Comparison of the isomerization mechanisms of humanmelanopsin and invertebrate andvertebrate rhodopsinsSilvia Rinaldia,b, Federico Melaccioa, Samer Gozemb, Francesca Fanellic, and Massimo Olivuccia,b,1

aDipartimento di Biotecnologie, Chimica e Farmacia, Università di Siena, I-53100 Siena, Italy; bChemistry Department, Bowling Green State University, BowlingGreen, OH 43403; and cDulbecco Telethon Institute, Department of Life Sciences, University of Modena and Reggio Emilia, I-41125 Modena, Italy

Edited by Arieh Warshel, University of Southern California, Los Angeles, CA, and approved December 18, 2013 (received for review June 13, 2013)

Comparative modeling and ab initio multiconfigurational quan-tum chemistry are combined to investigate the reactivity of thehuman nonvisual photoreceptor melanopsin. It is found that boththe thermal and photochemical isomerization of the melanopsin11-cis retinal chromophore occur via a space-saving mechanisminvolving the unidirectional, counterclockwise twisting of the=C11H-C12H= moiety with respect to its Lys340-linked frame as pro-posed byWarshel for visual pigments [Warshel A (1976) Nature 260(5553):679–683]. A comparison with the mechanisms documentedfor vertebrate (bovine) and invertebrate (squid) visual photorecep-tors shows that such a mechanism is not affected by the diversityof the three chromophore cavities. Despite such invariance, trajec-tory computations indicate that although all receptors display lessthan 100 fs excited state dynamics, human melanopsin decays fromthe excited state∼40 fs earlier than bovine rhodopsin. Some diversityis also found in the energy barriers controlling thermal isomerization.Human melanopsin features the highest computed barrier whichappears to be ∼2.5 kcal mol−1 higher than that of bovine rhodopsin.When assuming the validity of both the reaction speed/quantumyield correlation discussed byWarshel, Mathies and coworkers [WeissRM, Warshel A (1979) J Am Chem Soc 101:6131–6133; Schoenlein RW,Peteanu LA, Mathies RA, Shank CV (1991) Science 254(5030):412–415]and of a relationship between thermal isomerization rate and ther-mal activation of the photocycle, melanopsin turns out to be a highlysensitive pigment consistent with the low number of melanopsin-containing cells found in the retina and with the extraretina locationof melanopsin in nonmammalian vertebrates.

ultrafast isomerization | thermal noise in photoreceptors |conical intersection | QM/MM method | computational photobiology

For a long time it was assumed that the human retina containsonly two types of photoreceptor cells: the rods and cones

responsible for dim-light and daylight vision, respectively.However, recent studies have revealed the existence of a smallnumber of intrinsically photosensitive retinal ganglion cells(ipRGCs) that regulate nonvisual photoresponses (1). ipRGCsexpress an atypical opsin-like protein named melanopsin (2, 3)which plays a role in the regulation of unconscious visual reflexesand in the synchronization of endogenous physiological responsesto the dawn/dusk cycle (circadian rhythms) (4, 5).Melanopsins are unique among vertebrate photoreceptors be-

cause their amino acid sequence shares greater similarity to in-vertebrate than vertebrate rhodopsin (i.e., the photoreceptor ofrods) (6, 7). Like rhodopsins, melanopsins feature an up–downbundle architecture of seven transmembrane α-helices incorpo-rating the 11-cis isomer of retinal as a covalently bound pro-tonated Schiff base (PSB11 in Fig. 1A). Light-induced (i.e.,photochemical) isomerization of PSB11 to its all-trans isomer(PSBAT) triggers an opsin conformational change that, ultimately,activates the receptor and signaling cascade (8, 9). However,similar to invertebrate and in contrast to vertebrate rhodopsins,melanopsins are bistable (10). Indeed, although vertebrate rhodop-sins need a retinoid cycle (11) to regenerate PSB11, melanopsins

have an intrinsic light-driven chromophore regeneration functionvia PSBAT back-isomerization. Furthermore, past studies haveshown that melanopsins use an invertebrate-like signal trans-duction cascade (12).Melanopsins are held responsible for photoentrainment, using

the changes of irradiance and spectral composition to adjust thecircadian rhythm (13). The different studies carried out so far onmelanopsin light sensitivity do not lead to consistent results.Although Do et al. (14) argue that ipRGCs work at extremelylow irradiation intensities showing a single-photon response largerthan rods, Ferrer et al. (15) conclude that the melanopsin hasa reduced sensitivity relative to visual pigments. On the otherhand, these photoreceptors would be expected to display highlight sensitivity (14). In the vertebrate retina their density is 104

times lower than that of rhodopsins. Moreover, the receptor isnot confined in a dedicated cellular domain such as the outersegment of rods and cones, resulting in a ipRGCs photon capturemore than 106-fold lower than that of rods and cones per unit ofretina illumination. A high sensitivity of melanopsins would alsobe consistent with their presence in extraretina locations such asin pineal complex, deep brain, and derma of nonmammalian ver-tebrates (e.g., amphibian) (16–18). The amount of light that canpenetrate into such regions is limited and enriched in the redcomponent due to light scattering by the surrounding tissues (14).The molecular-level understanding of the primary light re-

sponse of melanopsin is a prerequisite for the comprehensionof more complex properties such as its activation and sensitivity.Despite numerous studies carried out since its discovery (16),

Significance

In this work, ab initio multiconfigurational quantum chemicalcomputations are used to understand how light sensitivity iscontrolled in biological photoreceptors. The authors focus onmelanopsin: the recently discovered nonvisual retina pigmentresponsible for the regulation of unconscious visual reflexesand the synchronization of endogenous physiological responsesto the dawn/dusk cycle (circadian rhythms). The investigationindicates that the light-induced isomerization of melanopsin isup to several tens of femtoseconds faster than the analogueisomerization of invertebrate and vertebrate visual pigments. Itis also revealed that its thermal isomerization is controlled by anenergy barrier higher than the barrier of dim-light visual pig-ments. These properties support the hypothesis of an extremelight sensitivity of melanopsins.

Author contributions: M.O. designed research; S.R., F.M., S.G., and F.F. performed re-search; S.R., F.M., S.G., F.F., and M.O. analyzed data; and S.R., F.M., S.G., and M.O. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. E-mail: molivuc@bgsu.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1309508111/-/DCSupplemental.

1714–1719 | PNAS | February 4, 2014 | vol. 111 | no. 5 www.pnas.org/cgi/doi/10.1073/pnas.1309508111

Dow

nloa

ded

by g

uest

on

Aug

ust 2

4, 2

020

there is presently little information on the molecular mechanismof melanopsin activation. The common PSB11 chromophore ofmelanopsins and rhodopsins does not guarantee that the samemechanism operates in both photoreceptors. This not only con-cerns light-induced activation but also thermal activation: a pro-cess whose rate limits the photoreceptor light sensitivity and thatis currently associated with thermal, rather than photochemical,PSB11 isomerization (19–24).The mechanism of light-induced PSB11 isomerization in ver-

tebrate rhodopsins has been extensively investigated. Spectro-scopic studies have shown that in bovine rhodopsin (Rh) theisomerization occurs on a subpicosecond timescale (25–27). More-over, the observation of ground state (S0) vibrational coherence (28)is consistent with a direct transfer of the excited state (S1) pop-ulation to the photoproduct (Fig. 1B) passing through a conicalintersection (CI). Such a path has been located along the S1 po-tential energy surface by constructing a multiconfigurationalquantum chemistry (MCQC) based computer model of thephotoreceptor (29–31) and spectroscopically supported by prob-ing in the infrared (31). More recently (32), the same computermodel has been used to map the Rh thermal isomerization path(Fig. 1B) providing information on the transition states control-ling the reaction.Here we present a computational study focusing on the mecha-

nism of photochemical and thermal isomerization of human mel-anopsin (hMeOp). This would require the construction ofa computer model of hMeOp starting from the receptor crystalstructure. However, the lack of hMeOp crystallographic datadoes not allow the use of the protocol previously applied in Rhstudies. The significant sequence similarity between squid rho-dopsin (sqRh), whose crystal structure is available (PDB code:2Z73) (33), and hMeOp (40%, SI Appendix, Fig. S1) provides thefundamentals for constructing a structural model of hMeOp ata significant atomic resolution. Building on a study by Batista andcoworkers (34) on murine melanopsin, we combine comparative

modeling of hMeOp with MCQC to construct a quantum me-chanics/molecular mechanics (QM/MM) computer model capa-ble of simulating the photochemical and thermal isomerizationreactions of hMeOp. The results are then compared with thosefound using Rh and sqRh models constructed using the sameprotocol. Such a comparison is expected to provide informationon the differences in spectral and functional properties of theseevolutionary distant pigments. As we will show below, the modelsindicate that hMeOp has a faster photochemical isomerizationdynamics and a higher thermal isomerization barrier than bothRh and sqRh.

Results and DiscussionAbsorption Maxima. As reported in Table 1, the ground stateequilibrium hMeOp, sqRh, and Rh models, displayed in Fig. 2Aand constructed using the common protocol described in theMethods section, reproduce the vertical excitation energies(ΔES1-S0) associated with the observed absorption maxima (λmax)(35–38) within a few kcal mol−1. However, although the observedλmax for mammalian melanopsins fall in the 470–480-nm range,the computed ΔES1-S0 values for the hMeOp model and fora murine melanopsin model (34) do not fall within the corre-sponding ∼2 kcal mol−1 range. This is attributed to the differentprotocols used to build the corresponding QM/MM models (SIAppendix provides details). To minimize the dependence of ourresults on methodological details, below we focus on the geo-metrical and ΔES1-S0 changes displayed by our three consistentmodels. These changes are qualitatively stable when samplingthe chromophore cavity configuration (Methods).Comparison of the PSB11 dihedral angles of hMeOp with the

corresponding sqRh and Rh quantities (Fig. 2A) shows that thehMeOp chromophore has an enhanced torsional distortion. In-deed, in such a structure the C8-C9, C10-C11, and C12-C13single bonds are all more than −10° twisted consistently witha backbone featuring a strong counterclockwise helicity. Becausesingle-bond twisting breaks the π-conjugation and increases theS1-S0 energy gap, we have evaluated the effect of the torsionaldistortion on ΔES1-S0 along the hMeOp, sqRh, and Rh series bycomputing the ΔES1-S0 of the isolated PSB11 chromophoresextracted from their equilibrium opsin structures. From Table 1it is apparent that, indeed, hMeOp displays a 3 kcal mol−1 largerΔES1-S0 relative to Rh, but just 1 kcal mol−1 larger comparedwith the sqRh homolog.In our models the electrostatic effect of the protein is responsible

for a 7–8 kcal mol−1 increased ΔES1-S0 with respect to the isolatedchromophores. To disentangle the factors responsible for suchchange, the qualitative effect of each cavity residue [defined asthe residues with at least one atom within 4.0 Å from any atom ofPSB11 (Fig. 2 B–D)] has been studied. The hMeOp cavity con-tains 21 residues (excluding the retinal-bound lysine) and twowaters. Among these, there are 14 apolar residues mostly locatedin the vicinity to the β-ionone ring (Fig. 2B). The putative E215counterion is located 6.1 Å away from PSB11. Notice that suchresidue is not homologous to the Rh E113 counterion but toE181 (39). The effect of each residue is evaluated by recomputingthe ΔES1-S0 after setting to zero its charges. SI Appendix, Fig. S6,shows that in hMeOp almost all sizable effects induce an increase

11

13

12

109

87

65

14 109

1113

1214

Rh

TS

Ene

rgy

CI

photocycle

bathoRhEaT

E(S1-S0)

cis trans

A

B

15

Fig. 1. PSB11 chromophore reactivity. (A) Chromophore structure andisomerization to PSBAT. (B) Schematic representation of the photochemical(full arrows) and thermal (dashed arrows) isomerization paths. The CI is lo-cated energetically above the TS, features a different geometrical structure,and drives a far-from-equilibrium process. ΔES1-S0, τcis→trans, and EaT (in red)are the fundamental quantities computed in the present work.

Table 1. Computed and experimental vertical excitation energies (ΔES1-S0) and absorption maxima (λmax) values

Model, kcal mol−1 (nm)Oscillator strength,

S0→S1 Isolated PSB11, kcal mol−1 (nm) Observed, kcal mol−1 (nm)

hMeOp 61.9 (462) 0.9 53.6 (533) 59.6–61.2 (467–480)sqRh 60.0 (476) 0.9 52.6 (543) 58.5 (489)Rh 57.3 (499) 0.9 50.5 (566) 57.4 (498)

The λmax values are given in brackets. hMeOp, human melanopsin; sqRh, squid rhodopsin; Rh, bovine rhodopsin.

Rinaldi et al. PNAS | February 4, 2014 | vol. 111 | no. 5 | 1715

CHEM

ISTR

YBIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Aug

ust 2

4, 2

020

in ΔES1-S0 (a blue shift in λmax), and the total contribution of thecavity is an increase of 12 kcal mol−1. Despite its large distancefrom the Schiff base linkage, the largest effect comes from E215which is responsible for a ∼5 kcal mol−1 increase. As shown inFig. 2B, other sizable contributions of the same sign come fromQ123 and the WAT367 and WAT368 waters. S220 has instead anopposite effect.The computed ΔES1-S0 changes are explained by considering

that upon vertical excitation, ∼30% of the positive charge lo-calized in the C=N region of the backbone is translocated towardthe β-ionone ring (40). Therefore, residues creating a negative(positive) electrostatic potential in the Schiff base region wouldstabilize (destabilize) the S0 state with respect to S1. On the otherhand, residues creating a negative (positive) potential on theβ-ionone would stabilize (destabilize) S1 with respect to S0. ForhMeOp the blue-shifting effect appears to be mainly due toa cluster formed by relatively close residues. These constitutea sort of “distributed” counterion including the distant ionizedresidue E215 imposing a negative potential on the Schiff base.As shown in SI Appendix, Fig. S6, only a few cavity residues leadto a decrease (a red shift) of the ΔES1-S0 which would contrastthe counterion effect.The sqRh cavity contains 23 amino acids (Fig. 2C). Similar to

hMeOp, the 15 apolar residues are mostly around the β-iononering. The largest effect on ΔES1-S0 is computed for the putativeE172 counterion (again, the homolog of E181 in Rh) whosecharge increases ΔES1-S0 by ∼9 kcal mol−1. As shown in Fig. 2C,sizable effects are also attributed to N79, Y103, and S179. Althoughthe sqRh cavity generates a blue shift similar to that of hMeOpthe counterion structure shows some differences. In fact, sizableblue-shifting effects are now more limited, the cavity waters havelittle contributions and, as reported in SI Appendix, Fig. S6, morered-shifting residues contrast the counterion effect. Further-more, the ionized E172 residue is closer to PSB11 leading to analmost doubled ΔES1-S0 change with respect to E215 in hMeOp.This points to a more localized counterion structure.Finally, Rh contains 26 amino acids and two waters in its

cavity. In the study we also included the E181, the homologue ofthe putative counterions of hMeOp and sqRh (Fig. 2D). As alsoobserved in hMeOp and sqRh, nonpolar amino acids are mainlylocated around the β-ionone ring. In line with previous results

(40–42), the Rh model shows that the largest electrostatic effectis due to the putative E113 counterion (SI Appendix, Fig. S6). Infact, the E113 carboxylate is much closer to the Schiff base withrespect to E172 and E215 carboxylates of sqRh and hMeOp,respectively, thus leading to a larger ΔES1-S0 increase. The othersizable contributions come from the residues T118, S186, E181,W265, and the two waters but are all positive. Globally, the 27residues in the cavity generate a blue shift of 10 kcal mol−1,which results from large blue-shifting E113 effect contrasted bya cluster of residues creating a positive potential on the Schiffbase region. This is consistent with the idea of a substantially“localized” counterion structure in Rh whose effect is partiallyquenched by the remaining opsin residues as also reported inprevious studies using similar models (29, 40, 42).

Photochemical Isomerization. The mechanism of light-inducedisomerization has been investigated by running single S1 trajec-tory computations starting from the S0 equilibrium structuresdescribed above (without initial velocities) (30, 43). These simu-lations were stopped whenever a near-degeneracy region (ΔES1-S0less than few kcal mol−1) was approached and associated to aconical intersection (CI) between S1 and S0. Given the ultrashorttimescale spanned by the trajectory, it is assumed that it repre-sents the average behavior of the S1 population as previouslyassessed by comparison with trajectory sets (30, 31). Such anassumption has been tested by computing, for each model, eighttrajectories featuring different initial configurations as well as bylooking at the evolution of 100 trajectories for a reduced chro-mophore model (Methods and SI Appendix).The results reported in Fig. 3 show that hMeOp, sqRh, and Rh

all decay within 100 fs. However, the nonvisual pigment hMeOpis the fastest, reaching a CI point ∼40 fs earlier than the verte-brate pigment Rh which reaches the CI only after 90 fs consis-tently with previous results (30, 31, 43). Despite these differ-ences, as well as the differences in structure described above, wefound that the isomerization mechanism is substantially thesame. As detailed in SI Appendix, the isomerization begins witha large double-bond/single-bond inversion and is followed bymultimode torsional deformations dominated by the skeletalC11=C12 (the reactive double bond) and the C9=C10 (the ad-jacent double bond) torsional coordinates. These changes arealso documented by comparing the values of the corresponding

B

C

A

D

E113 (17.6)

E181 (-2.2)

T118 (-1.5)

S186 (-1.3)

W265 (-1.0)

WAT355 (-2.1)

WAT359 (-1.1)

2.8 Å5.5 Å

6.1 Å

6.8 Å

E215 (5.4)

S220 (-1.7)

Q123 (1.2)

WAT367 (3.3)

WAT368 (1.8)

5.0 Å

6.0 Å

E172 (8.5)

Y103 (1.0)

N79 (2.8)

S179 (-1.3)

-6.2-9.6

-10.4

175.8

170.6-180.0

178.3175.7

175.0

-55.5

-61.7-61.4

172.3178.4

-178.5169.4

169.2163.1

157.2155.5

164.2-174.0

167.1179.0

1.29

1.43

1.29

1.421.43

1.29

1.37

1.47

1.36

1.47

1.36

1.36

1.361.36

1.47 Fig. 2. Computed ground state equilibrium structureand cavity orientation of PSB11. (A) Geometries ofPSB11 in hMeOp (blue), sqRh (red), and Rh (green).The relevant bond lengths and backbone dihedralangles are given in Å and degrees, respectively. Curlyarrows indicate alternating counterclockwise and clock-wise isomerizations about the C9=C10 and C11=C12retinal double bonds. (B–D) Retinal-binding pocket ofhMeOp (B), sqRh (C) and Rh (D). Conventional apolarand polar residues are reported in gray and cyan,respectively. The chromophore, putative counterionsand E181 residue of Rh are shown in tube represen-tation. The cavity residues inducing a ΔES1-S0 changelarger than 1 kcal mol−1 or smaller than −1 kcal mol−1

are marked, on the β-carbon, with blue and red spheres,respectively. The corresponding change value is given inparenthesis in kcal mol−1. Notice the increase in coun-terion localization and parallel distance decrease withrespect to the Schiff base when going from hMeOp tosqRh to Rh. This trend is accompanied by an increasein counterbalancing red-shifting residues (SI Appendix,Fig. S6, provides further information).

1716 | www.pnas.org/cgi/doi/10.1073/pnas.1309508111 Rinaldi et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 2

4, 2

020

dihedral angles at the S0 equilibrium and CI structures in Figs.2A and 3B. All three CIs display a 82°–88° twisted C11=C12bond, a 27°–38° twisted C9=C10 bond, and an almost planar1°–7° C10–C11 bond. This motion is consistent with a counter-clockwise rotation of the -C10H-C11H- plane with respect to therest of the backbone. As previously discussed for the case of Rh(30, 31, 44), the resulting reaction coordinate is consistent witha variant of the so-called bicycle-pedal mechanism originallyproposed by Warshel (45) and here extended to nonvisual andinvertebrate pigments.The sensitivity of the isomerization to the distributed and

distant or localized and close counterion structure has beenprobed via trajectory computations (SI Appendix providesdetails). The hMeOp and Rh trajectories have been recomputedafter setting to zero the charges of the corresponding putativeE215 and E113 counterions, respectively. It is found (SI Ap-pendix, Fig. S11) that the E215 charge only slightly affects theisomerization timescale of hMeOp. However, the removal of theE113 charges blocks the Rh isomerization. Further analysis showsthat removal of the charges of the set of residues forming thedistributed hMeOp counterion significantly increases the modelisomerization timescale. Also, adding the charges of a glutamate tothe neutral residue E181 after removal of the E113 charges in Rh(to mimic the putative hMeOp counterion position) restores thereactivity. Overall, these effects show that the PSB11 S1 dynamicsis very sensitive to the details of the counterion configurations.

Thermal Isomerization. Recently, we have reported on the geo-metrical and electronic structure of the transition states control-ling the thermal isomerization in Rh (32). Two transition stateshave been located featuring a charge transfer (TSCT) and a

diradical (TSDIR) electronic structure and characterized by dif-ferent bond length alternation (BLA) values (this is the differ-ence between the average length of single and double bonds asassigned in Fig. 1A), respectively. Although these transitionstates have been optimized in a series of Rh mutants, they havenever been computed for other opsins. Accordingly, below wereport the TSCT and TSDIR structures for hMeOp and sqRh.In Fig. 4A we compare the TSCT and TSDIR structures for Rh,

sqRh, and hMeOp. It is evident that the six transition states displaya 90° through 92° twisted C11-C12 bond, a 22° through 28° twistedC9=C10 bond, and an almost planar 0° through 8° C10-C11 bond.Again, this motion is consistent with a torsional, counterclockwiserotation of the -C10H-C11H- plane with respect to the rest of thebackbone, further extending the validity of Warshel’s (45) mech-anism to thermal isomerizations. Although TSCT and TSDIR havevery similar torsional deformations, they mediate electronicallydifferent processes. TSCT mediates a heterolytic C11=C12 breakingconsistent with a neutral closed-shell -C12-C13-C14-C15-NH-moiety, an inverted BLA pattern (compare Figs. 2A and 4A), anda charge translocation with respect to the corresponding equilib-rium structures. In contrast, TSDIR mediates a homolytic breakingconsistent with a -C12-C13-C14-C15-NH- radical-cation moietyand a BLA closer to the reactant (also SI Appendix, Fig. S4).The energy barriers (EaT) associated with TSCT and TSDIR

are reported in Fig. 4B as a function of 1/λmax for Rh, sqRh, andhMeOp and indicate that TSCT controls the isomerization in allthree cases. It is also apparent that the TSCT EaT increases whenthe λmax is decreasing. The results are consistent with the Barlow(46, 47) correlation which establishes an inverse proportionality

hMeOpsqRhRh

A

-86.4

-172.4-178.6

-173.3-87.9

-179.0177.1

173.2

-152.6-152.0-142.4

-166.8-162.0

-167.7

-82.0

B time (fs)

Fig. 3. Photochemical isomerization. (A) Energy profiles along scaled-CASSCF/Amber trajectories (see Methods and SI Appendix) of hMeOp (bluesquare), sqRh (red triangle), and Rh (green circle). S1 and S0 energies areassociated with full and empty symbols, respectively. The gray full symbolsrepresent the CASPT2 energies calculated along the trajectory which furthersupports the timescales evaluated at the scaled-CASSCF level (30). The shortdark and light gray arrows indicate the conical intersection points for thecomputed CASSCF and CASPT2 energies, respectively. The time required toreach the conical intersection represents the “reaction timescale” τcis→trans.(B) Conical intersection geometries of retinal chromophore of hMeOp(blue), sqRh (red), and Rh (green). The main computed dihedral angles arereported in degrees.

-90.6-90.7

-153.1-154.1-153.5

171.6

-173.8

175.5

-174.9

-179.7

-160.5-178.8

-179.3177.3

-90.3

1.371.381.36

1.351.351.36

1.451.461.46

1.361.361.36

1.47

1.471.47

A

-92.3-91.4

-157.6-151.6-155.5

178.1 -178.8

179.2 -172.1

-176.3-169.9

179.6

179.7178.7

-92.4

1.29

1.42

1.30

1.411.42

1.29

1.40

1.39

1.40

1.391.39

1.41

1.471.481.47

B

Fig. 4. Thermal isomerization. (A) PSB11 geometries for the transitionstates mediating the thermal isomerization of hMeOp (blue), sqRh (red),and Rh (green). The relevant dihedral angles are reported in degrees. (B)CASPT2//CASSCF/Amber values of the energy barriers of the same transitionstates plotted as a function of the inverse of the corresponding absorptionmaxima. The dashed line corresponds to the linear fitting of the barriers(open circles) from the bovine rhodopsin mutants (with the A1 chromophore)computed in ref. 32.

Rinaldi et al. PNAS | February 4, 2014 | vol. 111 | no. 5 | 1717

CHEM

ISTR

YBIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Aug

ust 2

4, 2

020

between thermal activation kinetic constants (measuring thermalnoise) and λmax for vertebrate pigments. Recently, this pro-portionality has been investigated by constructing a set of QM/MMmodels for bovine Rh variants (mutants and derivatives wherethe A2 chromophore featuring an extended π-conjugation replacesthe Rh native A1 (i.e., PSB11) chromophore) (32).In Fig. 4B we show that the computed EaT vs. 1/λmax slopes for

the TSCT and TSDIR of our Rh, sqRh, and hMeOp models aresimilar to those previously reported for the Rh variants (32) thusextending the Barlow correlation to sqRh and hMeOp. Notice,however, that hMeOp represents a borderline case where theTSCT and TSDIR have close EaT values and where the TSCT andTSDIR correlation lines would cross for higher 1/λmax values.(Fig. 4B) Thus, our models indicate that mutations extending theabsorption of hMeOp further to the blue are not expected tolead to a decrease of thermal isomerization rate, as EaT willdecrease (32) after TSDIR starts to control the isomerization.

Light Sensitivity in Melanopsin. The light-induced isomerizationquantum yield and the thermal isomerization rate are basic factorsdetermining the level of light sensitivity of opsin photoreceptors.To enhance sensitivity, quantum yields must be maximized,whereas thermal rates, which create the “background noise” in thesignal, must be minimized. Although Rh quantum yield compu-tations have been reported (48), their significance is limited bythe affordable numbers of semiclassical trajectories and, mostimportantly, by the cost of accurate MCQC energy gradients (49).To avoid these limitations, here we discuss the relative quantumyields of hMeOp, sqRh, and Rh using the correlation proposedby Weiss and Warshel (50) and Mathies and coworkers (25) whoalso provided supporting experimental evidence (51).The idea is that a large fraction of the S1 population following

the S1 isomerization path would decay to S0 via or in the vicinityof a CI. If this is the case, the larger the velocity of the S1 re-actant moving toward the CI, the higher is the reaction quantumyield consistently with a Landau–Zener model (52). Althoughrecent studies have denied the existence of such a correlation forthe retinal chromophore in solution (53, 54), this does not ex-clude that such correlation holds in the protein environment andwhen the S1 lifetime is below a 100 fs threshold. In fact, this is inline with the more complete semiclassical treatment introducedby Weiss and Warshel (50) whose validity is also supported bythe sudden change in charge distribution seen at decay in ourmodels (SI Appendix, Fig. S8). Thus, the change in quantum yieldshould correlate with the inverse of the S1 lifetime which can beestimated by computing the τcis→trans time required to reach theCI (Fig. 1B). On the other hand, the trend in thermal isomeri-zation rate may be estimated more directly by assuming that thechange in rate will parallel the change in the corresponding po-tential energy barrier. This can be computed as the energy dif-ference EaT between the pigment lowest-lying transition state andits S0 equilibrium structure (Fig. 1B).Both τcis→trans and EaT trends displayed in Figs. 3A and 4B

point to hMeOp as the highest sensitive pigment, followed bysqRh and then Rh. In fact, our trajectories indicate that the S1dynamics of hMeOp is faster than that of Rh whereas its thermalisomerization barrier is 2.5 kcal mol−1 higher with respect to Rh.This difference in sensitivity is not paralleled by a change inreaction mechanism which remains substantially invariant forboth the light-induced isomerization and the thermal isomeri-zation despite the different opsin environment (especially thedifferent counterion location and configuration).The calculations above also indicate that, in hMeOp, the two

S0 transition states TSCT and TSDIR are closer in energy than insqRh and in Rh. Thus, hMeOp would mark the border between aBarlow correlation and an anti-Barlow correlation where PSB11-based pigments with a λmax absorbing further to the blue wouldlead to an increase of the background noise. On the basis of our

models one can therefore conclude that colors in the 470–480 nmrange are the best perceived for both visual and nonvisual pigmentsincorporating an A1 chromophore.In conclusion, to provide information on the events preceding

the activation of nonvisual photoreceptors, we have comparedconsistent QM/MM models of human melanopsin and inver-tebrate and vertebrate rhodopsins. The models allowed studyingthe differences in structure, excitation energy, excited state life-time, and thermal isomerization barriers. Assuming the validity ofa correlation between isomerization speed and quantum yields fortimes below 100 fs and of a relationship between rate of thermalisomerization and thermal activation (20, 21), our results areconsistent with an extreme light sensitivity of ipRGCs and thedetection of low levels of radiance. Although more work is re-quired to precisely relate the opsin cavity residue compositionwith the simultaneous modulation of quantities such as λmax,τcis→trans, and EaT, our comparative analysis indicates the needto go beyond single-point charge models (55) and learn sys-tematically the effect of more complex charge configurations.Indeed, the transition from distributed (a cluster of blue-shiftingresidues with a precise spatial configuration) to localized coun-terions deserves further investigation as distinct charge config-urations may characterize nonvisual, visual invertebrate, and visualvertebrate photoreceptors.

MethodsThe hMeOp model was constructed by comparative modeling, by means ofthe MODELLER program (56) using the crystallographic structure of sqRh(PDB code: 2Z73, chain A, 40% primary sequence similarity) (33) as a tem-plate (SI Appendix, Fig. S2). The sequence alignment used for comparativemodeling is shown in SI Appendix, Fig. S1. All protein portions but theN-term and C-term were modeled. One thousand models were built byrandomizing all of the Cartesian coordinates of standard residues in theinitial model. A high degree of model refinement was set. Finally, amongthe top 15 models showing the lowest violation of spatial restraints (i.e.,as accounted for by the OBJECTIVE FUNCTION), model #490 was selected,being characterized by the highest value of both 3D-Profile score andstereochemical quality concerning the main chain dihedral angles.

The model was subjected to adjustment of the side-chain torsion angleswhen in nonallowed conformation and optimized by means of the Chemistryat Harvard Molecular Mechanics (CHARMM) force field, using the generalizedBorn with simple switching function (GBSW) implicit membrane/watermodel (57). The retinal chromophore coordinates extracted from 2Z73 werekept fixed.

The resulting comparative model was used to start hybrid QM/MM (58, 59)calculations. QM/MM studies on structures obtained by comparative mod-eling have already been proposed in several studies (34, 60, 61) for proteinswithout an available crystal structure, such as the three human retina conepigments. We supported the above protocol by building the model ofhuman rhodopsin using the crystallographic structure of Rh as a template(95% sequence similarity) and reproducing the observed λmax.

Although the QM/MM model of hMeOp was constructed from its com-parative model, the corresponding Rh and sqRh models were constructedstarting from their crystallographic structures. The same protocol was used inall cases for consistency. It was shown that a hMeOP model featuring twocavity waters is thermally more stable than the one where the waters areabsent. For consistency, twowaters were then added to the homologous sqRhmodel which was found structurally stable (SI Appendix). In all modelsthe retinal chromophore was treated quantum mechanically using the com-plete-active-space self-consistent field (CASSCF) method (62) and embeddedin a protein environment described by the molecular mechanics AMBER forcefield. CASSCF is an ab initio MCQC method (i.e., with no empirically derivedparameters and avoiding single-reference wavefunctions) offering a bal-anced description of the electronic and geometrical structure of a reactingmolecule. The CASSCF wavefunction can be used for subsequent multi-configurational second-order perturbation theory (CASPT2) computations ofthe dynamic correlation energy (63) of each state thus allowing for a morequantitative evaluation of energy barriers and excitation energies. In pre-vious work we have shown that CASPT2//CASSCF-based (i.e., CASSCF geom-etry optimization and CASPT2 energy evaluation) QM/MM models reproducespectroscopic properties within a few kcal mol−1 error (29, 64). All final modelsand the resulting properties described above were tested for stability on

1718 | www.pnas.org/cgi/doi/10.1073/pnas.1309508111 Rinaldi et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 2

4, 2

020

the basis of additional models generated via a thermal sampling of thechromophore cavity residues. The details of the chosen QM/MM protocoland stability testing are reported in SI Appendix which also includes thedetails of trajectory and transition state computations.

ACKNOWLEDGMENTS. This work was supported in part by the NationalScience Foundation (NSF) under Grant CHE-1152070 and the Human Frontier

Science Program Organization under Grant RGP0049/2012CHE09-56776. M.O.is grateful to the Center for Photochemical Sciences and School of Arts andSciences of the Bowling Green State University. F.F. was supported by theTelethon–Italy Grant GGP11210/S00068TELC. The European Cooperation inScience and Technology Action CM1002 (CODECS) is also acknowledged. Theauthors are indebted to the NSF-Extreme Science and Engineering DiscoveryEnvironment and Ohio Supercomputer Center for granted computer time.

1. Berson DM, Dunn FA, Takao M (2002) Phototransduction by retinal ganglion cells thatset the circadian clock. Science 295(5557):1070–1073.

2. Provencio I, et al. (2000) A novel human opsin in the inner retina. J Neurosci 20(2):600–605.

3. Hattar S, Liao HW, Takao M, Berson DM, Yau KW (2002) Melanopsin-containingretinal ganglion cells: Architecture, projections, and intrinsic photosensitivity. Science295(5557):1065–1070.

4. Ruby NF, et al. (2002) Role of melanopsin in circadian responses to light. Science298(5601):2211–2213.

5. Güler AD, et al. (2008) Melanopsin cells are the principal conduits for rod-cone inputto non-image-forming vision. Nature 453(7191):102–105.

6. Arendt D (2003) Evolution of eyes and photoreceptor cell types. Int J Dev Biol 47(7-8):563–571.

7. Terakita A, Kawano-Yamashita E, Koyanagi M (2012) Evolution and diversity of op-sins. WIREs Membr Transp Signal 1(1):104–111.

8. Berson DM (2007) Phototransduction in ganglion-cell photoreceptors. Pflügers Arch.EJP 454(5):849–855.

9. Lok C (2011) Vision science: Seeing without seeing. Nature 469(7330):284–285.10. Mure LS, et al. (2009) Melanopsin bistability: A fly’s eye technology in the human

retina. PLoS ONE 4(6):e5991.11. Parker RO, Crouch RK (2010) Retinol dehydrogenases (RDHs) in the visual cycle. Exp

Eye Res 91(6):788–792.12. Isoldi MC, Rollag MD, Castrucci AMDL, Provencio I (2005) Rhabdomeric photo-

transduction initiated by the vertebrate photopigment melanopsin. Proc Natl AcadSci USA 102(4):1217–1221.

13. Peirson SN, Halford S, Foster RG (2009) The evolution of irradiance detection: Mela-nopsin and the non-visual opsins. Philos Trans R Soc Lond B Biol Sci 364(1531):2849–2865.

14. Do MTH, et al. (2009) Photon capture and signalling by melanopsin retinal ganglioncells. Nature 457(7227):281–287.

15. Ferrer C, Malagón G, Gomez MdelP, Nasi E (2012) Dissecting the determinants of lightsensitivity in amphioxus microvillar photoreceptors: Possible evolutionary im-plications for melanopsin signaling. J Neurosci 32(50):17977–17987.

16. Provencio I, Jiang G, De Grip WJ, Hayes WP, Rollag MD (1998) Melanopsin: An opsin inmelanophores, brain, and eye. Proc Natl Acad Sci USA 95(1):340–345.

17. Drivenes Ø, et al. (2003) Isolation and characterization of two teleost melanopsingenes and their differential expression within the inner retina and brain. J CompNeurol 456(1):84–93.

18. Chaurasia SS, et al. (2005) Molecular cloning, localization and circadian expression ofchicken melanopsin (Opn4): Differential regulation of expression in pineal and retinalcell types. J Neurochem 92(1):158–170.

19. Baylor DA, Matthews G, Yau KW (1980) Two components of electrical dark noise intoad retinal rod outer segments. J Physiol 309(1):591–621.

20. Aho AC, Donner K, Hydén C, Larsen LO, Reuter T (1988) Low retinal noise in animalswith low body temperature allows high visual sensitivity. Nature 334(6180):348–350.

21. Barlow HB (1988) The thermal limit to seeing. Nature 334(6180):296–297.22. Kefalov V, Fu Y, Marsh-Armstrong N, Yau KW (2003) Role of visual pigment prop-

erties in rod and cone phototransduction. Nature 425(6957):526–531.23. Lórenz-Fonfría VA, Furutani Y, Ota T, Ido K, Kandori H (2010) Protein fluctuations as

the possible origin of the thermal activation of rod photoreceptors in the dark. J AmChem Soc 132(16):5693–5703.

24. Luo DG, Yue WW, Ala-Laurila P, Yau KW (2011) Activation of visual pigments by lightand heat. Science 332(6035):1307–1312.

25. Schoenlein RW, Peteanu LA, Mathies RA, Shank CV (1991) The first step in vision:Femtosecond isomerization of rhodopsin. Science 254(5030):412–415.

26. Kukura P, McCamant DW, Yoon S, Wandschneider DB, Mathies RA (2005) Structuralobservation of the primary isomerization in vision with femtosecond-stimulatedRaman. Science 310(5750):1006–1009.

27. Kandori H, Shichida Y, Yoshizawa T (2001) Photoisomerization in rhodopsin. Bio-chemistry (Mosc) 66(11):1197–1209.

28. Wang Q, Schoenlein RW, Peteanu LA, Mathies RA, Shank CV (1994) Vibrationallycoherent photochemistry in the femtosecond primary event of vision. Science266(5184):422–424.

29. Andruniów T, Ferré N, Olivucci M (2004) Structure, initial excited-state relaxation, andenergy storage of rhodopsin resolved at the multiconfigurational perturbation the-ory level. Proc Natl Acad Sci USA 101(52):17908–17913.

30. Frutos LM, Andruniów T, Santoro F, Ferré N, Olivucci M (2007) Tracking the excited-state time evolution of the visual pigment with multiconfigurational quantumchemistry. Proc Natl Acad Sci USA 104(19):7764–7769.

31. Polli D, et al. (2010) Conical intersection dynamics of the primary photoisomerizationevent in vision. Nature 467(7314):440–443.

32. Gozem S, Schapiro I, Ferré N, Olivucci M (2012) The molecular mechanism of thermalnoise in rod photoreceptors. Science 337(6099):1225–1228.

33. Murakami M, Kouyama T (2008) Crystal structure of squid rhodopsin. Nature453(7193):363–367.

34. Sekharan S, Wei JN, Batista VS (2012) The active site of melanopsin: The biologicalclock photoreceptor. J Am Chem Soc 134(48):19536–19539.

35. Walker MT, Brown RL, Cronin TW, Robinson PR (2008) Photochemistry of retinalchromophore in mouse melanopsin. Proc Natl Acad Sci USA 105(26):8861–8865.

36. Matsuyama T, Yamashita T, Imamoto Y, Shichida Y (2012) Photochemical propertiesof mammalian melanopsin. Biochemistry 51(27):5454–5462.

37. Shichida Y, Tokunaga F, Yoshizawa T (1979) Squid hypsorhodopsin. PhotochemPhotobiol 29(2):343–351.

38. Hurley JB, Ebrey TG, Honig B, Ottolenghi M (1977) Temperature and wavelengtheffects on the photochemistry of rhodopsin, isorhodopsin, bacteriorhodopsin andtheir photoproducts. Nature 270(5637):540–542.

39. Terakita A, et al. (2004) Counterion displacement in the molecular evolution of therhodopsin family. Nat Struct Mol Biol 11(3):284–289.

40. Coto PB, Strambi A, Ferré N, Olivucci M (2006) The color of rhodopsins at the ab initiomulticonfigurational perturbation theory resolution. Proc Natl Acad Sci USA 103(46):17154–17159.

41. Altun A, Yokoyama S, Morokuma K (2008) Spectral tuning in visual pigments: anONIOM(QM:MM) study on bovine rhodopsin and its mutants. J Phys Chem B 112(22):6814–6827.

42. Tomasello G, et al. (2009) Electrostatic control of the photoisomerization efficiencyand optical properties in visual pigments: On the role of counterion quenching. J AmChem Soc 131(14):5172–5186.

43. Schapiro I, et al. (2011) The ultrafast photoisomerizations of rhodopsin and bath-orhodopsin are modulated by bond length alternation and HOOP driven electroniceffects. J Am Chem Soc 133(10):3354–3364.

44. Schapiro I, Weingart O, Buss V (2009) Bicycle-pedal isomerization in a rhodopsinchromophore model. J Am Chem Soc 131(1):16–17.

45. Warshel A (1976) Bicycle-pedal model for the first step in the vision process. Nature260(5553):679–683.

46. Barlow HB (1957) Purkinje shift and retinal noise. Nature 179(4553):255–256.47. Ala-Laurila P, Pahlberg J, Koskelainen A, Donner K (2004) On the relation between

the photoactivation energy and the absorbance spectrum of visual pigments. VisionRes 44(18):2153–2158.

48. Weingart O, et al. (2011) Product formation in rhodopsin by fast hydrogen motions.Phys Chem Chem Phys 13(9):3645–3648.

49. Gozem S, et al. (2012) Dynamic electron correlation effects on the ground state po-tential energy surface of a retinal chromophore model. J Chem Theory Comput 8(11):4069–4080.

50. Weiss RM, Warshel A (1979) A new view of the dynamics of singlet cis-trans photo-isomerization. J Am Chem Soc 101(20):6131–6133.

51. Kim JE, Tauber MJ, Mathies RA (2003) Analysis of the mode-specific excited-stateenergy distribution and wavelength-dependent photoreaction quantum yield inrhodopsin. Biophys J 84(4):2492–2501.

52. Zener C (1932) Non-adiabatic crossing of energy levels. Proc R Soc A 137(833):696–702.53. Zgrabli�c G, Novello AM, Parmigiani F (2012) Population branching in the conical in-

tersection of the retinal chromophore revealed by multipulse ultrafast optical spec-troscopy. J Am Chem Soc 134(2):955–961.

54. Sovdat T, et al. (2012) Backbone modification of retinal induces protein-like excitedstate dynamics in solution. J Am Chem Soc 134(20):8318–8320.

55. Nakanishi K (1991) 11-cis-retinal, a molecule uniquely suited for vision. Pure ApplChem 63(1):161–170.

56. Sali A, Blundell TL (1993) Comparative protein modelling by satisfaction of spatialrestraints. J Mol Biol 234(3):779–815.

57. Im W, Feig M, Brooks CL, 3rd (2003) An implicit membrane generalized born theoryfor the study of structure, stability, and interactions of membrane proteins. Biophys J85(5):2900–2918.

58. Warshel A, Chu ZT (2001) Nature of the surface crossing process in bacteriorhodopsin:Computer simulations of the quantum dynamics of the primary photochemical event.J Phys Chem B 105(40):9857–9871.

59. Gao J (1996) Hybrid quantum and molecular mechanical simulations: An alternativeavenue to solvent effects in organic chemistry. Acc Chem Res 29(6):298–305.

60. Fujimoto K, Hasegawa J, Nakatsuji H (2008) Origin of color tuning in human red,green, and blue cone pigments: SAC-CI and QM/MM study. Chem Phys Lett 462(4-6):318–320.

61. Trabanino RJ, Vaidehi N, Goddard WA, 3rd (2006) Exploring the molecular mechanismfor color distinction in humans. J Phys Chem B 110(34):17230–17239.

62. Roos BO, Lawley KP (1987) Ab initio methods in quantum chemistry II. Adv Chem Phys69:399–446.

63. Andersson K, Malmqvist P-A, Roos BO, Sadlej AJ, Wolinski KJ (1990) Second-orderperturbation theory with a CASSCF reference function. J Phys Chem 94(14):5483–5488.

64. Ferré N, Olivucci M (2003) Probing the rhodopsin cavity with reduced retinal modelsat the CASPT2//CASSCF/AMBER level of theory. J Am Chem Soc 125(23):6868–6869.

Rinaldi et al. PNAS | February 4, 2014 | vol. 111 | no. 5 | 1719

CHEM

ISTR

YBIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Aug

ust 2

4, 2

020