Visualize TFAPs

The "Visualize TFAPs" tab is useful for visually inspecting patterns of condition-specific regulation for multiple TFs across multiple microarray experiment conditions. It presents this TFAP information in color matrix displays that were originally developed for visualizing microarray data [8] and later adapted to visualizing TFAPs [9]. To generate such a display for the sporulation data and our chosen TFs, we first select the experiments by scrolling down to GSM87588 and highlighting all of the experiments through GSM87623. Next, we go to the list of all PSAMs at the bottom of the page and mark the check box next to each PSAM that were of interest from the "Sort PSAMs" pages. After we have finished selecting PSAMs, we may choose to alter some of the display options for the color matrix. We choose to cluster PSAMs by the similarity of their TFAPs by checking the "Cluster" check box next to "PSAMs." This way, PSAMs that have similar regulatory profiles will be displayed next to each other. We could choose to cluster experiments too by clicking on the respective box, but the samples are already in chronological order, which will be easier to interpret than clusters. Thus, we leave the "Cluster Experiments" box unchecked. We can also choose the clustering metric to be used. All of the metrics available in the Algorithm::Cluster package [10] are also available here. We find the default metric (the Pearson correlation) is fine for most purposes; however, the user may refer to the documentation for Algorithm::Cluster to learn about other options [10]. Lastly, we have two cosmetic options: "Contrast" changes the absolute t-value that receives the saturated colors in the color matrix and "Increments" changes how many intermediate color levels are between black (t-value = 0) and the saturated color. We leave these two options at the defaults. Finally, we push the "Show TFAPs" button to cause the Yeast Transfactome system to dynamically generate our color matrix result from the database contents.

We then receive a blue and yellow color matrix displaying the TFAPs for our selected PSAMs across our selected experiments.

Each colored box in this display represents the t-value for the fit of the predicted promoter occupancies for the TF (given its PSAM) to the measured expression values for the respective experiment. We can compare the color of the box with the legend colorbar to get an approximate t-value. We can also get a "tooltip" that displays the t-value if we mouse over individual squares. Finally, we can view the t-value, P-value, experiment and PSAM names, and affinity logo for a particular color box by clicking on it.

Here, we comment on the proper interpretation of the t-values in TFAPs. If the microarray experiment is performed using two color arrays and its values are reported as log-ratios, then the sign of the t-values can only be interpreted in the context of how the microarray experiment was designed and analyzed. Usually, such experiments have a control RNA sample and a test RNA sample. When the data is analyzed, the log-ratios are generated such that increased gene expression in the test sample relative to the control sample yields a positive log-ratio and decreased expression yields a negative log-ratio. For two color arrays, the interpretation of TFAP color matrices is similar. Yellow color (positive t-value) corresponds to a positive correlation between the predicted regulatory region occupancies (and thus predicted regulation) and expression log-ratios -- the TF target genes were upregulated in the test sample relative to the control sample. Conversely, blue (negative t-value) corresponds to a negative correlation between predicted regulatory region occupancies and gene expression in the test sample. Of course, if the test sample was called the control and the log-ratios were reversed, positive correlations would become negative and vice versa. However, in the case of microarray experiments that measure absolute expression, positive t-values always correspond to increased absolute expression and negative t-values correspond to decreased absolute expression for increasing predicted TF occupancy. For the series of experiments in our example, the control sample was always yeast cells growing in acetate media before transfer to sporulation media. Thus, yellow means that the TF targets were expressed more in the test condition than in acetate media and blue means that the target genes were expressed less in the test condition.

Many regulatory events occurring in response to both sporulation and changes in media conditions are present in this TFAP color matrix. We will highlight a few of the trends to demonstrate the utility of the Yeast Transfactome website. First, we notice a large block of highly similar TFAPs for PSAMs SFP1_SM, RAP1_YPD, FHL1_H2O2Hi, CHA4_SM, and HMS1_YPD, which have negative and gradually increasing t-values across the normal sporulation time course (GSM87588 to GSM87596) and positive and sharply increasing t-values upon each shift to YPD media (GSM87596 to GSM87623). Since these TFAPs change little over the normal sporulation time course, it is likely that these TFs are responding to the changes in media and not the sporulation transcriptional program. It is strange, however, that all of the TFAPs are so similar. By clicking on each PSAM name in this block (not shown), we see the affinity logo of the PSAM as well as all of the selected experiments sorted by t-value. We notice that the PSAMs for all but FHL1_H2O2Hi contain roughly the same consensus of YCYDNDCM. A closer inspection of the FHL1_H2O2Hi reveals that it also contains this consensus in reverse complement (which is an equivalent solution to the MatrixREDUCE algorithm, when it scores both strands). Thus, the reason that all of these PSAMs have roughly the same TFAPs across these conditions is that they have roughly the same sequence specificities and thus predicted promoter occupancies. This common consensus sequence represents the well-characterized specificity of Rap1p. This is both an interesting observation and a demonstration of a limitation of ChIP-chip experiments that is reflected in the Yeast Transfactome Database. In developing the database, we derived each PSAM from ChIP-chip experiments assaying each named factor. Thus, there are three possible explanations for the similarity of these PSAMs. First, the antibodies used for immunoprecipitation of the TFs may all have some affinity for Rap1p. Second, although unlikely, all of these proteins may have the same or similar sequence specificities. Finally, since these experiments are performed in vivo, the immunoprecipitated factor may be cross-linked to other proteins. While the chromatin that is used for the microarray experiment is the chromatin that was associated with the assayed factor, the actual DNA-binding specificity may reflect that of a physically associated, unknown cofactor. Therefore, these results may suggest that each of these TFs was associated with Rap1p in the ChIP-chip experiment conditions and was immunoprecipitated while cross-linked to Rap1p and its associated chromatin. Likewise, we see similar TFAPs and PSAMs for the paralogs Msn2p and Msn4p, but we also see an Msn2/4p-like PSAM for PHO2_Hi. Also, UME1_YPD has a Ume6p-like PSAM and a similar TFAP. It is possible that Ume1p and Ume6p exist in a complex since they have been found to be physically associated in vivo [11].

Since in this example we are interested in the the transcriptional basis of the commitment to sporulation, we are looking for transcription factors whose response changes significantly over the sporulation time course. From this perspective, the most interesting PSAMs are UME6_H2O2Hi, SUM1_YPD, RME1_YPD, MSN2_H2O2Lo, and MSN4_H2O2Hi, all of which are for TFs known to be involved in regulation of sporulation [12]. We will discuss Ume6p as an example. Ume6p is responsible for the repression of early meiotic genes during mitotic growth and induction of the same genes during sporulation [13]. Incidentally, a deletion of Ume1p also results in a derepression of early meiotic genes [14], which is consistent with Ume1p existing in complex and aiding the function of Ume6p. As expected from previous knowledge, predicted promoter occupancies for UME6_H2O2Hi correlated with increased expression during the first several time points of the sporulation time course (see image above, 2h to 6h). Friedlander et al. [2] reported that cells that had not yet undergone meiosis I (about 5h) returned to mitotic growth after transfer to YPD media. Consistent with that observation, based on the t-values for media transfers before the 5h time point (GSM87597 to GSM87605), the targets of Ume6p are downregulated within 20 min. after transfer to YPD media (see image above).