© 2006 Nature Publishing Group The information in complete genome sequences1 and the identification and system-atic cloning of human cDNAs are providing us with the challenging opportunity to analyse the complexity of biological processes on a large scale, with the goal of reaching a more complete description of their molecular regulation. For this purpose, high-throughput techniques — such as protein analysis by mass spectrometry, or expression and transcription profiling by protein or DNA microarrays — have been developed and successfully applied to diverse biological questions. However, despite their great usefulness, these techniques cannot provide adequate temporal or spatial resolution and, most importantly, they do not directly show whether the identi-fied molecules have a functional role in the cellular process that is under investigation.Fluorescence-based imaging assays in intact living cells overcome these limitations because they can probe the function of macromolecules in their natural environment with exquisite and ever increasing spatial and temporal resolution2–4. Fluorescence-imaging assays, in principle, also have the potential to be applied to large-scale analyses, and simple assays have already been applied to cell-biological problems in high-throughput fluorescence-microscopy experiments5,6. Standardized reagents that interfere with cellular functions and high-throughput transfection methods, such as cell arrays7,8, are becoming available. Using these, large-scale fluorescence imaging at single-cell or even subcellular resolution can be combined with genome-wide RNA interference (RNAi) approaches9,10, small-molecule-based perturbations11 or overexpression strategies8 to reveal comprehensively the regulatory networks that underlie the functions of intact cells.Carrying out functional high-throughput microscope-based experiments that can pro-vide data for such systems-biology questions is currently still a challenge. It requires auto-mation and the precise coordination of vari-ous steps in an integrated workflow (FIG. 1). In this article, we highlight recent developments in high-throughput fluorescence microscopy and focus on future requirements for imag-ing and image analysis, which are currently two of the most important areas of new technology development.Assays: live cells are the futureThe key to any high-throughput fluorescence-microscopy approach is the development of an appropriate imaging assay that spe-cifically reads out the biological function of interest and is robust enough to provide reproducible, quantitative data using high-throughput image acquisition and analysis. Large-scale projects using high-throughput microscopy that have been reported so far almost exclusively used fixed-cell assays (see, for example, BOX 1 for a detailed example). Unfortunately, such endpoint experiments do not provide any temporal information and results might be misinterpreted if, for example, the final state of the examined cells is an indirect consequence of a number of sequentially occurring events. Experiments using live-cell assays (for example, REF. 10) and high-throughput time-lapse microscopy can overcome this problem, and they provide much more detailed phenotypic information than fixed-cell assays.However, the high-throughput automated fluorescence imaging of biological processes in living cells is currently technically chal-lenging (see later), and requires robust and simple fluorescent labelling techniques. The fluorescent reporters need to be specific, they must interfere as little as possible with the biological process being visualized and they must not perturb the global physiological conditions of the cells. Most importantly, the labelling and detection pro-cedures have to be quantitative and highly reproducible so that different experiments that have been carried out at different times can be compared, and so that image data can be evaluated using an automated phenotypic analysis — an absolute requirement for large-scale projects (see later). Numerous green fluorescent protein (GFP)-based protein markers for a vast array of cellular functions have been successfully devel-oped in the past years for manual, single live-cell experiments (see, for example, REFS 2,4,13–16). Monoclonal cell lines that stably express one or more of such proteins that are tagged with spectral variants of GFP will be one way to develop robust assays for high-throughput fluorescence-microscopy experiments in living cells. In addition to enabling the detection of the dynamic spatial distribution of the respective fusion proteins, GFP-based reporters have also been the basis for more sophisticated assays that monitor protein interactions, enzyme activities, pH, cyclic AMP or Ca2+ concentrations in living cells (reviewed, for example, in REFS 17–19). Although such reporters have so far only been used in single-cell experiments, they have the potential to be used in high-throughput fluorescence-microscopy experi-ments once the appropriate hardware and software for data acquisition and analysis are in place.INNOVATIONHigh-throughput fluorescence microscopy for systems biologyRainer Pepperkok and Jan EllenbergAbstract | In this post-genomic era, we need to define gene function on a genome-wide scale for model organisms and humans. The fundamental unit of biological processes is the cell. Among the most powerful tools to assay such processes in the physiological context of intact living cells are fluorescence microscopy and related imaging techniques. To enable these techniques to be applied to functional genomics experiments, fluorescence microscopy is making the transition to a quantitative and high-throughput technology.690 | SEPTEMBER 2006 | VOLUME 7 www.nature.com/reviews/molcellbioPERSPECTIVES© 2006 Nature Publishing Group FRET FRAP FCSCFPYFPABMulticolour 3D Time-lapsea Sample preparation b Image acquisition c Data handlinge Data mining and modellingd Image analysisHigh-throughput imagingHigh-throughput fluorescence-microscopy experiments comprise at least five independ-ent steps — sample preparation, image acquisition, data handling, image analysis and data mining combined with bioinfor-matic modelling — and these steps need to be tightly coordinated to sustain a rapid flow of work and data (FIG. 1). Therefore, any image-acquisition system that is used for large-scale experiments has to meet require-ments that are different from those for traditional