Transcutaneous carbon dioxide profile during sleep reveals metabolic risk factors in post-menopausal females

There is increasing evidence to suggest that sleep disorders and cardiovascular diseases are linked. Aging and menopause increase the risk of sleep-disordered breathing (SDB) and poor sleep quality, and each of these are also associated with metabolic disorders 1–3. Although the mechanisms of interaction between sleep disorders and cardiovascular diseases are not fully understood, an autonomic nervous system imbalance (increased sympathetic and decreased parasympathetic activity) and endothelial inflammation are likely to be involved.

The transcutaneous carbon dioxide (TcCO₂) sensor has been developed for noninvasive estimation of the partial pressure of arterial carbon dioxide. The method has not gained wide acceptance because the correlations between TcCO₂ and arterial carbon dioxide tension are affected by haemodynamic events, such as vasoconstriction and vasodilatation 4, 5. However, measurement of local carbon dioxide is of special interest, since it is an important regulator of vascular nitric oxide production 6. Using our algorithms to analyse the nocturnal TcCO₂ plateaus and sudden TcCO₂ descents, we have previously been able to predict nitric oxide mediated vasodilatation in pre-menopausal females 7 and metabolic status in patients with suspected sleep apnoea 8.

In females, the risk of metabolic syndrome (defined as insulin resistance, abdominal obesity, dyslipidemia and elevated blood pressure), increases during the menopausal transition 9, 10. The early detection of developing metabolic abnormalities in this population would be key to preventing or reducing the effects of metabolic syndrome and its complications. Therefore, we performed sleep studies, including all-night TcCO₂ recordings, in a so far healthy population of post-menopausal females with two specific aims. First, we wanted to evaluate further the performance of our TcCO₂ analysis in predicting metabolic variables such as glycosylated haemoglobin A1C (GHbA1C), blood pressure (BP), and plasma lipoprotein cholesterols and triglycerides in a group of individuals whose pre-test probability of metabolic syndrome was lower than that of the patients with suspected sleep apnoea in our previous study. Our second aim was to screen a number of nocturnal TcCO₂ features, along with other sleep parameters, for their potential to predict metabolic abnormalities and increased cardiovascular risk in these subjects.

METHODS

Processing of the TcCO₂ signal

Details of the TcCO₂ signal processing are provided in the supplementary data, and are only described briefly herein. First, obvious artefacts at the beginning or at the end of the TcCO₂ recordings were manually removed. The artefacts in the middle of the recordings were replaced with constant line segments. The median overnight TcCO₂ levels were calculated for each sleep stage (S1, S2, SWS, REM sleep and wakefulness), as well as the percentage of time the signal stayed above 7 kPa. Special attention was paid to abrupt TcCO₂ descents, referred to as pit patterns. Each pit pattern was characterised by its descent (amplitude, duration and slope). The highest TcCO₂ plateau (maximal plateau) was defined visually from each curve. An example of a TcCO₂ curve with a pit pattern and maximal plateau is shown in figure 1⇓.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1—

Representative overnight recording of the transcutaneous carbon dioxide (TcCO₂) as function of sleep stages and events. Episodes of TcCO₂ pit patterns, apnoea, hypopnoea, desaturation and arousals per hour are illustrated below the hypnogram. The TcCO₂ signal insert corresponds to a single pit pattern. The pit patterns were analysed for amplitude, duration and slope. The TcCO₂ level during evening wakefulness and the maximal TcCO₂ plateau are indicated. - - - -: the 7 kPa threshold. Other features extracted from the TcCO₂ signal include median TcCO₂ levels in the various sleep stages. REM: rapid eye movement.

RESULTS

Each of the 22 females participated in the overnight measurements (fig. 1⇑). The demographic data of the study group is presented in table 1⇓. The females were mildly overweight (mean BMI 25.3 kg·m⁻²). Two of the females had a BMI >30 kg·m⁻². Seven females had a systolic BP (SBP) >140 mmHg and six females had a diastolic BP (DBP) >90 mmHg. One subject had a GHbA1C of 6.1 %, and 17 subjects had total cholesterol >5 mmol·L⁻¹. The inter-relationships between the metabolic variables used in the linear regression models are presented in table 2⇓. The FEV₁ ranged from 77% to 146% of predicted values and the median FEV₁ was 95%. In all of the females, FSH was in the post-menopausal levels ranging from 31 to 140 IU·L⁻¹. The subjects included four habitual smokers, two occasional smokers and four habitual snorers (snored at least three nights per week). None of the subjects had chronic obstructive pulmonary disease or asthma. Two females regularly used acetylsalicylic acid, one used cetirizine for allergic symptoms and gastric mucoprotective drugs and one female was on citalopram.

The mean±sd values of the sleep architecture measures from the sleep study are shown in table 3⇓. The S_a,O2 and TcCO₂ measurements are presented in table 4⇓. None of the subjects had an ODI₄ of >5 events·h⁻¹. Two subjects had an AHI of >5 events·h⁻¹ (7 and 8 events·h⁻¹, respectively). The nocturnal frequency of pit patterns was computed individually for both REM and NREM sleep states, with the pit index being considerably higher in REM sleep (p<0.001). In the final multivariate results, age was not taken as a predictor because it did not show a significant correlation with any of the metabolic variables (correlations ranged from 0.233 to 0.303 and the p-values from 0.171 to 0.947).

DISCUSSION

In our study population of generally healthy 55-yr-old post-menopausal females, nocturnal TcCO₂ features were the most important predictors of GHbA1C, BP and cholesterol levels (table 5⇑). A longer sleep period was linked with a lower GHbA1C, and fragmented sleep with lower HDL cholesterol, as suggested by previous studies 17, 18. Mean and nadir S_a,O2, ODI₄, AHI, BMI values and waist circumference were worse predictors of metabolic variables in this population. Subjects were all generally healthy, although marginally overweight, and with BP and waist circumference slightly exceeding the International Diabetes Federation reference values 19 (table 1⇑). Despite this, the overnight TcCO₂ features were systematically associated with the metabolic variables. Our results provide further support to our earlier findings that, irrespective of the study population, the nocturnal TcCO₂ profile contains risk factor information on metabolic diseases 8. It is possible that increasing TcCO₂ levels during sleep indicate decent vasodilatation capacity in the silence of sympathetic drive, whereas sudden TcCO₂ decreases (pit patterns) indicate surges of sympathetic activity that cause exaggerated vasoconstrictive responses when endothelial dysfunction is present. The feasibility of interaction between the local CO₂ events and endothelial dysfunction is further strengthened by the observation that the CO₂ mediated autoregulatory vasodilatation is mediated through nitric oxide 6. The nocturnal TcCO₂ features have a high predictive power on daytime endothelial function tests 7, suggesting that local CO₂ is a major controller of local nitric oxide production. Further studies are needed to establish to what extent nocturnal TcCO₂ events display local nitric oxide production.

Menopause increases the risk of SDB 20, 21, which in turn is also known to be a risk factor for metabolic syndrome and cardiovascular disorders 2, 22. Our study population was a group of clinically healthy females with age and post-menopausal status as risk factors, which gave us a unique setting to search for signs of emerging metabolic syndrome. CO₂, which is the final metabolic end product, played a major role in our results. A TcCO₂ sensor measures the CO₂ that diffuses through the skin, and its measurements are affected by central respiratory drive, peripheral vascular perfusion and local tissue metabolism 12, 23. Conventional severity indexes of sleep apnoea (AHI, ODI₃, ODI₄ or S_a,O2) were not found to be important in this population with relatively little SDB. This suggests that the TcCO₂ features predict metabolic variables independently of hypoxaemia or SDB.

The high overnight levels of TcCO₂ were one of the new key predictors for protective metabolic variables in our earlier study in patients with sleep apnoea 8. This association was also found in the present study, despite an essentially different study population (healthy post-menopausal females). The longer the subjects maintained their TcCO₂ at >7 kPa, the lower their levels of GHbA1C and triglycerides. Furthermore, the visually detected maximal plateau of the TcCO₂ curve associated positively with the HDL/total cholesterol ratio. These results are encouraging, yet they should be interpreted with caution, since we have no explicit data about “the normal” TcCO₂ ranges during sleep. It is probable that nocturnal TcCO₂ variables differ between sexes and change with increasing age, after the menopause 11 or during oestrogen therapy 24. However, the findings are in line with the earlier TcCO₂ profile results and, in particular, with the results from our previous work that the proportion of high TcCO₂ levels measured during sleep were one of the most important features for classifying insulin resistance 8.

Obstructive sleep apnoea (OSA) and snoring are associated with insulin resistance and an impaired lipid profile 3, 25, 26. OSA may decrease the arterial CO₂ tension due to repetitive arousals and hyperventilation following each apnoea. Therefore, it is possible that people with even mild sleep disturbances cannot achieve as high TcCO₂ levels as coeval healthy people because episodes of apnoea, snoring or frequent arousals interrupt sleep. This is in line with the finding that awakening is usually followed by a notable descent in the TcCO₂ tension (fig. 1⇑), while falling asleep is typically related to a rise in TcCO₂ 11.

In the present study, we classified the TcCO₂ features according to the sleep stages and found that TcCO₂ pit patterns occur significantly more often in REM sleep and wakefulness than in other sleep stages (fig. 1⇑). Normally sympathetic activity dominates during wakefulness and appears as bursts during REM sleep. NREM sleep is characterised by parasympathetic dominance. Sympathovagal imbalance is common in metabolic syndrome, but its definitive causative role has not been demonstrated 27. Pit patterns may result from sudden bursts of sympathetic activity that produce peripheral vasoconstriction. These bursts may appear more consistently during sympathetic dominance. Subjects with a high pit index had a lower HDL cholesterol concentration, confirming our previous findings 8. Low HDL cholesterol has previously been linked with a greater frequency of arousals 17. Likewise, in our study, sleep fragmentation was the other predictor of HDL cholesterol. As arousals and awakenings increase sympathetic activation during sleep, it is possible that they also produce the pit patterns. Spiegel et al. 28 have shown that glucose metabolism is impaired with increased sympathetic tone, which has been induced by partial short-term sleep deprivation. If TcCO₂ reflects sympathovagal balance, then during parasympathetic dominance when periferic blood vessels are dilated, TcCO₂ levels are high.

Another feature of the TcCO₂ patterns is the amplitude and sharpness of the pits. The fast and deep descents appear to be associated with low LDL and total cholesterol levels. In addition, high amplitude of pit pattern was the only predictor of low DBP. This further supports the idea of TcCO₂ as a reflector of sympathovagal balance. By monitoring the transient TcCO₂ events against the prevailing parasympathetic tone, the bursts of sympathetic nervous activity can be distinguished more clearly. The dominance of sympathetic activity may diminish the amplitudes of the pit patterns.

The only predictor for SBP was the evening wakefulness level of TcCO₂. As CO₂ is a known vasodilatator 29, this may explain why subjects with a high TcCO₂ had a lower SBP. The possible protective effect of higher TcCO₂ is also in line with the findings in females with SDB. Those with a predominantly partial upper airway obstruction during sleep (flow limitation) combined with increased TcCO₂ levels had less hypertension than in patients with OSA 30. However, high TcCO₂ did not associate with DBP. Anderson et al. 31 have previously demonstrated that high end-tidal CO₂ predicts high SBP in females. End-tidal CO₂ measures the CO₂ concentration in the alveoli and is strongly affected by changes in ventilation, whereas TcCO₂ is also affected by tissue metabolism and local vasodynamics. Hence, end-tidal CO₂ and TcCO₂ are likely to measure different phenomena. However, the results of Anderson et al. 31 show that breathing is an important contributor in the development of hypertension. The BP of OSA patients is higher than in healthy controls 32 and nocturnal hypoxia elevates BP 33. Generally, poor sleep seems to be an important risk factor for hypertension because sleep fragmentation, arousals and short self-reported sleep duration are associated with high blood pressure 17, 34, 35. These variables did not predict BP in our study. This may be due to the relatively small and healthy study population.

The risk of metabolic syndrome increases around the time of menopause 9, 10, 36. Therefore, one would expect to find some variation in our study population, even if all of the females were generally healthy. Central obesity is a known risk factor for insulin resistance 37. However, in our study, it did not turn out to be an important contributor, probably because most of our subjects were rather lean (table 1⇑). Neither the duration nor the proportion of SWS turned out to be important predictors in our study population, even though van Tasali et al. 1 recently showed that a short SWS duration is associated with an increased risk of diabetes. This may be because of our healthy sample, as only one female had a GHbA1C >6%. In addition to the SWS duration, self-reported short sleep has been linked with obesity and diabetes 18, 38. These studies are in line with our results, that a long sleep period (assuming that the length of a subject's sleep period in a sleep laboratory reflects their normal sleep period) was the most important predictor of lower GHbA1C. In addition, sleep fragmentation was a predictor of decreased HDL cholesterol. Recently, Ekstedt et al. 17 showed that the number of arousals predicted lower HDL cholesterol. Together, these results suggest the importance of adequate length and quality of sleep in the prevention of metabolic disorders. Short and fragmented sleep may disturb the TcCO₂ signal, which seems to be sensitive to subtle nocturnal changes. This may explain why in our results the TcCO₂ features play such an important role. However, more studies are still needed to confirm the results.

Our study has some potential confounders and limitations. The cross-sectional study design does not allow us to confirm whether CO₂ plays a causative role, or whether the TcCO₂ profile is a marker of some other underlying pathophysiological process. Moreover, the number of subjects was relatively small and heterogeneous in terms of hormone replacement therapy. Although the subjects were all around the same age, the time from menopause varied (table 1⇑). The time from menopause was not included in linear regression models as only 13 (59%) females remembered their exact time of menopause. This, together with the cross-sectional study design, makes it impossible to separate the influence of age and menopause. In addition, oestrogen deficiency affects metabolism 10. Oestrogen usage was not selected as an exclusion criterion because our main goal was not to study the effects of oestrogen but to find the predictors for a wide range of metabolic variables. Moreover, grouping the subjects based on the oestrogen usage did not reveal differences in any of the measurements (unpaired t-test p>0.1). Some potential pitfalls in the subject selection could be criticised as well. The subjects were recruited through newspaper announcements, calling healthy post-menopausal females for a sleep and cardiovascular study. Some members of this group of “healthy” subjects may have been compelled to participate in such a study due to subclinical hidden sleep problems or cardiovascular family risk factors. However, none of the subjects regularly used hypnotic drugs.

Support statement

This work was supported by grants from the Finnish Anti-Tuberculosis Association Foundation, Tampere Tuberculosis Foundation, Aarne and Aili Turunen Foundation, Väinö and Laina Kivi Foundation, Academy of Finland, Paulo Foundation, the Jalmari and Rauha Ahokas Foundation, Orion-Farmos Research Foundation, the Research Foundation for Respiratory Diseases and the governmental EVO-grant of the Turku University Hospital (All Finland).

Statement of interest

None declared.