The prevalence of overweight and obesity rises dramatically with age and is associated with increased morbidity and reduced quality of life (Kalyani et al., 2017). Hyperinsulinemia and insulin resistance are potential causes and consequences of obesity. Both are negatively affected by age and both play a pivotal role for the development of type 2 diabetes and other age-related diseases (Palmer and Kirkland, 2016). The mechanisms and directions of these interactions are under debate. For instance, it has long been assumed that insulin resistance in aging precedes the development of hyperinsulinemia, while recent data suggest a reverse direction (Janssen, 2021). Moreover, there is evidence from animal and human research of an weight-independent effect of aging on insulin sensitivity (Ehrhardt et al., 2019; Petersen et al., 2003).

Besides the importance of peripheral insulin for the glycemic control in the body, recent findings also highlight the role of insulin action in the brain for the metabolic and hedonic control of food intake (Kullmann et al., 2020a). Findings in rodents and humans indicate that, apart from signaling in hypothalamic neurocircuits regulating energy homeostasis, central insulin mediates non-homeostatic feeding for pleasure by signaling within mesolimbic reward circuits (Davis et al., 2010; Murray et al., 2014; Tiedemann et al., 2017). Specifically, insulin action in the ventral tegmental area (VTA) reduces hedonic feeding in rodents (Labouèbe et al., 2013; Mebel et al., 2012) and decreases hedonic value signals in the VTA and nucleus accumbens (NAc) in lean subjects (Tiedemann et al., 2017). On the other hand, insulin-mediated long-term depression of VTA dopamine neurons is reduced in hyperinsulinemia (Liu et al., 2013) and aberrant insulin action in VTA–NAc pathways has been observed in insulin-resistant participants (Tiedemann et al., 2017).

There is consensus that the improvement of hyperinsulinemia and insulin resistance, as achieved by caloric restriction (CR), is key in the prevention and treatment of obesity, type 2 diabetes, or cardiovascular diseases in aging (Janssen, 2021; Ryan, 2000). Evidence from human studies for such effects, however, are sparse, especially when it comes to central nervous insulin signaling. High insulin sensitivity in MEG theta activity was predictive for long-term weight management in 15 young adults (Kullmann et al., 2020b) suggesting the critical role of central insulin action for future feeding regulation and as a major target of treatment intervention. Whether such intervention can modulate brain insulin sensitivity in older age is particularly questionable given that changes in function and distribution of adipose tissue can trigger metabolic alterations such as hyperinsulinemia (Palmer and Kirkland, 2016; Tchkonia et al., 2010) and relevant brain circuits for central insulin action like the dopaminergic mesolimbic pathway undergo age-related changes (Karrer et al., 2017).

In the current longitudinal study, we investigated the role of peripheral and central insulin resistance regarding their predictive value for dietary success in older adults and whether both can be modified by weight changes. Fifty older (>55 years) overweight, non-diabetic individuals were randomly assigned to a 3-month, CR intervention or an active waiting group (WG). Before and after the intervention, overnight fasted participants took part in a crossover, placebo-controlled, double-blind pharmacological fMRI examination in which they rated the palatability of high and low sugar food pictures and the attractiveness of non-food items (control) after receiving intranasal insulin (INI) or placebo. Fasting c-peptides and blood glucose were assessed to calculate peripheral insulin sensitivity. We tested several predictions: (1) successful weight loss can be predicted by peripheral and central insulin sensitivity, the latter indicated by an insulinergic inhibition of mesolimbic responses to hedonic food stimuli at baseline (Tiedemann et al., 2017), and (2) both, peripheral and central insulin sensitivity improve with successful weight loss at follow-up. Moreover, we explored the common and distinct impact of both markers on weight (changes) in older age.

Fifty overweight and obese older adults (age: 63.7 ± 5.9 years, range 55–78 years; body mass index [BMI]: 32.7 ± 4.3 kg/m², range 25.8–32.4 kg/m²; 20 men) with an explicit wish to lose weight participated in this study. Of these, 30 randomly selected participants underwent a 3-month CR diet (diet group, DG, 14 men), while 20 participants were randomly assigned to a 3-month active WG (6 men). Before (T0) and after (T1) the intervention phase, we assessed anthropometrics and blood measures. Normal HbA1C values confirm the exclusion of manifest diabetes in overweight and obese participants who are at risk for T2D but in whom elevated insulin release may still compensate for reduced insulin sensitivity (mean HOMA-2: 2.2 ± 0.08, range 1.2–3.5). All participants underwent a double-blind, randomized, placebo-controlled fMRI paradigm on food and non-food liking combined with an INI application before and after the intervention phase (Figure 1). Thus, each participant attended a total of four scanning sessions. This longitudinal, within-subject design allowed us to evaluate peripheral and food-related central insulin action linked to overweight and weight loss in older adults.

Figure 1

Download asset Open asset

CR significantly reduces weight in dieters

Glucose, insulin, and c-peptide levels were assessed on all four study days in the morning after an overnight fast of at least 10 hr. Fasting glucose levels confirmed fasting state in all participants. Groups were well balanced regarding gender, age, overnight fasting times, and days between sessions (all p > 0.28). At T0, BMI, weight, bodyfat, waist, and HOMA-2 did not differ between groups (all p > 0.06, Table 1).

Table 1

	DG (N = 30)			WG (N = 20)
	T0	T1	p time	T0	T1	p time	p time × group
BMI (kg/m²)	32.1 (0.7)	30.8 (0.6)	***	32.8 (1.1)	32.8 (1.2)	N.S.	***
Waist (cm)	103.7 (1.8)	98.1 (1.9)	***	103.1 (2.7)	99.7 (2.8)	N.S.	N.S.
Bodyfat	37.4 (1.3)	36.7 (1.4)	N.S.	39.5 (1.6)	39.0 (1.7)	N.S.	N.S.
Blood
HOMA-2	2.1 (0.1)	1.9 (0.1)	*	2.3 (0.2)	2.3 (0.1)	N.S.	N.S.
Glucose (mmol/l)	5.5 (0.1)	5.4 (0.1)	N.S.1	5.7 (0.1)	5.8 (0.1)	N.S.1	N.S.²
Insulin (pmol/l)	78.8 (4.0)	70.3 (4.4)	+	95.5 (8.7)	94.2 (7.7)	N.S.	N.S.
C-peptide (nmol/l)	0.9 (0.03)	0.8 (0.04)	*	1.1 (0.06)	1.0 (0.05)	N.S.	N.S.
HbA1C	5.4 (0.03)	5.4 (0.04)	N.S.	5.5 (0.07)	5.5 (0.07)	N.S.	N.S.

1. ***p < 0.001, *p < 0.05, ⁺p < 0.10, s.e.m. in parantheses

2. DG, diet group; WG, waiting group; PL, placebo; IN, insulin; T0, baseline; T1, follow-up; ¹ Wilcoxon-rank Test; ² Mann-Whitney-U-Test.

3. BMI, body mass index; HOMA-2, c-peptide-based Homeostatic Model Assessment for Insulin Resistance; N.S, not significant.

After 3 months (mean 96 ± 10 days, no group differences: p = 0.91), follow-up measurements showed a significant mean weight loss compared to baseline of 3.61 kg (±3.06, T₍₂₉₎ = 6.47, p < 0.001, d = 1.18) in the DG, reflecting on average a 4% loss of baseline bodyweight and BMI, respectively (T₍₂₉₎ = 6.66, p < 0.001, d = 1.22). Twenty-one dieters (70%) lost more than 2 kg (range: 2.5–9.6 kg), only one dieter gained more than 0.4 kg (3.1 kg). In the WG, mean weight change was 0.07 kg (±1.5 kg). Sixteen participants of the WG were able to maintain their weight ±2 kg. Two gained weight (2.3 and 2.7 kg), two lost weight of 2.3 and 4 kg, respectively. Accordingly, percentage BMI change differed significantly between groups (BMI_%change: both T₍₄₈₎ = 5.45, p < 0.001, d = 1.39; Figure 2, Appendix 1—figure 1).

Figure 2

Download asset Open asset

Baseline peripheral insulin sensitivity predicts dietary success

Fasted serum c-peptide and plasma glucose levels were used for the calculation of an effective measure of peripheral insulin resistance (HOMA-2, Levy et al., 1998; https://www.dtu.ox.ac.uk/homacalculator/index.php). To test the predictive value of baseline insulin resistance in the periphery for dietary success after 3 months, we correlated individual HOMA-2 scores assessed on the placebo session from T0 with BMI_%change in dieters and found a significant correlation (r = −0.37; p = 0.046, n = 30, Pearson’s correlation). That is, higher insulin sensitivity at baseline predicted more subsequent weight loss in dieters (Figure 3a). Control analyses showed no such correlation in the WG and no correlation was observed between BMI_%change and baseline BMI (all p > 0.16).

Figure 3

Download asset Open asset

Midbrain insulin effects during sugar liking predict weight loss in dieters

To examine the neural mechanisms of how insulin influenced the brain’s mesocorticolimbic reward circuitry, we analyzed blood oxygenation level-dependent (BOLD) activity measured during the preference task using a two-level random effects model. As expected from our previous study (Tiedemann et al., 2017), the analysis of differences in BOLD responses to food compared to non-food items in the placebo session at T0 yielded highly significant activations across all participants in a network of reward-related brain regions including the bilateral insula, medial OFC, and amygdala (Figure 4a). Also in line with our previous findings, regions that encode the subjective value of items, that is, regions that show a positive correlation between the amplitude of the BOLD response and subjective preference values, comprised regions of the brain’s valuation network including the vmPFC and NAc (Figure 4b). BOLD signals in these regions did not differ between groups. Furthermore, valuation of HS compared to LS food items evoked significantly stronger correlations between BOLD signal and preference values in the ACC/vmPFC (Figure 4c), the right caudate nucleus and thalamus (all p < 0.05 FWE corrected) and as trend in the right NAc (9, 10, −7; FWE = 0.09, Figure 4c).

Figure 4

Download asset Open asset

We then investigated the effects of INI on these value signals. Here, in line with behavioral findings, analyses across and between groups yielded no significant changes of neural value signals for both general food items and HS versus LS items (see uncorrected results in Appendix 1—figure 3). There was also no relation between HOMA-2 and neural insulin effects across individuals or within the DG. Following up on the behavioral findings, we next analyzed whether individual insulin effects on neural signals during HS compared to LS food valuation predicted subsequent weight loss in the DG. Simple regression analysis including BMI_%change as a covariate yielded insulin-induced signal changes in the left VTA to predict subsequent weight loss across all participants as well as within subjects from the DG alone (−8, −13, −13, p < 0.05 FWE corrected; Figure 3b). Results in the left VTA were still significant when including age and BMI as covariates in the analysis (p < 0.05 FWE corrected). This indicates that participants in whom INI reduced the HS-specific valuation signal in the midbrain at T0 are more likely to benefit from CR by weight loss as assessed at T1.

Improvement of peripheral insulin sensitivity is related to increased insulin effects on NAc HS value signals after weight loss

We finally investigated metabolic and neurobehavioral changes due to successful weight loss. Within participants from the DG, HOMA-2 scores were significantly improved at follow-up (T₍₂₉₎ = 2.33; p = 0.027, d = 0.43). Moreover, the improvement in HOMA-2 scores was directly correlated with successful weight change within dieters (r = 0.43; p = 0.017; N = 30, Pearson’s correlation) and across all participants (r = 0.33; p = 0.020; N = 50, Pearson’s correlation, Figure 5).

Figure 5

Download asset Open asset

We next tested whether successful weight loss also improved central insulin sensitivity as assessed with our pharmacological fMRI design (for characteristics of the T1 fMRI sessions see Appendix 1—table 1, Appendix 1—table 4). In behavior, participants from the DG showed a significantly reduced sweet food preference (i.e., percentage of sweet foods in preferred foods) under insulin compared to placebo at follow-up (T₍₂₉₎ = 2.59; p = 0.015, d = 0.47) that tended to be stronger compared to the WG (T₍₄₉₎ = 1.80; p = 0.08) and to baseline (T₍₂₉₎ = 1.67; p = 0.11) (Figure 6a).

Figure 6

Download asset Open asset

On the neural level, behavioral insulin effects in the DG at follow-up were reflected by a stronger reduction of sugar-specific value signals in the NAc under insulin in the DG compared to the WG (peak right: 10, 8, −7, p = 0.028 FWE corrected and peak left: −10, 12, −8, p = 0.043 FWE corrected, t-test). The effect in the right NAc was also significantly stronger when directly comparing follow-up to baseline valuation responses between groups (peak: 10, 8, −6, p = 0.019 FWE corrected; two-sample t-test). Exploration of extracted BOLD signals (Figure 6b) indicate that this effect was at least partly driven by an opposite effect in the WG, that is, sweet food signals in the NAc relatively increased under insulin at T1. There was no significant insulin effect across all participants at T1 (see uncorrected results in Figure 3b).

We therefore more directly focused on the DG and explored whether insulin-mediated neural signal changes after weight loss were related to changes in peripheral insulin sensitivity following dietary intervention. To this end, changes in HOMA-2 scores (post > pre) were entered as a covariate in a simple regression of BOLD signals from the contrast HS > LS_{PL>IN_post} > HS > LS_{PL>IN_pre} within participants from the DG. Results revealed positive correlations between improvement of insulin sensitivity measured in the blood and increased insulin effects in the NAc (peak left: −10, 10, −7, p = 0.03 FWE corrected; peak right: 10, 10, −7; Figure 7). No significant brain correlates for this analysis were found in the WG. Results in the left NAc were still significant when including age and BMI as covariates in the analysis.

Figure 7

Download asset Open asset

Our findings indicate an independent predictive value of peripheral and central insulin sensitivity for dietary success in overweight elderlies and an improvement of both after losing weight. In non-diabetic, overweight, and obese older participants who underwent a 3-month CR, significant weight loss could be predicted by baseline measures of c-peptide-based insulin sensitivity as well as acute insulinergic inhibition of VTA responses to high sugar food items. Both markers of insulin function made an independent contribution to weight loss prediction emphasizing the necessity to take both aspects into account when assessing predictors and consequences of overweight in the elderly. At follow-up, weight loss in dieters was associated with improved peripheral insulin sensitivity which was directly related to a stronger insulinergic inhibition in the NAc during hedonic food valuation. These findings extend work in rodents (Mebel et al., 2012) and first studies in humans (Kullmann et al., 2020b) about the critical role of insulin sensitivity for future feeding regulation. It is also, to our knowledge, the first study that demonstrates positive effects of CR on central insulin function in humans using a longitudinal within-subject design. The observation of such an effect in older adults is particularly important given age-related metabolic and neural changes and the potential, detrimental consequences of hyperinsulinemia, overweight, and obesity in later life (Janssen, 2021).

As was expected from overweight and obese individuals, both peripheral and central markers of insulin sensitivity were low at baseline and both selectively improved with weight loss after CR in dieters. At baseline, participants with higher hyperinsulinemia demonstrated a specifically enhanced sugar preference which fits animal and human data on the critical role of sugar-enriched diets on whole-body insulin functioning (Macdonald, 2016). Baseline insulin markers in the blood, however, were not related to behavioral or neural INI responses to HS items. This may be driven by the general lack of central insulin effects across participants at baseline which fits with data in overweight younger adults (Tiedemann et al., 2017) and which might result from an attenuated insulin transport into the cerebrospinal fluid in individuals with reduced whole-body insulin sensitivity (Heni et al., 2014). However, there was also no such association in the subgroup of successful dieters in whom there was a significant response to INI already before the intervention. This indicates that, even though peripheral insulin might be a good proxy for central insulin functioning in younger lean adults (Tiedemann et al., 2017), pathophysiological changes due to overweight and aging might at least in part independently affect peripheral and central insulin effects. This is further underlined by the independent predictive value of both markers for weight changes after 3 months of CR. The impact of these predictive values could not be explained by body mass, which demonstrates that insulin sensitivity, but not necessarily obesity, is predictive for future weight management. Indeed, there is evidence from animal and human research of an adiposity-independent effect of aging on insulin sensitivity (Ehrhardt et al., 2019; Petersen et al., 2003), that, for instance, may result from age-related changes in mitochondrial energy metabolism (Petersen et al., 2003) or increased systemic inflammation (Ehrhardt et al., 2019).

Inhibitory INI effects on VTA value signals to sweet food items were selectively predictive for subsequent success of CR. The VTA plays a central role in the insulinergic modulation of hedonic eating behavior (Labouèbe et al., 2013; Mebel et al., 2012; Tiedemann et al., 2017). Direct administration of insulin into the VTA reduces hedonic feeding and depresses somatodendritic DA in the VTA which has been attributed to the upregulation of the number or function of DA transporter in the VTA (Mebel et al., 2012). Connectivity analyses of fMRI data further suggest that INI can suppress food value signals in the mesolimbic pathway by negatively modulating projections from the VTA to the NAc (Tiedemann et al., 2017). Insulinergic effects at baseline and follow-up were specifically restricted to HS food stimuli. The palatability of sugar has been linked to DA release in the NAc in rodents (Hajnal et al., 2004) and there is evidence for neural adaptations in the NAc in response to excessive sugar intake (Klenowski et al., 2016). For instance, higher sugar preference in overweight individuals has been related to stronger white matter connectivity within the VTA–NAc pathway (Francke et al., 2019). Our data indicate that insulinergic functionality in this network may be critical for hedonic feeding regulation as the reduction of sugar intake is substantial for the success of a dietary intervention.

A reduced insulinergic functionality of this network in older overweight individuals may not only be the consequence of adiposity (Mattson and Arumugam, 2018) but may also result from age-related metabolic and neural changes. Aging is associated with a decrease of cortical insulin concentration, reduced insulin receptor binding ability and reduced insulin transport across the blood–brain barrier (Cholerton et al., 2011). Moreover, target systems of metabolic–hedonic networks relevant for insulin action undergo age-related changes (Mattson and Arumugam, 2018; Smith et al., 2020). The dopamine system, for example, is particularly vulnerable to aging which might lead to functional changes in subcortical reward circuits (Dreher et al., 2008; Karrer et al., 2017). There is a significant loss of dopaminergic neurons in the basal ganglia including the VTA (Siddiqi et al., 1999). Given that insulin acts via glutamatergic synaptic transmission onto VTA DA neurons (Labouèbe et al., 2013) this might have direct consequences on the insulinergic suppression of subsequent DA release in mesolimbic regions. The potential negative impact of adiposity and age on described dysfunctions are thereby probably not simply additive. For instance, chronic metabolic morbidities like obesity can further accelerate brain aging (Mattson and Arumugam, 2018). A chronic positive energy balance thereby adversely affect brain function (Beyer et al., 2017) and structure (Janowitz et al., 2015) and is related to many of the cellular and molecular hallmarks of brain aging such as oxidative damage and neuroinflammation (Mattson and Arumugam, 2018).

Intriguingly, while there was no association between blood parameters and central insulin action at baseline, weight-change related improvements in peripheral and central insulin sensitivity in our sample of older dieters were directly correlated at follow-up, indicating a common modulator. Moreover, changes in central insulin sensitivity were restricted to an increased inhibition of value signals in the NAc but not the VTA. Thus, one could speculate that weight change specifically normalized adiposity-related dysfunctions while variability due to aging itself were less affected. Improvement in insulin sensitivity and glucose homeostasis is a broadly observed metabolic effect of CR in rodents (Yu et al., 2019; Zhang et al., 2021) as well as young and older adults (Fontana and Klein, 2007; Johnson et al., 2016; Most and Redman, 2020; Most et al., 2017). The mechanisms behind these effects are not fully understood yet but have been related to significantly increased hepatic insulin clearance (Bosello et al., 1990), reduced levels of thioredoxin-interacting protein (TXNIP; Johnson et al., 2016), and generally decreased oxidative stress and inflammatory processes (Fontana and Klein, 2007). Animal data about CR effects on brain functioning suggest that CR can induce adaptive cellular responses that can enhance neuroplasticity and stress resistance, for example, by the upregulation of neurotrophic factor signaling, suppression of oxidative stress and inflammation, stabilization of neuronal calcium homeostasis, and stimulation of mitochondrial biogenesis (Mattson, 2012; Mattson and Arumugam, 2018). In addition, recent work in rodents demonstrate improved insulin sensitivity following CR that was associated with enhanced brain monoamine concentrations such as increased DA levels in the striatum (Portero-Tresserra et al., 2020). Our data extend these beneficial neural effects of CR in animals to improved central insulin functioning in the human brain. This is particularly intriguing with regard to our non-diabetic sample of elderlies, in whom weight-related brain dysfunction is not only a risk factor for metabolic disorders but also for cognitive decline and neurodegeneration (Ekblad et al., 2017; Janssen, 2021; Mattson and Arumugam, 2018).

We chose a relatively mild dietary intervention that reduced participants’ individual caloric intake by 10–15% with a minimal intake set to 1200 kcal per day. This was done to increase compliance and to provide elderlies with a feasible long-term strategy to lose and maintain weight. Accordingly, there was only a mild-to-moderate average weight loss of 4%. Even this mild weight change was related to significant improvement of insulin sensitivity in the periphery and in the brain which underline that adiposity-related dysfunctions in later life are able to normalize. This is especially promising given new evidence for hyperinsulinemia preceding insulin resistance (Janssen, 2021) which makes it a key target for early interventions. It is now critical to understand the long-term effects of such changes with a special focus on food intake assuming that long-term effects are probably particularly dependent on prefrontal mediated psychological strategies including self-control during eating decisions (Hare et al., 2009; Phelan et al., 2020). In conclusion, we provide data demonstrating that peripheral insulin sensitivity as well as central hedonic feeding regulation predict and normalize with dietary success in overweight elderlies. Our results of an independent contribution peripheral and central insulin sensitivity make for successful feeding regulation emphasize the necessity to control for both when treating individuals at risk for metabolic disorders.

All data for analyses and figures in this study are provided at Dryad.

1. 1. Brassen S

   (2022) Dryad Digital Repository

   Data from: Insulin sensitivity in mesolimbic pathways predicts and improves with weight loss in older dieters.

   https://doi.org/10.5061/dryad.8cz8w9gsn

1. 1. Beyer F
   2. Kharabian Masouleh S
   3. Huntenburg JM
   4. Lampe L
   5. Luck T
   6. Riedel-Heller SG
   7. Loeffler M
   8. Schroeter ML
   9. Stumvoll M
   10. Villringer A
   11. Witte AV

   (2017) Higher body mass index is associated with reduced posterior default mode connectivity in older adults

   Human Brain Mapping 38:3502–3515.

   https://doi.org/10.1002/hbm.23605

   • PubMed
   • Google Scholar
2. 1. Born J
   2. Lange T
   3. Kern W
   4. McGregor GP
   5. Bickel U
   6. Fehm HL

   (2002) Sniffing neuropeptides: a transnasal approach to the human brain

   Nature Neuroscience 5:514–516.

   https://doi.org/10.1038/nn849

   • PubMed
   • Google Scholar
3. 1. Bosello O
   2. Zamboni M
   3. Armellini F
   4. Zocca I
   5. Bergamo Andreis IA
   6. Smacchia C
   7. Milani MP
   8. Cominacini L

   (1990) Modifications of abdominal fat and hepatic insulin clearance during severe caloric restriction

   Annals of Nutrition & Metabolism 34:359–365.

   https://doi.org/10.1159/000177610

   • PubMed
   • Google Scholar
4. 1. Cholerton B
   2. Baker LD
   3. Craft S

   (2011) Insulin resistance and pathological brain ageing

   Diabetic Medicine 28:1463–1475.

   https://doi.org/10.1111/j.1464-5491.2011.03464.x

   • PubMed
   • Google Scholar
5. 1. Davis JF
   2. Choi DL
   3. Benoit SC

   (2010) Insulin, leptin and reward

   Trends in Endocrinology and Metabolism 21:68–74.

   https://doi.org/10.1016/j.tem.2009.08.004

   • PubMed
   • Google Scholar
6. 1. Dreher JC
   2. Meyer-Lindenberg A
   3. Kohn P
   4. Berman KF

   (2008) Age-Related changes in midbrain dopaminergic regulation of the human reward system

   PNAS 105:15106–15111.

   https://doi.org/10.1073/pnas.0802127105

   • PubMed
   • Google Scholar
7. 1. Ehrhardt N
   2. Cui J
   3. Dagdeviren S
   4. Saengnipanthkul S
   5. Goodridge HS
   6. Kim JK
   7. Lantier L
   8. Guo X
   9. Chen YDI
   10. Raffel LJ
   11. Buchanan TA
   12. Hsueh WA
   13. Rotter JI
   14. Goodarzi MO
   15. Péterfy M

   (2019) Adiposity-independent effects of aging on insulin sensitivity and clearance in mice and humans

   Obesity 27:434–443.

   https://doi.org/10.1002/oby.22418

   • PubMed
   • Google Scholar
8. 1. Ekblad LL
   2. Rinne JO
   3. Puukka P
   4. Laine H
   5. Ahtiluoto S
   6. Sulkava R
   7. Viitanen M
   8. Jula A

   (2017) Insulin resistance predicts cognitive decline: an 11-year follow-up of a nationally representative adult population sample

   Diabetes Care 40:751–758.

   https://doi.org/10.2337/dc16-2001

   • Google Scholar
9. 1. Fontana L
   2. Klein S

   (2007) Aging, adiposity, and calorie restriction

   JAMA 297:986–994.

   https://doi.org/10.1001/jama.297.9.986

   • PubMed
   • Google Scholar
10. 1. Francke P
    2. Tiedemann LJ
    3. Menz MM
    4. Beck J
    5. Büchel C
    6. Brassen S

    (2019) Mesolimbic white matter connectivity mediates the preference for sweet food

    Scientific Reports 9:4349.

    https://doi.org/10.1038/s41598-019-40935-6

    • PubMed
    • Google Scholar
11. 1. Hajnal A
    2. Smith GP
    3. Norgren R

    (2004) Oral sucrose stimulation increases accumbens dopamine in the rat

    American Journal of Physiology. Regulatory, Integrative and Comparative Physiology 286:R31–R37.

    https://doi.org/10.1152/ajpregu.00282.2003

    • PubMed
    • Google Scholar
12. 1. Hare TA
    2. Camerer CF
    3. Rangel A

    (2009) Self-Control in decision-making involves modulation of the vmpfc valuation system

    Science 324:646–648.

    https://doi.org/10.1126/science.1168450

    • PubMed
    • Google Scholar
13. 1. Heni M
    2. Schöpfer P
    3. Peter A
    4. Sartorius T
    5. Fritsche A
    6. Synofzik M
    7. Häring HU
    8. Maetzler W
    9. Hennige AM

    (2014) Evidence for altered transport of insulin across the blood-brain barrier in insulin-resistant humans

    Acta Diabetologica 51:679–681.

    https://doi.org/10.1007/s00592-013-0546-y

    • PubMed
    • Google Scholar
14. 1. Hou X
    2. Harel J
    3. Koch C

    (2012) Image signature: highlighting sparse salient regions

    IEEE Transactions on Pattern Analysis and Machine Intelligence 34:194–201.

    https://doi.org/10.1109/TPAMI.2011.146

    • PubMed
    • Google Scholar
15. 1. Janowitz D
    2. Wittfeld K
    3. Terock J
    4. Freyberger HJ
    5. Hegenscheid K
    6. Völzke H
    7. Habes M
    8. Hosten N
    9. Friedrich N
    10. Nauck M
    11. Domanska G
    12. Grabe HJ

    (2015) Association between waist circumference and gray matter volume in 2344 individuals from two adult community-based samples

    NeuroImage 122:149–157.

    https://doi.org/10.1016/j.neuroimage.2015.07.086

    • PubMed
    • Google Scholar
16. 1. Janssen J

    (2021) Hyperinsulinemia and its pivotal role in aging, obesity, type 2 diabetes, cardiovascular disease and cancer

    International Journal of Molecular Sciences 22:7797.

    https://doi.org/10.3390/ijms22157797

    • PubMed
    • Google Scholar
17. 1. Johnson ML
    2. Distelmaier K
    3. Lanza IR
    4. Irving BA
    5. Robinson MM
    6. Konopka AR
    7. Shulman GI
    8. Nair KS

    (2016) Mechanism by which caloric restriction improves insulin sensitivity in sedentary obese adults

    Diabetes 65:74–84.

    https://doi.org/10.2337/db15-0675

    • PubMed
    • Google Scholar
18. 1. Jones AG
    2. Hattersley AT

    (2013) The clinical utility of C-peptide measurement in the care of patients with diabetes

    Diabetic Medicine 30:803–817.

    https://doi.org/10.1111/dme.12159

    • PubMed
    • Google Scholar
19. 1. Kalyani RR
    2. Golden SH
    3. Cefalu WT

    (2017) Diabetes and aging: unique considerations and goals of care

    Diabetes Care 40:440–443.

    https://doi.org/10.2337/dci17-0005

    • PubMed
    • Google Scholar
20. 1. Karrer TM
    2. Josef AK
    3. Mata R
    4. Morris ED
    5. Samanez-Larkin GR

    (2017) Reduced dopamine receptors and transporters but not synthesis capacity in normal aging adults: a meta-analysis

    Neurobiology of Aging 57:36–46.

    https://doi.org/10.1016/j.neurobiolaging.2017.05.006

    • PubMed
    • Google Scholar
21. 1. Klenowski PM
    2. Shariff MR
    3. Belmer A
    4. Fogarty MJ
    5. Mu EWH
    6. Bellingham MC
    7. Bartlett SE

    (2016) Prolonged consumption of sucrose in a binge-like manner, alters the morphology of medium spiny neurons in the nucleus accumbens shell

    Frontiers in Behavioral Neuroscience 10:54.

    https://doi.org/10.3389/fnbeh.2016.00054

    • PubMed
    • Google Scholar
22. 1. Kullmann S
    2. Kleinridders A
    3. Small DM
    4. Fritsche A
    5. Häring HU
    6. Preissl H
    7. Heni M

    (2020a) Central nervous pathways of insulin action in the control of metabolism and food intake

    The Lancet. Diabetes & Endocrinology 8:524–534.

    https://doi.org/10.1016/S2213-8587(20)30113-3

    • PubMed
    • Google Scholar
23. 1. Kullmann S
    2. Valenta V
    3. Wagner R
    4. Tschritter O
    5. Machann J
    6. Häring HU
    7. Preissl H
    8. Fritsche A
    9. Heni M

    (2020b) Brain insulin sensitivity is linked to adiposity and body fat distribution

    Nature Communications 11:1841.

    https://doi.org/10.1038/s41467-020-15686-y

    • PubMed
    • Google Scholar
24. 1. Labouèbe G
    2. Liu S
    3. Dias C
    4. Zou H
    5. Wong JCY
    6. Karunakaran S
    7. Clee SM
    8. Phillips AG
    9. Boutrel B
    10. Borgland SL

    (2013) Insulin induces long-term depression of ventral tegmental area dopamine neurons via endocannabinoids

    Nature Neuroscience 16:300–308.

    https://doi.org/10.1038/nn.3321

    • PubMed
    • Google Scholar
25. 1. Leighton E
    2. Sainsbury CA
    3. Jones GC

    (2017) A practical review of C-peptide testing in diabetes

    Diabetes Therapy 8:475–487.

    https://doi.org/10.1007/s13300-017-0265-4

    • PubMed
    • Google Scholar
26. 1. Levy JC
    2. Matthews DR
    3. Hermans MP

    (1998) Correct homeostasis model assessment (HOMA) evaluation uses the computer program

    Diabetes Care 21:2191–2192.

    https://doi.org/10.2337/diacare.21.12.2191

    • PubMed
    • Google Scholar
27. 1. Liu S
    2. Labouèbe G
    3. Karunakaran S
    4. Clee SM
    5. Borgland SL

    (2013) Effect of insulin on excitatory synaptic transmission onto dopamine neurons of the ventral tegmental area in a mouse model of hyperinsulinemia

    Nutrition & Diabetes 3:e97.

    https://doi.org/10.1038/nutd.2013.38

    • PubMed
    • Google Scholar
28. 1. Macdonald IA

    (2016) A review of recent evidence relating to sugars, insulin resistance and diabetes

    European Journal of Nutrition 55:17–23.

    https://doi.org/10.1007/s00394-016-1340-8

    • PubMed
    • Google Scholar
29. 1. Mattson MP

    (2012) Energy intake and exercise as determinants of brain health and vulnerability to injury and disease

    Cell Metabolism 16:706–722.

    https://doi.org/10.1016/j.cmet.2012.08.012

    • PubMed
    • Google Scholar
30. 1. Mattson MP
    2. Arumugam TV

    (2018) Hallmarks of brain aging: adaptive and pathological modification by metabolic states

    Cell Metabolism 27:1176–1199.

    https://doi.org/10.1016/j.cmet.2018.05.011

    • PubMed
    • Google Scholar
31. 1. Mebel DM
    2. Wong JCY
    3. Dong YJ
    4. Borgland SL

    (2012) Insulin in the ventral tegmental area reduces hedonic feeding and suppresses dopamine concentration via increased reuptake

    The European Journal of Neuroscience 36:2336–2346.

    https://doi.org/10.1111/j.1460-9568.2012.08168.x

    • PubMed
    • Google Scholar
32. 1. Most J
    2. Tosti V
    3. Redman LM
    4. Fontana L

    (2017) Calorie restriction in humans: an update

    Ageing Research Reviews 39:36–45.

    https://doi.org/10.1016/j.arr.2016.08.005

    • PubMed
    • Google Scholar
33. 1. Most J
    2. Redman LM

    (2020) Impact of calorie restriction on energy metabolism in humans

    Experimental Gerontology 133:110875.

    https://doi.org/10.1016/j.exger.2020.110875

    • PubMed
    • Google Scholar
34. 1. Murray S
    2. Tulloch A
    3. Gold MS
    4. Avena NM

    (2014) Hormonal and neural mechanisms of food reward, eating behaviour and obesity

    Nature Reviews. Endocrinology 10:540–552.

    https://doi.org/10.1038/nrendo.2014.91

    • PubMed
    • Google Scholar
35. 1. Okura T
    2. Nakamura R
    3. Fujioka Y
    4. Kawamoto-Kitao S
    5. Ito Y
    6. Matsumoto K
    7. Shoji K
    8. Sumi K
    9. Matsuzawa K
    10. Izawa S
    11. Ueta E
    12. Kato M
    13. Imamura T
    14. Taniguchi SI
    15. Yamamoto K

    (2018) CPR-IR is an insulin resistance index that is minimally affected by hepatic insulin clearance-A preliminary research

    PLOS ONE 13:e0197663.

    https://doi.org/10.1371/journal.pone.0197663

    • PubMed
    • Google Scholar
36. 1. Palmer AK
    2. Kirkland JL

    (2016) Aging and adipose tissue: potential interventions for diabetes and regenerative medicine

    Experimental Gerontology 86:97–105.

    https://doi.org/10.1016/j.exger.2016.02.013

    • PubMed
    • Google Scholar
37. 1. Petersen KF
    2. Befroy D
    3. Dufour S
    4. Dziura J
    5. Ariyan C
    6. Rothman DL
    7. DiPietro L
    8. Cline GW
    9. Shulman GI

    (2003) Mitochondrial dysfunction in the elderly: possible role in insulin resistance

    Science 300:1140–1142.

    https://doi.org/10.1126/science.1082889

    • PubMed
    • Google Scholar
38. 1. Phelan S
    2. Halfman T
    3. Pinto AM
    4. Foster GD

    (2020) Behavioral and psychological strategies of long-term weight loss maintainers in a widely available weight management program

    Obesity 28:421–428.

    https://doi.org/10.1002/oby.22685

    • PubMed
    • Google Scholar
39. 1. Portero-Tresserra M
    2. Rojic-Becker D
    3. Vega-Carbajal C
    4. Guillazo-Blanch G
    5. Vale-Martínez A
    6. Martí-Nicolovius M

    (2020) Caloric restriction modulates the monoaminergic system and metabolic hormones in aged rats

    Scientific Reports 10:19299.

    https://doi.org/10.1038/s41598-020-76219-7

    • PubMed
    • Google Scholar
40. 1. Ryan AS

    (2000) Insulin resistance with aging: effects of diet and exercise

    Sports Medicine 30:327–346.

    https://doi.org/10.2165/00007256-200030050-00002

    • PubMed
    • Google Scholar
41. 1. Siddiqi Z
    2. Kemper TL
    3. Killiany R

    (1999) Age-Related neuronal loss from the substantia nigra-pars compacta and ventral tegmental area of the rhesus monkey

    Journal of Neuropathology and Experimental Neurology 58:959–971.

    https://doi.org/10.1097/00005072-199909000-00006

    • PubMed
    • Google Scholar
42. 1. Smith SM
    2. Elliott LT
    3. Alfaro-Almagro F
    4. McCarthy P
    5. Nichols TE
    6. Douaud G
    7. Miller KL

    (2020) Brain aging comprises many modes of structural and functional change with distinct genetic and biophysical associations

    eLife 9:e52677.

    https://doi.org/10.7554/eLife.52677

    • PubMed
    • Google Scholar
43. 1. Tchkonia T
    2. Morbeck DE
    3. Von Zglinicki T
    4. Van Deursen J
    5. Lustgarten J
    6. Scrable H
    7. Khosla S
    8. Jensen MD
    9. Kirkland JL

    (2010) Fat tissue, aging, and cellular senescence

    Aging Cell 9:667–684.

    https://doi.org/10.1111/j.1474-9726.2010.00608.x

    • PubMed
    • Google Scholar
44. 1. Tiedemann LJ
    2. Schmid SM
    3. Hettel J
    4. Giesen K
    5. Francke P
    6. Büchel C
    7. Brassen S

    (2017) Central insulin modulates food valuation via mesolimbic pathways

    Nature Communications 8:16052.

    https://doi.org/10.1038/ncomms16052

    • PubMed
    • Google Scholar
45. 1. Witte AV
    2. Fobker M
    3. Gellner R
    4. Knecht S
    5. Flöel A

    (2009) Caloric restriction improves memory in elderly humans

    PNAS 106:1255–1260.

    https://doi.org/10.1073/pnas.0808587106

    • PubMed
    • Google Scholar
46. 1. Yu D
    2. Tomasiewicz JL
    3. Yang SE
    4. Miller BR
    5. Wakai MH
    6. Sherman DS
    7. Cummings NE
    8. Baar EL
    9. Brinkman JA
    10. Syed FA
    11. Lamming DW

    (2019) Calorie-restriction-induced insulin sensitivity is mediated by adipose mTORC2 and not required for lifespan extension

    Cell Reports 29:236–248.

    https://doi.org/10.1016/j.celrep.2019.08.084

    • PubMed
    • Google Scholar
47. 1. Zhang L
    2. Huang YJ
    3. Sun JP
    4. Zhang TY
    5. Liu TL
    6. Ke B
    7. Shi XF
    8. Li H
    9. Zhang GP
    10. Ye ZY
    11. Hu J
    12. Qin J

    (2021) Protective effects of calorie restriction on insulin resistance and islet function in STZ-induced type 2 diabetes rats

    Nutrition & Metabolism 18:48.

    https://doi.org/10.1186/s12986-021-00575-y

    • PubMed
    • Google Scholar

Decision letter after peer review:

Thank you for submitting your article "Insulin sensitivity in mesolimbic pathways predicts and improves with weight loss in older dieters" for consideration by eLife. Your article has been reviewed by 1 peer reviewer, and the evaluation has been overseen by a Reviewing Editor and Carlos Isales as the Senior Editor. The reviewer has opted to remain anonymous.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

Abstract

(1) Line 22: Whole body insulin sensitivity was not assessed, please write peripheral (same for line 406).

Results

(2) Line 76 ff: Please indicate the age range, BMI range, and HOMA IR range of participants. Please add the number of men and women in the DG and WG group separately. Was sex equally distributed for the intervention and control groups? Were the women in menopause?

(3) The authors present the nucleus accumbens and VTA BOLD response to insulin. As this is quite a novel study, I would recommend showing the whole brain response (uncorrected for display) for Insulin versus placebo for T0 and T1 for both groups (for sweet versus non-sweet contrast). This could be presented in the supplementary material.

Discussion

(4) Line 302: the first statement of the discussion is quite strong. I agree that the regression model shows that peripheral insulin sensitivity and central insulin response explained variance independent of one another, this is a first indication that peripheral and central responses might be independent, but the study also shows that they correlate. Furthermore, insulin resistance was only based on fasting measurements and not during a clamp or oGTT.

Methods

(5) Participants: in the abstract and in the discussion (line 392), the authors refer to their study sample as prediabetic. However, based on the description of the participants in the methods section, persons overweight and with obesity at the age of 55 and higher were recruited for the study. No official criteria for prediabetes (as described for example by the ADA: https://www.diabetes.org/diabetes/a1c/diagnosis) were employed. Please use the term non-diabetic as in other passages of the manuscript.

(6) I agree that C-peptide-based assessments of peripheral insulin resistance can be superior to insulin-based calculations. However, the CPR-IR is rarely used (only 9 citations of the original paper on that index). I suggest using the commonly used C-peptide-based HOMA-2 instead (DOI: 10.2337/diacare.21.12.2191 see: https://www.dtu.ox.ac.uk/homacalculator/index.php).

Why did the authors use plasma glucose of 128 mg/dl as a cut-off for non-fasting? A commonly used cut-off is 126 mg/dl. Another reliable way to detect non-fasted individuals is unusually high C-peptide concentrations.

(7) Line 420: "two showed task behavior indicating...". I think there is a word missing.

(8) Line 422: "substantial movement". How was this defined? This could also be indicated in paragraph 522 ff.

(9) The authors conducted a median split on sugar content to divide the stimulus set into a high sugar (HS) and low sugar (LS) condition. What was the distribution of the sugar content in the food images? Who evaluated the sugar content of the stimuli?

(10) Line 433: The exclusion criteria include certain illnesses. Was this assessed by a physician or self-reported? Please add this information.

(11) Line 440 ff: What is meant by "additional financial incentives were provided [..] for successful weight loss"? Did participants receive more money if they lost more weight? Please clarify.

(12) Line 465: How was the individual reduction by 10 to 15% calculated? Please specify. Was resting energy expenditure used? Or was it based on their daily prior consumption? The authors mention dietary records. How many days were recorded and at which time points of the study? How was compliance assured?

(13) Line 516: it is not clear what 30_ steeper means. Please clarify.

(14) Line 502: The fMRI food cue paradigm relies on images selected from the internet. Please indicate from what platform these images were selected. The images were then categorized based on sweetness. Were the pictures matched for physical and psychological variables? The authors only mention that the mean preference ratings of the stimuli were investigated in an independent sample, but other properties could play an important role (e.g. arousal, valence, and complexity of image).

(15) The fMRI data analyses section is described in detail up to the second level model.

Line 540: For the second level model, the authors indicate that they used one-sample t-tests and two-sample t-tests. Please be more specific about which contrasts (differential images) went into the model. (16) Based on the described results, I would have expected paired t-tests (T0 versus T1) or a flexible factorial model to investigate within-subject and between-subject effects in one model. Please clarify.

(16) The method section is lacking a statistical methods paragraph for the regression analysis and behavioral data analyses. What tool was used, etc.?

(17) The authors used one-way ANOVAs for analyzing parametric liking scores for the different picture sets. What is the distribution of the liking ratings? Are they normally distributed? The authors computed several rmANOVAs. Were the assumptions for such analyses tested?

BMI reduction was correlated with several measures (predictors of subsequent weight loss). In this (18) case, you must correct your p-values in relation to the number of tests (Bonferroni correction). Were the correlations between the BOLD response and peripheral insulin resistance still significant after adjusting for BMI and age?

(19) Was the study registered, for example at clinical trials.gov? Were the hypothesis and primary outcomes or study design preregistered? How was weight loss success defined?

(20) The study by Tiedmann et al., 2017 Nat Comm used the same fMRI paradigm to investigate insulin response of the reward circuitry in young adults of normal weight as well as overweight and obesity. In that study, the authors used HOMA-IR as a measure of peripheral insulin resistance. Why change this in the current study? Furthermore, in the previous study by Tiedemann, DCM model was used to show insulin-induced changes in connectivity between the nucleus accumbens and VTA. This connection was significantly modulated by intranasal insulin. Why wasn't this evaluated in the current study?

(21) The study of Tiedemann published in 2017 investigated the response of intranasal insulin in young individuals, while the current study included elderly persons. The impact of age independent of BMI has not been investigated thus far on neural insulin processing. Have you looked at the effect of age on the nucleus accumbens or VTA BOLD response to insulin?

(22) The authors use the term insulinergic inhibition. Is it possible to evaluate neural inhibition with BOLD fMRI?

Essential revisions:

Abstract

(1) Line 22: Whole body insulin sensitivity was not assessed, please write peripheral (same for line 406).

This has been changed.

Results

(2) Line 76 ff: Please indicate the age range, BMI range, and HOMA IR range of participants. Please add the number of men and women in the DG and WG group separately. Was sex equally distributed for the intervention and control groups? Were the women in menopause?

We have now added age range, BMI range, and HOMA-2 (+range). Number of men and women in DG and WG were also added. We also mention that there were no sex differences between groups (chi-square = 1.39, P = .38).

We did not specifically assess menopause status by hormonal measurements. The youngest woman was 55 years old, so it is likely that most, if not all, women in this study were in the middle or postmenopausal phase.

(3) The authors present the nucleus accumbens and VTA BOLD response to insulin. As this is quite a novel study, I would recommend showing the whole brain response (uncorrected for display) for Insulin versus placebo for T0 and T1 for both groups (for sweet versus non-sweet contrast). This could be presented in the supplementary material.

We agree that sub-threshold activation patterns are helpful to inform future designs given the novelty of our study. Consistent with the reviewer’s suggestion, we have therefore included uncorrected SPMs for each time point for all participants in the appendix (Appendix Figure 3). In addition, we would like to point out that whole-brain T-maps of these and the other results are available at Dryad (https://datadryad.org/stash/share/_sUG1RvcWB3uCpSxeiYRWcKBtUZmCnzSjir4dWGv16c) which can be used, for example, to suggest neural assumptions in future studies.

Discussion

(4) Line 302: the first statement of the discussion is quite strong. I agree that the regression model shows that peripheral insulin sensitivity and central insulin response explained variance independent of one another, this is a first indication that peripheral and central responses might be independent, but the study also shows that they correlate. Furthermore, insulin resistance was only based on fasting measurements and not during a clamp or oGTT.

We have toned down this statement in accordance with the reviewer’s suggestion in “Our findings indicate…”. However, we would like to point out that the regression model indeed shows independent effects of the two parameters (based on partial correlations) and that the two variables indeed were not correlated at T0. Both aspects strongly suggest that we observed two independent predictive mechanisms in this context.

We agree that an OGTT or clamp based assessment of insulin sensitivity would have been an interesting addition to this study When we designed the study, we decided against an euglycemic clamp procedure or oGTT, mainly for logistic and compliance reasons. The latter, in particular, could have had an impact on participants’ performance in the scanner. Participants already had to fast overnight for at least 10 hours, meaning that scanning took place in the early morning. Considering that one study day already lasted 2 hours (before participants were finally allowed to eat), a preceding clamp or oGTT would have probably overstretched the physical and psychological compliance of some participants. Reassuringly, there is an extensive literature demonstrating the high validity (e.g., reported correlations of HOMA-IR with clamp assessment ~.88) of fasting measurements in non-diabetic participants (reviewed in Wallace et al., Diabetes Care, 2004).

Methods

(5) Participants: in the abstract and in the discussion (line 392), the authors refer to their study sample as prediabetic. However, based on the description of the participants in the methods section, persons overweight and with obesity at the age of 55 and higher were recruited for the study. No official criteria for prediabetes (as described for example by the ADA: https://www.diabetes.org/diabetes/a1c/diagnosis) were employed. Please use the term non-diabetic as in other passages of the manuscript.

This has been changed.

(6) I agree that C-peptide-based assessments of peripheral insulin resistance can be superior to insulin-based calculations. However, the CPR-IR is rarely used (only 9 citations of the original paper on that index). I suggest using the commonly used C-peptide-based HOMA-2 instead (DOI: 10.2337/diacare.21.12.2191 see: https://www.dtu.ox.ac.uk/homacalculator/index.php).

In accordance with this suggestion, we now report C-peptide-based HOMA-2 values. This resulted in minor changes in the results (HOMA-2 contains a non-linear transformation that altered the ranking of the T0-T1 difference values) but did not affect the main findings of the study. CPR-IR and HOMA-IR values were removed.

Why did the authors use plasma glucose of 128 mg/dl as a cut-off for non-fasting? A commonly used cut-off is 126 mg/dl. Another reliable way to detect non-fasted individuals is unusually high C-peptide concentrations.

Please excuse that this phrase was misleading. Our cut-off was actually 126, but what was meant was that all three excluded individuals had a value of > 128 (i.e., 129, 131, 134). In the remaining sample, there was no glucose value higher than 125 on any of the four study days and we have now included this information in the Methods section: Three participants were excluded because of elevated glucose levels (≥126)

(7) Line 420: "two showed task behavior indicating...". I think there is a word missing.

This has been corrected.

(8) Line 422: "substantial movement". How was this defined? This could also be indicated in paragraph 522 ff.

The following information was added to the suggested paragraph: Image diagnostics was performed using visual inspection of image-to-image variability (tsdiffana, https://imaging.mrc-cbu.cam.ac.uk/imaging/DataDiagnostics).

This visualization revealed massive distortions in one participant that matched the logged observations from the scanning sessions. The reason for these artifacts was most likely due to the pronounced head circumference of this very large participant, which prevented him from being optimally positioned in the head coil.

(9) The authors conducted a median split on sugar content to divide the stimulus set into a high sugar (HS) and low sugar (LS) condition. What was the distribution of the sugar content in the food images? Who evaluated the sugar content of the stimuli?

The sugar content in all four stimulus sets was evaluated using the food database on fddb.info. Importantly, the four parallel stimulus sets did not differ significantly w.r.t. mean and median sugar content. This is particularly important for the sets used at placebo and insulin at T0 and T1 respectively. The use of the first two sets was counterbalanced for IN/PL at T0, the third and fourth sets were counterbalanced for IN/PL at T1. Information on the distribution of the sugar content for each set has now been added to the Methods and the Appendix (Appendix Figure 5).

(10) Line 433: The exclusion criteria include certain illnesses. Was this assessed by a physician or self-reported? Please add this information.

Initial screening as well as all clinical measurements in this study were performed by a physician (P.F., K.G.). This information was added in lines 433 ff.

(11) Line 440 ff: What is meant by "additional financial incentives were provided [..] for successful weight loss"? Did participants receive more money if they lost more weight? Please clarify.

Participants received an extra bonus of 50€. We have now added this information.

(12) Line 465: How was the individual reduction by 10 to 15% calculated? Please specify. Was resting energy expenditure used? Or was it based on their daily prior consumption? The authors mention dietary records. How many days were recorded and at which time points of the study? How was compliance assured?

The individual reduction was based on participants’ daily prior consumption as assessed by dietary records. For these records, participants protocolled their daily food intake for one week (7 days).

Compliance was assured by three telephone contacts after 2, 6 and 12 weeks during which participants confirmed that they continued to adhere to the diet and had the opportunity to clarify any questions regarding the dietary regimen. We have added this information.

(13) Line 516: it is not clear what 30_ steeper means. Please clarify.

Thank you, it should be 30° steeper, and has been corrected.

(14) Line 502: The fMRI food cue paradigm relies on images selected from the internet. Please indicate from what platform these images were selected. The images were then categorized based on sweetness. Were the pictures matched for physical and psychological variables? The authors only mention that the mean preference ratings of the stimuli were investigated in an independent sample, but other properties could play an important role (e.g. arousal, valence, and complexity of image).

Stimuli were not selected from a particular platform. They were selected via a Google search from a predefined list that we had created to ensure that a similar amount of savory and sweet foods were included. Sugar in all four stimulus sets was evaluated using the food database on fddb.info. Sugar content did not differ between sets. The correspondence of “valence” between the four stimulus sets was previously validated in an independent sample of 16 participants using liking ratings. We have now made a further subdivision into HS and LS stimuli and, as a kind of approximation to “arousal” / visual complexity, we have done the same for picture saliency (line 516 ff). The data are presented in the appendix (Appendix Table 2), and show no difference between stimulus categories and the sets.

(15) The fMRI data analyses section is described in detail up to the second level model.

Line 540: For the second level model, the authors indicate that they used one-sample t-tests and two-sample t-tests. Please be more specific about which contrasts (differential images) went into the model.

The following was added to the Methods section: Contrast images of interest comprised onset and parametric regressors for food > non-food and HS > LS from the placebo session at T0 (general paradigm-induced activation) and parametric regressors for HS > LS covering insulin effects at baseline (PL > IN) and compared to follow-up (T1 > T0).

In addition, we made sure that all contrasts that went into second level summary statistics are now specified in the result section and the figure legends.

(16) Based on the described results, I would have expected paired t-tests (T0 versus T1) or a flexible factorial model to investigate within-subject and between-subject effects in one model. Please clarify.

We preferred the more conservative summary statistics (T-Tests) to factorial designs primarily for the following reasons: Mixed-effects models (i.e., with both within-subjects and between-subject factors, as in our study) require full variance partitioning. Repeated measures ANOVAs (i.e., flexible designs) in SPM provide only one variance term, in this case, within-subject variance. This means that the results of group main effects as well as individual condition main effects can be inflated and therefore require separate models (for more details on this topic, see McFarquhar: Modeling group-level repeated measurements of neuroimaging data using the univariate general linear model. Frontiers in Neuroscience, 13, 2019). Summary statistics are robust to potential misspecification, and since we were primarily interested in analyses with covariates (i.e., simple regressions), we set up all the relevant contrasts at the single subject level and submitted each of the contrast images in a separate one- or two-sample T-test which corresponds to results from an ANOVA with partitioned error terms.

(16) The method section is lacking a statistical methods paragraph for the regression analysis and behavioral data analyses. What tool was used, etc.?

Thank you, this paragraph has been added.

(17) The authors used one-way ANOVAs for analyzing parametric liking scores for the different picture sets. What is the distribution of the liking ratings? Are they normally distributed? The authors computed several rmANOVAs. Were the assumptions for such analyses tested?

There are three main assumptions for rmANOVAs:

1) Independent observations: Given for individual subjects.

2) Sphericity of all difference scores among test variables: We used only two-level within subject factors (i.e., pre vs. post, PL vs. IN, HS vs. LS), that is, testing for sphericity of all differences (e.g., using Mauchly’s Test) is not applicable. In other words, there is no possible violation of this assumption.

3) Normality: We now have tested for normality for all variables of interest both across participants and within groups using the Kolmorgorov-Smirnov-Test (line 502 ff). The main results here are:

HS and LS liking scores before and after the scan, at T0 and T1, are normally distributed across participants and within groups (all P >.098).

HOMA-2 scores at T0 and T1 and difference scores across and within groups are normally distributed (all P >.20)

We also tested all other variables included in control analyses (hunger, time fasted, glucose, insulin, c-peptides). Here, we found at least one single measure (from pre or post, PL or IN, T0 or T1) was not normally distributed for fasting time, hunger and glucose. Nonparametric tests are now given for these variables (Table 1, Methods line 503 ff, Appendix Tables 1 and 5). As expected, the results did not differ from those of the parametric testing.

BMI reduction was correlated with several measures (predictors of subsequent weight loss). In this (18) case, you must correct your p-values in relation to the number of tests (Bonferroni correction).

We have now restricted the reported correlations to those related to our main hypotheses on weight loss prediction, i.e., one correlation with peripheral insulin sensitivity and one correlation with neural insulin sensitivity. The neural results are already corrected by family wise error (FWE) correction. In addition, the multiple regression model was used to test the predictive value of both parameters within a single model (partial correlations). To address the reviewer’s concern, however, we Bonferroni corrected the regression coefficients. Both coefficients survived the corrected threshold in our model of interest. In the control analysis with BMI as a third predictor, peripheral insulin sensitivity survived only an uncorrected threshold.

Were the correlations between the BOLD response and peripheral insulin resistance still significant after adjusting for BMI and age?

As suggested by the Reviewer, we performed a control analysis including BMI and age as covariates in the regression model of central and peripheral insulin changes. This analysis again revealed significant results in the left NAc with an identical peak voxel (-10, 10, -7, P <.05, FWE corrected). This information was added to the results. For standardization, we also performed this control analysis for baseline findings in the VTA. Again, the addition of the covariates age and BMI did not result in a difference for the significant BOLD response in the VTA.

(19) Was the study registered, for example at clinical trials.gov? Were the hypothesis and primary outcomes or study design preregistered? How was weight loss success defined?

Unfortunately, when the study was initiated in 2015, we did not pre-register the project (as we do by default today). This study is based on a peer-reviewed grant proposal where hypotheses are specified and that can be provided by request. It is now post-hoc –registered, but we admit that this is not very satisfactory. Basically, the project was not designed as a clinical trial with no primary focus on clinical outcomes but as a basic research study of how peripheral and central insulin function relates to weight changes in this particular population. For this reason, weight loss (outcome) was only defined by BMI changes.

(20) The study by Tiedmann et al., 2017 Nat Comm used the same fMRI paradigm to investigate insulin response of the reward circuitry in young adults of normal weight as well as overweight and obesity. In that study, the authors used HOMA-IR as a measure of peripheral insulin resistance. Why change this in the current study?

The reason for using a c-peptide based measure was that in the current study we observed a surprisingly high within-subject variability in insulin compared to c-peptide levels in some participants. This was also statistically relevant, as evidenced by significantly higher coefficients of variance compared with c-peptide levels. Due to this variability together with reports that c-peptide based indices are a more reliable indicator of insulin secretion, we preferred a c-peptide based measure and are grateful for the reviewer’s suggestion to use HOMA-2 as a further updated model of HOMA-IR.

Furthermore, in the previous study by Tiedemann, DCM model was used to show insulin-induced changes in connectivity between the nucleus accumbens and VTA. This connection was significantly modulated by intranasal insulin. Why wasn't this evaluated in the current study?

In Tiedemann et al., 2017, we were interested in whether consistent animal findings on insulin action in mesolimbic networks / connections can be translated into the human brain and our DCM results mainly confirmed this translation. In the current study, we used these results to define regions of interest for studying individual predictors and consequences of weight changes. In fact, we had no hypotheses for a generally different basic insulinergic mechanism in this particular sample. But if there are hypotheses of age-related changes in basic insulin function, this should first be tested in healthy, lean older (compared with younger) adults, perhaps involving structural measures. We agree that this could be an interesting question to address in the future. However, the focus of the current project was on longitudinal aspects of the interactions between insulin and weight, and the amount of data and analyses reported here are already quite complex.

(21) The study of Tiedemann published in 2017 investigated the response of intranasal insulin in young individuals, while the current study included elderly persons. The impact of age independent of BMI has not been investigated thus far on neural insulin processing. Have you looked at the effect of age on the nucleus accumbens or VTA BOLD response to insulin?

We agree that age effects per se are very interesting. At the moment, we do not have data from healthy, lean, older adults to examine this aspect, but we hope to address this question in an upcoming project. At least, as suggested by the reviewer and addressed in response #18, our results hold when including age and BMI as covariates.

(22) The authors use the term insulinergic inhibition. Is it possible to evaluate neural inhibition with BOLD fMRI?

The term insulinergic inhibition is based primarily on our neural and behavioral findings showing (a) reduced food preference, and (b) reduced BOLD signal under insulin compared to placebo. This fits strikingly with data in rodents showing insulinergic inhibition of neural activity in eating-relevant mesolimbic systems (e.g., Stuber and Wise, Nature Neuroscience 2016). Of course, we are unable to differentiate direct or indirect inhibitory effects of insulin, but we believe that our current results, together with our previous data and animal findings, support this “inhibitory” interpretation.

Author details

1. Lena J Tiedemann

   Department of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Contribution

   Data curation, Formal analysis, Investigation, Visualization, Writing - original draft, Project administration

   Competing interests

   No competing interests declared

2. Sebastian M Meyhöfer

   1. Institute for Endocrinology & Diabetes, University of Lübeck, Lübeck, Germany
   2. German Center for Diabetes Research (DZD), Ingolstädter Landstraße, Germany

   Contribution

   Methodology, Writing - original draft

   Competing interests

   No competing interests declared

3. Paul Francke

   Department of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Contribution

   Data curation, Investigation, Writing - original draft, Project administration

   Competing interests

   No competing interests declared

4. Judith Beck

   Department of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Contribution

   Data curation, Investigation, Project administration

   Competing interests

   No competing interests declared

5. Christian Büchel

   Department of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Contribution

   Conceptualization, Funding acquisition, Writing - original draft

   Competing interests

   Senior editor, eLife

   "This ORCID iD identifies the author of this article:" 0000-0003-1965-906X

6. Stefanie Brassen

   Department of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Contribution

   Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Visualization, Methodology, Writing - original draft, Project administration

   For correspondence

   sbrassen@uke.de

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8884-7593

• 359

  Page views

• 70

  Downloads

• 2

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.