As a case study, we selected three Iranian medicinal Salvia species, which are all locally rare, and exhibiting a diversity of distributional and altitudinal ranges (Jafari Foutami and Akbarlou, 2017; Figure 1). For each of the four species we attempted to sample across the geographical distribution of the species. Salvia sclarea L. (clary sage; N = 10) is native to the Northern Mediterranean but has also become a weed outside its native range including North America, where it is called European sage (Ghani et al., 2010; Dickinson and Royer, 2014). Salvia multicaulis Vahl (syn. S. acetabulosa L.; N = 8) is native to Turkey and bordering countries (Tepe et al., 2004). Salvia aethiopis L. (N = 4) is naturally occurring in the Northern Mediterranean but has become a noxious weed in North America, where it is referred to as Mediterranean sage (Chalchat et al., 2001; Dickinson and Royer, 2014). Additionally, we included S. officinalis (N = 12) allowing for comparison with extensive literature and verified reference material adhering to the European Pharmacopeia standards (Council of Europe, 2014).

Salvia officinalis L. Ph.Eur 8.3. quality leaf reference material was obtained from Alfred Galke GmbH, Germany. Fresh samples of S. aethiopis, S. multicaulis and S. sclarea were collected in Iran in 2017 and air-dried at room temperature. Herbarium material of S. aethiopis, S. multicaulis, S. officinalis, and S. sclarea was obtained from Herbarium C of the Natural History Museum of Denmark, University of Copenhagen spanning 150 years of collecting across the Mediterranean (Figure 2).

Taxonomic identity of specimen was confirmed to species morphologically by Isa Jafari Foutami. Collection year was recorded from labels and are listed in Table 1 together with altitude and GPS coordinates. However, in most cases, GPS coordinates and in a few cases altitude had to be inferred from the locality description on the labels. Details of all plant materials are listed in Table 1.

Twenty-Five Milligram leaf material was ground using a mortar and pestle under liquid N2 and transferred to a vial (Supelco, Pennsylvania, USA). Seven Hundred Microliter analytical grade hexane (Sigma-Aldrich, Denmark) was added, the sample was vortexed for 10 s and then incubated on a shaker at 37°C for 2 h. Samples were then vortexed again for 10 s and left for 24 h to allow tissue to settle. Approximately 200 μL of the hexane extract was transferred into a weighed glass GC vial with a limited volume insert and kept at −20°C until GC-MS analysis. The remainder of the sample was left uncapped at 65°C until the solvent was evaporated. Vials were reweighed to obtain the dry mass of leaf material. Samples were analyzed in triplicate.

GC–MS analyses were carried out using an Agilent 6890N Gas Chromatograph equipped with a split/splitless injector (200°C), a HP-5MS capillary column (30 m × 0.25 mm; film thickness 0.25 μm), and coupled with an Agilent 5975 MS Detector (MSD), operating in the electron impact (EI) mode at 70 eV. The carrier gas was helium (1.0 mL/min), and the oven temperature was programmed to increase from 60°C to 280°C at a rate of 3°C/min. The injected volume was 2 μL.

Chromatograms were analyzed both as untargeted data using tentative identifications based on the mass spectra in the NIST 8.0 library (National Institute of Standards and Technology, Gaithersburg, MD, USA) and as targeted data using 20 standard compounds.

Commercially available pure chemical certified reference material standards of common Salvia constituents (Russo et al., 2013; Hatipoglu et al., 2016; Craft et al., 2017) were obtained from Sigma-Aldrich, Germany (Table 2). Purity of the commercial standards was not investigated experimentally and the approach does therefore not guarantee no degradation of the standards could have happened. However, for the targeted approach, all samples were analyzed using the same standards.

A dilution series of the standard compounds in hexane was made and analyzed as samples: linear regression was used to calculate the concentrations of samples (R² ≥ 0.886 for individual compounds). The set of standards included monoterpene hydrocarbons (α-Pinene, Camphene, β-Pinene, β-Myrcene, β-Cymene, β-Phellandrene, Limonene, Terpinolene, 3-Carene, α-Ocimene, γ-Terpinene, Aromandendrene), oxygenated monoterpenes (Linalool, Camphor, Borneol, Bornyl acetate, Eucalyptol), and sesquiterpene hydrocarbons (Aromadendrene, β-Caryophyllene, Humulene, Nerolidol).

GC-MS chromatograms were processed using PARADISe, a PARAFAC2 based deconvolution and identification system for direct analysis of complex raw GC-MS data (Johnsen et al., 2017). The targeted and untargeted data matrices are provided as Supplementary Material online (Supplementary Tables S1, S2). Triplicates were averaged and a number of samples did not have triplicates due to sample failure (sample 1, 2, 28, and 30). Compounds with retention times between α-Pinene (8.3 min) and Camphor (28.4 min) were retained for further analysis and compounds with zero or negative values were excluded.

All statistical analyses were performed using R statistical software (version 2.14.9). Initially untargeted data was converted into a Bray-Curtis similarity matrix and non-metric multidimensional scaling was performed using the Vegan package (Oksanen et al., 2013), which was visualized using ggplot2 (Wickham, 2016). A significant species effect on the untargeted chemical composition was tested using multivariate generalized linear modeling followed by analysis of variance (MGLM-ANOVA), which was performed using the manyglm and anova functions within the MVABUND package (Wang et al., 2012). Given the strong expected effects of species on plant chemistry (e.g., Hegnauer, 1962–1973; Rønsted et al., 2012), PERMANOVAs were ran to test for the effects of collection year, altitude and geography on untargeted chemical composition, using the adonis function with the Vegan package. Due to incomplete datasets, the effect of each was tested for individually in a PERMANOVA. Furthermore, sample GPS coordinates were converted into a principle coordinates of neighbor matrix (PCNM), with the resulting first two principle components (PCNM1 and PCNM2) used within the PERMANOVA testing for geographical effects on the untargeted chemical composition.

One-way analysis of variance (ANOVA) was performed to test for significant differences in untargeted chemical richness between species. Meanwhile, mixed linear modeling (MLM) was performed using the lmer function from the package lme4 (Bates et al., 2014). In these, the significance of random effects (altitude and geography (as PCNM1 and PCNM2) were determined using the drop1 function, which performed likelihood ratio tests (chi-square), whilst accounting for variation associated with the random effect (species).

The targeted dataset consisted of 20 compounds for which standards were obtained as described above (Table 2). The targeted chemical composition was analyzed as before, undergoing nMDS for visualization), MGLM-ANOVA and PERMANOVAs to assess for compositional effects, and ANOVA and MLM to assess for differences in chemical richness. Given that standards allowed for reliable quantification, each individual compound also underwent generalized linear modeling (GLM) to test for species effects, with compound serving as the independent variable and species serving as the dependent using the glm function of the native stats package of R. MLM was performed to test for the effects of year of collection, altitude and geography (as PCNM1 and 2) whilst accounting for potential species effects. Individual compounds from the targeted data served as the independent variable, species served as the random effect, and either year of collection, altitude, or PCNM1 and PCNM2 (together) serving as the fixed effect.

After quality filtering a total of 285 compounds were within the untargeted dataset (Supplementary Table S2). Nearly all samples contained detectable levels of the majority of compounds, with untargeted chemical richness varying from 249 to 283 between samples, with no clear effect of species on the richness (ANOVA; df = 3, F-value = 1.16, P-value = 0.338). However, the untargeted chemical composition demonstrated clear clustering in the nMDS plot (Figure 3A), which proved highly significant (MGLM-ANOVA; Table 3).

Within the targeted dataset, samples varied from 14 to 20 in the number of the 20 compounds analyzed, strongly differed between species (ANOVA; df = 3, F-value = 29.95, P-value = <0.001), with Salvia multicaulis and S. officinalis having significantly higher targeted chemical richness (19.7 and 19.4, respectively) than S. sclarea and S. aethiopis (15.6 and 16.0, respectively). Targeted chemical composition clearly clustered by species within the nMDS plot (Figure 3B), which was also highly significant (MGLM-ANOVA; Table 3).

Within the targeted dataset, S. officinalis and S. multicaulis consistently contained significantly higher concentrations when compared to the other two species (Figure 4). For example, borneol was the most abundant compound, ranging from 172 to 96,234 ng/μL, and were significantly higher in S. officinalis (31,954 μg/g) and S. multicaulis (21,964 μg/g) than averages in S. sclarea (436 μg/g), and S. aethiopis (675 μg/g) (Table 2). p-cymene was the second most abundant compound that was also concentrated in S. officinalis (11,559 μg/g) and S. multicaulis (6,774 μg/g) when compared to S. sclarea (51 ng/μL) and S. aethiopis (47 μg/g). Whilst camphor was present in lower abundance than borneol and p-cymene in S. officinalis (8291 μg/g) and S. multicaulis (5,273 μg/g), it was also present in comparable quantitates in S. sclarea (3,246 μg/g) and in two of the five S. aethiopis samples (average 2,696 μg/g). Terpinolene was also in abundance within S. officinalis (2,387 μg/g) whilst being almost absent from the other species. The other 15 compounds were present in relatively low amounts across all samples.

Given the confirmation of species effects on plant chemical composition, we next investigated age effects on the chemical composition. PERMANOVAs for both the targeted and untargeted datasets did not demonstrate a significant age effect on chemistry (PERMANOVA; Table 3). There were no significant effects of age of sample on either targeted or untargeted chemical richness (MLM; Untargeted χ₂ = 3.79, P-value = 0.051 and targeted χ₂ = 0.07, P-value = 0.787). Two individual compounds significantly declined in concentration over time (MLM; Table 4), α-phellandrene (Figure 5A), which was a very minor constituent, and camphor that was present in all four species (Figure 5B). Total amount of the compounds in the targeted dataset also significantly declined over time (MLM; χ₂ = 4.11, P-value = 0.042) (Figure 5C).

Altitude did not significantly affect the targeted or untargeted chemical composition (PERMANOVA; Table 3) and chemical richness (MLM; χ² = 0.182, P-value = 0.670 and χ₂ = 0.563, P-value = 0.453 in the untargeted and targeted datasets, respectively). Altitude also did not affect a single individual compound in the targeted dataset (MLM; Table 4). Geographic variation too did not affect chemical composition (PERMANOVA; Table 3) nor did it affect a single individual compound in the targeted dataset (MLM; Table 4). However, geographic variation significantly affected chemical richness in the untargeted dataset (PCNM1: χ₂ = 8.37, P-value = 0.004, PCNM2: χ₂ = 7.19, P-value = 0.007) but not the targeted (PCNM1: χ₂ = 0.06, P-value = 0.794, PCNM2: χ₂ = 3.34, P-value = 0.068).

The biological properties of essential oil of Salvia officinalis are attributed mainly to α-thujone and ß-thujone, camphor, and 1,8-cineole as reviewed by Raal et al. (2007). In the present study of herbarium samples, we also found monoterpenes to be the main fraction of compounds mainly consisting of borneol, p-cymene, camphor, and to a lesser extent camphene and terpinolene (Table 2). Whereas, high levels of borneol and camphor correspond to the findings of Raal et al. (2007), the high levels of p-cymene may reflect some degree of degradation of terpenoids, as p-cymene is often identified in aged oils (Turek and Stinzing, 2013). However, no age effect was observed for p-cymene.

The monoterpenes borneol, α-pinene, p-cymene, camphor, and camphene, were the main compounds found in S. multicaulis in the present study (Table 2). α-pinene, borneol, camphor, and camphene are also main compounds reported in other studies of S. multicaulis (e.g., Rustaiyan et al., 1999; Tepe et al., 2004; Morteza-Semnani et al., 2005; Bagci and Kocak, 2008; Karamian et al., 2013; Hatipoglu et al., 2016) in addition to 1,8-cineole, eucalyptol, bornyl acetate, and sesquiterpenes such as ß-caryophyllene and germacrene-D.

Salvia sclarea and S. aethiopis generally contained fewer compounds and in lower concentration than S. officinalis and S. multicaulis. In the survey of 45 Turkish Salvia species by Hatipoglu et al. (2016), both species were found to be rich in sesquiterpenes relative to monoterpenes. Our targeted dataset only included comparably few sesquiterpenes compared to monoterpenes. In S. aethiopis, we mainly found the oxygenated monoterpene camphor and lower amounts of borneol as well as the monoterpene hydrocarbon terpinolene and the sesquiterpenes ß-caryophyllene and α-humulene (Table 2). Other studies have primarily reported a range of sesquiterpenes from Salvia aethiopis including β-caryophyllene, bicyclogermacrene, germacrene-D, α-humulene, and α-copaene (e.g., Torres et al., 1997; Rustaiyan et al., 1999; Chalchat et al., 2001; Gulluce et al., 2006; Hatipoglu et al., 2016, and references therein). In S. sclarea, we found mainly the oxygenated monoterpenes camphor and borneol, which is in line with other studies reporting oxygenated monoterpenes such as linalyl acetate, linalool, and α-terpineol to be the main components (e.g., Gülçin et al., 2004; Farka et al., 2005; DŽamić et al., 2008; Ghani et al., 2010; Sharopov and Setzer, 2012). Previously, germacrene-D, β-caryophyllene, bicyclogermacrene sesquiterpenes have been reported from this species (Gülçin et al., 2004; Farka et al., 2005; DŽamić et al., 2008).

Among 108 volatile compounds across 45 Turkish Salvia species (Hatipoglu et al., 2016), both total yield and percentage of individual compounds were highly variable between species. α-pinene, camphene, β-pinene, 1,8-cineole, camphor, and borneol were the main chemical markers of monoterpene rich species, whereas β-caryophyllene, germacrene-D, β-bisabolene, bicyclogermacrene, caryophyllene oxide, and spathulenol were the main chemical markers in sesquiterpene rich species. In summary, several studies of essential oil components of different Salvia species have reported presence of a huge range of compounds, many in trace amounts, and other consistently present in higher amounts in specific species.

In the present study of herbarium specimens, we observed a significant species effect using both a targeted and an untargeted approach. Variation in study design including the number of compounds, geographical origin of samples, samples per species, and plant parts analyzed, makes it difficult to directly compare our findings with the literature. However, in general the number and amount of compounds as well as the main individual chemical compounds found in the present study are also reflected in the literature confirming that the species profiles we observe across 150-years of herbarium specimens are comparable to modern samples.

We observed a strong species effect and a weak geographical effect on terpenoid chemical composition of four Salvia species across the Mediterranean and no overall effect of sampling age, although two individual compounds did change with age of samples. These findings suggests that chemical composition appears to be well preserved in herbarium samples over at least 100–150 years suggesting herbarium samples can be used for chemical screening as well as for preliminary studies of pharmacological activity, which is also supported by previous studies reporting little effect of sampling age on overall chemical profiles (Eloff, 1999; Zangerl and Berenbaum, 2005; Cook et al., 2009; Yilmaz et al., 2012; Jafari et al., 2018). Herbarium samples may therefore provide an untapped resource for preliminary studies as well as for large-scale comparative surveys across taxonomical and geographical ranges. Whereas overall chemical composition appears to be well preserved over 100–150 years, the concentrations of individual compounds may be affected and we would therefore suggest herbarium samples may be well-suited for qualitative studies, whereas caution is needed in the use of herbarium samples for quantitative studies depending also on the type of compounds of interest (Rønsted et al., 2017). At a larger scale, a wide and thorough phytochemical screening of herbarium samples from all over the world could give us a better understanding also of chemotaxonomic issues, and comparison with climatic data would allow for testing of long-term environmental impact.

We observed a geographical effect on chemical richness in the untargeted but not in the targeted dataset of Salvia species across the Mediterranean. Geographic variation had no effect on chemical composition or on individual compounds. Russo et al. (2013) found chemical composition of S. officinalis essential oils to vary depending on environmental factors such as altitude, water availability and soil conditions in Italy. We did not observe an effect of altitude in the present study. However, our sampling covered a wide geographical range and therefore the potential effects of altitude alone may have been confounded by sampling across multiple ranges across such large scales as the Mediterranean. Future studies may disentangle geographical and environmental effects by more intensive sampling across single or multiple altitudinal gradients. However, it should be noted that sampling from herbaria is limited by the number of specimens available in the collection and herbarium specimens are mostly collected as individual specimens rather than representing populations, which limits the prospects of comparative environmental studies.

Many other studies have shown that the yield and chemical composition of essential oils varied with ecological factors and geographical areas (e.g., Uribe-Hernández et al., 1992; Salgueiro et al., 1997; Viljoen et al., 2006; Liu et al., 2015; Rezende et al., 2015; Jaradat et al., 2017). Essential oils play important biological functions related to environmental adaptation, protection against biotic and abiotic stresses, and pollinator attraction (Jassim and Naji, 2003; Zangerl and Berenbaum, 2005). Therefore, plants of the same species growing under different environmental conditions may differ in the composition of their essential oils in response to different environmental pressures (Weiss and Edwards, 1980; Salimpour et al., 2011). Over evolutionary time scales, unique chemotypes may eventually develop into separate genotypes adapted to different environmental conditions (Verpoorte et al., 2000; Heywood, 2002; Wink, 2003; Beccera, 2015). In addition to further comparison of environmental parameters, future studies may benefit from comparing genetic profiles of samples.

Salvia comprises nearly 1,000 species with distributions spread across the globe (Drew et al., 2017). Several Salvia species are used as food flavoring owing to their content of essential oils as well as in folk medicine for a range of conditions including microbial infections, inflammation, cancer, and malaria (Lu and Foo, 2002; Kamatou et al., 2008; Russo et al., 2013; European Medicines Agency, 2016; Hatipoglu et al., 2016).

Some species like Salvia officinalis are common and widespread, whereas others are narrow and potentially threatened endemics (Wood and Harley, 1989; Viney et al., 2006; Kahraman et al., 2012). A better understanding of the chemical composition of different Salvia species and how their chemical diversity is affected by environmental factors such as altitude can help inform both their local medicinal use as well as conservation policies and efforts of the threatened species.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.