Cerebrospinal Fluid Orexin in Alzheimer’s Disease: A Systematic Review and Meta-Analysis

Abstract

Objective/Background:

A growing body of evidence suggests that sleep and Alzheimer’s disease (AD) have a bi-directional relationship. Emerging research also suggests that orexin, a key neurotransmitter involved in sleep-wake regulation, may be altered in persons with AD, however results have not been consistent across prior studies. This investigation was conducted to both evaluate the aggregate literature to minimize the risk of bias and identify potential factors associated with heterogeneity across studies.

Methods:

Systematic review identified relevant investigations that compared cerebrospinal fluid orexin in persons with AD and controls. Meta-analysis (random effects model) compared effect size (Hedge’s g) for orexin between AD and controls. Meta-regression was additionally performed for key variables of interest to evaluate potential causes of heterogeneity among studies.

Results:

17 studies were identified that met inclusion/exclusion criteria. Evidence of publication bias was not identified. Non-significant increases in orexin were observed in AD relative to controls, with moderate to large heterogeneity among studies (Hedge’s g=0.20, p=0.136, I²=72.6%). Meta-regression demonstrated both year of publication (β=0.055, p=0.020) and effect size for phosphorylated tau in AD versus controls (β=0.417, p=0.031) were associated with differences in orexin.

Conclusions:

Results do not support broad differences in orexin in AD compared to controls, however, evolving diagnostic criteria may have affected findings across studies. Future research that examines orexin in AD over the longitudinal course of the disorder and explores potential links between phosphorylated tau and orexin are indicated.

1. Introduction

Alzheimer’s disease (AD) is the most common cause of dementia, characterized pathologically by extracellular amyloid-beta plaques and intracellular tau neurofibrillary tangles that synergistically interact and progressively affect broader and key brain areas, ultimately leading to cognitive and functional decline¹. Multiple emerging lines of research indicate that abnormalities in sleep are associated with the development of AD. Although disturbed sleep was previously thought to merely be an epiphenomenon of AD progression, there is mounting evidence that sleep and AD have a bidirectional relationship^2-4. Under this paradigm, disordered sleep increases the longitudinal risk of developing AD, and once AD has developed, AD increases the severity and impact of the sleep disturbance. Thus, understanding the causes of sleep disturbance in AD is an important area of investigation, as it may allow for the development of novel and preventative sleep-related treatment strategies for this devastating illness.

A growing body of research has focused on the role of orexin (hypocretin) in AD. Orexin is a neuropeptide produced by neurons in the lateral hypothalamus, with wide-spread projections throughout the brain^5-7. Although orexin is involved in several regulatory functions including feeding behavior, energy homeostasis, stress response, and the reward system, its role in sleep-wake regulation is central to normal and pathological processes^5-7. Broadly, orexin promotes wakefulness and stabilizes transitions between sleep and wake in large part through innervation of wake promoting neurons in the central nervous system (CNS)^{5, 6, 8}. Orexin has two active isoforms, orexin-A and orexin-B, with the former binding to both orexin 1 and 2 receptors with relatively equal affinity, and the latter more selectively binding to orexin 2 receptor⁹. Loss of orexin-containing neurons through an autoimmune process is central to type 1 narcolepsy, a neurological disorder characterized by excessive daytime sleepiness, cataplexy, and other symptoms that reflect inappropriate admixtures of rapid eye movement (REM) sleep and waking physiology, such as sleep paralysis and hypnagogic/hypnopompic hallucinations^{10, 11}. Interestingly, persons with type 1 narcolepsy, who have low levels of cerebrospinal fluid (CSF) orexin, demonstrate reduced cortical amyloid burden on positron emission tomography (PET) imaging compared to healthy controls¹². Additionally, persons with type 1 narcolepsy have lower CSF amyloid-β (Aβ), total tau (t-tau), and phosphorylated (p-tau) relative to unaffected controls¹³. Prior research has also demonstrated that CSF orexin-A levels are significantly and positively associated with these three key biomarkers of AD pathology in cognitively normal subjects¹⁴.

The observation that CSF orexin, either occurring within the normal physiological range or as part of more profound pathological processes such as type 1 narcolepsy, are associated with reduced AD pathology, has also led to several levels of research into the effect of orexin and its antagonists in AD. From a clinical perspective, a recent randomized clinical trial demonstrated benefit of suvorexant, a dual orexin-receptor antagonist (DORA), in treating insomnia associated with mild-to-moderate probably AD dementia¹⁵. Murine models of AD have demonstrated orexin infusion increases wake and prevents normal decreases in Aβ during sleep relative to waking phases¹⁶. Orexin gene knockout in AD mouse models also results in reduction in Aβ, and subsequent local overexpression of orexin in the bilateral hypothalamus leads to increased wake and Aβ, without parallel increases in either Aβ or wake from orexin overexpression in the hippocampus¹⁷. The concept that increases in amyloid burden are mediated by increased wake and decreased sleep rather than orexin itself, is further supported by AD murine models that demonstrate chronic sleep restriction increases Aβ, while DORAs increase sleep and mitigate increases in Aβ over time^{16, 18}. These findings parallel human studies that have demonstrated increases in Aβ measured via CSF and PET from acute sleep loss^{19, 20}.

In summary there are multiple converging lines of research that suggest CSF orexin may be related to AD pathology, though its role in human AD has not been fully elucidated. Several investigations have demonstrated differences in CSF orexin in patients with AD compared to controls, however, results have not been consistent across studies²¹. Thus, the aims of this study were to conduct a systematic review and meta-analysis of the aggregate literature, to more critically evaluate CSF orexin in human AD.

2. Methods

2.1. Types of participants

Studies that evaluated persons on the Alzheimer’s continuum were considered for inclusion. This included those with Alzheimer’s dementia, mild cognitive impairment (MCI), and/or persons without cognitive impairment who had positive AD biomarkers, consistent with the National Institute on Aging and Alzheimer’s Association (NIA-AA) research framework²².

2.2. Types of studies

All studies that measured and reported CSF orexin (or its precursor protein preproorexin) in patients on the Alzheimer’s continuum and in healthy controls were considered for inclusion. Investigations that utilized plasma measures of orexin were not considered because orexin is expressed in multiple source tissues outside of the CNS and measures are highly variable across studies^{23, 24}. Studies were included that did not consider orexin comparisons between groups as primary study endpoints, so long as relevant data were reported.

2.3. Search strategy

Searches were conducted using PubMed, PsycINFO, and CINAHL. There were no limitations on year of publication or language of the articles. The following search terms were utilized: (orexin OR hypocretin) AND (AD OR Alzheimer* OR Dementia OR MCI OR Mild Cognitive Impairment). Peer reviewed publications and unpublished literature were included. The authors conducted all searches. Searches were conducted on May 8, 2021.

2.4. Eligibility

The following criteria were required for inclusion in the study: 1) orexin must have been quantified from cerebrospinal fluid; 2) orexin levels from a healthy comparison group were reported; 3) studies were conducted in human subjects; and 4) orexin levels from person on the AD continuum were reported. Studies were excluded if they 1) did not measure orexin or used a sample other than CSF (e.g., blood); 2) no healthy controls were utilized; 3) human subjects were not used; or 4) the experimental group was not on the AD continuum.

2.5. Data extraction

The authors extracted all data. Extracted data included: author/journal, year of publication, patient demographics (age and sex), type of assay used to measure orexin, definition AD and controls, and mean and standard deviation of orexin levels for AD patients and controls. Authors of articles that met inclusion/exclusion criteria but did not report data sufficient for meta-analysis were contacted for clarification of orexin data (e.g. data reported in graph format, median/interquartile range of orexin reported, etc.). If requests for primary orexin data were not successful, additional methods were employed to minimize the risk of bias through exclusion of a given study. This included extrapolation of graphically presented data using Digitizeit (Braunschweig, Germany) if needed to extract data from an article that otherwise met inclusion/exclusion criteria. Additionally, if mean and standard deviation could not be obtained, methods to estimate these values from other descriptive statistics were employed²⁵. Other data that were extracted included amyloid-beta 1-42 (Aβ₄₂), total tau, and phosphorylated tau (p-tau) levels for AD patients and controls, when available.

2.6. Analysis

Potential articles were screened by title, abstract, and full text for inclusion/exclusion criteria. Included articles were analyzed both qualitatively and quantitatively. Quality of evidence was evaluated using the methodological index for non-randomized studies (MINORS) rating scale method, which rates the quality of evidence of case-control and other non-randomized study designs²⁶. Meta-analysis was performed using a random-effects model. The primary variable of interest was CSF orexin (AD relative to controls). Hedge’s g was utilized as the effect size for the meta-analysis to minimize the potential impact of interassay variability in the measure between studies. I² was used to assess heterogeneity among studies, with cutoffs 0%, 25%, 50%, and 75% used to define no, small, medium, and large heterogeneity. Sensitivity analyses considered on the AD continuum, only those with AD dementia, and only those with mild cognitive impairment. One study utilized a primary sample and a replication sample in their methods²⁷. Due to high interassay variability between primary and replication samples in this study, quantitative methods were applied using the primary sample, with additional sensitivity analysis performed using the secondary sample. Publication bias was evaluated using the nonparametric “trim and fill” method²⁸. Effects of sex, age, year of publication, orexin assay type (radioimmunoassay vs. not), Aβ₄₂, total tau, and p-tau were considered as relevant variables for meta-regression in the context of elevated heterogeneity across studies. The ratio of females to males between AD and control groups was calculated to examine potential effects of sex. Similarly, the ratio of age between AD and control groups was calculated to examine effects of age. Hedge’s g values between AD and control groups for Aβ₄₂ tau, and p-tau were calculated for each study to control for inter-study variability in assay methods for these biomarkers.

3. Results

3.1. Study Inclusion and assessment

The Preferred Reporting Items for Systematic Reviews (PRISMA) flow diagram is depicted in Figure 1, which details inclusion/exclusion of studies²⁹. There were 43 full text articles assessed for eligibility, of which 26 were excluded. The remaining 17 that met all inclusion/exclusion criteria were analyzed in the qualitative and quantitative assessment of the literature. One study utilized patients with remitted depression as a control group²⁷. Since CSF orexin levels in depressed patients have not been demonstrated to be significantly different from controls^30-32, this study was included in the systematic review and meta-analysis, with sensitivity testing to determine if its inclusion substantially altered results. Of the four studies that were not included due to absence of healthy control data, one was a correlational analysis of orexin and daytime napping in AD³³, one compared AD and other dementias to patients with narcolepsy³⁴, one compared CSF orexin between AD with and without neuropsychiatric symptoms³⁵, and one compared actigraphy data between AD and controls, however the control group did not have CSF orexin measured as part of the protocol³⁶.

Figure 1.

Preferred Reporting Items for Systematic Reviews (PRISMA) Flow Diagram.

Primary study details meeting inclusion criteria are detailed in the table of evidence (Table 1). Sample size was generally small to moderate across investigations. Studies utilized a wide range of criteria to define AD, including clinical diagnoses (with and without AD biomarker data), as well as other systems such as the National Institute on Aging-Alzheimer’s Association (NIA-AA)²², the National institute of Neurological and Communicative Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA)³⁷; and the National Institute of Neurological Disorders and Stroke-Association Internationale pour la Recherche et l'Enseignement en Neurosciences (NINDS-AIREN) system³⁸. Eleven studies utilized radioimmunoassay to quantify CSF orexin^{27, 39-48}, with other studies utilizing multiplex panels^{49, 50}, enzyme-linked immunosorbent assay (ELISA)^{43, 51, 52}, or fluorescence immunoassay (FIA)⁵³.

Table 1.

Table of Evidence. AD (Alzheimer’s Disease); C (Control); DSM-IV (Diagnostic and Statistical Manual, 4^th edition); ELISA (enzyme-linked immunosorbent assay); F (female); FIA (fluorescence immunoassay); M (male); MCI (mild cognitive impairment); n/a (not available); NIA-AA (National Institute on Aging-Alzheimer’s Association); NINCDS-ADRDA (National institute of Neurological and Communicative Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association); NINDS-AIREN (National Institute of Neurological Disorders and Stroke-Association Internationale pour la Recherche et l'Enseignement en Neurosciences); RIA (radioimmunoassay); SMCI (suspected mild cognitive impairment). Orexin levels reported as pg/mL.

Study [Minors Rating Scale Score]	Patient Age mean (SD) & M:F	Orexin Assay	AD Definition	AD Orexin level mean (SD)	Control Definition	Control Orexin Level mean (SD)	Aβ, tau, p-tau mean (SD)
Baumann et al.,2004 [14]	AD: 72 (12.9) M:F n/a Control: 44 (16.6) M:F n/a	RIA	clinical diagnosis	n=7 472.5 (87.7)	Patients w/o signs of a disorder of the nervous system	n=20 496.3 (62.9)	n/a
Deuschle et al., 2014 [14]	AD: 66.4 (7.2) M:F 3:7 Control: 66.3 (9.3) M:F 4:6	RIA	NINCDS/A DRDA criteria	n=10 417 (49)	Depressed patients in full remission without cognitive impairment	n=10 408 (33)	AD-Aβ42: 459 (170) AD-tau: 369 (111) AD-p-tau: 76.0 (32.0) C-Aβ42: 687 (231) C-tau: 237 (90) C-p-tau: 53.9 (22.9)
Fronczek et al., 2012 [14]	AD: 76.3 (12.4) M:F 11:13 Control: 76.4 (12) M:F 12:13	RIA	clinical diagnosis	n = 24 256.6 (93.7)	matched for age, sex, and post mortem delay	n = 25 322.2 (79.3)	n/a
Gabelle et al., 2017 [14]	AD: 69.5 (6.45) M:F 18:23 MCI: 69.7 (6.2) M:F 20:21 AD/MCI: 69.6 (6.3) M:F 38:44 Control: 65.3 (10) M:F 12:12	RIA	used NIA diagnosis guidelines	AD: n = 41 288.76 (45.99) MCI: n = 41 271.78 (49.55) AD/MCI: n = 82 280.27 (48.27)	no cognitive complaints and normal neuropsych measurements	n = 24 243.2 (37.78)	AD-Aβ42: 614.2 (166.8) AD-tau: 684.3 (219.1) AD-p-tau: 104 (31.3) MCI-Aβ42: 855.4 (276) MCI-tau: 587.1 (230.1) MCI-p-tau: 83 (30.2) AD/MCI-Aβ42: 734.8 (257.1) AD/MCI-tau: 635.7 (228.6) AD/MCI-p-tau: 93.5 (32.3) C-Aβ42: 1150.7 (196.7) C-tau: 252.6 (91.9) C-p-tau: 37 (8.2)
Heywood et al., 2018 [15]	AD: 72.1 (6.8) M:F 5:8 Control: 61 (9.9) M:F 5:10	multiplex panel	abnormal Aβ, t-tau, and p-tau levels with a clinical diagnosis	n = 13 0.00793 (0.0037)	patients who underwent lumbar puncture on suspicion of neurological diseases that were ruled out	n = 15 0.00234 (0.0015)	AD-Aβ42: 425.2 (138.7) AD-tau: 470.3 (333.3) AD-p-tau: 81.7 (40.63) C-Aβ42: 940.3 (272.8) C-tau: 101.7 (77.5) C-p-tau: 21.8 (8.9)
Hoglund et al., 2017 [12]	AD: (Aβ42≤530 pg/mL) 82.5 (3.6) M:F 20:23 Control (Aβ42>530 pg/mL): 81.6 (3.3) M:F 36:53	RIA	All participants were cognitively normal with AD defined by Aβ42 threshold	n = 43 691.1 (159.4)	Aβ42 >530 pg/rnl	n = 86 724.1 (189.4)	AD-Aβ42: <530 AD-tau: 609.1 (230.4) AD-p-tau: 83.6 (25.5) C-Aβ42: >530 C-tau: 428.2 (163.6) C-p-tau: 65.2 (19.7)
Johannson et al., 2015 [17]	AD: 74.3 (4.6) M:F 15:17 SMCI: 71.7 (4.1) M:F 5:8 AD/SMCI: 73.5 (4.6) M:F 20:25 Control: 74.3 (6.3) M:F 10:10	multiplex panel	DSM-IV and NINDS-AIREN criteria	AD: n = 32 1599 (411) SMCI: n = 13 1278.3 (247.1) AD/SMCI: 1506.4 (396.4)	patients with no subjective symptoms of cognitive impairment	n = 20 1588 (378.6)	AD-Aβ42: 416.3 (121.5) AD-tau: 588.3 (242.3) AD-p-tau: 96.3 (27.1) SMCI-Aβ42: 683.3 (267.8) SMCI-tau: 295.7 (140.1) SMCI-p-tau: 58 (31.5) AD/SMCI-Aβ42: 493.4 (211.8) AD/SMCI-tau: 503.8 (254.2) AD/SMCI-p-tau: 85.2 (33.1) C-Aβ42: 938 (200) C-tau: 316 (138.9) C-p-tau: 64.7 (23)
Liguori et al., 2017 [19]	AD: 71.56 (3.92) M:F 8:10 Control: 74.1 (2.8) M:F 7:11	ELISA	NIA-AA criteria	n = 18 165.33 (58.07)	nondemented controls matched for age and sex. These patients were admitted for suspected cognitive impairment but this was ruled out.	n = 18 145.26 (36.91)	AD-Aβ42: 246.89 (106.61) AD-tau: 693.28 (245.67) AD-p-tau: 101.44 (51.58) C-Aβ42: 870.45 (267.66) C-tau: 217.09 (66.05) C-p-tau: 36.64 (12.22)
Liguori et al., 2019 [19]	AD: 66.3 (4.18) M:F 7:13 Control: 63.8 (8.46) M:F 8:7	RIA	Dementia with biomarkers consistent with AD pathology	n = 20 537.5 (59.25)	patients admitted for suspected polyneuropat hy which was ruled out	n = 15 486.25 (88.54)	AD-Aβ42: 327.4 (132.6) AD-tau: 1059.7 (598.95) AD-p-tau: 136.2 (57.67) C-Aβ42: 889.4 (423.16) C-tau: 176.2 (89.27) C-p-tau: 38.3 (6.84)
Liguori et al., 2016 [17]	MCI: 72.7 (4.81) M:F 9:11 Control: 68.8 (2.97)	ELISA	MCI was defined by cognitive concerns; objective evidence of cognitive impairment, normal functional activities, and absence of dementia	MCI: n = 20 169.52 (63.9)	controls were age and sex matched and were originally admitted for suspected polyneuropat hies, but this was ruled out.	n = 26 143.11 (35.38)	MCI-Aβ42: 310.4 (116.39) MCI-tau: 592.25 (405.12) MCI-p-tau: 85 (49.68) C-Aβ42: 802.5 (230.5) C-tau: 188.69 (71.15) C-p-tau: 32.6 (9.9)
Liguori et al., 2014 [15]	AD 70.5 (7.6) M:F 25:23 Control: 70.4 (9.9) M:F 15:14	ELISA	diagnosed using NIA-AA criteria then categorized using MMSE	n = 48 137.69 (31.56)	nondemented patients of similar age and sex	n = 29 131.03 (26.55)	AD-Aβ42: 323.46 (215.11) AD-tau: 670.33 (273.7) AD-p-tau: 96.6 (41.32) C-Aβ42: 953.97 (190.2) C-tau: 224.14 (69.98) C-p-tau: 44.86 (11.72)
Ripley et al., 2001 [12]	AD: 69 (7.7) M:F n/a Control: 41 (16) M:F n/a	RIA	the AD patient data for this study was provided by multiple institutions	n = 24 300 (54.1)	all controls were healthy and had no clinical sleep abnormalities	n = 48 345 (107)	n/a
Schmidt et al., 2013 [15]	AD: 73.76 (8.07) M:F 11:22 Control: 52.03 (17.24) M:F 19:14	FIA	clinical diagnosis using Dubois criteria	n = 33 86.5 (18.82)	subjects without any psychiatric or neurological disorders	n = 33 87.91 (14.95)	AD-Aβ42: 633.8 (290.2) AD-tau: 449.5 (245) AD-p-tau: 79.2 (38.8) C-Aβ42: 1157.3 (260.6) C-tau: 133.8 (48.5) C-p-tau: 32.6 (12)
Shimizu et al., 2020 [13]	AD: 73.9 (8) M:F 11:11 Control: 71.4 (11.2) M:F 15:10	RIA	AD patients met the NIA-AA criteria	n = 22 322.2 (84.4)	normal neuropsychol ogical measuremen ts who underwent clinical neurological investigation, neuroimagin g, and lumbar puncture for diagnostic purposes	n = 25 301 (64)	AD-Aβ42: 162.4 (97.4) AD-p-tau: 89.9 (28.5) C-Aβ42: 185.6 (83.4) C-p-tau: 42.6 (10.8)
Slats et al., 2012 [14]	AD: 71 (5.13) M:F 3:3 Control: 71 (10.26) M:F 3:3	RIA	diagnosed using NINCDS-ADRDA criteria	n = 6 407 (19.8)	healthy elderly controls	n = 6 401 (35.2)	AD-Aβ42: 118 (29.6) C-Aβ42: 216 (114)
Trotti et al., 2021 [13]	AD: 65.5 (7.7) M:F 31:29 Control 69.6 (9.2) M:F 8:17	RIA	Diagnosed by consensus, with participants meeting consensus criteria for MCI whose CSF were consistent with pathologic AD included	n = 60 256.8 (59.0)	Age-matched normal cognition participants	n = 25 248.1 (53.5)	AD-Aβ42: 118,8 (63.1) AD-tau: 93.8 (71.8) AD-p-tau: 50.2 (26.0) C-Aβ42: 301.0 (137.5) C-tau: 37.9 (20.7) C-p-tau: 24.8 (19.6)
Wennstrom et al., 2012 [13]	AD: 73 (6) M:F 15:11 Control: 72 (8) M:F 16:8	RIA	diagnosed using NINCDS-ADRDA criteria	n = 26 AD Men: 590 (244) AD Women: 786 (175) AD: 672.9 (235.3)	nondemented controls	n = 24 C Men: 542 (105) C Women: 617 (109) Control: 567 (110)	AD-Aβ42: 410 (83) AD-tau: 649 (305) AD-p-tau: 76 (35) C-Aβ42: 707 (196) C-tau: 326 (142) C-p-tau: 61 (18)

3.2. Qualitative synthesis

The included studies varied in the type of CSF orexin assay and AD spectrum disorder considered (see Table 1 for details). Fourteen studies reported a single AD group relative to a control group^{27, 39, 40, 42-46, 48, 49, 51-53}. One of these studies included participants with MCI whose CSF biomarkers were consistent with AD pathology in the larger AD group⁴⁷. Two studies had separate AD, MCI, and control groups^{41, 50}, with one study solely comparing MCI to controls⁵⁴. One study considered cognitively normal subjects with AD defined by abnormal Aβ₄₂⁴². One study utilized post-mortem samples⁴⁰, with the remaining examining samples collected from living participants.

Eleven studies reported Aβ₄₂, total tau, and p-tau levels in AD and control groups^{27, 41, 43, 47-54}. One study reported Aβ₄₂ and p-tau⁴⁵; one study only reported Aβ₄₂⁴⁶. One study defined their AD group and control groups by having a beta amyloid level below or above 530 pg/ml respectively, not reporting Aβ₄₂ for each group but reporting total and phosphorylated tau⁴². Three studies did not include beta amyloid, total tau, or phosphorylated tau^{39, 40, 44}. Only five studies reported APOEε4 status^{42, 43, 48, 49, 53}.

3.3. Quantitative synthesis

Meta-analysis considering all participants on the AD continuum compared to healthy comparison participants demonstrated a non-significant increase in CSF orexin compared to controls [(Hedge’s g=0.20,(95%CI: −0.06-0.47), p=0.136; Figure 2]. Moderate to large heterogeneity was observed across studies (I²=72.6%). Similar findings were observed considering only those with Alzheimer’s dementia [Hedge’s g=0.25, (95%CI: −0.06-0.55, p=0.113, I²=74.5%, Supplemental Figure 1)]. Inclusion or exclusion of the study that included MCI with AD pathologic biomarkers in the sample did not alter these findings. Pooled data from the three studies that reported MCI also demonstrated similar findings [Hedge’s g=0.10, (95%CI: −0.84-1.04), p=0.828, I²=86.6%, Supplemental Figure 2]. Findings were not appreciably altered by exclusion or use of the replication sample of Deuschle, et al. (2014). No evidence of publication bias was identified using visual inspection of Funnel plot or the “trim and fill” method, with no filled studies generated (Figure 3). However, post hoc analysis excluding the lone study that utilized postmortem specimens⁴⁰, did demonstrate a significant increase in CSF orexin relative to controls [[(Hedge’s g=0.26,(95%CI: 0.01-0.51), p=0.044; Supplemental Figure 3], with minimal effect on heterogeneity among studies, which remained moderate to high ((I²=67.1%).

Figure 2.

Forest plot of effect size (Hedges’s g) for CSF orexin in Alzheimer’s disease continuum relative to controls.

Figure 3.

Funnel plot for effect size (Hedge’s g) of CSF orexin in AD relative to controls against standard error.

Meta-regression was conducted to determine if a priori variables were related to heterogeneity across studies. A significant relationship was observed between Hedge’s g for orexin and year of publication, such that more recent studies were associated with increased orexin between AD and controls [β=0.055, p=0.020; Figure 4]. This relationship remained significant excluding the study with postmortem data from the analysis [β=0.049, p=0.025]. An association between Hedge’s g for orexin and Hedge’s g for p-tau was also observed, such that larger differences in p-tau between AD and controls were associated with larger differences in orexin between AD and controls [β=0.417, p=0.031; Figure 5]. No significant associations between orexin and other variables of interest (sex, age, type of orexin assay, Aβ₄₂, or total tau) were observed.

Figure 4.

Scatterplot of moderator analysis demonstrating a significant effect of publication year on effect size (Hedges’s g) for CSF orexin in AD relative to controls.

Figure 5.

Scatterplot of moderator analysis demonstrating a significant effect of effect size (Hedges’s g) for CSF phosphorylated tau in AD relative to controls on effect size (Hedges’s g) for CSF orexin in AD relative to controls.

Since four studies came from the same research group^{43, 51, 52, 54}, an additional post hoc moderator analysis was performed that did not demonstrate a significant effect on the primary outcome measure for this site’s findings compared to others. Additionally, given associations between orexin and both year of publication and p-tau, an exploratory meta-regression was conducted examining the relationship between these variables in the dataset. This analysis demonstrated a non-significant relationship between p-tau and year of publication, [β=0.0996, p=0.106].

4. Discussion

This systematic review and meta-analysis evaluated differences in CSF orexin in persons with AD relative to healthy controls. The principal finding was a non-significant increase in orexin among AD in pooled estimates, which was unchanged across multiple sensitivity analyses that considered only those with Alzheimer’s dementia, MCI, or any AD spectrum disorder. Despite absence of publication bias in the identified studies, exclusion of the lone study that examined postmortem specimens did demonstrate a significant increase in CSF orexin in AD compared to controls in pooled estimates. Differences in orexin were highly heterogeneous across studies, with meta-regression demonstrating two variables, year of publication and effect size for p-tau between AD and control participants, associated with the primary outcome variable. Other moderators including sex, age, Aβ₄₂, total tau, and type of orexin assay did not have significant effects on findings. These findings have important implications for sleep and AD research, and point to future key areas of inquiry.

First, consistent with the complex relationships between sleep and AD frequently observed in the literature⁴, the findings of this systematic review and meta-analysis do not support the presupposition that AD is broadly associated with changes in CSF orexin. However, the association of the primary outcome variable (Hedges’g for orexin in AD versus controls) with both year of publication and p-tau suggests these factors are likely to have influenced differences in findings across studies. The association between orexin and year of publication suggests that over time, as AD diagnosis has refined, the magnitude of the difference in orexin between AD and controls observed in the scientific literature has also increased. The diagnostic criteria and methods used to diagnose AD have evolved substantially^{22, 55}. AD biomarkers that evaluate amyloid pathology, pathologic tau, and neurodegeneration have increased the ability to accurately diagnose AD, and currently these biomarkers are considered part of the pathologic continuum of the disease, even in the absence of cognitive impairment²². Since the diagnosis of AD in living patients can be difficult due to overlap with other dementia syndromes (e.g. vascular dementia, etc.), these AD biomarkers, and particularly p-tau, which more specifically correlates with neurofibrillary tangles in AD, has become increasingly important in identifying the disorder⁵⁶.

The finding that the effect size for p-tau among AD relative to controls was associated with differences in orexin between AD and controls also suggests that heterogeneity across studies may have been due to variability of CSF p-tau over the course of AD. CSF p-tau demonstrates steep increases several years prior to the onset of AD dementia, with levels following a logistic “S-shaped” curve plateauing after onset of dementia^{57, 58}. Thus, while theoretical, studies that consider persons with AD dementia who have had a longer timecourse of cognitive symptoms would potentially be less likely to demonstrate large differences in CSF orexin between AD and controls. This hypothesis would be supported by the observation that the largest negative effect size (i.e., orexin was less in AD compared to controls) was observed in a post-mortem study that examined AD patients with end-stage disease⁴⁰. However, this particular study did not have p-tau values for inclusion in the meta-regression, and so potential effects on orexin over the longitudinal trajectory of the illness remain speculative. Because exclusion of this study from pooled estimates did result in a significant increase in orexin relative to controls, it is also possible that changes in orexin occurring postmortem affected findings. Thus, research that more critically evaluates orexin in AD as a function of AD time course, and considers pre- and postmortem specimens, is likely to be a fruitful area of investigation.

While this investigation identified an association between p-tau and orexin, the causal mechanisms that underlie the association cannot be determined from this study. Prior animal research designed to probe causal mechanisms have suggested that relationships between AD and orexin are mediated through effects on sleep rather than direct effects of orexin on AD pathology^{16, 17}. However, cellular models of AD have also suggested that orexin may directly decrease Aβ clearance and/or affect mitochondrial function^{59, 60}. It is also possible that there are overlapping mechanisms connecting phosphorylation of tau protein and upregulation of the orexin gene. The two active forms of orexin (A&B) are both derived from the same gene and cleaved from a common precursor protein (prepro-orexin)⁶¹. In this context, it is notable that the greatest effect size observed for orexin (i.e., greater orexin in AD than controls) occurred in the investigation that assayed the precursor protein prior to cleavage into the two active neuropeptide isoforms (orexin-A and orexin-B), rather than orexin-A (hypocretin-1). While speculative, this observation could support the hypothesis that increased transcription and translation of the orexin gene may occur in parallel with other aspects of AD pathology related to tau phosphorylation. Under such a scenario, orexin and AD pathology would have an intertwined relationship that could feed into one another, accelerating pathologic processes over the course of AD.

There are limitations of this study that merit discussion. First, there was insufficient data across studies to examine relationships between sleep-wake variables (e.g. sleep efficiency, total sleep time, etc.) and differences in CSF orexin between AD and controls. Given previous studies that have demonstrated an association between these variables and AD biomarkers^{51, 54}, future research in this area should carefully consider sleep phenotypic data when evaluating the relationship between orexin and AD. Relatedly, there may have been other unmeasured/unconsidered variables that affected findings within individual studies that may have affected aggregate results. For example, concurrent use of opiate medications, which are frequently used in the elderly, may have affected findings through recruitment of orexin producing neurons⁶². While radioimmunoassay (RIA) is considered the current standard for quantifying CSF orexin, the accuracy of both RIA and alternatives such as enzyme-linked immunosorbent assay (ELISA) is a debated area in the field^63-66. Despite the absence of an observed effect of assay type in this study, it is possible emerging methods to quantify orexin such as mass spectrometry could affect findings if applied to future studies in AD^{66, 67}. Finally, inherent to any systematic review and meta-analysis, there is the possibility that our methods failed to identify relevant research that would otherwise meet inclusion/exclusion criteria. However, the comprehensive search strategy and efforts made to include all available data, minimizes this possibility.

5. Conclusions

In summary, this systematic review and meta-analysis demonstrated non-significant increases in CSF orexin between AD and controls. Findings were heterogenous across the literature, and related to year of publication and p-tau. Future research that carefully considers sleep data, relies on differences in specific AD biomarkers that enhance diagnosis of the disorder, and carefully examines relationships between orexin and p-tau cross-sectionally and longitudinally are likely to further advance this important area of research at the nexus of sleep and AD.

Highlights.

• Orexin may play a key role in sleep disturbance in Alzheimer’s disease.

• Meta-analysis demonstrated non-significant increases in orexin.

• Associations between phosphorylated tau and orexin were identified.

• Year of publication was also associated with orexin levels.

• Future research will further clarify these relationships.