Neuron Pruning Research Articles

Rab11 activation by Ik2 kinase is required for dendrite pruning in Drosophila sensory neurons.

Neuronal pruning is a commonly observed phenomenon for the developing nervous systems to ensure precise wiring of neural circuits. The function of Ik2 kinase and its downstream mediator, Spindle-F (Spn-F), are essential for dendrite pruning of Drosophila sensory neurons during development. However, little is known about how Ik2/Spn-F signaling is transduced in neurons and ultimately results in dendrite pruning. Our genetic analyses and rescue experiments demonstrated that the small GTPase Rab11, especially the active GTP-bound form, is required for dendrite pruning. We also found that Rab11 shows genetic interactions with spn-F and ik2 on pruning. Live imaging of single neurons and antibody staining reveal normal Ik2 kinase activation in Rab11 mutant neurons, suggesting that Rab11 could have a functional connection downstream of and/or parallel to the Ik2 kinase signaling. Moreover, we provide biochemical evidence that both the Ik2 kinase activity and the formation of Ik2/Spn-F/Rab11 complexes are central to promote Rab11 activation in cells. Together, our studies reveal that a critical role of Ik2/Spn-F signaling in neuronal pruning is to promote Rab11 activation, which is crucial for dendrite pruning in neurons.

Investigation and analysis of hyper and hypo neuron pruning to selectively update neurons during unsupervised adaptation

Unseen or out-of-domain data can seriously degrade the performance of a neural network model, indicating the model's failure to generalize to unseen data. Neural net pruning can not only help to reduce a model's size but can improve the model's generalization capacity as well. Pruning approaches look for low-salient neurons that are less contributive to a model's decision and hence can be removed from the model. This work investigates if pruning approaches are successful in detecting neurons that are either high-salient (mostly active or hyper) or low-salient (barely active or hypo), and whether removal of such neurons can help to improve the model's generalization capacity. Traditional blind adaptation techniques update either the whole or a subset of layers, but have never explored selectively updating individual neurons across one or more layers. Focusing on the fully connected layers of a convolutional neural network (CNN), this work shows that it may be possible to selectively adapt certain neurons (consisting of the hyper and the hypo neurons) first, followed by a full-network fine tuning. Using the task of automatic speech recognition, this work demonstrates how the removal of hyper and hypo neurons from a model can improve the model's performance on out-of-domain speech data and how selective neuron adaptation can ensure improved performance when compared to traditional blind model adaptation.

The membrane protein Raw regulates dendrite pruning via the secretory pathway.

Neuronal pruning is essential for proper wiring of the nervous systems in invertebrates and vertebrates. Drosophila ddaC sensory neurons selectively prune their larval dendrites to sculpt the nervous system during early metamorphosis. However, the molecular mechanisms underlying ddaC dendrite pruning remain elusive. Here, we identify an important and cell-autonomous role of the membrane protein Raw in dendrite pruning of ddaC neurons. Raw appears to regulate dendrite pruning via a novel mechanism, which is independent of JNK signaling. Importantly, we show that Raw promotes endocytosis and downregulation of the conserved L1-type cell-adhesion molecule Neuroglian (Nrg) prior to dendrite pruning. Moreover, Raw is required to modulate the secretory pathway by regulating the integrity of secretory organelles and efficient protein secretion. Mechanistically, Raw facilitates Nrg downregulation and dendrite pruning in part through regulation of the secretory pathway. Thus, this study reveals a JNK-independent role of Raw in regulating the secretory pathway and thereby promoting dendrite pruning.

NeST: A Neural Network Synthesis Tool Based on a Grow-and-Prune Paradigm

Deep neural networks (DNNs) have begun to have a pervasive impact on various applications of machine learning. However, the problem of finding an optimal DNN architecture for large applications is challenging. Common approaches go for deeper and larger DNN architectures but may incur substantial redundancy. To address these problems, we introduce a network growth algorithm that complements network pruning to learn both weights and compact DNN architectures during training. We propose a DNN synthesis tool (NeST) that combines both methods to automate the generation of compact and accurate DNNs. NeST starts with a randomly initialized sparse network called the seed architecture. It iteratively tunes the architecture with gradient-based growth and magnitude-based pruning of neurons and connections. Our experimental results show that NeST yields accurate, yet very compact DNNs, with a wide range of seed architecture selection. For the LeNet-300-100 (LeNet-5) architecture, we reduce network parameters by 70.2x (74.3x) and floating-point operations (FLOPs) by 79.4x (43.7x). For the AlexNet and VGG-16 architectures, we reduce network parameters (FLOPs) by 15.7x (4.6x) and 30.2x (8.6x), respectively. NeST's grow-and-prune paradigm delivers significant additional parameter and FLOPs reduction relative to pruning-only methods.

A Layer Decomposition-Recomposition Framework for Neuron Pruning towards Accurate Lightweight Networks

Neuron pruning is an efficient method to compress the network into a slimmer one for reducing the computational cost and storage overhead. Most of state-of-the-art results are obtained in a layer-by-layer optimization mode. It discards the unimportant input neurons and uses the survived ones to reconstruct the output neurons approaching to the original ones in a layer-by-layer manner. However, an unnoticed problem arises that the information loss is accumulated as layer increases since the survived neurons still do not encode the entire information as before. A better alternative is to propagate the entire useful information to reconstruct the pruned layer instead of directly discarding the less important neurons. To this end, we propose a novel Layer DecompositionRecomposition Framework (LDRF) for neuron pruning, by which each layer’s output information is recovered in an embedding space and then propagated to reconstruct the following pruned layers with useful information preserved. We mainly conduct our experiments on ILSVRC-12 benchmark with VGG-16 and ResNet-50. What should be emphasized is that our results before end-to-end fine-tuning are significantly superior owing to the information-preserving property of our proposed framework. With end-to-end fine-tuning, we achieve state-of-the-art results of 5.13× and 3× speed-up with only 0.5% and 0.65% top-5 accuracy drop respectively, which outperform the existing neuron pruning methods.

Abnormal Behavior of Zebrafish Mutant in Dopamine Transporter Is Rescued by Clozapine.

SummaryDopamine transporter (SLC6A3) deficiency causes infantile Parkinson disease, for which there is no effective therapy. We have explored the effects of genetically deleting SLC6A3 in zebrafish. Unlike the wild-type, slc6a3−/− fish hover near the tank bottom, with a repetitive digging-like behavior. slc6a3−/− fish manifest pruning and cellular loss of particular tyrosine hydroxylase-immunoreactive neurons in the midbrain. Clozapine, an effective therapeutic for treatment-resistant schizophrenia, rescues the abnormal behavior of slc6a3−/− fish. Clozapine also reverses the abnormalities in the A8 region of the mutant midbrain. By RNA sequencing analysis, clozapine increases the expression of erythropoietin pathway genes. Transgenic over-expression of erythropoietin in neurons of slc6a3−/− fish partially rescues the mutant behavior, suggesting a potential mechanistic basis for clozapine's efficacy.

Individual components of the SWI/SNF chromatin remodelling complex have distinct roles in memory neurons of the Drosophila mushroom body.

ABSTRACTTechnology has led to rapid progress in the identification of genes involved in neurodevelopmental disorders such as intellectual disability (ID), but our functional understanding of the causative genes is lagging. Here, we show that the SWI/SNF chromatin remodelling complex is one of the most over-represented cellular components disrupted in ID. We investigated the role of individual subunits of this large protein complex using targeted RNA interference in post-mitotic memory-forming neurons of the Drosophila mushroom body (MB). Knockdown flies were tested for defects in MB morphology, short-term memory and long-term memory. Using this approach, we identified distinct roles for individual subunits of the Drosophila SWI/SNF complex. Bap60, Snr1 and E(y)3 are required for pruning of the MBγ neurons during pupal morphogenesis, while Brm and Osa are required for survival of MBγ axons during ageing. We used the courtship conditioning assay to test the effect of MB-specific SWI/SNF knockdown on short- and long-term memory. Several subunits, including Brm, Bap60, Snr1 and E(y)3, were required in the MB for both short- and long-term memory. In contrast, Osa knockdown only reduced long-term memory. Our results suggest that individual components of the SWI/SNF complex have different roles in the regulation of structural plasticity, survival and functionality of post-mitotic MB neurons. This study highlights the many possible processes that might be disrupted in SWI/SNF-related ID disorders. Our broad phenotypic characterization provides a starting point for understanding SWI/SNF-mediated gene regulatory mechanisms that are important for development and function of post-mitotic neurons.

Delayed Spine Pruning of Direct Pathway Spiny Projection Neurons in a Mouse Model of Parkinson's Disease.

In animal models of Parkinson’s disease (PD), principal striatal spiny projection neurons (SPNs) lose axospinous synapses. However, there has been a disagreement about whether this loss is restricted to a specific type of SPN or not, as some studies have reported pruning in both direct pathway SPNs and indirect pathway SPNs, while others have found this pruning to be restricted to indirect pathway SPNs. One possible explanation for the discrepancy is the period between the induction of the parkinsonian state and the assessment of spine loss. To test this hypothesis, transgenic mice were subjected to unilateral 6-hydroxydopamine (6-OHDA) lesions of nigrostriatal dopaminergic neurons and then direct pathway SPNs examined in ex vivo brain slices using two photon laser scanning microscopy either one or 2 months afterwards. These studies revealed that 1 month after the lesion, there was no loss of spines in direct pathway SPNs. However, 2 months after the lesion, spine loss was significant in direct pathway SPNs. In addition to reconciling the existing literature on the impact of the parkinsonian state on axospinous synapse elimination in SPNs, our results suggest that the delayed spine loss in direct pathway SPNs is not driven by homeostatic mechanisms [as posited for indirect pathway (iSPNs)], but rather by network pathophysiology.

A Novel Dynamic Network Pruning via Smooth Initialization and its Potential Applications in Machine Learning Based Security Solutions

With the development of artificial intelligence technology, the demand for new digital security and privacy solutions is growing importantly. Inspired by the synaptic pruning in mammalian brains, we develop a network pruning method, called dynamic network pruning (DNP) method, which reduces the necessary number of free parameters in the convolutional neural networks. The DNP can be perfectly integrated with the gradient descent training and performed at any point, even multiple times, during training. We show that the pruning of connections (filters) is more intrinsic than the pruning of neurons (channels), and we relate the significance of filters to the dispersion of their values. With our proposed weight initialization technique, called smooth initialization (SI), unimportant filters can be easily identified based on the simple thresholds. The DNP method does not need pre-training to learn the connectivity of the network nor requires a very long time of fine-tuning to restore the performance. The DNP can also lead to better data privacy under distributed environment due to improved learning efficiency and convergency. The experiments show that our method outmatches several weight reduction methods in terms of reduction ratio and test accuracy on various models and datasets, and the generalization abilities of the pruned models are not damaged.

Purification of Recombinant ESCRT-III Proteins and Their Use in Atomic Force Microscopy and In Vitro Binding and Phosphorylation Assays.

The endosomal sorting complex required for transport (ESCRT)-III proteins are known to assemble into filaments that mediate membrane remodeling and fission in various biological processes, including the formation of endosomal multivesicular bodies, viral budding, cytokinesis, plasma membrane repair, nuclear pore quality control, nuclear envelope reformation, and neuron pruning. The study of the regulation and function of ESCRT-III proteins is therefore crucial to understand these events and requires a combination of in vivo and in vitro experimental techniques. Here we describe two protocols for the purification of human and Drosophila ESCRT-III proteins from bacteria and their use in in vitro phosphorylation assays and atomic force microscopy experiments on membrane lipid bilayers. These protocols can also be applied for the purification of other proteins that are insoluble when expressed in bacteria.

Regularization of deep neural networks with spectral dropout

The big breakthrough on the ImageNet challenge in 2012 was partially due to the ‘Dropout’ technique used to avoid overfitting. Here, we introduce a new approach called ‘Spectral Dropout’ to improve the generalization ability of deep neural networks. We cast the proposed approach in the form of regular Convolutional Neural Network (CNN) weight layers using a decorrelation transform with fixed basis functions. Our spectral dropout method prevents overfitting by eliminating weak and ‘noisy’ Fourier domain coefficients of the neural network activations, leading to remarkably better results than the current regularization methods. Furthermore, the proposed is very efficient due to the fixed basis functions used for spectral transformation. In particular, compared to Dropout and Drop-Connect, our method significantly speeds up the network convergence rate during the training process (roughly ×2), with considerably higher neuron pruning rates (an increase of ∼30%). We demonstrate that the spectral dropout can also be used in conjunction with other regularization approaches resulting in additional performance gains.

Prd1 associates with the clathrin adaptor α-Adaptin and the kinesin-3 Imac/Unc-104 to govern dendrite pruning in Drosophila.

Refinement of the nervous system depends on selective removal of excessive axons/dendrites, a process known as pruning. Drosophila ddaC sensory neurons prune their larval dendrites via endo-lysosomal degradation of the L1-type cell adhesion molecule (L1-CAM), Neuroglian (Nrg). Here, we have identified a novel gene, pruning defect 1 (prd1), which governs dendrite pruning of ddaC neurons. We show that Prd1 colocalizes with the clathrin adaptor protein α-Adaptin (α-Ada) and the kinesin-3 immaculate connections (Imac)/Uncoordinated-104 (Unc-104) in dendrites. Moreover, Prd1 physically associates with α-Ada and Imac, which are both critical for dendrite pruning. Prd1, α-Ada, and Imac promote dendrite pruning via the regulation of endo-lysosomal degradation of Nrg. Importantly, genetic interactions among prd1, α-adaptin, and imac indicate that they act in the same pathway to promote dendrite pruning. Our findings indicate that Prd1, α-Ada, and Imac act together to regulate discrete distribution of α-Ada/clathrin puncta, facilitate endo-lysosomal degradation, and thereby promote dendrite pruning in sensory neurons.

Neuroprotective Action of the CB1/2 Receptor Agonist, WIN 55,212-2, against DMSO but Not Phenobarbital-Induced Neurotoxicity in Immature Rats.

The developing brain is uniquely susceptible to drug-induced increases in programmed cell death or apoptosis. Many compounds, including anticonvulsant drugs, anesthetic agents, and ethanol, when administered in a narrow postnatal window in rodents, result in increased pruning of neurons. Here, we report that dimethyl sulfoxide (DMSO) triggers widespread neurodegeneration in the immature (postnatal day, P7) rat brain, an effect consistent with a prior report in neonatal mice. We found that the synthetic cannabinoid receptor agonist WIN 55,212-2 (WIN) exerts a neuroprotective effect against DMSO-induced cell death. We extended these findings to determine if WIN is neuroprotective against another drug class known to increase developmental cell death, namely antiseizure drugs. The antiseizure drug phenobarbital (PB) remains the primary treatment for neonatal seizures, despite significantly increasing cell death in the developing rodent brain. WIN exerts antiseizure effects in immature rodent seizure models, but increases the toxicity associated with neonatal ethanol exposure. We thus sought to determine if WIN would protect against or exacerbate PB-induced cell death. Unlike either the prior report with ethanol or our present findings with DMSO, WIN was largelywithout effect on PB-induced cell death. WIN alone did not increase cell death over levels observed in vehicle-treated rats. These data suggest that WIN has a favorable safety profile in the developing brain and could potentially serve as an adjunct therapy with phenobarbital (albeit one that does not attenuate PB-induced toxicity).

Differential Requirement for Translation Initiation Factor Pathways during Ecdysone-Dependent Neuronal Remodeling in Drosophila

Dendrite pruning of Drosophila sensory neurons during metamorphosis is induced by the steroid hormone ecdysone through a transcriptional program. In addition, ecdysone activates the eukaryotic initiation factor 4E-binding protein (4E-BP) to inhibit cap-dependent translation initiation. To uncover how efficient translation of ecdysone targets is achieved under these conditions, we assessed the requirements for translation initiation factors during dendrite pruning. We found that the canonical cap-binding complex eIF4F is dispensable for dendrite pruning, but the eIF3 complex and the helicase eIF4A are required, indicating that differential translation initiation mechanisms are operating during dendrite pruning. eIF4A and eIF3 are stringently required for translation of the ecdysone target Mical, and this depends on the 5' UTR of Mical mRNA. Functional analyses indicate that eIF4A regulates eIF3-mRNA interactions in a helicase-dependent manner. We propose that an eIF3-eIF4A-dependent alternative initiation pathway bypasses 4E-BP to ensure adequate translation of ecdysone-induced genes.

Differences in lumbar motor neuron pruning in an animal model of early onset spasticity.

Motor neuron (MN) development in early onset spasticity is poorly understood. For example, spastic cerebral palsy (sCP), the most common motor disability of childhood, is poorly predicted by brain imaging, yet research remains focused on the brain. By contrast, MNs, via the motor unit and neurotransmitter signaling, are the target of most therapeutic spasticity treatments and are the final common output of motor control. MN development in sCP is a critical knowledge gap, because the late embryonic and postnatal periods are not only when the supposed brain injury occurs but also are critical times for spinal cord neuromotor development. Using an animal model of early onset spasticity [ spa mouse (B6.Cg- Glrbspa/J) with a glycine (Gly) receptor mutation], we hypothesized that removal of effective glycinergic neurotransmitter inputs to MNs during development will influence MN pruning (including primary dendrites) and MN size. Spa (Glrb-/-) and wild-type (Glrb+/+) mice, ages 4-9 wk, underwent unilateral retrograde labeling of the tibialis anterior muscle MNs via peroneal nerve dip in tetramethylrhodamine. After 3 days, mice were euthanized and perfused with 4% paraformaldehyde, and the spinal cord was excised and processed for confocal imaging. Spa mice had ~61% fewer lumbar tibialis anterior MNs ( P < 0.01), disproportionately affecting larger MNs. Additionally, a ~23% reduction in tibialis anterior MN somal surface area ( P < 0.01) and a 12% increase in primary dendrites ( P = 0.046) were observed. Thus MN pruning and MN somal surface area are abnormal in early onset spasticity. Fewer and smaller MNs may contribute to the spastic phenotype. NEW & NOTEWORTHY Motor neuron (MN) development in early onset spasticity is poorly understood. In an animal model of early onset spasticity, spa mice, we found ~61% fewer lumbar tibialis anterior MNs compared with controls. This MN loss disproportionately affected larger MNs. Thus number and heterogeneity of the MN pool are decreased in spa mice, likely contributing to the spastic phenotype.

Myopic (HD-PTP, PTPN23) selectively regulates synaptic neuropeptide release

Neurotransmission is mediated by synaptic exocytosis of neuropeptide-containing dense-core vesicles (DCVs) and small-molecule transmitter-containing small synaptic vesicles (SSVs). Exocytosis of both vesicle types depends on Ca2+ and shared secretory proteins. Here, we show that increasing or decreasing expression of Myopic (mop, HD-PTP, PTPN23), a Bro1 domain-containing pseudophosphatase implicated in neuronal development and neuropeptide gene expression, increases synaptic neuropeptide stores at the Drosophila neuromuscular junction (NMJ). This occurs without altering DCV content or transport, but synaptic DCV number and age are increased. The effect on synaptic neuropeptide stores is accounted for by inhibition of activity-induced Ca2+-dependent neuropeptide release. cAMP-evoked Ca2+-independent synaptic neuropeptide release also requires optimal Myopic expression, showing that Myopic affects the DCV secretory machinery shared by cAMP and Ca2+ pathways. Presynaptic Myopic is abundant at early endosomes, but interaction with the endosomal sorting complex required for transport III (ESCRT III) protein (CHMP4/Shrub) that mediates Myopic's effect on neuron pruning is not required for control of neuropeptide release. Remarkably, in contrast to the effect on DCVs, Myopic does not affect release from SSVs. Therefore, Myopic selectively regulates synaptic DCV exocytosis that mediates peptidergic transmission at the NMJ.

Yif1 associates with Yip1 on Golgi and regulates dendrite pruning in sensory neurons during Drosophila metamorphosis.

Pruning that selectively removes unnecessary neurites without causing neuronal death is essential for sculpting the mature nervous system during development. In Drosophila, ddaC sensory neurons specifically prune their larval dendrites with intact axons during metamorphosis. However, the important role of endoplasmic reticulum (ER)-to-Golgi transport in dendrite pruning remains unknown. Here, in a clonal screen, we have identified Yif1, an uncharacterized Drosophila homolog of Yif1p that is known to be a regulator of ER-to-Golgi transport in yeast. We show that Yif1 is required for dendrite pruning of ddaC neurons but not for apoptosis of ddaF neurons. We further identify that the Yif1-binding partner Yip1 is also crucial for dendrite pruning. Yif1 forms a protein complex with Yip1 in S2 cells and ddaC neurons. Yip1 and Yif1 colocalize on ER/Golgi and are required for the integrity of Golgi apparatus and outposts. Moreover, we show that two GTPases, Rab1 and Sar1, which are known to regulate ER-to-Golgi transport, are essential for dendrite pruning of ddaC neurons. Finally, our data reveal that ER-to-Golgi transport promotes endocytosis and downregulation of the cell-adhesion molecule Neuroglian and thereby dendrite pruning.

Defect in Synaptic Pruning of Motor Cortex Neurons is Associated with Early Perturbed Dopaminergic System

Background: Early stress exposure during neurodevelopmental stages has been linked to some adult neuropsychiatric disorders. The dopaminergic system which has been implicated in movement and reward system has been linked to movement and mood disorders when perturbed at early development. This study is designed to check the mechanisms involved in movement disorders such as dyskinesia, associated with early perturbed dopaminergic system in the motor cortex Methods: Haloperidol was used to block D2R in neonatal albino Wistar rats in utero by administering 20 mg/kg BW (intraperitoneally) to pregnant adult wistar rats (n=8) in the third week of gestation. Behavioural studies such as the rotarod test were carried out on the neonatal animals (n=5) to test their motor function at postnatal day twentyeight (P28). Electrophysiological recordings were carried out on the motor cortex (M1) to determine the significance of D2R inhibition on calcium neural activity. Immunofluorescence was done to demonstrate synaptic vesicle protein (SV) and microtubule associated protein kinases (MAP K) as a measure of synapses count and microtubule phosphorylation respectively. Results: Behavioural studies showed a decline in motor function of animals exposed to haloperidol in utero compared to the control. This motor deficit was accompanied by a significant increase in the Ca2+ neural activity of the motor cortex as shown by electrophysiological recordings. Immunofluorescence staining showed there was significant increase in the number of MAPK+ and SV+ cells in the motor cortex of haloperidol exposed animals compared to the control. Conclusion: These findings showed that early perturbation in dopaminergic system is associated with an increase in synapses and neuronal density, as well as an increase in phosphorylation of microtubules of neurons in the motor cortex.

Building selective ensembles of Randomization Based Neural Networks with the successive projections algorithm

Randomization based methods for training neural networks have gained increasing attention in recent years and achieved remarkable performances on a wide variety of tasks. The interest in such methods relies on the fact that standard gradient based learning algorithms may often converge to local minima and are usually time consuming. Despite the good performance achieved by Randomization Based Neural Networks (RNNs), the random feature mapping procedure may generate redundant information, leading to suboptimal solutions. To overcome this problem, some strategies have been used such as feature selection, hidden neuron pruning and ensemble methods. Feature selection methods discard redundant information from the original dataset. Pruning methods eliminate hidden nodes with redundant information. Ensemble methods combine multiple models to generate a single one. Selective ensemble methods select a subset of all available models to generate the final model. In this paper, we propose a selective ensemble of RNNs based on the Successive Projections Algorithm (SPA), for regression problems. The proposed method, named Selective Ensemble of RNNs using the Successive projections algorithm (SERS), employs the SPA for three distinct tasks: feature selection, pruning and ensemble selection. SPA was originally developed as a feature selection technique and has been recently employed for RNN pruning. Herein, we show that it can also be employed for ensemble selection. The proposed framework was used to develop three selective ensemble models based on the three RNNs: Extreme Learning Machines (ELM), Feedforward Neural Network with Random Weights (FNNRW) and Random Vector Functional Link (RVFL). The performances of SERS-ELM, SERS-FNNRW and SERS-RVFL were assessed in terms of model accuracy and model complexity in several real world benchmark problems. Comparisons to related methods showed that SERS variants achieved similar accuracies with significant model complexity reduction. Among the proposed models, SERS-RVFL had the best accuracies and all variants had similar model complexities.

Sec71 functions as a GEF for the small GTPase Arf1 to govern dendrite pruning of Drosophila sensory neurons

Sec71 functions as a GEF for the small GTPase Arf1 to govern dendrite pruning of Drosophila sensory neurons