Abstract
Mesopic (dim light) conditions are prevalent in everyday environments, yet most human vision research is conducted under idealized, photopic (bright) conditions. Electrophysiological studies suggest that under mesopic conditions, contrast-encoding retinal ganglion cell receptive fields expand their center width while diminishing surround inhibition. These retinal modifications enhance light capture by increasing the summation area but they limit spatial resolution. However, the impact of mesopic conditions on human cortical spatial integration mechanisms remains unclear. To address this, we investigate how mesopic conditions affect early visuocortical processing, specifically spatial summation and surround suppression. Across two experiments, we acquired fMRI BOLD responses from 11 normally sighted participants of both sexes under photopic and mesopic conditions in visual areas V1–V3. The first experiment estimated population receptive field (pRF) properties while the second experiment assessed cortical surround suppression. Photopic and mesopic psychophysical surround suppression, contrast sensitivity function (CSF), and visual acuity were also measured. At the cortical level, mesopic conditions were associated with smaller pRF sizes, while surround suppression remained robust. At the perceptual level, mesopic conditions led to reduced acuity, lower CSF, and weaker suppression, diverging from the observed cortical effects. Importantly, individual differences linked these findings: participants who exhibited greater mesopic reductions in visual acuity also showed larger decreases in early visuocortical surround suppression, underscoring its role in contrast coding and spatial resolution. Altogether, our fMRI findings contrast with retinal electrophysiology and suggest that early visual cortex may employ distinct, perhaps compensatory, mechanisms in response to reduced retinal input under mesopic conditions.
- contrast sensitivity function
- cortical adaptation
- fMRI
- mesopic conditions
- pRFs
- primary visual cortex
- spatial summation
- surround suppression
Significance Statement
Despite the prevalence of mesopic (dim light) environments, their impact on human visuocortical processing remains understudied. Electrophysiological studies suggest that mesopic conditions lead to larger receptive fields and reduced surround inhibition in retinal ganglion cells, enhancing light summation at the cost of spatial resolution. Using fMRI and psychophysical measurements, we investigate how mesopic conditions impact spatial summation and surround suppression across early visual cortex. We find that under mesopic conditions, population receptive fields become smaller, and cortical surround suppression remains robust. However, individual differences revealed a correlation between mesopic visual acuity impairment and changes in V1 surround suppression. These findings contrast with retinal electrophysiological findings, pointing to potential cortical refinement mechanisms that help preserve visual function under degraded viewing conditions.
Introduction
The human visual system functions under a wide range of ambient light levels, spanning photopic (10–108 cd/m2, daylight), mesopic (0.01–3 cd/m2, dim light), and scotopic (10−6–10−3 cd/m2, dark; Roufs, 1978). In photopic conditions, only cone photoreceptors are active, while mesopic conditions represent an intermediate luminance range where both rods and cones drive vision, leading to a dynamic interplay between photoreceptor types. Many artificial lighting environments, such as streetlights, indoor spaces, and overcast settings, fall within the mesopic range (Grimes et al., 2018), making mesopic vision critical for daily activities, including driving, navigation, and reading (Bhorade et al., 2013; Kimlin et al., 2017; Grimes et al., 2018; Penaloza et al., 2025). Nonetheless, most vision research has focused on photopic conditions, leaving mesopic cortical visual processing largely unexplored. This study aims to address this critical knowledge gap by investigating the impact of mesopic viewing conditions on early visual cortical processing.
The integration of visual information is key to constructing a coherent representation of the visual world, enabling pattern recognition and complex scene perception. This process relies on two complementary mechanisms: differentiation and convergence. Differentiation enhances sensory information by refining edges and contours through lateral inhibition and surround suppression, improving feature segmentation. Convergence integrates inputs from multiple neurons, leading to a progressive increase in receptive field size at each synaptic relay. Together, these mechanisms enable neurons to process and respond to increasingly complex visual stimuli, facilitating high-order perception.
Ambient light levels impact these fundamental visual mechanisms as early as the retina, where the center–surround receptive field (RF) organization of retinal ganglion cells (RGCs) plays a key role in contrast coding to detect local edges and contours (Barlow, 1958; Thoreson and Mangel, 2012). As retinal illumination decreases, the RF center summation area enlarges (Barlow, 1958; Andrews and Hammond, 1970; Derrington and Lennie, 1982; Chan et al., 1992; Hunter et al., 2023), while the antagonistic surround weakens or even disappears (Barlow et al., 1957; Chan et al., 1992; Cowan et al., 2017; Fig. 1). These changes are believed to compensate for reduced light levels but come at the expense of spatial resolution (Derrington and Lennie, 1982; Cowan et al., 2017). Human psychophysical studies showed that both visual acuity and contrast sensitivity decline in dim light (Hecht, 1928; Hertenstein et al., 2016), accompanied by changes in the contrast sensitivity function (CSF). Under dim light, the CSF exhibits a reduced bandwidth, lower peak sensitivity, and a shift in peak spatial frequency toward lower spatial frequencies, suggesting increased summation area and decreased inhibitory influences (Patel, 1966; Van Meeteren and Vos, 1972; Smith, 1973).
Hypothetical changes in cortical receptive fields (RF) under mesopic conditions. Schematic diagram of an early visual receptive field model illustrating the predicted effects of reduced luminance from photopic to mesopic conditions. A, RFs are modeled based on a difference of Gaussians framework (Barlow, 1953). B, Top, The orange RF represents photopic conditions, characterized by a well-defined center–surround organization with strong surround inhibition. Middle, The blue RF depicts a hypothetical mesopic RF organization based on animal retinal ganglion cell findings (Barlow et al., 1957; Barlow, 1958; Andrews and Hammond, 1970; Derrington and Lennie, 1982; Chan et al., 1992), showing an expanded RF center, diminished surround, and reduced center response amplitude. Bottom, The black RF represents an alternative mesopic RF organization in which cortical adaptation mechanisms compensate for weakened retinal input, leading to a narrower RF center and stronger surround. These schematics contrasts retinal driven versus cortical compensatory mechanisms in mesopic vision, providing a framework for testing whether cortical RFs passively reflect changes or actively adjust to maintain functional vision in dim light.
Then, the question arises as to how cortical spatial integration mechanisms adapt to mesopic conditions. If feedforward signals primarily dictate the behavior of early visuocortical RFs, cortical responses should mirror those observed in the retina: larger RF sizes and weakened surround suppression (Fig. 1). However, it is also plausible that the cortex actively refines feedforward input to compensate for reduced retinal signals, potentially through low-level visual adaptation (Kwon et al., 2009; Zhang et al., 2009). Alternatively, higher-order feedback or attentional mechanisms could enhance perceived stimulus intensity to compensate for reduced visibility (Lu and Dosher, 1998; Carrasco and Barbot, 2019). In such a scenario, cortical compensatory mechanism counteracting the retinal effects of dim light could result in decreased spatial summation, indicated by reduced RF sizes, and stronger surround suppression (Fig. 1).
To test these hypotheses, we investigate how mesopic conditions affect cortical spatial summation and surround suppression across early visual areas V1–V3. Cortical surround suppression is assessed with BOLD responses to a central grating surrounded by a high-contrast annulus, a well-established paradigm (Williams et al., 2003; Zenger-Landolt and Heeger, 2003). Voxel-wise spatial summation is estimated using population receptive field (pRF) mapping techniques (Dumoulin and Wandell, 2008; Kay et al., 2013). To establish a link between cortical and perceptual signatures of mesopic visual processing, we also psychophysically measured visual acuity, CSFs, and surround suppression.
Materials and Methods
Participants
Thirteen participants (9 female, 4 male, mean age 25 ± 8 years) were enrolled in the MRI study. A subset of three participants (2 female, 1 male) also completed the psychophysical CSF assessment. In addition, six participants, including three who did not take part in the MRI sessions, completed the psychophysical surround suppression experiment. Participants were recruited through advertisements within the Northeastern University community and the Psychology undergraduate student pool. Compensation was provided in the form of course credits or monetary payment. All participants had normal or corrected-to-normal visual acuity and wore their best correction for both the psychophysical and MRI sessions (MR-safe glasses were provided during scanning). The study was approved by the Northeastern University Institutional Review Board. All participants provided written informed consent and underwent screening for MRI-related contraindications.
Three participants’ data were excluded from each MRI experiment to ensure rigorous data quality control: Two participants were removed due to drowsiness during scan sessions, leading to consistently low central fixation task accuracy (below 75% accuracy; see Results). One exhibited poor performance across both sessions, while the other was excluded only from the surround suppression session but retained in the pRF dataset, as their performance in that session was sufficient. Another participant was removed from the pRF analysis due to outlier pRF size values (>2.5 SD) but remained in the surround suppression dataset. An additional participant was excluded due to poor quality pRF maps, likely caused by eye movement during scanning. As a result, the final sample consists of 11 participants (10 in each experiment), with nine participating in both the surround suppression and pRF experiments.
Luminance adjustment
Mesopic conditions were achieved by dimming (for visual acuity assessments) or eliminating (for MRI and psychophysics) all ambient lighting and placing a neutral density (ND) filter (Kodak) in front of the stimulus presentation apparatus. Luminance attenuation was verified using photometric measurements from a Minolta LS-110 Luminance Meter (Konica Minolta) to ensure precise control of luminance levels.
MRI measurements
Stimuli
Stimuli were generated in MATLAB (2017a) running Psychophysics Toolbox (Brainard, 1997) on a MacBook Pro and displayed on a linearized MRI-compatible monitor (BOLDscreen 32 LCD, Cambridge Research Systems) at the rear of the scanner bore. Participants viewed the BOLDscreen through a mirror mounted on the head coil, with a total viewing distance of 157 cm, subtending a visual angle of 14.2°. The maximum luminance under photopic conditions was 138 cd/m2. For mesopic conditions, a custom-made cover screen with a 1.8 optical density ND filter was applied, reducing the maximum luminance to 2.15 cd/m2.
Population receptive field mapping
We employed stimuli and code from the analyzePRF toolbox for MATLAB (Kay et al., 2013) to map pRFs. The stimuli consisted of color images of objects and faces at various spatial scales over a pink noise background, presented in apertures that systematically swept the visual field against a mean luminance background (Fig. 2C). The stimulus refresh rate was 15 Hz. Each scan included two types of pRF runs: bar sweeps and a combination of rotating wedge alternating with expanding and contracting rings. The stimulus covered a radius from 0.2 to 7.1° from center fixation. During scanning, participants performed a central fixation task, reporting color changes (black to white) that occurred in a 0.4° central fixation dot approximately every 2.5 s. Accuracy feedback was provided at the end of each run. Participants maintained high fixation accuracy, with an average accuracy of 95.7% (±1.1%) under mesopic conditions and 93.5% (±1.9%) under photopic conditions, confirming that participants remained engaged and compliant with central fixation.
pRF surface maps and example time series. A, Occipital surface maps from both hemispheres for a representative participant, under photopic (left) and mesopic (right) conditions. The open side cut represents the calcarine sulcus. Top, Preferred polar angle maps; preferred polar angle is color-coded as shown in inset. Lines drawn on surface indicate the boundaries of areas V1 (closest to calcarine sulcus), V2, and V3. Bottom, Preferred eccentricity in dva. Gray dots indicate the locations of three example voxels whose time series are shown in B. The color bar represents eccentricity in degrees of visual angle. B, Example pRF time-course from the three voxels. In each gray-framed panel, top time series is taken from the bar runs, and bottom from the wedge/ring runs. Each panel represents the time-course of an identical voxel in the mesopic and photopic condition. Time courses were averaged across repetitions of each run in each luminance condition. C, BOLD time courses from an example set of voxels located between 2 and 2.5° eccentricity, in bar (top) and wedge/ring (bottom) runs, averaged across all 10 participants. Only identical voxels across conditions, whose eccentricity estimate fell within this band in both luminance conditions, were included in this comparison. Shaded error bars represent within-subjects SEM (Loftus and Masson, 1994; Kwon et al., 2014).
Surround suppression
The surround suppression stimuli consisted of 2 cycles/degree (cpd) gratings at 98% Michelson contrast, centrally presented on a mean luminance background. Stimuli were presented in an alternating 16 s on–16 s off blocked design, with each run beginning and ending with an off block (Fig. 4A,B). During on blocks, the stimulus was either a central grating alone (inner radius 0.6°, outer radius 2.9°) or a central grating surrounded by an annulus (inner radius 3.1°, outer radius 7°). Both center and surround were iso-oriented and phase-matched to maximize surround suppression effects. Stimuli were counterphase flickering at 8 Hz to continually drive visual cortex activity. Each block contained one of four stimulus orientations: 0, 45, 90, or 135°, with orientation presented in a pseudo-randomized order. Across two experimental runs, each orientation appeared six times (three in the center only and three in the center + surround condition). The initial on block type (center only or center + surround) was counterbalanced across participants. The order of orientations remained identical for each observer across mesopic and photopic conditions.
Prior to the surround suppression scans, independent functional localizer data were collected to identify voxels selectively responsive to the center and surround stimuli. The localizer stimulus consisted of a 100% contrast 2 cpd checkerboard pattern, counter-flickering at 10 Hz, with inner and outer radii matching the center or surround gratings (Fig. 3C,D). The localizer runs used a 16 s blocked design, alternating between the localizer stimulus (center or surround) and a mean luminance blank screen. As in the pRF mapping experiment, participants monitored the central fixation dot for color changes and responded via key press as quickly and accurately as possible. Accuracy feedback was provided. Participants demonstrated high fixation compliance, with an average accuracy of 96% (±1.1%) in the mesopic condition and 95% (±1.4%) in the photopic condition, confirming attentional engagement and fixation reliability.
Surround suppression functional localizer and surround suppression experiment time courses. A, B, Occipital surface map representing GLM statistical significance value for (A) center and (B) surround localizer relative to baseline (mean luminance), under photopic (left) and mesopic (right) luminance; thresholded at p = 0.05. The black outlines represent the extent of the center localizer ROI, based on the eccentricity (voxel center at under 2.9° eccentricity) and localizer significance (p < 0.05 to center localizer and p ≥ 0.95 to surround localizer) criteria. The center and surround localizer stimuli are shown to the right of the cortical maps. The color bar reflects the central localizer p-value. C, D, Average time series across cortical regions responsive to (C) center stimulus and (D) surround stimulus. V1 time-courses are shown; V2–V3 time-courses show the same pattern with lower % signal change amplitudes. Shaded error bars represent within-subjects SEM.
MRI data acquisition
The two experimental sessions were conducted separately, with an interval of 1–10 d between them. Each session lasted between 75 min and 2 h. All MRI data were acquired at the Northeastern University Biomedical Imaging Center using a 3 Tesla Siemens Prisma MR scanner equipped with a 64-channel head coil. Whole-brain anatomical images were acquired using a multiecho T1-weighted sequence with 1 mm isotropic voxel size (FOV, 256 mm; flip angle, 8°; TR, 2,500 ms; TE, 1.62 ms). Functional images were obtained using a T2*-weighted simultaneous multislice echo planar imaging (EPI) sequence (Moeller et al., 2010; Xu et al., 2013) using a multiband acceleration factor of 4 with 2 mm isotropic voxel resolution (FOV, 208 mm; flip angle, 57°; TR, 1,000 ms; TE, 30 ms).
Experimental procedure
In both experiments, the two luminance conditions (mesopic and photopic) were tested in a single session. The session always began with the mesopic condition, allowing participants to adapt to dim light during setup and initial scans. Once the mesopic portion was completed, the neutral density filter was removed, and field map scans were acquired, providing participants with a brief period to adjust to photopic luminance before the second half of the session, which consisted of photopic stimulus runs.
Population receptive field mapping experiment
The first experimental session was dedicated to pRF mapping under both mesopic and photopic conditions. The session began with a whole-brain T1-weighted anatomical scan, lasting ∼8 min, during which participants continued adapting to the mesopic luminance. Following the anatomical scan, each participant completed five functional runs (300 TRs per run) per luminance condition, consisting of three bar sweep runs alternating with two rotating wedge and expanding/contracting ring runs.
Surround suppression experiment
The session began with a 6 min resting-state scan, which was not analyzed for the current study. To ensure equal adaptation times between sessions, we waited 8 min after the participants entered the scanner before starting the task runs. Participants then completed two functional localizer runs (208 TRs each), with a center localizer run followed by a surround localizer run, before proceeding to two runs of the surround suppression task (400 TRs each) per luminance condition.
MRI data processing
Anatomical data
T1-weighted anatomical data were processed with FreeSurfer (Fischl, 2012). Scans were first anonymized using the mri_deface function (Bischoff-Grethe et al., 2007) before undergoing automated cortical reconstruction using FreeSurfer's recon-all pipeline. The resulting cortical surface reconstructions were then used for coregistration of anatomical and functional images. An occipital lobe surface label was also generated based on an intrinsic functional connectivity atlas (Yeo et al., 2011), which guided voxel selection for pRF analysis.
Functional data
Functional data were processed in each session's native space using FreeSurfer and FSL (Smith et al., 2004). Echoplanar imaging (EPI) distortions were first corrected using a reverse phase-encoding method (Andersson et al., 2003). Following distortion correction, data were processed with the FreeSurfer Functional Analysis Stream (FS-FAST), which included standard motion correction, Siemens slice timing correction, and boundary-based registration to anatomical scans (Greve and Fischl, 2009). To maintain voxel-level precision, spatial smoothing was omitted (FWHM = 0). Voxel-to-voxel correspondence within each scan session was ensured using robust rigid registration (Reuter et al., 2010), aligning all runs in a session based on the middle time point of each run, with the first run as the registration target. Mesopic and photopic data were analyzed separately to assess luminance-dependent effects on functional responses.
Population receptive field mapping
Population receptive field data were analyzed using the analyzePRF toolbox (Kay et al., 2013), which employs a compressive spatial summation (CSS) pRF model to estimate receptive field properties. Each voxel's aggregate pRF is modeled as an isotropic 2D Gaussian (Dumoulin and Wandell, 2008), with a compressive power-law nonlinearity to account for subadditive spatial summation observed in the retinotopic visual cortex (Pihlaja et al., 2008; Kay et al., 2013; Winawer et al., 2013). The CSS model has been shown to improve pRF estimation compared with the linear 2D Gaussian model (Kay et al., 2013). pRF size was estimated as follows:
Only voxels within the occipital lobe surface ribbon were included in the analysis. Early visual cortical areas (V1/V2/V3) were manually delineated based on polar angle preference reversals observed in photopic pRF data (Fig. 2A), as pRF mapping in the literature is typically conducted under photopic conditions. As expected from previous work (Carvalho et al., 2022) and as verified by visual inspection, the coarse organization of the mesopic and photopic polar angle maps showed strong correspondence (Fig. 2A, representative observer). Additionally, previous work demonstrates that even severe visual deficits (e.g., visual field defects or scotoma) do not generally lead to large-scale changes in visual field maps, with visual area borders remaining stable (Papanikolaou et al., 2014). This provided confidence in using the same ROI labels for voxel selection across both photopic and mesopic conditions.
Surround suppression
Functional localizer scans were analyzed with a standard GLM analysis in FreeSurfer to evaluate each voxel's response to center and surround stimulation (Fig. 3A,B). Surround suppression data were processed with custom MATLAB scripts, including detrending, high-pass filtering (0.01 Hz cutoff), and normalization to % signal change by referencing each time point to the time series average. An event-triggered average was computed for each condition (center only/no surround vs center + surround, across grating orientations), aligned to stimulus onset. To compare % signal change across conditions, data were averaged over a 6–16 s poststimulus window (Fig. 4C, gray shaded areas), accounting for hemodynamic delay and capturing the peak BOLD response.
Surround suppression design, stimuli, and results. A, Depiction of the 16 s on–16 s off block design alternating center only (left stimulus in B) and center + surround (right stimulus in B). The starting condition was counterbalanced between participants and identical for each participant between the photopic and mesopic conditions. C, Mesopic (left) and photopic (right) event-triggered average % BOLD signal change in the center only and center + surround conditions in area V1 (top), V2 (middle), and V3 (bottom). The shaded gray area highlights the temporal averaging window. Shaded error bars represent within-subjects SEM. D, Left, Averaged center BOLD activation to no surround versus surround stimuli in the three ROIs, compared between mesopic and photopic conditions. Error bars represent within-subjects SEM. Right, Data are further summarized by plotting the % signal change difference between center only and center + surround conditions across mesopic and photopic luminance. Surround suppression effects were statistically significant in V1 and V2 and marginally significant in V3 (*** indicates p < 0.001, ** indicates p < 0.01, + indicates marginal significance, i.e., p = 0.05, n.s., indicates a nonsignificant result. PSC = % signal change). The absence of an interaction effect indicates no differences in ROI-wide surround suppression magnitude between photopic and mesopic conditions.
Voxel selection
Surround suppression experiment
Voxel selection was conducted within the manually defined regions of interest (ROIs) for V1, V2, and V3. Voxels located at the boundaries, and thus assigned to multiple areas, were excluded to prevent duplicate data points. Voxels with excessive noise, defined as % signal change fluctuations exceeding 50%, were discarded. For surround suppression analysis, we specifically focused on voxels with pRFs that fell within the center stimulus (0.6–2.9° from fixation). Voxel selection was based on independent localizer GLM results, with inclusion criteria requiring a central localizer significance value of p < 0.05 and a surround localizer significance value of p ≥ 0.95. This was done to ensure that selected voxels responded predominantly to the central stimulus and not to the surrounding grating.
Voxel selections were performed separately for the mesopic and photopic conditions in each ROI, resulting in two distinct groups per condition. The average numbers of voxels selected per visual area were as follows: photopic condition—90 ± 38 voxels in V1, 44 ± 19 voxels in V2, and 30 ± 13 voxels in V3; mesopic condition—96 ± 53 voxels in V1, 52 ± 22 voxels in V2, and 37 ± 19 voxels in V3.
pRF experiment
As with the surround suppression dataset, independent voxel groups were used for mesopic and photopic conditions. Voxel selection criteria included discarding voxels assigned to multiple labels or with nonconverging pRF estimates. Additional exclusion criteria included pRF sizes near the pRF model's lower bound (<0.01°), estimated eccentricities exceeding the maximal pRF stimulus eccentricity (>7.1° from fixation), and voxels within the foveal confluence (<0.75° from fixation). A bootstrapping thresholding procedure was applied to exclude voxels with poor pRF model goodness-of-fit (R2) and to increase the contribution of high R2 voxels. This approach was taken to enhance model reliability by mitigating potential biases associated with using a fixed, yet arbitrary, single-threshold criterion. For each condition, observer, and visual area, R2 values were systematically varied from 10 to 80% in 5% increments, resulting in 15 iterations of voxel selection. A 10% pRF R2 threshold is commonly used in the literature (Harvey and Dumoulin, 2011; Benson et al., 2018; Welbourne et al., 2018), while an 80% threshold is considered excellent (Kay et al., 2013). In each iteration, voxels with R2 values below the threshold were discarded. As a result, the pooled results from all 15 iterations emphasized voxels with the highest fitting models (>80% R2). Subsequent analyses were performed on this resampled pool for each condition, observer, and ROI. pRF model goodness-of-fit metrics were comparable between the two luminance conditions. The observer-averaged R2 values were as follows: photopic condition—51% (±4.5%) in V1, 50.6% (±5.9%) in V2, and 47.8% (±4.8%) in V3; mesopic condition—49.5% (±3%) in V1, 48.9% (±4.6%) in V2, and 45.8% (±4.1%) in V3.
Psychophysical measurements
Visual acuity
Binocular near visual acuity (VA) was assessed at a 40 cm viewing distance using an Early Treatment Diabetic Retinopathy Study (ETDRS) chart under both mesopic (2.49 cd/cm2) and photopic (151.7 cd/cm2) conditions. We did not implement a formal adaptation period to mesopic conditions prior to acuity testing. Instead, participants naturally adapted to the dim lighting while the experimenter was setting up the acuity test. Following the consenting process, the testing room lights were dimmed and participant completed the acuity task under mesopic conditions. After the mesopic acuity test, lights were turned back on and participants performed the photopic acuity test.
Contrast sensitivity experiment
Stimuli
A vertical sine wave grating was centrally presented at spatial frequencies of 0.5, 1, 2, 4, 6, and 12 cpd at a 62 cm viewing distance against a uniform gray background. The grating radius was 2.9° to match the extent of the MRI stimulus. Maximum luminance was 100.72 cd/m2 in the photopic condition and in 2.16 cd/m2 in the mesopic condition. Stimuli were generated and controlled using MATLAB (version 8.4; MathWorks) and Psychophysics Toolbox extensions (Brainard, 1997) for Windows 10, running on a PC desktop computer (Dell Precision Tower 5810; Dell). The stimuli were rendered on a 32″ Display++ LCD monitor (Cambridge Research Systems) with a 120 Hz refresh rate and a 1,920 × 1,080 pixels resolution. The monitor was set to Mono++ mode to achieve up to 16 bit grayscale precision.
Procedure
Detection contrast thresholds for each spatial frequency were measured binocularly under mesopic and photopic conditions, with the two luminance conditions completed on two separate days. A two-interval forced-choice (2IFC) task was used to determine the threshold for contrast detection. Each experimental session was divided into several blocks, with each block containing 30 trials of a single spatial frequency. Each luminance condition was repeated twice, once with ascending spatial frequency order and once with descending spatial frequency order. To counterbalance potential order effects, we also alternated the starting luminance condition and spatial frequency order across participants. Each trial sequence consisted of the following: a 500 ms central fixation interval, a first 250 ms stimulus interval (auditory-cued), 600 ms blank interval, a second 250 ms stimulus interval (auditory-cued), a 600 ms blank interval, and a response interval. The grating was presented in one of the two stimulus intervals. The participants were instructed to maintain central fixation throughout the experiment and report which stimulus interval contained the stimulus grating by pressing a key. Auditory feedback was provided for correct responses. Contrast thresholds were measured using a 3-down-1-up staircase procedure, with step size of 1 dB, yielding a detection accuracy of 79.4% (Wetherill and Levitt, 1965). The detection contrast threshold of each spatial frequency condition was determined by taking the geometric mean of the last 8 staircase reversals (out of 10 reversals). Each testing session lasted ∼25 min. As with the fMRI sessions, participants underwent at least 8 min of mesopic adaptation before beginning the experiment. Chin and forehead rests were used to minimize head movements and maintain a fixed viewing distance throughout the experiment.
Data analysis
The final threshold contrast for each spatial frequency was averaged across the ascending and descending runs for each observer and luminance condition and plotted against spatial frequency to produce a CSF. The resulting CSF was individually fit with an asymmetric parabolic function (Chung and Legge, 2016) separately per luminance condition, modeled as follows:
Perceptual surround suppression
Stimuli
This experiment was designed to measure the psychophysical counterpart of the cortical surround suppression metric assessed in the fMRI experiment. In the center-only condition, a vertical 2 cpd sine wave grating (radius, 2.9°) was centrally presented at a 62 cm viewing distance. In the surround suppression condition, the central grating was surrounded by a high-contrast (98%) annulus (inner radius, 3.1°; outer radius, 10°) with the same orientation and spatial frequency as the center grating.
Procedure
Detection contrast thresholds were measured using a 2IFC paradigm for the center-only and center + surround configurations under both mesopic and photopic conditions. Each condition was tested in a separate block, with the order counterbalanced across participants. As with the fMRI sessions, participants adapted to the mesopic conditions for at least 8 min before beginning the mesopic sessions. The entire experimental procedure lasted ∼50 min. Luminance attenuation levels and the experimental procedure were otherwise identical to the CSF experiment.
Data analysis
Suppression strength was quantified by subtracting each observer's contrast threshold in the center-only condition from their threshold in the center + surround condition for both luminance conditions. Thus, positive values indicate an increase in contrast threshold (i.e., reduced sensitivity) in the presence of a surround, with larger positive values reflecting stronger surround suppression.
Experimental design and statistical analyses
fMRI
To evaluate the effects of luminance in each visual area in this within-subjects design, we employed linear mixed-effects models using lme4 and lmerTest packages in R and RStudio (R Core Team, http://www.r-project.org/). Individual participants were included as random effects. Each statistical test included 10 observations. F tests were used to determine the statistical significance of the main and interaction effects. In the pRF experiment, separate models were used for each outcome variable and visual area. The independent variable was luminance (mesopic vs photopic), and the dependent variables included pRF size, pRF R2, and BOLD signal gain. In the surround suppression experiment, the model included two predictors: luminance (mesopic vs photopic) and presence of surround (surround vs no surround). The dependent variable was the % signal change of the BOLD signal. Additionally, to further confirm evidence for or against the effect of mesopic conditions, we conducted a Bayesian repeated-measures ANOVA in each visual area using the BayesFactor package for R, with individual participants again included as random effects.
Psychophysics
For psychophysical measurements of surround suppression (N = 6), we used Wilcoxon signed-rank tests to compare mesopic versus photopic conditions, treating luminance as the independent variable. To compare visual acuity between mesopic and photopic conditions for all participants (N = 11), we used paired t tests. Finally, to examine relationships between perceptual and cortical measures, we conducted Pearson’s correlation analyses between visual acuity, pRF size, surround suppression, and other fMRI parameters of interest, separately for each ROI.
Data and code accessibility
Data and analysis code can be found on Open Science Framework at the following link: https://osf.io/fzemj/.
Results
Cortical signatures of mesopic visual processing
Surround suppression remained robust under mesopic conditions: divergence from retinal neurophysiological findings
For surround suppression analysis, we averaged the time-window mean BOLD response across voxels in each ROI and experimental condition. The % signal change in voxels corresponding to the center stimulus was modeled as a function of surround presence (surround vs no surround) and luminance condition (mesopic vs photopic). To test whether mesopic conditions affect surround suppression, we included an interaction term between surround presence and luminance and in our model. Figure 4C summarizes the results. In V1, the presence of the surround significantly reduced the BOLD signal (F(27) = 12.3, p = 0.002), confirming the effect of surround suppression on the central grating. There was also a significant main effect of luminance, with lower BOLD responses under mesopic compared with photopic conditions (F(27) = 13.7, p < 0.001). However, the interaction between surround and luminance was not significant, indicating that surround suppression remains stable across luminance conditions. A similar pattern was observed in V2, with significant main effects of surround (F(27) = 20.3, p < 0.001) and luminance (F(27) = 15.5, p < 0.001), but no interaction. In V3, the surround effect was marginally significant (F(27) = 4.2, p = 0.05), with no significant effects of luminance or interaction. Although mesopic conditions reduced the overall BOLD signal in early visual areas (V1 and V2), no evidence suggested a weakening of surround suppression across the center region. Figure 4D further summarizes this by plotting a direct comparison of suppression strength (computed as the BOLD response in the no surround conditions minus surround condition) in each luminance condition and visual area.
Given that perceptual suppression is typically strongest near the boundary between target and suppressive stimuli (Snowden and Hammett, 1998), we conducted an additional analysis focusing on voxels within ±1° of the outer boundary of the center stimulus (2.9°). Statistical tests on this limited subset similarly revealed no interaction between surround and luminance.
To further confirm our findings, we computed Jeffreys–Zellner–Siow Bayes factors (JZS BF) using the BayesFactor package for R (Morey and Rouder, 2011). Bayesian repeated-measures ANOVA (participants as random effects) showed the strongest evidence for a model with main effects of luminance and surround but no interaction in both V1 and V2 (estimated JZS BF10 in V1, 211 ± 1.5%, and in V2, 11,612 ± 1.5). In V3, the strongest evidence supported a model with a main effect of surround only (JZS BF10 = 1.54 ± 0.8%). Similarly, Bayesian analyses found no support for an interaction of luminance and surround in the boundary voxel subset.
Taken together, mesopic luminance did not appear to alter surround suppression strength, either across the entire center stimulus or at the center–surround boundary, where surround suppression is typically strongest. These results suggest that cortical surround suppression is preserved despite significant reductions in luminance and increased rod involvement. This contrasts with retinal neurophysiological studies, which reported weakened suppression at low luminance levels.
Finally, we aimed to rule out the possibility that our suppression findings were artifacts introduced by the alternating block design (i.e., the 16 s ON–16 s OFF block design alternating between center-only and center + surround; Fig. 5A). To test this, we conducted an additional fMRI surround suppression experiment using a control blocked paradigm, in which the center-only and center + surround conditions were tested in separate runs (Fig. 5B). The session structure, including the independent localizer scans, remained identical to the main surround suppression experiment, as did the data analysis procedure. As shown in Figure 5A,B, the results were consistent across paradigms, further confirming the robustness of surround suppression under mesopic conditions, regardless of the experimental design. A Bland–Altman plot (Bland and Altman, 1986; Fig. 5D) directly compares suppression effects across luminance conditions by plotting the % signal change difference between the two designs against the mean % signal change across designs for each observer and ROI. The analysis showed agreement between paradigms, demonstrating consistent suppression of BOLD signals in the center + surround condition across both photopic and mesopic conditions.
Comparison of two fMRI blocked design paradigms used for surround suppression. Panels A and B show the averaged BOLD activation for the same three participants under two different blocked design paradigms, the main experimental paradigm, and a control experimental paradigm. A, Top, Alternating blocked design paradigm used for our main experiment in which no surround and surround conditions were presented within the same scan run. Error bars represent within-subjects SEM. B, Top, Control blocked design paradigm, in which no surround and surround conditions were presented in separate runs. Underneath both panels is the averaged BOLD signal between no surround and surround conditions and both luminance conditions, in each visual area, for the same three participants who completed both the original and the control scans. Error bars represent within-subjects SEM. C, Stimuli used in both surround suppression experiments. D, Bland–Altman plot comparing the % signal change difference between the two paradigms against the mean % signal change across the two paradigms, for each observer and ROI. The analysis was conducted separately for the photopic and mesopic conditions. Each marker shape represents an individual participant. Results demonstrate agreement between paradigms, confirming the robustness of surround suppression for both mesopic and photopic conditions regardless of experimental paradigm.
Cortical pRF size under mesopic conditions is reduced compared with photopic conditions: divergence from retinal neurophysiological findings
To investigate how spatial summation mechanisms are affected by mesopic conditions, we compared pRF sizes in each visual area between mesopic and photopic conditions (Fig. 6A). Average pRF size estimates were computed per observer for each ROI and luminance condition. We also analyzed pRF sizes as a function of eccentricity by fitting a linear model to the combined voxel dataset across all 10 participants (voxels with pRF R2 > 10%, whose eccentricity estimate fell between 0.75 and 7.1° from fixation), in each luminance condition and visual area (Fig. 6B).
pRF size comparison between mesopic and photopic conditions. A, Average pRF size in each luminance condition and ROI. In all three ROIs, pRFs were significantly larger under photopic compared with mesopic conditions (*** indicates p < 0.001, ** indicates p < 0.01). Error bars represent within-subjects SEM. B, Top, scatterplots showing individual voxel pRF size as a function of eccentricity in each visual area. Bottom, Best-fit linear regression lines describing the relationship between pRF size and eccentricity for mesopic (blue color) and photopic (orange color) conditions.
As expected, pRF size increased with eccentricity and visual hierarchy, consistent with previous findings (Clavagnier et al., 2015; Dumoulin and Wandell, 2008; Fig. 6). However, pRF sizes were systematically smaller under mesopic conditions. Linear mixed-effects models revealed a significant main effect of luminance across all three ROIs (Fig. 6A; V1: F(9) = 31.1, p < 0.001; V2: F(9) = 34.1, p < 0.001; V3: F(9) = 22.4, p = 0.001). Despite this overall reduction, the slope of the best-fit lines relating pRF size to eccentricity did not differ significantly between mesopic and photopic condition in any visual area. This indicates that while mesopic conditions reduce pRF sizes, the scaling relationship between pRF size and eccentricity remains preserved across luminance levels.
Given the unexpected reduction in pRF size under mesopic conditions, we conducted control analyses to ensure that this effect was not an artifact of lower mesopic BOLD signal strength or differences in voxel selection criteria. If these factors were responsible, we would expect to see significant and consistent differences in pRF model fit quality (R2) and/or BOLD signal amplitude between photopic and mesopic conditions. First, we compared pRF R2 values across luminance conditions in each visual area. While average R2 values did not differ significantly between luminance conditions in V1 (F(9) = 1.6, p = 0.2), significant differences were found in V2 (F(9) = 11.2, p = 0.009) and V3 (F(9) = 5.6, p = 0.04). To further control for differences in R2, we repeated the pRF size analysis using only voxels with a high R2 threshold (≥70%) in either condition, thus placing stringent criteria on pRF data quality. Even when R2 values were matched across conditions (i.e., R2 differences were n.s. in all ROIs), pRF size remains larger in the photopic condition across ROIs (Fig. 7A). These differences are significant in V1 (F(9) = 5.24, p = 0.048) and V2 (F(9) = 9.6, p = 0.01), but not in V3 (F(9) = 1.5, p = 0.25).
Relationship between pRF size, pRF model goodness-of-fit, and pRF BOLD signal amplitude. A, Left, pRF size averaged across participants in each visual area and luminance condition, from only voxels with an R2 of 70% or higher. Right, Histogram of all voxels meeting our original pRF voxel selection criteria (see Materials and Methods) in each visual area. The arrows at 70% represent the threshold R2 used for our model goodness-of-fit control analysis. Error bars represent within-subjects SEM. B, Left, pRF size averaged across participants in each visual area and luminance condition, from only voxels with a BOLD amplitude of 0.4% signal change or higher. Right, Histogram of all voxels meeting our original pRF voxel selection criteria (see Materials and Methods) in each visual area. The red arrow at 0.4% marks the threshold BOLD amplitude used for voxel selection in our BOLD amplitude control analysis. ** indicates p < 0.01, * indicates p < 0.05, n.s., indicates a nonsignificant result.
We also examined whether differences in BOLD signal amplitude could account for the observed effects; specifically, we aimed to test the possibility that higher BOLD signal gain could lead to larger pRF sizes. The average BOLD response to the pRF stimuli was significantly higher in the photopic condition only in V1 (F(9) = 10.3, p = 0.01), but not in V2 (F(9) = 0.3, p = 0.6) or V3 (F(9) = 1.6, p = 0.2). To further control for this factor, we isolated voxels with the highest pRF gain (≥0.4%) in either condition and compared pRF sizes while ensuring that BOLD % signal change was matched across conditions. Even with this subset of high-BOLD amplitude voxels, average pRF size remains larger in the photopic compared with mesopic conditions (Fig. 7B). These differences remained significant in V1 (F(9) = 11, p = 0.009) and V3 (F(9) = 11.6, p = 0.008), though they were no longer significant in V2 (F(9) = 3.6, p = 0.09).
Both analyses provide strong evidence that differences in % signal change and model fit quality do not account for the observed mesopic reductions in pRF size. Furthermore, when selecting voxels with higher BOLD signal strength, average pRF size actually decreased (Fig. 7B), suggesting that larger pRF size estimates are not merely a byproduct of stronger BOLD responses. Instead, these results support our conclusion that mesopic conditions systematically reduce cortical pRF size independent of signal quality differences.
To further ensure that our findings were not driven by voxel selection criteria, we conducted two additional analyses on the pRF data using alternative voxel selection approaches. First, we applied a conventional 10% pRF R2 threshold, a commonly used criterion in previous studies (Harvey and Dumoulin, 2011; Benson et al., 2018; Welbourne et al., 2018). We found that our results remained unchanged, confirming that our primary voxel selection method did not bias the observed effects. In a second analysis, we selected identical voxels across mesopic and photopic conditions, rather than conducting voxel selection separately for each condition. This ensured that the same voxel population were compared between luminance conditions. Even with this stricter voxel matching, the statistical comparisons of pRF size remained consistent with our original results, reinforcing the robustness of our findings.
Perceptual effects of mesopic viewing conditions
Mesopic conditions reduce spatial resolution, as evidenced by declines in both visual acuity and a shift in peak spatial frequency of the contrast sensitivity function
We next examined the impact of mesopic conditions on three key perceptual measures: visual acuity, CSF and surround suppression. Consistent with previous psychophysical findings (Hecht, 1928; Hertenstein et al., 2016), spatial resolution decreased under mesopic conditions, as indicated by both visual acuity and CSF. First, as shown in Figure 8A, average acuity declines by a factor of 1.8, from −0.07 ± 0.09 logMAR (20/17 Snellen) under photopic conditions to 0.18 ± 0.14 logMAR (20/30 Snellen) under mesopic conditions (t(10) = 5.74, p < 0.001). This decline corresponds to a loss of more than two lines on a standard acuity chart. Second, a decrease in spatial resolution was also evident in the CSF curve (Fig. 8B), measured using detection contrast thresholds required for a vertical sine wave grating across six spatial frequencies. To quantify changes from photopic to mesopic conditions, CSF data were fitted with the asymmetric parabolic function (see Materials and Methods), and the resulting parameters were compared between luminance conditions. As shown in Figure 8B, peak contrast sensitivity reduced by a factor of 1.68, and peak spatial frequency shifted from 3.0 cpd (photopic) to 1.9 cpd (mesopic), further confirming a decline in contrast sensitivity and spatial resolution.
Psychophysical results and perceptual—fMRI correlation. A, Visual acuity (N = 11). Average visual acuity in logMAR units is compared for mesopic (blue) and photopic (red) conditions. B, Contrast sensitivity function (N = 3, a subset of fMRI participants). Contrast sensitivity (inverse of detection threshold contrast) measured with a contrast detection task, fit with an asymmetric parabola function (see Materials and Methods). Data points represent sensitivity averaged across participants, and the line represents the best-fit model. Right four panels, CSF model parameters. Individually fit parameters of the CSF model fit, with markers representing individual participants. C, Surround suppression (N = 6; only 3 are a subset of fMRI participants). Surround suppression was computed as the difference in contrast thresholds between center-only and center + surround conditions. Positive values thus indicate stronger surround suppression. Individual data points represent participants. D, Relationship between the change in visual acuity (mesopic minus photopic) and the change in surround suppression (photopic minus mesopic) for each participant. The data points represent individual participants. For all panels (A–D), black error bars denote ±1 SEM. *** denotes p < 0.001, * denotes p < 0.05.
We also measured surround suppression effects psychophysically using similar stimuli—sine wave grating with a 2 cpd spatial frequency—as in the fMRI experiment to link cortical and perceptual suppression effects. However, a key difference in stimulus and experimental design was that, for fMRI, both the center and surround gratings were presented at high contrast (98%) and counterphase flickered at 8 Hz, whereas in the psychophysical experiment, only the surround grating was high contrast and both gratings were stationary. This allowed us to measure participants’ contrast thresholds for detecting the center stimulus in the presence and absence of the surround, providing a behavioral measure of surround suppression. Surround suppression strength was quantified as the difference in detection thresholds between the center-only and center + surround conditions, with positive values indicating stronger suppression. As shown in Figure 8C, surround suppression was significantly weaker under mesopic conditions, as determined by a Wilcoxon signed-rank test (z = 2.2, p = 0.03). This finding aligns with our initial predictions and retinal electrophysiological findings, which suggest a weakening of suppression under low luminance levels. However, it contrasts with the fMRI results, which showed no significant differences in surround suppression between luminance conditions.
Behavioral visual acuity correlates with V1 surround suppression, suggesting potential cortical compensatory mechanism under mesopic conditions
To further explore the relationship between cortical processing and perceptual performance, we examined whether changes in visual acuity under mesopic conditions were associated with changes in pRF size or cortical surround suppression. For each observer, we computed the following:
Visual acuity decrement, defined as the difference in VA between mesopic and photopic conditions (mesopic–photopic). Larger positive values indicate greater acuity impairment under mesopic conditions.
Cortical surround suppression, calculated as the difference in % BOLD signal change between center-only and center + surround conditions. The luminance-related difference in V1 surround suppression was then computed by subtracting mesopic surround suppression from photopic surround suppression. Positive values indicate weaker surround suppression under mesopic conditions.
pRF size differences, determined by subtracting mesopic pRF size from photopic pRF size for each observer and ROI.
As shown in Figure 8D, we found a significant positive correlation between visual acuity decrement and the luminance-related difference in V1 surround suppression (r(8) = 0.68, p = 0.03). This indicates that a greater reduction in surround suppression under mesopic conditions was associated with a larger decline in visual acuity. This correlation was not significant in V2 (r(8) = 0.2, p = 0.57) or V3 (r(8) = 0.62, p = 0.06). Moreover, pRF size differences did not correlate with changes in visual acuity in any ROI (V1: r(8) = 0.32, p = 0.36; V2: r(8) = −0.04, p = 0.92; V3: r(8) = −0.02, p = 0.96).
These results suggest that, while cortical surround suppression did not differ significantly between luminance conditions at the group level, individual differences in V1 surround suppression were linked to visual acuity. Specifically, stronger surround suppression in V1 under mesopic conditions was associated with better mesopic visual acuity. This finding suggests the possibility that preserving cortical surround suppression under mesopic conditions may help maintain visual sensitivity despite diminished feedforward signals in mesopic conditions.
Discussion
Understanding how the human visual system adapts to rod-dominant mesopic conditions is important, as many daily activities are carried out in dim-light environments, from reading under indoor lighting to nighttime driving (Bhorade et al., 2013; Kimlin et al., 2017; Grimes et al., 2018; Penaloza et al., 2025). Yet, most vision research, particularly human neuroimaging studies, has focused on cone-dominant photopic conditions, leaving cortical processing under dim light largely unexplored. Retinal physiology suggests that under mesopic conditions, RGCs expand their RF centers and reduce surround inhibition, increasing light capture at the cost of spatial resolution (Barlow et al., 1957; Barlow, 1958; Derrington and Lennie, 1982; Chan et al., 1992). However, the extent to which mesopic retinal changes shape cortical mechanisms of spatial integration and surround suppression has not been studied in humans. If early visuocortical processing merely reflects these retinal feedforward changes, we would expect larger cortical integration areas and weak surround inhibition in early visual cortex as well. An alternative possibility is that the visual cortex may instead engage compensatory mechanisms to optimize its processing of the inherently weak and noisy rod-driven signals, for instance, by restricting integration zone and maintaining robust surround inhibition.
Using fMRI and psychophysical measures, we examined how mesopic viewing conditions affect spatial summation and surround suppression in human early visuocortical areas (V1–V3).
At the perceptual level, we confirmed that mesopic conditions reduced spatial resolution, as evidenced by both a decline in visual acuity (by a factor of 1.8) and a shift in the peak spatial frequency of the CSF to a lower spatial frequency (from 3.0 to 1.9 cpd). Mesopic conditions also diminished contrast sensitivity, as indicated by reduced peak CSF sensitivity (by a factor of 1.68). These findings align with previous psychophysical studies demonstrating various visual deficits under mesopic conditions in both normal and clinical populations (Hecht, 1928; Hertenstein et al., 2016; Owsley et al., 2020; Goddin et al., 2023). Furthermore, mesopic conditions weakened surround suppression, as indicated by increased contrast thresholds for detecting a central target grating when presented with a surround. To our knowledge, this is the first report of perceptual surround suppression under rod-dominant luminance conditions, providing novel evidence that mesopic luminance alters contextual modulation in human vision.
At the cortical level, we observed an overall reduction in BOLD signal under mesopic conditions in V1, consistent with evidence of lower gain in V1 cell responses from nonhuman primate electrophysiology (Duffy and Hubel, 2007). A limited number of human neuroimaging studies showed that the early visual cortex responds to luminance increments with increased BOLD signal; however, these studies either only included photopic luminance levels or were limited to BOLD responses to uniform fields or temporal flicker, not directly examining spatial processing mechanisms (Goodyear and Menon, 1998; Haynes et al., 2004; Vinke and Ling, 2020). Second, pRF size increased with eccentricity and visual hierarchy under both luminance conditions, with a similar rate of increase across eccentricity (e.g., 0.09° vs 0.10° pRF size increase per 1° eccentricity in V1). Third, consistent with previous work (Williams et al., 2003; Zenger-Landolt and Heeger, 2003; Schallmo et al., 2016), our fMRI surround suppression paradigm indeed induced significant suppression effects in early visual areas under photopic conditions. Importantly, surround suppression was present under mesopic conditions as well. It is noteworthy that we also observed the noticeable dissociation between psychophysical and fMRI suppression strength noted in previous work (Nurminen et al., 2009; Schumacher and Olman, 2010; Schallmo et al., 2016). Fourth, pRF sizes were significantly smaller under mesopic conditions (reduced by 18% in V1, p < 0.001). Interestingly, these cortical results contrast with predictions derived from retinal studies (Barlow et al., 1957; Barlow, 1958; Andrews and Hammond, 1970; Derrington and Lennie, 1982). Instead, our findings suggest that early visual cortex mechanisms, particularly in V1, may actively compensate for reduced retinal input by minimizing the loss of spatial differentiation. Indeed, a closer examination of individual differences further revealed an important link between the two: participants with greater mesopic reductions in visual acuity also exhibited the largest decreases in V1 suppression magnitude, whereas those with minimum acuity impairment showed little or no change in cortical suppression. This suggests that V1 may employ compensatory mechanisms under mesopic conditions, helping to preserve visual resolution despite degraded retinal input. Prior studies further support this notion, showing that V1 is particularly sensitive to luminance reductions compared with extrastriate cortex, responding more strongly to variations in luminance (Haynes et al., 2004; Vinke and Ling, 2020; Cicero et al., 2024).
Several mechanisms may account for the visuocortical changes observed under mesopic conditions. One possibility is low-level visual adaptation, which is known to optimize neural responses to frequently encountered inputs, such as luminance contrast (Gardner et al., 2005; Kwon et al., 2009; Bao and Engel, 2012), spatial resolution (Webster et al., 2002), orientation (Zhang et al., 2009), or color [Webster and Leonard, 2008; also see a review paper on visual adaptation (Webster, 2015)]. For instance, prolonged exposure to a low-contrast environment has been shown to increase V1 and V2 BOLD responses and enhance low-contrast discrimination, suggesting compensatory mechanisms for efficient contrast coding under contrast-deprived conditions (Kwon et al., 2009). Perhaps, similar contrast-preserving mechanisms may be at play under mesopic conditions. Rod-mediated neural signals are known to exhibit a reduced dynamic range and be more susceptible to noise, resulting in a decreased signal-to-noise ratio (SNR; Barlow, 1956; Banks et al., 1987; Donner, 1992; Wilkinson et al., 2020). The visual cortex may thus enhance surround suppression and reduce pRF size to selectively filter out noise and maintain spatial differentiation despite degraded input. Our cortical findings can be interpreted as an adaptative cortical strategy to enhance signal fidelity in a low-SNR environment. Such adaptation would allow the cortex to optimize signal reliability even when retinal inputs are compromised. Another potential mechanism is top-down feedback from higher cortical areas, such as endogenous attention. Attention has been shown to enhance perceived stimulus contrast, effectively acting as a compensatory mechanism to improve stimulus visibility (Lu and Dosher, 1998; Carrasco and Barbot, 2019), thought to achieve this by increasing the SNR of feedforward sensory signals (Cohen and Maunsell, 2009; Briggs et al., 2013). Attention is also known to affect receptive field properties in the early visual system, shifting their spatial preference (Womelsdorf et al., 2006; Klein et al., 2014) and potentially altering their center–surround organization, leading to changes in RF size (Anton-Erxleben et al., 2009). Furthermore, feedback from higher cortical areas can modulate surround suppression (Angelucci et al., 2017; Nurminen et al., 2018). While the aforementioned top-down and bottom-up mechanisms are plausible, we acknowledge that identifying the exact underlying processes of mesopic adaptations would require further investigations.
We acknowledge several limitations of the current study. First, although correlations between surround suppression in V1 and visual acuity suggest individual differences in mesopic adaptation, the relatively small sample size (N = 10) limits the ability to generalize findings. Including a broader range of participants, such as older individuals or those with visual disorders who may be more vulnerable to dim light, could provide deeper insights into individual variability in mesopic cortical processing. Second, the pRF mapping method may influence the estimated pRF size. pRF mapping relies on population-level activity within a voxel, typically using high-contrast stimuli on a mean luminance background to maximize response amplitude and SNR. This results in larger pRF estimates compared with single-cell measures (Dumoulin and Wandell, 2008; Land et al., 2013). A recent fMRI study (Yildirim et al., 2018) showed that pRF estimates vary based on stimulus properties, with second-order orientation contrast yielding smaller pRF estimates compared with luminance contrast. A similar effect could have contributed to our findings, as mesopic conditions may have activated a smaller subset of neurons per voxel, leading to reduced pRF sizes. However, we found no systematic evidence that BOLD signal amplitude or model fit (pRF R2) differences explain the observed pRF size effects, suggesting that this factor alone does not account for the observed pRF size reductions (Fig. 7). Future studies could address this by using feature-selective stimuli, as in Yildirim et al. (2018), to better control neural populations engaged under different luminance conditions. Third, the luminance levels in our experiments were at the higher end of the mesopic range, whereas more pronounced retinal RF reorganization has been observed at lower mesopic levels (Andrews and Hammond, 1970). Nevertheless, nearly all our participants exhibited significant changes at both perceptual and cortical levels, suggesting that the luminance reduction was sufficient to elicit meaningful impact. Moreover, the higher mesopic levels used in the current study more closely resemble real-world nighttime environment with artificial lighting, such as streetlights (Wood, 2020).
In summary, our study provides novel insights into how mesopic conditions affect spatial integration and contextual modulation mechanisms in the visual cortex. The findings suggest that, rather than passively reflecting mesopic retinal changes, cortical mechanisms may actively refine spatial representations to compensate for degraded retinal input. This highlights a potential dissociation between retinal and cortical processing and underscores the role of the early visual cortex in adapting to low-light conditions.
Footnotes
This work was supported by National Institutes of Health/National Eye Institute Grant R01 EY027857 and Research to Prevent Blindness (RPB)/Lions’ Clubs International Foundation (LCIF) Low Vision Research Award to M.Kw. The sponsor or funding organization had no role in the design or conduct of this research. The authors alone are responsible for the content and writing of the article. We thank Traci-Lin Goddin, Da Hae Choi, Hanze Liu, Ashley Kim, and Joonsik Moon for their help and assistance with data collection.
The authors declare no competing financial interests.
- Correspondence should be addressed to Michaela Klimova at m.klimova{at}northeastern.edu or MiYoung Kwon at m.kwon{at}northeastern.edu.
