Muscleblind‐like 2 knockout shifts adducin 1 isoform expression and alters dendritic spine dynamics of cortical neurons during brain development

Muscleblind‐like 2 (MBNL2) plays a crucial role in regulating alternative splicing during development and mouse loss of MBNL2 recapitulates brain phenotypes in myotonic dystrophy (DM). However, the mechanisms underlying DM neuropathogenesis during brain development remain unclear. In this study, we aim to investigate the impact of MBNL2 elimination on neuronal development by Mbnl2 conditional knockout (CKO) mouse models.


INTRODUCTION
Myotonic dystrophy (dystrophia myotonica, DM) is the most common muscular dystrophy in adults with a prevalence ranging from 0.5 to 180 in 100,000, with a relatively higher prevalence in Europeans and their descendants. [1][2][3] DM is an autosomal dominant inherited disease with multi-systemic involvement, including skeletal muscle atrophy and myotonia, respiratory failure, heart conduction defects and dysfunction of the central nervous system (CNS). 4,5 Two types of DM, type 1 (DM1) and type 2 (DM2), are caused by mutations on two different genes, 6-10 either (CTG) trinucleotide repeats in the 3 0 untranslated region (3 0 -UTR) of DMPK (myotonic dystrophy protein kinase) gene in DM1 or (CCTG) tetranucleotide repeats in the first intron of CNBP (cellular nucleic acid-binding protein) in DM2. 1,5,11 The molecular pathogenesis of DM is through direct sequestration of RNA-binding protein Muscleblind-like (MBNL) 12 or indirect stabilisation of CUGBP Elav-like family member 1 (CELF1, also known as CUGBP) by the microsatellite repeat expansions. 13 Particularly, the MBNL family of proteins colocalises with RNA foci, which accumulate in the nuclei, and subsequently, the function of MBNLs on RNA processing is affected. 12,14,15 To test the hypotheses of MBNL loss of function in DM, Mbnl KO mouse models were generated and successfully reproduced myotonia, muscle pathology, cardiac conduction delay, respiratory distress, REM sleep misregulation and impaired spatial learning ability. [16][17][18][19] CNS manifestations severely compromise DM patients' capability of handling complex tasks and their quality of life 2,5,20 ; either in congenital, infantile and adolescent DM patients 5 who show learning disability, autism spectrum disorders (ASD) and attention-deficit hyperactivity disorder (ADHD) or excessive daytime sleepiness, depression, apathy, dysexecutive syndrome, social avoidance, and fatigue in adult DM1 and DM2. 5,21 Although advances in magnetic resonance (MR) brain imaging on DM patients and microstructural analysis in mouse models have been achieved in the past few years, [22][23][24][25][26] there is still a huge gap between DM brain phenotypes and the misregulations of gene expression. In particular, it remains unclear how dysfunctions in MBNL-mediated splicing cause DM CNS phenotypes and affect neuronal microstructures.
In this study, we used in utero electroporation (IUE) of Crerecombinase in Mbnl2 flox/flox (Mbnl2 f/f ) mice to knock out Mbnl2 in neural progenitors and their progeny cells and aimed to examine the dynamics of synapses by two-photon microscopy in live animals. We found evidence of dendritic spine abnormality, both in spine density and dynamics, during the course of brain development. We found the MBNL2 deficiency caused a splicing shift of Add1 mRNA toward the fetal isoform, which in turn altered the phosphorylation and binding of Add1 mRNA product adducin 1 (ADD1) to cytoskeletal protein alpha-II spectrin (SPTAN1). These morphological and molecular changes caused by MBNL2 elimination in the developing neurons may consequently contribute to DM brain pathogenesis.

Animal model
The Mbnl2 f/f mouse model was originally generated and gifted by Dr. Maurice Swanson in the Department of Molecular Genetics and Microbiology, University of Florida (Gainesville, FL, USA). 16 The wildtype (WT), heterozygous and homozygous floxed Mbnl2 alleles were confirmed by genotyping and noted as Mbnl2 +/+ , Mbnl2 +/f and Mbnl2 f/f , respectively. Mbnl2 conditional knockout (Mbnl2 f/f ; Nestin-Cre +/À or Nestin-Cre CKO or CKO) models were generated by mating the Nestin-Cre line with conditional Mbnl2 f/f , as previously reported. 26,27 IUE The detailed procedures of IUE have been published elsewhere. [28][29][30] 30 Briefly, mice were anaesthetised with isoflurane (induction: 4% supplied by chamber; maintenance: 1.5%-2% supplied by mask), and an incision was made through the skin and the abdominal muscle in

Key Points
• Loss of MBNL2 in cortical neurons leads to a reduction of dendritic spine density and abnormal spine dynamics in adolescent mice.
• Mbnl2 KO shifts ADD1 to fetal isoform, which reduced the ability to interact with SPTAN1.
• Expression of ADD1 adult isoform restores the dendritic spine density in Mbnl2 KO neurons.
• ADD1 isoform shift may contribute to neuronal defects in the mouse model of myotonic dystrophy. order to expose the underlying viscera. Each embryo was randomly injected with $0.5 μL of plasmid DNA (final concentration for Mbnl2 knockout: 1.5 μg/μL pCALNL-GFP + 0.5 μg/μL pCAG-Cre; or final concentration for Add1 expression: 1.5 μg/μL pCALNL-GFP + 1.5 μg/μL pCALNL-Add1-IRES-GFP + 0.5 μg/μL pCAG-Cre) into the lateral ventricle on one side of the brain. Electroporation was then carried out at a voltage of 40 V, with five 50 ms pulses separated by 450 ms intervals. After electroporation, the uterine horns were placed back into the abdominal cavity carefully, and the incision was closed by sutures. The embryos were allowed to develop in the uterus, and pups were born through normal spontaneous delivery. The brains of the electroporated mice were harvested at E18.5, P1, P7, P14, P21, P30 and P90.

Constructs for knockout and overexpression
pCAG-Cre and pCALNL-GFP were gifts from Connie Cepko (Addgene, plasmid #13775 and #13770) and were used for Mbnl2 IUE-mediated knockout. The pCAG-Cre expresses Cre recombinase only in the electroporated neural progenitors (i.e., radial glial cells). The pCALNL-GFP is composed of SV40 poly(A) stop signal floxed by two LoxP sites, upstream of the IRES-GFP sequence that allows the expression of GFP and labels the electroporated cells. Add1 cDNA construct was designed, synthesised and then cloned into pCALNL-GFP by GenScript (New Jersey, USA). This gene was termed pCALNL-Add1-IRES-GFP.

Brain section and immunofluorescence staining
Immunofluorescence staining of brain slices was performed as previously described. 31,32 Brains were perfused and immersed with 4% paraformaldehyde (PFA), embedded in agarose gel and sectioned by vibratome (Leica). Slices were washed with phosphate buffered saline (PBS), followed by phosphate buffered saline with Triton X-100 (PBST, 0.2% of Triton X-100 in PBS) permeabilization for 30 min.

Protein lysis and western blot analysis
The cortex tissue was lysed in RIPA buffer (Sigma-Aldrich, R0278) containing 10% protease inhibitor (Sigma-Aldrich, 04693159001) and

Dendritic spine imaging and analysis
Mouse brains electroporated with pCALNL-GFP and pCAG-Cre constructs were fixed and sectioned. Six to eight brain slices of each mouse were then mounted. Next, 10 to 15 basal dendrites of cortical layer II/III neurons from the somatosensory cortex were randomly selected and imaged with confocal microscopy (LSM700/LSM880, Carl Zeiss), and the dendrites were analysed using semi-automated 33 and artificial intelligence (AI)-assisted algorithms (NYCU-ASUSTeK).
Spine density results were acquired through the analysis of thousands of dendritic spines from more than three mice in each group. Since the structure of dendritic spines was small, the frame size of each image was unified as 1024 (X)*1024 (Y), and the size of each pixel was controlled as 0.078 μm*0.078 μm*0.42 μm (X, Y, Z).

Thinned-skull cranial window and in vivo two-photon imaging
The procedure of thinned-skull preparation was described previously. 34 Briefly, a round region of the skull 1 mm in diameter above the somatosensory cortex was thinned into 20-30 μm in each mouse without causing microbleeds or skull damage. All mice underwent surgeries twice on experimental day 0 and day 2. Apical dendrites extending from GFP + pyramidal neurons (about 20 apical dendrites from more than three mice in each group) were randomly selected and monitored by two-photon microscopy (7MP, Carl Zeiss). All images were scanned with frame size 512(X)*512(Y) and then analysed by using a semi-automated algorithm as described above. The size of each   method for generating GST-tagged protein using Escherichia coli has been previously described . The E18.5 WT mouse brain lysates were incubated with glutathione resins conjugated with neck and tail domains of ADD1 Àex15 or ADD1 +ex15 overnight at 4 C. Western blotting was used to identify the interaction between ADD1 and SPTAN1 with an anti-alpha-II spectrin antibody (Santa Cruz, sc-48382).

In situ hybridization
A digoxigenin-labelled antisense RNA probe for Mbnl2 mRNA was prepared according to the manufacturer's instructions (Roche). The sequence of the Mbnl2 probe was obtained from the Allen Institute for Brain Science (https://portal.brain-map.org/). Brain samples for in situ hybridization were derived from 2 months old Mbnl2 +/+ and  (dT) and random primers (Solis BioDyne, 06-20-00500). Isoforms of gene targets were amplified from cDNA by PCR. Specific primers for targets were as previously described. 16 Amplified cDNAs were further analysed by DNA electrophoresis and ImageJ (NIH, USA).

Statistical analysis
All data were statistically analysed with GraphPad Prism (version 8.

Macro-structures of the hippocampus and cerebral cortex in Mbnl2 KO mice
Since spatial learning deficits were observed in Mbnl2 ΔE2/ΔE2 mice, 16 we first examined the development of hippocampal neurons, which have been implicated in the formation of spatial memory. 35  including the thickness and area in brain sections ( Figure S1C). We found no significant differences between these mice ( Figure S1D), consistent with our previous study. 26 Therefore, from a macrostructural perspective, the early development from immature to mature neurons was not affected in the mouse hippocampal neurons that were devoid of MBNL2 in neural progenitor cells.
Because cognitive deficits in DM patients may also be related to cortical function, we then investigated the effects of MBNL2 loss of function on cortical development. We applied IUE to introduce cDNA expressing Cre-recombinase (pCAG-Cre) into neural progenitors  Figure 3B). The complexity of dendritic branches was analysed by Sholl analysis at 1 and 3 months. We found the numbers of dendritic branches did not show significant differences among all groups ( Figure S2A, B). Transcallosal axon projections towards the contralateral brain could also be readily observed ( Figure S2C). The general structure of axons also did not show an apparent difference in all groups although ultrastructural changes may exist but could not be detected due to the limitation of optical resolution ( Figure S2D).
The densities of dendritic spines that originated from these dendrites were analysed at P14, P21, P30 and P90 among Mbnl2 +/+ ,   We further investigated whether Mbnl2 knockout also affects different spine types, which are typically classified into filopodia, thin, stubby, and mushroom, based on morphology 44 ( Figure S3A). We found that there was no significant difference in the composition of spines, either immature (e.g., filopodia) or mature (e.g., mushroom) type of spines, between Mbnl2 +/+ and Mbnl2 f/f neurons at four different time points throughout the period from P14 and P90 ( Figure S3B).
Since MBNL2 depletion in cell types other than neurons may interfere with the observed neuronal phenotypes, we tested whether depriving MBNL2 through IUE in Mbnl2 f/f mice may affect glial cell density and morphologies using immunofluorescence staining. Astrocytes were stained with GFAP (glial fibrillary acidic protein) in electroporated cortices at P21. We found that the morphology and density of astrocytes in electroporated Mbnl2 f/f cortices did not show differences compared to those in Mbnl2 +/+ cortices ( Figure S4A, B). Mbnl2 +/+ and Mbnl2 f/f cortices ( Figure S5).

Alternative splicing changes of Add1 in Mbnl2 KO mice
Previous studies revealed that constitutive loss of MBNL2 causes thousands of splicing changes in its downstream targets by RNA-seq and microarray. Forty-two highly ranked mis-spliced genes were found in both data sets. We found most affected genes switched the splicing pattern back to the fetal isoform in the adult mouse hippocampi. 16 Among them, Add1 (adducin 1), Tanc2 (tetratricopeptide repeat, ankyrin repeat and coiled-coil containing 2) and Dlg2 (discs large MAGUK scaffold protein 2, also known as PSD-93) were involved in dendritic spine development [45][46][47][48] ( Figure 4A). Thus, we first examined adult and fetal mRNA isoforms of Add1, Tanc2 and Dlg2 in Cre-electroporated brain tissue in Mbnl2 f/f mice at P30 using RT-PCR and validated a shift from adult to fetal isoform in all 3 genes ( Figure S6).
It was previously demonstrated that MBNL2 inactivation leads to an Add1 isoform shift with the inclusion of exon 15 (37 nucleotides), 16 which causes a frameshift and introduction of a premature stop codon in exon 16 ( Figure 4B). This change may potentially produce a truncated adducin 1 (ADD1) protein isoform of 632 a.a. (ADD1 +ex15 at $70 kDa) instead of the normal size of 735 a.a (at $81 kDa). Therefore, we examined the expression of these two isoforms in Mbnl2 CKO mice at P7, P14, P21, P30 and P90 by western blotting ( Figure 4C). In WT mice, significantly higher levels of the fetal ADD1 +ex15 isoform were expressed in the cortex at P7 (before the robust expression of MBNL2). The relative level of the adult ADD1 Àex15 isoform increased dramatically from P7 to P14. Remarkably, in the cortex of Mbnl2 CKO mice, the protein expression level of ADD1 +ex15 fetal isoform was significantly higher compared to the WT mice between P14 and P90, while total ADD1 expression remained unchanged ( Figure 4D). This result suggested that loss of MBNL2 indeed leads to changes in two different ADD1 protein isoforms. This

Increased spinogenesis by ADD1 expression in Mbnl2 KO neurons
Based on these results, we hypothesized that the isoform switch of Adducin 1 belongs to the adducin cytoskeletal protein family and forms a heterodimer with adducin 2 (ADD2, encoded by Add2). Previously, ADD2 was known to play an important role in synapse formation and dendritic spine maturation by interacting with spectrin through its C-terminal tail domain. [48][49][50] To examine the impact of exon 15 inclusion on adducin 1 and its binding capability with spectrin, we generated cDNAs that express the C-terminus of adducin 1 adult and fetal isoforms with a GST tag ( Figure 6A). GST pull-down assay was performed by using E18.5 WT mouse brain lysates in vitro.
We found that the adducin 1 adult isoform possesses a stronger binding affinity with alpha-II spectrin (SPTAN1), compared to the fetal isoform, in which the C-terminal 103 amino acids are missing ( Figure 6B, C). Interestingly, the C-terminus of the adducin 1 adult isoform contains several major phosphorylation sites; therefore, we examined whether overall protein phosphorylation was also changed in fetal F I G U R E 4 Mbnl2 knockout affected ADD1 isoform expression in the developing cortex. (A) Flow chart of screening for potential target genes associated with Mbnl2 knockout effects on dendritic spines. Add1, Tanc2 and Dlg2 are genes encoding synaptic proteins and are misspliced during brain development in  Figure 6D). EGFP-tagged ADD1 Àex15 and ADD1 +ex15 were expressed in cells and then pulled down to detect the phosphorylation level by anti-phospho-(Ser/Thr) antibody. We found that the relative phosphorylation level was significantly decreased in fetal adducin 1 compared to the adult isoform ( Figure 6D, E).

Alterations in dendritic spine dynamics in Mbnl2 KO neurons
Since dendritic spine density in Mbnl2 KO mice was altered, we further investigated the dendritic spine dynamics by utilising transcranial two-photon imaging, a routine experiment in our lab. 34 We performed thinned-skull preparation on Mbnl2 +/+ and Mbnl2 f/f mice electroporated with pCAG-Cre and pCALNL-GFP ( Figure 7A). The initial images were taken on Day 0 (P30) and the same dendrites and spines were re-evaluated on Day 2 (P32) (Figure 7B, C). We found spine elimination rate was significantly higher than the formation rate in Mbnl2 +/+ mice (elimination rate: 32.07 ± 2.17%; formation rate: 21.76 ± 2.75%; mice n = 4, dendrites n = 17) ( Figure 7C, D) (Figure 7C, D). These results suggest that Mbnl2 ablation in neurons may negatively impact the normal pruning process of dendritic spines during adolescence.

DISCUSSION
(2) -4head. (C) Bar and dot graphs show the relative SPTAN1 interaction between each ADD1 isoform. n = 5 mice in each group. Error bars represent SEM. One-way ANOVA test. ***p < 0.001. (D) Phosphorylation of ADD1 in HEK293T cells transfected with pEGFP-Add1 Àex15 and pEGFP-Add1 +ex15 . ADD1 was pulled down by GFP-trap magnetic beads and blotted by anti-Ser/Thr and anti-GFP antibodies. (E) Bar and dot graphs show a decrease in the phosphorylation of fetal ADD1 +ex15 isoform. Error bars represent SEM. Student's t-test with Welch's correction. ***p < 0.001. cells are mostly fated to become cortical layer II/III neurons. Since moderate neuronal migration defects were found previously in Nestin-Cre double knockouts (Mbnl1 ΔE3/ΔE3 ; Mbnl2 f/f ; Nestin-Cre +/À ) and Nestin-Cre CKOs, 26  Here, we studied the somatosensory area of the cortex. Clinically, structural changes of the somatosensory cortex have also been reported in DM patients. 54,55 In a study of DM2 patients, atrophy was found in the white matter of the cingulate gyrus, medial frontal cortex and primary somatosensory cortex. 54 A more recent study also reported that the number of CTG repeats in the leukocytes of DM1 patients was significantly correlated to the thickness of the left primary somatosensory cortex. 55 These findings based on structural magnetic resonance imaging implicated that the function of the somatosensory cortex may also be affected in DM patients. In our study, we reported the reduction of dendritic spine density and dynamics in the somatosensory cortex in the DM mouse model. These results may provide insights into the cellular basis for the pathogenesis of these areas in DM brains.
Although our study focused on pyramidal neurons in the cortex, changes in inhibitory interneurons may also contribute to brain dysfunction in DM patients. Some DM patients display epileptic assaults [56][57][58] and mild cortical changes in the density and distribution of inhibitory interneurons were found in the Mbnl2 f/f ; Nestin-Cre +/À mice. 26 Abnormality in the small group of inhibitory interneurons may result in improper synaptic transmission in the brain. In addition, although our IUE-mediated Mbnl2 KO mice did not show obvious changes in morphologies and densities of glial cells in the cortex, glial cells may also play important roles in modulating multiple neuronal functions, including neurotransmitter metabolism, synaptic structure, and blood-brain barrier integrity. 59 Recently, mis-splicing of genes in primary astrocytes derived from the DMSXL mouse model of DM was Spine dynamics was determined by spine formation and elimination, the ratio between formation rate and elimination rate will lead to F I G U R E 8 Schematic model of MBNL2 in the regulation of spinogenesis. In WT cortical neurons, MBNL2 protein modulates alternative splicing for many synaptic proteins after birth and promotes dendritic spine development. Among them, ADD1 plays a crucial role in spine formation through phosphorylation in the tail domain and interaction with F-actin and SPTAN1 (top). In the Mbnl2 KO cortical neurons, mis-splicing of Add1 causes a premature stop in transcription and truncated ADD1 protein in translation. The lack of ADD1 C-terminal tail leads to less phosphorylation and prohibits the interaction between ADD1 (adducin 1) and SPTAN1 (alpha-II spectrin). These changes may disturb the formation of the adducin-spectrin-actin complex and further lead to a decrease in spine dynamics and density in the mouse model of DM (bottom).
an increase or decrease in spine density. 66 We found the percentage of stable spines was significantly higher in Mbnl2 KO than in WT, and the proportion of eliminated spines was relatively low in Mbnl2 KO neurons. The changes in spine dynamics suggested that spine plasticity may also be affected around P30 and eventually lead to the unexpected increase of spine density in Mbnl2 KO mice at P90. A moderate pruning process after spinogenesis has been considered important for fine-tuning neural circuits, memory consolidation of learning and maturation of cognitive function. 43,66 Deficient or excessive adolescent spine pruning may impair cognitive function and lead to neurodevelopmental and psychiatric disorders. 43,66 Our finding in abnormal dendritic spine dynamics in adolescent Mbnl2 knockout mice may implicate that CNS symptoms may result from abnormal spine formation at early developmental stages, as well as improper spine pruning during adolescence.
The number of dendritic branches is one of the key factors associated with brain connectivity and intelligence quotient (IQ) score. 67 Based on Sholl analysis in the Cre-electroporated neurons in Mbnl2 f/f mice at 1 and 3 months old, we did not find significant abnormality in dendritic branches. In our previous study, a significant reduction of basal dendritic branches and a relatively minor decrease in apical dendrites were found in the layer II/III cortical neurons in Mbnl2 f/f ; Nestin-Cre mice at 2 to 4 months old. 26 In another DM mouse model, EpA960/CaMKII-Cre mouse, which expresses 960 CUG repeats, apical dendrites were not affected until 6 months old but present shortened dendrites at 9 months old. 68 In primary hippocampal neurons transfected with human DMPK 3 0 UTR with CUG repeats (DMPK-CUG 960 ), they showed decreases in dendrite number compared to control neurons. 69 These results implicated that age, neuronal type and cell non-autonomous effects may contribute to different complexity of dendritic morphology.
Previously, Mbnl2 knockout mice enhanced fetal exon inclusion of Add1 at the RNA level. 16 HITS-CLIP was used to detect target RNAs containing direct binding sites for MBNL2 in vivo and identified potential binding sites within the Add1 gene. Here, we further showed the shift of ADD1 isoforms in the protein level ( Figure 4). Mis-splicing and change of reading frame of ADD1 caused a truncated adducin 1 protein, which lacks MARCKS (myristoylated alanine-rich C-kinase substrate) domain in the C-terminal tail. The MARCKS domain is required for the interaction between adducin 1 and spectrin-actin complex, which is critical in dendritic spine formation/maturation and brain development. [70][71][72][73] By in vitro pull-down assay, we demonstrated a reduced binding affinity between fetal adducin 1 isoform and alpha-II spectrin, compared to the adult isoform. Lack of tail domain also led to less phosphorylation of ADD1 fetal isoform. We have examined the sequences of TANC2 and DLG2 fetal isoforms and found that the premature stop codon did not eliminate known/ predicted phosphorylation sites. However, it is still possible phosphorylation or other post-translation modification could be changed on TANC2 or DLG2. In fact, knockout of Add2, which usually forms dimer/tetramer with adducin 1, has been reported to cause decreases in mushroom spine density and learning/coordination deficits in mice. 50,74,75 Our results are compatible with previous reports and implicated that Add1 may be one of the important genes involved in spinogenesis mediated by MBNL2 and may at least in part account for CNS manifestations in DM patients. It was also quite interesting that there was no significant difference in the spine density at P14, whereas the ADD1 isoform shift has already shown a difference. Therefore, other players may contribute to the defects in dendritic spine development in the context of MBNL2 deficiency at this stage.
Although ADD1 has been shown to associate with cerebrovascular/ cardiovascular diseases and cancers through its roles in red blood cells and cancer cells, 76

CONFLICT OF INTEREST
None of the authors reports any conflict of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.

ETHICS STATEMENT
All animal studies followed the protocol approved by the Institutional Animal Care and Use Committee at Chang Gung Memorial Hospital, Keelung branch (IACUC No.2016060302).