Transfer learning to leverage larger datasets for improved prediction of protein stability changes

Amino acid mutations that lower a protein's thermodynamic stability are implicated in numerous diseases, and engineered proteins with enhanced stability can be important in research and medicine. Computational methods for predicting how mutations perturb protein stability are, therefore, of great interest. Despite recent advancements in protein design using deep learning, in silico prediction of stability changes has remained challenging, in part due to a lack of large, high-quality training datasets for model development. Here, we describe ThermoMPNN, a deep neural network trained to predict stability changes for protein point mutations given an initial structure. In doing so, we demonstrate the utility of a recently released megascale stability dataset for training a robust stability model. We also employ transfer learning to leverage a second, larger dataset by using learned features extracted from ProteinMPNN, a deep neural network trained to predict a protein's amino acid sequence given its three-dimensional structure. We show that our method achieves state-of-the-art performance on established benchmark datasets using a lightweight model architecture that allows for rapid, scalable predictions. Finally, we make ThermoMPNN readily available as a tool for stability prediction and design.

Axon Initial Segment GABA Inhibits Action Potential Generation throughout Periadolescent Development

Neurons are remarkably polarized structures: dendrites spread and branch to receive synaptic inputs while a single axon extends and transmits action potentials (APs) to downstream targets. Neuronal polarity is maintained by the axon initial segment (AIS), a region between the soma and axon proper that is also the site of action potential (AP) generation. This polarization between dendrites and axons extends to inhibitory neurotransmission. In adulthood, the neurotransmitter GABA hyperpolarizes dendrites but instead depolarizes axons. These differences in function collide at the AIS. Multiple studies have shown that GABAergic signaling in this region can share properties of either the mature axon or mature dendrite, and that these properties evolve over a protracted period encompassing periadolescent development. Here, we explored how developmental changes in GABAergic signaling affect AP initiation. We show that GABA at the axon initial segment inhibits action potential initiation in layer (L)2/3 pyramidal neurons in prefrontal cortex from mice of either sex across GABA reversal potentials observed in periadolescence. These actions occur largely through current shunts generated by GABAA receptors and changes in voltage-gated channel properties that affected the number of channels that could be recruited for AP electrogenesis. These results suggest that GABAergic neurons targeting the axon initial segment provide an inhibitory "veto" across the range of GABA polarity observed in normal adolescent development, regardless of GABAergic synapse reversal potential.Significance Statement GABA receptors are a major class of neurotransmitter receptors in the brain. Typically, GABA receptors inhibit neurons by allowing influx of negatively charged chloride ions into the cell. However, there are cases where local chloride concentrations promote chloride efflux through GABA receptors. Such conditions exist early in development in neocortical pyramidal cell axon initial segments (AISs), where action potentials (APs) initiate. Here, we examined how chloride efflux in early development interacts with mechanisms that support action potential initiation. We find that this efflux, despite moving membrane potential closer to action potential threshold, is nevertheless inhibitory. Thus, GABA at the axon initial segment is likely to be inhibitory for action potential initiation independent of whether chloride flows out or into neurons via these receptors.

Motor control complexity can be dynamically simplified during gait pattern exploration using motor control-based biofeedback

Understanding how the central nervous system coordinates diverse motor outputs has been a topic of extensive investigation. Although it is generally accepted that a small set of synergies underlies many common activities, such as walking, whether synergies are equally robust across a broader array of gait patterns or can be flexibly modified remains unclear. Here, we evaluated the extent to which synergies changed as nondisabled adults (n = 14) explored gait patterns using custom biofeedback. Secondarily, we used Bayesian additive regression trees to identify factors that were associated with synergy modulation. Participants explored 41.1 ± 8.0 gait patterns using biofeedback, during which synergy recruitment changed depending on the type and magnitude of gait pattern modification. Specifically, a consistent set of synergies was recruited to accommodate small deviations from baseline, but additional synergies emerged for larger gait changes. Synergy complexity was similarly modulated; complexity decreased for 82.6% of the attempted gait patterns, but distal gait mechanics were strongly associated with these changes. In particular, greater ankle dorsiflexion moments and knee flexion through stance, as well as greater knee extension moments at initial contact, corresponded to a reduction in synergy complexity. Taken together, these results suggest that the central nervous system preferentially adopts a low-dimensional, largely invariant control strategy but can modify that strategy to produce diverse gait patterns. Beyond improving understanding of how synergies are recruited during gait, study outcomes may also help identify parameters that can be targeted with interventions to alter synergies and improve motor control after neurological injury.NEW & NOTEWORTHY We used a motor control-based biofeedback system and machine learning to characterize the extent to which nondisabled adults can modulate synergies during gait pattern exploration. Results revealed that a small library of synergies underlies an array of gait patterns but that recruitment from this library changes as a function of the imposed biomechanical constraints. Our findings enhance understanding of the neural control of gait and may inform biofeedback strategies to improve synergy recruitment after neurological injury.

Mapping the timescale of suicidal thinking

This study aims to identify the timescale of suicidal thinking, leveraging real-time monitoring data and a number of different analytic approaches. Participants were 105 adults with past week suicidal thoughts who completed a 42-d real-time monitoring study (total number of observations = 20,255). Participants completed two forms of real-time assessments: traditional real-time assessments (spaced hours apart each day) and high-frequency assessments (spaced 10 min apart over 1 h). We found that suicidal thinking changes rapidly. Both descriptive statistics and Markov-switching models indicated that elevated states of suicidal thinking lasted on average 1 to 3 h. Individuals exhibited heterogeneity in how often and for how long they reported elevated suicidal thinking, and our analyses suggest that different aspects of suicidal thinking operated on different timescales. Continuous-time autoregressive models suggest that current suicidal intent is predictive of future intent levels for 2 to 3 h, while current suicidal desire is predictive of future suicidal desire levels for 20 h. Multiple models found that elevated suicidal intent has on average shorter duration than elevated suicidal desire. Finally, inferences about the within-person dynamics of suicidal thinking on the basis of statistical modeling were shown to depend on the frequency at which data was sampled. For example, traditional real-time assessments estimated the duration of severe suicidal states of suicidal desire as 9.5 h, whereas the high-frequency assessments shifted the estimated duration to 1.4 h.

Lipid droplets are intracellular mechanical stressors that impair hepatocyte function

Matrix stiffening and external mechanical stress have been linked to disease and cancer development in multiple tissues, including the liver, where cirrhosis (which increases stiffness markedly) is the major risk factor for hepatocellular carcinoma. Patients with nonalcoholic fatty liver disease and lipid droplet-filled hepatocytes, however, can develop cancer in noncirrhotic, relatively soft tissue. Here, by treating primary human hepatocytes with the monounsaturated fatty acid oleate, we show that lipid droplets are intracellular mechanical stressors with similar effects to tissue stiffening, including nuclear deformation, chromatin condensation, and impaired hepatocyte function. Mathematical modeling of lipid droplets as inclusions that have only mechanical interactions with other cellular components generated results consistent with our experiments. These data show that lipid droplets are intracellular sources of mechanical stress and suggest that nuclear membrane tension integrates cell responses to combined internal and external stresses.

Heterosynaptic long-term potentiation of non-nociceptive synapses requires endocannabinoids, NMDARs, CamKII, and PKCζ

Noxious stimuli or injury can trigger long-lasting sensitization to non-nociceptive stimuli (referred to as allodynia in mammals). Long-term potentiation (LTP) of nociceptive synapses has been shown to contribute to nociceptive sensitization (hyperalgesia) and there is even evidence of heterosynaptic spread of LTP contributing to this type of sensitization. This study will focus on how activation of nociceptors elicits heterosynaptic LTP (hetLTP) in non-nociceptive synapses. Previous studies in the medicinal leech (Hirudo verbana) have demonstrated that high-frequency stimulation (HFS) of nociceptors produces both homosynaptic LTP as well as hetLTP in non-nociceptive afferent synapses. This hetLTP involves endocannabinoid-mediated disinhibition of non-nociceptive synapses at the presynaptic level, but it is not clear if there are additional processes contributing to this synaptic potentiation. In this study, we found evidence for the involvement of postsynaptic level change and observed that postsynaptic N-methyl-d-aspartate (NMDA) receptors (NMDARs) were required for this potentiation. Next, Hirudo orthologs for known LTP signaling proteins, CamKII and PKCζ, were identified based on sequences from humans, mice, and the marine mollusk Aplysia. In electrophysiological experiments, inhibitors of CamKII (AIP) and PKCζ (ZIP) were found to interfere with hetLTP. Interestingly, CamKII was found to be necessary for both induction and maintenance of hetLTP, whereas PKCζ was only necessary for maintenance. These findings show that activation of nociceptors can elicit a potentiation of non-nociceptive synapses through a process that involves both endocannabinoid-mediated disinhibition and NMDAR-initiated signaling pathways.NEW & NOTEWORTHY Pain-related sensitization involves increases in signaling by non-nociceptive sensory neurons. This can allow non-nociceptive afferents to have access to nociceptive circuitry. In this study, we examine a form of synaptic potentiation in which nociceptor activity elicits increases in non-nociceptive synapses. This process involves endocannabinoids, "gating" the activation of NMDA receptors, which in turn activate CamKII and PKCζ. This study provides an important link in how nociceptive stimuli can enhance non-nociceptive signaling related to pain.