Medial prefrontal cortex stimulation abolishes implicit reactions to threats and prevents the return of fear

Down-regulating emotional overreactions toward threats is fundamental for developing treatments for anxiety and post-traumatic disorders. The prefrontal cortex (PFC) is critical for top-down modulatory processes, and despite previous studies adopting repetitive Transcranial Magnetic Stimulation (rTMS) over this region provided encouraging results in enhancing extinction, no studies have hitherto explored the effects of stimulating the medial PFC (mPFC) on threat memory and generalization. Here we showed that rTMS applied before threat memory retrieval abolishes implicit reactions to learned and novel stimuli in humans. These effects were not due to inhibition of electrodermal reactivity and enduringly persisted one week later in the absence of rTMS. No effects were detected on explicit recognition. Critically, we observed stronger attenuation of defensive responses in subjects stimulated over the mPFC than the dlPFC. Our findings uncover a prefrontal region whose modulation can permanently hamper implicit reactions to learned dangers, representing an advance to long-term deactivating overreactions to threats.


28
Emotional memories related to past threat experiences allow humans to predict future dangers and 29 trigger adaptive defensive reactions when encountering learned threat-signaling cues 1 . However, 30 extremely dangerous situations may represent points of origin for trauma and lead to maladaptive 31 behaviors and psychological disorders 2 . In this scenario, intrusive traumatic memories may be 32 explained by over-time enduring conditioned responses to trauma reminders 3 . Attempting to down- underlying threat memory processes through non-invasive brain stimulation 6 . 41 Previous studies adopted transcranial direct current stimulation (tDCS) or transcranial electrical Results 71 72 mPFC-focused rTMS effects on implicit reactions toward threat-predictive cues 73 To explore the effects of an mPFC-centered rTMS on the defensive responses to a learned threat, we 74 designed a three-session experiment starting with a threat learning followed by an implicit retention 75 test and a follow-up implicit re-test ( Figure 1). 76 During the conditioning session, participants learned to associate an auditory cue (conditioned 77 stimulus, CS, 800Hz) with a mild electric stimulation (unconditioned stimulus, US, individually 78 calibrated intensity) in a given environment (context A). We adopted a single-cue conditioning 79 paradigm because it more ecologically reflects real-life traumatic experiences [26][27][28][29][30] . To validate the 80 between-groups homogeneity in the painful stimuli perception, we computed a one-way ANOVA on 81 post-conditioning US ratings, and we observed no significant differences amongst groups (F(2,60) = 82 0.6108, P > 0.05, ηp 2 = 0.01995) ( Table 1). We also did not observe significant differences amongst 83 groups in SCRs to the CS during the preconditioning phase (F(2,60) = 1.539, P > 0.05, ηp 2 = 0.04879) 84 nor to the US during the conditioning phase (F(2,60) = 1.237, P > 0.05, ηp 2 = 0.03961) (Supplementary 85 Figure 1). 86 One week later, we tested the implicit memory of the learned association in control subjects 87 and in those that received rTMS over the mPFC. To locate this brain region, which corresponds to 88 the Brodmann area 10 31 , we positioned the coil over Fpz adopting the 10-20 EEG coordinate system, 89 since previous rTMS studies 15,32,33 ensured this placement reached the mPFC. An offline 10-min 90 session of 1Hz-rTMS targeting this neural site (mPFC, n = 21) was applied immediately before 91 memory retrieval. Besides proximal inhibitory effects on local brain activity, 1Hz-rTMS protocols 92 have been shown to induce downstream distal effects through the potentiation of resting-state 93 functional connectivity with remote brain areas inside the stimulated neural network 34 . When applied 94 over prefrontal regions, 1Hz-rTMS seems to increase regional cerebral blood flow (rCBF) 35,36 . 95 Control subjects underwent a 10-min sham stimulation procedure over the same cortical area (sham, 96 6 n = 21). To ascertain the topographical selectivity, in one further condition (OC, n = 21) we applied 97 the rTMS over the left occipital cortex as an active control site. 98 Memory retention was tested in a different environment from that where the learning had 99 occurred (context B) to avoid any contextual influence on retrieval [37][38][39][40][41] . Indeed, the context shift for 100 this session mirrors a real-life treatment setting -which unlikely takes place in the threatening 101 location. To test implicit threat memory, we performed an implicit recognition task in which subjects 102 were exposed to the CS while being recorded in their evoked autonomic reactions (i.e., electrodermal 103 skin conductance responses, SCRs). No US shocks were delivered during this phase. Besides the CS, 104 participants were presented with two novel but perceptually similar tones (NS1, 1000Hz; NS2, 600Hz) 105 to study threat generalization. Auditory frequencies of NSs were selected to obtain a slowly decaying 106 gradient of defensive tunings 39,42,43 . A one-way ANOVA (F(2,60) = 8.793, P < 0.001, ηp 2 = 0.2267) 107 revealed that subjects that received rTMS over the mPFC exhibited weakened SCRs than those 108 observed in the sham (P < 0.001) and OC (P = 0.0074) groups, where the CS evoked similarly strong 109 autonomous reactions (P > 0.05) ( Figure 2A). These data were obtained by averaging all four CS 110 trials. To test potential differences in the genesis of each CS-related response, we performed a trial-111 by-trial analysis through a 3 × 4 mixed ANOVA. This analysis revealed a significant main effect of 112 group (F(2,60) = 8.793, P < 0.001, ηp 2 = 0.2267), a significant main effect of trial (F(3,180) = 12.328, P < 113 0.001, ηp 2 = 0.17) and a not significant group × trial interaction (F(6,180) = 1.155, P > 0.05, ηp 2 = 0.037).

114
Notably, we observed that in Trial 1 the mPFC group already exhibited weaker reactions than sham 115 (P = 0.002) and OC (P = 0.046) conditions, which did not differ from each other (P > 0.05) ( Figure   116 2B). This data indicates that the rTMS procedure immediately affected SCRs triggered by memory 117 retrieval, rather than interfering with reactivation-induced reconsolidation processes, as previously 118 reported 17 . To the best of our knowledge, this is the first evidence that brain stimulation may promptly 119 attenuate implicit defensive reactions during memory retrieval, without requiring repeated exposure 120 to the CS (i.e., extinction learning). In the same session, we also analyzed threat generalization to the NSs, and we found that the 122 mPFC group displayed attenuated responses to the NS1 relative to the other two conditions, which did 123 not differ between each other (F(2,60) = 5.856, P = 0.0048, ηp 2 = 0.1633; mPFC vs sham: P = 0.0051; 124 mPFC vs OC: P = 0.0483; sham vs OC: P > 0.05) ( Figure 2C). Autonomous responses to the NS2 125 (F(2,60) = 5.20, P = 0.0083, ηp 2 = 0.1477) were reduced in the mPFC group relative to the sham group 126 (P = 0.0091) but not relative to the OC group (P = 0.067), whereas sham and OC groups were similar 127 (P > 0.05) ( Figure 2D). We then explored fear tunings (i.e. the decay of defensive responses typically  We next sought to disambiguate whether the rTMS effects were due to a general down-138 regulation of electrodermal responsivity, or whether they specifically targeted the threat memory. To 139 this end, subjects were presented with an unconditioned threatening stimulus consisting of a female 140 scream sample (unconditioned stimulus 2, US2) while being recorded in their SCRs. A one-way 141 ANOVA revealed no differences amongst conditions (F(2,60) = 2.045, P > 0.05, ηp 2 = 0.06381), 142 indicating that the rTMS did not cause an overall inhibition of electrodermal reactivity ( Figure 2E). To test whether and to what extent rTMS-related outcomes endured beyond the short-term 146 after-effect window and persisted in a long-term period, we planned a follow-up session. One week 147 8 after rTMS session and threat memory retrieval test, all participants returned to the conditioning room 148 (context A) and underwent a re-testing phase, which was identical to the testing one except for the 149 absence of rTMS administration. This phase allowed us to evaluate potential long-term rTMS effects, 150 and to test a possible renewal effect since other approaches based on extinction learning are context-151 dependent 44 and here subjects were re-exposed to the original threatening environment.

152
Concerning the implicit responses to the CS, participants of the mPFC group persisted in  presented with a random sequence of tone pairs, each composed of the CS and one of the two NSs. 174 Here, subjects were asked to consciously identify and refer which stimulus of each pair was the one 175 previously paired with the US (i.e., the CS), and to provide a subjective confidence level for each 176 choice using a scale ranging from 0 (completely unsure) to 10 (completely sure) 39,45 . Explicit   Figure 3B).

184
Next, we implemented a 2AFC perceptual task in which we investigated the ability of 185 participants to sensory discriminate between the CS and the two NSs by collecting binary 'same or

207
These findings suggest that the rTMS procedure over the mPFC did not affect the long-term 208 capacity to consciously identify and perceptually discriminate the learned threat-predictive cue.  Figure 4C). 229 We found no between-groups differences in the implicit responses to the NS1 (t(40) = 1.72, P 230 > 0.05, ηp 2 = 0.06883) ( Figure 4D In this study, we found that implicit reactions to both learned and novel stimuli were reduced 246 following a 1Hz-rTMS procedure over the mPFC. recall (24h later), which was specific for the cue that had been paired with rTMS.

262
In our study, we tested both immediate and remote (one week) rTMS effects and we did not 263 include extinction training before retrieval. We observed a significant decrease in defensive reactions 264 even in the first CS trial of the test session, i.e. the first time that subjects were re-exposed to the CS 265 after the formation of the CS-US association, and this effect was maintained in the follow-up session. 266 Hence, since we found this outcome in absence of extinction training, it is likely that the rTMS 267 procedure directly modulated the defensive responses activated by the implicit threat memory trace.  To potentiate the neural activity of the PFC, both the aforementioned studies 15,16 adopted high-283 frequency rTMS protocols -which are conventionally considered excitatory of proximal brain 284 activity 54 . In our study, we adopted a low-frequency rTMS protocol -which is conventionally 285 considered inhibitory 54 . Recent evidence, however, challenged this common frequency-dependent 286 rule 55 . Indeed, resting-state functional magnetic resonance imaging (fMRI) studies demonstrated that 287 1Hz-rTMS protocols may also induce distal effects and enhance functional connectivity amongst the 288 brain regions located underneath the coil and remote brain areas of the stimulated functional 289 network 34 . Additionally, some studies 35,36 reported that 1Hz-rTMS procedures delivered over the PFC 290 may paradoxically increase local brain activity.

291
The dorsolateral PFC is another prefrontal region that is assumed to be critically involved in   Although several studies enlightened the role of the mPFC and mPFC-hippocampus 322 connectivity in emotion-related components of episodic memory 83 , we did not detect any rTMS-323 driven effect on explicit recognition memory. The observed divergence between autonomous and 324 declarative patterns might have been due to a selective rTMS action upon the neural system 325 supporting implicit threat processing, which has been widely dissociated from the neural system 326 underlying explicit memory processes 39,46-48 . Critically, an rTMS procedure that shapes implicit 327 overreactions to learned threats without affecting conscious knowledge of the same stimuli might 328 represent a strategic advantage for therapeutic applications.   Table 1 for all groups' mean State-Trait Anxiety 343 Inventory scores). Participants were then randomly assigned to each experimental condition, based 344 on sex and age (see Table 1 for all groups' mean age and sex distribution). We discarded seven     Table 1 for all 386 groups' US current intensity and US analog ratings). was set at 80% of the rMT for subjects whose rMT was ≤ 50% of the machine's maximum deliverable 397 power (e.g., the intensity corresponded to 40% of the maximum power when the rMT was equal to 398 50% of the same parameter). For subjects with an rMT > 50%, the stimulation intensity was always 399 set to a ceiling corresponding to 40% of the machine's maximum deliverable power (see Table 1 for 400 each group's mean rMT and mean stimulation intensity). During the rTMS procedure participants 401 were seated in a comfortable recliner that we adjusted to allow their upper body to be in a sloped 402 position, thus ensuring an optimal positioning of the coil.

Two-alternative forced-choice (2AFC) explicit recognition test 432
This procedure involves the presentation of two stimuli on each trial and the subject chooses the one 433 that was previously encoded (i.e. the first or the second one). As in our previous works 39,45 , a 2AFC 434 design was preferred over a new-old paradigm, which involves one single stimulus on each trial and 435 the subject judges whether the stimulus has been previously encoded (old), or whether it is new. Our 436 choice was motivated by the evidence that a 2AFC task improves recognition performance and 437 20 discourages response biases such as the familiarity-based decision bias, namely the heuristic to 438 endorse novel cues as 'old' when their familiarity is high 99 .

439
The task consisted of the presentation of 16 tone-pairs, each composed of the CS (800Hz) and one of 440 the two NSs (NS1, 1000Hz or NS2, 600Hz) in a completely random sequence: 4 × CS vs NS1, 4 × 441 NS1 vs CS, 4 × CS vs NS2, 4 × NS2 vs CS. On each trial, the two stimuli were presented with an intra-442 trial-interval of 1000ms. After each pair offset, an ITI randomly ranging between 21s and 27s 443 occurred. Participants were explained that in each couple of sounds there was a tone that they had 444 heard on the first session (one week before or, in the case of the follow-up session, two weeks before) 445 and a new tone. Participants were then instructed to recognize and verbally refer which one (the first 446 or the second) was the tone heard in the first session, paired with the US-shock (CS). Participants 447 were further asked to verbally provide a confidence rating about each response, on a scale from 0 448 (completely unsure) to 10 (completely sure). No feedback was supplied. As in the implicit task, the 449 stimulating electrode was kept attached, but no shock was delivered. completely random sequence (ITI randomly ranging between 21s and 27s). For each pair, subjects 455 were asked to refer whether the two tones were "the same tone or different tones", and to provide a 456 confidence rating on an analog scale from 0 (completely unsure) to 10 (completely sure). No feedback 457 was supplied.  Responses that did not fit these criteria were scored zero. Raw SCR data were square-root transformed  Statistical analyses 479 We computed the appropriate sample size based on a power analysis performed through G*Power  To test the between-groups differences in the explicit recognition and respective confidence ratings, 496 as well as in the perceptual discrimination and respective confidence ratings during the test and the 497 follow-up sessions, we performed Student's unpaired t tests. To test whether explicit recognition 498 levels were significantly higher than the 50% chance level for each condition during the test and the 499 follow-up sessions, we calculated Student's one sample t tests against 0.50.

500
To test the between-groups (dlPFC vs mPFC) differences in autonomic reactions to the CS and the 501 NSs during the test, and to the CS during the follow-up, we adopted Student's unpaired t tests. To 502 test potential differences in the time course of CS-related responses, we performed 2 × 4 mixed 503 ANOVAs with Group (dlPFC and mPFC) as between-subjects variable and Trial (1-4) as within-504 subjects variable. For each ANOVA we assessed the Sphericity assumption through Mauchly's Test.

505
Where it was violated, we applied the Greenhouse-Geisser correction accordingly.

506
The null hypothesis was rejected at P < 0.05 significance level. All statistical analyses were performed 507 using SPSS Statistics 22 (IBM) and Prism 9 (GraphPad). 508 23 Acknowledgements 509 We thank all the subjects for their participation in this study. We also thank Melania Lattuada,                       State-Trait Anxiety Inventory Form Y (STAI-Y) State subscale score during session 1 (S1), session 774 2 (S2), and session 3 (S3), and Trait subscale score, US current intensity (mA), post-conditioning US 775 rating, TMS resting motor threshold (rMT), and TMS power. All data are mean ± standard deviation.