Cells from each of the 4,848 distinct haploid yeast deletion strains, grown in rich media (YPD), were printed onto glass microscope slides coated with poly-L-lysine or concanavalin A (ConA) using a custom-built high-speed robotic arrayer that is normally used to manufacture DNA microarrays 28. Figure 1b shows an image of a cell microarray printed using this methodology. Each spot normally contains around 20-40 cells from a single deletion strain, as seen in Figure 1c using a standard microscope. Our preliminary data indicate that arrayed cells remain viable and physiologically normal after printing and washing, although cells are typically fixed for imaging purposes. A cell chip is analyzed using an automated fluorescence microscope to sequentially autofocus and image each spot.

As an initial proof-of-concept, we first performed genome-wide differential interference contrast (DIC) imaging to examine the effects of deleting each yeast gene on basic aspects of cellular morphology such as cell shape, size, budding pattern and clumping, from which we expected to find genes controlling fundamental cell growth processes. Systematic analysis of the haploid yeast deletion strain phenotypes on two slides (around 10,000 images) reveals that about 2,000 of the 4,848 strains exhibited atypical morphologies of varying degree. Two independent graders manually assigned numerical scores to phenotypes by severity, penetrance in the population, and type (large, small, elongated, round, and clumped 27, as well as polarized bud growth and pseudohyphal-like morphology). Control experiments were performed by constructing cell chips from known morphology mutants and wild-type strains, and grading these in the same grading scheme. Of these deletion strains, 381 (8%) were considered to have severe morphology defects (Figure 2) to a degree considered significant in the control experiments, with an estimated precision of 82% and recall of 26% (see Additional data file 1).

Genes deleted from strains with an observed morphology defect were often functionally diverse. Nonetheless, certain general functions were enriched, which we evaluated by comparing the sets of strains exhibiting a given phenotype with the sets of strains previously known to exhibit characteristic cell morphologies 27 or with sets of genes associated with distinct Munich Information Center for Protein Sequences (MIPS) functions 29 or Gene Ontology 30 functions. Elongated strains were enriched (p < 0.01, as calculated using FunSpec 31) for genes operating in nucleic acid metabolism, cell-cycle defects, transcription, and meiosis; large strains were enriched for transporter defects; round strains for cell wall, budding, cell polarity, and cell-differentiation genes; small strains for mitochondrial, carbohydrate metabolism, and phosphate-transport genes; and strains with polarized bud growth defects for budding, cell polarity, and filament-formation genes. Large and elongated strains significantly (p < 0.01) overlapped strains previously identified with these phenotypes during analysis of the homozygous diploid yeast deletion strains 27. Additional data files 2 and 3 summarize the morphological defects and functional enrichment, respectively.

Having established the typical morphology of each haploid deletion strain, we examined the primary morphological differentiation pathway in budding yeast - the response of the cells to the mating pheromone alpha factor during sexual conjugation.

Although this pathway is well studied 7, it has yet to be analyzed to completion. We reasoned that additional genes affecting the pheromone-response pathway, either directly or indirectly, could be identified by examining shmoo phenotypes when the deletion collection was treated with alpha factor. We treated the entire mating type a haploid yeast deletion collection with alpha factor, then constructed and imaged spotted cell microarrays from the treated and fixed cells. Two graders manually examined the cell images for the absence of shmoos, grading the images on a numerical scale. Consistency between graders was high, and no systematic grading differences were apparent (see Additional data file 1). Defects in shmooing were found in 142 strains; these either formed no shmoos or formed barely detectable shmoos in the imaged fields of cells (Figure 2b and see Additional data files 2 and 4).

These 142 strains represent a mixture of genes participating in the pathway and false-positive results in the large-scale screen, primarily arising from stochastic sampling of cells from image fields with limited penetrance of shmoos and from ambiguity in identifying cells with mating projections. In practice, we explicitly included ambiguous cases for later retesting, thereby increasing the false-positive rate of the genome-wide screen but decreasing the false-negative rate (see Additional data file 1). We filtered this set for reproducible shmoo defects by manually retesting the 142 strains twice via alpha factor addition (to both mid-log and late-log phase cells) and microscopic imaging; 54 of the 142 strains showed consistent shmoo defects. Of these strains, ten were previously identified as diploid or MATalpha strains in the MATa haploid strain collection (A. Tong and C. Boone, personal communication), which correctly appear insensitive to alpha factor in this screen. Removing these strains (accounting for all diploid and MATalpha contaminants) and six strains whose deletions could not be confirmed by PCR or whose phenotype failed to reproduce in a reconstructed strain (see Additonal data file 1) leaves 38 MATa haploid strains reproducibly defective in shmoo formation. Note that deletion of one of these genes, GPA1 , is thought to be lethal except in the presence of additional pheromone pathway mutations 32, implying either strain-specific viability or additional suppressor mutations in the library strain.

To validate the involvement of these genes in the pheromone-response pathway, we followed up the high-throughput cell-chip screen by conducting growth assays measuring the tendency of the strains to arrest growth upon pheromone exposure. We tested the complete set of 142 deletion strains (that is, the 38 reproducibly defectively shmooing strains and the remaining strains whose defects failed to reproduce) plus 271 additional deletion strains as controls with either normal shmooing (wild-type like, as determined from the cell microarray screen) or enhanced shmooing (marked by increased frequency of shmoos in the cell population), as well as strains deleted for 28 of the 41 genes previously known to be involved in the pheromone-response pathway. The positive controls are clearly differentiated from the normally shmooing strains in this assay (Figure 3), except for those deleted for five inhibitors of the pathway that arrest growth strongly in this assay (that is, they fail to show defective growth-arrest phenotypes). These include strains deleted for BAR1 , the protease that degrades mating pheromone 33, and DIG2 , which inhibits pheromone-responsive transcription 34.

Figure 3 shows that 30 of the 38 reproducible shmoo-defective strains fail to arrest growth upon exposure to alpha factor to an extent comparable to the positive controls. Lack of growth arrest agreed well with reproducible shmoo defects. These strains were defective in both shmoo formation and growth arrest, implicating the deleted genes in the pathway. An additional four MATa haploid strains first identified as shmoo defective, but not among the 38 reproducibly shmoo-defective strains, also fail to arrest growth upon exposure to alpha factor, implicating the deleted genes in the pathway (see Additional data file 4). Enhanced shmooing strains arrest even more strongly and appear systematically hypersensitive to the pheromone (Figure 3). Thus, the extent of growth arrest in this assay correlates well with the penetrance of shmooing across the populations of cells as measured with the cell-chip assay.

As two distinct, although correlated, phenotypes were assayed (growth arrest and shmoo formation), we expected to find genes defective in either or both pathways - a defect in both implicates the gene in the initial alpha-factor response pathway or in both downstream pathways, whereas a defect in only one implicates the gene in the corresponding downstream pathway (Figure 4). We first investigated mutants exhibiting both defects (termed ASD, for arrest and shmoo defective), implicated in pheromone detection and signaling.

Comparison with the known pathway (Figure 5, see also 7) shows that of the 41 genes previously known to be in the pathway, 15 were recovered in the cell microarray experiment. Examination of the remaining genes is revealing: ten genes are not represented in the deletion library (many are essential), 13 genes are inhibitors of the pathway and are thus not expected to be observed in either screen, as the deletion strains still shmoo, and the remaining three genes were missed for technical reasons related to image focus or low cell count. Thus, of the 31 genes expected to be found in this screen, 15 (48%) were correctly identified, including components of the receptor-coupled heterotrimeric G protein (STE4 , GPA1 ), the MAP kinase signal transduction cascade (STE20 , STE11 , STE5 , STE7 , FUS3 , FAR1 , STE50 ), and silencers of mating loci (SIR1 , SIR2 , SIR3 ). Recognizing that negative regulators may not be found in this screen raises the recovery rate to 15/18 genes, or 83%. Interestingly, strains with deletions of certain negative regulators such as HSL7 and DIG1 are shmoo defective and we correctly identify them in the screen (Figure 5).

Beyond the known signal transduction pathway, 15 genes were found that fail to shmoo and fail to arrest growth upon exposure to alpha factor. Examples include genes with clear functions in polarized growth (BEM4 and BNI1 ), as well as regulatory functions (the histone deacetylase SDS3 and the ubiquitin protein ligase UBR2 ). We separately validated the BNI1 and UBR2 involvement by reconstructing and retesting the deletion strains. Other strains were PCR-confirmed for the identities of the deleted genes, but not reconstructed (see Additional data file 1), and thus should be validated by strain reconstruction before confirming the definite involvement of these genes in the pheromone-response pathway. There is a general implication of genes affecting membrane properties, including PDR17 , which controls phospholipid synthesis/transport 35 and LAS21 , which controls glycosylphosphatidylinositol-linked protein transport/remodeling 36. Several genes encoding plasma-membrane transporters are identified (QDR2 and DAL5 ), as well as a cell-wall biosynthetic enzyme (YEA4) and mannoprotein (TIR3). Loss of any of these genes disrupts pheromone response, possibly indicating membrane properties feeding back into control of mating response, consistent with the important role of plasma-membrane reorganization in shmooing 37.

Such comparisons with known and literature-associated pathway components, as well as strain reconstructions, allow us to estimate the false-positive rate of this screen. Of the 40 original genes (after removing MATalpha, diploids, and strains not verified by PCR), 15 are known pathway components, three (BEM4 , ISY1 , SDS3 ) can be reasonably implicated in polarized growth and pheromone response from literature, two (BNI1 , UBR2 ) were confirmed with reconstructed strains, and two were eliminated as false positives in reconstructed strains. Therefore, 20 of 40 genes were confirmed and two were false positives, placing the false positive rate at 2/22, or 9%. Not considering the three components implicated from the literature raises this to 2/19, or 11%. Nevertheless, as with any genome-wide screen we advise reconstruction of deletion strains before unequivocally concluding that these genes are implicated in the pheromone-response pathway.

Finally, we identified strains defective in only one of the two assayed phenotypes, implicating the genes in downstream pathways. The set of strains that fail to arrest yet shmoo properly (termed AD for arrest defective) was functionally diverse as well as small (in part because only around 8% of the deletion collection was tested for growth arrest - we expect more such mutants given a complete screen for growth arrest). These strains were deleted for FMP35, RPL37B , YHL042W , YDR360W , YGL214W , PUB1 , PMT2 , TRX2 , SFK1 , MUP3 , SPL2 , and STM1. Conversely, eight genes were identified arresting normally yet failing to shmoo (termed SD for shmoo defective). Interestingly, six of these (VPS8 , VPS21 , VPS22 , VPS23 , VPS28 , VPS36 ) are involved in vacuolar protein sorting, with all but VPS8 and VPS21 specific to class E sorting and resulting in inefficient transport out of the endosome 38, suggesting a critical role of this system in shmoo formation (that is, downstream of pheromone signaling), possibly related to plasma-membrane reorganization 3739. The remaining two proteins are involved in polarized growth (ECM33 ) and transcriptional regulation (the histone acetylase EAF3 ).

All methods are described in full in Additional data file 1. In brief, cell microarrays were constructed by contact deposition of suspensions of yeast cells from the arrayed collection of S. cerevisiae haploid deletion strains (BY4741 genetic background; MAT a his3 Δleu2 Δmet15 Δura3 Δ) onto Con A 50 or poly-L-lysine-coated glass slides using a custom-built DNA microarray printing robot 28. In about 12 h, more than 100 slides can be printed, each containing the entire deletion collection as well as the isogenic wild-type parent strain as a control. Cell arrays may be used for imaging immediately after printing or stored at 4°C or -80°C, provided that the cells are printed with glycerol. Centrifugation enhances adherence of cells to the slide, permitting washing before staining and imaging. Cell images were collected via automated microscopy, using a Nikon E800 microscope with computer-controlled X-Y stage and piezoelectric-positioned objective, by scanning to the position of each spot, autofocusing, and capturing the image with a Coolsnap CCD camera (Photometrics, Tucson, USA). Images were stored in a custom cell microarray image database (Cellma) 51 for manual examination or automated image analysis. Using Perl scripts and custom MetaMorph (Universal Imaging Corporation, Sunnyvale, USA) journals, a full set of approximately 5,000 images can be collected from a slide in bright-field mode in less than 4 hours or for fluorescent images in around 10 h. Control cell chips, grading schemes, and morphology analysis details are described in the Additional data file 1.

To examine cell morphology after stimulation with alpha factor, each yeast deletion strain was subcultured into fresh YPD medium in 96-well Costar tissue culture plates (Corning, Corning, USA), grown for 36 hours at 30°C, centrifuged, and washed with YPD, pH 3.5, to inactivate the Bar1p protease 52. Alpha factor (350 μg/ml) was added to each sample well, a concentration measured by titration (as in 53) to induce shmoo formation in around one half of the cells in the majority of deletion strains (see Additional data file 1). Cells were incubated for 4 hours at 30°C, fixed with 3.7% formaldehyde for 1 hour at room temperature, washed with YPD containing 17% (w/v) glycerol, supplemented with 20 mM CaCl₂, 20 mM MnSO₄, then spotted onto Con A-coated glass slides. Slides were stained with DAPI, imaged by automated microscopy, and manually scored by two independent graders for extent of shmooing.

Four hundred and twenty-six selected deletion strains were grown overnight in YPD, centrifuged, and washed with YPD, pH 3.5 52. The cultures were split into replicate 96-well plates of YPD, with and without alpha factor at a final concentration of 25 μg/ml, maintaining cells at an optical density at 600 nm (OD₆₀₀) of around 0.2-0.5 52. Plates were incubated at 30°C for 10 h, recording OD₆₀₀ hourly from each strain. The slope of each growth curve was calculated from a plot of log(OD₆₀₀) versus time. The effect of alpha factor on the strains was obtained as the ratio of the slope from the untreated sample to that of the alpha-factor-treated sample. Average ratios were calculated from two or three independent assays.