Multiple allostery in the regulation of PDGFR beta kinase activities

Abstract

Platelet-derived growth factor receptor beta (PDGFRβ), a type III receptor tyrosine kinase (RTK) with a featured kinase insert, regulates important cellular functions. Dysregulation of PDGFRβ is associated with cardiovascular and fibrosis diseases. Thus, its kinase activity needs to be precisely regulated under physiological conditions. Early studies demonstrated that its kinase is autoinhibited by its juxtamembrane segment and activated by transphosphorylation. However, additional mechanisms are required for the comprehensive regulation of the receptor kinase. Herein, we provide evidence that dimerization of activated kinases, autoinhibition by the kinase insert, and dimerization of inactive kinase, all contribute to the regulation of the receptor kinase. Moreover, we find such multiple allosteric regulation is also conserved in other type III RTKs, including colony stimulating factor 1 receptor (CSF1R). Impaired allosteric regulation of CSF1R is associated with malfunctions of microglia and demyelination of neurons in hereditary diffuse leukoencephalopathy with spheroids (HDLS).

Introduction

Platelet-derived growth factor receptor alpha (PDGFRα) and PDGFRβ, together with colony stimulating factor 1 receptor (CSF1R), mast/stem cell growth factor receptor (SCFR or c-Kit), and Fms-like tyrosine kinase 3 (Flt-3) receptors, belong to the type III receptor tyrosine kinase subfamily [ 1– 5]. All of them are composed of 5 extracellular Ig-like domains, a single-pass transmembrane domain (TM), an intracellular juxtamembrane segment (JM), a kinase domain splatted apart by a not-well characterized insert (KI), and a C-terminal tail bearing multiple phosphorylation sites ( Figure 1A) [ 1– 5].These receptors regulate important biological functions, including cell proliferation, differentiation, and migration. Dysregulation of these receptors is associated with cancer, cardiovascular and fibrosis diseases [ 1– 5]. Thus, a clear understanding of the precise regulation of these receptors is essential. The receptors in this subfamily are regulated by ligand-induced dimerization and subsequent activation through transphosphorylation ( Figure 1B) [ 5– 8]. In the absence of ligand, these receptors are monomeric and autoinhibited [ 5, 7– 9]. Upon binding to the dimeric ligands, these receptors are assembled into dimers so that two kinases in the dimer could trans-phosphorylate each other and be activated [ 5, 7, 8].

Figure 1 .

Overview of the structure and activation process of PDGFRβ

(A) Schematic diagram of the primary structure of PDGFRβ. sp: signal peptide; D1–5: Ig-like ectodomain; TM: transmembrane domain; JM: juxtamembrane segment; KD: kinase domain; KI: kinase insert. (B) Diagram of the activation process of PDGFRβ. Dimerized PDGF ligands bind to two PDGFRβs and activate their kinases. However, the precise regulation of the kinases has not been fully defined. (C) Autoinhibited state of PDGFRα kinase. By forming a hairpin structure, the JM segment binds to the active site of the kinase and competitively blocks the kinase from binding to its substrates. In addition, the JM interacts with the kinase αC helix. Through these interactions, the JM segment stabilizes the kinase in an inactive conformation. (D) Structure of the transphosphorylation complex of c-Kit kinases. The JM of one kinase is attached to the C-lobe of the other and places the tyrosine on the JM into the active site of the latter. When the kinase is activated, such autoinhibition of the JM segment needs to be released.

Crystallographic studies provided further details for these receptor kinases in autoinhibited and transphosphorylation states [ 10– 13]. In the autoinhibited state, the JM of PDGFRα, FLT3, or c-Kit receptor forms a hairpin winding around the αC helix of the kinase and poses the JM tyrosine-containing segment into the active site of the kinase ( Figure 1C) [ 10, 12, 13]. As such, the JM blocks the kinase from binding to the substrate [ 10, 12, 13]. In the transphos-phorylation state of c-Kit kinases, the JM of one kinase is latched onto the C-lobe of another kinase so that the two JM Tyrs from the former kinase can get into the active site of the latter for transphosphorylation ( Figure 1D) [11]. Moreover, in this trans-phosphorylation complex, both kinases are adopted in active conformation [11].

Although these early studies provided a basic framework for understanding the regulation of these receptor kinases, some conceptual gaps regarding the transition and stabilization of different conformational states of the kinase are still missing. For example, it was largely unknown: how is the kinase kept in a precise balance between the autoinhibited state and activated state? In addition, if the kinase is activated merely by auto-phosphorylation, how to prevent uncontrolled trans-phosphorylation and activation which can be introduced constantly by random collision of these receptors on the cell surface? And how to prevent the self-stimulated amplification of external stimuli through lateral phosphorylation of unliganded receptors by ligand-bound receptors ( Figure 1B)?

Accompany with these missing conceptual gaps, mounting experimental evidences also suggested that there are additional layers in the tight regulation of the kinase activities of PDGFRs and their subfamily members. It has been shown that, by rotating the transmembrane helixes of engineered PDGFRβ receptors, the kinases of ligand-bound receptors can be periodically activated [14]. This finding suggests that bringing the kinases into proximity is not enough to activate the kinases. Moreover, chimeric fusions of PDGFRα or PDGFRβ with proteins in oligomers were found in cancer patients [ 15– 18]. However, the kinase activities of these constitutively oligomerized chimeras were not similarly upregulated but were rather dependent on the presence of PDGFR JM and other regulatory elements [ 19– 21]. In addition to ligand-bound dimers, unliganded PDGFRβ tetramers have also been proposed to exist on the cell surface [ 22, 23].

To identify these gaps, we used chemical cross-linking mass spectrometry (CXMS), structure-based mutagenesis, and Rosetta modeling to study the regulation of PDGFRβ kinase activities. We found that the kinase activities of PDGFRβ and its subfamily members are regulated by multiple allostery. These types of allosteric regulation include the formation of an activated symmetric kinase dimer, autoinhibition by the kinase insert, and dimerization of inactive kinases. Disruption of these regulations impairs the functions of these receptors. In addition, we found that some HDLS-associated mutations in CSF1R destabilizes the dimerization of the active kinases and thus provides molecular insights for understanding the dominant inheritance of HDLS.

Materials and Methods

Materials

Antibodies were obtained from the following sources: protein C-tag antibody (HPC4) (rabbit polyclonal) from Genscript (Piscataway, USA); anti-phosphotyrosine antibody, clone 4G10 (mouse monoclonal) from Merck Millipore (Darmstadt, Germany); human PDGF-BB (PDB-H4112) and human M-CSF/CSF-1 (MCF-H5218) from ACRO Biosystems (Newark, USA). Dovitinib was from MedChemExpress (Monmouth Junction, USA); and disuccinimidyl suberate (DSS) and sulfosuccinimidyl suberate (BS ³) were from Thermo Fisher Scientific (Waltham, USA).

Construction of plasmids

The genes encoding wild-type full-length human PDGFRβ (from Dr. Jiahuai Han laboratory, Xiamen Universty, China) and CSF1R (Sino Biological, Inc., Beijing, China) were subcloned and inserted into the pEF1-puro plasmid (from Dr. Timothy A. Springer, Harvard Medical School, Boston, USA) at BamHI and XhoI restriction sites. All mutations were constructed by using Gibson assembly kit (NEB, Ipswich, USA). The signal peptide of the murine immunoglobulin κ chain was used for the expression of PDGFRβ and CSF1R receptors. For purification and western blot analysis, a protein C tag and an SBP tag were fused in tandem to the C-terminus of PDGFRβ or the CSF1R receptor. To monitor the transfection efficiency, an IRES element and the EGFP gene were added after the receptor coding sequence. All the constructs were verified by Sanger sequencing.

Cell culture, transfection and immunoblotting

HEK293T cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% FBS at 37°C with 5% CO ₂. When cells reached 40%‒50% confluency, 0.5 μg of plasmid was mixed with 1.5 μg of linear polyethyleneimine (PEI) and then added to a well of cultured HEK293T cells in 12-well plates. After the cells were transfected for 5 h, the culture media was replaced by serum-free DMEM. Twenty hours posttransfection, the cells were treated with 50 ng/μL PDGF-BB or CSF (ACRO Biosystems) for 20 min in a CO ₂ incubator. The treated cells were lysed and prepared for western blot analysis as described previously [24]. The expressions of PDGFRβ and CSF1R were detected by western blot analysis with an anti-protein C antibody (GenScript). The phosphorylation levels of PDGFRβ and CSF1R were detected by western blot analysis with a 4G10 antibody (Merck Millipore).

Expression, purification of PDGFRβ and kinase assay in vitro

To express PDGFRβ, HEK293F cells (1.5 × 10 ⁶ cells/mL) were seeded into 1 L of SMM-293TI medium in a 2-L conical flask rotating at 120 rpm in a 5% CO ₂ incubator at 37°C. One milligram of plasmid and 3 mg of PEI were mixed and incubated at room temperature for 30 min. Then, the mixture was added to the flask. After 3 days, the transfected cells were harvested and stored at ‒80°C.

To purify PDGFRβ, harvested cells were resuspended and lysed as previously described [24]. The lysate was cleared by ultracentrifugation at 50,000 g for 30 min at 4°C. The cleared lysate was loaded onto a strep-tactin column (IBA, Göttingen, Germany) and purified as previously described [24]. For the chemical cross-linking experiments, the lysis buffer was replaced by 50 mM HEPES (pH 8.0), 400 mM NaCl, 1 mM EDTA, 10% glycerol, 0.15% Triton X-100, 100 mM Na ₃VO ₄, 0.5 mM TCEP, and 1× complete protease inhibitor cocktail. The wash buffer was replaced by 50 mM HEPES, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 0.1% Triton X-100, 0.5 mM TCEP, and 2 mM PMSF.

For the kinase activity assay, purified PDGFRβ was treated with 0.8 μM PDGF-BB in 25 μL of reaction buffer containing 8 mM HEPES and 160 μM ATP on ice for 30 min. The reaction was stopped by the addition of 6× SDS sample buffer. To inhibit PDGFRβ kinase activity, 10 μM Dovitinib together with 0.8 μM PDGF-BB was added to the purified receptor and incubated on ice for 5 min. The expression and phosphorylation levels of PDGFRβ were detected by western blot analysis with anti-protein C and anti-4G10 antibodies, respectively.

Chemical cross-linking of proteins coupled with mass spectrometry (CXMS) analysis

The purified protein was concentrated by using an Amicon Ultra 0.5 mL 50-kDa filter (Millipore) at 12,000 g for 10 min. The protein concentration was calibrated by using BSA as a standard (~0.3–1 mg/mL).

Approximately 8 μg of purified PDGFRβ was incubated with 0.18 μM PDGF-BB at 4°C for 1.5 h. Then, the proteins were crosslinked with BS ³ (1 μg BS ³ per 1 μg protein) or DSS (the molar ratios of protein to DSS were 1:50 and 1:150) for 1 h at room temperature. The reaction was quenched by the addition of 20 mM NH ₄HCO ₃. To stabilize the kinase of PDGFRβ in an inactive conformation, 10 μM Dovitinib was added to the protein solution and incubated on ice for 5 min. The cross-linking experiment was performed as described previously.

Next, the cross-linked proteins were precipitated with 4 volumes of ice-cold acetone, resuspended in 8 M urea and 100 mM Tris (pH 8.5), and digested with 0.16 μg of trypsin (Promega, Madison, USA) in 2 M urea and 100 mM Tris (pH 8.5). LC-MS/MS analysis was performed on an Easy-nLC 1000 II HPLC (Thermo Fisher Scientific) coupled to a Q-Exactive HF mass spectrometer (Thermo Fisher Scientific). Peptides were loaded and desalted on a pre-column (75 μm ID, 6 cm long, packed with ODS-AQ 120 Å–10 μm beads from YMC Co., Ltd., Kyoto, Japan) and subsequently separated on an analytical column (75 μm ID, 12 cm long, packed with Luna C18 1.9 μm 100 Å resin from Welch Materials Inc. (Shanghai, China)). The sample on the analytical column was eluted with a linear reverse-phase gradient from 100% buffer A (0.1% formic acid in H ₂O) to 30% buffer B (0.1% formic acid in acetonitrile) over 56 min at a flow rate of 200 nL/min. The top 15 most intense precursor ions from each full scan (resolution 60,000) were isolated for HCD MS2 (resolution 15,000; normalized collision energy 27) with a dynamic exclusion time of 30 s. Precursors with 1+, 2+, 7+ or above, or unassigned charge states were excluded. pLink 2 [25] software was used to identify cross-linked peptides with a precursor mass accuracy of 20 ppm and a fragment ion mass accuracy of 20 ppm. The results were filtered by applying a 5% FDR cutoff at the spectral level and then an E-value cutoff at 0.001 [26].

Structural modeling

The global docking of the PDGFRβ kinase dimer was calculated with Cluspro [27]. The local docking of the kinase dimer was calculated with Rosetta as described previously [28]. Each cross-linked residual pair was used individually as a distance constraint in the calculation of the kinase dimer. In the calculation, the pair-wise distance was calculated with Xwalk [29]. Five hundred models with the lowest energy scores were selected. These models were subsequently screened for their ability to satisfy cross-linking constraints (< 30 Å) and the requirement for buried interface area (> 900 Å ²). The two largest clusters were selected from these screened models. The structural models calculated with each distance restraint were compared. Only those pairs that generated similar converged structural models were selected as inter-molecular constrains for the next round of calculation. Next, every two selected cross-linked pairs were combined as two distance constraints in the local docking calculation. Those that yielded converged models were selected. Finally, three out of five cross-linked pairs were combined as inter-molecular distance constraints in the calculation of the final docking models.

Results

A symmetric, active kinase dimer interface was identified in ligand-bound PDGFRβ receptors by CXMS and Rosetta modeling

To find out how PDGFRβ kinases in ligand-bound, dimeric receptors interact with each other to stabilize their activated state, we used CXMS and Rosetta modeling to determine the kinase dimerization interface [30]. The functional full-length PDGFRβ receptor was purified as previously described [24]. The kinase activities of the purified receptors in detergent were analyzed by western blot analysis in the presence or absence of ATP, PDGF-BB, and Dovitinib. Dovitinib is a PDGFR kinase inhibitor which can stabilize the kinase in the inactive conformation [ 31, 32]. The kinase activities of the purified receptors could be stimulated by treatment with 0.8 μM PDGF-BB. In addition, the stimulated activities of the receptors could be inhibited by the addition of 10 μM Dovitinib, suggesting that the purified receptors were functional ( Figure 2A). Then, we cross-linked purified, ligand-bound receptors using 0.8 mM BS ³ or 0.15‒0.45 mM disuccinimidyl suberate (DSS) for 1 h. The cross-linked dimers were separated from the monomers by SDS-PAGE and analyzed by mass-spectrometry ( Figure 2B).

Figure 2 .

Symmetric dimerization of activated PDGFRβ kinases

(A) Purified PDGFRβ is functional. The activity of purified PDGFRβ was analyzed in the presence or absence of ATP, PDGF-BB, and Dovitinib. The phosphorylation and protein levels of the samples were analyzed by western blot analysis using 4G10 and protein C antibodies, respectively. (B) Cross-linking of purified PDGFRβ with BS3 and DSS. The purified PDGFRβ was treated with PDGF-BB at 4°C for 1.5 h, and then the mixture was cross-linked with BS3 (1 μg BS3/μg protein) or DSS (the molar ratio of protein to DSS was 1:50 or 1:150) at room temperature for 1 h. The reaction was stopped with 20 mM NH4HCO3. The samples were separated by SDS-PAGE. The arrow indicates cross-linked bands of PDGFRβ. (C) Dimeric structures of activated PDGFRβ kinases. The model of the activated PDGFRβ kinase dimer was calculated with Rosetta. K637s are shown as red sticks. The αC helix is in yellow, and the activation loop is in green. (D) The crystallographic dimer of FGFR2 kinase in the active conformation (PDB 3B2T).

Nine cross-linked peptide pairs were identified from ligand-bound dimeric PDGFRβ receptors ( Table 1 and Supplementary Figure S1). In particular, two Lys387s on the extracellular subdomain D4s were cross-linked together in ligand-bound receptors ( Table 1). This finding is consistent with experimental evidence that the proximity of D4 is critical for PDGFRβ activation. In addition, it is compatible with the crystal structure of the ligated c-Kit ectodomain (PDB 2E9W) [ 33, 34], suggesting that our cross-linking experiments are valid.

Table 1 Cross-linked peptide pairs of ligand-bound PDGFRβ

	PDGFRβ			PDGFRβ			Spectral counts	Best E value
	Sequence	Cross-linked lysine		Sequence	Cross-linked lysine
	VTDPQLVVTLHE KK	163		SYIC KTTIGDR	191		2	4.49E-04
•	V KVAEAGHYTMR	387		V KVAEAGHYTMR	387		8	4.83E-04
	ML KSTAR	637		ML KSTAR	637		3	8.32E-05
•	ML KSTAR	637		SSE KQALMSELK	645		4	5.36E-04
•	ML KSTAR	637		DSNYIS KGSTFLPLK	860		1	3.39E-07
	ML KSTAR	637		KYQQVDEEFLR	969		3	3.78E-05
	VAV KMLK	634		SSE KQALMSELK	645		4	9.07E-04
	SSE KQALMSELK	645		DSNYIS KGSTFLPLK	860		1	4.13E-03
•	LLGEGY KK	967		KYQQVDEEFLR	969		3	4.09E-05

• represents that the same cross-linked peptide pair was identified in the presence and absence of Dovitinib.

On the intracellular side, we reasoned that K637-K637, K637-K969, and K967-K969 represent the inter-molecularly cross-linked pairs. In the threading model of PDGFRβ kinase, which is based on the structure of active c-Kit kinase (PDB 1PKG), K969 is located too far from K637 to be intra-molecularly cross-linked. Similarly, K969 is located too close to K967 to be intra-molecularly cross-linked. Other cross-linked Lys pairs were individually screened in silico as inter- or intra-molecular pairs according to the criteria described in the Methods section.

By using the proximities of inter-molecularly cross-linked Lys pairs as distance constraints, we calculated a model of activated PDGFRβ kinase dimer using Rosetta Docking protocols [30]. As K637-K637, K637-K969, and K967-K969 were reasoned to be inter-molecularly cross-linked pairs, they impose a nearly 2-fold symmetry constraint on dimeric kinase assembly. As such, we used a 2-fold symmetry constraint in the global docking step and released such a constraint in the final local docking step.

The calculated active PDGFRβ kinase dimer is symmetric ( Figure 2C). The total buried surface area of the dimer is 2290 Å ². The binding energy and P value of the assembly were estimated to be ‒9.0 kcal/mol and 0.315, respectively, using the PISA server [35]. In this dimer, two kinases are interlocked together through the interactions between N-lobe residues (G603, R604, T605, S608, Q613, S638, T639, A640, S642, K645, G674 and G675) and C-lobe residues (R830, P866, L867, K868, W869, P905, E906, P908 and M909).

This symmetric dimer is reminiscent of the crystallographic dimer of FGFR2 kinase (PDB 3B2T) ( Figure 2D), in which the activity-compromised mutant of FGFR2 kinase adopts an active conformation [36]. A similar symmetric assembly was also found in the crystal structures of the active RET (PDB 2X2L) and FGFR1 kinases (PDB 5FLF) ( Supplementary Figure S2) [ 37, 38]. Moreover, the dimer interface is well conserved among the type III receptor tyrosine kinases [39] ( Supplementary Figure S3).

Disruption of the symmetric kinase dimer interface abolished the ligand-stimulated activities of PDGFRβ and CSF1R

To validate the physiological relevance of the identified symmetric kinase dimer interface, 11 out of 21 residues buried at the PDGFRβ kinase dimer interface were randomly selected and mutated ( Figure 3A). With the exception of P908A, all the other mutations impaired the stimulated activity of PDGFRβ ( Figure 3B). The R604E, G603E, A640E, and K645E mutations exhibited moderate effects, while other mutations had more significant effects on activity. The residues R830 and P866 are located near the ATP binding pocket, whereas the other residues are not involved in ATP or substrate binding.

Figure 3 .

Disruption of activated kinase dimerization impairs ligand-stimulated activation of PDGFRβ

(A) Calculated structural model for the activated PDGFRβ kinase dimer. For clarity, two activated kinases in the dimer were rotated clockwise and counterclockwise by 90°, respectively. N-lobe residues at the dimer interface are shown as blue sticks on the right, whereas C-lobe residues at the dimer interface are shown as red sticks on the left. (B) Mutations of the residues at the symmetric dimer interface impair the activation of PDGFRβ. Eleven residues at the symmetric kinase dimer interface were selected and mutated. With the exception of P908A, every mutation reduced the stimulated activity of PDGFRβ. (C) Mutations in the CSF1R kinase dimer interface are linked to the pathogenesis of HDLS and impaired kinase activation. All selected mutations impair the ligand-dependent activation of CSF1R. (D) Effects of the expression levels of PDGFRβ mutants on the activity of co-expressed WT receptor. The inactive mutants of PDGFRβ were co-transfected with the WT receptor. During transfection, the amount of the plasmid encoding WT PDGFRβ was fixed, while the amount of the plasmid encoding the inactive mutant increased from 0.5- to 4-fold that of the WT. Increasing the expression of the I875A mutant could proportionally inhibit the stimulated activity of the WT receptor. (E) The impact of the expression levels of the CSF1R mutant on the activity of co-expressed WT receptor. The inactive mutant of CSF1R, I827A, was co-transfected with the WT receptor. During the transfection, the amount of the plasmid encoding WT CSF1R was fixed, while the amount of the plasmid encoding I827A was increased from 0.5- to 4-fold that of the WT. Increasing the expression of the I827A could proportionally inhibit the stimulated activity of the WT receptor.

Strikingly, the conserved kinase dimer interface overlaps with CSF1R mutations associated with HDLS, a rare autosomal dominant neurodegenerative disease [ 40– 42]. We hypothesized that these disease-associated mutations affect the stability of activated kinase dimers and thus impair the stimulated activity of CSF1R. To test this hypothesis, eight residues associated with HDLS were mutated, and all of them impaired the stimulated activity of CSF1R ( Figure 3C).

It is unlikely that our selected mutations all interfere with substrate recognition via trans-phosphorylation. However, we carefully examined this possibility by co-transfecting each selected mutant with the wild-type receptor. For the co-transfection experiments, the K634A/WT combination was used as a control. K634 is the key residue required for catalyzing the phospho-transfer reaction. Presumably, the K634A mutation should have little effect on the equilibrium of the kinase conformations. If auto-phosphorylation and dimerization of activated kinases are all required for PDGFRβ activation, co-expression of WT receptor with a mutant that can impair both auto-phosphorylation and active kinase dimerization should have a stronger impact on kinase activity than co-expression of K634A mutant with WT receptor. Indeed, we found that the stimulated activity of co-expressed WT and I875A was lower than that of co-expressed WT and K634A ( Supplementary Figure S4). Our structural analysis revealed that the I875 residue is located on the αEF helix in proximity to the dimerization interface. In addition, I875 was predicted to be at the transphosphorylation interface.

To further validate this result, we performed a titration experiment. During transfection, we fixed the amounts of the plasmids encoding WT PDGFRβ but increased the amounts of the plasmids encoding the mutants ( Figure 3D). With increasing amounts of the plasmids encoding I875A mutant, the stimulated activities of the co-transfected I875A/WT receptors decreased proportionally. In comparison, the stimulated activities of co-transfected L867E/WT as well as K634A/WT receptors were not decreased proportionally.

The above experimental evidences are consistent with our calculated structure of the activated PDGFRβ kinase dimer. Based on our structure, I875 is located at the substrate recognition site of the kinase. Moreover, I875A mutation could destabilize the active kinase dimerization. As such, I875A mutation exerts a greater effect on the kinase activity of co-transfected receptors than the K634A mutation.

I875 is conserved across type III RTKs ( Supplementary Figure S3). Mutation of its equivalent residue in CSF1R, I827, is linked to the onset of HDLS [42]. To investigate the significance of this mutation in the pathogenesis of HDLS, we mutated CSF1R I827 to Ala and performed a titration experiment as described above ( Figure 3E). The stimulated activities of the samples co-transfected with I827A and WT CSF1Rs were reduced in a manner dependent on the amount of co-transfected I827A plasmids. This result indicated that I827A mutation could inhibit the activity of WT CSF1R in trans, which has implications for the dominant inheritance of this disease.

The kinase insert could autoinhibit the ligand-induced activation of PDGFRβ

To understand how inactive kinases are associated together in ligand-bound PDGFRβ receptors, we used a PDGFRβ kinase inhibitor, Dovitinib, to stabilize the kinase in the inactive conformation ( Figure 4A,B) and studied the kinase association by CXMS [43]. Purified PDGFRβ receptors treated with both PDGF-BB and Dovitinib showed a different cross-linking spectrum from the sample treated with PDGF-BB alone ( Table 2 and Supplementary Figure S5). Especially, many cross-linking pairs were identified between the kinase insert and the kinase domain. In addition, the cross-linking pair between the two K387s on the receptor extracellular D4 domains was maintained, indicating that the proximity of the D4 domains in ligand-bound PDGFRβs was not affected by stabilizing the kinase in an inactive conformation.

Figure 4 .

Autoinhibition of PDGFRβ by its kinase insert

(A) PDGFRβ kinase can be stabilized in an inactive conformation by Dovitinib. Superimposition of the structure of activated PDGFRβ kinase with that of inactive FGF1 kinase complexed with Dovitinib (PDB ID: 5A46). Activated PDGFRβ is in light blue, and its αC helix is in yellow; inactive FGF1 is in gray, and its αC helix is in green. (B) Dovitinib-bound PDGFRβ was cross-linked by BS3 and DSS. Purified PDGFRβ was incubated with PDGF-BB at 4°C for 1.5 h and then treated with 10 μM Dovitinib for 5 min. The mixtures were crosslinked with BS3 or DSS at room temperature for 1 h. The arrow indicates cross-linked bands of PDGFRβ. (C) Schematic diagram of the PDGFRβ kinase insert. Using cross-linking data as constraints, the structure of the kinase insert was docked onto the inactive kinase structure using Rosetta. The kinase insert was docked to the kinase C-lobe near the αF helix. The binding site of the kinase insert overlaps with the substrate recognition site and the activated kinase dimer interface. (D) Mutations in the kinase insert impaired the activity of PDGFRβ. Five KI residues that interact with the kinase were mutated. Three of them (D737A, Y751A, and M741A) could enhance the activity of the kinase.

Table 2 Cross-linked peptide pairs of ligand-bound PDGFRβ inhibited by Dovitinib

	PDGFRβ			PDGFRβ			Spectral counts	Best E value
	Sequence	Cross-linked lysine		Sequence	Cross-linked lysine
	TDPQLVVTLHE KK	163		K KGDVALPVPYDHQR	164		1	8.45E-05
•	V KVAEAGHYTMR	387		V KVAEAGHYTMR	387		1	9.47E-04
	ML KSTAR	637		GDV KYADIESSNYMAPYDNYVPSAPER	762		9	1.39E-08
•	ML KSTAR	637		SSE KQALMSELK	645		4	9.65E-07
•	ML KSTAR	637		DSNYIS KGSTFLPLK	860		8	9.82E-06
	DSNYIS KGSTFLPLK	860		YNAI KR	918		3	5.06E-06
	RPPSAELYSNALPVGLPLPSHVSLTGESDGGYMDMS KDESVDYV	745		PMLDM KGDVK	758		3	1.66E-09
	GQVVEATAHGLSHSQATM KVAVK	630		DV KYADIESSNYMAPYDNYVPSAPER	762		10	3.72E-07
•	LLGEGY KK	967		KYQQVDEEFLR	969		4	2.16E-11
	KYQQVDEEFLR	969		KYQQVDEEFLR	969		1	7.52E-04

• represents that the same cross-linked peptide pair was identified in the presence and absence of Dovitinib.

To model the inactive kinase association in ligand-bound, Dovitinib-inhibited receptors, we took three steps in Rosetta computation: Ab initio modeling of the kinase insert, docking of the insert core onto the kinase, and docking of the insert-bound inactive kinases [7].

In the Ab initio modeling of the kinase insert, all-atom models were generated and refined [30]. These models vary significantly in their structures. In the top cluster, the RMSDs of the calculated models converged to 8 Å ( Supplementary Figure S6A). However, these models share a conserved core structure composed of 4 β-strands. The RMSDs of the core structures were converged to 1 Å ( Supplementary Figure S6B). In addition, the sequence of this kinase insert core is well conserved across Metazoa ( Supplementary Figure S7).

Using the cross-linking data and the calculated structure of the insert core, we modeled the association between the insert and the inactive kinase domain with Rosetta [5]. In docking models, the insert core is mapped to the kinase C-lobe near the αF helix ( Figure 4C). At this position, the kinase insert could sterically block the kinase from recognizing its substrates and forming the activated kinase dimer.

To verify this point, we mutated 5 residues on the kinase insert, which interact with the kinase domain in our model ( Figure 4D). Compared with the WT receptor, the stimulated activities of 3 out of the 5 mutants were significantly enhanced, while the activities of the other mutants remained similar, confirming the autoinhibitory function of the kinase insert ( Figure 4D).

Unfortunately, we could not unambiguously determine the interface for inactive kinase dimerization due to the limited number of identified cross-linking pairs ( Table 2). However, two Lys969s on the kinase C-terminal αJ helices were cross-linked, suggesting that these two helices should be in proximity to each other in the inactive kinase dimer. This result is consistent with a previous report that the C-terminus of PDGFRβ is only exposed and detectable by a conformational-specific antiserum in the presence of ligand [44].

The formation of a tetrameric complex is not required for the transactivation of PDGFRβ kinase

As it has been proposed that PDGFRβ may exist as a tetramer on the cell surface, we examined this possibility by analyzing the crystal packing of known RTK III kinase structures [ 22, 23]. From these structures, we found that activated c-Kit kinases could form a circular tetrameric complex via transphosphorylation through four nearly identical head-to-tail interactions (PDB 1PKG) ( Figure 5A) [11]. The interface of these interactions is well conserved among RTK III kinases [39] ( Figure 5B and Supplementary Figure S3), indicating that this assembly might be functionally relevant.

Figure 5 .

The tetrameric assembly of PDGFRβ is not required for its ligand-dependent activation

(A) The crystal packing of tetrameric c-Kit kinases during transphosphorylation. Residues at the tetrameric interface are shown in close-up views. JM residues are shown as blue sticks, and C-lobe residues are shown as purple sticks. (B) Model of tetrameric PDGFRβ kinases and the sequence alignment of c-Kit and PDGFRβ. The tetrameric interfaces are located at two different regions: the JM segment and the C-lobe. Residues at the JM interface are shown in blue stars, and residues at the C-lobe interface are shown in green stars. (C) Effects of mutations at the C-lobe interface on PDGFRβ activation. Mutants with decreased activity are marked with hollow green stars. (D) Effects of mutations at the JM interface on PDGFRβ activation. JM mutants with decreased activity are marked with hollow blue stars. Residues that are not at the interface but have decreased activity are marked by hollow purple stars. (E) Double mutation of the electrostatic interaction pair could not restore the impaired kinase activity of singly mutated receptors. Residues that form electrostatic interactions at the tetrameric interface were mutated to residues with opposite charges. A pair of non-interacting residues (644/912) was used as a control. Left, comparison of the activities of single mutants with those of double mutants. Right, comparison of the activities of single mutants with those of paired co-transfected single mutants.

Therefore, we tested whether the formation of this tetramer is required for stabilizing PDGFRβ kinases into activated dimer-to-dimer interactions during transactivation. To this end, we mutated a series of residues at the transphosphorylation interface. On the basis of the localization of these residues, they can be divided into two different groups: one at the kinase C-lobe and the other at the juxtamembrane ( Figure 5B). Six out of eight mutations in the kinase C-lobe dramatically reduced the ligand-stimulated activity of PDGFRβ ( Figure 5C). In contrast, only 2 (Y579A and Y581A) of the 7 JM mutations at the interface reduced the ligand-stimulated activity of PDGFRβ ( Figure 5D), whereas three JM mutations (V582A, Q586A and L587E) at the interface increased kinase activity. We also found that 2 other mutations (D583A and M585A) not at the interface could reduce kinase activity ( Figure 5D). Taken together, these results do not support the possibility that the formation this transphosphorylation tetramer is required for PDGFRβ transactivation.

To further validate this conclusion, we introduced an electrostatic interaction pair (911/653) into the transphosphorylation interface and studied whether double mutations could restore the impaired activities of singly mutated receptors. As a control, a pair of non-interacting residues (644/912) was selected and mutated. Except Q912E, every single mutation reduces the stimulated activity of the kinase. However, neither of the double mutations restored the kinase activity to the average level of the two singly mutated receptors ( Figure 5E, left). Similar results were obtained when we compared the average activity of two single mutants with that of paired, cotransfected single mutants ( Figure 5E, right).

These results suggest that the formation of the tetrameric transphosphorylation complex is not required for the transactivation of PDGFRβ.

Discussion

The regulation of PDGFRβ kinase activity is far more complicated than expected. In addition to previously identified autoinhibition by JM and activation by transphosphorylation [ 5, 6, 12], we identified additional regulatory layers that are required for the precise regulation of the activities of PDGFRβ and its subfamily members. These regulatory layers include the formation of a specific activated kinase dimer, autoinhibition by the kinase insert, and the formation of an inactive kinase dimer ( Figure 6). Identification of these additional layers would clarify the comprehensive regulation of PDGFRβ and other receptors in its subfamily.

Figure 6 .

Schematic diagram of the potential mechanism of PDGFRβ activity regulation

Precise regulation of PDGFRβ activation necessitates multiple regulatory layers. In the absence of ligands, PDGFRβ is autoinhibited either by the kinase insert or through the formation of an inactive kinase dimer. Upon ligand binding, the receptors undergo transphosphorylation and are subsequently stabilized in a specific active dimer configuration.

Allosteric activation might be a general theme in the regulation of RTKs. Such mechanism has been demonstrated by Kuriyan et al. [ 45, 46] in the study of EGFR, a non-canonical receptor tyrosine kinase. Unlike PDGFRβ, phosphorylation of EGFR kinase activation loop is not required for its activation. Instead, two kinases of ligated EGFRs form a cyclin/CDK-like asymmetric dimer to activate the “CDK-like” receiver kinase [ 45, 46]. Similar asymmetric dimerization was also found in the tetrameric complex of c-Kit kinases [11], but such asymmetric interaction is not required for the activation of PDGFRβ. Instead, two activated kinases in ligand-bound PDGFRβ receptors form a symmetric dimer. This symmetric dimer interface is well conserved among type III RTKs, including CSF1R ( Supplementary Figure S3). Similar symmetric dimerization can be found in the crystal packing of FGFR1, RET, and FGFR2 kinases [ 37, 38, 47].

Asymmetric and symmetric dimerization have different impacts on receptor oligomerization and signaling. Two distinct interfaces of asymmetric dimerization allow the receptors to be assembled continuously, forming a linear polymer for lateral signal propagation [ 45, 46]. However, if these two distinct interfaces were used for symmetric dimerization, the receptors could only be assembled into dimers but not high-order oligomers. This self-constrained dimerization could prevent uncontrolled signaling caused by random collision of receptors, which occurs constantly on the cell surface. Indeed, many kinases, including Ser/Thr kinases, are assembled into symmetric dimers in their signaling [48].

Dysregulation of RTKs is often associated with various human diseases, including cancer and neurodegenerative disease [ 1, 3]. While many cancers are associated with gain-of-function mutations in kinases, neurodegenerative diseases are often associated with loss-of-function mutations in kinases [ 40, 41]. HDLS, a dominant neurodegenerative disorder, is caused by microglial dysfunction, which in turn leads to demyelination of neurons [41]. Many missense mutations in CSF1R have been identified in HDLS patients [41]. However, their impacts on CSF1R functions and their underlying pathogenic basis are largely unknown [ 40, 41]. In our studies, we found that some HDLS-associated CSF1R mutations impaired the stimulated activity of the receptor and the dimerization of the activated kinases. In addition, the I827A mutation not only impaired the kinase activity of itself but also inhibited the activity of co-expressed WT receptors in a dose-dependent manner. Our results provide molecular insights for understanding the regulation of CSF1R function in microglia and the molecular basis for the dominant heritance of HDLS.

A striking feature of type III RTKs is the presence of a kinase insert that splits the kinase domain into two separate halves [ 3, 5]. The KI of PDGFRβ has been shown to be important for PDGF-induced mitogenesis [ 49, 50]. However, less is known about the structures and functions of these KIs in kinase activity regulation. In limited studies on the KIs of type III RTKs, controversial results have been reported [ 49, 51]. To investigate the biological functions of PDGFRβ-KI, we used Dovitinib to stabilize PDGFRβ kinase in an inactive conformation. We found that the KI of PDGFRβ could sterically block substrate recognition and dimerization of activated kinases. As such, the KI of PDGFRβ autoinhibits the activation of the receptor. However, our studies were limited by the uncertainty of the complete structure of the KI. Further studies are required to define the atomic details of the KI and its interactions with the splatted kinase domain.

Another limitation of our study is that the inactive kinase dimer interface was not defined. However, a clear contrast between Dovitinib-bound and unbound kinase dimer interactions was detected in CXMS experiments. From the limited cross-linking pairs detected in the inactive kinase dimer, we found that the two C-terminal αJ helices should be in proximity to each other.

In summary, we found that multiple allostery is involved in the precise regulation of the activities of PDGFRβ and its subfamily members. Disruption of these regulatory layers could impair the functions of these receptors, which might have clinical implications.

Supplementary Data

Supplementary data is available at Acta Biochimica et Biophysica Sinica online.

COMPETING INTERESTS

The authors declare that they have no conflict of interest.

Funding Statement

This work was supported by the grants from the National Natural Science Foundation of China (Nos. 31670738 and 31470730 to L.M.), and the Tianjin Municipal Science and Technology Bureau and Tianjin Natural Science Foundation (No. 21JCQNJC01660 to X.Q.).