Neuromorphic one-shot learning utilizing a phase-transition material.

Design of hardware based on biological principles of neuronal computation and plasticity in the brain is a leading approach to realizing energy- and sample-efficient AI and learning machines. An important factor in selection of the hardware building blocks is the identification of candidate materials with physical properties suitable to emulate the large dynamic ranges and varied timescales of neuronal signaling. Previous work has shown that the all-or-none spiking behavior of neurons can be mimicked by threshold switches utilizing material phase transitions. Here, we demonstrate that devices based on a prototypical metal-insulator-transition material, vanadium dioxide (VO2), can be dynamically controlled to access a continuum of intermediate resistance states. Furthermore, the timescale of their intrinsic relaxation can be configured to match a range of biologically relevant timescales from milliseconds to seconds. We exploit these device properties to emulate three aspects of neuronal analog computation: fast (~1 ms) spiking in a neuronal soma compartment, slow (~100 ms) spiking in a dendritic compartment, and ultraslow (~1 s) biochemical signaling involved in temporal credit assignment for a recently discovered biological mechanism of one-shot learning. Simulations show that an artificial neural network using properties of VO2 devices to control an agent navigating a spatial environment can learn an efficient path to a reward in up to fourfold fewer trials than standard methods. The phase relaxations described in our study may be engineered in a variety of materials and can be controlled by thermal, electrical, or optical stimuli, suggesting further opportunities to emulate biological learning in neuromorphic hardware.

Offline memory replay in recurrent neuronal networks emerges from constraints on online dynamics.

During spatial exploration, neural circuits in the hippocampus store memories of sequences of sensory events encountered in the environment. When sensory information is absent during 'offline' resting periods, brief neuronal population bursts can 'replay' sequences of activity that resemble bouts of sensory experience. These sequences can occur in either forward or reverse order, and can even include spatial trajectories that have not been experienced, but are consistent with the topology of the environment. The neural circuit mechanisms underlying this variable and flexible sequence generation are unknown. Here we demonstrate in a recurrent spiking network model of hippocampal area CA3 that experimental constraints on network dynamics such as population sparsity, stimulus selectivity, rhythmicity and spike rate adaptation, as well as associative synaptic connectivity, enable additional emergent properties, including variable offline memory replay. In an online stimulus-driven state, we observed the emergence of neuronal sequences that swept from representations of past to future stimuli on the timescale of the theta rhythm. In an offline state driven only by noise, the network generated both forward and reverse neuronal sequences, and recapitulated the experimental observation that offline memory replay events tend to include salient locations like the site of a reward. These results demonstrate that biological constraints on the dynamics of recurrent neural circuits are sufficient to enable memories of sensory events stored in the strengths of synaptic connections to be flexibly read out during rest and sleep, which is thought to be important for memory consolidation and planning of future behaviour. KEY POINTS: A recurrent spiking network model of hippocampal area CA3 was optimized to recapitulate experimentally observed network dynamics during simulated spatial exploration. During simulated offline rest, the network exhibited the emergent property of generating flexible forward, reverse and mixed direction memory replay events. Network perturbations and analysis of model diversity and degeneracy identified associative synaptic connectivity and key features of network dynamics as important for offline sequence generation. Network simulations demonstrate that population over-representation of salient positions like the site of reward results in biased memory replay.

Functional comparison of the effects of TARPs and cornichons on AMPA receptor trafficking and gating.

Glutamate receptors of the AMPA subtype (AMPARs) mediate fast synaptic transmission in the brain. These ionotropic receptors rely on auxiliary subunits known as transmembrane AMPAR regulatory proteins (TARPs) for both trafficking and gating. Recently, a second family of AMPAR binding proteins, referred to as cornichons, were identified and also proposed to function as auxiliary subunits. Cornichons are transmembrane proteins that modulate AMPAR function in expression systems much like TARPs. In the present study we compare the role of cornichons in controlling AMPA receptor function in neurons and HEK cells to that of TARPs. Cornichons mimic some, but not all, of the actions of TARPs in HEK cells; their role in neurons, however, is more limited. Although expressed cornichons can affect the trafficking of AMPARs, they were not detected on the surface of neurons and failed to alter the kinetics of endogenous AMPARs. This neuronal role is more consistent with that of an endoplasmic reticulum (ER) chaperone rather than a bona fide auxiliary subunit.

Probing TARP modulation of AMPA receptor conductance with polyamine toxins.

The properties of synaptic AMPA receptors (AMPARs) depend on their subunit composition and association with transmembrane AMPAR regulatory proteins (TARPs). Although both GluA2 incorporation and TARP association have been shown to influence AMPAR channel conductance, the manner in which different TARPs modulate the mean channel conductance of GluA2-containing AMPARs is unknown. Using ultrafast agonist application and nonstationary fluctuation analysis, we found that TARP subtypes differentially increase the mean channel conductance, but not the peak open probability, of recombinant GluA2-containing AMPARs. TARP γ-8, in particular, enhances mean channel conductance to a greater degree than γ-2, γ-3, or γ-4. We then examined the action of a use-dependent antagonist of GluA2-containing AMPARs, philanthotoxin-74 (PhTx-74), on recombinant AMPARs and on GluA2-containing AMPARs in cerebellar granule neurons from stargazer mice transfected with TARPs. We found that the rate and extent of channel block varies with TARP subtype, in a manner that correlates linearly with mean channel conductance. Furthermore, block of GluA2-containing AMPARs by polyamine toxins varied depending on whether channels were activated by the full agonist glutamate or the partial agonist kainate, consistent with conductance state-dependent block. Block of GluA2-lacking AMPARs by PhTx-433 is also modulated by TARP association and is a function of agonist efficacy. Our data indicate that channel block by polyamine toxins is sensitive to the mean channel conductance of AMPARs, which varies with TARP subtype and agonist efficacy. Furthermore, our results illustrate the utility of polyamine toxins as sensitive probes of AMPAR channel conductance and suggest the possibility that TARPs may influence their channel properties by selectively stabilizing specific channel conformations, rather than altering the pore structure.

Liprinalpha1 degradation by calcium/calmodulin-dependent protein kinase II regulates LAR receptor tyrosine phosphatase distribution and dendrite development.

Neural activity regulates dendrite and synapse development, but the underlying molecular mechanisms are unclear. Ca(2+)/calmodulin-dependent protein kinase II (CaMKII) is an important sensor of synaptic activity, and the scaffold protein liprinalpha1 is involved in pre- and postsynaptic maturation. Here we show that synaptic activity can suppress liprinalpha1 protein level by two pathways: CaMKII-mediated degradation and the ubiquitin-proteasome system. In hippocampal neurons, liprinalpha1 mutants that are immune to CaMKII degradation impair dendrite arborization, reduce spine and synapse number, and inhibit dendritic targeting of receptor tyrosine phosphatase LAR, which is important for dendrite development. Thus, regulated degradation of liprinalpha1 is important for proper LAR receptor distribution, and could provide a mechanism for localized control of dendrite and synapse morphogenesis by activity and CaMKII.

TARP modulation of synaptic AMPA receptor trafficking and gating depends on multiple intracellular domains.

Previous work has established stargazin and its related family of transmembrane AMPA receptor regulatory proteins (TARPs) as auxiliary subunits of AMPA receptors (AMPARs) that control synaptic strength both by targeting AMPARs to synapses through an intracellular PDZ-binding motif and by modulating their gating through an extracellular domain. However, TARPs gamma-2 and gamma-8 differentially regulate the synaptic targeting of AMPARs, despite having identical PDZ-binding motifs. Here, we investigate the structural elements that contribute to this functional difference between TARP subtypes by using domain transplantation and truncation. We identify a component of synaptic AMPAR trafficking that is independent of the TARP C-terminal PDZ-binding motif, and we establish previously uncharacterized roles for the TARP intracellular N terminus, loop, and C terminus in modulating both the trafficking and gating of synaptic AMPARs.

Pungent agents from Szechuan peppers excite sensory neurons by inhibiting two-pore potassium channels.

In traditional folk medicine, Xanthoxylum plants are referred to as 'toothache trees' because their anesthetic or counter-irritant properties render them useful in the treatment of pain. Psychophysical studies have identified hydroxy-alpha-sanshool as the compound most responsible for the unique tingling and buzzing sensations produced by Szechuan peppercorns or other Xanthoxylum preparations. Although it is generally agreed that sanshool elicits its effects by activating somatosensory neurons, the underlying cellular and molecular mechanisms remain a matter of debate. Here we show that hydroxy-alpha-sanshool excites two types of sensory neurons, including small-diameter unmyelinated cells that respond to capsaicin (but not mustard oil) as well as large-diameter myelinated neurons that express the neurotrophin receptor TrkC. We found that hydroxy-alpha-sanshool excites neurons through a unique mechanism involving inhibition of pH- and anesthetic-sensitive two-pore potassium channels (KCNK3, KCNK9 and KCNK18), providing a framework for understanding the unique and complex psychophysical sensations associated with the Szechuan pepper experience.

Recurrent Excitatory Feedback From Mossy Cells Enhances Sparsity and Pattern Separation in the Dentate Gyrus via Indirect Feedback Inhibition.

It is generally appreciated that storing memories of specific events in the mammalian brain, and associating features of the environment with behavioral outcomes requires fine-tuning of the strengths of connections between neurons through synaptic plasticity. It is less understood whether the organization of neuronal circuits comprised of multiple distinct neuronal cell types provides an architectural prior that facilitates learning and memory by generating unique patterns of neuronal activity in response to different stimuli in the environment, even before plasticity and learning occur. Here we simulated a neuronal network responding to sensory stimuli, and systematically determined the effects of specific neuronal cell types and connections on three key metrics of neuronal sensory representations: sparsity, selectivity, and discriminability. We found that when the total amount of input varied considerably across stimuli, standard feedforward and feedback inhibitory circuit motifs failed to discriminate all stimuli without sacrificing sparsity or selectivity. Interestingly, networks that included dedicated excitatory feedback interneurons based on the mossy cells of the hippocampal dentate gyrus exhibited improved pattern separation, a result that depended on the indirect recruitment of feedback inhibition. These results elucidate the roles of cellular diversity and neural circuit architecture on generating neuronal representations with properties advantageous for memory storage and recall.

TARP subtypes differentially and dose-dependently control synaptic AMPA receptor gating.

A family of transmembrane AMPA receptor regulatory proteins (TARPs) profoundly affects the trafficking and gating of AMPA receptors (AMPARs). Although TARP subtypes are differentially expressed throughout the CNS, it is unclear whether this imparts functional diversity to AMPARs in distinct neuronal populations. Here, we examine the effects of each TARP subtype on the kinetics of AMPAR gating in heterologous cells and in neurons. We report a striking heterogeneity in the effects of TARP subtypes on AMPAR deactivation and desensitization, which we demonstrate controls the time course of synaptic transmission. In addition, we find that some TARP subtypes dramatically slow AMPAR activation kinetics. Synaptic AMPAR kinetics also depend on TARP expression level, suggesting a variable TARP/AMPAR stoichiometry. Analysis of quantal synaptic transmission in a TARP gamma-4 knockout (KO) mouse corroborates our expression data and demonstrates that TARP subtype-specific gating of AMPARs contributes to the kinetics of native AMPARs at central synapses.

Bidirectional synaptic plasticity rapidly modifies hippocampal representations.

Learning requires neural adaptations thought to be mediated by activity-dependent synaptic plasticity. A relatively non-standard form of synaptic plasticity driven by dendritic calcium spikes, or plateau potentials, has been reported to underlie place field formation in rodent hippocampal CA1 neurons. Here, we found that this behavioral timescale synaptic plasticity (BTSP) can also reshape existing place fields via bidirectional synaptic weight changes that depend on the temporal proximity of plateau potentials to pre-existing place fields. When evoked near an existing place field, plateau potentials induced less synaptic potentiation and more depression, suggesting BTSP might depend inversely on postsynaptic activation. However, manipulations of place cell membrane potential and computational modeling indicated that this anti-correlation actually results from a dependence on current synaptic weight such that weak inputs potentiate and strong inputs depress. A network model implementing this bidirectional synaptic learning rule suggested that BTSP enables population activity, rather than pairwise neuronal correlations, to drive neural adaptations to experience.

A new housing development in a familiar neighborhood, a wrong turn that ends up lengthening a Sunday stroll: our internal representation of the world requires constant updating, and we need to be able to associate events separated by long intervals of time to finetune future outcome. This often requires neural connections to be altered. A brain region known as the hippocampus is involved in building and maintaining a map of our environment. However, signals from other brain areas can activate silent neurons in the hippocampus when the body is in a specific location by triggering cellular events called dendritic calcium spikes. Milstein et al. explored whether dendritic calcium spikes in the hippocampus could also help the brain to update its map of the world by enabling neurons to stop being active at one location and to start responding at a new position. Experiments in mice showed that calcium spikes could change which features of the environment individual neurons respond to by strengthening or weaking connections between specific cells. Crucially, this mechanism allowed neurons to associate event sequences that unfold over a longer timescale that was more relevant to the ones encountered in day-to-day life. A computational model was then put together, and it demonstrated that dendritic calcium spikes in the hippocampus could enable the brain to make better spatial decisions in future. Indeed, these spikes are driven by inputs from brain regions involved in complex cognitive processes, potentially enabling the delayed outcomes of navigational choices to guide changes in the activity and wiring of neurons. Overall, the work by Milstein et al. advances the understanding of learning and memory in the brain and may inform the design of better systems for artificial learning.

Regulation of AMPA receptor gating and pharmacology by TARP auxiliary subunits.

Presynaptic glutamate release elicits brief waves of membrane depolarization in neurons by activating AMPA receptors. Depending on its precise size and shape, current through AMPA receptors gates downstream processes like NMDA receptor activation and action potential generation. Over a decade of research on AMPA receptor structure and function has identified binding sites on AMPA receptors for agonists, antagonists and allosteric modulators as well as key residues underlying differences in the gating behavior of various AMPA receptor subtypes. However, the recent discovery that AMPA receptors are accompanied in the synaptic membrane by a family of auxiliary subunits known as transmembrane AMPA receptor regulatory proteins (TARPs) has revealed that the kinetics and pharmacology of neuronal AMPA receptors differ in many respects from those predicted by classical studies of AMPA receptors in heterologous systems. Here, we summarize recent work and discuss remaining questions concerning the structure and function of native TARP-AMPA receptor complexes.

GRIP1 controls dendrite morphogenesis by regulating EphB receptor trafficking.

The function of the multi-PDZ domain scaffold protein GRIP1 (glutamate receptor interacting protein 1) in neurons is unclear. To explore the function of GRIP1 in hippocampal neurons, we used RNA interference (RNAi) to knock down the expression of GRIP1. Knockdown of GRIP1 by small interfering RNA (siRNA) in cultured hippocampal neurons caused a loss of dendrites, associated with mislocalization of the GRIP-interacting proteins GIuR2 (AMPA receptor subunit), EphB2 (receptor tyrosine kinase) and KIF5 (also known as kinesin 1; microtubule motor). The loss of dendrites by GRIP1-siRNA was rescued by overexpression of the extracellular domain of EphB2, and was phenocopied by overexpression of the intracellular domain of EphB2 and extracellular application of ephrinB-Fc fusion proteins. Neurons from EphB1-EphB2-EphB3 triple knockout mice showed abnormal dendrite morphogenesis. Disruption of the KIF5-GRIP1 interaction inhibited EphB2 trafficking and strongly impaired dendritic growth. These results indicate an important role for GRIP1 in dendrite morphogenesis by serving as an adaptor protein for kinesin-dependent transport of EphB receptors to dendrites.

New transmembrane AMPA receptor regulatory protein isoform, gamma-7, differentially regulates AMPA receptors.

AMPA-type glutamate receptors (GluRs) mediate most excitatory signaling in the brain and are composed of GluR principal subunits and transmembrane AMPA receptor regulatory protein (TARP) auxiliary subunits. Previous studies identified four mammalian TARPs, gamma-2 (or stargazin), gamma-3, gamma-4, and gamma-8, that control AMPA receptor trafficking, gating, and pharmacology. Here, we explore roles for the homologous gamma-5 and gamma-7 proteins, which were previously suggested not to serve as TARPs. Western blotting reveals high levels of gamma-5 and gamma-7 in the cerebellum, where gamma-7 is enriched in Purkinje neurons in the molecular layer and glomerular synapses in the granule cell layer. Immunoprecipitation proteomics shows that cerebellar gamma-7 avidly and selectively binds to AMPA receptor GluR subunits and also binds to the AMPA receptor clustering protein, postsynaptic density-95 (PSD-95). Furthermore, gamma-7 occurs together with PSD-95 and AMPA receptor subunits in purified postsynaptic densities. In heterologous cells, gamma-7 but not gamma-5 greatly enhances AMPA receptor glutamate-evoked currents and modulates channel gating. In granule cells from stargazer mice, transfection of gamma-7 but not gamma-5 increases AMPA receptor-mediated currents. Compared with stargazin, gamma-7 differentially modulates AMPA receptor glutamate affinity and kainate efficacy. These studies define gamma-7 as a new member of the TARP family that can differentially influence AMPA receptors in cerebellar neurons.

Persistent Sodium Current Mediates the Steep Voltage Dependence of Spatial Coding in Hippocampal Pyramidal Neurons.

The mammalian hippocampus forms a cognitive map using neurons that fire according to an animal's position ("place cells") and many other behavioral and cognitive variables. The responses of these neurons are shaped by their presynaptic inputs and the nature of their postsynaptic integration. In CA1 pyramidal neurons, spatial responses in vivo exhibit a strikingly supralinear dependence on baseline membrane potential. The biophysical mechanisms underlying this nonlinear cellular computation are unknown. Here, through a combination of in vitro, in vivo, and in silico approaches, we show that persistent sodium current mediates the strong membrane potential dependence of place cell activity. This current operates at membrane potentials below the action potential threshold and over seconds-long timescales, mediating a powerful and rapidly reversible amplification of synaptic responses, which drives place cell firing. Thus, we identify a biophysical mechanism that shapes the coding properties of neurons composing the hippocampal cognitive map.

Hippocampal Dentate Mossy Cells Improve Their CV and Trk into the Limelight.

The impact of dentate mossy cells on hippocampal activity remained uncertain despite a long history of investigation. In this issue of Neuron, Hashimotodani et al. (2017) discover a presynaptically expressed form of long-term potentiation at mossy cell outputs, shedding light on their mysterious function.

Behavioral time scale synaptic plasticity underlies CA1 place fields.

Learning is primarily mediated by activity-dependent modifications of synaptic strength within neuronal circuits. We discovered that place fields in hippocampal area CA1 are produced by a synaptic potentiation notably different from Hebbian plasticity. Place fields could be produced in vivo in a single trial by potentiation of input that arrived seconds before and after complex spiking. The potentiated synaptic input was not initially coincident with action potentials or depolarization. This rule, named behavioral time scale synaptic plasticity, abruptly modifies inputs that were neither causal nor close in time to postsynaptic activation. In slices, five pairings of subthreshold presynaptic activity and calcium (Ca2+) plateau potentials produced a large potentiation with an asymmetric seconds-long time course. This plasticity efficiently stores entire behavioral sequences within synaptic weights to produce predictive place cell activity.

Inhibitory suppression of heterogeneously tuned excitation enhances spatial coding in CA1 place cells.

Place cells in the CA1 region of the hippocampus express location-specific firing despite receiving a steady barrage of heterogeneously tuned excitatory inputs that should compromise output dynamic range and timing. We examined the role of synaptic inhibition in countering the deleterious effects of off-target excitation. Intracellular recordings in behaving mice demonstrate that bimodal excitation drives place cells, while unimodal excitation drives weaker or no spatial tuning in interneurons. Optogenetic hyperpolarization of interneurons had spatially uniform effects on place cell membrane potential dynamics, substantially reducing spatial selectivity. These data and a computational model suggest that spatially uniform inhibitory conductance enhances rate coding in place cells by suppressing out-of-field excitation and by limiting dendritic amplification. Similarly, we observed that inhibitory suppression of phasic noise generated by out-of-field excitation enhances temporal coding by expanding the range of theta phase precession. Thus, spatially uniform inhibition allows proficient and flexible coding in hippocampal CA1 by suppressing heterogeneously tuned excitation.

Axonal Filtering Allows Reliable Output during Dendritic Plateau-Driven Complex Spiking in CA1 Neurons.

In CA1 pyramidal neurons, correlated inputs trigger dendritic plateau potentials that drive neuronal plasticity and firing rate modulation. Given the strong electrotonic coupling between soma and axon, the >25 mV depolarization associated with the plateau could propagate through the axon to influence action potential initiation, propagation, and neurotransmitter release. We examined this issue in brain slices, awake mice, and a computational model. Despite profoundly inactivating somatic and proximal axon Na(+) channels, plateaus evoked action potentials that recovered to full amplitude in the distal axon (>150 µm) and triggered neurotransmitter release similar to regular spiking. This effect was due to strong attenuation of plateau depolarizations by axonal K(+) channels, allowing full axon repolarization and Na(+) channel deinactivation. High-pass filtering of dendritic plateaus by axonal K(+) channels should thus enable accurate transmission of gain-modulated firing rates, allowing neuronal firing to be efficiently read out by downstream regions as a simple rate code.

Inhibitory Gating of Input Comparison in the CA1 Microcircuit.

Spatial and temporal features of synaptic inputs engage integration mechanisms on multiple scales, including presynaptic release sites, postsynaptic dendrites, and networks of inhibitory interneurons. Here we investigate how these mechanisms cooperate to filter synaptic input in hippocampal area CA1. Dendritic recordings from CA1 pyramidal neurons reveal that proximal inputs from CA3 as well as distal inputs from entorhinal cortex layer III (ECIII) sum sublinearly or linearly at low firing rates due to feedforward inhibition, but sum supralinearly at high firing rates due to synaptic facilitation, producing a high-pass filter. However, during ECIII and CA3 input comparison, supralinear dendritic integration is dynamically balanced by feedforward and feedback inhibition, resulting in suppression of dendritic complex spiking. We find that a particular subpopulation of CA1 interneurons expressing neuropeptide Y (NPY) contributes prominently to this dynamic filter by integrating both ECIII and CA3 input pathways and potently inhibiting CA1 pyramidal neuron dendrites.

Conjunctive input processing drives feature selectivity in hippocampal CA1 neurons.

Feature-selective firing allows networks to produce representations of the external and internal environments. Despite its importance, the mechanisms generating neuronal feature selectivity are incompletely understood. In many cortical microcircuits the integration of two functionally distinct inputs occurs nonlinearly through generation of active dendritic signals that drive burst firing and robust plasticity. To examine the role of this processing in feature selectivity, we recorded CA1 pyramidal neuron membrane potential and local field potential in mice running on a linear treadmill. We found that dendritic plateau potentials were produced by an interaction between properly timed input from entorhinal cortex and hippocampal CA3. These conjunctive signals positively modulated the firing of previously established place fields and rapidly induced new place field formation to produce feature selectivity in CA1 that is a function of both entorhinal cortex and CA3 input. Such selectivity could allow mixed network level representations that support context-dependent spatial maps.