The biochemical activation mechanism of NDR kinases has been established: MST3 kinase phosphorylates NDR1/2 at its C-terminal hydrophobic residue T444 to activate it (Stegert et al., 2005). NDR1/2 can be activated by okadaic acid (OA) via inhibition of protein phosphatase 2A, facilitating phosphorylation at T444 and the autophosphorylation at S281 (Stegert et al., 2005). MOB1/2 binding to N-terminal region of NDR kinases is required for the release of autoinhibition and maximal activity (Bichsel et al., 2004). Autophosphorylation site S281 is critical for NDR1/2 kinase activity. In order to test NDR1/2's role in dendrite development we first generated dominant negative and constitutively active NDR1 mutants (Figure 1D & E). For dominant negative NDR1 we mutated Ser281 and Thr444 to Alanine (S281A; T444A, NDR1-AA) or catalytic lysine to alanine (K118A, NDR1-KD); both mutants have no kinase activity (Millward et al., 1999; Stegert et al., 2004). To obtain constitutively active NDR1, we replaced the C-terminal hydrophobic domain with that of PRK2 (PIFtide), similar to the generation of constitutively active NDR2 (Stegert et al., 2004).

Kinase activity levels of NDR1 kinase dead (NDR1-KD) and constitutively active (NDR1-CA) mutants were confirmed by in vitro kinase assay with immunoprecipitated NDR1 using an NDR1 substrate peptide as the kinase target (Stegert et al., 2005) (Figure S4A). We then expressed mutant NDR1 proteins together with GFP, to test for their effect on the morphology of cultured hippocampal neurons. Neurons were transfected at DIV6-8 to perturb NDR1/2 function during dendrite development and analyzed at DIV16. With low transfection efficiency it was possible to investigate the cell-autonomous function of NDR1/2 (Figure 2A). We found that NDR1-KD resulted in increased proximal dendrite branching, whereas NDR1-CA caused a major reduction in proximal dendritic branching (Figure 2A & B). Total dendrite branch points were also increased in NDR1-AA and NDR1-KD and reduced in NDR1-CA (Figure 2D). In addition, NDR1-CA resulted in a larger number of branch crossing at 340 μm in Sholl analysis (Figure 2B), indicating that NDR1 activity may produce longer main dendrites at the expense of proximal dendrite branches. Total dendrite length was increased with NDR1-KD and the reduction with NDR1-CA was nearly significant (p = 0.05) (Figure 2F). These results indicate that NDR1 activity inhibits proximal dendrite growth and branching during development. We found that mutant NDR2 expressions in neurons yielded comparable results (data not shown).

To corroborate these findings, we next used NDR1 and NDR2 siRNA to knockdown NDR1/2 function. SiRNA sequences were chosen based on knockdown efficiency of overexpressed NDR1 or NDR2 in HEK293 cells (Figure S2A). These siRNAs partially knocked down the endogeneous protein and were compatible with neuronal viability (Figure S7A). We find that expression of NDR1 and NDR2 siRNA together (but not alone) increased proximal branching, total branch points and total length (Figure 2A, C, E & G), as did dominant negative mutants, supporting NDR1/2's role on inhibiting exuberant growth. This effect was rescued by co-expression of siRNA resistant NDR1 (NDR1*) (Figure 2C, E & G) or siRNA resistant NDR2 (Figure S2F & G), indicating that the effect was indeed due to loss of NDR1/2 kinase function. Our data suggests that NDR1 and NDR2 could have redundant functions in dendrite development. However, it is possible that reduction of NDR1 or NDR2 with their respective siRNA does not bring the protein level below a threshold at which neuronal morphology is altered but cumulative reduction of both leads to the observed defects, and there could be synergistic interaction between NDR1 and NDR2. Taken together with Trc's role on dendrite development of sensory neurons in fly, where trc mutants show increased branching and increased total length of dendrites (Emoto et al., 2004), our findings reveal an evolutionarily conserved function of NDR1/2 in dendrite morphogenesis.

We next employed chemical genetics to manipulate NDR1function in hippocampal cultures. We first mutated the ATP binding pocket gatekeeper Methionine to Alanine (M166A) to make an analog-sensitive NDR1 (NDR1-as), which can use bulky ATP analogs instead of ATP and can be blocked by kinase inhibitors such as 1-Na-PP1 (Bishop et al., 2000). We further introduced two rescue mutations in the kinase domain (M152L and S229A) to increase kinase activity since NDR1-M166A had reduced ATP usage (Figure 1D & E) (Zhang et al., 2005). Although M166A gatekeeper mutation resulted in reduced ATP-γ-S usage (Figure 5B), M152L and S229A rescue mutations led to recovery of ATP-γ-S usage albeit at a lower level than Benzyl-ATP-γ-S, as expected, which were blocked by 1-Na-PP1 (Figure S2B). We transfected neurons with activated NDR1-as at DIV8 and investigated the effect of NDR1 on dendrite development with or without 1-Na-PP1 inhibition from DIV8 to DIV16. We found that 1-Na-PP1 inhibition of NDR1-as resulted in increased proximal branching (50 μm), total branch points and total length (Figure 2A, H, I & J), likely due to a dominant negative effect. Activated NDR1-as treated with the vehicle DMSO resulted in larger dendrite arbor with greater number of branch crossings at 340 μm in Sholl analysis (Figure 1A & H), likely due to increased NDR1 activity. These results further confirm that NDR1 functions to reduce proximal dendrite branching and NDR1 activity may in turn facilitate dendrite arbor expansion distally.

We transfected neurons at DIV6-8 and analyzed them at DIV16. Expression of dominant negative NDR1 (NDR1-KD or NDR1-AA) caused a robust increase of filopodia and atypical protrusion density together with a reduction in mushroom spine density (Figure 3A-C), indicating that NDR1 function is necessary for mushroom spine formation. In contrast, NDR1-CA drastically reduced the total dendritic protrusion density, as a result of significant reduction in mushroom spines, filopodia and stubby spines (Figure 3A-C). Although there was variability in the absolute densities of dendritic spine categories among cultures, decreasing or increasing NDR1 activity consistently induced comparable changes as illustrated here. Robust inhibition of dendritic protrusions by NDR1-CA suggests that excessive NDR1 activity reduces all actin-rich dendritic protrusions. Similar to the dominant negative effects of NDR1 mutants, NDR1siRNA + NDR2siRNA also resulted in increased filopodia and atypical protrusions and decreased mushroom spine densities, which was rescued by co-expression of siRNA resistant NDR1 (NDR1*) (Figure 3A & D). The difference in the extent of filopodia/atypical protrusion increase between dominant negative mutants and siRNA might be due to incomplete knockdown by siRNAs. In addition, the total numbers of dendritic protrusions were not completely restored by NDR1*, suggesting a small non-specific effect of siRNA expression. These data indicate that NDR1/2 are required for efficient formation and/or maturation of mushroom spines. Expression of NDR2-KD and NDR2-CA yielded alterations similar to those induced by the corresponding NDR1 mutants (data not shown).

To determine whether changes in spine morphologies reflected defects in synaptic function, we recorded miniature excitatory postsynaptic currents (mEPSCs) in cultured hippocampal neurons transfected the same way (Figure 3E). We found that mEPSC frequency was reduced by NDR1-KD and by NDR1siRNA + NDR2siRNA, and the effect due to siRNA knockdown was rescued by siRNA resistant NDR1, indicating that NDR1/2 are necessary for active synapse formation (Figure 3F). Interestingly, NDR1-CA also caused a reduction in mEPSC frequency indicating that uncontrolled NDR1 activity can also inhibit active synapse formation (Figure 3F). We did not find a difference in mEPSC amplitude (Figure 3G), suggesting that NDR activity affects the number of active synapses rather than the strength of each synapse. Furthermore, co-immunostaining with post and presynaptic markers indicate that synapses are mostly made directly on dendritic shaft in NDR1-CA expressing neurons in contrast to neurons expressing NDR1-KD or GFP alone (Figure S3A). These observations indicate that mEPSCs in NDR1-CA neurons could originate from synapses on dendritic shafts, and support the notion that reduction in total number of synapses in NDR1-KD and NDR1-CA expressing neurons leads to reduced mEPSC frequency.

In perfused brains from animals subjected to in utero elecroporation we did not observe variable dendritic protrusion morphologies at the resolution imaged (Figure 4C). The predominant protrusion type observed in vivo was mushroom spine. Therefore, we measured spine head diameters and spine density as the functional parameters of these postsynaptic structures. We found that NDR1-KD and NDR1siRNA + NDR2siRNA decreased dendritic spine head diameter (Figure 4G), suggesting that NDR1/2 is required for dendritic spine development in vivo. These results are in agreement with our hippocampal culture results where NDR1/2 promoted mushroom spines and active synapses and limited immature protrusions. It is possible that certain factors that contribute to spine formation and stabilization which are present in vivo are largely absent in cultures. Such differences between cultures and in vivo studies might result in the different spine phenotypes we observe (Figure 3A-D & 4G) caused by similar manipulations of NDR1/2 activity.

We also found that dendritic spine density was reduced in NDR1-CA expressing neurons and was increased by NDR1siRNA + NDR2siRNA in vivo, while we did not observe a significant change with NDR1-KD (Figure 4H). It is possible that NDR1-KD expression level was not sufficient to cause increased spine density in vivo. NDR1/2 participates in limiting dendritic spine density as demonstrated in NDR1-CA expressing cultured neurons (Figure 3A-D). Our data supports that NDR1 activity is necessary in limiting dendritic spine numbers in vivo, as well.

We generated two analog sensitive NDR1-CA's (M166A and M166G). In order to identify which bulky ATP-γ-S analog is most compatible with the mutant kinase, we performed an in vitro kinase reaction with these mutants using NDR1's target peptide as described previously (Stegert et al., 2005). The thiophosphorylated substrate is detected by anti-thiophosphate ester antibody on western blot after esterification by para-nitro benzyl mesylate (PNBM) (Allen et al., 2007). We found that NDR1-CA M166A used Benzyl-ATP-γ-S, however, the ATP analog usage was reduced (Figure S4B). To rescue NDR1 kinase activity we mutated two residues known to be suppressor mutations that can rescue kinase activity when the gatekeeper residue is mutated (Zhang et al., 2005) and obtained NDR1-as-CA with increased kinase activity (NDR1-CA with M166A, M152L and S229A mutations) (Figure 1D & E, Figure 5B). We used this kinase (NDR1-as-CA) in subsequent substrate identification experiments. To perform labeling reactions in which NDR1-as-CA would thiophosphorylate substrates with Benzyl-ATP-γ-S we reacted 10 μg of purified kinase with 1 mg brain lysate protein. Labeled lysate was treated by covalent capture for substrate identification (Blethrow et al., 2008; Hertz et al., 2010). Briefly, labeled protein lysate is digested by trypsin, then thiol containing peptides (including thiophosphorylated substrates and cysteine containing peptides) are captured by thiol reactive resin while non-thiol containing peptides are washed away. In the third step, beads are treated with Oxone to oxidize sulphur and elute phosphopeptides by spontaneous hydrolysis of thiophosphate linkage, whereas cysteine containing peptides remain attached to the beads by thioether bonds. Finally, the eluted peptides are analyzed by LC/MSMS to identify not only the substrates but also the phosphorylation sites, which is a major advantage of the method (Figure 5D).

In each experiment, we included two negative controls (lysate alone and lysate reacted with NDR1-KD) in parallel, with these controls we could disregard abundant proteins that are detected non-specifically. We have carried out substrate labeling from brain lysates 8 times using P3 (2X), P8 (5X) and P13 (1X) brains, to identify potential NDR1 targets. We identified 5 phospho-proteins that are specific to NDR1-as-CA and are detected in more than one experiment (Table 1). Strikingly, 4 of these contained the consensus sequence of HXRXXS/T which is highly similar to the one reported for the NDR1 homolog Cbk1p (HXRRXS/T) (Mazanka et al., 2008) (Table 1). The remaining candidate was not included in the table because the phosphorylation site was preceded by acidic amino acids, rendering it an unlikely NDR1 substrate. In addition, we cultured dissociated cortical neurons on transwell insert culture dishes in order to harvest neuronal processes but not cell bodies to simplify total protein content. We identified one additional candidate with the same consensus site: Rab11fip5 (Rab11 family interacting protein 5) (Table 1). Proteins without the consensus sequence were not included in the table for this experiment.

Thus, we have identified five putative NDR1 substrates: AAK1 (AP-2 associated kinase 1) and Rabin8 (Rab8-GEF), both of which are known to function in vesicle trafficking (Henderson and Conner, 2007; Stenmark, 2009), PI4Kbeta (Phosphatidyl inositol 4 kinase beta), which catalyzes the formation of phosphatidyl inositol 4 phosphate that can give rise to other phosphatidyl inositols (De Matteis et al., 2005), Pannexin-2, a large pore ion channel expressed in the brain (MacVicar and Thompson, 2010) and Rab11fip5 which regulates Rab11 (Horgan and McCaffrey, 2009). Given their high sequence homology especially in the kinase domain and indistinguishable biochemical properties as so far tested, taken together with the ability of NDR1 to rescue for NDR1/2 reduction, NDR1 and NDR2 probably have common substrates.

We were particularly interested in the two most prevalent candidates AAK1 and Rabin8 because both function in intracellular vesicle trafficking. AAK1 was identified in 7 out of 8 experiments and Rabin8 was identified in 3 out of 8 experiments. Moreover, the yeast Rabin8 homolog Sec2p is phosphorylated by yeast NDR kinase Cbk1p (Kurischko et al., 2008) indicating that this kinase regulation might be evolutionarily conserved. We confirmed that AAK1 and Rabin8 were indeed phosphorylated by NDR1 by using direct kinase assay (Figure 5E & F). We reacted purified NDR1-as-CA with purified substrate proteins using Benzyl-ATP-γ-S and detected phosphorylation by anti-thiophosphate ester antibody after esterification with PNBM (Figure 5E & F), a method that avoids the background caused by AAK1 autophosphorylation when using radioactive ATP for detection. We confirmed that the AAK1 phosphorylation site was indeed S635 as identified in mass spectrometry (Figure 5G) since S635A mutant was not phosphorylated (Figure 5E). Furthermore, we generated an antibody which targets AAK1 phosphorylated at S635 (anti AAK1 P-S635). When co-expressed in COS-7 cells NDR1-CA specifically phosphorylated S635 of AAK1 in intact cells (Figure S5E). However, it should be noted that this antibody did not exclusively stain the endogeneous phosphorylated AAK1 by immunocytochemistry.

To explore if AAK1 is downstream of NDR1/2 in dendrite development, we performed epistasis experiments. Total plasmid DNA concentration was kept constant between conditions. Control neurons were transfected with GFP expressing empty siRNA plasmid (pGmir) together with HA expressing empty plasmid (prk5). To observe the effect of NDR1/2 loss of function we transfected NDR1siRNA and NDR2siRNA together with equal amount of empty prk5 vector. This treatment caused an increase in proximal dendrite branching (Figure 6E and F), total dendrite branching (Figure 6G) and length (Figure 6H) as expected. In order to test epistasis, NDR1siRNA and NDR2siRNA were co-transfected with the AAK1-SD-HA construct in prk5 vector. This treatment led to the rescue of dendrite phenotypes induced by NDR1siRNA + NDR2siRNA (Figure 6E-H). In complementary experiments we transfected NDR1-CA with GFP expressing empty siRNA plasmid and observed robust reduction in proximal dendrite branching (Figure 6E & F), total dendrite branching (Figure 6G) and length (Figure 6H). The reduction in dendrite branching and length with NDR1-CA was more pronounced than previous results due to the higher plasmid concentration used here. These effects of NDR1-CA were partially rescued with co-expression of AAK1siRNA (instead of empty siRNA plasmid), indicating that AAK1 activity was necessary to limit dendrite branching. These experiments indicate that AAK1 is downstream of NDR1 for limiting dendrite branching.

We thank Mark Wessels, Peter Soba and Hye-Young Lee for technical help, we thank Chao Zhang for the valuable suggestion of kinase activation mutations. We thank Jon Trinidad for advice on phosphoproteomics and David Maltby for mass spectrometer instrumentation advice. We thank Jan and Shokat lab members for discussion and critical reading of the manuscript. Mass spectrometry was made possible by NIH grants NCRR RR015804 and NCRR RR001614. Financial support for purchase of the LTQ Velos Orbitrap mass spectrometer was provided by HHMI and NIH NCRR01614 (ALB).This work was supported by NARSAD Young Investigator Award to SKU, NARSAD Distinguished Investigator Award to YNJ, HFSP fellowship to WPG, NIH grants R37NS040929 and 5R01MH084234 to YNJ and NIH grant RO1EB001987 to KMS and Genentech predoctoral fellowship to NTH. KMS, LYH and YNJ are investigators of Howard Hughes Medical Institute.