Salvia africana-lutea samples were collected from five different sites: Stellenbosch (S 33° 55.120′ E 18° 51.360′), Brackenfell Nature Reserve (S 33° 52.845′ E 18° 42.784′), Koeberg Nature Reserve (S 33° 40.128′ E 18° 26.524′), Silwerstroomstrand (S 33o 34.632′ E 18o 22.349′) and Yzerfontein (S 33° 22.309′ E 18° 10.871′). All locations are situated in the Western Cape Province of South Africa (Additional file 1). In total, 25 samples were collected from the different sites. Samples were identified by Dr Petra Wester and voucher specimens were deposited at the Stellenbosch University Herbarium. The arboreal plant parts were collected at two different times: April 2009 and June 2011. The plant material was oven-dried in closed brown bags at 50°C then ground to a fine powder using a mortar and pestle with liquid nitrogen prior to storage in the dark at room temperature.
Extraction of plant material
For each site, extraction was carried out on the dried powdered aerial parts (5 g) with 20 ml of a 1:1 (v/v) methanol:dichloromethane mixture in a 60 ml glass test tube. These were then sonicated for 35 min (Bransonic 220, USA) before filtering with Whatman filter paper number 1. The extraction was repeated twice and pooled extracts were dried using a rotary evaporator (Buchi, Germany) at 55°C. Extracts were then stored in a desiccator prior to use. Micro-extraction was done on 0.5 g of dried powder using 10 ml of solvent mixture. These were then vortexed for 1 min and sonicated for 30 min. This step was repeated twice prior to centrifuging for 2 min at 4750 revolutions per minute (rpm). All extracts were filtered using cotton wool in a Pasteur pipette and collected in a 10 ml tube. Thereafter, they were evaporated to dryness in vacuo. Five extractions were performed for each site at one particular time. The experiments were repeated at least twice; unless otherwise stated.
Fungal isolates and microtitre assays
Isolates of two fungal species Fusarium verticillioides (MRC 826 and 8267) and F. proliferatum (MRC 7140 and 6908) kept at the PROMEC Unit Culture Collection of the South African Medical Research Council (MRC) were used. The Fusarium isolates utilized are classified as high fumonisin B1 producers. Fungal isolates were grown on Carnation Leaf Agar (CLA) slants for 21 days at 25°C to induce spore production and stored in a cold room at 4°C prior to use. Fungal suspensions were prepared by dislodging the conidia in a 20 ml sterile 0.85% (w/v) saline solution. Conidia suspensions were standardized to a 0.5 McFarland concentration. The reference method for broth dilution antifungal susceptibility testing of filamentous fungi as described by the M38-A2 guide of the Clinical and Laboratory Standards Institute  was used to determine the minimum inhibitory concentration (MIC) for plant extracts. Each plant extract was resuspended in dimethyl sulfoxide (DMSO) to obtain stock solutions at a concentration of 50 mg ml-1. These were further diluted in the Roswell Park Memorial Institute RPMI-1640 medium at a 1:50 (v/v) ratio to obtain final concentrations of 1.0, 0.5, 0.25, 0.125, 0.063, 0.031, 0.016, 0.008, 0.004 and 0.002 mg ml-1 in the 10 wells. Voriconazole (Vfend®, Pfizer) was used as a positive control. A row of DMSO and medium was used as a solvent control, while the last negative control had the medium only (growth control).
Five extracts from the different study sites were resuspended in 1 ml of a 50% (v/v) mixture of acetonitrile and H2O containing 0.1% (v/v) formic acid. The suspensions were vortexed for 1 min then sonicated for 5 min, vortexed again for 1 min prior to spinning at 10 000 rpm for 10 min. The supernatant (3 μl) was injected into the LC-MS instrument. Metabolites were separated using a gradient of H2O with 0.1% formic acid (solvent A) and acetonitrile (solvent B), using a Waters UPLC at a flow rate of 0.4 ml min-1 on a Waters BEH C18, 2.1×50 mm column. Mass spectrometry was obtained on a Waters SYNAPT™ G2 MS (Manchester, England) using electron spray ionization (ESI) running in positive mode with a cone voltage of 15 V. The injections were repeated once to ensure repeatability.
Twenty one dried micro-extracts from five different locations (four from Stellenbosch, four from Yzerfontein, three from Silwerstroomstrand, five from Koeberg and five from Brackenfell) were vortexed for 15 s after the addition of 2.5 ml of DMSO-d6 (Eurisotop, France). The mixture was filtered and 550 μl of the filtrate were analyzed. A 10 mM solution of sodium 2,2,3,3-tetradeutero-3-trimethylsilylpropionate (TSP) (Sigma-Aldrich, St. Louis MO, USA) (10 μl) was added as an internal chemical shift reference before the NMR analysis. One dimensional (1D) 1H NMR spectra were recorded at 298 K on a Bruker Avance 500 NMR instrument operating at 500.13 MHz, equipped with a 5-mm TCI cryoprobe. The 1H NMR experiments were acquired using a relaxation delay-pulse-acquisition sequence. Acquisition parameters were as follows: pulse width of 2.8 μs (flip angle ≈ 30°), relaxation delay of 4 s, 64 K data points, spectral width of 9500 Hz (19 ppm) and 128 scans. All FIDs were processed using the Bruker TopSpin 2.1 software with one level of zero-filling and a line broadening of 0.7 Hz. Baseline correction was performed on each spectrum and spectra were referenced to the signal of TSP at δ 0.00 ppm.
Chemometric analysis of the data
The 1D 1H NMR spectra were transferred to the KnowItAll® software (Bio-Rad, USA). The bin area method was used to segment the spectra between 0 and 13.1 ppm with the variable size intelligent bucketing tool included in the KnowItAll® package. Bucket sizes ranged from 0.01 to 0.30 ppm. The spectral regions containing the NMR signals of DMSO (δ 2.47-2.57 ppm) and its 13C satellites (δ 2.36-2.40 ppm, and 2.63-2.67 ppm), methanol (δ 3.17-3.20 ppm), H2O (δ 3.33-3.45 ppm) and dichloromethane (δ 5.75-5.80 ppm) were removed. A manual filtering procedure was applied to the whole spectrum to exclude buckets that contained only noise. A total of 119 variables were considered for subsequent statistical analyses. For spectrum normalization, integrated regions were divided by the total area of the spectrum and multiplied by the mean value of the corresponding family previously calculated. Data were preprocessed by mean-centering. The unit variance (UV) scaling method was applied prior to analysis.
Both principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) were performed with the SIMCA-P + 12.0 software (Umetrics, Umeå, Sweden) and for the t-tests, the R software (R Development Core Team, 2012) was used. The predictive ability of the PLS-DA models was assessed from the values of Q2
cum (> 0.5), R2Ycum (> 0.7) and R2Xcum parameters. The statistical significance of R2Y and Q2 parameters was also estimated through the response permutation test where the Y matrix was 999 times randomly permuted when the X matrix was fixed . For determining the discriminating variables between classes, loading plots, coefficient plots, variable importance in the projection (VIP) from PLS-DA models, and p-values (<10-3) of the t-tests on the variables arising from the coefficient plot and VIP were considered. Several PLS-DA models were built: (i) from variables of the whole 1H NMR spectrum (119); (ii) from variables of the 13-5 ppm region of the 1H NMR spectrum (69); and, (iii) from a combination of the 69 variables of the 13-5 ppm region of the 1H NMR spectrum with the 39 variables corresponding to the main peaks of the LC-MS chromatograms.
For each sample 100 mg of ground plant material was utilized. A similar protocol to that used by Glassop et al.  was employed with only a minor change for the derivatization of solvent extracts. Myo-inositol (2 mg ml-1) was added to the ground plant material then dissolved in 350 μl of methanol:chloroform (1:1, v/v) in a 2 ml microcentrifuge tube. All sample tubes were placed in a sonicator (Bransonic 220, USA) at room temperature for 45 min. Samples were then centrifuged at 1 200 rpm for 10 min at room temperature in a centrifuge (Biofuge pico, Germany). Ribitol (Sigma-Aldrich; Germany) was included as an internal standard after derivatization. One μl of the samples was injected for a splitless run with an initial temperature of 70°C (5 min) and a maximum oven temperature of 330°C (equilibration time of 0.25 min) was used. Analysis was performed using a network GC system (6890 N) coupled to inert XL EI/CI Mass Selective Detector (MSD) 5975B (Agilent Technologies Inc., Palo Alto, CA) equipped with a CTC Analytics PAL autosampler had. Separation was achieved with a capillary column (Restek RTX200; trifluoropropylmethyl (30 m in length; 250 μm diameter; 1 μm in thickness)).
The temperature was increased from 76°C (1 ramp min-1) to 320°C (4 ramp min-1) The run time was 72 min and helium gas was used as a carrier at a flow rate of 53.7 ml min-1. The instrument was set to the following conditions: pressure of 62.6 kPa, purge flow of 50 ml min-1 for 2 min, flow rate of 1 ml min-1 (37 cm sec-1) and a data rate of 20 Hz.
The mass spectrometer was operated in electron ionization (EI) mode at ionization energy of 70 eV, scanning from 35 to 600 m/z in positive mode. Caffeic acid, rosmarinic acid, myo-inositol, glucose, galactose and mannose were used as standards to aid with identification of constituents in the extracts. All standards were purchased from Sigma-Aldrich (Germany) except for the mannose which was provided by Merck (Germany). Data were analyzed using the MSD Chemstation software which was linked to the National Institute of Standards and Technology (NIST) mass spectral search program library ver. 2.0 d (2005; standard reference data program of the National Institute of Standards and Technology, USA) for peak identification of metabolites. A library match of 80% and above for metabolites from the NIST library were regarded as likely hits. The relative abundance of metabolites was recorded using the total ion chromatogram peak integration (Additional file 2).