Fresh leaf samples of T. terrestris L. were separately harvested from three different locations of Mpumalanga province, Bushbuckridge (24º50’23.28’’S, 31º2’47.04’’E), Kamagugu (25º27’23.76’’S, 31º0’12.24’’E) and Nkomatipoort¹ (25º27’3.6’’S, 31º57’31.32’’E). The samples were harvested between October and December 2020. All the harvested samples were growing in the locations as weeds (not cultivated). To minimise variation, young tender leaves, specifically from the apex of the shoot, free of dirt or soil particles, decay and/or damage, were carefully plucked from 20 young flowering plants according to the method reported by Ntuli (2019), with modifications. To prevent contamination and moisture loss, the samples were put in Ziploc™ bags. On arrival in the laboratory, the samples were separately washed under running distilled water and freeze-dried (−85°C, LyoQuest-55/Telstar, Shanghai, China) and ground into fine powder for analysis according to the method reported by Moloto et al. (2020).

Freeze-dried leaf samples (0.1 g) were homogenised in water with 2 mL of 80% methanol, then sonicated at 30°C for 30 min. The mixture was then centrifuged at 3000 × g for 20 min at 4°C using a Hermle Z326k (Hermle Labortechnik GmbH, Wehingen, Germany). The collected supernatants were used for the UHPLC-QTOF-MS analysis.

A 0.22-µm polytetrafluorethylene filter was then used to filter the supernatants. A Quadrupole 120 time-of-flight (QTOF) mass spectrometer UPLC-QTOF/MS (Waters, Milford, Massachusetts, United States of America [USA]) was used to identify and quantify predominant secondary metabolites. An ACQUITY UPLC BEH C18 column (2.1 mm × 100 mm i.d., 1.7 × 10⁻⁶ m; Waters) was used for all analyses. The mobile phase was composed of acetonitrile (A) and 0.1% formic acid, v/v (B), with the following gradient elution: (1) 0–8 min, 95% – 80% A; (2) 8–12 min, 80% – 70% A; (3) 12–15 min, 70 A – 65 A; (4) 15–18 min, 65% A; (5) 18–21 min, 65% – 20% A; (6) 21–23 min, 20% – 5% A; (7) 23–24 min, 5% A; (8) 25–30 min, 95% A. The flow rate of the mobile phase was 0.4 mL/min and the temperatures of the column and autosampler were maintained at 30°C and 10°C, respectively. Data were processed using Mass Spectrometry – Data Independent Analysis (MS-DIAL) and MS-FINDER (RIKEN Center for Sustainable Resource Science: Metabolome Informatics Research Team, Kanagawa, Japan) (Lai et al. 2018; Tsugawa et al. 2015). Functions 1 (unfragmented channel) and 2 (fragmented channel) of the Waters MSe data were processed by MS-DIAL to produce MS1 and MS2 spectra as well as extracted ion chromatograms with associated peak height intensity data. As calibration standards are not available for the majority of these compounds, the peak height intensity was converted to concentration in a semiquantitative manner by interpolation of a calibration curve for catechin acquired under the same instrumental conditions. Each deconvoluted feature (alignment in MS-DIAL), together with its associated MS1 and MS spectra, were exported from MS-DIAL to MS-FINDER. Based on the accurate mass elemental compositions, possible compounds were identified from the listed databases and then subjected to in silico fragmentation. According to the spectral match between the in silico and measured spectra, a score (out of 10) is assigned to each of the possible compound matches, with the highest score being accepted as the most likely (assuming a score of at least four).

Secondary metabolites that could be the elements driving the antiproliferative effect of T. terrestris L. leaves were identified by nontargeted profiling using ultra-performance liquid chromatography coupled with high-resolution Q-TOF-MS. The UHPLC mass spectrometry (MS) examination of T. terrestris L. leaves taken from several locations in Mpumalanga province revealed 50 secondary metabolites, the majority of which were flavonoids, phenolic acids and saponins (Table 1). The annotations or identifications were tentative based on accurate mass elemental composition and fragmentation spectra. Based on ESI positive data, the results from this study revealed that quercetin derivatives (quercetin 3-gentiobioside and rutin) and F-gitonin with concentrations of 10 886, 9307 and 7296.7 mg/kg DW, peak 7; 9068, 7118 and 5885.3 mg/kg DW, peak 9; and 10 387, 7613.5 and 5707.7 mg/kg DW, peak 23 in Kamagugu, Nkomati and Bushbuckridge, respectively, were predominant compounds in the leaves of T. terrestris L. Peak 39, m/z 426.301, was discovered as compound jervine in the leaves of T. terrestris L., which had the lowest concentrations (296.7, 440, and 422.5 mg/kg DW) from Bushbuckridge, Kamagugu and Nkomati, respectively. The leaves of T. terrestris L., as reported in this study, were high in steroidal saponins, which are compounds likely to be responsible for the libido effects in humans and possibly the testosterone mimicry in animals (Ștefănescu et al. 2020). Enrichment of the saponins in T. terrestris L. reported in this study poses them as an ideal carbohydrate source for diabetic patients because of their ability to potentiate glucose-induced insulin release (Pereira-Leal & Seabra 2001). Saponins, because of their amphiphilic molecule, have membrane-permeabilising properties; thus, they could increase the absorption of other compounds. This property is of great importance because toxic effects could appear in patients with multiple conditions who undergo chronic treatments. The current findings contrast with those of Lazarova et al. (2011), who found significant differences (different amounts of chemicals and the absence of some compounds) between samples taken from the same geographic location. A rise in the use of T. terrestris L. supplements has also been seen among athletes, who are always looking for natural ways to improve their performance (Ștefănescu et al. 2020). Saponins have biological and therapeutic activities that include: (1) haemolytic factor, (2) anti-inflammatory, (3) antibacterial, (4) antifungal, (5) antiviral, (6) insecticidal, (7) anticancer, (8) cytotoxic and (9) molluscicidal effects (Abdel Gawad 1999; Armelle et al. 2018; Cheng et al. 2011; Ellen et al. 2007; Hassan et al. 2010; Just et al. 1998; Simões, Amoros & Girre 1999; Sindambiwe et al. 1998; Sparg, Light & Van Staden 2004).

As a result of the complexity and multidimensionality of metabolomic data, appropriate statistical and chemometric tools were used to obtain chemical information and convert it into biological knowledge (Boccard & Rudaz 2014). Chemometrics is the science of extracting valuable information from complex data sets through pattern recognition and machine learning algorithms (Brereton 2015; Tebani, Afonso & Bekri 2018). The unsupervised (PCA) obtained data from the UPLC-QTOF/MS showed that the phenolic metabolites in the leaves of T. terrestris L. from different locations of Mpumalanga province differed in distribution. Principal component analysis is a multivariate technique that increases interpretability and minimises the loss of biological information by reducing the dimensionality of complex data sets (Jolliffe & Cadima 2016). The underlying structures and characteristics of the data are thus revealed by this unsupervised, explorative method. The PCA plot from this study revealed three district clusters based on the metabolites shown by the leaves of T. terrestris L. in different locations of Mpumalanga province (Bushbuckridge, Kamagugu and Nkomati village) (Figure 1 and Figure 2). In this study, the sample size was too small to develop robust chemometric models. However, as the aim was not to develop a prediction model but merely to investigate inter- and intraspecies variation, chemometric analysis was nevertheless used to analyse the liquid chromatography mass spectrometry (LC-MS) data. The two principal component (PC) models were able to distinguish clearly between the locations, with 22.9% separation along PC2 and 37.6% separation along PC1. Two locations, namely Bushbuckridge and Kamagugu, were distinctly separated by the first principal component (PC1), with Bushbuckridge samples situated on the negative (PC1) and Kamagugu on the positive (PC1) areas of the plot, while one sample from the Nkomati locality clustered along the negative (PC1), and the rest clustered in the positive (PC1). The relationship of the observed variation to locality was further evaluated by performing classification and discriminant analysis using orthogonal projections to latent structures discriminant analysis (OPLS-DA). This supervised pattern recognition technique was used to identify variables (in this case, peaks on the chromatogram) responsible for the differences between the groups. To obtain more in-depth information regarding the differential metabolic profiles revealed by PCA modelling, a supervised method, OPLS-DA, was used. It can be seen from the plot that the greatest variation was along PC2 (Figure 3). The OPLS-DA scores plot confirmed that the samples from different locations were very different in composition. Figure 4 shows the box-and-whisker plot representing an in-depth, semiquantitative visualisation of the distribution of selected metabolites ([A] parvispinoside B, [B] F-gitonin and [C] gitonin) in the leaves of T. terrestris L. from three different locations of Mpumalanga province plots. The interlocation variations in metabolite profiles are illustrated by the increased or relative decrease of the mentioned metabolites. Figure 4 reveals a relative increase in the metabolite concentration of T. terrestris L. leaves from the Bushbuckridge region, while the leaves from the Kamagugu region revealed a decreased concentration of these metabolites. The location-based quantitative and qualitative metabolite distribution was further evidenced in the group-wise metabolite selection, biplot-based discrimination analysis and the variable importance in projection (VIP), showing metabolic markers contributing to metabolic phenotypes in Figure 4. These are secondary metabolites, which were responsible for the differentiation in PCA detected. The performance of the developed model was validated by the computed parameters ‘R² (0.9856)’ and ‘Q² (0.9577)’, indicating its covered variance and good prediction power, respectively (Figure 5). Metabolites facilitating the discrimination of the leaves of T. terrestris L. from three different locations are shown in Figure 4. The PLS-DA also finds the characteristics (metabolites) with the highest VIP scores. As a result, the contribution of each metabolite to group separation was calculated using VIP ratings. Variable importance in projection scores is calculated by adding the squares of the partial least squares (PLS) loadings, which measure the amount of Y-variance explained across all dimensions and the weighted sum of the PLS-regression coefficients (Carvalho et al. 2021). Thus, metabolites were ranked according to their VIP scores, and only the top metabolites with the highest VIP scores were deemed to provide the most meaningful interpretation of the data (Carvalho et al. 2021). Hierarchical cluster analysis was performed on the whole data set of discovered metabolites to visualise clusters of samples with comparative chemical composition, which gave additional evidence of the related metabolites associated with T. terrestris L. from different regions. The results of this study were merged with a heat map structure created from metabolite concentrations across all samples (Figure 6). Blue in the heat map denotes metabolite downregulation or lower expression, whereas red depicts metabolite upregulation or higher expression. Heat maps also make patterns that might not be evident or visible.