Browse the corpus

Brain activity associated with breakthrough food preoccupation in an individual on tirzepatide. Obesity and related conditions are associated with distressing food preoccupation that often culminates in dysregulated eating behaviors. Incretin-based therapies can reduce excessive weight in obesity, but their impact on dysregulated eating behaviors remains largely unexamined. Understanding how these pharmacologics engage the brain's mesolimbic circuitry may inform the expansion of their therapeutic potential. We report a rare, first-in-human exploration of the physiological action of these therapies by examining the electrophysiology directly within the human nucleus accumbens. After a short-term course of tirzepatide, the patient-participant exhibited increased severe food preoccupation episodes, which were preceded by an increased delta-theta frequency (≤7 Hz) power in the nucleus accumbens region. We propose that the effects of an incretin-based therapy (tirzepatide) on food preoccupation may be associated with modulation of aberrant activity within this key hub of human mesolimbic circuitry.

Eating behaviors are regulated by homeostatic (for example, eating based on energy needs) and hedonic (for example, eating based on pleasure) processes, involving the hypothalamic and brain stem circuits as a hub for the former, and a mesolimbic circuit (including the nucleus accumbens (NAc)) for the latter1–3. These systems are highly interactive and are further influenced by other intermediate brain regions to include the complex motivational processes of ingestion1,4,5. As such, the distinction between homeostatic and hedonic eating as entirely separate entities is increasingly viewed as a conceptual oversimplification1,4. There is a preponderance of receptors of incretin-based therapies (for example, glucose-dependent insulinotropic polypeptide (GIP) and glucagon-like peptide-1 (GLP-1)-based receptor agonists) in central nervous system nuclei, including the hypothalamus and NAc, which regulate energy balance and reward processing6,7, underlying their therapeutic potential for obesity and type 2 diabetes8–10.

ample, glucose-dependent insulinotropic polypeptide (GIP) and glucagon-like peptide-1 (GLP-1)-based receptor agonists) in central nervous system nuclei, including the hypothalamus and NAc, which regulate energy balance and reward processing6,7, underlying their therapeutic potential for obesity and type 2 diabetes8–10. However, the physiological action of incretin-based therapies specifically on the mesolimbic circuitry to alter human eating behaviors remains unexplored. In concordance with homeostatic processes, the mesolimbic system contributes to food-related motivation and its dysregulation underlies disturbances in food preoccupation (that is, heightened or persistent reactivity to food cues11). Food preoccupation is often associated with dysregulated eating behaviors, ranging from loss-of-control eating (that is, eating with a subjective feeling of loss of control and associated distress) to binge eating (that is, the most extreme bout of loss-of-control eating). These debilitating symptoms affect up to 60% of patients with obesity and related eating disorders11–13. Although aberrations in the mesocorticolimbic system, hypothalamus and brain stem are implicated in both obesity and binge eating disorder3,4,14, patients with binge eating disorder may be more prone to these symptoms than those with obesity in the absence of binge eating disorder because of the degree of reward hypersensitivity and food impulsivity involving mesocorticolimbic dysregulation14–17. While incretin-based therapies have exhibited some promise in ameliorating food preoccupation and dysregulated eating behaviors9–11, early data suggest a tolerance effect for food preoccupation18,19. Direct measures of neural activity could yield insights into how incretin-based therapies engage the mesolimbic circuitry and help broaden their therapeutic scope to related eating disorders, possibly by identifying a target engagement biomarker (that is, a neural signal that reflects functional modulation of a brain region in response to treatment).

ty could yield insights into how incretin-based therapies engage the mesolimbic circuitry and help broaden their therapeutic scope to related eating disorders, possibly by identifying a target engagement biomarker (that is, a neural signal that reflects functional modulation of a brain region in response to treatment). Intracranial electroencephalography (iEEG), acquired using implanted depth electrodes, provides a rare opportunity to directly measure neural activity within human brain circuitry. iEEG has recently been used to identify electrographic biomarkers of neuropsychiatric disorders20–22. An ongoing early feasibility trial (ClinicalTrials.gov registration NCT03868670) has recruited participants with treatment-refractory obesity and loss-of-control eating to identify related iEEG activity23. Participants’ dysregulated eating episodes were classified as loss-of-control eating rather than binge eating because they did not consistently meet the criteria for eating an objectively large amount, probably because of their restricted gastric volume after bariatric surgery. We previously reported iEEG activity within a low-frequency band (2–8 Hz), which ramped up during periods of loss-of-control eating from previous participants21. In the present study, we used a first-of-its-kind opportunity to report a case study provided by participant 3 (Fig. 1a–d) to investigate an electrographic biomarker associated with the frequency of severe food preoccupation while taking tirzepatide, using preliminary findings from participants 1 and 2 as a reference. The preliminary findings presented in this article for participants 1 and 2 differ from a previous report21 that aimed to guide responsive deep brain stimulation (rDBS) with specificity for hedonic states. Instead, in the present study, we focused on food preoccupation, reflecting a conceptual shift that dysregulated eating behaviors are a result of disruption in both hedonic and homeostatic processes (for more information, see Methods)1,4,11.Fig. 1Background information of participant 3 and association of increased delta–theta power in the ventral NAc with severe food preoccupation in participants 1 and 2.a, Two quadripolar depth electrodes were placed bilaterally in the ventral NAc of participant 3 with a neurostimulator fully implanted subgaleally in the skull.

ackground information of participant 3 and association of increased delta–theta power in the ventral NAc with severe food preoccupation in participants 1 and 2.a, Two quadripolar depth electrodes were placed bilaterally in the ventral NAc of participant 3 with a neurostimulator fully implanted subgaleally in the skull. b, Anatomical figure of participant 3 (view from posterior to anterior): three-dimensional rendering of DBS electrodes and their position in basal forebrain structures in the participant’s native space: ventral NAc, magenta; dorsal NAc, white; putamen, green; caudate, blue; anterior limb of internal capsule (ALIC), light pink. c, Anatomical magnetic resonance imaging of participant 3 (Supplementary Fig. 1). L, left hemisphere; R, right hemisphere. Prior participants in this ongoing trial had similar electrode placement21. d, Timeline of crucial events for participant 3. A larger illustration and specifics for data collection and use are described in the Methods and Supplementary Fig. 2. e, Participant 1 (left ventral NAc): power spectrum (mean ± s.e.m.) during the biomarker discovery (left) and stimulation (right) phases when the participant was relaxing (control, blue) or in a severe food preoccupation state (pink). The bottom black lines indicate frequencies with statistically significant differences in power values between the control and severe food preoccupation conditions after two-sided permutation testing (P < 0.05) with cluster correction. f, Data from participant 2 (left ventral NAc), using the same format as in e. g, Participant 1 (left ventral NAc): delta–theta band (≤7 Hz) power during the biomarker discovery (left) and stimulation (right) phases when the participant was relaxing (control, blue) or in a severe food preoccupation state (pink). The center line of the box indicates the median; the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The individual points outside the whiskers are considered as outliers. The top black line with a single asterisk in the box plot shows statistically significant differences between two conditions after two-sided permutation testing (*P < 0.05). h, Data from participant 2 (left ventral NAc), using the same format as in g. For more information, including the right NAc, see Supplementary Fig. 3 and Supplementary Tables 1 and 2. Illustrations in a and d created using BioRender.com.

ifferences between two conditions after two-sided permutation testing (*P < 0.05). h, Data from participant 2 (left ventral NAc), using the same format as in g. For more information, including the right NAc, see Supplementary Fig. 3 and Supplementary Tables 1 and 2. Illustrations in a and d created using BioRender.com. a, Two quadripolar depth electrodes were placed bilaterally in the ventral NAc of participant 3 with a neurostimulator fully implanted subgaleally in the skull. b, Anatomical figure of participant 3 (view from posterior to anterior): three-dimensional rendering of DBS electrodes and their position in basal forebrain structures in the participant’s native space: ventral NAc, magenta; dorsal NAc, white; putamen, green; caudate, blue; anterior limb of internal capsule (ALIC), light pink. c, Anatomical magnetic resonance imaging of participant 3 (Supplementary Fig. 1). L, left hemisphere; R, right hemisphere. Prior participants in this ongoing trial had similar electrode placement21. d, Timeline of crucial events for participant 3. A larger illustration and specifics for data collection and use are described in the Methods and Supplementary Fig. 2. e, Participant 1 (left ventral NAc): power spectrum (mean ± s.e.m.) during the biomarker discovery (left) and stimulation (right) phases when the participant was relaxing (control, blue) or in a severe food preoccupation state (pink). The bottom black lines indicate frequencies with statistically significant differences in power values between the control and severe food preoccupation conditions after two-sided permutation testing (P < 0.05) with cluster correction. f, Data from participant 2 (left ventral NAc), using the same format as in e. g, Participant 1 (left ventral NAc): delta–theta band (≤7 Hz) power during the biomarker discovery (left) and stimulation (right) phases when the participant was relaxing (control, blue) or in a severe food preoccupation state (pink). The center line of the box indicates the median; the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The individual points outside the whiskers are considered as outliers. The top black line with a single asterisk in the box plot shows statistically significant differences between two conditions after two-sided permutation testing (*P < 0.05). h, Data from participant 2 (left ventral NAc), using the same format as in g. For more information, including the right NAc, see Supplementary Fig.

he top black line with a single asterisk in the box plot shows statistically significant differences between two conditions after two-sided permutation testing (*P < 0.05). h, Data from participant 2 (left ventral NAc), using the same format as in g. For more information, including the right NAc, see Supplementary Fig. 3 and Supplementary Tables 1 and 2. Illustrations in a and d created using BioRender.com.

he top black line with a single asterisk in the box plot shows statistically significant differences between two conditions after two-sided permutation testing (*P < 0.05). h, Data from participant 2 (left ventral NAc), using the same format as in g. For more information, including the right NAc, see Supplementary Fig. 3 and Supplementary Tables 1 and 2. Illustrations in a and d created using BioRender.com. We analyzed ambulatory iEEG recordings from the NAc in participants 1 and 2 during the biomarker discovery phase. During this phase for both participants, the delta–theta band (≤7 Hz) power in the ventral NAc during the severe food preoccupation states was significantly higher than that of control states in both the left hemisphere (permutation testing, P = 2.1035 × 10−6 (participant 1) and 2.48443 × 10−11 (participant 2)) and right hemisphere (P = 4.7013 × 10−5 (participant 1) and 4.2414 × 10−8 (participant 2)) (Fig. 1e–h left, Supplementary Fig. 3 and Supplementary Tables 1 and 2). Moreover, both participants reported a high number of severe food preoccupation episodes (that is, moments of feeling intense food noise; Supplementary Fig. 4). After the biomarker discovery phase, and thus after a few months of responsive stimulation triggered by this biomarker detection, the delta–theta band power in the ventral NAc during severe food preoccupation states was indistinguishable from that of control states in both hemispheres of both participants (left hemisphere: P = 0.0519 (participant 1) and 0.6433 (participant 2); right hemisphere: P = 0.5129 (participant 1) and 0.4227 (participant 2); Fig. 1e–h right, Supplementary Fig. 3 and Supplementary Tables 1 and 2). The number of severe food preoccupation episodes during the stimulation phase also decreased (Supplementary Fig. 4). Thus, including the previous findings21,24, we postulated that the delta–theta band power (≤7 Hz) could serve as a biomarker reflecting a state of heightened propensity for severe food preoccupation, as observed in the changes of the number of episodes. We hypothesized that the effects of tirzepatide on food preoccupation are related to modulation of this delta–theta band biomarker in the NAc, a key hub of the mesolimbic reward circuitry where incretin receptors are also expressed6,7.

for severe food preoccupation, as observed in the changes of the number of episodes. We hypothesized that the effects of tirzepatide on food preoccupation are related to modulation of this delta–theta band biomarker in the NAc, a key hub of the mesolimbic reward circuitry where incretin receptors are also expressed6,7. Unlike participants 1 and 2, participant 3 exhibited a lengthy absence of severe food preoccupation in months 2–4 after surgery (Fig. 2a and Supplementary Fig. 7), coinciding with a tirzepatide dose increase that occurred before surgical implantation. During this period, the delta–theta band (≤7 Hz) power during the severe food preoccupation states was indistinguishable from that of control states (Fig. 2b, under the green bar, and Fig. 2c,d) in both the left (permutation testing, P = 0.8105) and right (P = 0.1011) hemispheres. There were also no differences in other higher frequencies. These findings are markedly different from those from participants 1 and 2 (Fig. 1e–h, left) and our prior reports21,24. The length of this quiescent period (months 2–4) was later corroborated by using an algorithm that identified the transition point corresponding to the most pronounced change in power values within the delta–theta frequency band (≤7 Hz; Supplementary Fig. 8; see Methods for more detail)25,26.Fig. 2Association of increased delta–theta power in the ventral NAc with severe food preoccupation in participant 3.a, Number of severe food preoccupation episodes per month after surgery (excluding month 1 after surgery because of a potential implantation effect27; month 4 had no episode). The green block denotes months 2–4. b, Participant 3 (left ventral NAc): spectrograms from magnet swipes for the control (left) and severe food preoccupation (right) conditions during the biomarker discovery phase (months 2–7). These spectrograms show power values per frequency from 1 to 10 Hz when the participant was relaxing (control; n = 33) or when the participant was in a severe food preoccupation state (n = 10). Each column corresponds to one magnet swipe episode, which is a participant-triggered iEEG recording (90 or 180 s). The red dashed line denotes the end of month 4 and the beginning of month 5 (see the green and pink bars above each spectrogram). For the results from the right ventral NAc, see Supplementary Fig. 5.

10). Each column corresponds to one magnet swipe episode, which is a participant-triggered iEEG recording (90 or 180 s). The red dashed line denotes the end of month 4 and the beginning of month 5 (see the green and pink bars above each spectrogram). For the results from the right ventral NAc, see Supplementary Fig. 5. Note that the apparent similarity in the number of severe food preoccupation episodes across months 2–4 versus months 5–7 arises from technical limitations (for example, device storage or trigger failure). Data reflect the number of available electrophysiological recordings. During months 2–4, the reported number of severe food preoccupation episodes was five but only four iEEG data were available, as shown on the right (under the green bar). Likewise, during months 5–7, the reported number was 11 but only six iEEG recordings were collected, as shown on the right (under the pink bar). c,d, Participant 3 (ventral NAc): power spectrum (mean ± s.e.m.) (c) and delta–theta band (≤7 Hz) power (d) when the participant was relaxing (control, blue) or in a severe food preoccupation state (pink) during months 2–4. c, The bottom black line shows statistically significant differences in power values between the control and severe food preoccupation (two-sided permutation testing (P < 0.05) conditions with cluster correction). There was no significant difference except at 7.2–8.8 Hz where the power value from severe food preoccupation was lower than the control. d, The center line of the box indicates the median; the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The individual points outside the whiskers are considered as outliers. The top black line with a single asterisk shows statistically significant differences between two conditions after two-sided permutation testing (*P < 0.05). During months 2–4, there was also a 7% decrease in body weight relative to the baseline before surgery (138–128 kg; Supplementary Fig. 6). e,f, Power spectrum (mean ± s.e.m.) (e) and delta–theta band (≤7 Hz) power (f) from months 5–7, formatted as in c,d. c–f, For more information, see Supplementary Table 3.

P < 0.05). During months 2–4, there was also a 7% decrease in body weight relative to the baseline before surgery (138–128 kg; Supplementary Fig. 6). e,f, Power spectrum (mean ± s.e.m.) (e) and delta–theta band (≤7 Hz) power (f) from months 5–7, formatted as in c,d. c–f, For more information, see Supplementary Table 3. a, Number of severe food preoccupation episodes per month after surgery (excluding month 1 after surgery because of a potential implantation effect27; month 4 had no episode). The green block denotes months 2–4. b, Participant 3 (left ventral NAc): spectrograms from magnet swipes for the control (left) and severe food preoccupation (right) conditions during the biomarker discovery phase (months 2–7). These spectrograms show power values per frequency from 1 to 10 Hz when the participant was relaxing (control; n = 33) or when the participant was in a severe food preoccupation state (n = 10). Each column corresponds to one magnet swipe episode, which is a participant-triggered iEEG recording (90 or 180 s). The red dashed line denotes the end of month 4 and the beginning of month 5 (see the green and pink bars above each spectrogram). For the results from the right ventral NAc, see Supplementary Fig. 5. Note that the apparent similarity in the number of severe food preoccupation episodes across months 2–4 versus months 5–7 arises from technical limitations (for example, device storage or trigger failure). Data reflect the number of available electrophysiological recordings. During months 2–4, the reported number of severe food preoccupation episodes was five but only four iEEG data were available, as shown on the right (under the green bar). Likewise, during months 5–7, the reported number was 11 but only six iEEG recordings were collected, as shown on the right (under the pink bar). c,d, Participant 3 (ventral NAc): power spectrum (mean ± s.e.m.) (c) and delta–theta band (≤7 Hz) power (d) when the participant was relaxing (control, blue) or in a severe food preoccupation state (pink) during months 2–4. c, The bottom black line shows statistically significant differences in power values between the control and severe food preoccupation (two-sided permutation testing (P < 0.05) conditions with cluster correction). There was no significant difference except at 7.2–8.8 Hz where the power value from severe food preoccupation was lower than the control.

statistically significant differences in power values between the control and severe food preoccupation (two-sided permutation testing (P < 0.05) conditions with cluster correction). There was no significant difference except at 7.2–8.8 Hz where the power value from severe food preoccupation was lower than the control. d, The center line of the box indicates the median; the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The individual points outside the whiskers are considered as outliers. The top black line with a single asterisk shows statistically significant differences between two conditions after two-sided permutation testing (*P < 0.05). During months 2–4, there was also a 7% decrease in body weight relative to the baseline before surgery (138–128 kg; Supplementary Fig. 6). e,f, Power spectrum (mean ± s.e.m.) (e) and delta–theta band (≤7 Hz) power (f) from months 5–7, formatted as in c,d. c–f, For more information, see Supplementary Table 3.

P < 0.05). During months 2–4, there was also a 7% decrease in body weight relative to the baseline before surgery (138–128 kg; Supplementary Fig. 6). e,f, Power spectrum (mean ± s.e.m.) (e) and delta–theta band (≤7 Hz) power (f) from months 5–7, formatted as in c,d. c–f, For more information, see Supplementary Table 3. In contrast, during months 5–7, the delta–theta band biomarker emerged and the participant began to report breakthroughs in severe food preoccupation despite the maximum dose of tirzepatide (Fig. 2a,b, under the pink bar; increased power values in the delta–theta band (≤7 Hz) are noted in yellow). During this period, a prominent delta–theta oscillatory waveform was observed (Supplementary Fig. 9). Moreover, the delta–theta band (≤7 Hz) power from severe food preoccupation states was significantly higher than that of control states in the left hemisphere (permutation testing, P = 1.5310 × 10−22) and right hemisphere (P = 1.0887 × 10−6) (Fig. 2e,f). After the change in biomarker, the number of severe food preoccupation episodes increased to seven per month (Fig. 2a).

re food preoccupation states was significantly higher than that of control states in the left hemisphere (permutation testing, P = 1.5310 × 10−22) and right hemisphere (P = 1.0887 × 10−6) (Fig. 2e,f). After the change in biomarker, the number of severe food preoccupation episodes increased to seven per month (Fig. 2a). We present a unique case that provided a serendipitous opportunity to investigate the associated electrophysiology of an incretin-based pharmacologic in the human NAc. A profoundly low number of severe food preoccupation episodes (and a reduction in body weight) during months 2–4 (excluding month 1 after surgery because of a potential implantation effect27) was consistent with a concomitant increase in tirzepatide for diabetes management11,28. Importantly, this lengthy absence of severe food preoccupation after surgery contrasted with participants 1 and 2 (Supplementary Fig. 7). During this period, participant 3 also exhibited an absence of the expected delta–theta band (≤7 Hz) biomarker in the ventral NAc. The delta–theta band biomarker emerged during months 5–7 in participant 3, which preceded a breakthrough in severe food preoccupation despite tirzepatide18,19. Therefore, the delta–theta band power during severe food preoccupation states may reflect a state of heightened propensity for severe food preoccupation, as observed in the changes of the number of episodes. The biomarker is present (or increased compared to control) when severe food preoccupation occurs more frequently and is absent (or indistinguishable from control) when severe food preoccupation occurs less frequently.

tened propensity for severe food preoccupation, as observed in the changes of the number of episodes. The biomarker is present (or increased compared to control) when severe food preoccupation occurs more frequently and is absent (or indistinguishable from control) when severe food preoccupation occurs less frequently. These preliminary results suggest that tirzepatide administration may be associated with the modulation of the delta–theta band (≤7 Hz) biomarker in the human NAc. All three participants exhibited a substantial increase in delta–theta power during severe food preoccupation states in both hemispheres (that is, six hemispheres) during the biomarker discovery phase. For participant 3, there was a temporal lag between the emergence of the biomarker in month 5 and the most severe breakthrough of food preoccupation in month 7, which could be specific to tirzepatide. In particular, a supplementary cross-correlation analysis suggested a 7-week lag (Supplementary Fig. 10). Additionally, the effect size of the power difference between severe food preoccupation and control states was more pronounced, and the transition point was only identified in the left NAc of participant 3, suggesting a potential laterality bias.

ross-correlation analysis suggested a 7-week lag (Supplementary Fig. 10). Additionally, the effect size of the power difference between severe food preoccupation and control states was more pronounced, and the transition point was only identified in the left NAc of participant 3, suggesting a potential laterality bias. Our findings raise the possibility that this delta–theta band oscillation could serve as a target engagement biomarker, but its relationship to food preoccupation warrants more controlled investigation. Moreover, the phenomenon of a biomarker preceding actual behavioral change has been reported previously in the context of other behaviors relevant to psychiatric illness22,29. Thus, the early findings reported in this study could provide the foundations of developing such a biomarker-based approach for tirzepatide administration for dysregulated eating, a strategy garnering interest for neuropsychiatric disorders30,31. Although the invasive nature of monitoring this biomarker may limit scalability, the results reported in this study could inform preclinical studies given that the low-frequency nature of this biomarker is conserved at least when recorded from the NAc across mouse and human studies24. Further, noninvasive strategies can be developed to capture relevant brain dynamics; a parallel can be found in patients with Parkinson disease, where a prominent beta band signal has been detected both within the subthalamic nucleus and via scalp EEG32,33. Thus, a biomarker-based approach holds promise as a strategy to optimize incretin-based therapies for food preoccupation18,19.

vant brain dynamics; a parallel can be found in patients with Parkinson disease, where a prominent beta band signal has been detected both within the subthalamic nucleus and via scalp EEG32,33. Thus, a biomarker-based approach holds promise as a strategy to optimize incretin-based therapies for food preoccupation18,19. This study has some limitations. As this is a single uncontrolled case study, it is unknown whether the findings will generalize, for example, to other incretin-based therapies. Given the potential compulsive component in patients exhibiting dysregulated eating behaviors, the delta–theta biomarker may not be applicable to the broader population with obesity. In addition, we cannot determine whether the effects of tirzepatide are due to direct action in the NAc or identify which incretin receptor is involved (that is, GLP-1 or GIP). Lastly, analyzing electrophysiological data in the absence of tirzepatide use was not feasible, as it was part of the ongoing diabetes management of participant 3. Thus, the possibility that the observed electrophysiological changes may be attributable to other confounding factors (for example, postoperative recovery, elapsed time since surgery or unrelated behavioral changes) cannot be excluded. Discontinuation of tirzepatide could pose both clinical and ethical challenges, but future studies may be designed to directly investigate the physiological impact of incretin-based therapies on brain reward circuitry.

Written informed consent was obtained from participant 3 to participate in an early feasibility trial of ‘responsive deep brain stimulation (rDBS) for patients with treatment-refractory obesity and loss-of-control eating’ (ClinicalTrials.gov registration NCT03868670)23. This trial was approved by the U.S. Food and Drug Administration (FDA) and the institutional review board (IRB) of the University of Pennsylvania (IRB no. 850489 and investigational device exemption no. G180079). Participant 1 is a 51-year-old woman with severe treatment-resistant obesity (body mass index (BMI) = 46.5 kg m−2) and distressing food preoccupation despite bariatric surgery. After Roux-en-Y gastric bypass surgery, she lost 52.16 kg but she gradually regained the weight, returning to her before surgery weight at the time of enrollment. She had the comorbidities of neoplasm, lower back pain, kyphoscoliosis/scoliosis, hypertension, esophageal reflux, dyslipidemia, complicated migraine and anxiety at the time of enrollment. Participant 2 is a 61-year-old woman with severe treatment-resistant obesity (BMI = 47.1 kg m−2) and distressing food preoccupation despite bariatric surgery. After Roux-en-Y gastric bypass surgery, she lost 68.95 kg but regained the weight and was back to within 9% of her before surgery weight at the time of enrollment. She had the comorbidity of migraine at the time of enrollment. After bariatric surgery and before the enrollment, both participants tried many other weight loss strategies, including behavior therapy, support groups and medication, which were unsuccessful.

thin 9% of her before surgery weight at the time of enrollment. She had the comorbidity of migraine at the time of enrollment. After bariatric surgery and before the enrollment, both participants tried many other weight loss strategies, including behavior therapy, support groups and medication, which were unsuccessful. Both participants reported severe food preoccupations particularly related to emotional-related and stress-related triggers that often led to loss-of-control eating episodes (approximately five episodes per week and four episodes per week, respectively, as measured using the Eating Disorder Examination34). Neither participants reported previous testing or diagnosis regarding monogenic obesity, which is not part of our clinical standard. These participants were enrolled at Stanford University and the study was approved by Stanford University’s IRB (IRB no. 46563) at that time. Informed consent was obtained from both participants (please refer to ref. 21 for more details).

iagnosis regarding monogenic obesity, which is not part of our clinical standard. These participants were enrolled at Stanford University and the study was approved by Stanford University’s IRB (IRB no. 46563) at that time. Informed consent was obtained from both participants (please refer to ref. 21 for more details). Participant 3 is a 60-year-old woman with severe treatment-resistant obesity (BMI = 46.1 kg m−2) despite bariatric surgery and comorbid type 2 diabetes. She did not report any previous testing or diagnosis regarding monogenic obesity. She presented to us reporting substantial distress from food preoccupation. Her frequent food preoccupations led to unwanted eating behaviors, including many loss-of-control eating episodes. Before laparoscopic Roux-en-Y gastric bypass surgery, she weighed 154 kg and she had cravings for calorically dense food choices. After her bariatric surgery, she reached a nadir weight of 115 kg (BMI = 38.7 kg m−2), but near the time of enrollment, her weight had increased to 137 kg (BMI = 46.1 kg m−2). She stated that she was often preoccupied with thoughts of food, which led to ordering a meal out or to continual snacking, even though she wanted to resist. Her preoccupation focused on both sweet and salty foods, such as prepackaged cupcakes and roast beef sandwiches with french fries. She reported 19 loss-of-control episodes in the previous month on study entry. The participant endorsed eating until uncomfortably full, eating large amounts when she was not hungry and feeling guilty after these episodes, with high levels of distress associated with them.

s and roast beef sandwiches with french fries. She reported 19 loss-of-control episodes in the previous month on study entry. The participant endorsed eating until uncomfortably full, eating large amounts when she was not hungry and feeling guilty after these episodes, with high levels of distress associated with them. The participant fulfilled all eligibility criteria of the trial, which are mainly: (1) BMI = 40–60 kg m−2; (2) unsuccessful intervention with of bariatric surgery, behavioral therapy and pharmacological therapy for dysregulated eating behavior and weight loss and (3) loss-of-control eating episodes at least four times per week. Importantly, these included unsuccessful use of a GLP-1 receptor agonist (dulaglutide), which resulted in no relief in her weight or food preoccupation. She was switched to a GLP-1-GIP dual receptor agonist (tirzepatide) for its FDA-approved indication treating type 2 diabetes. There was no reported impact on weight and food preoccupation at 7.5 mg per week of tirzepatide at the time of baseline assessment. She reported temporary weight loss followed by subsequent return to her initial weight with no reported impact to her food preoccupation and loss-of-control eating episodes. As this agent was intended for treating her type 2 diabetes, the participant was enrolled into the study and underwent implantation of the rDBS system (NeuroPace) bilaterally in the NAc. The patient increased the tirzepatide dose to optimize her diabetes management as suggested by the clinical team given the known risk of this medical comorbidity on surgical outcomes, particularly infection, given the medical device implantation. At the time of implantation, the patient was receiving 12.5 mg per week and it was further increased to 15 mg per week after approximately 4 months for continued optimization of glucose control. She had comorbidities of hypertension, hyperlipidemia, coronary artery disease, nonalcoholic steatohepatitis, irritable bowel, migraine and asthma at the time of enrollment.

eiving 12.5 mg per week and it was further increased to 15 mg per week after approximately 4 months for continued optimization of glucose control. She had comorbidities of hypertension, hyperlipidemia, coronary artery disease, nonalcoholic steatohepatitis, irritable bowel, migraine and asthma at the time of enrollment. The surgical procedure has been reported previously23,35. Briefly, probabilistic tractography was used to guide surgical targeting of the NAc as described previously36. On the day of surgery, implantation of bilateral DBS electrodes in the NAc was performed under awake conditions as per our standard institutional practice. Using a personalized appetitive provocation task, microelectrode recording was performed intraoperatively to identify single-unit or multiunit appetitive neural activity35. After confirmation of the target with electrophysiology and imaging, a quadripolar depth electrode was placed and macroelectrode monopolar stimulation mapping was conducted to confirm positive effects and no adverse effects. After securing the DBS electrodes, the electrodes were connected to the neurostimulator pulse generator, which was placed in the right parietal skull region of the patient.

polar depth electrode was placed and macroelectrode monopolar stimulation mapping was conducted to confirm positive effects and no adverse effects. After securing the DBS electrodes, the electrodes were connected to the neurostimulator pulse generator, which was placed in the right parietal skull region of the patient. iEEG recordings were acquired from the FDA-approved rDBS device (NeuroPace) as reported previously21. An rDBS device is different from a regular DBS device in that it stimulates only when it detects a predefined biomarker rather than stimulating continuously. Neural recordings acquired from the rDBS device were used to identify biomarkers differentiating severe food preoccupation swipes from control swipes. iEEG data were recorded at a 250-Hz sampling rate and bipolar re-referenced online. We used data from channels 1 and 3, which were referred to as the data from the left and right ventral NAc in this article. Four electrode contacts were located in the following order: (1) ventral NAc (the most ventral contact; presumed NAc shell36); (2) dorsal NAc (presumed NAc core36); (3) and (4) ALIC. For all participants, an electrode in the left hemisphere, one channel (channel 1) was bipolar re-referenced between contacts 1 and 3; the other channel (channel 2) was between contacts 2 and 4. Likewise, an electrode in the right hemisphere, one channel (channel 3) was bipolar re-referenced between contacts 1 and 3, and the other channel (channel 4) was between contacts 2 and 4.

channel (channel 1) was bipolar re-referenced between contacts 1 and 3; the other channel (channel 2) was between contacts 2 and 4. Likewise, an electrode in the right hemisphere, one channel (channel 3) was bipolar re-referenced between contacts 1 and 3, and the other channel (channel 4) was between contacts 2 and 4. The biomarker discovery phase data of participants 1 and 2 overlaps with ambulatory data used in our previous report21. However, the focus and analytical approach in the previous report were fundamentally different from the current study. In the previous report, the analysis was centered around pure ‘craving’ in the absence of hunger, aiming to dissociate hedonic and homeostatic eating by stratifying data based on ‘craving’ and ‘hunger’ ratings. This allowed us to explore the NAc electrophysiology in states seemingly dominated by hedonic versus homeostatic derives, given that we were performing biomarker discovery to guide an rDBS. Thus, we hypothesized that stimulation would be more behaviorally specific to hedonic states.

d on ‘craving’ and ‘hunger’ ratings. This allowed us to explore the NAc electrophysiology in states seemingly dominated by hedonic versus homeostatic derives, given that we were performing biomarker discovery to guide an rDBS. Thus, we hypothesized that stimulation would be more behaviorally specific to hedonic states. However, in the current report, we focused on a distinct construct, that is, food preoccupation, as defined by ‘heightened and/or persistent reactivity to food cues’11. This encompasses a broader range of influences, including hunger and other sensorial, environmental and social cues11. The excessive food preoccupation observed in our participants reflected combined alterations in both homeostatic and hedonic eating rather than hedonic eating alone1. To better align with this conceptual shift, data were stratified to ‘severe food preoccupation’ only using the craving rating regardless of the hunger rating. This stratification included episodes with both high craving and hunger ratings.

ations in both homeostatic and hedonic eating rather than hedonic eating alone1. To better align with this conceptual shift, data were stratified to ‘severe food preoccupation’ only using the craving rating regardless of the hunger rating. This stratification included episodes with both high craving and hunger ratings. For participants 1 and 2, the bipolar reference montage was adjusted during early recording period and the stimulation safety testing period. We confined our analysis to the periods across all participants when the same recording montage was used. Thus, the biomarker discovery phase used for participant 1 was limited to data collected for approximately 1 month before final initiation of stimulation, which took place during months 8–9 after surgery. There was no apparent change in food preoccupation during this brief period of safety testing. To examine the effect of stimulation, we used a dataset collected from the moment immediately after the stimulation parameter had been set to a maximum dose to month 18 after surgery (that is, the stimulation phase). To reduce confounders to our electrophysiological analysis due to stimulation, we used magnet swipes that had no stimulation within the magnet swipe time window. The same was done for participant 2.

fter the stimulation parameter had been set to a maximum dose to month 18 after surgery (that is, the stimulation phase). To reduce confounders to our electrophysiological analysis due to stimulation, we used magnet swipes that had no stimulation within the magnet swipe time window. The same was done for participant 2. For participant 2, we defined the period from when electrophysiology data were collected between month 3 and month 6 after surgery before stimulation initiation as the ‘biomarker discovery phase’. To examine the effect of stimulation, we again used a dataset collected from the moment immediately after the stimulation parameter had been set to a maximum dose to month 18 after surgery (that is, the stimulation phase). For participant 3, stimulation was not delivered throughout the data acquisition reported in this article. Moreover, we excluded data from up to a month from the surgery date (month 1) because of confounding with an implantation effect, as we did for the prior participants, although implantation effects typically last less than a month in patients with Parkinson disease27. Thus, months 2–7 corresponded to the ‘biomarker discovery phase’. We limited this case study’s interim analysis to the 6-month recording phase planned by the investigational device exemption trial to avoid further trial-related confounders.

effects typically last less than a month in patients with Parkinson disease27. Thus, months 2–7 corresponded to the ‘biomarker discovery phase’. We limited this case study’s interim analysis to the 6-month recording phase planned by the investigational device exemption trial to avoid further trial-related confounders. With the rDBS system, participants can initiate iEEG recordings by swiping their magnet over the surgically implanted device under the scalp (Supplementary Fig. 2). The magnet swipe triggers the iEEG recordings, which record a preset length of time before and after the magnet swipe with a two-to-one ratio. For instance, if the preset length is 90 s, it records 60 s before the magnet swipe and 30 s after it. For this study, the length was set to 90 s (participants 2 and 3) or 180 s (participants 1 and 3). For the control condition of participants 1 and 2, data were automatically recorded at a preset time (12:00 for participant 1 and 17:00 for participant 2), the time participants answered that they were most likely to be at rest. This was to reduce their study burdens. For participant 3, the control condition consisted of magnet swipes collected when she was relaxing and not feeling food cravings. For control episodes from scheduled recordings, we removed them if craving swipes were present near the scheduled recording time.

kely to be at rest. This was to reduce their study burdens. For participant 3, the control condition consisted of magnet swipes collected when she was relaxing and not feeling food cravings. For control episodes from scheduled recordings, we removed them if craving swipes were present near the scheduled recording time. For the food preoccupation condition, we asked participants to swipe the magnet when they felt cravings for food and before eating. Considering food preoccupation as ‘heightened or persistent reactivity to food cues’, we focused on the extent of food preoccupation regardless of hunger level, unlike our prior report21, because cues that could elicit food preoccupation include not only hunger and craving but also other sensorial, environmental and social aspects11. We asked participants to swipe when they were feeling a sense of craving to capture the most relevant moment of food preoccupation. For all magnet swipes, participants were asked to keep a magnet swipe diary, recording their craving, hunger and thirst levels, the extent they felt a loss of control and the extent they felt compelled to eat in a 5-point Likert scale (1 for none/not at all versus 5 for extreme/extremely). Therefore, magnet swipes with intense craving were classified as severe food preoccupation based on craving ratings per participant. We collected participants’ number of severe food preoccupation episodes per month through the magnet swipe diary but used ecological momentary assessment or verbal or written reports as a supplement if they forgot to keep them in the diary.

For the food preoccupation condition, we asked participants to swipe the magnet when they felt cravings for food and before eating. Considering food preoccupation as ‘heightened or persistent reactivity to food cues’, we focused on the extent of food preoccupation regardless of hunger level, unlike our prior report21, because cues that could elicit food preoccupation include not only hunger and craving but also other sensorial, environmental and social aspects11. We asked participants to swipe when they were feeling a sense of craving to capture the most relevant moment of food preoccupation. For all magnet swipes, participants were asked to keep a magnet swipe diary, recording their craving, hunger and thirst levels, the extent they felt a loss of control and the extent they felt compelled to eat in a 5-point Likert scale (1 for none/not at all versus 5 for extreme/extremely). Therefore, magnet swipes with intense craving were classified as severe food preoccupation based on craving ratings per participant. We collected participants’ number of severe food preoccupation episodes per month through the magnet swipe diary but used ecological momentary assessment or verbal or written reports as a supplement if they forgot to keep them in the diary. In the offline analysis, standard preprocessing techniques were conducted in MATLAB (v.R2022b) using the FieldTrip Toolbox37, which involved the application of a 1–124-Hz band-pass filter. Static spectral analysis was performed using a multi-taper method with four Slepian multi-tapers per epoch. These epochs were acquired using chunking magnet swipes every 5 s without overlaps. A full power spectrum (1–124 Hz) was acquired with 0.5-Hz frequency resolution; delta–theta band powers were acquired using a ≤7-Hz window. Epochs containing artifacts were removed if the artifact was larger than the six standard deviations of the data points of a corresponding magnet swipe. For the power spectrums in Fig. 2b, power values per frequency were averaged per a magnet swipe.

esolution; delta–theta band powers were acquired using a ≤7-Hz window. Epochs containing artifacts were removed if the artifact was larger than the six standard deviations of the data points of a corresponding magnet swipe. For the power spectrums in Fig. 2b, power values per frequency were averaged per a magnet swipe. Power spectral density values from the control and severe food preoccupation swipes were tested using two-sided permutation testing with 1,000 permutations and P = 0.05. Then, it underwent cluster correction using a cluster size threshold of the top 2.5%. Delta–theta band power values from two swipe conditions were tested using two-sided permutation testing with 10,000 permutations and P = 0.05. The transition point, which was predefined based on the emergence of the biomarker, was further validated using a model that identifies the transition point corresponding to the most pronounced change in power values within the delta–theta frequency band (≤7 Hz)25,26. More specifically, the transition point between two periods was determined by minimizing the total residual error, calculated from deviations of each time point from the root mean square estimate of the period to which it belongs. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Participant 1 is a 51-year-old woman with severe treatment-resistant obesity (body mass index (BMI) = 46.5 kg m−2) and distressing food preoccupation despite bariatric surgery. After Roux-en-Y gastric bypass surgery, she lost 52.16 kg but she gradually regained the weight, returning to her before surgery weight at the time of enrollment. She had the comorbidities of neoplasm, lower back pain, kyphoscoliosis/scoliosis, hypertension, esophageal reflux, dyslipidemia, complicated migraine and anxiety at the time of enrollment. Participant 2 is a 61-year-old woman with severe treatment-resistant obesity (BMI = 47.1 kg m−2) and distressing food preoccupation despite bariatric surgery. After Roux-en-Y gastric bypass surgery, she lost 68.95 kg but regained the weight and was back to within 9% of her before surgery weight at the time of enrollment. She had the comorbidity of migraine at the time of enrollment. After bariatric surgery and before the enrollment, both participants tried many other weight loss strategies, including behavior therapy, support groups and medication, which were unsuccessful.

The surgical procedure has been reported previously23,35. Briefly, probabilistic tractography was used to guide surgical targeting of the NAc as described previously36. On the day of surgery, implantation of bilateral DBS electrodes in the NAc was performed under awake conditions as per our standard institutional practice. Using a personalized appetitive provocation task, microelectrode recording was performed intraoperatively to identify single-unit or multiunit appetitive neural activity35. After confirmation of the target with electrophysiology and imaging, a quadripolar depth electrode was placed and macroelectrode monopolar stimulation mapping was conducted to confirm positive effects and no adverse effects. After securing the DBS electrodes, the electrodes were connected to the neurostimulator pulse generator, which was placed in the right parietal skull region of the patient.

iEEG recordings were acquired from the FDA-approved rDBS device (NeuroPace) as reported previously21. An rDBS device is different from a regular DBS device in that it stimulates only when it detects a predefined biomarker rather than stimulating continuously. Neural recordings acquired from the rDBS device were used to identify biomarkers differentiating severe food preoccupation swipes from control swipes. iEEG data were recorded at a 250-Hz sampling rate and bipolar re-referenced online. We used data from channels 1 and 3, which were referred to as the data from the left and right ventral NAc in this article. Four electrode contacts were located in the following order: (1) ventral NAc (the most ventral contact; presumed NAc shell36); (2) dorsal NAc (presumed NAc core36); (3) and (4) ALIC. For all participants, an electrode in the left hemisphere, one channel (channel 1) was bipolar re-referenced between contacts 1 and 3; the other channel (channel 2) was between contacts 2 and 4. Likewise, an electrode in the right hemisphere, one channel (channel 3) was bipolar re-referenced between contacts 1 and 3, and the other channel (channel 4) was between contacts 2 and 4.

With the rDBS system, participants can initiate iEEG recordings by swiping their magnet over the surgically implanted device under the scalp (Supplementary Fig. 2). The magnet swipe triggers the iEEG recordings, which record a preset length of time before and after the magnet swipe with a two-to-one ratio. For instance, if the preset length is 90 s, it records 60 s before the magnet swipe and 30 s after it. For this study, the length was set to 90 s (participants 2 and 3) or 180 s (participants 1 and 3). For the control condition of participants 1 and 2, data were automatically recorded at a preset time (12:00 for participant 1 and 17:00 for participant 2), the time participants answered that they were most likely to be at rest. This was to reduce their study burdens. For participant 3, the control condition consisted of magnet swipes collected when she was relaxing and not feeling food cravings. For control episodes from scheduled recordings, we removed them if craving swipes were present near the scheduled recording time.

kely to be at rest. This was to reduce their study burdens. For participant 3, the control condition consisted of magnet swipes collected when she was relaxing and not feeling food cravings. For control episodes from scheduled recordings, we removed them if craving swipes were present near the scheduled recording time. For the food preoccupation condition, we asked participants to swipe the magnet when they felt cravings for food and before eating. Considering food preoccupation as ‘heightened or persistent reactivity to food cues’, we focused on the extent of food preoccupation regardless of hunger level, unlike our prior report21, because cues that could elicit food preoccupation include not only hunger and craving but also other sensorial, environmental and social aspects11. We asked participants to swipe when they were feeling a sense of craving to capture the most relevant moment of food preoccupation. For all magnet swipes, participants were asked to keep a magnet swipe diary, recording their craving, hunger and thirst levels, the extent they felt a loss of control and the extent they felt compelled to eat in a 5-point Likert scale (1 for none/not at all versus 5 for extreme/extremely). Therefore, magnet swipes with intense craving were classified as severe food preoccupation based on craving ratings per participant.

We collected participants’ number of severe food preoccupation episodes per month through the magnet swipe diary but used ecological momentary assessment or verbal or written reports as a supplement if they forgot to keep them in the diary.

In the offline analysis, standard preprocessing techniques were conducted in MATLAB (v.R2022b) using the FieldTrip Toolbox37, which involved the application of a 1–124-Hz band-pass filter. Static spectral analysis was performed using a multi-taper method with four Slepian multi-tapers per epoch. These epochs were acquired using chunking magnet swipes every 5 s without overlaps. A full power spectrum (1–124 Hz) was acquired with 0.5-Hz frequency resolution; delta–theta band powers were acquired using a ≤7-Hz window. Epochs containing artifacts were removed if the artifact was larger than the six standard deviations of the data points of a corresponding magnet swipe. For the power spectrums in Fig. 2b, power values per frequency were averaged per a magnet swipe.

Power spectral density values from the control and severe food preoccupation swipes were tested using two-sided permutation testing with 1,000 permutations and P = 0.05. Then, it underwent cluster correction using a cluster size threshold of the top 2.5%. Delta–theta band power values from two swipe conditions were tested using two-sided permutation testing with 10,000 permutations and P = 0.05.

The transition point, which was predefined based on the emergence of the biomarker, was further validated using a model that identifies the transition point corresponding to the most pronounced change in power values within the delta–theta frequency band (≤7 Hz)25,26. More specifically, the transition point between two periods was determined by minimizing the total residual error, calculated from deviations of each time point from the root mean square estimate of the period to which it belongs.

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgments, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41591-025-04035-5.