Controlling neural network responsiveness: tradeoffs and constraints

Engineered Biological Neural Networks on High Density CMOS Microelectrode Arrays

In bottom-up neuroscience, questions on neural information processing are addressed by engineering small but reproducible biological neural networks of defined network topology in vitro. The network topology can be controlled by culturing neurons within polydimethylsiloxane (PDMS) microstructures that are combined with microelectrode arrays (MEAs) for electric access to the network. However, currently used glass MEAs are limited to 256 electrodes and pose a limitation to the spatial resolution as well as the design of more complex microstructures. The use of high density complementary metal-oxide-semiconductor (CMOS) MEAs greatly increases the spatial resolution, enabling sub-cellular readout and stimulation of neurons in defined neural networks. Unfortunately, the non-planar surface of CMOS MEAs complicates the attachment of PDMS microstructures. To overcome the problem of axons escaping the microstructures through the ridges of the CMOS MEA, we stamp-transferred a thin film of hexane-diluted PDMS onto the array such that the PDMS filled the ridges at the contact surface of the microstructures without clogging the axon guidance channels. This method resulted in 23 % of structurally fully connected but sealed networks on the CMOS MEA of which about 45 % showed spiking activity in all channels. Moreover, we provide an impedance-based method to visualize the exact location of the microstructures on the MEA and show that our method can confine axonal growth within the PDMS microstructures. Finally, the high spatial resolution of the CMOS MEA enabled us to show that action potentials follow the unidirectional topology of our circular multi-node microstructure.

Plasticity and Adaptation in Neuromorphic Biohybrid Systems

Neuromorphic systems take inspiration from the principles of biological information processing to form hardware platforms that enable the large-scale implementation of neural networks. The recent years have seen both advances in the theoretical aspects of spiking neural networks for their use in classification and control tasks and a progress in electrophysiological methods that is pushing the frontiers of intelligent neural interfacing and signal processing technologies. At the forefront of these new technologies, artificial and biological neural networks are tightly coupled, offering a novel "biohybrid" experimental framework for engineers and neurophysiologists. Indeed, biohybrid systems can constitute a new class of neuroprostheses opening important perspectives in the treatment of neurological disorders. Moreover, the use of biologically plausible learning rules allows forming an overall fault-tolerant system of co-developing subsystems. To identify opportunities and challenges in neuromorphic biohybrid systems, we discuss the field from the perspectives of neurobiology, computational neuroscience, and neuromorphic engineering.

A Biohybrid Setup for Coupling Biological and Neuromorphic Neural Networks

Developing technologies for coupling neural activity and artificial neural components, is key for advancing neural interfaces and neuroprosthetics. We present a biohybrid experimental setting, where the activity of a biological neural network is coupled to a biomimetic hardware network. The implementation of the hardware network (denoted NeuroSoC) exhibits complex dynamics with a multiplicity of time-scales, emulating 2880 neurons and 12.7 M synapses, designed on a VLSI chip. This network is coupled to a neural network in vitro, where the activities of both the biological and the hardware networks can be recorded, processed, and integrated bidirectionally in real-time. This experimental setup enables an adjustable and well-monitored coupling, while providing access to key functional features of neural networks. We demonstrate the feasibility to functionally couple the two networks and to implement control circuits to modify the biohybrid activity. Overall, we provide an experimental model for neuromorphic-neural interfaces, hopefully to advance the capability to interface with neural activity, and with its irregularities in pathology.

Control of cortical oscillatory frequency by a closed-loop system

BACKGROUND

We present a closed-loop system able to control the frequency of slow oscillations (SO) spontaneously generated by the cortical network in vitro. The frequency of SO can be controlled by direct current (DC) electric fields within a certain range. Here we set out to design a system that would be able to autonomously bring the emergent oscillatory activity to a target frequency determined by the experimenter.

METHODS

The cortical activity was recorded through an electrode and was analyzed online. Once a target frequency was set, the frequency of the slow oscillation was steered through the injection of DC of variable intensity that generated electric fields of proportional amplitudes in the brain slice. To achieve such closed-loop control, we designed a custom programmable stimulator ensuring low noise and accurate tuning over low current levels. For data recording and analysis, we relied on commercial acquisition and software tools.

RESULTS

The result is a flexible and reliable system that ensures control over SO frequency in vitro. The system guarantees artifact removal, minimal gaps in data acquisition and robustness in spite of slice heterogeneity.

CONCLUSIONS

Our tool opens new possibilities for the investigation of dynamics of cortical slow oscillations-an activity pattern that is associated with cognitive processes such as memory consolidation, and that is altered in several neurological conditions-and also for potential applications of this technology.

Autonomous Optimization of Targeted Stimulation of Neuronal Networks

Driven by clinical needs and progress in neurotechnology, targeted interaction with neuronal networks is of increasing importance. Yet, the dynamics of interaction between intrinsic ongoing activity in neuronal networks and their response to stimulation is unknown. Nonetheless, electrical stimulation of the brain is increasingly explored as a therapeutic strategy and as a means to artificially inject information into neural circuits. Strategies using regular or event-triggered fixed stimuli discount the influence of ongoing neuronal activity on the stimulation outcome and are therefore not optimal to induce specific responses reliably. Yet, without suitable mechanistic models, it is hardly possible to optimize such interactions, in particular when desired response features are network-dependent and are initially unknown. In this proof-of-principle study, we present an experimental paradigm using reinforcement-learning (RL) to optimize stimulus settings autonomously and evaluate the learned control strategy using phenomenological models. We asked how to (1) capture the interaction of ongoing network activity, electrical stimulation and evoked responses in a quantifiable ‘state’ to formulate a well-posed control problem, (2) find the optimal state for stimulation, and (3) evaluate the quality of the solution found. Electrical stimulation of generic neuronal networks grown from rat cortical tissue in vitro evoked bursts of action potentials (responses). We show that the dynamic interplay of their magnitudes and the probability to be intercepted by spontaneous events defines a trade-off scenario with a network-specific unique optimal latency maximizing stimulus efficacy. An RL controller was set to find this optimum autonomously. Across networks, stimulation efficacy increased in 90% of the sessions after learning and learned latencies strongly agreed with those predicted from open-loop experiments. Our results show that autonomous techniques can exploit quantitative relationships underlying activity-response interaction in biological neuronal networks to choose optimal actions. Simple phenomenological models can be useful to validate the quality of the resulting controllers.

Electrical stimulation of the brain is increasingly used to alleviate the symptoms of a range of neurological disorders and as a means to artificially inject information into neural circuits in neuroprosthetic applications. Machine learning has been proposed to find optimal stimulation settings autonomously. However, this approach is impeded by the complexity of the interaction between the stimulus and the activity of the network, which makes it difficult to test how good the result actually is. We used phenomenological models of the interaction between stimulus and spontaneous activity in a neuronal network to design a testable machine learning challenge and evaluate the quality of the solution found by the algorithm. In this task, the learning algorithm had to find a solution that balances competing interdependencies of ongoing neuronal activity with opposing effects on the efficacy of stimulation. We show that machine learning can successfully solve this task and that the solutions found are close to the optimal settings to maximize the efficacy of stimulation. Since the paradigm involves several typical problems found in other settings, such concepts could help to formalize machine learning problems in more complex biological networks and to test the quality of their performance.

A Review of Control Strategies in Closed-Loop Neuroprosthetic Systems

It has been widely recognized that closed-loop neuroprosthetic systems achieve more favorable outcomes for users then equivalent open-loop devices. Improved performance of tasks, better usability, and greater embodiment have all been reported in systems utilizing some form of feedback. However, the interdisciplinary work on neuroprosthetic systems can lead to miscommunication due to similarities in well-established nomenclature in different fields. Here we present a review of control strategies in existing experimental, investigational and clinical neuroprosthetic systems in order to establish a baseline and promote a common understanding of different feedback modes and closed-loop controllers. The first section provides a brief discussion of feedback control and control theory. The second section reviews the control strategies of recent Brain Machine Interfaces, neuromodulatory implants, neuroprosthetic systems, and assistive neurorobotic devices. The final section examines the different approaches to feedback in current neuroprosthetic and neurorobotic systems.

Closed-Loop Neuroscience and Non-Invasive Brain Stimulation: A Tale of Two Loops

Closed-loop neuroscience is receiving increasing attention with recent technological advances that enable complex feedback loops to be implemented with millisecond resolution on commodity hardware. We summarize emerging conceptual and methodological frameworks that are available to experimenters investigating a “brain in the loop” using non-invasive brain stimulation and briefly review the experimental and therapeutic implications. We take the view that closed-loop neuroscience in fact deals with two conceptually quite different loops: a “brain-state dynamics” loop, used to couple with and modulate the trajectory of neuronal activity patterns, and a “task dynamics” loop, that is the bidirectional motor-sensory interaction between brain and (simulated) environment, and which enables goal-directed behavioral tasks to be incorporated. Both loops need to be considered and combined to realize the full experimental and therapeutic potential of closed-loop neuroscience.

Recovery of Dynamics and Function in Spiking Neural Networks with Closed-Loop Control

There is a growing interest in developing novel brain stimulation methods to control disease-related aberrant neural activity and to address basic neuroscience questions. Conventional methods for manipulating brain activity rely on open-loop approaches that usually lead to excessive stimulation and, crucially, do not restore the original computations performed by the network. Thus, they are often accompanied by undesired side-effects. Here, we introduce delayed feedback control (DFC), a conceptually simple but effective method, to control pathological oscillations in spiking neural networks (SNNs). Using mathematical analysis and numerical simulations we show that DFC can restore a wide range of aberrant network dynamics either by suppressing or enhancing synchronous irregular activity. Importantly, DFC, besides steering the system back to a healthy state, also recovers the computations performed by the underlying network. Finally, using our theory we identify the role of single neuron and synapse properties in determining the stability of the closed-loop system.

Brain stimulation is being used to ease symptoms in several neurological disorders in cases where pharmacological treatment is not effective (anymore). The most common way for stimulation so far has been to apply a fixed, predetermined stimulus irrespective of the actual state of the brain or the condition of the patient. Recently, alternative strategies such as event-triggered stimulation protocols have attracted the interest of researchers. In these protocols the state of the affected brain area is continuously monitored, but the stimulus is only applied if certain criteria are met. Here we go one step further and present a truly closed-loop stimulation protocol. That is, a stimulus is being continuously provided and the magnitude of the stimulus depends, at any point in time, on the ongoing neural activity dynamics of the affected brain area. This results not only in suppression of the pathological activity, but also in a partial recovery of the transfer function of the activity dynamics. Thus, the ability of the lesioned brain area to carry out relevant computations is restored up to a point as well.

Interaction of electrically evoked activity with intrinsic dynamics of cultured cortical networks with and without functional fast GABAergic synaptic transmission

The modulation of neuronal activity by means of electrical stimulation is a successful therapeutic approach for patients suffering from a variety of central nervous system disorders. Prototypic networks formed by cultured cortical neurons represent an important model system to gain general insights in the input–output relationships of neuronal tissue. These networks undergo a multitude of developmental changes during their maturation, such as the excitatory–inhibitory shift of the neurotransmitter GABA. Very few studies have addressed how the output properties to a given stimulus change with ongoing development. Here, we investigate input–output relationships of cultured cortical networks by probing cultures with and without functional GABA_Aergic synaptic transmission with a set of stimulation paradigms at various stages of maturation. On the cellular level, low stimulation rates (<15 Hz) led to reliable neuronal responses; higher rates were increasingly ineffective. Similarly, on the network level, lowest stimulation rates (<0.1 Hz) lead to maximal output rates at all ages, indicating a network wide refractory period after each stimulus. In cultures aged 3 weeks and older, a gradual recovery of the network excitability within tens of milliseconds was in contrast to an abrupt recovery after about 5 s in cultures with absent GABA_Aergic synaptic transmission. In these GABA deficient cultures evoked responses were prolonged and had multiple discharges. Furthermore, the network excitability changed periodically, with a very slow spontaneous change of the overall network activity in the minute range, which was not observed in cultures with absent GABA_Aergic synaptic transmission. The electrically evoked activity of cultured cortical networks, therefore, is governed by at least two potentially interacting mechanisms: A refractory period in the order of a few seconds and a very slow GABA dependent oscillation of the network excitability.

In vitro studies of neuronal networks and synaptic plasticity in invertebrates and in mammals using multielectrode arrays

Brain functions are strictly dependent on neural connections formed during development and modified during life. The cellular and molecular mechanisms underlying synaptogenesis and plastic changes involved in learning and memory have been analyzed in detail in simple animals such as invertebrates and in circuits of mammalian brains mainly by intracellular recordings of neuronal activity. In the last decades, the evolution of techniques such as microelectrode arrays (MEAs) that allow simultaneous, long-lasting, noninvasive, extracellular recordings from a large number of neurons has proven very useful to study long-term processes in neuronal networks in vivo and in vitro. In this work, we start off by briefly reviewing the microelectrode array technology and the optimization of the coupling between neurons and microtransducers to detect subthreshold synaptic signals. Then, we report MEA studies of circuit formation and activity in invertebrate models such as Lymnaea, Aplysia, and Helix. In the following sections, we analyze plasticity and connectivity in cultures of mammalian dissociated neurons, focusing on spontaneous activity and electrical stimulation. We conclude by discussing plasticity in closed-loop experiments.