To have the ability to respond rapidly to abrupt and dramatic fluctuation in the external conditions is crucial for cell survival. Sudden changes in the external environment can perturb the internal system of the cells, disrupting cellular functions and preventing growth. Therefore, unicellular organisms such as yeasts have evolved complex adaptation mechanisms to cope with environmental stresses. One aspect of this cellular adaptation is the reorganization of genomic expression (Gasch and Werner-Washburne, 2002). The genome-wide stress-responsive genes in yeast have been discovered by DNA microarrays (Gasch et al. 2000
; Causton et al. 2001
). However, the complete network of the regulators of stress responses and the details of their actions remain to be elucidated (Gasch et al. 2000
In this study, we developed STFIA which integrates gene expression and TF-gene association data to identify stress TFs of six kinds of stresses. Some general stress TFs that are in response to various stresses and some specific stress TFs that are in response to one specific stress are identified. For example, STFIA found out the general stress TFs Msn2 and Msn4 (Cherry et al. 1998
) and the well-known heat shock TF Hsf1 (Cherry et al. 1998
), oxidative shock TFs Skn7 and Yap1 (Güldener et al. 2005
), osmotic shock TF Hot1 (Rep et al. 1999
), nitrogen depletion TFs Gln3 and Dal80 (Hofman-Bang, 1999
), and amino acid starvation TFs Gcn4, Met4, Met28 and Met31 (Cherry et al. 1998
). The ability to find out these well-known stress TFs validates the power of STFIA.
STFIA identified 24 distinct TFs (Arr1, Dal80, Dal81, Gat1, Gcn4, Gln3, Hot1, Hsf1, Ifh1, Ino2, Ino4, Leu3, Met28, Met31, Met4, Msn2, Msn4, Pdr3, Rpn4, Sfp1, Skn7, Stp1, Stp2, and Yap1) to be in response to at least one of the six stresses under study (see ). That is, a small number of TFs may be sufficient to control a wide variety of expression patterns in yeast under different stresses. Two implications can be inferred from this observation. First, the adaptation mechanisms to different stresses may have a bow-tie structure (Csete and Doyle, 2004). As shown in , the core stress TFs make up the ‘knots’ of a bow tie, facilitating the fan in of a large of variety of environmental stresses through signal transduction pathways and fan out of an even larger variety of stress-adapting proteins through activating stress-responsive target genes. Actually, approximately two-thirds of the yeast genome (about 4000 genes) is involved in responding to the changes in environment (Causton et al. 2001
). Second, there exists extensive regulatory cross-talk between different stress responses (see ). We found that heat shock, oxidative shock, osmotic shock, and acidic stress all can trigger the stress TFs Msn2, Msn4 and Pdr3, indicating these four stresses share a similar stress adaptation mechanism. Moreover, we found that nitrogen depletion and amino acid starvation both can trigger the stress TFs Gcn4, Gln3, Dal80, Stp1 and Stp2, indicating a cross-talk between the cellular responses to these two stresses. This is not surprising because both nitrogen depletion and amino acid starvation belong to the nutrient deprivation stress and could have a similar stress adaptation mechanism. The fact that different stress adaptation mechanisms share some, but not all, of their regulators suggests a higher level of modularity of the yeast stress response network (Segal et al. 2003
Figure 2. The transcriptional regulatory network of the yeast stress response. Environmental stresses are represented by octagons, stress TFs are represented by ellipses, and stress-responsive genes are represented by rectangles. Solid (Dashed) lines indicate the (more ...)
Figure 3. Regulatory cross-talk among different stress responses. The cellular responses to heat shock, oxidative shock, osmotic shock and acidic stress have significant regulatory cross-talk. They all trigger stress TFs Msn2, Msn4 and Pdr3. Besides, nitrogen depletion (more ...)
Step 3 of STFIA is very crucial for filtering out the low-confidence stress TFs that may be found only under some specific parameter setting. For example, 15 TFs (Arr1, Fhl1, Gcn4, Hsf1, Ifh1, Msn2, Msn4, Pdr3, Rap1, Rgt1, Rpn4, Rtg1, Rtg3, Sfp1, Stp1) are identified in Step 2 of STFIA as the heat shock TFs with the parameter setting T = 1, F = 2, and pthreshold = 0.01. However, nine (Arr1, Fhl1, Gcn4, Ifh1, Rap1, Rgt1, Rtg1, Rtg3, Stp1) of these 15 TFs are with low confidence because no known evidence shows that they are involved in responding to heat shock. As shown in , eight of the nine low-confidence stress TFs are eliminated in Step 3 of STFIA because they are not on the top 25% of the ranked list. The reason for choosing only the top 25% of the ranked list is as follows. If we choose a more stringent cutoff threshold, we will not have enough stress TFs (say five) for each stress to see whether there exists cross-talk among different stresses. However, if we choose a looser cutoff threshold, we may include some low-confidence stress TFs into our results. In , we show that when the cutoff threshold equals 25%, STFIA has the best performance in terms of the trade-off between the false positive and false negative rates to find out the high-confidence heat shock TFs.
Figure 4. Statistics of the performance of STFIA using different cutoff threshold. The false positive and false negative rates of STFIA using different cutoff threshold are shown. When the cutoff threshold equals 25%, STFIA has the best performance in terms of (more ...)