Magnetic Resonance Spectroscopy of Hypoxic-Ischemic Encephalopathy After Cardiac Arrest

Abstract

Background and Objectives

To correlate brain metabolites with clinical outcome using magnetic resonance spectroscopy (MRS) in patients undergoing targeted temperature management (TTM) after cardiac arrest and assess their relationships to MRI and EEG variables.

Methods

A prospective cohort of 50 patients was studied. The primary outcome was coma recovery to follow commands. Comparison of MRS measures in the posterior cingulate gyrus, parietal white matter, basal ganglia, and brainstem were also made to 25 normative controls.

Results

Fourteen of 50 patients achieved coma recovery before hospital discharge. There was a significant decrease in total N-acetylaspartate (NAA/Cr) and an increase in lactate/creatine (Lac/Cr) in patients who did not recover, with changes most prominent in the posterior cingulate gyrus. Patients who recovered had decrease in NAA/Cr as compared to controls. NAA/Cr had a strong monotonic relationship with MRI cortical apparent diffusion coefficient (ADC); Lac level exponentially increased with decreasing ADC. EEG suppression/burst suppression was strongly associated with Lac elevation.

Discussion

NAA and Lac changes are associated with clinical/MRI/EEG changes consistent with hypoxic-ischemic encephalopathy (HIE) and are most prominent in the posterior cingulate gyrus. NAA/Cr decrease observed in patients with good outcomes suggests mild HIE in patients asymptomatic at hospital discharge. The appearance of cortical Lac represents a deterioration of aerobic energy metabolism and is associated with EEG background suppression, synaptic transmission failure, and severe, potentially irreversible HIE.

Classification of Evidence

This study provides Class IV evidence that in patients undergoing TTM after cardiac arrest, brain MRS–determined decrease in total NAA/Cr and an increase in Lac/Cr are associated with an increased risk of not recovering.

Hypoxic-ischemic encephalopathy (HIE) brain injury represents the leading cause of mortality and disability after cardiac arrest. More than 377,000 adults have fatal cardiac arrest annually in the United States,¹ and 80% do not regain consciousness and may remain in a coma for an unknown period of time.² As clinical examination may not be informative after targeted temperature management (TTM),³ ancillary tests are used to assess prognosis.

Although magnetic resonance spectroscopy (MRS) has been utilized in assessing prognosis,⁴ few studies have examined the range of changes observed in brain-specific metabolic markers through MRS in adults undergoing TTM. In neonatal patients undergoing TTM for hypoxic-ischemic injury, changes in MRS N-acetylaspartate (NAA) and choline (Cho) levels were associated with survival and neurodevelopmental outcome after TTM.^5,6 Previous studies have focused on brain regions such as the pons, basal ganglia, and thalamus, which for technical reasons may not provide accurate measure for MRS changes. The wide array of metabolites assessed allows for simultaneous evaluation of neuronal integrity and aerobic metabolism, both of which are primary pathologies in HIE.

We examined metabolites, as measured by MRS, in patients unconscious after cardiac arrest, and evaluate for differences between (1) patients with good vs poor coma recovery and (2) patients with hypoxic-ischemic injury vs normative control participants in the posterior cingulate gyrus (PCG), parietal white matter (PWM), basal ganglia (BG), and brainstem. These results are placed in the context of their EEG and conventional MRI changes to identify and quantify markers of HIE. This study examined whether, in patients undergoing TTM after cardiac arrest, brain MRS-determined decrease in total NAA/Cr and an increase in lactate/creatine (Lac/Cr) are associated with an increased risk of lack of coma recovery.

Methods

Patient Cohort and Clinical Outcome

This is a single-center prospective study of patients who underwent TTM for unconsciousness after in- or out-of-hospital cardiac arrest and return of spontaneous circulation (ROSC) at the Brigham and Women's Hospital between October 2016 and November 2020. Consecutive patients were screened for inclusion: age >18, admitted after cardiac arrest of any etiology regardless of pulseless rhythm, unconscious and unable to obey commands, and clinical indication for an MRI scan. Exclusion criteria were concurrent significant brain injury that may have resulted in cardiac arrest (intracranial hemorrhage, ischemic stroke). Normative controls consisted of 25 male volunteers without a history of head trauma or neurologic disease including brain tumors, epilepsy, dementia, or psychiatric disorders including posttraumatic stress, bipolar disorder, or psychosis, and between ages 45 and 70 (mean 57.9 ± 7.0) scanned from another study that used the same scanner and MRS parameters (DOD CDMRP WX81-XWH-10-1-0835).

Clinical data collected included age, sex, and initial pulseless rhythm (ventricular fibrillation/shockable vs pulseless electrical activity/asystole). Clinical examination findings, performed by the study team or the neurology consultation service, included best pupillary light response, corneal reflex, and motor response off sedation. Motor findings were dichotomized as flexor response or better (including flexion, withdrawal localization, or normal function) vs extensor response, triple-flexion, or no response.

The main clinical outcome was defined as coma recovery in those patients who regained the ability to open eyes and follow commands before hospital discharge (Glasgow Coma Scale M6). Secondary outcomes were as follows: (1) as some patients were able to follow commands but were still cognitively impaired, we evaluated cognitive return to baseline; (2) cerebral performance category (CPC) at discharge was binarized to good (CPC 1–2) or poor (CPC 3–5) outcome.⁷

Cardiac Arrest Protocol

TTM was performed according to the previously published local protocol at 33°C or 36°C for 24 hours.⁸ MRI scans were obtained per local clinical protocol if the patient did not recover to follow verbal commands within 24 hours after full rewarming. Withdrawal of life-sustaining treatment (WLST) was performed on a case-by-case basis through informed decision-making by the patient's family or guardian in close collaboration with the medical team.

EEG Monitoring

Continuous video EEG (cEEG) monitoring (Natus XLTEK system) was placed in the international 10–20 system as soon as possible and recorded for a minimum of 24 hours after normothermia. EEG data were interpreted using the American Clinical Neurophysiology Society critical care EEG terminology⁹ and classified into 3 categories per Westhall et al.¹⁰: highly malignant (suppressed background with or without continuous periodic discharges), malignant (abundant periodic discharges, electrographic seizures discontinuous/low voltage background, reversed antero-posterior gradient, lack of reactivity), or benign.

Neuroimaging

MRI Scanning

MRI/MRS scans were performed on a 3T Siemens Verio scanner using a 32-channel head coil. Structural imaging included 3D T1-weighted (magnetization-prepared rapid gradient echo: 1 × 1 × 1 mm³, repetition time 2,530 ms, echo time 3.36 ms) and diffusion-weighted images. Regional analysis of apparent diffusion coefficient (ADC) images was carried out using FSL (Oxford Center for Functional MRI of the Brain).¹¹ ADC images were first coregistered to the T1-weighted image for each participant. The T1 images were then nonlinearly registered to the Montreal Neurologic Institute (MNI) Brain Atlas, and the same transformation was applied to the ADC images. The success of automated registration was then manually confirmed. Artifacts were removed by filtering out signal intensity less than 200 mm²/s, while CSF was removed by filtering signal intensity greater than 2000 mm²/s. Regional anatomical masks were created using a 50% probability map from the MNI atlas. Mean ADC signal intensity was then calculated for each region. The percentage of whole brain and cortical voxels with ADC values <650 mm²/s were similarly calculated from whole brain and cortical anatomical maps.

Magnetic Resonance Spectroscopy

MRS was obtained only with a concurrent clinically indicated MRI. MRS was performed in 4 brain regions using a point-resolved spectroscopy (PRESS) sequence (echo time 30 ms, repetition time 2,000 ms, voxel size 20 × 20 × 20 mm³, 128 averages) in the PCG, PWM, left BG, and brainstem, as shown in Figure 1. Voxel placement utilized anatomical landmarks from the T1-weighted images, and subsequently, automated optimization (3D shimming, transmit gain, frequency adjustment, and water suppression) was performed. When necessary, MRI technicians manually shimmed to a linewidth of <15 Hz of the full-width half maximum of the unsuppressed water spectrum in the BG and brainstem to ensure consistent results.^12,13 Manual shimming was not required for the PCG or PWM as both of these regions have excellent B0 homogeneity. MRS of the BG (2 patients) and brainstem (12 patients) was not obtained due to time constraints.

Figure 1. Magnetic Resonance Spectroscopy Voxel Placement and Representative Spectra in Patients With Good and Poor Outcome.

Placement of the magnetic resonance spectroscopy voxels: (A) posterior cingulate gyrus; (B) posterior white matter; (C) basal ganglia; (D) brainstem. Spectra of illustrative patients with good outcomes (A.b, B.b, C.b, D.b) or poor outcomes (A.c, B.c, C.c, D.c). Cho = choline; Glx = glutamate + glutamine; Lac = lactate; NAA = N-acetylaspartate.

Single-voxel MRS raw data were preprocessed using singular value decomposition (SVD)–based channel combination, spectral registration to correct for frequency drift, and water suppression using the Hankel SVD method in OpenMRSLab.¹⁴ Metabolites were then fit and quantified using linear combination models (LCModel)¹⁵ yielding the following measurements: total NAA (including NAA + NAA–glutamate [Glu]), total Cr (including Cr + PhosphoCr), total Cho (including PhosphoCho + Cho), Glu, glutamine (Gln), Glu + Gln (Glx), myoinositol (mI), and Lac. Cr concentrations were not significantly different between the different cohorts and therefore were used to normalize the data across the study. Normative control scans used for comparison were acquired using the same scan protocol (PCG and PWM) and on the same scanner.

In all participants, major metabolites of NAA, Cr, Cho, and mI had Cramer-Rao lower bounds (CRLB) of less than 20% and thus no participants were excluded. Due to the high CRLB of Lac at very low concentrations, the data were not filtered to include participants with low Lac to avoid any study bias.¹⁶

Statistical Analysis

Comparative statistics (Student t test, Fisher exact test, Wilcoxon rank-sum test) were used as appropriate. Significant results were evaluated using a multivariable logistic regression model with dichotomized coma recovery as well as CPC at hospital discharge (1–2 vs 3–5) as dependent variables. Correlations between variables of interest were assessed using Spearman correlation. Multiple comparisons were adjusted for using a false discovery rate set at 0.05.¹⁷ p Values presented in the text are the raw p values; all p values are also presented in Tables 1 and 2 and eTable 1, links.lww.com/WNL/B746, with indications of their statistical significance after multiple comparisons adjustments. Calculations were performed using R 3.3.2.

Table 1.

Characteristics of the Study Population

Table 2.

MRS Peaks vs Outcome

Standard Protocol Approvals, Registrations, and Patient Consents

The local institutional review board (Partners Human Research Committee, 2014P001623) approved this study and consent was obtained for each patient from the legally authorized representative unless MRS was performed as part of clinical care.

Data Availability

Data will be shared at the request of other investigators.

Results

Characteristics of Study Cohort

Of 233 unconscious patients after cardiac arrest, 51 underwent cEEG, MRI, and MRS. One patient was excluded due to presentation with a large posterior circulation stroke immediately before cardiac arrest, resulting in a total of 50 patients who underwent analysis. Of these, 14 had coma recovery, of whom 7 returned cognitively to baseline; all patients had benign EEGs (Table 1). The mean time to coma recovery was 14.7 ±14.9 days (range 2–49 days). In patients who died, time to WLST was 9.2 ± 5.5 days. Three patients who did not have coma recovery survived, remaining unconscious at hospital discharge (CPC 4). One patient was awake, spontaneously moving extremities, and noted to follow 1-step commands approximately 75% of the time (CPC 3). There were no significant differences in age, sex, or race (eTable 1, links.lww.com/WNL/B746). No differences in time to ROSC were found (mean 27.2 minutes for patients with coma recovery vs 29.6 for others) but times were only available for 27 patients and most were estimated. Patients who recovered were more likely to have intact corneal reflexes (100% vs 55.6%; p = 0.002) and motor movements better than flexion (100% vs 11.1%; p < 0.001).

MRI/MRS was obtained at a mean of 6.4 ± 6.3 days (median 4.8 days) after cardiac arrest. MRS peaks were evaluated for effect of time elapsed to scan after cardiac arrest. Time to scan revealed a positive correlation with NAA/Cr (r_s = 0.43 [95% CI 0.17, 0.63]; p = 0.002) and a negative correlation with log (Lac/Cr) (r_s = −0.40 [−0.11, −0.60]; p = 0.007). No correlation between NAA/Cr (r_s = −0.0.3 [−0.55, 0.51]; p = 0.92) or log (Lac/Cr) (r_s = 0.14 [−0.42, 0.63]; p = 0.61) was found with time to coma recovery. EEG assessments were made within 24 hours of the MRI/MRS study in 47 patients; in 3 patients, there was a mean delay of 5 days. One patient was initially medically unstable and did not recover; the other 2 patients had more persistent mental status changes than expected leading to MRI scans, both of whom eventually recovered.

Persistent Lack of Coma Recovery Is Associated With Reduced NAA/Cr and Increased Lac/Cr Compared to Patients With Coma Recovery and Controls

Patients who had persistent lack of coma recovery had lower PCG NAA/Cr those who did recover (0.82 vs 1.26; p < 0.001, Tables 1 and 2, Figure 2). None of the patients with an NAA/Cr ratio under 0.98 recovered; 33% of patients between 0.98 and 1.2 recovered, and all patients above 1.2 recovered. PCG NAA/Cr was significantly lower in patients who died, as compared to those who survived (mean NAA/Cr ratio of 0.79 vs 1.20; p < 0.001).

Figure 2. Posterior Cingulate Gyrus NAA/Cr Ratios in Patient Populations.

Boxplots of posterior cingulate gyrus N-acetylaspartate (NAA)/creatine (Cr) levels in patients who had no coma recovery (did not follow commands), patients who had recovery (followed commands), and normative control participants.

Patients who lacked coma recovery had higher PCG Lac/Cr than those who recovered (0.39 vs 0.04; p < 0.001, Table 2). There was one outlier survivor whose Lac/Cr was higher than expected (0.136). Although included for purposes of completeness, there were technical difficulties with the scan; a water reference was not obtained, which may have affected scaling ratios.

PCG Lac/Cr was also compared to peak serum Lac during admission, and serum Lac level in closest proximity to the MRS scan, all but one of which was within 72 hours. There was no significant relationship between PCG MRS Lac/Cr peak and peak serum Lac (r_s = 0.09, 95% CI [−0.20, 0.36]) or with serum Lac in the proximity of the MRS (r_s =0.17 [−0.12, 0.42]).

In the PWM, BG, and brainstem, there were smaller differences in NAA/Cr and Lac/Cr between patients who did or did not achieve coma recovery. Only differences in Lac/Cr peaks reached statistical significance before multiple comparisons correction in the WM and BG.

Overall, MRS assessment had the highest association with coma recovery in the PCG as compared to PWM, BG, or brainstem. Poor coma recovery was associated with a decrease in PCG NAA/Cr by approximately 35% and elevated Lac production.

Patients With Coma Recovery Have Reduced NAA/Cr as Compared to Controls

Patients were compared to control participants with no history of brain injury or trauma. In patients without coma recovery, the following changes were observed:

• Decrease in NAA/Cr (PCG [0.82 vs 1.40, p < 0.001]; PWM [1.48 vs 1.87, p < 0.001]; adjusted for multiple comparisons)

• Decrease in GABA (PCG [0.19 vs 0.27, p < 0.001]; PWM [0.20 vs 0.23, p < 0.001])

• Decrease in Glu (PCG [0.97 vs 1.24, p < 0.001])

• Decrease in Glx (PCG [1.74 vs 2.10, p = 0.010]; PWM [1.18 vs 1.36, p < 0.001])

• Decrease in mI/Cr (PCG [0.55 vs 0.75, p < 0.001])

• Increase in Lac/Cr (PCG [0.39 vs 0.073, p < 0.001] and PWM [0.43 vs 0.22, p < 0.001])

• Increase in Gln/Cr (PCG [1.12 vs 0.48, p < 0.001]; PWM [0.47 vs 0.30, p < 0.001])

In patients with coma recovery, NAA/Cr was lower (PCG [1.26 vs 1.40, p < 0.001]; PWM [1.71 vs 1.87, p = 0.008]), in addition to a decrease in Glu (PCG [1.12 vs 1.24, p = 0.030]) and an increase in Gln (PWM [041 vs 0.30, p = 0.003]). No other peaks were different between the 2 populations; specifically, no difference in Lac/Cr was seen. These changes were still present in patients with coma recovery who had a cognitive return to baseline; NAA/Cr was lower in PCG (1.26 vs 1.40, p = 0.001) and PWM (1.76 vs 1.87, p = 0.035) and Gln/Cr elevated (0.43 vs 0.30, p = 0.020). These findings suggest that patients with coma recovery still sustained a HIE, but not at the same scale as the nonrecovered patients.

Reduced MRI ADC Signal Intensity Is Associated With NAA/Cr Reduction and Lac Elevation

There were significant reductions in MRI ADC measures in patients who did not have coma recovery in the whole brain, cortex, hippocampus, and white matter, with the cortex being most significantly affected (Table 1). Correlation between PCG NAA/Cr and ADC values were examined for mean total brain ADC as well as mean ADC values for individual brain regions (eTable 1, links.lww.com/WNL/B746). There were significant correlations with the cerebellum, cortex, hippocampus, globus pallidus, putamen, BG (combined), and white matter regions. There was a monotonic relationship between NAA/Cr and ADC (Figure 3), and it was strongest with cortical ADC values (r_s = 0.65 [0.45, 0.79]; p < 0.001). In comparison, there was a negative monotonic relationship with the log-transformed PCG Lac/Cr and cortical ADC (r_s = −0.74 [−0.58, −0.84]; p < 0.001). Overall, there is a monotonic relationship between cortical ADC values on MRI and PCG NAA/Cr and there is an exponential increase in PCG Lac with decreasing cortical ADC (e.g., increasing cortical diffusion restriction).

Figure 3. Relationship Between Cortical ADC vs PCG NAA/Cr, PCG Lac, and Coma Recovery.

(A) Monotonic relationship between cortical apparent diffusion coefficient (ADC) and posterior cingulate gyrus (PCG) N-acetylaspartate (NAA)/creatine (Cr). The red line represents the optimal cutoff for recovery to follow commands. (B) Inverse monotonic relationship between cortical ADC and log of PCG lactate (Lac). The red line represents the optimal cutoff for recovery to follow commands. (C, D) Relationship between % of cortical voxels with ADC <650 mm²/s and NAA/Cr (C) and log of PCG lactate (D).

White matter ADC values were correlated to PWM Lac/Cr (r_s = −0.46 [−0.35, −0.74]; p = 0.0044) with a trend towards correlation with PWM NAA/Cr (r_s = 0.28 [−0.0001, 0.52]; p = 0.052). Basal ganglia ADC values were correlated to NAA/Cr (r_s = 0.39 [0.008, 0.68]; p = 0.047) with a tend towards correlation with Lac/Cr (r_s = −0.41 [−0.023, 0.69] but p = 0.06 after multiple comparisons). Correlations between PWM and BG metabolites and ADC values from corresponding brain regions were otherwise not statistically significant.

Continuous Background on EEG Is Associated With Higher NAA/Cr; Suppressed or Burst Suppressed EEGs Are Associated With Lac/Cr Production

Patients who had a cEEG background had higher NAA/Cr than patients with a discontinuous background (0.79 vs 1.18; p < 0.001) and less likely to have a Lac/Cr over 0.14 (5 of 21 vs 26 of 29; p < 0.001). Patients who did not have coma recovery nearly always had discontinuous, burst suppressed, or completely suppressed EEGs whereas all patients who recovered had a continuous background on EEGs (Table 1). Reactivity to stimulation, though not tested or determinable in all patients, was more often present in patients who recovered (p < 0.001). There were no differences in the number of patients who had status epilepticus, generalized periodic discharges, or myoclonus. All patients with suppressed EEG backgrounds at the time of MRS/MRI scans had elevated PCG Lac/Cr peaks of at least 0.331. All patients with burst suppressed EEGs had PCG Lac/Cr peaks of at least 0.167.

In patients who did not achieve coma recovery, Lac/Cr was lower in patients who experienced a seizure during any time of their hospitalization than patients who did not (0.452 vs 0.163; p = 0.0048). No changes in Lac in relation to seizures were seen in patients who achieved coma recovery.

This study provides Class IV evidence that in patients undergoing TTM after cardiac arrest, brain MRS-determined decrease in total NAA/Cr and an increase in Lac/Cr are associated with an increased risk of not recovering.

Discussion

This study examines the spectrum of metabolic changes on MRS in patients after HIE and relationships with clinical, MRI, and electrographic markers of HIE. We demonstrate a tight coupling between NAA/Cr and MRI ADC values. ADC measures the impedance of water molecule diffusion and assesses cell membrane integrity.¹⁸ NAA, a molecule found in the brain at high concentrations, is sensitive in a nonspecific manner to several neurologic disorders and has been postulated to be a marker of neuronal health and bioenergetic dysfunction.^19,20 The high degree of correlation of these 2 disparate measurements in the cortex suggests that the cortex has the greatest sensitivity to HIE as compared to other sampled brain regions.

This study highlights the importance of the appearance of a Lac peak. Although there was an overlap of NAA/Cr and ADC values in patients who had coma recovery vs those who did not, a Lac concentration in the PCG of 0.14 was more consistently associated with the absence of coma recovery. Electrographically, complete background suppression or burst suppression, which are patterns associated with poor outcome in multiple studies,^10,21 were consistently associated with a Lac peak. As such, we hypothesize that MRS changes provide an in vivo view of the major pathophysiologic processes in HIE that may occur at different severities (Figure 4). NAA/Cr may be representative of early injury that may be most sensitive but potentially nonspecific for poor outcome. Most patients with a mild decrease in NAA/Cr and ADC achieve coma recovery. Increasing severity of injury results in only a portion of patients recovering. Injury severe enough to increase Lac represents severe, potentially irreversible HIE. MRS Lac levels have been demonstrated to be a marker of poor outcome after stroke.²²

Figure 4. Illustration of the Relationship Between Magnetic Resonance Spectroscopy, MRI, EEG, and Outcome After Cardiac Arrest.

Illustrative summary of magnetic resonance spectroscopy and apparent diffusion coefficient (ADC) peaks after targeted temperature management. The appearance of lactate resulted in a poor outcome. Patients between these values (in purple) have the potential for coma recovery. For illustrative purposes, diffusion-weighted images are shown rather than ADC maps. Cr = creatine; NAA = N-acetylaspartate; PCG = posterior cingulate gyrus.

The mechanism of elevated Lac peak in severe HIE may be viewed in terms of the integrity of oxidative metabolism. Rapid depletion of ATP after cessation of cerebral perfusion causes failure in the membrane ATP-dependent Na/K pumps, resulting in a massive influx of Na, efflux of K, and membrane depolarization. This results in opening voltage-gated Ca channels resulting in large increases in intracellular Ca, and activation of Ca-dependent K channels, resulting in further loss of selective membrane permeability.²³ Lac levels rise almost immediately as a result of a switch to anaerobic glycolysis, which quickly returns to baseline if perfusion is restored and normal mitochondrial function is resumed. Accumulation of Lac after re-establishment of perfusion is due to secondary energy failure in which neurons that survive the initial insult develop energy depletion from mitochondrial failure^24,25; elevated MRS Lac at 48 hours is associated with cell death and microglial activation in animal models,²⁶ and the accumulation of Lac portended an extremely poor outcome in pediatric HIE.^27,28 Failure to re-establish membrane gradient results in cytotoxic edema, cessation of cerebral perfusion, and neuronal transmission.

The failure of oxidative metabolism, as evidenced by Lac generation, and its relationship to synaptic activity may account for the consistent findings of either burst suppression or severe background voltage suppression seen in this cohort. Approximately 75%–80% of the brain's energy requirements are for signal processing, predominantly for the generation of action potentials and postsynaptic actions of neurotransmitters, predominantly Glu.^29,30 The generation of EEG signals is due to a summation of local field potentials, the most important source of which is synaptic activity.³¹ Varying levels of damage to the excitatory synapses have been demonstrated to recapitulate the EEG abnormalities seen in postanoxic EEGs, including burst suppression and discontinuous low-voltage recordings.³² Although it is likely that presynaptic ischemic failure is the initial cascade in inhibiting synaptic transmission in mild to moderate ischemia,³⁰ severe and widespread failure in synaptic transmission is likely coupled to the impairment of oxidative metabolism.

One unexpected finding is the increase in Gln in patients with poor outcomes. Glu and Gln are preferentially used as metabolic and cataplerotic substrates in the Krebs cycle during anoxia and ischemia. They may also function as reservoirs to protect against postischemic reduction in cardiac output by maintaining metabolic intermediates. The biosynthesis of Gln from Glu is catalyzed by the enzyme glutamine synthetase (GS); after the release and reuptake of Gln into neurons and glia, Gln is catalyzed back to Glu and ammonium by mitochondrial phosphate-activated glutaminase (PAG). GS activity increases in response to acute HIE in children and other compensatory mechanisms prevail in the case of chronic HIE.³³ In cell culture models of hypoxia, PAG activity is inhibited, potentially by the acidic pH induced by lactic acidosis.^34,35 The combination of increased GS activity and inhibition of phosphate-activated glutaminase may play a rescue role in preventing Glu-induced injury. Separating the Gln and Glu resonances remains challenging and further studies are required to confirm these findings.

Our study is in agreement with a large European study that demonstrated that NAA/Cr ratios measured in the pons and thalami were significantly lower in patients with unfavorable outcomes for MRI/MRS obtained between 7 and 28 days after cardiac arrest.⁴ Differences from our study include that their study reported only MRS values of the thalamus and pons, included both patients who did and did not undergo TTM, and included only patients who had survived for 7 days after cardiac arrest. The reason for selecting the PCG and PWM is that these regions have been shown to be highly sensitive to HIE³⁶ and are sites of reduced cerebral perfusion.³⁷ In addition, these regions do not require additional time for manual shimming and provide consistent results such that no participants needed to be excluded, compared with the exclusion of 17% of participants due to poor quality in the pons in the European study. Consequently, our results show that PCG is more sensitive for identifying MRS changes when compared to the PWM, BG, and brainstem. Another difference is that the previous study utilized long echo times (135 ms) whereby the use of a short echo time (30 ms) allows for the characterization of additional metabolites other than NAA, Cr, and Cho, such as Glu, Gln, and mI, due to their relatively fast relaxation times. Furthermore, we demonstrate a robust correlation between single-voxel spectra of the PCG and both cortical and whole-brain ADC, suggesting that spatial sampling bias may be less relevant as loss of perfusion as well as reperfusion is global. Future studies with multivoxel spectroscopy will help elucidate regional variations.

We observed a decrease in NAA/Cr values in patients with coma recovery as compared to normative control participants. This decrease persisted even in patients who had cognitive return to baseline at bedside examination, although full neuropsychological evaluations were not performed. We hypothesize that this is likely the result of mild HIE, rather than TTM itself exerting nonspecific changes in the MRS spectra once the patient has been rewarmed. It is unclear whether these patients experienced more subtle longer-term cognitive or other deficits, despite their good functional outcome at hospital discharge. Previous studies have demonstrated long-term cognitive deficits in >40% of survivors of cardiac arrest.³⁸ Interventions in addition to TTM to preserve structural and metabolic integrity after resuscitation to optimize neurologic recovery, even in patients who appear to have excellent in-hospital outcomes, should be explored in future studies. Recent preclinical and clinical studies have suggested promising effects of noble gases³⁹ and citicoline⁴⁰ as neuroprotective agents.

The sample size is too small to develop a multimodal outcome prediction model. This also likely explains the lack of association between coma recovery and status epilepticus, generalized periodic discharges, or myoclonus. Voxel selection did not target regions of greatest ADC or T2 changes on MRI scans, potentially decreasing MRS effect size. However, this minimized variability in signal/noise characteristics of different brain regions and allowed the generalizability of findings across participants. As MRS scans were only obtained with a clinically necessary MRI scan, there is variability in the timing of the image acquisition and potential selection bias. Although there were correlations between time elapsed to scan and MRS peaks, this may reflect a strong bias for patients with clinical suspicion of poor outcome to be scanned earlier in their course of hospitalization. Furthermore, we also do not find any difference in coma recovery in patients who only received MRI scans without MRS (28 of 86 [32.6%]) and no MRI scans (30 of 96 [31.3%]) during the study period at our institution. Previous MRS studies in stroke²² and neonatal HIE⁴¹ demonstrated a sustained decrease in NAA and a more transient increase in Lac. Without longitudinal MRS data, we are unable to determine the nature of MRS peak changes over time. Cortical ADC values of survivors are higher than typically seen in other studies, likely due to our selection of a more conservative cutoff value of 2000 mm²/s to remove artifacts,⁴² likely resulting in the inclusion of high ADC CSF voxels. Nonetheless, our results remained robust across a range of other cutoff values (1,000 mm²/s and 1,500 mm²/s).

Neuron-specific enolase and somatosensory evoked potentials were not routinely obtained on every patient and a pupillometer was not utilized. We assessed cognitive return to baseline to the patient's estimated premorbid baseline on bedside examination. A detailed neuropsychological evaluation, which was not performed, would be required to fully assess residual deficits. WLST was performed on a case-by-case basis and may potentially conflict with objective prognostic measures and treating clinicians were aware of the MRS results. The choice of coma recovery defined as following commands, rather than CPC, as the primary clinical outcome minimized the effect of these variabilities. On the other hand, the primary outcome of following commands may overestimate the patient's general recovery, as the patient may still have an overall poor outcome. The control cohort consisted of healthy male participants, and as such, may introduce bias related to sex and underlying brain health.

As with any testing modality, care must be taken in avoiding overreliance on a single measure, including the MRS, as demonstrated by 1 patient who underwent 2 MRS scans. A 40-year-old woman sustained a pulseless electrical activity arrest and thereafter experienced myoclonic status epilepticus. MRI/MRS was obtained 4 days and 7 days after cardiac arrest; neither scan revealed Lac or ADC changes, but NAA/Cr of the basal ganglia decreased 20.9% between the 2 scans. The patient eventually made a full recovery, without chronic seizures or observable cognitive deficits on long-term follow-up.

We demonstrate differences in several MRS-detectable metabolites in patients with cardiac arrest undergoing TTM who achieved coma recovery as compared to those who did not. NAA/Cr and diffusion restriction changes are highly correlated. Increasing severity of the injury is associated with the emergence of a rapidly increasing Lac peak, which was associated with either burst or complete suppression on EEG. MRS changes may be seen as compared to normative controls even in patients with presumed excellent recovery; some neuronal injury may be tolerated, whereas compromise to aerobic oxidative metabolism represents severe, potentially irreversible HIE.

Glossary

ADC

apparent diffusion coefficient

BG

basal ganglia

cEEG

continuous video EEG

Cho

choline

Cr

creatine

CPC

cerebral performance category

CRLB

Cramer-Rao lower bounds

Gln

glutamine

Glu

glutamate

Glx

glutamate + glutamine

GS

glutamine synthetase

HIE

hypoxic-ischemic encephalopathy

Lac

lactate

mI

myoinositol

MNI

Montreal Neurologic Institute

MRS

magnetic resonance spectroscopy

NAA

N-acetylaspartate

PAG

phosphate-activated glutaminase

PCG

posterior cingulate gyrus

PWM

parietal white matter

ROSC

return of spontaneous circulation

SVD

singular value decomposition

TTM

targeted temperature management

WLST

withdrawal of life-sustaining treatment

Appendix. Authors

Study Funding

The authors report no targeted funding.

Disclosure

J. Lee has performed contract work for Teladoc and Bioserenity, was the site PI for Engage Therapeutics, has received research funding from NINDS, and is the cofounder of Soterya, Inc. L. Sreepada reports no disclosures relevant to the manuscript. M. Bevers is supported by grants from the American Academy of Neurology and National Institute of Neurologic Disorders and Stroke and reports research funding and personal fees from Biogen outside the scope of the current work. K. Li reports no disclosures relevant to the manuscript. B. Scirica reports institutional research grant to Brigham and Women's Hospital from AstraZeneca, Eisai, Novartis, and Merck; consulting fees from AbbVie, Allergan, AstraZeneca, Boehringer Ingelheim, Covance, Eisai, Elsevier Practice Update Cardiology, GlaxoSmithKline, Lexicon, Medtronic, Merck, NovoNordisk, Sanofi, and equity in Health [at] Scale; and is a member of the TIMI Study Group, which has received institutional research grant support through Brigham and Women's Hospital from Abbott, Amgen, Aralez, AstraZeneca, Bayer HealthCare Pharmaceuticals, Inc., BRAHMS, Daiichi-Sankyo, Eisai, GlaxoSmithKline, Intarcia, Janssen, MedImmune, Merck, Novartis, Pfizer, Poxel, Quark Pharmaceuticals, Roche, Takeda, The Medicines Company, and Zora Biosciences. D. Silva, G.V. Henderson, and C. Bay report no disclosures relevant to the manuscript. A. Lin is a consultant for Agios Pharmaceuticals, Biomarin Pharmaceuticals, and Moncton MRI; is co-founder of BrainSpec; and receives research funding from NINDS, NIA, the Department of Defense, and the Alzheimer's Association. Go to Neurology.org/N for full disclosures.