Research to date strongly suggests the influence of biological factors in panic disorder (1–4). There is evidence that the disorder is familial and may be genetically transmitted (5, 6). Individuals with panic disorder are susceptible to having panic attacks in response to various metabolic challenges, ranging from sodium lactate infusion (7–9), caffeine ingestion (10), and carbon dioxide (CO₂) inhalation (11, 12) to sustained hyperventilation (13). In subjects with panic disorder, sustained hyperventilation commonly elicits slower end-tidal CO₂ (pCO₂) recovery after cessation (3, 14). From several studies, respiratory dysregulation also persists following treatment (15, 16), implicating “trait” respiratory components to the disease.

During alkalotic challenges, such as hyperventilation, a number of studies have investigated peripheral blood lactate response. In most investigations, blood plasma lactate abnormally increases in subjects with panic disorder to these challenges, even when there is control for subject fitness (13) or glucose availability (17). However, the time course and magnitude of acid-base response are different for the peripheral (blood) and central (brain) compartments (18). In prior work with single-voxel hydrogen (¹H) magnetic resonance spectroscopy (MRS) to measure brain lactate response during regulated hyperventilation (20 mm Hg) (19), more rapid and prolonged brain lactate increases were observed in asymptomatic medication-treated subjects with panic, a group specifically studied to evaluate the underlying features of the disorder. This evidence for abnormal compensation during alkalosis and increasing discussion of acid-base systems in panic disorder (20) led to the current investigation aimed at characterizing brain pH response to hyperventilation directly.

Phosphorous (³¹P) MRS provides a noninvasive measure of intracellular brain pH derived from the frequency shift between metabolites, most commonly inorganic phosphate relative to phosphocreatine (21). To improve time-resolution while ensuring a sufficient ³¹P signal-to-noise ratio, a larger volume of interest (a 30-mm slab) was employed compared to our single-voxel ¹H MRS work (19). The diffuse brain lactate response to hyperventilation (22) and the homogeneous pH distribution in the brain (23) suggest MRS voxel localization need not be markedly specific to assess this regulatory system. Two metabolic models in panic disorder were posited to explain the previously observed excess lactate increase to hyperventilation challenge: 1) subjects with panic disorder would demonstrate greater alkalosis to hyperventilation, implicating lactate as directly compensatory to an exaggerated brain pH response; 2) subjects with panic disorder would demonstrate reduced alkalosis, implicating lactate as overly compensatory to a normal pH response. In both models, we hypothesized that delayed pCO₂ recovery in subjects with panic disorder, if present, would be related to slower pH normalization in the posthyperventilation phase.

Subjects

Nine asymptomatic, medication-treated subjects with panic disorder (five men, four women; mean age=42.7 years, SD=15.3) and 11 healthy comparison subjects (three men, eight women; mean age=31.5 years, SD=9.6) were studied using ³¹P MRS during hyperventilation. One additional comparison subject was studied but was excluded from analyses because of poor-quality MRS data. Panic disorder diagnosis, based on DSM-IV criteria, was established using a structured psychiatric interview developed at the University of Washington, consistent with prior work (24, 25). The comparison subjects were similarly evaluated and had no history of panic attacks or a family history of panic disorder in first-degree relatives, as assessed by subject reports. All subjects were free of other axis I disorders. All subjects gave written informed consent for participation in the study, which was approved by the University of Washington Human Subjects Review Committee. Eight subjects with panic disorder were under treatment with fluoxetine (20–40 mg/day), and one subject was under treatment with gabapentin (3600 mg/day). The subjects with panic disorder were symptom free for at least 12 weeks before a magnetic resonance imaging (MRI) examination. While unmedicated, eight of the nine subjects with panic disorder had previously undergone a lactate infusion challenge and experienced a positive panic response, demonstrating group sensitivity to this alkalotic perturbation.

Symptom Assessment

Before and after completion of the MRI procedure, the subjects were administered the Acute Panic Inventory (26) (range=0–51). At similar time points, the subjects were also asked to rate their anxiety on a scale of 0 (none) to 4 (severe) and panic severity on a scale of 0 (none) to 10 (severe) with analog scales.

MR Hyperventilation Procedures

The MRI protocol consisted of a 12-minute baseline period (five scans of 2 minutes, 20 seconds each) in which subjects were instructed to “relax and breathe normally.” After this phase was a 12-minute hyperventilation period (five scans), during which subjects were instructed to breathe rapidly while fully inhaling and exhaling. A nasal cannula was used to sample expired CO₂ (pCO₂) throughout the experiment; tubing was run through a conduit in the magnet room and pCO₂ was analyzed with a NellCor 1000 monitor (Nellcor, Pleasanton, Calif.). Before entering the scanner, the subjects were trained to exhale through their noses (ensuring cannula pCO₂ measurement) and to inhale and exhale fully during the hyperventilation period. This latter instruction was aimed at avoiding shallow breathing during hyperventilation, which, by primarily clearing the dead space in the lungs, may affect measurement of pCO₂. Hyperventilation duration and instructions were similar to a respiratory study that induced a marked pCO₂ recovery delay in patients with panic disorder (14). During the hyperventilation phase, frequent feedback for all subjects at 30-second intervals was provided to aid in the maintenance of end-tidal CO₂ near 20 mm Hg. At the end of the hyperventilation period, the subjects were instructed to resume normal breathing without further feedback during a 24-minute (10-scan) recovery period.

MRI/MRS Procedures

³¹P MRS studies were performed with a clinical 1.5-T General Electric Signa (Milwaukee) whole-body scanner equipped with version 5.8 Genesis operating software. An experimental dual-tuned ³¹P-¹H birdcage coil, adapted from a modified circuit board and based on published work (27), was built at the University of Washington and used for all studies. The coil was optimized for both ³¹P and ¹H after substantial bench tuning.

Proton imaging was performed for detailed localization of the phosphorous MRS axial slab (sagittal T₁ localizer and an axial dual-echo sequence, TE=13/104 msec, 22 cm field of view, and 3 mm slice thickness). An axial section just inferior to the tops of the lateral ventricles was selected as the midpoint of the 30 mm slab examined with ³¹P MRS. This anatomical location was chosen to include the anatomical loci used in prior work (19) and to ensure optimal field homogeneity by avoiding sinus inclusion. Furthermore, this large sample volume included minimal signal contribution from nonbrain signals (28). Before switching to the ³¹P section of the coil, two passes of ¹H gradient shimming were conducted to optimize field homogeneity over the axial section.

For ³¹P MRS data collection, a spin-echo sequence detailed by Lim et al. (29) was used, employing a TE of 6 msec, a TR of 2 seconds, and 64 scan averages (2 minutes, 20 seconds of acquisition). After acquisition, ³¹P data were transferred to a workstation and converted to the magnetic resonance user interface (MRUI) (30) format. For each subject, the first spectrum in the series was used to establish starting parameters for subsequent spectra. Prior knowledge of peak line widths, frequencies, and couplings was included in the variable projection time-domain fitting of Lorentzian lines performed in MRUI with the Bolstad algorithm (Figure 1). The phosphate-phosphocreatine shift was used to compute pH in MRUI (pH=6.75 + [log¹⁰ ([phosphate-phosphocreatine] – 3.27)/(5.63 – [phosphate-phosphocreatine]), similar to the formula described by Petroff et al. (21). At each scan time point, pCO₂ measures were sampled. Data (pH and pCO₂ as within-subject factors) were compared with a repeated-measures analysis of variance (ANOVA), often referred to as a doubly multivariate model, by using SPSS 10 (SPSS, Chicago). To assess other metabolite changes over time, fitted amplitudes of phosphodiesters, phosphomonoesters, phosphate, and the high-energy phosphates phosphocreatine and ATP were referenced to the total ³¹P signal and also analyzed with repeated-measures ANOVA.

There were no group differences in gender distribution (p=0.36, Fisher’s exact test) (df=1, p=0.21), although the healthy subjects were somewhat younger than the panic disorder group (t=–1.99, df=18, p=0.06). Symptom differences between groups in the Acute Panic Inventory at baseline and after the hyperventilation procedure were demonstrated (baseline—healthy subjects: mean=0.50, SD=0.92; subjects with panic disorder: mean=4.56, SD=4.77) (t=2.77, df=18, p=0.01) (posthyperventilation—healthy subjects: mean=1.63, SD=2.06; subjects with panic disorder: mean=7.39, SD=7.19) (t=–2.54, df=18, p=0.02) and for anxiety ratings (baseline—healthy subjects: mean=0.36, SD=0.67; subjects with panic disorder: mean=1.33, SD=1.00; Mann-Whitney U=21.5, df=18, p=0.03; posthyperventilation: healthy subjects: mean=0.09, SD=0.30; subjects with panic disorder: mean=1.56, SD=1.33; Mann-Whitney U=13.5, df=18, p=0.002). No differences were found at baseline for the panic ratings (healthy subjects: mean=0.23, SD=0.75; subjects with panic disorder: mean=1.11, SD=1.83; Mann-Whitney U=37.0, df=18, p=0.18), although significant differences were present after the hyperventilation paradigm (healthy subjects: all=0; subjects with panic disorder: mean=2.33, SD=1.94; Mann-Whitney U=16.5, df=18, p=0.002). Despite greater endorsement of anxiety symptoms in the panic disorder group, no subject reported a response similar to a panic attack, nor did any subject meet DSM-IV criteria for a panic attack in response to hyperventilation.

In both groups, pCO₂ decreases were demonstrated across the hyperventilation phase for the effect of time (F=191.62, df=19, 342, p<0.001; Greenhouse-Geisser effect=0.001), with significantly different group response-by-time interaction (F=2.84, df=19, 342, p<0.001; Greenhouse-Geisser effect=0.02). The group main effect for CO₂ differences was less than significant (F=3.84, df=1, 18, p=0.07). Post hoc testing of the group-by-time effect demonstrated greater decreases in pCO₂ by scan nine in the panic disorder group (t=1.96, df=18, p=0.07) and significant decreases on the last scan of the hyperventilation period (scan 10) (t=2.72, df=18, p=0.01). In the posthyperventilation phase, pCO₂ recovery was significantly delayed in the panic disorder group on scans 11–17 (all t>2.17, df=18, p<0.04), with a difference observed on scans 18 and 19 (all t<2.01, df=18, p<0.06) (Figure 2). To ensure that the delayed recovery was not simply a result of different group pCO₂ minima at the end of the hyperventilation period, data were transformed for each subject to a percent recovery value (B–R)/(B–H), where B was the mean baseline pCO₂, R was the pCO₂ at each point in recovery, and H was the pCO₂ at the end of hyperventilation. Correction for the greater pCO₂ minimum in the panic disorder group at the end of hyperventilation with this percent recovery formula did not markedly affect pCO₂ results, with group differences observed on scans 11, 12, and 20 (all t>2.75, df=18, p<0.10) and significant differences in pCO₂ recovery for scans 13–19 (all t>2.37, df=18, p<0.03).

Across the respiratory challenge, brain pH increases were demonstrated for both groups (time F=5.36, df=19, 342, p<0.001; Greenhouse-Geisser effect=0.001). However, no differential effects of group (F=1.14, df=1, 18, p=0.30) or group by time were demonstrated (F=0.75, df=19, 342, p=0.77; Greenhouse-Geisser effect=0.64). Despite the greater pCO₂ decrease during the hyperventilation phase, the panic disorder group did not demonstrate increased brain alkalosis; instead, pH values for subjects with panic disorder were not significantly different from comparison values throughout the hyperventilation protocol (Figure 2). In the posthyperventilation phase, there was slower pCO₂ recovery, but pH recovery was not significantly different between the subjects with panic disorder and the healthy comparison subjects.

The discordance between pCO₂ and pH measurements in the panic disorder group led to exploratory analyses aimed to model the degree of pH blunting. With healthy comparison subjects’ pCO₂ and pH values, linear regression models were created for the hyperventilation (five scans) and posthyperventilation (10 scans) phases. Comparison group pCO₂ and pH values were entered into a linear regression for each phase, and slope and intercept terms were generated. Resultant linear regression models were as follows: hyperventilation: pH=–0.0124 * pCO₂ + 7.4178; posthyperventilation: pH=–0.0023 * pCO₂ + 7.228. pCO₂ data for subjects with panic disorder were then entered into these models for healthy comparison subjects, generating estimated pH response values for the panic disorder group. The predicted panic disorder pH curve from these models is shown in Figure 2, illustrating the discrepancy between the estimated and measured acid-base response.

No differences were found when we compared phosphomonoesters, phosphodiesters, or inorganic phosphate amplitudes by group (all F<1.88, df=1, 18, p>0.19), time (all F<1.06, df=19, 342, p>0.39; Greenhouse-Geisser effect>0.39), or group by time (all F<0.83, df=19, 342, p>0.67; Greenhouse-Geisser effect>0.39). Similarly, no differences were found for phosphocreatine or ATP by group (all F<1.21, df=1, 18, p>0.28), time (all F<1.43, df=19, 342, p>0.11; Greenhouse-Geisser effect>0.19), or group by time (all F<1.29, df=19, 342, p>0.19; Greenhouse-Geisser effect>0.25).

Despite similar training and feedback provided during the hyperventilation phase, the subjects with panic disorder had difficulty maintaining a hyperventilation rate that produced a pCO₂ of 20 mm Hg. This overbreathing led to more robust pCO₂ changes in the panic disorder group during the hyperventilation phase. Across this same interval, however, the magnitude of alkalosis was similar between groups. This physiological response pattern, wherein a substantially greater decrease in pCO₂ was not accompanied by a similar magnitude of alkalotic response (a pH increase), supports a model of greater alkalotic buffering in the panic disorder group. Consistent with this idea, estimating panic disorder pH response with data from the healthy comparison subjects as a model acid-base buffering system illustrates that the induced alkalosis should have been greater in the panic disorder group, reaching a maximum equal to the healthy comparison subjects at an earlier hyperventilation time point. This predicted asymptotic pH response with decreasing pCO₂ levels is similar to the pH curve observed in animal work (31), adding support for the validity of this modeling approach and data interpretation.

In the posthyperventilation period, there was delayed pCO₂ recovery in the panic disorder group assessed with either raw or percent recovery values, but the time course of pH normalization was not substantially different between groups. Recent evidence suggests that this may reflect the insensitivity of pH as a metric to predict hyperventilation maintenance. From a study of hypoxia in rats, dichloroacetate, administered to block lactate production, markedly reduced hyperventilation (32). If the converse is operating in subjects with panic disorder, it is possible that elevated lactate in response to hyperventilation might be driving or initiating the persistent respiratory behavior after voluntary hyperventilation cessation. Heuristically, it remains undetermined whether the exaggerated lactate response reported in prior work reflects increased production in panic disorder or whether lactate, once produced, is cleared more slowly, perhaps because of altered enzymatic activity, such as lactate dehydrogenase. This latter point has some support from the genetic literature investigating subjects sensitive to anxiety provocation by CO₂ inhalation (33). Further work measuring lactate in parallel with brain pH at high time resolution will be necessary to clarify a mechanistic description of events.

Similar to past work, medication-treated subjects with panic disorder were studied to minimize state-specific anxiety contributions to physiological measures and to allow assessment of underlying trait features (19). Although the panic disorder group showed increased levels of anxiety at baseline and after hyperventilation challenge in relation to comparison subjects, symptom severity was substantially lower than would be expected among untreated patients with panic disorder (24). Future work evaluating medication-treated healthy comparison subjects or a panic group that symptomatically improved with cognitive behavior therapy will aid in evaluating what effect, if any, medication had on the measured results.

A further point to consider when we interpret these results is the rapid cerebral blood flow (CBF) reduction that occurs in response to hypocapnea (34, 35). Altered blood flow response to hyperventilation in subjects with panic disorder has been implicated in two studies (36, 37). On this point, pilot work conducted in 1998 on six of seven subjects with panic disorder who participated in the ¹H MRS investigation by Dager et al. (19) did not demonstrate differential middle cerebral artery flow velocity changes using transcranial Doppler during an identical hyperventilation challenge (unpublished data of S. Dager and K. Beach). In the present study, high energy phosphates (ATP and phosphocreatine) were assessed as an indirect indication of hypoxia. Despite the increased hypocapnea in the panic disorder group, ATP and phosphocreatine levels were not significantly altered, suggesting that this factor, per se, was not a major contributor to the observed pH results.

Further study is necessary to evaluate whether lactate increase is a direct buffering response, as we have postulated, or instead a byproduct of another process affected by the acid-base perturbation. For example, it is likely that hyperventilation-induced increases in pH alter the redox state toward glycolysis (25). In this model, lactate increases would reflect a redox shift instead of a primary buffering process. Quantifying the time course and buffering components of such acid-base changes will be important to delineate in future work.

Whether pH and CO₂ dysregulation are also present in panic disorder subgroups not susceptible to lactate infusion (which also causes an alkalotic state) or for patients susceptible to only acidotic perturbations (e.g., CO₂ inhalation) (38) remains an interesting research question. It is also possible that subgroups of panic subjects sensitive to only alkalotic or acidotic paradigms have pH regulatory abnormalities that are specific in direction and a further subgroup sensitive to both challenge types has altered buffering in general. Future work exploiting higher field strengths to increase signal-to-noise ratio and improve the sensitivity of measuring small differences in pH response and energy metabolism will be helpful to further test and extend these observations.