A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action.

Current machine learning techniques enable robust association of biological signals with measured phenotypes, but these approaches are incapable of identifying causal relationships. Here, we develop an integrated "white-box" biochemical screening, network modeling, and machine learning approach for revealing causal mechanisms and apply this approach to understanding antibiotic efficacy. We counter-screen diverse metabolites against bactericidal antibiotics in Escherichia coli and simulate their corresponding metabolic states using a genome-scale metabolic network model. Regression of the measured screening data on model simulations reveals that purine biosynthesis participates in antibiotic lethality, which we validate experimentally. We show that antibiotic-induced adenine limitation increases ATP demand, which elevates central carbon metabolism activity and oxygen consumption, enhancing the killing effects of antibiotics. This work demonstrates how prospective network modeling can couple with machine learning to identify complex causal mechanisms underlying drug efficacy.

A high-throughput synthetic biology approach for studying combinatorial chromatin-based transcriptional regulation.

The construction of synthetic gene circuits requires the rational combination of multiple regulatory components, but predicting their behavior can be challenging due to poorly understood component interactions and unexpected emergent behaviors. In eukaryotes, chromatin regulators (CRs) are essential regulatory components that orchestrate gene expression. Here, we develop a screening platform to investigate the impact of CR pairs on transcriptional activity in yeast. We construct a combinatorial library consisting of over 1,900 CR pairs and use a high-throughput workflow to characterize the impact of CR co-recruitment on gene expression. We recapitulate known interactions and discover several instances of CR pairs with emergent behaviors. We also demonstrate that supervised machine learning models trained with low-dimensional amino acid embeddings accurately predict the impact of CR co-recruitment on transcriptional activity. This work introduces a scalable platform and machine learning approach that can be used to study how networks of regulatory components impact gene expression.

A self-propagating, barcoded transposon system for the dynamic rewiring of genomic networks.

In bacteria, natural transposon mobilization can drive adaptive genomic rearrangements. Here, we build on this capability and develop an inducible, self-propagating transposon platform for continuous genome-wide mutagenesis and the dynamic rewiring of gene networks in bacteria. We first use the platform to study the impact of transposon functionalization on the evolution of parallel Escherichia coli populations toward diverse carbon source utilization and antibiotic resistance phenotypes. We then develop a modular, combinatorial assembly pipeline for the functionalization of transposons with synthetic or endogenous gene regulatory elements (e.g., inducible promoters) as well as DNA barcodes. We compare parallel evolutions across alternating carbon sources and demonstrate the emergence of inducible, multigenic phenotypes and the ease with which barcoded transposons can be tracked longitudinally to identify the causative rewiring of gene networks. This work establishes a synthetic transposon platform that can be used to optimize strains for industrial and therapeutic applications, for example, by rewiring gene networks to improve growth on diverse feedstocks, as well as help address fundamental questions about the dynamic processes that have sculpted extant gene networks.

Network-level allosteric effects are elucidated by detailing how ligand-binding events modulate utilization of catalytic potentials.

Allosteric regulation has traditionally been described by mathematically-complex allosteric rate laws in the form of ratios of polynomials derived from the application of simplifying kinetic assumptions. Alternatively, an approach that explicitly describes all known ligand-binding events requires no simplifying assumptions while allowing for the computation of enzymatic states. Here, we employ such a modeling approach to examine the "catalytic potential" of an enzyme-an enzyme's capacity to catalyze a biochemical reaction. The catalytic potential is the fundamental result of multiple ligand-binding events that represents a "tug of war" among the various regulators and substrates within the network. This formalism allows for the assessment of interacting allosteric enzymes and development of a network-level understanding of regulation. We first define the catalytic potential and use it to characterize the response of three key kinases (hexokinase, phosphofructokinase, and pyruvate kinase) in human red blood cell glycolysis to perturbations in ATP utilization. Next, we examine the sensitivity of the catalytic potential by using existing personalized models, finding that the catalytic potential allows for the identification of subtle but important differences in how individuals respond to such perturbations. Finally, we explore how the catalytic potential can help to elucidate how enzymes work in tandem to maintain a homeostatic state. Taken together, this work provides an interpretation and visualization of the dynamic interactions and network-level effects of interacting allosteric enzymes.

Blood volume measurement using cardiovascular magnetic resonance and ferumoxytol: preclinical validation.

BACKGROUND: The hallmark of heart failure is increased blood volume. Quantitative blood volume measures are not conveniently available and are not tested in heart failure management. We assess ferumoxytol, a marketed parenteral iron supplement having a long intravascular half-life, to measure the blood volume with cardiovascular magnetic resonance (CMR). METHODS: Swine were administered 0.7 mg/kg ferumoxytol and blood pool T1 was measured repeatedly for an hour to characterize contrast agent extraction and subsequent effect on Vblood estimates. We compared CMR blood volume with a standard carbon monoxide rebreathing method. We then evaluated three abbreviated acquisition protocols for bias and precision. RESULTS: Mean plasma volume estimated by ferumoxytol was 61.9 ± 4.3 ml/kg. After adjustment for hematocrit the resultant mean blood volume was 88.1 ± 9.4 ml/kg, which agreed with carbon monoxide measures (91.1 ± 18.9 ml/kg). Repeated measurements yielded a coefficient of variation of 6.9%, and Bland-Altman repeatability coefficient of 14%. The blood volume estimates with abbreviated protocols yielded small biases (mean differences between 0.01-0.06 L) and strong correlations (r2 between 0.97-0.99) to the reference values indicating clinical feasibility. CONCLUSIONS: In this swine model, ferumoxytol CMR accurately measures plasma volume, and with correction for hematocrit, blood volume. Abbreviated protocols can be added to diagnostic CMR examination for heart failure within 8 min.

Modulating the evolutionary trajectory of tolerance using antibiotics with different metabolic dependencies.

Antibiotic tolerance, or the ability of bacteria to survive antibiotic treatment in the absence of genetic resistance, has been linked to chronic and recurrent infections. Tolerant cells are often characterized by a low metabolic state, against which most clinically used antibiotics are ineffective. Here, we show that tolerance readily evolves against antibiotics that are strongly dependent on bacterial metabolism, but does not arise against antibiotics whose efficacy is only minimally affected by metabolic state. We identify a mechanism of tolerance evolution in E. coli involving deletion of the sodium-proton antiporter gene nhaA, which results in downregulated metabolism and upregulated stress responses. Additionally, we find that cycling of antibiotics with different metabolic dependencies interrupts evolution of tolerance in vitro, increasing the lifetime of treatment efficacy. Our work highlights the potential for limiting the occurrence and extent of tolerance by accounting for antibiotic dependencies on bacterial metabolism.

An engineered live biotherapeutic for the prevention of antibiotic-induced dysbiosis.

Antibiotic-induced alterations in the gut microbiota are implicated in many metabolic and inflammatory diseases, increase the risk of secondary infections and contribute to the emergence of antimicrobial resistance. Here we report the design and in vivo performance of an engineered strain of Lactococcus lactis that altruistically degrades the widely used broad-spectrum antibiotics ß-lactams (which disrupt commensal bacteria in the gut) through the secretion and extracellular assembly of a heterodimeric ß-lactamase. The engineered ß-lactamase-expression system does not confer ß-lactam resistance to the producer cell, and is encoded via a genetically unlinked two-gene biosynthesis strategy that is not susceptible to dissemination by horizontal gene transfer. In a mouse model of parenteral ampicillin treatment, oral supplementation with the engineered live biotherapeutic minimized gut dysbiosis without affecting the ampicillin concentration in serum, precluded the enrichment of antimicrobial resistance genes in the gut microbiome and prevented the loss of colonization resistance against Clostridioides difficile. Engineered live biotherapeutics that safely degrade antibiotics in the gut may represent a suitable strategy for the prevention of dysbiosis and its associated pathologies.

Sequence-to-function deep learning frameworks for engineered riboregulators.

While synthetic biology has revolutionized our approaches to medicine, agriculture, and energy, the design of completely novel biological circuit components beyond naturally-derived templates remains challenging due to poorly understood design rules. Toehold switches, which are programmable nucleic acid sensors, face an analogous design bottleneck; our limited understanding of how sequence impacts functionality often necessitates expensive, time-consuming screens to identify effective switches. Here, we introduce Sequence-based Toehold Optimization and Redesign Model (STORM) and Nucleic-Acid Speech (NuSpeak), two orthogonal and synergistic deep learning architectures to characterize and optimize toeholds. Applying techniques from computer vision and natural language processing, we 'un-box' our models using convolutional filters, attention maps, and in silico mutagenesis. Through transfer-learning, we redesign sub-optimal toehold sensors, even with sparse training data, experimentally validating their improved performance. This work provides sequence-to-function deep learning frameworks for toehold selection and design, augmenting our ability to construct potent biological circuit components and precision diagnostics.

A CRISPR-based assay for the detection of opportunistic infections post-transplantation and for the monitoring of transplant rejection.

In organ transplantation, infection and rejection are major causes of graft loss. They are linked by the net state of immunosuppression. To diagnose and treat these conditions earlier, and to improve long-term patient outcomes, refined strategies for the monitoring of patients after graft transplantation are needed. Here, we show that a fast and inexpensive assay based on CRISPR-Cas13 accurately detects BK polyomavirus DNA and cytomegalovirus DNA from patient-derived blood and urine samples, as well as CXCL9 messenger RNA (a marker of graft rejection) at elevated levels in urine samples from patients experiencing acute kidney transplant rejection. The assay, which we adapted for lateral-flow readout, enables-via simple visualization-the post-transplantation monitoring of common opportunistic viral infections and of graft rejection, and should facilitate point-of-care post-transplantation monitoring.