Abstract
Background Bipolar disorder and epilepsy are treated with anti-seizure medications. Preliminary evidence indicates that a ketogenic diet (a metabolic treatment for epilepsy) may be effective in the treatment of bipolar disorder.
Aims To explore the impact of a ketogenic diet on clinical outcomes, and metabolomic and brain magnetic resonance spectroscopy biomarkers.
Methods Euthymic individuals with bipolar disorder (N=27) were recruited to a 6-8 week single-arm trial of a modified ketogenic diet and a range of metabolic and clinical outcome measures were assessed.
Results There was a positive correlation between daily ketone levels and ecological momentary assessments of mood (p < 0.001) and energy (p < 0.001) and an inverse correlation between ketone levels and impulsivity (p < 0.001) and anxiety (p < 0.001). Mean weight fell by 4.2kg (p < 0.001), mean BMI fell by 5.3% (p < 0.001) and mean systolic blood pressure was reduced by 7.4 mmHg (p < 0.041). Brain Glx (a putative marker of treatment response to anti-seizure medication) decreased by 13.1% in the posterior cingulate cortex (PCC) (p < 0.001) and by 9.2% in the anterior cingulate cortex ACC (p = 0.02). At the dietary cessation period, one third of participants, who reported reduced mood lability, opted to remain on a ketogenic diet.
Conclusion The majority of participants had improved cardiometabolic risk parameters and approximately one third experienced improvement in psychiatric symptoms. The study findings are consistent with our a priori hypothesis of ketone bodies acting as an alternative metabolic substrate under conditions of impaired insulin signalling in a subgroup of patients with bipolar disorder.
Practitioner Points A ketogenic diet may be effective in improving cardiometabolic risk factors and managing medication-induced weight gain. 95% of participants lost weight with a mean of 4.2kg (p < 0.001) over 8 weeks.
Ketone level was correlated to improved daily ecological momentary assessments of mood (p < 0.001) and energy (p < 0.001) and decreased impulsivity (p < 0.001) and anxiety (p < 0.001).
Participants who opted out of the dietary cessation period to remain on the diet, and who had reduced Affective Lability Scores, were characterised by more pronounced insulin resistance (HOMA-IR + 27.19%) and higher fasting insulin (+ 19.26%) at baseline.
Study Registration Number: ISRCTN61613198
Background
A ketogenic diet is established as a metabolic therapy for refractory epilepsy, supported by data from 13 randomised controlled trials and over a century of clinical use.1, 2 Preliminary data from case reports and retrospective analyses have suggested that a ketogenic diet may also have beneficial effects in bipolar disorder.3, 4, 5
Epilepsy and bipolar disorder may share some common pathophysiological mechanisms, and anti-seizure medications are effective for both conditions. Research into the influence of a ketogenic diet in neuropsychiatric disorders has highlighted effects that may be relevant to bipolar disorder pathophysiology, including neuroprotective, anti-inflammatory, metabolic and mood-stabilising properties.6 We have previously described a proposed mechanism of action through which a ketogenic diet may reduce deficits in brain cellular energetics due to impaired insulin signalling and dysregulated glucose metabolism.7 We sought to explore evidence for these ideas in this pilot study.
Bipolar disorder carries a substantial metabolic burden, with high rates of obesity, type 2 diabetes, cardiovascular disease and life expectancy that is reduced by at least 10 years.8, 9, 10 Several first-line medications for bipolar disorder increase cardiometabolic risk and so adjunctive treatment strategies to ameliorate these effects are urgently needed.11 A ketogenic diet has been demonstrated to improve cardiometabolic health outcomes in general population samples and may therefore have utility as an adjunctive treatment for SMI.12 However, there are currently no published trials of a ketogenic diet in bipolar disorder.
Our study was based on the hypothesis that a ketogenic diet might improve inter-episode mood stability and functioning in euthymic individuals with bipolar disorder. The study was conducted to assess feasibility and acceptability as well as effects on a range of clinical outcomes and metabolomic and brain magnetic resonance spectroscopy biomarkers, to inform the protocol for a larger phase 2 or 3 trial.
Methods
Participant details, recruitment, retention, feasibility and adherence in this study have been published previously.13 A brief summary is provided here alongside complete clinical, metabolomics and magnetic resonance spectroscopy (MRS) methods.
Participants
Twenty-seven participants aged 18-70 years who met DSM-IV diagnostic criteria for bipolar disorder for at least 1 year and were in a euthymic state were recruited.
Intervention
Participants followed the UK modified ketogenic diet formulated by an NHS dietitian for 6-8 weeks, with a brief adaptation period and 2-week dietary cessation period. The macronutrient ratios were 60-75% fat, 5-7% carbohydrate and the remainder of calories from protein.
Before starting the ketogenic diet intervention, participants underwent baseline assessments, including psychiatric assessment, fasting venepuncture and fasting brain imaging. Participants’ height, weight, and blood pressure were measured, and they received instruction on wearing an actigraph and using a glucose/ketone measuring device to record daily levels throughout the study period.
Measures of Mental Health
At baseline and follow up, medical and medication history were reviewed and a diagnostic interview was completed. Participants completed a range of mental health measures including Affective Lability Scale 18 (ALS); Beck’s Depression Inventory (BDI); and Young Mania Rating Scale (YMRS).13
Ecological Momentary Assessment and Analysis
Text messages were used for daily ecological momentary assessment (EMA) of mood, energy, speed of thought, impulsivity, and anxiety. Initial EMA rating instructions delivered to participants 1-12 were highlighted as difficult to follow due to varied interpretation of neutral, normal and pathological ranges on the 100 point scale. Thresholds for normal and pathological ranges were added to the instructions for patients 13-27 (see supplementary material) and EMA analysis was performed on this group.
Descriptive statistics, investigating correlations and (Granger) causality relationships between ketone levels (mmol/L) and self-reported mental health variables (mood, energy, thought, impulsivity and anxiety) were performed. The Dumitrescu-Hurlin test for Granger (non)causality in panel data was applied.14 Participants with more than 20 missing EMA ratings were excluded from the analysis. The remaining missing values were imputed using the Amelia R library15 which implements a multiple imputation method based on an expectation maximisation algorithm with bootstrapping. We tested individually for all mental-health response variables, namely: mood, energy, thought, impulsivity and anxiety and plotted the time series for averaged ketone and response variables to the same scale.
Brain Magnetic Resonance (MR) Imaging and Spectroscopy
Brain MR imaging and spectroscopy were acquired using a 3-T clinical MRI scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) with a 32-channel receive head coil. A 3D T1-weighted image (T1w; TR/TE/TI = 2500/4.37/1100 ms, flip angle = 7°, 1.0-mm isotropic resolution) was acquired to facilitate placement of MRS voxels (Figure 4) covering the anterior cingulate cortex (ACC; 20 x 20 x 20 mm3, 128 averages, acquisition time 4m26s), right dorsolateral prefrontal cortex (RDLPFC; 30 x 15 x 15 mm3, 96 averages, acquisition time 3m22s) and posterior cingulate cortex (PCC; 20 x 20 x 20 mm3, 128 averages, acquisition time 4m26s). The ACC and DLPFC were selected as regions of interest (ROIs) based on previous literature implicating these regions in the pathophysiology of BD,16, 17 whereas the PCC was selected as it typically produces high-quality MRS spectra, due to reduced magnetic susceptibility effects which can impact ACC and DLPFC scans. At follow up, voxels were automatically placed using the AutoAlign feature to replicate baseline placement. MRS was obtained using point-resolved spectroscopy (PRESS; TR/TE = 2000/30 ms, flip angle = 90°, receive bandwidth 4000 Hz, 4096 data points, with a single transient acquired without water suppression).
MRS data were processed using Osprey version 2.4.0,18 including frequency alignment and phasing of transients, to the T1w anatomical image and segmentation of tissues within the voxels. Data were fitted in the frequency domain using a linear combination modelling approach, including default basis functions for lipid and macromolecular resonances, in addition to ascorbate, aspartate, creatine, gamma-aminobutyric acid, glycerophosphorylcholine, glutathione, glutamine, glutamate, myo-inositol, lactate, N-acetylaspartate, N- acetylaspartylglutamate, phosphocholine, phosphocreatine, phosphoethanolamine, scyllo-inositol and taurine. Metabolite levels were quantified as water-scaled, relaxation-corrected and tissue-corrected estimates of molal concentration.19 Spectra were automatically excluded if either the Cr or water resonance linewidth (full width at half maximum) was equal to or greater than 0.1 ppm; in addition, spectra and model fits were inspected visually and excluded in case of poor model fit, excessive lipid contamination, spurious signals, low signal, poor water suppression or baseline distortion.
Metabolomics
Fasting blood samples were taken before and after the intervention and biochemical parameters were measured including HbA1c, C-reactive Protein (CRP), beta-hydroxybutyrate (BHB), insulin, glucose and lipid levels.
In a more detailed untargeted metabolomic analysis, metabolites were extracted from 25 µL serum by adding 1000 µL of ice-cold chloroform/methanol/water (1/3/1 ratio) and mixing vigorously on a vortex for 5 min at 4 °C. The extraction mixtures were centrifuged at 4 °C and 13,000 ×g for 3 min. A 200 µL aliquot of the extracts were transferred into sterile screw-top vials and stored at −80 °C until metabolomics analysis. The untargeted metabolomics analysis was performed using liquid chromatography (LC) coupled to ion mobility (IM) quadrupole time of flight (qTOF) mass spectrometry (MS) as described previously.20 Please see supplementary materials for further details.
Results
Summary of Feasibility and Adherence
Participants were considered adherent to the diet if their daily ketone levels reached 0.5 mmol/L or above, as measured by a glucose/ketone measuring device. Of all available daily readings for participants who completed the dietary intervention (N=20), 91% indicated ketone levels above 0.5 mmol/L, indicating a high level of adherence to a ketogenic diet. Of 27 recruited participants, 26 commenced the modified ketogenic diet and 20 completed follow up after 6 weeks on the diet.13
Mental Health Measures
Available case analysis of all participants who had baseline and follow up mental health measures was performed. The sample sizes for each variable are detailed in supplementary materials.
Participants were euthymic at baseline with the following median psychometric scores: BDI: 8.5 (IQR = 11), ALS: 15 (IQR = 16.5) and YMRS: 0. Subsyndromal symptoms within the euthymic state were measured over time using daily EMA. Median scores at week 8 remained within euthymic range with no significant changes: ALS: 15 (IQR = 16.5) to 12.5 (IQR = 13.25) (p = 0.80), YMRS: 0 to 0 (p = 0.86), and BDI: 8.5 (IQR = 11) to 9 (IQR = 9.25) (p = 0.55). Out of the 20 participants who reached the dietary cessation period at 8 weeks, over one third (n=7) chose to remain on the diet instead of being supported through a 2 week dietary cessation period. Participants who opted to remain on the diet had a statistically significant reduction in mean ALS score (18.2 ± 8.1 to 11.0 ± 6.8, p = 0.021) compared with no significant change in participants who did not continue with the diet (13.00 ± 10.87 to 15.67 ± 9.06, p = 0.177).
Over the 8 week period a significant positive correlation was observed between daily ketone level and mood (r = 0.21, p < 0.001) and energy (r = 0.19 p < 0.001) scores and a negative correlation of ketone level with impulsivity (r = -0.32, p < 0.001) and anxiety (r = -0.18, p < 0.001). No significant correlation was observed with speed of thought (r = -0.08 p > 0.05). (Fig.1).
Daily Ecological Momentary Assessment and Ketone Level Scatterplots
Time series were plotted to examine the trajectory of ketone levels and EMA scores throughout the 8 week intervention and 2 week dietary cessation period. Covariation between ketone level and mood, energy and anxiety level were observed (Fig.2), with indications of an effect becoming established over a 2 week adaptation period.
Time Series Plot of Ketone Level and Ecological Momentary Assessment Scores
The Dumitrescu-Hurlin test, an extension of the Granger causality test for panel data, was applied to determine if changes in ketone levels could predict changes in EMA scores. Significant Granger causality was observed for ketone level and mood (Z = 3.19 p < 0.01) and anxiety (Z = 2.29, p < 0.05). The time series in fig. 2 shows aggregated ketone levels and EMA scores over the sampling period.
Times series were plotted separately for the participants remaining on a ketogenic diet with reduced ALS scores (fig. 3). The Granger causality test for these participants indicated clearer patterns of significance for mood (Z = 4.45 p < 0.001) and anxiety (Z = 3.37, p < 0.001).
Time Series Plot of Ketone Level and Ecological Momentary Assessment Scores in Responder Group.
To further explore any relationship between daily metabolic status and mental health assessments, we investigated a metric of relevance to ketogenic metabolic therapy for epilepsy: the glucose ketone index (GKI): glucose level (mmol/L) divided by ketone level (mmol/L). Time series for GKI and all mental-health response variables were plotted to the same scale (see supplementary materials). This analysis revealed covariation and significant granger causality between GKI level and mood (Z = 2.32, p < 0.05), energy (Z = 2.28, p < 0.05) and anxiety (44.51, p < 0.001). We then explored GKI in the responder group and observed stronger covariation and evidence of Granger causality for mood (Z = 3.11 p < 0.01), energy (Z = 3.60 p < 0.001) and anxiety (Z = 55.14 p < 0.001).
Metabolic
Available case analysis of all participants who had baseline and follow up measures was performed. The sample sizes for each variable are detailed in the supplementary materials.
At baseline 70.3% of participants were obese or overweight and 29.7% were of normal or healthy BMI. 19 of 20 participants who reached follow up reduced their bodyweight. Mean body weight decreased by 5.2% (4.2kg) from 81.2 ± 13.2kg to 77.0 ± 11.7kg (p < 0.001). Mean BMI decreased by 5.3% from 28.2 ± 4.2 kg/m2 to 26.7 ± 3.9 kg/m2 (p < 0.001). Mean systolic blood pressure decreased by 7.4mmHg from 133.1 ± 16.2 mmHg to 125.7 ± 11.9 mmHg (p = 0.041). Mean diastolic blood pressure was 81.5 ± 12.20 mmHg at baseline and 77.4 ± 7.7 mmHg (p = 0.162) at follow up.
The changes in mean values of other metabolic parameters (reported below as mean values at baseline then at follow up) were not significant: fasting insulin 75.3 ± 44.0 to 68.4 ± 49.0 pmol/L (p = 0.540); fasting glucose 5.0 ± 0.6 to 4.7 ± 0.4 mmol/L (p = 0.12); HbA1c: 34.9 ± 2.7 to 33.9 ± 2.3 mmol/mol (p = 0.153); HDL cholesterol: 1.6 ± 0.43 mmol/L to 1.5 ± 0.51 mmol/L (p = 0.72); LDL cholesterol: 3.0 ± 0.71 mmol/L to 3.6 ± 1.4 mmol/L (p = 0.09); total cholesterol: 5.3 ± 0.9 mmol/L to 5.8 ± 1.7 mmol/L (p = 0.21); triglycerides 1.5 ± 0.8 mmol/L to 1.6 ± 0.9 mmol/L (p = 0.98); and CRP 2.4 ± 1.8 mg/L to 2.8 ± 3.6 mg/L (p = 0.738).
MR Spectroscopy (MRS)
Brain Metabolite Levels
Acetoacetate and acetone were largely undetected across ACC (acetoacetate: 0.05±0.09; acetone: 0.005±0.02), PCC (acetoacetate: 0.02±0.08; acetone: 0±0), and RDLPFC (acetoacetate: 0±0; acetone: 0±0) session one scans conducted before the KD-intervention commenced. Likewise, acetoacetate and acetone remained largely undetectable post-intervention across the ACC (acetoacetate: 0.09±0.13; acetone: 0±0), PCC (acetoacetate: 0.01±0.04; acetone: 0±0), and RDLPFC (acetoacetate: 0±0; acetone: 0±0). Conversely, small concentrations of beta-hydroxybutyrate were detected in both session one (ACC: 1.90±1.52; PCC: 0.67±0.57; RDLPFC: 0.51±0.41) and session two scans (ACC: 2.03±1.54; PCC: 0.69±0.65; RDLPFC: 0.33±0.51). Detected beta-hydroxybutyrate levels did not significantly differ between session one and session two scans. However, the reported concentration values likely include other neurometabolites due to spectral overlap, such as lactate, as we did not use spectral editing techniques to be able to confidently detect beta-hydroxybutyrate. Therefore, to prevent overfitting during linear combination modelling, ketone bodies were excluded from the basis spectrum.
Neurometabolite concentration estimates are reported in Table 1 and Figure 5. Glx (glutamate+glutamine) decreased significantly from baseline in both the ACC (concentration change: -1.50±3.92 95% CI [0.22,2.78] mM) and PCC (-2.14±4.81 95% CI [1.21,3.08] mM). A small, but significant, decrease in total choline was observed in both the ACC (-0.17±0.41 95% CI [0.06,0.28] mM) and PCC (-0.12±.27 95% CI [0.06,0.18] mM). A decrease in myoinositol was also observed in the PCC (-0.39±1.05 95% CI [0.06,0.72] mM).
Representative MR spectra (gold), fitted model spectra (blue) and model components (blue, below) from a single participant before and after ketogenic diet intervention. NAA = N-acetylaspartate. Glx = glutamate + glutamine. tCr = total creatine. tCho = total choline. mI = myo-inositol.
Metabolite concentrations in the DLPFC, ACC and PCC, as measured by MR spectroscopy before and after ketogenic diet intervention. Red point = mean. *=p<.05. **=p<.005 (FDR corrected).
An analysis to explore changes to glutamate (Glu) and glutamine (Gln) separately was carried out and revealed a significant decrease in Glu in the ACC (-0.93±2.52 95% CI [0.06,1.79] mM) and PCC (-1.04±2.36 95% CI [0.55,1.52] mM). Additionally, a significant decrease in Gln was observed in the PCC (-1.11±2.54 95% CI [0.57,1.65] mM).
No significant differences were observed in the right DLFPC. After using the false discovery rate correction to correct for multiple comparisons, significant differences before and after intervention remained only for total choline in the ACC and total choline, Glx, Glu, Gln, and myoinositol in the PCC (Table 1).
No significant differences in brain metabolite concentrations pre-and post-intervention were observed between those defined as good and poor responders across ROIs. Please see supplementary materials for details of MRS Quality Assessment.
Serum Metabolomics
Global metabolomics analysis was performed on a total of 36 serum samples, corresponding to 18 participants who gave a baseline and follow up blood sample. Rapid HILIC-Z Ion mobility mass spectrometry (RHIMMS) analysis was used to probe the serum metabolome and a total of 358 metabolic features were annotated. As expected with human metabolome studies, large variations were noted between the metabolic profiles of individual participants. However, multivariate statistical analysis of the baseline and follow-up sample groups showed a clear separation of the study groups in PLS-DA, suggesting that reliable trends in participant metabolism can be observed within this dataset (see supplementary materials).
The serum metabolome analysis confirmed ketogenesis in all participants, except one. Increased levels of the ketone bodies, β-hydroxybutyrate (p = 8.84E-06) and β-ketopentanoate (p = 0.361), were found while levels of acetoacetate (p = 0.054) were reduced in the follow up samples (Figure 6 b-d). Acetoacetate is the direct precursor to β-hydroxybutyrate and can be non-enzymatically broken down to acetone, which could not be unambiguously annotated in the dataset and therefore is not reported here. Probing central carbon metabolism involving the most common energy generating pathways in the body, revealed a decrease in lactate (p = 0.02) and increase in citrate (p = 0.02) levels in the follow up samples (Figure 6 e-g). The statistically significant decrease in levels of myo-inositol and glutamine in brain imaging studies were not found in the serum metabolome, although (non-significant) differences between baseline and follow up samples were in the same direction. (Figure 6 h-i).
(a) Partial Least Squares—Discriminant Analysis (PLS-DA) score plot shows separation of the serum metabolome at baseline (red circle) and follow up (blue triangle). The first and second components explain 7.4 and 3.6% of the variation, respectively. (b-i) The boxplots represent peak intensities of metabolites for baseline (red) and follow up (blue) groups (N = 18). Metabolites with statistically significant (p < 0.05) differences between the groups are highlighted in blue text.
Discussion
Mental health outcomes
Over the 8 week period, significant positive correlation was observed between daily ketone level and daily ecological momentary assessments of mood (r = 0.21, p < 0.001) and energy (r = 0.19 p < 0.001) scores and there was a negative correlation between ketone levels and both impulsivity (r = -0.22, p < 0.001) and anxiety (r = -0.18, p < 0.001). There are indications from the literature that the strength of the anti-seizure effect of ketosis is related to ketone and glucose levels,21 and guidelines are provided by clinicians and dieticians on achieving optimal ketone and glucose ranges. For example, the recommendation from Liao et.al to maintain blood ketone level around 2.0 - 4.0 mmol/L, blood glucose around 4.0 mmol/L and ratio of blood glucose/ketone level (GKI) between 1:1 - 2:1.22
Visual inspection of the scatterplots shows some evidence of a threshold at around 2mmol/L with fewer scores below 50 for mood and energy and fewer scores above 50 for anxiety and impulsivity above 2mmol/L. A threshold effect at approximately 4mmol/L has been noted for seizure reduction in children with epilepsy, with more than 90% seizure control occurring for only 8% of patients below 4mmol/L and 48% of patients over 4mmol/L.21 However, children appear to more readily achieve high ketone levels, and the threshold is likely to be significantly lower in older age groups.
In order to assess association between ketone level and EMA scores in more detail, time-series of ketone levels and EMA scores were plotted and Granger causality test applied. We observed co-variation of time-series plots of ketone level with EMA scores and significant Granger causality for mood (Z = 3.19 p < 0.01) and anxiety (Z = 2.29, p < 0.05) suggesting that changes in ketone level may predict changes in EMA scores. Evidence of an initial adaptation period when initiating ketosis (which we have previously noted from patient reports)23 was observed, with onset of effects becoming established within the first 10-15 days.
Epilepsy research indicates that glucose ketone index may be more strongly correlated to seizure control than either ketone or glucose level alone.24, 25 In an analysis of GKI we observed covariation and significant granger causality between GKI level and mood (Z = 2.32, p < 0.05), energy (Z = 2.28, p < 0.05) and anxiety (44.51, p < 0.001) and particularly in the responder group who remained on the diet (mood (Z = 3.11 p < 0.01), energy (Z = 3.60 p < 0.001) and anxiety (Z = 55.14 p < 0.001)). The GKI plots exhibit further evidence of an adaptation period where mood and energy increase as GKI falls (indicating an increasing ratio of ketones to glucose). Conversely, in the same period, anxiety falls as GKI decreases.
Out of the 20 participants who reached the dietary cessation period at 8 weeks, one third (n=7) chose to remain on the diet instead of being supported through a 2-week dietary cessation period. Participants who opted to remain on the diet had a statistically significant reduction in mean ALS score (18.2 ± 8.1 to 11.0 ± 6.8, p = 0.021) compared with no significant change in participants who did not continue with the diet (12.92 ± 10.87 to 15.67 ± 9.06, p = 0.177).
Given that the intervention represents a substantial commitment to participants, we carried out further exploratory analysis on this “responder” group who had reduced ALS scores and opted to remain on the diet. The largest differences in the baseline responder and non-responder values were HOMA-IR (+27.19% in responders) and insulin (+19.26% in responders), indicating more significant insulin resistance in this group. Responders also exhibited elevated HbA1c and glucose compared to non-responders, but not weight or BMI (see supplementary materials for details). Higher glucose and HbA1c may further indicate a role of insulin resistance in this group.
We have previously hypothesised that disruption of PI3K/AKT insulin signalling in the brain, may underpin a metabolic phenotype of bipolar disorder7 and have proposed that Lithium may act on insulin signalling pathways.26 However, it should be noted that markers of peripheral insulin resistance such as HOMA-IR do not necessarily indicate the presence of insulin resistance in the central nervous system (CNS).
Our observations in this study are consistent with several preliminary case reports and observational studies noting improvement of psychiatric symptoms in bipolar disorder and schizophrenia patients on a ketogenic diet 3, 4, 27, 28, 29, 30
Metabolic
Significant changes in risk factors for cardiometabolic disease were observed among the participants, 70% of whom were overweight or obese. Participants reduced their mean weight by 4.2kg, mean BMI by 5.3% and mean systolic blood pressure by 7.4 mmHg. Changes in other metabolic parameters were not statistically significant in this study. Metabolic dysfunction (manifesting as insulin resistance, metabolic syndrome, type 2 diabetes and related conditions) is highly prevalent in people living with bipolar disorder.10 The effects of metabolic dysfunction on peripheral tissues of the body are well characterised. However, effects on the brain and CNS are only beginning to be understood, and may encompass some aspects of the pathophysiology of serious mental illness. New and effective metabolic interventions are required solely on the basis of reducing the cardiometabolic burden of SMI. However, it is also possible that such interventions may prove effective in addressing psychiatric symptoms by improving metabolic function of the brain and CNS.
MRS
MRS brain imaging was performed to assess whether a ketogenic diet may change neurometabolite levels in BD. Elevated glutamate (an excitatory neurotransmitter) is among the most consistent findings in brain MRS studies of bipolar disorder.31, 32, 33 Glutamatergic mechanisms have been a focus of pharmaceutical research in both epilepsy and bipolar disorder (which share anti-seizure medications such as Lamotrigine) and altered glutamate metabolism is proposed as a mechanism of action of several anti-seizure medications.
Glx (glutamate+glutamine) decreased from baseline by 9.22% in the ACC and by 13.13% in the PCC, total choline decreased from baseline by 3.53% in the ACC and 4.86% in the PCC, and myoinositol decreased by 3.77% in the PCC. Furthermore, glutamate alone decreased from baseline by 8.31% in the ACC and 8.23% in the PCC, whereas glutamine alone decreased from baseline by 34.47% in the PCC. The reduction in Glx observed in the RDLPFC did not reach conventional statistical significance. This may reflect lower spectral quality obtained for this ROI, which is challenging to place accurately and was acquired late in the examination when participant movement is more prevalent. Our observation of decreased Glx post-intervention is in line with previous research detecting a reduction in glutamate, glutamine, and Glx in people with bipolar disorder who were good responders to both pharmacological interventions34, 35, 36, 37,38 and total sleep deprivation and light therapy39, 40 interventions. Therefore, acute treatment-induced brain glutamate reduction could be explored as a potential biomarker to predict treatment response.
We recently described a novel hypothesis of the role of glutamate in bipolar disorder,41 noting that under conditions of impaired oxidative glucose metabolism, glutamate, generated in states of hyperglycolysis, may act as a metabolic substrate in neurons. Glutamate is readily converted to alpha-ketoglutarate, a citric acid cycle intermediate, and may thereby participate in cerebral metabolism. Ketone bodies fulfil a similar metabolic role via conversion to the citric acid cycle intermediate acetyl-CoA. According to this hypothesis, under conditions of impaired glucose metabolism ketones act as an alternative metabolic substrate in neurons, precluding the need for upregulation of glutamate metabolism.
The reduction in PCC myo-inositol is also interesting as myo-inositol has also been a significant focus of MRS research in bipolar disorder. One hypothesis regarding the mechanism of action of lithium posits that lithium depletes myo-inositol, a sugar alcohol with important signalling functions in the brain.42 Myo-inositol plays a central role in the phosphatidylinositol 3-kinase insulin signalling mechanism we have proposed to underlie dysregulation of glucose metabolism in a sub-group of bipolar disorder patients.7
We did not reliably observe signals corresponding to β-hydroxybutyrate, acetone and acetoacetate in the brain, and secondary analyses that included these metabolites in the basis set did not detect changes following intervention. We therefore excluded these ketone bodies from the basis set to avoid overfitting the spectra. β-hydroxybutyrate was reliably detected in blood plasma samples, and we speculate that concentration of ketone bodies in the brain is low due to rapid metabolism of these energy substrates. In addition, the characteristic β-hydroxybutyrate doublet signal at 1.2 ppm overlaps with macromolecular and lipid resonances and may be difficult to quantify reliably using short-echo-time MRS. Lactate, another possible alternative energy substrate has a characteristic doublet resonance at 1.3 ppm, and is also challenging to detect. Therefore, future MRS studies could utilise longer echo times or spectral editing techniques to investigate changes in β-hydroxybutyrate and lactate concentration.
Metabolomics
Systematic review of metabolomics biomarkers in bipolar disorder indicates that lactate is consistently found to be increased, and that it is among the most significantly altered biomarkers in people with bipolar disorder compared with controls.43 Increased lactate is observed in several mitochondrial disorders and is used as a diagnostic marker, particularly in disorders with accompanying neurologic symptoms.44 We specified lactate as a biomarker of interest to our study hypothesis a priori noting that increased level can indicate a disruption of oxidative phosphorylation and a compensatory increase in glycolysis. We have described mechanisms through which ketones may restore oxidative phosphorylation and reduce glycolysis, thereby reducing lactate.7, 23
A significant reduction in lactate, an increase in citrate (which participates in oxidative phosphorylation through the tri-citric acid cycle) and an increase in ketone body beta-hydroxybutarate were among the significant metabolomics findings of the study. These findings may indicate that ketosis is associated with improved mitochondrial function and a shift toward improved oxidative phosphorylation and reduced glycolysis.
Limitations
There are a number of important limitations to report. This pilot experimental study was designed to generate data to inform the planning of a subsequent larger scale clinical trial. Rather than detailing a number of specific research questions a priori, we approached this study as one which would generate mental health and metabolic data which could be used to describe changes associated with the ketogenic diet to explore “signals of effect” and generate hypotheses to be addressed in future larger studies. The study was not powered to demonstrate statistically significant differences in mental health or metabolic outcomes.
Participants were recruited from a bipolar disorder charity and volunteered to participate in the study. Therefore, these participants may not represent the response of patients recruited from a clinical setting. The limited sample size results in less precise parameter estimates, increased risk of type 1 errors and false negative findings in comparisons between groups. In addition, we only conducted clinical assessments (including administering psychometric scales) on two occasions, at baseline and follow up, which does not account well for the variability in these measures. Blood pressure readings were taken as single baseline and follow up readings, and should therefore be interpreted with caution. Despite this, some changes in metabolic and mental health parameters were large enough to be noteworthy even in this small study. After analysing the MRS and metabolomics data, we discovered that one participant had not adhered to the fasting requirement in the protocol. Despite this, their data did not present as an outlier and had no significant impact on the overall results. Consequently, we excluded their data from the subsequent analysis reported in the "Metabolic" results section but retained it in the completed MRS and metabolomics analyses.
The data derived from actigraphy had significant day-to-day variability both within and between subjects, precluding any assessment of the effect of the diet intervention on sleep and/or rest-activity rhythmicity.
The design of the experimental study was single-arm with no control group and was unblinded [for both participants and research staff] and so results may be subject to bias. Dietary interventions, such as the ketogenic diet, are not possible to blind in a trial which is an important limitation. However, we suggest that the Granger causality analyses which assess the relationship between daily ketone and GKI levels and mental health parameters are less subject to bias and give support to our interpretation of the data.
We did not specify detailed research questions to address a priori so that some of the changes that are reported are generated from post hoc analyses and so may be subject of over-interpretation. Nevertheless, our analyses addressed hypotheses on deficits in insulin signalling pathways and glucose metabolism that we have published a priori and were referenced in the study protocol and ISRCTN registration.
We have not corrected for multiple testing throughout most of the study (false discovery rate was employed in the MRS analysis) but rather we present these data as exploratory and requiring further validation in larger studies.
Conclusions
This was a pilot study of a novel ketogenic diet intervention, seeking to generate data to inform planning for a future trial and metabolic studies investigating underlying mechanisms of action. Most of the participants had improved cardiometabolic risk parameters and approximately one third experienced improvements in mood stability as measured by Affective Lability Scale, sufficient that they opted to remain on a ketogenic diet at the end of the study. This subgroup was characterised by elevated HOMA-IR, fasting insulin, glucose and HbA1c levels at baseline. Daily EMA self-reports of mental health status demonstrated significant correlations with ketone levels and significant granger causality for mood and anxiety, supporting the inference that ALS changes may have been mediated by the ketogenic diet intervention rather than biases in an unblinded study. Some evidence of a possible threshold effect above 2 mmol/L ketone level was observed in the EMA data and should be further explored.
Given the substantial cardiometabolic risk associated with bipolar disorder and the urgent need for new adjunctive (and non-medication) treatment strategies, a randomised controlled trial is now warranted to establish whether a ketogenic diet may be a viable treatment option for some people with bipolar disorder. The study findings are consistent with our a priori hypothesis of there being a metabolic phenotype of bipolar disorder, perhaps characterised by impaired insulin signalling in the brain and CNS.7
Declaration of Interest
Dr Iain Campbell has a diagnosis of bipolar disorder and follows a ketogenic diet to manage his symptoms. His current fellowship is funded by the Baszucki Brain Research Fund.
Funding
This study was funded by the Baszucki Brain Research Fund. MJT acknowledges funding from the Scottish Chief Scientist Office through the NHS Lothian Research and Development Office.
Author Contributions
IC, HC, and DS were responsible for conceptualising the research question. IC, NN, HG, SS, EM, TM, KB, MT, MM-G, JN, AM, CF, HC and DS were responsible for designing the study. NN, HG, BM, FC were responsible for data collection. IC, HG, IK, SL, SS, GT, MT, MG, TM, KB, HC and DS were responsible for analysis and interpretation of the data. IC, MT, MG, TM, KB were responsible for drafting the paper. All authors were involved in subsequent revisions, and agree to be accountable for all aspects of the work.
Data availability statement
The data that support the findings of this study are not publicly available as explicit consent was not sought from participants.