Recently published work from the Infrared beamline at the Diamond Light Source has the potential to speed up and simplify investigations in cell biology. The report, highlighted on the front cover of the 14th November 2012 issue of The Analyst, details a technique that has been developed to rapidly and effectively screen cell samples, replacing a key step in single cell analysis that is usually carried out by the human eye.
Right: Cover page published 14 November 2012. (Courtesy: Analyst)
The study of single cells is key to understanding many diseases and is therefore an important area for modern biology and medicine. The intrinsic brightness of synchrotron sources in the Infrared (IR) region of the spectrum enables the investigation of cellular biochemistry at a subcellular level. Imaging of tiny details at a fine spatial resolution (<5 thousandths of a millimetre) is time consuming and therefore limits the area and the number of samples that can be measured. In a cultured cell sample of hundreds or thousands of cells, an initial screening step is necessary to select the relevant regions for a high resolution measurement. This is normally done by selecting the individual cells with the eye, which can bias the experiment and make the results difficult to reproduce.
In collaboration with Dr. Josep Sule-Suso of Keele University and the University Hospital of North Staffordshire, the scientists on the ‘Multimode IR Imaging And Microspectroscopy’ (MIRIAM) beamline B22 at Diamond Light Source have developed automated hyperspectral image analysis algorithms to allow cell samples to be rapidly screened prior to measurement at a synchrotron source. The method uses an infrared focal plane array detector to rapidly image large areas of the sample, by creating a set of comprehensive image data, a so-called hyperspectral data set, which contains key information about the sample’s chemistry and morphology (the cell size, shape, optical density etc).
In general, these images are difficult to analyse due to the large amount of data they contain and there is a risk that essential information can be lost during the data reduction process. Complex multivariate analysis methods can be used to extract all the chemical information from the data, but the morphological information often remains unused. The algorithms developed on MIRIAM make best use of these hyperspectral image data sets by separating the chemical from the morphological information for further specific analysis steps.
Dr. Jacob Filik, the postdoctoral researcher on the MIRIAM beamline during this project and now with the Scientific Software group, describes the new data processing: “The key step is recognising that each cell can be identified as a separate entity in the hyperspectral image. If the cells can be accurately separated by eye, then common image manipulation algorithms should be able to achieve the same ends. Having identified the single cells, it is a trivial process to obtain the average spectrum of each, which reduces the data set from tens or hundreds of thousands of spectra to a few hundred. This smaller data set is more manageable to work with but the real advantage is that the morphological information, the size and shape of the cell, is not lost and can be used further down the analysis pipeline.”
Left: Location and separation of individual cell regions, images show the same sample area (a) Visible microscope image of human lung carcinoma (CALU-1) cells, (b) Infrared hyperspectral image reduced to show the total absorbance [each pixel represents one IR spectrum], (c) The cell separation process starts by identifying the individual cells. This is done by identifying local regions of maximum absorbance [red markers], (d) These markers are then used in a watershed transform to calculate the location of the cell-to-cell boundaries, (e) The separated cell regions are labelled with different colours making the identification of all the pixels [spectra] belonging to each cell simple. This image can also be used to determine the size and shape of each individual cell. (Courtesy: Analyst).
The next challenge for the team is to make the analysis algorithms easily accessible to the IR users at Diamond. “Although all the processing was done using open source software and image processing libraries, and the analysis process code was outlined in the supplementary material of the article, this data reduction method may be challenging to be implemented by someone new to programming.”, Dr. Filik concludes.
The MIRIAM research team hopes to make this in-house developed software more easily accessible by using the data analysis framework put in place by the Scientific Software group here at Diamond in collaboration with other facilities. The ‘Data Analysis WorkbeNch’ (DAWN, www.dawnsci.org) is a free, open source product that can act as a simple interface to complex custom algorithms such as those described here, and has the potential to speed up and facilitate the analysis of data from many fields, not just cell biology.
Morphological analysis of vibrational hyperspectral imaging data
J. Filik, A. V. Rutter, J. Sule-Suso and G. Cinque,
Published by Analyst online, 21 September 2012.
Science