Metabolic derangements are associated with impaired glucose delivery following traumatic brain injury

Abstract Metabolic derangements following traumatic brain injury are poorly characterized. In this single-centre observational cohort study we combined 18F-FDG and multi-tracer oxygen-15 PET to comprehensively characterize the extent and spatial pattern of metabolic derangements. Twenty-six patients requiring sedation and ventilation with intracranial pressure monitoring following head injury within a Neurosciences Critical Care Unit, and 47 healthy volunteers were recruited. Eighteen volunteers were excluded for age over 60 years (n = 11), movement-related artefact (n = 3) or physiological instability during imaging (n = 4). We measured cerebral blood flow, blood volume, oxygen extraction fraction, and 18F-FDG transport into the brain (K1) and its phosphorylation (k3). We calculated oxygen metabolism, 18F-FDG influx rate constant (Ki), glucose metabolism and the oxygen/glucose metabolic ratio. Lesion core, penumbra and peri-penumbra, and normal-appearing brain, ischaemic brain volume and k3 hotspot regions were compared with plasma and microdialysis glucose in patients. Twenty-six head injury patients, median age 40 years (22 male, four female) underwent 34 combined 18F-FDG and oxygen-15 PET at early, intermediate, and late time points (within 24 h, Days 2–5, and Days 6–12 post-injury; n = 12, 8, and 14, respectively), and were compared with 20 volunteers, median age 43 years (15 male, five female) who underwent oxygen-15, and nine volunteers, median age 56 years (three male, six female) who underwent 18F-FDG PET. Higher plasma glucose was associated with higher microdialysate glucose. Blood flow and K1 were decreased in the vicinity of lesions, and closely related when blood flow was <25 ml/100 ml/min. Within normal-appearing brain, K1 was maintained despite lower blood flow than volunteers. Glucose utilization was globally reduced in comparison with volunteers (P < 0.001). k3 was variable; highest within lesions with some patients showing increases with blood flow <25 ml/100 ml/min, but falling steeply with blood flow lower than 12 ml/100 ml/min. k3 hotspots were found distant from lesions, with k3 increases associated with lower plasma glucose (Rho −0.33, P < 0.001) and microdialysis glucose (Rho −0.73, P = 0.02). k3 hotspots showed similar K1 and glucose metabolism to volunteers despite lower blood flow and oxygen metabolism (P < 0.001, both comparisons); oxygen extraction fraction increases consistent with ischaemia were uncommon. We show that glucose delivery was dependent on plasma glucose and cerebral blood flow. Overall glucose utilization was low, but regional increases were associated with reductions in glucose availability, blood flow and oxygen metabolism in the absence of ischaemia. Clinical management should optimize blood flow and glucose delivery and could explore the use of alternative energy substrates.

Higher plasma glucose was associated with higher microdialysate glucose. Blood flow and K 1 were decreased in the vicinity of lesions, and closely related when blood flow was 525 ml/100 ml/min. Within normal-appearing brain, K 1 was maintained despite lower blood flow than volunteers. Glucose utilization was globally reduced in comparison with volunteers (P 5 0.001). k 3 was variable; highest within lesions with some patients showing increases with blood flow 525 ml/100 ml/min, but falling steeply with blood flow lower than 12 ml/100 ml/min. k 3 hotspots were found distant from lesions, with k 3 increases associated with lower plasma glucose (Rho -0.33, P 5 0.001) and microdialysis glucose (Rho -0.73, P = 0.02). k 3 hotspots showed similar K 1 and glucose metabolism to volunteers despite lower blood flow and oxygen metabolism (P 5 0.001, both comparisons); oxygen extraction fraction increases consistent with ischaemia were uncommon. We show that glucose delivery was dependent on plasma glucose and cerebral blood flow. Overall glucose utilization was low, but regional increases were associated with reductions in glucose availability, blood flow and oxygen metabolism in the absence of ischaemia. Clinical management should optimize blood flow and glucose delivery and could explore the use of alternative energy substrates.

Introduction
Acute derangements in cerebral glucose metabolism (CMRG) are common following traumatic brain injury (TBI). [1][2][3][4] These have been attributed to anaerobic glycolysis in the presence of classical ischaemia, but other pathophysiological mechanisms may also be responsible. [5][6][7] Microdialysis shows marked reductions in brain glucose 8 with modest decreases in plasma glucose within the normal range resulting in metabolic crisis 9 and poor outcome. 10 In addition to altered relationships between plasma and brain glucose, 8 substantial spatial and temporal heterogeneity in glucose utilization exists, 2 with increases in non-oxidative glucose utilization seen across white matter, 11 and in relation to brain lesions. 2,12 The changes in brain glucose metabolism represent complex interactions between glucose availability (plasma glucose), delivery (blood flow), uptake across the blood-brain barrier (transport), entry into the glycolytic pathway (phosphorylation), and oxidative metabolism of glucose (characterized as the metabolic ratio of oxygen-to-glucose utilization). Addressing these complex inter-relationships, while accounting for spatial variations in physiology, requires appropriate tools.
PET using the glucose analogue 18 F-fluorodeoxyglucose ( 18 F-FDG) and multi-tracer oxygen-15 (H 15 2 O, 15 O 2 and C 15 O) allow imaging of regional CMRG and kinetic parameters, [13][14][15] and comparison with cerebral blood flow (CBF), oxygen metabolism (CMRO 2 ) and oxygen extraction fraction (OEF). Hattori et al. 2 showed that 18 F-FDG transport into brain, expressed by the rate constant K 1 , was reduced within the vicinity of contusions following TBI. Glucose uptake depends on supply (plasma glucose and CBF) with transport facilitated via GLUT-1. 16 Reductions within the vicinity of lesions could be explained by a decrease in glucose availability, delivery, transport capacity, or a combination of these factors. Increases in glucose phosphorylation have also been seen following TBI, 2,12 and tend to occur on the background of an overall reduction in phosphorylation activity (measured by the 18 F-FDG rate constant k 3 ). 11 An integrated understanding of these acute metabolic derangements following clinical TBI is critical for development of treatments that improve outcome. Using a combination of 18 F-FDG and 15 O PET, with brain microdialysis in a subset of patients, we investigate how derangements in 18 F-FDG kinetic parameters relate to visible TBI lesions, and changes in plasma and microdialysis glucose, CBF, CMRO 2 , and OEF increases consistent with classical ischaemia.

Materials and methods
In this single-centre observational cohort study, 26 TBI patients requiring intracranial pressure (ICP) monitoring within intensive care and 47 healthy volunteers underwent 18 F-FDG and 15 O PET between February 1998 and July 2014. We excluded volunteers aged 460 years, technically inadequate imaging studies and those with large variations of arterial pressure or PaCO 2 . Some data have been reported previously, 6,17 but none for the aims of the current manuscript.
Studies were conducted in accordance with 1964 Declaration of Helsinki and later amendments following approval by Cambridgeshire Research Ethics Committee (REC 97/290) and UK Administration of Radioactive Substances Advisory Committee. Volunteers provided informed consent, and assent obtained from patient representatives with consent at follow-up if capacity was regained.

Healthy volunteers
Blood pressure, pulse oximetry, and arterial blood gases were monitored to ensure physiological stability during imaging. phosphorylation (k 3 ), and influx rate constant (K i ) ( Supplementary  Fig. 1), together with CMRG were calculated at the voxel and regional level as previously described. [21][22][23] A more detailed description is provided within the Supplementary material. We categorized PET studies into those performed within 24 h (early), between 2-5 days (intermediate), and 6-12 days (late) following TBI. Parametric maps were co-registered to acute anatomy using structural X-ray CT. In addition, and where available, MRI obtained at follow up were co-registered to early CT.

Microdialysis
Microdialysis catheters (CMA70 10-mm membrane, CMA) were inserted into the cerebral parenchyma as part of standard clinical monitoring. Vials were collected every 20 min and glucose, glutamate, lactate and pyruvate were determined using the CMA600 bedside microdialysis analyser. 24 We calculated the mean microdialysis and plasma glucose during 18 F-FDG PET.

Imaging analysis
Imaging data were analysed using a combination of customdesigned automated software (PETAn) 25 and several software packages, including Statistical Parametric Mapping (SPM, Wellcome Department of Imaging Neuroscience, University College London), MATLAB (MathWorks, Inc), Analyze (AnalyzeDirect, Inc) and Advanced Normalization Tools (ANTs). 26 Structural data (CT or T 1 -weighted MRI) were edited to extract a mask that identified brain tissue voxels and excluded extracranial tissue, CSF, and extra-axial haematomas. All whole brain and region of interest data were masked prior to calculation of PET parameters. In healthy volunteers a whole brain region was used incorporating mixed grey and white matter. In patients, T 1 -weighted MRI obtained at follow-up were non-rigidly registered with acute CT using ANTs.
Structural T 1 -weighted MRI acquired in healthy volunteers, and at follow-up in patients, were processed on a specific TBI pipeline 27 incorporating segmentation into grey and white matter regions of interest. These regions of interest were applied to the Jacobian determinant images 28 providing the amount of contraction or expansion required to normalize the subjects brain into template space (Supplementary material).

Structural regions of interest
Lesion-based regions of interest were drawn on CT using a bespoke segmentation tool (Imseg V1.8, BioMedIA, London, UK). We defined haemorrhagic contusion core and low signal intensity penumbra (Fig. 1A). In addition, we drew a 1 cm cuff of normalappearing tissue adjacent to penumbra (peri-penumbra). When present, we drew a 10 mm circular cuff of tissue centred on the microdialysis catheter tip. All other normal-appearing tissue was defined as the 'normal-appearing' region of interest.

Physiological regions of interest
We used the mean and standard deviation (SD) of k 3 voxels within normal-appearing brain of TBI patients to identify regional hotspots. These k 3 hotspot regions of interest were defined as the region where k 3 voxels were found above the upper 95% confidence interval threshold calculated for each individual patient using the mean plus 1.96 SDs of normal-appearing brain voxels.
We estimated an individualized critical OEF threshold using a previously validated technique to define and calculate the ischaemic brain volume (IBV). 17 A more detailed description is provided in the Supplementary material. We examined the spatial location, underlying physiology and mismatch between k 3 hotspots and the IBV. MRI obtained at follow-up were co-registered to acute CT. The k 3 hotspot regions of interest (minus lesion core and penumbra) were superimposed to explore possible relationships between increased k 3 and tissue fate.

Statistical analyses
Data were analysed using R (R Foundation for Statistical Computing, Vienna, Austria), and expressed as median (interquartile range, IQR) unless otherwise stated. Correlations were calculated using linear regression and, in the case of small datasets, Spearman rank. For comparisons between groups, we used the Mann-Whitney U-test and Kruskal-Wallis with Dunn's test for post hoc comparisons. The Dice similarity coefficient was used to determine the degree of spatial overlap between k 3 hotspots and the IBV. 29 Locally weighted scatterplot smoothing (LOWESS) 30 was used to plot relationships using whole brain voxel data using a 2.5% weighted sample. Results were interpreted taking into consideration that a Bonferroni correction for use of 11 variables [K 1 , k 2 , k 3 , K i , CMRG, lumped constant, CBF, CMRO 2 , OEF, CBV, and CMRO 2 /CMRG metabolic ratio (MR)] would require P 5 0.005 for formal statistical significance.

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

Results
Additional data are presented within the Supplementary material.

Study population
We analysed data from 20 healthy volunteers following 15 O PET and nine healthy volunteers following 18 F-FDG PET, with no significant neurological/psychiatric illness, and 34 combined PET sessions ( 18 F-FDG and 15 O) in 26 TBI patients with stable ventilation and CPP as per protocolized clinical management. Eleven healthy volunteers were excluded for age 460 years, three for movementrelated artefact and four with physiological instability during imaging (variations in PaCO 2 4 0.5 kPa).
Patients had severe TBI with a Glasgow Coma Score (GCS) 4 8 or presented with moderate TBI (GCS 9-12), but deteriorated, requiring sedation and ventilation for management of ICP. Of the 34 PET sessions in patients, 12 were within 24 h (early), eight between 2-5 days (intermediate), and 14 between 6-12 days (late) after TBI. Eight subjects underwent imaging on two occasions, but for one of these only the CBF from the late session was analysed. (P 5 0.001). Microdialysis data were available for 10 PET sessions from eight patients. On two occasions the microdialysis catheter was inserted within the vicinity of injured tissue. Plasma and microdialysis glucose in this subset ranged from 4.9-8.9 mmol/l and 0.4-4.1 mmol/l, respectively, and increases in plasma glucose were associated with higher microdialysis glucose (Rho = 0.64, P = 0.047; Supplementary Fig. 5). Brain tissue glucose concentration calculated using the 18 F-FDG kinetic parameters was related to plasma glucose within healthy volunteers (R 2 = 0.47, P = 0.04) and all TBI regions (R 2 = 0.16, P 5 0.001). In patients, this relationship was strongest within normal-appearing brain and peri-penumbra (R 2 = 0.66, P 5 0.001 and R 2 = 0.69, P 5 0.001, respectively), in comparison with core and penumbra (R 2 = 0.06, P = 0.17 and R 2 = 0.25, P 5 0.01, respectively) ( Supplementary Fig. 6).
Glucose delivery varies between normal-appearing, perilesional and lesion core tissue CBF and K 1 were decreased in core and penumbra as compared to peri-penumbra and normal-appearing brain in patients (P 5 0.001; Table 2). When compared with healthy volunteers, normalappearing brain in patients had lower CBF (P 5 0.001), but similar K 1 . There were no significant changes in CBF and K 1 within lesion and normal-appearing brain regions at different time points (Fig. 2  and Supplementary Table 1).

Cerebral blood flow is an important determinant of brain glucose transport
Region of interest data in patients showed a clear relationship between CBF and glucose transport (as measured by K 1 ) across all tissue classes (R 2 = 0.65, P 5 0.001. Fig. 3). This relationship was driven by core and penumbra (R 2 = 0.81 and 0.48, respectively, P 5 0.001), where both K 1 and CBF were below the lower limit of values found within healthy volunteers. There was no significant relationship within peri-penumbra and normal-appearing tissue (both R 2 = 0.1, P = 0.08 and P = 0.07, respectively).
Whole-brain voxel-wise analysis showed that where CBF was maintained within the normal range, K 1 was higher than within healthy volunteers, but reductions in CBF below 25 ml/100 ml/min were associated with steep reductions in K 1 (Fig. 4 and Supplementary Fig. 7).

Glucose utilization
Regional glucose utilization and phosphorylation rates are discordant and variable Data are displayed in Table 2. There was a general decrease in phosphorylation activity (k 3 ) in TBI patients with normal-appearing brain showing lower activity than healthy volunteers 0.029 (0.024-0.034) versus 0.057 (0.055-0.063)/min, P 5 0.001. In patients, k 3 was highest within core 0.044 (0.036-0.057)/min with values significantly higher than peri-penumbra 0.030 (0.025-0.037)/min and Figure 1 Peri-lesional hyperglycolysis in the absence of cerebral ischaemia. CT, CBF, CMRO 2 , OEF, and 18 F-FDG kinetic parameters obtained in a 29-year-old male within 24 h of severe TBI following a road traffic accident. Plasma glucose during imaging was 6.6 mmol/l. (A) CT demonstrates bilateral temporal and parietal haemorrhagic contusions. Regions of interest are defined for haemorrhagic lesion (core, red), hypodense tissue (penumbra, blue), and a 1 cm border zone of normal-appearing tissue (peri-penumbra, green). (B) Co-registered parametric maps of CBF, CMRO 2 , OEF, K 1 , K i and k 3 . The regions with a critical increase in OEF above the individually calculated threshold (ischaemic brain volume) and region of increased phosphorylation activity (k 3 hotspot) are both shown in red overlying the CT scan. The ischaemic brain and k 3 hotspot volumes were 7 ml and 56 ml, respectively, and overlap between these two tissue classes in this patient was 0.4 ml.
normal-appearing brain (P 5 0.001 for both comparisons). Penumbra k 3 values 0.039 (0.033-0.043)/min were higher than normal-appearing brain (P = 0.002). Increases in k 3 were associated with a fall in metabolic ratio, particularly within core and penumbra where metabolic ratio was 54 (Table 2 and Supplementary Fig. 8).
Glucose metabolism and K i were lowest in core and highest in normal-appearing brain, but significantly lower when compared to healthy volunteers (P 5 0.001 for both comparisons). When dephosphorylation (k 4 ) was included within regional kinetic modelling the results were similar (Supplementary Table 2). A similar pattern was found for CBF, CBV, CMRO 2 and metabolic ratio, with lowest values found within the core, and greater values found with increasing distance from the lesion core. The temporal pattern of regional data is displayed in Fig. 2 and Supplementary  Table 1. There was a non-significant trend for higher k 3 values at late time points post-injury within core (P = 0.09), and higher metabolic ratio values within all regions of interest at early time points (which did not survive correction for multiple comparisons). In comparison, regional CMRO 2 and OEF showed higher values at early times points post-injury, with CMRO 2 values for normalappearing brain (P = 0.003) and OEF values in penumbra (P = 0.004) remaining significant after correction for multiple comparisons.
Phosphorylation activity is dependent on the adequacy of cerebral blood flow The voxel-wise relationship between k 3 and CBF showed variability both within and between subjects ( Supplementary Fig. 7), but demonstrated clear patterns across the dataset. k3 remained constant across the range of CBF values found in healthy volunteers, but showed a small rise at CBF 20 ml/100 ml/min, followed by a precipitous fall with CBF 512 ml/100 ml/min (Fig. 4). A similar relationship was seen between k 3 and CMRO 2 ( Supplementary Fig. 9).

Relationships between plasma and microdialysis glucose and phosphorylation activity
Across all structural regions in patients, a lower plasma glucose was associated with an increase in k 3 and K i (Rho = -0.33 and -0.50, respectively, P 5 0.001). Using the modelled 18 F-FDG kinetic parameters, we calculated the proportion of brain tissue glucose phosphorylated across all patient regions and related this to plasma and calculated brain glucose concentration. The fraction of tissue glucose undergoing phosphorylation increases when plasma glucose falls to 56-8 mmol/l and/or brain glucose 51-2 mmol/l, particularly within the vicinity of brain lesions ( Supplementary Fig. 10).     Fig. 11) as well as K i (Rho = -0.71, P = 0.03). There were no significant relationships between microdialysis glucose and glutamate, pyruvate, lactate or lactate/pyruvate ratio. No microdialysis catheters were inserted in areas where k 3 or OEF values were above hotspot or IBV thresholds.
Hyperglycolysis is found in perilesional and normal brain, and is largely independent of ischaemia Cerebral physiology differed between k 3 hotspots and the ischaemic brain volume The volume of brain meeting the criteria for a k 3 hotspot was larger than IBV: 64 (55-96) versus 7 (2-19) ml (P 5 0.001, Mann-Whitney U-test), and the volume of k 3 hotspot and IBV did not differ between early, intermediate and late groups. Similar proportions of k 3 hotspots [34 )%] and IBV [29 (15-54)%] were within 1 cm of a lesion (P = 0.75), but regions of physiological derangement consistent with both tissue classes were seen within normalappearing brain. Evidence of overlap between k 3 hotspots and IBV was uncommon [0 (0-1) ml], with a Dice similarity co-efficient of 0.01 (0-0.02), and even where such overlap was substantial, it represented the minority of each physiological tissue class (Fig. 5).
More commonly, k 3 increases were found in the absence of ischaemia, both within the vicinity of injured (Fig. 1) and normal-appearing brain (Supplementary Fig. 12).
The temporal pattern of changes in physiology within k 3 hotspots and the IBV are shown with Supplementary Fig. 13 and Supplementary Table 4. Within k 3 hotspots, CMRO 2 , OEF and metabolic ratio tended to reduce with time post-TBI, but only comparisons for OEF survived correction for multiple comparisons (P = 0.002).

Association between the ischaemic brain volume and k 3 hotspots and outcome
Within 24 h, IBV showed a non-significant trend with worse outcome quantified using the GOS (Rho = -0.51, P = 0.11), and there was no association across the whole cohort (Rho = -0.28; P = 0.13). The volume of k 3 hotspot was not associated with the GOS (Rho = -0.18, P = 0.31).
Eight patients underwent follow-up MRI at 43 months postinjury (median 9, range 4-35 months), of which seven had T 1 -weighted volumetric images for comparison with similar data acquired within the nine 18 F-FDG PET healthy volunteers. Mean normalized cortical grey matter volume was similar (0.42 versus 0.39, P = 0.05) in patients and healthy volunteers, while white matter volume was lower in patients (0.42 versus 0.47, P 5 0.01). The mean Jacobian determinant values for cortical grey and white matter regions in patients compared with healthy volunteers were 0.04 versus -0.01 (P = 0.01) and 0.02 versus -0.01 (P = 0.06), respectively, with higher values indicative of smaller volumes requiring expansion to match the brain template. Early increases in k 3 outside of lesion core and penumbra were often found in close proximity to areas showing evidence of late structural injury on MRI obtained at follow-up (Fig. 6). There was no consistent voxel-wise relationship between k 3 and volume contraction using Jacobian determinant values across the normal-appearing brain in TBI patients ( Supplementary Fig. 14).

Discussion
We used 18 F-FDG and multi-tracer 15 O PET together with brain microdialysis in a cohort of acute TBI patients to image the impact of glucose availability, delivery and utilization on cerebral metabolism. Brain glucose delivery (K 1 ) is dependent on both plasma glucose and the adequacy of CBF, particularly within regions where CBF 525 ml/100 ml/min. An overall global reduction in glucose utilization following TBI appears to be driven by low glucose delivery and low phosphorylation activity, but regional hotspots were found within, and distant from, visible brain lesions. Such increases were explained by preserved glucose delivery (K 1 ) despite reductions in CBF, and in plasma and brain glucose. Increases in glucose utilization were typically accompanied by CMRO 2 reductions, but without OEF increases consistent with ischaemia. These metabolic derangements represent a potential therapeutic target. Figure 3 Relationship between glucose transport and CBF based on region of interest data. Scatterplot of 18 F-FDG transport (K 1 ) and CBF in patient regions of interest: lesion core (black), penumbra (purple), peripenumbra (blue) and normal-appearing (orange). The black vertical and horizontal dotted lines indicate the full range for healthy volunteer CBF and K 1 values, respectively.

Glucose availability, delivery and transport
We confirm previous findings that brain glucose is associated with serum glucose, but variable across the injured brain, 8,31 and that K 1 is similar to healthy volunteers in normal-appearing brain and reduced around lesions following TBI. 2 However, we also show for the first time, that K 1 was related to CBF, particularly at CBF 525 ml/100 ml/min (Figs 3 and 4). Microdialysis shows a linear relationship between plasma and brain glucose in non-injured areas after TBI, 8,31 but glucose transport is impaired within the injured brain. 8 CBF may be the rate-limiting factor for glucose transport into the brain, but we found that normal-appearing brain in TBI patients had similar K 1 to that found within healthy volunteers despite lower CBF, suggesting that increases in GLUT-1 expression may also contribute. 32

Glucose utilization
In comparison with healthy volunteers, TBI patients showed a global decrease in glucose utilization, but derangements in kinetic parameters were heterogeneous across the injured brain, in keeping with past reports, 33 with reductions attributed to a combination of sedation, 34,35 decreased consciousness and/or TBI itself. 36 All patients who underwent PET imaging received variable rate infusions of sedatives and analgesics to facilitate metabolic suppression as part of intensive care management and ICP-directed therapy, but despite this, there were regional increases in glucose utilization. The metabolic ratio was reduced within visible brain lesions, but also within normal-appearing brain, as reported previously. 15,37 Lower values within lesion core and penumbra imply increased anaerobic glycolysis, but the lack of OEF increase is against classical ischaemia and is suggestive of microvascular ischaemia and mitochondrial dysfunction. 7 Whilst visible lesions had the lowest K i and CMRG, this was driven by low glucose delivery (K 1 and CBF) rather than a reduction in k 3 , which was higher than that found within normal-appearing brain. A sharp fall in k 3 was associated with CBF 512 ml/100 ml/ min and cerebral oxygen metabolism consistent with irreversible tissue injury (540 mmol/100 ml/min). 38 Prior to reaching this threshold, increases in k 3 within individual subjects ( Supplementary Fig. 7) were seen below the lower limit of CBF found in healthy volunteers (525 ml/100 ml/min). Such findings support measures to preserve or optimize CBF and glucose delivery within the vicinity of brain lesions to maintain metabolism and prevent neuronal loss. In addition, studies highlight that alternative substrates may be utilized as an energy source, via the tricarboxylic acid cycle, following TBI and are a potential therapeutic approach. [39][40][41][42] Increases in k 3 and K i in TBI patients were associated with both lower plasma glucose and microdialysis glucose, raising the possibility that low brain glucose is driven by increased consumption as well as reduced delivery. Our data show that the proportion of brain glucose utilized increases as plasma and brain glucose falls below 6-8 and 1-2 mmol/l, respectively, particularly within the vicinity of lesions (Supplementary Fig. 10). Other studies have shown CMRG increases with tight glycaemic control, associated Figure 4 Relationships between 18 F-FDG kinetic parameters and CBF based on voxel data. For each PET imaging session in patients the relationship between CBF and 18 F-FDG transport (K 1 ; left) and phosphorylation activity (k 3 ; right) is plotted for voxels across the whole brain. The fitted light blue lines represent modelling of the relationship between each parameter using locally weighted scatterplot smoothing (LOWESS), with the 95% confidence interval shown in grey. The blue vertical and horizontal dotted lines indicate the full range for healthy volunteer CBF and 18 F-FDG kinetic parameters, respectively.
with an increase in lactate/pyruvate ratio indicative of metabolic stress. 43 We found no association between microdialysis lactate or lactate/pyruvate ratio and k 3 in our current study, and have previously shown an increase in glucose metabolism to both lactate and pyruvate with no change in the lactate/pyruvate ratio following TBI. 44 Randomized controlled trials of intensive insulin therapy in TBI show no benefit, [45][46][47] and though a recent metaanalysis is inconclusive, 48 our data suggest that plasma glucose levels should be carefully controlled and not be allowed to fall below 6-8 mmol/l, ideally in combination with monitoring of brain glucose to ensure values of 1-2 mmol/l are maintained across the injured brain.
Previous 18 F-FDG PET studies show hyperglycolysis following TBI, 12 driven by regional phosphorylation increases. 2 We delineated the burden and spatial pattern of k 3 hotspots and compared this with evidence of ischaemia (IBV). The k 3 hotspot regions were found both in the vicinity of, and distant from visible lesions. The volume of k 3 hotspots was larger than the IBV, the underlying physiology different, and there was little spatial overlap between these two tissues classes. While ischaemia and hyperglycolysis sometimes overlapped within 24 h of injury (Fig. 5), this was uncommon. The typical metabolic signature of a k 3 hotspot was higher CMRG with preserved K 1 and OEF, despite lower CBF, CMRO 2 and metabolic ratio. A late (Day 6-12 post-TBI) trend for lower CMRO 2 , OEF and metabolic ratio within k 3 hotspots was suggestive of worsening or persistent metabolic derangements and risk of neuronal loss with CMRO 2 values close to the threshold for tissue viability (540 mmol/100 ml/min). 38 While such changes could reflect differences in intensive care management at different phases of recovery following TBI, all received standardized care. This pattern of metabolic derangements is also consistent with microvascular ischaemia 7 and/or mitochondrial dysfunction, 49 and could be associated with such pathophysiological processes as spreading depolarization, 50 cytotoxic oedema, 51 and neuroinflammation. 52 Such processes are common following TBI, 7 and are consistent with our finding of a low metabolic ratio secondary to a decrease in oxygen metabolism within regions showing increases in glucose utilization. In patients, follow-up imaging of brain metabolism, with structural and functional tissue outcome evaluated using MRI, and functional outcome assessments would be required to definitively address the significance of early metabolic derangements.
We compared 18 F-FDG PET obtained within a group of TBI patients to identical data obtained in healthy volunteers using standard kinetic modelling. [53][54][55] Previously published values for 18 F-FDG kinetic parameters in healthy volunteers and TBI patients 2,56 are similar to the data we report, and the available test-retest reproducibility data suggest a co-efficient of variation of up to 10-15% for glucose metabolism. 57 While test-retest variability is important in assessing the significance of interventional or serial studies within the same subject, in this observational study we demonstrate regional changes in 18 F-FDG kinetic parameters and glucose metabolism between TBI subjects and healthy volunteers of up to 50% or more across the injured brain. At the voxel level, to avoid high variability in the 18 F-FDG kinetic parameters, we used a basis function version of a two-tissue compartment Figure 5 Evidence of cerebral ischaemia and hyperglycolysis. CT, CBF, CMRO 2 , OEF, and 18 F-FDG kinetic parameters obtained in a 40-year-old female within 24 h of severe TBI resulting from a fall. Imaging was obtained following evacuation of a subdural haematoma. Plasma glucose during PET was 3.7 mmol/l. (A) CT demonstrates residual subdural blood with minimal midline shift and underlying haemorrhagic contusions. Regions of interest are defined for haemorrhagic lesion (core, red), hypodense tissue (penumbra, blue), and a 1 cm border zone of normal-appearing tissue (peri-penumbra, green). (B) Co-registered parametric maps of CBF, CMRO 2 , OEF, K 1 , K i and k 3 . The regions with a critical increase in OEF above the individually calculated threshold (ischaemic brain volume) and increased phosphorylation activity (k 3 'hotspot') are both shown in red overlying the CT scan. The ischaemic brain volume and k 3 hotspot volume were 96 ml and 174 ml, respectively, and the volume of overlap between these two tissue classes in this patient was 38 ml. model, which assumes irreversible phosphorylation of 18 F-FDG. 22,54,58 Modelling the data on a regional basis with the inclusion of dephosphorylation (k 4 ) showed similar results (Supplementary Table 2). In contrast to Hattori et al., 2 we found no differences in k 4 between patient regions on interest, or in comparison with healthy volunteers. The lumped constant within normal-appearing brain in TBI patients (0.41) was lower than that found within healthy volunteers (0.56), and is comparable to previous reports. 36 While this value is used to calculate CMRG, we also report the influx rate constant (K i ).

Outcome impact
We have previously shown that early ischaemia (524 h post-TBI) was associated with poor outcome. 17 We were unable to confirm this finding, perhaps because IBV was not increased within 24 h of injury in the current cohort. k 3 hotspots were found throughout the first 12 days post-TBI, but there was no association with poor outcome, as defined by GOS. Those patients who underwent follow-up MRI demonstrated lower normalized global white matter volume compared with healthy volunteers and a trend for lower cortical grey and white matter volumes. While k 3 hotspots in structurally normal brain (on CT) may be associated with MRI visible lesions at follow-up, the relationship between k 3 and volume contraction was variable. Metabolic derangements resulting in both increases and decreases in k 3 may be associated with late structural injury and further study is required to confirm the significance of these derangements.

Limitations
We examined the impact of TBI on glucose delivery and metabolism using PET obtained within a typical patient cohort, rather than examine the temporal pattern and evolution within individual subjects. As such, we did not acquire follow-up MRI in all subjects and the analyses of the possible relationship between early metabolic derangements and structural outcome presented are illustrative of how future studies could address this issue. Future sequential imaging studies incorporating MRI and spectroscopy, 59 diffusion tensor imaging and multi-tracer PET ( 15 O, 18 F-FDG, and SV2A) 60 could delineate evidence of pan-necrosis, cortical atrophy and selective neuronal loss related to early metabolic derangements, and correlate these with functional deficits resulting in poor outcome post TBI.
Acute MRI was not available for TBI patients and evidence of structural injury was defined on the basis of early CT. While we may have missed structural changes not visible on CT, we were able to demonstrate significant physiological differences between structural regions of interest and used PET to define metabolic regions of interest for analyses.

Conclusions
The data presented in this study confirm previous work 61 and show that, up to 12 days post-injury, there is an overall reduction in oxygen and glucose metabolism across the brain. Our results show that K 1 is preserved within normal-appearing brain in the face of CBF reduction and this may help maintain glucose delivery, Figure 6 Association between early metabolic derangements and tissue fate. Imaging obtained in a 17-year-old male with moderate TBI following a road traffic accident with initial GCS of 12, but who required sedation and ventilation for control of raised ICP. Co-registered CBF, CMRO 2 , OEF, 18 F-FDG kinetic parameters and CT imaging obtained on Day 8 following TBI. Plasma glucose during imaging was 5.4 mmol/l. The CT demonstrates left frontal and temporal haemorrhagic contusions and is shown with the T 1 -weighted MRI obtained at 9 months post-injury; the MRI has been non-rigidly registered to the CT. Both structural images are displayed with the region of k 3 hotspot outlined in red, but both lesion core and penumbra have been excluded from the hotspot to ensure that only increases in k 3 outside of the lesion identified on CT imaging are shown. The volume of brain within the k 3 hotspot in this subject was 149 ml and, in comparison with brain that appeared structurally normal, had CBF 19.6 versus 19.6 ml/100 ml/ min, CMRO 2 55.2 versus 60.2 mmol/100 ml/min, OEF 37.8 versus 42.4%, K 1 0.097 versus 0.087 ml/ml/min, K i 0.018 versus 0.012 ml/ml/min and k 3 0.058 versus 0.024/min, respectively. The T 1 -weighted MRI demonstrates established lesions within the left frontal and temporal brain regions in close proximity with k 3 increases outside of lesion core and penumbra identified on the CT image. and mitigate the impact of modest plasma glucose reductions. However, this compensation fails when CBF and/or glucose delivery are critically low. Despite this, such regions can show high k 3 values, providing evidence of metabolically active tissue, particularly within the vicinity of brain lesions at risk of expansion. Regions of increased glucose utilization were associated with non-ischaemic reductions in CBF and CMRO 2 consistent with microvascular ischaemia and/or mitochondrial dysfunction. The primary physiological target in TBI has been the maintenance of CBF to ensure oxygen delivery, but we show that it may be just as important to ensure adequate glucose delivery. Where CBF elevation is either impossible or ineffective, our data provide a physiological framework for maintaining glucose at high normal levels (8-10 mmol/l) to optimize substrate delivery. Finally, these data make a case for the development and evaluation of alternative substrates (such as ketones) whose delivery is independent of transporters used to deliver glucose. 42