Pathogenic, Total Loss-of-Function DYRK1B Variants Cause Monogenic Obesity Associated With Type 2 Diabetes

Obesity is a complex multifactorial disorder characterized by a strong genetic basis influenced by environmental factors. This disorder encompasses monogenic forms triggered by a solitary rare mutation exerting a strong effect capable of initiating the disease. These forms, which account for <5% of the population with obesity, occur early in life and are generally severe (1). However, monogenic forms of obesity may also present with distinct phenotypic features, giving rise to what is commonly referred to as a syndrome. One such example is abdominal obesity associated with metabolic syndrome (AOMS), encompassing central obesity, type 2 diabetes, early-onset coronary artery disease, and hypertension (2). A study involving three families highlighted that pathogenic, monoallelic variants located within DYRK1B encoding the dual-specificity tyrosine phosphorylation-regulated kinase 1B caused the development of AOMS 3 (3).

DYRK1B belongs to the protein kinase family, which plays pivotal roles in regulating various cellular processes, including cell cycle progression, cell differentiation, cell survival, motility, embryonic development, and gene transcription through phosphorylation on serine and threonine residues of target proteins (4,5). For instance, DYRK1B directly phosphorylates cell regulators such as cyclin D1 (CCND1) and cyclin-dependent kinase inhibitor 1B (CDKN1B), triggering their degradation and stability, leading to growth arrest and the promotion of quiescence (5). Additionally, DYRK1B is involved in a complex cross talk with the hedgehog (Shh) and Wnt signaling pathways, exerting dual roles in modulating these pathways, either positively or negatively. This modulation depends on the cell type and the specific involvement of canonical or noncanonical signaling mechanisms (4,5). DYRK1B exhibits ubiquitous expression, and its function has primarily been explored in the differentiation of skeletal muscle and adipocyte, spermatogenesis, and in cancers (4,5).

The initial study highlighting the role of DYRK1B in AOMS described two rare, missense DYRK1B variants (encoding p.R012C and p.H90P [NM_004714.3]) in three Iranian families and five unrelated European individuals (3). These mutations induced higher expression of glucose-6-phosphatase enzyme (G6PC1), a key hepatic gluconeogenic enzyme. The variant encoding p.R102C had a more potent inhibitor effect on the Wnt pathway and a stronger stimulating effect on adipogenic transcription factors in mouse preadipocytes, consequently promoting adipogenesis (3). However, in contrast to these seemingly gain-of-function effects, in vitro assays from another study actually demonstrated a reduction of the kinase activity of the variant encoding p.R102C, suggesting a loss-of-function effect instead (6). Two rare, heterozygous DYRK1B variants (encoding p.K68Q and p.R252H [NM_004714.3]) that were predicted to be deleterious according to in silico programs were then found in two families with AOMS (7). More recently, a rare, heterozygous null DYRK1B variant (c.520 + 1G>A [NM_004714.3]), inherently linked to the loss-of-protein function, has been identified in a father and his two daughters who presented with AOMS and varying degrees of intellectual disability (8).

Nevertheless, the comprehensive impact of DYRK1B variants on metabolic traits has yet to be extensively explored. In particular, the mechanisms through which DYRK1B variants affect metabolic homeostasis (whether involving gain- or loss-of-function) remain elusive. Here, we performed a large-scale functional genomic study focusing on rare variants of DYRK1B with the aim of unraveling their effects on metabolic traits, especially in relation to obesity and type 2 diabetes risk in humans.

In the context of a case-control analysis for type 2 diabetes and obesity, we investigated 9,353 blood samples from the RaDiO (Rare variants involved in Diabetes and Obesity) study including the D.E.S.I.R. (Data from Epidemiological Study on Insulin Resistance Syndrome) 9-year prospective study, the Department of Diabetes of the Corbeil-Essonnes Hospital (Corbeil-Essonnes, France), the CNRS UMR8199 unit (Lille, France), the Department of Nutrition of Hôtel-Dieu Hospital (Paris, France), the Centre d’Etude du Polymorphisme Humain (CEPH, Paris, France), and the French Fleurbaix-Laventie Ville Santé study, as previously described (9–11). For the obesity case-control study, case subjects included adults or children with obesity, and control subjects consisted of adults or children with normal weight (Supplementary Table 1). In individuals >18 years, normal weight was defined as BMI <25 kg/m², overweight as 25 ≤ BMI <30 kg/m², and obesity as BMI ≥30 kg/m². For children and adolescents <18 years old, normal weight was defined as BMI-for-age <85th percentile and obesity as BMI-for-age ≥95th percentile. Regarding the case-control study for type 2 diabetes, case subjects included adults with type 2 diabetes, while control subjects had normal glucose levels (<6.1 mmol/L) at age >30 years without diabetes treatment (Supplementary Table 1). Participants with type 2 diabetes had fasting glucose ≥7.0 mmol/L and/or were undergoing diabetes treatment (e.g., insulin, sulfonylureas, metformin) and were negative for islet and insulin autoantibodies. The study protocols received approval from local ethics committees, and all participants >18 years provided written informed consent. For children and adolescents, oral assent was obtained, and their parents (or legal guardians) signed an informed consent form. Study enrollment was voluntary for all participants.

DNA sequencing of DYRK1B (NM_004714.3) was done by next-generation sequencing, as previously described (9,10). All rare coding DYRK1B variants (with minor allele frequency <1%) were covered with >40 reads. Most of the DYRK1B variants had a variant quality (QUAL) score >150, and those with a QUAL score <150 were confirmed by Sanger sequencing (3730xl DNA Analyzer, Applied Biosystems). Furthermore, no variants had >5% missing genotypes across the participants, and no participants had >5% of missing genotypes across DYRK1B. To assess the pathogenicity of the rare variants, we applied the standards and guidelines of the American College of Medical Genetics and Genomics (ACMG) (12). In particular, we used our in vitro functional results to address the strong pathogenic criterion PS3. Furthermore, to address the moderate pathogenic criterion PM2, the variant had a minor allele count <5 in the Genome Aggregation Database (GnomAD) browser (version 2.1.1). To address the moderate criterion PM4, the variant was an in-frame deletion or insertion in a nonrepeat region. To address the moderate criterion PM5, the variant was at an amino acid residue where a different missense change was previously determined to be pathogenic in the literature or ClinVar. To address the supporting criterion PP3, the variant was predicted deleterious according to REVEL (rare exome variant ensemble learner) (score > 0.5) (13). To address the supporting criterion PP5, the variant was reported as pathogenic by the literature or ClinVar on one occasion.

DYRK1B sequencing for cosegregation analysis was performed by Sanger sequencing in two available families. PCR conditions and primer sequences are available upon request.

We identified 65 rare DYRK1B variants (i.e., 1 nonsense and 64 missense variants) in the RaDiO study and 1 negative control variant (harboring both c.812A>T and c.818A>T mutations encoding p.Y271F and p.Y273F, respectively [noted p.Y271/273F afterward] with inactivated kinase domain) (14) were created by site-directed mutagenesis. They were inserted into plasmids from the wild-type (WT) DYRK1B plasmid (Origene) using the Quik Change II XL Site Directed Mutagenesis Kit (Agilent). Sequence of each plasmid was checked by Sanger sequencing.

Human embryonic kidney 293 (HEK293) cells were maintained in standard culture conditions, as described in the Supplementary Material.

HEK293 cells were transfected using FuGENE HD (Promega) with 200 ng/mL of plasmid containing the β-galactosidase gene, 350 ng/mL of TOPflash plasmid (i.e., T-cell factor [TCF] reporter plasmid; Sigma-Aldrich), and 450 ng/mL of DYRK1B plasmid (WT or with a variant) and then seeded in a polylysine–coated 96-well plate at a concentration of 0.5 × 10⁶ cells/mL. Two days after transfection, the cells were treated with increasing concentrations (0, 10, 30, and 100 ng/mL) of WNT3A ligand (R&D Systems) for 6 h. After treatment, cells were lysed, and the β-galactosidase and luciferase activity were measured, as previously described (9). Each experiment was performed in technical triplicate and was repeated four times. To validate the results, three different controls were used in each experiment: a positive control with the WT DYRK1B plasmid, a negative control with the plasmid including p.Y271/273F, and a negative control where DYRK1B was not transfected. Luciferase measures were normalized using β-galactosidase measures. Fold change was calculated by dividing the normalized luciferase activity by the mean of the baseline luciferase activity of WT under the control condition (0 ng/mL). The effect of each variant (vs. WT condition) was analyzed using a linear regression model adjusted for WNT3A concentrations and plate effects. Owing to the observed deviation from normality in the residuals of certain linear regression models, we ensured the validity of the statistical outcomes using the robust Scheirer-Ray-Hare test, a nonparametric ANOVA, adjusted for WNT3A concentration and the interaction between the variant and WNT3A concentration. These analyses were performed using R 4.2.1 software. P/LP variants with a β < −0.6, along with a significant P value following both the regression linear model and ANOVA, or null P/LP variants, were categorized as fully inhibitory P/LP variants (noted P/LP-null afterward).

HEK293 cells were transfected with CCND1 plasmid (pcDNA3.1-HA-CCND1; Addgene, 172649) (15) and DYRK1B plasmid (WT or harboring a P/LP-null variant). The protein expression of DYRK1B, CCND1, and phosphorylated (phospho-)CCND1 was assessed by Western blotting in transfected HEK293 cells. Further details are provided in the Supplementary Material. Three different controls were used in each experiment: a positive control containing the WT DYRK1B plasmid, a negative control containing the p.Y271/273F mutation plasmid, and another negative control with an empty vector without the DYRK1B gene (Origene). Four independent experiments for each mutation were performed. The protein expression levels were quantified using ImageJ software. The relative expression of phospho-CCND1 was calculated by dividing the intensity of phospho-CCND1 by the intensity of CCND1, followed by normalization with β-actin. Fold change was obtained by dividing the normalized protein expression by the mean of the protein expression of the WT condition. The effect of each variant (vs. WT condition) was analyzed using a Mann-Whitney test.

Participants’ ancestry was assessed using the first five genotypic principal components (from PC1 to PC5), as previously described (10). The significant P-value threshold was set at 0.05. These statistical analyses were performed using a custom code (available following this link: https://github.com/umr1283/DYRK1B/tree/main) and R 4.0.2 software.

The meta-analysis between the current study and both 52 K and Trans-Omics for Precision Medicine (TOPMed) cohorts available in the Type 2 Diabetes Knowledge Portal (17) (accessed in August 2023) was performed using the generic inverse variance method from the R package meta (18). Only P/LP-null DYRK1B variants were included in this meta-analysis.

All coding exons of DYRK1B were sequenced in 9,353 participants (including 7,268 adults and 2,085 children) from the RaDiO study. We detected 65 rare, heterozygous variants in DYRK1B, including 1 nonsense variant (encoding p.R358*) and 64 missense variants (Supplementary Table 2). Before functional investigations, the burden of all rare DYRK1B variants was not significantly associated with adiposity, type 2 diabetes risk, or fasting glucose (Table 1).

We used the ACMG guidelines to assess the pathogenicity of each detected variant (Supplementary Table 2). To determine the strong pathogenic ACMG criterion PS3, we assessed the functional effect of each DYRK1B variant on Wnt signaling by TCF/LEF luciferase reporter assays, in response to increasing concentrations of Wnt ligand (WNT3A) within HEK293 cells (Fig. 1 and Supplementary Fig. 1). Compared with the WT condition, we found that 31 DYRK1B variants significantly decreased Wnt signaling. Using the ACMG criteria (including PS3 criterion), we identified 20 P/LP variants of DYRK1B (Fig. 1). Among these P/LP variants, six variants (encoding p.G120V, p.H179L, p.D259N, p.P282L, p.R349 W, and p.R358*) showed a fully inhibitory impact (P/LP-null) on Wnt signaling. These effects paralleled the magnitude observed in the negative control (i.e., DYRK1B carrying the p.Y271/273F variant causing an inactivation of its kinase domain) (Fig. 1). Kinase activity of all P/LP-null DYRK1B variants on CCND1 was then assessed by Western blotting. Compared with the WT condition, five of six P/LP-null variants (encoding p.G120V, p.H179L, p.D259N, p.P282 L, and p.R358*) showed a significant decrease in the expression of phospho-CCND1, suggesting reduced kinase activity of DYRK1B on CCND1 (Fig. 2, and Supplementary Fig. 2). Notably, all of these P/LP-null variants were situated within the profoundly conserved kinase domain of DYRK1B (Fig. 3).

We assessed the effect of P/LP and P/LP-null DYRK1B variants on adiposity and type 2 diabetes using the MiST method adjusted for sex, age, and ancestry. Among adults from the RaDiO study, we found that the burden of P/LP variants was significantly associated with higher BMI (31 ± 9.1 kg/m² in carriers vs. 27 ± 6.5 kg/m² in noncarriers; P_π = 0.0084 with an effect [π] of 4.2 ± 1.6) (Table 1) and was significantly associated with an increased risk of obesity (9 of 14 [64%] individuals with obesity among carriers vs. 35% of individuals with obesity among noncarriers; P_π = 0.016 with an odds ratio [OR] of 3.3, 95% CI 1.2–10) (Table 1). Importantly, the burden of P/LP-null variants showed a stronger effect on BMI than P/LP variants (36 ± 15 kg/m² in carriers vs. 27 ± 6.5 kg/m² in noncarriers; P_π = 0.013 with an effect [π] of 8.0 ± 3.2), despite a slight heterogeneity of the burden (P_τ = 0.016) (Table 1). Similarly, the effect of the burden of DYRK1B P/LP-null variants was stronger on obesity risk than P/LP variants (four of five [80%] individuals with obesity among carriers vs. 35% of individuals with obesity among noncarriers; P_π = 0.028, with an OR of 7.9, 95% CI 1.2–155) (Table 1). Among the 2,085 children and adolescents from the RaDiO study, we identified only 3 child participants carrying P/LP variants, with no instances of P/LP-null variant carriers, significantly curtailing our statistical power. Nevertheless, two of the three children carrying DYRK1B P/LP mutations presented with obesity.

We found a nominal effect of P/LP-null DYRK1B variants on type 2 diabetes risk (four of six [67%] individuals with type 2 diabetes among carriers vs. 34% of individuals with type 2 diabetes among noncarriers; P_π = 0.066 with an OR of 4.8, 95% CI 0.85–37) (Table 1); but the P/LP variants did not have any effect (Table 1). We then analyzed the association between P/LP-null variants in DYRK1B (NM_004714.3 [ENST00000323039] transcript) and type 2 diabetes risk in the Type 2 Diabetes Knowledge Portal (using the genetic association interactive tool) (17). In 43,125 individuals from 52 K and in 44,083 individuals from TOPMed, we found two and seven P/LP-null variants in 52 K and TOPMed, respectively (i.e., null variants with a very low minor allele frequency in the GnomAD browser) (Supplementary Table 3). No P/LP or P/LP-null variants identified in our study were observed in TOPMed or the 52 K study. Whereas P/LP-null variants in 52 K had no effect on disease risk due to a low statistical power (P = 0.13 with an OR of 1.7; collapsing burden test), P/LP-null variants in TOPMed showed a significant association with type 2 diabetes (P = 0.030 with an OR of 4.2; collapsing burden test). Moreover, through a meta-analysis of the RaDiO, TOPMed, and 52 K studies, we found an enrichment of P/LP-null DYRK1B variants among case subjects with type 2 diabetes (P = 0.0040 with an OR of 2.9, 95% CI 1.4–6.0; random effects model). In addition, we found that P/LP-null DYRK1B variants were significantly associated with higher fasting glucose levels (9.4 ± 5.9 mmol/L in carriers vs. 6.2 ± 2.2 mmol/L in noncarriers; P_π = 0.0051 with an effect [π] of 2.9 ± 1.0) (Table 1), although the burden presented with heterogeneity (P_τ = 5.8 × 10⁻⁵) (Table 1).

Finally, we had the opportunity to analyze the cosegregation of P/LP-null variants (encoding p.H179 L and p.R358*) with obesity, type 2 diabetes, and other metabolic traits in two families, with index case subjects originating from RaDiO (Supplementary Fig. 3). The index case subject carrying the variant encoding p.R358*, who exhibited severe obesity (BMI of 39 kg/m²), transmitted the mutation to her daughter, who also had severe obesity (BMI of 38 kg/m²). Her two sons did not carry the mutation; one of them had normal weight and the other one had obesity (Supplementary Fig. 3). The second family exhibited greater limitations, with the variant encoding p.H179 L found exclusively in the index case subject (Supplementary Fig. 3).

Our main result is that P/LP, total loss-of-function DYRK1B variants cause a monogenic form of obesity (average BMI of 36 kg/m²) associated with type 2 diabetes in the large population studies we analyzed. Notably, based on our in vitro assays, we have demonstrated that P/LP, total loss-of-function DYRK1B variants have a much more pronounced impact on obesity and type 2 diabetes compared with P/LP DYRK1B mutations. Similarly, in a previous functional genetic study focusing on PCSK1 variants, we demonstrated that only total loss-of-function (or null) PCSK1 variants caused monogenic obesity, whereas missense variants with partial or neutral effect according to our in vitro assays did not contribute to obesity risk (11). These outcomes underscore the importance of functional in vitro analyses to discriminate P/LP, total loss-of-function DYRK1B variants from P/LP variants. The identification of individuals with these P/LP, total loss-of-function variants is crucial because future advancements in pharmacological interventions to modulate metabolic function of DYRK1B may enable precision medicine for these patients.

Interestingly, all identified P/LP, total loss-of-function mutations in this study were located in the catalytic domain (Fig. 3), implying a direct impact on the kinase activity. In contrast, most of the mutations previously identified in DYRK1B were situated within the DYRK1B-homology box or the N-terminal domain of DYRK1B. The effects on these mutations can be explained by structural alterations that induce conformational instability and increase of misfolding of DYRK1B, leading to the intracellular aggregation of a nonfunctional protein (7,19).

Of note, the variant encoding p.K68Q, which we have classified neutral, was identified as pathogenic in a prior study (7). However, the carrier of this variant in our study was a 45-year-old man who did not present with overweight or type 2 diabetes.

Our study has initiated an inquiry into the genuine impact of DYRK1B variants causing AOMS, specifically whether they result in gain- or loss-of-function concerning DYRK1B activity. The first study described the DYRK1B variants that were P/LP in AOMS as gain-of-function (3). Nevertheless, various investigations have demonstrated a loss-of-function effect of DYRK1B variants via diminished kinase activity and DYRK1B expression, and heightened misfolding, degradation, and aggregation of DYRK1B (6,8,19). These hypotheses suggesting P/LP loss-of-function variants, as opposed to P/LP gain-of-function variants, are in line with our findings. Indeed, we observed that our P/LP-null variants exhibited a similar effect on Wnt signaling as the negative control where the kinase domain was deactivated (p.Y271/273F), as well as the nonsense variant (p.R358*), which inherently leads to loss-of-function. Moreover, both variants encoding p.Y271/273F and p.R358* reduced strongly the expression of phospho-CCND1, suggesting a loss-of-function effect. It should be noted, however, that DYRK1B can also act as a transcription factor independently of its kinase activity (3).

The primary limitation of our study was the absence of cardiac status data for carriers from the RaDiO study. Consequently, we lack information on whether P/LP-null variants are associated with an increased risk of early coronary artery disease. Furthermore, we focused solely on few signaling pathways involving DYRK1B. Consequently, the potential effect on other target proteins in different tissues cannot be ruled out and requests further investigation.

Acknowledgments. The authors are grateful to all individuals included in the cohort studies. The authors thank Frédéric Allegaert (University of Lille) and Timothée Beke (University of Lille) for technical assistance and Michele Pagano (New York University) for providing the CCND1 plasmid.

Funding. This study was funded by the French National Research Agency (Agence Nationale de la Recherche (ANR10-LABX-46 [European Genomics Institute for Diabetes] and ANR-10-EQPX-07-01 [Lille Integrated Genomics Advanced Network for personalized medicine] to A.B. and P.F.) and France Génomique consortium (ANR-10-INBS-009), European Research Council (OpiO 101043671 to A.B.), the European Union’s Horizon Europe Research and Innovation Programme (OBELISK grant agreement 101080465 to A.B. and P.F.), and the National Center for Precision Diabetic Medicine (PreciDIAB to A.B. and P.F.), which is jointly supported by the French National Agency for Research (ANR-18-IBHU-0001), European Regional Development Fund, Hauts-de-France Regional Council, and the European Metropolis of Lille.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. L.F., M.B., P.F., and A.B. conceptualized and designed the study. L.F., V.S., H.L., and A.De. contributed to the functional analyses of each variant. L.F. and A.B. wrote the first draft of the manuscript. B.T., E.V., A.De., F.D.P., and R.B. contributed to next-generation sequencing and/or Sanger sequencing. M.B., A.Di., L.N., M.D., and A.B. contributed to statistical and/or computer analyses. B.B., G.C., S.F., M.M., P.F., and A.B. contributed to collection of cohort data. All authors provided input on the interpretation of results and revised and approved the final manuscript. P.F. and A.B. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.