Early Systemic Glycolytic Shift After Aneurysmal Subarachnoid Hemorrhage is Associated with Functional Outcomes

Abstract

Background:

Aneurysmal subarachnoid hemorrhage (aSAH) leads to a robust systemic inflammatory response. We hypothesized that an early systemic glycolytic shift occurs after aSAH, resulting in a unique metabolic signature and affecting systemic inflammation.

Methods:

Control patients and patients with aSAH were analyzed. Samples from patients with aSAH were collected within 24 h of aneurysmal rupture. Mass spectrometry–based metabolomics was performed to assess relative abundance of 16 metabolites involved in the tricarboxylic acid cycle, glycolysis, and pentose phosphate pathway. Principal component analysis was used to segregate control patients from patients with aSAH. Dendrograms were developed to depict correlations between metabolites and cytokines. Analytic models predicting functional outcomes were developed, and receiver operating curves were compared.

Results:

A total of 122 patients with aSAH and 38 control patients were included. Patients with aSAH had higher levels of glycolytic metabolites (3-phosphoglycerate/2-phosphoglycerate, lactate) but lower levels of oxidative metabolites (succinate, malate, fumarate, and oxalate). Patients with higher clinical severity (Hunt-Hess Scale score ≥ 4) had higher levels of glyceraldehyde 3-phosphate and citrate but lower levels of α-ketoglutarate and glutamine. Principal component analysis readily segregated control patients from patients with aSAH. Correlation analysis revealed distinct clusters in control patients that were not observed in patients with aSAH. Higher levels of fumarate were associated with good functional outcomes at discharge (odds ratio [OR] 1.76, 95% confidence interval [CI] 1.15–2.82) in multivariable models, whereas higher levels of citrate were associated with poor functional outcomes at discharge (OR 0.36, 95% CI 0.16–0.73) and at 3 months (OR 0.35, 95% CI 0.14–0.81). No associations were found with delayed cerebral ischemia. Levels of α-ketoglutarate and glutamine correlated with lower levels of interleukin-8, whereas fumarate was associated with lower levels of tumor necrosis factor alpha.

Conclusions:

Aneurysmal subarachnoid hemorrhage results in a unique pattern of plasma metabolites, indicating a shift toward glycolysis. Higher levels of fumarate and lower levels of citrate were associated with better functional outcomes. These metabolites may represent targets to improve metabolism after aSAH.

Introduction

Aneurysmal subarachnoid hemorrhage (aSAH) affects approximately 50,000 people per year in the United States and causes significant morbidity and mortality [1]. Those who survive the initial bleed are at risk for delayed secondary complications. The effects of aSAH are not limited to the brain, with severe systemic medical complications occurring in 40% of patients and commonly causing cardiomyopathy, pulmonary edema, renal injury, hyperglycemia, and coagulopathy [2].

Central and systemic metabolic changes occur early after aSAH, with impaired systemic substrate availability affecting brain metabolism [3, 4]. Additionally, a robust systemic inflammatory reaction is initiated [5], which has been implicated in the occurrence of delayed cerebral ischemia (DCI) and worse functional outcomes [6, 7]. Although not well characterized, this increased inflammatory response is likely due to an early activation of the innate immune system after aSAH accompanied by the mobilization of proinflammatory monocytes and subsequent activation of astrocytes and microglia. Concomitantly, an increase in proinflammatory cytokines is seen systemically and is found to peak at 24–48 h after injury [6, 8].

Metabolism and inflammation are intricately linked [9]. When activated by tissue injury or infection, immune cells undergo metabolic reprogramming characterized by glycolytic shift [10]. In activated macrophages, the tricarboxylic acid (TCA) cycle is broken at two places (after citrate and after succinate), resulting in citrate accumulation for fatty acid synthesis and a glycolytic shift [11]. No studies have examined the systemic metabolic response to aSAH. We hypothesize that a systemic glycolytic shift occurs early after aSAH, promoting the development of a systemic proinflammatory state and contributing to poor outcomes after aSAH. We used a mass spectrometry–based metabolomics approach to study metabolic shifts after aSAH and bioinformatics methods to define a specific metabolic signature occurring after aSAH. We further determine the relationship between metabolic changes and inflammation to predict functional outcomes.

Methods

Study Population

Samples were obtained from a biobank consisting of patients with acute aSAH admitted to the neuroscience intensive care unit who were prospectively enrolled into an observational study between July 2013 and March 2015. The diagnosis of aSAH was established based on admission computed tomography (CT) and CT angiography. Patients with spontaneous aSAH were enrolled within 24 h of ictus. Patients with SAH due to trauma or arterial venous malformation and patients with isolated cortical SAH were excluded. Patients with comorbidities affecting baseline inflammation including autoimmune disease, chronic infection, malignancy, and pregnancy were excluded. Control patients without neurological issues were obtained from an outpatient clinic for routine follow-up.

Standard Protocol Approvals, Registrations, and Patient Consent

The study was approved by the institutional review board at the University of Texas McGovern School of Medicine (HSC-MS-12–0637). Written informed consent was obtained from each patient or surrogate.

Demographic, Imaging, and Outcome Data

Demographic and clinical data were collected, including age, sex, medical comorbidities, and Hunt-Hess Scale (HHS) score. Admission CT images were adjudicated for intraventricular hemorrhage, hydrocephalus, and cerebral infarctions. All CT scans were independently evaluated by a study neurologist for the amount and location of blood characterized by modified Fisher score [12]. DCI was defined as described (Supplementary Methods) [13] and diagnosed after exclusion of all other possible causes. DCI was adjudicated through consensus of at least two attending neurointensivists. Modified Rankin Scale (mRS) at discharge was assessed by the attending neurointensivist caring for the patient [14]. Trained research personnel obtained 3-month mRS scores through follow-up phone calls by using standardized questionnaires. Favorable outcome was defined as mRS score 0–3. Outcome data were available for all patients.

Sample Collection and Processing

All analyses were conducted using samples collected within 24 h of aSAH ictus, with the goal of assessing early systemic changes occurring after injury. Samples were stored at − 80 °C until use.

Cytokine Analysis

Concentrations of 41 serum cytokines were measured following the manufacturer’s protocol by using a MAG-PIX magnetic bead–based enzyme-linked immuno-sorbent assay 41-plex assay (EMD Millipore, Billerica, MA) (Table S1).

Metabolomics

Targeted metabolomics were performed on plasma samples using liquid chromatography/mass spectrometry, as described [15]. A total of 16 metabolites were assessed and represented the following: glycolysis, TCA cycle, and the pentose phosphate pathway (Supplementary Methods).

Bioinformatics Analyses

Informatics methods were used to provide systemic insight into metabolite-metabolite and metabolite-cytokine interactions. The methods included principal component analysis (PCA), hierarchical clustering [16], and network visualization techniques [17]. The hierarchical clustering algorithm was used to group patients who have similar metabolite and cytokines profiles. Weighted correlation network analysis is a network visualization tool used to identify pairwise correlations among data features, in which nodes are represented by metabolites/cytokines and interactions are represented by Pearson’s correlation coefficients. Box-Cox transformation was applied to normalize cytokine data. Metabolite data already followed a normal distribution. To reduce false positive correlations, only highly significant (raw P < 0.01) positive correlations greater than + 0.4 and negative correlations less than − 0.1 were included. Analyses were performed by using open-source software packages (R, version 4.0.3). Cytoscape (version 3.2.1, Institute for Systems Biology, Seattle, WA) was used for network visualization.

Predictive Models

A baseline logistic regression model (including age, sex, and HHS) was established to predict functional outcomes at discharge and 3 months. Functional outcomes were quantified by using mRS (dichotomized as “good” [mRS ≤ 3] and “bad” [mRS ≥ 4]). The data set was randomly stratified but matched for confounders into a “training” set (65% of the data) and “test” set (35% of the data). Models were developed by using the training set, and the performance of the models was validated and reported on the test set to reduce the effects of model overfitting. Area under the receiver operating curves (AUCs) were statistically compared by using the DeLong test [18]. Models were developed by using “scikitlearn” in Python (version 3.6).

Statistical Analysis

Continuous variables were compared by using unpaired Student’s t-tests, and categorical variables were compared by using χ² or Fisher’s exact tests. Student’s t-tests were used to compare differences in metabolites comparing control patients and patients with aSAH and within patients with aSAH according to disease severity. Multiple comparisons were controlled for by using the Benjamini–Hochberg procedure with a false discovery rate of 0.05 [19]. Multivariable logistic regression analyses were performed to identify independent predictors of poor functional outcome and DCI. Statistical analyses were performed using R (version 4.0.3).

Data Availability

Anonymized data not published within this article will be made available on request from qualified investigators.

Results

Demographics

A total of 122 patients with aSAH and 38 control patients were analyzed (Table 1). No significant differences were observed between aSAH and control groups in age, sex, hypertension, hyperlipidemia, coronary artery disease, prior stroke, chronic kidney disease, type 2 diabetes, and smoking status. Significant differences were observed in race and ethnicity (P = 0.008). More Hispanic patients were represented in the aSAH group than in the control group, whereas the control group had more non-Hispanic, White patients.

Table 1.

Demographics

Demographic	Control (n = 38)	aSAH (n = 122)	P value
Age, median (IQR)	54.3 (47.3–64.0)	53.2 (44.0–61.8)	0.636
Female sex, n (%)	22 (58)	95 (78)
Race, n (%)
White	24 (63.2)	45 (36.9)	0.008 *
Hispanic	5 (13.2)	46 (37.8)
Black	7 (18.4)	26 (21.3)
Asian	2 (5.3)	5 (4.1)
HTN, n (%)	18 (47.4)	78 (63.9)	0.103
HLD, n (%)	10 (26.3)	17 (13.9)	0.126
CAD, n (%)	3 (7.9)	6 (4.9)	0.770
Prior stroke, n (%)	0 (0.0)	5 (4.1)	0.134
CKD, n (%)	0 (0.0)	2 (1.6)	0.435
DM2, n (%)	2 (5.3)	19 (15.6)	0.171
Smoking, n (%)	9 (23.7)	50 (41.0)	0.08
HHS, median (IQR)	–	3 (2–4)
mFS, median (IQR)	–	3 (3–3)

^*Statistically significant differences are bolded

aSAH aneurysmal subarachnoid hemorrhage, CAD coronary artery disease, CKD chronic kidney disease, DM2 type 2 diabetes mellitus, HLD hyperlipidemia, HHS Hunt Hess Scale, HTN hypertension, IQR interquartile range, mFS modified Fisher Scale

Analysis of Metabolites

Metabolites with significant differences between control patients and patients with aSAH are shown in Fig. 1a (P value < threshold). Levels of 3-phosphoglycerate and 2-phosphoglycerate (3PG/2PG) and lactate were significantly higher in patients with aSAH. Conversely, levels of glucose/fructose, ribose/ribulose/xylulose-5P, malate, fumarate, and succinate were lower in patients with aSAH. Differences in each metabolite tested are shown in Fig. S1. Metabolites were compared between patients with low (HHS ≤ 3) and high (HHS ≥ 4) clinical severity. Clinically significant differences are shown in Fig. 1a. Levels of glyceraldehyde 3-phosphate and citrate were significantly higher in patients with higher clinical severity. Conversely, levels of α-ketoglutarate and glutamine were lower in patients with higher clinical severity. Differences in each metabolite tested according to clinical severity are shown in Fig. S2. PCA using the metabolites identified in Fig. 1a revealed significant differentiation between control patients and patients with aSAH (Fig. 1b). Poor differentiation resulting when performing PCA with all metabolites tested (Fig. S3). Raw and corrected P values for each comparison are shown in Table S2.

Fig. 1.

Metabolomics comparing control patients and patients with aSAH. a, Metabolites showing significant differences comparing control patients and patients with aSAH. Metabolites are also shown with significant differences comparing low clinical severity (HHS ≤ 3) and high clinical severity (HHS ≥ 4). b, PCA analysis comparing control patients and patients with aSAH based on metabolites identified in a to differ between aSAH and control. Dendrograms representing correlations between metabolites in control patients (c) and patients with aSAH (d). Each pixel in the dendrogram is a correlation coefficient value between a pair of metabolites corresponding to that row and column. Green pixels indicate a positive correlation, red pixels indicate a negative correlation, and black pixels indicate a near-zero correlation. The clustering algorithm groups metabolites that are most correlated with each other. *P < 0.05, **P < 0.001. 2PG, 2-phosphoglycerate, 3PG, 3-phosphoglycerate, aSAH, aneurysmal subarachnoid hemorrhage, G6P/F6P, glucose 6-phosphate/fructose 6-phosphate, HHS, Hunt-Hess Scale, PCA, principal component analysis, R/R/X-5P, ribose/ribulose/xylulose 5-phospahte

Hierarchical clustering algorithms were used to group metabolites. The algorithm identified distinct correlated metabolite clusters in control patients (Fig. 1c) and patients with aSAH (Fig. 1d). In control patients, the TCA metabolites cis-aconitate, succinate, fumarate, malate, oxalate, glutamic acid, and hydroxyglutarate; glycolytic metabolites glucose/fructose, glucose-6-phosphagte/fructose-6-phosphate, and lactate; and the pentose phosphate pathway metabolites ribose/ribulose/xylulose formed a cluster (cluster 1). α-Ketoglutarate, glutamine, 3PG/2PG, and glyceraldehyde 3-phosphate were also strongly positively correlated, forming a cluster (cluster 2), while having neutral correlations with the previous cluster. Citrate formed strongly negative correlations with cluster 1 while having neutral to slightly positive correlations with cluster 2. By contrast, distinct clusters were not seen in patients with aSAH (Fig. 1d). However, α-ketoglutarate and glutamine remained positively correlated. α-Ketoglutarate and glutamine also formed negative associations with cis-aconitate. All correlations and levels of significance are shown in Table S3 for control patients and Table S4 for patients with aSAH.

Associations with Outcome

Univariate models were initially developed to assess the association between each metabolite and outcome variable (Table S5). Higher citrate was associated with worse discharge mRS (odds ratio [OR] 0.87 [95% confidence interval {CI} 0.78–0.97], P = 0.012) and 3-month mRS (OR 0.89 [95% CI 0.81–0.99], P = 0.043). Higher α-ketoglutarate was associated with better discharge mRS (OR 1.12 [95% CI 1.03–1.21], P = 0.038); however, this effect was not observed at 3 months (OR 1.05 [95% CI 0.97–1.13], P = 0.56). Similarly, higher glutamine was associated with improved discharge mRS (OR 1.11 [95% CI 1.02–1.23], P = 0.025), with this effect losing significance at 3 months (OR 1.05 [95% CI 0.97–1.15], P = 0.56). No metabolites were associated with DCI.

Multivariable logistic regression models were developed accounting for HHS, sex, Glasgow Coma Scale, and age (Table 2). Higher citrate was associated with worse discharge (OR 0.36 [95% CI 0.16–0.73], P = 0.008) and 3-month mRS (OR 0.35 [95% CI 0.14–0.81], P = 0.022), whereas higher fumarate (OR 1.77 [95% CI 1.15–2.82], P = 0.009) was associated with better discharge mRS. Higher HHS and Glasgow Coma Scale were associated with worse discharge and 3-month mRS. No metabolite was associated with DCI. Higher HHS was associated with DCI (OR 0.14 [95% CI 0.03–0.53], P = 0.007), and male sex was associated with absence of DCI (OR 3.58 [95% CI 1.04–1.71], P = 0.041).

Table 2.

Associations with outcome

Parameter	Good mRS (discharge)	Good mRS (3 months)	Absence of DCI
Citrate	0.36 (0.12–0.69) P = 0.008 *	0.35 (0.18–0.81) P = 0.022	0.90 (0.47–1.72) P = 0.748
Fumarate	2.04 (1.23–3.64) P = 0.009	1.08 (0.66–1.84) P = 0.764	1.00 (0.69–1.50) P = 0.999
Glutamine	0.54 (0.02–12.6) P = 0.0702	3.39 (0.15–9.69) P = 0.452	0.88 (0.07–12.1) P = 0.923
α-Ketoglutarate	3.06 (0.20–5.64) P = 0.433	0.34 (0.02–541) P = 0.453	0.73 (0.07–6.96) P = 0.792
HHS	0.14 (0.02–0.66) P = 0.019	0.79 (0.20–0.94) P = 0.020	0.14 (0.03–0.53) P = 0.007
Sex (male)	1.77 (0.49–6.92) P = 0.391	1.79 (0.15–3.18) P = 0.639	3.53 (1.02–1.70) P = 0.041
GCS	0.16 (0.03–0.64) P = 0.013	0.05 (0.009–0.23) P = 0.0002	2.77 (0.75–13.7) P = 0.158
Age	0.89 (0.85–0.94) P = 3.35 × 10−5	0.95 (0.90–0.99) P = 0.049	1.03 (0.99–1.07) P = 0.180

Results are presented as OR (95% CI). The four metabolites shown were included in each multivariable model

^*Statistically significant values are bolded

CI confidence interval, DCI delayed cerebral ischemia, GCS Glasgow Coma Scale, HHS Hunt Hess Scale, mRS, modified Rankin Scale, OR odds ratio

Associations with Cytokines/Chemokines

Hierarchical clustering algorithms were used to group correlated metabolites and cytokines/chemokines. Algorithms were applied to metabolite and inflammation panel data obtained from all patients with aSAH (Fig. 2a), those with low disease severity (HHS ≤ 3) (Fig. 2b), and those with high disease severity (HHS ≥ 4) (Fig. 2c). We focus here on significant correlations between cytokines/chemokines and those metabolites found to have associations with outcomes, with all significant correlations and corrected P values listed in Table S4 for all patients, Table S6 for low grade patients, and Table S7 for high grade patients.

Fig. 2.

Association between metabolites and cytokines. Each pixel in a dendrogram is a correlation coefficient value between a cytokine pair corresponding to that row and column. White, red, and black pixels indicate positive, negative, or near-zero correlations, respectively. Because the correlational matrix is symmetric, the upper triangle is a duplicate of the lower triangle. The diagonal elements are self-correlations and are equal to one. The clustering algorithm groups cytokines that are most correlated with each other. α-KG, alpha-ketoglutarate, CCL5, C–C motif chemokine ligand 5, CCL11, C–C motif chemokine ligand 11, CSF2, colony stimulating factor 2, G3P, glyceraldehyde 3-phosphate, IFN, interferon, IL-6, interleukin-6, IL-8, interleukin-8, MIP, macrophage inflammatory protein, PDGFAB1, platelet derived growth factor AB1, sCD40L, soluble CD40 ligand, TNFα, tumor necrosis factor alpha

A strong positive correlation between α-ketoglutarate and glutamine persisted among patients with all clinical severities (all: r = 0.98, P = 0.00; low grade: r = 0.98, P = 0.00; and high grade: r = 0.98, P = 0.00). Both α-ketoglutarate and glutamine were negatively correlated with the proinflammatory cytokine interleukin-8 (IL-8) (α-ketoglutarate: r = − 0.24, P = 0.0084; glutamine: r = − 0.27, P = 0.0028), with glutamine also being negatively correlated with macrophage inflammatory protein-1 alpha (MIP1A) (r = − 0.25, P = 0.0049). Negative correlations were present with other proinflammatory cytokines including IL-6, interleukin-8 (IL-8), C–C motif chemokine ligand 11 (CCL11), colony stimulating factor 2 (CSF2), colony stimulating factor 3 (CSF3), MIP1A, MIP-1B, interferon gamma (IFN-γ), tumor necrosis factor alpha (TNF-α), and the soluble CD40 ligand (CD40L) but did not reach significance. Fumarate was negatively correlated with the proinflammatory cytokine TNF-α (r = − 0.24, P = 0.0072). These correlations were not present among patients with low clinical severity, although citrate demonstrated a strong negative associated with colony stimulating factor 2 (r = − 0.34, P = 0.0017). Among patients with high clinical severity, both α-ketoglutarate and glutamine were negatively associated with IL-10 (α-ketoglutarate: r = − 0.48, P = 0.0030, glutamine: r = − 0.50, P = 0.0016). A negative correlation was also seen between citrate and C–C motif chemokine ligand 11 (r = − 0.47, P = 0.0034).

Predictive Models Including Systemic Metabolites

Logistic regression models were developed using systemic metabolites to predict outcomes (good [mRS ≤ 3] and bad [mRS ≥ 4]). Baseline model included age, sex, and HHS at admission. Models were developed for both discharge (Fig. 3a) and 3-month outcomes (Fig. 3b).

Fig. 3.

Receiver operating characteristics curves. Machine learning models were developed in to determine the ability of metabolites to predict outcomes after aSAH. Baseline models were developed by using sex and Hunt-Hess score at admission. Models were developed by using α-ketoglutarate, citrate, or a combination of α-ketoglutarate and citrate. Models are shown for both discharge (a) or 3-month (b) functional outcomes. Considering discharge functional outcomes, the model using α-ketoglutarate had a significantly higher area under the curve than the baseline model (difference of 0.106, 95% confidence interval 0.011–0.202)

AUC for the baseline model to predict discharge outcomes was 0.768 (95% CI 0.626–0.877). Adding citrate to the baseline model improved AUC to 0.882 (95% CI 0.757–0.956) (P = 0.0143). Adding fumarate to the baseline model resulted in a nonsignificant increase in AUC to 0.848 (95% CI 0.717–0.935) (P = 0.186). Adding a combination of citrate and fumarate resulted in a nonsignificant increase in AUC to 0.846 (0.714–0.933) (P = 0.089).

For the prediction of 3-month outcomes, the AUC for the baseline model was 0.745 (95% CI 0.584–0.868). Adding citrate (0.724 [95% CI 0.562–0.852], P = 0.768), fumarate (0.747 [95% CI 0.587–0.870], P = 0.970), or a combination of citrate and fumarate (0.739 [95% CI 0.578–0.863], P = 0.935) did not significantly improve AUC above the baseline model.

Network Models

Networks were developed to visualize the interactions between metabolites and cytokines in control patients (Fig. 4a) and patients with aSAH (Fig. 4b). Associations between cytokines and metabolites are only shown for citrate, α-ketoglutarate, glutamine, and fumarate. Significantly more associations were seen in control patients compared with patients with aSAH, resulting in a much denser network. Among controls, citrate was an important node and showed strong negative correlations with key TCA cycle metabolites including malate, fumarate, succinate, oxalate, and glutamine. Conversely, among patients with aSAH, citrate positively correlated only with the downstream TCA cycle metabolites cis-aconitate. Among both control patients and patients with aSAH, α-ketoglutarate and glutamine were strongly correlated with each other but without significant interactions with other TCA cycle metabolites. In both control patients and patients with aSAH, α-ketoglutarate and glutamine were negatively correlated with IL-8 and MIP1A.

Fig. 4.

Network models. Networks were developed showing correlations between metabolites. Correlations were added for cytokines for selected metabolites only (citrate, fumarate, glutamine, and α-ketoglutarate). Each node represents a metabolite while each interaction represents a Pearson’s correlation coefficient. Positive correlations are shown in black while negative correlations are shown in red. Metabolites are color coded according to pathway as shown in the legend. 2PG, 2-phosphoglycerate, 3PG, 3-phosphoglycerate, aSAH, aneurysmal subarachnoid hemorrhage, CCL2, C–C motif chemokine ligand 2, CCL11, C–C motif chemokine ligand 11, CSF2, colony stimulating factor 2, CSF3, colony stimulating factor 3, CSF5, colony stimulating factor 5, G6P/F6P, glucose 6-phosphate/fructose 6-phosphate, IFNg, interferon-gamma, IL-6, interleukin-6, IL-8, interleukin-8, IP-10, interferon gamma–induced protein 10, MIP1A, macrophage inflammatory protein-1 alpha, PDGAFAA, platelet-derived growth factor-AA, PPP, pentose phosphate pathway, RRX-5P, ribose/ribulose/xylulose 5-phospahte, sCD40L, soluble CD40 ligand, TCA, tricarboxylic acid

Fumarate was positively correlated with downstream members of the TCA cycle in both control patients and patients with aSAH. However, associations with cytokines shifted, with fumarate being negatively correlated with TNF-α only in patients with aSAH. Negative correlations between fumarate and platelet-derived growth factor-AA (PDGFAA) and CCL5 seen in controls were not seen in patients with aSAH.

Discussion

Herein, we report differences in systemic metabolism that occur early after aSAH and suggest links with the systemic inflammatory response. In addition to detailing changes in levels of individual metabolites, we show that significant changes in metabolic networks occur after aSAH, resulting in a unique metabolic signature. Our results demonstrate that lower levels of citrate and higher levels of fumarate and are associated with better functional outcomes. These metabolites also have associations with levels of circulating cytokines/chemokines and may represent targets for treatments aimed at modulating systemic metabolism after aSAH.

Systemic metabolic changes were able to readily distinguish control patients from patients with aSAH (Fig. 1b) creating a unique metabolic signature. Overall, patients with aSAH demonstrated increased levels of glycolytic metabolites (3PG/2PG and lactate) and decreased levels of TCA cycle metabolites (succinate, fumarate, malate, and oxalate) (Fig. 1a). This glycolytic shift was more evident among patients with high grade injury (HHS ≥ 4), with these patients having higher levels of the glycolytic metabolite G3P and lower levels of the TCA cycle metabolites α-ketoglutarate and glutamine (Fig. 1a). Control patients demonstrated discrete clusters of metabolites (clusters 1 and 2) with citrate having numerous negative correlations [20], particularly on cluster 1 (Fig. 1c). Clear clusters of metabolites were not seen in patients with aSAH (Fig. 1d), suggesting a disjointed TCA cycle with loss of regulation. This metabolic signature may provide insight into systemic pathophysiology occurring after aSAH. A glycolytic shift has been reported in numerous disease states. Cancer cells have long been recognized to undergo metabolic reprogramming termed the Warburg effect in order support rapid growth and proliferation [21]. A similar glycolytic shift is seen in both myeloid and lymphoid cells in response to proinflammatory signals, with activated M1 polarized macrophages having a particular dependence on glycolysis [22]. The changes in levels of particular metabolites can also drive the production of proinflammatory cytokines [23].

Key metabolites had associations with clinical outcomes, with higher levels of fumarate and lower levels of citrate being correlated with better discharge functional outcomes after controlling for confounding variables (Table 2). Only citrate also had an association with 3-month functional outcomes in multivariate models. However, predictive models including citrate in addition to baseline variables only showed an improved ability to predict discharge functional outcomes and not 3-month outcomes (Fig. 3). We propose that the changes in systemic metabolites, especially citrate, represent key pathophysiological processes occurring after aSAH. Future studies will be needed to elucidate the functional role of each metabolite after injury. Studies with larger sample sizes will also be necessary to refine predictive models combining clinical variables with key metabolites and other biomarkers to improve predictions of longer term functional outcomes.

Our results show associations with citrate and outcomes as well as numerous associations with other metabolites that are lost after aSAH (Fig. 4). The functions of citrate are multifactorial and complex [20]. Citrate is required for fatty acid synthesis, which can be used for cell growth and proliferation, membrane repair, and production of proinflammatory cytokines. Citrate derived acetyl-CoA can generate prostaglandins while citrate derived oxaloacetate produces reduced nicotinamide adenine dinucleotide phosphate (NADPH) that can generate nitric oxide via inducible nitric oxide synthase (iNOS) and reactive oxygen species via NADPH oxidase (Fig. 5) [20]. Citrate is also able to generate the metabolite itaconate, which is anti-inflammatory and has been shown to decrease production of IL-1β, IL-12p70, and IL-6 in activated macrophages [24]. Itaconate is able to inhibit succinate dehydrogenase (which converts succinate to fumarate) and citrate is able to inhibit pyruvate dehydrogenase and indirectly inhibit pyruvate kinase, thereby limiting pyruvate utilization by the TCA cycle [25]. While we are not able to draw mechanistic conclusions from our results, it is possible that these inhibitory effects on the TCA cycle may account for the negative correlations between citrate and downstream metabolites seen in control patients. Given the association between citrate levels and poor outcomes in patients with aSAH, we suspect that after aSAH citrate plays a more proinflammatory role and may be required for lipogenesis to generate intermediates for membrane repair.

Fig. 5.

Schematic overview. A schematic of TCA cycle metabolites and proposed changes occurring after aSAH is shown. Citrate has numerous associations with downstream TCA cycle metabolites in controls patients, with breaks occurring after citrate and succinate in patients with aSAH, allowing citrate to be used for lipogenesis and membrane repair as well as production of cytokines and reactive oxygen production. Glutamine is able to provide an alternative point of entry into the TCA cycle via α-ketoglutarate. aSAH, aneurysmal subarachnoid hemorrhage, OAA, oxaloacetate, PGE₂, prostaglandin E₂, TCA, tricarboxylic acid

In activated cells (e.g., M1 polarized macrophages), flow through the TCA cycle is impaired after citrate and after succinate (Fig. 5) [26]. Importantly, entry into the TCA cycle can occur at other points, as well. Glutamine is converted to glutamate by glutaminase and then into α-ketoglutarate, which can be used in the TCA cycle (Fig. 5) [27]. In certain cells, especially rapidly proliferating cancer cells, glutaminolysis can be an important point of entry into the TCA cycle [28]. The aspartate-arginosuccinate shunts is also able to provide entry into the TCA cycle at fumarate by synthesizing asginosuccinate from aspartate and citrulline, which can then generate fumarate [9]. Significant differences in the interconnectivity of metabolites after aSAH (Fig. 4) suggests a shift in the use of particular metabolites. The strong correlation between α-ketoglutarate and glutamine was present in control patients and patients with aSAH, suggesting that the close relationship between these metabolites may still play an important role in supporting oxidative metabolism after aSAH.

Metabolites exhibited strong associations with the proinflammatory cytokines (Figs. 2, 4). Particularly, α-ketoglutarate and glutamine showed significant negative correlations with the proinflammatory cytokine IL-8. IL-8 plays a critical role in acute inflammation, providing a potent signal to attract neutrophils to sites of injury [29]. Peripheral levels of IL-8 have been shown to peak by 30 min after aneurysm rupture in an animal model [30], and IL-8 has been shown to be elevated in the CSF early after aSAH during operative aneurysmal repair [31]. Higher levels of IL-8 have been shown to associated with higher injury severity (HHS ≥ 4) and poor functional outcomes [7]. Fumarate was negatively associated with TNFα. TNFα plays a pivotal role in the innate inflammatory response and is able to induce proinflammatory phenotypes in other cell types, such as monocytes [32]. TNFα is able to induce IL-8 production in a nuclear factor-κB (NF-κB) dependent fashion [33]. Increased plasma levels of TNFα have been linked with poor outcomes after aSAH; however, no effect was observed on the occurrence of vasospasm [34]. Supplementation with α-ketoglutarate has been shown to attenuate NF-κ signaling and decrease levels of TNFα [35], and this may represent a therapeutic target after aSAH to bolster oxidative metabolism.

Some of the key metabolites described herein have previously been shown to have potential pathophysiological effects. Supplementation with α-ketoglutarate has been shown to extend lifespan in C elegans by interacting with adenosine triphosphate synthase [36]. α-Ketoglutarate also has been shown to extend lifespan in a murine model and decrease levels of proinflammatory cytokines (IL-1α/β, IL-6, IL-8, CCL2, and CXCL-1) [37]. Administration of ornithine α-ketoglutarate was also able to improve brain oxygen utilization under hypoxic conditions in canines [38]. Similarly, supplementation with glutamine has been shown to decrease proinflammatory cytokines production (IL-6, IL-8, IL-1β) in a murine model of acute lung injury by modulating toll-like receptor-4 signaling [39]. Importantly, in a murine model, microglia were able to readily metabolize glutamine under conditions of hypoglycemia [40], suggesting the relevance of glutamine as fuel source for the central nervous system. The anti-inflammatory effects of fumarate have been well described. Dimethyl fumarate (DMF) is a fumaric acid ester that has immunomodulatory properties and is an effective treatment for multiple sclerosis. DMF is hydrolyzed to monomethyl fumarate in the small intestine [41]. Monomethyl fumarate is subsequently hydrolyzed inside of cells to fumarate, which can enter the TCA cycle [42]. DMF is thought to modulate inflammation by activating nuclear factor (erythroid-derived 2)-like 2 and inhibiting NF-κB [43]. Increased fumarate levels may also be able to reverse the direction of succinate dehydrogenase, further attenuating inflammation [44]. Fumarate has also been shown to modulate neuroinflammation and play a neuroprotective role in models of ischemic stroke [45-47]. Preclinical models in SAH are currently lacking, and future studies will be required to determine the mechanistic role of these metabolites in SAH and whether metabolite supplementation may be able to impact clinical outcomes.

This study has several limitations. Because all patients are from one tertiary care center, generalizability may be limited. However, standard care was provided to all patients with aSAH, and the results consisted of two separate cohorts. All results were obtained from plasma samples; therefore, it is impossible to determine the cellular origin of metabolic perturbations described herein. There was an imbalance in ethnicity, with Hispanic patients being overrepresented in the aSAH cohort. Given the different prevalence of obesity and metabolic syndrome as well as potentially shifted metabolomic profile among Hispanics [48, 49], this imbalance may have impacted the differences seen between control patients and patients with aSAH. This study also looked at metabolites at a single, early timepoint. Given the occurrence of DCI often a week or later after injury, it is possible that our early timepoint is not the most relevant to the pathophysiology of DCI. This may account for why no associations were found with DCI and limits our ability to detect later changes in metabolites that may affect functional outcomes. It is also not possible for us to determine metabolic flux (i.e., the balance between synthesis and breakdown of each metabolite). Future studies will be required using multiple timepoints after injury to elucidate metabolic flux after aSAH and to detect later metabolic changes that may play key pathophysiological roles. This will be particularly important to understand the mechanistic effects of α-ketoglutarate, glutamine, citrate, and fumarate after aSAH.

Conclusions

We have demonstrated a unique metabolic signature occurring systemically after aSAH. Changes in key metabolites may have important effects on the systemic proinflammatory response to brain injury. We have identified metabolites (citrate, fumarate, α-ketoglutarate, and glutamine) that are associated with early clinical outcomes.