Succinic semialdehyde dehydrogenase deficiency in mice and in humans: An untargeted metabolomics perspective

Succinic semialdehyde dehydrogenase deficiency (SSADHD) is a rare neurometabolic disorder caused by disruption of the gamma‐aminobutyric acid (GABA) pathway. A more detailed understanding of its pathophysiology, beyond the accumulation of GABA and gamma‐hydroxybutyric acid (GHB), will increase our understanding of the disease and may support novel therapy development. To this end, we compared biochemical body fluid profiles from SSADHD patients with controls using next‐generation metabolic screening (NGMS). Targeted analysis of NGMS data from cerebrospinal fluid (CSF) showed a moderate increase of aspartic acid, glutaric acid, glycolic acid, 4‐guanidinobutanoic acid, and 2‐hydroxyglutaric acid, and prominent elevations of GHB and 4,5‐dihydroxyhexanoic acid (4,5‐DHHA) in SSADHD samples. Remarkably, the intensities of 4,5‐DHHA and GHB showed a significant positive correlation in control CSF, but not in patient CSF. In an established zebrafish epilepsy model, 4,5‐DHHA showed increased mobility that may reflect limited epileptogenesis. Using untargeted metabolomics, we identified 12 features in CSF with high biomarker potential. These had comparable increased fold changes as GHB and 4,5‐DHHA. For 10 of these features, a similar increase was found in plasma, urine and/or mouse brain tissue for SSADHD compared to controls. One of these was identified as the novel biomarker 4,5‐dihydroxyheptanoic acid. The intensities of selected features in plasma and urine of SSADHD patients positively correlated with the clinical severity score of epilepsy and psychiatric symptoms of those patients, and also showed a high mutual correlation. Our findings provide new insights into the (neuro)metabolic disturbances in SSADHD and give leads for further research concerning SSADHD pathophysiology.


| INTRODUCTION
Succinic semialdehyde dehydrogenase deficiency (SSADHD; MIM #271980) is a rare neurometabolic disorder, with $450 patients described worldwide. 1It is caused by disruption of the gamma-aminobutyric acid (GABA) pathway (Figure 1).GABA-transaminase converts GABA into succinic semialdehyde (SSA), which is further metabolized by SSADH into succinic acid.In case of SSADHD, SSA is mainly converted into the alternative product gamma-(or 4-) hydroxybutyric acid (GHB).This metabolite can be found in urine, blood, and CSF of patients and is a hallmark of the disease, which is therefore also known as 4-hydroxybutyric aciduria. 2nother biomarker that is considered pathognomonic for SSADHD is 4,5-dihydroxyhexanoic acid (4,5-DHHA).It is thought to be formed from a reaction between SSA and pyruvate, catalyzed by pyruvate dehydrogenase, but this remains to be proven. 3he clinical symptoms of SSADHD are diverse, including motor and mental delay, hypotonia, speech delay, and epilepsy. 4This nonspecific clinical phenotype warrants biochemical and genetic confirmation when SSADHD is suspected.This is usually done by analysis of GHB in urine and by sequencing for variants of ALDH5A1, the gene encoding SSADH. 5The connection between the clinical symptoms and the metabolic disruptions is still largely unclear.An effective (curative) treatment is not yet available. 6Several new therapies based on the SSADHD disease mechanism are currently under development, mostly in the preclinical stage.Some clinical trials (phase I or II) have been performed, but without beneficial outcomes.For example, the only phase II clinical trial to date, which tested the GABA B receptor antagonist SGS-742, neither showed any improvement of cognition nor of cortical excitability. 7 more detailed understanding of the pathophysiology of the disease, beyond the accumulation of GABA and GHB, may support novel therapy development for SSADHD.Whereas the deficiency directly affects GABA metabolism, it is possible that related metabolic pathways are affected as well.Finding such clues may unravel novel targets for therapeutic intervention.Since SSADHD is predominantly characterized by neurological symptoms, studies of brain tissue and cerebrospinal fluid (CSF) are likely the most promising approaches to discover novel disease biomarkers.Recently, postmortem brain tissue of a single SSADHD patient was studied using metabolomics, which showed disruption of shortchain fatty acid, creatine, and amino acid metabolism, apart from the expected changes in GABA-associated metabolites.3 Given the limited availability of human brain tissue, CSF provides an interesting alternative source for study.CSF is generally considered a good metabolic reflection of brain metabolism.The study of CSF from SSADHD patients so far has focused on a limited number of metabolites such as GHB, GABA and 4-guanidinobutanoic acid (4-GBA).8,9 Untargeted metabolomics in CSF provides a broader view on the metabolic disruptions in the brain and allows the identification of additional biomarkers that have not been previously linked to SSADHD.
We previously described next-generation metabolic screening (NGMS) as an untargeted metabolomic approach in CSF. 10 While this study already included a demonstration the diagnostic use of NGMS for SSADHD patients, we now aim to apply it to extend our understanding of SSADHD pathophysiology.First, we studied which known biomarkers are deviant in CSF of SSADHD patients.Since the description of exact concentrations of 4,5-DHHA is lacking in literature we quantified this metabolite in CSF.Furthermore, we tested the epileptogenic potential of 4,5-DHHA in a zebrafish model.By untargeted metabolomics, we searched for features potentially serving as new SSADHD biomarkers in body fluids.Lastly, the levels of selected features were correlated to clinical severity outcomes, using plasma and urine samples.

| Sample collection
We collected CSF samples from five unrelated SSADHD patients (four males and one female; age range 0.6-22.6years).All samples from these patients were leftover material from diagnostics performed for regular clinical follow-up in the period 2010-2019.Four patients had no medication at the time of lumbar puncture while one was under treatment with vigabatrin.Controls included 10 CSF samples from our previously described control cohort, with ages ranging from 0 to 15 years old, 10 and 11 samples from patients with inherited metabolic disorders unrelated to GABA metabolism, who were 2 to 25 years old.For patients 1, 2, and 4 the diagnosis SSADHD was genetically confirmed.They had homozygous ALDH5A1 mutations: c.698C>T (p.Thr233Met); c.621delC (p.(Ser208-Valfs*3)); and c.1234C>T (p.Arg412*), respectively.The mutations in patients 3 and 5 were unknown.In the 1.6 year-old patient three strongly increased concentrations of GHB were measured in urine, plasma and CSF with gas chromatography-mass spectrometry and in the 14 year-old patient 5 an increased concentration of GHB was measured in CSF using proton nuclear magnetic resonance spectroscopy (NMR; patient 437 μM; normal 0-2 μM 11 ).
Heparin plasma and urine samples were obtained from the Washington State University SSADHD Biorepository, and collected from patients enrolled in the SSADHD Natural History study (NIH/NICHD 1R0-1HD91142, Gibson, Principal Investigator). 12Of note, the subjects in this still ongoing Boston Childrens Hospital (BCH) Natural History study do not include any of the five patients of whom CSF was measured.The first batch of BCH samples consisted of first visit plasma and urine samples from five patients with SSADHD and five controls, which were unaffected family members.Age, sex, and medication of these subjects are listed in Table S1.These were used to check whether the selected features identified in CSF were also present and showing the same deviation in plasma and urine.For the same purpose, we also obtained mouse brain tissue from three SSADHD knockout (KO; Aldh5a1 À/À ) mice and three wildtype (WT) mice (Aldh5a1 +/+ ) from the same colony from Washington State University. 13The mice in both groups were male (one KO-mouse: sex-unknown).The animals were sacrificed at 3 and 6 weeks, respectively; KO-mice do not survive past 3 weeks. 13 second batch of plasma and urine samples from the Natural History Study biorepository consisted of plasma (n = 25) and urine (n = 23) samples of patients with SSADHD.These included second visit samples from four patients included in the first batch, and 21 (plasma)/19 (urine) first visit samples from other patients.As extensive clinical data was available for these samples (Data S1), they were used for calculating the correlation of feature intensities to clinical severity outcomes.

| Sample preparation and UHPLC-QTOF-MS analysis
CSF and plasma samples were deproteinated and subsequently analyzed with ultra-high-performance liquid The GABA pathway.Simplified schematic representation of gamma-aminobutyric acid (GABA) metabolism.In SSADHD, accumulating succinic semialdehyde (SSA) will be converted in a nonenzymatic, reversible reaction to form gamma-hydroxybutyric acid.Also, SSA is thought to react with pyruvate, catalyzed by pyruvate dehydrogenase (PDH), to form 4,5-dihydroxyhexanoic acid.
chromatography-quadrupole time of flight mass spectrometry (UHPLC-QTOF-MS) as described previously. 10,14For urine samples, the creatinine concentration was first measured using the Cobas 8000 modular analyzer series (Roche Diagnostics, Basel, Switzerland).Subsequently, the samples were diluted to the lowest creatinine concentration observed in all samples (0.85 mM) before being prepared using the same deproteination protocol as for CSF and plasma.Mouse brain tissues were pre-treated by submerging the tissue in methanol (400 μL per 1030 mg brain tissue to match a previous study 15 ) and centrifuging the sample for 30 min at 18 600g at room temperature.The tissue was then pulverized and the sample was centrifuged again for 20 min at the same conditions.To increase the supernatant volume, another 400 μL/1030 mg methanol was added and the sample was centrifuged for 10 more minutes.After this, 100 μL of the supernatant was collected and processed using the same protocol as for CSF and plasma.For all measurements, a procedure blank was included, which consisted of 100 μL Milli-Q processed using the same protocol.

| Targeted analysis of SSADH deficiency biomarkers from NGMS data
We performed a literature search to identify biomarkers that have previously been associated with SSADHD in humans.From these, we selected those metabolites that were part of our in-house inborn errors of metabolism (IEM) panel and for which the exact retention time (RT) was known 14 to allow identification based on two orthogonal properties (classified as level 1 in the guidelines of the Metabolomics Standards Initiative [MSI]). 16The RTs were combined with the molecular formulas to search the raw UHPLC-QTOF-MS data for these metabolites using Skyline version 20.2.0.343.The corresponding intensities were extracted and further analyzed using R version 4.2.In case the procedure blank showed a signal for a metabolite, this intensity was subtracted from the intensities in the samples before statistical analysis.The fold change (FC) of the feature intensities were calculated by dividing the median intensity in the patients by the median intensity in the controls.The mean intensities of the two groups were compared using a Kruskal-Wallis test with subsequent Benjamini-Hochberg correction to obtain adjusted p-values (q), which were considered to indicate a significant difference when <0.05.

| Characterization of 4,5-DHHA in CSF
The concentrations of 4,5-DHHA in CSF of patients or controls were so far unknown. 9Using proton NMR spectroscopy, we quantified 4,5-DHHA in the CSF of patient 5. 17 For quantification in the UHPLC-QTOF-MS analysis 4,5-DHHA was synthesized (Figure S1).Using the relative intensities from UHPLC-QTOF-MS, we estimated the concentrations in the other samples: concentration sample X = concentration sample patient 5/UHPLC-QTOF-MS intensity sample patient 5 Â UHP LC-QTOF-MS intensity sample X.Additionally, we estimated the 4,5-DHHA concentration of the quality control (QC) control pool sample using the standard addition method, for which the CSF QC sample was spiked with 28, 56, and 281 nM 4,5-DHHA.Furthermore, using data from our control cohort, we studied the age, sex, and fraction dependency of 4,5-DHHA as described previously. 10We also studied the correlation between 4,5-DHHA and GHB in this control cohort, as well as in the patient samples from the current study.

| Epileptogenic potential of 4,5-DHHA in zebrafish larvae
WT Tüpfel long fin zebrafish were bred and raised as described previously. 15At 4 days post fertilization (dpf), larvae were transferred to separate wells of a 96-wells plate.In each well, 4,5-DHHA was added to the medium to reach final concentrations of 100 nM, 1 μM, 10 μM, 100 μM, or 1 mM (16 larvae/condition).Subsequently, we analyzed the movement of the larvae by video tracking, using a DanioVision Observation Chamber (Noldus Information Technology, Wageningen, The Netherlands).While kept at 28.5 C, the larvae underwent a seizureinducing protocol consisting of 10 min of adaptation to the dark, a 5 min baseline measurement in the dark (phase 1), and 5 min of light stimuli (5 cycles of 10 s 100% light and 50 s light off; phase 2). 18Behavioral responses were analyzed with EthoVision XT15 software (Noldus Information Technology).Stage I seizure-like behavior, that is, increased swimming activity, 19 was assessed by quantitating the total distance of all movement (with thresholds of 1.00 mm/s start velocity and 0.00 mm/s stop velocity) and the total distance of high-speed movements (>20 mm/s) in the different phases.For the assessment of Stage II and III epileptic activity, consisting of rapid "whirlpool-like" circling swimming and clonus-like convulsions leading to a loss of posture, 19 we monitored the behavior of the larvae using video recordings.

| Untargeted metabolomics in CSF
For untargeted analysis, raw data acquired from the UHPLC-QTOF-MS runs were aligned using the R package "xcms" (XCMS version 3.4.4running under R version 4.2).The extracted features (i.e., the combination of an accurate mass-to-charge-ratio (m/z), RT, and for each sample an intensity) were preprocessed by selecting those within an m/z range of 70-700, RT between 0.4 and 16 min, intensity in the procedure blank lower than 25% of the maximum intensity of the CSF samples, and feature intensity of ≥10 000 in at least one sample.Features only present in the QC samples were removed.Before applying any statistics, we averaged duplicate measurements.
We compared the feature intensities of the five CSF samples from SSADHD patients to those of the 21 controls using a Kruskal-Wallis test with Benjamini-Hochberg correction using R version 4.2.Initially, we selected features that were significantly different between the two groups (q < 0.05).Given the high number of significant features, we applied additional prioritization criteria to extract the features with the highest biomarker potential.First of all, features were grouped when they correlated with each other (r > 0.9) and had a similar RT (ΔRT <0.05 within mode, ΔRT <0.6 between positive/ negative mode); grouped features were considered to be from the same metabolite.Within each group, the feature with the highest intensity was selected for further analyses.The groups of features originating from GHB or 4,5-DHHA were subsequently removed.Remaining features were selected if they had no overlap in intensity between patient and control groups, the intensity was ≥50 000 in at least one sample, and the XCMS data could be reproduced in the raw data using MassHunter Qualitative Analysis B.07.00.We also checked if the feature was also deviant (nonoverlapping increase or decrease) in the data of a previously measured CSF sample from a SSADHD patient. 10Lastly, for features that were only present in patient samples and thus absent in controls, we checked whether they were also absent in the data of the previously reported control cohort. 10or features included in the final selection, we also checked their presence in the human plasma and urine samples (five patients vs. five controls) and the mouse brain tissue samples (three KO vs. three wildtype mice).To this end, we inspected the raw data using MassHunter Qualitative Analysis B.07.00, taking into account both m/z and RT of the features.If present, we compared the intensities of patients against controls to determine whether a similar change (nonoverlapping increase or decrease) as in CSF was present.
For all selected features, a putative annotation consisting of a molecular formula and adduct was determined from the raw data based on the m/z value and the distance and relative intensity of the isotopes using MassHunter Qualitative Analysis B.07.00.We then checked the presence of peaks at the same RT that had an m/z fitting with other adducts ([M + H] + , [M + Na] + , [M-H] À , [M + Cl] À ) of the same molecular formula to corroborate this putative annotation.Furthermore, we compared the found m/z and RT values to those present in our in-house IEM panel and searched the m/z in the Human Metabolome Database 20 and METLIN database 21 to identify possible molecular structures of the metabolite.We also considered the products of common chemical reactions of known accumulating metabolites SSA, GABA, and GHB.When feasible, the hypothesized structures were synthesized and analyzed using UHPLC-QTOF-MS to test whether they had a matching RT: RTs within 0.1 min of each other were considered a match.If so, the hypothesized structure was considered a definite identification.This procedure was followed for 4,5-dihydroxyheptanoic acid (Figure S1).

| Correlation between selected features and clinical severity
For selected features that were also present and deviant in plasma and urine of patients with SSADHD, we calculated the correlation of these features with several outcomes of clinical severity.The intensities of the features were assessed in 25 plasma samples and 23 urine samples from patients with SSADHD.For measures of clinical severity, we used the Clinical Severity Score based on five separate components: communication, cognition, motor skills, psychiatric presentation, and epilepsy: (CSS; ranges from 5 to 25, with five being the most severe). 12Scores for the individual components ranged from 1 to 5 with decreasing severity of the symptoms.For the correlation of cognition with the biomarkers we have used the IQ of the patient instead of the 1-5 cognition score in the overall CSS to get a more accurate correlation.All correlations were calculated using R version 4.2 by Spearman's rank correlation coefficient followed by Benjamini-Hochberg correction and considered significant if q < 0.05.

| Targeted analysis of SSADH deficiency biomarkers in NGMS data
We found 46 metabolites in our IEM panel which have been reported as SSADHD biomarkers in literature; 29 of these could be identified in our data.For a full list including adducts, m/z and RT values see Data S1. Figure 2 shows the seven metabolites for which the intensities were significantly different in patient CSF samples compared to controls.Two metabolites, GHB and 4,5-DHHA, showed a very clear increase with FCs of 206 and 161 (q < 0.01), respectively.The associated features also had high intensities for both metabolites, with a saturated signal for the [M-H] À feature (Figure 2 shows the [M + Na] + adduct, the highest unsaturated signal).Whereas a genetic confirmation of the diagnosis was not available for patients 3 and 5, their clearly elevated levels for GHB and 4,5-DHHA confirmed the diagnosis of SSADHD.The intensities of the other five metabolites, aspartic acid, glutaric acid, glycolic acid, 4-guanidinobutanoic acid, and 2-hydroxyglutaric acid, were also increased in the CSF from SSADHD patients.However, these differences were more subtle, with FCs ranging from 1.6 to 5.4 (q < 0.05 for all).

| Characterization of 4,5-DHHA in CSF
First, we confirmed the identity of 4,5-DHHA by comparing the m/z, the RT, the adducts and the infrared spectrum of a QC sample spiked with synthesized 4,5-DHHA with a SSADHD CSF sample (m/z 171.0628sodium adduct; RT 2.40 min; data not shown).Given the remarkable increase in 4,5-DHHA in the CSF of SSADHD patients, with both the FC and the peak intensities comparable to that of the hallmark biomarker GHB, we quantified 4,5-DHHA in patient and control CSF.Using proton NMR spectroscopy, we estimated the concentration of 4,5-DHHA in the CSF of patient 5 at 64 μM, while the GHB concentration was 2.4 mM.Combining this with the UHPLC-QTOF-MS data, we estimated the 4,5-DHHA concentrations in CSF of the remaining patients at 63-80 μM, of controls at 0.13-0.83μM, and of the QC pool sample at 0.34 μM.The 4,5-DHHA concentration of the QC pool sample estimated via standard addition was 0.38 μM (Figure S1), in line with the first estimation.
Figure 3A-C shows the evaluation of the effects of age, sex, and CSF fraction on the level of 4,5-DHHA in CSF samples from our control cohort.Based on the comparison of age groups of 0-2, 2-5, 5-10, and 10-15 years, age had a significant effect on the 4,5-DHHA level (Kruskal-Wallis test; p = 1.9e-7; Figure 3A).Neither sex nor the CSF fraction influenced the 4,5-DHHA level (Figure 3B,C).Furthermore, there was a significant positive correlation between CSF intensities of 4,5-DHHA and GHB within this cohort (R = 0.75; p = 1.2e-16; Figure 3D).However, this was not the case when only studying the samples of the five SSADHD patients (R = À0.32,p = 0.6; Figure 3E).

| Epileptogenic potential of 4,5-DHHA in zebrafish larvae
We tested the epileptogenic potential of 4,5-DHHA by tracking the movement of zebrafish larvae (4 dpf) after exposure to various concentrations of 4,5-DHHA (ranging from 100 nM-1 mM) to the medium.Only at 1 mM, an effect was observed, with an increase of both total and high-speed movements compared to control, fitting with stage I seizure-like behavior (Figure 4).This increase was present in both Phase 1 (5 min baseline, dark) and 2 (5Â 10 s light/50 s dark) and thus seemed independent of the seizure-inducing light impulses.This increased mobility may reflect limited epileptogenesis.Using video recordings, we observed whirlpool-like swimming (stage II) only in a single larva of the 1 mM group, at the start of the first light impulse.Stage III behavior (clonus-like convulsions) was not observed at all.

| Untargeted metabolomics in CSF
Initially, 35 401 features were extracted by XCMS, of which 19 699 features remained after preprocessing.Comparing patient and control CSF samples, 1174 features (6%) showed a significant difference, as listed in the Data S1.Using the additional prioritization criteria described in the Methods, we selected 12 features as candidate biomarkers (Table 1).These features all had significantly increased intensities in SSADHD CSF compared to control CSF (q < 0.05), with FCs ranging from 6.9 to 669.Two features, p7167 and n7964, were also present at increased levels in SSADHD plasma, 8/12 were present in urine of these patients, and 10/12 in brain tissue of SSADH KO mice (compared to control mouse brain).The features p7167 and n7964 were the only ones present in increased concentration in all three SSADHD body fluids and as well in mouse brain tissue samples, thus qualifying as potential additional SSADHD biomarkers.The FCs of both features were >100 and both discriminated well between controls and SSADHD in CSF samples (for p-values: Table 1).In urine, 4,5-dihydroxyheptanoic acid and the features n3205, p6016, and p9323 potentially qualify as additional biomarkers of the disease.The feature with the highest signal intensity was n1245 for CSF, n7964 for plasma, 4,5-dihydroxyheptanoic acid (n1202) for urine and n2784 for mouse brain.All of these reached intensity levels of 10 6 .In mouse brain the features n2784 and n7964 stood out; both had more than 100-fold higher signal intensities than the other selected features.
For all features, we determined a putative formula and adduct matching with the exact m/z of the features (Table 1).For eight features, we identified additional adducts to support this preliminary annotation.For seven, at least one hypothesized structure was proposed and four features were actually compared to synthesized standards using UHPLC-QTOF-MS (Figure 5).Feature n1202 was compared to a synthesized 4,5-dihydroxyheptanoic acid standard.The synthesized model compound showed a matching RT.Also, the MS/MS spectrum of this feature in the synthesized 4,5-dihydroxyheptanoic acid was identical F I G U R E 4 4,5-DHHA epileptogenic potential.Assessment of Stage I seizure-like behavior (increased swimming activity) in zebrafish by quantification the total distance of all movements (with thresholds of 1.00 mm/s start velocity and 0.00 mm/s stop velocity) and the total distance of high-speed movements (>20 mm/s) in the different phases.Phase 1: baseline measurement, 5 min, dark.Phase 2: five cycles of 10 s light (100%) and 50 s dark.ns: q > 0.05, *: q < = 0.05, **: q < = 0.01, ***: q < = 0.001, ****: q < = 0.0001.to its MS/MS spectra in SSADHD-CSF and in SSADH KO-mice brain tissue extract (Figure S3).Interestingly, a stereoisomeric form of 4,5-dihydroxyheptanoic acid was seen at RT 3.84 min in SSADH KO-mouse brain extract (Figure S4).The two stereoisomers were present in equal amounts in the brain extract and in the synthesized model compound.The 3.84 min form occurred only in minute amounts in the patient's body fluids.The FC of 4,5-dihydroxyheptanoic acid in SSADH CSF amounted to 339 (4,5-DHHA and GHB: 161 and 206, respectively, as shown in Figure 2).Taken together these data provide definite level 1 identification (MSI) for feature n1202 as 4,5-dihydroxyheptanoic acid.Figure 5 gives hypothesized structures for other unidentified features from Table 1.Feature n2784 in SSADHD CSF was compared to synthesized GABA-1O-Glu, GABA-5O-Glu, and γ-Glu-GABA.However, these synthetic metabolites all had significantly lower RTs than n2784 (>0.1 min difference).The positive mode features p6016 and p8392 were compared with the synthesized model compounds Lys-έ-GABA and N2-succinoylarginine, respectively, but their RTs did not match.

| Correlation between selected features and clinical severity
The intensities of known biomarkers GHB and 4,5-DHHA and features n1202, p6016, n3205, p7167, p9323, and n7964 in plasma and/or urine samples of patients with SSADHD were correlated to various clinical severity outcome measures (Figure 6).The clinical severity score, CSS, was associated with the intensity of 4,5-DHHA and of GHB in plasma and the intensity of n3205 in urine.The clinical scores for psychiatric impairment and epilepsy correlated positively with almost all selected feature intensities (Figure 6).Since higher scores correspond to a milder phenotype, this indicated that higher feature intensities were associated with less severe symptoms.Of note, all used features intensities had significant positive mutual correlations (q < 0.05) with Spearman's correlation coefficient ranging from 0.57 to 0.97 (Figure 6).

| DISCUSSION
The main findings of this study were (1) a moderate increase of aspartic acid, glutaric acid, glycolic acid, 4-guanidinobutanoic acid, and 2-hydroxyglutaric acid, and prominent elevations of GHB and 4,5-DHHA in CSF of patients with SSADHD; (2) in an established zebrafish epilepsy model, 4,5-DHHA showed increased mobility that may reflect limited epileptogenesis; (3) using untargeted metabolomics, we identified 12 features in CSF with high biomarker potential, all with significantly increased intensity in the CSF of SSADHD patients and for 10 of these features, a similar increase was found in plasma, urine and/or mouse brain tissue; (4) one of the features was identified as the novel biomarker 4,5-dihydroxyheptanoic acid; (5) the intensities of selected features in plasma and urine of SSADHD patients correlated positively with the clinical severity score of epilepsy and/or psychiatric symptoms but also showed a high mutual positive correlation.
While GHB as hallmark biomarker in CSF is wellknown, 4,5-DHHA in CSF has not been described as such before, except for a brief notion of a 100-fold increase in a single patient. 22We detected 4,5-DHHA also in control CSF with an estimated concentration in the high nanomolar range.These levels were influenced by age, but not by sex or CSF fraction.In CSF of SSADHD patients, the 4,5-DHHA concentration was estimated at 63-80 μM, which is lower than GHB concentrations which range from the high micromolar to the low millimolar range. 9nterestingly, we reproduced the correlation between these two metabolites in CSF (previously identified in F I G U R E 5 Novel SSADHD biomarker features.Structural identification of unknown features.Hypothesized structures for n1202, n2784, p6016, and p8392 measured using LC-Qtof MS to compare the retention time of the feature in CSF (RT) with the RT in the synthesized products.RTs within 0.1 min of each other were considered a match.This was only the case for feature n1202 and the synthesized standard of 4,5-dihydroxyheptanoic acid.GABA, gamma-aminobutyric acid; Glu, glutamic acid; Lys, lysine.
human brain tissue 3 ).In our five SSADHD patient samples, this relation between the GHB and 4,5-DHHA was not observed, indicating that either the numbers are too small or in case of SSADHD, the biological relation between these metabolites may be altered.
Using a previously established zebrafish epilepsy model, 18,19 we found that exposure to 1 mM 4,5-DHHA in the medium led to a significant increase in movement of zebrafish larvae.This seizure-like hyperactivity (stage I behavior) may be seen as indicative of limited epileptogenesis of 4,5-DHHA but this would require further electrophysiological evidence.Also, we remain cautious to conclude this, as the hyperactivity only occurs at doses higher than the concentration in CSF and as well because stage II and III seizure behavior was (mostly) absent.It therefore remains to be explored if and how the observed high levels of 4,5-DHHA in CSF contribute to the pathophysiology of SSADHD.
Apart from elevated GHB and 4,5-DHHA, targeted analysis revealed five more metabolites with moderately increased levels in the CSF of SSADHD patients.While the increases of 4-guanidinobutyric acid and 2-hydroxyglutaric acid in CSF have been reported before, 9 the finding is new for aspartic acid, glutaric acid, and glycolic acid.Aspartic acid was also increased in human SSADHD brain tissue and considered especially interesting because of its role as an excitatory neurotransmitter. 3Glutaric acid and glycolic acid were so far only described as urine biomarkers for SSADHD. 23Glutaric acid is known to be neurotoxic, albeit in the context of glutaric aciduria type I with a more pronounced accumulation of this metabolite. 24Conversely, glycolic acid has been described as neuroprotective, preventing ischemiainduced neuronal death in in vitro and in vivo models of stroke. 25Despite the less distinct increases, all three metabolites could be of interest for our understanding of SSADHD pathophysiology.
The high number of significantly different features in untargeted analysis fits with the idea that in SSADHD, the initial disturbance of the GABA breakdown pathway may affect other pathways, including the Krebs cycle and β-oxidation. 3The 12 selected features with the highest biomarker potential were all clearly increased in patients with low to very low signals in control CSF, similar to GHB and 4,5-DHHA, indicating that there are many accumulating metabolites that, if toxic, could contribute to the pathophysiology of SSADHD.We hypothesize that the newly identified biomarker, 4,5-dihydroxyheptanoic acid, is formed from SSA and 2-ketobutyrate in a reaction similar to the formation of 4,5-DHHA (Figure 7).Features p2362, n1245, and n2101, with formulas C 5 H 10 O 4 , C 6 H 12 O 5 and C 8 H 14 O 6 , could theoretically be formed from SSA with the keto acids glyoxylate, hydroxy pyruvate and alpha-ketoglutarate, respectively, to form various hydroxyl acids (Figure 7 and Table 1).Furthermore, we postulate that features n2784, p6016, and p8392 may arise from conjugation of accumulating metabolites in SSADHD with amino acids (Figure 7).These hypotheses remain to be confirmed.For the remaining five features from Table 1, no likely structures could be proposed.
Finding the exact identification for the features from untargeted analysis was the major challenge in our study, as is the case for many metabolomics studies. 26Further work on feature identification using MS/MS fragmentation and infrared ion spectroscopy is required for a better understanding of SSADHD pathophysiology. 27 limitation in our study was the availability of CSF samples from SSADHD patients.The available clinical data were scarce for the five CSF samples that we collected.We used urine and plasma instead of CSF for correlation between the intensities of selected features and the clinical symptomatology.Not all features found in CSF could be studied in this way, because they were not present in urine or plasma.Many of the clinical signs and symptoms are neurological and it would therefore have been especially interesting to correlate them with CSF parameters.For epilepsy and for psychiatric impairment though, several features in plasma and urine positively correlated with the clinical severity scores.Because of the high correlation between the features themselves, it is impossible to indicate which individual features are the most relevant for SSADHD pathophysiology.Interestingly, since high scores correspond to less severe symptoms, the positive correlations indicate that higher biomarker levels are accompanied by milder symptoms.Although this may seem counterintuitive, a recent study also showed that the presence of seizures was associated with lower levels of GABA and GHB.28 Partly this may be explained because these molecules become reduced from supraphysiological levels over time.
In conclusion, we report new findings about the SSADHD metabolome.We showed that 4,5-DHHA is strongly elevated in CSF of patients with SSADHD, while also present at low levels in CSF of controls.Furthermore, we identified 12 features that are clearly increased in CSF of SSADHD patients, one of which was identified as the novel biomarker 4,5-dihydroxyheptanoic acid.Intensities of several of these features as well as levels of GHB and 4,5-DHHA in plasma and urine, showed significant correlations with the severity score of epilepsy and psychiatric symptoms in patients with SSADHD.Counterintuitively SSADHD patients with higher biomarker levels had milder symptoms.Our findings give new insights in the (neuro) metabolic disturbances in SSADHD and provide leads for further research into the pathophysiology of this disease.

T A B L E 1
Selected features with high biomarker potential, identified by untargeted metabolomics analysis comparing cerebrospinal fluid (CSF) next-generation metabolic screeningprofiles of Succinic semialdehyde dehydrogenase deficiency (SSADHD) patients to controls.Feature a

"
All 12 features in this table were found significantly increased in SSADHD CSF samples.a Feature-ID: features are marked "p" or "n": positive-or negative-mode of the MS.b Feature found increased in SSADHD patient plasma (P), patient urine (U), or KO-mouse brain tissue (MB)."-": feature not present/no increase found.c Other adducts found, in order of intensity.Checked for [M Tested" indicates that the hypothesized structure was synthesized and compared to the feature with regard to RT as measured by ultra-high-performance liquid chromatography-quadrupole time of flight mass spectrometry: RTs within 0.1 min of each other were considered a match.e A clear increase was found in three out of five patient samples; the other two samples showed feature intensities comparable to controls.m/z = mass-to-charge ratio, RT = retention time, P = SSADHD plasma, U = SSADHD urine, MB = KO-mouse brain tissue homogenate.

F
I G U R E 6 SSADHD biomarkers and clinical severity.Correlation between selected features from targeted and untargeted analysis and clinical severity outcome scores (CSS).Colors represent the Spearman's rank correlation coefficient scores ranging from À1 to 1. Asterisks indicate significant correlations (q < 0.05).