Dynamics of visual contextual interactions is altered in Parkinson’s disease

Over the last decades, psychophysical and electrophysiological studies in patients and animal models of Parkinson’s disease (PD), have consistently revealed a number of visual abnormalities. In particular, specific alterations of contrast sensitivity curves, electroretinogram (ERG), and visual evoked potentials (VEP), have been attributed to dopaminergic retinal depletion. However, fundamental mechanisms of cortical visual processing, such as normalization or “gain-control” computations, have not yet been examined in PD patients. Here we measured electrophysiological indices of gain control in both space (surround suppression) and time (sensory adaptation) in PD patients based on steady-state VEP (ssVEP). Compared to controls, patients exhibited a significantly higher initial ssVEP amplitude that quickly decayed over time, and greater relative suppression of ssVEP amplitude as a function of surrounding stimulus contrast. Meanwhile, EEG frequency spectra were broadly elevated in patients relative to controls. Thus, contrary to what might be expected given the reduced contrast sensitivity often reported in PD, visual neural responses are not weaker; rather, they are initially larger but undergo an exaggerated degree of spatial and temporal gain control and are embedded within a greater background noise level. We conclude that compensatory cortical mechanisms may play a role in determining dysfunctional center-surround interactions at the retinal level.


Introduction
Parkinson's disease (PD) is a progressive neurological disorder characterized by dopamine deficiency most famously in striatal circuits of the basal ganglia, but also in other dopamine-regulated systems, including the retina. [1][2][3] The involvement of circuits other than the basal ganglia could explain some of the non-motor symptoms including sleep regulation problems, autonomic dysfunction, hyposmia and visual abnormalities. 4 Visual symptoms in PD range from blurry vision, decreased ability to discern color, loss of contrast sensitivity to circadian dysregulation and visual hallucinations. [5][6][7] Systematic abnormalities have also been reported by electrophysiological and psychophysical studies.
Pattern electroretinogram (PERG) responses in patients and animal models are decreased in amplitude and increased in latency compared to controls [8][9][10][11][12] , particularly in an intermediate range of spatial frequencies (2-5 cpd) where contrast sensitivity is highest. 9,11 Pattern Visualevoked potential (PVEP) studies have shown parallel effects [13][14][15] , including delayed latency of the N70 and P100 component in a similar range of spatial frequencies and for temporal frequencies between 4-5 Hz 16 , and phase delay of a low-frequency steady state response. 17 Acute levodopa administration in general reversed both PERG and PVEP abnormalities, suggesting that dopamine has an important role in visual processing. [11][12][13][14][15]17 Psychophysical studies of spatiotemporal tuning curves in PD patients have confirmed a major loss of contrast sensitivity for spatial frequencies in the 2-4 cpd range and temporal frequencies between 4 and 8Hz. [18][19][20] Given the overlap in the affected spatial and temporal frequency ranges, it is plausible that these perceptual abnormalities are linked to the retinal dysfunction reflected in the PERG, and both may arise from the dopaminergic cell degeneration that has been observed in the retina in PD. 2,[21][22][23] Dopaminergic amacrine cells are present in the retinal interplexiform layer and are key players in neural transmission from photoreceptors to ganglion cells and in the generation of center-surround interactions of ganglion cell's receptive fields, thus contributing to spatial frequency tuning. [23][24][25] Further, mouse models of retinal dopamine deficiency have shown loss of contrast sensitivity and visual acuity, resembling the visual abnormalities described in PD. 26 While those findings provide an extensive characterization of the retinally-mediated visual abnormalities present in PD, relatively little is known regarding potential differences in visual processing at the early cortical level, and, to our knowledge, there has yet been no systematic study of fundamental aspects of visual processing such as spatial and temporal gain control. Here we addressed this gap by examining cortically-generated, steady-state visualevoked potentials (ssVEP) in PD and their modulation by surrounding spatial context and temporal adaptation.
We employed a recently developed paradigm designed to probe visual surround suppression whereby the perceived intensity of a stimulus decreases under the presence of a surrounding pattern 27,28 , a phenomenon that relies on lateral, feedback and feedforward connections from the earliest stages of information processing in the retina to visual cortex. 29,30 Since the focus of our investigation was on cortical processing, we used spatial and temporal frequencies lying outside the ranges that typically reveal retinal abnormalities in PD (1cpd and 25Hz). We reasoned that, if the only alterations in cortex are those inherited from the retina, one should expect little or no difference in cortical responses.
We found several differences in patients with PD compared to healthy controls: 1) a greater background noise level in the frequency spectrum; 2) a higher visual response (ssVEP amplitude) that quickly adapts over time, and 3) a stronger relative suppression of the ssVEP amplitude under the presence of surrounding stimuli. These overall signatures of visual contextual interactions indicate that contrast gain control is abnormal in PD and thus may serve as biomarkers to aid diagnosis and to evaluate therapeutic efficacy.

Results
Data were analyzed from a short surround-suppression paradigm recorded as part of a larger study of 28 PD patients and 30 healthy age-matched controls. In this paradigm, subjects passively viewed a series of 2.4-s long trials in which four "foreground" grating stimuli flickered on-and-off at 25Hz, embedded within a static (non-flickering) surround grating pattern, with both the foreground (FG) and surround (SS) contrasts varying randomly from trial to trial.
Background (no-stimulus) Frequency spectrum profiles. Since differences in background spectral EEG amplitude have previously been reported in PD [31][32][33][34] and would impact the estimation of ssVEP amplitudes driven by the external flicker-stimulation, it was important first to characterize such background spectral differences. EEG spectra computed from trials with zero foreground contrast and zero surround contrast (blank screen), averaged across a cluster of six parieto-occipital electrodes where ssVEPs were measured, revealed a broadly elevated spectrum in PD relative to controls ( Figure 1). A bootstrap statistical test at the critical frequency of 25Hz revealed a significant difference in the means (p<0.05). Importantly, the spectral difference extended through the θ , α  and β  bands.

Figure 1. Background frequency spectrum profiles in patients versus controls. Fast
Fourier Transforms were computed on the last 2.24 s of all trials with 0% foreground and 0% surround contrast (blank screen, shown in inset) and then averaged across trials and across subjects within each group. Patients showed significantly higher spectral amplitudes, marked with asterisks for each of the corresponding frequency bins (bootstrapping test, p<0.05).
Shaded error bars indicate mean ± standard error of the mean (s.e.m.).
The broad spectral elevation seen under no stimulation conditions would also have the effect of elevating the amplitude of the 25-Hz ssVEP under the flicker-stimulus conditions when measured in the same way, even if there were no underlying differences in the amplitude of the exogenous response to the flicker. To eliminate this potential confound we took two measures.
First, rather than averaging the single-trial spectra, we computed Fourier spectra on the average across trials per condition, thus averaging-out the background activity levels while retaining the ssvEP signal because it is strictly phase-locked to the stimulus ( Figure 2A). Second, we further used the blank-screen condition to determine a lower number of trials to average within each control subject such that the background noise was elevated to the same average level as the PD patients. Tracing background noise levels in the 25Hz frequency bin during the no-flicker stimulus condition revealed that equal levels of background noise were achieved in both groups when the maximum number of available trials per condition was included for the PD group and about half of the total trials (3 to 4) for the controls ( Figure 2B). This random trial rejection policy was then applied in the same way in the flicker conditions. Surround suppression effects. Applying this noise equalization approach, we examined ssVEP contrast response functions and temporal profiles (Figures 3 and 4). These analyses were carried out on subsets of age-matched patients and controls without cognitive impairment (eleven per group, see methods). Contrast response functions show that ssVEP amplitude increased as a function of foreground contrast in both groups, and decreased as the contrast of the surround stimulus increased ( Figure 3A,B). and 100%. Surround stimulus varied across three levels of contrast: 0% contrast (black trace), 50% contrast (brown trace), and 100% contrast (yellow trace). Insets show stimulus configurations for FG 100% embedded in SS of 0% and 50% contrast. c-Surround suppression ratio. Each point corresponds to a subject. The suppression ratio is computed as the ratio between the ssVEP amplitude for FG=100%, SS=100% and the ssVEP amplitude for FG=100%, SS=0%. Error bars indicate s.e.m. d-Topographical distribution of the 25Hz steady state visual evoked potential amplitude as a grand average between patients and control subjects, for trials with 75 or 100% contrast foreground embedded in a midgray surround (0% contrast). The cluster of electrodes chosen to measure and statistically test ssVEP amplitude are marked with star symbols.
As in our study of younger healthy subjects 27 , we computed a summary metric of the surround suppression effect as the ratio between ssVEP amplitude corresponding to FG=100%, SS=100% and the ssVEP amplitude for FG=100%, SS=0% ( Figure 3C). for each stimulus condition. We found a steep increase of the visual response in the first 500ms after flicker onset, followed by a decay that reflected temporal adaptation. This was more evident for the high foreground contrasts. (Figure 4A,B). The results of an ANOVA showed in Table 1 (see Methods) revealed significant differences across groups that depended on contrast levels and time. To unpack this, we conducted a bootstrapping analysis testing for group differences at each timepoint and for each contrast combination. We found significant differences only for time points between 420-840ms in the 75% FG, 0% SS condition, and for timepoints between 140-1400ms in the 100% FG, 0% SS condition (see asterisks, Figure 4). Since the amplitude increase in PD appeared to be restricted to earlier timepoints for the un-surrounded, high contrast stimuli, we compared with post-hoc test the rates of temporal decline. A Wilcoxon rank-sum test on the slope of a line fit over all timepoints from 500 to 1000 ms after stimulus onset indicated a significantly steeper, faster decay of the visual response in PD subjects (Z=-2.37; p=0.0179).

Discussion
In this study, we Third, the EEG spectrum in PD showed a significantly higher level of background activity, or noise. These three findings challenge the notion that reduced visual contrast sensitivity and visual deficits observed in PD mainly stem from retinal dysfunction.
The higher spectral power in patients over the theta, alpha and beta bands (Figure 1) derived from occipital electrodes are in line with previous reports of increased spectral background over centroparietal regions in alpha and beta bands, mostly during dopaminergic therapy. [31][32][33][34] The visual response in PD reached a higher initial amplitude and more quickly decayed However, we found a significantly greater surround suppression effect in the cortical responses of PD patients as well as steeper temporal adaptation. Thus, we speculate that compensatory mechanisms arising at the cortical level may account for the lack of center surround interactions at the retinal level, and feedback from extrastriate areas to V1 may play a major role in such compensatory mechanisms, 29,30,40 as it is also suggested by recent imaging studies. 41 In our previous work, we found a relationship between the degree of electrophysiological surround suppression and perception. 27

Materials and Methods
Subjects. EEG data were collected on the surround suppression paradigm as part of a larger multi-test study of 28 patients with PD (7 women) and 30 healthy control subjects (20 women), which adopted inclusive recruitment criteria that only required subjects to be right-handed and to show no signs of dementia. As described below, all subjects also received a complete neuropsychological assessment. Subjects were examined by a movement disorder neurologist In the present study our final analysis was conducted on a subset of age-matched, cognitively unimpaired PD patients and control subjects who had sufficient trials for unbiased analysis of visual response magnitudes. Specifically, 10 controls and 10 PD patients were excluded for having MoCA scores of less than 26 47 and/or a diagnosis of depression, and a further 4 controls and 7 PD patients were excluded on the basis that they did not complete a sufficient number of artifact-free trials to measure response amplitudes unconfounded by differences in background spectral EEG levels (see analysis details below). The final 11 control subjects were selected from the remaining 16 so that the two groups of 11 were matched for age and gender, with selections made randomly wherever multiple options for inclusion existed. Surround suppression effects were tested using steady-state visual responses (ssVEP) to flickering foreground stimuli embedded in static surrounds, where stimuli in the upper field were flickered out-of-phase relative to those in the lower field. This method exploits anatomical and signal-summation principles to produce robust ssVEPs. 28 The central "foreground" stimulus (FG) was composed of four vertically-oriented circular gratings, located at an eccentricity of 5 degrees of visual angle, at polar angles of 20° above (two upper disks) and 45° below (two lower disks) the horizontal meridian. Disks flickered on and off at 25 Hz, embedded within a non-flickering full-screen static "surround" (SS) also with vertical orientation, parallel to the FG.
In this configuration, the "foreground" flickering stimulus is embedded into a "surrounding" pattern in order to induce maximum visual surround suppression effects using peripheral flickering foreground stimuli. 27 In all cases, foreground and surround patterns were sinusoidally modulated luminance gratings with a spatial frequency of 1 cycle per degree. The average luminance of all gratings was equal to that of the midgray, 0% contrast surround, which was 65 cd/m 2 . Foreground contrasts could be 0% (midgray), 25%, 50%, 75% or 100% (black and white stripes) and static surround stimulus contrast could be 0%, 50% or 100%, both randomly assigned on each trial (see Figure 3 insets, stimulus configurations for FG 100% embedded in SS of 0% and 50% contrast).
For each of the contrast conditions, surround stimuli were presented spatially in-phase and opposite-phase relative to the foreground, interleaved pseudorandomly, and we collapsed across spatial phase conditions within each configuration to reduce border effects (see Vanegas,et al. 27 ). The full paradigm included a total of 120 trials (15 contrast combinations, 4 repetitions each for surround stimuli in and out of phase with foreground). Each trial started with the presentation of a fixation spot for 500ms, after which the flickering foreground and static surround stimuli were simultaneously presented for 2,400ms. Subjects were instructed to maintain their fixation on the center of the screen throughout each trial.
EEG recordings. During each block, high-density electroencephalography data (HD EEG, Electrical Geodesics Inc., Eugene, OR) were recorded from 256 electrodes at a sample rate of 1,000Hz. We applied an online notch filter at 60Hz. Impedances were stable below 50kΩ. We selected 183 channels on the scalp and downsampled to 500Hz. Eye gaze was monitored continually using an EyeLink1000 (SR-Research) eye tracker to ensure gaze was within 1 degree of visual angle around the fixation spot during stimulus presentation. referenced all channels to average mastoids by simple subtraction. We used Fourier decomposition to examine frequency spectra and compute contrast response functions from ssVEP amplitude corresponding to increasing levels of "foreground" contrast with and without "surrounding" patterns. The paradigm was designed to elicit visual sensory responses at a high frequency (25Hz), clear of the strongest endogenous rhythms such as alpha (10Hz). However, spectral amplitudes at any frequency bin are influenced by differences in levels of background noise. In PD, spectral power is known to be increased compared to controls. 31,33 The power increase is mainly evident over centro-parietal regions in the alpha and beta ranges in levodopa-treated patients. 34 We therefore began our analysis by examining the full EEG spectrum in the entire cohort of subjects by computing the Fast Fourier Transform (FFT) for the blank-screen (baseline) condition of FG=0% and SS=0% contrast (Figure 1 inset). We computed the FFT for a 2240-ms window beginning 160ms after each stimulus onset and then averaged across trials, and across electrodes over the central posterior midline region: Pz, POz and Oz, and neighboring electrodes, just as in Vanegas et al. 27 This analysis using the entire subject cohort confirmed the background spectral difference in PD subjects as has been previously reported in the literature (Figure 1). 31,33 This difference in background EEG noise levels could potentially confound comparisons of ssVEP amplitude between the two groups because it causes all frequencies, including the precise frequency of the ssVEP, to be elevated in PD. We thus carried out a noise-equalization procedure to preclude this confound, whereby we averaged the ssVEP measurements across fewer trials in the control subjects than in the PD patients in such a way that background spectral levels were equalized. Since background brain rhythms are asynchronous to the flickering onset, the background spectrum (noise) is expected to decrease with trial number when the FFT is computed on the average across trials. To determine the trial numbers required for noise-matching, we computed the frequency spectra for the blank-screen (0% foreground and 0% surround) condition as a function of number of trials. For each number of trials, we randomly selected this number 1,000 times from the available trials for this blank-screen condition and then averaged the results. We found that spectral background levels matched  Model fit. As in our original study using this paradigm in young healthy subjects, 27 we fitted the grand-average time-resolved contrast response functions with a model accounting for variations over time as well as foreground and surround contrast. First, the basic amplitude increase as a function of foreground contrast was captured by the standard Naka-Rushton function. 50 Second, the influence of the spatial surround was encapsulated in an additive term in the denominator.
Third, the temporal adaptation of both the foreground and surround drives was described by decaying exponentials with the same time constant across contrasts levels and stimuli (foreground and surround) but asymptotes that were permitted to differ. The complete model is where r is the ssVEP amplitude at time t, Rm is the maximal response, R0 is the spectral baseline noise, σ is the contrast at which half of the maximum response is achieved, β is the coefficient of suppression, which scales the influence of surround contrast in the denominator and n is the exponent that accounts for non-linearity of the function. Since in spectral EEG measurements the baseline level R0 reflects the noise floor (25-Hz amplitude even in the absence of any stimulation), we use the max function to reflect the fact that measurements cannot pass below this noise floor. The time-dependent foreground drive Df(t) and suppressive drive Ds(t) were modeled as decaying exponential functions beginning at the veridical physical contrast of the stimulus and asymptotically tending towards a fraction of that value, captured in the relations: where τ is the mutual time constant for adaptation of foreground and surround drives and Cf and Cs are the foreground and surround contrasts, respectively. K∞,f and K∞,s represent factors by which the foreground and suppressive drive are asymptotically reduced relative to the initial value. The fit was carried out on the data over the interval 420 to 2240ms so that FFT windows stayed within the bounds of the stimulation period and using the method of least-squares. The best fit parameters of the model (see equations 1, 2 and 3) were as follows: PD subjects: condition. To test for spectral differences in the blank-screen condition, we performed a bootstrap analysis at each frequency bin. In this method, statistical samples were obtained by resampling from the shuffled original sample 1,000 times. Surround suppression effects and clinical scores were also tested using independent samples t-tests in cases where measures met the Kolmogorov-Smirnov test for normality. When samples were not normally distributed (i.e., slope of the line, temporal response), we conducted a nonparametric Wilcoxon rank-sum test for independent groups. Correlation analyses (Pearson) were computed to test for relationships between two clinical scores (i.e., duration of the disease and motor score in the UPDRS scale, in patients) and two electrophysiological metrics: 1-ssVEP amplitude in the stimulus condition of FG=100% and SS=0%, and 2-surround suppression effect or suppression ratio, calculated as the ratio between ssVEP amplitude corresponding to FG=100%, SS=100% and the ssVEP amplitude for FG=100%, SS=0%. In order to test for significant differences between the model parameter values fit to the grand average of each group, we used shuffle statistics; specifically, we randomly reassigned group membership 500 times and fit the model to the two false 'groups' each time to form a null distribution of parameter value differences across groups, and compared the true parameter 'value' difference against this null distribution.
The level of significance was set at 0.05 in all tests.
Medical Sciences (SC2-GM-099626 to SPK) and from the National Science Foundation (BCS-1358955 to SPK).

Data availability
The datasets generated and analyzed in this study are available from the corresponding authors on reasonable request.