Differential mechanisms involved in the formation of long-term and short-term representations
It is well established that, following one-trial learning, long-term memory (LTM) consolidation requires protein synthesis whereas short-term memory (STM) does not. A great deal of work has examined the role of specific protein products, including brain-derived neurotrophic factor (BDNF), in the consolidation for one-trial learning. What remains unclear is the differential involvement of plasticity-related mechanisms in STM and LTM for multi-trial learning. In this study, undergraduate researcher Patrick Kim and I trained mice to learn an odor-reward association over 20-trials. Prior to training, we gave olfactory bulb (OB)-specific infusions of BDNF receptor antagonist, K252a. We, then, tested their memory for the association 2 (STM) or 48 (LTM) hours later. We found that mice given K252a did not differ from controls in learning rate, but showed impaired memory when tested 48 hours, but not 2 hours, after training. We also found that this impaired memory performance was associated with low certainty of the learned association. The finding suggests that BDNF in uniquely involved in LTM consolidation for incrementally-acquired memories. This work was done as part of Patrick's honors thesis project. It was presented at the annual conference of the Canadian Association for Neuroscience in 2014.
The role of Brain-Derived Neurotrophic Factor (BDNF) in the formation of sensory representations
This project is part of my ongoing collaboration with Professor Francis Lee at Weill-Cornell Medical College in New York City, where I spent a year as part of a NIH-sponsored translational research grant. Francis’s lab successfully created the Val66Met knock-in mouse, which has near-normal levels of constitutive BDNF secretion but compromised activity-dependent secretion. Thus, the model allows us to differentiate between the unique contributions of the two modes of secretion. Previous work from our lab showed that the odor generalization task can be used to behaviorally approximate the probability density function of an odor representation (Cleland et al., 2012). Using this task in conjunction with the bdnf genetic variants (or with pharmacological manipulations) allows us to study the mechanisms and their effect on representations (or the connection between the implementional and representational levels, in the language of Marr, 1982). Previously, we showed using the odor generalization task that Val66Met mice were able to learn specific odor representations, whereas bdnf heterozygote mice (which have half of the normal levels of BDNF) were not. The results suggest that regardless of the mode of secretion, the ability to form specific odor representations depends on the absolute levels of BDNF.
Towards a spatiotemporal "map" of plasticity-related activity following multi-trial associative learning
Patrick and I identified BDNF as an important protein for LTM formation in multi-trial learning. What we have not yet addressed is the specific timeline of its involvement. Previous studies have contributed to a strong understanding of the role of many molecular mechanisms involved in LTM, including describing the effects of those mechanisms on longer-term structural changes to neuron ensembles involved in LTM consolidation. However, this research has also suggested that LTM depends as much on the temporal specificity of these mechanisms as their downstream effects. In addition, it is yet unclear from this work how the timing of these mechanisms is coordinated across the multiple brain regions involved in learning. In this study, we take a first step toward characterizing the timecourse of several molecular mechanisms across multiple brain regions. We train mice on an associative odor learning task for 1, 2, 4, or 6 days. We collected the OB, striatum, hippocampus, cortex, and cerebellum from the mice on each day prior to training, immediately after training, or 15, 30, or 60 minutes after training. We then used high-throughput RT-PCR to analyze mRNA levels for several plasticity-related proteins (PRPs), including bdnf, intracellular signaling cascades, erk1 and erk2, transcription factor, creb1, and immediate early genes, arc, fos, and erg1. We found that learning-responsive transcription differed be- tween genes, and that PRP timecourses differed as a function of brain region. Future studies could use this “spatiotemporal” map to discover the functional consequences of these patterns.
Changes in representations from learning to memory
Traditionally, studies in behavioural neuroscience have used test/pre-test comparisons of measures like latency, digging time, or percent correct as metric of memory retention. While these metrics are effective in assessing differences in memory expression between time points, they fail to provide insight into the underlying representations responsible for the observed memory differences. In my work with Patrick, we very much used traditional metrics to show that BDNF blockade disrupted LTM, but not STM, for an odor-reward association. My training in cognitive psychology, however, compels me to consider memory at the representational level (Marr, 1982). Here, differences in memory performance, as seen by traditional metrics, can be said to arise from representations which differ in form or shape. For example, animals could show poor memory performance as a result of a memory representation with equally decayed synaptic connections that maintains its high specificity (e.g. "I sort of remember it being grapefruit juice and not orange juice, but I'm not sure at all"). Or poor memory could result from disruptions to the specificity of the memory representation without influencing the strength of existing connections (e.g. "I am absolutely certain that it was some kind of citrus, but I don't know which"). In this scenario, poor memory performance is the result of a strong representation whose form no longer precisely represents the learned stimulus.
A natural next question for me, is to bridge the gap between behaviour and mechanism, by investigating the role of plasticity-related mechanisms in modifying memory representations. A simple, but impactful, initial experiment uses the olfactory generalization task in mice with infusions BDNF receptor blockers. Students and I can train animals to learn an odor-reward association and we would test their representations 2 and 48 hours after training. The results would be able to inform how memory strength corresponds to representation structure. For example, are the LTM deficits we observed after K252a infusion in Patrick's experiments the result of weakened, but accurate representations or does the BDNF-pathway play some role in maintaining representation specificity?
Future studies can vary learning schedules (e.g. spaced versus massed learning) in order to ask questions about how different training parameters influence the shape of representations. Additionally, we can probe representations along various timepoints of the learning process, such as encoding, consolidation, and even, reconsolidation. Reconsolidation would be an interesting one for me to pursue, particularly because we know that some mechanisms involved in initial LTM consolidation are not involved in the reconsolidation of the same memory (Lee et al., 2004). So comparisons of these timepoints will contribute to further understanding of the temporal specificity of mechanisms. Finally, our ultimate goal is to understand the role of molecular mechanisms in representation specificity, thus, we can extend our investigation to other plasticity-proteins and neuromodulators. I'd also like to improve the specificity of our manipulations by using oligodeoxynucleotides or designer receptors exclusively activated by designer drugs (DRREADs; Ferguson et al., 2012)
Spatiotemporal analysis of the molecular mechanisms of STM and LTM
The results of work with Madhura and the Pleiss Lab provide an invaluable map that allows us to target our investigation of memory mechanisms and, specifically, to begin to form a timeline of known molecular and structural changes. When we talk about the structural mechanisms of LTM, we immediately think of dendritic growth, long-term potentiation (LTP), the survival of adult-born neurons, or changes to the extracellular matrix. We also have some understanding of the molecular mechanism involved in these structural changes.
One question that could be a strong first experiment is what exactly is the role of timing specificity in LTM consolidation? We would select a mechanism of interest (e.g. erg1). From the work with the Pleiss Lab, we would know when and in what brain region erg1 has its peak activity. As we trained animals, we could artificially delay or advance the timing of the erg1 peak in various brain regions and observe the effects on memory performance.
Another research question that I am particularly interested in is to better understand the relationship between the different types of structural plasticity. However, what is the relationship between these different forms? Certainly, for any one consolidation event, all or some of these changes occur across the whole brain, so what is their individual contribution to LTM? Do they occur co-currently or is there a meaningful order?