Contrary to our hypothesis, significant interactions between stimulation target and fear generalization did not emerge for any index besides avoidance, in which HNT-TMS produced a less linear (steeper) avoidance response gradient. However, this effect was driven by non-significant HNT-TMS driven avoidance decreases and increases in different aspects of the fear continuum, precluding clear conclusions about the effect of HNT-TMS on avoidance generalization. Significant effects of HNT-TMS on fear generalization did emerge in analyses that assessed interactions between TMS site, fear generalization, and non-associative fear responding. These analyses revealed that HNT-TMS yielded steeper, less linear generalization gradients, indicative of reduced generalization (i.e., increased discrimination), as non-associative fear levels diminished. This interaction was detected in both self-report indices (anxiety and risk ratings). Follow-up analyses homing in on specific aspects of the generalization gradient showed that HNT-TMS yielded a steeper decline in anxiety and risk ratings from the CS + to its closest approximation (GS3), and the GS3 to its closest approximation (GS2), as non-associative fear levels declined. This effect did not emerge for control stimulation. Complementing this finding, HNT-TMS (but not vertex TMS) promoted a steeper decline in anxiety and risk ratings from the CS + to the GS3 as associative fear learning increased (i.e., greater CS + vs. CS- acquisition differential). Taken together, these findings suggest that HNT-TMS selectively boosted discrimination in participants showing a stronger associative versus non-associative fear response pattern.
The hippocampus’ schematic matching function (i.e., 11,12) plays a critical role in current models of conditioned fear generalization (6, 8). According to these models, the hippocampus matches cues that resemble the CS+ (i.e., GSs) against the stored CS + representation, triggering pattern separation (fear discrimination) to cues with insufficient CS + similarity. Hippocampal pattern separation is most taxed for cues with high similarity to the CS+. In contrast, cues with no perceptual association to the CS+ (i.e., CS-o) are not predicted to engage the hippocampal schematic matching function. Reactivity to such stimuli is thought to be driven by non-associative fear sensitization rather than CS + schematic matching (i.e., pattern completion). Consistent with schematic matching models of fear generalization, HNT-TMS effects emerged when the schematic matching process was putatively most actively engaged, whereby HNT-TMS yielded stronger discrimination of the CS + from its closest approximation when associative learning was more robust and non-associative fear levels were lower. This finding may have clinical import, as excessive responsivity to the closest CS + approximation has been found in anxiety-related disorders (panic disorder: 65; GAD: 66, but see 67; and PTSD: 68) making it a target with high potential translational value.
Anxiety-related disorders are heterogeneous and may be driven by a variety of associative and non-associative learning abnormalities with partially distinct etiologies (68). Although the neurobiology of sensitization has received less attention than associative learning processes (for recent review, see: 69) sensitization is thought to result from a stressor induced cascade of corticotropin releasing factor and norepinephrine, which heightens arousal and potentiates rapid defensive circuits (i.e., amygdala: 70,71). Such potentiation putatively biases processing toward these circuits and away from the hippocampus (72). Consistent with this account, heightened arousal has been linked to reduced hippocampal activation during fear processing in PTSD (73, 74). In the present study, sensitization-related arousal increases may have blunted the effects of HNT-TMS by reducing engagement of the hippocampal schematic matching function. Alternatively, arousal increases may have directly impinged on the hippocampal schematic matching function. Indeed, our updated neural account of fear generalization posits that overgeneralization is driven in part by locus coeruleus mediated noradrenergic transmission (8), which biases the hippocampus toward retrieval of the conditioned fear memory (i.e., pattern completion) by increasing neural gain in CA3/CA1 circuits (75–77). HNT-TMS – or at least one session of HNT-TMS – may have been insufficient to counter this effect. These intriguing possibilities merit investigation in future studies.
Discordance between defensive response indices is a well replicated finding (78, 79) potentially linked to divergence in the neural circuits that undergird subjective fear, physiological threat, and avoidance responses (80, 81) or the unique sources of error associated with specific measures. In the present study, HNT-TMS reduced subjective anxiety and risk ratings as a function of non-associative anxiety/risk levels but did not yield the same effect on physiological threat responsivity (fear potentiated startle, FPS) or avoidance. For FPS, there was no interaction between stimulation target and conditioned FPS responses; nor was there an interaction between stimulation target, conditioned FPS responses, and non-associative FPS responses. In contrast, an interaction between stimulation target and avoidance responses emerged, whereby HNT-TMS yielded steeper, less linear avoidance response gradients. Although decreases in linearity can indicate decreased generalization, linearity decreases in the present study emerged due to a combination of (non-significantly) decreased avoidance of the closest CS + approximation (GS3) and (non-significantly) increased avoidance of cues with low CS + resemblance (GS1, CSm) following HNT versus control stimulation. This mixed result precludes straightforward conclusions regarding the effect of HNT-TMS on avoidance.
Despite cross-species evidence implicating the hippocampus (particularly the dentate gyrus sub-region) in pattern separation (15, 82, 83), HNT-TMS did not modulate non-affective pattern separation as measured by the MST in the present study. One interpretation of this result is that the human hippocampus does not mediate pattern separation, a claim recently made by Quiroga (2020) (84). However, this claim has been challenged by others (85, 86) and is inconsistent with evidence showing that MST pattern separation performance is linked to hippocampal activation/connectivity (87, 88), and disrupted by hippocampal lesions (89, 90) and disorders that impair hippocampal integrity such as Alzheimer’s disease (91). Another possibility is that HNT-TMS has a stronger effect on the encoding of associations versus individual stimuli. Indeed, previous studies have shown that HNT-TMS strengthens recollection of associative memory (i.e., object-location/object-scene) without affecting memory for individual stimuli (26, 30). While associations between stimuli and outcomes (shock/no-shock) are encoded during the Farmer Task, individual stimuli are encoded during the MST. Therefore, differences in the memory processes indexed by these tasks may explain why HNT-TMS modulated affective pattern separation as a function of sensitization without affecting neutral pattern separation.
The present study has several strengths including its use of relatively dense sampling which facilitated individualized, precision targeted TMS. However, the study also has noteworthy limitations. For example, although our sample-size is comparable to previous fear-based (92–94) and neutral memory (25–27, 29–32, 95) hippocampal TMS studies, future larger sample studies are necessary to replicate the present results. Additionally, the lack of a post-stimulation neuroimaging component precluded investigation of the neural effects of stimulation. Future studies are therefore necessary to confirm hippocampal engagement and investigate links between neural and behavioral effects of HNT-TMS. Finally, although participants in the present study had PTSS, many did not meet full PTSD criteria (62.5% met PTSD criteria). Given that TMS has been shown to produce different effects on stimulated circuits in individuals with and without psychiatric disorders (96–98), it is important that future studies assess the effects of HNT-TMS in PTSD samples.
In conclusion, our results indicate that HNT-TMS may selectively sharpen hippocampal mediated fear discrimination when the hippocampal schematic matching process is most engaged by highly similar threat-like cues and associative versus non-associative learning patterns are relatively strong. If replicated in larger clinical samples, these findings could motivate investigation into the clinical potential of HNT-TMS as a mechanism-based treatment for a subset of patients with anxiety-related disorders, consistent with contemporary biomarker based stratification approaches (99, 100).