Correction: Evolutionary principles of modular gene regulation in yeasts

[This corrects the article DOI: 10.7554/eLife.00603.].

A tree-like Bayesian structure learning algorithm for small-sample datasets from complex biological model systems

Background

There are increasing efforts to bring high-throughput systems biology techniques to bear on complex animal model systems, often with a goal of learning about underlying regulatory network structures (e.g., gene regulatory networks). However, complex animal model systems typically have significant limitations on cohort sizes, number of samples, and the ability to perform follow-up and validation experiments. These constraints are particularly problematic for many current network learning approaches, which require large numbers of samples and may predict many more regulatory relationships than actually exist.

Results

Here, we test the idea that by leveraging the accuracy and efficiency of classifiers, we can construct high-quality networks that capture important interactions between variables in datasets with few samples. We start from a previously-developed tree-like Bayesian classifier and generalize its network learning approach to allow for arbitrary depth and complexity of tree-like networks. Using four diverse sample networks, we demonstrate that this approach performs consistently better at low sample sizes than the Sparse Candidate Algorithm, a representative approach for comparison because it is known to generate Bayesian networks with high positive predictive value. We develop and demonstrate a resampling-based approach to enable the identification of a viable root for the learned tree-like network, important for cases where the root of a network is not known a priori. We also develop and demonstrate an integrated resampling-based approach to the reduction of variable space for the learning of the network. Finally, we demonstrate the utility of this approach via the analysis of a transcriptional dataset of a malaria challenge in a non-human primate model system, Macaca mulatta, suggesting the potential to capture indicators of the earliest stages of cellular differentiation during leukopoiesis.

Conclusions

We demonstrate that by starting from effective and efficient approaches for creating classifiers, we can identify interesting tree-like network structures with significant ability to capture the relationships in the training data. This approach represents a promising strategy for inferring networks with high positive predictive value under the constraint of small numbers of samples, meeting a need that will only continue to grow as more high-throughput studies are applied to complex model systems.

Electronic supplementary material

The online version of this article (doi:10.1186/s12918-015-0194-7) contains supplementary material, which is available to authorized users.

From genome-scale data to models of infectious disease: A Bayesian network-based strategy to drive model development

High-throughput, genome-scale data present a unique opportunity to link host to pathogen on a molecular level. Forging such connections will help drive the development of mathematical models to better understand and predict both pathogen behavior and the epidemiology of infectious diseases, including malaria. However, the datasets that can aid in identifying these links and models are vast and not amenable to simple, reductionist, and univariate analyses. These datasets require data mining in order to identify the truly important measurements that best describe clinical and molecular observations. Moreover, these datasets typically have relatively few samples due to experimental limitations (particularly for human studies or in vivo animal experiments), making data mining extremely difficult. Here, after first providing a brief overview of common strategies for data reduction and identification of relationships between variables for inclusion in mathematical models, we present a new generalized strategy for performing these data reduction and relationship inference tasks. Our approach emphasizes the importance of robustness when using data to drive model development, particularly when using genome-scale, small-sample in vivo data. We identify the use of appropriate feature reduction combined with data permutations and subsampling strategies as being critical to enable increasingly robust results from network inference using high-dimensional, low-observation data.

Low-cost, point-of-care biomarker quantification

Low-cost, point-of-care (POC) devices that allow fast, on-site disease diagnosis could have a major global health impact, particularly if they can provide quantitative measurement of molecules indicative of a diseased state (biomarkers). Accurate quantification of biomarkers in patient samples is already challenging when research-grade, sophisticated equipment is available; it is even more difficult when constrained to simple, cost-effective POC platforms. Here, we summarize the main challenges to accurate, low-cost POC biomarker quantification. We also review recent efforts to develop and implement POC tools beyond qualitative readouts, and we conclude by identifying important future research directions.

Differential metabolite levels in response to spawning-induced inappetence in Atlantic salmon Salmo salar

Atlantic salmon Salmo salar undergo months-long inappetence during spawning, but it is not known whether this inappetence is a pathological state or one for which the fish are adapted. Recent work has shown that inappetent whale sharks can exhibit circulating metabolite profiles similar to ketosis known to occur in humans during starvation. In this work, metabolite profiling was used to explore differences in analyte profiles between a cohort of inappetent spawning run Atlantic salmon and captively reared animals that were fed up to and through the time of sampling. The two classes of animals were easily distinguished by their metabolite profiles. The sea-run fish had elevated ɷ-9 fatty acids relative to the domestic feeding animals, while other fatty acid concentrations were reduced. Sugar alcohols were generally elevated in inappetent animals, suggesting potentially novel metabolic responses or pathways in fish that feature these compounds. Compounds expected to indicate a pathological catabolic state were not more abundant in the sea-run fish, suggesting that the animals, while inappetent, were not stressed in an unnatural way. These findings demonstrate the power of discovery-based metabolomics for exploring biochemistry in poorly understood animal models.

LK-DFBA: a linear programming-based modeling strategy for capturing dynamics and metabolite-dependent regulation in metabolism

BACKGROUND

The systems-scale analysis of cellular metabolites, "metabolomics," provides data ideal for applications in metabolic engineering. However, many of the computational tools for strain design are built around Flux Balance Analysis (FBA), which makes assumptions that preclude direct integration of metabolomics data into the underlying models. Finding a way to retain the advantages of FBA's linear structure while relaxing some of its assumptions could allow us to account for metabolite levels and metabolite-dependent regulation in strain design tools built from FBA, improving the accuracy of predictions made by these tools. We designed, implemented, and characterized a modeling strategy based on Dynamic FBA (DFBA), called Linear Kinetics-Dynamic Flux Balance Analysis (LK-DFBA), to satisfy these specifications. Our strategy adds constraints describing the dynamics and regulation of metabolism that are strictly linear. We evaluated LK-DFBA against alternative modeling frameworks using simulated noisy data from a small in silico model and a larger model of central carbon metabolism in E. coli, and compared each framework's ability to recapitulate the original system.

RESULTS

In the smaller model, we found that we could use regression from a dynamic flux estimation (DFE) with an optional non-linear parameter optimization to reproduce metabolite concentration dynamic trends more effectively than an ordinary differential equation model with generalized mass action rate laws when tested under realistic data sampling frequency and noise levels. We observed detrimental effects across all tested modeling approaches when metabolite time course data were missing, but found these effects to be smaller for LK-DFBA in most cases. With the E. coli model, we produced qualitatively reasonable results with similar properties to the smaller model and explored two different parameterization structures that yield trade-offs in computation time and accuracy.

CONCLUSIONS

LK-DFBA allows for calculation of metabolite concentrations and considers metabolite-dependent regulation while still retaining many computational advantages of FBA. This provides the proof-of-principle for a new metabolic modeling framework with the potential to create genome-scale dynamic models and the potential to be applied in strain engineering tools that currently use FBA.

Improved metabolite profile smoothing for flux estimation

As genome-scale metabolic models become more sophisticated and dynamic, one significant challenge in using these models is to effectively integrate increasingly prevalent systems-scale metabolite profiling data into them. One common data processing step when integrating metabolite data is to smooth experimental time course measurements: the smoothed profiles can be used to estimate metabolite accumulation (derivatives), and thus the flux distribution of the metabolic model. However, this smoothing step is susceptible to the (often significant) noise in experimental measurements, limiting the accuracy of downstream model predictions. Here, we present several improvements to current approaches for smoothing metabolite time course data using defined functions. First, we use a biologically-inspired mathematical model function taken from transcriptional profiling and clustering literature that captures the dynamics of many biologically relevant transient processes. We demonstrate that it is competitive with, and often superior to, previously described fitting schemas, and may serve as an effective single option for data smoothing in metabolic flux applications. We also implement a resampling-based approach to buffer out sensitivity to specific data sets and allow for more accurate fitting of noisy data. We found that this method, as well as the addition of parameter space constraints, yielded improved estimates of concentrations and derivatives (fluxes) in previously described fitting functions. These methods have the potential to improve the accuracy of existing and future dynamic metabolic models by allowing for the more effective integration of metabolite profiling data.

Point-of-Care Analyte Quantification and Digital Readout via Lysate-Based Cell-Free Biosensors Interfaced with Personal Glucose Monitors

Field-deployable diagnostics based on cell-free systems have advanced greatly, but on-site quantification of target analytes remains a challenge. Here we demonstrate that Escherichia coli lysate-based cell-free biosensors coupled to a personal glucose monitor (PGM) can enable on-site analyte quantification, with the potential for straightforward reconfigurability to diverse types of analytes. We show that analyte-responsive regulators of transcription and translation can modulate the production of the reporter enzyme β-galactosidase, which in turn converts lactose into glucose for PGM quantification. Because glycolysis is active in the lysate and would readily deplete converted glucose, we decoupled enzyme production and glucose conversion to increase the end point signal output. However, this lysate metabolism did allow for one-pot removal of glucose present in complex samples (like human serum) without confounding target quantification. Taken together, our results show that integrating lysate-based cell-free biosensors with PGMs enables accessible target detection and quantification at the point of need.

Dramatic transcriptomic differences in Macaca mulatta and Macaca fascicularis with Plasmodium knowlesi infections

Plasmodium knowlesi, a model malaria parasite, is responsible for a significant portion of zoonotic malaria cases in Southeast Asia and must be controlled to avoid disease severity and fatalities. However, little is known about the host-parasite interactions and molecular mechanisms in play during the course of P. knowlesi malaria infections, which also may be relevant across Plasmodium species. Here we contrast P. knowlesi sporozoite-initiated infections in Macaca mulatta and Macaca fascicularis using whole blood RNA-sequencing and transcriptomic analysis. These macaque hosts are evolutionarily close, yet malaria-naïve M. mulatta will succumb to blood-stage infection without treatment, whereas malaria-naïve M. fascicularis controls parasitemia without treatment. This comparative analysis reveals transcriptomic differences as early as the liver phase of infection, in the form of signaling pathways that are activated in M. fascicularis, but not M. mulatta. Additionally, while most immune responses are initially similar during the acute stage of the blood infection, significant differences arise subsequently. The observed differences point to prolonged inflammation and anti-inflammatory effects of IL10 in M. mulatta, while M. fascicularis undergoes a transcriptional makeover towards cell proliferation, consistent with its recovery. Together, these findings suggest that timely detection of P. knowlesi in M. fascicularis, coupled with control of inflammation while initiating the replenishment of key cell populations, helps contain the infection. Overall, this study points to specific genes and pathways that could be investigated as a basis for new drug targets that support recovery from acute malaria.

Metabolic Dynamics in Escherichia coli-Based Cell-Free Systems

The field of metabolic engineering has yielded remarkable accomplishments in using cells to produce valuable molecules, and cell-free expression (CFE) systems have the potential to push the field even further. However, CFE systems still face some outstanding challenges, including endogenous metabolic activity that is poorly understood yet has a significant impact on CFE productivity. Here, we use metabolomics to characterize the temporal metabolic changes in CFE systems and their constituent components, including significant metabolic activity in central carbon and amino acid metabolism. We find that while changing the reaction starting state via lysate preincubation impacts protein production, it has a comparatively small impact on metabolic state. We also demonstrate that changes to lysate preparation have a larger effect on protein yield and temporal metabolic profiles, though general metabolic trends are conserved. Finally, while we improve protein production through targeted supplementation of metabolic enzymes, we show that the endogenous metabolic activity is fairly resilient to these enzymatic perturbations. Overall, this work highlights the robust nature of CFE reaction metabolism as well as the importance of understanding the complex interdependence of metabolites and proteins in CFE systems to guide optimization efforts.

SCOUR: a stepwise machine learning framework for predicting metabolite-dependent regulatory interactions

BACKGROUND

The topology of metabolic networks is both well-studied and remarkably well-conserved across many species. The regulation of these networks, however, is much more poorly characterized, though it is known to be divergent across organisms-two characteristics that make it difficult to model metabolic networks accurately. While many computational methods have been built to unravel transcriptional regulation, there have been few approaches developed for systems-scale analysis and study of metabolic regulation. Here, we present a stepwise machine learning framework that applies established algorithms to identify regulatory interactions in metabolic systems based on metabolic data: stepwise classification of unknown regulation, or SCOUR.

RESULTS

We evaluated our framework on both noiseless and noisy data, using several models of varying sizes and topologies to show that our approach is generalizable. We found that, when testing on data under the most realistic conditions (low sampling frequency and high noise), SCOUR could identify reaction fluxes controlled only by the concentration of a single metabolite (its primary substrate) with high accuracy. The positive predictive value (PPV) for identifying reactions controlled by the concentration of two metabolites ranged from 32 to 88% for noiseless data, 9.2 to 49% for either low sampling frequency/low noise or high sampling frequency/high noise data, and 6.6-27% for low sampling frequency/high noise data, with results typically sufficiently high for lab validation to be a practical endeavor. While the PPVs for reactions controlled by three metabolites were lower, they were still in most cases significantly better than random classification.

CONCLUSIONS

SCOUR uses a novel approach to synthetically generate the training data needed to identify regulators of reaction fluxes in a given metabolic system, enabling metabolomics and fluxomics data to be leveraged for regulatory structure inference. By identifying and triaging the most likely candidate regulatory interactions, SCOUR can drastically reduce the amount of time needed to identify and experimentally validate metabolic regulatory interactions. As high-throughput experimental methods for testing these interactions are further developed, SCOUR will provide critical impact in the development of predictive metabolic models in new organisms and pathways.

Gas Chromatography-Mass Spectrometry Microbial Metabolomics for Applications in Strain Optimization

Metabolomics is the systems-scale measurement of biochemical intermediates in biological systems; by virtue of its deep and broad study of metabolism, it has great potential for applications in metabolic engineering. While a number of the analytical techniques used widely in metabolomics are familiar to metabolic engineers performing post hoc analyses of product titers, the requirements for accurately capturing metabolism at a systems scale rather than just measuring a single secreted product are much more complicated. Nonetheless, metabolomics (which is still not widely available as an affordable consumer service like many molecular biology services) is within reach of many properly equipped metabolic engineering groups. To this end, we present a detailed metabolomics protocol with application to strain optimization. Specifically, we focus on characterizing metabolism in the yeast Saccharomyces cerevisiae using gas chromatography coupled to mass spectrometry. The measurement of metabolic intermediates that results from such approaches has the potential to enable more informed and rational efforts towards pathway engineering and strain optimization.

Systems biology-based analysis of cell-free systems

Cell-free expression systems are becoming increasingly widely used due to their diverse applications in biotechnology. Despite this rapid expansion in adoption, many aspects of cell-free systems remain surprisingly poorly understood. Systems biology approaches make it possible to characterize cell-free systems deeply and broadly to better understand their underlying complexity. Here, we review recent systems biology studies that have provided insight into cell-free systems. We focus on characterization of the cell-free proteome, including its dependence on preparation protocol and host strain, as well as the cell-free metabolome and the relationship of endogenous metabolism to system performance. We conclude by highlighting promising future research directions.

Nucleic acid partitioning in PEG-Ficoll protocells

The phase separation of aqueous polymer solutions is a widely used method for producing self-assembled, membraneless droplet protocells. Non-ionic synthetic polymers forming an aqueous two-phase system (ATPS) have been shown to reliably form protocells that, when equipped with biological materials, are useful for applications such as analyte detection. Previous characterization of an ATPS-templated protocell did not investigate the effects of its biological components on phase stability. Here we report the phase diagram of a PEG 35k-Ficoll 400k-water ATPS at baseline and in the presence of necessary protocell components. Because the stability of an ATPS can be sensitive to small changes in composition, which in turn impacts solute partitioning, we present partitioning data of a variety of nucleic acids in response to protocell additives. The results show that the additives-particularly a mixture of salts and small organic molecules-have profound positive effects on ATPS stability and nucleic acid partitioning, both of which significantly contribute to protocell function. Our data uncovers several new areas of optimization for future protocell engineering.

Metabolomics Analysis of Cell-Free Expression Systems Using Gas Chromatography-Mass Spectrometry

Metabolomics is the systems-scale study of the biochemical intermediates of metabolism. This approach has great potential to uncover how metabolic intermediates are used and generated in cell-free expression systems, something that is to date not well understood. Here, we present a detailed metabolomics protocol for characterization of the small molecules in cell-free systems. We specifically focus on the analysis of Escherichia coli lysate-based cell-free systems using gas chromatography coupled to mass spectrometry. Measuring and monitoring the metabolic changes in cell-free systems can provide insight into the ways that metabolites affect the productivity of cell-free reactions, ultimately allowing for more informed engineering and optimization efforts for cell-free systems.

MaHPIC malaria systems biology data from Plasmodium cynomolgi sporozoite longitudinal infections in macaques

Plasmodium cynomolgi causes zoonotic malarial infections in Southeast Asia and this parasite species is important as a model for Plasmodium vivax and Plasmodium ovale. Each of these species produces hypnozoites in the liver, which can cause relapsing infections in the blood. Here we present methods and data generated from iterative longitudinal systems biology infection experiments designed and performed by the Malaria Host-Pathogen Interaction Center (MaHPIC) to delve deeper into the biology, pathogenesis, and immune responses of P. cynomolgi in the Macaca mulatta host. Infections were initiated by sporozoite inoculation. Blood and bone marrow samples were collected at defined timepoints for biological and computational experiments and integrative analyses revolving around primary illness, relapse illness, and subsequent disease and immune response patterns. Parasitological, clinical, haematological, immune response, and -omic datasets (transcriptomics, proteomics, metabolomics, and lipidomics) including metadata and computational results have been deposited in public repositories. The scope and depth of these datasets are unprecedented in studies of malaria, and they are projected to be a F.A.I.R., reliable data resource for decades.

Introduction to Engineering Biology: A Conceptual Framework for Teaching Synthetic Biology

As the impacts of engineering biology grow, it is important to introduce the field early and in an accessible way. However, teaching engineering biology poses challenges, such as limited representation of the field in widely used scientific textbooks or curricula, and the interdisciplinary nature of the subject. We have created an adaptable curriculum module that can be used by anyone to teach the basic principles and applications of engineering biology. The module consists of a versatile, concept-based slide deck designed by experts across engineering biology to cover key topic areas. Starting with the design, build, test, and learn cycle, the slide deck covers the framework, core tools, and applications of the field at an undergraduate level. The module is available for free on a public website and can be used in a stand-alone fashion or incorporated into existing curricular materials. Our aim is that this modular, accessible slide deck will improve the ease of teaching current engineering biology topics and increase public engagement with the field.

A Cell-Free Biosensor for Assessment of Hyperhomocysteinemia

Hyperhomocysteinemia─a condition characterized by elevated levels of homocysteine in the blood─is associated with multiple health conditions including folate deficiency and birth defects, but there are no convenient, low-cost methods to measure homocysteine in plasma. A cell-free biosensor that harnesses the native homocysteine sensing machinery of Escherichia coli bacteria could satisfy the need for a detection platform with these characteristics. Here, we describe our efforts to engineer a cell-free biosensor for point-of-care, low-cost assessment of homocysteine status. This biosensor can detect physiologically relevant concentrations of homocysteine in plasma with a colorimetric output visible to the naked eye in under 1.5 h, making it a fast, convenient tool for point-of-use diagnosis and monitoring of hyperhomocysteinemia and related health conditions.

A modular cell-free protein biosensor platform using split T7 RNA polymerase

Conventional laboratory protein detection techniques are not suitable for point-of-care (POC) use because they require expensive equipment and laborious protocols, and existing POC assays suffer from long development timescales. Here, we describe a modular cell-free biosensing platform for generalizable protein detection that we call TLISA (T7 RNA polymerase-linked immunosensing assay), designed for extreme flexibility and equipment-free use. TLISA uses a split T7 RNA polymerase fused to affinity domains against a protein. The target antigen drives polymerase reassembly, inducing reporter expression. We characterize the platform and then demonstrate its modularity by using 16 affinity domains against four different antigens with minimal protocol optimization. We show that TLISA is suitable for POC use by sensing human biomarkers in serum and saliva with a colorimetric readout within 1 hour and by demonstrating functionality after lyophilization. Altogether, this technology has the potential to enable truly rapid, reconfigurable, modular, and equipment-free detection of diverse classes of proteins.

Bacillus subtilis engineered for topical delivery of an antifungal agent

Fungal skin infections are a common condition affecting 20-25 percent of the world population. While these conditions are treatable with regular application of an antifungal medication, we sought to develop a more convenient, longer-lasting topical antifungal platform that could increase patient adherence to treatment regimens by using Bacillus subtilis, a naturally antifungal bacteria found on the skin, for drug production and delivery. In this study, we engineered B. subtilis for increased production of the antifungal lipopeptide iturin A by overexpression of the pleiotropic regulator DegQ. The engineered strain had an over 200% increase in iturin A production as detected by HPLC, accompanied by slower growth but the same terminal cell density as determined by absorbance measurements of liquid culture. In an in vitro antifungal assay, we found that despite its higher iturin A production, the engineered strain was less effective at reducing the growth of a plug of the pathogenic fungus Trichophyton mentagrophytes on an agar plate compared to the parent strain. The reduced efficacy of the engineered strain may be explained by its reduced growth rate, which highlights the need to address trade-offs between titers (e.g. measured drug production) and other figures of merit (e.g. growth rate) during metabolic engineering.

Diverse classes of constraints enable broader applicability of a linear programming-based dynamic metabolic modeling framework

Current metabolic modeling tools suffer from a variety of limitations, from scalability to simplifying assumptions, that preclude their use in many applications. We recently created a modeling framework, Linear Kinetics-Dynamic Flux Balance Analysis (LK-DFBA), that addresses a key gap: capturing metabolite dynamics and regulation while retaining a potentially scalable linear programming structure. Key to this framework's success are the linear kinetics and regulatory constraints imposed on the system. However, while the linearity of these constraints reduces computational complexity, it may not accurately capture the behavior of many biochemical systems. Here, we developed three new classes of LK-DFBA constraints to better model interactions between metabolites and the reactions they regulate. We tested these new approaches on several synthetic and biological systems, and also performed the first-ever comparison of LK-DFBA predictions to experimental data. We found that no single constraint approach was optimal across all systems examined, and systems with the same topological structure but different parameters were often best modeled by different types of constraints. However, we did find that when genetic perturbations were implemented in the systems, the optimal constraint approach typically remained the same as for the wild-type regardless of the model topology or parameterization, indicating that just a single wild-type dataset could allow identification of the ideal constraint to enable model predictivity for a given system. These results suggest that the availability of multiple constraint approaches will allow LK-DFBA to model a wider range of metabolic systems.