Enzymatic catalysis is essential to cell survival. In many instances, enzymes that participate in reaction cascades have been shown to assemble into metabolons in response to the presence of the substrate for the first enzyme. However, what triggers metabolon formation has remained an open question. Through a combination of theory and experiments, we show that enzymes in a cascade can assemble via chemotaxis. We apply microfluidic and fluorescent spectroscopy techniques to study the coordinated movement of the first four enzymes of the glycolysis cascade: hexokinase, phosphoglucose isomerase, phosphofructokinase and aldolase. We show that each enzyme independently follows its own specific substrate gradient, which in turn is produced by the preceding enzymatic reaction. Furthermore, we find that the chemotactic assembly of enzymes occurs even under cytosolic crowding conditions.

Motor proteins such as myosin and kinesin play a major role in cellular cargo transport, muscle contraction, cell division, and engineered nanodevices. Quantifying the collective behavior of coupled motors is critical to our understanding of these systems. An excellent model system is the gliding motility assay, where hundreds of surface-adhered motors propel one cytoskeletal filament such as an actin filament or a microtubule. The filament motion can be observed using fluorescence microscopy, revealing fluctuations in gliding velocity. These velocity fluctuations have been previously quantified by a motional diffusion coefficient, which Sekimoto and Tawada explained as arising from the addition and removal of motors from the linear array of motors propelling the filament as it advances, assuming that different motors are not equally efficient in their force generation. A computational model of kinesin head diffusion and binding to the microtubule allowed us to quantify the heterogeneity of motor efficiency arising from the combination of anharmonic tail stiffness and varying attachment geometries assuming random motor locations on the surface and an absence of coordination between motors. Knowledge of the heterogeneity allows the calculation of the proportionality constant between the motional diffusion coefficient and the motor density. The calculated value (0.3) is within a standard error of our measurements of the motional diffusion coefficient on surfaces with varying motor densities calibrated by landing rate experiments. This allowed us to quantify the loss in efficiency of coupled molecular motors arising from heterogeneity in the attachment geometry.

DNA encoded libraries (DELs) are used for rapid large-scale screening of small molecules against a protein target. These combinatorial libraries are built through several cycles of chemistry and DNA ligation, producing large sets of DNA-tagged molecules. Training machine learning models on DEL data has been shown to be effective at predicting molecules of interest dissimilar from those in the original DEL. Machine learning chemical property prediction approaches rely on the assumption that the property of interest is linked to a single chemical structure. In the context of DNA-encoded libraries, this is equivalent to assuming that every chemical reaction fully yields the desired product. However, in practice, multistep chemical synthesis sometimes generates partial molecules. Each unique DNA tag in a DEL therefore corresponds to a set of possible molecules. Here, we leverage reaction yield data to enumerate the set of possible molecules corresponding to a given DNA tag. This paper demonstrates that training a custom GNN on this richer dataset improves accuracy and generalization performance.

Queueing approaches can capture the stochastic dynamics of chemical reactions and provide a more accurate picture of the reaction kinetics than coupled differential equations in situations where the number of molecules is small. A striking example of such a situation is an enzyme cascade with substrate channeling, where reaction intermediates are directly passed from one enzyme to the next via tunnels or surface paths with limited capacity. In order to better understand the contribution of the stochastic dynamics to the observed enhancement in cascade throughput as a result from substrate channeling, we compare the results of a model using differential equations to describe concentration changes with a queueing model. The continuum model and the queueing model yield identical results, except when the maximum rate of reaction of the enzymes are similar. In two enzyme cascades, the queueing model predicts at most a 50% smaller throughput than the continuum model even if the waiting room size (the maximum number of molecules that can fit in the tunnel or surface path between enzymes) is limited to only one molecule and the enzymes are perfectly matched in their kinetic rates. In longer cascades, the discrepancy increases, reaching a 5-fold difference for a 10 enzyme cascade. In line with theoretical results from queueing theory, stochastic effects are found to always reduce cascade throughput, which means they cannot contribute to the experimentally observed enhancement in throughput due to channeling.

DNA-encoded libraries (DELs) provide the means to make and screen millions of diverse compounds against a target of interest in a single experiment. However, despite producing large volumes of binding data at a relatively low cost, the DEL selection process is susceptible to noise, necessitating computational follow-up to increase signal-to-noise ratios. In this work, we present a set of informatics tools to employ data from prior DEL screen(s) to gain information about which building blocks are most likely to be productive when designing new DELs for the same target. We demonstrate that similar building blocks have similar probabilities of forming compounds that bind. We then build a model from the inference that the combined behavior of individual building blocks is predictive of whether an overall compound binds. We illustrate our approach on a set of three-cycle OpenDEL libraries screened against soluble epoxide hydrolase (sEH) and report performance of more than an order of magnitude greater than random guessing on a holdout set, demonstrating that our model can serve as a baseline for comparison against other machine learning models on DEL data. Lastly, we provide a discussion on how we believe this informatics workflow could be applied to benefit researchers in their specific DEL campaigns.

In the past twenty years, our ability to manipulate and engineer devices at the nano-scale has grown exponentially. As the fabrication of autonomous systems at these scales becomes a reality, the observation of biological structures can help us understand general design principles at the nano-scale. The gliding motility assay is an excellent model system for the observation of collective behavior of coupled motors. Indeed, hundreds of surface-adhered kinesin motors propel one microtubule filament (Figure 1). Filament motion has been observed using fluorescence microscopy, revealing fluctuations in gliding velocity [3; 4]. We here theoretically characterize the motional diffusion coefficients through the heterogeneity factor proposed by Sekimoto and Tawada [5], and use a Brownian dynamics simulation of kinesin head diffusion under an anharmonic potential to determine a theoretical value of 0.3 for this heterogeneity factor.

DNA-encoded libraries (DELs) provide the means to make and screen millions of diverse compounds against a target of interest in a single experiment. However, despite producing large volumes of binding data at a relatively low cost, the DEL selection process is susceptible to noise, necessitating computational follow-up to increase signal-to-noise ratios. In this work, we present a set of informatics tools to analyze DEL selection data so that subsequent DEL screens probe productive regions of chemical space. Our approach segments DEL data at the individual building block level to identify productive building blocks in a library. We show how similar building blocks have a similar probability of binding, which we then employ to predict the behavior of untested building blocks. Lastly, we build a model from the inference that the combined behavior of individual building blocks is predictive of the activity of an overall compound. We report a performance of more than an order of magnitude greater than random guessing on a holdout set, demonstrating that our model can serve as a baseline for comparison against other machine learning models on DEL data.

No abstract