Human neuronal stem cells (hNSCs) from two genetic backgrounds (SA001/hNSC₁ and GMO1869/hNSC₂) were obtained from I-Stem and are described in previous studies (Boissart et al., 2012; Courraud et al., 2023). hNSCs were seeded on poly-ornithine-and laminin-coated dishes and maintained in N2B27 medium (DMEM/F12 and Neurobasal medium [1:1] supplemented with N2, B27, 2-mercaptoethanol [all from Invitrogen]), BDNF (20 ng/mL), FGF-2 (10 ng/mL; both from PeproTech), and EGF (R&D Systems; 10 ng/mL). Culture and quality controls of hNSCs were performed as described in Quartier et al. (2018). hNSCs were transfected using INTERFERin reverse transfection protocol (Polyplus-transfection), with Scramble siRNA, a pool of DYRK1A siRNA (120 nM final concentration), or the transfecting agent only. Cells were harvested 48 h after transfection for RNA and protein extractions. All cell lines were grown at 37°C in 5% CO₂. For proliferation assay, reverse transfection of hNSCs (20,000 cells/cm2) was performed in 96-well plates using INTERFERin reagent and siRNA according to the manufacturer’s recommendations. Each day, a plate was fixed and stained with DAPI, and the number of nuclei in each condition was counted using CellInsight automated microscope and HCS Studio software (Thermo Fisher Scientific).

Proteins extracted from untreated hNSC₁ were immunoprecipitated with DYRK1A antibodies targeting the N-terminal and C-terminal regions of the protein, respectively (Cohesion Biosciences #CPA1357, Immunogen sequence aa 39-51 and Abnova H00001859-M01, aa 674–763). As negative controls, we used free beads without antibodies as well as mouse anti-rabbit (MAR, 211–002-171, Jackson ImmunoResearch) and rabbit anti-mouse (RAM, 315–005-044, Jackson ImmunoResearch) antibodies as described (Mattioli et al., 2019). Immunoprecipitations were validated using Western blot as previously described (Mattioli et al., 2019) before the mass spectrometry analyses (Proteomic platform, IGBMC). In brief, the samples were treated with LysC/trypsin for digestion and injected into Orbitrap ELITE/C18 Accucore 50 cm (20 μL 0.1%TFA/1 μL) for 2 h runs in triplicate. Data were processed using Proteome Discoverer 2.2 software with Homosapiens_190716_reviewed. fasta and contaminants_190528.fasta databases. Thresholds were set at 1% FDR with a minimum of two unique peptides per protein. To consider a protein as a candidate interactor, we applied the thresholds to keep only proteins with (1) peptide-spectrum matching (PSM) sum of any of the control conditions (Beads, MAR, GAR) inferior to 5, (2) a sum of PSM for the three replicates of the test conditions (DYRK1A C-term or DYRK1A N-term) greater or equal to 5 with a positive value for each replicate, and (3) a ratio of NSAF between anti-DYRK1A and control conditions ≥1.5. Enrichment analysis was realized using Database for Annotation, Visualization, and Integrated Discovery (DAVID) using proteins detected in the mass spectrometry experiments (minimum of two peptides per protein) as background. The list of already known DYRK1A interactors was obtained from BioGRID and from previously published proteomic studies (Varjosalo et al., 2013; Li et al., 2015; Menon et al., 2019; Roewenstrunk et al., 2019; Guard et al., 2019; Huttlin et al., 2021). Network analysis was performed using Poteo3Dnet.¹ Information about pathologies caused by mutations in genes encoding DYRK1A interactors was retrieved using Human Phenotype Ontology. The probability of the genes being intolerant to the loss-of-function variant was obtained from GnomAD (pLI). Gene expression profiles during brain development were obtained from BrainSpan data and were correlated with the DYRK1A expression profile (Pearson’s correlation).

Cells were lysed in RIPA buffer (50 mM Tris–HCl pH 7.5, 150 mM NaCl, 0.25% sodium deoxycholate, 1% NP-40) supplemented with protease inhibitor cocktail and phosphatase inhibitor cocktail. A total of 5 to 50 μg of protein lysate was separated on 10% SDS-PAGE and transferred to a polyvinylidene fluoride (PVDF) membrane. Membranes were blocked in 5% non-fat dry milk diluted in tris-buffered saline with tween 20 (50 mM Tris, 150 mM NaCl, 0.05% Tween 20) and probed using the antibodies overnight at 4°C. GAPDH was used as a loading control. Incubation with appropriate HRP-conjugated secondary antibody was followed by detection using Immobilon Western Chemiluminescent HRP Substrate (Merck Millipore, Darmstadt, Germany). The antibodies used were anti-GPADH (G9545 Sigma Aldrich), anti-DYRK1A N-ter (CPA1357 Cohesion Biosciences), anti-DYRK1A C-ter (H00001859-M01 Abnova), pan ERK1/2 (4696S Cell signaling technology), p21/CDKN1A (2947S Cell signaling technology), PTBP2 (H00058155-M01 VWR International), RNF114 (Fisher Scientific, PA5-112261), phospho ERK1/2 (9,101 Cell signaling technology), pan mTOR (2,972 Cell signaling technology), phospho mTOR (2,974 Cell signaling technology), pan PKC (P5704 clone MC5 Sigma), and phospho PKCα (sc-208 Santa Cruz).

RNAs were extracted from two series (biological replicates) of hNSC₁ treated with siRNA targeting DYRK1A, Scramble siRNA, or transfection agent alone for 48 h and sequenced as previously described (Mattioli et al., 2020). DESeq was used to detect changes in gene expression, and splicing changes were identified using LeafCutter. Genes with a change in mRNA level with an adjusted p-value of <0.1 were considered differentially expressed genes (DEG). Gene ontology enrichment was performed using DAVID tools on all the DEG using all the genes expressed in hNSCs as background (number of reads normalized and divided by the median of transcripts length in kb >50). A list of E2F and TGFB1 targets was retrieved from the MSigDB Molecular Signatures Database.² Expression changes were confirmed for the most significant DEG in a third series of hNSC₁ and in three series (biological replicates) of hNSC₂: After RNA isolation, reverse transcription reactions (Invitrogen SuperScript IV) were performed using 200–500 ng RNA, followed by qPCR on cDNA on LightCycler 480 II (Roche) using the QuantiTect SYBR Green PCR Master Mix (Qiagen) or RT-qPCR multiplex using 48*48 array Fluidigm Biomark technology (GenomEast platform). A list of primers is available upon request.

Cells were fixed 4 days after siRNA transfections with 0.5% glutaraldehyde solution for 15 min at room temperature and washed twice with PBS/MgCl₂ pH 5.95. Fixed cells were incubated with freshly prepared X-gal staining solution (1 mg/mL X-gal), 5 mM potassium ferrocyanide, and 5 mM potassium ferricyanide in PBS/MgCl₂ pH 6 at 37°C for 20 to 48 h. Cells were washed three times with H₂O and pictured in a bright field (20X magnification). Counting was done manually using the Cell Counter plugin in FIJI software, and 1,400 to 3,200 cells were counted for each condition. As a positive control of SA-β galactosidase labeling, a batch of untreated cells was irradiated at 8Gy using the “CellRad System Precision,” then fixed and labeled as previously described.

DYRK1A interactome has been studied in different cell types, including HeLa, HEK293T, human glioblastoma T98G, and human neuroblastoma SH-SY-5Y cell lines (Varjosalo et al., 2013; Li et al., 2015; Menon et al., 2019; Roewenstrunk et al., 2019; Guard et al., 2019; Huttlin et al., 2021), which led to a list of 552 interactors (BioGRID). To identify protein partners of DYRK1A in human neural stem cells (hNSC₁), we performed immunoprecipitations on endogenous DYRK1A using antibodies recognizing either the C-terminal (Cter) or the N-terminal (Nter) parts of the protein. Mass spectrometry analysis of immunoprecipitated proteins (IP-MS) revealed, in addition to DYRK1A, 35 protein interactors with 7 detected with both antibodies. Among them, 15 had previously been identified as DYRK1A partners in other cell types in the literature, and 20 were novel interactors (Table 1; Figure 1A). Phosphorylation analysis on IP-MS data identified phosphorylation events on DYRK1A (Tyr321) and in three partners: FAM117B (Ser106), GLCCI1 (Ser30 and Ser303), and FAM53C (Ser324). The partners identified were significantly enriched in proteins involved in the regulation of cell cycle and protein ubiquitination (Figure 1B), such as for instance members of the anaphase-promoting complex (APC) but also RNF114, alias ZNF313, a zinc-finger E3 ligase involved in cell cycle progression and senescence repression known to ubiquitinate p21/CDKN1A. Among the five members of the APC identified as DYRK1A’s partners in hNSC, three (CDC27/ANAPC3, ANAPC4, and CDC23/ANAPC8) had already been identified in previous studies (Lepagnol-Bestel et al., 2009; Li et al., 2018) but two (CDC16/ANAPC6 and ANAPC7) had not previously been described as DYRK1A interactors (Figure 1A; Supplementary Figure S1). Given that all these proteins are components of the APC complex, immunoprecipitation likely captures the entire complex rather than reflecting direct interactions between DYRK1A and each subunit. Apart from APC, additional clusters of interactions of proteins involved in DNA repair, mRNA regulation at the transcriptional and posttranscriptional level, and intracellular signaling pathways were also identified (Supplementary Figure S1).

We compared the expression of DYRK1A and its partners during brain development across the different brain structures (data from BrainSpan) and found a moderate-to-high positive correlation (Pearson’s coefficient) for one-third of them (11/35; Table 1). Half of DYRK1A’s partners identified in hNSCs (16/35) are known to be involved in human diseases, with neurodevelopmental manifestations for nine of them (PRKAR1A, CYFIP2, CNOT1, DHCR7, SMC3, CDC42BPB, ANAPC7, AGO1, and ADAR; Table 1). Interestingly, of the 73 clinical signs listed in the Human Phenotype Ontology (HPO) as part of DYRK1A syndrome, 80% are also described in diseases caused by mutations in partners of DYRK1A (Supplementary Figure S2). The diseases caused by pathogenic variants in DHCR7, SMC3, and CDC42BPB show the strongest overlap as they share more than 25% of the clinical signs of DYRK1A syndrome (Figure 1C). Among DYRK1A’s newly described partners that are not yet linked to human pathology, some exhibit high intolerance to loss-of-function (LoF) variants (high pLI in the gnomAD database) and present an expression profile during brain development that strongly correlates with DYRK1A. This suggests that these partners could be relevant candidates for NDDs and may contribute to the pathophysiology of DYRK1A syndrome. Notable examples include the well-known partner DCAF7 (pLI = 0.93, Pearson’s correlation = 0.81) as well as GSPT1, a GTP-binding protein involved in translation termination and the regulation of cell cycle progression (1; 0.85).

DYRK1A has previously been described as both a direct and indirect transcriptional regulator. Indeed, DYRK1A has been shown to modulate chromatin remodeling complexes and epigenetic marks (Lepagnol-Bestel et al., 2009; Li et al., 2018) and also interacts with various transcription factors such as NFAT (Gwack et al., 2006) and GLI1 (Ehe et al., 2017) and with RNA Pol II (Yu et al., 2019). To identify the effect of DYRK1A loss on the regulation of gene expression in hNSCs, we used a pool of siRNAs to knock down (KD) DYRK1A expression in hNSC₁ (Supplementary Figure S3) and performed mRNA sequencing. This transcriptomic analysis revealed 91 significantly deregulated protein-coding genes (DEG) following DYRK1A-KD compared to the control condition, with 85 of these genes being downregulated; DYRK1A was the second most significantly downregulated gene (log2 fold change = −0.87, adjusted p-value =1.10E-05; Figure 2A; Supplementary Tables S2, S3). In comparison, no DEGs were found in hNSC₁ treated with non-specific siRNA (Scramble). Uniprot and GO terms analyses revealed an enrichment of genes related to extracellular matrix (ECM) components and the plasma membrane (e.g., COL6A1, COL6A3, LUM, and THBS2), as well as calcium binding (e.g., CALB1, SGC2, and UNC13A) among the DEGs (Figure 2B). Only six genes (EGR1, EGR3, E2F2, HMRB2 and TYMS) were found to be significantly upregulated. Among them, two were members of the early growth factor (EGR) transcription factors families, EGR1 and EGR3. We then checked the expression of the most significant DEGs using multiplex RT-qPCR in a new series of the same hNSC line (hNSC₁) and three series of an additional hNSC line from another genetic background (hNSC₂) treated by DYRK1A siRNA. We confirmed significant changes in gene expression in both cell lines for 23 genes (Figure 2C). No significant motif enrichment was found in the promoter region (−1,000/+100 bp) of the DEGs using the Analysis of Motif Enrichment, version 5.1.1 (AME) tool. Among the DEGs, strong evidence of involvement in neurodevelopmental disorders was reported for 15 of them (“definitive” gene list from SysNDD, e.g., PLP1, RGS6, and SCN3A) and moderate evidence for 10 more (“limited” gene list from SysNDD, e.g., UNC13A, NMNAT2, and CHL1). Some of the DEG are not yet associated with human disease but are highly intolerant to LoF. Among them, PTBP2 has a high pLI and exhibits a strong expression correlation with DYRK1A during brain development (Pearson’s correlation = 0.8), suggesting that it could be a novel candidate gene for NDDs. Interestingly, if an increase of PTPB2 mRNA was observed after DYRK1A KD in both hNSC lines, a significant decrease in protein level was observed (Supplementary Figure S4). The same observation was made for the gene CDKN1A, which encodes the p21 protein, as a decrease in protein level was detected after DYRK1A KD, while an increase of mRNA was detected using Leafcutter, a program designed to detect changes in exon/intron ratios (Supplementary Table S5; Figure 3A; Supplementary Figure S5), affecting a weakly expressed non-canonical isoform with unknown function (NM_001291549.1). An overexpression of E2F2 mRNA was observed, which prompted us to analyze the expression of all known target genes of the E2F signaling pathway. Among the 200 genes belonging to this molecular signature, 91% were overexpressed in the DYRK1A-KD condition (Supplementary Figure S6), suggesting incomplete repression of E2F-mediated gene transcription, similar to what has been reported in human lymphocytes (Thompson et al., 2015). In contrast, we checked the expression of TGFB target genes as both TGFB2 and TGFBI mRNA were found to significantly decrease after DYRK1A-KD and observed that they are mainly downregulated (42/54 with a log2FC <0), highlighting repression of the TGFβ pathway after DYRK1A loss.

The interaction of DYRK1A with APC members and other proteins involved in cell cycle regulation, as well as the changes in expression of p21, E2F, and others identified after DYRK1A-KD suggest a role of DYRK1A in the proliferation of hNSCs, consistent with what was observed in other cell types (Park et al., 2010). Therefore, we analyzed the ability of hNSCs to proliferate after DYRK1A-KD and observed a significant decrease in cell proliferation 72 h following DYRK1A-KD in hNSCs, compared to a control condition or Scramble siRNA (Figure 3B; Supplementary Figure S7A). Because of the implication of p21/CDKN1A in cellular senescence, we tested whether DYRK1A-KD affects the number of cells undergoing senescence using the SA-β galactosidase assay 4 days post-transfection. Counting revealed that few cells turned positive after SA-β galactosidase labeling and that neither siScramble nor siDYRK1A treatments appeared to increase the number of senescent cells (Supplementary Figure S8). As one of the novel DYRK1A partners we identified, RNF114 is a ubiquitin ligase that regulates cell cycle progression by degrading p21 (Han et al., 2013). We tested the consequences of its KD in hNSCs (Supplementary Figure S9A). However, if a decrease in p21 protein levels was observed after RNF114-KD (Supplementary Figure S9B), no obvious effect on proliferation was observed (Supplementary Figure S8C), suggesting that the decrease in hNSC proliferation caused by DYRK1A-KD is not mediated by its interaction with RNF114. Recently, a study conducted in a conditional Dyrk1a heterozygous mouse model identified several alterations in growth signaling cascades, such as ERK and mTOR (Levy et al., 2021). Interestingly, a significant decrease in ERK phosphorylation was observed in hNSCs 48 h after treatment with siDYRK1A (Figure 3A).