Activation of MAP3K DLK and LZK in Purkinje cells causes rapid and slow degeneration depending on signaling strength

The conserved MAP3K Dual-Leucine-Zipper Kinase (DLK) and Leucine-Zipper-bearing Kinase (LZK) can activate JNK via MKK4 or MKK7. These two MAP3Ks share similar biochemical activities and undergo auto-activation upon increased expression. Depending on cell-type and nature of insults DLK and LZK can induce pro-regenerative, pro-apoptotic or pro-degenerative responses, although the mechanistic basis of their action is not well understood. Here, we investigated these two MAP3Ks in cerebellar Purkinje cells using loss- and gain-of function mouse models. While loss of each or both kinases does not cause discernible defects in Purkinje cells, activating DLK causes rapid death and activating LZK leads to slow degeneration. Each kinase induces JNK activation and caspase-mediated apoptosis independent of each other. Significantly, deleting CELF2, which regulates alternative splicing of Map2k7, strongly attenuates Purkinje cell degeneration induced by LZK, but not DLK. Thus, controlling the activity levels of DLK and LZK is critical for neuronal survival and health.


Introduction
Mitogen-activated protein kinase (MAPK) signaling pathways play important roles in neuronal development and function, and aberrant regulation of MAP kinases is associated with many neurological diseases, such as Parkinson's disease (PD), amyotrophic lateral sclerosis (ALS) and Alzheimer's disease (AD) ( Thomas and Huganir, 2004;Schellino et al., 2019;Hotamisligil and Davis, 2016;Hollville et al., 2019;Adib et al., 2018). The MAPK cascade involves MAP3Ks (MAP kinase kinase kinases), MAP2Ks and MAPKs that together form a phosphorylation relay and activate downstream signaling events in response to external or internal stimuli. The mammalian MAP3K DLK (Dual leucine zipper kinase, or MAP3K12) and LZK (Leucine zipper kinase, or MAP3K13) are members of an evolutionarily conserved family that includes C. elegans DLK-1 and Drosophila DLK/Wallenda (Jin and Zheng, 2019). These MAP3Ks act as upstream kinases for JNK and p38 MAP kinase, and are now known as key players in neuronal stress response network both under acute injury and in chronic neurodegenerative diseases (Jin and Zheng, 2019; Adib et al., 2018;Farley and Watkins, 2018). An emerging theme is that while activation of these kinases triggers seemingly common pathways, the outcome is highly context-specific both in terms of cell types and forms of insults.
Both DLK and LZK show broad expression in the nervous system. Several studies have investigated roles of DLK in the development of the nervous system. Constitutive Dlk knockout mice die perinatally, and different regions of developing brain display varying degrees of altered axon fibers, abnormal synapses and increased neuronal survival (Hirai et al., 2006;Hirai et al., 2011;Nakata et al., 2005;Collins et al., 2006;Lewcock et al., 2007).
However, mice with adult deletion of Dlk survive and show no detectable abnormalities (Le Pichon et al., 2017;Tedeschi and Bradke, 2013). Under traumatic insults, DLK activity is reported to increase and trigger a variety of cellular responses. For example, sciatic nerve injury induces DLK-dependent pro-regenerative responses in dorsal root ganglia (DRG) sensory neurons (Shin et al., 2012;Shin et al., 2019). In the central nervous system (CNS), optic nerve injury up-regulates DLK expression in retinal ganglion cells (RGC), which triggers cell death in many RGCs and also promotes axon growth from surviving RGCs (Watkins et al., 2013;Welsbie et al., 2013). In a mouse model for stroke, increased DLK expression in pre-motor cortex is suggested to promote motor recovery (Joy et al., 2019). Increased DLK activity is also reported in animal models of neurodegeneration, and genetically or pharmacologically inhibiting DLK in the aged PS2APP mice for AD and the SOD1 G93A mice for ALS has resulted in some neuroprotective effects (Chen et al., 2008;Le Pichon et al., 2017). Intriguingly, in human iPSC derived neurons treated with ApoE4, a protein associated with an increased risk for AD, DLK is rapidly up-regulated and enhances transcription of APP (Huang et al., 2017). As numerous approaches now target DLK for drug discovery (Siu et al., 2018), it is important to investigate how the pleiotropic effects of manipulating DLK in different cell types influence disease progression. In comparison, despite the fact that LZK was also discovered 20 years ago (Sakuma et al., 1997;Holzman et al., 1994), the in vivo roles of LZK are only beginning to be explored. In a mouse model of spinal cord injury, LZK is upregulated in astrocytes and mediates reactive astrogliosis (Chen et al., 2018). Emerging studies show that LZK can cooperate with DLK in RGC to promote cell death after optic nerve injury and in DRG for axon degeneration (Welsbie et al., 2017;Summers et al., 2020).
Here, we dissect the roles of the two kinases in the cerebellar Purkinje cells. We analyzed genetic deletion mice for each kinase, and also developed transgenic mice that allow for Cre-mediated expression of DLK or LZK. Biochemical studies have shown that DLK and LZK undergo auto-activation via leucine-zipper mediated dimerization, and such autoactivation is dependent on the protein abundance (Nihalani et al., 2000;Ikeda et al., 2001b).
Therefore, elevating expression of DLK or LZK is a proxy to its activation of the downstream signal transduction. We find that deletion of DLK and/or LZK, singly or in combination, from Purkinje cells, does not affect their development and postnatal growth. In contrast, induced expression of DLK in Purkinje cells causes rapid degeneration, whereas elevating LZK expression induces a slow degeneration. Strikingly, we find that deleting the RNA splicing factor CELF2 ameliorates Purkinje cell degeneration induced by LZK, but not DLK, partly via regulating alternative splicing of Mkk7, a MAP2K. These findings provide important insights to the understanding of neurodegenerative processes.

Normal development of cerebellar Purkinje cells in the absence of DLK and LZK
Both DLK and LZK are expressed in cerebellar neurons, with high levels of DLK observed in the molecular layer of adult cerebellum (Hirai et al., 2005;Suenaga et al., 2006;Goodwani et al., 2020). Dlk knockout (KO) mice die soon after birth, and cerebellar architecture is grossly normal (Hirai et al., 2006). The roles of Lzk in neuronal development remain unknown. To address the function of Lzk and further probe into the interactions between the two kinases, we generated an Lzk KO mouse line using CRISPR-editing to delete the entire kinase domain (

Elevating DLK expression in Purkinje cells causes rapid degeneration via apoptosis
Increased expression of DLK or LZK has been reported under traumatic injury or other stress conditions (Shin et al., 2012;Shin et al., 2019;Watkins et al., 2013;Welsbie et al., 2013;Joy et al., 2019;Chen et al., 2018;Huang et al., 2017). We next investigated how elevating expression of DLK and LZK, hence activation of these kinases, affects neurons. To this end, we generated two transgenic mouse lines by inserting a Cre-inducible transgene of Dlk or Lzk at the Hipp11 (H11) locus, named H11-Dlk iOE or H11-Lzk iOE , respectively ( . The molecular layer of the cerebellum in these mice was significantly thinner than that in the littermate control mice from P10 to P21 (Figure 2 F). Purkinje cell degeneration is known to be associated with increased reactivity of astrocytes and microglia (Cvetanovic et al., 2015;Lobsiger and Cleveland, 2007;Lattke et al., 2017). Indeed, we observed increased expression of GFAP and IBA1 (detecting both microglia and macrophage) in these mice at P21, compared to control mice ( To determine if targeted expression of DLK induced activation of the JNK signaling, we co-immunostained for phospho-c-Jun (p-c-Jun) and Calbindin on cerebellar tissues of P6 mice. While many p-c-Jun signals were likely from granule neurons as they did not overlap with Calbindin + Purkinje cells in both mutant and control mice, we observed that DLK activation induced substantially increased p-c-Jun in Purkinje cells of PV Cre/+ ;Dlk iOE/+ mice, compared to littermate controls (Figure 3 A-B). We also asked if the loss of Purkinje cells involved apoptosis using the TUNEL assay. During early postnatal cerebellar development, multiple types of cells undergo apoptosis, including those in the granular and the molecular layers of the cerebellar cortex (Cheng et al., 2011). Indeed, in the control littermates, we observed many TUNEL signals at P5, which decreased over the following postnatal days ( Figure 3-figure supplement 1 A-B). In the PV Cre/+ ;Dlk iOE/+ mice, the number of apoptotic cells at P5 was comparable to that in control, but continued to rise over the next 10 days, reaching peak levels around P15 (Figure 3-figure supplement 1 A-B). Importantly, some TUNEL signals detected in P15 PV Cre/+ ;Dlk iOE/+ mice co-localized with tdTomato-labeled Purkinje cells (Figure 3-figure supplement 1 C). Furthermore, a significant portion of the Purkinje cells in the P15 PV Cre/+ ;Dlk iOE/+ mice were positively stained for cleaved (and thus activated) caspase-3 (Figure 3 C-D), a molecular marker for apoptosis (Elmore, 2007).
Collectively, these results show that elevating DLK expression in Purkinje cells activates the JNK pathway and causes early-onset, rapid degeneration through apoptotic cell death.

DLK activation disrupts dendritic cytoskeleton
DLK is known to be localized to neuronal processes (Suenaga et al., 2006) and regulates microtubule stability (Simard-Bisson et al., 2017;Valakh et al., 2015;Hirai et al., 2011). We performed immunostaining using anti-DLK antibodies on cerebellar section of PV Cre/+ ;Dlk iOE/+ mice and detected DLK expression in the somas, dendrites and axons of Purkinje cells Dendrite swelling is associated with major disorganization of the cytoskeleton network (Cupolillo et al., 2016;Liu et al., 2015;Hoskison et al., 2007). We next assessed how the microtubule cytoskeleton was altered by immunostaining for microtubule-associated protein 2 (MAP2), which is expressed in dendrites of Purkinje cells (Dehmelt and Halpain, 2005).
We found that MAP2 levels were significantly decreased in dendrites of Purkinje cells in PV Cre/+ ;Dlk iOE/+ mice, compared to the levels of Calbindin as well as to control mice at P15 (Figure 3 G). The neurofilament protein NF-200 is present in both dendrites and axons and implicated in axon growth and regeneration (Wang et al., 2012). By immunostaining, we observed decreased levels of NF-200 in axons and dendrites of Purkinje cells in PV Cre/+ ;Dlk iOE/+ mice (Figure 3 H). Together, these data are consistent with the notion that DLK regulates the neuronal cytoskeleton, and further suggest that the dendritic cytoskeleton in Purkinje cells is highly susceptible to disruption upon aberrant activation of DLK.

Elevating LZK expression in Purkinje cells causes progressive degeneration
In contrast to the early lethality of PV Cre/+ ;Dlk iOE/+ pups, the PV Cre/+ ;Lzk iOE/+ mice survived to older adults (observed up to 8 months). The adult mice had low body weight, compared to control mice (  Biochemical studies have shown that two MAP2K, MKK4 and MKK7 act downstream of DLK and LZK to activate JNK (Hirai et al., 2011;Le Pichon et al., 2017;Huang et al., 2017;Ikeda et al., 2001a;Chen et al., 2016b;Ikeda et al., 2001b;Holland et al., 2016;Merritt et al., 1999). However, in vivo evidence for how each MAP2K contributes to DLK and LZK induced signal transduction cascade in neurons is limited (Yang et al., 2015). Recent studies of T-cell activation have reported that the activity of MKK7 is regulated through alternative splicing of its exon 2, which encodes a small peptide within the JNK docking site in MKK7 (Martinez et al., 2015) ( Figure 5-figure supplement 1 A). During T-cell activation, the RNA splicing factor CELF2 promotes skipping of this exon, favoring the production of a short isoform of MKK7 that has high potency to activate JNK (Martinez et al., 2015;Ajith et al., 2016).
To test if this regulation of Mkk7 alternative splicing has functional significance in neurons, we generated PV Cre/+ ;Celf2 fl/fl ;Dlk iOE/+ and PV Cre/+ ;Celf2 fl/fl ;Lzk iOE/+ mice, along with To address whether CELF2 was involved in LZK signaling in Purkinje cells, we examined Mkk7 exon 2 splicing. By qRT-PCR analysis we detected that Celf2 deletion reduced the ratio of mRNA of the short isoform (Mkk7-S) to the long isoform (Mkk7-L) by ~20% in cerebellum of PV Cre/+ ;Celf2 fl/fl ;Lzk iOE/+ mice, compared to PV Cre/+ ;Lzk iOE/+ mice ( , which as a full-length protein prevents caspase activation, but the cleaved product promotes apoptosis (Gross et al., 1999). Western blot analysis of cerebellar protein extracts from P21 mice when minimal Purkinje cell degeneration was detected in PV Cre/+ ;Lzk iOE/+ showed increased pro-apoptotic cleavage products of Bcl-xL, compared to control samples ( Figure 6-figure supplement 1 A, C). All together, these data support a conclusion that Celf2 deletion attenuates LZK-induced JNK signaling, and provide in vivo evidence that MKK7 is a functional mediator of LZK signaling in Purkinje cells.

DLK and LZK can induce Purkinje cell degeneration independent of each other
DLK and LZK have a nearly identical kinase domain, and are reported to bind and be coimmunoprecipitated from mouse brain (Pozniak et al., 2013). Recent studies have shown that in injured RGCs or DRGs the two kinases may have redundant or synergistic interactions (Welsbie et al., 2017;Summers et al., 2020). We next addressed whether the

Discussion
In this study, we have used cerebellar Purkinje cells to gain a systematic understanding of the function of DLK and LZK, two closely related kinases that have emerged as key players in neural protection under injury and disease (Adib et al., 2018;Jin and Zheng, 2019;Farley and Watkins, 2018). We employed both conditional KO and transgenic mouse models to manipulate levels of DLK and LZK expression. We find that while deleting one or both kinases in Purkinje cells postnatally did not affect neuronal development and animal health, activating DLK or LZK, through elevating their expression, causes Purkinje cell degeneration. Our Cre-inducible DLK and LZK transgenes have the same design and are inserted in the same H11 locus to avoid position effect on transgene expression. Despite the similarly targeted transgenes, we found that DLK elevation triggers rapid degeneration of Purkinje cells, while LZK elevation causes slow degeneration. Each kinase activates JNK signaling, measured by increased phosphorylated c-Jun, and induces apoptosis. Each kinase can induce neuron degeneration in the absence of the other. Importantly, we show that deletion of Celf2 strongly attenuates Purkinje cell degeneration caused by LZK, but not DLK, activation, providing further evidence for a signaling pathway-specific effect for each kinase activation rather than a generic, secondary effect of overexpressing any kinase. As Purkinje cells and cerebellum are not essential for animal viability, we interpret that the lethality of PV Cre/+ ;Dlk iOE/+ pups is likely due to disruption of other parvalbumin-expression neurons, with the underlying basis remaining to be addressed in future studies. All together, these data demonstrate the utility of our transgenic mice for dissecting cell-type specific roles of these kinases and their signaling pathways.
DLK and LZK share a kinase domain that is ~90% identical and can activate the JNK signaling pathway through two MAPKK, MKK4 or MKK7 (Hirai et al., 2011;Le Pichon et al., 2017;Huang et al., 2017;Ikeda et al., 2001a;Chen et al., 2016b;Ikeda et al., 2001b;Holland et al., 2016;Merritt et al., 1999). Several studies have supported MKK4 as a major mediator for DLK in RGCs and DRGs (Yang et al., 2015). Currently, little is known which MAPKK mediates LZK signaling. Our data show that DLK activation in Purkinje cells leads to robust JNK signaling, compared to LZK activation. The observation that Celf2 deletion did not affect any phenotypes caused by DLK activation could be due to a combination of the strong JNK activation and the rapid time course of cell death. In contrast, deletion of CELF2 significantly reduced the activation of c-Jun and almost completely rescued the Purkinje cell degeneration caused by LZK activation. These data are consistent with the role of Celf2 in regulating alternative splicing of Mkk7 (Martinez et al., 2015), and support that MKK7 is a functional downstream kinase for LZK in vivo.
Numerous studies have revealed roles of DLK in axon growth, regeneration and/or degeneration (Tedeschi and Bradke, 2013; Jin and Zheng, 2019). However, not much is known about roles of LZK in neurons. Our data show that LZK activation decreased neurofilament levels in the molecular layer of cerebellum and caused disorganization of the pinceau at the axon initial segment of Purkinje cell. DLK is known to regulate microtubule stability (Simard-Bisson et al., 2017;Valakh et al., 2015;Hirai et al., 2011), and several microtubule-associated proteins such as SCG10, DCX and MAP2 are JNK substrates (Chang et al., 2003;Gdalyahu et al., 2004;Tararuk et al., 2006;Björkblom et al., 2005). We find that both DLK and LZK overexpression significantly decreased MAP2 levels in dendrites of Purkinje cells, and that Celf2 deletion restored MAP2 levels in PV Cre/+ ;Celf2 fl/fl ;Lzk iOE/+ mice.
In addition, DLK activation caused dendrite swelling of Purkinje cells, and many of the swollen dendrites also had cleaved caspase-3 signals. Activated caspase-3 in dendrites has been shown to cause cleavage of microtubules and local pruning of dendrites and spines (Ertürk et al., 2014;Khatri et al., 2018). These data are consistent with known roles of JNK regulation of microtubule associated proteins.  Chen et al., 2008;Le Pichon et al., 2017;Huang et al., 2017;Chen et al., 2018), indicating that altered signaling of this pathway may be a prevalent feature in CNS injury and diseases. Along this line, genetic studies of DLK in both invertebrate and vertebrate species revealed prominent developmental defects with DLK activation rather than inactivation (Zhen et al., 2000;Schaefer et al., 2000;Wan et al., 2000;Grill et al., 2016). As DLK and LZK activity exhibits high cell-type and context-dependent specificity, our transgenic mice offer valuable gain of function models to study their signaling pathways with the ease for temporal and spatial manipulation. The knowledge learned will advance our understanding of how diverse neuronal types respond to insults to the nervous system.

Mice
All animal protocols were approved by the Animal Care and Use Committee of the University of California San Diego. Wild-type C57BL/6J mice and PV Cre mice (Stock No: 017320) were purchased from The Jackson Laboratory.
Lzk knockout mice were generated in the UCSD Transgenic and Knockout Mouse Core, using CRISPR-Cas9 technology (Ran et al., 2013). Briefly, sgRNA sequences targeted to the kinase domain were designed using online tools (http://crispr.mit.edu) ( Table 1). The selected sgRNAs were annealed, and then cloned into PX330 backbone digested with BbsI.
Effectiveness of sgRNAs was tested using Surveyor nuclease assay (Surveyor Mutation Detection Kit, IDT, 706020). To make sgRNAs, DNA fragments containing T7 promoter followed by sgRNA were first amplified using primers YJ12532-12535. The purified DNAs were then in vitro transcribed using MEGAscript T7 Transcription Kit (Invitrogen, AMB13345), and the resulting transcripts were purified using MEGAclear-96 Transcription Clean-Up Kit (Invitrogen, AM1909). The sgRNAs and Cas9 mRNA were injected into zygotes from C57BL6, which were then implanted into the CD1 surrogate mothers. Two KO mouse lines were obtained and the one containing a deletion of the entire kinase domain was used in this study. Lzk fl and Dlk fl mice were reported in (Chen et al., 2016b). Dlk fl mice were a kind gift of Dr. Lawrence B. Holzman (Univ. Penn). Celf2 fl mice were described previously (Chen et al., 2016a). Primers for genotypes are listed in Table 2.
Transgenic conditional overexpressing Dlk iOE and Lzk iOE mice were made by Applied StemCell, Inc (Milpitas, CA), using TARGATT TM Technology (Tasic et al., 2011). A mixture of plasmid pBT378-LSL-3X Flag-Dlk-T2A-tdTomato, or pBT378-LSL-1X Flag-Lzk-T2A-tdTomato DNA, and in vitro transcribed ϕC31 integrase mRNA was microinjected into the pronucleus of zygotes from a FVB strain that has the Att recombination landing site inserted in H11 locus, which were then implanted into the CD1 surrogate mothers. The founder heterozygous mice were bred three times to pure C57BL/6J background.

Histology and immunocytochemistry
Mice were transcardially perfused with 0.9% saline solution and then 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) (pH 7.2-7.4). Brains were dissected, post-fixed in 4% PFA overnight at 4℃, and then transferred to 30% sucrose in PBS, prior to embedding using O.C.T compound (Fisher Healthcare, 4585) on dry ice. 25 μm thick sagittal sections were collected on a cryostat (Leica, CM1850) into PBS with 0.01% NaN3. For histology analysis, free floating tissue sections were loaded to the slide, and then sequentially stained with hematoxylin and eosin (H&E staining Kit, abcam, ab245880). For immunostaining, free floating tissue sections were washed twice in PBS with 0.2% Triton X-100, blocked with 5% goat serum in PBS with 0.4% Triton X-100 for 2 hours at room temperature, then incubated with primary antibodies (Table 3) diluted in PBS with 0.2% Triton X-100 and 2% goat serum overnight at 4℃. Alexa Fluor 488-conjugated and Alexa Fluor 647-conjugated secondary antibodies (Invitrogen) were incubated for 2 hours at room temperature, followed by staining with DAPI (14.3 μM in PBS, Thermo Fisher Scientific, D1306) for 10 min. Stained sections were mounted with prolong diamond antifade mountant (Thermo Fisher Scientific, P36970).

TUNEL staining
The DeadEnd Fluorometric TUNEL System (Promega, G3250) was used with a modified protocol. Free floating tissue sections were washed twice in PBS, and then loaded to the slide. The slides were dried at 65℃ for 5 min, then immersed into PBS with 0.5% Triton X-100 and incubated at 85℃ for 20 min, followed by three times rinsing with PBS. The slides were incubated with equilibration buffer at room temperature for 5 min, and then incubated with reaction mix (equilibration buffer: nucleotide mix: rTdT enzyme = 45:5:1) at 37℃ for 1 hour in a humidity chamber. The reactions were stopped by incubating the slide with 2X saline-sodium citrate (SSC) buffer at room temperature for 10 min. After three washes with PBS, the slides were incubated with DAPI (14.3 μM in PBS) at room temperature for 15 min, followed by three washes with ddH2O, and then mounted with prolong diamond antifade mountant (Thermo Fisher Scientific, P36970).

Image acquisition and analysis
The slides of H&E staining were scanned with Nanozoomer 2.0-HT digital slide scanner (Hamamatsu) in brightfield at 20X magnification. The images were processed using NDP.view2 Viewing software (Hamamatsu) and ImageJ software (NIH). Fluorescence images of paired WT and mutant samples were acquired on a Zeiss LSM 710 confocal microscope. The images were taken as Z-stack under identical settings, and the maximum intensity projection images were processed using ImageJ software (NIH). For image quantification, three midline parasagittal sections per brain and at least 3 brains per genotype of given age were analyzed and data was averaged. Cells were counted using the cell counter plugin for ImageJ (NIH). Analyses of cell numbers for Calbindin + PCs, p-c-Jun + PCs, dendrite swelling + PCs or cleaved caspase-3 + PCs were performed by counting the soma of each PC in the entire lobules. The thickness of the molecular layer visualized by Calbindin staining was assessed for lobule V/VI in midline sections by measuring the perpendicular distance from the molecular layer-facing edge of a Purkinje cell soma to the outer edge of the molecular layer. Cerebellum area was calculated by outlining the perimeter of the outer edges of the sagittal sections of cerebellum. TUNEL + cells were counted by analyzing particles after adjustment of threshold and watershed. The particles with area larger than 8 μm 2 were measured. TUNEL + cell density was calculated the number of TUNEL + cells in entire cerebellum divided by entire cerebellum area. GFAP or IBA1 immunofluorescence intensity density was calculated by dividing the GFAP or IBA1 immunofluorescence intensity of entire cerebellum by the entire cerebellum area. For p-c-Jun immunofluorescence intensity quantification, 50 sampling area surrounding a single p-c-Jun + nucleus (ROI being 347.543 μm 2 ) per section were measured. For LZK immunofluorescence intensity quantification, 30 sampling area surrounding a single soma of Purkinje cells (ROI being 352.943 μm 2 ) were measured in the region of interest per section.
Integrated density was averaged after subtraction of background signal and adjustment of threshold.

Immunoprecipitation and western blotting
Dissected cerebella from mice of indicated age were homogenized in an appropriate volume of cell lysis buffer (50 mM Tris.Cl (pH 7.4), 1% Triton X-100, 0.1% SDS, 1 mM EDTA (pH 7.0), 150 mM NaCl, 1% n-Octyl β-D-glucopyranoside, 1 x protease inhibitor cocktail (Roche, 05892970001)) using TissueRuptor II (QIAGEN, 9002756), then lysed for 1 hour on ice, and cleared by centrifugation at 13,000 rpm for 10 min at 4℃. Supernatants were collected and protein concentrations were determined by Pierce BCA protein assay kit (Thermo Fisher Scientific, 23225). For immunoprecipitation experiments, antibody-bound beads were prepared using 2 μg rabbit anti-MAP3K13 polyclonal antibody (Sigma-Aldrich, HPA016497) in 800 μl of lysis buffer with 50 μl of 50% Protein G agarose bead slurry, with gentle rotation at 4℃ for 1 hour. ~1 mg protein lysates were pre-cleared with 50 μl of 50% Protein G agarose bead slurry, then incubated with the antibody-bound beads overnight at 4℃. The beads were washed three times with lysis buffer, and then resuspended in 60 μl lysis buffer and 20 μl 4 x Laemmli Sample Buffer (Bio-RAD, 161-0747), heat shocked in a thermomixer (Eppendorf) at 95℃ for 10 min and analyzed by western blotting. Immunoprecipitated samples were separated by SDS-PAGE using Any kD Mini-PROTEAN TGX Precast Protein Gels (Bio-Rad, 4569034), and then blotted to a PVDF membrane (0.2 μm, Bio-RAD, 1620177) by Mini Trans-Blot Cell (Bio-RAD, 170-3930) at 100 mA for 1 hour. Blots were blocked in 10% non-fat dry milk in PBST (PBS with 0.05% Tween-20) for 1 hour at room temperature, and then incubated with an appropriate concentration of primary antibody in 1% non-fat BSA in PBST at 4℃ for overnight. Afterwards, the membrane was incubated with Horseradish Peroxidase (HRP)-conjugated secondary antibody (GE healthcare, NXA931V or NA934V) in 1% non-fat BSA in PBST at room temperature for 1 hour, followed by detection using enhanced chemiluminescence (ECL) reagents (GE Healthcare, RPN2106).

Animal behavior tests
Mouse behavioral analysis was scored in a genotype blind manner following the protocol described in (Guyenet et al., 2010). Briefly, ledge walking, hind limb clasping, gait, and kyphosis were scored with a scale of 0-3 in each category, resulting in total score of 0-12 points for all four measures at P30, P45, P60, P75, P90, P105 and P120. A score of 0 represents absence of the relevant phenotype and 3 represents the most severe phenotype.
Each test was performed 3 times to ensure reproducibility. For data analysis, the score was calculated for each measure by taking the mean of the three measurements in each mouse.

Statistics
GraphPad Prism 6.0 (GraphPad Software, Inc) was used for all statistical analysis. After assessing for normal distribution, statistical analyses between two groups were calculated with the two-tailed t-test for normally distributed data. For comparison of more than two groups, normally distributed data was calculated with a one-way ANOVA. performed the experiments and data analyses. All authors contributed to writing, reviewing, editing of the manuscript.
Statistics: One-way ANOVA; ns. no significant. Color representation for genotypes in C, E, and F is the same as in B.
G. Representative images of Calbindin staining of cerebellar sections from P120 mice of genotypes indicated. Scale bars: 100 μm.
H. Quantification of total Purkinje cells in all cerebellar lobules at P120.
I. Quantification of cerebellum area at P120.
J. Quantification of the molecular layer thickness of P120 mice in cerebellar lobules V-VI.
F. Quantification of the ratio of DLK relative to β-actin protein levels.
G. The LZK protein levels in cerebellar extracts from P21 mice of genotypes indicated were determined by immunoprecipitation and western blot.
H. Quantification of the ratio of LZK relative to β-actin protein levels.