|Home | About | Journals | Submit | Contact Us | Français|
We tested hypothesis that cerebral deoxygenation near maximal exercise intensity is mediated by hyperventilation, via hypocapnia-induced reductions in cerebral blood flow, by utilizing canonical correlation analysis (CCA) to determine the relative influence of cardiopulmonary changes on cerebral oxygenation, as assessed by near infrared spectroscopy (NIRS). Twenty-three subjects performed incremental exercise tests under normoxic and hypoxic conditions. Changes in ventilation (E) were strongly correlated with end-tidal CO2 (PET CO2) and NIRS after the respiratory compensation point (RCP) (r2 >0.97). However, in contrast to our expectations, CBF velocity (CBFv) shared the least amount of variance with NIRS measurements (r2 < 0.56) and the reduction in CBFv was not accompanied by a reduction in cerebral blood volume. These results demonstrate that while cerebral deoxygenation was associated with hyperventilation, it was not solely explained by hypocapnia-induced reductions in CBFv. CCA revealed that a relative increase in the venous contribution to NIRS explained a larger amount of variation in cerebral oxygenation than reductions CBFv.
Reductions in regional cerebral oxygenation during high intensity aerobic exercise may limit performance, particularly at high altitude (Amann et al. 2007; Subudhi et al. 2007; Subudhi et al. 2008). The mechanisms responsible for reduced regional cerebral oxygenation are unknown, but have been theorized to stem from hypocapnia-induced cerebral vasoconstriction subsequent to hyperventilation (Bhambhani et al. 2007; Querido et al. 2007; Thomas et al. 2008). Support for this theory is based on a series of bivariate relationships which show that during high intensity exercise, a disproportionate increase in pulmonary ventilation (E) relative to metabolic production CO2 decreases arterial partial pressure of CO2 (PaCO2). Decreased Pa CO2 leads to vasoconstriction of cerebral vessels which in turn reduces cerebral blood flow (CBF) and oxygenation. While these relationships demonstrate a linkage between hyperventilation and cerebral deoxygenation, they do not describe the relative influence that each step has on cerebral oxygenation.
Canonical Correlation Analysis (CCA) is a statistical method which analyzes relationships between multiple predictor and criterion variables, as opposed to multiple regression which evaluates multiple predictors against a single criterion (Sherry et al. 2005). CCA is well suited for physiological investigations since multiple variables are logically grouped to describe an overall effect on an organ system (e.g. individual spirometry measurements are grouped to assess overall pulmonary function). CCA is a particularly promising tool for analyzing cerebral oxygenation data obtained via near infrared spectroscopy (NIRS), because the criterion oxygenation measurement is a function of concurrent changes in oxygenated, deoxygenated and total hemoglobin concentrations (O2Hb, HHb and THb). Moreover, CCA offers an advantage over multiple bivariate analyses because it yields ‘cross loading’ terms which explain the relative influence that each predictor has on an overall criterion value while controlling for Type I error.
In this study, we illustrate how CCA was used to test the hypothesis that hyperventilation influences cerebral oxygenation via hypocapnia-induced cerebral vasodilation in normoxic and hypoxic conditions. We hope that this work will inspire others to consider the use of CCA in physiological studies.
Athletic subjects were recruited from the local community as part of a larger study investigating acute mountain sickness. All participants provided informed consent conforming to the Declaration of Helsinki and underwent physical examinations to assess general health prior to participation.
Subjects performed two randomized and blinded incremental exercise trials in a hypobaric chamber under normoxic (PB = 610 mmHg; PIO2 = 118 mmHg) and hypoxic conditions (PB = 425 mmHg; PIO2 = 79 mmHg) with 30 min of rest between. After reaching the first target pressure, 15 min were taken to instrument the subject. Resting data were collected for 2 min prior to a 5-min warm up period at 50W. Workrate was then increased continuously (25W/min) until subjects could not maintain 60 rpm (Wmax). Cool-down exercise was performed at 50W for 5 min before chamber pressure was adjusted and the protocol was repeated.
A continuous wave NIRS (Oxymon MK III, Artinis Medical Systems, The Netherlands) was used to monitor cerebral oxygenation throughout the chamber session, as previously described (Subudhi et al. 2007). NIRS transmitters and detectors were spaced 40 to 50mm apart, depending on individual head geometry, just above the supra-orbital ridge and were tightly strapped with a custom made headset. Tissue absorbance measurements of two NIR light sources (780 and 850 nm) were used to calculate O2Hb and HHb concentrations (in μM) using the modified Beer-Lambert law and a path-length factor of 5.93. Total Hb (THb) was calculated as the sum of O2Hb and HHb and used as an indicator of cerebral blood volume (CBV). All values were expressed as changes from the rest period prior to warm up, arbitrarily defined as 0 μM Data were recorded at 125Hz and averaged in 5 sec windows prior to analysis. Transcranial Doppler (DWL Multi Dop T2, Germany) was used to measure cerebral blood flow velocity of the left middle cerebral artery (MCA). The custom NIRS headset was modified to hold a 2-MHZ Doppler probe over the left temporal window to insonate the MCA and record the maximal velocity envelope at 125Hz. Time averaged mean velocity (CBFv) was calculated beat-by-beat and reported as % change from rest.
Minute ventilation (E) and mix expired gases were analyzed via a metabolic measurement system (TrueMax 2400 analyzer, Parvomedics, USA). Oxygen consumption (O2) and carbon dioxide production (CO2) were calculated in 15 sec increments using the Haldane transformation. End tidal O2 and CO2 were sampled at the mouth with a separate analyzer (O2Cap, Oxigraf, USA). Subjects cycled in an aerodynamic position (Velotron, Racermate, Seattle, WA) which allowed fingertip oxygen saturation (SpO2) measurements from the left middle and ring finger (Nellcor, N-200 oximeter, USA; Criticare, 503 oximeter, USA). The average of both SpO2readings was used for analysis. Mean arterial blood pressure (MABP) was measured continuously from the left index finger (Nexfin HD, BMEye BV, The Netherlands). Analog signals from all instruments were converted to digital format (Powerlab, ADInstruments, Colorado Springs) and merged into one datafile with a samping rate of 200Hz for offline analysis.
For each individual, data were divided into 100 bins based on maximal power output (1 to 100% Wmax). A two-segmental piecewise linear regression identified the downward inflection point in PETCO2 (Sigmaplot 10.0, Systat Software). This downward inflection point represents hyperventilation in response to exercise intensity and is known as the respiratory compensation point (RCP). Repeated measures ANOVAs were run to evaluate differences across workrate (rest, RCP, 100% Wmax) at each PB (610 and 425 mmHg). A canonical correlation analysis (CCA) determined relationships between cardiopulmonary and cerebral oxygenation measurements before and after the RCP. The CCA grouped individual predictors and criteria to calculate overall predictor and overall criterion variable. The canonical correlation was the bivariate correlation between these two overall variables. In this study, MABP, E, SpO2, PETCO2 and CBFv were considered predictors of the CCA derived index of cerebral oxygenation (COX), calculated from concurrent changes in O2Hb, HHb and THb. To study the individual influence of each cardiopulmonary variable on COX, the cross loadings derived from the CCA were analyzed. Cross loadings represented the bivariate correlations between individual predictors and COX. Statistical significance was evaluated using the multivariate Wilks Lamba statistic to control for type I error at the P < 0.05 level.
Data from 23 subjects (21 males, 2 females) were used for analysis. Subjects were 28.6 ± 7.9 years of age 181.7 ± 8.0 cm tall, and weighed 73.7 ± 10.0 kg.
O2-max was 3.49 ± 0.76 L/m and Wmax was 305 ± 52 W. The average RCP occurred at 199 ± 47 W or 65± 11% of Wmax (Figure 1). Prior to the RCP, changes in E, PETCO2, CBFv, and NIRS variables (O2Hb, HHb and THb) were highly correlated (r2 > 0.62); however, after the RCP, PETCO2 – CBFv and CBFv – NIRS variables correlations were reduced (r2 < 0.29). The CCA(rCCA > 0.99; Wilks Lamba < 0.05) showed that E and PETCO2 were strongly associated with COX throughout exercise (r2 > 0.62), yet changes in CBFv explained the least amount of variance after the RCP (r2< 0.23) (Table 1).
Absolute O2-max (2.74 ± 0.60 L/m), Wmax (256± 38 W) and RCP (171 ± 26 W) were 21.5 %,16.2% and 14.1% lower than normoxia (P < 0.01), yet the RCP occurred at a similar relative work rate (66 ± 8 % of hypoxia Wmax; P = 0.89 vs. normoxia). Prior to the RCP changes in E were highly correlated with changes in O2Hb (r2 = 0.93) and HHb (r2 = 0.85) while after the RCP PETCO2 (r2 = 0.84) and E (r2 = 0.84) shared the most variance with cerebral oxygenation. After the RCP the correlation between CBFv and NIRS increased (r2 = 0.18 to 0.56), but was only moderately strong. The CCA (rc > 0.99; Wilks Lamba < 0.05) showed that changes in COX were more closely related to changes in PETCO2 and E than CBFv (Table 1).
We used CCA to test the hypothesis that hyperventilation influences cerebral oxygenation via hypocapnia-induced cerebral vasodilation under both normoxic and hypoxic conditions. Although hyperventilation resulted in concurrent decreases in PETCO2, CBFv and COX, results of the CCA caused us to reject the hypothesis. Our rationale for rejecting the hypothesis based on CCA is discussed below.
The strong canonical correlations between grouped cardiopulmonary and cerebral oxygenation variables indicate that the model was successful (rc2 > 0.99). To assess the relative influence of each cardiopulmonary variable on the overall measurement of cerebral oxygenation (COX) we evaluated the cross loading terms (Table 1). Specifically, if COX was mediated by a reduction in CBF then we would have expected to see strong cross loading between CBFv – COX. This was not the case. CBFv shared the least amount of variance with COX, showing that factors other than reduced CBFv explain decreased cerebral oxygenation near maximal exertion. In fact, bivariate analyses obtained with CCA showed that reductions in CBFv were associated with increased THb (r = −0.69, P < 0.01), an estimation of regional CBV. This paradoxical finding was the result of an increased HHb contribution to the overall NIRS signal, suggesting that an increase in the relative proportion of venous blood in the cerebral volume illuminated by NIRS provides a more accurate measure of NIRS evaluated cerebral deoxygenation during high intensity exercise. We propose that simultaneous increases in HHb and THb may be due to a reduction in cerebral venous outflow (Stolz et al. 2009), leading to venous pooling at exercise intensities above the RCP. Future research should thus be directed towards supporting or refuting this hypothesis..
CCA provided a statistical tool to analyze complex interactions between multiple variables while controlling for type I error. We have demonstrated how results can provide new insights on the mechanisms responsible for cerebral deoxygenation during exercise and propose that additional physiological understanding may be gained by applying CCA to other questions.
The authors express their gratitude to Brittany Miramon and Matthew Granger for their managerial and technical assistance during data collection. This project was funded in part by National Heart, Lung and Blood Institute Grant HL-070362
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.