Disentangling Dynamic Changes of Multiple Cellular Components during the Yeast Cell Cycle by <italic>in Vivo</italic> Multivariate Raman Imaging

Cellular processes are intrinsically complex and dynamic, in which a myriad of cellular components including nucleic acids, proteins, membranes, and organelles are involved and undergo spatiotemporal changes. Label-free Raman imaging has proven powerful for studying such dynamic behaviors in vivo and at the molecular level. To construct Raman images, univariate data analysis has been commonly employed, but it cannot be free from uncertainties due to severely overlapped spectral information. Here, we demonstrate multivariate curve resolution analysis for time-lapse Raman imaging of a single dividing yeast cell. A four-dimensional (spectral variable, spatial positions in the two-dimensional image plane, and time sequence) Raman data “hypercube” is unfolded to a two-way array and then analyzed globally using multivariate curve resolution. The multivariate Raman imaging thus accomplished successfully disentangles dynamic changes of both concentrations and distributions of major cellular components (lipids, proteins, and polysaccharides) during the cell cycle of the yeast cell. The results show a drastic decrease in the amount of lipids by ∼50% after cell division and uncover a protein-associated component that has not been detected with previous univariate approaches. T cell cycle plays a pivotal role in reproduction of all living organisms. The molecular mechanisms of the cell cycle have been intensively studied over the last 40 years by genetic and molecular biological approaches using fission yeast and budding yeast as model organisms. The cell cycle, like any other cellular processes, is highly dynamic in nature and its mechanistic study requires time and space specificity as well as chemical specificity. Among various microspectroscopic methods developed to date, Raman spectroscopy has emerged as a potential bioimaging tool to meet these requirements with nondestructive, less invasive, and label-free characteristics. We have very recently achieved the first multimode timelapse Raman imaging of a single dividing Schizosaccharomyces pombe cell, in which Raman images for nine vibrational modes in the fingerprint region (800−1800 cm−1) were obtained simultaneously at different cell-cycle stages of the S. pombe cell. We showed that the concentrations and distributions of lipids and proteins varied in a concerted manner as the cell cycle proceeded. In that work, we used a univariate data analysis to construct Raman images. The univariate Raman image based on a single Raman band is readily obtained by integrating the Raman intensity under the band contour at each position in the image plane, followed by assembling these Raman intensities calculated at all positions to generate a two-dimensional (2D) map of the intensity distribution. Because Raman intensity is proportional to the concentration of a molecular species that gives rise to the Raman band, the Raman image so obtained displays a relative concentration map of the molecular species. Despite the simplicity of the principle, univariate Raman imaging often suffers from a limitation inherent to cellular vibrational spectra. The limitation arises from severely overlapped spectral information. An enormous variety of biomolecules can contribute to a cellular Raman spectrum and each of the biomolecules in turn exhibits many (broad) Raman bands, thus lowering chemical specificity. For example, the amide I band of proteins completely overlaps with the cisCC stretch band of unsaturated lipids, with both peaking at around 1655 cm−1. Therefore, the Raman image at 1655 cm−1 (see below) contains contributions from both lipids and proteins. This limitation indeed prevented us from fully utilizing the nine Raman images; only four of them that can be assigned unambiguously were discussed in detail. Without careful consideration on band assignments, univariate Raman images could lead to erroneous interpretations of the underlying biochemistry. Multivariate data analysis, such as principal component analysis (PCA), cluster analysis, and multivariate curve resolution (MCR, also known as nonnegative matrix factorization), is a more preferred approach, because it is capable of extracting maximum chemical information from complicated Raman spectra without a priori knowledge about spectral characteristics. In a multivariate data Received: March 27, 2012 Accepted: May 31, 2012 Published: May 31, 2012 Article