Anxiety, Respiration and Cerebral Blood Flow: Implications for Functional Brain Imaging

Abstract

Brain functional imaging methods, such as fMRI, are sensitive to changes in cerebral blood flow (CBF) that are normally associated with changes in regional neural activation. However, other endogenous and exogenous factors can alter CBF independently of brain neural activity, thus complicating the interpretation of functional imaging data. The presence of an anxiety disorder, as well as change in state anxiety, is often accompanied by respiratory alterations that affect arterial CO₂ tensions and produce significant changes in CBF that are independent of task-related neural activation. Therefore, the effects of trait and state anxiety need to be given close consideration in interpreting functional imaging findings. In this paper, we review the dependence of most brain functional imaging methods on localized changes in CBF and the potentially confounding effects of anxiety-related alterations of respiration on interpreting patterns of functional activation. Approaches for addressing for these effects are discussed.

Introduction

Methods for imaging brain function are increasingly being used to investigate brain functional pathways, regulatory mechanisms, and changes associated with treatment in psychiatric disorders. This research has produced valuable insights regarding the pathophysiology of psychiatric disorders. Earlier functional imaging modalities, such as positron emission tomography (PET) and single photon emission computed tomography (SPECT), have been supplemented by functional magnetic resonance techniques (1), most commonly blood oxygen level-dependent (BOLD) fMRI. BOLD fMRI derives contrast from deviations in blood oxygenation that accompanies neural activity (2). During activation, the general increase in local blood flow results in an increased ratio of oxygenated to deoxygenated hemoglobin in that region. Oxyhemoglobin and deoxyhemoglobin have different magnetic properties that influence the magnetic susceptibility of surrounding tissues. In a typical experimental design, the so-called “boxcar design,” a rest and experimental state are alternated for a number of trials. The resultant data analyses average these relative signal differences (e.g., percent change) across periods to yield an activation map. This map shows areas with large changes (statistically inferred) in activity. By contrast, measuring resting state fluctuations in signal intensity (3) reveals what event-related BOLD analyses remove, constant activations readying the system for efficient processing.

BOLD fMRI is powerful investigational tool, and provides a number of advantages over other functional imaging methods. For example, compared to PET, BOLD fMRI provides greater sensitivity and spatial and temporal resolution, allowing the acquisition of more reliable event-related measurements in single subjects. However, a number of limitations are recognized for the method, as well. First, BOLD fMRI does not deliver quantitative measures. This makes changes in signal response sensitive to within- and between-group differences in baseline activity and reactivity (4). More recent developments have sought to quantify fMRI acquisition, but these methods are not widely used (5–7). Some newer approaches, which calibrate the BOLD signal using breath-hold techniques, hold promise for standardizing functional measurements across experiments and scanner field strengths (8). Second, the physiological interaction between neurons and glia during neuronal firing is elegantly complicated (9), with BOLD fMRI changes measuring only a single component of the interaction. The precise relationship between BOLD signal intensity changes, blood flow, oxygenation, baseline metabolic rates and neurotransmitter shuttling are very difficult to investigate (10). However, some recent studies utilizing new or combined techniques have begun to make substantial inroads and largely support assumptions of a qualitative relationship between changes in brain activation and blood flow (11) (fMRI based on changes in vascular space occupancy (12), for example). Third, BOLD signal intensity changes may not be precisely co-localized with neural activity. For example, differences in the magnetic resonance properties of blood and tissue can lead to detection of activity-related signal changes in draining veins at a distance from the site of brain activity (13). Other sources of spatial distortion may include incidental neural activation or blood flow increases extending beyond the area of target activation (14). Recent data suggest also that there may be only a small window of time (e.g., 2–3 sec) following neuronal activation during which spatially specific functional images may be acquired (15).

Since the blood oxygen changes on which BOLD signal responses rely are mediated by changes in CBF, factors affecting CBF that are independent of local brain activity need to be considered. A number of studies have shown that factors such as age (16), gender (17), hematocrit level (18), menstrual cycle (19, 20) , caffeine (21, 22) and nicotine (23) use and withdrawal, can all have significant impacts on the relationship between brain metabolic activity, CBF and BOLD fMRI signal response. One factor that can have an even greater effect on this interrelationship, but has not been widely accounted for, is the change in respiration associated with acute and chronic anxiety. Respiratory change is a sensitive index of both trait and state-anxiety and even small changes in respiration can significantly alter arterial carbon dioxide tension (P_CO2) which, in turn, has a large effect on CBF. P_CO2 has a large regulating influence on CBF, even greater than other intrinsic physiological factors, such as cerebral metabolism, cerebral perfusion pressure, or cardiac output. Global CBF is positively correlated to PCO2, with an approximately linear relationship between 20–60 mmHg PCO2. Thus, for each 1 mmHg increase or decrease in PCO2 over the range of 20–60 mmHg, there is a corresponding CBF change in the same direction of approximately 1–2 ml/100g/min, or 2–5% (24, 25). So, even small alterations in respiration can produce significant changes in global CBF. This may be particularly true in certain groups, such as those with panic disorder, who may experience greater changes in CBF during periods of hypocapnia (26).

Neuroimaging Studies of Anxiety

A comprehensive review of neuroimaging findings in anxiety states and disorders is beyond the scope of this paper. A number of recent reviews on the topic have been published (27–31). To summarize, consistent with many neurobiologic theories of pathological fear and anxiety (32), most functional neuroimaging studies have found evidence of abnormal activity in frontal and limbic-paralimbic cortical areas in patients with a variety of anxiety disorders. In panic disorder, for example, abnormal hippocampal activity at baseline (33), and higher insular activation and decreased prefrontal activation during symptomatic periods, have been shown (34, 35). Generally, patients with post traumatic stress disorder (PTSD) exhibit lower basal activation of the anterior cingulate and increased activation of paralimbic regions and the amygdalae, as well as greater decreases in frontal cortical regions during stressful tasks, such as trauma cues (36–40).

Anxiety, Respiration and CBF

Respiratory dysregulation is characteristic of anxiety and related symptoms are a diagnostic feature of several anxiety disorders. Patients with panic disorder, for example, typically exhibit chronic hyperventilation (41), breath-to-breath respiratory instability and frequent sighing (42, 43), even during panic-free periods (44). Sustained hyperventilation causes large reductions in global CBF, but even a single sigh can produce a 1–3 mmHg decrease in PCO2, enough to decrease CBF. Respiratory pattern differences are similarly present, albeit less consistently, in other anxiety disorders (e.g., generalized anxiety disorder) (44). Even for subjects without an anxiety disorder, increases in state-anxiety can be accompanied by acute increases in respiration rate and a corresponding hypocapnia (45–47) . Individuals with high trait-anxiety may be more likely to respond to psychological stress with exaggerated respiratory increases (48). Since even mild hyperventilation produces significant decreases in CBF that are unrelated to task-evoked activation, the effects of both trait and state anxiety need to need carefully considered when interpreting functional imaging studies based on localized changes in CBF. While it is more common for patients with anxiety disorders to increase ventilation (hypocapnia), the effects of breath holding are equally important to consider in the context of induced signal changes. A number of recent studies have focused on breath holding and BOLD signal changes as a metric to evaluate vascular tone, demonstrated to differ between children, adults (49) and aged samples (50). While vascular tone is an interesting concept to consider separately when evaluating BOLD signal changes, the impact of even short breath-holds (e.g., 3 seconds) at the start of an activation block (as may occur with increasing task demand) may confound fMRI results (51).

The MRI scanning process itself has been shown to increase anxiety and reduce CBF in a substantial portion of healthy subjects (52, 53). More than one-third of non-psychiatric general clinical patients report moderate to severe levels of anxiety when undergoing a clinical MRI procedure, with some experiencing severe anxiety or new onset of panic attacks (54–58). One recent study reported that almost half of patients experienced distress related to the noise and confinement during MRI scaning (59). It was also found that subjects differ significantly in how they cope to the stress of the MR environment (60). Marked changes in CBF associated with procedural distress appear to habituate after the first session in most healthy subjects undergoing multiple sessions (61).

However, patients with anxiety disorders habituate physiologically more slowly to stressful experimental stimuli, and so might be expected to show less habituation to the scanning process, as well (62). There is also some evidence to suggest that CBF effects due to anxiety follow a Yerkes-Dodson-type curve (52), with low anxiety subjects demonstrating an increase in CBF with anxiety, whereas highly anxious subjects show linear decrease in CBF with increasing anxiety(52). This may be due to an intrinsic tendency for greater respiratory reaction in anxiety patients, since state-related mood effects on respiratory responses are proportional to trait anxiety levels (63). These data also suggest that greater anxiety-related CBF responses can be expected in other psychiatric populations with overlapping features, such as major depressive disorder. These respiratory changes may have both long and short time components, for example, the initiation of a mild hyperventilation state which persists for minutes, interspersed with individual breath-holds or sighing during task initiation.

The confounding effects of arousal and state anxiety on functional contrast have been shown in several functional neuroimaging studies (64). However, in the published functional neuroimaging studies investigating situational anxiety or pathological anxiety states, there have been inconsistent efforts to control or account for the effects of respiration on imaging data. From efforts to account for respiratory influences, respiratory differences between anxious and non-anxious subjects have been found to exert such a large effect that, when CBF values are adjusted to PCO2 levels, any functional activation became non-significant (65). Others have found that CBF changes associated with anxiety are not wholly explained by changes in PCO2 (66, 67). Factors associated with anxiety states, such as systemic sympathetic activation also have a small direct effect on cerebral vasoconstriction (68) and can modulate CO2-mediated CBF reactivity (69) and additionally contribute to CBF changes associated with anxiety. Thus, differences in respiratory patterns or PCO2 cannot entirely account for CBF changes associated with anxiety, but the importance of PCO2 effects on CBF remains critical for interpreting functional imaging data.

Mechanisms of effect

The pathway through which PCO2 changes produce alterations in CBF is well delineated. PCO2 affects extracellular and perivascular pH, which alters the properties of vascular smooth muscle (70, 71). In the presence of carbonic anhydrase, CO2 is quickly hydrated to form carbonic acid and its dissociation products, bicarbonate and hydrogen ions. The resultant acidic extracellular environment enhances the vasodilatory effect of adenosine (72) and increases potassium ion conductance across vascular smooth muscle (73), resulting in blood vessel dilation and thus increased flow (74). The effect of pH on cerebral vascular tone may be mediated by a variety of factors, including nitric oxide (NO) (75), lactate (76), prostanoids (77), cyclic nucleotides (78), potassium channels (79) and intracellular calcium (80). Further, the degree to which CBF is altered by changes in PCO2 is also influenced by various intrinsic factors in addition to anxiety state, such as baseline CBF, age, cardiopulmonary status and the presence of cerebrovascular pathology. For example, CBF responsivity to changes in PCO2 progressively decreases with age and during periods of hypotension (81–83) . Regional cerebral ischemia also can affect CBF responses to PCO2, with graded reductions in reactivity occurring as a function of proximity to the site of occlusion and related to CBF reduction in the area (84). In contrast, chronic hypertension does not appear to alter PCO2 reactivity when subjects with hypertension are compared to normotensive subjects (85).

Cerebrovascular reactivity to changes in PCO2 also varies by brain region. For example, greater decreases in CBF occur in grey matter than in white matter during hypocapnia (86). Significant regional heterogeneity of cerebral vascular response to changes in PCO2 has also been reported. Using PET, Ito and colleagues (87) observed large relative increases in CBF (i.e., >10% per mmHg increase in PCO2) in the pons, cerebellum, thalamus and putamen during 5 minutes of hypercapnia induced by inhalation of 7% CO2 gas which were significantly greater than that measured for the brain as a whole. In contrast, smaller than average relative CBF increases were measured in temporal, temporo-occipital, and occipital cortical regions. During hyperventilation-induced hypocapnia, greater than average CBF decreases were measured in the pons, putamen temporal, temporo-occipital and occipital cortical regions relative to the brain as a whole. In contrast, smaller than average CBF reductions in the precentral gyrus, prefrontal cortex and cerebellum were observed during hypocapnia. However, variable regional decreases in response to hypocapnia were less pronounced (i.e., 3% per mmHg change in PCO2) than the variation in regional increases observed during hypercapnia. Similarly, evaluation of the regional distribution of BOLD fMRI signal decreases to lower PCO2 during sustained hyperventilation has observed greater reductions in the insular cortex, thalamus, striatum and cerebellum than in the frontal, parietal and occipital areas (88). In total, these observations suggest that cerebrovascular responsivity to PCO2 varies significantly by region, an additional factor to be taken into account when assessing the impact of respiration on functional imaging results.

Global versus regional effects

A related factor in considering the effects of anxiety on functional imaging contrast is how global alterations in CBF can influence localized changes in cerebrovascular reactivity related to neural activity. For example, Posse and colleagues (89) found that graded hypo- and hypercapnia produced significant changes in task-related localized fMRI signal contrast that was inversely proportional to PCO2. An unresolved controversy is the question of whether changes in global CBF affect local fMRI response patterns in simple (i.e., additive) or more complex ways. The resolution of this question has important implications for the analysis of neuroimaging data to account for the confounding effects of global CBF changes (as well as other factors that affect global signal intensity). Different approaches for analyzing fMRI BOLD responses have either assumed that regional signals are directly proportional to global signal intensity (90), or, conversely, that the effects are independent and additive (91).

In most of the studies that have sought to address this question, alterations in global CBF have been produced by manipulating PCO2 via CO2 inhalation, breath holding, or overbreathing. For example, Ramsey et al (92) used PET to measure CBF responses in the insular cortex, centrum semiovale and visual cortex to visual stimulation under conditions of hypocapnia, normocapnia and hypercapnia. They found greater increases in CBF in the insular cortex (grey matter) than in the centrum semiovale (white matter) associated with rising PCO2 from CO2 inhalation. In contrast, no significant differences in the magnitude of visual stimulus responses were observed between the three breathing conditions. Thus, it appeared that the increase in signal associated with stimulus-induced regional activation was independent of that associated with CO2 inhalation-induced increases in CBF. Li et al.(93) used fMRI to study BOLD response to photic stimulation under normo- and hypercapnic (using breath holding) conditions. They similarly found signal increases in visual cortex during breath holding, but BOLD response to visual stimulation did not differ from normocapnic response conditions. Also, Corfield et al.(94) reported no significant interaction between the effects of visual stimulation and PCO2 level on the intensity of BOLD signal response in occipital cortex.

In contrast to the above noted findings, Shimosegawa et al. (95) reported that absolute increases in PET signal in visual cortex during photic stimulation were significantly lower during hypocapnia than under normocapnic conditions. However, fractional increases (i.e., percent of change relative to baseline) produced by visual stimulation did not differ between hypocapnic, normocapnic and hypercapnic conditions. Bandettini and Wong (96), found that hypercapnia (from breathing 5% CO2) reduced activation-induced fMRI signal changes in motor cortex relative to normocapnic conditions. Weckesser and colleagues (97) observed that hypocapnia induced by voluntary hyperventilation significantly reduced BOLD fMRI signal response in visual cortex such that, in 3 out of 6 subjects in their study, signal changes were not different from those in unactivated areas (Figure 1). Additionally, Posse et al. (89) found that fMRI contrast changes due to visual stimulation more than doubled when PCO2 was increased from 20 mmHg to 50 mmHg, whereas, at a PCO2 of 70 mmHg, functional contrast was virtually abolished (Figure 2).

Figure 1.

(a) Results of the pixel-based analysis of the effect of visual stimulation during normocapnia (left) and hypocapnia (right). All pixels shown in white exhibit a positive correlation to the reference function (correlation coefficient .0.6). Four adjacent axial echoplanar images through the occipital cortex are shown for each condition. An extensive activation of primary and secondary visual areas is evident in normocapnia; no significant activation remains during hypocapnia. (b) Time course of signal intensity in the experiment with a long repetition time (3 sec), displaying visual activation during normocapnia and the effect of hyperventilation and visual activation during hypocapnia [visual stimulation indicated by the gray background; hyperventilation (HV) indicated by the fat horizontal line]. (Reproduced from(97), with permission).

Figure 2.

(top) Averaged time course of fMRI signal before, during and after a 3-second light flicker stimuls at different PETCO2. The prestimulus signal has been set to 100%. Time in seconds after the start of 8 Hz flicker light indicated by the gray bar. (bottom) Averaged time course for 20, 40, and 60 mmHg, normalized to the individual and global maximal amplitude, including standard deviation. (Reproduced from (98), with permission).

Other aspects of the event-related BOLD response appear also to be significantly affected by changes in arterial gases. Kemna and Posse (98) reported that the temporal dynamics of the event-related fMRI signal changed in accordance with PCO2 changes. For example, response curve full-width-at-half-maximum values increases with increasing PCO2 during visual stimulation. Likewise Cohen et al. (99) reported that under conditions of hypocapnia, normocapnia or hypercapnia, the magnitude of BOLD response to visual stimulation, as well as the temporal dynamics of those responses (e.g., onset time, time-to-peak), was inversely related to PCO2 levels. Kashikura and colleagues (100) also examined the effect of hyperoxia on visual event-related BOLD fMRI signals with findings that, in comparison with normoxic conditions, hyperoxia is associated with larger peak height and shorter peak time of BOLD response curves measured in V1 (101). However, those findings are in contrast to those of Lindauer et al. (102), who found a decreased stimulus-induced response during hyperoxia.

Thus far, we have discussed the influence of arterial gas tensions on CBF and, so, BOLD signal response. However, it may also be the case that changes in gas tensions affect neural activity itself. For example, increased extracellular pH has been associated with increased neural excitability. And, conversely, extracellular acidosis is accompanied by decreased neural activity (103, 104) . Thus, Huttunen et al. (105) found that certain EEG auditory evoked potentials are reduced during hypocapnia. They postulated that a rise in pH in neuronal tissues increases neural excitability and leads to greater spontaneous, asynchronous cortical neuron activity, thus, decreasing stimulus-locked neural responses. Recently, Friedman et al. (106) observed that patients with panic disorder exhibited a blunted brain pH response to hypocapnia during carefully regulated hyperventilation. This finding is consistent with prior reports of excessive brain lactate production in the disorder (107–109) , and might imply different neuronal activation response to PCO2 changes (from differential pH response) in this population. Though speculative, these results suggest that PCO2 changes can impact BOLD signal change by modulating both CBF and, potentially, neuronalglial activity, as well.

PCO2 and CBF oscillations

Periodic oscillations are another factor impacting signal measurement. Spontaneous periodic oscillations in cerebrovascular and metabolic resting states that are not directly related to brain-activity can be a significant source of noise in neuroimaging studies (assuming the oscillations are not themselves the subject of study, e.g. (3, 110)) as they can reach similar magnitudes as stimulus-induced changes (111). Periodic oscillations observed in CBF appear to be similar to flow changes seen in the peripheral arteries and occur in a least four distinct frequency ranges: heart rate, respiration rate, low frequency (~ 0.1 Hz) and very low frequency (~0.01–0.05 Hz) (112). Oscillations at these same frequencies have been observed in oxy- and deoxyhemoglobin and cytochrome-oxydase levels (111). Low and very low frequency range oscillations are also observed for PETCO2 (113). It has been posited that CBF oscillations facilitate cerebral autoregulation (114) in much the same way that peripheral vasculature regulatory control may be facilitated by periodic oscillatory behavior (115, 116), as has been described in more general terms of dynamic autoregulation (117–119). However, because spontaneous periodic oscillations in cerebrovascular and metabolic elements are not random, signal variability due to these activities is not significantly reduced by time averaging of data sets. In this context, oscillations in CBF can be significantly influenced by varying PCO2 levels with periods of hypercapnia suspending periodic fluctuations in CBF that, conversely, are augmented during hypocapnia (120). Therefore, some have sought to improve functional signal-to-noise ratios by filtering spontaneous physiological oscillations of known frequencies (e.g., heart rate, respiration rate) out of imaging signals (121). More recently, it was demonstrated that regressing out BOLD signal changes linked to periodic variability in respiratory tidal volume (or pacing subjects’ respiratory rate and depth) significantly improved detection of task-related activation and deactivation(122).

Approaches to accounting for PCO2 effects on BOLD fMRI response

There are several approaches for managing the effects of anxiety and PCO2 changes on functional neuroimaging studies. All require that investigators obtain continuous measurements of PCO2, or end-tidal CO2, which has been shown to be a reliable and accurate estimate of PCO2(123), even during hyperventilation (124). One approach has been to acquire imaging data only during periods when PCO2 and CBF have reached steady state levels, both at baseline and during challenges (86). This may be especially important for provocation studies, such as those that employ hyperventilation or breathing CO2-enriched air to induce anxiety or panic. However, steady states might not be achievable with some study designs or for highly anxious subjects. During hyperventilation, CBF begins to decrease rapidly within 20 seconds of onset of the drop in PCO2 (125), and then begins to level off within 1–2 minutes. With sustained hypocapnia, CBF values then gradually rise back toward baseline values. Similarly, during CO2 inhalation, CBF rapidly increases with increasing PCO2 and then slowly declines toward baseline levels during sustained hypercapnia. Thus, both under conditions of hypercapnia and hypocapnia it is unclear whether a true steady-state period in CBF can be attained. However, available data suggest relatively stable CBF for at least 2 minutes (88, 126), and possibly as long as 10 minutes (127), following the initial change in PCO2. Within 1–2 minutes of the termination of short periods of hyperventilation or CO2 inhalation, CBF appears to return to baseline levels, although a brief “overshoot” past baseline values may also occur (128). It should be emphasized, though, that characterizations of the CBF time course during experimental hypo- and hypercapnia have come from studies of healthy subjects and changes in CBF were estimated from measures of blood velocity changes in the middle cerebral artery. Therefore, it is unclear whether these time course data can be generalized to specific patient groups and to all brain regions. For example, this time course probably differs for patients with panic disorder, who exhibit slower recovery from hyperventilation episodes (129) (Figure 3).

Figure 3.

Means and standard errors of self-reported anxiety and minute-by-minute mean end-tidal pCO₂ for the three groups over the course of the experiment. BASE is the baseline. 6 x FB/REC refers to 6 1-min fast breathing periods and 1-min recovery periods, the following FB refers to a 3-min fast breathing period, and REC refers to a 10-min final recovery period. Anxiety ratings were not evaluated during the initial 6 recovery periods. (Modified from (129), with permission.)

A second approach has been to adjust for the effect of PCO2 on global CBF (74, 130). General bivariate adjustment formulas based on previously published findings can be utilized. However, large individual variability exists in the quantitative relationship between PCO2 and CBF. Therefore it is likely that this approach introduces new sources of error. An alternative approach to accounting for the effects of PCO2 on BOLD fMRI signal response is to calibrate the two variables within each subject by using transfer function analysis as the basis for regressing out changes in CBF due to PCO2 changes. For example, during a pre-experiment calibration session, images are acquired during each of three breathing conditions: resting PCO2 and then at sustained breathing levels adequate to maintain PCO2 above (e.g., 45 mmHg) and below (e.g., 35 mmHg) a subject’s resting level. A regression line of MR signal on PCO2 can then be computed as a with-in subject estimate of the respiratory influence on BOLD signal intensity by region that is then used as a scaling factor in interpreting subsequent experimental data for that subject.

Summary

Changes in respiration associated with anxiety and anxiety disorders can produce local and global changes in cerebral blood flow that is unrelated to regional neuronal activation. These changes confound the interpretation of neuroimaging study results and increase the risk of either identifying spurious results or masking potentially real findings. By integrating measurement of real-time PCO2 and fMRI indices, this important factor can be minimized, thereby ensuring that specific targeted experimental features are more clearly identified.