The myth of TMS-induced ipsilateral enhancement in visual detection paradigms: A systematic review and meta-analysis of inhibitory parietal TMS studies in healthy participants

Spatial attention control involves specialized functions in both hemispheres of the brain, leading to hemispheric asymmetries. Neuropsychological models explain this lateralization mainly based on patient studies of hemi-neglect. Studies in healthy volunteers can mimic hemineglect using transcranial magnetic stimulation (TMS) by disrupting the left/right posterior parietal cortex (PPC) during visual detection tasks, enabling a comparison of hemispheric contributions to stimulus detection in the contra-versus ipsilateral hemifields. Kinsbourne ’ s opponent processor model and Heilman ’ s hemispatial model present contrasting hypotheses regarding the behavioral consequences of unilateral PPC disruption. A pivotal prediction in distinguishing between these models is the occurrence of ipsilateral enhancement. Our meta-analysis assessed inhibitory TMS effects on PPC during visual detection tasks across ten studies (1994 – 2022). PPC disruption caused contralateral impairment for bilateral stimuli, but no ipsilateral enhancement for unilateral or bilateral stimuli. These results are at odds with influential reports of ipsilateral enhancement after PPC disruption in healthy volunteers that have shaped the field of spatial attention research and should prompt a re-evaluation of current theoretical models of attention and their application to novel brain stimulation-based therapeutic interventions.


Introduction
Among the most common consequences of brain damage is a neuropsychological syndrome called spatial hemineglect that can be observed after lesions to frontal, parietal, or sub-cortical structures (Corbetta et al., 2005).Spatial neglect is characterized by impaired attentional processing within the space contralateral to the lesion's location.As a consequence, patients typically fail to attend, explore, and act upon one side of space which also leads to peculiar effects in the perceptual domain.The phenomenon of spatial neglect is more common and severe after right hemisphere damage than after left hemisphere damage, suggesting a functional asymmetry of the mechanisms underlying spatial attention (Mesulam, 1981).Based on this observation, two competing theories regarding the neural dynamics underlying spatial attention have emerged, namely Heilman's "hemispatial" theory (Heilman and Abell, 1980) and Kinsbourne's "opponent processor" theory (Kinsbourne, 1977), both accounting for this asymmetry but proposing very different mechanisms.See Fig. 1.
According to Heilman's "hemispatial" model, the right hemisphere contributes to attention allocation to both left and right hemifield, whereas the left hemisphere only contributes to attention allocation to the right hemifield.In this model, left parietal lesions can be compensated for by the right hemisphere but not vice versa.According to Kinsbourne's "opponent processor" model, both hemispheres induce attentional biases toward their respective contralateral visual hemifield, while at the same time actively inhibiting each other via interhemispheric suppression in order to maintain a system balance.In this model, the left hemisphere-induced bias towards the right hemifield is somewhat stronger and once disinhibited after right hemispheric lesion (loss of interhemispheric balance), this stronger functional spatial attention bias towards the right side causes left hemineglect.Taken together, the explanatory approaches of both theories are very different but the predicted behavioral consequences of left and right hemisphere damage are very similar making it difficult to dissociate them empirically on behavioral level alone.Taken together, the explanatory approaches of both theories are very different but the predicted behavioral consequences of left and right hemisphere damage are very similar making it difficult to dissociate them empirically on behavioral level alone.
In addition to neuropsychological evidence based on brain lesion studies that indicate hemispheric asymmetries in the behavioral functional relevance of left versus right parietal cortex for spatial attention, a more rigorous and controlled experimental investigation of these asymmetries in healthy volunteers using neuroscientific research tools is paramount.Although extensive neuroimaging work on human visual attention has addressed the role of the right and left hemisphere (e.g., (Corbetta and Shulman, 2002;Driver et al., 2004;Serences and Yantis, 2006)), standard neuroimaging studies struggle to assess differential hemispheric functional contributions to attention.
Transcranial magnetic stimulation (TMS), as a functional intervention, allows the temporary disruption of local neural activity in healthy individuals (Bien et al., 2012b;Cohen Kadosh et al., 2007;Duecker et al., 2013;Gurel et al., 2018;Jeurissen et al., 2014;Pascual-Leone, 2000;Sack et al., 2007).Several such TMS experiments have been conducted to investigate the behavioral consequences of disrupting either the left or right hemisphere in the context of visual detection and spatial attention paradigms.Many of these studies reported results seen as evidence in favor of Kinsbourne's "opponent processor" model (Dambeck et al., 2006;Hilgetag et al., 2001;Silvanto et al., 2009).As a prime and highly cited example, Hilgetag et al. (2001) found contralateral impairments and ipsilateral enhancements of target detection after inhibitory TMS over left and right parietal cortex.Especially the reported ipsilateral enhancement effects in visual detection following unilateral inhibitory TMS seem to be in accordance with predictions made on the basis of Kinsbourne's opponent processor model.According to this model, the inhibitory TMS protocol applied over, e.g., left parietal cortex, not only suppresses the left hemisphere causing contralateral, i.e. right hemifield, impairments in visual detection tasks, but at the same time also dis-inhibits the right hemisphere (because it is now released from inhibition from the suppressed left hemisphere) causing left hemifield enhancement in visual detection.This combination of contralateral impairment and ipsilateral enhancement after unilateral suppressive TMS is exactly what was found and reported in this study.Also in accordance with this interpretation, Dambeck et al. (2006) revealed that target detection is unaffected when suppressive TMS is applied simultaneously over left and right parietal cortex because a second contralateral lesion should restore interhemispheric balance with consequent recovery of the attention deficit.
Importantly, based on these findings, a consensus seems to have emerged in the scientific community that most TMS studies largely confirm the specific predictions made by the opponent processor model with unilateral TMS causing both, contralateral impairment but also ipsilateral enhancement of visual detection.Consequently, current clinical TMS protocols to treat hemineglect in stroke patients are based on this notion, using suppressive TMS protocols applied over the unaffected healthy hemisphere in stroke patents to enhance their ipsilateral, i.e. contralesional detection / attention deficits and thus to alleviate their hemineglect symptoms.This is in fact the now common and recommended clinical practice of using TMS in stroke rehabilitation (Lefaucheur et al., 2020).And while this clinical application of TMS in stroke rehabilitation is very promising (Valero-Cabré et al., 2020), the underlying mechanisms are by no means established and generalization across intact and lesioned brains do not necessarily hold.
Here, we present findings of a systematic classical and Bayesian meta-analysis where we aggregated all parietal inhibitory TMS studies in healthy volunteers targeting either the left hemisphere, right hemisphere, or both, to systematically assess to what extent this notion of TMS-induced ipsilateral enhancements in visual detection paradigms is really supported by the existing empirical evidence.We focus on TMS studies in the context of visual detection paradigms which allow to analyze the separate functional role of each hemisphere for hemispherespecific attention biases, gains, and costs towards the ipsilateral versus contralateral side of space, and thus to allow for a systematic assessment of detection performance in each hemifield in isolation (Duecker and Sack, 2015).Based on the here aggregated data, we aim to objectify the current empirical evidence base for a reliable TMS-induced i) contralateral impairment, ii) ipsilateral enhancement and iii) attentional selection in visual detection paradigms.

Methods
The current review was not registered but followed the PRISMA guidelines.
Fig. 1.Schematic representation of competing models of attention and their behavioral predictions.Please note that 'stimulation' always refers to the applications of TMS, whereas 'stimulus' always refers to the visual target presented in one or both hemifields. A. Heilman's hemispatial model predicts that only right PPC disruption causes a contralateral impairment.B. Kinsbourne's opponent processor model predicts that left and right PPC disruption both cause an imbalance between hemispheres causing contralateral impairment and ipsilateral enhancement.

Literature search and study selection
The literature search was conducted on PubMed, Elsevier, and Web of Science using the search codes (TMS OR "transcranial magnetic stimulation") AND (PPC OR "parietal") AND ("extinction" OR "visual detection" OR "visuospatial attention" OR "spatial attention" OR "spatial neglect").The search was restricted to journal articles written in English, between 1994 and February 2022.Two researchers searched articles fully independently according to the PRISMA guideline and inconsistencies in the search results were resolved in team discussions.

Inclusion and exclusion criteria
Search results were imported into Endnote and duplicates removed in the first round of screening.Then two researchers conducted the abstract and full-text review respectively.Inclusion criteria were: (1) healthy human participants (at least 5); (2) TMS targeting any region of the PPC (labeled as intraparietal sulcus or close to the MNI coordinates reported by (Fox et al., 2006)); (3) using the visual detection/extinction task; (4) comparison of active stimulation to baseline condition (no TMS or sham stimulation or control site stimulation (e.g., the vertex)); ( 5) inclusion of at least one detection performance measure (e.g.detection rate, error rate).

Data extraction and management
One researcher used a standardized data extraction form specifically designed for this review to collect data from the included studies.Extracted data included the following information: author; published year; a detailed description of the participants, their age, sex, handedness; TMS related parameters, frequency, offline or online TMS application, duration, TMS localization; detection performance change compared with baseline.For articles only reporting data in figures, numerical results were extracted from the figures using GetData Graph Digitizer 2.24.All of these extraction steps were double-checked by the senior researchers.

Statistical methods
Statistical analyses were done with SPSS (IBM), Stata, and R software (Metafor and Meta package).Changes in detection performance in participants were analyzed using paired or one-sample t-tests.For studies that reported the F value, the formula t = √F was used to estimate the t-test statistic from the one-way analysis of variance.For studies that reported the mean value and standard error of baseline and TMS conditions, but not the t-test statistic, we used the paired sample tstatistic and formulas(see below ( 1) and ( 2)) outlined in Morris and DeShon (Morris and DeShon, 2002) to derive the correlation between outcome measures.

ES
The former part of formula 1 is suitable for independent t-test, and the latter part is also suitable for paired t-test but should combine formula 2 to get the pooled standard deviation.
The effect sizes (cohen's d) were then calculated in these studies, including studies using repetitive TMS (rTMS), continuous theta burst stimulation (cTBS) and event-related TMS stimulation.For studies that reported t or F value, effectsize::t_to_d function was used to get the Cohen's d.For studies that reported data as mean value and standard error, the escalc function was used to obtain the (bias-corrected) standardized mean differences and corresponding sampling variances and transformed it to Cohen's d according to the 'Metafor 'package.Since few of these studies reported the change of detection performance (compared with baseline, baseline as 0 or 1), then Stata Software was used to get the corresponding t value.
After calculating the merged effect sizes, we evaluated their significance by checking whether the 95% confidence interval included 0 or not.Besides a homogeneity analysis was carried out, followed by Q Test measures for heterogeneity, and the Egger test and funnel plot to determine publication bias.In case of a significant publication bias, trim and fill methods were used.Lastly, we generated forest plots to visually display results and conducted a sensitivity analysis and subgroup analyses by 'Meta' package in R.

Multilevel meta-analysis
For studies that include multiple nested interventions in a single sample, dependence might be introduced.Statistical independence is one of the core assumptions of meta-analytic pooling (Harrer et al., 2021), and dependency between effect sizes (i.e., the effect sizes are correlated) could artificially reduce heterogeneity and lead to false-positive results.These dependencies that may exist in these nested designs can be handled using a multilevel meta-analysis.In multilevel meta-analysis, the variance in observed effect sizes is decomposed into sampling variance (level 1), between-study variance (level 2) and variance between groups of studies (level 3), and the moderating effect of characteristics of studies (at level 2) and groups of studies (at level 3) can be explored.This analysis was conducted using the metafor package in R.

Bayesian meta-analysis
In addition to the conventional meta-analysis outlined above, we also performed a Bayesian meta-analysis.Within the R statistical computing environment, we used the "brms" package (Bürkner, 2017) based on Stan software (Carpenter et al., 2017), to fit Bayesian multilevel models.The first step of Bayesian analysis was defining a prior distribution of standardized mean difference (SMD) as SMD~N(0,1) and heterogeneity(τ) as τ~HC(0,0.5),then set up the formula for the model and the MCMC algorithm run 4000 iterations to fit the model.Before evaluating the model fitting, convergence was assessed by posterior predictive checks and the R-hat values of the parameter estimates.Based on the obtained Bayesian model, we calculated the exact probabilities that the meta-analytic effect will be smaller/larger than a given effect size value by looking at the empirical cumulative distribution function (ECDF) of the posterior distribution for the pooled effect size.

Overview
The initial literature search resulted in 813 articles (duplicates removed) of which 144 were included in the full-text review (details in Fig. 2).Ultimately, a total of 10 different studies met the inclusion criteria.Stimulation parameters and participants' information are shown in Table1.
As illustrated in Fig. 2, from the third round screening to the coding stage, several studies were excluded.Five studies did not use a simple detection task (Battelli et al., 2009;Hung et al., 2005;Leitão et al., 2017;Mahayana et al., 2014;Muggleton et al., 2006) One study (Bien et al., T. Wang et al. 2012a) was excluded because the control condition in the study was considered inadequate because trials without TMS were interleaved with TMS trials (Duecker and Sack, 2013).Two studies did not report the required data and there was no way to retrieve them (Hilgetag et al., 2003;Pascual-Leone et al., 1994).Two high frequency offline TMS studies (i.e., excitatory instead of inhibitory TMS effects) were excluded, namely Dombrowe et al. (2015) and Jin, Hilgetag (2008).Eventually, ten studies with almost identical stimulation sites, experimental tasks, and conceptually matching TMS protocols remained that were included in the final meta-analysis.
Among the final 10 studies (Cazzoli et al., 2009;Dambeck et al., 2006;Gießing et al., 2020;Hilgetag et al., 2001;Koch et al., 2005;Koivisto et al., 2017;Müri et al., 2002;Petitet et al., 2015;Schauer et al., 2016;Vesia et al., 2015), some conducted right hemisphere stimulation as well as left hemisphere stimulation, so there were two datasets within one study (Dambeck et al., 2006;Hilgetag et al., 2001;Koivisto et al., 2017;Vesia et al., 2015).The authors of two studies (Schauer et al., 2016;Vesia et al., 2015) supplied raw data and additional details on request to enable inclusion in our analyses.For the study by Dambeck et al. (2006), the bilateral TMS stimulation was excluded as well as a longer ISI conditions because no other studies explored these parameters.For Koch et al. (2005), two control conditions and the paired-pulse TMS condition with different ISI were excluded.To keep some parameters as consistent as possible across all included studies, for the study of Schauer et al. (2016), only the stimulation site in posterior SPL was included as this was most similar to the other PPC targets in this meta-analysis.For the study by Vesia et al. (2015), the cTBS effect peaked after 20-25 min, so only blocks three to six were combined and included.

Meta-analysis of a contralateral TMS effect on a unilateral stimulus
To check the contralateral TMS effect on a unilateral stimulus, 11 datasets from previous studies were merged to yield a pooled effect size for left and right PPC stimulation.The test for heterogeneity was not significant (I2 = 14%; τ2 = 0.0233, p = 0.31), supporting the rationale for computing a fixed effect model.The fixed effect model showed a nonsignificant negative Cohen's d − 0.13, 95% CI ranged from − 0.35-0.1 (See Fig. 3A).The Egger test showed that there was no significant publication bias (p = 0.23).After omitting any single study, the effect was essentially the same.Thus, the current result was robust and not impacted by other variables.
The subgroup analysis showed that there was no significant difference between right PPC and left PPC stimulation (Q=0.14, df=1, p = 0.7).For right PPC stimulation, the fixed effect model showed a nonsignificant effect size (Cohen's d = − 0.09; 95% CI ranged from − 0.37 to 0.19), and the same was true for left PPC stimulation (Cohen's d = − 0.2; 95% CI ranged from − 0.58 to 0.18).Therefore, the result suggested TMS over RPPC/LPPC does not impaired detection performance in the contralateral hemifield when a unilateral stimulus was presented.
As mentioned above, given the potential dependencies caused by multiple datasets stemming from single studies, we considered such dependencies by integrating a third layer into the structure.We used a 3level model including the sampling variation for each ES (level 1), variation across ESs within a study (level 2), and variation across studies (level 3).Here, several datasets from four studies (Dambeck et al., 2006;Hilgetag et al., 2001;Koivisto et al., 2017;Vesia et al., 2015), which collected data from multiple sites, were added.The full model showed that the pooled effect size was − 0.21, 95% CI ranged from − 0.52-0.11.After checking the variance distribution of the full model, it was   observed that layer 1 to layer 3 accounted for 79%, 0%, and 21% variance, respectively.From the comparison between the full model and the leave-level 3-out model, no significant model fitting difference were found (lower AIC and BIC; p = 0.48).

Meta-analysis of an ipsilateral TMS effect on a unilateral stimulus
To check the ipsilateral TMS effect on a unilateral stimulus, 10 datasets were merged to yield a pooled effect size for left and right PPC stimulation.The test for heterogeneity was not significant (I2 = 43%; τ2 = 0.1141, p = 0.06), supporting the rationale for computing a fixed effect model.The fixed effect model showed a non-significant positive Cohen's d 0.14, 95% CI ranged from − 0.09-0.37(See Fig. 3B).The Egger test showed there was no significant (p = 0.1) publication bias, but the funnel plot (see Fig. 4A) suggested that there was some degree of publication bias.Using the trim and fill method, three virtual studies were added (Fig. 4B), the effect was non-significant (Cohen's d − 0.01, 95% CI ranged from − 0.36 to 0.33).After omitting any single study, the effect was still essentially the same.Thus, the current result revealed was robust and not impacted by other variables.
The subgroup analysis showed that there was no significant difference between the right PPC and left PPC stimulation (Q=0.18,df=1, p = 0.67).For the left PPC stimulation, the fixed effect model showed a non-significant positive effect size (Cohen's d = 0.17, 95% CI ranged from − 0.12 to 0.46), and the same was true for right PPC stimulation (Cohen's d= 0.1, 95% CI ranged from − 0.27 to 0.47, See Fig. 3B).Therefore, the result suggested TMS over RPPC/LPPC did not enhance detection performance in the ipsilateral hemifield when a unilateral stimulus was presented.
The multiple meta-analysis full model showed that the pooled effect size was 0.08, 95% CI ranged from − 0.22-0.38.After checking the variance distribution of the full model, it was observed that layer 1 to layer 3 accounted for 75%, 25% and 0% variance respectively.From the comparison between the full model and the leave-level 2-out, no significant model fitting difference were found (lower AIC and BIC) (p = 0.59).

Meta-analysis of TMS induced impairment effect on a bilateral stimulus
In order to assess whether TMS induced an impairment effect on a bilateral stimulus, 11 datasets were merged to yield a pooled effect size for left and right PPC stimulation.The test for heterogeneity was not significant (I2 = 42%; τ2 = 0.1114, p = 0.07), supporting the rationale for computing a fixed effect model.The fixed effect model showed a nonsignificant Cohen's d − 0.24, 95% CI ranged from − 0.47 to − 0.01 (see Fig. 5A).The Egger test showed that there was no significant publication bias (p = 0.47).After omitting any single study, the effect was still essentially the same.
For the right PPC stimulation, the fixed effect model showed a significant negative Cohen's d − 0.39, 95% CI ranged from − 0.68 to − 0.1.For the left PPC stimulation, the fixed effect model showed a nonsignificant negative Cohen's d − 0.03, 95% CI ranged from − 0.36-0.41.In other words, only right PPC TMS impaired detection of a bilateral stimulus.In order to further investigate the hemifield-specific contributions to this effect (i.e., potential contralateral impairment and ipsilateral enhancement effects), follow-up analyses focused exclusively on the right PPC.

Meta-analysis of a contralateral TMS effect on bilateral stimulus (right PPC only)
As shown in Fig. 5B, 7 datasets were merged into a pooled effect size.
The test for heterogeneity was not significant (I2 = 51%; τ2 = 0.1779, p = 0.06), supporting the rationale for computing a fixed effect model.The fixed effect model showed a significant negative Cohen's d − 0.87, 95% CI ranged from − 1.17 to − 0.58.The Egger test showed that there was no significant publication bias (p = 0.09).After omitting any single study, the effect was still essentially the same, i.e., TMS over right PPC impaired detection performance in the contralateral hemifield when a bilateral stimulus was presented.

Meta-analysis of an ipsilateral TMS effect on bilateral stimulus (right PPC only)
As shown in Fig. 5C, 6 datasets were merged into a pooled effect size.
The test for heterogeneity was not significant (I2 = 33%; τ2 = 0.0848, p = 0.21).The fixed effect model showed a non-significant Cohen's d 0.05, 95% CI ranged from − 0.33-0.44.The Egger test showed there was no significant (p = 0.1) publication bias.After omitting any single study, the effect was still essentially the same, i.e., TMS over right PPC did not enhance detection performance in the ipsilateral hemifield when

Bayesian meta-analysis of ipsilateral enhancement effects
A critical aspect of the results reported above is the absence of an enhancement effect in the ipsilateral hemifield.We also performed Bayesian meta-analyses to obtain additional support for the null hypothesis.In addition to conventional confidence intervals of effect sizes, a Bayesian meta-analysis can also provide a distribution of effect sizes and thus estimate the probability that the true effect is larger than x, given the data.Lastly, this analysis is particularly suited when the number of included studies is relatively small, as is the case here (Harrer et al., 2021).
For a unilateral stimulus presented in the ipsilateral hemifield, we combined the data of left and right PPC stimulation, whereas only right PPC stimulation conditions were considered for the performance in the ipsilateral hemifield for a bilateral stimulus, thus intended as direct follow-up analyses of the results outlined above.After confirming convergence (Ȓ =1), results showed essentially the same pooled effects for both analyses (unilateral stimulus: a non-significant positive Cohen's d 0.17, 95% CI ranged from − 0.13 to 0.49; bilateral stimulus: a nonsignificant positive Cohen's d − 0.11, 95% CI ranged from − 0.42 to 0.75).The ECDF function was then used to obtain the probability of the pooled effect being greater than a fixed effect size of 0.4 (medium effect size).Critically, there is only a 6% probability of an ipsilateral enhancement effect size of that magnitude for the unilateral condition, and a 13% probability for the bilateral condition.Taken together, this provides additional support for the absence of an ipsilateral enhancement effect, mirroring the outcome of the conventional meta-analysis.

Discussion
The objectives of this meta-analysis were to quantify the current empirical evidence base for a reliable TMS-induced i) contralateral impairment, ii) ipsilateral enhancement and iii) attentional selection bias in visual detection paradigms.We focused on studies using visual detection paradigms that allowed to assess the specific functional contributions of parietal cortex in both hemispheres to attentional processing in each hemifield separately, and thereby to evaluate diverging predictions of two influential theories of spatial attention control: the Heilman hemispatial theory and Kinsbourne's opponent processor theory.This is not only important for basic research on the cognitive neuroscience of spatial attention control, but also has direct clinical relevance as the currently common and officially recommended clinical TMS protocols in stroke rehabilitation are largely based on an assumed notion that most TMS studies are in agreement with the opponent processor model by revealing not only an inhibitory TMS-induced contralateral impairment in visual detection, but also ipsilateral enhancement.
We here included 10 studies using inhibitory/disruptive TMS applied over left or right parietal cortex in the context of visual detection/ extinction paradigms.Our meta-analysis of these studies globally revealed that inhibitory/disruptive TMS applied over posterior parietal cortex does functionally impact visual detection performance, confirming previous studies regarding the functional role of parietal cortex for visual detection performance (Bien et al., 2012a;Dambeck et al., 2006;Gießing et al., 2020;Hilgetag et al., 2001).More specifically, we found a significant contralateral impairment effect after TMS over right PPC in bilateral visual stimulus conditions, i.e. when two simultaneously presented visual stimuli are competing for attention.In the unilateral visual stimulus conditions, however, no significant contralateral impairment effect was observed.Importantly, we also could not find any significant ipsilateral enhancement effects in both, bilateral and unilateral visual stimulus condition.In addition, our findings revealed a hemisphere asymmetry in the bilateral stimulus conditions, where only stimulation of the right parietal cortex resulted in a significant impairment effect.
These findings nicely support and complement our previous metaanalysis (Wang et al., 2023) on left versus right parietal TMS effects on line-bisection and landmark tasks, which measured the attention bias.In this previous meta-analysis, we could confirm that studies using inhibitory/disruptive TMS over the right parietal cortex provided evidence for the functional relevance of the posterior parietal cortex in successfully executing these attentional tasks, but not the left parietal cortex.Similar to the previous meta-analysis, the current meta-analysis also clearly demonstrates that TMS applied over the right parietal cortex results in a significant contralateral impairment in bilateral stimulus conditions, while no such effect was observed in the left parietal cortex.Both findings are nicely replicating the hemispheric functional asymmetry also reported in neglect patients.
In comparison to the previous meta-analysis (Wang et al., 2023), the current meta-analysis addressed several limitations.Line bisection tasks may engage not only attentional mechanisms but also magnitude processing/comparison, which were predominantly a right parietal process (Cantlon et al., 2006;Faillenot et al., 1999;Piazza et al., 2006;Pinel et al., 2004;Sack et al., 2009;Seydell-Greenwald et al., 2019) and had a mixed attentional effect.More importantly, however, only the current meta-analysis allowed us to segregate the exact differential contributions of each hemisphere for each hemifield specifically, which is indispensable when referring to the separate functional role of each hemisphere for hemisphere-specific attention biases, gains, and costs towards the ipsilateral versus contralateral side of space (Duecker and Sack, 2015).As concluded in our previous meta-analysis, this can only be achieved by using a visual detection tasks.These tasks not only allow the calculation of attention bias (i.e.difference in detection performance left versus right for unilateral stimuli, or proportion of left versus right responses in case of bilateral stimuli) but furthermore allow assessment of detection performance in each hemifield in isolation (i.e.detection rates of left stimuli or right stimuli considered in isolation).Focusing on those studies employing such visual detection paradigms in the context of parietal TMS, our current meta-analysis indeed again also revealed TMS-induced contralateral impairment effects, but only for bilateral visual stimuli and not for unilateral visual stimulus conditions.This difference between TMS effects on bilateral versus unilateral conditions could be due to differences in task difficulty, with bilateral trials being more demanding for perceptual and attentional processes thus making them more susceptible to disruption by TMS.Alternatively, this result could also hint at a dissociation of unilateral and bilateral detection performance potentially related to a specific involvement of the targeted parietal regions in visual extinction.Vossel et al. (2011) have argued that unilateral spatial neglect and visual extinction can be linked to distinct lesion profiles and our results may also reflect this dissociability of the two phenomena.
Regarding the assumed and often referred to TMS-induced ipsilateral enhancement effect, we here could not find any supporting evidence for the existence of such an enhancement effect, neither during unilateral nor bilateral visual stimulus conditions.The conventional and Bayesian meta-analytical results statistically challenge the ipsilateral enhancement effect after parietal TMS.The pooled effect size in our metaanalysis, twenty years after the first highly influential reports of ipsilateral enhancement (Hilgetag et al., 2001), is simply not in support of such an enhancement effect when considering all available data.We conclude that the TMS-induced attentional shift consistently reported in line bisection or landmark tasks is probably exclusively caused by a contralateral impairment effect without any contribution of the ipsilateral hemifield.Accordingly, these meta-analytic findings of TMS-induced visual extinction in healthy volunteers are also not in support of Kinsbourne's opponent processor theory according to which the inhibition of one hemisphere by suppressive TMS should not only lead to contralateral impairments, but also cause a disinhibition of the contralateral hemisphere (which is released from its inhibition in the context of inter-hemispheric competition), causing ipsilateral enhancement.But this is not what the current evidence from these TMS studies T. Wang et al. suggests.These results clearly invite further discussion, not only on scientific grounds but also because Kinsbourne's model (Kinsbourne, 1977), including the notion of interhemispheric balance/competition, seems core to the rationale of clinical TMS treatment in stroke rehabilitation.Some clinical studies on neglect patients report ipsilateral enhancement after parietal stimulation targeting the healthy, unaffected hemisphere (Nyffeler et al., 2009;Yi et al., 2016), but in many clinical studies, the evaluation of improvement is based on paradigms that are not suited to separate ipsilateral and contralateral effects and/or the hemisphere-specific contribution in isolation (Lim et al., 2010;Oliveri et al., 2001;Song et al., 2009;Sparing et al., 2009) also see in the review (Mylius et al., 2012).While it remains difficult to compare healthy brains to those of lesion patients, the current meta-analysis raises concerns regarding the assumed TMS-induced enhancement in visual detection paradigms.We at this point and based on the currently available literature in fact need to acknowledge that there is no strong evidence base for claiming such a specific TMS-induced ipsilateral enhancement of attention after parietal stimulation in healthy volunteers.This often referred to and repeatedly communicated notion is based on an early and highly influential study that could not be replicated ever since.However, it continues to fuel the narrative for using inhibitory/disruptive TMS interventions targeted at the unaffected healthy hemisphere in neglect patients.While the verdict regarding the clinical efficacy of this specific TMS approach is still out (as an injured brain may respond very differently to TMS as compared to a healthy brain), the assumed empirical support for TMS-induced ipsilateral enhancements in visual detection paradigms in healthy volunteers seems to be a myth.
The predictions from Heilman's hemispatial theory (Heilman and Abell, 1980) are more in line with the current results.Heilman's model predicts contralateral deficits, which we found here, and makes no prediction of ipsilateral enhancement.However, our results do not explicitly support the 'soft supposition' of a larger contralateral impairment effect induced by right, as compared to left, parietal cortex disruption.Again, it is difficult to compare healthy brains to those of lesion patients, and TMS insults are fundamentally weaker and different by nature.There might be a difference between both hemispheres in their ability to compensate for an insult to the contralateral side, which remains hidden because TMS insults are not severe enough.
In sum, the current meta-analysis study challenges the Kinsbourne's opponent processor model and supports Heilman's hemispatial theory.The findings contribute to a better understanding of the visual attention system, hemispheric asymmetries, and highlights the importance of building strong empirical and theoretical foundations when translating fundamental research to clinical application.

CRediT authorship contribution statement
TW, FD & ATS conceived the study, which lead to TW & JW extracting the data under the supervision of FD & ATS.TW, FD and TG drafted the work, and TW& ZW conducted the analyses.TW, FD, TG, TS & ATS made substantial contributions to the interpretation of the results.All authors revised the manuscript critically for important intellectual content, approved the final version for publication and agreed to be accountable for all the aspects of the work.

Fig. 2 .
Fig. 2. Flowchart of data extraction.The flowchart includes database searches, the screening of study abstracts and full-texts, and the reasons for excluding studies.

Fig. 3 .
Fig. 3. Unilateral stimulus contralateral impairment and ipsilateral enhancement effects.Subgroup comparison of the Mean effect size (Cohen's d) and 95% confidence intervals for the 11 datasets for visual detection performance changes a unilateral stimulus presented in the contralateral hemifield (Fig. 3A, which correspond to the expectation Fig1, unilateral stimulus blue bar) and ipsilateral hemifield (Fig. 3B, which correspond to the expectation Fig1, unilateral stimulus orange bar) after left vs. right PPC stimulation.The red color squares represent the mean effect of each single study, and the black squares represent the pooled effects.

Fig. 4 .
Fig. 4. The funnel plot of TMS unilateral stimulus ipsilateral enhancement effect.A. The original funeral plot.Every dot represents a single study.B. The funeral plot after trim virtual studies.Black dots represent real studies and white dots represent virtual studies.

Fig. 5 .
Fig. 5. Bilateral stimulus TMS-induced effect.A. Subgroup comparison of the Mean effect size (Cohen's d) and 95% confidence intervals for the 11 datasets for visual detection performance change on bilateral trials after left vs. right PPC stimulation; B. the Mean effect size (Cohen's d) and 95% confidence intervals for the 7 datasets for visual detection contralateral hemifield performance change on bilateral trials (which correspond to the expectation Fig. 1, bilateral stimulus blue bar) after right PPC stimulation; C. the Mean effect size (Cohen's d) and 95% confidence intervals for the 4 datasets for visual detection ipsilateral hemifield performance change on bilateral trials (which correspond to the expectation Fig. 1, bilateral stimulus orange bar) after right PPC stimulation.The red color squares represent the mean effect of each single study, and the black squares represent the pooled effects.