Xiaoliang Sun

COVAIN: a toolbox for uni- and multivariate statistics, time-series and correlation network analysis and inverse estimation of the differential Jacobian from metabolomics covariance data

Metabolomics emerges as one of the cornerstones in systems biology by characterizing metabolic activities as the ultimate readout of physiological processes of biological systems thereby linking genotypes with the corresponding phenotypes. As metabolomics data are high-dimensional, statistical data analysis is complex. No single technique for statistical analysis and biological interpretation of these ultracomplex data is sufficient to reveal the full information content of the data. Therefore a combination of univariate and multivariate statistics, network topology and biochemical pathway mapping analysis is in all cases recommended. Therefore, we developed a toolbox with fully graphical user interface support in MATLAB© called covariance inverse (COVAIN). COVAIN provides a complete workflow including uploading data, data preprocessing, uni- and multivariate statistical analysis, Granger time-series analysis, pathway mapping, correlation network topology analysis and visualization, and finally saving results in a user-friendly way. It covers analysis of variance, principal components analysis, independent components analysis, clustering and correlation coefficient analysis and integrates new algorithms, such as Granger causality and permutation entropy analysis that are not implemented in other similar softwares. Furthermore, we provide a new algorithm to reconstruct a differential Jacobian matrix of two different metabolic conditions. The algorithm is based on the assumptions of stochastic fluctuations in the metabolic network as described by us recently. By integrating the metabolomics covariance matrix and the stoichiometric matrix N of the corresponding pathways this approach allows for a systematic investigation of perturbation sites in the biochemical network based on metabolomics data. COVAIN was primarily developed for metabolomics data but can also be used for other omics data analysis. A C language programming module was integrated to handle computational intensive work for large datasets, e.g., genome-level proteomics and transcriptomics data sets which usually contain several thousand or more variables. COVAIN can perform cross analysis and integration between several datasets, which might be useful to investigate responses on different hierarchies of cellular contexts and to reveal the systems response as an integrated molecular network. The source codes can be downloaded from http://www.univie.ac.at/mosys/software.html.

mzGroupAnalyzer-Predicting Pathways and Novel Chemical Structures from Untargeted High-Throughput Metabolomics Data

The metabolome is a highly dynamic entity and the final readout of the genotype x environment x phenotype (GxExP) relationship of an organism. Monitoring metabolite dynamics over time thus theoretically encrypts the whole range of possible chemical and biochemical transformations of small molecules involved in metabolism. The bottleneck is, however, the sheer number of unidentified structures in these samples. This represents the next challenge for metabolomics technology and is comparable with genome sequencing 30 years ago. At the same time it is impossible to handle the amount of data involved in a metabolomics analysis manually. Algorithms are therefore imperative to allow for automated m/z feature extraction and subsequent structure or pathway assignment. Here we provide an automated pathway inference strategy comprising measurements of metabolome time series using LC- MS with high resolution and high mass accuracy. An algorithm was developed, called mzGroupAnalyzer, to automatically explore the metabolome for the detection of metabolite transformations caused by biochemical or chemical modifications. Pathways are extracted directly from the data and putative novel structures can be identified. The detected m/z features can be mapped on a van Krevelen diagram according to their H/C and O/C ratios for pattern recognition and to visualize oxidative processes and biochemical transformations. This method was applied to Arabidopsis thaliana treated simultaneously with cold and high light. Due to a protective antioxidant response the plants turn from green to purple color via the accumulation of flavonoid structures. The detection of potential biochemical pathways resulted in 15 putatively new compounds involved in the flavonoid-pathway. These compounds were further validated by product ion spectra from the same data. The mzGroupAnalyzer is implemented in the graphical user interface (GUI) of the metabolomics toolbox COVAIN (Sun & Weckwerth, 2012, Metabolomics 8: 81–93). The strategy can be extended to any biological system.

Metabolite changes with induction of Cuscuta haustorium and translocation from host plants

Cuscuta is a stem holoparasitic plant without leaves or roots, parasitizing various types of host plants and causing major problems for certain crops. Cuscuta is known as a generalist and, thus, must have unique parasite strategies to cope with different host plants. For elucidating metabolic responses and mechanisms of parasitization, metabolomic approaches using GC/MS were applied. We compared five stages of Cuscuta japonica: early stage seedlings, with far red light (FR) cue, with contact signal, haustorium induced seedlings by both signals and adult plant parasites on host plants. Sugars, amino acids, organic acids, nucleic acids, and polyols were identified from the polar phase fraction. The apical part contained metabolite profiles different from the haustorium induced part or the basal part. Amino acid and some organic acids were up-regulated for haustorium induction but decreased after parasitization. After attachment to different host plants, metabolite profiles of Cuscuta japonica changed dramatically due to the absorption of specific host plant metabolites such as pinitol. Cuscuta seedlings attached to pinitol rich host plants contained more pinitol and showed different profiles from those attached to plants having less or lacking pinitol.

Solving the differential biochemical Jacobian from metabolomics covariance data.

High-throughput molecular analysis has become an integral part in organismal systems biology. In contrast, due to a missing systematic linkage of the data with functional and predictive theoretical models of the underlying metabolic network the understanding of the resulting complex data sets is lacking far behind. Here, we present a biomathematical method addressing this problem by using metabolomics data for the inverse calculation of a biochemical Jacobian matrix, thereby linking computer-based genome-scale metabolic reconstruction and in vivo metabolic dynamics. The incongruity of metabolome coverage by typical metabolite profiling approaches and genome-scale metabolic reconstruction was solved by the design of superpathways to define a metabolic interaction matrix. A differential biochemical Jacobian was calculated using an approach which links this metabolic interaction matrix and the covariance of metabolomics data satisfying a Lyapunov equation. The predictions of the differential Jacobian from real metabolomic data were found to be correct by testing the corresponding enzymatic activities. Moreover it is demonstrated that the predictions of the biochemical Jacobian matrix allow for the design of parameter optimization strategies for ODE-based kinetic models of the system. The presented concept combines dynamic modelling strategies with large-scale steady state profiling approaches without the explicit knowledge of individual kinetic parameters. In summary, the presented strategy allows for the identification of regulatory key processes in the biochemical network directly from metabolomics data and is a fundamental achievement for the functional interpretation of metabolomics data.

System-Level and Granger Network Analysis of Integrated Proteomic and Metabolomic Dynamics Identifies Key Points of Grape Berry Development at the Interface of Primary and Secondary Metabolism

Grapevine is a fruit crop with worldwide economic importance. The grape berry undergoes complex biochemical changes from fruit set until ripening. This ripening process and production processes define the wine quality. Thus, a thorough understanding of berry ripening is crucial for the prediction of wine quality. For a systemic analysis of grape berry development we applied mass spectrometry based platforms to analyse the metabolome and proteome of Early Campbell at 12 stages covering major developmental phases. Primary metabolites involved in central carbon metabolism, such as sugars, organic acids and amino acids together with various bioactive secondary metabolites like flavonols, flavan-3-ols and anthocyanins were annotated and quantified. At the same time, the proteomic analysis revealed the protein dynamics of the developing grape berries. Multivariate statistical analysis of the integrated metabolomic and proteomic dataset revealed the growth trajectory and corresponding metabolites and proteins contributing most to the specific developmental process. K-means clustering analysis revealed 12 highly specific clusters of co-regulated metabolites and proteins. Granger causality network analysis allowed for the identification of time-shift correlations between metabolite-metabolite, protein- protein and protein-metabolite pairs which is especially interesting for the understanding of developmental processes. The integration of metabolite and protein dynamics with their corresponding biochemical pathways revealed an energy-linked metabolism before veraison with high abundances of amino acids and accumulation of organic acids, followed by protein and secondary metabolite synthesis. Anthocyanins were strongly accumulated after veraison whereas other flavonoids were in higher abundance at early developmental stages and decreased during the grape berry developmental processes. A comparison of the anthocyanin profile of Early Campbell to other cultivars revealed similarities to Concord grape and indicates the strong effect of genetic background on metabolic partitioning in primary and secondary metabolism.

From Proteomics to Systems Biology: MAPA, MASS WESTERN, PROMEX, and COVAIN as a User-Oriented Platform

Genome sequencing and systems biology are revolutionizing life sciences. Proteomics emerged as a fundamental technique of this novel research area as it is the basis for gene function analysis and modeling of dynamic protein networks. Here a complete proteomics platform suited for functional genomics and systems biology is presented. The strategy includes MAPA (mass accuracy precursor alignment; http://www.univie.ac.at/mosys/software.html ) as a rapid exploratory analysis step; MASS WESTERN for targeted proteomics; COVAIN ( http://www.univie.ac.at/mosys/software.html ) for multivariate statistical analysis, data integration, and data mining; and PROMEX ( http://www.univie.ac.at/mosys/databases.html ) as a database module for proteogenomics and proteotypic peptides for targeted analysis. Moreover, the presented platform can also be utilized to integrate metabolomics and transcriptomics data for the analysis of metabolite-protein-transcript correlations and time course analysis using COVAIN. Examples for the integration of MAPA and MASS WESTERN data, proteogenomic and metabolic modeling approaches for functional genomics, phosphoproteomics by integration of MOAC (metal-oxide affinity chromatography) with MAPA, and the integration of metabolomics, transcriptomics, proteomics, and physiological data using this platform are presented. All software and step-by-step tutorials for data processing and data mining can be downloaded from http://www.univie.ac.at/mosys/software.html.

Draft Genome Sequence of the Growth-Promoting Endophyte Paenibacillus sp. P22, Isolated from Populus

ABSTRACT Paenibacillus sp. P22 is a Gram-negative facultative anaerobic endospore-forming bacterium isolated from poplar hybrid 741 (♀[Populus alba × (P. davidiana + P. simonii) × P. tomentosa]). This bacterium shows strong similarities to Paenibacillus humicus, and important growth-promoting effects on in vitro grown explants of poplar hybrid 741 have been described.

Challenges of Inversely Estimating Jacobian from Metabolomics Data

Inferring dynamics of metabolic networks directly from metabolomics data provides a promising way to elucidate the underlying mechanisms of biological systems, as reported in our previous studies (Weckwerth, 2011; Sun and Weckwerth, 2012; Nägele et al., 2014) by a differential Jacobian approach. The Jacobian is solved from an overdetermined system of equations as JC + CJT = −2D, called Lyapunov Equation in its generic form,1 where J is the Jacobian, C is the covariance matrix of metabolomics data, and D is the fluctuation matrix. Lyapunov Equation can be further simplified as the linear form Ax = b. Frequently, this linear equation system is ill-conditioned, i.e., a small variation in the right side b results in a big change in the solution x, thus making the solution unstable and error-prone. At the same time, inaccurate estimation of covariance matrix and uncertainties in the fluctuation matrix bring biases to the solution x. Here, we first reviewed common approaches to circumvent the ill-conditioned problems, including total least squares, Tikhonov regularization, and truncated singular value decomposition. Then, we benchmarked these methods on several in silico kinetic models with small to large perturbations on the covariance and fluctuation matrices. The results identified that the accuracy of the reverse Jacobian is mainly dependent on the condition number of A, the perturbation amplitude of C, and the stiffness of the kinetic models. Our research contributes a systematical comparison of methods to inversely solve Jacobian from metabolomics data.

Combined Metabolomic Analysis of Plasma and Urine Reveals AHBA, Tryptophan and Serotonin Metabolism as Potential Risk Factors in Gestational Diabetes Mellitus (GDM)

Gestational diabetes mellitus during pregnancy has severe implications for the health of the mother and the fetus. Therefore, early prediction and an understanding of the physiology are an important part of prenatal care. Metabolite profiling is a long established method for the analysis and prediction of metabolic diseases. Here, we applied untargeted and targeted metabolomic protocols to analyze plasma and urine samples of pregnant women with and without GDM. Univariate and multivariate statistical analyses of metabolomic profiles revealed markers such as 2-hydroxybutanoic acid (AHBA), 3-hydroxybutanoic acid (BHBA), amino acids valine and alanine, the glucose-alanine-cycle, but also plant-derived compounds like sitosterin as different between control and GDM patients. PLS-DA and VIP analysis revealed tryptophan as a strong variable separating control and GDM. As tryptophan is biotransformed to serotonin we hypothesized whether serotonin metabolism might also be altered in GDM. To test this hypothesis we applied a method for the analysis of serotonin, metabolic intermediates and dopamine in urine by stable isotope dilution direct infusion electrospray ionization mass spectrometry (SID-MS). Indeed, serotonin and related metabolites differ significantly between control and GDM patients confirming the involvement of serotonin metabolism in GDM. Clustered correlation coefficient visualization of metabolite correlation networks revealed the different metabolic signatures between control and GDM patients. Eventually, the combination of selected blood plasma and urine sample metabolites improved the AUC prediction accuracy to 0.99. The detected GDM candidate biomarkers and the related systemic metabolic signatures are discussed in their pathophysiological context. Further studies with larger cohorts are necessary to underpin these observations.