Systematic measurement of transcription factor-DNA interactions by targeted mass spectrometry identifies candidate gene regulatory proteins Hamid Mirzaeia,b, Theo A. Knijnenburga,c, Bong Kima, Max Robinsona, Paola Picottid,e, Gregory W. Cartera,f, Song Lia, David J. Dilwortha, Jimmy K. Enga,g, John D. Aitchisona, Ilya Shmulevicha, Timothy Galitskia,h, Ruedi Aebersoldd,e,1, and Jeffrey Ranisha,1 a

Institute for Systems Biology, Seattle, WA 98109; bDepartment of Biochemistry, University of Texas Southwestern Medical Center, Dallas, TX 75390; Bioinformatics and Statistics, Division of Molecular Biology, Netherlands Cancer Institute, 1066 CX Amsterdam, The Netherlands; dDepartment of Biology, Institute of Molecular Systems Biology, Eidgenössische Technische Hochschule Zürich, CH-8093 Zürich, Switzerland; eFaculty of Science, University of Zürich, CH-8006 Zürich, Switzerland; fThe Jackson Laboratory, Bar Harbor, ME 04609; gDepartment of Genome Sciences, University of Washington, Seattle, WA 98195; and hEMD Millipore Corporation, Billerica, MA 01821 c

Regulation of gene expression involves the orchestrated interaction of a large number of proteins with transcriptional regulatory elements in the context of chromatin. Our understanding of gene regulation is limited by the lack of a protein measurement technology that can systematically detect and quantify the ensemble of proteins associated with the transcriptional regulatory elements of specific genes. Here, we introduce a set of selected reaction monitoring (SRM) assays for the systematic measurement of 464 proteins with known or suspected roles in transcriptional regulation at RNA polymerase II transcribed promoters in Saccharomyces cerevisiae. Measurement of these proteins in nuclear extracts by SRM permitted the reproducible quantification of 42% of the proteins over a wide range of abundances. By deploying the assay to systematically identify DNA binding transcriptional regulators that interact with the environmentally regulated FLO11 promoter in cell extracts, we identified 15 regulators that bound specifically to distinct regions along ∼600 bp of the regulatory sequence. Importantly, the dataset includes a number of regulators that have been shown to either control FLO11 expression or localize to these regulatory regions in vivo. We further validated the utility of the approach by demonstrating that two of the SRM-identified factors, Mot3 and Azf1, are required for proper FLO11 expression. These results demonstrate the utility of SRM-based targeted proteomics to guide the identification of gene-specific transcriptional regulators.

C

ritical to understanding gene regulation is the ability to determine the composition of the regulatory complexes that assemble at specific genes and to determine how the composition of these complexes change in response to cellular, genetic, and environmental signals. Despite considerable efforts to address these key questions, the lack of methods for routine analysis of the ensemble of transcription factors (TFs) associated with specific transcriptional regulatory elements (TREs) remains a significant limitation. Current approaches for studying TF–TRE interactions include the EMSA (1, 2), protein binding microarrays (PBMs) (3), the yeast one-hybrid method (4), and chromatin immunoprecipitation (ChIP)-based methods (5, 6). Although each of these methods can provide information about TF–DNA interactions, their utility for routine analysis of TF–DNA interactions and complexes assembled at TREs is limited by the need for genetic engineering, protein detection reagents, and/or purified proteins. Notably, ChIP-chip has been used to systematically study the localization of most transcriptional regulators (TRs) across the yeast genome (7), but this tour de force required the creation and independent assay of 203 TR-specific, epitope-tagged strains. Another approach for studying TF–TRE interactions that does not require genetic engineering or protein detection reagents, and can readily provide information about the ensemble of TFs associated with a TRE, is DNA affinity purification followed by “shotgun” mass spectrometry (MS) analysis (8–10). This approach takes advantage www.pnas.org/cgi/doi/10.1073/pnas.1216918110

of the ability of MS to identify large numbers of proteins in complex mixtures, and, when performed in a quantitative manner, can identify specific TF–DNA binding events even in the presence of a high background of nonspecifically copurifying proteins. Although this is a powerful method to study TF–DNA interactions, the use of data-dependent acquisition routines during shotgun MS can limit sampling to the most abundant peptides eluting from the liquid chromatography (LC) column at any given time. In addition, the reproducibility of the method can be limited due to the stochastic nature of precursor ion selection before collision-induced dissociation (11). Although significant effort has been focused on developing methods for the identification of TF–TRE interactions, less effort has been exerted toward improving the reproducibility, limit of detection, and dynamic range of the platforms used to analyze the samples. However, selected reaction monitoring (SRM) permits highly reproducible measurements of a target set of peptides, with a low limit of detection and a wide dynamic range. SRM uses a series of quadrupole mass analyzers to detect and quantify specific, predetermined peptides in complex mixtures (11, 12). The reproducibility and wide dynamic range of SRM originate from its focus on a predefined set of targets irrespective of their abundance relative to the rest of the sample, and its sensitivity is enhanced by integration of targeted signals over extended periods of time. Additionally, more time is typically allocated to acquisition of data corresponding to the targeted analyte in SRM compared with shotgun MS experiments. Shorter duty cycles also contribute to SRM quantification accuracy by permitting acquisition of sufficient data points over the chromatographic elution profile of a specific peptide to support accurate peak reconstruction (13, 14). The favorable sensitivity, dynamic range, reproducibility, quantification accuracy, and relative insensitivity to chemical noise make SRM an attractive protein analysis platform for characterization of systems such as the TF–TRE interactome where most of the potential interacting proteins are known. To improve our ability to systematically study the ensemble of TFs associated with TREs, we developed an array of SRM assays that targets most known and putative proteins that function at RNA polymerase II (Pol II) TREs in Saccharomyces cerevisiae.

Author contributions: H.M., T.A.K., M.R., P.P., J.D.A., I.S., T.G., R.A., and J.R. designed research; H.M., T.A.K., B.K., M.R., P.P., G.W.C., S.L., and D.J.D. performed research; H.M., T.A.K., M.R., P.P., J.K.E., T.G., and R.A. contributed new reagents/analytic tools; H.M., T.A.K., M.R., G.W.C., T.G., and J.R. analyzed data; and H.M., T.A.K., B.K., M.R., G.W.C., T.G., R.A., and J.R. wrote the paper. The authors declare no conflict of interest. This article is a PNAS Direct Submission. Data deposition: The SRM coordinates are provided in Dataset S1. 1

To whom correspondence may be addressed. E-mail: [email protected] or rudolf. [email protected].

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. 1073/pnas.1216918110/-/DCSupplemental.

PNAS Early Edition | 1 of 6

SYSTEMS BIOLOGY

Edited by Mark Groudine, Fred Hutchinson Cancer Research Center, Seattle, WA, and approved January 4, 2013 (received for review October 12, 2012)

We selected three to five proteotypic peptides for each targeted protein and developed optimized LC and MS settings for sensitive and reliable detection of each peptide in a complex sample. The final set of assays, which includes 464 proteins, 1,639 peptides, and 9,834 SRM transitions (mass spectral coordinates to trace a peptide via its fragment ions), is the largest set of SRM assays deployed to measure a subproteome to date. Moreover, because the SRM assays are transferrable to other laboratories, and even to other instruments with minimal optimization (15), they will serve as a valuable resource for future systematic studies of the TF proteome. We first used the assays to probe a yeast nuclear extract, where we reproducibly detected and accurately quantified more than 42% of all target proteins without any fractionation. To demonstrate the utility of the assays to identify TF–TRE interactions, we applied them to systematically assess the ability of 222 known and putative TRs to specifically bind to 642 bp of the FLO11 gene regulatory region in cell extracts. FLO11 encodes a cell surface glycoprotein that is required for yeast to execute important developmental decisions, such as the transition from round cell growth to pseudohyphal or invasive growth in response to diverse environmental and internal signals (16, 17). The FLO11 gene regulatory region is one of the most complex in yeast, containing at least four upstream activation sequences (UASs) and nine repression elements that together span ∼2.8 kb (17). In addition, a broad array of regulators has been implicated in the control of FLO11 expression (16–23). Because of its central role in integrating cellular and environmental signals, the complexity of its promoter, and a large body of research on its regulation, the FLO11 gene was an attractive target for testing the utility of the SRM assays. The set of SRM assays identified 17 FLO11-specific binding events involving 15 TRs including the known FLO11 regulator Msn1. Comparison of the SRM dataset with results from motif scanning, PBMs (3), a large-scale ChIP-chip study (7), and functional data revealed prior supporting evidence for 11 of the binding events. Based on this analysis, as well as our network analysis of filamentation (20), we selected two candidates for follow-up studies: Mot3 and Azf1. These studies established activating and repressing roles for Azf1 and Mot3, respectively, in the control of FLO11 expression, and they validated the localization of Azf1 to a previously characterized UAS in the FLO11 promoter. The results highlight the effectiveness of the SRM-based approach to systematically measure the TF proteome, and, when used in combination with a DNA affinity purification step, to systematically screen for TF–TRE interactions.

analysis resulted in validated SRM assays for 464 proteins and 1,639 peptides (Fig. S1). Eighty-three percent of the proteins can be detected by monitoring transitions from more than one peptide, and ∼40% can be detected by monitoring transitions from five peptides. We then calculated the SRM transitions for all peptides containing isotopically heavy lysine (6C13, 2N15) or arginine (6C13, 4N15) at their C termini. Our final list contained 9,834 heavy and light SRM transitions (Dataset S1). Due to the large number of SRM assays, it was necessary to use a scheduled mode of measurement to increase the capacity for simultaneous monitoring of peptides (27). In addition, halogenated peptides as internal standards (H-PINS) technology was used to ensure reliable detection of targeted peptides during SRM analysis (28). H-PINS consist of a set of halogenated peptides which are used as standards to calibrate the retention times (RTs) of targeted peptides throughout the analysis. The use of H-PINS is important because during scheduled SRM, a change in peptide RT could lead to complete loss of signal if the peptide elutes outside of the window of time when it is expected to elute. All mass spectral coordinates required to use the assays are provided in Dataset S1. Quantifying the TF Proteome in Nuclear Extracts. We assessed the reproducibility, limit of detection, and false discovery rate (FDR) of the TF SRM assay (SRMA), as well its accuracy and precision of quantification, by monitoring all 9,834 target transitions (1,639 peptides, 464 proteins) as well as 3,720 decoy transitions (620 “decoy peptides”; decoy peptides were generated by reversing the sequences of randomly selected target peptides) in yeast nuclear extracts. Extracts prepared from cells grown in media containing either isotopically heavy or light lysine and arginine were mixed in a 2:1 ratio (light:heavy), digested, and analyzed in triplicate by LC-SRM in scheduled mode. SRM peptide identification was based on detecting three coeluting transitions for both heavy and light peptides with expected relative transition ratios within a projected elution window. We reproducibly identified 327 peptides representing 196 of the 464 (42%) target proteins, but only 1 decoy peptide, yielding an estimated FDR of <1% (Fig. 1A and Dataset S2). Reproducibility was excellent as 92% (327 of 355) of the

Results TF SRM Assay: Array of Validated SRM Assays Targeting the Yeast TF Proteome. To develop a system to systematically study TF–TRE

interactions, we first generated an array of definitive and quantitative SRM assays to unambiguously detect and quantify most known and putative proteins that function at or near RNA Pol II promoters in yeast. We initially created a list of these proteins using the Saccharomyces Genome Database (www.yeastgenome.org) and literature to include known and putative DNA binding TRs, coregulator complexes, chromatin remodeling and modifying complexes, and the general transcription machinery (Dataset S1). We then searched the PeptideAtlas database (www.peptideatlas.org) (24, 25) for mass spectral evidence for each target protein. For proteins with existing mass spectral evidence, target peptides were selected from the list of previously detected peptides using previously described selection criteria (12, 25). For proteins without existing mass spectral evidence, we either generated the data by enriching target proteins using protein or promoter DNA affinity purification followed by trypsin digestion and shotgun MS analysis, or we used the PeptideSieve algorithm to predict the best target peptides (26). We then chemically synthesized the peptides selected from PeptideAtlas and PeptideSieve and analyzed them by LC-SRM to determine the three most intense transitions to be used in our assays. For enriched proteins, peptides derived from tryptic digestion of the samples were used to generate the assays. This 2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1216918110

Fig. 1. Application of TF SRMA to detect proteins in nuclear extracts. To test the assay, yeast nuclear extract was monitored for the entire set of target proteins as described in the text. (A) Assay reproducibility. Venn diagrams depicting the number of peptides and proteins that were detected in triplicate analyses of the samples are shown. (B) Detection limit. A plot depicting the range of abundances for detected proteins is shown. (C) Quantification precision. The coefficient of variation for protein light:heavy (L:H) ratios was used as the measure of protein quantification precision. (D) Quantification accuracy was determined by fitting the protein L:H ratios to a Gaussian standard curve; see also Dataset S2.

Mirzaei et al.

Identification of FLO11 Promoter Binding Factors Using TF SRMA. We next used TF SRMA to systematically measure the propensity of 222 known or putative DNA binding TRs (Datasets S3 and S4) to interact with specific segments of the FLO11 promoter. As described above, the FLO11 promoter is an attractive target for testing the utility of the SRM assays because it contains a large number of cis-regulatory elements and numerous trans-acting factors have been implicated in its regulation (16–23). A previous study identified FLO11 regulatory elements by construction of FLO11:: lacZ reporter constructs containing either serial ∼200-bp deletions in the promoter or overlapping 400-bp promoter segments (17). We focused on the 642 bp contained in segments 5, 6, and 7 because it contains several cis-regulatory elements that are required for proper FLO11 expression (17), and a number of TRs, including Flo8, Sfl1, Ste12, Tec1, Msn1, and Gcn4 have been implicated in controlling FLO11 expression by acting on sequences within this region (17, 18, 23). However, of these TRs, only Flo8 and Sfl1 have been shown to directly bind to this region of the FLO11 promoter (18). Thus, we expected to identify some of these known TRs as well as previously unrecognized FLO11 regulators. To identify FLO11 promoter binding factors, we used quantitative TF SRMA to compare the abundance levels of TRs isolated by either FLO11 promoter affinity purifications or by a control purification. The control consisted of a set of five intergenic regions located between convergently transcribed genes. Proteins specifically enriched in the FLO11 promoter purifications could be identified based on their relative abundances from the two purifications during quantitative SRM analysis. Quantification was achieved by preparing two SILAC-labeled extracts (30), one each from cells grown to postdiauxic conditions on media containing isotopically heavy or light lysine and arginine. These are conditions that activate FLO11 expression (17). We performed two experiments for each FLO11/control comparison (Fig. S2). In the “forward” experiment, the FLO11 promoter segment was incubated with the heavy labeled extract and the control DNA was incubated with the light labeled extract, whereas in the “reverse” experiment the extracts were reversed. After combining the eluate from each FLO11 promoter purification with eluate from the control purification, the mixtures were digested and prepared for SRM analysis. Protein ratios and corresponding P values were computed separately for the forward and reverse experiments by combining all heavy/light transition pair ratios corresponding to the protein (Table 1, Datasets S3 and S4, and SI Text). The P values from the forward and reverse experiment for each protein were combined using Fisher’s method, and proteins with an FDR of <1% were called significant binding events. Of 666 potential TR–DNA interactions (222 proteins, 3 promoter segments) assayed, we detected 17 instances of preferential binding to a FLO11 Mirzaei et al.

Table 1. Proteins showing preferential binding to FLO11 promoter segments FWD

REV

Segment

Protein

n

Log ratio

CV

n

Log ratio

CV

FDR

5

AZF1 RTG1 CRZ1 DIG1 RTG3 AZF1 HAA1 MOT3 WAR1 USV1 MCM1 MSN1 UPC2 YAP6 HMO1 YAP1 MOT3

10 2 6 6 3 11 6 3 2 6 3 3 3 7 12 3 2

2.0 1.8 1.0 0.3 1.6 1.4 0.3 1.8 2.6 1.3 0.6 1.6 1.5 4.3 0.9 3.7 2.9

68 11 55 33 67 71 48 77 28 23 56 32 31 47 18 20 25

7 2 5 6 6 11 6 3 4 9 3 3 3 5 9 3 3

1.8 4.1 0.9 1.0 0.2 1.7 2.5 2.7 1.8 0.6 3.3 1.3 0.9 3.1 1.1 0.7 0.2

75 11 73 56 115 101 76 77 52 60 28 30 18 36 29 67 69

3.8E-08 6.2E-04 0.0012 0.0024 0.0068 1.5E-08 5.1E-05 3.1E-04 4.5E-04 7.8E-04 0.0010 0.0016 0.0048 2.2E-08 1.1E-05 1.7E-04 0.0059

6

7

Shown are all proteins with FDR of <1% and a binding preference [log2 (ratio) > 0] for a FLO11 promoter segment relative to control in both biological replicates, along with the number of transitions (n) contributing to each protein measurement and the coefficient of variation (CV) of the ratios. FWD, forward; REV, reverse.

promoter segment, involving 15 unique proteins (Table 1). Although most of the enriched TRs were not known to regulate the FLO11 promoter, the known FLO11 regulator Msn1 was significantly enriched in the segment 6 purification. This is direct, physical evidence for an Msn1–DNA interaction in this region; however, this segment overlaps precisely with previously described MSN1 responsive elements (19, 23). Taken together, two independent lines of evidence suggest that Msn1 regulates FLO11 expression by acting either directly or indirectly on segment 6. The results show that the systematic SRM assay is capable of identifying a known FLO11 TR as well as potential TR–FLO11 promoter interactions. The known FLO11 regulators Flo8, Ste12, and Gcn4 were quantified by TF SRMA, but they were not significantly enriched in the FLO11 promoter purification (Dataset S3). Potential reasons for not detecting enrichment of TRs that have been implicated in regulating FLO11 via segments 5, 6, or 7 could be that the proteins do not bind efficiently to the immobilized promoter segments under the conditions used in our study, they do not actually bind to the FLO11 promoter, or the peptides chosen for studying these proteins were not detected due to the presence of unanticipated modifications. We note that this is a proof-of-principle experiment designed to demonstrate the feasibility of using SRM to systematically measure the propensity of components of the TR proteome to bind to specific TREs. Further optimization of the cell growth and protein isolation conditions and the SRM assays could uncover additional specific binding events. Comparison of SRM Results with Other Systematic TR–DNA Interaction Studies. To guide our selection of TRs for functional validation

experiments, we looked for independent TR–DNA interaction data that supported the SRM results. Specifically, we compared the SRM TR–DNA binding data for each FLO11 promoter segment with TR–DNA interaction data from motif scanning, PBMs (3), and a large-scale ChIP-chip study (7, 31), which are in silico, in vitro, and in vivo techniques, respectively (Figs. S3 and S4, Dataset S5, and SI Text). Eleven of 17 interactions identified by SRM had independent support. Seven are supported by all available data [Azf1::5, Azf1::6, Mot3::6, Mot3::7, War1::6, Yap6::7, Hmo1::7 (numbers after colon indicate the FLO11 DNA segment)]. PNAS Early Edition | 3 of 6

SYSTEMS BIOLOGY

identified peptides and 94% (196 of 209) of the identified proteins were detected in all three analyses. Interestingly, seven of the detected proteins are not represented in PeptideAtlas (24), and four of these proteins were not visualized in a global protein quantification study by Western blotting (29). It is likely that the target proteins that were not detected either were not expressed under the growth conditions used, were below the limit of detection, or were not detected due to the presence of unanticipated modifications to their representative peptides. To assess our limit of detection, we checked the concentration of detected proteins in terms of copy per cell as determined in a Western blotting study (29). TF SRMA is capable of quantifying proteins that are expressed at a wide range of abundance levels, from <50 to 213,000 copies per cell (Fig. 1B). The precision of the analysis is reflected by the mean coefficient of variation (CV) of relative quantifications of all proteins measured by triplicate analyses, which was 27% (Fig. 1C). The relative abundance for each protein was expected to be 2:1; our assay resulted in estimated ratios of 2.15 (Fig. 1D). The measurement was therefore accurate to 7.5% [2.15 − 2)/2]. TF SRMA therefore provided reproducible, accurate, and precise quantification for 42% of the measured TF proteome in unfractionated nuclear extracts.

Mcm1::6 is supported by PBM, and motif scanning but not ChIP; Crz1::5, Rtg3::5, Msn1::6, Yap1::7 are supported by only one other source. Among the TRs involved in binding events supported by all available data, Mot3 and Azf1 are especially interesting, because they are found on two segments (5 and 6 for Azf1, and 6 and 7 for Mot3). Based on the results of this comparative analysis, and our previous network analysis of filamentation (20), Mot3 and Azf1 were selected for follow-up studies to determine whether they regulate FLO11 expression. To assess how TF SRMA compares with these other systematic approaches for detection of TR–DNA interactions, we performed an enrichment test for all pairs of data from our SRM study, motif scanning, PBM, and ChIP-chip (SI Text). Although we find a significant overlap of binding events determined by motif scanning and each of the other methods (P < 0.05), we did not find a significant overlap of binding sites identified among the other approaches. This analysis indicates that TF SRMA, PBM, and ChIP-chip provide largely distinct TR–DNA interaction information for the FLO11 regulatory regions studied here. However, integration of this information, as done in this study, can help guide identification of gene-specific TRs. Mot3 Is a Repressor of FLO11 Expression. Mot3 is an attenuator of signaling responses (32) that negatively regulates filamentous and invasive growth (20, 33). Our SRM evidence indicates that Mot3 binds segments 6 and 7 of the FLO11 promoter (Table 1 and Fig. S5). These findings are supported, in part, by motif scanning (segments 5–7) and ChIP-chip data (7) (the probe used in this study covers segments 5–7) (Fig. S6 and Datasets S5 and S6). Segment 7 has also been functionally identified as a promoter sequence that can repress FLO11 expression (17). Based on these results, we hypothesized that Mot3 functions as a transcriptional repressor of FLO11 upon binding to its promoter. We initially tested this hypothesis by assaying the expression of green fluorescent protein (GFP) driven by a copy of the FLO11 promoter, inserted into the genome adjacent to the intact FLO11 gene, in the presence and absence of a galactose-inducible MOT3 expression construct. Results are shown in Fig. 2A. We found that FLO11 expression is increased after growth on low-nitrogen solid media (SLAD), and this increase in expression is suppressed upon galactose induction of MOT3 expression. Furthermore, by microarray analysis, deletion of MOT3 leads to a threefold increase in FLO11 gene expression on SLAD media (Fig. 2B). These results confirm that Mot3 is a repressor of FLO11 expression. We next assayed the filamentation phenotype of a MOT3 gene deletion strain and a strain inducibly overexpressing MOT3 (GAL1MOT3). Because FLO11 is the prototype filamentation gene, we predicted that filamentation competence would correlate with effects on GFP expression. We found that deletion of MOT3 in cells leads to hyperfilamentation, as well as hyperinvasion on SLAD agar (Fig. S7), and hyperadhesion on YPD agar (Fig. 2C). We also found that galactose-induced MOT3 overexpression suppressed cell elongation and hyperfilamentation (Fig. S7). These data further support our SRM-derived hypothesis that Mot3 binds the FLO11 promoter, either directly or indirectly, and

negatively regulates its expression. Together with the SRM and ChIP-chip data, the results suggest that Mot3 is a repressor of FLO11 expression that functions by localizing to promoter segments 6 and 7. Azf1 Is an Activator of FLO11 Expression. Azf1 is a zinc finger TR that can affect expression of different classes of genes depending on the available carbon sources (34, 35). SRM data indicates that Azf1 can bind to FLO11 promoter segments 5 and 6 (Table 1 and Fig. S5), and these findings are supported, in part, by motif scanning (segments 5–7) and the ChIP-chip data (segments 5–7) (7) (Datasets S5 and S6). To further validate the Azf1 binding results, we performed segment-specific ChIP-qPCR studies on chromatin isolated from cells expressing a myc-tagged Azf1 protein after growing cells to postdiauxic conditions (Fig. 3A). Our ChIP study revealed that Azf1 localizes to segment 6 in vivo (P = 0.0001). It is possible that Azf1 does not localize to segment 5 in vivo under postdiauxic conditions, perhaps due to the chromatin environment, or that the ChIP assay is not able to detect Azf1 localization at segment 5. Because segment 6 activates FLO11 expression under both postdiauxic and exponential growth conditions (17), we hypothesized that Azf1 controls FLO11 expression via binding to segment 6. We initially tested this hypothesis by assessing the expression levels of FLO11 in wild-type cells and an AZF1 deletion strain after growing cells under postdiauxic conditions. We were unable to reliably detect a difference in FLO11 expression levels between the strains under these conditions. Subsequently, we assessed the expression levels of an integrated GFP reporter driven by the FLO11 promoter in either a wild-type strain or an AZF1 deletion mutant after transferring stationary phase cells to glucose-rich media. These are conditions that also induce FLO11 transcription due to a transient increase in cAMP levels and subsequent activation of the protein kinase A (PKA) pathway (36). Under these conditions, we found that induction of GFP expression was severely compromised in the AZF1 deletion strain compared with a wild-type strain and expression of Azf1 from a plasmid restored GFP expression to near wild-type levels (Fig. 3B). Together with our SRM and ChIP results, the data indicates that Azf1 is an activator of FLO11 expression that functions, at least in part, by binding to segment 6. Furthermore, the results highlight the utility of the SRM-based in vitro promoter binding assay combined with ChIP-qPCR as a way to pinpoint TR interaction sites in TREs.

Discussion The ability to decipher the ensemble of TFs associated with specific TREs and their dynamics is critical to understanding gene regulation. To address this issue, we have developed a set of SRMbased assays for the systematic measurement of the yeast TF proteome. SRM is an attractive technology for this purpose because it is sensitive, reproducible, quantitative, fast, and has a wide dynamic range. We first demonstrated the utility of these assays by deploying them to systematically detect a large fraction of the TF proteome in unfractionated nuclear extracts with high reproducibility and quantitative accuracy and precision (Fig. 1). Next, we applied the assay to identify potential regulators and their binding locations

Fig. 2. Mot3 is a repressor of FLO11 expression and an attenuator of adhesive growth. (A) MOT3 overexpression suppresses GFP expression driven by the FLO11 promoter. Data are shown for yeast cells (S288c background) containing either an empty vector, grown on synthetic low-ammonium solid media containing either glucose (SLAD, black line) or galactose (SLAG, red line), or a GAL1-driven MOT3 expression construct, grown on SLAD (blue line) or SLAG (violet line). (B) FLO11 gene expression intensity relative to wild-type (WT) yeast-form growth. Three biological replicates are shown for WT and mot3Δ. The mean expression ratio of WT SLAD is 4.6, and mean expression ratio of mot3Δ SLAD is 12.7. (C) MOT3 deletion strains are hyperadhesive on YPD (Σ1278b background for B and C ).

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1216918110

Mirzaei et al.

along ∼600 bp of the FLO11 promoter. TF SRMA identified 15 unique regulators involved in 17 TR binding events that were significantly enriched in FLO11 promoter purifications. In support of the effectiveness of the approach, one of these TRs, Msn1, was previously shown to regulate FLO11 expression, and the segment where Msn1 was enriched overlaps precisely with previously described MSN1 responsive elements (19, 23). Furthermore, there is supporting evidence from either ChIP-chip, motif scanning, or PBM data for 10 of the 16 additional binding events that we detected. We note that this is likely an underestimate of the effectiveness of the SRM-based approach because we only analyzed ∼20% of the known regulatory region and cell extracts were prepared from only one growth condition. Although we found a significant overlap of binding events determined by motif scanning and SRM, PBM, or ChIP-chip, we did not find a significant overlap of binding events identified among the other approaches. This result is perhaps not surprising given the numerous differences between the techniques—ChIP is an in vivo technique, whereas TF SRMA and PBM are in vitro techniques; PBM relies on the use of purified proteins, whereas TF SRMA and ChIP use crude extracts or whole cells, respectively; and each technique relies on different protein detection technologies. Furthermore, the ChIP probe used to detect binding events on FLO11 segment 5 in the Harbison et al. study (7), only overlapped with <50% of the segment. This could also account for the lack of overlap between some of the TF–DNA interactions detected by SRM and ChIP on segment 5. Overall, this analysis indicates that TF SRMA, PBM, and ChIP-chip provide largely distinct TR–DNA interaction information for the regulatory regions studied here, and it highlights the utility of having multiple approaches for detecting these interactions. It is worthwhile to point out that, whereas ChIP-chip and PBM can measure the binding of a single TF to all promoters in a single experiment, TF SRMA permits binding measurements of all TFs Mirzaei et al.

PNAS Early Edition | 5 of 6

SYSTEMS BIOLOGY

Fig. 3. Azf1 is an activator of FLO11 expression and localizes to the FLO11 promoter. (A) ChIP-qPCR analysis on chromatin derived from cells grown to postdiauxic conditions using myc-tagged Azf1 and PCR primers that amplify the indicated FLO11 promoter segments. (B) Azf1 induces GFP expression driven by the FLO11 promoter. FLO11-GFP expression was measured by FACS in a wild-type strain, an azf1 deletion strain, and an azf1 deletion strain transformed with an AZF1 expression plasmid after transferring stationary-phase cells to fresh glucose-rich media for the indicated times (S288c background).

at a single promoter in an experiment. As such, the approaches are complementary. TF SRMA required 11 MS analyses to systematically assay the TR proteome enrichment at each of the FLO11 TREs. The requirement for multiple runs is dictated by a need to balance sensitivity with the analytical capacity of the QQQ instrument. We expect even further improvements in throughput and sensitivity due to recent software and hardware advancements in QQQ instrumentation. Thus, TF SRMA provides an attractive method to efficiently assay the TF proteome without the need for antibodies or genetic engineering. We note that a limitation of the approach is its inability to detect proteins/peptides that are not targeted. This concern will be alleviated as improvements in instrumentation and continued SRM assay development will permit monitoring of increasing numbers of proteins. Also, SRM complements shotgun MS approaches that can be used to discover unanticipated proteins with the caveat that undersampling may limit the sensitivity and reproducibility of shotgun experiments (11). We prioritized the list of SRM identified factors for follow-up studies by looking for independent evidence of TF–DNA binding to the FLO11 promoter segments. Mot3 was a particularly attractive candidate for functional studies based on this analysis combined with previous studies that revealed Mot3 to be a repressor of both filamentation and invasive growth (20, 33). Consistent with our prediction that Mot3 functions as a repressor of FLO11 expression, we found that overexpression of Mot3 represses FLO11 expression when cells are grown on low-nitrogen, SLAD medium (Fig. 2A) and deletion of MOT3 leads to a threefold increase in FLO11 gene expression on SLAD medium (Fig. 2B). Mot3 has also been shown to function as a repressor of a diverse set of genes (32, 37). Although the mechanism is unclear, Mot3 appears to function in a chromatin-dependent manner, which can involve cooperation with the repressor Rox1 to facilitate recruitment of the Ssn6-Tup1 corepressor complex (38), or by imposing a requirement for the Rpd3L deacetylase complex (39). Similar mechanisms may exist at FLO11 given that Tup1 and components of the Rpd3L complex have been shown to affect FLO11 expression (21, 40). In addition, ChIP-chip results suggest that Rox1 can localize to segments 5–7, which overlap with the Mot3 binding segments that we detected and, like Mot3, Rox1 is a repressor of filamentation and haploid invasive growth (41). We also functionally validated the SRM binding results for Azf1 by showing that it localizes to the FLO11 promoter (Fig. 3A) and is required for proper expression of FLO11 when cells are exposed to fresh glucose-containing media (Fig. 3B). These conditions result in activation of the cAMP–PKA signal transduction pathway, which has been shown to impinge on the FLO11 promoter via the Flo8 activator and the Sfl1 repressor (17, 18, 23). Interestingly, the Flo8 and Sfl1 binding region in the FLO11 promoter overlaps with the Azf1 binding region that we identified. This raises the intriguing possibility that Azf1 may cooperate with Flo8 during glucosedependent activation of FLO11. Our observation that Azf1 binds to the FLO11 promoter and regulates its expression upon exposure to fresh glucose-containing media expands our knowledge of the array of regulators that integrate environmental signals at the FLO11 promoter. It will be interesting to determine whether Azf1 regulation of FLO11 expression is mediated by the cAMP–PKA pathway and whether Azf1 cooperates with Flo8 during glucose-dependent activation of FLO11. We note that whereas Azf1 binding to the FLO11 promoter was detected by both SRM and ChIP under postdiauxic conditions, AZF1 was not required for FLO11 expression under these conditions. Together, the results suggest that the functional requirement for the Azf1–FLO11 promoter interaction is condition dependent. Our identification of unique and potential FLO11 regulators, and localization of their DNA interactions, will permit refinement of models of the molecular interaction network controlling FLO11 expression. Future studies will be directed at testing predictions based on these models to clarify the mechanisms involved in the regulation of FLO11 expression. Furthermore, the methods

presented here, and applied to the FLO11 locus in yeast, can be readily applied to other TREs to identify key regulators of gene expression. Materials and Methods Yeast Strains. Yeast strains used in this study are listed in Dataset S7, and a discussion of the rationale for the use of the strains is presented in SI Text. Immobilized FLO11 Promoter DNA Affinity Chromatography. Immobilized promoter DNA affinity chromatography was performed essentially as described previously (8) with exceptions described in SI Text.

LC-SRM using a 4000QTrap in 18 batches (Dataset S8) as described in SI Text. SRM data were processed using ABI’s Multiquant software. Relative quantifications of peptides were based on the average ratio between the peak heights from the heavy and light transitions. Relative quantification of proteins, where multiple peptides were detected per protein, was based on the average light to heavy peptide ratios. SRM Analysis of FLO11 Promoter DNA Binding Purified Samples. SRM analysis was performed as described above with exceptions described in SI Text and Dataset S9.

SRM Analysis of Yeast Nuclear Extracts. Nuclear extracts from cells grown in light and heavy SILAC media were mixed in a 2:1 ratio, respectively, digested with trypsin, and purified. Peptides (2 μg) were analyzed in triplicate by

ACKNOWLEDGMENTS. This work was funded with federal (US) funds from the National Heart, Lung, and Blood Institute via Seattle Proteome Center Contract N01-HV-8179 (to R.A.); National Institute of General Medical Sciences Grant P50 GMO76547/Center for Systems Biology (to J.D.A., T.G., J.R., and I.S.); and National Institute of General Medical Sciences Grant K25 GM079404 (to G.W.C.). The study was also supported in part by SystemsX.ch, the Swiss initiative for systems biology via the projects, by the European Research Council advanced grant “Proteomics v3.0” (Grant 233226), European Union Seventh Framework Programme grant “Unicellsys” (Grant 201142), and a contract from the University of Luxembourg.

1. Garner MM, Revzin A (1981) A gel electrophoresis method for quantifying the binding of proteins to specific DNA regions: Application to components of the Escherichia coli lactose operon regulatory system. Nucleic Acids Res 9(13):3047–3060. 2. Fried M, Crothers DM (1981) Equilibria and kinetics of lac repressor-operator interactions by polyacrylamide gel electrophoresis. Nucleic Acids Res 9(23):6505–6525. 3. Zhu C, et al. (2009) High-resolution DNA-binding specificity analysis of yeast transcription factors. Genome Res 19(4):556–566. 4. Li JJ, Herskowitz I (1993) Isolation of ORC6, a component of the yeast origin recognition complex by a one-hybrid system. Science 262(5141):1870–1874. 5. Ren B, et al. (2000) Genome-wide location and function of DNA binding proteins. Science 290(5500):2306–2309. 6. Johnson DS, Mortazavi A, Myers RM, Wold B (2007) Genome-wide mapping of in vivo protein-DNA interactions. Science 316(5830):1497–1502. 7. Harbison CT, et al. (2004) Transcriptional regulatory code of a eukaryotic genome. Nature 431(7004):99–104. 8. Ranish JA, et al. (2003) The study of macromolecular complexes by quantitative proteomics. Nat Genet 33(3):349–355. 9. Mittler G, Butter F, Mann M (2009) A SILAC-based DNA protein interaction screen that identifies candidate binding proteins to functional DNA elements. Genome Res 19(2): 284–293. 10. Himeda CL, Ranish JA, Hauschka SD (2008) Quantitative proteomic identification of MAZ as a transcriptional regulator of muscle-specific genes in skeletal and cardiac myocytes. Mol Cell Biol 28(20):6521–6535. 11. Domon B, Aebersold R (2010) Options and considerations when selecting a quantitative proteomics strategy. Nat Biotechnol 28(7):710–721. 12. Picotti P, et al. (2010) High-throughput generation of selected reaction-monitoring assays for proteins and proteomes. Nat Methods 7(1):43–46. 13. Lange V, et al. (2008) Targeted quantitative analysis of Streptococcus pyogenes virulence factors by multiple reaction monitoring. Mol Cell Proteomics 7(8):1489–1500. 14. Wolf-Yadlin A, Hautaniemi S, Lauffenburger DA, White FM (2007) Multiple reaction monitoring for robust quantitative proteomic analysis of cellular signaling networks. Proc Natl Acad Sci USA 104(14):5860–5865. 15. Addona TA, et al. (2009) Multi-site assessment of the precision and reproducibility of multiple reaction monitoring-based measurements of proteins in plasma. Nat Biotechnol 27(7):633–641. 16. Lo WS, Dranginis AM (1998) The cell surface flocculin Flo11 is required for pseudohyphae formation and invasion by Saccharomyces cerevisiae. Mol Biol Cell 9(1):161–171. 17. Rupp S, Summers E, Lo HJ, Madhani H, Fink G (1999) MAP kinase and cAMP filamentation signaling pathways converge on the unusually large promoter of the yeast FLO11 gene. EMBO J 18(5):1257–1269. 18. Pan X, Heitman J (2002) Protein kinase A operates a molecular switch that governs yeast pseudohyphal differentiation. Mol Cell Biol 22(12):3981–3993. 19. Braus GH, Grundmann O, Brückner S, Mösch HU (2003) Amino acid starvation and Gcn4p regulate adhesive growth and FLO11 gene expression in Saccharomyces cerevisiae. Mol Biol Cell 14(10):4272–4284. 20. Carter GW, et al. (2007) Prediction of phenotype and gene expression for combinations of mutations. Mol Syst Biol 3(96):96. 21. Barrales RR, Jimenez J, Ibeas JI (2008) Identification of novel activation mechanisms for FLO11 regulation in Saccharomyces cerevisiae. Genetics 178(1):145–156. 22. Lambrechts MG, Bauer FF, Marmur J, Pretorius IS (1996) Muc1, a mucin-like protein that is regulated by Mss10, is critical for pseudohyphal differentiation in yeast. Proc Natl Acad Sci USA 93(16):8419–8424.

23. Gagiano M, Van Dyk D, Bauer FF, Lambrechts MG, Pretorius IS (1999) Divergent regulation of the evolutionarily closely related promoters of the Saccharomyces cerevisiae STA2 and MUC1 genes. J Bacteriol 181(20):6497–6508. 24. Deutsch EW, Lam H, Aebersold R (2008) PeptideAtlas: A resource for target selection for emerging targeted proteomics workflows. EMBO Rep 9(5):429–434. 25. Picotti P, et al. (2008) A database of mass spectrometric assays for the yeast proteome. Nat Methods 5(11):913–914. 26. Mallick P, et al. (2007) Computational prediction of proteotypic peptides for quantitative proteomics. Nat Biotechnol 25(1):125–131. 27. Stahl-Zeng J, et al. (2007) High sensitivity detection of plasma proteins by multiple reaction monitoring of N-glycosites. Mol Cell Proteomics 6(10):1809–1817. 28. Mirzaei H, et al. (2009) Halogenated peptides as internal standards (H-PINS): Introduction of an MS-based internal standard set for liquid chromatography-mass spectrometry. Mol Cell Proteomics 8(8):1934–1946. 29. Ghaemmaghami S, et al. (2003) Global analysis of protein expression in yeast. Nature 425(6959):737–741. 30. Ong SE, et al. (2002) Stable isotope labeling by amino acids in cell culture, SILAC, as a simple and accurate approach to expression proteomics. Mol Cell Proteomics 1(5): 376–386. 31. MacIsaac KD, et al. (2006) An improved map of conserved regulatory sites for Saccharomyces cerevisiae. BMC Bioinformatics 7(113):113. 32. Grishin AV, Rothenberg M, Downs MA, Blumer KJ (1998) Mot3, a Zn finger transcription factor that modulates gene expression and attenuates mating pheromone signaling in Saccharomyces cerevisiae. Genetics 149(2):879–892. 33. Jin R, Dobry CJ, McCown PJ, Kumar A (2008) Large-scale analysis of yeast filamentous growth by systematic gene disruption and overexpression. Mol Biol Cell 19(1):284–296. 34. Newcomb LL, Hall DD, Heideman W (2002) AZF1 is a glucose-dependent positive regulator of CLN3 transcription in Saccharomyces cerevisiae. Mol Cell Biol 22(5):1607–1614. 35. Slattery MG, Liko D, Heideman W (2006) The function and properties of the Azf1 transcriptional regulator change with growth conditions in Saccharomyces cerevisiae. Eukaryot Cell 5(2):313–320. 36. Gagiano M, Bauer FF, Pretorius IS (2002) The sensing of nutritional status and the relationship to filamentous growth in Saccharomyces cerevisiae. FEMS Yeast Res 2(4): 433–470. 37. Hongay C, Jia N, Bard M, Winston F (2002) Mot3 is a transcriptional repressor of ergosterol biosynthetic genes and is required for normal vacuolar function in Saccharomyces cerevisiae. EMBO J 21(15):4114–4124. 38. Mennella TA, Klinkenberg LG, Zitomer RS (2003) Recruitment of Tup1-Ssn6 by yeast hypoxic genes and chromatin-independent exclusion of TATA binding protein. Eukaryot Cell 2(6):1288–1303. 39. Sertil O, Vemula A, Salmon SL, Morse RH, Lowry CV (2007) Direct role for the Rpd3 complex in transcriptional induction of the anaerobic DAN/TIR genes in yeast. Mol Cell Biol 27(6):2037–2047. 40. Conlan RS, Tzamarias D (2001) Sfl1 functions via the co-repressor Ssn6-Tup1 and the cAMP-dependent protein kinase Tpk2. J Mol Biol 309(5):1007–1015. 41. Kadosh D, Johnson AD (2001) Rfg1, a protein related to the Saccharomyces cerevisiae hypoxic regulator Rox1, controls filamentous growth and virulence in Candida albicans. Mol Cell Biol 21(7):2496–2505. 42. Reiter L, et al. (2011) mProphet: Automated data processing and statistical validation for large-scale SRM experiments. Nat Methods 8(5):430–435.

SRM Assay Development. SRM assay development and implementation of H-PINS technology was performed essentially as described previously (12, 28) with exceptions described in SI Text. Decoy peptides were generated using the mProphet algorithm (42).

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1216918110

Mirzaei et al.