Neuronal SETD2 activity links microtubule methylation to an anxiety-like phenotype in mice

See Andrieux and Sadoul (doi:10.1093/brain/awab265) for a scientific commentary on this article.

Koenning et al. show that disabling microtubule methylation by SETD2 causes an autism-associated phenotype in mice without disrupting histone methylation or gene expression. These data suggest a new mechanism by which mutations in SETD2 and other chromatin remodelers may contribute to ASD via cytoskeletal rather than chromatin defects.

Abstract

Gene discovery efforts in autism spectrum disorder have identified heterozygous defects in chromatin remodeller genes, the ‘readers, writers and erasers’ of methyl marks on chromatin, as major contributors to this disease. Despite this advance, a convergent aetiology between these defects and aberrant chromatin architecture or gene expression has remained elusive. Recently, data have begun to emerge that chromatin remodellers also function directly on the cytoskeleton. Strongly associated with autism spectrum disorder, the SETD2 histone methyltransferase for example, has now been shown to directly methylate microtubules of the mitotic spindle. However, whether microtubule methylation occurs in post-mitotic cells, for example on the neuronal cytoskeleton, is not known. We found the SETD2 α-tubulin lysine 40 trimethyl mark occurs on microtubules in the brain and in primary neurons in culture, and that the SETD2 C-terminal SRI domain is required for binding and methylation of α-tubulin. A CRISPR knock-in of a pathogenic SRI domain mutation (Setd2^SRI) that disables microtubule methylation revealed at least one wild-type allele was required in mice for survival, and while viable, heterozygous Setd2^SRI/wtmice exhibited an anxiety-like phenotype. Finally, whereas RNA-sequencing (RNA-seq) and chromatin immunoprecipitation-sequencing (ChIP-seq) showed no concomitant changes in chromatin methylation or gene expression in Setd2^SRI/wtmice, primary neurons exhibited structural deficits in axon length and dendritic arborization. These data provide the first demonstration that microtubules of neurons are methylated, and reveals a heterozygous chromatin remodeller defect that specifically disables microtubule methylation is sufficient to drive an autism-associated phenotype.

See Andrieux and Sadoul (doi:10.1093/brain/awab265) for a scientific commentary on this article.

Introduction

SETD2 (SET domain containing 2 histone lysine methyltransferase) is a dual-function lysine methyltransferase responsible for both the histone H3 lysine 36 trimethyl (H3K36me3) mark on histones and the α-tubulin lysine 40 trimethyl (α-TubK40me3) mark on microtubules.^1,2 SETD2 has also recently been shown to trimethylate actin on lysine 68 (ActK68me3), an activity mediated by interaction with the Huntington protein (HTT) and the actin-binding adapter protein HIP1R.³ These cytoskeletal activities of SETD2 play several roles in the cell. The ActK68me3 mark is seen at the leading edge of the cell, and is involved in actin polymerization and cell migration. α-TubK40me3 methylation occurs on spindle and midbody microtubules and is required for proper chromosome segregation during mitosis and cytokinesis. However, whether microtubule methylation plays a role in interphase cells, or post-mitotic cells such as neurons, has not been explored.

In addition to its catalytic SET domain, the Set2-Rpb1 interaction (SRI) domain of SETD2 contributes to its methyltransferase activity by facilitating substrate recognition. The SRI domain binds to the highly phosphorylated C-terminal domain (CTD) of RNA polymerase II, which facilitates delivery of SETD2 to chromatin and methylation of histones during transcription.⁴ The SRI domain is also important for SETD2 methylation of α-tubulin, as pathogenic mutations in this domain abrogate the ability of SETD2 to methylate microtubules.^2,5 In both in vitro methyltransferase assays and cell-based rescue experiments with wild-type versus SETD2 constructs containing SRI domain mutations, SRI-domain mutant SETD2 is able to methylate histones but is deficient for α-tubulin methylation.^2,5,6

Homozygous inactivation of Setd2 is embryonic lethal,⁷ and monoallelic inactivation occurs in several disease settings. For example, in clear cell renal cell carcinoma, monoallelic inactivation of SETD2 due to 3p loss occurs in >95% of all tumours, and has been shown to result in haploinsufficiency for microtubule, but not histone, methylation.^6,8 In this setting, SETD2 haploinsufficiency causes genomic instability, and is thought to be an early driver of cancer progression. Haploinsufficiency is also thought to underly the pathogenicity of heterozygous SETD2 mutations associated with autism spectrum disorder (ASD). SETD2 is classified as a high confidence category 1 gene in the latest SAFARI database release (https://gene.sfari.org).⁹ While SETD2 is expressed in the brain (https://mouse.brain-map.org/gene/show/88100) and linked to ASD, little is known about SETD2 function(s) in the brain, whether microtubule structures in post-mitotic cells are methylated, or how heterozygous mutations in SETD2 contribute to neurodevelopmental disease.

We report here that cytoskeletal methylation by SETD2 occurs in the mouse brain, and is prominent in neurites and growth cones of post-mitotic neurons. Homozygosity or hemizygosity for a mutant SRI allele, Setd2^SRI/SRI or Nestin-Cre:Setd2^flox/SRI, respectively, was lethal even though this SRI domain mutation preserves catalytic methyltransferase activity and binding to phosphorylated RNA polymerase II. While heterozygous Setd2^SRI/wt mice with only one wild-type (wt) allele lived to adulthood, male Setd2^SRI/wt mice exhibited an anxiety-like phenotype, and cultured neurons from Setd2^SRI/wt mice showed significant deficits in both axon length and dendritic arborization. These data demonstrate that heterozygosity for an α-tubulin methylation-defective allele is sufficient to cause an ASD-associated co-morbidity (anxiety), and suggest that defects in cytoskeletal processes involved in neuronal differentiation and dendritic maturation may contribute to disease pathogenesis in individuals carrying heterozygous SETD2 mutations.

Materials and methods

Generation of Setd2^SRI/wt mice and genotyping

Point missense mutant R2483H allele of Setd2 (Setd2^SRI/wt) mice were generated by the CRISPR/Cas9 system using C57BL/6J (GRCm38) mice with the single guide RNA (sgRNA) sequence T TTC AAG CAC CTC GCC CGA AAG G in exon 20. The R2483H point variant (CGA → CAT) was created. Two silent mutations L2481L (CTC → CTT) and A2482A (GCC → GCG) were added to interrupt the sgRNA site and create a novel FspI restriction site for genotyping. PCR was performed using genomic DNA from tail biopsies. The primers were TTG TTG GCC TAG ACA GCA GC and GAT TGG GGC AAG CTG GTA CA. Digestion of the PCR product (408 bp) with FspI generates 214 bp and 194 bp DNAs in R2483H mutants. All mice were maintained in the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC)-accredited facility and sacrificed according to the protocols approved by Baylor College of Medicine institutional animal care and use committee (IACUC).

Immunohistochemistry on mouse brains

For immunohistochemistry on adult brains, animals were paraformaldehyde (PFA) perfused and tissue was washed progressively in 70–100% ethanol, chloroform and paraffin. Tissue was embedded in paraffin and sections were cut. Slides were incubated at 65°C for 15 min, then dewaxed in xylene and rehydrated through alcohols in distilled water. Antigen retrieval was performed by microwaving in citric acid for 30 min. Slides were washed in PBS, incubated in 1.8% hydrogen peroxide for 10 min at room temperature, washed in PBS and incubated in 10% normal horse serum with 0.1% Tween-20 for 30 min. Primary antibody was applied overnight in a humidity chamber. Following three PBS washes, goat anti-rabbit was applied at 1:200 for 30 min and ABC Elite Kit (Vector Laboratories) was used at 1:100 dilution. Slides were incubated in 3,3′-diaminobenzidene (DAB) solution, dehydrated through alcohols, cleared in xylene and mounted using Cytoseal^TM.

Immunoblotting and immunoprecipitation assay

Mouse brain tissues were lysed using TissueLyser II (Qiagen) in RIPA buffer (150 mM NaCl, 1.0% IGEPAL^® CA-630, 0.5% sodium deoxycholate, 0.1% SDS, 50 mM Tris, pH 8.0) containing 1 mM podophyllotoxin (Santa Cruz Biotechnology) to depolymerize microtubules and protease phosphatase inhibitor cocktail (MilliporeSigma) according to the manufacturer’s manual. Methylated tubulin was immunoprecipitated from cell extracts using anti-α-TubK40me3 (Me3^K40) antibody^2,10 and Pierce Protein G magnetic beads (Thermo Fisher Scientific) in PBST buffer (3.2 mM Na₂HPO₄, 0.5 mM KH₂PO₄, 1.3 mM KCl, 135 mM NaCl, pH 8.0) containing 1% Triton^TM X-100 with protease inhibitor cocktail (MilliporeSigma). The immunoprecipitated sample was subjected to SDS-PAGE gel electrophoresis under denaturing conditions, and subsequently immunoblotted using antibodies against α-tubulin (12G10) (DSHB). The immunoblots were performed using antibodies against H3K36me3 (Me3^K36, Active Motif), or histone H3 (Cell Signaling Technology).

Primary hippocampal neuron culture

Hippocampal and cortical neurons were isolated from mice at day of gestation (P0) as previously described.¹¹ After isolation, cortices and hippocampi were dissociated via trypsin and DNase incubation for 15 min and 5 min, respectively, then hand-triturated through a fire rounded pipette. Cells were plated at a density of 5 × 10⁵/well on 22-mm glass coverslips (VWR) in 6-well plates (Corning). Coverslips were coated with 0.5 mg/ml poly-d-lysine (PDL; Sigma) in borate buffer at 37°C overnight prior to plating cells. Cultures were maintained in Neurobasal^TM medium supplemented with B-27^TM, GlutaMAX^TM, and penicillin/streptomycin (all from Life Technologies) in a humidified 37°C incubator with 5% CO₂.

Primary hippocampal neuron culture and immunofluorescence microscopy

Hippocampal neurons were isolated from P0 mouse brains and cultured on PDL-coated coverslips for the indicated days. Neurons or mouse embryonic fibroblasts (MEFs) were fixed using 4% PFA solution in PBS buffer at 37°C temperature for 30 min, and followed by permeabilization using a 0.3% Triton^TM X-100 solution in PBS buffer for 30 mins. The cells were washed using PBS buffer and blocked in blocking buffer (10% goat serum, 0.3% Triton^TM X-100) for 1 h at room temperature. The cells were incubated in blocking buffer using primary antibodies overnight at 4°C. Primary antibodies were diluted as follows: SETD2 (Sigma, 1:500), rabbit anti-Me3^K36 (H3K36me3, 1:1000, Active motif), Me3^K40 (α-TubK40me3, 1:1000), MAP2 (1:1000, Encore), α-tubulin 12G10 (1:30, DSHB), acetylated α-tubulin (1:1000, Santa-Cruz), and tyrosinated α-tubulin (1:500, Millipore). After five rounds of washing for 10 min each using PBS buffer, cells were incubated for 1 h in blocking buffer with corresponding secondary antibodies [Alexa Fluor^® (H+L) goat anti-mouse 546 and goat anti-rabbit 488 (Invitrogen)] at a dilution of 1:1000. Following more than three washes, for 10 min each using PBS buffer, cells were then post-fixed to stabilize the signal for 5 min using 4% PFA in PBS buffer. The cells were counterstained using DAPI or phalloidin (1:50) for subsequent visualization of nuclei or actin, respectively. Coverslips were mounted in SlowFade Gold (Molecular Probes) and imaged using deconvolution and confocal microscopy [DeltaVision Elite (GE), Ti Eclipse (Nikon)].

Glutathione S-transferase pull-down assays

Glutathione S-transferase (GST )-SRI fusion protein expression vector containing the SRI domain (2438–2564 amino acids) of human SETD2 was generated in a pGEX-4T-1 (GE Healthcare) vector by cloning fragments of human SETD2 cDNA (GenScript) into the BamHI and XhoI sites via PCR using In-Fusion^® HD Cloning Kit (Clontech). GST-SRI-R2510H were generated using QuikChange II Site-directed mutagenesis kit (Agilent). GST-SRI constructs were transformed into BL21 E. coli and induced using 0.5 mM IPTG overnight at 15°C. Cells were collected by centrifugation and digested with 0.5 mg/ml lysozyme (Thermo Scientific) and 250 U Benzonase^® (Millipore) on ice for 30 min in PBST buffer containing 1% Triton^TM X-100 with protease inhibitor cocktail (MilliporeSigma). The cells were subsequently sonicated for 15 s three times. The GST-SRI fusion proteins were purified using glutathione-agarose beads (Pierce) and analysed on SDS-PAGE gels stained with Coomassie blue. The purified GST-SRI and GST-SRI-R2510H fusion proteins (5 µg) were incubated with 2 μg of purified tubulin proteins from porcine brain (Cytoskeleton) or 200 µg of mouse brain or cell extracts in PBST pH 7.4 containing 5% BSA overnight at 4°C. Following washing using PBST, the protein complexes were run on SDS-PAGE gels. Immunoblot analyses were performed using a mouse anti-α-tubulin antibody, 12G10 (DSHB), or antibodies against total RNA polymerase II or phospho-Rpb1 CTD (Ser2/Ser5) (D1G3K) (Cell Signaling Technology), which recognizes endogenous levels of Rpb1 only when the CTD heptapeptide repeats are dually phosphorylated at Ser2 and Ser5 but not singly phosphorylated at Ser2, Ser5 or Ser7.

Mouse behavioural studies

Behavioural tests were performed similar to prior studies.^12,13 Two different cohorts of animals at 5–9 months of age were used. Procedures were performed during the animals’ light cycle. For each test, animals were habituated to the test room for 30 min; room lighting was set to 150 lx and ambient sound was provided by white noise generators set for 60 dB of white noise. Animals were tested across each test in the battery (testing order: elevated plus maze, open field, light-dark box) with at least 24 h between tests. Animal testing order within a test was organized to prevent animals from being single housed immediately prior to being tested. The experimenter was blinded to the genotype of animals during testing.

Elevated plus maze

After the habituation period, animals were placed in the centre of a maze consisting of two closed arms and two open arms (each arm 25 × 7.5 × 0.5 cm) elevated 50 cm above ground level; the arms of the maze were equidistant from the centre platform. The position of the animal was tracked using an overhead camera and ANY-maze software for 10 min. Following testing, the animals were returned to their home cage.

Open field assay

After the habituation period, animals were placed in the centre of a 40 × 40 × 30 cm chamber equipped with photobeams (Accuscan) to record activity during a 30-min test period. Following testing, the animals were returned to their home cage.

Light-dark box

After the habituation period, animals were placed in the light side (36 × 20 × 26 cm) of a chamber separated from the dark side, which was a chamber covered with black plastic (15.5 × 20 × 26 cm) with a 10.5 × 5 cm opening. The position of the animals was tracked using occurrence of photobeam breaks for 10 min (Accuscan). For light dark statistical analysis, one male Setd2^wt/wt mouse was excluded due to an equipment problem-induced artefact.

Sholl analysis

Hippocampal neurons of heterozygous SRI mutants and littermates were fixed at 7 days in vitro (DIV7) and immunostained for MAP2 and DAPI as described above. Eight-bit images of neurons were acquired on a 40× objective (Nikon) and traced using the NeuriteTracer plugin¹⁴ in ImageJ (Fiji version, NIH). The resulting tracing stack was split, and individual images were evaluated using the Sholl analysis feature of ImageJ.

Statistical analysis

Scholl curves were analysed using two-way ANOVA with Bonferroni post hoc correction in GraphPad Prism. Behavioural data were analysed using commercially available SPSS statistics analysis software. Behavioural data were analysed by non-parametric Mann-Whitney U-test with genotype as a factor. Because the presence of anxiety-like features would also predict the direction of effect, the one-sided significance was used for outcome measures with expected directions of effects for a model with anxiety and autism-like features (elevated plus maze: reduced open entries, increased closed entries, reduced open time, increased closed time, increased latency to enter light; open field assay: increased foecal boli, decreased vertical activity, increased movement stereotypies, decreased centre to total distance ratio; light-dark box: decreased light entries, decreased light duration). All other outcome measures, such as total distance travelled during open field assay and light-dark box, which would not have a predicted direction of effect, were analysed using the two-sided significance. Summary figures were prepared using commercially available GraphPad Prism software to illustrate individual data-points, mean values, and standard error of the mean (SEM).

Data availability

The authors confirm that the data supporting the findings of this study are available within the article and its Supplementary material. The RNA-sequencing (RNA-seq) and the chromatin immunoprecipitation-sequencing (ChIP-seq) data that support the findings of this study are available from the corresponding author, upon reasonable request at NCBI’s Gene Expression Omnibus site (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE173093).

Results

SETD2 expression and cytoskeleton methylation occur in the brain and primary neurons

We found SETD2 was highly expressed in embryonic and neonatal mouse brains, and while detectable, expression dramatically decreased in the adult mouse brain (Fig. 1A). While expression decreased in the adult brain, immunohistochemistry confirmed SETD2 was expressed in individual cells of most brain regions, including the cortex, hippocampus, and cerebellum (Fig. 1B). SETD2 was most prominent in the nucleus, although dendrites of Purkinje cells also exhibited immunoreactivity (Fig. 1B, arrows). In primary neurons isolated from neonatal (P0) brains, SETD2 expression could also be seen in cortical (Fig. 1C) and hippocampal (Fig. 1D) neurons by immunostaining.

Figure 1.

SETD2 expression in mouse brain and primary neurons. (A) Western blots for SETD2 using brain lysates of embryo at 17.5 days post coitum (E17.5), neonate (P0), and adult. (B) Representative immunohistochemistry images of cortex, hippocampus and cerebellum of mouse adult brain using an antibody against SETD2. Scale bar = 100 µm; 10 µm in box. (C and D) Representative immunohistochemistry images of SETD2 expression in primary cortical (C) and hippocampal (D) neurons at DIV7 using antibodies against SETD2 (red) and MAP2 or α-tubulin (green), respectively.

Similar to the expression pattern seen for SETD2, we also observed the α-TubK40me3 mark in the brain and in neurons. Methylation of α-tubulin was seen using a previously validated methyl-specific antibody (Me3^K40) raised against the SETD2 K40 trimethyl epitope of α-tubulin,¹⁰ and was highest in embryonic and neonatal brains, relative to the adult brain (Fig. 2A). Although SETD2 levels were reduced in adult brains, with commensurate reduction in α-tubulin methylation, histone H3K36me3 methylation was equivalent at time points examined, including embryonic, postnatal and adult brain (Fig. 2A). Immunostaining using the Me3^K40 antibody confirmed methylation in the adult brain, including the soma and apical dendrites of cortical neurons (Fig. 2B and Supplementary Fig. 1A), as well as axon bundles of mossy fibres in the CA3 region of the hippocampus. Corresponding to SETD2 localization in the cerebellum, methylation was also detected in the dendrites of Purkinje cells (Fig. 2B).

Figure 2.

Detection of α-TubK40me3 in mouse brain. (A) α-TubK40me3 detection by western blots from mouse brain lysates using antibodies against Me3^K40. (B) Representative immunohistochemistry images of cortex, hippocampus and cerebellum of mouse adult brain using an antibody against Me3^K40. Scale bar = 100 µm; 10 µm in box.

Primary neurons cultured from the cortex and hippocampus of P0 embryos exhibited prominent cytoskeletal methylation using both the Me3^K40 antibody (Fig. 3) and a second commercially available antibody (Me3^K36) raised against the SETD2 methyl epitope on K36 of histone H3 that recognizes the K68me3 SETD2 methyl epitope of actin³ and K40me3 of α-tubulin¹⁰ (Supplementary Fig. 1B and C). As shown in Fig. 3A, α-TubK40me3 in DIV7 cortical neurons was highest in MAP2-positive neurites, a dendrite marker. Staining intensity with the Me3^K40 antibody (green peaks) relative to α-tubulin (red peaks) was much lower in MAP2-negative axons relative to MAP2-positive dendrites (Fig. 3A). The α-TubK40me3 methylation in these neurons was confirmed by mass spectrometry (Fig. 3B). Primary cultured hippocampal neurons also exhibited higher methylation in dendrites (MAP2-positive) than in the proximal axons (MAP2-negative) (Fig. 3C). In these neurons, the methyl mark extended past acetylated α-tubulin into the neurite tip (Fig. 3D and E). Methylation also co-localized with tyrosinated α-tubulin (Supplementary Fig. 1B), especially in small branches of axons that did not stain for acetylated tubulin (Fig. 3E). In the growth cone, immunoreactivity for the SETD2 methyl mark was also seen in filopodia (Supplementary Fig. 1B and C) and co-localized with phalloidin-stained polymerized actin (Supplementary Fig. 1C), suggesting SETD2 is also methylating the actin cytoskeleton in these structures, consistent with previously reported data in cell lines.³ On neurites, the α-TubK40me3 mark co-localized with both tyrosinated and acetylated α-tubulin in mouse cortical (Supplementary Fig. 2A) and hippocampal (Supplementary Fig. 2B) neurons, but in growth cones the methyl mark could be seen co-localized with tyrosinated α-tubulin in microtubules distinct from acetylated α-tubulin. While the K40 residue on α-tubulin can be methylated or acetylated, any given lysine residue can only have one of these modifications. The stoichiometry of K40 methylation and acetylation within the microtubule polymer is not known, but western analysis with quantitation (Supplementary Fig. 2C) showed the expected inverse relationship between αTubK40me3 (which decreases between embryonic Day 17.5 and adulthood) and α-TubK40ac (which increases during this period).

Figure 3.

Detection of α-TubK40me3 in primary neurons. (A) Representative immunostaining images of mouse cortical neurons at DIV7 using antibodies recognizing α-TubK40me3 (Me3^K40), total α-tubulin and MAP2. Arrowhead = axon. The fluorescence intensity profiles of α-TubK40me3 (green) and α-tubulin (red) at the arrow is shown. Scale bar = 50 µm. (B) Mass spectrometry analysis of primary cortical neurons at DIV30. All the red peaks show trimethylation on K40 of α-tubulin. (C) Representative immunostaining images of mouse hippocampal neurons at DIV31 using antibodies recognizing α-TubK40me3 (Me3^K40), α-tubulin, and MAP2 (a dendrite marker). Scale bar = 50 µm. (D) Representative immunostaining images of mouse hippocampal neurons at DIV2 using antibodies recognizing α-TubK40me3 (Me3^K40), acetylated (Ac)-α-tubulin, and actin (phalloidin). Scale bar = 50 µm. (E) Representative immunostaining images of rat hippocampal neurons at DIV8 using antibodies recognizing H3K36me3 as well as α-TubK40me3 and ActK68me3 (Me3^K36 from Active Motif), tyrosinated (Tyr) and acetylated (Ac) α-tubulin. Scale bar = 50 µm; 5 µm in box. The circled area shows that methylation co-locates with tyrosinated α-tubulin but not acetylated α-tubulin (Ac-α-Tubulin).

SETD2 SRI domain mutation disrupts SETD2 binding to α-tubulin

The SRI domain of SETD2 plays a role in methylation of two of its substrates: histone H3 and α-tubulin. SETD2 methylation of histone H3 is facilitated by binding of the SRI domain to the highly repetitive CTD of RNA polymerase II (Fig. 4A), which when transcriptionally active is phosphorylated at serine 2 (pS2) and 5 (pS5) of the hexapeptide CTD repeats.⁴ The SRI domain also plays a critical role in methylation of microtubules as mutations in the SRI domain abrogate α-tubulin methylation in cells as well as in in vitro methylation assays,^2,5 although the basis for the dependency on the SRI domain for α-tubulin methylation is unknown.

Figure 4.

SRI domain mediates SETD2 binding to RNA polymerase II and α-tubulin. (A) A hypothesis of R2510H mutation effect on α-TubK40me3. R2510H mutation impairs the methylation of α-TubK40me3 on α-tubulin but not those of H3K36me3 on histone H3 and ActK68me3 on actin. (B) Phosphomimetics of phosphorylated C-terminal domain (pCTD) repeats of RNA polymerase II. The C-terminal tails (CTT) of all human α-tubulin isotypes are aligned to two or three repeats of pCTD heptapeptides. Aspartic acids (D) or glutamic acids (E) were regarded as phosphorylated serine 2 and serine 5. (C) R2510H missense mutation generated in Setd2 GST-SRI domain construct.

We noted that phosphomimetic aspartic and glutamic acid residues in the C-terminal tail (CTT) of α-tubulin align with the phosphorylated pS2 and pS5 residues of the hexapeptide CTD repeat (Fig. 4B), suggesting that the SETD2 SRI domain could participate in methylation of α-tubulin by binding to the α-tubulin CTT. To test this hypothesis, we utilized an SRI domain R2510H missense mutation (Fig. 4C), which occurs frequently in cancer, and is known to selectively impede tubulin but not histone methylation by SETD2.^2,5,6 We confirmed that a SETD2 construct with an R2510H mutation (tSETD2-R2510H) could rescue chromatin methylation (H3K36me3) but not microtubule methylation (α-TubK40me3), even when expressed at levels much higher than endogenous SETD2 (Fig. 5A). Because SETD2 also methylates actin,³ we wanted to determine if the R2510H mutation also impacted actin methylation. We found the R2510H and several other SETD2 SRI mutant alleles bound actin as efficiently as wild-type SETD2 (Supplementary Fig. 3A), and that a tSETD2-R2510H mutant was able to methylate actin as efficiently as wild-type tSETD2 (Supplementary Fig. 3B–E).

Figure 5.

Impact of SRI domain mutation on SRI binding to CTD of RNA polymerase II and CTT of α-tubulin. (A) Rescue of histone H3K36me3 methylation by tSETD2-R2510H mutant protein but not α-TubK40me3 methylation. The R2510H missense mutant protein expression vector, which was generated in a fully functional N-terminal truncated SETD2 (tSETD2) containing catalytic SET domain and SRI domain (1418–2562 amino acids), was transfected into SETD2 null 786–0 cells. The methylated tubulin was immunoprecipitated using antibody against α-TubK40me3 (Me3^K40) and immunoblotted using the indicated antibodies. (B) GST pull-down assay using GST, GST-SRI domain, and GST-SRI domain with R2510H mutation. Purified GST fusion proteins were incubated with porcine brain tubulin proteins and analysed by western blot using an α-tubulin antibody after pulling down. Three independent experiments were quantitated. Data are presented as all points and were analysed using Student’s t-test. (C) GST pull-down assay using mouse embryonic brain lysates at embryonic Day 17.5. Western blots were performed using antibodies recognizing total RNA polymerase II, phosphorylated CTD of RNA polymerase II at both serine 2 and serine 5, or α-tubulin. At least three independent experiments were quantitated. Data are presented as all points and were analysed using Student’s t-test.

To test the hypothesis that attenuation of α-tubulin methylation by the SETD2 SRI domain mutation was due to decreased binding to α-tubulin, we performed GST pull-down assays using purified wild-type or SRI-R2510H mutant GST-fusion proteins. The wild-type GST-SRI domain bound α-tubulin protein purified from porcine brain (Fig. 5B) as well as α-tubulin in mouse embryonic brain lysates (Fig. 5C). Both of these interactions were significantly reduced with the GST-SRI-R2510H mutant fusion protein (Fig. 5B and C). In contrast, binding to the phosphorylated fraction of RNA polymerase II was slightly affected by the presence of the SRI-R2510H mutation. Whereas an antibody that recognizes both phosphorylated and non-phosphorylated RNA polymerase II showed a smaller fraction of total RNA polymerase II bound to the GST-SRI-R2510H fusion protein, there was only a slight difference between the amount of phosphorylated RNA polymerase II bound by wild-type or SRI-R2510H mutant GST-fusion proteins (Fig. 5C). Together, these results indicate that the SRI domain plays a critical role in SETD2 engagement with its tubulin substrate.

Heterozygosity for a Setd2 α-tubulin methylation-deficient allele is haploinsufficient

ASD-linked mutations in SETD2 are deleterious and heterozygous, pointing to haploinsufficiency as the driver for pathogenesis in this disease.^9,15-23 Of the SETD2 mutations observed in ASD, 65% are nonsense/frameshift mutations occurring upstream of the SRI domain and/or missense mutations occurring in the set domain, both of which would cause loss of α-tubulin methylation due to loss of catalytic activity and/or inability to bind the CTT of α-tubulin (Supplementary Fig. 4). In addition, 4/20 frameshift/nonsense mutations occur downstream of the catalytic domain, resulting in loss of the C-terminal SRI domain (Supplementary Fig. 4). Thus, we asked if an R2510H SRI domain mutation that specifically inactivates α-tubulin methylation would phenocopy the predicted loss of SETD2 α-tubulin methyltransferase activity caused by the majority of ASD-linked SETD2 mutations, while preserving catalytic function (as well as chromatin and actin methylation), allowing separation of these key SETD2 functions.

To assess the impact of a heterozygous SRI domain SETD2 mutation, we developed a knock-in mouse model carrying the murine equivalent substitution to the human R2510H mutation (R2483H) using CRISPR-Cas9. Exon 22 of Setd2 was targeted via a sgRNA, and donor DNA carrying the codon substitution (CGA>CAT) as well as silent codon changes to introduce an additional FspI restriction site for genotyping (Fig. 6A and B). Sequencing and genotyping confirmed the presence of the knock-in or wild-type (Fig. 6B and C). While heterozygous Setd2^SRI/wt animals were viable, no adult homozygous Setd2^SRI/SRI animals were obtained from Setd2^SRI/wt × Setd2^SRI/wt crosses, suggesting the Setd2^SRI/SRI genotype was lethal. To investigate if a functional SRI domain was specifically required in the nervous system, we generated a conditional knockout using Nestin-Cre and wild-type floxed and SRI mutant Setd2 alleles. Of a total of 97 animals, no Setd2-null (Nestin-Cre;:Setd2^flox/flox) mice were seen, and only a single Nestin-Cre:Setd2^SRI/flox mouse (which died shortly after weaning) was observed, far below the expected ratio of 12.5% for those genotypes (Fig. 6D). These findings indicate that at least one fully functional SETD2 SRI domain is required in cells of the nervous system for viability.

Figure 6.

Generation of SRI mR2483 mutant mice and MEFs. (A) Generation of Setd2 R2483H (R2510 in human) mutant mice using CRISPR/Cas9. R2483H point variant (CGA → CAT) was created. (B) Sequencing for generation of mR2483H knock-in with CRISPR-CAS9. Silent mutations L2481L (CTC → CTT) and A2482A (GCC → GCG) were added to interrupt the sgRNA site and create a novel FspI restriction site (TGCGCA). (C) Representative genotyping data. PCRs were performed using genomic DNAs from mouse tails and their amplified DNAs were digested using the FspI restriction enzyme. The FspI digestion generates 214 bp and 194 bp in R2483H alleles from 408 bp PCR product. (D) Panel shows mouse litter numbers of each genotype from Setd2^SRI/wtand Nestin-Cre:Setd2^flox/SRI. (E) Representative immunostaining images of MEFs isolated from Setd2^wt/wt and Setd2^SRI/wt embryos using antibodies recognizing α-TubK40me3 (Me3^K40) and acetylated α-tubulin (Ac-Tubulin) with DAPI for DNA. (F and G) Representative images and line profiles of α-TubK40me3 through the midbodies of MEFs isolated from Setd2^wt/wt and Setd2^SRI/wt embryos. (H) Quantitation of α-TubK40me3 methylation across the midbodies of Setd2^wt/wt and Setd2^SRI/wt MEFs collected from three independent cultures (n ≥ 50). Highest fluorescent intensities on each midbodies were counted. Data are shown as all points and analysed using Student’s t-test.

Since both in vitro methyltransferase assays and cell line rescue experiments fail to preserve the normal methyltransferase-target stoichiometry, we performed additional Setd2^SRI/wt × Setd2^SRI/wt crosses to assess the impact of heterozygosity for the Setd2^SRI allele when a Setd2^wt allele is present. Assessing the impact of loss of SETD2 in neurons is confounded by potential full or partial redundancy between SETD2 and another recently discovered H3K36me3 methyltransferase in the brain, SETD5.²⁴ In addition, there are also other recently discovered α-tubulin methyltransferases, SET8²⁵ and SYMD2²⁶, and it is not known if these are redundant for mono-, di- or tri-methylation of different α-tubulin isotypes or microtubule structures in neurons.^27,28

However, SETD2 has been shown to be non-redundant for methylation of midbody microtubules.^2,6 Therefore, we assessed the impact of heterozygosity for the Setd2^SRI allele during mitosis in MEFs isolated from Setd2^SRI/wt × Setd2^SRI/wt crosses. As shown in Fig. 6, MEFs derived from Setd2^SRI/wt embryos exhibited a significant loss of the α-TubK40me3 mark on midbody microtubules relative to Setd2^wt/wt MEFs (P < 0.0001, Student’s t-test) (Fig. 6E and F). In contrast, chromatin methylation was essentially equivalent (<20% difference) in Setd2^SRI/wt compared to Setd2^wt/wt MEFs (Supplementary Fig. 5A), and in the brains of Setd2^wt/wt and Setd2^SRI/wt mice when normalized for total H3 (Supplementary Fig. 5B).

Furthermore, transcriptomic profiling via RNA-Seq of P0 brains from seven littermates (four Setd2^wt/wt and three Setd2^SRI/wt) showed that of the 10 571 genes expressed in the brain, animals carrying the SRI mutation differed from their wild-type littermates in expression of only 22 genes (Supplementary Fig. 5C). We also performed ChIP-sequencing for the transcription elongation histone modification H3K36me3 on the brains of these mice, and as shown in Supplementary Fig. 5D, only four genes were identified from this analysis with differential H3K36me3 peaks at the promoter/gene body, none of which overlapped with differentially expressed genes identified by RNA-seq. Furthermore, none of the differentially expressed genes identified by RNA-Seq exhibited differences in H3K36me3 (Supplementary Fig. 5D). These data indicated that in contrast to microtubule methylation, heterozygosity for the SRI-domain mutation had a negligible effect on SETD2 activity on chromatin and gene expression. Thus, the presence of the Setd2^SRI allele severely disabled the cytoskeletal but not chromatin activity of this methyltransferase, demonstrating Setd2^SRI/wt haploinsufficiency for microtubule methylation.

Heterozygous Setd2^SRI/wt mice exhibit an anxiety-like phenotype

To determine how Setd2 heterozygosity impacted behaviour, Setd2^SRI/wt and Setd2^wt/wt mice were held until adulthood for behavioural testing. We performed a focused battery of tests, including elevated plus maze, open field, and light-dark box, which revealed significant differences in endophenotypes consistent with an anxiety phenotype in Setd2^SRI/wt mice. The anxiety-like phenotype is more prominent in male Setd2^SRI/wt mice than those females. For male mice, these endophenotypes included reduced number of entries into the open arm of an elevated plus maze (P < 0.05) (Fig. 7A and B) but not the number of entries into the closed arm and the overall ratio of entries into the open arm (Supplementary Fig. 6A and B). The time spent in the arm also trended towards a reduction but did not reach significance (Fig. 7C and D). The latency to enter the open arm was increased in male Setd2^SRI/wt mice (P < 0.05) (Fig. 7E and F) but not to the closed arm (Supplementary Fig. 6C). Consistent with the avoidance of the open arm, male Setd2^SRI/wt mice spent more time in the closed arm of the maze (P < 0.05) (Fig. 7G and H). Further, the number of entries to the light side in the light-dark test was significantly reduced in both males and female Setd2^SRI/wt mice relative to Setd2^wt/wt mice (P < 0.05) (Fig. 7I and J). The time spent in the light side was significantly reduced in only male Setd2^SRI/wt mice (Supplementary Fig. 6D), while the total moved distance and the ratios of the time spent in light side to time spent in light and dark sides were not significant (Supplementary Fig. 6E and F). In the open field assay, Setd2^SRI/wt mutant mice showed similar locomotor activity to control Setd2^wt/wt littermates, as evident in no significant differences in the total distance travelled (Supplementary Fig. 6G). The moved distance ratio in the centre to total distance was reduced in Setd2^SRI/wt mice (Fig. 7K and L) due to a reduced distance in the centre, especially in male Setd2^SRI/wt mice (Supplementary Fig. 6H). Setd2^SRI/wt mice exhibited a reduced number of stereotypies (Supplementary Fig. 6I), and both male and female Setd2^SRI/wt mice exhibited reduced vertical rearing behaviour (P < 0.05) (Fig. 7M and N). Male Setd2^SRI/wt mice produced an increased amount of foecal boli (P < 0.05) during the open field assay (Fig. 7O and P), an indicator of stress.

Figure 7.

An anxiety-like phenotype of heterozygous Setd2^SRI/wt mutant mice. Mouse behavioural tests were performed using 37 Setd2^wt/wt mice (19 males and 18 females) and 36 Setd2^SRI/wt mice (13 males and 23 females). (A–H) Elevated plus maze was performed and selected outcome measures included number of entries into the open arm (A and B), time spent in the open arm (C and D), latency to entering the open arm (E and F), and time spent in the closed arm (G and H). Data were analysed in the entire cohort (A, C, E and G) and in the cohort split by sex (B, D, F and H). (I and J) Light-dark box test counting the number of entries into the light side in the entire cohort (I) or in the cohort spilt by sex (J). (K–P) Open field assay. Selected outcome measures include centre to total distance ratio (K and L), vertical rearing activity (M and N), and faecal boli produced during the test (O and P). Data were analysed for the entire cohort (K, M and O) or in the cohort split by sex (L, N and P). All plots are represented as minimum to maximum showing all points and were analysed using Mann-Whitney U-test.

This anxiety-like phenotype was independently reproduced in Setd2^SRI/wt versus Setd2^wt/wt mice at a second research institution (Supplementary Fig. 7). In this independent cohort of mice, while total distance travelled was again similar, both male and female Setd2^SRI/wt mice exhibited significantly decreased total centre distance (P < 0.05) in an open field test (Supplementary Fig. 7A). The time spent in the centre was significantly reduced in total Setd2^SRI/wt mice (P < 0.05) (Supplementary Fig. 7A). In this repeat, only male Setd2^SRI/wt mice showed the decreased vertical rearing behaviour (P < 0.05) (Supplementary Fig. 7B) probably due to the reduced sample size in the independent assays. We could not detect any significant behavioural phenotype in Setd2^SRI/wt mice by three chamber social interaction assays, cued fear conditioning, novel object recognition, and Y-maze spontaneous test (Supplementary Fig. 8). Overall, the behaviour of heterozygous Setd2^SRI/wt mice carrying one α-tubulin methylation-defective allele was consistent with an anxiety-like phenotype, a co-morbidity associated with autism, similar to what has been seen in other rodent ASD models.²⁹

Setd2^{SRI / wt} mutant neurons exhibit defects in axon length and dendrite branching

SETD2 shows a higher methyltransferase activity towards α-tubulin in the soluble α/β-tubulin dimer state than in the polymerized microtubule state.⁵ This predicts that haploinsufficiency could manifest in two ways: an overall decrease in α-TubK40me3 incorporated into polymerized microtubules (as seen in MEFs above), or alternatively, formation of fully methylated, but abnormal cytoskeletal structures if methylated α-tubulin is incorporated into polymers, but is rate-limiting due to decreased availability of methylated α/β-tubulin dimers. We did not detect a dramatic difference in the α-TubK40me3 mark on polymerized microtubules in primary neurons from Setd2^wt/wt versus Setd2^SRI/wt mice (Supplementary Fig. 5E). Rather, primary neurons from Setd2^SRI/wt mice showed significantly reduced complexity in dendritic arborization when compared to their wild-type Setd2^wt/wt littermates, as evidenced by Sholl analysis at DIV7 (Fig. 8A). While Setd2^wt/wt neurons had an average of ∼10 intersections at 200 µm distance from the cell centre, the average number of intersections in Setd2^SRI/wt neurons was significantly less, not exceeding an average of approximately eight intersections. These differences in intersections between Setd2^wt/wt and Setd2^SRI/wt animals was highly significant over multiple comparison points (P < 0.0001, two-way ANOVA, Bonferroni post hoc test). Setd2^SRI/wt hippocampal neurons also exhibited differences in axon extension as determined by length of the longest neurite at DIV5. Neurons of Setd2^SRI/wt mutant mice showed a 50% reduction compared to wild-type Setd2^wt/wt littermates: 380 µm versus 800 µm, respectively, in total 110 neurons from Setd2^wt/wt and total 119 neurons from Setd2^SRI/wt mice (P = 0.018, Student’s t-test of average neurite lengths in three independent cultures) (Fig. 8B). Thus, in the setting of a heterozygous SETD2 mutation and associated haploinsufficiency for α-tubulin methylation, defects in neurite outgrowth and dendritic arborization occur in neurons of Setd2^SRI/wt mutant mice.

Figure 8.

Reduced dendritic arborization and axon length in primary hippocampal neurons of heterozygous Setd2^SRI/wtmutant mouse. (A) Representative immunostaining images of dendrite branching in primary hippocampal neurons at DIV7 from three Setd2^wt/wt (total 68 neurons) and three Setd2^SRI/wt (total 67 neurons) P0 mouse neonates (top). At least 20 neurons were analysed in each mouse. Scholl analysis (bottom) was performed using neurons stained for MAP2 and DAPI. P < 0.0001, two-way ANOVA, Bonferroni post hoc test. (B) Representative immunostaining images of primary hippocampal neurons at DIV5 stained for acetylated (Ac) α-tubulin from Setd2^wt/wt and Setd2^SRI/wt mice (left). Quantification (right) of the distance from cell body to the terminal of the longest neurite were measured using NIS-Elements AR software. Data for analysis of neurite lengths were obtained from three independent primary neuron cultures. In these biological replicates, we analysed 28, 32 and 51 neurons from Setd2^wt/wt mice (total 110 neurons), and 29, 41 and 49 neurons from Setd2^SRI/wt mice (total 119 neurons), respectively. All plots are represented as minimum to maximum showing all points and were analysed using Student’s t-test. Scale bars = 100 µm.

Discussion

We report here that α-TubK40me3 is a new post-translational modification (PTM) of neuronal microtubules. Heterozygosity of Setd2, which writes this mark, leads to deficits in axon extension and dendrite arborization. Mice carrying a mutation in the SETD2 SRI domain show behavioural deficits, suggesting a potential dual dependency of both epigenetic and cytoskeletal regulation for proper brain development. The K40 residue of α-tubulin was previously shown to be important for dendritic arborization in fruit flies.³⁰ While this was thought to be due to a requirement for acetylation of K40, a microtubule PTM known to be involved in axonal transport,³¹ our findings of decreased dendritic arborization in SRI domain mutant mice suggest that K40 methylation could be playing an equally important role in this process.

Whereas methylation of α-tubulin occurs primarily on unpolymerized dimers, K40 acetylation by α-tubulin acetyltransferase (ATAT1/MEC-17) occurs on polymerized microtubules.^32–35 We observed global α-TubK40me3 levels inversely correlated with age, being higher in the embryonic and neonatal brain relative to the adult brain, whereas α-TubK40ac levels increased over this same interval. In some cases, localization of α-TubK40me3 also differed from α-TubK40ac, being co-localized with tyrosination, a marker for dynamic microtubules, at the tip of neurites and growth cones where α-TubK40ac is absent.

We also observed that dendrites were less branched and neurite growth retarded/delayed in Setd2^SRI/wt primary hippocampal neurons. Consistent with our data, loss of α-tubulin acetylation in primary hippocampal neurons (potentially increasing opportunities for methylation), has been shown to increase axon branching and growth in Atat1^−/− mice.³⁶ This suggests that the stoichiometry of acetylation and methylation along the microtubule polymer may play an as yet to be determined role in regulating neurite growth and arborization.

An important difference between the studies reported here and previous work on SETD2 methyltransferase activity was the preservation of methyltransferase: target stoichiometry. In Setd2^SRI/wt mice, SETD2 is expressed from two endogenous alleles, only one of which carries an SRI domain mutation; the catalytic SET domain activity is retained in both alleles. Remarkably, an SRI-domain mutation in only one Setd2 allele was sufficient to disrupt α-tubulin methylation on midbody microtubules even when a fully functional wild-type allele was present, demonstrating that SETD2 is haploinsufficient for α-tubulin methyltransferase activity. Although the basis for SETD2 haploinsufficiency for microtubule methylation is not known, one possibility is that the SRI-domain mutant acts as a dominant negative, competing with the wild-type allele for the CTT of α-tubulin. This has been seen in other settings, where for example the K36M oncohistone mutation binds and sequesters SETD2 to inhibit chromatin methylation by this methyltransferase in trans.³⁷ This appears unlikely to be the case for the SRI domain mutation of SETD2, however, as we found the R2510H mutation could not bind α-tubulin, and therefore would not obstruct binding of the wild-type allele to the CTT of α-tubulin.

The importance of the SRI domain for binding and methylation of α-tubulin may have been missed in previous studies for several reasons. First, in vitro methyltransferase assays typically utilize recombinant SET domain, as methyltransferases such as SETD2 are very large (>250 kDa), and recombinant full-length protein difficult to isolate. However, using truncated but functional SETD2 recombinant protein, we were able to observe a dependence on the SRI domain for α-tubulin methylation in vitro that was abrogated by several different SRI domain mutations. Second, because the isolated SRI domain has a 5-fold lower binding affinity for α-tubulin relative to the SET domain, previous studies using different SETD2 domains may not have observed this interaction. However, in the present study, increasing the amount of purified GST-fusion protein used in the pull-down assays allowed us to detect the SRI:α-tubulin interaction, and showed that the R2510H mutation disrupted binding to α-tubulin but not phosphorylated RNA polymerase II. The retention of SRI domain binding to phosphorylated CTD even in the presence of R2510H mutation could be explained by the multiple CTD repeats in RNA polymerase II (52 CTD repeats)³⁸ relative to the α-tubulin CTT (2–3 repeats), and the fact when active, multiple residues in the Pol II CTD become highly phosphorylated, providing the opportunity for a much stronger interaction than the two phosphomimetic repeats in the α-tubulin CTT.³⁹

De novo mono-allelic SETD2 mutations have been found in ASD patients and Sotos-like syndrome with autism phenotypes.^9,15–23 Given that the vast majority of ASD-linked mutations in chromatin remodellers occur in only one allele, pathogenicity is ascribed to either dominant-negative activity of the mutant allele or haploinsufficiency associated with monoallelic loss-of-function.^40–43 Both dominant-negative effects on the epigenome (e.g. BAF) and haploinsufficiency (e.g. SETD5) have been demonstrated for ASD-linked chromatin remodellers,^44,45 although the direct mechanism by which such defects contribute to ASD pathogenesis remains ill-defined.

As well-known regulators of gene expression, research on ASD-associated chromatin remodeller defects has focused on their impact on the epigenome and gene expression. However, there is little evidence that SETD2 is haploinsufficient for its activity on chromatin. For example, inactivation of both SETD2 alleles is required for loss of H3K36me3 on chromatin in cancers such as clear cell renal cell carcinoma.^6,8 We found when heterozygous, SETD2 is haploinsufficient for the α-TubK40me3 methyl mark on microtubules, and causes an anxiety-like phenotype in mice. Anxiety-like endophenotypes have also been seen in other rodent ASD models²⁹ using elevated plus maze and light-dark exploration, assays that also produced the significant differences between the behaviour of wild-type and SRI-domain mutant mice. Importantly, haploinsufficiency resulting in loss of microtubule methylation occurred in the absence of any global changes in H3K36me3 or gene expression in heterozygous Setd2^SRI/wt mice. Thus, these studies point to a potential cytoskeletal mechanism by which heterozygous defects in SETD2, and possibly other chromatin remodellers, could contribute to ASD and/or ASD-associated co-morbid symptoms such as anxiety.⁴⁶

Glossary

ASD

autism spectrum disorder

CTD

C-terminal domain

GST

glutathione-S-transferase; MEFs = mouse embryonic fibroblasts