Comparison of oxidative stress response of in vitro retinal cells exposed to blue light from emissive versus reflective displays

During COVID‐19, there was increased use of handheld displays in educational settings. There is growing concern that eye health may be affected by prolonged exposure to the light‐emitting diodes used as frontlights or backlights in handheld displays. The potential impact of light exposure from tablet‐sized devices with different display technologies and various spectral outputs was assessed in an in vitro model using human retinal epithelial (ARPE‐19) cells. Cellular response was quantified by measuring reactive oxidative species (ROS) and by analyzing mitochondrial morphology. Control experiments established a baseline ROS response to hazardous blue light exposure and also that red light resulted in no detectable ROS response. Under identical conditions, ROS response increased with time for all devices. However, different device spectra caused ROS to accumulate at different rates. When operating the devices in the same mode (day or night), cells accumulated ROS two to three times more slowly on exposure to frontlit electronic paper displays compared to backlit liquid crystal displays. With increasing ROS accumulation, mitochondrial morphology shifted from elongate interconnected features typically observed under normal conditions to rounded disconnected features associated with oxidative stress response.


| INTRODUCTION
Light emitted from the screens of handheld tablet-sized devices used for prolonged periods of reading and study is suspected to have a significant impact on eye health, in particular on retinal cells. Compared to television screens, these small screens are used at close reading distances. Light emitted from such displays not only is more intense than light reflected off objects in the ambient environment but also contains a higher proportion of blue light in comparison to natural light. This spectral difference is mostly due to the light-emitting diode (LED) sources that are used to illuminate their screens, as well as the increased intensity typically required to overcome ambient light reflected off the screen when viewing the device.
According to the Eyesafe ® Display Requirements 2.0, "blue light, also known as high-energy visible (HEV) light, is the portion of the visible light spectrum that has the shortest wavelengths and therefore the highest amount of energy. Blue light is visible light with wavelengths ranging from 400 to 500 nm." 1 Because of the growing concern that prolonged exposure from LED sources used in handheld displays may contribute to visual impairment, the Chinese Government in 2020 mandated an upper limit for online study of 4 h/day for middle school children. 2 Further complicating the issue, children's corneas absorb less short-wavelength light than adults'. Thus, when presented with the same spectrum, more blue light reaches children's retinas than adults'. 3 Laboratory studies theorized that blue light exposure reduces cell viability because blue light is absorbed by the retinal pigment epithelium generating localized oxidative and thermal stress. 4,5 Cellular exposure to oxidative stress results in increased levels of reactive oxidative species (ROS) in intracellular biochemical pathways within cells, altering mitochondrial metabolism response to reduce energy producing cellular components necessary for intracellular repair and maintenance. Mitochondrial morphology is altered in the presence of increased levels of ROS in cells changing from elongated interconnected structures to disjointed mitochondrial fragments. It has been demonstrated that mitochondria fragmented morphology is linked with a decrease in mitochondrial length and an increase in mitochondrial circularity. [6][7][8][9] Emissive displays, for example, liquid crystal displays (LCDs), use LED sources to backlight the display screen. The white light emitted from the LEDs must pass through a stack of functional optical layers that include diffuser, polarizer, the array of electronically actuated liquid crystal shutters, color filter array, and touch panel; thus, the backlighting LEDs must be quite intense. Typically, white LEDs have a blue-emitting diode with maximum radiance at around 450 nm, coated with yellow yttrium aluminum garnet (YAG) phosphor. In some instances, the color gamut of the display can be increased by replacing the yellow phosphors with narrow-band green and red phosphors. 10,11 In contrast to emissive displays, information displayed on electronic paper (ePaper) devices is viewable due to the reflection of ambient light off the display screen, just like normal paper. In electrophoretic ePaper displays (EPDs), white pigments diffusely reflect ambient light while black or color pigments modulate this reflection according to the information to be displayed. 12,13 To make EPD usable in dark viewing environments, however, a light source must be provided, just like with normal paper. Frontlights consisting of an edge-lit light guide plate (LGP) with white LED light sources are added to the optical layer stack above the reflective pigment layer. This is different to LCDs, where the LEDs are backlighting the transmissive liquid crystal array.
Thus, the effect of reflected ambient light is fundamentally different for frontlit reflective and backlit emissive displays. For a reflective display, the reflected ambient light carries useful information modulated by the screen's pigments. Unless the frontlight is turned on, the eye never receives more light than present in the ambient environment. However, for an emissive display, reflected ambient light is a disturbance because it carries no information and its reflection off the screen makes it harder to view details shown by the display. In fact, emissive displays, such as phones and tablets, are designed to mitigate the contrastreducing effect of reflected bright daylight by increasing white screen luminance (brightness) to 500 cd/m 2 or more. 14 Such levels of luminance are striking because handheld displays are viewed directly, from a close distance, and for long periods of time up to several hours. In contrast, when a frontlight is used on an EPD in the presence of ambient light, the white EPD pigments reflect light from both light sources so the total display luminance is the linear combination of the luminance components from the reflected ambient and reflected frontlight. The frontlight of an EPD is only necessary to replace ambient light in dark environments, so its white screen luminance is much lower, typically only about 100 cd/m 2 .
A large body of literature has accumulated on how to quantify interactions with, for example, digital display screens. A human observer viewing a display experiences brightness and color, measurable as spectral radiance L (λ) of the light emitted and reflected from the display. The display's spectral radiance depends on the displayed color Q. The highest level of spectral radiance is emitted when displaying white; the spectral radiance of the white screen is L W (λ). When the optical system of the human eye images the display onto the retina, the spectral radiance of the light emitted from the display is transformed into a spectral irradiance E W (λ) incident on the retina. The conversion from radiometric spectral radiance L W (λ) or spectral irradiance E W (λ) to the photometric equivalents luminance L W or illuminance E W is performed via spectral weighting with the CIE standard spectral luminous efficiency function V(λ) for photopic vision as defined in ISO 23539: with K m as the maximum luminous efficacy for photopic vision of 683 lm/W at 555 nm. The International Commission on Non-Ionizing Radiation Protection (ICNIRP) developed guidelines and metrics for retinal exposure limits from incoherent broadband visible and infrared sources. 15 It defined a weighting function for blue-light photoretinopathy with lens (phakic) B(λ), also referred to as blue-light hazard function, with peak sensitivity to blue-light hazard at 435 to 440 nm. The blue-light hazard function was endorsed by the International Safety Equipment Association (ISEA) and adopted by the American National Standards Institute (ANSI) in the ANSI/ISEA Z87.1-2020 standard. 16 The ICNIRP studied potential biological hazards of white LED light sources with different color temperatures and stated that these sources should be judged by the safety standards for broadband lamps, not lasers. 17 The Eyesafe ® Display Requirements 2.0 focuses on HEV light emitted by digital devices. It defines blue light as "visible light with wavelengths ranging from 400 to 500 nm," also referred to as HEV light. To be certified as Eyesafe ® , a display must "limit light emissions in the blue-violet segment of the blue light spectrum" while "maintaining Color Gamut and CCT (correlated color temperature) requirement values." 1 The Eyesafe requirements and certification are based on the Blue Light Toxicity Factor (BLTF). 18 BLTF compares a display's white screen spectral radiance L W (λ) between 380 and 780 nm, weighted with the blue-light hazard function B(λ), to its spectral radiance weighted with the CIE 1931 photopic luminous efficacy function V (λ): According to Eyesafe ® guidelines, the ratio of the display's total hazardous blue radiance over its luminance should be less than 0.085. 18 For the purpose of certification, the Eyesafe ® Display Requirements 2.0 introduced the Radiance Protection Factor (RPF ® ) for Display, which translates the BLTF measured for the display into an RPF scale of 0 to 100.
RPF normalizes the BLTF measured for the display to the BLTF of the CIE D65 illuminant. Similar to the Sun Protection Factor (SPF), the RPF evaluates the eye protection against potential risk of damage from the blue light emission of displays, with higher RPF indicating better protection. However, this study will focus on BLTF, not RPF, because EPD frontlights are typically used at lower luminance levels than the 200 cd/m 2 specified for RPF comparisons. 1 2 | MATERIALS AND METHODS

| Study display devices for spectral exposures
The study tested the effect of spectral exposure from different handheld displays on basic cellular functions, with the aim to establish whether the direct exposure from these display devices triggers measurable biological effects in a controlled environment. The four study devices listed in Table 1  of various spectral outputs: one tablet having a color LCD screen with backlighting and three eReaders having B&W or color EPD screens with front lighting. Devices 1, 2, and 3 each have user-controlled options for day and night modes that change the color of the display light from cold white for daytime viewing to a warm white with less blue light for nighttime viewing. Device 3 has an experimental frontlight that differs from the others by using LEDs with the blue emission shifted to 460 nm, away from the peak of the blue-light hazard function. Device 4 was chosen as a commercially available color EPD. This set of displays was chosen to cover a wide range of illumination spectra for studying the effect of the display spectrum on the retinal cell response. The technical details of the display illumination spectra will be discussed in Section 3.1.
For all experiments, each study display was set to a full-screen, full signal white, using an 8-bit input signal of R = G = B = 255. The screen luminance was set to its maximum level. For the day and night modes, the control settings of the user interface, for example, sliders, were set to their maximum levels.
Spectral radiances of study displays were measured between 380 and 780 nm at 1-nm intervals in darkroom conditions using a factory-calibrated spectroradiometer CS-2000 (Konica-Minolta Sensing, Ramsey, NJ, USA). Each measurement was taken at the center of the display from a normal direction at a distance of 500 mm. The spectroradiometer was set to a measurement field angle of 1 and focused at the display surface.

| Spectral light source for control exposures
For control exposures of ARPE-19 cells, the spectral light source SpectralLED ® RS-7-3 (Gamma Scientific, Inc., San Diego, CA, USA) was used. It synthesizes custom irradiation spectra by combining the spectral output of 35 different color LEDs. This allowed comparing the cell response to spectral stimuli within versus above the spectral range of the blue-light hazard function B(λ) (see Section 3.2). To enable cell plate exposures, the source was combined with a fiberoptic backlight A08925 (SCHOTT North America, Inc., Rye Brook, NY, USA) having an emissive area of 102 mm Â 124 mm. When placed in an incubator, the backlight was connected to the source via a 1016-mm-long fiber bundle through a port in the incubator. Gamma Scientific performed a National Institute for Standards and Technology (NIST)traceable calibration of the spectral radiance emitted by the backlight.

| Cell culture
For in vitro cell culture experiments, the human retinal pigment epithelial (RPE) cell line ARPE-19 was used. ARPE-19 catalog number (ATCC CRL-2302) was purchased from American Type Culture Collection (Manassas, VA, USA). The rationale for using this cell line is that (1) it is a neurosensory cell derived from the eye, (2) it is responsive to oxidative stress challenges, and (3) it can be maintained as a monolayer. The purpose of this cell line was to investigate oxidative stress in the cell line only. All experiments were performed by using ARPE-19 between Passages 10 and 20. All reagents used in this study were purchased from Sigma-Aldrich (St. Louis, MO, USA), or listed and noted from specific vendors.
Cells were cultured and maintained in Dulbecco's modified Eagle's medium: Nutrient Mixture F-12 (DMEM: F-12 Medium [ATCC 30-2006]) supplemented with 10% fetal bovine serum (Gibco, Life Technologies) at 37 C in humidified 5% CO 2 incubator. Cell culture media were replaced every 2 days for culture maintenance and maintained for 4-6 months. As standard practice, cultured cells were constantly monitored for bacterial contamination infection. For splitting between passages and seeding into multi-well plates for use during their exponential growth phase, cells were passaged using Accutase (STEMCELL Technologies, Cambridge, MA, USA) when they reached about 80% confluence. To test for cell viability, cultured cells were released from tissue culture substrate and centrifuged at 300 Â g for 5 min. Ten-microliter aliquots of cell suspension were mixed with Trypan blue dye in a 1:1 ratio and counted using Countess II FL Automated Cell Counters (ThermoFisher Scientific, Waltham, MA, USA).

| Setup for cell exposure from study display devices
To mimic retinal irradiance from display screens in cell cultures, cell plates are irradiated by the screens of each study device listed in Table 1. For control experiments using synthesized spectra, the calibrated fiberoptic backlight connected to the spectral light source was used.
If a screen with a diagonal dimension D is mounted close to a cell plate at a distance z, the spectral irradiance E W (λ, z) the cell plate receives from the display when emitting a spectral radiance of L W (λ) can be expressed as For close contact between screen surface and cell plate, the distance z becomes so small relative to the screen diagonal D that Equation (4) can be approximated as 19 This near-distance approximation simplifies the comparison of displays with different screen sizes. Light emitted from a display with a spectral radiance L W (λ) is stimulating the retinal cells as spectral irradiance E W (λ). The photochemical and photobiological response to this light stimulus is cumulative, increasing over the time the stimulus persists. Therefore, the cumulative time dependence of the light stimulus is expressed as spectral exposure H W (λ), the product of spectral irradiance E W (λ) and exposure time t 20 : At a 20-mm distance between screen and cell plate, a 3.6inch white screen with a luminance of 100 cd/m 2 will illuminate the cell plate with 300 lx, which corresponds to an irradiance of 0.44 J/(s m 2 ) at 555nm. The exposure from hazardous blue light H HB (λ) is calculated from the total hazardous blue irradiance E HB (λ), by weighted integration of the measured spectral radiance L W (λ) with the blue-light hazard function B(λ): Photobiological cell responses to light follow a sigmoidshaped exposure-response curve, the four parameters of which can be fitted to measured data of ROS response versus exposure by least-square regression. The ROS response to an exposure H is where ROS 0 is the minimum ROS response, a is the difference between the maximum and minimum ROS responses, H 0 is the exposure for a half-maximal response (the 50% value), and b is the slope parameter. 20 During the exposure experiments, the incubator was set to 37 C but the study display devices (Table 1) had to be maintained at an operating temperature of 25 C. This was achieved by placing each device on top of an aluminum heat exchange plate with recirculating water bath (Thermo NESLAB RTE 7) set to 25 C, as shown in Figure 1. The cell plates with 96 wells containing cells were supported by a 3D printed plastic frame and positioned to maintain a constant distance of 20 mm between the cellular monolayers and display surfaces. Small airrecirculating fans were built into the frame to maintain constant temperature of 37 C inside all well compartments containing adherent cells and culture media.
Exposure times were set to four time points in geometrical succession: 30, 60, 120, and 240 min (4 h). Predefined sections of wells were shielded from irradiation with aluminum foil to obtain dark control data (no light exposure). Experiments longer than 4 h were not conducted due to technical equipment limitations as well as recommendations by government agencies for maximum daily display viewing times. 1 For each exposure experiment, the cell plates were filled with cultured retinal cells. ROS response to exposure was measured with five replicates (five sections of the cell plate) and one dark control (the sixth section of the cell plate shielded from irradiation). Each exposure experiment was then repeated four times for statistical analysis (shown as error bars in the diagrams showing ROS responses). Although more experimental repeats might have been desirable, replicates were limited by the cell plate area (= display size), and repeats by the life span of cell cultures. The measured ROS response at each time point was F I G U R E 1 Plate cooling system designed and operational within cell incubator with study display on top of cooling plate (left). Custom-designed 3D printed cooling fan assembly (center); temperature monitoring of experiment (right). During the exposure experiments, the incubator is closed, and light exposure of cells is only from the study display. normalized by subtracting the dark ROS response, generated by cells in the wells shielded from irradiation.
Each display was set to its maximum white screen luminance (see Table 1). In order to determine the exact exposure for each cell plate, the white screen luminance L W was measured before and after each cell plate exposure in the incubator, using the contact luminance meter Mavo Monitor M504G (Gossen Foto-und Lichtmesstechnik GmbH, Nürnberg, Germany).

| Oxidative stress analysis
ARPE-19 cells were seeded at 20Á10 3 cells per well in black-walled 96-well plates (Millipore Sigma, St. Louis, MO, USA) overnight at 37 C. Reactive oxidative species (ROS) expression levels were measured immediately following exposures from study devices at the four exposure times using ROS-Glo™ H 2 O 2 luminescence assay (Promega, Madison, WI, USA) according to the manufacturer's instructions. 21 ROS (e.g., superoxide, singlet oxygen, and H 2 O 2 ) have a short half-life in solution and can be converted to H 2 O 2 in cells by enzymatic or nonenzymatic reaction rapidly. 22 ROS-Glo™ H 2 O 2 is a rapid, robust, and sensitive method to measure the H 2 O 2 levels directly in cell culture. 23 In this assay, a positive control was used by incubating cells with 10-μM menadione, a known biochemical inducer of oxidative stress that was diluted in cell culture medium for 4 h at 37 C. To activate the fluorescent response designed into the assay, H 2 O 2 substrate was added to cell wells and incubated for 2 h. The luminescent signal was measured after 20-min incubation with ROS-Glo reagent at room temperature. The final luminescence of samples was measured using a Biotek Synergy 2 luminometer (Winooski, VT, USA) and represented with relative luminescence unit (RLU). Three biological replicates were analyzed in quadruplicates. Each measurement was normalized to the dark (no light exposure) ROS response control.

| Mitochondrial morphology
Sterile round 12 mm, diameter #1.5 thick gelatin-coated coverslips (NeuVitro, Camas, WA, USA) were placed in 24-well plates and seeded with 40,000 ARPE-19 cells maintained in culture medium until $50% confluency. For mitochondrial labeling, cells were exposed to the study devices for 240 min, as described in Section 2.3.2, and then MitoTracker™ Fluorescence dye (Invitrogen, Waltham, MA, USA) was added to coverslip containing wells and stained for 60 min according to the manufacturer's instructions. Adherent cells were washed with pre-warmed dye-free Hanks Balanced Salt Solution (HBSS) and fixed in the 4% paraformaldehyde at room temperature for 10 min. The coverslips were mounted onto Superfrost Plus microscope slides (VWR, Radnor, PA, USA) with 40-μL VECTASHIELD Vibrance™ Antifade Mounting Medium with DAPI (Vector Laboratories, Newark, CA, USA) to counterstain nuclei. Cells were analyzed using a Zeiss LSM 780 confocal laser scanning microscope (Carl Zeiss AG, Jena, Germany). Images of stained cells were captured using the oil immersion objective C Plan-Apochromat 63x/1.4 Oil DIC M27. DAPI stained nuclear profiles and mitochondria features were imaged with a 405-nm diode laser, argon ion laser, and red shifted diode laser within the microscope. Cell profiles beginning at the coverslip surface were collected as serial optical sections at 0.5-μm step intervals and then converted to three-dimensional reconstructions using arivis Scientific Imaging Platform (arivis Imaging Inc., Boston, MA, USA).
Image processing and morphological measurements of mitochondria circularity (elongation) and interconnectivity were obtained from stacks of serial sections using arivis. Circularity is calculated as the deviation from the sphere measured by the maximum chord length in threedimensional space. Interconnectedness is the number of volumes physically touching each other with a minimal number of five voxels. Between 10 and 30 observations were made of mitochondrial morphologies from different cells depending on the sample. These data were used for qualitative analysis only.

| BLTF from measured spectral radiance
To determine the BLTF for each study display in the available day and night light modes, each screen was set to its maximum luminance level (see Table 1), and the white screen spectral radiances L W (λ) were measured in darkroom conditions.
The results are shown in Figure 2. For comparison, each spectrum was normalized to the same luminance of white L W = 100 cd/m 2 using Equation (1), which is similar to tuning each display to the same level of perceived brightness. Although the devices show significant differences from each other in the spectra of the blue emitters as well as the phosphors, blue emission is generally reduced in night light mode.
From the measured spectral white screen radiances L W (λ) shown in Figure 2, luminance L W , BLTF, CCT, and white screen colors were calculated (see Table 2).
The white screen luminance of the EPD samples with frontlight could not reach the 200 cd/m 2 specified for RPF comparisons, so no RPF comparisons are made in this study. EPDs illuminated by ambient light or frontlight are viewed at lower luminance levels of around 100 cd/m 2 . In comparison, print paper of 84% reflectance illuminated by 300 lx has a luminance of 80 cd/m 2 . The white screen colors illustrate the trade-off between lower BLTF (less blue light) and color quality of white (yelloworange tint).
Device 1 (color LCD) has a backlight white LED with a blue emitter at 450 nm, a green phosphor at 536 nm, and a red KSF phosphor with triple peaks at 614, 631, and 649 nm. It switches to night mode by changing the LCD screen color, reducing the white screen color temperature from about 6500 to about 2700 K. Device 2 (B&W EPD) has a frontlight with two types of LED. It fails BLTF in day mode where a cold-light LED (7400 K) with a strong blue emitter at 450 nm and a YAG phosphor is used, but passes BLTF by a very wide margin in night mode where an amber LED (2300 K) with a low-intensity blue emitter at 447 nm and an amber phosphor is used.
Device 3 (color EPD) has a frontlight with two types of LED, cold white with green and red KSF phosphors, and warm white with an amber phosphor. For both LEDs, the blue emission is shifted to 460 nm, away from the peak of the blue-light hazard function. To reduce the color shift between day and night modes, light from the cold-and warm-white LEDs is mixed at 75% cold plus 25% warm in day mode and 25% cold plus 75% warm in night mode. Device 3 is passing BLTF in both day and night modes with comfortable margins while showing a less tinted white in night mode compared to Devices 1 and 2.
Device 4 is similar to Device 3 in day mode, but with a blue emitter at 450 nm, passing BLTF with a narrower margin. 3.2 | ROS response to exposure from the calibrated spectral light source The tunable spectral backlight was used for model experiments to determine which part of the visual spectrum triggers the ROS response. Two custom spectra were used for exposure in the two distinct spectral regions shown in Figure 3: blue light within the range of the blue-light hazard function B(λ) and peak emission at 450 nm, and red light above 600 nm, that is, above the spectral range of the blue-light hazard function. The exposures were calculated using Equation (5). The measured ROS responses were normalized by subtracting the ROS base rate, generated by cells kept in cell plate wells shielded from radiation.
The plots of ROS response versus exposure in Figure  4 show that the cells responded to blue light exposure but not significantly to red light exposure at wavelengths above the range of the blue-light hazard function. Fitting the data to the sigmoid-shaped exposure-response model in Equation (8) shows that the measured exposure range covers only the toe and linear portion of the sigmoid model.
The model experiments using a calibrated spectral light source yield the baseline ROS response of retinal cells to blue light exposure in units of Joules per square centimeter.

| ROS response to exposure from the study devices
Display devices emit light that is perceived as white, not blue light only. The following experiments using the study display devices explore the effect of their complex spectral stimuli on the cellular ROS response. The measured ROS responses were each corrected by subtracting the measured dark-level ROS response of the cells in wells shielded from exposure. Figure 5 summarizes the corrected ROS responses for all devices in day and night light modes.
The levels of ROS increase over time and at different rates for the different devices. The highest increase of ROS over time is exhibited for Devices 1 and 2, each in day mode. This corresponds to both of them failing the BLTF < 0.085 requirement (see Table 2). The ROS responses to Devices 1 and 2 in night mode are much smaller but still increase over time. The ROS responses of Device 3 (in day and night modes) and Device 4 are not significantly different from each other, which corresponds to each passing the BLTF requirement.
The model experiments with the spectral illuminator demonstrated that the retinal cells respond to blue light exposure. But the study devices differ in their blue light spectral distributions; some contain more hazardous blue light, whereas others less. As BLTF evaluates the entire device spectrum, a correlation between BLTF and ROS response would indicate that not only hazardous blue emission but also the entire device spectrum has an effect on ROS response. Figure 6 plots ROS response versus BLTF for the three devices that have day and night modes and for each of the four exposure times. It shows that ROS response increases with BLTF linearly, with R 2 ranging from 66% to 72% depending on time.
For comparison of the different study devices, the ROS responses were plotted over the exposures to hazardous blue light, using weighed integration of the measured spectral radiances L W (λ) with the blue-light hazard F I G U R E 3 Spectral radiances of the spectral backlight tuned to blue light (400 to 500 nm) and red light (>600 nm).
F I G U R E 4 Reactive oxidative species (ROS) response (normalized by subtracting dark response) versus exposure to the blue and red light shown in Figure 3, fitted to the sigmoid response model. function B(λ) in Equation (7). Referring the ROS response to the exposure accounts for the differences in luminance between the different devices. Figure 7 shows the ROS response to hazardous blue light exposure from each of the devices compared to the baseline response. The device response curves differ from the baseline response. They are shifted horizontally along the exposure axis, suggesting that the cells respond differently to the same levels of hazardous blue light exposure received from the different devices in different light modes (day and night). As each device differs in its radiance spectrum, there is indication that the different ROS responses to the same hazardous blue light exposure are related to the differences in the device spectra. In Section 3.2, we studied the ROS response to blue and red light exposure, each in isolation, demonstrating that ROS response is triggered by blue but not red light exposure. But the mechanisms of cell response to the complex spectral stimuli received from the devices containing simultaneous stimuli of blue, green, and red light are not understood and were outside the scope of this study.

| ROS-equivalent hazardous blue exposure times
Assuming that the cells respond primarily to hazardous blue exposure, not to the entire spectrum, then their ROS response curves versus hazardous blue exposure should match regardless of device and light mode. Figure 8 examines the example of Device 2 in day (D) and night (N) modes, comparing the ROS responses to the baseline ROS response. It shows that the response curves as measured are shifted horizontally along the exposure axis. A match of each device response to the baseline response can be achieved by multiplying the measured hazardous blue exposures H HB with exposure scale factors, obtained by least-square regression. For Device 2, this results in scale factors of 1.55 in day mode and 3.4 in night mode. As exposure is proportional to irradiance E and exposure time t, these scale factors can be conceptualized as relative exposure times. A higher equivalent exposure means that cells can be exposed longer to a device spectrum before reaching the baseline ROS response to hazardous   The ROS-equivalent exposure times obtained by leastsquare matching allow a comparison of the different device spectra to the baseline response to blue light only. Relative exposure times greater than one mean the cells can be exposed longer to that device spectrum before accumulating an ROS response equal to the baseline response. Figure 9 shows the ROS-equivalent hazardous blue exposure times relative to the baseline response.
The ROS-equivalent hazardous blue exposure times account for the effect of the different device spectra on the ROS response. Differences between devices indicate that the different amounts of red light relative to emitted hazardous blue light may even have some protective and restorative effects, but proving this is beyond the scope of this study. Figure 10 shows that there is some correlation between BLTF and ROS-equivalent exposure times. Of the devices passing the BLTF criterion, Device 1N is an outlier with lower than expected time. A possible explanation is that the luminance of 1N at 285 cd/m 2 is almost three times higher than those of the frontlit EPD, indicating that the cells' ROS response is disproportionally stronger at higher radiance levels, reducing the beneficial effect of the warmer device spectrum. Figure 11 shows changes in mitochondria morphology after 4 h of exposure from Devices 1D, 1N, 2D, and 2N compared to the dark control (no light exposure). Mitochondrial response is immediate upon light exposure and it accumulates over time. Green elongate profiles (marked with white arrowheads) are commonly observed within cells exposed to the light modes of dark (no light control) and night (N) under experimental conditions. Mitochondrial fragmentation (marked with white arrows) and loss of interconnection were observed in both types of study devices in the display mode of day (D). Nucleus is shown in blue. Scale lines are at 4-μm intervals on the reconstructed images.

| ROS response versus cell morphology
The morphological analysis of the mitochondrial response in Figure 12 compares mitochondrial interconnectivity and circularity in cells exposed to Devices 1 and F I G U R E 8 Reactive oxidative species (ROS) response to hazardous blue exposure from Device 2 as measured (top) and after exposure matching to the baseline response. For matching, the exposures from Device 2D were multiplied with 1.55, and those from Device 2N with 3.4.
F I G U R E 9 Relative exposure times for an equivalent reactive oxidative species (ROS) response to hazardous blue light, normalized to the baseline response. 2 in day and night modes to the control dark response of cells shielded from light exposure. These two parameters were chosen for comparison because cells show stress from hazardous blue light exposure by losing mitochondrial connectivity and increasing mitochondrial circularity as the disconnected mitochondria fractionate. As Figure 12 shows, circularity increases relatively from night mode (N) to day mode (D) and from Device 1 to Device 2. Interconnectivity is reduced in day mode relative to night mode, but there is no significant difference between Devices 1 and 2.

| CONCLUSION
In vitro human retinal epithelial (ARPE-19) cells served as a model to assess the potential impact of light exposure from tablet-sized displays. These cells were exposed to light from displays with different technologies (LCD with backlight vs. EPD with frontlight) and varied spectral outputs (day mode with cold-white light vs. night mode with warm-white light). Cellular response to exposure was quantified by measuring ROS and by analyzing mitochondrial morphology.
Control experiments with a spectral illuminator established a baseline ROS response to blue light-only exposure, but none to red light. ROS response to light F I G U R E 1 1 Mitochondrial morphologic response after 4 h of exposure from Devices 1 (top) and 2 (bottom row). White arrowheads mark mitochondria with elongated profiles in the dark control and after illumination from devices in night mode. White arrows mark examples of mitochondrial break and circularity increase after illumination from devices in day mode. The grid lines denote 4-μm intervals.
F I G U R E 1 2 Mitochondrial circularity and interconnectivity relative to dark control. exposure from the study devices generally increased with time, but differences in the device spectra caused different ROS responses to the same hazardous blue exposures. In other words, if two devices expose the cells to the same level of hazardous blue light but their spectra differ otherwise, it may take different times for the cells to accumulate equivalent levels of ROS. Cells accumulated ROS two to three times as fast when being exposed to backlit LCD compared to frontlit EPD, each operated in the same light mode (day or night).
With increased accumulation of ROS, mitochondrial morphology shifted from elongate interconnected features typically observed under normal conditions to rounded disconnected features associated with oxidative stress response. While the amount of mitochondrial morphology change seemed to roughly correlate with ROS response, it is unclear how these changes may impact long-term cell viability and/or inflammatory microenvironments.
Extrapolating from the data, it would seem that devices that cause ROS accumulation at a lower rate (i.e., EPD with frontlight) can be used for longer times before the same levels of ROS are reached. Differences in ROS responses suggest that different relative amounts of red light in the device spectra are mitigating the effects of hazardous blue exposure, but it requires further study to confirm and quantify such protective effect.