We have described the development of a high-throughput platform capable of measuring transcriptional dynamics for many genes in large numbers of single cells exposed to the same experimental conditions. Existing methods for measuring gene regulation, such as DNA microarrays and chromatin immunoprecipitation, have excelled at identifying members of transcriptional networks and the interactions that occur between them at specific points in time. However, these techniques acquire data from homogenized samples derived from populations of cells and cannot provide accurate temporal and spatial resolution at a scale appropriate to characterize time-dependent transcriptional responses [28, 29].
Here we used a high-throughput cell-based assay to specifically transfect hundreds of constructs into adherent cell lines and tracked each fluorescent cell over 24 hours. We adapted the technique to use reporter constructs, so that changes in transcription from different promoters could be assessed at single-cell resolution. In its current form, 600 spatially distinct clusters of cells can be specifically transfected in a single chamber. This removes many sources of well-to-well variability that are frequently associated with high-content screening, such as differences in seeding density, ligand concentration and temperature.
Using the ECFP channel provides an internal control that enables single-cell normalization of the target promoter activity levels without requiring co-transfection. Tracking uninduced cells was also possible using this channel, such that we could investigate each cell's history during the experiment. As with the Venus reporter, destabilizing the ECFP protein by fusing it to a PEST domain could provide a more sensitive method of detecting loss of the plasmid and relating this to variability in the Venus channel.
Despite identifying several cell lines suitable for our platform, we determined that a subset of lines (such as F442-A, AtT-20, MCF-7 and HepG2) either did not grow as monolayers or could not be transfected at high enough levels to be used (Additional file 1, Table S1). Viruses offer an attractive alternative to transfection-mediated delivery of DNA into cells that are difficult to transfect, particularly primary cells. Previous studies using VSV-G lentiviruses  and more recently adenoviruses  have demonstrated the feasibility of the reverse infection approach that would be compatible with our live cell techniques. Expressing the reporter from a single integration site would provide the added benefit of removing reporter variability from differences in plasmid copy number between cells. Moreover, adenovirus systems can accommodate larger insert sizes compared to retroviral ones, such that our dual fluorophore vector could be re-used.
The measurement of single-cell expression offers much potential and has been studied using other methods, such as the dynamic proteomics approach  where coding sequences are randomly fused to an EYPF cassette in their endogenous context. The primary advantage to this technology is that expression can be studied from a single-copy integrant that would presumably retain the endogenous regulation of the gene at both the transcriptional and translational levels. However, a significant disadvantage of the method is that specific genes cannot be targeted using this method, as it generates a library of randomly integrated YFP cassettes that subsequently requires extensive screening. The maintenance of multiple cell lines, each harbouring a different YFP-tagged gene, is also a limitation that is addressed in our method, where the expression of hundreds of genes can be studied in parallel within a single well.
The Living Microarray platform can be applied to many systems. Expanding the number of synthetic reporters could provide parallel measurements for the activities of dozens of transcription factors in response to various stimuli, as has been shown with pooled measurements of sequencing and luciferase-based reporters [27, 31]. The technology could also be used to identify genomic regulatory sequences. Given the high-throughput nature of this technology, it would be possible to screen large regions of non-coding DNA for transcriptional activity by generating a library of reporters from tiled PCR products. In particular, this approach could be useful for validating putative regulatory variants from genome-wide association studies. Constructing a Living Microarray with known promoters could also profile complex processes, for example in studying the transcriptional changes that occur during the cell cycle [32–34], where temporal profiling of transcription in mammalian cells has been limited to microarray analysis using pooled RNA from synchronized cultures [35–37]. Such studies suffer from limited time resolution and fail to capture cell-cell variability or the heritability of expression patterns over successive generations. Moreover, chemical synchronization of cells may introduce spurious expression patterns in addition to only being partially effective . A recent report performed genome-wide siRNA knockdowns in reverse transfection microarrays and scored the phenotypes of the transfected clusters by computationally classifying time-lapse images of cytoplasmic and nuclear fluorophores during mitosis . In this way, hundreds of genes were identified as being required for the cell cycle. An attractive extension to these findings would be to construct a Living Microarray with reporters for these genes to precisely map their temporal regulation.
Gene expression is an inherently stochastic process, both within single cells and among cells of a population, owing to the many sources of intrinsic and extrinsic noise [39, 40]. For instance, clonal populations of mouse haematopoietic stem cells display heterogeneity in transcriptional profiles; these differences in genetically identical cells are responsible for each cell's propensity to differentiate into myeloid or erythroid lineages . Cell-fate decisions, such as pheromone switching in isogenic yeast, can also arise from heterogeneity that is more related to the individual cell's signal transmission and expression capacities, rather than on random fluctuations in gene expression . Nuclear receptor signaling is an ideal model for studying dynamic transcriptional processes, since the timing of induction can be tightly controlled. We found considerable variability in response at the single-cell level that is consistent with previous reports using single-cell measurements. For instance, reporter studies on the GH gene promoter revealed that when activated, only 25% of cells displayed a sustained response, while 50% showed only a transient one and 25% were not induced at all . Single-cell studies of the prolactin promoter in pituitary cells have indicated that the apparently stable transcription rate in a population may represent the overall sum of dynamically variable patterns of promoter activity among the individual cells [44–46]. This suggests that within populations of cultured cells (and perhaps normal tissues), there exists a mixture of cells that have different capacities to respond to external stimuli. Whether this heterogeneity reflects the presence of distinct subpopulations of cells, or results from normal fluctuations in cell physiology (possibly resulting from changes in cell cycle or metabolism) are questions that merit further investigation.