Intranasal insulin administration decreases cerebral blood flow in cortico‐limbic regions: A neuropharmacological imaging study in normal and overweight males

To assess and compare the effects of 160 IU intranasal insulin (IN‐INS) administration on regional cerebral blood flow (rCBF) in healthy male individuals with normal weight and overweight phenotypes.


| INTRODUCTION
Obesity has been classified as a major public health issue and its prevalence continues to increase, with more than approximately twothirds of adults in the UK currently living with obesity and overweight (OW). 1 Coupled with this is the increase in type 2 diabetes (T2D), a disease associated and largely defined through an insensitivity to the peripheral effects of insulin, termed insulin resistance. 2 Insulin effects in the peripheral system have been well evaluated and researched, however, insulin's effects on brain function have yet to be fully elucidated. A body of behavioural and neuroimaging literature has shown a modulatory role of insulin in mechanisms linked to appetite control and food intake. 3 Some of these studies have delivered human insulin solution via the nasal cavity, a procedure commonly termed intranasal insulin (IN-INS) administration. This method permits direct brain administration, circumventing peripheral blood glucose regulation and control, 4 increasing cerebrospinal fluid (CSF) levels within 30-60 minutes. 5 In comparison with intravenous and oral methods, IN-INS is an attractive tool for assessing the effects of insulin on brain function because of its limited effects on peripheral glucose concentration and other metabolic gut-derived hormones. 6,7 Previous functional neuroimaging research with IN-INS has shown little direct interaction with cerebral vasculature, 7 suggesting that insulinassociated effects seen from neuroimaging techniques, which take advantage of neurovascular coupling, are a product of neuronal insulin signalling. 8 Given this, functional MRI (fMRI) methods in combination with intranasal delivery offer a valuable way to investigate the effects of insulin on brain function.
OW is a term that describes an individual with a body mass index (BMI) above the range considered healthy or desirable. The OW phenotype is defined as a BMI range of 25-29.9 kg/m 2 and in medicine is considered a precursor to obesity (a BMI of 30 kg/m 2 or higher). Obesity is associated with many morbidities, including T2D and increased cardiovascular risk. 9 Importantly, population studies have shown increased T2D diagnoses, 10 cardiovascular-related death 11 and overall mortality 12 with incremental BMI increases above 25 kg/m 2 , classifying those who are OW (but not obese) as an at-risk population for the aforementioned morbidities. In England, 40% of adult males are OW as assessed by BMI, with 33% of individuals classified as normal weight and 26% as obese, therefore the majority of males in England are classified as OW. 1 There is considerable interest in the impact of IN-INS in those individuals with impaired appetite control. Reports show a lack of effective weight loss following long-term IN-INS administration in men with obesity compared with men of normal weight who achieved a significant weight loss. 13 To our knowledge, this is the first study of the impact of IN-INS on people in this intermediate non-obese category exclusively. Studying this group of individuals is important for establishing possible markers and differences in brain function and/or central insulin sensitivities that may be associated with this healthy, but potentially at-risk population.
A small number of experiments looking at the modulatory effects of IN-INS have employed arterial spin labelling (ASL) to explore regional cerebral blood flow (rCBF) changes in normal weight, 6 obese, insulin-resistant 14 and also elderly individuals. 15 Measures of rCBF with ASL are quantitative (ml/100 g of brain tissue/minute) and seemingly physiologically relevant when examining drug-induced effects in the brain. 16 The present study is, to our knowledge, the first to measure rCBF in response to IN-INS versus IN-placebo (IN-PLA) in a group of healthy normal weight and OW individuals as determined by BMI.
Based on the previous literature, using this dose, we hypothesized that 160 IU IN-INS would produce significant changes in rCBF within the limbic and cortical structures that express insulin receptors. 17 We further predicted that OW individuals may display decreased central insulin sensitivity and this would be observed as less pronounced changes in rCBF in contrast to normal weight comparators.

| Participants
The study followed the guidelines in the Declaration of Helsinki and was approved by the King's College London Psychiatry Nursing and Midwifery Ethics Committee (RESCM-17/18-2282). Written, informed consent was signed prior to any study procedures. The study comprised three visits. The first was a screening session and the remaining two visits were experimental imaging sessions. The two imaging sessions were separated by 1 week.
Healthy right-handed male participants were screened to ensure they had no history of psychiatric illness or diabetes, no cardiacrelated complications, no history of any eating disorders, asthma or allergies associated with breathing difficulties, and did not smoke more than five cigarettes per day. During the screening, height and weight measurements were taken to ascertain BMI. Only men with a BMI of between 18.5 and 30 kg/m 2 were recruited then stratified into two age-matched groups defined by BMI as either below (normal weight: lean) or above (OW) a BMI of 25 kg/m 2 , for analysis, respectively.

| Imaging sessions
For both of the imaging sessions, participants were instructed to follow an overnight (12-hour) fast, with their last meal to be consumed no later than 10:00 PM the night before the study visit. Participants abstained from alcohol consumption the night before and caffeine consumption each morning. Shortly after their arrival, participants provided a blood sample via venepuncture from the cubital vein (referred herein as predose) and a second sample after the MRI protocol (referred herein as postscan), 2.5 hours apart. Blood samples were analysed for plasma glucose, serum insulin and serum C-peptide to assess the effect of IN-INS administration on peripheral concentrations.

| Intranasal administration
Thirty minutes prior to functional image acquisition, participants received either 160 IU insulin (Humulin, 500 IU/mL, Eli Lily, USA) or saline solution 0.9% w/v (placebo) using a commercial spray device that had been tested and characterized with the used dosage. 18 Administration was timed so that data acquisition coincided with peak insulin concentrations in the central nervous system, in accordance with the pharmacokinetics of IN-INS previously reported. 5 Administration was performed using a commercial pump with suitable spray characteristics for nose to brain delivery of insulin solution, using an identical dose to this report. 18

| Blood analysis
Venous blood samples were spun in a centrifuge (10 minutes at 1000 rpm). Plasma and serum were extracted into aliquots and stored at −20 C. Biochemical analysis was performed using routine assays to ascertain serum insulin and C-peptide (Siemens Healthcare Centaur Assays) and plasma glucose (Siemens Healthcare AVIDA) concentrations, respectively. Baseline measures (predose) of insulin sensitivity for each participant were calculated using the homeostatic model assessment of insulin resistance (HOMA-IR) 2 model. 19 HOMA-IR can be calculated using plasma glucose and serum insulin concentrations or plasma glucose with serum C-peptide concentrations. For this study, the latter was implemented using the online, publicly available HOMA-IR 2 calculator v. 2.2.3 (https://www.dtu.ox.ac.uk/ homacalculator/). Average HOMA-IR scores across visits were calculated and compared between groups (unpaired t-test).
The change (Δ) in concentration between predose and postdose collection periods was calculated for each metabolite. These Δ values, for each metabolite, were entered into a mixed effects analysis of variance (ANOVA) model to interrogate the main effects of 'Treatment', 'Group' and any interaction effects. Significance thresholds for main effects and interaction effects were set to P less than .017 (0.05/3) to correct for multiple comparisons. Main effects were interrogated with planned comparison t-tests as a post hoc analysis (within-group or within-treatment), and interaction effects with Tukey tests. Planned comparison test significance thresholds were set to P less than .025 to account for the two tests.

| Hunger scores
Hunger scores were assessed in the scanner immediately after perfusion image acquisition using a visual analogue scale ('How hungry do you feel right now?': '0' = not hungry at all to '100' = very hungry).
Similar to the blood analysis above, these scores were run through a factorial model (ANOVA) to look for treatment, group and interaction effects. Significance thresholds were set to P greater than .05. Post hoc analysis was the same as implemented for blood analysis.

| Image acquisition
Scanning was conducted using a 3 T MR750 GE Discovery Scanner Following structural image acquisition, whole-brain CBF data were collected using a 3D pseudo-continuous ASL (pCASL) sequence acquired with a fast spin echo stack of spiral readout. The following parameters were used for the readout: 10 spiral arms, 600 points per arm, leading to an equivalent in-plane resolution of 2.94 × 2.94 mm 2 , slice thickness = 3 mm, 54 slices and TE/TR = 11.8/7325 ms. For the perfusion labelling module, the following parameters were used: label duration = 3500 ms, postlabel delay (PLD) = 2025 ms, four background suppression pulses and two pairs of 'control and labelled' images. The total acquisition time was 5 minutes and 37 seconds. A 3D proton density (PD) image was acquired as part of the same image series, using identical readout parameters. This permitted rCBF quantification in standard physiological units, following the methodology recently recommended by the ASL community. 20 Participants were instructed to look at a fixation cross displayed to them via a projector screen.

| Perfusion image processing and analysis
Quantitative CBF maps were computed from perfusion-weighted (PW) and PD images using the single compartment model and online GE scanner software, according to the formula for single PLD pCASL data from the ASL consensus paper. 20

| Image registration and processing
Image processing was conducted using a bespoke pipeline consisting of a mixture of processing software. First, 3D T1-weighted images were combined to create a group template (templatecreate, advanced normalization tools software 21,22 [ANTs]), which was registered to standardized Montreal Neurological Institute (MNI) space. All individual subject transformation and warp matrices from these steps were saved for later application.
PD images, which are in perfect registration, and boast higher tissue contrast, to both CBF maps and PW images, were co-registered to subject-specific T1-weighted images (epi-reg, FMRIB Software Library, v. 3.20, University of Oxford, UK; http://www.fmrib.ox.ac.uk/fsl). Subject CBF maps were normalized to MNI space by applying the saved transformation matrices (antsapplytransforms, ANTs) and smoothed using a full width at half maximum Gaussian kernel of 6 × 6 × 6 mm 3 with statistical parametric mapping (SPM) software (SPM-12, Wellcome Trust Centre for Neuroimaging, University College London, UK; www. fil.ion.ucl.ac.uk/spm). This smoothing kernel, of approximately twice the acquired spatial resolution, was implemented in reference to recommendations for group statistical inferences with functional blood oxygen level-dependent (BOLD) data. 23  contrasts, 'Treatment' x 'Group'. Voxel-wise whole-brain analysis results were created from a cluster-forming threshold of P less than .001. Significant clusters were determined based on correction for multiple comparisons computed from 'cluster extent' statistics 24 using a family-wise error threshold (FWE) of P less than .05.

| CBF image analysis
In response to a significant treatment or interaction effect (P < .05, FWE), second-level, whole-brain, post hoc paired t-tests were created for each group for the appropriate treatment directionally, using the same cluster-forming and significance criteria described above.
Whole-brain group effects were assessed using separate withintreatment, two-sample t-tests with the same statistical criteria mentioned above and with global GM added as a covariate. Median CBF values for each individual subject and ROI were extracted (3dmaskave, Analysis of Functional Neuro Images) and entered into a repeated measure ANOVA (rm-ANOVA) statistical model. To correct for multiple comparisons, a Bonferroni significance threshold was set to P less than .0125 (0.05/no. of ROIs) for these tests. Following a significant main effect of treatment or group, planned comparison t-tests were conducted to interrogate treatment (paired) or group (unpaired) related changes. These t-tests were referenced as planned comparisons and formed a post hoc analysis, as opposed to testing all possible combinations. Significance for these planned comparisons was set to P less than .025. Finally, we investigated if global GM CBF was affected by IN-INS or differed between groups to ascertain whether significant effects from ROI analysis may be attributed to global changes and not regional changes in CBF. To this end, GM CBF was extracted from the GM mask employed during the whole-brain analysis and run through an rm-ANOVA model, just like the ROIs above.

| Presentation of statistical results
Summary data are presented as mean ± standard deviation (SD), tabulated and in graphical formats. Blood and ROI statistical analyses were conducted using R statistical analysis software (Rstudio v. 1.1453; Boston, MA, USA; http://www.rstudio.org/).

| RESULTS
Thirty participants completed the study. Of these, three were excluded for violation of the fasting study protocol as judged by serum insulin levels (>50 mIU/L); another participant was excluded after presenting with an extreme lack of sleep prior to one of the study visits. The remaining 26 subjects were divided into two groups: normal weight (lean, n = 12, BMI = 22.40 ± 1.89 kg/m 2 ) and OW (n = 14, BMI = 27.76 ± 1.92 kg/m 2 ). The two groups were matched for age (27.00 ± 5.44 and 24.76 ± 4.30 years, respectively; P = .30, unpaired t-test) (demographic data are presented in Table 1).

| Blood analysis
HOMA-IR values did not significantly differ between groups (Table 1). Changes in plasma glucose and serum insulin did not reveal any significant group, treatment or interaction effects ( Table 2).
Analysis of Δ serum C-peptide concentration revealed a main treatment effect (F [1,46]

| Hunger score analysis
Summary hunger scores for each group and each treatment are displayed in Table 2. Hunger scores were comparable across treatment and groups and no significant effects differences were identified or interaction effects.

| Whole-brain analysis
Whole-brain, two-sample tests (within treatment) did not provide any significant clusters for either contrast (lean > OW or lean < OW) fol- Post hoc testing of the subcontrasts forming this interaction is described below. The opposite interaction contrast did not provide any significant results.  Table 3; statistical maps are presented in Figure 1. Note: Changes in C-peptide were significantly different between treatments, a greater suppression of C-peptide following IN-INS compared with that of IN-PLA (ANOVA). Post hoc within-group analysis (paired t-test) showed that this treatment effect was significant only in the lean group but not OW (highlighted in bold). For blood analysis whole group n = 25, lean n = 12, OW = 13. *P < .017, data are presented as mean ± standard error. Hunger scores recorded show no treatment, group or interaction effects.

| ROI analyses
All ROI CBF values are displayed in Table 3. There were no significant treatment, group or interaction effects on global GM CBF. ROI analysis was performed using four previously defined anatomical ROIs.
These values were compared across treatment conditions and also across groups to examine rCBF differences through creation of individual ROI rm-ANOVA models.
No interaction effects or main treatment effects were reported from any of the anatomical ROIs tested; however, significant grouprelated differences in rCBF were seen for the insula (F [1,47]   we have decided that the perfusion differences discussed here can be interpreted to indicate changes in regional neuronal activity.
From an anatomical perspective, the treatment-related changes in rCBF observed from the whole-brain analyses accord with the findings of previous investigations of the effects of centrally acting insulin and also insulin receptor distribution in the brain. 6,7,14,15,26,27 As seen from the results presented in this study there is a difference in observed effects on rCBF between normal weight and OW individuals, and this will be discussed.
Whole-brain statistical parametric maps in OW group analysis dis- In addition, whole-brain analysis in the OW group showed a large cluster of significant rCBF change at the boundary between the left putamen and the left insula. The putamen, a prominent region within T A B L E 4 Mean extracted rCBF values calculated from the ROIs tested for each group and treatment arm Note: ROI analysis consisted of a group × treatment ANOVA model for each region with HOMA-IR as an added covariate. P-values displayed for main group, treatment and group × treatment interaction effects. The insula showed a significant effect of group on rCBF. *P < .05. **P < .01; lean n = 12, OW n = 14.
F I G U R E 2 rCBF measures extracted from the insula. A, ROI analysis revealed a significant main effect of group where the OW group displayed greater rCBF than the lean group. B, post hoc testing showed that under IN-PLA conditions this group effect is significant but this difference significance under IN-INS conditions. Group average median rCBF values plotted ± SD, with individual subject values (white dots) transposed to show rCBF variability. *P < .025; **P < .01; lean, n = 12; OW, n = 14; OW, overweight the limbic system and reward circuitry, has been shown previously to exhibit increases in rCBF following nasal administration of 40 IU insulin in participants of normal weight. 6 The results reported in our study are comparable with regionality but differ in the directionality of rCBF change. This difference could be a result of the higher dose that was implemented in this study (160 IU) or possibly highlights a differential response to IN-INS in this OW group compared with a normal weight group, despite no significant changes being observed in the normal weight group from our study. The higher dosage implemented in this study was decided based on a previous paper by Kullmann 27 The authors showed and summarized that acute quantification of regional insulin effects with fMRI requires higher doses (160 IU) and that there was a prominent dose-dependent effect with the strongest effects identified from 160 IU dosages. 27 Using this summary and recommendation we opted to implement a 160 IU dose in this study to try and achieve a pronounced fMRI-CBF effect.
The insula is known for its role as a hub for integration of visceral stimuli such as taste and odour, commonly referred to as the primary gustatory cortex, 36 and boasts a high density of insulin receptors. 37 As central insulin activity has shown anorexigenic effects, 38  within a food cue paradigm. 39 The directionality of the treatment effects supports our data but it should be noted that these observations were made in lean individuals while engaged in a visually stimulating task. The fusiform gyrus has been termed an 'insulin-sensitive' brain region from several reviews, [40][41][42] for which the data reported here provide further support.
Despite OW group treatment effects seen at the whole-brain level, ROI analysis failed to provide any treatment-related differences in rCBF. An explanation for this could be that the a priori regions were bilateral and the results from the whole-brain analysis revealed lateralized IN-INS effects.
The stratification of individuals using BMI allowed the study of individuals in a metabolic state between normal weight and obesity, OW, who could be at risk of developing obesity along with its associated complications. 43 These two groups did not differ in age or peripheral insulin sensitivity. Studying this group is important for establishing possible markers and differences in brain function and/or central insulin signalling that may be associated with this otherwise healthy population. The authors note that by comparing two groups that do not dramatically differ in BMI there is a potential reduction in contrast between groups. In light of this we find that the grouprelated effects from the ROI analysis are particularly relevant and of interest. No group effects were seen from the whole-brain analysis and could be attributed to a loss of statistical power when identifying peaks across the whole brain in comparison with a more refined ROI which was greater than in the lean group, may explain why this was not significant. In addition, insulin resistance in the OW group (although not statistically significant) was slightly higher versus lean and could arguably be a reason why this effect is not seen.
In addition to the dose-dependent effects presented by Kullmann et al., there appear to be differential regional effects on resting state brain activity, as illustrated by IN-INS-related increases in the prefrontal cortex and amygdala, and decreases in the caudate and hypothalamus. 27 Taking all these results into consideration, this makes interpretation of our results, which differ in dosage, cohort and type of insulin, somewhat difficult to situate within the published literature.
Of note, there are methodological differences in the literature, not just concerning the aforementioned points but also the implementation of different types of pumps (powder-and solution-based), which ultimately lead to different administration profiles and subsequent central effects. 18 The results gathered from this examination of the effects of

IN-INS administration on cerebral perfusion indicate that insulin is
centrally active in this group of OW individuals. We cannot exclude that the lack of average effect in the lean group is possibly attributable to an insufficiently large sample required for the underlying effect size; however, we would prefer to posit a more biological explanation.
IN-INS administration may best represent, physiologically, the insulin concentrations that may be seen in the postprandial state following a mixed meal, albeit without the changes in carbohydrate, protein, fat, ghrelin and the plethora of orexigenic gustatory hormones. Regional CBF responses in the insula, striatum (putamen and caudate) and also the hippocampus and fusiform gyrus have been shown to be significantly reduced in response to glucose and fructose ingestion, respectively, in men and women of normal weight. 49 Under this framework and given the lack of effect in the normal weight group, we could suggest that gustatory innervation and/or associated changes in postprandial hormone profiles are necessary for insulin signalling effects on rCBF and that resting neural activity is unaltered by IN-INS in the fasted state. On the other hand, the OW group may, however, have higher resting rCBF in the regions seen from this analysis as a result of factors, not measured, which may lead to insulin-induced reduction in rCBF, similar to that seen under postprandial conditions. 49 This work was performed exclusively in male participants. Previous work has shown the effects of central insulin in men and women to occupy a differential response, 38