LRRK2-mediated phosphorylation of HDAC6 regulates HDAC6-cytoplasmic dynein interaction and aggresome formation

Mutations in LRRK2 are the most common cause of dominantly inherited Parkinson’s disease (PD). A proportion of LRRK2 PD exhibits Lewy pathology with accumulations of α-synuclein and ubiquitin in intracellular aggregates that are indistinguishable from idiopathic PD. LRRK2 is a multi-domain protein with both GTPase and kinase activities that has been shown to affect various cellular processes including protein homeostasis, however how PD mutations in LRRK2 may lead to accumulation of ubiquitinated protein aggregates remains unclear. A main cellular pathway to remove aggregated ubiquitinated proteins is aggrephagy: the histone deacetylase HDAC6 recognizes ubiquitinated misfolded proteins and recruits them to the molecular motor cytoplasmic dynein which transports them to the perinuclear region where they are trapped in aggresomes that are subsequently removed by macroautophagy. Here we identified HDAC6 as a novel LRRK2 substrate and show that LRRK2 regulates HDAC6-dependent aggresome formation. LRRK2 directly interacted with the HDAC6 deacetylase domains via its Roc domain and phosphorylated HDAC6 on serine-22. Serine-22 phosphorylation of HDAC6 enhanced its interaction with cytoplasmic dynein and stimulated recruitment of ubiquitinated proteins to aggresomes. Knockdown or knockout of LRRK2 impaired HDAC6-mediated aggresome formation. PD mutant LRRK2 G2019S showed reduced interaction with HDAC6 and did not support aggresome formation to the same extend as wild type LRRK2. This was recapitulated in LRRK2 G2019S patient-derived iAstrocytes that showed an aggresome formation defect. In conclusion our data reveal HDAC6 as a target of LRRK2 and suggest that deregulation of HDAC6-mediated aggresome formation and aggrephagy could contribute to the pathology of PD.

homeostasis, however how PD mutations in LRRK2 may lead to accumulation of ubiquitinated protein aggregates remains unclear.
A main cellular pathway to remove aggregated ubiquitinated proteins is aggrephagy: the histone deacetylase HDAC6 recognizes ubiquitinated misfolded proteins and recruits them to the molecular motor cytoplasmic dynein which transports them to the perinuclear region where they are trapped in aggresomes that are subsequently removed by macroautophagy.
Here we identified HDAC6 as a novel LRRK2 substrate and show that LRRK2 regulates HDAC6-dependent aggresome formation. LRRK2 directly interacted with the HDAC6 deacetylase domains via its Roc domain and phosphorylated HDAC6 on serine-22. Serine-22 phosphorylation of HDAC6 enhanced its interaction with cytoplasmic dynein and stimulated recruitment of ubiquitinated proteins to aggresomes. Knockdown or knockout of LRRK2 impaired HDAC6-mediated aggresome formation. PD mutant LRRK2 G2019S showed reduced interaction with HDAC6 and did not support aggresome formation to the same extend as wild type LRRK2. This was recapitulated in LRRK2 G2019S patient-derived iAstrocytes that showed an aggresome formation defect.
In conclusion our data reveal HDAC6 as a target of LRRK2 and suggest that deregulation of HDAC6-mediated aggresome formation and aggrephagy could contribute to the pathology of PD.
Aggresomes are pericentriolar inclusion bodies where misfolded proteins accumulate prior to removal by macroautophagy. When the ability of cells to refold proteins or to remove misfolded proteins via the ubiquitin-proteasome system (UPS) is exceeded, ubiquitinated misfolded proteins are transported by cytoplasmic dynein to aggresomes. Histone deacetylase 6 (HDAC6) plays a key role in aggresome formation. HDAC6 is a member of a family of HDACs containing 11 Zn 2+ -dependent enzymes (HDAC1-11) and 7 NAD + -dependent proteins (Sirtuin1-7) that are subdivided into 4 classes, Class I (HDAC1, 2, 3, and 8), Class IIa (HDAC4, 5, and 7) and IIb (HDAC6 and 10), Class III (Sirtuin1-7), and Class IV (HDAC11). HDAC6 stands out among the other HDACs because it is predominantly cytosolic, contains two catalytic domains and has a ubiquitin binding domain. Cytosolic substrates of HDAC6 include α-tubulin, Hsp90, tau, cortactin, and peroxiredoxin. In respect to aggrephagy, HDAC6 interacts with cytoplasmic dynein and recruits polyubiquitinated misfolded proteins to dynein motors for transport to aggresomes through its ubiquitin binding domain 1 .
Lewy bodies are a hallmark pathology of Parkinson's disease and it has been proposed that impaired handling of misfolded proteins and aggregates may contribute to their formation [2][3][4] . HDAC6 is present in Lewy bodies, indicating aggrephagy may be involved in their formation 1,5 . Mutations in the Roc and COR domain diminish the GTPase activity of LRRK2, whereas mutations in the kinase domain enhance its kinase activity. It is generally assumed that kinase hyperactivity is linked to neurotoxicity, but it is less clear how diminished LRRK2 GTPase activity contributes to disease 7 .
In most cases LRRK2-associated Parkinson's disease is clinically and pathologically indistinguishable from idiopathic late-onset PD but this may vary depending on the type of pathogenic mutations 8 . Our previous studies have linked LRRK2 to HDAC6 and microtubule acetylation 9 , and LRRK2 has been proposed to be involved in proteostasis and aggrephagy, but reports are conflicting and the molecular mechanisms involved poorly understood [10][11][12] .
Here we investigated the role of LRRK2 in aggrephagy. We report that phosphorylation of HDAC6 by LRRK2 regulates HDAC6-dependent delivery of ubiquitinated proteins to the aggresome and show that this novel function of LRRK2 is impaired by the PDassociated G2019S mutation and in LRRK2 G2019S patient-derived iAstrocytes.

LRRK2 kinase regulates HDAC6-dependent aggresome formation
To characterise the possible role of LRRK2 in aggresome formation, we depleted LRRK2 in HEK293 cells using siRNA and quantified aggresome formation using two well-characterised aggresome reporters, EGFP-CFTR∆F508 13 and GFP-250 14 . The cystic fibrosis causing allele of the cystic fibrosis transmembrane conductance regulator (CFTR), ∆F508, interferes with its ability to fold; misfolded CFTR∆F508 is ubiquitinated and degraded by the proteasome. Inhibition of proteasome activity in cells expressing CFTR∆F508 causes accumulation of stable, ubiquitinated aggregates of CFTR∆F508 in aggresomes 13 . Similar to CFTR∆F508, GFP-250, GFP fused at its COOH terminus to a 250-amino acid fragment of the Golgi protein p115, is sequestered in aggresomes upon inhibition of the proteasome. However, unlike CFTR∆F508, GFP-250 in aggresomes is not ubiquitinated 14 .
In line with previous publications 13,14 , in control HEK293 cells treated with nontargeting control (NTC) siRNA, both EGFP-CFTR∆F508 and GFP-250 were recruited to distinctive perinuclear inclusion bodies after treatment with MG132 to inhibit the proteasome. Co-staining with vimentin confirmed that these structures were aggresomes. SiRNA-mediated LRRK2 depletion had no effect on the formation of GFP-250 aggresomes but fully prevented EGFP-CFTR∆F508 aggresomes ( Fig. 1A-B; SFig. 1). To verify the specificity of the siRNA treatment we re-expressed wild type LRRK2 and as expected this completely rescued CFTR∆F508 aggresome formation in LRRK2 siRNA treated cells (Fig. 1A). To further substantiate these findings, we turned to LRRK2 knockout (KO) MEFs 15 , reasoning that if LRRK2 is indeed required for aggresome formation that these cells would be defective in aggresome formation.
As expected, LRRK2 KO MEFS did not form EGFP-CFTR∆F508 aggresomes, and this could be rescued by expression of wild type LRRK2 (Fig. 1C). Interestingly, in LRRK2 KO MEFs we did observe perinuclear vimentin-positive spots that were reminiscent of the vimentin cages associated with aggresomes despite the absence of perinuclear EGFP-CFTR∆F508 accumulation (Fig. 1C), suggesting that loss of LRRK2 may impair recruitment of EGFP-CFTR∆F508 to aggresomes rather than the assembly of vimentin-positive aggresomes per se. This is consistent with the observation that LRRK2 was not required for formation of ubiquitin-independent GFP-250 positive aggresomes (Fig. 1B).
Finally, to confirm that these observations were not just a consequence of overexpression of EGFP-CFTR∆F508 we inhibited the proteasome in NTC or LRRK2 siRNA treated HeLa cells and observed the formation of endogenous aggresomes using anti-ubiquitin antibodies. In NTC siRNA-treated HeLa accumulations of ubiquitin in the perinuclear region were readily observed after proteasome inhibition and these accumulations stained positive for vimentin, identifying them as aggresomes (Fig. 1D).
In contrast, after proteasome inhibition in LRRK2 siRNA-treated HeLa, ubiquitinpositive spots were observed throughout the cytoplasm and did not accumulate in the perinuclear region, consistent with a failure to recruit ubiquitinated proteins into aggresomes (Fig. 1D).
Ubiquitinated, misfolded proteins such as CFTR∆F508 are recruited to the aggresome by HDAC6 whereas the recruitment of non-ubiquitinated proteins such as GFP-250 is HDAC6 independent 1 . HDAC6-dependent aggresome formation requires both HDAC6 deacetylase and ubiquitin binding activity 1 . The data above suggested a role for LRRK2 in HDAC6-dependent recruitment of ubiquitinated proteins to aggresomes.
To begin to investigate this, we inhibited HDAC6 using the highly selective inhibitor to support aggresome formation (Fig 1A, C).

HDAC6 is a LRRK2 substrate
The data above strongly suggested that LRRK2 regulates HDAC6's function in aggresome formation. To further investigate this, we checked if LRRK2 and HDAC6 may interact and if HDAC6 is a substrate of LRRK2.
We first investigated the possible interaction of LRRK2 and HDAC6 in coimmunoprecipitation assays in HEK293 cells. Untagged HDAC6 coimmunoprecipitated with myc-tagged LRRK2 (myc-LRRK2) from cells co-transfected with HDAC6 and myc-LRRK2, but not from HDAC6 or myc-LRRK2 only transfected cells ( Fig. 2A).
To determine if HDAC6 and LRRK2 could directly interact we performed GST pulldown assays in which we incubated a GST-tagged LRRK2 fragment (aa 970-2527) with recombinant His-tagged HDAC6. HDAC6 was readily pulled down with GST-LRRK2 but not with the GST control (Fig. 2B). Because both HDAC6 and LRRK2 have been shown to interact with tubulin, we verified the absence of tubulin in the reaction (SFig.

2).
We next determined the domains of LRRK2 and HDAC6 involved in their interaction in co-immunoprecipitation assays from HEK293 cells ( Fig. 2C and D). We found that HDAC6 efficiently interacted with the Roc domain in these assays, and to a much lesser extend with the COR domain; the interaction did not require the ARM/Ankyrin repeat, LRR, or WD40 domains. The kinase domain per se did not interact with HDAC6 ( Fig. 2C). Interestingly, while the Roc-COR-Kinase fragment interacted to a similar extend as the Roc domain, interaction of HDAC6 with Roc-COR fragment was markedly reduced and more similar to the COR domain only fragment. Thus, the COR domain appears to impair efficient binding of HDAC6 to the Roc domain, and this can be overcome by the presence of the kinase domain.
HDAC6 interacted with the LRRK2 Roc-COR-Kinase fragment via the histone deacetylase domains and both domains bound LRRK2 to a similar extend. The ubiquitin-binding domain (ZnF-UBP) of HDAC6 was not required to bind the LRRK2 Roc-COR-Kinase fragment, confirming that the interaction was not due to HDAC6 binding to ubiquitinated LRRK2 (Fig. 2D).
We next performed in vitro phosphorylation assays combined with mass spectrometry to identify possible LRRK2 phospho-sites in HDAC6. LRRK2 was found to phosphorylate HDAC6 serine-22 (Fig. 3A). To verify this phosphorylation in cells we co-expressed HDAC6 and wild type LRRK2 or kinase dead LRRK2 D1994A 17 in HEK293 cells and determined HDAC6 phospho-serine-22 (pSer-22) levels using phospho-specific antibodies (SFig. 3). Consistent with LRRK2-mediated phosphorylation of expressed HDAC6, pSer-22 increased upon overexpression of wild type but not kinase dead LRRK2 (Fig. 3B). Similarly, we found that overexpression of wild type LRRK2 increased phosphorylation of endogenous HDAC6 (Fig. 3C). Thus, HDAC6 interacts with and is a substrate of LRRK2.
The above data suggested that LRRK2 mediated phosphorylation of HDAC6 may regulate aggresome formation.
We first checked if LRRK2 kinase activity was required for HDAC6 aggresome formation by rescue of LRRK2 siRNA-treated HEK293 cells with wild type LRRK2 or kinase dead LRRK2 D1994A. While as above wild type LRRK2 efficiently rescued LRRK2 siRNA treatment, kinase dead LRRK2 D1994A was unable to do so (Fig. 4A).
Since HDAC6 was a substrate of LRRK2 (Fig. 3), we reasoned that HDAC6 was likely to be downstream of LRRK2. To test this, we overexpressed HDAC6 on a LRRK2 deficient background and analysed EGFP-CFTR∆F508 aggresome formation. As expected if HDAC6 was downstream of LRRK2, increasing HDAC6 expression partially rescued loss of LRRK2. HDAC6 lacking its ZnF-UBP domain and thus unable to bind ubiquitinated cargo, did not rescue loss of LRRK2 confirming the specificity of the rescue (Fig. 4C). The partial restoration of aggresome formation by HDAC6 expression indicated the possibility that a LRRK2-dependent step was required for full activity. Therefore, using the same experimental paradigm, we next tested the role of HDAC6 pSer-22 by expressing phospho-deficient S22A or phospho-mimicking S22E forms of HDAC6. HDAC6 S22A did not rescue loss of LRRK2, whereas HDAC6 S22E completely restored EGFP-CFTR∆F508 aggresome formation (Fig. 4C). The latter is consistent with a model in which HDAC6 requires LRRK2 phosphorylation on serine-22 for full activity and aggresome formation.

LRRK2 mediated HDAC6 serine-22 phosphorylation regulates HDAC6 interaction with cytoplasmic dynein
Phosphorylation of HDAC6 on serine-22 has previously been associated with increased deacetylase activity 18 . Since HDAC6 deacetylase activity is required for aggresome formation 1 we compared the deacetylase activity of HDAC6 S22A to that of wild type HDAC6 by quantifying acetylated tubulin on immunoblot and by immunofluorescence. Expression of both HDAC6 S22A and wild type HDAC6 markedly decreased the levels of acetylated tubulin. The effect of HDAC6 S22A was equivalent to wild type HDAC6 in these assays, making it unlikely that its effect on aggresome formation is due to loss of deacetylase activity ( Fig. 5A and B).
The binding of HDAC6 to cytoplasmic dynein is crucial for the delivery of ubiquitinated cargo to the aggresome 1,19 . To test if HDAC6 pSer-22 affects its ability to interact with cytoplasmic dynein, we next inhibited the proteasome in HEK293 cells to induce aggresome formation and evaluated the interaction of wild type HDAC6 and HDAC6 S22A with endogenous cytoplasmic dynein. Induction of aggresome formation increased the interaction of wild type HDAC6 with cytoplasmic dynein as was previously reported (Fig. 5C) 1 . In contrast, the interaction of HDAC6 S22A with cytoplasmic dynein did not increase after induction of aggresome formation (Fig. 5C).
Thus, HDAC6 pSer-22 appears to be instrumental in recruitment of HDAC6 to cytoplasmic dynein after induction of the aggresome pathway. To confirm if the increase in HDAC6 serine-22 phosphorylation and resulting recruitment to cytoplasmic dynein depended on LRRK2 kinase activity we knocked down LRRK2. In absence of LRRK2, wild type HDAC6 was no longer recruited to cytoplasmic dynein after induction of aggresome formation (Fig. 5C).

Aggresome formation is impaired in LRRK2 PD
LRRK2 G2019S is the most common genetic cause of Parkinson's disease 6 . To test if the G2019S mutation affects LRRK2 function in aggresome formation we reconstituted LRRK2 deficient HEK293 cells with LRRK2 G2019S and monitored CFTR∆F508 aggresome formation after proteasome inhibition. While wild type LRRK2 fully restored aggresome formation, LRRK2 G2019S only partially did, indicating that the G2019S mutation reduces LRRK2's ability to mediate aggresome formation in response to accumulation of ubiquitinated proteins (Fig. 6A). As also noted above in LRRK2 deficient cells (Fig. 1), LRRK2 G2019S did not prevent the formation of vimentin cages per se, but rather appeared to affect the recruitment of CFTR∆F508 to aggresomes.
To further investigate how the G2019S mutation may affect LRRK2-mediated aggresome formation we tested if the G2019S mutation affected the interaction of LRRK2 with HDAC6, HDAC6 serine-22 phosphorylation, or HDAC6 deacetylase activity. We found that LRRK2 G2019S interacted significantly less with HDAC6 compared to wild type LRRK2 in co-immunoprecipitation assays (Fig. 6B). In contrast, the kinase dead LRRK2 D1994A mutant bound stronger to HDAC6 in these assays ( Fig. 6B). Thus, LRRK2 kinase activity inversely correlated with HDAC6 binding. The G2019S mutation did not significantly increase HDAC6 phosphorylation compared to wild type LRRK2, even though numerous groups have shown that the G2019S mutation markedly increases LRRK2 kinase activity (Fig. 6C). Possibly the diminished interaction of LRRK2 G2019S reduces phosphorylation efficiency. Finally, LRRK2 G2019S and wild type LRRK2 stimulated HDAC6 deacetylase activity to the same extend ( Fig. 6D). Thus, we propose that decreased interaction of LRRK2 G2019S with HDAC6 reduces the efficiency of serine-22 phosphorylation, and this impairs LRRK2mediated aggresome formation.

Aggresome formation is impaired in LRRK2 G2019S patient-derived iAstrocytes
To test if endogenous LRRK2 G2019S impairs aggresome formation in a diseaserelevant model we turned to LRRK2 G2019S patient-derived iAstrocytes 20 . We treated two matched non-disease control and two patient-derived iAstrocyte lines with proteasome inhibitor and monitored endogenous aggresome formation by immunofluorescence microscopy of ubiquitin and vimentin. Both controls consistently formed ubiquitin and vimentin positive aggresomes that also contained HDAC6 (Fig.   7). In contrast, in the two patient-derived iAstrocyte lines, aggresome formation was significantly impaired (Fig. 7). In the patient cells, clusters of ubiquitin that colocalised with HDAC6 were observed but no characteristic accumulation in a perinuclear aggresome occurred (Fig. 7). Thus, in agreement with the data above, while HDAC6 was recruited to ubiquitinated proteins, subsequent recruitment of cytoplasmic dynein appears to be impaired in a model of LRRK2 G2019S PD.

Discussion
Accumulation of protein aggregates is a hallmark of many neurodegenerative diseases including PD. Protein homeostasis is maintained by the proteostasis network, a complex regulatory network that controls protein biosynthesis, folding, trafficking, and clearance. Failure of the proteostasis network to deal with misfolded and potentially toxic proteins ultimately is deleterious to cells; this appears to be especially the case for neurons, as exemplified by the common occurrence of aberrant protein folding and aggregate deposition in neurons in neurodegenerative disease 21 .
Aggregated proteins are mostly removed by autophagy, a process that is also called aggrephagy 22 . In aggrephagy, ubiquitinated aggregated proteins are recruited to HDAC6 which by binding to cytoplasmic dynein enables the transport of the ubiquitinated protein aggregates along microtubules to the aggresome. The aggresome is insoluble and metabolically stable, and is enclosed by intermediate filaments. The contents of aggresomes may subsequently be degraded by aggrephagy. Thus, the aggresome likely represents a means to sequester aggregated proteins from the cytosol so as to prevent toxicity while awaiting clearance 23 .
In case of PD, the classic histopathological feature is the Lewy body that comprises mainly ubiquitinated α-synuclein filaments but also numerous other proteins. Lewy bodies share many morphological and biochemical similarities with aggresomes and may form via an aggresome-like mechanism 2-4 . Several groups have linked HDAC6 to Lewy body formation and have shown that HDAC6 is a constituent of Lewy bodies 1,5,24 . A role for dynactin was also described in Lewy body formation 25 .
Since our previous studies had linked LRRK2 to HDAC6 9 , and LRRK2 itself is present in Lewy bodies 26 we hypothesised that LRRK2 may be involved in aggresome formation via HDAC6. Confirming our hypothesis, we found that LRRK2 is required for sequestration of ubiquitinated but not of non-ubiquitinated protein aggregates to aggresomes (Fig. 1). The former is dependent on HDAC6 1 while the latter relies on Bag3 to connect the protein aggregates to cytoplasmic dynein 27 . We found that HDAC6 binds to LRRK2 and that this interaction is mediated by the HDAC6 deacetylase domains and the LRRK2 Roc domain (Fig. 2). LRRK2 kinase activity appears to regulate HDAC6/LRRK2 interaction ( Fig. 2 and 6B). Autophosphorylation of the Roc domain is well established and may regulate GTPase activity and in turn kinase activity [28][29][30] . Thus, our data suggests autophosphorylation of the Roc domain may regulate its interaction with HDAC6. Phosphorylation may directly affect binding, or, alternatively, HDAC6/LRRK2 interaction may also be regulated by phosphorylationinduced changes in LRRK2 GTPase activity.
Phosphorylation of serine-22 has been reported in a number of large-scale proteomic studies, but its role is not yet clear. It has been suggested that GSK3β-dependent HDAC6 serine-22 phosphorylation increases its deacetylase activity in hippocampal neurons 18 . However, using a HDAC6 S22A mutant that cannot be phosphorylated, we did not observe an appreciable effect of serine-22 phosphorylation on α-tubulin acetylation levels (Fig. 5A). On the other hand, the same mutant completely failed to rescue aggresome formation (Fig. 4C). Thus, we believe that HDAC6 serine-22 phosphorylation is primarily involved in aggresome formation. Possibly the effect of Our data show that LRRK2 is required for aggresome formation (Fig. 1) and that the most common PD-associated LRRK2 mutant, G2019S, did not support aggresome formation to the same extend as wild type LRRK2 (Fig. 6A). This finding seemed counterintuitive because the G2019S mutant has increased kinase activity, and our data show that phosphorylation of HDAC6 by LRRK2 drives aggresome formation (Fig.   5). However, compared to wild type LRRK2 expression of LRRK2 G2019S did not additionally increase HDAC6 phosphorylation (Fig. 6C) while its interaction with HDAC6 was significantly reduced (Fig. 6B). Thus, we suggest that at physiological expression levels, the G2019S mutation behaves as a loss-of-function mutant that reduces LRRK2-mediated phosphorylation of HDAC6 and causes impairment of aggresome formation. In agreement with this, we found a marked defect in aggresome formation in LRRK2 G2019S patient-derived iAstrocytes (Fig. 7).
Inhibition of HDAC6 has been shown to increase α-synuclein levels 24

Immunofluorescence
Immunostaining was performed as described previously 39

Microscopy
Images were recorded using appropriate filtersets (Omega Optical and Chroma

Image analysis
All image analysis was performed using ImageJ 41  Where possible operators where blinded to the identity of the samples analysed.
SDS-PAGE and immunoblotting were performed as described previously (       antibodies. Immune pellets were probed for total HDAC6 and HDAC6 pSer-22.

Data availability.
All data files and files produced for statistical analysis are available on request.       (0) (0) ns **** C t r l 2 9 C t r l 2 9 P a t 6 8 P a t 6 8 Ctrl Mg132