Distinctive properties of cones in the retinal clock and the circadian system

Multiple circadian clocks dynamically regulate mammalian physiology. In retina, rhythmic gene expression serves to align vision and tissue homeostasis with daily light changes. Photic input is relayed to the brain to entrain the master circadian clock, the suprachiasmatic nucleus, which matches behaviour to environmental changes. Circadian organization of the mouse retina involves coordinated, layer-specific oscillators, but so far little is known about the cone photoreceptor clock and its role in the circadian system. Using the cone-only Nrl-/- mouse model we show that cones contain a functional self-sustained molecular clockwork. By bioluminescence-combined imaging, we also show that cones provide substantial input to the retinal clock network. Furthermore, we found that light entrainment and negative masking in cone-only mice are subtly altered and that constant light displayed profound effects on their central clock. Thus, our study demonstrates the contribution of cones to retinal circadian organisation and their role in finely tuning behaviour to environmental conditions.


Introduction
to the LD cycle at low light intensities [26][27][28] and also for cones [29][30][31]. In addition, recent results in mice suggested that cones also play a role in entrainment mechanisms by perceiving spectral changes characteristic of dusk or dawn [32,33]. However, these functions have not been investigated with gain of function mutants, in particular for cones.
Here we investigate the role of murine cones in the circadian system. We show that cones in  Figure 1A). A total cone photoreceptor area of 3 mm 2 was collected per eye. In order to prevent RNase reactivation and RNA degradation, the microdissection was carried out within maximum 60 min for each slide. 3-4 caps/eye were collected into the same reaction tube which contained RLT + lysis buffer (Qiagen, Hilden, Germany) and stored at −80°C.
which were removed for the final statistical analysis (n = 1 for Per1, Per2 and Per3 quantification).

Real-time bioluminescence recordings
Bioluminescence recordings from whole retinas and isolated photoreceptor layers were obtained in several successive experiments and data were analysed all together. Only samples generating a bioluminescence signal above the background level (similar level for both genotypes: 80 % samples for whole retina and 30 % samples for photoreceptor layers) were retained in the study.

Whole retina explant cultures
WT and KO mice (5)(6) week-old, Per2 Luc background), were euthanized with CO2 (progressive increase up to 20% in an airtight box) during the light phase and enucleated.
Eyeballs were kept at room temperature in HBSS [1 x HBSS (Sigma-Aldrich, Steinheim, Germany) containing antibiotics (100 U/mL penicillin and 100 mg/ml streptomycin, Sigma-Aldrich, Steinheim, Germany), 100 mM HEPES (Sigma-Aldrich, Steinheim, Germany) and 4.2 mM sodium bicarbonate (Sigma-Aldrich, Steinheim, Germany)] for whole retina dissection. The eye ball was incised under the ora serrata and the cornea and lens were cut out. Retinas were carefully detached from the retinal pigment epithelium and flattened with small radial incisions.

Photoreceptor layer explant cultures
Retinas were dissected as described above. Photoreceptor layers were isolated using the vibratome technique and cultured as reported previously [9]. WT (n = 6 samples, 6 mice) and KO (n = 9 samples, 8 mice) photoreceptor explants were recorded for at least 5 days and the photons were integrated for 112 s every 15 min. Exceptionally, when layers of insufficient size were collected, samples from both retinas were cultured together (2 samples in WT group, 1 sample in KO group).

Transversal retinal slice imaging
Flattened retinas (4 week-old KO mice, n = 3, Per2 Luc background) were mounted with warm (37°C) 5% gelatin on top of a 10% gelatin block. The whole retina-embedded block was glued on the tissue holder and then placed into the tissue bath (containing HBSS, Sigma-Aldrich) of a Vibroslice MA752 (Campden Instruments, Loughborough, England). A transversal 100 µm thick slice was cut, placed carefully on a semipermeable membrane in a 35 mm culture dish and pre-incubated with Neurobasal A medium for 24 h. Just before imaging the medium was replaced with pre-warmed recording medium under dim red light.
The sealed dish was placed into the culture chamber (37°C) of a Luminoview 200 microscope (Olympus, Hamburg, Germany) equipped with an EM-CCD camera (Hamamatsu, Japan) cooled to −76°C. Bioluminescence images (20x objective, EM gain = 80, 1 × 1 binning of pixels) were taken every 2 h over minimum 3 days.
Culture dishes were sealed with vacuum grease. The bioluminescence was recorded using the LumiCycle for 112 s in 15 min intervals and during at least 6 days.

Bioluminescence data analysis
Whole retina and SCN explant PER2::LUC raw data were subtracted with a 24 h running average (removal of the baseline drift) using the LumiCycle analysis software (Actimetrics, Wilmette, IL, USA). The first cycle was removed and the analysis was performed on the following 4 (retina) or 5 (SCN) cycles. The robustness of the rhythms (relative rhythmic power [36]) and the phase were also calculated using the LumiCycle analysis software. The period, amplitude and damping rate were determined using a cosinor derived sine wave function: f = y0 + a * exp (-x/d) * sin [2 * π * (x + c) / b] where a is the amplitude (counts/s), b is the period (h), c is the phase-related term (h) and d is the damping rate (days) and assuming that damping follows an exponential pattern. Baseline for each individual peak in retinal samples was estimated as the baseline from LumiCycle analysis taken at the peak time.
Photoreceptor layer data were analyzed as previously described, on 4 successive cycles [9].
Bioluminescence data from whole retinas and photoreceptor layers were obtained over several series of recordings: samples for which activity did not exceed the background of Lumicycle were excluded from the study.
Transversal retinal images were analyzed with ImageJ (open source software https://imagej.nih.gov/). A median 3D filter was applied to remove the hotspots. The ganglion cell layer (GCL), inner nuclear layer (INL) and photoreceptor layer (PRL) were defined as regions of interest (ROI) and the bioluminescence levels (grey levels) were measured and exported for the analysis of rhythmicity. The periods were determined using the cosinor derived sine wave function: f = y0 + a * exp (-x/d) * sin [2 * π * (x + c) / b] as above.

Locomotor activity recordings
For behavioural recordings, male and female mice (WT and KO combined or not with the Per2 Luc knock-in allele) were housed in individual standard cages equipped with a 10-cmdiameter stainless steel running wheel [37] or with infrared detectors placed above the cage and linked to an automated recording system (CAMS, Circadian Activity Monitoring System, Lyon, France) as previously described [38]. Data were collected in 5 min bins and analysed with the ClockLab Software (Actimetrics, Wilmette, IL, USA). Locomotor activity data were represented as double-plotted in actograms.

Circadian phenotype
To determine the daily and circadian rhythm of locomotor activity in Nrl mutant mice, 5-6 month-old mice (WT n = 4, KO n = 7, Per2 Luc background) were initially maintained for 10 days under LD 12:12 and then 16 days under constant darkness (DD). Total activity and rhoand alpha-phase activity levels were calculated over the last 3 days in LD and the endogenous period (Chi-square Periodogram method) was determined over 10-day interval after 7 days from the transition to DD.

Behavioural phase-shifts to light pulses
To evaluate phase shifting in response to light pulses 5-6 week-old mice (WT n = 5, KO n = 8) were initially maintained in LD 12:12 (100 lux) and then challenged by 3 alternating DD (9-14 days) -LD (14-18 days) cycles. On the day before each light-pulse, the room lights went off at ZT12. On the following day a 15 min light pulse (LP) was applied at the projected ZT15. Then lights remained off for at least 9 days before re-exposing animals to LD condition. The intensity of the light pulses decreased one order of magnitude as indicated in Figure 3A, B. To determine phase changes in control and Nrl mutant mice, a linear regression analysis of the activity onsets was performed by projecting the onset phase of the free run in DD back to the mean onset phase under LD condition (ClockLab).

Exposure to constant light
Effects of constant light exposure (light/light, LL) were assessed by wheel running activity in 6 month-old mice (WT n = 6, KO n = 7, Per2 Luc background). Thus, after 10 days in LD 12:12 animals were transferred to LL for 70 days (130 ± 34 lux on average). Total activity per cycle, period and relative rhythmic power were measured by using ClockLab. Mice were then exposed to a second LD cycle (LD2: 10 days) to evaluate if entrainment and locomotor activity returned to baseline levels.

Statistical analysis
Results are expressed as means ± SEM, except for qPCR data. Statistical analyses were performed by using SigmaPlot 12 software (Systat Software, San Jose, CA, USA).
Comparison of two groups was performed by using the Student's t test. Comparison of several groups was performed by using 1-way or 2-way ANOVA for independent and repeated measures, followed by post hoc test (Holm-Sidak test).
A statistically significant difference was assumed with p values less than 0.05.

A functional clock in cone photoreceptors
We first aimed to characterize the cone molecular clock on microdissected photoreceptors isolated from the Nrl KO mice over 24 h in DD ( Figure 1A). We found that all core clock gene transcripts examined, Bmal1, Clock, Per1, Per2, Per3, Cry1, Cry2, Rev-Erbα, Rorβ are expressed in cones ( Figure 1B, top panel). Significant rhythmic levels of expression were determined for Bmal1, Per1, Per2, Per3, Rev-Erbα (Table 1 and Supplementary table 2).
Interestingly, the expression profile of Bmal1 was in opposite phase in comparison to the profiles of Per transcripts, as described in the SCN [40,41] and other peripheral tissues such as liver [42,43]. 24 h profiles in cones were also similar, at least for Bmal1 and Per1 transcripts, to those reported for mouse whole retinas sampled in DD [44].
We also investigated the expression of several well-known or putative target genes of the  (Table 1 and Supplementary table 2).
To further evaluate the capacity of cones to sustain rhythmicity, we used a vibratome-based sectioning of the retina to isolate photoreceptor (cone-only) layers from the KO mice raised on the Per2 Luc reporter background [34] for real-time bioluminescence recordings.
Photoreceptor layers from WT mice were used as controls. As previously described [9] the latter showed robust PER2::LUC oscillations with a 26.46 ± 0.02 h period ( Figure 1C). Cone layers from the KO retinas also proved robustly rhythmic in culture, but yet with a significantly longer period: 29.07 ± 0.03 h (n = 6 for WT, n = 9 for KO; genotype effect: p = 0.018) ( Figure 1C).
Finally, we examined how cone layers oscillate within the context of the whole retina by using in vitro real-time bioluminescence combined with imaging. 100 µm transversal sections were cut using the vibratome technique illustrated in Figure 1D, transferred on a semipermeable membrane, then cultured and imaged for several days in a temperature controlled microscope chamber. PER2 bioluminescence signal emerged from all layers, with higher intensity in ganglion and inner cell layers and weaker signal in the outer, photoreceptor layer ( Figure 1E). Moreover, the PER2 signal was rhythmic in all layers ( Figure 1F Table 1 and Supplementary Table 2 Results are represented as mean ± SEM. *: p < 0.05.    Figure 1C) showed no significant differences between wild-type and mutant retinas. By contrast, we observed a 2-fold increase in the level of expression of Opn1mw in the Nrl mutants (p < 0.0001), as previously described [20,45]. Despite normal number and phenotype of dopaminergic cells, KO retinas displayed altered dopamine metabolism in response to light (Supplementary Figure 1D), as known for rodless retinas [46].

Gene
Thus, the changes related to rhythms measured in vitro in the Nrl -/retinas are likely to result mainly from their specific photoreceptor composition.

SCN-driven rhythms are preserved in the cone-only mouse
SCN explants from Nrl -/-Per2 Luc mice produced autonomous and sustained PER2::LUC oscillations for at least 6 days in vitro similar to those from WT (Figure 2A). The robustness of rhythms was similar between genotypes, based on the relative rhythmic power (p = 0.344), indicating that in the cone-only mutant the master clock is not impaired ( Figure 2B).
Moreover, there was no effect

Mild effects of acute light exposure
To evaluate behavioral response to acute light exposure, WT and KO mice were first exposed to a phase-resetting protocol.

Nrl -/mice do not re-entrain to phase-shifted LD cycle at low light intensity
Animals were challenged with four successive 6 h phase-delayed LD cycles combined with a reduction of light intensity (100 lux, 10 lux, 1 lux, 0,1 lux) ( Figure 4A, Supplementary Figure   4). WT animals were able to entrain to each shifted LD cycle at different light intensities while the KO mice needed longer time to entrain at 1 lux (jet-lag 3, p = 0.026) and were not able to entrain at 0.1 lux (jet-lag 4, p < 0.001) even after 50 days ( Figure 4B). Subsequent exposure to total darkness (DD, 22 days) confirmed that almost all mutant animals were freerunning in the previous condition ( Figure 4A and Supplementary Figure 4; data not shown).
When animals were subsequently exposed to LD at 100 lux during the light phase, animals from both genotypes were able to re-entrain ( Figure 4B), confirming that there was no overt loss of visual function.

Discussion
In the present study we used different approaches to determine the role of cones in the circadian system. We show that photoreceptor layers lacking rods but having normal cone and cone-pathway contain a molecular machinery characteristic of a functional clockwork and likely contribute, together with the inner and ganglion cell layers, to the overall clock rhythmicity in the retina. We bring evidence that the Nrl -/retina also displays novel distinctive properties regarding light impact on the central clock, providing new insight into the role of cones in the circadian system.
Rhythmic functions in mammalian cones have been only poorly documented [13,[47][48][49][50], likely because of the scarcity of this cell type in nocturnal rodents [18,19]. To circumvent this limitation, we used the Nrl -/animal model in which all rods are replaced by cones [20,51].
These photoreceptors were previously shown to have major characteristics of native S-cones regarding morphology, molecular content, nuclear architecture and light response [20,[51][52][53] and constitute an adequate model to question the properties of cones without the interference from rods. Moreover, and unlike what was shown in other models with impaired rod phototransduction pathway [54,55] these retinas show no sign of alteration of other cellular populations such as dopaminergic amacrine cells or ipRGCs, which are known to contribute to clock properties in the retina [8,13].
Cellular localization of the circadian clock in the mammalian retina is still a matter of debate.
The literature agrees on a main contribution from the inner retina [9,44] and several reports exclude rod-type photoreceptors from the circadian network [13,56,57] although the presence of sustained clock gene rhythms in rods has been suggested elsewhere [8][9][10]58].
Upon immunofluorescence analysis of clock protein factors, cones appeared the most evident cell-autonomous clock in the mouse retina [13]. In agreement with this study, we here describe robust rhythms in expression of clock genes from the main (Bmal1, Per1, Per3) and secondary (Rev-Erbα) loops of the well described molecular machinery [59] in Nrl -/photoreceptor layers laser-microdissected throughout the 24 h cycle in constant dark condition. However, unlike what was described for immunostained clock factors, their mRNAs show distinct phases, as observed at the level of the whole retina [44], which might be due to the enrichment in S-versus M-cones in the KO retinas or suggest posttranscriptional regulation of clock factors. We previously described that cones are the photoreceptor site of robust oscillations in Aanat (the enzyme responsible for melatonin rhythm) expression by using a diurnal, cone-rich rodent, Arvicanthis ansorgei [11,47].
Besides, circadian rhythms in cone-specific genes have essentially been investigated in chicken [60,61] and zebrafish [62]. In particular, robust rhythms in phototransduction genes in zebrafish cones appear driven by key transcription factors (Neurod, Crx) themselves regulated by the clock [63]. In our study, cones express major phototransduction elements in a rhythmic manner with high amplitudes but we did not detect any rhythm in Crx expression, indicating that in mammalian cones phototransduction elements retain clock regulation but with mechanisms distinct from the zebrafish. Importantly, when isolated by vibratomesectioning of fresh retinas, cone layers express sustained rhythms with a specific period, distinct from the period measured in photoreceptor layers from control mice. This observation probably reflects the differences in clock machinery and associated signalling occurring in rods (97% of photoreceptors in WT) versus cones. It might also reflect a difference in coupling strength within the respective photoreceptor populations, as previously described in the retina [9]. Communication through gap junctions might be reduced in the S-cone enriched photoreceptor layers of the KO, since expression of connexion 36 was shown to be absent in this cone population in mammals [64]. This might be responsible for the increased period in the KO [9]. Taken together with our demonstration of rhythmic phagocytosis of cone outer segments [21] our data strongly suggest the presence of a functional, autonomous circadian clock within cones.  Figure 2). We observed that the high number of cones does not provide any increased response capacity of the circadian system to the diverse light stimulation paradigms used here. This result corroborates previous discussion in the field, suggesting that the light adaptation properties of cones preclude their participation in the input of long light exposure to the circadian system, including phase shift experiments [26,28]. Indeed, no defect was detected in Nrl -/mice under the phase shift paradigm (as also seen in [45]). By contrast, our model rather displays some features typical of rodless animals, such as reduced capacity to shift at low light intensity (1 or 0.1 lux) in a jet-lag experiment, as previously observed with the Gnat1 -/model [26]. Physiological features of rodless retina are also reflected in dopamine metabolism (Supplementary Figure 1D), with the loss of daily rhythm of DOPAC generation in KO retinas [46,68] as previously described for the rds strain. The discrepancy between the results from light pulse and the jet-lag experiments might be due to the fact that the threshold levels required for entrainment constitute a more sensitive test of deficit in entrainment than phase shift following a light pulse [69].
Rats or mice with outer retinal impairment were repeatedly reported to exhibit total loss of positive masking by light and (consequently) enhanced inhibition of locomotor activity (negative masking), especially at low light intensities [70][71][72]. By contrast, melanopsin phototransduction appears indispensable for negative masking [73]. Using monochromatic light, Thompson et al. also provide evidence that cones (short-and medium-wavelength sensitive) contribute to negative masking and influence its dynamic range [74]. In the present study, the Nrl -/animals show reduced negative masking behaviour specifically at low light intensities (between 0.5 and 10 lux), despite a normal ipRGC population and unlike most rodless mice. The discrepancy between this result and the literature might be explained by the fact that we used global activity recordings and not wheel running activity. Indeed, positive masking might be more pronounced when using wheel running activity and hence introduce a confounding effect (increased negative masking in rodless animals) at low light intensity.
Furthermore, some data also indicate that rods contribute, at least transiently, to negative masking at light intensities too low to excite ipRGC [75]. Thus, the behaviour triggered in the Nrl -/animals by acute light stimulation probably reflects the combined absence of rods and integrity of ipRGC.
Increase of the endogenous period in constant light has also been partly attributed to rod signalling [26,28] and requires the integrity of ipRGC [23]. In our experiments we observed that the free running periods in LL were first increased to a similar extent for both the WT and mutant mice, suggesting that mechanisms distinct of the rod-pathway are involved. However, periods then decreased, with WT reaching a mean value around 24 h and the mutants rather lower periods (23.25 h on average). In addition, KO mice retained high level of activity in constant light, indicating loss of masking or of light aversion, a feature which is also shared with mice devoid of ipRGC [23] and has been proposed to mainly involve these cells [71] but also other photosensitive systems in the retina [76]. Thus, the exact mechanism underlying this phenotype here remains to be determined. Short free-running period values have been rarely described in LL, except in Per2 clock gene mutants of different backgrounds [77][78][79].
The phenotype in the Nrl -/mice could be explained by distinct hypotheses: 1, their high wheel running activity in LL might feedback on the clock and induce period shortening [80]; 2, the sustained activity might simply counteract the effects of (constant) light on neuronal activity in the SCN and counterbalance the period-lengthening mechanism [81]; 3, cone abundance could trigger another, yet unknown signalling towards the central clock.
Identification of the mechanisms by which excess of cones alters properties of the circadian system will require further investigation.
In conclusion, by using the Nrl -/cone-only mouse model we provide compelling evidence that cones contain a circadian clock part of the retinal oscillating network. Our data confirm the interest of visual system mutants in the understanding of retinal pathways regulating the central clock. Exposure of Nrl -/mice to specific experimental paradigms highlighted their particularities, namely loss of entrainment and negative masking induced by the absence of rods and, importantly, altered behaviour in constant light, specific to the enlarged cone population. Thus, cones are important players in the circadian system with distinctive properties and contributions, both as light sensors for the central clock and as elements of the retinal circadian system.