Relationship Between the Visual Evoked Potential and Structure in the Primary Visual Cortex in Healthy Individuals and in Patients with Severe Mental Disorders

Schizophrenia (SCZ) spectrum and bipolar disorder (BD) are severe mental disorders with unknown pathophysiology. Altered visual evoked potential (VEP), an electroencephalogram signal reecting function in the primary visual cortex (V1), abnormal visual processing and visual hallucinations reported in these patients, all point towards V1 dysfunction. While the mechanisms contributing to V1 dysfunction remain unknown, structural alterations are possible candidates. Lack of insight into neural substrates of structure and functional in V1 has limited our ability to determine implications of altered V1 function. While combining VEP and magnetic resonance imaging has increased our understanding of the structure-function relationship in V1 in healthy individuals, no previous study has examined the same structure-function relationship in patients with SCZ spectrum and BD. Here, we aimed to conrm previous ndings of a selective positive correlation between the amplitude of the P100 component of the VEP and V1 surface area (SA) in 307 healthy individuals and to examine whether this relationship was altered in patients with SCZ spectrum (n=30) and BD (n=45). The correlation between the P100 amplitude and the total, (r=0.16, p=0.006), right (r=0.14, p=0.013) and left V1 surface area (r=0.13, p=0.02) was signicant in healthy individuals, but not in patients. The current results support previous ndings of a selective relationship between P100 amplitude and V1 surface area in healthy individuals and suggests that other factors than V1 surface area or thickness explain V1 dysfunction reported in these patients.


Introduction
Schizophrenia (SCZ) spectrum and bipolar disorders (BD) are severe mental disorders with largely unknown pathophysiology [1][2][3][4] . Studies on how the brain processes speci c stimuli using electroencephalography (EEG) 5 in the form of event related potential (ERP) techniques have reported abnormal brain function in the primary visual cortex (V1) in these disorders [6][7][8][9][10][11][12][13][14][15][16][17][18] . The visual evoked potential (VEP) is a robust ERP signal primarily re ecting post-synaptic potentials of pyramidal cells in V1 and provides information on the functional integrity of the visual system 16,19,20 . In humans, checkerboard reversal stimulation triggers a VEP with three main components, the N75, the P100 and the N145 5,21 , that can be extracted from scalp-recorded EEG. While the exact cortical sources of the VEP components remain to be fully clari ed, previous studies suggest that the VEP components are generated mainly in V1 areas 22,23 . See gure 1 illustrating how checkerboard reversal triggers a VEP response that can be recorded with scalp-EEG. The amplitude of the VEP, re ecting function in V1 20 , is reduced in SCZ spectrum disorder 13 and altered VEP plasticity -a correlate of long-term potentiation-like visual cortex plasticity -is described in SCZ spectrum and BD 7,8,14,18 . In addition to altered VEP, abnormal visual processing 24,25 , impaired neural activity in V1 areas 26 and visual hallucinations [27][28][29][30] all point towards dysfunction in V1 areas in these patients. While the precise mechanisms contributing to such V1 dysfunction remains unknown, structural alterations are possible candidates. While magnetic resonance imaging (MRI) studies show group-level alterations in brain cortical structure [31][32][33][34][35][36][37][38][39][40][41][42] , no studies have reported speci c alteration in the V1 surface area or thickness in patients with SCZ spectrum or BD. This suggest that other factors than surface area or thickness in V1 may explain V1 dysfunction, as re ected by altered VEP in these patients [43][44][45][46] . While previous studies have investigated structure and function in the visual cortex with functional MRI (fMRI) [47][48][49][50] , studies utilized the enhanced spatio-temporal resolution obtained when combining structural MRI (sMRI) and VEP to explore the structure-function relationship in V1 are lacking. While we previously found a positive correlation between the amplitude of the P100 component of the VEP and the V1 surface area 51 in a small sample of healthy individuals, no previous study has examined the same relationship in patients with SCZ spectrum and BD 44,47,48,50 . Combining VEP and sMRI may provide new insights into neural substrates of functional alterations in V1 in these disorders. However, to determine implications of V1 dysfunction in these patients, we rst need to gain more knowledge on the structure-function relationship in V1 in healthy individuals. Here, we examined 307 healthy individuals and participants with SCZ spectrum (n=30) or BD (n=30) with a standardized VEP paradigm and brain sMRI with the following main aims: (1) to test the reproducibility of a previously reported positive and selective relationship between the amplitude of the P100 component of the VEP and the V1 surface area in a larger sample of healthy individuals 51 , and (2) to assess whether the same structure-function relationship was altered in patients with SCZ spectrum and BD. We hypothesized to nd a positive and selective relationship between the P100 amplitude and the surface area, but not thickness of V1 in healthy individuals 51 . Further, we hypothesized to nd an altered V1 structure-function relationship in patients with SCZ spectrum and BD. No previous study has shown evidence for reduced V1 surface area or thickness in this patient group. Thus, we hypothesized that other factors than V1 surface area or thickness explain V1 dysfunction as re ected by altered VEP.

Sample description
Our study sample included healthy individuals and patients with a diagnosis of SCZ spectrum or BD who participated in the TOP-Study at the Norwegian Centre for Mental Disorders Research (NORMENT) located in Oslo, Norway. Patients with a DSM-IV diagnosis of SCZ spectrum disorder (including schizophrenia, schizophreniform, schizoaffective and psychotic disorder not otherwise speci ed (NOS)) or BD (including bipolar type 1, bipolar type 2 and bipolar NOS) were included, provided that they were deemed capable of giving formal consent. Healthy individuals were recruited from the national population register in Norway or through advertisement in the local newspaper. Only participants with a normal or corrected-to-normal vision, assessed using a Snellen eye chart, were included. All participants with a history of head trauma with loss of consciousness, an IQ < 70, or a somatic or neurological disorders believed to in uence brain function were excluded. Exclusion criteria for healthy individuals were as follows: 1.) any history of mental disorders, severe mental disorders in rst degree relatives, 2.) any history of alcohol-and substance abuse or dependence 3.) use of cannabis in the last three months prior to inclusion. Further healthy individuals and patients with any contraindications for MRI (e.g., metallic implants, pacemakers or MRI-incompatible prosthetic heart valves) or EEG examinations (e.g., history of epilepsy), with any incidental ndings on the MRI caput indicating brain pathology potentially in uence our analyses, with poor quality of V1 area mask and with a time interval between the EEG and MRI examinations > 11 months were further excluded. In our main and secondary analyses participants with a diagnosis of SCZ spectrum or BD were referred to as patients and analyzed together as one group. An overview of the nal study sample is provided in Table 1. The study was approved by the Regional Committees for Medical and Health Research Ethics of South -Eastern Norway. All methods were performed in accordance with the Declaration of Helsinki and the project leader was an authorized clinical psychiatrist, legally obliged to follow the Health Personnel Act. The project was granted with license from Norwegian Information Council (Datatilsynet) to collect and store sensitive information and human biological material until 2050. All participants provided written informed consent. All participation was voluntary and independent of any treatment provided. The participants could whenever they want, withdraw from the study without this having any consequences for the treatment..  52 . In brief, processing steps included removal of nonbrain tissue, automatic Talairach transformation and intensity correction. Intensity information was used to reconstruct the inner (i.e., the gray/white matter boundary) and outer (i.e., the gray matter/cerebrospinal uid boundary) surface of the cerebral cortex through a series of processing steps as previously described [53][54][55] 55 to obtain measures of the total (right + left), right and left V1 surface area, mean (average) V1 cortical thickness ((right + left V1 cortical thickness) / 2)) in addition to estimated total intracranial volume (eTIV). This method locates the V1 using a surface-based probabilistic atlas derived from high-resolution sMRI of the stria of Gennari. Manual editing of surface reconstruction errors was performed and V1 masks were visually inspected to ensure satisfactory placement. In addition to these quality assurance steps we included FreeSurfer Euler numbers as covariates in our analyses to evaluate MRI image quality 56 . We calculated "the total corrected Euler number" from numbers of holes in the original (non-corrected) cortical surface, using the following formula: 2*(#holes -1) for both the left and right V1 surface area.

EEG data acquisition, analyses and quality control
EEG data was collected according to the international 10-20 system from a BioSemi ActiveTwo ampli er. Ag-AgCl scalp electrodes (n=64) recorded brain activity, four external electrodes recorded lateral and vertical movement of the eyes and two external electrodes recorded the heart rhythm (electrocardiography). Potentials at each channel were sampled at 2048 Hz with respect to a common mode sense with a driven right leg electrode minimizing common mode voltages. The VEP paradigm was adopted from Normann et al. 57 . In the current study we focused on VEPs elicited during a prolonged period of pattern reversing checkerboards, where checkerboards changed color from white to black with a spatial frequency of 1 cycle/degree over ~ 28 ° visual angle and a xed frequency of 2 reversals per second. We extracted N75, P100 and N145 amplitudes from the intervention phase of the VEP paradigm that lasted 10 min, yielding a total of 1200 reversals. After collecting o ine VEPs, the EEG data was processed as described previously 18,58,59 using MATLAB and the EEGLAB toolbox for MATLAB 60 . After down-sampling to 512 Hz, noisy channels were interpolated using the Prep-Pipeline algorithms 61 with default mode while muscle, eye blinks and eye movement artefacts were eliminated using SASICA defaults 62 . We decomposed the EEG data into independent components using the SOBI algorithm 63 . Epochs with amplitude exceeding 100 µV were removed and the AFz electrode was used as the reference electrode for all channels. ERPs were extracted from the Oz channel. As a quality index for EEG, we computed the standard error (across trials) of the VEP amplitudes for each participant, and standardized this index by dividing them by the mean VEP amplitudes (i.e., expressing the cross-trial variance as a percentage of the total amplitude). See gure 1 illustrating the VEP experiment.

Statistical analyses
All statistical analyses were conducted in R version 3.6. (https://www.r-project.org, R Core Team, 2014) and gures were produced using the ggplot2 package implemented in R (Wickham, 2009 To replicate previous ndings of an intimate relationship between V1 structure and function in healthy individuals, we ran partial Pearson correlation analyses between the P100 amplitude and the total, the right and the left V1 surface area. Prior to running correlation analyses we corrected for the effect of age, sex, eTIV and Euler number on V1 surface area, and for effects of age, sex and standard error for P100 amplitude on the P100 amplitude, in healthy individuals (n= 307). To test our hypothesis that the same structure-function relationship was altered in patients with SCZ spectrum and BD, we ran the same correlation analyses in our patient group and performed the paired r test implemented in R to assess whether the P100-V1 correlation was signi cantly different between healthy individuals and patients.

Vertex-wise whole brain analyses in healthy individuals
To test if a putative relationship between the P100 amplitude was speci c to the surface area in V1, we ran whole cortex analyses in our group of healthy individuals. General linear models (GLMs), implemented in FreeSurfer 6.0 were used to test the associations between P100 amplitude and surface area in each vertex of the cortical surface, correcting for the effects of age, sex, eTIV and Euler number. A different offset, same slope model was speci ed. Similar models were used to test all associations speci ed above. To correct for multiple testing across space, cluster-wise correction was performed using the following parameters: Cluster-forming threshold: p<0.05, unsigned; precomputed Monte Carlo simulation (--cache); cluster-wise probability: p<0.05; correction for analyses of two hemispheres (--2spaces) applied.

Correlation between the P100 amplitude and V1 surface in healthy individuals and in patients
We found a positive correlation between the P100 amplitude and the total, the right and the left V1 surface area in healthy individuals (n=307), but not in patient (n=75

Correlation between the amplitude of the other VEP components and V1 structure (surface area and cortical thickness) in healthy individuals
We did not nd any signi cant correlation between the amplitude of the N75 and the total (r=-0.007, p=0.9), the right (r=0.02, p=0.73) or the left (r=-0.033, p=0.57) V1 surface area or V1 cortical thickness (r=-0.0002, p=1). Further, we observed no signi cant correlation between the N145 and the total (r=-0.085, p=0.14), the right (r=-0.042, p=0.47) or the left (r=-0.11, p=0.062) V1 surface area or V1 cortical thickness (r=-0.098, p=0.087). A corrected p-value of <0.008 was considered signi cant.

Vertex-wise whole brain analyses in healthy individuals
Before correction for multiple testing, we observed a bilateral positive association between the P100 amplitude and V1 surface area (p< 0.01, uncorrected). No other cortical regions were observed to have similar correlations in both hemispheres. However, after cluster-wise correction for multiple testing, ndings of a positive association in the V1 were no longer signi cant. See gure 4 illustrating signi cant correlation between P100 amplitude and surface area in each hemisphere.

Discussion
The current study con rms our hypothesis and previous ndings of an intimate relationship between V1 surface area, but not thickness, and V1 function, re ected by P100 amplitude, in healthy individuals 51 . The non-signi cant correlation between P100 amplitude and V1 structure (surface area and thickness) in our patient group, support our hypothesis of an altered structure-function relationship in SCZ spectrum and BD.

Underlying mechanism explaining the correlation between surface area and function in V1 in healthy individuals
The exact mechanism explaining the positive correlation between V1 surface area and function in the same area, as re ected by P100 amplitude in healthy individuals remains elusive. However, the pool of synapses and the width and/or the number of cortical columns in V1 potentially increases with enlarged surface area, enhancing the summation of postsynaptic potentials (PSP) generated by V1 pyramidal neurons. Since the P100 amplitude primary re ects PSP in V1 pyramidal cells, a greater summation of PSP with a larger surface area may explain the positive correlation between V1 surface area and the P100 amplitude 44 . Further, the level axonal myelination in V1 may correlate with the V1 surface area in healthy individuals. Enhanced axonal myelination with larger V1 surface area may enable V1 neurons to re more synchronously 45,49,64−67 . Since scalp EEG re ect only synchronized neuronal activity, a larger surface area and enhanced myelination in V1 may result in a larger P100 amplitude. Further, the level of myelination in uences the conduction velocity of action potentials (AP) through presynaptic axons. If APs from a large group of presynaptic cells arrive at the presynaptic terminal simultaneously, neurotransmitters are released from the presynaptic cell into the synaptic cleft, bind to the receptors on the postsynaptic pyramidal neurons and generate PSPs. The summation of a PSPs from large groups of synchronized neurons can be recorded with scalp electrodes (EEG) with a larger P100 amplitude re ecting enhanced summation of PSPs. Since the V1 is highly myelinated, the level of myelination in V1 may mirror function in the same area as re ected by P100 amplitude, either by enhancing summation of PSPs or indirectly by increasing the diameter of pyramidal axons, or both 64,65,67−70 . Precisely how the level of myelination in V1 mirrors P100 amplitude remains elusive, but are research topics we plan to study in the future. Further, the level of neurotransmitters released by presynaptic neurons, either in the form of glutamate or gamma aminobutyric acid (GABA) possibly indirectly affects generation of PSPs in pyramidal cells 71,72 . A larger number of synapses may correlate with an increased release of excitatory neurotransmitters (glutamate) and enhanced summation of PSPs, further explaining our ndings of a positive relationship between P100 amplitude and V1 surface area. However, while we know that the level of GABA and glutamate in uences the activation of neurons, exactly how these neurotransmitter in uence VEP amplitudes, remains elusive.

4.2.
The structure-function relationship in V1 in patients with SCZ spectrum and BD While reduced P100 amplitude and V1 dysfunction has been reported in patients with SCZ spectrum and BD, no previous study has demonstrated altered V1 surface area or thickness in these patients, suggesting that other factors than altered V1 surface area or thickness explain V1 dysfunction. Of interest, in the current study we found no signi cant difference in mean total V1 surface area or mean P100 amplitude between patients and healthy individuals (see Supp Fig. 7 in supplementary  information). However, to our knowledge, no previous study has combined sMRI and VEP to investigated whether and how V1 structure mirrors V1 function in patients with SCZ spectrum and BD. While the P100-V1 correlation was not signi cantly different between healthy individuals and patients, the current and previous ndings support our hypothesis that other factor than reduced V1 surface area explains reduced VEP amplitude 13 and altered VEP plasticity 7,8,14,18 in these patients. To this end, we can only speculate about what factors might explain V1 dysfunction that manifested clinically as visual hallucinations [27][28][29][30] and altered visual processing 24,25 . While patients with SCZ spectrum show reduced number and degeneration of V1 neurons in addition to reduced V1 volume, whether and how these factors correlate with V1 function, as re ected by VEP, remains unknown 46,73,74 . Synaptic dysfunction 75 , aberrant synchronization 76,77 , abnormal regulation of neurotransmitters 72,78 and /or abnormal myelination in V1 79-83 may also explain altered function in V1 in these patients. Further, diminished dendritic arborization of V1 neurons and altered synaptic pruning in SCZ spectrum disorders may result in reduced summation of PSPs and hence reduced P100 amplitude. Since the process of synaptic pruning is preceded by the development of the V1 cortex (including V1 surface area and thickness) abnormal synaptic pruning could explain altered VEP with normal V1 surface area and V1 cortical thickness in patients with SCZ spectrum and BD 84-86 . Studies examining the exact mechanisms to which all factors mentioned above might in uence VEP characteristics are missing. Knowledge on how these (structural) factors in uence function in V1 may provide new insight into the neural substrates of altered brain function in severe mental disorders. While previous studies show evidence for altered myelination in the visual cortex in patients with SCZ spectrum and BD [79][80][81][82][83] , whether and how altered myelination explains V1 dysfunction, remains unknow. Combining MRI measures of myelination and VEP measures re ecting function in V1 may reveal whether abnormal myelination in V1 can explain V1 dysfunction, exhibited clinically as visual hallucinations and altered visual processing. 4.3. Why is V1 surface area, but not V1 cortical thickness associated with the P100 amplitude? Current and previous evidence for a correlation between the P100 amplitude and the surface area, but not the thickness of V1 in healthy individuals indicate that surface area and thickness in uence cortical function in different ways. Cortical surface area and thickness are both heritable traits, but differ phenotypically and genetically and result from different ontogenetic stages during development of the cortex 38,87−95 . While cortical thickness is considered to be in uenced by environmental and neurodegenerative factors, cortical surface area appears more determined by genetics and is in uenced by early neurodevelopmental factors 38,88, [92][93][94]96 . Further, cortical surface area is determined by the total number of cortical columns while cortical thickness is in uenced by the number of cells within each cortical column 90,91 . The selective relationship between P100 amplitude and the surface area, but not thickness in V1 might be due to the organization of the cortex in columns and lamina, where the number and/or width of vertical columns correlated positively with surface area, but not thickness in V1 91 . Schwarzkopf and colleagues argued that their ndings of a selective relationship between cortical surface area, but not thickness and the gamma-band frequency EEG signal was unrelated to gray matter volume, and that the distance across the cortical sheet was the relevant factor for the association 44 . Although speculative, increased surface area may re ect an increased number of vertically organized pyramidal cells in V1 while an increase in cortical thickness may mirrors increased length of pyramidal axons, which does not result in enhanced PSPs re ected by increased P100 amplitude. Intriguingly, one of few previous ERP-MRI studies revealed a signi cant, but negative correlation between the auditory N1 amplitude, an ERP component re ecting function in the auditory cortex, and the thickness, but not surface area, in the auditory cortex. These ndings support previous reports a positive correlation between thinner cortex and function in the auditory cortex 97,98 . Albeit ndings of a positive correlation between N1 amplitude and thickness in the auditory cortex is not directly comparable to the current study ndings, these results may suggest that increased surface area and thinner cortex mirror increased neural activity in some cortical brain areas 99 .

Strengths and limitations
Compared to the previous study by Elvsåshagen and colleagues 51 , we corrected for the effect of additional potential confounders on our MRI and ERP measures prior to running analyses. Further, we performed extensive quality control on our structural and functional measures, including adding Euler numbers as a quality index for MRI and standard error for P100 amplitude to control for poor EEG quality. Lack of power due to a limited number of patients included complicated the interpretation of our results. Further, we were not able to control for potential effects of skull thickness, cerebrospinal uid and meninges on the conductance of electromagnetic elds from their neural sources to scalp electrodes. This should preferably be accounted for when analyzing ERP data. Further, we did not correct for plausible effects of medication use, disease states or symptom severity, including prevalence of visual hallucinations for the patients included.

Conclusion
The current study ndings support the hypothesis of an intimate relationship between the surface area and function in V1 in healthy individuals and shows evidence for an altered structure-function relationship in patients with SCZ spectrum and BD. We need to investigating the relationship between V1 function, as re ected by VEP, and structural components in V1 other than surface area and thickness to gain new insight into neural substrates of altered V1 function in these patients.
Declarations honoraria from Lundbeck and Janssen Cilag and is a consultant to BrainWaveBank. O.A.A. has received speaker's honorarium from Lundbeck and Sunovion and is a consultant to HealthLytix. The other authors report no con icts of interest. EEG data is currently not openly available due to ethical and privacy issues of clinical data. All codes used for the statistical analyses are available through R studio.  Figure 1 Visual Evoked Potential. The gure illustrates how visual stimuli in the form of checkerboard reversals stimulates the visual system all the way from the retina to the primary visual cortex (V1) and how the visual evoked potentials (VEPs) can be recorded with scalp-EEG.

Figures
Step 1. Scalp-EEG is recorded while the participant is focusing on a red dot in the middle of a computer screen (placed approximately 50 cm in front of them) while exposed to reversals of checkerboards.
Step 2. The reversal of checkerboards (white squares change color to black and vice versa) stimulates the visual system to send electrical impulses from the retina, through the optic nerve and towards V1.
Step 3. When the electrical impulses reach V1, presynaptic neurons (interneurons) generate action potentials (Aps) that trigger the presynaptic cells to release neurotransmitters from the presynaptic terminals into the synaptic clefts.
Step 4. Binding of the neurotransmitters to receptors on the postsynaptic neurons (pyramidal cells) generate temporary changes in the postsynaptic membrane potentials, also known as postsynaptic potentials (PSPs).
Step 5. Summation of PSPs from a large number of synchronized pyramidal neurons can be recorded from scalp electrodes overlying the V1 area.
Step 6. Grand average VEPs measured at the occiput (Oz) with anterior reference (AFz) during intervention phase of the rst VEP paradigm that lasted for 10 min, yielding a total of 1200 reversals. Ms=milliseconds; µV=microvolts. Parts of the symbols used in gure 1 are taken from Servier Medical Art (https://smart.servier.com/).

Figure 2
a Pearson correlation between the P100 amplitude and total V1 surface area in healthy individuals and in patients. b Pearson correlation between the P100 amplitude and right V1 surface area in healthy individuals and in patients. c Pearson correlation between the P100 amplitude and left V1 surface area in healthy individuals and in patients. Results from the paired r test revealed a non-signi cant difference in the correlation between the P100 amplitude and the V1 surface area (total, right and left) between healthy individuals and patients (p>0.02).

Figure 3
Pearson correlation between P100 amplitude and the V1 cortical thickness in healthy individuals (n= 307) and patients (n=75) with SCZ spectrum and BD.