Additive contributions of melanopsin and both cone types provide broadband sensitivity to mouse pupil control

We first set out to identify neurons in the mouse pretectum that received input from ipRGCs. To this end, we performed multielectrode (32 channel) recordings from the PON and surrounding pretectum of 26 anaesthetised Opn1mw^R mice. Although the pretectum receives input from both ipRGCs and other RGC types [35, 36], a characteristic feature of melanopsin phototransduction is a sluggish and sustained elevation in firing in response to high intensity short-wavelength (‘blue’) light [5]. Accordingly, to screen for cells likely to receive input from ipRGCs, we first evaluated responses to monochromatic 460-nm light steps (10 s duration from darkness) across a range of intensities (Fig. 1a; 14–16 Log melanopsin effective photons/cm²/s; termed here ‘Mel High’) predicted to robustly activate melanopsin-based responses in all known classes of ipRGCs [37–39].

Across all primary retinorecipient targets, visual responses driven by rods and cones are characterised by an acute increase in firing rate within a few hundred milliseconds following stimulus onset that typically decays substantially on prolonged stimulation [11, 12, 16, 17, 27, 40]. By contrast, where present, response elements originating with melanopsin build up over several seconds and persist even under very long exposure [7, 11, 16, 17, 40]. We first then quantified early and later components of the response to bright 460-nm light steps (0–500 ms and 5–10 s following stimulus onset respectively) to identify neurons with the sustained changes in firing consistent with melanopsin input (Fig. 1b).

As expected, based on previous data [11], a substantial number of the pretectal neurons we isolated from these multielectrode recordings (n = 72/230 visually responsive cells) exhibited sustained increases in firing in response to such stimuli (Fig. 1b, c). By contrast, the remaining cells exhibited responses incompatible with a significant contribution from ipRGCs (Fig. 1b, c): the majority (n = 121/158) exhibited only very transient increases in firing at the onset and offset of the light step (as reported previously; [11]), while a smaller subset (n = 37) exhibited light-driven decreases in firing rate (OFF responses).

To more specifically establish that the population of cells exhibiting sustained responses to Mel High stimuli did indeed correspond to those receiving input from ipRGCs, we next evaluated their responses to polychromatic light steps, matched to provide identical stimulation of L- and S-cones but a significantly weaker impact on melanopsin (Fig. 1a; ~ 500-fold weaker, termed here ‘Mel Low’). We chose this approach, initially, since we predicted it would allow us to evaluate the impact of much larger variations in melanopsin excitation than are achievable using more selective melanopsin-isolating stimuli, without encountering the rod-intrusion that can occur following light-adaptation [16, 40, 41]. Hence, while the Mel Low/High stimuli used here also differed substantially in their impact on rods, our expectation was that, at the two higher intensities tested, both should be sufficiently bright to produce a saturating rod response when presented as light steps from darkness (≥ 12.9 and 15 Log rod effective photons/cm²/s for Mel Low and High respectively; c.f. [42]).

Consistent with our expectation that Mel High and Low stimuli should produce equivalent rod/cone-mediated responses, light-responsive cells whose properties were incompatible with ipRGC input (those with transient or OFF responses) exhibited equivalent responses to both stimuli (Additional file 1: Figure S1a,b). Similarly, as expected given the very sluggish nature of melanopsin phototransduction, for the majority of the cells with sustained responses (n = 60/72), the initial increase in firing (peak occurring during first 500 ms of light step) was statistically equivalent for Mel Low/High at one or both of the top two intensities tested (within-cell t tests across 10 trials at each stimulus). Further analysis of this group of cells indicated that, in fact, there was no detectable effect of stimulus on early response components (mean firing during first 500 ms) at any intensity (Fig. 1d, e; two-way RM ANOVA with Sidak’s post-tests, all P > 0.05). Importantly, however, Mel High/Low responses diverged at later timescales (Fig. 1d) such that, over the last 5 s of the response, the Mel High-evoked increase in firing was significantly greater at all intensities tested (Fig. 1e; two-way RM ANOVA with Sidak’s post-tests). Based on this pronounced and selective enhancement of late components of the Mel High response (across a range where melanopsin phototransduction is active; [37]), we consider this group of cells melanopsin responsive (MR). These observations align well with the kinetics of PON neuronal responses observed previously in rodless/coneless mice [11]. This conclusion is also further supported by additional data reported below and later in the manuscript (Figs. 5 and 7 and associated additional files), including the anatomical location of such cells which, as expected based on known ipRGC projections [43, 44], were strongly clustered in the region of the PON.

By contrast to the above, we also observed a small number of cells (n = 12) with sustained responses where both early and later components of the responses were significantly enhanced for Mel High stimuli, an effect that was most apparent at the highest intensity tested (Additional file 2: Figure S2a,b). Although we cannot definitively rule out a melanopsin contribution to the responses of this group of cells, this difference even in the earliest portions of the response suggests contributions from a photoreceptor other than melanopsin (presumably an unexpected contribution from rods; see below). Accordingly, for subsequent analysis, we consider this rare group of cells (~ 5% of the light-responsive cells identified here) as non-MR.

To provide further confidence in our classification of cells as melanopsin responsive, we also performed parallel sets of experiments in melanopsin knockout red cone animals (Opn1mw^R; Opn4^−/−; n = 5 mice) and mice lacking functional cone photoreception (Cnga3^−/−; n = 4). Among both groups of mice, a small population of cells (n = 8/41 and n = 3/24 for Opn1mw^R; Opn4^−/− and Cnga3^−/− respectively) exhibited behaviour equivalent to that described above: a global enhancement in responses to the Mel High vs. Low stimulus that emerged at higher intensities (Additional file 2: Figure S2c-f). Since we observe this behaviour in recordings from animals lacking either cone or melanopsin phototransduction, we conclude this must reflect an unexpected difference in rod-mediated responses that appears in certain cells under high light intensities. This may reflect the emergence of bleaching adaptation [41] or perhaps even the contribution of an atypical phototransduction pathway [45]. In either case, the impact of such a mechanism appears to be sufficiently restricted (being essentially absent from the other classes of visually responsive cells we recorded; Fig. 1f–i; Additional file 1: Figure S1) that it does not significantly interfere with our ability to identify MR cells.

Of particular note here then are the more commonly encountered sustained cells exhibiting statistically identical initial increases in firing to Mel High/Low steps at high stimulus intensities (n = 9 and n = 5 for Opn1mw^R; Opn4^−/− and Cnga3^−/− respectively). As for the large population of Opn1mw^R cells matching these criteria (i.e. MR cells), early components of the responses of both Opn1mw^R; Opn4^−/− (Fig. 1f, g) and Cnga3^−/− cells (Fig. 1h, i) to Mel High vs. Low stimuli were in fact equivalent at all intensities (two-way RM ANOVAs with Sidak’s post-tests, all P > 0.05). Cells in Cnga3^−/− animals (where melanopsin remains functional) also retained the pronounced enhancement across later components of the Mel High response at all intensities (Fig. 1h, i; two-way RM ANOVA with Sidak’s post-tests). By contrast, responses of Opn1mw^R; Opn4^−/− cells were qualitatively different (Fig. 1f, g), instead showing statistically equivalent responses at all but the highest intensity where Mel Low responses were very marginally reduced (Sidak’s post-test, P = 0.03; presumably reflecting the threshold appearance of a mechanism equivalent to that illustrated in Additional file 2: Figure S2). Collectively then, these data support our classification of cells as MR and provide further confidence that elements of the Opn1mw^R MR-cell responses ascribed to melanopsin do not instead reflect significant stimulus-related difference in rod/cone activation.

Having identified pretectal neurons that received input from ipRGCs, we next aimed to comprehensively define how their activity was modulated by cone input and determine whether any exhibited the colour-opponent behaviour observed previously in another major ipRGC target—the SCN [27]. To this end, starting with a polychromatic background stimulus that resembled a wildtype mouse’s experience of natural daylight (Fig. 2a), we adjusted the spectral composition to generate carefully calibrated pairs of stimuli designed to differ in apparent brightness for one or both cone opsins without significantly altering rod or melanopsin excitation (Fig. 2b; see also ‘Methods’ for further details). In particular, we generated four stimulus pairs where transitions between each element selectively modulated excitation of just L- or S-cone opsin in isolation, or both opsins in unison (‘L + S’) or in antiphase (‘L − S’). We then applied 0.25 Hz square-wave cycles of these cone-isolating stimulus pairs at a range of contrasts up to 75% Michelson (sevenfold change in apparent brightness).

The majority of MR units tested with these stimuli (n = 50/60) exhibited robust responses to at least a subset of the cone-isolating conditions tested (Fig. 2c–g). Moreover, a substantial proportion of these (n = 24/50) changed their firing rates in opposite directions when we selectively modulated L- or S-cone excitation in isolation; that is, almost half of these MR cells exhibited colour-opponency. Among this group, S-ON/L-OFF behaviour was substantially more common than the converse (n = 17 vs. n = 7 with L-ON/S-OFF responses); however, both classes of opponent MR cells usually exhibited a significant bias (on average ~ 3-fold larger responses) towards the excitatory (ON) component of the response (Fig. 2c, d, g).

We next inspected contrast response functions for ON and OFF responses among these colour opponent cells, by quantifying the percentage variance in firing rate accounted for by stimuli providing 15–75% contrast for the target cone population (see ‘Data analysis’ section in ‘Methods’ for further details). This analysis indicated, while OFF responses were reliably smaller at all tested contrasts (Fig. 2e; two-way RM ANOVA with Sidak’s post-tests), the sensitivity of ON and OFF responses was similar. Consistent with this behaviour, while these opponent MR cells were capable of responding to stimuli that modulated cone ‘illuminance’ without changing chromaticity (L + S stimulus), their responses were significantly larger when activation of the two cone opsins was modulated in antiphase to produce pronounced changes in colour (Fig. 2e; two-way RM ANOVA with Sidak’s post-tests). This chromatic response enhancement was evident even under the lowest contrast tested (15% Michelson).

Anatomically intermingled with the cell populations described above (Additional file 3: Figure S3), the remaining MR cells that exhibited robust responses to our cone-isolating stimuli lacked any clear evidence of opponent behaviour. In most cases, the responses of this group of cells were strongly biased towards one of the two opsins, with similar numbers of cells primarily driven by L-cone or S-cone opsin (Fig. 2g, n = 12 and n = 14 respectively). At the level of the population average then (Fig. 2d), achromatic (L + S) modulations in cone illuminance drove much more robust changes in firing than antiphasic (L − S) modulations (due to differential modulation of L- and S-biased subpopulations). By contrast, at the level of individual cells, responses to L + S stimuli tended to be only marginally elevated relative to those produced by L − S stimuli (Fig. 2e, two-way RM ANOVA with Sidak’s post-tests), consistent with the typically very weak responses driven by the ‘non-dominant’ opsin.

This diversity in the responses of pretectal MR units and, especially, the prevalence of cells exhibiting S-ON/L-OFF type colour opponent responses recapitulates our previous findings in the SCN [27], where retinal input comes almost exclusively via ipRGCs. By contrast, among the pretectal neurons identified here where we did not find clear evidence of melanopsin-dependent responses (Additional file 4: Figure S4), S-ON/L-OFF responses were significantly less common (Fig. 2g, n = 16/170, vs. 17/60 MR cells; Fisher’s exact test, P = 0.001) while the prevalence of L-ON/S-OFF and non-opponent cone-driven responses was similar (P = 0.136 and P = 0.446 respectively). Instead, across this group, there was a higher proportion of cells lacking detectable responses (n = 81/170 vs. 10/60 MR cells, P < 0.0001), indicating that either such cells primarily receive rod input or, perhaps more likely, that they are specialised to detect specific stimulus feature, e.g. motion or spatial contrast [46]. These differences in the relative proportions of response types held true also when we analysed data separately for non-MR cell populations (Additional file 4: Figure S4 g) designated ‘transient’ or ‘OFF’ but not the very rarely encountered group with sustained responses (where it is harder for us to definitively rule out a melanopsin contribution). Collectively then, alongside the recent identification of a subtype of mouse ipRGC (M5) with colour opponent input [28] and our earlier data [27], the present findings indicate a close association between the presence of melanopsin input and S-ON/L-OFF chromatic input.

To rule out the possibility that our assessment of cone preferences (and especially our inability to detect such responses in subsets of cells) simply reflected the relatively high background light intensity used, in a subset of experiments (n = 10 Opn1mw^R mice), prior to collecting the data reported in Fig. 2, we also applied identical cone-isolating stimuli but at a background 10× dimmer (ND1; intensity ~equivalent to sunrise/sunset). In fact, we found all cell groups tested displayed remarkably consistent responses at both background light intensities (Additional file 5: Figure S5a), barring a small number of cells that only responded under one of the two intensities (n = 12 and n = 6 responding only at ND1 and ND0 respectively from 133 cells tested). Response amplitudes were, however, modestly increased at lower irradiance (Additional file 5: Figure S5b, paired t test P < 0.0001). Since, for MR cells, peak firing rates were also at least nominally higher at the lower intensity (Additional file 5: Figure S5c), we suspect the slight reduction in response amplitude at higher irradiance reflects some degree of contrast adaption (since the higher irradiance also followed the lower in these experiments). Most importantly, however, the fundamental nature of the cone-dependent modulations in firing (opponent vs. non-opponent) was retained across all cell types.

Given our data above, we next sought to better understand the temporal properties of cone inputs to MR units in the PON. In particular, a number of previous studies have provided evidence that S-cone contributions to ipRGC-dependent physiological responses (including PON neuronal activity) may exhibit temporal properties that are distinct from those originating with longer-wavelength-sensitive cones [11, 30, 55]. Here then, we examined in more detail the temporal tuning properties of cone inputs to MR cells by applying cone-isolating stimuli as sinusoidal oscillations across a range of temporal frequencies (0.025 to 5 Hz, 75% contrast, ND0).

We first examined the responses of MR cells to their ‘optimal’ stimulus type (L − S and L + S stimuli for colour-opponent and non-opponent cells respectively) and found that the majority of MR neurons that responded to square-wave modulations also exhibited significant responses across a wide range of sinusoidal stimulus frequencies (Fig. 6a, b). Indeed, substantial numbers of MR cells continued to track modulations in cone-derived illuminance or chromatic signals even at the lowest temporal frequency tested (Fig. 6b, n = 32/50). This observation applied similarly to both colour-opponent and non-opponent MR neurons (n = 18/24 and 14/26 respectively; Fisher’s exact test, P = 0.15). By contrast, non-MR cells (Additional file 9: Figure S9) generally showed significant modulations in the firing rate over a narrower range of temporal frequencies, with a much lower proportion of cells capable of tracking the lowest temporal frequency we tested (Fig. 6b; n = 32/89 cells that responded to square-wave stimuli, Fisher’s exact test, P = 0.002 vs. MR units). Similarly, while a substantial proportion of MR cells in fact exhibited the most robust modulations in firing rate at the lowest frequency tested (n = 18/50; quantified as percent variance in firing rate explained by a 0.025-Hz stimulus), this was less commonly observed across non-MR cells (n = 15/89; Fisher’s exact test P = 0.013). Collectively then, these data support previous suggestions that cone inputs to ipRGCs are unusually sustained [14, 15, 53] and indicate that MR neurons in the pretectum can use cone inputs to track even quite gradual changes in illuminance or colour.

We next examined whether the L- and S-opsin-driven responses of MR units differed in their ability to track low-frequency modulations. In line with the data reported above, we found that low-frequency sinusoidal modulations of the cone opsin class to which that cell was biased (as determined from response to square-wave stimuli, Fig. 2) reliably evoked robust responses in a significantly larger population of MR vs. non-MR neurons (Fig. 6c; Kolmogorov-Smirnov test, P < 0.0001). Further analysis classes confirmed that, as expected based on data presented above (Fig. 2), there were near equivalent proportions of both MR and non-MR neurons that preferentially responded to low-frequency stimuli targeting L- or S-opsin (Fig. 6d; MR: n = 22 L-biased vs. 28 S-biased, non-MR: 38 vs. 51; Fisher’s exact test vs. equal proportions—both P > 0.05). Hence, for low temporal frequency stimuli, colour opponent cells almost always exhibited more robust variations in the firing rate when we modulated the opsin providing the excitatory/ON component of their response (Fig. 6e), whereas subsets of achromatic cells preferentially responded to low-frequency L- or S-opsin-isolating stimuli (Fig. 6d, e).

The ipRGCs provide a major but not exclusive source of input to the PON [35] and the wider pretectum receives substantial input from non-melanopsin-expressing RGCs (c.f. [36, 43]). As such, the influence of cones on MR neurons described here need not directly recapitulate those of ipRGCs. Nonetheless, we here find that cone inputs to MR units (both colour-opponent and non-opponent) are characteristically more sustained than among non-MR cells, consistent with previous comparisons of cone influences on ipRGCs vs. other RGC types [15]. We also find that colour opponent responses are particularly common among MR cells in the PON, whereas such properties are significantly less so across PON cells that lack evidence of melanopsin input (although not altogether absent) and are very rare at the level of RGCs in general [56].

Of particular note here then, while an earlier investigation [15], failed to find evidence of colour opponency at the level of mouse ipRGCs, recent data indicate that a particular subtype of ipRGC (M5) displays the S-ON type colour opponency [28] which we commonly find in PON MR neurons (and previously found also in the SCN; [27]). One plausible origin for the colour opponency we observe in PON MR units then is that it is directly inherited from M5 cells. Since the responses of M5 cells are believed to rely on a specific S-ON bipolar cell, this mechanism cannot directly account for the smaller subset of MR units in the pretectum that exhibit the opposite (L-ON/S-OFF) colour preference, however, suggesting an alternative or additional origin.

Although ipRGCs with S-OFF properties have been identified in primates [29], they have yet to be observed in mouse retina. Nonetheless, one previous study has provided evidence that a subset of alpha-ganglion cells (and therefore potentially M4 ipRGCs; [57]) exploit the gradient in mouse cone opsin expression [58] to generate chromatic responses [59]. Such a mechanism could generate either S-ON or S-OFF type chromatic responses. Accordingly, the properties of PON MR units observed here could arise either via input from M4 cells or a combination of M4 and M5 ipRGCs. Either way, the proposed source of colour-opponency involves a centre-surround mechanism such that the chromatic influence is most apparent for stimuli larger than the cells receptive field centre. Centre-surround opponency of this type generally seems to result in stronger centre-driven ON vs. surround-driven OFF responses, which matches the bias we observe also in PON MR units with colour opponent responses. It should be noted, however, that while both M4 and M5 cells are known to project to the visual thalamus [28, 60] it is so far unclear whether any provide input to the PON. As such, it remains possible that the colour opponent responses we observe centrally arise via some alternative mechanism (e.g. involving central processing or convergent input from multiple RGC types). A similar situation is true also for the non-MR colour opponent neurons we observed. Such responses could in principle be directly inherited from one or more RGC types that do not express melanopsin but we cannot exclude the possibility that they instead arise via central processing. We do however note that the temporal tuning of cone-opponent input to MR and non-MR units differs (Fig. 6b), suggesting that the retinal/central mechanisms that generate their opponency also differ.

Although, to our knowledge, there have been no other investigations of chromatic processing in mouse brain regions involved in NIF visual responses, a final point to note here is that previous studies have suggested a qualitative difference in types of visual information provided by S- vs. M/L-cone opsin [11, 55]. In particular, it has appeared that S-opsin-driven responses are more sustained/better able to track lower temporal frequency events. Our data reveal that, in fact, both cone opsin types can readily track low-frequency stimuli, but that individual cells generally display strong preferences towards just one class. In aggregate then, our present data support the previous suggestion that mouse PON neurons are better able to track gradual changes in light at shorter wavelengths [11] but suggest this is not because of any fundamental difference in the temporal properties of S- vs. L-cone pathways that impinge on ipRGCs. Instead, it seems simply that, because both cone types show significant sensitivity in the UV-blue region of the spectrum (e.g. see Fig. 1a), shorter-wavelength stimuli recruit more neurons (i.e. both S- and L-cone biased non-opponent cells and S-ON/L-OFF colour opponent cells).

Our data regarding the timescales over which melanopsin contributions to PON activity/pupil control emerge align well with previous data from macaque and human [30, 65], suggesting a well-conserved contribution of this photoreceptor channel. By contrast, our finding that chromatic signals do not noticeably contribute to mouse pupil regulation suggests a clear distinction from humans, where S- (and potentially also M-) cones contribute opponent signals to those arising from the other opsins [30–32, 66].

Insofar as mice are clearly less specialised towards chromatic photopic vision compared to humans, such a divergence might simply be written off as an expected consequence of the differences between mouse and human retina. Nonetheless, mice retain the S-cone-specific circuitry required to generate a chromatic channel equivalent to human blue-yellow colour vision [67], in addition to mechanisms of colour opponency analogous to those proposed to underlie primate red-green colour discrimination [59]. As such, mice can use colour both as a circadian timing cue [27] and for discrimination at the behavioural level [68], including adapting to use information from novel photopigments based on existing retinal architecture [69]. Most significantly though, we directly demonstrate here that colour information is readily apparent at the level of mouse PON neurons just as it is presumed to be in humans [29].

Given the points raised above, if including colour information in pupil regulation were of value to mice, we see no reason why this would not be apparent. As such, we assume the divergence in the contribution of cones to mouse vs. human pupil control instead reflects the distinct sensory requirements of these two species. While the value of including a chromatic component in human pupil regulation is currently unclear, in the case of mice, we see clear value in using short- and longer-wavelength-sensitive cones in an additive manner. Indeed, as discussed above, this enables rapid pupil constriction to light increments across the visible spectrum, an arrangement that is perhaps more important for preventing photoreceptor saturation in the rod dominated retina in this species vs. the cone-dominant retina of humans.

All experiments were conducted in accordance with the Animals (Scientific Procedures) Act of 1986 (United Kingdom) and received approval from the institutional ethics committee at the University of Manchester. Mice were bred and housed at the University of Manchester in a 12:12-h light dark cycle at 22 °C with food and water available ad libitum. Most experiments were performed in adult male red cone knockin (Opn1mw^R) mice [34]. For a subset of experiments, we used hemi/homozygous offspring of a red cone knockin/melanopsin knockout [5] cross (Opn1mw^R; Opn4^−/−). A further subset employed mice lacking the cone-specific cyclic nucleotide-gated channel (Cnga3^−/−; [70]).

Animals were anaesthetised with urethane (1.55 g/kg i.p.; Sigma-Aldrich, Dorset, UK) and prepared for stereotaxic surgery as described previously [71]. In brief, a small craniotomy (< 1 mm diameter) was placed 1.1 mm lateral and 2.7 mm posterior to the bregma (based on the coordinates of the OPN from the stereotaxic mouse atlas [72]), and atropine (1% solution in saline) followed by mineral oil (both from Sigma-Aldrich) was applied to the eyes to achieve pupil dilation and to prevent corneal drying without modulating optical transmission. Silicon multielectrode recording probes (32 channel) were then coated in CM-DiI (V22888; Fisher Scientific, Loughborough, UK) to allow post hoc visualisation in histological images, before being inserted into the brain using a hydraulic micromanipulator (MO-10, Narishige International Ltd., London, UK) at 10° to the dorsal-ventral axis and to a depth of 2.3 mm from the pial surface. Experiments employed a variety of silicon recording array designs; either 4 × 8 probes (n = 14 experiments; 200 × 50 μm spacing; A4x8-5mm-50-200-177; Neuronexus Technologies Inc., USA), 4 × 2 tetrode arrays (n = 19; 200 × 150 μm inter-tetrode spacing, 25 μm intra-tetrode spacing; A4x2tet-5mm-150-200-121; Neuronexus) or 2 × 16 parallel arrays (n = 2; 250 × 20 μm spacing; Cambridge NeuroTech, Cambridge, UK). In all cases, prior to neurophysiological recording, mice were left for 30 min to dark to adapt and allow neural activity to stabilise.

Neuronal data was acquired using a Recorder64 data acquisition system (Plexon, USA), amplified for a gain of 3500×, digitised at 40 kHz and stored continuously in a 16-bit format. Single unit activity was isolated using an automated template-matching-based algorithm (Kilosort; [73]) and manually refined using Phy [74]. For initial experiments (n = 8), we also extracted virtual tetrode waveforms and sorted manually in Offline Sorter (Plexon, TX, USA), as described previously [71]. Single unit isolation was confirmed by reference to MANOVA F statistics, J3 and Davies-Bouldin validity metrics and the presence of a distinct refractory period (> 1.5 ms) in the interspike interval distribution. Unit isolation resulting from the semi-automated procedure was essentially identical to that derived manually.

All light measurements were performed using a calibrated spectroradiometer (Bentham instruments, Reading, UK) and quantified according to the known opsin sensitivities after correction for prereceptoral filtering [75, 76] as described previously [27, 54]. For modelling (Fig. 3), we adjusted S- and L-cone opsin sensitivity by ± 25 nm relative to λmax used for calculations below (S-opsin: 365 nm, L-opsin: 556 nm). For this analysis, we used a two-phase saturating function that approximated published measurement mouse lens sensitivity [76] so that we could assess the impact of varying shortwave cutoff (again by ± 25 nm relative to a reported value of 337 nm). Penumbral cone sensitivity was estimated using the method previously presented [50].

Visual stimuli were created using a bespoke light source (components from Thorlabs: NJ, USA, and Edmund Optics; York, UK) which combined signals from three LEDs (λmax 410, 470 and 617 nm; see Fig. 1) via dichroic mirrors. Light was supplied to the subject via a 7-mm diameter flexible fibre optic light guide positioned 5 mm from the contralateral eye and enclosed within an internally reflective plastic cone to provide approximately full field illumination and prevent any light reaching the ipsilateral eye. A similar apparatus was positioned over the ipsilateral eye, providing light (where required) from a single 410-nm LED. LED intensity was controlled dynamically via a PC running LabVIEW and a USB-6343 DAQ board (National Instruments, TX, USA). Further control was provided by neutral density (ND) filter wheels, allowing for spectrally neutral 10 to 10,000-fold (ND1-ND4) reductions in light intensity.

For identification of cells that received input from ipRGCs, we compared responses to two stimuli: a simple monochromatic 460-nm light step (‘Mel High’) and a combination of 410 and 617 nm light (‘Mel Low’; Fig. 1a). Stimuli were presented in interleaved fashion as 10-s steps from a background of darkness (60 s ISI; 10 repeats/stimulus at ND2–0). Intensities of the two stimuli were calibrated to provide identical activation of mouse S- and L-(Opn1mw^R) cone opsins (12.8 and 15.5 Log effective photons/cm²/s respectively at ND0) but to differ greatly in their ability to activate melanopsin (16.0 vs. 13.4 Log effective photons/cm²/s for Mel High and Low). Although these stimuli, in principle, also differ substantially in apparent brightness for rods, effective rod flux for both stimuli was expected to be sufficiently high by ND1 (12.9 Log effective photons/cm²/s for Mel Low) to evoke a maximal rod response [42, 77, 78].

For selective manipulation of cone photoreception, we first calibrated a mixture of each LED to approximate a wildtype mouse’s experience of natural daylight (Fig. 2a, b), based on previously reported data [27]. Input-output relationships for modulations in LED intensity relative to this background were measured and linearized by 4th order polynomial fits. We then calculated, as above, modulations in the intensity of each LED that provided changes in the activation of L- and/or S- cone opsins (in isolation, unison or antiphase) by ± 75% relative to the background (equivalent to a 0.85 log unit or sevenfold change in apparent brightness for the stimulated opsin). For L- and S-cone-isolating stimuli, effective change in brightness for the silenced opsin was calculated to be < 0.2% Michelson and for all stimuli effective changes in rod and melanopsin activation were below 6% (full details of all stimuli are provided in Additional file 12). We also produced stimuli providing spectrally neutral modulations in light intensity (± 75% relative to the background in Fig. 1c) that allowed for concurrent modulations in the activation of all opsins and stimuli designed to provide a large spectral change but to be silent for cones (Fig. 3a; < 0.05% Michelson contrast for cones, 45.0 and 43.2% respectively for rods and melanopsin).

For assessment of contrast response relationships, we applied cone-isolating and all-opsin stimuli as 0.25-Hz square-wave modulations providing 15–75% Michelson contrast (albeit with a smooth 40-ms transition between ‘bright’ and ‘dim’ phases). Opsin-specific stimuli were presented as interleaved blocks of 6 cycles running from low to high contrast to control for any effect of contrast adaptation. The full protocol was then repeated five times to provide 30 repeats for each stimulus. Stimuli were presented well within the photopic regime (as shown in Fig. 2b) to allow for assessment of melanopsin-cone interactions (comparison of L + S and all opsin stimuli) without rod intrusion [40]. In a subset of experiments, we first also applied the same protocol at ND1, to confirm cone-based responses were stable across the photopic range. For experiments assessing temporal frequency tuning, cone-isolating and all-opsin stimuli were applied exclusively at ND0 as sinusoidal modulations (0.025–5 Hz) at a fixed contrast of 75% Michelson (interleaved as above although here the number of stimulus cycles ranged from 10 at 0.025 Hz to 80 at 5 Hz).

For all stimuli involving cone-isolating stimuli, background light intensity at the ipsilateral eye was set to approximate that for the stimulated eye using a 410-nm LED (which provides near equal activation of all mouse opsin classes; [71]). The one exception here was for an additional protocol applied in a subset of experiments (n = 5 Opn1mw^R mice, n = 4 Opn1mw^R;Opn4^−/−) where cone-isolating stimuli were applied using identical parameters to those used for pupillography (below), interleaved single cycle square-wave modulations (75% contrast, 5 s/phase, positive first and negative first polarity).

After each experiment, brains were removed and placed into 4% paraformaldehyde for 48 h before cryoprotection in 30% sucrose (24 h). Brains were then frozen with dry ice and sectioned coronally using a freezing sledge microtome at widths of 100 μm before mounting with Vectashield (Vector laboratories, UK) to glass slides and coverslipping. Sections were imaged under an upright light microscope (BX51; Olympus, UK) with appropriate filter sets for visualisation of DiI fluorescence and images acquired with a Coolsnap HQ camera (Photometrics, USA). Resulting images were scaled and aligned with best matching coronal panels from the mouse atlas [72] with the anatomical location for each cell estimated based on the known geometry of the recording array and the corresponding recording site location were largest spike amplitudes were detected. For display (Additional file 3: Figure S3), estimated unit locations were mapped onto a single coronal anatomical template, based on the projected centre of the PON from each recording.

Animals were habituated prior to handling and had their whiskers trimmed prior to experiments to reduce interference in the images. All experiments were performed during the middle of the animal’s light phase and during experiments the room was maintained in constant darkness. For assessment of consensual pupil responses, the left pupil was dilated by topical application of tropicamide (1% solution; Minimus, UK), mice were gently restrained and the same light source used for neurophysiological recording placed over the dilated pupil. A NoIR Raspberry Pi camera (V2.1; Raspberry Pi Foundation, UK) with a macro lens and visible light filter (695 nm longpass; Thorlabs, USA) was then placed at a fixed distance (5 cm) from the right eye and an 880-nm infrared light (Thorlabs, NJ, USA) was then placed above to enhance pupil-iris contrast. Images were acquired at 2 Hz, synchronised to visual stimulus presentation via LabVIEW.

Mice were exposed to the background stimulus for 30 s prior to application of cone-isolating and all photoreceptor stimuli (as above) as single cycle square-wave modulations (75% contrast, 5 s/phase) at ND1. Stimuli were presented to each animal in randomised order and at both possible polarities (i.e. transition from background to ‘bright’ followed by ‘dim’ and vice versa) with individual stimuli separated by 10 s of exposure to the background. Each animal received two to four trials of each stimulus/polarity (mean = 3.1 trials/stimulus).

For analysis, images (2592 × 1944 pixels) were virtually stacked and stabilised using After Effects (Adobe Software Systems Ltd., Ireland) then cropped to the horizontal and vertical boundaries of the eye. The images were then processed using custom MATLAB scripts (Mathworks, USA), based on published methodology [79], to extract pupil diameter. In brief, images were first converted to greyscale and a 3 × 3 median filter applied to reduce noise. A region-growing algorithm was run from an initial, user-defined, seed point. The resulting binary image then had any holes generated by reflections filled before undergoing a morphological opening operation with a circular 5 × 5 structuring element for smoothing. The image boundary was subsequently extracted by subtraction from a morphologically dilated representation (using a 3 × 3 structuring element) and an ellipse fitted using the direct fit method [80]. Pupil diameter (in pixels) was taken as the long axis of the fitted ellipse and normalised according to eye size (long axis of an ellipse fitted to the whole eye).

For initial cell classification, units were considered to exhibit sustained responses when the change in firing evoked by the Mel High stimulus remained above 25% of its initial value (mean over the first 500 ms) over the last 5 s of the 10-s light step (and did not fall below 1.2 spikes/s above pre-step values). Cells that otherwise exhibited significant increases in firing at the onset of the light step were designated ‘transient’ and cells that exhibited significant decreases as ‘OFF’. We further separated cells with sustained responses based on the presence of equivalent initial responses to Mel High vs. Low (100 ms bin occurring within the first 500 ms of light onset with the highest firing rate) at the two higher intensities (indicating equivalent rod/cone-driven components of the response). For subsequent analysis, we then extracted the mean change in the firing rate for each cell (relative to the preceding 10-s baseline) across early (first 500 ms) or late (last 5 s) elements of the response and analysed with two-way RM ANOVA (GraphPad Prism 7; GraphPad Software, Inc., CA, USA).

For analysis of neuronal responses to cone-isolating stimuli, spike counts were binned (100 bins/stimulus cycle; smoothed with a 5-bin boxcar filter) and peak-trough amplitudes extracted. To remove the effect of random variations in baseline firing, we also produced similar estimates based on shuffled data (spike counts shuffled in time independently for each trial). Cells were considered responsive when the measured response amplitude exceed the 95% confidence limits of responses assessed from shuffled data (100 repeats), with the mean shuffled response subsequently subtracted from the true response. Response polarity (ON vs. OFF) was assessed based on the stimulus phase where we observed the largest absolute deviation in spike rates from the mean and the sign (positive vs. negative) of that response. To determine cone opsin preference, the mean response across all tested contrasts was determined across each of the four cardinal stimulus dimensions, normalised according to the largest absolute magnitude and represented as vectors with ON and OFF responses arbitrarily designated as vectors at 0–180°, 45–225°, 90–270° and 135–315° for L, L + S, S and L − S stimuli respectively. Overall preference was then taken as the mean of the resulting vectors. Cells were categorised as colour opponent when we observed significant responses (as above) of opposite sign for L- and S-opsin-isolating stimuli or when we did not detect a significant response from one of the two cone opsins in isolation but found that the mean response to the L − S stimuli was significantly greater than that to the L + S stimulus (t test, P < 0.05).

For comparisons of contrast response functions and temporal frequency tuning we also used a modification of the χ²-peridogram [27, 81] to quantify the percentage variance within the timeseries that was accounted for by a rhythmic process with the same periodicity as the stimulus. This provides an unbiased assessment of relative response strength that has the advantages that it makes no assumptions about the shape of neuronal responses, allows for direct significance testing and allows for comparison of responses occurring across greatly differencing timescales. As above, for group comparison of responses between stimuli, we used two-way RM ANOVA.

For quantification of pupillary responses, we extracted the mean pupil size across 5-s epochs corresponding to ‘bright’ and ‘dim’ stimulus phases (expressed as a percent change relative to baseline) and compared via paired t test. Further analyses were performed on the raw difference between bright and dim stimulus phases (normalised to eye size) using two-way and one-way RM ANOVA as indicated in the manuscript text.

Throughout the manuscript, appropriate post hoc tests were employed wherever main ANOVA reported a significant effect of stimulus or interaction and significant differences (P < 0.05) between stimuli indicated wherever they occurred. Full details of all statistical test results and the underlying raw data is provided in the relevant worksheets of the accompanying data (Additional files 12 and 13).

This work was funded by the Biotechnology and Biological Sciences Research Council UK grants BB/I017836/1 and BB/N014901/1 to TMB.

All data generated or analysed during this study are included in this published article [and its Additional files].

Not applicable.

Not applicable.

The authors declare that they have no competing interests.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

All data generated or analysed during this study are included in this published article [and its Additional files].