Protein phosphatase 2A – structure, function and role in neurodevelopmental disorders

Protein phosphorylation is arguably the most common post-translational modification (PTM) in cells. It is estimated that there are 230,000 phosphorylation sites in humans (Vlastaridis et al., 2016), the majority of which (>96%) are serine (Ser) and threonine (Thr) residues (Murray, 2011). Interestingly, for Ser/Thr-modifying enzymes, the number of protein kinases are more than 10-fold greater than the number of protein phosphatases (Manning et al., 2002). This difference has contributed to the misconception that protein Ser/Thr phosphatases are promiscuous and unregulated enzymes that merely reset proteins for regulation by kinase. However, the complexity and specificity of the Ser/Thr phosphatases arises from the association of the catalytic subunit with a large array of regulatory subunits. One of the most abundant Ser/Thr phosphatases is protein phosphatase 2A (PP2A) (Clark and Ohlmeyer, 2019; Wlodarchak and Xing, 2016). PP2A is composed of a catalytic subunit (PP2A-C) and a scaffolding subunit (PP2A-A). This core dimer (PP2A_D) binds one of several regulatory subunits (PP2A-B) to form the PP2A heterotrimeric holoenzyme (Fig. 1). There are four gene families encoding the canonical regulatory PP2A-B subunit: PP2A-B, PP2A-B′, PP2A-B″ and PP2A-B‴ encoded by PPP2R2, PPP2R5, PPP2R3 and PPP2R6, respectively. The three to five isoforms in each regulatory subunit family display high sequence conservation, and neither sequence nor structural similarities exists between gene families. Regulatory subunits are further diversified by alternative splicing, such that any given cell type likely expresses more than a dozen distinct PP2A holoenzymes (Seshacharyulu et al., 2013).

PP2A is recognized as a tumor suppressor and negative regulator of cell growth, mostly because of the direct inhibition of PP2A-C by cancer-promoting okadaic acid (Bialojan and Takai, 1988; Janssens et al., 2005; Li et al., 1995; Pallas et al., 1992). Additionally, deletions of PP2A-subunit-encoding genes and inactivating mutations have been identified in various cancers (Calin et al., 2000; Colella et al., 2001; Ruediger et al., 2001; Suzuki and Takahashi, 2003; Takagi et al., 2000; Wang et al., 1998). PP2A dysfunction has also been implicated in Alzheimer's disease (AD). Indeed, PP2A is the main phosphatase targeting the hyperphosphorylated tau, the main component of neurofibrillary tangles (Sontag et al., 2004a,b; Trojanowski and Lee, 1995; Vafai and Stock, 2002). PP2A expression and activity are reported to be much reduced in postmortem AD brains samples, whereas the methylation state of PP2A controls the sensitivity and resistance to β-amyloid-induced cognitive and electrophysiological impairments. Consistent with this, multiple in vivo studies have found that inhibiting PP2A produces AD-like tau pathology and cognitive impairment (Gnanaprakash et al., 2021; Nicholls et al., 2016; Staniszewski et al., 2020).

With whole exome (WES) or genome sequencing (WGS) becoming an affordable diagnostic tool, an increasing number of PP2A subunit mutations have been identified as likely causes of neuropsychiatric and neurodevelopmental disorders (NDDs). Most of these are de novo mutations (DNMs) that are present in the affected individual but not the parents. DNMs typically arise during oogenesis or spermatogenesis, although they can also occur during embryonic development, in which case they result in mosaicism (Acuna-Hidalgo et al., 2015; Biesecker and Spinner, 2013; Campbell et al., 2015; Dal et al., 2014; Lupski, 2010). Here, we review recent discoveries of pathogenic mutations in the PP2A holoenzyme, sourced from peer-reviewed literatures and additionally tabulate novel pathogenic single nucleotide variants (SNVs) identified from publicly available clinical databases (DECIPHER, ClinVar and SFARI; see Table S1). We summarize the genetic and clinical features of the associated NDDs and discuss implications of these mutations on the structure and function of PP2A as a regulator of signaling cascades in neuronal development.

In mammals, the two isoforms of PP2A-A – PP2A-Aα and PP2A-Aβ (encoded by PPP2R1A and PPP2R1B, respectively) – are ubiquitously expressed and highly conserved (Groves et al., 1999; Hemmings et al., 1990; Khew-Goodall and Hemmings, 1988; Stone et al., 1987; Zhou et al., 2003) (Table 1). PP2A-A is composed of 15 tandem Huntington elongation A-subunit TOR (HEAT) repeats (HRs). Each HR is composed of two α-helices connected by inter- and intra-repeat loops (Fig. 1). HR11 to HR15 anchors PP2A-C, while the variable regulatory B subunit binds to HR1 to HR10. Recent WGS and WES studies (Houge et al., 2015; Lenaerts et al., 2021; Wallace et al., 2019; Zhang et al., 2020) have reported a total of 16 SNVs on the PP2A-Aα isoform (Fig. 2A, red; Table S1). Houge et al. (2015) identified individuals with P179L, R182W and R258H mutations that show severe ID and display neurological deficits such as microcephaly, agenesis of the corpus callosum (ACC), and speech and motor disabilities. These symptoms together in patients are sometimes referred to as autosomal mental retardation 36 (MRD36). Recently, Lenaerts et al. (2021) reported 12 novel mutations in individuals who commonly displayed language delay, hypotonia and hypermobility in joints, but the severity of developmental delay (DD) ranged from mild to severe ID, correlating well with the diverse range of biochemical dysfunctions observed, as is briefly discussed below (Table S1). The one exception was that the individual harboring the S152F mutation did not have ID, possessing a full-scale IQ of 86, but did display autism spectrum disorder (ASD) (Deciphering Developmental Disorders Study, 2015; Houge et al., 2015; Lenaerts et al., 2021; Zhang et al., 2020). Among all these mutations, six recurrent mutations (M180T, M180V, R182W, S219L, V220M and R258H) have been identified in almost 75% of the total PPP2R1A patients (Lenaerts et al., 2021).

Structurally, most PP2A-Aα mutations (Fig. 2A, red) map to the HR1 to HR7 region that engages PP2A-B subunits. Most of these mutations reside adjacent to each other and/or affect the same amino acid within the intra-repeat loops between HR4 and HR7, suggesting this region as a hotspot for the diverse biochemistry and phenotypes associated with PP2A-Aα-mediated NDD. The unique S152F mutation residing within HR4 shows no functional impairments biochemically. However, neurons with this mutation show a significant decrease in dendritic spines indicating its pathogenicity via altered functionality (Lenaerts et al., 2021). Most PP2A-Aα mutations (refer to Table S1 for specific mutation effects) show no or less binding to the PP2A-Bα (encoded by PPP2R2A) but not to PP2A-B′δ (encoded by PPP2R5D) or PP2A-B‴ (encoded by PPP2R6B). Interestingly, these variants bind strongly to STRN3 (encoded by PPP2R6B), but it is unclear how the interaction is associated with the severe ID, seizures and microcephaly in patients. Several PPP2R1A mutations (F141L, P179L, R182W, R183W, S219L, V220M and R258H/S) lead to significantly less binding to PP2A-C despite being present on HR1 to HR8, which are not known to mediate PP2A-C subunit binding, thereby suggesting that some B subunits indirectly mediate binding between PP2A-A and PP2A-C via its interaction with PP2A-A subunit. In a recent cryo-EM structure, STRN3 has been shown to interact via HR1 of the PP2A-A subunit, wherein a mutation within this interface causes diminished PP2A-A and PP2A-C binding suggesting that there would be a similar effect on holoenzyme assembly mediated through the B subunit (Jeong et al., 2021). Notably, functionally disruptive mutations, such as P179L, R182W, R183W, S219L, V220M, R258H/S, are also found in the Catalogue of Somatic Mutations in Cancer (COSMIC), which shows that these exact PPP2R1A mutations cause cancer growth and migration and drug resistance (Haesen et al., 2016; Jeong et al., 2016; O'Connor et al., 2020). Additionally, functionally impaired mutations can cause insufficient dephosphorylation of PP2A substrates (Houge et al., 2015). In HEK293 cells overexpressing the PP2A-Aα R182W mutant, an increased phosphorylation of Ser9 and inactivation of GSK-3β has been observed (Houge et al., 2015). GSK-3β, being an established PP2A-B′δ substrate, plays a pivotal role in controlling the neuronal progenitor proliferation and regulating neuronal polarity (Houge et al., 2015; Jiang et al., 2005; Liu and Eisenman, 2012). With severe biochemical and phenotypic effects of these PPP2R1A mutations, further studies need to be performed in both in vitro and in vivo systems.

In addition to PP2A-Aα, seven de novo or inherited mutations have been reported in the PP2A-Aβ isoform. Six individuals display mild to severe ASD (Fromer et al., 2014; Leblond et al., 2019; Ruzzo et al., 2019; Sanders et al., 2012; Wu et al., 2020) and one individual has Tourette's syndrome (Willsey et al., 2017) (Fig. 2A, blue; Table S1). One missense variant, R270C, is homologous to R258H/S mutation (Fig. 2A, purple) on PP2A-Aα, suggesting a likely similar effect on holoenzyme assembly and pathogenicity. Three frameshift mutations, V115C/G and D211V, and a nonsense mutation, R261*, result in early protein termination of PP2A-Aβ and likely lead to nonsense-mediated decay, hinting towards haploinsufficiency, although this requires further investigation and biochemical characterization (Leblond et al., 2019; Wu et al., 2020).

The two isoforms of PP2A-C, PP2A-Cα and PP2A-Cβ (encoded by PPP2CA and PPP2CB, respectively) are highly conserved among mammals (Table 1) (Orgad et al., 1990; Seshacharyulu et al., 2013). PP2A-C adopts a compact spherical structure composed of α/β folds; the two central β-sheets are connected via protein loops comprised of six highly conserved residues – D57, H59, D85, N117, H167 and H241 – that chelate the catalytic manganese ion (Xing et al., 2006) (Fig. 1). PP2A activity is regulated by two important PTMs – phosphorylation (Chen et al., 1992) and methylation (Lee and Stock, 1993) – on its highly conserved C-terminal tail (Janssens et al., 2008) mediated by an array of proteins (Box 1). PP2A-C has been reported to dephosphorylate and activate various proteins that are critical in neuronal growth, such as actin-depolymerizing factor (ADF), cofilin (Kuhn et al., 2000), collapsin response mediator protein 2 (CRMP2) (Zhu et al., 2010) and Ca²⁺/calmodulin-dependent protein kinase II (CaMKII) (Fink et al., 2003; Monroe and Heathcote, 2013). Indeed, overexpression of PP2A-C enhances the formation of neurites in neuroblastoma-2a cells, whereas inhibition of the PP2A-C activity in neuroblastoma-2a cells and hippocampal neurons reduces axonal growth and neurogenesis (Liu et al., 2012; Prinetti et al., 1997). Recently, a WES and microarray-based copy-number analysis have identified 16 patients with 15 different variants caused by DNMs in PP2A-Cα (Reynhout et al., 2019). These include eight missense non-recurrent mutations (G60V, D88G, Q122H, Y127C, D131H, D223V/H, and Y265C), one recurrent missense mutation (H191R), three nonsense mutations (Q125*, R214*, and R295*), a single amino acid duplication (F308dup), a frameshift (F146L fs*29), and a partial gene deletion (del in Chr5) (Reynhout et al., 2019) (Fig. 2B). These individuals have mild to profound ID and DD, severe language impairment, hypotonia, epilepsy, ASD and worsening behavioral problems (Table S1).

Functional studies have shown that the PP2A activity is completely inhibited by the D88G, Y127C, R214*, Y265C, and F308dup mutations, while a significant reduction is seen with the Q122H, H191R, D223H, and R295* mutations (Reynhout et al., 2019). The decrease in catalytic activity is likely due to disruption of methylation by mutations of key residues in the catalytic center (D88, Y127 and Y265), which make direct contact with leucine methyl transferase (LCMT-1), or due to the absence of C-terminal tail in the truncating variants R214* and R295*. These mutations therefore are likely to cause pathologies by a loss-of-function mechanism (Reynhout et al., 2019).

Biochemically, most catalytically impaired mutations retain the ability to interact with endogenous PP2A-A but vary vastly in their ability to form heterotrimers with different PP2A-B regulatory subunits, suggesting different mechanisms of pathogenicity. For example, D88G, R295* and F308dup variants only form heterotrimers with PP2A-B′δ (encoded by PPP2R5D), the Y127C mutation forms heterotrimers only with the PP2A-B and PP2A-B‴ subunits, suggesting a dominant-negative mode of action against PP2A-B′δ substrates and PP2A-B and PP2A-B‴, respectively. Haploinsufficiency is also proposed based on the premature stop mutations, partially deleted PPP2CA and poorly expressing gene product. H191R, a recurrent missense mutation, is located in the conserved ‘loop switch’ (Jiang et al., 2013) (Fig. 2B) region that determines the proper conformation of the active site. A missense mutation therefore would cause PP2A-C inactivation and decrease its binding to PP2A-B′δ, hinting towards haploinsufficiency (Reynhout et al., 2019).

Despite the high expression of PP2A-C in the brain, especially the hippocampus (a region of lifelong neurogenesis) (Liu et al., 2010), and of the role of PP2A-C in neuronal growth, it is still unclear how these specific PP2A-Cα mutations alter signaling cascade to cause pathologies. Brain-specific PPP2CA knockout (KO) mice have phenotypic features similar to individuals with DNMs, including shrinkage of cortical neurons, synaptic plasticity impairments, and learning and memory deficits (Liu et al., 2018), thereby offering a great in vivo model system for further studies.

The mammalian B family of regulatory PP2A-B subunits (also referred to as B55 or PR55) consists of four isoforms: α, β, γ and δ, which are encoded by the genes PPP2R2A–PPP2R2D, respectively. They share high protein sequence similarity (>90% for humans) except at their divergent N-termini, and show specific expression patterns and subcellular localizations (Table 1) (Healy et al., 1991; Mayer et al., 1991; Strack et al., 1999; Zolnierowicz and Hemmings, 1994). The crystal structure of Bα within the PP2A holoenzyme reveals a seven-bladed β-propeller made up of WD40 (tryptophan-aspartate) repeats (Strack et al., 2002). A β-hairpin extending from the second blade contributes to binding of the A subunit, as do residues at the bottom of the centrifuge rotor-shaped structure/β propeller structure of the regulatory subunit (Fig. 1) (Xu et al., 2008). Several kinase signaling cascades are known to be regulated by the different PP2A-B subunits. Specifically, PP2A-Bα inhibits AKT signaling by dephosphorylating phosphorylated T308 (Kuo et al., 2008), PP2A-Bγ promotes neuronal differentiation by activating the MAPK cascade (Strack, 2002), and PP2A-Bα and PP2A-Bδ directly dephosphorylate ERK proteins (Van Kanegan et al., 2005). In Xenopus egg extracts, PP2A-Bα and PP2A-Bβ are involved in multitude of metabolic processes, such as mitochondria function, cell division, apoptosis, cytoskeleton organization and DNA damage repair (Cundell et al., 2016; Wang et al., 2018), whereas PP2A-Bδ controls mitotic entry and exit (Mochida et al., 2009). Among the four B family genes, a possibly pathogenic DNM and a germline breakpoint mutation have been identified in PPP2R2B and PPP2R2C, respectively, as discussed below (Table S1).

PP2A-Bα (encoded by PPP2R2A) and PP2A-Bδ (encoded by PPP2R2D) are expressed in many tissues, including the brain. PPP2R2A is an essential gene, as homozygous KO mice die during embryonic development (Panicker et al., 2020). Specifically, PP2A-Bα is a critical regulator of epidermal stratification and ectoderm development (Panicker et al., 2020). Moreover, dysregulation of PP2A-Bα expression has been implicated in the pathogenesis of neurodegenerative diseases, such as Alzheimer's disease (AD) (Sontag et al., 2004a,b) and Parkinson's disease (PD) (Park et al., 2016). The PP2A-Bα holoenzyme is also the primary tau phosphatase; its downregulation is correlated with increased tau phosphorylation in vivo and in AD patients (Sontag et al., 2004a,b). Similarly, PP2A-Bα plays an important role in dephosphorylating hyperphosphorylated α-synuclein, a hallmark of PD, which leads to misfolding and aggregation of α-synuclein into proto-fibrils (Lee et al., 2011). However, currently, there is no direct link between PP2A-Bα or PP2A-Bδ and NDDs, despite their high expression in brain. However, many commercially available antibodies against the B subunits do not actually differentiate between subunits of this regulatory family, which may lead to difficulty in attributing specific effects of these genes.

PP2A-Bβ (encoded by PPP2R2B) is expressed specifically in neurons and is associated with the autosomal-dominant neurodegenerative disease spinocerebellar ataxia type 12 (SCA12) (Fujigasaki et al., 2001; Holmes et al., 1999). SCA12 is caused by a CAG repeat expansion (>50 repeats) in a 5′-UTR or promoter region of PPP2R2B, but whether and how the repeat expansion affects PPP2R2B expression is currently unknown. SCA12 patients present cerebral and cerebellar atrophy, bradykinesia, ataxic gait, dysarthria, tremors and dementia (Holmes et al., 2001, 2003; Srivastava et al., 2001). PPP2R2B undergoes alternative splicing and alternative promoter use, which gives rise to several N-terminal sequence variants. The two most abundant isoforms, PP2A-Bβ1 and PP2A-Bβ2, are targeted via their divergent N-terminal tails to the cytosol and the mitochondrial surface, respectively (Dagda et al., 2003; Merrill et al., 2013). During neuronal stress in vitro, PP2A-Bβ2 induces mitochondrial fragmentation by dephosphorylating S637 and activating the mitochondrial fission enzyme Drp1 (also known as DNM1L), which increases susceptibility to neuronal insults and cell death (Dagda, 2006; Dagda et al., 2003, 2008; Merrill et al., 2013). This highly conserved inhibitory phosphorylation site confers neuroprotection by outer-mitochondrial protein kinase A (PKA) in cerebral ischemic stroke (Flippo et al., 2018). Excessive mitochondrial fission is associated with multiple neurodegenerative diseases (Serasinghe and Chipuk, 2017). We recently reported that knocking out PP2A-Bβ2 in mice provides protection against stroke damage by maintaining neuroprotective Drp1 phosphorylation (Flippo et al., 2020). We also showed that this reversible phosphorylation of Drp1 at S637 by PP2A-Bβ2 and AKAP1–PKA (AKAP1 is a kinase anchoring protein for PKA) affects dendrite and synapse development in cultured hippocampal neurons. Phosphorylation of Drp1 by AKAP1–PKA increases mitochondrial length and dendritic occupancy, enhances the dendritic outgrowth, and reduces the synapse number and density, whereas the dephosphorylation of Drp1 by PP2A-Bβ2 has the opposite effect (Dickey and Strack, 2011). A recent WES study has identified a de novo missense mutation at R149P (Bβ1 numbering) in PPP2R2B that is associated with moderate to severe ID and pervasive DD not otherwise specified (PDD-NOS), along with poor eye contact, intractable seizures, autistic features and aggressive behaviors (Hamdan et al., 2014). This mutation is situated within the extended β-hairpin that contacts PP2A-A and thus may block association of PP2A-Bβ with the holoenzyme (Fig. 3A); however, further biochemical and in vivo studies need to be performed.

PP2A-Bγ (encoded by PPP2R2C) is also predominantly expressed in the brain (Zolnierowicz and Hemmings, 1994). Forced expression of PP2A-Bγ promotes neuronal differentiation via transient activation of MAPK signaling in a neuronal cells line (Strack, 2002). PP2A-Bγ also regulates cancer cell proliferation by suppressing the mTOR signaling pathway, and low levels of PP2A-Bγ activity have been linked to brain cancer (Fan et al., 2013; Li et al., 2015). Additionally, the PPP2R2C gene is located within a region of chromosome 4p16 that is associated with bipolar disorder (BP) and single nucleotide polymorphisms (SNPs), and other BP risk haplotypes have been reported in PPP2R2C (Borsotto et al., 2007). Providing a potential pathogenic mechanism, PP2A-Bγ associates, dephosphorylates and activates the K⁺ channel KCNQ2 (Borsotto et al., 2007; Hu et al., 2000). In another reported family pedigree, PPP2R2C is disrupted by an autosomal dominant familial reciprocal translocation (4;6) (p.16.1; q22) on chromosome 4 (Backx et al., 2010). Affected individuals suffer from mild ID, learning and behavior problems, including aggressive, obsessive, transgressive behaviors and tonic-clonic epilepsy episodes (Backx et al., 2010).

The B′ family members (α, β, γ, δ and ε), encoded by PPP2R5A–PPP2R5E, are highly conserved and contain divergent N- and C-termini that are subject to alternative splicing (see sequence alignment in Fig. S1A). The crystal structure of the PP2A-B′γ holoenzyme reveals that B′-family members are curved superhelical structures containing eight tandem α-helical HEAT-like repeats (Xu et al., 2006) (Fig. 1). The concave surface facing PP2A-C is negatively charged, while the convex side contacting PP2A-A contains conserved hydrophobic residues (Xu et al., 2006). The PP2A-B′-family members have ubiquitous tissue expression with the greatest abundance in the brain and testis (McCright et al., 1996). They also display distinct localization, with PP2A-B′δ and PP2A-B′γ containing a predicted nuclear-localization motif (McCright et al., 1996; Tehrani et al., 1996), suggestive of specific regulatory functions (Nijenhuis et al., 2014; Vallardi et al., 2019). Indeed, the B′ subunits are important for cell cycle regulation, where PP2A-B′γ and PP2A-B′ε have been shown to control anaphase onset and cytokinesis, whereas PP2A-B′α, PP2A-B′β and PP2A-B′δ mediate chromosome alignment and segregation (Bastos et al., 2014). All five B′-family members are phosphorylated proteins (McCright et al., 1996). Both PP2A-B′δ and the splice variant PP2A-B′γ3 contain a consensus motif for PKA-mediated phosphorylation (RRKS) (McCright et al., 1996; Tehrani et al., 1996), which in the case of PP2A-B′δ has been shown to activate the PP2A holoenzyme (Yu and Ahn, 2010).

The mechanisms by which PP2A-B′-containing holoenzymes recognize specific substrates are beginning to be unraveled. We previously reported charge–charge interactions as a general mechanism for substrate recognition by the B′-family members (Jong et al., 2020; Saraf et al., 2010). More recently, the short linear motif (SLiM) LxxIxE has been identified in B′-interacting proteins, such as BuBR1 and KIF4A (Hertz et al., 2016; Kruse et al., 2020; Wang et al., 2016). BuBR1 is a multidomain protein kinase that regulates mitotic spindle checkpoint (Bolanos-Garcia and Blundell, 2011; Takahashi et al., 2016), while KIF4A is a chromokinesin that regulates mitosis, as shown by its function in chromosome condensation and segregation (Takahashi et al., 2016). These SLiMs are recognized by a well-conserved hydrophobic binding pocket on B′ subunit between HEAT repeats 3 and 5 (Hertz et al., 2016). Phosphorylation of residues within the SLiM further strengthens the interaction with B′ subunits (Hertz et al., 2016). A recent study described a highly conserved acidic surface adjacent to the SLiM binding pocket on B′ subunits that electrostatically interacts with a basic patch motif on substrates, suggesting that charge–charge interactions enhance the dephosphorylation of SLiM-containing substrates by PP2A-B′ (Wang et al., 2020). Below, we briefly described each of these five B′ subunit-encoding genes and its associated NDDs (Table 1; Table S1).

Primarily studied in relation to cancer progression (Janghorban et al., 2017; Liu et al., 2020; Zhuang et al., 2017), the role of PP2A-B′α in neuronal development and physiology is poorly understood. In non-neuronal cells, PP2A-B′α regulates cell cycle progression to promote mitosis (Foley et al., 2011; Porter et al., 2013) and initiates DNA repair response following DNA damage (Freeman et al., 2010). In addition, PP2A-B′α regulates cell growth and proliferation by inhibiting Wnt signaling and mediates degradation of c-Myc and β-catenin (Li et al., 2001). Indeed, several cancer-related studies have identified c-Myc as a direct substrate of PP2A-B′α (Arnold and Sears, 2006; Leonard et al., 2020). At present, no NDDs associated with PP2A-B′α mutations have been reported.

Primarily expressed in neurons with predominant somatodendritic distribution (Saraf et al., 2007), PP2A-B′β has been implicated in dopamine synthesis (Saraf et al., 2007), hippocampal long-term potentiation (LTP) (Fukunaga et al., 2000) and signal transduction in response to stimulation with various growth factors (Brandt et al., 2008; Van Kanegan and Strack, 2009). PPP2R5B has been recently linked to NDDs as two chromosomal deletions encompassing PPP2R5B have been reported (Boyle et al., 2015; Mohrmann et al., 2011). In addition, one de novo missense mutation in PPP2R5B (S161L) is reported as a pathogenic variant that causes moderate ID and human overgrowth (Loveday et al., 2015) (Table S1). Based on its close proximity to the catalytic subunit, it is postulated that the PP2A-B′β S161L mutation compromises substrate binding and dephosphorylation, thus enhancing phosphoinositide 3-kinase (PI3K)/AKT signaling to cause human overgrowth syndrome. N-terminal phosphorylation of PP2A-B′β has shown to regulate neuronal signaling (Bidinosti et al., 2016). In induced pluripotent stem cell (iPSC)-derived neurons from Phelan–McDermid syndrome (PMDS) patients harboring SHANK3 deletions, PP2A-B′β was found to be hyperphosphorylated at S46 due to elevation of its kinase CLK2, which inhibits AKT–mTORC1 signaling (Bidinosti et al., 2016).

The PP2A-B′γ isoform (encoded by PPP2R5C), consists of three splice variants (γ1, γ2, and γ3), is the most widely studied isoform in the B′ family. Various cellular functions of the PP2A-B′γ-containing holoenzyme have been documented, including a role in cell cycle regulation through interaction with shugoshins (Xu et al., 2009), CYK4 (Wu et al., 2017), KIF4A (Bastos et al., 2014), BuBR1 (Varadkar et al., 2017), p53 (also known as TP53) (Li et al., 2007) and the APC/C–CDC20 complex (Fujimitsu and Yamano, 2020). The PP2A-B′γ-containing holoenzyme also regulates Wnt/β-catenin signaling during lung airway morphogenesis (Everett et al., 2002) and vascular remodeling (Jeon et al., 2010), which is likely to be mediated by interaction with Naked cuticle (NKD1), an inhibitor of the Wnt signaling cascade (Creyghton et al., 2005). The PP2A-B′γ subunit has also been suggested to act as an adaptor protein by interacting with liprin-α, a binding partner of the LAR receptor tyrosine phosphatase (Arroyo et al., 2008). When bound to liprin-α, the PP2A-B′γ subunit does not promote cell transformation, but may regulate cell morphology by facilitating the recruitment of substrates to membrane structures containing liprin-α (Arroyo et al., 2008). Pathologically, PP2A-B′γ has been associated with cancer as a tumor suppressor (Chen et al., 2004; Ihara et al., 2002; Ito et al., 2000; Nobumori et al., 2012; Sablina et al., 2010; Shouse et al., 2011; Zheng et al., 2011). While rare, one case of a DNM in PP2A-B′γ is linked to ID and human overgrowth (Loveday et al., 2015) (Table S1). Specifically, an in-frame 3-bp deletion (c.468_470delAAC) causes loss of T157 (based on NP_001133197), which is homologous to T181 (based on NP_001339842), a residue that disrupts the contact of PP2A-B′γ with the PP2A-C subunit and subsequent substrate recognition (Houge et al., 2015; Shang et al., 2016). With regard to the possible pathogenic mechanism, it has been proposed that the PP2A-B′γ T157 deletion disrupts PI3K–AKT–mTOR signaling (Loveday et al., 2015). As there are no published biochemical or functional studies of PP2A-B′γ mutations, further characterization on PPP2R5C mutations is highly warranted.

The PP2A-B′δ (encoded by PPP2R5D) isoform is the largest member of the B′ family with unique N- and C-terminal sequence extensions (McCright et al., 1996). PP2A-B′δ contains a PKA phosphorylation motif ⁵⁷⁰RRKSEL⁵⁷⁵ (McCright et al., 1996), and phosphorylation at Ser573 has been shown as essential for activation of the PP2A-B′δ-containing holoenzyme (Ahn et al., 2007a; Ranieri et al., 2018; Yu and Ahn, 2010). A number of studies report roles of the PP2A-B′δ-containing holoenzyme. First, the PP2A-B′δ-containing holoenzyme regulates cell cycle progression by dephosphorylating the mitotic activator Cdc25 at Thr138 and releasing 14-3-3 to promote mitosis (Margolis et al., 2006). Second, it facilitates the nuclear translocation of voltage-dependent L-type Ca²⁺ channel subunit β4 isoform (CACNB4) to regulate neuronal excitability (Ronjat et al., 2013). Third, the PP2A-B′δ subunit interacts with and promotes cytoplasmic localization of PPM1G, a nuclear-localized Ser/Thr type 2C phosphatase family member, to dephosphorylate α-catenin and regulate the proper assembly of adherens junctions (Kumar et al., 2019). Finally, dopaminergic neurotransmission is regulated via cAMP-dependent phosphorylation of the PP2A-B′δ subunit and activation by PKA, leading to dephosphorylation of DARPP-32 at T75 (Ahn et al., 2007a; Yu and Ahn, 2010), or by PKC-mediated phosphorylation of the PP2A-B′δ subunit, which in this case leads to dephosphorylation of tyrosine hydroxylase at S40 (Ahn et al., 2011).

Over the past decade, mutations in PPP2R5D have been reported to cause various pathologies, such as human overgrowth (Loveday et al., 2015), ID (de Ligt et al., 2012; Gilissen et al., 2014; Stessman et al., 2017), Parkinsonism (Hetzelt et al., 2021; Kim et al., 2020; Walker et al., 2021) and ASD (Abrahams et al., 2013; Basu et al., 2009; Firth et al., 2009; Iossifov et al., 2014; Satterstrom et al., 2020). ID caused by de novo PPP2R5D mutations is referred to as MRD35 or Jordan's Syndrome after one of the first patients diagnosed in the United States by WES. Jordan's Guardian Angels (JGA) was established in 2017 to support research on Jordan's Syndrome (https://jordansguardianangels.org/). Currently, JGA has identified more than 130 individuals with Jordan's Syndrome (Biswas et al., 2020; Mirzaa et al., 2019). Ten pathogenic PPP2R5D missense mutations have been reported so far, P53S, E197K, E198K, E200K, E250K, D251A, P201R, W207R, W207L and E420K (Fig. 3B; Table S1) (Houge et al., 2015; Loveday et al., 2015; Mirzaa et al., 2019; Reijnders et al., 2017; Shang et al., 2016; Yeung et al., 2017). E198K mutation is the most common, accounting for approximately one-third of all cases. Except for P53S, E250K, D251A and E420K, most mutations lie in the highly conserved acidic loop of the PP2A-B′δ subunit (Fig. 3B, red). This acidic loop faces and directly contacts PP2A-C (Shang et al., 2016) (Fig. 1). Acidic-to-basic mutations in this region have been reported to destabilize the PP2A-B′δ-containing holoenzyme (Houge et al., 2015), but may also affect substrate specificity (Saraf et al., 2007). Further functional studies have suggested that most PPP2R5D mutations cause Jordan's syndrome by dysregulating PI3K–AKT–mTOR signaling, which regulates cellular proliferation and growth (Loveday et al., 2015; Yeung et al., 2017). Furthermore, a recent study has shown that holoenzyme containing E420K PP2A-B′δ forms a complex with AKT family proteins and suppresses the dephosphorylation of AKT (Papke et al., 2021). Interestingly, mice with global KO of both PPP2R5D alleles are viable and fertile, and display no behavioral or cognitive abnormalities (Louis et al., 2011). In the brain, the KO has no apparent effect on phosphorylation of AKT family proteins although S9 GSK3β phosphorylation is paradoxically decreased (Louis et al., 2011), perhaps because of developmental compensation by other B′ regulatory subunits. PPP2R5D KO mice also exhibit spatially restricted tau hyperphosphorylation that is linked to hyperactivation of CDK5 and GSK3β in the brain (Louis et al., 2011). In addition, hyperactivation of GSK3β is also observed in the liver of the KO mice, which would explain the spontaneous tumor development, validating the role of PP2A as a tumor suppressor (Lambrecht et al., 2018). More recently, the PPP2R5D missense mutations E198K, E200K and E250K have been identified in several patients with early-onset atypical PD (Kim et al., 2020; Walker et al., 2021). These observations suggest that PPP2R5D dysregulation in the brain can cause both neurodevelopmental and neurodegenerative diseases.

The functional role of PP2A-B′ε (encoded by PPP2R5E) is not well studied. However, initial studies have described roles for PP2A-B′ε in the regulation of embryonic development via Wnt/β-catenin signaling (Jin et al., 2010; Vinyoles et al., 2017; Yang et al., 2003), microtubule organization through interaction with microtubule crosslinking factor 1 (MTCL1) (Hyodo et al., 2016), vertebrate eye development via the PI3K/AKT pathway (Rorick et al., 2007), and KLHL42-regulated and TGFβ-mediated profibrotic signaling (Lear et al., 2020). While downregulation of PPP2R5E has been associated with cancers (Cristobal et al., 2013, 2014; Liu et al., 2014; Yang et al., 2013), PPP2R5E mutations have recently been linked to mild ID and global DD (Murcia Pienkowski et al., 2019). De novo balanced chromosomal translocations involving chromosomes 5q11.2 and 14q23.2 appears to be caused primarily by the PPP2R5E gene deletion on chromosome 14 (Table S1) (Murcia Pienkowski et al., 2019). Further analyses of the PPP2R5E gene reveal a high probability of loss-of-function intolerant (pLI) score (0.9994) and a high DOMINO score, which measures the likelihood of a dominant inheritance (0.86), both of which are consistent with a dominant effect (Murcia Pienkowski et al., 2019). How PPP2R5E haploinsufficiency alters PP2A-mediated dephosphorylation events in neurodevelopment is not known.

The B″ family members are extended α-helical proteins with an N-terminal hydrophobic region, two EF hand Ca²⁺-binding domains and a substrate-binding domain at their C-terminus (Fig. 1) (Ahn et al., 2007b; Hendrix et al., 1993; Li and Virshup, 2002; Wlodarchak et al., 2013). Both the hydrophobic region and the first EF hand contact PP2A-A, whereas the second EF hand and an adjacent α-helix of B″ contact PP2A-C (Wlodarchak et al., 2013). The B″ family consists of three isoforms encoded by three genes, B″α1/α2 (also known as PR130 and PR72, encoded by PPP2R3A) (Janssens et al., 2003, 2016), B″β1/β2 (PR70, encoded by PPP2R3B) (Dovega et al., 2014; Yan et al., 2000; Zwaenepoel et al., 2008), and B″γ (G5PR, encoded by PPP2R3C) (Xing et al., 2006; Zwaenepoel et al., 2008). B″ subunits are implicated in various cellular events, such as cell cycle progression (Voorhoeve et al., 1999; Wlodarchak et al., 2013; Yan et al., 2000), cell migration (Janssens et al., 2016), B-cell survival (Xing et al., 2005), neuronal signaling (Ahn et al., 2007b) and Wnt signaling (Creyghton et al., 2006). A sequence alignment of B″ family members is shown in Fig. S1B. Among the three isoforms, PPP2R3C has recently been associated with autosomal recessive mutations in four girls with syndromic 46, XY gonadal dysgenesis (Swyer syndrome), who also exhibited extragonadal abnormalities including neuromotor delay and ID (Guran et al., 2019). The parents of these individuals are healthy heterozygous carriers of the PPP2R3C pathogenic mutations L103P, L193S and F350S, of which F350S maps to the first EF hand (Guran et al., 2019) (Fig. 3C; Fig. S1). In the dysgenetic gonads of the affected individuals, SOX9 phosphorylation is greatly diminished (Guran et al., 2019). PKA- and PKC-mediated SOX9 phosphorylation have important roles in testicular development and chondrogenesis (Guran et al., 2019; Huang et al., 2000; Matta and Mobasheri, 2014), and the reduced SOX9 phosphorylation is therefore a plausible mechanism for gonadal dysgenesis and facial deformity in the affected individuals. These findings also further suggest that homozygous PPP2R3C mutations might enhance the catalytic activity of PP2A to impair SOX9 signaling in gonadal and bone development (Guran et al., 2019). Further studies are required to understand the effects of PPP2R3C mutations on PP2A holoenzyme assembly and activity.

PP2A_D also forms stable complexes with the large multidomain scaffolding proteins of the striatin family (termed the B‴ family), which consists of three members, striatin (STRN, encoded by PPP2R6A), STRN3 (encoded by PPP2R6B) and STRN4 (encoded by PPP2R6C) (Hwang and Pallas, 2014; Moreno et al., 2000). Striatins contain four protein–protein interaction domains – a caveolin-binding domain, a coiled-coil domain, a calmodulin-binding domain, and a WD-repeat domain, which folds into a β-propeller structure (Bartoli et al., 1999; Castets et al., 2000). The coiled-coil domain mediates striatin dimerization and its association with PP2A-A and PP2A-C (Chen et al., 2014). Striatins are also the eponymous core components of striatin-interacting phosphatase and kinase (STRIPAK) complexes, which include kinases and other signaling and regulatory proteins besides PP2A-A and PP2A-C. A recent cryo-EM structure of human STRIPAK identifies a non-canonical PP2A complex that comprises of four copies of STRN3 coiled-coil, one PP2A_D, STRIP1 and MOB4 (Fig. 1) (Jeong et al., 2021).

STRN and STRN4 are major components of the synaptosomal brain fraction (Kuck et al., 2019), and regulate dendritic arborization and spine development in striatal neurons and hippocampal neurons, respectively (Li et al., 2018; Lin et al., 2017). Consistent with them having important roles in synaptic development and function, downregulation of striatin has been reported to impair locomotor activity of rats (Bartoli et al., 1999). The first case of a de novo missense variant of an alanine to valine change in STRN was recently reported to cause a mild bipolar disorder BD (II) (Toma et al., 2020). A recent review has described the function of STRIPAK complexes in numerous developmental signaling pathways (Kuck et al., 2019), and the recent cryo-EM structure of human STRIPAK has revealed that mutations within the interface of the non-canonical PP2A complex are the cause of aberration to the hippo signaling pathway (Jeong et al., 2021). Mutations in STRIPAK complexes have been identified in individuals with autism (Iossifov et al., 2012), specifically a de novo 2-bp deletion in CTTNBP2 causes a frameshift and disruption of protein function (Iossifov et al., 2014). CTTNBP2 is stably localized to neuronal dendrites regulating spinogenesis, and this mutation likely causes an enhanced CTTNBP2 function as evidenced by increased spined density in ASD patients (Chen et al., 2012). Another STRIPAK protein, CCM3, is mutated in familial cerebral carvernous malformations, and causes seizures, recurrent hemorrhages, strokes and focal neurological deficits (Goudreault et al., 2009; Bergametti et al., 2005; Guclu et al., 2005). In addition to neurological disorders, roles of striatins and STRIPAK in heart diseases, diabetes and cancer have been extensively studied and reviewed (Hwang and Pallas, 2014).

PP2A has important roles in the development of the nervous system as evidenced by the plethora of recently discovered DNMs associated with NDDs. As discussed in this Review, these mutations occur in the majority of PP2A subunits (A, B and C) and their regulators, most of which are ubiquitously expressed, yet cause largely neurodevelopmental problems. Patients are generally heterozygous for these de novo mutations, indicating either haploinsufficiency of the disease gene, or a dominant-negative or toxic gain-of-function of the mutated gene product. Genetically engineered animals carrying the human mutations will provide useful model systems to tease out the specific roles and substrates of the different PP2A holoenzymes during neurodevelopment. In addition, the use of patient-derived induced pluripotent stem cells, and their differentiation into neurons and organoids, offers complementary opportunities for discovery and validation. Recent advances in phospho-proteomics and other unbiased approaches have finally allowed the protein phosphatase field to catch up with the protein kinase field in terms of insights into regulation and substrate specificity (Peti et al., 2013; Wang et al., 2020; Xing et al., 2006, 2008). In combination with new animal and human model systems, this recently gained understanding can now be leveraged to consider therapeutic interventions into NDDs associated with PP2A.

The authors acknowledge the support of the National Institute for Health Research, through the Comprehensive Clinical Research Network (https://www.ncbi.nlm.nih.gov/clinvar). This study makes use of data generated by the DECIPHER community. A full list of centres who contributed to the generation of the data is available from https://decipher.sanger.ac.uk/about/stats and via email from [email protected]. Funding for the DECIPHER project was provided by Wellcome.