How Pupil Size Affects the Link Between Blue Light Exposure and Melatonin Reduction in Young Men

Data were recorded continuously between December 2019 and July 2021 with interruptions between mid-March and mid-May 2020 due to the global COVID-19 pandemic. A total of 72 healthy young men between the ages of 19 and 35 years (intensity 1: 24.5 ± 3.8 years; intensity 2: 25.4 ± 5.5 years; intensity 3: 24.3 ± 4.3 years; intensity 4: 24.7 ± 3.5 years) were included in the study. To avoid possible effects of the menstrual cycle on sleep (Driver et al., 1996) and melatonin secretion (Greendale et al., 2020), we did not include female participants (Vidafar and Spitschan, 2023). Participants were seen by a study physician and a graduated optometrist to exclude volunteers with visual impairments. Participants with monocular visual acuity <0.5 (Freiburg Visual Acuity Test Version 3.10.2 [Bach, 1996]), with color vision deficiencies (Ishihara < 17 of 21 plates [Ishihara, 1998], 100-hue error score >40, Farnsworth, 1943]), and with reduced stereoscopic vision (Lang II <200 arc sec [Lang and Lang, 1988]) were excluded from the study. Note that all inclusion and exclusion criteria of this study have been already published in Schöllhorn et al. (2023), where we reported melatonin, subjective alertness, and sleep latency data.

The study protocol was approved by the ethics commission northwest/central Switzerland (2019-00571) and conformed to the Declaration of Helsinki. All participants provided written informed consent and received compensation for participation. The study consisted of one habituation night and two experimental days. The experimental nights took place on the same day of the week and were usually exactly 1 week apart. Participants were balanced randomly and assigned to one of the four light intensity groups (n = 18, per group). Within each light intensity group, we counterbalanced the order of the two light conditions (HM and LM). The laboratory environment was controlled without any external time cues, including clocks, smartphones, and daylight. Note that the study protocol and light conditions have been already reported in Schöllhorn et al. (2023).

During the evenings of the two 17-h experimental visits, which only differed in the light condition (HM vs LM), participants came to the laboratory 7 h prior to habitual bedtime (see Figure 1). One hour of 67 lx fluorescent light (Philips Master TL5 HO 54W/830, CRI 80, 3000 K) was followed by two adaptation periods of complete darkness (~0.1 lx), separated by a period of dim light (~0-7 lx), which were used for photoreceptor sensitivity adaptation and baseline measurements. Participants were exposed to LM and HM conditions for approximately 3.5 h, starting 4 h before their usual bedtime. Before, during, and in the morning after LM and HM exposure, saliva samples were collected, electroencephalography was continuously recorded, participants performed cognitive tasks (i.e., psychomotor vigilance task, go/no-go performance task, word-pair learning task), subjective and objective alertness was assessed, and pupil size was measured. Here, we will only report on the results of the pupil data and the previously published melatonin data from Schöllhorn et al. (2023). During task-free periods, participants listened to prepared audio books and were instructed to look at the center of the screen. The light exposure was followed by 8 h of sleep in complete darkness and 1 h of dim light the next morning. Data on subjective sleep quality, which was assessed in the morning, will be published elsewhere.

The light was presented on a custom 27-inch visual display. The screen contained five different LED types (dominant wavelength: 430, 480, 500, 550, and 630 nm). The metameric screen was designed and calibrated using the method of silent substitution (Estévez and Spekreijse, 1982; Spitschan and Woelders, 2018) implemented in MATLAB (The Mathworks, Natick, MA). LEDs with wavelengths of 480, 500, and 630 nm were used for HM and LEDs with wavelengths of 430, 550, and 630 nm were used for LM. The light settings were matched in terms of cone excitation (L, M, and S) based on the 10° cone fundamentals using CIE S 026 (CIE S 026/E:2018, 2018). Adjustment of these five LEDs allowed for between 200% and 300% melanopsin contrast between the LM and HM conditions (for an overview of the light characteristics, see Table 1 and spectra are shown in Figure 1). Calibration measurements were performed with the JETI spectraval 1501 (JETI Technische Instrumente GmbH, Jena, Germany) at an eye level at a distance of 60 cm from the center of the screen. The spectrometer was inclined at an angle of 15° to the center of the monitor (Schöllhorn et al., 2023).

The participants’ pupil diameter of the right eye was measured during the test sessions using an eye-tracking device (Pupil Labs GmbH, Berlin, Germany), which allows for the recording of pupil sizes at 200 Hz. When participants arrived in the laboratory, screen marker calibration choreography provided by Pupil Labs was performed. The algorithms of Pupil Core automatically run two detection pipelines in parallel, two-dimensional (2D) and three-dimensional (3D). The median pupil size of 20-min bins during each of the five test sessions (go/no-go task, Karolinska drowsiness test, and psychomotor vigilance task) was computed for statistical analyses. Therefore, we used the provided data of the 3D Pupil Detection (Dierkes et al., 2019; Swirski and Dodgson, 2013). Prior to statistical analysis, pupil measurements with a confidence of <0.6 and pupil diameters less than 2 mm and greater than 10 mm were excluded to remove artifacts and eye blinks. We then calculated the median pupil size per part for statistical analysis. Due to technical problems (i.e., loss of data due to storage problems), we lost the pupil recordings from three test visits (~2%) and a further seven test parts. In addition, we excluded data from 24 test parts (3.33%) due to poor data quality (i.e., <2% of available data). Therefore, for statistical analysis, the recordings of 674 (94%) test parts were used.

Retinal irradiance cannot be measured directly in visual experiments. Therefore, to calculate a conventional retinal irradiance (measured in trolands), the product of luminance and pupil area is used (Thibos et al., 2018; Wyszecki and Stiles, 1982). Here, we additionally determined the α-opic retinal irradiances, measured in trolands, for the five human photoreceptor classes (melanopsin, rods, L-, M-, and S-cones) using spectral radiances, pupil areas, and spectral sensitivity functions based on the 10° cone fundamentals using CIE S 026 (CIE S 026/E:2018, 2018).

All statistical analyses were conducted in R (Version 4.1.1, R Core Team, 2020). Linear mixed model (LMM) analyses were performed separately for each light intensity group. “Light Condition” and “Time of day” were included as fixed effects and repeated measures per participant were modeled as random intercept. LMM analyses were followed by an analysis of variance (ANOVA) (Type = III) function. LMM analyses were calculated using the packages lme4 (Bates et al., 2015) and lmerTest (Kuznetsova et al., 2017). As an effect size measure, partial omega squared (ωp2) was calculated using the effectsize package (Ben-Shachar et al., 2020). It can be interpreted as follows: small effect: ωp2 ≥ 0.01, medium effect: ωp2 ≥ 0.06, and large effect: ωp2 ≥ 0.14. A value of p < 0.05 was considered indicating statistical significance. Contrast tests for significant main effects of “Time of day” or significant interactions were performed using the emmeans package (Lenth, 2021) with Kenward-Roger degrees of freedom and a Tukey adjustment for multiple comparisons. Linear regression models were used to describe the relationship between mEDI and the observed variables.

The time course of median pupil diameter is illustrated in Figure 2a. During light exposure, pupil diameter was significantly larger in the LM compared to the HM condition in all four light intensity groups (p < 0.001, large effects, see Table 2). The main factor “time of day” was only significant in light intensity group 4 (F_1,80 = 3.21, p < 0.05, ωp2 = 0.05). However, post hoc comparisons did not show significant differences (p > 0.05). Within the four light intensity groups, the approximately 3× reduction in melanopic irradiance caused an average increase in pupil size between 16% and 23%, depending on the different light intensity groups (i.e., intensity 1: 22.6%, intensity 2: 20.4%, intensity 3: 21.1%, and intensity 4: 15.6%). The interaction between “Light Condition” and “Time of day” was only significant in intensity group 2 (Light condition * time of day: F_2,81 = 3.43, p = 0.04, ωp2 = 0.19; session 2: HM − LM: t₍₈₁₎ = −5.44, p < 0.001, session 4: HM − LM: t₍₈₁₎ = −4.04, p = 0.002).

Grouping the data for pupil size according to the mEDI resulted in a significant dose-response relationship (R = −0.7, p < 0.01), such that higher mEDI levels were associated with a smaller pupil size (see Figure 2b).

Retinal irradiances of HM versus LM showed a high correlation (R = 0.92, p < 0.01). Given that pupil size, and therefore retinal irradiance, was higher in LM than in HM within each light intensity group, the regression line was constantly shifted in a more positive direction (see Figure 2c).

Using α-opic spectra for weighting retinal (i.e., melanopic trolands) and corneal illuminance (i.e., mEDI) of the different photoreceptor types indicates higher melanopic trolands with increasing mEDI independent of the LM and HM condition (see Figure 3d and 3e). While our light conditions were matched for L-, M-, and S-cone-opic weighted EDIs, retinal irradiances for the three cone types were slightly higher during LM compared to HM, thanks to the difference in pupil size (see Figure 3a - 3c).

Although pharmacological pupil dilation has been shown to be a significant determinant of melatonin suppression, it is not a condition that occurs under natural lighting conditions (Giménez et al., 2022). Whether the use of melanopsin-weighted retinal irradiance by including pupil size is a better predictor of melatonin concentration than melanopic irradiance (i.e., mEDI) at relatively low light levels (<90 lx) remains to be determined. Therefore, we used the melatonin area under the curve (AUC) previously published by Schöllhorn et al. (2023) and compared linear fits using mEDI and melanopic trolands to predict evening melatonin AUCs. Dose-response relationships for melanopic EDI or melanopic trolands and melatonin AUCs showed similar fits by linear regression (R = −0.37, p < 0.01; R = −0.38, p < 0.01, respectively). Hence, the model fit could not be improved qualitatively by including pupil area (i.e., melanopic trolands) in the estimation of evening melatonin AUCs compared to mEDIs (see Figure 4).