Integrated Multiomics Analysis Identifies a Novel Biomarker Associated with Prognosis in Intracerebral Hemorrhage

Existing treatments for intracerebral hemorrhage (ICH) are unable to satisfactorily prevent development of secondary brain injury after ICH and multiple pathological mechanisms are involved in the development of the injury. In this study, we aimed to identify novel genes and proteins and integrated their molecular alternations to reveal key network modules involved in ICH pathology. A total of 30 C57BL/6 male mice were used for this study. The collagenase model of ICH was employed, 3 days after ICH animals were tested neurological. After it, animals were euthanized and perihematomal brain tissues were collected for transcriptome and TMT labeling-based quantitative proteome analyses. Protein-protein interaction (PPI) network, Gene Set Enrichment Analysis (GSEA), and regularized Canonical Correlation Analysis (rCCA) were performed to integrated multiomics data. For validation of hub genes and proteins, qRT-PCR and Western blot were carried out. The candidate biomarkers were further measured by ELISA in the plasma of ICH patients and the controls. A total of 2218 differentially expressed genes (DEGs) and 353 differentially expressed proteins (DEPs) between the ICH model group and control group were identified. GSEA revealed that immune-related gene sets were prominently upregulated and significantly enriched in pathways of inflammasome complex, negative regulation of interleukin-12 production, and pyroptosis during the ICH process. The rCCA network presented two highly connective clusters which were involved in the sphingolipid catabolic process and inflammatory response. Among ten hub genes screened out by integrative analysis, significantly upregulated Itgb2, Serpina3n, and Ctss were validated in the ICH group by qRT-PCR and Western blot. Plasma levels of human SERPINA3 (homologue of murine Serpina3n) were elevated in ICH patients compared with the healthy controls (SERPINA3: 13.3 ng/mL vs. 11.2 ng/mL, p = 0.015). Within the ICH group, higher plasma SERPINA3 levels with a predictive threshold of 14.31 ng/mL (sensitivity = 64.3%; specificity = 80.8%; AUC = 0.742, 95% CI: 0.567-0.916) were highly associated with poor outcome (mRS scores 4-6). Taken together, the results of our study exhibited molecular changes related to ICH-induced brain injury by multidimensional analysis and effectively identified three biomarker candidates in a mouse ICH model, as well as pointed out that Serpina3n/SERPINA3 was a potential biomarker associated with poor functional outcome in ICH patients.


Introduction
Intracerebral hemorrhage (ICH), which accounts for 15-20% of all strokes, is the most devastating stroke subtype, contributing to high mortality and disability in the world [1][2][3]. However, no effective pharmaceutical or surgical treatments, which can treat ICH-induced brain injury, are available, and as such, novel exploration of ICH-related biomarkers and therapeutic targets is urgently needed [4].
Secondary injury develops soon after ICH onset and is mostly driven by the inflammatory response, edema, oxidative stress, and excitotoxicity [5,6]. Preclinical studies have demonstrated that cascades of inflammatory processes occur around the hematoma [7][8][9][10]. Almost simultaneously, mitochondrial dysfunctions, oxidative stress, and neurotoxicity mediated injury could result in the blood-brain barrier (BBB) impairment and aggregate the formation of perihematomal edema [11][12][13][14]. Generally speaking, secondary brain injury following ICH is not caused by a single factor, but a cascade of amplified reactions caused by multiple pathological factors working in concert. In this regard, deepened insight into the underlying mechanism of ICH injury is required to aid in identifying molecular indicators of ICH and provide potential intervention targets for clinical treatment.
Progress in omics-based approach with high-throughput data is emerging as an available approach to enhance the understanding of the pathophysiology of ICH. Transcriptomics technologies, which focus on both coding and noncoding sequences [5], have been applied to brain tissue and peripheral blood. Previous studies have identified that several transcript panels are related to ICH diagnosis and prognosis [15][16][17]. Meanwhile, proteomics approaches have made it an ideal strategy for molecular mechanism studies in ICH [18][19][20][21]. However, given the complexity and variability of pathophysiological processes involved in brain injury after ICH, such independent analysis from each level of omics may overlook the crosstalk between different molecular entities and could miss biologically relevant information [22]. In this context, integrated analysis of multiomics (intergromics) has emerged as a novel approach to facilitate interpretation of the multidimensional data and provide a global-wide insight into the pathological changes of ICH.
In the present study, we firstly explored the main transcriptomics and proteomics changes that occur in the brains of ICH mice. Then, we combined experimental data of two omics simultaneously by using integrative analysis and further identified the key significant expression molecules in the acute phase of ICH. Finally, we assessed the association between the plasma level of highlighted candidate molecule and clinical characteristics of ICH patients.

Materials and Methods
2.1. Animals. 30 C57BL/6 male mice (8)(9)(10)(11)(12) weeks old, 20-25 g) were purchased from the Laboratory Animal Center of Chongqing Medical University. All the animals were individually housed under the standard conditions (temperature of 21-24°C, humidity of 52°, 12 h light-dark cycle). All animal experiments were approved by the Ethical Commit-tee of Chongqing Medical University (ECCMU, Chongqing, China) and conducted in compliance with the National Institutes of Health Guidelines for animal research. Utmost efforts were taken to reduce the number and minimize suffering of animals.
2.2. Experimental ICH Model. ICH was induced in mice by collagenase injection under inhalation anesthesia with isoflurane as previously reported [23]. The mice were fixed in a stereotactic frame (RWD, Life Science, China) in a prone position. The scalp was incised along the midline, and a burr hole was drilled. The mice were injected in the right striatum (coordinates: 0.6 mm anterior, 3.7 mm ventral, and 2.3 mm lateral to the bregma) with type IV-S clostridial collagenase (Sigma, St. Louis, MO, USA; 0.075 U in 0.4 μL sterile 0.9%NaCl) by using a 26-gauge Hamilton syringe needle. The needle was left for 10 min and then slowly withdrawn after it. Control surgeries were performed using the same surgical procedures without injection of type IV collagenase. Exclusion criterion was death during or after surgery.
2.3. Behavioral Assessment. Neurological function evaluation was performed using the modified Garcia test and beam walking test at 3 days after ICH, as previously described [24]. The modified Garcia test consisted of 6 subtests including vibrissae touch, trunk touch, spontaneous activity, spontaneous movement of the four limbs, forelimbs outstretching, and climbing capacity. A maximal score of 18 was given to animals without apparent neurological dysfunctions. The beam walking score was determined by observing the walking distance of mice on a wooden beam within 1 minute: five points, reaching the opposite platform or walking ≥ 45 cm within 30 s; four points, reaching the opposite platform or walking ≥ 45 cm within 60 s; three points, walking ≥ 22:5 cm within 60 s; two points, walking < 22:5 cm within 60 s; one point, no walking but no falling over 30 s (any posture); and zero points, falling within 30 s [25]. The higher score, the better neurological function. All tests were performed by a blinded investigator.

Sample Collection and Preparation.
Mice were deeply anesthetized with pentobarbital sodium. Mouse brain samples were quickly removed and sliced into 1 mm coronal sections. The perihematomal brain tissues were obtained and then stored at -80°C for subsequent experiments, including microarray analysis, liquid chromatography-mass spectrometry/mass spectrometry (LC-MS/MS) analysis, quantitative RT-PCR (qRT-PCR), and Western blot.
2.5. Hematoxylin-Eosin (HE) Staining. Mice were deeply anesthetized and then transcardially perfused with 40 mL precold phosphate buffer saline (PBS). The brains were removed, fixed with 4% paraformaldehyde and embedded in paraffin. Then, 3 μm thick brain tissue sections were prepared. After the deparaffinization, the brain tissue sections were stained with hematoxylin/eosin according to the manufacturer's instructions. Sections were viewed using a light microscope (Eclipse Ci-L, Nikon, Japan).
2.6. Transcriptomics Study. Total RNA extraction was performed from brain tissues (n = 5, each group). 2.9. Gene Set Enrichment Analysis (GSEA). GSEA is mainly used to determine whether a predefined gene set can show significant differences between two biological states. We conducted GSEA analysis through the local version (http:// www.broadinstitute.org/gsea/index.jsp) and obtained important gene sets through the following parameters: NES > 1, NOM p value < 0.01, and FDR q value < 0.25.
2.10. Integrative Analysis. All DEGs and DEPs screened from transcriptomics data and proteomics data, respectively, were included in the follow-up analysis. To reveal the interaction networks underlying multidimensional data, different integrated bioinformatics approaches were used through the following four steps. Firstly, to obtain an overview of the expressive and regulated characteristics of DEGs and DEPs, the R package (version 3.6.3) was used to create a Venn diagram, and hierarchical clustering analysis was performed based on the normalized values of proteins. Then, Pearson's correlation analysis was performed to assess the correlation of expression levels between two high-throughput datasets.
Secondly, for those persistently dysregulated DEGs and DEPs, a protein-protein interaction (PPI) network with a strict combined score > 0:7 was created by the online database STRING (version 11.0) (https://string-db.org/) and was visualized by Cytoscape (version 3.8.0), a public bioinformatics analysis platform. Furthermore, CytoHubba was used to identify hub genes based on degree algorithms.
Thirdly, the mixOmics R package (version 6.3.1, http:// mixomics.org) was used to perform a regularized Canonical Correlation Analysis (rCCA) to investigate the most significantly correlated module between transcriptomics and proteomics data. As reported previously [26], omics data were repeatedly shrunk and optimized according to regularize the empirical covariance matrices of variables (X, Y) and tune two parameters (λ1, λ2) using cross-validation until 3 Oxidative Medicine and Cellular Longevity parameter λ1 = 0:05, parameter λ2 = 0:2, and the crossvalidation score = 0:87. Then, Biological Networks Gene Ontology (BiNGO) package (version 3.0.3) was used to conduct the biological process analysis of highly relevant networks (parameter setting: significance level: 0.05, weight: 0.8). The rCCA network and PPI network were combined with integrate consideration, and hub molecules were finally screened out.
2.11. qRT-PCR. Total RNA was extracted from mouse brain tissues by Trizol (Invitrogen, United States). For cDNA synthesis, the PrimeScript™ RT reagent kit was performed followed the manufacturer's instructions (RR047A, TaKaRa, Shiga, Japan). qRT-PCR was carried out using the SYBR Green detection system (Roche, Germany) according to the manufacturer's instructions. β-Actin was used as the internal control in all samples. The values of fold change for mRNA expression levels were calculated according to formula 2 -ddCT . The primer sequences are presented in Supplementary Table S1.

Western Blot.
Briefly, whole-cell lysates were obtained by gently homogenizing of brain samples in RIPA lysis buffer (P0013, Beyotime, China) and centrifuging (14,000 g at 4°C for 30 min). The supernatant was collected, and the protein concentration was determined using a bicinchoninic acid (BCA) assay (Bio-Rad, Dc protein assay). Equal amounts of protein (30 μg) were loaded and subjected to electrophoresis on a 10% SDS-PAGE gel. After being transferred onto a PVDF membrane (Bio-Rad), the membrane was blocked with 0.5% skimmed milk in TBST for 3 h at room temperature and was incubated with the primary antibody overnight at 4°C. Then, the membrane was incubated with corresponding secondary antibodies for 2 h at room temperature. Internal loading control β-actin was used. The primary antibodies were Serpina3n (1 : 500, AF4709, R&D Systems), Itgb2 (1 : 1000, 10554-1-AP, ProteinTech Group), Serping1 (1 : 1000, 12259-1-AP, ProteinTech Group), and Ctss (1 : 1000, DF8246, Affinity). β-Actin (1 : 10000, ab8226, Abcam) was used as the internal loading control. The blots were visualized using an ECL Kit (Millipore) and ChampChemi (Sage Creation Science, Beijing, China). The density of the bands was analyzed by Quantity One software (Bio-Rad, California, USA).

Human Blood Samples and Biomarker Measurement.
All blood samples of ICH patients were withdrawn and collected in EDTA tubes within 24 hours from ICH onset. Afterward, blood samples and controls were centrifuged at 3000 rpm for 10 minutes. Separated plasma from the top layer was preserved at -80°C until measurements. Plasma concentrations were gauged in duplicate samples with commercial enzyme-linked immunosorbent assay (ELISA) kits (KALANG Corp).
Clinical information was further collected, including baseline demographics, clinical admission status, laboratory data, and imaging data. A routine 90-day follow-up was conducted for ICH patients. Functional outcomes of ICH patients were assessed by modified Rankin Scale (mRS) at a 90-day follow-up, as described previously [27]. mRS scores 4-6 were defined as poor outcomes.
2.14. Statistical Analysis. Most of the methods of bioinformatics analysis have been described above. Besides, the SPSS package (version 24.0) and R (version.3.6.3) software were used for statistical analyses. All data were presented as the mean and standard deviation (mean ± SD). Student's t-test or one-way ANOVA was performed for statistical evaluation. The two-side p value < 0.05 was considered statistically significant at a 95% confidence level.

Histological and Functional Evaluation of Experimental ICH Model.
The overall workflow of this study is shown in Figure 1(a). Briefly, the workflow included three phases: (i) ICH modeling and tissue preparation, (ii) discovery study, and (iii) validation study. Of the 30 mice, 2 mice in the ICH group died due to excessive hematoma. The total mortality of the surgeries was 6.67% (2/30). Figure 1(b) shows representative images of ICH and sham brains, which verify the morphological and pathological changes. Figure 1(c) demonstrates the representative brain slices after H&E staining. Of note, macrophage infiltration (green arrows) and hemorrhages (black arrow) could be seen on the left side of the striatum in ICH mice compared with the controls. ICH induced severe neurological dysfunction, which was evaluated by modified Garcia and beam walking tests compared with the controls. (Figure 1(d)).

3.2.
DEGs in the Perihematomal Brain following ICH. The ipsilateral hemisphere was obtained to perform transcriptomic analysis. The box plot showed the distribution of gene expression levels in the ten perihematomal brain tissues samples ICH compared to the controls (Figure 2(a)). The correlation of gene expression levels between all samples is shown in Figure 2(b). The principal component analysis (PCA) plots showed that the samples in the ICH group and control group were separated in both 2D and 3D dimensions, indicating the gene expression pattern between these two groups was different (Figures 2(c) and 2(d)). These analysis results showed the reliability of the omics data was guaranteed.
A total of 2218 genes (2065 upregulated and 153 downregulated) were identified between the ICH group and control group as DEGs (Figure 2(e)). Hierarchical clustering analysis showed that these DEGs could significantly distinguish ICH vs. sham groups (Figure 2(f)). To further identify the function of DEGs, GO enrichment analysis and KEGG pathway enrichment analysis were subjected by using clusterProfiler R package. As shown in Figure 2(g) and Table 1, DEGs were significantly enriched in 1044 GO biological process (BP) terms, which identified leukocyte migration, myeloid leukocyte activation, and positive regulation of cytokine production as the top BP terms. The DEGs were significantly enriched in 56 GO cellular component (CC) terms, which identified collagen-containing extracellular matrix, extracellular matrix and condensed chromosome, 4 Oxidative Medicine and Cellular Longevity  5 Oxidative Medicine and Cellular Longevity and centromeric region as the top CC terms. Furthermore, the DEGs were observed in 101 GO molecular function (MF) terms, which identified chemokine activity, cytokine receptor activity, and cytokine binding as the top MF terms. As for KEGG pathway enrichment analysis, the results demonstrated that a total of 56 pathways were significantly enriched, which were mainly enriched in cytokine-cytokine receptor interaction, viral protein interaction with cytokine and cytokine receptor, and hematopoietic cell lineage. These findings indicated that immune-related changes were deeply involved in the acute stage of ICH.
3.3. DEPs in the Perihematomal Brain Proteome following ICH. The TMT-labelled quantitative proteomics analysis of perihematomal brain tissues was performed. First, we matched the peptides in the database, removed duplicate proteins, and eliminated the peptides and proteins with FDR greater than 1%. Then, a total of 6012 proteins were identified in the ICH group and control group, of which 353 proteins (198 upregulated and 155 downregulated) were screened out as DEPs (Figure 3(a)). Hierarchical clustering analysis results showed that these DEPs could significantly distinguish the ICH and control groups (Figure 3(b)). As shown in Figure 3(c) and Table 2, GO enrichment analysis and KEGG pathway enrichment analysis were performed using clusterProfiler R package. DEPs were significantly enriched in 670 GO BP terms, which identified blood coagulation, hemostasis, and coagulation as the top BP terms. DEPs were also significantly enriched in 73 GO CC terms, which identified collagen-containing extracellular matrix, extracellular matrix, and actin cytoskeleton as the top CC terms. Furthermore, these DEPs were enriched in 41 GO MF terms, which played essential roles in terms of actin binding, enzyme inhibitor activity, and endopeptidase regulator activity. Next, KEGG pathway enrichment analysis showed the results that DEPs were enriched 6 KEGG pathways, such as complement and coagulation cascades and lysosome.

Functional Analysis through GSEA.
To validate omics results and search for those gene sets with important biological functions, GSEA was subsequently performed based on overall genes following GO annotation. As shown in Figure 4(a), the top 20 enriched GO terms were presented ranking by enrichment scores (ES). The enrichment maps were obtained and summarized in Figure 4(b), which showed that immune-related gene sets were prominently upregulated during the ICH process compared with the controls, involving pathways including inflammasome complex, negative regulation of interleukin-12 production, and pyroptosis.
3.5. Integrative Multiomics Analysis following ICH. We performed integrative analysis to reveal the post transcriptomics regulation of gene expression and explore ICH-related hub molecules. We explored the corresponding relationship between the mRNA information obtained by the transcriptome and the protein information identified by the proteome. The result showed that a total of 5789 comolecules simultaneously changed in both transcriptomic and proteomics datasets. Among these comolecules, 62 molecules were identified as simultaneously differentially expressed in both two omics data, which might serve as molecular markers of ICH ( Figure 5(a)). Further, the transcriptome and proteome expression correlation analysis revealed that Pearson's correlation coefficient of comolecules was 0.332 (p = 0:011) (Figures 5(b) and 5(c)).
For the 62 comolecules, we established a PPI network with 44 nodes and 109 interactions ( Figure 5(d)). To further identify the key ICH-related molecules of PPI network, we screened the top 10 molecules ranked by degree via the CytoHubba plug-in, namely, Lyz2, C3, Serpina3n, Cfp,    Meanwhile, by constructing a gene-protein connection from DEGs and DEPs, we found two highly correlated clusters in the rCCA network, of which cluster 1 contained 67 nodes with 275 interactions and cluster 2 contained 21 nodes with 27 interactions (Figure 5(f)). To further identify the fundamental function of the molecular in two clusters, GO-BP analysis was performed using BiNGO plug-in. In cluster 1, relative molecules were greatly involved in the sphingolipid catabolic processes, membrane lipid catabolic process, cerebellar cortex morphogenesis, neurotransmitter metabolic process, and cerebellum morphogenesis. In cluster 2, the top 5 GO-BP terms were mainly response to wounding, response to stress, inflammatory response, acute inflammatory response, and response to stimulus, respectively (Table 3), which indicated their importance in the process of intracerebral hemorrhage. Then, we matched 10 hub molecules with genes/protein in the rCCA network. In this regard, three candidates (Serpina3n, Serping1, and Cfp) were screened out, which presented a tight interaction with other molecules.
3.6. Validation of Hub DEGs/DEPs following Integrative Analysis. A total of 10 hub genes were chosen to further validate their expression in mRNA levels by qRT-qPCR ( Figure 6). The results showed significant difference between the ICH and control groups in the expression of the follow- Given the high correlation between mRNA expression and behavioral assessments and no relevant report in ICH field, we selected four novel candidates (Serpina3n, Serping1, Ctss, and Itgb2) from ten hub molecules and explore their protein expression. Results showed significant upregulation  of Serpina3n (p = 0:006), Itgb2 (p = 0:014), and Ctss (p = 0:001) in brain tissues of ICH compared to the control (Figure 7(a)). The results of correlation analysis showed that protein expression of Serpina3n and Ctss showed a strong correlation with mGarcia scores. Additionally, Itgb2 protein expression exhibited a moderate correlation with results of the mGarcia test. Serpina3n, Itgb2, and Ctss protein expressions were all strongly correlated with beam walking scores. However, neither mGarcia scores nor beam walking scores had a significant correlation with Serping1 protein expression (Figures 7(b) and 7(c)).

Exploration of the Diagnostic and Prognostic Value of
Serpina3n as a Biomarker. In the current study, we enrolled 40 ICH patients and 38 healthy controls. The demographic and clinical information of those two groups was collected and is presented in Supplementary Table S2. The control group matches ICH patients in terms of age and gender. Of interest, plasma concentrations of SERPINA3 were significantly increased in ICH patients compared with the controls (13.3 ng/mL vs. 11.2 ng/mL, p = 0:015) (Figure 8(a) and Supplementary Table S2).

Discussion
In this study, we used integrative analyses of transcriptomics and proteomics to explore molecular mechanisms, underlying the development of ICH-induced brain injury. Using the collagenase model of ICH on mice, we determine the following: (1) immune inflammatory response and coagulation cascade activation were deeply involved in the acute stage of ICH; (2) an ICH-related rCCA network involving two highly connective clusters was presented by integrative multiomics analyses; and (3) Serpina3n, Itgb2, and Ctss were identified as hub molecules at both omics levels, and Serpi-na3n could be a promising biomarker in the diagnosis and prognosis of ICH.
Based on bioinformatics analysis of transcriptomic and proteomics profiles, our results showed that cytokinecytokine receptor interaction, complement, and coagulation 11 Oxidative Medicine and Cellular Longevity cascade pathways were significantly enriched at both mRNA and protein levels in ICH compared with the sham operated animals, which demonstrated an activation of numerous inflammatory cascades occurred in the perihematomal area of brains post-ICH. The cascades include the activation of microglia, astrocytes, and the infiltration of leukocytes [28]. Activated immune cells release a mass of proinflammatory cytokines and chemokines into the brain. These molecules aggravate the initial inflammatory response, which lasts over days or even weeks and is able to affect the subsequent brain repair processes [28]. Besides, the hematoma around the ruptured blood vessel could trigger the coagulation cascade and release a large amount of thrombin, which eventually leads to perihematomal edema and neurological deficit [12]. Strikingly, our results showed that pyroptosis, a newly discovered inflammatory programmed cell death pathway, might be involved in the occurrence and progression of ICH. Previous studies reported that inflammasomemediated pyroptosis was an important event that led to neu-ronal and microglial cell death and consequently to the release of proinflammatory cytokines and neuroinflammation [29][30][31]. The results mentioned above suggested that immune and agglutination-related processes played important roles in ICH pathology.
Given the gap in the depth and breadth between proteome and transcriptome, separated omics might overlook the crosstalk between different molecular entities and could miss biologically relevant information [22]. In this context, we constructed an integrated rCCA network of ICH through a multiomics approach of combing DEG and DEP datasets, which revealed pathological changes involved in ICH from a more comprehensive perspective. We constructed the rCCA network of DEGs and DEPs, which identified important modules related to ICH pathology. In general, we found some well-known pathological changes after ICH, such as wounding response and acute inflammatory response, which had been comprehensively described in previous researches. That fact supports the reliability of our results and the  14 Oxidative Medicine and Cellular Longevity showed that catabolic disorder of sphingomyelin could decrease cell membrane integrity and hinder neural differentiation, mainly due to the reduction of SMPDL3B [32]. Sphingosylphosphorylcholine (SPC), a kind of sphingolipids, could mediate vascular inflammation by increasing the release of MCP-1 and might contribute to vasospasm events after SAH [33]. Besides, cleavage of membrane sphingomyelin by sphingomyelinase generated the second messenger ceramide (Cer), which is involved in an injuryresponsive signaling pathway and act as a trigger of mediated neuronal apoptosis and IL-6 expression after ischemic stroke [34,35]. Thus, we hypothesized that accumulation of sphingolipids might impair neural differentiation and cause excessive Cer production and eventually contribute to neuroinflammation and induce cell apoptosis in ICH.
Integromics-based candidates could be biologically relevant regardless of whether the changes at each single omics level are large or small [22]. In our study, we found three potential candidates (Serpina3n, Itgb2, and Ctss) with high degree scores in the PPI network or high weight in the rCCA network. We also successfully validate that these candidates were significantly upregulated at both mRNA and protein levels following ICH.
Itgb2 (CD18), a cell-surface glycoproteins, belongs the integrin family and acts as a coordinator of extra-and intracellular signaling. It has been reported that itgb2 could interact with intercellular adhesion molecule-1 (ICAM-1) to promote firm adhesion of neutrophils to the endothelium in acute inflammation [36]. Moreover, itgb2 deficiency could also exert a neuroprotective effect via decreasing numbers of infiltrating leukocytes, reducing the release of inflammatory cytokines and chemokines [37][38][39]. We have observed a significant increase of itgb2 in both gene expression and protein abundance in ICH brain tissues, indicating that itgb2 might play a role in regulating signal transduction of immune cells and mediating neuroinflammation in the acute phase of ICH. However, the signal downstream involved in itgb2 after ICH remained obscure and needed to be further explored.
Ctss is a lysosomal protease with highly specific proteolytic activity [40]. Its activation might result in the initiation of cancer, abdominal aortic aneurysm, and atherosclerosis through regulating tumor-promoting immune microenvironment and extracellular matrix degradation [41][42][43][44]. As a proinflammatory mediator, Ctss is able to secrete from activated microglia into the extracellular space and subsequently contribute to the impairment of the BBB, and Ctss inhibitor is reported to ameliorate BBB injury and neurological dysfunction in ischemic stroke [45]. Given the fact that cerebral inflammation and BBB disruption are major pathophysiological events in ICH, it is worth to explore the underlying mechanism of Ctss following hemorrhagic brain injury in future studies. Besides, the inhibition of Ctss to restore BBB function and neurological deficits might be an effective therapeutic intervention after ICH.
Murine Serpina3n, the homologue of human SERPINA3 (a1-antichymotrypsin (ACT)), is a member of the superfamily of serpins, which is highly expressed in the brain, testis, lungs, thymus, and spleen [46,47]. Serpina3n is reported to predominantly targets multiple types of immune cells and inhibits the release of serine protease, such as cathepsin G (CatG), leukocyte elastase (LE), granzyme B (GrB), and matrix metalloproteinase 9 (MMP9), exerting the function in the maintenance of immune balance accordingly [48][49][50]. In our study, we observed a rapid upregulation of Serpina3n mRNA and protein levels in brain tissues of   ICH mice at the acute phase, suggesting the contribution of Serpina3n to participate in regulating early immune response following the insult. Indeed, it has been previously reported that overexpressed Serpina3n could be induced by glia cells in the central nervous [51]. Another research on ischemic stroke has also demonstrated that Serpina3n could be a potent marker in place of conventional biomarker GFAP for its specific identification of activated astrocytes [52]. However, no relationship has been identified between Serpina3n and ICH-induced injury yet. Accordingly, there is a need to dig into the exact mechanism by which Serpi-na3n can produce a benefit or harmful effect on neurological function following ICH onset.
Based on preliminary findings of the studies mentioned above and Serpina3n results of our study, we subsequently measured its expression levels in the plasma of ICH patients to decipher whether the Serpina3n/SERPINA3 could achieve a translation from animal discovery to human research. Of interest, we found that ICH patients exhibited elevated plasma Serpina3 levels compared with the controls, and the increased level of SERPINA3N in plasma of ICH patients was highly associated with poorer outcome (mRS score 4-6). It has been previously reported that SERPINA3 was identified as a potential diagnostic biomarker in other neurological disorders, such as multiple sclerosis and Alzheimer's disease [53,54]. In fact, after ICH onset, peripheral lymphocytes could be soon activated and subsequently infiltrated into the brain parenchymalocal, accompanied by the release of massive cytokines, chemokines, and proteases [55], which might prompt the rapid and persistent secretion of Serpi-na3n to regulate the balance of the immune system. In this context, circulating Serpina3n/SERPINA3 levels might be a promising monitoring marker of prognosis after ICH.
Several limitations of the present study need to be borne in mind. First, although integrative approaches are implemented to gain a deep insight into ICH pathology underlying multiple molecular levels, there is a need for further exploration of methods that combine omics data with ICH phenotypic data. Second, due to the acquirement limitation of clinical samples, we were not able to exhibit cerebrospinal fluid (CSF) SER-PINA3 concentrations during ICH progression. Lastly, the study cohort of ICH patients and the controls was relatively small, a larger-scale enrollment should be taken to further investigate the diagnostic efficacy of candidate SERPINA3.

Conclusions
Using integrative multiomics approaches, we effectively identified hub genes and proteins and revealed an ICHrelated molecule network in brain of mice. Moreover, we demonstrated that Serpina3n/SERPINA3 is a potential biomarker, in which expression is associated with the diagnosis and prognosis for ICH patients.

ICH:
Intracerebral hemorrhage BBB: Blood-brain barrier ECCMU: Ethical Committee of Chongqing Medical University LC-MS/MS: Liquid chromatography-mass spectrometry/ mass spectrometry qRT-PCR: Quantitative RT-PCR H&E: Hematoxylin eosin PBS: Phosphate-buffered saline DEGs: Differentially expressed genes FDR: False discovery rate DEPs: Differentially expressed proteins GO: Gene ontology KEGG: Kyoto Encyclopedia of Genes and Genomes GSEA: Gene Set Enrichment Analysis PPI: Protein-protein interaction network rCCA: Regularized Canonical Correlation Analysis BiNGO: Biological Networks Gene Ontology BCA: Bicinchoninic acid ELISA: Enzyme-linked immunosorbent assay mRS: Modified Rankin Scale PCA: Principal component analysis   Table S1: primer sets for quantitative real-time PCR. Table  S2: comparison of the demographic and clinical characteristics between patients with ICH and health control group. Table S3: comparison of the demographic and clinical characteristics in ICH patients according to 90-day outcome. Figure S1: the correlation analysis between ten hub gene expression and behavior tests. (Supplementary Materials)