Spotted cell microarrays were first developed and applied to functional genomic screens in Saccharomyces cerevisiae 
and bacteria 
. To print yeast cell chips, we used a contact microarray printing robot to draw a microsample from a suspension of fixed cells in a 96-well microplate (the “source plate”) and deposit it on a poly-L-lysine coated glass slide. To print human cell lines, we used custom microarray pins with blunt tips and wide slots, and after experimenting with other adhesion protocols, we determined that printing biotin-decorated cells on streptavidin-coated glass slides ensured cell adhesion and reproducibility. An overview of the process is shown in . To demonstrate achievable array densities, we printed eight replicate HeLa cultures repeatedly onto a slide. A total of 4,608 spots were successfully printed on a single slide (), using eight spotting pins and a spot pitch of 400 µm. Chips of much higher density, exceeding 8,000 spots per slide, could be achieved by decreasing spot pitch ~10% and increasing to 12 spotting pins.
Overview of spotted cell chip process.
Contact microarray technology, typically used to print DNA oligonucleotides or cDNA sequences for use in RNA hybridization assays, is optimized around printing the smallest spots that can be consistently delivered. During the development of the yeast spotted cell microarray technique it was observed that better performance was achieved using microarray pins that had been “blunted” by repeated use in printing cDNA arrays. The blunted pins gave a larger spot size, a greater volume of medium deposited and, typically, a larger number of cells in the spot. However, the degree of blunting and therefore the quality of spots delivered varied widely among these well-used pins.
To adapt cell chips to human cells, we initially used the same microarray pins as in the yeast cell chip, and printed on poly-L-lysine (poly-K) coated slides. Early testing was conducted using the Jurkat T-cell leukemia cell line, as these cells are easy to grow in large quantities and a successful cell chip would provide a new platform for assaying suspension cells. We immediately observed that the larger human cells – which are typically spheroids 10–20 µm in diameter, many times larger than ovoid yeast cells that measure 3–8 µm on the long axis – did not print consistently onto poly-K coated slides, and that the inconsistency was in part attributable to how deformed the microarray pins were. To address this issue in a more systematic manner we acquired microarray pins with sharp or blunt tips in three sizes (Majer Precision MicroQuill 2000, part nos. 11077-1, 11077-2, and 11077-3). The 11077-1 pins were sharp and yielded spots <100 um in diameter, while the -3 pins had the largest blunt area and gave spots ~200 um across. Quantity of cell deposition was further improved by using custom pins, based on the 11077-3 form factor, but with a slot width of 0.030″ (76 µm) vs. the standard 0.015″ (38 µm). The smaller slot is only 2–3 cell diameters in width and may have induced shear effects and clumping as cells were loaded and deposited by the pins; these effects appear to have been largely mitigated by using the wider slots. The custom 11077-3 pin with 0.03″ slot width consistently delivers a spot ~200 µm in diameter and was used for all subsequent human cell chip prints.
Although we achieved regularity in spot sizes by selecting the appropriate microarray pins, the number of spots delivered was found to be highly dependent on the concentration of cells in the 384-well source plate. Depositing 50 cells in a spot ~1 nl in volume implies a concentration of ~50,000 cells/µl, or 106 cells in 20 µl suspension in each well of the source plate. However, during the time required to print ~100 samples onto each of 10–20 slides – roughly 30 minutes – the cell suspension settles into a loose pellet at the bottom of the well. In an effort to maintain the cells in suspension during printing, we increased the viscosity of print media using glycerol (15–50%) and sucrose (30–50%).
We tested the cell chip's ability to detect cellular state by inducing apoptosis in Jurkat cells. We grew the cells under normal tissue culture conditions. Separate cultures were treated with staurosporine, a potent inhibitor of protein kinase C and other essential cellular kinases, and fixed with formaldehyde after 1, 2, or 4 hours. Treated and untreated cells were collected in several wells of a 384-well plate at a concentration of >105 cells/µl and printed on poly-L-lysine coated slides such that each sample was printed several times on each of several replicate slides.
Immediately after printing, slides were imaged with transmitted light to analyze print quality; printed spots were discrete and typically contained 20–50 cells. Three slides were then probed for signs of apoptosis by immunofluorescence with antibodies against cleaved caspase 3, cleaved caspase 9, and cleaved PARP. Each slide was also labeled with a nuclear stain, and each spot was imaged using automated microscopy. Images of Jurkat cells immediately after printing, and of labeled cells after probing for cleaved caspase 3, are shown in .
15X images of an early print of Jurkat cells onto poly-L-lysine slides without the use of WGA-biotin. Left panels: DIC images acquired immediately after printing.
Although the immunofluorescence data supported the prototype cell chip's ability to detect cellular state, we observed that a significant number of cells – perhaps 10% of the cells in some spots — had shifted on the slide during the wash steps of the immunofluorescence protocol. This translocation is evident in when the pre-probe DIC images are compared to the fluorescent images (see arrows in figure). Given the relatively small numbers of cells in each spot, cross-contamination of even individual cells could dramatically reduce the dynamic range of the cell chip as an assay tool. To alleviate this problem, we tested an alternate adherence technique involving an adaptor molecule instead of relying on electrostatic interaction. After fixation, we decorated cells with a biotinylated lectin, wheat germ agglutinin (WGA-biotin), and printed the cells on streptavidin-coated slides. Under this protocol, increased print buffer viscosity is not required; cells were resuspended at 106 cells in 20 µl PBS (without glycerol or sucrose) in each well of the 384-well source plate and allowed to settle into a loose pellet. The microarray robot was calibrated to dip the pins into the pellet during loading. We printed WGA-biotin-labeled Jurkat and DG-75 suspension cells as well as trypsinized HeLa and HEK293 adherent cells onto replicate chips. Using the wide-slot pins and a standard wash cycle between loads, we observed neither cell clumping in the pins nor cross-contamination of cells into adjacent spots. After printing on a streptavidin-coated slide and allowing the print to dry, we observed no cell translocation throughout many repeated washing steps. The WGA-biotin/streptavidin slide combination was used for all subsequent prints.
To demonstrate the multiplex capability of the cell chip, we printed chips with both A549 non-small-cell lung cancer cells and HeLa cervical cancer cells. Each cell line was divided into three cultures: one treated with anisomycin (1 µM, 30′), one with TNFα (10 ng/ml, 60′), and one untreated control. Anisomycin, a translation inhibitor, activates (by phosphorylation) both the p38 and c-Jun N-terminal kinase (JNK) stress kinases. Among the effects of TNFα exposure are JNK activation and NFκB translocation to the nucleus. NFκB is maximally concentrated in the nucleus at about an hour after TNFα exposure 
, while JNK activation peaks after about 15 minutes and degrades to background levels about half an hour later 
. Multiple replicate chips were printed, each carrying all six conditions printed in multiple replicate spots.
Individual chips were probed for phospho-p38 kinase, phospho-JNK, and the p65/RelA subunit of NFκB. Each slide was counterstained with Hoechst 33342 nucleic acid stain and a high resolution image of each spot was captured in the corresponding fluorescent wavelengths. illustrates representative nuclear stain and immunofluorescence images from the chip probed for phospho-p38; the two spots show the increase in signal in HeLa cells treated with anisomycin compared to controls. The translocation of NFκB to the nucleus in response to TNFα in both cell lines was apparent in the images ( shows p65 translocation in HeLa cells).
Cell chips can be printed with multiple cell types, treatments, and time points on the same arrays, enabling extensive multiplexing of experiments.
We analyzed the set of treated spots from each cell line for an increase in signal relative to that cell type's control spots on the same slide by comparing the set of mean bias-corrected signal intensities of each set of spots (two-sample one-tailed T-test; see Methods
). Calculated p-values are shown in . JNK was phosphorylated in response to anisomycin treatment in both cell lines but TNFα-treated cells showed a weak response only in HeLa, consistent with the expected dynamics of TNF-induced JNK activation and deactivation. Anisomycin also activated p38 in HeLa cells, as expected, but surprisingly the response was much weaker in A549s; the p-value of 0.01 is not significant after multiple-hypothesis correction. P65 translocation to the nucleus is represented as an increase in nuclear signal in HeLa cells.
Measuring cellular response to drug treatments (Anisomycin, 1 µM, 30′; TNFα, 10 ng/ml, 60′) on A549 and HeLa cells printed on the same chips.
To explore the utility of cell chip technology for pathway analysis, we examined the chip's ability to recapitulate the interferon response of A549 cells. Exposure to interferon activates the JAK/STAT signal transduction cascade, resulting in up-regulation of interferon response genes, including dsRNA-activated protein kinase (PKR), the 2′–5′ oligoadenylate synthetases (OAS), and the Myoxovirus resistance gene (Mx) 
. We chose two assay targets, PKR and phospho-STAT1, to further validate the accuracy of the cell chip technology and to explore its dynamic range. We grew cells and exposed them to IFN-α (1000 U/ml) for 15 minutes or 18 hours before trypsinizing and formaldehyde fixing cells, along with untreated control cells. Technical and biological repeats were printed on the same slide (‘print 1’). At the same time, an equal number of cultures were prepared but stored in −20°C methanol for seven weeks before printing in an identical manner (‘print 2’). After printing, one slide from each print was immunoprobed for phosphorylated STAT1, counterstained with nuclear stain, and each spot was imaged at 40X. A second slide from each print was probed for PKR.
shows representative nuclear stain and immunofluorescence images of an individual spot from a slide probed for phospho-STAT1. Results of quantitative analysis are shown in . A fifteen-minute interferon exposure gave very strong signal for both prints, as well as weaker signal after 18 hours, indicating no loss of signal due to storage of fixed cells prior to printing. A third chip was probed after 30 days of storage at 4°; it showed less overall signal strength across all spots but also less variance, resulting in p-values nearly identical to the other two chips.
Measuring cellular response to interferon treatment and signal variance due to experimental methods.
The slides probed for PKR showed a small increase in signal at the 18-hour timepoint (; ), but, as expected, no response in the 15-minute samples. The small increase probably reflects the fact that PKR under these conditions is only up-regulated ~3-fold (measured by Western blotting; data not shown). This low relative signal may bound the sensitivity of the current state of this technology.
The presence of multiple cell lines and treatment conditions on the same slide can be exploited as internal controls for both experimental conditions and probes. In the anisomycin/TNFα chips, for example, p38 kinase showed lower response to anisomycin treatment in A549 cells than in HeLa. The anti-phospho-p38 antibody gave the expected response for the HeLa cells on the same slide, which serves as a positive control for the probe. JNK kinase responded to anisomycin in both the A549 and HeLa cells, which are drawn from the same population as those probed for phospho-p38, indicating the drug treatment worked properly. Therefore it is reasonable to conclude that the observed difference in p38 activation reflects a biological phenomenon rather than an experimental artifact. This property of multiplex controls can in principle be applied to much larger screens including a wider variety of experimental conditions and probes.