All rats consumed the normal rat chow from Specialty Feeds (Glen Forrest, Australia). The phosphate buffer saline (PBS), lipopolysaccharide (LPS) derived from Escherichia coli 026: B6, and dextromethorphan hydrobromide (DXM) were obtained from Sigma Aldrich (St. Louis, United States). The NMR analysis used solvents in deuterated form, including deuterium oxide (D₂O, 99.9%), deuterated methanol (CD₃OD, 99.9%), deuterated sodium hydroxide and potassium dihydrogen phosphate were purchased from Merck (Darmstadt, Germany), and 3-Trimethylsilyl propionic acid (TSP) was purchased from Sigma Aldrich (St. Louis, United States). The LCMS analysis acquired LC-MS grade methanol, acetonitrile, water, and 0.1% formic acid from Merck (Darmstadt, Germany).

The procedures followed the methods described by Ahmad Azam et al. (2020b). In brief, the CN plant was authenticated by a botanist and the voucher specimen (SK2883/15) was kept at the herbarium of the Institute of Bioscience, Universiti Putra Malaysia. The CN leaves collected in December 2015 at Sendayan, Negeri Sembilan, Malaysia (coordinates: 2°38'03.4"N 101°53'20.5"E) was extracted with ionized water by immersing it for 3 days at a ratio of 1 g dried leaves: 50 ml solvent. The process was repeated two times, and the extract was lyophilized (extraction yield: 30%w/w) and kept frozen in −80°C. The relative quantification of chemical markers for the CN aqueous extract via ¹H NMR metabolomics approach was conducted as reported by Ahmad Azam et al. (2020b).

The animal tests were conducted, handled, and performed following the ethical guidelines approved by Universiti Putra Malaysia Animal Ethics Committee (Approval number: UPM/IACUC/AUP189 R070/2015). The animal experiments were carried out following the housing specification of Animal Biosafety Level-2 (ABSL-2) at Laboratory of Animal Resources, Universiti Kebangsaan Malaysia (Bangi, Malaysia) with the room temperature maintained at 24 ± 2°C, a light cycle of both dark and light for 12/12 h, and free access to food and ad libitum water.

All rats were brought in from the in-house service of the same laboratory and were acclimatized for 7 days before the experiment. Only 12 male Sprague Dawley (SD) rats of 13 weeks of age (300 ± 50 g) from the overall 35 rats published in Ahmad Azam et al. (2019) were used in this study. All rats, three in each group, were divided into the following: Group 1 was normal rats injected with phosphate-buffered saline (PBS)+water as the control group (N); Group 2 was LPS-induced rats administered with water (LPS + water); Group 3 was LPS-induced rats treated with 500 mg/kg of BW CNE (LPS+500CN); and Group 4 was LPS-induced rats treated with dextromethorphan 5 mg/kg of BW (LPS + DXM).

The induction of either 10 μl phosphate buffer saline (PBS) to the normal rat groups or lipopolysaccharides (LPS, 1 μg/1 μl) to the neuroinflammed groups have been described in previously published manuscripts (Ahmad Azam et al., 2019; Ahmad Azam et al., 2020a). To summarize, the rats were anesthetized with ketamine-xylazine (K-X); K: 80 mg/kg BW; X: 10 mg/kg of BW through intraperitoneal (i.p) route and they underwent stereotaxic surgery with a single intracerebroventricular (ICV) injection at the location of substantia nigra at the right side of the brain with a consistent rate of 3 μl per minute using a Harvard Apparatus Pump 11 elite infusion syringe via a Hamilton syringe (Holliston, Massachusetts, United States). In total, 1 week after the injection, the rats were administered with each treatment by oral gavage for 14 consecutive days. The CNE was prepared 3 days before each use and preserved at 4°C whereas DXM was freshly prepared before each use.

All rats from the groups were fasted for 14 h and then euthanized with K-X. Then, the terminal process via exsanguination was done by cardiac puncture. The serum was procured from the collected blood sample in a plain vacutainer, and centrifuged for 10 min at 4°C. The collected supernatant was stored at −80°C until analysis.

The ¹H NMR spectroscopic serum data obtained as described in a previously published report was further utilized in this study. This was due to the success of the OPLS–DA model that revealed the potential ameliorative effects of CNE by four selected groups of normal (N), LPS + water, LPS+500CN, and LPS + DXM rats. Three representatives were randomly picked from each of the selected groups in the OPLS–DA data analysis of sera samples (Ahmad Azam et al., 2019).

A total of 12 serum samples (3 samples from each four selected groups of N, LPS + water, LPS+500CN, and LPS + DXM) were used. Each of the 50 µl serum samples were added with 400 µl methanol in a sample tube placed on ice. The samples were vortexed for 2 min, centrifuged at 5000 rpm for 4 min at 4^oC before 400 µl of the supernatant was transferred, and freeze-dried for 8 h. The samples were kept in −80°C for not more than 2 weeks until analyzed.

The freeze-dried samples were reconstituted with 200 µl acetonitrile: water (95:5) prior to usage. The samples were vortexed for 2 min, centrifuged for 5 min at 1,000 rpm, and then syringe-filtered using 0.45 µm, 13 mm, and PTFE filter membranes. The samples were transferred into autosampler vials from which 20 µl of the sample was injected into the Agilent 1,290 Infinity LC system coupled to Agilent 6,550 Accurate-Mass Quadrupole Time-of-Flight (Q-TOF) mass spectrometer (Agilent Technologies, United States). The HILIC column (2.1 × 100 mm, 1.7 µm, Waters, Milford, MA) was eluted using an isocratic solution of MeOH with 0.1% formic acid as the mobile phase at a constant flow rate of 300 µl/min for 15 min including the equilibration time. This method was modified from the procedure for UPLC-TOF MS on HILIC column in a broad-spectrum analysis [Center Specific Procedure (CSP) no: RTI-RCMRC-LCMS-01 ver.00] by the NIH Eastern Regional Comprehensive Metabolomics Resource Core at the Research Triangle Institute International, 2011 United States (www.rti.org/rcmrc, accessed on April 10, 2017).

An electrospray ionization (ESI) source interface was used and set in dual-mode of positive and negative. The following parameters were employed: capillary voltage, 3.5 kV; drying gas flow, 14 L/min; gas temperature: 200^oC; nebulizer pressure, 35 psig.; fragmentor voltage, 175 V; and skimmer voltage, 65 V. The data were collected in a centroid mode, and the mass range was set at m/z 50–1,000 using the extended dynamic range. The trace of metabolites was referred to several main available computational tools for LC-MS data, namely, Agilent Masshunter Qualitative Analysis Ver. B.050.00 (MQA, RRID:SCR_019081) and Agilent Mass Profiler Professional B.05.01 (MPP), and open-source softwares, such as MZmine (RID:SCR_012040) (Pluskal et al., 2010), XCMS (RRID:SCR_015538) (Smith et al., 2016), and Metabolite Automatic Identification Toolkit (MAIT) of R package (RRID:SCR_001905) (Fernández-Albert et al., 2014). Lastly, each of the metabolite's mass representative was matched and annotated via MPP, HMDB MS search (HMDB, RRID:SCR_007712), KEGG (RRID:SCR_012773) compounds, Chemspider (RRID:SCR_006360) and MassBank (RRID:SCR_015535).

All processes were conducted in R wherein the LC-MS spectral baselines were corrected, normalized, aligned, and grouped by isotope. The data were binned into integrated regions with a width of 0.6934 corresponding to the aligned point. The processed data were then used for multivariate pattern recognition analysis.

The overall possible metabolic pathways were constructed using the Metabolic Pathway Analysis (MetPa, RRID:SCR_015539) (http://www.metaboanalyst.ca) (Xia et al., 2015), MetScape 3.1.1 (RRID:SCR_014687) of pathway filter and MetDisease plug-in application (Karnovsky et al., 2012; Duren et al., 2014) of the JAVA software Cytoscape ver. 3.7.1. (RRID:SCR_003032) (Nishida et al., 2014). The possible pathways were also annotated via the KEGG (https://www.genome.jp) and HMDB (http://www.hmdb.ca/metabolites) libraries through metabolites pathway search.

The PCA model for multi-platform integration data in Supplementary Figure S4 yielded an R ² value of 0.699 and Q² value of 0.495. The first PC explained 36.6% of the total variation, mainly separating the normal and LPS-induced neuroinflammed rats. Notable grouping between the groups of LPS + water, LPS + DXM, and LPS + CN500 could be observed via PC2, which explained 16.6% of the total variation. However, the random distribution of LPS + CN500 variables made this group difficult to be clearly discriminated. Thus, compared to the PLS-DA, the supervised method gave a clearer observation for pattern recognition, leading to the identification of the important metabolites responsible for the discrimination between groups.

PLS-DA uses the classification technique of linear combination of original variables to classify label prediction by using zeros and ones (Westerhuis et al., 2010). Figure 1 reveals a good metabolomics model of PLS-DA with R ² and Q² values of 0.743 and 0.731, respectively, for the combination of both analytical methods of LC-MS and NMR. The variations that could be explained by each PC were 34.6% (PC1) and 14.5% (PC2) with a total of 49.1% of explainable variations in this model. As shown in Figure 1A, the score plot of PLS-DA revealed a clear discrimination along PC1 between normal (N) and LPS groups treated with DXM, CN500, and water. Rats treated with CN500 clustered together with LPS + DXM on the right side of PC1 but separated from the LPS + water group by PC2. This grouping implied the possibility of CNE positively affecting the neuroinflammatory condition as it clustered closely to the positive control (DXM).

Figure 1B reveals the complexity of the overall data compilation in the loading scatter plot. The VIP score is a weighted sum of squares of the PLS loadings, considering the amount of explained Y-variation in each dimension. The metabolites scoring VIP values higher than one were the most influential contribution to the model (Eriksson et al., 2006). However, the selection of VIP > 1 resulted in 112 putative metabolites, an amount considered as too many and difficult to be explained. Hence, the VIP indicator was increased to higher than 1.3, which appropriately categorized metabolites that yield the most influential contribution for the model discrimination. Consequently, 35 metabolites were listed as the potential biomarkers for the model (Figure 1C). The metabolites with variable importance of projection (VIP) values of 1.3 and above in the PLS-DA model are visualized in Supplementary Figure S5.

The biomarkers were determined through metabolite alteration, which was interpreted by the VIP values in MVDA. The importance of 35 putative biomarkers from three sets of analytical data is tabulated in Table 1. The application of chemometrics data, which is the mathematical statistics, from the analytical chemical analysis data is vital in a metabolomics approach (Polanski et al., 2017) that involves both the univariate (UVDA) and multivariate data analyzes (MVDA) (Pinto, 2017). When a large number of metabolites is obtained, it is natural to primarily use MVDA. Afterwards, the application of the univariate method becomes necessary to protect against the increased probability of obtaining false positives resulting from the significance testing for metabolites ranging from the tens to the hundreds by correcting multiple tests (Storey and Tibshirani, 2003; Broadhurst and Kell, 2006). The univariate method is used when only one variable is analyzed at a time.

Therefore, the UVDA was conducted to obtain the metabolites alteration expressed in fold change (FC). The summary of the metabolite properties such as mass data, chemical shifts (ppm), chemical formulations, related pathways, and their FC value was listed. In Table 1, the FC displayed only focused on LPS-induced neuroinflammed rats treated with DXM or CN500. The herb (CN500) and drug (DXM) treatments yielded consistent results of amelioration without normalization. Hence, the comparison was focused only on LPS + water treatment to understand the relationship of the chemical interventions on the neuroinflammed rats. Among the 35 VIP > 1.3 metabolites, only nine of them differed in pattern alteration (fold change increase or decrease, bolded; Table 1) between the biomarker of DXM and CN500. The metabolites that increased in CN treatment were identified as choline, allantoin, deoxycholic acid, 5-diphosphomevalonic acid, 1-methylinosine, cholesterol sulfate, unknown 4 (U4), LysoPE(24:0/0:0), and C16 sphinganine. Similar significant metabolite alteration was also observed in the DXM group, suggesting that both DXM and CN treatments produced a similar pattern in ameliorating the neuroinflammed condition of LPS-induced rats. Only allantoin, 1-methylinosine, U4, and C16 sphinganine were significantly different between the treatments of CN500 and DXM.

Among the remaining 26 metabolites, only seven metabolites in either one of the group treatments (LPS + CN500 or LPS + DXM) significantly differed in intensity (FC value in italic in Table 1). The comparison between these two treatments showed a significant difference in the increase of 2-octenoylcarnitine, U1 in the CN treatment group, whereas PI(18:1(9Z)/20:3(8Z,11Z,14Z)), DHA ethyl ester, and arachidonic acid decreased in the DXM group. The significant decrease of lactate could only be observed by CN treatment. For the remaining 19 metabolites, androstenedione, chenodeoxycholic acid disulfate, LPA(18:2(9Z,12Z)/0:0), and lacto-N-triaose have uniformly signified increase in both groups. The significant fold change (FC) decrease for both treatments was observed in isoleucine, ethanol, 7-8-dihydropteroic acid, canrenone, trans-2-dodecenoylcarnitine, leucine, 2,4-dimethyl-tetradecanoic acid, and 2-amino-4-oxo-4-alpha-hydroxy-6-(erythro-1',2',3'-trihydroxypropyl)-5,6,7,8-tetrahydroxypterin. However, 8-hydroxy-deoxyguanosine, l-phenylalanine, palmitic acid methyl ester, and 7'-carboxy-gamma-chromanol equally increased without any significance. The rest of the compounds, including U2, U3, and PE(22:4(7Z,10Z,13Z,16Z)/19:O) decreased in intensity, yet these changes were not significant.

One of the problems mostly encountered in building PLS models in MVDA is the greediness of the technique, leading to the overfitting of the data. This can be controlled by using other appropriate cross-validation strategies (Kjeldahl and Bro 2010; Westerhuis et al., 2010), such as permutation test, CV-ANOVA test, misclassification Fisher probability and double cross-validation. Hence, the present PLS-DA of the multi-platform model was validated by the confirmation of the significant CV-ANOVA value of p = 0.034. The validation was further observed in the misclassification table in which the Fisher probability value obtained was 6.5e-5. The correct classification ratio was calculated to be 100%, meaning that all rats were correctly predicted by the model. Both CV-ANOVA and Fisher probability values of less than 0.05 proved the validity of the model (Eriksson et al., 2006). The CV-ANOVA and misclassification table were tabulated in Supplementary Tables S6, S7, respectively.

The internal cross-validation of the model can be referred back to R ² and Q² values whereby the model was proven to be an established metabolomic model. Unfortunately, the permutation test did not fit into the first regulations of R ² < 0.3 and Q² < 0.05 (Eriksson et al., 2008). The R ² values of the model for each group were 0.56, 0.54, 0.57, and 0.47 while the Q² values were −0.03, −0.22, −0.29, and −0.23 for normal, LPS + CN500, LPS + water, and LPS + DXM, respectively. However, there was another criterion for metabolomic model validation in which all Q² values on the permuted data set to the left would be lower than the Q² value of the actual data set on the right or the regression line (line joining the point of observed Q² to the centroid of a cluster of permuted Q²) had a negative value of Y-intercept. Hence, this criterion for the permutation test proved the validity of the PLS-DA model as all four permuted classes yielded negative Q² intercept values (Mahadevan et al., 2008; Sedghipour and Sadeghi-Bazargani, 2012). The permutation tests are shown in Supplementary Figure S8A–D in the Supplementary Material.

A published study of the same project on ¹H NMR serum has shown seven intervened biomarkers: choline, allantoin, ethanol, isoleucine, acetate, lactate, and leucine (Ahmad Azam et al., 2019). However, the use of UVDA on the quantitative results (Table 1) revealed a complexity due to the inconsistent pattern changes of each biomarker when two analytical tools were combined. Hence, the MVDA of the component analysis was referred to holistically summarize the pattern changes of both analytical observations. The PLS-DA model cluster analysis resulting in the groups of nested trees could also assist in recognizing the variations in the metabolites of LCMS and NMR data.

The results shown by the ingenuity metabolic pathway analysis (MetPA) index by MetaboAnalyst suggested seven most impacted pathways based on the input of 31 metabolites (four unknowns were excluded). The valine, leucine and isoleucine biosynthesis; phenylalanine, tyrosine, and tryptophan biosynthesis; phenylalanine metabolism; arachidonic acid metabolism; glycerophospholipid metabolism; terpenoid backbone biosynthesis; and sphingolipid metabolism were pathways showing an impact value higher than 0.1. This value was required to be categorized as the most relevant pathway. The summary of pathway analysis suggested by MetPA was visualized and tabulated in the Supplementary Figure S9 and Supplementary Table S10, respectively. Only two of the pathways (glycolysis and gluconeogenesis; and valine, leucine, and isoleucine biosynthesis) were in agreement with the previous ¹H NMR sera profiling results (Ahmad Azam et al., 2019). However, all of the other suggested relevant pathways from the previous and recent reports were compatible with the overall systemic map suggested by MetScape ver. 3.1.3. Figure 5 summarizes the shortest route to explain the interactions among the metabolites. Further analysis was conducted with another plugin in Cytoscape called MetDisease. This plugin matched the metabolites to those reported in PubChem Compound record under their medical subject headings (MeSH) (Duren et al., 2014). As a result, Figure 6 is a list of the promising outcomes from MetDisease analysis from which the compatibility of the metabolites with nervous system diseases was recorded to be the highest by 36 nodes of successful linkages. Hence, the model of this study has been demonstrated as an established neuroinflammation model since the metabolites were mostly well-matched with the central nervous system diseases by 36 nodes.

Almost half of the resulted (18/31) biomarkers were categorized in the chemical taxonomy as a superclass of lipid and lipid-like molecules (Human Metabolome Database (HMDB), 2018). They were identified as 2-octenoylcarnitine, androstenedione, deoxycholic acid, chenodeoxycholic acid disulfate, LPA(18:2(9Z,12Z)/0:0), PE(22:4(7Z,10Z,13Z,16Z)/19:O), LysoPE(24:0/0:0), canrenone, arachidonic acid, DHA ethyl ester, palmitic acid methyl ester, trans-2-dodecenoylcarnitine, PI(18:1(9Z)/20:3(8Z,11Z,14Z )), 7'-carboxy-γ-chromanol, 2,4-dimethyl-tetradecanoic acid, cholesterol sulfate, choline, and C16 sphinganine. Most of these metabolites were located at the right side of the loading scatter plot (Figure 1C) which belonged to the LPS-induced rats. Lipids are a crucial member of the cellular membrane and source of stored energy metabolism, and they act as a signaling transductor at the membrane cell (van Meer et al., 2008).

The model pathway analysis results showed a substantial influence among lipid metabolisms via glycerophospholipid, glycosphingolipid and arachidonic acid metabolism alteration. Glycerophospholipids, such as phosphatidylethanolamine (PE) and LysoPE are important components of membrane bilayers. Lysophosphatidic acid (LPA) is extremely important in the biochemical process as a lipid mediator that controls growth, mortality, and differentiation of chemotaxis and ultrastructure of human neutrophils for the innate immune system (hmdb.ca, 2019). The separation by PC2 in Figure 1C shows the location of PE and lysoPE to be higher in LPS + water while LPA was observed among LPS + DXM, LPS + CN500, and normal rats. Consequently, the LPS induction has successfully resulted in lipid rafts on the serum lipid membranes since an amphiphilic structure of LPS allowed the lipids to be rapidly inserted into the cell membranes, forming into the disordered phase of lipid ratios (Martins, 2016). Sphingolipid is also the backbone of neural tissue. The induction of LPS has been reported to interrupt the blood-brain barrier (BBB), increasing the level of the sphingolipid C16 sphinganine in the serum of LPS + water rats (Figure 1C). Furthermore, the roles of squalene/terpenoid and cholesterol biosynthesis in the ameliorative effects of DXM and 500CN treatments were demonstrated by the putative biomarker of 5-diphosphomevalonic acid via the mevalonate pathway. This pathway leads to the synthesis of sterol and isoprenoids, which have been recognized to be essential for cell proliferation for the survival of various types of cancer cells (Ameer et al., 2018). The high abundance of arachidonic acid, which is a polyunsaturated and essential fatty acid, among the LPS + water rats was reported to be important in the inflammation-associated disease as it could act as a mediator to regulate leukocyte chemotaxis, inflammatory cytokines, and immune function (Thakur et al., 1998). Nevertheless, further lipids analysis needs to be confirmed with a proper lipidomic analysis.

Another class of the possible biomarkers was nucleotides and analogs (hydroxyl-deoxyguanosine and 1-methylinosine); organic acid and derivatives (phenylalanine, isoleucine, leucine, acetate, and lactate); organoheterocyclic compounds (allantoin, 7,8-dihydropteroic acid and 2,4-dimethyl-tetradecanoic acid); and organooxygen compounds (ethanol, 5-diphosphomevalonic acid and lacto-N-triose). As the parallel relationships between the metabolites in the pathways involved in the nervous system diseases were deciphered, we suggest looking into the signaling of neurotransmission. The strongest impact obtained from MetPa was observed on the valine, leucine and isoleucine biosynthesis, and phenylalanine, tyrosine, and tryptophan metabolism. The production of essential amino acids, such as phenylalanine and branched-chain amino acid (BCAA) (isoleucine and leucine), could influence the neurotransmitter levels in the brain. This study revealed the elevation of BCAA in the serum of LPS + water group, confirming the inflammation as an activation of the nervous system. Unfortunately, due to the lack of LCMS databases, the identification of some important metabolites in the spectra could not be completely accomplished. Thus, the overall suggested metabolic system was only based on the 31 identified biomarkers (Figure 5).