**3. Results**

#### *3.1. Spatial Variability*

The 14 plants analyzed to assess the spatial variability of floral scent showed the presence of 70 compounds with a high variability between samples (min 14; max 28; mean 19.4). Remarkably, no compound was found to be present in all the samples. This variability has been observed considering both the composition and abundance of VOCs. Ethyl dodecanoate was the most frequent compound; it was determined in 11 samples. Hexadecane and β-bisabolene (Figure 4a) were detected in 10 samples, δ-selinene and β-sesquiphellandrene (Figure 4e) were detected in 9 samples, whereas caryophyllene (Figure 4b), *cis*-<sup>α</sup>-bergamotene (Figure 4f), and heptadecane were found in 8 samples. Considering the most abundant compounds, each sample afforded a different result. Verbenone was the main component in Pisticci 1 and β-sesquiphellandrene in Pisticci 2; Calciano gave pristane as prevalent; in Sant'Arcangelo, the main components were alternatively α-zingiberene, verbenone, and pristane. Caryophyllene was the principal component in Tolve 1, *i*-propyl 14-methyl-pentadecanoate was the main component in Tolve 2, whereas farnesol has the same role in Tolve 3. The Pomarico sample gave *p*-menth-8-en-1-ol as a main component (Figure 4c); in the two samples of Vietri, we found β-sesquiphellandrene and citronellol (Figure 4d) as dominant, whereas the Potenza plant gave pristane as the most abundant compound. β-Sesquiphellandrene was also the main component in the sample of Savoia. Only two compounds were found as dominant in more than two samples: pristane in S. Arcangelo 3, Calciano and Potenza and β-sesquiphellandrene in Pisticci 2, Vietri 1 and Savoia. Notably, except for Vietri 1 and Savoia, these groups comprise samples located rather distantly geographically. Coherently, the overall similarities between samples were rather low, with a mean value of 18.97 ± 1.6 SE (min 0.96, max 60.94). The full VOC compositions for each sample are reported in Table 1.

The NMDS ordination resulted in a two-dimensional solution with a final stress of 0.14 (Figure 5a). The VOCs most strongly correlated with the first axis were β-sesquiphellandrene (positively) and longipinene (negatively), whereas *p*-menth-8-en-1-ol (positively) and *i*-propyl 14-methyl-pentadecanoate (negatively) were the most correlated compounds with the second axis.

We did not identify a clear geographical structure in the dataset, except for the samples from Savoia and Vietri, which seemed to group together; these are also very close to each other geographically (Figure 1), and there was a weak correlation between axis 1 and altitude. However, in the NMDS, most of the samples coming from nearby locations (e.g., Tolve 1 vs. Tolve 3 and Pisticci 1 vs. Pisticci 2) were strongly separated along the two axes, and the position of the Potenza sample, located at the higher altitude, is not coherent.

Furthermore, in the hierarchical clustering (Figure 5b) within the subcluster including samples from Vietri and Savoia a sample from Pisticci was also unexpectedly included, the locality placed at the maximum geographical distance from Vietri and Savoia (Figure 1). The lack of a significative geographical driver underlying our dataset was also confirmed by the results of the Mantel test, which showed that there is no correlation between composition in VOCs and geographical distances between samples (Mantel test: r = 0.17; *p* = 0.19).

The same result was obtained from analyzing the correlation between VOC composition and the environmental distances (i.e., ecological niche) of sample sites (Mantel test: r = 0.10; *p* = 0.28).

#### *3.2. Temporal Variability*

As for temporal differences in VOCs composition during the flowering phase, the comparison between floral volatiles emitted at the beginning and at the end of the flowering exhibited the presence of a similar number of compounds (min 15; max 22; mean 18.3). The mean value is in agreemen<sup>t</sup> with findings from the sampling performed to assess the spatial variability, where a mean value of 19.4 compounds was found. The VOC compositions for each sample are reported in Table 2. In particular, comparing the two flowering phases, we found the same number of compounds in sample 2 and a slight decrease in sample 1 at the end of the flowering. Additionally, in this case, despite the fact they come from the same population (S. Arcangelo), composition quantitative analysis confirmed the presence of a strong association between individuals' variability. In sample 1, the two most abundant compounds were caryophyllene and 2,3-dihydrofarnesol, whereas in sample 2, they were citronellol and β-sesquiphellandrene. Caryophyllene remained the dominant compound in the sample 1 even at the end of flowering, whereas the abundance of 2,3-dihydrofarnesol strongly decreased; then, the second most abundant compound became 4-methyltetradecane. In sample 2, instead, the dominance between the first two compounds was reversed at the end of the flowering, with β-sesquiphellandrene becoming the most abundant compound followed by citronellol. The decrease in the number of compounds in sample 1 (from 22 to 15) is linked to the non-detection at the end of flowering of various compounds which, previously, had low abundance (area% < 2), whereas only nonadecane, with an area value of 0.37%, was found in addition at the end of flowering. On the other hand, in sample 2, five compounds were no longer identified at the end of flowering, replaced by an equal number of compounds not present in the first flowering phase. Among these, α-terpineol acetate stands out for its abundance, characterized with a value of 13.83% in sample 2 at the end of flowering.

**Figure 4.** Mass spectra of (**a**) β-bisabolene; (**b**) caryophyllene (**c**) *p*-menth-8-en-1-ol; (**d**) citronellol; (**e**) β-sesquiphellandrene; (**f**) *cis*-<sup>α</sup>-bergamottene.


#### *Compounds* **2022**, *2*


**Table 1.** *Cont.*

**Figure 5.** Multivariate analysis of VOCs emitted by flowers of *Barlia robertiana* based on the betweensamples Bray–Curtis similarity: (**a**) non-metric dimensional scaling; (**b**) UPGMA hierarchical clustering. Labels of samples are accordance with Figure 1.


**Table 2.** Comparison of VOCs emitted by two plant samples of *Barlia robertiana* from the S. Arcangelo site (Basilicata region, S-Italy) detected during the first part of the flowering phase with respect to the last part of the flowering.

As for inter-annual variations in VOC compositions, unexpectedly, a huge difference was found in the pairwise comparison of the three plants consecutively analyzed in 2018 and 2019 (Table 3). These plants came from the populations of Calciano, Potenza and S. Arcangelo 2; the 2018 samples were also used for spatial variability analyses. The plant from Calciano, as described above, had pristane and ethyl tetradecanoate as main

components; surprisingly, these two compounds were completely absent from the sample of the following year, substituted by α-terpinolene and *(E)*-β-farnesene. Additionally, the other two individuals analyzed gave similar results for the main components: in the plant from Potenza, verbenone and D-carvone were not detected in the second year of sampling, which gave pristane and farnesol as the two dominant VOCs. Regarding the plant from S. Arcangelo, verbenone, the main component in 2018, was almost absent in 2019 (0.81%), whereas δ-selinene, which was the second most abundant, was no more detected. Instead, the main component in 2019 resulted citronellol and β-sesquiphellandrene. Consequently, the inter-annual between-sample similarity, measured through the Bray–Curtis index, reached only very low values, lower than 9% (Calciano 4.6%, Arcangelo 5.1%, Potenza 8.9).

**Table 3.** Inter-annual variations in VOCs emitted by three individuals of *Barlia robertiana* sampled in two consecutive years.



**Table 3.** *Cont.*

## **4. Discussion**

In this study, we identified a very high number of VOCs emitted by *Barlia robertiana*: considering all the analyzed samples, more than 100 compounds were identified. These results largely encompass the findings of our preliminary study [31].

As for VOC compositions, notably, in a study performed in Spain [32], Gallego et al. found α-pinene, β-pinene, and limonene as the main components of the floral scent of *B. robertiana*. In our samples, instead, α-pinene and limonene were detected, and only at low percentages, and they were not always present (Table 1), whereas β-pinene was found

with even less frequency and lower percentages. On the other hand, the compounds that most characterize the plants from Italian populations have been not detected in the samples carried out in Spain or, therein, they were only present with very low values; this was the case with verbenone, for example. Considering the influence that analytical tools and sampling procedures may have on the results obtained from VOC analysis, we argue that most of detected differences in the floral scents of *B. robertiana* are due to an intrinsic extreme capability of this species to vary its floral emissions, both qualitatively and quantitatively. At present, the data collected do not allow hypothesizing how much of this variability is under genetic control and how much depends on contingent environmental conditions. What it was possible to ascertain in this study, confirming the preliminary data shown by [31] based on a minor number of samples, is that the variability of floral scents is not related to the geographical distance between populations, nor to the main environmental characteristics of the growth sites. However, even if based on only three samples, the observed variation of the VOCs emitted by the same individual in two consecutive years would seem to indicate a poor genetic determinism for this phenomenon. The mechanisms behind this variability could be related to an intrinsic plasticity of metabolic pathways that lead to the synthesis of VOCs in *B. robertiana*, but no studies have specifically investigated this aspect thus far.

Several of the detected compounds, such as verbenone and α-zingiberene, are known to act as pheromones [33,34]; however, this specific function probably is not specifically used by *B. robertiana*. These results may sugges<sup>t</sup> that there is no adaptation of floral scent to local environments or specific communities of pollinators. In fact, the wide spectrum of VOCs emitted can allow *B. robertiana* to attract different species of insects also belonging to very distant taxonomic groups, as evidenced by some studies on its pollinators [24,25] relying on a large plethora of possible pollinators. The strategy of *B. robertiana* to attract pollinators manly involves early flowering, showiness and long-lasting inflorescence, traits that can be advantageous for exploiting the first insects that emerged from winter hibernation. Floral scent, a key trait for interaction between plants and insects [35,36], plays an important role for floral mimicry in deceptive species. In this context, a huge variation in flora scent, such as that highlighted by *B. robertiana*, can be considered an effective strategy for a rewardless, but allogamous, species to avoid that visiting insect learn to avoid such flowers. In fact, some studies [37] has been highlighted as rewardlessness can be a dangerous strategy [38]. Despite the causes of rewardlessness are still little known, just a study on *B. robertiana* showed for the first time the reproductive advantage of the lack of nectar [24]. However, it must be stressed again that this reproductive advantage can only occur if pollinating insects do not learn to associate the floral signals of a species with the lack of nectar inside the flowers.
