No evidence for an S cone contribution to acute neuroendocrine and alerting responses to light

Summary Exposure to even moderately bright short-wavelength light in the evening can strongly suppress the production of melatonin and delay our circadian rhythm. These effects are mediated by the retinohypothalamic pathway, connecting a subset of retinal ganglion cells to the circadian pacemaker in the suprachiasmatic nucleus (SCN) in the brain. These retinal ganglion cells express the photosensitive protein melanopsin, rendering them intrinsically photosensitive (ipRGCs). But ipRGCs also receive input from the classical photoreceptors — the cones and rods. Here, in human participants, we examined whether the short-wavelength-sensitive (S) cones contribute to the neuroendocrine response to light by using stimuli which differed exclusively in the amount of S cone excitation by almost two orders of magnitude (ratio 1:83), but not in the excitation of long-wavelength-sensitive (L) and medium-wavelength-sensitive (M) cones, rods, and melanopsin. We specifically examined the S cones since the previously published action spectra for melatonin suppression [1,2] pointed to a possible role of S cones in addition to melanopsin. We find no evidence for a role of S cones in the acute alerting and melatonin-supressing response to evening light exposure.

Data exclusion and missing data. We had to exclude two of the 17 participants in our analysis. Our analyses of the melatonin, sleepiness (KSS), and vigilant attention (auditory RT) data do not include these two participants. One participant was excluded due to stimulus mistiming (melatonin concentrations never exceeded 5 pg/mL). The other one was excluded due to implausible high melatonin levels (>300 pg/mL) in one session, possibly pointing to contamination of the samples.
In some samples of the remaining participants, the assays returned implausibly high melatonin samples (e.g. >30 pg/mL six hours before habitual bedtime), in which case we detected and removed outliers across participants but within-condition using the iterative generalized extreme Studentized deviate test for outliers (implemented in MATLAB's isoutlier function). This affected 20 samples of the total 420 samples (15 participants × 2 sessions × 14 samples per sessions), leading to an exclusion rate of <5% of samples.
We generated our S-cone-selective stimuli using the method of silent substitution [S6,S7]. In the method of silent substitution, pairs of spectra are generated as mixtures of the ten primaries lights which produce a difference in only one photoreceptor class (in this case, the stimulated S cones), while there is no difference in the other photoreceptors (in this case, the silenced L and M cones, rods, and melanopsin). This method has previously been used to examine the effect of melanopsin-only differences in lighting on melatonin suppression [S8, S9] (but has a long history in vision science, see [S7]).
To produce calibrated stimuli, we first measured the spectral radiance of each LED independently at 19 intensity levels (spaced at 5% increments from 5% to 100%, where 100% is maximum intensity) using a spectroradiometer (spectroval 1511, JETI Instruments GmbH, Jena, Germany). We addressed the typical changes in spectrum with increasing intensity by relying on an interpolation-based forward model of our primaries (interpolating at unmeasured primary settings). Using this model, we generated two sets of settings for our primaries which would have the feature that they yielded maximum differential stimulation on the S cones, with minimal change in L and M cone, rod and melanopsin stimulation. These settings were simultaneously found using constrained minimisation routines implemented in MATLAB (fmincon SQP solver with global optimisation; 1000 trial points). In this procedure, we used the cone, rod and melanopsin spectral sensitivities [S10] comprising the 10° Stockman-Sharpe cone fundamentals [S11], the CIE V'(λ) function for the rods, and the standard curve for melanopsin [S12]. Irradiance spectra measured in the corneal plane from the observer's point of view are given in Table S1.
We quantified the difference in S cone excitation by calculating the Weber contrast (percentage change in S cones from S-to S+) and as a factor (ratio of S cone excitation between S+ and S-). Both numbers are equivalent representations of the stimulus change. We report both numbers for completeness. For clarity, in the main text, we round the factor down to the nearest integer so as not to overstate the stimulus change.
We achieved a stimulus with a difference of 8268% (factor 83.68×), or equivalently almost two log units (~1.92 log difference), in S cone stimulation, with minimal stimulation of L and M cones, rods and melanopsin. The photopic illuminances were 168 lux for the S-condition (0.48, 0.26; 'orange' appearance) and 173 lux for the S+ condition (0.61, 0.37; 'pink' appearance). The melanopic irradiance was 59 mW/m 2 for the S+ condition and 53 mW/m 2 for the S-condition. These background radiances correspond to moderate photopic light levels.
Validating the spectra from this optimisation procedure, our stimuli demonstrated excellent silencing for the L and M cones ( Fig. 1; -3% L cone contrast, -1% M cone contrast), and very good silencing for rods (-18%) and melanopsin (+11%), while providing an almost two log unit difference S cone stimulation. It is unlikely that these small nominal differences produce a meaningful physiological difference, given the very large and to our knowledge unparalleled difference in S cone stimulation.
Protocol. The study took place in a dedicated light-, temperature-and humidity-controlled apartment comprising a double-room as well as a dedicated bathroom (see Appendix A in [S13] for photograph). Upon arrival (30 minutes prior to protocol start), participants gave a urine sample for drug test (multi-drug panel test for AMP, BZD, COC, MOR/OPI, MTD and THC; exclusion if positive; nal von minden, Den Haag, Netherlands) and accommodated to the laboratory. Then, the protocol began, lasting from 6.5 hours before habitual bedtime to habitual bedtime. Every 30 minutes, participants completed an alertness assessment using a simple auditory reaction time task, the Karolinska Sleepiness Scale (KSS), and gave a saliva sample using Salivettes in dim light provided by room illumination (photopic illuminance in the corneal plane <8 lux). From 2.5 hours to 0.5 hours prior to the habitual bedtime, participants were either exposed to the S-or S+ stimuli in 20-minute sections under steady fixation, yielding a total of 80 minutes of light exposure to the experimental stimuli. Between the 20-minute sections, participants completed the questionnaire, performed the PVT and gave the saliva sample under the dim <8 lux lighting. This protocol balanced feasibility of light exposure with the possibility that cone responses might adapt during long-term light exposure.
Fixation and eye opening were verified using a video-based head-mounted eye tracker (Pupil Labs GmbH, Berlin, Germany). Participants had access to water throughout the experiment but no food or other drinks. Participants were allowed to spend their time reading, studying, playing Nintendo GameBoy (illuminance at cornea <8 lux), or other activities not involving additional light exposure. Smartphones and other electronic devices were removed from the experiment suite. All experiments took place between November 2018 and June 2019. All sessions took place one week from another and condition order was randomised between participants. From one week prior to the experiment to the second session, participants were instructed to adhere to regular bedtimes (±30 minutes) and wore actigraphy devices (Condor Instruments, São Paolo, Brasil). On the day of the experiment, participants were asked to refrain from caffeine consumption after noon.
Vigilant Attention. Vigilant Attention was measured using a custom-made simple auditory reaction time task programmed in Psychtoolbox and MATLAB (The Mathworks, Natick, MA). Participants were presented with a tone emitted from a loudspeaker and were instructed to press as quickly as possible to the tone using a PlayStation-like gamepad. ISI was randomly set to 5-8 seconds. Median reaction times were calculated from 50 trials.
Statistical analysis. We modelled our data using a linear mixed-effects model, modelling subjects as a random-effects, and condition (S+ or S-) and sample number (with sample #14 corresponding to habitual bedtime) as fixed effects, along with the interaction between condition and sample. In Wilkinson-Rogers notation, the full model (M1) is specified as outcome ~ Sample + Condition + (1|participant).
The null model (M0, no effect of S cone manipulation) is specified as To estimate the evidential strength for an S cone manipulation, we calculated Bayes factors (BF) using the R package 'BayesFactor' (version 0.9.12-4.2) [S15-S17]. Compared to traditional null hypothesis significance testing, this approach allows for assessing the evidential strength of competing models. Bayes factors specify the ratio of the marginal likelihood of two competing models. We used standard scales for interpreting the Bayes factor [S18] and considered both the full data (all data points) and the data points only during the light exposure. The resulting Bayes factors are given below. Ethical approval. This study was approved by the cantonal ethics commission (Ethikkommission Nordwest-und Zentralschweiz, PB_2018-00164 -280/90) and was conducted in accordance with the Swiss law and according to the Declaration of Helsinki.