Associations between nocturnal urinary 6-sulfatoxymelatonin, obstructive sleep apnea severity and glycemic control in type 2 diabetes

ABSTRACT

Reduced nocturnal secretion of melatonin, a pineal hormone under circadian control, and obstructive sleep apnea have been both identified as risk factors for the development of type 2 diabetes mellitus. Whether they interact to impact glycemic control in patients with existing type 2 diabetes is not known. Therefore, this study explores the relationships between obstructive sleep apnea, melatonin and glycemic control in type 2 diabetes. As diabetic retinopathy may affect melatonin secretion, we also explore the relationship between retinopathy, melatonin and glycemic control. Fifty-six non-shift workers with type 2 diabetes, who were not using beta-blockers, participated. Most recent hemoglobin A1c (HbA1c) levels and the results of ophthalmologic examinations were obtained from medical records. Obstructive sleep apnea was diagnosed using an ambulatory device. Sleep duration and fragmentation were recorded by 7-day wrist actigraphy. The urinary 6-sulfatoxymelatonin/creatinine ratio, an indicator of nocturnal melatonin secretion, was measured in an overnight urine sample. Mediation analyses were applied to explore whether low nocturnal urinary 6-sulfatoxymelatonin/creatinine ratio could be a causal link between increasing obstructive sleep apnea severity [as measured by an Apnea Hypopnea Index (AHI)] and poorer glycemic control, and between the presence of retinopathy and glycemic control. AHI and HbA1c were log-scale (ln) transformed. Obstructive sleep apnea was found in 76.8%, and 25.5% had diabetic retinopathy. The median (interquartile range) of urinary 6-sulfatoxymelatonin/creatinine ratio was 12.3 (6.0, 20.1) ng/mg. Higher lnHbA1c significantly correlated with lower 6-sulfatoxymelatonin/creatinine ratio (p = 0.04) but was not directly associated with OSA severity. More severe obstructive sleep apnea (lnAHI, p = 0.01), longer diabetes duration (p = 0.02), retinopathy (p = 0.01) and insulin use (p = 0.03) correlated with lower urinary 6-sulfatoxymelatonin/creatinine ratio, while habitual sleep duration and fragmentation did not. A mediation analysis revealed that lnAHI negatively correlated with urinary 6-sulfatoxymelatonin/creatinine ratio (coefficient = −2.413, p = 0.03), and urinary 6-sulfatoxymelatonin/creatinine negatively associated with lnHbA1c (coefficient = −0.005, p = 0.02), after adjusting for covariates. Mediation analysis indicated that the effect of lnAHI on lnHbA1c was indirectly mediated by urinary 6-sulfatoxymelatonin/creatinine ratio (B = 0.013, 95% CI: 0.0006, 0.0505). In addition, having retinopathy was significantly associated with reduced nocturnal urinary 6-sulfatoxymelatonin/creatinine ratio, and an increase in HbA1c by 1.013% of its original value (B = −0.013, 95% CI: −0.038, −0.005). In conclusion, the presence and severity of obstructive sleep apnea as well as the presence of diabetic retinopathy were associated with lower nocturnal melatonin secretion, with an indirect adverse effect on glycemic control. Intervention studies are needed to determine whether melatonin supplementation may be beneficial in type 2 diabetes patients with obstructive sleep apnea.

KEYWORDS:

• Type 2 diabetes
• melatonin
• glycemic control
• obstructive sleep apnea

Introduction

Melatonin is a neurohormone secreted by the pineal gland during the biological night under control by the central circadian clock in the suprachiasmatic nucleus (SCN) (Brzezinski, 1997). In humans, its secretion follows a diurnal pattern with low levels during the day, an increase initiated 1–2 h prior to habitual sleep onset, a peak in the middle of the night and a gradual fall during the second half of the night (Brzezinski, 1997). Melatonin release decreases with age. Melatonin exerts its effects through two G-protein coupled receptors, MT1 and MT2 (found in the SCN, non-SCN brain tissues, peripheral tissues including pancreatic β- and α-cells) (Peschke & Muhlbauer, 2010). The 24 h rhythm of melatonin levels serves as an internal synchronizer of the circadian system. Melatonin also has potent anti-oxidant and anti-inflammatory properties (Mauriz et al., 2013).

Reduced nocturnal melatonin levels have been shown to be a risk factor for incident diabetes. In a nested case–control study within the Nurses’ Health Study Cohort, women in the lowest tertile of urinary 6-sulfatoxymelatonin (aMT6s, a major urine melatonin metabolite) to creatinine ratio had a nearly 50% increase in the risk of developing diabetes compared with those in the highest tertile (McMullan et al., 2013). In addition, several genetic studies have linked variants of the gene encoding MT2, MTNR1B, to elevated glucose and diabetes risk (Bonnefond et al., 2012; Sparso et al., 2009). These data support a role for melatonin in glucose metabolism. The 24 h profile of plasma melatonin in type 2 diabetes (T2D) patients has only been examined in two small studies (Mantele et al., 2012; Peschke et al., 2006), both found lower nocturnal levels in the patients as compared with controls. The largest study to examine nocturnal melatonin secretion so far involved 2821 older men, including 370 with diabetes. Participants who had an overnight urinary aMT6s/creatinine ratio in the lower rather than higher quartile were significantly more likely to have diabetes (Saksvik-Lehouillier et al., 2015). Whether melatonin plays a role in glycemic control in T2D patients is unknown. Lastly, two studies have documented significantly lower nocturnal melatonin levels in patients with proliferative diabetes retinopathy compared with those without (Chen et al., 2014; Hikichi et al., 2011). This is possibly due to a dysfunctional light perception associated with retinopathy. Whether lower nocturnal melatonin levels in patients with diabetic retinopathy may affect glycemic control has not been explored.

Another factor that should be considered in the relationship between melatonin and T2D is obstructive sleep apnea (OSA). OSA is a complex sleep disorder characterized by repetitive episodes of upper airway closures or partial collapse during sleep, resulting in intermittent hypoxia and fragmented sleep. OSA is a risk factor for insulin resistance and is common in T2D (Reutrakul & Van Cauter E., 2014). Moreover, OSA is a risk factor for cardiovascular morbidity including hypertension, heart disease and stroke (Somers et al., 2008), and recently emerged as a putative risk factor for diabetic microvascular complications (Pallayova et al., 2014). Increased oxidative stress and inflammation resulted from intermittent hypoxia are thought to be prominent underlying mechanisms leading to OSA-related complications (Lavie, 2015). Interestingly, in animal models of intermittent hypoxia, melatonin administration was shown to reduce damages related to increased oxidative stress and inflammation, including endothelial dysfunction and hypertension (Hung et al., 2013), and insulin resistance (Bertuglia & Reiter, 2009). However, positive effects of melatonin on glucose metabolism were not unanimous as a recent study revealed that melatonin blocks insulin release in mouse islets and clonal insulin-secreting cells (Tuomi et al., 2016).

In humans with OSA, melatonin secretion has been explored in a few studies, mostly in non-diabetic participants, with inconsistent results. Compared with non-OSA participants, some found abnormal serum melatonin patterns in patients with OSA including absent nighttime peak (Hernandez et al., 2007), later peak time (Zirlik et al., 2013) or higher levels in the afternoon (Ulfberg et al., 1998). However, some studies found no differences in melatonin rhythm between the two groups (Papaioannou et al., 2012; Wikner et al., 1997). A large study of 2821 older men in whom OSA was assessed by home polysomnography concluded that the presence of moderate to severe OSA was not a significant predictor of nocturnal excretion of urinary aMT6s (Saksvik-Lehouillier et al., 2015). Lastly, melatonin secretion may be affected by additional factors, including sleep duration (Saksvik-Lehouillier et al., 2015), sleep quality (Haimov et al., 1994;Saksvik-Lehouillier et al., 2015) and depression (Srinivasan et al., 2006).

Thus, the relationships between melatonin, OSA and T2D remain poorly understood. It is not known whether there is a relationship between reduced nocturnal melatonin and glycemic control in T2D. While several studies have reported an adverse impact of OSA on glycemic control (Aronsohn et al., 2010; Pillai et al., 2011), whether this adverse impact is mediated by reduced nocturnal melatonin has not been determined so far. In addition, different associations between OSA and glycemic control may exist in Asian population. Therefore, the first objective of this study was to explore the relationship between glycemic control and nocturnal excretion of urinary aMT6s in T2D patients, along with other parameters that may affect glycemic control. The second objective was to explore the relationship between nocturnal excretion of urinary aMT6s and sleep-related (including OSA severity) and diabetes-related parameters (including retinopathy). Lastly, we explored if nocturnal excretion of urinary aMT6s could be a mediator between OSA severity and glycemic control, and between the presence of retinopathy and glycemic control by applying mediation analyses. We hypothesized that OSA severity and the presence of retinopathy might affect nocturnal excretion of urinary aMT6s and influenced glucose metabolism.

Methods

T2D patients attending the endocrinology clinic at the Faculty of Medicine Ramathibodi Hospital (Bangkok, Thailand), were invited to participate. Exclusion criteria included use of melatonin supplement, use of beta-blockers (known to inhibit melatonin synthesis), previously diagnosed OSA, currently pregnant or performing shift work. Additionally, patients with a history of congestive heart failure, chronic obstructive pulmonary disease, end-stage renal disease or severe chronic liver disease, stroke, permanent pacemaker placement and use of certain medications (e.g. opioids, alpha adrenergic blockers, clonidine, methyldopa, nitroglycerin) were excluded in order to obtain valid results from the OSA diagnostic method utilized (see below). All participants gave written informed consent. The protocol was approved by the Committee on Human Rights Related to Research, Faculty of Medicine Ramathibodi Hospital.

Assessment of diabetes history and glycemic control

Research personnel interviewed participants about their diabetes history. Height, age and current medications, and most recent hemoglobin A1c (HbA1c) values (within 3 months) were extracted from medical records. Weight was measured, and body mass index (BMI) was calculated as weight (kg)/height (m)². History of retinopathy was obtained from ophthalmologic examinations, and categorized into those with retinopathy versus no retinopathy. Nephropathy was defined as the presence of albuminuria (urine albumin/creatinine ratio ≥30 mg/g or overt proteinuria) or an estimated glomerular filtration rate <60 mL/min/1.73 m²). Neuropathy was defined as presence of symptoms of peripheral neuropathy such as pain or burning sensation.

Subjective sleep and depressive symptoms assessments

Sleep quality in the past month was assessed using a Thai version of the Pittsburgh Sleep Quality Index (PSQI), with a higher score reflecting poorer sleep quality (Sitasuwan et al., 2014). Depressive symptoms were assessed using a Thai version of the Center for Epidemiologic Studies Depression Scale (CES-D) (Trangkasombat et al., 1997), with a higher score reflecting more depressive symptoms.

Assessment of OSA

OSA was diagnosed using a FDA-approved portable diagnostic device, the WatchPAT 200 (Itamar Medical, Caesarea, Israel), which has been validated against PSG in a diabetic population (Zou et al., 2006). This device is shaped similar to a large watch, worn on the non-dominant wrist immediately before bedtime and removed upon awakening in the morning. It has two probes connecting the device to the participants’ fingers to measure changes in peripheral arterial tone (PAT) and oxygen saturation, along with a built-in actigraph to measure sleep time. The severity of OSA was assessed by the PAT Apnea Hypopnea Index (AHI) that is automatically generated by the software. OSA was considered present if AHI ≥ 5. Other parameters included oxygen desaturation index (ODI) (the average number of times per hour of sleep that the oxygen saturation drops ≥3%), T90 (percent of total sleep time in which the oxygen saturation was <90%) and mean O₂ (mean oxygen saturation during the recorded sleep period).

Because this device relies on changes in PAT, use of certain medications was not allowed. Further, the WatchPAT cannot differentiate obstructive from central apnea events. Therefore, we excluded patients with conditions associated with a high risk for central apnea, as indicated in the exclusion criteria.

Sleep duration, fragmentation and timing measurement

Participants wore an Actiwatch 2 wrist activity monitor (Philips Respironics, Bend, OR) for 7 days. Actiwatches use highly sensitive omnidirectional accelerometers to count the number of wrist movements in 30 s epochs. The software scores each 30 s epoch as sleep or wake based on a threshold of activity counts that are estimated using activity within the epoch being scored as well as the epochs 2 min before and after that epoch. Bedtime and wake time are set by the researcher using event markers, sleep log data and in-person reviews of sleep timing with the participants. Sleep duration, efficiency and fragmentation were calculated using Actiware 6.0 software. Sleep duration was defined as the amount of actual sleep obtained at night, and sleep efficiency as percentage of time in bed spent sleeping. Sleep fragmentation was an index of restlessness during the sleep period, expressed as a percentage. Mid-sleep time was the midpoint of sleep (between sleep onset and sleep end time). For each participant, the mean across all available nights was used.

Nocturnal urinary excretion of 6-sulfatoxymelatonin assessment

The participants were instructed to discard the last void prior to bedtime and collect each subsequent void until the first next-morning void. The samples were stored in a dark bottle at room temperature, the total volume measured and then stored at –20°C until assayed. Urinary 6-sulfatoxymelatonin (aMT6s), a major melatonin metabolite, was measured by a competitive enzyme-linked immunosorbent assay (Bühlmann Laboratories AG, Schönenbuch, Switzerland). Urine creatinine was analyzed on an automated biochemical analyzer (Dimension RxL; Dade Behring, Newark, DE). Urinary aMT6s/creatinine ratios were calculated by dividing urinary aMT6s levels by urine creatinine concentration, expressed as nanograms per milligram. These values were referred to as nocturnal urinary aMT6s.

Statistical analysis

Continuous variables were expressed as mean ± SD if they were normally distributed, otherwise they were expressed as median (range). Categorical variables were expressed as frequencies/percentages. A simple linear regression was applied to determine the associations between independent variable (i.e. demographic and sleep variables) and dependent variable (i.e. nocturnal urinary aMT6s and HbA1c). Sleep variables and HbA1c were natural log (ln) transformed where appropriate to meet normality assumptions.

Mediation analysis was applied to determine whether the effects of OSA parameters (lnAHI, ODI, T90 and mean O₂) on HbA1c could be partly explained by nocturnal urinary aMT6s (Imai et al., 2010; MacKinnon et al., 2002; MacKinnon et al., 2007). A mediation diagram was constructed as displayed in Figure 1. Two equations were constructed as follows:

(1)

(2)

Figure 1. A mediation diagram of the associations between lnAHI, nocturnal urinary aMT6s and lnHbA1c. In this mediation diagram, lnHbA1c is the outcome variable, lnAHI is the causal variable and nocturnal urinary aMT6s is the mediator. Coefficients are noted for each path.

Nocturnal urinary aMT6s, a mediator, was first regressed on each sleep parameter (lnAHI, ODI, T90, mean O₂) one by one (path a). The lnHbA1c was then regressed on nocturnal urinary aMT6s and lnAHI (path b). Covariables whose p-values were <0.1 in the simple linear regression were considered in these two equations. Only significant covariables (p < 0.05) were kept in the final models. Causal association effects were then estimated using the product-of-coefficient method (MacKinnon et al., 2002, 2007). A bootstrap analysis with 1000 replications was applied to estimate average mediation effects (Preacher & Hayes, 2008). The mediation effect was estimated for each bootstrap, averaged across 1000 replications and its corresponding 95% confidence interval (CI) was determined using a bias-corrected bootstrap technique.

In addition, the causal relationship pathway of retinopathy → nocturnal urinary aMT6s → glycemic control was explored (Figure 2). Therefore, a two-stage least square instrumental variable regression, which allows an estimation of a causal relationship among variables, was applied. In attempting to estimate a causal effect of a particular phenotype on an outcome of interest, an instrument is an independent variable which affects the outcome only through its effect on a phenotype. Retinopathy was considered as the instrumental variable, nocturnal urinary aMT6s level was considered as the intermediate phenotype and lnHbA1c was the outcome of interest (Didelez & Sheehan, 2007; Staiger & Stock, 1997). The analysis was carried out as follows: first, nocturnal urinary aMT6s was regressed on the instrumental variable retinopathy, adjusted for diabetes duration (path a in Figure 2). The predicted nocturnal urinary aMT6s was then estimated from the first equation. F-statistic (hereafter called F-first) was applied to assess whether the retinopathy qualified as the instrumental variable (i.e. F-first > 10 (Didelez and Sheehan, 2007)). Second, the outcome variable of lnHbA1c was fitted on the predicted nocturnal urinary aMT6s value, adjusted for age (path b in Figure 2). Analyses were performed using STATA 14.0 software. A p-value of <0.05 was considered statistically significant.

Figure 2. A causal association diagram between retinopathy, nocturnal urinary aMT6s and lnHbA1c. In this diagram, lnHbA1c is the outcome variable, aMT6s is a mediator phenotype and retinopathy is an instrumental variable.

Results

Baseline demographic, glycemic and sleep characteristics

Fifty-six patients, mean age 52.4 years (range: 20–71), participated. Their demographics, diabetes and sleep characteristics are given in Table 1. Among the 56 participants, 17 (30.3%) were male and 16 (28.6%) were using insulin, and the median HbA1c was 7.2 % (range: 5.3–12.6). OSA prevalence was 76.8% (95% CI: 65.4%, 88.2%) with a median AHI (range) of 10.0 (0.4, 70.2), ODI of 5.6 (0.2, 63.3), T90 of 0.15 (0, 24.3) and mean O₂ of 96 (92, 97). The prevalence of retinopathy, nephropathy and neuropathy was 25.5% (95% CI: 13.6%, 37.3%), 22.6% (95% CI: 10.9%, 34.3%) and 22.2% (95% CI: 10.8%, 33.7%), respectively. The median nocturnal urinary aMT6 was 12.3 ng/mg (range: 0.6–36.5 ng/mg).

Table 1. Participants’ demographic, diabetic and sleep characteristics.

	Results
Demographic and diabetic characteristics
Age (years, mean ± SD)	52.4 ± 11.7
Male (number (%))	17 (30.3)
BMI (kg/m^2, mean ± SD)	28.7 ± 5.2
Diabetes duration (years, median (range))	7.0 (1.0, 35.0)
Insulin use (number (%))	16 (28.6)
HbA1c (%, median (range))	7.2 (5.3, 12.6)
HbA1c (mmol/mol, median (range))	55 (34, 114)
Retinopathy (number (%))	14 (25.5)
Nephropathy (number (%))	12 (22.6)*
Neuropathy (number (%))	12 (22.2)**
Nocturnal urinary aMT6s (ng/mg, median (range))	12.3 (0.6, 36.5)
Subjective sleep and depressive symptoms
PSQI score (mean ± SD)	7.5 ± 3.4
CES-D score (mean ± SD)	13.0 ± 6.2
Objective sleep measures
OSA diagnosis (number (%))	43 (76.8)
AHI (events/h, median (range))	10.0 (0.4, 70.2)
ODI (events/h, median (range))	5.6 (0.2, 63.3)
T90 (median (range))	0.15 (0, 24.3)
Mean O2 ((%), median (range))	96 (92, 97)
Sleep duration (h, mean ± SD)	6.0 ± 1.0
Sleep fragmentation (mean ± SD)	21.7 ± 7.2
Sleep efficiency (%, mean ± SD)	81.2 ± 6.4
Bedtime^§ (mean ± SD)	22:38 ± 01:13
Mid-sleep time,^a mean ± SD	02:27 ± 01:01

*n = 53; **n = 54.

^aTime is presented in 24 h clock time.

Bivariate correlations between the variables are listed in Table 1, and nocturnal urinary aMT6s and HbA1c are shown in Table 2. Longer diabetes duration, insulin use, the presence of OSA and more severe OSA (higher lnAHI and ODI) significantly correlated with lower nocturnal urinary aMT6s levels, whereas higher mean O₂ was significantly correlated with higher urinary aMT6s levels. There was no correlation between urinary aMT6s and age (r = 0.001, p = 0.991), or between urinary aMT6s and T90 (p = 0.112). There were no significant bivariate associations between nocturnal urinary aMT6s and any other variables in Table 1.

Table 2. Bivariate analysis between individual variables and nocturnal urinary aMT6s and lnHbA1c.

	Nocturnal urinary aMT6s		lnHbA1c
Variables	Coefficient	p Value	Coefficient	p Value
Age	−.001	0.991	−.005	0.014
Female versus male	3.032	0.264	−.032	0.530
BMI	−.151	0.537	.005	0.318
Diabetes duration	−.329	0.022	.000	0.949
Insulin use	−5.997	0.027	.067	0.192
Retinopathy	−6.981	0.014	.139	0.008
Nephropathy	−5.327	0.083	.070	0.228
Neuropathy	−3.017	0.323	.019	0.738
Nocturnal urinary aMT6s			–0.005	0.036
Subjective sleep and depressive symptoms
PSQI score	−.005	0.892	−.005	0.395
CES-D score	−.268	0.168	.005	0.197
Objective sleep measures
OSA diagnosis	−6.517	0.025	.038	0.491
lnAHI	−2.998	0.014	.013	0.567
ODI	−.229	0.026	.002	0.381
T90	−.529	0.112	.002	0.692
Mean O2	2.955	0.004	−.027	0.163
Sleep duration	.511	0.687	−.018	0.364
Sleep fragmentation	−.283	0.155	−.004	0.180
Sleep efficiency	.246	0.211		0.976
Bedtime	1.405	0.192	.000	0.990
Mid-sleep time	1.161	0.345	−.004	0.863

For glycemic control, there was a significant association between lower nocturnal urinary aMT6s and higher lnHbA1c levels, whereas the presence or severity of OSA was not correlated with lnHbA1c. None of the other sleep variables was significantly associated with lnHbA1c. Younger age was associated with poorer glycemic control.

The presence of retinopathy, but not of other diabetic complications, was significantly associated with both lower nocturnal urinary aMT6s and higher lnHbA1c levels.

Relationship between osa severity, nocturnal urinary excretion of 6-sulfatoxymelatonin and glycemic control

Because in bivariate analyses, more severe OSA was significantly associated with lower nocturnal urinary aMT6s levels, and lower nocturnal urinary aMT6s was significantly associated with poorer glycemic control, we further investigated if lower nocturnal urinary aMT6s level could be a mediator between increasing severity of OSA and poorer glycemic control. The mediation pathway higher lnAHI → lower nocturnal urinary aMT6s → higher lnHbA1c was therefore constructed to further explore this relationship (see Figure 1), adjusting only for significant covariables (Table 3). The mediation model suggested that lower lnAHI was associated with higher nocturnal urinary aMT6s levels (coefficient = −2.413, p = 0.031, path a) after adjusting for diabetes duration. The outcome (i.e. lnHbA1c) model indicated that nocturnal urinary aMT6s was negatively associated with lnHbA1c (coefficient = −0.005, p = 0.02, path b) after adjusting for age, whereas a direct association between lnAHI and lnHbA1c was not significant.

Table 3. Mediation analysis of lnAHI, nocturnal urinary aMT6s and lnHbA1c.

	Factors	b	SE	t	p	95% CI
lnAHI
lnAHI → aMT6s	lnAHI	−2.413	1.119	−2.156	0.031	−4.606, −0.219
Path a (Figure 1)	Diabetes duration	−0.265	0.135	−1.958	0.050	−0.531, 0.000
aMT6s → lnHbA1c	aMT6s	−0.005	0.002	−2.320	0.020	−0.010, −0.001
Path b (Figure 1)	lnAHI	−0.001	0.021	−0.045	0.964	−0.042, 0.040
	Age	−0.005	0.002	−2.725	0.006	−0.008, −0.001
ODI
ODI → aMT6s	ODI	−0.190	0.097	−1.968	0.049	−0.380, −0.001
	Diabetes duration	−0.278	0.135	−2.053	0.040	–0.544, –0.013
aMT6s → lnHbA1c	aMT6s	−0.005	0.002	−2.276	0.023	−0.010, −0.001
	ODI	0.000	0.002	0.171	0.864	−0.003, 0.004
	Age	−0.005	0.002	−2.719	0.007	−0.008, −0.001
T90
T90 → aMT6s	T90	−0.438	0.311	−1.410	0.159	−1.046, 0.171
	Diabetes duration	−0.305	0.136	−2.235	0.025	−0.572, −0.037
aMT6s → lnHbA1c	aMT6s	−0.006	0.002	−2.426	0.015	−0.010, −0.001
	T90	−0.001	0.006	−0.133	0.894	−0.012, 0.010
	Age	−0.005	0.002	−2.733	0.006	−0.008, −0.001
Mean O2
Mean O2 → aMT6s	Mean O2	2.611	0.938	2.784	0.005	0.773, 4.449
	Diabetes duration	−0.263	0.131	−2.008	0.045	−0.520, − 0.006
aMT6s → lnHbA1c	aMT6s	−0.004	0.002	−1.777	0.076	−0.009, 0.000
	Mean O2	−0.024	0.019	−1.299	0.194	−0.061, 0.012
	Age	−0.005	0.002	−2.987	0.003	−0.009, −0.002

aMT6s, nocturnal urinary aMT6s.

A mediation effect was then estimated using a 1000-replication bootstrap (see Table 4). This indicated that the effect of lnAHI on lnHbA1c was significantly mediated by nocturnal urinary aMT6s such that every one unit increase in lnAHI would be associated with decreased aMT6s, and an increase in lnHbA1c of approximately 0.013, which is equivalent to 1.013% of its original value (95% CI: 1.000, 1.052). The direct pathway of lnAHI → lnHbA1c revealed no significant effect.

Table 4. Mediation effects of AHI on HbA1c through nocturnal urinary aMT6s.

	Pathway	B	SE	Bias	95%CI
lnAHI
Indirect	lnAHI → nocturnal urinary aMT6s → lnHbA1c (path ab, fig 1)	0.0132	0.0118	0.0015	0.0006, 0.0505
Direct	lnAHI → lnHbA1c (path c, fig 1)	−0.0009	0.026	−0.0008	−0.0648, 0.0394
ODI
Indirect	ODI →nocturnal urinary aMT6s →lnHbA1c	0.0010	0.0008	0.0001	0.0001, 0.0033
Direct	ODI → lnHbA1c	0.0003	0.0018	−0.0001	−0.0041, 0.0033
T90
Indirect	T90 → nocturnal urinary aMT6s → lnHbA1c	0.0024	0.0029	0.0010	0.0004, 0.0107
Direct	T90 → lnHbA1c	−0.0074	0.0089	−0.0028	−0.0227, 0.0074
Mean O2
Indirect	Mean O2 → nocturnal urinary aMT6s → lnHbA1c	−0.0111	0.0068	0.0001	−0.0287, −0.0010
Direct	Mean O2 → lnHbA1c	−0.0244	0.0209	0.0018	−0.0668, 0.0153

Mediation effect was estimated using a bootstrap with 1000 replications. For each bootstrap, indirect effect of sleep parameter on lnHbA1c through nocturnal urinary aMT6s (i.e. ab) was estimated. Then, it was averaged among 1000 replications.

In addition, other sleep parameters including ODI, T90, mean O₂ were also explored (see Tables 3 and 4). This revealed that every one unit increase in T90 and ODI would be associated with decreased nocturnal urine aMT6s, and an increase in lnHbA1c of 0.0024 (95% CI: 0.0004, 0.0107) and 0.001 (95% CI: 0.0001, 0.0033). Moreover, every one unit increase in mean O₂ would be associated with a significant increase in nocturnal urine aMT6s, and a decrease in lnHbA1c of 0.0111 (95% CI: −0.0287, −0.0010).

Relationship between retinopathy, nocturnal urinary amt6 and glycemic control

Because the presence of retinopathy was associated with both lower nocturnal urinary aMT6s level and poorer glycemic control, and lower nocturnal urinary aMT6s level was significantly associated with poorer glycemic control, we further explored if the nocturnal urinary aMT6s could be associated with retinopathy and resulted in poor glycemic control. Thus, the pathway retinopathy → nocturnal urinary aMT6s → lnHbA1c was assessed using instrumental variable regression (Figure 2).The result revealed that having retinopathy was significantly associated with reduced nocturnal aMT6s level, and an increase in HbA1c by 1.013% of its original value (B = −0.013, 95% CI: −0.038, −0.005). However, an estimated F-first was 5.34 (p = 0.008), indicating that retinopathy was not a strong instrumental variable of nocturnal urinary aMT6s and suggesting that other factors may also influence nocturnal urinary aMT6s.

Discussion

This study explored the relationship between nocturnal urinary aMT6s, OSA severity and glycemic control in Asian patients with T2D. We found that more severe OSA was significantly associated with lower nocturnal urinary aMT6s level, which was in turn significantly associated with poorer glycemic control. Our mediation analysis suggested that nocturnal urinary aMT6s could serve as a mediator between OSA severity and glycemic control, that is one unit increase in lnAHI would be associated with lower aMT6s levels, and an increase in HbA1c of approximately 1.013% of its original value. For instance, patients aged 55 years with T2D having an AHI of 5, 10, 30 (reflecting no, mild and severe OSA) would have nocturnal urinary aMT6s of 15.9, 14.3 and 11.7 ng/mg, respectively. Consequently, his/her HbA1c would be 7.1%, 7.2% and 7.3% (54, 55 and 56 mmol/mol), respectively. The analyses on other sleep parameters (ODI, T90 and mean O₂) revealed similar associations. Although statistically significant, the effect size is relatively small and the results should be interpreted with caution. While the direct effect between OSA severity and glycemic control was not seen, the mediation effects through nocturnal urinary aMT6s were detected using the mediation analysis with 1000-replication bootstrap. Our data possibly suggest that melatonin may serve as an additional pathway in which OSA severity is related to glycemic control. While the direct effect of OSA severity on HbA1c was not seen as previously reported in Western population, it could partly be due to different associations between BMI, OSA and glycemic control as Asians tend to have lower BMI.

While the underlying mechanisms between OSA severity and nocturnal urinary aMT6s are not explored, there are several possible explanations. Intermittent hypoxia and EEG arousals due to OSA may disrupt melatonin synthesis through an activation of hypothalamic pituitary adrenal axis seen in patients with OSA (Vgontzas et al., 2007). In healthy volunteers, corticotropin-releasing hormone administration exerted an inhibitory effect on melatonin secretion (Kellner et al., 1997). Sleep disruptions by OSA may also alter circadian rhythmicity and melatonin synthesis. Some studies, mostly in non-diabetic participants, found similar patterns of melatonin secretion between those with and without OSA (Papaioannou et al., 2012; Saksvik-Lehouillier et al., 2015; Wikner et al., 1997), however, some found differences. In 33 men with OSA, more severe OSA was associated with higher peak serum melatonin levels at night (Brzecka et al., 2001), while others found that melatonin peaked later in those with OSA and returned to normal after treatment with continuous positive airway pressure (CPAP) (Hernandez et al., 2007; Zirlik et al., 2013). However, another small study found no effect of CPAP on nocturnal urinary melatonin excretion (Wikner et al., 1997). Of note, it is possible that the relationship between OSA severity and nocturnal melatonin secretion could be affected by geographic locations as nocturnal urinary aMT6s in the current study, conducted in Thailand (latitude 13N), is less likely to be subjected to seasonal variations compared with those collected in countries located at higher latitudes. The mechanisms underlying the altered melatonin synthesis in OSA and whether these differ between those with and without diabetes, as well as the effects of CPAP treatment require further investigations.

To our knowledge, this is the first time that the relationship between nocturnal urinary aMT6s and glycemic control is described. Melatonin rhythm serves as internal hormonal synchronizer of central and peripheral circadian oscillators. Thus, lower amplitude of melatonin may lead to more circadian misalignment resulting in adverse metabolic outcomes. Experimental circadian misalignment in healthy volunteers has been shown to result in glucose intolerance and insulin resistance (Leproult et al., 2014; Scheer et al., 2009). Furthermore, both isoforms of melatonin receptors are found in the pancreatic β-cells and α-cells (Peschke & Muhlbauer, 2010), and melatonin has been shown to modulate insulin secretion (Peschke & Muhlbauer, 2010). The current findings further support the role of melatonin in glucose metabolism beyond that of an increased diabetes risk as previously described (McMullan et al., 2013; Sparso et al., 2009).

The current data highlight the complex relationship between melatonin, OSA and glucose metabolism. This could partially help explain the increased diabetes risk seen in those with OSA as well as poorer glycemic control in T2D patients with more severe OSA as previously described (Reutrakul & Van Cauter E., 2014). It also raises the question whether melatonin supplement might be beneficial in T2D patients OSA. Melatonin supplementation in diabetes patients deserves a careful consideration. A study revealed that 2 months of melatonin supplementation in 30 patients with metabolic syndrome resulted in a significant decrease in blood pressure and LDL cholesterol, along with reduction in oxidative stress markers but glucose levels did not change (Kozirog et al., 2011). Additionally, melatonin can modulate sleep which is often disturbed in patients with diabetes. In a randomized, double-blinded, crossover study involving 36 T2D patients with insomnia, prolonged-release melatonin administration significantly improved sleep efficiency at 3 weeks but without changes in glucose levels (Garfinkel et al., 2011). However, HbA1c improved significantly at 5 months during the open-labeled phase, but this was not predicted by sleep improvements. This is in contrast to a more recent study in which the effects of acute melatonin administration on glucose metabolism in 21 healthy volunteers were investigated (Rubio-Sastre et al., 2014). Immediate-release melatonin administration both in the morning and evening resulted in an increase in the incremental area under the curve and maximum concentration of plasma glucose following an oral glucose tolerance test, as compared with placebo. Therefore, even though melatonin appears to be beneficial in some animal experiments, more data are needed in humans before this can be widely applied. Whether different metabolic effects exist between short and prolonged release melatonin formulations, different doses of melatonin, time of administration (day or night) or differences in study population (diabetes vs. non-diabetes) remain to be investigated.

In addition, melatonin has been shown to exert anti-inflammatory and anti-oxidant effects, mediated through a variety of cellular pathways (Mauriz et al., 2013). In animal models of intermittent hypoxia, melatonin ameliorated damages related to intermittent hypoxia, including endothelial dysfunction and hypertension (Hung et al., 2013), and insulin resistance (Bertuglia & Reiter, 2009). While CPAP is the gold standard treatment of OSA (Somers et al., 2008), adherence could be a challenge. Whether melatonin supplementation could help decrease OSA-associated inflammation and improve clinical outcomes, especially in CPAP-intolerant patients, remains a subject of future research.

Our findings of lower nocturnal urinary aMT6s levels in those with retinopathy are consistent with previous data (Chen et al., 2014; Hikichi et al., 2011). This is possibly due to impaired retinal light perception by intrinsically photosensitive ganglion cell. While poor glycemic control can lead to retinopathy, we hypothesize that once retinopathy occurs it may adversely affect glycemic control through a reduction of melatonin synthesis. Our analysis revealed for the first time that the presence of retinopathy was associated with decrease nocturnal aMT6s level, and an increase in HbA1c by 1.3% of its original value. This suggested a possibility of a vicious cycle between poor glycemic control and retinopathy, and will require further investigation in a larger study.

The strengths of our study are that many confounding factors known to affect melatonin secretion were considered, and that it was conducted exclusively in T2D. However, there are several limitations. The causal relationship could not be established given the cross-sectional nature of the study without manipulating one of the variables such as exogenous administration of melatonin. The number of participants was small, which yielded powers of test of 56.4% and 63.6% for path lnAHI → nocturnal urinary aMT6s and nocturnal urinary aMT6s → lnHbA1c, respectively. We did not find the relationship between sleep duration or quality with nocturnal urinary aMT6s as previously described (Saksvik-Lehouillier et al., 2015) which could be due to the differences in studied populations, the number of the participants (which limited the power of test) or that the average sleep duration of 6 h in our study was relatively short. In addition, OSA was not diagnosed using the gold-standard polysomnography. Lastly, the relationship between nocturnal urinary aMT6s, glycemic control and OSA severity may differ between insulin and non-insulin users, and male and female. Subgroup analyses would be informative but could not be performed due to small number of participants in each group in the current study. This should be addressed in a future larger study.

In summary, OSA severity was associated with lowering of nocturnal urinary aMT6s in T2D patients. Lower nocturnal urinary aMT6s level was associated with higher HbA1c level and served as a mediator between OSA severity and glycemic control. Whether melatonin supplementation may be beneficial in these patients remains to be investigated.

Declaration Of Interest

S. Reutrakul received speaker honoraria from Sanofi Aventis, Medtronic, Novo Nordisk and Astra Zeneca; research equipment support from ResMed, Thailand; and a research grant from Merck Sharp and Dohme. There is no financial interest related to the subject matter of this manuscript. Rest of the authors reported no conflicts of interest.

Acknowledgements

We thank Dr. Eve Van Cauter, Ph.D., Section of Adult and Pediatric Endocrinology, Diabetes, and Metabolism, and Sleep, Metabolism and Health Center, Department of Medicine, The University of Chicago, Chicago, IL, USA, for her guidance and critical input to the manuscript; and Dr. Piyarat Govitrapong, Mahidol University, Bangkok, Thailand, for her assistance in reviewing the manuscript.

Funding

This study received grant support from Mahidol University, Bangkok, Thailand.

Additional information

Funding

This study received grant support from Mahidol University.

Notes on contributors

Sirimon Reutrakul

S. Reutrakul conceptualized the study, researched and analyzed the data, wrote manuscript, contributed to discussion, reviewed/edited manuscript and is the guarantor of this work and, as such, had full access to the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. N. Siwasaranond, H. Nimitphong, S. Saetung, L. Chailurkit researched data and reviewed/edited manuscript; N. Chirakalwasan contributed to discussion and reviewed/edited manuscript; K. Srijaruskul researched data; B. Ongphiphadhanakul analyzed data, contributed to discussion and reviewed/edited manuscript. A. Thakkinstian analyzed data, wrote manuscript, contributed to discussion and reviewed/edited manuscript.