Modulation of thermal noise and spectral sensitivity in Lake Baikal cottoid fish rhodopsins

Lake Baikal is the deepest and one of the most ancient lakes in the world. Its unique ecology has resulted in the colonization of a diversity of depth habitats by a unique fauna that includes a group of teleost fish of the sub-order Cottoidei. This relatively recent radiation of cottoid fishes shows a gradual blue-shift in the wavelength of the absorption maximum of their visual pigments with increasing habitat depth. Here we combine homology modeling and quantum chemical calculations with experimental in vitro measurements of rhodopsins to investigate dim-light adaptation. The calculations, which were able to reproduce the trend of observed absorption maxima in both A1 and A2 rhodopsins, reveal a Barlow-type relationship between the absorption maxima and the thermal isomerization rate suggesting a link between the observed blue-shift and a thermal noise decrease. A Nakanishi point-charge analysis of the electrostatic effects of non-conserved and conserved amino acid residues surrounding the rhodopsin chromophore identified both close and distant sites affecting simultaneously spectral tuning and visual sensitivity. We propose that natural variation at these sites modulate both the thermal noise and spectral shifting in Baikal cottoid visual pigments resulting in adaptations that enable vision in deep water light environments.

Scientific RepoRts | 6:38425 | DOI: 10.1038/srep38425 An alternative explanation for the λ max blue-shift observed in the deeper Baikal rhodopsins may be based on the existence of a Barlow-like correlation 9,10 . This is an inverse proportionality relationship between rhodopsin λ max and chromophore thermal isomerization rate. By competing with the photoisomerization triggering the rhodopsin function, the thermal isomerization must contribute to the thermal noise 11,12 decreasing visual acuity 13 . When assuming the validity of such an isomerization-noise link, the Barlow correlation implies that the λ max blue-shift of the abyssal rhodopsins would reflect the need to reduce the noise in habitats with low light intensities 5,14 .
Here we investigate the thermal noise hypothesis through a combination of multiconfigurational quantum chemistry (MCQC) calculations and experimental studies on a set of Baikal rhodopsins. More specifically, we demonstrate the existence of a Barlow correlation between λ max , expressed in terms of the chromophore vertical excitation energy (Δ E), and isomerization rate, related to the chromophore activation energy (E a T ) and derive an atomistic model of the Δ E and E a T variation. The results show that the amino acid substitutions found in the sequences of the selected Baikal rhodopsin set, modulate Δ E and E a T simultaneously in an interdependent parallel fashion suggesting that a reduction in thermal noise may have evolved in Lake Baikal fish pigments as a dim-light adaptation for increased photosensitivity.

Results and Discussion
The sequence similarity and marked λ max variation of Baikal rhodopsins facilitate the study of the effect of single amino acid substitutions on Δ E and E a T . Accordingly, we consider four species representative of different habitats (in order of depth): the littoral (1-5 m) depth (Paracottus kneri, λ max = 516 nm), the sub-littoral (1-120 m) depth (Paracottus jettelesi, λ max = 505 nm), the supra-abyssal (50-450 m) depth (Cottocomephorus inermis, λ max = 495 nm), and the abyssal (400-1500 m) depth (Abyssocottus korotneffi, λ max = 484 nm) 5 . For each species The amino acid of Rh belong either to the "cavity" or "extra-cavity" region. (b) Structure of the Rh Lys 296 -Chromophore (green-blue) and its cavity (red) comprising the conserved E113 counterion. (c) Structure of the A1 (11-cis retinal) and A2 (11-cis 3-dehydroretinal) chromophores. The arrows indicates the double-bond isomerization triggering the pigment function. (d) Effect of a negative (red) point charge located in proximity of the chromophore β -ionone ring. The charge would stabilize the electronically photo-excited state (S 1 ) dominated by a charge-transfer electronic configuration (φ CT ) with respect to the ground state (S 0 ) dominated by a covalent electronic configuration (φ COV ) leading to an increase in λ max (i.e. a decrease in Δ E as shown in the red energy level diagram on the right). A charge of the opposite sign placed in the same location would lead to the opposite effect (blue energy level diagram). (e) Relationship between the Δ E (proportional to 1/λ max ) and the E a T controlling the chromophore thermal isomerization according to the point-charge model of ref. 8. A schematic representation of the chromophore charge distribution at the transition state (TS) is also given. The same negative/positive point charge would decrease/increase the barrier respectively. a MCQC-based quantum-mechanics/molecular-mechanics (QM/MM) model of the corresponding rhodopsin is constructed (Methods section and SI Appendix, Fig. S1) using, as a template,the crystallographic structure of bovine rhodopsin (Rh). The model quality is assessed by reproducing: (i) the observed λ max changes along the set plus the Rh template and (ii) the observed linear relationships between the λ max of A1/A2 pairs of pigments 15,16 . This second test is carried out by preparing and spectroscopically characterizing in vitro Baikal rhodopsins where the A1 chromophore is replaced by the A2 chromophore forming red-shifted analogs 17 . The validated QM/MM models are then used to study the effects of the amino acid substitutions differentiating the four species through a computational implementation 18 of the point-charge model proposed by Nakanishi and coworkers 19 (Fig. 1d and e).
Origin of the excitation energy changes. Three Lake Baikal pigments were expressed and purified in vitro with both the A1 and A2 chromophores. The measured λ max of the A1 rhodopsins were found to be almost identical to the literature values measured via microspectroscopy (MSP) 2 . C. inermis λ max was identical to MSP measurements (495 nm), while A. korotneffi was found to absorb maximally at 482 nm (− 2 nm from MSP values) and P. jettelesi absorbed at 501 nm (− 4 nm from MSP values) (SI Appendix, Fig. S2). As expected, the λ max of the A2 rhodopsins was found to be red-shifted in comparison to the corresponding A1 rhodopsin value. A. korotneffi A2 pigment shifted to 499 nm, a total red-shift of 17 nm. P. jettelesi rhodopsin expressed with A2 chromophore shifted by 19 nm to 520 nm. C. inermis was red-shifted by 21 nm to 516 nm in the A2 pigment (SI Appendix, Fig. S2). All A2 rhodopsins were also successfully light bleached and their MII intermediate also showed the characteristic observed blue-shifted λ max with respect to the dark adapted state (SI Appendix, Fig. S3) as expected for functional pigments.
As reported in Fig. 2a (see also SI Appendix, Table S1), the observed A1 and A2 rhodopsin λ max trends as well as the related A1/A2 linear relationship are reproduced by the QM/MM models. Furthermore, the computed A1/A2 slope only modestly deviate from that established experimentally by Dartnall and Lythgoe 15 showing a 5 nm error (i.e. < 1 kcal mol −1 ).
In order to investigate the origin of the λ max trend, we computed the Δ E values ( Fig. 2a and SI Appendix, Table S1) for the isolated (in vacuo) chromophores of the four Baikal pigments. In these computations, the geometrical parameters of the chromophore are fixed at the values of the S 0 equilibrium structure of the QM/MM model. The results provide information on the Δ E variations due to the changes in chromophore geometry. Within the A1 and A2 sets, the Δ E values show only limited ≤ 1 kcal mol −1 variations consistently with the limited geometrical changes displayed in Fig. 2b (i.e. with dihedral angle changes ≤ 4 degrees). Thus, the model indicates that the λ max variations are not due to progressive chromophore distortion (except for a fraction in the case of C. inermis) and must be dominated by electrostatic effects (i.e. by the variations in the point charges of cavity and extra-cavity amino acids).
Effect of cavity and extra-cavity amino acids. The Δ E change between the most red-shifted model (P. kneri) and the most blue-shifted model (A. korotneffi) is 1.9 and 3.3 kcal mol −1 for the A1 and A2 chromophore respectively. This value (see Fig. 1d) reflects the stabilizing effect of the P. kneri and A. korotneffi protein environments on the difference in S 1 and S 0 charge distribution of the chromophore (the S 1 /S 0 charge difference of Fig. 2c). Since the S 1 /S 0 charge difference is similar in all pigments, we focused on the larger Δ E changes of the A2 rhodopsins.
The Δ E decrease (red-shift) or increase (blue-shift) associated with a specific side-chain, can be evaluated by setting its point charges to zero and recomputing the excitation energy (Δ E off ). The largest Δ E-Δ E off differences computed for the cavity residues are displayed in the balloon diagrams of Fig. 3a. When comparing the effects of side-chain substitutions, one finds that a Δ E change may have two components. The first is a direct component due to the change in number, magnitude and position of the corresponding side-chain point charges. The second component is indirect and originates from the reorganization of the hydrogen bond network (HBN) induced by the same substitution. This second component/effect explains why conserved residues and water molecules may display large Δ E-Δ E off changes and contribute to the total Δ E variation significantly.
When comparing the extreme cases of P. kneri (reddest) and A. korotneffi (bluest), the sequence data shows that the amino acid substitutions G114A and Y261F remove two red-shifting residues in P. kneri (see Fig. 3a) which directly contribute to blue-shifting the A. korotneffi absorption. While the same data shows that A292S does not change the Δ E-Δ E off , below we will see that this substitution modifies the HBN which then blue-shifts the λ max indirectly. Thus variations in the composition of the rhodopsin cavity modulates the λ max between littoral and abyssal habitats through direct and indirect changes. The same analysis indicates that, due to a cancellation of Δ E-Δ E off of opposite signs (e.g. the sizable R140C red-shifting replacement is counterbalanced by the smaller T209I, L176S, T297S blue-shifting replacements in Fig. S6), the substitution of extra-cavity residues contributes only modestly to the λ max change from P. kneri to A. korotneffi.
The sub-littoral and supra-abyssal species P. jettelesi and C. inermis feature the same amino acid cavity composition and similar cavity Δ E-Δ E off values (SI Appendix, Fig. S6). In contrast, the extra-cavity substitutions T297S, D83N, T166S relating these species are associated with direct blue-shifting changes. This suggests that spectral tuning among species in the closer sub-littoral and supra-abyssal habitats may be controlled by extra-cavity amino acids. On the other hand, the Δ E variations computed between sub-littoral and littoral and between abyssal and supra-abyssal are modulated by both cavity and extra-cavity substitutions and by direct and indirect changes (SI Appendix, Table S4 and S5) as we will discuss below.

Activation energy changes.
In order to find out if the blue-shift observed when passing from the littoral to the abyssal habitat reflects the need to reduce the rhodopsin thermal noise, we built the QM/MM models for the S 0 transition states (TS, Fig. 1e) that control thermal isomerization. The models allow to compute the corresponding E a T , thermal activation energy. The results yield a linear relationship between E a T and 1/λ max (Fig. 2d) with the most blue-shifted A1 rhodopsin (from A. korotneffi) displaying an E a T 5.4 kcal mol −1 higher than the E a T of the most red-shifted rhodopsin (from P. kneri). Notice that the present work is not aimed at reproducing the absolute values of the observed barriers but only their variation among different Baikal species. This is discussed in Section 6 of the SI Appendix which highlights a non-Arrhenius behavior as a source of discrepancy between computed and available observed E a T values. In the same section, an additional source of inaccuracy is associated with the fact that reactant and transition state structures are computed as single points on the rhodopsin potential energy surface without explicitly accounting for the protein dynamics at body temperature. However, this error is expected to be systematic and therefore unable to affect the computed trends.
The geometrical structures of the chromophore at the TSs of A. korotneffi and P. kneri (see Fig. 2e) at the transition state are similar and consistent with those reported for Rh 18 . The structures indicate that the isomerization occurs via an aborted bicycle-pedal reaction coordinate 20,21 involving the -C9= C10-C11= C12-segment of the chromophore backbone. Such motion is coupled with a substantially complete charge translocation from the  Fig. 1d and e respectively).
Similar to what was found for Δ E, the E a T of the chromophores in vacuo, i.e. the energy difference between the chromophores extracted from the QM/MM models of the TS and S 0 reactant, are close (see Fig. 2d and Table S6). It is therefore concluded that the changes in E a T are due to variations in the protein environment. Furthermore in Fig. 2d we show that electrostatic interactions prevail over steric (e.g. van der Waals) interactions. In order to isolate the steric effects, we zeroed all protein charges of the models and recomputed the E a T values. The A. korotneffi value is found to be lower than the corresponding P. kneri value showing that steric effects would, as confirmed by the P. jettelesi and C. inermis E a T values, result in a trend opposite to the one observed when both steric and electrostatic effects are considered (see also SI Appendix). It is thus concluded that the protein electrostatics determines the E a T trend. According to the point charge model, E a T is modulated by the residue charges which "stabilize" or "destabilize" the TS/S 0 charge changes (Fig. 2f). Such difference is qualitatively similar to the S 1 /S 0 charge change (compare Fig. 2c and f). Thus, we investigate the differences in E a T between P. kneri and A. korotneffi by applying the same analysis employed for Δ E. Accordingly, the effect of each residue is evaluated by computing the quantity E a T -E a T off (E a T off being the barrier obtained after zeroing the charges of a specific residue). When a residue is replaced such quantity is expected to display variations similar to the one seen for Δ E-Δ E off but the significance of which is more complex to interpret. In fact, while Δ E-Δ E off reflects, by definition, the effect of the residue charges, E a T -E a T off also incorporates the effect of the geometrical difference between the TS and the S 0 reactant. The E a T -E a T off variations induced by the rhodopsin cavity substitutions relating P. kneri to A. korotneffi (Y261F, A292S and G114A), are given in Fig. 3b and c and Table S7. Y261F leads, through a direct change, to an increase of E a T in A. korotneffi (2.7 kcal mol −1 ) consistently with the effect reported above for Δ E. As shown in Fig. 3c  A292S leads, again through a direct change, to a large increase (4.8 kcal mol −1 ) in E a T of A. korotneffi. Although this variation parallels the corresponding Δ E increase, the modeled E a T change is due to HBN modification rather than a direct change as for Δ E. Finally, while G114A (see Fig. S7) leads to a negligible E a T variation in A. korotneffi, it causes a limited direct Δ E increase (0.8 kcal mol −1 ). In conclusion, while the overall variation induced by the three substitutions show the same trend for both Δ E and E a T , their contributions may be mechanistically distinct as we detail below.
Mechanisms of thermal noise modulation and spectral tuning. As reported above the combined effects of three cavity substitutions (see Fig. 4a) play a substantial role in establishing the differences between the Δ E and E a T of P. kneri and A. korotneffi. The Y261F substitution blue-shifts the λ max of all species relative to P. kneri by effectively changing the side-chain point charges. In fact, Y261F loses a dipole (the OH group of tyrosine) pointing its negative pole towards the β -ionone ring (see Fig. 4a top). This destabilizes the S 1 /S 0 charge difference of Fig. 2c increasing the Δ E and leading to a blue-shift. As shown in Fig. 4b the same mechanism is seen when comparing P. kneri and P. jettelesi.
A parallel mechanism explains the increase of E a T in A. korotneffi. with respect to P. kneri. In fact, similar to the Δ E effect, the Y261F substitution in A. korotneffi destabilizes the TS/S 0 charge shift laid out in Fig. 2f and thus increases the E a T . However, in contrast to Δ E, E a T is also modulated via an indirect effect of the same Y261F substitution. In fact, the loss of OH in position 261 in A. korotneffi, which used to form a hydrogen bond with the backbone oxygen of the conserved G121 residue in P. kneri (compare bottom and top in Fig. 3b), induces an HBN change. This change affects the stability of the TS and S 0 reactant differently and contributes to the E a T increase in A. korotneffi.
As seen in Fig. 3c, the A292S substitution relating P. kneri to A. korotneffi does not blue-shift the λ max through a direct change, but through a modification of the HBN. In fact, A292S induces a relocation/reorientation of WAT2 which displaces it away from the Schiff base region (see Figs 3c and 4a bottom). Since the positive pole of WAT2 points towards the -C15= NH-moiety and destabilizes the S 1 /S 0 charge difference, such WAT2 relocation increases the Δ E in A. korotneffi. The same mechanism, which is also responsible for the P. jettelesi to A. korotneffi λ max blue-shift (see Fig. 4c), explains the increased E a T in A. korotneffi through a decreased destabilization of the TS/S 0 charge difference. However, the A292S induced WAT2 relocation also mediates a secondary indirect change of E a T . In fact, it perturbs an HBN connecting the conserved residues E181, S186 and Y268 (see bottom and top in Fig. 4b) which thus contribute to modulate E a T . This is demonstrated by the 1.3 kcal mol −1 increase of S186 and − 2.1 and − 1.4 kcal mol −1 decrease of E181 and Y268 respectively in A. korotneffi compared to P. kneri. Notice that, although individually E181 and Y268 induce a reduction in E a T , such HBN modulation is dominated by the 4.6 kcal mol −1 increase due to WAT2 (see Fig. 3c).
Finally, the G114A substitution, which replaces a non-polar residue with a sterically larger residue, shows a contrasting effect in A. korotneffi. As shown in Fig. 4a top and c, the G114 hydrogen of P. kneri and P. jettelesi is close to the Schiff base linkage and stabilizes the S 1 /S 0 charge difference. Thus, the G114A substitution in A. korotneffi contributes to increase the Δ E. Such Δ E change is not paralleled E a T which instead decreases. Nevertheless, due to the limited change in polarity, the decrease (see Fig. S7) is smaller than the E a T increase due to the Y261F and A292S substitutions.
In conclusion, point-charge analysis has revealed a set of substitutions which simultaneously modulate Δ E and E a T via cooperative direct and indirect HBN mediated mechanisms. While the magnitude of the described changes is expected to be sensitive to the details of our basic QM/MM models, the same substitutions have (b) Spectral-tuning mechanism related to the Y261F substitution between P. kneri (red) and P. jettelesi (yellow). (c) Spectral-tuning mechanism related to the A292S and G114A substitutions between P. jettelesi and A. korotneffi (blue). (d) Cavity and extra-cavity substitutions associated with the transitions between different rhodopsins. The full arrows indicate the proposed evolutionary relationship between the corresponding species when assuming P. jettelesi to be the closest to the ancestor. In contrast, the grey arrow indicates the substitutions involved in the transition between littoral and abyssal species.
Scientific RepoRts | 6:38425 | DOI: 10.1038/srep38425 been detected in other contexts. In fact, Y261F has been shown to be responsible for the spectral differentiation between green and red cone pigments in primates 22 . G114A has been shown to cause a blue-shift also in Rh when expressed in vivo 23,24 and in spite of the limited polarity change. A292S has also been detected in blue-shifted rhodopsin from other fish 25 , marine mammals 26 , and monotremes 27 . Light-sensitivity in related species. Above we have employed MCQC-based QM/MM models of rhodopsins reconstituted with both A1 and A2 retinals to investigate the relationship between spectral tuning and thermal isomerization rate in different species of cottoid fish. The results support the existence of a direct proportionality relationship between Δ E and E a T for pigments of closely related species which evolved in the confined environment of Lake Baikal. This expands the validity of the Barlow correlation discussed for rod and cone pigments of distant species 9,10,28 and provides a link with the observed inverse proportionality relationship between λ max and isomerization rate in proton-pumping rhodopsins 29 and even in the extreme case of 13-cis retinal chromophore salts in solution 30 .
The Δ E and E a T proportionality originates at the electronic level. Indeed, the similarity between the S 1 /S 0 and TS/S 0 charge differences, (see Fig. 2c and f) due to the changes in chromophore π -electron density, makes Δ E and E a T sensitive to the same substitutions. At a more fundamental level, such similarity originates from the fact that the same charge transfer configuration (φ CT ) of the chromophore dominates the rhodopsin vertical S 1 state and S 0 transition state. As previously shown 18 , this is a consequence of a quantum mechanical property of the conical intersection of the rhodopsin chromophore 18,31 . Therefore the λ max changes observed in Baikal rhodopsins reflects the biological exploitation of a quantum effect to increase light sensitivity 32 .
The analysis of the QM/MM models indicates that the variation of Δ E and E a T in phylogenetically closely related rhodopsins is controlled by the electrostatic characteristics of the protein. Our implementation of Nakanishi's point charge analysis has identified 8 rhodopsin substitutions, over a total of 20, modulating light sensitivity from red-shifted P. kneri to the blue-shifted A. korotneffi. The same analysis also produced an "atomistic model" of dim-light adaptation through specific side-chain substitutions. Through this model, specific mechanisms can be associated to the proposed phylogeny 5 assumed to originate from P. jettelesi as its λ max matches that of the ancestor. While the modification of the point charges associated with a cavity substitution have a direct impact on Δ E and E a T (e.g. F261Y when comparing P. jettelesi and P. kneri in Fig. 4b), it would be impossible to model the observed trends without taking into account the HBN modifications associated with the same substitution (e.g. A292S comparing P. jettelesi and P. korotneffi in Fig. 4c) or the effect of extra-cavity substitutions (e.g. D83N, T297S and T166S when comparing P. jettelesi and C. inermis and, additionally, S298A in P. jettelesi and A. korotneffi). Also, in our QM/MM models, extra-cavity substitutions display large effects when an ionized residue replaces a neutral one (e.g. C140R replacing cysteine in P. jettelesi to a arginine in P. kneri).
In conclusion, when assuming that the thermal isomerization of rhodopsin dominates its thermal noise, the regular Baikal rhodopsin blue-shift observed when moving from littoral to abyssal habitats may be a byproduct of visual adaptations to extremely low levels of illumination. In fact, our study shows that for Baikal fishes, these two aspects of visual pigment function are interdependent: the isomerization rate (which would determine the amount of thermal noise) and the wavelength of maximal absorbance. Amino acid substitutions have evolved in these fishes that shift both quantities simultaneously for adaptations that would contribute to better visual sensitivity and enable colonization of the dimly lit blue-shifted deepwater environments of Lake Baikal. Our results suggest that it is possible similar mechanisms may underlie colonization of other deepwater dimly lit environments such as those inhabited by deep sea fishes in marine habitats.

Molecular biology methods.
No experiments on live vertebrates were carried out in this study. Incomplete Baikal cottoid RH1 sequences were taken from 5 and completed with wildtype bovine sequences for the N-and C-termini. The full-length hybrid RH1 genes were synthesized by GeneArt (Invitrogen) with BamHI and EcoRI restriction sites at the 5′ and 3′ ends, respectively. The synthesized sequences were then inserted into the p1D4-hrGFP II expression vector which tags expressed rhodopsin sequences with the nine amino acid 1D4 peptide sequence (TETSQVAPA) at the carboxy terminus 33 . This enables immunoaffinity purification of expressed proteins from HEK293T cells as previously described 34,35 . UV-vis absorption spectra of purified rhodopsin samples were measured at room temperature both in the dark, and following light-bleaching for 60 seconds using a fiber optic lamp. Difference spectra were calculated by subtracting the light-bleached spectra from respective dark spectra. To provide accurate estimates of λ max , dark absorbance spectra were fit to standard templates for either A1 or A2 visual pigments 36 . Computational methods. The QM/MM models of both A1 and A2 fish rhodopsins were prepared starting with a structures obtained via comparative modeling. To do so, the chain A of the 1U19 structure of bovine rhodopsin 37 was used as a template. The models were then constructed by relaxing the cavity-counterion-chromophore complex in its protein environment via molecular dynamics and geometry optimization. The chromophore was treated using the complete-active-space self-consistent field (CASSCF) method 38 with an active space corresponding to the entire π -system and the 6-31G* basis set. The protein environment was instead described using the AMBER force field. To account for the dynamic electron correlation, the model equilibrium CASSCF/AMBER geometries and wavefunctions were used for single-point multiconfigurational second-order perturbation theory (CASPT2) calculations with a two-root state average zeroth-order wavefunction 39 . The Δ E values are computed at the CASPT2//CASSCF/AMBER level. The transition states controlling the thermal isomerization were located via restricted-step rational-function-optimizations 40 at the CASSCF/AMBER level. The corresponding E a T values were computed at the CASPT2//CASSCF/AMBER level. See the SI Appendix for further details.