A ubiquitin-like domain controls protein kinase D dimerization and activation by trans-autophosphorylation

Protein kinase D (PKD) is an essential Ser/Thr kinase in animals and controls a variety of diverse cellular functions, including vesicle trafficking and mitogenesis. PKD is activated by recruitment to membranes containing the lipid second messenger diacylglycerol (DAG) and subsequent phosphorylation of its activation loop. Here, we report the crystal structure of the PKD N terminus at 2.2 Å resolution containing a previously unannotated ubiquitin-like domain (ULD), which serves as a dimerization domain. A single point mutation in the dimerization interface of the ULD not only abrogated dimerization in cells but also prevented PKD activation loop phosphorylation upon DAG production. We further show that the kinase domain of PKD dimerizes in a concentration-dependent manner and autophosphorylates on a single residue in its activation loop. We also provide evidence that PKD is expressed at concentrations 2 orders of magnitude below the ULD dissociation constant in mammalian cells. We therefore propose a new model for PKD activation in which the production of DAG leads to the local accumulation of PKD at the membrane, which drives ULD-mediated dimerization and subsequent trans-autophosphorylation of the kinase domain.


Figure S3. The ULD forms concentration-dependent dimers.
A. Static light scattering (SEC-MALS) of the human PKD1 ULD (PKD1 ULD , light blue) and PKD3 ULD (PKD3 ULD , dark blue) and their corresponding molar mass (dotted). The table summarizes the polydispersity (Mn/Mw) as well as the experimentally determined molecular weight (Mexp), the theoretical monomeric molecular weight (Mtheo) and the oligomeric state (Mexp/Mthe) of both proteins (see as well Fig 2B).
C. Intensities of identical peptides between rat and human PKD1 in INS-1 lysates targeted by parallel reaction monitoring mass spectrometry. Lysates were spiked with an equimolar amount (5, 50, 500 ng) of human PKD1 ULD-C1a and PKD1 PH-CAT (black symbols) or un-spiked (cyan symbols). The intensity of the un-spiked lysates was subtracted from the spiked samples prior to regression analysis. The linear regressions (red, solid line) were used to calculate the amounts of endogenous PKD1 protein. The experiment was repeated twice (replicate 1/replicate 2). Note that one of the peptides (peptide 1) could only be reliably detected in all samples in one replicate, and that some of the endogenous peptides are below the lowest amount of spiked protein and therefore cannot be reliably quantified.

Figure S4. Electron density maps of the ULD-C1a interface.
A. 2Fo-Fc electron density map surrounding Arg17 of the ULD. Arg17 makes a bifurcated hydrogen bond to the side chain carboxylate of Glu85 (ULD) and a hydrogen bond to the backbone carbonyl of Arg145 (C1a). Clear electron density is observed for Phe114 of the C1a domain, which forms a hydrogen bond between its backbone carbonyl and the epsilon nitrogen of Arg17. B. 2Fo-Fc electron density map surrounding both Arg17 and Arg22 of the ULD. In contrast to Arg17, the electron density for Arg22 is less well defined, though a clear region of electron density could be observed at a location, size, and shape consistent with the guanidino group of the side chain. Contouring of the 2Fo-Fc map at 0.7σ reveals contiguous electron density for the side chain.

Figure S5. The PKD1 CAT autophosphorylation site can be mapped to S 742 .
A. ArgC digest of ATP incubated protein (see Figure 7D) analysed by LC-MS/MS. Identified peptides mapped onto the protein sequence shown in magenta. The table lists all PKD-specific phospho-sites identified with assigned localization probability, score and mass error (in parts per million), in order of signal intensity. The phosphorylation site within the respective peptide is highlighted in magenta. Score refers to the Andromeda (Max Quant) score of the best scoring peptide spectrum match assigned to each site. For each phospho-site the ratio of phosphorylated peptide to unmodified peptide signal (ratio mod/base) is reported. B. Overlay of intact mass spectra of recombinant PKD1 CAT and stoichiometrically phosphorylated PKD1 CAT (P-PKD1 CAT ) which was generated as described in the methods. The shift in the molecular mass of about 80 Da indicates the protein is predominantly monophosphorylated, and to a smaller extent also bisphosphorylated. LC-MS/MS analysis of ArgC digested protein as described in Fig S3A, showing that monophosphorylation for PKD1 CAT is primarily located on S 742 .

Extended Experimental Procedures
Peptide mass spectrometry and phosphorylation site mapping 1 µg of protein was denatured in 8 M urea 50 mM ammonium bicarbonate (ABC), reduced with 10 mM DTT for 15 min at room temperature, alkylated with 20 mM IAA for 20 min at room temperature in the dark, excessive IAA quenched with 10 mM DTT and then diluted to 0.6 M urea with 50 mM ABC. After addition of 10 mM CaCl2 the protein was digested overnight using mass-spec grade ArgC (Roche) at 37°C. The digestion was stopped with 1% trifluoroacetic acid (TFA) and the peptides were desalted using custom-made C18 stagetips (1).
The peptides were separated on an Ultimate 3000 RSLC nano-flow chromatography system (Thermo-Fisher), The spectral library for PRM data analysis in Skyline was generated as follows: peptides from one replicate with the highest spike-in concentration were identified using X!Tandem version Vengeance (2015.12.15.2) (4). The search was conducted using SearchGUI version 3.3.5 (5) against the PKD1 target sequence alone, as a target/decoy approach is not applicable to PRM data. The identification settings were as follows: Trypsin, specific, with a maximum of 2 missed cleavages; 10.0 ppm as MS1 and 20.0 ppm as MS2 tolerances; Carbamidomethylation of cysteine was set as a fixed and acetylation of protein N-term and oxidation of methionine as variable modifications. Peptides and proteins were inferred from the spectrum identification results using PeptideShaker version 1.16.31 (6). Given the small database size, FDR validation was omitted and instead the PKD1 spectra manually inspected in PeptideShaker, exported as mzident.ml files and imported to Skyline using a score cut-off of 0.95. The spectra matching transitions at the peak apex were further manually validated in Skyline, with all peptides showing plausible fragmentation spectra, consistent with the shotgun measurements and concentration-dependent behaviour in the dilution series.
Data analysis, including manual validation of all peptides and their transitions (based on retention time, relative ion intensities, and mass accuracy), and relative quantification was performed in Skyline. Three of the eight target peptides were removed from the analysis due to strong interfering signals from the background matrix at lower concentrations, or in case there were less than three transitions detectable over the whole concentration range. Up to ten of the most intense non-interfering transitions of the five target peptides were selected and their peak areas were summed for peptide quantification (total peak area). Peptide intensities were summed to protein intensities. To correct for minor variations in sample injection amounts and instrument stability, the total ion current of all MS1 scans between 10 and 60 min of the gradient was summed for each run and used as a global normalization factor. The signals of un-spiked samples were subtracted from those of the dilution series of spiked samples and scaled for easier interpretation. The decimal logarithm of the signal intensity was plotted against the decimal logarithm of the spiked protein concentration and Origin 7 was used to fit the double log-plot data points to a linear regression model with the formula y = A +B*x (Fig 2D, Fig S2D). The linear regression was used to calculate the amount of endogenous protein per cell. The molar concentration of cellular PKD1 was then calculated assuming a total cell volume of 1800 µm 3 for HEK293T cells (7). Since a large proportion of these cells can be assigned to the nucleus, a substantially smaller volume of 822 µm 3 was assumed for the cytosolic volume of these cells (7).
As a rough assessment of the cellular concentration range of PKD1 in INS-1 cells the PRM assay was repeated in two biological replicates based upon the three peptide sequences shared between rat and human. Of these three peptides two could be used for quantification according to the same criteria as described above, although only one peptide could be detected in all spiked and un-spiked samples in both biological replicates. Therefore, peptide intensities were not summed but analysed individually and the data was not globally normalized. The unspiked peptide signals were subtracted from the spiked samples before performing linear regression analysis.
From the regression the amount of endogenous protein per lysate was calculated. The cellular molar concentration of PKD1 in INS-1 cells was calculated using the molecular weight of the spiked proteins, the number of cells that had been counted prior to lysis and the volume of a rat beta cell of 1020 µm 3 which has been previously reported (8).

Experimental design and statistical rational of mass spectrometry experiments
The applied PRM analysis falls in the Tier 3 category (9). Lysates and external reference curve standards were prepared and measured in three biological replicates for HEK293T cells, which should allow a reasonable estimation of the concentration range. The additional experiment on INS-1 cells should serve only as a rough estimate of the concentration range without offering the full quantitative accuracy of the HEK293T experiment.
The non-spike-in lysate samples were run before the spike-in standards to prevent carry-over and blank runs were used to determine baseline signal. The standards were measured in the order of increasing concentration.
Injection of 25 ng of HeLa digest on a 30 min gradient were used to monitor system performance before and after the sample batch by determining number of PSMs and other performance parameters as described in (10) showing no significant differences over the course of the experiment. In addition, all samples were globally normalized to total ion current as described above.
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository (11) with the dataset identifier PXD013216 and PXD013232.