Proton magnetic resonance spectroscopy assessment of neonatal brain metabolism during cardiopulmonary bypass surgery

Abstract

Here, we report on the development and performance of a robust 3-T single-voxel proton magnetic resonance spectroscopy (¹H MRS) experimental protocol and data analysis pipeline for quantifying brain metabolism during cardiopulmonary bypass (CPB) surgery in a neonatal porcine model, with the overall goal of elucidating primary mechanisms of brain injury associated with these procedures. The specific aims were to assess which metabolic processes can be reliably interrogated by ¹H MRS on a 3-T clinical scanner and to provide an initial assessment of brain metabolism during deep hypothermia cardiac arrest (DHCA) surgery and recovery. Fourteen neonatal pigs underwent CPB surgery while placed in a 3-T MRI scanner for 18, 28, and 37° C DHCA studies under hyperglycemic, euglycemic, and hypoglycemic conditions. Total imaging times, including baseline measurements, circulatory arrest (CA), and recovery averaged 3 h/animal, during which 30–40 single-voxel ¹H MRS spectra (sLASER pulse sequence, TR/TE = 2000/30 ms, 64 or 128 averages) were acquired from a 2.2-cc right midbrain voxel. ¹H MRS at 3 T was able to reliably quantify (1) anaerobic metabolism via depletion of brain glucose and the associated build-up of lactate during CA, (2) phosphocreatine (PCr) to creatine (Cr) conversion during CA and subsequent recovery upon reperfusion, (3) a robust increase in the glutamine-to-glutamate (Gln/Glu) ratio during the post-CA recovery period, and (4) a broadening of the water peak during CA. In vivo ¹H MRS at 3 T can reliably quantify subtle metabolic brain changes previously deemed challenging to interrogate, including brain glucose concentrations even under hypoglycemic conditions, ATP usage via the conversion of PCr to Cr, and differential changes in Glu and Gln. Observed metabolic changes during CPB surgery of a neonatal porcine model provide new insights into possible mechanisms for prevention of neuronal injury.

1 |. INTRODUCTION

The potential for neuronal injury remains an ongoing concern for both children and adult patients requiring cardiopulmonary bypass (CPB) heart surgery.^1,2 Postoperative MRI findings have consistently reported multiple abnormalities believed to be associated with ischemic events,^3–5 and studies of cognitive function have consistently found neurodevelopment deficits.⁶ For decades, cooling via deep hypothermia circulatory arrest (DHCA) has been used to reduce brain metabolism and associated metabolic needs during these surgeries. More recently, multiple variations in CPB flow conditions, including antegrade cerebral perfusion (ACP) and retrograde cerebral perfusion (RCP),^7–10 have been used in combination with intraoperative neurological monitoring.¹¹

Survival rates for neonatal congenital cardiac surgery are currently 90% or higher, but 25%–50% of these patients later developed neurodevelopmental deficits, depending on the complexity of their disease, surgery, and multiple other factors.^3,12 Unfortunately, despite the theoretical advantages of CPB surgery using hypothermia with added cerebral perfusion, recent studies of neurodevelopment outcomes have not shown a significant difference,¹³ suggesting that details regarding brain metabolism and causes of neurological injuries during these procedures are not fully understood.^14,15 Improved measurements of brain injury during CPB surgery are likely to enable methods to reduce long-term negative neurodevelopmental outcomes for these patients.

Swine have been the most widely employed preclinical models, with measurements of brain blood flow and metabolism during CPB surgery obtained using a variety of techniques, including laser Doppler,¹⁶ optical imaging,¹⁷ microdialysis,^14,15 histopathology,^13,18 MRI (notably diffusion-weighted imaging),¹⁹ and magnetic resonance spectroscopy (MRS).²⁰ Among these methods, ¹H MRS is one of the most powerful, in principle allowing the noninvasive real-time monitoring of up to 20 brain metabolites.²¹ In addition to brain temperature being measurable via the temperature-dependent chemical shift of water,²² a typical ¹H MRS brain spectrum contains large peaks from N-acetylaspartate (NAA, a neuronal metabolite), creatine (Cr, an energy substrate), and choline (a cell membrane constituent), as well as smaller signals from myo-Iinositol (mI, an osmolyte and glial marker), glutamate (Glu) and glutamine (Gln) (involved in neuronal function), lactate (Lac, a measure of anaerobic metabolism), GABA (γ-aminobutryic acid, inhibitory neurotransmitter), glucose (Glc, the brain’s primary energy substrate), phosphocreatine (PCr, an energy substrate), and the antioxidants glutathione and ascorbate.²³

Although MRS performance, in terms of signal-to-noise ratio (SNR) and spectral separation, increases at higher magnetic fields, performance on widely available clinical 3-T scanners is an important benchmark for clinical and translational studies. A recent consensus paper for ¹H MRS brain studies reported semi-localized by adiabatic selective refocusing (sLASER²⁴) with least-square fitting of the resulting spectra as the best single-voxel acquisition method, and metabolite concentrations reported as ratios to either tissue water or total creatine (tCr; Cr + PCr), as well as SNR, linewidths, and estimated Cramér–Rao lower bounds (CRLBs), to be important metrics for quality control.²⁵

Current in vivo 3-T ¹H MRS methods do, however, have important limitations, particularly with respect to robust quantification of metabolites with highly overlapped spectral patterns. Important examples include Glu and Gln most often only reported as the sum of these signals (Glx = Glu + Gln),²⁶ suggestions to only report Glc as the sum of Glc and taurine (Tau),²⁷ and 3-T reports of only tCr rather than differentiating the Cr and PCr signals.²⁶

Perhaps the most fundamental challenge for validating in vivo 1H MRS data is that, given the lack of alternative noninvasive techniques to measure the targeted metabolites, there are typically no gold standards to use for quantification. Estimated CRLBs of variance (valid for unbiased estimators) are routinely available from most MRS processing software^26,28; unfortunately, there are typically no available metrics to assess estimator bias. However, the pig model of CPB surgery described here offers an important opportunity to induce a robust metabolic challenge (in this case circulatory arrest [CA] driving a well-characterized shift from aerobic to anaerobic metabolism) and observe the 3-T ¹H MRS-detectable metabolic changes prior to, during, and postchallenge with each animal serving as its own control.

This paper reports on our MRS methods and initial observations using this animal model. The primary goal of these experiments was to establish a robust single-voxel ¹H MRI/MRS acquisition and data-processing pipeline for measuring changes in brain metabolism during CPB surgery in a neonatal porcine model, and, utilizing this unique perturbation-response experimental animal model, to assess which metabolic processes can be reliably quantified via single-voxel brain MRS on a 3-T clinical scanner. Surgical parameters were chosen to establish the range of quantitative metabolite changes to be expected in these types of procedures.

2 |. METHODS

All animal procedures were conducted under an approved Stanford University Institutional Animal Care and Use Committee protocol. Following median sternotomy and central cannulation, neonatal pigs (N = 14, 2 weeks old, 10.4 ± 1.2 kg weight) were placed in a GE MR750 3-T scanner (GE Healthcare, Waukesha, WI, USA) for 18, 28, and 37°C DHCA studies under hyperglycemic, euglycemic, and hypoglycemic conditions with CA periods of 30–50 min (Figure 1).

FIGURE 1.

Representative 18°C deep hypothermia cardiac arrest study: (A) Equipment setup; (B) Selected voxel; and (C) Experimental timings. CPB, cardiopulmonary bypass

2.1 |. Surgical protocol

Specifically, as previously described,²⁰ for each piglet, after induction of anesthesia and a surgical plane of anesthesia was achieved, a median sternotomy was performed. The ascending and aortic arch, brachiocephalic artery and its three branches (right subclavian artery and both common carotid arteries) and left subclavian artery were dissected and mobilized for cannulation and isolation of perfusion of the head. Heparin (300 IU/kg) was administered. The right atrial appendage (venous) and ascending aorta (arterial) were cannulated. The piglet was then transported to the MRI suite and placed in the MRI magnet for imaging and spectroscopy. Members of the hospital perfusion team provide CPB support for all our cases to ensure the highest level of expertise and consistency for this critical procedure. Blood glucose levels were adjusted via the addition of dextrose. Both fentanyl (10 mcg/kg/h infusion) and 1%–2% isoflurane were continued on pump, and phentolamine 0.5 mg/kg was added to the perfusate to assist with cooling and rewarming.

Core cooling was commenced at a pump flow of 200 ml/kg/min. Brain temperature was measured using the chemical shift between NAA and water using the formula: ‒ 97:26 × Δ (H₂O ‒ NAA) + 293:28:²² Using this approach, the precision of the temperature measurement is +/− the inverse of the sampling window, which was +/− 1.25 Hz or about +/− 0.01 ppm. This corresponds to an approximate error of +/− 1.0°C. Once the brain reached the targeted temperature, baseline spectra at the targeted DHCA temperature were collected, and then total body perfusion rate was reduced to 100 ml/kg/min, the ascending aorta was clamped, and cold cardioplegia (Plegisol²⁹) was administered via the aortic root. CA was continued for approximately 50 min (a benchmark time period for completion of typical complex arch repairs), after which all clamps were removed, and rewarming was commenced at a total body flow rate of 200 ml/kg/min until a temperature of 37°C was achieved. The post-CA recovery of the piglet was observed for approximately 2 h, after which time the animal was euthanized. Table 1 lists representative CPB parameters and timings for a given piglet, and Table 2 lists the distribution of the 14 pig studies completed under this study in terms of brain glucose level and DHCA temperature.

TABLE 1.

CPB parameters and timings for a representative 18°C DHCA study

MRI study				Full flow (150 ml/kg/min): 1350 ml					Prime:	NaHC03 (mEq): 30
Date: 12 Sep 2020									Drugs:	Heparin (units): 10,000 units
Location:	Beckman			Deep hypothermia circulatory arrest						Mannitol (gm): 4.5
Cool to:	18°C			Temperature on this sheet is nasopharyngeal						Regitine (mg): 1.8 (not in prime)
Pig#:	9 Dec 2020			Weight: 9 kg						Calcium (mg): 200
Time	Flow (ml/min)	Iso	Hct	K+	Glucose (mg/dl)	Blood lactate	Ca++	Temp (°C)	ACT	Comment
9:10 AM	N/A	N/A	23	4.3	306	N/A	<0.25	N/A	999	Circuit gas, 25 g dextrose added after ABG
9:23 AM	1320	1	26	4.2	547	3.25	0.68	36.5	999	ABG, ACT, 12.5 g dextrose added
9:58 AM	1460	1	28	3.6	536	3.18	0.96	37.2	999	VBG, ACT, 12 g dextrose added
10:26 AM	1460	1	28	3.3	671	3.12	1.17	36.6	999	ABG, ACT
10:44 AM	1460	1	27	3.2	531	2.92	1.18	36.6	999	VBG, ACT, 7 g dextrose added, X-clamp at 1055
11:05 AM	1480	1	26	3.1	578	3.19	1.35	36.2	617	ABG, ACT, start cooling, pH stat, Regitine, 2000 units heparin
11:25 AM	1480	1	27	2.4	526	3.53	1.34	20.0	999	VBG, ACT, 7 g dextrose added
11:41AM	1480	1	29	2.4	671	3.76	1.25	18.3	999	ABG, ACT, Temp is from CDI
12:36 PM	1400	2	30	3.7	593	6.22	1.3	18.5	999	VBG, ACT, Rewarming at 1240, Regitine
12:59 PM	1390	2	30	3.4	582	6.85	1.2	34.2	857	ABG, ACT, 2000 units heparin
1:29 PM	1400	2	32	2.8	489	6.45	1.21	36.6	999	VBG, ACT
CPB time				On:	9:14 AM	Off:	1:49 PM	Total:	275 min
CircArrest time				On:	11:44 AM	Off:	12:29 PM	Total:	45 min

Abbreviations: ABG, arterial blood gas; ACT, activated clotting time; CDI, CDI-500 arterial blood gas shunt sensor (Terumo Cardiovascular Systems, Ann Arbor, MI).

TABLE 2.

Distribution of pig studies (number of animals) with respect to brain glucose level and deep hypothermia cardiac arrest (DHCA) temperature (total studies = 14 animals)

	DHCA temperature
Glucose level	37° C	28° C	18° C
hyperglycemic	-	2	3
euglycemic	2	2	3
hypoglycemic	-	2	-

2.2 |. MRS acquisition

An eight-channel ¹H RF knee coil was used to first acquire anatomical T₁- and T₂-weighted baseline MRI scans (T1 parameters: 3D BRAVO, 450 ms prep time, TE/TR: 2.6/6.2 ms, 1 mm slice thickness: 150 locations per slab, 256 × 256 matrix, and 4 min 46 s scan time; T2 parameters: 2D Fast Spin Echo, ETL 12, 13.4 ms echo spacing, TE/TR = 60/2000 ms, 1 mm slice thickness: 72 locations per slab, 256 × 192 matrix, and 4 min 1 s scan time). Single-voxel ¹H MRS data were then acquired from a 12 × 12 × 15 mm³ right midbrain voxel (Figure 1B). The supraventricle voxels used in this study were fairly large relative to pig brain structures, and segmented MRI data were used to measure an ~ 50/50 mixture of gray/white matter, with minimal cerebrospinal fluid (CSF) contribution. On average, 30 to 40 2.5-min (64 averages, TR 2000 ms) or 5-min (128 averages, TR 2000 ms) spectra were collected during a 3–4 h experiment. Baseline spectra at 37°C were acquired in the first 30 min before cooling; 2.5-min spectra were used during cool down and warming ramps to avoid averaging over too large a temperature shift and for better temporal resolution of brain temperature. The remainder of the time was taken up by additional imaging sequences, alternate spectroscopic sequences, and an individual prescan for each spectrum, as well as time for the surgeons to start and stop CA. Spectra were acquired with TE 30 sLASER²⁴ with 5000 Hz bandwidth and 4096 datapoints. VAPOR water suppression²⁴ with reduced flip angle was used with default 75 Hz water stop-band for residual water-by-design. In addition to nasal temperature feedback normally used for the bypass experiment, the brain temperature was measured as soon as each MRS acquisition was completed using the chemical shift difference between the water and the 2.04-ppm NAA resonance, and was used to avoid the overshoot of brain temperature observed when only using nasal temperature feedback. After the brain temperature equilibrated, baseline spectra at the target temperature were collected and then the pump was turned off for ~ 50 min of CA. The pump was then restarted, and rewarming was initiated until the brain temperature reached 37°C for a 1.5-h recovery period. The total elapsed time, including baseline measurements, CA, and recovery, was ~ 3.5 h/animal.

2.3 |. MRS processing

A hybrid LCModel basis set was constructed for quantifying these CPB studies; 3-T basis spectra created from density matrix simulations using chemical shifts and coupling constants extracted from experimental high-field NMR data do not account for the impact of chemical exchange, sequence, or scanner-dependent effects. In particular, chemical exchange effects are both temperature- and field strength-dependent. Hence, while the density matrix simulation strategy could be used if high-field spectra were available at all the temperatures of interest, these may yet be incorrect at 3 T due to the field strength dependence of chemical shift on exchange. Hence, we acquired experimental spectra for 21 of the 23 metabolites listed in Table 3 using phantoms warmed to both 20 and 37°C. Density matrix simulations were used to complete the LCModel basis set for the two remaining metabolites, scyllo-inositol (SYL) and N-acetylaspartylglutamate (NAAG). The simulated NAAG and SYL singlets were line broadened and scaled (taking account of the number of protons) to match the linewidth and peak area of the measured NAA methyl singlet. NAAG was excluded due to high costs, and the ¹H spectrum of SYL consists of a temperature-independent singlet.

TABLE 3.

Construction of hybrid 37°C LCModel basis set for CPB studies. For each of these solutions, we used a buffer consisting of 72 mM potassium phosphate dibasic, 28 mM potassium phosphate monobasic, with 0.5 mM of DSS as an NMR reference, 0.5 ml/L Magnevist gadolinium chelate to adjust NMR relaxation times, and 0.1% sodium azide as a preservative

Chemical	Name	Conc (mM)	MW	g/L	Acquired
ALA	L-Alanine	50.00	89.09	4.455	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
ASC	(+)-Sodium L-ascorbate	45.00	198.11	8.915	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
ASP	L-Aspartic acid sodium salt monohydrate	50.00	173.1	8.655	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
CRE	Creatine monohydrate	50.00	149.15	7.458	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
CRN	Creatinine	50.00	113.12	5.656	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
GABA	γ-aminobutryic acid	50.00	103.12	5.156	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
GLC	Glucose	50.00	180.16	9.008	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
GLN	Glutamine	50.00	146.14	7.307	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
GLU	L-Glutamic acid monosodium salt hydrate	50.00	169.11	8.456	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
GLY	Glycine	50.00	75.07	3.754	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
GPC	sn-Glycero-3-Phosphocholine	20.00	257.22	5.144	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
GSH	L-Glutathione reduced	50.00	307.32	15.366	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
GSSG	L-Glutathione oxidized	25.00	612.63	15.315	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
LAC	L-Lactate	50.00	112.06	5.603	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
MYO	myo-Inositol	50.00	180.16	9.008	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
NAA	N-acetylaspartic acid	50.00	175.14	8.757	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
NAAG	N-acetylaspartylglutamate				Simulated as single N-acetyl peak from HR chemical shift at 37°C
PCH	Phosphocholine Ca salt	40.00	329.73	13.189	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
PCR	Phosphocreatine	50.00	255.08	12.754	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
PE	Phosphoethanol-amine	50.00	141.06	7.053	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
SUC	Sodium succinate dibasic	50.00	162.05	8.103	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3
SYL	syllo-inositol				Simulated as single N-acetyl peak from HR chemical shift at 37°C
TAU	Taurine	50.00	125.15	6.258	Experimental 3-T sLASER 4000/30 10 KHz BW 8192 points 37°C pH 7.3

Abbreviations: CPB, cardiopulmonary bypass; DSS, dimethyl silylpentase sulfonate; MW, molecular weight; sLASER, semilocalized by adiabatic selective refocusing.

Additional care was used for quantifying Cr and PCr. Differentiating these two metabolites is strongly dependent on the small relative chemical shift differences of their CH₂ peaks, which are of the same order of magnitude as temperature-dependent shifts over our targeted 20 to 37°C temperature range.^30,31 Accordingly, while 20 and 37°C experimental 3-T Cr and PCr spectra were used as acquired, linear interpolation of the CH₂ and CH₃ singlet chemical shifts from 20 to 37°C were used to generate simulated Cr and PCr basis spectra at these intermediate temperatures. Absolute amplitudes of the basis sets were also adjusted to account for the impact of temperature on polarization.

To process the in vivo data, eddy current correction was performed on each free induction decay (FID) using the linear and cubic splined phase estimated from the water-unsuppressed FIDs low pass-filtered with a 25-Hz cutoff frequency.³² A pure water spectrum (PWS) was estimated by subtracting the spectrum of the water-suppressed spectrum from the water-unsuppressed spectrum. The PWS was subtracted from the water-suppressed spectrum for artifact reduction.³³

The preprocessed spectra, acquired from each pig study, were then quantified using LCModel utilizing the basis set at the temperature closest to the measured temperature at the time of data acquisition. All metabolites were then quantified in terms of the ratio to tCr (Cr + PCr), assuming a nominal brain tCr concentration of 8 mM, where the value for tCr was obtained by averaging the tCr estimate from each spectrum over the entire 3.5-h experiment. Figure 2 shows representative in vivo piglet spectra and corresponding LCModel fits.

FIGURE 2.

Representative 18°C porcine deep hypothermia cardiac arrest data from a single study showing ¹H spectra and LCModel fitting results at (A) Start and (B) End of circulatory arrest (CA). MRS parameters: sLASER, TR/TE = 2000/30 ms, 128 averages, 4 min/spectrum, 12 × 12 × 15 mm³ voxel. The most prominent peaks are from lipids, N-acetylaspartate (NAA), glutamine + glutamate (Glx), creatine + phosphocreate (tCr), choline (Cho), and myo-Inositol (mI). sLASER, semilocalized by adiabatic selective refocusing

To help visualize and quantify the metabolite time curves shown in Figures 3–5 (and the repeatability data presented in the supporting information [Supplemental Materials]), measured datapoints were resampled by linear interpolation onto a vector of uniformly sampled 1-min time points. The curves were then smoothed using a moving average filter of 5 min in length. Standard errors for the metabolite values were estimated by computing the standard deviation of the residuals resulting from subtracting the measured values from the smoothed curve.

FIGURE 3.

Data from representative porcine studies showing the effects of deep hypothermia cardiac arrest (DHCA) temperature (37, 28, and 18°C) on (A) Glucose (Glc) and lactate (Lac) concentrations, (B) Phosphocreatine (PCr) and total creatine (tCr) concentrations, and (C) Glutamine/glutamate (Gln/Glu) ratios during DHCA. Measured datapoints are marked by symbols ◇, ∘, and ▲, and smoothed curves correspond to 5-min time averages. Vertical lines denote the start and end of the circulatory arrest (CA) period

FIGURE 5.

Effects of brain glucose levels on glutamine/glutamate (Gln/Glu) ratios during deep hypothermia cardiac arrest (DHCA). Representative 28°C porcine DHCA data under (A) Hyperglycemic and (B) Hypoglycemic conditions (the same animals as shown in Figure 4; the euglycemic case is shown in the middle column of Figure 3). Brain glucose levels were manipulated by changing the blood glucose concentrations. The post-circulatory arrest (CA) rise in the Gln/Glu ratio increases with brain glucose levels

3 |. RESULTS

The spectral quality across all studies was generally excellent, with an LCModel-reported average linewidth of 4.2 ± 1.2 Hz (mean ± standard deviation) and SNR of 19 ± 2. Among the 23 metabolites quantified from each spectrum, we noted the most significant DHCA-related temporal changes in Lac, Glc, PCr, Cr, Glu, Gln, and alanine (Ala), and these were selected for a more careful analysis. In these experiments, we measured an average blood/brain glucose ratio of 3.0 ± 0.2, consistent with previously reported values.³⁴ For these experiments, we defined hyperglycemic, euglycemic, and hypoglycemic conditions as brain glucose concentrations at the start of CA as being Glc ≥ 6 mM, 3 mM ≥ Glc ≤ 6 mM, and Glc ≤ 3 mM, respectively. Summary data of the metabolite changes observed during CA are given in Table 4.

TABLE 4.

Summary data (mean ± standard deviation) for all N = 14 animals of metabolite changes during circulatory arrest (CA). Brain glucose at the start of CA was grouped into three categories: hyperglycemic ≥ 6 mM; 6 mM > euglycemic ≥ 2 mM; and 2 mM > hypoglycemic. Data are averaged over N = 2 animals for all cases, except for hyperglycemic 18°C deep hypothermia cardiac arrest (DHCA), where N = 3. For individual metabolites, the value given is the mM change in concentration at the start versus the end of CA. For Gln/Glu, the table reports the value of this ratio at the start of CA arrest and the maximum value for this ratio with the time to max given in minutes poststart of CA. Note that the duration of CA for 37°C studies is averaged for 30 min compared with 50 min for the lower temperature studies

Blood glucose	DHCA temp (°C)	ΔGlc (mM)	ΔLac (mM)	ΔPcr (mM)	ΔGlu (mM)	ΔGln (mM)	Gln/Glu @ start CA	Maximum Gln/Glu (t = min post-CA)
Hypoglycemic	28	−0.4 ± 0.3	3.7 ± 0.4	−2.2 ± 0.3	−2.6 ± 0.6	−0.5 ± 0.5	0.3 ± 0.1	0.4 ± 0.1 @ t = 95 min
Euglycemic	37	−1.7 ± 0.4	4.6 ± 0.4	−1.4 ± 0.3	−1.8 ± 0.6	−0.3 ± 0.2	0.3 ± 0.2	1.0 ± 0.2 @ t = 113 min
Euglycemic	28	−2.5 ± 0.4	5.7 ± 0.4	−1.8 ± 0.7	−2.3 ± 0.5	−0.6 ± 0.3	0.4 ± 0.2	0.9 ± 0.2 @ t = 97 min
Euglycemic	18	−3.6 ± 0.4	8.5 ± 0.5	−2.3 ± 0.4	−1.1 ± 0.6	−0.3 ± 0.4	0.3 ± 0.2	0.4 ± 0.1 @ t = 118 min
Hyperglycemic	28	−5.4 ± 0.3	9.0 ± 0.3	−2.2 ± 0.3	−2.0 ± 0.5	0.1 ± 0.3	0.3 ± 0.1	1.1 ± 0.1 @ t = 62 min
Hyperglycemic	18	−5.5 ± 0.4	8.4 ± 0.3	−1.8 ± 0.4	−3.1 ± 0.6	−0.1 ± 0.4	0.3 ± 0.1	0.6 ± 0.1 @ t = 70 min

3.1 |. DHCA temperature effects

Performing CPB at temperatures above 18°C reduces surgical times by shortening cooling and warming periods, but the effects of increased temperature on brain metabolism are not fully understood. We sought to examine this effect. An immediate loss of brain glucose and corresponding increase in Lac were clearly seen during CA in all pigs, and Figure 3 shows representative porcine DHCA ¹H MRS data acquired at CA brain temperatures of (A) 37°C, (B) 28°C, and (C) 18°C. On these, and subsequent metabolic plots, LCModel-computed datapoints are indicated by the symbols, with the solid lines representing 5-min time averages. Brain temperature was determined by the NAA to water chemical shift. Under euglycemic conditions, Lac concentrations at the end of CA were Lac_{end CA} = 10.0 ± 0.8 mM at 37°C, Lac_{end CA} = 9.0 ± 0.9 mM at 28°C, and Lac_{end CA} = 10.0 ± 0.7 mM at 18°C. These were not significantly different, despite the differing C. These were not significantly different, despite the differing CA temperatures. Post-CA Lac recovery rates, as estimated by the % reduction from the Lac signal observed at the end of CA compared with that observed at 50 min postend of CA, varied significantly between animals; and we did not detect a statiticially significant correlation with DHCA temperature.

The ATP buffering role of PCr is of considerable importance in the brain³⁵ and we looked for corresponding changes in the ¹H MRS PCr and Cr resonances during DHCA. All animals showed a significant decrease in PCr (and corresponding increase in Cr) during CA, with recovery upon reperfusion (Figure 3). The z-scores, computed by measuring the variance of the corresponding PCr time courses for each animal, and then averaging across animals studied under a similar DHCA temperature (N = 2 or 3) for the change in PCr at the beginning versus the end of CA, were z = 4.7, 2.6, and 5.8 for temperatures of 37, 28, and 18°C, respectively (p-values all less than 0.005). tCr levels remained approximately constant (within SNR limits) throughout all studies.

We further noted the effects of DHCA temperature on Gln and Glu concentrations. As shown by the Gln/Glu ratio curves in Figure 3C, there was a temperature-dependent increase in Gln/Glu, beginning at the start of reperfusion and peaking 1–1.5 h after blood flow was restored to the animal. Specifically, the z-scores for the increase in the Gln/Glu ratio observed at its maximum post-CA value versus the start of CA were: 37°C, z = 3.5 (p < 0.005); 28°C, z = 2.5 (p < 0.005); and 18°C, z = 1.3 (p = 0.10). Possible explanations for the observed Gln/Glu changes are presented in the Discussion.

3.2 |. Brain glucose effects

Patient blood glucose level at the start of DHCA is another surgical parameter subject to manipulation. The question is if having higher glucose concentrations at the start of CA provides additional fuel for the brain, and hence generates a neuroprotective effect. To start to examine this question, DHCA studies were performed, whereby the animal blood glucose levels prior to CA were adjusted via the addition or withholding of dextrose. These studies were performed at 28°C DHCA. Figure 4 shows the effect of Lac production during CPB surgery as a function of Glc levels at the start of CA. There was an observed decrease in peak Lac concentrations with decreasing starting brain glucose levels. However, the ratios of the change in Lac from the beginning to the end of CA, ΔLac, to change-in-glucose, ΔGlc, as a function of brain glucose levels, were: hyperglycemic: ΔLac/ΔGlc = 1:7 with 95% CI (1.3–2.2); euglycemic: ΔLac/ΔGlc = 2:3 with 95% CI (2.1–2.5); and hypoglycemic: ΔLac=ΔGlc / 9:3 with 95% CI (6.9–14.0). The effects of start-of-CA brain glucose levels on the observed Gln/Glu ratios are shown in Figure 5. The post-CA rise in the Gln/Glu ratio was observed to increase with increasing brain glucose levels. Potential sources for both the large observed deviation under hypoglycemic conditions from the expected stoichiometric glycolytic value of ΔLac/ΔGlc = 2:0, as well as possible effects on the Gln/Glu ratios, can be found in the Discussion.

FIGURE 4.

Effect of brain glucose levels on lactate production during deep hypothermia cardiac arrest (DHCA). Representative 28°C porcine DHCA data under (A) Hyperglycemic and (B) Hypoglycemic conditions (the euglycemic case is shown in the middle column of Figure 3). Precirculatory arrest (CA) brain glucose levels were manipulated by changing the blood glucose concentrations. Smoothed metabolite curves correspond to 5-min time averages. The change-in-lactate (Lac_{end CA} ‒ Lac_{start CA} = ΔLac) to change-in-glucose (Glc_{end CA} ‒ Glc_{start CA} = ΔGlc) ratios during circulatory arrest were (A) Hyperglycemic ΔLac/ΔGlc = 1:7 with 95% CI (1.3–2.2), and (B) Hypoglycemic ΔLac/ΔGlc = 9:3 with 95% CI (6.9–14.0). The large deviation under hypoglycemic conditions from the stoichiometric predicted glycolytic value of ΔLac/ΔGlc=2:0 suggests the utilization of an alternative energy source beyond glucose, possibly glycogen

3.3 |. Water linewidth effects

In addition to the reported changes in brain metabolites, we also noted distinct effects of DHCA on the water signal. Residual water line-shapes are known to be sensitive to the level of vascular deoxy-hemoglobin³⁶ and, based on modeling vessels and capillaries as cylinders randomly distributed in space, signal from deoxygenated intravascular water is theoretically predicted to exhibit a classic NMR “powder pattern” by Springer et al.³⁷

During CA, we observed a mild broadening of the unsuppressed water peak, which contrasted with a strongly asymmetric broadening at the level of the metabolites and was especially evident in the PWS-corrected residual suppressed water signal (Figure 6). The broadening of the water signal recovered after blood flow was resumed. At its broadest, the signal exhibited the shape of an NMR powder pattern, the observation of which was most clearly seen in the partially suppressed water spectra, especially after using PWS to remove interference from low-frequency eddy current water sidebands (Figure 6A). As seen in Figure 6B, via the first point into CA and the first point after flow restoration, the line broadening at the metabolic level did not appear to be confounded by the 75-Hz water suppression band used for these studies. The observed water line broadening and corresponding powder pattern shape are qualitatively consistent with an increase in deoxyhemoglobin during CA and a decrease in deoxyhemoglobin with reperfusion, as predicted by Springer et al. (see Figure 6C and the supporting information).

FIGURE 6.

Residual water powder pattern. Observation of asymmetric broadening of the (A) Residual water spectrum and its (B) Time course from a brain voxel in a neonatal pig model during circulatory arrest (CA). (C) Based on a model of randomly oriented veins/capillaries containing deoxyhemoglobin, a “powder pattern” is theoretically predicted from the intravascular water signal.37 Theoretical values for the frequency shifts of || and ⊥ blood vessels are 0.63 and ‒0.27 ppm, assuming O₂ sat = 0% and hematocrit = 0.6 (see the supporting information). LW, linewidth

4 |. DISCUSSION

This study reports on the development of a MR spectroscopy protocol optimized to study brain metabolism during pediatric CPB surgery. While N = 2 in each group (groups divided by DHCA temperature and initial blood glucose levels) is not sufficient to make definitive comparisons between groups, we believe these numbers to be sufficient to identify key trends, estimate the size of the anticipated effects, and to provide an initial assessment of repeatability between animals.

DHCA surgeries at high temperatures (e.g., ≥28°C) are not clinically viable,³⁸ but were chosen in this study to identify the maximal expected metabolic changes. There is, however, a clear trend towards CPB surgeries with moderate hypothermia (24–28°C) using alternative approaches, such as ACP, wherein blood flow is maintained to the brain but arrested throughout the rest of the body during surgery.^{39 1}H MRS studies of brain metabolic changes during these alternative surgical approaches are a clear next step for this work. Another important limitiation of our results is that these experiments did not assess functional outcome.

Our relatively large voxel was chosen to interrogate a brain volume containing both white and gray matter and to look for global metabolic changes. There are, however, published ¹H MRS studies on hypoxic ischemic encephalopathy that focus on deep gray matter changes for predicting outcomes.^40,41 The ultimate size of our targeted 12 × 12 × 15 mm³ (2.1 cc) voxel was driven primarily by SNR and temporal resolution needs; however, future studies focusing on smaller anatomical regions are certainly warranted. In addition to the larger metabolic peaks, single-voxel 3-T ¹H MRS can reliably quantify more subtle in vivo brain metabolic changes previously deemed challenging to interrogate, including brain glucose concentrations even under hypoglycemic conditions, ATP usage via the conversion of PCr to Cr, and differential changes in Glu and Gln. In this work, the observed metabolic changes during CPB surgery of a neonatal porcine model provide new insights into possible mechanisms of neuronal injury during these procedures.

The combination of the sLASER acquisition and LCModel fitting using experimentally measured metabolite basis sets (with additional corrections to the Cr and PCr basis functions at intermediate temperatures) yielded repeatable findings consistent with the metabolic changes expected during CPB surgery. Differentiation of Cr and PCr spectra was not readily visualized on individual spectra; however, the width of the CH₂ associated line is notably broader than that of the CH₃.

In computing mM concentrations for the ¹H MRS-detected brain metabolites, the implicit assumption was that the concentration of tCr was 8 mM. By using the average value for tCr rather than tCr from each spectrum reported by LCModel, variations in tCr during the overall experiment, and any physiologic changes in tCr during the 4-h experiment, are measured. Replacing individual tCr values with average tCr did not change other metabolite estimates over the course of a given experiment.

With respect to temperature-dependent metabolite basis spectra, the approach of using linear interpolated basis sets for Cr and PCr within our targeted 20–37°C temperature range was likely sufficient because of an absence of spin–spin coupling in these two metabolites. However, linear interpolation of selected peaks is likely inadequate for more complicated spectra, such as those for Glu and Gln. Because 28°C is now a target bypass temperature for many pediatric cardiac surgeries,⁴² experimental basis sets may need to be expanded to more temperatures to achieve the most accurate results.

There are some further caveats with respect to LCModel fitting (likely also common to other similar spectral fitting software). The contributions from baseline and macromolecules present a potential source of bias, along with linewidth and SNR (see Marjańska et al.⁴³). For this report, estimates of baseline and macromolecules are limited to the default LCModel software (version 6.3–1 J). On average, metabolite standard errors were approximately two times larger than the CRLBs predicted by LCModel, probably because of contributions from both additive noise and physiological processes. More significantly, as pointed out by Kreis,²⁸ the choice of % coefficient-of-variance (%CVs) as a metric of performance, as often used in the literature (e.g., many researchers reject metabolite estimates with %CVs > 25%²⁸), would have been a poor choice in the data presented here. For example, the calculated %CVs for PCr and Glu at the end of CA arrest were routinely greater than 25%, yet time curves for these metabolites yielded data that were highly consistent with the expected metabolic changes associated with hypoxia. Finally, for the data presented here, all metabolites were quantified with respect to their ratio to brain tCr and unadjusted for differential in vivo metabolite relaxation times, which could also be temperature dependent. Although ¹H MRS studies at short echo times such as 35 ms are certainly less relaxation time-weighted than studies at 168 or 288 ms, without more detailed measurements of metabolite relaxation times in pig brain, we cannot exclude that some of the observed changes could include contributions from relaxation effects. The tCr values showed considerable temporal stability for all animals; however, concentrations based on ratios to tissue water represent an alternative approach to absolute quantitation.²⁵

In addition to providing insight into brain metabolism during DHCA surgery and recovery, an additional goal of these studies was to assess which metabolic processes can be reliably interrogated by ¹H MRS on a 3-T clinical scanner by taking advantage of the perturbation-response experimental paradigm. Specifically, cutting off blood flow to the brain is expected to induce hypoxia with a compensatory increase in anaerobic metabolism, which was indeed observed via CA-induced loss of brain glucose, increase in Lac, and broadening of the residual water peak. Where previous reports by Joers et al. demonstrated robust ¹H MRS measurements of brain glucose under euglycemic and hyperglycemic conditions,²⁷ here we found that brain glucose could also be reliably measured under hypoglycemic conditions. Further, Joers et al. only reported the sum of Glc + Tau because of a high negative correlation between these metabolites, as reported by LCModel. We observed no changes in ¹H MRS-measured brain Tau values in any of our experiments (data not shown) despite the significant changes in Glc, suggesting that these two metabolites may be reliably quantified independently.

The finding of increased Lac during DHCA was consistent with previous reports of high Lac levels as measured by microdialysis⁴⁴; however, the dynamic MRS measurements also permitted measurement of both the Lac production rate and Lac/Glc ratios. Although reduction of brain glucose and increase of Lac were generally consistent with previous DHCA studies,⁴⁴ we observed a large deviation during CA from the stoichiometric predicted glycolytic value of ΔLac/ΔGlc of 2, whereby, under anaerobic conditions, each Glu molecule should generate two Lac molecules. This generation of excess Lac suggests the utilization of an alternative energy source beyond Glu, possibly glycogen. Indeed, Öz et al. reported human brain glycogen metabolism during and after hypoglycemia, although these data were reported for under normal circulatory conditions.⁴⁵

Because of their small spectral pattern differences, in vivo 3-T ¹H MRS measurements of PCr and Cr are typically only reported as a sum (i.e., tCr = PCr + Cr). Caution regarding individually quantifying these metabolites is warranted, due to (1) Cr and PCr not being readily distinguished in the raw spectra (see Figure 2), and (2) the significant changes in shape of the residual water peak with hypoxia (as shown in Figure 6) potentially biasing metabolite estimates. In the studies reported here, careful generation of appropriate LCModel basis functions enabled the successful differentiation of these important metabolites. Support for this claim is provided by observation of the expected conversion of PCr to Cr (and in the process maintaining ATP levels⁴⁶), using conditions whereby energy demands may be unmet by declining Glu levels, as occurs during CA. Exchange rates of creatine kinase metabolites have been quantified using chemical exchange saturation transfer MRI³⁰; however, here we report in vivo PCr changes measured with ¹H rather than via less sensitive ³¹P spectroscopy.^47,48 We also observed a trend towards a temperature-dependent decreasing PCr-to-Cr conversion rate, but more data are needed to accurately quantify this parameter.

The finding of a consistent post-CA rise across animals in the brain Gln/Glu ratio supports the claim that 3-T ¹H MRS can reliably differentiate Glu and Gln signals, rather than being limited to reporting Glu + Gln = Glx values.^25,49 We should note, however, that from simulations (not shown), LCModel appears to overestimate both Gln and Glu at lower SNR, due to inadequate estimation of the baseline. Unlike Glu and Gln, the macromolecule baseline also appears to be temperature sensitive, and poses an uncertainty. Fortunately, the Gln/Glu ratio should behave relatively well, but baseline and SNR biases make an accurate estimate of errors difficult.

Moreover, our Gln/Glu finding may have important pathophysiological implications. Under normal conditions, excess blood ammonia is eliminated by action of the liver. However, excess ammonia is known to be highly neurotoxic,⁵⁰ and the brain processes excess ammonia primarily through the conversion of Glu to Gln,⁵¹ with a secondary pathway via the generation of Ala.⁵² Hence, the post-CA increase in the Gln/Glu ratio observed in our studies suggests that a bolus of ammonia may be reaching the brain when the CPB pump is restarted post-CA.

Metabolically active gut bacteria are a primary source of the body’s ammonia production.⁵¹ One hypothesis for the source of the observed apparent brain ammonia spike in this pig model is that it could arise from anaerobic bacteria in the gut, which continues to produce ammonia during CA. Temporarily impaired due to an absence of blood flow during CA, the liver may be unable to perform its normal ammonia-detoxifying metabolic role. On reflow, the high level of ammonia in the gut and portal vein could temporarily overwhelm elimination by the liver and reach the brain, where it is converted to potentially toxic levels of Gln.

If this is a major driving factor of the observed increase in the brain Gln/Glu ratio, one would expect the effect to increase with increasing DHCA temperature, which is consistent with the data shown in Figure 4. For a given blood ammonia level, the amount of ammonia reaching the brain is also expected to be a function of the difference in blood and brain pH levels. A decrease in brain pH relative to blood may lead to higher ammonia concentrations entering the brain.⁵³ Accordingly, brain Lac concentrations may also play an important role during the immediate post-CA time period, and could explain our observed dependence on the Gln/Glu ratio on initial brain glucose levels and the corresponding Lac production, as shown in Figure 5. We have limited preliminary data (not shown) from a small number of animals showing a rise in blood ammonia levels post-CA, but significantly more data are needed to confirm this finding.

The observed broadening of the residual water signal during CA, which likely arises from B₀ inhomogeneities, also deserves some discussion. Following the theory developed by Springer et al.,³⁷ the theoretical powder pattern from fully deoxygenated intravascular water is shown in Figure 6C. We believe this may be the first in vivo observation of the predicted powder pattern from deoxygenated vessels. However, there are some caveats. Using the residual water peak to quantify absolute brain hypoxia levels may be limited, as our observed data were not fully consistent with the effect predicted by Springer et al. Quantitatively, our observed data were best fit assuming an O₂ saturation of 0% and 0.6 hematocrit; however, the blood hematocrit as measured from samples taken from blood exiting the CPB pump was ~0.3. The effective hematocrit for brain capillaries likely differs from blood hematocrit values,⁵⁴ and this also needs to be included in any quantitative calculations. O₂ saturation is also temperature dependent, and this should also be taken into account.⁵⁵

Although our findings support dynamic in vivo measurements of cerebral metabolism using single-voxel ¹H MRS as an important tool for generating new insights into the understanding of brain changes during CPB surgery, and offers significant advantages relative to other, more invasive, techniques, these studies have a number of important limitations. As with any animal studies, findings in a preclinical model may not accurately reflect those seen in humans. Moreover, with respect to the key clinical question of identifying sources of surgery-related brain injury, these preliminary studies provided no direct measures of neuronal injury or long-term outcomes. Further, we have only presented data for CPB using DHCA, and there are a number of other proposed and in-practice surgical methods for these patients, many of which include varying degrees of continuous cerebral perfusion, which we have not addressed here. Direct measures of neuronal injury, neurodevelopmental deficits, and comparisons with alternative surgical strategies are ongoing.

Abbreviations used:

Ala

alanine

ACP

antegrade cerebral perfusion

CA

circulatory arrest

CPB

cardiopulmonary bypass

Cr

creatine

DHCA

deep hypothermia cardiac arrest

FID

free induction decay

Glc

glucose

Gln

glutamine

Glu

glutamate

Glx

glutamine + glutamate

Lac

lactate

ml

myo-Inositol

PCr

phosphocreatine

PWS

pure water spectrum

RCP

retrograde cerebral perfusion

sLASER

semi-localized by adiabatic selective refocusing

Tau

taurine

tCr

total creatine

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.