Studying and modifying brain function with non-invasive brain stimulation

In the past three decades, our understanding of brain–behavior relationships has been significantly shaped by research using non-invasive brain stimulation (NIBS) techniques. These methods allow non-invasive and safe modulation of neural processes in the healthy brain, enabling researchers to directly study how experimentally altered neural activity causally affects behavior. This unique property of NIBS methods has, on the one hand, led to groundbreaking findings on the brain basis of various aspects of behavior and has raised interest in possible clinical and practical applications of these methods. On the other hand, it has also triggered increasingly critical debates about the properties and possible limitations of these methods. In this review, we discuss these issues, clarify the challenges associated with the use of currently available NIBS techniques for basic research and practical applications, and provide recommendations for studies using NIBS techniques to establish brain–behavior relationships. Polanía, Nitsche and Ruff summarize the state of non-invasive brain stimulation research in humans, discuss some current debates about properties and limitations of these methods, and give recommendations for how these challenges may be addressed.

In this review article, we outline the possibilities and limitations of NIBS methods 109 for investigations of brain-behavior relationships. We start with a concise overview of the 110 spatio-temporal properties of NIBS effects and the implications of these properties for the 111 use of these methods. In the second part, we will summarize and discuss recent debates 112 about the use of NIBS methods and provide recommendations for how these debates 113 may be addressed productively. Finally, we provide guidelines that may help to increase 114 both the conclusiveness of NIBS studies on brain-behavior relations and the potential 115 usefulness of NIBS protocols for possible translational applications. 116 117

Establishing brain-behavior relations with NIBS 118
While the evidence provided by brain imaging methods is purely correlative, it is 119 invaluable for identifying neural processes that may be targeted with causal manipulation 120 methods. In general, methods to causally manipulate neural activity can operate at 121 different levels of spatial specificity (micro-, meso-, and large-scale) and temporal 122 resolution (from milliseconds to days or even longer). In both these dimensions, NIBS 123 methods generally cover the middle ground, but specific ways of applying these methods 124 differ in their precise properties (Figure 1a). In terms of spatial resolution, the two most 125 popular methods (TMS and tES) lead to electric fields that span relatively large areas of 126 tissue compared to the effects of other, invasive methods ( Figure 1a and Box 1). 127 Therefore, claims about the spatial focality of the effects need to be interpreted with care 128 and should, whenever possible, be validated with combinations of neuroimaging methods 129 and computational modelling (we discuss this in more detail in the recommendations 130 section, below). Despite the relatively wide spatial spread of the electric fields across 131 large numbers of neurons, the "effective" spatial resolution for modulating various types 132 of behaviors is thought to be somewhat higher (Box 1 and Box Figure 1). This may reflect 133 that the behaviorally critical neural processes affected by the stimulation can themselves 134 be restricted to a relatively small number of cell groups within larger brain regions, and 135 that the stimulation can have different effects on neurons that are at rest or activated by 136 ongoing behavior 8, 9 . The functionally-relevant spatial resolution of NIBS methods may 137 therefore differ across different task contexts and may depend on the spatial extent of the 138 task-related ongoing neural processing. Moreover, different ways of applying the same 139 NIBS method can differ in their precise physical properties, which can set different limits 140 on their mechanism-of-action, physiological effects, and spatial/temporal specificity. 141 Different ways of applying NIBS methods are therefore suited to test different types of 142 hypotheses regarding physiology-behavior/cognition interactions. 143 For instance, online application of TMS (i.e., single-or double-pulse TMS, or short 144 bursts of TMS 10 ) elicits temporally restricted bursts of action potentials. The application 145 of such TMS pulses during task performance can be used to selectively interfere with 146 ongoing neuronal processes to study the temporal dynamics of brain function with high 147 temporal resolution (in the order of milliseconds). For examples, TMS pulses applied over 148 V1 at a specific latency from the onset of a visual stimulus can induce suppression of 149 conscious visual perception of this stimulus 11 and TMS pulses applied over cortical 150 language production areas can produce speech arrest within a specific timeframe 12 . 151 Additionally, simultaneous application of TMS pulses over different interconnected brain 152 areas 13 or during concurrent neuroimaging 14,15 (Figure 2c) allows tests of how action 153 potentials elicited in one brain area impact on processing in interconnected areas in a 154 top-down and/or context-sensitive manner; this allows direct study of how brain networks 155 dynamically operate at high temporal resolution and may make it possible to stimulate 156 deep cortical or subcortical areas indirectly via interconnected areas 14,15 . Moreover, 157 online TMS protocols that apply pulses at specific frequencies may facilitate 158 corresponding oscillations, thus allowing tests of the causal link between brain rhythms 159 and behavior [16][17][18] . Taken together, these studies demonstrate that online TMS protocols 160 exert influences on neural processing in a highly task-, context-, and time-dependent 161 manner; these protocols can therefore be tailored to affect specific aspects of neural 162

activity. 163
Other applications of TMS have focused on neuromodulatory after-effects 164 following repetitive TMS protocols 10 (rTMS). Depending on their specific frequency and/or 165 patterning, different rTMS protocols result in excitatory or inhibitory after-effects lasting 166 several minutes, which have been linked to long-term potentiation or long-term 167 depression (LTP/LTD, see Box 2), respectively. These after-effects are thought to reflect 168 rTMS influences on the strength of glutamatergic synapses via NMDA receptor, AMPA 169 receptor, and calcium channel effects 10,19-21 . Other possible mediators of these effects 170 may reflect non-linear time-dependent influences on inhibitory GABAergic neurons, non-171 synaptic mechanisms including alterations of the brain-derived neurotrophic factor 172 (BDNF, see Box 2), and even neurogenesis 22 . Given these modulatory impacts of rTMS 173 protocols on brain physiology, their effects by definition critically depend on brain state 174 during the stimulation 23 . The duration of the physiological aftereffects makes these behavioral alterations in the immediate aftermath of the rTMS protocol, thereby testing 178 the functional consequences of the temporary excitability modulation for behavior. 179 The second family of methods -tES -produces its neuromodulatory effects not 180 via magnetic fields (as TMS does) but rather by means of weak electrical currents applied 181 on the scalp. The most popular variant is transcranial direct current stimulation (tDCS), 182 introduced about two decades ago (Figure 1b). This method applies a weak tonic direct 183 current between electrodes mounted on the head, which partially passes through the 184 cortical tissue and affects relatively large cortical areas (on the order of centimeters, see 185 Box 1). This current de-or hyperpolarizes neuronal resting membrane potentials and 186 thereby alters cortical excitability 30,31 . The primary effects of tDCS do not include synaptic 187 mechanisms but instead involve voltage-dependent ion channels 32 . However, stimulation 188 extending over a few minutes leads to LTP-or LTD-like plasticity 32,33 that can extend to 189 inter-connected cortical and subcortical structures 34,35 . The temporal resolution of this 190 technique is low, as the online neuromodulatory effects start to take place few seconds 191 after the begin of the stimulation and continue throughout current application, whereas 192 the physiological aftereffects can last for several hours and even days if accompanied by 193 pharmacological interventions 32 . Thus, considering the physiology and neuromodulatory 194 characteristics of tDCS, the functional specificity of the intervention largely relates to its 195 capability to modulate task-related neural processing rather than to the spatial and 196 temporal specificity of the electric fields produced by the stimulation itself 36 . 197 While tDCS has low temporal resolution and is indiscriminate as to which aspects 198 of neural processing are modulated, other variants of tES methodology can be used to 199 target more specific aspects of neural function at higher temporal scales. One such 200 method was specifically developed to investigate the role of neural oscillations in 201 designated frequency bands for behavior 37 . This technique -known as transcranial 202 alternating current stimulation (tACS) -employs oscillatory electrical stimulation with the 203 aim of facilitating neuronal activity in specific frequency bands 38-40 , thereby allowing study 204 of causal links between brain rhythms and specific aspects of behavior 41-44 . For instance, 205 tACS can be used to study the causal role of theta-gamma cross-frequency coupling for 206 working memory performance 45 , the contributions of beta and gamma oscillations to 207 motor behavior 41,43 , the role of frontal gamma oscillations during high level cognitive 208 tasks 46 , or the causal contributions of alpha oscillations to the generation of visual and 209 crossmodal perceptual illusions 42,44 . 210 tACS can also be used to investigate how oscillatory coherence between spatially 211 distinct nodes of functional networks contributes to behavior 47-50 , by simultaneously 212 applying oscillatory currents over distinct regions at the same frequency, but using 213 different oscillatory phases to facilitate or hamper synchronization in the functional 214 networks ( Figure 2a). As mentioned before, the link between rhythmic oscillations and 215 behavior can also be investigated using rTMS protocols that apply pulses at specific 216 frequencies to facilitate corresponding oscillations [16][17][18] . Crucially, emerging work starts to 217 suggest that TMS pulses may have very different effects if they are applied at different 218 phases of ongoing neural oscillations 51 . This shows directly that some of the variability of 219 neural NIBS effects may relate to the precise temporal relation between the NIBS protocol 220 and ongoing neural activity, suggesting that this information could be used to design more 221 efficient stimulation protocols in the context of closed-loop systems 52-54 . 222 A limitation of the frequency-specific protocols mentioned above (and tES methods 223 in general) is that they can only directly affect activity in cortical regions. Direct stimulation 224 of deeper structures typically requires invasive procedures, for example deep brain 225 stimulation (DBS). However, there are attempts to develop specific TMS hardware -e.g. 226 the TMS H-coil 55 -to modulate the excitability of brain areas lying further away from the 227 cortical surface (possibly up to 6 cm) 56 . Moreover, a recent study showed in mice that a  However, this study could not directly demonstrate neural entrainment due to technical investigation of how tACS entrains or modulates oscillatory activity in the human brain will 246 require the development of multi-modal NIBS-recording techniques and well-validated 247 artifact rejection methods capable of identifying neural oscillations during stimulation 58,59 . 248 Another related tES technique called transcranial random noise stimulation (tRNS) 249 focuses on the link between behavior and frequency-specific noise inherent in neural 250 processing 60 . Compared to other stimulation methods, relatively little is known about the 251 physiological impact of this method. However, only 10 minutes of tRNS applied over M1 252 can enhance motor cortex excitability for about 60 minutes after the end of stimulation, 253 suggesting that this method may induce neuroplastic effects 60 of similar strength as those 254 induced by anodal tDCS. Applied in conjunction with cognitive tasks, tRNS protocols may 255 enhance learning performance even more strongly than anodal tDCS does 61,62 . 256 Interestingly, the effects of tRNS are strongest when used at intensities thought to induce 257 optimal noise levels 63 (Figure 2b), consistent with the stochastic resonance principle (see 258 Box 2). tRNS may thus prove useful for investigating the stochastic dynamics of neuronal 259 processing in the intact human brain 64 . 260 Standard NIBS studies using the approaches mentioned above typically apply 261 these protocols in purely behavioral settings, targeting brain areas identified by previous 262 neuroimaging research and assuming that the NIBS methods exert uniform and clearly 263 interpretable physiological effects on these areas. This standard approach has been used 264 for studying causal brain−function relationships in numerous domains, including vision 65  continues to yield very interesting demonstrations that specific aspects of behavior can 268 be changed by stimulation, and therefore causally relate to the affected neural processes, 269 it has also triggered critical debates about the properties and possible limitations of these 270 methods. We will discuss these in the following section. 271

273
Current controversies associated with the use of NIBS 274 Over the past few years, critical discussions have arisen about the replicability of effects 275 reported in various scientific fields 84,85 . For studies using NIBS, this discussion has 276 focused on both physiological and behavioral effects of these techniques. However, this 277 general discussion often has not explicitly differentiated between deterministic and 278 neuromodulatory NIBS approaches. The former methods -e.g., single-or double-pulse 279 TMS, or short bursts of TMS 10 -directly elicit action potentials that may have relatively 280 uniform physiological and behavioral effects (even though some intra-and interindividual 281 variability can be observed 86 ). The latter -e.g., offline rTMS or tES methods -mainly 282 operate by modulating ongoing brain activity, so that the effects of these methods will by 283 definition depend critically on brain state and task context. This state-dependency of 284 neuromodulatory NIBS effects is confirmed by animal studies showing, for instance, that 285 the ability to induce LTP and LTD is critically shaped by the previous learning experience 286 of the targeted cortical area 87 . Indeed, in humans, the effects of rTMS and tES on cortical 287 excitability (as monitored by TMS-generated MEPs) varies between individuals, as do 288 stimulation effects on other physiological and cognitive-behavioral variables 88-92 . 289 However, precise estimates of this variability are so far lacking, as the objectives and 290 methodical procedures of NIBS applications differ considerably between studies. This 291 severely complicates the use of meta-analytic procedures to estimate effect sizes 292 associated with NIBS applications: Such procedures can only validly be applied to 293 logically coherent sets of effects generated with the same well-defined methodical 294 procedures in the same task contexts. Preliminary attempts at quantifying effect sizes 295 associated with NIBS methods per se 93,94 have therefore been inconclusive, as they have 296 mostly pooled many different studies using this research method in very different ways. 297 The sources of the reported variability of NIBS effects have hardly been explored 298 systematically, but include brain-intrinsic, task-related, and methodological factors. Therefore, the individual variability of NIBS effects is not surprising, as NIBS protocols 305 induce plasticity by affecting glutamatergic, calcium-dependent mechanisms that are 306 affected by various neuromodulatory agents. By definition, these effects will therefore 307 vary between different tasks and brain regions (see below). As for methodological 308 aspects, variations of NIBS protocols in terms of intensity, duration, electrode position, 309 and coil orientation can alter stimulation effects, even in a non-linear fashion 100,101 (see 310 also Box 1). Additionally, the physiological effect of NIBS methods can strongly depend 311 on characteristics of the testing situation, as clearly illustrated by the fact that even MEPs 312 elicited from motor cortex following modulatory NIBS protocols can differ in strength 313 depending on what participants were doing at the time of stimulation (e.g., whether they 314 engaged in motor behavior or not 102 ). Finally, subject-specific aspects can also play a 315 role, such as differences in arousal or attentional state, ceiling or floor effects with regard to task performance, or differences in group size, just to name a few 103 . However, it is 317 important to highlight that many of these sources of variability are not unique to NIBS 318 studies and equally apply to many other research approaches attempting to relate 319 physiology and behavior in the biological and social sciences 104 (Figure 3). 320 The variability of reported NIBS effects need not be disadvantageous, but may 321 instead provide important information about how interventions may be personalized and 322 optimized 105,106 . Moreover, this natural variability may help to identify factors that affect 323 naturally occurring plasticity, thereby further elucidating the brain physiology underlying 324 cognitive processes. Future meta-analyses of NIBS effects should therefore attempt to 325 systematically identify the factors that determine the variability of NIBS effects; at the very 326 least, these analyses should only pool studies that indeed investigated the same specific 327 brain-behavior relationship with closely comparable NIBS procedures 93,104 . 328 The sources of physiological variability discussed above show that one cannot 329 assume that protocols known to result in enhancement or reduction of primary motor 330 cortex excitability -the most frequently-used assay of physiological NIBS effects -will 331 have the same physiological effect when applied to another brain area. Another factor 332 that may affect the variability of NIBS effects relates to possible non-linear interactions 333 with task-related neural processing. For instance, if NIBS methods and task performance 334 have synergistic effects on the same neuronal populations, neurons may be activated too 335 strongly, thereby resulting in antagonistic NIBS effects 101,107 . Finally, the link between 336 behavioral performance and physiological measures -such as TMS-generated 337 excitability measures or cerebral activation monitored by functional imaging -may in itself 338 not always be straightforward. For instance, improved performance during motor learning is known to result in activity reductions in motor cortex networks 108,109 . However, these 340 reductions obviously do not indicate that the functional relevance of this network has 341 decreased; instead, they may reflect that the selectivity of task-relevant networks has 342 increased 43 . NIBS protocols may therefore affect performance in opposite ways during 343 different stages of learning, as shown e.g. for visuo-motor coordination 110 . 344 One crucial, currently unresolved issue is the question whether tES protocols 345 always elicit their strongest effects under the electrodes, since computational models 346 suggest that the peak of the electric field should lie between the electrodes for some 347 montages (Box 1). Such computational models of tES-induced electric fields may 348 ultimately prove crucial for optimizing the efficiency of NIBS protocols 106,111 , but it will be 349 crucial to validate their computational predictions both physiologically and behaviorally, Critics believe that it may be too early to employ NIBS methods as routine neuro-359 enhancement tools, because the physiological effects vary between individuals (see 360 above) and because important translational questions needed for everyday use of NIBS 361 remain unaddressed. Most of the existing NIBS studies were conducted in controlled 362 laboratory settings, did not specifically aim for maximal and homogeneous effects, did not 363 explore long-term (and possibly performance-reducing) effects, and did not focus on 364 possible late-occurring side effects or side effects that might be caused by intensified use. 365 Obviously, this cautionary statement does not mean that NIBS will never be suitable for Some of the problems discussed in the previous section might relate to the variability of 387 methodical procedures employed in NIBS studies. This variability may reflect a lack of 388 clear guidelines on how conclusive NIBS evidence can be, given the details of how the 389 specific NIBS method was employed and how the resulting effects are interpreted. In this 390 section, we propose some tentative guidelines that may help in both assessing the 391 strength of evidence for brain-behavior relations in NIBS studies and for designing and 392 conducting NIBS studies. These guidelines may provide a starting point for overcoming 393 some of the limitations discussed in the previous section. Note that we focus these 394 guidelines on studies of brain-behavior relations; our recommendations may be neither 395 sufficient nor necessary for basic neurophysiology research using NIBS methods. 396 Overcoming the limitations of NIBS methods will require both specific methodical 397 procedures as well as combinations of NIBS procedures with other research methods. In 398 our eyes, the more these two strategies are adhered to in a given NIBS study, the more 399 conclusive the evidence for a specific brain-behavior relation can be ( Figure 4). For 400 instance, most exploratory and least conclusive may be those studies that acquire only 401 behavioral measures in combination with NIBS application over a target site that is 402 defined purely based on scalp measurements (using for instance the 10-20 system). We 403 expect this type of studies to result in the highest level of variability in effect size. On the 404 other hand, most conclusive (and least exploratory) about a brain-behavior relation may 405 be studies that incorporate the following methodical procedures: First, neuro-navigation 406 in order to more precisely locate the NIBS region of interest in each participant, e.g. based 407 on functional neuroimaging evidence or based on clearly defined anatomical criteria. This is arguably more critical for TMS studies than for studies employing tES with its relatively 409 coarser spatial resolution. However, tES studies may also benefit from this step since this 410 ensures more homogenous positioning of the areas of interest in the induced fields, in 411 particular for emergent tES protocols that offer higher spatial resolutions (see BOX 1 for 412 a discussion on this topic). Second, control tasks or behavioral measures that ascertain 413 that the NIBS effects are indeed specific for the behavior under study. Third, stimulation 414 of control regions/frequencies in order to test the functional specificity of the target 415 area/neural process of interest. Fourth, combination with neuroimaging in order to directly 416 quantifiy the strength of the NIBS effect on the local neural effect of interest, and to 417 measure how connected brain networks are affected by the application of the stimulation. 418 Fifth, characterization of the NIBS-induced changes with theory-driven models whose 419 mechanistic latent variables can capture changes in both behavioral and brain activity 420

modulations. 421
The multi-method approach we propose here may be impractical for clinical use 422 and may have poor ecological validity for standard clinical settings. However, we think it 423 may be decisive for basic research in order to provide conclusive evidence for the 424 effectiveness of a given NIBS protocol. This step appears essential to inform subsequent 425 translational and/or applied clinical use of these methods, which would not have to employ 426 the demanding research pipeline described in Figure 4 but could follow the exact protocol 427 established as effective in prior basic studies. 428 Adopting the type of multi-method strategies mentioned above are labor-intensive 429 and challenging, but this approach is increasingly adopted and therefore feasible 18,117-119 . 430 One example study 118 that utilized many of the methodical procedures suggested in 431 Figure 4 tested the hypothesis that working memory information is temporarily stored via 432 "activity silent" synaptic mechanisms (Figure 5a). This study used fMRI to localize cortical 433 areas that represent category specific working memory contents, and TMS combined with 434 EEG to characterize the temporal dynamics of the hypothesized memory reactivation. 435 Another study 18 utilizing similar procedures investigated the causal role of theta 436 oscillations (~6 Hz) on the dorsal stream for working memory maintenance (Figure 5b). 437 The authors used MEG to identify for each individual the cortical generators of theta 438 oscillations related to memory maintenance, and then tested the causal role of these 439 temporal-spatial oscillatory signatures supporting working memory maintenance with 440 combinations of rhythmic TMS and EEG that can test for neural entrainment 120 . A third 441 example study 117 demonstrated a causal role for the temporoparietal junction (TPJ) in 442 guiding strategic social behavior, by combining computational modeling of behavior, 443 neural activity recordings with fMRI, and transcranial magnetic stimulation (TMS) guided 444 by neuronavigation (Figure 5c). Notably, in all these studies, the documented effects were 445 shown to be specific for a given task context, brain region, or stimulation frequency. Thus, 446 these example studies demonstrate that NIBS studies can deliver conclusive evidence 447 for a specific, mechanistically defined brain behavior relationship (rather than being purely 448 exploratory) if researchers employ a methodical framework similar to the one illustrated 449 in Figure 4. As hypothesized, the new experiment revealed that cortical memories were re-exposed 463 during anodal tDCS, thereby illustrating how NIBS in combination with different 464 neuroimaging modalities (MRS and fMRI) can be used to reveal a more comprehensive 465 picture of the neurophysiological mechanisms underlying cognitive processes. 466 Shifting the field from more exploratory behavioral demonstrations to the multi-467 method approaches illustrated above requires careful planning of all stages of a NIBS 468 study ( Figure 6). That is, during the design stage of the experiments, the researchers 469 must already clearly define the area that should be stimulated, the cognitive process that 470 should be modulated, and how this NIBS influence on behavior can be measured 471 conclusively. This latter step requires a-priori considerations of including a control 472 task/behavioral measure to establish context-specificity and selecting a control brain 473 region to test the spatial selectivity of the intervention effect. Additionally, in order to

Implications for translational applications 508
Beyond studies employing NIBS methods to reveal causal brain-behavior relations, 509 important applications of NIBS protocols have always attempted to identify and potentially 510 ameliorate pathophysiological mechanisms underlying neurological and psychiatric 511 diseases. The problems discussed above apply in a similar manner to these more clinical 512 and translational applications of NIBS methods. While the use of NIBS for therapeutic 513 applications has been extensively investigated, the corresponding treatment effects have 514 been moderate and variable in most cases; beyond the use of prefrontal rTMS for 515 treatment of major depression, no NIBS protocol has developed into a routinely-used 516 treatment tool so far 134 . This does not necessarily reflect limited therapeutic potential of 517 NIBS interventions. However, it does suggest that research strategies in this field so far 518 may not have been well suited to develop and identify NIBS protocols with optimal 519 efficacy. At least three lines of research may advance the field in this respect. First, it will 520 be important to base any intervention protocol on solid mechanistic knowledge about the 521 causal and specific contribution of brain areas and networks to clinical symptoms. In would have to be derived with combinations of brain stimulation, neuroimaging, solid 524 experimental designs, and modeling work (as attempted e.g. in computational 525 psychiatry 135 ). Such initial studies in healthy participants should lead to further 526 translational treatment-validation studies that should not only monitor clinical symptoms 527 but also physiological data, to validate the precise neurophysiological mechanisms 528 causally mediating the intervention effects. Second, promising treatment protocols 529 identified with the strategy discussed above should be further optimized by systematic 530 evaluation of the optimal stimulation areas and parameter settings for the stimulation; this 531 should initially be performed in healthy surrogate populations but should importantly be 532 directly validated in the target patient groups (to account for the state-dependency of 533 neuromodulatory NIBS protocols discussed above). This optimization of intervention 534 protocols may not be restricted to the group level, but should include individual 535 optimization of the protocols dependent on brain state, lesions, clinical symptoms, and 536 other factors. Third, the field is currently characterized by a multitude of studies with 537 relatively small sample sizes. While this may be helpful for exploratory and screening 538 purposes, it is not sufficient for establishing the clinical relevance of an intervention and 539 for decisions about its implementation in clinical routine. Thus, larger and preferably multi-540 center randomized clinical trials should be conducted to establish with adequate statistical 541 power which protocols may have clinically relevant effects, and on whom. All these steps 542 would be important to provide solid evidence for the usefulness of applying these 543 validated protocols in more basic and less research-oriented clinical settings.

Conclusions 548
In the last 30 years, NIBS methods have become indispensable tools for elucidating how 549 behavior causally depends on specific aspects of neural activity in the healthy human 550 brain. There is presently no alternative to these techniques for the study of causal brain-551 behavior relationships in humans, but current controversies highlight that the use of NIBS 552 for research purposes requires responsible scientific practice. This may necessitate a 553 shift in focus from simplistic assumptions about how NIBS methods generally affect the 554 brain towards more physiologically informed multi-method approaches that test specific 555 hypotheses about how NIBS influences on behavior are mediated by modulation of well-556 defined neural processes. These approaches should explicitly consider various intrinsic, 557 task-related, and methodological factors that can potentially influence the variability of 558 behavioral and physiological outcomes. Moreover, more attention should be devoted to 559 the precise reporting of methods, protocols and results to allow more accurate 560 interpretations and future summary of the data. Of course, these considerations are not 561 only important for NIBS research but also for other fields of experimental sciences. But 562 the current debates highlight that NIBS research in particular may be at a crossroads 563 where the field would strongly benefit from coordinated methodological efforts to optimize 564 the conclusiveness of findings on brain-behavior relations. This step appears vital for 565 successful translational applications of these methods for cognitive enhancement and 566 improved mental health. increasingly conclusive and mechanistically-informed evidence for the relationship between behavior and a well-defined neural process (for examples, such a scheme was 630 followed in REFs. 18, 117 and 118; see also Figure 5). It is important to note that the 631 scheme is illustrative rather than fully prescriptive, as the precise order of these 632 procedures is not necessarily the same for all studies and as one or several of the 633 illustrated procedures may not apply or be available in particular contexts. Moreover, it 634 should be noted that clinical or translational studies may not necessarily benefit from 635 following these procedures if they apply well-validated protocols. However, the more of 636 these methodical procedures that can be included in a given study, the more conclusive 637 and mechanistically informed the resulting evidence. represents beliefs about how our choices will influence those of others we interact with 117 . 658 The authors first identified the region of interest using fMRI and computational modelling 659 (left panel). The authors then used rTMS to inhibit the activity of the right temporoparietal 660 junction (rTPJ), which was hypothesized to implement the social influence signal (middle 661 panel). Additionally, the authors also used a remote control region (vertex) to test the 662 regional specificity. After rTMS, participants performed the social task during fMRI and 663 used computational modelling to study how mechanistic latent variables of behavior where affected by the inhibitory rTMS protocol to the rTPJ compared to the control region 665 (right panel). This multimethod approach, thus, allowed the authors to reveal a regional 666 and functional specific causal role of the rTPJ in computing social influence signals. 667 668 669 670 Figure 6. Multi-method approaches can be used to gain fundamental and more reliable 671 insights on brain-behavior relations via NIBS. However, in order to carry out such studies 672 involving high methodological effort (see Figure 4), it is crucial to have a clear work plan 673 before conducting the study. This scheme shows an example of important aspects to 674 consider in such a work plan before, during and after the execution of NIBS studies. One strategy that has been proposed to address these issues is to estimate 694 computational models of the most likely induced electric fields, which has led to the 695 development of novel electrode configurations 139 that may help to predict NIBS-induced 696 effects with greater accuracy 106,111 (Box Figure 1a,b,c). For instance, modelling work 697 suggests that conventional electrode montages might induce effects not only under the 698 electrodes but also between them, and that for some montages the strongest fields may 699 actually not lie under the electrodes (Box Figure 1a, top). While these efforts at modelling tES-induced electric fields and effects on neurons may ultimately prove crucial for 701 optimizing the efficiency of NIBS protocols, it is important to note that such models need 702 to be physiologically validated 9,36,106 and will need to be able to fully account for the well-  Figure 1c). However, the neurophysiological underpinnings of 721 these tFUS-induced changes of cortical excitability still need to be understood in much 722 more detail before this method can be put to safe routine use. 723 In an attempt to answer the question "which aspects of neural processing are 724 influenced by NIBS?", researchers have tried to measure the neurophysiological 725 influences of NIBS using a variety of methods including in vitro 141 , in vivo 9,40,140  In humans, brain-derived BDNF gene polymorphisms have been shown to have an be taken into account when considering potential sources of behavioral and physiological 746 variability in NIBS-induced effects (Figure 3). 747

Long-term potentiation (LTP): A facilitation of synaptic transmission that is considered 748
to be one of the major mechanisms underlying learning and memory formation. The 749 opposite phenomenon, long-term depression (LTD), refers to inhibition of synaptic 750 transmission. LTP and LTD are thought to be expressed at possibly every synapse in the 751 mammalian brain 147 . Long-lasting neurophysiological facilitation or inhibition induced by 752 NIBS (depending on the method and protocol used and additional factors such as brain 753 state and cognitive task) is believed to relate to LTP-or LTD-like changes. is too weak to be detected by a sensor is enhanced by adding an optimal level of noise. 767 For instance, it has been shown that visual detection performance can be increased by