Direct Thalamic Inputs to Hippocampal CA1 Transmit a Signal That Suppresses Ongoing Contextual Fear Memory Retrieval

SUMMARY Memory retrieval of fearful experiences is essential for survival but can be maladaptive if not appropriately suppressed. Fear memories can be acquired through contextual fear conditioning (CFC) which relies on the hippocampus. The thalamic subregion Nucleus Reuniens (NR) is necessary for contextual fear extinction and strongly projects to hippocampal subregion CA1. However, the NR-CA1 pathway has not been investigated during behavior, leaving unknown its role in contextual fear memory retrieval. We implement a novel head-restrained virtual reality CFC paradigm and show that inactivation of the NR-CA1 pathway prolongs fearful freezing epochs, induces fear generalization, and delays extinction. We use in vivo sub-cellular imaging to specifically record NR-axons innervating CA1 before and after CFC. We find NR-axons become selectively tuned to freezing only after CFC, and this activity is well-predicted by an encoding model. We conclude that the NR-CA1 pathway actively suppresses fear responses by disrupting ongoing hippocampal-dependent contextual fear memory retrieval.

inactivation of the NR-CA1 pathway prolongs fearful freezing epochs, induces fear 23 generalization, and delays extinction. We use in vivo sub-cellular imaging to specifically 24 record NR-axons innervating CA1 before and after CFC. We find NR-axons become 25 selectively tuned to freezing only after CFC, and this activity is well-predicted by an 26 encoding model. We conclude that the NR-CA1 pathway actively suppresses fear 27 responses by disrupting ongoing hippocampal-dependent contextual fear memory 28 retrieval.

32
Flexibly encoding and retrieving memories of fearful events is a critically conserved 33 survival behavior, as a single failure can be deadly. However, failing to suppress 34 inappropriate fear responses can also have devastating consequences, manifesting as 35 negative affective states in generalized anxiety disorder and post-traumatic stress 36 disorder 1,2 . One way in which fear memories can be studied in the laboratory is through 37 contextual fear conditioning (CFC), in which a spatial context, the conditioned stimulus 38 (CS), is repeatedly paired with a noxious unconditioned stimulus (US), generally a mild 39 shock 3-56 . Freezing is a species-specific fear response, and a quantifiable readout of contextual fear memory retrieval (CFMR) of the learned association 7,8 . With continued 41 exposure to the CS in the absence of the US, freezing generally decreases and 42 exploratory behavior increases -a process termed fear extinction. Fear extinction occurs 43 as animals learn over time that the context no longer predicts shocks 9-12 . For extinction 44 to occur, mice must therefore suppress CFMR during each fearful freezing epoch to avoid 45 excessive freezing, which would be detrimental to survival. Therefore, mechanisms must 46 exist in the brain to suppress CFMR as it is occurring. 47 48 Contextual fear in both mice and humans relies on coordinated brain regions including 49 the Medial Prefrontal Cortex (mPFC), Thalamus, Amygdala, and Hippocampus 13 . The 50 contextual component of these memories relies on the hippocampus, which retrieves and 51 updates contextual fear memories 6,14-19 . Experimental inhibition of a subset of 52 hippocampal neurons tagged using immediate early genes active during CFC is sufficient 53 to suppress CFMR [20][21][22] . This suggests that natural suppression of ongoing CFMR must 54 involve a circuit that can modulate hippocampal activity. 55 56 One potential source of this modulation is the ventral midline thalamic subregion, Nucleus 57 Reuniens (NR). Sometimes termed 'limbic thalamus' for its diverse set of inputs from The mPFC-NR projection and NR itself is necessary for both fear extinction and for 65 preventing fear generalization to a neutral context, a process in which mice fail to form 66 context-specific memory and additionally associate a non-shocked context with fear 29-35 . 67 NR stimulation reduces contextual fear-induced immediate early gene expression in both 68 mPFC and CA1 35,36 . While the roles of the mPFC-NR pathway and NR itself have been 69 explored during CFMR, the role of the NR-CA1 pathway is unknown. We hypothesize that 70 NR transmits a signal to CA1 to suppress ongoing CFMR, thereby reducing fear 71 responses (freezing) and promoting exploratory behavior (movement).

73
To test our hypothesis, we used a chemogenetic approach to directly inhibit the NR-CA1 74 pathway, and 2-photon calcium imaging in head-restrained mice to record NR-axons in in the control context, and returned to near baseline in the shocked context and baseline 122 in the control context (Extended Data Fig. 2B: day 3). 123 124 While freezing quantity differed between contexts and days of retrieval, freezing position 125 was distributed evenly across all track locations in both contexts on all retrieval days. This 126 shows that mice associated fear with the entire context, and not specific locations along 127 the track or near specific objects in VR (Extended Data Fig. 1C). As an additional control, 128 a separate group of mice went through the same process but were never shocked in 129 either context. These mice froze significantly less on retrieval days 1-3 (on average 11.6 130 ± 8.7%) without any significant differences to freezing on day 0 or between contexts ( Fig.   131 1F, Extended Data Fig. 2H). These spontaneous freezing events in both the control 132 condition and the pre-shocked contexts (i.e. before the delivery of any shocks) could 133 potentially be caused by the lack of water reinforcement, the presence of the tail coat 134 itself, or an unrelated temporary disinterest in running, and provide a within-mouse 135 comparison to post-shock fear-evoked freezing. 136 137 To further quantify freezing behavior, we measured the duration of each individual 138 freezing event (freezing epoch) and found that freezing epochs were longer in the 139 shocked versus the control context on retrieval day 1 (Fig. 1G). Freezing epochs 140 remained longer on day 2 in the shocked compared to the control context, however, they To test the involvement of NR-CA1 projecting neurons in CFMR, we designed a designer 150 receptor exclusively activated by designer drugs (DREADD) based inhibition paradigm 41-151 43 (Fig. 1b). We injected a Cre-expressing virus bilaterally in NR, and a retrograde Cre-152 dependent virus carrying the inhibitory G(i)-coupled DREADD receptor, hM4Di-DREADD, 153 bilaterally in the SLM of dorsal CA1 where hippocampal-projecting NR-axons terminate 154 ( Fig. 2A) 44 . This enabled us to intraperitoneally (IP) inject the hM4Di agonist, 155 deschloroclozapine dihydrochloride (DCZ) 43 , before the first post-shock re-exposure to 156 the contexts on retrieval day 1, therefore selectively inhibiting a subset of NR-CA1 and received saline instead of DCZ on retrieval day 1 (Extended Data Fig. 3A). In both 162 control groups (Extended Data Fig. 3A), freezing behavior was similar to the experimental 163 mice shown in Fig. 1D, and the groups were thus combined and termed the NR-CA1 164 intact group for further analysis. We then compared the behavioral impact of inhibiting the 165 NR-CA1 pathway on day 1, and on subsequent retrieval days 2 and 3 with the NR-CA1 166 intact group (Fig. 1B Bottom). 167 168 We found that in the shocked context on retrieval day 1, NR-CA1 inhibited mice (N = 5) 169 spent ~57% more time freezing than NR-CA1 intact mice (77.8% ± 12.4% versus 49.3% 170 ± 10.1%; Fig. 1D  Given these findings, we asked if mice could still discriminate well between the shocked 187 and the control context after NR-CA1 inhibition. To do so, we calculated a discrimination 188 index 29 which revealed a significant decrease in discrimination between the shocked and 189 control contexts on day 1 from 32.2% ± 6.6% with NR-CA1 intact to 18.1% ± 6.3% with 190 NR-CA1 inhibited (Extended Data Fig. 4B, P = 0.0034), suggesting that inhibition of the 191 NR-CA1 pathway reduces fear-induced contextual discrimination. significantly different from pre-shock levels in either context (Extended Data Fig. 3C), 202 indicating successful fear extinction. Thus, the absence of the NR-CA1 input on retrieval 203 day 1 caused an increase in CFMR on day 2 in the shocked context, reducing fear 204 extinction, but reinstatement of the NR-CA1 pathway on day 2 allowed extinction to occur 205 on day 3.

207
We wanted to ensure that the DCZ-induced increase in freezing was not due to a general 208 decrease in movement. To do so, we exposed NR-CA1 inhibited and intact mice to a 209 'dark' context (devoid of any visual cues) for ~5 minutes after they were exposed to both 210 the shocked and control contexts on retrieval days 1-3. In this dark context, mice quickly 211 recovered their running behavior, with NR-CA1 inhibited mice freezing on average only 212 4.4 ± 4.0% of the time across all 3 days of retrieval. Both within and across-mice controls 213 froze at comparably low levels (on average under 5%; Extended Data Fig. 3B). Therefore, 214 neither DREADD inhibition, nor DCZ itself, impacted the mouse's ability to move, and the 215 increase in freezing behavior is therefore specific to when mice are navigating in VR 216 contexts. This indicates that the increase in post-shock freezing that we observe in the 217 control context in both NR-CA1 intact and inhibited mice over baseline could not be due 218 to an overall decrease in movement, but is specific to the VR context. It additionally 219 indicates that the increase in fear generalization to the control context in NR-CA1 inhibited 220 mice is due exclusively to NR-CA1 pathway inhibition. Our results indicate that the NR-221 CA1 pathway sends a potent fear suppression signal, critical for shortening the length of 222 freezing epochs, preventing fear generalization, and inducing contextual fear extinction. We limited our analysis to a putative single axon per animal, since all identified axonal 242 segments within the field of view with above-baseline activity were highly correlated (see 243 Method Details). We were additionally able to track a subset of the same NR-axons (N = 244 4) across days (Extended Data Fig. 5A). 245 246 We found that NR axons switched their activity from untuned sparse activity ( Fig. 2B; Top) 247 pre-shocks, to activity highly selective for freezing epochs post-shocks, even after filtering 248 for axons with detectable pre-shock activity ( Third, we characterized the dynamics of NR-axon activity within each freezing epoch on 284 pre-shocks day 0 and compared to retrieval day 1. To do so we aligned NR-axons by 285 dividing each freezing or a running epoch into 5 even bins, each containing a mean 286 normalized Δf/f of NR-axon peaks, then took the within-bin mean across all epochs pre 287 and post-shocks. This enabled us to effectively 'stretch' or 'shrink' all epoch lengths to a 288 uniform standard. Using this method, we found that mean axon activity ramped up rapidly 289 in the beginning of a freezing epoch, plateaued, then fell right before freezing transitioned 290 to running (Fig. 2E). Such temporal dynamics were absent during the freezing epochs 291 pre-shocks (Fig. 2E). These dynamics were all similarly observed in axons tracked across 292 days (Extended Data Fig. 5B-E). These results collectively show that NR-axons projecting 293 to CA1 strongly tune their activity to fearful freezing epochs during CFMR, and this post-294 shock activity is context-independent. Because there was variability in the fluorescence signal recorded from the axons, we 315 checked whether model accuracy was related to the signal-to-noise. Indeed, model 316 accuracy was correlated with axon activity -the greater the change in the normalized 317 fluorescence signal from baseline, the better the model performed (Fig. 3B). The model 318 performed significantly above chance in predicting NR-axon signal in 8/10 mice, on 319 retrieval days 1-3. In 2/10 mice, model prediction was poor on retrieval days, due to lower 320 signal-to-noise ratio (SNR) of the fluorescence signal. However, changes in SNR did not account for the poor model performance pre-shock, as model accuracy was still low in 322 animals with higher axon activity. Although overall activity was higher in post-shock days, 323 pre-shock activity in longitudinally-tracked axons reached similar peak heights as in post- 324 shock days (Extended Data Fig. 4A), and all mice included in analysis had at least 2 325 peaks reaching a minimum of 0.1 Δf/f in the recording session, ensuring that poor model 326 performance was not simply due to a lack of signal to predict. In summary, using an 327 encoding model, we demonstrated that NR-axon activity recorded in hippocampal CA1 328 can be predicted from freezing behavior during CFMR, but not before the animal is fear-329 conditioned, revealing the development of predictable structure in NR-axon activity tuned 330 to CFMR. Our findings expand on a previous canon of work that indicates both the mPFC-NR 335 projection and NR itself are required for contextual fear extinction and preventing fear 336 overgeneralization 29,33-35 . Our results suggest that in addition to these roles, NR reduces 337 time spent freezing following CFC by suppressing CFMR as it is occurring during freezing 338 epochs. We found that the NR-CA1 pathway is a key component of the circuit responsible 339 for mediating the fear suppressive function of NR. This is supported by our observation 340 that NR axons in CA1 become selectively tuned to freezing epochs following CFC and 341 inhibiting the NR-CA1 pathway lengthens freezing epochs. The function of the NR-CA1 342 pathway in CFMR suppression is not restricted to the context in which shocks were 343 presented, but extends to similar contexts where shocks never occurred. This seems to 344 limit overgeneralization as shown by NR-CA1 inhibition reducing context discrimination. 345 Lastly, the process of suppressing ongoing CFMR by the NR-CA1 pathway also has 346 longer term effects, as shown by reduced fear extinction a day following NR-CA1 347 inhibition. In summary, our observations support a framework in which the NR-CA1 348 pathway actively suppresses fear responses by disrupting ongoing hippocampal-349 dependent CFMR to promote non-fearful behavior, and this process also limits 350 overgeneralization and promotes fear extinction.

352
Interestingly, we did not observe a significant difference in NR-axon activity during CFMR 353 between contexts on retrieval days, despite NR-CA1 inactivation reducing discrimination 354 between these contexts. It could be that while the NR-axons in CA1 are not contextually 355 modulated, their activity induces postsynaptic dynamics in CA1 that encode differences 356 in context. This is supported by previous work showing that CA1 is specifically necessary 357 for the context-dependence of fear extinction 52 . We also found the difference between 358 NR-axonal activity in freezing and running epochs following CFC does not decrease over 359 days, even as mice decrease their time spent freezing. Previous work shows that 360 extinction does not erase previously-learned contextual fear memories, as reactivation of hippocampal fear memories rapidly reinduces fear behavior 21,53 . This suggests that fear 362 memories are retained but are dormant after extinction. Continued differential activity of 363 NR-axons between freezing and running epochs in CA1, even after extinction, may be 364 necessary to prevent the maladaptive retrieval of dormant fear memories, therefore 365 enabling successful extinction learning. The input driving the NR-CA1 pathway is most likely from the mPFC, encompassing both 387 the prelimbic (PL) and infralimbic (IL) regions. While PL is needed for fear acquisition and 388 retrieval, IL is necessary for the opposing task of fear suppression and preventing 389 overgeneralization 56-59 . The likely opposing influences of IL and PL on NR during CFMR 390 illustrates the importance of understanding NR output pathways. Our results indicate that 391 a fear suppression signal circuit may be transmitted from IL, through NR, and into CA1 392 during CFMR. Of note, a small population of NR neurons that project both to CA1 and 393 either PL or IL may have a key role in facilitating cross-regional theta synchrony 394 associated with CFMR 28,60 . While we cannot rule out that some of our recorded NR-axons 395 collaterally project to mPFC, since this population makes up a small subset of all NR 396 neurons (~3-9% 60 ), we would expect the majority of our recordings to be from non-dual 397 projecting neurons. It additionally remains to be seen if the NR-CA1 exclusively projecting 398 versus the NR-CA1 dual projecting populations have distinct dynamics during CFMR. 399 A key question that arises from our work is how the NR-CA1 pathway potentially disrupts Abuse awarded to S.K. We thank the University of Chicago imaging core for assistance 450 with confocal imaging, and the University of Chicago animal care staff for ensuring the 451 well-being of experimental animals. We thank Chad Heer for early help with imaging 452 protocols. We thank Valerie Barreto for helping to train animals and collect confocal data. 453 We thank Timothy Ratigan for assistance with data analysis. We thank Rossten Rad for  470 One or more of the authors of this paper self-identifies as an underrepresented ethnic 471 minority in science. One or more of the authors of this paper self-identifies as a member 472 of the LGBTQ+ community. While citing references scientifically relevant for this work, we 473 also actively worked to promote gender balance in our reference list. We support 474 inclusive, diverse, and equitable conduct of research.      Once mice met training criteria, they were habituated to the injection process. They were   Fig 1b. Fig 2a, Fig 3a, and Extended Data Fig 1a), some figure text, and figure layouts 875 were made with BioRender (https://biorender.com/).