From local counterfactuals to global feature importance: efficient, robust, and model-agnostic explanations for brain connectivity networks.

BACKGROUND: Explainable artificial intelligence (XAI) is a technology that can enhance trust in mental state classifications by providing explanations for the reasoning behind artificial intelligence (AI) models outputs, especially for high-dimensional and highly-correlated brain signals. Feature importance and counterfactual explanations are two common approaches to generate these explanations, but both have drawbacks. While feature importance methods, such as shapley additive explanations (SHAP), can be computationally expensive and sensitive to feature correlation, counterfactual explanations only explain a single outcome instead of the entire model. METHODS: To overcome these limitations, we propose a new procedure for computing global feature importance that involves aggregating local counterfactual explanations. This approach is specifically tailored to fMRI signals and is based on the hypothesis that instances close to the decision boundary and their counterfactuals mainly differ in the features identified as most important for the downstream classification task. We refer to this proposed feature importance measure as Boundary Crossing Solo Ratio (BoCSoR), since it quantifies the frequency with which a change in each feature in isolation leads to a change in classification outcome, i.e., the crossing of the model's decision boundary. RESULTS AND CONCLUSIONS: Experimental results on synthetic data and real publicly available fMRI data from the Human Connect project show that the proposed BoCSoR measure is more robust to feature correlation and less computationally expensive than state-of-the-art methods. Additionally, it is equally effective in providing an explanation for the behavior of any AI model for brain signals. These properties are crucial for medical decision support systems, where many different features are often extracted from the same physiological measures and a gold standard is absent. Consequently, computing feature importance may become computationally expensive, and there may be a high probability of mutual correlation among features, leading to unreliable results from state-of-the-art XAI methods.

Neuronal network activity and connectivity are impaired in a conditional knockout mouse model with PCDH19 mosaic expression.

Mutations in PCDH19 gene, which encodes protocadherin-19 (PCDH19), cause Developmental and Epileptic Encephalopathy 9 (DEE9). Heterogeneous loss of PCDH19 expression in neurons is considered a key determinant of the disorder; however, how PCDH19 mosaic expression affects neuronal network activity and circuits is largely unclear. Here, we show that the hippocampus of Pcdh19 mosaic mice is characterized by structural and functional synaptic defects and by the presence of PCDH19-negative hyperexcitable neurons. Furthermore, global reduction of network firing rate and increased neuronal synchronization have been observed in different limbic system areas. Finally, network activity analysis in freely behaving mice revealed a decrease in excitatory/inhibitory ratio and functional hyperconnectivity within the limbic system of Pcdh19 mosaic mice. Altogether, these results indicate that altered PCDH19 expression profoundly affects circuit wiring and functioning, and provide new key to interpret DEE9 pathogenesis.

Three-dimensional unsupervised probabilistic pose reconstruction (3D-UPPER) for freely moving animals.

A key step in understanding animal behaviour relies in the ability to quantify poses and movements. Methods to track body landmarks in 2D have made great progress over the last few years but accurate 3D reconstruction of freely moving animals still represents a challenge. To address this challenge here we develop the 3D-UPPER algorithm, which is fully automated, requires no a priori knowledge of the properties of the body and can also be applied to 2D data. We find that 3D-UPPER reduces by [Formula: see text] fold the error in 3D reconstruction of mouse body during freely moving behaviour compared with the traditional triangulation of 2D data. To achieve that, 3D-UPPER performs an unsupervised estimation of a Statistical Shape Model (SSM) and uses this model to constrain the viable 3D coordinates. We show, by using simulated data, that our SSM estimator is robust even in datasets containing up to 50% of poses with outliers and/or missing data. In simulated and real data SSM estimation converges rapidly, capturing behaviourally relevant changes in body shape associated with exploratory behaviours (e.g. with rearing and changes in body orientation). Altogether 3D-UPPER represents a simple tool to minimise errors in 3D reconstruction while capturing meaningful behavioural parameters.

Look-up and look-down neurons in the mouse visual thalamus during freely moving exploration.

Visual information reaches cortex via the thalamic dorsal lateral geniculate nucleus (dLGN). dLGN activity is modulated by global sleep/wake states and arousal, indicating that it is not simply a passive relay station. However, its potential for more specific visuomotor integration is largely unexplored. We addressed this question by developing robust 3D video reconstruction of mouse head and body during spontaneous exploration paired with simultaneous neuronal recordings from dLGN. Unbiased evaluation of a wide range of postures and movements revealed a widespread coupling between neuronal activity and few behavioral parameters. In particular, postures associated with the animal looking up/down correlated with activity in >50% neurons, and the extent of this effect was comparable with that induced by full-body movements (typically locomotion). By contrast, thalamic activity was minimally correlated with other postures or movements (e.g., left/right head and body torsions). Importantly, up/down postures and full-body movements were largely independent and jointly coupled to neuronal activity. Thus, although most units were excited during full-body movements, some expressed highest firing when the animal was looking up ("look-up" neurons), whereas others expressed highest firing when the animal was looking down ("look-down" neurons). These results were observed in the dark, thus representing a genuine behavioral modulation, and were amplified in a lit arena. Our results demonstrate that the primary visual thalamus, beyond global modulations by sleep/awake states, is potentially involved in specific visuomotor integration and reveal two distinct couplings between up/down postures and neuronal activity.

Nonlinear machine learning pattern recognition and bacteria-metabolite multilayer network analysis of perturbed gastric microbiome.

The stomach is inhabited by diverse microbial communities, co-existing in a dynamic balance. Long-term use of drugs such as proton pump inhibitors (PPIs), or bacterial infection such as Helicobacter pylori, cause significant microbial alterations. Yet, studies revealing how the commensal bacteria re-organize, due to these perturbations of the gastric environment, are in early phase and rely principally on linear techniques for multivariate analysis. Here we disclose the importance of complementing linear dimensionality reduction techniques with nonlinear ones to unveil hidden patterns that remain unseen by linear embedding. Then, we prove the advantages to complete multivariate pattern analysis with differential network analysis, to reveal mechanisms of bacterial network re-organizations which emerge from perturbations induced by a medical treatment (PPIs) or an infectious state (H. pylori). Finally, we show how to build bacteria-metabolite multilayer networks that can deepen our understanding of the metabolite pathways significantly associated to the perturbed microbial communities.

A High-Dimensional Quantification of Mouse Defensive Behaviors Reveals Enhanced Diversity and Stimulus Specificity.

Instinctive defensive behaviors, consisting of stereotyped sequences of movements and postures, are an essential component of the mouse behavioral repertoire. Since defensive behaviors can be reliably triggered by threatening sensory stimuli, the selection of the most appropriate action depends on the stimulus property. However, since the mouse has a wide repertoire of motor actions, it is not clear which set of movements and postures represent the relevant action. So far, this has been empirically identified as a change in locomotion state. However, the extent to which locomotion alone captures the diversity of defensive behaviors and their sensory specificity is unknown. To tackle this problem, we developed a method to obtain a faithful 3D reconstruction of the mouse body that enabled to quantify a wide variety of motor actions. This higher dimensional description revealed that defensive behaviors are more stimulus specific than indicated by locomotion data. Thus, responses to distinct stimuli that were equivalent in terms of locomotion (e.g., freezing induced by looming and sound) could be discriminated along other dimensions. The enhanced stimulus specificity was explained by a surprising diversity. A clustering analysis revealed that distinct combinations of movements and postures, giving rise to at least 7 different behaviors, were required to account for stimulus specificity. Moreover, each stimulus evoked more than one behavior, revealing a robust one-to-many mapping between sensations and behaviors that was not apparent from locomotion data. Our results indicate that diversity and sensory specificity of mouse defensive behaviors unfold in a higher dimensional space, spanning multiple motor actions.

Short-Term Classification Learning Promotes Rapid Global Improvements of Information Processing in Human Brain Functional Connectome.

Classification learning is a preeminent human ability within the animal kingdom but the key mechanisms of brain networks regulating learning remain mostly elusive. Recent neuroimaging advancements have depicted human brain as a complex graph machinery where brain regions are nodes and coherent activities among them represent the functional connections. While long-term motor memories have been found to alter functional connectivity in the resting human brain, a graph topological investigation of the short-time effects of learning are still not widely investigated. For instance, classification learning is known to orchestrate rapid modulation of diverse memory systems like short-term and visual working memories but how the brain functional connectome accommodates such modulations is unclear. We used publicly available repositories (openfmri.org) selecting three experiments, two focused on short-term classification learning along two consecutive runs where learning was promoted by trial-by-trial feedback errors, while a further experiment was used as supplementary control. We analyzed the functional connectivity extracted from BOLD fMRI signals, and estimated the graph information processing in the cerebral networks. The information processing capability, characterized by complex network statistics, significantly improved over runs, together with the subject classification accuracy. Instead, null-learning experiments, where feedbacks came with poor consistency, did not provoke any significant change in the functional connectivity over runs. We propose that learning induces fast modifications in the overall brain network dynamics, definitely ameliorating the short-term potential of the brain to process and integrate information, a dynamic consistently orchestrated by modulations of the functional connections among specific brain regions.

Single unit activities recorded in the thalamus and the overlying parietal cortex of subjects affected by disorders of consciousness.

The lack of direct neurophysiological recordings from the thalamus and the cortex hampers our understanding of vegetative state/unresponsive wakefulness syndrome and minimally conscious state in humans. We obtained microelectrode recordings from the thalami and the homolateral parietal cortex of two vegetative state/unresponsive wakefulness syndrome and one minimally conscious state patients during surgery for implantation of electrodes in both thalami for chronic deep brain stimulation. We found that activity of the thalamo-cortical networks differed among the two conditions. There were half the number of active neurons in the thalami of patients in vegetative state/unresponsive wakefulness syndrome than in minimally conscious state. Coupling of thalamic neuron discharge with EEG phases also differed in the two conditions and thalamo-cortical cross-frequency coupling was limited to the minimally conscious state patient. When consciousness is physiologically or pharmacologically reversibly suspended there is a significant increase in bursting activity of the thalamic neurons. By contrast, in the thalami of our patients in both conditions fewer than 17% of the recorded neurons showed bursting activity. This indicates that these conditions differ from physiological suspension of consciousness and that increased thalamic inhibition is not prominent. Our findings, albeit obtained in a limited number of patients, unveil the neurophysiology of these conditions at single unit resolution and might be relevant for inspiring novel therapeutic options.

Alternating Dynamics of Segregation and Integration in Human EEG Functional Networks During Working-memory Task.

Brain functional networks show high variability in short time windows but mechanisms governing these transient dynamics remain unknown. In this work, we studied the temporal evolution of functional brain networks involved in a working memory (WM) task while recording high-density electroencephalography (EEG) in human normal subjects. We found that functional brain networks showed an initial phase characterized by an increase of the functional segregation index followed by a second phase where the functional segregation faded after the prevailing the functional integration. Notably, wrong trials were associated with different or disrupted sequences of the segregation-integration profiles and measures of network centrality and modularity were able to identify crucial aspects of the oscillatory network dynamics. Additionally, computational investigations further supported the experimental results. The brain functional organization may respond to the information processing demand of a WM task following a 2-step atomic scheme wherein segregation and integration alternately dominate the functional configurations.

The thalamo-cortical complex network correlates of chronic pain.

Chronic pain (CP) is a condition with a large repertory of clinical signs and symptoms with diverse expressions. Though widely analyzed, an appraisal at the level of single neuron and neuronal networks in CP is however missing. The present research proposes an empirical and theoretic framework which identifies a complex network correlate nested in the somatosensory thalamocortical (TC) circuit in diverse CP models. In vivo simultaneous extracellular neuronal electrophysiological high-density recordings have been performed from the TC circuit in rats. Wide functional network statistics neatly discriminated CP from control animals identifying collective dynamical traits. In particular, a collapsed functional connectivity and an altered modular architecture of the thalamocortical circuit have been evidenced. These results envisage CP as a functional connectivity disorder and give the clue for unveiling innovative therapeutic strategies.

Electrophysiological and Anatomical Correlates of Spinal Cord Optical Coherence Tomography.

Despite the continuous improvement in medical imaging technology, visualizing the spinal cord poses severe problems due to structural or incidental causes, such as small access space and motion artifacts. In addition, positional guidance on the spinal cord is not commonly available during surgery, with the exception of neuronavigation techniques based on static pre-surgical data and of radiation-based methods, such as fluoroscopy. A fast, bedside, intraoperative real-time imaging, particularly necessary during the positioning of endoscopic probes or tools, is an unsolved issue. The objective of our work, performed on experimental rats, is to demonstrate potential intraoperative spinal cord imaging and probe guidance by optical coherence tomography (OCT). Concurrently, we aimed to demonstrate that the electromagnetic OCT irradiation exerted no particular effect at the neuronal and synaptic levels. OCT is a user-friendly, low-cost and endoscopy-compatible photonics-based imaging technique. In particular, by using a Fourier-domain OCT imager, operating at 850 nm wavelength and scanning transversally with respect to the spinal cord, we have been able to: 1) accurately image tissue structures in an animal model (muscle, spine bone, cerebro-spinal fluid, dura mater and spinal cord), and 2) identify the position of a recording microelectrode approaching and inserting into the cord tissue 3) check that the infrared radiation has no actual effect on the electrophysiological activity of spinal neurons. The technique, potentially extendable to full three-dimensional image reconstruction, shows prospective further application not only in endoscopic intraoperative analyses and for probe insertion guidance, but also in emergency and adverse situations (e.g. after trauma) for damage recognition, diagnosis and fast image-guided intervention.

Results of a prospective study (CATS) on the effects of thalamic stimulation in minimally conscious and vegetative state patients.

OBJECTIVE Deep brain stimulation of the thalamus was introduced more than 40 years ago with the objective of improving the performance and attention of patients in a vegetative or minimally conscious state. Here, the authors report the results of the Cortical Activation by Thalamic Stimulation (CATS) study, a prospective multiinstitutional study on the effects of bilateral chronic stimulation of the anterior intralaminar thalamic nuclei and adjacent paralaminar regions in patients affected by a disorder of consciousness. METHODS The authors evaluated the clinical and radiological data of 29 patients in a vegetative state (unresponsive wakefulness syndrome) and 11 in a minimally conscious state that lasted for more than 6 months. Of these patients, 5 were selected for bilateral stereotactic implantation of deep brain stimulating electrodes into their thalamus. A definitive consensus for surgery was obtained for 3 of the selected patients. All 3 patients (2 in a vegetative state and 1 in a minimally conscious state) underwent implantation of bilateral thalamic electrodes and submitted to chronic stimulation for a minimum of 18 months and a maximum of 48 months. RESULTS In each case, there was an increase in desynchronization and the power spectrum of electroencephalograms, and improvement in the Coma Recovery Scale-Revised scores was found. Furthermore, the severity of limb spasticity and the number and severity of pathological movements were reduced. However, none of these patients returned to a fully conscious state. CONCLUSIONS Despite the limited number of patients studied, the authors confirmed that bilateral thalamic stimulation can improve the clinical status of patients affected by a disorder of consciousness, even though this stimulation did not induce persistent, clinically evident conscious behavior in the patients. Clinical trial registration no.: NCT01027572 ( ClinicalTrials.gov ).

Electrophysiological effects of non-invasive Radio Electric Asymmetric Conveyor (REAC) on thalamocortical neural activities and perturbed experimental conditions.

The microwave emitting Radio Electric Asymmetric Conveyor (REAC) is a technology able to interact with biological tissues at low emission intensity (2 mW at the emitter and 2.4 or 5.8 GHz) by inducing radiofrequency generated microcurrents. It shows remarkable biological effects at many scales from gene modulations up to functional global remodeling even in human subjects. Previous REAC experiments by functional Magnetic Resonance Imaging (fMRI) on healthy human subjects have shown deep modulations of cortical BOLD signals. In this paper we studied the effects of REAC application on spontaneous and evoked neuronal activities simultaneously recorded by microelectrode matrices from the somatosensory thalamo-cortical axis in control and chronic pain experimental animal models. We analyzed the spontaneous spiking activity and the Local Field Potentials (LFPs) before and after REAC applied with a different protocol. The single neuron spiking activities, the neuronal responses to peripheral light mechanical stimuli, the population discharge synchronies as well as the correlations and the network dynamic connectivity characteristics have been analyzed. Modulations of the neuronal frequency associated with changes of functional correlations and significant LFP temporal realignments have been diffusely observed. Analyses by topological methods have shown changes in functional connectivity with significant modifications of the network features.

Quantifying the Number of Discriminable Coincident Dendritic Input Patterns through Dendritic Tree Morphology.

Current developments in neuronal physiology are unveiling novel roles for dendrites. Experiments have shown mechanisms of non-linear synaptic NMDA dependent activations, able to discriminate input patterns through the waveforms of the excitatory postsynaptic potentials. Contextually, the synaptic clustering of inputs is the principal cellular strategy to separate groups of common correlated inputs. Dendritic branches appear to work as independent discriminating units of inputs potentially reflecting an extraordinary repertoire of pattern memories. However, it is unclear how these observations could impact our comprehension of the structural correlates of memory at the cellular level. This work investigates the discrimination capabilities of neurons through computational biophysical models to extract a predicting law for the dendritic input discrimination capability (M). By this rule we compared neurons from a neuron reconstruction repository (neuromorpho.org). Comparisons showed that primate neurons were not supported by an equivalent M preeminence and that M is not uniformly distributed among neuron types. Remarkably, neocortical neurons had substantially less memory capacity in comparison to those from non-cortical regions. In conclusion, the proposed rule predicts the inherent neuronal spatial memory gathering potentially relevant anatomical and evolutionary considerations about the brain cytoarchitecture.

A novel wireless recording and stimulating multichannel epicortical grid for supplementing or enhancing the sensory-motor functions in monkey (Macaca fascicularis).

Artificial brain-machine interfaces (BMIs) represent a prospective step forward supporting or replacing faulty brain functions. So far, several obstacles, such as the energy supply, the portability and the biocompatibility, have been limiting their effective translation in advanced experimental or clinical applications. In this work, a novel 16 channel chronically implantable epicortical grid has been proposed. It provides wireless transmission of cortical recordings and stimulations, with induction current recharge. The grid has been chronically implanted in a non-human primate (Macaca fascicularis) and placed over the somato-motor cortex such that 13 electrodes recorded or stimulated the primary motor cortex and three the primary somatosensory cortex, in the deeply anaesthetized animal. Cortical sensory and motor recordings and stimulations have been performed within 3 months from the implant. In detail, by delivering motor cortex epicortical single spot stimulations (1-8 V, 1-10 Hz, 500 ms, biphasic waves), we analyzed the motor topographic precision, evidenced by tunable finger or arm movements of the anesthetized animal. The responses to light mechanical peripheral sensory stimuli (blocks of 100 stimuli, each single stimulus being <1 ms and interblock intervals of 1.5-4 s) have been analyzed. We found 150-250 ms delayed cortical responses from fast finger touches, often spread to nearby motor stations. We also evaluated the grid electrical stimulus interference with somatotopic natural tactile sensory processing showing no suppressing interference with sensory stimulus detection. In conclusion, we propose a chronically implantable epicortical grid which can accommodate most of current technological restrictions, representing an acceptable candidate for BMI experimental and clinical uses.

Radio electric asymmetric conveyer: a novel neuromodulation technology in Alzheimer's and other neurodegenerative diseases.

Global research in the field of pharmacology has not yet found effective drugs to treat Alzheimer's disease (AD). Thus, alternative therapeutic strategies are under investigation, such as neurostimulation by physical means. Radio electric asymmetric conveyer (REAC) is one of these technologies and has, until now, been used in clinical studies on several psychiatric and neurological disorders with encouraging results in the absence of side effects. Moreover, studies at the cellular level have shown that REAC technology, with the appropriate protocols, is able to induce neuronal differentiation both in murine embryonic cells and in human adult differentiated cells. Other studies have shown that REAC technology is able to positively influence senescence processes. Studies conducted on AD patients and in transgenic mouse models have shown promising results, suggesting REAC could be a useful therapy for certain components of AD.

Functional correlates of preserved naming performance in amnestic Mild Cognitive Impairment.

Naming abilities are typically preserved in amnestic Mild Cognitive Impairment (aMCI), a condition associated with increased risk of progression to Alzheimer's disease (AD). We compared the functional correlates of covert picture naming and word reading between a group of aMCI subjects and matched controls. Unimpaired picture naming performance was associated with more extensive activations, in particular involving the parietal lobes, in the aMCI group. In addition, in the condition associated with higher processing demands (blocks of categorically homogeneous items, living items), increased activity was observed in the aMCI group, in particular in the left fusiform gyrus. Graph analysis provided further evidence of increased modularity and reduced integration for the homogenous sets in the aMCI group. The functional modifications associated with preserved performance may reflect, in the case of more demanding tasks, compensatory mechanisms for the subclinical involvement of semantic processing areas by AD pathology.

The Compression Flow as a Measure to Estimate the Brain Connectivity Changes in Resting State fMRI and 18FDG-PET Alzheimer's Disease Connectomes.

The human brain appears organized in compartments characterized by seemingly specific functional purposes on many spatial scales. A complementary functional state binds information from specialized districts to return what is called integrated information. These fundamental network dynamics undergoes to severe disarrays in diverse degenerative conditions such as Alzheimer's Diseases (AD). The AD represents a multifarious syndrome characterized by structural, functional, and metabolic landmarks. In particular, in the early stages of AD, adaptive functional modifications of the brain networks mislead initial diagnoses because cognitive abilities may result indiscernible from normal subjects. As a matter of facts, current measures of functional integration fail to catch significant differences among normal, mild cognitive impairment (MCI) and even AD subjects. The aim of this work is to introduce a new topological feature called Compression Flow (CF) to finely estimate the extent of the functional integration in the brain networks. The method uses a Monte Carlo-like estimation of the information integration flows returning the compression ratio between the size of the injected information and the size of the condensed information within the network. We analyzed the resting state connectomes of 75 subjects of the Alzheimer's Disease Neuroimaging Initiative 2 (ADNI) repository. Our analyses are focused on the 18FGD-PET and functional MRI (fMRI) acquisitions in several clinical screening conditions. Results indicated that CF effectively discriminate MCI, AD and normal subjects by showing a significant decrease of the functional integration in the AD and MCI brain connectomes. This result did not emerge by using a set of common complex network statistics. Furthermore, CF was best correlated with individual clinical scoring scales. In conclusion, we presented a novel measure to quantify the functional integration that resulted efficient to discriminate different stages of dementia and to track the individual progression of the impairments prospecting a proficient usage in a wide range of pathophysiological and physiological studies as well.

A simple stimulatory device for evoking point-like tactile stimuli: a searchlight for LFP to spike transitions.

Current neurophysiological research has the aim to develop methodologies to investigate the signal route from neuron to neuron, namely in the transitions from spikes to Local Field Potentials (LFPs) and from LFPs to spikes. LFPs have a complex dependence on spike activity and their relation is still poorly understood(1). The elucidation of these signal relations would be helpful both for clinical diagnostics (e.g. stimulation paradigms for Deep Brain Stimulation) and for a deeper comprehension of neural coding strategies in normal and pathological conditions (e.g. epilepsy, Parkinson disease, chronic pain). To this aim, one has to solve technical issues related to stimulation devices, stimulation paradigms and computational analyses. Therefore, a custom-made stimulation device was developed in order to deliver stimuli well regulated in space and time that does not incur in mechanical resonance. Subsequently, as an exemplification, a set of reliable LFP-spike relationships was extracted. The performance of the device was investigated by extracellular recordings, jointly spikes and LFP responses to the applied stimuli, from the rat Primary Somatosensory cortex. Then, by means of a multi-objective optimization strategy, a predictive model for spike occurrence based on LFPs was estimated. The application of this paradigm shows that the device is adequately suited to deliver high frequency tactile stimulation, outperforming common piezoelectric actuators. As a proof of the efficacy of the device, the following results were presented: 1) the timing and reliability of LFP responses well match the spike responses, 2) LFPs are sensitive to the stimulation history and capture not only the average response but also the trial-to-trial fluctuations in the spike activity and, finally, 3) by using the LFP signal it is possible to estimate a range of predictive models that capture different aspects of the spike activity.

Neuronal functional connection graphs among multiple areas of the rat somatosensory system during spontaneous and evoked activities.

Small-World Networks (SWNs) represent a fundamental model for the comprehension of many complex man-made and biological networks. In the central nervous system, SWN models have been shown to fit well both anatomical and functional maps at the macroscopic level. However, the functional microscopic level, where the nodes of a network are represented by single neurons, is still poorly understood. At this level, although recent evidences suggest that functional connection graphs exhibit small-world organization, it is not known whether and how these maps, potentially distributed in multiple brain regions, change across different conditions, such as spontaneous and stimulus-evoked activities. We addressed these questions by analyzing the data from simultaneous multi-array extracellular recordings in three brain regions of rats, diversely involved in somatosensory information processing: the ventropostero-lateral thalamic nuclei, the primary somatosensory cortex and the centro-median thalamic nuclei. From both spike and Local Field Potential (LFP) recordings, we estimated the functional connection graphs by using the Normalized Compression Similarity for spikes and the Phase Synchrony for LFPs. Then, by using graph-theoretical statistics, we characterized the functional topology both during spontaneous activity and sensory stimulation. Our main results show that: (i) spikes and LFPs show SWN organization during spontaneous activity; (ii) after stimulation onset, while substantial functional graph reconfigurations occur both in spike and LFPs, small-worldness is nonetheless preserved; (iii) the stimulus triggers a significant increase of inter-area LFP connections without modifying the topology of intra-area functional connections. Finally, investigating computationally the functional substrate that supports the observed phenomena, we found that (iv) the fundamental concept of cell assemblies, transient groups of activating neurons, can be described by small-world networks. Our results suggest that activity of neurons from multiple areas of the rat somatosensory system contributes to the integration of local computations arisen in distributed functional cell assemblies according to the principles of SWNs.