A novel role for GalNAc-T2 dependent glycosylation in energy homeostasis

Objective GALNT2, encoding polypeptide N-acetylgalactosaminyltransferase 2 (GalNAc-T2), was initially discovered as a regulator of high-density lipoprotein metabolism. GalNAc-T2 is known to exert these effects through post-translational modification, i.e., O-linked glycosylation of secreted proteins with established roles in plasma lipid metabolism. It has recently become clear that loss of GALNT2 in rodents, cattle, nonhuman primates, and humans should be regarded as a novel congenital disorder of glycosylation that affects development and body weight. The role of GALNT2 in metabolic abnormalities other than plasma lipids, including insulin sensitivity and energy homeostasis, is poorly understood. Methods GWAS data from the UK Biobank was used to study variation in the GALNT2 locus beyond changes in high-density lipoprotein metabolism. Experimental data were obtained through studies in Galnt2−/− mice and wild-type littermates on both control and high-fat diet. Results First, we uncovered associations between GALNT2 gene variation, adiposity, and body mass index in humans. In mice, we identify the insulin receptor as a novel substrate of GalNAc-T2 and demonstrate that Galnt2−/− mice exhibit decreased adiposity, alterations in insulin signaling and a shift in energy substrate utilization in the inactive phase. Conclusions This study identifies a novel role for GALNT2 in energy homeostasis, and our findings suggest that the local effects of GalNAc-T2 are mediated through posttranslational modification of the insulin receptor.


INTRODUCTION
Genome-wide association studies (GWAS) identified GALNT2, encoding polypeptide N-acetylgalactosaminyltransferase 2 (GalNAc-T2), to be associated with plasma levels of high-density lipoprotein (HDL) cholesterol and triglycerides [1]. In the same study, a causal role for GalNAc-T2 in HDL metabolism was shown in mouse models [1]. The GALNT2 gene is abundantly expressed in all tissues [2] and its translation results in GalNAc-T2, which is a glycosyltransferase that initiates mucin-type O-linked glycosylation at serine or threonine residues of target proteins. Although GalNAc-T2 modifies many target proteins, only a few have been discovered. Holleboom et al. were the first to identify apolipoprotein (apo) C-III (encoded by APOC3) as a potential substrate of GalNAc-T2 mediating the effect on HDL cholesterol and postprandial triglyceride levels in human heterozygous carriers of a rare variant in GALNT2 [3]. In a subsequent study, angiopoietin-like 3 (encoded by ANGPTL3) was identified as another potential substrate of GalNAc-T2 [4]. Furthermore, Khetarpal et al. showed that GALNT2 deficiency reduces HDL cholesterol levels in humans, mice and rats, and in nonhuman primates [5]. The authors confirmed that apoC-III and ANGPTL3 are substrates of GalNAc-T2 in humans and reported phospholipid transfer protein (PLTP) as an additional new substrate of GalNAc-T2 in humans and rodents [5]. While regulation of HDL metabolism has been at the center of key GALNT2 studies, clues suggest that GalNAc-T2 plays roles beyond the regulation of plasma lipids. Recently, human GALNT2 deficiency was presented as a syndromic disorder called GALNT2-congenital disorder of glycosylation (GALNT2-CDG). The affected individuals were shown to exhibit dysmorphic facial features, global developmental delay and intellectual disability, with mouse and rat models mimicking this phenotype [6]. Similarly, a recent study in cattle shows that homozygote carriers of splice-disrupting GALNT2 variants display reduced body weight and stature [7]. To date, however, the association of GALNT2 with metabolic syndrome [8], developmental delay and reduced body weight [6,7] have not been mechanistically investigated. Kettunen et al. were the first to show an association between GALNT2 genetic variants and metabolic syndrome [8]. Furthermore, it has been shown that GALNT2 over-expression in pre-adipocytes improved insulin signaling, stimulated adipocyte maturation and led to enlarged mature adipocytes [9,10]. In line with these studies is a recently published review that describes several studies in adipocytes with a focus on the role of GALNT2 in insulin sensitivity and adipose tissue homeostasis [11]. Here, we addressed the role of GALNT2 in body weight regulation, insulin sensitivity and whole-body energy metabolism and provide new insights into the pathophysiology of murine Galnt2 deficiency.

Genetic association with continuous traits and biomarkers
Genetic associations with biological traits were extracted from the Panancestry genetic analysis of the UK Biobank by the Pan-UK Biobank team [released June 15, 2020] [12]. In short, genetic and phenotypic data from w500,000 participants in the UK Biobank (https://www. ukbiob4ank.ac.uk) were used to conduct a Genome-Wide Association Study (GWAS). Genotypes were imputed from the Haplotype Reference Consortium plus UK10K & 1000 Genomes reference panels as released by UK-Biobank in March 2018. This research was conducted using the UK Biobank Resource (project ID 31063), and the use of these data is bound by all terms of usage of the UK Biobank. We made use of data related to 152 traits (listed in Table S1). Benferroni correction, based on the number of independent tests, was used to correct for multiple testing (significant threshold p ¼ 3.29E-04 (0.05/ 152 traits)).

Generation of Galnt2-deficient mice and genotyping
Plasmids and E. coli strains for gene targeting were obtained from the Frederick National Laboratory for Cancer Research. The PL253 targeting vector replaces exon 7 of the Galnt2 gene with a neo cassette flanked by loxP sites. Embryonic stem cell culture, screening, blastocyst injection and chimeric mice production were performed by the University of Cincinnati Gene Targeting and Transgenic Mouse Models Core. Galnt2 heterozygous mice (Galnt2 þ/À ) were backcrossed into the C57BL/6NHsd inbred mouse strain (The Jackson Laboratory) for over >15 generations before further analysis. Male and female Galnt2 þ/À mice were used for crosses to obtain all three genotypes: Galnt2 þ/þ , Galnt2 þ/À and Galnt2 À/À . The following three primers were used for PCR genotyping: P1 (5 0 GGTCCTGACCTTCCTAGACAGTCACTGC 3 0 ), P2 (5 0 GCACTCTCCAAGGGCATGACAGAGC 3 0 ) and P3 (5 0 GGGGGAG-GATTGGGAAGACAATAGC 3 0 ). The wild-type allele resulted in a 1029 bp product, while the Galnt2 mutant allele resulted in a 386 bp product.

Animals
Heterozygous breeding was used to obtain Galnt2 À/À mice and wildtype littermates. The genetically modified mice and wild-type littermates used in this study were on a 99% C57BL/6J genetic background and fed a standard chow diet (RM1; SDS Diets, Woerden, The Netherlands). Mice were housed in a light-and temperature-controlled facility with a 12-h light/dark cycle at 21 C with free access to water and control diet consisting of 10% fat (D12450J, Research Diets) or high-fat diet consisting of 60% fat (D12492i, Research Diets). Animal experiments were performed with the approval of the National Ethics Committee for Animal Experiments of The Netherlands, in accordance with relevant guidelines and regulations (including laboratory and biosafety regulations).

Animal experiments
Both male and female mice between 3-and 9-week-old were grouphoused with littermates, weighed weekly and fed a standard chow diet. Since the estrous cycle in female mice is known to affect metabolic regulation, 12-week-old male mice were used in all other experiments. Body weight and food intake of individually housed male mice were measured weekly. Measurements of body composition, including fat mass and lean tissue mass, were assessed at several time points (i.e., before indirect calorimetry measurements and before sacrifice) in nonanesthetized mice using nuclear magnetic resonance (NMR) MiniSpec (MiniSpec LF90 BCA-analyzer, Bruker). Blood was collected via retroorbital bleeding under isofluorane anesthesia in citrate-EDTA tubes, and plasma was isolated by centrifugation at 1000Âg for 10 min at 4 C and stored at À80 C until further analysis. Mice were sacrificed under isofluorane anesthesia after a 4-to 6-h fasting period in the morning, after which blood was drawn by cardiac puncture followed by cervical dislocation. Fat pads, liver and quadriceps were weighed, and tissues for mRNA and protein expression analysis were snap-frozen in liquid nitrogen and stored at À80 C until further analysis.
2.5. Glucose homeostasis Glucose homeostasis was assessed by glucose and insulin tolerance tests (GTT, ITT). For glucose tolerance, mice were fasted for 6 h after the dark phase. Mice received an intraperitoneal injection of glucose solution of 2g dextrose/kg of body weight (SigmaeAldrich, #G6152). Blood glucose levels were measured using a glucometer (Accu-Check Performa, Roche) before and 15, 30, 60, 90, and 120 min post injection. To assess insulin tolerance, mice were fasted for 6 h after the dark phase. Mice received an intraperitoneal injection containing insulin at a dose of 0.75 units/kg of body weight (insulin aspart, NovoRapid). Blood glucose levels were measured before and 15, 30, 60, 90, and 120 min post injection.

Expression and purification of GalNAc-T2
The gene expressing N-terminal His-tagged human GalNAc-T2 (hGalNAc-T2, aa 75e571) was introduced into the Pichia pastoris his4D strain Bg12 (BioGrammatics) by linearizing the expression vector pKN55-6XHis-TEV-hGalNac-T2 [18] with PmeI (New England Biolabs) followed by electroporation (Biorad). The cells were grown to an OD 600 w 10 at 30 C in MGY-case media (1% yeast extract, 1% casamino acids, 1% yeast nitrogen base with ammonium sulfate, without amino acids, 1% glycerol) and cleared by centrifugation at 2000Âg for 10 min. Cell pellets were resuspended in 1/5 volume of MMY-case media (1% glycerol is replaced with 2% methanol) for induction at 20 C for 24 h. The supernatant was cleared by centrifugation at 2000Âg for 10 min and filtered, and its pH was adjusted by adding 50 mM Tris pH 7.5 and 10 mM b-mercaptoethanol. To purify His 6 -TEV-hGalNAc-T2, the supernatant was applied to a 5 mL HisTrap column (GE Healthcare) pre-equilibrated with buffer containing 250 mM NaCl, 10 mM bME, 25 mM Tris, pH 7.5, and eluted with equilibration buffer containing 500 mM imidazole over 10 CV. The peak fractions were pooled and incubated with His 6 -TEV protease overnight at 4 C while dialyzing into equilibration buffer containing 25 mM imidazole. To separate GalNAc-T2 from the cleaved His 6 -Tag and His 6 -TEV, the sample was manually loaded onto a 1 mL HisTrap column (GE Healthcare), pre-equilibrated with dialysis buffer, and the column was washed with 4 mL of dialysis buffer. The flow-through and wash containing pure hGalNAc-T2 were pooled, and 15% glycerol was added before aliquoting and snap freezing in LN 2 for storage at À80 C. 2.18. O-glycosylation site-specific analysis of INSR in Galnt2 À/À and wild type livers Site-containing O-glycopeptides were enriched using the EXoO method [19]. The O-glycopeptides (w80 mg) were fractionated to 24 fractions followed by LC-MS/MS analysis using the Orbitrap FusionÔ LumosÔ TribridÔ Mass Spectrometer with the HCD-pd-EThcD fragmentation method [20]. MSfragger-Glyco [21] and pGlyco3 [22] were used for the identification of site-specific O-glycopeptides. A total number of 1327 and 1573 glycopeptides containing 1061 and 1212 O-glycosites were identified from livers obtained from wild-type and Galnt2 À/À mice, respectively. The total peptide-to-spectrum matches (PSM) were 11,721 and 12,019 for livers derived from wild-type and Galnt2 À/À mice, so the PSM ratio between samples obtained from wild-type and Galnt2 À/À mice was 0.975, suggesting that data normalization was unnecessary. Furthermore, proteomic analysis of the paired wild-type and Galnt2 À/À samples using TMTproÔ 16plex Label Reagent revealed equivalent insulin receptor protein levels. In addition, the median number of PSM of peptides was both three in the wild type and Galnt2 À/À samples providing further support that normalization was not required.

Quantitative real-time PCR (qPCR)
The qPCR was used to determine the level of Galnts gene expression in livers from wild-type and Galnt2 À/À mice. DNase-free RNA was isolated using the PureLinkÒ RNA Mini Isolation Kit (Ambion). cDNA synthesis was performed using the iScript cDNA Synthesis Kit (Bio-Rad). PCR primers used were published previously [23] using Beacon Designer software (BioRad). The qPCR was performed on a CFX96 real time PCR thermocycler (Bio-Rad) using the SYBR-Green PCR Master Mix (Bio-Rad). The qPCR was performed in triplicate, and four independent experiments were performed. Gene expression levels were normalized to 29S rRNA and displayed as relative expression levels.

Targeted proteomics
Targeted proteomics was used to quantify proteins in plasma or mitochondrial proteins in homogenized skeletal muscles tissues via isotopically labeled peptide standards (containing 13 C-labeled lysines or arginines) derived from synthetic protein concatamers (QconCAT) (PolyQuant GmbH, Germany) as described for the lipid metabolic proteins [24] and the mitochondrial proteins [25]. For the detection and quantification of Galnt2 in liver and visceral WAT, the isotopically labeled synthetic peptide NVPYGNIQSR (containing a 13 C 15 N-labeled arginine, AQUA Basic, Thermo Scientific) was used. For the plasma protein targets, the presented quantification results in this paper were limited to the following subset of peptides/proteins: AVEPQLEDDER and GVQIPLPEGINFVR for phospholipid transfer protein (PLTP), LESLLEEK and IYAIVQQSNYILR for angiopoietin-related protein 3 (ANGPTL3) and TVQDALSSVQESDIAVVAR for apolipoprotein C-III (APOC3). Plasma adiponectin (ADIPOQ) levels were measured using the following peptides: IFYNQQNHYDGSTGK and VTVPNVPIR. These peptides were selected not only for their optimal detection properties and uniqueness of the sequence toward the intended protein targets but also to avoid inclusion of previously reported glycosylation sides as reported in

Original Article
UniProt (www.uniprot.org); the determined protein concentrations are therefore not affected by potential changes in glycosylation.

Statistical analysis
Statistical analysis was performed using the GraphPad Prism 9.0 software package (GraphPad Software, San Diego, Ca, USA). Significance was determined using an unpaired t-test when comparing two groups. Longitudinal data were analyzed using two-way ANOVA with repeated measures, and the area under the curve was calculated and plotted in a separate figure. CalR was used for statistics of indirect calorimetry experiments [13]. All values are given as means AE SEM, and a P-value of less than 0.05 was considered statistically significant. Significance was indicated as * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001.

RESULTS
3.1. GALNT2 is associated with changes in plasma lipids and body fat By using the top-associated single nucleotide polymorphism (SNP) from previous studies [1,5] (rs4846914) as a genetic tool, we tested the association of GALNT2 in a selection of 152 traits (see Supplementary Table 1) measured in the UK Biobank. Rs4846914 is a common variant, with an A allele frequency of 61% in the European (non-Finnish) population (https://gnomad.broadinstitute.org) and 59% in the UK Biobank cohort. In addition to its very strong association with HDL cholesterol (p ¼ 2.72E-245) and triglyceride (p ¼ 4.06E-142) levels ( Figure 1A), we found that rs4846914 is also significantly associated with body mass index (BMI), body fat percentage and whole-body fat mass (p-values < 2.68E-06) ( Figure 1B), indicating a potential role for GALNT2 in the regulation of body weight and wholebody energy homeostasis.
3.2. Galnt2 À/À mice present with reduced plasma lipids To gain insight into the association between GALNT2 and metabolicrelated traits, we generated a Galnt2 À/À mouse model (Fig. S1A). We observed a dramatic reduction of hepatic Galnt2 gene expression, with no effects on expression of other abundant GalNAc-transferase genes, i.e., Galnt1, Galnt4, Galnt10 and Galnt11 (Figs. S1B and C).
This indicates that there is no compensatory upregulation at the mRNA level by other members of the GalNAc-transferase family. Using targeted proteomics, we confirmed the absence of GalNAc-T2 protein levels in liver and visceral white adipose tissue (WAT) where Galnt2 is normally abundantly expressed (Figs. S1D and E) [26,27]. Galnt2 À/À mice have reduced fasting plasma cholesterol levels, on both control and high-fat diet (HFD), compared with wild-type littermates due to a reduction of cholesterol in both HDL and low-density lipoproteins (LDL) (Figs. S2AeD) as described previously [4]. We furthermore observed a trend for decreased fasting plasma triglyceride levels when Galnt2 À/À mice were fed a control diet but not when fed a HFD (Figs. S2E and F).
3.3. Galnt2 deficient mice are smaller and gain less weight on control and high-fat diets Corroborating results with another Galnt2 À/À mouse line [6], we observed a similar embryonic lethality following heterozygous breeding, while homozygous breeding was unsuccessful (data not shown). Early in lifedin fact, from the earliest point that we could measure, i.e., from three to nine weeks of ageda reduction in body weight was observed in both surviving male and female Galnt2 À/À mice compared with wild-type littermates (Fig. S3). Given the similarity of this effect in both genders, further studies were performed in male mice only. On a control diet, adult male Galnt2 À/À mice of 10e12 weeks of age remained smaller and gained less weight compared with control mice over a period of 12 weeks (Figure 2A,B). This was characterized by a significant reduction in both lean and fat mass ( Figure 2C). When adult male Galnt2 À/À mice of 10e12 weeks of age were challenged with a HFD for 12 weeks, we observed a consistently reduced body weight and body weight gain compared with controls ( Figure 2D,E). This was again characterized by reductions in lean mass and fat mass ( Figure 2F) that were more pronounced compared with Galnt2 À/À mice fed the control diet ( Figure 2F versus Figure 2C). In addition to decreased body weight gain, Galnt2 À/À mice also displayed reduced body, tibia and snout length (Figure 2GeJ). The reduced body weight and stature phenotype have been previously observed in Galnt2 deficient mice, as well as in humans and dairy cattle [6,7]. In line with our genetic observations in humans, our mouse data show that Galnt2 plays a role in whole-body energy homeostasis: Galnt2 À/À mice remain smaller and gain less weight on both control and HFD compared with wild-type littermates, and this can be explained by reduction in lean and fat mass.

Galnt2 deficiency is associated with reduced visceral WAT mass
To delineate the causes of reduced body weight gain in Galnt2 À/À mice and found a trend toward reduced liver weight (p ¼ 0.08) but no differences in liver-to-body weight ratio when compared with wild-type littermates ( Figure 3A,B). Also, no differences were observed in hepatic levels of triglycerides, free cholesterol, cholesterol esters and phospholipids between genotypes (Fig. S4). In line with reduced lean and fat mass ( Figure 2C,F), Galnt2 À/À mice showed significant decreases in quadriceps and visceral white adipose tissue (WAT) weight. However, only the change in visceral WAT weight remained significant after normalization to body weight (Figure 3CeF). No differences between genotypes were observed in both the interscapular brown adipose tissue (BAT) and subcutaneous WAT weight (Figure 3GeJ). The reduction in visceral WAT in Galnt2 À/À mice was associated with an almost two-fold decrease in visceral adipocyte size compared with wild-type littermates ( Figure 3K,L). Akt/mTORC1 signaling is important for normal adipose tissue growth and adipogenesis [28], so we tested whether the Akt/mTORC1 signaling pathway was affected in visceral WAT of Galnt2 À/À mice. Upon immunoblotting of Akt and mTORC1 substrates, we observed a decrease in phosphorylated glycogen synthase kinase 3 beta (Gsk3b), ribosomal protein S6 kinase (p70 S6 kinase), ribosomal protein S6 and eukaryotic translation initiation factor 4E-binding protein 1 (4E-BP) (Figs. S5A and B). These findings suggest reduced Akt/mTORC1 signaling in absence of Galnt2 in visceral WAT. Taken together, murine Galnt2 deficiency leads to changes in visceral WAT mass, adipocyte size and Akt signaling, and this extends the in vitro findings by Marucci and Di Paola et al., who suggested that GalNAc-T2 plays a role in adipocyte insulin signaling, maturation and adipogenesis [9,10]. Overall, these findings suggest a novel role for GALNT2 in visceral WAT. Figure 2: Galnt2 À/À mice are smaller and exhibit a GALNT2-CDG like phenotype. Male Galnt2 À/À mice and wild-type littermates were followed on a control diet (CD) whereby (A) body weight, (B) body weight gain and (C) body composition was monitored (n ¼ 8e9). Male Galnt2 À/À mice and wild-type littermates at 12 weeks of age were followed on a high-fat diet (HFD) whereby (D) body weight, (E) body weight gain and (F) body composition was monitored (n ¼ 9e10). (G) Body length (n ¼ 8e10) and (H) tibia length (n ¼ 8e 10) along with photographic representations of (I) the body and (J) head of a Galnt2 À/À mouse and wild-type littermate (Galnt2 þ/þ ). The experiments on control diet were repeated 3 times (A-C), and on HFD 2 times (D-F), (G-J) was determined in one cohort, but the effect was also observed in all other cohorts. Data are presented as mean values AE SEM with *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Galnt2 deficiency does not alter insulin and glucose tolerance but affects the insulin receptor and insulin signaling
The combined effects on body weight gain and composition as well as visceral WAT prompted us to study potential effects of Galnt2 deficiency on glucose homeostasis. While fasting plasma glucose levels were reduced in Galnt2 À/À mice, insulin levels were not affected ( Figure 4A,B). Despite being significantly smaller, no pronounced differences in glucose or insulin tolerance were observed in Galnt2 À/À mice compared with wild-type littermates (Figure 4CeF).
Next, we assessed the ex vivo insulin signaling in visceral WAT as changes in insulin signaling have been shown to alter adipocyte lipid storage and breakdown [29]. We found a rapid and enhanced Akt phosphorylation upon 5 and 15 min of insulin treatment in visceral WAT depots of Galnt2 À/À mice ( Figure 4G,H) compared with controls. Downstream levels of phosphorylated Gsk3b, p70 S6 kinase, S6 and 4E-BP were also increased after 15 min of insulin stimulation (Fig. S5C), suggesting more sensitive activation of the Akt/mTORC1 pathway in visceral WAT of Galnt2 À/À mice compared with wild-type littermates. These findings suggest that loss of Galnt2 leads to changes in response to insulin, resulting in aberrant rapid insulin receptor activation followed by increased Akt signaling. Interestingly, when immunoblotting for the insulin receptor, we observed that the apparent molecular weight of the mature insulin receptor was reduced in visceral WAT of Galnt2 À/À mice ( Figure 4G, arrow) compared with controls, suggesting that it may be a target of GalNAc-T2. Collectively, these data indicate that Galnt2 deficiency has only mild effects on glucose and insulin tolerance. However, loss of Galnt2 results in altered insulin signaling and a change in the molecular weight of the insulin receptor, suggesting that the insulin receptor may be a direct target of GalNAc-T2.

The insulin receptor is a new target of GalNAc-T2 mediated Oglycosylation
The shift in the apparent molecular weight of the mature insulin receptor was identified in visceral WAT, liver and quadriceps derived from Galnt2 À/À mice ( Figure 5A). This pointed toward a difference in the glycosylation state of the insulin receptor between genotypes in different metabolic organs. To study this, we performed in silico analysis using the ISOGlyP (Isoform Specific O-Glycosylation Prediction, http://isoglyp.utep.edu/index.php) server. This revealed three potential O-linked glycosylation sites in the insulin receptor, at T930, T1089 and T1122. O-glycosylation assays using synthetic peptide substrates suggest that GalNAc-T2 modifies the three putative sites in vitro ( Figure 5B). Next, we used a mass spectrometry based-method to map site-specific O-GalNAcylation sites, called extraction of Olinked glycopeptides (EXoO), using livers derived from wild-type and Galnt2 À/À mice [19]. A total of six O-glycosites were detected at the insulin receptor. These included T763, T764, T766, S774, S775 and T776 that are part of an extracellular domain (amino acids 753e946) and in close proximity to the insulin-binding domain (amino acids 735e743) (Schematically represented in Figure 5C). Differential analysis revealed that T763 was detected in livers derived from wildtype mice with a total peptide-to-spectrum matches (PSM) of 43, but it was not detected in livers derived from Galnt2 À/À mice ( Figure 5D). The PSM number of other O-glycosylated sites was significantly decreased in liver samples derived from Galnt2 À/À mice ranging from 1.3 to 6-fold decrease ( Figure 5D). There was an exception for S774, which was only detected in the liver samples derived from Galnt2 À/À mice, but the low PSM number suggests a low abundance. In conclusion, the insulin receptor harbored decreased levels of Oglycosylation and an absence of O-glycosylation at T763 in livers of Galnt2 À/À mice, supporting the hypothesis that the insulin receptor is a novel target of GalNAc-T2.
3.7. Galnt2 À/À mice exhibit increased plasma NEFA levels but normal adipose tissue lipolysis To further explore the mechanism underlying reduced visceral WAT mass, we assessed whether Galnt2 deficiency affects adipocyte function. Interestingly, we found markedly increased fasting plasma non-esterified fatty acid (NEFA) levels in Galnt2 À/À mice ( Figure 6A); this may be due to increased adipocyte lipolysis. Fasting plasma adiponectin levels, an adipocyte-derived adipokine involved in glucose metabolism and fatty acid oxidation, were increased almost two-fold ( Figure 6B). In line with increased NEFA levels, we observed a more than two-fold increase in b-hydroxybutyrate, one of the main ketone bodies ( Figure 6C). Intrinsic ex vivo lipolysis of visceral WAT pads as measured by the release of glycerol ( Figure 6D,E) and free fatty acids ( Figure 6F), however, was not affected in Galnt2 À/À mice. Accordingly, no differences were observed in proteins involved in the adipose tissue lipolysis pathway, including phosphorylated and total hormonesensitive lipase, perilipin 1 and CD36 ( Figure 6G,H). In addition, we did not observe a strong impairment in insulin-mediated suppression of lipolysis, with Galnt2 À/À mice only showing a mildly delayed insulinmediated turnover of free fatty acids compared with wild-type littermates ( Figure 6I,J). As adipose tissue lipolysis is also under the control of sympathetic activity, we also tested whether there were changes in plasma levels of lipolytic hormones, including glucagon, catecholamines, corticosterone and intermediate products. We found no significant differences in these hormones apart from a slight but significant decrease in 11-deoxycorticosterone in Galnt2 À/À mice compared with controls (Fig. S6). Taken together, our studies do not support abnormal adipose tissue lipolysis in Galnt2 À/À mice but suggest altered peripheral NEFA uptake or metabolism.

3.8.
Galnt2 deficient mice rely more on lipids as a substrate for energy Galnt2 À/À mice display reduced visceral WAT mass and adipocyte size, while plasma NEFA, adiponectin and b-hydroxybutyrate levels are increased without apparent changes in intrinsic adipocyte lipolysis. To gain more insight into the mechanisms underlying this metabolic phenotype, we examined the energy metabolism of Galnt2 À/À mice using indirect calorimetry. Food intake was similar between Galnt2 À/À mice and wild-type littermates ( Figure 7A). Interestingly, we found differences in oxygen consumption, carbon dioxide production, energy expenditure and respiratory exchange ratio (RER) between genotypes (Figure 7BeE). Compared with wild-type littermates, Galnt2 À/À mice showed a marked decrease in RER in the inactive (light) phase, indicating increased preference for lipids as a fuel source. Indeed, when calculating the oxidative substrate preference, we found a profound increase in lipid oxidation ( Figure 7F) and a decrease in glucose oxidation ( Figure 7G) in Galnt2 À/À mice during the inactive phase. These differences, however, were not observed in the active (dark) phase and were not related to differences in protein levels of mitochondrial respiration markers in quadriceps, liver or visceral WAT (Fig. S7). Our studies reveal a surprising novel role for GalNAc-T2 in the use of lipids as oxidative fuel in the inactive phase.
3.9. Galnt2 and its protein substrates Finally, we focused on substrates of Galnt2 that have thus far been described in the literature, i.e., apoC-III and ANGPTL3 (both inhibitors of lipoprotein lipase (LPL), which normally mediates the lipolysis of triglyceride-rich lipoproteins) and PLTP [3e5]. To assess potential effects of Galnt2 deficiency on protein concentrations, we used a targeted proteomics approach in which all substrates were quantified simultaneously in plasma samples. This analysis did not reveal significant differences in the concentrations of apoC-III and PLTP ( Figure 8A). The earlier proposed altered glycosylation of these proteins, therefore, did not appear to have resulted in quantitative differences in plasma between genotypes in our study. We did observe increased levels of ANGPTL3 in plasma of Galnt2 À/À mice ( Figure 8A). ANGPTL3 is known to be cleaved at the proprotein convertase processing site, resulting in a separate N-terminal and C-terminal site, and it is suggested that the N-terminal site contains the LPL binding motif to inhibit LPL in the circulation [30]. Previous studies suggested that GalNAc-T2 is an important regulator in ANGPTL3 cleavage [4,31].
We observed a significant increase in both the N-and C-terminal site of ANGPTL3 in plasma and a higher concentration of the C-terminal site compared with the N-terminal site ( Figure 8B). The detected fragments can be part of the complete protein or the two separate fragments.
These results cannot rule out qualitative effects due to loss of GalNAc-T2 mediated O-linked glycosylation of these substrates or subsequent downstream effects such as lipolysis. To assess the actual physiological consequences of all above findings, we measured postprandial triglyceride levels following an oral fat tolerance test. Galnt2 À/À mice display delayed triglyceride absorption as well as delayed clearance compared to controls ( Figure 8C), suggesting alterations in the postprandial triglyceride response. Although the current study also reveals that the insulin receptor is yet another substrate for GalNAc-T2, the findings combined highlight the complexity of the metabolic phenotype of a complete loss of GalNAc-T2 in mice.

DISCUSSION
We showed that GALNT2 plays an important dynamic role in wholebody energy homeostasis. First, we revealed significant associations between GALNT2 gene variation, BMI and fat mass in humans. Our studies in Galnt2 À/À mice subsequently reveal decreased adiposity, altered insulin signaling and a shift in energy substrate utilization during the inactive phase. GalNAc-T2 has previously been shown to exert systemic effects on plasma lipids through post-translational modification, i.e., O-linked glycosylation, of proteins that are secreted by the liver [3e5]. The current study provides the first evidence that GalNAc-T2 can also have local effects through O-glycosylation of the insulin receptor. We performed genetic associations with biological traits using a common variant of GALNT2, namely rs4846914, extracted from UK Biobank GWAS data. The A allele was significantly associated with increased HDL cholesterol levels and decreased triglyceride levels and overall reduced adiposity (traits like body fat percentage, whole-body fat mass and body mass index were decreased). In contrast, carriers of rare loss of function (LOF) variants were previously shown to have both decreased plasma HDL cholesterol and triglyceride levels [5]. Furthermore, GALNT2-CDG patients not only exhibit reduced levels of plasma HDL cholesterol and triglycerides, but also dysmorphic facial features, short stature and developmental delay [6]. The latter clinical findings are recapitulated in various Galnt2-deficient animal models, and those models also show decreased adiposity. There is a discrepancy in the directionality between common SNPs in GWAS studies, rare LOF variants and knock-out animal models. Thus, common variants result in milder phenotypes as compared with the severe pathophysiological phenotypes observed in carriers of rare LOF variants.
We observed reduced body weight in three to nine-week-old male and female Galnt2 À/À mice and decreased body length and body weight gain in adult male Galnt2 À/À mice upon control or HFD feeding. This observation is in line with previous studies showing reduced body weight and decreased stature in human carriers of GALNT2 variants, male Galnt2 À/À mice and cattle carrying Galnt2 variants [6,7]. We demonstrate that the reduction in body weight is the result of a decrease in both lean and fat mass; this was exacerbated when Galnt2 À/À mice were fed a HFD. However, when adjusted for body weight, visceral WAT was the only tissue that was significantly reduced in Galnt2 À/À mice, and this abnormality coincided with reduced visceral WAT adipocyte size. In line, previous studies in pre-adipocytes also showed that GalNAc-T2 overexpression improved insulin signaling, stimulated adipocyte maturation and led to enlarged mature adipocytes [9,10]. The absence of changes in subcutaneous WAT in our study can be explained by differences in plasticity, metabolic function and immune cell composition between visceral and subcutaneous WAT [32e34]. For instance, visceral WAT is considered to exert more adverse effects on health, is strongly associated with insulin resistance and represents the primary fat storage depot during the initial phase of obesity [32,33]. One possible explanation for the observed differences is that GalNAc-T2 exerts local actions via downstream effects of its substrates in visceral but not subcutaneous WAT. For example, it has been shown that ANGPTL3, one of the proposed GalNAc-T2 substrates, plays an important role in the uptake of fat into visceral WAT, but this is not described for subcutaneous WAT [35]. Adipose tissue is an important regulator of insulin sensitivity and glucose tolerance, and smaller adipocytes are often linked to improved insulin sensitivity and glucose tolerance [34]. Standard glucose and insulin tolerance tests, however, did not reveal notable changes in these parameters. We did observe significant reductions in plasma glucose levels: this could be due to decreased gluconeogenesis, reduced hepatic glucose production or glucose transport issues. The smaller adipocytes, reduced visceral WAT and body weight of Galnt2 À/À mice resemble whole-body and adipocyte-specific S6k1 À/À mice and other models in which mTORC1 signaling is disrupted [36e 38]. In visceral WAT harvested from fasted mice on a control diet, we likewise found reduced phosphorylated protein levels of Akt and downstream targets of Akt and mTORC1 (Gsk3b, Foxo1, S6K1, S6, and 4E-BP) in visceral WAT; this lends support for the involvement of changes in Akt/mTORC1 signaling in explaining smaller adipocytes, visceral WAT and body size (Figs. S5A and B). Interestingly, ex vivo insulin stimulation of visceral WAT led to a transient increase in phosphorylated Akt in Galnt2 À/À mice compared with wild-type littermates ( Figure 4G) and was accompanied by an increase in downstream Akt and mTORC1 signaling (Fig. S5C). This suggests increased activity of the insulin receptor caused by the loss of O-glycosylation. Studying Akt and mTORC1 signaling in the liver of Galnt2 À/À mice compared with wild-type littermates only revealed increased phosphorylation of p70 S6K with no further increase of S6 activation (Figs. S8A and B). In quadriceps, we observed an overall increase in Akt/mTORC1 signaling (Figs. S8C and D), which was not further studied. However, as GalNAc-T2 is expressed only at low levels in muscle (data not shown), it remains unclear whether this is a direct or indirect effect. Taken together, our Galnt2 mouse studies reveal that GalNAc-T2 plays a new physiological role in adipocyte-specific insulin signaling in visceral WAT. Our in vitro studies show that the insulin receptor can be O-glycosylated by GalNAc-T2 at three threonine residues. Although the insulin receptor has been reported to be heavily glycosylated with both N-and O-glycans, the key enzymes responsible for mediating the glycosylation process are currently unknown. Though it is generally assumed that N-linked glycosylation is essential for insulin binding and autophosphorylation [39e41], a role for GalNAc-T enzymes in O-linked glycosylation of the insulin receptor has, to our knowledge, not been described. Using a recently developed EXoO method [19], we show that the insulin receptor lacks O-glycosylation at T763 accompanied by reduced O-glycosylation at T764, T766, S775 and T776 in Galnt2 À/À mice compared with wild-type littermates. To our knowledge, we are the first to show that the insulin receptor is a substrate of GalNAc-T2.
The respective threonine and serine residues are in close proximity to an insulin-binding domain, and alterations in O-linked glycosylation at these sites may impact the insulin receptor activation by changes in insulin binding and subsequent Akt signaling. Galnt2 À/À mice presented with increased plasma NEFA and adiponectin levels. In line, we observed increased plasma b-hydroxybutyrate levels as a measure of ketone bodies ( Figure 6C), in Galnt2 À/À compared with wild-type mice [42]. Although augmented ketone bodies are indicative for increased mitochondrial oxidation in the livers of Galnt2 À/À mice, no changes were seen in mitochondrial proteins in the liver (Fig. S7). However, we cannot rule out changes of protein levels at other time points, such as after prolonged fasting, throughout the light/dark cycle or related to changes in the activity of mitochondrial proteins, e.g., by posttranslational modifications. While these findings point at altered adipocyte triglyceride lipolysis and fatty acid oxidation [43e45], we did not observe abnormal intrinsic adipose tissue lipolysis. However, we identified profound changes in substrate utilization in indirect calorimetry studies. Galnt2 À/À mice predominantly use lipids for oxidation in the inactive phase but rapidly switch to glucose oxidation during the active phase as if they were wild-type animals. The increased use of lipids as substrates during the fasting/resting period is consistent with the decreased fasting plasma glucose levels in Galnt2 À/À mice, meaning that less glucose is available to be used as substrate. These findings point toward a problem in the production, uptake or utilization of glucose during fasting and a problem with NEFA metabolism or uptake by adipose tissue. A swift transition in the use of substrates between fasting and refeeding is usually attenuated in pathophysiological models [46]. Increased lipid oxidation may be due to the activation of NEFA mobilization from white adipose tissue and fatty acid transport into mitochondria for oxidation in energy-demanding tissues. Mice are normally fasting in the light phase, which lowers insulin levels, leading to enhanced adipose tissue triglyceride hydrolysis and delivery of NEFA [46]. This leads us to speculate that this effect could be related to impaired glucose production or to our finding that Galnt2 À/À mice display altered adipocyte-specific insulin action. Our observations suggest enhanced metabolic flexibility with regard to substrate use, which is in line with the observed elevated plasma adiponectin levels in Galnt2 À/À mice [46,47]. GalNAc transferases all catalyze the same enzymatic reaction but exert distinct effects due to differential tissue expression, enzyme activity and substrate specificity [48]. When considering the actual potential mediators of the various observations in our mouse model, i.e., the substrates of GalNAc-T2, the studies published thus far have focused on the association between GALNT2 and plasma HDL cholesterol levels. This has rendered apoC-III and Angptl3 as GalNAc-T2 known substrates [3e5], but this will only be a fraction of its total number of targets. To explain the plasma lipid phenotype, Khetarpal et al. also identified PLTP as a third substrate mediating the HDL cholesterol phenotype [5]. The post-translational modifications of ApoC-III, ANGPTL3 and PLTP by GalNac-T2 can affect protein concentrations, for example, through changing further protein processing, synthesis or catabolism. In the above studies, antibodies were used to detect proteins of interest, which showed that GALNT2 affects isoforms of apoC-III and cleavage of the N-terminal part of Angptl3 [3e5,31]. To address effects in a quantitative manner in the current study, we used targeted proteomics in which unique peptides labeled with stable isotopes are used to quantify the concentrations of peptides of endogenous proteins in one single sample preparation [24,25]. Our Galnt2 À/À mice presented increased plasma ANGPTL3 protein levels.
Several studies have reported a role for GalNac-T2 in regulating ANGPTL3 cleavage through O-linked glycosylation of a threonine amino acid adjacent to a proprotein convertase processing site, thereby blocking cleavage in vitro and in vivo [4,31]. Actual cleavage yields an N-terminal domain containing the LPL binding motif and a C-terminal domain. Mice overexpressing the N-terminal fraction of ANGPTL3 indeed exhibit an increased plasma LPL inhibitory capacity and increased plasma triglyceride levels, however, a cleavage-resistant mutant of ANGPTL3 is still able to inhibit LPL, suggesting that the unprocessed protein is also able to inhibit LPL [30]. Additionally, inhibiting or promoting ANGPTL3 cleavage, by Galnt2 overexpression or suppression, did not alter plasma triglyceride levels in mice, again suggesting that ANGPTL3 cleavage is not essential for its LPL inhibitory capacity [31]. In our mass spectrometry measurements, we used one peptide that detects the N-terminal part, upstream of a proprotein convertase cleavage site, and one peptide that detects the C-terminal part of ANGPTL3 and both were increased in Galnt2 À/À mice. With the previous studies in mind, an increase in ANGPTL3 (full protein or cleaved fragments) in Galnt2 À/À mice is expected to increase inhibition of LPL and reduce hydrolysis of triglyceride-rich lipoproteins. However, in the fasting state as well as after an oral fat load, Galnt2 À/À mice do not present with marked changes in plasma triglycerides. Apparently, the role of GalNac-T2 in regulating triglycerides via ANGPTL3 in our mice is limited. This may be due to quantitative changes in other GalNac-T2 substrates, which brings us to apoC-III and PLTP concentrations that were not found to be different between Galnt2 À/À mice and controls. Attenuated O-linked glycosylation may, however, change the biochemical and biological properties of proteins that are not reflected by changes in concentrations. For example, Holleboom et al. showed that in the post-prandial phase, hypoglycosylated apoC-III of individuals with a loss-of-function mutation in GALNT2 is not residing on triglyceride-rich lipoproteins but on HDL [3]. Clearly, it is very challenging to pursue the downstream effects of attenuated GalNAc-T2 mediated O-linked glycosylation of multiple substrates that occur simultaneously, especially in a whole body knockout setting.

CONCLUSION
In humans, cattle and mice, loss of GALNT2 results in a syndrome characterized by an intriguing metabolic phenotype with decreased body weight and stature. In Galnt2 À/À mice, we observed a metabolic phenotype that combines decreased adiposity, changes in adipocyte function and altered energy substrate utilization. Our findings show that the insulin receptor is another GalNAc-T2 substrate that may be responsible for local rather than systemic effects. The local effects are of special interest when considering the tissue-specific role of GalNAc-T2 and other members of the GalNAc-T glycosyltransferase family that catalyze the same reaction. They may or may not be capable of compensating for the loss of GalNAc-T2 activity. Tissue-specific knockout mouse models are needed to further dissect the role of GalNAc-T2 and its substrates in various metabolic and non-metabolic tissues. The current study reveals that GalNAc-T2 has a previously unknown role in whole-body energy homeostasis: this may be due to glycosylation of the insulin receptor and thereby affect insulin signaling.