BDNF Val66Met gene polymorphism modulates brain activity following rTMS-induced memory impairment

The BDNF Val66Met gene polymorphism is a relevant factor explaining inter-individual differences to TMS responses in studies of the motor system. However, whether this variant also contributes to TMS-induced memory effects, as well as their underlying brain mechanisms, remains unexplored. In this investigation, we applied rTMS during encoding of a visual memory task either over the left frontal cortex (LFC; experimental condition) or the cranial vertex (control condition). Subsequently, individuals underwent a recognition memory phase during a functional MRI acquisition. We included 43 young volunteers and classified them as 19 Met allele carriers and 24 as Val/Val individuals. The results revealed that rTMS delivered over LFC compared to vertex stimulation resulted in reduced memory performance only amongst Val/Val allele carriers. This genetic group also exhibited greater fMRI brain activity during memory recognition, mainly over frontal regions, which was positively associated with cognitive performance. We concluded that BDNF Val66Met gene polymorphism, known to exert a significant effect on neuroplasticity, modulates the impact of rTMS both at the cognitive as well as at the associated brain networks expression levels. This data provides new insights on the brain mechanisms explaining cognitive inter-individual differences to TMS, and may inform future, more individually-tailored rTMS interventions.


Materials and methods
Participants. The methods and experimental procedures (except for magnetic resonance imaging [MRI] procedures, see below) used in the present investigation are described in full in our previous report 10 . In brief, we adapted a TMS-memory interference protocol 6 and demonstrated replicability of the induced effects across a two-center study. In the present report we focused on functional MRI (fMRI) activation changes during the recognition phase as a function of the BDNF Val66Met gene variation. From our previous study 10 , we included all participants with available fMRI data (N = 43; mean age: 23.47 ± 3.5 years). No additional subjects were recruited. All participants were healthy right-handed young males with at least secondary educational attainment. Subjects were recruited from two centers, Barcelona (BCN; N = 31) and Marseille (MRS; N = 12), where the experimental protocols had been harmonized in the context of the IMI FP7 European 'Pharmacog' research project (https:// www. imi. europa. eu/ proje cts-resul ts/ proje ct-facts heets/ pharma-cog). The study protocol was approved by the French ethics committee "SUD MÉDITERRANÉE I", the French regulatory authority Agence Nationale de Sécurité du Médicament (ANSM) and the Spanish committee "Comitè d'Ètica d'Investigació Clínica de l'Hospital Clínic de Barcelona" (CEIC) in Barcelona. The study was in accordance with the Declaration of Helsinki. All volunteers were properly informed and gave written consent. Thereby, this study constitutes a sub-analysis of our previous investigation 10 , conducted within the IMI FP7 European 'Pharmacog' project, herein leveraging brain imaging and genetic data.
Experimental design. Participants attended their respective center (BCN or MRS) and undertook an encoding visual memory task while receiving short and high frequency rTMS trains (see next section for further detail) either over the left frontal cortex (LFC; experimental condition; see "TMS protocol" section) or the cranial vertex area (control condition). Subsequently, they performed the recognition of the visual memory task while brain imaging data was acquired (Fig. 1). TMS protocol. TMS was applied using figure-eight shaped coils. A MagPro X100 magnetic stimulator (MagVenture A|S, Denmark) combined with an eXimia Navigated Brain Stimulation system (Nexstim, Finland) and a Magstim stimulator (Magstim Company Limited, USA, CE certification) combined with the neuronavigation system Brainsight 2.2 (Rogue Research Inc., Montreal, QC, Canada) were used for the BCN and MRS subsamples, respectively. Resting motor threshold (RMT) was assessed as described in the International Standard Guidelines 33 for all participants before conducting the encoding task. High frequency (20 Hz) 900 ms rTMS trains were delivered synchronously with the encoding task (500 ms after onset of pictures presentation) at 90% intensity of individual RMT. LFC stimulation site was extracted from a previous comparable fMRI memory study, as the maximum peak-activity coordinate within left frontal cortex (MNI coordinates: X = − 42; Y = 10; Z = 30 34 ; see also Supplementary Material [SM] for anatomical details). Vertex stimulation control point was determined according to the 10-20 electrode placement (Cz 35 ). Structural MRI was acquired for all participants (see below) and stimulation point was then transformed from MNI space to each native image using FLIRT tool from FSL (see below). To guide stimulation and ensure accuracy both in spot localization and coil positioning, neuronavigation was conducted in each volunteer on the basis of his structural MRI with stereotactic registration.
MRI protocol. The (3) HF > HV were defined (see also SM). First level analyses were further fit into higher-level statistics using the FMRIB's Local Analysis of Mixed Effects (FLAME 39 ). For the main contrast of interest (i.e., HF > HV), we created group GLM designs to evaluate: (1) differences between BDNF subgroups (Val vs. Met); and (2) subgroups mean activity maps. Additionally, group differences (Val vs. Met) were tested for the other first-level contrasts calculated (i.e., mean HF and mean HV; see SM for further details). All analyses were performed in the whole brain at a voxel-wise level, and a Z > 2.3 was used to define contiguous clusters of activity, then cluster significance levels were estimated and corrected using family-wise error (FWE) correction. The significance threshold was set at a corrected p < 0.05.

Statistical analysis.
Non-imaging data analyses were performed using IBM SPSS (IBM Corp. Released 2020. IBM SPSS Statistics for Windows, Version 27.0. Armonk, NY: IBM Corp). For cognitive performance data, accuracy (percentage of pictures correctly categorized as indoor or outdoor), and reaction time (RT; time elapsed from the presentation of a picture to the subsequent response) during encoding and hits (percentage of correctly recognized pictures), and RT (time elapsed between picture appearance and yes/no motor response) during recognition, were collected for both experimental conditions (LFC and vertex). For these outputs, oneway repeated measures ANOVAs were conducted with experimental conditions (i.e., LFC vs. vertex) as withinsubject factor and BDNF Val66Met gene polymorphism (i.e., Val vs. Met) as between-subject factor. When interactions or main effects emerged, subsequent pair-wise analyses were conducted to investigate directionality of the data, or to explore if a specific group or condition driven the observed results. Hence, paired-samples t-test analyses were performed comparing the performance between experimental conditions in each genetic subgroup. Additional independent-samples t-test analyses were conducted comparing genetic subgroups in each experimental condition performance' measure. Finally, we extracted the mean values of the blood oxygen level dependent (BOLD) within the significant regions of interest (ROI) derived from the fMRI analyses, in order to test associations between functional activity and cognitive performance estimates using Pearson correlations. All statistical analyses were two-tailed and α was set at 0.05 (see SM for further details and analyses). Due to techni-  68 , created for visual purposes, pointed with a yellow arrow). After a 30 min rest, participants completed the recognition of the visual memory task (i.e., discrimination between seen/not seen items in encoding phase) within the MRI scanner. RMT, resting motor threshold; rTMS, repetitive transcranial magnetic stimulation; LFC, left frontal cortex; fMRI, functional magnetic resonance imaging. www.nature.com/scientificreports/ cal issues during the recording of the encoding phase, data from three subjects was not available for conducting accuracy statistical analyses (see also Table S1). Regarding RT, a main effect of condition was observed for accuracy' RT (F (1,38) = 5.240, p = 0.028, η p 2 = 0.121), but not an interaction (F (1,38) = 3.658, p = 0.063, η p 2 = 0.088). Subsequent paired-samples t-test analyses revealed that the accuracy' RT was lower following vertex as compared to LFC stimulation for the Met allele carrier subgroup (t (17) = 2.365, p = 0.030), but not for Val/Val allele carriers (t (21) = 0.348, p = 0.731). These differences did not survive when controlling RT for accuracy performance (interaction: F (1,38) Table S1, reporting all behavioral data, and Fig. S1, showing individual data).

Results
In summary, at the cognitive level we observed greater rTMS interference effects amongst Val/Val individuals compared to Met allele carriers, both at the encoding as well at the recognition phase. fMRI findings. First, we investigated whether there were fMRI BOLD activation differences for memory performance (hits) associated to both rTMS conditions (HF and HV) as a function of genetic background (Val vs. Met). For the contrast of interest (i.e., HF > HV), we found BDNF group differences in the sense of more fMRI BOLD activation for the Val/Val than the Met allele carrier subgroup over a right frontal cluster comprising the frontal pole and the middle and superior frontal gyri (Fig. 3A). Subsequent analyses corroborated that these effects were driven by increased frontal fMRI BOLD activation amongst Val/Val subgroup during memory recognition for items corresponding to the LFC condition relative to vertex. The clusters identified were allocated bilaterally including the frontal pole, the paracingulate gyrus, and the anterior division of the cingulate gyrus (Fig. 3B). No significant differences appeared for the Met subsample in the same HF > HV contrast of interest. All these results survived when controlling for the center variable (no effect of center was observed). Additional analyses exploring genetic group fMRI differences during both mean HF and mean HV contrasts can be found in SM and Fig. S2.  Fig. 3D). The associations between the behavioral measure and the means of activity within the stated imaging clusters did not reach statistical significance in the Met group (r = 0.445, p = 0.056; r = 0.453, p = 0.052). Further exploratory correlative imaging-behavior analyses can be found in SM. Hence, those subjects in the Val group (which, at the cognitive level, were more vulnerable to rTMS interference than the Met allele carrier subgroup) exhibiting greater overactivations in the stated frontal brain areas, were those in whom the TMS-related cognitive interference was lower.

Discussion
The present study confirmed the expected episodic encoding interference using an adaptation of a rTMS memory paradigm [6][7][8][9] . Specifically, we found that cognitive interference effects were driven by the Val/Val homozygous. Furthermore, this effect was observed in the context of a greater brain activity pattern during the LFC condition in Val/Val subjects, which in turn was positively associated with memory performance. Present results, therefore, are consistent with the BDNF Val66Met genetic variation being a relevant factor for inter-individual variability of rTMS effects on cognitive function, and provides novel evidence of the underlying neurophysiological mechanisms subtending such effects.

Cognitive findings. Cellular and molecular studies have stablished that BDNF exerts an important role in
regulating the functional dynamics of brain regions relevant for learning and memory 40 and that the Met allele of the BDNF Val66Met genetic variation is associated with a decreased activity-dependent secretion of BDNF 18 . Early influential findings (e.g., 18 ), indicated that Met over Val/Val allele carriers exhibited lower memory performance. However, a recent systematic review observed that only around half of studies reported positive associations, leading the authors to conclude that the effects of this genetic variation on human cognition remain inconclusive 40 . www.nature.com/scientificreports/ Previous inconsistent findings may obey to several conditions, including sample characteristics (i.e., sample sizes and healthy individuals vs. patients 40 , age effects 41 , or specific cognitive tasks 42 ). In our study, we did not find a difference in memory recognition dependent on Val/Met variation when groups were compared under the control condition. However, we did observe a clear genetic effect when memory was assessed as a function of rTMS interference. Therefore, present results highlight that using rTMS as a perturbation approach 43 represents a sensitive approximation to reveal the effects of factors contributing to individual differences in cognitive performance that may not be apparent during conventional cognitive assessments.
More specifically, our findings emphasize that rTMS delivered over the LFC decreased performance in Val/ Val carriers both during the initial attentional tasks as well as at subsequent memory recognition performance. While memory effects have been consistently reported in the literature using the original interference paradigm (i.e., 6 ) the effects during the encoding process were not reported previously. A possibility regarding this discrepancy may be related to the specific stimulation site used in our adaptation of the protocol. Here, and guided by a previous fMRI pilot study 34 , our stimulation target overlapped with the left frontal eye field (L-FEF) node of the dorsal attention network (DAN), which is associated to the capacity of orienting one's focus to a particular task and specifically contributes to top-down guided voluntary allocation of attention to locations or features [44][45][46] . In contrast, F3 site of the 10-20 EEG system, used in the original studies [6][7][8][9] , is mainly located within the frontal areas of the left executive-control network (LECN), which subtends executive functions including working memory and decision-making processes (e.g., 47 ; for a visual comparison of stimulation sites, see Fig. S3). Hence, the effect we observed at encoding accuracy could be explained by a more specific rTMS-induced perturbation over dorsal attention systems. In these regards, it can be conceived that the TMS-induced modulation at the attentional memory stage constrained the behavioral TMS effects observed during the retrieval, and therefore in the current study are difficult to dissociate from its preceding codification phase.
Altogether, our findings align with previous literature conducted with electromyographic recordings of MEPs during stimulation of the motor cortex (e.g., 22,48 ) as well as with a previous report focusing on decision-making processes 49 , confirming that BDNF Val66Met genetic polymorphism is a relevant factor explaining variability in response to cognitive effects of TMS, and suggesting greater susceptibility for Val/Val homozygous.

Brain activation findings. Previous neuroimaging studies conducted during memory tasks revealed
greater medial temporal lobe (MTL) activity amongst Val/Val allele carriers 17,18 or similarly, negative associations between brain activity and the dose of Met-BDNF allele 50 . In the same line, during a working memory task, greater brain activity with increasing memory loads in the medial prefrontal areas, in particular within the superior frontal gyrus, was observed amongst Val/Val but not Met carriers 51 . However, the directionality of these findings is not consistent across the studies, as for example Dennis et al. 52 reported greater MTL activity amongst Met/− allele bearers.
As in the cognitive studies referred above, a number of factors, including the type of memory task employed (i.e., item vs. relational), design (i.e., blocks vs. event-related) or differences in performance between genetic groups may determine the significancy and directionality of the observed functional imaging findings. Furthermore, it should be noted that a meta-analysis on this topic concluded that while a biological effect of this genetic variation detectable in humans is plausible, its analysis using imaging protocols will require large samples sizes to identify consistent effects 53 . In our study, with a comparable or larger sample size than previous investigations, we did not observe differences between genetic subgroups in the control condition. However, differences were clearly amplified when reflecting recognition memory performance for items where rTMS interference was introduced at encoding. Hence, aligned with our conclusions at the cognitive level, present findings reinforce the idea that rTMS combined with neuroimaging techniques represents a useful experimental approach to highlight the underlying brain mechanisms through which factors that influence variability of cognitive functions in humans operate.
Across the different imaging analyses, present findings revealed that Val/Val subjects exhibited a more extended pattern of brain activity than Met group after rTMS over the LFC. First, increased activity in parietal, temporal, occipital and motor regions were found corresponding to LFC-linked recognition events (Fig. S2A). Topographically, some of these areas are embedded within the general brain network subtending memory recognition processes 54 or have been observed during visual working memory fMRI studies (e.g., 55 ). The pattern also included areas typically non-related to episodic memory processes, such as pre and posterior central gyri regions contralateral to the hand used to press the 'correct' (i.e., hits) button during the task. When looking at more specific effects, we identified increased brain activity over the right frontal region for the HF > HV contrast. In subsequent analyses conducted to explore the directionality of such results, we observed that Val/Val group (but not Met carriers) showed greater activity specific to rTMS-LFC (i.e., HF > HV) in bilateral frontal and anterior cingulate regions. These findings indicate that increased bilateral frontal activity, and particularly in contralateral areas to the stimulation site (i.e., rTMS was delivered at encoding over the LFC), was observed amongst Val/ Val subjects and that this was only evident during the effortful memory recognition process where, in average, rTMS induced greater cognitive deficits in Val compared to Met group. In this sense, increased brain activity in frontal regions including bilateral activations (i.e., more widespread allocation of functional resources) have been frequently reported in the neuroimaging literature, in particular attributed to compensatory processes (see below). Furthermore, previous rTMS findings revealed that amongst young but not older individuals 7 , perturbation of brain activity over the right frontal cortex during memory recognition results in clear interference effects, advocating for a causal role of this region during these cognitive processes [6][7][8][9] . Overall, present fMRI findings suggest that the greater permeability of Val/Val subjects to rTMS effects during memory encoding resulted in over recruitment of brain regions specifically subtending memory recognition. www.nature.com/scientificreports/ Associations between fMRI findings and cognitive performance. We found that the observed increases of brain activity amongst Val/Val allele carriers were positively associated with memory performance. This was evidenced in a specific way through right and bilateral frontal activations, after subtracting both brain activity and memory performance related to the vertex control condition from LFC (Fig. 3C,D, respectively). Also, in a general manner, brain activity under the LFC condition was associated to hits performance (Fig. S2B). Such findings suggest evidence of compensation, a concept particularly developed within the field of cognitive neuroscience of aging, and which refers to the capacity of engaging additional neural resources (i.e., such as reflected by increased fMRI BOLD signal) to counteract the lack of functionality of the typical brain resources in a given situation [56][57][58] . Compensation has been frequently associated with increases in contralateral recruitment as an adaptative shift [57][58][59] and is a term associated to enhanced cognitive performance. However, a task-related increase in cognitive demands may be completely counteracted by the recruitment of additional neural resources, or it may reduce but do not completely eliminate the gap between task demands and available resources 57 . This latter scenario, which may also be conceptualized under the term 'attempted compensation' 60 , appears to fit particularly well with our findings of positive associations between the expression of neural resources amongst Val/ Val allele carriers, within the context of an overall greater rTMS-induced interference on memory recognition, as compared to Met/− individuals. Compensatory processes have also been reported in the context of task-performance studies combining TMS with functional neuroimaging data. For example, in the language domain, bilateral reorganizations in homologous right-hemispheric areas after disruption of left-hemispheric language regions have been reported (reviewed in 61 ). In the case of memory studies, rTMS-induced bilateral fronto-parietal reorganizations amongst older individuals concomitant with an improvement in a visual memory task were reported 62 . Furthermore, Davis and colleagues 59 delivered 1 and 5 Hz rTMS trains over the left middle frontal gyrus and investigated the effects during a subsequent fMRI source memory task. First, they observed that 1 Hz, an inhibitory stimulation, resulted in decreased 'memory-related' activity in the prefrontal area under stimulation while 5 Hz, an excitatory protocol, increased it. Second, and using multivariate network analyses, the authors found that induced decreases in the stimulated node (i.e., following 1 Hz rTMS) resulted in memory-related increases in connectivity with other cortical regions (frontal, temporal, parietal), in particular with homologous contralateral right PFC areas, thus providing direct support for the compensatory hypothesis (i.e., that local dysfunction is counterbalanced by a more global recruitment; for a review on this topic, see also 63 ).
The present findings using rTMS as a technique to directly manipulate neural activity are aligned with previous rTMS studies conducted amongst older adults (i.e., 59,62,64 ) showing that a disruption of local resources resulted in a shift of brain activity positively associated with task performance. However, it should be noted that in the present study we employed an rTMS 'online' stimulation protocol during the memory encoding phase, using very brief (i.e., 900 ms) trains of a high frequency stimulation (20 Hz), that in principle should not induce after-effects. Hence, and since our memory recognition task was conducted 30 min later, it is very unlikely that the observed fMRI findings are confounded by direct TMS effects (i.e., 'after-effects') on the reorganization of brain networks. Therefore, our fMRI findings reveal genuine differences on brain network expression directly associated to the BDNF Val66Met human variation. Nonetheless, it cannot be ruled out the possibility that our results, focused on functional systems, might be partially explained by brain structural differences related to the BDNF Val66Met polymorphism. In this vein, previous studies have claimed that a critical role of the BDNF polymorphism on brain plasticity might be associated with the robustness of white matter structural networks, being Val homozygote subjects those showing more resilience under targeted attacks on central nodes 65 . However, in a more general manner, it should be noted that structural and functional connectivity relationships appear to be weak in healthy populations 66 . Altogether, and according to strong cellular and molecular evidence linking this polymorphism with synaptic transmission and brain plasticity mechanisms 32 , the most likely explanation for the observed overactivations suggest the expression of rapid adaptative plasticity responses amongst Val/Val allele carriers allowing to−partially−counteract greater interference effects at encoding.

Limitations
Our study has some limitations. First, in the present investigation, all subjects were male. This fact, which limits the generalization of our results, is clearly acknowledged and was chosen because an eventual objective of our research, within the framework of the European IMI-PharmaCog European initiative, was to compare our TMS findings with those obtained using sleep deprivation, as a second planned controlled sets of experiments designed to transiently disrupt memory processes. Gender effects have been described to interact with sleep restriction in their impact on cognitive performance 67 , and therefore only male participants were included in all studies. Further, and due to its low relative occurrence, we could not analyze event-related brain activity differences associated to the induced errors in memory performance (i.e., misses or false alarms). Finally, it should be noted that albeit a direct perturbation role of rTMS can be assumed for the interference of memory encoding process, the proof that overexpression areas are reflecting compensatory efforts (i.e., causally associated with task performance) would equally require to target such areas/networks with an additional right-sided TMS perturbation approach and observe the behavioral effects.

Conclusion
Through the induction of encoding interference, the present rTMS-fMRI investigation allowed to increase evidences of the study of genetic influences on the expression of brain networks involved in memory functions. Present results also suggested that Val/Val individuals may be particularly sensitive to TMS effects, indicating that future studies with individuals bearing this genetic profile could be a useful approach to detect proof of