We developed a high-content image-profiling system to identify drug targets systematically. The multiparameter profiling method uses 501 morphologic parameters and a dataset of 4718 nonessential deletion mutants (). We used four well-characterized compounds that affect various cellular functions as test cases (). The drug target candidates were successfully screened into 2% to 8% from 4718 mutants on the panel, and the previously reported targets of the compounds and gene products associated with pathways that were functionally related to the target were enriched among the candidates. These results indicate that our high-content image-profiling method can detect targets of drug candidate and uncover their potential mechanisms of action.
The proposed method allows systematic, high-content screening independent of previous information about the drug candidate. The systematic approach with defined statistical criteria means that this method can theoretically be used not only to identify candidates of the drug targets but also to assess the effects of any condition that induces morphologic changes (e.g
., genetic mutations, nutritional starvation, and temperature shifts, among others). Further, because the analysis is statistics-based, the methodology can be expanded to higher eukaryotes if a morphologic phenotype database is available. The statistical power and utility of the method are enhanced by the multiple parameters that are extracted from high-resolution images. Examination of multiple intracellular activities and detailed phenotypic information for each mutant allow assessments of the effects of a drug on a wide variety of cellular functions. Our high-content image-profiling method also allows direct and systematic identification of drug targets from 4718 nonessential genes without any prior knowledge of about the mechanism of action of the candidate drug. This contrasts with conventional high-content imaging approaches, which focus on specific bioactivities (e.g.
, translocation of fluorescently labeled cellular targets between intracellular compartments) to assign the drug candidate into well-characterized groups 
. In addition, the off-target effects may be estimated by comparing results from low concentration of the drug treatment (reflecting target-specific effect) with that from high concentration (reflecting non-specific effects).
Compendium approaches that have been used with microarray technology to identify genetic targets comprehensively and systematically 
are similar approach to ours. In the compendium approaches, multiparametric profiles similar to that under the query conditions are surveyed from a collection of profiles under various conditions (drug, deletion mutants and etc). The profiles based on the relative gene expression abundance (~300 profiles of ~6000 genes) 
, the synthetic lethality (~1700 profiles of ~3900 genes) 
and the fitness (~80 profiles of ~3400 genes) 
are used as compendium data set. These compendium approaches used hydroxyurea as a test case to assess feasibility of target inference, and their results were consistent with that of this study (
and data not shown). These approaches and our high-content method may be complementary for target identification because of different screening criteria (i.e
., fitness versus morphology). Moreover, our approach may be particularly useful to identify targets of drugs that have no apparent effect on cellular fitness.
A key point for our high-content image-profiling is to detect the morphological changes. Therefore, if there are no or very little morphologic changes, it becomes difficult to identify drug targets. As expected, candidates with minor morphologic changes could not be detected. For example, when the drug target is the gene product possessing functionally redundant proteins (e.g., VPH1
and STV1 
) and/or is associated with relatively lower enzymatic activity (e.g., Rnr3p 
, Hmg2p 
, and Fks2p 
). In addition, in the case of a multifunctional protein (e.g., Fks1p 
), detection of the target was difficult even with marked morphologic changes 
. Nevertheless, as Hmg1p, which showed almost no morphologic change 
was identified as the target candidate of lovastatin, even with very weak phenotype it may be possible to identify the candidate, suggesting that strong statistical power is needed to detect target-related mutants which show weak phenotypes (e.g., vma1
). In order to better the statistical power, improvement in the reference data set of deletion mutants (e.g., preparation of the replicated data sets) would be required.
The current version of our high-content image-profiling system is limited to nonessential genes because morphologic phenotypes of mutants carrying deletions in essential genes are not available. Some genetic techniques may allow us to overcome this limitation. For example, heterozygous deletion mutants in diploid yeast may cause a haploinsufficient phenotype and enable essential genes to be screened 
. Alternatively, a comprehensive set of temperature-sensitive mutants for essential genes 
, including multiple mutant alleles of the same gene, may provide intragenic insights into the inhibitory mechanisms of compounds even if the target genes are multifunctional (e.g., FKS1 
). Finally, morphologic information from overexpression mutants may facilitate the identification of functionally promoted genetic targets 
To further improve our method, we can examine more image parameters from of other organelles. The new version of CalMorph can be used to examine 1111 morphologic parameters, including additional characteristics of various cellular components such as mitochondria, vacuoles and spindle pole bodies 
. The new version of CalMorph should improve the inference accuracy and generality of this approach for a variety of chemicals by expanding the available morphologic data from the mutant panel.