Nemo-like kinase regulates Progranulin levels in the brain through the microglial endocytosis-lysosomal pathway

Genetic variants in Granulin (GRN), which encodes the secreted glycoprotein Progranulin (PGRN), are associated with several neurodegenerative diseases including frontotemporal lobar degeneration, neuronal ceroid lipofuscinosis, and Alzheimer’s disease. These genetic alterations manifest in pathological changes due to a reduction of PGRN expression; therefore, identifying a factor that can modulate PGRN levels in vivo would enhance our understanding of PGRN in neurodegeneration, and could reveal novel potential therapeutic targets. Here, we report that Nemo-like kinase (Nlk) regulates Pgrn levels and its associated neuropathophysiology. Genetic interaction studies in mice show that Grn heterozygote mice on an Nlk heterozygote background display pathological and behavioral phenotypes which mimic Grn knockout mice. Furthermore, biochemical and cell biological studies suggest that Nlk reduction promotes Pgrn degradation via the endocytosis-lysosomal pathway, specifically in microglia. Our results reveal a new mechanism for the regulation of Pgrn in the brain and provide insight into the pathophysiology of PGRN-associated diseases.

Haploinsufficiency of PGRN, caused by heterozygous loss-of-function mutations in one GRN allele, leads to the development of familial frontotemporal lobar degeneration (FTLD), a neurodegenerative disease characterized by atrophy in the frontal and temporal lobes of the brain (Baker et al., 2006;Cenik et al., 2012;Chen-Plotkin et al., 2010;Cruts et al., 2006;Gijselinck et al., 2008). Homozygous loss-of-function mutations in both alleles of GRN cause neuronal ceroid lipofuscinosis (NCL), a neurodegenerative disease in which lipofuscins aberrantly accumulate in the affected tissues (Smith et al., 2012). Finally, specific single nucleotide variants in GRN have been shown to decrease plasma and brain expression levels of PGRN, and are risk factors for Alzheimer's disease (AD), the most common form of dementia (Perry et al., 2013;Wojtas et al., 2012).
Although the involvement of GRN genetic variants in human disease appears to be straightforward, the precise effect of changes to Grn in animal models is far more elusive. In mice, Grn heterozygosity causes little or no change in behavior and neuropathology up to 23 months of age (Ahmed et al., 2010). In contrast, mice lacking both copies of Grn display many behavioral and neuropathological abnormalities, which recapitulate several features of PGRNdeficient FTLD (FTLD-PGRN) and NCL (Filiano et al., 2013;Hafler et al., 2014;Klein et al., 2017;Yin et al., 2010b). In addition, Pgrn has been shown to be a modulator in animal models of neurodegenerative diseases including AD Takahashi et al., 2017) and Parkinson's disease (Van Kampen et al., 2014). Together these studies suggest that decreased expression of PGRN is a direct cause, or indirect risk factor, for a variety of neurodegenerative diseases.
The mechanism through which reductions in PGRN expression lead to diverse neurodegeneration phenotypes remains a largely unanswered question in the field. Pgrn is a secreted glycoprotein which is endocytosed through receptor-and clathrin-dependent endocytosis and delivered to the endosome/lysosome pathway. Previous works suggest that Pgrn plays important roles in the regulation of neurite growth, innate immunity, and lysosome function, all of which are cellular processes studied extensively in the context of neurodegeneration (Ahmed et al., 2007;Gass et al., 2012;Kessenbrock et al., 2008;Klein et al., 2017;Lui et al., 2016;Tanaka et al., 2017;Van Damme et al., 2008). Loss of Pgrn in neurons results in the inhibition of neurite outgrowth (Gass et al., 2012;Van Damme et al., 2008), and Pgrn is involved in the modulation of Wnt and Notch signaling pathways, which regulate axon regeneration, neuronal differentiation, and neuronal survival (Altmann et al., 2016;Raitano et al., 2015;Rosen et al., 2011). In addition, it has been shown that Pgrn plays a role in the regulation of neuroinflammation as an anti-inflammatory factor (Ahmed et al., 2007;Kessenbrock et al., 2008). Lysosomal defects and excessive complement production, which triggers selective synaptic pruning by microglia, occur with loss of Pgrn (Lui et al., 2016). Several other studies also suggest that Pgrn plays active roles in the control of lysosome formation and function Tanaka et al., 2017).
Besides its role in the regulation of neurite outgrowth, neuroinflammation, and lysosomes, several studies have focused on Pgrn catabolism. Pgrn is expressed in a wide range of tissues and cell types, including neurons and activated microglia in the brain (Hu et al., 2010;Petkau et al., 2010). It undergoes receptor-mediated endocytosis and subsequent trafficking to the lysosomes. Extracellular Pgrn can be endocytosed through a Sortilin receptor-dependent orindependent manner and degraded by endosome-mediated delivery to lysosomes (Hu et al., 2010;Zhou et al., 2015). Thus, identifying factors capable of modulating the functioning and/or expression levels of Pgrn itself will be critical, as knowledge of such regulatory factors will not only impart invaluable information regarding the underlying pathophysiology of PGRNassociated neurodegenerative diseases, but also uncover novel therapeutic targets.
Nemo-like kinase (Nlk) is an evolutionarily conserved serine/threonine kinase that plays roles in various signaling pathways including Notch, Wnt, and DNA damage response pathways Meneghini et al., 1999;Ota et al., 2012;Rocheleau et al., 1999;Rottinger et al., 2006;Zhang et al., 2014a). Several lines of evidence suggest that Nlk could act as a potential key protein to modulate the pathogenesis of PGRN-associated neurological diseases and the expression levels of PGRN in the brain. First, changes in Nlk and Pgrn levels affect similar cellular phenotypes. For example, both Nlk and Pgrn are known to be important for neurite outgrowth (Gass et al., 2012;Ishitani et al., 2009;Van Damme et al., 2008) and for the regulation of neuroinflammation (Ahmed et al., 2007;Cenik et al., 2012;Kessenbrock et al., 2008;Saijo et al., 2009). Second, Nlk and Pgrn influence similar molecular pathways that may underlie shared cellular functions. Both Nlk and Pgrn play a role in the modulation of Wnt and Notch signaling pathways, which are important for neuronal survival and axon regeneration (Altmann et al., 2016;Ishitani et al., 1999;Meneghini et al., 1999;Ota et al., 2012;Raitano et al., 2015;Rocheleau et al., 1999;Rosen et al., 2011;Rottinger et al., 2006). Third, and most importantly, deletion of a component in the Nlk-mediated molecular signal transduction pathways leads to neuropathological phenotypes that are similar to AD and FTLD-PGRN (Christopher et al., 2017). Specifically, Nlk-mediated phosphorylation of Nurr1 is essential for the prevention of neuroinflammation by inhibiting neurotoxic gene expression in microglia via the recruitment of CoREST/LSD1 complex for the transrepression of NF-kB activity (Saijo et al., 2009). Subsequently, mice lacking the histone demethylase Lsd1 display behavioral, neuropathological, and molecular phenotypes that highly overlap with those seen in human AD and FTLD-PGRN cases (Christopher et al., 2017).
In this study, we investigated the potential role of Nlk in modulating Pgrn expression and function by using mouse models and cultured cells. Our genetic interaction studies in mice demonstrate that Grn heterozygote mice display neuropathological and behavioral abnormalities in an Nlk heterozygote background at 1 year of age, whose phenotypes are similar to those previously observed in Grn knockout mice. The phenotypes observed in the compound heterozygote arise due to Nlk's function as a positive regulator of brain Pgrn levels, as we show that loss of Nlk leads to a reduction in Pgrn levels. Furthermore, Nlk-mediated regulation of Pgrn levels in the brain occurs mainly through microglia by controlling the clathrin-dependent Pgrn endocytosis and degradation via lysosomes. These results reveal a new mechanism for the regulation of Pgrn levels in the brain and provide insight into the pathophysiology of several neurodegenerative diseases associated with PGRN expression levels, such as AD, FTLD, and NCL.

Decreased Nlk expression induces neuropathological phenotypes in heterozygous Grn mice
Previous studies have documented that Grn +/heterozygous mice do not display any obvious pathological and behavioral phenotypes up to 23 months of age (Ahmed et al., 2010). In contrast, Grn -/null mice exhibit mild to moderate cognitive impairment and pathological abnormalities at 12-18 months of age (Ghoshal et al., 2012;Petkau and Leavitt, 2014;Yin et al., 2010b).
Although Grn +/mice are phenotypically normal, it is possible that Grn +/mice may exhibit a synergistic phenotype in certain genetic backgrounds, such as with reduced Nlk expression. To explore the possibility that the neuropathological phenotypes associated with loss of Grn are affected by Nlk, we conducted a genetic interaction study using mouse models of Grn and Nlk.
Since both genes are located on the same chromosome, we crossed Grn +/mice with Nlk +/animals and evaluated the progeny on a pure C57BL/6J background, investigating whether Nlk +/-Grn +/double heterozygote mice could develop pathological and behavioral changes.
We first investigated whether a 50% reduction of Nlk expression generates pathological phenotypes in Grn +/mice. Previous studies have demonstrated that Grn -/mice recapitulate several key features of PGRN-FTLD and NCL, including microgliosis, accumulation of lipofuscin, and retinal degeneration (Ahmed et al., 2010;Filiano et al., 2013;Hafler et al., 2014;Klein et al., 2017;Yin et al., 2010a). We therefore examined these phenotypes in 1-year-old Nlk +/-Grn +/mice to determine if loss of Nlk promotes pathology in the phenotypically normal To identify changes in microgliosis, Iba1 and CD68 antibodies labeling microglia and microglial lysosomes, respectively, were used for immunohistochemistry. There was a significant increase in the number of Iba1-positive cells in the thalamus of 1-year-old Nlk +/-Grn +/mice compared to age-matched littermate controls ( Figure 1A,B). In addition, CD68positive immunostaining showed enlarged lysosome volume in Iba1-positive microglia in 1-yearold Nlk +/-Grn +/mice ( Figure 1A,C). For both Iba-1 and CD68-positive immunostaining, there was no significant difference between Nlk +/and Grn +/mice with wild-type (WT) control littermates ( Figure 1A-C), further indicating that partial loss of Nlk and Grn contributes to microgliosis. These results are consistent with a previous report that showed microglia activation in Grn -/mice (Lui et al., 2016). To examine lipofuscin deposition, autofluorescence was assessed in the retina, thalamus, and cortex in 1-year-old mice. There was a significant accumulation of lipofuscin in the retina of Nlk +/-Grn +/mice compared to littermate controls Whitney t-test (P<0.05). To explore retinal degeneration in this mouse model, immunohistochemistry was conducted for Brn3a, a transcription factor expressed in most retinal ganglion cells. We found substantial degeneration of retinal ganglion cells in 1-year-old Nlk +/-Grn +/mice ( Figure 1F,G). Taken together, these neuropathological results show that Nlk may play an important role in modulating phenotypes typified by Pgrn reduction.

Nlk reduction promotes behavioral alterations in phenotypically unaffected Grn +/mice
We next examined the effect of Grn +/-Nlk +/compound heterozygosity on mouse behavior. A recent report has documented that Grn -/deficient mice show disinhibition in the context of the elevated plus maze test , which is in agreement with FTLD patient behavioral changes (de Vugt et al., 2006). To determine if the genetic interaction between Nlk and Grn induces this behavioral phenotype in the otherwise normal Grn +/mice, we assessed 1-year-old Grn +/-Nlk +/mice performance on the elevated plus maze (Figure 2A). Since previous reports have found genotype-dependent alterations in levels of anxiety specifically in male Grn -/mice, only males were tested (Kayasuga et al., 2007;Petkau et al., 2012). Nlk +/-Grn +/male mice were less anxious and disinhibited, as they spent more time in the open arms compared to littermate controls ( Figure 2A). Grn +/and Nlk +/mice alone did not have any significant behavioral phenotype relative to WT controls ( Figure 2A).
We also examined cognitive behavior by performing a Morris water maze (MWM) test at 1 year of age. Nlk +/-Grn +/mice had a significant impairment in identifying the hidden platform on trials 4-6 of testing relative to WT controls ( Figure 2B). As expected, Grn +/mice performed similarly to WT controls ( Figure 2B). During the probe trial, Grn +/-Nlk +/mice spent less time in the target quadrant than their WT littermates ( Figure 2C) (p=0.0741, one-way ANOVA with Tukey's multiple comparisons post-hoc test), which was significant with non-parametric Mann-Whitney t-test (P<0.05). To verify that these results were not an effect of alterations in locomotor behavior and visual acuity, swim speed, and vision were also assessed ( Figure 2D,E). There was no significant difference between Nlk +/-Grn +/mice and their littermates in swim speed across all six trials of MWM testing ( Figure 2D). Although we observed significant retinal degeneration in Nlk +/-Grn +/mice ( Figure 1F,G), this did not produce an observable behavioral deficit in vision, as there was no difference in the time spent to identify a visual cue ( Figure 2E). Collectively, these data indicate that Nlk +/-Grn +/mice have normal sensorimotor function. These results are correlated to the observed trending pathological phenotypes (Figure 1), as the thalamus, at least in part, contributes to the anxiety phenotypes and the cortex is important for the recognition phenotype. In addition, consistent with our pathological studies, these behavioral assays support the hypothesis that partial loss of Nlk and Grn additively or synergistically contributes to the appearance of pathological and behavioral phenotypes found in Grn -/mice.

Decreased expression of Nlk reduces Pgrn levels in vivo
We next investigated the molecular basis by which Nlk reduction in Grn +/mice promotes the appearance of pathological and behavioral phenotypes. Since Nlk +/-Grn +/mice show phenotypes similar to Grn -/mice, we wondered whether partial loss of Nlk decreases the expression levels of Pgrn. We therefore measured both intracellular and extracellular Pgrn expression, since it is a secreted protein. We found that the expression levels of both intracellular and extracellular Pgrn proteins were significantly reduced in the cortex of 1-year-old Nlk +/-Grn +/mice compared to Grn +/littermates ( Figure 3A-C). Interestingly, Nlk +/mice alone had a significant reduction in extracellular Pgrn expression relative to WT control littermates, with comparable Pgrn expression between Nlk +/and Grn +/mice ( Figure 3A-C). Immunofluorescence staining for Pgrn in the mouse cortex confirmed our western blotting results, showing a significant decrease in Pgrn expression in Nlk +/-Grn +/relative to Grn +/-, and a decrease in Nlk +/relative to WT controls ( Figure 3D,E). These data suggest that Nlk positively regulates Pgrn expression in vivo.
To determine if a complete loss of Nlk enhances the down-regulation of Pgrn, we temporally deleted Nlk in all cell types during adulthood to circumvent Nlk knockout (KO) prenatal lethality (Ju et al., 2013). We took advantage of mice with a flox-Nlk allele (Canalis et al., 2014) and crossed them to mice expressing tamoxifen (TMX)-inducible Cre driven ubiquitously by a CMV-IE enhancer and chicken b-actin promoter (Hayashi and McMahon, 2002). We generated Nlk fl/fl ; Actin-cre ERT2 mice and their littermate mice on a pure C57BL/6J background. After intraperitoneal TMX injection for 7 consecutive days starting at 6 weeks old, we verified that the expression levels of Nlk mRNA and protein were significantly reduced in the cortex of Nlk fl/fl ; Actin-cre ERT2 mice ( Figure 3F-H). Furthermore, there was a significant reduction of intracellular and extracellular Pgrn protein levels in the cortex of Nlk fl/fl ; Actin-cre ERT2 mice compared to their littermates ( Figure 3F,G).
The down-regulation of Pgrn protein expression in Nlk fl/fl ; Actin-cre ERT2 mice may be due to a similar down-regulation at the transcriptional level. To determine if Grn mRNA was also down-regulated in Nlk fl/fl ;Actin-cre ERT2 mice, we utilized quantitative real-time reverse transcription polymerase chain reaction (RT-qPCR) to assess Grn expression in the cortex of 7week-old mice. Grn mRNA level was not significantly altered by Nlk deletion ( Figure 3H).
Together, these studies indicate that Nlk regulates Pgrn expression in the adult brain posttranscriptionally.

Nlk regulates Pgrn levels via microglia, but not through neurons
Since we have observed that constitutive deletion of a single Nlk allele ( Figure 3A-E) or temporally induced deletion of both copies of Nlk in all cell types ( Figure 3F-H) leads to a significant reduction of Pgrn protein expression in the adult mouse cortex, we next wanted to determine which cell types are essential for the Nlk-mediated regulation of Pgrn expression. Nlk and Grn are expressed in both neurons and glia in the cortex (Zhang et al., 2014b), so we first tested whether Nlk-mediated regulation of Pgrn primarily occurs in neurons, and generated Nlk fl/fl ; Nex-cre mice and their littermate controls on a pure C57BL/6J background. Nex-cre mice express Cre recombinase in the principal neurons of the cortex and the hippocampus and persists through adulthood (Goebbels et al., 2006). The expression levels of Nlk mRNA and protein were significantly reduced in the cortex of Nlk fl/fl ; Nex-cre mice at 6 weeks of age ( Figure 4A-C).
However, we failed to find any significant alterations in the expression levels of Pgrn protein or its mRNA in the cortex of Nlk fl/fl ; Nex-cre mice in vivo ( Figure 4A-C).
Since Nex-cre is expressed mainly in excitatory neurons, but not in inhibitory interneurons (Goebbels et al., 2006), we assayed Pgrn expression in constitutive Nlk +/and Nlk -/primary neuron cortical culture, which contains both excitatory and inhibitory neurons (Lodato and Arlotta, 2015). There was no significant difference in Pgrn protein or mRNA expression levels between WT, Nlk +/-, and Nlk -/cortical neurons in vitro ( Figure 4D-F). Taken together, these data suggest that the Nlk-mediated Pgrn regulation observed in vivo may not occur in neurons.
Transcriptomics studies have indicated that Grn is expressed highly in microglia (Zhang et al., 2014b). To test if microglia could be the primary cell type in Nlk-mediated Pgrn regulation, we generated Nlk fl/fl ; Cx3cr1-cre mice and their littermate mice on a pure C57BL/6J background. Cx3cr1-cre mice express Cre recombinase in microglia in the brain (Yona et al., 2013). We failed to detect a statistically significant decrease in Nlk expression in the whole cortex of Nlk fl/fl ; Cx3cr1-cre mice ( Figure 5A-C), which is likely due to the low fraction of microglia and therefore, minor contribution of Nlk expression from microglia in the whole mouse cortex. Furthermore, we are unable to detect the reduction or loss of Nlk in microglia by immunostaining due to non-specific signals. However, we found that the expression levels of intracellular and extracellular Pgrn protein were significantly reduced in Nlk fl/fl ; Cx3cr1-cre mice compared to littermate controls at 6 weeks of age ( Figure 5A,B). Consistent with results from Nlk fl/fl ; Actin-cre ERT2 mice ( Figure 3H), there was no significant change in Grn mRNA in Nlk fl/fl ; Cx3cr1-cre mice compared to littermate controls ( Figure 5C). To further confirm the role of microglia in Nlk-mediated regulation of Pgrn levels, we also performed primary microglia culture experiments from the cortex of WT and Nlk +/mice and monitored Pgrn expression.
Consistent with our finding in vivo, Nlk haploinsufficiency resulted in a marked reduction of Pgrn expression in primary microglia culture in vitro ( Figure 5D,E).
Since loss of Nlk mediates a down-regulation of Pgrn, increasing Nlk expression should in turn elevate Pgrn expression. To test this, we utilized a murine microglial cell line, BV2. To first verify that BV2 cells recapitulate Nlk-mediated regulation of Pgrn as seen in primary microglia, we intended to generate Nlk-KO cells using CRISPR/Cas9 (Cong et al., 2013). Due to the additional chromosomes that result from the non-diploid nature of the BV2 cancer cell line, incomplete Nlk gene-targeted BV2 cell clones were identified and one clone was selected as Nlk knockdown (Nlk-KD) cells for our analyses, which led to a consistent knockdown efficiency of Nlk (about 70% reduction) at the protein level ( Figure 6A,B) and a significant reduction at the mRNA level ( Figure 6C). Consistent with our results from primary microglia ( Figure 5D,E), there was a significant reduction in intracellular and medium/extracellular Pgrn in Nlk-KD BV2 cells ( Figure 6A,B), and no change in Grn mRNA ( Figure 6C), validating BV2 cells as a model for identifying the mechanism underlying Nlk's moderating effect on Pgrn levels.
To determine if overexpression of Nlk, and the kinase activity of Nlk, affect the regulation of Pgrn expression, we transiently transfected Flag-tagged Nlk and examined the expression levels of Pgrn protein in BV2 cells ( Figure 6D,E). Consistent with the loss-offunction studies, increased expression of wild-type Nlk (Nlk-WT) strongly increased extracellular, but not intracellular, Pgrn expression levels in BV2 cells ( Figure 6D,E). In contrast, kinase-inactive Nlk (Nlk-T298A) had no effect on the regulation of Pgrn in BV2 cells. T298A is a threonine to alanine substitution at residue 298 in the catalytic domain of Nlk, leading to defective kinase activity. Taken together, these data strongly suggest that Nlk deficiency decreases, and its overexpression increases, Pgrn levels in a kinase activity-dependent manner in microglia both in vivo and in vitro.

Nlk controls Pgrn endocytosis in microglia in a clathrin-dependent manner
We next sought to elucidate the cellular mechanism through which Nlk regulates Pgrn expression levels in microglia. We first examined the specific subcellular localization of Pgrn in WT and Nlk +/primary microglia by immunofluorescence staining. We found that the expression of Pgrn in WT primary microglia was weak and mainly diffuse ( Figure  . This suggests that Pgrn may be trafficked into the endosome/lysosome-like vesicles in Nlk +/primary microglia, since Pgrn is known to co-localize with lysosome-associated membrane protein 1 (Lamp1) in neurons (Hu et al., 2010;Tanaka et al., 2017). To test this idea, we co-stained for Pgrn and several lysosome and endosome markers in WT and Nlk +/primary microglia. As expected, Pgrn partially co-localized with Lamp1-and Lamp2-positive lysosomes, as well as with early (Rab5), late (Rab7), and recycling (Rab11) endosomes, which were significantly increased in Nlk +/primary microglia compared to WT littermate controls ( Figure   7A and Figure 7-figure supplement 1), suggesting Nlk mediates trafficking of Pgrn into the endosome/lysosome pathway. Due to the lack of available antibodies for immunofluorescence detection of Nlk, we have not been able to demonstrate the subcellular co-compartmentalization of endogenous Nlk protein to similar intracellular structures as Pgrn.
Collectively, the knowledge of the high degree of co-localization of Pgrn with endosome markers in Nlk +/mice ( Figure 7A and Figure 7-figure supplement 1), the enhanced Pgrn levels in extracellular medium following Nlk overexpression in BV2 cells ( Figure 6D,E), and the fact that Pgrn is a secreted protein that can be endocytosed and delivered to lysosomes (Hu et al., 2010), raise the possibility that Nlk regulates microglial endocytosis. Thus, we decided to test whether indicating that Nlk reduction enhances endocytosis activity in microglia, but not in neurons.
Next, to directly test if Nlk affects microglial endocytosis of Pgrn specifically, we treated WT and Nlk-KD BV2 microglial cells with exogenous Flag-tagged recombinant PGRN proteins.
We measured Flag-tagged PGRN proteins remaining in the medium after a 15 minute incubation and found a significantly decreased level of Flag-tagged PGRN remaining in media from the Nlk-KD BV2 cells compared to that from WT controls ( Figure 7G,H). Since endocytosis can be divided into phagocytosis, pinocytosis, and receptor-mediated endocytosis, we decided to investigate which specific pathway is likely regulated by Nlk. Using Pitstop 2, a clathrinmediated endocytosis inhibitor, we found that Nlk-regulated endocytosis of Flag-tagged PGRN was completely blocked. Although we cannot exclude the possibility that Nlk regulates the exocytosis of PGRN, these data suggest that Nlk plays a role in receptor-mediated endocytosis of PGRN ( Figure 7G,H).

Nlk controls Pgrn levels via lysosome-dependent degradation in microglia
Considering our evidence that Nlk regulates Pgrn uptake via endocytosis in microglia ( Figure   7G,H), it is likely that Pgrn is then degraded via lysosome-dependent degradation. To reconcile the fact that Nlk reduction promotes microglial Pgrn endocytosis, resulting in a transient increase in vesicular Pgrn staining ( Figure 7A,B), and that we observe an overall reduction in intracellular and extracellular Pgrn protein levels in our compound heterozygote animals ( Figure 3A-C), we examined the involvement of Nlk-mediated lysosomal degradation of Pgrn. We treated primary microglia with the lysosome inhibitor Bafilomycin A1 (BafA1), which blocks the final lysosome-mediated degradation step by inhibiting vATPase-mediated acidification of lysosomes, and examined the expression patterns of Pgrn in WT and Nlk +/primary microglia. The increased number and intensity of Pgrn puncta in Nlk +/primary microglia relative to WT controls was completely abrogated by the BafA1 treatment, which further enhanced intensity in both groups ( Figure 8A Taken together, these studies strongly suggest that Nlk regulates Pgrn levels in the brain in a lysosome-dependent manner through microglia, but not through neurons.

Discussion
Heterozygous and homozygous loss-of-function mutations in the GRN gene are common causes of familial FTLD and NCL, respectively (Baker et al., 2006;Cruts et al., 2006). Furthermore, reduced PGRN levels have been linked to other neurodegenerative disorders including AD (Jian et al., 2016;Perry et al., 2013;Wojtas et al., 2012). Despite extensive studies, the biological and molecular mechanisms underlying Pgrn function and metabolism have yet to be completely understood. Furthermore, there is no cure or effective therapeutic to reverse or slow progression of PGRN-associated neurodegenerative diseases. This underlines the need for basic research to better understand Pgrn neurobiology and to identify effective strategies or proteins that could increase Pgrn levels in the brain, which will eventually be useful for the development of effective therapeutics for several neurodegenerative disorders associated with PGRN levels. In this study, we identified such a protein, Nlk, that regulates Pgrn levels and the pathophysiological phenotypes of Grn haploinsufficiency. By combining mouse genetic approaches and cell biological studies, we show that Nlk is a key protein regulating Pgrn levels in the brain via control of the receptor-mediated endocytosis-lysosomal pathways in microglia, and thus modulates the pathophysiology of Grn haploinsufficiency in the brain (Figure 9). Different from GRN haploinsufficiency in humans, mice with a heterozygous loss of Grn fail to produce any obvious behavioral or neuropathological FTLD-PGRN phenotypes (Ahmed et al., 2010;Kayasuga et al., 2007). As a consequence, more attention has been directed towards characterizing mice that are null for Grn, which reproduces several features of human FTLD-PGRN and NCL (Filiano et al., 2013;Hafler et al., 2014;Klein et al., 2017;Yin et al., 2010b).
Studies using Grn -/mice would enhance our understanding of Pgrn function and biology in the brain, and provide some insights into the pathophysiology of FTLD-PGRN. However, there are clear limitations using Grn-deficient mice for such studies, one of which is whether and how increased expression levels of Pgrn would be able to rescue the phenotypes specific to FTLD-PGRN or NCL. Also, the lack of neuropathological phenotypes in Grn +/heterozygous mice or the absence of Grn gene and Pgrn protein in Grn -/null mice lead to an unequivocal desire for the development and identification of more physiologically relevant animal models for FTLD-PGRN study in order to better understand the molecular and cellular basis of neurodegeneration caused by Grn haploinsufficiency.
In this regard, it is very intriguing that this study shows that Grn +/heterozygous mice develop several behavioral and neurodegenerative phenotypes, including cognitive impairments and altered anxiety, as well as lipofuscin accumulation, microgliosis, and retinal neurodegeneration, in the Nlk +/heterozygous background (Figures 1,2), some of which are similarly observed in human patients with GRN haploinsufficiency (Viskontas et al., 2007;Ward et al., 2014;Woollacott and Rohrer, 2016). Consistent with previous studies, the Grn +/heterozygous mice in the WT (Nlk +/+ ) background do not develop any phenotypes (Figures 1,2) (Ahmed et al., 2010;Kayasuga et al., 2007). To our knowledge, this is the first report showing that the genetically heterozygous Grn +/mice can develop relevant cognitive impairments and neurodegeneration phenotypes in a certain genetic background, specifically a 50% Nlk expression reduction, which together leads to 70-80% Pgrn reduction ( Figure 3A-C). Although several pathological phenotypes assessed in the present study do not quite reach statistical significance (p<0.05) by one-way ANOVA in some brain regions, we believe that the statistically significant and trending data (significant by Mann-Whitney t-test) together represent a potential biologically significant threshold of required Pgrn levels below which pathological phenotypes begin to appear in mice. We speculate that the reduction of Pgrn by Nlk haploinsufficiency in the already Grn heterozygote lowers Pgrn protein levels to near threshold levels, and that perhaps a slight decrease in Pgrn dosage even further would generate significant pathological changes in all assessed brain regions.
Grn knockin mouse models have been generated with mutations found in human FTLD patients (PGRN-R504X and PGRN-R493X) (Fujita et al., 2018;Nguyen et al., 2018), and in particular, PGRN-R504X heterozygous mice exhibited Tau-mediated synaptic pathology and cognitive impairment, but not neurodegeneration (Fujita et al., 2018). Further studies are needed to investigate the phenotypic variance in mice of 50% Pgrn level in Grn +/mice versus 50% Pgrn-R504X (with 50% WT Pgrn) level in Pgrn-R504X/+ heterozygous mice. Nevertheless, utilizing diverse Grn mutant mouse models and different genetic approaches will help us to better understand how Grn mutations affect the function and metabolism of Pgrn and its role in the pathogenesis of diverse neurodegenerative diseases including FTLD-PGRN.
The importance of expression levels and the functions of Pgrn in the brain, in particular in neurons and in microglia, have been well established (Kao et al., 2017;Petkau and Leavitt, 2014). A fundamental aim is to understand how Pgrn levels are regulated. In the brain, Pgrn protein is detected highly in neurons and activated microglia by immunostaining (Hu et al., 2010), while Grn mRNA is expressed highly in microglia and lowly in neurons and astrocytes (Zhang et al., 2014b). Conditional deletion of Grn specifically either in neurons or in microglia is not sufficient to cause phenotypes associated with Pgrn deficiency in mice (Petkau et al., 2017a;Petkau et al., 2017b), which may be due to extracellular Pgrn levels being maintained at functionally sufficient levels by different cell types. Our studies provide three significant findings on the regulation of Pgrn levels in the brain: (1) Nlk is critical for the regulation of Pgrn levels in the brain; (2) Nlk regulates Pgrn levels via microglia, but not via neurons; and (3) Nlk-mediated control of Pgrn levels occurs at the level of protein, but not the level of Grn mRNA expression. These findings support the main conclusion that Nlk controls Pgrn protein levels in the brain via microglia, which cannot be compensated for by other cell types. Consistent with this notion, Pgrn reduction levels (about 50% reduction) in the whole cortex of the microgliaspecific KO (Nlk fl/fl ; Cx3cr1-cre) mice ( Figure 5A-C) are equivalent to those in Nlk deletion in all cell types (Nlk fl/fl ; Actin-cre ERT2 ) ( Figure 3F-H), suggesting that microglia are the major cell types by which Nlk plays a fundamental role in regulating Pgrn levels in the brain.
Another important outcome of this study is to reveal the cell biological mechanism underlying the Nlk-mediated regulation of Pgrn levels in microglia. Specifically, Nlk regulates Pgrn levels via control of the endocytosis-lysosomal pathways in microglia, but not in neurons: (1) Nlk negatively regulates microglial endocytosis ( One possibility is that Nlk localizes into endosomes/lysosomes and regulates their dynamics.
Although we have not successfully identified endogenous Nlk localization by antibody staining due to technical issues, it is worth noting that Nlk localizes to both the nucleus and cytoplasm when overexpressed in cell lines, without preferential localization to any specific organelle.
Several important questions are raised by our study. First, what is the molecular basis of the Nlk-mediated cell-type specific regulation of endocytosis? In other words, why does Nlk regulate endocytosis specifically in microglia, but not in neurons, and what are the direct substrate(s) of Nlk that control microglial endocytosis? Our preliminary studies suggest Nlk can phosphorylate several proteins involved in endocytosis, some of which are expressed specifically or very highly in microglia, but not in neurons. Second, what receptor mediates Pgrn endocytosis in microglia? Is it Sortlin since it is known to act as a Pgrn receptor for endocytosis and is able to regulate Pgrn levels (Hu et al., 2010)? Third, what effect does Nlk have on Pgrn secretion?
Finally, considering the potential promising outcomes of Nlk overexpression in a microglia cell culture system ( Figure 6D,E), can the increased expression of Nlk specifically in microglia upregulate Pgrn levels in the brain in vivo in order to rescue or improve the pathophysiology associated with reduced PGRN levels? Given that this regulation is kinase activity dependent, is Nlk a potential druggable target? Collectively, the findings of our study warrant a further, detailed mechanistic examination of the precise Nlk-mediated microglia-modulating effects, including characterization of the effects of varying Nlk levels on synaptic pruning and neuronal numbers in a brain circuit-specific manner, which may be involved in the behavioral changes we report here.
In conclusion, our study provides strong genetic and cell biological evidence that Nlk can control the microglial endocytosis-lysosomal pathway and modulate Pgrn levels and biology in the brain of a Grn +/mouse model. It reveals a new regulatory mechanism for the control of Pgrn levels and, consequently, its associated neurobiology in the brain, which provide insight into the pathophysiology and the therapeutic development of PGRN-associated neurodegenerative diseases. This study also supports the future investigation of potential roles of Nlk and Nlkmediated microglia modifications in diverse neurodegenerative diseases including AD.

Mouse behavioral tests
Elevated plus maze: Elevated plus maze was set at a height of 65cm and consisted of two open white Plexiglas arms, each arm 8cm wide x 30cm long and two enclosed arms (30cm x 5cm) with 15cm high walls, which were connected by a central platform (5cm x 5cm). Individual mice were placed at the center of the maze, facing one of the closed arms and observed for 5 minutes.
All apparatus arms were cleaned with 70% ethanol after every trial. Data acquisition was recorded on a JVC Everio, G-series camcorder, and analysis of time spent on open and closed arms was performed using Panlab's Smart tracking and analysis program, v2.5.
Morris Water Maze Test: Mice at 12 months were tested in a cylindrical tank of 100cm in diameter and 60cm in height. The tank was filled with water at 25°C and the platform was submerged 1cm below the water surface. The tank was divided into four quadrants with different navigational landmarks for each quadrant. The midpoint of the wall in each quadrant was used as the starting location from which animals were released into the water. In the hidden platform acquisition test, mice were allowed to swim freely to search for the escape platform within 60 seconds. The platform location remained constant throughout the test. The time taken to reach the platform was recorded as the escape latency. The mouse was kept on the platform for 10 seconds after it found the hidden platform. If mice failed to find the platform within 60 seconds, it was guided to the platform and placed on the platform for 10 seconds; in this case, the escape latency was recorded as 60 seconds for this trial. The same animal was released from a new starting quadrant 4 minutes after the previous trial. The experiment was repeated with 6 trials per mouse, with one trial every morning and afternoon for 3 consecutive days. The mean escape latency was measured to evaluate the spatial learning ability.
Twenty-four hours after the hidden platform acquisition test, probe trials were conducted by removing the platform. Mice were placed in the quadrant opposite to the removed hidden platform and were allowed to swim freely in the pool for 60 seconds. The percentage of time spent in the area around the original hidden platform was used to indicate long-term memory maintenance. At the end of the probe trial, a colorful flag was placed on the top of the hidden platform, which was opposite to the testing quadrant. The mouse was released from 3 different quadrants 7 times, and the time spent to find the flag was recorded as a measure of visual ability.
The probe trials were recorded on a JVC Everio, G-series camcorder and analyzed using Panlab's Smart tracking and analysis program, v2.5. The observer was blinded to genotype for the duration of behavioral testing.

Western blot analysis
To examine protein expression levels, samples were prepared from the whole cortex of each mouse genotype or the cultured cells. To prepare extracellular protein samples from tissue, dissected mouse cortex was homogenized in PBS with a dounce homogenizer and centrifuged for 10 minutes at 6,000rpm at 4°C, with the resulting supernatant collected for extracellular proteins. Supernatant culture medium from BV2 and primary microglia was also collected for extracellular protein analysis as shown previously (Capell et al., 2011). To prepare intracellular protein samples, the pellet was lysed in lysis buffer (50mM Tris [pH7.5], 150mM NaCl, 0.1% SDS, 0.5% Triton X-100, 0.5% NP-40 and Roche complete protease inhibitor cocktail), rotated at 4°C for 20 minutes, and then centrifuged for 10 minutes at 13,000rpm at 4°C. The supernatant was collected as intracellular proteins. All samples were quantified and 20 or 40µg total protein from each sample was boiled for 10 minutes in sample buffer (BioRad #161-0737), loaded onto a SDS-PAGE, and transferred to a nitrocellulose membrane for western blot analysis. Membrane was blocked by incubation in 5% skimmed milk powder in TBST, incubated with primary antibodies in TBST containing 5% skimmed milk, then incubation in secondary antibodies conjugated with HRP, before detection with ECL reagents. The following primary antibodies were used: mouse anti-Vinculin (Sigma, V9264), sheep anti-Pgrn (R&D, AF2557), and rabbit anti-Nlk (Abcam, ab26050).

Quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR)
The RNA extraction, cDNA synthesis, and qRT-PCR were similarly performed as described . Total RNA was extracted from mouse cortex with the optional DNase digest step according to the manufacturer's instructions (Qiagen, #74136). cDNA was synthesized using the iScript cDNA Synthesis Kit (Bio-Rad, #170-8891). qRT-PCR was performed using the C1000 Thermal Cycler and quantified using the CFX96 Real-Time System (Bio-Rad). The TaqMan gene expression assay and the iQ supermix (Bio-rad, #170-8862) were used for PCR amplification and real-time detection of PCR products. All RNA samples were analyzed in triplicate and normalized relative to ACTB expression levels. The following probes from Invitrogen were used: Grn (Mm00433848_m1), Nlk (Mm00476435_m1), and mouse ACTB (4352933E).

Primary cortical neuron culture
Primary cortical neurons were prepared from neonatal pups at P0. Cerebral cortices were isolated and freed from meninges. Tissues were first digested with papain (Worthington, LS003126) and DNase I (Sigma, 10104159001) in HBSS (Gibco, 14170-112) for 30 minutes at 37°C, then washed three times with HBSS, and triturated with fire-polished glass pipettes until single cells were obtained. Cell suspensions were then filtered through a 40µm cell strainer and seeded at different densities according to the experimental design. Neurons were first plated in Neurobasal medium (Gibco, 21103-049) supplemented with B27 (Gibco, 17504-044) and 1% FBS (Gibco, 16140-071) and changed to non-serum medium 1 day later. Half media changes were performed every 3 days. Neurons were treated and collected at 7 to 8 days in vitro (DIV-7/8) for endocytosis analysis or collected at DIV-14 for protein and mRNA expression.

Primary microglial cell culture
Primary cultured microglia were prepared from mouse brains, mainly from the cortex and the hippocampus, at P2 or P3. Meninges were removed mechanically, and the cells were dissociated by 0.25% Trypsin (Gibco, 25200-056) in HBSS, then cultured in PDL (Sigma, P6407)-coated T25 flask with DMEM (Gibco, 11965-092) supplemented with 10% heat-inactivated FBS (65°C, 30 minutes) and penicillin/streptomycin. After 14 days , the culture flasks were shaken at 200rpm for 3 hours to collect microglia. Microglia from three flasks having the same genotype were put together and counted as one sample. In total, three independent experiments were repeated. All experiments were performed at DIV-16 to DIV-18.

Generation of CRISPR-mediated Nlk knockdown cells
The CRISPR/Cas9 technology (Cong et al., 2013) was used to generate Nlk knockdown (Nlk-KD) cells. Two guide RNAs were cloned separately into px462_v2 plasmid (Addgene, #62987, a gift from Dr. Feng Zhang). 250ng of each plasmid was co-transfected into the murine microglial cell line BV2 using Nucleofector Kits (Lonza, VPI-1006). At 24 hours post transfection, cells were enriched by the treatment of 4µg/ml puromycin (Thermo Fisher Scientific, #A1113802) for an additional 48 hours. Living cells were dissociated and diluted to 1 cell/100µl in DMEM and cultured in 96-well plate to obtain clonal population. Gene-targeting efficiency for each clone was analyzed by western blot analysis. Due to the incomplete targeting of Nlk genes on the nondiploid BV2 cancer cell line, a single Nlk KD BV2 cell clone was identified and selected for our analyses. The two guide RNA sequences used were, #1; 5'-CCCATCCCCGGCACCGGGTC -3' and #2; 5'-AACAACGGGTCCCAAATTGT -3'.

Cell culture
The murine microglial BV2 cells were grown in DMEM supplemented with 10% heatinactivated FBS and maintained at 37°C and 5% CO2. Cells were transiently transfected, using Amaxa Nucleofector ® (Lonza, Walkersville, MD) according to the manufacturer's instructions, with the pcDNA3.1, Flag-mNlk-WT or Flag-mNlk-T298A constructs. For lysosomal-mediated degradation inhibition experiments, BV2 cells were grown in a 12-well plate to 70 to 80% confluence before treatment. Lysosomal degradation was blocked by Bafilomycin-A1 (LC laboratories, B-1080, 400nM) for 4 hours. To investigate the endocytosis-lysosome-mediated degradation of PGRN in WT or Nlk-KD BV2 cells, cells were treated first with BafA1 or DMSO for 4 hours, the culture medium was removed, and cells were washed twice with PBS before treating with 20nM of recombinant human PGRN (Flag-PGRN recombinant, AdipoGen, AG-40A-0068) for 15 minutes. For primary cortical neurons, the recombinant human PGRN incubation time was increased to 1 hour with a 20nM concentration. For the clathrin-mediated endocytosis study, cells were treated first with Pitstop2 (Sigma, SML1169, 30µM) or DMSO for 1 hour, then the culture medium was removed, and cells were treated with 10nM of recombinant human PGRN for 15 minutes before collecting medium.

Dextran uptake
BV2 microglial cells were plated on 24-well plates with a round glass coverslip (Electron Microscopy Sciences, #72196-12) for 2 days prior to experiment. The cells were placed on ice for 10 minutes, followed by incubation with Alexa 647-conjugated Dextran (Molecular Probes, #D22914) at 1mg/ml in DMEM for 20 minutes at 37°C. After washing twice with PBS, cells were fixed with 4% paraformaldehyde (PFA, Sigma, 158127) for 10 minutes.

Transferrin uptake
BV2 cells, primary microglia, or primary cortical neurons were plated on PDL-coated glass bottom dishes (MatTek, P35G-1.5-14-C) 2 days before experiments. Cells were placed on ice for 10 minutes, followed by incubation with Alexa 568-conjugated Transferrin (Molecular Probes, T23365) at 25µg/ml in Live Cell Imaging Solution (LCIS, Invitrogen, A14291DJ) containing 20mM Glucose and 1% BSA for 15 minutes at 37°C. Cells were washed twice with cold LCIS before live cell image acquisition.

Immunofluorescence staining
Immunofluorescence staining and confocal microscopy analyses were performed using frozen mouse tissues or cultured cells as described previously (Ju et al., 2013;Todd et al., 2015).

Autofluorescence analysis
For autofluorescence analysis on mouse brains, sectioned tissues were washed in PBS. Without primary or secondary antibody incubation, sectioned brain tissues were coverslipped with Vectashield with DAPI and visualized on a Zeiss LSM710 Spectral confocal microscope with multiple excitation wavelengths including 488 nm and 543 nm. Images are z-stack composites encompassing the entire section.

Retinal ganglion cell counts
Retinas from both eyes were dissected and processed as described . Briefly, retinas were post-fixed for 1 to 3 hours in 4% PFA, cryoprotected in 30% sucrose solution overnight at 4 o C, embedded using OCT, and rapidly frozen on dry ice. 15µm sections were sliced on a cryostat and collected immediately on microscope slides. Retinal sections were then stained on slides. Primary antibody (mouse anti-Brn3a, Santa Cruz, 14A6, sc-8429) was applied overnight at 4℃. The secondary antibody used was Alexa Fluor 488-conjugated goat anti-mouse IgG (Invitrogen, A11001). TO-PRO-3 Iodide (642/661) was used to stain nuclei. Fluorescent images were scanned using a Zeiss LSM710 Spectral confocal microscope with a 40X objective and processed with Zen 2.1 software (Carl Zeiss). Within the ganglion cell layer, the number of immunostained Brn3a cells was counted per 100mm for each retinal section from the central regions. Data were averaged from 3 slices of each retina.

Image acquisition and data analyses
For cultured cells, images were captured using Zeiss spinning disk confocal microscopy (SDC) with a 60X oil objective lens. For mouse brain sections, images were captured using a Zeiss LSM710 Spectral confocal microscope with a 20X or 40X objective and processed with Zen 2.1 software (Carl Zeiss), followed by z-stack composites encompassing the entire section. Three sections per animal were analyzed. Quantification of fluorescence intensity was performed by Volocity software (PerkinElmer).

Statistical analysis
Data are presented as mean ± standard errors of measurement (SEM). The statistical significance was assessed using the t-test, one-way ANOVA, or two-way ANOVA on the GraphPad Prism 6.0 software (GraphPad Software). A value of P<0.05 was considered statistically significant.

Data availability
The datasets generated during and/or analyzed in the current study are available in the main and supplementary figures, as well as available from the corresponding author on reasonable request.

Author Contributions
T.D. and J.L. conceived the project. T.D. and H.K. performed experiments. T.D., H.K., T.M.D., L.T. and J.L. analyzed and interpreted data, and wrote the manuscript. All authors reviewed the manuscript and discussed the work.