Ethics statement

All experimental procedures were approved by the Institutional Review Board of the University of Texas at Dallas. Data from twenty four healthy subjects 12 younger (6 Male and 6 Female; mean age = 24years; range: 19–27 years) and 12 older (5 Male and 7 Female; mean age = 58 years; range = 55–71 years) with no history of head trauma and neurological diseases obtained as part of our earlier study [18] was used for the present study.

Experimental Tasks

Experimental tasks are explained in greater detail in our earlier study [18]. Briefly each subject performed a bilateral fingertap and a digit symbol substitution task (DSST) for the functional motor and cognitive tasks respectively. fMRI scans in the resting state and BH scans were also obtained in the same session for all subjects. The resting state scan was the first functional measurement, followed by the motor, BH and cognitive tasks. This order of scans was performed to avoid confounds that BH or cognitive tasks might introduce to the resting state response. Subjects remained relaxed with their eyes closed during the resting state scans. For the motor task, participants sequentially touched each finger of each hand to its respective thumb making one touch and release, at an easy rate of 0.5 sec in synchrony with the flashing circle. The experimenters observed the subjects prior to and during the scan where there was no group specific inability to maintain the minimal performance rate. The motor paradigm consisted of an initial 10 sec rest period followed by four repetitions of alternate epochs of 20 sec of bilateral fingertap and 20 sec of rest. BH consisted of a 40 second normal breathing period followed by three repetitions of alternate epochs of 20 seconds BH and 40 seconds normal breathing. Subjects performed an end-inspiration BH after inhaling a normal volume of air. Subjects were instructed not to inhale excessively prior to BH, but maintain their normal breathing volume [1], [21]. Using this style of end-inspiration BH, we have repeatedly demonstrated a non-biphasic response [1], [17], [21]. Normal inspiration significantly avoids the initial negative BOLD response typically observed in maximal-inspiration BH. An important reason we adopted end-inspiration BH is that it can be comfortably performed by younger and older subjects without reaching a breakpoint [15]. End-inspiration also minimizes reduction of lung volume and resulting alveolar contraction during the BH episode [22], [23]. This can be a significant confound as reduced lung volume reduces gas exchange between blood and the contracted alveoli during BH episodes. While the selective advantage of end-inspiration over end-expiration is still under debate, fMRI studies comparing the two observe significantly larger changes in ETCO₂, BOLD and CBF responses during end-inspiration BH compared to end-expiration BH [14]. A modified Digit Symbol Substitution (DSST) paradigm consisting of a code table representing pairs of digits and nonsense symbols was used for the cognitive task [18], [24] (Figure 1A). In the DSST subjects were told that, if one of the digit-symbol pairs from the code table matched a probe-pair presented simultaneously beneath the table, they were to press a right-thumb button; otherwise, they were to press a left-thumb button. There were 52 trials in a single scanning run. On half the trials, the probe pair matched one of those in the code table and on the other half the probe-pair did not match one of those in the code table. The stimuli stayed on the screen for 4 sec followed by randomly varying inter-trial intervals of 0, 4, 8, or 12 sec. The cognitive task responses were well controlled. To wit, inter-subject variability of the DSST-induced BOLD responses within our study design has been observed to be relatively smaller than that during the fingertap motor task [25].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Schematic of the DSST task paradigm and RSFA estimates.

A. During the DSST task, a digit-symbol key and digit-symbol probe appearing simultaneously for 4 seconds in each trial. The trials were jittered with 4-second periods of rest. B. Estimate of RSFA from the temporal standard deviation of the resting state BOLD time series.

https://doi.org/10.1371/journal.pone.0088751.g001

Magnetic Resonance Image (MRI) acquisition

MRI measurements were performed on a 3T PHILIPS scanner. Gradient echo-EPI images were subsequently obtained during resting state, motor (fingertap), cognitive (DSST) and BH tasks. 32 slices were obtained in the axial plane covering the entire brain. Imaging parameters were: FOV = 220 mm, matrix size = 64×64, TR/TE = 2000/30 msec, and slice thickness = 4 mm. Number of acquisitions were motor task = 110, cognitive task = 150, BH = 120 and resting state = 110.

Data Analysis

All analysis of the functional images were performed using AFNI [26]. Preprocessing and motion correction steps have been described in more detail in [18]. Briefly, a gamma-variate function was convolved with the task reference function and cross correlated with the BOLD signal on a voxel-wise basis to determine activation. During BH, the reference function was appropriately shifted to take into account the large hemodynamic delay during the BH response (Kannurpatti et al., 2002). Successful performance of the BH task by each subject was confirmed by a voxel-wise inspection of each subject data set during BH and observing the BOLD response to each epoch using the graph utility in AFNI. A threshold of p<0.01; Bonferroni corrected was considered for determining active voxels. This statistical threshold was optimal since it not only maximized the correlation between BH versus task and RSFA versus task responses with fewer outliers [17], but also preserved activity from tissue to a large extent. At this threshold, motor task-induced BOLD signal changes in the range 1–3% were the most prominent with 95% of activated voxels in the BOLD response range of <3%, confirming that the threshold was not biased towards large vessel signals [27]. Subject-level mean values were obtained after averaging across voxels within the task activated regions. Fractional changes in BOLD signal during task (% change) were estimated for the motor, cognitive and BH tasks which was different from the SD estimate obtained in our previous study [18]. The main reason for estimating the % change was to keep the analysis method consistent with other fMRI studies of CVR estimates. Fractional change in BOLD signal in response to task was estimated from the fit magnitudes of the general linear model (GLM). The temporal standard deviation (SD) was estimated from the resting state time series [18] that determined RSFA, (Figure 1B). All analyses were performed within the active voxels during the motor and cognitive tasks determined for every subject without any spatial normalization. For each subject, the frequency distribution of %BOLD change within the respective activated regions of interest during the motor and cognitive tasks was determined. Voxels making up the frequency count in frequency distribution analyses were obtained from the original functional images. Mean and SD of the frequency distributions were obtained for the younger and older subject groups. Coefficient of variation (CV = SD/mean) was estimated for various %BOLD changes within the motor and cognitive task activated regions. A similar procedure was carried out to obtain the distribution of RSFA within the activated regions. Equality of variance between the younger and older subject groups was tested using the Bartlett's Test with a p<0.05 required for a significant difference.

BH performance/physiological confounds in CVR estimation

BH has been frequently used in fMRI to test CVR [1], [2], [3], [11], [33]. However, when considering aged individuals or with disease, BH-induced BOLD responses may vary due to physiological factors such as differences in oxygen and CO₂ gas exchange in addition to performance differences in the BH task itself [12]. In younger and more homogeneous groups of human subjects, CVR assessed as a change in BOLD or CBF per end-tidal CO₂ change can be more accurate when using CO₂ inhalation methods [34]. However, continuous end-tidal CO₂ during BH durations are not available and need to be assumed as time-linear [35]. An important subject-related physiological factor is lung volume contraction during BH, impacting gas exchange between bood and alveoli [22], [23]. This is a difficult variable to account for using external measures in the magnet environment further compounded by age or disease. As several factors can affect the CVR value determined by ΔCBF/Δend-tidal CO₂ [34], the usefulness of capnometric measures in the magnet environment are still actively being researched, but have not explained BOLD outcome variability in aged or diseased individuals performing BH tasks. Other factors affecting BOLD outcome during BH is inspirational volume prior to an end-inspirational BH. While chest expansion is a good surrogate to monitor consciously varied inspiration depths prior to BH [12], its success in streamlining natural BH performance in real time is not known. Due to multiple physiological variables involved during the BH, additional physical task burden on subjects are unlikely to eliminate variations in outcome, no matter how hard the subject or patient may try [15]. If fMRI methods improve and evolve towards standard clinical applications, experimental designs with minimal subject discomfort or task burden will assure greater success. To date, there are no studies that have successfully decreased chest expansion variability in subjects naturally performing the BH task. From our prior experience with the BH task [1], [18], [19], [34], we optimized the 20 sec time length which could be completed comfortably by a large number of younger and older subjects. In post-study debriefing and time series inspection of the BH data in every subject, we ascertained that all subjects, irrespective of age, completed the BH task. Inspiration depth-induced BOLD variation was not a concern as subjects avoided inhaling excess air than they would normally when they spontaneously breathed. Inspiration depths requiring no extra efforts were perfected by subjects during a practice session prior to scanning. This practice procedure resulted in consistent BH-induced response with no significant differences in the mean activation volume between the younger and older groups (Figure 7). From prior studies, only deliberate efforts of 90–100% chest expansion (resembling a Valsalva maneuver) produce marked differences in BOLD, while BH inspiration depths variations between 40–80% produce very small BOLD differences [12]. Thus in trained subjects, small inspiration depth variations might not significantly affect BH-induced BOLD responses. Despite the younger and older subjects satisfactorily performing the simple 20 sec end-inspirational BH task, the observed loss of linear relationship with task in the older subjects indicated a different BH-related physiological cause than inspiration depth variations.

Vasomotive component: a major constituent of low frequency resting state BOLD fluctuations

Resting state low frequency BOLD fluctuations and their temporal characteristics have been frequently used as correlates of intrinsic neural activity [36], [37], [38], [39]. However, low frequency BOLD fluctuations in the resting state also represent end-tidal CO₂ variations [20] and their temporal and frequency spectral amplitudes have a high voxel-wise correlation with BH-induced BOLD changes [18], [19], [21], [40]. As BOLD signals are a convolution of neural and vascular variables and BOLD responses in the task-related and resting state correlate with each other [17], low frequency BOLD fluctuations cannot be exclusively interpreted as neural or vascular markers but might represent both [41]. The low frequency BOLD is bound to the vasomotive component related to respiration in spontaneously breathing human subjects [21], [42]. Our prior studies, along with other independent evidences correlating end-tidal CO₂ fluctuations with low frequency BOLD fluctuations have established RSFA as a strong CVR correlate, available from a simple resting state fMRI scan of spontaneously breathing humans with their eyes closed [18], [19], [20]. As decreased resting- state functional connectivity occurs in eyes closed states compared to eyes open [43], [44], state variations in neural activity are minimal during estimation of RSFA in the current ‘eyes closed’ design and ideal for CVR estimates. Low frequency BOLD fluctuations, however, can vary during mind-wandering which is difficult to control. But the overwhelming majority (almost 80–90%) of the low frequency spectral amplitude in spontaneously breathing humans can be attributed to vasomotive factors [42]. Even a subtle respiratory state change such as ‘cued breathing’ during the resting state can tremendously reduce the low frequency vasomotive component by more than 50% [42]. Hence subtle cognitive changes as associated with closed eyes, open eyes or mind-wandering that might alter coherence in neuronal circuits negligibly impact BOLD fluctuation amplitudes compared to vasomotive changes. It is for this reason vasomotive signals (termed as noise) are generally removed to improve resting state functional connectivity analyses of brain functional networks [42]. In other words, the resting state vasomotive component that RSFA represents can be imagined as the maximal response to randomized event-related BH trials of varying short durations (eg., 1, 2, 3 or 4 seconds) depending upon the time-dependent respiratory variability (RVT) within the resting and spontaneously breathing subject.

Differences between RSFA and BH

RSFA can be considered to be independent of task performance and behavioral variability unlike BH. Our results indicated that the linear relationship between RSFA and task was intact in the younger and older subjects (Figure 3). CVR variability represented by RSFA increased in the older subjects unlike BH. Thus these two variables behave quite differently. A comparative analysis of the subject-wise relationships between BH and RSFA showed significant to near significant linear relationships in the younger subjects within the motor and cognitive task activated areas (Figure 6A and 6B) whereas these relationships were much in the older group (Figure 6C and 6D). Because younger and older subjects performed the motor and BH tasks consistently, the results from Figures 7C and 7D indicated that different subjects within the older group, with quite variable CVRs, might lead to closely clustered BOLD responses during BH that might attenuate the accuracy of CVR estimation in older individuals. These results together indicate that BH might not accurately measure CVR in older individuals, and to a lesser extent, within younger individuals (Figure 6B). Overall, the primary reason for the absence of the linear relationship between BH-motor and BH-cognitive task responses in the older subjects appears, on the basis of our results, to be subject-specific performance and physiological outcome to BH and might not reflect age-related CVR changes per se within brain tissue. A notable result was an increased dispersion in RSFA in the motor areas compared to cognitive areas (that included the visual cortex) in older subjects (Figure 3B). This increased dispersion could not be attributed to CSF pulsations as there was also a proportional change in BOLD response to the motor task in these subjects. In an earlier analysis [19], we observed that low pass filtering of the resting state BOLD fluctuations (to remove CSF pulsations) still preserved 85–100% of RSFA's amplitude across the range of gray matter and white matter voxels. This indicated that CSF pulsations do not contribute significantly to RSFA within non-ventricular brain regions and for the same reason, RSFA was considered as a CVR marker without further signal processing such as filtering [19]. While further studies are required to attribute such a difference to selective age-related tissue atrophy in the motor areas, with the present results, one can only speculate to an age-related regional cerebrovascular difference of an unknown origin within motor cortical areas compared to visual cortical areas.

Preprocessing steps such as global mean-subtraction may influence RSFA. In a recent study of young adult humans, global mean-subtracted signals showed a low subject-level correlation between RSFA and task-evoked BOLD amplitude in the visual cortex [29], whereas correlation between coherence of the resting state fMRI signals and task-evoked BOLD amplitude at the voxel- or subject-levels were not affected by global mean removal [29], [45]. However, global mean-removal is unnecessary when estimating the CVR component represented by RSFA [18], [19].

Normalization of task-evoked BOLD response and RSFA

Hemodynamic scaling or normalization of BOLD responses can account for vascular-related changes within the brain and can correct for vascular differences among inhomogeneous population. Scaling or normalization can be performed in different ways. The first is by dividing the task-induced response by the CVR response on a voxel wise manner [21], [46]. A second approach is to divide the average task-induced response by the average CVR response from the subject-level response [16]. A third approach is to consider the CVR response as a covariate and remove it from the task-induced response [16]. However, a linear relationship between CVR and task response is a prerequisite to estimate normalized or scaled fMRI responses using CVR as a covariate [16]. The present results using a large age range indicate that BH may not satisfy this prerequisite as the linear relationship between BH versus task was disrupted. Hence, the unconstrained RSFA, with an intact linear relationship with tasks, can be used for within-subject normalizations [18] or between-subject normalizations [16] without any age-related or methodological constraints. Furthermore, in choosing a practical normalization variable, RSFA performs better than BH or CO₂ breathing for several reasons. Generally, vasomotive-related hypercapnic challenges whether BH or CO₂ breathing are not completely free of neuronal activity components. RSFA also contains neuronal activity components but only ‘spontaneous’ as opposed to ‘evoked’ in the other two cases. For example, BH evokes air hunger leading to excitation of respiratory-related neurons and hypercapnia with CO₂ ranging from 2% to 5% typically evoke neuronal activity within the brain [19]. Hence, the neuronal components in the BH and CO₂ challenges are relatively stronger (being evoked neuronal activity) than the spontaneous neuronal components present within RSFA.

Conclusions

Subject-related differences in physiological changes during BH can profoundly affect CVR determination in large age ranges. The unconstrained CVR marker such as RSFA available from the resting state fMRI signal is suggested as a better alternative to BH. The unconstrained assessment of subtle CVR differences in the higher and lower order brain areas were sensitive enough to detect regional vascular differences and well-suited for fMRI designs that include comparisons between healthy individuals, older individuals, patient- and special-populations [47].