Abstract

Background

   Obstructive sleep apnoea (OSA) is the most frequent form of
   sleep-disordered breathing in patients with Alzheimer’s disease (AD).
   Available evidence demonstrates that both conditions are independently
   associated with alterations in lipid metabolism. However, it is unknown
   whether the expression of lipids is different between AD patients with
   and without severe OSA. In this context, we examined the plasma
   lipidome of patients with suspected OSA, aiming to identify potential
   diagnostic biomarkers and to provide insights into the
   pathophysiological mechanisms underlying the disease.

Methods

   The study included 103 consecutive patients from the memory unit of our
   institution with a diagnosis of AD. The individuals were subjected to
   overnight polysomnography (PSG) to diagnose severe OSA (apnoea-hypopnea
   index ≥30/h), and blood was collected the following morning. Untargeted
   plasma lipidomic profiling was performed using liquid chromatography
   coupled with mass spectrometry.

Results

   We identified a subset of 44 lipids (mainly phospholipids and
   glycerolipids) that were expressed differently between patients with AD
   and severe and nonsevere OSA. Among the lipids in this profile, 30 were
   significantly correlated with specific PSG measures of OSA severity
   related to sleep fragmentation and hypoxemia. Machine learning analyses
   revealed a 4-lipid signature (phosphatidylcholine PC(35:4),
   cis-8,11,14,17-eicosatetraenoic acid and two oxidized triglycerides
   (OxTG(58:5) and OxTG(62:12)) that provided an accuracy (95% CI) of 0.78
   (0.69–0.86) in the detection of OSA. These same lipids improved the
   predictive power of the STOP-Bang questionnaire in terms of the area
   under the curve (AUC) from 0.61 (0.50–0.74) to 0.80 (0.70–0.90).

Conclusion

   Our results show a plasma lipidomic fingerprint that allows the
   identification of patients with AD and severe OSA, allowing the
   personalized management of these individuals. The findings suggest that
   oxidative stress and inflammation are potential prominent mechanisms
   underlying the association between OSA and AD.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s13195-022-01102-8.

   Keywords: Alzheimer’s disease, Biomarker, Diagnosis, Lipidomics,
   Obstructive sleep apnoea, STOP-Bang questionnaire

Introduction

   Alzheimer’s disease (AD) is the main cause of dementia and one of the
   main causes of disability and death after 75 years of age. Given the
   increase in life expectancy, its prevalence is expected to increase
   drastically in the coming years [[47]1]. Currently, there is a lack of
   drugs that can cure or manage the disease long-term. Therefore, it is
   important to identify all the factors that we can modify to avoid both
   the appearance and progression of this disease.

   In recent years, different sleep disorders, including their duration
   and quality, have been shown to be risk factors for the development of
   AD. Of all known sleep disturbances, the presence of obstructive sleep
   apnoea (OSA) has been consistently identified as a risk factor for AD
   in population studies AD [[48]2, [49]3]. Studies on cognitively healthy
   elderly subjects with OSA have reported increased amyloid β-42 (Aβ-42)
   and phosphorylated tau (p-tau) levels as measured in the cerebrospinal
   fluid (CSF) or on positron emission tomography (PET) [[50]4]. In
   addition, longitudinal studies have reported an increase in the speed
   of cerebral amyloid accumulation promoted by OSA [[51]5]. Its
   prevalence in patients with AD ranges between 45 and 90%, presenting as
   severe OSA in up to 40% of AD patients [[52]6].

   Clinically, although some studies have shown that the presence of OSA
   advances the age of diagnosis of mild cognitive impairment (MCI) and
   AD, suggesting that it could accelerate the progression of these
   diseases in the early stages [[53]7], and that continuous positive
   airway pressure (CPAP) can improve short-term cognitive performance
   [[54]8], other longitudinal studies have not observed that the presence
   of OSA worsens the cognitive evolution of patients with mild–moderate
   AD [[55]9].

   The high prevalence of OSA in patients with AD makes the diagnosis of
   OSA essential, given that it is also a risk factor for hypertension,
   diabetes, heart failure, stroke or depression, all of which are risk
   factors for AD [[56]10, [57]11]. However, the study of sleep in these
   patients is complex. Although polysomnography (PSG) is the technique of
   choice, the need to go to the hospital to sleep one or two nights often
   limits its usefulness in patients with AD and can generate sleep data
   that do not correspond to the patient’s usual routine at home [[58]12].
   In addition, simple screening questionnaires, including the STOP-Bang
   questionnaire (SBQ) [[59]13] and the Berlin questionnaire (BQ)
   [[60]14], have been shown to be insufficient for identifying subjects
   at risk of OSA in this type of population [[61]15]. Thus, the search
   for new screening tools to detect OSA in this population remains.

   Lipidomics is the science of the large-scale determination of
   individual lipid species in biological samples and has demonstrated
   great potential in the search for disease-associated biomarkers. OSA is
   a pathologic condition that is strongly associated with systemic lipid
   dyshomeostasis [[62]16]. In addition, OSA can increase lipoxidation, as
   has been evidenced in both the brain and blood of AD patients [[63]17,
   [64]18]. Therefore, the identification of systemic alterations in lipid
   species using high-performance lipidomic platforms could contribute to
   finding OSA-associated lipid profiles in AD and increase our
   understanding of the relationship between these two complex
   pathological conditions.

   Therefore, the aim of our study was (i) to evaluate whether the
   expression pattern of circulating lipids is different between AD
   patients with and without severe OSA, which would be of great relevance
   for the practical and noninvasive screening of OSA among patients with
   AD; (ii) to investigate whether the severity of OSA is correlated with
   changes in plasma lipid levels; and (iii) to evaluate the diagnostic
   performance of lipidomics findings in joint use with classic screening
   tests such as the SBQ.

Materials and methods

Study population

   This is an ancillary study from trial [65]NCT02814045 that was
   conducted in the Cognitive Disorders Unit of the Hospital Universitari
   Santa Maria (Lleida, Spain) from November 2014 to November 2017 to
   evaluate the cognitive evolution of AD patients with and without OSA
   after 1 year of follow-up. The patients were recruited prospectively
   and consecutively according to the eligibility criteria: (1) males and
   females above 60 years without specific treatment for dementia at the
   moment of inclusion and with a new diagnosis of mild or moderate AD
   (Mini-Mental State Examination (MMSE) score ≥20) according to the
   National Institute on Aging–Alzheimer’s Association (NIA-AA) criteria
   [[66]19]; (2) absence of visual or hearing problems that, in the
   investigator’s judgement, would decrease compliance with the
   neuropsychological examination; (3) signed informed consent from the
   patient and the responsible caregiver (and/or if applicable, the legal
   representative if different from the responsible caregiver); and (4) a
   knowledgeable and reliable caregiver accompanying the patient to all
   clinic visits during the study.

   The exclusion criteria were as follows: (1) a previous diagnosis of OSA
   treated with CPAP; (2) severe AD, other types of dementia or patients
   with mild–moderate AD with current acetylcholinesterase inhibitor
   treatment or memantine; (3) presence of any previously diagnosed sleep
   disorder: narcolepsy, severe insomnia or chronic lack of sleep; (4)
   comorbidities such as cancer, severe depression, severe renal or
   hepatic insufficiency and severe cardiac or respiratory failure; and
   (5) the presence of excessive somnolence for unknown reasons. All
   exclusion criteria are available in the paperwork for [67]NCT02814045.

Study design

   Patients with mild–moderate AD who gave consent to participate in the
   study underwent a detailed interview regarding personal history, a
   general clinical examination for associated conditions and
   comorbidities and anthropometric data collection. At baseline,
   participants were evaluated by a polysomnographic study, and blood and
   CSF samples were obtained to determine the APOE genotype and the levels
   of Aβ42, total tau (t-tau) and p-tau, respectively.

   Eligible individuals were selected and classified as severe OSA
   (apnoea-hypopnea index [AHI] ≥30/h) and nonsevere OSA (AHI <30/h)
   patients based on the PSG findings. Only those with a complete PSG and
   available blood samples for determining the plasma lipidome were
   included in the present study.

   All participants underwent cognitive assessment using the MMSE
   [[68]20]. Seventy patients also underwent a semistructured sleep
   questionnaire that included the SBQ for the detection of OSA. The SBQ
   comprises 8 items requiring dichotomous responses related to OSA,
   snoring, tiredness, observed sleep apnoea, high blood pressure, body
   mass index (BMI), age, neck circumference and sex. The score ranges
   from 0 to 8, with the highest scores associated with a high probability
   of OSA. A cut-off score of ≥ 3 is considered high risk of
   moderate/severe OSA, and <3 is considered low risk [[69]13].

Clinical variables

   The following variables were collected: age, sex, years of education,
   unhealthy habits (alcohol consumption and smoking), vascular risk
   factors (hypertension, diabetes mellitus, dyslipidaemia, stroke and
   heart diseases) and personal psychiatric history. BMI was calculated as
   body weight (in kg)/height (in m^2). Excessive daytime sleepiness was
   evaluated by the Epworth Sleepiness Scale (ESS) and was defined as a
   total ESS score > 10 [[70]21].

Polysomnography (PSG)

   PSG was performed according to international guidelines to classify the
   patients as nonsevere OSA (AHI <30/h) or severe OSA (AHI ≥30/h)
   patients. The following devices were used: an Embletta® sleep monitor
   (Embla, Canada), a Sibelmed Exea Series 5 (Sibel SAU, Spain), a Philips
   Respironics Alice 6 LDx (Philips, USA) and an ApneaLink Resmed (Resmed,
   Canada).

   Apnoea was defined as the absence of airflow for more than 10 s.
   Hypopnea was defined as a reduction in airflow that lasted more than 10
   s leading to arousal or oxygen desaturation (represented by a decrease
   in oxygen saturation greater than 3%). The AHI was defined as the
   number of apnoea and hypopnea events per hour during the time spent
   sleeping. CT90 was defined as the percentage of cumulative sleep time
   with oxyhaemoglobin saturation (SpO2) <90%. The arousal index was
   defined as the number of awakening events per hour after sleep onset.

Genetic analysis

   DNA was extracted from buffy coat cells using a Maxwell® RCS blood DNA
   kit (Promega, USA). APOE genotyping was performed using TaqMan® SNP
   genotyping assays (C_3038793_20 and C_904973_10) and real-time
   polymerase chain (PCR) according to the manufacturer’s user guide
   (Publication No. MAN0009593, revision B.0).

Cerebrospinal fluid (CSF) biomarkers

   All patients underwent lumbar puncture between 8:00 and 10:00 am to
   avoid variations related to the circadian rhythm. Samples were
   collected in polypropylene tubes, centrifuged at 2000 × g for 10 min at
   4°C and stored at −80°C until use. The levels of CSF Aβ42 (Innotest®
   β-Amyloid (1-42)), t-tau (Innotest® hTAU Ag) and p-tau (Innotest®
   Phospho-Tau (181P)) were determined by the enzyme immunoassay method
   according to the manufacturer’s instructions (Fujirebio Europe, Ghent,
   Belgium). All samples were measured in duplicate and expressed in
   pg/ml. Samples were obtained with support from IRBLleida Biobank
   (B.0000682) and PLATAFORMA BIOBANCOS PT17/0015/0027.

Lipidomic profiling

   The plasma lipidome of patients was determined using untargeted
   lipidomic analysis. The lipids were extracted based on a previously
   published and validated method [[71]22]. Lipid extracts were analysed
   via ultrahigh-performance liquid chromatography (UHPLC) coupled with
   electrospray ionization quadrupole time of flight (ESI-Q-TOF) tandem
   mass spectrometry (MS/MS) according to a previously published method
   [[72]23, [73]24] using an Agilent 1290 liquid chromatography system
   (Agilent Technologies, Santa Clara, CA, USA) coupled with a 6520
   ESI-Q-TOF mass spectrometer (Agilent Technologies, Santa Clara, CA,
   USA) was used. Data were acquired in both positive and negative
   ionization modes.

Lipidic identification

   The differentially expressed features were identified in the Human
   Metabolome Database (HMDB) [[74]25] according to the exact mass and
   retention time, while the molecular weight tolerance was adjusted to 30
   ppm. Potential identities were confirmed by comparison of the exact
   mass, retention time and MS/MS spectral fragmentation pattern of the
   class representative internal standards, when available, with a public
   database using the LC–MS/MS search module of the HMDB web server, as
   well as Lipidmatch and MSDIAL software [[75]26].

Pathway enrichment analysis

   The annotated differential lipids were searched against the KEGG
   library of H. sapiens. Pathway enrichment analysis was performed
   through the MetaboAnalyst web service
   ([76]http://www.metaboanalyst.cat/) [[77]27]. A hypergeometric test was
   applied for overrepresentation analysis. p values regarding
   significantly affected pathways were adjusted for the false discovery
   rate (FDR).

Statistical analyses

   Descriptive statistics were used to summarize the characteristics of
   the study population. Continuous variables were summarized using the
   mean (standard deviation) for normally distributed data and the median
   (25th percentile; 75th percentile) for nonnormally distributed data.
   The normality of the distributions was assessed by the Shapiro–Wilk
   test. Categorical data were summarized using frequency (percentage).
   Clinical and sociodemographic characteristics of the patients were
   compared between groups separated according to the OSA status (AHI ≥ 30
   vs. AHI <30) using the t test (or an equivalent nonparametric test) or
   the chi-squared test depending on whether the variables were
   quantitative or categorical, respectively. Lipid levels were
   log-transformed for statistical purposes. Linear models with empirical
   Bayes statistics were used to evaluate differences in lipid levels
   between groups [[78]28]. Models for differential expression between
   groups were adjusted for age, sex and body mass index (BMI). Lipids
   with a significant difference (p value <0.05) between groups and a fold
   change (FC) higher than 1.25 (or lower than 0.8 for downregulated
   lipids) were considered differentially expressed. Differential
   expression between study groups was displayed in volcano plots.
   Correlations between differentially expressed lipids and PSG parameters
   were evaluated using Pearson’s correlation coefficient. Furthermore,
   the variable importance, calculated as the average of 50 runs of random
   forests, was calculated for each differentially expressed lipid.

   A feature selection process based on the random forest algorithm
   [[79]29] was performed to construct a lipidomic signature that
   predicted severe OSA. This feature selection process is suitable for
   high-dimensional data and was applied to the differentially expressed
   lipids identified and repeated 10 times to account for variability in
   the selection process. The lipids selected in some executions of the
   process were included in the candidate set for the final predicted
   model. The candidates were included as predictors in a logistic model
   with OSA status as a response. The best model, based on the Akaike
   information criterion (AIC), included the lipids that composed the
   final lipidomic signature. The accuracy (95% confidence interval (CI))
   of the model was estimated and compared. Receiver operating
   characteristic (ROC) curves were constructed for the lipidomic
   signature and the reference questionnaire (STOP-Bang), and the area
   under the ROC curve (AUC) was used as the global discrimination value
   measure. p values <0.05 were considered to indicate statistical
   significance. All statistical analyses were performed using R software,
   version 4.0.2.

Results

Characteristics of the sample

   A total of 103 consecutive mild–moderate AD patients with clinical data
   and plasma samples were included in the study. The mean (SD) age of the
   population was 75.49 (5.62) years; 59 (57.28%) participants were women,
   and the MMSE score was 23.5 (2.38) points. Arterial hypertension was
   the most frequent vascular risk factor, present in 60 (58.25%)
   patients, followed by dyslipidaemia in 47 (45.63%) participants and
   diabetes in 19 (18.44%) participants. Regarding the sleep parameters,
   the mean ESS was 5.57 (4.09), and the mean AHI was 29.53 (21.85).
   Sixty-three patients were considered to have nonsevere OSA (AHI <30),
   and 40 patients were considered to have severe OSA (AHI≥30). The
   characteristics of the patients at baseline were similar between the
   severe OSA and nonsevere OSA groups. The baseline characteristics by
   OSA status are summarized in Table [80]1.

Table 1.

   Characteristics of Alzheimer’s disease patients according to their
   obstructive sleep apnoea (OSA) status. BMI body mass index, AD
   Alzheimer’s disease, AHI apnoea-hypopnea index per hour, CSF
   cerebrospinal fluid, APOE Ɛ4 apolipoprotein epsilon 4, MMSE Mini-Mental
   State Examination, ACE angiotensin-converting enzyme, OSA obstructive
   sleep apnoea, SaO2 oxygen saturation, CT90 time with SaO2 <90%
   All (n=103) Non-severe OSA (AHI<30/h) (n = 63) Severe OSA (AHI≥30/h) (n
   = 40) p value
   Demographic characteristics
    Age at baseline visit (years), median [IQR] 76.0 [72.0; 80.0] 75.0
   [71.5; 79.5] 78.0 [72.8; 80.0] 0.14
    Gender (female), n (%) 61 (59.2%) 42 (66.7%) 19 (47.5%) 0.085
    Education (years), mean (SD) 7.31 (2.68) 7.46 (2.32) 7.08 (3.19) 0.511
    BMI (kg/m^2), mean (SD) 27.7 [25.0; 31.1] 27.6 [24.7; 30.6] 28.1
   [26.8; 32.4] 0.081
   Medical disorders
    Hypertension (yes), n (%) 60 (58.3%) 37 (58.7%) 23 (57.5%) 0.999
    Diabetes (yes), n (%) 19 (18.4%) 13 (20.6%) 6 (15.0%) 0.999
    Hypercholesterolaemia (yes), n (%) 42 (40.8%) 24 (38.1%) 18 (45.0%)
   0.625
    Depression (yes), n (%) 29 (28.2%) 18 (28.6%) 11 (27.5%) 0.999
    Smoker 0.662
     Nonsmoker, n (%) 82 (79.6%) 48 (76.2%) 34 (85.0%)
     Current, n (%) 1 (0.97%) 1 (1.59%) 0 (0.00%)
     Former, n (%) 20 (19.4%) 14 (22.2%) 6 (15.0%)
    Family history of AD (yes), n (%) 40 (38.8%) 21 (33.3%) 19 (47.5%)
   0.219
    Use of acetylcholinesterase inhibitors or memantine (yes), n (%) 98
   (95.1%) 60 (95.2%) 38 (95.0%) 0.999
   Polysomnography parameters
    AHI (events/h), median [IQR] 23.6 [12.2; 47.7] 15.0 [7.47; 20.3] 52.2
   [42.3; 62.0] <0.001
    CT90, % 2.20 [0.31; 9.31] 1.12 [0.18; 4.86] 5.80 [1.07; 14.9] 0.004
    Mean SaO2, % 93.0 [92.0; 94.0] 93.0 [92.0; 94.0] 93.0 [92.0; 93.0]
   0.269
    Minimum SaO2, % 84.0 [79.0; 87.0] 86.0 [82.0; 88.0] 81.5 [78.0; 85.0]
   0.003
    Arousal index, events/h 37.6 [26.2; 49.5] 29.0 [19.7; 40.6] 46.1
   [40.0; 55.8] <0.001
   Epworth Sleepiness Scale (0–24), median [IQR] 5.00 [2.50; 8.00] 5.00
   [2.00; 8.00] 5.00 [3.00; 8.00] 0.773
   STOP-Bang score
   MMSE score 23.0 [22.0; 25.0] 23.0 [22.0; 25.0] 24.0 [22.0; 25.0] 0.512
   AD biomarkers
    Aβ42 CSF (pg/ml), median [IQR] 493 [399; 580] 489 [393; 584] 505 [406;
   564] 0.679
    Total tau CSF (pg/ml), median [IQR] 494 [350; 696] 494 [369; 707] 469
   [346; 684] 0.676
    Phospho-tau CSF (pg/ml), median [IQR] 81.0 [55.4; 97.5] 80.0 [58.0;
   95.0] 81.0 [55.1; 98.0] 0.929
    ApoE Ɛ4 (carrier), n (%) 55 (53.4%) 32 (50.8%) 23 (57.5%) 0.644
   Medications
    ACE inhibitors, % 32 (31.1%) 21 (33.3%) 11 (27.5%) 0.707
    Beta-blockers, % 16 (15.5%) 7 (11.1%) 9 (22.5%) 0.202
    Diuretic agents, % 30 (29.1%) 21 (33.3%) 9 (22.5%) 0.339
    Calcium-channel blockers, % 13 (12.6%) 9 (14.3%) 4 (10.0%) 0.738
    Lipid-lowering agents, % 43 (41.7%) 24 (38.1%) 19 (47.5%) 0.376
    Insulin, % 2 (1.94%) 1 (1.59%) 1 (2.50%) 0.999
   [81]Open in a new tab

Untargeted lipidomic analysis

   The first objective of the study was to evaluate the differences in the
   lipidome in patients with and without severe OSA. Nondirected
   lipidomics was performed by LC–MS. After quality control, 1022 lipids
   were detected and included in the analyses. After adjusting for
   confounding factors (age, sex and BMI), 44 differentially expressed
   lipid species were identified, 11 with reduced (FC from 0.65 to 0.75)
   and 33 with increased plasma levels in patients with severe OSA (FC
   from 1.26 to 1.90) (Fig. [82]1A and Fig. [83]S1). Subsequently, we
   identified a lipidomic prediction model for the detection of severe OSA
   based on random forest analysis. In Fig. [84]1B, we can see the
   importance of each lipid in the classification of the study groups
   (severe OSA vs. nonsevere OSA).

Fig. 1.

   [85]Fig. 1
   [86]Open in a new tab

   Untargeted lipidomic profiling in AD patients with severe OSA. A
   Volcano plots of the FC (x-axis) and p value (y-axis) for each detected
   lipid in the comparison of severe OSA vs. nonsevere OSA subjects. Red
   dots represent significantly downregulated (FC <0.80) molecules, and
   blue dots represent significantly upregulated (FC> 1.25) molecules in
   severe OSA patients. The results were adjusted for confounding factors
   (age, sex and BMI). The p value threshold defining statistical
   significance was <0.05. B Top 20 significant lipids in the
   classification of the study groups (severe OSA vs. nonsevere OSA) based
   on random forest. C Significant correlations between PSG parameters of
   OSA severity and the differentially expressed lipids. The colour scale
   illustrates the degree of correlation and ranges from red to blue,
   indicating negative and positive correlations, respectively. Unknown
   features are presented as exact mass @ retention time. Definition of
   abbreviations: AHI apnoea-hypopnea index, LysoPC
   lysophosphatidylcholine, PG phosphatidylglycerol, FC fold change, OSA
   obstructive sleep apnoea, PC phosphatidylcholine, PE
   phosphatidylethanolamine, OxTG oxidized triglyceride, CT90 time with
   oxygen saturation <90%. For interpretation of the references to colour