COLMARq: A Web Server for 2D NMR Peak Picking and Quantitative Comparative Analysis of Cohorts of Metabolomics Samples

Abstract

Highly quantitative metabolomics studies of complex biological mixtures are facilitated by the resolution enhancement afforded by 2D NMR spectra such as 2D 13 C– 1 H HSQC spectra. Here, we describe a new public web server, COLMARq, for the semi-automated analysis of sets of 2D HSQC spectra of cohorts of samples. The workflow of COLMARq includes automated peak picking using the deep neural network DEEP Picker, quantitative cross-peak volume extraction by numerical fitting using Voigt Fitter, the matching of corresponding cross-peaks across cohorts of spectra, peak volume normalization between different spectra, database query for metabolite identification, and basic univariate and multivariate statistical analyses of the results. COLMARq allows the analysis of cross-peaks that belong to both known and unknown metabolites. After a user has uploaded cohorts of 2D 13 C– 1 H HSQC and optionally 2D 1 H– 1 H TOCSY spectra in their preferred format, all subsequent steps on the web server can be performed fully automatically, allowing manual editing if needed and the sessions can be saved for later use. The accuracy, versatility, and interactive nature of COLMARq enables quantitative metabolomics analysis, including biomarker identification, of a broad range of complex biological mixtures as is illustrated for cohorts of samples from bacterial cultures of Pseudomonas aeruginosa in both its biofilm and planktonic states.

Differential metabolism between biofilm and suspended Pseudomonas aeruginosa cultures in bovine synovial fluid by 2D NMR-based metabolomics

“…By contrast, CFUs for all other cultures showed a uniform cellular phenotype with colonies of similar color, shape, and size. Metabolomics spectral data was normalized using a median ratio analysis method to adjust to global concentration differences between sample replicates 22 . This method is based on spectral peak volume differences between housekeeping metabolites among samples, which reflect variation in cell numbers and sample preparation.…”

“…Due to the high complexity of metabolism, many metabolites and pathways are not yet fully elucidated, especially in pathogens like P. aeruginosa which has high metabolic versatility 29 , 30 . In addition, in an untargeted analysis such as that utilized here, we are limited in identification of metabolites by the availability of NMR spectral database information 22 . Because the COLMAR database is mostly composed of primary metabolites, most of the metabolites identified in this work are primary metabolites.…”

DEEP Picker1D and Voigt Fitter1D: A versatile tool set for the automated quantitative spectral deconvolution of complex 1D NMR spectra

“…Recent applications of machine-learning methods, in particular of deep neural networks (DNN), have shown qualitative progress in the ability to deconvolute complex multidimensional NMR spectra (Li et al, 2022b). In the case of "DEEP Picker", training was exclusively based on 70 a library containing 5000 synthetic 1D test spectra consisting of 3 to 9 individual Voigt-shaped peaks with random amplitudes and positions amounting to a collection of training spectra with a wide range of spectral overlap (Li et al, 2021).…”

Absolute Metabolite Quantification Using Pure Shift NMR: Toward Quantitative Metabolic Profiling of Aqueous Biological Samples

“…Promising results have been reported in this field. − Semiautomated approaches have been recently demonstrated as efficient tools for performing a quantitative comparative analysis of complex metabolite mixtures . However, determining the best analytical workflow to quantify real intact samples (i.e., with the presence of macromolecules, highly overlapped spectra) from 2D in a reasonable amount of time is still an open question, as illustrated by the time needed to acquire 2D data with satisfying resolution in ref .…”

“…This approach faces three challenges that are (i) acquiring multidimensional data on biological samples in a limited amount of time, (ii) achieving sufficient sensitivity, and (iii) performing high-resolution and limiting artifacts to preserve the quality of 2D data, which amounts to squaring the circle. Promising results have been reported in this field. − Semiautomated approaches have been recently demonstrated as efficient tools for performing a quantitative comparative analysis of complex metabolite mixtures . However, determining the best analytical workflow to quantify real intact samples (i.e., with the presence of macromolecules, highly overlapped spectra) from 2D in a reasonable amount of time is still an open question, as illustrated by the time needed to acquire 2D data with satisfying resolution in ref .…”