Normative tDCS over V5 and FEF reveals practice-induced modulation of extraretinal smooth pursuit mechanisms, but no specific stimulation effect

The neural networks subserving smooth pursuit eye movements (SPEM) provide an ideal model for investigating the interaction of sensory processing and motor control during ongoing movements. To better understand core plasticity aspects of sensorimotor processing for SPEM, normative sham, anodal or cathodal transcranial direct current stimulation (tDCS) was applied over visual area V5 and frontal eye fields (FEF) in sixty healthy participants. The identical within-subject paradigm was used to assess SPEM modulations by practice. While no specific tDCS effects were revealed, within- and between-session practice effects indicate plasticity of top-down extraretinal mechanisms that mainly affect SPEM in the absence of visual input and during SPEM initiation. To explore the potential of tDCS effects, individual electric field simulations were computed based on calibrated finite element head models and individual functional localization of V5 and FEF location (using functional MRI) and orientation (using combined EEG/MEG) was conducted. Simulations revealed only limited electric field target intensities induced by the applied normative tDCS montages but indicate the potential efficacy of personalized tDCS for the modulation of SPEM. In sum, results indicate the potential susceptibility of extraretinal SPEM control to targeted external neuromodulation (e.g., personalized tDCS) and intrinsic learning protocols.


S1 Eye tracking data acquisi on and processing
Eye movements were recorded using a video-based eye tracking system (Eyelink 1000Plus, SR Research Ltd., O awa, Canada) using Eyelink host so ware (version 5.17) with the "standard" heuris c single-stage filter 1 .Par cipants were placed 65 cm in front of an LCD monitor (XL2720, BenQ, Taipeh, Taiwan; 1920 x 1080 pixel, i.e., 49.3° x 28.9° visual size; 120 Hz refresh rate) with their chins stabilized on a chin-forehead rest in a closed room with lights off.Binocular 13-point calibra on (calibra on point posi ons [x, y]  Supplementary figure 1. Eye movement parameters based on the velocity signal.Three tasks were conducted to assess different aspects of smooth pursuit eye movements (SPEM), namely a con nuous pursuit (TRI), con nuous pursuit with blanking (TRIBL) and foveopetal step-ramps (SR).Top: For each task, experimental parameters and posi on plots are shown for le ward and rightward ramps.Bo om: Median velocity traces (black lines) and individual ramps (thin gray lines) for one exemplary subject and all three tasks.Intervals for the velocity gain computa on are indicated by gray shaded areas.Exemplary latencies are marked by blue dots and accelera on/decelera on es mates are marked by green regression lines.For TRI, maintenance gain was computed to quan fy the predic ve SPEM performance with ongoing visual input.In TRIBL trials, velocity gain was computed before the target was switched off (preblank gain, 200 to 400 ms), during blanking (residual gain, 700 to 1000 ms), and a er re-occurrence of the visual target (postblank gain, 1150 to 1450 ms).During SR, early maintenance gain was computed (300 to 700 ms).For TRIBL and SR, accelera on/decelera on es mates at different s mulus intervals are indicated by green lines.Latencies that indicate the start of decelera on/accelera on are marked by blue dots.
Using a custom semi-automa c pre-processing procedure, the raw gaze coordinate data of one eye (missing data in experiment 1 (V5): 0.9 ± 0.7 %; experiment 2 (FEF): 1.2 ± 1.1 %) were corrected for offset and posi on gain errors before con nuous data were lowpassfiltered at 50 Hz using a Gaussian filter.By default, the le eye was used for analysis.However, the right eye was used in rare occasions for recording blocks where limited data quality restricted the analysis of the le , but not the right eye (the right eye was used in 0.9 % (experiment 1, V5) and 5 % (experiment 1, FEF) of oculomotor blocks).The correct detec on of eyeblinks, saccades and intervals of artefactual signal was validated manually (details described in 4 ).SPEM parameters were computed for each of the three oculomotor tasks (TRI, TRIBL, SR), as described in the main manuscript.Experimental details as well as data from one exemplary subject are shown in supplementary figure 1 for illustra on.
Data from N = 8 par cipants were excluded from the analysis of tDCS effects, due to limited eye tracking data quality (V5: N = 3, FEF: N = 4).Specifically, N = 5 par cipants (V5: N = 2, FEF: N = 3) showed overall limited SPEM performance as indicated by the TRI maintenance gain, averaged across mepoints and recording sessions (TRI maintenance gain < 0.85 and zvalue < -1.96, rela ve to the sample in the V5 and FEF experiment, respec vely).Two more par cipants (V5: N = 1, FEF: N = 1) showed considerably low maintenance gain values during one isolated recording session (TRI maintenance gain < 0.8) and, in addi on to qualita vely bad signal quality that prevented or hampered the quan fica on of accelera on and decelera on parameters.

S2 Comparing norma ve with personalized electric fields: Data acquisi on and analysis
A subsample (N = 6) of par cipants from the experiment 1 and 2 furthermore completed a comprehensive assessment that allowed for individual head models and the es ma on of the func onal loca on and orienta on of areas V5 and FEF in the right hemisphere.Based on this informa on, individual electric field simula ons were computed for the norma ve tDCS montages of experiment 1 (V5) and experiment 2 (FEF; see Fig. 1 in the main manuscript), as well as personalized tDCS montages for both areas.This procedure enabled the assessment of individual electric field intensi es with respect to the individual func onal targets in V5 and FEF for both the norma ve and the personalized case.
T1 and T2-weighted MRI data were used to construct individual six-compartment head models (including scalp, skull compacta, skull spongiosa, cerebrospinal fluid (CSF), gray ma er, and anisotropic white ma er).A er registering the T2 onto the T1 using FSL FLIRT 5 , ssues were segmented using CAT12 6 .From the T1, gray ma er, white ma er and scalp were segmented while the T2 was used for CSF, and skull compacta.The spongiosa segmenta on was created by performing Otsu thresholding 7 on the by 2mm eroded skull mask.Overlap of brain ssues and skull/CSF were removed and unrealis c holes within the masks were detected and filled using custom MATLAB-scripts including Boolean and morphological opera ons cf. 8,9.Following the recommenda ons of 10 , the model was cut using an axial plane 4 cm below the skull.From the segmenta ons, geometry adapted hexahedral meshes with a node shi of 0.33 were created 11 .Anisotropic white ma er tensors were computed based on an effec ve medium approach 12 using the DTI data.For this, eddy current and nonlinear suscep bility ar facts were removed using FSL and HySCO 13 .
A 2 mm resolu on source space was constructed in the middle of the gray ma er compartment without restric on to source orienta ons (no normal-constraint).As we used a Venant source model to represent dipolar neural sources, the so-called Venant condi on must be fulfilled meaning that the node closest to the source should be located en rely within gray ma er 14 .All lead fields were computed using the DUNEuro toolbox 15 .For computa on of the EEG lead field, a skull conduc vity calibra on was performed.
Individual skull conduc vity was calibrated based on the dis nct effect of volume conduc on on somatosensory evoked poten als (SEP) and somatosensory evoked fields (SEF) 16,17 .The somatosensory EEG and MEG data were filtered between 20 to 250 Hz.A 50 Hz notch filter was applied (considering harmonics) to account for the power line artefact.Data was epoched (-50 to 150 ms rela ve to onset of electrical pulses) and bad channels and trials were removed semi-automa cally (rejected trials: 9.6 ± 2.8 %).Finally, the data was averaged over the trials and individual P20 components were determined.An MEG dipole scan was performed, and the resul ng dipole loca on was saved.For the EEG, the source space was restricted to include only the result of the MEG dipole scan.Changing only the skull conduc vity (the conduc vity ra o of spongiosa to compacta was fixed to 3.6) we then minimized the residual variance based on Brent's algorithm 18 to es mate the most likely skull conduc vity.The resul ng conduc vi es for skull compacta ranged from 0.005 to 0.032 S/m (0.015 ± 0.01 S/m).For skull spongiosa, compacta conduc vi es were mul plied with a constant of 3.6 and norma ve values were assigned to the other ssues (scalp: 0.43 S/m, CSF: 1.79 S/m, gray ma er: 0.33 S/m, white ma er: 0.14 S/m).For more informa on on head model crea on and skull conduc vity calibra on, see 16,17 .EEG lead fields were recomputed using the calibrated individual skull conduc vi es.

S2.3 Func onal MRI data acquisi on and defini on of s mula on target loca ons
Par cipants were presented with two runs of horizontal smooth pursuit eye movement tasks, while EPI blood oxygen level dependent (BOLD) ac vity was recorded (SMS 4, 307 Volumes, TR = 980 ms, TE = 30 ms, FA = 70°, resolu on 3 x 3 x 3 mm, 68 x 68 x 56 mm).
Par cipants performed smooth pursuit by fovea ng a red dot on an LCD monitor (size 0.5°, black background; screen resolu on 1920 x 1080, refresh rate 60 Hz, NordicNeuroLab, Bergen, Norway) either in the framework of a con nuous pursuit task (18.7°/s, ±15° amplitude, four blocks of each 8 ramps to the le and right), con nuous pursuit with blanking (18.7°/s, ±15° amplitude, blanked from 300 to 1000 ms a er ramp onset, four blocks of each 7 ramps to the le and right, preceded by one con nuous triangular wave), con nuous oscilla ng pursuit with sta onary background (four blocks of 40 s; red dot of size 0.5° oscilla ng at 0.2 Hz, ± 15° amplitude; background: 70 sta onary white dots with size 0.5° (2.5° spacing) and fixa on with moving background (four blocks of 40 s; central fixa on red dot, size 0.5°; background: 70 white dots with size 0.5° (2.5° spacing) moving at 0.2 Hz).Eight blocks with foveopetal stepramps were presented block-wise directed either to the le or the right side of the screen (8 ramps per block, 18.7 °/s, ± 15° amplitude, ± 2.5° step size, inter-trial central fixa on ji ered between 1 and 1.5 s).Each block was preceded by 12 s fixa on intervals with a centrally presented red dot (12 s, size 0.5°).Eye movements were recorded using a video-based eyetracker system (Eyelink 1000Plus, 1000 Hz sampling rate; SR Research Ltd., Ontario, Canada).Func onal images were smoothed using a 6 mm FWHM Gaussian kernel, corrected for slice-ming and co-registered to the normalized T1 image.Individual loca on vectors of right visual area V5 and right Frontal Eye Field (FEF) were determined based on sta s cal maps showing increased BOLD ac vity during con nuous pursuit blocks in contrast to fixa on intervals (T-contrast, p < .05across 5 adjacent voxels, FWE-corrected).The loca on vectors were defined as the local maxima near puta ve V5 and FEF regions that have previously been shown to signal brain ac vity during smooth pursuit eye movement 19,20  A notch filter at 50 Hz (considering harmonics) was applied to account for the power line artefact.A filtering, the data were cut into the final 320 epochs (0 to 1.59 s length rela ve to ramp onsets to le and right) and demeaned.Invalid channels and epochs were rejected semi-automa (rejected trials: 14.5 ± 8.6 %).Finally, the EEG data was re-referenced to the average reference.
For each of 1000 bootstrap samples, the covariance matrices for both EEG and MEG were computed based on a randomly drawn sample of mepoints (with replacement; only samples in the me window between 0.3 to 1.3 s rela ve to ramp onset were considered) comprising the same length as the final MEG/ EEG data of the respec ve par cipant.We used a combined EEG/MEG approach, using a Linearly Constrained Minimum Variance (LCMV) beamformer with Unit-Noise-Gain constraint, and a regulariza on of 5%, which showed the most accurate orienta on es mate in a previous simula on study 21 .The direc on of maximum power was defined as the orienta on 22 .The analysis was performed for both EEG and for MEG, where we used the 3D lead field for EEG as described in S2.2 but reduced the MEG lead field to the two direc ons tangen al to the skull surface, using a singular value decomposi on 23 .We then recombined the tangen al orienta on resul ng from the MEG analysis with the radial orienta on, which we extracted from the resul ng EEG orienta on to obtain a combined EEG/MEG orienta on es mate in each of 1000 repe ons, normalized to the vector length.
Across all 1000 itera ons, the orienta on medians for each direc on (x, y, z) were extracted as orienta on values and the orienta on was normalized to the vector length.

S3 Direc on of SPEM is modulated by a en on
During the assessment of prac ce effects, we observed a direc on-specific facilita on of SPEM performance for le ward ramps during TRIBL with shorter re-accelera latency and faster re-accelera on, both highly related to an cipa on of the re-appearance of the visual target at the end of the blanking interval.SPEM performance is highly related to hemispherebrain ac vity that might explain subtle differences in the direc on of ramps.For example, by unilateral lesion of the FEF, SPEM performance was impaired for pursuit of ipsiversive moving targets 24,25 Electrical micros mula of the V5 homologue region in monkeys induced an accelera on of ipsiversive and a decelera on of contraversive SPEM 26 and chemical lesion resulted in impaired velocity when pursuing ipsiversive mo on s muli 27 .
A facilitated SPEM performance for ramps with le ward direc on might be explained by asymmetries of visuo-spa al a en on that affect this hemisphere-specific representa on of SPEM [28][29][30][31] .First, visuo-spa al a en on is involved in the motor prepara on, and thus the an cipa on of upcoming visual s muli (TRIBL re-accelera on in this study), both during covert shi s of a en on and overt eye movements towards a visual target s mulus [32][33][34][35][36] .During the maintenance of SPEM, visuo-spa al a en on has been shown to constantly shi closely ahead of the pursued visual target 37 .Thus, for le ward SPEM, a en on shi s to the le hemifield, rela ve to the moving target while informa on in the le hemifield is processed in the contralateral hemisphere (and vice versa for rightward SPEM).However, visuo-spa al a en on has been repeatedly shown to be biased towards the le hemifield, a phenomenon o en referred to as pseudoneglect 38 which in turn is related to a right-hemispheric lateraliza on of the structural network underlying (exogenous) a en on 28 .In this study, facilitated SPEM performance for le ward SPEM might reflect those asymmetries in visuospa al a en on that also seem to affect overt eye movements to some extent (Fig. 3).
However, a paradoxical effect was observed for SR pursuit latency, specifically shorter (i.e., facilitated) latency for rightward ramps, compared to le ward ramps.One might speculate that the SPEM ini a on might involve a disengagement of a en on from the non-target direc on, due to the unknown target direc on of the SR (le ward or rightward).Like the mechanisms that facilitate contraversive SPEM in predictable situa ons like during TRIBL or during closed-loop processing (SR early maintenance gain), a right-hemispheric dominance of the a en on network might also facilitate the disengagement of a en on from the le hemifield to allow a faster ini a on of rightward SPEM during ini a on in less predictable situa ons.

table 1 .
and were validated across the applied tasks.Target loca ons for V5 and FEF.Loca on vectors were determined based on the contrast compu ng larger BOLD ac vity during con nuous pursuit compared to central fixa on.MNI-coordinates [x, y, z] and Tvalues of local maxima are depicted for the right V5 and the right FEF during con nuous pursuit for subjects S1 to S6 (* indicate p < .05acrossat least 5 adjacent voxels, FWEcorrected).S mulus presenta on was generated by PsychToolbox (version 3.0.18,Brainard,1997; Kleiner, Brainard & Pelli, 2007) on Matlab (R2019b, The Mathworks, Na ck/MA).hps://www.fil.ion.ucl.ac.uk/spm, version 7771 on Matlab R2021b, The Mathworks, Na ck/MA).S2.4 Combined EEG/MEG data acquisi on during smooth pursuit to define target orienta onIn addi on to the above-described data acquisi on for skull conduc vity calibra on, (To compute the orienta on for the V5 or FEF loca on determined by fMRI, the gray ma er source grid points (see S2.2) with the smallest Euclidian distance to the fMRI target loca (right V5 and right FEF) were defined as regions of interest.EEG and MEG data were cut into 20 epochs (one per block of the pursuit task; 27 s length) and highpass-filtered at 0.1 Hz.
6.In both experiments and across tDCS condi ons (including sham tDCS) mainly absent to slight somatosensory and pain percep on was reported.

table 4 .
Prac ce effects observed in experiment 1 (V5,N = 27).Results of linear mixed model analysis for each es mated oculomotor parameter indica ng learning effects across sessions (day) and within sessions ( mepoint).F-values and p-values for main effects and mepoint * day interac on effects are reported.Besides a significant mepoint * direc on interac on for re-accelera on (F = 2.96, p = .032),no significant effects were observed for the remaining interac on effects in the saturated model, thus, these effects are omi ed in the table.Asterisks indicate significant effects with p < .05.TRI = con nuous pursuit.TRIBL = con nuous pursuit with blanking.SR = foveo-petal step-ramps.

table 5 .
Prac ce effects observed in experiment 2 (FEF,N = 25).Results of linear mixed model analysis for each es mated oculomotor parameter indica ng learning effects across sessions (day) and within sessions ( mepoint).F-values and p-values for main effects and mepoint * day interac effects are reported.No significant effects were observed for the remaining interac on effects in the saturated model, thus, these effects are omi ed in the table.Asterisks indicate significant effects with p < .05.TRI = con nuous pursuit.TRIBL = con nuous pursuit with blanking.SR = foveo-petal step-ramps.Supplementary table 6. tDCS side-effects during experiment 1 (V5) and experiment 2 (FEF).Rela ve frequencies (in %) of subjec vely perceived somatosensory and pain sideeffects.