ECCB 2002 Poster sorted by: Author | Number

Next | Previous poster (in order of the view you have selected)
Title: Gene expression profiles in disease: From gene expression in isolated cells to gene expression in whole tissues and vice versa
P60
Hoffmann, Martin (1); Pohlers, Dirk (2); Kinne, Raimund W.(2); Kroll, Torsten (3); Wölfl, Stefan (3)

mhoffman@pmail.hki-jena.de, Dirk.Pohlers@med.uni-jena.de, Raimund.W.Kinne@uni-jena.de, Torsten.Kroll@med.uni-jena.de, stefan@imb-jena.de
Hans Knöll Institute for Natural Products Research, Dep. of Applied Microbiology, Jena; Friedrich Schiller University, Experimental Rheumatology Unit, Jena; Friedrich Schiller University, Molecular Biology Group, FZL, Jena;

One of the most prominent applications of gene expression profiling is the comparative analysis of healthy and diseased tissues, aimed at the detection of the underlying pathological processes. An advantage of analysing whole tissue samples compared to isolated cell fractions is that cell communication and metabolic interactions between different cell types can be monitored. However, major drawbacks of the whole tissue technique have to be considered:

· Cell type fractions may change from sample to sample.
· Only a sub-fraction of a given cell type may undergo changes in gene expression.
· The contribution of different cell types to the pathological process may be variable/unknown.
· Changes in gene expression may result from single cell processes or from changes in cell-cell communication.

Measuring the expression profiles of each individual cell, or at least each individual cell type separately, is very difficult or impossible for currently available laboratory methods. Therefore, the question of gene expression profiling in whole tissues and purified cell fractions is presently addressed using simple mathematical models. Prior to analysis, the following general limitations have to be taken into account:

· A reliable detection of down-regulation is possible only for highly expressed genes. Referring to the common terminology, an n-fold increase is equivalent to a larger absolute change than an n-fold decrease. Experimental noise is more important for low-expressed genes.
· Whole tissue samples: A reliable detection of changes in gene expression is possible only for cell types making up a significant fraction of the tissue.
· Purified cell fractions: The successful purification of specific cellular sub-fractions must be ensured. This can best be verified by down-regulation of marker genes of contaminating cells.
· Gene expression profiling has to be validated by independent follow-up analysis.

The most basic and at the same time most important issue in the analysis of gene expression patterns from tissue samples is the determination of the relative fractions of the constituting cell types. Due to spatial variations and different pathological states, these fractions can vary considerably even within the same tissue sample. Immunohistochemical staining is thus of limited use and can provide only a rough estimate of cell type fractions. One potential way for determining these fractions by computational modelling is the reconstitution of tissue gene expression profiles from expression profiles of purified cell fractions. This concept can, however, be based only on those genes whose expression level is very robust and comparable in tissue and purified cell fractions. The aim of bioinformatics is thus to determine the cellular fractions of a given tissue sample using the robustly expressed subset of genes and identify candidate genes involved in cell-cell communication or pathological processes from the pool of genes disturbing an optimal reconstitution.

The proposed concept of computational reconstitution is tested using Affymetrix GeneChip® [1] expression data from inflamed rheumatoid synovial tissue and purified synovial cell fractions of adherent macrophages, adherent fibroblasts, and non-adherent cells. The relative cell fractions were estimated by means of weighted least squares (sum of squared errors) and least modulus (sum of absolute errors) procedures using top-down (successively eliminating genes), bottom-up (successively adding genes) or genetic clustering (iteratively exchanging genes) approaches to determine the fraction of robust genes that allow for a partial reconstitution of the synovial tissue profile. The employed algorithms are checked by analysing "synovial tissue" profiles artificially assembled from defined portions of the purified cell fractions and an increasing number of disturbing genes from the real synovial tissue. The results obtained indicate that a reconstitution of the synovial tissue gene expression profile from the purified fractions is indeed possible for about 7000 out of 12559 monitored genes. An important issue in this process is the relative weighting of the expression data, which span 3 orders of magnitude. So far the best results are obtained by weighting with the inverse square root of the expression level, compatible with previous reports on the scaling of the experimental standard deviation of expression profiles [1,2].
[1] http://www.affymetrix.com.
[2] D. M. Rocke, B. Durbin, JCB, Vol 8 (6), pp. 557-569, 2001