Toward automation of flow data analysis

Abstract

IN the last decade, there has been a growing demand for quantitative multiparametric characterization of clinical and biological samples (1). Fluorescence assays are those that have the highest versatility to study multifactorial processes in high-content analysis (see also Ref. 2 and the respective journal issue). High-content and high-throughput cytometric measurements usually provide data in matrix format (FCS file standard) for every event and every measured value. The conventional cytometric results such as mean fluorescence intensity, coefficient of variance, percentage of positive population are raised by analysis of FCS files with the aid of dedicated software. The same stands true, if data from image cytometry are acquired (3). However, with an imaging cytometer the number of measured and analyzed parameters can be substantially higher (4,5) than with standard polychromatic flow cytometry (6–8). In all cases usage of gating cascades and other analysis strategies needs manual work and is traditionally based on the judgment of an expert. The process is generally time consuming and often subjective (9). The number of two-dimensional dot-plots is increasing quadratically with the number of measured parameters. Similarly, the number of potential cell populations and subpopulations increases exponentially. As a consequence, highly multiplexed measurements result in a tremendous number of cell phenotype combinations and of two parameter dot-plots making manual analysis extremely laborious and confusing. Multiplexing is inexorably increasing by development of new dyes (e.g. Ref. 10) and new instrumentation and their combinations that allow increasingly multiplexed measurements for high-content analysis, e.g. by new lasers and detectors (11,12). Nowadays we become aware that the cellular systems (such as the immune system) are far more complex as we believed in the past (e.g. Ref. 13–18). More and more new cell types have been identified that have often relevance for clinical diagnosis. As a consequence, multicolor analysis becomes day-to-day routine in clinical diagnostic laboratories (6,7), and the demand in basic and clinical research increases to characterize specific cells as accurately as possible. Today, we face flow and image cytometry experiments with more than eight different markers. In such cases, eight dimensions of the data come only from the markers’ mean intensity. From slide-based cytometric and in flow imaging systems substantially more parameters are taken into account, such as area, eccentricity, perimeter, major, and minor axis length of the nucleus or of the cell (19,20). We can easily imagine that soon we will have 20 markers (resulting in 380 dotplots) or even more (21), and as we know from the past this number can only increase in the future (22). Therefore, manual analysis of data from highly multiplexed cytometry measurements is more art than science. Automation could be a solution to reach the complete information in an objective way. For these reasons a reliable automated approach to flow cytometric analysis is desirable, and this demand has been recognized already several decades ago. Different batch analyses for flow cytometry were already available in the 1980s (e.g. Consort 30 software) but the gates and markers were fixed and could not follow the usual fine intensity changes of scattered and fluorescent light and movements of the populations among the samples and series. Later on other programs for cluster analysis (e.g. Attractors) came to the market. These programs had the major improvement to move gates according to changes in population intensity and do multidimensional gating and clustering (23). Still they suffered from difficulties in reliably identifying minute subpopulations of cells in the proximity to very frequent cell population. Actually more flexible algorithms are under development. Automatic gating of flow cytometric data was published recently (24). This robust model-based clustering was applicable for analysis of proliferative and apoptotic populations and also phenotyping measurements. One pivotal requirement is to automatically detect and discriminate cell cluster. Required