Lineage-specific splicing regulation of MAPT gene in the primate brain

Summary

Divergence of precursor messenger RNA (pre-mRNA) alternative splicing (AS) is widespread in mammals, including primates, but the underlying mechanisms and functional impact are poorly understood. Here, we modeled cassette exon inclusion in primate brains as a quantitative trait and identified 1,170 (∼3%) exons with lineage-specific splicing shifts under stabilizing selection. Among them, microtubule-associated protein tau (MAPT) exons 2 and 10 underwent anticorrelated, two-step evolutionary shifts in the catarrhine and hominoid lineages, leading to their present inclusion levels in humans. The developmental-stage-specific divergence of exon 10 splicing, whose dysregulation can cause frontotemporal lobar degeneration (FTLD), is mediated by divergent distal intronic MBNL-binding sites. Competitive binding of these sites by CRISPR-dCas13d/gRNAs effectively reduces exon 10 inclusion, potentially providing a therapeutically compatible approach to modulate tau isoform expression. Our data suggest adaptation of MAPT function and, more generally, a role for AS in the evolutionary expansion of the primate brain.

Graphical abstract

Highlights

• •OU models identified 1,170 cassette exons with lineage-specific splicing
• •These splicing shifts are under stabilizing selection in primate brain evolution
• •MAPT exon 2 and exon 10 underwent anticorrelated, two-step splicing shifts
• •MAPT exon 10 splicing divergence is mediated by divergent MBNL-binding sites

Recinos et al. identified over 1,000 alternative exons showing evolutionarily divergent splicing under stabilizing selection in primate brains. Among them, they found microtubule-associated protein tau (MAPT) exon 2 and exon 10 underwent anticorrelated, two-step splicing shifts, which may suggest adaptation of tau function.

Introduction

Eukaryotic genes undergo splicing of precursor messenger RNA (pre-mRNA) to remove introns and join exons, which is required to produce mature mRNA for protein translation.^1,2 When RNA splicing was discovered over four decades ago, it was hypothesized that the split gene structure allows multiple transcripts and protein isoforms to originate from single genes by alternative splicing (AS). AS not only dramatically amplifies the complexity of the genetic information encoded in the genome but also accelerates evolution, contributing to both speciation and phenotypic variation among individuals within a species.³ This hypothesis is appealing as mammals, including humans, have a similar number of protein-coding genes to lower vertebrates yet exhibit higher organismal complexity. For example, brain size and functions are substantially expanded in primates, especially when comparing humans to other closely related primates. At the molecular level, these predictions have been confirmed. With the advent of splicing profiling technologies such as deep RNA sequencing (RNA-seq), it is clear that AS is not only ubiquitous,^4,5 and frequently under cell-type and developmental-stage-specific regulation,^6,7 but many events have diverged across mammalian species due to both creation and loss of exons^8,9 or quantitative changes in exon inclusion levels.^{10,11,12,13,14,15} However, the functional implications of these divergent AS events are under debate,^16,17 and it remains poorly understood how sequence changes in the splicing-regulatory elements affect their recognition by RNA-binding proteins (RBPs) and lead to a divergence in splicing patterns of different species or lineages.

The microtubule-associated protein tau (MAPT) gene is located in chromosome 17q21, one of the most structurally complex and evolutionarily dynamic genomic regions.^18,19 The encoded protein tau is normally localized to neuronal axons. However, in multiple neurodegenerative diseases, including Alzheimer’s disease (AD), frontotemporal lobar degeneration (FTLD), progressive supranuclear palsy (PSP), and Pick’s disease, tau is frequently found in cytoplasmic aggregates named neurofibrillary tangles (NFTs); thus, these diseases are collectively referred to as tauopathies.^20,21 The MAPT gene contains three developmentally regulated alternative exons (i.e., exons 2, 3, and 10), and different splicing patterns can result in six major tau isoforms. Exons 2 and 3 encode two positively charged N-terminal inserts in the tau protein. Exon 2, included alone or concomitant with exon 3, produces 1N and 2N tau isoforms, respectively. Exon 10 encodes the second of four negatively charged microtubule-binding repeats, and its AS generates tau containing three (3R) or four repeats (4R), which differ in microtubule-binding affinity.²² Embryonic human brains express only the 0N3R tau isoform, which skips exons 2, 3, and 10. Adult human brains maintain a balance of tau isoforms with 0N/1N/2N inserts and 3R/4R, with 3R and 4R isoforms expressed at about an equal molar ratio.^23,24,25 The splicing of MAPT exon 10 has been studied extensively in the literature, as mutations within or around this exon cause certain familial forms of FTLD.^26,27 Of particular importance, a subset of these mutations are synonymous or located in the flanking intronic regions, which do not affect protein sequences directly but rather lead to an increase of exon 10 inclusion, suggesting that perturbation in the balance of 3R/4R tau isoform ratio is sufficient to cause the disease (see review²⁵). In addition, an increase of 4R tau that results in nuclear tau aggregation was also observed in Huntington’s disease (HD).²⁸ Intriguingly, exon 10 is initially skipped in embryonic mouse brains as in humans. However, it increases to almost complete inclusion in the adult mouse brain,²⁹ and, despite the predominant 4R tau expression, it does not result in neurodegeneration. It is unclear when the human-mouse splicing divergence arose during evolution.

To address these questions, we modeled the alternative cassette exon inclusion in the primate brain as a quantitative trait using the Ornstein-Uhlenbeck (OU) process to characterize random drift and selection constraints within the phylogenetic tree. We identified over a thousand cassette exons with lineage-specific splicing shifts under stabilizing or adaptive selection. Surprisingly, we found a two-step evolutionary change in the splicing of MAPT exons 2 and 10 in the primate brains. MAPT exon 2 and exon 10 inclusion levels are conserved in New World monkey, prosimian monkey, and mouse brains. The two-step evolutionary change led to the reduction of exon 10 inclusion and the increase of exon 2 inclusion in Old World monkeys, and the change progressed further in hominoids. The splicing of these two exons is negatively correlated across different lineages, suggesting that they co-evolved to provide an adaptation to new functional constraints. Furthermore, our mechanistic studies focusing on MAPT exon 10 splicing revealed that mutations previously implicated in FTLD and differing between humans and mice do not explain this evolutionary divergence. Instead, divergent, distal regulatory sequences in intron 10 recognized by muscleblind-like (MBNL) proteins, critical for the exon 10 developmental splicing switch, are a major contributor to the evolutionary shifts in the splicing pattern. Building on these findings, we also demonstrated that exon 10 splicing and tau isoform ratios can be effectively modulated by targeting the MBNL-binding sites using guide RNAs (gRNAs) in complex with CRISPR-dCas13d. This strategy is potentially relevant in the treatment of FTLD and other tauopathies.

Results

Systematic identification of lineage-specific splicing shifts under stabilizing selection

To study the quantitative changes of AS during the evolution of the brain, we quantified the exon inclusion levels (percentage spliced in [PSI] or Ψ) of over 42,000 cassette exons using RNA-seq data with high read coverage from seven primate species, including two hominoids (human and chimpanzee), three Old World monkeys (rhesus macaque, crab-eating macaque, and baboon), and two New World monkeys (marmoset and squirrel monkeys) (see STAR Methods). At least two independent samples were available for each species, allowing for assessment of intra-species variation, and exons with minimal splicing variation across species were excluded in the downstream analysis (Tables S1 and S2). Unsupervised clustering readily identified groups of exons with differential splicing in specific lineages, such as exons with increased or decreased inclusion levels in humans and chimpanzees compared to the other primate species (Figure S1).

To identify divergent AS events with a potential functional impact, we used an OU model to distinguish changes in selective constraints that result in a lineage-specific splicing shift under stabilizing selection in a phylogenetic tree against random drifts modeled as Brownian motion (Figure 1A; see STAR Methods). For each exon, five statistical tests were performed, including comparisons between humans vs. non-humans, hominoids vs. non-hominoids (i.e., Old World and New World monkeys), and catarrhines (hominoids + Old World monkeys) vs. New World monkeys. Tests were performed in the phylogenetic tree of all seven primate species or the five catarrhine species for human- or hominoid-specific splicing shifts. Overall, we identified 1,170 exons showing a significant shift in at least one test (false discovery rate [FDR] < 0.1; Figures 1B and S2A; Table S1). Among them, the largest number of exons showed a splicing shift in hominoids (n = 725 compared with Old World monkeys and n = 541 compared with all other primates; Figure S2B). Simulation analysis confirmed that the OU model was well calibrated (Figure S3A) and that it had reasonable statistical power (e.g., over 80% power to detect a shift in exon inclusion level |ΔΨ| > 0.25 in hominoids as compared to Old World monkeys or non-hominoid primate species at p = 0.05; Figure S3B).

Figure 1.

Global identification and characterization of exons with lineage-specific splicing shifts under stabilizing selection in the primate brain

(A) Schematic illustration of testing lineage-specific splicing shifts using an Ornstein-Uhlenbeck (OU) model.

(B) Heatmap showing adult brain splicing profiles for exons with lineage-specific splicing shifts in humans, hominoids (human + chimpanzee), and catarrhines (hominoid + Old World monkeys) compared to the other primate species. Each exon is a row, and each sample is a column. The exon inclusion level (percentage spliced in, Ψ) is centered by the mean across samples. The number of significant exons (false discovery rate [FDR] ≤0.1) and MAPT exon 10 are indicated on the right.

(C) Percentage of exons with lineage-specific splicing shifts showing developmental splicing switches. The percentage among all cassette exons is used for control. Lineage-specific exons tested within catarrhines are indicated in parentheses by “in C.”

(D) Gene Ontology (GO) terms associated with exons showing lineage-specific splicing shifts. Significant GO terms were identified using genes with splicing shifts in each specific lineage or the combined gene list. Only GO terms that are significant (Benjamini FDR ≤0.05) in at least one test are shown and color coded. Gray indicates non-significant GO terms for the respective gene list.

(E) Enrichment or depletion of splicing-regulatory elements associated with exons showing a lineage-specific increase (left) or decrease (right) in exon inclusion. ESE/ESS, exonic splicing enhancer/silencer; ISE/ISS, intronic splicing enhancer/silencer; EIE/IIE, exonic/intronic identity element; IIE5ss/IIE3ss, IIE associated with 5′ or 3′ splice sites. The dotted lines indicate p < 0.05.

(F) Concordance between splicing changes from RNA-seq quantifications and predicted changes in splice site strengths for exons displaying lineage-specific splicing shifts. Splice site strengths were predicted either by MaxEntScan based on splice site motifs or by SpliceAI based on both local and distal pre-mRNA sequences. Different subsets of exons with varying magnitude in splicing divergence were used for analysis (ΔΨ ≥ 0.1–0.4). ND, not determined due to lack of mutations in splice site motifs.

See also Figures S1–S4 and Tables S1, S2, and S3.

About 50%–60% of exons with lineage-specific splicing shifts maintain the reading frame, which is comparable to all cassette exons used as a control (Figure S4A). Exons displaying a more recent shift (e.g., human vs. all other primates) less frequently have conserved AS patterns in mice. However, the frequency is nevertheless higher compared to all cassette exons, suggesting that many of these exons with splicing shifts have a relatively ancient origin (Figure S4B). Moreover, we found that the exons with lineage-specific shifts are more frequently under developmental-stage-specific regulation; this is especially prominent for exons with human-specific shifts (Figure 1C). Gene Ontology (GO) analysis of genes harboring exons with human-specific shifts suggests a significant enrichment of genes involved in cilia assembly and morphogenesis (12 genes, 8-fold enrichment, p < 5E−3 after Benjamini correction; Figure 1D; Table S3) in addition to GO terms reflecting general features of alternatively spliced genes (e.g., “protein binding”). Primary cilia are protrusions from the surface of most mammalian cells that function as sensory organelles, which play an important role in multiple aspects of brain development, including the expansion of cortical neural progenitor pools. Mutations in several of these cilial genes cause a range of multi-system developmental disorders known as ciliopathies^30,31 (Table S3). These data support the notion that lineage-specific AS contributes to not only human-specific brain development but also human-specific susceptibility to brain developmental disorders.

To further evaluate the reliability of the identified lineage-specific splicing shifts and gain mechanistic insights, we asked whether these splicing changes could be predicted from changes in splicing-regulatory signals, as previous studies have suggested that divergence in cis-acting splicing-regulatory elements is likely the major driving force of splicing divergence in mammals.^11,32 We expect that mutations that create splicing enhancers and/or disrupt splicing silencers likely lead to an increase in exon inclusion, while mutations that disrupt splicing enhancers and/or create splicing silencers likely lead to a decrease in exon inclusion (Figure 1E, upper panels). Indeed, when we examined the frequency of putative splicing-regulatory elements in the cassette exons and flanking intronic sequences, we found that a positive shift in exon inclusion is associated with an increase in the density of splicing enhancer elements and a decrease in the density of silencer elements in the respective lineage (Figure 1E, bottom panels). We also examined whether changes in the splice site strengths, predicted by using either MaxEntScan (an entropy-based method that scores the consensus splice site motifs)³³ or SpliceAI (a deep learning method that predicts splice site strength based on the extended primary RNA sequences),³⁴ were consistent with changes in exon inclusion levels. In a majority of the cases, the predicted change in splice site strength is consistent with the change in exon inclusion. MaxEntScan predicted ∼35% of exons showing consistent splicing changes vs. ∼18% of exons showing inconsistent exons, while ∼46% of exons do not have mutations in the splice site motif (i.e., among exons with changes in splice site motifs, 66% show splicing shifts in the predicted direction). Similarly, SpliceAI predicted ∼70% of exons with consistent splicing changes vs. ∼30% with inconsistent splicing changes (Figure 1F). The concordance between splicing changes and genomic sequence divergence supports the reliability of the exons we identified. These data suggest that mutations affecting both the splice site motifs and splicing-regulatory elements outside the splice sites contribute to the lineage-specific splicing shifts in primates.

A two-step lineage-specific splicing shift in MAPT exons 2 and 10

To dissect the underlying regulatory mechanisms that drive divergent splicing and the potential functional impact in depth, we focused on the MAPT gene, which contains multiple developmentally regulated alternative exons, exons 2, 3, and 10 (Figures S5A and S5B). Analysis using OU models showed a significant shift in exon 10 inclusion in hominoids as compared to Old World monkeys (FDR < 0.05) and all other primate species (FDR < 0.1) (Figure 1B). Exon 2 was excluded in formal OU modeling because of missing exon-level quantification values due to insufficient read coverage in marmoset and squirrel monkey data. However, exon 2 inclusion levels showed dramatic divergence across humans, non-human primates, and mice (Figures S5A, S5B, and 2A below).

Figure 2.

MAPT exon 10 splicing undergoes a two-step evolutionary shift during primate evolution

(A) Top: the genomic locus where the MAPT gene is located and the MAPT gene structure. The major alternative exons 2, 3, and 10 are highlighted. Bottom: adult brain inclusion level of exon 2 and exon 10 in different lineages of primate species as quantified using RNA-seq data. Error bars in the bar plots represent the standard error of the mean (SEM). The source of the RNA-seq data is indicated in parentheses: L, Lister et al.³⁵; N, NHPRTR³⁶; B, Brawand et al.³⁷; W, Weyn-Vanhentenryck et al.⁶.

(B) Anti-correlation of exon 2 and exon 10 inclusion across species. Error bars represent SEM.

(C) A stem-loop structure at the border of exon 10 and intron 10 that overlaps with the 5′ splice site was previously predicted to repress exon 10 inclusion. This stem-loop structure is predicted to be destabilized by mutations in FTLD patients (red nucleotides) and in mice (green nucleotides), increasing in exon 10 inclusion.

(D) Multiple alignments of sequences near the 5′ splice site of exon 10 containing the predicted stem-loop structure. Note the perfect conservation of the stem-loop structure in primates (shaded).

See also Figure S5.

The lineage-specific splicing shifts of MAPT exons 2 and 10 detected in primates prompted us to perform an extended analysis to determine the evolutionary history of this splicing pattern divergence. Specifically, we examined exon inclusion in additional primate species with available adult brain RNA-seq data, including three hominoids (bonobo, gorilla, and orangutan) and mouse lemur, resulting in a total of 11 primate species analyzed (Figure 2A); mouse brain was also included as an outgroup. This analysis revealed that MAPT exon 10 splicing underwent a two-step lineage-specific splicing shift in primates during evolution. In New World monkeys and mouse lemur, the exon is almost completely included (>90%), while it reduces to 75%–78% in Old World monkeys and ultimately to 41%–54% in hominoids (Figure 2A). Intriguingly, exon 10 inclusion level is inversely correlated with exon 2 level across primate species, which is about 37%–59% in hominoids, 24%–28% in Old World monkeys, and 17% in marmoset (and 12% in mouse) (Figure 2 A and B). Together, these data suggest the co-evolution of exon 2 and exon 10 splicing regulation and the potential adaptation of tau function in different lineages.

An extensive effort has been made to identify splicing-regulatory elements involved in MAPT exon 10 splicing as disease-causing mutations increase its inclusion. Multiple serine and arginine-rich (SR) proteins and heterogeneous nuclear ribonucleoproteins (hnRNPs) were shown to regulate exon 10 splicing by binding to splicing enhancers and silencers in or near exon 10, and some of these elements overlap with FTLD-associated mutations (summarized in Figure S6 and reviewed in Andreadis²⁵ and Qian and Liu³⁸). In addition, it has been previously proposed that the sequence bordering exon 10 and intron 10 forms a stem-loop structure that functions as a splicing repressor by blocking the accessibility of the 5′ splice site.^{26,27,32,39,40} FTLD-associated mutations overlapping with this region were predicted to destabilize this structure, increasing exon 10 inclusion in those patients. Furthermore, several nucleotide positions that differ in the orthologous sequence of mouse Mapt were also predicted to destabilize the structure and thus could explain the higher exon 10 inclusion in the adult mouse brain^40,41,42 (Figure 2C). However, we found that the entirety of exon 10 (except for one nucleotide mutation in green monkey and another in bushbaby) and the predicted stem-loop structure are conserved in all sequenced primate species (Figures 2D and S6), disqualifying these splicing-regulatory elements from explaining the observed splicing divergence across primates. In addition, given the conservation of exon 10 inclusion levels in New World monkeys, mouse lemur, and rodents, the mutations disrupting the stem-loop sequence in mice alone are insufficient to cause the splicing pattern divergence observed between humans and mice. Lastly, these previously identified mechanisms do not explain the drastic splicing switch during brain development, suggesting the presence of additional critical splicing-regulatory elements.

Divergence in MBNL-dependent regulation is a major driving factor of MAPT exon 10 lineage-specific splicing

In searching for the regulatory mechanisms driving the lineage-specific splicing shift of MAPT exon 10, we noted that the exon is almost completely skipped in the embryonic brain in both humans and mice, despite the drastic difference in exon inclusion in the adult brain of the two species (Figure 3A). Also, the 3R isoform is more abundant in newborn marmosets, although earlier time points were not examined.⁴³ We thus hypothesized that divergence in splicing-regulatory elements important for developmental splicing switches of exon 10 may contribute to its splicing pattern divergence. Based on the following observations, we focused on the MBNL splicing factors. We previously determined that MBNL1 and MBNL2 (referred to as MBNL1/2), whose expression increases throughout brain development, contribute to the developmental splicing switch of exon 10, as MBNL depletion in the mouse brain leads to a more “embryonic brain-like” splicing pattern with a reduction in exon 10 inclusion.⁴⁴ Similarly, exon 10 inclusion was reduced in the postmortem brains of patients with myotonic dystrophy (DM), in which MBNL proteins are functionally sequestered due to an expansion of repetitive RNA containing MBNL-binding sites.⁴⁵

Figure 3.

The divergence in developmentalstage-specific regulation by MBNL drives the MAPT exon 10 splicing shift between human and mouse

(A) Cassette exons with conserved, mouse-specific, or human-specific late developmental splicing switches (left). Regulation of these exons by MBNL, as determined by comparison of wild-type (WT) vs. Mbnl1/2 double knockout (DKO) brains in mice as well as control and myotonic dystrophy type 1 (DM1) brains in humans is shown (right). Exons with MBNL-dependent inclusion are indicated by red bars, while exons with MBNL-dependent skipping are indicated by blue bars on the right. Note that most of these exons with developmental increases are activated by MBNL, which is consistent with the MBNL expression pattern. The proportion of exons in each group that are MBNL targets is also indicated below each heatmap. pcw, post-conception weeks.

(B) Changes of MBNL2 mRNA expression level (left axis) and MAPT exon 10 inclusion (right axis) in human cortices at different developmental stages (left) as quantified by RNA-seq. Exon 10 inclusion in control and DM1 patient brains is also shown and the statistical significance of the difference is indicated (right).

(C) Changes of Mbnl2 mRNA expression level (left axis) and Mapt exon 10 inclusion in mouse cortex at different developmental stages (left) as quantified by RNA-seq. Exon 10 inclusion in WT and Mbnl1/2 DKO brains is also shown and the statistical significance of the difference is indicated (right).

(D) MBNL expression drives exon 10 inclusion in human and mouse MAPT minigenes. Exon 10 inclusion in human and mouse MAPT minigenes (hMAPT and mMAPT) in HEK293T cells transfected with empty pcDNA vector (pcDNA), cells with MBNL1/2 knockdown using siRNAs (siMBNL1/2), and cells with overexpression of FLAG-MBNL2 (pFLAG-MBNL2). Representative images show the confirmation of protein expression using immunoblots (control: GAPDH) and exon 10 inclusion using RT-PCR. Quantification of exon 10 inclusion (mean ± SEM from n = 3 replicates) is shown at the bottom.

See also Figures S6–S9.

To investigate the MBNL-dependent splicing regulation and its impact on the divergence of developmental splicing switches, we performed RNA-seq to identify the MBNL target exons by comparing mouse wild-type (WT) and Mbnl1; Mbnl2 double-knockout (DKO) brains, as well as human control and DM type 1 (DM1) brains. We previously demonstrated that MBNL is a master regulator of the second of two waves of developmental splicing switches (i.e., late splicing switches),⁶ which occur during cortical development, corresponding to the first few months post birth in humans and between postnatal day 4 (P4) and P15 in mice. During this period, 42% and 65% of exons with conserved developmental splicing switches in humans and mice, respectively, rely on MBNL and exhibit a direction of MBNL-dependent splicing consistent with that of developmental switches (Figure 3A). Importantly, for exons with human-specific developmental changes and no switches in mice, there is a significant reduction in MBNL-dependent splicing in mice (18% vs. 65%, p = 1.6E−6, Fisher’s exact test) but not in humans (42% vs. 42%, p = 1, Fisher’s exact test), when compared with exons exhibiting conserved splicing switch patterns. For exons with mouse-specific developmental changes and no switches in humans, there is a significant reduction of MBNL-dependent splicing in humans (7% vs. 42%; p = 5E−5, Fisher’s exact test) but not in mice (60% vs. 65%, p = 0.55, Fisher’s exact test) (Figure 3A). These data suggest that divergence in MBNL-dependent regulation significantly contributes to the divergence of developmental splicing switches in mammals.

When we focused on MAPT exon 10, multiple lines of evidence suggested that MBNL is a dominant regulator of its developmental splicing switch. First, the exon 10 splicing switch occurs precisely at the time point when the late developmental splicing switch occurs, and the human-mouse divergence is maximized in the adult brain. The developmental splicing changes coincide with the increased expression of MBNL1/2, especially MBNL2 (Figures 3B and 3C). Second, we found that the exon 10 developmental splicing switch occurs uniformly across different brain regions (Figure S7A), consistent with the broad expression pattern of MBNL. Third, exon 10 is consistently included at a high level (∼95%) across diverse neuronal and glial cell types in the adult mouse cortex (Figure S7B), consistent with the expression and activity of MBNL in both neurons and glia. In line with this notion, a developmental splicing switch of exon 10 was also observed in rodent oligodendrocytes.⁴⁶

The magnitude of MBNL-dependent splicing activation is also greater in mice than in humans (ΔΨ = 0.23 in humans and 0.56 in mice; Figures 3B and 3C), which led us to hypothesize that the attenuation of MBNL-dependent regulation may explain the lineage-specific splicing shifts in hominoids and Old World monkeys when compared to more distally related primates and rodents. However, it is important to note that exon 10 skipping is incomplete in Mbnl1/2 DKO mouse brains or DM human brains, likely due to an incomplete depletion or functional sequestration of MBNL. To examine the range of exon 10 inclusion that can be driven by different levels of MBNL expression and potential species differences, we generated human and mouse MAPT (hMAPT and mMAPT) minigenes encompassing exon 9 to exon 11. These minigenes contain full-length sequences of the downstream intron, in contrast to previous exon 10 splicing minigenes that contain minimal intronic sequences.^32,47 Each minigene was transfected into HEK293T cells, which, compared to the brain, have a relatively high expression level of MBNL1 but low levels of MBNL2 (Figures S8A and S8B). In this system, exon 10 is included at 11%–25% for hMAPT and 24%–54% for mMAPT (Figures 3D and 4 below), which are comparable to levels observed in developing brains. Notably, knockdown of MBNL1/2 by co-transfection of specific siRNAs led to nearly complete skipping of both hMAPT and mMAPT exon 10 (<2%; p < 0.01 for hMAPT and p = 2.2E−4 for mMAPT), while overexpression of a FLAG-tagged MBNL2 (FLAG-MBNL2) increased the exon inclusion level to 40%–58% for hMAPT (p = 1.8E−4) and 60%–75% for mMAPT (p = 3.1E−4) (n = 3 per group, two-sided t test; Figure 3D; see also Figures 4 and S9). These data suggest that MBNL-dependent regulation can largely explain the range of MAPT exon 10 inclusion levels observed during brain development, and the extent of this regulation is attenuated in human compared to mouse.

Figure 4.

Divergence in the distal MBNL-binding sites in the downstream intron 10 contributes to the lineage-specific shifts of MAPT exon 10 splicing

(A) MBNL-binding sites around MAPT exon 10 as identified by MBNL2 CLIP data and bioinformatically predicted YGCY (Y = C/U) clusters. Shaded boxes highlight the two major distal binding sites in the downstream intron 10. Multiple alignments of site 2 sequences in five representative primate species analyzed with mMAPT and hMAPT minigene splicing reporters are shown at the bottom, with YGCY elements highlighted.

(B and C) MBNL-binding site deletion (B) or replacement (C) experiments using mMAPT minigene with (pFLAG-MBNL2) or without (pcDNA) MBNL2 overexpression. Representative gel images show confirmation of protein expression using immunoblots (control: GAPDH) and exon 10 inclusion using RT-PCR. Quantification of exon 10 inclusion (mean ± SEM from n = 3 replicates) is shown at the bottom.

(D and E) MBNL-binding site 1 or 2 replacement experiments using hMAPT minigene with (pFLAG-MBNL2) or without (pcDNA) MBNL2 overexpression. (D) Replacement of site 1 or site 2 from human to mouse sequences. (E) Replacement of human site 2 to sequences from different primate species and mouse. Representative gel images show confirmation of protein expression using immunoblots (control: GAPDH) and exon 10 inclusion using RT-PCR. Quantification of exon 10 inclusion (mean ± SEM from n = 3 replicates) is shown at the bottom.

See also Figures S6–S9.

MAPT exon 10 splicing divergence is modulated by divergence in MBNL-binding sites

We next searched for splicing-regulatory elements required for MBNL-dependent inclusion of MAPT exon 10. In particular, we examined whether the divergence of these elements can explain the splicing divergence observed both across primate species and between primates and mice. MBNL binds to clusters of YGCY (Y = C/U) elements to regulate AS and typically activates exon inclusion by binding in the downstream intron.^48,49,50 Leveraging the MBNL2 crosslinking and immunoprecipitation (CLIP) data we previously generated from mouse and human brains^44,45 and bioinformatically predicted YGCY clusters,⁴⁸ we found two major MBNL-binding sites (denoted as site 1 and site 2) in intron 10 supported by both sets of data (Figure 4A). Site 1 is located deep in the middle of intron 10 and appears to have a higher MBNL-binding affinity in humans, while site 2 is ∼550 bp upstream of exon 11 and appears to have a stronger MBNL-binding affinity in mice. Extensive creations and losses of YGCY elements occurred in different primate lineages and may have led to differences in MBNL-binding affinity (e.g., four YGCYs in humans and eight YGCYs in mice for site 2; Figure 4A). These data suggest a possibility that divergence in the MBNL-binding sites in the downstream intron can alter the magnitude of MBNL-dependent splicing activation and thus explain the lineage-specific splicing shifts observed in adult brains.

To validate the importance of the identified MBNL-binding sites in the regulation of exon 10 splicing and their contribution to lineage-specific splicing shifts, we performed a series of mutagenesis experiments in the predicted MBNL-binding sites using the hMAPT and mMAPT minigenes. We first generated truncation mutants by eliminating MBNL-binding site 1, site 2, or both in the mMAPT minigene (Figure 4B). The WT or mutant minigene reporters were transfected into HEK293T cells, with or without co-transfection of FLAG-MBNL2 (n = 2 per group). Individual deletion of site 1 or site 2 reduced exon 10 inclusion from 42% to about 15% (p = 0.06, two-sided t test), and deletion of both sites resulted in a further reduction to <9% (p = 0.04, two-sided t test; Figure 4B). Overexpression of MBNL2 increased exon 10 inclusion in all the minigene constructs, especially for the WT and single-site deletion mutants, data consistent with MBNL-dependent exon activation (Figure 4B). Deleting either site 1 or site 2 resulted in a partial reduction of exon 10 inclusion (p = 0.03 and 0.02, respectively; two-sided t test), as compared with the WT minigene, and the largest reduction was again observed when both binding sites were deleted (p = 0.003, two-sided t test). These data suggest that both site 1 and site 2 are important for MBNL-dependent splicing regulation of exon 10. Moreover, the MBNL-dependent increase in exon 10 inclusion is diminished for the double-deletion mutants, suggesting that these two sites represent the most critical regulatory elements for MBNL-dependent splicing regulation.

We hypothesized that site 2 might be more relevant for the lineage-specific splicing shifts of exon 10 across species, as MBNL is predicted to have weaker binding sites in humans than in mice. This pattern is consistent with the divergent exon inclusion levels and the relative proximity of site 2 to the downstream exon. To test this hypothesis, we replaced the mouse site 2 sequence with the orthologous human sequence (Figure 4A). Interestingly, exon 10 inclusion levels in the chimeric mMAPT minigene with the human site 2 sequence did not rescue the reduction of exon inclusion caused by the site 2 deletion in HEK293T cells, with or without MBNL2 overexpression (p = 0.94 and 0.49, respectively, n = 3 per group, two-sided t test), confirming that the human MBNL-binding site 2 has reduced activity compared to the mouse counterpart (Figure 4C).

Next, we performed additional MBNL-binding site replacement experiments using the hMAPT minigene (n = 3 replicates per group). Replacing the human site 2 sequence with the orthologous mouse sequence increased exon 10 inclusion (p = 0.006, two-sided t test), although the difference is more subtle upon MBNL2 overexpression (p = 0.21, two-sided t test). These data are consistent with our hypothesis that mouse site 2 has a higher binding affinity to MBNL than human site 2 (Figure 4D). Interestingly, the effect of site 1 replacement appears to depend on the expression level of MBNL. Without overexpression of MBNL2, the chimeric minigene with human site 1 replaced with the mouse sequence exhibited reduced exon 10 inclusion levels (p = 0.004, two-sided t test). However, upon MBNL2 overexpression, which resulted in MBNL expression levels similar to those in the adult brain, the site 1 replacement led to a slight increase, if any, in exon 10 inclusion (Figure 4D). Together, these experiments confirmed that the MBNL-binding sites, especially site 2, are more potent in activating exon 10 inclusion in the mouse Mapt gene than in the human MAPT gene.

We then examined how the sequence divergence of site 2 affects exon 10 splicing divergence in primates. Specifically, we replaced the site 2 sequence in the hMAPT minigene with its orthologous sequences from rhesus macaque (a representative of Old World monkeys) and marmoset and squirrel monkey (representatives of New World monkeys) (n ≥ 2 replicates per group). Replacement of the human site 2 with marmoset and squirrel monkey sequences increased exon 10 inclusion compared to the WT hMAPT minigene without MBNL2 overexpression. Although the significance level of the increase is marginal due to some technical variability (p = 0.06, two-sided t test; p = 0.03 for marmoset and p = 0.02 for squirrel monkey, if pairwise t test was used to control systematic technical variability across replicates), it is again consistent with our hypothesis. The difference diminished upon MBNL2 overexpression. Replacement with the rhesus monkey sequence did not significantly affect this experimental setup without MBNL2 overexpression (p = 0.93 and 0.68, t test; Figure 4E). Overall, these data support our model that attenuation of MBNL-dependent splicing regulation caused by the weakening of MBNL-binding sites is a major contributing factor to the lineage-specific splicing shift of MAPT exon 10 in hominoids and Old World monkeys.

Modulation of MAPT exon 10 splicing by steric hindrance using CRISPR-dCas13d/gRNAs

Given the clinical relevance of MAPT exon 10 inclusion in FTLD, Huntington’s disease, and potentially other tauopathies, previous studies have attempted to modulate exon 10 splicing by targeting splice sites and/or exonic splicing-regulatory elements using antisense oligonucleotides (ASOs)⁵¹ or the RNA-targeting CRISPR system composed of the enzymatically dead Cas13d (dCas13d) in complex with the gRNAs.^52,53 We thus tested whether the identified MBNL-binding sites can be used to modulate exon 10 splicing. We used mMAPT as a model for this experiment since its exon 10 inclusion levels are more comparable to FTLD patients than the WT human exon 10 reporter. Specifically, we used dCas13d/gRNA to target site 2 sequences in the mMAPT minigene, block MBNL binding by steric hindrance, and thus inhibit MBNL-mediated splicing. A non-targeting (NT) gRNA was used as a control. Four gRNAs were designed to target the MBNL site 2 sequence containing YGCY elements (Figure 5A; Table S4). These gRNAs, individually or all together, were co-transfected with the mMAPT minigene in HEK293T cells (n = 3 replicates per group). We found that the expression of dCas13d/gRNAs effectively reduced exon 10 inclusion in comparison to the control gRNA, with the strongest inhibition observed when gRNA3 or all four gRNAs were expressed together (p = 0.07 for gRNA1 and 0.01 for gRNA 4, two-sided t test; Figure 5B). When the experiment was performed with overexpression of MBNL2, gRNA 3 was also able to reduce exon 10. Notably, co-expression of all four gRNAs most effectively decreased exon 10 inclusion from about 60% to 27% (p = 0.002, two-sided t test). These data confirm that MBNL-binding sites can be targeted to modulate exon 10 splicing efficiently.

Figure 5.

Modulation of MAPT exon 10 splicing by steric hindrance of MBNL-binding sites using dCas13d/gRNAs

(A) Schematic of splicing modulation using dCas13d/gRNAs targeting specific sequences in distal intronic MBNL-binding site 2. Four gRNAs targeting MBNL-binding YGCY elements in site 2 were tested in HEK293T cells. A non-targeting (NT) gRNA was used for control.

(B) Exon inclusion with or without dCas13d/gRNA targeting using individual gRNAs or all four gRNAs together. Representative immunoblot images show confirmation of protein expression (control: GAPDH) and exon 10 inclusion using RT-PCR. Quantification of exon 10 inclusion (mean ± SEM from n = 3 replicates) is shown at the bottom.

Discussion

In this study, we investigated divergent AS in primate brains and demonstrated that ∼3% of cassette exons (representing ∼900 genes) show evidence of lineage-specific splicing shifts under stabilizing selection. Our focus on exons under selective constraints, differing from previous studies aimed at identifying divergent AS events (e.g., several reviews^{10,11,12,13,14,15}), maximizes the chances of finding evolutionary innovations with functional significance. In our analysis, the largest group shifted splicing in hominoids, correlated with the brains’ rapid evolution and functional/structural expansion in this lineage.⁵⁴ These exons, especially those showing human-specific splicing shifts, tend to be regulated during brain development, indicating their potential functional relevance. Mechanistically, the splicing shifts can be explained by mutations in the splice sites and numerous splicing-regulatory elements outside the splice sites, although precise identification of mutations that cause divergent splicing of individual exons is currently challenging.

This study led to the unexpected finding of the splicing divergence in the MAPT gene. Since it was first cloned in 1975 as a microtubule-binding protein,⁵⁵ MAPT/tau and its splicing variants have been the subject of intensive investigations in the context of neurodegeneration. However, after five decades of studies, molecular mechanisms controlling tau expression are poorly understood, and the physiological function of MAPT remains enigmatic. The disruption of MAPT exon 10 splicing in FTLD and its divergence between humans and mice have been well documented. In addition, exon 10 inclusion levels in chimpanzee, gorilla, gibbon, and marmoset have been analyzed in separate studies (e.g., Sharma et al. and Holzer et al.^43,56), reaching conflicting conclusions regarding the divergence of splicing patterns and genomic sequences. To our knowledge, the evolutionary divergence of exon 2 has yet to be noted in the literature. By systematic cross-species comparison of 12 primate species using uniform data sources and analysis pipelines, this study demonstrated that the MAPT gene exhibits drastic two-step splicing shifts during primate evolution. Exon 10 decreases in Old World monkeys and hominoids, while exon 2 exhibits the inverse pattern. As a result, the inclusion levels of these two exons are inversely correlated across primate species. The pattern of the splicing shifts is consistent with the phylogenetic relationship of the compared species, and regulatory mechanisms involved in splicing these two exons likely co-evolve.

Examining MAPT exon 10 splicing-regulatory mechanisms from the evolutionary perspective allowed us to identify previously uncharacterized splicing-regulatory elements. It is clear that the previously identified splicing-regulatory elements, including those in exon 10 recognized by multiple SR proteins and hnRNPs or the stem-loop structure at the border of exon 10 and intron 10,^{25,40,41,42,57} are insufficient to explain the developmental-stage- and lineage-specific splicing switch of this critical exon. While we and others have previously demonstrated that MBNL regulates exon 10 splicing,^44,45 the extent to which such regulation affects the developmental splicing switch is unclear. Our data provide compelling evidence that the dynamic expression of MBNL does not simply contribute to but largely drives the developmental splicing changes of MAPT exon 10 during brain development. Through systematic mutagenesis analysis, we further determined that the regulation is mediated through two distal binding sites in the downstream intron, whose divergence contributes significantly, if not entirely, to the lineage-specific splicing pattern in primates. This discovery creates an opportunity to effectively reduce exon 10 inclusion by targeting the MBNL-binding sites, which may find applications in clinical settings, where modulating the 3R/4R tau ratio could ameliorate tau pathology. However, it is important to note that our data do not argue against the importance of previously identified splicing-regulatory elements for splicing regulation or their relevance in aberrant splicing observed in FTLD patients. Instead, it is likely that alterations of these elements by FTLD-causing mutations, together with MBNL-dependent splicing activation, result in the increased exon 10 inclusion level observed in FTLD. The impact of several FTLD-causing mutations that increase 4R tau, such as N279K, P301L, and 10+16C>T, has been modeled using neurons differentiated from patient-derived induced pluripotent stem cells (iPSCs).^58,59 It has been demonstrated that these mutations accelerate the expression of the exon 10 inclusion isoforms during neuronal maturation in vitro. However, owing to the immaturity of in vitro neurons, they predominantly express exon 10-skipping isoforms even after extended culture. Therefore, exon 10 is not constitutively included in these patients but is most likely developmentally regulated by MBNL in these patients. Targeting MBNL-binding sites could serve as a viable approach to modulating 3R/4R tau ratio in these patients with FTLD and other tauopathies. In this study, we provide proof of the principle of this strategy by targeting MBNL-binding sites using dCas13d/gRNA.

What are the functional implications of the lineage-specific splicing regulation of exons 2 and 10? Exon 10 encodes the second microtubule-binding repeat, and its varying exon inclusion level likely contributes to varied microtubule-binding affinities in different species (Figures 6A and 6B). However, the functional roles of exon 2 (and the nearby exon 3) are less well characterized. While tau is a highly soluble protein, it undergoes liquid-liquid phase separation in vitro and in cells in specific conditions, which might be instrumental in its physiological function with implication in the formation of tau aggregation during neurodegeneration.^60,61,62 Recent studies demonstrated that the bipolar charge distribution of tau, with a positively charged N-terminal region and a negatively charged C-terminal region, plays a critical role in its phase separation through electrostatic interactions^60,63 (Figure 6B). Importantly, compared to truncation mutants lacking the positively charged N insert or the negatively charged microtubule repeats, the full-length 2N4R tau isoform, resulting in extended regions for electrostatic interactions, is the most prone to phase separation. This suggests that the distinct biochemical properties of different tau isoforms can modulate phase separation.⁶⁰ We note that, due to the splicing divergence of both exon 2 and exon 10, the abundance of 1N4R and 2N4R isoforms varies in the brain of different species, with the highest abundance found in hominoids (∼25%), compared to Old World monkeys (∼19%) and New World monkeys/mouse (10%). The abundance difference implies that the propensity of tau phase separation and pathological aggregation differs in these lineages (Figure 6C). This notion can potentially explain two intriguing observations in the literature. First, in a humanized tau mouse model, replacing the mouse Mapt gene with a human MAPT gene, which recapitulates human MAPT splicing patterns of exons 2 and 10, accelerates pathological tau aggregation and propagation upon induction with exogenous pathological tau.⁶⁴ Second, the aging brain of rhesus macaques (Old World monkeys) is much more vulnerable to neurotoxicity upon induction of pathological amyloid-β protein than that of marmosets (New World monkeys) and rats; this vulnerability is evident in the presence of amyloid β plaques and hyperphosphorylated tau.⁶⁵ In addition, individual or coordinated alterations of exons 2, 3, and 10 splicing have been reported in postmortem brains of AD, PSP, and HD patients (e.g., several articles^28,66,67 and references therein). These data support that tau has acquired intrinsic functional differences during evolution, leading to distinct disease susceptibilities in different species.

Figure 6.

The proposed model of MAPT functional adaptation by lineage-specific exon 10 splicing shifts

(A) Schematic of the model showing that the formation of tau liquid-liquid phase separation (LLPS) or NFT is dictated by the type of intermolecular interactions. The transition between different states can be modulated by the type of tau isoforms or post-translational modifications affected by mutations or other cell stressors. The negatively charged and positively charged regions in tau mediating electrostatic interactions are shaded in blue and red, respectively.

(B) Reduction of exon 10 inclusion in Old World monkeys and hominoids is expected to alter tau’s microtubule-binding affinity during neurodevelopment and homeostasis.

(C) Co-evolution of exon 2 and exon 10 splicing resulted in an increase in the full-length (1N4R or 2N4R) tau isoforms in the adult hominoid brain. These isoforms have a higher propensity of phase separation and susceptibility to aggregation upon disease-associated mutations or stresses.

Despite these translational implications, we note that the evolutionary divergence of MAPT AS is unlikely to be driven by the involvement of tau in adult-onset neurodegeneration but rather by its impact on neurodevelopment and homeostasis. As a conserved microtubule-binding protein involved in microtubule stability and axonal transport, its functional adaptation may underlie certain aspects of brain evolution, such as the expansion in size or specific requirements on synaptic plasticity. The consequence of tau depletion in different species is worth noting. In humans, a 17q21.3 microdeletion encompassing the MAPT gene causes severe developmental delays and intellectual disability.^68,69 However, depletion of tau in rodents, especially in haploinsufficiency models, results in moderate phenotypes,^70,71 suggesting that the essentiality of tau in different organisms has also diverged. Altogether, these data suggest functional adaptation of tau in different primate species during evolution, with critical contribution from divergent AS.

Limitations of the study

While we demonstrated the role of MBNL-binding sites in regulating developmental-stage- and lineage-specific splicing of MAPT exon 10, these elements are unlikely to function alone, but rather, they coordinate with additional regulatory sequences. Mechanistic details on how the distal MBNL-binding sites and the proximal sequences near the 5′ splice site interact are unclear. Studies of endogenous genes in different species (for example, using iPSC-derived models) will provide further validation of the results presented in this work, although such experiments are currently challenging. Another limitation of the current study concerns modulating MAPT exon 10 splicing using dCas13d/gRNA. As a next step toward translational applications, the efficacy demonstrated in vitro has to be confirmed using more therapeutically compatible approaches, such as antisense oligonucleotides (ASOs) in more physiological systems in a future study.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Primary antibody: rabbit α-MBNL1	Charles Thornton Lab72	A2746
Primary antibody: mouse α-MBNL2	Santa Cruz Biotechnology	Cat# sc-136167; RRID:AB_2140469
Primary antibody: mouse α-FLAG (Clone M2)	Sigma-Aldrich	Cat# P2983; RRID:AB_439685
Primary antibody: mouse α-GAPDH	Millipore	Cat# MAB374; RRID:AB_2107445
Bacterial and virus strains
NEB Stable Competent E. coli	NEB	Cat #C3040
Biological samples
Human: RNA-seq human cortices control	This study	N/A
Human: RNA-seq Myotonic Dystrophy type 1 (DM1) patients	This study	N/A
Mouse: genomic DNA from wildtype B6/C57 mouse	The Jackson Laboratory	JAX:000664; RRID:IMSR
Mouse: RNA-seq cortices from P4, P7, P30	Ref. 73	N/A
Chemicals, peptides, and recombinant proteins
Lipofectamine 3000	Thermo Fisher	Cat# L3000001
SuperScript III reverse transcriptase	Thermo Fisher	Cat# 18080044
TRIzol reagent	Thermo Fisher	Cat# 15596026
Directzol Miniprep Kit	Zymo Research	Cat# R2050
Fetal Bovine Serum (FBS)	Sigma-Aldrich	Cat# F0926
Dulbecco’s modified Eagle’s medium (DMEM)	Corning	Cat# MT10013CV
Gibson Assembly® Master Mix	NEB	Cat# E2611
Deposited Data
Human and mouse: Developing and adult brain RNA-seq data	NCBI SRA35	SRP026048
Non-human primate: Adult brain RNA-seq data from non-human primate reference transcriptome resource (NHPRTR)	Ref. 36	N/A
Non-human primate: Adult brain RNA-seq data additional primate species	NCBI SRA37	SRP007412
Human: Developing and adult brain RNA-seq data	dbGAP74	phs000755
Mouse: Developing and adult brain RNA-seq data	NCBI SRA73	SRP055008
Mouse: RNA-seq of wildtype and Mbnl1/2 DKO cortices	NCBI SRA6	SRP142522
Human: RNA-seq of human cortices control and Myotonic Dystrophy type 1 (DM1) patients	NCBI SRA (this study)	PRJNA1093165
Experimental models: Cell lines
Human: HEK293T cells	ATCC	CRL-3216
Oligonucleotides
Human MBNL1 siRNA:/5Phos/rUrCrUrCrUr ArCrArUrArCrUrUrCrCrArGrUrGdTdT	Ref. 75	N/A
Human MBNL2 siRNA:/5Phos/rGrArGrCrGr UrGrArGrCrArUrGrUrUrCrCrUrCdTdT	Ref. 75	N/A
Primers and gblock sequences for human and mouse MAPT minigene cloning and RT-PCR analysis, as well as gRNA sequences, see Table S4	This study	N/A
Recombinant DNA
pcDNA5/FRT	Thermo Fisher	Cat# V601020
pCAGGS-3xFLAG-MBNL2	Ref. 6	N/A
pXR002: EF1a-dCasRx-2A	Addgene52	109050
pLentiRNAGuide_001 - hU6-RfxCas13d-DR1-BsmBI-EFS-Puro-WPRE	Addgene76	138150
Software and algorithms
Quantas pipeline	Ref. 73	https://zhanglab.c2b2.columbia.edu/index.php/Quantas
OLego	Ref. 77	http://zhanglab.c2b2.columbia.edu/index.php/OLego
R package ‘pcaMethods’	Ref. 78	https://www.bioconductor.org/packages/release/bioc/html/pcaMethods.html
Expression Variance and Evolution (EVE) model	Ref. 79	https://rohlfslab.weebly.com/software.html
Gene ontology (GO) analysis using DAVID	Ref. 80	https://david.ncifcrf.gov
RESCUE ESE	Ref. 81	http://hollywood.mit.edu/burgelab/rescue-ese/
FAS-ESS	Ref. 82	http://hollywood.mit.edu/fas-ess/
Chasin-PESE and PESS	Ref. 83	N/A
Exon identity element (EIE) and intron identity elements (IIE)	Ref. 84	N/A
MaxEntScan	Ref. 33	http://hollywood.mit.edu/burgelab/maxent/Xmaxentscan_scoreseq.html
SpliceAI	Ref. 34	https://github.com/Illumina/SpliceAI

Resource availability

Lead contact

As lead contact, Chaolin Zhang is responsible for all reagent and resource requests. Please contact Chaolin Zhang at cz2294@columbia.edu with requests and inquiries.

Materials availability

All DNA and primer sequences are listed in key resources table and Table S4. Plasmid vectors are available upon request.

Data and code availability

RNA-seq data from wild-type and Mbnl1/2 DKO mouse brains, as well as control and DM1 human cortices, have been deposited to NCBI Short Read Archive (SRA) (accession: SRP142522 and PRJNA1093165).

Experimental model and subject details

Primate brain RNA-seq data analysis

To model splicing shifts of cassette exons in a specific lineage within a phylogenetic tree, we used adult brain RNA-seq data from human³⁵ and six other primate species (data from non-human primate reference transcriptome resource, NHPRTR³⁶) with sufficient read coverage. For extended analysis of MAPT exon 10 splicing pattern divergence, we used additional RNA-seq data from the adult brain of primate species³⁷ to cover more species and increase the number of biological replicates from the same species. Only species with sequenced reference genomes and at least two biological replicates in each RNA-seq dataset were included in our analysis.

For each RNA-seq dataset, we mapped raw reads to the human reference genome (hg19) using OLego,⁷⁷ allowing 8 mismatches (for 101-nt reads). The number of allowed mismatches was relaxed here compared to standard analysis to accommodate mismatches caused by evolutionary changes in genomic sequences. This approach was found to be preferable over mapping to the individual reference genomes of respective species due to the varying quality and completeness across the reference genomes and the complication of mapping errors due to the presence of pseudogenes that affect each species differently. This method also avoided the complication of mapping AS events in humans to other species, since AS annotation is incomplete for non-human primates.

Quantification of splicing for 42,761 previously annotated cassette exons was performed using the Quantas pipeline (http://zhanglab.c2b2.columbia.edu/index.php/Quantas), as we described previously.⁷³ In brief, the inclusion level of each cassette exon (percent spliced in, PSI or Ψ) was calculated from the number of supporting exon junction reads for the inclusion and skipping isoforms. Only quantifications with ≥20 supporting junction reads were used for downstream analysis (and missing value was assigned otherwise).

Method details

Detecting lineage-specific splicing shifts under stabilizing selection using an OU model

For data preprocessing, we first excluded exons unless they could be quantified in at least one sample for each of the seven primate species. For the remaining exons, missing values were imputed by the Bayesian PCA method using R package ‘pcaMethods’.⁷⁸ Exons with minimal variation across samples (as measured by standard deviation (SD) < 0.05) were then excluded. In total, 3,690 cassette exons passed these filtering steps and were used for Ornstein-Uhlenbeck (OU) modeling (Tables S1; S2).

Compared to other methods, such as generalized linear models (GLMs), the OU model provides a statistical framework to model evolutionary changes of a quantitative trait and distinguish stabilizing (adaptive) selection from random drift within large phylogenies with complex covariance structures. This method has been previously applied to analyze the evolution of gene expression,³⁷ but not splicing. In brief, the OU process used to model the inclusion level of each exon across species in a phylogeny can be viewed as a random walk (Brownian motion) plus a pull towards an optimal value.^85,86 More formally, dΨ_t =α(θ-Ψ_t) dt+σdW_t, where Ψ_t is the quantitative trait value (I.e., exon inclusion level in our case) at time t, α parameterizes the strength of pull towards the optimal value θ, which can shift in specific lineages. dW_t models evolutionary drift using a normally distributed random variable with variance dt, and σ parameterizes the strength of drift. In this framework, the change of the quantitative trait over a time interval dt is the sum of a stochastic component (σdW_t, drift) and a deterministic component (α(θ-Ψ_t), stabilizing selection). For a phylogeny with known topology and branch lengths, we define r selective regimes acting on the phylogeny, each regime defined by an optimal value θ∈(θ₁, …, θ_r). We can then assign an OU process with parameters α, σ², and θ_i to model the trait value of each taxa Ψ_i. Now let Ψ_i = [Ψ₁,Ψ₂, …,Ψ_N] be the state of the OU processes at the N terminal taxa; Ψ follows a multivariate normal distribution. These model parameters can be estimated using the standard maximum likelihood methods. In particular, using this framework, we can identify AS exons under differential stabilizing selection in particular lineages (i.e., two selective regimes in the simplest case as described in this study) by testing the null hypothesis (in which all branches share the same optimum parameters θ_shared) against the alternative hypothesis (in which a different optimum θ₁≠θ _shared acted on a particular lineage).

We used an extended OU model named Expression Variance and Evolution (EVE) model,⁸⁷ which considers within-species variation. A likelihood ratio (LR) test was used to assess the fitness of the null and alternative models using a chi-square distribution, and we ran five branch-specific (BS)-LR tests on different selective regimes (i.e., human vs. non-humans, hominoids vs. non-hominoids, Catarrhines vs. New World monkeys, either in the whole phylogenetic tree composed of seven species or in Catarrhines composed of five species). The phylogenetic tree used for these tests was downloaded from the UCSC genome browser (https://genome.ucsc.edu). Chi-square pp-values were adjusted by Benjamini & Hochberg multiple test correction to obtain the false discovery rate (FDR) for the tests performed for each lineage. Results from these tests were summarized in Table S1.

Power estimation in detecting AS exons with lineage-specific splicing shifts

We performed a series of simulations to evaluate the pp-value calibrations and estimate the power of the EVE model. For each BS-LR test, we simulated the inclusion levels of 100 exons under a null model of θ_shared across the phylogeny, using parameters estimated from real RNA-seq data with the same BS-LR test. All simulated Ψ values, sampled from a normal distribution, were limited between 0 and 1 (Ψ values <0 were set to 0 while Ψ values > 1 were set to 1). The simulated data was then subjected to BS-LR test, and the pp-value for each exon was calculated using a chi-square test. The observed and expected pp-values were compared using a qq-plot (Figure S3) to confirm whether the chi-square pp-values were properly calibrated. To estimate the power, we simulated 100 exons with an exon inclusion shift (Δθ) in one lineage (branch) relative to the other branch in the phylogenetic tree. We calculated the power of the EVE model in detecting lineage-specific splicing shifts by counting how many times the simulated exons were called significant (chisq pp-value<0.05). These simulations were performed for varying magnitudes of lineage-specific shifts (Δθ⁠ ranging from 0.01 to 0.4), reflecting weak or drastic lineage-specific selection.

Functional annotation of cassette exons

To annotate cassette exons with detected lineage-specific splicing shifts, we obtained conservation of AS patterns between humans and mice and the impact of AS on protein-coding (i.e., whether AS causes frame-shifts that result in unproductive transcripts subject to nonsense-mediated decay (NMD)) from a previous study.⁷³ Developmental splicing changes were evaluated using human brain RNA-seq data from ref. ³⁵ (Lister) and BrainSpan.⁷⁴ Briefly, pairwise differential splicing analyses were performed using samples from different age groups for the Lister and BrainSpan datasets separately, using the Quantas pipeline (read coverage≥20, |ΔΨ|≥0.2, and Benjamini FDR≤0.05). We excluded exons with significant changes only in one pairwise comparison to focus on more robust changes and avoid potential confounding factors (such as variation in sample quality or in genetic backgrounds among individuals). In total, 4,651 cassette exons were called to have significant developmental splicing switches. Gene ontology (GO) analysis was performed for each list of genes containing cassette exons with lineage-specific splicing shifts using DAVID,⁸⁰ and GO terms that are significant in at least one gene list (Benjamini FDR<0.05) were shown in Figure 1D.

Analysis of splicing-regulatory elements among divergent cassette exons

Exonic and flanking intronic sequences of all human cassette exons were extracted from the human reference genome (hg19). We also mapped the coordinates of cassette exons from hg19 to the other reference genomes using UCSC’s liftOver tool to retrieve the orthologous sequences in the other non-human primate species (the assembly version of non-human primates can be found in Table S2).

For exons showing lineage-specific splicing shifts, we sought to test whether an increase in exon inclusion is associated with an increase in splicing enhancers and/or a decrease in splicing silencers while a decrease in exon inclusion is associated with a decrease in splicing enhancers and/or an increase in splicing silencers. To this end, we analyzed lists of computationally identified splicing-regulatory elements from previous studies, including RESCUE ESE,⁸¹ FAS-ESS,⁸² Chasin-PESE and PESS,⁸³ exon identity element (EIE), and intron identity elements (IIE).⁸⁴ For each group of exons showing a specific divergence pattern (e.g., higher inclusion in hominoids compared to other primate species), we calculated the density of each type of feature in the exon or upstream/downstream flanking intronic regions (100 nt) using sequences from the two compared lineages (observed counts normalized by the total possible counts determined by total sequence lengths). The densities between the two lineages were compared using a two-sided Binomial test.

We also tested whether changes in splice site strength can be explain the splicing shifts. For this analysis, splice site strength was calculated using MaxEntScan, which scores the local splice site motifs,³³ and SpliceAI, a machine learning-based algorithm that predicts splice site using both local and distal sequences (5kb on each side of the splice site).³⁴ For MaxEntScan, the prediction scores of 3′ and 5′ splice sites were averaged as a measure of the splice site strength of the exon. For SpliceAI, we used the maximum of the 3′ and 5′ splice site scores as a measure of the splice site strength of the exon, as recommended by the original study.³⁴ For each exon with splicing shifts, we consider pairs of species (one from each branch) and determine whether the direction of splicing divergence measured by RNA-seq is consistent with that predicted from the splice site strength. The number of exon pairs with consistent and inconsistent directions with varying thresholds on |ΔΨ| is summarized in Figure 1F.

Comparative analysis of developmental splicing switches in human and mouse cortices

We compared the developmental splicing profiles of 3,583 pairs of orthologous cassette exons in human and mouse developing brains using two RNA-seq datasets of human brain development^35,74 and two RNA-seq datasets of mouse brain development.^35,73 Exons with developmental splicing switches at specific time points were identified using weighted gene co-expression network analysis (WGCNA)⁸⁸ analysis in combination with sigmoidal fits, similar to a previous study.⁶ The more detailed procedure is described elsewhere. In this work, we focused on exons showing late splicing switches and increased exon inclusion during brain development (similar to MAPT exon 10), including 81 exons with conserved switches, 42 exons with mouse-specific switches and 38 exons with human-specific switches (Figure 3A).

RNA-seq analysis of MBNL-dependent splicing in human and mouse cortices

To identify the comprehensive list of MBNL-dependent exons, we previously generated a Mbnl1^−/−; Mbnl2 ^loxP/loxP; Nestin-Cre mouse line to deplete both MBNL1 and MBNL2 in neurons and glial cells (referred to Mbnl1/2 double-KO or DKO). Deep RNA-seq was performed using RNA extracted from adult Mbnl1/2 DKO and control cortices using the standard Illumina TruSeq platform (PE 101-nt reads; NCBI SRA accession: SRP142522). In parallel, we also performed RNA-seq analysis of human cortices obtained from control and myotonic dystrophy type 1 (DM1) patients, each group in triplicates (data are being deposited to NCBI SRA/GEO). Raw RNA-seq reads were processed using the same Quantas pipeline as described above. Differential splicing was called in mice and humans with the following criteria (read coverage≥20, |ΔI|≥0.1, and Benjamini FDR≤0.05).

Plasmid cloning of MAPT exon 10 minigenes

The human MAPT (hMAPT) and mouse Mapt (mMAPT) minigenes containing exon 9, exon 10, and exon 11 with relevant intronic sequences were cloned using Gibson assembly into the backbone vector pcDNA5/FRT (Thermo Fisher Scientific, Waltham, MA) using the HindIII and XhoI restriction sites. For the mMAPT minigene, the genomic region of interest was PCR-amplified from purified DNA obtained from wild-type B6/C57 mouse tail. Part of intron 9 containing deep intronic sequences (chr11: 104,310,938-104,317,156, mm10) was truncated to limit the size of the minigene. Caution was taken to avoid excluding conserved regions or major RBP binding sites based on CLIP data, while the entirety of intron 10 sequence was included in the minigene. In addition, a total of four mutant mMAPT minigenes with MBNL binding site deletions or replacements were cloned using Gibson assembly: i) deleted mouse MBNL binding site 1, ii) deleted mouse MBNL binding site 2, iii) both mouse MBNL binding sites 1 and 2 deleted, and iv) mouse MBNL binding site 2 replaced with the human MBNL binding site 2. For the hMAPT minigene, the genomic region of interest was PCR-amplified using DNA purified from HEK293T cells, and part of intron 9 (chr17:44,074,434-44,086,750, hg19), which is orthologous to the truncated mouse sequence, was also truncated. In addition to the wild-type hMAPT minigene, we also generated five mutant hMAPT minigenes, including one mutant minigene in which i) the MBNL binding site 1 was replaced with the orthologous mouse sequence and four mutant minigenes in which the MBNL binding site 2 was replaced with the orthologous binding site 2 sequence from ii) rhesus macaque, iii) squirrel monkey, iv) marmoset monkey, and v) mouse. All primer and gblock sequences used for cloning are listed in Table S4. All plasmids generated were confirmed by Sanger sequencing (Eton Bioscience Inc., Union, NJ).

MAPT exon 10 minigene splicing reporter assay

For hMAPT and mMAPT splicing reporter assays, HEK293T cells maintained in Dulbecco’s modified Eagle’s medium (DMEM) and supplemented with 10% FBS were seeded one day before transfection (about 3.5 x10⁵ cells per well of a 6-well plate). Plasmid DNA was transfected using Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA) according to the manufacturer’s instructions. For most experiments, 0.125 μg minigene and 1.5 μg pCAGGS-3xFLAG-MBNL2 expression vector (or pcDNA control) were used for transfection, except for experiments presented in Figure 4E, which used 1 μg pCAGGS-3xFLAG-MBNL2 expression vector for MBNL2 overexpression. Twenty-four hours post transfection, cells were scrapped in ice-cold 1× PBS and spun down. Half of the cells were used for total RNA isolation using Trizol reagent (Thermo Fisher Scientific, Waltham, MA) and Direct-zol RNA kits (Zymo Research). The remaining cells were resuspended in 60 μL lysis buffer (50 mM HEPES, pH 7.4, 100 mM NaCl, 1% Triton X-100, 0.1% sodium dodecyl sulfate (SDS), 1 mM EDTA, 1 mM dithiothreitol (DTT), cOmplete protease inhibitors (Roche Diagnostics, Mannheim, Germany)) for protein extraction. To compare MBNL1/2 expression, we also used protein lysates from mouse cortices at different ages (P4, P7, P30) generated in a previous study.⁷³

Protein samples were prepared with 4x Laemmlin Sample Buffer (Bio-Rad Laboratories, Hercules, CA) and 89mM β-mercaptoethanol, boiled, and loaded into 4–12% SDS-polyacrylamide gel electrophoresis (SDS-PAGE) using Novex Bis-Tris gels (Thermo Fisher Scientific, Waltham, MA) to confirm protein expression using immunoblots. After protein transfer onto 0.45 μm nitrocellulose membrane (GE Healthcare, Marlborough, MA), the following primary antibodies were used: rabbit α-MBNL1 (A2746, generous gift from Charles Thornton, 1:1000), mouse α-MBNL2 (3B4, Santa Cruz Biotechnology, 1:100), mouse α-FLAG (Sigma-Aldrich, 3B4, 1:4000), and mouse α-GAPDH (EMD Millipore, 6C5, 1:10,000).

For RT-PCR analysis of the splicing products, cDNA was prepared using SuperScript III reverse transcriptase (Thermo Fisher Scientific, Waltham, MA) with random hexamer primers. Alternative exons of interest were amplified with primers listed in Table S4 to measure exon inclusion. PCR products were resolved on 1.2–1.5% agarose gel.

MBNL1/2 RNA interference and MBNL2 overexpression

For samples with knockdown of MBNL1 and MBNL2, siRNAs (Millipore Sigma; MBNL1: 100 nM; MBNL2: 100 nM) were transfected into HEK293T cells seeded the day before (about 3.5x10⁵ cells per well of a 6-well plate) using Lipofectamine 3000 (ref. 75). The MAPT minigene plasmids (125 ng) were transfected 24 h after siRNA transfection using Lipofectamine 3000, as well. For overexpression of MBNL2, 1.5 μg pCAGGS-3xFLAG-MBNL2 expression vector was transfected at the same time point with MAPT minigene plasmids.⁸⁹ RNA and protein were collected 48-h post minigene transfection for immunoblots and RT-PCR analysis of the splicing products as described above.

Blocking of MBNL binding sites using dCas13d/gRNA

A total of four individual gRNAs targeting MBNL binding site 2 in the mMAPT minigene splicing reporter were cloned. The oligonucleotides were synthesized (IDT, Coralville, IA) and cloned into the CasRx gRNA cloning backbone using BsmBI restriction site (Addgene plasmid #138150). A non-targeting (NT) gRNA containing a spacer without a match in the human or mouse transcriptomes was used for control.⁵³ gRNA targeting sequences can be found in the Table S4. All constructs were confirmed by Sanger sequencing (Eton Bioscience Inc., Union, NJ). The dCas13d expression plasmid was kindly provided by Patrick Hsu (Addgene plasmid #109050).

For dCas13d/gRNA transfections, HEK293T cells maintained in DMEM and supplemented with 10% FBS were seeded one day before transfection (about 3.5 x10⁵ cells per well of a 6-well plate). A plasmid ratio of mMAPT:gRNA:dCas13d (50ng:1600ng:1600 ng) was transfected using Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA) according to manufacturer’s instructions. To test the effect of MBNL2 overexpression, 625 ng of pCAGGS-3xFLAG-MBNL2 expression vector was co-transfected, along with the mMAPT minigene and dCas13d/gRNA plasmids. Control samples included 625 ng of pcDNA plasmid. Cells were collected 48 h post-transfection for RNA and protein extraction. Immunoblots and RT-PCR analysis of splicing products were performed as described above.

Quantification and statistical analysis

Statistical tests

Two-sided Student’s t-test with equal variants was used to evaluate splicing differences between conditions as measured by RT-PCR analysis.

Published: May 20, 2024

Supplemental information

Table S3. Exons with human-specific splicing shifts are enriched in ciliopathy genes, related to Figure 1 in the main text

Exon coordinates are based on hg19. NMD_in and NMD_ex, NMD upon exon inclusion or exclusion; IFT, intraflagellar transport subcomplex.