Whole Brain Neuronal Abnormalities in Focal Epilepsy Quantified with Proton MR Spectroscopy

1. Introduction

Epilepsy, a chronic neurological condition characterized by recurrent seizures, affects ~50 million worldwide, nearly 1% of the population (Walsh et al., 2016). Its most common adult form is localization-related epilepsy, in which EEG and MRI can localize an epileptogenic focus, often within the temporal lobe. Despite its inferred localized nature, however, it is well documented that abnormalities can extend beyond the epileptogenic zone (Doucet et al., 2016; Hetherington et al., 2007; Ji et al., 2013). Indeed, the contemporary view of epilepsy is of a network disorder, with EEG, neuroanatomical and functional changes consistent with widely distributed dysfunction. Its pathophysiologic nature has multiple hypotheses, including overlapping contributions from inflammation (Butler et al., 2016), aberrant neurocircuitry (Ji et al., 2013), redistribution of excitatory and inhibitory function (Chowdhury et al., 2015), and metabolic dysfunction (Hetherington et al., 2007; van Veenendaal et al., 2015). For the latter, there is strong clinical and experimental evidence; e.g., FDG-PET use is predicated on identification of hypometabolic regions as possible seizure onset sites (Butler et al., 2016; Hodolic et al., 2016). Experimentally, mitochondrial dysfunction was found to relate to cellular excitability via calcium mishandling (Kann and Kovacs, 2007). Furthermore, reactive oxygen species that contribute to mitochondrial dysfunction, are increased in epilepsy models (Rowley et al., 2015).

In this report, we use proton MR spectroscopy (¹H-MRS) to assess the epileptic brain. This modality facilitates in vivo evaluation of brain metabolism, most prominently, N-acetyl-aspartate, total creatine and choline (NAA, Cr and Cho), used as surrogate markers for neuronal integrity, energy metabolism and membrane-turnover (Benarroch, 2008; Mountford et al., 2010; Zhu and Barker, 2011). The most common single- and multi-voxel variants of localized ¹H-MRS, however, are subject to some limitations: (a) they cover ‹10% and rarely up to 65% of the brain (Govind et al., 2010; Posse et al., 2013), precluding assessment of total disease loads (Rigotti et al., 2007); (b) therefore, must rely on an assumption that metabolic changes occur at MRI-visible abnormalities, missing MRI-occult pathologies; (c) are prone to repositioning errors; (d) require long, 4 – 30 minutes, acquisition to achieve sufficient signal-to-noise ratio (SNR); and (e) assume same T₁ and T₂ relaxation-times in patients and controls that could bias metabolic quantification (Zaaraoui et al., 2007).

We address all these issues with a non-localizing ¹H-MRS sequence that acquires the whole-head signal (Gonen et al., 1998; Soher et al., 2014), and that has recently been shown to yield contributions from only the whole-brain (WB), the compartment of interest (Kirov et al., 2017). In this paper, we use the recently developed full spectral-modeling for WB ¹H-MRS (Soher et al., 2014), in order to test the hypothesis that despite lack of atrophy, brains of patients with localization-related epilepsy incur neuronal dysfunction well beyond the ictal zone, reflected by a decrease in their global NAA concentration. To this end, we compared patients to matched healthy controls in terms of their WB ¹H-MRS NAA, Cr and Cho concentrations, as well as total brain volume, gray and white matter (GM, WM) fractions.

2. Materials and Methods

2.1. Human subjects

Thirteen patients diagnosed with localization related epilepsy (7 men, 6 women) mean age 40±13, range 23 – 64 years old, 8.3±13.4 years disease duration (range: 1–45 years), were recruited prospectively and scanned interictally. Localization-related seizure onset was defined based on seizure semiology, MRI, scalp EEG and video monitoring, where available. None had known neurological co-morbidities or history of substance abuse. Their demographics, are given in Table 1. Fourteen age and sex matched controls (7 men and 7 women, age 38±11, range 22–63 years old) also underwent the same protocol. The study took place at an outpatient MRI facility between 12/2014 and 7/2016. The study was approved by the Institutional Review Board and written informed consent was obtained from all participants.

Table 1:

Patient demographics and clinical history.

Pt. ID	aAge/ sex	bOnset age	Seizure frequency	cSeizure Locus	Medication	Surgical History	Last Seizure to Scan (months)
1	26/M	25	Auras weekly	Bi-temporal	Levetiracetam & Lacosamide	no	0.25
2	29/F	29	Seizure-free	L temporal	Valproic acid	no	2
3	58/F	54	Seizure-free	L temporal	Lamotrigine	no	3
4	60/M	58	Seizure-free	R temporal	Valproic acid	no	>24
5	43/F	42	Seizure-free	R temporal	Lamotrigine	no	>24
6	28/F	24	1–4 per month	L temporal	Levetiracetam	no	unknown
7	30/M	25	0–1 per month	R frontal	Oxcarbazepine & Levetiracetam	no	0.25
8	46/F	43	2 CPS/month	R temporal	Eslicarbazepine acetate & Lamotrigine	Right craniectomy for placement of subdural electrodes, grid, and depth electrodesd	0.25
9	54/M	early 20s	1–2 auras/week	R centro-temporal	Lamotrigine	Right temporal lobectomye	0.5
10	23/F	15	0–1 per month	R frontal temporal	Lamotrigine & Carbamazepine	no	2
11	28/F	Unknown	1 per month	R mesial temporal	Lamotrigine	no	unknown
12	53/F	8	Seizure-free	Bitemporal R>L	Lamotrigine & Primidone	no	>24
13	41/F	40	Seizure-free	L temporal	Lamotrigine	no	6

^aage (years) and sex (F=female, M=male);

^bage at seizure onset (years);

^cR=Right, L=Left.

^d~1 year ago;

^e~24 years ago.

Note: (a) that only 3 of the patients had a disease duration of ›5 years; and (b) the surgical history of patients #8 and 9 does not affect the metabolic concentrations, because they are corrected for the (remaining) brain volume via Eq. [3]; see also (Cohen et al., 2005).

2.2. MRI

All experiments were done in a 3 T whole-body MR/PET scanner (Biograph mMR, Siemens AG, Erlangen, Germany) with a quadrature transmit-receive head-coil (TEM3000, MRInstruments, Minneapolis, MN) producing ±15% radio-frequency (RF) field homogeneity over the brain (Soher et al., 2014). Subjects were placed head-first supine into the scanner and its magnetic field (B₀) homogeneity over the brain (excluding extraneous tissue) optimized by our non-iterative map based BOLERO (B0 loop encoded readout) software automatically adjusted its 1^st and 2^nd order shims (Hetherington et al., 2006). It was followed by sagittal 3D T₁-weighted Magnetization-Prepared RApid Gradient-Echo (MP-RAGE) MRI: TE/TI/TR=2.6/800/1360 ms, 256×256 matrix, 256×256 mm² field-of-view, 160 1 mm thick slices, formatted into 192 axial images at isotropic 1 mm³ resolution, for tissue segmentation.

2.3. Brain Volumetry

To determine the WB tissue volumes, the MP-RAGE MRI was segmented into cerebro-spinal fluid (CSF), GM and WM masks using SPM12 (Ashburner and Friston, 2005), as shown in Fig. 2, in 5–7 minutes on an i7 class PC workstation. Their global volumes were obtained from the masks with our custom IDL (ITT Visual Information Solutions, Boulder CO) software:

$$V_{WM}=v\cdot \sum _{\begin{array}{c}i=all\\ pixels\end{array}}{p}_i(WM),V_{GM}=v\cdot \sum _{\begin{array}{c}i=\text{all}\\ pixels\end{array}}{p}_i(GM),V_{CSF}=v\cdot \sum _{\begin{array}{c}i=all\\ pixels\end{array}}{p}_i(CSF)\left[{\text{mm}}^3\right],$$ [1]

where V_WM, V_GM, V_CSF and v are that subject’s global WM, GM, CSF and pixel (1 mm³) volumes and p_i-s the mask’s i-th pixel probability to be of that tissue type: p_i(WM)+p_i(GM)+p_i(CSF)=1 inside the brain, “0” outside. Our software then estimated the brain volumes (V_B=V_GM+V_WM) and the brains’ global, GM and WM volume fractions (fBV, fGM, fWM), by dividing their absolute values, from Eq. [1], with the total intracranial volume – ICV (=V_GM+V_WM+V_CF). The fractional metrics, fBV, fGM, fWM are brain-size independent, suitable for inter-subject comparisons.

Fig. 2. Brain segmentation masks.

Axial, T1-weighted, MP-RAGE slice (a), and the GM (b), WM (c) and CSF (d) tissue masks derived from it with SPM12. Note the overall tissue-differentiation performance; and that the masks are probabilistic, i.e., p_i(WM+p_i(GM)+p_i(CSF)=1 in the i-th pixel in the brain, and “0” outside.

2.4. Whole brain (WB) non-localized ¹H-MRS

Subsequently, the WB amounts of NAA, Cr and Cho were obtained with a non-localizing, TE/TI/TR=5/940/10⁴ ms ¹H-MRS sequence, described previously (Gonen et al., 1998; Soher et al., 2014). The long, TR»T₁, and short, TE≈5 ms, of this sequence minimize sensitivity to the metabolites’ T₁ and T₂ variations, which are often unknown, especially in patients (Hovener et al., 2008). The acquisition takes less than 3 minutes.

The WB ¹H-MRS spectral fitting were done with the Versatile Simulation, Pulses and Analysis (VeSPA) package [https://scion.duhs.duke.edu/vespa/project (Soher et al., 2011; Soher et al., 2014)]. Its VeSPA-Analysis application applies a standard set of preset processing and spectral fitting parameters to fit the data parametrically, as shown in Fig. 1 (Soher et al., 2014; Soher et al., 1998; Young et al., 1998). It yields the relative levels of the i^th (=NAA, Cr, Cho) metabolite in the j^th (=1..27) subject, S_ij, in 1–2 minutes on an i7 class PC workstation. The fit basis set (synthesized by the VeSPA-Simulation application using the MRS sequence actual RF pulses and timings) comprised total-NAA [NAA+ NAA-glutamate at 7:1 ratio (Pouwels and Frahm, 1997)], Cho, Cr, glutamate+glutamine (Glx) and myo-inositol (mI). [Note that the last two were only included in the simulation to more accurately account for the total spectral lineshape; however, because the mI peak is partially suppressed by the bandwidth of the 1331 readout pulse of the sequence (Soher et al., 2014) and the Glx coefficient of variation is ~30% larger than the major metabolites’ (Davitz et al., 2017), both were excluded from the results]. S_ij were scaled into absolute amounts, Q_ij, against a 2 L sphere of Q_i^vitro=25, 20, and 6 millimoles NAA, Cr and Cho in water at physiological ionic strength:

$$Q_{ij}=Q_i^{vitro}\cdot \frac{S_{ij}}{S_{iR}}\cdot \frac{V_k^{{180}^{°}}}{V_R^{{180}^{°}}}\text{millimoles},$$ [2]

where S_iR is the VeSPA-Analysis derived sphere’s metabolites’ signals; and V_j^180°, V_R^180° subject and sphere RF voltages for non-selective 1 ms 180° pulses. (No T₁ or T₂ weighting corrections made due to the sequence’s long TR and very short TE). To account for variations in brain size, each Q_ij was divided by that individual’s V_B, to yield the WB concentration:

$$C_{ij}.=Q_{ij}/V_B\text{millimolar}(\text{mM}).$$ [3]

Fig. 1. Whole-brain (WB) spectra.

Top: WB ¹H-MRS from three patients: numbers 5, 7 and 13 in Table 1 (thin black line), all on the same frequency and intensity scales, overlaid with their VeSPA fit (thick gray line), baseline (dashed black line) and residual (raw – fit) from 0 – 4 ppm.

Bottom: Same from their age and gender matched controls (numbers 18, 22 and 23 in Table 2). Note (i) the small residuals (reflecting VeSPA fit’s quality), that are likely due to the limited number of metabolites in the fitting basis set; and possible lineshape differences between the measured in vivo peaks and the assumed Voight basis functions used by the VeSPA software; (ii) excellent lipid suppression performance; and (iii) while the acquisition is non-localizing, the metabolites’ spectra come from the brain only (Kirov et al., 2017).

These C_ijs are therefore specific metrics, suitable for inter-subject comparisons, and are also insensitive to tissue losses, e.g., due to resections, as shown previously (Cohen et al., 2005).

2.5. Statistical analyses

The patients and controls groups were compared in terms of gender using a Fisher exact test. A Schapiro-Wilk test showed that all imaging measures could be reasonably assumed to follow a normal distribution within each group except for fBV, fGM and fWM within the patient group. As a result, the groups were compared in terms of each metric using an unequal variance t test with results confirmed by a Mann-Whitney test to account for any violation of the t test assumptions. Analysis of covariance was used to compare the groups adjusted for age and gender with a Shapiro-Wilk test confirming the underlying assumption that the model residuals measures could be reasonably assumed to follow a normal distribution. All statistical tests were conducted at the two-sided 5% significance level using SAS 9.3 (SAS Institute, Cary, NC).

3. Results

Women comprised 50% (7/14) of the controls; and 46% (6/13) of the epilepsy patients. The Fisher exact test indicated the groups were not different in terms of gender composition (p›0.2).

3.1. Brain Volumetry

All 13 patients and 14 controls yielded MP-RAGE images of sufficient quality for brain segmentation. The ICV, GM and WM volumes (mean±SD) were: 1433±109, 664±109 and 320±107 cm³ for the patients; and 1521±144, 696±64 and 475±64 cm³ for the controls, as shown in Table 2. The fBV, fGM and fWM were: 0.81±0.07, 0.47±0.04, 0.31±0.04 in the patients and 0.79±0.05, 0.48±0.04, 0.32±0.02 in the controls. Their distributions, are shown in Fig. 3a-c. The patients were not statistically different from the controls in any of these metrics (p›0.05 for all).

Table 2:

Quantitative MRI and MRS metrics of patients (#1 – 13) and matched controls (#14 – 27).

Subjects		Volumes (cm3)			Concentrations (mM)
aNo./Cohort	Age/sex	bVGM	bVWM	bVCSF	cNAA	cCho	cCr
1/P	26/M	771	534	420	11.7	7.8	1.4
2/P	29/F	659	455	356	14.1	8.2	1.5
3/P	58/F	580	393	244	13.3	7.4	1.9
4/P	60/M	669	561	574	8.9	4.8	1.2
5/P	43/F	601	441	246	11.6	6.2	1.5
6/P	28/F	819	506	278	12.0	8.5	1.2
7/P	30/M	755	481	466	11.5	6.7	1.2
8/P	46/F	716	468	259	11.4	7.2	1.2
9/P	54/M	581	441	277	11.4	6.7	1.0
10/P	23/F	832	537	296	9.6	7.3	1.4
11/P	28/F	589	377	257	9.8	5.7	1.0
12/P	53/F	434	182	374	-	-	-
13/P	41/F	604	386	166	12.4	8.4	1.3
Mean:	40±13	664±109	320±107	324±111	11.5±1.5	7.1±1.1	1.3±0.2
14/C	28/F	752	446	214	13.0	7.4	1.4
15/C	51/M	736	532	452	12.7	7.6	1.4
16/C	40/F	754	521	330	13.2	8.1	1.4
17/C	31/M	777	536	413	11.7	6.8	1.2
18/C	31/M	724	452	340	13.8	8.1	1.4
19/C	33/F	566	325	391	13.2	7.7	1.4
20/C	28/M	722	529	369	12.8	7.5	1.3
21/C	33/M	689	531	353	12.6	7.3	1.3
22/C	45/F	656	442	395	14.2	8.8	1.6
23/C	43/F	644	446	355	13.9	8.1	1.4
24/C	62/M	667	432	271	11.6	7.2	1.3
25/C	58/M	632	377	244	12.5	8.5	1.6
26/C	26/M	762	454	245	14.0	7.8	1.5
27/C	29/F	711	474	236	13.2	6.2	1.2
Mean:	38±11	696±64	475±64	329±76	13.0±0.8	7.7±0.7	1.4±0.1

^aP=patient, C=control,

^bvolume in cm³,

^cconcentrations in millimolar (mM).

Note: that the order of the patients is the same as in Table 1; and that only the NAA concentration is statistically different (12% lower) from the controls.

Fig. 3. Brain volumetry and Whole-Brain (WB) metabolic concentrations.

Left: Box plots showing the first, second (median) and third quartiles, ±95^th percentiles (whiskers), outliers (*) and means (□) of the fBV (a), fGM (b) and fWM (c) in the N=13 epilepsy patients and their N=14 matched controls. Note the (i) median and mean proximity, suggesting normal distributions; (ii) the outlier is the same patient in all three metrics - #11 in Table 1; (iii) the lack of statistical differences between the two cohorts, suggesting no atrophy, globally or in the GM or WM moieties of the patients.

Right: Box plots showing the global WB NAA (d), Cr (e) and Cho (f) concentrations distribution in the N=12 epilepsy patients with usable MRS data, and their N=14 matched controls. Note that (i) only the WB NAA is statistically lower (−12%, p=0.004) in the patients, suggesting widespread neuronal damage; (ii) median and mean proximity, suggesting normal distributions; and (iii) the global Cho and Cr concentrations are not statistically different between the patients and their controls.

3.2. Whole brain (WB) ¹H-MRS

The BOLERO auto-shim yielded 22±3 Hz WB water linewidth in under 5 minutes for all 27 participants. All 14 controls and 12 of 13 patients yielded MRS data suitable for post processing, as shown in Fig. 1. Data from patient #12 in Tables 1 and 2, were unusable, likely due to motion. The WB NAA linewidth was 12±2 Hz, and metabolites’ Cramer-Rao lower bounds were ‹0.1% in all subjects, reflecting the very high, ›500, SNRs (see Fig. 1), as described by (Bolan et al., 2003). Patients’ global NAA, Cr and Cho concentrations (Eq. [3]), were: 11.5±1.5, 7.1±1.1 and 1.3±0.2 mM; versus controls’ 13.0±0.8, 7.7±0.7 and 1.4±0.1 mM. Individuals’ metrics are compiled in Table 2, and their distributions are shown in Fig. 3d-f. While patients’ and controls’ Cho and Cr levels were not statistically different, patients’ mean NAA concentration was 12% lower (p=0.004), retaining significance even after multiple comparison Bonferroni correction (Abdi, 2007). No correlation was found between the NAA concentrations and patients’ disease duration.

4. Discussion

Whole brain global metrics are by their nature insensitive to local pathology whose partial volume is smaller than their sensitivity threshold. Therefore, a 12% decline in the global NAA concentration, above the threshold of the methodology (Soher et al., 2014), may represent either of two scenarios: either total localized loss of this neuronal marker, presumably representing complete destruction of the neurons in this volume, ~150 cm³, of the brain; or, alternatively, it may reflect deficits that could be greater in some region(s) and lesser in others, widespread throughout the brain. The first scenario is less probable, since this magnitude localized injury is unlikely not to go undetected by MRI, either locally, or in fGM and fWM. This observation, therefore, supports the hypothesis and notion that the “epileptic brain” may be globally different from its healthy counterpart, at least in its neuronal cell integrity metabolic marker, if not in global morphologic metrics. We note, however, that the term “global” used in the ¹H-MRS context, refers to the average concentrations over the entire brain, and therefore lower WB NAA does not necessarily indicate that all brain tissue is affected. The most likely scenario is that of multifocal injury, given previous imaging data (Doucet et al., 2016; Finnema et al., 2016; Hetherington et al., 2007), which do not support the assertion of homogeneous injury over the entire brain.

In assessing this NAA decline we also consider two additional observations. First, there were no statistically significant volumetric differences, specifically in fBV, fGM or fWM, between the cohorts. This suggests that there was no collateral cellular loss(s), i.e., atrophy, which replicates the findings from a larger cohort of localization-related epilepsy patients of short disease duration (Pardoe et al., 2017). Second, we find no spectroscopic evidence of differences in the global cellular environment of the brain, reflected in unchanged Cho and Cr. This metabolic finding suggests no evidence of global gliosis, which would have led to Cho and Cr elevation (whose concentrations are higher in glial cells than in neurons) (Zhu and Barker, 2011); nor of abnormal global membrane turnover (changes in Cho). Consequently, a consistent interpretation is that the patients’ NAA deficits do not represent neuronal loss, but rather a decline in the neurons’ health. One possible cause could be mitochondrial dysfunction, a notion with strong literature support, since NAA is synthesized in the mitochondria (Benarroch, 2008; Paling et al., 2011); and its oxidative stress has been consistently implicated as a key contributing factor to neuronal injury in localization-related epilepsy (Kann and Kovacs, 2007; Waldbaum and Patel, 2010).

It is noteworthy that 10 of the 13 patients are relatively early in their disease course, ‹5 years from onset and more than half have their seizures under control (cf. Table 1). The absence of tissue atrophy, therefore, is not necessarily surprising (Pardoe et al., 2017). The reduced NAA finding and its lack of correlation with disease duration, suggest that: (a) NAA decline may have preceded the known onset of the disease, not an uncommon feature of chronic neurological disorders, e.g., multiple sclerosis and mild cognitive impairment (Falini et al., 2005; Filippi et al., 2003). (b) That this global metabolic metric may be a more sensitive marker than global morphological ones. However, these patients were not scanned prior to the onset of their disease, nor followed afterwards (at sufficiently large time intervals to be sensitive to WB NAA change) (Davitz et al., 2017). It is, therefore, impossible to determine from our data whether the NAA deficits preceded the disease onset, or occurred afterwards. We are also unable to ascertain whether the NAA signal was affected by the various medication regimens.

Admittedly, the proposed method also has some limitations. First, the lack of localization of the WB sequence, a consequence of the tradeoff between high SNR, acquisition speed and total spatial coverage for the ability to discern regional from global changes. This forces the assumption that changes in NAA, Cr and Cho are widespread enough to be detected. Consequently, second, localized changes in restricted region(s) will be lost to the complete partial volume effect and may end up not being detected, even if their (localized) metabolic changes are dramatic. Third, although the scans were done interictally, the time elapsed from the last known seizure was variable: less than a month for four patients, less than a year for another four, and longer for the rest, as shown in Table 1. Consequently, there is insufficient power to analyze whether the relative decline in NAA is a (transient) result of a proximity to a seizure. Finally, fourth, the patient cohort size and composition (in variable seizure types, frequency, and being mostly of relatively short disease duration) impedes the discovery of relationships between metabolism and these clinical metrics. More homogeneous cohorts and higher sample sizes, therefore, are needed to confirm these findings and investigate their relationship with epilepsy subtypes and clinical outcome.

5. Conclusions

The observation of global NAA decline in the absence of other metabolic or structural abnormalities indicates compromised global neuronal health and supports the view of localization-related epilepsy as a network disorder. Longitudinal follow-up, along with assessment of seizure outcomes in these patients, will determine if WB NAA correlates with clinical measures and can, therefore, play a role in epilepsy prognosis and treatment monitoring.