Deletion of the Brain-Specific α and δ Isoforms of Adapter Protein SH2B1 Protects Mice From Obesity

Human mutations in SH2B1 are associated with severe, early-onset obesity, hyperphagia, and often disproportionately high insulin resistance (1–3). Similarly, mice null for Sh2b1 (Sh2b1 knockout [KO] mice) exhibit hyperphagic obesity and impaired glucose homeostasis (4,5). Reintroduction of SH2B1β into neurons in Sh2b1 KO mice largely restores normal body weight and glucose homeostasis (6), suggesting the importance of neuronal SH2B1 for metabolic control. SH2B1 is an adapter protein that regulates responses to multiple hormones and neurotrophic factors that regulate the nervous system. For example, SH2B1 binds to and modulates the activity of the tyrosine kinase JAK2, which forms a complex with several cytokine family receptors including the leptin receptor (LepRb). SH2B1 also binds to and modulates actions of the insulin receptor; TrkA and TrkB, receptors for nerve growth factor (NGF) and brain-derived neurotrophic factor (BDNF), respectively; and RET, a coreceptor for glial cell line–derived neurotrophic factor (GDNF) and growth/differentiation factor-15 (GDF-15) (7,8). When activated, these kinases recruit SH2B1 via its SH2 domain. SH2B1 then facilitates a variety of cellular responses, including changes in the actin cytoskeleton and gene expression (9–11). Consistent with SH2B1’s involvement in neurotrophic factor signaling and energy balance and the importance of neurons for appetite control, SH2B1 promotes outgrowth of dendrites and/or axons of cultured primary neurons (12–14) and neurite outgrowth and neuron-specific gene expression in PC12 cells (9,15,16). Hence, SH2B1 influences both structural and functional aspects of the nervous system.

Previous work suggests that SH2B1 contributes to energy balance by modulating LepRb/JAK2 activity (7). Leptin, secreted primarily from white adipose tissue (WAT) in approximate proportion to triacylglycerol content, signals the repletion of energy stores to the brain (17). Normal leptin concentrations indicate an adequate fat supply and thus suppress hunger and allow for energy use; conversely, low leptin levels signal low fat stores and thus increase hunger and conserve existing energy stores (17). Consistent with SH2B1 enhancing leptin signaling, Sh2b1 KO mice exhibit reduced leptin-induced inhibition of food intake and weight gain and dysregulated control of hypothalamic leptin-sensitive gene expression (4,6).

Alternative splicing yields four previously described SH2B1 isoforms—α, β, γ, and δ—which differ only in their COOH-terminal tails (Fig. 1A and B). In humans, SH2B1β and SH2B1γ are expressed ubiquitously, whereas SH2B1α and SH2B1δ are restricted almost exclusively to the brain (1). In vitro studies indicate that some SH2B1 isoforms possess unique cellular properties. For example, while SH2B1β and SH2B1γ promote neurite outgrowth in PC12 cells (9,15,16), SH2B1α does not, and even inhibits the ability of SH2B1β to do so (2,18). However, before this work, functions of the various SH2B1 isoforms had not been assessed in vivo. Because both SH2B1α and SH2B1δ exhibit the unique characteristic of being expressed almost exclusively in the brain, the central regulator of metabolism, we investigated the combined contributions of these two brain-specific SH2B1 isoforms to energy balance and glucose homeostasis. Our results suggest not only that SH2B1α and/or SH2B1δ are critical regulators of body weight but also that the different alternatively spliced SH2B1 isoforms perform different functions in vivo.

Mice were housed in ventilated cages (∼22°C) on a 12-h light/dark cycle (∼6:00 am–6:00 pm) in a pathogen-free animal facility at the University of Michigan. Food and water were available ad libitum except as noted. Experimental mice were fed a standard chow diet (9% fat, cat. no. 5058; LabDiet) or a high-fat diet (HFD) (60% fat, D1249; Research Diets). Experiments were approved by the University of Michigan Institutional Animal Care & Use Committee.

CRISPR/Cas9 editing was used to delete the regions in Sh2b1 required for expression of SH2B1α and SH2B1δ (Sh2b1^DELαδ). RNA guides were selected as described previously (19). A 4-kb donor template was designed to juxtapose exon 9 of Sh2b1, the sequence in exon 10 that contains the β/γ stop codons, and the region of exon 10 downstream of the α stop codon (Fig. 1B). A guide/donor package was injected into C57BL/6J × SJL F2 oocytes, which were implanted into C57BL/6J × SJL F2 mice (University of Michigan Transgenic Animal Model Core). Founders were backcrossed against C57BL/6J mice. See details in Supplementary Material.

Genomic DNA was isolated from tails prior to weaning, and diagnostic fragments were amplified by PCR with Taq polymerase and two primer sets (Supplementary Table 1). Heterozygous Sh2b1^DELαδ/+ (αδHET) mice were backcrossed to C57BL/6J mice for three to five generations. Nonsibling αδHET mice were bred to produce experimental mice. Experimenters were blind to genotypes. Mice were regenotyped after harvest.

Body weight was assessed weekly. Food intake was monitored once or twice weekly (details in Supplementary Material). Body composition was measured using nuclear magnetic resonance (NMR) (Minispec LF90II; Bruker Scientific). Oxygen consumption (VO₂), carbon dioxide production (VCO₂), and spontaneous locomotor activity were monitored by the Comprehensive Lab Animal Monitoring System (CLAMS) (Columbus Instruments) (Michigan Mouse Metabolic Phenotyping Center) as described previously (3).

Blood was collected between 9:00 am and 11:00 am (glucose, insulin) or 10:00 am and 2:00 pm (leptin). Glucose levels were measured via tail vein bleeding and Bayer Contour glucometer. Insulin and leptin levels in tail vein blood and terminal trunk blood were measured using Crystal Chem Mouse Insulin ELISA (cat. no. 90080) and Crystal Chem Mouse Leptin ELISA (90030) kits, respectively.

Mice were subjected to a 5- to 6-h morning fast followed by intraperitoneal (i.p.) injection of d-glucose or human insulin. Blood glucose levels were measured as described above.

Mice were i.p. injected twice daily (6:00 am, 6:00 pm) with vehicle (PBS, pH 7.4) (days 1–3, 7–8) or recombinant mouse leptin (days 4–6).

Mice were anesthetized by isoflurane between 10:00 am and 2:00 pm and decapitated, and terminal blood was collected. Tissues were dissected, weighed, and fixed in 10% neutral buffered formalin and stored (4°C) for histology or cryopreserved in liquid nitrogen and stored (−80°C) for RNA or protein extraction. Hypothalami (3-mm cubes) were dissected from the ventral diencephalon immediately caudal to the optic chiasm with a coronal brain matrix. Other brain sections were dissected under a dissecting microscope. Fixed liver tissue was paraffin embedded, sectioned (5 μm thickness), stained with hematoxylin-eosin, and imaged, as previously described (20).

RNA was isolated from frozen tissue with QIAGEN RNeasy Mini Kits (cat. nos. 74104 and 74804). RNA was reverse transcribed into cDNA using TaqMan Reverse Transcription Reagents (Fig. 1E) or iScript cDNA Synthesis Kit (Figs. 1C and 7A and Supplementary Fig. 4A). Taq polymerase was then used for PCR reactions in Fig. 1E (primers listed in Supplementary Table 2). Relative levels of mRNA transcripts encoding SH2B1 isoforms and leptin-regulated genes were determined with TaqMan Gene Expression Assays (details in Supplementary Table 3 and Supplementary Material).

RNA samples had integrity numbers ≥7.5. cDNA library preparation and sequencing were performed by the University of Michigan DNA Sequencing Core. See Supplementary Material for details of preparation and analyses.

cDNAs encoding mouse GFP-SH2B1α (GenBank accession no. AF421138) (18) and rat GFP-SH2B1β (accession no. NM_001048180) (15) have previously been described. cDNA encoding mouse GFP-SH2B18c was created from cDNA encoding mouse GFP-SH2B1γ (accession no. NM_011363.3) (details in Supplementary Material).

PC12 cells (ATCC) were grown and treated and neurite outgrowth experiments were completed as described previously (18) with modifications described in Supplementary Material.

Frozen tissues were lysed and homogenized with a glass Dounce homogenizer containing lysis buffer, described previously (3). PC12 cells were transfected and lysed as described previously (18). Equal amounts of protein were immunoblotted with antibody to SH2B1 (cat. no. sc-136065; Santa Cruz Biotechnology), β-tubulin (sc-55529; Santa Cruz Biotechnology), or ERK1/2 (4695S; Cell Signaling Technology) as described previously (3) with modifications in Supplementary Material. Expression levels of SH2B1 isoforms in Fig. 1I were quantified using LI-COR Image Studio Lite (version 5.2.5).

Statistical analyses of phenotyping data were performed using GraphPad Prism or CalR (21), a custom package of the R programming language designed to analyze indirect calorimetry using ANCOVA. Body weights were compared by two-way repeated-measures ANOVA with Dunnett or Sidak multiple comparisons tests. Cumulative food intake was analyzed by linear regression. See the Supplementary Material for details of statistics used to analyze RNA-seq data. All other comparisons were analyzed by one-way ANOVA with Dunnett multiple comparisons test or two-tailed Student t test. P < 0.05 was considered significant.

The data generated during this study are available in the Gene Expression Omnibus repository, GSE145202, or available from the corresponding author upon reasonable request. The mouse model generated during this study is available from the corresponding author upon reasonable request.

Exon skipping and an alternative donor site produce four Sh2b1 splice variants (22,23) (Fig. 1B). Consistent with the human tissue expression profile of SH2B1 mRNA (1), we detected substantial Sh2b1α and Sh2b1δ mRNA in mouse brain tissue, including hypothalamus, cortex, and cerebellum, but little to none in peripheral tissues including gonadal WAT, liver, and testes; all of these tissues contained Sh2b1β and Sh2b1γ mRNA (Fig. 1C).

To gain insight into the importance of Sh2b1 splicing in the brain and the contributions of the brain-specific SH2B1α and SH2B1δ isoforms to metabolism in vivo, we used CRISPR/Cas9 to generate mice lacking SH2B1α/δ (Fig. 1B). Genotyping (Fig. 1D) and DNA sequencing (data not shown) of founder animals and their progeny identified mice containing correctly edited Sh2b1^DELαδ alleles. αδHET intercrosses produced pups with Sh2b1 genotypes at the expected Mendelian ratio (data not shown). PCR confirmed that Sh2b1δ mRNA transcripts were absent in brain tissue from Sh2b1^{DELαδ/DELαδ} (αδKO) mice (Fig. 1E). Additionally, RNA sequencing (RNA-seq) confirmed the absence of Sh2b1α/δ transcripts in αδKO hypothalami and the presence of similar levels of Sh2b1β/γ transcripts in Sh2b1^+/+ (wild type [WT]) and αδKO hypothalami (Fig. 1F). We confirmed the absence of SH2B1α/δ proteins, and the continued presence of SH2B1β/γ proteins, in αδKO brain lysates (Fig. 1G). As expected, only β/γ isoforms were detected in liver tissue from WT and αδKO mice (Fig. 1H). SH2B1β/γ protein levels were similar between WT and αδKO mice in liver tissue (Fig. 1H) and between WT, αδKO, and αδHET mice in brain tissue (Fig. 1G and I), whereas SH2B1α/δ protein levels were decreased by ∼40% in αδHET compared with WT brain lysates (Fig. 1I). Thus, αδKO mice exhibit the predicted isoform-specific ablation of SH2B1α/δ, and αδHET mice exhibit reduced SH2B1α/δ, without compensatory alterations in expression of SH2B1β/γ isoforms.

To determine whether SH2B1α/δ influence energy balance, we fed male and female αδKO, αδHET, and WT littermates standard chow and measured their body weight weekly. Both male and female αδKO mice appeared thinner and weighed considerably less than WT littermates by 8 weeks of age (males) (Fig. 2A and B) or 20 weeks of age (females) (Supplementary Fig. 1A). Body weight of male and female αδHET mice did not diverge from that of controls before 25 weeks; however, the weight of αδHET males was significantly decreased compared with controls at 38–42 weeks (Fig. 2C and Supplementary Fig. 1B).

While αδKO males exhibited slightly decreased body length and lean mass, these measures were not different from those of sex-matched WT littermates for αδKO females or αδHET mice of either sex (Fig. 2D and E and Supplementary Fig. 1C and D), suggesting that differences in overall body size were unlikely to have mediated the decreased body weight of αδKO mice. Indeed, compared to controls, fat mass was decreased in male and female αδKO and male αδHET mice (Fig. 2F and Supplementary Fig. 1E). Additionally, percent lean mass was increased and percent fat was decreased in αδKO mice of both sexes (Fig. 2G and H and Supplementary Fig. 1F and G). Similarly, leptin concentrations and adipose tissue weights were decreased in αδKO mice of both sexes (Fig. 2I and J and Supplementary Fig. 1H and I). Thus, while body length was slightly decreased in αδKO males compared with controls, the major effect across sexes was of decreased adiposity in αδKO mice. αδKO liver weight was also decreased compared with controls, which may reflect a decrease in triacylglycerol content (as for animals on HFD [see below]) because the mass of other tissues (e.g., brain) was unchanged (Fig. 2J and Supplementary Fig. 1I). Furthermore, the decreased body weight and fat mass of αδHET males compared with controls suggest that reduced expression of SH2B1α/δ associated with haploinsufficiency of SH2B1α/δ in males is sufficient to impact body weight and adiposity.

To determine whether the decreased body weight of αδKO mice resulted from reduced caloric intake, increased energy expenditure, or both, we measured weekly food intake of αδKO and WT littermates when their body weights were diverging. Male (Fig. 3A and B) and female (Supplementary Fig. 2A) αδKO mice consumed less food than controls. We observed no statistically significant alteration in food intake in αδHET versus control mice, consistent with the minor alteration in their body weight. Metabolic cages revealed no differences by genotype in VO₂, respiratory exchange ratio (RER), or locomotor activity (Fig. 3C–F and Supplementary Fig. 2B–E). There were also no differences by genotype in VO₂, VCO₂, or energy expenditure when body weight, lean mass, or fat mass were taken into account as covariates in the ANCOVA. These data suggest that the decreased weight of chow-fed αδKO mice results primarily from decreased caloric intake.

To determine whether removal of SH2B1α/δ affords protection from diet-induced obesity (DIO), we challenged αδKO and WT littermates with an HFD. Males were used because of their enhanced susceptibility to DIO relative to females (24,25). HFD-fed αδKO mice gained significantly less weight than controls (Fig. 4A). In fact, their body weight remained near to or below that of chow-fed WT mice for the entire study (compare with Fig. 2A). HFD-fed αδKO mice consumed amounts of food similar to those of controls (Fig. 4B and C). The finding that the body weight of HFD-fed αδKO mice was reduced but their food intake was normal suggests that they expended more energy than controls. Body composition measurements revealed that, as for chow-fed mice, the decreased body weight of HFD-fed αδKO mice resulted primarily from decreased fat mass (Fig. 4D–G). While leptin levels were highly variable and not significantly different by genotype, they trended lower in αδKO mice (Fig. 4H). Similar to chow-fed animals, HFD-fed αδKO mice exhibited reduced weight of inguinal WAT and liver (Fig. 4I). Histological analysis of liver tissue detected less lipid in livers of HFD-fed αδKO mice compared with controls, suggesting that decreased steatosis may underlie the lower weight of αδKO livers (Fig. 4J).

To determine whether deletion of SH2B1α/δ impacts glucose metabolism, we first measured glucose and insulin in ad libitum–fed or overnight-fasted αδKO and WT mice (22–27 weeks old). Fed αδKO males displayed normal blood glucose levels but low insulin concentrations (Fig. 5A and B). When fasted, αδKO males exhibited low blood glucose and insulin concentrations (Fig. 5C and D). Glucose and insulin tolerance tests revealed improved glucose tolerance in 28- to 32-week-old chow-fed αδKO males (Fig. 5E and F) and improved glucose tolerance and insulin sensitivity in 18- to 20-week-old HFD-fed αδKO males (Fig. 5G and H). In chow-fed αδKO females, parameters of glucose metabolism were unchanged except that their insulin concentrations were decreased in the fed state (Supplementary Fig. 3A–F). Thus, αδKO mice generally exhibit improved glucose tolerance and insulin responsiveness compared with their more obese littermate controls. Together with the improved glucose homeostasis in the lean αδKO males, the relatively normal glucose metabolism in αδKO females, which exhibit a more modest reduction in body weight, suggested that alterations in glucose homeostasis in αδKO mice might result not directly from the lack of SH2B1α/δ but, rather, from decreased adiposity. We therefore assessed glucose homeostasis in young, 10- to 12-week-old chow-fed male αδKO mice, when their fat content was similar to that of controls (Fig. 6A). Their decreased leptin levels (Fig. 6B) suggested that some differences in adiposity were beginning to develop. We observed no differences by genotype in glucose or insulin concentrations, glucose tolerance, or insulin sensitivity in these young animals (Fig. 6C–H). Thus, the improved glucose homeostasis observed in older αδKO mice is likely to be secondary to their reduced adiposity rather than the result of any direct effect of removal of SH2B1α/δ.

Because SH2B1 modulates leptin signaling and SH2B1α/δ expression is restricted to the brain, the site of leptin action on food intake and energy balance, we examined the possibility that increased leptin signaling might underlie the leanness of αδKO mice. The arcuate nucleus of the hypothalamus is a primary target for leptin (26,27). Leptin action in the arcuate nucleus inhibits the expression of appetite-stimulating neuropeptides agouti-related peptide (AgRP) and neuropeptide Y (NPY) and promotes the expression of appetite-suppressing neuropeptides proopiomelanocortin (POMC) and cocaine- and amphetamine-regulated transcript (CART) (28). Leptin also regulates the expression of other genes in LepRb-expressing cells in the hypothalamus (29). We measured expression of eight relevant leptin-regulated, protein-encoding genes (Agrp, Npy, Pomc, Cartpt, Asb4, Irf9, Ghrh, Serpina3n) in αδKO and WT hypothalami using quantitative PCR (qPCR). To avoid potential confounding by differences in adiposity and/or circulating leptin levels, which would impact leptin action, we assessed young, 10- to 12-week-old male mice. We found no differences by genotype in expression of these genes (Fig. 7A).

To more directly test the effects of SH2B1α/δ on leptin sensitivity, we examined the responses of 6-week-old αδKO and WT mice to low-dose leptin injections. We have previously used this assay to identify increased leptin sensitivity in other mouse lines (3,30,31). At this age, there was no difference in circulating leptin levels between αδKO and WT littermates (Fig. 7B). We observed no difference by genotype in rate of weight loss or food intake reduction (Fig. 7C and D) in response to leptin treatment. In conjunction with the WT-like expression levels of leptin-responsive genes in αδKO hypothalami, these results suggest that removal of SH2B1α/δ does not alter leptin action but, rather, protects mice from obesity primarily via leptin-independent mechanisms.

To gain additional insight into mechanisms underlying the leanness of αδKO mice, we examined the impact of deleting SH2B1α/δ on hypothalamic gene expression using RNA-seq. Given that WT and αδKO mice had similar fat content at the time of hypothalamic dissection (Fig. 6A), changes in gene expression were expected to be adiposity independent. The combined expression of all Sh2b1 transcripts annotated in the Ensembl mouse genome was substantially decreased in αδKO mice, as predicted (Supplementary Fig. 5B). RNA-seq identified 59 additional genes that exhibited statistically significant differential regulation by genotype. Of these, 29 were upregulated and 30 were downregulated in αδKO mice (Fig. 8A and Supplementary Tables 4 and 5). qPCR analysis validated the gene expression changes of a representative sample (Slc5a11, C1qa, Gsg1l, Kdm8) of the up- and downregulated genes identified by RNA-seq (Supplementary Fig. 4A). We compared the hypothalamic transcriptional changes that we observed in αδKO mice with published results from leptin-treated mice (29). There was no correlation between data sets among all genes that were differentially expressed or among the genes that were statistically significantly regulated (Fig. 8B and C). These results provide further evidence that removal of SH2B1α/δ protects mice from obesity via leptin-independent mechanisms.

Interestingly, Gene Ontology (GO) revealed that changes in the transcriptional profile of hypothalamic tissue from αδKO mice were significantly associated with several genes linked to microglial function (Supplementary Fig. 4B). Similarly, querying MouseMine revealed that many of the mouse phenotypes that significantly associate with differentially expressed genes in αδKO mice were related to microglia (Supplementary Fig. 4C). Also, comparison with a previously published mouse hypothalamic single-cell RNA-seq data set that contained expression profiles for all hypothalamic cell types including neurons, microglia, and macrophages (32) indicated that most of the genes with statistically significant differential regulation in αδKO hypothalami were predominantly expressed in microglia and/or macrophages (Supplementary Fig. 4D). In addition, Sh2b1 was expressed in all hypothalamic cell types including neurons, microglia, and macrophages (32). These findings provide evidence that deletion of SH2B1α/δ impacts hypothalamic microglial function. Interestingly, most of the microglia-related genes (C1qa, C1qb, Cx3cr1, Grn, Trem2, Tyrobp) that exhibited statistically significantly upregulated expression in our data set have previously been identified as contributors to complement-mediated synaptic pruning (33–37), presenting the possibility that SH2B1α/δ may regulate structural changes at synapses. The ability of SH2B1α/δ to regulate structural changes at synapses would be consistent with previous in vitro findings that SH2B1 regulates the actin cytoskeleton (10,11,14).

A byproduct of our RNA-seq analysis was greater insight into Sh2b1 transcripts. Our initial quantification of Sh2b1 transcripts using StringTie included only transcripts that had previously been annotated in the Ensembl mouse genome (Supplementary Fig. 5A). However, this analysis suggested that almost no mRNA encoding SH2B1α was detected in WT hypothalami (Supplementary Fig. 5B), which contradicted our qPCR data (Fig. 1C). We reanalyzed the data to include previously unannotated transcripts. This analysis revealed six potential novel Sh2b1 transcripts in WT mice (Supplementary Fig. 5A). Inclusion of unannotated alongside annotated transcripts provided a more accurate measurement of Sh2b1 transcript abundance (Supplementary Fig. 5C). This revised sum of RNA-seq transcripts more closely matches total Sh2b1 gene expression measured by qPCR (Supplementary Fig. 5D).

While four of six potential novel (unannotated) Sh2b1 transcripts are slight variations of known transcripts, the remaining two differ notably from known transcripts in that their eighth intron is not spliced out (Supplementary Figs. 5A and 6). We named these predicted transcripts “Sh2b18c” for their unique eighth exon.

Sh2b18c mRNA contains an in-frame stop codon located 263 base pairs downstream of exon 8 and is predicted to be targeted for nonsense-mediated decay (38). Regardless, to examine potential expression of Sh2b18c-encoded proteins, we generated a GFP-SH2B18c expression vector and transfected it into PC12 cells. GFP-SH2B18c migrated at its predicted molecular size on SDS-PAGE gels (Supplementary Fig. 5E) and enhanced neurite outgrowth to an extent similar to that of SH2B1β (Supplementary Fig. 5F), suggesting that if it were expressed, it might function similarly to SH2B1β or SH2B1γ. However, αδKO brain lysates showed no increase in SH2B1 protein migrating at the expected size of SH2B18c on SDS-PAGE gels (Supplementary Fig. 5E), despite the increased Sh2b18c mRNA in these mice. Thus, when combined with our prediction that the Sh2b18c transcripts would be targeted for nonsense-mediated decay, this lack of an appropriately migrating band makes it unlikely that Sh2b18c transcripts contributed to the αδKO phenotype.

Here we create a novel mouse model that lacks two isoforms of adapter protein SH2B1, α and δ, both of which are expressed almost exclusively in brain tissue. We demonstrate that αδKO mice gain less weight than controls and are protected against DIO. The decreased body weight of the chow-fed αδKO mice results primarily from suppressed food intake. Additionally, the αδKO mice exhibit adiposity-dependent improvements in glucose homeostasis. Thus, SH2B1α and/or SH2B1δ are critical determinants of energy balance and, indirectly, glucose homeostasis.

Most isoform-specific KO models exhibit phenotypes that are similar to, yet less severe than, those of animals with the complete KO (39). Thus, given the obesity of Sh2b1 KO mice, the simplest prediction for mice lacking two of four known SH2B1 isoforms would be an overweight phenotype. However, αδKO mice, lacking two of four known SH2B1 isoforms, and approximately half the normal SH2B1 brain protein content, are underweight. These findings suggest that the different SH2B1 isoforms make nonredundant, even opposing, contributions to the regulation of energy balance. Furthermore, normal expression of SH2B1β/γ is not sufficient to maintain normal energy balance, suggesting that the ratio of SH2B1 isoforms must be carefully titrated for the body to appropriately balance its energy. We investigated whether altering the ratio of the isoforms could be a defense against famine or overeating. However, preliminary studies indicated that the ratios of the various isoforms do not appear to be regulated by a 24-h fast or an HFD (data not shown). Nevertheless, our finding that in the brain WT mice have a high ratio of SH2B1α/δ to SH2B1β/γ protein levels, αδHET mice have a medium ratio, and αδKO mice have a ratio of zero suggests that the lower the ratio of SH2B1α/δ to SH2B1β/γ, the stronger the resistance to weight gain. Therefore, we propose that manipulating the ratio of SH2B1 isoforms, perhaps by identifying molecular targets that could be modified to alter SH2B1 splicing activity or disrupt the expression of SH2B1α/δ specifically, may serve as the basis for new obesity therapeutics. It is also possible that other physiological stressors and/or genetic variants or mutations may impact the SH2B1 isoform ratio and, as a consequence, energy balance.

The lean phenotype of αδKO mice is consistent with known cellular actions of the individual SH2B1 isoforms. Specifically, in PC12 cells, SH2B1α was shown to inhibit SH2B1β enhancement of NGF-induced neurite outgrowth by a process that appears to be dependent on NGF-induced phosphorylation of a specific tyrosine unique to the COOH terminus of SH2B1α (18). Thus, the simplest explanation of the αδKO phenotype is that in these mice, removal of SH2B1α enhances the activity of SH2B1β, and presumably SH2B1γ, to enhance the actions of neurotrophic factor receptors known to recruit SH2B1 isoforms. One can envision that the competing actions of SH2B1 isoforms enable the highly specialized fine-tuning between neurite outgrowth versus retraction that is required for the formation and maintenance of the neuronal synapses that are needed to properly regulate energy balance. Cells in nonbrain tissues may not have a need for such specialized fine-tuning, perhaps explaining why the α isoform evolved almost exclusively in the brain. Indeed, in a variety of assays in other nonbrain cell types, in response to a variety of ligands, SH2B1α functions similarly to SH2B1β/γ (1,40,41). Whether SH2B1δ also has unique functions relevant to the brain is not yet known. Unfortunately, it was not possible to parse the individual contributions of SH2B1α or SH2B1δ by deleting either isoform individually because of the Sh2b1 gene structure. However, based on protein levels of SH2B1 detected by Western blotting, the α isoform is present at much higher levels than the δ isoform, suggesting that removal of the α isoform may make the major contribution to the αδKO phenotype.

Which brain cells might be driving the αδKO phenotype? Despite prior evidence suggesting that SH2B1 contributes to energy balance by modulating LepRb/JAK2 signaling (7), our data suggest that deletion of SH2B1α/δ does not alter cellular responses to leptin. Our RNA-seq data suggest that removal of SH2B1α/δ has the greatest impact on genes associated with cells of the immune system, including microglia. These are the first results to our knowledge that associate SH2B1 isoforms with microglial functions and, thus, highlight a novel focal point for follow-up studies. Microglia could contribute to the αδKO phenotype via complement-mediated synaptic pruning of appetite-regulating neuronal synapses. In fact, most of the microglia-related genes within our significantly differentially regulated gene set (C1qa, C1qb, Cx3cr1, Grn, Trem2, Tyrobp) have been identified as contributors to complement-mediated synaptic pruning (33–37). Whether the lack of SH2B1α/δ affects microglia function directly or indirectly, perhaps via neurons, remains to be determined.

Given the ability of SH2B1 to enhance signaling of neurotrophic factors and affect neuronal architecture, the αδKO phenotype could also be a consequence, at least in part, of altered neuronal activity. BDNF-sensitive neurons are attractive candidates for neurons affected in αδKO mice. BDNF and TrkB are critical regulators of metabolism (42). BDNF/TrkB activity is important for many neuronal processes that contribute to energy balance, including regulation of mature neural circuits through structural changes of dendritic spines at excitatory synapses. SH2B1β has been shown to enhance BDNF-induced neurite outgrowth of PC12 cells, regulate the actin cytoskeleton, and interact with actin-binding protein IRSp53 in primary cultured neurons to regulate the formation of dendritic filopodia—the small membranous protrusions that often develop into dendritic spines (10,11,13,14). Thus, SH2B1 isoforms could collectively fine-tune appetite-regulating neuronal synapses by mediating cytoskeletal rearrangement within dendritic spines of TrkB-expressing neurons, and perhaps the synaptic pruning carried out by microglia. In αδKO mice, we would predict that upregulated SH2B1β/γ activity would increase cytoskeletal rearrangement and thereby improve communication between appetite-regulating neurons to decrease appetite. Consistent with our hypothesis that activity downstream of TrkB is upregulated in αδKO hypothalami, administration of BDNF into various hypothalamic nuclei in mice reduces food intake and weight gain (43). Future experiments will be necessary to clarify whether TrkB or other receptor tyrosine kinases have altered activity in αδKO mice. Additional work will also be required to fully understand why the lean phenotype of αδKO mice appears to have arisen through distinct physiological mechanisms when mice were fed normal chow versus a HFD.

Alternative splicing generates much of the molecular and cellular diversity that exists, spatially and temporally, in the brain (44,45). However, alternative splicing is delicately regulated and has been shown to go awry. Indeed, prior studies have linked alternative splicing aberrations to neurological disorders (46) and metabolic dysregulation (47,48). Our work is unique in that it presents alternative splicing not as a potential cause for disease but, rather, as an opportunity to treat disease. In other words, regardless of the exact cellular mechanisms involved, the bottom line of this report is that disrupting brain-specific alternative splicing of Sh2b1 to delete SH2B1α/δ generates mice that are resistant to weight gain and seemingly healthy otherwise.

To summarize, our findings advance our understanding in four major ways. First, while removal of all SH2B1 isoforms induces obesity in mice (4), we show that deletion of the brain-specific SH2B1α/δ isoforms has the opposite effect, offering protection against obesity. Thus, for the first time, our data demonstrate unique, nonredundant functions of SH2B1 isoforms in vivo. Second, these results indicate that disrupting the alternative splicing of Sh2b1 to delete the α and δ isoforms protects against weight gain, presenting a potential target for obesity therapeutics. Third, our research suggests that the α and δ isoforms of SH2B1 regulate energy homeostasis primarily via leptin-independent mechanisms. Fourth, our RNA-seq data set suggests potential cellular mechanisms by which isoforms of SH2B1 control energy balance—perhaps by refining synapses between appetite-regulating neurons in the hypothalamus. Together, these findings highlight the importance of alternative splicing in regulating brain function relevant to energy balance and illuminate several novel pathways that researchers might follow to gain more information about the mechanism(s) by which SH2B1 isoforms work together to regulate body weight.

Acknowledgments. The authors thank Dr. Ray Joe, Dr. Lei Yin, Dr. Xin (Tony) Tong, and Dr. Deqiang Zhang (University of Michigan) for feedback on experimental design and data analysis and Sarah Cain (University of Michigan) for administrative assistance. The authors thank Dr. Miriam Meisler, Dr. Carey Lumeng, and Dr. Malcolm Low (University of Michigan) for helpful discussions, Dr. Liangyou Rui (University of Michigan) for the gift of the Sh2b1 KO strain, and Dr. Heimo Riedel (West Virginia University) for the cDNA encoding SH2B1γ. The authors acknowledge the Michigan Diabetes Research Center Molecular Genomics Core and Dr. Thomas Saunders, Galina Gavrilina, and Dr. Wanda Filipiak of the University of Michigan Transgenic Animal Model Core for help generating the αδKO mouse model.

Funding. This research was supported by National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health (NIH), grants R01-DK-054222 and R01-DK-107730 (to C.C.-S.), R01-DK-056731 (to M.G.M.), and R01-DK-062876 and R01-DK-092759 (to O.A.M.). J.L.C. was supported by a National Science Foundation Graduate Research Fellowship and a Rackham Predoctoral Fellowship from the Horace H. Rackham School of Graduate Studies at the University of Michigan. A.F. was supported by predoctoral fellowships from the Horace H. Rackham School of Graduate Studies at the University of Michigan (Rackham Merit Fellowship), the Systems and Integrative Biology Training Program (National Institute of General Medical Sciences, NIH, T32-GM-008322), and the Howard Hughes Medical Institute (Gilliam Fellowship for Advanced Study). L.C.D. was supported by an Endocrine Society Summer Research Fellowship and an American Physiological Society Undergraduate Summer Research Fellowship. D.P.B. was supported by predoctoral fellowships from the Medical Scientist Training Program (National Institute of General Medical Sciences, NIH, T32-GM-007863), Training Program for Organogenesis (National Institute of Child Health and Human Development, NIH, T32-HD-007605), and Horace H. Rackham School of Graduate Studies at the University of Michigan (Rackham Merit Fellowship). Mouse body composition and CLAMS studies were partially supported by the Michigan Diabetes Research Center (National Institute of Diabetes and Digestive and Kidney Diseases, NIH, P30-DK-020572), Michigan Nutrition Obesity Research Center (National Institute of Diabetes and Digestive and Kidney Diseases, NIH, P30-DK-089503), and Michigan Mouse Metabolic Phenotyping Center (National Institute of Diabetes and Digestive and Kidney Diseases, NIH, U2C-DK-110678). Generation of the CRISPR/Cas9 mice was partially supported by the Michigan Diabetes Research Center Molecular Genomics Core (NIH, P30-DK-020572). The authors thank MedImmune for the gift of recombinant mouse leptin.

Duality of Interest. D.P.B. was supported by a TYLENOL Future Care Scholarship. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.L.C. directed and conducted experiments and prepared the manuscript including all figures. L.S.A. and A.F. designed and generated the αδKO mice. A.C.R. performed the initial bioinformatics analysis of RNA-seq data and generated plots for Fig. 8B and C and Supplementary Figs. 4B–D, 5A, and 6. J.M.C. performed qPCR assays (Figs. 1C and 7A and Supplementary Figs. 4A and 5D), ran Western blots (Fig. 1G and H and Supplementary Fig. 5E), and made the construct encoding GFP-SH2B18c (Supplementary Fig. 5E). L.C.D. and A.H.B. assisted with body weight and food intake studies, blood sample collections, harvests, and genotyping. D.P.B. prepared and imaged liver samples (Fig. 4J). P.B.V. performed neurite outgrowth experiments (Supplementary Fig. 5F). A.M.C. and E.S.C. assisted with genotyping and mouse colony maintenance. G.C. assisted with body weight studies and genotyping. All authors reviewed and approved the final content. J.L.C., L.S.A., O.A.M., M.G.M., and C.C.-S. developed the concepts and hypotheses, designed the experiments, and interpreted the data. L.S.A., O.A.M., M.G.M., and C.C.-S. made revisions to the manuscript. C.C.-S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this work were presented in poster form or as short oral presentations at the Keystone Symposium: Neuronal Control of Appetite, Metabolism and Weight, Copenhagen, Denmark, 9–13 May 2017; the 48th Annual Scientific Meeting of the Michigan Chapter of the Society for Neuroscience, Ann Arbor, MI, 22 May 2017; the 2017 Gordon Research Conference on Neurotrophic Factors, Newport, RI, 4–9 June 2017; the Keystone Symposium: Functional Neurocircuitry of Feeding and Feeding Disorders, Banff, Alberta, Canada, 10–14 February 2019; ENDO 2019: the Endocrine Society Annual Meeting, New Orleans, LA, 23–26 March 2019; and the 2019 Gordon Research Conference on Neurotrophic Mechanisms in Health and Disease, Newport, RI, 2–7 June 2019.