Driver or passenger? A new assessment of genes in the schizophrenia-associated 3q29 deletion locus for contribution to neurodevelopmental disorders

Abstract

3q29 deletion (3q29Del) syndrome is caused by a 1.6 Mb copy number variant (CNV) located near the telomeric end of the long arm of the third human chromosome. Hemizygosity of this set of 22 protein-coding genes significantly increases risk for schizophrenia and autism spectrum disorders among other neurodevelopmental conditions, but it is not known which genes in this CNV interval are responsible for these phenotypes. We have evaluated existing literature and public genomic resources for this set of genes, categorizing them based on known cellular functions and assessed their potential as phenotypic drivers. We provide a comprehensive, synthetic review of the essential known functions of 3q29 deleted genes, and how multiple 3q29-encoded proteins may functionally interact. Our analysis reveals that ubiquitination/SUMOylation stands out among processes potentially compromised due to compound haploinsufficiency of four 3q29Del genes (UBXN7, FBXO45, RNF168, SENP5). The available genomic evidence indicates that no single gene in the 3q29 locus is solely responsible for the neurodevelopmental phenotypes of 3q29Del syndrome. Overall, we propose that functional, expression, and gene constraint evidence supports six genes (TFRC, UBXN7, FBXO45, PAK2, NCBP2, DLG1) as the most likely phenotypic drivers in 3q29Del syndrome. Haploinsufficiency of these proteins would likely disrupt metabolic, synaptic, and signaling mechanisms in developing and mature neurons, which collectively may impair neural circuit differentiation and function.

Supplementary Information

The online version contains supplementary material available at 10.1186/s11689-026-09696-y.

Introduction

3q29 microdeletion (3q29Del) syndrome (OMIM # 609425) is a copy number variant (CNV) disorder that occurs in approximately 1:30,000–40,000 live births. This disorder is autosomal dominant, typically occurring de novo, and is caused by a hemizygous 1.6 Mb microdeletion near the telomeric end of the long arm of human chromosome 3 [1, 2]. The deleted segment (~ hg38 chr3:196,000,000–197,600,000) most commonly contains 22 protein-coding genes, but it is not known which of these genes, when hemizygously deleted, influence neural development or neuron and circuit function in the mature brain. Each of the 3q29Del genes may have a relatively consistent impact on brain development or physiology, or subsets of these genes may interact with one another as well as additional genes to yield a broader range of phenotypes. Indeed, 3q29Del-associated phenotypes vary widely from mild or subtle symptoms to major neurodevelopmental disability suggesting substantial interaction of 3q29Del genes [3, 4]. Nevertheless, these interactions as well as potential genetic or environmental modifying factors have not been identified. Understanding the developmental and neurobiological effects of 3q29Del could also have a broad impact as this CNV is the strongest known genetic risk factor for schizophrenia (SCZ) [5, 6], and may point to neurodevelopmental mechanisms of SCZ pathogenesis relevant to idiopathic cases.

The natural history of 3q29Del syndrome remains an active area of investigation, but several detailed phenotyping studies have begun to provide insights. Infants with 3q29Del often experience failure to thrive [7], followed by delayed development and mild to moderate intellectual disability (ID) [8]. Clinical characteristics of 3q29Del often include neurodevelopmental and psychiatric conditions, with deletion individuals experiencing approximately 19x greater chance of autism spectrum disorder (ASD) diagnosis and a 40x increased likelihood of developing SCZ [9]. Other comorbid neuropsychiatric conditions include attention deficit/hyperactivity disorder (ADHD), generalized anxiety disorder (GAD), and graphomotor weakness. Additional common symptoms include sleep disturbance, congenital heart defects, gastroesophageal reflux disease, feeding difficulties, and ocular and dental anomalies [7, 10]. These phenotypes are often shared across neurodevelopmental anomalies [11]. The reciprocal 3q29 duplication has also been described and is associated with intellectual disability and risk for ASD [12]. While the 3q29 duplication is apparently less severe, these associations nonetheless further underscore the neurodevelopmental sensitivity to gene dosage at this locus.

Several genes within the 3q29Del interval (Fig. 1) have been individually associated with ASD (gene.sfari.org), such as PAK2 [13], DLG1 [14], and TM4SF19 [15], but no studies have demonstrated comparable effect sizes to individually account for the entire risk conferred by 3q29Del. Oligogenic CNVs can be more complex than single-gene variants, but may be a more tractable way to understand the polygenic nature of complex neuropsychiatric conditions [16]. Previous studies of other variants have demonstrated that CNVs can elicit phenotypes in several ways [17]. Haploinsufficiency of a single driver gene within the deletion locus may account for all or most phenotypes as appears to be the case for SHANK3 at 22q13 (OMIM # 606232) [18, 19]. Alternatively, one gene may be associated with a certain phenotype and another gene may be responsible for a separate phenotype—sometimes referred to as “contiguous gene effects”—as in the Williams-Beuren syndrome (OMIM # 194050) locus (7q11.23). In 7q11.23 deletion, haploinsufficiency of the gene Gtf2ird1 was found to produce the auditory hyperacuity phenotype in mice [20] whereas loss of ELN causes the arterial disease supravalvular aortic stenosis [21]. Another possible mechanism is that compound haploinsufficiency of multiple genes within a CNV locus (to varying degrees) produces a phenotype. The craniofacial and neuroanatomical abnormalities that result from copy number variation at 16p11.2 and 22q11.2 seem to be produced by multiple genes in this manner [22, 23].

Fig. 1.

The genes in the 3q29Del locus are displayed in categories assigned based on known protein functions

To date, genomic studies do not support the hypothesis that any single gene within the 3q29Del locus is responsible for the entire phenotype and analysis of single gene loss-of-function or heterozygous mutations in multiple animal models have sought to determine which genes may be the primary phenotypic drivers. For example, mice heterozygous for the 3q29Del gene Dlg1, a candidate gene that encodes a multi-function synaptic signaling scaffold protein, did not recapitulate phenotypes of the full syntenic 3q29Del mouse which include spatial memory and social interaction deficits [24]. A combinatorial knockdown of Drosophila homologs demonstrated that diminished expression of the 3q29Del gene NCBP2, which encodes an mRNA cap-binding protein, exacerbates phenotypes associated with losses of other 3q29 genes [25]. A bioinformatic analysis of the human cortical transcriptome indicated that the 3q29Del gene UBXN7 may play a similar role as a cortical “hub gene” [26]. None of these three candidate genes can explain the entirety of ASD or SCZ risk conferred by 3q29Del.

Beyond top candidates, many of the genes in this locus remain poorly understood [26]. Revisiting available literature in combination with bioinformatic analyses provides insight into likely phenotypic disruptions due to 3q29Del. Clues as to which 3q29 genes may be essential for neurodevelopment could be uncovered through analysis of genomic data from healthy individuals. Thus, we have leveraged the publicly available Genome Aggregation Database (gnomAD) and the Allen Institute’s BrainSpan developmental atlas to analyze mutational constraint metrics and brain-specific mRNA expression for each protein-coding gene in the 3q29Del interval. The Loss-of-function Observed/Expected Upper bound Fraction (LOEUF) is a continuous variable estimate that reflects an approximation of the constraint faced by each gene. Smaller values indicate that less missense and nonsense variation is observed in genes than expected based on synonymous variation rates [27, 28], consistent with negative selection for loss or gain of function variants. Low LOEUF scores can be indicative of broad essentiality but does not specifically indicate relevance to neurodevelopment. When paired with the neurodevelopmental specificity of the BrainSpan database, a moderate-to-low LOEUF score with moderate-to-high levels of RNA expression suggests that the gene is both intolerant to loss-of-function variation and involved in neurodevelopmental processes, thus increasing the likelihood of being a phenotypic driver in 3q29Del. (Fig. 2; Table 1). This multifactorial analysis based on mutational constraint, neurodevelopmental expression levels, functional annotations, and computational estimates of haploinsufficiency [42] has led us to group these 22 genes into 4 Tiers. Tier 1 genes (6) are likely phenotypic drivers based on multiple lines of evidence. Next are those genes that are likely (Tier 2, 7 genes) or possibly (Tier 3, 3 genes) involved as modifiers of certain phenotypes. We have found no compelling evidence to support the involvement of the remaining 6 genes.

Fig. 2.

a The Loss-of-function Observed/Expected Upper bound Fraction (LOEUF) is plotted for each 3q29Del locus gene arranged by chromosomal position (centromere to telomere) from the Genome Aggregation Database (gnomAD) version 4.1. Asterisks indicate brain expression ascertained from BrainSpan [43] (b). Data in b are mean RPKM + SEM across brain region samples and individuals from BrainSpan

Table 1.

Protein-coding genes of the 3q29 deletion locus in chromosomal order with four layers of evidence for involvement in neuropsychiatric conditions: mutational constraint (LOEUF), probability of haploinsufficiency (pHaplo), human brain expression levels (BrainSpan), and clues from homozygous null mouse phenotypes. Driver Likelihood summarizes this evidence into four tiers (Tier 1 = strongest)

Gene	LOEUF (v4.1.1)	pHaplo1	Brain Expression	Homozygous Mouse Phenotype	Driver Likelihood
TFRC	0.541	0.888	+++	Lethal, E12.5 [29]	Tier 1
ZDHHC19	1.199	0.262	-	Male infertility [30]	No evidence
SLC51A	0.904	0.773	-	Disrupted bile biosynthesis [31]	No evidence
PCYT1A	0.799	0.838	++	Lethal, E3.5 [32]	Tier 2
DYNLT2B	1.123	NA	+++	Male infertility [33]	Tier 2
TM4SF19	1.168	0.566	+	Alterations in adipocytes [34]	Tier 3
UBXN7	0.271	0.979	+++	Lethal, pre-weaning (IMPC; [35])	Tier 1
RNF168	1.296	0.802	++	Immunodeficiency, multi-system effects [36]	Tier 2
SMCO1	2.087	0.577	-	Unknown	No evidence
WDR53	1.308	0.588	+	Altered metabolism (IMPC; [35])	No evidence
FBXO45	0.304	0.966	+++	Lethal, perinatal [37]	Tier 1
NRROS	0.984	0.819	+	Reduced lifespan [38]	Tier 3
PIGX	1.175	0.461	+++	Lethal, E12.5 (IMPC; [35])	Tier 2
CEP19	1.250	0.348	++	Obesity, metabolic syndrome [39]	Tier 3
PAK2	0.459	0.963	+++	Lethal, E12.5 (IMPC; [35])	Tier 1
SENP5	0.407	0.973	+++	No significant phenotypes (IMPC; [35])	Tier 2
NCBP2	0.672	0.899	+++	Lethal, E9.5 (IMPC; [35])	Tier 1
NCBP2AS2	NA	NA	-	Unknown	No evidence
PIGZ	1.479	0.747	++	Altered metabolism (IMPC; [35])	Tier 2
MELTF	0.981	NA	+	Unknown	No evidence
DLG1	0.440	0.993	+++	Lethal, perinatal [40]	Tier 1
BDH1	1.228	0.811	++	Lethal, perinatal, incomplete penetrance [41]	Tier 2

¹From Collins et al. [42]

Our bioinformatic analysis of 3q29Del genes as well as re-assessment of the literature on function of individual genes based upon mutant analyses in multiple models provides an up-to-date description of 3q29Del gene functional categories based on their inferred primary or obligate functions as was previously done for the neuropsychiatric disorder-associated CNV 22q11.2 [44]. These functional categories are aligned to possible cellular mechanisms that impact neural development and may therefore indicate neurodevelopmental pathways that are vulnerable to 3q29Del, which can be prioritized for experimental investigation.

Protein modifications

Ubiquitination and SUMOylation

Ubiquitination stands out as a broad cellular process that may be compromised by haploinsufficiency of multiple 3q29 genes. Ubiquitin and ubiquitin-like proteins (such as SUMO) are eukaryotic post-translational modifications that are essential for many cellular processes including regulation of protein homeostasis [45]. Interestingly, four genes within the 3q29Del interval are involved in this process and, while it is not yet clear if any genetic interactions exist between these genes, the ubiquitin-like system may be a point of vulnerability in 3q29Del. Numerous studies have demonstrated critical roles of other ubiquitination proteins in neuronal development [46]. For example, the E3 ubiquitin ligase encoded by UBE3A (15q11.2) illustrates a key connection between the ubiquitination system and neurodevelopmental disorders. Alterations in the dosage of this gene results in Dup15q syndrome (OMIM #608636) [47, 48], Angelman Syndrome (OMIM #105830), or Prader-Willi Syndrome (OMIM #176270), which are all associated with neurodevelopmental disorders [49]. Ube3a has been found to be required for experience-dependent plasticity and visual cortex circuit development [50]. Furthermore, genes encoding members of the Cullin family of E3 ligases have also been recently associated with neuropsychiatric risk and are discussed further below [6, 51].

UBXN7 (Tier 1)

UBXN7 (aka UBXD7) encodes the UBA-UBX domain protein family member UBXN7 [52], which is widely expressed across numerous tissues including the brain. As a UBA-UBX protein, UBXN7 can function as an adaptor for interacting with ubiquitinated proteins through its UBA domain, and to bind via its UBX domain to the p97 ATPase, which is involved in diverse cellular functions such as protein homeostasis, membrane remodeling, and chromatin regulation [53].

UBXN7 has been found to play a key role in the regulation of cellular hypoxia and antioxidant responses. The master regulator of the hypoxia response, HIF1α, is constitutively degraded in normoxic conditions by the CUL2^VHL E3 ubiquitin ligase complex [54]. UBXN7 was found to recruit ubiquitinated HIF1α to p97 to facilitate its degradation [55]. A lack of UBXN7 in 3q29Del may therefore be expected to compromise the cellular hypoxia response. Additional studies have found that UBXN7 is a promiscuous interactor with CUL1, CUL2, and CUL3-based E3 ubiquitin ligases [56]. These interactions are notable as rare variants in CUL1 were found to confer high risk for schizophrenia [6] and CUL3 has been strongly associated with risk for ASD [51, 57]. Moreover, CUL3 is part of the complex that regulates cellular levels of the antioxidant response protein NRF2 [58] whereas CUL2 regulates HIF1α [59]. UBXN7 binds to both complexes and the cellular protein level of UBXN7 has been found to dictate the balance of NRF2 [60] and HIF1α in immortalized cell lines [61].

Levels of UBXN7 are regulated by an E3 ubiquitin ligase called MUL1 [61] that is localized to the outer mitochondrial membrane and can signal mitochondrial status [62]. When MUL1 is inactive, levels of UBXN7 increase, which promotes HIF1α accumulation and a shift toward glycolysis to meet cellular energy demands [63]. However, when UBXN7 levels are diminished, as is the case in 3q29Del, NRF2 levels may increase, HIF1α levels may decrease, and cells shift toward mitochondrial oxidative phosphorylation (OxPhos) to meet energy needs [61]. These NRF2-HIF1α experiments were performed in engineered human embryonic kidney cells, and it is not known how 3q29Del would impact these pathways, particularly in the developing central nervous system. These studies are intriguing in light of the recent finding that 3q29Del human neural progenitor cells do not switch from glycolysis to OxPhos like intact cells from neurotypical controls [64]. UBXN7 may be predicted to be involved in the critical metabolic reprogramming phase of neurogenesis and neuronal differentiation.

A genetic screen aimed at identifying proteins with a role in maintaining stem cell identity through mitosis revealed that UBXN7 is involved in recruiting the p97 ATPase to ubiquitinated histones for rapid histone degradation and initiation of cell-type specific transcriptional programs [65], which may compromise the stability of the neural stem/progenitor cell state. Precocious neuronal differentiation from neural stem/progenitor cells has been described in several models of neurodevelopmental disorders including Fragile X Syndrome [66] and 16p11.2 deletion [67]. It is not yet known if haploinsufficiency of UBXN7 in 3q29Del would similarly impact neural differentiation trajectories.

Lastly, UBXN7 has the lowest LOEUF score of any 3q29Del gene (Fig. 2), and is expressed in human brain at particularly high levels prenatally [43]. Thus, UBXN7 is a strong candidate as a phenotypic driver gene and may be involved in multiple aspects of 3q29Del associated neurodevelopmental phenotypes.

RNF168 (Tier 2)

RNF168 encodes an E3 ubiquitin ligase that plays a critical role in DNA double-strand breaks (DSBs) [68–70]. In the case of a DSB, RNF168 cooperates with RNF8 to ubiquitinate histone proteins and, like UBXN7 above, recruits the ATPase p97 to the site, though for a different purpose [71]. Homozygous loss-of-function mutation of RNF168 results in RIDDLE syndrome, a human immunodeficiency disorder caused by dysfunctional DNA repair [72], though it is not clear if hemizygosity of RNF168 (as in 3q29Del) is associated with known phenotypes. RNF168 has also been identified as a candidate gene for both Parkinson’s disease [73] and Crohn’s disease [74]. RNF168 has not been found to be significantly reduced at the protein level in 3q29Del mice due to apparent post-transcriptional compensation [24] and LOEUF analysis and relatively low expression in developing and adult brain indicates it is unlikely to be a phenotypic driver in the hemizygous state. However, due to its involvement in ubiquitination pathways, hemizygotic effects of RNF168 may be amplified by loss of other 3q29 genes.

FBXO45 (Tier 1)

FBXO45 encodes a member of the F-box protein family, that typically functions as an E3 ubiquitin ligase via the formation of SCF (Skp1, Cullin, F-box) complexes [75]. However, FBXO45 was found to not form SCF complexes and instead binds to the RING finger-type ubiquitin ligase PAM (aka MYCBP2) [37]. One ubiquitination target of the FBXO45 complex was reported to be the conserved pro-apoptotic transcription factor p73 [76]. Biochemical subcellular fractionation, immunocytochemistry, and immunoelectron microscopy experiments demonstrated that Fbxo45 localizes to pre- and post-synaptic compartments of neurons [77]. In mice, Fbxo45 protein was found almost exclusively in the nervous system. While Fbxo45 heterozygous mice showed no phenotypic differences from their wild type counterparts, full knockout of Fbxo45 resulted in perinatal lethality due to respiratory distress [37]. Fbxo45-null mice also showed significant abnormalities in synapse formation, diaphragm innervation, axon development, and neuronal migration indicating that the protein could have a key role in neurodevelopment [37]. Knockdown of Fbxo45 in hippocampal neurons increased excitatory postsynaptic currents, possibly due to a disruption of Fbxo45-mediated ubiquitination of the synaptic vesicle-associated protein Munc13-1 [77]. In addition to its role in ubiquitination, Fbxo45 was surprisingly found to be secreted and to bind to N-cadherin [78].

The synaptic localization of FBXO45 and its link to the Tuberous Sclerosis Complex proteins via PAM/MYCBP2 [79] prompted sequencing of FBXO45 in an ASD trio study, but no variants were identified that were over-represented in ASD probands [80]. A sequencing study in individuals with SCZ found one rare variant in FBXO45 [81], though this variant has not been identified in larger, subsequent studies [6]. A genome-wide association study reported that FBXO45 may also have a role in emotional regulation [82].

Based on its synaptic localization and essential role in mammalian CNS development FBXO45 is an intriguing candidate driver gene. Gene constraint analysis from gnomAD indicates that it is in the top decile of the most constrained human genes and ranks second in the 3q29 locus (to UBXN7), and thus likely to be haploinsufficient. Further studies are needed to understand how loss of one copy of this gene may disrupt neurodevelopment in 3q29Del syndrome.

While each of these three proteins appear to handle largely distinct roles in ubiquitination, it is likely that we do not understand the full range of their functions, which may overlap in critical ways in specific cell types or states at certain phases of development.

SENP5 (Tier 2)

SENP5 encodes a protease for the SUMO protein (small ubiquitin-like modifier) that is involved in mitosis [83, 84]. SUMO proteins are covalently attached as a post-translational modification, which can be reversed by SUMO-specific proteases such as SENP5 [85]. SUMOylation impacts numerous cellular processes like DNA repair, protein stabilization, apoptosis, and signal transduction [86]. SENP5 knockdown by shRNA was found to alter mitochondrial morphology and increase reactive oxygen species production [87]. Further investigation revealed that SENP5 shuttles to mitochondria during cell division and is involved in mitochondrial fragmentation [88]. SENP5 is not among canonical mitochondrial proteins [89], and this cell cycle-dependent translocation behavior may explain why. Mitochondrial fragmentation is also an essential phase of apoptosis, which involves SUMOylation of the mitochondrial fission protein DRP1 [90]. Increased SUMOylation of DRP1, which could result from a lack of SENP5, would increase the likelihood of initiating apoptotic signaling cascades [87]. Interestingly, the E3 ubiquitin ligase MUL1 (aka MAPL), which was found to ubiquitinate and regulate levels of UBXN7 (see above), also functions as an E3 SUMO ligase that SUMOylates DRP1 [87]. The effect of haploinsufficiency of both UBXN7 and SENP5 on mitochondrial dynamics and function is not clear.

An examination of Senp proteins Senp3 and Senp5 in mouse brain found that Senp3 expression is restricted to cell nuclei whereas Senp5 was found in the neuropil and to co-localize with presynaptic, postsynaptic, and mitochondrial markers [91]. Further studies found that the balance between two Senp5 splice variants, Senp5L and Senp5S, regulates the balance of mitochondrial fission/fragmentation as the smaller “S” isoform lacks the protease catalytic domain and competes with full-length Senp5 for substrate binding [92]. Additionally, in vivo knockdown of Senp5 was found to disrupt the radial migration of neurons into the cortical plate. These findings suggest that SENP5 could be involved in 3q29Del phenotypes and that compound hemizygosity of SENP5 and UBXN7 could be more deleterious than loss of one copy of either gene alone.

PAK2 (Tier 1)

PAK2 is a ubiquitously expressed serine/threonine protein kinase that is crucial to numerous signaling pathways such as apoptosis [93], cell division, cytoskeleton regulation, and cell motility [94]. Interestingly, BrainSpan data shows very high prenatal expression of PAK2 (Fig. 2b), especially at very early phases of neurodevelopment (Supp. Figure 1) that diminishes postnatally.

PAK2 is typically activated by small GTPases such as Rac1 and Cdc42 [95], but can also be activated by the TSC and mTOR-related small GTPase Rheb [96]. Heterozygous Pak2 knockout in mice resulted in autism-like behaviors, defective long-term potentiation (LTP), and reduced spine density [13], and it is now considered a strong ASD candidate gene. Notably, heterozygous Pak2^+/− deletion did not recapitulate all behavioral phenotypes of 3q29Del mice [24, 97]: spatial memory and startle responses appeared to remain intact. However, diminished Pak2 expression in cortex and hippocampus of Pak2^+/− mice was associated with reduced phosphorylation of Limk1 and Cofilin-1, which are key regulators of dendritic spine cytoskeletal dynamics and possibly underlying the reported reduction in hippocampal dendritic spines and theta burst-induced LTP [13]. LIMK1, which is directly activated by PAK2 [98], is encoded by LIMK1 in the Williams-Beuren Syndrome locus 7q11.23 (deletion), illustrating a potential molecular link between these CNV syndromes that are both associated with social differences.

PAK2 also has an emerging connection to cellular and organismal energy homeostasis. This metabolic role for PAK2 could be related to direct phosphorylation by the glycolytic enzyme pyruvate kinase M2 [99] and the cellular energy balance sensor AMPK [100, 101]. Fasting increases AMPK-mediated PAK2 phosphorylation in the mouse arcuate nucleus of the hypothalamus, a crucial food intake signaling hub [101]. At the cellular level, genetic or pharmacologic disruption of PAK2 can shift cells toward more oxidative phosphorylation-based metabolism [102] and overexpression of PAK2 can promote aerobic glycolysis and cell proliferation [103]. The full 3q29Del was found to impair metabolic flexibility in developing neural progenitor cells and complete knockout of PAK2 in HEK cells partially recapitulated that phenotype [64].

While complete knockout of Pak2 in mice results in embryonic lethality due to neural tube defects [94, 104], Schwann cell-specific knockout was found to severely disrupt myelination, decrease nerve conduction velocity, and produce peripheral nerve dysfunction [105]. A biallelic PAK2 missense mutation was linked to Knobloch Syndrome type 2, a rare disorder that presents with severe vision problems and craniofacial malformation [106]. PAK2 has also been linked to developmental language disorder [107], cardiac, inflammatory, and immunological diseases [108], and is an autosomal homolog of PAK3, a gene linked to neurodevelopmental delays and behavioral abnormalities [109]. The LOEUF score for PAK2 is moderate at 0.662 but findings from the heterozygous mice indicate that heterozygous loss of Pak2 alone disrupts neurodevelopment and therefore is likely involved in at least some phenotypic manifestations of 3q29Del Syndrome.

DLG1 (Tier 1)

DLG1 encodes a ubiquitously expressed, large, multi-functional protein necessary for multiple domains of development [110]. DLG1 (aka SAP97) is widely expressed in various cellular compartments and throughout many tissues, but is notably found at synapses and is involved in the stabilization and maturation of newly-formed dendritic spines [111]. As a member of the Membrane-Associated Guanylate Kinase (MAGUK) family, DLG1 can act as a signaling hub facilitating the interactions of multiple proteins including cell surface receptors and effector molecules. DLG1 has three PDZ (PSD95/DLG/ZO-1) domains and has been reported to bind to G protein-coupled receptors with PDZ binding intracellular C-termini including the β1-adrenergic receptor [112], ADGRB1 (BAI1) [113, 114], ADGRA3 (GRP125) [115], ADGRE5 (CD97) [116], the serotonin 5-HT2A receptor [117], the corticotropin-releasing factor 1 receptor [118], and Frizzled-4 [119]. These interactions open the possibility that reductions in DLG1 protein resulting from 3q29Del [24] could impact numerous signaling cascades relevant to neurodevelopment and adult circuit function.

The DLG1 gene produces multiple isoforms, which have differential effects on synaptic plasticity [120]. Indeed, DLG1 protein interacts with subunits of key glutamate receptors involved in synaptic plasticity. It binds to the NR2A subunit of NMDA receptors via one of its three PDZ domains. NMDA receptor activation promotes DLG1 phosphorylation by CaMKII, which disrupts the DLG1-NR2A interaction [121]. CaMKII phosphorylation also disrupts the interaction between DLG1 and the astrocytic glutamate transporter EAAT2 [122], possibly indicating a generalized mechanism of interactions.

DLG1 binds to the AMPA receptor subunit GluR1 and the complex travels together from the endoplasmic reticulum to postsynaptic compartments [123, 124]. DLG1 L27 domain-mediated interaction with the X-linked neurodevelopmental disorder associated protein CASK (calcium/calmodulin-dependent serine protein kinase) influences DLG1 conformation and interaction with AMPA or NMDA receptor subunits [125, 126]. In addition to glutamate receptor subunits, which form Na⁺ and Ca²⁺ channels, DLG1 interacts with the inward rectifier potassium channel Kir2.2 [127] and the voltage-gated potassium channel Kv4.2 at the post-synaptic density [128]. DLG1 is found in both pre- and post-synaptic compartments and, interestingly, postsynaptic expression of Dlg1 was found to affect presynaptic function [129]. DLG1 has also been found to have key non-synaptic roles in neural function. For example, DLG1 interacts with PTEN to inhibit AKT and serve as a brake on peripheral axon myelination [130].

For a ubiquitously expressed, multi-functional protein like DLG1, the cellular context is critically important for understanding specific functional roles. For example, a recent study found that knockdown of Dlg1 expression had no measurable effect on hippocampal area CA1 neurotransmission but dramatically increased AMPA receptor-mediated activity in the dentate gyrus [131]. Thus, whole-body heterozygosity of DLG1 as in 3q29Del may only elicit phenotypes in certain neural cell types and circuits.

Given the large interactome of DLG1 and its formation of signaling complexes, it has been hypothesized that DLG1 may interact directly or indirectly with other 3q29Del locus proteins. Indeed, there is suggestive evidence of a genetic interaction between DLG1 and PAK2 in a Drosophila model. Heterozygous deletion of either homolog dlg or pak did not elicit an observable phenotype but dlg/pak transheterozygotes were found to have disrupted sleep patterns and a loss of neuromuscular junction synapses [132].

The direct connection of DLG1 to schizophrenia risk, however, remains unclear. Among 3q29Del locus genes, DLG1 has received by far the most attention as a potential individual driver of neuropsychiatric disease risk, likely due to the extensive literature describing the role of DLG1 at synapses and the link between synaptic genes and schizophrenia risk [133, 134]. Notably, a Dlg1 heterozygous mouse was found to not recapitulate the behavioral phenotypes of the 3q29Del mouse model [24]. Rare, de novo mutations in DLG1 have been found in cohorts of individuals with schizophrenia [135] though larger, genome-wide analyses have not implicated rare DLG1 variants. Targeted analysis of single-nucleotide polymorphisms (SNPs) in DLG1 have identified several variants enriched in schizophrenia cohorts, but it is unclear if these loci meet genome-wide significance thresholds [136–139]. The most recent schizophrenia GWAS found a SNP linked to DLG1 to be associated with modest (~ 3% increase) but significant risk for SCZ, though the impact of this SNP on gene expression or protein function has not been reported [134]. Genomic (LOEUF) evidence indicates that DLG1 may be mutationally constrained, though the complexity of its many isoforms and extensive protein-protein interactions could complicate this analysis. Interestingly, BrainSpan data indicates that DLG1 expression is rather modest prenatally and increases postnatally (Fig. 2). Together, the evidence suggests that DLG1 is likely involved as a phenotypic driver gene though haploinsufficiency of DLG1 alone is not responsible for the entirety of 3q29Del syndrome.

PIGX and PIGZ (Tier 2)

Two Phosphatidyl Inositol Glycan genes, PIGX and PIGZ are found in the 3q29Del interval, both of which encode proteins that are involved in mannose attachment to glycosylphosphatidylinositol (GPI) anchors and are expressed during brain development (Fig. 2). PIGX encodes a subunit of the GPI mannosyltransferase complex. GPI is a glycolipid that anchors numerous proteins to the cell surface [140]. Within the GPI complex, the function of PIGX is to stabilize the PIGM enzyme [141].

PIGZ is responsible for attaching a fourth mannose to the GPI complex. mRNA analysis has revealed that PIGZ is expressed across most human tissues, with the highest levels found in the brain and colon [142]. Mouse knockout strains of PIGZ resulted in increased levels of amyloid beta peptide, a key factor in Alzheimer’s disease [143]. While it is intriguing that two 3q29Del locus genes are involved in the process of GPI anchor biosynthesis, it remains unclear whether hemizygosity of both genes would have any deleterious effects in the developing nervous system. However, at least 19 GPI anchor-related genes are known to cause GPI biosynthesis defect disorders which can result in seizures and developmental delay [144] though only when both alleles are mutated. While gnomAD analysis does not indicate either PIGX or PIGZ are constrained, it is possible that hemizygosity of both genes could disrupt GPI anchor biosynthesis. Further studies would be required to test this possibility.

ZDHHC19 (No evidence)

ZDHHC19 encodes a protein found in the Golgi membrane, endoplasmic reticulum, and perinucleolar compartment, with high expression in adult testis and minimal to no expression in other tissues [30]. Knockout studies in mice revealed that ZDHHC19 has a role in male fertility. Spermatozoa in knockout mice presented with defects of the head and tail, impacting motility and oocyte fertilization [145]. The available evidence and gnomAD analysis indicate it is unlikely that hemizygosity of ZDHHC19 would impact neurodevelopment.

Metabolism

SLC51A (No evidence)

SLC51A (aka OSTA) encodes a protein involved in intestinal transport of bile acids (aka OST-α). SLC51A also has the ability to transport steroids through the hepatic system [146]. Knockout of mouse homolog Osta showed no physical differences from wild-type but signs of disrupted bile biosynthesis were identified [31]. In humans, biallelic deficiency of OST-α results in severe hepatic dysfunction, including congenital diarrhea and fat malabsorption [147, 148]. Public databases indicate that SLC51A expression is restricted to the adrenal gland, gut, and liver.

PCYT1A (Tier 2)

PCYT1A encodes the enzyme CTP: phosphocholine cytidylyltransferase A (CCTa/PCYT1A) [149]. PCYT1A regulates the formation of phosphatidylcholine (PC) via the Kennedy pathway [150] and also maintains PC levels within cellular and nuclear membranes. PCYT1A is expressed in all tissues but is highly concentrated in the liver, kidney, and heart [151]. Biallelic mutations of PCYT1A can cause spondylometaphyseal dysplasia with cone-rod dystrophy, a rare disorder that presents with musculoskeletal malformation, delayed growth, and progressive vision loss [152]. Other vision-related disorders have been linked to PCYT1A, such as Leber congenital amaurosis (LCA) and retinal dystrophy [153]. PCYT1A mutations have also been linked to congenital lipodystrophy and fatty liver disease, which are caused by the disruption of phosphatidylcholine synthesis [154]. Complete knockout of Pcyt1a in mice resulted in embryonic lethality by day 3.5 and a failure to implant. Heterozygous adult mice had decreased levels of Pcyt1a in the liver and other tissues, while embryos displayed signs of delayed development at the morula stage [32].

BDH1 (Tier 2)

BDH1 is a member of the short-chain dehydrogenase/reductase gene family and utilizes phosphatidylcholine to initiate enzymatic activity [155], illustrating a potential connection to PCYT1A haploinsufficiency. BDH1 is highly expressed in the liver and gut and initiates the catalysis of ketone bodies formed in the liver, and its dysfunction can lead to jaundice [41]. Ketone bodies serve as an alternative fuel source in glucose-dependent tissues, like the brain, when glucose is absent. In mice, Bdh1 assists in the reduction of cardiac stress following an apical cardiac infarction [41], and attenuates diabetes-induced atherosclerosis [156]. In light of recent findings of altered fat metabolism in 3q29Del mice [157] and mitochondrial energy metabolism dysfunction in human 3q29Del cells [64], BDH1 is a reasonable candidate to contribute as a phenotypic driver gene. In addition, it is notable that BDH1 is the only 3q29Del gene that is listed in the Mitocarta3.0 inventory of mitochondrial proteins [89]. However, gnomAD (v4.1) data does not indicate genic constraint on BDH1, suggesting that BDH1 may interact with other 3q29Del genes to produce metabolic and/or mitochondrial phenotypes.

Iron homeostasis

TFRC (Tier 1)

TFRC encodes the transferrin receptor (TFRC, aka TFR1), which binds to and internalizes iron-bound transferrin and thereby mediates the primary route for cellular iron import [158]. Homozygous Tfrc knockout mice are not viable past embryonic day 12.5, suffering from severe anemia, edema, necrosis, and neurological abnormalities [29]. Heterozygous Tfrc knockout animals were reported to be grossly normal but were found to have reduced iron levels in liver and spleen and evidence of iron deficiency in erythrocyte analysis. Barring compensation from another 3q29 locus gene – or chromosome and gene-regulatory differences in the two models – 3q29Del mice could be expected to have similar phenotypes, though this has not been reported. Homozygous Tfrc knockout in neural crest cells (Wnt1^Cre) led to severe embryonic craniofacial abnormalities and reduced TGF-β and BMP signaling [159], which could impact neural differentiation and axon specification [160, 161]. Knockdown of TFRC in human neural progenitor cells was found to decrease levels of mitochondrial fission-associated proteins DRP1 and FIS1 and increase levels of fusion-associated proteins OPA1 and MFN2 [162]. Furthermore, TFRC was found to interact with GSK3β, another signaling molecule important for neurogenesis, neuronal polarization, and axon growth [163] and levels of GSK3β were found to correlate with TFRC protein levels [162]. Additional experiments will be required to determine how hemizygosity of TFRC as in 3q29Del may impact these pathways in developing neural cells.

From a translational perspective, iron deficiency, which is a potential risk for individuals with 3q29Del syndrome due to hemizygosity of TFRC, is associated with a similar spectrum of neurodevelopmental syndromes as 3q29Del including significantly elevated risk for schizophrenia [164–166]. The mechanisms by which iron deficiency increases risk for these NDDs in humans are not clear, but studies in mammalian experimental systems have shown that genetic or environmental iron deficiency can impair neuronal metabolism and maturation [167–169]. Further studies are needed to understand if 3q29Del indeed increases neurodevelopmental vulnerability to iron deficiency and if iron deficiency could be an environmental mediator of 3q29Del phenotype severity.

MELTF (No evidence)

MELTF (aka MFI2) encodes the cell-surface expressed, iron-binding glycoprotein melanotransferrin (MTf) [170]. MTf is highly conserved across species and its genetic sequence shares 40% similarity to other iron-binding proteins such as lactoferrin and transferrin [171]. While it is intriguing that a second 3q29 locus gene is potentially involved in iron homeostasis – and could represent a “second hit” to the pathway – it remains unclear whether MTf participates in cellular iron homeostasis, particularly in the brain [172].

Immune System

TM4SF19 (Tier 3)

Tetraspanins are integral membrane proteins involved in numerous cellular functions, such as proliferation, apoptosis, migration, and signal transduction, while also critical for cell-cell fusion [173]. TM4SF19 is a tetraspanin family protein found to regulate osteoclast function [174]. TM4SF19 has been shown to be a lysosomal membrane protein expressed in macrophages, involved in reduction of inflammation of adipose tissue. Tm4sf19 null mice displayed altered metabolic function, with KO mice showing improved glucose and insulin tolerance [34]. Interestingly, splice-site and missense variants of TM4SF19 have been identified in ASD candidate gene research [15, 175], indicating that this poorly-annotated protein deserves further investigation.

NRROS (Tier 3)

NRROS is a leucine-rich repeat containing transmembrane protein that was named for its function as a negative regulator of reactive oxygen species in immune cells [176]. Complete knockout of Nrros resulted in enhanced bactericidal activity by phagocytes but oxidative damage to the CNS [176]. In fact, knockout of Nrros resulted in early mortality, developmental delays, and spontaneous neurological disorders in mice [38] and biallelic human mutations were found to cause demyelination, neuronal loss, and astrogliosis [177]. NRROS expression is limited to microglia in the parenchyma of the central nervous system and is most critical at embryonic and postnatal stages of development [177]. Homozygous or compound heterozygous mutations of NRROS have been linked to early-onset seizure disorders in humans, with brain calcification and neurodegeneration [177, 178]. LOEUF does not indicate that NRROS is a constrained gene (Fig. 2) and further experimentation will be required to determine if hemizygosity of NRROS as in 3q29Del would contribute to neuropsychiatric phenotypes.

Ciliary

DYNLT2B (Tier 2)

DYNLT2B (aka TCTEX1D2) is part of a family of cytoplasmic dyneins responsible for intracellular motility and is specifically necessary for retrograde ciliary transport [179]. Skeletal ciliopathies are commonly associated with mutations in this gene family. DYNLT2B has been linked to Short-Rib Thoracic Dysplasia-17 (SRTD17) with or without polydactyly, a group of autosomal recessive disorders that include Ellis-van Creveld syndrome (EVC), Jeune syndrome, and asphyxiating thoracic dystrophy (ATD). SRTD17 presents with numerous skeletal abnormalities like extra fingers and toes, a bell-shaped chest, shortened ribs and limbs, and an abnormal pelvis. Symptoms also can include renal dysfunction, and pancreatic, hepatic, and retinal abnormalities [180]. ATD is the most severe SRTD17 disorder, causing severe hypoxia that often leads to death in early childhood [181]. Knockdown of DYNLT2B in zebrafish resulted in numerous physiological abnormalities, including ventrally curved bodies, hydrocephalus, malformed otoliths, and small eyes [182]. Few embryos survived past four days post fertilization; those that did displayed abnormal craniofacial formation, severe edema, pronephric cysts, and mild developmental delays [182]. LOEUF analysis does not indicate DYNLT2B is constrained or a likely phenotypic driver, though it appears to be expressed at very high levels in the developing human brain (Fig. 2, Supp. Figure 1). It is not known if hemizygosity of this gene in combination with another 3q29 gene, such as CEP19 (below), could contribute to human neurodevelopmental phenotypes.

CEP19 (Tier 3)

CEP19 is expressed throughout the body and at high levels in the testes. CEP19 is also a ciliary protein, forming a pathway with RABL2 to activate the formation of cilia and is necessary for microtubules to anchor along centrosomes [183]. A human familial study suggested that heterozygous mutation of CEP19 has no metabolic impact, however homozygous carriers displayed symptoms of metabolic disorder, most commonly Type 2 diabetes, and males had little to no sperm count [39]. CEP19 homozygous truncating mutation has also been linked to Bardet-Biedl syndrome (BBS), which is a rare disorder resulting in the dysfunction of primary cilia [184]. Symptoms of BBS include obesity, kidney abnormalities, intellectual disability, developmental delays, metabolic disorders, and skeletal malformations of the fingers and toes. Homozygous knockout in mice results in spermatogenic failure as well as morbid obesity, while heterozygous mice displayed no apparent differences compared to wild-type animals [39]. While the connection to glucose metabolism is intriguing given metabolic abnormalities in 3q29Del [157], it is not yet clear how hemizygosity of this gene would contribute to 3q29Del phenotypes.

RNA processing

NCBP2 (Tier 1)

NCBP2 (aka CBP20) encodes a nuclear cap binding protein involved in the process of binding monomethylated caps at the 5’ end of pre-mRNAs, which allows mRNA to bind to ribosomes in the cytoplasm [185, 186]. It is highly expressed in the human brain, particularly during prenatal development (Fig. 2). Cap-binding complexes are conserved from yeast to primates and are required for cell growth and proliferation [187]. NCBP2 works directly with the RNA cap [188], is believed to be recruited and stabilized by NCBP1 [189], and has been found to regulate the binding of NCBP3 [190]. Ncbp2 homozygous deletion results in embryonic lethality prior to organogenesis in mice, and heterozygosity was found to possibly increase anxiety (International Mouse Phenotyping Consortium; IMPC) [35].

In human cells, NCBP1 knockdown resulted in loss-of-function while silenced NCBP2 was apparently compensated for by NCBP3 [188]. In a Drosophila and Xenopus study, homologous knockdown of NCBP2 resulted in enhanced phenotypes in combinatorial knockdown of other 3q29 homologs [25], though Drosophila are not known to express a homolog of the potential compensatory protein NCBP3. Loss of NCBP2 could have a profound effect on transcript export, but it is not clear that protein levels are reduced. In at least one study in mouse brain, Ncbp2 mRNA levels were reduced to match gene copy number but Ncbp2 protein levels were not significantly altered compared to wild type, suggesting a post-transcriptional compensatory mechanism [24].

Unknown functions

SMCO1, WDR53, and NCBP2AS2 are genes of essentially unknown function with little genomic evidence of mutational constraint or brain expression. NCBP2AS2 encodes a protein that was found to be induced by hypoxia in tumors and secreted to promote angiogenesis [191], though it is unclear if it has a role in the CNS.

Discussion

In the quest to understand the biological basis of neurodevelopmental psychiatric conditions, rare CNVs such as 3q29Del have emerged as promising and important targets for investigation. The high disease risk conferred by 3q29Del suggests that this variant would produce strong, observable biological effects using experimental systems such as mouse models [24, 97] and induced-pluripotent stem cell lines [64]. These observations will lead to a better understanding of the pathogenesis of this genetic disorder and may produce new insights into the underlying biology of conditions such as SCZ and ASD. While 3q29Del syndrome certainly qualifies as a rare genetic disorder, estimates indicate a similar prevalence to Huntington’s disease (1:~20,000) [192] and Rett Syndrome (1:10–20,000 females) [193], which have been much more widely studied. PubMed results for ‘Huntington’s’ (> 19,000) and ‘Rett Syndrome’ (> 4,800) far surpass those for ‘3q29’ (~ 250). Thus, there is much more to learn about the natural history and phenotypic expression of 3q29Del syndrome, in addition to molecular neurobiology, which will be essential to designing evidence-based interventions to improve well-being and clinical outcomes for people living with this condition.

The oligogenic nature of CNVs increases the complexity of experimental analysis, but also likely makes CNVs a more natural model of disorders such as SCZ, which are considered to have a polygenic basis [16]. Distinguishing potential driver genes from likely “passengers” (i.e. hemizygous genes that do not contribute to phenotype expression) is an important goal for this line of investigation. The available evidence does not support the concept of a single driver gene in the deletion interval producing all or even most neurodevelopmental phenotypes. Therefore, developing a better appreciation of how deficits of multiple genes may impinge on common cellular pathways or in certain cell types will improve our understanding of 3q29Del Syndrome and the basis for psychiatric disease risk associated with this variant.

We conclude that several potential cellular processes may be disrupted by loss of multiple 3q29 genes including ubiquitination/SUMOylation pathways (UBXN7, RNF168, FBXO45, SENP5), immune functions (NRROS, TM4SF19), and metabolism (BDH1, PCYT1A, TFRC). Multiple “hits” to these pathways could have different effects in certain cell types and at discrete stages of development adding to the complexity and opportunity to discover new neurobiology with relevance to human neurodevelopmental and psychiatric conditions.

To provide insight into the functional imperative for each 3q29 gene as a foundation for further analysis, we have included gene constraint data from the gnomAD database along with gene expression data from BrainSpan to aid in prioritizing potential driver genes. These analyses, in addition to in-depth literature review of functional annotations have led us to prioritize six genes – TFRC, UBXN7, FBXO45, PAK2, NCBP2, and DLG1 – as likely (Tier 1) drivers of 3q29Del-associated neurodevelopmental phenotypes. This conclusion aligns closely with a genome-wide analysisthat applied machine-learning approaches to estimate the probability of haplo- or triplosensitivity [42]. They concluded that these same six genes plus SENP5 are likely sensitive to haploinsufficiency, though that analysis was not focused specifically on neurodevelopmental disorders. Notably, only two 3q29 genes – PAK2 and UBXN7 – were predicted to be triplosensitive, which is consistent with findings that the 3q29 duplication is associated with less severe neurodevelopment outcomes [12, 194].

These analyses, naturally, have limitations. The BrainSpan developmental transcriptome does not provide single-cell resolution, which can obscure the contribution of genes such as NRROS and TM4SF19 that seem to be expressed only in certain cell lineages. Moreover, constraint metrics are somewhat limited by gene-level analysis that does not account for compound haploinsufficiency. For example, neither BDH1 nor PCYT1A show evidence of mutational constraint in gnomAD (v4.1.0), but PCYT1A produces phosphatidylcholine, which is required for BDH1 enzymatic activity. Thus, the combined loss of these genes in 3q29Del could have additive or synergistic effects on this pathway. Indeed, a limitation of our analysis is that gene-gene interactions can be unpredictable from single gene functional annotations.

Experimental directions

Gene functional annotations are continually improving but experimentation will be necessary to resolve mechanisms of multi-gene interactions in specific cell types. For example, compound haploinsufficiency of UBXN7 and TFRC could compromise cellular hypoxia responses in the developing CNS. The balance of proliferation and differentiation in radial glia is regulated in part by hypoxia signaling [195]. Hemizygosity of TFRC may lead to cellular iron deficiency, which can initiate hypoxia signaling and stabilization of HIF1α. UBXN7 is involved in regulating levels of HIF1α [54, 55]. How do these or other compound hemizygosities due to 3q29Del alter neural circuit development or function? There is evidence that reduction in Dlg1 expression specifically alters glutamatergic function in the hippocampal dentate gyrus [131]. Would diminished Pak2 and Fbxo45 expression in 3q29Del further disrupt this circuit?

Mouse models of 3q29Del exhibit reduced brain mass [24, 97] and recent human neuroimaging studies have reported striking reductions in cerebellar volume [10, 196]. It is not yet known if specific cell types or circuits are compromised in these phenotypes or if these are global phenomena in 3q29Del. Mutations in CASK, which encodes a key binding partner of DLG1 [197], are known to cause profound cerebellar hypoplasia [198]. Alternatively, recent findings of altered neural progenitor metabolism [64] could be extrapolated to a deficit in proliferation and differentiation. The human cerebellum may be particularly vulnerable to a reduction in neural progenitor proliferation capacity due to the exponential expansion of granule cell precursors required for healthy development that far exceeds that of mice [199]. Understanding these mechanisms will require further in vivo and in vitro investigation in multiple experimental systems. Approaches such as CRISPRi may help uncover gene-gene interactions [200], and cell-lineage specific mouse driver lines can illuminate gene contributions in certain cell types.

Conclusion

Population-level genomic studies have not yet identified any single gene within the 3q29Del interval as a strong risk factor for SCZ or ASD, which suggests that hemizygosity of multiple genes in the deletion locus is required. Our analyses suggest a model in which haploinsufficiency of multiple 3q29 genes converges on several cellular pathways including ubiquitination, metabolism, and synaptic function. These early cellular perturbations may alter neural progenitor behavior, fate specification, and circuit function, ultimately contributing to the neurodevelopmental and psychiatric phenotypes associated with the deletion.

Authors’ contributions

ARH and RHP wrote the manuscript text and Table 1. RHP prepared the figures. Both authors reviewed the manuscript.

Funding

This work was supported by National Institute of Mental Health project K01MH133970 (R.H.P.).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare no competing interests.