GlobalApproaches to Quantitative Analysis ofGene-Expression Patterns Observed byUseofTwo-Dimensional Gel Electrophoresis

A majordifficultyintheuseoftwo-dimensional proteinmaps toidentify andclassify celltypesistheproblem ofacquinng, selecting, and analyzingquantitative data on hundredsof proteinspots.Herewe usemethods of multivanate statistics toanalyzethedifferences amonga panelofhumancelllines, insomecasesinvolving quantitative dataonmorethan250 proteins.Principal-component and cluster-analysis tech- niquesshowthatthelinescanbe easilydistinguished, even by usingthe subsetof proteinspresentin all cells. A preliminary analysis oftheprotein changesbrought aboutby phorbolester-induced differentiation of the line U937 is included. Two-dimensional (2-D) electrophoresis of proteins has been used primarily to detect small differences in protein composition between pairs of related samples. It is useful in this context because sequential isoelectric focusing (with the sample in a medium containing NP-40 detergent and urea) and pore gradient electrophoresis of samples treated with sodium dodecyl sulfate (SDS) can resolve thousands of proteins (1), and thus one can detect changes in the abun- dance, modification, or rate of synthesis of proteins never before observed. In most comparisons of two samples, visual inspection alone allows detection of the alteration or disap- pearance of a single protein in a field of 2000. This accounts for the widespread usefulness of the technique without computerized data reduction. In at least three situations, however, visual analysis has proved inadequate: (a) inter-comparison of a large series of gels, to evaluate a panel of differences (where the observer's memory and notation methods are the major limitations); (b) interpretation of complex qualitative differences (such as the multiple charge changes seen in comparison of species); and (c) comparison of patterns in search of complex quanti- tative pattern differences (as in comparisons of gene-expres- sion patterns in cell types from one species). The first situation can be dealt with by using a computer system capable of quantifying and stretching gels into registration, combined with a simple database capability for remembering facts about spots (2). The second case has been attacked by using visual inter- pretation (3-5), often with discouraging results (3, 5). We will describe more-promising approaches to this problem in a separate paper. The third situation requires multivariate statistical anal- ysis of computer-generated quantitative data from two- dimensional gels, an initial exploration of which forms the