A research team at the National University of Singapore and other institutions has tested an implantable prototype called NeuroAxis in diabetic rats. The device generates power from glucose in the body and uses it to stimulate the vagus nerve. The voltage produced by this power generation also automatically determines when stimulation occurs. Weijun Xiu, Qian Wang and colleagues published the peer-reviewed study in Nature on November 18, 2026 (DOI: 10.1038/s41586-026-11251-7).

The device's output voltage correlated strongly with blood glucose measured from blood samples. The researchers examined the effect of nerve stimulation on blood glucose separately, using a control device that had the same fuel cell but delivered no stimulation. How did the team separate the process of drawing power from glucose from the effect of electrical stimulation of the vagus nerve?

AD

Fuel cell voltage determines when stimulation occurs

The authors point out that existing autonomic nerve stimulation devices lack a mechanism for measuring metabolic state and factoring it into stimulation decisions. NeuroAxis generates power from glucose in the body and stores it in a capacitor. It needs no battery or external wireless power supply, and it indirectly estimates glucose concentration in the body from the fuel cell's terminal voltage.

The terminal voltage is measured every 30 minutes. When five consecutive valid readings exceed 0.6 V, an electrode delivers three minutes of electrical stimulation to the vagus nerve, which runs near the esophagus below the diaphragm. If a reading falls to 0.6 V or below, or once stimulation ends, the count of consecutive readings resets to zero. This kind of control, in which stimulation is regulated according to measurements, is called closed-loop control. An example control flow the authors published on GitHub matches the decision conditions shown in Fig. 3b of the paper. Bluetooth Low Energy is used to send measurements to external devices.

To verify that the circuit delivers stimulation as configured, the team immersed the device in glucose solutions of different concentrations. No stimulation occurred in a 3 mM solution, whereas in a 12 mM solution a voltage satisfying the decision conditions persisted and stimulation occurred. In a calibration using three independent samples, the voltage changed by 0.034 V for every 1 mM change in glucose concentration. The 0.6 V threshold used for control is the fuel cell's terminal voltage, and these results show the circuit's response in solution.

In the rat experiments, the Pearson correlation coefficient between the device's output voltage and blood glucose measured from blood samples was r = 0.96 across 50 measurements. The correlation between glucose concentration in interstitial fluid under the skin of the abdomen and blood glucose was r = 0.98 across 20 measurements. These figures refer to the number of measurement points; a separate 28-day voltage-tracking experiment used five diabetic rats. The results indicate that, under the experimental conditions, fuel cell voltage can serve as an indirect indicator of blood glucose. However, determining whether nerve stimulation changes blood glucose requires comparing groups with and without stimulation.

Experiments in a type 2 diabetes model using a non-stimulating control device

To examine the effect of the glucose the fuel cell consumes while generating power, the team left a working fuel cell in the non-stimulating device as well. In this configuration, only the output to the stimulation electrode was physically disconnected. Diabetic rats were divided into a stimulation-active group, a stimulation-inactive group and a control group with no implanted device, and healthy rats were added for a four-group comparison. According to the authors, there were no statistically significant differences in body weight or blood glucose between the stimulation-inactive group and the diabetic control group without a device. The authors therefore consider it unlikely that glucose consumption from power generation alone explains the changes observed in the stimulation-active group.

The experiments used male Sprague-Dawley rats that were 8 weeks old at purchase. The type 2 diabetes model was created with a high-fat diet, and the rats were switched to standard chow before the intervention. While feeding the rats standard chow, the researchers measured non-fasting blood glucose, body weight and food intake in seven rats per group over six weeks. On day 42, tolerance tests were performed on five rats per group. The glucose tolerance test examined the course of blood glucose after glucose administration, and the insulin tolerance test examined the blood glucose response after insulin administration. The authors report that in both tests the stimulation-active group responded better than the diabetic control group without a device and the stimulation-inactive group, and that both glucose tolerance and insulin sensitivity improved.

The team also investigated the pathways through which nerve stimulation might affect blood glucose. In addition to a comparison with matched food intake, they performed interventions that block the vagus nerve pathway and that inhibit a receptor involved in suppressing inflammation. The authors suggest that both changes in food intake and inflammation-related pathways may have contributed to the observed effects.

AD

Blood glucose tracked for eight weeks in a type 1 diabetes model

In the type 1 diabetes model, pancreatic beta cells, which secrete insulin, were damaged with a drug called streptozotocin. This model chemically induces insulin deficiency in male rats, and its mechanism differs from human type 1 diabetes, in which beta cells are damaged by autoimmunity. Four main groups of five rats each were followed for eight weeks, and blood glucose trajectories differed clearly depending on whether stimulation was delivered. On day 56, mean blood glucose was 194.4 mg/dL in the stimulation-active group and 439.0 mg/dL in the stimulation-inactive group.

Group Day 0 Day 56
Healthy control 129.8 ± 19.8 130.6 ± 22.1
Diabetic control without device 352.4 ± 19.0 440.2 ± 16.7
Stimulation-inactive 346.2 ± 28.1 439.0 ± 17.6
Stimulation-active 357.6 ± 13.9 194.4 ± 37.3

The table shows mean ± sample standard deviation of blood glucose (mg/dL) for five rats per group. The arithmetic means and sample standard deviations were calculated from the source data for Fig. 6a (Sheet1, rows 3 and 21), which corresponds to Nature's Figure 6, and rounded to one decimal place.

Mean blood glucose in the stimulation-active group fell from day 0, whereas it rose in the diabetic control group without a device and in the stimulation-inactive group. Even so, mean blood glucose on day 56 in the stimulation-active group was still higher than in the healthy control group. In tissue analyses, the team examined immune cells in the spleen and pancreas at week 3 and beta cell mass in the pancreas at week 8, using five rats per group. At week 8, insulin-positive beta cell mass was greater in the stimulation-active group than in the diabetic control group without a device and the stimulation-inactive group. The authors suggest that beta cell proliferation may have contributed, and that reduced inflammation may have helped preserve beta cells.

Does the device keep working after blood glucose falls?

In the type 2 diabetes model, the team reports that improved glycemic control continued for about a month after stimulation ended. The rats were again fed standard chow during this period. In a separate experiment, the researchers examined over time the proportion of devices that kept operating, and according to the supplementary figure it used 10 independent devices. The authors cite both falling blood glucose and declining output of the fuel cell itself as reasons the active devices stopped producing stimulation output after several weeks. If blood glucose or fuel cell output falls and the terminal voltage no longer meets the trigger condition, the next stimulation does not begin.

The team randomly assigned rats to groups and blinded personnel to group information in behavioral assessments and quantitative tissue analyses. The number of animals per group was determined from preliminary experiments and similar studies. However, formal blinding was not used for measurements such as blood glucose. Within the research team, each experiment was repeated at least three times with similar results. Independent replication by another research group has not been confirmed, and the study did not evaluate safety or efficacy in humans.

To assess long-term performance, an important question is how the device responds if blood glucose rises again after stimulation has stopped. How much the voltage response to glucose concentration varies between individuals, and whether calibration drifts over long-term use, also remain to be tested. Cross-checking blood glucose from blood samples, the fuel cell's terminal voltage and the stimulation output to the electrode would show whether the device can detect a renewed rise in blood glucose and resume stimulation.