Human Visual Cortex and Deep Convolutional Neural Network Care Deeply about Object Background.

Deep convolutional neural networks (DCNNs) are able to partially predict brain activity during object categorization tasks, but factors contributing to this predictive power are not fully understood. Our study aimed to investigate the factors contributing to the predictive power of DCNNs in object categorization tasks. We compared the activity of four DCNN architectures with EEG recordings obtained from 62 human participants during an object categorization task. Previous physiological studies on object categorization have highlighted the importance of figure-ground segregation-the ability to distinguish objects from their backgrounds. Therefore, we investigated whether figure-ground segregation could explain the predictive power of DCNNs. Using a stimulus set consisting of identical target objects embedded in different backgrounds, we examined the influence of object background versus object category within both EEG and DCNN activity. Crucially, the recombination of naturalistic objects and experimentally controlled backgrounds creates a challenging and naturalistic task, while retaining experimental control. Our results showed that early EEG activity (< 100 msec) and early DCNN layers represent object background rather than object category. We also found that the ability of DCNNs to predict EEG activity is primarily influenced by how both systems process object backgrounds, rather than object categories. We demonstrated the role of figure-ground segregation as a potential prerequisite for recognition of object features, by contrasting the activations of trained and untrained (i.e., random weights) DCNNs. These findings suggest that both human visual cortex and DCNNs prioritize the segregation of object backgrounds and target objects to perform object categorization. Altogether, our study provides new insights into the mechanisms underlying object categorization as we demonstrated that both human visual cortex and DCNNs care deeply about object background.

Variance analysis as a method to predict the locus of plasticity at populations of non-uniform synapses.

Our knowledge on synaptic transmission in the central nervous system has often been obtained by evoking synaptic responses to populations of synapses. Analysis of the variance in synaptic responses can be applied as a method to predict whether a change in synaptic responses is a consequence of altered presynaptic neurotransmitter release or postsynaptic receptors. However, variance analysis is based on binomial statistics, which assumes that synapses are uniform. In reality, synapses are far from uniform, which questions the reliability of variance analysis when applying this method to populations of synapses. To address this, we used an in silico model for evoked synaptic responses and compared variance analysis outcomes between populations of uniform versus non-uniform synapses. This simulation revealed that variance analysis produces similar results irrespectively of the grade of uniformity of synapses. We put this variance analysis to the test with an electrophysiology experiment using a model system for which the loci of plasticity are well established: the effect of amyloid-ß on synapses. Variance analysis correctly predicted that postsynaptically produced amyloid-ß triggered predominantly a loss of synapses and a minor reduction of postsynaptic currents in remaining synapses with little effect on presynaptic release probability. We propose that variance analysis can be reliably used to predict the locus of synaptic changes for populations of non-uniform synapses.

A Critical Test of Deep Convolutional Neural Networks' Ability to Capture Recurrent Processing in the Brain Using Visual Masking.

Recurrent processing is a crucial feature in human visual processing supporting perceptual grouping, figure-ground segmentation, and recognition under challenging conditions. There is a clear need to incorporate recurrent processing in deep convolutional neural networks, but the computations underlying recurrent processing remain unclear. In this article, we tested a form of recurrence in deep residual networks (ResNets) to capture recurrent processing signals in the human brain. Although ResNets are feedforward networks, they approximate an excitatory additive form of recurrence. Essentially, this form of recurrence consists of repeating excitatory activations in response to a static stimulus. Here, we used ResNets of varying depths (reflecting varying levels of recurrent processing) to explain EEG activity within a visual masking paradigm. Sixty-two humans and 50 artificial agents (10 ResNet models of depths -4, 6, 10, 18, and 34) completed an object categorization task. We show that deeper networks explained more variance in brain activity compared with shallower networks. Furthermore, all ResNets captured differences in brain activity between unmasked and masked trials, with differences starting at ∼98 msec (from stimulus onset). These early differences indicated that EEG activity reflected "pure" feedforward signals only briefly (up to ∼98 msec). After ∼98 msec, deeper networks showed a significant increase in explained variance, which peaks at ∼200 msec, but only within unmasked trials, not masked trials. In summary, we provided clear evidence that excitatory additive recurrent processing in ResNets captures some of the recurrent processing in humans.

Calcium signaling in individual APP/PS1 mouse dentate gyrus astrocytes increases ex vivo with Aß pathology and age without affecting astrocyte network activity.

Astrocytes are critical for healthy brain function. In Alzheimer's disease, astrocytes become reactive, which affects their signaling properties. Here, we measured spontaneous calcium transients ex vivo in hippocampal astrocytes in brain slices containing the dentate gyrus of 6- (6M) and 9-month-old (9M) APPswe/PSEN1dE9 (APP/PS1) mice. We investigated the frequency and duration of calcium transients in relation to aging, amyloid-ß (Aß) pathology, and the proximity of the astrocyte to Aß plaques. The 6M APP/PS1 astrocytes showed no change in spontaneous calcium-transient properties compared to wild-type (WT) astrocytes. 9M APP/PS1 astrocytes, however, showed more hyperactivity compared to WT, characterized by increased spontaneous calcium transients that were longer in duration. Our data also revealed an effect of aging, as 9M astrocytes overall showed an increase in calcium activity compared to 6M astrocytes. Subsequent calcium-wave analysis showed an increase in sequential calcium transients (i.e., calcium waves) in 9M astrocytes, suggesting increased network activity ex vivo. Further analysis using null models revealed that this network effect is caused by chance, due to the increased number of spontaneous transients. Our findings show that alterations in calcium signaling in individual hippocampal astrocytes of APP/PS1 mice are subject to both aging and Aß pathology but these do not lead to a change in astrocyte network activity. These alterations in calcium dynamics of astrocytes may help to understand changes in neuronal physiology leading to cognitive decline and ultimately dementia.

Interaction of Cortical and Amygdalar Synaptic Input Modulates the Window of Opportunity for Information Processing in the Rhinal Cortices.

The perirhinal (PER) and lateral entorhinal (LEC) cortex function as a gateway for information transmission between (sub)cortical areas and the hippocampus. It is hypothesized that the amygdala, a key structure in emotion processing, can modulate PER-LEC neuronal activity before information enters the hippocampal memory pathway. This study determined the integration of synaptic activity evoked by simultaneous neocortical and amygdala electrical stimulation in PER-LEC deep layer principal neurons and parvalbumin (PV) interneurons in mouse brain slices. The data revealed that both deep layer PER-LEC principal neurons and PV interneurons receive synaptic input from the neocortical agranular insular cortex (AiP) and the lateral amygdala (LA). Furthermore, simultaneous stimulation of the AiP and LA never reached the firing threshold in principal neurons of the PER-LEC deep layers. PV interneurons however, mainly showed linear summation of simultaneous AiP and LA inputs and reached their firing threshold earlier. This early PV firing was reflected in the forward shift of the evoked inhibitory conductance in principal neurons, thereby creating a more precise temporal window for coincidence detection, which likely plays a crucial role in information processing.

Excitation-Inhibition Dynamics Regulate Activity Transmission Through the Perirhinal-Entorhinal Network.

The perirhinal (PER) - lateral entorhinal (LEC) network plays a pivotal role in the information transfer between the neocortex and the hippocampus. Anatomical studies have shown that the connectivity is organized bi-directionally: the superficial layers consist of projections running from the neocortex via the PER-LEC network to the hippocampus while the deep layers form the output pathway back to the neocortex. Although these pathways are characterized anatomically, the functional organization of the superficial and deep connections in the PER-LEC network remains to be revealed. We performed paired recordings of superficial and deep layer principal neurons and found that a larger population of superficial neurons responded with action potential firing in response to superficial cortical input, compared to the deep layer population. This suggested that the superficial network can carry information from the cortex towards the hippocampus. The relation between the excitatory and inhibitory input onto the deep and superficial principal neurons showed that the window of net excitability was larger in superficial principal neurons. We performed paired recordings in superficial layer principal neurons and parvalbumin (PV) expressing interneurons to address how this window of opportunity for spiking is affected in superficial principal neurons. The PV interneuron population initiated inhibition at a very consistent timing with increasing stimulus intensity, whereas the excitation temporally shifted to ensure action potential firing. These data indicate that superficial principal neurons can transmit cortical synaptic input through the PER-LEC network because these neurons have a favorable window of opportunity for spiking in contrast to deep neurons.

Parvalbumin interneuron mediated feedforward inhibition controls signal output in the deep layers of the perirhinal-entorhinal cortex.

The perirhinal (PER) and lateral entorhinal (LEC) cortex form an anatomical link between the neocortex and the hippocampus. However, neocortical activity is transmitted through the PER and LEC to the hippocampus with a low probability, suggesting the involvement of the inhibitory network. This study explored the role of interneuron mediated inhibition, activated by electrical stimulation in the agranular insular cortex (AiP), in the deep layers of the PER and LEC. Activated synaptic input by AiP stimulation rarely evoked action potentials in the PER-LEC deep layer excitatory principal neurons, most probably because the evoked synaptic response consisted of a small excitatory and large inhibitory conductance. Furthermore, parvalbumin positive (PV) interneurons-a subset of interneurons projecting onto the axo-somatic region of principal neurons-received synaptic input earlier than principal neurons, suggesting recruitment of feedforward inhibition. This synaptic input in PV interneurons evoked varying trains of action potentials, explaining the fast rising, long lasting synaptic inhibition received by deep layer principal neurons. Altogether, the excitatory input from the AiP onto deep layer principal neurons is overruled by strong feedforward inhibition. PV interneurons, with their fast, extensive stimulus-evoked firing, are able to deliver this fast evoked inhibition in principal neurons. This indicates an essential role for PV interneurons in the gating mechanism of the PER-LEC network.

Distinct Spatiotemporal Activation Patterns of the Perirhinal-Entorhinal Network in Response to Cortical and Amygdala Input.

The perirhinal (PER) and entorhinal cortex (EC) receive input from the agranular insular cortex (AiP) and the subcortical lateral amygdala (LA) and the main output area is the hippocampus. Information transfer through the PER/EC network however, is not always guaranteed. It is hypothesized that this network actively regulates the (sub)cortical activity transfer to the hippocampal network and that the inhibitory system is involved in this function. This study determined the recruitment by the AiP and LA afferents in PER/EC network with the use of voltage sensitive dye (VSD) imaging in horizontal mouse brain slices. Electrical stimulation (500 µA) of the AiP induced activity that gradually propagated predominantly in the rostro-caudal direction: from the PER to the lateral EC (LEC). In the presence of 1 µM of the competitive γ-aminobutyric acid (GABAA) receptor antagonist bicuculline, AiP stimulation recruited the medial EC (MEC) as well. In contrast, LA stimulation (500 µA) only induced activity in the deep layers of the PER. In the presence of bicuculline, the initial population activity in the PER propagated further towards the superficial layers and the EC after a delay. The latency of evoked responses decreased with increasing stimulus intensities (50-500 µA) for both the AiP and LA stimuli. The stimulation threshold for evoking responses in the PER/EC network was higher for the LA than for the AiP. This study showed that the extent of the PER/EC network activation depends on release of inhibition. When GABAA dependent inhibition is reduced, both the AiP and the LA activate spatially overlapping regions, although in a distinct spatiotemporal fashion. It is therefore hypothesized that the inhibitory network regulates excitatory activity from both cortical and subcortical areas that has to be transmitted through the PER/EC network.

Carbamazepine modulates the spatiotemporal activity in the dentate gyrus of rats and pharmacoresistant humans in vitro.

INTRODUCTION: Human hippocampal tissue resected from pharmacoresistant epilepsy patients was investigated to study the effect of the antiepileptic drug CBZ (carbamazepine) and was compared to similar experiments in the hippocampus of control rats. METHODS: The molecular layer of the DG (dentate gyrus) of human epileptic tissue and rat nonepileptic tissue was electrically stimulated and the evoked responses were recorded with voltage-sensitive dye imaging to characterize the spatiotemporal properties. RESULTS: Bath applied CBZ (100 µmol/L) reduced the amplitude of the evoked responses in the human DG, albeit that no clear use-dependent effects were found at frequencies of 8 or 16 Hz. In nonepileptic control DG from rats, CBZ also reduced the amplitude of the evoked response in the molecular layer of the DG as well as the spatial extent of the response. CONCLUSIONS: This study demonstrates that CBZ still reduced the activity in the DG, although the patients were clinically diagnosed as pharmacoresistant for CBZ. This suggests that in the human epileptic brain, the targets of CBZ, the voltage-gated Na(+) channels, are still sensitive to CBZ, although we used a relative high concentration and it is not possibility to assess the actual CBZ concentration that reached the target in the patient. We also concluded that the effect of CBZ was found in the activated region of the DG, quite comparable to the observations in the nonepileptic rat.

Delay and impairment in brain development and function in rat offspring after maternal exposure to methylmercury.

Maternal exposure to the neurotoxin methylmercury (MeHg) has been shown to have adverse effects on neural development of the offspring in man. Little is known about the underlying mechanisms by which MeHg affects the developing brain. To explore the neurodevelopmental defects and the underlying mechanism associated with MeHg exposure, the cerebellum and cerebrum of Wistar rat pups were analyzed by [(18)F]FDG PET functional imaging, field potential analysis, and microarray gene expression profiling. Female rat pups were exposed to MeHg via maternal diet during intrauterinal and lactational period (from gestational day 6 to postnatal day (PND)10), and their brain tissues were sampled for the analysis at weaning (PND18-21) and adulthood (PND61-70). The [(18)F]FDG PET imaging and field potential analysis suggested a delay in brain activity and impaired neural function by MeHg. Genome-wide transcriptome analysis substantiated these findings by showing (1) a delay in the onset of gene expression related to neural development, and (2) alterations in pathways related to both structural and functional aspects of nervous system development. The latter included changes in gene expression of developmental regulators, developmental phase-associated genes, small GTPase signaling molecules, and representatives of all processes required for synaptic transmission. These findings were observed at dose levels at which only marginal changes in conventional developmental toxicity endpoints were detected. Therefore, the approaches applied in this study are promising in terms of yielding increased sensitivity compared with classical developmental toxicity tests.

Efficacy of a new charge-balanced biphasic electrical stimulus in the isolated sciatic nerve and the hippocampal slice.

Most deep brain stimulators apply rectangular monophasic voltage pulses. By modifying the stimulus shape, it is possible to optimize stimulus efficacy and find the best compromise between clinical effect, minimal side effects and power consumption of the stimulus generator. In this study, we compared the efficacy of three types of charge-balanced biphasic pulses (CBBPs, nominal duration 100 µs) in isolated sciatic nerves and in in vitro hippocampal brain slices of the rat. Using these two models, we tested the efficacy of several stimulus shapes exclusively on axons (in the sciatic nerve) and compared the effect with that of stimuli in the more complex neuronal network of the hippocampal slice by considering the stimulus-response relation. We showed that (i) adding an interphase gap (IPG, range 100-500 µs) to the CBBP enhances stimulus efficacy in the rat sciatic nerve and (ii) that this type of stimuli (CBBP with IPG) is also more effective in hippocampal slices. This benefit was similar for both models of voltage and current stimulation. In our two models, asymmetric CBBPs were less beneficial. Therefore, CBBPs with IPG appear to be well suited for application to DBS, since they enhance efficacy, extend battery life and potentially reduce harmful side effects.

The retrosplenial cortex: intrinsic connectivity and connections with the (para)hippocampal region in the rat. An interactive connectome.

A connectome is an indispensable tool for brain researchers, since it quickly provides comprehensive knowledge of the brain's anatomical connections. Such knowledge lies at the basis of understanding network functions. Our first comprehensive and interactive account of brain connections comprised the rat hippocampal-parahippocampal network. We have now added all anatomical connections with the retrosplenial cortex (RSC) as well as the intrinsic connections of this region, because of the interesting functional overlap between these brain regions. The RSC is involved in a variety of cognitive tasks including memory, navigation, and prospective thinking, yet the exact role of the RSC and the functional differences between its subdivisions remain elusive. The connectome presented here may help to define this role by providing an unprecedented interactive and searchable overview of all connections within and between the rat RSC, parahippocampal region and hippocampal formation.

Spatiotemporal analyses of interactions between entorhinal and CA1 projections to the subiculum in rat brain slices.

The subiculum and the entorhinal cortex (EC) are important structures in processing and transmitting information between the neocortex and the hippocampus. The subiculum potentially receives information from the EC through two routes. In addition to a direct projection from EC to the subiculum, there is an indirect polysynaptic connection. The latter uses a number of possible pathways, which all converge onto the final projection from the hippocampal field CA1 to the subiculum. In this series of experiments we investigated to what extent activity in both pathways influences population activity of subicular neurons. We used voltage sensitive dyes in combined hippocampal-EC slices of the rat to measure the spatio-temporal activity patterns. To activate the two inputs to the subiculum, stimulation electrodes were placed in the stratum oriens/alveus of CA1 and in layer III of the medial EC. The response patterns evoked in the subiculum after electrical stimulation of each of these input pathways separately were compared with the response patterns after simultaneous stimulation of both areas (medial EC + CA1). A comparison of the computed added responses of the two individual stimulations with the measured responses after simultaneous stimulation suggests that both inputs are linearly added in the subiculum with very little nonlinear interactions. This strongly suggests that in the subiculum interaction at a single cell level of the direct and the indirect pathways from the EC is an unlikely scenario.

Acceleration of cisplatin ototoxicity by perilymphatic application of 4-methylthiobenzoic acid.

The antitumor agent cisplatin has dose-limiting side effects such as ototoxicity. Systemical co-treatment with anti-oxidants like 4-methylthiobenzoic acid (MTBA) and sodium thiosulfate (STS) provides protection against cisplatin ototoxicity. However, systemically administered protective agents may reduce the chemotherapeutic effect of cisplatin. Local application of the protective agents could avoid this undesirable effect. In the present study, we aimed at suppressing cisplatin-induced ototoxicity in guinea pigs by administering MTBA or STS perilymphatically through cochlear perfusion. Guinea pig cochleas were perfused for 10 min with artificial perilymph (ArtP) containing cisplatin at 0.3 mg/ml, either alone, or in combination with MTBA (0.1 or 1.0 mg/ml) or STS (0.75 or 3.0 mg/ml). The compound action potential (CAP) and the summating potential (SP), evoked by 8 kHz tone bursts, and the endocochlear potential (EP; MTBA only) were measured just before and 1, 2, 3 and 4 h after perfusion. Cisplatin gradually reduced the CAP amplitude in time. Adding MTBA only accelerated this ototoxic effect. After cisplatin treatment a decline was found in the EP, irrespective of co-treatment, i.e., addition of MTBA did not accelerate the EP decrease. In contrast to MTBA, STS ameliorated the ototoxic effect of cisplatin. In conclusion, local application of anti-oxidants can ameliorate cisplatin ototoxicity but this is not a feature of all anti-oxidants.

Differential susceptibility of rats and guinea pigs to the ototoxic effects of ethyl benzene.

The present study was designed to compare the ototoxic effects of volatile ethyl benzene in guinea pigs and rats. Rats showed deteriorated auditory thresholds in the mid-frequency range, based on electrocochleography, after 550-ppm ethyl benzene (8 h/day, 5 days). Outer hair cell (OHC) loss was found in the corresponding cochlear regions. In contrast, guinea pigs showed no threshold shifts and no OHC loss after exposure to much higher ethyl benzene levels (2500 ppm, 6 h/day, 5 days). Subsequently, a limited study (four rats and four guinea pigs) was performed in an attempt to understand these differences in susceptibility. Ethyl benzene concentration in blood was determined in both species after exposure to 500-ppm ethyl benzene (8 h/day, 3 days). At the end of the first day, blood of the rats contained 23.2+/-0.8-microg/ml ethyl benzene, whereas the concentration in guinea pig blood was 2.8+/-0.1 microg/ml. After 3 days, the concentration in both species decreased with respect to the first day, but the ethyl benzene concentration in rat blood was still 4.3 times higher than that in guinea pig blood. Thus, the difference in susceptibility between the species may be related to the ethyl benzene concentration in blood.

Noise-induced hearing loss in rats.

The effect of noise exposure on the auditory system is well known from animal studies. However, most of the studies concern short-term exposure conditions. The purpose of the present research was to find the dose-effect curve for hearing loss in rats following 5 days of noise exposure. Three groups of eight Wag/Rij rats were exposed to broad band noise at levels of 90, 100 and 110 dB SPL for 8 hours/day and 5 consecutive days. An additional group of eight rats served as the control group. Between three and seven weeks after the exposure, hearing was tested by electrocochleography (CAP) and distortion product otoacoustic emissions (DPOAE). Subsequently, the cochleas were morphologically examined. Only the highest two exposure levels affected hearing. The DPOAE growth curves at 4, 8 and 16 kHz and the CAP growth curves at 4, 8, 12, 16 and 24 kHz were affected after the 110 dB SPL broad band noise. After the 100 dB SPL noise, only the 12 kHz CAP growth curve was affected. At the light-microscopic level, OHC damage was not detected in this study.