The context in which sudden fearful events occur can be poorly encoded into memory. Yet, the consequences of the resulting context-impoverished memories remain unknown. We demonstrate that restricting the time available for context encoding during contextual fear conditioning causes maladaptively overgeneralized and inextinguishable fear. However, post-conditioning context exposure enables further context encoding through hippocampal reconsolidation-dependent memory updating. Updating in the conditioning context alleviates overgeneralization and restores capacity for extinction. However, updating in a similar safe context erroneously shifts fear from the dangerous to the safe context. We argue that these phenomena can be explained by uncertainty about where events occurred. Moreover, we show that a hippocampal-neocortical neurocomputational model based on this assumption successfully simulates and explains our observations. These findings reveal that context-impoverished memories are maladaptive and can be improved or distorted after recall, with implications for basic memory theory, memory distortion, and treatment of disorders like post-traumatic stress disorder.

Memories are rarely acquired under ideal conditions, rendering them vulnerable to profound omissions, errors, and ambiguities. Consistent with this, recent work using context fear conditioning has shown that memories formed after inadequate learning time display a variety of maladaptive properties, including overgeneralization to similar contexts. However, the neuronal basis of such poor learning and memory imprecision remains unknown. Using c-fos to track neuronal activity in male mice, we examined how these learning-dependent changes in context fear memory precision are encoded in hippocampal ensembles. We found that the total number of c-fos-encoding cells did not correspond with learning history but instead more closely reflected the length of the session immediately preceding c-fos measurement. However, using a c-fos-driven tagging method (TRAP2 mouse line), we found that the degree of learning and memory specificity corresponded with neuronal activity in a subset of dentate gyrus cells that were active during both learning and recall. Comprehensive memories acquired after longer learning intervals were associated with more double-labeled cells. These were preferentially reactivated in the conditioning context compared with a similar context, paralleling behavioral discrimination. Conversely, impoverished memories acquired after shorter learning intervals were associated with fewer double-labeled cells. These were reactivated equally in both contexts, corresponding with overgeneralization. Together, these findings provide two surprising conclusions. First, engram size varies with learning. Second, larger engrams support better neuronal and behavioral discrimination. These findings are incorporated into a model that describes how neuronal activity is influenced by previous learning and present experience, thus driving behavior.SIGNIFICANCE STATEMENT Memories are not always formed under ideal circumstances. This is especially true in traumatic situations, such as car accidents, where individuals have insufficient time to process what happened around them. Such memories have the potential to overgeneralize to irrelevant situations, producing inappropriate fear and contributing to disorders, such as post-traumatic stress disorder. However, it is unknown how such poorly formed fear memories are encoded within the brain. We find that restricting learning time results in fear memories that are encoded by fewer hippocampal cells. Moreover, these fewer cells are inappropriately reactivated in both dangerous and safe contexts. These findings suggest that fear memories formed at brief periods overgeneralize because they lack the detail-rich information necessary to support neuronal discrimination.