Central insulin modulates food valuation via mesolimbic pathways

Abstract

Central insulin is thought to act at the neural interface between metabolic and hedonic drives to eat. Here, using pharmacological fMRI, we show that intranasal insulin (INI) changes the value of food cues through modulation of mesolimbic pathways. Overnight fasted participants rated the palatability of food pictures and attractiveness of non-food items (control) after receiving INI or placebo. We report that INI reduces ratings of food palatability and value signals in mesolimbic regions in individuals with normal insulin sensitivity. Connectivity analyses reveal insulinergic inhibition of forward projections from the ventral tegmentum to the nucleus accumbens. Importantly, the strength of this modulation predicts decrease of palatability ratings, directly linking neural findings to behaviour. In insulin-resistant participants however, we observe reduced food values and aberrant central insulin action. These data demonstrate how central insulin modulates the cross-talk between homeostatic and non-homeostatic feeding systems, suggesting that dysfunctions of these neural interactions may promote metabolic disorders.

Figure 1. Outline of study design and experimental task.

(a) Experimental protocol. (b) Schematic representation and timing of the experimental paradigm. On each trial, a food or non-food picture (pseudo-randomized) was presented for 4 s. During the first 3.5 s, participants had to indicate the general liking of the depicted item, by pressing one of two buttons. Feedback of the chosen answer was provided for 0.5 s. After a fixation period of 3 s and during a maximum duration of 3 s, participants were asked to detail their preference rating, that is, how much they like (this example) or dislike (see Supplementary Fig. 1) the item using a four-point rating scale, by pressing one of four buttons. After another feedback of 0.5 s, the trial ended with a random fixation period of 4–6 s. (c) Examples of less palatable food and non-food stimuli.

Figure 2. Behavioural results in the placebo condition.

(a) Groups means and s.e.m. of percentage of liked food and non-food items during placebo demonstrate reduced food value scores in IR. (b) Group means and s.e.m. of combined preference values ranging from 1 (‘not at all’) to 8 (‘very much’). The dashed line separates ‘yes’ from ‘no’ decisions. (c) Correlation between individual post-scan plasma insulin levels and preference values for food items during placebo. Only in NIR plasma insulin predicted preference values.

Figure 3. Behavioural insulin effects and autonomic data.

(a) Group means and s.e.m. of preference values for food and non-food stimuli during placebo (PL) and insulin (IN). rmANOVA revealed significantly reduced preference values specifically of food items under INI only in NIR, whereas food values tend to increase in IR. (b) INI-mediated changes in food preference scores (placebo session minus insulin session) were directly correlated to individual peripheral insulin sensitivity as defined by the HOMA index, across all participants. The lower the HOMA score the more food values were decreased under insulin. Group means and s.e.m. of heart rate (c) and respiration (d) recorded during the placebo and the insulin session.

Figure 4. Paradigm-induced activation patterns.

(a) Mesolimbic ROI in the NAc (red) and the VTA (blue) overlaid on the mean structural image of all participants. (b) Categorical effect of food stimulus presentation. Greater activity in the insula, amygdala, orbitofrontal cortex, VTA and hypothalamus was observed in the food compared to the non-food condition across both groups. (c) Neural representation of preference values (parametric analysis). Regions in which the correlation with the preference value was significant in both the food and the non-food conditions across all participants included the ventromedial prefrontal cortex, posterior cingulate cortex and the bilateral NAc. (d) Group differences in food value signals. NIR demonstrated increased food value signals in the left NAc relative to non-food signals compared to insulin-resistant individuals (IR). Plotted contrast: NIR_{food>non-food}>IR_{food>non-food}. The bar graph shows group means and s.e.m. of mean parameter estimates extracted from the ROI of the left NAc. All peaks are P<0.05 FWE corrected. Activations are overlaid on the mean structural image of all participants (display threshold P<0.005 uncorrected).

Figure 5. Neural insulin effects.

General linear modelling of value signals from the placebo and the insulin session revealed group by session interactions in the bilateral NAc (a) and the left VTA (b). In both regions, food value signals were decreased in the insulin session only in NIR, while signals were increased under INI in IR. Bar plots show group means and s.e.m. of mean contrast estimates extracted from ROIs from the comparison NIR_PL>IN > IR_PL>IN. Displayed P-values are FWE corrected for bilateral masks. Activations are overlaid on the mean structural image of all participants (display threshold P<0.005 uncorrected).

Figure 6. DCM results.

(a) The full model that was defined and inverted for each participant included all (intrinsic) self-connections and (extrinsic) forward and backward connections, food and non-food stimuli as driving inputs, and insulin modulation of all extrinsic and intrinsic connections. (b) The winning model: the reduced model with the highest evidence across all participants as identified through post-hoc optimization includes modulation of the VTA-NAc forward connection and of self-connections by insulin. Data are given as Bayesian parameter averages. These parameters indicate positive connection strength between the VTA and the NAc, and insulin to selectively inhibit forward VTA-NAc connections as well as intrinsic self-connections of both regions. (c) The upper graph shows the range of log-posterior probabilities of all possible models examined in the left hemisphere. The lower graph shows the posterior probability of the reduced model which had the posterior probability of (almost) 1 suggesting that the reduced model had more evidence than any other variant. The next most probable model’s probability was very low (almost 0, the log-probability was −76.88). (d) Correlation between neural and behavioural insulin effects. Individual parameter estimates of INI modulation on the forward connection from the VTA to the NAc correlated with insulin-mediated changes of preference scores for food relative to non-food items (Δ insulin effect food−non-food). Inhibitory modulation predicted stronger decrease of food values under INI across all participants. (e) The winning model selected for the different groups: within IR, there was no significant modulation by INI of the VTA to NAc forward connection.