Code-specific learning rules improve action selection by populations of spiking neurons

Population coding is widely regarded as a key mechanism for achieving reliable behavioral decisions. We previously introduced reinforcement learning for population-based decision making by spiking neurons. Here we generalize population reinforcement learning to spike-based plasticity rules that take account of the postsynaptic neural code. We consider spike/no-spike, spike count and spike latency codes. The multi-valued and continuous-valued features in the postsynaptic code allow for a generalization of binary decision making to multi-valued decision making and continuous-valued action selection. We show that code-specific learning rules speed up learning both for the discrete classification and the continuous regression tasks. The suggested learning rules also speed up with increasing population size as opposed to standard reinforcement learning rules. Continuous action selection is further shown to explain realistic learning speeds in the Morris water maze. Finally, we introduce the concept of action perturbation as opposed to the classical weight- or node-perturbation as an exploration mechanism underlying reinforcement learning. Exploration in the action space greatly increases the speed of learning as compared to exploration in the neuron or weight space.

“Fitness Fingerprints” Mediate Physiological Culling of Unwanted Neurons in Drosophila

BACKGROUND: The flower gene has been previously linked to the elimination of slow dividing epithelial cells during development in a process known as "cell competition." During cell competition, different isoforms of the Flower protein are displayed at the cell membrane and reveal the reduced fitness of slow proliferating cells, which are therefore recognized, eliminated, and replaced by their normally dividing neighbors. This mechanism acts as a "cell quality" control in proliferating tissues. RESULTS: Here, we use the Drosophila eye as a model to study how unwanted neurons are culled during retina development and find that flower is required and sufficient for the recognition and elimination of supernumerary postmitotic neurons, contained within incomplete ommatidia units. This constitutes the first description of the "Flower Code" functioning as a cell selection mechanism in postmitotic cells and is also the first report of a physiological role for this cell quality control machinery. CONCLUSIONS: Our results show that the "Flower Code" is a general system to reveal cell fitness and that it may play similar roles in creating optimal neural networks in higher organisms. The Flower Code seems to be a more general mechanism for cell monitoring and selection than previously recognized.

Antagonizing Nogo-receptor 1 promotes the number of cultured dopaminergic neurons and elongates their neurites

The myelin-associated protein Nogo-A and its receptor Nogo-receptor 1 (NgR1) are known as potent growth inhibitors of the adult central nervous system (CNS). Nogo-A is mostly expressed on the surface of oligodendrocytes, but is also found in neurons of the adult and developing CNS. This observation suggests that Nogo-A serves additional functions in the brain. Hence, in the present study, we investigated the effects of antagonizing NgR1 on cultured organotypic and dissociated dopaminergic neurons. For that purpose ventral mesencephalic cultures from E14 rat embryos were grown in absence or presence of the NgR1 antagonist NEP1-40 for 1 week. Treatment with NEP1-40 significantly increased cell densities of tyrosine hydroxylase-immunoreactive neurons. Moreover, organotypic ventral mesencephalic cultures displayed a significantly bigger volume after NEP1-40 treatment. Morphological analysis of tyrosine hydroxylase-positive neurons disclosed longer neurites and higher numbers of primary neurites in dissociated cultures incubated with NEP1-40, whereas soma size was not changed. In conclusion, our findings demonstrate that interfering with Nogo-A signaling by antagonizing NgR1 modulates dopaminergic neuron properties during development. These observations highlight novel aspects of the role of Nogo-A in the CNS and might have an impact in the context of Parkinson's disease.

Quantitation of contacts among sensory, motor, and serotonergic neurons in the pedal ganglion of aplysia.

Present models of long-term sensitization in Aplysia californica indicate that the enhanced behavioral response is due, at least in part, to outgrowth of sensory neurons mediating defensive withdrawal reflexes. Presumably, this outgrowth strengthens pre-existing connections by formation of new synapses with follower neurons. However, the relationship between the number of sensorimotor contacts and the physiological strength of the connection has never been examined in intact ganglia. As a first step in addressing this issue, we used confocal microscopy to examine sites of contact between sensory and motor neurons in naive animals. Our results revealed relatively few contacts between physiologically connected cells. In addition, the number of contact sites was proportional to the amplitude of the EPSP elicited in the follower motor neuron by direct stimulation of the sensory neuron. This is the first time such a correlation has been observed in the central nervous system. Serotonin is the neurotransmitter most closely examined for its role in modulating synaptic strength at the sensorimotor synapse. However, the structural relationship of serotonergic processes and sensorimotor synapses has never been examined. Surprisingly, serotonergic processes usually made contact with sensory and motor neurons at sites located relatively distant from the sensorimotor synapse. This result implies that heterosynaptic regulation is due to nondirected release of serotonin into the neuropil.

TGF-beta1 in Aplysia: role in long-term changes in the excitability of sensory neurons and distribution of TbetaR-II-like immunoreactivity.

Exogenous recombinant human transforming growth factor beta-1 (TGF-beta1) induced long-term facilitation of Aplysia sensory-motor synapses. In addition, 5-HT-induced facilitation was blocked by application of a soluble fragment of the extracellular portion of the TGF-beta1 type II receptor (TbetaR-II), which presumably acted by scavenging an endogenous TGF-beta1-like molecule. Because TbetaR-II is essential for transmembrane signaling by TGF-beta, we sought to determine whether Aplysia tissues contained TbetaR-II and specifically, whether neurons expressed the receptor. Western blot analysis of Aplysia tissue extracts demonstrated the presence of a TbetaR-II-immunoreactive protein in several tissue types. The expression and distribution of TbetaR-II-immunoreactive proteins in the central nervous system was examined by immunohistochemistry to elucidate sites that may be responsive to TGF-beta1 and thus may play a role in synaptic plasticity. Sensory neurons in the ventral-caudal cluster of the pleural ganglion were immunoreactive for TbetaR-II, as well as many neurons in the pedal, abdominal, buccal, and cerebral ganglia. Sensory neurons cultured in isolation and cocultured sensory and motor neurons were also immunoreactive. TGF-beta1 affected the biophysical properties of cultured sensory neurons, inducing an increase of excitability that persisted for at least 48 hr. Furthermore, exposure to TGF-beta1 resulted in a reduction in the firing threshold of sensory neurons. These results provide further support for the hypothesis that TGF-beta1 plays a role in long-term synaptic plasticity in Aplysia.

SIGNALING MECHANISMS THAT CONTROL GAP JUNCTIONAL COUPLING BETWEEN RETINAL NEURONS

Gap junctions between neurons form the structural substrate for electrical synapses. Connexin 36 (Cx36, and its non-mammalian ortholog connexin 35) is the major neuronal gap junction protein in the central nervous system (CNS), and contributes to several important neuronal functions including neuronal synchronization, signal averaging, network oscillations, and motor learning. Connexin 36 is strongly expressed in the retina, where it is an obligatory component of the high-sensitivity rod photoreceptor pathway. A fundamental requirement of the retina is to adapt to broadly varying inputs in order to maintain a dynamic range of signaling output. Modulation of the strength of electrical coupling between networks of retinal neurons, including the Cx36-coupled AII amacrine cell in the primary rod circuit, is a hallmark of retinal luminance adaptation. However, very little is known about the mechanisms regulating dynamic modulation of Cx36-mediated coupling. The primary goal of this work was to understand how cellular signaling mechanisms regulate coupling through Cx36 gap junctions. We began by developing and characterizing phospho-specific antibodies against key regulatory phosphorylation sites on Cx36. Using these tools we showed that phosphorylation of Cx35 in fish models varies with light adaptation state, and is modulated by acute changes in background illumination. We next turned our focus to the well-studied and readily identifiable AII amacrine cell in mammalian retina. Using this model we showed that increased phosphorylation of Cx36 is directly related to increased coupling through these gap junctions, and that the dopamine-stimulated uncoupling of the AII network is mediated by dephosphorylation of Cx36 via protein kinase A-stimulated protein phosphatase 2A activity. We then showed that increased phosphorylation of Cx36 on the AII amacrine network is driven by depolarization of presynaptic ON-type bipolar cells as well as background light increments. This increase in phosphorylation is mediated by activation of extrasynaptic NMDA receptors associated with Cx36 gap junctions on AII amacrine cells and by Ca2+-calmodulin-dependent protein kinase II activation. Finally, these studies indicated that coupling is regulated locally at individual gap junction plaques. This work provides a framework for future study of regulation of Cx36-mediated coupling, in which increased phosphorylation of Cx36 indicates increased neuronal coupling.

Inhibition of dendritic Ca2+ spikes by GABAB receptors in cortical pyramidal neurons is mediated by a direct Gi/o- -subunit interaction with Cav1 channels

Voltage-dependent calcium channels (VDCCs) serve a wide range of physiological functions and their activity is modulated by different neurotransmitter systems. GABAergic inhibition of VDCCs in neurons has an important impact in controlling transmitter release, neuronal plasticity, gene expression and neuronal excitability. We investigated the molecular signalling mechanisms by which GABAB receptors inhibit calcium-mediated electrogenesis (Ca2+ spikes) in the distal apical dendrite of cortical layer 5 pyramidal neurons. Ca2+ spikes are the basis of coincidence detection and signal amplification of distal tuft synaptic inputs characteristic for the computational function of cortical pyramidal neurons. By combining dendritic whole-cell recordings with two-photon fluorescence Ca2+ imaging we found that all subtypes of VDCCs were present in the Ca2+ spike initiation zone, but that they contribute differently to the initiation and sustaining of dendritic Ca2+ spikes. Particularly, Cav1 VDCCs are the most abundant VDCC present in this dendritic compartment and they generated the sustained plateau potential characteristic for the Ca2+ spike. Activation of GABAB receptors specifically inhibited Cav1 channels. This inhibition of L-type Ca2+ currents was transiently relieved by strong depolarization but did not depend on protein kinase activity. Therefore, our findings suggest a novel membrane-delimited interaction of the Gi/o-βγ-subunit with Cav1 channels identifying this mechanism as the general pathway of GABAB receptor-mediated inhibition of VDCCs. Furthermore, the characterization of the contribution of the different VDCCs to the generation of the Ca2+ spike provides new insights into the molecular mechanism of dendritic computation.

Polysensory interneuronal projections to foot contractile pedal neurons in Hermissenda.

A Pavlovian-conditioning procedure may produce modifications in multiple behavioral responses. As an example, conditioning may result in the elicitation of a specific somatomotor conditioned response (CR) and, in addition, other motor and visceral CRs. In the mollusk Hermissenda conditioning produces two conditioned responses: foot-shortening and decreased locomotion. The neural circuitry supporting ciliary locomotion is well characterized, although the neural circuit underlying foot-shortening is poorly understood. Here we describe efferent neurons in the pedal ganglion that produce contraction or extension of specific regions of the foot in semi-intact preparations. Synaptic connections between polysensory type Ib and type Is interneurons and identified foot contractile efferent neurons were examined. Type Ib and type Is interneurons receive synaptic input from the visual, graviceptive, and somatosensory systems. Depolarization of type Ib interneurons evoked spikes in identified tail and lateral foot contractile efferent neurons. Mechanical displacement of the statocyst evoked complex excitatory postsynaptic potentials (EPSPs) and spikes recorded from type Ib and type Is interneurons and complex EPSPs and spikes in identified foot contractile efferent neurons. Depolarization of type Ib interneurons in semi-intact preparations produced contraction and shortening along the rostrocaudal axis of the foot. Depolarization of Is interneurons in semi-intact preparations produced contraction of the anterior region of the foot. Taken collectively, the results suggest that type Ib and type Is polysensory interneurons may contribute to the neural circuit underlying the foot-shortening CR in Hermissenda.

Blocking brain-derived neurotrophic factor inhibits injury-induced hyperexcitability of hippocampal CA3 neurons

Brain trauma can disrupt synaptic connections, and this in turn can prompt axons to sprout and form new connections. If these new axonal connections are aberrant, hyperexcitability can result. It has been shown that ablating tropomyosin-related kinase B (TrkB), a receptor for brain-derived neurotrophic factor (BDNF), can reduce axonal sprouting after hippocampal injury. However, it is unknown whether inhibiting BDNF-mediated axonal sprouting will reduce hyperexcitability. Given this, our purpose here was to determine whether pharmacologically blocking BDNF inhibits hyperexcitability after injury-induced axonal sprouting in the hippocampus. To induce injury, we made Schaffer collateral lesions in organotypic hippocampal slice cultures. As reported by others, we observed a 50% reduction in axonal sprouting in cultures treated with a BDNF blocker (TrkB-Fc) 14 days after injury. Furthermore, lesioned cultures treated with TrkB-Fc were less hyperexcitable than lesioned untreated cultures. Using electrophysiology, we observed a two-fold decrease in the number of CA3 neurons that showed bursting responses after lesion with TrkB-Fc treatment, whereas we found no change in intrinsic neuronal firing properties. Finally, evoked field excitatory postsynaptic potential recordings indicated an increase in network activity within area CA3 after lesion, which was prevented with chronic TrkB-Fc treatment. Taken together, our results demonstrate that blocking BDNF attenuates injury-induced hyperexcitability of hippocampal CA3 neurons. Axonal sprouting has been found in patients with post-traumatic epilepsy. Therefore, our data suggest that blocking the BDNF-TrkB signaling cascade shortly after injury may be a potential therapeutic target for the treatment of post-traumatic epilepsy.

ROLE OF THE GCN5 HISTONE ACETYLTRANSFERASE IN SPINOCEREBELLAR ATAXIA TYPE 7 AND IN IMMATURE NEURONS

Spinocerebellar Ataxia type 7 (SCA7) is a neurodegenerative disease caused by expansion of a CAG repeat encoding a polyglutamine tract in ATXN7, a component of the SAGA histone acetyltransferase (HAT) complex. Previous studies provided conflicting evidence regarding the effects of polyQ-ATXN7 on the activity of Gcn5, the HAT catalytic subunit of SAGA. Here I showed that reducing Gcn5 expression accelerates both cerebellar and retinal degeneration in a mouse model of SCA7. Deletion of Gcn5 in Purkinje cells in mice expressing wild type Atxn7, however, causes only mild ataxia and does not lead to the early lethality observed in SCA7 mice. Reduced Gcn5 expression strongly enhances retinopathy in SCA7 mice, but does not affect the transcriptional targets of Atxn7, as expression of these genes is not further altered by Gcn5 depletion. These findings demonstrate that loss of Gcn5 functions can contribute to the time of onset and severity of SCA7 phenotypes, but suggest that non-transcriptional functions of SAGA may play a role in neurodegeneration in this disease.

Long-term injury induced plasticity in {\it Aplysia\/} neurons: Behavioral, electrophysiological, and morphological studies

Nerve injury is known to produce a variety of electrophysiological and morphological neuronal alterations (reviewed by Titmus and Faber, 1990; Bulloch and Ridgeway, 1989; Walters, 1994). Determining if these alterations are adaptive and how they are activated and maintained could provide important insight into basic cellular mechanisms of injury-induced plasticity. Furthermore, characterization of injury-induced plasticity provides a useful assay system for the identification of possible induction signals underlying these neuronal changes. Understanding fundamental mechanisms and underlying induction signals of injury-induced neuronal plasticity could facilitate development of treatment strategies for neural injury and neuropathic pain in humans.^ This dissertation characterizes long-lasting, injury-induced neuronal alterations using the nervous system of Aplysia californica as a model. These changes are examined at the behavioral, electrophysiological, and morphological levels. Injury-induced changes in the electrophysiological properties of neurons were found that increased the signaling effectiveness of the injured neurons. This increase in signalling effectiveness could act to compensate for partial destruction of the injured neuron's peripheral processes. Recovery of a defensive behavioral response which serves to protect the animal from further injury was found within 2 weeks of injury. For the behavioral recovery to occur, new neural pathways must have been formed between the denervated area and the CNS. This was found to be mediated at least in part by new axonal growth which extended from the injured cell back along the original pathway (i.e. into the injured nerve). In addition, injury produced central axonal sprouting into different nerves that do not usually contain the injured neuron's axons. This could be important for (i) finding alternative pathways to the periphery when the original pathways are impassable and (ii) the formation of additional synaptic connections with post-synaptic targets which would further enhance the signalling effectiveness of the injured cell. ^