{"paper_id":"2c47500a-ced1-4fbd-9c49-48f6303a0004","body_text":"The effects of action-based predictions in early visual cortex\nBianca M. van Kemenade1,2, Lars F .Muckli1\n1School of Psychology and Neuroscience, University of Glasgow, UK\n2Center for Psychiatry, Justus Liebig University Giessen, Germany\nAbstract\nDuring voluntary movement, predictions about the sensory consequences of an action\ntypically result in reduced sensory sensitivity. The forward model theory proposes that this\nreduced sensitivity is due to neural suppression or cancellation of sensory action\noutcomes. However, recently this theory has been challenged by three alternative theories:\nthe pre-activation account, sharpening, and the opposing processes theory. In this fMRI\nstudy, we directly tested and compared these four theories using univariate and\nmultivariate analyses both prior to and during stimulus presentation. Participants\nperformed a visual orientation discrimination task on two sequentially presented gratings,\nwhich were either presented automatically (passive condition) or triggered by their own\nbutton press (active condition). Auditory cues at trial onset indicated the overall grating\norientation, followed by a preparatory phase in which participants anticipated the upcoming\nstimuli. During this phase, the predicted stimulus orientation was decodable from early\nvisual cortex activity in both active and passive conditions at levels significantly above\nchance, indicating pre-activation of the predicted orientation, but with no difference\nbetween active and passive conditions. BOLD responses did not emerge earlier in active\nconditions, arguing against the pre-activation theory. During stimulus presentation, actively\ngenerated stimuli elicited larger BOLD responses than passively presented ones,\ncontradicting the forward model theory, which predicts overall response suppression.\nDecoding accuracy did not differ between conditions, inconsistent with the sharpening\nhypothesis, which predicts enhanced neural precision for actively generated stimuli.\nInstead, our findings align most closely with the opposing processes theory, which posits\npre-activation in both conditions. However, the stronger BOLD response for actively\ngenerated stimuli is not predicted by any existing theory, suggesting that additional\nmechanisms—such as heightened attention or motor-related enhancement—may\ncontribute to sensory processing during self-initiated actions.\nIntroduction\nOur actions produce sensory outcomes, contributing substantially to the constant stream\nof sensory input that we encounter every day. It was proposed decades ago that we\nsuppress sensory consequences of our own action using predictions generated by an\ninternal forward model (Wolpert et al., 1995). According to this theory, this internal forward\nmodel uses the efference copy, a copy of the motor command, to generate predictions\nabout the sensory outcomes of our action (Sperry, 1950; von Holst & Mittelstaedt, 1950).\nThese predictions are then used to cancel out sensory action consequences, or generate\nprediction errors in the case of a mismatch between prediction and sensory feedback. This\nmechanism is thought to aid us in guiding our actions, learn new motor skills, save\nresources, and distinguish self from other (S. Blakemore & Frith, 2003; Wolpert &\nFlanagan, 2001). In line with this idea are numerous studies that show reduced neural\nactivity elicited by self-generated stimuli compared to identical but externally generated\nstimuli (S.-J. Blakemore et al., 1998; Lubinus et al., 2022; Martikainen et al., 2005; Shergill\net al., 2013; Straube et al., 2017). For many years, this theory has shaped our thinking\nabout sensorimotor mechanisms. However, in recent years this theory has been\nchallenged by three alternative theories. First of all, the preactivation account posits that\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nsuppression of self-generated stimuli is not due to a true dampening of neural activity, but\nrather a pre-activation of sensory areas (Roussel et al., 2013). The idea is that due to this\npre-activation, the baseline activity prior to stimulus onset is increased, making it harder to\ndistinguish the actual stimulus activity from the baseline activity, leading to an apparent\nsuppression during stimulus presentation. However, evidence for this account using\nneuroimaging techniques is lacking. Secondly, studies on sensory predictions have\nsuggested that the apparent suppression induced by sensory expectations, so called\n“expectation suppression” (Summerfield et al., 2008), may instead be due to a sharpening\nof the underlying neuronal populations. This neural sharpening may resemble suppression\nwhen using univariate analysis, but can be detected using multivariate pattern analysis\nshowing better decoding of expected stimuli (Kok et al., 2012). It has been proposed that\naction also leads to neural sharpening (Y onet al., 2018), but to our knowledge no-one has\ndirectly compared active with passive conditions using this approach to confirm that the\neffects of action on neural activity can be specifically attributed to neural sharpening. Lastly,\nthe recently proposed „opposing process theory“ suggests that action-based predictions\nand general predictions about our environment are based on the same mechanism and\ninvolve different time scales (Press et al., 2019). According to this theory, any prediction\nabout an upcoming event firstly pre-activates neurons tuned to the expected features.\nDuring stimulus presentation, “cancellation” may occur in the form of boosting the\nunexpected event compared to the expected event, but only when the unexpected event is\nlikely to be informative. This theory unites the fields of action-based and general sensory\npredictions, but as it was proposed only very recently, the support for this theory is still\npreliminary. Evidence for pre-activation of neurons tuned to expected features exists in the\nperception literature (Kok et al., 2014, 2017), but evidence for neural pre-activation for\naction-based predictions, as well as a supposed similarity between these two types of\npredictions, is lacking.\nIn this fMRI study, we combined univariate and multivariate analysis techniques both prior\nto and during stimulus presentation in order to directly test and compare all four theories.\nThis strategy allowed us to test for changes in both overall neural amplitude and\nrepresentational content during both prediction and perception phases (see T able 1). If\naction cancels out sensory outcomes through dampening as the forward model theory\nsuggests, we would expect both reduced neural activity and reduced decoding accuracy\nfor self-generated stimuli. If action instead sharpens neural responses, we would expect\nreduced neural activity but increased decoding accuracy for self-generated stimuli. If\naction modulates neural activity through pre-activation, we would expect increased activity\nand increased decoding accuracy during preparation, and suppressed neural activity and\ndecreased decoding accuracy during stimulus presentation. Lastly, if action-based and\ngeneral sensory predictions follow the same principles as outlined in the opposing process\ntheory, we would expect a pre-activation of expected units through significant above-\nchance decoding of the upcoming stimulus during preparation, with no differences\nbetween self-generated and externally generated conditions. During stimulus presentation,\nwe would expect no differences in activity or decoding accuracy, at least when using highly\npredictable stimuli.\nT ab.1 Preparatory\nactivity\nPreparatory\ndecoding\nStimulus activity Stimulus decoding\nForward Model A- < P A- < P\nSharpening A- < P A+ > P\nPre-activation A+ > P A+ > P A- < P A- < P\nOpposing process A = P A+ = P+ A = P A+ = P+\nTable 1. Four models‘ hypotheses about activity level and decodability of visual features during the\npreparation phase and in response to stimulus presentation. Marked in blue are unique predictions.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nResults\nParticipants performed both a behavioural and an fMRI experiment, that used the same\nstimuli and experimental paradigm (see Fig. 1). Participants had to perform an orientation\njudgment task on pairs of gratings that they either caused themselves to appear (active\nblocks) or that appeared automatically (passive blocks). Each trial started with an auditory\ncue that predicted with 100% validity the overall orientation of two consecutively presented\ngratings (45° leftward or 45° rightward). There was a preparation phase of eight seconds\nbetween the auditory cue and the active/passive stimulus generation, during which only a\nfixation dot was presented on the screen, but participants already had a prediction about\nthe upcoming stimuli due to the cue. We performed our fMRI analyses for this preparation\nphase and for the stimulus phase separately. In the behavioural experiment, the\npreparation phase after the cue was reduced to three seconds.\nFig. 1. Experimental paradigm. Active and passive conditions were presented in separate blocks. Each trial\nstarted with a predictive cue indicating the overall orientation of the upcoming gratings and informing\nparticipants which button to press in the active condition. After the cue, a preparatory phase started during\nwhich only a fixation dot was shown. Then the fixation dot briefly turned red, which meant that participants\nhad to now press a button with their index or middle finger to elicit the presentation of the gratings (active\ncondition) or had to wait for the gratings to be presented automatically (passive condition). The two gratings\ndiffered slightly in their orientation (+/- 0.5-6.5°), and participants were asked to judge whether the second\ngrating was tilted clockwise or counterclockwise compared to the first one. They had to report their answer\nwhen the dot turned green after grating offset. Any time that remained from the button press window in the\nactive condition to elicit grating presentation was added to the ITI (7-11 s, mean 9 s). This means that trials\nhad on average the same duration (24,1 s). The mapping between cue frequency, button, and grating\norientation was counterbalanced across participants.\nBehavioural effects\nWe fitted psychometric functions to the data from the behavioural experiment for active\nand passive conditions separately and derived thresholds and slopes. Furthermore, we\ncalculated the overall accuracy. Paired-samples t-tests testing for differences between\nactive and passive conditions revealed no significant differences, neither for thresholds\n(t(27)=0.209, p=0.836), nor slopes (t(27)=-0.216, p=0.831) or overall accuracy\n(t(27)=0.310, p=0.759), as displayed in Fig. 2.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nFig. 2. Behavioural results. Boxplots showing the thresholds and slopes derived from psychometric\nfunctions for each condition, as well as the overall accuracy.\nWe applied the same approach to the behavioural data obtained during the fMRI\nexperiment. Replicating the findings from the behavioural experiment, we found no\nsignificant differences between active and passive conditions, neither for thresholds\n(t(25)=0.613, p=0.546), nor slopes (t(25)=0.155, p=0.878) or overall accuracy (t(25)=0.059,\np=0.954), see Fig. 3.\nFig. 3. Behavioural results from the\nfMRI experiment. Boxplots showing\nthe thresholds and slopes derived\nfrom psychometric functions for each\ncondition, as well as the overall\naccuracy.\nUnivariate analyses\nSelf-generated stimuli evoked more activity than externally generated stimuli in early visual\ncortex (t(25) = -4.463, p = 0.00015). During preparation, no significant differences between\nactive and passive conditions were found (t(25) = -1.392, p = 0.176), see Fig. 4. The\nresults were comparable across all ROIs (see Supplementary Material).\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nFig. 4. Univariate results. We performed paired t-tests to test for differences between active and passive\nconditions. Boxplots show the extracted beta values from early visual cortex (EVC). n.s. = not significant, ***\n= p < 0.001.\nNext, in order to test whether there were any differences in the timing of the BOLD\nresponse, we performed a deconvolution GLM. As can be seen in Fig. 5, responses did\nnot differ in their timing. Instead, amplitude differences were found in the late preparatory\nphase, with more sustained activity in the passive condition (V3: p = 0.0018; all other p <\n0.001), and in the stimulus phase, with a larger response in the active condition (all p >\n0.001). Responses did not differ in amplitude in the early preparatory phase (EVC: t(25) = -\n1.109, p = 0.267; V1: t(25) = -0.724, p = 0.469; V2: t(25) = -1.699, p = 0.089; V3: t(25) = -\n1.870, p = 0.062).\nFig. 5. Results from deconvolution GLM. Time courses are plotted for each condition, time-locked to the\nstart of a trial (auditory cue). Error bars represent the standard deviation. n.s. = not significant, ** = p < 0.01,\n*** = p < 0.001.\nOur analysis of eye tracking data showed that the differences between active and passive\nconditions cannot be explained by differences in fixation behaviour, neither during the\npreparatory phase (t(13)=-1.345, p = 0.202) nor during the stimulus phase (t(13)=-1.752, p\n= 0.103).\nMultivariate analyses\nFig. 6 illustrates the results of our multivariate analyses. The orientation of the gratings\ncould be decoded significantly above chance during the stimulus phase in all ROIs (active\nconditions: EVC 71.2%, V1 66.7%, V2 70.7%, V3 65.9%; passive conditions: EVC 74.5%,\nV1 70.8%, V2 70.8%, V3 67.8%; all p < 0.001), as was to be expected based on previous\nstudies (Kamitani & T ong, 2005). Importantly, the orientation of upcoming, not yet\npresented stimuli in the preparatory phase could also be decoded significantly above\nchance, again in all visual areas (active conditions: EVC 54.5%, V1 54.4%, V2 54.1%, V3\n52.0%; passive conditions: EVC 56.2%, V1 56.3%, V2 56.3%, V3 54.8%, V3 active\nconditions p = 0.025; all other p < 0.001). There were no significant differences between\nactive and passive conditions in any visual area, neither during the preparatory nor the\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nstimulus phase (PrepPhase: all p > 0.195; StimPhase: EVC t(25) = -1.723, p = 0.097, V1\nt(25) = -1.840, p = 0.078, V2 t(25) = -0.022, p = 0.983, V3 t(25) = -1.247, p = 0.224).\nFig. 6. Results from multivariate analysis. Decoding accuracy is plotted for each single participant. Error\nbars represent the 95% confidence interval. Chance level is indicated by the dotted line at 50%. * = p < 0.05,\n*** = p < 0.001.\nIn order to test for similarities between active and passive conditions, we performed cross-\nclassifications (Fig. 7). Training the classifier on active trials and testing on passive trials\n(and vice versa) revealed very similar results, showing significant above-chance decoding\nin both the preparatory and stimulus phase, indicating the classifier could generalise\nacross conditions (Training on active and testing on passive trials, PrepPhase: 55.4%;\nStimPhase: 73.4%. Training on passive and testing on active trials, PrepPhase: 54.1%,\nStimPhase: 71.8%; all p < 0.001). In contrast, training on the preparatory phase and\ntesting on the stimulus phase (and vice versa) showed chance-level decoding for both\nactive and passive conditions, suggesting that the underlying patterns differed between\nthese time points (Training on the PrepPhase and testing on the StimPhase, active\nconditions: 48.9%, p = 0.869, passive conditions: 48.8%, p = 0.892. Training on the\nStimPhase and testing on the PrepPhase, active conditions: 49.1%, p = 0.941, passive\nconditions: 49.8%, 0.620). Results were comparable across all ROIs (see Supplementary\nMaterial).\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nFig. 7. Results from multivariate analysis, cross-classification. Decoding accuracy is plotted for each\nsingle participant. The figure on the left shows results for training on active and testing on passive conditions\n(train act – test pas) and vice versa (train pas – test act). The figure on the right shows results for training on\nthe preparatory phase and testing on the stimulus phase (train PrepPhase – test StimPhase) and vice versa\n(train StimPhase – test PrepPhase). Error bars represent the 95% confidence interval. Chance level is\nindicated by the dotted line at 50%. n.s. = not significant, *** = p < 0.001.\nT ab.2 Preparatory\nactivity\nPreparatory\ndecoding\nStimulus activity Stimulus decoding\nForward Model A- < P A- < P\nSharpening A- < P A+ > P\nPre-activation A+ > P A+ > P A- < P A- < P\nOpposing process A = P A+ = P+ A = P A+ = P+\nResults A =< P+ A+ = P+ A+ > P A+ = P+\nTable 2. Results compared with the hypotheses from each theory. The above chance decoding during the\npreparatory and stimulus phases, with no differences between active and passive conditions, match the\npredictions made by the opposing process theory. The univariate results during the preparatory phase partly\nmatch the opposing process predictions. Only the enhancement during the stimulus phase does not match\nany prediction.\nDiscussion\nThis study investigated how action-based predictions modulate activity in early visual\ncortex. Overall, our results do not support the cancellation, sharpening, and pre-activation\naccounts. Instead, our results mostly align with the opposing process theory. In line with\nour predictions for this theory, we show that both action-based and general sensory\npredictions induce similar stimulus-specific representations prior to stimulus presentation\nas evidenced by significant decoding and cross-decoding. During stimulus presentation we\nalso found no differences in decoding accuracy between conditions, and cross-\nclassification between conditions was possible. However, we observed enhanced – rather\nthan suppressed – BOLD responses for self-generated stimuli, a finding inconsistent with\nall the aforementioned theories. Altogether, these results suggest that 1) action-based and\ngeneral sensory predictions are based on similar mechanisms, and that 2) action\nexecution can boost the amplitude, but not precision, of neural responses.\nOur finding of enhanced rather than suppressed responses to self-generated stimuli\nchallenges the forward model, also known as the cancellation account, which posits that\nefference-copy-based predictions suppress sensory consequences of actions. Although\nthis theory has strong behavioral and neural support – showing reduced perceived\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nintensity and neural responses for self-generated stimuli (Arikan et al., 2019; Bäß et al.,\n2008; Bays et al., 2005; S. Blakemore et al., 1999; S.-J. Blakemore et al., 1998; Cardoso-\nleite et al., 2010; Kilteni et al., 2020; Kilteni & Henrik Ehrsson, 2020; Lubinus et al., 2022;\nMartikainen et al., 2005; Straube et al., 2017; Voudouris & Fiehler, 2022; Weiss et al.,\n2011), we found no behavioral differences in orientation performance and an increased\nBOLD response for self-generated stimuli. Our findings align with other studies reporting\nneural enhancement (Csifcsák et al., n.d.; Krala et al., 2019; Reznik et al., 2014, 2015). It\nis still unclear why action sometimes suppresses and sometimes enhances neural activity.\nIt is possible that the effect of action depends on task demands. In most studies showing\nsuppression, participants had either no task at all, or had to judge the intensity of the\nstimuli. In the studies showing enhanced neural responses, participants had to closely\nmonitor their actions, as they either had to play tones in the correct order or time (Reznik\net al., 2014, 2015), or had to reproduce a specific distance traveled (Krala et al., 2019).\nAlthough our study did not require close action monitoring, participants performed a\ndemanding discrimination task. Having control over when the stimuli would appear in the\nactive condition may have increased attention and boosted neural responses in visual\ncortex. Variability across sensory modalities may also be relevant: suppression is more\nconsistent in tactile and auditory domains, whereas visual studies report mixed effects. For\nexample, while (Cardoso-leite et al., 2010) found reduced sensitivity for self-generated\nvisual stimuli, (Schwarz et al., 2017) failed to replicate this result. Other behavioral studies\nshow either enhancement or suppression depending on task timing (Desantis et al., 2014;\nY on& Press, 2017). On the neural level, studies report reduced responses in visual areas\n(Benazet et al., 2016; Gentsch & Schütz-Bosbach, 2011; Lubinus et al., 2022; Ody, Straube, et\nal., 2023; Roussel et al., 2014; Straube et al., 2017), enhanced responses (Hughes & Waszak,\n2011; Mifsud et al., 2016; Wen et al., 2018), or mixed effects depending on participants or\nERP components (Buaron et al., 2020; Csifcsák et al., n.d.). Our study cannot determine the\ncause of the enhanced BOLD response in visual cortex for actively generated stimuli and\nthe lack of behavioural differences. Future research should explore different tasks to test\nhow task demands, sensory modalities, and stimulus predictability interact to drive\nsuppression or enhancement of self-generated stimuli.\nOur study also found no support for the pre-activation account, which proposes that\nlearned action effects pre-activate sensory regions, resulting in early activation followed by\nperceptual suppression (Roussel et al., 2013). Several behavioural studies support this\nidea, showing early facilitation and later suppression of perception of self-generated stimuli\n(Desantis et al., 2014; Roussel et al., 2013; Y on& Press, 2017). Furthermore, it has been\nreported that active movements lead to an earlier BOLD response than passive\nmovements (Kavroulakis et al., 2022). However, our univariate GLM could not find any\nsignificant differences between active and passive conditions in the preparatory phase. A\ndeconvolution GLM revealed divergence just before stimulus onset, but with stronger\nsustained activity in the passive condition. This may either suggest a pre-activation in the\npassive condition, or a stronger initial dip in the active condition. If a stronger initial dip at\nthe end of the prepatory phase is taken as evidence for more activity then there would be\nsome indirect support to the pre-activation model. Multivoxel pattern analysis showed that\nthe orientation of predicted upcoming gratings could be decoded from preparatory visual\ncortex activity. This finding aligns with previous research showing that predictions outside\nthe context of action induce stimulus representations prior to their occurrence (Kok et al.,\n2017) or during omission of an expected stimulus (Kok et al., 2014). However, decoding\nperformance was similar across conditions, and significant cross-classification suggests\nshared neural patterns. While EEG studies have reported stronger pre-activation for active\nconditions a few hundred milliseconds prior to action execution (Ody, Kircher, et al., 2023;\nReznik et al., 2018), these short-lived effects may have been undetectable with fMRI.\nOverall, our results suggest that both action-based and general sensory predictions lead to\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nsimilar pre-activation in early visual regions. This challenges the idea that action uniquely\nenhances pre-activation, highlighting the need for further investigation into the interplay\nbetween action, prediction, and neural dynamics.\nThe sharpening account was also not supported by our results. The concept of\npredictions leading to sharpening originates from the general perception literature. Kok et\nal. (2012) demonstrated that expected stimuli result in a reduced neural response yet\nimproved decoding performance. This finding suggests that expectations sharpen the\nneural response, an effect that may appear as suppression when examined using\nunivariate measures. (Y onet al., 2018) extended this analysis to self-generated stimuli and\nobserved similar results: expected sensory action outcomes led to reduced activity but\nimproved decoding compared to unexpected outcomes. These findings were taken as\nevidence that action does not suppress neural activity, but instead sharpens neural\nresponses. However, we found no difference in decoding accuracy between active and\npassive conditions. Unlike Y on et al., who compared expected and unexpected action\noutcomes (all self-generated), we compared self-generated and externally generated\nstimuli, both of which were expected and temporally predictable. The absence of\ndifferences in decoding performance between conditions suggests that action-based and\ngeneral predictions may operate similarly under predictable conditions. Since our study did\nnot include unexpected events, we cannot confirm whether predictions lead to a\nsharpening of neural responses for expected compared to unexpected stimuli. However,\nour results indicate that action-based predictions do not sharpen neural responses more\nthan general sensory predictions do.\nOverall, our results largely align with the opposing process theory (Press et al.,\n2019). We observed pre-activation for both action-based and general predictions, and no\ndifference in decoding accuracy during stimulus presentation – consistent with the theory’s\nprediction for expected stimuli. However, we did find enhanced BOLD responses for self-\ngenerated stimuli, which the theory does not fully explain in this context. One possibility is\nthat nonspecific factors like increased attention or arousal associated with action boosted\nneural responses. Another explanation might be linked to motor-induced responses in the\nvisual cortex as observed for example in mice (Fiser et al., 2016; Muzzu & Saleem 2021;\nVasilevskaya et al., 2023; for an overview see Schneider, 2020). Our current design\ncannot rule out the influence of these additional processes. Future studies should\nmanipulate the predictability of stimuli as well as attention to further disentangle these\neffects.\nConclusion\nAltogether, we have shown that action-based and general sensory predictions pre-activate\nneurons in visual cortex by inducing a representation of the predicted upcoming stimulus.\nThere representations were similar for self-generated and externally generated stimuli,\nsuggesting similar mechanisms. Furthermore, we have shown that action can boost the\namplitude, but not precision, of neural responses to self-generated stimuli. These results\nmostly support the opposing process theory (Press et al., 2019).\nAcknowledgements\nBvK received support from the Deutsche Forschungsgemeinschaft (DFG, grant KE\n2016/2-1 and SFB/TRR 135, grant number 222641018, project A10). LFM has received\nfunding for this project from the Biotechnology and Biological Science research Council\n(BBSRC) BB/V010956/1 ('Layer-specific cortical feedback dynamics'). We thank Francis\nCrabbe and Veronika Penkova for their support in the fMRI data collection.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nMethods\nParticipants\nA total of 28 healthy, right-handed participants were invited to take part in the study, which\nwas comprised of three sessions: one behavioural experiment, one fMRI experiment, and\none fMRI session with retinotopic mapping and functional localisers needed to create\nindividual regions of interest. All 28 participants completed the behavioural experiment (8\nmale, 20 female, mean age: 23.6 years, SD = 4.7). Furthermore, 26 participants\nadditionally took part in the fMRI experiment, and 25 completed all three sessions. For the\nparticipant that took part in the fMRI experiment, but not the localiser session, we were\nable to generate regions of interest based on the main experiment. The final sample for\nthe fMRI study thus included 26 participants (7 male, 19 female, mean age: 23.5 years, SD\n= 4.5). For three participants, one run each had to be excluded; one due to excessive\nhead motion, one due to technical error causing button presses to not be recorded, and\none due to the participant pressing the button at incorrect times in the active condition.\nStimuli\nThe stimuli were sinusoidal gratings with 1.5 cycles per degree (cpd), presented at 50%\ncontrast. They were presented in an annulus (outer diameter: 8° of visual angle, inner\ndiameter: 0.5°) surrounding a black presentation dot of 0.3°, on a midgrey background.\nAuditory cues were pure tones of either 400 Hz or 1000 Hz, presented through MRI-\ncompatible headphones. Stimuli were presented with PsychoPy 3.2.4 on a Windows 10\nPC. The monitor had a refresh rate of 60 Hz.\nExperimental design\nThe behavioural and fMRI experiment used the same experimental design, with some\nsmall differences in timing. Each trial started with an auditory cue, presented for 300 ms,\nthat predicted the overall orientation of the upcoming gratings. The mapping between the\ncue and the grating orientation was counterbalanced across participants, and trained in the\nbehavioural experiment prior to the fMRI experiment. After the auditory cue, there was a\npreparation phase, during which only the fixation dot was present on the screen. This\nphase lasted three seconds in the behavioural experiment and eight seconds in the fMRI\nexperiment. When the preparatory phase had passed, the fixation dot turned red for 300\nms. In active conditions, this served as a cue for the participant to press the button with\ntheir right index or middle finger, after which the gratings were presented immediately as a\nresult of the button press. Participants were given a maximum of 3 s to press the button. In\npassive conditions, the fixation dot turning red signalled to participants that the gratings\nwould be presented soon. The interval between fixation cue and presentation of gratings in\nthe passive condition was the average time the participant took to press the button in\nactive conditions in the training of the behavioural experiment. The two gratings were\npresented for 500 ms each, separated by an interstimulus interval of 500 ms. The\norientation of the first grating was always +45° or -45° from vertical, depending on which\nauditory cue was presented at the start of the trial, and the second grating deviated slightly\nfrom this orientation by one of five angles (± 0.5° - 6.5°). After the offset of the second\ngrating, participants were asked to judge whether the second grating was oriented\nclockwise or counterclockwise with respect to the first grating. A green fixation dot\nindicated the interval during which participants could respond (behavioural: until they\nanswered, fMRI: max. 2 s). They indicated their answer using MRI-compatible buttons\n(clockwise = left index finger, counterclockwise = left middle finger). In the behavioural\nexperiment, the next trial started after an intertrial interval of 2.5 seconds. In the fMRI\nexperiment, the next trial started after a jittered intertrial interval, with a minimum mean\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nduration of 9 s (range: 7-11) to which remaining time from the button press intervals were\nadded.\nProcedure\nParticipants were first invited to the behavioural experiment. Here, they first trained the\nmapping between auditory cue, button, and grating orientation, and were then familiarised\nwith the task. They then performed a training session with the main experimental paradigm,\nin which only the largest angle difference (6.5°) was used, and feedback on their\nperformance was given. If performance was below 75%, this training session was\nrepeated. After training, all participants reached at least 75% accuracy and were able to\nproceed with the main experiment. The fMRI experiment took part on another day. At the\nstart of this session, participants performed a brief training session outside the scanner to\nrefresh their memory. They then proceeded with the fMRI experiment inside the scanner.\nOn a third day, retinotopic mapping and functional localiser data were collected.\nMRI acquisition\nFunctional MRI data were acquired using a 3T TIM Trio scanner (Siemens, Erlangen,\nGermany), using a 32-channel head coil. A multiband sequence was used to collect whole-\nbrain functional data with the following settings: 45 slices interleaved; field of view [FoV] =\n222 mm; repetition time [TR] = 1000 ms; echo time [TE] = 36 ms; flip angle = 60°; PAT\nmode = GRAPPA, multiband acceleration factor = 3, slice thickness = 3.0 mm, distance\nfactor = 10%, and voxel size = 3 x 3 x 3 mm. A total of 995 volumes were collected per run\nin the main experiment, 800 volumes for the polar angle retinotopic mapping, and 196\nvolumes for the functional localiser. A T1-weighted anatomical scan (MPRAGE) was\ncollected at the end of each scanning session.\nMRI analysis\nfMRI data were analysed using BrainVoyager 22.0 for Linux and custom Matlab scripts in\nMatlab R2019b. Data preprocessing included slice time correction, motion correction, high-\npass temporal filtering, and coregistration to each participant‘s anatomical image. Both the\nanatomical and functional data were then normalised to MNI space. For the univariate\nanalysis, functional data were additionally smoothed with 8mm. For the multivariate\nanalysis, the unsmoothed data were used. A GLM was created with predictors modelling\nthe two different time points of interest for each condition (active/passive and\nleftward/rightward gratings). The preparatory phase was modeled from the onset of the\nauditory cue to the end of the preparatory phase, and the stimulus phase from the onset of\nthe first stimulus until the offset of the second stimulus. An additional predictor was\nincluded to model the time during which participants could answer in response to the\norientation discrimination task, from the onset until the offset of the green fixation dot. For\nthe univariate analysis, each predictor contained all trials of that condition, leading to a\ntotal of 9 predictors per run. For the multivariate analysis, each trial was modelled with a\nseparate predictor, leading to a total of 81 predictors per run.\nVOI definition\nVisual areas V1, V2, and V3 were defined using standard retinotopic mapping procedures\nbased on the polar angle sequence. An additional region called „early visual cortex“ (EVC)\nwas created by combining V1/2/3. Within these regions, voxels responsive to our stimuli\nwere selected using our functional localiser (t-contrast stimulus > fixation, p < 0.01\nuncorrected).\nUnivariate analysis\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nA VOI GLM was run on our regions of interest using the predictors described in MRI\nanalysis. Enhancement for active trials in the preparatory phase was tested with the\ncontrast (Active prep > Passive prep). We tested for suppression for active trials in the\nstimulus phase with the contrast (Passive stim > Active stim). An exploratory whole-brain\nunivariate analysis was performed in addition using the same contrasts (see\nSupplementary Material).\nMVPA\nMVPA was performed with custom Matlab scripts, implementing the LibSVM software\n(http://www. csie.ntu.edu.tw/wcjlin/libsv). A linear support vector machine was trained to\ndiscriminate leftward from rightward gratings using the beta images resulting from the GLM.\nClassifier performance was tested using a leave-one-run-out cross-validation approach,\nand permutation testing was performed to determine statistical significance. MVPA was\nperformed separately for active and passive conditions, as well as preparatory and\nstimulus phases. Paired sample t-tests were performed to determine differences in\ndecoding accuracy between active and passive conditions, separately for the preparatory\nand stimulus phases. In addition, cross-classification analyses were performed. First, the\nclassifier was trained on active trials and tested on passive trials, and vice versa. Second,\nthe classifier was training on the preparatory phase and tested on the stimulus phase, and\nvice versa.\nEye tracking analysis\nEye movements were recorded with an MR-compatible EyeLink 1000 eyetracker system\n(SR Research, Osgoode, ON, Canada). We analysed the gaze coordinates during the\npreparatory and stimulus phase for active and passive conditions separately. The data\nwere first smoothed with a smoothing window of 50 ms after which they were detrended\nand mean-corrected to remove linear drifts and correct for calibration inaccuracies. We\nthen defined a fixation region-of-interest by a circle around the center of the screen with a\nradius of 1.5° and calculated the percentage of samples in which participants looked\noutside of this fixation area. We used paired-sample t-tests to determine whether the\nnumber of samples outside of the fixation area differed between active and passive\nconditions.\nReferences\nArikan, B. E., van Kemenade, B. M. V .,Podranski, K., Steinsträter, O., Straube, B., & Kircher, T.\n(2019). Perceiving your hand moving: BOLD suppression in sensory cortices and the role of\nthe cerebellum in the detection of feedback delays. Journal of Vision, 19(14), 1–22.\nhttps://doi.org/10.1167/19.14.4\nBäß, P .,Jacobsen, T., & Schröger, E. (2008). Suppression of the auditory N1 event-related potential\ncomponent with unpredictable self-initiated tones: Evidence for internal forward models\nwith dynamic stimulation. International Journal of Psychophysiology, 70(2), 137–143.\nhttps://doi.org/10.1016/j.ijpsycho.2008.06.005\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nBays, P .M., Wolpert, D. M., & Flanagan, J. R. (2005). Perception of the consequences of self-\naction is temporally tuned and event driven. Current Biology, 15(12), 1125–1128.\nhttps://doi.org/10.1016/j.cub.2005.05.023\nBenazet, M., Thénault, F., Whittingstall, K., & Bernier, P .-M. (2016). Attenuation of visual\nreafferent signals in the parietal cortex during voluntary movement. Journal of\nNeurophysiology, 116(4), 1831–1839. https://doi.org/10.1152/jn.00231.2016\nBlakemore, S., & Frith, C. (2003). Self-awareness and action. Current Opinion in Neurobiology, 13,\n219–224. https://doi.org/10.1016/S0959-4388(03)00043-6\nBlakemore, S., Frith, C. D., & Wolpert, D. M. (1999). Spatio-Temporal Prediction Modulates the\nPerception of Self-Produced Stimuli. Journal of Cognitive Neuroscience, 11(5), 551–559.\nBlakemore, S.-J., Wolpert, D. M., & Frith, C. D. (1998). Central cancellation of self-produced tickle\nsensation. Nature Neuroscience, 1(7), 635–640. https://doi.org/10.1038/2870\nBuaron, B., Reznik, D., Gilron, R., & Mukamel, R. (2020). V oluntaryactions modulate perception\nand neural representation of action-consequences in a hand-dependent manner. Cerebral\nCortex, 30(12), 6097–6107. https://doi.org/10.1093/cercor/bhaa156\nCardoso-leite, P .,Mamassian, P .,Schütz-Bosbach, S., & Waszak, F. (2010). A New Look at Sensory\nAttenuation: Action-Effect Anticipation Affects Sensitivity , Not Response Bias.\nPsychological Science, 21(12), 1740–1745. https://doi.org/10.1177/0956797610389187\nCsifcsák, G., Balla, V .R., Dalos, V .D., Kilencz, T., Biró, E. M., Urbán, G., & Szalóki, S. (2019.).\nAction‐associated modulation of visual event‐related potentials evoked by abstract and\necological stimuli. Psychophysiology, 56(e13289), 1:15.\nDesantis, A., Roussel, C., & Waszak, F. (2014). The temporal dynamics of the perceptual\nconsequences of action-effect prediction. Cognition, 132(3), 243–250.\nhttps://doi.org/10.1016/j.cognition.2014.04.010\nFiser, A., Mahringer, D., Oyibo, H.K., Petersen, A.V .,Leinweber, M., & Keller, G. Experience-\ndependent spatial expectations in mouse visual cortex. Nature Neuroscience 19, 1658–1664\n(2016). https://doi.org/10.1038/nn.4385\nGentsch, A., & Schütz-Bosbach, S. (2011). I Did It: Unconscious Expectation of Sensory\nConsequences Modulates the Experience of Self-agency and Its Functional Signature.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nJournal of Cognitive Neuroscience, 23(12), 3817–3828.\nhttps://doi.org/10.1162/jocn_a_00012\nHughes, G., & Waszak, F. (2011). ERP correlates of action effect prediction and visual sensory\nattenuation in voluntary action. NeuroImage, 56(3), 1632–1640.\nhttps://doi.org/10.1016/j.neuroimage.2011.02.057\nKamitani, Y .,& Tong, F. (2005). Decoding the visual and subjective contents of the human brain.\nNature Neuroscience, 8(5), 679–685. https://doi.org/10.1038/nn1444\nKavroulakis, E., V anKemenade, B. M., Arikan, B. E., Kircher, T., & Straube, B. (2022). The effect\nof self‐generated versus externally generated actions on timing, duration, and amplitude of\nblood oxygen level dependent response for visual feedback processing. Human Brain\nMapping, 43(16), 4954–4969. https://doi.org/10.1002/hbm.26053\nKilteni, K., Engeler, P .,& Ehrsson, H. H. (2020). Efference Copy Is Necessary for the Attenuation\nof Self-Generated Touch. iScience, 23(2), 100843.\nhttps://doi.org/10.1016/j.isci.2020.100843\nKilteni, K., & Henrik Ehrsson, H. (2020). Functional connectivity between the cerebellum and\nsomatosensory areas implements the attenuation of self-generated touch. Journal of\nNeuroscience, 40(4), 894–906. https://doi.org/10.1523/JNEUROSCI.1732-19.2019\nKok, P .,Failing, M. F., & de Lange, F. P .(2014). Prior expectations evoke stimulus templates in the\nprimary visual cortex. Journal of Cognitive Neuroscience, 26(7), 1546–1554.\nhttps://doi.org/10.1162/jocn\nKok, P ., Jehee, J. F. M., & de Lange, F. P . (2012). Less Is More: Expectation Sharpens\nRepresentations in the Primary Visual Cortex. Neuron, 75(2), 265–270.\nhttps://doi.org/10.1016/j.neuron.2012.04.034\nKok, P ., Mostert, P ., & De Lange, F. P . (2017). Prior expectations induce prestimulus sensory\ntemplates. Proceedings of the National Academy of Sciences of the United States of America,\n114(39), 10473–10478. https://doi.org/10.1073/pnas.1705652114\nKrala, M., van Kemenade, B., Straube, B., Kircher, T., & Bremmer, F. (2019). Predictive coding in a\nmultisensory path integration task: An fMRI study. Journal of Vision, 19(11), 13.\nhttps://doi.org/10.1167/19.11.13\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nLubinus, C., Einhäuser, W., Schiller, F., Kircher, T., Straube, B., & V anKemenade, B. M. (2022).\nAction-based predictions affect visual perception, neural processing, and pupil size,\nregardless of temporal predictability. NeuroImage, 263, 119601.\nhttps://doi.org/10.1016/j.neuroimage.2022.119601\nMartikainen, M. H., Kaneko, K., & Hari, R. (2005). Suppressed Responses to Self-triggered Sounds\nin the Human Auditory Cortex. Cerebral Cortex, 15, 299–302.\nhttps://doi.org/10.1093/cercor/bhh131\nMifsud, N. G., Oestreich, L. K. L., Jack, B. N., Ford, J. M., Roach, B. J., Mathalon, D. H., &\nWhitford, T. J. (2016). Self-initiated actions result in suppressed auditory but amplified\nvisual evoked components in healthy participants. Psychophysiology, 53, 723–732.\nhttps://doi.org/10.1111/psyp.12605\nMuzzu, T. & Saleem, A.B. (2021). Feature selectivity can explain mismatch signals in mouse visual\ncortex. Cell Reports, 37, 109772.\nOdy, E., Kircher, T., Straube, B., & He, Y . (2023). Pre‐movement event‐related potentials and\nmultivariate pattern of EEG encode action outcome prediction. Human Brain Mapping,\n44(17), 6198–6213. https://doi.org/10.1002/hbm.26506\nOdy, E., Straube, B., He, Y ., & Kircher, T. (2023). Perception of self‐generated and\nexternally‐generated visual stimuli: Evidence from EEG and behavior. Psychophysiology,\n60(8), e14295. https://doi.org/10.1111/psyp.14295\nPress, C., Kok, P ., & Y on, D. (2019). The Perceptual Prediction Paradox. Trends in Cognitive\nSciences, 24(1), 13–24. https://doi.org/10.1016/j.tics.2019.11.003\nReznik, D., Henkin, Y .,Schadel, N., & Mukamel, R. (2014). Lateralized enhancement of auditory\ncortex activity and increased sensitivity to self-generated sounds. Nature Communications,\n5(May), 1–11. https://doi.org/10.1038/ncomms5059\nReznik, D., Ossmy, O., & Mukamel, X. R. (2015). Enhanced Auditory Evoked Activity to Self-\nGenerated Sounds Is Mediated by Primary and Supplementary Motor Cortices. Journal of\nNeuroscience, 35(5), 2173–2180. https://doi.org/10.1523/JNEUROSCI.3723-14.2015\nReznik, D., Simon, S., & Mukamel, R. (2018). Predicted sensory consequences of voluntary actions\nmodulate amplitude of preceding readiness potentials. Neuropsychologia, 119(January),\n302–307. https://doi.org/10.1016/j.neuropsychologia.2018.08.028\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nRoussel, C., Hughes, G., & Waszak, F. (2013). A preactivation account of sensory attenuation.\nNeuropsychologia, 51(5), 922–929. https://doi.org/10.1016/j.neuropsychologia.2013.02.005\nRoussel, C., Hughes, G., & Waszak, F. (2014). Action prediction modulates both neurophysiological\nand psychophysical indices of sensory attenuation. Frontiers in Human Neuroscience,\n8(February), 1–6. https://doi.org/10.3389/fnhum.2014.00115\nSchneider, D.M. (2020). Reflections of action in sensory cortex. Current Opinion in Neurobiology,\n64:53–59.\nSchwarz, K. A., Pfister, R., Kluge, M., Weller, L., & Kunde, W. (2017). Do We See It or Not?\nSensory Attenuation in the Visual Domain. Journal of Experimental Psychology: General,\n0(999). https://doi.org/10.1037/xge0000353\nShergill, S., White, T. P ., Joyce, D. W., Bays, P . M., Wolpert, D. M., & Frith, C. D. (2013).\nModulation of somatosensory processing by action. NeuroImage, 70, 356–362.\nhttps://doi.org/10.1016/j.neuroimage.2012.12.043\nSperry, R. W. (1950). Neural basis of the spontaneous optokinetic response produced by visual\ninversion. Journal of Comparative and Physiological Psychology, 482–489.\nStraube, B., van Kemenade, B. M., Arikan, B. E., Fiehler, K., Leube, D. T., Harris, L. R., & Kircher,\nT. (2017). Predicting the Multisensory Consequences of One’s Own Action: BOLD\nSuppression in Auditory and Visual Cortices. PLoS ONE, 12(1), 1–25.\nhttps://doi.org/10.1371/journal.pone.0169131\nSummerfield, C., Trittschuh, E. H., Monti, J. M., Mesulam, M. M., & Egner, T. (2008). Neural\nrepetition suppression reflects fulfilled perceptual expectations. Nature Neuroscience, 11(9),\n1004–1006. https://doi.org/10.1038/nn.2163\nvon Holst, E., & Mittelstaedt, H. (1950). The principle of reafference: Interactions between the\ncentral nervous system and the peripheral organs. PC Dodwell (Ed. and Trans.), Perceptual\nProcessing: Stimulus Equivalence and Pattern Recognition, 41–72.\nV asilevskaya,A., Widmer, F.C., Keller, G., Jordan, R. (2023). Locomotion-induced gain of visual\nresponses cannot explain visuomotor mismatch responses in layer 2/3 of primary visual\ncortex. Cell Reports, 42(3), 112096. https://doi.org/10.1016/j.celrep.2023.112096\nV oudouris,D., & Fiehler, K. (2022). The role of grasping demands on tactile suppression. Human\nMovement Science, 83, 102957. https://doi.org/10.1016/j.humov.2022.102957\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nWeiss, C., Herwig, A., & Schütz-Bosbach, S. (2011). The self in action effects: Selective attenuation\nof self-generated sounds. Cognition, 121, 207–218.\nhttps://doi.org/10.1016/j.cognition.2011.06.011\nWen, W., Brann, E., Di Costa, S., & Haggard, P .(2018). Enhanced perceptual processing of self-\ngenerated motion: Evidence from steady-state visual evoked potentials. NeuroImage,\n175(April), 438–448. https://doi.org/10.1016/j.neuroimage.2018.04.019\nWolpert, D. M., & Flanagan, J. R. (2001). Motor prediction. Current Biology, 11(18), R729–R732.\nhttps://doi.org/10.1016/S0960-9822(01)00432-8\nWolpert, D. M., Ghahramani, Z., & Jordan, M. I. (1995). An internal model for sensorimotor\nintegration. Science, 269, 1880–1882. https://doi.org/10.1126/science.7569931\nY on,D., Gilbert, S. J., Lange, F. P .D., & Press, C. (2018). Action sharpens sensory representations\nof expected outcomes. Nature Communications, 9(4288), 1–8.\nhttps://doi.org/10.1038/s41467-018-06752-7\nY on,D., & Press, C. (2017). Predicted Action Consequences Are Perceptually Facilitated Before\nCancellation. Journal of Experimental Psychology. Human Perception and Performance,\nMarch. https://doi.org/10.1037/xhp0000385\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nSupplementary material\nResults\nUnivariate analyses - ROI\nUnivariate  results  for  each  individual  visual  region  depicted  in  Fig.  1  resembled  the  \nfindings found in our early visual cortex ROI reported in the main manuscript. There were  \nno significant differences between active and passive conditions during the preparatory  \nphase in any of the visual areas (V1: t(25) = -1.634, p = 0.115; V2: t(25) = -1.425, p =  \n0.167; V3: t(25) = -1.034, p = 0.311). During the stimulus phase, there was significantly  \nmore activity in the active compared to the passive condition in all visual areas (V1: t(25) =  \n-4.445, p = 0.000157; V2: t(25) = -4.439, p = 0.000159; V3: t(25) = -4.355, p = 0.000198).\nFig. 1. Univariate ROI analysis results.  We performed paired t-tests to test for differences between active  \nand passive conditions. Boxplots show the extracted beta values from V1, V2, and V3. n.s. = not significant,  \n*** = p < 0.001.\nUnivariate analyses – Whole-brain\nWhole-brain univariate analysis showed significantly more activity for active compared to  \npassive conditions during the preparatory phase in supplementary motor area, left primary  \nmotor cortex, and right cerebellum (Fig. 2A). These areas are typically involved in motor  \npreparation and therefore expected, as participants were preparing to press a button with  \ntheir right hand in active conditions during this phase. Aligning with our ROI analysis, no  \nsignificant differences were found in visual cortex. During the stimulus phase, we observed \nsignificantly  more  activity  for  active  conditions  in  a  large  network  with  peaks  in  the  \ncerebellum, supplementary motor area, primary motor and somatosensory cortex, middle  \nand inferior frontal gyrus, insula, and precuneus (Fig. 2B). As participants pressed a button \nin active conditions during this phase, it is no surprise that many areas related to motor  \nexecution were observed. Supporting our ROI analysis, active conditions also enhanced  \nactivity in visual areas.\nActive Passive Active Passive Active Passive Active Passive\n Active Passive Active Passive\n*********\nn.s. n.s.n.s.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nFig. 2. Univariate whole-brain analysis results.  Active conditions were contrasted with passive conditions  \nfor the preparatory phase (PrepPhase, panel A) and for the stimulus phase (StimPhase, panel B). FDR: false  \ndiscovery rate.\nMultivariate analyses – Cross-classification\nCross-classification results for each individual visual region (Fig. 3) resembled the findings  \nfound in our early visual cortex ROI reported in the main manuscript. Training the classifier  \non active trials and testing on passive trials (and vice versa) showed significant above-\nchance decoding in both the preparatory and stimulus phase, indicating the classifier could \ngeneralise across conditions (Training on active and testing on passive trials, PrepPhase:  \nV1 52.5%, p = 0.011, V2 54.2%, p = 0.002, V3 53.0%, p = 0.006; StimPhase: V1 67.5%, p  \n< 0.001, V2 69.7%, p < 0.001, V3 67.9%, p < 0.001. Training on passive and testing on  \nactive trials, PrepPhase: V1 54.4%, p < 0.001, V2 54.2%, p< 0.001, V3 53.0%, p = 0.005;  \nStimPhase: V1 66.6%, p < 0.001, V2 69.7%, p < 0.001, V3 64.8%, p < 0.001).\nFig. 3. Results from multivariate analysis, cross-classification, for areas V1, V2, and V3.  These are the \nresults for training on active and testing on passive conditions (train act – test pas) and vice versa (train pas  \n– test act). Decoding accuracy is plotted for each single participant. Error bars represent the 95% confidence  \ninterval. Chance level is indicated by the dotted line at 50%. * = p < 0.05, ** = p < 0.01, *** = p < 0.001.\nIn contrast, training on the preparatory phase and testing on the stimulus phase (and vice  \nversa) showed chance-leve l decoding for both active and passive conditions, suggesting  \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint \n\nthat the underlying patterns differed between these time points (Training on PrepPhase  \nand testing on StimPhase, active conditions: V1 47.5%, p = 0.998, V2 48.7%, p = 0.902,  \nV3 48.7%, p = 0.915; passive conditions: V1 48.6%, p = 0.910, V2 49.8%, p = 0.584, V3  \n48.4%, p = 0.938. Training on StimPhase and testing on PrepPhase, active conditions: V1  \n49.2%, p = 0.904, V2 49.6%, p = 0.778, V3 49.3%, p = 0.839; passive conditions: V1  \n51.1%, p = 0.03, V2 49.5%, p = 0.759, V3 49.3%, p = 0.871).\nFig. 4. Results from multivariate analysis, cross-classification, for areas V1, V2, and V3.  These are the \nresults for training on the preparatory phase and testing on the stimulus phase (train PrepPhase – test  \nStimPhase) and vice versa (train StimPhase – test PrepPhase). Decoding accuracy is plotted for each single  \nparticipant. Error bars represent the 95% confidence interval. Chance level is indicated by the dotted line at  \n50%. n.s. = not significant, * = p < 0.05.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted September 1, 2025. ; https://doi.org/10.1101/2025.08.28.672792doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}