Pruning recurrent neural networks replicates adolescent changes in working memory and reinforcement learning

Navigating Pubertal Goldilocks: The Optimal Pace for Hierarchical Brain Organization

Adolescence is a timed process with an onset, tempo, and duration. Nevertheless, the temporal dimension, especially the pace of maturation, remains an insufficiently studied aspect of developmental progression. The primary objective is to estimate the precise influence of pubertal maturational tempo on the configuration of associative brain regions. To this end, the connection between maturational stages and the level of hierarchical organization of large-scale brain networks in 12-13-year-old females is analyzed. Skeletal maturity is used as a proxy for pubertal progress. The degree of maturity is defined by the difference between bone age and chronological age. To assess the level of hierarchical organization in the brain, the temporal dynamic of closed eye resting state high-density electroencephalography (EEG) in the alpha frequency range is analyzed. Different levels of hierarchical order are captured by the measured asymmetry in the directionality of information flow between different regions. The calculated EEG-based entropy production of participant groups is then compared with accelerated, average, and decelerated maturity. Results indicate that an average maturational trajectory optimally aligns with cerebral hierarchical order, and both accelerated and decelerated timelines result in diminished cortical organization. This suggests that a "Goldilocks rule" of brain development is favoring a particular maturational tempo.

Microbiome depletion prior to repeat mild TBI differentially alters social deficits and prefrontal cortex plasticity in adolescent and adult rats

Although aging, repeat mild traumatic brain injury (RmTBI), and microbiome modifications independently change social behavior, there has been no investigation into their cumulative effects on social behavior and neuroplasticity within the prefrontal cortex. Therefore, we examined how microbiome depletion prior to RmTBI affected social behavior and neuroplasticity in adolescent and adult rats. Play, temperament analysis, elevated plus maze, and the hot/cold plate assessed socio-emotional function. Analyses of perineuronal nets (PNNs) and parvalbumin (PV) interneurons was completed. Social-emotional deficits were more pronounced in adults, with microbiome depletion attenuating social behavior deficits associated with RmTBI in both age groups. Microbiome depletion increased branch length and PNN arborization within the PFC but decreased the overall number of PNNs. Adults and males were more vulnerable to RmTBI. Interestingly, microbiome depletion may have attenuated the changes to neuroplasticity and subsequent social deficits, suggesting that the microbiome is a viable, but age-specific, target for RmTBI therapeutics.

Changes in Behavior and Neural Dynamics across Adolescent Development

Adolescence is an important developmental period, during which substantial changes occur in brain function and behavior. Several aspects of executive function, including response inhibition, improve during this period. Correspondingly, structural imaging studies have documented consistent decreases in cortical and subcortical gray matter volume, and postmortem histologic studies have found substantial (∼40%) decreases in excitatory synapses in prefrontal cortex. Recent computational modeling work suggests that the change in synaptic density underlie improvements in task performance. These models also predict changes in neural dynamics related to the depth of attractor basins, where deeper basins can underlie better task performance. In this study, we analyzed task-related neural dynamics in a large cohort of longitudinally followed subjects (male and female) spanning early to late adolescence. We found that age correlated positively with behavioral performance in the Eriksen Flanker task. Older subjects were also characterized by deeper attractor basins around task related evoked EEG potentials during specific cognitive operations. Thus, consistent with computational models examining the effects of excitatory synaptic pruning, older adolescents showed stronger attractor dynamics during task performance.SIGNIFICANCE STATEMENT There are well-documented changes in brain and behavior during adolescent development. However, there are few mechanistic theories that link changes in the brain to changes in behavior. Here, we tested a hypothesis, put forward on the basis of computational modeling, that pruning of excitatory synapses in cortex during adolescence changes neural dynamics. We found, consistent with the hypothesis, that variability around event-related potentials shows faster decay dynamics in older adolescent subjects. The faster decay dynamics are consistent with the hypothesis that synaptic pruning during adolescent development leads to stronger attractor basins in task-related neural activity.

Shaping dynamical neural computations using spatiotemporal constraints

Dynamics play a critical role in computation. The principled evolution of states over time enables both biological and artificial networks to represent and integrate information to make decisions. In the past few decades, significant multidisciplinary progress has been made in bridging the gap between how we understand biological versus artificial computation, including how insights gained from one can translate to the other. Research has revealed that neurobiology is a key determinant of brain network architecture, which gives rise to spatiotemporally constrained patterns of activity that underlie computation. Here, we discuss how neural systems use dynamics for computation, and claim that the biological constraints that shape brain networks may be leveraged to improve the implementation of artificial neural networks. To formalize this discussion, we consider a natural artificial analog of the brain that has been used extensively to model neural computation: the recurrent neural network (RNN). In both the brain and the RNN, we emphasize the common computational substrate atop which dynamics occur-the connectivity between neurons-and we explore the unique computational advantages offered by biophysical constraints such as resource efficiency, spatial embedding, and neurodevelopment.

Attractor dynamics reflect decision confidence in macaque prefrontal cortex

Decisions are made with different degrees of consistency, and this consistency can be linked to the confidence that the best choice has been made. Theoretical work suggests that attractor dynamics in networks can account for choice consistency, but how this is implemented in the brain remains unclear. Here we provide evidence that the energy landscape around attractor basins in population neural activity in the prefrontal cortex reflects choice consistency. We trained two rhesus monkeys to make accept/reject decisions based on pretrained visual cues that signaled reward offers with different magnitudes and delays to reward. Monkeys made consistent decisions for very good and very bad offers, but decisions were less consistent for intermediate offers. Analysis of neural data showed that the attractor basins around patterns of activity reflecting decisions had steeper landscapes for offers that led to consistent decisions. Therefore, we provide neural evidence that energy landscapes predict decision consistency, which reflects decision confidence.

Attractor dynamics reflect decision confidence in macaque prefrontal cortex

Decisions are made with different degrees of consistency, and this consistency can be linked to the confidence that the best choice has been made. Theoretical work suggests that attractor dynamics in networks can account for choice consistency, but how this is implemented in the brain remains unclear. Here, we provide evidence that the energy landscape around attractor basins in population neural activity in prefrontal cortex reflects choice consistency. We trained two rhesus monkeys to make accept/reject decisions based on pretrained visual cues that signaled reward offers with different magnitudes and delays-to-reward. Monkeys made consistent decisions for very good and very bad offers, but decisions were less consistent for intermediate offers. Analysis of neural data showed that the attractor basins around patterns of activity reflecting decisions had steeper landscapes for offers that led to consistent decisions. Therefore, we provide neural evidence that energy landscapes predict decision consistency, which reflects decision confidence.

Distinctive properties of biological neural networks and recent advances in bottom-up approaches toward a better biologically plausible neural network

Although it may appear infeasible and impractical, building artificial intelligence (AI) using a bottom-up approach based on the understanding of neuroscience is straightforward. The lack of a generalized governing principle for biological neural networks (BNNs) forces us to address this problem by converting piecemeal information on the diverse features of neurons, synapses, and neural circuits into AI. In this review, we described recent attempts to build a biologically plausible neural network by following neuroscientifically similar strategies of neural network optimization or by implanting the outcome of the optimization, such as the properties of single computational units and the characteristics of the network architecture. In addition, we proposed a formalism of the relationship between the set of objectives that neural networks attempt to achieve, and neural network classes categorized by how closely their architectural features resemble those of BNN. This formalism is expected to define the potential roles of top-down and bottom-up approaches for building a biologically plausible neural network and offer a map helping the navigation of the gap between neuroscience and AI engineering.

The neuroconnectionist research programme

Artificial neural networks (ANNs) inspired by biology are beginning to be widely used to model behavioural and neural data, an approach we call 'neuroconnectionism'. ANNs have been not only lauded as the current best models of information processing in the brain but also criticized for failing to account for basic cognitive functions. In this Perspective article, we propose that arguing about the successes and failures of a restricted set of current ANNs is the wrong approach to assess the promise of neuroconnectionism for brain science. Instead, we take inspiration from the philosophy of science, and in particular from Lakatos, who showed that the core of a scientific research programme is often not directly falsifiable but should be assessed by its capacity to generate novel insights. Following this view, we present neuroconnectionism as a general research programme centred around ANNs as a computational language for expressing falsifiable theories about brain computation. We describe the core of the programme, the underlying computational framework and its tools for testing specific neuroscientific hypotheses and deriving novel understanding. Taking a longitudinal view, we review past and present neuroconnectionist projects and their responses to challenges and argue that the research programme is highly progressive, generating new and otherwise unreachable insights into the workings of the brain.

Repetitive mild traumatic brain injury alters central and peripheral clock gene expression in the adolescent rat

Mild traumatic brain injury (mTBI) or concussion is a common injury worldwide leading to substantial medical costs and a high burden on society. In adolescents, falls and sports related trauma are often the causes of mTBI. Importantly, critical brain growth and development occurs during this sensitive period making the prospect of a brain injury a worrying phenomenon. Upwards of 70% of patients report circadian disruption following these injuries and this has been shown to impede recovery. Therefore, we sought to determine if core circadian clock gene expression was disrupted in rat model of repetitive mTBI (RmTBI). Male and female adolescent rats (n = 129) received sham or RmTBI. The animals were then euthanized at different times throughout the day and night. Tissue from the hypothalamus, cerebellum, hippocampus, liver, and small intestine were evaluated for the expression of per1, per2, cry1, clock, bmal1 and rev-erb-α. We found most clock genes varied across the day/night indicating circadian expression patterns. In the hypothalamus we found RmTBI altered the expression of cry1 and bmal1 in addition to sex differences in per2, cry1, clock, bmal1 and rev-erb- α. In the cerebellum, per1, per2, cry1, clock, bmal1 and rev-erb-α rhythms were all knocked out by RmTBI in addition to sex differences in cry1, clock and bmal1 expression. We also detected a significant decrease in overall expression of all clock genes in males in the middle of the night. In the hippocampus we found that RmTBI changed the rhythm of rev-erb-α expression in addition to sex differences in bmal1 expression. In the liver we detected strong rhythms in all genes examined, however only per2 expression was knocked out by RmTBI, in addition we also detected sex differences in per2 and cry1. We also detected an overall decrease in female clock gene expression in the early night. In the small intestine, RmTBI altered cry1 expression and there were sex differences in rev-erb-α. These results indicate that RmTBI alters core circadian clock gene expression in the central and peripheral nervous system in a time, tissue and sex dependent manner. This may be disrupting important phase relationships between the brain and peripheral nervous system and contributing to post-injury symptomology and also highlights the importance for time and sex dependent assessment of injury outcomes.

Strength of Excitatory Inputs to Layer 3 Pyramidal Neurons During Synaptic Pruning in the Monkey Prefrontal Cortex: Relevance for the Pathogenesis of Schizophrenia

BACKGROUND

In schizophrenia, layer 3 pyramidal neurons (L3PNs) of the dorsolateral prefrontal cortex exhibit deficits in markers of excitatory synaptic inputs that are thought to disrupt the patterns of neural network activity essential for cognitive function. These deficits are usually interpreted under Irwin Feinberg's hypothesis of altered synaptic pruning, which postulates that normal periadolescent pruning, thought to preferentially eliminate weak/immature synapses, is altered in schizophrenia. However, it remains unknown whether periadolescent pruning on L3PNs in the primate dorsolateral prefrontal cortex selectively eliminates weak excitatory synapses or uniformly eliminates excitatory synapses across the full distribution of synaptic strengths.

METHODS

To distinguish between these alternative models of synaptic pruning, we assessed the densities of dendritic spines, the site of most excitatory inputs to L3PNs, and the distributions of excitatory synaptic strengths in dorsolateral prefrontal cortex L3PNs from male and female monkeys across the periadolescent period of synaptic pruning. We used patch-clamp methods in acute brain slices to record miniature excitatory synaptic currents and intracellular filling with biocytin to quantify dendritic spines.

RESULTS

On L3PNs, dendritic spines exhibited the expected age-related decline in density, but mean synaptic strength and the shape of synaptic strength distributions remained stable with age.

CONCLUSIONS

The absence of age-related differences in mean synaptic strength and synaptic strength distributions supports the model of a uniform pattern of synaptic pruning across the full range of synaptic strengths. The implications of these findings for the pathogenesis and functional consequences of dendritic spine deficits in schizophrenia are discussed.

Using Nonhuman Primate Models to Reverse-Engineer Prefrontal Circuit Failure Underlying Cognitive Deficits in Schizophrenia

In this chapter, I review studies in nonhuman primates that emulate the circuit failure in prefrontal cortex responsible for working memory and cognitive control deficits in schizophrenia. These studies have characterized how synaptic malfunction, typically induced by blockade of NMDAR, disrupts neural function and computation in prefrontal networks to explain errors in cognitive tasks that are seen in schizophrenia. This work is finding causal relationships between pathogenic events of relevance to schizophrenia at vastly different levels of scale, from synapses, to neurons, local, circuits, distributed networks, computation, and behavior. Pharmacological manipulation, the dominant approach in primate models, has limited construct validity for schizophrenia pathogenesis, as the disease results from a complex interplay between environmental, developmental, and genetic factors. Genetic manipulation replicating schizophrenia risk is more advanced in rodent models. Nonetheless, gene manipulation in nonhuman primates is rapidly advancing, and primate developmental models have been established. Integration of large scale neural recording, genetic manipulation, and computational modeling in nonhuman primates holds considerable potential to provide a crucial schizophrenia model moving forward. Data generated by this approach is likely to fill several crucial gaps in our understanding of the causal sequence leading to schizophrenia in humans. This causal chain presents a vexing problem largely because it requires understanding how events at very different levels of scale relate to one another, from genes to circuits to cognition to social interactions. Nonhuman primate models excel here. They optimally enable discovery of causal relationships across levels of scale in the brain that are relevant to cognitive deficits in schizophrenia. The mechanistic understanding of prefrontal circuit failure they promise to provide may point the way to more effective therapeutic interventions to restore function to prefrontal networks in the disease.

Dendritic Spines: Synaptogenesis and Synaptic Pruning for the Developmental Organization of Brain Circuits

Synaptic overproduction and elimination is a regular developmental event in the mammalian brain. In the cerebral cortex, synaptic overproduction is almost exclusively correlated with glutamatergic synapses located on dendritic spines. Therefore, analysis of changes in spine density on different parts of the dendritic tree in identified classes of principal neurons could provide insight into developmental reorganization of specific microcircuits.The activity-dependent stabilization and selective elimination of the initially overproduced synapses is a major mechanism for generating diversity of neural connections beyond their genetic determination. The largest number of overproduced synapses was found in the monkey and human cerebral cortex. The highest (exceeding adult values by two- to threefold) and most protracted overproduction (up to third decade of life) was described for associative layer IIIC pyramidal neurons in the human dorsolateral prefrontal cortex.Therefore, the highest proportion and extraordinarily extended phase of synaptic spine overproduction is a hallmark of neural circuitry in human higher-order associative areas. This indicates that microcircuits processing the most complex human cognitive functions have the highest level of developmental plasticity. This finding is the backbone for understanding the effect of environmental impact on the development of the most complex, human-specific cognitive and emotional capacities, and on the late onset of human-specific neuropsychiatric disorders, such as autism and schizophrenia.