Divergence of Melanopsin’s Intrinsic Properties and Downstream Actions

SUMMARY

Opsins are light-activated receptors. A basic question is how the molecular properties of an opsin shape its effects on the organism. In mammals, melanopsin regulates perception, cognition, physiology, and mood. Because these actions are vital, an international standard has been formulated to express light in terms of melanopsin stimulation. Short wavelengths are absorbed best by melanopsin and should drive its effects most strongly. We report that melanopsin’s cellular and behavioral effects show little or even opposite wavelength sensitivity in some regimes. This divergence stems from melanopsin’s persistent activity, which generates adaptation. Wavelengths absorbed strongly by melanopsin produce high persistent activity and adaptation, giving net desensitization, whereas wavelengths absorbed weakly can produce sensitization instead. We provide a quantitative model that captures these transformations. This work indicates how the downstream actions of a molecule may diverge starkly from its intrinsic properties.

INTRODUCTION

In the mammalian eye, melanopsin triggers electrical signals that influence perception and are especially critical for driving “non-image” visual functions, such as regulation of the pupil, circadian rhythms, sleep, and mood^1–3. Like other photopigments, melanopsin absorbs various wavelengths of light to different degrees^4–9. The axiom is that a photopigment confers its absorption spectrum on downstream processes. This is foundational for understanding light’s biological effects, designing research, and developing artificial lighting. Indeed, melanopsin-driven behaviors have been reported to mirror melanopsin’s absorption spectrum^10–12. These pioneering studies were conducted when melanopsin was understood to have one ground state and thus one activation spectrum.

In mammals, melanopsin is now recognized to have three highly stable and photoconvertible states^6–9:

$$\text{Melanopsin}\left(\text{R}\right)\rightleftharpoons \text{Metamelanopsin}\left(\text{M}\right)\rightleftharpoons \text{Extramelanopsin}\left(\text{E}\right)$$

The ground state (“R”), signaling state (“M”), and third state (“E,” electrically silent) have distinct absorption spectra. In studies of purified melanopsin, their spectra peak at 467, 476, and 446 nm, respectively⁶; cellular experiments largely agree^7–9. Under short-wavelength illumination, the R and E states absorb photons best and therefore depopulate. Because the M state absorbs these wavelengths less effectively, it populates. Thus, at photoequilibrium, the M state dominates. It has high thermodynamic stability; after illumination ceases, it persists and drives cellular activation for minutes^7,8,13. Conversely, longer wavelengths produce a photoequilibrium in which the M state is depopulated. Different spectra of illumination may therefore produce different levels of melanopsin activation^7,8. Melanopsin tristability appears unique and highly conserved⁸. It has been recognized for a decade, but its consequences for the organism are largely unexplored.

Prior studies comparing the spectral sensitivity of melanopsin and its downstream processes were designed to minimize the effect of stimulus history. However, stimulus history is a natural influence on sensory responses¹⁴ and is now understood to have enduring actions on melanopsin signaling. For example, light drives mechanisms of negative regulation—for brevity, “adaptation”—that reduce melanopsin signaling. These include Ca²⁺ influx, phosphorylation, and arrestin binding^15–21. The availability of melanopsin molecules for photoactivation also changes with light history, as some accumulate in the signaling state and others become insensitive (e.g., by losing their photosensitive chromophores)^7,8,22,23. How adaptation shapes the expression of melanopsin’s intrinsic molecular properties is unclear^24,25.

Here, we query the nature of melanopsin signaling in the context of tristability and adaptation. Investigating this topic provides insight into melanopsin, the critical processes it supports, the influence of photopigment properties on the organism, and how lighting may affect health^2,26. We find that the interaction of tristability and adaptation shapes the manifestation of melanopsin’s effective spectral sensitivity downstream, in behavioral and cellular responses to light. Rather than short and long wavelengths always driving more or less activation, respectively, their effects appear equal or even inverted in some conditions.

RESULTS

Effective Spectral Sensitivity of a Melanopsin-Driven Behavior

We examined a behavioral response that depends on melanopsin and operates on timescales that facilitate quantitative study and comparison to melanopsin-driven cellular responses. This is the pupillary light reflex (PLR). It supports vision as light intensifies by sharpening the image and lowering sensitivity^22,27–29. To focus on melanopsin signaling, we evaluated pupil responses in mice with rod and cone phototransduction disabled (Gnat1^−/−; Gnat2^−/−)^30,31. We delivered pulses of light at a frequency that should allow light history to accumulate^7,15,16. We used brief pulses, approximating impulse stimuli, to focus on system dynamics rather than their convolution with ongoing stimulation^22,32. We delivered intensities that drove pupil constriction from threshold (the lowest tested intensity corresponds to melanopsin’s photon absorption rate when the sun is 2.5° above the horizon) to beyond the physiological range (melanopsin’s absorption reaches 8 log₁₀ photons μm⁻² s⁻¹ in daylight)³³.

We tested two wavelengths, matching their intensities at the retina by accounting for the eye’s transmission spectrum³³. The first is short (440 nm) and produces a photoequilibrium of melanopsin states in which the M state dominates; daylight and other kinds of white light produce similar photoequilibria⁷. The second is long (560 nm), which minimizes the M state at photoequilibrium⁷ (Figure S1). Exact wavelengths and all details for specific experiments are given in Table S1.

We constructed intensity-response (I-R) relations for short- and long-wavelength light (Figure 1A–D). After prolonged darkness, all melanopsin holopigment (opsin bound to chromophore) should be in the R state, such that sensitivities to light of these wavelengths should have a ratio of ~10.7 (440 and 560 nm; Methods)⁷. At the foot of pupil constriction's I-R relation (threshold), we measured a short:long sensitivity ratio of 7.5 (95% confidence interval, CI, of 5.1–9.0; Figure 1E). This smaller value is likely due to prior light pulses that activated the pathway but did not evoke detectable constriction³⁵. As the I-R relations rise to their saturated maxima, they gradually converge and their response ratio approaches unity (Figure 1D–E). In this regard, the spectral sensitivity of melanopsin-driven pupillary constriction naturally diminishes with response saturation.

Figure 1. Melanopsin-driven behavior and neuronal activity diverge from melanopsin’s spectral sensitivity.

A. The pupil in darkness and following pulses of light (stimulus monitor at the top) delivered to the contralateral eye. Here and in subsequent plots, short and long wavelengths are designated by blue and red coloring, respectively. To isolate melanopsin signaling, phototransduction is genetically inactivated in rods and cones. Single trials from one example mouse shown.

B. Quantification of pupil diameter for single trials from the animal in panel A. Upward deflections indicate constriction. The stimulus monitor is at the top.

C. Intensity-Response (I-R) relations for peak constriction (normalized to maximum). Each curve shows data from one animal (mean of 3 sessions). The short-wavelength stimulus produces a region of pronounced negative slope over higher intensities. Open circles mark the example animal of A and B.

D. Mean I-R relations for peak constriction (normalized to maximum mean constriction). The repeated highest intensity is shown on a separate axis to the right. Here and in H, open circles indicate regions of negative slope (responses at least 5% smaller than a prior response). Sigmoid fits are to the region of positive slope. Error bars are standard error of the mean.

E. Left, Sensitivity of pupil constriction (mm photons⁻¹ μm²) under short or long wavelength, normalized to the maximum for the former and plotted against cumulative light history. If long-wavelength sensitivity followed the prediction from melanopsin's dark-adapted (R state) absorption spectrum (‘Abs. Spec.’), it would fall on the gray dashed line. The prediction for melanopsin's E state is even lower and is not shown. The first two stimuli are not shown, being consistently below threshold. Individual trials (dots) are averaged (diamonds) and interpolated (lines, with 95% confidence intervals; see Methods). Right, Short:long sensitivity ratio with confidence intervals. With cumulative light history, the spectral sensitivity shifts away from the prediction from melanopsin's absorption spectrum (~1) to equalization (0).

F. Bottom, Patch-clamp recordings of voltage from two example intrinsically photosensitive retinal ganglion cells (ipRGCs, all of the M1 type in this and subsequent figures; 35 °C). Top, Timing of the light pulses.

G. I-R relations of peak subthreshold voltage (normalized). Note the region of pronounced negative slope for that obtained with short-wavelength light. Open circles mark example cells in F.

H. As in D, but for peak subthreshold voltage.

I. As in E, but for peak subthreshold voltage. For each plot, the response to the first pulse was undetectable and not shown.

See Table S1 for additional methodological details and Table S2 for statistical results. Antagonists of synaptic transmission were included in all ex vivo experiments.

We found an additional drive toward spectral equalization. Melanopsin's absorption spectrum indicates that short-wavelength pulses of increasing intensity should activate larger fractions of melanopsin to evoke stronger pupil constrictions, up to saturation^10,28,29. Instead, constriction usually peaked at an intermediate intensity and declined thereafter, producing an unusual region of negative slope at the higher end of the intensity-response (I-R) relation (Figure 1A–D). Repeating the highest intensity caused constriction to decline even further, to 0.70 ± 0.21 of the maximum (mean ± SD, 3 sessions for each of 4 animals). This indicates adaptation (i.e., negative regulation) because the same stimulus produces an attenuated response. By contrast, long-wavelength pulses of increasing intensity produced increasing pupillary constriction, and repeating the brightest pulse tended to give the same response (0.97 ± 0.06 of the maximum; p = 2 × 10⁻⁴, Nested Wilcoxon rank test³⁴; Figure 1A–D). Indeed, for the last pulse tested, absolute pupil constrictions were similar between wavelengths (0.99 ± 0.31 and 1.13 ± 0.22 mm for short and long, respectively; p = 0.27). Thus, when comparing across these lighting conditions, pupil constriction depends far less on wavelength than expected from melanopsin's absorption spectra.

Effective Spectral Sensitivity of Melanopsin-Driven Voltage Changes

To investigate where the effective spectral divergence between melanopsin and pupillary constriction originates, we considered that M1 intrinsically photosensitive retinal ganglion cells are key mediators of this process (hereafter, “ipRGCs”)^1,36–38. These neurons sense light using melanopsin and by receiving synaptic input driven by the classic rod and cone photoreceptors¹. We examined responses of ipRGCs in the ex vivo mouse retina near body temperature. We preserved and isolated their melanopsin-driven responses by using perforated-patch electrophysiological recording and by providing antagonists of synaptic transmission, respectively^7,8,39. We focused on the subthreshold membrane voltage because of its relatively straightforward dependence on melanopsin phototransduction, as compared to spikes. For comparison with pupillary measurements, we delivered the same stimulus to the retina, accounting for the eye’s transmission spectrum³³, except for use of a brighter last pulse to probe a broader range of intensities.

The subthreshold photovoltage of ipRGCs was more sensitive than pupil constriction, likely reflecting downstream thresholding of ipRGC signals for the latter³⁵. Like pupillary constriction, photovoltage was more similar between wavelength conditions than expected from melanopsin’s absorption spectra. Over higher intensities, photovoltage I-R relations exhibited a stronger region of negative slope for short- than long-wavelength light (p = 0.0082; Wilcoxon rank-sum test for unpaired data unless otherwise noted; Methods). At the highest intensity tested, photovoltages were similar across these wavelengths (36.8 ± 10.7 and 39.2 ± 6.0 mV, p = 0.58). Turning to the short:long sensitivity ratio, it was 2.74 (95% CI: 1.84–3.98) at threshold, consistent with photovoltages appearing after some light history. With additional light history, the ratio fell gradually to 0.85 (95% CI: 0.77–0.93; 6 and 7 cells; Figure 1F–I). Therefore, from threshold to saturation of ipRGC photovoltage, the difference between short- and long-wavelength conditions progressively diminishes. It then inverts, which is unexpected from melanopsin's absorption spectra.

Effective Spectral Sensitivity of Melanopsin Phototransduction

To examine melanopsin phototransduction directly, we clamped the membrane voltage to measure current through the cascade’s TrpC6/TrpC7 cation channels^8,29,40 and to suppress voltage-sensitive channels. We tested both wavelengths in each cell, examining all four combinations of pulse sets across cells (long-long, long-short, short-short, and short-long; Figure 2A–C and Figure S2). Doing so shows whether wavelength-specific response properties are robust to light history, and controls for cellular heterogeneity^41,42. These extended experiments require exceptional stability, so we worked at room temperature, increased inter-stimulus intervals to accommodate the slower kinetics at this temperature^22,41, and focused on the upper end of the intensity range.

Figure 2. Melanopsin phototransduction diverges from melanopsin’s spectral sensitivity.

A. Example ipRGCs in voltage clamp (−80 mV) and given intensifying light pulses of either short (left; blue) or long wavelength (right; red). Dotted lines (gray) mark the pre-stimulus baseline.

B. Intensity-response (I-R) relations for peak transient current for short and long-wavelength pulse sets. Open circles mark example cells of A.

C. Mean I-R relations for peak current, normalized within each wavelength to the maximum mean response across cells. Open circles mark regions of negative slope (responses at least 5% smaller than a prior response). Sigmoid fits are to the region of positive slope. Error bars are standard error of the mean.

D. Left, Sensitivity (pA photons⁻¹ μm², normalized) to the set of short- and long-wavelength pulses plotted against cumulative light history. Measurements from individual cells (points) are averaged (diamonds) and interpolated (lines flanked by 95% confidence intervals; see Methods). If long-wavelength sensitivity followed the prediction from melanopsin's dark-adapted (R state) absorption spectrum (‘Abs. Spec.’), it would fall on the gray dashed line. Right, Short:long sensitivity ratio (thick line flanked by 95% confidence intervals). Spectral sensitivities shift from the log ratio predicted by the absorption spectrum (~1) to equalization (0) and then inversion (< 0). Response saturation alone (‘Sat.’; orange dashed line; see Methods) produces equalization, but not inversion.

In A, 10-mV hyperpolarizing steps (not shown) were given periodically to probe recording parameters and the resulting capacitance transients are often visible. See Table S1 for additional methodological details and Table S2 for statistical results. Antagonists of synaptic transmission were included in all experiments.

Overall, melanopsin photocurrent resembled photovoltage and pupil size. Analyzing the first relations for all cells (set 1, initial condition of dark adaptation), we found that negative slope over high intensities was stronger for short- than long-wavelength pulses (p = 6.9 × 10⁻⁶). At the highest intensity tested, photocurrents were indistinguishable across wavelengths (158 ± 56 and 160 ± 82 pA, p = 0.86; 20 and 15 cells for short and long wavelengths, respectively). The short:long sensitivity ratio began at 10.99 (95% CI: 8.18–15.17), near expectation from melanopsin's absorption spectrum (Methods). With cumulative light history, the ratio fell to a minimum of 0.65 (95% CI: 0.57–0.71; Figure 2D). This is the monotonic progression toward spectral equalization, followed by the unexpected inversion where the long-wavelength condition ends with higher effective sensitivity.

Comparing two sets of I-R relations within cells provides a similar view. For instance, negative slope was pronounced for short-wavelength light whether it was used for the first set, second set, or both sets of pulses (Figure 3A–B and Figure S2). Likewise, negative slope was milder or absent for long-wavelength light whenever it was used (Figure S2). Therefore, the divergence of effective spectral sensitivity between the melanopsin molecule and its phototransduction cascade is robust to light history and cellular heterogeneity.

Figure 3. Dependence of melanopsin phototransduction on persistent activity and adaptation.

A. Example ipRGC given two consecutive sets of intensifying light pulses of long or short wavelength. The cell received an additional pair of high-intensity, short-wavelength pulses (black arrowheads) with an intervening long-wavelength pulse to suppress persistent activity. During this long-wavelength suppression pulse, the current was initially large and then declined, evincing adaptation (connected red arrows; observed in 15 additional cells). The transient response amplitude was largest at an intermediate intensity (1), smaller at the repeated highest intensity (2), and recovered by the long-wavelength period (3). Dotted lines mark the pre-stimulus baseline (gray), magnitude of the largest persistent activity evoked by long wavelength (red), and that evoked by short wavelength (blue).

B. Intensity-response relations for peak transient (top) and persistent current (bottom) for the stimulus in A. Open circles mark the example cell.

C. Excerpts of persistent currents from panel A, measured in a 3-s window prior to the next light pulse. “After Supp.” is after the long-wavelength pulse that suppressed persistent activity.

D. Left, Peak transient photocurrents from short-wavelength, second pulse sets. Shown are the maximum response to any pulse before long-wavelength suppression (1) and the response to the pulse preceding suppression (2) or after suppression (3). Solid and dashed lines indicate cells for which the first pulse set was short- or long-wavelength, respectively. Open circles indicate the example cell in A. Right, The same transient photocurrents normalized within-cell to the pre-suppression transient response.

E. Repeated, saturating pulses of short-wavelength light with an intervening period of long-wavelength light to suppress persistent activity. Inset, Expanded view of the two short-wavelength pulses bracketing the long-wavelength pulse. Dashed lines are for comparison of response magnitudes.

F. Continuation of the experiment for the cell in E (<1 min delay). Persistent activity is ongoing. Long-wavelength pulses, calibrated to be sub-saturating, are delivered with an intervening short-wavelength pulse. Inset, Expanded view of the two long-wavelength pulses bracketing the short-wavelength pulse. Dashed lines are for comparison of response magnitudes.

G. Response magnitudes, normalized to the first, for all 12 cells tested with short-wavelength pulses (example shown as open symbols).

H. Same as G but for long-wavelength pulses. Red and blue shadings indicate timings of intervening long- and short-wavelength pulses, respectively.

In A, 10-mV hyperpolarizing steps (not shown) were given periodically to probe recording parameters and the resulting capacitance transients are often visible. See Table S1 for additional methodological details and Table S2 for statistical results (* p < 0.05, ** p < 0.01, *** p < 0.001). Antagonists of synaptic transmission were included in all experiments.

Persistent Activity Produces Adaptation

Examination of melanopsin photocurrent suggests how sensitivities under short- and long-wavelength conditions come to appear similar. Short-wavelength pulses of increasing intensity evoked cumulative increases in persistent photocurrent—that which long outlasts the stimulus (also see Figure S3A)^7,8. At higher intensities, this increase was accompanied by a drop in the transient photocurrent, which is reflected in the I-R relation’s region of negative slope (Figure 2A–C). By contrast, long-wavelength pulses caused little accumulation of persistent photocurrent and the transient photocurrent magnitude increased to a plateau (Figure 3A–D). These observations suggest that persistent activity shapes the I-R relation.

To test this idea, following the second set of short-wavelength pulses, we delivered a period of long-wavelength light (100–140 s) to depopulate the M state and thus suppress persistent current (Figures 3A, 3C, S1, and S3A). This suppression increased the transient current evoked by a subsequent short-wavelength pulse (potentiation of 236 ± 95%, n = 8 cells; Figure 3D; also see Figure S2D). Suppression itself drove adaptation, as expected^15,16,41 and as indicated by the decay of photocurrent during the period of long-wavelength illumination (Figure 3A and Figure S3A). Nevertheless, the net effect of suppression is sensitization. These experiments indicate that short-wavelength light produces high persistent activity, which drives adaptation that in turn reduces responsiveness.

In separate experiments, we delivered a sequence of identical, short-wavelength pulses that should each drive a saturated response, reaching a photoequilibrium with high occupancy of melanopsin’s signaling M state (Figure 3E and G; also see Figure S1). The first pulse induced a transient response and persistent activity. The following pulses evoked transient responses, riding atop persistent activity, that were attenuated or undetectable (9 and 3 of 12 cells, respectively). We then delivered a period of long-wavelength light, which suppressed persistent activity (10 of 12 cells). A subsequent short-wavelength pulse produced a larger response; it also restored persistent activity (in those 10 cells).

We then delivered long-wavelength pulses that were each calibrated to partially suppress persistent activity, so that melanopsin fractions should proceed incrementally toward a photoequilibrium where M-state occupancy is low (Figure 3F and H; also see Figure S1). Each decrement of persistent activity was accompanied by a larger transient response to the next pulse (Figure 3H). A subsequent short-wavelength pulse restored persistent activity, and the transient response to the next long-wavelength pulse was diminished (3 of 3 cells). Persistent activity appears to drive graded levels of adaptation in ipRGCs (also see Figure 2A and Figure S2).

Adaptation Shapes Persistent Activity

Having found evidence that persistent activity produces adaptation, we asked if adaptation acts in turn on persistent activity. We previously observed that Ca²⁺ influx drives adaptation in melanopsin phototransduction, such that its acute removal increases the sensitivity and magnitude of light responses in ipRGCs¹⁶. Therefore, we removed extracellular Ca²⁺ and asked if persistent activity increased. We continued to use voltage clamp and synaptic antagonists to isolate the effect of this manipulation on phototransduction¹⁶. Persistent activity doubled in size (increasing from 8.5 ± 4.8 to 16.9 ± 7.2 pA, 9 cells; Figure 4A, B and D). For an internal control, we measured dim-flash responses in the same experiment. A dim-flash response is in the linear range of phototransduction; its waveform reflects that of a single-photon (single-melanopsin) response and its amplitude, divided by the flash intensity, gives a measure of absolute sensitivity (S_F, in pA photons⁻¹ μm²)^22,43. Ca²⁺ removal did not change the initial trajectory of these responses, indicating little or no dependence of activation on Ca²⁺ influx. However, the response in 0 Ca²⁺ rose further to reach a ~2-fold larger peak, indicating that Ca²⁺ influx usually curtails activation, as expected from a mediator of adaptation (i.e., driving a reduction of activity; Figure 4A, C and E). These experiments indicate that adaptation strongly regulates persistent activity (also see Figure S3).

Figure 4. Adaptation shapes persistent activity.

A. Protocol to evaluate the effect of Ca²⁺ removal on persistent activity (top) and its effect on an example cell (bottom; same cell in B and C). A pre-induction period is followed by induction (short-wavelength pulse) and then suppression (long-wavelength pulse) of persistent activity. In each period, extracellular Ca²⁺ is acutely removed and then returned, and the effect on the holding current (which includes any persistent activity) assessed. In the pre-induction and post-suppression periods, dim-flash responses are assessed. 10-mV hyperpolarizing steps (not shown) were given periodically to probe recording parameters and the resulting capacitance transients are sometimes visible.

B. Effect on the persistent current of removing and replenishing extracellular Ca²⁺.

C. Effect on the dim-flash response of extracellular Ca²⁺ manipulations in pre-induction (left) and post-suppression (right) periods.

D. Left, Persistent activity (the change in holding current after the induction pulse) increases in 0 Ca²⁺. Right, 0 Ca²⁺ increases the holding current before induction and after suppression of persistent activity, but this increase is greater during persistent activity, indicating an effect on persistent activity itself.

E. In the same sample as D, dim-flash sensitivity (S_F) increases in 0 Ca²⁺ whether in the baseline or post-suppression periods (left and middle, respectively). The effect of 0 Ca²⁺ is similar between these periods (right).

See Table S1 for additional methodological details and Table S2 for statistical results (* p < 0.05, ** p < 0.01, *** p < 0.001). Antagonists of synaptic transmission were included in all experiments.

Transformations of Melanopsin Signaling by the Interaction of Persistent Activity and Adaptation

Persistent activity appears to produce adaptation, and adaptation appears to regulate persistent activity. We sought to predict how this interaction shapes the effective spectral sensitivity of melanopsin signaling, iterating across three models of increasing complexity and accuracy. We began with our existing computational model of melanopsin’s three photoconvertible states (R, M, and E; Figure 5A and Figure S1)^6,7. This model produces persistent activity, in the form of M state occupancy that remains after illumination ceases. It also incorporates some adaptation. For example, as molecules are activated to the M state, fewer are available for activation. Moreover, long-wavelength light increases the fraction of E, which reduces responsiveness to wavelengths that this state absorbs more poorly than the R state. This model captures the known photoconversions of melanopsin.

Figure 5. Gain shifting is required for steady irradiance encoding.

A. Schematics of melanopsin models without (top row) and with (bottom row) adaptation implemented as shifts to low-gain states (R’, M’, and E’). Horizontal transitions are light-dependent, while vertical transitions are light-independent. “Melanopsin activity” is the fraction of molecules in the M state (f^M) for the no gain-shifting (no GS) model. For the GS model, it is $f^M+0.038f^{M^{\text{'}}}$ (see Methods).

B. Simulated melanopsin activity for a set of flashes varying in intensity (100 ms, 7–10 log₁₀ photons μm⁻² in 0.5-log₁₀ increments). The addition of gain shifting (bottom) allows for the activity to decay.

C. Without gain shifting, steps (3–8 log₁₀ photons μm⁻² s⁻¹ in 0.5-log₁₀ increments) do not produce steady activity that grades with light intensity. Flashes (200 ms, 7.5 log₁₀ photons μm⁻²) are provided for reference.

D. The protocol used in Figure 3E (see Table S1 for stimulus details). When persistent activity is saturated, subsequent pulses do not generate additional changes in melanopsin activity (top) unless gain shifting is implemented (bottom).

We found four model behaviors that indicate a need for additional adaptation. First, flashes produce stepwise increases in the M fraction and thus phototransduction (Figure 5B). Measured flash responses differ, having a large transient component²² (Figure 4). Second, different intensities of sustained light produce the same steady-state level of phototransduction, which is set by the photoequilibrium fraction of M (Figure 5C and Figure S1). By contrast, measured step responses settle at steady levels that grade with light intensity^16,41. Third, activity during illumination is the same magnitude as that persisting in subsequent darkness (Figure 5B–C). Cellular measurements show that persistent activity is far smaller⁷ (Figure 2A–C and Figures S2 and S3). Fourth, pulses of light given during saturated persistent activity produce no response (Figure 5D). In actuality, “increment” responses can be driven atop saturated persistent activity⁷ (Figure 3E). These observations underscore the divergence between melanopsin’s photoconversions, which the model simulates, and their cellular effects.

We therefore added to the model a tier of adapted states (R’, M’, and E’; Figure 5A), which may be considered analogous to those produced by forms of adaptation like receptor phosphorylation and arrestin binding^17–21,44. Molecules in the M state drive phototransduction; those in the M’ state do so with low gain (little activation per molecule); and those in the remaining states do not (no activation per molecule). The magnitude of phototransduction increases with the fraction of molecules in M and M’. This simple, “gain shifting” (GS) model accounts for the four features of melanopsin phototransduction mentioned above (Figure 5B–D). First, flash responses are transient because M, once formed, transitions into the low-gain M’ state (Figure 5B). Second, during sustained illumination, melanopsin activity reaches steady levels that grade with light intensity (Figure 5C). This outcome depends on a dynamic equilibrium of melanopsin states. At low intensities, molecules in the M’ state are photoconverted into R’ and E’ at low rates, yielding little recovery to R and E. Because of this and the low photon absorption rate, photoconversions of R and E to the high-gain M state are sparse. On the other hand, at high intensities, the cycling of molecules from the low-gain M’ state to the high-gain M state (via R’/E’ and R/E) is faster. In other words, during illumination, melanopsin activity is greater in brighter light because the system is operating at higher, continually replenished gain—a process of “gain cycling.” Third, persistent activity is smaller than the preceding activity during illumination. This is because persistent activity arises entirely from the low-gain M’ state, while activity during illumination has gain cycling (Figure 5C). Fourth, light given during persistent activity drives gain cycling, producing increment activity (Figure 5D). In addition, the GS model mimics recovery of the saturated response following suppression of persistent activity (Figure 5D; see Figure 3 for experimental data). To summarize, simulating principal features of melanopsin phototransduction requires balancing persistent activity with adaptation.

Despite the GS model’s strengths, it misses aspects of the empirical I-R relations (Figure 6B–E): modeled persistent activity saturates at relatively modest intensities, negative slope at the top of the relation is mild or absent under short-wavelength illumination, the short:long sensitivity ratio diminishes with light history but does not invert, and response kinetics are oversimplified. Augmenting the GS model reproduces these features (Figure 6A–H). Adding loss and recovery of melanopsin (analogous, for example, to pigment bleaching/regeneration^20,22,23,45) gives appropriately graded levels of persistent activity. Allowing persistent activity to drive adaptation, thus reducing responses to additional illumination, produces negative slope under short- but not long-wavelength light (Figures 2–3; see Figures S2 and S3 for additional evidence that persistent activity drives adaptation). Incorporating an additional adaptation mechanism, based on total activity, improves response kinetics (Figure S6). This augmented GS (aGS) model fits the data well (Figure 6B–H and Figures S4 and S5). It is as accurate as any single ipRGC's measured I-R relation for predicting the population's I-R relations (Figure 6E and Figure S10). Accordingly, the aGS model recapitulates the short:long sensitivity ratio and its change with cumulative light history (from 7.9 to 0.57; Figure 6G–H). It also recapitulates shifts in spectral sensitivity previously measured in ipRGCs following conditioning with long-wavelength light⁷ (Figure S7). The need for complexity in the aGS model provides a rationale for the numerous mechanisms of adaptation observed in melanopsin signaling^{17,18,20–23,44,46–49} (Figures S5 and S6).

Figure 6. An augmented gain-shifting model predicts spectral divergence.

A. Schematic of the augmented GS model (aGS). Left, All states transition to a nonphotoconvertible state (O) that has minuscule signaling gain and gradually returns to R. Light gray arrows indicate O’s potential return to E and M (disabled here; see Methods). Right, Two gain controls accumulate through leaky integration of persistent activity, corresponding to the fraction of melanopsin in M’ (∫f^M’) and the summed activity of all states (∫Activity). A nonlinearity governs each control’s reduction of melanopsin signaling gain. See Methodsfor further details.

B. Model responses (GS: green; aGS: orange) inverted for comparison to example ipRGC photocurrents (gray; data are from Figure 2). Short- and long-wavelength pulse series are on the left and right, respectively, with light monitors at the top.

C. Short-wavelength intensity-response (I-R) relations for the GS (green) and aGS (orange) models, compared with those of ipRGCs (gray lines). See Figure S5 for other stimuli.

D. Same as C, for long-wavelength pulse series.

E. Loss (mean squared error between predicted and cellular I-R relations) with predictions from the GS model (green squares), aGS model (orange squares), or individual ipRGCs (gray dots). Black bars indicate the mean loss for individual cellular predictions.

F. I-R relations for short- (blue) and long-wavelength (red) pulse series predicted by the GS (left) and aGS (right) models. Open circles indicate regions of negative slope (responses at least 5% smaller than a prior response). Sigmoid fits are to the region of positive slope.

G. Simulated sensitivity (normalized to sensitivity at the first short-wavelength pulse) of GS (inset; green lines) and aGS (orange lines) responses to the short- and long-wavelength pulses (solid and dashed lines). Light history is expressed as a cumulative photon count. Predictions (gray dashed lines) are the short-wavelength sensitivity of GS and aGS models divided by the empirical short:long sensitivity ratio for melanopsin’s dark-adapted absorption spectrum (R state; 10.7; see Methods).

H. Short:long sensitivity ratios predicted by the GS model (green line) and aGS model (orange line) compared to the measured ratios from voltage clamp experiments (heavy black line, from Figure 2). Thin lines are 95% confidence intervals. With cumulative light history, relative spectral sensitivity shifts away from that predicted by melanopsin’s absorption spectrum (‘Abs. Spec.’; ~1) to equalization (0) and then inversion (< 0).

Transformations of Melanopsin Signaling under Naturalistic Illumination

We have used short and long wavelengths to probe the spectral sensitivity of melanopsin and its downstream actions, but common light sources have a broad spectrum. Thus, we tested xenon (Xe) light, which has a spectrum resembling that of daylight³³. We modified its spectrum to mimic filtering by the eye’s optics⁵⁰, matched its intensity to that of the short-wavelength light used in our other experiments, and delivered it to ipRGCs (Methods). The I-R relations were qualitatively similar to those elicited by short-wavelength light, including high persistent activity and a region of prominent negative slope at higher intensities (Figure S8). These features were robust, being similar for Xe delivered without ocular filtering, which enriches for short wavelengths⁵⁰ (Figure S8). Our aGS model simulates these effects of Xe light accurately (Figure S9). These observations indicate that persistent activity and adaptation shape melanopsin signaling under natural conditions.

DISCUSSION

We find that the spectral sensitivities of cellular and behavioral responses can diverge from the absorption spectrum of the photopigment that drives them. In the case of melanopsin, which drives many vital processes^1,2,26,51,52, this divergence results from an interaction of persistent activity and adaptation, both of which follow from light history^2,26,51. The interaction makes melanopsin-driven responses more agnostic to wavelength than melanopsin itself. This favors encoding of the overall light level rather than image details (i.e., spectral composition). Such downstream transformations of photopigment activity are likely to be important considerations in studies of the visual system, lighting design, and optogenetic applications.

Persistent activity reflects melanopsin’s stable activation^7,8. This stability has advantages; for example, it blurs spatiotemporal contrasts to favor encoding of the overall light intensity^7,8. It also can be problematic because light of any intensity eventually causes the same number of activated melanopsin molecules to accumulate, which contrasts with the role of ipRGCs in steadily encoding light intensity^8,28,42,53. Our study indicates that intensity encoding may emerge from the cycling of gain in a dynamic photoequilibrium of melanopsin states. Gain cycling could exist beyond ipRGCs, such as in invertebrate photoreceptors that use photoconvertible pigments and display adaptation⁵⁴. Indeed, persistent activity and adaptation are widespread and arise from molecular, cellular, circuit, and systems mechanisms^14,55,56. Their interaction may shape biological signaling in diverse contexts.

Additional investigations may deepen the understanding of melanopsin signaling in context. The challenges include obtaining a quantitative understanding of melanopsin phototransduction’s many mechanisms of adaptation and their influences^15–21; learning how these mechanisms interact with ipRGC properties that are dynamic, exhibit unconventional features like depolarization block, and span timescales from milliseconds to hours^8,42,53; capturing the variations of environmental illumination that occur across the day, seasons, and latitudes⁵⁷; and defining how animals sample these conditions^58,59. The experimental observations and computational model provided here are a framework for such efforts. They also support the rational design of lighting to support research and promote health.

Supplementary Material

Supplemental Information

Key Resources Table

Methods

Figures S1 to S10

Tables S1 and S2