Inbal Goshen, Matthew Brodsky, Rohit Prakash, Jenelle Wallace, Viviana Gradinaru, Charu Ramakrishnan, Karl Deisseroth. Dynamics of Retrieval Strategies for Remote Memories. Cell, Volume 147, Issue 3, 28 October 2011, Pages 678-689.

Prevailing theory suggests that long-term memories are encoded via a two-phase process requiring early involvement of the hippocampus followed by the neocortex. Contextual fear memories in rodents rely on the hippocampus immediately following training but are unaffected by hippocampal lesions or pharmacological inhibition weeks later. With fast optogenetic methods, we examine the real-time contribution of hippocampal CA1 excitatory neurons to remote memory and find that contextual fear memory recall, even weeks after training, can be reversibly abolished by temporally precise optogenetic inhibition of CA1. When this inhibition is extended to match the typical time course of pharmacological inhibition, remote hippocampus dependence converts to hippocampus independence, suggesting that long-term memory retrieval normally depends on the hippocampus but can adaptively shift to alternate structures. Further revealing the plasticity of mechanisms required for memory recall, we confirm the remote-timescale importance of the anterior cingulate cortex (ACC) and implicate CA1 in ACC recruitment for remote recall.

In a delayed-nonmatch-to-sample task, a rat was shown one of two levers, then there was a delay during which the rat was distracted, then the rat was shown both levers and was supposed to press the one it hadn’t been shown at first. The model of CA1 was trained on the most difficult, successful trials, then replayed later to stimulate CA1.

Stimulation occurred in two conditions: normal, and when glutamate transmission was blocked. In both conditions, the prosthesis augmented performance by about 20%. I couldn’t tell from the paper whether they had different models depending on which lever was about to be pressed, and chose to play the correct model to stimulate recall; if they did, then this is really just showing that the prosthesis can affect which memory is recalled, not that it can actually substitute for CA1.

Theodore W Berger, Robert E Hampson, Dong Song, Anushka Goonawardena, Vasilis Z Marmarelis and Sam A Deadwyler. A cortical neural prosthesis for restoring and enhancing memory. 2011 J. Neural Eng. 8 046017

We blogged about this project a few years ago: http://neurodudes.com/2007/04/04/interview-on-usc-hippocampal-prosthetic/

http://www.nature.com/nature/journal/vaop/ncurrent/full/nature09817.html

Abstract after the break.

Adult hippocampal neurogenesis is a unique form of neural circuit plasticity that results in the generation of new neurons in the dentate gyrus throughout life. Neurons that arise in adults (adult-born neurons) show heightened synaptic plasticity during their maturation and can account for up to ten per cent of the entire granule cell population. Moreover, levels of adult hippocampal neurogenesis are increased by interventions that are associated with beneficial effects on cognition and mood, such as learning, environmental enrichment, exercise and chronic treatment with antidepressants. Together, these properties of adult neurogenesis indicate that this process could be harnessed to improve hippocampal functions. However, despite a substantial number of studies demonstrating that adult-born neurons are necessary for mediating specific cognitive functions, as well as some of the behavioural effects of antidepressants, it is unknown whether an increase in adult hippocampal neurogenesis is sufficient to improve cognition and mood. Here we show that inducible genetic expansion of the population of adult-born neurons through enhancing their survival improves performance in a specific cognitive task in which two similar contexts need to be distinguished. Mice with increased adult hippocampal neurogenesis show normal object recognition, spatial learning, contextual fear conditioning and extinction learning but are more efficient in differentiating between overlapping contextual representations, which is indicative of enhanced pattern separation. Furthermore, stimulation of adult hippocampal neurogenesis, when combined with an intervention such as voluntary exercise, produces a robust increase in exploratory behaviour. However, increasing adult hippocampal neurogenesis alone does not produce a behavioural response like that induced by anxiolytic agents or antidepressants. Together, our findings suggest that strategies that are designed to increase adult hippocampal neurogenesis specifically, by targeting the cell death of adult-born neurons or by other mechanisms, may have therapeutic potential for reversing impairments in pattern separation and dentate gyrus dysfunction such as those seen during normal ageing.

When we learn new information we use only a tiny fraction of the neurons in our brain for that particular memory trace. In order to allow the molecular study of those specific neurons we combined elements of the tet system with a promoter that is activated by high level neural activity (the cfos promoter) to generate mice in which a genetic tag can be introduced into neurons that are active at a given point in time. The tag can be maintained for a prolonged period, creating a precise record of the neural activity pattern at a specific point in time. Using fear conditioning we found that the same neurons activated during learning were reactivated when the animal recalled the fearful event. We also found that these neurons were no longer activated following memory extinction, consistent with the idea that extinction modifies a component of the original memory trace.

That quote is from an abstract for a talk by Mark Mayford that will be given next week at UCSD. However, the following paper seems to report those results:

Leon G. Reijmers, Brian L. Perkins, Naoki Matsuo, Mark Mayford. Localization of a Stable Neural Correlate of Associative Memory. Science 31 August 2007: Vol. 317. no. 5842, pp. 1230 – 1233.

In addition, here’s the rest of the talk abstract, which seems to report new results:

One fundamental question in memory research has been how this nuclear to synaptic communication occurs. Using the cfos transgenic approach to specifically focus on activated circuits, we found in a recent study that glutamate receptors get specifically targeted to synapses that are altered with learning. That is, learning produces a sort of molecular tag at certain synapses that allows them to capture the newly synthesized receptors arriving from the nucleus hours after the learning event. Thus, the synapses that are altered in strength to produce a short-term memory must be primed, or tagged, to receive new receptor in order for that memory to be maintained long-term.

Supplementary Figure 1: A schematic illustrating the main finding. Slow gamma is maximal on the descending portion of the theta wave, and fast gamma peaks near the trough. Slow gamma serves to synchronize CA1 with inputs arriving from CA3, and fast gamma synchronizes CA1 with MEC input.

Laura Lee Colgin, Tobias Denninger, Marianne Fyhn, Torkel Hafting, Tora Bonnevie, Ole Jensen, May-Britt Moser & Edvard I. Moser. Frequency of gamma oscillations routes flow of information in the hippocampus. Nature 462, 353-357 (19 November 2009)

Gamma oscillations are thought to transiently link distributed cell assemblies that are processing related information^{1, 2}… This ‘binding’ mechanism requires that spatially distributed cells fire together with millisecond range precision^{7, 8}; however, it is not clear how such coordinated timing is achieved given that the frequency of gamma oscillations varies substantially across space and time, from 25 to almost 150 Hz^{1, 9, 10, 11, 12, 13}. Here we show that gamma oscillations in the CA1 area of the hippocampus split into distinct fast and slow frequency components that differentially couple CA1 to inputs from the medial entorhinal cortex, an area that provides information about the animal’s current position^{14, 15, 16, 17}, and CA3, a hippocampal subfield essential for storage of such information^{14, 18, 19}. Fast gamma oscillations in CA1 were synchronized with fast gamma in medial entorhinal cortex, and slow gamma oscillations in CA1 were coherent with slow gamma in CA3. Significant proportions of cells in medial entorhinal cortex and CA3 were phase-locked to fast and slow CA1 gamma waves, respectively. The two types of gamma occurred at different phases of the CA1 theta rhythm and mostly on different theta cycles.
…
Hippocampal gamma oscillations are thought to arise from two sources, one in the entorhinal cortex (EC)^{9, 21} and another intrinsic to the hippocampus^{9, 10}. The estimated current sources during hippocampal gamma oscillations closely match the currents that result from stimulation of the perforant path projection from EC to the hippocampus⁹, indicating that hippocampal gamma may be entrained by direct inputs from EC. Entorhinal gamma has been reported to be relatively fast (90 Hz)²², and high-frequency gamma (80 Hz) has been reported also in the hippocampus⁹. However, in animals with EC lesions, a slower gamma rhythm (40 Hz) becomes more apparent in the hippocampus. The pattern of current dipoles for this slower oscillation matches the current profile associated with activation of the Schaffer collateral/commissural pathway from CA3 to CA1⁹. Collectively, these observations indicate that hippocampal gamma oscillations have multiple origins and raise the possibility that variations in gamma frequency in CA1 reflect alternating synchronization with slow gamma in CA3 and fast gamma in EC (Supplementary Fig. 1). To test this idea, we sampled neural activity simultaneously from CA1 and either CA3 or layer III of medial entorhinal cortex (MEC) in freely moving rats.
…
We recorded a total of 169 CA1 place cells and 17
putative CA1 interneurons… Only 6% of CA1 place cells showed significant phase-locking to both slow and fast gamma (Supplementary Table 1). The fast gamma-modulated place cells tended to spike near the gamma trough and the slow gamma-modulated place cells preferred to fire closer to the gamma peak (Fig. 4d, 4e; Supplementary Fig. 12). These observations indicate that slow and fast gamma-modulated place cells correspond to separate populations of CA1 neurons firing at different gamma phases and carrying different temporal codes²³.
…
The data support the notion that direct input from layer III of MEC is important for activating CA1 place cells, particularly in the centre of their firing field. A significantly higher proportion of place cells in CA1 was driven by fast gamma from MEC than by slow gamma from CA3. Additionally, fast gamma power in CA1 was maximal near the theta trough, the theta phase when CA1 place cells are most likely to fire. These observations fit well with previous studies showing that place-selective firing in CA1 depends on direct inputs from layer III of MEC^{14, 17} and indicate that transmission of spatial information between these regions is facilitated by fast gamma oscillations. Considering that the power of fast gamma oscillations in the prefrontal and parietal cortices has recently been found to be modulated by CA1 theta phase¹³, the fast gamma-mediated coupling between CA1 and MEC may extend further to other cortical regions.

A significantly lower percentage of CA1 place cells was phase-locked to slow gamma than to fast gamma. Slow gamma occurs primarily at a theta phase when CA1 place cells fire with relatively low probability and is probably driven by feedforward inhibition from CA3 that transiently suppresses CA1 firing¹⁰. Perhaps more easily able to overcome inhibition during slow gamma are cell ensembles with synapses that had previously undergone long-term potentiation, a process that lastingly strengthens the responses of neurons to inputs and is believed to underlie memory storage²⁵. During periods of slow gamma, CA1 place cells were activated across larger spatial regions, indicating that the cells fired at earlier stages of trajectories through the firing fields, possibly as a consequence of long-term potentiation of CA3–CA1 synapses^{26, 27}. Together, these findings raise the possibility that slow gamma conveys information from memory stores in the CA3 or CA3–CA1 network^{18, 19}. In line with such an idea, coherence between gamma activity in CA3 and CA1 is increased during retrieval of spatial information in a hippocampus-dependent memory task⁶.

The results are consistent with previous studies reporting that inputs from EC and CA3 arrive in CA1 at different phases of the theta cycle²⁸. Long-term potentiation in CA1 is most easily induced at a particular phase of theta²⁹, corresponding to the phase when EC input is maximal²⁸. This indicates that the theta phase when EC inputs preferentially arrive may coincide with the time when memory encoding occurs optimally and raises the possibility that the EC-coupled CA1 fast gamma observed in the present study serves to facilitate memory encoding³⁰. Retrieval of information is thought to occur at a different theta phase than memory encoding, during which time CA3 input to CA1 is maximal and incoming signals from EC are suppressed²⁸. This idea fits well with the above-hypothesized memory retrieval function for slow gamma. Separation of afferent inputs to CA1 on different phases of theta is probably important for avoiding re-encoding of previously stored memories and also for reliably distinguishing perceptions of ongoing experiences from internally evoked memories. The present results raise the possibility that slow and fast gamma play an important role in this separation of inputs by filtering out improperly timed signals from one afferent while facilitating transfer of coherent activity from another. Considering that broadband gamma oscillations occur in other areas^{1, 2, 3, 4, 11, 24}, separation of gamma oscillations into discrete frequency channels may be used throughout the brain to enhance interregional communication.

Replay of behavioral sequences in the hippocampus during sharp wave ripple complexes (SWRs) provides a potential mechanism for memory consolidation and the learning of knowledge structures. Current hypotheses imply that replay should straightforwardly reflect recent experience. However, we find these hypotheses to be incompatible with the content of replay on a task with two distinct behavioral sequences (A and B). We observed forward and backward replay of B even when rats had been performing A for >10 min. Furthermore, replay of nonlocal sequence B occurred more often when B was infrequently experienced. Neither forward nor backward sequences preferentially represented highly experienced trajectories within a session. Additionally, we observed the construction of never-experienced novel-path sequences. These observations challenge the idea that sequence activation during SWRs is a simple replay of recent experience. Instead, replay reflected all physically available trajectories within the environment, suggesting a potential role in active learning and maintenance of the cognitive map.

Anoopum S. Gupta, Matthijs A.A. van der Meer, David S. Touretzky, A. David Redish. Hippocampal Replay Is Not a Simple Function of Experience. Neuron, Volume 65, Issue 5, 695-705, 11 March 2010

And here’s an interesting detail from the discussion section of the paper:

Furthermore, during left-only and right-only half-
sessions, trajectories along the nonrecent (opposite-side) loop
were replayed with a similar frequency to trajectories on the
recent (same-side) loop. This observation was in contrast to
alternation half-sessions in which opposite-side loops were re-
played less frequently. These observations indicate that current
proposals for potential mechanisms of replay that rely on
recency or frequency of experience are inadequate.

What’s going on? One tantalizing possibility is that the enzyme itself is directly related to the memory trace. This contradicts the (unproven) assumption of modern neuroscience that memories are stored solely in the synaptic strengths (ie. membrane-bound receptors) of a neuron. The other suggestion is that PKM-zeta is actively maintaining synapses and that enzymatic inhibition disrupts the precise maintenance of receptors or synaptic machinery. The effects happen quite fast (within 2 hours after drug injection), which seems short for receptor recycling but perhaps long enough for structural change to occur. I’m no expert on receptor movement: Is 2 hours long enough to add/remove a significant number of receptors?

Fascinating work but the method is blunt, wiping all experimentally-induced memories (and probably others too). Last month, another group reported (also in Science) selective erasure of a fear-conditioned memory using an interesting new genetic tool. Here, neurons in the amgydala that overexpressed CREB were found to be preferentially recruited into a fear memory trace (as shown in a previous Science paper). Incorporation into the memory trace was assayed by expression of the immediate-early gene (ie. activity-dependent) Arc. In the present study, they combine overexpression of CREB in a subset of neurons with cell death (via Diphtheria toxin in a transgenic mouse vulnerable to diphtheria). Apparently, normal mice lack the receptor (here a simian version is used) that confers pathogenicity for diphtheria. Thus, the viral construct both overexpresses CREB in a subset of neurons and selectively makes the same subset vulnerable to diphtheria. Ablation of just these neurons causes a permanent loss of the memory. Subsequent similar learning proceeds just fine (using the remaining neurons).

Can we say that the race is officially on to ablate just the synapses involved in the memory? I think so. Extra points if the ablation is reversible too!

Atul Gawande also recently wrote a New Yorker article about treating phantom itch with Ramachandran’s mirror box. I found this part of Gawande’s article on statistical inference in perception most interesting:

You can get a sense of this from brain-anatomy studies. If visual sensations were primarily received rather than constructed by the brain, you’d expect that most of the fibres going to the brain’s primary visual cortex would come from the retina. Instead, scientists have found that only twenty per cent do; eighty per cent come downward from regions of the brain governing functions like memory. Richard Gregory, a prominent British neuropsychologist, estimates that visual perception is more than ninety per cent memory and less than ten per cent sensory nerve signals. When Oaklander theorized that M.’s itch was endogenous, rather than generated by peripheral nerve signals, she was onto something important.

I’m not familiar with this field but I wonder if anyone has tried to quantify what percent of our conscious experience that we normally believe to be 100% due to sensory input is actually recall from memory/inference based on past observation. Also, can this percentage adaptively change? Perhaps there are situations where the brain chooses to rely more heavily on memory and other cases where it relies more on primary sensory input.

Click on for some of my favorite passages from the Abbott piece.

Abbott uses the following problem of input decoding

Spike counts and neuronal firing rates are positive quantities. This simple fact has important implications for neural coding and neural circuits that provide a framework for thinking about a number of research directions taken over the past 20 years.

to highlight new work in synchrony, dendritic compartments, and balanced excitation-inhibition. This is probably the best part of the whole article. With some simple arithmetic, he motivates and explains solutions to the problem of correlating neural activity with real events.

The successes of circuit models (and principles of circuit models) in primary visual cortex:

We now have plausible mechanisms for how simple and complex cells obtain their basic response characteristics. Although no single consensus abouthow the circuits of primary visual cortex operate has arisen from this body of work, this may simply reflect the fact that multiple mechanisms contribute. In other words, many of these ideas are probably correct in one way or another, and the wealth of ideas in this field should be viewed as a success. Circuit-level modeling is now advancing beyond primary sensory areas (for example, Cadieu etal., 2007) and to the consideration of phenomena such as working memory through sustained activity (Amit and Brunel, 1997,Compte etal., 2000,Seung etal., 2000) and decision making (Wang, 2002,Machens etal., 2005).

And the dangers of an unhealthy obsession with connectomics:

What can we learn from the complete connectome or, indeed, a complete mathematical description of a complex artificial network model?

First, what can’t we learn? It is unlikely, for example, that we could deduce the task that the network was constructed to perform even if we were given the complete equations and connections of the model. If, along with this information, we were told what this task was, it is unlikely that we could figure out how the network performs it. If we somehow managed to make any progress along these lines, the people who constructed the network could probably provide us with another one that performs the same task but has a different connectome. In a similarway, biological systems may operate in a more variable manner than we have suspected, as has been stressed by Eve Marder (Marder etal., 2007). These issues are particularly true of a class of network models known as liquid state or echostate networks (Maass etal., 2002,Jaeger, 2003). In these models, the vast majority of interneuronal connections are not directly related to the task being performed (they are typically chosen randomly and left unchanged), the exceptions being synapses onto the output units of the network. Nevertheless, the tuned values of the synapses onto the output units can only be understood through their relationships to the random synapses. Such systems represent enormous challenges for conventional anatomical and physiological approaches.

The fact that the connectome of an artificial neural network does not typically tell us what the network does or how it does it should not be taken as an indication that this information is useless. Far from it. But we must be willing to be more abstract in our thinking. The important issue for an artificial network is not how it works but how it was constructed, which means what training procedures and modification rules were used to get it to perform a task. Although this information is not provided directly by the connectome, much can be inferred. For example, it is important to know whether the network has a feedforward architecture or has strong feedback loops. Other features of the network layout, whether it has hubs or bottlenecks, how many layers it contains, and its degree of heterogeneity, provide important clues as well. Obtaining a high-resolution connectome in neuroscience will be of great value, but artificial neural networks provide a cautionary tale that reminds us that scientific revolutions tend to render uninteresting as many questions as they answer. We will be fortunate if the connectome project does this for neuroscience, butas we launch ourselves into it we should appreciate that, as artificial neural networks appear to suggest, we may be asking the wrong questions.

Finally a major challenge for the future:

This is where I think the future lies in theoretical investigations of cognitive function. We must learn how to build models that construct hypotheses through their internally generated activity while remaining sensitive to the constraints provided by externally generated sensory evidence.

http://news-service.stanford.edu/news/2008/october8/hamster-100808.html:

The hamsters were first exposed to two hours of bright light late at night. Then the next day the researchers delayed the usual light/dark cycle by three hours.

This disrupted their circadian rhythm, and made it hard for them to learn.

http://www.pnas.org/content/105/40/15593:

Control hamsters exhibited normal circadian modulation of performance in a delayed novel-object recognition task. By contrast, arrhythmic animals could not discriminate a novel object from a familiar one only 20 or 60 min after training.

Memory performance was not related to prior sleep history as sleep manipulations had no effect on performance.

Because GABA is the primary neurotransmitter of the SCN, we hypothesized that the normal pattern of GABA output from the SCN may have been altered in arrhythmic hamsters in such a way as to increase inhibitory input at SCN target sites involved in cognition.

“What I thought was happening was that our animals were having chronically high levels of GABA because they had lost their circadian rhythm,” Ruby said. “So instead of rhythmic GABA, it is just constant GABA output.”
(from the press release (the first link))

So to test this, they blocked GABA, and indeed,

The GABA antagonist pentylenetetrazol restored learning without restoring circadian rhythms…

One interesting thing this is that it adds another item to the list of functions of the circadian rhythm, a list that is so far is surprisingly short:

…the good health and longevity of suprachiasmatic nucleus (SCN)-lesioned animals support the notion that the circadian system is of little consequence to their overall physiology. One notable exception to this trend is reproduction in rodents. Elimination of circadian timing by SCN ablation eliminates estrous cycles and thereby prevents reproduction… (from the journal article)