Profiling of Chlorogenic Acids from Bidens pilosa and Differentiation of Closely Related Positional Isomers with the Aid of UHPLC-QTOF-MS/MS-Based In-Source Collision-Induced Dissociation

Bidens pilosa is an edible herb from the Asteraceae family which is traditionally consumed as a leafy vegetable. B. pilosa has many bioactivities owing to its diverse phytochemicals, which include aliphatics, terpenoids, tannins, alkaloids, hydroxycinnamic acid (HCA) derivatives and other phenylpropanoids. The later include compounds such as chlorogenic acids (CGAs), which are produced as either regio- or geometrical isomers. To profile the CGA composition of B. pilosa, methanol extracts from tissues, callus and cell suspensions were utilized for liquid chromatography coupled to mass spectrometric detection (UHPLC-QTOF-MS/MS). An optimized in-source collision-induced dissociation (ISCID) method capable of discriminating between closely related HCA derivatives of quinic acids, based on MS-based fragmentation patterns, was applied. Careful control of collision energies resulted in fragment patterns similar to MS2 and MS3 fragmentation, obtainable by a typical ion trap MSn approach. For the first time, an ISCID approach was shown to efficiently discriminate between positional isomers of chlorogenic acids containing two different cinnamoyl moieties, such as a mixed di-ester of feruloyl-caffeoylquinic acid (m/z 529) and coumaroyl-caffeoylquinic acid (m/z 499). The results indicate that tissues and cell cultures of B. pilosa contained a combined total of 30 mono-, di-, and tri-substituted chlorogenic acids with positional isomers dominating the composition thereof. In addition, the tartaric acid esters, caftaric- and chicoric acids were also identified. Profiling revealed that these HCA derivatives were differentially distributed across tissues types and cell culture lines derived from leaf and stem explants.


Introduction
Bidens pilosa L. (B. pilosa) is a flowering edible herb from the family Asteraceae, commonly known as "Blackjack" and as "Beggar's tick" or "Needle grass" and has been reported to be a nutritious food and medicine source [1][2][3]. This plant is thought to have originated from South America but has spread to most tropical hot areas including African countries [4]. B. pilosa has been shown to treat over 41 diseases and conditions [2], such as malaria [5], diabetes [6], hypertension [4], obesity [7] and syphilis [8]. B. pilosa has various bioactivities owing to its diverse phytochemical constituents, of pressure (between high pressure and high vacuum) and applied voltages accelerate ions, promoting collision with the surrounding gas. This permits multiple activation-fragmentation episodes before ions reach the mass analyzer, allowing for efficient structural elucidation [36,37]. ISCID proves to be superior to MS E , as in this method in-source dissociation is performed prior to regular MS/MS. This then emulates pseudo-MS 3 , acquiring fragmentation patterns typical for IT mass spectrometers [38].
In the present study, an optimized ISCID method was used, in combination with metabolomic tools and approaches, to differentiate between CGAs regio-isomers (including di-esters) in tissues, callus and suspension-cultured cells of B. pilosa. The results of the current study also contribute to the profiling of CGAs in differentiated tissues vs. undifferentiated cells and contribute to possible identification of any underlying biochemical mechanisms with regard to CGA biosynthesis within tissues and cell cultures of B. pilosa.

Profiling of Chlorogenic Acid Derivatives in Tissues and Cell Cultures of B. pilosa
In this study, the distribution profile of HCA derivatives in tissues (leaves, stems, and roots), callus and cell suspensions derived from stems and leaf tissues were studied with the aid of a high-throughput analytical method, UHPLC-QTOF-MS ( Figure 1 and Figure S1A,B respectively). Structural elucidation and putative annotations were achieved through an ISCID method [34,35]. The HCA derivatives were shown to be a prominent group of metabolites in methanol extracts of B. pilosa tissues ( Figure 1) and extracts from stem-and leaf-derived callus and cell suspensions ( Figure S1A,B), shown in base peak intensity (BPI) chromatograms and indicated with yellow rectangles. Furthermore, the BPI chromatograms indicated differences in intensities of the HCA derivatives as shown in Figure 1. Within a specific plant species, the distribution of secondary metabolites is expected to vary not only as a function of developmental stage, but also among plant tissues, hence some tissue-specific metabolite variations are observed [39].
Metabolites 2020, 10, x 3 of 22 pressure (between high pressure and high vacuum) and applied voltages accelerate ions, promoting collision with the surrounding gas. This permits multiple activation-fragmentation episodes before ions reach the mass analyzer, allowing for efficient structural elucidation [36,37]. ISCID proves to be superior to MS E , as in this method in-source dissociation is performed prior to regular MS/MS. This then emulates pseudo-MS 3 , acquiring fragmentation patterns typical for IT mass spectrometers [38].
In the present study, an optimized ISCID method was used, in combination with metabolomic tools and approaches, to differentiate between CGAs regio-isomers (including di-esters) in tissues, callus and suspension-cultured cells of B. pilosa. The results of the current study also contribute to the profiling of CGAs in differentiated tissues vs. undifferentiated cells and contribute to possible identification of any underlying biochemical mechanisms with regard to CGA biosynthesis within tissues and cell cultures of B. pilosa.

Profiling of Chlorogenic Acid Derivatives in Tissues and Cell Cultures of B. pilosa
In this study, the distribution profile of HCA derivatives in tissues (leaves, stems, and roots), callus and cell suspensions derived from stems and leaf tissues were studied with the aid of a highthroughput analytical method, UHPLC-QTOF-MS ( Figure 1 and Figures S1-A,B respectively). Structural elucidation and putative annotations were achieved through an ISCID method [34,35]. The HCA derivatives were shown to be a prominent group of metabolites in methanol extracts of B. pilosa tissues ( Figure 1) and extracts from stem-and leaf-derived callus and cell suspensions ( Figure S1-A,B), shown in base peak intensity (BPI) chromatograms and indicated with yellow rectangles. Furthermore, the BPI chromatograms indicated differences in intensities of the HCA derivatives as shown in Figure 1. Within a specific plant species, the distribution of secondary metabolites is expected to vary not only as a function of developmental stage, but also among plant tissues, hence some tissue-specific metabolite variations are observed [39]. The yellow rectangles indicate the chromatographic regions where hydroxycinnamic acid derivatives are present across the three tissue types with some visible differences in intensities.

Multivariate Data Analysis to Reveal Tissue-Specific and Cell Line-Specific Differences Within Tissues and Cell Cultures of B. pilosa
To analyze the variability within and between the tissue extracts (Figure 2A  To analyze the variability within and between the tissue extracts (Figure 2A), cell callus ( Figure S2A) and cell suspensions ( Figure S2B), principal component analysis (PCA) was performed for all metabolites within each sample group. PCA is an unsupervised, explorative chemometric tool for the reduction of dimensionality of complex datasets to provide insights into variations and systematic trends among sample groups [40,41]. The computed model (score plot) of the PCA for tissues indicate that 41.1% and 24.3% of the variation were explained by PC1 and PC2, respectively ( Figure 2A). Statistical validation was described using R 2 and Q 2 which explains the goodness-of-fit of the model and model predictability, respectively. The model computed was acceptable for metabolomic analysis of the phytochemical data as the R 2 > 0.7 and the Q 2 > 0.4 [42]. For the computed model, R 2 = 0.890 and Q 2 = 0.874 respectively were found to be statistically adequate to make relevant biological interpretations. The PCA model revealed an obvious separation among the three tissue types, as shown in Figure 2A. This indicates that the metabolic constituents and their distribution in leaf, stem and root tissues of B. pilosa varied significantly. This observation is also reflected in the variation of intensities of separated metabolites observed in the BPI chromatograms shown in Figure 1. Similarly, PCA scores plots were computed and validated for stem-and leaf-derived callus and cell suspensions as indicated in Figure S2A and Figure S2B respectively.
Metabolites 2020, 10, x 4 of 22 metabolites within each sample group. PCA is an unsupervised, explorative chemometric tool for the reduction of dimensionality of complex datasets to provide insights into variations and systematic trends among sample groups [40,41]. The computed model (score plot) of the PCA for tissues indicate that 41.1% and 24.3% of the variation were explained by PC1 and PC2, respectively ( Figure 2A). Statistical validation was described using R 2 and Q 2 which explains the goodness-of-fit of the model and model predictability, respectively. The model computed was acceptable for metabolomic analysis of the phytochemical data as the R 2 > 0.7 and the Q 2 > 0.4 [42]. For the computed model, R 2 = 0.890 and Q 2 = 0.874 respectively were found to be statistically adequate to make relevant biological interpretations. The PCA model revealed an obvious separation among the three tissue types, as shown in Figure 2A. This indicates that the metabolic constituents and their distribution in leaf, stem and root tissues of B. pilosa varied significantly. This observation is also reflected in the variation of intensities of separated metabolites observed in the BPI chromatograms shown in Figure 1. Similarly, PCA scores plots were computed and validated for stem-and leaf-derived callus and cell suspensions as indicated in Figure S2-A and Figure S2-B respectively. Ward's minimum variance as a dissimilarity and linkage rule, respectively) shows tissue-specific clustering into two major groups, grouping roots and stems tissues together.
To determine correlations/similarities among the different tissues, agglomerative hierarchical cluster (HC) analysis models were computed. Each observation is initially treated as an individual cluster; then after processing, groups are merged to indicate group similarities [43]. These were represented by means of a dendrogram shown in Figure 2B which indicates differences in metabolomic profiles respective to the different tissues of B. pilosa. The dendrogram indicates that the metabolomic profiles of stems and roots are closely related, contrasting with that of leaf tissues [44]. Dendrograms were also constructed to indicate similarities within different cell cultures of B. pilosa as shown in Figure S2-C and Figure S2-D.

Metabolite Annotations
Mass spectral data were obtained in both ESI (+/-) modes, but negative ionization was preferred as the majority of metabolites were found to ionize better in the negative MS mode [31,32] and since the hierarchical scheme proposed by Clifford et al. [31] was also generated using negative ionization. In this study a combined total of 33 CGA derivatives (both regio-and geometric isomers) were identified in B. pilosa tissues, callus and cell suspensions as listed in Table 1 with corresponding MS . The quality parameters of the model are: explained variation/goodness-of-fit R 2 = 0.890 and the predictive variance Q 2 = 0.874. The ellipse in the PCA score scatterplot indicates the Hotelling's T 2 at 95% confidence interval. (B) Hierarchical cluster analysis of the hierarchical structure of the data in dendrogram format. The model computed (using Euclidean distance and Ward's minimum variance as a dissimilarity and linkage rule, respectively) shows tissue-specific clustering into two major groups, grouping roots and stems tissues together.
To determine correlations/similarities among the different tissues, agglomerative hierarchical cluster (HC) analysis models were computed. Each observation is initially treated as an individual cluster; then after processing, groups are merged to indicate group similarities [43]. These were represented by means of a dendrogram shown in Figure 2B which indicates differences in metabolomic profiles respective to the different tissues of B. pilosa. The dendrogram indicates that the metabolomic profiles of stems and roots are closely related, contrasting with that of leaf tissues [44]. Dendrograms were also constructed to indicate similarities within different cell cultures of B. pilosa as shown in Figure S2C,D.

Metabolite Annotations
Mass spectral data were obtained in both ESI (+/-) modes, but negative ionization was preferred as the majority of metabolites were found to ionize better in the negative MS mode [31,32] and since the hierarchical scheme proposed by Clifford et al. [31] was also generated using negative ionization. In this study a combined total of 33 CGA derivatives (both regioand geometric isomers) were identified in B. pilosa tissues, callus and cell suspensions as listed in Table 1 with corresponding MS data in  and structures presented in Figure 6. CGAs are classically described as a subclass of phenylpropanoids formed when HCAs are esterified to QA, while a wider scope of the definition describes the CGAs as conjugates of many other compounds such as malic acid, succinic acid, fumaric acid, tartaric acid and sugars [27,32,45]. QA has axial hydroxyl groups attached to carbon positions 1 and 3 and equatorial hydroxyl groups attached to carbons 4 and 5 where the HCAs attach to form CGAs [32]. HCAs are naturally synthesized with trans configurations but it has been shown that the cis configuration can occur as a result of UV radiation [46,47] and under stress conditions [25]. CGAs are chemically diverse and the various regioand geometrical isomers can make discrimination challenging. UHPLC coupled to CID during tandem MS/MS procedures (such as UHPLC-Q-TOF-MS) can partially solve this problem and identify CGAs through the use of fragmentation patterns as outlined in the discussion [34,45,48]. Metabolites were putatively identified to level 2 of the Metabolomics Standards Initiative (MSI) [49]. Figure 6. CGAs are classically described as a subclass of phenylpropanoids formed when HCAs are esterified to QA, while a wider scope of the definition describes the CGAs as conjugates of many other compounds such as malic acid, succinic acid, fumaric acid, tartaric acid and sugars [27,32,45]. QA has axial hydroxyl groups attached to carbon positions 1 and 3 and equatorial hydroxyl groups attached to carbons 4 and 5 where the HCAs attach to form CGAs [32]. HCAs are naturally synthesized with trans configurations but it has been shown that the cis configuration can occur as a result of UV radiation [46,47] and under stress conditions [25]. CGAs are chemically diverse and the various regio-and geometrical isomers can make discrimination challenging. UHPLC coupled to CID during tandem MS/MS procedures (such as UHPLC-Q-TOF-MS) can partially solve this problem and identify CGAs through the use of fragmentation patterns as outlined in the discussion [34,45,48]. Metabolites were putatively identified to level 2 of the Metabolomics Standards Initiative (MSI) [49].

Discussion
The number system used below refers to the sequence of presentation in Table 1 where metabolites are listed according to increasing m/z values.

Characterization of Mono-Acyl Chlorogenic Acids (CGAs)
The hierarchical scheme keys for LC-MS n identification of CGAs [31] was used to assist in the identification of metabolite

Discussion
The number system used below refers to the sequence of presentation in Table 1 where metabolites are listed according to increasing m/z values.

Characterization of Mono-Acyl Chlorogenic Acids (CGAs)
The hierarchical scheme keys for LC-MS n identification of CGAs [31] was used to assist in the identification of metabolite

Discussion
The number system used below refers to the sequence of presentation in Table 1 where metabolites are listed according to increasing m/z values.

Characterization of Mono-Acyl Chlorogenic Acids (CGAs)
The hierarchical scheme keys for LC-MS n identification of CGAs [31] was used to assist in the identification of metabolite Previous work [34,35,47] was also used as references in putatively identifying these metabolites listed in Table 1. Mass spectrometric data and chromatographic elution order were also considered while determining the regioand geometric isomers of the annotated metabolites. Notably, cis-isomers of 4and 3-CQAs elute before their trans-isomers, while the cis-isomer of 5-CQA elute after its trans-isomer and after the 4-CQA [46]. 4-CQAs ( Figure S4B [31]. Fragmentation of 5-CQAs ( Figure 4C) result in the formation of a single base peak product ion at m/z 191. Hence, based on the elution order and fragmentation patterns, (4-7) were identified as trans-3-CQA (4), trans-5-CQA (5), trans-4-CQA (6), and cis-5-CQA (7) respectively. In tissues, all the trans-CQAs were present with the exception of trans-3-CQA which was absent in roots. In callus cultures only trans-5-CQA was found to be absent. The mono-CQAs were observed in cell suspensions derived from both stems and leaves of B. pilosa; however, the cis-5-CQA was only present in cell suspensions derived from leaf tissue.
A similar approach was followed in the identification two feruloylquinic acid (FQA) isomers (8 and 9) which were identified by their precursor ion [M-H]at m/z 367 and based on their fragmentation patterns and Rt shown in Table 1. Although differing in intensities, the two FQA regio-isomers were identified in all tissues and cell culture systems. As described in the hierarchical scheme keys for LC-MS n identification of CGAs [31],

Characterization of Caffeoylglycoside
HCAs in nature may occur as soluble forms conjugated to organic acids and/sugars [50]. In this study, one caffeoylglycoside (3) was identified which had a precursor ion, [M-H] -, at m/z 341 as shown in Table 1. The molecular ion fragmented to give ions at m/z 179 [CFA-H]due to loss of a glycosyl residue and an ion at m/z 135[CFA-CO 2 ]shown in Figure S6 [35]. The caffeoylglycoside was found to be present in all the tissues of B. pilosa.

Characterization of Hydroxycinnamoyl-Tartaric Acid Esters
Hydroxycinnamoyl-tartaric acid esters such as the di-ester of two caffeic acids to tartaric acid (chicoric acid, CA) and the mono-ester of caffeic acid to tartaric acid (caftaric acid, CTA), are biologically active compounds which are shown to have various health benefits and antioxidant properties [51,52]. These HCA derivatives are the main caffeic acid derivatives in Echinacea purpurea but have been also identified in leaves of B. pilosa and in more than 60 plant genera [53,54]. Caftaric acid ( Figure S7A [29]. Profiling of the tartaric esters revealed that they are only present in the aerial parts of the plants and absent in the roots. This could suggest exclusivity in the biosynthesis of tartaric acid esters, suggesting localized biosynthesis of these esters in the aerial parts of the plant. These esters were also absent in cell cultures of B. pilosa.

Characterization of Di-Caffeoylquinic Acids (Di-CQAs)
In this study, six di-CQAs shown in Table 1  Although it may be present in the fragmentation pattern of 4,5-di-CQA, the intensity is comparatively lower. Differences observed in the intensities of fragment ions can be ascribed to variances in energy distribution which cause the structurally similar isomers to behave differently under the described MS conditions. On a reverse-phase column the elution order of di-CQA regio-isomers is expected to be as follows: 3,4-di-CQA, 3,5-di-CQA, followed by 4,5-di-CQA eluting the latest, validating the annotation of di-CQAs shown in Table 1 [21,29]. These di-CQA were differentially present throughout tissues and cell cultures of B. pilosa.  Table 1 [55].

Characterization of Tri-Caffeoylquinic Acids (Tri-CQAs) and Di-Caffeoylquinic Acid Glycosides
Metabolites The Rt was also considered as tri-CQAs are expected to elute later than 4,5-di-CQA as these are more hydrophobic [56]. However, positions of acylation on tri-caffeoylquinic acids/tri-acylated glycosides were not fully characterized as description of triand tetra-acylation would require MS 4 and/or MS 5 spectra [32]. These tri-acylated HCA derivatives were observed to be only present in the leaves of B. pilosa.

Characterization of p-Coumaroyl-Caffeoylquinic Acids (pCo-CQAs)
As mentioned earlier, CGAs are a complex group of compounds that may also comprise of mixed di-esters. Characterization of p-coumaric acid-containing di-acyl-CGAs has been previously done for green coffee beans [53]. To the authors' knowledge, these have never been identified in B. pilosa. pCo-CQAs were identified by a parent ion [M-H]at m/z 499 [22,57]. The elution order of these metabolites was considered to assist in their annotation. Six of these isomers were observed (11-16) and they occurred in pairs as shown in Figure 3A and Table 1. These isomers were found to follow an elution order similar to that of di-CQAs where 3,4-di-esters elute first, followed by the 3,5-di-esters and with the 4,5-di-esters eluting last. The 20 eV collision energy level (MS E ) was considered when annotating these metabolites.
The first pair was annotated as 3-pCo-4-CQA (11) and 3-C-4-pCoQA (12) shown in Figure 4A The presence of a peak at m/z 173 was indicative of acylation at C4 of the QA, the peak at m/z 335 indicated a dehydrated caffeoylquinic acid and its ratio of approximately 30% to the base peak was characteristic of a 3,4-di-chlorogenic acid. An intense product ion at m/z 163 is characteristic of a coumaric residue at C3 of the QA. The 3-C-4-pCoQA (12) fragmented to give a base peak m/z 337, which is indicative of a loss of a caffeoyl residue and secondary peak at m/z 173 which indicated that the coumaroyl residue was acylated at C4 of the QA.
The next pair of isomers were identified as 3,5-di-esters due to their lack of a product ion at m/z 173. Figure 4C indicates the fragmentation pattern of a metabolite annotated as 3-pCo-5-CQA (13) with a base peak at m/z 337 indicating that the caffeoyl residue was extensively lost and which suggests that the caffeoyl residue is attached at position C5 of QA. According to [31] the acylation at position C5 is the easiest to remove followed by that at position C3, whilst the one at C4 is the hardest to remove. A product ion at m/z 163 was also observed which indicated that the coumaroyl residue was attached at position C3, hence this metabolite was annotated as 3-pCo-5-CQAs (13). The other isomer of this pair was annotated as 3-C-5-pCoQA (14). Its fragmentation pattern is indicated in Figure 4D, showing a base peak at m/z 353 and secondary ions of m/z 337, 191 and 179. The caffeoyl residue was assigned position C3 as the secondary ions m/z 191 and 179 showed behavior analogous to that of 3-CQA were a base peak at m/z 191 is observed and an ion at m/z 179 is present at 50% intensity compared to the m/z 191.
The last two pairs were annotated as 4,5-di-esters and are indicated in Figure 4E and Table 1. The isomer that eluted first in this pair was annotated as 4-coumaroyl-5-caffeoylquinic acid (15) and the fragments observed were a base peak at m/z 337 and secondary ions at m/z 173 and m/z 163, also shown in Figure 4E. Absence of the m/z 353 suggested that the caffeoyl residue was more extensively lost and most likely attached at position C5 of QA. The base peak at m/z 337 gave a fragment at m/z 173, suggesting that the coumaroyl residue was attached to position C4 of QA. The last isomer was annotated as 4-C-5-pCoQA (16). The fragmentation pattern of this metabolite is shown in Figure 4F, with a base peak at m/z 353 and secondary ions at m/z 337, 191, 179 and 173. All these isomers were present in both stem and leaf cell suspensions but absent in callus cultures. Five of these isomers were observed in leaf tissue of B. pilosa but were undetected in the stem and root tissues.

Characterization of Feruloyl-Caffeoylquinic Acids (F-CQAs)
The F-CQAs (23-29) were identified by their parent ion at m/z of 529 and the chemical structures are illustrated in Figure 6 [22,31,57]. Chromatographically, six isomers were observed which eluted in pairs shown in Figure 3B. Figure 5A Figure 5B shows the fragmentation of 3-C-4-FQA (24) which gave a base peak of m/z 367 and secondary ion at m/z 173, indicating that the feruloyl residue is attached at position C4 of QA.

The first two isomers were identified as 3-F-4-CQA (23) and 3-C-4-FQA (24). A typical fragmentation pattern of 3-F-4-CQA (23) shown in
The next two isomers were annotated as 3,5-di-esters as they lacked a fragment ion of m/z 173 which indicated no acylation at position C4 of QA. These were annotated as 3-F-5-CQA (25) and 3-C-5-FQA-1 (26). The fragmentation pattern of 3-F-5-CQA (25 - Figure 5C) shows a base peak at m/z 367 which indicates extensive loss of the caffeoyl residue. This suggests acylation with a caffeoyl residue at position C5, while the intense secondary ion at m/z 193 indicates that the feruloyl residue was attached to position 3. Figure 5D shows the fragmentation pattern of 3-C-5-FQA (26) and fragments observed were m/z 353, 337, 191 and 179. The caffeoyl residue was assigned to position C3 as ions m/z 191 and 179 showed behavior similar to that of 3-CQA where a base peak at m/z 191 is observed and an m/z at 179 is present at 50% intensity compared to the m/z 191 ions. The feruloyl moiety was thus assigned to position C5 of QA.
The last two isomers were annotated as 4-F-5-CQA (27) and 4-C-5-FQA (28). The first eluting isomer showed a base peak m/z 367 and secondary ions m/z 193 and an intense m/z 173 shown in Figure 5E, hence the feruloyl residue was attached at position C4. Absence of m/z 353 indicated acylation of the caffeoyl residues at position C5 of QA. The fragmentation pattern of 4-C-5-FQA (22) is shown in Figure 5F. A later eluting F-CQA was also identified in stems and leaf tissues. This metabolite was identified as 3-C-5-FQA-2 (29) as it had similar fragmentation ions to (26). Table 1. Characterization of chlorogenic acids (CGAs) consisting of hydroxycinnamic acid (HCA) derivatives of quinic acid (QA) and tartaric acid from tissues (L-leaves, S-Stems and R-Roots) and from two callus cell lines (C-L and C-S) and two cell suspension lines (S-L and S-S) of Bidens pilosa.

Distribution of HCA Derivatives in Tissues and Cell Cultures of B. pilosa
Differential distribution of HCA derivatives were observed amongst the various tissue types of B. pilosa as indicated in Table 1. Most annotated HCA-derivatives (specifically quinic acid esters) were found to be mostly distributed in stem and leaf tissues as opposed to the roots. As described in [58], CGAs were found to be distributed mainly in chlorenchyma cells and appeared to be associated with chloroplast and were implicated to confer protection to chloroplasts against light. These metabolites were also found to be localized in the vascular bundles and this could suggest that they are transported throughout plant organs. From the observations of the distribution of CGAs in B. pilosa tissues, this suggests that the synthesis of these metabolites could be localized in the aerial parts of the plant and possibly translocated to other plant organs. Furthermore, this suggests that in B. pilosa, the hydroxycinnamoyl-CoA/quinate hydroxycinnamoyl transferase (HQT), an enzyme responsible for the biosynthesis of quinic acid esters [59] could be localized in or near the chloroplast, hence accumulation of HCA-derivatives was higher in the aerial parts of the plant.
In contrast, the tartaric acid esters were only to be distributed in the aerial parts of the plant and mostly in leaves whilst absent in root tissues. This corresponded to the analysis of different tissues of E. purpurea where CA was found to be present more in the apical parts of that plant [60]. Absence of the tartaric acid esters (CA and CTA) in the root tissues of B. pilosa could suggest a few possibilities that would require further investigations. Metabolite distribution patterns may differ, due to differences in expression of genes and localization of enzymes responsible for their biosynthesis [61]. The hydroxycinnamoyl-CoA/tartaric acid hydroxycinnamoyl transferase (HTT), an enzyme responsible for the biosynthesis of the tartaric esters [62] could be localized in the aerial parts of the plant considering the apparent complete absence of these esters in roots, therefore suggesting its localization in the chloroplast. However, further proteomics studies are required to validate the localization of HTT in the chloroplast and measurements of posttranscriptional regulation would need to be conducted to connect the distribution of these esters to localization of HTT. Deducing from the observations, HCA derivatives were generally abundant in the aerial parts of the plant; this could provide a chemical basis for the distinct usage of different tissues of B. pilosa, to maximally harness its bioactivities. Plant cell cultures have been shown to be an attractive approach for the controlled production of bioactive natural products like phenylpropanoids, compared to the use of wild plants [11]. In cell suspensions of B. pilosa, 23 HCA-derivatives were identified while only 14 of these were identified in the callus cultures as presented in Table 1. Although plant cell culture is a promising alternative for metabolite production and provides numerous advantages, a significant limitation of using cell culture is that undifferentiated cells may accumulate secondary metabolites to a lesser extent compared to the parent plant [63,64]. As observed in this study, fewer HCA-derivatives were accumulated in cell cultures compared to plant tissues where 30 HCA-derivatives were identified. However, various strategies can be employed to improve yield of secondary metabolites in plant cell cultures such as optimizing cultural conditions (medium modifications) [65], precursor feeding [66], immobilization techniques [11], elicitation [67], and screening for high-producing cell lines [14]. As mentioned above, fewer HCA-derivatives were identified in cultured callus cells compared to cell suspensions. This could be because callus is maintained on semi-solid media while cell suspensions are maintained in liquid medium which is agitated to enhance oxygenation, promote better growth and transfer of nutrients [68,69].
Cell suspensions of B. pilosa were shown to be better starting material for in vitro cultivation for production of HCA-derivatives such as CGAs. Cultured cells of B. pilosa were noted to possess some form of a genetic memory and totipotency for the biosynthesis of some CGAs, similar to that of the parent plant. This observation may in future provide possibilities for bioreactor-based large-scale production of these biologically important secondary metabolites. Cultured cells are comparable to undifferentiated meristematic cells and lack chloroplasts. This could explain the apparent lack of tartaric acid esters in B. pilosa cultures [70]. These esters were hypothesized to be exclusively biosynthesized by the enzyme HTT as mentioned above which possibly could be localized in the chloroplasts of B. pilosa. Although making photosynthetic cell cultures with functional chloroplasts has been deemed difficult and time-consuming, recent advances have been made where medium modification in Arabidopsis thaliana cultures resulted in chloroplast formation [71].

Plant Cultivation Tissues and Undifferentiated Cells
B. pilosa seeds were collected from matured plants in the wild (Venda area of South Africa) and air-dried at room temperature. The seeds were cold shocked at 4 • C for 48 h; this was performed as a way of cold-wet stratification to break seed dormancy in summer perennials [72]. The seeds were then sown in Culterra germination mix (Culterra, Muldersdrift, South Africa, http://culterra.co.za). Germinated plants were then grown under greenhouse conditions at 28 • C for a period of two months. The plants were watered twice a week and fertilized once every two weeks with a fertilizer containing 90 mg/L mono-potassium phosphate, 150 mg/L Soluptase, 20 mg/L Microples, 40 µL/L Kelp-P-Max, 650 mg/L CaNO3, 400 mg/L KNO3, 300 mg/L MgSO4, and 90 mg/L mono-ammonium phosphate. Plants stems, roots and leaves were harvested and immediately shock-frozen in liquid nitrogen to quench all metabolic reactions [73][74][75]. The frozen plant tissues were stored at −80 • C, pending metabolite extractions.

Callus Initiation and Cell Suspension Cultures
B. pilosa callus cultures were established from leaf and stem explants on Murashige and Skoog medium with Murashige and Skoog vitamins containing 100 mg/L myo-inositol, 1 g/L hydrolyzed casein and 30 g/L sucrose with agar. The medium contained growth regulators, 0.45 mg/L 2,4-dichlorophenoxyacetic acid (2,4-D) and 1 mg/L 6-benzylaminopurine (BAP) at pH 5.8. Friable callus was sub-cultured into the medium with 0.45 mg/L 2,4-D and 1.0 mg/L BAP and grown in Erlenmeyer flasks on an orbital shaker at 120 rpm at room temperature with a light/dark cycle of 12 h/12 h and maintained at a low light intensity of 30 µmol/m 2 /s.

Metabolite Extraction
Two grams (2 g) of the samples; cell suspensions (harvested using filter paper (70 mm) on a vacuum filtration system), callus and plant tissues (crushed frozen with liquid nitrogen using a mortar and pestle) were homogenized at 5100 rpm in 20 mL (1:10 m/v) of 80% methanol (Romil SpS, Cambridge, UK). Samples were sonicated for 30 min at 30 • C and 100% intensity in a sonicator bath (Branson CPX, Fischer Scientific, Waltham, MA, USA). The crude extracts were centrifuged at 5100 rpm in a benchtop centrifuge (Beckman Coulter, Midrand, South Africa), for 15 min and supernatants were evaporated under vacuum using a rotary evaporator (Heidolph Instruments, Schwabach, Germany), at 55 • C to~1 mL of samples. Samples were transferred to 2 mL Eppendorf tubes and dried to completion overnight in a dry bath at 55 • C, reconstituted with 500 µL of 50% methanol and sonicated for 30 min at 30 • C followed by filtration using 0.22 µm nylon filters into HPLC glass vials with 500 µL inserts. Samples were stored at 4 • C until future analysis. To ensure experimental reproducibility, eight independent biological replicates were prepared, and three instrumental technical replicates were analyzed.

Ultra High-Performance Liquid Chromatography-Quadrupole Time-of-Flight Mass Spectrometry (UHPLC-QTOF-MS/MS)
Extracts were analyzed on an ultrahigh-performance liquid chromatography-quadrupole time-of-flight MS instrument (UHPLC-QTOF SYNAPT G1 system, Waters Corporation, Manchester, UK) fitted with an Acquity HSS T3 C18 column (150 mm × 2.1 mm with particle size of 1.7 µm) (Waters, Milford, MA, USA). An injection volume of three µL was used and a binary solvent system was used consisting of solvent A: 0.1% formic acid in Milli-Q water (both HPLC grade, Merck, Darmstadt, Germany) and solvent B: acetonitrile (UHPLC grade, Romil SpS, Cambridge, UK) with 0.1% formic acid. A binary solvent gradient (with solvent A and B) with a flow rate of 0.4 mL/min was used to separate analytes over 30 min. The separation conditions were: 2% B over 0.0-1.0 min, 2%-60% B over 2.0-24 min, 60%-95% B over 24-25 min, from 25-27 min the conditions were maintained at 95% B and the column was washed with 95%-2% B over 27-28 min. The column was allowed to re-equilibrate with 5% B over a 2 min isocratic wash. The chromatographic effluents were further analyzed utilizing the SYNAPT G1 Q-TOF high definition mass spectrometer (Waters Corporation, Manchester, UK). Separate injections (using the same chromatographic settings and conditions) were performed for positive and negative electrospray ionization (ESI) modes. The MS conditions were set as follows: capillary voltage of 2. Prior to MVDA, UHPLC-QTOF-MS raw data was processed using MassLynx XS™ software's MarkerLynx application (Waters, Manchester, UK). The MarkerLynx application performs accurate peak detection and alignment using the patented ApexTrack algorithm [41]. The following parameters were used: Retention time (Rt) range of 0.50-22 min with a Rt window of 0.2 min, mass range of 100-1000 Da and the mass tolerance as 0.05 Da. After the peaks were detected the corresponding ions (consisting mostly of the pseudomolecular ion peaks before fragmentation) were analyzed (maximum intensity, Rt and m/z) and recorded for all the samples. Data normalization was done by using total ion intensities of each defined peak. Prior to calculating intensities, the software performs a patented modified Savitzky-Golay smoothing and integration. The data matrix obtained from MassLynx was exported into SIMCA-15.0 software (Umetrics Corporation, Umea, Sweden) for statistical modeling. Statistical models computed were principal component analysis (PCA) and hierarchical cluster (HC) analysis which are both unsupervised models that show trends, clusters and similarities between samples. Agglomerative HC models were computed using Ward's linkage method (incremental sum of squares method) that considers between-and within-cluster distances when forming clusters, and the tree was sorted based on size [37]. Unless stated otherwise, Pareto scaling was applied for all computed models to reduce the relative importance (masking effects) of large values from abundant metabolites, but partially preserve data structure [44,76,77]. Metabolite annotation was executed based on mass spectral information from MS E and/or MS 2 experiments, accurate mass information, elemental composition calculations and searches in various databases such as ChemSpider and Dictionary of Natural Products (DNP, dnp.chemnetbase.com). Fragmentation patterns were also compared to available literature such as the hierarchical scheme keys for LC-MS n identification of CGAs [31]. Metabolites were putatively identified to level 2 of the Metabolomics Standards Initiative (MSI) [49]. The surrogate standard approach, through comparison with already analyzed plant extracts, was also followed to validate the identity of metabolites of which authentic standard are not available [78].

Conclusions
In this study tissues, callus and cell suspensions of B. pilosa were shown to contain diverse HCA derivatives of quinic acid which substantiates the reported health benefits of the plant as an alternative food source and, moreover, the use of cell culture for production of biologically important secondary metabolites. These HCA derivatives were shown to have differential distribution across tissues and cell cultures. Given this point, undifferentiated cells of B. pilosa indicated cell line-specific differences in distribution of HCA derivatives as a result of their inherent metabolite memory. The protocol outlined in this study offers possibilities for sustainable production of biochemically important phenolic compounds in cell suspension cultures of B. pilosa as reported for the first time in this study. Although the CGAs identified in this study are structurally diverse in terms of their geometric and regio-isomerism, the applied UHPLC-QTOF-MS/MS in-source collision-induced dissociation method assisted in annotating and differentiating between the CGA isomers found the cellular extracts. Minor differences in the fragmentation patterns gave characteristic diagnostic peaks that can be used to efficiently elucidate the regio-isomers of the HCA derivatives. Fragmentation patterns similar to those described in the hierarchical scheme keys for LC-MS n identification of CGAs [31] were obtained by use of in-source collision-induced dissociation. This method (ISCID) provides an analytical avenue that allows for efficient discrimination of CGA regio-isomers unaccompanied by the use of advanced MS technologies. In addition to the CGAs, HCAs esterified to tartaric acid and sugars are also reported in tissues of B. pilosa which indicates diversity in metabolite composition of this plant.  Figure S3: A typical mass spectrum of the fragmentation pattern of 5-coumaroylquinic acid. Figure S4: Typical mass spectra of the fragmentation patterns of 3-caffeoylquinic acid (A), 4-caffeoyquinic acid (B) and 5-caffeoylquinic acid (C). Figure S5: Typical mass spectra of the fragmentation patterns of 3-feruloylquinic acid (A) and 4-feruloylquinic acid (B). Figure S6: A typical mass spectrum of the fragmentation patterns of caffeoylglycoside. Figure S7: Typical mass spectra of the fragmentation patterns of caftaric acid (A) and chicoric acid (B). Figure S8: Typical mass spectra of the fragmentation patterns of 3,4-di-caffeoylquinic acid (A), 3,5-di-caffeoylquinic acid (B) and 4,5-di-caffeoylquinic acid (C). Figure S9: Typical mass spectra of the fragmentation patterns of tri-caffeoylquinic acid (A) and di-caffeoylquinic acid glycoside (B).