Search

Biological neuronal networks (BNNs) are a source of inspiration and analogy making for researchers that focus on artificial neuronal networks (ANNs). Moreover, neuroscientists increasingly use ANNs as a model for the brain. Despite certain similarities between these two types of networks, important differences can be discerned. First, biological neural networks are sculpted by evolution and the constraints that it entails, whereas artificial neural networks are engineered to solve particular tasks. Second, the network topology of these systems, apart from some analogies that can be drawn, exhibits pronounced differences. Here, we examine strategies to construct recurrent neural networks (RNNs) that instantiate the network topology of brains of different species. We refer to such RNNs as bio-instantiated. We investigate the performance of bio-instantiated RNNs in terms of: (i) the prediction performance itself, that is, the capacity of the network to minimize the cost function at hand in test data, and (ii) speed of training, that is, how fast during training the network reaches its optimal performance. We examine bio-instantiated RNNs in working memory tasks where task-relevant information must be tracked as a sequence of events unfolds in time. We highlight the strategies that can be used to construct RNNs with the network topology found in BNNs, without sacrificing performance. Despite that we observe no enhancement of performance when compared to randomly wired RNNs, our approach demonstrates how empirical neural network data can be used for constructing RNNs, thus, facilitating further experimentation with biologically realistic network topologies, in contexts where such aspect is desired.

International audience; The ‘Sprague Effect’ described in the seminal paper of James Sprague (Science 153:1544 e1547, 1966a) is an unexpected paradoxical effect in which a second brain lesion reversed functional deficits induced by an earlier lesion. It was observed initially in the cat where severe and permanent contralateral visually guided attentional deficits generated by the ablation of large areas of the visual cortex were reversed by the subsequent removal of the superior colliculus (SC) opposite to the cortical lesion or by the splitting of the collicular commissure. Physiologically, this effect has been explained in several ways-most notably by the reduction of the functional inhibition of the ipsilateral SC by the contralateral SC, and the restoration of normal interactions between cortical and midbrain structures after ablation. In the present review, we aim at reappraising the ‘Sprague Effect’ by critically analyzing studies that have been conducted in the feline and human brain. Moreover, we assess applications of the ‘Sprague Effect’ in the rehabilitation of visually guided atten- tional impairments by using non-invasive therapeutic approaches such as transcranial magnetic stimulation (TMS) and transcranial direct-current stimulation (tDCS). We also review theoretical models of the effect that emphasize the inhibition and balancing be- tween the two hemispheres and show implications for lesion inference approaches. Last, we critically review whether the resulting inter-hemispheric rivalry theories lead toward an efficient rehabilitation of stroke in humans. We conclude by emphasizing key challenges in the field of ‘Sprague Effect’ applications in order to design better therapies for brain-damaged patients.; L’effet Sprague décrit dans le document fondateur de James Sprague (Science 153: 1544 e1547, 1966a) est un effet paradoxal inattendu dans lequel une deuxième lésion cérébrale inversait les déficits fonctionnels induits par une lésion antérieure. Il a été observé initialement chez le chat où les déficits attentionnels controlatéraux sévères et permanents générés par l'ablation de grandes zones du cortex visuel ont été inversés par l'ablation ultérieure du colliculus supérieur (SC) opposé à la lésion corticale ou par la division de la commissure colliculaire. Physiologiquement, cet effet a été expliqué de plusieurs façons, notamment par la réduction de l'inhibition fonctionnelle de la SC ipsilatérale par la SC controlatérale et la restauration des interactions normales entre les structures corticales et du mésencéphale après l'ablation. Dans la présente revue, nous visons à réévaluer «l’effet Sprague» en analysant de manière critique les études qui ont été menées sur le cerveau félin et humain. De plus, nous évaluons les applications de «l’effet Sprague» dans la réadaptation des troubles de l’attention guidés visuellement en utilisant des approches thérapeutiques non invasives telles que la stimulation magnétique transcrânienne (TMS) et la stimulation transcrânienne à courant continu (tDCS). Nous passons également en revue les modèles théoriques de l'effet qui mettent l'accent sur l'inhibition et l'équilibre entre les deux hémisphères et montrent les implications pour les approches d'inférence des lésions. Enfin, nous examinons de manière critique si les théories de rivalité interhémisphérique qui en résultent conduisent à une réhabilitation efficace de l'AVC chez l'homme. Nous concluons en mettant l’accent sur les principaux défis dans le domaine des applications de «l’effet Sprague» afin de concevoir de meilleures thérapies pour les patients atteints de lésions cérébrales.

Uncovering organizational principles of the cerebral cortex is essential for proper understanding of this prominent structure of the mammalian brain. The theory of the dual origin of the cerebral cortex offers such organizational principle. Here, we demonstrate that a duality pertains to the connectional architecture of the cerebral cortex of different mammals. This dual structure also constitutes a major axis of organization of the transcriptional architecture of the cortex and reflects the expression of different morphogens stemming from distinct patterning centers in the developing pallium. The duality of the cortex is also reflected in its spatial dimension, highlighting cortical areas as spatially ordered constellations that are centered around the paleocortex and archicortex, with the later primordial moieties reminiscent of antipodal points in the cortical sheet. The ontogeny of the uncovered dual connectional structure might be rooted in heterochronous neurodevelopmental gradients in the developing pallium, a suggestion corroborated by computational modeling. In all, the current results exemplify the duality of the cerebral cortex as an overarching organizational principle, reflected across the different levels of cortical architecture of different mammalian species, defining a natural axis of mammalian cortical organization.

A key component of current theories of brain structure and function is the layer-specific origin of structural connections of the cerebral cortex. This fundamental connectional feature pertains to different mammalian cortices, and recent neuroimaging advancements have started to pave the way for its function-based mapping in humans. Here, we propose a framework that systematically explains the characteristic layer-specific origin of structural connections and its graded variation across the cortical sheet and across mammalian species. The framework unifies seemingly dispersed observations on multiple levels of cortical organization, including the cellular, connectional, and functional level. Moreover, the framework allows the prediction of the layer-specific origin of connections in a spectrum of mammals, from rodents to humans.

Intrinsic coupling constitutes a key feature of ongoing brain activity, which exhibits rich spatiotemporal patterning and contains information that influences cognitive processing. We discuss evidence for two distinct types of intrinsic coupling modes which seem to reflect the operation of different coupling mechanisms. One type arises from phase coupling of band-limited oscillatory signals, whereas the other results from coupled aperiodic fluctuations of signal envelopes. The two coupling modes differ in their dynamics, their origins, and their putative functions and with respect to their alteration in neuropsychiatric disorders. We propose that the concept of intrinsic coupling modes can provide a unifying framework for capturing the dynamics of intrinsically generated neuronal interactions at multiple spatial and temporal scales.

How signals propagate through a network as a function of the network architecture and under the influence of noise is a fundamental question in a broad range of areas dealing with signal processing - from neuroscience to electrical engineering and communication technology. Here we use numerical simulations and a mean-field approach to analyze a minimal dynamic model for signal propagation. By labeling and tracking the excitations propagating from a single input node to remote output nodes in random networks, we show that noise (provided by spontaneous node excitations) can lead to an enhanced signal propagation, with a peak in the signal-to-noise ratio at intermediate noise intensities. This network analog of stochastic resonance is not captured by a mean-field description that incorporates topology only on the level of the average degree, indicating that the detailed network topology plays a significant role in signal propagation.

The dorsal fronto-parietal attention network of the human brain plays a central role in tasks with conflicting visual stimuli, such as the flanker task. It is assumed to mediate spatial attention on the target and the distractor and thus influence the saliency of the target and resulting response times (RT) of subjects. Here we causally localized contributing ‘hot spots’ within the parietal and frontal cortical regions of the network using online Transcranial Magnetic Stimulation (TMS). Using a visual flanker task, subjects (n = 12) were instructed to identify a central target (‘>’ vs). Subjects responded faster to congruent than incongruent stimuli, indicating the presence of a flanker effect. Subject-specific hot spots were identified where TMS pulses significantly reduced the flanker effect for contralateral but not ipsilateral flankers. One, or two adjacent, hot spots were identified for all subjects in parietal as well as frontal regions. We conclude that online TMS can selectively disrupt attentional control of the contralateral visual field and identify contributing cortical hot spots with high spatial precision.

Many neural networks, such as the complex cortical networks of the mammalian brain, are organized in multiple clusters, with many connections within but few links between clusters. To generate this organization, we explored a wiring rule in which the establishment of a connection between two areas depended on the areas' distance as well as their respective time windows for connection establishment. The concept of time windows was based on the observation that the development of cortical areas partially overlaps in time. Our algorithm was able to generate multiple network clusters depending on the number and overlap of the time windows.

Identifying the functional roles of elements of a neural network is one of the fundamental challenges in understanding neural information processing. Aiming at this goal, lesion studies have been used extensively in neuroscience. Most of these employ single lesions and hence, limited ability in revealing the significance of interacting elements. This paper presents the multi-perturbation Shapley value analysis (MSA), an axiomatic, scalable and rigorous method, addressing the challenge of determining the contributions of network elements from a data set of multi-lesions or other perturbations. The successful workings of the MSA are demonstrated on artificial and biological data. MSA is a novel method for causal function localization, with a wide range of potential applications for the analysis of reversible deactivation experiments and TMS-induced “virtual lesions”.

The brain shows a highly specific structural organization, also in its connectivity. How may genetic specification or epigenetic self-organization influence the layout of cortical systems networks? We developed a computational model of spatial network growth to investigate the conditions leading to similar network density and clustering as found in the cortical connectivity of cat and macaque monkey. An essential factor in the model was the introduction of spatial borders. We further compared metric aspects of wiring in the macaque brain and the simulated networks, which suggest that cortical networks may trade perfect minimal wiring length for shorter average shortest paths among cortical areas.

Organization and function of the primate prefrontal cortices have been the subject of much discussion and speculation. We quantitatively analyzed the architecture of primate prefrontal cortices in order to provide more reliable foundations for this debate. Stereological data were obtained for more than 20 prefrontal areas in the rhesus monkey, including quantitative information on the densities of neurons, glia and different classes of neurons expressing calcium binding proteins as well as laminar depth. The results of multivariate data analyses indicated that the anatomical organization of prefrontal cortices is determined by categorical factors, such as the existence of a granular layer IV in some areas, and is also shaped by structural gradients, such as a systematic shift of cellular density from deep to superficial cortical layers progressing from medial and orbital to lateral areas.

We explored the mechanisms of spatial attention in the cat with a combination of permanent and reversible lesion techniques. Animals with large unilateral visual cortical lesions displayed profound blindness of contralesional visual space. However, cooling deactivation of either posterior parietal cortex or the superior colliculus (SC) in the contralesional half of the brain restored visual orienting responses in the blind hemifield. Interestingly, gradual SC deactivation by decreasing cooling loop temperatures produced a topographic restoration of orienting responses, starting at the midline and progressing towards more peripheral eccentricities in the previously blind field. These effects can be explained within a computational model for spatial orienting, which represents the interhemispheric competition of bilateral brain structures that possess a topographic, center-magnified representation of the external visual space.

According to the structural model of prefrontal cortex, the pattern of corticocortical connections is intricately linked to cortical architecture (Cereb. Cortex 7 (1997) 635). We further explored this model by using quantitative methods to describe the structure and connections of prefrontal cortices. Multi-parameter analyses distinguished different cortical types, positioning at one extreme medial and orbitofrontal (limbic) cortices, and at the other extreme, lateral (eulaminate) cortices. The structural model accurately predicted the laminar pattern of connections, and the relative distribution of connections within cortical layers, based on cortical type. This model may provide the foundation to predict the nature of corticocortical processing and its disruption in psychiatric and neurologic diseases, where neuropathology affects specific types of neurons and layers.

Systematic collations and computational analyses of connectivity data from anatomical tract tracing can yield reliable and valuable insights into the principles governing cortical network organization. Based on an innovative databasing approach (http://www.cocomac.org) we present a preliminary analysis of the connectional organization of primate prefrontal cortex. Multivariate statistics of the connections between Walker's areas reveal distinct orbito-medial and lateral networks linked by anterior cingulate and dorsal prefrontal areas. Novel indices of areal network participation show that ventrolateral prefrontal cortex is a unique relay of information from lateral areas affecting the entire region.

While the anatomical connectivity of the macaque visual system is organized hierarchically, many dorsal cortical visual areas respond nearly simultaneously to flashed stimuli. Does this contradict hierarchical processing? To address this issue, we constructed an integrate-and-fire network, consisting of M and P layers of LGN, and of several visual cortical areas for which latency data were available. The neural network implemented the areas’ known laminar connectivity. The model displayed simultaneous onset of activity in several dorsal stream areas, closely matching experimental observations. Thus, anatomical organization and neurophysiological data are not contradictory, and hierarchical processing does not imply sequential response timing.

Progress in the understanding of brain function depends on comprehensive knowledge of neural structure and organization, based on reliable quantitative data. We quantified and analysed data about long-range connections between different visual cortical areas of the cat. The analyses were based on a numeric evaluation of the retrograde distribution of anatomical label in 16 visual cortical areas after injections of the middle suprasylvian cortical region. Although connection profiles and connection densities varied substantially between injections in different individual animals, characteristic features of corticocortical connectivity emerged. Our results suggest that long-range connections in the cortex obey global constraints, such as a linear relationship between connection density and area size of connection origin, or a sequential organization of laminar connectivity between different visual areas. Within the limits set by these global constraints, individual areas also possess specific connectivity characteristics, which hint on specialized functional interactions between cortical regions.

Spatial attention is an important biological tool that helps animals to orient in their environment as well as to detect food and predators. We have investigated the neural basis of attentional mechanisms with a combined approach of behavioral testing, ‘virtual lesion’ techniques and computational modeling. Cats with unilateral deactivations of either posterior parietal cortex or the superior colliculus exhibited a robust hemineglect for all novel visual stimuli presented in the contralateral space. Bilateral deactivation of the same neural structures, however, led to restoration of spatial attentional behavior. A simple mathematical model, based on known cortical and midbrain connectivity of the cat, was used to simulate the neural mechanisms underlying intact, unilaterally impaired and bilaterally restored attentional behavior, as well as to produce predictions for further experiments. Our results outline the contributions of cortical and subcortical stations to attention-guided behavior, and characterize functional interactions between structures in either half of the brain.

Spatial visual attention in humans and mammals is severely disrupted after deactivation of specific cortical or subcortical regions. Such dysfunction has frequently been associated with impaired unilateral cortical mechanisms. Alternatively, I propose that essential mechanisms of spatial attention are based on the bilateral competition between subcortical midbrain structures. Consequently, visuospatial hemineglect and hemi-extinction arise from unbalanced bilateral competition. I present a simple mathematical model, based on known connectivity of the cat midbrain, that exemplifies the principle of competition through mutual inter-hemispheric inhibition. The model represents a linear, topographic integration of several cortical and subcortical inputs in the two halves of the feline superior colliculus. Such a system reproduced a variety of neglect symptoms and also explained the paradoxical effects observed in some lesion experiments where the consequences of a primary lesion could be reversed through a secondary lesion in structures of the contralateral hemisphere. The model generated predictions for future experiments.

Studies of the effects of brain lesions on behaviour have informed brain sciences for more than 100 years. Paradoxical results from some experiments, however, contradict widely accepted logic for imputing function to structure, including the `gold standard'; double dissociation. Orienting systems have produced clear examples of paradoxical lesion effects and provide the opportunity to identify reliable inference for imputing function. Our models reproduced the experimental effects. Analysis of the models demonstrated why single and double dissociation studies do not yield reliable conclusions, but suggested more reliable methods for understanding the delegation of functions to neuroanatomical structures in large-scale neural networks.

The brain effortlessly recombines information about the shape, colour, motion and so on of objects in the visual scene, but how it does so is not known. Synchronous neuronal firing has seemed an attractive solution to this problem, but new results and theoretical insights cast doubt on its functional role.

In his thought-provoking review of current concepts in neuroscience, Pessoa [1] addresses the ongoing paradigm shift of the field, in which the perspective has moved from individual nodes to distributed networks in order to account for distributed brain function. Within this perspective, Pessoa describes diverse aspects and topological features of brain networks that are potentially relevant for brain function. As he notes, however, the shift to networks does not solve all problems of linking brain function to structure. In particular, Pessoa considers one issue to be central, the correspondence problem of mapping many functions to many elements, where a single element may contribute to multiple functions, and a particular function may involve many system elements. This problem is not simply resolved by attributing function to network features, such as modules, hubs, rich clubs, etc., instead of nodes. Nonetheless, it seems that the problem can be addressed in a straightforward manner by defining indices for participation, similar to the diversity profiles described by Pessoa, or comparable measures for localization and specialization derived from functional contribution matrices, e.g. [2]. In such approaches, which consider the functional contributions of a set of neural elements to a set of tasks, elements that are contributing highly and significantly to a particular task can be said to be specialized for the task, while functions that show particularly high involvement of some elements might be said to be localized in these elements. (Naturally, as also pointed out by Pessoa, an important question is how functions of neural elements should be defined in the first place. Any definition that is not matched to the actual biophysical, computational or functional properties of the elements is of little value [3].) While the correspondence problem may be addressed in a straightforward way, there exist some other, fundamental problems of the network perspective, of which we outline two. First, the role of statistical procedures in network research needs to be discussed on a much more fundamental level. Current research into brain networks is largely exploratory and focuses on issues such as computing centrality indices or evaluating whether the brain possesses small world features. It is important to note that there is little value per se in such explorative findings. The purpose of exploratory research is to help forming intuitions about the

It is widely known that the human frontal cortex is involved in the allocation of visual attention, however, the exact causal functional contributions of individual subregions are not well understood. In the present study, we used a simple visual localization task and applied rTMS pulses to map frontal cortical subregions likely to generate significant visuo-spatial biases during the spatial deployment of attention prior to perception. A group of nine adult healthy subjects executed a task based on the localization of small rectangular black dots. Targets were displayed unilaterally (left or right) or bilaterally (left and right) for a brief period (40 ms). In a systematic mapping approach, a stimulation grid of 9 (3 × 3) sites was anchored 2 cm rostral to the motor hand area. Three pulses of real or sham 10 Hz rTMS were delivered in a counterbalanced manner, at each of the grid locations 50 ms post target onset to interfere with the ongoing neural processing. As a main finding, significant deterioration of detection performance, as measured by effective reaction time, for stimuli in the contralateral hemifield and increased performance for ipsilateral targets were observed for two grid regions anatomically associated with the right FEF and right middle frontal gyrus. We conclude that the disruptive effects of TMS on a simple spatial localization task for which a well-balanced deployment of attention is required are exquisitely spatially selective, and are found in specific frontal cortical subregions.

Repetitive transcranial magnetic stimulation (rTMS) in ‘offline’ experiments is commonly administered to participants that are in a relaxed ’resting state’. Clearly, this approach leads to an ill controlled baseline state of the participants’ brain, in face of evidence that the mental baseline influences how the brain is affected by TMS (e.g., Silvanto and Pascual-Leone (2008) Brain Topography 21). The aim of the present study was to assess the effect of uncontrolled vs. controlled behavior during rTMS administration on participants’ performance in a subsequent visual localization task. We compared the offline effects of 10 Hz rTMS for standard TMS administration (uncontrolled ‘resting’ behavior without task) to 10 Hz rTMS administered while the participants were performing a visual localization task (controlled behavior with task). Tasks during the online and offline period were identical, testing the participants’ ability to allocate attention to the detection of peripheral unilateral (left or right) or bilateral (left and right) peri-threshold visual stimuli. rTMS or sham stimulation (by vertically tilted coil) were applied to the 10–20 coordinate P4, located in the right posterior parietal cortex. Performance after rTMS in the ‘resting’ condition appeared similar to the performance in the ‘with task’ condition. Importantly, however, when compared to sham stimulation conditions, rTMS induced a deterioration of offline performance relative to sham stimulation in the ‘resting’ condition, whereas it led to better performance relative to sham in the ‘with task’ condition ( Fig. 1 ). We conclude that in the present paradigm, the manipulation ‘with task’ versus ‘resting’ during the administration of rTMS might be too weak to induce brain states that are distinct enough to induce different levels of performance in the subsequent task. From this perspective, controlling the participants’ behavior during rTMS administration does not appear to matter for their offline behavior after rTMS. However, the results of the study also demonstrate that whether rTMS leads to a deterioration or improvement of performance critically depends on the baseline to which the offline performance is compared. Offline performance after rTMS or sham stimulation applied to participants during a behavioral task or at rest. Performance was evaluated as inverse efficiency, computed as reaction time/ proportion of correct trials. Shown are the results for bilateral visual stimuli. Within-subject error bars were created using the method of Cousineau (2005).