3.50), which in rod opsin is an adRP mutation site, forms a double salt bridge with both the adjacent glutamate, E134(3.49) and E247(6.30) (Fig. 1b) [118]. Both the E3.49-R3.50 and R3.50-E6.30 interactions would contribute to maintain the inactive state of rhodopsin family GPCRs, based on the results of in vitro and computational experiments (reviewed in [35]). The higher degree of conservation of D/E3.49 than E6.30 would, however, make the former residue more relevant than the latter in stabilizing the resting state. Together with H3, H6 is the second longest helix. Its cytosolic half is almost perpendicular to the membrane plane, whereas its extracellular half is bent, because of P267(6.50), which in rod opsin is an adRP mutation site (Fig. 1a) [6, 7, 102]. H6 holds the conserved FxxCWxPY motif forming an aromatic cluster, with W265(6.48) and Y268(6.51) participating in 11-cis-retinal binding. H7 contains a 3.10 helical stretch in the region around the retinal attachment site, K296(7.43), and two prolines, P291(7.38) and P303(7.50), the latter belonging to the highly conserved NPxxY motif. The tyrosine of this motif is involved in π-π stacking interactions with F313 in H8 (Fig. 1b).

Packing of the cytosolic regions diverges from that of the intradiscal regions, the latter being significantly associated with each other and well resolved, contrarily to the former. Indeed, the Nt holds a β-hairpin forming a four-stranded antiparallel β-sheet with the β-hairpin in E2 that plugs the retinal, which has to enter through pores opened in between the TM helices [139]. E2 is engaged in a disulfide bridge with C110(3.25). Both C110(3.25) and C187 are sites of adRP misfolding mutations, indicating the relevance of the bridge in rhodopsin stability.

The most resolved rhodopsin structure, 1U19, holds the complete backbone [112, 113, 118, 159]. Li et al. resolved a rhodopsin structure at 2.65 Å (PDB: 1GZM) [87] by molecular replacement from the 1F88 structural model [118]. Like 1U19, 1GZM also resolves all the interhelical loops. The most significant main chain differences between 1GZM and 1U19 reside in I2, I3, and the C-terminal segment following H8 (Fig. 1a), the latter being disordered in 1GZM [87, 113]. Other differences between 1U19 and 1GZM concern the retinal ring, whose conformation in 1GZM is the same as in 1F88 [87, 113, 118] (Fig. 1a).

Insights into vertebrate rod opsin photobleaching from spectroscopic and structure determinations

Vertebrate rod opsin is the GPCR best characterized at atomic resolution in its inactive and active states. To date, a number of crystal structures of bovine rhodopsin have been released, which correspond to different states in the photoactivation cascade (Table 1). These include (a) rhodopsin in the dark state [87, 94, 112, 113, 118, 135, 150, 159]; (b) rhodopsin containing the non-native chromophore 9-cis-retinylidene (isorhodopsin) [104], or a 6-carbon-ring retinal [53], or octyl beta-D-glucopyranoside; (c) two early photointermediates [105, 106]; (d) a thermostabilized mutant in the dark state [149]; (e) a photoactivated deprotonated intermediate [135]; (f) the active opsin (Ops*) apoprotein both in its free state [120] and in complex with the Ct of transducin (GtCt) [13, 141]; (g) Ops* in complex with non-retinoid ligands [13, 95, 119]; (h) the E113Q CAM bound to all-trans-retinal and the GtCt peptide [148]; (i) MII both in its free and GtCt-bound forms [24]; (j) G90D and T94I CSNB mutants [146, 147]; (k) the cryoEM structure of an inactive opsin dimer [169]; (l) rhodopsin in complex with G proteins [46, 63, 68, 161, 162]; (m) dimeric opsin in complex with the finger loop of Arr1 [156]; and (n) opsin bound to Arr1 [67, 170, 171] (Table 1).

Table 1 X-ray structures of vertebrate rod opsins
Full size table

Photochemical experiments allowed definition of the reaction coordinates of rhodopsin activation [97, 111]. Photoactivation of rod opsin include (1) cis-trans isomerization of the retinal, (2) thermal relaxation of rhodopsin complex, and (3) formation of the signaling active states (reviewed in refs [59, 97, 111]).

Photobleaching of rhodopsin involves different intermediates (Fig. 2) [9, 55, 77, 111, 167].

Fig. 2
figure 2

Rhodopsin photobleaching. The process of rhodopsin photoactivation is schematized here. The retinal and selected surrounding amino acid residues are represented as sticks for the dark, BATHO, LUMI, and MIIb H+ states (highlighted in red)

Full size image

Following photon absorption and electronic excitation, fast isomerization of the chromophore leads to formation of bathorhodopsin (BATHO, 529 nm) (Fig. 2) on the fs time scale (i.e., in 200 fs). BATHO is in equilibrium with the blue-shifted intermediate (BSI, 477 nm), which decays to lumirhodopsin (LUMI, 492 nm) in 150 ns. Whereas the BATHO to BSI transition involves conformational changes of retinal, the BATHO to LUMI transition is accompanied by a movement of the β-ionone ring away from W265(6.48) (Fig. 2). Indeed, the distorted all-trans retinal in the BATHO structure (PDB: 2G87) dislocates its β-ionone ring in the LUMI intermediate (PDB: 2HPY) [105, 106]. The adRP mutation site, E181 (in E2), would be charged in the BATHO state [136]. LUMI transforms to metarhodopsin I (MI, 478 nm) in 10 μs. MI formation would not involve large helix movements, but only a rearrangement close to the bend of H6, at the level of the retinal chromophore [133]. These results indicate that rhodopsin remains in a ground state-like conformation until late in the photobleaching process, the gross conformational changes occurring in the MI to MII transition, the only one occurring on a slow time scale (i.e., 1 ms) [133]. The transition depends on and probably involves protonation of E134(3.49), which would require the presence of transducin [3, 34, 111]. Protonation of E134(3.49) of the E/DRY motif would destabilize the salt bridge between E134(3.49) and R135(3.50), releasing an important constraint of the inactive state. This hypothesis is supported by the findings that the E134Q mutation induces constitutive activation of rod opsin [26]. Proton translocations to E134(3.49) and E113(3.28) can be decoupled, corresponding to the MIIa and MIIb states (Fig. 2) [3, 111]. According to this model, the transition of MI to MIIa is accompanied by translocation of the SB proton to E113(3.28), whereas MIIb formation requires also proton uptake from the cytoplasm, with a pH-dependent ΔG [111]. Formation of the signaling active state would merely be a release of constraints in the helix bundle, leading to the exposure of cytosolic portions deputed to G protein recognition. Spectroscopic studies on rhodopsin allowed an extension of the MI↔MIIa↔MIIb to MI↔MIIa↔MIIb↔MIIbH+ equilibrium scheme, meaning that H6 motions accompanying MIIb formation lead to the MIIbH+ state following proton uptake by E134(3.49) from the cytosol (Fig. 2) [78]. The latter would be important for GDP release from Gt [138].

Initial deactivation of MII relies on opsin interaction with its kinase, which phosphorylates the receptor promoting tight binding to Arr1 [114, 124]. Full deactivation requires rhodopsin regeneration [38, 56]. This requires the hydrolysis of the all-trans SB and depletion of all-trans-retinal from the active site, a step that requires the nucleophilic attack of structural water in the hydrophobic retinal binding site [38, 56]. Formation of opsin is accompanied by increase in intrinsic tryptophan fluorescence after releasing all-trans-retinal from the active site [38, 56]. The retinal remains associated with opsin membranes and is converted by endogenous NADPH-dependent retinol dehydrogenase (RDH, reviewed in ref [97]) to all-trans-retinol. During the MII decay, a storage form of rhodopsin, metarhodopsin III (MIII), is generated. The formation of MIII can be induced by blue light absorption in MII, passing through the anti-syn isomerization form or “reverted-Meta” intermediate MIII, and the subsequent reprotonation of the Schiff base [55, 132]. It has been inferred that, in addition to the retinylidene pocket (site I), there are other two retinoid-binding sites within opsin: site II, the entrance site, and site III, the exit site. The latter is occupied when retinal remains bound after its release from site I [139]. The MIII form is suggested to hold photolyzed all-trans-retinal bound in site III. Opsin finally returns to the ground state via a transiently formed opsin-11-cis-retinal complex, which contains both retinal isomers bound to site II and site III [139].

The activity of ligand-free opsin is 106 times lower than the activity of the all-trans-retinal-bound active MII state, but it is higher than the activity of dark rhodopsin, thus indicating that 11-cis-retinal acts as an inverse agonist [98, 111]. Lower pH enhances opsin activity likely due to protonation of E134(3.49) [59, 141].

X-ray determination of bovine Ops* structure (PDB: 3CAP [120]) provided a significant advancement in our understanding of ligand-independent GPCR activation. SB hydrolysis makes MII unstable and difficult to crystallize. The crystal structure of the MII state was, therefore, obtained by soaking the Ops* crystals with all-trans-retinal forming a SB with K296(7.43) [24]. Such structure, released both in its free state (PDB: 3PXO [24]) and in complex with the GtCt peptide (PDB: 3PQR [24]), as well as the structure of the E113Q CAM in complex with GtCt (PDB: 2X72 [148]), are almost identical to the corresponding free and GtCt-bound states of Ops*.

Major structural differences between dark rhodopsin and Ops*/MII essentially concern the cytosolic halves of H5, H6, H7, and I3 (Fig. 3). Indeed, H5 is longer in Ops*/MII than in dark rhodopsin, with its cytoplasmic end shifted by 2–3 Å towards H6. The cytosolic end of H6 is tilted from H3 by 6–7 Å (Fig. 3c). These displacements are associated with the breakage of both the salt bridge interactions involving R135(3.50) in the resting dark state. This is accompanied by the establishment of novel interactions between, (a) R135(3.50) and the conserved Y223(5.58) and (b) E247(6.30) and K231(5.66). Moreover, the conserved Y306(7.53) of the NPxxY motif rotates to face the helix bundle, thereby blocking H6 from moving back towards H3 in an inactive state conformation (Fig. 3b).

Fig. 3
figure 3

Structural comparisons of active and inactive rod opsin structures. Cartoons of the superimposed structures of dark rhodopsin (PDB: 1GZM, violet [87]) and MII (PDB: 3PXO, green [24]) are shown. The helix bundles are seen both in a direction parallel to the membrane surface with the cytosolic side being at the top (a and b) and from the cytosolic side in a direction perpendicular to the membrane surface (c). The side chains of selected highly conserved and of retinal binding site amino acids and the backbone of E2 are shown (c). The 11-cis-retinal and all-trans-retinal are represented as black and red sticks, respectively

Full size image

The opening of a cytosolic crevice in between H3 and H6, which follows the outward movement of H6, is instrumental in forming the docking site for transducin, as shown by the crystallographic complexes between Ops*/MII/E113Q and GtCt (PDB: 3DQB [141], 3PQR [24], and 2X72 [148], respectively).

The striking similarity between Ops* and MII structures contrasts with differences in their coupling efficiency to transducin.

The structures of activated opsin in complex with stabilizing detergents or other small compounds bound onto or near the orthosteric retinal binding site provide valuable targets for structure-based discovery/design of stabilizing non-retinoid ligands [13, 95, 119].

Structures of signaling active rod opsin states in complex with G proteins or visual arrestin

Since very recently, the structures of rod opsin in ternary (i.e., with bound all-trans-retinal) or binary (in the apo state) complexes with heterotrimeric Go, Gi, or Gt have been released providing insights into a fundamental step in visual phototransduction [46, 63, 68, 162].

Heterotrimeric G proteins belong to the α-family of the Ras GTPase superfamily. Gα proteins are made of two domains, the Ras-like or GTPase (G) domain, which is shared by all Ras GTPases, being deputed to GDP/GTP binding and GTP hydrolysis, and a helical (H) domain [165].

The following structural description of the α-subunit employs the Noel’s nomenclature (Fig. 4a, b) [108].

Fig. 4
figure 4

OFF and GEF-bound states of heterotrimeric Gi. The inactive GDP-bound state (a) and the nucleotide-free opsin-bound states (b) of heterotrimeric Gi are shown. The G and H domains of the α-subunit are green and violet, respectively, the β- and γ-subunits are aquamarine and orange, respectively, and opsin is gray. In the inactive-state structure (a), the GDP bound in between the G and H domains of the α-subunit is represented as red sticks. In all panels, the secondary structure elements are labeled according to Noel’s nomenclature [108]

Full size image

The G domain holds a Rossmann fold, characterized by a 3-layer(αβα) sandwich architecture holding a five-stranded parallel β-sheet sandwiched between two layers of α-helices, α1 and α5, on one side, and α2, α3, and α4 on the other side. The β1/α1, α1/β2 (αF/β2 in the Gα proteins), β3/α2, β5/α4, and β6/α5 loops (i.e., G boxes 1-5 (G1-G5)) hold sequence conservation and are deputed to nucleotide binding (Fig. 4). G1, the β1/α1 loop, is also named P-loop (phosphate-binding loop) as it contacts the phosphates through main-chain NH groups and lysine side chain. G2, the α1/β2 loop (αF/β2 loop in the Gα proteins), is also called switch I (swI). G3 is part of the switch II (swII), which is made of the β3/α2 loop plus the α2-helix. G4 and G5 are, respectively, the β5/α4 and β6/α5 loops, which make contacts with the guanine base.

Differently from small Ras GTPases, Gα proteins hold a third switch, swIII, that is the loop connecting β4 to α3 [108] as well as the H domain made of six helices (αA-αF) organized in an orthogonal bundle architecture and a Gi alpha 1, domain 2-like topology. The loops that connect G and H domains are linker1 (the α1/αA loop) and linker 2 (the αF-β2 loop corresponding to the swI) (Fig. 4). The nucleotide, GDP or GTP, binds at the interface between G and H domains (Fig. 4a).

In heterotrimeric G proteins, the β-subunit holds a 7-propellor architecture and a methylamine dehydrogenase topology, whereas the γ-subunit holds an irregular architecture of two α-helices connected by a loop and a G Protein Gi Gamma 2 topology.

The signaling active states of GPCRs, in general, and of rod opsin, in particular, recognize the Ct of α5 and the α/β loops of the Gα G domain, which are distal from the nucleotide binding site [129]. The first insights into the mechanism of GPCR-catalyzed GDP release arose from molecular simulations and biophysical determinations as well as from structure determination of the ternary complex made of agonist-bound β2-adrenergic receptor (β2AR) and nucleotide-free heterotrimeric Gs [41, 49, 74, 126, 129, 140, 153, 163]. These experiments indicated that, by acting on the orientation of α5 and on the conformation of G5, GPCRs weaken the contacts between GDP and G domain accompanied by a detachment of the H domain from the G domain, ultimately leading to GDP release [41, 49, 74, 126, 129, 140, 153, 163].

The structures of rod opsin in complex with heterotrimeric G proteins confirmed the gross binding mode of the G domain to the cytosolic regions of the receptor as found in the β2AR-Gs complex (PDB: 6CMO, 6FUF, 6OYA/B, 6QNO [46, 63, 68, 162]). In particular, α5 of Gα docks in between H3 and H6 and contacts H8 of rod opsin, whereas αN of Gα interacts with I2 of rod opsin (Fig. 4b). Differences, however, occur in the way GsCt and GiCt/GtCt dock onto the cytosolic cleft of β2AR and rod opsin, respectively [46, 63, 68, 129, 162]. Indeed GiCt or GtCt make a more extended hydrophobic interface with H6 of rod opsin than GsCt does with H6 of β2AR. This may be responsible for the different conformational behavior of H6, whose outward motions would be less pronounced for Gi-bound rod opsin than Gs-bound β2AR [46, 63, 68, 129, 162]. This different dynamics, likely linked to chemico-physical differences in the GαCt-H6 interface, might determine receptor-G protein coupling specificity [68].

Three out of the four opsin-G protein complexes released so far concern mini G proteins bearing only the G domain [46, 63, 68, 162]. In the cryoEM structure of rod opsin in complex with heterotrimeric Gi holding both domains, a dominant-negative mutant form of Gi was used to promote the nucleotide-free form of the G protein [68]. Such complex displays a detachment of the H domain from the G domain, though less marked compared to the one seen in the β2AR-Gs complex [46, 63, 68, 129, 162]. These structural models strengthen the evidence that interdomain motion is an essential step in the process of GDP depletion from Gα.

The cryoEM structure of rod opsin and heterotrimeric mini Gi shows that the receptor Ct interacts with the Gβ subunit of the G protein, providing a structural foundation for the role of Ct in GPCR signaling and of Gβ as a scaffold for recruiting Gα subunits and G protein-receptor kinases [63]. The study suggests that the observed interaction between the central part of opsin Ct (OpsCt) and Gβγ is instrumental in localizing the G protein heterotrimer to the active receptor first. After dissociation of the Gα subunit, a transient Gβγ-receptor complex could provide the adequate molecular framework for receptor phosphorylation by bringing the kinase close to the OpsCt [63].

Providing insights into opsin GEF action and the linked mechanism of G protein activation, structure determinations of opsin-G protein complexes represent invaluable frameworks to infer the determinants of inherited retinal diseases affecting opsin-transducin coupling. In this respect, non-misfolded adRP rod opsin mutants such as M44T and V137M (in positions 1.39 and 3.52, respectively), both located in the cytosolic end of rod opsin helix bundle, have been found to alter opsin-transducin coupling by accelerating the early steps of transducin activation, without increasing the basal activity of the G protein [2]. Moreover, rod opsin mutants linked to CSNB, e.g., G90D, constitutively activate transducin by perturbing the SB [147].

Upon phosphorylation by receptor kinases, rod opsin undergoes fast desensitization by binding to visual arrestin (Arr1). Arr1, which, differently from Arr2 and Arr3, is uniquely deputed to recognize opsins, holds a β-sandwich architecture and an immunoglobulin-like topology made of N-terminal (residues 8-180) and C-terminal (residues 188-362) domains (ND and CD, respectively) (Fig. 5a).

Fig. 5
figure 5

Visual arrestin structures in the free state and in complex with rod opsin. a The inactive state of Arr1 is shown (PDB: 1CF1 [58]). The finger, middle, lariat, and C loops are, respectively, magenta, blue, orange, and yellow. The Ct-tail is red, and the ND and CD are, respectively, green and aquamarine. b The superimposed structures of Arr1 in the inactive (violet, PDB: 1CF1 [58]) and opsin-activated (green, PDB: 5W0P [171]) states are shown. c The complex between opsin-activated Arr1 (colored as in a) and opsin (gray) (PDB: 5W0P [171]) is shown

Full size image

Each domain is made of a seven-stranded β-sandwich plus an additional lateral strand [58]. The sandwich comprises a four-stranded β-sheet packed against a three-stranded β-sheet. ND contains a single α-helix, while CD contains two short 310 helices (Fig. 5a) [58]. The terminal residues of the protein, “C-tail” (residues 372–404), are connected by a flexible linker (residues 362-371) to CD (Fig. 5a). In the resting state, the C-tail forms a parallel β-sheet with the lateral strand of ND. The topology of the two domains is such that three loops, i.e., finger loop (residues 68–78), middle loop (residues 133–142), and C-loop (residues 248–253) form a central crest. The finger loop is a key receptor-binding element (Fig. 5a) [142]. The interface of ND and CD is stabilized by a number of interactions including those in the so-called polar core. Receptor-arrestin recognition would occur in two steps. In the first step, the phosphorylated receptor Ct (OpsCt, in the case of opsin) displaces the C-tail of arrestin and recognizes positively charged amino acids on ND, thus forming a low affinity pre-complex. The second step, triggered by interdomain movements, is characterized by binding of a number of arrestin loops, including finger, middle, and lariat (residues 281–321) loops, to the cytosolic regions of the receptor, thus forming a high-affinity complex [57, 142].

Advances in structure determinations shed light on the mechanisms of arrestin activation. The active-state structures of Arr1 concern a naturally occurring constitutively active form called p44 (PDB: 4J2Q) [75] and the opsin-bound state. The recent structure of phosphorylated opsin in complex with Arr1 (PDB: 5W0P) unveiled the structural characteristics of the high affinity receptor-arrestin complex (Fig. 5c) [171]. The complex is characterized by two interfaces. The primary interface is contributed by interactions between OpsCt and a positively charged crevice in Arr1 ND. The secondary interface concerns the interactions between Arr1 and the receptor core and includes (a) I1, H6, and H8 of opsin and the finger loop of Arr1; (b) I2 of opsin and the middle, lariat, and the 236–265 β-hairpin-strands of Arr1; and (c) H5 of opsin and the C-loop of Arr1 (Fig. 5c). In the opsin-bound state, the finger loop of Arr1 acquires two-turns of α-helix compared to the receptor-free form (Fig. 5b). Other hallmarks of Arr1 activation, inferred from the structural model of opsin-Arr1 complex, include breakage of the polar core, ~21° interdomain rotation about an intramolecular pseudo two-fold axis, and conformational changes within the central crest loops.

Remarkably, Arr1 competes for the same binding site of the G protein, by docking into the cytosolic crevice formed by the outward motion of H6, as an effect of receptor activation (Figs. 4b and 5c). In this respect, structure determinations suggest mutually exclusive binding of G protein and Arr1 to activated opsin. Similarly to the opsin-G protein complexes, the structure of opsin in complex with Arr1 may help inferring the determinants of adRP linked to mutations that affect opsin desensitization and endocytosis. These include mutants at R135(3.50) of the highly conserved E/DRY motif, which is involved both in transducin and Arr1 recognition. The adRP rod opsin mutant R135W was found to be hyperphosphorylated and bound with high affinity to Arr1 in the absence of chromophore, being constitutively internalized [25]. This causes accumulation into rod inner segment and prevents traffic to the outer segment contributing to disease. The presence of a tryptophan in place of R135 would favor the formation of a hydrophobic crevice in opsin that would tightly bind the hydrophobic face (i.e., contributed by I73, M76, and L78) of the finger-loop helix of Arr1, thus explaining in part the high affinity of mutant opsin for Arr1. Remarkably, the adRP-linked mutation of R135 into another hydrophobic amino acid like leucine displays the same phenotype as R135W [1, 25].

Supramolecular organization of rod opsin

The oligomeric state in native membranes was demonstrated years ago for rhodopsin [42, 43]. According to AFM experiments, the native arrangement of rhodopsin in mouse disk membranes would consist in paracrystalline rows of dimers [43]. The authors also demonstrated that rhodopsin dimers hasten the process of transducin activation by accelerating rhodopsin-transducin recognition [44, 64, 65, 152]. Concentration-dependent oligomerization was also observed by Förster energy transfer (FRET) spectroscopy in Chinese hamster ovary (CHO) cells [52, 103]. Whether rhodopsin oligomerization is an artifact or a physiologic condition has long been debated [19,20,21, 29, 121].

AFM determinations permitted the building of a semiempirical structural model of oligomeric rhodopsin (PDB: 1N3M, Fig. 6a) [89], made of repeats of the same monomeric unit (i.e., a completed 1HZX structure [159]). According to this model, two monomers of rhodopsin make contacts mediated, for both monomers, by E2, I2, H4, and H5 (Fig. 6a). Contacts between dimers involve I3 and both I1 and Ct (Fig.5a). This oligomeric model was supported by cysteine cross-linking experiments indicating that rhodopsin dimers would rely on W175-W175 (in E2) and Y206(5.41)-Y206(5.41) contacts [81].

Fig. 6
figure 6

Structural models of rhodopsin oligomers. Rhodopsin hexamers are shown, which have been obtained by fitting the 1U19 structure onto each monomer onto the 1N3M oligomer [89] (a) and onto the oligomeric model inferred from cryoEM (kindly provided by K. Palczewski) (b) [169]

Full size image

Successively, the crystal structure of Ops* (PDB: 3CAP [120]) revealed a divergent dimeric architecture characterized by H1-H1 and H8-H8 contacts, all compatible with AFM images [92]. Consistently, the recent cryoEM structures of a cross-linked rhodopsin dimer and of a rhodopsin dimer reconstituted into nanodiscs from purified monomers show a dimer interface mediated by H1-H1 and H8-H8 contacts (PDB: 6OFJ [169]). An oligomer was also inferred, made of rows of dimers held together by E2-I3 and H4-H6,H5 contacts (Fig. 6b). The authors proposed that the dimer interface and the arrangement of two protomers are necessary for the formation of the rows of dimers [169].

The more recent oligomeric model is consistent with the experimental evidence that the adRP rod opsin mutant F45L in H1 (position 1.40) is impaired in dimerization [123]. Indeed, in the cryoEM opsin oligomer, native F45 contributes to the H1-H1 contacts funding the dimer interface.

Although the role of rod opsin oligomerization awaits clarification, functional characterization of rhodopsin monomers and dimers in detergents demonstrated that monomeric rhodopsin is able to activate transducin, though the oligomeric form is more active [64, 65]. In vitro experiments in solution and in nanodiscs containing one and two rhodopsin molecules strengthened the evidence that the receptor monomer holds the structural determinants for transducin activation and, in this regard, it is the functional unit [10, 20, 21, 33].

Oligomerization may be essential for the ontogeny and/or desensitization of rod opsin and, hence, in the control of light signaling. Aggregation between WT rod opsin and some partially misfolded adRP RHO mutants may have implications in the etiology of the disease and on responsiveness to retinoid chaperones [50, 51]. Addressing these aspects may have important implications in unraveling the molecular determinants of retinal degenerative diseases.

Insights into retinal diseases from structure determinations and computational experiments

Rod cells of human CSNB patients [96, 145] and of animal models carrying the G90D mutation are functionally desensitized, as if they were under constant low basal stimulation that perturbs dim-light vision [32, 66, 144]. The determinants of CSNB linked to CAM opsin would concern (1) constitutive activation of transducin by G90D opsin [66, 128], (2) constitutive activation of opsin caused by thermal isomerization of retinal [145], and (3) constant basal activation by a pre-activated dark state [32].

The crystal structures of G90D (PDB: 4BEY and 4BEZ [147]) and T94I (PDB: 5DYS and 5EN0 [146]) CSNB RHO mutants, in their unliganded and all-trans-retinal-bound states, provided insights into the structural determinants of the disease. The structural models of the mutants are identical to those of WT apo and all-trans-retinal-bound opsin (Fig. 7 a–d). The structure of G90D is characterized by a salt bridge between D90(2.57) and K296(7.43), displacing the retinal counterion E113(3.28) (Fig. 7c). The D90-K296 salt bridge would both stabilize the active state and reduce binding of opsin to the desensitizing Arr1 [147]. Moreover, the interference of G90D with the site of retinal attachment would also increase the rate of thermal isomerization [147]. Several retinal isomers, including all-trans- and 11-cis-retinal, can transiently activate opsin until protein deactivation by SB formation [72, 80]. Perturbation of SB by CSNB mutations can lead to constitutive activation in the presence of non-covalently bound cis-retinal [147].

Fig. 7.
figure 7

Crystal structures of WT and CSNB mutant rod opsins. The crystal structures of WT Ops* (PDB: 3CAP [120]) (a), MII (PDB: 3PXO [24]) (b), D90D (PDB: 4BEZ [147]) (c), and T95I (PDB: 5DYS [146]) (d) are shown

Full size image

In T94I rod opsin mutant, the replacing I94 establishes direct van der Waals interactions with K296(7.43), thus prolonging the lifetime of the MII signaling active state [146]. Collectively, a common feature of G90D and T94I would be the ability of both replacing amino acids in H2 to interact with K296(7.43), thus altering the dark state by weakening the interaction between the SB and its counterion E113(3.28) [146, 147].

Retinal dystrophies due to adRP RHO mutations represent a challenge in developing therapeutic intervention also because many of those mutations cause different degree of misfolding of the opsin protein. The inherent instability of those mutants impedes high-resolution structure determination. So far, atomic-level investigations relied on structural bioinformatics and molecular simulations.

The protein design algorithm FoldX was challenged to estimate the effects of adRP mutations on the stability of native rhodopsin [127]. In spite of the general usefulness of the tool, neglecting structural changes in free energy calculations makes the approach inadequate to estimate the effects of adRP rod opsin mutations that are linked to protein misfolding.

Understanding the molecular bases of conformational diseases requires the characterization of the interacting forces that drive folding, stabilize the native state, and, once impaired, lead to misfolding. We employed a number of computational strategies to face this issue. Replica exchange molecular dynamics (REMD) simulations served to investigate the effects of 40 rhodopsin mutations located in E2 on the conformation of the β-hairpin [39]. All the adRP RHO mutants in E2 were included in the simulated set. The free energy landscape of β-hairpin formation was found to align with in vitro mutational effects. Indeed, marked effects on the stability of E2 were exhibited only by those mutations causing severe impairments in folding/expression of the opsin protein [39]. Further insights were inferred from computational experiments based on the combination of mechanical unfolding simulations and protein structure network (PSN) analysis, an ever increasingly used approach to investigate structural communication, folding, and stability [36, 37, 164]. The study suggested that adRP misfolding RHO mutations tend to weaken a number of highly connected amino acids in the proximity to the retinal binding site and the G protein-binding regions [37].

In a more recent study, molecular simulations (i.e., 300K and melting unfolding simulations) coupled to the graph-based PSN analysis were combined with in vitro subcellular localization analyses to infer the effects of 33 adRP RHO mutations on stability and transport of the protein in the absence and presence of the retinal ligand [12].

The working hypothesis was that misfolded adRP mutants represent “quasi-native” states of the opsin protein.

While in silico experiments on the rhodopsin state employed 11-cis-retinal, present in the selected 1GZM structure, in vitro experiments employed 9-cis-retinal. The ability of 9-cis- and 11-cis-retinal to promote proper folding and trafficking has been previously demonstrated for the adRP mutants P23H, T17M, and Q28H [76, 82, 88, 99, 109, 110]. The employment of 9-cis-retinal in our subcellular localization analyses was dictated by the fact that the 9-cis-isomer is more stable than the 11-cis-isomer and shows similar behaviors, thus making it a good substitute [76, 117].

Remarkably, for both the opsin and rhodopsin states, a linear correlation was obtained between an index of structure network perturbation (NP) and the Pearson correlation coefficient (PCC) of calnexin/rhodopsin co-localization accounting for ER retention (i.e., the higher the PCC, the higher the co-localization between rod opsin and the ER-resident protein calnexin is [12]). Such a linear correlation between indices of misfolding and mislocalization allowed inferring common structural defects held by the mutants in relation to their ER retention.

The NP index served to divide the adRP RHO mutants into four clusters. Mutant clustering based upon the NP index computed on the opsin and rhodopsin states and the ΔNP was in line with ER retention [12]. In synthesis, mutant clusters 1 and 4 show, respectively, the lowest and highest destabilization of the native structure network and ER retention independent of the presence of retinal (i.e., either 11-cis or 9-cis), whereas cluster 2 resembles cluster 4 in the absence of retinal and cluster 1 in the presence of retinal. In other words, for cluster 2, the presence of retinal hinders structural impairment and mislocalization by mutation. The structural and cellular behaviors of cluster-3 mutants lie in between those of clusters 2 and 4 [12]. In this respect, cluster-1 mutants are T4K, M44T, F45L, G51A, T58R, G89D, G106R, L125R, V137M, A164V, and C167R; cluster-2 mutants are T17M, Q28H, V87D, R135W, P171Q, E181K, D190Y, and P267L; cluster-3 mutants are P23H, Y178C, P180A, S186P, and D190G; and cluster-4 mutants are L46R, P53R, C110Y, G114D, C187Y, G188R, H211P, H211R, and C222R.

The correlation between NP and PCC revealed a bridge between the classification of mutants based on their structural perturbation and the classification based on their ER retention [12]. Earlier classifications of adRP RHO mutants essentially relied on mutational effects on ER retention and the ability of rod opsin to bind retinal (reviewed in [18, 82]. Our in vitro data on mutant opsin agreed substantially with previous data on ER retention (reviewed in [18, 82]). Apparent discrepancies regard two mutants (L125R and C167R) assigned to classes IIb/II [18, 82, 151] and IIa/II [17, 18, 82, 154], respectively, but localized at the plasma membrane in our experiments and one mutant (L46R) assigned to class I [18] but retained in the ER in our experiments. As for L125R and C167R, molecular simulations showed that the replacing arginine makes a salt bridge with E122, which, consistent with poor ER retention and good plasma membrane targeting, does not perturb the native network in the opsin state. Simulations of C167R and L125R rhodopsin, however, showed that the replacing arginine, especially in L125R, induced deformations in the retinal ligand [12]. This may suggest that mutant opsins are able to reach plasma membrane even if impaired in retinal binding, consistent with L125R opsin showing WT-like electrophoretic pattern [47]. Thus for cluster-1 mutants, which reach plasma membrane and show low ER retention either in the opsin or rhodopsin states, the chaperone effect of the chromophore, if any, is masked. Therefore, our subcellular localization analysis could not distinguish cluster-1 mutants able to bind retinal from cluster-1 mutants unable to do so.

More discrepancies concern mutant responses to retinal, possibly due to divergences in the experimental approaches. Whereas previous studies employed spectroscopic analyses to evaluate rod opsin regeneration with retinal, we evaluated the ability of retinal to improve the subcellular localization of those mutants mostly retained into ER and not targeted to the plasma membrane in their opsin state. In our study, five (P171Q, Y178C, E181K, S186P, and D190Y) out of eleven (C110Y, G114D, P171Q, Y178C, E181K, S186P, C187Y, D190Y, H211P, H211R, and C222R) mutants, previously found unable to reconstitute with retinal (i.e., falling in class IIa/II [2, 17, 18, 60, 62, 69, 71, 82, 131, 154, 166]), could instead improve their plasma membrane targeting and relieve ER retention as an effect of retinal binding.

By considering the adRP RHO mutant grouping into seven classes by Athanasiou and co-workers, while class 2 comprises possibly misfolded mutants because of their ER retention and impairment in retinal binding, the other classes comprise mutants thought to be properly folded but impaired in different aspects of rod opsin function [5]. Consistently, cluster-2, 3, and 4 mutants from our study, which are retained in the ER at least in their opsin form, fall into class 2. Consistency was also found for a number of mutants in our clusters 1 and 2, which fall in classes 3 (R135W [25]), 4 (T4K and T17M [22, 134, 157]), 5 (M44T and V137M [2]), and 7 (F45L [123]) by Athanasiou et al. and include also a mutant considered as a benign single-mutant polymorphism (G51A) [5]. Indeed, for those mutants showing membrane targeting at least in the presence of retinal, defects other than ER retention may cause the pathologic phenotype. In particular, (a) R135W is hyperphosphorylated and bound with high affinity to Arr1 [25]; (b) T4K and T17M, located in structured portions of the Nt, by affecting backbone conformation, may alter N-linked glycosylation [22, 134, 157]; (c) M44T and V137M were found to increase initial activation rates of transducin, though not bearing any constitutive activity [2]; and (d) F45L is expected to impair rod opsin dimerization [123]. Six mutants from our cluster 1 (T58R, G89D, G106R, L125R, A164V, and C167R) were ascribed to class-2 retinal non-responsive mutants by Athanasiou and co-workers [5, 17, 60, 130, 151, 154, 155]. In line with the statements above, cluster-1 mutants are poorly retained into ER and reach plasma membrane independent of retinal presence. Therefore, those mutants may be not retained into ER, consistent with our data, but impaired in retinal binding, consistent with mutant classification by Athanasiou and co-workers [5].

Early mass spectrometry determinations on four adRP mutants indicated partial (for G89D and A164V) or complete (for L125R and H211P) misfolding of rod opsin due to the formation of a non-native disulfide bridge between C185 and C187 in E2 [61]. Three of those mutants fall in our cluster 1, suggesting that, if present, such a non-native disulfide bridge would be compatible with a protein state targetable to the plasma membrane even in the absence of retinal.

Collectively, our investigation revealed some retinal responsiveness in 13 out of 22 adRP rod opsin mutants found retained into ER and not targeted to the plasma membrane in their opsin form. Those mutants may be target candidates of small chaperones recognizing the orthosteric retinal binding site. Remarkably, some cluster-1 mutants while not mislocalized may not respond to the retinal.

Effective structural signatures of each cluster, as inferred from the structure networks, were the native stable links that undergo an average frequency reduction ≥ 25% in the mutant trajectories belonging to the cluster (i.e., ≥ 40% frequency reduction in the perturbed networks shown in Fig. 8 a and b). By focusing on the intradiscal half of the receptor, which hosts the retinal binding site, it was observed that, in the opsin state, the lowest perturbation of native stable links concerns cluster 1 (Fig. 8b). Noticeably, 11-cis-retinal acts as a chaperone by preventing major link weakening shared in the opsin state by the mutants of cluster 1 and, more prominently, of clusters 2 and 3. In contrast, the ligand is less effective as a chaperone for clusters-4 mutants (Fig. 8b) [12]. Collectively, all mutants show marked link destabilization in the opsin state as an effect of mutation on protein structure and dynamics. Major structural differences among the four clusters concerned destabilization of native links in the intradiscal half of the receptor and in the ability of 11-cis-retinal to impede such perturbations, thus acting as a small chaperone. The structural effects of 11-cis-retinal were maximal for clusters 2 and 3 but minimal for cluster 4. The significant perturbation in the retinal binding site suggested that cluster-4 mutants are deficient in retinal binding in line with the inability of the retinal to relieve ER retention of mutant opsins.

Fig. 8
figure 8

Structural signatures of the four clusters of adRP mutants. a The mutation sites of the four clusters of mutants (clusters 1, 2, 3, and 4 are, respectively, light green, marine, gray, and red) are shown as spheres centered on the Cα-atoms. As already stated in the main text, cluster-1 mutants are T4K, M44T, F45L, G51A, T58R, G89D, G106R, L125R, V137M, A164V, and C167R; cluster-2 mutants are T17M, Q28H, V87D, R135W, P171Q, E181K, D190Y, and P267L; cluster-3 mutants are P23H, Y178C, P180A, S186P, and D190G; and cluster-4 mutants are L46R, P53R, C110Y, G114D, C187Y, G188R, H211P, H211R, and C222R

Full size image

The stability core of rod opsin inferred from the PSN analysis was targeted by virtual screening of over 300,000 anionic compounds, which led to the discovery of a novel chaperone, 5,8-epoxy-13-cis retinoic acid (13-cis-5,8-ERA), able to bind bovine opsin with 14-fold better EC50 (8.6 ± 0.2 nM) than 9-cis-retinal (123.9 ± 7.0 nM). Acting as a reversible orthosteric inhibitor of retinal binding, the compound proved more effective than 9-cis-retinal in promoting membrane localization of three adRP RHO mutants from the retinal-responsive clusters 2 and 3, i.e., T17M, P23H, and E181K [12]. In the same period, other two studies reported on the discovery of small chaperones able to bind bovine rod opsin and to promote membrane localization of P23H opsin though with EC50 comparable or worse than that of 9-cis-retinal [23, 95].

A recent extension of mechanical unfolding simulations coupled to PSN analysis to the same set of 33 adRP RHO mutants allowed us to develop a structure-based automated approach able to predict ER retention of novel adRP mutants and the ability of small chaperones to mitigate such ER retention with therapeutic implications (Felline et al. under review). In this framework, the average distributions of stable native hubs (i.e., nodes with at least four links) in the four clusters of mutants as a function of hub-link weakening and amino acid conservation provided valuable fingerprints of rhodopsin misfolding by mutation. Such fingerprints could distinguish mutants that do not respond to small chaperones from mutants either non-misfolded or responsive to small chaperones.

All these studies did not consider the tendency of partially misfolded mutants to aggregate with the WT as a possible determinant of the disease.

Collectively, thermal and mechanical unfolding simulations converge into the inference that misfolding rod opsin mutants weaken the stability core of the protein, which is participated by amino acids in the retinal binding site. The effects of such perturbations on the secondary structure elements and their spatial arrangements still remain undefined.

Conclusions and perspectives

Vertebrate rod opsin is the GPCR best characterized at the atomic level of detail. Since the release of the first crystal structure 20 years ago, a huge number of structures have been released that, in combination with valuable spectroscopic determinations, unveiled most aspects of the photobleaching process.

Retinal dystrophies due to adRP RHO mutations still represent a challenge because the inherent instability of such mutants so far impeded high-resolution structure determinations.

The existent structures of WT rod opsin in its apo and retinal-bound forms were exploited as input structures for unfolding simulations and analysis of adRP RHO mutants with valuable hints. Those simulations, indeed, converged into the inference that misfolding rod opsin mutations weaken the stability core of the protein, which is participated by amino acids in the retinal binding site. The effects of such perturbations on the secondary structure elements and their architecture still remain undefined. The ability of some misfolded mutants to aggregate with the WT as a possible determinant of the disease awaits elucidation as well. The combination of in silico and in vivo experiments suggest that at least all mutants in clusters 1, 2, and 3 are in quasi-native states and that mutants from clusters 2 and 3 are recovered by small chaperones like 11-cis-retinal and 13-cis-5,8-ERA.

As already shown, a computational approach based on the combination of misfolding simulations of rod opsin mutants with PSN analysis and virtual screening of compounds targeting the stability core of the protein may lead to the discovery of small chaperones, even better than the one discovered so far. This would have important implications in personalized therapeutic interventions for adRP linked to rod opsin mutations. Actually, our efforts are going in that direction.