Identification of Upstream Kinases by Fluorescence Complementation Mass Spectrometry.

Protein kinases and their substrates comprise extensive signaling networks that regulate many diverse cellular functions. However, methods and techniques to systematically identify kinases directly responsible for specific phosphorylation events have remained elusive. Here we describe a novel proteomic strategy termed fluorescence complementation mass spectrometry (FCMS) to identify kinase-substrate pairs in high throughput. The FCMS strategy employs a specific substrate and a kinase library, both of which are fused with fluorescence complemented protein fragments. Transient and weak kinase-substrate interactions in living cells are stabilized by the association of fluorescence protein fragments. These kinase-substrate pairs are then isolated with high specificity and are identified and quantified by LC-MS. FCMS was applied to the identification of both known and novel kinases of the transcription factor, cAMP response element-binding protein (CREB). Novel CREB kinases were validated by in vitro kinase assays, and the phosphorylation sites were unambiguously located. These results uncovered possible new roles for CREB in multiple important signaling pathways and demonstrated the great potential of this new proteomic strategy.

Introduction

Protein kinase activities determine the phenotypes of all cells including cancer cells.[1] To dissect cellular signaling pathways, it is critical to identify direct relationships between kinases and their substrates. To date, several systems biology approaches have been applied to identify kinase substrates on a large scale.[2−4] However, there are few high throughput methods available to identify, given a specific substrate, the upstream kinases that phosphorylate it and regulate its activity. Attempts have been made to apply affinity pull-down methods in which an immobilized substrate is used to adsorb potential interacting kinases from cell lysates.[5,6] However, this strategy has considerable limitations because it is well established that many interactions between kinases and their substrates are weak and transient. Shokat and co-workers have devised clever substrate-trapping methods to convert transient enzyme–substrate interactions into covalent complexes.[7,8] However, this technique has been demonstrated only with a model peptide as the substrate and has not been applied to the discovery of novel upstream kinases for their natural substrates in living cells. To date, the upstream protein kinases directly responsible for thousands of phosphorylation events known from the phosphoproteome remain to be discovered.

Here we present a general proteomic strategy, termed fluorescence complementation mass spectrometry (FCMS), to identify the upstream kinases of a given phosphoprotein. This approach uses fluorescent protein fragments as originally developed for the bimolecular fluorescence complementation (BiFC) assay. BiFC is widely used for the visualization of protein–protein interactions in living cells.[9] In this technique, a fluorescent protein is split into two fragments, each of which is fused to one of two putative interacting proteins, which are then coexpressed in cells. Once the proteins interact, the fluorescent protein fragments are brought together to form a complex that emits fluorescence. Once formed, this fluorescent complex is stable in vivo, which makes BiFC a unique method to study transient or weak protein–protein interactions.[10,11] Here, we selected the fluorescent protein Venus and split it into N-terminal (VN) and C-terminal (VC) fragments, which were fused to a substrate and to a library of kinases, respectively. The substrate and kinases were coexpressed in mammalian cells. Any transient interaction between a kinase and its substrates will be stabilized by the association of the fluorescent protein fragments. Instead of detection of fluorescence signal and dependence on fluorescence-activated cell sorting (FACS) in BiFC, FCMS specifically isolates kinase–substrate pairsfor mass spectrometric analyses to identify, in a single experiment, multiple kinases capable of interacting with a single substrate.

We chose the transcription factor cAMP response element-binding protein (CREB) as a model substrate for the identification of its upstream kinases. CREB can be phosphorylated on multiple sites, and its activity is highly regulated by several known upstream kinases.[12] CREB binds to specific DNA sequences called cAMP response elements (CRE) and, once phosphorylated, interacts with coactivators to alter gene transcription.[13] Although several protein kinases are known to participate in the stimulus-induced phosphorylation of CREB, evidence for direct in vivo interactions between these kinases and CREB is often lacking. Furthermore, the complete repertoire of kinases capable of phosphorylating and regulating CREB activity is unknown. Given the important roles of CREB-mediated gene transcription in multiple disease states including neuropsychiatric disorders[14] and cancer,[15] this is an important question to explore. Finally, as a transcription factor, CREB lies at the base of multiple cellular signaling cascades, which allows us to study its upstream kinases without interference from downstream kinases.

Results

The flowchart of fluorescence complementation mass spectrometry (FCMS) for the identification of upstream kinases is shown in Figure . To screen for potential upstream kinases, we constructed an expression library that potentially expresses 559 human kinases fused to the Venus C-terminal fragment VC. We devised two strategies to improve the specificity. First, a mutant or truncated substrate that does not interact with the kinases is used as a control, allowing stable isotope labeling based on amino acids in cell culture (SILAC) to be applied to quantitatively measure the interaction of kinases with the wild-type versus the mutated or truncated substrate. Second, GFP nanobody, which recognizes only the intact VN–VC complex and does not bind either VN or VC fragment alone, was used to purify the kinase–substrate complexes for mass spectrometric analyses.

Figure 1

Flowchart of FCMS to identify upstream kinases. Wild type substrate (SUB) or mutated substrate (mSUB) in BiFC vector Myc-VN155 and human kinase cDNA expression library in HA-VC155 vector are cotransfected to different SILAC cells. Cells are combined, and protein complexes are immunoprecipitated with GFP nanobody. Proteins on beads are reduced, alkylated, and tryptic digested, and the resulting peptide samples are analyzed by LC–MS for both protein identification and quantitation.

One of the major challenges in affinity-purification based mass spectrometry (IP-MS) is nonspecific binding to the bait protein and the solid phase support.[16] Furthermore, the self-assembly of fluorescent protein fragments can contribute to high false-positives.[17] To address these issues, in the first step, we constructed CREB segments that differentially bind to upstream kinases and then applied quantitative proteomics to distinguish specific from nonspecific interactions. As shown in Figure a, CREB contains two glutamine-rich domains Q1 and Q2 and a kinase-inducible domain (KID) that together constitute the transcriptional activation domain. The leucine zipper and the basic region form the dimerization and DNA-binding regions of the protein. It was reported that KID is the main domain in which phosphorylation occurs.[12] Thus, we divided CREB into two fragments, one containing the Q1-KID (Q1K) region and the other the Q2-Basic domain-Leucine zipper (Q2L). Q1K should interact with the catalytic sites of most CREB kinases, while Q2L should not. The two fragments were cloned separately into the Myc-VN155 vector to generate the corresponding fusion proteins. The expression levels of Myc-Q1K-VN155 and Myc-Q2L-VN155 when transfected into HEK293T cells were analyzed by Western blotting using a Myc-epitope antibody (Figure b). Each was expressed at a similar level, which is critical for quantitative studies. To determine if the constructs were functional for BiFC assays, we cotransfected (1) Myc-Q1K-VN155 + HA-PKA-VC155 or (2) Myc-Q2L-VN155 + HA-PKA-VC155 into HEK293T cells, respectively. A construct coding for PKA was selected due to the well characterized interaction between this kinase and CREB. Strong fluorescence was observed in group 1 cells, while the fluorescent signal from group 2 cells was weak (Figure c). This functional validation confirmed an interaction between VC and VN and between PKA and Q1K, but not Q2L. Therefore, Q1K and Q2L could be used as positive and negative CREB fragments, respectively, in our FCMS experiment.

Figure 2

CREB fragment generation and validation. (a) CREB sequence structure and the constitution of CREB truncates. The glutamine rich domains Q1 and Q2 and the kinase-inducible domain (KID) constitute the transcription activation domain of CREB, while the basic region and the leucine zipper domain form the dimerization and DNA-binding region of the protein. Two truncated CREB fragments are Q1-KID (Q1K) and Q2-Basic domain-Leucine zipper (Q2L). Both were cloned into BiFC vector Myc-VN155 separately. (b) Expression of CREB mutants. Expression of Myc-Q1K-VN155 and Myc-Q2L-VN155 in 293T cells was examined with anti-Myc antibody. Note that Myc-Q1K-VN155 is 39 kDa and Myc-Q2L-VN155 is 41 kDa. (c) Fluorescence microscope imaging of BiFC assay of Myc-Q1K-VN or Myc-Q2L-VN cotransfected with HA-PKA-VC. (d) Western blotting against anti-Flag and anti-HA showed that when Flag-CREB-VN or HA-PKA-VC expressed alone, GFP nanobody did not capture either of them but instead captured the VN–VC complex.

Generation of a Kinase Expression Library

The second step was to generate the kinase expression library. The cDNAs coding for 559 distinct human kinases were transferred from entry clones (Addgene) to the HA-VC155 plasmid to obtain the library of HA-kinase-VC155 expression vectors. To test the quality of the generated library, we transformed the plasmids to Escherichia coli DH5a and cultured in a LB plate with selective antibiotic. The plasmids from a total of 24 individual colonies plus mixed library plasmids were digested and analyzed by electrophoresis. Different molecular sizes of released cDNAs were observed from individual clones while the library mixture yielded smeared cDNA bands, indicating that the expression clones were successfully generated with high complexity. The mixed expression clones were transfected into HEK293T cells, and kinases were immunoprecipitated with immobilized anti-HA antibody beads, followed by mass spectrometric analysis. We identified over 350 kinases from the transfected cells in a single LC–MS experiment (data not shown).

Specific Isolation of Kinase–Substrate Complexes

In a typical BiFC experiment, the formation of a BiFC complex is typically quantified by measuring the fluorescence intensities of the complexes and of the intact fluorescent protein in the same cells using either fluorescence microscopy or flow cytometry. The ratio of the fluorescence intensities can be used to confirm the formation of specific BiFC complexes. However, such BiFC-based screening requires the use of fluorescence-activated cell sorting (FACS) to isolate highly homogeneous cells in which BiFC complexes have formed. This procedure is laborious and not compatible with MS-based experiments due to the limited quantity of material that can be isolated via FACS. We sought to develop a highly specific method to capture kinase–substrate complexes for MS analyses. After screening multiple anti-GFP antibodies,[18] we identified an engineered single-chain anti-GFP antibody, GFP nanobody, that specifically recognized BiFC complexes, but not individual fragments. GFP nanobody (or nanotrap) was originally developed to isolate GFP, YFP, Venus, and Citrine with high affinity.[19] Compared with traditional antibodies, GFP nanobody has a smaller size and higher affinity and can survive in harsh conditions such as high salt, low/high pH, and high temperature.[19] To demonstrate its specificity, we expressed Flag-CREB-VN155/HA-PKA-VC155 and Flag-CREB-VN155/HA-PKA-VC155 in the cells separately and together. Then protein complexes were immunoprecipitated with GFP nanobody and analyzed by both Western blotting and MS. Both results indicated that the GFP nanobody captured fusion proteins only when the VN–VC fragments of Venus had reassociated (Figure d). Since the GFP nanobody-based isolation is based on a high affinity interaction,[19] we applied stringent washes to reduce nonspecific binding while preserving specific interactions. Different stringent washing conditions were examined. Western blotting and mass spectrometric analyses were applied to identify the optimum washing condition that includes the use of RIPA buffer, 5 M NaCl, and 500 mM glycine (pH 4.0) to remove nonspecific bindings while preserving VN–VC complexes. We reported the initial results in 2014,[18] and high specificity of GFP nanobody has been also reported recently by a separate research group to identify interacting proteins of the epidermal growth factor receptor (EGFR) family member ERBB2.[20]

Identification of CREB Kinases

We then applied FCMS to identify the upstream kinases interacting with CREB. Myc-Q1K-VN155 and the kinase library were transfected into SILAC heavy cells, while Myc-Q2L-VN155 and the library were transfected into SILAC light cells. After protein expression, cells were observed under fluorescence microscopy. As we expected, Myc-Q1K-VN155 interacted with kinases, leading to the emission of a fluorescence signal, while any fluorescence signal from Myc-Q2L-VN155 expressing cells was too weak to be visible. Cells were harvested and lysed, and the GFP nanobody was used to isolate CREB–kinase complexes. The Western blot result was consistent with fluorescence imaging, showing the formation of VN–VC complexes primarily in Q1K-containing cells. Direct on-bead digestion with trypsin was performed to recover peptides for LC–MS analysis. Three biological replicate experiments were conducted for reciprocal SILAC experiments. The SILAC result, shown in Figure a, revealed that the ratio of the majority of endogenous proteins centers to 1:1 (i.e., they are found at equivalent levels in the heavy- and light-isotope labeled samples) while the ratios of kinases deviated greatly from 1:1. Kinases that prefer binding with Q1K rather than with Q2L are potential CREB-interacting kinases. The protein ratios of the reciprocal SILAC experiments were clustered to generate a heat map (Figure 3b). At the bottom of the heat map, proteins have higher ratios in group 1 (heavy cells with Myc-Q1K-VN, light cells with Myc-Q2L-VN), and lower ratios in group 2 (heavy cells with Myc-Q2L-VN, light cells with Myc-Q1K-VN). The majority of kinases were found in this area, which is consistent with our expectation that this method can identify specific interacting CREB kinases. One-tailed one sample t tests were conducted for each triplicate experiment. Kinases with p < 0.01 and an abundance change of at least 4-fold were chosen as the candidates with highest confidence (Figure 3c). From overlapping results of 3 reciprocal SILAC experiments, 23 protein kinases were identified (Table ). Among the candidate proteins, 7 are known CREB kinases, representing almost half of all previously known CREB kinases identified in different systems and cellular states. These 7 known CREB kinases are cAMP-dependent protein kinase catalytic subunit alpha (PRKACA),[21] LIM domain kinase 1 (LIMK1),[22] calcium/calmodulin-dependent protein kinase type 1 (CaMKI),[23] ribosomal protein S6 kinase (RPS6KA1),[24] protein kinase D1 (PRKD1),[25] glycogen synthase kinase-3 alpha (GSK3A),[26] and 5′ AMP-activated protein kinase (AMPK).[27,28] Several different subunits of AMPK were found in reciprocal SILAC lists.

Figure 3

MS data acquisition and analysis. (a) Protein counts across different ratio ranges. The figure shows the condition of light cells with Myc-Q2K-VN155 and heavy cells with Myc-Q1L-VN155. Histograms show protein counts across the ratio range in three experiments. Red column represents kinases, and blue column represents proteins other than kinases. (b) Heat map represents the quantified proteins in the reciprocal SILAC experiments. Protein fold changes from the experiment were normalized and clustered. Kinases are accumulated in the area in the red rectangle. (c) One tail t test plot of SILAC experiments of light cells with Myc-Q2K-VN155 and heavy cells with Myc-Q1L-VN155. Proteins were identified at least twice in three biological replicates. Proteins within the cutoff area (ratio has 4-fold change, P < 0.01) were considered as significant. Kinases are represented by red dots.

Table 1

Identified Candidate Upstream Kinases of CREB

gene name	protein description	previous knowledge
CDK3	cyclin-dependent kinase 3	regulates same family member of CREB[29−31]
STK24	serine/threonine-protein kinase 24
MAPK1	mitogen-activated protein kinase 1	in the same pathway with CREB[32]
PHK	phosphorylase b kinase
CDK6	cyclin-dependent kinase 6	in the same pathway with CREB[21,33]
BRSK2	serine/threonine-protein kinase BRSK2	in the same pathway with CREB[27,34]
TSSK2	testis-specific serine/threonine-protein kinase 2	its family member TKK2 is a CREB upstream kinase[12]
RPS6KL1	ribosomal protein S6 kinase-like 1
TNNI3K	serine/threonine-protein kinase TNNI3K
PHKB	phosphorylase b kinase regulatory subunit beta
CAMK1	calcium/calmodulin-dependent protein kinase type 1	CREB upstream kinase[23]
CDK4	cyclin-dependent kinase 4
RPS6KA1 (p90RSK)	ribosomal protein S6 kinase	CREB upstream kinase[24]
LIMK1	LIM domain kinase 1	CREB upstream kinase[22]
CHEK2	serine/threonine-protein kinase Chk2
MAP2K2	dual specificity mitogen-activated protein kinase kinase 2	in the same pathway with CREB[32]
PRKACA	cAMP-dependent protein kinase catalytic subunit alpha	CREB upstream kinase[21]
EEF2K	eukaryotic elongation factor 2 kinase	is a substrate of CREB kinase p90RSK[35]
MAP4K1	mitogen-activated protein kinase kinase kinase kinase	in the same pathway with CREB[36,37]
PRKD1	serine/threonine-protein kinase D1	CREB upstream kinase[25]
PIM2	serine/threonine-protein kinase pim-2	in the same pathway with CREB[38−40]
PRKY	putative serine/threonine-protein kinase PRKY
GSK3α	glycogen synthase kinase-3 alpha	CREB upstream kinase[26]
AMPK	5′-AMP-activated protein kinase	CREB upstream kinase[27,28]

We selected six kinases, brain-selective kinase 2 (BRSK2), mitogen-activated protein kinase kinase kinase kinase 1 (MAP4K1), PIM2, cyclin-dependent protein kinase 3 (CDK3), cyclin-dependent protein kinase 4 (CDK4) and cyclin-dependent protein kinase 6 (CDK6), all of which were reported to be associated with CREB but never implicated as the CREB kinase, to determine if they might be novel CREB kinases. In vitro kinase assays indicated that, with the exception of CDK4, the other 5 kinases all could phosphorylate CREB in vitro (Figure.4a). LC–MS detection also revealed their putative CREB phosphorylation sites. Accordingly, we generated CREB phosphorylation resistant mutants by site-directed mutagenesis and carried out in vitro kinase assays. We successfully confirmed that BRSK2 phosphorylated CREB on S98 while PIM2 and CDK3 phosphorylated CREB on S133 (Figure.4b).

Figure 4

CREB upstream kinase selection and validation. (a) CREB kinases were validated by 32P-ATP based autoradiography. Kinase only (−CREB) or with CREB (+CREB) were applied to in vitro kinase assay. PKA was used as a positive control. Result showed that BRSK2, CDK6, MAP4K1, PIM2, and CDK3 phosphorylated CREB. (b) CREB mutants S133A, S98A/S133A, S100A/S133A, and S98A/S100A/S133A were used for selective in vitro kinase assay to locate the CREB phosphorylation sites. The autoradiography detection showed that, after S98A mutation, BRSK2 did not phosphorylate CREB; on the other hand, after mutating 133S to 133A, PIM2 and CDK3 could not phosphorylate CREB.

Discussion

Kinase–substrate networks are the main components of many signal transduction pathways. Although proteome-wide studies have been successful in elucidating many important biological events including the cataloging of thousands of sites of protein phosphorylation, there is a lack of a universal method to identify upstream kinases responsible for many of these modifications. FCMS presents a general high throughput approach that identifies potential upstream kinases in living cells by stabilizing and capturing the kinase–substrate pairs. Using FCMS, we identified several known and novel kinases of CREB.

The five newly discovered CREB kinases all have some indirect connections to CREB according to previous studies. BRSK2 is a serine/threonine protein kinase of the CAMK group, and it plays a key role in polarization of neurons and axonogenesis,[41,42] cell cycle progress,[43] and insulin secretion.[44] BRSK2 is phosphorylated and activated by liver kinase B1 (LKB1) and AMPK,[34] which are also CREB kinases.[27,28] Both BRSK2 and CREB are in the LKB1/AMPK pathway.[27,34] The CREB phosphosite we have identified for BRSK2 is S98, a site that also can be phosphorylated by PRKD1.[25] The sequence surrounding S98 is consistent with the substrate specificity of BRSK2 and is related in primary sequence to a previously reported phosphorylation site on CDK16.[45] The biological function of this site still remains to be investigated.

Both CDK3 and PIM2 phosphorylate CREB on S133. CDK3 has been reported to phosphorylate and activate transcription factor ATF-1,[29] which is in the same CREB/ATF family with CREB[30,31] and which it phosphorylates on S63, a site with high sequence similarity to the region surrounding CREB S133. PIM2 is distantly related to the family of calcium/calmodulin-dependent protein kinases and has positive effects on cell cycle progression while inhibiting apoptosis. The substrate specificity of PIM2 matches well with the sequence of amino acids surrounding S133, and this site is the major site that regulates CREB activity. Kinases including PRKACA,[46] ATM,[47] PRKD1,[25] and p90RSK[48] are all reported to phosphorylate CREB on S133. The phosphorylation of this site is related to the regulation of apoptosis,[49] differentiation,[47] and its transcriptional activities[50−53] and is important in diseases such as prostate cancer,[54] neuroblastoma,[55] and many others.

MAP4K1 is reported to activate the c-JUN N-terminal kinase (JNK) pathway,[36] a pathway that also mediates CREB phosphorylation and activation,[37] but has not been reported previously to interact directly with CREB. Similarly, CDK6 has not been previously identified as a CREB kinase, although interesting roles for CDK6 and cyclin D in transcriptional regulation have recently come to light[56] including the identification of cyclin D interactions with motifs in promoters recognized by CREB.[57] It is interesting to speculate that CDK6 might phosphorylate and modulate the activity of CREB as a mechanism to regulate transcription.

Some other interesting kinases were also in our short list. Examples include mitogen-activated protein kinase 1 (MAPK1) and its upstream activator mitogen-activated protein kinase kinase 2 (MAP2K2). Both have previously been reported as functioning upstream of the activation of CREB, but not directly interacting with the substrate.[32,58,59] These CREB kinases remain to be further verified.

FCMS is a novel proteomics method in several aspects. First, it works by stabilizing transient protein–protein interactions, enabling complexes to be isolated and analyzed by mass spectrometry. Previous studies showed that certain kinase–substrate interactions could be isolated using affinity purification while others could not.[2,60] Before we applied FCMS as a high throughput strategy to identify substrates and their kinases, we compared the method with the traditional AP-MS method using a known kinase and substrate pair, PKA–CREB. We found that FCMS detected PKA–CREB interaction, while traditional AP-MS did not (data not shown). Second, this stabilization is specific, allowing us to focus on interacting kinase–substrate molecules of interest with low interference from other proteins in the lysate. This is especially advantageous when compared to current cross-linking-based AP-MS methods, in which the linker may nonspecifically cross-link all proteins within close proximity.[61] Third, we introduced an effective negative control to enhance further the high specificity of FCMS. Unlike normal negative controls in typical AP-MS experiments in which tag only is used, a fragment of the substrate itself or a mutant version in its native state can be employed. Using a negative control of similar protein size to the wild type substrate limits the potential size effect for protein–protein interactions. Lastly, we successfully captured assembled kinase–substrate pairs specifically in our purification process by using GFP nanobody. Instead of capturing all bait proteins no matter if binding to preys or not, as is the case when using tag-based methods, our method greatly reduces the sample complexity by capturing only interacting protein pairs. This is especially advantageous in proteomic studies since low abundant, but biologically significant, proteins often are difficult to identify in a protein mixture in the presence of large amounts of high abundant proteins.

The limitation of FCMS includes the use of an overexpression system which may introduce false positives. Activation of a kinase may also need a specific extracellular stimulation, which is difficult to implement in a high throughput experiment. As a high throughput screening method to identify upstream kinases, further validation of candidates is expected in different cellular systems. In addition, FCMS detects interacting kinases, not necessarily the upstream kinases, although in this study, CREB is a transcription factor, and we reason that few kinases are downstream of CREB signaling. Furthermore, one critical issue in FCMS is to make the mutant or segment of substrate with the disrupted interacting region between interactors as the negative control. In high throughput experiments, ideally a negative control should be equally effective for eliminating the identification of all nonspecific interactions. If there is a lack of knowledge related to generic interacting domains in the target proteins, it is recommended that one design and construct several mutant candidates at the same time, and compare them with the wild-type protein. For the case of CREB, we used two CREB segments that differentially interact with upstream kinases. After we tested the strategy, results in fluorescence imaging, Western blot, and quantitative mass spectrometry all confirmed our assumption that the C-terminal had significantly fewer interacting kinases than did the N-terminal. This can be a general strategy for any future high throughput FCMS experiment, in which the knowledge about the substrate’s kinase interaction domain or phosphorylation sites is limited.

Methods

Online Method and Additional Information

Methods and any associated references are available as Supporting Information. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) with Project accession number PXD004739 via the PRIDE partner repository.[62] Correspondence and requests for materials should be addressed to W.A.T.