Threat Learning Impairs Subsequent Associative Inference Part 1

Despite it being widely acknowledged that the most important function of memory is to facilitatethe prediction of significant events in a complex world, no studies to date have investigated how ourability to infer associations across distinct but overlapping experiences is affected by the inclusionof threat memories. To address this question, participants (n= 35) encoded neutral predictiveassociations (A→B). 

First, neutral predictive associations help improve memory. Research has found that during the learning process if people can associate neutral things with cognitive tasks, they can improve their memory. Take learning a foreign language as an example. If you associate neutral words with real-life situations during learning (such as associating "apple" with actual apples), you can better remember new words, and you can also communicate in actual situations. Retrieve information from memory faster.

Finally, neutral predictive associations can also reduce stress. In daily life, we often encounter various pressures and challenges. By predicting the appearance of neutral stimuli, people can self-regulate their emotions and reduce stress. Some studies have found that some people tend to have problems such as memory decline and insufficient decision-making ability when they are under long-term stress. Therefore, neutral predictive associations can also perform memory exercises while relieving long-term stress.

Click know 10 ways to improve memory

The following day these memories were reactivated by pairing B with a newaversive or neutral outcome (B→ ­CTHREAT/NEUTRAL) while pupil dilation was measured as an index ofemotional arousal. Then, again 1 day later, the accuracy of indirect associations (A→C?) was tested.Associative inferences involving a threat to learning memory were impaired whereas the initial memorieswere retroactively strengthened, but these effects were not moderated by pupil dilation at encoding. 

The last two decades have seen a dramatic shift in how episodic memories are understood, from rigid, passiverecords of the past to flexible, actively constructed representations in service of the future1,2. This change in conceptualization is underlined by the recent surge in studies investigating relational memory and the neurocognitive mechanisms that enable it3. The ability to recombine information across distinct but overlapping episodes referred to as associative inference, is considered a key feature of relational memory4. 

Crucially, it functions toinform us in novel situations when habits and memories of single experiences are insufficient to generate thepredictions needed to guide action and decision-making5. From an evolutionary point of view, the most valuable memory traces to construct and maintain are those that help predict aversive experiences, such that thesecan be avoided in the future. 

However, despite the central role that negatively arousing memories are thoughtto play in adaptive future behavior6, investigations of associative inference have rarely researched the effects ofemotion7 or accounted for its fundamentally prospective purpose. It is thus unknown how associative inferenceis affected when one of the recombined memories constitutes a threatening experience, and whether such aneffect is moderated by noradrenergic arousal.

Contemporary neuroscientific studies demonstrate that non-emotional memories of distinct experiencesthat share a common element are integrated at the representational level. 

Importantly, these findings implythat the representations that facilitate associative inference are established before they are required10. Asassociations between emotionally arousing events and their learned predictors are privileged in memory11, it canbe hypothesized that the integration of threatening events with pre-existing related memories is prioritized, resultingin enhanced associative inference for judgments that include a memory of threat (prioritization hypothesis). 

Inline with this idea, specific retroactive emotional enhancements have previously been demonstrated: recognitionmemory was enhanced for neutral items that later, through new learning, acquired emotional significance asinstances of a semantic category that predicts a mild shock12. As for associative memories, Zhu et al.13 recentlyrevealed that when one element of an existing memory is paired with an emotional stimulus, associations withinthe original memory are strengthened. Similar evidence comes from a study using monetary reward as a reinforcing stimulus, showing that a hippocampus-dependent mechanism allows positive value to spread to reactivatedmemories, thereby subconsciously biasing subsequent decision-making14. 

Experiments using higher-order threatconditioning paradigms have similarly demonstrated that conditioned responses generalize to stimuli that areonly indirectly predictive of threat–18. However, note that these studies were not designed to test the effectof threat-predictive value on associative inference, which is a declarative memory ability, as the emotionallyrelevant stimuli used (shock or monetary reward) overlapped with many episodes instead of being unique. As a result, such previous study designs do not enable testing of indirect associations across episodes (associativeinference) with threat stimuli.

In sharp contrast with this prioritization hypothesis, research into the effects of negative emotion on episodicassociative memory paints a different picture. Memory for associations between items is typically impaired whena negative stimulus is involved19. 

This effect is thought to be driven primarily by noradrenergic arousal20. Studiesby Bisby et al.21–23 have shown that the presence of a negative element during encoding decreases the accuracy andcoherence of subsequent associative memory, despite recognition memory for the negative element itself beingenhanced. If the formation of cross-memory associations between reactivated memories and novel experiencesrelies on the same processes that bind elements within memories, associative inference following threat learningshould be impaired. Specifically, the integrated representations that support associative inference may be affectedby emotion in the same way that simple associative memories are (impairment hypothesis).

Here we hypothesized that predictive threat learning can, once consolidated, affect associative inference.However, as the literature is unclear on the direction of this effect, this was left open. We further hypothesizedthat, regardless of the direction, the magnitude of this effect is amplified by noradrenergic arousal responsesduring threat learning. To test whether and how threat learning impacts associative inference, we developedthe 'Predictive Relational Emotional Memory (PRE-Memory) paradigm'. 

Unlike most previous studies of emotional associative memory, we spread out learning and testing over several days. The effects of emotion on episodic memory typically require time to emerge24, as alteration via synaptic consolidation is a protein-synthesis-dependent process25. Further, the formation of cognitive-map-like overlapping representations and extraction ofregularities from them has always been thought of as a time-dependent process26,27. Another core difference withearlier work is that here, the neutral and emotional elements that form an episode were presented sequentially,rather than simultaneously. This 'episodic threat learning' procedure is likely to elicit defensive preparations toactively predict and cope with impending threats, triggering motivational systems for future-oriented action.Specifically, participants first encoded neutral premise memories (A→B), which on the following day werelinked to multimodal stimuli that were either highly arousing or neutral (B→ CTHREAT/NEUTRAL). 

Then, on thethird and final day, participants made associative inferences, recombining the indirectly linked elements acrosspremise memories (A→ CTHREAT/NEUTRAL). Noradrenergic arousal responses to B and C items during episodicthreat learning were indexed using pupillometry29. Employing the novel PRE-Memory paradigm, thepresent study sheds light on the functioning of a complex feature of episodic memory in those situations whenit may matter the most.

Due to the current unavailability of tools for determining the necessary sample size in logistic multilevel regressions, no a priori power analysis was conducted. Instead, we reasoned that the sample sizeshould be (1) comparable to those reported in key studies that inspired this experiment, and (2) sufficient to reliably estimate parameters in a two-level model. Bisby et al. 

consistently demonstrated impairments in emotionalassociative memory across three experiments with sample sizes between 17 and 2721, whereas enhancements ofneutral memories following threat learning have been shown in samples of 30 participants12,13. We thus soughtto include at least 30 participants, such that both an impairing or enhancing effect of threat on associative inference could be detected. 

Moreover, sample sizes of 30 or more participants are likely to yield unbiased estimates inmultilevel models. Anticipating some drop-out and missing data for the pupil measure, 47 healthy individualswere recruited via the university's online system and gave written informed consent to participate in this study.

Exclusion criteria as assessed via a screening based on self-report were recreational drug use at a frequencyhigher than once a month, the average consumption of 21 or more units of alcohol per week, having experiencedtrauma, and having received treatment for a mental disorder listed in the DSM-5 by a psychologist or psychiatristin the past year. 

Participants were rewarded either with course credits or 45 euros for completing the experiment,or according to the total amount of time spent in the lab in case of dropping out. Due to the high aversivenessof the picture-sound combinations on day two, it was emphasized that they were free to quit the experiment atany time without having to state a reason. 

Ten participants made use of this option on day two. Additionally,one participant aborted the computer task on day two by accidentally pressing the escape button, and anothermissed the appointment on the third day. The final sample thus includes data from 35 participants (mean age=21.4,SD=2.6, range= [19–28]; 24 women) before analysis. This study was performed by the Declaration of Helsinki and approved by the local ethics committee of the University of Amsterdam.

For more information:1950477648nn@gmail.com