Single cell cytometry of protein function in RNAi treated cells and in native populations
© LaPan et al; licensee BioMed Central Ltd. 2008
Received: 14 December 2007
Accepted: 01 August 2008
Published: 01 August 2008
Skip to main content
© LaPan et al; licensee BioMed Central Ltd. 2008
Received: 14 December 2007
Accepted: 01 August 2008
Published: 01 August 2008
High Content Screening has been shown to improve results of RNAi and other perturbations, however significant intra-sample heterogeneity is common and can complicate some analyses. Single cell cytometry can extract important information from subpopulations within these samples. Such approaches are important for immune cells analyzed by flow cytometry, but have not been broadly available for adherent cells that are critical to the study of solid-tumor cancers and other disease models.
We have directly quantitated the effect of resolving RNAi treatments at the single cell level in experimental systems for both exogenous and endogenous targets. Analyzing the effect of an siRNA that targets GFP at the single cell level permits a stronger measure of the absolute function of the siRNA by gating to eliminate background levels of GFP intensities. Extending these methods to endogenous proteins, we have shown that well-level results of the knockdown of PTEN results in an increase in phospho-S6 levels, but at the single cell level, the correlation reveals the role of other inputs into the pathway. In a third example, reduction of STAT3 levels by siRNA causes an accumulation of cells in the G1 phase of the cell cycle, but does not induce apoptosis or necrosis when compared to control cells that express the same levels of STAT3. In a final example, the effect of reduced p53 levels on increased adriamycin sensitivity for colon carcinoma cells was demonstrated at the whole-well level using siRNA knockdown and in control and untreated cells at the single cell level.
We find that single cell analysis methods are generally applicable to a wide range of experiments in adherent cells using technology that is becoming increasingly available to most laboratories. It is well-suited to emerging models of signaling dysfunction, such as oncogene addition and oncogenic shock. Single cell cytometry can demonstrate effects on cell function for protein levels that differ by as little as 20%. Biological differences that result from changes in protein level or pathway activation state can be modulated directly by RNAi treatment or extracted from the natural variability intrinsic to cells grown under normal culture conditions.
RNAi has become a widely used method for conducting gene perturbation studies [1, 2]. Studies using RNAi to investigate gene function can be highly specific as well as scalable, including whole-genome screens [3–10]. While RNAi can be robust, there are challenges inherent to any RNAi experiment [11, 12]. These challenges arise from problems in predicting the specificity of an individual siRNA a priori, as well as directly linking the reduced target protein levels with the observed effects [13, 14]. Despite these challenges, RNAi is the most versatile and robust method for broadly testing gene function in most eukaryotes .
High content screening (HCS), or automated quantitative immunofluorescence, is being used to an increasing extent in the target validation stage of drug development, as well as in basic science [16, 17]. Image analysis is used to identify, quantitate and track multiple measures of individual cells [18–20]. Usually, these data are averaged, which is analogous to whole-well assays such as caspase activity or reporter gene expression. The advantage of HCS even in analyses at the whole-well level is that cells can be individually screened for inclusion in the well average according to parameters such as the health of the cell, stage in the cell cycle or activation state of a signaling pathway.
Single cell cytometry (or single cell analysis) has been used historically to analyze complex populations of cells, such as the study of differentiating immune cells by flow cytometry [21, 22]. Recently, the use of flow cytometry and single cell analysis has been applied to signaling pathways within cancer cell lines [23–26]. These studies highlight two advantages to flow cytometry-based single cell analysis. First, the ability to integrate the study of more than one cell-signaling pathway into an assay allows the classification of cancer cells according to perturbation responses, rather than static pathway activation levels. This better recapitulates the complex stimuli cancer cells encounter in vivo. Furthermore, advanced solid-tumor cancers are comprised of multiple subpopulations of cells, based on their genetic fluctuations and their interactions with host cells and tissues. Single cell analysis is capable of measuring changes within each of these subpopulations [25, 27–29]. The methods developed to analyze interrelationships between thousands of data points in each of multiple samples are advancing biological and pharmaceutical research beyond the study of single pathways, and towards the study of outcomes that arise from complex interactions between multiple pathways [24, 30, 31]. Such approaches are gaining favor because single-pathway studies show only limited correlations across cell lines or clinical samples, whereas the integration of multiple pathways and over complex sets of stimuli, enable more accurate understandings of cell signaling by addressing direct signaling as well as cross-pathway regulation .
We have used HCS to characterize the effects of genetic and chemical perturbations on cells by single cell analysis. We find that the wide range of protein expression levels in unperturbed cells is a significant complication for RNAi experiments, but that this complication can be addressed directly by analyzing such experiments at the single cell level. These methods allow the study of protein function by measuring the response in distinct subpopulations of cells in culture that result from stochastic variability of a target protein in a culture of cells.
To investigate the extent to which transfection and other sources of variability play a role in the analysis of GFP knockdown by an siRNA, we analyzed the same data at the single cell level. The data for one well where the siRNA was transfected at 3.13 nM are presented in Figure 1C. These data are reported as single cell values that correlate the expression of GFP with the amount of siRNA taken up on a per-cell basis for the GFP siRNA, which was labeled with Rhodamine. The siRNA shows a clear ability to reduce GFP levels. It can also be readily observed that the sample treated with the NTC siRNA includes a significant numbers of cells that intrinsically express low levels of GFP. The number of cells that express low levels of GFP in the control sample affects the mean level of GFP for the pool of untreated cells, and therefore, the extent of knockdown of the treated sample. While the effectiveness of the siRNA in reducing GFP levels is scored as roughly 60% using a whole-well analysis, gating on data within GFP-positive regions (analogous to the gating of cell populations in flow cytometry), the experimental effect is 10-fold, or a 90% reduction in high GFP-expressing cells, with 457 GFP expressing cells in the NTC siRNA treated sample, and 48 in the GFP siRNA treated sample. Heterogeneity of GFP expression is observed by other investigators. In particular, it has been noted that a variety of factors contribute to the perception of stochastic effects on protein expression levels when individual cells are examined. These effects contribute to the observed variability in lines developed from clonally expanded isolates , and from constitutive promoters .
We have examined the variability in intrinsic protein levels in cells, including a potential role for bias during the fixation and staining process, by dual-color staining (Figure 2D–2F). We observe that for many pairs, the extent of covariation is low, as observed for p53 and BRCA1 (r = 0.379) and Rb and HDAC3 levels (r = 0.353) in T47D cells. These data indicate that fixation and permeabilization do not play dominant roles in the distribution of antigen intensity. We do observe a higher correlation between c-MYC and BRCA1 levels (r = 0.814), in this particular case, the co-variation may reflect a biological correlation. In addition to the analytical comparison of co-staining patterns, we have examined several pairs of antigen staining to determine whether the staining patterns themselves are independent in cases where abundances are independent, by high-resolution confocal microscopy (results not shown). We find that in cases where two antigens are characterized in the same cells, the patterns are consistent for each antigen, regardless the level of staining for the second antigen. For example, the extent of nuclear staining and the degree of punctate staining observed were independent for the pairs examined (pairwise combinations of HDAC3, Rb and p53), further indicating that artifactual factors, such as uneven permeabilization or fixation, are not the cause of the wide range in antigen levels observed for these cells.
Figure 3C also shows the complexity of the AKT/mTor pathway when each sample is examined at the single cell level. That is to say, the effect observed in the whole well analyses, a decrease in PTEN results in an increase in phospho-S6 levels, would be expected to cause a negative correlation between these two proteins at the single cell level. Instead, a moderate positive correlation is observed, similar to the correlation observed in the unperturbed endogenous protein levels studied in Figure 2. Although often depicted as a linear pathway that leads to the activation of transcription, translation and metabolic activity, this pathway is under multiple levels of positive and negative feedback regulation of PI-3 kinase, AKT, mTor and ribosomal protein S6 kinase [53–55], which complicates strict correlations between any two points that are separated by one or more of these additional regulatory channels (discussed below). The extensive number of interactions between the AKT/mTor pathway and other regulatory pathways means that cells in culture are in a large number of discrete states. This has been observed elsewhere by our laboratory , and has been noted as a complicating factor in therapies that target this pathway, including those that target Her-2 NEU , PI3K and ERK [57, 58]. The use of single cell analysis to track multiple signaling states presents a valuable advance in the study of current and novel theapeutics.
While strong changes in average protein levels are required for experiments at the whole well level, analysis at the single cell level shows that STAT3 levels vary over a broad range under both control and STAT3 siRNA treatments. As such, comparisons between low and high STAT3 levels can be made by single cell analysis in cases where whole well differences are less dramatic. As an example, the effect of reducing STAT3 levels by RNAi can be analyzed in the experiment shown in Figure 4. Specifically, STAT3 is constitutively activated in many cancer cell lines, and reduction in STAT3 levels or activity (through RNAi or inhibitors of the JAK/STAT pathways) have been shown to result in growth arrest and apoptosis [60–62].
Proliferation inhibition is the result of the essential role of the protein in growth, but the induction of apoptosis or other forms of cell death has been ascribed to more complex interactions, such as oncogene dependency  or oncogenic shock . In these models, cancer cell death results from a release in apoptosis suppression mediated by the signal transduction pathway. The data in Figure 4A can be used to determine whether reducing STAT3 levels through RNAi results in a change in cell health that is distinct from cells with equivalent levels of STAT3 as a result of expression adjustments made during growth in standard culture conditions. This was done through comparing the distribution of cells through the cell cycle in the entire dataset versus a subset of cells where STAT3 levels were low in the STAT3 siRNA-treated sample. For the cells treated with the STAT3 siRNA, 22034 cells were analyzed in the complete dataset and 5471 cells were analyzed in the low-STAT3 population, as indicated in the annotation of Figure 4A. Samples were initially compared for DNA content, as a measure of cell cycle distributions. The data for the entire STAT3 siRNA-treated sample is shown in Figure 4B, and that for the low-STAT3 subset are shown in Figure 4C.
The data in Figure 4B shows that the cells are proliferating, with a significant number of cells in the G2/M phases of the cell cycle. For the low STAT3-containing cells (Figure 4C), the distribution shows a reduction in cells in these phases of the cell cycle, and a majority of the cells in G1. The cell cycle distribution is similar for the low-STAT3 cells of the NTC treated samples, but there are fewer cells and the histogram is not as smooth (not shown). Looking at subgroups with higher levels of STAT3, the proportion of cells in G2 increases somewhat.
In addition to measuring the effect on the cell cycle, the effect of lowering STAT3 levels through RNAi on cell stress and cell death can be determined as well. In this case, such effects would indicate a dependence on high STAT3 levels for survival, either through oncogene addiction or oncogenic shock, two models derived from observations that reduction in oncogene activity can induce cell death. Severe cell stress and cell death are manifest in several ways, including changes to the chromatin and nuclei [28, 65–67], which can be quantitated in image-based assays. In the present example, an effect of lowering STAT3 levels on viability would manifest itself as a change in nuclear size in the STAT3 siRNA-treated cells as compared to the NTC siRNA-treated cells. This has been noted in cytometry-based profiling studies [28, 68, 69], and is shown for SW480 colon carcinoma cells as a function of etoposide treatment in Additional File 1 (details are provided in the Methods section). Nuclear size as a function of DNA content is shown in Figure 4D and Figure 4E for the entire dataset and for the low STAT3-expressing fraction of cells, respectively. Nuclear size increases as a function of DNA content through the cell cycle, as shown for both panels, with increasing nuclear size as cells progress into S phase and again in late G2, immediately prior to anaphase. For the data shown in Figure 4, the relationship between DNA content and nuclear size is essentially identical for the NTC siRNA-treated sample (in blue) and the STAT3 siRNA-treated sample (in red) in both analyses, indicating that cells that have had STAT3 levels reduced through RNAi treatment are not undergoing cell death to a greater extent than control cells. If STAT3 levels were critical to the suppression of apoptosis or necrosis, the nuclear diameter of the cells with low STAT3 abundance would change, relative to the control cells. They would increase in size as a general function of cell stress [27–29], but would typically shrink and become more variegated in classical apoptosis [65, 70]. None of these changes are observed in any of the subsets. Taken together, these results suggest that STAT3 is playing an important role in the proliferation of SW480 cells, but is not acting as an essential oncogene through the suppression of apoptosis or necrosis, as would be evident if the nuclei were significantly different.
For the sample treated with the NTC siRNA, the amount of p53 per cell was used to divide the cells into groups, and the fraction of cells for each group as a function of adriamycin concentration is shown in Figure 5C. Cells with high levels of p53 are compared to cells with low levels of p53 for each dose of adriamycin. The data shows that cells expressing low levels of p53 are sharply reduced as adriamycin concentrations increase, and to an extent comparable to the reduction of the total cell numbers. This suggests that cells with low levels of p53 are particularly sensitive to adriamycin treatment. Since p53 levels can rise as a direct result of DNA damage, it is also possible that cells with low levels of p53 initially are actually stabilizing p53 and levels are increasing. Therefore, we sought to resolve these two factors in p53-mediated cell survival mechanisms.
We have addressed the question of whether adriamycin sensitivity is affected by p53 levels at the time of DNA damage by looking at how cells respond to treatment prior to when cell death and increased p53 levels are observed. In Figure 5D, the level of p53 in cells treated with siRNAs targeting p53 and the NTC control are shown for cells treated with increasing concentrations of adriamycin for 6 hours. At this time, we do not observe cell death (as reported by the number of cells per well), or a significant increase in average p53 levels (as shown in the figure). However, DNA damage can be observed in these cells in a dose-dependent manner, as determined by changes in DNA and nuclear morphology (data not shown). We have binned these cells by p53 level for each concentration of adriamycin treatment, and measured the levels of γ-H2A-x phosphorylation for each group, as shown in Figure 5E. Phosphorylation of this variant histone occurs in cells following DNA damage  independently of changes in p53 level or modification [72, 73]. The data shows that cells with higher levels of p53 show stronger DNA damage responses, as evidenced by increased γ-phosphorylated histone-H2A-x levels. Since these are independent responses to DNA damage, it suggests that cells with higher p53 levels may result from a stronger (or more activated) DNA damage response pathway prior to the onset of DNA damage itself, up until a point where the damage is beyond the ability of the cells to respond effectively (1.2 μM and higher concentrations). At high concentrations, significant cell death is observed for all cells (>85% cell killing), and no differential is observed between untreated cells and those treated with an siRNA targeting p53. At concentrations where the dependence of p53 status on adriamycin sensitivity can be observed, single cell analysis has been able to correlate the extent of the DNA damage response induction with p53 levels in cells where p53 levels have not been altered prior to DNA damage. The same general response can be observed in separate experiments using DLD-1 cells that have not been treated with any siRNA prior to that with adriamycin (i.e. no mock or control siRNA transfection at all), shown in Figure 5F. The cells are somewhat more resistant to adriamycin in general, possibly a result of no treatment with liposomes in a transfection, but the pattern of higher p53 levels correlating with higher DNA damage response is still evident.
We have applied the general concept of multiparametric single cell analysis to the use of RNAi, and to the relationship between protein levels and chemotherapeutic response. High Content Screening is becoming an important and general approach to biological and therapeutic studies. In addition to increasing the options available for cell-based assays in general, it is opening up new approaches to biological processes and drug development, such as cytological profiling [28, 29, 66]. Inherent in the latter approaches is the use of single cell cytometry to analyze complex patterns in cellular responses . We have generalized the use of single cell cytometry in several experimental systems and have found that it generally improves experimental analysis, and in some cases, enables challenging questions to be addressed directly. We have used single cell cytometry to address four biological problems: identifying the relevant cells in a knockdown of GFP, correlating the knockdown of PTEN with the increase in activity of pS6 kinase, the effect of knockdown of STAT3 on proliferation and death of colon carcinoma cells and the relationship between p53 levels and responsiveness to DNA damage (both as manipulated by RNAi and as occur intrinsically through standard cell culture conditions).
For RNAi screening in general, there are two applications of single cell cytometry that are potentially valuable. First is a general analysis of knockdown phenotypes by number of cells showing an altered phenotype, rather than average phenotypic change for the two samples. This approach is more in line with other distribution-based methods such as sectoring samples in flow cytometry, and can present data in more biologically-relevant way than reporting as percent-of-control (discussed below). Rigorous analysis of RNAi screening data is currently challenging [15, 74], and would benefit from clearer definitions of what constitutes a hit [9, 75]. The second benefit of single cell cytometry is the capacity to score cells as a function of the amount of siRNA effectively introduced in cells, as evidenced by the accumulation of the (non-functional) sense strand in P-bodies following efficient transfection. Transfection of siRNAs are frequently associated with off-target effects [76–78], particularly at concentrations typically used for library-based screening (>20 nM) [79, 80]. Off-target effects result in many false positive hits in RNAi screens, and impose a significant burden on the post-screening confirmation phase of a project . Transfection at low concentrations (< 10 nM) has been shown to reduce such artifacts, however library screening is performed with many siRNAs that have not been well-validated, particularly for off-target effects. Library screening typically involves higher concentrations because a productive screen requires that cells be reliably transfected, and some balance between the efficiency of transfection and a lack of specificity can be tolerated in the initial screen , as long as an effective strategy exists for demonstrating authentic gene-phenotype connections [81, 82]. Therefore, off-target effects resulting from high concentrations of siRNA transfections are a common and perhaps unavoidable complication of running siRNA screens. Reduced off-target effects have been associated with pooling or multiplexing siRNAs, particularly in highly complex pools such as are generated by enzymatic preparation of gene-specific siRNA pools (esiRNAs, ), at least in part because the concentration of any single siRNA is low.
Reverse-transfection, including the live cell array [7, 84, 85], is frequently used in functional screens. This format spots the siRNA (or dsRNA for screens in Drosophila cells) onto a surface prior to use with cultured cells, and therefore cells are not transfected at a specific concentration, strictly speaking. Single cell analysis can be readily performed on assays following reverse transfection, since these explicitly require image-based readouts. Selecting a subpopulation with consistent siRNA uptake for each siRNA is computationally intensive, and therefore would be difficult to use directly in the primary screen endpoint, but could be used to analyze data from a primary screen that uses a high content (image-based) assay. The siRNAs need to be labeled directly or co-transfected with a labeled siRNA, in order for siRNA levels to be quantitated. However, the benefit of this is that knockdown phenotypes can be scored for cells within specific thresholds of siRNA accumulation, and these thresholds can be adjusted as the data is reviewed, rather than during image analysis.
Scoring perturbations by fraction of responding cells (in the case of GFP knockdown at the single cell level) and by response magnitude as a function of target level (such as in the example of DNA damage response as a function of p53 levels) highlight important characteristics of biological samples, particularly in the development of human diseases such as cancer. Clinically important roles are played by minor populations within cell types, such as the growth of solid tumors through tumor-initiating cells (cancer stem cells) and the importance of regions within tumors that control angiogenesis and chemoresistance (the hypoxic core of cells within solid tumors). These properties can be observed in cell culture models, but this differentiation is lost in whole-well methods. Tracking effects of candidate therapeutics among rare cells or cells that have reduced proliferation rates can focus decisions on how well promising a strategy may be by limiting analysis to the cells that play the biggest role in disease progression.
A similar situation occurs with pathway analyses. An assay that measures a change in a complex pathway, such as the PI3K/AKT/mTor pathway, cannot help but exclude important factors that contribute to a diverse set of outputs. This heterogeneity may be as much a part of the discordance between target inhibition and clinical response as widely cited factors, such as tumor heterogeneity as a result of genetic instability. In both cases, variability in the cells that constitute a tumor enable a significant number of cells to escape death. The difference between these two scenarios is that genetic instability suggests a somatic evolutionary process, whereas signaling heterogeneity suggests that insufficient control of the pathway results in escape from a therapeutic. In such cases, single cell analysis could improve the search for combination therapeutic strategies. mTor activity is subject to multiple levels of feedback regulation [86, 87] and to cross-talk with other pathways, particularly the influence of amino acid and cellular energy levels on mTor activity . As such these influences would need to be measured in a multiparametric assay system, to track changes between two points in such a complex pathway. Taken together, the results presented here suggest that pathways that are quiescent (such p53 during periods of low DNA damage) or truly linear (such as activation of STAT signaling by JAK kinases) should show correlations between two points at the single cell level. This correlation could be used to validate results from RNAi experiments by providing a separate method of linking protein levels to pathway function.
Studies that integrate complex signaling interactions, as opposed to linear events within single pathways, are at the root of systems biology [31, 32], and are better able to characterize pathway states in their biological contexts. Such approaches are being shown to be of direct relevance to signaling in disease biology [25, 88]. HCS is a strong complement to flow cytometry as a method of single cell analysis because signaling pathway responses can be integrated with cytological dynamics, and as such will extend systems biology into areas such as cancer cell motility and invasion [27, 29, 89]. These approaches will lead to more innovative approaches to treating disease , including complex molecular studies which can be integrated with genetic and epidemiological studies that show subtle but important interactions between common disease loci.
Immortalized breast cell lines 184A1, and 184B5 were generously provided by Martha Stampfer (LBNL, Berkeley, CA). The C19 derivative of T/C-28a2 was developed and generously provided by Manas Majumdar (Wyeth Research, Cambridge, MA). MCF-7, T47D, MDA-MB-235, DLD-1, RWPE-1 were obtained from ATCC (Mannasas, VA). RWPE-1-GFP was developed by transduction of a lentivirus that encodes the GFP gene under the control of the CMV promoter. Media used for each cell line were according to instructions from the source.
Antibodies against γ-phosphorylated histone H2A-x, were obtained from Upstate Biotechnologies (Lake Placid, NY); antibodies against caspase-cleaved PARP and p53 were obtained from Cell Signaling Technologies (Beverly, MA). Fluorescent probes, including DAPI, and antibodies conjugated to Alexa dyes, were obtained from Molecular Probes/Invitrogen (Carlesbad, CA). Adriamycin, 16% paraformaldehyde, and Tween-20 were obtained from Sigma, Inc. (St. Louis, MO). siRNAs targeting p53 were obtained from Ambion, Inc (Austin, TX). Custom synthesized and unmodified siRNAs targeting GFP were obtained from Qiagen (Valencia, CA).
siRNAs were transfected as complexes with cationic liposomes from one of several manufacturers. For each experiment 3–5 commercially-available lipids were tested in a series of concentrations and siRNA:lipid ratios, according to manufacturers instructions. Transfections were four hours long and terminated by a change in media. For each cell line used in each experiment, the optimal lipid and siRNA:lipid ratios were determined using a test siRNA that targets GAPDH and GAPDH enzyme activity was measured for each condition, using the KD Alert kit from Ambion (Austin, TX). Optimal conditions were chosen as those that gave the greatest reduction in GAPDH activity when treated with the GAPDH-targeting siRNA, but minimal toxicity as identified by the NTC siRNA. Optimal conditions for each experiment are listed in Additional File 2.
Cells labeled as described in the figures were fixed with 4% paraformaldehyde, washed, permeabilized with 0.2% Triton X-100 and stained with 300 nM DAPI, primary and secondary antibodies and washed again. Antibodies were titrated for optimal imaging, and the lowest concentration that gave a highly-specific labeling of the antigen was used. Sources, dilution levels and fluorescence conditions are listed in Additional File 2.
Antigen intensities and localizations within cells following fixation and staining were imaged using an ArrayScan VTI (Cellomics, Pittsburgh, PA), using a 20 × 0.63 NA objective. Images were analyzed using the Target Activation and Compartmental Analysis image analysis applications from Cellomics. Cellular imaging was accomplished by first locating cell nuclei using DAPI-chromatin fluorescence and expanding the diameter of the nuclei to encompass the cytoplasmic region. Specific adjustments are required for each cell line. Cytoplasmic regions of neighboring cells were optimized in an iterative cycle of algorithm modifications and testing. Fluorescence intensity was captured and interpreted by one of several methods, typically mean fluorescence intensity per cell. Fluorescence measurements were well within the linear range of the image capture system (illumination, light filtering and detection using a cooled-CCD camera), so relative changes in protein levels could be made using relative changes in fluorescence between cells and samples. Non-specific detection is low, as shown in Additional File 3, and this enabled relative changes in protein levels to be determined from the fluorescence intensities.
Nuclear morphology was used as an indicator of cell health. Specifically, changes in nuclear area are indicative of severe cell stresses that result in necrosis or apoptosis. The identification of cells lethally treated with etoposide using nuclear area as an indicator of imminent cell death has been used by several laboratories in both classical apoptosis studies without the use of automation and in cytological profiling approaches. The change in nuclear area following treatment with an inducer of apoptosis is shown in Additional File 1. SW480 cells were treated with 5 μM etoposide for 24 hr, fixed and stained as described above. Cells treated with 10 μM and 20 μM etoposide showed similar distributions of nuclear area.
We have used HCS to examine protein levels within cells, and how these levels are manipulated by RNAi, at the single cell level. Data extraction and processing were performed using the statistical programming language R http://cran.r-project.org. Data from individual cells were extracted directly from the Cellomics' STORE database using a custom R function getCellData(), which uses a SQL query provided by Cellomics. The getCellData() function allows single cell data to be queried by well, row, column, or plate, one feature at a time, and is described in Additional File 4.
R scripts utilizing the getCellData() function are executed on a LINUX cluster. An auxiliary text file lists the plates and wells to be extracted, as well as the annotation associated with each well. The R script reads the auxiliary file 1nd replicates and merges the annotation with the single cell data as it is extracted from the database. Averaging, normalizations, and transformations are performed in R prior to export as a flat text file. Data is visualized either directly in R or imported into Spotfire for interactive analysis.
We thank Spyro Mousses, Kim Arndt and Robert T. Abraham for helpful comments on the manuscript. We thank John Morris, Bill Hussey, Tom Cannon and Ming Cui for information management support.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.