Blood-Brain Glucose Transfer in Alzheimer’s disease: Effect of GLP-1 Analog Treatment

There are fewer than normal glucose transporters at the blood-brain barrier (BBB) in Alzheimer’s disease (AD). When reduced expression of transporters aggravates the symptoms of AD, the transporters become a potential target of therapy. The incretin hormone GLP-1 prevents the decline of cerebral metabolic rate for glucose (CMRglc) in AD, and GLP-1 may serve to raise transporter numbers. We hypothesized that the GLP-1 analog liraglutide would prevent the decline of CMRglc in AD by raising blood-brain glucose transfer, depending on the duration of disease. We randomized 38 patients with AD to treatment with liraglutide (n = 18) or placebo (n = 20) for 6 months, and determined the blood-brain glucose transfer capacity (T max) in the two groups and a healthy age matched control group (n = 6). In both AD groups at baseline, T max estimates correlated inversely with the duration of AD, as did the estimates of CMRglc that in turn were positively correlated with cognition. The GLP-1 analog treatment, compared to placebo, highly significantly raised the T max estimates of cerebral cortex from 0.72 to 1.1 umol/g/min, equal to T max estimates in healthy volunteers. The result is consistent with the claim that GLP-1 analog treatment restores glucose transport at the BBB.

Group Regressions and Correlations at Baseline. Linear regressions among the variables determined at baseline are shown in Fig. 2: Analysis of CMR glc versus duration of AD revealed a negative correlation (P = 0.006, R 2 = 0.23, Fig. 2A) at baseline of both groups pooled. We observed a positive correlation between CMR glc and the total cognition score at baseline of all members of the two groups pooled (P = 0.006, R 2 = 0.23, Fig. 2B). We found tendencies of correlation between the T max estimates and the duration of disease (P = 0.05, R 2 = 0.12, Fig. 2C) at baseline of all members of the groups pooled, nor between the T max estimates and total cognitive score (P = 0.83, Figure 1. Estimates of T max of two groups before and after liraglutide and placebo treatment. Ordinate T max estimates. GLP-1 analog treatment very significantly (P < 0.0001) raised the average T max estimate in cerebral cortex as a whole. The resulting value of T max significantly exceeded the value reached by placebo treatment. R 2 = 0.002, not shown). We found positive linear correlation of glucose utilization fraction (GUF) and cognition (P = 0.01, R 2 = 0.21, Fig. 2D) at baseline of both groups pooled, but no correlation between GUF and duration of disease (P = 0.19, R 2 = 0.06, not shown), also of all members of the two groups pooled, nor between the unidirectional glucose extraction fraction (GEF) and cognition or duration (P > 0.35, R 2 < 0.03, not shown). We also found a positive linear correlation of CMR glc and T max (P = 0.02, R 2 = 0.18, Fig. 2E).
We observed a positive correlation between the net clearance of [ 18 F]FDG (and therefore of glucose as well) and the total cognitive score at baseline of both groups pooled (P = 0.02, R 2 = 0.17, Fig. 2F). The correlation of the net clearance of [ 18 F]FDG (and therefore of glucose) and the duration of AD was significantly negative (P = 0.007, R 2 = 0.22, not shown) at baseline for the members of both groups pooled.
Group Variables, Changes, and Differences During and After Treatment. Gejl et al. 27 reported a decrease in fasting plasma glucose in both groups and a significant difference of fasting plasma glucose levels between the two groups after six months of treatment (5.6 mM in the placebo group and 5.1 mM in the GLP-1 analog group). Fasting plasma glucose levels in the heathy group was 5.8 30 . We also noted significant reduction of systolic and diastolic blood pressures in the liraglutide group at the end of the study period. Relationship between duration of AD and CMR glc (A), total cognitive score and CMR glc (B), duration of AD and T max , (C), total cognitive score and GUF (D), T max and CMR glc (E), and total cognitive score and net clearance of [ 18 F]FDG. The points representing the averages of the healthy control group were not included in the analysis.
SCIENTIfIC REpoRTs | 7: 17490 | DOI:10.1038/s41598-017-17718-y We observed no change of the cortical T max estimate (P = 0.24, mean difference 0.093 µmol/g/min, 95% CI of difference: −0.037; 0.22) in the placebo group during the six months of treatment. In the GLP-1 analog treatment group, the T max estimate increased significantly after the 6 months of treatment (P < 0.0001, mean difference 0.34 µmol/g/min, 95% CI of the difference 0.20; 0.49), as shown in Fig. 1 of both estimates. After the six months, the T max estimates of the two groups no longer differed significantly (P = 0.24). The estimates of T max increased significantly more in the liraglutide treated group than in the placebo group (P = 0.0002, mean difference 0.25 µmol/g/ min, 95% CI of the difference 0.13; 0.37), as shown in Figs 1 and 3. In healthy volunteers, the T max estimates averaged 1.022 umol/g/min, as shown in Fig. 3.
The maximum observable clearance is the ratio of an estimate of T max to the corresponding estimate of the Michaelis half-saturation concentration K t , i.e., the T K / t max ratio, equal to the abscissa intercept of the Eadie-Hofstee plot. The ratio is a measure of the highest attainable transport capacity at the present affinity of the transporters to glucose. The maximum clearance estimates increased with the duration of disease, as shown in Fig. 4 for the placebo group at baseline and 6 months, and for liraglutide treatment group at baseline. The six months of liraglutide treatment reversed the trend by returning the ratio to the value of the placebo group at baseline. The corresponding estimates of the Michaelis half-saturation concentration are listed in the legend to Fig. 1.
The estimates of net clearance of [ 18 F]FDG and glucose (K * and K) did not change in the placebo group (P = 0.53), but increased with borderline significance in the liraglutide group (P = 0.049), as listed in Table 1. We calculated the estimates of CMR glc listed in Table 1 by regional kinetic analysis. The estimates confirmed the decrease in the placebo group (P = 0.05) and unaltered CMR glc in the liraglutide group (P = 0.58) reported by Gejl et al. 27 . The CMR glc estimates of the healthy control group was 0.31 ± 0.062 umol/g/min 29 .

Discussion
Analogs of the glucagon-like peptide-1 (GLP-1) hold promise as a novel approach to the treatment of neurodegenerative disorders such as AD 31 . The analogs target cerebral metabolism and neurodegeneration in animal models, with potential translation to afflicted humans 27 . In the present analysis of findings obtained in patients with AD, we discovered a highly significant effect of the GLP-1 analog liraglutide on the blood-brain glucose transport capacity. In the patients, the estimates of blood-brain glucose transport capacity correlated with the estimates of glucose consumption (CMR glc ). We confirmed the natural disease progression in the study population, determined by a negative correlation between disease duration and neuronal activity measured as glucose metabolism and the tendency of the T max estimates to decline with increasing duration of disease. The estimates of metabolism in turn correlated positively with measures of cognition, including the trend towards decreased estimates of cortical CMR glc in the members of the placebo treatment group. The net clearances of [ 18 F]FDG and glucose correlated positively with cognition and negatively with duration at baseline and increased significantly Figure 3. The GLP-1 analog treatment appeared to reduce the effects of disease duration. The results are consistent with the claims (1) that the maximum blood-brain transfer capacity declines with duration of Alzheimer's disease, and (2) that GLP-1 analog treatment raises the GLUT1 activity in the barrier as a potential future target for treatment of the neurovascular dysfunction in Alzheimer's disease. Estimates of K t averaged 9.3 mM (± 0.28 RSDR, robust standard deviation of residuals) in the placebo treated group at baseline, 10.1 mM (± 0.19 RSDR) at six months, and 5.9 mM (± 0.18 RSDR) in the liraglutide group at baseline, and 11.1 mM (± 0.25 RSDR) at six months of treatment. The points representing the averages of the healthy control group were not included in the analysis.
with treatment in the liraglutide group. The glucose utilization fraction (GUF) correlated positively with cognition and in the treatment group numerically exceeded the corresponding estimates of placebo group.
Glucose transport across the BBB is the net result of glucose fluxes in the two directions across both membranes of the capillary endothelium, mediated by glucose transporters of which GLUT1 predominates 32 . The fluxes across the membranes of the BBB can be regulated (1) by changing the concentration gradients in the directions of the tissue, thereby changing the differences between the fluxes in both directions across the endothelial membranes, (2) by changing the affinity of the transporters separately or jointly for the multiple substrates 33 , or (3) by changing the number or density of transporters by insertion of new proteins independently of the capillary surface area 34 . The last of the three mechanisms determines the magnitude of the maximum transport capacity (T max ). The present results indicate that changes associated with the GLUT1 transporter can occur by changes of the density of GLUT1 or by changes of the half-saturation or affinity constants. In animal studies, reports suggest that regulation of GLUT1 is essential to preservation of proper brain capillary networking, blood flow, and endothelial integrity, as well as to neuronal function and structure 25 . Recent results imply that reduced glucose availability in the central nervous system directly triggers behavioral deficits related to the development of neuropathology and synaptic dysfunction mediated by hyperphosphorylated tau proteins 35 . The GLUT1 deficiency syndrome introduced by, among other authors of the report by DeVivo and Harik 36 , is an important example of an extreme version of GLUT1 deficiency that leads to a number of neurodegenrative features, of which Alzheimer's disease may be said to be a milder example.
The present results agree with reports of healthy subjects 26 , although little is known of the specific effects of GLP-1 and its analogs on brain glucose transport. The claim of a direct effect of GLP-1 or its analogs on the transport of glucose across the BBB therefore remains speculative, but recent work by Jais et al. 37 unveils the novel mechanism that vascular endothelial growth factor (VEGF) from macrophages at the BBB restores the presence of GLUT1 to baseline values, maintains CMR glc , and prevents loss of function. It is of particular interest that  GLP-1 is coupled to augmented VEGF generation 38 . The most abundant GLUT in the BBB is GLUT-1, but low levels of other GLUTs (including the insulin sensitive GLUT-4) have been reported 6 . The present observation of increased estimates of T max in principle can be explained by increase of the number of GLUTs, prompted by liraglutide, or by subsequent increased postprandial insulin levels. Liraglutide raises GLUT-4 levels in the periphery via an AMPK-dependent mechanism that is independent of insulin 39 , but recent reports suggest that astrocytic insulin receptors modulate GLUT1 expression, and consequently GLUT1 protein levels at the cell membrane 40 . Moreover, insulin signalling in hypothalamic astrocytes supposedly contributes to glucose sensing in the central nervous system and systemic glucose metabolism by regulation of glucose uptake across the BBB 41 , in agreement with the present results, as GLP-1 potentially co-regulates the expression of hypothalamic insulin, although the mechanism is debated 42 . We note that liraglutide does not appear to cross the BBB of cerebral cortex 43 , suggesting direct effects on the barrier, or peripheral effects of receptor stimulation, e.g., anti-inflammatory 44 , potentially interfering with the reported decline of CMR glc in AD.
The average value of the T max of 1.02 umol/g/min in healthy volunteers agreed with values from Choi et al. 45 who fitted the standard Michaelis-Menten equation to the measured brain glucose concentrations. As a function of plasma glucose, the regression yielded values of Michaelis constant K t of 11.8 ± 1.6 mmol/L and a T max /CMR glc ratio of 4.7 ± 0.14. Similarly, De Graaf et al. 46 reported values of the T max /CMR glc ratio that averaged 3.2 ± 0.10 and 3.9 ± 0.15 for gray matter and white matter using the standard transport model, with K t of 6.2 ± 0.85 and 7.3 ± 1.1 mmol/L for gray matter and white matter. For an average whole-brain glucose consumption rate of 0.25 umol/g/min, the corresponding T max estimates average close to 1 umol/g/min with an average Michaelis constant of 8. Measurements of blood-brain unidirectional cleance of glucose (K1) [48][49][50][51][52][53] , averaged 0.08 ml/g/min for K1 and 0.34 umol/g/min for the blood-brain glucose flux (J1 = K1 Ca). The canonical T max estimate for glucose transport across the human blood-brain barrier was listed by Gjedde 54 as 1 umol/g/min.
The prevention of decline of CMR glc is associated with an effect of liraglutide on the T max estimates. The effect may represent a possible reversal of the expected down-regulation of the brain glucose transporters that are required for glucose uptake and metabolism in the brain 19 . We note that the finding of reduced glucose transport capacity in AD presented here may also to some extent reflect lower neuronal activity and/or loss of neuronal cell mass, raising the issue of the unresolved identification of cause and effect. Although the present population of patients only revealed a tendency towards a direct correlation with disease duration, animal studies show that BBB breakdown occurs before the development of functional deficits 25 . The reversal of the duration-related increase of the estimates of maximum clearance (T K / t max ) is a further indication that a potentially important mechanism of disease advance with duration is the increase of affinity of the transporters to glucose that causes the maximum clearance to increase, as the maximum transport capacity declines. The treatment with liraglutide reverses both of these trends, jointly expressed as a normalization of the estimates of maximum clearance.
Aging and neurodegenerative disorders have been shown to be associated with increased lactate concentration 55 , and lactate has been observed to play an important role in the regulation of cerebral blood flow 56,57 , implying that higher lactate production is associated with higher blood flow rates. One prediction from this association is higher blood flow rates relative to glucose consumption in these states, including lowering of the glucose utilization or net extraction fraction (GUF) in aging and neurodegenerative disorders. In agreement with these findings, we report a positive correlation between cognitive scores and GUF estimates in the patients from both groups at baseline. In addition, we report that the GLP-1 analog treatment is associated with insignificantly higher GUF estimates and significantly increased net clearances of [ 18 F]FDG and glucose in the treatment group, compared to the placebo group, with numerical tendencies towards an increase of GUF estimates in the treatment group and a decline of GUF estimates in the placebo group (although neither trend is significant).
As predicted, the estimates of CMR glc declined with disease duration, indicating the natural progression of the disease in this group of patients, signifying the conclusion that GLP-1 receptor stimulation halts the progression of Alzheimer's disease, in association with a very significant increase of the T max estimates. The decline of CMR glc estimates coincided with the decline of cognitive functioning, as previously reported 11 . As the restoration of brain glucose levels and metabolism is likely to positively influence AD pathology 35 , this is a potentially novel approach to prevention or termination of disease progression.

Limitations.
The AD duration measure determined as the time from definite diagnosis naturally has considerable uncertainty. The fact that the inclusion criteria were solely clinical, potentially limits the accuracy of the duration measure 58 . A further caveat is the small subject sample sizes. Regardless of blinding and randomization, minor differences in the baseline characteristics of the treated and placebo groups influence subsequent disease progression and hence affect the therapeutic impact. In larger studies, randomization would be more likely to balance the treated and untreated groups. Despite the extensive evidence of the relation between CMR glc and cognition, also as reported in a study of this size, the interpretation of measures of cognition in the current paper is of course speculative. While the trial was conducted as a randomized, placebo-controlled, double-blinded intervention study, and the PET analysis was carried out by an author blinded to the group and subject definitions (KV), the findings are the results of post-hoc analysis and as such were not obtained in a blinded fashion.

Conclusion
In the present study, we report the evidence that prevention of the decline of CMR glc is associated with improvement of BBB glucose transport capacity, detected as differences of estimates of the maximum capacity of blood-brain transfer of glucose (T max ), interpreted as a crucial element of disease progression and duration, responsible for possible limitations of nutrient delivery. The evidence extends the previous discovery of an effect of GLP-1 on maximum blood-brain glucose transfer capacity in healthy human volunteers, reported by Gejl et al. 26 . The change of T max estimates occurred in relation to changes of the net clearance of glucose, and net glucose consumption (CMR glc ), both as functions of disease duration and cognitive ability. Thus, the restoration of brain glucose availability and neuronal metabolism with GLP-1 or an analog potentially protects against cognitive impairment in Alzheimer's disease.

Methods
Study design and participants. In the present investigation, we completed a 26-week, randomized, placebo-controlled, double-blinded intervention with liraglutide or placebo in patients with AD, recruited from dementia clinics in Central Denmark, with key clinical inclusion and exclusion criteria and administrative details listed by Gejl et al. 27 .
We assigned 38 patients to receive either the GLP-1 analog liraglutide (n = 18) or placebo (n = 20), as described by Gejl et al. 27 . Of these, 14 patients had PET with [ 18 F]FDG before and after treatment, compared to 19 patients who received placebo, all of whom completed the cognitive examination. Tomography sessions for CMR glc were incomplete in two patients, leaving 17 patients from the placebo group and 14 patients from the liraglutide group in the final analysis of [ 18 F]FDG-derived radioactivity. Demographic and clinical characteristics are described in Gejl et al. 27 . The AD duration measure was determined as the time from definite diagnosis.
Patients willing to participate gave written informed consent. Safety data were monitored independently throughout the study period. The study was conducted according to the principles of the Helsinki Declaration. The Central Denmark Regional Committees on Biomedical Research Ethics, the Danish Data Protection Agency, and the Danish Medicines Agency approved the protocol 59 , with trial registration at ClinicalTrials.gov: NCT01469351, November 1, 2011. Participants attended a screening visit to assess eligibility followed by randomization to liraglutide or placebo for 26 weeks. Liraglutide was administered as 0.6 mg subcutaneously daily for one week; hereafter 1.2 mg daily for one week, before final increase to 1.8 mg daily. The placebo group members received saline in similar volumes. Co-registration. We acquired anatomical images for co-registration with the 3 T Magnetom Tim Trio system (Siemens Healthcare, Erlangen, Germany) with 3D T1-weighted high-resolution anatomic scan of magnetization-prepared rapid acquisition gradient echo (MPRAGE) sequence. We co-registered PET images with individual MR images to an MR template, and evaluated the quality of each co-registration by visual inspection in 3 planes. PET and MR-images were co-registered and entered in Talairach space, and anatomical volumes of interest were used to extract time-activity-curves (TACs) from the dynamic PET images for the [ 18 F]FDG analyses of cortex as a whole.

Positron Emission
Cognitive Testing. We evaluated cognition by the "Brief cognitive examination" from the Wechsler Memory Scale (WMS-IV) 61 , the test examines examining orientation, time estimation, mental control, clock drawing, incidental recall, inhibition and verbal reproduction.
Kinetic Analysis. The primary outcome was the maximum glucose transport capacity of the BBB (T max .) assessed with [ 18 F]FDG. We also determined magnitudes of unidirectional glucose extraction fraction (GEF) from measures of [ 18 F]FDG uptake and blood flow, the latter as reported by Gejl et al. 27 where K * and ⁎ K 1 are the net and unidirectional blood-brain clearances of [ 18 F]FDG, and ⁎ k 2 , and ⁎ k 3 are the corresponding rate constants of [ 18 F]FDG-derived radioactivity exchanges between circulation and the [ 18 F]FDG precursor pool mediated by GLUT1 and hexokinase. The unidirectional glucose flux from blood into brain was calculated from the unidirectional [ 18 F]FDG clearance as, where J 1 is the unidirectional flux of glucose (rather than [ 18 F]FDG) from blood to brain, C a is the arterial plasma glucose concentration, and τ is the affinity ratio of [ 18 F]FDG to glucose for blood-brain transfer across the BBB. The net clearance of glucose by definition is where LC is the "lumped constant", defined above as the ratio of the net clearances of [ 18  Here, the unidirectional glucose extraction fraction (GEF) by definition is K F / 1 where F is cerebral blood flow, while the glucose utilization fraction (GUF), also by definition, is FC CMR /( ) a glc . We determined the blood-brain unidirectional clearance, defined as, from the values of flux and concentration, while the glucose clearance from brain tissue to the circulation (K 2 ) likewise was determined from the brain-blood flux and brain tissue concentration of glucose, as, from which we calculated the parameters T max and K t by multilinear regression, yielding a shared estimate of K t for all members of the group and individual estimates of T max for each member.

Control group.
We obtained a control estimate of the maximum glucose transport across the blood-brain barrier from published measurements by Kuwabara et al. 29 that we analyzed in the same manner as the estimates in the four groups of patients with Alzheimer's disease, after correction for the lower resolution and greater partial volume effect of PET before the 21st Century (radioactivities increased by 1.33 for cerebral cortex).

Statistics.
We analyzed treatment group data in two ways: We analyzed changes of kinetic variables by paired t-tests within groups, and changes of T max with 2-way ANOVA of qui-squared with Tukey's correction for multiple comparisons within and between groups. P-values less than 0.05 were considered indicative of significant difference. Spearman and Pearson's r tests were used to evaluate correlations, implemented in GraphPad Prism (GraphPad Software, San Diego, CA) and PMOD (PMOD Technologies Ltd., Zürich, Switzerland).