Serum proteomic analysis of novel predictive serum proteins for neurological prognosis following cardiac arrest

Early prognostication of neurological outcome in comatose patients after cardiac arrest (CA) is vital for clinicians when assessing the survival time of sufferers and formulating appropriate treatment strategies to avoid the withdrawal of life‐sustaining treatment (WLST) from patients. However, there is still a lack of sensitive and specific serum biomarkers for early and accurate identification of these patients. Using an isobaric tag for relative and absolute quantitation (iTRAQ)‐based proteomic approach, we discovered 55 differentially expressed proteins, with 39 up‐regulated secreted serum proteins and 16 down‐regulated secreted serum proteins between three comatose CA survivors with good versus poor neurological recovery. Then, four proteins were selected and were validated via an enzyme‐linked immunosorbent assay (ELISA) approach in a larger‐scale sample containing 32 good neurological outcome patients and 46 poor neurological outcome patients, and it was confirmed that serum angiotensinogen (AGT) and alpha‐1‐antitrypsin (SERPINA1) were associated with neurological function and prognosis in CA survivors. A prognostic risk score was developed and calculated using a linear and logistic regression model based on a combination of AGT, SERPINA1 and neuron‐specific enolase (NSE) with an area under the curve of 0.865 (P < .001), and the prognostic risk score was positively correlated with the CPC value (R = 0.708, P < .001). We propose that the results of the risk score assessment not only reveal changes in biomarkers during neurological recovery but also assist in enhancing current therapeutic strategies for comatose CA survivors.

circulation (ROSC). 2,3 Survivors of cardiac arrest (CA) often require lengthy intensive care admission, rehabilitation and ongoing therapy of chronic complications because of poor functional outcome. 4,5 Thus, early and accurate identification of neurological outcome among CA patients is vital for clinicians when assessing the survival time of sufferers and formulating appropriate treatment strategies to avoid the withdrawal of life-sustaining treatment (WLST) from patients who may have a good neurological outcome and the use of futile treatments in patients who may have a poor neurological outcome.
Repeated clinical examination in combination with electroencephalography, electrophysiological tests, computed tomography (CT) and magnetic resonance imaging (MRI) are generally employed for early judgments of neurological outcome. 6 However, the operation of most of these devices requires great skill and expertise for accurate judgment, 7 and the evidence derived from these methods is not consistent. 8,9 In addition, neuron-specific enolase (NSE), a dimeric isoform of the glycolytic enzyme enolase found mainly in neurons, is present in serum following different brain disorders. 10 NSE levels have been found to increase in comatose CA survivors with severe brain injury, and the levels of NSE at 72 h after CA have been regarded as an effective prognostic marker, according to the European Resuscitation Council and European Society of Intensive Care Medicine (ERC-ESICM) guidelines for prognostication after CA. 11,12 Targeted temperature management (TTM) in combination with several drugs has become the standard treatment for comatose CA patients, as TTM effectively improves the recovery of neurological function; nevertheless, the above-mentioned data originated from a cohort study performed prior to the widespread implementation of TTM. False-positive rates have been reported for nearly one-third of comatose CA patients after TTM. 13 Furthermore, newly developed biomarkers such as protein S-100B do not add any real value to current prognostication models with NSE. 14 Therefore, the development of neuronal injury biomarkers must be approached from a new angle so that clinical therapeutic strategies for comatose CA survivors can be improved.
Recent successes reported in the literature have illustrated the role of mass spectrometry-based proteomics as an indispensable analytical tool in the workflow for disease biomarker discovery based on liquid biopsy. 15 In our work, we performed an investigation of serum proteins that were differentially expressed in comatose CA survivors with good versus poor neurological recovery using an isobaric tag for relative and absolute quantitation (iTRAQ)-based proteomic approach. Differentially expressed proteins were validated via an enzyme-linked immunosorbent assay (ELISA) approach in a larger-scale independent sample, and a prognostic risk score was developed and calculated using a linear and logistic regression model based on a combination of select proteins and NSE. We propose that the results of the risk score assessment not only reveal changes in biomarkers during neurological recovery but also assist in enhancing current therapeutic strategies for comatose CA survivors.

| Patient sample collection
From January 2014 to December 2019, we conducted a retrospective observational study of OHCA patients at Nanjing Drum Tower Hospital (Nanjing, China). The participants were biologically unrelated, and informed written consent was obtained before sample collection. The exclusion criteria were as follows: failure to achieve the ROSC; age < 18 years; past history of irreparable brain damage; the recovery of consciousness after the ROSC without undergoing TTM; death within 72 h after the ROSC; did not receive TTM; and known haematological disease. Blood samples were immediately collected from venous blood puncture into EDTA tubes during the course of routine intensive care at 72 h after the ROSC. From each sample, 2 mL of blood was allowed to clot at 4°C for at least 2 h and then was centrifuged at 1000 × g for 30 min to precipitate blood cells. Then, the serum was collected and divided into aliquots followed by freezing at -80°C until analysis.

| Proteomic digestion and iTRAQ labelling
Before the proteomic approach, a Human 14 Multiple Affinity Removal LC Column (Agilent, USA) was used to deplete highly abundant proteins according to the manufacturer's instructions.
Then, concentrated serum samples were mixed with 1 × SDT lysis buffer (4% SDS, 100 mM Tris-HCl, 1 mM DTT, pH 7.6) and then heated in boiling water for 15 min. After centrifugation at 14,000 rpm for 20 min, the supernatant was collected. Depleted serum protein concentrations were detected using a BCA protein assay kit (Pierce, USA). A total of 100 μg of sample was reduced and alkylated by a final concentration of 10 mM dithiothreitol and 50 mM iodoacetamide, respectively. Then, the digested samples with 2.5 μg of trypsin (Sigma, USA) were individually labelled with iTRAQ reagents at room temperature for 1 h. Finally, the six labelled peptide aliquots were combined for subsequent fractionation.

| ELISA validation and measurement
Serum angiotensinogen (AGT), alpha-1-antitrypsin (SERPINA1), transgelin-2 (TAGLN2) and Talin-1 (TLN1) concentrations were determined by specific human ELISA kits (Abcam, USA), and experimental steps were performed according to the manufacturer's instructions. In addition, ELISAs were performed in a clinical laboratory to measure the NSE level.

| Diagnostic scores for neurological outcome development
The serum level of a candidate protein detected by an ELISA was subjected to receiver operating characteristic (ROC) analysis to evaluate the sensitivity and specificity to distinguish a poor neurological outcome from a good neurological outcome and to elucidate the prognosis of neurological outcome in comatose CA survivors. Linear and logistic regression patterns were established for the risk score of neurological outcome diagnosis.

| Statistical analysis
Statistical analysis was performed for the normal distribution and homogeneity of variance test with SPSS software 22.0. Student's t test (two tails) and a one-way analysis of variance (ANOVA) were employed to examine the difference between subjects with a normal distribution and a homogeneity of the variance, while the Benjamini-Hochberg approach was utilized to test the FDR.
The Chi-Square-test or the Fisher exact test was used to compare categorical data. ROC curves were generated to differentiate a poor neurological outcome from a good neurological outcome by the standard method of sensitivity and specificity. The results are shown as the means ± the standard deviation (SD). The generally accepted level of significance was P < .05. The relative intensity value was acquired by comparing the intensity of a sample to that of the sample with the lowest intensity.

| Patient characteristics
In the present work, we explored whether serum fingerprint proteins in early-stage CA patients could predict neurological outcome. The Then, three GNO patients and three PNO patients were randomly selected and assigned to these two groups mentioned above. Detailed demographics of the enrolled patients are presented in Table 1.

| Proteomic discovery using the iTRAQ approach
Subsequently, we carried out iTRAQ labelling-based proteomics to identify differentially expressed proteins that are related to the prognosis of neurological recovery using serum samples collected from three GNO and three PNO patients. First, we depleted high-abundance serum proteins by an immunodepletion kit before proteomic analysis because thousands of dynamic proteins from extremely high abundance to low abundance exist in serum. In total, 5,738 peptides

| Biomarker validation by ELISA and the predictive power of serum proteins for neurological prognosis
The validation of candidate protein biomarkers is crucial for clinical applications. A set of significantly differentially expressed proteins (AGT, SERPINA1, TAGLN2 and TLN1) in the cohorts (n = 78, 32 GNO patients and 46 PNO patients) were selected and measured using ELISAs. Selection was dependent on the following basic selection experimental guidelines: containing more than one unique peptide being identified and quantified, differential expression fold change ≥ 2, multiple hypothesis testing (FDR < 0.05) and significance based on a standard Student's t test (P < .05). As shown in Figure 3A, the serum concentrations of AGT and SERPINA1 were predominantly up-regulated, whereas no significant difference in TAGLN2 or TLN1 was found in PNO patients. To evaluate the power of these differentially expressed serum proteins for the prediction of neurological prognosis, we performed a ROC curve analysis. The  Figure 3B). It has been widely assumed that the corresponding multiple effective biomarkers could be combined to generate a single score to assist clinicians in making a better diagnostic assessment, 16 with an increase in the utilization of multiplex biomarkers. Thus, multivariate logistic regression analysis was carried out for these two markers combined with NSE. As shown in Figure 3B, the AUC of the combined results for NSE and AGT was 0.819 (95% CI = 0.728 to 0.911), and the AUC of the combined results for NSE and SERPINA1 was 0.832 (95% CI = 0.741 to 0.922), which was superior to that of the NSE biomarker alone. The AUC of the combined results of AGT, SERPINA1 and NSE was 0.859 (95% CI = 0.778 to 0.939); this combination showed the best prognostic efficiency with the highest sensitivity and specificity to differentiate a poor neurological prognosis from a good neurological prognosis.

| Development of a prognostic panel for neurological outcome
To further elucidate the signature of the combination of AGT, SERPINA1 and NSE for predicting neurological prognosis, we Subsequently, all CA participants were divided into low-and highrisk score groups according to the threshold point value (1.748). The proportion of PNO individuals in the high-risk score group was predominantly higher than that in the low-risk score group according to an analysis of the association between the risk score distribution and the neurological outcome status ( Figure 4C). Interestingly, patients with higher risk scores were in the PNO group. Further investigation showed that the high-risk score group had a significantly higher rate of poor neurological outcome (89.19%, 33/37) than the low-risk score group (31.70%, 13/41) ( Figure 4D). Then, a linear correlation analysis demonstrated that the risk score was positively correlated with the CPC value (R = 0.708, P < .001) ( Figure 4E).

| D ISCUSS I ON
Disturbances in neurocognitive performance are a core feature among CA survivors. Over the last few years, the therapeutic management of patients who are comatose after CA has been widely improved. However, few therapeutic interventions are available to completely alleviate neurological damage after CA [7]. Determining The principal advantage of the use of this specific proteomic method to search for candidate biomarkers is that it permits the simultaneous measurement of multiple proteins within a single sample. 18 In addition, it is not essential to define ahead of time which biomarkers should be analysed. For example, NSE, which is released into the bloodstream once the nervous system is injured, serves as a prognostic biomarker after CA. Recently, other potential biomarkers, including S100, 14 B-type natriuretic peptide, 19 non-coding RNA 20 and cell-free DNA, 21 have been investigated. However, their suboptimal sensitivity and specificity limit their application. 22 Moreover, the unifying principal rule of these prior research studies is that these proteins were known in advance. In several recent studies, increasing attention has been focused on the identification and description of disease progress by non-invasive surrogate markers from fluid biopsy, such as serum or plasma. The advantages of fluid biopsy biomarkers include the following: a) the biomarkers can be detected in a non-invasive way; b) fluid biopsy can be performed in ambulatory settings; and c) fluid biopsy is a reliable source of biomarkers and can be repeatedly analysed. 17 Using serum proteomics as a "discovery-based approach" to protein profiling in disease biomarker research has also been recommended recently. 23 Indeed, previous work revealed novel predictive biomarkers that could be used to determine neurological recovery using two-dimensional gel electrophoresis (2D-GE) coupled with matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) MS. 22 In the present study, we compared the variation in serum protein differences using iTRAQ labelling-based quantitative proteomics, and we identified TA B L E 1 Detailed demographics of the enrolled patients. Values are presented as median (interquartile range) or number (%). additional proteins and more differentially expressed proteins than were reported in prior work. 22 In this regard, the global-tagging iTRAQ technique was more sensitive than 2D-GE, as the comigration and partial comigration of proteins could compromise the repeatability and accuracy of quantification in the 2D-GE approach. 24 AGT, expressed as a constitutive protein by the liver and various other tissues, including the brain, interacts with renin to generate the prohormones angiotensin Ⅱ and angiotensin. 25 An increasing amount of evidence has suggested that most AGT mRNA is expressed in the brain, comprising approximately one-third of mRNAs originating in the liver that are present in ependymal and astrocytes. 26,27 A series of studies suggests that AGT produced by astrocytes could be constitutively secreted into perivascular space.

Detailed demographics of the enrolled patients
However, the relative proportion of AGT secreted into the circulation versus that located intracellularly and the role of the secreted extracellular AGT are unknown. In addition, the existence of a high level of AGT in astrocytes may also indicate a more comprehensive function for the brain renin-angiotensin system, 27 and AGT is required for the initiation of the brain renin-angiotensin system cascade, providing an impetus for a better understanding of its regulation in the brain. 28 The relationship between the renin-angiotensin system and brain function has been investigated and confirmed in previous work. 29  found to interfere with the activity of caspases 1 and 3, protecting pancreatic beta cells, 31 lung alveolar and endothelial cells 32,33 and skin fibroblasts 34 from the effects of apoptosis. In addition, SERPINA1 has been shown to down-regulate the levels of nitric oxide, leukotriene B4 and pro-inflammatory cytokines, 35 including interleukin-1b, interleukin-6, interleukin-8, interleukin-32, tumor necrosis factor (TNF)-a and monocyte chemoattractant protein-1, without preventing the delivery of the anti-inflammatory cytokine interleukin-10 into the blood. SERPINA1 was also found to protect host cells from the influence of microorganisms, preventing their replication and infectivity. 36 However, the fact that oxidants, 37 metalloproteinases 38 and the excision of its reactive site loop inactivate SERPINA1 by oxidation may cause it to lose some of its effectiveness under inflammatory conditions. Recently, SERPINA1 was reported to be a chaperone in amyloidoses, and it plays a role in amyloid aggregation, which may lead to the pathogenesis of amyloidotic diseases, for example amyotrophic lateral sclerosis (ALS), familial amyloid polyneuropathy (ATTR) and Alzheimer's disease. [39][40][41] Previous findings have revealed that AGT belongs to another ser- would also like to thank American Journal Experts for proofreading the article.

CO N FLI C T O F I NTE R E S T
The authors confirm that there are no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.