Neural circuits

This page is being created.

Introduction

Figure 1. Golgi rendition of the cerebellar Purkinje cell by Santiago Ramon y Cajal, illustrating his theory of 'dynamic polarization'. (Public domain image accessed on August 30, 2023)

Since ancient times people have sought to understand how the brain works! As early as the early 1900s the prevailing view was that the nervous system was a web of interconnected nerve cells. Indeed, through decades of careful research and magnificent discoveries it is apparent that the nervous system is an ensemble of myriad, intricate connections made between neurons. The complex circuits these connections form lay down the basic architecture for brain function, crucially engineering how intelligence and consciousness emerge from the inner workings of the brain. However, during those early days there was absent an understanding of mechanisms underlying cell function and no notion of the direction of cytoplasmic flow. This narrative extended to the brain, where the nervous system and its interconnected cells were thought to fire randomly, lacking direction! At the turn of the twentieth century, the Spanish physician and Nobel laureate, Santiago Ramon y Cajal, considered the father of modern neuroscience, was armed with his exquisite drawings of neurons and proceeded to provide the pivotal directional decree for signaling in neurons! Crucially, Cajal advocated that the neuron was a "polarized" functional entity, with branches of dendrites and the cell body that received information on one end and axon cylinders that conducted nerve impulses and terminals that transmitted information at the other end. This polarization enabled signals to flow from one end of the neuron to the other, representing a functional processing unit—the neuron doctrine of Cajal juxtaposed against Camilo Golgi’s reticular theory of neurons. Hence, the argument in favor of the existence of circuits to route input and output of neuronal impulses has existed since Cajal and his theory of "dynamic polarization" (Figure 1). This view was augmented by the work of the British neurophysiologist and Nobel Prize winner, Sir Charles Sherrington, who coined the term "synapse" to describe the specialized junction through which neurons communicate. His work on the receptive field properties of the skin for the scratch reflex further posited neurons as the elementary and fundamental functional units of neural circuits. Cajal’s drawings and description of neurons and circuits as a means of signal propagation in the visual, olfactory, auditory, somatosensory, cerebellum, and the spinal cord served as the basis for his insistence of similar circuitry in higher neocortical regions. This neuron-centric view was justified, as the anatomical and physiological experimental approaches that existed at the time largely limited the investigation to the level of single neurons. Subsequently, sophisticated advancements in technology and research enabled recordings from multiple neurons and highlighted neuronal ensembles as functional physiological units, unveiling distinct emergent properties not realized by individual neurons.

Figure 2. High-resolution image of a neuron and its neurites generated by Scanning electron microscope (SEM). (Public domain image accessed on August 30, 2023)

Cajal formulated his "neuron doctrine" by using the Golgi method to stain individual neurons, and consequently his doctrine was based on static images that established the neuron as the basic organizational and signaling unit of the nervous system. Cajal regarded the neuron as an individual entity that established contacts with other neurons without being continuous with them, and that neurons served as the basic building block of neural circuits (owing to their dynamic polarization, vis a vis inputs and output) Yet it would take till the electron microscope in the 1950s and electrophysiological single-cell recordings in the 1950s and 60s to provide conclusive evidence of the growing significance of the neuron as a functional unit of the nervous system (Figure 2).

In the 18^th and early 19^th centuries, there was some understanding of localization of function with regard to the organization of the brain and more broadly the nervous system. Function was allocated to certain sensory input areas and the output was generated from the motor cortex, but there was no significant knowledge of information processing between the inputs and outputs. Later, in the mid-nineteenth century, Hebb DO, laid out his neurophysiological concepts on the formation of neural circuits in his influential book The organization of behavior: A Neuropsychological Theory (1949) His postulates had a profound impact and through the ensuing decades have continued to guide and extend our understanding of neural networks, learning, memory, and synaptic plasticity. Hebbian constructs beautifully encapsulate the creation, maintenance and plasticity of ‘cell assemblies’ in local neural networks. In this regard, 3 postulates that Hebb propounded underlie the neurophysiological changes that occur with learning and memory. His primary and core postulate and learning rule was the strengthening of synaptic connections between cells when one cell persistently drives another —cells that fire together wire together’! In essence, this implies that the cell had ‘learned’ a particular correlated pattern of activity, translating into changes in synaptic strength. Secondly, this coordinated activity organizes these neurons into reverberating, synaptically-coupled ‘cell assemblies’, a trace, that is shaped by and is subject to change by experience. Hebb’s third construct postulated that these neuron ‘cell assemblies’, a circuit, is essentially maintained by the synchronous firing and wiring between a series of synaptically-coupled clusters that converge and activate other cortical regions to generate a ‘phase sequence’’, i.e., the neurophysiological correlate underlying learning and memory that can be linked to thought and mind, a purview that was under intense consideration by cognitive psychologists.

The opposite coin of the Hebbian learning rule maintained that connections between unused synapses would weaken and be eliminated implying a process for active depression. This proposal extended the Hebbian learning rule and provided a mechanism whereby convergent and coactive pre-synaptic inputs to a post-synaptic cell would strengthen and uncoordinated inputs would weaken. Later studies dramatically demonstrated this point for ocular representation in the cortex. Studies on monkeys and cats showed that monocular deprivation resulted in thalamic projections from the intact eye occupying a greater extent of layer 4 (L4, the cortical input layer) in the visual cortex at the expense of representation from the deprived eye. Furthermore, excessive synaptic pruning has been associated with negative consequences in neurological disorders, such as Alzheimer's disease.

Hebbian learning rule based on synchronous activity as a coincidence detector has been discovered mechanistically in long-term potentiation (LTP) in the hippocampus. The axons of the excitatory neurons of the entorhinal cortex, via the perforant pathway, synapse with the granule cells of the hippocampal dentate gyrus (an area crucial for conscious memory formation), and exhibit activity-dependent synaptic potentiation as as result of coordinated activity. LTP causes an immediate and sustained increase in synaptic efficiency that is dependent on the glutamatergic N-methyl-D-aspartate (NMDA) receptor as a coincidence detector of pre- and post-synaptic activity. It is experimentally induced by delivering trains of high-frequency stimulation to the monosynaptic excitatory pathways in the hippocampus, and compellingly manifests Hebbian plasticity.

Markram H, et al. would go on to demonstrate how the temporal order and time window of pre- and post-synaptic activity could impact the strength of connections between two cells and hence add a more dynamic element to synaptic plasticity and the Hebbian learning rule. These studies laid the groundwork for the plasticity mechanism, appropriately named spike timing-dependent plasticity (STDP), expounding the strengthening of synaptic connections with correlated activity between cells.

Neural circuits are formed by integrating many local networks that are specialized for a particular function into a larger coherent framework. The functionally specialized and anatomically segregated networks are dynamically organized into a distributed and hierarchical systems of neurons, and the spatiotemporal firing patterns of neurons in local networks is synchronized for efficient information processing within and between cortical areas in the hierarchy. The integrity of computations within the local networks in the hierarchy and the efficiency of communication between different cortical circuits is critical for information processing in the brain. Computations in these modular systems is manifested as distinct patterns of spiking activity that continues downward through the hierarchy. In this regard, distinct cortical areas are specialized to subserve different cognitive functions and conduct definitive neural computations specific for those cognitive functions. This necessitates specific firing patterns of presynaptic circuits elicit a distinct, detectible, and reliable response in the post-synaptic networks for the effective routing of information between these functional networks.

It is important to note that the computations that occur in neural networks are an emergent property that derives from the collective activity of neurons in the circuit and is not necessarily resident in the individual cells. Arguably, the cells can participate in different functional networks, displaying combinatorial flexibility (proposed by Hebb) as a consequence of plasticity. Indeed, this plasticity grants neuronal assemblies the ability to engage dynamically with smaller and larger “local-global” networks, and hence change the pattern of activity in the population of neurons. This could mean that the repeat presentation of the stimulus may not elicit the same network operations on subsequent occasions, suggesting a fluidity to the workings of the neuronal assemblies. This could ostensibly be an emergent property of the different networks and an intrinsic mechanism that could enrich and represent how the brain encodes and tells time, by indexing the unique timing of events and experiences, keeping them spatiotemporally separate and discrete.

Why is it important to study neural circuits? Firstly, they are critical for cortical and subcortical information processing. Neural circuit dysfunction provides clues into the etiology of brain diseases. Furthermore, since psychiatric disorders are considered to be caused by underlying neural circuit dysfunction of cortical and subcortical areas in humans, an understanding of defects in the hierarchy could shed light on the pathogenesis of these disorders.

Current technologies have impactfully changed the playing field for the study of neural circuits, and cellular recordings have moved from single cells to registering the activity of populations of neurons forming the Hebbian cell assemblies. In Hebb’s view it was the interconnected network of neurons that facilitated the encoding of all the essential aspects of an object, event, scenario—not as it were, the individual neurons! In accordance with this view, it is evident from past and current studies that individual neurons are not just ‘feature detectors’ but demonstrate a mixed selectivity for features within a perceptual, behavioral, or cognitive domain, and tune their responses according to the changing conditions. Thus, the neural circuit made up of these cell assemblies, weaves together and amalgamates all the facets and the sum of the experience. For example, the ventral stream of neural processing for vision culminates in neurons in the monkey inferotemporal (IT) cortex and its human cognate the lateral occipital complex (LOC), that process highly specific and complex features of visual stimuli, whereby they extract and represent two-dimensional shape. However, they also respond to multiple elements of the scene, and images are clustered according to some category structure not observed at earlier stages of visual processing, implying that the specific, high-fidelity identification of objects occurs at the population level higher up the hierarchy. These neuronal assemblies operate similarly in humans and non-human primates in the way they categorize objects. Similarly, in the hippocampus, the place cells maintain a neural representation of the environment, a cognitive map, and adapt their responses to the changing physical space, essentially ‘remapping’, while retaining certain elements of the prior environment and  encoding features of the new space. These and other studies highlight how multiple domains of information are stored and represented in different regions and how context and change continuously reorganizes the coding of neurons and their assemblies.  

Advanced and Emerging Technologies to Study Neural Circuits

Figure 3. Stained Pyramidal neurons exemplifying the complexity of neural circuits. The figure showcases the intricacy of axons and dendritic arbors. (Public domain image accessed on August 30, 2023)

Research with an emerging tableau of technologies ostensibly seeks to clarify the complicated organization of neural circuits embedded in different cortical areas and further decode their operations (Figure 3). Tissue volumes ideally containing all elements of a fully functional neuronal circuit is approximately on the order of millimeter (mm)³ in mammals. The synaptic contacts on the other hand would require a spatial granularity down to the ~10 nanometer (nm)³ resolution. The approaches used to define the structural and functional features of a unit of tissue need to be robust and reliable and should be complemented with sophisticated computational modeling. Technological advancements have enabled recordings from neurons with multi-scale resolution from the circuit down to finer, individual synaptic connections. Large-scale recording techniques in freely moving animals monitor the activity of a great number of neurons in different networks across the brain in behaving animals. Studies have drawn on these techniques to interrogate circuit activity and the interactions between networks in different regions associated with natural and specific behaviors. Complementary non-invasive, genetic based approaches enable optical stimulation of distinct neuronal populations, coding activity at the spatiotemporal level of single spikes and synapses. Optogenetics techniques use light to drive neural activity in neurons that are genetically engineered to express certain peculiar ion channels, naturally found in certain algae. These procedures precisely exploit the distinctive and idiosyncratic optical properties of these channels called Channelrhodopsins, that respond rapidly to light. Thus, optogenetic techniques enable the study of circuit maps and allows for the  probing of distinct neurons within the confines of a network of diverse cell types, while also manipulating and probing the concurrent activity of multiple neurons.

In animal models, 2-photon imaging is used to examine a functionally defined, unit volume of brain tissue, with a combination of techniques to structurally resolve the relevant subcellular features—proceeding from image capture with techniques such as, synchrotron X-ray computed tomography with propagation-based phase contrast (SXRT) to targeted acquisition of multiple volumes of tissue sections using serial block-face electron microscopy (SBEM). Since no single technique can furnish the structural, physiological, and functional features of a circuit, these correlative multimodal imaging (CMI) and other correlative methods can provide us with better insight into a region of interest, with broader implications for gaining an understanding of disease processes.

Concurrently, high-resolution functional MRI (fMRI) enables distinguishing individual circuit features of cortical lamina in humans, allowing researchers to test existing hypothesis and generate testable predictions of value for different aspects of brain function. CMI techniques allow for the integration of functional, temporal, structural, and molecular information gathered from the same tissue, using different technologies, thereby, enriching our understanding of and making the underlying mechanisms of neural circuits more tractable.

Organization of the Cortex

Laminar Organization

Figure 4. Cytoarchitectonic 'areas' of human brain, delineated by Brodmann (1909). (Public domain image accessed on August 30, 2023)

Ever since the investigations in the early 1900’s, neuroanatomists and neurophysiologists have appreciated the laminar or horizontally layered structure of the cerebral cortex. It was by and large recognized that the superficial layers were to a great extent receptive to incoming inputs, and that the deeper layers constituted chiefly the efferent, outgoing component of the cortex. At the turn of the 19^th century the German physician and neurologist, and arguably the founder of brain cartography, Korbinian Brodmann undertook a pioneering and indisputably to say the least, a monumental and excellent effort to anatomically map and elucidate the cytoarchitectonics of the mammalian cortex, including charting the surface of the entire human cerebral cortex (Figure 4). Brodmann diagrammed and elegantly parcellated the cortex into 52 ‘areas’ that shared cellular and laminar structure and set forth the structural basis for the ‘localization’ of function in the brain. He broadly delineated the 6 layered cortical architecture, with the number of layers featuring differently across different cortical regions, depending on their functional correlate. In the following years, with the advent of retrograde tracers, the projection pathways and laminar organization of the cortex was further refined and validated, and it became evident that dominant inputs to cortical lamina are derived locally from excitatory and inhibitory neurons, standing in contrast to the minor contribution from afferents originating from other cortical and sub-cortical areas. Brodmann’s laudatory map with its trademark structural detail has been highly regarded throughout the last century. Even today, it is widely used by neuroscientists and neurologists alike as a tool to mark individual areas of the brain and link them to function, for instance, area 17 denotes the visual cortex, while area 4 distinguishes the motor cortex. Speak about the lasting influence and impact on the brain!

Figure 5. BigBrain Project. The. first microsccopic resolution 3-D model of the human brain. The model can be examined in the saggital, coronal, axial. and any oblique angle. (Image accessed on August 30, 2023) BigBrain is licensed under:Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License.

However, these manually constructed brain maps are limited and manifest the brain at the macroscopic two-dimensional (2-D) level, biased to regional cortical neuroanatomy. Recently, projects that rely on digitized techniques, like the openly accessible BigBrain (Figure 5), based on histological samples of over 7000 human brain sections (cell bodies stained, coronally sectioned blockface images, digitized, reconstructed, and MRI registered), offer a fully automated, non-invasive, ultrahigh-resolution three-dimensional (3-D) model of the human brain that enables the interrogation of cortical laminar organization. BigBrain offers cellular resolution at the microscopic level (20 micrometer, in three orthogonal planes—saggital, coronal, axial, and any oblique ), enhancing our ability to resolve cortical connectivity and spatial patterns, and empowering us to tease out novel structure-function relationships not previously observed. These kinds of projects are exceedingly valuable as tools to model normal brain function, and are even more valuable for characterizing brain pathologies.

Vertical Organization

Subsequently, other cortical features were deciphered. The overwhelmingly intricate and pain-staking work conducted in the mid-19^th century specified a crucial organizational feature of the cerebral cortex, delineating vertical (radial) columns and revealing another very important structural facet of the cortex. Owing to their vertical partitioning, the columns span all the layers (L) of the cortex (L2 to L6, with L1 at the top), from pia at the cortical surface to the white matter, with resident neurons sharing extrinsic connectivity that enables them to effectively serve as a functional unit (Figure 6). These cells cluster in a column in bespoke sensory cortices and the response of neurons lying within them is tuned to a particular attribute of the receptive field of different sense organs. The functionally relevant vertical columns with their stereotypical intrinsic structure and configuration, is repeated throughout the cortex, in sensory, motor, and association areas, and underlies the unique processing subserving different sensory modalities and higher cognitive functions. Cortical columns have thus been conceptualized as discrete anatomical and functional entities, that exhibit periodicity, and are endowed with intrinsic computational facility duplicated across different sensory and higher-order cognitive cortices in the mammalian brain.

Figure 6. Rodent Vibrissal Columns in the somatosensory cortex. This is a 3-D receonstruction of 5 cortical columns with the 6. layers clearly demarcated (color coded) from the pial surface to the white matter. (Public domain image accessed on August 31, 2023) This file is licensed under the Creative Commons Attribution-Share Alike 4.0 International license.

At the mesoscopic level, the Spanish neuroscientist Lorente de Nó, who was Cajal's diciple, set forth two rules of processing within cortical circuits, i.e., the plurality of connections and reciprocal, recurrent connections representing feedback. The synaptic coupling within vertical columns exemplified these processes and were essential for the formation of interconnected modules. The properties of plurality and recurrent connections endowed these circuits with the ability to dynamically reconfigure a neuronal circuit, and so representing the physical and biological basis for providing feedback in a network of connected cells. The vertical  setup with different and adaptive configurations of cortical processing could presumably lead to different behavioral outcomes. These insights into the structural elements of vertical ‘minicolumns’ essentially laid the groundwork for determining the functional dynamics of sensory columns and delineating the sensory profiles, exemplified by the somatosensory and visual cortices. Mountcastle’s work on the similarity of the receptive field properties found in cells within a column in the somatosensory cortex, and Hubel and Wiesel’s work on orientation and ocular dominance columns of the mammalian visual cortex widely established and epitomized vertical columns as a crucial structural feature of cortical processing (Figure 7). In 1981, Hubel and Wiesel were awarded the Nobel Prize in Physiology or Medicine for their seminal work on the columnar organization of and the processing within the visual cortex—"for their discoveries concerning information processing in the visual system". Roger Sperry shared the award "for his discoveries concerning the functional specialization of the cerebral hemispheres".

Figure 7. Illustration of ocular dominance columns in the visual cortex. The alternating columns selectively receive inputs related to the left or right eye via the thalamic lateral geniculate nucleus. (Public domain image accessed on August 31, 2023)

However, since the 1990s, it has become increasingly apparent that cortical columns are not useful as a concept anymore. A review of half a century of research concluded that columns have no function and have no utility in describing uniformity in cortical processing, and that the column has ‘failed as a unifying principle for understanding cortical function’. Strong criticism that echoed the ongoing debates since the 1990s, questioning the value of the columnar architecture, erroneously designated as containing 110 neurons in a volume of 30 micrometer (μm)-wide and 25 μm-deep cortical section. This led to the proposal that the concept that columns were intended for, i.e., to serve as building blocks of the neocortex, has not held up and therefore should be retired. This argument however is not completely debunked and does still appreciate the column as both a ‘guide and cautionary tale’ when studying the organization of the neocortex. Recently two-photon imaging studies deploying genetically engineered Ca²⁺ indicators (GECIs) have interrogated individual neurons in individual layers, L2 - L6, providing a 3-D map of functional columns. These imaging studies overcome the shortcomings of electrical recordings that are biased towards recording only from active neurons, sampling from a minute population of cells in a given area. In the primary auditory cortex, a recent study deploying two-photon Ca²⁺imaging showed response-specific (best-frequency) neurons were present in all layers, and generally clustered in distinct columns along the tonotopic axis, that also crossed over to other columns. This spatial distribution over few hundred micrometers results in heterogeneity at the microscale while maintaining stable response patterns at the larger, macroscale level. This is conjectured to be a reflection of the reliance of the auditory cortex on developmental and learning-dependent wiring rules.

Interest in modular processing units has also been focused on the microcolumns, that are smaller than minicolumns, i.e., the mini orientation or ocular dominance columns of the visual cortex. Cells in adjacent orientation minicolumns all respond to similar orientations, whereas cells in adjacent microcolumns respond to different orientations. Microcolumns with similar selectivity are presumed to lay close together in an orientation microcolumns and assume a hexagonal arrangement. Distinct excitatory and inhibitory cells have been shown to form a hexagonal lattice like structure in layer 5 of major cortical areas, including the somatosensory, visual, motor, and language areas. Microcolumns have their own modular synaptic connectivity and neuronal activity -  synchronized activity exhibiting similar temporal dynamics—acting as independent processing units. Cells within a microcolumn of sub-cerebral projection neurons connect only to each other and those of callosal projection neurons connect only to each other. Parallel processing of numerous and repeated microcolumns subserves multiple cortical functions, such as sensory perception, motor function, and language.

Altogether, it is important to note that traditional cortical columns have utility, it is also important to understand that they are not entirely stereotypical and are best distinguished by their uniqueness. Columns fundamentally differ in cellular composition, patterns of intrinsic and extrinsic connectivity - both ipsilateral, contralateral, and cortical and sub-cortical connectivity - and synaptic organization, proteins and gene expression, signaling molecules, spatiotemporal dynamics, and are modality specific. Thus, it goes without saying that many types of columns exist and they remain functionally relevant at a conceptual level.

Cortical Organization in Humans

Recently, a systematic review of peer-reviewed articles based on MRI imaging of human and non-human primates was completed in order to derive a data-driven model of cortical laminar connectivity. The effort aimed at addressing the morphological and laminar organization and cortical interconnectivity both locally and globally, across the cortex. The study endeavored to provide exactly the mesoscale resolution detailed by Lorento de Nó and others at the turn of the 19^th century. Essentially, resolution at the mesoscale level offers a way to reconcile and bridge the gap between microscale imaging of the cortical lamina and macroscale imaging of axonal connections - connectomics - and advances an exciting way to model whole-brain connectivity.

Figure 8. A representative high-resolution MRI of the human occipital lobe shown in coronal (top) and axial (bottom) view. Algorithms parcellate the cortex into similar laminar structures. The arrowheads in the left hemisphere and the green clusters in the right hemisphere, respectively, depict the Stria of Gennari, calcarine fissure (CF), and demarcate the primary visual cortex (V1). The. same clusters in axial view (bottom) are shown on the medial surface (left) and occipital pole (left). (Public domain image accessed on August 31, 2023) This file is licensed under the Creative Commons Attribution 4.0 International license.

The review found that the cortical laminar connectivity was most robustly related to the granularity index, i.e., the level of granularity in cortical layers and cortical regions, estimated by the proportions of granule cells. This was best laid out in the granularity-based atlas of von Economo-Koskinas that divided the cortex based on cytoarchitectonics at the levels of granule cells into low-order agranular cortex and higher-order granular cortex. The granular cortices are representative of sensory areas, epitomized by the distinctive layer 4 granule cells of the visual cortex that receive sensory inputs from the thalamus. According to the review of MRI neuroimaging studies, the 6 layered cortex can also be roughly organized into 3 groups, the supragranular (L1-L3; lamina L), granular (L4) , and infragranular (L5-L6) layers. This grouping is both relevant to the different anatomical structures the lamina connect to and is furthermore significantly associated with the functioning of the individual layers. The data-driven model also proposed probabilities of interconnectedness within the vertical microcolumn for each of the groupings that are mostly in line with the connections formed in the ‘canonical microcircuit’ (see below). Briefly, the granular L4 receives inputs from the thalamus, and in turn projects to more superficial supragranular L2/3 that then form connections within the supragranular layer and with deeper L5 that in turn drives L6, the latter then projects back to the thalamus.

The modeling based on data from MRI (Figure 8), and other studies of both the global white matter axonal connections (tractography) and grey matter laminar organization reconciles the two MRI neuroimaging approaches – diffusion weighted imaging for tractography and T1 imaging for cortical lamina analysis, respectively—and puts them in the framework of whole-brain laminar connectivity. This manner of modeling is a great addition to the neuroimaging arsenal and bolsters the study of connectivity across the entire brain, including for the study of connectomes and brain networks.

The Canonical Neural Circuit

The 6-layered cortex is a salient feature of cortical architecture across species, and the laminar organization defines cortical circuits and markedly increases computational efficiency. The layout features the essential afferent, intrinsic, and efferent projection neurons that have a discrete, biased, and non-random arrangement in the different cortical layers. It has long been observed that majority of the connections are made horizontally, parallel to the lamina, and this is emblematic of a representative block of a six-layered cortex with a surface area of a square millimeter that fundamentally represents a functional and elementary ‘processing unit’. It is arguably this fundamental ‘processing unit’ that is evolutionarily conserved across species, and the highly ordered basic motif is repeated many times over, lattice-like, contributing vastly to the increase in cortical surface area. Decades of neuroanatomical and neurophysiological studies have elucidated the features of cortical circuits and have found a similar pattern of excitatory and inhibitory neurons, and their organization and morphologies playing out in different cortical regions. This motif represents a ‘canonical circuit’ that incorporates these elements (Figure 9). These ubiquitous biological circuits have been overwhelmingly described for the unimodal granular sensory cortices, especially the columns of the primary visual striate cortex and somatosensory cortex, while similar networks have been encountered in other cortical regions. In these circuits, signal is transmitted through feedforward pathways with several nested positive and negative feedback loops at various levels that serve to modulate and amplify the signal. The serial feedforward processing engenders the orientation-selective receptive fields of simple cells in layer 4 (L4) in the primary visual cortex, V1, and the generation of more complex features of the receptive fields of cells in the superficial layers 2/3 (L2/3). This difference in the neural coding in different lamina have been reported in the primary somatosensory cortex (S1) as well.

Figure 9. Illustration of the canonical local circuit in the somatosensory (barrel) cortex, depicting the 6 layers. On the left is a potrayal of the morphology and on the right is a cartoon of the cortical processing in the different layers. Thickness of arrows indicate strength of  connection. (Public domain image accessed on August 31, 2023) This file is licensed under the Creative Commons Attribution-Share Alike 3.0 Unported license.

Simply put, the canonical circuit is comprised of thalamic or subcortical inputs primarily to L4 granule cells (granular layer in sensory cortices) that then send afferents to the superficial pyramidal neurons of the L2/3 that receive and process the incoming information, subserving ‘receptive and associational’ functions. L2/3 pyramidal cells then project to the deeper layer 5 (L5) pyramidal neurons that represent the main output units of the cortical circuit (except feedback connection to the thalamus), comprising ‘corticofugal and commissural’ projections. L2/3 neurons further provide feedforward inputs to L4 of ‘higher’ cortical areas. L5 neurons also provide inputs to layer 6 (L6) that then completes the cortical loop by sending projections to L4-L2, and along with other layers (except L4) sends feedback to the areas that sent inputs to the circuit. L6 afferents to the thalamus, constitute a significant proportion of corticothalamic feedback, and for example in the visual system, serve to change the receptive field properties of the neurons in the thalamic nucleus and importantly gate the information flow back to the cortex. Layer 1 is populated with inhibitory interneurons and the dendrites of excitatory neurons from lower layers and is tasked with integrating information from other ‘higher’ cortical areas and the thalamus. This pattern of connectivity is repeated in different regions and adapt neocortical regions to processing distinct formats of information. There are area- and species differences to this motif. This circuit is heralded as ‘canonical’ as it is a stereotypical motif, but the cortex is anything but stereotypical and there are other circuit motifs deployed in different areas.

As referred to above, along with anatomists’ appreciation of the laminar organization of the cortex, they acknowledged from early the prevalence of cortical columns running vertically or radially in the cerebral cortex, orthogonal to the horizontal lamina. Though its presence has a functional unit has come into question (see above), the vertical column as a functional correlate of the morphologically characterized column was beautifully demonstrated and established by Mountcastle in the somatosensory cortex. Mountcastle showed that neurons arranged in a narrow vertical column, or cylinder, spanning L2 to L6 in the cat primary somatosensory cortex responded to the same tactile stimuli with the same peripheral receptive fields. Thalamic inputs to L4 were further processed by cells in other layers lying at superficial and deep levels within the column with brief latencies. Thus, the modular organization of cells in vertical columns, represents an array of iterative neurons extending from layers 2 through layer 6, and constituted the elementary processing unit in the cerebral cortex, including in primates and humans.

Similary, influential work done by Hubel & Wiesel in the 1950's, 60's and 70's on the mammalian visual system, illustrated the receptive field properties of neurons in cortical columns. For example, the esteemed pair arrived at certain conclusions based on their observation of electrophysiological recordings from the macaque monkey primary visual cortex (V1, striate cortex, area 17). They deduced that inputs received from the multilayered lateral geniculate nucleus - the thalamic relay nucleus that itself receive separate inputs from both eyes - are 'rearranged' in V1, such that cells respond to specific orientations of line segments. In an elegant series of studies, they demonstrated that the visual cortex featured orientation columns, with cells that preferred certain orientations aligning vertically, with shifts in preference for orientations advancing in an orderly manner, clockwise or counter-clockwise, along the cortex. Furthermore, inputs from both eyes tended to 'converge' onto single cells, requisite for binocular vision. This processing occurs in multiple stages with cells with shared physiological properties organized in vertical cortical columns. Hubel and Wiesel advanced the functional role of columnar processing, by showing ocular dominance columns, with alternating groups of vertically clustered cells responding preferentially to either the left or right eye. This advanced the notion of the mammalian cortex being segregated vertically and horizontally. Cells lying within a particular vertical columns shared similar properties, such as preference for orientation or eye dominance, directionality, etc. Cells organized into different horizontal layers were arranged hierarchically, and processed information differently. Simple cells in L4, early in the hierarchy, were monocularly driven, where ascomplex cells in L2 and L3, and further up the hierarchy in L5 and L6 were binocularly driven.

Transmodal Association Cortices

Although the canonical circuit is optimized for the sensory cortices, a modified dynamic circuitry has been proposed for higher association cortices. In contrast to the sensorimotor cortices, the association cortices are to a great extent multi- or trans-modal and functionally flexible, drawing from available sensorimotor inputs and reliance on representations from memory. The association cortex comprises the prefrontal, cingulate, inferior parietal, precuneal, and middle temporal areas in humans. They occupy large areas of the cerebral cortex and receive and recapitulate information, integrating inputs from the primary and secondary sensory and motor cortices, thalamus and brainstem, and implement complex cognitive computations for the generation of behavior, and attention and memory. The association cortices are spatially distant from the primary cortices and their functional attributes of increasing abstraction are a consequence of being unconstrained from the confines of network circuitry of the functionally specialized cortices. Imaging studies, including resting state functional MRI (rsfMRI) have demonstrated the existence of hierarchical progression of inputs form the local, typical circuits of the sensory and motor areas devised for perception and action to the larger and more distributed transmodal areas that serve to integrate converging unimodal signals for abstract cognition. Studies using rsfMRI reveal a principal topographic gradient in the connectome running from the sensorimotor cortices - visual, somatosensory, auditory, and motor cortices - at one extreme of the hierarchy to the transmodal association cortices, that include the default-mode network (DMN) in humans at the other extreme. The hierarchical organization thus represents a structural feature that dovetails with function and indicates a course for the DMN to play a role in cognition, acting as a hub for integrating and abstracting transmodal information without being reliant on immediate sensory input. The topographic connectivity pattern of the principal gradient from sensorimotor to transmodal cortices provides a spatial framework for processing within multiple large-scale networks, and serves as another emergent motif of neural circuitry. Morphometric fMRI and other analysis on humans, strongly indicate that psychiatric disorder such as schizoplrenia, bipolar, and other cognitive disorders are linked to structural differences in the transmodal association cortices, DMN, disrupting their functional connectivity to sensorimotor and other cortices.

The network hierarchy answers the question structurally, how signals are propagated and more functionally relevant, why the networks are organized the way they are. Segregation of processing along the hierarchy is essential for adaptive cognition going from the unimodal sensorimotor cortices when responding to immediate environmental demands and switching to transmodal cortical areas when behavior needs to be guided by internal representations. A recent human fMRI study showed that in a visual decision-making task that either depended on real-time perception or reliance on memory, how the unimodal sensory cortices–medial and lateral visual cortices–were recruited when processing immediate sensory inputs while transmodal, DMN regions were activated for memory reliant decisions. The more marked the shift in neural processing in the hierarchy from unimodal to transmodal areas when relying on representations from memory the more efficiently the task was performed. This macroscale hierarchical organization of neural processing facilitates efficient generation of behavior, anchored either on perception and action at one end, and memory and internal representation at the other and is crucial for adaptive cognition.

The transmodal association cortices also feed back information to the unimodal cortices for purposes of memory recall and for recapturing some of the nested associations regarding the memory and its target. This has been elegantly demonstrated in the temporal lobe, and other cortices. A model referred to as the cortical “dynamic multimode module” or D3M, posits that the association cortex encompasses and augments the ‘canonical circuits’ and their operations and furthermore it conforms to the particular cognitive demands, such as sensory cue processing or memory recall. The D3M model includes the feedforward pathways of canonical circuits and incorporates cortico-cortical projections important for integrating inputs between the association cortex and different cortical and sub-cortical regions related to information processing and cognition. For example, for processing sensory cues this model incorporates the feedfoward connections of the classic canonical circuit, namely L4 receives feedforward projections from the sensory cortices and projects to the supragranular L2/3 that then send afferents to infragranular L5. In contrast, for memory recall or feedback mode, studies conducted on primate association area, (perirhinal cortex) the inputs are processed between the infragranular L5 and L6 with some local information processing to L2/3, following which the retrieved information is sent back from Layer 6 through ‘backpropogation’ to the earlier cortical and other cortical structures for additional processing and retrieval of nested associations. Therefore, the D3M model, is proposed as a canonical network for association cortices that execute high level computations essential for complex cognitive tasks such as memory and attention that rely heavily on both the traditional canonical feedforward circuitry and equally on the back propogating feedback circuitry across cortical areas.

Tremendous amount of work done using tract-tracing with markers and electrophysiological recordings have extended our understanding of the laminar locations of neurons, their interareal connections, cortical hierarchy, and how they integrate with local circuits. These and other studies on cortical and subcortical areas strongly suggest that rostrally directed pathways mediate feedforward connections, and conversely, caudally directed ones submit feedback. 

Patterns of Synaptic Connections

The last quarter of the twentieth century witnessed a heightened interest in systematically describing features of the mammalian cortex. It is widely recognized that network activity is essential for information processing, and that the specific patterns of synaptic connections provide the physical infrastructure for communication between neurons and their activity. It has also been revealed that information received either from external, peripheral sources or from other neurons in a circuit is synaptically weighted and the weights are summed to inform the outputs and the resulting circuit motifs formed.  

The different connectivity patterns between neurons enlisted in neuronal networks form the physiological basis of and provide the general principles for information processing in the nervous system. In general, similar network motifs are used in combination in all areas of the brain and offer a useful level of abstraction for describing information processing in local and interareal circuits. Numerous studies have investigated network motifs using multiple techniques, including neuroanatomical studies, tract tracing and electrophysiological recordings in different species, enabling an appreciation of cortical hierarchy, to the extent that interareal connections integrate with local circuits. Human data from high resolution fMRI measuring blood oxygenation level-dependent (BOLD) signals in different cortical layers during cognitive, sensory and motor actions is accumulating and these kinds of techniques enable us to study the activity of different brain regions non-invasively.

It is abundantly evident that in order to gain a rich understanding of cortical function it is incumbent upon us to correctly and definitively interpret and model circuit motifs and processing of inputs -both feedforward and feedback synaptic inputs—received by different neuronal compartments and interneurons spanning multiple layers. This includes inputs and outputs with regard to the particular location of the cell bodies and dendrites, both basal and apical, and axonal connection, either local or distal, over the thickness of the cortex. Below is a summary of these circuit motifs of communication between neurons.

Feedforward Networks

Feedforward Excitation

The simplest form of information flow in the nervous system recruits excitatory neurons at every single relay in a linear, feedforward network, with no feedback connections to neurons early in the relay. By engaging excitatory neurons in this manner, this kind of network topology enables propagation of information over great distances in the nervous system. Feedforward networks integrate information at successive levels and include convergent and divergent patterns of synaptic coupling that selectively decipher the information flowing at each stage and implement computations for the output disseminated to the next level. Convergent excitation of a neuron by many pre-synaptic neurons enables it to extract features that may not be specifically or explicitly expressed by the individual inputs, and this enhances the signal-to-noise ratio of the message at the level of the recipient target neuron. Divergent excitation, meanwhile, feeds forward the message from one neuron to a multitude of post-synaptic neurons in the network for efficient information processing.

Very good examples of these circuit motifs are prevalent in sensory systems that recruit multiple parallel channels to transmit information, maximizing the amount, accuracy and speed of the conveyed message. Sensory perception, be it visual, auditory, or olfactory, begins with many sensory receptors, located peripherally—at sites where they encounter the stimuli. Divergence is commonly seen in sensory systems, such as in the mammalian retina, where photoreceptors send signals to multiple bipolar cells. Non-linear processing in the visual system also includes convergence of feed-forward input throughout the retina. These operations are essential to compress and selectively represent large amounts of inbound information at the level of the photoreceptors (~100 million photoreceptor to ~1 million ganglion cells, 100:1 ratio) and accommodate it within the confines of the optic nerve. The axons of ganglion cells, the retinal output neurons, band together in the optic nerve which presents a bottleneck, and carry the signals form the periphery, centrally to higher order areas. Model systems encompassing the neural circuitry of the retina in silico have further advanced our understanding of the computations required for ensuring the fidelity of information flow from photoreceptors to the ganglion cells.

Other studies have similarly established patterns of convergence in other sensory systems and species. For instance the olfactory receptor projections in the olfactory glomeruli of the fruit fly Drosophila, enhance signals from single neurons by averaging inputs for increased accuracy and odor detection. The subsequent reconvergence of this information onto higher order neurons in the pathway is optimized for speed. These motifs are recurring themes in the nervous systems of various species, where compression and consolidation of information flowing in parallel, enhances the signal and reduces the noise, and efficiently and seemingly effortlessly guides behavior. It follows that deploying these different circuit motifs enables greater, faster and more efficacious transmission of information over a window of time.

Convergence also occurs at the macro level, whereby information proceeds from purely sensory cortices successively to sensory-specific association cortices higher up in the hierarchy that ultimately intersects with other sensory inputs in multimodal association cortices that integrate all the inputs (see above). This convergent pattern is reciprocated with complementary divergence from these multisensory cortices back to the earlier sensory cortices. For example, an early study showed impressive evidence of sensory convergence and divergence in monkey entorhinal cortex that receives afferents from multimodal regions of the frontal and temporal lobes that themselves receive inputs from classical sensory areas.

Feedforward Inhibition

Feedforward information processing also relies heavily on inhibition, which is ubiquitous in the nervous system. Inhibitory interneurons play a fundamental role in sculpting the activity of neural circuits, presenting a counterpoint to the long- or short-range excitatory inputs. Cortical activity is massively reliant on this form of inhibition for maintaining a balance between excitation and inhibition. The networks engage interneurons to generate feedforward inhibition, where the presynaptic excitatory neurons simultaneously target post-synaptic excitatory neurons and inhibitory interneurons. Once activated the interneurons inhibit the post-synaptic excitatory neurons, cutting of their activity. This form of inhibition is extremely effective in curtailing the spikes of post-synaptic neurons as it operates mostly at the dendrites, the site of excitatory inputs. Feedforward inhibition is fast, disynaptic, and is proportional to the strength of the input, and in return it functions to enhance downstream signals and precision of outputs. It has for example been shown to enhance the precision and coding of temporal frequencies in auditory circuits. Feedforward inhibition has also been encountered in other sensory systems and species, such a the locust olfactory system for odor discrimination.

The mammalian cortex has a crucial role in transforming pertinent sensory inputs into performance of purposefully driven motor acts. The anterior cingulate cortex (ACC) located in the frontal lobe, directs appropriate behavior and is involved in guiding goal-directed decision making by iteratively monitoring actions to update outcomes. It receives inputs from sensory cortices and enhances perceptual operations via top-down modulation of sensory areas, like the visual cortex. To interrogate the cortical circuitry involved in converting visual inputs into motor acts, a recent study provided a visual stimulus (flashing lights) to reward a lick response in rats. Certain premotor neurons in the ACC suppress motor acts, such as licking in rodents. Bottom-up inputs from the visual cortex feed into sensory neurons in the ACC that use feedforward inhibition to suppress the sustained activity of the motor neurons and unleash the lick response. Thus, the internal activity of neurons in the ACC gates the inputs from the visual cortex to inhibit the motor neurons and elicit the motor (lick) response.

Interneurons acting at other sites, such as the soma (eg., basket cells) and axon initial segment (eg., chandelier cells) also participate in feedforward inhibition. The cortical and hippocampal fast-spiking (FS) parvalbumin (PV)-expressing GABAergic interneurons (basket cells) engage in feedforward inhibition to maintain the excitatory/inhibitory balance. This elegant inhibitory motif also plays out between different brain areas, where the interneurons inhibit post-synaptic excitatory neurons outside of the realm of operation of the primary excitatory neurons. The robustness of excitatory input on post-synaptic neurons and interneurons and the inhibition mediated by the latter ultimately determines the shape of the discharge. This mechanism is observed in CA1 hippocampal circuits where oxytocin exerts its effects through feedforward inhibition to sharpen the signal-to-noise ratio, enhancing the fidelity of spike transmission and information processing. In cortico-hippocampal circuits, cholecystokinin (CCK)-expressing interneurons provide strong feedforward inhibition to the CA1 pyramidal neurons and have an enduring influence on synaptic plasticity, a crucial mechanism for information processing and storage. Suppression of CCK-expressing interneurons by the coordinated activity of the coincident afferents to the CA1 pyramidal neurons illustrate the impact of this form of inhibition on information flow. Specifically, inputs to the CA1 pyramidal neurons arrive from the entorhinal cortex via two paths, a heterosynaptic route, via the dentate gyrus and CA3 Schaffer collaterals, and direct perforant pathway afferents. The temporally precise activity in these pathways suppresses the inhibitory action of CCK-expressing interneurons and shapes the activity in the local circuit. The study demonstrates the power of feedforward inhibition and its inhibition by the dual pathways and its dynamic impact on gating information flow in the local network, establishing a temporally precise synaptic learning rule for associational plasticity and processing and storage of information.

Feedback Inhibition

Feedback inhibition occurs when a population of excitatory neurons drive a group of inhibitory neurons that themselves inhibit the same set of input neurons, mostly at their presynaptic terminals. Feedback or recurrent inhibition, as it is also known, recasts the temporal aspects of the excitatory neurons output by forestalling or restricting the discharge in a manner proportional to the strength of the output. The reciprocal excitatory-inhibitory motif is encountered in numerous cortical circuits. For example, the feature appears in the feedback circuit in the cerebellum, between Purkinje cells and molecular layer interneurons that are crucial for cerebellar learning. Reciprocal feedback excitatory-inhibitory connections have also been evidenced in the prefrontal cortex, an area critical for working memory. Electrophysiological recording of the reciprocal connection between L2/3 pyramidal cells and inhibitory fast-spiking (FS) interneurons in the prefrontal cortex showed the connections to be highly efficient and more reliable, with low failure rates of neurotransmission as compared to unidirectional connections. These connections greatly contribute to the unique processing required from these regions underlying key cognitive functions.

As previously mentioned, there is a vast repository of continuing studies on cortical and subcortical areas, using a variety of techniques, such as tract-tracing with markers and electrophysiological recordings that have elucidated the hierarchical arrangement of neurons in cortical layers, and the integration of their connections with local and other, regionally diverse cortical circuits. It is evidenced that feedforward operations are subserved by rostrally directed pathways, and feedback is routed via caudally directed ones. The hierarchically organized visual cortex serves as an example with feedforward connections transmitting sensory information from lower to higher areas, whilst feedback connections communicate downstream effects. Feedforward and successive hierarchical connections of the visual system are activated immediately after encountering an image. This constitutes the 'feedforward sweep' that extracts features ‘hardwired’ in the brain, such as orientation, direction of motion, color, and their co-occurrence (base-grouping). The feedforward sweep rapidly spreads to higher areas, and is purported to be pre-attentive, i.e., lacking conscious awareness of the image. Recurrent connections are required for attending to the features of the image and the subsequent conscious appreciation of them. A recent imaging study (fMRI) in humans supports the role of fast feedforward perceptual pathways and highlights the parsimonious allocation of neural resources for efficient object recognition. Recurrent connections are recruited and adapt to the complexity of the scene, with nominal feedback deployed for simple scenes and the feedback intensifying as the scene gets more complex and cluttered. This demonstrates the parsimonious and dynamic deployment of neural resources adapted to the complexity of the visual setting. The mixed set of feedback and feedforward activities subserving vision have been found to be encoded by 4 distinct functional streams, two each for feedforward and feedback connections, with defined laminar projections, marked frequency and temporal dynamics, and hierarchical organization. This dual counterstream architecture lends support to the predominance of higher, gamma frequency dominated feedforward connections emanating from supragranular cortical layers, and lower frequency band dominated feedback connections arising from infragranular layers. These separate modes generally operate sequentially at different time scales, with the faster feedforward pathways ascending the hierarchy earlier than the slower latency descending feedback streams. Feedforward and feedback connections are well suited to render different aspects of vision, such as contrast, receptive field, location, and distinct spatial features.  

GABAergic inhibitory interneurons have a high-powered role in circuit activity and represent a smaller percentage of cells in neural networks, ~20-30% compared to the larger cohort of glutamatergic principal neurons, ~70-80%. Nevertheless, the heterogenous populations of interneurons are intimately involved in network operations and participate acutely in sculpting the activity and complex response properties of circuits, such as expansion or contraction of the dynamic activity of neurons lying in circuits and playing a key role in generation of neuronal oscillations. For instance, the GABAergic, fast-spiking(FS) parvalbumin (PV)-expressing interneurons, have singular properties in feedforward and feedback inhibition. This is owed to the rapidity with which they act, transforming, within a millisecond, excitatory input to inhibitory output! Recent technological advances have made it feasible to tease out precisely the involvement of these and other phenotypically different local circuit interneurons in orchestrating network function and to further investigate their critical role in the etiology of brain disease.

Lateral Inhibition

Lateral inhibition is a related circuit motif that is widely encountered in the nervous system. Neurons in parallel pathways, independently activate interneurons that then selectively inhibit the information flow through those separate channels, enhancing the difference in activity between the two. Principle neurons utilize this mechanism to inhibit or reduce the activity of neurons that they are connected to or of neighboring principal cells, but unlike recurrent inhibition they do not cause inhibition of their own activity. Iterations of lateral inhibition involving principal cells that are connected to a common interneuron that inhibits the surrounding principal cells are greatly dependent on the strength and timing of the input of the principal neurons. This mode of inhibition basically amplifies the differences between the parallel pathways, and is powerfully deployed in sensory systems, such as in the retina and the olfactory bulb. The winner-take-all configuration of neural computing systems is modelled after lateral inhibition.

Regional differences in how this mechanism is deployed have been noted, and lateral inhibition contributes significantly to the unidirectional inhibition in the dentate gyrus, which is ~10 times more prevalent in this structure than reciprocal inhibitory connections, as compared to areas such as the neocortex where recurrent inhibition predominates. In hippocampal memory systems, the dentate gyrus performs pattern separation, whereby it transforms multiple overlapping inputs about similar experiences into non-overlapping discrete events. Patch-clamp recordings from the dentate gyrus elucidate a mechanism for indexing these inputs, revealing that the granule cells restrict the activity of nearby granule cells via GABAergic parvalbumin (PV)-expressing inhibitory interneurons. The winner-take-all motif in the dentate gyrus, lends itself to processing of incoming information and establishes a circuit for achieving pattern separation. Essentially the ‘winner cells’ actively suppress the activity of a population of ‘non-winner’ cells, and in the process leave their own activity unaltered. As mentioned, this activity is supplemented by the interaction of the principal cells with inhibitory interneurons to accomplish pattern separation. The execution of this operation decorrelates incoming inputs for distinguishing between similar experiences. Thus, the dentate gyrus plays a critical role, processing inputs essential for downstream memory storage in the CA1 pyramidal cells. This lends credence to the view that the activity in different brain regions is shaped by different patterns of local connections between excitatory principal cells and inhibitory interneurons.

Lateral inhibition is a highly effective and favored mechanism for resolving minutely contrasting inputs and is widely prevalent in neural circuits of different brain regions. In sensory systems the brain processes and encodes different aspects of information in separate, parallel streams. The sensory networks avail themselves of lateral inhibition to enhance the contrast in parallel pathways. For example, in the olfactory system it is effectively deployed to discriminate between similar odors based on intensity. In this system, sensory neurons in the olfactory bulb activate two specific projection neurons, the mitral and tufted cells in the olfactory bulb that process information in parallel streams. Lateral inhibition serves to segregate the information in these distinct, parallel paths based on the concentration of the odor. Mitral cells are best at distinguishing between similar odors at high concentrations, whereas tufted cells are better at lower, near-threshold odors. The olfactory and other sensory systems process information in parallel pathways, and tactically deploy lateral inhibition to extract distinct features of stimuli whose intensities may vary by many orders of magnitude.

Similarly, the visual system employs numerous strategies for processing information about the outside world, segregating inputs attributed to motion and color. Parallel processing of information is elegantly showcased in this system, right from the processing at the level of the retina to higher-order processing by cortical structures. In macaque retinas, the input of 4 million cones and millions of intervening neurons is compacted into about 1.6 million retinal ganglion cells (RGCs; 2.5- cones:1 RGC) whose axons form the centrally directed optic nerve. The RGCs project to a commensurate number of neurons in the six-layered lateral geniculate nucleus (LGN) in the thalamus, the intermediate relay structure to the visual cortex. Similar cone to ganglion cell ratios have been reported for humans, based on eccentricity from the fovea where there is parity (1 cones:1 RGC). RGCs are sensitive to different levels of contrast, i.e., high or low, and they send parallel projections to magnocellular and parvocellular layers of the LGN. Specifically, the the 4 dorsal parvocellular layers of the LGN receive inputs from RGCs (midget RGCS) conveying red-green color opponency and low contrast information, whereas the two ventral magnocellular layers receive mostly achromatic, high luminance contrast information from RGCs (parasol RGCs). The visual system uses lateral inhibition to code different aspects of the visual scene in an efficient manner, and impacts how the cell activity differs from its surround, and how one perceives contrast and achieves spatial discrimination. The classic center-surround receptive field (RF) properties of RGC are generated by lateral inhibition incurred at earlier stages in the pathway. Specifically in the retina, the feedforward connections of the presynaptic photoreceptors and bipolar cells, excite horizontal and amacrine interneurons, respectively. The interneurons, essentially amalgamate the signals from the presynaptic neurons and send inhibitory signals back to them and their neighbors. The feedforward neurons effectively produce an excitatory center in the RFs of RGCs, while the lateral inhibition by the interneurons generates an inhibitory surround. The center-sorround organizations of RGC receptive field is essential for perceptual processes, for example, like resolving contrast in a spatial setting, and discerning the perceived brightness of an object based on the luminance of the background. (Figure)

A combination network motif i.e., mutual inhibition of lateral inhibition, serves as an elementary computational building block and contributes significantly to simple decision making. For instance, in zebrafish, a simple behavioral choice like the direction of escape from an incoming stimulus is guided by this circuit motif. Neurons (Mauthner cells; M-cells) in the hindbrain, - an evolutionary conserved region, crucial for appropriate response to threatening or abrupt events - that descend and activate the contralateral spinal motor neurons, receive ipsilateral excitatory inputs from the auditory nerve and bilateral inhibition from feed-forward neurons. The feed-forward inhibitory neurons compute the difference in inputs coming in from both sides and relay it to the M-cells that elicit an appropriate and adaptive response from the spinal motor neurons. This elegant motif is played out again and again in the nervous system to make these simple and adaptive choices. Similarly, the midbrain superior colliculus (optic tectum in non-mammalian vertebrates) is important for orienting and selective attention and uses mutual inhibition of lateral inhibition in its circuitry to pivot to the most salient aspect of the visual field. These and many other studies on this evolutionarily conserved circuit strongly suggest that it may be used more broadly in the nervous system and more broadly for functions including, making behavioral choices, noise reduction, pattern completion, categorization, and associative memory.

Conclusion

In conclusion, it is imperative to emphasize that these circuit motifs are deployed widely in the brain, and are adapted by each individual region for their specific, idiosyncratic purpose. Modeling done to tease out specific aspects of cellular circuitry is constrained by the complexity of cell types, connections and synapses, but have been able to recapitulate previous in vivo and in vitro experimental data. Studies that rely on electrophysiological recordings and morphological reconstruction of different neurons have limitations, such as the tissue tends to be just a few hundred micrometer thick, with severed thalamic, sub- and cortico-cortical connection. These connections and their neuromodulatory inputs massively influence individual cell function and the activity of the whole circuit. Furthermore, the same neuromodulator could have different, excitatory or inhibitory effects on different cell types in the same cortical region based on the expression of particular receptors.

The work on neural circuits has spanned centuries and now with the availability of and further advancements in technology it is feasible to conduct highly sophisticated experiments on behaving animals to gain further insights into the inner workings of cortical circuits. These experiments undertaken to tease out different components of a cellular circuit that act in concert to define its phenotype combine, - wide-scale and simultaneous extracellular recordings from thousands of cells, with state of the art imaging and cell manipulations, along with genetic characterization of molecular markers of individual cells - and have already begun to bear seed, and promise exciting and rapid advances in our understanding of brain and its diseases.