GWAS Identifies Risk Locus for Erectile Dysfunction and Implicates Hypothalamic Neurobiology and Diabetes in Etiology

Erectile dysfunction (ED) is a common condition affecting more than 20% of men over 60 years, yet little is known about its genetic architecture. We performed a genome-wide association study of ED in 6,175 case subjects among 223,805 European men and identified one locus at 6q16.3 (lead variant rs57989773, OR 1.20 per C-allele; p = 5.71 × 10−14), located between MCHR2 and SIM1. In silico analysis suggests SIM1 to confer ED risk through hypothalamic dysregulation. Mendelian randomization provides evidence that genetic risk of type 2 diabetes mellitus is a cause of ED (OR 1.11 per 1-log unit higher risk of type 2 diabetes). These findings provide insights into the biological underpinnings and the causes of ED and may help prioritize the development of future therapies for this common disorder.

Main Text

Erectile dysfunction (ED) is the inability to develop or maintain a penile erection adequate for sexual intercourse.

ED has an age-dependent prevalence, with 20%–40% of men aged 60–69 years affected.

The genetic architecture of ED remains poorly understood, owing in part to a paucity of well-powered genetic association studies. Discovery of such genetic associations can be valuable for elucidating the etiology of ED and can provide genetic support for potential new therapies.

We conducted a genome-wide association study (GWAS) in the population-based UK Biobank (UKBB) and the Estonian Genome Center of the University of Tartu (EGCUT) cohorts and hospital-recruited Partners HealthCare Biobank (PHB) cohort. Subjects in UKBB were of self-reported white ethnicity, with subjects in EGCUT and PHB of European ancestry, as per principal components analyses (Supplemental Material and Methods).
ED was defined as self-reported or physician-reported ED using ICD10 codes N48.4 and F52.2, or use of oral ED medication (sildenafil/Viagra, tadalafil/Cialis, or vardenafil/Levitra), or a history of surgical intervention for ED (using OPCS-4 codes L97.1 and N32.6) (Supplemental Material and Methods). The prevalence of ED in the cohorts was 1.53% (3,050/199,352) in UKBB, 7.04% (1,182/16,787) in EGCUT, and 25.35% (1,943/7,666) in PHB (Table S1). Demographic characteristics of the subjects in each cohort are shown in Table S2. The reasons for the different prevalence rates in the three cohorts may include a higher median cohort age for men in PHB (65 years, compared to 59 years in UKBB and 42 years in EGCUT; Table S2), “healthy volunteer” selection bias in UKBB,

a lack of primary care data availability in UKBB, and intercultural differences, including “social desirability” bias.

Importantly, we note that the assessment of exposure-outcome relationships remains valid, despite the prevalence likely not being representative of the general population prevalence.

GWASs in UKBB revealed a single genome-wide significant (p < 5 × 10−8) locus at 6q16.3 (lead variant rs57989773, EAFUKBB [C-allele] = 0.24; OR 1.23; p = 3.0 × 10−11). Meta-analysis with estimates from PHB (OR 1.20; p = 9.84 × 10−5) and EGCUT (OR 1.08; p = 0.16) yielded a pooled meta-analysis OR 1.20; p = 5.71 × 10−14 (heterogeneity p value = 0.17; Figures 1A–1C). Meta-analysis of all variants yielded no further genome-wide loci. Meta-analysis of our results with previously suggested ED-associated variants also did not result in any further significant loci (Supplemental Material and Methods; Table S3), nor did X chromosome analysis in UKBB.

Figure thumbnail gr1
Figure 16q16.3 (Lead Variant rs57989773) Is an Erectile Dysfunction-Associated Locus and Exhibits Pleiotropic Phenotypic Effects


The association of rs57989773 was consistent across clinically and therapy defined ED, as well as across different ED drug classes (Figures 1C and S1). No further genome-wide significant loci were identified for ED when limited to clinically or therapy defined case subjects (2,032 and 4,142 case subjects, respectively).
A PheWAS of 105 predefined traits (Table S4) using the lead ED SNP rs57989773 found associations with 12 phenotypes at a p value < 5 × 10−4 (surpassing the Bonferroni-corrected threshold of 0.05/105), including adiposity (nine traits), adult height, and sleep-related traits. Sex-stratified analyses revealed sexual dimorphism for waist-hip ratio (WHR; unadjusted and adjusted for body mass index) and systolic and diastolic blood pressure (Figure 1D; Table S5).
The lead variant at the 6q16.3 locus, rs57989773, lies in the intergenic region between MCHR2 and SIM1, with MCHR2 being the closest gene (distances to transcription start sites of 187 kb for MCHR2 and 284 kb for SIM1). Conditional and joint analysis (Supplemental Material and Methods) revealed no secondary, independent signals in the locus. Previous work has implicated the MCHR2-SIM1 locus in sex-specific associations on age at voice-breaking and menarche.

The puberty timing-associated SNP in the MCHR2-SIM1 region (rs9321659; ∼500 kb from rs57989773) was not in LD with our lead variant (r2 = 0.003, D’ = 0.095) and was not associated with ED (p = 0.32) in our meta-analysis, suggesting that the ED locus represents an independent signal.

To identify the tissue and cell types in which the causal variant(s) for ED may function, we examined chromatin states across 127 cell types

for the lead variant rs57989773 and its proxies (r2 > 0.8, determined using HaploReg v.4.1) (Supplemental Material and Methods). Enhancer marks in several tissues, including embryonic stem cells, mesenchymal stem cells, and endothelial cells, indicated that the ED-associated interval lies within a regulatory locus (Figure 2A; Table S6).

Figure thumbnail gr2
Figure 2Functional Analysis of 6q16.3 Implicates SIM1 in ED Pathogenesis


To predict putative targets and causal transcripts, we assessed domains of long-range three-dimensional chromatin interactions surrounding the ED-associated interval (Figure 2B). Chromosome conformation capture (Hi-C) in human embryonic stem cells

showed that MCHR2 and SIM1 were in the same topologically associated domain (TAD) as the ED-associated variants, with high contact probabilities (referring to the relative number of times that reads in two 40-kb bins were sequenced together) between the ED-associated interval and SIM1 (Figures 2B and S2). This observation was further confirmed in endothelial precursor cells,

where Capture Hi-C revealed strong connections between the MCHR2-SIM1 intergenic region and the SIM1 promoter (Figure 2C), pointing toward SIM1 as a likely causal gene at this locus.

We next used the VISTA enhancer browser

to examine in vivo expression data for non-coding elements within the MCHR2-SIM1 locus. A regulatory human element (hs576), located 30-kb downstream of the ED-associated interval, seems to drive in vivo enhancer activity specifically in the midbrain (mesencephalon) and cranial nerve in mouse embryos (Figure 2D). This long-range enhancer close to ED-associated variants recapitulated aspects of SIM1 expression (Figure 2D), further suggesting that the ED-associated interval belongs to the regulatory landscape of SIM1. Taken together these data suggest that the MCHR2-SIM1 intergenic region harbors a neuronal enhancer and that SIM1 is functionally connected to the ED-associated region.

Single-minded homolog 1 (SIM1) encodes a transcription factor that is highly expressed in hypothalamic neurons.

Rare variants in SIM1 have been linked to a phenotype of severe obesity and autonomic dysfunction,

including lower blood pressure. A summary of the variant-phenotype associations at the 6q16 locus in human and rodent models is shown in Table S7. Post hoc analysis of association of rs57989773 with autonomic traits showed nominal association with syncope, orthostatic hypotension, and urinary incontinence (Figure S3). The effects on blood pressure and adiposity seen in individuals with rare coding variants in SIM1 are recapitulated in individuals harboring the common ED-risk variants at the 6q16.3 locus (Figure 1D), suggesting that SIM1 is the causal gene at the ED-risk locus. SIM1-expressing neurons also play an important role in the central regulation of male sexual behavior as mice that lack the melanocortin receptor 4 (encoded by MC4R) specifically in SIM1-expressing neurons show impaired sexual performance on mounting, intromission, and ejaculation.

Thus, hypothalamic dysregulation of SIM1 could present a potential mechanism for the effect of the MCHR2-SIM1 locus on ED.

An alternative functional mechanism may be explained by proximity of the lead variant (rs57989773) to an arginase 2 processed pseudogene (LOC100129854), a long non-coding RNA (Figure 2A). RPISeq

predicts that the pseudogene transcript would interact with the ARG2 protein, with probabilities of 0.70–0.77. Arginine 2 is involved in nitric oxide production and has a previously established role in erectile dysfunction.

GTEx expression data

demonstrated highest mean expression in adipose tissue, with detectable levels in testis, fibroblasts, and brain. Expression was relatively low in all tissues, however, and there was no evidence that any SNPs associated with the top ED signal were eQTLs for the ARG2 pseudogene or ARG2 itself.

As a complementary approach, we also used the Data-driven Expression Prioritized Integration for Complex Traits and GWAS Analysis of Regulatory or Functional Information Enrichment with LD correction (DEPICT and GARFIELD, respectively; Supplemental Material and Methods)

tools to identify gene-set, tissue-type, and functional enrichments. In DEPICT, the top two prioritized gene-sets were “regulation of cellular component size” and “regulation of protein polymerization,” whereas the top two associated tissue/cell types were “cartilage” and “mesenchymal stem cells.” None of the DEPICT enrichments reached an FDR threshold of 5% (Tables S8–S10). GARFIELD analyses, which assesses enrichment of GWAS signals in regulatory or functional regions in different cell types, also did not yield any statistically significant enrichments, therefore limiting the utility of these approaches in this case.

ED is recognized to be observationally associated with various cardiometabolic traits and lifestyle factors,

including type 2 diabetes mellitus (T2D), hypertension, and smoking. To further evaluate these associations, we first conducted LD score regression

to evaluate the genetic correlation of ED with a range of traits. LD score regression identified ED to share the greatest genetic correlation with T2D, limb fat mass, and whole-body fat mass (FDR-adjusted p values < 0.05; Table S11).

Next we performed Mendelian randomization

(MR) analyses to evaluate the potential causal role of nine pre-defined cardiometabolic traits on ED risk (selected based on previous observational evidence linking such traits to ED risk

), i.e., T2D, insulin resistance, systolic blood pressure, LDL cholesterol, smoking heaviness, alcohol consumption, body mass index, coronary heart disease, and educational attainment (Tables S12–S15). MR identified genetic risk to T2D to be causally implicated in ED: each 1-log higher genetic risk of T2D was found to increase risk of ED with an OR of 1.11 (95% CI 1.05–1.17, p = 3.5 × 10−4, which met our a priori Bonferroni-corrected significance threshold of 0.0056 [0.05/9]), with insulin resistance likely representing a mediating pathway

(OR 1.36 per 1 standard deviation genetically elevated insulin resistance, 95% CI 1.01–1.84, p = 0.042). Sensitivity analyses were conducted to evaluate the robustness of the T2D-ED estimate (Figure S5, Table S13), including weighted median analyses (OR 1.12, 95% CI 1.02–1.23, p = 0.0230), leave-one-out analysis for all variants (which indicated that no single SNP in the instrument unduly influenced the overall value derived from the summary IVW estimate

), and a funnel plot (showing a symmetrical distribution of single-SNP IV estimates around the summary IVW causal estimate). The MR-Egger regression (intercept p = 0.35) provided no evidence to support the presence of directional pleiotropy as a potential source of confounding.

We also identified a potential causal effect of systolic blood pressure (SBP), with higher SBP being linked to higher risk of ED (MR-Egger OR 2.34 per 1 standard deviation higher SBP, 95% CI 1.26–4.36, p = 0.007, with MR-Egger intercept [p = 0.007] suggesting presence of directional pleiotropy). LDL cholesterol (LDL-C) showed minimal evidence of a causal effect (OR 1.07 per 1 standard deviation higher LDL-C, 95% CI 0.98–1.17, p = 0.113), and there was limited evidence to support a role for smoking heaviness or alcohol consumption (Table S15). Genetic risk of coronary heart disease (CHD) showed weak effects on risk of ED, suggesting that pathways leading to CHD may be implicated in ED (OR 1.08, 95% CI 1.00–1.17, p = 0.061). Further, we identified no causal effects of BMI (using a polygenic score or a single SNP in FTO) or education on risk of ED.
Genetic variants may inform drug target validation by serving as a proxy for drug target modulation.

ED is most commonly treated using phosphodiesterase 5 (PDE5) inhibitors such as sildenafil. To identify potential phenotypic effects of PDE5 inhibition (e.g., to predict side effects or opportunities for repurposing), we looked for variants in or around PDE5A, encoding PDE5, which showed association with the ED phenotype. Of all 4,670 variants within a 1 Mb window of PDE5A (chromosome 4:119,915,550–121,050,146 as per GRCh37/hg19), the variant with the strongest association was rs115571325, 26 kb upstream of PDE5A (ORMeta 1.25, nominal p value = 8.46 × 10−4; Bonferroni-corrected threshold [0.05/4,670] = 1.07 × 10−5; Figure S6). Given the weak association with ED, we did not evaluate this variant in further detail.

We have gained insight into ED, a common condition with substantial morbidity, by conducting a large-scale GWAS and performing several follow-up analyses. By aggregating data from 3 cohorts, including 6,175 ED-affected case subjects of European ancestry, we identified a locus associated with ED, with several lines of evidence suggesting SIM1, highly expressed in the hypothalamus, to be the causal gene at this locus. Our findings provide human genetic evidence in support of the key role of the hypothalamus in regulating male sexual function.

Mendelian randomization implicated risk of T2D as a causal risk factor for ED with suggestive evidence for insulin resistance and systolic blood pressure, corroborating well-recognized observational associations with these cardiometabolic traits.

Further research is needed to explore the extent to which drugs used in the treatment of T2D might be repurposed for the treatment of ED. Lack of evidence for a causal effect of BMI on ED risk in MR analysis (using multiple SNPs across the genome) suggests that the association of the lead SNP (rs57989773) with BMI arises from pleiotropy and that the association of this variant with ED risk is independent of its association with adiposity.

In conclusion, in a large-scale GWAS of more than 6,000 ED-affected case subjects, we provide insights into the biological underpinnings of ED and have elucidated causal effects of various risk factors, including pathways involved in the etiology of T2D. Further large-scale GWASs of ED are needed in order to provide additional clarity on its genetic architecture and etiology and to shed light on potential new therapies.

Optimal Prevention of Dysplasia Requires Complete Ablation of All Intestinal Metaplasia

This meta-analysis demonstrated that the rate of recurrent dysplasia was doubled when residual Barrett epithelium remained after endoscopic ablation.

Current guidelines suggest that endoscopic ablation be offered to patients with confirmed dysplasia in a segment of Barrett esophagus (BE). The authors of this meta-analysis examined long-term outcomes (almost 13,000 patient follow-up years) after ablation of dysplastic BE, comparing the 86% of patients who had complete remission of intestinal metaplasia with the 14% of patients who had eradication of dysplasia but with persistent metaplasia.

Dysplasia recurred in 5% of those with complete ablation of all metaplasia versus 12% of those who had ablation of dysplasia with persistent metaplasia. The development of high-grade dysplasia or cancer was also twice as likely when metaplasia persisted (3% vs.6%)


This important insight into the management of BE patients after ablation makes it clear that the goal should be complete eradication of all Barrett metaplasia, which decreases the risk for recurrent dysplasia and, more importantly, for developing high-grade dysplasia or cancer. Careful follow-up and retreatment of any persistent metaplasia is part of the eradication process. BE can also recur after ablation, which likewise increases the risk for an adverse outcome. Thus, regular surveillance is mandatory. Finally, earlier study findings suggest that high-dose proton-pump inhibitor therapy is another important component in preventing BE recurrence. This detailed process must be followed if optimal outcomes are to be achieved.


Rethinking Bone Disease in Kidney Disease


Renal osteodystrophy (ROD) is the bone component of chronic kidney disease mineral and bone disorder (CKD‐MBD). ROD affects bone quality and strength through the numerous hormonal and metabolic disturbances that occur in patients with kidney disease. Collectively these disorders in bone quality increase fracture risk in CKD patients compared with the general population. Fractures are a serious complication of kidney disease and are associated with higher morbidity and mortality compared with the general population. Furthermore, at a population level, fractures are at historically high levels in patients with end‐stage kidney disease (ESKD), whereas in contrast the general population has experienced a steady decline in fracture incidence rates. Based on these findings, it is clear that a paradigm shift is needed in our approach to diagnosing and managing ROD. In clinical practice, our ability to diagnose ROD and initiate antifracture treatments is impeded by the lack of accurate noninvasive methods that identify ROD type. The past decade has seen advances in the noninvasive measurement of bone quality and strength that have been studied in kidney disease patients. Below we review the current literature pertaining to the epidemiology, pathology, diagnosis, and management of ROD. We aim to highlight the pressing need for a greater awareness of this condition and the need for the implementation of strategies that prevent fractures in kidney disease patients. Research is needed for more accurate noninvasive assessment of ROD type, clinical studies of existing osteoporosis therapies in patients across the spectrum of kidney disease, and the development of CKD‐specific treatments. © 2018 The Authors. JBMR Plus published by Wiley Periodicals, Inc. on behalf of the American Society for Bone and Mineral Research.


Renal bone disease or renal osteodystrophy (ROD) is a complex disorder of bone in patients with chronic kidney disease (CKD).13 Progressive kidney disease results in metabolic and hormonal disturbances that impair bone quality and is characterized by abnormal remodeling (low, normal or high turnover) with or without abnormalities in mineralization. Altered bone and mineral metabolism in kidney disease is part of a broader systemic disorder defined by the Kidney Disease Improving Global Outcomes (KIDGO) as CKD‐mineral and bone disorder (CKD‐MBD).4 CKD‐MBD is manifested by either one or a combination of: 1) abnormalities of calcium, phosphate, parathyroid hormone (PTH), or vitamin D metabolism; 2) abnormalities of bone turnover, mineralization, volume or strength, and linear growth; and 3) vascular or soft tissue calcification. Manifestations of CKD‐MBD begin early in CKD, with near‐normal kidney function, and the severity of CKD‐MBD and its clinical outcomes of increased fracture risk increase in parallel with declining renal function.57

In practical terms, our ability to confidently diagnose ROD type in CKD and to initiate strategies that could prevent fractures remains limited by the lack of accurate and noninvasive diagnostic tools. The measurement of calcium, phosphate, vitamin D, and PTH identify metabolic abnormalities associated with CKD‐MBD, but these remain poor markers of ROD and have insufficient discrimination of ROD turnover type and mineralization.8, 9 Bone turnover markers (BTMs), routinely used in non‐CKD patients to monitor fracture risk and osteoporosis therapy, lack validation in CKD and end‐stage chronic kidney disease (ESKD) and are consequently used infrequently. Transiliac crest bone biopsy remains the gold standard tool to assess bone quality in metabolic bone diseases; however, bone biopsy is invasive, expensive, painful, requires time‐consuming measurements, and is available at only a few centers worldwide.

Over the past decade, advancements in the field of metabolic bone diseases have the potential to alter the paradigm of how we approach renal bone disease. In 2017, KDIGO updated its guidelines to recommend risk classification of patients with kidney disease for fracture by measurement of areal bone mineral density (BMD) by dual‐energy X‐ray absorptiometry (DXA).10 Noninvasive measurement of cortical and trabecular microarchitecture is now possible with both high‐resolution imaging techniques and novel algorithms that reinterpret grayscale variation in DXA images. The diagnostic utility of BTMs in kidney disease is also being refined and fracture risk assessment tools based on clinical risk factors alone have been developed. Below we review the state of the field pertaining to the epidemiology, pathology, diagnosis, and management of renal bone disease in patients with CKD 3‐5D. Furthermore, we provide our personal interpretation of the current issues and advocate for a greater clinical awareness of this condition and the pressing need to develop and test strategies to prevent fractures in these patients.

Pathophysiology of Renal Osteodystrophy

ROD is the bone component of CKD‐MBD and is defined as alterations in bone morphology associated with progressive CKD that can be quantified by bone histomorphometry.(4) It is important to note that although CKD is defined as an estimated glomerular filtration rate (eGFR) <60 mL/min/1.73m2, changes of CKD‐MBD are present even with mild renal impairment (eGFR 60–90 mL/min/1.73m2).1113 The historically complex nomenclature pertaining to ROD was simplified by the KDIGO working group and aligned with the standard nomenclature recommended by the American Society for Bone and Mineral Research.4, 14 Classification of ROD type was based on the turnover (high, normal, or low), mineralization (normal or abnormal), and volume (high, normal, or low) of bone (TMV classification system, Table 1). The aim of the revised classification was to encompass the most significant bone abnormalities in kidney disease that would inform management decisions.

Table 1. Bone Turnover, Mineralization, and Volume (TMV) Classification System for Renal Osteodystrophya
Turnover Mineralization Volume
Low Low
Normal Normal
High High
  • Reprinted with permission from aMoe et al.4

Historically, bone abnormalities in patients with kidney disease were attributed to alterations in PTH and 1,25 dihydroxyvitamin D [1,25(OH)2D]. The characterization of circulating and bone‐derived hormones and the changes that occur with progressive kidney disease have fundamentally altered our understanding of the changes of CKD‐MBD and the pathogenesis of ROD. A detailed discussion of these changes is beyond the scope of this review. However, progressive CKD is generally associated with increased levels of PTH, fibroblast growth factor 23 (FGF‐23), osteoprotegerin, sclerostin, and Dickkopf‐related protein 1 (DKK1) and reductions in α‐Klotho, serum 25‐hydroxyvitamin D [25(OH)D], and [1,25(OH)2D].12, 13, 1522 The expression of bone turnover such as bone‐specific alkaline phosphatase (BSAP), procollagen type 1 N‐terminal (P1NP), C‐terminal telopeptide of type 1 collagen (CTX), and tartrate‐resistant acid phosphatase (TRAP‐5b) is more variable and dependent not only on the degree of renal impairment but also bone turnover.23 Together, these changes impact bone formation, resorption, and mineralization through their effects on osteoblast and osteocyte function.

Expression of skeletal proteins such as FGF‐23, dentin matrix protein 1, and matrix extracellular phosphoglycoprotein is also altered in CKD. In a study by Periera and colleagues,24 FGF‐23 and DMP1 expression was increased across all stages of CKD (compared with healthy controls), whereas there was no difference in MEPE expression. FGF‐23 and DMP1 were inversely related to osteoid accumulation, whereas MEPE was inversely related to bone volume, suggesting a role for FGF‐23 and DMP1 in bone mineralization and MEPE in the regulation of bone volume. The Wnt/β‐catenin pathway is essential for normal osteoblast differentiation and function and therefore normal bone formation. Sclerostin and Dickkopf‐1 (DKK1) are two circulating inhibitors of this pathway; these inhibit lipoprotein receptor‐related protein 5/6 activation of Wnt signaling and impair normal osteoblast differentiation.21, 2527 Sclerostin is primarily expressed in skeletal tissue, and its expression is maintained during aging; in contrast, expression of DKK1 is more general and decreases in bone with age.2830 Inhibition of sclerostin and DKK1 leads to increased bone formation in humans and animal models.3135 Furthermore, animal studies suggest that DKK1 overexpression negatively impacts bone healing, suggesting a role for DKK1 inhibition during the fracture repair process.36, 37 Sclerostin and DKK1 levels are elevated in CKD, with sclerostin levels increased early in CKD, and generally preceding the rise of FGF‐23 and β‐catenin.27, 38 In a study of ESKD patients, sclerostin levels were inversely associated with reduced bone formation and bone loss over a 1‐year period.39, 40 These studies highlight the complex endocrine and paracrine bone‐renal signaling pathways and suggest that the pathogenesis of ROD is driven by changes within osteocytes that occur early in CKD.

In 2001, the National Institutes of Health defined osteoporosis as a skeletal disorder characterized by compromised bone strength, predisposing to an increased risk of fracture.41 Bone strength is a combined measure that reflects both bone density and quality. Bone density can be determined by DXA; however, bone quality is more broadly defined and pertains to bone material properties, such as bone remodeling, microdamage, microarchitecture, and collagen and mineral characteristics.42 Disorders of bone quality accumulate with age and directly affect the mechanical properties of bone and therefore fracture risk. In health, damaged areas of bone are constantly targeted for ongoing remodeling and repair. In CKD, some or all aspects of bone quality may be impaired (Table 2).4348 This includes defective mineralization (osteomalacia), abnormal remodeling (low‐ or high‐turnover bone disease), increased microarchitectural impairment (cortical porosity), and accumulation of microdamage and advanced glycation end products (AGE) cross‐linking.43, 44

Table 2. Bone Changes Associated With Hormonal and Metabolic Changes of End‐Stage Kidney Disease
Decreased bone density
Alterations in bone microarchitecture
• Cortical porosity
• Cortical thinning and trabecularization
• Trabecular thinning and dropout
• Disruption in balance and orientation of newly formed and mature bone
Decreased bone quality
• Mineralization (osteomalacia)
• Abnormal remodeling (loss of normal repair processes)
о Adynamic bone disease
о Low turnover
о High turnover
• Microdamage accumulation
о Reduced resistance to impact
• Advanced glycation end products cross‐linking
о Loss of elasticity and tissue embrittlement

Kidney disease patients have both traditional and CKD‐specific risk factors for bone disease and fractures (Table 3). For example, older age, low body weight, postmenopausal status, a history of previous fractures, increased risk of falls, and the use of immunosuppressive medications that promote bone loss are all common in CKD cohorts.4952 Metabolic disturbances that occur due to CKD, including decreased levels of nutritional and activated vitamin D, disordered calcium and phosphorous metabolism, premature hypogonadism, hyperparathyroidism, and metabolic acidosis all contribute to abnormalities in bone strength.4, 5356 Patients with kidney disease are also more likely to have reduced physical activity, postural hypotension, and decreased muscle mass, which increase susceptibility to falls and fractures.57, 58 The age‐specific risk of fracture associated with CKD is higher in younger age groups, but the absolute risk of fracture increases with age, suggesting an interaction between CKD‐specific and traditional risk factors for fracture in older CKD and ESKD patients.7, 59, 60

Table 3. General and CKD‐Specific Risk Factors for Bone Loss and Fractures
General risk factors CKD‐specific
Patient‐related (non‐modifiable) • Hyperparathyrodism
• Age • Low nutritional and activated vitamin D
• Sex • Disordered mineral metabolism
• Ethnicity • Chronic inflammation
• Past history of fracture • Metabolic acidosis
• Premature hypogonadism
General (modifiable) • Medications
• Low physical activity о Steroids
• Smoking о Phosphate binders (eg, aluminium)
• Alcohol о CNI
• Medications (eg, streoids) • Dietary restriction
• Diabetes • Dialysis‐related amyloidosis
• Sarcopenia
• Chronic inflammatory disorders • Higher prevalence of general risk factors for osteoporosis
  • CKD = chronic kidney disease; CNI = calcineurin inhibitor.

Bone biopsy studies in CKD patients have provided important insights into the patterns of ROD observed in patients across the spectrum of CKD. In patients with CKD stages 3 to 5 (non‐dialysis), some data suggest that up to three‐quarters of patients have histologic evidence of renal osteodystrophy.1, 6166 Depending on the cohort studied, there is considerable variation in the prevalence of ROD types, and findings include a predominance of high or adynamic bone disease and even normal bone. In ESKD, recent large bone biopsy studies have characterized tissue‐level impairments in bone quality that reflect current CKD‐MBD management strategies. In a seminal study, Malluche and colleagues used the TMV classification system to evaluate 630 bone biopsies from adult hemodialysis patients from Europe and the United States.9 For turnover, 58%, 25%, and 18% of patients had low, high, and normal turnover, respectively. There were clear racial differences in turnover: low bone turnover predominated in whites (62%) and normal or high turnover predominated in blacks (68%). For mineralization, defects were uncommon (3% of patients). For volume, low, normal, or high cancellous bone volume was equally distributed among whites, but high volume predominated in blacks. Furthermore, blacks had normal cortical thickness with higher porosity, but whites had an equal distribution of low or normal thickness with high or normal porosity. Trabecular microarchitecture was also examined, with trabecular thickness being low in 37% of patients, normal in 40% of patients, and high in 13% of patients. Trabecular separation was normal in most (78%) of patients. Interestingly, both black and white patients with high bone turnover had increased porosity, and more than 80% of patients with low cancellous bone volume had thin trabeculae. These findings were supported in a recent study, where bone histomorphometric assessment of turnover (bone formation rate/bone surface [BFR/BS]) was performed in 492 dialysis patients.8 Low turnover was the dominant lesion being present in 59% (n = 289), whereas high turnover was present in 17% (n = 83). In a smaller study of 35 hemodialysis patients, those with low bone turnover had more microstructural abnormalities (lower cancellous bone volume and trabecular thickness) than those with high or normal turnover.67 Conversely, those with high turnover had reduced mineral content and reduced stiffness.

These data suggest that early CKD is characterized by high bone turnover. In ESKD, low‐turnover disease is the dominant lesion; however, high‐turnover lesions are present in a significant proportion of patients, whereas mineralization defects are comparably low in both groups. Importantly, the studies above highlight that ROD is more than bone turnover but a global disorder of bone quality and strength. It has been proposed that ROD and fractures related to kidney disease be considered as a subtype of osteoporosis (analogous to steroid‐induced osteoporosis), as bone quality and strength are impaired to a greater extent than age‐related osteoporosis.68 Although intentionally provocative, this definition highlights the need for a more practical and easily translatable definition of ROD. This should not only incorporate bone biopsy findings but surrogate parameters of bone strength (such as BTMs, DXA) that are easily accessible in daily clinical practice, facilitate diagnosis and inclusion in therapeutic trials, and inform treatment decisions.

Fracture Epidemiology in CKD and ESKD

The incidence and prevalence of fractures increases with CKD stage and has been reported to be 2‐ to 17‐fold higher in CKD patients compared with the general population.7, 59, 6972 Recent studies have provided important insights into secular trends of fracture epidemiology in ESKD, highlighting differences in fracture rates between general population and ESKD cohorts and in fracture incidence rates at the central and peripheral skeleton.

Longitudinal studies using USRDS and US Medicare data have compared the incidence of hip fractures in ESKD and non‐ESKD cohorts.73, 74 In general, hip fracture rates in ESKD increased steadily from the early 1990s until the mid 2000s (an increase of 43%), with a nonsignificant reduction in incidence after 2004 (Fig. 1).73 More recent data from the US Nationwide Inpatient Sample examined age‐ and sex‐standardized hip fracture rates in ESKD and found a 12.6% decrease between 2003 and 2011.75 However, in 2010, hip fracture rates in ESKD remained 27% higher compared with 1996, and individuals aged 66 years or older with ESKD had a markedly higher incidence of hip fractures compared with those without ESKD (31.9 versus 8.0 per 1000 patient‐years). Furthermore, although these data suggest that hip fracture rates are decreasing, peripheral (arm and leg) fractures have more than doubled over the corresponding period.60 It is important to put these data in perspective, and the small decrease in hip fractures reported must be interpreted in the context of a persistently high overall fracture rate in ESKD patients compared with the general population. All studies report that current fracture rates are significantly higher in ESKD compared with general population cohorts, and fractures remain more prevalent today than they were in 1996.

Adjusted hospitalized fracture rates per 1000 person‐years in Medicare point‐prevalent hemodialysis and non‐ESKD patients aged 66 years or older. Reprinted with permission from Arneson et al.73

The morbidity, mortality, and health care costs associated with fractures are higher for patients with kidney disease compared with the general population.68 Patients with CKD and ESKD experience longer hospitalization after a fracture, with reported mortality rates between 16% and 60%.74, 76, 77 Many patients do not return to their premorbid level of function after a hip fracture, with as many as 80% discharged to a long‐term‐care facility after a fracture.5 It was estimated that in 2010 hip fracture‐associated expenses in patients with CKD and ESKD were in excess of $600 million USD,(5) and in the general US population, the costs of fractures and associated morbidity are projected to increase to more than $25 billion USD by 2025.78

Noninvasive Assessment of Bone Quality and Fracture Risk in CKD

Bone biopsy is the gold standard for assessing bone quality in kidney disease and informs treatment options based on bone formation rates and mineralization characteristics. However, its utility as an everyday clinical tool is limited by lack of availability and long duration of time required to process and analyze bone tissue. Therefore, there is great interest in the use of noninvasive approaches to assess bone quality in kidney‐related bone disease so that fracture risk classification and the selection of antifracture treatments can be implemented broadly in the renal clinic. Imaging modalities such as DXA, Trabecular bone score (TBS), conventional Quantitative computed tomography (QCT), High‐resolution peripheral QCT (HR‐pQCT) and micro magnetic resonance imaging (MRI) assess bone density and/or structural aspects of bone quality, whereas PTH and BTMs assess aspects of bone quality that cannot be assessed by imaging, such as bone formation rates and mineralization. We will discuss the role of each of these in CKD.

Bone Density and the Fracture Risk Assessment Tool

DXA is the clinical standard to determine BMD and measure fracture risk in the general population and is integral to the World Health Organization definition of osteoporosis (T‐score ≤–2.5).79 Historically the role of DXA to assess bone health and fracture risk in CKD3‐5D was controversial, as small cross‐sectional studies did not demonstrate that DXA discriminated prevalent fractures, and BMD measurements did not predict type of ROD. However, several recent longitudinal studies in patients across the spectrum of CKD and ESKD have demonstrated that low BMD at the hip and forearm do predict incident fractures.8083 These studies reported that the WHO T‐scores perform similarly in patients with and without CKD, with regard to fracture prediction, and resulted in the revised KDIGO recommendation to include BMD measurement in patients with CKD 3‐5D to assess fracture risk.

In the general population, estimates of fracture prediction are improved by the addition of clinical risk factors to BMD measurements. The Fracture Risk Assessment (FRAX) was developed to provide 10‐year absolute risk of major osteoporotic or hip fracture by combining 10 clinical risk factors, with or without femoral neck BMD, into a fracture risk algorithm.51 The clinical relevance of FRAX in CKD remains unclear. In a study from the Canadian Multicentre Osteoporosis Study, 320 individuals with an eGFR <60 mL/min, and 1787 with an eGFR ≥60 mL/min were followed for a mean of 4.8 years.81 This study showed that FRAX did not differ in its ability to predict major osteoporotic fractures, despite differences in underlying renal function. It is important to note that the incidence of fractures in the CKD patients was low and most patients did not have evidence of CKD‐MBD. In a cross‐sectional study of 353 patients with CKD (mean eGFR of 28 mL/min), approximately 30% had prevalent fractures; FRAX with femoral neck BMD discriminated those with and without fractures but was not superior to femoral neck BMD alone.84 All three study groups (FRAX alone, FRAX with BMD, BMD alone) were better discriminants of fracture than age alone. In a study of 485 Japanese hemodialysis patients, FRAX did not predict increased fracture risk over a 3.3‐year median follow‐up.80 It is important to note the relatively short follow‐up of this study, along with concerns about the accuracy of the fracture risk assessment data. In a recent study of 718 hemodialysis patients who were followed for a period of 2 years, the area under the curve (AUC) for FRAX was 0.76 (95% confidence interval [CI] 0.69–0.82) for major fractures and 0.70 (95% CI 0.69–0.84) for hip fractures, and FRAX discriminated fractures better than individual elements in the FRAX algorithm, although this did not include BMD.85 These studies suggest that more research is needed to determine the usefulness of FRAX in CKD and ESKD. More specifically, some of the clinical factors included in current FRAX algorithms may not be relevant in these patients and CKD‐specific fracture assessment tools need to be developed. These should incorporate predictors of fracture specific to kidney disease, for example, bone alkaline phosphatase (BSAP) or PTH.86

Assessment of Bone Microarchitecture—High‐Resolution Imaging

An important limitation of DXA is that it does not assess the 3‐dimensional structure of bone. Therefore, high‐resolution imaging methods such as QCT, HR‐pQCT, and micro‐MRI have been developed to provide 3‐dimensional imaging of bone density and microarchitectural aspects of bone quality, including cortical and trabecular volumetric BMD, geometry, microarchitecture, and strength. The TBS, although not a high‐resolution imaging modality, has also emerged as an important tool in the assessment of bone microarchitecture and will be discussed below. Micro‐MRI assessment of bone microarchitecture has been evaluated in general and kidney disease cohorts; however, studies of fracture discrimination are lacking, as such no further discussion has been included in this review.8793

Conventional QCT has a resolution of 300 μm3 and quantifies volumetric BMD and cortical and trabecular geometry. In CKD and ESKD, studies utilizing QCT have shown that cortical deficits predominate and these could discriminate and predict future fractures.9496 HR‐pQCT has a higher nominal resolution (60 to 82 μm3), which allows for quantification of trabecular number, thickness, and separation (Fig. 2). Finite element analysis (FEA) has been used in biomechanics to determine the mechanical behavior (and therefore strength) of bone.97 The advent of high‐resolution imaging has allowed FEA to be used in the assessment of bone strength, stiffness, and failure load, either in its entirety or in individual (cortical or trabecular) compartments.98, 99 Recent developments in HR‐pQCT processing methods have been developed to characterize cortical porosity and trabecular rod and plate structure, which have been strongly associated with bone strength.100102

HR‐pQCT provides detailed images of bone microarchitecture at the radius (left) and tibia (right). Scout view (A) reference line position (solid line) and the measurement site (dotted line). Images from healthy, postmenopausal white female (B). Images from female with CKD, no fractures (C). Images from female with CKD and prevalent fractures (D). Reprinted with permission from Nickolas et al.103

The ability of HR‐pQCT analysis of bone microarchitecture at the distal radius and tibia to assess fracture risk and bone quality has been evaluated in kidney disease. Specifically, HR‐pQCT was reported to discriminate and predict fractures, define abnormalities in bone quality that adversely affect bone strength, and identify microstructural abnormalities that account for reduced bone density as measured by DXA.103108 Measures from DXA and HR‐pQCT were associated with prevalent fractures in patients with CKD.103, 104 Patients with fractures had lower BMD by DXA and lower cortical and trabecular volumetric BMD and thinner cortices and trabeculae by HR‐pQCT. These abnormalities were more severe with longer duration and severity of CKD. A study of 211 men and women with CKD 3–5 assessed the ability of areal BMD by DXA and volumetric BMD by HR‐pQCT to discriminate fractures.105 Both tests discriminated fracture status and the addition of HR‐pQCT measures to BMD by DXA did not improve discrimination. In a study of 74 prevalent hemodialysis patients, those with fractures had reduced cortical and trabecular microarchitecture compared with those without fractures.106 Changes to bone microarchitecture and calculated bone strength were assessed in 33 ESKD patients and 33 age‐matched controls.107 Cortical and trabecular bone microarchitecture and calculated bone strength were altered in ESKD patients; these changes were more pronounced in females. HR‐pQCT has also been used to determine the microarchitectural mechanisms of bone loss in CKD. In a longitudinal study of 54 patients with CKD 2‐5D, the mean annualized loss of bone density by DXA at the radius was 2.9%.108 With HR‐pQCT, this was characterized by loss of cortical area, density, and thickness and a significant increase in cortical porosity.

The use of HR‐pQCT in the clinic is not currently practical because of cost and limited availability. Thus, methods to assess bone microarchitecture that can be implanted in the clinic have high potential to assist with the diagnosis and management of patients with metabolic bone diseases. TBS was developed to assess trabecular microarchitecture (ie, bone quality) by using software analysis that measures grayscale homogeneity from lumbar DXA images.109 In early studies, TBS has been shown to correlate with trabecular microarchitecture as measured by micro‐computed tomography, HR‐pQCT, and iliac crest bone biopsy.110112 TBS was also shown to predict fractures independently of clinical risk factors and areal BMD by DXA113115 and has been incorporated into international fracture risk guidelines and predictive algorithms such as FRAX.116 TBS has also been investigated in patients with kidney disease. In a recent study of 50 patients with CKD 3‐5D, TBS was associated with trabecular bone volume, width, and spacing, as well as cortical width as measured by bone biopsy.117 Luckman and colleagues112 reported that TBS measures correlated with both cortical and trabecular microarchitecture by HR‐pQCT in ESKD patients before kidney transplantation and that changes in TBS measurements reflected changes in trabecular microarchitecture and failure load but not cortical microarchitecture 12 months after transplantation. In a Canadian cohort of 1476 patients, those with an eGFR <60 mL/min and a TBS value below the median (<1.277) had a higher 5‐year fracture probability that was independent of BMD and clinical risk factors for fracture (hazard ratio [HR] = 1.62 per SD decrease in TBS).118 Further verification and qualification of TBS as a fracture prediction tool in patients with kidney disease is needed.

Bone Turnover Markers

In patients without CKD, BTMs can be used to assess fracture risk and monitor osteoporosis therapy. In patients with CKD and ESKD, they provide mild to moderate accuracy in the noninvasive assessment of bone turnover and mineralization. In some cases, BTMs can be used instead of bone biopsy to inform antifracture strategies, although the utility of BTMs to improve patient‐related outcomes such as fractures remains unproven. Markers of bone formation (osteoblast function) include BSAP, osteocalcin, and P1NP. Bone resorption markers (osteoclast function) include TRAP‐5b and CTX.

In kidney disease, PTH and BSAP are the most widely tested BTMs to assess bone turnover, and BSAP, vitamin D, calcium, and phosphorus are used to assess mineralization.810, 67, 119 Generally, extremes of PTH and BSAP identify bone turnover type based on bone biopsy finding in CKD and ESKD. In a study of 132 patients with CKD 3–5, plasma PTH measurements effectively distinguished patients with and without low bone turnover, AUC 0.96 for CKD stages 3 and 4, and 0.86 for CKD stage 5.119 In a study of 141 hemodialysis patients, an intact PTH < 420 pg/mL increased the positive predictive value (PPV) for low bone turnover from 74% to 90% in white patients, whereas an intact PTH < 340 pg/mL increased the PPV for low bone turnover from 48% to 90% in black patients.120 In 492 ESKD patients, the AUC for discriminating low versus non‐low bone turnover was 0.701 for PTH, 0.757 for BSAP, and 0.718 for PTH and BSAP in combination.8 In another study of patients with ESKD, PTH values within the middle range (150 to 300 pg/mL) less reliably identified underlying histology.121 In age‐related osteoporosis, BSAP reflects bone formation and correlates with bone histology and other BTMs.122124 In a study of 42 ESKD patients, BSAP and PTH were compared with bone biopsy.122 BSAP levels were higher in patients with high compared with low turnover (66.9 versus 10.8 ng/mL, p < 0.0005), were more strongly correlated with bone formation than PTH, and levels > 20 ng/mL reliably excluded adynamic bone disease. Total ALP has sometimes been used as a surrogate marker of BSAP, particularly in the absence of liver disease. However, ALP has high biological variation (>20%), and a high BSAP can still result in a normal ALP level.125, 126

PTH and BTMs may also have clinical utility in predicting bone loss and fractures. In a cross‐sectional study of 82 CKD patients, higher levels of PTH and BTMs were associated with lower cortical and trabecular density and increased cortical and trabecular thinning.104 In the same study, higher levels of PINP, osteocalcin, and TRAP discriminated fracture. In a prospective 2‐year study of 89 hemodialysis patients, baseline BSAP was strongly associated with loss of BMD cortical mass and volume.127 In prospective studies of patients with CKD before (n = 52) and after renal transplantation (n = 47), the microarchitectural and biochemical mechanisms of bone loss were examined by BTMs, DXA, and HR‐pQCT.108, 128 Higher levels of PTH, BSAP, osteocalcin, PINP, TRAP, and CTX predicted the loss of cortical area, density, and thickness, increase in cortical porosity, and decreased bone strength. The ability of PTH and BTMs to predict fractures was assessed in a prospective study of 485 ESKD patients. Incident fracture was associated with PTH levels either <150 pg/mL or >300 pg/mL compared with 150 to 300 pg/mL (p < 0.01).80 In the same study, BSAP was a very useful surrogate marker for any type of incident fracture risk (AUC = 0.766, p < 0.0001).

These data suggest that PTH and BTMs may have greater clinical utility in assessing bone quality and fracture risk in CKD and ESKD, in particular when used in conjunction with bone imaging methods. For example, fracture risk can be estimated by DXA. However, deciding on a treatment (vitamin D, calcimimetic, antiresorptive, anabolic) will also require consideration of bone turnover. PTH and BTMs can potentially be used to inform which pharmacologic agent is most appropriate. However, before BTMs can be widely used to manage renal bone disease, significant barriers need to be overcome. BSAP lacks a readily available automated assay, and there are concerns of cross‐reactivity with the liver iso‐enzyme,129 there are no validated reference ranges of BTMs in patients with CKD and ESKD, many BTMs are cleared by the kidney (monomeric P1NP, osteocalcin, and CTX) and BTMs have high intra‐ and interassay variability and biological variability.130

Treatment of ROD and Preventing Fractures in CKD

Management of kidney‐associated bone disease for fracture risk reduction is controversial. First, until the release of the 2017 KDIGO guidelines, fracture risk classification in CKD3‐5D was not recommended. Second, there are no antifracture treatments that have been developed specifically for patients with CKD‐MBD. However, since emerging data and anecdotal experience with existing antifracture agents suggest that they are safe, we expect that these agents will be more widely used in patients with CKD 3‐5D, especially as more patients undergo DXA screening. In this section, we will briefly review the antifracture treatments that are in current clinical use and their potential application to patients with kidney disease.

Management of CKD‐MBD

Management of the abnormalities associated with CKD‐MBD must occur before initiating specific antifracture therapy in patients with CKD 3‐5D. A detailed discussion of this is comprehensively articulated in the updated KIDGO guidelines and associated publications.10, 53, 131 In brief, supplementing with vitamin D (nutritional and/or active), lowering phosphate, initiating calcimimetics, and deciding on the need for parathyroidectomy are critically important to treating kidney‐bone disease and can have some antifracture benefits.132137 In our opinion, after optimized management of CKD‐MBD, one should consider additive treatment with an agent demonstrated to have antifracture efficacy in the general population (Fig. 3). The updated 2017 KDIGO CKD‐MBD Guidelines clearly endorse the use specific antifracture therapies in CKD 1–2 and to a lesser extent in CKD stage 3 in the absence of abnormalities of CKD‐MBD (recommendation 4.3.1 and 4.3.2).10 However, In patients with CKD 3–5 and evidence of CKD‐MBD, the use of osteoporosis therapies is not directly addressed (recommendation 4.3.3): “In patients with CKD G3a–G5D with biochemical abnormalities of CKD‐MBD and low BMD and/or fragility fractures, we suggest that treatment choices take into account the magnitude and reversibility of the biochemical abnormalities and the progression of CKD, with consideration of a bone biopsy (2D).” It is noteworthy that the KDIGO update no longer mandates that a bone biopsy should be obtained before starting osteoporosis treatment, in part because of the increasing experience with the use of osteoporosis medications in CKD patients.

Algorithm for fracture risk screening and initiation of antifracture strategies in patients with CKD. DXA = dual‐energy X‐ray absorptiometry; CKD‐MBD = chronic kidney disease‐mineral and bone disorder; PTH = parathyroid hormone; BSAP = bone‐specific alkaline phosphatase; LLN = lower limit of the normal reference range; ULN = upper limit of the normal reference range. Lifestyle factors include weight‐bearing exercise, cessation of smoking, adequate nutrition, moderate alcohol intake, and fall prevention strategies. Management of CKD‐MBD includes phosphate lowering, vitamin D supplementation (nutritional and active), calcimimetics, and parathyroidectomy. Anabolic agents include teriparatide and abaloparatide. Antiresorptive agents include bisphosphonate and denosumab.

Antiresorptive Agents

Bisphosphonates and denosumab are commonly used in the treatment of senile and glucocorticoid‐induced osteoporosis; these agents lower remodeling rates and may be helpful in preventing bone loss and fracture in normal and high‐turnover bone disease. These should be avoided in patients with adynamic bone disease, but to date there are no studies in CKD that definitively demonstrate that bisphosphonates cause adynamic bone disease. There are no primary safety and efficacy data on the use of antiresorptive therapies in patients with CKD‐MBD, but there are post hoc analyses of the registration trials in patients with mild to moderate CKD (without biochemical manifestations of CKD‐MBD). These agents can therefore be used in CKD patients who have similar characteristics to the inclusion criteria of the registration trials, but in CKD 3‐5D patients with evidence of CKD‐MBD, further evaluation of the underlying ROD type needs to be undertaken before the use of these agents for fracture prevention.


Bisphosphonates are selectively taken by osteoclasts, where they inhibit farnesyl pyrophosphate synthase and the synthesis of isoprenoid compounds, which are essential for osteoclast activity, thereby reducing osteoclast‐mediated bone resorption. Oral bisphosphonates have poor bioavailability (<1%); around 50% of the drug is not taken up by osteoclasts and cleared by the kidney. The actual amount of bisphosphonate retained in bone depends on the remodeling space, the GFR, and bone turnover rate.52 Given the lack of clinical trial data, oral bisphosphonates are not recommended in patients with eGFR < 30 mL/min because of concerns about excessive accumulation in bone and long‐term suppression of bone remodeling.

However, data from recent fracture intervention trials suggest that these drugs can be safely used in patients with reduced eGFR. In a post hoc analysis of 9 randomized, double‐blind trials comparing risedronate to placebo, 8996 females were identified as having kidney impairment, with the majority having a creatinine clearance of 30 to 80 mL/min.138 Risedronate effectively preserved BMD and reduced the incidence of vertebral fractures, and overall and renal function‐related adverse effects were similar in both groups and independent of renal function. Bone biopsy data of 57 patients with renal impairment showed no evidence of adynamic bone disease or mineralization defects. In a secondary analysis of the Fracture Intervention Trial, 581 women (9.9%) had a GFR < 45 mL/min.139 Women with a reduced eGFR had a 5.6% increase in total hip BMD, and alendronate increased BMD regardless of eGFR. Treatment with alendronate reduced clinical and spine fractures to a similar degree in those with and without renal impairment. There were no differences in adverse outcomes by renal function. A recent systematic review by Wilson and colleagues140 compared bisphosphonate to placebo in six studies (n = 1013) of patients with CKD (four of these were in renal transplant recipients).139, 141145 Given the heterogeneity of the studies included, the use of bisphosphonates was associated with a nonsignificant reduction in fracture risk and moderate evidence that bisphosphonates may reduce loss of lumbar spine BMD but not femoral neck BMD. A recent post hoc analysis of three Japanese RCTs146148 compared risedronate to placebo in 852 patients with an eGFR from 30 to >90 mL/min (with 228 having an eGFR <60).149 Patients were analyzed according to eGFR level ≥90, 60 to <90, and ≥30 to <60 mL/min/1.73 m2, and lumbar BMD, BTMs (CTX, P1NP, and BSAP) were evaluated at 48 weeks. The increase in lumbar BMD (p < 0.001) and inhibition of BTMs (p < 0.001) with the use of risedronate did not differ in the eGFR subgroups. There are no data on the optimal duration of bisphosphonate therapy in CKD or ESKD. General population data suggest 3 to 5 years of therapy, with a lack of evidence on fracture reduction beyond 5 years and concerns about atypical femoral fractures.150152


Denosumab is a humanized monoclonal antibody against the Receptor activator of nuclear factor kappa‐Β ligand (RANKL), reducing bone turnover by inhibiting osteoclast proliferation and development. The Fracture Reduction Evaluation of Denosumab in Osteoporosis Every 6 Months (FREEDOM) Trial recruited 7808 postmenopausal women and demonstrated that denosumab administered twice yearly for 36 months was associated with a reduction in the risk of vertebral (68%), nonvertebral (20%), and hip fractures (40%) compared with placebo.153 Denosumab is not cleared by the kidney, so unlike bisphosphonates, there is no risk of excess drug accumulation. Its effectiveness and safety in CKD (without evidence of CKD‐MBD) were assessed in a secondary analysis of patients from the FREEDOM trial, where treatment effect was compared across CKD categories (CKD 4 = 73, CKD 3 = 2817, CKD 2 = 4069).154 There was no difference in treatment efficacy and adverse effects by renal function, and denosumab increased BMD at the spine and hip and showed a 62% reduction in incident vertebral fractures.

The secondary analysis of the FREEDOM trial did not describe an association between hypocalcemia and decreased renal function. However, case reports and anecdotal experience suggest that this may occur in patients with CKD, hyperparathyroidism, and low vitamin D. In a case series of 8 patients with ESKD and 6 patients with CKD treated with denosumab, 8 patients developed severe hypocalcemia.155 In two studies of single‐dose denosumab in CKD and ESKD, hypocalcemia was the most common side effect; however, appropriate pretreatment with calcium and calcitriol was protective of clinically significant hypocalcemia.156, 157 These and other clinical data suggest that denosumab may have a role in patients with CKD 3‐5D, with appropriate calcium, vitamin D supplementation, and monitoring for hypocalcemia. Denosumab is metabolized in the reticuloendothelial system, and its clinical effect wanes after 6 months; frequency and duration of treatment in CKD 3‐5D remain unclear.

The increasing off‐label use of antiresorptive therapies in kidney disease patients reflects not only the decreasing concern regarding their short‐term safety but also the lack of specific antifracture treatments for patients with CKD‐MBD. No doubt further observational data will continue to guide and refine clinical practice, but ultimately prospective trials are needed to better define safety and antifracture efficacy of these drugs in patients with CKD‐MBD and an eGFR of <30 mL/min/1.73m2.

Osteoanabolic Agents

Osteoanabolic agents are forms of recombinant PTH, and their use in CKD remains controversial. In CKD, high‐baseline PTH levels promote cortical bone loss; therefore, these agents should not be used in patients with high‐turnover bone disease. However, they may increase bone turnover and bone density in patients with adynamic bone disease.

Teriparatide is a recombinant peptide of the first 34 amino acids of human PTH. In health, its antifracture efficacy is caused by an increase in osteoblast number, leading to increased bone formation and thickening of trabecular and cortical bone.158 Its efficacy in primary osteoporosis was demonstrated in two large clinical trials.159, 160 A post hoc analysis of Fracture Prevention Trial examined the safety and efficacy of teriparatide in 736 women with CKD (majority GFR 50 to 79 mL/min, and none <30 mL/min).161 Compared with placebo, teriparatide significantly increased lumbar spine and femoral neck BMD and had similar efficacy in preventing vertebral and nonvertebral fractures across all stages of CKD. The main adverse effects were hypercalcemia and hyperuricemia and these were more common with reduced eGFR. In a separate post hoc analysis from the same trial, improvements in BMD were correlated with improvements in trabecular microarchitecture.162 Teriparatide has also been used in patients with adynamic bone disease, where it improved BMD at the lumbar spine but not the femoral neck.163 A prospective, 48‐week study examined the effect of once‐weekly teriparatide in 22 hemodialysis patients with low BMD and hypoparathyroidism.164 Over the treatment period, there was an increase in lumbar spine but not femoral neck BMD. The observed increase in BSAP strongly correlated with improvements in BMD. There was no clinically significant hypercalcemia, and hypotension was the most common adverse event.

Abaloparatide is a newer parathyroid hormone‐related protein (PTHrP) analog, with stronger affinity for the transient state of the PTH1/PTH receptor and a more anabolic profile (compared with teriparatide) given less bone resorption and hypercalcemia.165, 166 In a study of 222 postmenopausal women, daily abaloparatide (20, 40, or 60 μg) was compared with teriparatide 20 μg or placebo.167 BMD was increased at the lumbar spine, femoral neck in a dose‐dependent manner, and to a greater extent than with teriparatide. Abaloparatide was compared with placebo and open‐label teriparatide in a study of 2461 postmenopausal women.168 New vertebral fractures occurred less frequently in the active treatment groups, and BMD increases were greater with abaloparatide than placebo. The incidence of hypercalcemia was lower with abaloparatide compared with teriparatide. In a secondary analysis of changes on bone histomorphometry, there was no evidence of abnormal mineralization, bone marrow abnormalities, or presence of excess osteoid.169 Patients treated with abaloparatide had lower eroded surface on histomorphometry versus placebo but similar increase in cortical porosity compared with teriparatide. This is supported by a smaller increase in CTX (a resorption marker) with abaloparatide compared with teriparatide. Based on the available data, abaloparatide has the ability to increase bone mass and formation, with less bone resorption and hypercalcemia. It has the potential to become an ideal agent for the treatment of patients with CKD and low to normal bone turnover; however, that data is currently lacking.

Sclerostin is a protein encoded by the SOST gene and secreted by osteocytes. Loss‐of‐function SOST mutations result in a high bone mass phenotype through increased bone formation, and sclerostin has predominantly anti‐anabolic effects on bone through inhibition of the Wnt‐signaling pathway. Inhibition of sclerostin is therefore of great interest in patients with CKD 3‐5D, particularly those with low bone mass and low bone turnover where antiresorptive therapies are contraindicated. In trials of postmenopausal women with osteoporosis, romosozumab increased BMD to a greater extent than existing anabolic agents and decreased vertebral and nonvertebral fractures.170172 In another study, comparing romosozumab to teriparatide, there was a greater increase in cortical BMD (compared with trabecular BMD) in the romosozumab arm compared with a reduction in cortical BMD in the teriparatide group.173 Increased cardiovascular events were reported in one (but no other) romosozumab study.171 As such, it remains unclear whether romosozumab increases cardiac risk, but given the large CVD burden in patients with kidney disease, further study of romosozumab in CKD and ESKD should be suspended until these issues are clarified in future studies. Inhibition of DKK1 is another target for potential novel anabolic agents, although the clinical development of these drugs lags behind that of sclerostin inhibitors. In animal models, inhibition of DKK1 by monoclonal antibodies (DKK1‐ab) generally increased bone formation and mass. In a mouse model, changes of CKD‐MBD including ROD were ameliorated after the administration of DKK1‐ab (in combination with phosphate binder therapy).174 In patients with multiple myeloma, administration of DKK1‐ab reduced bone resorption and reduced bone formation.34 Finally, the dual inhibition of sclerostin and DKK‐1 leads to synergistic bone formation in rodents and non‐human primates.175 These studies have important implications for patients with kidney disease, and clinical studies of DKK1‐ab are needed in CKD and ESKD cohorts.

Take‐Home Messages and Rethinking Bone Disease in CKD

In CKD patients, fracture rates are more than 10‐fold higher compared with age‐ and sex‐matched individuals without CKD. Although fracture incidence in the general population has fallen over the last two decades, it has increased in patients with ESKD. In the clinic, treatment of CKD‐MBD is focused on the correction of abnormalities associated with parathyroid hormone and phosphate, a strategy that has not mitigated the effects of CKD on fracture risk. Despite advances that have improved our ability to assess fracture risk in the clinic, we continue to lack noninvasive tools that predict turnover and diagnose ROD type, which would inform fracture prevention strategies. Furthermore, despite an ever‐increasing choice of antifracture treatments in the general population, we lack data on their safety and efficacy in CKD 3‐5D. Therefore, it is time to rethink bone disease in patients with kidney disease.

Kidney‐related bone disease is complex and multifaceted, as such any gains are likely to be incremental rather than revolutionary. A starting point could be implementation of the successful fracture screening, prevention, and treatment programs used in the general population to CKD 3‐5D. The 2017 KDIGO guidelines justify the use of DXA, a readily available and inexpensive fracture risk screening tool. Lifestyle measures such as smoking cessation, weight‐bearing exercise, alcohol moderation, and improved nutrition all have proven antifracture efficacy in the general population and could be argued as being even more important in CKD. Vitamin D deficiency is common in CKD and should be routinely supplemented given its skeletal benefits and low risk of adverse consequences. The role of calcium supplementation is less well defined in CKD and ESKD and cannot be recommended in these patients.

Our ability to initiate specific, mostly antiresorptive osteoporosis treatments in CKD 3‐5D remains limited by our inability to effectively assess bone turnover and a historical fear of propagating low turnover and atypical femoral fractures. In the clinic setting, the interpretation and treatment of the current markers of CKD‐MBD should not occur in isolation but in view of the broader endocrine and skeletal disturbances. Although imperfect, PTH and BSAP improve our ability to discriminate bone turnover and some of the microstructural derangements. These should be used with DXA and be further studied with novel algorithms such as TBS that can inform on bone microarchitecture, with only minimal modifications to existing imaging infrastructure. Finally, fracture prediction algorithms such as FRAX should be broadly utilized to increase awareness of fractures; however, these need further validation and refinement to improve their accuracy in patients with kidney disease. Many, if not all, of these measures are feasible and could be readily implemented in most developed countries.

Ongoing work in improving diagnosis of ROD type is needed. Bone biopsy will remain the gold standard; however, given the required expertise, its role will always be confined to specialized tertiary institutions. High‐resolution imaging techniques provide a glimpse of the future, where imaging could obviate the need for a physical biopsy. However, today, even the most advanced high‐resolution imaging techniques remain inadequate at evaluating bone turnover and mineralization, both of which remain essential to informing treatment decisions. We have no doubt that high‐resolution imaging will continue to evolve and there are countless examples in medicine where technology has diminished the need for invasive diagnostic tests but seldom have these been superseded. Therefore, we believe that bone biopsy will remain an important diagnostic and research tool, such as in validating studies of high‐resolution imaging techniques. As such, its availability should be fostered and rationally coordinated across regions and health care networks.

CKD patients are historically disadvantaged by their exclusion from large general population clinical trials, and this is clearly the case in osteoporosis. Much of the current data come from secondary analyses of large osteoporosis trials, where patients were excluded if they had evidence of CKD‐MBD. The safety and efficacy of existing and novel agents in CKD 3‐5D remain unclear, and studies of these drugs are urgently needed in these cohorts. Future studies also need to target patients at various stages in the evolution of renal bone disease; for example, prevention of bone loss in CKD, reduction of fractures in ESKD, as well as any effects on the systemic aspects of CKD‐MBD. For this to occur, stronger advocacy from kidney societies and a paradigm shift from drug companies around the world are required. As physicians, we need to recognize the burden of fractures and bone disease in our patients and work tirelessly to promote better awareness and utilization of available diagnostic and therapeutic resources, such as greater off‐label use of osteoporosis medications in CKD 3‐5D. To this end, collaborations with our endocrinology colleagues should be fostered in both the research and clinical settings. In the words of Lao‐Tzu, “A journey of a thousand miles begins with a single step.” No doubt the road ahead is long, harboring many challenges and false starts; however, to accept the status quo is simply not an option.

Effective weight control via an implanted self-powered vagus nerve stimulation device


In vivo vagus nerve stimulation holds great promise in regulating food intake for obesity treatment. Here we present an implanted vagus nerve stimulation system that is battery-free and spontaneously responsive to stomach movement. The vagus nerve stimulation system comprises a flexible and biocompatible nanogenerator that is attached on the surface of stomach. It generates biphasic electric pulses in responsive to the peristalsis of stomach. The electric signals generated by this device can stimulate the vagal afferent fibers to reduce food intake and achieve weight control. This strategy is successfully demonstrated on rat models. Within 100 days, the average body weight is controlled at 350 g, 38% less than the control groups. This work correlates nerve stimulation with targeted organ functionality through a smart, self-responsive system, and demonstrated highly effective weight control. This work also provides a concept in therapeutic technology using artificial nerve signal generated from coordinated body activities.


Obesity resulted from ingesting calories in excess of normal biological requirement is a major risk for a large number of chronic diseases, including cardiovascular disease1,2, diabetes mellitus1,3, chronic kidney disease1, gallbladder diseases4, certain cancers3,5, musculoskeletal disorders6, and even genetic variation7. Treatment of obesity imposes an enormous economic burden on the global healthcare system8,9,10. According to a recent global survey, over 710 million people worldwide, including 107.7 million children and 603.7 million adults, are plagued by obesity problems, and about 4 million people died of overweight- or obesity-related diseases in 201510. Common approaches for treating obesity include non-surgical and surgical treatments. Daily physical exercise and taking weight-loss drugs are common non-surgical weight reduction regimens, but there is a high potential of weight rebound or side effects from drugs11. Current bariatric surgical procedures such as gastric bypass, biliopancreatic diversion, and sleeve gastrectomy have demonstrated a significant impact on weight loss, but these procedures are invasive with the potential of serious complications12,13,14. The rising healthcare standards demand new obesity treatment strategies that are effective, easy to operate, and have less side effects.

Neuromodulation, as a non-destructive and reversible therapeutic strategy, can manipulate body functions by stimulating or influencing neurophysiological signals through the neural networks to achieve therapeutic purpose15,16. It has been known for a century that the vagus nerve (tenth cranial), a mixed parasympathetic nerve containing both afferent and efferent nerve fibers, acts as a signal bridge to transport information between the brain (the center of the nervous system) and the body (head, neck, thorax, and abdomen)17,18,19. Recent breakthroughs in neuromodulation for body weight control have provided potential opportunities for therapeutic interventions and brought renewed promises and vitality to the development of new anti-obesity strategies. A number of studies have demonstrated that pulsed electrical stimulations on vagus nerve could induce multiple physiologic functions related to food intake, energy metabolism, and glycemic control, which can result in appreciable weight loss20,21,22. An implantable vagus nerve stimulation (VNS) device for weight control was recently approved by Food and Drug Administration and commercialized20,23. Major concerns of current electrical stimulation are potential compensation mechanisms that blunt physiological responses14 and vicinity tissue damage that induces adverse effects17,24,25. In addition, the electrical system is bulky and complicated in operation. All the electrical stimulations need to be programed externally and the device needs to be charged periodically13,26. How to achieve real-time-responsive and self-sustainable stimulation remains a major challenge for this promising weight control strategy.

In this work, we present a correlated VNS system that is battery free and automatically generates electrical stimulations in correlation to stomach movement. A flexible nanogenerator device is developed to be attached to the stomach surface and produce biphasic electrical pulses in response to the peristalsis of stomach. The electric signals can stimulate the vagal afferent fibers to reduce food intake and eventually achieve weight control. We successfully demonstrated this strategy on rats and achieved 38% weight loss in as short as 15 days without further rebound, exceeding all current electrical stimulation approaches. This work provided an effective weight control strategy that is self-responsive, battery free, and directly correlating food intake to stomach movements.


Development and working principle of the VNS device

The correlated VNS system for weight control is designed following the principle depicted in Fig. 1a. The stomach motion is used as the sole source to generate pulsed voltage signals, which in response will stimulate the vagus nerves to reduce food intake. This self-responsive function is enabled by a triboelectric nanogenerator (TENG)27,28,29,30,31 attached on the surface of stomach, which generates biphasic electric pulses when the stomach is in peristalsis. Here, a bilateral VNS is implemented by wrapping the two gold (Au) leads around the anterior vagus nerves (AVNs) and posterior vagus nerves (PVNs) at the proximity of the gastroesophageal junction (Fig. 1b). The AVNs and PVNs were ~6 mm apart and could be clearly observed via multiple staining images (Fig. 1c, Supplementary Figure 1). Connecting at this position could provide a focused stimulation to the small unmyelinated C fibers and avoid stimulating fibers that join the trunk from the heart and lungs32.

Fig. 1
Fig. 1

The correlated vagus nerve stimulation (VNS) system and its biocompatibility. a Operation principle of the correlated VNS system schematically showing the pathway for biphasic electric signal generation and VNS. b An implanted VNS device with Au leads being connected to anterior and posterior vagal trunks. c Hematoxylin–eosin (H&E) staining of the implanted tissues (transverse section). Areas within the blue and red boxes are enlarged view of the anterior (red) and posterior (blue) vagus nerves (scale bar = 100 μm). d A packaged VNS device. e, f Fluorescence images of stained 3T3 cells that were cultured on a regular cell culture dish (e) and on the surface of a packaged device (f). g Comparison of normalized cell viability for 4 days showing excellent biocompatibility of the packaged device (n = 3 for each group). All data in g are presented as mean ± s.d.

To ensure the mechanical robustness and flexibility of implanted devices and to avoid potential erosion in the physiological environment, the entire VNS device was packaged by a multilayer film composed of polyimide, polydimethylsiloxane (PDMS), and ecoflex. Au leads were connected to the tips of Cu wire electrodes and partially exposed for electrical signal transmission (Fig. 1d, fabrication details are included in Supplementary Figure 2a and b, which was described in Supplementary Note 1). The TENG was able to generate reasonably high voltage and current output under normal contact-separation motions, with an optimal output power of ~40 µW at an external load of 20MΩ (Supplementary Figure 2c, d and e). Based on the impedance of the vagus nerve, the stimulation voltage was found to be around 200 mV (Supplementary Figure 2d). Similar outputs were obtained from various displacement motions, suggesting that the TENG was able to respond to complex stomach motions (Supplementary Figure 3). To confirm the biocompatibility of the packaged VNS device, mouse fibroblast 3T3 cells were cultured on the encapsulated device surface and in a reference cultural dish for 4 days to examine and compare the cell attachment, proliferation, and morphology. Cells in both media exhibited similar density and equivalent morphology. No dead or distorted cells were observed from the encapsulation material surface (Supplementary Figure 4). The fluorescent staining results showed that the 3T3 cells can spread and form intact cytoarchitecture in both groups (Fig. 1e, f). In addition, 3-{4,5-dimethylthiazol-2-thiazolyl}−2,5-diphenyl-2H-tetrazolium bromide (MTT) assay revealed that the relative viability of 3T3 cells on encapsulation material was more than 98% within 4 days, comparable to the cells cultured in the culture dish (Fig. 1g). These results confirmed that the encapsulated device is non-cytotoxic and biocompatible.

When the stomach is under peristalsis stomach33,34, the corresponding motion cycle of the triboelectric layers in the VNS device is depicted in Fig. 2a (i)–(iv). As the stomach is distended, the two triboelectric layers are pushed into contact, where oppositely charged surfaces are created based on their different electron affinity (stage i). The following contraction of the stomach pulls the bottom electrode layer (BEL) layer away from the polymer layer, and thus drives electrons flowing from the top electrode layer (TEL) electrode to the BEL through the two connections with the vagus nerve (stage ii). When the stomach is fully contracted, the triboelectric layers are fully separated, where maximum charge transfer is reached and the net current through the nerve drops back to zero (stage iii). The BEL layer is then brought back toward the polymer layer in the following stomach distention (phase iv), resulting in an opposite current flow through the vagus nerve until the stomach reaches the original distended stage (i) again. The recorded voltage output within one cycle at the frequency of 0.05 Hz is shown in Fig. 2b and corresponding stages are marked along the curve. This contributes to the cyclic alternating electrical signals as the stomach continues peristalsis.

Fig. 2
Fig. 2

Working principle and voltage signal of the vagus nerve stimulation device. a Schematics of the working principle of VNS device under different stomach motion stages. b A typical single-cycle voltage biphasic signal corresponding to the four stages of stomach movement at a frequency of 0.05 Hz. c Voltage signal measured in PBS solution under different agitation frequency when the VNS device was connected to an external load with the same impedance of the implanted area. d Voltage signal measured when the VNS device was implanted and connected to vagus nerves. e Long-term stability test of the VNS device, where the device was removed from the rats and the voltage was measured on an external load of 0.3MΩ. f Electrophysiological signals recorded from rats without implantation and with an active implanted VNS device on the same day, and 7 and 15 days post implantation. g Enlarged view of one group of electrophysiological signal highlighted in the dotted box in f

To investigate how the implanted VNS device functioned in response to stomach movements, voltage signals were first measured between the two Au leads. The stomach was arbitrarily deformed by cyclically pressurizing it via a gavage needle at a series of frequencies of 0.05, 0.1, 0.5, 1.0, 1.5, and 2.0 Hz (Fig. 2c). Correspondingly, the voltage signals were also measured when the two Au leads were connected to the vagus nerve in a rat’s body (Fig. 2d and Supplementary Movie 1). Both voltage signals exhibited similar amplitudes ranging from 0.05 to 0.12 V. It should be noted that the recorded voltage was lower than the actual operational voltage due to the finite internal impedance of the measurement system (1MΩ). Higher voltage outputs were recorded from higher frequency. While the stomach deformations (i.e., pressure change) remained constant, higher voltage signal could be attributed to a higher displacement rate, suggesting faster stomach motion is favorable for more intense stimulation. The implanted devices were removed from the rats 1 day, 7 days, 15 days, 4 weeks, 8 weeks, and 12 weeks after implantation, and their voltage output was measured accordingly. All the devices showed a good structural integrity without any observable defects (Supplementary Figure 6). The nearly unchanged voltage amplitude confirmed good stability and durability of the VNS device in the biological environment (Fig. 2e).

Electrophysiological signals were measured from the cervical vagal trunk on the same day, and 7 and 15 days after implantation (Supplementary Figure 7). As shown in Fig. 2f, when there was no external stimulation, a regular electrophysiological signal from the vagus nerve can be detected. The voltage amplitude was ~0.5 mV with 11 electric pulses per signal group (Fig. 2g). When the VNS device was activated, the amplitude increased to ~0.8 mV, and the number of electric pulse per group increased to 18–19. This measurement clearly showed that the VNS device can stimulate vagus nerves effectively. Similar electrophysiological signals could be detected at different time points post implantation, which evidenced the vagus nerves were effectively stimulated by the VNS device during the implantation period. Electrophysiological signals were further measured in response to a range of stimulation voltage from 50 to 740 mV. The stimulated state of the vagus nerve was detected as the voltage from VNS device was above 100 mV. The signal intensity from the vagus nerve increased monotonically following the stimulation voltage (Supplementary Figure 8, Supplementary Note 2).

Biocompatibility and biosafety of implanted VNS device

Rats with the VNS device implanted on stomach and the Au leads connected to the vagus nerve were defined as the VNS-active group. Small animal computed tomography (CT) was used to produce three-dimensional (3D) x-ray images of representative rat models as a function of implantation time to investigate the implantation stability when the rat was under normal daily activity (Fig. 3a, Supplementary Figure 9, and Movie 2). The high contrast spot (Au is a good contrast agent for CT) inside the rat was the implanted VNS device. No position shifting was observed during the entire 12 weeks of implantation period. This high stability could be attributed to the good biocompatibility of the packaged VNS device, which was observed being completely imbedded possibly by omentum and tightly fixed to the stomach surface post study (Supplementary Figure 10). The right panel of Fig. 3a shows an enlarged CT image of the implantation area, where the Au leads and exposed tip (with much brighter contrast) can be clearly identified, wrapping around at the vagus nerve region. In contrast to the VNS group, the sham group had the same VNS device implanted on stomach but without Au leads connecting the Cu wire electrodes to the vagus nerve (the completely packaged Cu wires were still placed at the same vicinity of the vagus nerve, Fig. 3b). 3D CT images showed the same stable VNS implantation in the sham group. The enlarged image revealed the insulated leads exhibiting a uniformly low contrast. As a comparison, rats in the laparotomy (Lap) group, which rats were subjected to surgery but without the VNS device implantation, were also imaged and showed only the skeleton of the rats (Fig. 3c). The Sham, Lap, and Intact (rats without any operation) groups are defined as the control groups.

Fig. 3
Fig. 3

Computed tomography (CT) 3D projection images and hematology data. ac Serial CT images over time of the VNS group, Sham group, and Lap group, respectively. Schematics on the left show the setup of each group. A series of CT images (coronal and sagittal) show a representative rat for each group at different time points. The enlarged views of the implantation site are shown at the end. di Hematology results of all four groups over time (n = 3 for each group). d Blood glucose (GLU) levels. e Infection-related lymphocytes (LYM) levels. f Hematopoietic function-related red blood cell (RBC) levels. g Hepatological function-related alanine aminotransferase (ALT) levels. h Renal function-related creatinine (CRE) levels. i Electrolyte metabolism-related calcium (Ca) levels. j H&E stains of vital organs (heart, lung, liver, spleen, kidney, bowel, stomach, and esophagus) at different time points (1 day, 7 days and 15 days) post implantation. All data in di are presented as mean ± s.d.

Whole blood and chemical analysis were performed on the four groups of rats for biosafety assessment during the implantation period. Compared to the hematology data in intact group, the blood glucose (GLU) concentration (Fig. 3d) in VNS and Sham group decreased at week 2 due to reduced food intake after surgery, and eventually recovered to the normal levels. The indicators of infection such as lymphocytes (Fig. 3e), hematopoietic function such as red blood cell (RBC) (Fig. 3f) and hemoglobin (Supplementary Figure 11a), hepatological function such as alanine aminotransferase (ALT) (Fig. 3g) and albumin (Supplementary Figure 11b), renal function such as creatinine (Fig. 3h) and blood urea nitrogen (Supplementary Figure 11c), and electrolyte metabolism such as calcium (Ca) (Fig. 3i) and phosphorus (Supplementary Figure 11d) all remained steady during the entire implantation period. In general, all the blood testing results were within the normal range shortly after the device implantation and did not show any abnormality35,36, suggesting that the VNS device is highly hemocompatible. The comprehensive blood analyses, together with the imaging results, confirmed that implanting the VNS device on stomach surface did not cause any measurable adverse effect to the rats. All rats with the VNS implantation exhibited normal daily behaviors, which were not different from the intact groups (Supplementary Movie 3).

Pathological tests were conducted on most vital organs, including heart, lung, liver, spleen, kidney, bowel, stomach, and esophagus. Hematoxylin and eosin (H&E) staining were collected from these organs at different time points (1 day, 7 days, and 15 days) post implantation. All the organs showed no deformation and abnormal lymphatic cell invasion (Fig. 3j), which further confirmed that all the rats were in a good health condition, and the VNS device had no systemic side effects. Histological analysis of the vagus nerves 15 days after implantation showed no signs of nerve cell shape change or invasion of inflammation cells (Supplementary Figure 12). This revealed the vagus nerves were not damaged by connecting the VNS device.

Weight control by implanted VNS device

The weight control performance was first examined in the four groups of rats (VNS, Sham, Lap, and Intact) that were fed and grown under the same conditions. The average initial weight of rats was 250 g, and their body weight and food intake were monitored on a daily basis. After 100 days, the body size of the VNS group was significantly smaller than all three control groups (Fig. 4a). The recorded body weight and corresponding daily food intake over time are shown in Fig.  4b, c, respectively. Since the implantation surgery was conducted on the seventh day of this study, all four groups exhibited the same body growth trend and the same amount of food consumption over the first week, indicating that all rats were under the identical growth conditions and their results were comparable. Immediately after the surgery, the VNS and Sham groups (both had VNS implanted) exhibited an obvious weight loss. Accordingly, their food intake was also largely reduced likely due to the VNS device attachment. The Lap group, which had the same surgical procedure/opening, did not show any abnormity in weight change or food intake when compared to the intact group, suggesting that surgery itself had minimal impact on weight control. As the rat’s body adapted to the VNS device implantation, the amount of food intake of the Sham group quickly recovered to the same level as the other two control groups after ~15 days of implantation. As a result, the average body weight of the Sham group bounced back after the initial 2–3 days, and increased following the same trend of the other two control groups (Lap and Intact). After 60 days, all three control groups exhibited a very close body weight of 535 ± 18 g (Sham), 538 ± 32 g (Lap), and 538 ± 32 g (Intact) (n = 6 for each control group), confirming that neither surgery nor simple stomach attachment had any effect on weight control. On the contrary, although the food intake of the VNS group recovered as well after the initial reduction and reached a steady level after ~15 days, the daily consumption of food was only ~2/3 of those consumed by the other control groups. Therefore, the average body weight of the VNS group exhibited a much slower growth rate. It reached a steady value of 350 ± 23 g (n = 6), about 63% of the other three control groups. Box plots were implemented to provide a statistic analysis of the final body weight (on the day of sacrifice, Fig. 4d) and food intake (on the last day before euthanasia, Fig. 4e). The differences between the VNS group and three control groups were statistically significant (P < 0.001) for both body weight and food intake, while the differences between all three control groups were not (P > 0.2). Such comparison clearly revealed that spontaneous nerve stimulation by the implanted VNS device had obvious impact on weight control.

Fig. 4
Fig. 4

Weight control during the weight-gaining growth stage of rats. a Representative images of body size of the VNS group (n = 6) and the control groups (Sham, Lap, and Intact group, n = 6 for each group). b Average body weight of rats in different groups over time (implantation was performed after a week of observation period under normal conditions). c Rat’s food intake over time in different groups. d Final body weight of rats in different groups. e Final daily food intake in steady state of rats in different groups. f Representative images of white adipose tissue (epididymal fat pad and perirenal fat pad) of the VNS group and the control group. g Adipose tissue weight of rats in different groups. h Epididymal fat pad/body weight ratio (EBR) of rats in different groups. i Percentage of weight loss over time (black dots) in comparison to the reported results by chronic electric stimulation with a rectangular waveform (voltage, frequency, and pulse duration are also shown). All data in b, c, and g are presented as mean ± s.d. In d, e, and h (box plots), dots are the mean, center lines are the median, box limits are the lower quartile (Q1) and upper quartile (Q3), and whiskers are the most extreme data points that are no more than 1.5× (Q3–Q1) from the box limits. Statistical analysis was performed by two-tailed unpaired Student’s t tests. n.s., non-significant (P > 0.05); *P < 0.05, **P < 0.01, and ***P < 0.001

All the rats were sacrificed after the 100-day weight control study for anatomical examinations. Similarly, the anatomical adipose tissues (epididymal fat pad and perirenal fat pad, representative of the body fat level)37 in the VNS group were significantly smaller than the control groups (Fig. 4f). The average weight of the epididymal fat pad and perirenal fat pad were only 4.01 and 1.36 g, 58 and 67% smaller than the control groups, respectively (Fig. 4g). The epididymal fat pad/body weight ratio (EBR) was calculated by dividing the fat pad weight by the total body weight. Average EBR was maintained at 1.14% in the VNS group, significantly lower than the control groups which all exhibited EBR of ~1.7% (Fig. 4h). By comparing the weight difference between the VNS and control groups, our correlated VNS system rapidly achieved 35% weight loss within 18 days and maintained a weight-loss ratio as high as 38% for remaining study period (75 days), which largely exceeded other reported electrical stimulation approaches based on similar rat models38,39,40,41,42,43 (Fig. 4i).

To further exploit the capability of the VNS system for weight loss, the same implantation surgery and analyses were conducted on grown adult rats that have been fed for 7 weeks and reached a steady average body weight of ~500 g. Little body weight gain was observed for the intact group during the 70-day study period, while the VNS group exhibited an obvious body size reduction (Fig. 5a). Similar as the previous study, the Lap group exhibited no difference when compared to the intact group, while the Sham group quickly recovered to the same levels after the initial drop in both body weight and food intake (Fig. 5b, c, respectively). Food intake of the VNS group exhibited a much slower recovery rate and eventually remained at the steady value that was ~2/3 of the control groups. Accordingly, the average body weight of the VNS group exhibited a steep drop over the first 25 days after implantation, followed by a small recovery and stabilized at ~400 g. The final body weight (Fig. 5d) and food intake (Fig. 5e) showed significant differences between the VNS group and control groups, and no significant difference were found among the control groups. The final average body weight was controlled at 410 ± 17 g (n = 4), significantly smaller than the control groups (Sham: 575 ± 22 g; Lap: 569 ± 39 g; and Intact: 574 ± 48 g, n = 4 for each control group). Significant differences were also found in the adipose tissue sizes (Fig. 5f) and weight (Fig. 5g). The final EBR was controlled at 1.26% in the VNS group, while the three control groups maintained a much high value from 1.69% to 1.77% (Fig. 5h). The calculated weight-loss percentage peaked at 38% at day 29 and gradually reached a stable 28% (Fig. 5i).

Fig. 5
Fig. 5

Weight loss of fully grown adult rats after implantation of the VNS device. a Representative images of rats in the VNS group (n = 4) and the control groups (n = 4 for each control group). b Average body weight of rats in different groups over time (implantation was performed after 7 days of observation under normal conditions). c Rat’s food intake in different groups over time. d Final body weight of rats in different groups. e Final daily food intake in steady state of rats in different groups. f Representative images of white adipose tissue (epididymal fat pad and perirenal fat pad) of the VNS group and the control group (scale bar = 2 cm). g Adipose tissue weight of rats in different groups. h Epididymal fat pad/body weight ratio (EBR) of rats in different groups. i Percentage of weight loss over time after implantation of the VNS device. All data in b, c, and g are presented as mean ± s.d. In d, e, and h (box plots), dots are the mean, center lines are the median, box limits are the lower quartile (Q1) and upper quartile (Q3), and whiskers are the most extreme data points that are no more than 1.5× (Q3–Q1) from the box limits. Statistical analysis was performed by two-tailed unpaired Student’s t tests. n.s., non-significant (P > 0.05); *P < 0.05, **P < 0.01, ***P < 0.001


In this work, we present a correlated VNS system as an effective therapeutic strategy for obesity, which provided correlated nerve stimulation signal in response to stomach peristalsis. The VNS device was built based on a flexible TENG that was attached to the stomach wall of rats and could generate biphasic electric pulses when the stomach wall moved. The TENG electrodes were directly connected to the vagus nerve, where the stomach motion-generated voltage signals stimulated the vagus nerve to reduce food intake. We envision that this correlated stimulation could provide less amount but more targeted stimulation so that the nerves might be more responsive to the stimulation, and thus more effective to control food intake. The VNS device exhibited excellent biocompatibility without any signs of side effects from the whole blood and chemical analysis. CT and hematology indicators revealed the implantation was very stable and remained at the same position during the entire testing period. The weight control performance was examined and compared among the VNS, Sham, Lap, and Intact groups of rats that were fed and grown under the same conditions. From the weight-gain test, the average body weight and EBR can be controlled at 350 g and 1.14% in the VNS group, compared to 559–564 g and 1.65–1.69% in the other three control groups with insignificant differences. The VNS system rapidly achieved 35% weight loss within 18 days, which was maintained 38% during the remaining 75-day testing period. From the weight-loss test, the average body weight and EBR was controlled at 410 g and 1.26% in the VNS group, compared to 570–575 g and 1.69–1.77% in the other three control groups. Rats in the VNS groups were also able to recover their normal weight pattern immediately after the implanted VNS devices were removed (Supplementary Figure 13, Supplementary Note 3). Our correlated VNS system demonstrates outstanding weight control results, which largely outperformed other reported chronic microchip VNS systems based on similar rat models. In addition, this correlated VNS is battery free and less invasive compared to the bariatric surgical strategies (e.g., gastric bypass, biliopancreatic diversion or sleeve gastrectomy) for weight control. For future clinical trials, a switch may be needed by the VNS device to control the treatment. This could be achieved by integrating a shutter switch to the electrical wires from the VNS device, as it has been proved that a disconnected device has no impact to food intake and weight change. More broadly, this development demonstrated a successful example of a self-responsive and real-time peripheral neuromodulation mechanism that may be more effective for achieving therapeutic purpose.


Device fabrication and encapsulation

Polyimide film (50 μm) was used as the core package layer that was proven to be biocompatible and corrosion resistant for bio-implanted devices44,45. Casting and curing a pre-polymer to PDMS (15:1 PDMS; Dow Corning, USA) covers the entire device as the shell package (1 mm) to enhance the leakproof and ensure good structural flexibility and stability. To further increase the flexibility to closely fit the non-planar surfaces of stomach and maintain sensitivity in response to stomach motions, a layer of ecoflex (200 μm) was coated onto the surface as another shell structure of the device46,47,48. The PTFE surface was treated by reactive ionic etching to introduce nanostructured features to enhance the electrical output49 (Supplementary Figure 2a). The overall VNS device dimensions are approximately 16 × 12 × 2.5 (L × W × T) mm3 and the weight was measured to be only ~0.8 g.

Electrical characterization of VNS devices

The electrical performance of all implanted VNS devices were measure by a portable oscilloscope (Agilent, DSO1012A, internal impedance is 1MΩ). The voltage signals shown in Fig. 2b, d were measured directly from an implanted VNS device where the two Au leads were connected to the vagus nerve. The stomach motions were induced by injecting water into the stomach using a gavage needle through the mouth and the injection volume difference was 2 mL. The voltage signal shown in Fig. 2c was measured by pressing the VNS devices in phosphate-buffered saline (PBS) solution under different frequency when the VNS device was connected to an external load with the same impedance of the implanted area. Voltage signals in Fig. 2e and Supplementary Figure 3 were measured by the pressing the VNS devices at frequency of 4 Hz when connected to an external load of 0.3MΩ, the same as the impedance of the vagus nerve. The voltage and current output of TENGs as a function of load (10Ω to 200MΩ) was measured by a Stanford Research Systems model SR 560 low-noise preamplifier (internal impedance is 100MΩ). The impedance of implanted VNS device was characterized from 0.01 to 10,000 Hz using an Autolab PGSTAT302N station (Supplementary Figure 5).

Electrophysiological properties of vagus nerve

A Sprague–Dawley rat was anesthetized and its right cervical vagal trunk was carefully exposed (Supplementary Figure 7a). A pair of bipolar platinum hook electrodes was then placed under the right vagal nerve immediately. The exposed nerve tissue was covered with warm (37 °C) paraffin oil. The electrical signals were recorded and analyzed by a Cambridge Electronic Design (CED) 1401 interface (Cambridge, UK) with Spike 2 software to monitor the change the electrical signal of the VNS device (Supplementary Figure 7b and c). The VNS device was implanted in rat’s body with Au leads being connected to anterior and PVNs. The implanted VNS devices were activated by gently pressing the abdomen of the rats at a frequency of 1 Hz. In addition, VNS devices with different sizes were fabricated to study the response of vagus nerve in corresponding to the amplitude of stimulation voltage (Supplementary Figure 8).

Animals and diets

All animal experiments were conducted under a protocol approved by the University of Wisconsin Institutional Animal Care and Use Committee. Seven- and eight-week-old male Sprague–Dawley rats were acquired from Envigo (New Jersey, USA). All rats were housed in separated cages at a temperature-controlled room (22 °C) with a 12-h light/dark cycle with free access to water and Purina PMI-certified rodent chow 5002 (LabDiet, MO, USA).

Food intake and body weight

Body weight and food intake was recorded at 8:00 p.m. every other day. The daily food intake was determined from the difference in food quality between each measurement and divided by two. All rats were deprived of food for 12 h before surgical implantation and blood test. Percentage of weight loss (Pweight loss) over time in Figs. 4i and 5i was calculated according to the formula: Pweight loss = (WIntactWVNS)/WIntact × 100%, where WIntact and WVNS represent the average weight of the Intact group and the VNS group, respectively.

Histological staining of vagus nerve and vital organs

Tissue slices of the bottom of esophagus and its surrounding tissue were prepared. H&E staining, immunohistochemical (Supplementary Figure 1a), and immunofluorescent (Supplementary Figure 1b) staining using anti-S-100 rabbit anti-rat poly-antibody50 showed clearly the anterior vagal trunk and posterior vagal trunk distributed on both sides of the esophagus. Vital organs including heart, lung, liver, spleen, kidney, bowel, stomach, and esophagus were retrieved from rats for H&E staining after euthanasia at different time points (1 day, 7 days, and 15 days) post implantation. In addition, vagus nerves were re-evaluated after the device being implanted for 15 days, and the additional H&E staining results were shown in Supplementary Figure 12.

Cell morphology and immunofluorescence staining

After 3T3 cells were cultured on encapsulation or cell plates in 24-well plates, cell morphology was observed directly using an inverted optical microscope (Nikon Eclipse Ti-U, Japan). The cytoskeleton and nucleus were stained with Texas Red-X phalloidin (591/608 nm) and blue fluorescent Hoechst (352/461 nm) (Thermo Fisher Scientific), respectively. The samples were fixed with 2–4% formaldehyde for 15 min and then rinsed three times with pre-warmed PBS. The samples were incubated with Texas Red-X phalloidin (100 nM) and Hoechst (50 nM) for 30 min at 37 °C. After staining, cells were rinsed with pre-warmed buffer for three times and imaged using a Nikon A1RS confocal microscope.

MTT assay

After 3T3 cells were cultured on the packaging film on 24-well plates, MTT assay (Thermo Fisher scientific) was performed to examine cell proliferation. After incubation at 37 °C in a humidified atmosphere with 5% CO2 for up to 4 days, 100 μL of MTT solution was added to each well. After 4-h incubation, the medium was removed and dimethyl sulfoxide (500 μL/well) was added to dissolve the precipitated fomazan. The optical density (n = 3) of the solution was evaluated using a microplate spectrophotometer at a wavelength of 490 nm.

Device implantation

In brief, anesthesia was induced by inhalation of 2–5% isoflurane and maintained with 2% isoflurane. Following anesthesia, rats were fixed in supine position. The abdomen of rats was scrubbed with iodine scrub, and then alcohol prior to surgery. An incision of 2–5 cm was made on the left upper abdomen of rats. The device was placed beside the stomach. The anterior and posterior vagus nerves were identified and separated from the gastroesophageal junction. For the VNS group, Au wires were placed in contact with the nerves and secured with sterile surgical tape. For the sham group, the device was implanted with insulated electrodes connecting to the vagus nerves. Afterwards, the muscle and skin were sutured layer by layer and the stitches were removed 2 weeks later. The implantation procedure is shown step by step in Supplementary Figure 14. The entire surgery lasted approximately 15 min.

CT scan

CT whole-body scan was performed to evaluate the position and integrity of the VNS device post implantation, which can generate 3D images and reconstruct a high-definition 2D projection image of rats with a resolution of up to 100 µm. In brief, rats were placed in the prone position after anesthesia and scanned by an Inveon micro positron emission tomography/CT scanner (Siemens Medical Solutions, USA) at 1 day, 2 weeks, 6 weeks, and 12 weeks post implantation. CT images were reconstructed and presented as 3D projection or slices.

Hematology data

Whole-blood and chemical analysis were performed for safety assessment pre-implantation and 2, 6, and 12 weeks post implantation. Blood was drawn from the tail vein of the rats and various tests were performed using an Abaxis VetScan HM5 Hematology Analyzer (Abaxis, USA) and VS2 Blood Chemistry Analyzer (Abaxis, USA). No centrifugation or other treatment were needed.

Anatomic examination and adipose tissue collection

After rats were euthanized at the end of the study, an incision was made on the abdomen. The stomach along with the VNS device were taken out for further analysis. In addition, epididymal fat pad and perirenal fat pad were removed and weighed for further analysis (Supplementary Figure 15).

Statistical analysis

For the final body weight, food intake, and adipose tissue weight, statistical analysis was performed by two-tailed unpaired Student’s t tests. n.s., non-significant (P > 0.05); *P < 0.05, **P < 0.01, ***P < 0.001. In box plots, dot is the mean, center line is the median, box limits are the lower quartile (Q1) and upper quartile (Q3), and whiskers are the most extreme data points that are no more than 1.5× (Q3–Q1) from the box limits.

Data availability

The authors declare that all data supporting the findings of this study are available within the Article and its Supplementary Information. The raw data generated in this study are available from the corresponding author upon reasonable request.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.


  1. 1.

    Wormser, D. et al. Separate and combined associations of body-mass index and abdominal adiposity with cardiovascular disease: collaborative analysis of 58 prospective studies. Lancet 377, 1085 (2011).


  • 2.

    Singh, G. M. et al. The age-specific quantitative effects of metabolic risk factors on cardiovascular diseases and diabetes: a pooled analysis. PLoS ONE 8, 65174 (2013).

  • 3.

    Colditz, G. A., Willett, W. C., Rotnitzky, A. & Manson, J. E. Weight gain as a risk factor for clinical diabetes mellitus in women. Ann. Intern. Med. 122, 481–486 (1995).

  • 4.

    Stampfer, M. J., Maclure, K. M., Colditz, G. A., Manson, J. E. & Willett, W. C. Risk of symptomatic gallstones in women with severe obesity. Am. J. Clin. Nutr. 55, 652 (1992).

  • 5.

    Lauby-Secretan, B. et al. Body fatness and cancer—viewpoint of the IARC Working Group. N. Engl. J. Med. 375, 794–798 (2016).

  • 6.

    Jiang, L. et al. The relationship between body mass index and hip osteoarthritis: a systematic review and meta-analysis. Jt. Bone Spine 78, 150–155 (2011).

  • 7.

    Wahl, S. et al. Epigenome-wide association study of body mass index, and the adverse outcomes of adiposity. Nature 541, 81–86 (2016).

  • 8.

    Wang, Y. C., Mcpherson, K., Marsh, T., Gortmaker, S. L. & Brown, M. Health and economic burden of the projected obesity trends in the USA and the UK. Lancet 378, 815 (2011).

  • 9.

    Ng, M. et al. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet 384, 766 (2014).

  • 10.

    Collaborators, G. O. et al. Health effects of overweight and obesity in 195 countries over 25 years. N. Engl. J. Med. 377, 13 (2017).

  • 11.

    Bray, G. A. & Tartaglia, L. A. Medicinal strategies in the treatment of obesity. Nature 404, 672–677 (2000).

  • 12.

    Listed, N. Treating diabetes with bariatric surgery. BMJ 346, f3710 (2013).

  • 13.

    Hwang, S. S., Takata, M. C., Ken, F. & William, F. Update on bariatric surgical procedures and an introduction to the implantable weight loss device: the Maestro Rechargeable System. Med. Devices 9, 291–299 (2016).

  • 14.

    Shikora, S. et al. Vagal blocking improves glycemic control and elevated blood pressure in obese subjects with type 2 diabetes mellitus. J. Obes. 2013, 245683 (2013).

  • 15.

    Krames, E. S., . & Peckham, P. H. & Rezai, A. & Aboelsaad, F. What is neuromodulation?. Neuromodulation 1, 13–18 (2009).

  • 16.

    Borovikova, L. V. et al. Vagus nerve stimulation attenuates the systemic inflammatory response to endotoxin. Nature 405, 458 (2000).

  • 17.

    Groves, D. A. & Brown, V. J. Vagal nerve stimulation: a review of its applications and potential mechanisms that mediate its clinical effects. Neurosci. Biobehav. Rev. 29, 493 (2005).

  • 18.

    Pavlov, V. A. & Tracey, K. J. The vagus nerve and the inflammatory reflex-linking immunity and metabolism. Nat. Rev. Endocrinol. 8, 743 (2012).

  • 19.

    Williams, E. K. et al. Sensory neurons that detect stretch and nutrients in the digestive system. Cell 166, 209–221 (2016).

  • 20.

    Ikramuddin, S. et al. Effect of reversible intermittent intra-abdominal vagal nerve blockade on morbid obesity: the ReCharge randomized clinical trial. JAMA 312, 915–922 (2014).

  • 21.

    Apovian, C. M. et al. Two-year outcomes of vagal nerve blocking (vBloc) for the treatment of obesity in the ReCharge Trial. Obes. Surg. 27, 169–176 (2016).

  • 22.

    Vallaillet, D., Biraben, A., Randuineau, G. & Malbert, C. H. Chronic vagus nerve stimulation decreased weight gain, food consumption and sweet craving in adult obese minipigs. Appetite 55, 245–252 (2010).

  • 23.

    Hampton, T. Proposed rule seeks to safeguard against phased-out chemicals. JAMA 313, 785 (2015).

  • 24.

    Privitera, M. D., Welty, T. E., Welge, J. & Ficker, D. M. Vagus nerve stimulation for partial seizures. Cochrane Database Syst. Rev. 1, CD002896 (2002).

  • 25.

    Zhang, Y. et al. Chronic vagus nerve stimulation improves autonomic control and attenuates systemic inflammation and heart failure progression in a canine high-rate pacing model. Circ. Heart Fail 2, 692–699 (2009).

  • 26.

    Imran, M. A. Implantable neurostimulator devices including both non-rechargeable and rechargeable batteries and methods of use therewith. US patent 9463320B2 (2016).

  • 27.

    Wang, Z. L., Chen, J. & Lin, L. Progress in triboelectric nanogenerators as a new energy technology and self-powered sensors. Energy Environ. Sci. 8, 2250–2282 (2015).

  • 28.

    Qiang, Z. et al. In vivo self-powered wireless cardiac monitoring via implantable triboelectric nanogenerator. Acs Nano 10, 6510 (2016).

  • 29.

    Ma, Y. et al. Self-powered, one-stop, and multifunctional implantable triboelectric active sensor for real-time biomedical monitoring. Nano Lett. 16, 6042 (2016).

  • 30.

    Yoo, H. G., Byun, M., Jeong, C. K. & Lee, K. J. Performance enhancement of electronic and energy devices via block copolymer self‐assembly. Adv. Mater. 27, 3982–3998 (2015).

  • 31.

    Hwang, G. T. et al. Self‐powered wireless sensor node enabled by an aerosol‐deposited PZT flexible energy harvester. Adv. Energy Mater. 6, 1600237 (2016).

  • 32.

    Roslin, M. & Kurian, M. Vagus nerve stimulation in the treatment morbid obesity. Vagus Nerve Stimul. (2002).

  • 33.

    Beck, C. S. & Mason, G. R. Gastric peristalsis. A study of regional rates of contraction. Am. J. Surg. 119, 217 (1970).

  • 34.

    Wang, X. Y., Lammers, W. J., Bercik, P. & Huizinga, J. D. Lack of pyloric interstitial cells of Cajal explains distinct peristaltic motor patterns in stomach and small intestine. Am. J. Physiol. Gastrointest. Liver Physiol. 289, G539 (2005).

  • 35.

    Giknis, M. & Clifford, C. Clinical Laboratory Parameters for Crl:CD(SD) Rats (Charles River Laboratories, Wilmington, MA 2006).

  • 36.

    Giknis, M. & Clifford, C. Clinical Laboratory Parameters for Crl:WI(Han). (Charles River Laboratories, Wilmington, MA, 2008).

  • 37.

    Whittle, A. J. et al. Soluble LR11/SorLA represses thermogenesis in adipose tissue and correlates with BMI in humans. Nat. Commun. 6, 8951 (2015).

  • 38.

    Krolczyk, G. et al. Effects of continuous microchip (MC) vagal neuromodulation on gastrointestinal function in rats. J. Physiol. Pharmacol. 52, 705–715 (2001).

  • 39.

    Laskiewicz, J. et al. Capasaicin induced deafferentation enhances the effect of electrical vagal nerve stimulation on food intake and body mass. J. Physiol. Pharmacol. 55, 155 (2004).

  • 40.

    Królczyk, G. et al. The effects of baclofen on the feeding behaviour and body weight of vagally stimulated rats. J. Physiol. Pharmacol. 56, 121 (2005).

  • 41.

    Bugajski, A. J. et al. Effect of long-term vagal stimulation on food intake and body weight during diet induced obesity in rats. J. Physiol. Pharmacol. 58(Suppl. 1), 5 (2007).

  • 42.

    Gil, K., Bugajski, A. & Thor, P. Electrical vagus nerve stimulation decreases food consumption and weight gain in rats fed a high-fat diet. J. Physiol. Pharmacol. 62, 637 (2011).

  • 43.

    Ziomber, A. et al. Magnetically induced vagus nerve stimulation and feeding behavior in rats. J. Physiol. Pharmacol. 60, 71–77 (2009).

  • 44.

    Jr, R. R. R., Miller, J. A. & Reichert, W. M. Polyimides as biomaterials: preliminary biocompatibility testing. Biomaterials 14, 627–635 (1993).

  • 45.

    Seo, J. M. et al. Biocompatibility of polyimide microelectrode array for retinal stimulation. Mater. Sci. Eng. C 24, 185–189 (2004).

  • 46.

    Kubo, M. et al. Stretchable microfluidic radiofrequency antennas. Adv. Mater. 22, 2749 (2010).

  • 47.

    Lipomi, D. J. et al. Skin-like pressure and strain sensors based on transparent elastic films of carbon nanotubes. Nat. Nanotechnol. 6, 788–792 (2011).

  • 48.

    Yeo, W. H. et al. Multifunctional epidermal electronics printed directly onto the skin. Adv. Mater. 25, 2773–2778 (2013).

  • 49.

    Zhu, G. et al. Triboelectric-generator-driven pulse electrodeposition for micropatterning. Nano Lett. 12, 4960–4965 (2012).

  • 50.

    Bleier, B. S. et al. The accessory posterolateral nerve: an immunohistological analysis. Am. J. Rhinol. Allergy 26, 271–273 (2016).



This publication was supported by the National Institute of Biomedical Imaging and Bioengineering of the National Institutes of Health under Award Numbers R01EB021336 and P30CA014520. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Author information

Author notes

  1. These authors contributed equally: Guang Yao, Lei Kang.


  1. Department of Materials Science and Engineering, University of Wisconsin-Madison, Madison, WI, 53706, USA

    • Guang Yao
    • , Jun Li
    • , Yin Long
    •  & Xudong Wang
  2. State Key Laboratory of Electronic Thin films and Integrated Devices, University of Electronic Science and Technology of China, Chengdu, Sichuan, 610054, People’s Republic of China

    • Guang Yao
    • , Yin Long
    •  & Yuan Lin
  3. Department of Radiology, University of Wisconsin-Madison, Madison, WI, 53705, USA

    • Lei Kang
    • , Hao Wei
    • , Carolina A. Ferreira
    •  & Weibo Cai
  4. Department of Nuclear Medicine, Peking University First Hospital, Beijing, 100034, People’s Republic of China

    • Lei Kang
  5. University of Wisconsin Carbone Cancer Center, Madison, WI, 53705, USA

    • Justin J. Jeffery


Honey bee Royalactin unlocks conserved pluripotency pathway in mammals


Royal jelly is the queen-maker for the honey bee Apis mellifera, and has cross-species effects on longevity, fertility, and regeneration in mammals. Despite this knowledge, how royal jelly or its components exert their myriad effects has remained poorly understood. Using mouse embryonic stem cells as a platform, here we report that through its major protein component Royalactin, royal jelly can maintain pluripotency by activating a ground-state pluripotency-like gene network. We further identify Regina, a mammalian structural analog of Royalactin that also induces a naive-like state in mouse embryonic stem cells. This reveals an important innate program for stem cell self-renewal with broad implications in understanding the molecular regulation of stem cell fate across species.


The regulation of self-renewal and differentiation potential in mouse embryonic stem cells (mESCs) occurs through complex transcriptional networks orchestrated by conserved transcription factors1,2. Although differences exist in the specific signaling pathways that control self-renewal and lineage development, culture conditions that allow for derivation and maintenance of stem cells have been identified3,4,5,6,7,8. In particular, use of two small molecule inhibitors targeting MAPK/ERK Kinase (Mek) and glycogen synthase kinase-3 (GSK3) in addition to Leukemia inhibitory factor (LIF) in serum-free media permitted derivation of germline-competent ESCs that resemble the mature mouse inner cell mass (ICM)9. However, recent findings suggest that prolonged Mek1/2 suppression may have detrimental effects on the epigenetic and genetic integrity of mESCs, effectively limiting their developmental potential10,11. As such, additional methods of maintaining mESCs in an ICM state are required.

Though best-known as an epigenetic driver of queen development in A. mellifera12, the functional component of royal jelly, Major Royal Jelly Protein 1 (MRJP1, also known as Royalactin), has been shown to modulate biological function in a broad range of species12,13,14,15. Indeed, conservation of increased growth stimulation and cellular proliferation phenotypes in response to MRJP1 has been observed in murine hepatocytes16,17,18. While this indicates a functionally important role for this royal jelly protein in regulating cell state and fate, the full scope of its effects has not yet been well characterized19.

In this study, we identify Royalactin as a potent activator of a pluripotency gene network through modulation of chromatin accessibility, that maintains mESC self-renewal in the absence of LIF. Royalactin cultured cells also occupy a more naive ground state capable of generating chimeric animals with germline transmission. Finally, we identify a mammalian structural analog of Royalactin possessing similar functional capacity, uncovering a molecular conservation that supports distinct evolutionary processes.


Royalactin maintains mESC self-renewal and pluripotency

As mESCs provide a powerful model with which to study cellular regulation through pluripotency and differentiation programs, we first employed this model to dissect Royalactin’s mechanisms of action on mammalian cells. Upon LIF withdrawal (serum/–LIF), mESCs previously cultured in serum/+LIF media readily demonstrated a differentiated morphology as expected (Fig. 1a). Surprisingly, addition of Royalactin in the absence of LIF (serum/–LIF + Royalactin) for multiple passages resulted in dose-dependent formation of undifferentiated colonies that demonstrated similar morphology to those grown in the presence of LIF (Fig. 1a, Supplementary Figure 1a). Gene expression profiles of the cells grown in serum/–LIF + Royalactin further demonstrated a high degree of similarity relative to those cultured in the presence of LIF (Fig. 1b, Supplementary Figure 1b).

Fig. 1
Fig. 1

Royalactin maintains stemness in murine embryonic stem cells. a Representative images of J1 and R1 mESCs cultured in serum/+LIF, serum/−LIF, or serum/−LIF + Royalactin for 10 and 20 passages. After LIF withdrawal, mESCs rapidly differentiated, whereas cells cultured with Royalactin supported self-renewal with negligible differentiation. Scale bar, 200 μm. b Quantitative expression of pluripotency and differentiation-associated genes from a. Data are means ± SD (n = 2). c Mice bearing mESC-derived teratomas from J1 mESCs cultured three passages in +LIF and −LIF + Royalactin demonstrated retained pluripotency, and on high magnification (400×) produced differentiated ectodermal, mesodermal, and endodermal tissues. Scale bar, 80 μm. d RNA-seq log2-fold change values in transcript level of all genes in serum/+LIF or serum/–LIF + Royalactin J1 mESCs (passage 10) relative to serum/–LIF. e GO term analysis of differentially expressed genes from d. f ATAC-seq activity in J1 mESCs at passage 10. Each column is a sample, each row is an element. Samples and elements are organized by unsupervised k-means clustering. g GO term analysis of differentially accessible regions from f. h Representative images of Stat3, Esrrb, and Tfcp2l1 knockdown in J1 mESCs with serum/+LIF and serum/−LIF + Royalactin conditions and qPCR analysis of pluripotency and differentiation-associated genes from the same cells. Data are means ± SD (n = 2). Scale bar, 200 μm. RylA Royalactin

Having observed a robust stemness-maintenance effect of Royalactin in vitro, we next examined whether mESCs treated with Royalactin in serum/–LIF retain embryonic identity and developmental potential in vivo. Indeed, mESCs grown in serum/–LIF + Royalactin were grafted subcutaneously and gave rise within 6 weeks to large multi-differentiated teratomas (Fig. 1c). Collectively, these results indicate that Royalactin can functionally maintain self-renewal and pluripotency in mESCs.

Royalactin modulates chromatin and pluripotent networks

In the interest of understanding Royalactin’s effects on the transcriptome, RNA-seq analyses of serum/+LIF, serum/–LIF, and serum/–LIF + Royalactin cells were conducted. These analyses revealed a strong enrichment for canonical pluripotency genes and a suppression of lineage-specific genes in serum/–LIF + Royalactin cells at levels similar to those in mESCs cultured in serum/+ LIF (Fig. 1d). Gene Ontology (GO) term analysis of all genes differentially expressed in the presence of Royalactin revealed strong enrichment for genes involved with proliferation and stemness in the upregulated gene set, and an overrepresentation of developmental processes in the downregulated gene set (Fig. 1e). Similarly, analysis of chromatin accessibility using the assay for transposase-accessible chromatin using sequencing (ATAC-seq)20 of mESCs cultured in serum/–LIF + Royalactin and serum/+LIF conditions revealed similar patterns of increases in ATAC-seq signal relative to mESCs grown in serum/–LIF conditions (Fig. 1f), specifically at promoters (TSS; 14234 total peaks with 7373 gaining accessibility and 6861 losing accessibility; Supplementary Figure 1c), traditional enhancers (TE; 5356 total peaks with 2571 gaining accessibility and 2785 losing accessibility; Supplementary Figure 1d), and super enhancer regions (SEs; 127 of 231 elements gaining accessibility; Supplementary Figure 1e). As expected, high correlation was observed between ATAC-seq changes and RNA-seq experiments (Supplementary Figure 1c–e), with functional annotation revealing that the ATAC-seq changes were located near genes associated with pluripotency, metabolism, and differentiation (Fig. 1g). Furthermore, motif enrichment analysis revealed that TFs such as KLF5, KLF4, and SOX2 bound at high frequency to the Royalactin-upregulated SE constituents (Supplementary Figure 1f). Collectively, this suggested that regulatory regions are sensitive to Royalactin culture conditions and cause subsequent changes in gene expression.

In order to gain a molecular understanding of patterns of gene expression and identify candidate regulators of pluripotency in response to Royalactin, a transcriptional network analysis was performed that identified Stat3, Tfcp2l1, Esrrb, and Nanog as the most significant nodes (Supplementary Figure 1g). Subsequent experimentation revealed a dose-dependent effect for Royalactin in stimulating phospho-Stat3 activation concomitant with other pluripotency factors (Supplementary Figure 1h), which was sustained after 10 and 20 passages (Supplementary Figure 1i). In addition, knockdown of these transcription factors greatly diminished the mESC response to Royalactin, with the most significant effects being observed following Stat3 knockdown (Fig. 1h). As these findings suggested that Royalactin triggers activation of a Stat3-driven LIF-independent pathway on mESC self-renewal, further analysis of gene expression profiles from serum/+LIF and serum/–LIF + Royalactin mESCs were compared to identify 519 genes that are specifically activated in response to Royalactin (Supplementary Figure 1j). GO term analysis showed enrichment of metabolic and biosynthetic processes (Supplementary Figure 1k), reminiscent of mESCs cultured without serum in the presence of inhibitors targeting mitogen-activated protein kinase kinase and GSK3 (2i)21. Global expression profiles from serum/–LIF + Royalactin and 2i-cultured cells also clustered together by principal component analysis away from serum/+LIF-cultured cells (Supplementary Figure 1l). GO term enrichment analysis found that genes involved in basic metabolism, transcription, and development were responsible for this separation (Supplementary Figure 1m). Collectively, this suggested that Royalactin may be driving ground-state pluripotency in mESCs.

Royalactin treated mESCs mimic ground-state pluripotency

We next sought to test the hypothesis that Royalactin was driving a ground-state-like pluripotency state in mESCs. As expected, mESCs cultured in 2i + LIF media maintained pluripotency and sustained expression of a Rex1 GFP pluripotency marker22, while those in 2i base media without inhibitors (0i) readily differentiated (Fig. 2a, b, Supplementary Figure 2). Remarkably, addition of Royalactin in the absence of inhibitors (0i + Royalactin) for multiple passages maintained undifferentiated GFP positive colonies with similar gene expression profiles to 2i + LIF cultured cells (Fig. 2a, b, Supplementary Figure 2). In addition, injection of 0i + Royalactin cultured cells into mouse blastocysts generated chimeric animals with germline transmission, highlighting the robust effects of this protein in vivo (Fig. 2c, Supplementary Table 1).

Fig. 2
Fig. 2

Royalactin drives a ground-state-like pluripotency state in mESCs. a Representative images of J1 and R1 mESCs cultured in serum-free media in the presence (2i + LIF) or absence (0i) of MAPKKi, GSK3i, and LIF for 10 and 20 passages. mESCs rapidly differentiated in 0i, whereas cells cultured with Royalactin (0i + Royalactin) supported self-renewal with negligible differentiation. Scale bar, 200 μm. b Quantitative expression of pluripotency and differentiation-associated genes from a. Data are means ± SD (n = 2). c Chimeras with germline transmission formed by CGR 8.8 mESCs treated for ten passages with 0i + Royalactin. d RNA-seq log2 fold change values in transcript level of all genes in 2i or 0i + Royalactin (0i + RylA) J1 mESCs (passage 10) relative to 0i. e GO term analysis of the differentially expressed genes from d. f J1 mESCs cultured in serum/+LIF, serum/−LIF + Royalactin, 2i + LIF, and 0i + Royalactin for ten passages are projected onto the first two principal components. All genes with mean normalized read counts larger than 10 were considered for principal component analysis (PCA). g Distribution of genes contributing to principal component 1 (PC1) in f, and GO enrichment analysis of genes most strongly contributing to PC1 separation. RylA Royalactin

RNA-seq and GO-term analyses further demonstrated marked similarities in enriched genes between 2i + LIF and 0i + Royalactin cells (Fig. 2d, e), and global expression profiles from 2i + LIF, 0i + Royalactin, serum/+LIF, serum/–LIF + Royalactin, and serum/–LIF cells demonstrated a clear clustering by principal component analysis of 0i + Royalactin cells nearer those cultured in 2i + LIF than those cultured in serum/+LIF (Fig. 2f). GO term enrichment analysis found that genes involved in basic metabolism, transcription, and development were responsible for this separation (Fig. 2g). These data suggest that Royalactin treatment is accompanied by a profound metabolic reprogramming resembling the 2i + LIF state, mimicking the environment of the mature mouse ICM.

Identification of Royalactin mammalian analog

We next wondered whether a homolog of Royalactin existed in mammals. Sensitive searches of sequence databases using iterative PSI-BLAST23, as well as aiming HHPRED sequence and structural profiles against the human and mouse proteomes24 did not reveal any Royalactin orthologs. However, the latter computational tool revealed that Royalactin is distantly related to an existing structure in the PDB database25, a secreted salivary gland protein (SGP) from the sand fly, L. longipalpis (PDB ID: 3Q6K)26. We then used this structure––a six-bladed β-propeller fold with no additional domains—as an accurate template for MODELLER27, yielding a high confidence model for the Royalactin fold (Fig. 3a). The resulting superposition of Royalactin and SGP sequences was then used to seed new and more precise HHPRED scans of the human proteome, in search of a possible structural and functional analog of the Royalactin/SGP β-propeller fold. Fitting the description of a secreted, single domain chain, with a predicted 6-bladed β-propeller architecture, only one protein, the provisionally named NHL Repeat Containing 3 (NHLRC3), arose as a potential candidate, with striking fold similarity to the Royalactin model (Fig. 3a). Although no known function of NHLRC3 has been identified to date, single-cell RNA-seq analyses of early mouse embryos revealed that it is expressed starting in E4.5 embryos, and that its expression increases steadily thereafter (Supplementary Figure 3a). To elucidate whether it served a functional purpose in stemness maintenance in mESCs similar to that observed with Royalactin, recombinant mouse NHLRC3 was added to mESC culture in the presence of serum/–LIF (serum/–LIF + NHLRC3) as well as 0i (0i + NHLRC3). As seen with Royalactin, NHLRC3 maintained mESCs in an undifferentiated state in both culture conditions for multiple passages (Fig. 3b, c, Supplementary Figure 3b, c), with expected changes in gene expression (Fig. 3d, e). Additionally, injection of 0i + NHLRC3 cultured cells into mouse blastocysts generated chimeric animals with germline transmission, highlighting the robust effects of this protein in vivo (Supplementary Figure 4, Supplementary Table 1). Thus, NHLRC3 appears to be a mammalian pluripotency maintenance factor, whose existence demonstrates a remarkable conservation of macromolecular structure and function. We renamed NHLRC3 as Regina due to its conservation of functions with those of Royalactin and the queenmaker Royal Jelly.

Fig. 3
Fig. 3

The mammalian structural analog of Royalactin maintains naive and ground-state pluripotency in mouse embryonic stem cells. a Computational modeling predicts the structure of Royalactin (left), allowing for identification of a highly structurally analogous protein, NHLRC3 (center). Superimposition of these models (right) demonstrates striking similarity between them. b Representative images of J1 and R1 mESCs cultured in serum/+LIF, serum/–LIF, or serum/−LIF + NHLRC3 for 10 and 20 passages. After LIF withdrawal, mESCs rapidly differentiated, whereas cells cultured with NHLRC3 supported self-renewal with negligible differentiation. Scale bar, 200 μm. c Representative images of J1 and R1 mESCs cultured in serum-free media in presence (2i + LIF) or absence (0i) of MAPKKi, GSK3i, and LIF for 10 and 20 passages. mESCs rapidly differentiated in 0i, whereas cells cultured with NHLRC3 (0i + NHLRC3) supported self-renewal with negligible differentiation. Scale bar, 200 μm. d Quantitative expression of pluripotency and differentiation-associated genes from b. Data are means ± SD (n = 2). e Quantitative expression of pluripotency and differentiation-associated genes from c. Data are means ± SD (n = 2)

In summary, our results demonstrate an unexpected capacity for Royalactin as a pluripotency factor that confers self-renewal and promotes emergence of the naive pluripotent gene regulatory network, and identify Regina as a factor that can promote ground-state pluripotency in mESCs. A better understanding of the interaction of Royalactin and Regina with genetic pathways and biochemical processes conserved in evolution, and how the different pluripotency networks function independently or synergistically to maintain stem cell self-renewal, will advance our efforts to better control stem cell fate, and provide a platform for dissection of the pluripotent state. Our findings and the discovery of Regina thus support the intriguing idea that profound molecular conservation underlies even the most evolutionarily novel traits. Future work in dissecting the mechanistic action of Royalactin in mammalian cells, including further characterization of its mammalian structural analog Regina, will shed new light on mammalian pluripotency and provide additional means to enhance optimal maintenance and derivation of ESCs for therapeutic application and regenerative medicine.


Embryonic stem cell culture

J1 and R1 mESCs (gift from Howard Y. Chang) were grown on 0.2% gelatin-coated (Sigma G6144) tissue culture plates. Cells were maintained for a minimum of ten passages in mESC serum medium containing KnockOut™ D-MEM (Life Technologies 10829), 15% HyClone™ fetal bovine serum (FBS; Thermo Scientific™ SH30396.03), 100 U/ml Penicillin-Streptomycin (P/S; Life Technologies 15140), 1% MEM non-essential amino acids (Life Technologies 11140), 1% GlutaMAX™ (Life Technologies 35050), and 0.1 mM 2-mercaptoethanol (Life Technologies 21985). Mouse leukemia inhibitory factor (LIF; Millipore ESG1107; 1000 U/mL), purified Royalactin (0.5 mg/mL), or purified NHLRC3 (0.125 mg/mL) were added to culture as indicated. Cells were passaged every 2 to 3 days using Trypsin-EDTA (0.25%; Life Technologies 25200) and seeded at 15,000 cells/cm2. Media and protein were changed daily.

For cultures under 2i conditions, J1, R1, and Rex1-GFP mESCs were grown on Poly-L-ornithine (Sigma-Aldrich)/Laminin (Fisher Scientific 23017–015) coated plates and CGR8.8 mESCs were grown on Matrigel (Corning 354277) plates coated according to manufacturer’s specifications. All cells were maintained a minimum of ten passages in serum-free media containing: 1:1 Neurobasal:DMEM-F12 base (Thermo Scientific 21103049), 1% Glutamax, 1% high insulin N2 (Thermo Scientific 17502001), 1% B27 supplement (Thermo Scientific 12587001), 0.1 mM 2-mercaptoethanol, 1% Penicillin-Streptomycin, 1% MEM non-essential amino acids, 1% sodium pyruvate (Thermo Scientific 11360070). Mouse leukemia inhibitory factor (LIF; Millipore ESG1107; 1000 U/mL), 1 μM PD0325901 (Selleckchem), 3 μM CHIR99021 (Selleckchem), purified Royalactin (0.5 mg/mL), or purified NHLRC3 (0.125 mg/mL) were added to cultures as indicated. Cells were passaged every 2 to 3 days using Accutase (Stemcell Technologies 07920) and seeded at 15,000 cells/cm2. Media and protein were changed daily.

Production of recombinant Royalactin and NHLRC3

FLAG-MRJP1-His (Genbank ID# NM_001011579.1) was cloned into LakePharma’s proprietary antibiotic selection vector and transfected by electroporation into suspension CHO parental cells. The FLAG-MRJP1-His stable cell line was generated after 2 weeks of antibiotic selection period. Purification of FLAG-MRJP1-His was achieved by a two-step chromatography method. Conditioned media was centrifuged, filtered, and loaded onto an anion exchange chromatography (AEX) resin pre-equilibrated with 20 mM Tris pH 7.5. FLAG-MRJP1-His was eluted by increasing sodium chloride concentration and fractions containing the protein were pooled together. This sample was further polished by a second immobilized metal (Ni) affinity chromatography (IMAC) step using increasing concentrations of imidazole for elution. SDS-PAGE was performed for each fraction and the ones containing FLAG-MRJP1-His were pooled together for dialysis. The final formulation for FLAG-MRJP1-His was 200 mM NaCl in 30 mM HEPES pH 7.0.

For production of supernatants containing recombinant NHLRC3, suspension CHO cells were seeded at 350,000 cells/mL into 90% CD OptiCHO Medium (Thermo Scientific 12681011) containing 6mM L-glutamine (Thermo Scientific 25030–081) and 10% CHO CD EfficientFeed (Thermo Scientific A1023401). CHO cells were grown at 37 °C for 5 days, shaking constantly. The cell suspension was centrifuged at 10,000 × g for 40 min and the supernatant run through a 0.22 μm filter. Presence of concentrated NHLRC3 was verified by western blot. The filtered supernatant was concentrated 35-fold and used directly in cell culture assays.

RNA extraction and quantitative PCR

Total RNA was isolated using TRIzol® (Life Technologies) and RNeasy Kit (QIAGEN) according to the manufacturer’s protocol. cDNA was made with Superscript VILO (Life Technologies). All primers (Supplementary Table 2) were tested for efficiency and single products confirmed. qPCR analyses were performed on the Light Cycler 480II (Roche).

Lentiviral expression and viral production

pLKO vectors were a gift of Alejandro Sweet-Cordero. N103 vector was a gift of Howard Y. Chang. Sequence verified constructs were used to produce lentivirus using pRSV (Addgene plasmid #12253), pMD2.G (Addgene plasmid #12259), and pMDLg/pRRE (Addgene plasmid #12251). 293Ts were maintained in G418 (Sigma-Aldrich G8168). Plasmids were transfected using Lipofectamine 2000 following manufacturer’s protocol (Thermo-Fisher Scientific 11668). After 12 h, media was changed to viral production media (DMEM, 10% FBS, 1% P/S, 20 mM HEPES). After 48 h, media was collected, spun to remove cell debris, and lentiviral-containing supernatant was added to Lenti-X™ Concentrator (CloneTech). Following incubation at 4 ˚C for >4 h, the mixture was spun at 500 × g for 45 min, the pellet resuspended in mESC media, and frozen at −80 ˚C.

Lentiviral transduction of mESCs

mESCs were plated at a density of 30,000/cm2. After 12 h, virus was added with 6 µg/mL polybrene. Media was changed 12 h later and puromycin selection began 48 h post-transduction.

Cell culture for teratoma formation assay

J1 mESCs were cultured in serum-free media (as described above) with addition of 1000 U/mL mouse LIF, 1 µM PD0325901, and 3 µM CHIR99021, or 0.5 mg/mL Royalactin for three passages. Cells were grown in suspension on Corning Costar Ultra-Low attachment plates (Sigma-Aldrich). Wells were seeded in duplicate at a cell density of 100,000/well, and media and protein were changed daily. To split, colonies were first pelleted by centrifugation at 1500 × g, suspended in TrypLE (Life Technologies), incubated at 37 °C for 5 min, diluted with PBS (Life Technologies) and pelleted. Cells were counted and re-seeded at a density of 100,000/mL.

Teratoma generation and histopathology

All animal studies were conducted in accordance with Stanford University animal use guidelines and were approved by the Administrative Panel on Laboratory Animal Care (APLAC). J1 mESCs were mixed with Matrigel (BD 356237) prior to being subcutaneously injected into 8-week-old female SCID/Beige mice (Charles River) on each flank. Four weeks after injection, the mice were euthanized and the teratomas were harvested. All animal studies were approved by Stanford University IACUC guidelines. For histological analysis, slides were stained with hematoxylin and eosin (H&E) following manufacturer’s instructions. Analyses were performed by a board-certified veterinary pathologist.

Chimera experiments

CGR 8.8 mESCs were grown in serum-free media (as described above) with addition of 1000 U/mL mouse LIF, 1 µM PD0325901, and 3 µM CHIR99021 (2i + LIF), 0i + Royalactin, or 0i + NHLRC3 (as noted) for ten passages. Media was changed daily. Cells were passaged using Accutase (Stemcell Technologies 07920) and suspended in M2 media for injection.

Protein extraction and western blot analysis

Cellular extracts were prepared using lysis buffer containing 50 mM Tris HCL (pH 7.5), 250 mM NaCl, 1% NP-40, 0.5% Na-deoxycholate, 0.1% SDS, 1 mM phenylmethylsulfonyl fluoride, Halt™ Protease, and Phosphatase Inhibitor Cocktail (ThermoFisher Scientific). Extracts were run on a 4–12% Bis-Tris gel (Novex) and transferred onto PVDF membranes. Blots were blocked in 5% milk PBS-T (TBS for phospho-specific) for 1 h at room temperature followed by overnight incubation at 4 ˚C with primary antibodies. HRP-conjugated secondary antibodies were used at 1:10,000. Antibodies used in this study include Nanog (ReproCELL, RCAB001P, 1:1000), Klf2 (Millipore, 09–280, 1:1000), Esrrb (Perseus Proteomics, PP-H6705–00, 1:500), Stat3 (Cell Signaling, 124H6, 1:1000), pStat3 Y705 (Cell Signaling, D3A7, 1:1000), Sox2 (Santa Cruz, sc-17320, 1:250), Tfcp2l1 (R&D, AF5726, 1:250), NHLRC3 (OriGene, TA336106, 1:500), anti-HA (Cell Signaling, C29F4, 1:1000), and Actin-HRP (Santa Cruz, sc-1616, 1:2500). Uncropped scans of the most important blots are included in Supplementary Figure 5.

RNA-seq library construction

RNA was extracted using TRIzol and purified on column with the RNeasy Mini Kit (Qiagen). Ribosomal RNA was depleted with the Ribo-Zero Gold rRNA Removal Kit (Illumina). RNA was lyophilized, suspended in 10 μL of water and fragmented to an average size of 200 base pairs using the Ambion RNA Fragmentation Kit (AM8740), and purified using Zymo clean and concentrator 5 columns. 3′ Phosphorylation, adapter ligation, reverse transcription, immunoprecipitation, circularization, amplification, and PAGE separation were performed using the FAST-iCLIP library construction method as previously described28. The quality of the libraries, including size distribution and molarity, was assessed on a BioAnalyzer High Sensitivity DNA chip (Agilent). The libraries were then multiplexed and sent for sequencing on an Illumina NextSeq 400 High Output machine for 1 × 75 cycles. Sequencing data deposited under GEO GSE81799.


ATAC-seq was performed on 50,000 J1 mESCs20. Cells were grown in serum mESC media (as described above) in serum/+LIF, serum/–LIF, or serum/–LIF + Royalactin for ten passages. Cells were washed with PBS (Life Technologies), trypsinized with Trypsin-EDTA (0.25%), quenched with serum mESC media, washed with PBS, before nuclear isolation with NP-40. Nuclei were resuspended in a transposase reaction mix containing 25µL 2× TD buffer, 2.5 µL Transposase (Illumina) and 22.5 µL of nuclease free water with sequencing adapters. Final libraries were purified on column using the QIAquick PCR Purification Kit (Qiagen) per the manufacturer’s protocol as well as with Agencourt AMPure XP beads (Beckman Coulter) to remove any remaining free adapters. The quality of the libraries, including size distribution and molarity, was assessed on a BioAnalyzer High Sensitivity DNA chip. Libraries were then multiplexed and sent for deep sequencing on the Illumina HiSeq 2500 machine for 2 × 50 cycles. Sequencing data deposited under GEO GSE81799.

RNA-seq data analysis

Reads were aligned to the mouse reference genome (build mm9) using Tophat. A maximum of a default 2 mismatches was allowed for read alignment. Gene counts were calculated using the HTSeq-count utility29 and used as an input for differential gene expression analysis with DESeq version 1.20.030. Genes with a p-value of 0.05, as well as a fold change of 2 were selected for further analysis. Validation of top differentially regulated genes was performed with quantitative reverse transcription polymerase chain reaction. Further network analysis on differentially significant genes was performed using NetworkAnalyst31. For RNA-seq analysis, GO terms were obtained using DAVID and its default parameters.

PCA analysis was performed using samples as indicated. The genes that led to the maximum amount of variance (PC1) were selected and GO terms obtained using the GO Consortium. Samples from different libraries were normalized using shifted log of normalized counts using DESeq. The ‘plotPCA’ function, which is a part of DESeq2, was used to construct the PCA plots.

ATAC-seq data analysis

Reads were aligned to the mouse reference genome (build mm9) using Bowtie. The ATAC-seq regions were divided into separate analyses: correlation with closest TSS, correlation with 5356 traditional enhancer regions present in the mm9 genome, and correlation with 361 super-enhancer regions discovered for the mm9 genome28. The ATAC-seq signals for serum/+LIF, serum/–LIF, and serum/–LIF + Royalactin after ten passages were compared using DESeq, and the results are represented in the heatmaps. The heatmaps for TSS regions, traditional enhancers, super-enhancers, and differential peaks were produced using unsupervised clustering methods, which used the normalized signal values obtained by quantile normalization, to extract transitions between two states: upregulated and downregulated. The differential peaks between serum/–LIF + Royalactin and serum/–LIF were used for correlation with RNA-seq. The peaks were filtered on the basis of a p-value threshold of 0.05 as well as fold change. Boxplots were produced using the ‘BOXPLOT’ function in R. p-value was calculated using the student’s t-test.

GO terms for peaks differentially expressed in serum/–LIF and serum/–LIF + Royalactin was performed using GREAT. The significant GO terms were filtered to only include GO terms associated with pluripotency and GO terms associated with metabolism. For pluripotency related GO terms, biological processes including morphogenesis, development, proliferation, and stem cell processes were analyzed. For metabolic GO terms, biological processes that were related to metabolism and biosynthetic processes were chosen. Motif analysis for the differential peak lists was performed using HOMER with all differential peaks used as background.

Structural modeling and Royalactin analog identification

As implemented at the MPI Toolkit (, HHPRED enables sensitive searches of sequence and structural databases through the assembly of profile Hidden Markov Models (HMMs) from a seed sequence, with multiple iterations of Hhblits (a more sensitive and faster program than PSI-BLAST) and PSIPRED (a very accurate secondary structure prediction program)24. The detection of a six-bladed β-propeller fold for Royalactin from the top salivary gland protein (PDB ID: 3Q6K) HHPRED hit was accompanied by a significant score of 177.4, an E-value of 6e−28, and a 28% amino acid identity from the structure-guided overlap of the mature, 413 residue honeybee Royalactin, with the 381 amino acid sand fly SGP. This structural alignment was used by MODELLER27 to guide a secure template-guided three-dimensional model of Royalactin (with a VERIFY3D score of 119.7)32, and also to nucleate a more sensitive search by HHPRED for a structurally analogous protein (to the greater Royalactin/SGP family) in the human and mouse proteomes. This latter screen, filtered by the signal peptide, single domain, and six-bladed β-propeller fold constraints common to Royalactin and SGPs, yielded NHLRC3 (Uniprot IDs: Q5JS37 and Q8CCH2, for the human and mouse orthologs, respectively) as the sole analog candidate. A three-dimensional model of the NHLRC3 β-propeller was then built by MODELLER on the best six-bladed β-propeller template recognized by HHPRED, Peptidyl-alpha-hydroxyglycine alpha-Amidating Lyase (PDB ID: 3FVZ; at a significant score of 125.6, E value of 1.7e-18, and amino acid identity of 24%). The comparison and visualization of Royalactin and NHLRC3 structural models were in turn performed by PyMOL ( We recognize that β-propeller folds in general (irrespective of ‘blade’ number) are consistently used as interaction scaffolds and preferred binding platforms in the cell33. The structural models are available upon request.


Antibiotic and acid-suppression medications during early childhood are associated with obesity


Objective Gut microbiota alterations are associated with obesity. Early exposure to medications, including acid suppressants and antibiotics, can alter gut biota and may increase the likelihood of developing obesity. We investigated the association of antibiotic, histamine-2 receptor antagonist (H2RA) and proton pump inhibitor (PPI) prescriptions during early childhood with a diagnosis of obesity.

Design We performed a cohort study of US Department of Defense TRICARE beneficiaries born from October 2006 to September 2013. Exposures were defined as having any dispensed prescription for antibiotic, H2RA or PPI medications in the first 2 years of life. A single event analysis of obesity was performed using Cox proportional hazards regression.

Results 333 353 children met inclusion criteria, with 241 502 (72.4%) children prescribed an antibiotic, 39 488 (11.8%) an H2RA and 11 089 (3.3%) a PPI. Antibiotic prescriptions were associated with obesity (HR 1.26; 95% CI 1.23 to 1.28). This association persisted regardless of antibiotic class and strengthened with each additional class of antibiotic prescribed. H2RA and PPI prescriptions were also associated with obesity, with a stronger association for each 30-day supply prescribed. The HR increased commensurately with exposure to each additional medication group prescribed.

Conclusions Antibiotics, acid suppressants and the combination of multiple medications in the first 2 years of life are associated with a diagnosis of childhood obesity. Microbiota-altering medications administered in early childhood may influence weight gain.

Corticosteroid therapy for sepsis: a clinical practice guideline

Three Steps Toward a More Sustainable Path for Targeted Cancer Drugs

Spending on cancer drugs in the United States has increased substantially over the past 5 years, from $26 billion in 2012 to more than $45 billion in 2016.1 Targeted cancer drugs, including small molecules, monoclonal antibodies, and other therapies based on genomic and related analyses, contributed 60% of this spending growth.2 One estimate suggests that by 2021, cancer drugs will comprise one-quarter of the US late-stage pharmaceutical research and development pipeline, and 87% of these products will be targeted agents.2

Imatinib, the small-molecule oral tyrosine kinase inhibitor for chronic myeloid leukemia (CML), is often cited as the model targeted cancer drug; imatinib is highly effective and has reduced toxicity vs previous therapies. For patients with CML treated with imatinib, overall 10-year survival is 84%.3 Introduction of imatinib was associated with a reduction in US age-adjusted CML deaths per 100 000 persons from 0.9 in 1996 to 0.4 in 2006. In 2015, the Medicare estimated monthly price for imatinib was $9299.4

Yet most targeted cancer drugs do not extend life to nearly the same degree. Even though many cancer drugs show improvement in surrogate measures, such as progression-free survival, substantial improvements in overall life expectancy have been rare. For example, in 2017, neratinib was approved for patients with early-stage breast cancer after improving invasive disease–free survival by 2% (from 92% to 94%) after 2 years of follow-up, without published survival data.5 The estimated monthly price of neratinib is $10 500.

Imatinib exemplifies the promise of targeted therapy, whereas neratinib exemplifies the concern: marginally effective treatments that, in aggregate, strain US health care spending. The distorted pricing of marginally effective drugs risks crowding out the capacity of the US health system to pay for highly effective cancer drugs or other therapies of public health importance, potentially jeopardizing valuable innovation and escalating out-of-pocket expenses. The combination of high prices and marginal effectiveness is unsustainable.

We propose 3 steps to promote targeted cancer drugs that yield meaningful clinical benefits while reducing overall price growth. The recommendations are informed by discussions of a group of experts from health care economics, policy, law, and regulation; cancer research and medicine; patient advocacy; and the pharmaceutical and insurance industries.

The FDA Should Define Minimum Clinically Meaningful Effect Sizes

The FDA has 2 pathways to approve new drug applications. The regular approval pathway is based on demonstration of “clinical benefit,” which is defined as prolongation of life, a better life, or “an established surrogate.” The accelerated approval pathway, in contrast, is based on a surrogate measure reasonably likely to predict clinical benefit. In 2007, the FDA issued guidance on cancer trial end points to support claims of benefit. The guidance referenced approval as being “highly dependent on … effect size” but did not specify minimum effect sizes.6 An essential question thus remains unanswered: what minimum effect size defines meaningful benefit? This ambiguity is particularly problematic in increasingly common scenarios, such as when new targeted cancer drugs demonstrate statistically significant but clinically questionable improvements in surrogate measures.

The FDA should develop guidance on minimum clinically meaningful effect sizes for cancer drugs. This would clarify prior FDA guidance and move from the current uncertain concept of meaningful clinical benefit to a consensus-driven definition. The agency could empanel multidisciplinary advisory councils of scientists, oncologists, patient advocates, and industry representatives to achieve this aim.

Clinical experts already support the principle. After convening work groups to define clinically meaningful outcomes for 4 malignancies, the American Society for Clinical Oncology (ASCO) endorsed a minimum absolute improvement of 3 to 6 months in overall survival over best available treatment for drug trials among patients with metastatic disease. Guidance could separately address cases in which, despite little or no change in median overall survival or hazard ratios, small proportions of patients experience large gains and the challenge of estimating benefits when pivotal trials involve head-to-head comparisons against active controls, thereby potentially underestimating the overall efficacy of novel agents.

By defining norms, the FDA would encourage manufacturers to design trials that demonstrate clinically meaningful benefits. The FDA could consider these thresholds in weighing benefits and risks for the purpose of approval decisions, payers could use them to better bargain on price and formulary with drug makers, guideline writers could use them to prioritize among drugs with similar indications, and clinicians and patients could use them to improve shared decision making.

Medicare Should Negotiate for Targeted Cancer Drugs

Medicare is the largest purchaser of cancer drugs. Medicare pays for targeted cancer drugs through Part B, which covers infused drugs, and Part D, which covers prescription drugs. The law, however, prevents Medicare from directly negotiating with manufacturers on drug prices. Instead, in Part B, Medicare pays for drugs under the buy-and-bill system, in which hospitals and physician offices purchase drugs and bill Medicare at 6% above the average sales price. In Part D, Medicare plan sponsors, typically insurance companies or pharmacy benefit managers, manage pricing negotiations. The law also effectively ensures that Medicare Parts B and D cover all FDA-approved cancer drugs for on-label indications as well as off-label indications listed in approved compendia, whether through “reasonable and necessary” language in Part B or “protected class” language in Part D. Consequently, Medicare cannot exercise pricing leverage through coverage determinations or formulary design.

Congress could direct the Centers for Medicare & Medicaid Services (CMS) to conduct a demonstration project in which Medicare negotiates the prices of targeted cancer drugs paid for by Parts B and D. The demonstration also should authorize Medicare to apply limited formulary tools, such as coverage restrictions or tiering, to marginally effective targeted cancer drugs or targeted cancer drugs with therapeutic alternatives. Alternative drugs should include not only biosimilars but also chemically or biologically different drugs with overlapping indications and benefit-risk profiles. Protected classes could be narrowed to permit exclusion of drugs with overlapping indications or mechanisms of action. An appeals process could address special cases. Advisory panels with diverse and relevant expertise, including patient advocates, could inform pricing discussions and formulary design.

Granting Medicare the ability to negotiate on price and use a formulary is politically challenging. The National Academies, the Medicare Payment Advisory Commission, and others have recommended that the federal government use its bargaining power to negotiate drug prices. This step could be operationalized either by Congress granting CMS authority as a single entity to negotiate with pharmaceutical companies or through competitive bidding. The program could phase in over multiple years, starting with clinical settings where therapeutic alternatives exist. This approach could foster evaluation and refinement based on lessons learned.

Guidelines Should Prioritize Drugs by Benefit and Price

Physicians and patients should be able to consider the prices of targeted cancer drugs along with their benefits and harms when selecting treatments. Evidence-based guidelines are best positioned to meet this need. Such guidelines should demarcate marginally effective from highly effective drugs. In addition, for cases in which 2 or more drugs or regimens have comparable benefit-harm profiles for an indication, guidelines should prioritize the lower-priced alternative.

Although organizations that produce practice guidelines have taken steps to incorporate costs, they should go further. ASCO could extend its value framework, which displays cost alongside net health benefit, to prioritize treatment regimens in its clinical practice guidelines. The National Comprehensive Cancer Network could rank-order treatment regimens in its practice guidelines, informed by its Evidence Blocks, which already evaluate affordability alongside other measures. In addition, other groups have developed reports on pricing, effectiveness, and value for cancer treatments and drugs more broadly that merit the attention of guideline writers.

Guidelines should also promote price transparency. To do so, they could report the estimated monthly price of cancer drugs, perhaps by using the amount reimbursed by Medicare. Although estimates of out-of-pocket expenses would be most useful to patients, and prices vary by payer, the Medicare payment amount correlates with patient co-insurance expenses and has the advantage of being a reference standard. Practice guidelines that rank-order targeted cancer drugs by clinical benefit and price and deprioritize marginally effective drugs could be influential, informing insurer value-based reimbursement programs and clinical pathways.


Successfully implementing steps to limit the use of high-priced, marginally effective drugs will be difficult; patients with life-threatening diseases may expect access to drugs despite their high costs and limited benefits. Nevertheless, the ultimate beneficiaries of these changes will be patients, who deserve the substantial efficacy, reduced toxicity and enhanced value that were the original promise of targeted cancer drugs.

Source: JAMA


Risk of Unnatural Mortality in People With Epilepsy

Key Points

Question  What is the risk and medication contribution to cause-specific unnatural mortality in people with epilepsy?

Findings  In this population-based cohort study, more than 50 000 people with epilepsy and 1 million matched individuals without epilepsy were identified in 2 data sets from the general populations of England and Wales. People with epilepsy had a 3-fold increased risk of any unnatural mortality and a 5-fold increased risk of unintentional medication poisoning; psychotropic and opioid, but not antiepileptic, drugs were most commonly used in poisoning.

Meaning  Clinicians should provide advice on unintentional injury and poisoning and suicide prevention and consider the toxicity of concomitant medication when prescribing drugs for people with epilepsy.


Importance  People with epilepsy are at increased risk of mortality, but, to date, the cause-specific risks of all unnatural causes have not been reported.

Objective  To estimate cause-specific unnatural mortality risks in people with epilepsy and to identify the medication types involved in poisoning deaths.

Design, Setting, and Participants  This population-based cohort study used 2 electronic primary care data sets linked to hospitalization and mortality records, the Clinical Practice Research Datalink (CPRD) in England (from January 1, 1998, to March 31, 2014) and the Secure Anonymised Information Linkage (SAIL) Databank in Wales (from January 1, 2001, to December 31, 2014). Each person with epilepsy was matched on age (within 2 years), sex, and general practice with up to 20 individuals without epilepsy. Unnatural mortality was determined using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes V01 through Y98 in the Office for National Statistics mortality records. Hazard ratios (HRs) were estimated in each data set using a stratified Cox proportional hazards model, and meta-analyses were conducted using DerSimonian and Laird random-effects models. The analysis was performed from January 5, 2016, to November 16, 2017.

Exposures  People with epilepsy were identified using primary care epilepsy diagnoses and associated antiepileptic drug prescriptions.

Main Outcomes and Measures  Hazard ratios (HRs) for unnatural mortality and the frequency of each involved medication type estimated as a percentage of all medication poisoning deaths.

Results  In total, 44 678 individuals in the CPRD and 14 051 individuals in the SAIL Databank were identified in the prevalent epilepsy cohorts, and 891 429 (CPRD) and 279 365 (SAIL) individuals were identified in the comparison cohorts. In both data sets, 51% of the epilepsy and comparison cohorts were male, and the median age at entry was 40 years (interquartile range, 25-60 years) in the CPRD cohorts and 43 years (interquartile range, 24-64 years) in the SAIL cohorts. People with epilepsy were significantly more likely to die of any unnatural cause (HR, 2.77; 95% CI, 2.43-3.16), unintentional injury or poisoning (HR, 2.97; 95% CI, 2.54-3.48) or suicide (HR, 2.15; 95% CI, 1.51-3.07) than people in the comparison cohort. Particularly large risk increases were observed in the epilepsy cohorts for unintentional medication poisoning (HR, 4.99; 95% CI, 3.22-7.74) and intentional self-poisoning with medication (HR, 3.55; 95% CI, 1.01-12.53). Opioids (56.5% [95% CI, 43.3%-69.0%]) and psychotropic medication (32.3% [95% CI, 20.9%-45.3%)] were more commonly involved than antiepileptic drugs (9.7% [95% CI, 3.6%-19.9%]) in poisoning deaths in people with epilepsy.

Conclusions and Relevance  Compared with people without epilepsy, people with epilepsy are at increased risk of unnatural death and thus should be adequately advised about unintentional injury prevention and monitored for suicidal ideation, thoughts, and behaviors. The suitability and toxicity of concomitant medication should be considered when prescribing for comorbid conditions.


%d bloggers like this: