An Augmented Two-Layer Model Captures Nonlinear Analog Spatial Integration Effects in Pyramidal Neuron Dendrites

Probing multiplexed basal dendritic computations using two-photon 3D holographic uncaging

Basal dendrites of layer 5 cortical pyramidal neurons exhibit Na+ and N-methyl-D-aspartate receptor (NMDAR) regenerative spikes and are uniquely poised to influence somatic output. Nevertheless, due to technical limitations, how multibranch basal dendritic integration shapes and enables multiplexed barcoding of synaptic streams remains poorly mapped. Here, we combine 3D two-photon holographic transmitter uncaging, whole-cell dynamic clamp, and biophysical modeling to reveal how synchronously activated synapses (distributed and clustered) across multiple basal dendritic branches are multiplexed under quiescent and in vivo-like conditions. While dendritic regenerative Na+ spikes promote millisecond somatic spike precision, distributed synaptic inputs and NMDAR spikes regulate gain. These concomitantly occurring dendritic nonlinearities enable multiplexed information transfer amid an ongoing noisy background, including under back-propagating voltage resets, by barcoding the axo-somatic spike structure. Our results unveil a multibranch dendritic integration framework in which dendritic nonlinearities are critical for multiplexing different spatial-temporal synaptic input patterns, enabling optimal feature binding.

Cellular mechanisms of cooperative context-sensitive predictive inference

We argue that prediction success maximization is a basic objective of cognition and cortex, that it is compatible with but distinct from prediction error minimization, that neither objective requires subtractive coding, that there is clear neurobiological evidence for the amplification of predicted signals, and that we are unconvinced by evidence proposed in support of subtractive coding. We outline recent discoveries showing that pyramidal cells on which our cognitive capabilities depend usually transmit information about input to their basal dendrites and amplify that transmission when input to their distal apical dendrites provides a context that agrees with the feedforward basal input in that both are depolarizing, i.e., both are excitatory rather than inhibitory. Though these intracellular discoveries require a level of technical expertise that is beyond the current abilities of most neuroscience labs, they are not controversial and acclaimed as groundbreaking. We note that this cellular cooperative context-sensitivity greatly enhances the cognitive capabilities of the mammalian neocortex, and that much remains to be discovered concerning its evolution, development, and pathology.

An operating principle of the cerebral cortex, and a cellular mechanism for attentional trial-and-error pattern learning and useful classification extraction

A feature of the brains of intelligent animals is the ability to learn to respond to an ensemble of active neuronal inputs with a behaviorally appropriate ensemble of active neuronal outputs. Previously, a hypothesis was proposed on how this mechanism is implemented at the cellular level within the neocortical pyramidal neuron: the apical tuft or perisomatic inputs initiate "guess" neuron firings, while the basal dendrites identify input patterns based on excited synaptic clusters, with the cluster excitation strength adjusted based on reward feedback. This simple mechanism allows neurons to learn to classify their inputs in a surprisingly intelligent manner. Here, we revise and extend this hypothesis. We modify synaptic plasticity rules to align with behavioral time scale synaptic plasticity (BTSP) observed in hippocampal area CA1, making the framework more biophysically and behaviorally plausible. The neurons for the guess firings are selected in a voluntary manner via feedback connections to apical tufts in the neocortical layer 1, leading to dendritic Ca2+ spikes with burst firing, which are postulated to be neural correlates of attentional, aware processing. Once learned, the neuronal input classification is executed without voluntary or conscious control, enabling hierarchical incremental learning of classifications that is effective in our inherently classifiable world. In addition to voluntary, we propose that pyramidal neuron burst firing can be involuntary, also initiated via apical tuft inputs, drawing attention toward important cues such as novelty and noxious stimuli. We classify the excitations of neocortical pyramidal neurons into four categories based on their excitation pathway: attentional versus automatic and voluntary/acquired versus involuntary. Additionally, we hypothesize that dendrites within pyramidal neuron minicolumn bundles are coupled via depolarization cross-induction, enabling minicolumn functions such as the creation of powerful hierarchical "hyperneurons" and the internal representation of the external world. We suggest building blocks to extend the microcircuit theory to network-level processing, which, interestingly, yields variants resembling the artificial neural networks currently in use. On a more speculative note, we conjecture that principles of intelligence in universes governed by certain types of physical laws might resemble ours.

Asymmetric Voltage Attenuation in Dendrites Can Enable Hierarchical Heterosynaptic Plasticity

Long-term synaptic plasticity is mediated via cytosolic calcium concentrations ([Ca2+]). Using a synaptic model that implements calcium-based long-term plasticity via two sources of Ca2+ - NMDA receptors and voltage-gated calcium channels (VGCCs) - we show in dendritic cable simulations that the interplay between these two calcium sources can result in a diverse array of heterosynaptic effects. When spatially clustered synaptic input produces a local NMDA spike, the resulting dendritic depolarization can activate VGCCs at nonactivated spines, resulting in heterosynaptic plasticity. NMDA spike activation at a given dendritic location will tend to depolarize dendritic regions that are located distally to the input site more than dendritic sites that are proximal to it. This asymmetry can produce a hierarchical effect in branching dendrites, where an NMDA spike at a proximal branch can induce heterosynaptic plasticity primarily at branches that are distal to it. We also explored how simultaneously activated synaptic clusters located at different dendritic locations synergistically affect the plasticity at the active synapses, as well as the heterosynaptic plasticity of an inactive synapse "sandwiched" between them. We conclude that the inherent electrical asymmetry of dendritic trees enables sophisticated schemes for spatially targeted supervision of heterosynaptic plasticity.

Parallel functional architectures within a single dendritic tree

The input-output transformation of individual neurons is a key building block of neural circuit dynamics. While previous models of this transformation vary widely in their complexity, they all describe the underlying functional architecture as unitary, such that each synaptic input makes a single contribution to the neuronal response. Here, we show that the input-output transformation of CA1 pyramidal cells is instead best captured by two distinct functional architectures operating in parallel. We used statistically principled methods to fit flexible, yet interpretable, models of the transformation of input spikes into the somatic "output" voltage and to automatically select among alternative functional architectures. With dendritic Na+ channels blocked, responses are accurately captured by a single static and global nonlinearity. In contrast, dendritic Na+-dependent integration requires a functional architecture with multiple dynamic nonlinearities and clustered connectivity. These two architectures incorporate distinct morphological and biophysical properties of the neuron and its synaptic organization.

Classical-Contextual Interactions in V1 May Rely on Dendritic Computations

A signature feature of the neocortex is the dense network of horizontal connections (HCs) through which pyramidal neurons (PNs) exchange "contextual" information. In primary visual cortex (V1), HCs are thought to facilitate boundary detection, a crucial operation for object recognition, but how HCs modulate PN responses to boundary cues within their classical receptive fields (CRF) remains unknown. We began by "asking" natural images, through a structured data collection and ground truth labeling process, what function a V1 cell should use to compute boundary probability from aligned edge cues within and outside its CRF. The "answer" was an asymmetric 2-D sigmoidal function, whose nonlinear form provides the first normative account for the "multiplicative" center-flanker interactions previously reported in V1 neurons (Kapadia et al., 1995, 2000; Polat et al., 1998). Using a detailed compartmental model, we then show that this boundary-detecting classical-contextual interaction function can be computed by NMDAR-dependent spatial synaptic interactions within PN dendrites - the site where classical and contextual inputs first converge in the cortex. In additional simulations, we show that local interneuron circuitry activated by HCs can powerfully leverage the nonlinear spatial computing capabilities of PN dendrites, providing the cortex with a highly flexible substrate for integration of classical and contextual information.

Avoiding Catastrophe: Active Dendrites Enable Multi-Task Learning in Dynamic Environments

A key challenge for AI is to build embodied systems that operate in dynamically changing environments. Such systems must adapt to changing task contexts and learn continuously. Although standard deep learning systems achieve state of the art results on static benchmarks, they often struggle in dynamic scenarios. In these settings, error signals from multiple contexts can interfere with one another, ultimately leading to a phenomenon known as catastrophic forgetting. In this article we investigate biologically inspired architectures as solutions to these problems. Specifically, we show that the biophysical properties of dendrites and local inhibitory systems enable networks to dynamically restrict and route information in a context-specific manner. Our key contributions are as follows: first, we propose a novel artificial neural network architecture that incorporates active dendrites and sparse representations into the standard deep learning framework. Next, we study the performance of this architecture on two separate benchmarks requiring task-based adaptation: Meta-World, a multi-task reinforcement learning environment where a robotic agent must learn to solve a variety of manipulation tasks simultaneously; and a continual learning benchmark in which the model's prediction task changes throughout training. Analysis on both benchmarks demonstrates the emergence of overlapping but distinct and sparse subnetworks, allowing the system to fluidly learn multiple tasks with minimal forgetting. Our neural implementation marks the first time a single architecture has achieved competitive results in both multi-task and continual learning settings. Our research sheds light on how biological properties of neurons can inform deep learning systems to address dynamic scenarios that are typically impossible for traditional ANNs to solve.

Parallel processing in active dendrites during periods of intense spiking activity

A neuron's ability to perform parallel computations throughout its dendritic arbor substantially improves its computational capacity. However, during natural patterns of activity, the degree to which computations remain compartmentalized, especially in neurons with active dendritic trees, is not clear. Here, we examine how the direction of moving objects is computed across the bistratified dendritic arbors of ON-OFF direction-selective ganglion cells (DSGCs) in the mouse retina. We find that although local synaptic signals propagate efficiently throughout their dendritic trees, direction-selective computations in one part of the dendritic arbor have little effect on those being made elsewhere. Independent dendritic processing allows DSGCs to compute the direction of moving objects multiple times as they traverse their receptive fields, enabling them to rapidly detect changes in motion direction on a sub-receptive-field basis. These results demonstrate that the parallel processing capacity of neurons can be maintained even during periods of intense synaptic activity.

GABAergic Interneurons with Nonlinear Dendrites: From Neuronal Computations to Memory Engrams

GABAergic interneurons (INs) are a highly diverse class of neurons in the mammalian brain with a critical role in orchestrating multiple cognitive functions and maintaining the balance of excitation/inhibition across neuronal circuitries. In this perspective, we discuss recent findings regarding the ability of some IN subtypes to integrate incoming inputs in nonlinear ways within their dendritic branches. These recently discovered features may endow the specific INs with advanced computing capabilities, whose breadth and functional contributions remain an open question. Along these lines, we discuss theoretical and experimental evidence regarding the potential role of nonlinear IN dendrites in advancing single neuron computations and contributing to memory formation.

Diverse processing underlying frequency integration in midbrain neurons of barn owls

Emergent response properties of sensory neurons depend on circuit connectivity and somatodendritic processing. Neurons of the barn owl’s external nucleus of the inferior colliculus (ICx) display emergence of spatial selectivity. These neurons use interaural time difference (ITD) as a cue for the horizontal direction of sound sources. ITD is detected by upstream brainstem neurons with narrow frequency tuning, resulting in spatially ambiguous responses. This spatial ambiguity is resolved by ICx neurons integrating inputs over frequency, a relevant processing in sound localization across species. Previous models have predicted that ICx neurons function as point neurons that linearly integrate inputs across frequency. However, the complex dendritic trees and spines of ICx neurons raises the question of whether this prediction is accurate. Data from in vivo intracellular recordings of ICx neurons were used to address this question. Results revealed diverse frequency integration properties, where some ICx neurons showed responses consistent with the point neuron hypothesis and others with nonlinear dendritic integration. Modeling showed that varied connectivity patterns and forms of dendritic processing may underlie observed ICx neurons’ frequency integration processing. These results corroborate the ability of neurons with complex dendritic trees to implement diverse linear and nonlinear integration of synaptic inputs, of relevance for adaptive coding and learning, and supporting a fundamental mechanism in sound localization.

Neurons at higher stages of sensory pathways often display selectivity for properties of sensory stimuli that result from computations performed within the nervous system. These emergent response properties can be produced by patterns of neural connectivity and processing that occur within individual cells. Here we investigated whether neural connectivity and single-neuron computation may contribute to the emergence of spatial selectivity in auditory neurons in the barn owl’s midbrain. We used data from in vivo intracellular recordings to test the hypothesis from previous modeling work that these cells function as point neurons that perform a linear sum of their inputs in their subthreshold responses. Results indicate that while some neurons show responses consistent with the point neuron hypothesis, others match predictions of nonlinear integration, indicating a diversity of frequency integration properties across neurons. Modeling further showed that varied connectivity patterns and forms of single-neuron computation may underlie observed responses. These results demonstrate that neurons with complex morphologies may implement diverse integration of synaptic inputs, relevant for adaptive coding and learning.

Burst control: Synaptic conditions for burst generation in cortical layer 5 pyramidal neurons

The output of neocortical layer 5 pyramidal cells (L5PCs) is expressed by a train of single spikes with intermittent bursts of multiple spikes at high frequencies. The bursts are the result of nonlinear dendritic properties, including Na+, Ca2+, and NMDA spikes, that interact with the ~10,000 synapses impinging on the neuron's dendrites. Output spike bursts are thought to implement key dendritic computations, such as coincidence detection of bottom-up inputs (arriving mostly at the basal tree) and top-down inputs (arriving mostly at the apical tree). In this study we used a detailed nonlinear model of L5PC receiving excitatory and inhibitory synaptic inputs to explore the conditions for generating bursts and for modulating their properties. We established the excitatory input conditions on the basal versus the apical tree that favor burst and show that there are two distinct types of bursts. Bursts consisting of 3 or more spikes firing at < 200 Hz, which are generated by stronger excitatory input to the basal versus the apical tree, and bursts of ~2-spikes at ~250 Hz, generated by prominent apical tuft excitation. Localized and well-timed dendritic inhibition on the apical tree differentially modulates Na+, Ca2+, and NMDA spikes and, consequently, finely controls the burst output. Finally, we explored the implications of different burst classes and respective dendritic inhibition for regulating synaptic plasticity.

A synaptic learning rule for exploiting nonlinear dendritic computation

Information processing in the brain depends on the integration of synaptic input distributed throughout neuronal dendrites. Dendritic integration is a hierarchical process, proposed to be equivalent to integration by a multilayer network, potentially endowing single neurons with substantial computational power. However, whether neurons can learn to harness dendritic properties to realize this potential is unknown. Here, we develop a learning rule from dendritic cable theory and use it to investigate the processing capacity of a detailed pyramidal neuron model. We show that computations using spatial or temporal features of synaptic input patterns can be learned, and even synergistically combined, to solve a canonical nonlinear feature-binding problem. The voltage dependence of the learning rule drives coactive synapses to engage dendritic nonlinearities, whereas spike-timing dependence shapes the time course of subthreshold potentials. Dendritic input-output relationships can therefore be flexibly tuned through synaptic plasticity, allowing optimal implementation of nonlinear functions by single neurons.

Optimizing a Neuron for Reliable Dendritic Subunit Pooling

In certain biologically relevant computing scenarios, a neuron "pools" the outputs of multiple independent functional subunits, firing if any one of them crosses threshold. Recent studies suggest that active dendrites could provide the thresholding mechanism, so that both the thresholding and pooling operations could take place within a single neuron. A pooling neuron faces a difficult task, however. Dendrites can produce highly variable responses depending on the density and spatial patterning of their synaptic inputs, and bona fide dendritic firing may be very rare, making it difficult for a neuron to reliably detect when one of its many dendrites has "gone suprathreshold". Our goal has been to identify biological adaptations that optimize a neuron's performance at the binary subunit pooling (BSP) task. Katz et al. (2009) pointed to the importance of spine density gradients in shaping dendritic responses. In a similar vein, we used a compartmental model to study how a neuron's performance at the BSP task is affected by different spine density layouts and other biological variables. We found BSP performance was optimized when dendrites have (1) a decreasing spine density gradient (true for many types of pyramidal neurons); (2) low-to-medium resistance spine necks; (3) strong NMDA currents; (4) fast spiking Na+ channels; and (5) powerful hyperpolarizing inhibition. Our findings provide a normative account that links several neuronal properties within the context of a behaviorally relevant task, and thus provide new insights into nature's subtle strategies for optimizing the computing capabilities of neural tissue.

Dendritic Computing: Branching Deeper into Machine Learning

In this paper, we discuss the nonlinear computational power provided by dendrites in biological and artificial neurons. We start by briefly presenting biological evidence about the type of dendritic nonlinearities, respective plasticity rules and their effect on biological learning as assessed by computational models. Four major computational implications are identified as improved expressivity, more efficient use of resources, utilizing internal learning signals, and enabling continual learning. We then discuss examples of how dendritic computations have been used to solve real-world classification problems with performance reported on well known data sets used in machine learning. The works are categorized according to the three primary methods of plasticity used-structural plasticity, weight plasticity, or plasticity of synaptic delays. Finally, we show the recent trend of confluence between concepts of deep learning and dendritic computations and highlight some future research directions.

Unique properties of dually innervated dendritic spines in pyramidal neurons of the somatosensory cortex uncovered by 3D correlative light and electron microscopy

Pyramidal neurons (PNs) are covered by thousands of dendritic spines receiving excitatory synaptic inputs. The ultrastructure of dendritic spines shapes signal compartmentalization, but ultrastructural diversity is rarely taken into account in computational models of synaptic integration. Here, we developed a 3D correlative light-electron microscopy (3D-CLEM) approach allowing the analysis of specific populations of synapses in genetically defined neuronal types in intact brain circuits. We used it to reconstruct segments of basal dendrites of layer 2/3 PNs of adult mouse somatosensory cortex and quantify spine ultrastructural diversity. We found that 10% of spines were dually innervated and 38% of inhibitory synapses localized to spines. Using our morphometric data to constrain a model of synaptic signal compartmentalization, we assessed the impact of spinous versus dendritic shaft inhibition. Our results indicate that spinous inhibition is locally more efficient than shaft inhibition and that it can decouple voltage and calcium signaling, potentially impacting synaptic plasticity.

Active dendrites enable strong but sparse inputs to determine orientation selectivity

The dendrites of neocortical pyramidal neurons are excitable. However, it is unknown how synaptic inputs engage nonlinear dendritic mechanisms during sensory processing in vivo, and how they in turn influence action potential output. Here, we provide a quantitative account of the relationship between synaptic inputs, nonlinear dendritic events, and action potential output. We developed a detailed pyramidal neuron model constrained by in vivo dendritic recordings. We drive this model with realistic input patterns constrained by sensory responses measured in vivo and connectivity measured in vitro. We show mechanistically that under realistic conditions, dendritic Na+ and NMDA spikes are the major determinants of neuronal output in vivo. We demonstrate that these dendritic spikes can be triggered by a surprisingly small number of strong synaptic inputs, in some cases even by single synapses. We predict that dendritic excitability allows the 1% strongest synaptic inputs of a neuron to control the tuning of its output. Active dendrites therefore allow smaller subcircuits consisting of only a few strongly connected neurons to achieve selectivity for specific sensory features.

Drawing inspiration from biological dendrites to empower artificial neural networks

This article highlights specific features of biological neurons and their dendritic trees, whose adoption may help advance artificial neural networks used in various machine learning applications. Advancements could take the form of increased computational capabilities and/or reduced power consumption. Proposed features include dendritic anatomy, dendritic nonlinearities, and compartmentalized plasticity rules, all of which shape learning and information processing in biological networks. We discuss the computational benefits provided by these features in biological neurons and suggest ways to adopt them in artificial neurons in order to exploit the respective benefits in machine learning.

The gradient clusteron: A model neuron that learns to solve classification tasks via dendritic nonlinearities, structural plasticity, and gradient descent

Synaptic clustering on neuronal dendrites has been hypothesized to play an important role in implementing pattern recognition. Neighboring synapses on a dendritic branch can interact in a synergistic, cooperative manner via nonlinear voltage-dependent mechanisms, such as NMDA receptors. Inspired by the NMDA receptor, the single-branch clusteron learning algorithm takes advantage of location-dependent multiplicative nonlinearities to solve classification tasks by randomly shuffling the locations of "under-performing" synapses on a model dendrite during learning ("structural plasticity"), eventually resulting in synapses with correlated activity being placed next to each other on the dendrite. We propose an alternative model, the gradient clusteron, or G-clusteron, which uses an analytically-derived gradient descent rule where synapses are "attracted to" or "repelled from" each other in an input- and location-dependent manner. We demonstrate the classification ability of this algorithm by testing it on the MNIST handwritten digit dataset and show that, when using a softmax activation function, the accuracy of the G-clusteron on the all-versus-all MNIST task (~85%) approaches that of logistic regression (~93%). In addition to the location update rule, we also derive a learning rule for the synaptic weights of the G-clusteron ("functional plasticity") and show that a G-clusteron that utilizes the weight update rule can achieve ~89% accuracy on the MNIST task. We also show that a G-clusteron with both the weight and location update rules can learn to solve the XOR problem from arbitrary initial conditions.

Local glutamate-mediated dendritic plateau potentials change the state of the cortical pyramidal neuron

Dendritic spikes in thin dendritic branches (basal and oblique dendrites) are traditionally inferred from spikelets measured in the cell body. Here, we used laser-spot voltage-sensitive dye imaging in cortical pyramidal neurons (rat brain slices) to investigate the voltage waveforms of dendritic potentials occurring in response to spatially restricted glutamatergic inputs. Local dendritic potentials lasted 200-500 ms and propagated to the cell body, where they caused sustained 10- to 20-mV depolarizations. Plateau potentials propagating from dendrite to soma and action potentials propagating from soma to dendrite created complex voltage waveforms in the middle of the thin basal dendrite, comprised of local sodium spikelets, local plateau potentials, and backpropagating action potentials, superimposed on each other. Our model replicated these voltage waveforms across a gradient of glutamatergic stimulation intensities. The model then predicted that somatic input resistance (Rin) and membrane time constant (tau) may be reduced during dendritic plateau potential. We then tested these model predictions in real neurons and found that the model correctly predicted the direction of Rin and tau change but not the magnitude. In summary, dendritic plateau potentials occurring in basal and oblique branches put pyramidal neurons into an activated neuronal state ("prepared state"), characterized by depolarized membrane potential and smaller but faster membrane responses. The prepared state provides a time window of 200-500 ms, during which cortical neurons are particularly excitable and capable of following afferent inputs. At the network level, this predicts that sets of cells with simultaneous plateaus would provide cellular substrate for the formation of functional neuronal ensembles.NEW & NOTEWORTHY In cortical pyramidal neurons, we recorded glutamate-mediated dendritic plateau potentials with voltage imaging and created a computer model that recreated experimental measures from dendrite and cell body. Our model made new predictions, which were then tested in experiments. Plateau potentials profoundly change neuronal state: a plateau potential triggered in one basal dendrite depolarizes the soma and shortens membrane time constant, making the cell more susceptible to firing triggered by other afferent inputs.

Contextual Modulation in Mammalian Neocortex is Asymmetric

Neural systems are composed of many local processors that generate an output given their many inputs as specified by a transfer function. This paper studies a transfer function that is fundamentally asymmetric and builds on multi-site intracellular recordings indicating that some neocortical pyramidal cells can function as context-sensitive two-point processors in which some inputs modulate the strength with which they transmit information about other inputs. Learning and processing at the level of the local processor can then be guided by the context of activity in the system as a whole without corrupting the message that the local processor transmits. We use a recent advance in the foundations of information theory to compare the properties of this modulatory transfer function with that of the simple arithmetic operators. This advance enables the information transmitted by processors with two distinct inputs to be decomposed into those components unique to each input, that shared between the two inputs, and that which depends on both though it is in neither, i.e., synergy. We show that contextual modulation is fundamentally asymmetric, contrasts with all four simple arithmetic operators, can take various forms, and can occur together with the anatomical asymmetry that defines pyramidal neurons in mammalian neocortex.

Illuminating dendritic function with computational models

Dendrites have always fascinated researchers: from the artistic drawings by Ramon y Cajal to the beautiful recordings of today, neuroscientists have been striving to unravel the mysteries of these structures. Theoretical work in the 1960s predicted important dendritic effects on neuronal processing, establishing computational modelling as a powerful technique for their investigation. Since then, modelling of dendrites has been instrumental in driving neuroscience research in a targeted manner, providing experimentally testable predictions that range from the subcellular level to the systems level, and their relevance extends to fields beyond neuroscience, such as machine learning and artificial intelligence. Validation of modelling predictions often requires - and drives - new technological advances, thus closing the loop with theory-driven experimentation that moves the field forward. This Review features the most important, to our understanding, contributions of modelling of dendritic computations, including those pending experimental verification, and highlights studies of successful interactions between the modelling and experimental neuroscience communities.

A Computational Theory for the Emergence of Grammatical Categories in Cortical Dynamics

A general agreement in psycholinguistics claims that syntax and meaning are unified precisely and very quickly during online sentence processing. Although several theories have advanced arguments regarding the neurocomputational bases of this phenomenon, we argue that these theories could potentially benefit by including neurophysiological data concerning cortical dynamics constraints in brain tissue. In addition, some theories promote the integration of complex optimization methods in neural tissue. In this paper we attempt to fill these gaps introducing a computational model inspired in the dynamics of cortical tissue. In our modeling approach, proximal afferent dendrites produce stochastic cellular activations, while distal dendritic branches–on the other hand–contribute independently to somatic depolarization by means of dendritic spikes, and finally, prediction failures produce massive firing events preventing formation of sparse distributed representations. The model presented in this paper combines semantic and coarse-grained syntactic constraints for each word in a sentence context until grammatically related word function discrimination emerges spontaneously by the sole correlation of lexical information from different sources without applying complex optimization methods. By means of support vector machine techniques, we show that the sparse activation features returned by our approach are well suited—bootstrapping from the features returned by Word Embedding mechanisms—to accomplish grammatical function classification of individual words in a sentence. In this way we develop a biologically guided computational explanation for linguistically relevant unification processes in cortex which connects psycholinguistics to neurobiological accounts of language. We also claim that the computational hypotheses established in this research could foster future work on biologically-inspired learning algorithms for natural language processing applications.

Representation of visual landmarks in retrosplenial cortex

The process by which visual information is incorporated into the brain’s spatial framework to represent landmarks is poorly understood. Studies in humans and rodents suggest that retrosplenial cortex (RSC) plays a key role in these computations. We developed an RSC-dependent behavioral task in which head-fixed mice learned the spatial relationship between visual landmark cues and hidden reward locations. Two-photon imaging revealed that these cues served as dominant reference points for most task-active neurons and anchored the spatial code in RSC. This encoding was more robust after task acquisition. Decoupling the virtual environment from mouse behavior degraded spatial representations and provided evidence that supralinear integration of visual and motor inputs contributes to landmark encoding. V1 axons recorded in RSC were less modulated by task engagement but showed surprisingly similar spatial tuning. Our data indicate that landmark representations in RSC are the result of local integration of visual, motor, and spatial information.

When moving through a city, people often use notable or familiar landmarks to help them navigate. Landmarks provide us with information about where we are and where we need to go next. But despite the ease with which we – and most other animals – use landmarks to find our way around, it remains unclear exactly how the brain makes this possible.

One area that seems to have a key role is the retrosplenial cortex, which is located deep within the back of the brain in humans. This area becomes more active when animals use visual landmarks to navigate. It is also one of the first brain regions to be affected in Alzheimer's disease, which may help to explain why patients with this condition can become lost and disoriented, even in places they have been many times before.

To find out how the retrosplenial cortex supports navigation, Fischer et al. measured its activity in mice exploring a virtual reality world. The mice ran through simulated corridors in which visual landmarks indicated where hidden rewards could be found. The activity of most neurons in the retrosplenial cortex was most strongly influenced by the mouse’s position relative to the landmark; for example, some neurons were always active 10 centimeters after the landmark.

In other experiments, when the landmarks were present but no longer indicated the location of a reward, the same neurons were much less active. Fischer et al. also measured the activity of the neurons when the mice were running with nothing shown on the virtual reality, and when they saw a landmark but did not run. Notably, the activity seen when the mice were using the landmarks to find rewards was greater than the sum of that recorded when the mice were just running or just seeing the landmark without a reward, making the “landmark response” an example of so-called supralinear processing.

Fischer et al. showed that visual centers of the brain send information about landmarks to retrosplenial cortex. But only the latter adjusts its activity depending on whether the mouse is using that landmark to navigate. These findings provide the first evidence for a “landmark code” at the level of neurons and lay the foundations for studying impaired navigation in patients with Alzheimer's disease. By showing that retrosplenial cortex neurons combine different types of input in a supralinear fashion, the results also point to general principles for how neurons in the brain perform complex calculations.

Enhanced Dendritic Compartmentalization in Human Cortical Neurons

The biophysical features of neurons shape information processing in the brain. Cortical neurons are larger in humans than in other species, but it is unclear how their size affects synaptic integration. Here, we perform direct electrical recordings from human dendrites and report enhanced electrical compartmentalization in layer 5 pyramidal neurons. Compared to rat dendrites, distal human dendrites provide limited excitation to the soma, even in the presence of dendritic spikes. Human somas also exhibit less bursting due to reduced recruitment of dendritic electrogenesis. Finally, we find that decreased ion channel densities result in higher input resistance and underlie the lower coupling of human dendrites. We conclude that the increased length of human neurons alters their input-output properties, which will impact cortical computation. VIDEO ABSTRACT.

How Dendrites Affect Online Recognition Memory

In order to record the stream of autobiographical information that defines our unique personal history, our brains must form durable memories from single brief exposures to the patterned stimuli that impinge on them continuously throughout life. However, little is known about the computational strategies or neural mechanisms that underlie the brain's ability to perform this type of "online" learning. Based on increasing evidence that dendrites act as both signaling and learning units in the brain, we developed an analytical model that relates online recognition memory capacity to roughly a dozen dendritic, network, pattern, and task-related parameters. We used the model to determine what dendrite size maximizes storage capacity under varying assumptions about pattern density and noise level. We show that over a several-fold range of both of these parameters, and over multiple orders-of-magnitude of memory size, capacity is maximized when dendrites contain a few hundred synapses-roughly the natural number found in memory-related areas of the brain. Thus, in comparison to entire neurons, dendrites increase storage capacity by providing a larger number of better-sized learning units. Our model provides the first normative theory that explains how dendrites increase the brain's capacity for online learning; predicts which combinations of parameter settings we should expect to find in the brain under normal operating conditions; leads to novel interpretations of an array of existing experimental results; and provides a tool for understanding which changes associated with neurological disorders, aging, or stress are most likely to produce memory deficits-knowledge that could eventually help in the design of improved clinical treatments for memory loss.

Dendritic Neuron Model With Effective Learning Algorithms for Classification, Approximation, and Prediction

An artificial neural network (ANN) that mimics the information processing mechanisms and procedures of neurons in human brains has achieved a great success in many fields, e.g., classification, prediction, and control. However, traditional ANNs suffer from many problems, such as the hard understanding problem, the slow and difficult training problems, and the difficulty to scale them up. These problems motivate us to develop a new dendritic neuron model (DNM) by considering the nonlinearity of synapses, not only for a better understanding of a biological neuronal system, but also for providing a more useful method for solving practical problems. To achieve its better performance for solving problems, six learning algorithms including biogeography-based optimization, particle swarm optimization, genetic algorithm, ant colony optimization, evolutionary strategy, and population-based incremental learning are for the first time used to train it. The best combination of its user-defined parameters has been systemically investigated by using the Taguchi's experimental design method. The experiments on 14 different problems involving classification, approximation, and prediction are conducted by using a multilayer perceptron and the proposed DNM. The results suggest that the proposed learning algorithms are effective and promising for training DNM and thus make DNM more powerful in solving classification, approximation, and prediction problems.

Scanless volumetric imaging by selective access multifocal multiphoton microscopy

Simultaneous, high-resolution imaging across a large number of synaptic and dendritic sites is critical for understanding how neurons receive and integrate signals. Yet, functional imaging that targets a large number of submicrometer-sized synaptic and dendritic locations poses significant technical challenges. We demonstrate a new parallelized approach to address such questions, increasing the signal-to-noise ratio by an order of magnitude compared to previous approaches. This selective access multifocal multiphoton microscopy uses a spatial light modulator to generate multifocal excitation in three dimensions (3D) and a Gaussian-Laguerre phase plate to simultaneously detect fluorescence from these spots throughout the volume. We test the performance of this system by simultaneously recording Ca2+ dynamics from cultured neurons at 98-118 locations distributed throughout a 3D volume. This is the first demonstration of 3D imaging in a "single shot" and permits synchronized monitoring of signal propagation across multiple different dendrites.

Human Cortical Pyramidal Neurons: From Spines to Spikes via Models

We present detailed models of pyramidal cells from human neocortex, including models on their excitatory synapses, dendritic spines, dendritic NMDA- and somatic/axonal Na⁺ spikes that provided new insights into signal processing and computational capabilities of these principal cells. Six human layer 2 and layer 3 pyramidal cells (HL2/L3 PCs) were modeled, integrating detailed anatomical and physiological data from both fresh and postmortem tissues from human temporal cortex. The models predicted particularly large AMPA- and NMDA-conductances per synaptic contact (0.88 and 1.31 nS, respectively) and a steep dependence of the NMDA-conductance on voltage. These estimates were based on intracellular recordings from synaptically-connected HL2/L3 pairs, combined with extra-cellular current injections and use of synaptic blockers, and the assumption of five contacts per synaptic connection. A large dataset of high-resolution reconstructed HL2/L3 dendritic spines provided estimates for the EPSPs at the spine head (12.7 ± 4.6 mV), spine base (9.7 ± 5.0 mV), and soma (0.3 ± 0.1 mV), and for the spine neck resistance (50–80 MΩ). Matching the shape and firing pattern of experimental somatic Na⁺-spikes provided estimates for the density of the somatic/axonal excitable membrane ion channels, predicting that 134 ± 28 simultaneously activated HL2/L3-HL2/L3 synapses are required for generating (with 50% probability) a somatic Na⁺ spike. Dendritic NMDA spikes were triggered in the model when 20 ± 10 excitatory spinous synapses were simultaneously activated on individual dendritic branches. The particularly large number of basal dendrites in HL2/L3 PCs and the distinctive cable elongation of their terminals imply that ~25 NMDA-spikes could be generated independently and simultaneously in these cells, as compared to ~14 in L2/3 PCs from the rat somatosensory cortex. These multi-sites non-linear signals, together with the large (~30,000) excitatory synapses/cell, equip human L2/L3 PCs with enhanced computational capabilities. Our study provides the most comprehensive model of any human neuron to-date demonstrating the biophysical and computational distinctiveness of human cortical neurons.

A Neurodynamic Model of Feature-Based Spatial Selection

Huang and Pashler (2007) suggested that feature-based attention creates a special form of spatial representation, which is termed a Boolean map. It partitions the visual scene into two distinct and complementary regions: selected and not selected. Here, we developed a model of a recurrent competitive network that is capable of state-dependent computation. It selects multiple winning locations based on a joint top-down cue. We augmented a model of the WTA circuit that is based on linear-threshold units with two computational elements: dendritic non-linearity that acts on the excitatory units and activity-dependent modulation of synaptic transmission between excitatory and inhibitory units. Computer simulations showed that the proposed model could create a Boolean map in response to a featured cue and elaborate it using the logical operations of intersection and union. In addition, it was shown that in the absence of top-down guidance, the model is sensitive to bottom-up cues such as saliency and abrupt visual onset.

Embedded ensemble encoding hypothesis: The role of the "Prepared" cell

We here reconsider current theories of neural ensembles in the context of recent discoveries about neuronal dendritic physiology. The key physiological observation is that the dendritic plateau potential produces sustained depolarization of the cell body (amplitude 10-20 mV, duration 200-500 ms). Our central hypothesis is that synaptically-evoked dendritic plateau potentials lead to a prepared state of a neuron that favors spike generation. The plateau both depolarizes the cell toward spike threshold, and provides faster response to inputs through a shortened membrane time constant. As a result, the speed of synaptic-to-action potential (AP) transfer is faster during the plateau phase. Our hypothesis relates the changes from "resting" to "depolarized" neuronal state to changes in ensemble dynamics and in network information flow. The plateau provides the Prepared state (sustained depolarization of the cell body) with a time window of 200-500 ms. During this time, a neuron can tune into ongoing network activity and synchronize spiking with other neurons to provide a coordinated Active state (robust firing of somatic APs), which would permit "binding" of signals through coordination of neural activity across a population. The transient Active ensemble of neurons is embedded in the longer-lasting Prepared ensemble of neurons. We hypothesize that "embedded ensemble encoding" may be an important organizing principle in networks of neurons.

Modelling plasticity in dendrites: from single cells to networks

One of the key questions in neuroscience is how our brain self-organises to efficiently process information. To answer this question, we need to understand the underlying mechanisms of plasticity and their role in shaping synaptic connectivity. Theoretical neuroscience typically investigates plasticity on the level of neural networks. Neural network models often consist of point neurons, completely neglecting neuronal morphology for reasons of simplicity. However, during the past decades it became increasingly clear that inputs are locally processed in the dendrites before they reach the cell body. Dendritic properties enable local interactions between synapses and location-dependent modulations of inputs, rendering the position of synapses on dendrites highly important. These insights changed our view of neurons, such that we now think of them as small networks of nearly independent subunits instead of a simple point. Here, we propose that understanding how the brain processes information strongly requires that we consider the following properties: which plasticity mechanisms are present in the dendrites and how do they enable the self-organisation of synapses across the dendritic tree for efficient information processing? Ultimately, dendritic plasticity mechanisms can be studied in networks of neurons with dendrites, possibly uncovering unknown mechanisms that shape the connectivity in our brains.

Synaptic plasticity in dendrites: complications and coping strategies

The elaborate morphology, nonlinear membrane mechanisms and spatiotemporally varying synaptic activation patterns of dendrites complicate the expression, compartmentalization and modulation of synaptic plasticity. To grapple with this complexity, we start with the observation that neurons in different brain areas face markedly different learning problems, and dendrites of different neuron types contribute to the cell's input-output function in markedly different ways. By committing to specific assumptions regarding a neuron's learning problem and its input-output function, specific inferences can be drawn regarding the synaptic plasticity mechanisms and outcomes that we 'ought' to expect for that neuron. Exploiting this assumption-driven approach can help both in interpreting existing experimental data and designing future experiments aimed at understanding the brain's myriad learning processes.

An Online Unsupervised Structural Plasticity Algorithm for Spiking Neural Networks

In this paper, we propose a novel winner-take-all (WTA) architecture employing neurons with nonlinear dendrites and an online unsupervised structural plasticity rule for training it. Furthermore, to aid hardware implementations, our network employs only binary synapses. The proposed learning rule is inspired by spike-timing-dependent plasticity but differs for each dendrite based on its activation level. It trains the WTA network through formation and elimination of connections between inputs and synapses. To demonstrate the performance of the proposed network and learning rule, we employ it to solve two-class, four-class, and six-class classification of random Poisson spike time inputs. The results indicate that by proper tuning of the inhibitory time constant of the WTA, a tradeoff between specificity and sensitivity of the network can be achieved. We use the inhibitory time constant to set the number of subpatterns per pattern we want to detect. We show that while the percentages of successful trials are 92%, 88%, and 82% for two-class, four-class, and six-class classification when no pattern subdivisions are made, it increases to 100% when each pattern is subdivided into 5 or 10 subpatterns. However, the former scenario of no pattern subdivision is more jitter resilient than the later ones.

The Effects of Realistic Synaptic Distribution and 3D Geometry on Signal Integration and Extracellular Field Generation of Hippocampal Pyramidal Cells and Inhibitory Neurons

In vivo and in vitro multichannel field and somatic intracellular recordings are frequently used to study mechanisms of network pattern generation. When interpreting these data, neurons are often implicitly considered as electrotonically compact cylinders with a homogeneous distribution of excitatory and inhibitory inputs. However, the actual distributions of dendritic length, diameter, and the densities of excitatory and inhibitory input are non-uniform and cell type-specific. We first review quantitative data on the dendritic structure and synaptic input and output distribution of pyramidal cells (PCs) and interneurons in the hippocampal CA1 area. Second, using multicompartmental passive models of four different types of neurons, we quantitatively explore the effect of differences in dendritic structure and synaptic distribution on the errors and biases of voltage clamp measurements of inhibitory and excitatory postsynaptic currents. Finally, using the 3-dimensional distribution of dendrites and synaptic inputs we calculate how different inhibitory and excitatory inputs contribute to the generation of local field potential in the hippocampus. We analyze these effects at different realistic background activity levels as synaptic bombardment influences neuronal conductance and thus the propagation of signals in the dendritic tree. We conclude that, since dendrites are electrotonically long and entangled in 3D, somatic intracellular and field potential recordings miss the majority of dendritic events in some cell types, and thus overemphasize the importance of perisomatic inhibitory inputs and belittle the importance of complex dendritic processing. Modeling results also suggest that PCs and inhibitory neurons probably use different input integration strategies. In PCs, second- and higher-order thin dendrites are relatively well-isolated from each other, which may support branch-specific local processing as suggested by studies of active dendritic integration. In the electrotonically compact parvalbumin- and cholecystokinincontaining interneurons, synaptic events are visible in the whole dendritic arbor, and thus the entire dendritic tree may form a single integrative element. Calretinin-containing interneurons were found to be electrotonically extended, which suggests the possibility of complex dendritic processing in this cell type. Our results also highlight the need for the integration of methods that allow the measurement of dendritic processes into studies of synaptic interactions and dynamics in neural networks.

Resonant Dynamics of Grounded Cognition: Explanation of Behavioral and Neuroimaging Data Using the ART Neural Network

Research on grounded cognition suggests that the processing of a word or concept reactivates the perceptual representations that are associated with the referent object. The objective of this work is to demonstrate how behavioral and functional neuroimaging data on grounded cognition can be understood as different manifestations of the same cortical circuit designed to achieve stable category learning, as proposed by the adaptive resonance theory (ART). We showed that the ART neural network provides a mechanistic explanation of why reaction times in behavioral studies depend on the expectation or attentional priming created by the word meaning (Richter and Zwaan, 2009). A mismatch between top-down expectation and bottom-up sensory data activates an orienting subsystem that slows execution of the current task. Furthermore, we simulated the data from functional neuroimaging studies of color knowledge retrieval that showed anterior shift (Chao and Martin, 1999; Thompson-Schill, 2003) and an overlap effect (Simmons et al., 2007; Hsu et al., 2011) in the left fusiform gyrus. We explain the anterior effect as a result of the partial activation of different components of the same ART circuit in the condition of passive viewing. Conversely, a demanding perceptual task requires activation of the whole ART circuit. This condition is reflected in the fMRI image as an overlap between cortical activation during perceptual and conceptual processing. We conclude that the ART neural network is able to explain how the brain grounds symbols in perception via perceptual simulation.

Structured Dendritic Inhibition Supports Branch-Selective Integration in CA1 Pyramidal Cells

Neuronal circuit function is governed by precise patterns of connectivity between specialized groups of neurons. The diversity of GABAergic interneurons is a hallmark of cortical circuits, yet little is known about their targeting to individual postsynaptic dendrites. We examined synaptic connectivity between molecularly defined inhibitory interneurons and CA1 pyramidal cell dendrites using correlative light-electron microscopy and large-volume array tomography. We show that interneurons can be highly selective in their connectivity to specific dendritic branch types and, furthermore, exhibit precisely targeted connectivity to the origin or end of individual branches. Computational simulations indicate that the observed subcellular targeting enables control over the nonlinear integration of synaptic input or the initiation and backpropagation of action potentials in a branch-selective manner. Our results demonstrate that connectivity between interneurons and pyramidal cell dendrites is more precise and spatially segregated than previously appreciated, which may be a critical determinant of how inhibition shapes dendritic computation.

Cognitive functions of intracellular mechanisms for contextual amplification

Evidence for the hypothesis that input to the apical tufts of neocortical pyramidal cells plays a central role in cognition by amplifying their responses to feedforward input is reviewed. Apical tufts are electrically remote from the soma, and their inputs come from diverse sources including direct feedback from higher cortical regions, indirect feedback via the thalamus, and long-range lateral connections both within and between cortical regions. This suggests that input to tuft dendrites may amplify the cell's response to basal inputs that they receive via layer 4 and which have synapses closer to the soma. ERP data supporting this inference is noted. Intracellular studies of apical amplification (AA) and of disamplification by inhibitory interneurons targeted only at tufts are reviewed. Cognitive processes that have been related to them by computational, electrophysiological, and psychopathological studies are then outlined. These processes include: figure-ground segregation and Gestalt grouping; contextual disambiguation in perception and sentence comprehension; priming; winner-take-all competition; attention and working memory; setting the level of consciousness; cognitive control; and learning. It is argued that theories in cognitive neuroscience should not assume that all neurons function as integrate-and-fire point processors, but should use the capabilities of cells with distinct sites of integration for driving and modulatory inputs. Potentially 'unifying' theories that depend upon these capabilities are reviewed. It is concluded that evolution of the primitives of AA and disamplification in neocortex may have extended cognitive capabilities beyond those built from the long-established primitives of excitation, inhibition, and disinhibition.

A simple transfer function for nonlinear dendritic integration

Relatively recent advances in patch clamp recordings and iontophoresis have enabled unprecedented study of neuronal post-synaptic integration ("dendritic integration"). Findings support a separate layer of integration in the dendritic branches before potentials reach the cell's soma. While integration between branches obeys previous linear assumptions, proximal inputs within a branch produce threshold nonlinearity, which some authors have likened to the sigmoid function. Here we show the implausibility of a sigmoidal relation and present a more realistic transfer function in both an elegant artificial form and a biophysically derived form that further considers input locations along the dendritic arbor. As the distance between input locations determines their ability to produce nonlinear interactions, models incorporating dendritic topology are essential to understanding the computational power afforded by these early stages of integration. We use the biophysical transfer function to emulate empirical data using biophysical parameters and describe the conditions under which the artificial and biophysically derived forms are equivalent.

Synaptic clustering within dendrites: an emerging theory of memory formation

It is generally accepted that complex memories are stored in distributed representations throughout the brain, however the mechanisms underlying these representations are not understood. Here, we review recent findings regarding the subcellular mechanisms implicated in memory formation, which provide evidence for a dendrite-centered theory of memory. Plasticity-related phenomena which affect synaptic properties, such as synaptic tagging and capture, synaptic clustering, branch strength potentiation and spinogenesis provide the foundation for a model of memory storage that relies heavily on processes operating at the dendrite level. The emerging picture suggests that clusters of functionally related synapses may serve as key computational and memory storage units in the brain. We discuss both experimental evidence and theoretical models that support this hypothesis and explore its advantages for neuronal function.

On the functions, mechanisms, and malfunctions of intracortical contextual modulation

A broad neuron-centric conception of contextual modulation is reviewed and re-assessed in the light of recent neurobiological studies of amplification, suppression, and synchronization. Behavioural and computational studies of perceptual and higher cognitive functions that depend on these processes are outlined, and evidence that those functions and their neuronal mechanisms are impaired in schizophrenia is summarized. Finally, we compare and assess the long-term biological functions of contextual modulation at the level of computational theory as formalized by the theories of coherent infomax and free energy reduction. We conclude that those theories, together with the many empirical findings reviewed, show how contextual modulation at the neuronal level enables the cortex to flexibly adapt the use of its knowledge to current circumstances by amplifying and grouping relevant activities and by suppressing irrelevant activities.

Model reduction of strong-weak neurons

We consider neurons with large dendritic trees that are weakly excitable in the sense that back propagating action potentials are severly attenuated as they travel from the small, strongly excitable, spike initiation zone. In previous work we have shown that the computational size of weakly excitable cell models may be reduced by two or more orders of magnitude, and that the size of strongly excitable models may be reduced by at least one order of magnitude, without sacrificing the spatio–temporal nature of its inputs (in the sense we reproduce the cell's precise mapping of inputs to outputs). We combine the best of these two strategies via a predictor-corrector decomposition scheme and achieve a drastically reduced highly accurate model of a caricature of the neuron responsible for collision detection in the locust.