Identification of novel targets of RBM5 in the healthy and injured brain.

Abstract

The tumor suppressor RNA-binding motif 5 (RBM5) regulates the expression levels and cassette exon-definition (i.e. splicing) of a select set of mRNAs in a tissue-specific manner. Most RBM5-regulated targets were identified in oncological investigations and frequently involve genes which mediate apoptotic cell death. Little is known about the role of RBM5 in the brain. Also, it is unclear if a brain injury may be required to detect RBM5 mediated effects on pro-apoptotic genes due to their low expression levels in the healthy adult CNS at baseline. Conditional/floxed (brain-specific) gene deleter mice were generated to elucidate CNS-specific RBM5 mRNA targets. Male/female mice were subjected to a severe controlled cortical impact (CCI) traumatic brain injury (TBI) in order to increase the background expression of pro-death mRNAs and facilitate testing of the hypothesis that RBM5 inhibition decreases post-injury upregulation of caspases/FAS in the CNS. As expected, a CCI increased caspases/FAS mRNA in the injured cortex. RBM5 KO did not affect their levels or splicing. Surprisingly, KO increased the mRNA levels of novel targets including casein kinase 2 alpha prime interacting protein (Csnka2ip/CKT2) – a gene not thought to be expressed in the brain, contrary to findings here. Twenty-two unique splicing events were also detected in KOs including increased block-inclusion of cassette exons 20–22 in regulating synaptic membrane exocytosis 2 (Rims2). In conclusion, here we used genome-wide transcriptomic analysis on healthy and injured RBM5 KO mouse brain tissue to elucidate the first known gene targets of this enigmatic RBP in this CNS.

Graphical Abstract

Introduction

RNA-binding motif 5 (RBM5) is a member of the RNA recognition motif (RRM) domain-containing family of RNA-binding proteins (RBPs) (Sutherland et al., 2005). Pioneering studies in cancer first demonstrated that it potently activates pro-death pathways in cells (Oh et al., 2006; Oh et al., 2002). The molecular mechanisms that mediate RBM5-dependent tumor suppression involve increased expression of pro-death genes and pre-mRNA splicing events that favor the translation of protein variants with greater toxicity. For instance, RBM5-dependent modulation of caspase-2 splicing in HeLa cells increased the translation of the long-variant (Casp2L), which is a potent inducer of apoptosis (Fushimi et al., 2008). Similarly, RBM5-dependent modulation of FAS splicing alters the activation of extracellular death receptors in HeLa cells (Bonnal et al., 2008). In addition, endogenous loss of RBM5, or its experimental knockdown, decreases total mRNA levels of caspases, which in turn promotes tumor survival and metastasis of cancers (Mourtada-Maarabouni et al., 2006; Oh et al., 2010). The extent that RBM5 augments the expression of canonical cell death genes (e.g. defined by its targets/effects in cancer) or rather if it modulates non-canonical pathways in normal tissues has not been determined for most organs.

The brain is enriched in RBM5 vs. other tissues yet its role in the CNS is poorly understood (O’Bryan et al., 2013). A total of four articles to date have investigated RBM5 functions in neurobiologically-relevant models (Jackson et al., 2015; Jackson et al., 2018b; Jackson et al., 2017; Zhang et al., 2015). Two studies used primary rat cortical neurons and two studies used immortalized/transformed neuronal cell lines. Three directly confirmed that RBM5 has neurotoxic effects. Collectively, they show that RBM5 promotes neuronal death in models of: (a) mechanical stretch-injury to induce trauma in primary neurons in vitro (Jackson et al., 2018b), (b) hydrogen peroxide injury to induce necrosis in rat neuronal pheochromocytoma (PC12) cells (Zhang et al., 2015), and (c) staurosporine injury to induce apoptosis in human differentiated neuronal SHSY5Y cells (Jackson et al., 2015). The gene programs regulated by RBM5 that led to enhanced neuronal vulnerability in these systems were not elucidated. Also, two of the articles included observational studies on the injured rodent CNS and found that endogenous RBM5 protein is increased in the hippocampus after experimental traumatic brain injury (TBI) and in spinal tissue after traumatic spinal cord injury (SCI) (Jackson et al., 2015; Zhang et al., 2015). Thus, RBM5 is increased after CNS damage in vivo, and it exacerbates neuronal injury in vitro, but the gene pathways mediating the latter effects are unknown.

The remaining (fourth) study investigated RBM5-dependent changes in the global transcriptome in healthy/uninjured primary rat cortical neurons pre-treated with RBM5 targeting RNAi (Jackson et al., 2017). Gene inhibition resulted in increased levels of the small GTPase Rab4a, which in turn increased endocytosis-mediated intracellular uptake of the monomeric serotonin transporter. No alterations in cell death related signaling pathways in RBM5 knockdown neurons were found; however, the omission of a post-injury comparison group may have precluded the detection of canonical cell death genes because their baseline expression is low in healthy neurons and is increased after damage (e.g. caspases). Thus, a paucity of data exists to address if RBM5-regulated pro-death genes and/or splicing events, which have mostly been observed in cancer cells, might manifest in the CNS only after an injury. Also, it has not been tested if RBM5 inhibition in vivo alters the expression of pro-death genes in the brain (healthy or injured), or if it alters the regulation of novel targets (e.g. Rab4a) recently identified in vitro (Jackson et al., 2017).

Here we addressed these knowledge-gaps by: (1) creating a novel transgenic conditional brain-targeted RBM5 knockout (KO) mouse, (2) subjecting it to an experimental severe TBI to trigger CNS damage and stimulate increased expression of cell death genes, and (3) performing global transcriptomic analysis in WTs vs. KOs to test the hypothesis that RBM5 inhibition robustly decreases the expression of pro-death genes in the brain in vivo in an injury-dependent manner. Novel conditional KOs were created for these studies because homozygous total body KO of RBM5 is postnatally lethal (Jamsai et al., 2017). As expected, TBI increased canonical pro-death RBM5 gene targets in the brain (e.g. caspases and FAS) but contrary to our hypothesis were unaffected by KO. We also found that Rab4a was not increased in the mouse cortex in vivo in KOs, which was contrary to expectations based on prior reports in rat cortical neurons in vitro (Jackson et al., 2017). Instead, we identified novel RBM5 targets. Genes in the cortex in which total mRNA levels were altered in KOs vs. WTs but in which splicing was unaffected included: casein kinase 2 alpha prime interacting protein (Csnka2ip/CKT2), Gm756, Serpina3n, and glial fibrillary acidic protein (GFAP). Genes in which cassette exon splicing was altered but in which total expression in the cortex was unaffected included: synaptic membrane exocytosis 2 (Rims2), solute carrier family 20 member 1 (Slc20a1), myocyte enhancer factor 2a (Mef2a), and N-myc downstream-regulated gene 2 protein (Ndrg2). In summary, here we report the first known targets of RBM5 in the brain in vivo.

Experimental Procedures

Study Design:

The goal of this study was to determine if RBM5 regulates, in the brain, cell death genes that are well-known targets in cancer. We created novel conditional RBM5 KO mice due to the inability to investigate gene expression changes in adult mice using previously established total body RBM5 KOs. Novel transgenics were crossed with Nestin-CRE mice to produce offspring (F2 generation) that were homozygous for RBM5 gene deletion in the brain. Mice were genotyped to confirm both CRE expression and RBM5 allele status. A breeding scheme was employed that ensured all group comparisons in downstream studies (WT, Het, KO sham vs. injury) harbored a single copy of the CRE transgene. A controlled cortical impact (CCI)-TBI was used to increase the expression of pro-death genes in the cortex and to modify the transcriptome in a manner which we hypothesized would optimize the detection of RBM5-mediated effects on caspases/FAS levels should they exist in the brain. A total of seven mice per group were targeted for RNA studies (6 groups in total), which is consistent with numbers used in prior studies to detect gene expression changes by microarray in transgenic mice (McClard et al., 2018; McGill et al., 2018; Medrano and Naya, 2017). A few genes were selected for further validation by quantitative Nanostring and for in situ hybridization (ISH) experiments. Acute cardiovascular physiology and weight were characterized in mice that survived to the 48h study endpoint and a total of ten animals per group were targeted. Experiments were designed to limit gender bias by including both male and female subjects in sham and CCI treatment groups for combined gene analysis. Allocation was randomized by: (a) sham vs. injury and (b) the age/sex that mice entered the CCI study based on availability. Acute physiological data and weight were collected and analyzed for animals that survived to the study conclusion (i.e. the 48h tissue collection timepoint). Physiological data was excluded from analysis for one male mouse in the WT-CCI group and for one male mouse in the KO-CCI group because they died prior to the predefined study endpoint (i.e. out of protocol). No sample data was excluded from microarray analysis. Naïve mice were used for ISH studies.

Animals:

All animal work was approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh. Protocols follow recommendations established by the American Medical Veterinary Association Guideline for Euthanasia. Studies were designed to minimize pain and suffering, and to use the minimum number of animals required to achieve significance. Animals were granted ab libitum accesses to food and water. All animals were maintained on 12h light/dark cycle. Animal surveillance of housing facilities and veterinary care are supervised by Division of Laboratory Animal Resources (DLAR) staff. B6.Cg-Tg(Nes-cre)1Kln/J mice were purchased from JAX (The Jackson Laboratory; Stock No: 003771). Nestin-CRE mice were maintained in the hemizygous (single copy) state for LoxP recombination studies and for colony maintenance. Novel transgenic conditional RBM5 KOs were created by goGermline technology (Ozgene; Bentley DC, WA, Australia). In brief, C57BL/6 ES cells were electroporated with a targeting vector which included LoxP-flanked exon 3 of the RBM5 gene and a FRT-flanked neomycin (Neo) cassette. Transgene insertion was confirmed by antibiotic resistance and southern blotting. ES mutant clones were injected into donor blastocysts and implanted into psudopregnant recipient female mice. Chimeric mice were identified by coat color and bred to germline transmission. Heterozygous C57BL/6 Rbm5^tm1Ozg mice were crossed with FLP deleters to remove the Neo cassette. Male founders were crossed with C57BL/6 female wild-type (WT) mice purchased from JAX laboratories. Heterozygous Rbm5^tm1Ozg mice were crossed to generate homozygous Rbm5^tm1Ozg mice. F1 colonies were generated by crossing homozygous Rbm5^tm1Ozg mice with hemizygous B6.Cg-Tg(Nes-cre)1Kln/J mice. F1 heterozygous Rbm5^tm1Ozg mice without CRE were breed with F1 heterozygous Rbm5^tm1Ozg mice with CRE, to generate F2 littermates for experiments.

Genotyping:

Tail snips were collected from pre-weaning (16–18d) pups and DNA was isolated for genotyping using Promega Wizard SV Genomic Purification System (Promega; Madison, WI, USA), per the manufacturer’s instructions. PCR reactions were run on a Biorad T100 Thermalcycler (BioRad; Hercules, CA, USA). Samples were loaded onto 3% agarose gels with Syber Safe Stain (ThermoFisher Scientific; Waltham, MA, USA). The following primer pair was used to screen pups for the absence or presence of the Cre transgene: GCGGTCTGGCAGTAAAAACTATC (forward) and GTGAAACAGCATTGCTGTCACTT (reverse). The temperature protocol steps were: 94°C/3min, 94°C/30sec, 51.7°C/1min, 72°C/1min repeated for 35 cycles, and 72°C/2min. The presence of the CRE transgene resulted in a ~100bp PCR product. The following primer pair was used to confirm Neo elimination after breeding with FLP deleter mice: ATACGCTTGATCCGGCTACCTG (forward) and TGCAGCCTATCTTCTATAAGGG (reverse). The following primer pair was used to genotype WT, conditional Het, and conditional KO Rbm5^tm1Ozg mice, prior to crossing with CRE mice: GCGGGACTCAGATTACAAAAGATC (forward) and TGCAGCCTATCTTCTATAAGGG (reverse). The temperature protocol steps were: 98°C/3min, 98°C/10sec, 62°C/30sec, 72°C/2min repeated for 29 cycles, and 72°C/10min. Samples were loaded onto 0.8% agarose gels with Syber Safe Stain. WT cDNA produced a single ~317pb product, whereas homozygous conditional Rbm5^tm1Ozg KO cDNA produced a single ~420bp PCR product.

Controlled Cortical Impact (CCI):

Male/female Nestin-CRE expressing RBM5 WTs, Hets, and KOs were used at 12–15 wks of age. On the day of surgery mice were anesthetized with 4% isoflurane (via nosecone) and decreased to 2% for maintenance anesthesia, in Nitrous Oxide (N₂O)/%O₂ (2:1). MAP and heart rate HR were continuously monitored via a femoral arterial catheter that was surgically inserted. Shams received surgical preparations but did not receive a craniotomy. Mice randomized to TBI were given a CCI as described by our group (Jackson et al., 2018a; Jackson et al., 2015). Our standard insult level (6m/s, 1.2mm depth) produces a moderate-severe TBI, resulting in a substantial ipsilateral cortical lesion post-injury (Jackson et al., 2018a). Tissues for total RNA were harvested 48h post-injury. A total of 42 cortices were harvested for total RNA (n=7/group) as follows: WT-Sham (n=3 male and n=4 female), WT-CCI (n=3 male and n=4 female), Het-Sham (n=3 male and n=4 female), Het-CCI (n=3 male and n=4 female), KO-Sham (n=4 male and n=3 female), KO-CCI (n=3 male and n=4 female). A total of 28 arrays were purchased to analyze the transcriptome in sham/injured WTs vs. KOs. In addition, WT, Het, and KO samples were analyzed in confirmatory Nanostring studies on selected genes (n=7/group).

Blood Chemistry and Weight:

A total of n=10/group were used for acute physiological analysis. Blood (~60μL) was sampled from the arterial catheter at baseline (2% isoflurane in N₂O/%O₂, 2:1), 5 min prior to a CCI (2% isoflurane in room air), 5 min post-CCI (2% isoflurane in room air), and 60 min post-CCI (2% isoflurane in room air). Blood was immediately analyzed via an ABL90 Radiometer Blood Gas Machine (Radiometer Medical ApS; Copenhagen, Denmark), as reported previously by our group in the CCI mouse model (Jackson et al., 2018a). Body weight was recorded prior to injury.

Western blot:

In brief, sham male (n=2 WT and n=2 KO) and sham female (n=2 WT and n=2 KO) hippocampal extracts were prepared in radioimmunoprecipitation assay (RIPA) buffer supplemented with EDTA, protease inhibitors, and phosphatase inhibitors (ThermoFisher Scientific). Protein concentration analysis was done by the BCA method (ThermoFisher Scientific). ~20μg/well total hippocampal extracts were loaded onto a 4–15% TGX^™ pre-cast SDS-PAGE gel (BioRad). Linearized proteins were transferred to a PVDF membrane. Total protein analysis was done using reversible Swift Membrane Stain^™ (ThermoFisher Scientific) and images collected via a 600dpi scanner to confirm equal protein loading/transfer. The membrane was blocked and then incubated overnight at 4°C with 1:1000 anti-RBM5 antibody (Proteintech; Chicago, IL, USA - Cat# 19930-1-AP, Lot#49598) diluted in tris-buffered saline tween-20 (TBST) and 7.5% milk. The following day, the membrane was washed, incubated 2h with a secondary antibody, washed again, incubated in ECL-2 detection substrate (Thermo Fisher Scientific), and imaged on film in a dark room. Images were compiled in Photoshop (Adobe; San Jose, CA, USA).

RNA Isolation:

Cortical brain tissues were homogenized in Trizol® in a BeadMill® (ThermoFisher Scientific) set to shake for 60 sec at 3 m/sec with 2.8 mM ceramic beads. A 1:5 volume of chloroform to Trizol was added and vigorously shaken. The mixture was incubated 15 mins at room temperature. Samples were centrifugated 15 mins at 12,000g to induce phase separation. The aqueous phase was isolated and placed in a new microcentrifuge tube. RNA was precipitated by addition of a 1:1 volume of isopropanol to sample (equal to the aqueous extraction volume). The RNA pellet was washed with 75% ethanol, air dried, and resuspended in nuclease-free water. The OD260, 260/280, and 260/230 ratio were measured by Nanodrop (ThermoFisher Scientific) to determine RNA concentration and purity. The integrity of RNA was confirmed by Tape Station® 4200 (Agilent; Santa Clara, CA, USA). The eRIN was >8.4 for all samples.

Microarray Analysis:

Complementary DNA (cDNA) was prepared using the WT Plus reagent kit (ThermoFisher Scientific) and used according to manufacturer instructions. Reverse transcription was done with 100ng total RNA and dNTP-T7 random primers. Second strand synthesis and in vitro transcription was done to amplify cRNA. A second reverse transcription reaction produced cDNA which was fragmented and end-labeled with biotin for hybridization onto Clariom D (MTA-1_0) arrays (5μg ds cDNA per array). After overnight (~16h) hybridization at 45°C, with rotational mixing at 60 rpm, arrays were processed on a GeneChip® 450 Fluidics Station (ThermoFisher Scientific) using manufacturer specified protocols. To remove unbound sample, arrays were washed with non-stringent wash buffer A. The GeneChips® were stained 10 mins in Stain cocktail 1. Excess stain was removed with wash buffer A. Signal amplification was achieved by 10 mins incubation with Stain Cocktail 2 followed by 10 min incubation with Stain Cocktail 1. The chips were washed with high stringency Buffer B and prior to removal from the fluidics station, loaded with Array Holding Buffer, then scanned using the GeneArray® 3000 scanner (ThermoFisher Scientific). Data quality-control and statistics were performed with Transcriptome Analysis Console (TAC) 4.0. Affymetrix/ThermoFisher Expression Console (EC) software was used for RMA normalization prior to DEG/splicing analysis. Microarray data (chp. and CEL. Files) were uploaded to the NCBI Gene Expression Omnibus database. All individual arrays passed labeling control-threshold QC, hybridization control-threshold QC, and Pos. vs. Neg. AUC Threshold QC.

Quantitative mRNA Analysis:

Custom NanoString nCounter gene expression codesets were developed. Brain tissue RNA was processed on the NanoString nCounter Analysis System according to the manufacturer’s protocol (NanoString Technologies; Seattle, WA, USA). Briefly, 100ng of RNA was hybridized to the reporter probes and capture probes in a thermal cycler for 16 hours at 65°C. Washing and cartridge immobilization was performed on the nCounter PrepStation (NanoString Technologies) and the cartridge was scanned at 555 fields of view on the nCounter Digital Analyzer (NanoString Technologies). The resulting RCC files containing raw target counts were imported into the NanoString nSolver Analysis Software v4.0 (NanoString Technologies) for downstream normalization and analysis.

High-Fidelity Single-Molecule ISH:

Naïve adult male WT and KO mice were used for RNA analysis on brain slices. FFPE brain tissue sections were prepared as described by our group and with minor modifications per manufacturer’s recommendations for ISH (Jackson et al., 2018a). For each ISH experiment, WT vs. KO brains were placed on the same glass slide to ensure equivalent manipulations across genotypes. RNA in situ hybridization experiments were performed using RNAscope® (Advanced Cell Diagnostics; Newark, CA, USA) as described by others (Wang et al., 2012). This technique enhances the rigor of ISH by robustly increasing the signal-to-noise ratio to detect individual mRNA molecules with high specificity by utilizing contiguous binding pairs of short target probes, which in the correct order, permits a subsequent preamplifier probe to bind and results in a signal (amplifier) (Wang et al., 2012). Paired double-Z oligonucleotide probes were designed against target RNA using custom software (Advanced Cell Diagnostics). The following probes were used: Mm-Csnka2ip, cat no. 824601, ENSMUST00000089279.4, 20zz pair, nt 134–1271; Mm-Rims2-O1, cat no. 824591, ENSMUST00000042917.9, 13zz pair, nt 2055–2842; BA-Mm-Rims2-E22E23, cat no. 824581, NM_001256382.1, 1zz pair, nt 4072–4113; BA-Mm-Rims2-E19E20, cat no. 824571, NM_001256382.1, 1zz pair, nt 3554–3598. The RNAscope® and BaseScope^™ Reagent Kit (Advanced Cell Diagnostics) was used according to the manufacturer’s instructions. High RNA integrity was confirmed for each sample with a probe specific to the housekeeping gene cyclophilin B (PPIB). Negative control background staining was evaluated using a probe specific to the bacillus subtilis dihydrodipicolinate reductase gene (dapB). Brain slices were counterstained with Hematoxylin. Total Csnka2ip and Rims2 mRNAs were detected with the 2.5 HD Red Assay Kit (Advanced Cell Diagnostics). Rims2 splice variant probes were detected with the BaseScope RED assay kit (Advanced Cell Diagnostics). The technician performing ISH staining was blind to genotype. The technician performing Image-J analysis was blind to genotype and to the target under investigation. Images were scanned at 4X and 20X. Randomly selected fields, within comparative brain sub-regions, were collected for enhanced viewing of 20X staining in ISH figures. Staining intensity was measured in Image-J. In brief, images were first deconvoluted to blue, green, and red signals. Pixels for the red probe-staining signal (color 2) were measured for each target. Thresholding was adjusted for each target and then consistently applied across all brain regions and genotype for a given probe. Three equivalent size boxes were drawn at random in each 20X image (i.e. three ROIs). Box sizes were identical across all images. The percent area containing signal within each box was calculated (unitless value) and the results averaged for each image. The fold-change was calculated by dividing the mean pixel content computed for each brain region/target in KOs by the mean pixel content in WTs. A decrease or increase in signal staining was defined as a fold-change < −2 or > 2, respectively.

Statistical Analysis:

Microarray: The filter criteria for significance for differentially expressed genes (DEGs) was: Fold-change > 2 or < −2 AND FDR-corrected p-value <0.05. The filter criteria for significance for splicing events was: (a) gene target expressed in both conditions, (b) exon/PSR within gene target expressed in at least one condition, (c) splicing Index > 2 or < −2, (d) FDR-corrected exon p-value <0.05, and (e) exon event score >0.2. A probe set (gene/exon) was considered expressed if ≥ 50% samples have detected above background (DABG) levels below threshold (0.05). Data were analyzed by ANOVA and FDR adjusted p-values were based on Benjamini-Hochberg Step-Up FDR-controlling Procedure. Exon event scores were calculated based on EventPointer (Romero et al., 2016). Nanostring: Significance was determined by a two-tailed t-test of the log-transformed normalized data within the nSolver Software Package. All samples passed QC parameters. Expression data was normalized to the geometric mean of: (a) spiked positive controls to adjust for technical variations across lanes and cartridges, and (b) to the geometric mean of five housekeeping genes (GAPDH, GUSB, LDHA, Olfr776, PGK1) for codeset content normalization. The reference gene list was collated based on a literature review of housekeeping targets that yielded best results for quantitative gene expression analysis studies and based on empirical confirmation of stably expressed genes in our microarray studies (Boda et al., 2009; Kang et al., 2018). Normalized counts of gene targets (n=7/group) were imputed into Prism and represented as box plots which show the minimum value, maximum value, interquartile range (IQR), and median (GraphPad Software Inc; La Jolla, CA, USA). Acute Physiology: MAP, HR, and blood chemistry data were analyzed by 3-Way-ANOVA. Body weight was analyzed by 2-Way ANOVA and using a Tukey’s Multiple Comparison post-hoc test. Data were significant at two tailed p<0.05.

Accession Codes:

The accession number for raw microarray data submitted to NCBI GEO is GSE128543.

Results

Validation of Novel Conditional RBM5 KO Mice.

A conditional gene targeting strategy was designed (Fig. 1A) and KO was confirmed by PCR genotyping (Fig. 1B) and by Western blot analysis in mouse hippocampal tissue extracts (Fig. 1C). RBM5 migrated at ~120 kDa on SDS-PAGE, consistent with our prior reports in primary rat cortical neurons (Jackson et al., 2018b; Jackson et al., 2017). Homozygous brain-specific RBM5 KOs were viable, well-appearing at birth, and showed no overt signs of health and/or behavioral abnormalities in adulthood. Mendelian ratios were calculated from F2 litters bred between August 2017 to January 2019 for mice that expressed a single copy of the CRE transgene (Fig. 1D). WTs (~28%) and Hets (~50%) followed the expected ratios. In contrast, the number of brain-specific KOs was lower than expected (~15%). Decreased KO numbers was not due to increased postnatal mortality, as was previously reported to occur in total body RBM5 KOs (Jamsai et al., 2017). Furthermore, the phenomenon was independent of sex - 13 male KOs vs. 10 females KOs survived into adulthood. In the course of studies two adult KOs died of unknown causes (1 Male and 1 Female; both >2mo of age), and a single male KO was euthanized postnatally because it was a runt. The mechanism(s) mediating a decrease in the number of KOs remains to be elucidated.

Fig. 1: Design Strategy of RBM5 KO Mice.

(A) Illustration of the design strategy. Exon 3 in the RBM5 gene is flanked with loxP sites. The neomycin cassette was removed prior to initiating studies by crossing novel transgenics with FLP deleters. Subsequent crossing to Nestin-CRE mice results in recombination mediated deletion of exon 3 in RBM5 and inhibition of a protein product. (B) PCR genotyping strategy to confirm homozygous conditional KOs. (C) Western blot of hippocampal extracts confirms the ~120 kDa RBM5 protein is present in WT-CRE mice but absent in homozygous KO-CRE mice. Total protein stain indicates equal loading across samples. (D) Mendelian ratios for brain-specific KOs in F2 litters is lower than expected.

Next, we measured mean arterial blood pressure (MAP), heart rate (HR) and hematological parameters to characterize baseline cardiovascular health in RBM5 KO shams and in injured mice. We focused on characterizing global cardiovascular indices here because they profoundly affect brain tissue survival and thus gene expression changes in the acute period after a TBI (Kroppenstedt et al., 1999). There were no differences in MAP (Fig. 2A) or HR (Fig. 2B) in WTs vs. KOs at baseline or after a CCI. KOs weighed significantly less vs. WTs (~3g). This effect was independent of sex and not seen in heterozygotes (Fig. 2C). There were no genotype differences in arterial pCO₂, or pO₂, or SaO₂, hematocrit, hemoglobin, HCO₃, potassium, or lactate (Fig. 2D). In contrast, blood sodium, chloride, glucose and osmolality were significantly (albeit modestly) increased in KOs vs. WTs (Fig. 2D). In addition, KOs had a slight decrease in blood pH. The differences were minor and all blood chemistry values in WTs and KOs were within the normal range for anesthetized mice. Heterozygotes also had significant (albeit minor) differences in blood sodium, chloride, glucose and osmolality vs. WTs but were detected as an interaction term and were within the normal range (Table S1). In summary, global measures of cardiovascular function and blood chemistry were comparable between WT vs. KOs before and immediately after a brain injury.

Fig. 2: Characterization of Acute Cardiovascular Parameters and Blood Chemistry in RBM5 KO Mice.

(A) KO does not alter mean arterial blood pressure (MAP) at baseline or after CCI-TBI (n=10/group). (B) KO does not alter heart rate (HR) at baseline or after CCI-TBI. (C) Body weight is decreased in male and female KOs but not in heterozygous mice. Red line depicts the mean (n=10/group). (D) Blood Chemistry and gas analytes in isoflurane anesthetized WT vs. KOs (n=10/group) at baseline, 5 min prior to CCI, 5 min after a CCI, and 60 min after a CCI. Graphs show mean + SEM. Black circles/squares in plots indicate WT mice. Red circles/squares in plots indicate KO mice. Data were significant at p<0.05. Blood chemistry variables that are significantly altered by genotype are highlighted in red text. Significant interaction terms are indicated. (I) = Injury and (S) = Sham. Arterial pH (pHA), arterial CO2 (PaCO2), arterial O2 (PaO2), arterial oxygen saturation (SaO2), hematocrit (HCt), hemoglobin (Hb), base excess (BE), bicarbonate (HCO3), sodium (Na), potassium (K), chloride (Cl), calcium (CA), osmolarity (osm).

RBM5 KO does not Decrease TBI-Induced Upregulation of Cell Death Genes in the Injured Cortex.

In WTs, CCI injury altered the expression of 154 coding-gene mRNAs (145 increased and 9 decreased) (Fig. 3A). Moreover, 203 multiple-complex genes were significantly altered by injury (194 increased and 9 decreased). Gene ontology (GO) analysis in PANTHER14.1 on the combined coding/multi-complex genes in WT mice mapped 166 genes across 57 distinct areas (Ashburner et al., 2000; Carbon et al., 2019; Mi et al., 2019) (Fig. 3B and Fig. S1). Injury-related pathways of note are highlighted in Fig. 3B. Microarray analysis confirmed a significant increase in caspase-8 expression levels in WT mice (CCI vs. sham; Table S2). In addition, caspase-4 and FAS levels trended to increase after a CCI (Table S2). Nanostring quantification confirmed a significant increase in the mRNA levels of caspase-4 (Fig. 3D), caspase-8 (Fig. 3G), and FAS (Fig. 3H) in the cortex in injured WTs vs. shams. Contrary to our hypothesis, RBM5 gene KO did not decrease the levels of caspases/FAS after a CCI (Fig. 4A–F and Table S2). In summary, global transcriptomic analysis suggests that RBM5 does not regulate mRNA expression of canonical cell death gene targets in the acute period after a TBI.

Fig. 3: A CCI-TBI Increases the Expression of Pro-Death Gene Targets in the Contused Cortex in WT Mice.

(A) The dendrogram shows hierarchical clustering in gene expression changes in WT shams vs. CCI-injured mice. The table below indicates the number of genes and the directionality of their change 48h post-injury. (B) Gene ontology of altered coding/multi-complex genes in injured WT mice. Areas of enrichment are denoted. (C-H) Increased levels of selected cell death gene targets in WT mice were confirmed by Nanostring quantification. Total mRNA levels of: (D) caspase-4, (G) caspase-8, and (H) the FAS receptor were significantly increased in the injured cortex. In contrast, levels of: (C) caspase-1, (D) caspase-6, and (F) caspase-7 did not significantly differ at the 48h endpoint in WT shams vs. CCI-injured mice. Box plots show the minimum, maximum, IQR, and median for the samples (n=7/group). Data were analyzed by a t-test and significant at p<0.05. Clear boxes indicate WT-Shams and grey boxes indicate WT-CCI.

Fig. 4: RBM5 KO does not Affect the Amplitude of Pro-death Gene Expression in the Contused Cortex after a CCI-TBI.

(A) Nanostring quantification of selected cell death genes in CCI-WT vs. CCI-KO mice. Total mRNA levels of: (A) caspase-1, (B) caspase-4, (C) caspase-6, (D) caspase-7, (E) caspase-8, and (F) FAS receptor were not significantly different by genotype in the contused cortex by 48h post-injury. Box plots show the minimum, maximum, IQR, and median for the samples (n=7/group). Data were analyzed by a t-test and significant at p<0.05. Clear boxes indicate WT-CCI and grey boxes indicate KO-CCI. Non-significant (N.S.).

Novel Gene Targets of RBM5 in the Brain: Csnka2ip is Expressed in the CNS & is Regulated by RBM5.

A total of four coding or multiple-complex genes were differentially expressed in WTs vs. KOs as determined by global transcriptomic analysis and included: casein kinase 2 alpha prime interacting protein (Csnka2ip/CKT2), Gm756, Serpina3n, and GFAP (Fig. 5A). We focused downstream confirmatory studies on Csnka2ip because it was the most significant coding gene identified (FDR p-value 10⁻⁷) and because prior studies reported it was not expressed in the brain. The mouse gene ensemble database predicts two Cnska2ip transcripts (201 and 202). Nanostring probes were designed to measure the levels of each transcript separately. Cnska2ip-201 was significantly increased in sham KOs vs. sham WTs (Fig. 5B) and in CCI-injured KOs vs. CCI-injured WTs (Fig. 5C), which was consistent with microarray findings. In contrast, Cnska2ip-202 was unaffected by KO and was not the dominant transcript expressed in the brain (Fig. 5B and 5C). Prior studies in rat cortical neurons found that RBM5 inhibition robustly increased mRNA levels of the small GTPase Rab4a (Jackson et al., 2017). Surprisingly, Rab4a was not among the differentially expressed genes detected in microarray studies on cortical tissue here. We confirmed this finding with Nanostring. The mouse gene ensemble database predicts three Rab4a transcripts and two are protein coding (201 and 202). Rab4a-202 was the dominant coding transcript in the brain. Consistent with the results of microarray, Nanostring analysis did not show increased levels of either transcript in the KO cortex (Fig. 5D and Fig. 5E). Finally, to further confirm and characterize Cnska2ip expression in the brain, ISH analysis using high-sensitivity RNAscope® technology was done to visualize the distribution of Cnska2ip-201 transcripts in sagittal brain sections. Csnka2ip staining was seen in the cerebellum, hippocampus (CA1, CA3, and DG), and in the cortex in both WTs (Fig. 6A–E) and KOs (Fig. 6F–J). Semi-quantitative Image-J analysis did not reveal a 2-fold increase in Cnska2ip staining in the KO vs. WT cortex as was seen with quantitative microarray and Nanostring (Fig. 11). In summary, brain-targeted RBM5 KO increased the levels of several novel gene targets including Cnska2ip in the cortex but did not alter Rab4a levels.

Fig. 5: Differentially Expressed Genes in the KO Cortex.

(A) The table shows the shortlist of genes that were significantly increased or decreased in KOs. Green columns compare sham-KO vs. sham-WT and show the associated FDR p-value. Orange columns compare CCI-KO vs. CCI-WT and show the associated FDR p-value. Empty blocks indicate that the gene of interest was not differentially expressed in KO vs. WT for that experimental group. The yellow column indicates the gene category. (B) Increased Cnska2ip-201 mRNA levels in sham-KOs was confirmed by quantitative Nanostring analysis (n=7/group). (E) Increased Cnska2ip-201 mRNA levels in CCI-KOs was confirmed by quantitative Nanostring analysis. (D and E) Nanostring analysis confirmed microarray findings that neither Rab4a-201 or Rab4a-202 transcripts were significantly altered in the cortex in sham-KOs or CCI-KOs vs. respective WT controls (n=7/group). Box plots show the minimum, maximum, IQR, and median for the samples (n=7/group). Data were analyzed by a t-test and significant at p<0.05. Clear boxes indicate WTs and grey boxes indicate KOs.

Fig. 6: Total Csnka2ip mRNA ISH Staining in the WT & KO Brain.

Sagittal brain sections were stained for Csnka2ip using RNAscope® technology. Transcripts were detected in the WT brain in the: (A) cerebellum, (B) CA1 hippocampus, (C) CA3 hippocampus, (D) cortex, and (E) hippocampal dentate gyrus. Csnka2ip staining was also found in the KO brain in the: (F) cerebellum, (B) CA1 hippocampus, (C) CA3 hippocampus, (D) cortex, and (E) hippocampal dentate gyrus. Square panels show 4X magnification of the entire stained brains. Circles show 20X magnification of sub-regions. The 20X CA3 image includes a representative scalebar (50μM distance) for reference. Hematoxylin staining is in purple. Probe-specific mRNA staining is in red.

Fig. 11: Image-J Analysis of ISH Data.

(Table 1). The left table shows fold-change RNAscope staining levels (KO/WT) for: total Csnka2ip, total Rims2, and total PPIB (positive control). (Table 2): The right table shows fold-changes BaseScope staining levels (KO/WT) for: Rims2 E19-E20, Rims2 E22-E23, and total PPIB. The mean pixel staining was averaged from three regions of interest (ROIs) randomly selected within each 20X image.

Novel Splicing Targets of RBM5 in the Brain

We first analyzed global splicing changes in WT mice (CCI vs. sham) to better characterize splicing events manifesting after injury in our experimental mouse model of TBI. 248 genes (out of 65,770) representing 0.37% of the total analyzable genome using current arrays passed filter criteria in support of alternative splicing (Fig. S2). A total of 564 events were detected among the 248 genes (Fig. S2). Most events were cassette exons (341/564). GO analysis in PANTHER14.1 mapped 174 of the 248 genes. (Fig. S2B and S2C). The effect of alternative splicing on altered protein function(s) are unknown for most mammalian genes. Nevertheless, Fig. S2B highlights a few notable splicing events which were robustly detected in our analysis and warrant future study as they may be important in understanding mechanisms of secondary brain injury after a TBI (i.e. beyond the primary goal of this study). Key targets included changes in the levels of cassette exons for caspase-8, matrix metallopeptidase 12 (MMP-12), serpina3g/Spi2A, TNF receptor superfamily member 1A (TNFR1), GFAP, tissue inhibitor of metalloproteinase 1 (TIMP-1).

Next we compared splicing changes in KOs vs. WTs in the cortex. A total of 22 events passed filter criteria and were affected by genotype in our primary analysis (Fig. 7A). The splicing events were largely independent of injury (i.e. the magnitude of the difference was similar in KO shams or KO CCI vs. WTs). Thus, we collapsed datasets by genotype for a secondary analysis (n=14 KOs and n=14 WTs) to screen for additional potential splicing events which may have evaded detection in the primary analysis. Three additional potential splicing events were found (Fig. 7A; Blue Columns). Fig. 7B shows an example of the microarray probe-level values for PSRs/junctions in the Rims2 gene. We were particularly interested in Rims2 splicing changes because RBM5 KO resulted in increased levels of three different cassette exons in that gene.

Fig. 7: Splicing Changes in the KO Cortex.

(A) The table shows the shortlist of splicing events that were significantly increased or decreased in KOs. Green columns compare sham-KO vs. sham-WT and show the associated FDR p-value. Orange columns compare CCI-KO vs. CCI-WT and show the associated FDR p-value. Blue columns show a secondary analysis of the combined datasets for sham and injured mice in each genotype (n=14/genotype). Empty blocks indicate that the splicing event for the gene of interest was not differentially expressed in KO vs. WT for that experimental group. The yellow column indicates the splicing event category. (B) PSR probe/junction results in TAC 4.0 software for Rims2 in KOs vs. WTs. (C) Altered Rims2 splicing in cortical brain samples was confirmed by quantitative Nanostring analysis. Transcripts harboring a contiguous E19-E23 spliced junction are decreased in KOs (Top blue panel). Transcripts harboring a contiguous E21-E22 spliced junction are increased in KOs (Bottom blue panel). Probes targeting a region outside of known splice sites show equivalent levels in KOs vs. WTs (Bottom pink panel). Box plots show the minimum, maximum, IQR, and median for the samples (n=7/group). Data were analyzed by a t-test and significant at p<0.05. Clear boxes indicate WTs and grey boxes indicate KOs.

Next we performed confirmatory NanoString studies on genes which harbored alternately spliced cassette exons found in the primary analysis and also verified the intron-retention event in myeloid leukemia factor (Mlf2) because it had a notably robust degree of significance (i.e. FDR p-value of 10⁻⁹). Custom probes were designed to anneal at specified junctions (Table S3): Rims2 (exons 19–23 , exons 21–22, and at a loci outside of known splice sites), Mef2a (exons 5–6), Ndrg2 (exons 2–3), Slc201a (exons 5–6), Lrrc49 (exons 3–5 and exons 9–10) and, Mlf2 (intron 8-exon 8 and intron 8-exon 9). A custom probe to measure the cassette exon event in Phkb could not be designed and was removed from further confirmatory analysis. Rims2 transcripts containing a contiguous exon 19–23 were decreased in KOs vs. WTs (Fig. 7C; Probe 1, top blue panel). Rims2 transcripts containing a contiguous exon 21–22 were increased in KOs vs. WTs (Fig. 7C; Probe 2, bottom blue panel). Rims2 transcripts targeting regions outside of known splice sites (i.e. measures total mRNA levels) were not statistically different in KOs vs. WTs (Fig. 7C; Probe 3, bottom pink panel). A protein alignment analysis of exons 20–22 in the mouse Rims2 transcript- 201 vs. the human Rims2 transcript-204 in LALIGN indicated 93.6% identity indicating conservation of these splice sites in humans (Fig. S3). Heterozygous mice also had significantly increased levels of Rims2 transcripts containing contiguous exons 21–22 vs. WTs (Fig. S4).

Mef2a Nanostring findings supported microarray results; a probe which overlapped with the spliced exon 5 was significantly decreased in KOs vs. WTs (Fig. S5B). Ndrg2 Nanostring findings also supported microarray results; a probe which measured the levels of transcripts that exclude (skip) exon 3 was decreased in KOs vs. WTs (i.e. suggesting that exon 3 is increased in KOs; Fig. S5C). Slc20a Nanostring findings supported microarray results – a probe which overlapped with the spliced exon 6 was significantly decreased in KOs vs. WTs (Fig. S5D). Lrrc49 Nanostring findings were incongruent with microarray findings and did not confirm altered splicing of exon 4 or exon 10 (Fig. S5F and S5G, respectively). Mlf2 Nanostring findings supported microarray results; a probe which measured the boundary of intron 8 and exon 9 was robustly decreased in KOs vs. WTs (Fig. S5H). Germane to Slc201a, heterozygous mice also had decreased levels of exon 6 (Fig. S4).

Rims2 changes were among the most robust cassette-exon splicing events detected in our transcriptomic analysis in KOs and was present in heterozygotes. Thus, we focused further ISH staining characterization efforts on Rims2 splicing changes. First, total Rims2 mRNA staining was visualized and found to be widespread in the cerebellum, hippocampus (CA1, CA3, and DG), and in the cortex in both WTs (Fig. 8A–E) and KOs (Fig. 8F–J). Semi-quantitative Image-J analysis revealed a ~4-fold increase in total Rims2 staining in KOs in the cerebellum and was associated with altered morphology of the cerebellar folia (Fig. 8A vs. 8F and Fig. 11). RNAscope® controls confirmed the specificity/sensitivity of the assay. No staining was seen in the negative control (bacterial gene dapB; Fig. S6). The positive control stain PPIB was equivalent and widespread in the cerebellum, hippocampus (CA1, CA3, and DG), and in the cortex in WTs and KOs (Fig. S7). Notably, PPIB staining was similar in the WT vs. KO cerebellum (Fig. S7A vs. S7F and Fig. 11).

Fig. 8: Total Rims2 mRNA ISH Staining in the WT & KO Brain.

Sagittal brain sections were stained for total Rims2 mRNA using RNAscope® technology. Rims2 expression is ubiquitous in the WT brain and was seen in the: (A) cerebellum, (B) CA1 hippocampus, (C) CA3 hippocampus, (D) cortex, and (E) hippocampal dentate gyrus. Rims2 expression was also ubiquitous in the KO brain but levels were markedly increased in the (F) cerebellum. In contrast, expression in the KO: (B) CA1 hippocampus, (C) CA3 hippocampus, (D) cortex, and (E) hippocampal dentate gyrus, were compared to levels in WTs. Square panels show 4X magnification of the entire stained brains. Circles show 20X magnification of sub-regions. The 20X CA3 image includes a representative scalebar (50μM distance) for reference. Hematoxylin staining is in purple. Probe-specific mRNA staining is in red.

Next, Rims2 transcripts harboring specific splicing events were visualized using BaseScope^™ technology. To further compliment Nanostring and microarray investigations, custom ISH probes were designed to selectively target the proximal and distal regions of the triexon block splice sites which were not analyzed in the Nanostring study (e.g. Probe 1: Rims2 exon junction 19–20, and Probe 2: Rims2 exon junction 22–23). Rims2 E19-E20 staining (Fig. 9 and Fig. 11) and Rims2 E22–23 staining (Fig. 10 and Fig. 11) were increased (> 2-Fold) in KOs vs. WTs. Notably, semi-quantitative Image-J analysis revealed that the largest increase was in all fields of the hippocampus. (e.g. Fig. 9B vs. 9G, 9C vs. 9H, 9E vs. 9J and Fig. 11). BaseScope^™ controls confirmed the specificity/sensitivity of the assay. No staining was seen with the negative dapB (Fig. S8). PPIB staining (positive control) was widespread in WT and KOs but did not differ by genotype (Fig. S9 and Fig. 11). In summary, brain-targeted RBM5 KO altered splicing of several novel gene targets including Rims2.

Fig. 9: Rims2 Exon Junction 19–20 ISH Staining in the WT & KO Brain.

Sagittal brain sections were stained for Rims2 splicing at the E19-E20 junction using Basescope® technology. Spliced Rims2 transcripts were detected at low levels in the WT brain in the: (A) cerebellum, (B) CA1 hippocampus, (C) CA3 hippocampus, (D) cortex, and (E) hippocampal dentate gyrus. Spliced Rims2 transcripts were increased in the KO: (F) cerebellum, (B) CA1 hippocampus, (C) CA3 hippocampus, and (E) hippocampal dentate gyrus. Levels were comparable to WTs in the (D) cortex. Square panels show 4X magnification of the entire stained brains. Circles show 20X magnification of sub-regions. The 20X CA3 image includes a representative scalebar (50μM distance) for reference. Hematoxylin staining is in purple. Probe-specific mRNA staining is in red.

Fig. 10: Rims2 Exon Junction 22–23 ISH Staining in the WT & KO Brain.

Sagittal brain sections were stained for Rims2 splicing at the E22-E23 junction using Basescope® technology. Spliced Rims2 transcripts were detected at low levels in the WT brain in the: (A) cerebellum, (B) CA1 hippocampus, (C) CA3 hippocampus, (D) cortex, and (E) hippocampal dentate gyrus. Spliced Rims2 transcripts were increased in the KO brain in the: (B) CA1 hippocampus, (C) CA3 hippocampus, (D) cortex, and (E) hippocampal dentate gyrus. Levels were comparable to WTs in the (F) cerebellum. Square panels show 4X magnification of the entire stained brains. Circles show 20X magnification of sub-regions. The 20X CA3 image includes a representative scalebar (50μM distance) for reference. Hematoxylin staining is in purple. Probe-specific mRNA staining is in red.

Discussion

RBM5 is increased in the mouse brain after an experimental TBI (Jackson et al., 2015). We speculated that its upregulation after trauma might promote the expression of pro-death genes in the CNS based on its well-known tumor suppressor functions in cancer. However, a single report on RBM5 gene targets in neurons did not find evidence that it regulates cell death pathways in uninjured-healthy cells (Jackson et al., 2017). Thus, we hypothesized that RBM5 may promote the expression and/or alter the splicing of cell death genes only after injury. Here we used the CCI-TBI model, a well-established tool to study a broad spectrum of pro-death genes/pathways in the brain, to test in vivo our primary hypothesis that RBM5 inhibition decreases pro-death gene expression in the injured cortex (Clark et al., 2000; Franz et al., 2002; Newcomb et al., 1999). Moreover, we created novel conditional KOs to minimize confounding cardiovascular effects or other peripheral changes induced by total body gene manipulations, and also to avoid postnatal death in RBM5 null pups which was seen in other strains (Jackson et al., 2018a; Jamsai et al., 2017).

As expected, CCI increased the expression of pro-death genes including caspase-8, caspase-4, and FAS. Brain-specific RBM5 gene deletion did not affect their expression level or RNA-splicing. Additional studies are needed to evaluate our primary hypothesis in different models of CNS injury and to measure potential changes in the transcriptome at later stages postinjury. In addition to cell death genes, studies in cortical neurons reported that Rab4a is a novel target regulated by RBM5 in vitro but we did not detect differences in the cortical levels of this gene in WTs vs. KOs in vivo; the reason for that discrepancy is unclear but may relate to differences in the experimental design between studies (e.g. species investigated, developmental stage of neurons, the influence of culture media on background gene expression, or the composition of the transcriptome in homogenous cortical neurons vs. heterogenous cortical brain tissue). Despite the lack of evidence to support the concept that RBM5 increases the expression of canonical pro-death genes in the injured brain, others have shown that it promotes neurotoxicity in diverse models of neuronal injury in vitro (Jackson et al., 2015; Jackson et al., 2018b; Zhang et al., 2015). Thus, it is possible that RBM5 enhances neuronal vulnerability via novel mechanisms which differ from those activated in cancer. That concept merits further investigation.

Csnka2ip was a surprising target identified in our screen. RBM5 KOs (shams and CCI-injured) had increased mRNA levels of Csnka2ip in the cortex vs. WTs. The seminal study on Csnka2ip reported that mRNA levels were undetectable in the mouse brain and found that it was exclusively expressed in the testes (Bai et al., 2009). The authors performed reverse-transcription PCR followed by gel electrophoresis to visualize PCR products. We do not know why our findings deviated from prior work. Both false-negative and false-positive findings in traditional RT-PCR studies have been widely discussed (Lion, 2001). Nevertheless, we used three markedly different and ultra-sensitive techniques (contemporary microarrays, Nanostring, and high-fidelity ISH) to confirm Csnka2ip expression in the mouse brain in WTs and in KOs.

The function of Csnka2ip in neurons is unknown. To our knowledge only two articles have reported on facets of its regulation (Bai et al., 2009; Liao et al., 2016). Mechanistically, it binds to the catalytic subunit of casein kinase 2 alpha prime (CK2α’) and is phosphorylated by the latter (Bai et al., 2009). Phosphorylated Csnka2ip is detectable in testicular lysates from WT mice but not in CK2α’ KOs (Bai et al., 2009). Germane to the brain, CK2α’ is widely distributed across the CNS and CK2 proteins phosphorylate over 300 substrates (Castello et al., 2017; Meggio and Pinna, 2003). Thus, Csnka2ip may play an important role in the regulation of signaling cascades in the brain. According to the mouse gene ensemble database there are two protein coding Csnka2ip transcripts. Transcript-201 encodes a small 276 amino acid (~30kDa) protein, and transcript-202 encodes a large 720 amino acid (~80kDa) protein. The small protein in specific was found to bind CK2α’ in mouse testes and modulate signal transduction (Bai et al., 2009). RBM5 KO increased the small (201) but not the large (202) transcript. Csnka2ip ZZ probe pairs in ISH studies were designed against regions within the mouse 201 transcript. Image-J analysis of ISH data did not indicate a > 2-fold increase of Csnka2ip levels in the KO vs. WT cortex as was seen with microarray and Nanostring. The lower sensitivity and semi-quantitative nature of staining analysis may explain the discrepancy. No commercial anti-Csnka2ip antibodies are currently available. Thus, future studies are needed to confirm the presence of brain Csnka2ip at the protein level pending the development and validation of the necessary reagents.

There is limited data on global gene-splicing changes in models of TBI (Meng et al., 2017). Therefore, we think it worthwhile to highlight some notable splicing alterations detected in our analysis in WT mice before discussing the effects of KO. Caspase-8: Exon 3 skipping in caspase-8 results in an anti-apoptotic short variant (caspase-8s) (Kisenge et al., 2003). In our analysis exon 2 and 3 in caspase-8 decreased 2.65-fold and 2.72-fold (i.e. splicing index/SI) after a CCI. MMP-12: MMP-12 promotes blood brain barrier damage after focal ischemic stroke (Chelluboina et al., 2015). We found that exon 4 in MMP-12 increased 4.48-fold (SI) after a CCI. Serpina3g: Cathepsin B is a cysteine protease which mediates cell death and contributes to TBI-induced tissue injury (de Castro et al., 2016; Hook et al., 2015). Relatedly, Serpina3g/Spi2A is a potent cathepsin B inhibitor (Liu et al., 2003). We found that Exon 4 in the Serpina3g transcript increased 43.32-fold (SI) after a CCI. TNFR1: TNFR1 redistributes to lipid rafts after TBI and mediates neuronal death (Lotocki et al., 2004). In our analysis exon 3 in TNFR1 increased 3.87-fold after a CCI. GFAP: GFAP is a clinical biomarker in TBI prognostication (Bazarian et al., 2018). Exon 1 in GFAP decreased 21.9-fold after CCI in our study. TIMP1: Overexpression of tissue inhibitor of metalloproteinase 1 (TIMP-1) decreases brain lesion volume after a CCI in mice (Tejima et al., 2009). Exon 3, exon 4, and exon 5 in TIMP-1 increased 7.01-fold, 5.57-fold, 3.52-fold, respectively, after a CCI. None of these injury-dependent splicing events were affected by RBM5 KO.

A total of 22 differentially expressed splicing events were detected in our primary analysis of cortical brain tissue in KOs vs. WTs. The limited number of splicing events that differed in KOs may be due to functional redundancies in other closely related RBM paralogues (e.g. RBM6/RBM10). Consistent with that possibility, RBM5 knockdown in HeLa cells failed to alter FAS splicing (as seen here in the brain) unless RBM6 and RBM10 were simultaneously coinhibited (Bonnal et al., 2008). Another possibility is that RBM5 regulates a limited set of gene targets. Indeed, Bechara and colleagues individually knocked down RBM5, RBM6, or RBM10 in breast cancer cells and found a total of 30, 252, or 302 alternative splicing events, respectively (Bechara et al., 2013). Thus, the total number of splicing events induced by selective RBM5 inhibition in the Bechara study was similar to our observations here. The authors found little overlap in the genes affected by inhibition of each RBM (Bechara et al., 2013).

Microarray discoveries on KO-dependent changes in Slc20a1, Mef2a, Ndrg2, and Rims2 splicing were validated by Nanostring. Rims2 captured our interest the most, in part, because it is involved in neurotransmission. RBM5 KOs had increased inclusion of exons 20, 21, and 22 in the Rims2 transcript. This finding was confirmed by three different approaches and four different probes targeting the: (a) E19-E22 junction, (b) E21-E22 junction, (c) E19-E20 junction, and (d) E22-E23 junction (Fig. S10). In addition, these splicing events were seen in multiple brain regions including in the cortex and in the hippocampus. The simultaneous block-splicing of exons 20–22 was predicted in an analysis of the mouse Celera genome database in 2002 and termed “Splice Site C” (Wang and Sudhof, 2003). The three exons referenced in that report were identified as 23, 24, and 25 and later revised (e.g. currently exons 20–22 in transcript Rims2–201). The authors also predicted that the block-exclusion of exons 20–22 had functional importance because it kept the transcript in-frame (yielded a protein product), whereas individual splicing of 20, 21, or 22 alone did not (Wang and Sudhof, 2003). To the best of our knowledge RBM5 is the first known regulator of Splice Site C. Future studies are needed to elucidate the precise molecular mechanisms mediating altered Rims2 splicing and if RBM5 exerts a direct vs. indirect effect on cassette exon definition. Notably, Splice Site C is predicted to be present in the human Rims2 gene (Wang and Sudhof, 2003). Also, protein alignment analysis of exons 20–22 in the mouse Rims2 transcript-201 vs. the human Rims2 transcript-204 indicated 93.6% shared identity.

The effect that Splice Site C has on the function of the Rims2 protein is unknown. Rims2 plays a critical role in neurotransmitter dynamics and tethers Ca²⁺ channels to the active zones of synapses. Rapid synchronous neurotransmitter release is severely impaired in double Rims1/2 KOs (Kaeser et al., 2011). Also, selective Rims2 KOs have decreased spontaneous and evoked neurotransmitter release in neurons but no change in Ca²⁺ responsiveness or release synchronization (Kaeser et al., 2012). The spliced region is near the c-terminal C₂B domain, which is essential to bind the ubiquitous β4 subunit of voltage-dependent calcium channels (Kiyonaka et al., 2007; Wang and Sudhof, 2003).

The clinical significance of altered Rims2 signaling in the brain is unclear but is emerging as a potential target in neurological conditions. Mice exposed to open-field blast-TBI showed changes in Rims2 phosphorylation in the brain (Chen et al., 2018). Schizophrenic patients have increased Rims2 levels in the brain (Weidenhofer et al., 2006; Weidenhofer et al., 2009). A loss-of-function point mutation in the Rims2 gene was the only variant risk factor for tardive dyskinesia in a recent whole-exome clinical study (Alkelai et al., 2019). Finally, polymorphisms in the Rims2 gene are associated with degenerative lumbar scoliosis (Kim et al., 2013).

Another unique feature of Rims2 neurobiology, which could have relevance germane to finding here, is its remarkably high ratio of circular RNAs (circRNAs) to linear mRNA in the human and mouse brain (20 fold higher) (Rybak-Wolf et al., 2015). Notably, circRims2 localizes primarily to the cerebellum in the adult mouse brain and is produced by “backsplicing” of pre-mRNAs which yields highly stable transcripts with long half-lifes (Holdt et al., 2018; Rybak-Wolf et al., 2015). RBPs were recently shown to play a major role in regulating circRNA expression in the brain (Ji et al., 2019). Given that: (a) RBM5 is an RBP which regulates Rims2 splicing (as shown here for the first time), (b) total Rims2 mRNA staining was markedly increased in the cerebellum, and (c) gross cerebellar morphology was altered in KOs - we speculate that RBM5 might regulate circRims2 splicing as well. Our assays were unable to investigate circRNAs but remains an important direction for future studies.

In summary, this is the first study to elucidate genes regulated by RBM5 in the intact brain. Surprisingly, RBM5 KO did not decrease the expression of pro-death genes following their induction with a CCI-TBI in vivo. The unexpected finding subverts our current thinking germane to the possible gene pathways regulated by RBM5 in the brain. Whereas it potently promotes the expression of cell death genes in cancer, we found that RBM5 regulates the expression and splicing of genes with a presumptive role in neurotransmission. More studies are needed to determine if blocking RBM5 in vivo is neuroprotective and whether it activates novel mechanisms which could have therapeutic benefits in the setting of a brain injury.

Highlights:

• In vivo elucidation of mRNA targets regulated by RBM5 in the brain.

• RBM5 KO increases total levels of Rims2 transcripts in the cerebellum.

• RBM5 KO increases Rims2 splicing in the cortex and in the hippocampus.

• Csnka2ip mRNA is expressed in the mammalian brain.

• RBM5 KO increases total levels of Csnka2ip mRNA in the cortex.

Abbreviations:

RBPs

RNA-binding proteins

RBM5

RNA-binding motif 5

RRM

RNA recognition motif

OCRE

Octamer repeat of aromatic residues

Csnka2ip/CKT2

Casein kinase 2 alpha prime interacting protein

Rims2

Regulating synaptic membrane exocytosis 2

TBI

Traumatic brain injury

CCI

Controlled cortical impact

GFAP

Glial fibrillary acidic protein

Slc20a1

Solute carrier family 20 member 1

Mef2a

Myocyte enhancer factor 2a

Ndrg2

N-Myc Downstream-Regulated Gene 2 Protein

ISH

In Situ Hybridization

PPIB

Cyclophilin B

dapB

bacillus subtilis dihydrodipicolinate reductase