Cell-specific exon methylation and CTCF binding in neurons regulate calcium ion channel splicing and function

Cell-specific alternative splicing modulates myriad cell functions and is disrupted in disease. The mechanisms governing alternative splicing are known for relatively few genes and typically focus on RNA splicing factors. In sensory neurons, cell-specific alternative splicing of the presynaptic CaV channel Cacna1b gene modulates opioid sensitivity. How this splicing is regulated is unknown. We find that cell and exon-specific DNA hypomethylation permits CTCF binding, the master regulator of mammalian chromatin structure, which, in turn, controls splicing in a DRG-derived cell line. In vivo, hypomethylation of an alternative exon specifically in nociceptors, likely permits CTCF binding and expression of CaV2.2 channel isoforms with increased opioid sensitivity in mice. Following nerve injury, exon methylation is increased, and splicing is disrupted. Our studies define the molecular mechanisms of cell-specific alternative splicing of a functionally validated exon in normal and disease states – and reveal a potential target for the treatment of chronic pain.

probe (Supplemental Fig. 2A). To establish specificity of binding, we incubated labelled e37a and e37b DNA probes (139 bp) alone; with recombinant CTCF; with CTCF plus CTCF monoclonal antibody; or with CTCF plus unlabeled probe (Fig. 2C, 2D). In the presence of 50 ng recombinant CTCF, ~30-40% of the e37a DNA probe migration was slowed, relative to no CTCF, as indicated by the presence of a second band which was shifted to higher molecular weights (Fig. 2C, 2D; compare lanes 1 and 2). Based on the following observations we conclude that CTCF binds with high specificity to Cacna1b e37a but not Cacna1b e37b: 1) CTCF bound e37a but not e37b DNA probes when studied using the same conditions (Fig. 2C, 2D; compare lanes 2 and 6); 2) the appearance of a third super-shifted band when e37a probe is incubated with both recombinant CTCF and a CTCF monoclonal antibody (Fig. 2C, 2D; lane 3); 3) total displacement of CTCF binding from the labeled e37a DNA probe on addition of unlabeled e37a DNA probe (Fig. 2C,2D;lane 4); and CTCF binding to e37a DNA is concentration dependent (Fig. 2B). We also tested the ability of various truncated e37a DNA probes to bind CTCF, but CTCF binding was reduced in all truncated constructs (Supplemental Fig. 2B). This is consistent with other reports that CTCF binding to DNA is strongly influenced by several factors including probe length (Fitzpatrick et al., 2007).

CTCF binds Cacna1b e37a locus in vivo and modifies e37a splicing in neuronal cells
Having shown that CTCF binds directly and specifically to Cacna1b e37a locus in vitro, but not e37b, we next tested if CTCF levels influence Cacna1b e37a splicing in neuronal cells.
We used the rat DRG/mouse neuroblastoma hybrid F11 cell line which expresses Cacna1b (Allen et al., 2017), to determine if CTCF binds Cacna1b e37a locus in a cellular context, and to test if CTCF levels influence Cacna1b e37a splicing. CTCF is localized to cell nuclei in F11 cells (Fig. 3A). We found that CTCF binds Cacna1b e37a locus in F11 cells by CTCF ChIP-qPCR (Fig. 3B). By contrast, CTCF does not bind e37b locus (Fig. 3B). Our findings in F11 cells, that CTCF associates with e37a but not e37b, are consistent with ChIP-seq data analyzed in 27 different human cell lines (see Fig. 1).
We next tested if CTCF levels in F11 cells influence the expression of Cacna1b e37a mRNAs. We conducted high-efficient qPCR using primer pairs for e37a, e37b, and e36 (constitutive exon), which were optimized and matched for specificity and accuracy, to quantify mRNAs (Fig. 3C). We either overexpressed CTCF, or knocked down CTCF in F11 cells (Fig.   3D, 3F) and then quantified levels of Cacna1b e37a mRNAs, relative to total Cacna1b mRNAs (e36; Fig. 3E, 3G). When overexpressed (Fig. 3D, 3E), recombinant CTCF tagged with GFP induced a ~2-fold increase in Cacna1b e37a levels within 2 days of cDNA transfection, as compared to cells expressing Gfp alone (Fig. 3E). siRNA targeting Ctcf in F11 cells (Fig. 3F,   3G) had the expected opposite effect; levels of e37a-containing Cacna1b mRNAs were ~30% reduced 3 days after transfecting Ctcf siRNA relative to control siRNA (Fig. 3G). Our findings show that CTCF promotes e37a inclusion in Cacna1b mRNAs, and that CTCF levels are rate limiting for Cacna1b e37a splicing in neuronal cells.
These experiments reinforce our hypothesis that CTCF is a factor in Cacna1b e37a recognition and that it promotes e37a inclusion during Cacna1b pre-mRNA splicing.
In the above experiments, global pharmacological manipulation of 5-mC gDNA reveals a negative correlation between 5-mC at the Cacna1b e37a locus and Cacna1b e37a mRNA levels in F11 cells. Our findings support the hypothesis that 5-mC within Cacna1b e37a locus occludes CTCF binding and impairs CTCF-mediated exon recognition during alternative pre-mRNA splicing.
Our results support a model in which Cacna1b e37a inclusion during alternative splicing is inhibited by DNMT3a and promoted by TET1 and TET2.

Cacna1b e37a is expressed in Trpv1-lineage sensory neurons of mouse dorsal root ganglia
In previous studies, we showed that Cacna1b e37a mRNAs are enriched in DRG of rat and, by single cell RT-PCR coupled to electrophysiological analyses, we found further enrichment in a subset of capsaicin-responsive nociceptors (Bell et al., 2004). Here we extend these studies and use a mouse strain expressing the TdTomato or Yfp reporter in Trpv1-lineage neurons for fluorescence-activated cell sorting (FACS). By endpoint RT-PCR, we found that The striking cell-specific expression pattern in Trpv1-lineage, but not in non Trpv1lineage neurons, offered a way to test our hypothesis in vivo: If 5-mC CpG of Cacna1b e37a locus is important for e37a recognition, then Trpv1-lineage neurons should have reduced 5-mC levels relative to non Trpv1-lineage DRG cells in Cacna1b e37a locus. We used targeted bisulfite sequencing and found that all 5 CpG sites in Cacna1b e37a locus in non Trpv1-lineage cells are methylated in the majority of sequences ( Fig 6B). By contrast, of 20 independent clones derived from Trpv1-lineage neurons, only one CpG site in Cacna1b e37a locus, in one sequence contained 5-mC ( Fig 6B).
Our results show that 5-mC CpGs of Cacna1b e37a locus are cell-specific -CpG sites in e37a are hypomethylated in Trpv1-lineage neurons relative to the close to fully methylated state of e37a locus in non Trpv1-lineage cells. These data provide strong independent support for our hypothesis, and they suggest that cell-specific control of local methylation regulates the level of

Peripheral nerve injury increases 5-mC within Cacna1b e37a locus in Trpv1-lineage neurons
We showed previously that Cacna1b e37a mRNA levels decrease in rat DRG following peripheral nerve injury, and this disruption in splicing is linked to reduced efficacy of morphine analgesia (Altier et al., 2007;Jiang et al., 2013). We therefore tested if nerve injury-induced disruption of Cacna1b e37a expression might reflect an underlying alteration in the local methylation state of Cacna1b e37a locus. We used the spared nerve injury (SNI) model of neuropathic pain which involves ligating and transecting two of three branches of the sciatic nerve (tibial and common peroneal) on one side of the animal, while leaving the sural nerve intact (Decosterd and Woolf, 2000). The SNI model results in stable, long-term hypersensitivity in the ipsilateral area of the hind paw which is innervated by the spared sural. We measured behavioral responses in hind paw, 5-mC and Cacna1b e37a levels in DRG, ipsilateral and contralateral to the surgery within the same animals. Hyperalgesia in ipsilateral hind paws developed within 2 days, and was sustained at least 8 days following SNI (Fig. 7A, 7B).
We found that Cacna1b e37a mRNA levels were reduced in Trpv1-lineage neurons isolated from L3-L4 DRG ipsilateral to the injured nerve in all four animals, relative to contralateral L3-L4 DRG (Fig. 7C). By contrast, there were no consistent changes in Cacna1b e37a mRNA levels in Trpv1-lineage neurons isolated from sham DRG compared to contralateral ( Fig 7C). We quantified 5-mC levels within Cacna1b e37a locus in Trpv1-lineage neurons and non Trpv1-lineage cells from ipsilateral L3-L5 DRG pooled from 3 sham and 3 SNI animals. We confirmed that Cacna1b e37a locus is hypomethylated in Trpv1-lineage neurons relative to the non Trpv1-lineage cell population in DRG (also see Fig. 6) but, after injury, 5-mC CpG at Cacna1b e37a locus is greater in ipsilateral Trpv1-lineage neurons (Fig. 7D). 5-mC CpG levels are unchanged in ipsilateral non Trpv1-lineage cell populations of DRG from SNI animals compared to sham (see Fig. 7D).
We conclude that 5-mC levels of CpG sites in Cacna1b e37a locus suppresses e37a inclusion during Cacna1b alternative pre-mRNA splicing in vivo in most neurons. Methylation of Cacna1b e37a locus occludes CTCF binding and accounts for the low expression of Cacna1b e37a mRNAs throughout the nervous system. In contrast, in Trpv1-lineage neurons, Cacna1b e37a locus is hypomethylated, permitting CTCF to bind and to promote Cacna1b e37a inclusion.
Differential local methylation dictates the cell-specific pattern of Cacna1b e37a expression particularly in Trpv1-lineage neurons of DRG. We showed that the increase in 5-mC at CpG sites in Cacna1b e37a locus following peripheral nerve injury, that induces long-lasting hyperalgesia, is associated with reduced Cacna1b e37a expression in Trpv1-lineage neurons.

DISCUSSION
Cell-specific alternative splicing is essential for normal cell function, and ~30% of all disease-causing mutations are related to RNA splicing (Montes et al., 2019). Our studies reveal the cell-specific mechanisms that govern alternative splicing of a critical synaptic calcium ion channel gene and they shed light on chronic pathology that follows nerve injury.
Cell-specific Cacna1b e37a splicing in Trpv1-lineage nociceptors underlies the unique sensitivity of voltage-gated CaV2.2 calcium channels to GPCRs important for defining their properties in vivo. Contrary to our initial expectations, we found that a ubiquitous DNA binding protein, CTCF promotes e37a inclusion in Cacna1b mRNAs in a highly cell-specific pattern.
This unexpected cell-specific action of CTCF is conferred by cell-specific hypomethylation within the CTCF binding motif in Cacna1b e37a locus. E37a locus demethylation is disrupted following nerve injury resulting in increased methylation levels and impaired Cacna1b e37a splicing in Trpv1-lineage nociceptors. Injury-induced increased methylation levels of Cacna1b e37a locus likely underlies some of the chronic pathophysiology associated with peripheral nerve injury.

CTCF and methylation regulate cell-specific alternative splicing
To date, the action of CTCF on alternative splicing has only been studied in immune cells CTCF is proposed to enhance exon recognition by slowing the elongation rate of Pol II (Shukla et al., 2011) and this co-transcription-splicing mechanism has been proposed to account for the influence of epigenetic modifications on pre mRNA splicing in neurons (Ding et al., 2017;Schor et al., 2013). CTCF binds ~60,000 sites on average on mammalian chromosomes (Maurano et al., 2015), most of which (~70%) reside in intergenic or near to promoter regions important for regulating chromatin structure and gene expression (Barski et al., 2007;Kim et al., 2007;Ong and Corces, 2014). CTCF binding to intragenic sites is reported to be relatively This study reveals a novel mechanism of cell-specific alternative splicing in neurons.
Remarkably, the ubiquitous DNA binding protein, CTCF, is critical for cell-specific inclusion of a mutually exclusive exon during Cacna1b pre mRNA splicing in neurons. The specificity of exon inclusion arises from a striking cell-specific exon methylation pattern, providing an exciting path for understanding cell-specific CTCF action and potentially identifying a major mechanism in neuronal alternative splicing.

Author contributions
EJLS & DL wrote the manuscript, designed experiments and performed analyses. EJLS performed all experiments.

GRAPHICAL ABSTRACT
Cell-specific epigenetic modifications in a synaptic calcium ion channel gene controls cellspecific splicing in normal and neuropathic pain. In naïve animals, in most neurons, Cacna1b e37a locus is hipermethylated (5-mC) and CTCF does not bind this locus. During splicing, e37a is skipped and Cacna1b mRNAs include e37b. In contrast, in Trpv1-lineage neurons, Cacna1b e37a locus is hypomethylated and is permissive for CTCF binding. CTCF promotes e37a inclusion and both Cacna1b e37a and e37b mRNAs are expressed. E37a confers strong sensitivity to the CaV2.2 channel to inhibition by -opioid receptors (OR). Morphine is more effective at inhibiting e37a-containing CaV2.2 channels. After peripheral nerve injury that results in pathological pain, methylation level of Cacna1b e37a locus is increased, CTCF binding is impaired, and Cacna1b e37a mRNA levels are decreased. This disrupted splicing pattern is associated with reduced efficacy of morphine in vivo.    (F) Western blot from F11 cells expressing control siRNA (Con) or Ctcf siRNA 3 days after transfection. In control cells, anti-CTCF identifies endogenous CTCF at ~135 kDa, and reduced CTCF levels in cells expressing Ctcf siRNA. Transfer membrane was cut at ~ 75 kDa (dotted red line) and the lower part treated with anti-GAPDH to measure GAPDH levels for protein expression and loading reference. CTCF protein levels relative to GAPDH (t-test P value = 0.0008 for Con vs Ctcf siRNA) (n = 3 per condition).
Biological replicates represent independent cell cultures, treatment and transfections.  (B) Thermal sensitivity assessment of mice prior to, and 8 days after surgery. Plantar, hind paw withdrawal latencies measured in response to a radiant heat source both ipsilateral and contralateral to the site of surgery in sham and SNI mice. SNI mice developed hyperalgesia in ipsilateral but not contralateral hind paws (two-way ANOVA P value = 0.0173 for side to the surgery factor, and P value = <0.0001 for time factor). By contrast, sham surgery had no effect on paw withdrawal latencies in response to heat source (two-way ANOVA P value = 0.2136 for side to the surgery factor, and P value = 0.7780 for time factor) (n = 4 mice per condition). (C) qRT-PCR of e37a relative to e36 in Trpv1-lineage neurons. Dotted lines connect contra and ipsilateral L3 and L4 DRG values from the same animals (Paired t-test P values for contra vs ipsilateral DRG = 0.3671 from sham, and P value = 0.0384 from SNI mice) (n = 4 mice per condition). truncated, and 5' truncated DNA probes containing e37a either alone (lanes 1, 4, 7, 10, 13, 16, 19, 22 and 25) or preincubated with 50 ng recombinant CTCF (lanes 2, 5, 8, 11, 14, 17, 20, 23 and 26), or CTCF plus 1000-fold excess unlabeled e37a competitor (lanes 3, 6, 9, 12, 15, 18, 21, 24 and 27).