You are here

  • Home
  • Archive
  • Volume 104, Issue 4
  • Proton magnetic resonance spectroscopy lactate/N-acetylaspartate within 2 weeks of birth accurately predicts 2-year motor, cognitive and language outcomes in neonatal encephalopathy after therapeutic hypothermia

Article Text

Article menu

Download PDFPDF

Original article
Proton magnetic resonance spectroscopy lactate/N-acetylaspartate within 2 weeks of birth accurately predicts 2-year motor, cognitive and language outcomes in neonatal encephalopathy after therapeutic hypothermia
  1. Subhabrata Mitra1,
  2. Giles S Kendall2,
  3. Alan Bainbridge3,
  4. Magdalena Sokolska3,
  5. Mary Dinan2,
  6. Cristina Uria-Avellanal1,
  7. David Price3,
  8. Katie Mckinnon2,
  9. Roxana Gunny2,
  10. Angela Huertas-Ceballos2,
  11. Xavier Golay4,
  12. Nicola J Robertson1,5
  1. 1 Institute for Women’s Health, University College London, London, UK
  2. 2 Women and Children, University College London NHS Foundation Trust, London, UK
  3. 3 Medical Physics and Engineering, University College London NHS Foundation Trust, London, UK
  4. 4 Institute of Neurology, Queen Square, University College London, London, UK
  5. 5 Division of Neonatology, Sidra Medicine, Doha, Qatar
  1. Correspondence to Professor Nicola J Robertson, Institute for Women’s Health, University College London, London WC1E 6HX, UK; n.robertson{at}ucl.ac.uk

Abstract

Objective Brain proton (1H) magnetic resonance spectroscopy (MRS) lactate/N-acetylaspartate (Lac/NAA) peak area ratio is used for prognostication in neonatal encephalopathy (NE). At 3 Tesla in NE babies, the objectives were to assess: (1) sensitivity and specificity of basal ganglia and thalamus (BGT) 1H MRS Lac/NAA for the prediction of Bayley III outcomes at 2 years using optimised metabolite fitting (Tarquin) with threonine and total NAA; (2) prediction of motor outcome with diffusion-weighted MRI; (3) BGT Lac/NAA correlation with the National Institute of Child Health and Human Development (NICHD) MRI score.

Subjects and methods 55 (16 inborn, 39 outborn) infants at 39w+5 d (35w+5d–42w+0d) with NE admitted between February 2012 and August 2014 to University College London Hospitals for therapeutic hypothermia underwent MRI and 1H MRS at 3T on day 2–14 (median day 5). MRIs were scored. Bayley III was assessed at 24 (22–26) months.

Results 16 babies died (1 inborn, 15 outborn); 20, 19 and 21 babies had poor motor, cognitive and language outcomes. Using a threshold of 0.39, sensitivity and specificity of BGT Lac/NAA for 2-year motor outcome was 100% and 97%, cognition 90% and 97% and language 81% and 97%, respectively. Sensitivity and specificity for motor outcome of mean diffusivity (threshold 0.001 mm2/s) up to day 9 was 72% and 100% and fractional anisotropy (threshold 0.198) was 39% and 94%, respectively. Lac/NAA correlated with BGT injury on NICHD scores (2A, 2B, 3).

Conclusion BGT Lac/NAA on 1H MRS at 3T within 14 days accurately predicts 2-year motor, cognitive and language outcome and may be a marker directing decisions for therapies after cooling.

  • imaging
  • neonatology
  • neurodisability
  • neurodevelopment

https://doi.org/10.1136/archdischild-2018-315478

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    What is already known on this topic?

    • Lactate/N-acetylaspartate peak area ratio acquired in the first 30 days in non-cooled babies with neonatal encephalopathy predicts motor outcome with a sensitivity of 82% and specificity of 95% and is more accurate and stable than mean diffusivity (MD).

    • Magnetic resonance spectroscopy is easy to acquire on modern MRI scanners. Peak area ratios are simple calculate and require only a single acquisition.

    • Improved metabolite peak definition may be seen at 3T versus 1.5T.

    What this study adds?

    • Basal ganglia and thalamus (BGT) Lac+Thr/tNAA acquired at 3T within 14 days after birth in babies who have been cooled predicts Bayley III motor, cognitive and language 2-year outcomes with a high degree of accuracy.

    • Higher field strength (3T) and inclusion of threonine with lactate in the metabolite basis set may contribute to better prediction.

    • BGT Lac+Thr/tNAA had improved sensitivity and specificity for motor outcomes than MD and fractional anisotropy during the period up to day 9 before MD pseudonormalisation.

    Background

    Intrapartum-related neonatal encephalopathy (NE) is a major healthcare problem across the world.1 The incidence of NE in Western Europe and North America is around 1–3/1000 term births depending on the definitions used.2 Although improvements in care might prevent NE and neonatal death in some cases,3 many cases cannot be prevented and therapies are limited. Therapeutic hypothermia (HT) initiated within 6 hours of birth improves outcome, yet despite this therapy 44%–53% of infants with NE die or suffer moderate to severe disabilities including cerebral palsy, developmental delay, epilepsy and visual impairment.4

    MRI is the imaging modality of choice for the assessment of injury pattern, severity and prognostication in NE with an optimal timing between 5 and 14 days. The pattern and severity of the National Institute of Child Health and Human Development (NICHD) Neonatal Research Network (NRN) MRI score5 correlates with the outcome of death or disability and with disability among survivors. In a nested MRI study in the TOBY cooling trial, major MRI abnormalities in the first 4 weeks after birth predicted death or severe disability at 18 months in cooled babies with a sensitivity of 88% and a specificity 82%.6 Proton (1H) magnetic resonance spectroscopy (MRS) is increasingly used as an independent quantitative tool for clinical prognostication in NE and has been used in the TOBY-Xenon clinical neuroprotection trial as a surrogate outcome measure.7 1H MRS in the neonatal brain has prominent signal peaks related to the presence of N-acetylaspartate (NAA), choline, creatine and lactate (Lac). Metabolite ratios (the signal amplitude of one metabolite vs another) are typically calculated for voxels positioned in the basal ganglia and thalamus (BGT). In the precooling era, a meta-analysis demonstrated that a Lac/NAA threshold of 0.29 (0.24 to 0.4) at 1.5T had a sensitivity of 82% and specificity of 95% for predicting an abnormal outcome8 and was more accurate than conventional MRI and diffusion-weighted imaging (DWI).

    Interest is now moving towards the tertiary phase of injury with therapies that can reduce neuroinflammation and improve regeneration and repair.9 A robust marker of brain injury and prediction of outcome following HT will be essential to detect babies who are likely to have adverse outcomes despite having had cooling therapy. An increased lactate and reduced NAA on MRS (translating to a high Lac/NAA peak area ratio) suggest brain mitochondrial impairment and impaired oxidative metabolism which may be amenable to late therapies. Unlike mean diffusivity (MD), which pseudonormalises around day 10 in babies who have been cooled,10 Lac/NAA remains stable over the first 2 weeks after birth and has potential to be used as a quantitative marker of brain injury after HT. Widespread hospital access to 3T MRI is now possible; a higher field strength as well as improved MRS analysis techniques are likely to further strengthen accuracy of Lac/NAA.

    1H MRS measurements of brain Lac have focused on the methyl proton resonance at 1.31 ppm with an optimised long echo time (TE) to reduce contamination with lipid signal.11 The methyl group of threonine (Thr), an essential amino acid present in the brain, also resonates close by at approximately 1.32 ppm.12 The proximity of the methyl resonances of Lac and Thr mean that with conventional MRS they are not independently resolvable in in vivo spectra.13 A potential benefit of including Thr, in addition to lactate, in the spectral analysis is improvement in the spectral fit in the region around 1.3 ppm in cases where lactate is raised. Similarly, including both NAA and NAA glutamate (NAAG, an abundant neurotransmitter released from axonal terminals with neuronal activity and hydrolysed to NAA and glutamate; NAA+NAAG= total NAA (tNAA)) improves fitting in the region around 2.0 ppm at field strength of 3T.14

    The objectives of this study were, in babies with NE who have been cooled, to assess: (1) BGT 1H MRS Lac+Thr/tNAA sensitivity and specificity at 3T for motor, cognitive and language Bayley III outcomes at 2 years; (2) prediction of motor outcome of Lac+Thr/tNAA versus other MRI biomarkers (MD and FA) before pseudonormalisation; (3) BGT Lac+Thr/tNAA correlation with the NICHD MRI score.

    Subjects and methods

    Approval was given from the University College London Hospital Research Ethics Committee for this retrospective service improvement project. Anonymised data that are routinely collected in our centre were used and requirement for informed consent was waived.

    Infants treated with therapeutic HT for moderate to severe NE in University College London Hospitals between February 2012 and August 2014 were reviewed. Infants with perinatal stroke, major congenital malformations and metabolic or congenital infection were excluded from the study. Neurodevelopmental follow-up data at 2 years were available in all 58 term newborn infants who were cooled. Of the 58 infants, 3 were excluded as they were part of another multicentre trial protocol. Fifty-five term neonates with a mean (range) gestational age of 39+5 (35+5–42+0) weeks and a birth weight of mean (±SD) 3260 (566) g were included. Thirty-nine infants were inborn and 16 were outborn. These neonates fulfilled criteria for HT based on the National Institute for Health and Care Excellence criteria.15 In five deliveries, there was uterine rupture and in two deliveries there was placental abruption. The mean±SD arterial gas (cord or within 60 min of birth) was 6.9±0.2 and base deficit was −15±7. Mean Apgar scores 2±2 at 1 min, 4±2 at 5 min and 6±3 at 10 min. All neonates required resuscitation at birth. Seven infants had a birth weight <10th centile. All infants received intravenous morphine sulfate (up to 30 μg/kg/hour) during therapeutic HT. All infants were monitored with continuous video electroencephalogram throughout cooling and rewarming. Following clinical or electrographic seizure activity, infants received phenobarbitone as first-line (up to 40 mg/kg total), phenytoin as second-line (20 mg/kg) and midazolam (bolus followed by infusion) as third-line antiepileptic drug. Seizure burden was not assessed as part of this study.

    MRI and spectroscopy

    Magnetic resonance imaging

    Scanning was performed at a median (range) age of 5 (2–14) days. In 47 infants, the MRI was performed after the 72-hour HT and following rewarming to normothermia; in 8 infants, the MRI was done during cooling to inform the direction of care. Scanning was performed either during natural sleep or after sedation with a morphine infusion. Most neonates (48) were ventilated during scanning and were transported and studied in a Lammers Medical MR Conditional transport incubator (LMT Medical Systems, Luebeck, Germany) with gentle head restraint. Throughout the examination, neonates were monitored by using MR-compatible pulse oximetry and ECG and supervised by an experienced neonatologist trained in clinical MR imaging safety.

    MRI was performed on a 3T MR system (Philips Medical Systems, Best, The Netherlands). The scanning protocol included T1-weighted imaging (inversion-prepared 3D gradient echo read-out: inversion time (TI)=1465 ms, repetition time (TR)=17 ms, TE=4.6 ms, sagittal slice thickness=1 mm, in-plane resolution=0.82×0.97 mm), T2-weighted imaging (T2W; coronal and axial, turbo spin echo: echo train length=11, TR=10 721 ms, TE=130 ms, slice thickness=3 mm, in-plane resolution=0.50×0.52 mm), DTI (32 directions, b=750, echo planar imaging (EPI) read-out: TR=7500 ms, TE=49 ms, slice thickness=2 mm, in-plane resolution=2.0×2.04 mm).

    Images were scored using the NICHD NRN MRI5 score by RG who was blind to the perinatal history and clinical outcome. NICHD NRN MRI score has been validated to predict death or IQ at 6–7 years of age following HT for NE. The score describes pattern 0 (normal MRI), 1A (minimal cerebral lesions), 1B (extensive cerebral lesions), 2A (basal ganglia thalamic, anterior limb of internal capsule (ALIC) or posterior limb of internal capsule (PLIC), or watershed infarction), 2B (2A with cerebral lesions) and pattern 3 (hemispheric devastation).

    Magnetic resonance spectroscopy

    MRS was performed using PRESS (15×15×15 mm voxel position on the left thalamus/basal ganglia, TR=2288 ms, TE=288 ms, a dynamic series of 16 spectra were acquired each with eight averages). The dynamic spectra were summed off-line after phase and frequency correction and rejection of motion-corrupted data. MRS analysis was performed using Tarquin.16 The basis set included Thr11 but did not include lipids or macromolecules. When the signal around 1.3 ppm is high, we have observed that the spectral fit in this region often has a significant residual. Including Thr in the basis set typically improves the fit in this region of the spectrum (figure 1). The ratio of lactate (Lac) plus Thr to total NAA (NAA+NAAG), (Lac+Thr)/tNAA, was calculated using the amplitudes of the fitted components. The position of the voxel and representative 1H MRS spectra in normal and abnormal outcomes is shown in figure 2.

    Figure 1
    Figure 1

    (A) Representative fitting of the 1H magnetic resonance spectrum using Tarquin and a basis metabolite set that includes both lactate and threonine. The ‘residual’ once the fitting has occurred is minimal (circle). (B) Removing threonine from the basis metabolite set leads to a large ‘residual’ to the fit (circle), especially if the lactate signal in the region of 1.3 ppm is elevated.

    Figure 2
    Figure 2

    Position of the thalamic 1H magnetic resonance spectroscopy voxel for spectral acquisition (sagittal A and axial B) from a neonate with normal outcome (Lac+Thr/tNAA 0.19) (C) and a neonate with poor outcome (Lac+Thr/tNAA 0.62) (D) after HT for NE. Cho, choline; Cr, creatine; HT, hypothermia; Lac, lactate; NE, neonatal encephalopathy; tNAA, total N-acetylaspartate; Thr, threonine.

    Mean diffusivity and fractional anisotropy

    The DTI volumes were analysed using tools from the FSL brain imaging software library17 and the DTI-TK toolbox.18 Non-brain tissues were first removed using image segmentation software and the FSL-EDDY tool used to correct the DTI volumes for eddy current induced distortions in the image and subject movement. After this, the diffusion datasets were inspected and those with severe uncorrectable motion artefacts were rejected from further analysis. There were 49 usable datasets. These were then processed using the FSL-FDT toolbox, which contains tools to fit the diffusion tensors from the DTI data and calculate MD and fractional anisotropy (FA) maps. MD and FA values for the BGT (figure 3) were segmented using the atlas described below.

    Figure 3
    Figure 3

    Region of interests (ROIs) for mean diffusivity (MD) and fractional anisotropy (FA) in basal ganglia and thalamus (BGT) were defined automatically based on both an atlas-based probabilistic tissue segmentation and a joint multi-atlas label propagation and fusion of 50 neonatal brain regions.

    Regional analysis

    Regional MD and FA were computed in BGT using BGT masks generated automatically based on both atlas-based probabilistic tissue segmentations of five tissue classes19 and a joint multi-atlas label propagation and fusion of 50 manual segmentations of neonatal brain regions.20 T2W atlas images were non-linearly registered to T2W image of each neonate, and the probabilistic maps and parcellations were propagated to individual subject space using the Nifty-Reg package.21 Probability maps were used to segment T2W images by the AdaPT22 algorithm using Nifty-Seg package.23 T2W subject images were coregistered to the MD maps using rigid registration, with tissue segmentations and brain parcellations propagated to MD space and resampled using point spread function matching.24 BGT mask was created by combining propagated parcellations of the thalami, subthalamic, caudate and lentiform nuclei.20 The internal capsule was not defined as part of the segmentation and no additional steps were undertaken to separate this structure. All generated segmentations were checked visually for correct anatomical correspondence. Voxels that were obviously mis-assigned (typically within vessels and less than 10 voxels per dataset) were manually edited if necessary. Finally, MD and FA were quantified within BGT segmentation mask.

    Tract-based spatial statistics (TBSS)

    TBSS is a method allowing for the voxel-wise statistical analysis of the white matter (WM) tract anatomy between subjects.25 It is therefore distinct from the measurement of FA within a specific brain region such as the BGT as described above. TBSS was performed on the corrected DTI volumes using an integration of the DTI-TK diffusion toolbox with the FSL-TBSS pipeline. This aimed to combine spatial normalisation provided by DTI-TK together with voxel-wise statistical inference for WM anatomy provided by TBSS. DTI-TK allows the tensor registration of each subject to a population-specific template.

    An initial template was formed from five well-aligned datasets. This was then refined through three stages of alignment: rigid, affine and deformable. Tensor-based registration has been shown to provide an improved alignment of WM tracts compared with using the derived FA maps in the standard TBSS pipeline.26 A population-specific template also may avoid biases associated with templates based on the most typical subject. The template and spatially normalised datasets were then used to create a mean FA skeleton and a 4D FA map, respectively. These were introduced into the final stages of the standard TBSS pipeline. This allowed for the voxel-wise analysis of WM FA between outcome groups (normal vs abnormal motor outcome). The FA threshold used in this study was 0.15. No additional steps were taken to remove non-WM voxels from the FA skeleton.

    Clinical data and outcome

    Neurodevelopmental outcome was assessed at 3, 6, 12 and 24 months as per local and regional protocols. At 24 (22–26) months, the Bayley Scales of Infant Development, Third Edition was used for assessment (Bayley III).27 The Bayley III has five scales (cognitive, language (receptive and expressive), motor (fine and gross motor), social emotional and adaptive behaviour). The Bayley III independently assesses cognitive and language skills (previous Bayley editions combined cognitive and language into one mental scale). Cognition’s raw score is transformed into a composite score. Language and motor raw scores are analysed independently for each subscale or as a composite score. Death or a composite score of <85 on language, motor or cognitive scales at 2 years of age was considered abnormal.

    Statistical analysis

    The values of (Lac+Thr)/tNAA peak area ratio are not normally distributed in this cohort. Therefore, in order to use parametric statistics on this dataset, (Lac+Thr)/tNAA was transformed to log10[(Lac+Thr)/tNAA] prior to analysis.

    The relationship between log10[(Lac+Thr)/tNAA] and the probability of motor, cognitive and language outcome was assessed using logistic regression modelling to each outcome separately. For each model, the value of log10[(Lac+Thr/tNAA)] that gave the optimum sensitivity and specificity was chosen for classification of ‘abnormal’ for values above the threshold. From this value, the optimum threshold of (Lac+Thr)/tNAA peak area ratio for classification of ‘abnormal’ could then be calculated. The relationship between MD and also WM FA and the probability of motor outcome was also assessed using logistic regression up to 9 days (before pseudonormalisation) and for all scans (up to 14 days).

    One-way analysis of variance (ANOVA) was used to compare (1) mean Lac+Thr/tNAA levels in ‘abnormal’ versus ‘normal’ outcome groups for each of motor, language and cognitive outcomes and (2) mean Lac+Thr/tNAA levels across NICHD MRI classification groups 2A, 2B and 3.

    Voxel-wise analysis of WM FA, performed using TBSS, is presented by colour-coding voxels with a significant difference (p<0.05) in FA between outcome groups.

    Results

    Of the 55 infants included, 16 infants died (1 inborn, 15 outborn), 20 infants had abnormal motor outcome, 19 had abnormal cognitive outcome and 21 had abnormal language outcome by 2 years of age (table 1). There were 18 infants who had abnormal outcomes in all three domains (motor, language and cognitive). Two other infants had abnormal motor outcome (total 20) without any cognitive or language problems. One other infant had an abnormal cognitive outcome (total 19), without any other motor or language problem. Three other infants had an abnormal language outcome (total 21) without any motor or cognitive problems.

    View this table:
    Table 1

    Patient characteristics

    Magnetic resonance imaging

    NICHD MRI score was 0 in 20 babies, 1A in 7 babies, 1B in 3 babies, 2A in 3 babies, 2B in 7 babies and 3 in 15 babies.

    1H MRS

    The Lac+Thr/tNAA and log10[(Lac+Thr)/tNAA] ratios by age at scan are shown in online supplementary figure 1. Using a log10[(Lac+Thr)/tNAA] cut-off threshold of −0.41, the sensitivity and specificity for predicting motor outcome in the Bayley III assessment was 100% and 97%, respectively, with an area under the curve (AUC) of 0.997 (figure 4A). This threshold predicted cognitive outcome with sensitivity of 90% and specificity of 97%, AUC of 0.967 (figure 4B), and language outcome with a sensitivity of 81% and specificity of 97%, AUC of 0.913 (figure 4C) (table 2). Comparing the group means of the normal and abnormal outcomes grouped by motor, cognitive and language outcomes, respectively, there were significant differences in the log10[(Lac+Thr)/tNAA] (p<0.0001 for all comparisons) (figure 4D–F). The log10[(Lac+Thr)/tNAA] threshold of −0.41 equates to an untransformed (Lac+Thr)/tNAA threshold of 0.39.

    Figure 4
    Figure 4

    Logistic regression analysis of log10[(Lac+Thr)/tNAA for outcome prediction. Using a (Lac+Thr)/tNAA cut-off threshold of 0.39 (equating to −0.41 for Log 10[(Lac+Thr)/tNAA], Tarquin analysis gave one false positive (A) for predicting poor motor outcome at 2 years (sensitivity 100%, specificity 97%, AUC 0.997). Analysis for cognitive outcome gave one false positive and two false negative (sensitivity 90%, specificity 97%, AUC 0.967 (B). Analysis for language outcome gave one false positive and four false negatives (sensitivity 81%, specificity 97%, AUC 0.913 (C). The normal and abnormal group means of log10[(Lac+Thr)/tNAA] were significantly different for motor (D), cognitive (E) and language (F) domains (p<0.0001). AUC, area under the curve; Lac, lactate; tNAA, total N-acetylaspartate; Thr, threonine.

    View this table:
    Table 2A

    The logistic regression model for all 55 subjects scanned between 2 and 14 days after birth

    MD and FA

    For all scans up to day 14, BGT MD predicted motor outcome with a sensitivity of 65% and specificity of 100% with an AUC of 0.769. BGT FA predicted motor outcome with a sensitivity of 45% and a specificity of 91% with an AUC of 0.544 (table 2A). Some correlation existed between Lac+Thr/tNAA and BGT MD (R2=0.51) but not with BGT FA (R2=0.01).

    Limiting the analysis to scans done up to day 9 after birth, before the return of MD to normal (pseudonormalisation), BGT MD predicted motor outcome with a sensitivity of 72% and specificity of 100% with an AUC of 0.819 (table 2B, figure 5A). BGT FA predicted motor outcome with a sensitivity of 39% and a specificity of 94% with an AUC of 0.507 (figure 5B). Comparing the means of the normal and abnormal groups for motor outcomes, there was a significant difference between MD (p=0.001) (figure 5C) but not for FA (p=0.707) (figure 5D).

    View this table:
    Table 2B

    The logistic regression model for subjects scanned between day 2 and 9 (n=50) after birth

    Figure 5
    Figure 5

    Logistic regression analysis of mean diffusivity (MD) (mm2/s) and fractional anisotropy (FA) restricted to scans acquired day 1–9 before pseudonormalisation for motor outcome prediction only (50 of 55 scans). MD threshold of 0.001 mm2/s gave a sensitivity of 72%, specificity of 100%, area under the curve (AUC) of 0.819) (A); FA threshold of 0.198 gave a sensitivity of 39%, specificity of 94% and AUC of 0.507) (B). The normal and abnormal group motor outcome means of MD were significantly different (p<0.001) (C); for FA, there was no difference (p=0.707) (D).

    Relation between Lac+Thr/tNAA and MRI score

    ANOVA to compare BGT mean log10[(Lac+Thr)/tNAA] across the different MRI scores showed an increase in BGT mean log10[(Lac+Thr)/tNAA] with increasing BGT severity on the NICHD MRI score from score 2A onwards (pattern 2A, any BGT, ALIC or PLIC involvement or watershed infarction noted without any other cerebral lesions); pattern 2B, involvement of either BGT, ALIC or PLIC or area of infarction and additional cerebral lesions; and pattern 3, cerebral hemispheric devastation (figure 6).5

    Figure 6
    Figure 6

    BGT Lac+Thr/tNAA correlated with BGT involvement in NICHD MRI score—scores 2A, 2B, 3. The BGT injury on MRI is indicated by the dotted box. Analysis of variance to compare BGT mean log10[(Lac+Thr)/tNAA] revealed significant differences between score 2A and B (p=0.0003) and score 3 and 2B (p=0.001). BGT, basal ganglia and thalamus; Lac, lactate; NICHD, National Institute of Child Health and Human Development; tNAA, total N-acetylaspartate; Thr, threonine.

    Tract-based spatial statistics

    There were 541 significantly lower FA voxels out of a total of 69 851 in the mean FA skeleton in the abnormal compared with the normal motor outcome group. There were no other statistically significant associations. The mean FA map was produced through the registration of all subjects’ diffusion datasets using DTI-TK. The mean FA skeleton was overlaid on the mean FA map, and the results of the statistical analysis for Bayley III motor outcome are shown in online supplementary figure 2. Voxels wherein infants with poor motor outcome had lower FA (p<0.05) are shown in red-yellow. On the axial view, there were significantly lower voxels in the ALIC and PLIC in the abnormal motor outcome group. On the coronal view, there were significantly lower voxels in optic radiations in the abnormal motor outcome group.

    Discussion

    This study shows that a 1H MRS spectrum acquired at 3T from the BGT using a long TE (288 ms) for Lac+Thr/tNAA peak area ratio within 14 days of birth in babies with moderate to severe NE who were cooled accurately predicts 2-year outcome. Using a threshold of 0.39, BGT Lac+Thr/tNAA peak area ratio predicts motor, cognition and language outcomes (sensitivity and specificity of 100% and 97%, 90% and 97%, 81% and 97%, respectively). Normal and abnormal motor, cognitive and language outcome groups had significantly different mean BGT log10[(Lac+Thr/tNAA)] levels (p<0.001). BGT log10[(Lac+Thr/tNAA)] correlated with BGT involvement in the NICHD MRI score (scores 2A, 2B, 3).

    In the 2010 meta-analysis of Lac/NAA peak area ratio, including 32 studies (860 infants) with NE who were not cooled, sensitivity and specificity was 82% and 95% for motor outcome for data acquired between day 1 and 30.8 In our study at 3T in babies in the cooling era, BGT Lac+Thr/tNAA had a sensitivity and specificity for predicting motor outcome of 100% and 97%, respectively. Therefore, BGT Lac+Thr/tNAA acquired within 14 days of birth and after HT remains a robust predictor of motor outcome in babies who have had HT. For the first time BGT Lac+Thr/tNAA, using a threshold of 0.39, is shown to predict cognitive and language outcomes as well as motor outcomes (table 2 and figure 3). The improved accuracy and prediction across other domains may be attributed to the improved spectral fitting with inclusion of Thr, acquiring MRS within 14 days and the use of higher field strengths allowing for a greater spectral separation. The prediction of specific neurodevelopmental domains with BGT Lac+Thr/tNAA further emphasises the central importance of injury to the BGT in determining outcome across several domains. The influence of BGT on outcome was also seen in the correlation between Lac+Thr/tNAA and the BGT involvement in the NICHD MRI score (pattern 2A, 2B, 3 BGT involvement). The importance of the BGT predicting outcome was observed in the MRI score described by de Vries and colleagues; in their study only the grey matter score was included, suggesting this location most closely determines outcome in NE.28 The sensitivity and specificity for 2-year neurodevelopmental outcome of this MRI score was 92% and 95% for the retrospective cohort and 42% and 98% for a prospective cohort. There was no 1H MRS included in the prospective cohort, which may explain the reduced accuracy.28

    1H MR spectra were analysed using Tarquin, a freely available spectroscopy analysis tool that fits signals in the time domain using a metabolite basis set. Tarquin is similar to LCModel in that it uses a basis set of metabolite signals to fit to the acquired spectrum. Tarquin has been shown to be robust, to produce similar results to LCModel in healthy volunteers16 and to have comparable errors to LCModel using Monte Carlo simulations.13 It has been used for the quantitative evaluation of brain and brain-tumour metabolites in humans29–31 and in an animal model.32

    Thr was included in the basis set for the spectral analysis. Thr is an essential amino acid, which is important for protein synthesis and folding. Foods such as meat, milk, beans and fish are a source of Thr. Thr has been detected in human brain on autopsy30 and also by 1H MRS.16 Thr brain levels are known to be affected by protein intake with low protein diets leading to reduced Thr levels in experimental rats.31 Chronic energy deficiency however was not associated with changes in brain Thr levels.33 Nevertheless, we have observed that including Thr in the metabolite basis set improves the signal fit in the region around 1.3 ppm, especially when the signal in this region is large (figure 1). Lac and Thr are not easily resolvable in in vivo spectra16 and so the sum of Lac+Thr is the most meaningful quantity to report.34 The inclusion of Thr has only a small effect on the fitting of other metabolites in the basis set other than Lac due to minimal overlap.

    Restricting to scans acquired up to day 9 when pseudonormalisation of MD occurs in cooled babies with NE,10 sensitivity and specificity for motor outcomes for MD (threshold 0.001 mm2/s) was 72% and 100% and FA (threshold 0.198) was 39% and 94%. These values were similar to MD and FA sensitivity and specificity for all scans up to day 14; as only five babies were scanned after day 9, these would have had a small influence. DTI analysed by TBSS has been previously used to show a treatment effect of HT in babies with NE35; babies who were not cooled (and likely to have adverse outcomes) had lower FA in the ALIC, PLIC and optic radiations. These findings are similar to the regional differences in our normal and abnormal motor outcome groups. In a subsequent study, Tusor et al report that infants with NE with unfavourable outcomes after HT had lower FA values (p<0.05) in centrum semiovale, corpus callosum, ALIC, PLIC, fornix, cingulum, cerebral peduncles, optic radiations and inferior longitudinal fasciculus.36 It is notable that the current cohort demonstrates a relatively limited distribution of lower FA after HT compared with that reported by Tusor. The reason this discrepancy is not clear as patient populations, acquisition and TBSS methods for the two studies are similar.

    There are some limitations to the study. First, the threshold of 0.39 Lac+Thr/tNAA was chosen as the threshold for normal/abnormal outcome for motor, cognitive and language outcomes by assessing the value that gave the optimum sensitivity and specificity on logistic regression. This Lac+Thr/tNAA threshold needs to be validated in another cohort of babies with NE. Second, the accuracy of Lac+Thr/tNAA for predicting cognitive and language as well as motor outcome may be influenced by overlap of poor outcome domains within individuals (18 infants had abnormal outcomes in all domains). Thirdly, the highest brain lactate levels were seen in the scans acquired during the first 4 days; this pattern may have been influenced by earlier scans done in the most severely affected infants to help with clinical management decisions. Lastly, the process of pseudonormalisation of MD is a biological continuum and the cut-off of day 9 is somewhat arbitrary and based on a small retrospective study.10 Variation in pseudonormalisation within the study population could therefore have confounded the diffusion-weighted MRI analysis (MD).

    Brain lactate has been observed to persist for months in the brain of babies with adverse outcome after NE37 38; this persisting brain lactate is associated with abnormal MRI and brain alkalosis.38 Mechanisms leading to persisting raised brain lactate include impaired mitochondrial function and oxidative phosphorylation leading to an increase in anaerobic glycolysis, an altered redox state, the presence of phagocytic cells, gliosis, loss of perfusion autoregulation and/or altered buffering mechanisms. This persisting abnormal brain metabolism may reflect tertiary brain injury9 and be a marker of ongoing injury, which may be a target for therapies. NAA is the second most abundant amino acid-related metabolite in the brain after glutamate. Both NAA and NAAG synthesis in neurons is ATP dependent and hypoxia–ischaemia initiates a decline in NAA/NAAG within hours. Much of the NAA synthesised in the neuron is transported to oligodendrocytes where it is a substrate source for myelin synthesis; it is likely that decreases in NAA following NE will lead to a reduced myelination.14 In adult stroke, NAA levels can predict the fate of an acute ischaemic lesion with more accuracy than DWI.39 40 The dissociation between NAA and DWI after hypoxia–ischaemia has been well described; restoration of ADC occurred despite the subsequent infarction of tissue, whereas NAA levels continued to show a decline in the same area.41 Our findings in babies with NE concur with the experience in adult stroke of better accuracy of 1H MRS in predicting outcome compared with MD and FA. We previously described the central importance of NAA in predicting outcome: absolute concentrations of NAA in babies with NE were more accurate than peak area ratios of Lac/NAA and could discriminate control, normal/mild and severe/fatal outcome groups during the first 4 days after birth.42 Time constraints, however, make collection of data for absolute concentration impractical in clinical MRI/MRS studies. Moreover, a significant advantage of peak area ratios, for example, with Lac+Thr/tNAA, is that they depend on both metabolite T2s and concentrations, both of which are pathologically modulated and hence injury severity prediction is improved.42

    In conclusion, we have shown in 55 babies with moderate to severe NE who have undergone HT, that BGT Lac+Thr/tNAA acquired at 3T within 14 days after birth predicts Bayley III motor, cognitive and language outcomes with a high degree of accuracy; the central importance of the BGT in prediction of all three domains was clear. Optimised spectral fitting with Tarquin and higher field strength may contribute to the improved accuracy compared with previous studies. Lac+Thr/tNAA correlated with MRI scores of BGT injury. Lac+Thr/tNAA is a robust marker of neurodevelopmental outcome on which decisions for adjunct therapies in the subacute phase after HT might be based.

    References

    1. ↵
      2. Lee AC ,
      3. Kozuki N ,
      4. Blencowe H , et al
      . Intrapartum-related neonatal encephalopathy incidence and impairment at regional and global levels for 2010 with trends from 1990. Pediatr Res 2013;74(Suppl 1):50–72.doi:10.1038/pr.2013.206
      OpenUrlCrossRefPubMed
    2. ↵
      2. Kurinczuk JJ ,
      3. White-Koning M ,
      4. Badawi N
      . Epidemiology of neonatal encephalopathy and hypoxic-ischaemic encephalopathy. Early Hum Dev 2010;86:329–38.doi:10.1016/j.earlhumdev.2010.05.010
      OpenUrlCrossRefPubMedWeb of Science
    3. ↵
      NPEU. MBRRACE-UK: Mothers and Babies: Reducing Risk from Audits and confidential Enquiries across the UK. https://www.npeu.ox.ac.uk/mbrrace-uk2017.
    4. ↵
      2. Jacobs S ,
      3. Berg M ,
      4. Hunt R , et al
      . Cooling for newborns with hypoxic ischaemic encephalopathy. Cochrane Database Sys Rev 2013;(1):CD003311.
    5. ↵
      2. Shankaran S ,
      3. Barnes PD ,
      4. Hintz SR , et al
      . Eunice Kennedy Shriver National Institute of Child Health and Human Development Neonatal Research Network. Brain injury following trial of hypothermia for neonatal hypoxic-ischaemic encephalopathy. Arch Dis Child Fetal Neonatal Ed 2012;97:F398–404.doi:10.1136/archdischild-2011-301524
      OpenUrlAbstract/FREE Full Text
    6. ↵
      2. Rutherford M ,
      3. Ramenghi LA ,
      4. Edwards AD , et al
      . Assessment of brain tissue injury after moderate hypothermia in neonates with hypoxic-ischaemic encephalopathy: a nested substudy of a randomised controlled trial. Lancet Neurol 2010;9:39–45.doi:10.1016/S1474-4422(09)70295-9
      OpenUrlCrossRefPubMedWeb of Science
    7. ↵
      2. Azzopardi D ,
      3. Robertson NJ ,
      4. Bainbridge A , et al
      . Moderate hypothermia within 6 h of birth plus inhaled xenon versus moderate hypothermia alone after birth asphyxia (TOBY-Xe): a proof-of-concept, open-label, randomised controlled trial. Lancet Neurol 2016;15:145–53.doi:10.1016/S1474-4422(15)00347-6
      OpenUrl
    8. ↵
      2. Thayyil S ,
      3. Chandrasekaran M ,
      4. Taylor A , et al
      . Cerebral magnetic resonance biomarkers in neonatal encephalopathy: a meta-analysis. Pediatrics 2010;125:e382–95.doi:10.1542/peds.2009-1046
      OpenUrlAbstract/FREE Full Text
    9. ↵
      2. Fleiss B ,
      3. Gressens P
      . Tertiary mechanisms of brain damage: a new hope for treatment of cerebral palsy? Lancet Neurol 2012;11:556–66.doi:10.1016/S1474-4422(12)70058-3
      OpenUrlCrossRefPubMedWeb of Science
    10. ↵
      2. Bednarek N ,
      3. Mathur A ,
      4. Inder T , et al
      . Impact of therapeutic hypothermia on MRI diffusion changes in neonatal encephalopathy. Neurology 2012;78:1420–7.doi:10.1212/WNL.0b013e318253d589
      OpenUrlCrossRefPubMed
    11. ↵
      2. Kreis R ,
      3. Hofmann L ,
      4. Kuhlmann B , et al
      . Brain metabolite composition during early human brain development as measured by quantitative in vivo 1H magnetic resonance spectroscopy. Magn Reson Med 2002;48:949–58.doi:10.1002/mrm.10304
      OpenUrlCrossRefPubMedWeb of Science
    12. ↵
      2. Govindaraju V ,
      3. Young K ,
      4. Maudsley AA
      . Proton NMR chemical shifts and coupling constants for brain metabolites. NMR Biomed 2000;13:129–53.doi:10.1002/1099-1492(200005)13:3<129::AID-NBM619>3.0.CO;2-V
      OpenUrlCrossRefPubMedWeb of Science
    13. ↵
      2. Choi C ,
      3. Coupland NJ ,
      4. Kalra S , et al
      . Proton spectral editing for discrimination of lactate and threonine 1.31 ppm resonances in human brain in vivo. Magn Reson Med 2006;56:660–5.doi:10.1002/mrm.20988
      OpenUrlCrossRefPubMed
    14. ↵
      2. Igarashi H ,
      3. Suzuki Y ,
      4. Huber VJ , et al
      . N-acetylaspartate decrease in acute stage of ischemic stroke: a perspective from experimental and clinical studies. Magn Reson Med Sci 2015;14:13–24.doi:10.2463/mrms.2014-0039
      OpenUrl
    15. ↵
      NICE, 2010. National Institute for Clinical Excellence: Therapeutic hypothermia with intracorporeal temperature monitoring for hypoxic perinatal brain injury: guidance. NICE guidelines: interventional procedures http://www.nice.org.uk/nicemedia/live/11315/48809/48809.pdf.
    16. ↵
      2. Wilson M ,
      3. Reynolds G ,
      4. Kauppinen RA , et al
      . A constrained least-squares approach to the automated quantitation of in vivo ¹H magnetic resonance spectroscopy data. Magn Reson Med 2011;65:1–12.doi:10.1002/mrm.22579
      OpenUrlCrossRefPubMed
    17. ↵
      2. Jenkinson M ,
      3. Beckmann CF ,
      4. Behrens TE , et al
      . FSL. Neuroimage 2012;62:782–90.doi:10.1016/j.neuroimage.2011.09.015
      OpenUrlCrossRefPubMedWeb of Science
    18. ↵
      2. Zhang H ,
      3. Avants BB ,
      4. Yushkevich PA , et al
      . High-dimensional spatial normalization of diffusion tensor images improves the detection of white matter differences: an example study using amyotrophic lateral sclerosis. IEEE Trans Med Imaging 2007;26:1585–97.doi:10.1109/TMI.2007.906784
      OpenUrlCrossRefPubMedWeb of Science
    19. ↵
      2. Kuklisova-Murgasova M ,
      3. Aljabar P ,
      4. Srinivasan L , et al
      . A dynamic 4D probabilistic atlas of the developing brain. Neuroimage 2011;54:2750–63.doi:10.1016/j.neuroimage.2010.10.019
      OpenUrlCrossRefPubMedWeb of Science
    20. ↵
      2. Gousias IS ,
      3. Edwards AD ,
      4. Rutherford MA , et al
      . Magnetic resonance imaging of the newborn brain: manual segmentation of labelled atlases in term-born and preterm infants. Neuroimage 2012;62:1499–509.doi:10.1016/j.neuroimage.2012.05.083
      OpenUrlCrossRefPubMed
    21. ↵
    22. ↵
      2. Cardoso MJ ,
      3. Melbourne A ,
      4. Kendall GS , et al
      . AdaPT: An adaptive preterm segmentation algorithm for neonatal brain MRI. Neuroimage 2013;65:97–108.doi:10.1016/j.neuroimage.2012.08.009
      OpenUrl
    23. ↵
    24. ↵
      2. Cardoso M ,
      3. Modat M ,
      4. Vercauteren T , et al.
      , 2015. Scale factor point spread function matching: beyond aliasing in image resampling. Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015: 18th International Conference, Munich, Germany;October 5-9; 675–83
    25. ↵
      2. Smith SM ,
      3. Jenkinson M ,
      4. Johansen-Berg H , et al
      . Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data. Neuroimage 2006;31:1487–505.doi:10.1016/j.neuroimage.2006.02.024
      OpenUrlCrossRefPubMedWeb of Science
    26. ↵
      2. Bach M ,
      3. Laun FB ,
      4. Leemans A , et al
      . Methodological considerations on tract-based spatial statistics (TBSS). Neuroimage 2014;100:358–69.doi:10.1016/j.neuroimage.2014.06.021
      OpenUrlCrossRefPubMed
    27. ↵
      San BNBSoIaTDre. Assessment: ATH, 2006.
    28. ↵
      2. Weeke LC ,
      3. Groenendaal F ,
      4. Mudigonda K , et al
      . A Novel Magnetic Resonance Imaging Score Predicts Neurodevelopmental Outcome After Perinatal Asphyxia and Therapeutic Hypothermia. J Pediatr 2018;192:33–40.doi:10.1016/j.jpeds.2017.09.043
      OpenUrl
    29. ↵
      2. Zarinabad N ,
      3. Abernethy LJ ,
      4. Avula S , et al
      . Application of pattern recognition techniques for classification of pediatric brain tumors by in vivo 3T 1 H-MR spectroscopy-A multi-center study. Magn Reson Med 2018;79:2359–66.doi:10.1002/mrm.26837
      OpenUrl
    30. ↵
      2. Vicente J ,
      3. Fuster-Garcia E ,
      4. Tortajada S , et al
      . Accurate classification of childhood brain tumours by in vivo ¹H MRS - a multi-centre study. Eur J Cancer 2013;49:658–67.doi:10.1016/j.ejca.2012.09.003
      OpenUrlCrossRefPubMedWeb of Science
    31. ↵
      2. Scott J ,
      3. Underwood J ,
      4. Garvey LJ , et al
      . A comparison of two post-processing analysis methods to quantify cerebral metabolites measured via proton magnetic resonance spectroscopy in HIV disease. Br J Radiol 2016;89:20150979.doi:10.1259/bjr.20150979
      OpenUrl
    32. ↵
      2. Doblas S ,
      3. He T ,
      4. Saunders D , et al
      . In vivo characterization of several rodent glioma models by 1H MRS. NMR Biomed 2012;25:685–94.doi:10.1002/nbm.1785
      OpenUrlCrossRefPubMedWeb of Science
    33. ↵
      2. Kaladhar M ,
      3. Rao BS
      . Experimental protein and energy deficiencies: effects on brain free amino acid composition in rats. Br J Nutr 1977;38:141–4.doi:10.1079/BJN19770070
      OpenUrlPubMed
    34. ↵
      2. Hofmann L ,
      3. Slotboom J ,
      4. Jung B , et al
      . Quantitative 1H-magnetic resonance spectroscopy of human brain: Influence of composition and parameterization of the basis set in linear combination model-fitting. Magn Reson Med 2002;48:440–53.doi:10.1002/mrm.10246
      OpenUrlCrossRefPubMed
    35. ↵
      2. Porter EJ ,
      3. Counsell SJ ,
      4. Edwards AD , et al
      . Tract-based spatial statistics of magnetic resonance images to assess disease and treatment effects in perinatal asphyxial encephalopathy. Pediatr Res 2010;68:205–9.doi:10.1203/PDR.0b013e3181e9f1ba
      OpenUrlCrossRefPubMedWeb of Science
    36. ↵
      2. Tusor N ,
      3. Wusthoff C ,
      4. Smee N , et al
      . Prediction of neurodevelopmental outcome after hypoxic-ischemic encephalopathy treated with hypothermia by diffusion tensor imaging analyzed using tract-based spatial statistics. Pediatr Res 2012;72:63–9.doi:10.1038/pr.2012.40
      OpenUrlCrossRefPubMedWeb of Science
    37. ↵
      2. Robertson NJ ,
      3. Cox IJ ,
      4. Cowan FM , et al
      . Cerebral intracellular lactic alkalosis persisting months after neonatal encephalopathy measured by magnetic resonance spectroscopy. Pediatr Res 1999;46:287–96.doi:10.1203/00006450-199909000-00007
      OpenUrlCrossRefPubMedWeb of Science
    38. ↵
      2. Robertson NJ ,
      3. Cowan FM ,
      4. Cox IJ , et al
      . Brain alkaline intracellular pH after neonatal encephalopathy. Ann Neurol 2002;52:732–42.doi:10.1002/ana.10365
      OpenUrlCrossRefPubMedWeb of Science
    39. ↵
      2. Loh PS ,
      3. Butcher KS ,
      4. Parsons MW , et al
      . Apparent diffusion coefficient thresholds do not predict the response to acute stroke thrombolysis. Stroke 2005;36:2626–31.doi:10.1161/01.STR.0000189688.95557.2b
      OpenUrlAbstract/FREE Full Text
    40. ↵
      2. Rivers CS ,
      3. Wardlaw JM ,
      4. Armitage PA , et al
      . Do acute diffusion- and perfusion-weighted MRI lesions identify final infarct volume in ischemic stroke? Stroke 2006;37:98–104.doi:10.1161/01.STR.0000195197.66606.bb
      OpenUrlAbstract/FREE Full Text
    41. ↵
      2. Kikuchi B
      . Mechanism of secondary deterioration after transient focal brain ischemia:1H MRS study. Niigata Med J 2008;122:271–9.
      OpenUrl
    42. ↵
      2. Cheong JL ,
      3. Cady EB ,
      4. Penrice J , et al
      . Proton MR spectroscopy in neonates with perinatal cerebral hypoxic-ischemic injury: metabolite peak-area ratios, relaxation times, and absolute concentrations. AJNR Am J Neuroradiol 2006;27:1546–54.
      OpenUrlAbstract/FREE Full Text

    Footnotes

    • Contributors SM, GSK, KM, CU-A recruited the babies and were the clinicians in charge of the clinical safety of the baby during the MRI/MRS study. SM wrote the first draft of the paper. MD assisted with the recruitment and care of the babies undergoing MRI. AB and MS performed the MRS analysis and the mean diffusivity assessment and interpreted the data. DP did the TBSS analysis and interpreted the data. RG scored the MRI images. XG provided advice on the MRI/MRS analysis. AH-C did the Bayley III assessment of the babies in the study. NJR designed and supervised the study and wrote the final drafts of the paper. All authors have been involved in finalising the manuscript and have seen and approved the final version.

    • Funding Funding support for this study was received from UK Department of Health’s NIHR BRC funding scheme.

    • Competing interests None declared.

    • Patient consent Not required.

    • Ethics approval University College London Hospital REC.

    • Provenance and peer review Not commissioned; externally peer reviewed.