Neural theta oscillations support semantic memory retrieval

Lexical–semantic retrieval emerges through the interactions of distributed prefrontal and perisylvian brain networks. Growing evidence suggests that synchronous theta band neural oscillations might play a role in this process, yet, their functional significance remains elusive. Here, we used transcranial alternating current stimulation to induce exogenous theta oscillations at 6 Hz (θ-tACS) over left prefrontal and posterior perisylvian cortex with a 180° (anti-phase) and 0° (in-phase) relative phase difference while participants performed automatic and controlled retrieval tasks. We demonstrate that θ-tACS significantly modulated the retrieval performance and its effects were both task- and phase-specific: the in-phase tACS impaired controlled retrieval, whereas the anti-phase tACS improved controlled but impaired automatic retrieval. These findings indicate that theta band oscillatory brain activity supports binding of semantically related representations via a phase-dependent modulation of semantic activation or maintenance.


Orthogonalization of the experimental conditions.
The participants were pseudo-randomly assigned into experimental condition tACS (sham, anti-phase, in-phase) and association chain test (ACT) block (A, B, C), which were properly orthogonalized across the three experimental sessions (see Table S1 and S2, respectively) and each other (see Table S4). Word stimuli in ACT blocks are shown in Table S4.

Statistical analyses (Linear mixed effect models)
Effects of the ACT conditions on the retrieval RT during sham conditions (i.e., not affected by active tACS conditions) were evaluated using LMEM as model = RT ~ RespT + SeqT + RespT:SeqT + (1 | ID), where RT were response times, RespT was response type (associative or dissociate), SeqT was sequence type (fixed or alternating), and 1 | ID was random intercept for subjects. This model was evaluated using Anova() function car package. The effects of tACS on the basic ACT measures (separately for category, associative fixed, associative alternating, dissociative fixed, and dissociative alternating measure) were assessed by comparing model0 = RT ~ Block + (1 | ID) and model1 = RT ~ tACS + Block + (1 | ID), using anova() function from stats package. The tACS factor included 3 conditions (sham, anti-phase, in-phase). Block factor had also 3 conditions (A, B, C), representing stimulus material, that were counterbalanced across the sessions (see Table S4). Additionally, in order to assess the moderating role of difficulty in category retrieval, we used model = RT ~ tACS * Half + Block + (1 | ID) formula. The Half factor had two conditions (first, second). The model was evaluated using Anova() function. The effects of tACS on response initiation were evaluated by comparing model0 = RT ~ (1 | ID) + (1 |Block:Condition) and model1 = RT ~ tACS + (1 | ID) + (1 |Block:Condition), using anova() function. In this model, the Condition factor had two levels (category, associative) and was used to account for the differences between these two ACT conditions. The same syntax was used for inhibition cost. In this model, the Condition factor included two levels (fixed, alternating). Finally, switching was evaluated by comparing model0 = RT ~ (1 | ID) + (1 | Block) and model1 = RT ~ tACS + (1 | ID) + (1 | Block), using anova() function. Notably, switching effect was only present in the dissociative response type and therefore the Condition term (associative, dissociative) was not included. The LMEM syntax is listed in see Table S5. RT ~ tACS + (1 | ID) + (1 | Block:Condition) athe basic measures were: category, associative fixed, associative alternating, dissociative fixed, and dissociative alternating RTs; bfor category RT, the interaction of tACS and serial position was assessed as RT ~ tACS * Half + Block + (1 | ID). cthe derived measures were: response initiation, inhibition cost, and switching cost. The main factor(s) of interest are marked in bold.
Computation of the effect size. For the anti-phase versus sham contrast and in-phase versus sham contrast, effect sizes were calculated as dRM = Mdiff / SDdif, where Mdiff is the sample mean change (average RTs in the respective active condition minus average RTs in the sham condition) and SDdiff represents the sample standard deviation of change scores 1 . Notably, winsorized means and standard deviation (10% quantile two-sided trimming) were used in order to obtain robust effect size estimates 2 .

Supplementary discussion
Targeting inter-regional phase synchronization using in-phase and anti-phase montages has been a critical concern in a number of previous studies using tACS [3][4][5][6][7][8] . The main issue is that stimulating in-phase requires a reference electrode, which introduces an additional field polarization within the underneath and nearby cortical regions. As a consequence, the in-phase and anti-phase usually yield more or less unbalanced electric field distributions, which may confound or interact with the phase-specific tACS effects.
On the other hand, however, several lines of recent evidence concur that neurobiological and cognitive effects of tES depend on the brain activation pattern that occurs during the stimulation rather than the distribution of induced electrical fields. In fact, it has been proposed and repeatedly verified that tES preferentially modulates functional networks that are engaged by the task performed during the stimulation 10,11 (see Bikson & Rahman, 2013 for a short review). As an important example, in a recent study, Pisoni et al. (2018) have provided evidence that even though the currents delivered using tDCS spread far from the stimulation sites (as assessed by field density models), the functional effects on neuronal processing and cognition are restricted to those areas which are engaged during the stimulation. In a very similar fashion, Violante et al. (2017) demonstrated that tACS-induced brain activity and connectivity effects were restricted to taskrelated functional brain network rather than the precise placement of electrodes. Notably, as in the case of our study, Violante and colleagues used an extra reference electrode to achieve in-phase stimulation, but the cortical areas underneath and nearby this site were not affected by the stimulation (as revealed by fMRI), despite the fact that these cortical areas received the strongest field polarization (i.e., the sum of the currents from the two in-phase electrodes). Therefore, in line with the evidence from these studies, although the field polarization of the brain in the in-phase and the anti-phase conditions were not completely overlapping, the minor differences should not account for the cognitive effects observed in our study. (Please also note that from the abovementioned studies also imply that even a single electrode montage may produce dissimilar pattern of excitability and/or connectivity if the brain is engaged by distinct cognitive tasks).
Furthermore, as recently emphasized by Saturnino and colleagues (2017), the unwanted field polarization near the reference electrode may represent a confound only if it overlaps with the functional network that implement the dependent behavioral measure. Since semantic memory retrieval is almost exclusively supported by a left-lateralized brain network, the central scalp was selected as the most appropriate site for the reference electrode in our study in order to minimize the relevance of the additional unwanted polarization.
Finally, it has been proposed that a dual-site high-definition tACS (HD-tACS) may represent a more appropriate solution for targeting inter-regional phase synchronization 13 (note that high-definition montages usually include 4x1 electrodes, where 4 outer electrodes surround 1 central electrode that has the opposite polarity/phase; dual-site HD montages have also been used for tDCS 14 ). The main reason is that by using dual-site HD-tACS it is possible to maintain the same electric field distributions for both the in-phase and the anti-phase condition and thus unambiguously interpret the phasic manipulation. However, as acknowledged by Saturnino et al. (2017), this applies insofar the distances among the outer electrodes of the two respective HD montages are sufficiently large (crucial is the smallest distance between the two outer electrode "rings"). Otherwise, provided that the polarity of the outer-ring electrodes is opposite (i.e., antiphase), the cortical areas between the two HD-tACS sites will be considerably polarized. Importantly, if the electrodes are close and have the same phase (i.e., in-phase), no such polarization would occur, resulting in slightly dissimilar field distributions between the conditions. Nevertheless, an optimized dual-site HD-tACS may be a preferable solution for studies targeting phase synchronization of more distant brain regions 15 . On the other hand, however, the dual-site HD-tACS solution may be less efficient for modulating white matter connections that mediate such inter-regional brain interactions. Further research is required to optimize tACS protocols to provide unambiguous manipulation of cortical phase-relations. Fig. S1. In-phase cable splitter. For the in-phase stimulation one electrode cable was split into two channels with 0° relative phase. One fixed and two variable resistors (rheostats, one for coarse and the second for fine-grained adjustments) were serially connected with both channels. The difference in voltage between the channels was indicated by a galvanometer, which could be adjusted to ensure even current intensity between the two in-phase channels. Please note that the circuit includes only passive elements and thus in no way increases the safety or tolerability risks of the stimulation. In both conditions, two 5x5 cm 2 electrodes were located over F3 and CP5 of the international 10-10 system of EEG electrode placement. The inphase condition included an additional 5x7 cm 2 reference electrode centered between Cz and CPz. The figure depicts left-lateral and dorsal view of the estimated field intensity.

Anti-phase tACS
In-phase tACS