Cerebral desaturation in heart failure: Potential prognostic value and physiologic basis

Cerebral tissue oxygen saturation (SctO2) reflects cerebral perfusion and tissue oxygen consumption, which decline in some patients with heart failure with reduced ejection fraction (HFrEF) or stroke, especially during exercise. Its physiologic basis and clinical significance remain unclear. We aimed to investigate the association of SctO2 with oxygen transport physiology and known prognostic factors during both rest and exercise in patients with HFrEF or stroke. Thirty-four HFrEF patients, 26 stroke patients, and 17 healthy controls performed an incremental cardiopulmonary exercise test using a bicycle ergometer. Integrated near-infrared spectroscopy and automatic gas analysis were used to measure cerebral tissue oxygenation and cardiac and ventilatory parameters. We found that SctO2 (rest; peak) were significantly lower in the HFrEF (66.3±13.3%; 63.4±13.8%,) than in the stroke (72.1±4.2%; 72.7±4.5%) and control (73.1±2.8%; 72±3.2%) groups. In the HFrEF group, SctO2 at rest (SctO2rest) and peak SctO2 (SctO2peak) were linearly correlated with brain natriuretic peptide (BNP), peak oxygen consumption (V˙O2peak), and oxygen uptake efficiency slope (r between -0.561 and 0.677, p < 0.001). Stepwise linear regression showed that SctO2rest was determined by partial pressure of end-tidal carbon dioxide at rest (PETCO2rest), hemoglobin, and mean arterial pressure at rest (MAPrest) (adjusted R = 0.681, p < 0.05), while SctO2peak was mainly affected by peak carbon dioxide production (V˙CO2peak) (adjusted R = 0.653, p < 0.05) in patients with HFrEF. In conclusion, the study delineates the relationship of cerebral saturation and parameters associated with oxygen delivery. Moreover, SctO2peak and SctO2rest are correlated with some well-recognized prognostic factors in HFrEF, suggesting its potential prognostic value.

Introduction Cerebral desaturation may occur in patients with heart failure with reduced ejection fraction (HFrEF) or stroke [1,2]. The former is due to insufficient cardiac output and dead space ventilation. The latter could be due to disruption of cerebral blood flow (CBF) from vascular abnormalities and neurologic deficit-related respiratory muscle weakness [3][4][5]. The brain utilizes 20% of the total oxygen consumption at rest [6]. Cerebral oxygenation is determined by arterial oxygen concentration, CBF, and cerebral tissue oxygen consumption, which represent oxygen supply and demand [7]. Insufficient hemoglobin (Hb) concentration, oxygen desaturation, and disruption of cerebral perfusion may result in decreased oxygen supply and thus, cerebral desaturation.
There is evidence that cerebral tissue deoxygenation may limit exercise performance in healthy people. Nielsen et al. demonstrated that when compared with 20% oxygen, 30% inspired oxygen concentration increased exercise performance by decreasing the extent of cerebral desaturation during strenuous exercise; meanwhile, muscle oxygenation was not changed [8]. This study concluded that an elevated inspiratory oxygen fraction increases exercise performance by maintaining cerebral oxygenation rather than through any effect on the working muscles [8]. It is likely that diminished cerebral perfusion may also limit exercise performance in patients with HFrEF. Our previous study revealed that cerebral hypoperfusion is associated with hyperventilation and diminished aerobic capacity in patients with HFrEF [9]. Additionally, frontal cortex dysfunction may impair executive function, resulting in central inhibition [1]. Conversely, poor physical fitness in patients with HFrEF was found to be associated with pathologic change of brain structure, including decreased grey matter volume and cortical thickness, while patients with good physical fitness preserve cerebral structure [10]. Therefore, it is a reasonable conjecture that cerebral tissue oxygen saturation (SctO 2 ) in the frontal lobe has a close relationship with exercise performance in patients with HFrEF or stroke, which has not been explored yet.
To go further, peak oxygen consumption ( _ VO 2peak ) is a well-recognized significant prognostic factor in patients with HFrEF [11]. If SctO 2 is related to exercise capacity, its potential prognostic value deserves investigation. In this study, the link between known prognostic factors and SctO 2 during both rest and peak exercise were explored.
Methodologically, SctO 2 was measured at the bilateral frontal region during the maximal incremental cardiopulmonary exercise test (CPET), in patients with systolic HFrEF, uni-hemisphere stroke, and healthy participants. Plasma level of brain natriuretic peptide (BNP) and complete blood cell count were collected in HFrEF and stroke groups. Physical parameters that determine SctO 2 during rest and exercise were investigated. We hypothesized that SctO 2 at rest and exercise have a close relationship with _ VO 2peak and are correlated with some known prognostic factors.

Participants
This is a case-controlled cross-sectional design. Thirty-four stable HFrEF patients, 26 firsttime uni-hemisphere ischemic stroke patients with hemiparesis, and 17 healthy controls were enrolled by convenience sampling in Linkou Chang Gung Memorial Hospital, a tertiary medical center. The enrolled stroke patients were at least 3 months post onset. All patients with HFrEF had a left ventricular ejection fraction 40% and a disease duration ! 3 months. Healthy controls were recruited by convenience sampling. Most of them were colleagues or caregivers of the hospitalized patients. The exclusion criteria included contraindications to stress exercise testing [12], or inability to ride a bike due to musculoskeletal problems or neurologic deficits including muscle strength less than 4 based on Medical Research Council scale. Patients with HFrEF with moderate to severe carotid artery stenosis or diseases that might affect ventilation, such as chronic obstructive pulmonary disease and pulmonary hypertension, were also excluded. In the stroke group, those with ventilation disorders were also excluded. In the healthy control group, those with any cardiovascular or respiratory diseases in the medical record were excluded. Written informed consent was obtained from every subject before the experiment. The study protocol was performed in accordance with the Declaration of Helsinki and approved by the ethics committee in Chang Gung Memorial Hospital, Linkou, Taiwan. Venous blood was sampled in the morning to determine BNP and complete blood cell count in the HFrEF and stroke groups.

Cardiopulmonary exercise testing
All participants received symptom-limited incremental CPET with upright position on a calibrated bicycle ergometer (Ergoselect 150P, Germany). CPET started with 2 min of rest and 1 min warm-up at a work rate of 10 W followed by a ramp increase of 10 W/min until exhaustion. Heart rate was calculated from the R-R interval recorded on a 12-lead electrocardiogram. Blood pressure was measured automatically every 2 min (Tango, SunTech Medical, UK). Gas analysis was measured breath by breath using a microprocessor-controlled system (MasterScreen CPX, Cardinal-health Germany). The _ VO 2peak was achieved as the examinee failed to keep the cadence above 50/min despite strong encouragement. Mean arterial pressure (MAP) was calculated by the following equation: MAP = [(2 x diastolic) + systolic] / 3. Arterial oxygen saturation was measured by finger pulse oximetry (model 9500, Nonin Onyx, Plymouth, Minnesota).

Ventilatory efficiency
Ventilation and carbon dioxide consumption ( _ VCO 2 ) responses, acquired from the initiation of exercise to peak values, were used to calculate the minute ventilation (V E )-_ VCO 2 slope using least-squares linear regression (y = mx + b, m = slope), where a more horizontal slope suggests better ventilation efficiency [13]. The oxygen uptake efficiency slope (OUES) was derived from the slope of a natural logarithm plot of V E vs. _ VO 2 . The OUES is an estimation of the efficiency of ventilation with respect to _ VO 2 , with greater slope indicating higher oxygen uptake efficiency [14].

Cerebral oximetry
SctO 2 was monitored using the FORE-SIGHT system (CAS Medical Systems, Inc., Branford, CT). The scanning frequency was 100 Hz and was averaged in one second for the value at rest and peak exercise. The SctO 2 value at rest was picked up when it reached a stable baseline during two-minute pretest before the exercise started as sitting upright on the cycle ergometry. The FORE-SIGHT device is a spatially resolved near-infrared cerebral oximeter that measures the absolute value of SctO 2 . Four continuous near-infrared (bandwidth < 1 nm) wave-length (690 nm, 780 nm, 805 nm, and 850 nm) of lights penetrated the brain from the prism of sensors (3.625" x 1.5"). Reflected light was then sampled by detectors on the sensor. Four wavelengths were employed to enhance measurement accuracy of oxyhemoglobin and deoxyhemoglobin levels by compensating for wavelength-dependent scattering losses and by eliminating interference from other background light absorbers (such as skin pigmentation and fluid) [15,16]. Both resting and peak SctO 2 were measured in the upright sitting posture on the stationary bike. Sensors were placed on the forehead bilaterally. The average value from bilateral forehead was analyzed along with the data from each side. They were used to represent tissue oxygenation at the frontal lobe. The validity has been established and approved by FDA in monitoring during cardiovascular surgery [16]. It has also been applied to measure cerebral oxygenation during exercise in patients with HFrEF [9, [17][18][19].

Echocardiography
Echocardiography was performed in HFrEF and stroke groups by a cardiologist using Vivid E90 (GE Healthcare, Milwaukee, WI) equipped with a 2.5-MHz transducer. Left ventricular ejection fraction, left ventricular end-diastolic, and end-systolic volumes were quantitated by M-mode and two-dimensional methods [20].

Statistical analysis
Statistical Package for the Social Sciences (SPSS) version 22.0 (SPSS, Inc., Chicago, IL, USA) was used to analyze the data. Continuous data were expressed as mean ± standard deviation. Three-group comparison was performed by ANOVA with Scheffe post hoc test in the continuous variables, and Chi-Squared test in the categorical ones. SctO 2 during rest and peak exercise in the three groups were compared by two-way repeated measure ANOVA with Scheffe post hoc test. Chi-Squared test was also applied to analyze the asymmetric pattern of SctO 2 . Pearson correlation was used to analyze SctO 2 versus other physical parameters, including Hb, BNP, and all the cardio-respiratory parameters listed in Table 1. The four variables with the highest correlation coefficients (all p < 0.05) were initially put into the linear stepwise univariate regression model to select the major variables associated with SctO 2 . Delta R-squared value was calculated to determine the change in R-squared value as adding another variable into the model. All probability values were two-tailed and the significance threshold was set at 0.05.
Oxygen saturation of cerebral tissue (rest; peak) were significantly lower in the HFrEF (66.3 ±13.3%; 63.4±13.8%) than that in the stroke (72.1±4.2%; 72.7±4.5%) and control (73.1±2.8%; 72±3.2%) groups as revealed by two-way repeated measure ANOVA, while those in the stroke group were close to the healthy control. Moreover, in the HFrEF group, SctO 2 decreased significantly at peak exercise, a phenomenon not observed in the stroke or healthy control groups (Fig 1, Table 3). Additionally, all the arterial oxygen saturation among three groups were 98~100% during both rest and peak exercise.
Linear stepwise univariate regression analysis was employed to identify the major physiologic determinants of SctO 2rest and SctO 2peak in patients with HFrEF. _ VO 2peak (r = 0.602), MAP rest (r = 0.597), partial pressure of end-tidal carbon dioxide at rest (P ET CO 2rest ) (r = 0.586), Hb (r = 0.51), and were entered into the model of SctO 2rest . OUES (r = 0.677), _ VO 2peak (r = 0.66), peak carbon dioxide production ( _ VCO 2peak ) (r = 0.651), and BNP (r = -0.561) were entered into the model of SctO 2peak (S1 Table). It revealed that SctO 2rest was determined by P ET CO 2rest , Hb, and MAP rest (R = 0.681, p < 0.05), while SctO 2peak was by VCO 2peak (R = 0.653, p < 0.05) ( Tables 4 and 5). It is worth mentioning that Hb is also significantly correlated with SctO 2peak (r = 0.39). However, it does not increase the predictive power significantly when it was entered after _ VCO 2peak (p-value of ß was 0.052). The experimental finding of bilateral comparison of SctO 2 in the HFrEF vs. control groups and its related discussion are appended in the supporting information (S1 Supporting information).

Discussion
To the best of our knowledge, the current investigation is the first to determine the potential prognostic value of SctO 2 and its possible physiologic basis in patients with HFrEF. Significant experimental findings were as follows: I. SctO 2 of the HFrEF group were significantly decreased compared with healthy control and stroke groups. II. Patients with HFrEF not only had diminished cerebral oxygenation at rest, but also showed further reduced oxygenation during incremental exercise testing, which was not observed in the stroke and healthy groups. III. Importantly, both SctO 2rest and SctO 2peak of the HFrEF group (especially SctO 2peak ) were correlated with _ VO 2peak , BNP, and OUES in moderate degree, which are well-established prognostic markers. IV. In the HFrEF cohort, linear regression analysis showed that SctO 2rest was primarily determined by P ET CO 2rest , Hb and MAP rest , while SctO 2peak was affected primarily by _ VCO 2peak . V. The numerical value of SctO 2 did not differ between the stroke and healthy control groups. Its association with other investigated biomarkers was not obvious in these two groups either.

Low cerebral oxygenation in patients with HFrEF
CBF is reduced in patients with mild to severe HFrEF even in resting state, as confirmed through 133 XE radionuclide angiography injection method, and transcranial and extracranial Doppler ultrasonography [21][22][23]. During dynamic exercise, cerebral perfusion can be sacrificed to active skeletal muscle in patients with HFrEF. Hellstrom et al. demonstrated that healthy subjects had a 20% increase in mean middle cerebral artery blood velocity (MCA V mean ) when performing one-legged exercise and this increase was maintained when performing two-legged exercise. However, in patients with HFrEF, MCA V mean was not increased during one-legged exercise and was significantly decreased during two-legged exercise [24]. For stroke patients, Robertson et al. reported that regional CBF was reduced after low-intensity exercise but increased after moderate-intensity exercise [25]. Similarly, the present investigation found that SctO 2 was lower in patients with HFrEF during resting state and decreased significantly from rest to peak exercise only in the HFrEF group. It is very possible that cerebral autoregulation fails to compensate CBF in face of low systemic BP commonly seen in HFrEF patients, leading to reduced SctO 2 [26] . The reason why our data did not demonstrate cerebral oxygenation reduction in the stroke group could be that only patients with mild severity of stroke were recruited. Their muscle strength in the hemiparetic limbs were 4 or 5-in manual muscle testing. Thus, the data pattern was similar in the stroke and healthy control groups. Meanwhile, the potential influence of medications should be considered. Calcium channel blockers (CCB) elevate CBF [27], while beta-blockers attenuate the increase in cardiac output, CBF and cerebral oxygenation during exercise [28,29]. In our study, significantly more patients took CCB in the stroke group, while more patients took beta-blockers in the HFrEF group. Although these medications may augment cerebral deoxygenation in HFrEF patients, the prescription was in line with the current treatment guidelines and reflects the real-world situation [30,31].

Potential prognostic value of SctO 2 in HFrEF
Our data showed that in patients with HFrEF, both SctO 2rest and SctO 2peak (especially SctO 2peak ) were correlated with _ VO 2peak , BNP, and OUES in moderate degree. BNP has been recognized as a prognostic and diagnostic factor for HFrEF and is used to assess the severity of HFrEF [32,33]. _ VO 2peak provides objective assessment for cardio-pulmonary fitness and is one Values are means ± SD; HFrEF, heart failure with reduced ejection fraction; Peak, peak exercise L't, Left; R't, Right; SctO 2 , cerebral tissue oxygen saturation †: p < 0.05, HFrEF vs. control Ã : p < 0.05, HFrEF vs. stroke; repeated measure ANOVA with Scheffe post hoc test https://doi.org/10.1371/journal.pone.0196299.t003 Cerebral desaturation in heart failure of the most significant short-term and long-term prognostic factors for patients with HFrEF [34,35]. OUES measures oxygen uptake efficiency in relation to ventilation during an incremental exercise test and has been shown to be a prognostic marker, which is even more predictive than _ VO 2peak in some of the HFrEF studies [14,36]. Koike et al. found that in patients with coronary artery disease, the change of cerebral oxyhemoglobin (O 2 Hb) from rest to peak exercise is prognostic for cardiovascular morbidity [37]. However, our results showed that the difference between SctO 2rest and SctO 2peak were not associated with the known prognostic factors investigated. Instead, it was the absolute value during peak or rest that might be related to prognosis. An explanation is that O 2 Hb and SctO 2 are essentially different. Decrease of O 2 Hb from rest to peak exercise may result from the reduction of CBF or increased oxygen consumption of the tissue. However, SctO 2 is decreased only when O 2 Hb has a relatively larger reduction than deoxyhemoglobin [24]. Probably, it is the reduction of CBF that is primarily associated with prognosis.

Determinants of SctO 2rest in HFrEF: P ET CO 2rest , Hb, and MAP rest
The linear stepwise regression model indicated that SctO 2rest of the HFrEF group was primarily determined by P ET CO 2rest , Hb, and MAP rest . The explanatory power was 46% (R 2 = 0.464) in this model. P ET CO 2 is very close to partial pressure of arterial carbon dioxide (PaCO 2 ) during rest since those with ventilation problems were excluded in the HFrEF group [38][39][40]. Therefore, our experimental findings of P ET CO 2 during rest may be interpreted as PaCO 2 . PaCO 2 was previously demonstrated to be positively linearly correlated to CBF from 15 to 60 mmHg [41]. Our data of P ET CO 2 in the HFrEF group and the control group were 30.5±7.7 and 37±2.2 mmHg, respectively. Therefore, the reduced P ET CO 2 in the HFrEF group may affect SctO 2rest due to decreased CBF. In addition, low P ET CO 2 is also a significant predictor of cardiac-related events in patients with HFrEF [42].
Cerebral oxygenation is also influenced by arterial oxygen concentration [43]. Anemia leads to a decreased oxygen supply to the cerebral tissue. It is also associated with an increased mortality rate in patients with HFrEF [44,45]. In healthy subjects, the disadvantage of anemia is compensated by increased cardiac output, plasma volume, reduced systemic vascular  resistance, and widened arteriovenous oxygen gradient. However, these compensatory mechanisms are impaired in patients with HFrEF [46,47]. MAP also influences cerebral oxygenation based on our results. Herholz et al. adopted 133 Xe clearance technique and demonstrated that the linear increase in CBF resulted not only from PaCO 2 but also MAP [41]. Rifai et al. also found a positive correlation between SctO 2 and MAP in patients with HFrEF at rest [48]. Higher MAP may be associated with higher cerebral perfusion [49]. As perfusion into the tissue vascular bed is increased, Hb density in the tissue increases and thereby, SctO 2 is less reduced. Accordingly, MAP rest , plays a role in determining CBF and thereby SctO 2rest . Meanwhile, MAP was inversely related to the total and cardiovascular mortality [50], which as well indicated potential prognostic value of SctO 2 . Nonetheless, MAP was not correlated with SctO 2peak . Previous investigations confirmed that CBF does not increase or even decline at peak exercise even though MAP rises in patients with HFrEF [9,51]. In other words, MAP does not reflect CBF during exercise in patients with HFrEF. In summary, P ET CO 2rest , Hb, and MAP rest are important determinants of SctO 2rest , as well as prognostic markers in patients with HFrEF.

Determinants of SctO 2peak in HFrEF: _ VCO 2peak
The stepwise linear regression showed that the most major determinant of SctO 2peak in the HFrEF group is _ VCO 2peak (R = 0.653; R 2 = 0.426). There are two explanations: first, _ VCO 2peak and _ VO 2peak have high collinearity. Therefore, low _ VCO 2peak suggests low oxygen delivery to the peripheral tissue, including brain and thus leads to decreased SctO 2 [43]. Second, patients with HFrEF with lower _ VCO 2peak (thus, lower _ VO 2peak ) tend to have lower PaCO 2 due to ventilation-perfusion mismatch [43], which results in cerebral vasoconstriction [52]. This could explain why SctO 2peak in the HFrEF group is primarily determined by _ VCO 2peak rather than _ VO 2peak .

SctO 2 in stroke
The numerical value of SctO 2 in the stroke and healthy control groups were not different based on the present data. Also, it was not associated with the investigated known prognostic factors. It can be explained by the mild severity in the included stroke patients. Their muscle strength in the hemiparetic limbs were 4 or 5-in manual muscle testing. In consideration of ergometer-riding being the modality of exercise testing, those with poorer muscle strength were not adequate to be included because maximal effort cannot be reached due to the neurologic deficits. In Table 3, the respiratory exchange ratio of the stroke group was not different from that of the control group (1.18±0.14 vs. 1.14±0.34), indicating that maximal exertion was nearly approached in the stroke group. The data showed that although patients with mild ischemic stroke were less fit (lower VO 2peak , HR peak and V Epeak ) than the control group, SctO 2 during rest or peak were not lower. Further study on stroke patients with higher severity and significant cerebrovascular stenosis is needed.

Limitations of NIRS cerebral oximetry
Some limitations concerning NIRS cerebral oximetry need to be taken into consideration. First, SctO 2 was measured without differentiating vascular bed as being arterial, capillary, or venous. Since it is estimated that more than 70% of the Hb in the brain is in venous bed, the measured SctO 2 may reflect larger proportion of venous saturation [53]. Second, extracranial contamination and melanin may absorb light and thus attenuate the signal, though all the participants are Asian. [54,55].

Study limitation
A longitudinal study is needed to confirm the prognostic value of SctO 2 . Moreover, P ET CO 2 rather than PaCO 2 was measured in the current study though patients with ventilation disorder were already excluded. However, in patients with advanced HFrEF, a certain degree of increased dead space ventilation may be present; thus, PaCO 2 might be slightly higher than P ET CO 2 , even at rest [43]. Additionally, lack of direct CO and CBF measurement and absence of echocardiographic data, Hb, and BNP in the control group limit precise interpretation and analysis. Also, it may influence the SctO 2 value in the stroke group to have the NIRS sensors placed on the fixed positions regardless of whether the infarction area was right underneath the sensor.

Conclusion
Cerebral oxygenation was reduced in patients with HFrEF compared with healthy controls. Moreover, cerebral oxygenation, especially at peak exercise, is correlated with _ VO 2peak , BNP, and OUES, which are well-recognized prognostic factors. Cerebral oxygenation during rest is determined mainly by P ET CO 2 , Hb, and MAP rest , while at peak exercise, is primarily affected by _ VCO 2peak in patients with HFrEF. These findings provided the physiologic basis of cerebral oxygenation and its potential prognostic value in patients with HFrEF, and may have clinical value in the future.
Supporting information S1 Table. Pearson correlation coefficients between cerebral tissue oxygen saturation and cardio-respiratory variables.
(DOCX) S1 Supporting information. Comparison of bilateral cerebral tissue oxygen saturation at peak exercise in HFrEF vs. control and stroke vs. control groups. (DOCX)