Neuronal human BACE1 knockin induces systemic diabetes in mice

Aims β-Secretase 1 (BACE1) is a key enzyme in Alzheimer’s disease pathogenesis that catalyses the amyloidogenic cleavage of amyloid precursor protein (APP). Recently, global Bace1 deletion was shown to protect against diet-induced obesity and diabetes, suggesting that BACE1 is a potential regulator of glucose homeostasis. Here, we investigated whether increased neuronal BACE1 is sufficient to alter systemic glucose metabolism, using a neuron-specific human BACE1 knockin mouse model (PLB4). Methods Glucose homeostasis and adiposity were determined by glucose tolerance tests and EchoMRI, lipid species were measured by quantitative lipidomics, and biochemical and molecular alterations were assessed by western blotting, quantitative PCR and ELISAs. Glucose uptake in the brain and upper body was measured via 18FDG-PET imaging. Results Physiological and molecular analyses demonstrated that centrally expressed human BACE1 induced systemic glucose intolerance in mice from 4 months of age onward, alongside a fatty liver phenotype and impaired hepatic glycogen storage. This diabetic phenotype was associated with hypothalamic pathology, i.e. deregulation of the melanocortin system, and advanced endoplasmic reticulum (ER) stress indicated by elevated central C/EBP homologous protein (CHOP) signalling and hyperphosphorylation of its regulator eukaryotic translation initiation factor 2α (eIF2α). In vivo 18FDG-PET imaging further confirmed brain glucose hypometabolism in these mice; this corresponded with altered neuronal insulin-related signalling, enhanced protein tyrosine phosphatase 1B (PTP1B) and retinol-binding protein 4 (RBP4) levels, along with upregulation of the ribosomal protein and lipid translation machinery. Increased forebrain and plasma lipid accumulation (i.e. ceramides, triacylglycerols, phospholipids) was identified via lipidomics analysis. Conclusions/interpretation Our data reveal that neuronal BACE1 is a key regulator of metabolic homeostasis and provide a potential mechanism for the high prevalence of metabolic disturbance in Alzheimer’s disease. Electronic supplementary material The online version of this article (doi:10.1007/s00125-016-3960-1) contains peer-reviewed but unedited supplementary material, which is available to authorised users.

alongside a fatty liver phenotype and impaired hepatic glycogen storage. This diabetic phenotype was associated with hypothalamic pathology, i.e. deregulation of the melanocortin system, and advanced endoplasmic reticulum (ER) stress indicated by elevated central C/EBP homologous protein (CHOP) signalling and hyperphosphorylation of its regulator eukaryotic translation initiation factor 2α (eIF2α). In vivo 18 FDG-PET imaging further confirmed brain glucose hypometabolism in these mice; this corresponded with altered neuronal insulin-related signalling, enhanced protein tyrosine phosphatase 1B (PTP1B) and retinol-binding protein 4 (RBP4) levels, along with upregulation of the ribosomal protein and lipid translation machinery. Increased forebrain and plasma lipid accumulation (i.e. ceramides, triacylglycerols, phospholipids) was identified via lipidomics analysis. Conclusions/interpretation Our data reveal that neuronal BACE1 is a key regulator of metabolic homeostasis and provide a potential mechanism for the high prevalence of metabolic disturbance in Alzheimer's disease.

Introduction
The incidence of type 2 diabetes and Alzheimer's disease is rising at an alarming rate. Type 2 diabetes increases the risk of Alzheimer's disease [1]; however, systemic glucose intolerance and insulin resistance are also reported in dementia patients without a history of diabetes [2]. Alzheimer's disease brains exhibit defective neuronal insulin signalling [3] and glucose hypometabolism in 2-[ 18 F]fluoro-2-deoxy-D-glucose positron emission tomography ( 18 FDG-PET) studies [4]. Further, mouse models of both diseases share similar cognitive phenotypes [5], and increased susceptibility to high-fat diet (HFD)-induced diabetes is observed in Alzheimer mouse models compared with controls [6]. Collectively, the overlapping pathology indicate common pathogenic factors, although their molecular foundations remain unclear. β-Secretase 1 (BACE1) is implicated in Alzheimer's disease as the enzyme responsible for the rate-limiting step in β-amyloid (Aβ) production [7]. Brain BACE1 levels increase with age [8], particularly in Alzheimer's disease [9], and following pathological events [10,11]. Bace1 deletion abolishes Aβ deposition and rescues cognitive deficits in APP mutant mice [12]. We recently confirmed that neuronspecific knockin of human (h)BACE1 induces Aβ accumulation, promotes brain inflammation and recapitulates Alzheimer's disease-like phenotypes in mice in the absence of mutant APP expression [13], suggesting that BACE1 represents a molecular risk factor for sporadic Alzheimer's disease.
Although studied principally for its role in amyloidosis, BACE1 has multiple substrates other than APP [14], comprising transmembrane proteins involved in intercellular signalling. BACE1 expression is predominantly neuronal, although BACE1 mRNA is also found in the liver, skeletal muscle and pancreas [15], with pancreatic mRNA encoding an inactive isoform [7,16]. A potential role for BACE1 in metabolic regulation has only recently emerged, as Bace1 knockout improved glucose metabolism and protected mice from HFD-induced obesity and diabetes [17]. Conversely, the induction of insulin deficiency via systemic streptozotocin (STZ) injection raised central BACE1 levels [18], and this was associated with endoplasmic reticulum (ER) stress [19]. Thus, BACE1 may contribute to metabolic regulation; however, it remains to be established whether BACE1 mediates the association between type 2 diabetes and Alzheimer's disease.
To elucidate the contribution of neuronal BACE1 to systemic glucose regulation and lipid metabolism, we characterised central and peripheral metabolic changes in brain-specific hBACE1 (PLB4) knockin mice [13]. Because global deletion of Bace1 protected mice from diet-induced diabetes [17], we hypothesised that neuronal BACE1 may regulate system metabolism in addition to inducing brain pathologies relevant to Alzheimer's disease [13]. We provide evidence that BACE1-induced hypothalamic dysregulation causes systemic diabetes, which may explain the high comorbidity of diabetes and Alzheimer's disease in ageing populations.

Methods
Animals All animals were housed and tested in accordance with European (Directive on the Protection of Animals used for Scientific Purposes, 2010/63/EU) and UK Home Office regulations, experiments were approved by the University of Aberdeen Ethics Board and performed in accordance with the Animal (Scientific Procedures) Act 1986 and following Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. The transgenic PLB4 and WT lines were generated and bred as previously described [13], experimenters were not blinded to genotype. All mice were housed in single sex groups, unless food and water consumption was being measured. Physiological assessments were performed in male mice at 3, 4, 5 and 8 months of age. All mice were killed by neck dislocation. Five-hour-fasted mice aged 3 and 8 months were used for liver assays, and all postmortem molecular analyses of signalling pathways (western blotting and quantitative real-time PCR [qPCR]) were performed using samples from 8-month-old mice. Glucose-stimulated insulin secretion (GSIS) was determined in 4-month-old mice at 0, 15 and 30 min time points during glucose tolerance tests (GTTs). Blood leptin content was determined from baseline reading at 3, 4 and 8 months in a 5 h fasted state. EchoMRI scans (EchoMRI, Houston, TX, USA) were performed on mice at ages 4, 5 and 8 months. Additional 6-month-old mice were used for lipidomic analysis and 18 FDG-PET imaging. Body weights were recorded and food and water consumption (in g) were measured for 2 weeks in 5-and 8-month-old male mice.
Comparative lipidomics plasma and brain analyses Frozen plasma and forebrain samples from 6-month-old male mice (WT, n = 8; PLB4, n = 8) were used for global lipid analysis using liquid chromatography-mass spectroscopy (LC-MS). Plasma lipids were extracted from samples according to the Folch method. Lipids were solvent extracted in methanol/ chloroform (2:1 vol./vol.). The lipid extracts were subsequently analysed by LC-MS in positive and negative ion modes with a C18 column and a water/acetonitrile/isopropanol gradient using an Exactive Orbitrap system (Thermo Scientific, Hemel Hempstead, UK) . Lipidomic datasets were processed using Progenesis QI software (version 2.0, Non-Linear Dynamics, Newcastle upon Tyne, UK) and searched against LIPID MAPS (www.lipidmaps.org/) and the Human Metabolome Database (www.hmdb.ca/).
Gene expression RNA was extracted from frozen hypothalamic and cortical samples using TRI Reagent, as previously described [23]. Target genes (hBACE1 and mouse Pomc, Mc4R, Npy, LepR, Ptp1b, Chop, Cd11 and Cd68) were amplified by qPCR using GoTaq Master Mix (Promega, Madison, WI, USA). The geometric mean of three commonly used reference mRNAs (Ywhaz, Nono and Actb) was used to normalise data.
Statistical analysis Prism 5 (GraphPad, La Jolla, CA, USA) was used for statistical analyses, and data are expressed as the means ± SEM. All molecular and genetic data were adjusted to loading controls or housekeeping genes and calculated relative to WT. Two-tailed t tests with Welch's correction were used for unbiased comparisons between transgenic mice and controls. Group analyses used two-way ANOVA (with repeated measures where appropriate), followed by Bonferroni post hoc tests.

Systemic glucose homeostasis is impaired in PLB4 mice
We first confirmed that human BACE1 expression was brain specific (Fig. 1a,b) and then established that body weights were normal in PLB4 mice aged up to 4 months ( Fig. 1c) but decreased compared with wild-type (WT) controls at 5 and 8 months. Food and water intake adjusted to body weight were not affected (Fig. 1d,e). Despite the lean phenotype, adipose tissue mass was mildly increased in PLB4 mice vs WT at 4 and 8 months of age (Fig. 1f), while lean mass was unaffected (Fig. 1g), as demonstrated by EchoMRI. Furthermore, 5-h fasted PLB4 mice exhibited elevated blood glucose levels, progressive severe glucose intolerance ( Fig. 1h-k) and impaired GSIS from 4 months of age (Fig. 1m,n). Serum insulin levels were also raised (hyperinsulinaemia) at 5 months but were similar to WT levels at 8 months (Fig. 1l), despite continued hyperglycaemia. Serum NEFA were increased in 3-month-old PLB4 mice, although this was not statistically significant (p = 0.06), and were comparable between groups at 8 months of age (Fig. 1o).
Altered plasma metabolic homeostasis and hyperleptinaemia Levels of the adipocyte-derived hormone, leptin, which strongly correlate with obesity and diabetes, were initially (q) Serum markers detected in 8-month-old PLB4 mice using an enzymatic multiplex assay. White bars, WT mice; black bars, PLB4 mice. Data represent means + SEM or (p, q) means + SEM normalised to WT values. *p < 0.05, **p < 0.01, ***p < 0.001 normal in PLB4 mice, but were drastically elevated at 4 and 8 months (Fig. 1p) when adiposity was increased.
Since body mass and long-term energy expenditure are tightly regulated by complex signalling networks between the CNS and peripheral systems, we screened serum samples from 8-month-old PLB4 mice for markers of homeostatic control using a multiplex assay (Fig. 1q). Concentrations of fasting C-peptide were low, indicative of defective proinsulin synthesis, while amylin levels (co-secreted with insulin) were enhanced. GIP was also upregulated in PLB4 mice compared with controls.
Impaired hepatic glycogen synthesis, insulin resistance and fatty liver phenotype Evidence of failing glucose storage in mice expressing hBACE1 was indicated by their reduced hepatic glycogen levels (Fig. 2a), increased hepatic triacylglycerol content (Fig. 2b) and inability to activate (i.e. dephosphorylate) GS in response to hyperglycaemic conditions (Fig. 2c,d). Heightened translational demand was also identified in PLB4 hepatocytes, suggested by increased phosphorylation of the rpS6 component of 40S ribosomes (Fig. 2c,d). Additionally, the ApoE, PTP1B and RBP4 lipometabolic regulators linked with obesity and insulin resistance [6,24,25] were upregulated in PLB4 mouse liver (Fig. 2c,d), suggesting decreased hepatic insulin sensitivity. Levels of these regulators were unchanged in white adipose tissue (WAT) and muscle (Fig. 2e-h).

Dysregulated central and plasma lipid composition
The non-alcoholic fatty liver phenotype of PLB4 mice corresponded to that observed in human type 2 diabetes [26], and called for further investigation of putative changes in lipid metabolism. Altered lipid composition was recently proposed to predict 'phenoconversion' from mild cognitive impairment to Alzheimer's disease, potentially offering a novel biomarker [27]. Hence, we applied a comparative global lipidomics approach and identified 321 altered species in PLB4 mice vs controls (see Table 1 for a summary and electronic supplementary material Tables 1-4 for details) comprising phospholipids, sphingomyelins and sphingolipids such as ceramides. Phospholipids such as phosphatidylethanolamine, lysophosphatidylethanolamine and phosphatidylserine were predominantly increased in both the brain and plasma; however only 11% of the altered lipid species were identical in both sample types. Fewer changes in lipid composition were detected in plasma samples than in brain samples.
Reduced brain glucose metabolism To further investigate whether the diabetic and metabolic phenotype of PLB4 mice was similar to that of human type 2 diabetes and dementia, we conducted an in vivo 18 FDG-PET study (Fig. 3). and confirmed global cerebral hypometabolism in PLB4 mice vs WT controls at the symptomatic age of 6 months. In contrast to that of the brain, the metabolic activity of brown adipose tissue (BAT) was drastically enhanced, while cardiac metabolism did not differ between groups. As BAT activation is inversely related to body weight and obesity, the hypermetabolic readouts in PLB4 mice are consistent with their leaner Hepatic triacylglycerol (μgl/mg) 3 8 Age (months) 3 8 Age ( phenotype [28] and indicative of a compensatory metabolic adjustment. Collectively, these data confirm that PLB4 mice have a diabetes-like phenotype. Abnormal neuronal insulin signalling Since hBACE1 was expression was forebrain specific, we next investigated whether neuronal insulin signalling was affected in   18 FDG-PET imaging in brain and BAT. (b) Quantification of upper body glucose uptake for the brain, heart and BAT. (c) Statistical parametric mapping shows reduced regional T scores (0-6) for brain metabolism in PLB4 vs WT animals. (d) Quantification of regional brain glucose uptake in WT and PLB4 mice. cBS, caudal brainstem; Cort, cortex; Cere, cerebellum; HC, hippocampus; Hypo, hypothalamus; STR, striatum; SUV, standardised uptake value. White bars, WT mice; black bars, PLB4 mice. Data show means + SEM. *p < 0.5, **p < 0.01, ***p < 0.001 symptomatic PLB4 mice by probing for markers of the insulin IR-Akt-GSK-3β transduction pathway (Fig. 4) in forebrain lysates of 8-month-old fasted mice. Total IR was upregulated in PLB4 vs WT mice, together with RBP4, thus corroborating links with insulin resistance [29]. PTP1B levels were also increased, indicating downregulation of central insulin and leptin signalling [6,24]. There were no differences in brain levels of Akt between PLB4 and WT mice, but trends for decreased baseline GSK-3β phosphorylation in PLB4 mice indicated early neuronal disinhibition of this pro-apoptotic kinase. Targets downstream of mTOR were also altered: although levels of mTOR did not differ in PLB4 forebrain vs controls, we detected increased levels of rpS6 and its kinase S6K, implying an overall increase in translational demand consistent with elevated brain lipid synthesis in PLB4 mice.
Altered hypothalamic melanocortin transcription and ER stress As the hypothalamus is the regulatory brain centre for metabolic control, we reasoned that its malfunction was a possible cause of the failing insulin signalling and diabetic phenotype of PLB4 mice. Hypothalamic function was assessed via gene expression analysis of neuro-and polypeptides that regulate the hypothalamic-pituitary-adrenal axis (Fig. 5a-c). Transcription of the appetite-suppressing pro-opiomelanocortin (Pomc) gene was drastically increased in PLB4 mice compared with controls. POMC-positive neurons respond to leptin or insulin to reduce food intake and maintain energy homeostasis via secretion of alpha melanocyte-stimulating hormone (αMSH) and melanocortin receptor 4 (MC4R). A consistent increase in Mc4r gene expression was detected in the PLB4 vs WT hypothalamus, suggesting a shift toward anorectic signalling in mouse brains expressing hBACE1. Importantly, hypothalamic ER stress (see Fig. 5d) was confirmed by a drastic rise in Chop transcription in PLB4 tissue compared with controls. Corresponding protein levels were also increased (Fig. 5e,f), as was phosphorylated (i.e. activated) eIF2α, a major CHOP regulator. These changes were not associated with transcriptional alterations in inflammatory markers (microglial Cd11 and Cd68).

Discussion
Diabetic complications can lead to cognitive dysfunction and are acknowledged risk factors of Alzheimer's disease, but little is known about the reverse scenario. Here, we demonstrate that the major amyloidogenic enzyme, BACE1, is sufficient to increase the risk of diabetes development when expressed in neurons only. Severe impairment in systemic glucose homeostasis and insulin sensitivity was evident in PLB4 mice from the age of 4 months onwards and progressively deteriorated with age. We previously reported that PLB4 mice develop mild cognitive deficits between 4 and 6 months. Our current findings therefore indicate that neuronal BACE1 induces global metabolic dysregulation along with brain inflammation and amyloidosis-related cognitive decline. Overall, the diabetic profile of PLB4 mice agrees with the improved glucose clearance and insulin sensitivity in mice lacking murine Bace1 globally [17,30], but pinpoints neuronal BACE1 as the major driver of this alteration.
Although both Bace1 knockout and our hBACE1 knockin mice were leaner compared with WT controls, there were differential effects on adiposity and leptin signalling. Deletion of Bace1 decreased adiposity and improved sensitivity to leptin, while neuronal hBACE1 expression promoted adipogenesis and induced hyperleptinaemia. Hence, it seems plausible that manipulation of (neuronal) BACE1 affects body weight via changing metabolic efficiency, adipose composition and signalling. Importantly, leptin production and signalling are regulated by neuronal PTP1B [24]. We propose that the elevated central PTP1B expression in PLB4 forebrains may explain the hyperleptinaemic profile and compensatory BAT hyperactivity in these mice, which confirms that leptin has opposing effects on WAT and BAT [31].
Further, while proinsulin synthesis appeared unchanged in PLB4 mice compared with controls, heightened levels of amylin indicate an attempt to downregulate hyperglycaemia in the insulin-deficient state, and suggests that pancreatic function was at least partially preserved in 8-month-old PLB4 mice. Glucose intolerance and the fatty liver phenotype in PLB4 mice were further associated with drastically elevated levels of serum GIP. Although GIP exhibits insulinotropic properties under physiological conditions, its incretin effects are thought to be blunted in the diabetic state, and elevated levels may promote fatty acid accumulation and induction of proinflammatory cytokines [32].
The fatty liver phenotype of PLB4 mice corresponded with high plasma triacylglycerol and elevated levels of phospholipids such as phosphatidylcholine and lysophosphatidylcholine, which are typically observed in type 2 diabetic patients [26]. Altered plasma lipid composition was also recently proposed to predict early memory impairments, thus potentially offering a novel approach to identify Alzheimer's disease [27,33,34].
Although lipid accumulation was also affected in neuronal tissue from symptomatic PLB4 mice, there was a poor species match for plasma, suggesting that (1) plasma markers may not be indicative of changes in brain lipid composition and (2) that the increase in neuronal lipids probably originated from the CNS. PLB4 mice had pronounced elevation of several classes of phospholipids such as phosphatidylethanolamine, lysophosphatidylethanolamine and phosphatidylserine, which are major components of neuronal membrane bilayers. Such alterations are reported in human Alzheimer's disease brains [35] and are proposed to affect mitochondrial function, signal transduction and receptor activation, hence interfering with neurotransmission and neuronal integrity [36,37]. Importantly, brain ceramide levels were substantially increased in PLB4 mice compared with controls. This sphingomyelin precursor is of particular interest because it increases naturally with age [38] and at an early stage of human Alzheimer's disease [39]. Furthermore, ceramides were previously shown to regulate BACE1 protein expression and promote APP β-site cleavage [40]. Although the mechanisms through which BACE1 may upregulate ceramide accumulation are largely unknown, initial evidence suggests that Aβ peptides may activate sphingomyelinase [41]. It therefore seems plausible that the introduction of BACE1 promotes ceramide biogenesis, and vice versa. In support of this, a recent study demonstrated that intracerebroventricular ceramide infusions induce lipotoxicity and hypothalamic ER stress associated with increased eIF2α and PERK phosphorylation, sympathetic inhibition, reduced weight gain and altered energy balance in rats [42].
Cerebral hypometabolism revealed via 18 FDG-PET imaging is common in early dementia patients as well as diabetic patients with or without mild cognitive impairment [43], and may ultimately contribute to cognitive pathology. We found that reduced glucose utilisation in PLB4 mice was associated with poor neuronal insulin sensitivity. Altered IR and PTP1B expression occurred in PLB4 mice at an advanced stage of systemic insulin resistance, hepatic dysfunction and Aβ-associated cognitive impairment [13]. Similar elevations in neuronal PTP1B were observed in response to HFD feeding in other Alzheimer's disease models [6], while deletion of neuronal PTP1B improved IR signalling and protected against HFD-induced obesity and insulin resistance [24].
Increased S6K phosphorylation in PLB4 vs WT brains and elevated expression of its substrate, rpS6, suggest an increased demand for protein and lipid synthesis. Additionally, the ribosomal element is regulated by eIF2, offering an alternative route for its modification [44]. Elevated rpS6 phosphorylation along with increased brain RBP4 levels were previously found in Alzheimer's disease mice on a HFD [6]. Neuronal pathways that mediate RBP4's action and toxicity are yet to be investigated, but have been linked to proinflammatory cytokines in macrophages and to activation of JNK, a major ER stress kinase [29].
Recent studies revealed reduced hypothalamic volume and accelerated atrophy of orexin neurons in early Alzheimer's disease [45,46]. In contrast to diet-induced obese and diabetic models, PLB4 mice displayed increased Pomc and Mc4r mRNA levels, suggesting a hypothalamic shift toward appetite suppression and increased energy expenditure. The increase in hypothalamic Mc4r and Pomc transcription in PLB4 mice further contrasts with the recently demonstrated Aβ-oligomer-induced elevation in Npy mRNA (but not in Pomc mRNA), which was not associated with changes in circulating leptin levels [47]. Here, elevated melanocortin transcription in the hypothalamus of PLB4 mice may be a downstream effect of persistently increased circulating levels of leptin and systemic hyperglycaemia.
An advanced state of ER stress was confirmed in the hypothalamus of PLB4 mice, resembling that induced by HFD feeding [48]. Pharmacologically induced hypothalamic ER stress resulted in systemic diabetes in mice [49]; this was also recently illustrated for Aβ oligomer infusions [47]. The pathology of PLB4 mice therefore agrees with a scenario of BACE1-driven elevations of the ER stress marker CHOP and the protein translation regulator eIF2α, which suggest induction of an integrated stress response. Mechanistically, neuronal expression of human BACE1 may promote hypothalamic ER stress via ceramide lipotoxicity, activation of eIF2α and an integrated stress response [42], in addition to driving Aβ production [13].
In conclusion, we demonstrate that neuronal expression of human BACE1 causes systemic diabetic complications. We propose that increased levels of central BACE1 promotes metabolic disturbance via inducing hypothalamic impairment, ER stress, and Aβ and lipid accumulation, leading to neuronal damage, insulin resistance, hepatic deficits and global glucose dyshomeostasis. The comorbid phenotype of the PLB4 mouse provides insight into the complex mechanistic interactions between diabetes and Alzheimer's disease. As an extension to the hypothesis that diabetic complications promote the onset and progression of Alzheimer's disease, we suggest that the reverse scenario may also apply.
Duality of interest statement The authors declare that there is no duality of interest associated with this manuscript.
Contribution statement BP and MD conceived the study; NM, DK and GR participated in the study design; KP and RD carried out the main experimental work, including western blotting brain and liver samples, physiological glucose experiments and plasma measurements; RD acquired and analysed EchoMRI and liver assay data; KS carried out and analysed the qPCR experiments; PW and MKD conducted the lipidomics study, and analysed brain and plasma lipid composition; MM and AW performed 18 FDG-PET imaging and analysed the glucose uptake data; KP, BP and MD performed the final data analysis and interpretation, with significant input from NM, DK and GR; KP, BP and MD co-wrote the paper; and all authors critically revised and approved the final version of the manuscript. BP and MD are responsible for the integrity of this work as a whole.
Open Access This article is distributed under the terms of the Creative Comm ons Attribution 4.0 International License (http:// creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.