Characterization of Constituents with Potential Anti-Inflammatory Activity in Chinese Lonicera Species by UHPLC-HRMS Based Metabolite Profiling

This study centered on detecting potentially anti-inflammatory active constituents in ethanolic extracts of Chinese Lonicera species by taking an UHPLC-HRMS-based metabolite profiling approach. Extracts from eight different Lonicera species were subjected to both UHPLC-HRMS analysis and to pharmacological testing in three different cellular inflammation-related assays. Compounds exhibiting high correlations in orthogonal projections to latent structures discriminant analysis (OPLS-DA) of pharmacological and MS data served as potentially activity-related candidates. Of these candidates, 65 were tentatively or unambiguously annotated. 7-Hydroxy-5,3′,4′,5′-tetramethoxyflavone and three bioflavonoids, as well as three C32- and one C34-acetylated polyhydroxy fatty acid, were isolated from Lonicera hypoglauca leaves for the first time, and their structures were fully or partially elucidated. Of the potentially active candidate compounds, 15 were subsequently subjected to pharmacological testing. Their activities could be experimentally verified in part, emphasizing the relevance of Lonicera species as a source of anti-inflammatory active constituents. However, some compounds also impaired the cell viability. Overall, the approach was found useful to narrow down the number of potentially bioactive constituents in the complex extracts investigated. In the future, the application of more refined concepts, such as extract prefractionation combined with bio-chemometrics, may help to further enhance the reliability of candidate selection.

Extensive research has been performed on Lonicera japonica [3,4]. L. japonica extracts and constituents have shown antibacterial, antiviral, anti-inflammatory, immunoregulatory, antitumor and hepatoprotective effects in vitro and in animal experiments [4,5]. The plant has also entered the spotlight in recent research due to its potential activity against SARS-CoV-2 infection. It is among the most commonly used herbs in TCM formulae recommended by Chinese health authorities in COVID-19 prevention programs [6,7]. Extensive phytochemical investigations of the plant have been performed, and more than 200 secondary metabolites have been identified from L. japonica flowers and aerial parts [3,4]. The plant parts mainly used for medicinal purposes are the flowers, followed by the stems [3,4]. The leaves have not commonly been used for medical purposes in the past but have recently attracted an increasing amount of attention, since their chemical profile has been found to be similar to that of L. japonica flowers [8]. The other species listed in the Chinese Pharmacopoeia and many other Lonicera species that occur as native or endemic representatives in China have been much less thoroughly investigated than L. japonica.
Classically, the discovery of bioactive natural products from plant sources involves activity-guided fractionation of active extracts. This is a time-and labor-intensive process with major shortcomings. Frequently, the bioactive constituents cannot be identified due to irreversible binding on chromatographic material or due to degradation during the separation process. The isolation of constituents that occur in low amounts also presents a challenge, since material amounts decrease with each fractionation step. Moreover, in many cases, already-known active constituents are isolated [9][10][11]. Through the development and broader application of advanced analytical instruments, metabolomics-based approaches have emerged as alternative tools for identifying active constituents present in complex herbal extracts [11][12][13]. One frequently used strategy is the metabolite profiling of bioactive natural extracts by means of ultra-high-performance liquid chromatography coupled to high-resolution mass spectrometry (UHPLC-HRMS). Since structural information can be obtained by taking this approach, the active constituents can be dereplicated (i.e., annotated) prior to isolation. In addition, linking metabolite profiling data with bioactivity data can enable the researcher to prioritize potentially bioactive natural products [12]. Chemometric analytical tools are currently being used to identify putative active constituents by correlating phytochemical and bioactivity data. These tools include partial least squares projections to latent structures (PLS), orthogonal PLS (OPLS) and orthogonal projections to latent structures-discriminant analysis (OPLS-DA) [9,14].
Our group previously applied attenuated total reflectance Fourier-transform infrared spectroscopy (ATR-FTIR) successfully in order to classify Lonicera samples displaying different degrees of in vitro anti-inflammatory activity [15]. However, ATR-FTIR cannot be used to trace back the active compounds in extracts of known activity. Thus, the aim of this study was to use UHPLC-HRMS and chemometrics to identify constituents in these extracts that could be associated with in vitro anti-inflammatory activity. We correlated UHPLC-HRMS phytochemical profiles with bioactivity data from the leaf extracts of eight different Lonicera species native to China by means of OPLS-DA. The candidate compounds derived from the OPLS-DA models were annotated on the basis of HRMS data. By taking this approach, we could annotate 65 potentially bioactive compounds and subsequently isolate eight of these candidate compounds. The in vitro anti-inflammatory properties of 15 candidate compounds present in different active Chinese Lonicera species were evaluated.

Lonicera Extracts Displayed High Chemical Diversity
The assignment of the plant samples to eight different Lonicera species by morphological analysis and DNA barcoding has been previously described [15] (see Supplementary  Table S1). The ethanolic leaf extracts were analyzed by UHPLC-HRMS. HESI-negative ion mode detection was chosen based on the results of pilot experiments, which indicated that the best metabolite coverage could be obtained under these conditions (data not shown). UHPLC-HRMS analysis results revealed that the extracts are complex mixtures containing constituents with a wide polarity range and that their composition was highly diverse (Supplementary Figure S1).
In order to obtain an overview of the chemical diversity of the studied extracts, the preprocessed normalized UHPLC-HRMS data were subjected to the principal component analysis (PCA). In the PCA score scatter plot (Figure 1), 14 principal components altogether explained 79.6% of the observed variation. Replicate extracts consistently clustered together, indicating the high reproducibility of the extraction method and UHPLC-HRMS analysis. All extracts derived from L. similis accessions formed a dense cluster, indicating that they had similar chemical compositions, while extracts from L. hypoglauca, L. acuminata, L. confusa and L. japonica formed broad clusters, indicating that these species had highly variable phytochemical profiles. Lonicera macrantha extracts tended to form two distinct clusters. The variation between these two clusters could clearly be seen in the UHPLC-HRMS chromatograms (see Supplementary Figure S2). In the case of L. japonica, the flower bud extract (L35) showed a clear separation from the L. japonica leaf extracts (L23-25 and L35) in the PCA score scatter plot [t1]/[t2] (Figure 1), indicating phytochemical differences between leaves and flower buds; such differences were also revealed by the UHPLC-HRMS chromatograms (Supplementary Figure S3).

Lonicera Extracts Displayed High Chemical Diversity
The assignment of the plant samples to eight different Lonicera species by morphological analysis and DNA barcoding has been previously described [15] (see Supplementary Table S1). The ethanolic leaf extracts were analyzed by UHPLC-HRMS. HESI-negative ion mode detection was chosen based on the results of pilot experiments, which indicated that the best metabolite coverage could be obtained under these conditions (data not shown). UHPLC-HRMS analysis results revealed that the extracts are complex mixtures containing constituents with a wide polarity range and that their composition was highly diverse (Supplementary Figure S1).
In order to obtain an overview of the chemical diversity of the studied extracts, the preprocessed normalized UHPLC-HRMS data were subjected to the principal component analysis (PCA). In the PCA score scatter plot (Figure 1), 14 principal components altogether explained 79.6% of the observed variation. Replicate extracts consistently clustered together, indicating the high reproducibility of the extraction method and UHPLC-HRMS analysis. All extracts derived from L. similis accessions formed a dense cluster, indicating that they had similar chemical compositions, while extracts from L. hypoglauca, L. acuminata, L. confusa and L. japonica formed broad clusters, indicating that these species had highly variable phytochemical profiles. Lonicera macrantha extracts tended to form two distinct clusters. The variation between these two clusters could clearly be seen in the UHPLC-HRMS chromatograms (see Supplementary Figure S2). In the case of L. japonica, the flower bud extract (L35) showed a clear separation from the L. japonica leaf extracts (L23-25 and L35) in the PCA score scatter plot [t1]/[t2] (Figure 1), indicating phytochemical differences between leaves and flower buds; such differences were also revealed by the UHPLC-HRMS chromatograms (Supplementary Figure S3).
Overall, the results of PCA indicated the broad phytochemical variability between and within the Lonicera species under investigation.  Overall, the results of PCA indicated the broad phytochemical variability between and within the Lonicera species under investigation.

Lonicera Extracts Displayed Distinct In Vitro Anti-Inflammatory Activities
To evaluate the pharmacological activity of the extracts, all extracts were assayed in three different cellular in vitro models that are relevant in the context of inflammation. The effects on the expression of IL-8 were screened in LPS-stimulated endothelial HUVECtert cells ( Figure 2A); the influence on NO release was investigated in LPS/IFN-γ-stimulated RAW 264.7 mouse macrophages ( Figure 2B), and modulation of the activity of the nuclear transcription factor NF-κB ( Figure 2C) was examined in HEK293 cells transfected with a luciferase reporter gene. In these cells, the potential impact of the extracts on cell viability was assessed ( Figure 2C).  The extracts showed some parallels regarding their influence on the release of NO and IL-8. The most pronounced activity was found for L. hypoglauca accessions L5 and L14, whereas L. macrantha, L. reticulata, L. acuminata and L. bournei samples showed no or only moderate activity in both assays. Furthermore, the activity of L. confusa and L. similis accessions was generally low with the exception of L1 and L18, which showed pronounced effects on the IL-8 levels and NO production, respectively. Lonicera japonica extracts showed generally higher activity in the IL-8 assay. Many of the extracts showed highly pronounced inhibitory activity on NF-κB activation. Cell viability was not or only moderately affected at screening concentrations of 50 µg/mL; only in the case of L14, a reduction to about 60% of the viability of control samples was observed.
Black dashed lines indicate the threshold below which extracts were considered as highly active in the respective assay.

Candidate Compounds from Different Compound Classes Were Annotated from OPLS-DA Models Correlating UHPLC-HRMS and Activity Data
In order to trace back the constituents that most likely correlate with anti-inflammatory activities, activity data were correlated with processed UHPLC-HRMS data by means of OPLS-DA, a supervised multivariate data analysis (MVDA) method. In OPLS-DA, a regression model is calculated between the multivariate data (i.e., the UHPLC-HRMS dataset) and a response variable (i.e., the class information). The class information is placed in the first predictive component, while the other components describe the variation orthogonal to the predictive component [16].
In the score scatter plots of all tree OPLS-DA models, a clear discrimination between active and inactive samples was observed (Supplementary Figures S4-S6), and the quality of the models can be considered as good (Table 1). Table 1. OPLS-DA model parameters and p-values of CV ANOVA of the OPLS-DA models generated from CD-processed LC-HRMS data and pharmacological assay data classified according to the parameters given in Supplementary Table S2 (extracts classified as moderately active were excluded from the models). A: number of principal components; N: number of observations; R2X (cum): cumulative sum of squares of the entire X explained by all extracted components. R2Y (cum): cumulative sum of squares of all the y-variables explained by the extracted components. Q2 (cum): cumulative Y variation predicted by the X model for the extracted components, according to cross-validation. From the OPLS-DA models, S-plots were generated to visualize the covariance and correlation between the scores for the predicted OPLS-DA component and the spectral variables, as well as to derive variables most likely correlated with class separation. High covariance and high correlation to the score on the predictive component indicate the most important variables [17]. Thus, the variables related to active extracts were sorted in a descending order according to covariance, and the 30 most reliable variables plus the interjacent, less reliable ones in the S-plots of the three models were selected as candidates. Since some of these were correlated with more than one activity, this resulted in the identification of 78 candidate features. The S-plots are shown in Supplementary Figures S4-S6 (candidate features marked in red). The complete candidate list is provided in Supplementary Table S4.

OPLS-DA Model Parameters
Four of the features were found to be dimers or adducts of other candidate compounds, and nine features could not be annotated. Of the remaining 65 features, seven could be unambiguously assigned, and 58 were initially tentatively annotated on the basis of their HRMS data; subsequently, eight of these tentatively annotated compounds were isolated from a L. hypoglauca ethanolic extract, and their structures were partially or fully elucidated by NMR spectroscopy (described in detail in Section 2.4). The mass spectrometric data of the annotated candidates are found in Table 2 and in the Supplementary Table S4.
The main compound classes annotated among the candidate compounds ranged from rather polar compounds, such as hydroxycinnamic acid derivatives, flavonoid glycosides and iridoids, to medium-polar compounds, such as flavonoid aglycones, bioflavonoids and fatty acids, and on to rather nonpolar lipids. Table 2. List of annotated OPLS-DA candidates; compounds written in italics were isolated from L. hypoglauca, and their structures were elucidated fully or partially by NMR spectroscopy (Section 2.4); priority: deduced from S-plots of respective OPLS-DA models; low number represents high priority; numbers written in gray represent candidates with low levels of reliability (jackknife-based confidence interval, including 0); annotation: phospholipid and sphingolipid nomenclature, see Reference [18], fragmentation patterns and identification sources and Supplementary Table S4. ID level ( [19,20]): 0 , isolated pure compound, unambiguous 3D structure; 1 , confident 2D structure, identified by reference standard match (retention time and MS/MS); 2 , tentative annotation by comparison of structural formula and MS/MS fragmentation pattern with data from databases or literature. 3 , no data available in literature or databases; compound tentatively assigned on the basis of the MS/MS fragmentation pattern (theoretical interpretation and/or comparison to related compounds). Concerning the hydroxycinnamic acid derivatives, dicaffeoylquinic acid isomers (11 and 12), coumaroylquinic acid isomers (2 and 3), dicoumaroylquinic acid isomers (15 and 17) and caffeic acid (1) could be annotated. Caffeoylquinic acid derivatives are widespread in the flower buds, stems and leaves of various Lonicera species [21][22][23], and also, several p-coumaroylquinic acid isomers have been detected [4]. This is the first time that di-O-pcoumaroylquinic acid derivatives are reported from this genus.

Nr
From the compound class of flavonoids, ten flavonoid-O-glycosides were annotated (5, 6, 7, 8, 9, 10, 13, 14, 16 and 20). Four of these were monoglycosides (7, 9, 13 and 14); four were diglycosides (5, 6, 8 and 10) and two were substituted with one sugar and one a caffeoyl (16) or coumaroyl (20) moiety. Compound 7 was unambiguously identified as luteolin-7-glucoside. In the MS/MS spectra of 5, 6, 8, 10, 13, 16 and 20, m/z 285.04 was detected as aglycone. However, since no aglycone breakdown fragments were detectable, it was impossible to determine whether the aglycone was luteolin or kaempferol on the basis of the MS/MS spectra generated with QExactive MS. Therefore, extracts containing the respective compounds were additionally analyzed on an LTQ-XL instrument equipped with a linear ion trap analyzer that provides MS n capabilities. Comparison of MS 3 spectra with reference spectra of luteolin and kaempferol allowed the annotation of the aglycones. Comparison with the literature led to a tentative annotation of this compound as kaempferol-3-hexoside-7-desoxyhexoside. Compounds 5, 6 and 10 were assigned as the dihexoside, hexoside-pentoside and hexoside-desoxyhexoside of luteolin, respectively; the luteolin hexoside desoxyhexoside lonicerin (luteolin-7-neohersperidoside) has been previously identified as possible marker for L. japonica [24]. For compounds 16 and 20, MS/MS fragmentation patterns clearly indicated the presence of caffeoyl (16) and coumaroyl (20) moieties. The aglycone was annotated as luteolin in both cases. Thus, compound 16 was tentatively identified as a luteolin caffeoyl hexoside, and compound 20, as luteolin coumaroyl hexoside. This is the first reported detection of these two compounds in Lonicera species.
In addition to the flavonoid glycosides, the flavone aglycones luteolin (18) and apigenin (21) were unambiguously identified. Compound 19 was annotated as the flavonol aglycone isorhamnetin. Compound 23 was tentatively identified as methylated flavone, but due to the high level of spectral similarity, it was impossible to discriminate between diosmetin or chrysoeriol on the basis of the MS/MS fragmentation pattern. The MS/MS spectrum of compound 22 indicated the presence of a polymethoxyflavone. This compound was subsequently isolated, and its structure identified by NMR spectroscopy (see Section 2.4).
Furthermore, a series of candidates were annotated as biflavonoids: amentoflavone (32) and podocarpusflavone A (33) were identified by comparison with authentic reference compounds. The fragmentation patterns of compounds 34 and 35 were highly similar and clearly indicated C-O-linked biapigenins, such as the 4 ,6 -O-biapigenin hinokiflavone [28]. A similar fragmentation can be expected for the 4 ,3 -O-biapigenin ochnaflavone, which has previously been isolated from L. japonica [29]. Compounds 26, 27 and 29 were tentatively identified as biapigenin, mono-and dimethylbiapigenin and were subsequently isolated and structurally identified by NMR spectroscopy (see Section 2.4).
Concerning nitrogen-containing compounds, we were able to annotate the alkaloid aurantiamide acetate (31) and to unambiguously identify the chlorophyll degradation product phaeophorbide A (52). Both compounds have not been identified in Lonicera species until now.
Finally, numerous OPLS-DA candidates were annotated as fatty acids and assigned to different types of lipids. Two trihydroxy octadecadienoic acid isomers (24 and 25) and an oxo-octadecadienoic acid isomer (36) were annotated, next to a C 24 monohydroxy fatty acid (58). Furthermore, a series of 12 thus far undescribed long-chain fatty acid derivatives were annotated (41-44, 46, 47, 49-51, 53, 54 and 56). Four of these (43, 44, 46 and 54) were subsequently isolated, and their structures were partially characterized (see detailed description in Section 2.4). Molecular formulas and MS/MS fragmentation patterns indicated that the 12 compounds possessed chain lengths of C 32 (41-44 and 46) or C 34 , two to four hydroxyl groups, as well as one to three acetyl moieties. To our knowledge, these compounds have not previously been described in literature.
With respect to lipids, candidates with phospholipid, glycolipid, cerebroside and ceramide structures were tentatively identified. Compound 39 was annotated as a glycolipid with an octadecatrienoic acid and a dihexoside moiety bound to glycerol, like gingerglycolipid A. Compound 48 was tentatively identified as glycerol bearing a palmitic acid moiety as well as a sulfur-containing sugar moiety, such as palmitoyl-sulfoquinovosylglycerol. Compounds 40, 45 and 55 were annotated as phospholipids, namely as lysophosphatidylcholine (LPC) bearing a C 18 fatty acid with three double bonds (40), a lysophosphatidylglycerol (LPG) bearing a C 16 -monounsaturated fatty acid (45) and an LPG with a saturated C 16 fatty acid (55). The structure of the LPC was additionally verified by the software Lipid Data Analyzer [33] (see Supplementary Figure S7). None of these glycolipids or phospholipids have been previously reported in Lonicera species.
Finally, seven candidate compounds were assigned to ceramides that contained either a hydroxysphingosine (59-63 and 65) or hydroxysphinganine isomer (64) as a long-chain base and saturated C 22 -, C 23 -or C 24 -mono-or dihydroxy fatty acids [34]. Their structures could be additionally verified by Lipid Data Analyzer [35] (see Supplementary Figures S7-S14). Finally, 57 was annotated as a cerebroside composed of a sphingadienine isomer, a saturated C 16 -dihydroxy fatty acid and a hexose moiety, such as soyacerebroside isomers. Soyacerebroside II, as well as several ceramides have been previously isolated from L. japonica [36,37]. The ceramide lonijaposide A3 has the same molecular formula as compound 61 (C 42 H 83 O 6 N. However, the MS/MS fragmentation pattern of compound 61 is more similar to that of a ceramide composed of a hydroxysphingosine isomer and a dihydroxytetracosanoic acid [34] than the structure of lonijaposide A3, consisting of a C 26 long-chain base and an α-hydroxy hexadecanoic acid [37].

Eight OPLS-DA Candidates Were Isolated from L. hypoglauca Leaves
To identify the structures of some of the candidate compounds that could not be unambiguously identified by UHPLC-HRMS analysis, a large-scale extract was prepared from L. hypoglauca leaves, since samples of this species had been found to contain several candidates at high levels (i.e., normalized peak area > average + 2 S.D. in OPLS-DA Xvar plot, Supplementary  Figure S15) allowed us to confirm that ring B is substituted with methoxy groups at positions 3 , 4 and 5 . A further methoxy group was assigned to position 5. On the one hand, no proton resonance was detected for position 5 and, on the other hand, carbon shifts were significantly reduced at positions C-2 and C-4 but significantly enhanced at position C-3. Comparing the 13 C shifts at these positions with the literature data on flavones with similar substitution pattern [38] allowed us to confirm the substitution of position 5 as a methoxy group. Consequently, 22 was identified as 7-hydroxy-5,3 ,4 ,5 -tetramethoxyflavone (Figure 3). Only one single patent describes the synthesis of this compound, but without disclosing spectroscopic data [39]. Accordingly, this is the first study reporting the isolation of 22 from a natural source and describing its spectroscopic assignment. Flavones with more than two methoxy groups in the molecule have not previously been described from L. hypoglauca; however, various highly methoxylated flavones have been identified in L. japonica flower buds [40].    Table S5). Thus, we could assign the A-ring-CH group with δH 7.04 ppm and δC 100.7 ppm to position C-6, and 26 was identified as the known 8,8"-biapigenin cupressuflavone (Figure 3). Cupressuflavone has been identified in several Lonicera species [41,42], but this represents a new report for L. hypoglauca.
The 1 H spectrum of 27 displayed a higher number of resonances than that of 26. Coupling patterns indicated a different substitution in one of the B-rings, while the other parts of the molecule displayed a pseudosymmetric arrangement. Based on HMBC correlations, the position of the additional OCH 3 group was identified as C-3 , allowing us to assign compound 27 to 3 -methoxycupressuflavone ( Figure 3). In the case of 29, the low number of proton resonances supported the symmetric substitution pattern indicated by the MS/MS data. HMBC correlations enabled the assignment of two methoxy groups to positions 3 and 3 in the two apigenin subunits, allowing us to identify the compound as 3 ,3 -dimethoxycupressuflavone ( Figure 3). Compounds 27 and 29 have only been described once before by Jang et al. [43] reported the isolation of these compounds from Zabelia tyaihyoni (Nakai) Hisauti & H. Hara (Caprifoliaceae). Consequently, this is the first report of the isolation of these compounds from a Lonicera species.  Figure S16). The HSQC spectrum (Supplementary Figure S17) determined the presence of five hydroxymethine groups. Two of these were found to bear acetyl moieties, the methyl group of which appeared as sharp singlet at a shift of 2.02 ppm in the proton spectrum. HMBC correlations indicated the presence of three carbonyl functional groups; two of these were assigned to the acetyl moieties (δc = 172.9 ppm) and the third to the carboxyl moiety at position 1 (δc = 176.9 ppm) (Supplementary Figure S18 and Table 4). Starting from the carboxyl group at position 1, a CH 2 group was assigned to position C-2 (δc = 43.7 ppm), and a CHOH group to position C-3 (δc = 69.6 ppm); via their HMBC correlations, the CH 2 groups at position C-4 (δc = 38.4 ppm) and C-5 (δc = 22.4 ppm) could be assigned as well. The terminal CH 3 group (δc = 14.5 ppm) displayed typical HMBC correlations for alkyl chains with a length of at least five carbon atoms. The remaining two CHOH groups showed only two HMBC correlations (δc = 38.1 ppm and 22.7 ppm). This indicates that the C atoms that couple to these two groups via two or three bonds display the same or a very similar chemical environment. For the two CHOCOCH 3 groups, three HMBC correlations (δc = 35.3 ppm, 26.4 ppm and 22.2 ppm) were detectable, indicating a slightly different chemical environment for the two groups (Supplementary Figure S19). The nonoverlapping HMBC correlations for these four groups indicate that at least four CH 2 groups are present; HSQC-TOCSY experiments did not provide useful data; thus, it was impossible to assign the exact position of the two CHOCOCH 3 groups and the two remaining CHOH groups in the molecule by means of NMR spectroscopy. GC-MS experiments performed in a similar way as described by Seipold et al. [44] did not allow the unambiguous assignment of these functional groups either (data not shown). In summary, compound 46 could be tentatively assigned as a saturated C 32 fatty acid hydroxylated at position 3. There are two more OH groups and two OCOCH 3 groups between position 3 and the terminal alkyl chain, which consists of at least five CH 2 groups. The exact positions of these functional groups could not be determined. 1 H NMR spectra of compounds 43, 44 and 54 were highly similar to those of compound 46, i.e., similar shifts were seen for the protons at positions 2, 3 and 32, as well as for the overlapping protons of the acetate groups, the CHOH groups and the acetylated CHOH groups (Supplementary Figures S20-S22). For 43 and 44, LC-HRMS data indicated the same molecular formula as for 46. Likewise, the MS/MS fragmentation patterns were highly similar. Therefore, 43 and 44 were assumed to be isomers of 46, containing a saturated C 32 alkyl chain as well. The molecular formula for compound 54 was calculated as C 38 H 72 O 9 , and, as for 43, 44 and 46, the MS/MS fragmentation pattern indicated the loss of three H 2 O and two CH 3 COOH molecules. Therefore, 54 obviously contains a saturated C 34 chain with the same functional groups as 43, 44 and 46.

Several Candidate Compounds Displayed In Vitro Anti-Inflammatory Activity
In order to assess the potential activities anticipated from the OPLS-DA models, 15 candidates that were available as pure compounds were tested in the same cellular in vitro assays which had been used for extract screening. The test compound set comprised flavonoid glycosides and aglycones, biflavonoids, an iridoid glycoside, fatty acids and the chlorophyll degradation product pheophorbide A. Compounds were tested at a screening concentration of 30 µM; if inhibitory activity was identified, the IC 50 values were determined (Table 5). Experimental results were compared with the priorities obtained from the respective OPLS-DA models ( Table 6). Table 5. Inhibitory activities of selected OPLS-DA hit compounds in cellular inflammation assays. # Compound numbers and identification; see Table 2. Effect on NF-κB activation was assessed in HEK/NFκB-luc cells stimulated with TNF-α; positive control: parthenolide (5 µM); results are given as means ± SD of 4 replicates (each in quadruplicate); IL-8 expression was assessed in HUVECtert cells stimulated with LPS; positive control: Bay 11-7082 (7.5 µM); results are given as means ± SD of 5 replicates; b in compounds significantly impairing cell viability, IL8 expression was determined one sample prepared by pooling the 5 replicates.; + compound significantly impaired cell viability at concentration of 30 µM; Effect on NO production was assessed in LPS/IFNγ-stimulated RAW macrophages.
Extract concentration: 50 µg/mL; positive control: L-NMMA (100 µM); results are given as the means ± SD of 3 experiments performed in duplicates.  Table 6. Comparison of experimentally determined in vitro anti-inflammatory activities of selected OPLS-DA candidate compounds with priorities deduced from OPLS-DA model; green fields indicate correctly predicted active compounds, and yellow fields indicate correctly predicted active compounds with negative impact on cell viability; Compound numbers, identification and priorities from OPLS-DA models; see Table 1. Numbers written in gray indicate low reliability (jackknife-based confidence interval including 0); activities determined in cellular inflammation assays; see Table 5. i = inactive; a = active; a/ct = active, but cytotoxic. In the OPLS-DA models, the flavone aglycone luteolin (18) was correlated with inhibitory activity in the NF-κB and IL-8 assay, and apigenin (21) was an NF-κB inhibitory candidate. The newly isolated 7-hydroxy-5,3 ,4 ,5 -tetramethoxyflavone (22) was indicated as potential IL-8 and NO inhibitor. Luteolin and apigenin indeed potently inhibited NF-κB activation (IC 50 6.84 and 4.11 µM) and moderately inhibited IL-8 production (IC 50 27.5 and 18.9 µM). Luteolin was only moderately active in the NO assay (IC 50 31.75 µM). 7-Hydroxy-5,3 ,4 ,5 -tetramethoxyflavone was inactive in all three models. The flavonol-O-glycoside luteolin-7-glucoside (7), which serves as a quality marker for Lonicerae japonicae flos according to the Chinese Pharmacopoeia [1], was also tested, although it was a candidate that was assigned low priority and reliability. The compound displayed no activity in any of the three assays. Five candidates with biflavonoid structure were subjected to pharmacological testing. While cupressuflavone (26) and 3 ,3"-dimethoxycupressuflavone (29) were inactive in all three assays, 3 -methoxycupressuflavone (27), amentoflavone (32) and podocarpusflavone A (33) displayed moderate to high activity in the NF-κB activation assay (IC 50 19.6, 6.45 and 3.44 µM for 23, 28 and 29, respectively). Compound 33 also displayed a weak inhibitory effect on IL-8 expression (IC 50 47.88 µM). The predicted high priorities concerning inhibitory activity of compounds 26, 27 and 29 on NO formation could not be verified by the experiments.

Nr
An iridoid was also among the candidate compounds, namely, secoxyloganin (4). However, its high priority on IL-8 expression could not be experimentally verified in this study. The chlorophyll breakdown product pheophorbide A (52) was correlated with inhibition of NF-κB activation, which could be experimentally verified (IC 50 3.04 µM). It also inhibited IL-8 production; however, this might be due to impaired HUVECtert cell viability.
Finally, several long-chain polyhydroxylated fatty acids represented high-ranked candidate compounds in particular with regard to their inhibition of IL-8 expression and NO formation. Four of these, compounds 43, 44, 46 and 54, which had been isolated from L. hypoglauca, were subjected to pharmacological testing. Compounds 46 and 54 indeed displayed inhibitory activity: compound 46 in all three models and compound 54 in the IL-8 and NO assays. However, all four fatty acids also impaired the viability of the cells used in the respective assays (Tables 5 and 6).
Taken together, the highest number of candidate compounds could be verified in the NF-κB transactivation assay. The most active compounds were flavonoid aglycones and biflavonoids. Pheophorbide A (52) and the newly identified long-chain polyhydroxy fatty acids (46,54) showed remarkable inhibitory activity both on NF-κB activation and NO formation, as well as IL-8 expression, respectively, but they also impaired the cell viability.

Discussion
In TCM, flowers and stems of L. japonica are most commonly used for anti-inflammatory therapies. In this study, we investigated leaf extracts from L. japonica, as well as from other Chinese Lonicera species that are less commonly used in TCM or not used at all. Besides L. japonica extracts, other less thoroughly investigated species also displayed considerable in vitro anti-inflammatory activity, such as L. hypoglauca and L. acuminata. Consequently, the leaves of these species may also be interesting sources for phytomedicines.
Annotation of the OPLS-DA candidate compounds on the basis of UHPLC-HRMS data and the structural elucidation of candidate compounds from L. hypoglauca after their isolation led to the tentative or complete identification of 65 compounds in total. Four of these were novel compounds (43, 44, 46 and 54), one was isolated for the first time from a natural source (22), and 28 compounds were identified in the genus Lonicera for the first time. Some of these may be new natural products.
For several annotated candidates, anti-inflammatory activity had been previously reported.
Concerning the flavonoids, aglycones usually display higher activities in inflammationrelated cell-based assays than their respective glycosides. In contrast, in animal models after oral treatment, the activity of glycosides has normally been shown to be similar or even higher than that of the respective aglycones [45]. The flavones luteolin and apigenin have shown potent in vitro and in vivo anti-inflammatory activity [46][47][48]. Luteolin-7-glucoside (cynaroside) has shown anti-inflammatory activity in septic [49] psoriatic animals [50]. These data underline the potentially important role of flavonoids and, in particular, of luteolin and its glycosides regarding the anti-inflammatory activity of Lonicera species. In our study, luteolin and apigenin displayed high anti-inflammatory activities, while luteolin-7-glucoside displayed no activity. In addition to these well-known flavonoids, 7-hydroxy-5,3 ,4 ,5 -tetramethoxyflavone could be isolated as a new natural product from L. hypoglauca. For this compound, only a weak inhibition of NF-κB activation was observed.
Biflavonoids such as the biapigenins ochnaflavone, podocarpusflavone A and amentoflavone have exhibited anti-inflammatory effects in vitro [51][52][53], while amentoflavone and cupressuflavone have shown these effects in animal models as well [54,55]. In this study, while cupressuflavone was inactive in all three assays, amentoflavone and podocarpusflavone A potently inhibited NF-κB activation. The rare biflavonoid 3 -methoxycupressuflavone moderately inhibited NF-κB activation, and podocarpusflavone A moderately inhibited IL-8 production. These findings indicate that biflavonoids are also relevant for the anti-inflammatory activity of Lonicera species.
Apart from these compound classes, which are well-known for Lonicera species, other activity-related candidates originated from compound classes that have been less commonly reported in Lonicera species or have not been previously described in this genus at all.
The chlorophyll degradation product pheophorbide A showed pronounced inhibition of NF-κB transactivation, a finding that is in accordance with literature data [56]. However, in contrast to literature reports [56], the compound did not inhibit NO production in RAW264.7 macrophages.
Moreover, we could isolate and partially identify the structures of four novel saturated C 32 -and C 34 -fatty acids containing different numbers of hydroxyl and acetyl moieties. Similar compounds, but with shorter chain lengths and lower numbers of hydroxy and acetyl substituents, have been identified in the floral oil of Malphigia coccigera [44]. There, the location of the substituents in the alkyl chains could be determined by interpreting GC-MS fragmentation patterns after methylation and trimethylsilylation. However, the same technique applied to our compounds that contained longer chains and more substituents led to ambiguous GC-MS fragments, which prevented the unequivocal detection of their substituent positions. Two of these displayed remarkable inhibitory activities in the applied cellular test models, but negatively impacted cell viability.
Further candidate compounds were identified from the compound classes of hydroxcinnamic acid derivatives, triterpenes, alkaloids and lipids. For many of these, the in vitro and in vivo anti-inflammatory effects have been previously described, but the absence of pure compounds prevented experimental assessments in this study. Thus, their role in the anti-inflammatory effects of the investigated extracts remains speculative.
Many of the annotated lipids reported in this study were detected in Lonicera species for the first time. They can be roughly stratified in three different categories: glycolipids, phospholipids and sphingolipids (glucocerebrosides and ceramides). One candidate compound was tentatively assigned to an acylglycerol bearing a sulfoquinovose. While diacylglyceryl sulfoquinovosides are associated with photosynthetic membranes of many photosynthetic organisms [57], palmitoyl glycerylsulfoquinovosides have only occasionally been isolated from natural sources, and their pharmacological activities have not been studied thoroughly [58].
Sphingolipids occur in essentially all eukaryotes and in certain prokaryotes. The glucocerebroside soyacerebroside II has been previously isolated from L. japonica [36]. An isomeric mixture of soyacerebrosides I and II has been shown to suppress the LPS-induced IL-18 release in human peripheral blood mononuclear cells [59].
Ceramides have been previously isolated from L. japonica [37]. However, their structures are not in concordance with those annotated in this study. Plant-derived ceramides are considered as interesting nutraceuticals [34], but their anti-inflammatory effects have not been systematically studied yet. Taking into consideration the fact that structural characteristics such as the number of hydroxyl groups determine whether a ceramide exhibits pro-or anti-inflammatory effects [60], the Lonicera ceramides annotated for the first time in this study may deserve a more thorough, systematic investigation to identify their structures and anti-inflammatory activity in the future.
In a subset of 15 OPLS-DA candidates that were available as pure compounds, the potential anti-inflammatory activities deduced from the models could be experimentally verified for several, but not for all, compounds. Some of the compounds exhibited not only anti-inflammatory activity but also impaired cell viability when applied as a pure substance. This side effect may be due to the fact that OPLS-DA-derived candidate compounds are typically among the more highly abundant bioactive compounds as a result of the high covariance criterion. In the complex, crude extracts used herein, obviously only the most powerful compounds are ranked highest; these, in turn, may also be toxic, while more easily tolerated, active trace compounds may remain unidentified [61].
To increase the identification rate of truly active compounds among MVDA-derived candidates and to unravel potential cytotoxic effects, it may be helpful to reduce the complexity of extracts by working with fractions in addition to or instead of crude extracts. For example, Kellogg et al. [44] suggested combining bioassay-guided fractionation with untargeted metabolite profiling by applying bio-chemometric methods. This approach enables the researcher to dereplicate the known bioactive compounds early on in the isolation process and to integrate several fractionation steps and the respective bioactivity data in one bio-chemometric analysis. Our results indicate that the successive addition of fractionation steps improved the quality of the biochemometric analysis [9]. Demarque et al. applied a partial least squares projection to latent structures and molecular networks to predict the presence of larvicidal compounds in pre-fractionated extracts from Annona crassiflora, then performed subsequent activity-guided fractionation to confirm this activity. According to these authors, crude extract prefractionation was a crucial step that enabled them to clean up the samples and to create unbalanced chemical profiles, thus facilitating the subsequent chemometric analysis of the samples [13].

Plant Material, Authentication and Extraction
Thirty-four (34) samples of the aerial parts of different Lonicera species were collected in the Guangxi Province (China) (samples 1-34). One sample of L. japonica flower buds was purchased at a TCM herb market in Nanning, China (sample 35), and one sample of L. japonica aerial parts was collected at the Botanical Garden in Graz (sample 36). Samples were authenticated by morphological analysis and by DNA barcoding as previously described [15]. Voucher specimens of all samples are deposited in the herbarium of the Guangxi Botanical Garden of Medicinal Plants (Nanning, China). A detailed sample list can be found in Supplementary Table S1.
The dried leaves were ground, extracted with 96% ethanol by accelerated solvent extraction (ASE) as previously described [15] and dried by vacuum evaporation. Extracts were prepared in duplicates. For sample 34, only a single extract was prepared due to the low amount of plant material. Extract yields are provided in Supplementary Table S1. Samples were frozen at −20 • C.
Raw data files were processed with Compound Discoverer 3.1 (Thermo Scientific) (CD), using the following workflow: retention time window: 0-60 min; total intensity threshold: 12,000,000. Alignment was performed with adaptive curve model. Maximum RT shift was 2 min, and maximum mass tolerance 5 ppm. For detecting and grouping unknown compounds, S/N threshold: 3; minimum intensity threshold: 12,000,000; RT tolerance: 1 min; S/N threshold for gap filling: 20. Processing resulted in a data matrix consisting of retention time and intensity of every feature in every sample that was exported for further data treatment and subsequent multivariate data analysis.

Cellular In Vitro Assays
Assays were performed as previously described [15,62]. Extracts were dissolved in dimethyl sulfoxide (DMSO) and further diluted in cell culture medium to obtain a final concentration of 50 µg/mL; DMSO at the respective final concentration (<0.1%) was used as vehicle control in each assay. OPLS-DA candidate compounds were also dissolved in DMSO to obtain a final screening concentration of 30 µM. For compounds active at this concentration, IC 50 values were determined.
Transactivation activity of a NF-κB-driven luciferase reporter gene was determined in TNF-α (2 ng/mL)-stimulated HEK293/NF-κB-luc cells (Panomics, RC0014), using 5 µM parthenolide (Sigma-Aldrich, St. Louis, MO, USA) as positive control. Differences in cell viability were detected by comparing the fluorescence of vehicle-treated cells and cells treated with sample solutions upon staining with CellTracker Green CMFDA (C2925; Invitrogen), a fluorescent probe that is retained inside living cells [63]. Screening samples were tested in three independent experiments, each performed in quadruplicate.
Release of nitric oxide (NO) from RAW 264.7 macrophages stimulated with LPS (0.5 µg/mL; E. coli 055:B5; Sigma-Aldrich) and interferon-γ (IFNγ; 50 U/mL, mouse recombinant E. coli, Roche Diagnostics) were tested as previously described [64], using the non-specific NOS-inhibitor L-NMMA at 100 µM as the positive control. After removing 50 µL of the supernatants to react with Griess reagent (Sigma Aldrich), cells were incubated with XTT reagent (XTT proliferation kit II, Sigma Aldrich). NO production (Griess reaction) and metabolic activity of the cells (XTT formazan product) were photometrically measured. Tests were performed in three independent experiments, each in duplicate.
Effects on IL-8 expression were tested in hTERT-immortalized HUVECtert cells stimulated with lipopolysaccharide (LPS, 100 ng/mL, Sigma-Aldrich), with BAY 11-7082 (5 µM) as positive control [15,62]. Activity screening was performed in sextuplicate. Impact of test samples on cell viability was determined after 6 h of stimulation by XTT. Incubation medium was removed, and 40 µL serum-free medium, containing XTT (Thermo Fisher Scientific, final concentration 200 µg/mL) and phenazine methyl sulfate (TCI Europe N.V., final concentration: 25 µM) were added to each well. After incubation for 2 h, absorbance was measured at 450 nm. Samples decreasing cell viability by more than 30% were considered to negatively influence cell viability.
To determine the IC 50 values, pure compounds showing significant inhibitory activity at the screening concentration were tested at 5-10 concentrations, each with 4 replicates. IC 50 values were calculated by applying the four-parameter logistic curve algorithm in Sigma Plot 13.0 (Systat Software Inc., Palo Alto, CA, USA) or GraphPad Prism 4.03 (Graph-Pad Software, Inc., San Diego, CA, USA).

Data Treatment and Multivariate Data Analysis (MVDA)
In order to compensate shifts in the sensitivity of the MS detector that possibly occurred in the course of the UHPLC-HRMS sequence, preprocessed UHPLC-HRMS data were normalized to the peak area of the external QC sample that had been injected after every series of six runs. Preprocessed, normalized UHPLC-HRMS data were imported to SIMCA 16.0.1 (Sartorius Stedim Data Analytics AB, Umeå, Sweden), pareto-scaled and logtransformed. Prior to OPLS-DA, bioactivity data were classified (active, medium-active and inactive; classification details are given in Supplementary Table S2). Moderately active samples were excluded from OPLS-DA models. To assess the reliability of the OPLS-DA models and to test the risk of overfitting, a cross-validated analysis of variance (CV-ANOVA) was performed [65]. Variables most likely correlated with bioactivity were deduced from the S-plots of the OPLS-DA models, and their reliability was deduced from the jackknife-based confidence interval in the loading plot of the S-plot [17]. Samples containing high levels of the candidate compounds were identified from the Xvar plots of the respective variables.

Annotation of OPLS-DA Candidates
OPLS-DA candidate features were annotated either by comparing their retention time, molecular formula and MS/MS fragmentation pattern with authentic reference standards (unambiguous assignment) or by comparing the data with data from the literature or mass spectral databases (tentative annotation). In cases where MS/MS fragmentation was not sufficient for tentative annotation, samples were additionally analyzed with the same UHPLC instrument and parameters as described above but while using the ion trap mass spectrometer (LTQ XL linear ion trap mass spectrometer equipped with H-ESI II probe, Thermo Fisher Scientific), which supports MS n experiments. Analyses were performed in the HESI-negative mode. Heater temperature was 250 • C, capillary temperature was 330 • C, sheath gas flow was 20 and aux gas flow 8 arbitrary units. Source voltage was 3 kV, capillary voltage −48 V and tube lens −147 V. Normalized collision energy for data-dependent MS 2 and MS 2 was 35, activation Q was 0.25 and activation time was 30 ms.

Isolation and Structure Elucidation of OPLS-DA Candidates
Aerial parts of L. hypoglauca were collected on 17 July 2013 in Guangxi Province, China and authenticated as described in Section 2.2. Voucher specimens are kept in the herbarium of Guangxi Botanical Garden of Medicinal Plants (GXMG, Nanning, China) and at the Department of Pharmacognosy in Graz (Austria) (voucher specimen Nr. L-102). One and a half kilograms of the dried, powdered material were extracted by percolation with CH 3 OH at room temperature. The extract was concentrated on a rotary evaporator and dried under nitrogen to yield 113 g (7.5%) CH 3 OH extract.
Of this, 108 g of the dried extract were subjected to liquid-liquid extraction, and the ethyl acetate subfraction was fractionated by performing several vacuum liquid chromatography (VLC) and solid-phase extraction (SPE) separation steps. Compounds were purified by means of semipreparative HPLC. The details of the isolation procedure are provided in the Supplementary  NMR spectra were recorded on a Bruker Avance III 700 MHz instrument; pyridine-d5 or MeOH-d4 (Eurisotop, France) were used as solvents; spectra were recorded at 25 • C; chemical shifts δ are relative to Me4Si as internal standard, J is in Hz.

Conclusions
Correlating pharmacological activity and UHPLC-HRMS data by means of OPLS-DA enabled us to narrow down the incredibly broad and diverse spectrum of secondary metabolites present in the investigated Lonicera species to a panel of candidates with putative pharmacological activity that could subsequently be annotated and further investigated. Among these candidates, we identified several new natural products and compounds that had been previously undiscovered in the genus Lonicera. In a set of tested candidate compounds, pharmacological activities could be verified in some but not in all cases. These findings confirm that the applied MVDA approach is feasible, enabling researchers to preselect a panel of potentially active candidates in active herbal extracts. However, when this approach is applied to complex crude extracts, the most potent and thus potentially toxic candidates fall predominantly within the top ranks. Therefore, we anticipate that this approach will need to be optimized by fractionating Lonicera extracts rich in putative antiinflammatory active compounds, accompanied by pharmacological testing, UHPLC-HRMS analysis, chemometric methods and molecular networks.
Supplementary Materials: The following supporting information can be downloaded at https://www. mdpi.com/article/10.3390/metabo12040288/s1: Table S1: List of Lonicera samples included in the metabolic profiling study, their collection dates and origins, and extract yields obtained by ASE extraction with 96% ethanol.  Figure 2). Table S4: Annotation of candidate compounds. Table S5: Figure S4: OPLS-DA model of CD-processed UHPLC-HRMS data in correlation with activity on transactivation of an NF-κB-driven luciferase reporter gene in TNF-α-stimulated HEK293/NF-κB-luc cells. (A) Score scatter plot. Blue: highly active samples; green: samples with low activity; moderately active samples were excluded from the model. (B) S-Plot of the OPLS-DA model. Variables most likely correlated with bioactivity are marked in red and listed in Table 2 and Table S3. Figure Table 2 and Table S3. Figure S6: OPLS-DA model of CD-processed UHPLC-HRMS data in correlation with the activity on NO production in LPS/IFNγ-stimulated RAW 264.  Table 2 and Table S3. Figure  Funding: This research was funded by the Austrian Science Fund FWF (NFN Drugs from Nature Targeting Inflammation, grant numbers S10704, S10705 and S10711).