Cerebral ketones detected by 3T MR spectroscopy in patients with high grade glioma on an Atkins-based diet

Abstract

Background and Purpose:

The objective of this study was to evaluate the feasibility of 3T magnetic resonance spectroscopy (MRS) to monitor brain ketone levels in patients with high-grade glioma who were on a ketogenic diet (KD- a modified Atkins diet) for 8 weeks.

Methods:

Paired pre- and post-KD MRS data from both the lesion and contralateral hemisphere were analyzed using LCModel software in 10 patients.

Results:

At baseline, the ketone bodies acetone (Ace) and β-hydroxybutyrate (bHB) were nearly undetectable, but by week 8 increased in the lesion for both Ace (0.06±0.03->0.27±0.06 i.u, p=0.005) and bHB (0.07±0.07->0.79±0.32 i.u, p=0.046). In contralateral brain, Ace was also significantly increased (0.041±0.01->0.16±0.04 i.u., p=0.004) but not bHB. Ace was detected in 9/10 patients at week 8 and bHB in 5/10. Acetone concentrations in contralateral brain correlated strongly with higher urine ketones (r=0.87, p=0.001) and lower fasting glucose (r=−0.67, p=0.03). Acetoacetate was largely undetectable. Small, but statistically significant, decreases in NAA were also observed in the contralateral hemisphere at 8 weeks.

Conclusion:

This study suggests that 3T MRS is feasible for detecting small cerebral metabolic changes associated with a ketogenic diet, provided that appropriate methodology is used.

Introduction

Ketogenic diets (KDs) have been used to treat epilepsy for almost 100 years ¹, and recently have been explored for many other neurological conditions including multiple sclerosis, Parkinson disease, Alzheimer disease, amyotrophic lateral sclerosis, migraine, autism, and glioma ^2–8. The physiological effects of these diets are incompletely understood, but it is clear that they modify the body’s energy metabolism, leading to lower systemic glucose levels and increased levels of ketone bodies. The ketone bodies β-hydroxybutyrate (bHB) and acetoacetate (AcAc) are produced in the liver from fatty acids under carbohydrate-restricted diets, and a third ketone body, acetone (Ace), is produced as the result of breakdown of AcAc. The ketones are water soluble and transported to other parts of the body, including the brain. However, a challenge in the application of the ketogenic diet is measuring its effect on cerebral metabolism. While dietary compliance may be estimated from measurements of urine ketones, weight and dietary food records, as well as measurement of serum ketone and glucose (and other) levels ^9–12, these measures provide a picture of the body’s ketogenic state but may not reflect the level of cerebral ketosis ¹³.

Since ketone bodies are known to accumulate in the brain in low mM concentrations during ketosis, they should be detectable using the non-invasive technique of proton magnetic resonance spectroscopy (MRS). The earliest demonstration of this was in patients recovering from diabetic ketoacidosis (DKA), where it was shown that short TE STEAM MRS at 1.5T was able to detect elevated levels of brain Ace ¹⁴. Subsequently both bHB and AcAc (as well as lactate (Lac)) were reported to be detected in children recovering from DKA using a long TE PRESS sequence at 1.5T ¹⁵. Increases in bHB and Lac using edited 4T MRS have also been observed during fasting in healthy subjects ¹⁶, and increased Ace (and possibly AcAc) observed during ketogenic diet treatment for epilepsy using 1.5T MRS ^{17, 18}. Increases in Ace and AcAc levels were reported using short TE 3T MRS in patients with primary brain tumors undergoing the ketogenic diet ¹⁹.

Overall, these prior studies indicate that ketone bodies are detectable by MRS in some cases, but that the results are often equivocal because of their low concentrations. In addition, it can be difficult to distinguish Ace from AcAc because their chemical shifts are very similar (2.22 vs. 2.27 ppm respectively), and the bHB doublet (at 1.20 ppm) is potentially obscured by overlap with lipids or lactate, as well as signal losses due to chemical shift displacement effects at intermediate TE values ²⁰. Therefore, there is no general concordance in the literature as to which ketone bodies can be most reliably detected in the brain using MRS, or which acquisition methodology is optimal.

The purpose of this paper is to evaluate the utility of short echo time 3T MRS to quantify cerebral ketone body and metabolite levels in patients with high-grade glioma enrolled in an on-going open-label, single arm, phase II clinical trial of the KD for 8 weeks. Note that this is an interim analysis which focuses only on the MRS data. The primary objective of the main study is to investigate feasibility of the dietary intervention as measured by dietary compliance. Here, the results of an interim analysis assessing feasibility of MRS for measuring cerebral ketones and the ability to detect KD-related changes are reported; the full clinical trial is still on going, and the results of the primary objective will be reported at a later date.

Methods

Study design and participants

The study was conducted at Johns Hopkins and the Wake Forest Baptist Medical Center with the approval of both institutional review boards. Written informed consent was obtained from all patients. The trial design was a single arm phase II study designed to assess the feasibility, safety, and activity of a modified Atkins-based diet to prevent tumor recurrence in glioma patients following the completion of adjuvant chemotherapy. Eligible patients were at least 18 years of age, had a Karnofsky performance status ≥ 60, a diagnosis of high-grade astrocytoma (WHO grade III or IV), and completed ≥ 80% of prescribed radiation therapy, concurrent temozolomide, and adjuvant temozolomide without CTCAE grade 4 leukopenia, neutropenia, or thrombocytopenia. Patients were excluded if they had a history of metabolic disorder, BMI >35.0 or < 20.0 kg/m², or a milk allergy.

Treatment

The study intervention consisted of an 8-week diet, which included 2 days ‘fasting’ and 5 days on a modified Atkins diet each week. The non-consecutive ‘fasting’ days had strict caloric restriction of up to 20% of the recommended daily caloric intake of the patient provided via a 4:1 ratio ketogenic liquid (e.g. KetoCal® drink). Modified Atkins diet days required net carbohydrate restriction to 20 grams per day with no caloric restriction. All diets were customized with the guidance of a registered dietitian.

MR Imaging

Patients underwent MRS at the beginning and end of the diet (8 week interval) on 3 Tesla MR scanners at either Johns Hopkins (12 patients) on a 3 Tesla Achieva MR system (Philips Medical Systems; Best, The Netherlands) or Wake Forest (2 patients) on a 3 Tesla Skyra MR system (Siemens Healthineers, Erlangen, Germany) using 32-channel head coils. Conventional MRI included a 1 mm isotropic T₁-weighted MPRAGE scan and axial FLAIR images. A 2×2×2 cm³ voxel was placed in a region maximally occupied by the lesion, while avoiding structures that might degrade spectral quality such as fluid filled spaces, regions of high magnetic susceptibility variation (e.g. due to hemorrhage, surgical clips, or air tissue interfaces), and lipid signals from the scalp. In patients with a gross total resection, the lesion voxel was placed adjacent to the resection cavity, often in perilesional regions of T₂ hyperintensity. A second voxel was placed in the contralateral hemisphere mirroring the location of the lesion voxel.

MRS was performed using semi-LASER (sLASER) localization ^{21, 22} (TR = 2.2 s, TE = 34 ms / 40 ms at Johns Hopkins / Wake Forest,). The sLASER sequence used broadband adiabatic refocusing pulses (3 kHz / 5 kHz for Johns Hopkins / Wake Forest), which have excellent slice profiles. In addition to acquiring water-suppressed data (number of excitations = 128), 4 excitations were recorded without water suppression. Scan time per voxel was 4 min 54s, including 2 dummy excitations to establish the steady state. Prior to acquisition, field homogeneity was optimized up to 2^nd order using either a FASTMAP-based technique ²³ or gradient-echo based shimming at Johns Hopkins and Wake Forest, respectively.

The ‘LCModel’ program ²⁴ was used to fit spectra, with basis sets for each site containing 3 ketone bodies (acetone (Ace), beta-hydroxybutyrate (bHB), and acetoacetate (AcAc)) (see Supplementary Figure 1) as well as standard MRS metabolites: alanine (Ala), ascorbate (Asc), aspartate (Asp), creatine (Cr), γ-aminobutyric-acid (GABA), glutamine (Gln), glutamate (Glu), glycine (Gly), myo-inositol (mI), lactate (Lac), glycerophosphocholine (GPC), phosphocreatine (PCr), phosphorylethanolamine (PE), scyllo-inositol (sI), taurine (Tau), glucose (Glc), glutathione (GSH), N-acetylaspartate (NAA), N-acetylaspartylglutamate (NAAG), and 2-hydroxyglutarate (2HG). Literature values for chemical shifts and coupling constants were taken from the following references ^25–27. Basis spectra were generated using density matrix simulations which incorporated real refocusing pulse information and 2D localization (MATLAB, Mathworks Inc). Macromolecular contribution in the spectra were modeled using simulated components available in LCModel. Metabolite concentrations were estimated relative to an internal water reference (assuming a bulk water concentration of 55.5 M), which were corrected for water T₂ decay differences in tumors ²⁸ and healthy brain ²⁹. Given the low concentrations of cerebral ketones in this study population, a rejection threshold using Cramér-Rao lower bounds (CRLBs) of metabolite fitting > 80% was applied in order to avoid statistical biasing of results when comparing increases of very low concentration metabolites ³⁰. Additionally, no correction for either tissue water or CSF content of the voxel was applied, thus reported concentrations are given in ‘institutional units’ (i.u.) which are approximately equivalent to mM.

Statistical analysis

Means and standard errors of the mean were presented for normally distributed continuous measures, and medians and ranges were presented for non-normally distributed continuous measures. Percentages and counts were presented for discrete measures. Approximate 95% confidence intervals on individual metabolite fits are calculated as 2xCRLB as indicated by ²⁴. After fitting spectral data with LCModel, statistical analysis was carried out using StataIC 15 (2017. College Station, TX: StataCorp LLC). For this analysis of feasibility, the primary outcome was the change in cerebral ketone concentrations from baseline to week 8. The primary analysis focused on the 3 ketone bodies Ace, bHB, and AcAc. Differences between groups at each time point were assessed by paired t-test for normally distributed continuous measures. After correcting for multiple comparisons using Bonferroni correction, an acceptance threshold of p <0.008 (=0.05/6 corrected for the 3 primary ketone metabolites of interest at 2 time-points) was considered significant. Spearman’s rank correlation coefficient was calculated to evaluate the association between continuous measures.

Results

Patient Characteristics

At the time of analysis, 14 subjects were enrolled. Two participants were excluded from the MRS analysis due to either scanner error or a large spectral linewidth arising from magnetic susceptibility gradients in the temporal lobe, respectively. Only baseline MRS data was available for two patients who did not complete the 8-week dietary intervention. Demographics of the 10 patients with evaluable pre- and post-intervention MRS are given in Table 1; 7 (70%) had WHO grade III anaplastic astrocytomas, and 3 (30%) had glioblastomas (Table 1). Two patients (20%) had previously undergone biopsy, 3 (30%) a subtotal resection, and 5 (50%) a gross total resection.

Table 1:

Demographic features of participants.

Age (years; mean, standard deviation)	49.2 (10.6)
Sex (male; n, %)	6 (60%)
World Health Organization (WHO) grade
- WHO 3	7 (70%)
- WHO 4	3 (30%)
Extent of Resection
- Biopsy	2 (20%)
- Subtotal	3 (30%)
- Gross Total	5 (50%)
IDH1/2 Mutational Status
- IDH wild type	4 (40%)
- IDH mutant	6 (60%)
MGMT Promoter Methylation Status
- Unmethylated	4 (40%)
- Methylated	4 40%)
- Unknown	2 (20%)
Concurrent TMZ (% completed; median, range)	100% (80–100%)
Adjuvant TMZ (# of cycles; median, range)	6 (6–12)

IDH = isocitrate dehydrogenase, MGMT = O⁶-methylguanine-DNA methyltransferase, TMZ = temozolomide.

With regards to biomarkers of dietary compliance, no participants had detectable urine ketones at baseline, and nine-of-the-ten patients who completed the study achieved some level of ketosis, measured as “trace” (5 mg/dL) or greater, during the study. Eight (80%) achieved moderate (40mg/dL) or greater urinary ketosis. Average fasting glucose levels in participants decreased modestly from 91 mg/dL at baseline to 84 mg/dL at 8 weeks.

Cerebral Ketone Levels

Figure 1 shows representative spectra from a single patient together with results of the LCModel fitting. Spectra and fitting results from each patient are provided in the Supplementary Figure 2. Quantitative MRS results for ketones bodies are summarized in Table 2 and for both ketones and other brain metabolites in Figure 2. Mean water linewidths from contralateral and lesion voxels were (6.7±0.9 Hz) and (6.1±1.9 Hz), respectively, indicating good shimming in quantified spectra.

Figure 1:

Representative MR spectra from one patient, together with spectral fitting results from the LCModel, at baseline and week 8. FLAIR MRI shows the voxel placement in the lesion in the left insular cortex and the corresponding voxel in the contralateral hemisphere. Individual fits of acetone (Ace) and β-hydroxybutyrate (bHB) at week 8 are also shown below the spectra, as well as those from lactate (Lac), glutamate (Glu), glutamine (Gln), and γ-aminobutyric acid (GABA).

Table 2:

Ketone body concentrations as measured by MRS before and after treatment with a ketogenic diet in both the lesion and in contralateral brain.

	Baseline (i.u.)	n	Week 8 (i.u.)	n	P value
Ace
- Contralateral Brain	0.04±0.01	5	0.16±0.04	9	0.004**
- Lesion	0.06±0.03	4	0.27±0.06	9	0.005**
bHB
- Contralateral Brain	0.09±0.06	2	0.28±0.08	6	0.12
- Lesion	0.07±0.07	1	0.79±0.32	5	0.046*
AcAc
- Contralateral Brain Lesion	0.07±0.04	3	0.06±0.03	3	0.76
- Lesion	0.03±0.03	1	0.02±0.02	1	0.72

P-value computed using paired t-test,

^*p ≤ 0.05,

^**p ≤ 0.008 (Bonferroni-adjusted threshold), n = number of spectra in which metabolite was fitted according to defined criteria. Ace = acetone, bHB = beta-hydroxybutyrate, AcAc = acetoacetate, i.u. = institutional units. Errors are provided as standard error in the mean.

Figure 2:

Ketone body (left panel) and metabolite (right panel) concentrations as measured by MRS at baseline and week 8 in the (A) contralateral brain and (B) lesion for all subjects. * p ≤ 0.05, ** p ≤ 0.008 (Bonferroni-adjusted threshold for ketone bodies).

In the lesion spectra, the ketone bodies Ace and bHB were detectable in a greater number of patient spectra following the dietary intervention according to the defined criteria. AcAc was only measured in one spectrum at baseline and week 8. Ketone bodies were largely undetectable in the lesion at baseline and increased significantly at week 8 for both Ace (baseline: 0.06±0.03, week 8: 0.27±0.06 i.u., p = 0.005) and bHB (baseline: 0.07±0.07, week 8: 0.79±0.32 i.u., p = 0.046, NS after Bonferroni correction, Table 2). Overall, Ace was detected in 90% and bHB in 50% of lesion scans at week 8. Mean CRLBs of ketone fitting after KD in the lesion were 29% for Ace and 34% for bHB, which were considerably lower than the rejection threshold.

Changes in the contralateral brain largely mirrored those seen in the lesion. Ace measures increased significantly from baseline (0.04±0.01 i.u., mean CRLB = 64%) to (0.16±0.04 i.u., mean CRLB = 39%) (p = 0.004) at week 8. bHB in the contralateral brain also increased from 0.09±0.06 to 0.28±0.08 i.u.; however, there was wider variability in bHB and this increase was not statistically significant (p = 0.12). Ace was detected in 90% and bHB in 60% of 8-week scans in the contralateral brain region.

Interestingly, there was significantly more Ace in the lesion than in the contralateral brain at week 8 (0.27±0.06 vs. 0.16±0.04 i.u., p = 0.012). A similar pattern was noted with bHB although this difference was not statistically significant (0.79±0.32 vs 0.28±0.08 i.u., p = 0.17).

Other Brain Metabolites

As expected, total NAA (tNAA) was lower in the lesion than contralateral brain both at baseline (p = 0.02) and at 8 weeks. Lactate (Lac) was significantly higher in the lesion compared to contralateral brain both at baseline and week 8 (2.5±0.74 vs 0.25±0.09 i.u., p = 0.004). Otherwise, no significant differences in indigenous cerebral metabolites were observed between the lesion and contralateral side, including total choline (p = 0.14) and total creatine (p = 0.09).

Contralateral tNAA concentrations decreased by a small but significant amount from 7.9±0.2 to 7.7±0.2 i.u. (p = 0.02) over the 8-week treatment period, while lesion tNAA levels were stable. There were no detectable changes in the concentrations of any other reported metabolite in either the lesion or contralateral brain, including lactate (p=0.8 for lesion, p = 0.3 for contralateral), over the 8-week period.

Correlation of brain ketone levels with peripheral markers of ketosis

Overall, patients with greater systemic measures of dietary compliance (as evidenced by urine ketones and fasting blood glucose at week 8) showed higher cerebral ketone concentrations. At the time of the week 8 MRS, higher contralateral brain Ace levels were significantly associated with greater urine ketones (Figure 3a, r = 0.87, p = 0.001) and lower week 8 fasting glucose (Figure 3b, r = −0.67, p = 0.03). A similar trend was also seen for lesion Ace concentrations and both urine ketones (r = 0.54, p = 0.11) and fasting glucose (r = −0.54, p = 0.10) but did not reach significance. No significant correlations were seen between bHB levels in either the lesion or contralateral brain and urine ketones or fasting glucose levels.

Figure 3:

Association between contralateral brain acetone levels estimated by MRS and systemic measures of ketosis at week 8. Spearman’s rank correlation between acetone concentrations (i.u., error bars represent the +/− 95% confidence interval as based on the LCModel CRLB values) in contralateral brain at week 8 plotted against A) urinary ketosis score and B) fasting serum glucose levels. Urine ketones are defined as 1 trace (~5 mg/dL), 2 small (~15 mg/dL), 3 moderate (~40 mg/dL), 4 large (≥ 80 mg/dL).

Discussion

The main finding of this study is that short echo time single voxel brain MRS performed at 3T using 32-channel receive head coils, in combination with LCModel analysis, was able to measure concentrations of ketone body Ace (90% of spectra) and, to a lesser extent, bHB (50–60% of spectra) in patients with glioma who were on a modified Atkins diet for 8 weeks. Furthermore, the brain Ace concentrations were found to correlate with concurrent urine ketone level measurements. These results suggest that localized MRS may provide a useful measure of brain ketosis, and possible noninvasive pharmacodynamics marker, in clinical trials of ketogenic diets in glioma or other neurological conditions.

It should be noted, however, that the amplitudes of the ketone body signals in the spectrum are very small in this dietary treatment, so robust acquisition and analysis methods are required in order to estimate their concentrations. In the current study, spectral quality was maximized through the combined use of a sLASER localization sequence, high-order shimming, and 32-channel receiver coils. In addition, a carefully simulated and constructed LCModel basis set for the specific acquisition and quantitative criteria for considering a peak ‘detectable’ or not were used for spectral analysis.

Sensitivity of detection of Ace, AcAc and bHB

The ketone body most reliably detected in this study was Ace, which is consistent with prior KD studies in patients with epilepsy ¹⁸ and also in patients recovering from DKA ¹⁴. However, other studies during fasting or ketogenic diet have also reported increases in bHB, AcAc or Lac ^{16, 17, 19}. The MRS detectability of these compounds is not uniform; it should be noted that the Ace signal is a singlet at 2.22 ppm which arises from 6 equivalent protons, whereas AcAc (a singlet at 2.27 ppm) arises from 3 equivalent protons, and therefore is half the amplitude for the same molecular concentration. The 1.2 ppm bHB signal also arises from 3 protons, but is a doublet due to J-coupling, so the peak height is further reduced by a factor of two. The lack of correlation of bHB with systemic measures is likely the result of its lower sensitivity of detection by MRS, resulting in fewer available data points and greater variability, and is not necessarily indicative of a true lack of relationship between cerebral bHB and urine ketones or fasting glucose levels. The detectability of bHB may also be partially compromised by overlap with lipid resonances, or reduced signal intensity due to J-modulation effects if intermediate or long TE values are used. Therefore, from a technical viewpoint, Ace is the ketone body most favorable for detection by MRS.

The detectability of the 1.19 ppm resonance of bHB might be improved by using an optimized long-TE sequence (e.g. 1/J = 160 ms, J = 6.3 Hz) to discriminate it from other resonances such as lipid, however, this would also result in significant signal loss due to T₂ decay, and likely worsen detection of the small concentration singlet resonances of Ace and AcAc. Since lactate and bHB have similar coupling constants, the evolution of their signals as a function of TE is also very similar (see Supplementary Figure 3) so altering TE is unlikely to improve separation of bHB and Lac. However, at 3 T, the methyl doublets of bHB and Lac at 1.19 and 1.31 ppm are usually sufficiently well resolved to be individually quantified by the LCModel, except perhaps in regions of poor B₀ field homogeneity.

Assignment of the singlet resonance at 2.2 ppm to either Ace or AcAc is quite challenging at typical in vivo linewidths; the 0.05 ppm chemical shift difference between Ace and AcAc corresponds to approximately 6 Hz at 3 T, which is comparable to the spectral linewidths observed in this study (6.1±1.9 Hz, in lesion). However, LCModel fitting also makes use of the 3.43 CH₂ peak of AcAc which is not present in Ace, and in the current study few of the LCModel analyses could identify this resonance (however, this peak may also be difficult to detect because it potentially overlaps with other peaks such as sI, Tau and mI). Therefore, it seems likely that the 2.2 ppm peak arises from Ace, not AcAc; this is consistent with a previous study in patients with diabetic ketoacidosis which reported that bHB or AcAc can be converted to Ace quite rapidly ¹⁵.

Choice of CRLB thresholds for detection of metabolites

Traditionally, a CRLB threshold of 20% has been suggested for determining whether a metabolite concentration determined by the LCModel should be considered reliable or not ²⁴. However, recently it has been shown that using a low threshold value such as this may lead to bias and non-normal distributions of metabolite concentrations, particularly for changes in low concentration compounds ³⁰. Reported sensitivities of detection will vary according to the threshold chosen; the 80% value chosen here represents a reasonable compromise between under- and over-fitting the data and avoiding bias for the small concentration ketone concentrations. Hence, although Ace was measured in 5/10 contralateral spectra at baseline with this criteria, the mean concentration was over 2x lower as compared to week 8 spectra (n = 9/10), and the associated error in fitting this low concentration higher (64% vs. 39%). In addition, the Ace measurements correlated well with peripheral measures of ketosis, suggesting that the 80% cutoff chosen did not result in erroneous values being reported.

Comparison of ketone bodies in lesion and contralateral brain

One of the findings of this study was that lesion ketone levels were higher than in contralateral brain; this is consistent with a KD study using ¹³C MRS in preclinical brain tumor models ³¹ which found that the ketone body monocarboxylate transporter was upregulated, facilitating uptake and oxidation of ketone bodies in the tumors. Other underlying causes for this observation might be increased delivery of ketones (increased blood volume) and/or increased blood brain barrier permeability in the lesions. One previous MR spectroscopic imaging ³² study also found that Ace was more detectable in fluid filled spaces (such as ventricular CSF) compared to normal brain, perhaps due to longer T₂ in these regions. It is therefore also possible that the lesion spectra in the current study (usually placed in regions of T₂ hyperintensity on MRI) may also show increases in Ace signal for this reason.

Comparison of other brain metabolites in lesion and contralateral brain

The lesion spectra observed in this study are very consistent with those previously reported in the literature ³³. NAA was significantly lower in the lesion, consistent with neuroaxonal loss either in the tumor or peri-tumor regions, while Lac was significantly elevated due to non-oxidative glycolysis often seen in brain tumors. While it may seem surprising that Cho levels were not significantly elevated in the lesion, it should be remembered that all cases were scanned post-surgery and chemoradiation, and that in 5 of 10 cases a gross total resection was performed. In these cases, the lesion voxel placed adjacent to the surgical cavity may or may not contain any tumor tissue.

Effects of KD on other brain metabolites and brain MRI

Overall, most brain metabolites and anatomical MRI findings were stable over the relatively short 8-week period of KD. In particular, no changes in the lesion Cho and NAA levels suggest that there was minimal tumor progression over this time. There were also no changes in Lac, the end product of non-oxidative glycolysis, despite lowering of systemic glucose levels. Further study will be required to determine if baseline tumor lactate levels have any predictive value in determining response to KD.

Interestingly, there was a small but statistically significant decrease in tNAA in the contralateral hemisphere at 8 weeks. Since NAA synthesis occurs in mitochondria and is dependent on TCA cycle metabolism, it is possible that decreased blood glucose levels during ketosis lead to reduced TCA cycle flux and NAA synthesis. It should be noted that progressive tNAA reductions were previously reported in a teenager experiencing repeated episodes for diabetic ketoacidosis ³⁴, and reduced tNAA compared to normal controls was reported in patients with epilepsy on KD ¹⁸.

Alternatively, the decrease in NAA seen in the contralateral hemisphere may represent delayed, ongoing changes in systemic brain metabolism as the result of the prior chemotherapy and radiation. Further studies will be required to investigate the origin of this effect.

This study has a number of limitations, including the small number of subjects which precluded analysis of sub-groups of patients (segregated, e.g., by tumor grade or genetic mutation status). Another limitation was that no attempt was made to correct metabolite concentrations for voxel water (or CSF) content. However, we feel that this did not significantly affect the results, since voxels were carefully placed in either the solid part of the lesion, or in the contralateral hemisphere avoiding fluid filled spaces. Voxel locations were also carefully matched between initial and follow-up scans, and brain MR images were stable over the 8 week duration of the diet, so it is unlikely that voxel water content changed over this period. Finally, the small ketone signals are of similar magnitude to the noise in the spectra, and thus difficult to ascertain visually. Increased numbers of signal averages (and associated scan time) may have resulted in improved detection of ketones; however this was not possible in the current study since the MRS was just one part of a lengthy clinical research protocol containing multiple other sequences. Future studies might use longer acquisition times and larger voxel sizes to increase conspicuity of the ketone signals.

Conclusions

This study suggests that 3T MRS is feasible for detecting small cerebral metabolic changes associated with a ketogenic diet, provided that appropriate methodology is used.

Supplementary Material

Supplemental Figure 1

On-line Figure 1: Simulated basis spectra (at A: 1 Hz linewidth, B: 6Hz linewidth) used for fitting ketone bodies AcAc (acetoacetate), Ace (acetate) and bHB (beta-hydroxybutyrate), which was incorporated into LCModel fitting. Simulations are for semi-LASER acquisition with TE=34ms.

Supplemental Figure 2A

On-line Figure 2: Patient Spectra

Supplemental Figure 3

On-line Figure 3: Simulated spectral patterns for equimolar concentrations lactate and bHB at varying echo times (30ms – 170ms), neglecting T₂ decay. The methyl groups in bHB and Lac behave very similarly as a function of echo time, given the J-coupling constants of 6.3 and 6.94 Hz, respectively. There is no optimum TE to separate bHB and Lac, however they are resolvable at 3 T with the 6 Hz Lorentzian linewidths shown here. Simulations were performed using an ideal PRESS localization. 144ms indicates the TE commonly used for detection of lactate.

Abbreviations

KDs

Ketogenic diets

bHB

β-hydroxybutyrate

AcAc

Acetoacetate

Ace

Acetone

DKA

Diabetic Ketoacidosis

Lac

Lactate

CTCAE

Common Terminology Criteria for Adverse Events

sLASER

semi-LASER

CRLBs

Cramér-Rao lower bounds