I completed my thesis in the Brain and Behaviour lab with Dr. Aldo Faisal.
Our nervous system, like most information processing systems, faces 4 fundamental physical constraints. It has to transmit information quickly (1. time) and reliably (2. noise) while keeping its energy consumption (3. energy) and size/weight (4. spatial scale) at a minimum, to support the behaviourally determined fitness of the organism. These constraints are likely to enforce trade-offs to be made in the evolution of the nervous system. Taking this view, I investigated myelinated and unmyelinated axons across systems and species. I developed user interfaces and simulation methods for the Modigliani stochastic simulation software, and made a number of findings highlighted in the following.
Myelinated axons make up the majority of long-range connections in CNS and PNS. I derived size-dependent relationships for metabolic costs of action potentials in myelinated axons. The high density of sodium channels at Nodes of Ranvier set lower limits on myelinated axons’ outer diameter (0.3 μm), which is 3-fold that for unmyelinated axons.
In contrast, thin, unmyelinated axons make up most of the local (cortical) connectivity. Using a variety of axon models and detailed models of synaptic calcium dynamics and vesicle release, I showed that the waveform of action potentials undergoes random changes whilst traveling along thin unmyelinated axons. These fluctuations translate into synaptic response variability. The diameter of unmyelinated axons sets energetic limits to signalling, and I derived diameter-dependent relationships for the maximum rate of burst and sustained firing. The latter depends on the density of pumps and metabolic cost of action potentials, but is counter intuitively independent of axon diameter.
My findings provide insights and scaling-relationships that relate the 4 fundamental design constraints for wiring brains. They allow us to quantitatively predict structure-function relationships, and form a basis for principled treatments of nerve disorders.
Post-synaptic potential (PSP) variability is typically attributed to mechanisms inside synapses, yet recent advances in experimental methods and biophysical understanding have led us to reconsider the role of axons as highly reliable transmission channels. We show that in many thin axons of our brain, the action potential (AP) waveform and thus the Ca++ signal controlling vesicle release at synapses will be significantly affected by the inherent variability of ion channel gating. We investigate how and to what extent fluctuations in the AP waveform explain observed PSP variability. Using both biophysical theory and stochastic simulations of central and peripheral nervous system axons from vertebrates and invertebrates, we show that channel noise in thin axons (<1 µm diameter) causes random fluctuations in AP waveforms. AP height and width, both experimentally characterised parameters of post-synaptic response amplitude, vary e.g. by up to 20 mV and 0.5 ms while a single AP propagates in C-fibre axons. We show how AP height and width variabilities increase with a ¾ power-law as diameter decreases and translate these fluctuations into post-synaptic response variability using biophysical data and models of synaptic transmission. We find for example that for mammalian unmyelinated axons with 0.2 µm diameter (matching cerebellar parallel fibres) axonal noise alone can explain half of the PSP variability in cerebellar synapses. We conclude that axonal variability may have considerable impact on synaptic response variability. Thus, in many experimental frameworks investigating synaptic transmission through paired-cell recordings or extracellular stimulation of presynaptic neurons, causes of variability may have been confounded. We thereby show how bottom-up aggregation of molecular noise sources contributes to our understanding of variability observed at higher levels of biological organisation.
The action potential (AP), the fundamental signal of the nervous system, is carried by two types of axons: unmyelinated and myelinated fibers. In the former the action potential propagates continuously along the axon as established in large-diameter fibers. In the latter axons the AP jumps along the nodes of Ranvier—discrete, anatomically specialized regions which contain very high densities of sodium ion (Na+) channels. Therefore, saltatory conduction is thought as the hallmark of myelinated axons, which enables faster and more reliable propagation of signals than in unmyelinated axons of same outer diameter. Recent molecular anatomy showed that in C-fibers, the very thin (0.1 µm diameter) axons of the peripheral nervous system, Nav1.8 channels are clustered together on lipid rafts that float in the cell membrane. This localized concentration of Na+ channels resembles in structure the ion channel organization at the nodes of Ranvier, yet it is currently unknown whether this translates into an equivalent phenomenon of saltatory conduction or related-functional benefits and efficiencies. Therefore, we modeled biophysically realistic unmyelinated axons with both conventional and lipid-raft based organization of Na+ channels. We find that APs are reliably conducted in a micro-saltatory fashion along lipid rafts. Comparing APs in unmyelinated fibers with and without lipid rafts did not reveal any significant difference in either the metabolic cost or AP propagation velocity. By investigating the efficiency of AP propagation over Nav1.8 channels, we find however that the specific inactivation properties of these channels significantly increase the metabolic cost of signaling in C-fibers.
Body-worn sensors have contributed to a rich and growing body of literature in public health and clinical research in the last decades. A major challenge in sensor research is the lack of consistency and standardization of the collection and reporting of the sensor data. The algorithms used to derive these activity counts can be vastly different between manufactures and not always transparent to the researchers. With Philips, one of the major research-grade wearable device manufacturers, discontinuing this product line, many researchers are left in need of alternative solutions and at the risk of not being able to relate their historical data using the Philips Actiwatch 2 devices to future findings with other devices. We herein provide a comparison analysis and conversion method that can be used to convert activity counts from Philips to those from ActiGraph, another major manufacturer who provide both raw acceleration data and count data based on their open-source algorithm to the research community. This work provides an approach to maximize the scientific value of historical actigraphy data collected by the Actiwatch devices to support research continuity in this community. The conversion, however, is not perfect and only offers an approximation, due to the intrinsic difference in the count algorithms between the two accelerometers, and the permanent information loss during data reduction. We encourage future research using body-worn sensors to retain the raw sensor data to ensure data consistency, comparability, and the ability to leverage future algorithm improvement.
Digital clinical measures based on data collected by wearable devices have seen rapid growth in both clinical trials and healthcare. The widely-used measures based on wearables are epoch-based physical activity counts using accelerometer data. Even though activity counts have been the backbone of thousands of clinical and epidemiological studies, there are large variations of the algorithms that compute counts and their associated parameters—many of which have often been kept proprietary by device providers. This lack of transparency has hindered comparability between studies using different devices and limited their broader clinical applicability. ActiGraph devices have been the most-used wearable accelerometer devices for over two decades. Recognizing the importance of data transparency, interpretability and interoperability to both research and clinical use, we here describe the detailed counts algorithms of five generations of ActiGraph devices going back to the first AM7164 model, and publish the current counts algorithm in ActiGraph’s ActiLife and CentrePoint software as a standalone Python package for research use. We believe that this material will provide a useful resource for the research community, accelerate digital health science and facilitate clinical applications of wearable accelerometry.
Activity counts have been used for over two decades with over 22,000 published scientific papers in public health and clinical research. ActiGraph recently released the algorithm for computing counts from raw accelerometer data as an open-source Python library, which is now ported by researchers to other languages, notably R. The current commentary presents historical overview of ActiGraph counts, and its development and evolution as a measure of physical activity. Further, we provide general recommendations on extracting counts from raw accelerometer data and discuss specific considerations with respect to device types, resampling, nonwear, axes orientations, and epoch length that may influence counts. Last, we provide a tutorial on how to use ActiGraph’s open-source Python library, agcounts, for consistent, accurate, and reproducible count. We expect this commentary will provide familiarity and transparency needed to adopt and produce activity counts in a consistent manner, allowing researchers to conduct statistical comparisons across multiple data sets and studies.
Wearable accelerometers are widely used as an ecologically valid and scalable solution for long-term at-home sleep monitoring in both clinical research and care. In this study, we applied a deep learning domain adversarial convolutional neural network (DACNN) model to this task and demonstrated that this new model outperformed existing sleep algorithms in classifying sleep–wake and estimating sleep outcomes based on wrist-worn accelerometry. This model generalized well to another dataset based on different wearable devices and activity counts, achieving an accuracy of 80.1% (sensitivity 84% and specificity 58%). Compared to commonly used sleep algorithms, this model resulted in the smallest error in wake after sleep onset (MAE of 48.7, Cole–Kripke of 86.2, Sadeh of 108.2, z-angle of 57.5) and sleep efficiency (MAE of 11.8, Cole–Kripke of 18.4, Sadeh of 23.3, z-angle of 9.3) outcomes. Despite being around for many years, accelerometer-alone devices continue to be useful due to their low cost, long battery life, and ease of use. Improving the accuracy and generalizability of sleep algorithms for accelerometer wrist devices is of utmost importance. We here demonstrated that domain adversarial convolutional neural networks can improve the overall accuracy, especially the specificity, of sleep–wake classification using wrist-worn accelerometer data, substantiating its use as a scalable and valid approach for sleep outcome assessment in real life.
Abstract For the last 40 years, actigraphy or wearable accelerometry has provided an objective, low-burden and ecologically valid approach to assess real-world sleep and circadian patterns, contributing valuable data to epidemiological and clinical insights on sleep and sleep disorders. The proper use of wearable technology in sleep research requires validated algorithms that can derive sleep outcomes from the sensor data. Since the publication of the first automated scoring algorithm by Webster in 1982, a variety of sleep algorithms have been developed and contributed to sleep research, including many recent ones that leverage machine learning and / or deep learning approaches. However, it remains unclear how these algorithms compare to each other on the same data set and if these modern data science approaches improve the analytical validity of sleep outcomes based on wrist-worn acceleration data. This work provides a systematic evaluation across 8 state-of-the-art sleep algorithms on a common sleep data set with polysomnography (PSG) as ground truth. Despite the inclusion of recently published complex algorithms, simple regression-based and heuristic algorithms demonstrated slightly superior performance in sleep-wake classification and sleep outcome estimation. The performance of complex machine learning and deep learning models seem to suffer from poor generalization. This independent and systematic analytical validation of sleep algorithms provides key evidence on the use of wearable digital health technologies for sleep research and care.
Research Objectives To investigate the usefulness of gait-related spectral features derived from accelerometry to assess the risk of falls in older adults (OA). Design Retrospective cohort. Setting Gait laboratory. Participants Publicly available data (Physionet.org) collected from 60 community-living OA categorized into either the Control Group (CG, n=34) or the Faller Group (FG, n=26) based on the self-reported history of falls (fallers: ≥ two falls in the previous year). Interventions N/A. Main Outcome Measures A total of 16 spectral features - mean frequency (fmean), median frequency (fmed), peak power spectral density (PSDp), and dominant frequency (fd) were derived from the tri-axial and the vector magnitude (VM) data recorded from a lower back-worn accelerometer during walking. The clinical outcomes included the Dynamic Gait Index (DGI), the Berg Balance Scale (BBS), and the Timed Up and Go (TUG) test. Results Mann-Whitney U tests showed that there was a significant difference in mean fd (anterior-posterior (AP) and VM (U=610.5, p=0.012), fmed (AP, VM) (U=590, p=0.028) between the groups. No significant group differences were found in BBS (p=0.285), TUG (p=0.109), and DGI (p=0.258). Pearson’s correlation between the spectral features and the clinical outcomes showed that 13 of 16 spectral features for the FG were significantly correlated (p< 0.05) with the BBS, TUG, and DGI. No significant correlations were observed for the CG (p>0.05), except for fd (BBS), fmean (DGI, TUG), and PSDp (TUG). Conclusions The current evidence supports the sensitivity of the accelerometry features to isolate gait characteristics specific to fallers, where the clinical outcomes may suffer from subjectivity, lack of sensitivity, and ceiling effects. For the FG, strong correlations with the clinical outcomes suggest the agreement between the spectral features and the functional status. Author(s) Disclosures None.
The energy necessary for propagating action potentials (APs) in axons is stored in the form of ionic concentration gradients across the membrane. It is commonly assumed that the number of ions crossing the membrane for each AP is very small compared to the total number of ions involved, as this is the case in classically studied axons e.g. the squid giant axon (SGA) with diameters of hundreds of microns. However, the mammalian nervous system contains much thinner axons e.g. C fibres or cortical axon collaterals (d=0.1-0.3 µm). The current due to ionic pumps is much smaller than that of ion channels in an action potential, which means that the potential energy pool may be depleted. We investigate how homeostatic and metabolic constraints limit neuronal activity using a Hodgkin-Huxley type model which tracks changes of ionic concentrations. A rough estimation yields a theoretical minimum of at least 0.3 mV of the potassium reversal potential (EK) in a C-fibre axon of diameter 0.3µm. Our simulations, which take into account the inefficiencies in channel kinetics, yield a higher value (1.2 mV) and show an inverse relationship between the change in reversal potentials and diameter. Based on basic physical considerations, we establish an equation describing the evolution of ionic concentrations and linking the maximum sustainable firing rate to the diameter. Using our model, we can predict the maximum firing rate that our model of the SGA can sustain for a given amount of time e.g. 10 seconds as a function of the axonal diameter. We conclude that in a similar way to noise, energetic considerations constrain the miniaturization of axons and limit the wiring density of neural circuits to energetically sustainable neural codes.
Recent advances in experimental methods have allowed to reconsider the role of axons as faithful transmission channels. Post-synaptic potential (PSP) variability is typically attributed to mechanisms inside the synapse, yet in the many thin axons of our brain, the action potential (AP) waveform and thus the Ca++ signal controlling vesicle release at the synapse will be significantly affected by the inherent stochasticity of ion channels. We investigate to what extent fluctuations in the AP waveform cause the observed PSP variability. We show, using both biophysical theory and stochastic simulations of invertebrate and vertebrate, CNS and PNS axons, that channel noise in axons below 1 µm causes random changes in the AP waveform. AP height and width, both experimentally characterised parameters of post-synaptic response amplitude, vary e.g. by 6 mV and 1.5 ms as a single AP propagates in 0.3 µm diameter axons (cortical axon collaterals). We show how AP height and width variabilities increase with a 3⁄4 power-law as diameter decreases and translate these fluctuations into post-synaptic response variability using biophysical data and models of synaptic transmission. We find that e.g. in synapses innervated by 0.2 µm diameter unmyelinated axons (cerebellar parallel fibres) half of PSP variability can be accounted for by axonal noise alone. Axonal variability may have considerable impact on synaptic response variability. However, it can be easily confounded with synaptic variability in many experimental frameworks investigating synaptic transmission through paired-cell recordings or extracellular stimulation of the presynaptic neuron and intracellular post-synaptic recordings.
The fundamental signal of the nervous system, the action potential, is carried by two types of nerve fibres, myelinated and unmyelinated axons. Myelinated axons feature a highly structured distribution of voltage-gated ion channels, with a characteristic clustering of Na channels at the Nodes of Ranvier. In contrast, unmyelinated axons are generally thought to feature uniformly distributed ion channels. The former type of axon is generally known to allow faster, more reliable and, as we recently showed (Neishabouri & Faisal, 2011, BMC Neurosci), considerably more energy efficient propagation of signals than unmyelinated axons. In contrast, only the latter can reach the physical limits to axon diameter at 0.1 µm, thus making the high connection densities of mammalian cortex possible (Faisal et al., 2005, Curr Biol). In C-fibres, we have recently discovered that NaV1.8, the voltage-gated Na channels of these 0.1 µm diameter unmyelinated axons (Baker, 2005, J.Physiol), are packed tightly together on lipid rafts instead of being uniformly distributed over the membrane, resembling the structure of Nodes of Ranvier in myelinated fibres. Dissolving the lipid rafts in these axons can result in the loss of action potential transmission. We investigated the effects of the lipid-raft clustering of Na channels in terms of function, by looking at the propagation velocity, reliability and metabolic cost of action potentials. Because ion channel noise is significant in these axons we used the Modigliani stochastic simulation framework (Faisal & Laughlin, 2007, PLoS CB). We simulated both uniformly distributed and clustered channels (0.2µm diameter clusters of channels every 3µm of axon) along the unmyelinated fibre (0.1 µm diameter), so average channel density was in both cases 125 per µm\^2. Metabolic cost was defined by the amount of ATP molecules necessary to reverse the Na+ current by Na-K-ATPase (Alle et al., 2009, Science). We found that clustering Na channels on lipid raft reduced metabolic cost by over 260% (2.86×10\^-6 pmol ATP/mm vs. 1.03×10\^-5 pmol ATP/mm), while affecting propagation velocity and reliability far less (~20%). Thus, it appears that lipid rafts act like molecular Nodes of Ranvier in C-fibers, which as we show here is an evolutionary advantageous design feature that may be general to the many thin fibers of the CNS and PNS.