Histone Deacetylase 6 Represents a Novel Drug Target in the Oncogenic Hedgehog Signaling Pathway

Medulloblastoma is a pediatric cancer affecting the cerebellum and representing the most frequent malignant brain tumor in children (1). Four molecular subtypes have been classified and one of these subgroups (which accounts for approximately one third of cases) is driven by excessive Hedgehog (Hh) signaling (2). In the developing cerebellum, Purkinje cell–derived Sonic Hh (SHH; one of three mammalian Hh ligands) serves as a potent mitogen for granule cell precursors (GCP), which transiently amplify and later constitute the internal granule cell layer (IGL) of the cerebellum. Persistent Hh signaling expands the number of GCPs in an uncontrolled manner, resulting in the formation of malignant medulloblastoma, a lethal condition if left untreated (3). Current treatment schemes using radiation, chemotherapy, and surgery lead to appreciable survival rates but are often associated with life-long impairments due to unspecific damage of the developing brain (1).

Mechanistically, the Hh ligands bind to Patched (PTCH1 or PTCH2) receptors situated in the membrane of a specialized organelle, the primary cilium, of receiving cells (4, 5). The immotile primary cilium represents a sensory structure present on most cell types and is held in place by a microtubule-based axoneme (6). Motor protein transport by intraflagellar transport (IFT) complexes along the microtubule tracks regulates the assembly, function, and the disassembly of this organelle (7). The tubulin subunits of the ciliary axoneme contain posttranslational modifications associated with long-lived microtubules, such as the acetylation of α-tubulin. An important modulator of tubulin acetylation is histone deacetylase 6 (HDAC6), a mainly cytoplasmic member of the HDAC family that uses tubulin as one of its major substrates (8, 9).

Binding of Hh ligands to PTCH1 allows for the entry of Smoothened (SMO) proteins into the cilium and the subsequent dissociation of SUFU (Suppressor of Fused; a cardinal negative regulator of Hh signaling) from GLI transcription factors (10). Specifically, free GLI2 and GLI3 are activated in the cilium and can subsequently enter the nucleus to induce target gene transcription (11, 12). The transcriptional targets include Hh pathway–specific genes, such as PTCH1 and GLI1, as well as more cell type–specific genes such as N-Myc, BMI1, or D-type Cyclins.

Recently, potent pharmacologic inhibitors targeting SMO, the key upstream bottleneck in the Hh cascade, have been introduced to the clinic (13). Although very promising first results have underscored the principle effectiveness of targeting Hh signaling in medulloblastoma, the rapid development of drug resistance has dampened the initial enthusiasm (14–16). These data encourage for the development of additional inhibitors that could be applied together with SMO inhibitors or as single-agent therapy in refractory situations. Because the major mechanisms of the development of drug resistance seems to involve SMO mutations affecting the drug binding site or the activation of SMO-bypassing downstream events (17), novel Hh pathway inhibitors would ideally target a non-SMO protein that is acting at a later step in the signaling cascade (18).

We found HDAC6 to be overexpressed in a murine model of Hh-driven medulloblastoma and were interested to see if this enzyme plays a role in the Hh pathway itself and in medulloblastoma. Our results show that HDAC6 significantly regulates Hh signaling, playing a dichotomous role: on the one hand, its enzymatic activity represses basal Hh target gene transcription; while, on the other hand, HDAC6 activity promotes the maximum expression of target genes. Using an allograft model of Hh-elicited medulloblastoma, we could show that despite this dichotomous effect of HDAC6 on the Hh pathway, specific inhibitors of this enzyme significantly reduced in vivo tumor growth. In light of the fact that the genetic knockout of Hdac6 does not show a prominent phenotype (19), selective HDAC6 antagonists represent an interesting new class of potential drugs in the treatment of medulloblastoma.

SMO agonist SAG was purchased from Calbiochem. Trichostatin A (TSA) and SANT (SANT-1) were from Sigma. Tubacin was kindly provided by Ralph Mazitschek and Stuart Schreiber [Initiative for Chemical Genetics, National Cancer Institute (NCI), Bethesda, MD, USA] and was also purchased from Enzo Life Sciences. CAY-10603 was purchased from Biomol. ACY-1215 for cell culture experiments was purchased from Selleck Bio and larger amounts for animal studies were bought from Xcess Bio. Lists of antibodies, siRNA sequences, and quantitative PCR (qPCR) primers used in this study can be found in Supplementary Tables S1–S3, respectively.

NIH3T3, Hek293A, ShhL2, and C3H10T1/2 cell lines were purchased from the ATCC in 2009 and were expanded and frozen in aliquots as passage 3 to 5. Cells were used for experiments until passage 25. Cell line authentication was performed by the ATCC. All cell lines were cultured in Dulbecco's modified Eagle medium (DMEM; high glucose plus glutamine and pyruvate; Invitrogen) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin at 37°C with 5% CO₂. Immortalized wild-type (wt) and Sufu^−/− mouse embryonic fibroblasts (MEF; 20) were a kind gift of Rune Toftgard (Karolinska Institute, Stockholm, Sweden). MEF^[SHH], MEF^[Smo*], MEF^[Gli2dN], and Gli23^−/− MEFs cells were kindly provided by Wade Bushman (Molecular and Environmental Toxicology Center, University of Wisconsin Medical School, Madison, WI; ref. 21). MEF lines (obtained in 2009 and frozen as aliquots) are regularly authenticated by us through functional testing (e.g., loss of Sufu in the case of Sufu^−/− MEFs).

Separation of lysates by SDS–PAGE was followed by subsequent Western blot analysis. SDS–PAGE gels were blotted on Immobilon-PVDF membranes (Millipore) and incubated with the respective primary antibody, followed by a horseradish peroxidase (HRP)–coupled secondary antibody. Detection of the HRP signal was performed using Pierce ECL Western Blotting Substrate (Thermo Scientific) according to the manufacturer's protocol.

Immunohistochemistry on formaldehyde-fixed and paraffin-embedded tissue sections was performed using an HDAC6 antibody (1:50) from Santa Cruz Biotechnology (sc-11420).

Total RNA was extracted using NucleoSpin RNA II kit (Macherey-Nagel) according to the manufacturer's protocol. cDNA synthesis of 1 μg total RNA was performed using the iScript cDNA Synthesis Kit (Bio-Rad) following the manufacturer's guidelines. qPCR reactions were performed using the Absolute QPCR SYBR Green Mix (ABGene). qPCR reactions were performed on 96-well qPCR plates (ABGene) using either the Mx3000P or Mx3005P qPCR systems (Agilent). Results were calculated as relative mRNA expression (2^ΔΔC_t). Data were obtained from at least three independent experiments and are shown as the mean ± SD.

ShhL2 cells were plated in triplicate and were grown to full confluence in solid white 96-well plates with clear bottom. Subsequently, cells were treated in full growth medium with 100 nmol/L SAG plus the indicated compounds for 48 hours. Cells were lysed in Passive Lysis Buffer (Promega) and Firefly and Renilla luciferase activity were measured using an Orion L microplate luminometer (Berthold Detection Systems) using Beetle- and Renilla-Juice reagents (both PJK, Kleinblittersdorf, Germany).

The mesenchymal progenitor cell line C3H10T1/2 was plated in triplicates and grown to full confluence in 96-well plates. Subsequently, cells were exposed to experimental compounds in full growth medium for 4 days. Afterward, cells were lysed in Passive Lysis Buffer (Promega). A part of the lysate was used for protein quantification (Bio-Rad Protein Assay; Bio-Rad) while the remaining lysate was used for measuring alkaline phosphatase (AP) activity (Alkaline Phosphatase Blue Microwell Substrate; Sigma).

The Neurod2-SmoA1 mouse line was purchased from JAX (stock number 008831) and is described previously (22).

Primary cells from SmoA1 medulloblastoma were obtained by mincing and trypsinization of tumor fragments. Subsequently, these cerebellum-derived cells were mixed 1:1 with Matrigel (Sigma E1270) and injected subcutaneously (s.c.) into C57BL/6J mice for amplification of the cells. After a few rounds of in vivo passaging, cells were pooled, mixed with Matrigel (one third of the volume), and injected s.c. into 15 C57BL/6J mice (3.5 months old/20 g weight). One week later, all mice had palpable s.c. tumors of approximately 25 mm². Mice were randomly divided into three groups (5 animals each, day 0) and were injected s.c. on days 0, 1, 2, 3, 4, 7, 8, 9, 10, and 11 each with 200 μL of compound [50 mg/kg; Solvent, vismodegib (GDC-0449; LC Laboratories), and ACY-1215 (Rocilinostat; Xcess Bio (ACY)]. The injections were peritumoral with an average distance to the tumor mass of approximately 0.5 to 1 cm. Mice were sacrificed on day 12 by cervical dislocation and tumors were removed for further analysis.

The compounds were first dissolved in 100% DMSO at a concentration of 70 mg/mL. Then, compounds were diluted in 45% (2-hydroxypropyl)-β-cyclodextrin (Sigma; #332607) in PBS for a final concentration of 5 mg/mL (1 mg/200 μL). All animal studies were approved and were in agreement with institutional and federal state laws.

Subconfluent MEF^[SHH] cells were transfected with 35 nmol/L siRNA (Dharmacon SMARTpools) on day 0 using Lipofectamine RNAiMAX (Invitrogen) as transfection reagent. On day 1 (24 hours after transfection), the cells were confluent and the siRNA solution was exchanged against full growth medium. Cells were grown for another 48 hours and were harvested for RNA preparation on day 3 (72 hours after transfection). The integrity of the isolated RNA was subsequently determined using the Experion Automated Electrophoresis System (Bio-Rad). The RNAs from two independent experiments were pooled in a 1:1 ratio. Labeling, hybridization to microarrays, and data analysis were performed by the IMT Genomics Core Facility. RNA was labeled with the two-color Quick-Amp Labeling kit (Agilent) and hybridized against Agilent-026655 microarrays. Raw microarray data were normalized using the “loess” method implemented within the marray package of R/Bioconductor. Agilent probes were reassigned to the Ensembl revision 70 genome annotations by aligning sequences with a short read aligner (Bowtie against both, the transcriptome and the genome) as described previously (23). The microarray data have been deposited at Array Express under the accession number E-MTAB-2440.

Data are shown as the mean of three independent experiments (unless otherwise indicated) ± SD. Statistical significance was calculated by applying a two-tailed Student t test (Microsoft Excel). *, P < 0.05; **, P < 0.005; ***, P< 0.0005. If not stated otherwise, the comparison is between the experimental condition and the corresponding control (DMSO, siCon).

HDAC6 had been implicated in the etiology of several cancers and in primary cilium biology (9, 24–27), which prompted us to analyze the role of HDAC6 in Hh-driven medulloblastoma. In line with an earlier report (28), we found endogenous Hdac6 to be overexpressed in cerebellar tumors arising in a mouse model of Hh-dependent medulloblastoma (SmoA1 model; Fig. 1A; refs. 22, 29). In wt animals, HDAC6 protein expression can mostly be seen in the non-tumor Purkinje cells. In contrast, in the transgenic mice, granule cell–derived tumor cell nests show high HDAC6 immunoreactivity (Fig. 1B). These findings raised the question about the functional importance for HDAC6 in Hh-driven medulloblastoma.

To address the relevance of HDAC6 in Hh-evoked disease, we were first interested to see whether this enzyme plays a role in the Hh pathway itself. To this end, we used three structurally distinct, specific HDAC6 antagonists (Tubacin, CAY-10603, and ACY-1215), which had been described and characterized before (Fig. 2A; refs. 26, 30, 31). We treated fibroblasts with these inhibitors because they represent a good model system for the analysis of the Hh signaling cascade and are one of the major Hh-responsive cell type known. As shown in Fig. 2B, analysis of NIH3T3 fibroblasts that stably carry a Hh-responsive luciferase construct in their genome (ShhL2 cells), demonstrated that the Hh pathway induction by a synthetic agonist (SAG; (32) could be blunted by coapplication of increasing amounts of HDAC6 inhibitor. The negative effect of HDAC6 blockade on endogenous Hh signaling could be observed with all three compounds, arguing against unspecific off-target effects. To rule out negative effects of the compounds on ciliogenesis, we analyzed the presence of primary cilia in treated cells. In line with HDAC6 stabilizing rather than destabilizing cilia (9, 27), we found no significant reduction of ciliogenesis or change in cilia morphology in cells exposed to HDAC6 inhibitors, irrespective of the serum concentrations used (Supplementary Fig. S1A–S1C). Moreover, HDAC6 inhibition was not associated with gross cellular toxicity (Supplementary Fig. S2A).

To extend our investigations, we tested the impact of HDAC6 inhibition on a physiologic Hh-driven process. Mesenchymal progenitor cells (C3H10T1/2 cells) differentiate along the osteogenic lineage upon treatment with Hh, a process that can be visualized by staining for AP activity, an osteogenic marker protein. As shown in Fig. 2C and D, the SAG-induced differentiation of C3H10T1/2 cells could be potently antagonized by application of the HDAC6 inhibitors CAY-10603 or Tubacin. Interestingly, the pan-HDAC inhibitor TSA also blocked SMO-induced cell differentiation, but in addition also resulted in an increase of basal induction (Fig. 2D). This suggests that nuclear HDAC family members repress basal Hh signaling and further underlines the specificity of the HDAC6 inhibitors used in this study. However, to demonstrate the requirement for HDAC6 in Hh signaling with a chemical compound-independent approach, we used a pool of four different RNAi constructs to selectively knockdown endogenous Hdac6 mRNA in MEFs stably transfected with SHH (MEF^[SHH]; ref. 21). Transfection of these constitutively signaling cells with siHdac6 resulted in a clear reduction of expression of the established Hh target genes Gli1, Ptch1, and Ptch2; indicative of Hh pathway inhibition (Fig. 2E and F). To rule out unspecific effects by members of the siRNA pool, we tested the individual siRNAs contained in the pool and could show that three of the four RNAi sequences repressed Hh signaling, supporting the previous findings (Supplementary Fig. S2B). Furthermore, the Hdac6-specific siRNA pool did not reduce the expression levels of Hdac family members other than Hdac6 (Supplementary Fig. S2C), demonstrating that the observed effects are in fact due to selective interference with Hdac6. Moreover, RNAi against Hdac6 also significantly reduced SAG-induced C3H10T1/2 cell osteogenic differentiation (Supplementary Fig. S2D–S2F). Finally, HDAC6 action was important for Hh signaling irrespective of the serum concentration (10% FBS; see Fig. 2; 1% FBS; see Supplementary Fig. S3A) and of the nature of the ligand (see recombinant SHH instead of SAG in Supplementary Fig. S3B). Taken together, our data show that HDAC6 functions in the Hh pathway and is required to achieve full pathway activity.

Next, we were interested to learn at which point in the Hh transduction cascade the HDAC6 functions would be integrated. To this end, we made use of MEF cell lines that harbor activating alterations to stimulate Hh signaling at different levels of the pathway and measured the abundance of the Hh target genes Gli1, Ptch1, Ptch2, and Hip1. As shown in Fig. 3A and B, signaling in MEF^[SHH] cells (which activate Hh signaling at the ligand step) could be blocked by application of different HDAC6 antagonists as demonstrated on the mRNA (Fig. 3A) and the protein level (Fig. 3B). Again, none of the used HDAC6 inhibitors significantly increased the levels of acetylated histone H3 (Fig. 3B and Supplementary Fig. S2G), in contrast to the pan-HDAC inhibitor TSA, arguing for HDAC6-selective mechanisms.

Interfering with endogenous HDAC6 function also blocked Hh signaling in Ptch1^−/− MEFs (Supplementary Fig. S3C) and in cells stably expressing a dominant active version of Smo (MEF^[Smo*]; Fig. 3C; ref. 21). Despite the fact that inhibition could be observed in all cell lines, it appeared as if cells harboring wt Smo (i.e., MEF^[SHH] and Ptch1^−/− MEFs) were more potently inhibited than cells with mutant Smo (MEF^[Smo*]).

Furthermore, we observed that in Ptch1^−/− cells, expression of the Hh target gene Hip1 was strongly repressed only if a pan-HDAC inhibitor was used (TSA; Supplementary Fig. S3C). In contrast, SANT and the HDAC6 inhibitors only reduced Hip1 levels very modestly, indirectly pointing to HDAC6 antagonists having no potent effect on histone acetylation, as was shown before.

To investigate whether HDAC6 also plays a role downstream of Ptch1 and Smo, we used MEF cells harboring a genetic deletion of the Sufu gene, leading to strong, but ligand- and receptor-independent constitutive signaling (20, 33). Although the SMO-selective inhibitor SANT (32) was ineffective in these cells, HDAC6 inhibition by ACY-1215, CAY-10603, or Tubacin resulted in a suppression of Hh target gene expression, albeit not to the same extent as in the MEF^[SHH] cells (Fig. 3D). Additional evidence for pathway inhibition in Sufu^−/− cells was also found upon RNAi-mediated knockdown of Hdac6 (Supplementary Fig. S3D).

Finally, we used cells with stable expression of low levels of an activated mutant of GLI2, the transcription factor mediating the final steps of the Hh signaling cascade (MEF^[Gli2dN] cells; ref. 21). As shown in Fig. 3E, HDAC6 blockade also repressed Hh target gene expression in these cells, whereas the upstream inhibitor SANT was inactive. Again, the inhibition achieved by HDAC6 blockade was not as pronounced as in cells with Hh pathway activation at the ligand step (e.g., MEF^[SHH]). Because direct deacetylation by nuclear HDACs had been shown to stimulate GLI function (34), we were interested if a similar mechanism applied to HDAC6 as well. Whereas HDAC1 and HDAC2 significantly promoted the activity of transfected GLI1 in luciferase reporter assays, HDAC6 remained inactive, despite prominent protein expression (Fig. 3F). This result demonstrates that HDAC6 does apparently not act through direct deacetylation of the GLI1 transcription factor and that it most likely functions through indirect mechanisms (which were overwhelmed by the high levels of transfected GLI1 in this assay).

In summary, we conclude that HDAC6 positively regulates Hh signaling, potentially at two levels: first, at the receptor level (PTCH1/SMO), and second, at a downstream point in the cascade at the level of the transcription factors.

After having observed the antagonistic effects of HDAC6 inhibition in specific Hh assays, we were interested to investigate the wider impact on the Hh-induced transcriptome. To this end, we treated MEF^[SHH] cells with control siRNA (targeting Firefly luciferase, siLuc) or RNAi against Hdac6. As reference constructs, we used siRNA targeting Smo, Gli1, or Gli2. RNA from these cells was subjected to microarray analysis. Confirming previous data, established Hh target genes, such as Ptch2 or Hip1 (Hhip), were downregulated in all sample groups compared with the control siRNA (not shown).

To compare global expression differences between siSmo and siHdac6 transfected samples, we used a scatter plot to represent the data. On the basis of the observed gene expression changes, we classified the transcripts into eight different categories, which can be seen as color-coded spots in the scatter plot in Fig. 4A. To investigate how these genes behave in a subsequent comparison with siGli1 and siGli2 conditions, we transferred these transcripts (and their color) into siGli1 and siGli2 scatter plots (Fig. 4A, middle and right).

Comparison of the expression changes induced upon siSmo and siHdac6 transfection revealed that, as expected, the well-established canonical Hh target genes (e.g., Ptch1 and Ptch2) fall into group 6 (Fig. 4A, left, purple spots) and were downregulated by both siRNAs. Not unexpectedly, these genes were also regulated in the same direction by siGli1 and siGli2 (purple and pink spots in Fig. 4A, middle and right). Interestingly, genes that were regulated in opposite directions by siSmo and siHdac6 (groups 1 and 8, orange and brown colored spots) were basically absent.

Another interesting set of genes fell into class 7 and was downregulated upon siHdac6, but was not affected by siSmo. However, a substantial fraction of these genes (red-colored spots in Fig. 4A) was shifted to the left in siGli1 and siGli2 plots, indicating that they were regulated by Gli transcription factors and by Hdac6, but not by Smo. Finally, sets of genes exist that were regulated by Smo, but not by Hdac6 (groups 4 and 5 (green and gray spots) in Fig. 4A). Intriguingly, a large fraction of these genes was not regulated by Gli1 or Gli2 either.

Given the fact that not all Smo-regulated genes were also affected by Gli1/2 (and vice versa), we defined a common set of genes termed the “Hh signature.” This gene set fulfilled the following stringent criteria: genes must have been (i) regulated in siSmo samples compared with siControl and (ii) in addition, siSmo-regulated genes had to be regulated in the same direction by either siGli1 or siGli2 transfection. This resulted in a set of genes encompassing 820 transcripts (the “Hh signature”). Comparing the Hh signature with genes regulated upon knockdown of Hdac6 revealed that 56% of the Hh signature were affected by loss of Hdac6 (Fig. 4B).

The expression data were validated by qPCR analysis of selected genes predicted to be regulated by “Gli1 and Hdac6” and by “Gli1 only.” As shown in Fig. 4C, canonical Hh targets, such as Hip1, Ptch1, and Ptch2, were all downregulated by siGli1 as well as by siHdac6 transfection in MEF^[SHH] cells. In contrast, the expression of Hsd11b1, Rasl11b, and Foxj1 were selectively affected by Gli1, but not by Hdac6, demonstrating that HDAC6 affects a subset of Gli target genes, as predicted by the microarray experiment.

In conclusion, HDAC6 activity affected more than half of the genes included in the Hh signature, but it did not affect all genes in the Hh signature. Interestingly, HDAC6 impinged also on the regulation of genes, which are Smo-independent yet Gli-dependent, arguing for some degree of downstream pathway control by HDAC6.

To learn more about the underlying molecular links between HDAC6 and Hh, we turned to GLI2 and GLI3, the principle downstream signal transducers in the cascade. Because HDAC6 had been implicated in protein degradation and clearance (35–41), we asked whether HDAC6 inhibition would affect the protein levels or the processing of GLI2/3. Treatment of NIH3T3 fibroblasts with the HDAC6 inhibitors ACY-1215 and CAY-10603 reduced the overall amount of full-length GLI2 protein as well as the amount of GLI3 activator (GLI3^A) and repressor (GLI3^R; Fig. 5A). On the mRNA level, only Gli2 expression was reduced, but Gli3 mRNA levels remained unchanged (Fig. 5B). These data suggest that HDAC6 plays a role in the stabilization of GLI3 protein and also takes part in the transcriptional control of Gli2 expression. In light of the finding that HDAC6 inhibition reduced the GLI3^R form, we speculated that this could derepress Hh target gene expression in nonstimulated cells. Therefore, we investigated Gli1 expression in wt MEF cells exposed to HDAC6 inhibitors. As shown in Fig. 5C, treatment with ACY-1215 or CAY-10603 indeed weakly stimulated the Hh pathway even in the absence of any Hh ligand or SAG. The target gene expression observed could not be reversed by the coadministration of the SMO antagonist SANT, which was able to block SAG-induced signaling (Fig. 5C). In contrast to the latter finding but in line with observations made before (Figs. 2 and 3), HDAC6 inhibition also reduced the maximal target gene expression induced by SAG. Similar findings were observed when Ptch1 expression was analyzed (Supplementary Fig. S4). Moreover, the induction of Gli1 transcription by HDAC6 inhibitors could also been demonstrated on the protein level in unstimulated cells. In line with qPCR data (Fig. 5C), HDAC6 blockade led to a reduction of maximal GLI1 protein levels in SAG-stimulated cells (Fig. 5D).

To delineate if the induction of Hh signaling by HDAC6 antagonists was mediated through GLI factors or through GLI-independent mechanisms, we used cells lacking functional Gli2 and Gli3 genes (double knockout Gli23^−/− MEFs; ref. 21). Treating these cells with pharmacologic HDAC6 inhibitors revealed that the Gli1 induction seen in wt cells was significantly reduced in the absence of Gli2 and Gli3 (Fig. 5E), suggesting that in particular the HDAC6-mediated effects on the GLI3^R protein levels might be important. However, positive effects through Gli1 might be needed for the ligand-independent Hh target gene induction as well. Therefore, we reduced endogenous Gli1 levels by RNAi-mediated knockdown of the remaining Gli1 expression in Gli23^−/− cells. As shown in Fig. 5F, siGli1 transfection reduced the Hh target gene induction (Ptch1 and Ptch2) compared with siControl cells, arguing also for the requirement of GLI1 activator function (Fig. 5F).

Taken together, we found a dichotomous behavior of HDAC6 inhibitors on Hh signaling. On the one hand, HDAC6 blockade results in a suppression of maximal Hh target gene expression; while on the other hand, HDAC6 interference stimulates basal target gene transcription. We hypothesize that the first observation might be caused by the reduced Gli2 mRNA expression upon HDAC6 inhibitor exposure, whereas the latter effect most likely involves the reduced GLI3^R protein in combination with GLI1 activator functions.

Next, we were interested to investigate whether an HDAC6-directed approach could in principle be exploited as a future therapy option in the treatment of medulloblastoma. To address this issue, we exposed primary mouse medulloblastoma cells (MB99–1 cells derived from the SmoA1 mouse model; refs. 22, 29) to increasing concentrations of HDAC6 inhibitors in culture and measured surviving cells by the CellTiter assays (Fig. 6A–C). As shown in Fig. 6A–C, we observed a striking reduction in cell number upon 48 hours of HDAC6 inhibition that correlated well with the increase in tubulin acetylation [r² = 0.927 (ACY-1215) and r² = 0.989 (CAY-10603) for inhibition of cell growth vs. acetylated tubulin]. This cytotoxicity seen was not limited to mouse medulloblastoma cells and was also observed in human cells (Daoy; data not shown). Importantly and in line with our previous data, histone acetylation was not increased by the HDAC6 inhibitors used and did not correlate with growth inhibition [r² = 0.377 (ACY-1215) and r² = 0.686 (CAY-10603) for inhibition of cell growth vs. acetylated histone H3; Fig. 6A–C].

These results prompted us to test the in vivo effects of small-molecule HDAC6 inhibition. To this end, we established s.c. allografts of primary SmoA1 medulloblastoma cells (MB99–1 cells). After tumors had reached a palpable size, mice (in groups of 5) were injected peritumoral s.c. with solvent or with 50 mg/kg of either vismodegib (GDC-0449, an FDA-approved SMO antagonist as a positive control) or ACY-1215 (see Materials and Methods for treatment scheme). Control (solvent-treated) tumors vastly increased in size over a 12-day period, whereas vismodegib- and ACY-1215-treated tumors remained small (Fig. 6D). This was also reflected in the mean tumor weight of dissected tumors taken at the end of the experiment (Fig. 6E). Notably, all drugs were well tolerated by the animals with no obvious signs of toxicity.

Immunohistologic examination of the tumor tissue revealed that in particular ACY-1215-treated tumors had high levels of cleaved caspase-3, a marker of apoptotic cell death (Fig. 6F), despite the fact that both drugs (vismodegib and ACY-1215) suppressed Hh pathway activity in the in vivo allografts, as measured by reduced Hh target gene expression (Supplementary Fig. S5).

HDAC6 belongs to the family of HDACs but mainly acts on nonhistone substrates. The enzyme can be found in the cytoplasm as well as in the nucleus of cells, depending on the differentiation status (42). In the cytosol, many substrates and interacting proteins have been described, including cortactin, Hsp90, and peroxiredoxins (43–49). One of the most prominent and best characterized protein substrate is the α-subunit of tubulin (8, 50), through which the enzyme regulates microtubule dynamics and vesicular transport (51–53). In addition, cytoplasmic functions of HDAC6 also include the formation of higher-order multiprotein complexes such as the protein-degrading aggresome as well as RNA/protein-containing stress granules (54, 55). In addition, HDAC6 is involved in the regulation of autophagy and several signal transduction cascades (39, 41, 56–59). Moreover, HDAC6 plays a role in the nucleus as a modifier of gene transcription (42, 60–62). Given the plethora of actions in which HDAC6 is involved, it is more than surprising to note that Hdac6-null animals are viable and healthy and have only a very gentle phenotype (19). It is currently not clear if the observed defects in immune response and the increased bone mineral density in these animals are linked to abnormal changes in Hh signaling, but it should be mentioned that this pathway plays a role in these processes.

In this study, we found HDAC6 to be overexpressed in Hh-induced medulloblastoma and we wondered whether this enzyme plays a functional role in the signal transduction of this pathway. In fact, HDAC6 overexpression has been observed in medulloblastoma before (28), but no link to the Hh system was made. Surprisingly, these authors saw no effect of the HDAC6 inhibitor Tubastatin on granule cell proliferation or allograft growth (28), which might be related to the relatively low concentrations used. In our hands, prominent Hh inhibition required the use of higher concentrations of HDAC6 inhibitors, indicative of the need to fully block HDAC6 to affect the Hh system. Despite the use of higher inhibitor concentrations, the compounds selectively inhibited HDAC6 and displayed no prominent effect on histone acetylation levels. We should point out that an allograft system always suffers from its artificial nature and that direct compound injections can potentially lead to high local inhibitor concentrations, although we tried to circumvent this issue by changing the injection sites and by not injecting directly into the tumor. It is worth noting that newer HDAC6 inhibitors are able to penetrate the blood–brain barrier and can therefore be tested under more physiologic conditions in autochthonous medulloblastoma mouse models in the future (63). Interestingly, these newer HDAC6 inhibitors have been shown to possess antidepressive characteristics, a phenomenon that we have linked to Hh inhibition before (64).

Collectively, our data revealed that endogenous HDAC6 exerts dual and opposing roles in Hh signaling: (i) HDAC6 is required for full pathway activation, while (ii) basal Hh signaling in the absence of ligands uses HDAC6 for full repression of target gene expression. This complex interplay between positive and negative actions might explain the limited phenotype in Hdac6 knockout animals. Furthermore, an induction of ligand-independent Hh signaling was not observed in the differentiation assays using C3H10T1/2 cells or in reporter assays with ShhL2 cells, suggesting either cell-type differences or that the HDAC6-induced derepression of Hh target genes is not sufficiently strong enough for certain biologic processes to occur. However, it is noteworthy that in our in vivo allograft model, mice receiving the SMO antagonist vismodegib presented with hair loss at the cutaneous injection sites (Supplementary Fig. S6). This is not unexpected as Hh signaling is well known to promote hair growth. The alopecia seen in vismodegib-treated mice was not observed in the ACY-1215 animal cohort, despite the fact that the Hh pathway was blocked. Our microarray analysis revealed that approximately half of the SMO/GLI–regulated genes are affected by HDAC6 and one hypothetical reason for the lack of alopecia might be that HDAC6 does not affect the Hh target genes required in hair cell biology. This observation warrants further investigation as it could be used to circumvent unwanted side effects associated with blocking Hh-dependent physiologic processes. In addition, an unexplored question as of now is whether HDAC6 inhibition affects SMO-dependent, but GLI-independent noncanonical signaling (65, 66).

We hypothesize that the impact of HDAC6 on Hh signaling could be linked to its effect on tubulin acetylation and is not mediated by direct deacetylation of GLI proteins through HDAC6. Posttranslational modifications of tubulin have been shown to affect motor protein-driven transport along microtubule. As Hh signaling critically requires IFT transport toward and from primary cilia (67–69), interfering with these transport processes might, in turn, suppress the precisely coordinated Hh signal transduction. As such, prolonged increased tubulin acetylation might functionally (not morphologically) resemble cilia defects such as those typically induced by loss-of-function of IFT components. In fact, certain HDAC6-related aspects of our data presented here, such as the reduced GLI3^R levels and the ligand-independent pathway activation, are in agreement with previous reports on cilia defects (70–74). The fact that HDAC6 inhibition negatively affects the Hh pathway in Sufu^−/− cells, which signal cilium independently (75, 76), suggests that other intracellular transport processes might be required downstream of SMO and primary cilia.

An alternative scenario would involve the nuclear HDAC6 fraction that might impinge on transcriptional complexes governing Hh target gene expression, such as Gli2 transcription. Importantly, we could demonstrate the HDAC6 specificity of our reagents in several instances, and thus exclude the role of histone-directed nuclear members of the HDAC family, which also modulate Hh signaling (34).

Interestingly, recent work implicates cilia-like processes at the immunologic synapse (77), a cellular structure that had previously been linked to HDAC6 (78). However, further investigations are needed to elucidate a function for HDAC6 in immunologic synapse-triggered Hh signaling.

HDAC6 has been shown to possess oncogenic traits in several tumors (25, 79, 80). In light of these findings and the requirement for HDAC6 to achieve maximal Hh signaling, we evaluated the impact of an HDAC6-directed therapy in an in vivo allograft model of medulloblastoma. HDAC6 inhibition had striking effects on medulloblastoma cell survival in vitro and was as effective as the positive control vismodegib in vivo. In light of the fact that ACY-1215 (but not vismodegib) induced strong and widespread apoptosis, longer treatment times might even induce complete tumor eradication.

In summary, we could identify an important function of HDAC6 in regulating mammalian Hh signaling. This provides an interesting novel drug target for the treatment of Hh-driven malignancies.

No potential conflicts of interest were disclosed.

Conception and design: M. Lauth

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.S. Holz, V. Fendrich

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.S. Holz, F. Finkernagel, M. Lauth

Writing, review, and/or revision of the manuscript: P.S. Holz, F. Finkernagel, M. Lauth

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.S. Holz, V. Fendrich

Study supervision: M. Lauth

Other (performed experiments): P.K. Dhanyamraju

The authors thank Julia Dick for expert assistance in animal work and Aninja Baier for help with histology.

This work was supported by grants obtained from the German Research Foundation (DFG LA2829/1-1 and LA2829/6-1, to M. Lauth), the University Medical Center Giessen and Marburg (UKGM, to M. Lauth), and the Head–Neck Tumor Research Foundation (to M. Lauth).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.