Bio-spectroscopy and bio-chemometrics:
High-throughput metabolic profiling for integrative genetics

Harald Martens¹ & Achim Kohler<sup>2

1,2</sup>Centre for Biospectroscopy and Data modeling ^1,2CIGENE/ IKBM/ IMT, Norwegian U. of Life Sciences,^1,2MATFORSK- Norwegian Food Research Institute, Ås, Norway, ¹LIFE/U. of Copenhagen, Denmark


Modern analytical instruments are usually multi-channel. Still, surprisingly many are used in only one channel at a time - like grand pianos played with only one finger at time. Real-life systems biology requires a given system to be studied with respect to many different properties, and under many different conditions. Lots of molecules, interactions and mechanisms need to be monitored in lots of samples. But some patterns of them will be found to be more important than others. Modern metabolic profiling can yield hundreds or thousands of descriptors, but often at a high price, and the amount of data can be overwhelming. This lecture concerns how bio-spectroscopy and bio-chemometrics provide useful tools for chemists and biologists interested in high-throughput, high-relevance screening of chemical and physical variations in biological systems. Metabolites “known” to be important in a system should of course be monitored, but it is equally important to discover unexpected effects. It is not cost-effective to try to identify and name the individual metabolites before useful metabolic information is searched for. Often, minute but critical genomic or metabolomic changes can cause large down-stream consequences that are easier to detect. On this basis, it is possible to home in on the most informative metabolomic variations samples in the system. Multi-channel spectrophotometry allows screening of large series of samples with respect to lots of properties at high speed and low cost. Multivariate data modeling helps the scientist convert the measured spectral profiles into interpretable maps that facilitate the discovery of unexpected metabolic variations, and into reliable predictions of some of the known metabolites. In bio-spectroscopy based on e.g. Near- or Mid-Infrared (NIR, FTIR) spectroscopy, more or less intact samples can be measured at high speed, low cost and with little work – sometimes in vivo. The individual molecular species are not measured directly, like in chromatography or immuno-based methods. Instead, many different types of chemical groups – e.g. –CH₃, -NH₂ or –COOH, in different molecular environments – e.g. more or less specific fatty acids, proteins or carbohydrates - are quantified at different wavelengths of light. Since this high-dimensional profiling is so cheap, lots of samples can be screened. If the samples studied are representative and sufficiently different, the resulting set of profiles or spectra is often highly informative about systematic variations in the biological state of the samples. Bio-chemometrical “soft” data modeling of such large sets of spectral profiles, in graphically oriented software, can then identify various combinations of molecular groups that vary systematically with each other. This will also reveal how these patterns of change (latent variables) vary along with e.g. the environment, or with other measurements along the functional genomic chain from DNA via mRNA to the proteome, the metabolome and the many other ‘-omes’ accessible. The multivariate data modeling reveals the underlying patterns of variation that seem relevant and reliable. This may help the scientist infer what chemical compounds or interaction effects are involved, test established hypotheses about them statistically, and get new ideas about why they arise and how to change them. Just as importantly, the multivariate data modelling of these screening profiles will identify groups of samples behaving similarly under various conditions. On this basis, a particularly informative subset of samples can be chosen, representing all the samples initially screened, but few enough to allow slower, more expensive analyses to be employed. Prof. Lars Munck and coworkers at U. Copenhagen, Denmark pioneered the use of NIR bio-spectroscopy for screening and inductive discovery in applied systems biology. They have served as inspiration for some of the on-going Norwegian applications to be outlined here. In the present lecture, FTIR will be used as bio-spectroscopy tool. But other high-speed, high-dimensional profiling methods can also be used – e.g. Raman or autofluorescence spectroscopy, or NMR or Maldi-TOF MS spectrometry. For bio-chemometrics methodology, preprocessing by EMSC, multivariate calibration by cross-validated PLSR and cross-disciplinary overview by multiblock PCA will here be employed, for:

• Separating different sources of variation in FTIR spectra of individual human cells
  
  
• Linking phenotype to genomics in FTIR spectra of single-gene knockout yeast strains
  
  
• Linking phenotype to genetic pedigree based on millions of FTIR spectra of milk