Research ArticleObesity

Long-acting MIC-1/GDF15 molecules to treat obesity: Evidence from mice to monkeys

See allHide authors and affiliations

Science Translational Medicine  18 Oct 2017:
Vol. 9, Issue 412, eaan8732
DOI: 10.1126/scitranslmed.aan8732

A bigger molecule to help slim down

Obesity is becoming increasingly common worldwide, and the available interventions do not fully address this problem. Surgery is currently the most effective intervention, especially for severe obesity, but it carries more risks than noninvasive treatments and produces permanent side effects. Xiong et al. searched for metabolically regulated proteins and identified the growth differentiation factor 15 (GDF15) pathway as a potential target for intervention. The loss of this protein in mice is associated with weight gain and worsened metabolic parameters. Conversely, the authors showed that treating with GDF15 improved metabolic health in mice, rats, and monkeys. They also designed a modified version of GDF15 (GDF15-Fc fusion) that has a longer half-life and would thus be a better candidate for clinical testing.


In search of metabolically regulated secreted proteins, we conducted a microarray study comparing gene expression in major metabolic tissues of fed and fasted ob/ob mice and C57BL/6 mice. The array used in this study included probes for ~4000 genes annotated as potential secreted proteins. Circulating macrophage inhibitory cytokine 1 (MIC-1)/growth differentiation factor 15 (GDF15) concentrations were increased in obese mice, rats, and humans in comparison to age-matched lean controls. Adeno-associated virus–mediated overexpression of GDF15 and recombinant GDF15 treatments reduced food intake and body weight and improved metabolic profiles in various metabolic disease models in mice, rats, and obese cynomolgus monkeys. Analysis of the GDF15 crystal structure suggested that the protein is not suitable for conventional Fc fusion at the carboxyl terminus of the protein. Thus, we used a structure-guided approach to design and successfully generate several Fc fusion molecules with extended half-life and potent efficacy. Furthermore, we discovered that GDF15 delayed gastric emptying, changed food preference, and activated area postrema neurons, confirming a role for GDF15 in the gut-brain axis responsible for the regulation of body energy intake. Our work provides evidence that GDF15 Fc fusion proteins could be potential therapeutic agents for the treatment of obesity and related comorbidities.


Obesity and its comorbidities are public health concerns that are epidemic worldwide (1). One of the most effective treatments for weight loss is Roux-en-Y gastric bypass surgery, which involves the reduction of the size of the stomach and bypassing a portion of the small intestine (2). Although the procedure results in a marked weight loss, it is an invasive complex surgery that leaves the patient with permanent undesirable side effects; therefore, there is a need for pharmacological therapies that are safe and efficacious and that can be administered long-term (3).

During a search for secreted factors that are metabolically regulated in liver and adipose tissues, we identified macrophage inhibitory cytokine 1/growth differentiation factor 15 (GDF15), an atypical member of the transforming growth factor–β (TGF-β) superfamily (47), to which we will refer as GDF15 going forward. GDF15 is a homodimer cysteine knot protein containing one interchain and eight intrachain disulfide bonds (4). GDF15 is implicated in multiple disorders of metabolism such as obesity, insulin resistance, and cancer cachexia (8). Ablation of GDF15 in genetically engineered mice resulted in increased body weight, whereas overexpression resulted in mice with lower body weight and fat mass and improved metabolic parameters (9, 10). These findings support GDF15 as a regulator of body weight and energy balance; however, many of the physiological effects of GDF15 remain unclear. A definitive receptor or clear signaling mechanism has not conclusively been described, and how GDF15 regulates body weight remains a question. Some insight into GDF15 function comes from studies where both the area postrema (AP) and the nucleus of the solitary tract were ablated in mice: After ablation of these regions, mice no longer responded to the anorectic actions of GDF15 (11). Neurons from this region participate in energy regulation and can be activated by GDF15 (12), suggesting that GDF15 may have a role in controlling the gut-brain–mediated neuronal pathway.

The clinical development of GDF15 as a therapeutic offers several challenges. The native molecule presents poor pharmacokinetics (PK) and drug-like properties, including a circulating half-life of 3 hours in mice and cynomolgus monkeys, low production yield, and manufacturing challenges. To investigate the role of GDF15 in metabolism and produce a potential therapeutic modality to treat obesity, we generated and tested recombinant GDF15 and half-life extended variants of GDF15 in obese mice, rats, and monkeys. These molecules demonstrated strong efficacy in lowering body weight in every species tested and improved multiple metabolic parameters. Therefore, GDF15-based therapeutics could become an efficacious therapy to treat obesity and related metabolic disorders, a major chronic health burden in modern society (1).


GDF15 is up-regulated with obesity, and AAV-GDF15 improves metabolic parameters in mice

GDF15 was identified as a highly up-regulated gene in the liver and fat of ob/ob mice in a microarray study to identify secreted proteins that were differentially expressed in animal models of obesity (Fig. 1A). Circulating concentrations of GDF15 in obese mice, rats, and humans were also significantly higher (P < 0.05) in the obese groups than normal lean controls across species (Fig. 1B). Increased GDF15 concentrations have been reported in various pathological conditions, but the physiological relevance of the increase is unclear (13). To assess whether GDF15 regulates energy metabolism, we administered adeno-associated virus–expressing human GDF15 (AAV-hGDF15) in several obese mouse models. AAV-hGDF15 reduced food intake and body weight and improved metabolic profiles in diet-induced obese (DIO) mice (Fig. 1C), as well as in ob/ob, db/db, and KKAy mice (fig. S1, A to C). The reduction in body mass of the AAV-hGDF15 DIO mice included both fat mass and lean mass, but the loss of fat mass was greater than the loss of lean mass (Fig. 1D). To get a better representation of the loss of body mass, we included a group of untreated 12-week-old mice fed a normal chow diet (12-week lean) as a control, and we subjected the mice to dual-energy x-ray absorptiometry (DEXA) measurements to determine fat and lean mass and bone mineral density (Fig. 1D). AAV-hGDF15 17-month-old DIO mice showed similar body mass composition to the lean 12-week-old chow-fed mice (Fig. 1D). No significant changes in bone mineral density were observed (Fig. 1D). Our findings align with data from other groups reporting that GDF15 transgenic mice were resistant to obesity induced by a high-fat diet (9, 14). Clinical and histopathology analyses of 1-year-old DIO mice treated with AAV-hGDF15 revealed no detrimental effects associated with GDF15 overexpression in the brain, heart, lung, liver, gallbladder, spleen, pancreas, kidney, skeletal muscle, stomach, duodenum, jejunum, ileum, colon, prostate gland, preputial gland, skin, or adipose tissue (table S1).

Fig. 1. GDF15 is up-regulated with obesity, and AAV-GDF15 improves metabolic parameters in DIO mice.

(A) Microarray signal intensity of GDF15 in the liver and fat tissues from C57BL/6 and ob/ob mice (n = 5 to 6). (B) Serum GDF15 concentrations in lean and obese mice (n = 5 to 8), rats (n = 5 to 6), and humans (n = 16). (C) Body weight, average daily food intake 1 month after AAV-hGDF15 injection; glucose, insulin, triglyceride, and cholesterol concentrations 5 months after AAV injection; and serum hGDF15 concentrations 12, 81, and 350 days after AAV injection in male B6D2F1 DIO mice injected with empty vector or AAV-hGDF15 (n = 5 to 15). (D) Fat mass, lean mass, and bone mineral density of 18-month-old male B6D2F1 DIO mice injected with empty vector 12 months after AAV injection, 18-month-old male B6D2F1 DIO mice injected with AAV-hGDF15 12 months after AAV injection, and a group of untreated 12-week-old mice on normal chow (n = 3). *P < 0.05, **P < 0.01, and ***P < 0.001 versus empty vector by analysis of variance (ANOVA).

Recombinant (r) GDF15 proteins and antibodies were tested in a variety of disease models. Both recombinant murine (rm) GDF15 and recombinant human (rh) GDF15 proteins demonstrated very similar efficacy and potency in reduction of food intake in ob/ob mice (fig. S2A). In these mice, treatment with rhGDF15 protein reduced food intake and body weight, whereas treatment with an mGDF15 antibody blocking endogenous GDF15 activity increased body weight and food intake, demonstrating that GDF15 regulates body energy homeostasis (fig. S2, B to E). In DIO mice, daily treatment with rhGDF15 for 4 weeks decreased food intake, body weight, blood glucose, serum insulin, serum triglyceride, and cholesterol concentrations and improved glucose tolerance in a dose-dependent manner (fig. S3A). Similarly, daily treatment with rhGDF15 for 6 weeks in obese cynomolgus monkeys reduced food intake, body weight, plasma insulin, and plasma triglyceride concentrations and improved glucose tolerance (fig. S3, B and C). PK properties of rhGDF15 are shown in fig. S4 (A, in mice, and B, in cynomolgus monkeys).

GDF15 crystal structure and Fc fusion GDF15 molecules

We used a structure-based approach to design GDF15 fusion proteins with a longer half-life and higher production yield than the native protein (11 mg per liter of mammalian culture) after we had solved the crystal structure of GDF15 at 2.3 Å resolution. This structure resembles those of other TGF-β superfamily members (15, 16), forming a homodimer that is stabilized by an interchain disulfide bond (Fig. 2, A and B). Although the GDF15 receptors are unknown, the interface for the type I receptor is speculated to be around the palm/wrist region and type II receptor around the fingertip, similar to the ternary complex of TGF-β/α–TbR1–TbRII (17, 18). The C terminus of GDF15 is packed against a hydrophobic patch of the opposite monomer (Fig. 2C). Ile-112, the last residue observed in the structure, is buried and constrained because of a disulfide bond between Cys-111 and Cys-48. A fusion protein introduced at the C terminus would negatively affect the dimer formation and consequently its function. In contrast, the N terminus of the two GDF15 monomers in the ternary complex are fully exposed and adopt the same orientation, making them desirable sites for engineering the fusion. Because of the distance mismatch between the N terminus of GDF15 dimer (38 Å) (Fig. 2B) and the C terminus of Fc dimer (20 Å) (fig. S5A), direct fusion of Fc to GDF15 subunit was challenging to produce, so a flexible linker was inserted between Fc and GDF15. A fusion protein with a modified Fc deleted hinge region (DhFc) (Fig. 2D) was active in the ob/ob mouse food intake assay [half-maximal effective dose (ED50), 2.1 μg/kg; Fig. 2D] but had a relatively low yield of ~1 mg per liter of culture.

Fig. 2. GDF15 crystal structure and enabled design of GDF15 molecules.

(A) Side view and (B) top view of the crystal structure of GDF15 dimer at 2.3 Å with individual subunits shown in gold and cyan. The N terminus and C terminus are highlighted by dotted spheres. (C) Close-up view of the C terminus of GDF15. One GDF15 monomer is shown in beige surface representation. The C-terminal residue Ile112 of the other monomer is shown in stick representation. The van der Waals radius of Ile112 is shown as dotted spheres. (D) Summary of the ED50 of GDF15 fusion molecules. DhFc, hinge-deleted Fc fragment; CpmFc, charge-paired mutant of Fc fragment; DhCpmFc, hinge-deleted charge-paired Fc fragment; ScFc, single-chain Fc fragment.

To increase protein production, we designed a fusion protein with one GDF15 subunit per Fc dimer by incorporating a set of complementary charges in the CH3 domains of the Fc region (CpmFc; Fig. 2D and fig. S5B) (19). This insertion relieved the Fc CH3 interface constraint on the GDF15 dimer and greatly enhanced the protein yield (270 mg per liter of culture), but the activity was substantially reduced (ED50, 112 μg/kg; Fig. 2D). However, truncating the N-terminal hinge region of the Fc (DhCpmFc; Fig. 2D) retained the protein yield (272 mg per liter of culture) and recovered most of the activity (ED50, 7.8 μg/kg; Fig. 2D). The single-chain Fc version (ScFc; Fig. 2D) demonstrated reasonable product yield and potency (ED50, 18.4 μg/kg; yield, 90 mg per liter conditioned medium; Fig. 2D).

Fc fusion GDF15 molecules improve metabolic parameters in obese mice, rats, and cynomolgus monkeys

PK analyses indicated that DhCpmFc and ScFc were suitable for weekly dosing in mice and cynomolgus monkeys (fig. S6, A and B). Weekly treatment with both analogs demonstrated similar efficacy in DIO mice (Fig. 3A) and obese cynomolgus monkeys (Fig. 3, B and C). In DIO mice, both molecules decreased body weight, serum insulin, triglyceride and total cholesterol concentrations, and food intake and decreased glucose area under the curve (AUC) during oral glucose tolerance testing (OGTT) (Fig. 3A). Similarly, in obese cynomolgus monkeys, both molecules decreased body weight, plasma glucose, insulin, triglyceride levels, and food intake (Fig. 3B). Plasma cholesterol concentrations were transiently increased only during the treatment (Fig. 3B). These transient increases were correlated with short-term (overnight) fasting in monkeys; the initial exposure to GDF15 and increase in cholesterol concentrations were lessened as food intake began to rebound during the treatment period and returned to the baseline cholesterol concentrations similar to the vehicle-treated group. Insulin AUC measured during OGTT improved at weeks 2 and 5 during the treatment period but was not significantly different at week 8 during the washout period (Fig. 3C). DhCpmFc was also tested in Zucker fatty rats. It decreased body weight, serum insulin concentrations, food intake, and glucose AUC during OGTT (fig. S7).

Fig. 3. Fc fusion GDF15 molecules improve metabolic parameters in obese mice and obese cynomolgus monkeys.

(A) Body weight, OGTT glucose AUC, insulin, triglyceride, cholesterol concentrations, and food intake of male C57BL/6 DIO mice treated weekly with vehicle, rosiglitazone, or Fc fusion GDF15 proteins (0.1, 1, or 10 nmol/kg) for 5 weeks (n = 12). *P < 0.05, **P < 0.01, and ***P < 0.001 versus vehicle by ANOVA. (B and C) Cynomolgus monkeys received weekly subcutaneous doses (days 0, 7, 14, 21, 28, and 35) of vehicle (closed square; n = 10), ScFc (open triangle; n = 5), or DhCpmFc (open circle; n = 8) for 6 weeks. (B) Plasma chemistries and body weights were measured 6 days after each weekly injection before the morning meal (days 6, 13, 20, 27, 34, and 41); food intake measurements occurred daily. Data are expressed as group means ± SEM. (C) OGTT data are represented as AUC (glucose AUC, mg/dl per hour; insulin AUC, ng/ml per hour; 0 to 180 min). Cynomolgus monkeys suspected to be positive for anti-GDF15 antibodies were excluded from data analysis (five monkeys for ScFc and two for DhCpmFc). Statistical analysis was performed by analysis of covariance (ANCOVA), and statistical significance is denoted as #P < 0.05, ##P < 0.01, and ###P < 0.01 versus vehicle for ScFc; *P < 0.05, **P < 0.01, and ***P < 0.01 versus vehicle for DhCpmFc.

GDF15 delays gastric emptying, changes food preference, and activates AP neurons

To understand how GDF15 affects energy intake, we measured gastric emptying in mice receiving vehicle or rhGDF15. Vehicle-treated groups emptied 20 and 31% of ingested phenol red solution at 5 and 15 min after oral gavage, respectively; however, rhGDF15-treated mice emptied 13 and 18% of the solution at these same time points (Fig. 4A). Vagal efferent and afferent signals play important roles in postprandial gastric emptying (20, 21), so we further tested GDF15 in mice after bilateral subdiaphragmatic vagotomy. As expected, vagotomized animals showed delayed gastric emptying compared to sham-operated animals (fig. S8A). GDF15-induced delay of gastric emptying was preserved in sham-operated animals but was not observed in mice after vagotomy, suggesting that GDF15 may act through a vagal-mediated mechanism (fig. S8A). We also investigated whether the GDF15 effect is partially mediated by the glucagon-like peptide 1 (GLP-1) pathway. GDF15 treatment did not change circulating GLP-1 concentrations (fig. S8B), suggesting that GDF15 most likely exerts its effect independently of GLP-1 production.

Fig. 4. GDF15 delays gastric emptying, changes food preference, and activates AP neurons.

(A) Percentage of liquid emptied out of the stomach 5 and 15 min after oral gavage (n = 8). Vehicle or GDF15 protein was dosed 30 min before oral administration of phenol red solution. *P < 0.05, **P < 0.01, and ***P < 0.001 versus vehicle by ANOVA. (B) Percentage of normal chow or condensed milk diet consumed before or after GDF15 treatment (n = 8). *P < 0.05, **P < 0.01, and ***P < 0.001, normal chow versus condensed milk diet, by ANOVA. (C) Representative thick sections of the AP showing c-FOS–positive neurons in mice dosed with amylin or high-dose GDF15 protein 30 min after intraperitoneal injection. Scale bars, 50 μm. (D) Fc immunoreactivity in colon sections after administration of DhCpmFc or controls. Scale bars, 25 μm. Corresponding plasma concentrations are indicated below the respective images. (E) Colocalization of Fc immunoreactivity with the neuronal marker pgp9.5 in colon sections. Scale bars, 25 μm.

We also observed that GDF15 treatment changed taste preferences. Without rhGDF15 treatment, animals preferred a condensed milk diet of moderately high fat content (32.5% calories from fat) over standard rodent chow when the two diets were presented at 1:1 ratio, with about 80% of total food consumed being condensed milk chow (Fig. 4B). After rhGDF15 treatment, the preference for the condensed milk chow was largely reduced, with about 55% of total food consumed being condensed milk chow and 45% standard rodent chow (Fig. 4B). When animals were given access to both diets, standard rodent chow consumed by the rhGDF15-treated group was actually more than what the vehicle-treated group consumed, although the total amount of food consumed by the rhGDF15-treated group was still less than for vehicle-treated group (fig. S8C), but rhGDF15 treatment reduced intake of standard rodent chow when it was the only food animals had access to (fig. S8D). When animals had access only to condensed milk diet, rhGDF15 reduced intake of condensed milk diet (fig. S8E).

To get a better understanding of action sites of GDF15, we conducted binding studies in the stomach, jejunum, colon, and liver and examined neuron activities after GDF15 treatment. We observed up-regulation of the immediate early gene c-Fos in the AP of the brain in response to exogenously administered rhGDF15 protein compared to a known activator of c-FOS in the AP, amylin (Fig. 4C and fig. S9). On average, 12 (±3) c-FOS–positive cells per 0.04 mm2 were counted in the AP after exposure to amylin or rhGDF15 (10 mg/kg) protein, as compared to 1 (±1) cells in the 1 mg/kg group and vehicle group.

The AP receives afferent output from the gut. Therefore, we investigated whether exogenously administrated GDF15 protein binds to the gut by immunohistochemical detection of the Fc portion of DhCpmFc after intravenous injection. Compared to Fc control, the binding pattern of Fc immunoreactivity indicated that DhCpmFc localized to neuronal beds in the colon, stomach, and jejunum (Fig. 4D and fig. S10, A and B) but not to the liver (fig. S10C). Although lower circulating concentrations of control Fc versus DhCpmFc could have contributed to differences in Fc staining intensity, the similarity of the staining intensity in DhCpmFc-treated gut sections at 2 hours and 4 days despite about 10-fold lower exposure at the latter time point (Fig. 4D) suggests that the labeling is specific because nonspecific bound Fc constructs are likely to be removed during perfusion. In DhCpmFc-treated gut sections, we conducted double labeling with antibodies to Fc and to PGP9.5, a neuronal specific marker. PGP9.5 staining overlapped with Fc immunoreactivity in putative neuronal structures, suggesting a connection between GDF15 and the myenteric plexus of the gut (Fig. 4E).


Obesity is one of the biggest health care problems society is currently facing, and treatments that do not involve invasive surgery are greatly needed. GDF15 is a potential approach to treat this problem; however, challenges exist in developing GDF15 as a viable therapeutic agent, particularly due to the short serum half-life of the native protein. Engineering a molecule with a long serum half-life while maintaining efficacy is the key to resolving these issues.

Fusion of proteins to a half-life–extending moiety is a common practice to improve the PK of therapeutic molecules. For example, fusion partners include proteins such as serum albumin and Fc, as well as nonprotein entities such as polyethylene glycol. To enhance the PK properties of GDF15, we evaluated the Fc fragment of an immunoglobulin G1 (IgG1) antibody as a fusion partner. The first design challenge was to identify the site where GDF15 could be fused, because analysis of the GDF15 crystal structure suggested that the C terminus of GDF15 is buried and not accessible for fusion. Fortunately, the N terminus of GDF15 provided opportunities for fusion partner design. The second challenge was to determine a stable placement of the Fc dimer with the GDF15 dimer. GDF15 forms a homodimer that is stabilized by an interchain disulfide bond, similar to other TGF-β superfamily members. To overcome the potential physical constraints caused by the Fc CH3 dimer, we designed a fusion protein where each GDF15 monomer is fused to an Fc dimer through a flexible linker (G4S)4. The flexible linker has additional benefits; that is, it allows the GDF15 receptor to access the canonical type I and type II receptor binding sites. The engineered Fc-GDF15 fusion protein showed increased yield and maintained biological activity.

Control of digestive system function is locally regulated by the enteric nervous system. This complex nervous system of the gut regulates many processes, including peristaltic contraction of smooth muscles (22), which in turn exerts control over digestive tract motility. Parasympathetic and sympathetic fibers connect the central nervous system with the digestive tract, and vagal nerves connecting the gut to the medulla oblongata are one of the main pathways controlling gut motility (23). Circumventricular organs of the medulla oblongata, particularly the AP, can be accessed by circulating factors, and GDF15 activation of these neurons and resulting effects on gastric motility could explain changes seen in gastric emptying. Also, bariatric surgery, a very effective treatment for obesity, is associated with delayed gastric emptying (24, 25) and loss of preference to nutrition-dense high-fat diet (25, 26). It is intriguing that GDF15 affected the same observed biological activities as bariatric surgery, and circulating concentrations of GDF15 are further increased in obese patients after bariatric surgery (27).

Previous studies have demonstrated a widespread hindbrain activation of c-FOS upon GDF15 administration (11), encompassing not only the AP but also the dorsal motor nucleus of the vagus and the nucleus of the solitary tract (11). Here, we detected c-FOS activation in the AP at the high dose only but did not observe consistent activation of the nucleus of the solitary tract. However, given the robust effects of GDF15 on metabolic parameters observed at lower doses, the lack of c-FOS signal is most likely due to low sensitivity of our immunohistochemical procedure. An earlier study showed evidence of c-FOS activation reaching into hypothalamic areas as well (12). This activation could be a result of both neuronal and humoral stimuli affecting the AP and surrounding structures. Amylin, another secreted protein that reduces gastric emptying, has also been suggested to work through the AP (28). Here, activation of AP neurons causes activation of hypothalamic areas involved in food intake and energy homeostasis (29). A complex network of neuronal activation that slows gastric emptying and induces satiety may be an underlying mechanism of GDF15.

Limitations associated with the development of any GDF15 therapeutic involve a lack of understanding of the relationship of GDF15 to its receptor and tissue localization of that receptor. In the basal state, GDF15 is found in the circulation in very low amounts; however, GDF15 concentrations increase in several types of cancer and many other disease states, which may imply GDF15 resistance (3032). We administered doses well above the concentrations seen in human circulation, possibly overcoming any state of resistance. Knowing and understanding the receptor of GDF15 and the associated signaling cascade may help explain GDF15 increases observed in different disease states. In addition, knowing tissue expression patterns of a GDF15 receptor could also help identify target-associated liabilities related to GDF15 receptor activation. It is also critical to identify and develop more effective and better-tolerated anti-obesity drugs that will maintain activity for a long period of time. Using an overexpression system with AAV, we found that hGDF15 decreased body weight for a year, suggesting that GDF15 pharmacological treatment could generate durable body weight lowering.

In conclusion, we demonstrated a strong effect of GDF15 in lowering body weight in obese mice and monkeys. Use of unconventional approaches allowed us to engineer long-acting forms of the GDF15 protein that could mimic the large effect size observed with the native protein in multiple species. We also identified the effects of GDF15 on gastric emptying, food preference, and AP neuron activities, confirming the role of GDF15 in the gut-brain axis. Our data support potential use of GDF15 in the clinic, with properties that could benefit individuals with obesity and associated metabolic comorbidities.


Study design

A microarray study was conducted to search for genes regulated by change of energy states. GDF15 was identified as a gene up-regulated in obesity. We examined the effects of AAV-mediated in vivo overexpression of GDF15 or daily dosing of rGDF15 protein on obesity and related metabolic parameters in various disease models. We also solved GDF15 crystal structure (details in table S2) and generated long-acting GDF15 molecules using unconventional Fc fusion design. We then tested the efficacy of weekly dosing with Fc-GDF15 molecules in rodents and nonhuman primates. To further understand the mechanism of action of GDF15, we conducted a series of rodent studies and examined the effect of GDF15 on gastric emptying, food preference, and activation of AP neurons. Investigators who performed rodent and cynomolgus experiments and analyzed the pattern and extent of binding and tissue pathology were randomly assigned to animals or samples. Investigators were blinded to the phenotypes of samples from different groups. For quantitative experiments, individual subject-level data are shown in table S3.


All rodent studies were conducted at Amgen Inc. and were approved by Amgen Institutional Animal Care and Use Committee (IACUC). Animals were maintained in rooms with a 12-hour light/dark cycle, temperature of 22°C, and humidity of 30 to 70%. Animals had free access to food and water and were maintained on standard rodent chow unless otherwise indicated.

The cynomolgus monkey efficacy study was conducted at Kunming Biomed International (KBI). All housing conditions and procedures were approved by and in compliance with the IACUC of KBI. Work was conducted under an IACUC-approved protocol in an assessment and accreditation of laboratory animal care (AALAC)–accredited facility. Cynomolgus monkeys were individually housed in stainless steel cages elevated from the floor and adequately sized to promote species-typical postures and behaviors, with daily environmental enrichment provided. Animal holding rooms were maintained on a 12-hour light/dark cycle with temperature between 18° and 26°C and humidity between 40 and 80%.

Human serum samples

Human samples were procured through Bioreclamation. Anonymized samples were collected after appropriate consent was obtained by Bioreclamation. Clinical sites contracted by Bioreclamation reviewed and approved the sample collection protocol. Age-matched obese patients [body mass index (BMI) > 27; n = 16] and healthy volunteers (n = 16) were enrolled. Fasting serum samples were collected.

Microarray study

Liver and epididymal fat tissues were collected from 10- to 11-week-old male C57BL/6 and ob/ob mice (The Jackson Laboratory) and snap-frozen in liquid nitrogen. Tissue RNA was isolated using TRIzol (Invitrogen) and RNeasy kit (QIAGEN). A custom array was built to contain probes for 4000 computationally annotated potential secreted factors, and array studies were conducted using SureScan microarray scanner (Agilent).

AAV vector generation

To generate the pAAV-GDF15 DNA construct, GDF15 cDNA was inserted between an elongation factor 1a promoter and a bovine growth hormone poly A, and an expression cassette was flanked by inverted terminal repeats of AAV. Recombinant AAV (rAAV) was produced by triple transfection of suspension-adapted human embryonic kidney–293 T cells, as previously described (33).

Cells from a 1-liter Erlenmeyer flask (corresponding to 0.8 × 109 cells) were collected, resuspended in 8 ml of serum-free Dulbecco’s modified Eagle’s medium, and lysed by three rounds of freezing and thawing. Before application onto an AVB Sepharose (GE Healthcare Life Sciences) column, the lysate containing the rAAV vector was treated as previously described (34). The filtered lysate containing rAAV vectors was then loaded onto an AVB Sepharose column and eluted with glycine HCl (pH 3.0). The eluted rAAV vector was immediately neutralized with 1 M tris-HCl (pH 8.0) and dialyzed against phosphate-buffered saline (PBS)–MK buffer (PBS with 1 mM MgCl2 and 2.5 mM KCl). Purified rAAV vectors were then concentrated in the dialysis cassette with Slide-A-Lyzer Concentrating Solution (Thermo Fisher Scientific). Purified rAAV vectors were titered using QuickTiter AAV Quantitation Kit (Cell Biolabs Inc.).

AAV injection

Six-month-old male B6D2F1 DIO mice (100006, The Jackson Laboratory) were started at 6 weeks of age and maintained on 60% high-fat diet (D12492, Research Diets Inc.); 7- to 8-week-old male ob/ob mice (000632, The Jackson Laboratory), 11- to 12-week-old male KKAy mice (002468, The Jackson Laboratory), and 8- to 9-week-old male db/db mice (000642, The Jackson Laboratory) were used in the AAV studies. Three to 4 days before AAV injection, animals were sorted by body weight, whole-blood glucose, serum insulin, triglyceride, and cholesterol concentrations for each treatment group to have similar baselines. Tail vein blood samples for this analysis were collected in DIO mice after 4 hours of fasting and in ob/ob, KKAy, and db/db mice after 1 hour of acclimation to the procedure room with free access to food. AAV was diluted in saline to 8 × 1012 genomic units/ml, and 100 μl was injected through the tail vein for each animal.

Body mass composition and bone mineral density measurements

Body mass composition (fat mass and lean mass) and bone mineral density were measured using a Lunar PIXImus DEXA device (PIXImus).

GDF15 and insulin ELISA

Serum GDF15 concentrations were measured using a mouse/rat GDF15 enzyme-linked immunosorbent assay (ELISA) and a human GDF15 ELISA kit (R&D Systems). In rodent studies, serum insulin concentrations were measured using a high range mouse insulin ELISA kit (ALPCO) or rat insulin ELISA kit (ALPCO). In cynomolgus monkey studies, serum insulin concentrations were measured using Roche cobas e411 automatic clinical analyzer.

Glucose, triglyceride, and cholesterol measurements

In rodent studies, blood glucose concentrations were measured using an AlphaTRAK 2 glucose strip (Abbott Diagnostics), serum triglyceride using a triglyceride quantitation kit (Sigma-Aldrich), and serum cholesterol using a cholesterol quantitation kit (Sigma-Aldrich). In cynomolgus monkey studies, plasma glucose, triglycerides, and cholesterol concentrations were measured using Roche cobas c311 automatic clinical analyzer.

Recombinant GDF15 protein

GDF15 is expressed in BL21DE3 cells as inclusion bodies, which were solubilized in 6 M guanidine and refolded in 20% glycerol, 0.15 M arginine HCl, 50 mM tris (pH 8.5), and 4 mM cysteine/cystamine for 4 days at 4°C. The refolded GDF15 proteins were purified on HiPrep DEAE FF 16/10 (GE Healthcare Life Sciences) anion exchange column, a Superdex 200 16/600 (GE Healthcare Life Sciences) size exclusion column, and a C4 protein reverse-phase column (Vydac).

GDF15 antibody production

C57BL/6 mice were immunized with the mGDF15 DNA construct and boosted with T cell epitope–conjugated mGDF15 protein. Immunized animals with high serum titers were sacrificed, and B cells were harvested from spleens and lymph nodes and fused with SP2/0 myeloma to generate hybridomas. Conditioned medium was screened for GDF15 binding affinity using ELISA. For the high-affinity binding antibodies, concentrated conditioned medium from hybridomas cultured to exhaustion was further purified through protein A/G affinity columns for epitope binning and affinity binding Biacore assays. Selected high-affinity hybridoma lines were subcloned to generate monoclonal hybridomas. A conditioned medium of a large-scale monoclonal hybridoma culture was purified for activity measurement.

Crystal structure

GDF15 protein was polished on a Supderdex G75 16/600 column equilibrated with a crystallization buffer [25 mM sodium acetate (pH 4.6) and 100 mM sodium chloride]. The peak fractions were pooled and concentrated to 1 mg/ml and screened against commercial high-throughput crystallization kits. GDF15 was crystallized using a sitting drop vapor diffusion method with the well solution containing 15% (w/v) tascimate (pH 7.0), 0.1 M Hepes (pH 7.0), and 2% polyethylene glycol 3350. For data collection, a single crystal was transferred to a cryoprotection solution containing the well solution supplemented with 25% ethylene glycol and frozen in liquid nitrogen.

Data collection and structure solution

The x-ray diffraction data sets were collected by synchrotron beamline 08ID at Canadian Light Source, processed using MOSFLM, and scaled using SCALA in CCP4 package (35). The structure was solved by molecular replacement method using program PHASER with the BMP6 monomer as a search model (Protein Data Bank access code: 2R52) (36). Model building was carried out in COOT (37), and structure refinement was carried out with REFMAC5 (38) in CCP4 package. The figures were prepared using PyMOL (The PyMOL Molecular Graphics System; Schrödinger LLC).

Fc fusion GDF15 proteins

Stable expression of Fc fusion molecules was completed by transfection in CHO-K1 host cells followed by puromycin/hygromycin antibiotic selection. Protein expression was carried out over 6 to 7 days. Conditioned medium was then directly loaded on to a MabSelect Sure column (GE Healthcare) and eluted using sodium acetate. The conditioned elution pool was purified using an SP-Sepharose HP column followed by a Superdex 200 26/60 column (GE Healthcare) and then formulated in 10 mM sodium acetate, 0.9% sucrose, and 0.005% polysorbate 80.

Ob/ob mice food intake assay

Six- to 7-week-old male ob/ob mice (The Jackson Laboratory) were sorted into different treatment groups, with each group having comparable pretreatment body weight and food intake. Animals were treated with different doses of proteins through subcutaneous injection, and food intake was measured for 20 to 22 hours. In studies using mGDF15 antibodies, animals were treated with GDF15 antibodies or control antibodies through subcutaneous injection 24 hours before rmGDF15 injection.

Chronic treatment in DIO mice

Chronic treatment studies were conducted with 19- to 20-week-old male C57BL/6 DIO mice (Taconic Biosciences) and maintained on 60% high-fat diet (Research Diets Inc.) from 6 weeks of age. Animals were acclimated to the dosing regimen by subcutaneous saline injection for 2 weeks. After acclimation, animals were sorted into treatment groups, with each group having comparable baseline body weight and 4-hour fasting whole-blood glucose and serum insulin, triglyceride, and cholesterol concentrations. Body weight and food intake were measured on days 7, 14, 21, 28, and 34. Four-hour fasting blood samples were collected at days 14 and 28, and overnight fasting blood samples were collected on day 35 after the first treatment. OGTTs (20% glucose, 10 ml/kg) were conducted on days 14 and 35 after blood sampling.

Efficacy studies in cynomolgus monkeys

Naïve male spontaneously obese cynomolgus monkeys with a BMI (>41 kg/m2) were prescreened for health. Cynomolgus monkeys were acclimated to single housing, experimental procedures, and handling for 6 weeks before treatment. During this acclimation period, plasma samples were collected from animals that fasted overnight (5:00 p.m. to 7:00 a.m.) for glucose, insulin, triglycerides, and total cholesterol measurements; a baseline OGTT (see procedure below) was conducted as well. Cynomolgus monkeys were randomized into two intravenous-treated groups receiving either vehicle (n = 10) or rhGDF15 (n = 10; 0.5 mg/kg) daily for 39 days (days 0 to 38), and three subcutaneously treated groups receiving either vehicle (n = 10), ScFc (n = 10; 1.5 mg/kg), or DhCpmFc (n = 10; 1.5 mg/kg) once weekly for 6 weeks (days 0, 7, 14, 21, 28, and 35). The randomization assured comparable baseline values for each of the characteristics. To minimize stress, cynomolgus monkeys receiving subcutaneous injections were held in separate holding rooms from cynomolgus monkeys receiving intravenous injections for the entire study (acclimation, treatment, and washout). A 5-week recovery and washout period followed the ~6-week treatment period. Overnight fasting (5:00 p.m. to 7:00 a.m.) blood samples were collected at predose days −24, −17, and −10, as well as on days 6, 13, 20, 27, 34, and 41 (6 days after each weekly dose) during the treatment phase. During the washout phase, blood samples were collected on days 48, 55, 62, 69, and 76.

A total of four OGTTs were performed, including during baseline (day −17), week 2 (day 13) and week 5 (day 34) of the treatment phase, and week 8 (day 55) of the washout phase. After overnight fasting (5:00 p.m. to 7:00 a.m.), monkeys going through OGTT were moved to the procedure room the morning of the challenge and secured in specially designed restraint chairs. Monkeys were alert/conscious and remained in restraint chairs for the entire OGTT procedure.

Plasma was collected 3 min before d-glucose oral gavage. Monkeys then received an oral gavage of freshly prepared 40% d-glucose (Sigma-Aldrich) solution at 4 g/kg (10 ml/kg dose volume) via nasogastric tube followed by 5 ml of saline flush. Timer was started at the end of the saline flush. Plasma samples were subsequently collected at 15, 30, 60, 120, and 180 min after the glucose administration. Monkeys were fed normal morning food rations after returning to cages in their home room after completion of the OGTT.

Body weight was measured once a week (before the morning meal), and food intake was monitored daily for each monkey throughout the study. Each monkey received unlimited food for a limited amount of time (1 hour) at the morning and evening feedings, about 8 hours apart. A 150-g apple snack, for a limited amount of time (1 hour), was provided between meals. The remaining food or apple was removed and weighed after each meal or snack to calculate food intake. Blood samples for determination of the drug concentration-time profile were collected throughout the treatment phase and washout phase (same points as shown above for blood chemistry determination). Plasma was analyzed for drug concentrations using ELISA. Cynomolgus monkeys suspected of having developed anti-GDF15 antibodies (n = 5 for ScFc and n = 2 for DhCpmFc) as evidenced by rapid clearance and an altered PK profile were excluded from data analysis.

Chronic treatment in Zucker fatty rats

Chronic treatment studies were conducted with 11-week-old male Zucker fatty rats (Charles River Laboratories) maintained on standard rodent chow (Teklad 2020 global diets, Envigo). Animals were acclimated to the dosing regimen by subcutaneous saline injection for 2 weeks. After acclimation, animals were sorted into treatment groups, with each group having comparable baseline body weight and 4-hour fasting whole-blood glucose and serum insulin, triglyceride, and cholesterol concentrations. Body weight and food intake were measured on days 5, 12, 19, 26, and 33. Four-hour fasting blood samples were collected on days 12 and 26, and overnight fasting blood samples were collected on day 35. OGTTs (20% glucose, 10 ml/kg) were conducted on days 12 and 35 after blood sampling.

Gastric emptying

Eight- to 9-week-old male C57BL/6 mice were fasted overnight and sorted into treatment groups, with each group having comparable body weight. RhGDF15 protein was injected 30 min before oral administration of 0.05% phenol red solution (1 mg/kg). Animals were euthanized by decapitation at 5, 15, or 30 min after administration of phenol red solution. The stomach was quickly ligated at the esophageal sphincter and the pyloric sphincter. The stomach was dissected and homogenized for 30 s in 8 ml of 0.1 N sodium hydroxide. 0.3 ml of 20% trichloroacetic acid was added to the homogenate. The mixture was briefly vortexed and centrifuged at 3000 rpm for 20 min. After centrifugation, 2 ml of the supernatant was collected and mixed with 0.2 ml of 4 N sodium hydroxide. Samples (200 μl) were transferred to a 96-well plate and read at optical density OD560 using a plate reader (Molecular Devices). A group of animals was sacrificed immediately after oral gavage of 0.05% phenol red solution; the phenol red reading of this group was used as 100% to calculate gastric emptying rate for the other groups of mice.

Food preference

Nine-week-old male Sprague Dawley rats (Charles River Laboratories) were single housed and provided daily with 60 g of standard rodent chow (Teklad 2020 global diets, Envigo) on one side of the wire cage top and 60 g of condensed milk chow (D12266B, Research Diets) on another side of the wire cage top. Fresh food was provided daily. Food intake was measured daily for 7 days to establish baseline and measured 1 day after GDF15 injection. Two groups of animals provided with 120 g of each type of food daily were used as controls.

c-FOS experiment

Male C57/B6 mice (n = 4 per group; Charles River Laboratories) were dosed intraperitoneally with amylin (25 μg/kg) (Bachem), or hGDF15 (1 or 10 mg/kg) or vehicle [5 mM sodium acetate buffer (pH 4.5)]. Thirty min after injection, mice were terminally anesthetized by intraperitoneal injection of Fatal Plus (100 mg/kg) (Western Medical Supply Inc.) and perfused with PBS followed by 4% paraformaldehyde in PBS. Brains were postfixed in 4% paraformaldehyde in PBS overnight and cryoprotected in 30% sucrose. Tissue sections (30 μm) were processed using a commercial kit (PerkinElmer Life Science) based on tyramide signal amplification with rabbit polyclonal anti–c-FOS antibody (sc-52, Cisbio) and 1:800 biotinylated goat anti-rabbit IgG (Jackson ImmunoResearch Laboratories Inc.).

Biodistribution experiment

Male C57/B6 mice (Charles River Laboratories) weighing 21 to 22 g were dosed in a blinded fashion via the tail vein with GDF15-Fc (10 mg/kg), human control Fc in buffer [5 mM sodium acetate (pH 4.5)], or vehicle [5 mM sodium acetate (pH 4.5)] (n = 3 per group). Mice were sacrificed either 2 hours (n = 1 per group) or 4 days (n = 2 per group) after injection. Blood was collected from the tail vein and then mice were perfused for immunohistochemistry as described above. Tissue sections (12 μm) were incubated with goat anti-human Fc antibody (Sigma-Aldrich) followed by 1:500 Alexa Fluor 488 donkey anti-goat IgG (Invitrogen). Double labeling was performed by co-incubation with rabbit anti-PGP9.5 (Sigma-Aldrich) followed by co-incubation with Alexa Fluor 555 donkey anti-rabbit IgG.

Statistical analysis

All data are means ± SEM. Comparison of statistical difference between two experimental groups was determined by unpaired t test. For statistical analysis of rodent experiments with more than two treated groups, ANOVA was performed, followed by Dunnett’s multiple comparison test. For statistical analysis of cynomolgus experiment: All data are means ± SEM. CAST-MD Application using SAS version 9.2 on a Linux system was used. Blood chemistry parameters were log-transformed before statistical analysis. The statistical analysis on parameters with baseline values was performed at each assessment separately using ANCOVA with the baseline value as the covariate. For both ANCOVA variations, homogeneous and heterogeneous models were fit using PROC MIXED, and the final model was selected on the basis of the lowest Akaike Information Criterion for finite sample sizes. In the event of nonconvergence, Levene’s test was used to assess homogeneity. If required, then a nonparametric rank-based ANCOVA was applied. Baseline by group interaction terms was included in the ANCOVA model for a given parameter if it was found to be statistically significant (P < 0.05) in greater than 30% of assessment models for that parameter.


Fig. S1. Improved metabolic parameters in AAV-hGDF15–injected ob/ob, db/db, and KKAy mice.

Fig. S2. Effects of rmGDF15, rhGDF15, and mGDF15 antibodies on food intake in ob/ob mice.

Fig. S3. Metabolic parameters in obese mice and obese cynomolgus monkeys treated with rhGDF15 protein.

Fig. S4. PK of GDF15 proteins.

Fig. S5. Model of DhCpmFc fusion protein.

Fig. S6. PK of long-acting GDF15 proteins.

Fig. S7. Reduced food intake and body weight in Zucker fatty rats treated with DhCpmFc protein.

Fig. S8. No delay in gastric emptying in GDF15-treated mice after vagotomy.

Fig. S9. Thick sections of the AP showing c-FOS–positive neurons in additional amylin and high-dose GDF15 peptide–dosed mice.

Fig. S10. Biodistribution of Fc immunoreactivity in jejunum, stomach, and liver.

Table S1. Pathology report of DIO mice 1 year after control AAV or AAV-GDF15 injection.

Table S2. Statistics of crystallographic data and refinement.

Table S3. Individual subject-level data of quantitative studies (provided as an Excel file).


Acknowledgments: We thank S. Pan, G. Cutler, L. Marshall, H. Tian, L. Ling, Z. Cao, and L. Yang for contributions to the secreted factor and microarray study, G. Shimamoto for purification, and A. Foreman-Wykert, PhD, Certified Medical Publication Professional (Amgen Inc.) for editorial support. Funding: This research was funded by Amgen Inc. Author contributions: Rodent studies were designed by Y.X. and executed by Y.X., T.T., and J.Y.; protein molecules were designed by K.W. and X.M. and produced by N.N. and D.K.; crystal structure was solved by X.M. and interpreted by X.M. and Z.W.; c-Fos and biodistribution studies were designed by S.M. and C.H. and executed by H.L.; cynomolgus studies were designed and analyzed by M.M.V. and R.K.; the cynomolgus PK study was designed and analyzed by J.D.; the study with human samples was designed by X.W.; AAV was produced by K.J.L.; the manuscript was written and revised by Y.X., K.W., X.M., C.H., S.M., R.K., K.L., Z.W., and M.M.V. Competing interests: All authors are employees of and hold stock or stock options in Amgen Inc. Data and materials availability: Recombinant GDF15 Fc fusion protein can be provided by and at Amgen’s sole discretion pending scientific review and a completed material transfer agreement with Amgen. Requests from an academic or nonprofit institution should be submitted to: Requests from a for-profit entity should be submitted to: BDopportunities{at}

Stay Connected to Science Translational Medicine

Navigate This Article