Российский физиологический журнал им. И.М. Сеченова, 2020, T. 106, № 4, стр. 421-435

DISCRETE “DARK” EVENTS IN RODS AS PROXIES FOR THERMAL ACTIVATION OF RHODOPSIN

In the late 1970’s, it became possible to “see” single-rhodopsin events electrophysiologically, taking advantage of the powerful molecular amplification of phototransduction in dark-adapted rod cells. The suction-electrode technique developed by K.W. Yau et al. [12], inspired by E. Neher and B. Sakmann [13], allowed recording the light-sensitive current of single rods essentially free from confounding effects of rod-rod coupling and voltage-sensitive channels. The first recordings were from the sturdy rods of the cane toad, Bufo marinus¹¹. Under very dim background illumination, discrete current bumps of fairly standardized shape and size could be discerned, Poisson-distributed in time, as expected from random arrival and absorption of photons [14]. The size of these quantal responses (SQRs) was ~1 pA at peak, representing a ~5% decrease of the circulating current in a dark-adapted rod. During the following decades, the molecular events that shape most aspects of rod and cone responses to light have been clarified in considerable detail through a fruitful interaction of biochemical and electrophysiological studies. Precisely how the fairly reproducible SQRs are generated has been one of the most challenging questions. Full consensus has not yet been reached, but the crucial variables of the amplification [15–18] and termination reactions are now known with reasonable quantitative precision [19–25].

D.A. Baylor et al. [26] found that even in absolute darkness there still occurred occasional discrete current bumps indistinguishable from the SQR. The obvious hypothesis was that these “dark events” originate at the same point as the SQR, at the very input to the amplification cascade, i.e., in the rhodopsin molecule, rather than arising e.g. from bursts of fortuitous synchronous activation of large numbers of intermediates in the cascade (G-proteins or PDE molecules). Thus they seemed to offer an exceptional window into the “dark life” of the rhodopsin molecule.

This notion faced a serious problem, though. From the temperature-dependence of the rates of dark events, D.A. Baylor and colleagues had determined an activation energy of ~22 kcal mol^–1 [26]. They stated that this value “seem(s) consistent with isomerization of the 11-cis retinal chromophore as the mechanism for thermal activation”, because it was close to that determined for thermal isomerization of the chromophore in aquaeous digitonin (24.5 kcal mol^–1) [27]. About the same time, however, A. Cooper [28] showed that the ground-state energy of the early photobleaching product, bathorhodopsin, is 35 kcal mol^-1 higher than that of rhodopsin, and argued for a 45 kcal mol^–1 energy barrier for the ground-state (thermal) 11-cis → all-trans transition. Neither did D.A. Baylor et al. cite two earlier studies of frog rhodopsins in solution: R.J. Lythgoe and J.P. Quilliam [29] had estimated an activation energy of 44 kcal mol^–1 for thermal bleaching, and R.C.C. St. George [30] had arrived at a photoactivation energy of 48.5 kcal mol^–1 based on the longest wavelength (590 nm) where photon energy alone sufficed for activation (see below). I shall hereafter denote by E_a and E_aH the energies for activation by light and by heat, respectively – estimates as well as the underlying entities, even though this may occasionally cause some confusion.

The discrepancy between apparent activation energies for photic and thermal activation caused much speculation on differing molecular routes, stressing that there is actually no reason why E_a and E_aH should be the same. R.B. Barlow and colleagues [31] first came up with a testable hypothesis. They proposed that thermal events originate in a small (<0.01%) subpopulation of rhodopsin molecules where the Schiff-base linkage between chromophore and opsin is unprotonated. According to their molecular modelling, this could lower the activation energy for ground-state 11-cis → all-trans isomerization by about half. Experimental testing is straightforward in principle, since the proportion of rhodopsin molecules with unprotonated Schiff base must increase with alkalinization and decrease with acidification in predictable manner. In Limulus, they found that correlated decreases in pH and in the activity of optic nerve fibers could be induced by efferent stimulation. However, the metric they used, spiking in the afferent nerve, is at least twice removed from the rhodopsin molecule. First, decreased activity may result from some factor other than a decreased rate of thermal “quantum bumps” in the photoreceptors (e.g. acidification as such). Second, “quantum bumps” in Limulus, originally reported by S. Yeandle [32], do not bear a straightforward relation to activation of single rhodopsin molecules [33]. Subsequent experiments on toad rods [34] and salamander cones [35] indicated no relevant effect of changing pH (intra- and extracellular) on thermal event rates. The deprotonation hypothesis could be rejected.

A second hypothesis disrupting the connection between SQR-like “dark” events in rods and intrinsic properties of the rhodopsin molecule has been advanced by I. Bókkon and R.L.P. Vimal [36]. They proposed that the events are in fact responses to real photons, “biophotons”, emitted by the retinal tissue. However, V.I. Govardovskii and coworkers [37] showed by direct measurements that biophoton emission rates in frog and sterlet retina are >100-fold too low to account for the discrete rod events recorded in the same species.

A third possibility that could cast doubt on the use of light-like noise as a measure of thermal rhodopsin activation is if cannot, after all, be distinguished from noise triggered at a later stage of the phototransduction cascade. A recent study [18] shows that the first amplification step, the number of G-proteins activated per activated rhodopsin, in dark-adapted mouse rods is only 12–14 rather than the commonly quoted number ~100, and that PDE-initiated events may be more similar to the SQR than previously thought. This could explain the difficulty of separating SQR-like events from “continuous noise” in mouse rods [38], but quantitative relations are likely to vary between species. In many species SQRs are much more distinct from continuous noise [39].

ESTIMATES OF THERMAL AND PHOTIC ACTIVATION ENERGY RECONCILED

The problem of the 2-fold discrepancy between estimates of photic [28, 40, 41] and thermal [26] activation energies remained unresolved for more than 20 years. The Gordian knot was cut by P. Ala-Laurila and collegues [42] who argued that the low thermal estimate was no more than an analytical artifact. R.J. Lythgoe and J.P. Quilliam [29] had already in 1938 considered whether rhodopsin activation kinetics would be affected by complexities dealt with in a recent treatise by C.N. Hinshelwood [43], and R.C.C. St. George [30] and P.R. Lewis [44] had applied some aspects of it, but only in [42] were its full implications developed. Briefly, the thermal energy distribution of complex molecules like rhodopsin, or even the 11-cis retinaldehyde chromophore, cannot be described by simple Boltzmann statistics, but must take into account the internal energy of the molecule present in a large number of vibrational modes. The chromophore alone consists of n = 49 atoms and has 3n – 6 = 141 kinetic degrees of freedom. The number of vibrational modes n (≤141) that actually contribute towards 11-cis → all-trans isomerization in a given opsin environment is unknown and could depend e.g. on the amino acid residues around the chromophore pocket. The predicted effect of temperature on the fraction F of molecules exceeding E_aH, and hence on dark event rates, will depend strongly on n, implying that the temperature-dependence of D.A. Baylor et al’s [26] data shown as an Arrhenius plot in Fig. 2 will yield very different values of E_aH depending on what n value is assumed. Based on Boltzmann statistics, the slope of the straight line indeed gives E_aH = 21.9 kcal mol^–1, but based on Hinshelwood statistics it may give, for example, E_aH = E_a = 44.3 kcal mol^–1 (where E_a is the estimated photoactivation energy of Bufo marinus rhodopsin in [45]) if n = 79, or E_aH = = 34.3 kcal mol^–1 if n = 44. The last example was given by P. Ala-Laurila et al. [42] to show the robustness of their model against a possible 5–10 kcal mol^–1 difference between the electronically excited state and the peak of the ground-state energy barrier (E_a – E_aH) as suggested by molecular modelling [46, 47]. Obviously, their approach did not enable actual estimation of E_aH, but it removed the supposed incompatibility with estimates of E_a, allowing that thermal activation may follow the same molecular pathway as photoactivation, starting from 11-cis → all-trans isomerization of the chromophore.

This is consistent with current molecular understanding. The energies of the ground-state barrier for thermal activation and the electronically excited state induced by photon absorption are expected to be close in view of the femtosecond transition from the latter to the earliest identified ground-state photoproduct [46–48]. Quantum chemical modelling [49] suggests that the transition state mediating thermal activation has the same electronic structure as the excited state, manifesting intrinsic chromophore features associated with the existence of a conical intersection between the ground and excited states (cf. [50]). Importantly, this gives a theoretical, molecular-level foundation for a correlation between the wavelength of maximum absorbance λ_max and the rate of dark events k (“the Barlow correlation”). The model has recently been applied to the opsins of the endemic cottoid fishes of Lake Baikal in an attempt to identify specific amino acid residues that may regulate both spectral and thermal properties in connection with the blue-shift of corresponding pigments between species with increasing habitat depth [51].

EMPIRICAL TESTING OF “BARLOW’S HYPOTHESIS”

The basically simple idea that long-wavelength sensitivity should correlate with high thermal activation rates first appears in the literature in a brief comment by H. de Vries in 1949 [52]: high sensitivity to long light wavelengths (low-energy photons) entails a low energy barrier for activation, and this will imply a high probability that the barrier be surmounted by molecular thermal energy alone. The conceptual relations between activation energy E_a, spectral absorbance (captured by λ_max) and the fraction F of rhodopsins with thermal energy exceeding E_a are shown by the scheme in Fig. 1. H.B. Barlow [53] gave the idea its classical formulation, proposing it as a teleological explanation for the ubiquitous blue-shift of night vision compared with daylight vision (the Purkinje shift). He pointed out that the shift does not increase photon catch at night, since star- and moonlight is in fact somewhat more “reddish” than daylight, but could be useful as a means of decreasing thermal noise.

H.B. Barlow’s hypothesis inspired several experimental studies of dark noise in rods with different spectral sensitivities even while the conflict between the estimated values E_aH and E_a remained unresolved. The first of these, on the blue-sensitive (433-nm) “green” rods of toad, provided a disappointment, as estimated dark event rates per pigment molecule were more than 4 times higher than in the regular 503-nm rods ([54] c.f. however [39] and below). We now know that the pigment of these blue-sensitive rods is not a rod rhodopsin (Rh1), but a cone (SWS2) pigment, albeit with a stabilizing mutation [55, 56]. Luckily, studies on bullfrog rods with A1 (502 nm) and A2 (525 nm) pigment [57] and sturgeon (A2) rods with λ_max = 538 and 549 nm [58] were more encouraging, showing a clear correlation between long-wavelength sensitivity and high rates of SQR-like dark events. This kept interest in Barlow’s hypothesis alive.

THE PHOTOACTIVATION ENERGY E_a

In the general scheme (Fig. 1), the activation energy E_a is in a pivotal position, being the determinant of both spectral absorbance and rates of thermal events. H.B. Barlow [53] initially assumed that E_a would be equal to the photon energy at the wavelength of maximum absorption or “maximum visibility” (E_a = hc/λ_max), but he envisaged improvements by taking into account, inter alia, “Lewis’s further development of Stiles’s theory”. One such development was the realization that E_a corresponds to the photon energy not at λ_max, but at some wavelength λ₀ > λ_max, recognizable as the longest wavelength where activation can still occur without supplementation by thermal energy [30, 44, 59]:

where C = λ_max/λ₀ < 1 may or may not differ between pigments (see below). All this built on Stiles’ (1948) “physical interpretation of the spectral sensitivity curve of the eye” [59], which also provided a rationale for estimating E_a through the effect of warming on long-wavelength sensitivities. At wavelengths λ corresponding to photon energies hc/λ > E_a, the activation probability should not depend on the thermal energy of the rhodopsin molecule, but beyond a critical wavelength λ₀ the insufficient photon energy requires added thermal energy to activate the pigment. In this domain, raising temperature will therefore increase the probability of activation (i.e., increase sensitivity), the more so the longer the wavelength. In electrophysiological experiments there is no upper limit for the range over which this effect can be measured other than the power of the light source, enabling fairly accurate determination of E_a (and λ₀) [60–64, 45, 39 ].

Summarizing results of their measurements on 12 photoreceptor species (both rods and cones with both A1 and A2 pigments, and two pigments in crustacean rhabdoms), Ala-Laurila et al. (2004) [64] could not confirm the simple relation expressed by eqn. (1), but still found a significant correlation between E_a and 1/λ_max described by the following linear regression equation:

The coefficient of determination was only 0.73, however, implying that 27% of the variance remained unexplained variation around the regression line. Moreover, the line itself is less steep than expected.

The data underlying eqn. (2) had been obtained by a combination of microspectrophotometry and transretinal ERG recording potentially susceptible to several sources of error, which moreover may differ between species. D.G. Luo et al. [39] reexamined the E_a – 1/λ_max relation in 7 species of vertebrate rods and cones, recording spectral sensitivities with the more precise suction-pipette technique. They did find a close agreement with eqn. (1), with λ₀ as a remarkably constant multiple of λ_max across species (mean ratio λ_max/λ₀ ± SD = 0.84 ± 0.01). Strict comparison between [64] and [39] is largely impossible, though, as they are based on mainly non-overlapping samples of photoreceptor species. For the Bufo marinus “red” rod included in both, the reported values differ significantly (E_a = 44.3 kcal mol^–1, λ_max = 503 nm, λ_max/λ₀ = 0.78 in [64] versus E_a = 48.0 kcal mol^–1, λ_max = 500 nm, λ_max/λ₀ = 0.84 in [39]).

While there is no doubt of the superior quality of the suction-pipette recordings in [39], the differences cannot be lightly dismissed as being due to poor quality of the ERG data. The presence of true variation in λ_max/λ₀ between species is suggested by a comparison of the E_a values of two A1–A2 pigment pairs (Fig. 3 based on ERG, cf. Fig. 3A in [64]). One pair consists of L-cone pigments of juvenile (A2 with λ_max = 629 nm) and adult (A1 with λ_max = 562 nm) Rana temporaria (original data from [61]), the other pair of rod pigments from adult Rana catesbeiana (A2 pigment with λ_max = 525 nm and A1 pigment with λ_max = = 502 nm; original data from [62]). The two straight lines plot eqn. (1) with λ_max/λ₀ = 0.89 for the R. temporaria L-cones (blue) and 0.81 for the R. catesbeiana rods (green). The good fit of the two different lines to the respective pair of points suggests three things. First, upon a chromophore switch in the same opsin, changes in E_a and λ_max may indeed be tightly coupled as described by eqn. (1). Second, random variation of the ERG-based estimates seems to be fairly small, because otherwise one would expect larger variation of λ_max/λ₀ within each pair. Third, as there were no clear sources of systematic error liable to differentially affect estimates for these two species of frogs, the ca 10% difference in λ_max/λ₀ between the two pairs appears significant. Pending new data, a cautious conclusion is that the relative shallowness of the regression equation (2) does reflect a real biological trend. One may hypothesize that the evolution of opsins that confer high long-wavelength-sensitivity has also involved selection against a “default” decrease in activation energy, as far as decoupling of the two by tinkering with the amino acid sequence is possible.

SYNTHESIS: RATES OF THERMAL ACTIVATIONS VS. SPECTRAL ABSORBANCE

From the viewpoint of visual function, what finally matters is the resultant relation between λ_max and the rate of randomly occurring thermal activations k (dashed arrow ③ in Fig. 1). P. Ala-Laurila et al. [42] compared the data then available for rods and cones with their model, where k was predicted by using the empirical equation (2) for the E_a – 1/λ_max relation, and the fraction of molecules with energy exceeding E_a was obtained from Hinshelwood’s distribution for n = 79 (the same value that made E_aH = E_a for Bufo marinus rods in Fig. 2). Fig. 4 plots the comparison as lg k against 1/λ_max for rods (A) and cones (B), reproduced from [42]. The solid lines show the model prediction, where the vertical positioning of each line is the only parameter freely fitted. This corresponds to fixing the “pre-exponential factor” in the Arrhenius equation, i.e., fixing the absolute rates of dark events (which are some three orders of magnitude higher in cones than in rods). The slopes of the lines provide an acceptable description of the admittedly sparse and scattered data for both rods and cones, in qualitative agreement with Barlow’s hypothesis. The prediction of Barlow’s original formulation (dashed line) is shown for comparison. The dotted lines show the “robustness test”, i.e. the model prediction assuming a 10 kcal mol^–1 difference between E_a and E_aH (whereby n = 44 is assumed, based on the fit to the temperature data in Fig. 2).

Again, D.G. Luo et al. [39] provided new data of reference quality against which the earlier results must be assessed. They mainly found a much tighter connection between lg k, 1/λ_max and theory. This might partly reflect the advantage of standardized protocols, avoiding the inter-laboratory variation affecting data assembled from many studies. However, precisely because of the persuasiveness of their elegant study, it seems important to keep up awareness that not everything can be so neatly wrapped up. By the examples in Figs. 5 and 6, I wish to emphasize that there exists real and substantial variation in dark event rates between rod rhodopsins that differ negligibly in λ_max. On a general level, this can be seen at a glance in Fig. 4 A from the scatter of k values near 1/λ_max ≈ 2 × 10⁶ m^–1, i.e., for rods having typical rod λ_max values ≈ 500 nm.

Fig. 5 shows dark current recordings from rods of two anuran species with roughly the same λ_max (502–503 nm). Panel A shows the iconic first example of dark rod events published by Baylor et al. (1980) [26]. In the 1050 seconds of recording from a Bufo marinus rod, a fair number of discrete “bumps” (some 20) can be identified by eye. This and similar recordings from 9 cells, analyzed by several different methods in their study, indicated 0.02–0.03 events per rod per second. Panel B shows a recording from an A1 rod of Rana catesbeiana [57]. In ca 20 minutes, not a single discrete bump can be seen (implying a rate of <0.0008 per second), nor did more sophisticated analyses indicate any. Two epochs of responses to light flashes delivering ca 1 photoisomerization on average are shown below the “dark” traces. The first of these was taken before all the dark epochs, the second between the two latter ones, showing that the absence of dark events is not due to SQRs being undetectable in this rod and recording. Such total silence in darkness was observed in 3 other rods out of 5 studied. In one rod a rate of 0.006 per second was determined from the current probability density histogram, and in a sixth rod recorded after the publication of [57], there was one event distinctly identified by eye in a ca. 1000 second epoch (implying 0.001 per second, or 2.5 × 10^–13 events per pigment molecule per second, R* s^–1). By any statistics, these results are incompatible with those from Bufo marinus, and even in the 2 out of 6 rods where any dark events at all were detected, it is doubtful whether they originate in the A1 pigment. Rana catesbeiana, as opposed to Bufo marinus, uses the inherently less stable A2 chromophore during the larval stage and even in the adult stage in the dorsal retina [65], but minimal amounts in other parts of the retina cannot be excluded. (It should be noted that SQRs produced by A1 and A2 pigments cannot be distinguished [66]). A natural hypothesis would be that the Rana opsin has evolved to limit noise when collaborating with A2, and that this results in exceptional stability of the A1 version of the pigment. Even a minute fraction of A2 remaining may significantly increase the rate of dark events. P. Ala-Laurila et al. [67] tested the relative noise contribution of A1 and A2 pigment systematically in experiments on rods of larval tiger salamander (Ambystoma tigrinum), where the native pigment is almost 100% A2. They measured changes in dark event rates upon replacing most of the A2 by A1, and plotted the rates as function of remaining A2 (Fig. 6). The dark events decreased linearly with decreasing A2 fraction, extrapolating to a rate of 0.0009 events per rod per second (~2.4 × 10^–13 R* s^–1) for zero A2 (pure A1). This suggests that A1 pigments of Ambystoma tigrinum (λ_max = 502 nm) and Rana catesbeiana (λ_max = 502 nm) have roughly the same, extremely low thermal activation rates, which are more than one order of magnitude lower than those in Bufo marinus (λ_max = 503 nm) ([26]: 1.2 × 10^–11 R*s^–1; [39]: 3.2 × 10^–12 R* s^–1) and in Bufo bufo (λ_max = 502 nm) ([68]: 5.4 × 10^–12 R* s^–1; [34]: 8.4 × 10^–12 R* s^–1). (All values refer to rates temperature-corrected to 21°C.) With respect to the existence of species differences, it may further be noted that Luo et al. [39] found a 16-fold higher temperature-corrected activation rate per molecule of visual pigment in mouse rods compared with Bufo marinus rods. A priori, toad and mouse pigments are expected to be roughly similar, and if anything, a possible difference would be expected to go in the opposite direction (mouse rod λ_max = 497 nm [69]).

Thus there can be little doubt that true interspecies variation exists around the average functional relationships of λ_max, k and E_a (Fig. 1). However, besides species differences and “acceptable” random differences between results from different laboratories, there are cases where results on the same photoreceptor species differ to a degree for which there is no clear explanation. It is tempting to think that there could be polymorphisms between populations/strains of the same species used in different laboratories. These amphibians, unlike laboratory rodents, have not been bred for global standardization of strains. For example, significant spectral polymorphism (not due to varying A2 admixture) has been found between individuals of the common frog Rana temporaria, with rod λ_max varying by 8 nm, far more than normal experimental variation [70]. Above, I have also referred to the significant differences in E_a and k values of Bufo marinus reported in [39] compared with those in [45] (E_a) and [26] (k).

A major unresolved conflict of this kind concerns the thermal stability of the blue-sensitive pigment of Bufo “green” rods (with λ_max ≈ 432–433 nm, [6]). Whereas G. Matthews [54] originally reported an unexpectedly high dark event rate, 4 times higher than in the “red” rods (see above), D.G. Luo et al. [39] found an event rate 100 times lower than in the “red” rods. The high value seemed to make sense when it was shown that the anuran green-rod pigment is an SWS2 cone pigment [55]. Cone pigments are generically less stable by 2–3 orders of magnitude [71–73, 39, 42 ], but the most short-wavelength-sensitive ones may approach the stability of rod pigments [74, 75]. The recent discovery of a stabilizing mutation in the anuran green-rod SWS2 pigment [56] might give reason to expect it to be particularly silent even compared with rod pigments, but really there is no easy explanation for the large discrepancy of the estimates in [54] and [39]. This question certainly merits further investigation.

TUNING OF SPECTRAL ABSORBANCE AND THERMAL STABILITY BY THE OPSIN

Over 30 years, a huge literature has accumulated on how amino acid residues in the opsin tune the chromophore for diffent spectral absorbances in single species and across species (e.g. [76–80]). On a general level, the results exemplify the “multiple realizability” of function in complex biological systems, meaning in this case that (for practical purposes) identical spectral absorbance can be achieved by different combinations of amino acids in critical positions. This also makes it seem likely that similar spectra can be associated with different thermal properties, and that mutations affecting spectral and thermal properties may be independent targets of natural selection. Recently there has been increasing interest specifically in the tuning of thermal stability [56, 81]. Earlier, N. Fyhrquist et al. [82] sequenced the rod opsins of the anuran species presented above as examples where the same λ_max is associated with different dark event rates: Rana catesbeiana (plus two other Rana species, as well as Xenopus laevis) vs. the two Bufo species. Although it was not possible to identify unique residues for stability, among sixteen non-conserved substitutions and six involving gain/loss of hydroxyl groups, a few clear contrasts between Bufo and Rana were found. Some of these were shared by the Rana opsins and the Xenopus laevis opsin (which also has to collaborate with the A2 chromophore). The resolution of such studies have improved since then. In [81], two residues were identified that explain at least some of the generically increased stability of rod compared with cone pigments, and in [56] a single threonine at position 47 was identified as responsible for the rod-like stability of the anuran (in contrast to the urodelan) blue-cone pigment (SWS2). The green rods equipped with this stabilized pigment allow frogs to make blue/green wavelength discriminations at the absolute threshold of vision [83].

These developments enable us to address new evolutionary questions. Does the rarity of strongly stabilized rod pigments like those of Rana catesbeiana and Ambystoma tigrinum (Figs. 5 and 6) indicate that pigment noise in e.g. Bufo and other typical vertebrate rods is already driven to an acceptably low level, given the presence of other noise sources? In other words, is the extreme silence of Rana and Ambystoma A1 pigments mainly a side effect of having to limit the noisiness of the A2 pigment? Could there be mutations making pigments with λ_max beyond the long-wavelength limit of ca. 630 nm found in nature stable enough to be useful? On a different track, it is interesting to follow the engineering of microbial (“type 1”) rhodopsins with the purpose of developing improved tools for optogenetics, for example, by moving their spectral absorbance further into the infrared [84–87].