Morphological and Chemical Diversity and Antioxidant Capacity of the Service Tree (Sorbus domestica L.) Fruits from Two Eco-Geographical Regions

Service tree, Sorbus domestica L., is a rare and neglected wild fruit tree species of southern and central Europe. Being distributed in different eco-geographical regions, with fragmented and low-density populations, S. domestica represents an interesting model case for investigating patterns of within- and between-population diversity at geographical and environmental scales. This study aimed to analyze the proximate composition, antioxidant activity, and morphometric fruit characteristics. We examined the diversity and population divergences of 49 S. domestica individuals originating from seven populations across continental and Mediterranean eco-geographical regions. In addition, tests of isolation by distance and environment were performed to detect the magnitude of divergence explained by geographic and environmental variables. Significant differences between the studied populations were found in almost all of the studied morphometric and chemical fruit characteristics. The studied service tree populations were characterized by high phenotypic variation despite the low number of trees per population. Model-based population structure analysis using morphometric and chemical fruit characteristics revealed three groups of service tree populations. We concluded that non-effective pollen and seed dispersal along with genetic drift and specific environmental factors resulted in a distinct phenotype with a specific chemical composition in the isolated island population. In addition, a pattern of isolation by the environment was revealed. We infer that morphological and chemical differences between the studied populations in the true service tree from different eco-geographical regions were mediated by adaptation to the specific environmental conditions.


Introduction
Forest fruit trees are species that play an important role in maintaining biodiversity, by enriching the gene pool of the forest ecosystem and with other trees and shrubs, ensuring the hardiness and health of the whole forest, as well as increasing the soil quality [1]. Due to their adaptability and resilience, the promotion of forest fruit tree species can be considered as one of the most promising efforts of climate change mitigation, and planting them is Chemical differentiation was examined using ten chemical characteristics, and morphological variation was assessed for ten fruit traits. Finally, a variety of multivariate analyses and Mantel tests were used to examine the roles of geographic and environmental isolation in determining the morphological and chemical diversity in wild fruit tree species.

Climate Differences among Sampling Sites
Nineteen bioclimatic variables were used to describe the environmental differences between seven studied populations (Table S1). The first principal component (PC1) explained 65.5% of the total variation and clearly separated the continental (P01 and P02) and Mediterranean (P03-P07) populations ( Figure S1 and Table S2). The continental region is characterized by cold winters, with a more even distribution of rainfall throughout the year. The annual mean temperature is lower than that in the Mediterranean region, as well as the mean temperature of the coldest month. In contrast, the Mediterranean region features hot to very hot summers and mild winters. A distinctive feature of the Mediterranean region is the seasonality of rainfall distribution, with the majority of precipitation in the coldest quarter of the year.

Fruit Morphometric Characteristics
Based on all parameters of fruit size, the continental populations were shown to have larger fruits than the Mediterranean populations (Table 1, Figure S2). Fruit mass had a mean value of 7.32 g, and it was the second most variable trait, with a mean coefficient of variation of 36.04%. The least variable traits were fruit (FL) and seed (SL) length and fruit (MFW) and seed (SW) width.
Fruit width and fruit length ratio (FW/FL) values for the continental populations and Mediterranean population Istria were approximately 1.0, and for other south Mediterranean populations, 1.1. Although fruits of slightly oblong shape, e.g., pear-like circular or oval, characterized southern Mediterranean populations, and apple-like shape continental and north Mediterranean population Istria, both types of fruits were found in all studied populations. Fruit shape was a far less variable characteristic, with a mean CV of 10.17%.
Populations from the continental region had wider seeds than the populations from the Mediterranean eco-geographical region. The mean seed number (NS) varied from 1.25 in population Novi Vinodolski to 3.66 in population Brač, with an overall mean value of 1.88. Seed number was by far the most variable trait, with a CV of 60.06%. In general, Mediterranean populations were characterized by a larger number of smaller and more variable seeds.
All fruit and seed parameters were significantly different between individual trees and populations, except fruit width and fruit length ratio (FW/FL) and seed length (SL) ( Table 2). The majority of tested parameters were characterized by having most of the variability linked to intra-population variability. Table 1. Arithmetic means (M) and coefficient of variations (CV/%) for fruit and seed morphometric characteristics. Different letters within one column denote statistically significant differences (p < 0.05) by Fisher's LSD test. Acronyms of populations: P01-Psunj; P02-Tounj; P03-Istria; P04-Novi Vinodolski; P05-Split; P06-Brač; P07-Konavle. Fruit and seed morphometric characteristics: m (g)-fruit mass; FL (cm)-fruit length; MFW (cm)-maximum fruit width; PMFW (cm)-fruit length, measured from the fruit base to the point of maximum fruit width; FW1 (cm)-fruit width at 15% of fruit length; FW2 (cm)-fruit width at 85% of fruit length; FW/FL-maximum fruit width/fruit length; SL (mm)-seed length; SW (mm)-maximum seed width; NS-number of seeds per fruit. P01 9.53f 27. 34 Table 2. Results of the hierarchical analysis of variance for fruit and seed morphometric characteristics. Acronyms for fruit and seed morphometric characteristics as in Table 1.

Proximate Analysis and Acidity
The analyzed service tree fruits were characterized by high water content and sugar levels and very low fat and acidity values (Table 3). Differences between the studied populations were confirmed for all studied proximate constitutes, except for acidity. The mean water content value was 65.14 g per 100 g dm. Values were relatively uniform, and the average coefficient of variation was low (CV = 6.70%). In general, Mediterranean populations had somewhat lower water content values than continental populations. Sugar content had a mean value of 49.36 g per 100 g dm and a coefficient of variation of 9.33%. The lowest values were found in population Brač (42.88 g per 100 g dm), and the highest in population Novi Vinodolski (55.60 g per 100 g dm). The mean protein value was 14.69 g per 100 g dm. Protein content was characterized by relatively high variability, with a mean coefficient of variation of 20.27%. Both ash and fat content were very low, with mean values of 2.12 and 0.72 g per 100 g dm, respectively. Both of these parameters were very variable, with mean coefficients of variation for fat of 19.52%, and ash content being the most variable parameter in general, with a coefficient of variation of 34.23%. Mediterranean populations were characterized by significantly higher fat content than the continental populations. Acidity values ranged from 0.64 to 0.74, with a mean value of 0.68. The coefficient of variations showed intermediate variability of data, with a mean value of 14.90% and a range of 5.24-28.20%. The cellulose content varied between the studied populations without a clear pattern. The average cellulose content was 6.24 g per 100 g dm, with a mean coefficient of variation of 27.95%. The population Brač was characterized by the highest cellulose content, 9.06 g per 100 g dm.

Total Phenolics and Antioxidant Capacity
Total phenolic content mean values ranged from 3.50 mg GAE/g dm to 12.10 mg GAE/g dm (Table 3). Continental populations have the highest values, 11.11 mg GAE/g dm (Psunj) and 12.10 mg GAE/g dm (Tounj), whereas island population Brač and population Split have the lowest, 3.50 mg GAE/g dm and 5.62 mg GAE/g dm, respectively. Populations Psunj and Brač have the highest coefficient of variation, 63.19%, and 57.44%, while population Tounj has the lowest variability of 16.80%, followed by population Istria, 38.23%.
Measured DPPH values ranged from 4.40 to 17.76%. Populations Istria (17.76%) and Split (13.76%) have the highest values, and Brač (11.70%) and Tounj (4.40%) have the lowest. Measured DPPH values were less variable than those of total phenolic content, with a coefficient of variation of 44.00%. Mean FRAP values ranged from 2.41 to 2.91 mmol Fe 2+ . The highest value was calculated for population Tounj, and the lowest for population Psunj.

Population Structure, Isolation by Distance, and Environment
The structure of the seven service tree populations was inferred by the K-means clustering method. The most probable division was detected at K = 3, and the estimated population structure is shown in Figures 1A and 2A. If the proportion of a certain population was equal to or higher than 0.70, it was assumed that the population belonged to one cluster, and if it was lower than 0.70, it was assumed that the population had a mixed origin. In both analyses, morphometric and chemical, we revealed the same geographical pattern. The samples from the populations Psunj and Tounj grouped into cluster A, the samples from island population Brač grouped into cluster B, and the samples from the coastal Mediterranean populations into cluster C. Only one population, P04 (Novi Vinodolski), was of mixed origin, with the dominant proportion from cluster C. The results obtained with the K-means clustering method were congruent with the results from the hierarchical clustering method (Figures 1B and 2B) and Barrier software ( Figure 3A,B). The existence of barriers was revealed between the continental and Mediterranean populations, as well as between the island population Brač and all other studied populations. Table 3. Arithmetic means (M) and coefficient of variations (CV/%) for water (g/100 g), crude protein (g/100 g), sugar (g/100 g), ash (g/100 g), crude fat (g/100 g), cellulose (g/100 g), acidity (%), total phenols (mg GAE/g), and antioxidant activity (DPPH %; FRAP mmol Fe 2+ ). All mass fractions were determined on a dry mass basis. Different letters within one column denote statistically significant differences (p < 0.05) by ANOVA and Fisher's LSD test. Acronyms of populations: P01-Psunj; P02-Tounj; P03-Istria; P04-Novi Vinodolski; P05-Split; P06-Brač; P07-Konavle. The first two components from a PC analysis of the phenotypic traits explained 59.2% and 20.6% of the total variation, respectively (Table S3 and Figure 4). PC1 was highly negatively correlated with the fruit size, and moderately positively with the seed number. On the other hand, PC2 was highly negatively correlated with the fruit shape, and moderately positively correlated with the seed size. Continental populations were associated with larger fruits with a smaller number of seeds, and Mediterranean populations with smaller fruits with a higher number of seeds.    PC analysis of all individuals showed that 50.2% of observed chemical variability was explained by the first two principal components (Table S4 and Figure 5). The first principal component (PC1) was strongly negatively correlated with the water and sugar content, and strongly positively with the ash and cellulose content. Strong negative correlations were observed between the second principal component (PC2) and the fat content and DPPH. Clear differences were revealed between the samples from the island Brač and other studied populations. Discriminant analysis was performed to determine which of the morphological and chemical traits were the most useful for maximum discrimination between the three groups of service tree populations established by the K-means clustering method. Four out of nine morphological traits (number of seeds, fruit length, position of maximum fruit width, fruit width at 15% of fruit length), and four out of ten chemical characteristics (water, crude fat, cellulose, DPPH), were determined by stepwise discriminant analysis to be the best differentiating variables between the studied groups of service tree populations (Tables S5 and S6). The discriminant function based on morphometric and chemical traits showed a classification success of 87.75% and 97.96%, respectively ( Figures 1C and 2C). Overall, these results confirmed the usefulness of morphometric and chemical fruit traits in discrimination of the studied groups of service tree populations.
Neither morphological nor chemical distances were related to geographical distances ( Figure 6A,C). On the other hand, there was an overall relationship between pairwise phenotypic and environmental distances matrices, suggesting a pattern of isolation by environment ( Figure 6B). Although we did not find significant correlations between pairwise chemical and environmental distances matrices ( Figure 6D), several chemical traits were closely related to the environmental variables, i.e., Mediterranean populations were characterized by lower water and total phenolic contents, and higher fat contents in comparison with the continental populations. In addition, the Mantel test identified significant correlations between the morphological and chemical distance matrices ( Figure 6E).
In general, populations from the continental eco-geographical region had larger fruits than populations from the Mediterranean region. The values of particular morphological traits found in our study did not differ considerably from the values reported in botany and dendrology textbooks [16,[40][41][42][43] or morphometric studies [18][19][20][21][22]. However, the range reported by other authors was significantly wider than the range obtained in this study, i.e., the maximum values for fruit size, noted by other authors, were somewhat higher than those obtained in this research. In previously published papers, analyzed fruits were collected in wild populations and orchards. Our samples, however, were collected only from wild populations, which most likely meant less human interference and suboptimal growing conditions. Interestingly, the levels of diversity in cultivated populations were similar to those in our study for the wild populations. High diversity in cultivated populations was probably preserved due to the fact that true service tree has never been subjected to selective breeding intensively, even though it has often been grown as a fruit tree [28]. According to Bignami [44], it was only propagated generatively by seeds for centuries and therefore preserved wide and underexploited variation in orchards as well as in natural populations.
The results of the analysis of variance (ANOVA) were in line with the expectations of high morphological variation within populations and low differentiation between populations, as observed in rowans [28,29,31] and other insect-pollinated and animal-dispersed tree species [45][46][47][48]. The relatively high level of among-tree variation within the populations is probably a result of both phenotypic plasticity to specific microenvironmental conditions experienced by each tree, and genetic differentiation among individual trees [28,29,49]. The within-population variation was of similar magnitude in all studied populations, except for the island population Brač. Even though isolated island populations could be expected to have significantly lower variability compared to the continental populations [50][51][52], island population Brač demonstrated the highest variability levels. This is more likely linked to the microsite features of its habitat than to genetic variability. Habitat conditions in which island samples were collected varied greatly, ranging from deeper soils to shallow, skeletal soils, stretched across varying altitudes. This indicates that phenotypic plasticity might be of greater importance with regard to microhabitat variation within service tree populations [53]. In addition, the island population Brač was morphologically the most distinguished population out of those analyzed. This population was characterized by having the smallest fruit with the highest number of seeds per fruit.

Proximate Composition
To the best of our knowledge, this is the first report on the proximate chemical composition of service tree fruits. Service tree fruits are sweet due to their high sugar content, are rich in proteins, and have a high content of cellulose, which acts as dietary fiber and as such can enrich a diet. We revealed that the parameters with the highest mean values were the least variable, e.g., water and sugar, and the most variable parameters were those with the lowest values-ash, cellulose, fat, and acidity. In comparison with the closely related rowan (Sorbus aucuparia L.), service tree fruits are characterized by lower water content, significantly higher levels of carbohydrates and cellulose, and similar, very low, fat content [54][55][56]. Results of the chemical composition analysis in this paper demonstrate great potential for further research on the specifics and benefits of service tree fruits and indicates potential future uses of fruits in both fresh and processed food supplements and products, like low-fat, protein-rich food.
This research has found significant differences between the populations in the analyzed chemical composition of service tree fruits. In general, the continental populations have somewhat higher water content and significantly lower fat values from cuticular wax. The differences between continental and Mediterranean populations are readily explicable by the specific ecological conditions in which populations grow. The Mediterranean populations are exposed to drought and extremely high temperatures, meaning their fruit need protection from the thicker cuticular wax, which acts as a shield against desiccation. It is reported that this protective layer in many plant species is a key evolutionary innovation in plants [57], and it has been implicated in protection mechanisms against different environmental stressors, such as excessive ultraviolet radiation, high temperatures, and salinity, low temperature during the vegetation season, etc. The current relationships between fruit chemical composition and environmental factors in S. domestica may be the result of long-term adaptation to different habitats.
Similar to morphological variability findings, intrapopulation variability was similar among all analyzed populations, with population Brač being the most variable. In addition, Brač was also the most specific population in terms of chemical composition. This island population was characterized by the lowest water content and highest ash and cellulose contents.

Total Phenolic Content and Antioxidant Activity
Several studies have demonstrated the importance of service tree fruits as a good source of different bioactive phytochemicals [26,41,58,59]. It was reported that those constitutes are mostly affected by geographic origin and genotype [26,60], climatic environment [61][62][63], and stage of maturity [64][65][66]. In general, it is well known that phenolic compounds represent a chemical interface between plants and the environment [67] and are under the influence of various environmental factors, including soil composition, temperature, and rainfall [68].
Significant differences in total phenols were found between the Mediterranean and continental populations, with higher values observed in the continental populations. This was somewhat opposite to our expectations since higher total phenolic content is expected in the sites where water deficit, along with high temperatures and increased UV radiation, causes stress in plants [69][70][71]. When subjected to environmental stress, plants respond by increasing the production of secondary metabolites, including phenolic compounds. The effect of moderate drought stress on the increased production of these compounds has been reported for several fruit species, namely Punica granatum L. [72][73][74], Olea europaea L. [75,76], and Prunus dulcis (Mill.) D.A. Webb [77]. Increased production of antioxidants and secondary metabolites, including phenols, neutralizes reactive oxygen and radicals [78]. However, prolonged drought leads to a reduction in the overall content of the metabolites due to greater growth reductions [71]. This could explain low phenolic values in the southernmost populations, island Brač, Split, and Konavle, since these populations grow in the Mediterranean climate, with prolonged drought and intense heat during the summer months. On the other hand, S. domestica is a tree species that prefers the Mediterranean climate. Therefore, in the continental eco-geographical region, where service tree grows in sub-optimal sites, adverse conditions that are more likely to become extreme could result in a higher total phenolic content in the fruits from continental populations. These adverse conditions can also be notable in pedoclimate, i.e., soil type. The continental populations grow on shallow, extremely dry, and basic soils, with carbonate, mostly marl-type parent material. The Mediterranean populations, although also facing shallow and dry conditions, grow on terra rossa of flysch or dolomite parent material with clayey texture, enabling better water retention and distribution, despite the increased amount of soil skeleton.
Measured DPPH values did not follow the same trend as total phenolic content. For example, population Tounj was characterized by the highest total phenolic content values but lowest DPPH, whereas the values for Psunj were very similar. DPPH free radical scavenging is a method for the evaluation of the antioxidant activity of all compounds in a sample, e.g., plant tissue. Natural antioxidants in plants include polyphenols (phenolic acids, flavonoids, anthocyanins, lignans, and stilbenes), carotenoids (xanthophylls and carotenes), and vitamins (vitamin E and C) [79]. Research on S. domestica has largely been focused on polyphenols, thus providing only a partial picture of all the components with antioxidative abilities and their relative contents. Apart from polyphenols, the biochemical composition of the fruits of S. domestica is characterized by 4 to 22% of carotenoids, 22.3 to 98.3% of ascorbic acid, as well as 1.8% of vitamin C [80]. It is therefore possible to assume phenols compromise only a small proportion of antioxidative compounds in S. domestica fruits, which would explain the variable data and relations between total phenolic content and DPPH assays. Furthermore, FRAP has not been shown to have a significant correlation with geographical or environmental distribution and is overall not significant as a parameter of distinction.

Population Structure, Isolation by Distance and Environment
Our results provide new insights into the morphological and chemical diversity and structure of service tree populations from two eco-geographical regions. In both cases, the studied populations were dived into three groups. The first group encompasses populations from the continental eco-geographical region, the second group samples from the island population Brač, and the third group populations from the coastal Mediterranean area. In both cases, island population Brač was specific, compared to other populations included in this research. Overall, these findings are consistent with the well-known fact that island populations are distinct because they have a unique array of traits/genes when compared with the mainland populations [81,82]. Scattered distribution and geographical isolation between those populations have probably resulted in a limited long-distance gene flow. Under such conditions, genetic drift may play a very important role in shaping the structure of the populations [28,[83][84][85]. This microevolutionary process is especially pronounced in such low-density populations. Similar results were reported for leaf morphometric analysis of service tree populations [29]. Island populations were highly divergent in leaflets shape in comparison with adjacent mainland populations. However, it is reported that in temperate, insect-pollinated, and animal-dispersed tree species such as S. domestica [17,86,87] and S. torminalis [88][89][90], fragmented subpopulations are functionally connected by gene flow through both pollen and seed. This can explain connectivity between other coastal populations since they were in both cases classified in the same group. Also, we cannot exclude the possibility of human-mediated gene-flow as this species has been cultivated since Roman times [8,9]. Nevertheless, the clear pattern of morphological and chemical diversity indicates a natural origin of the studied populations.
The 'mixed origin' of the Novi Vinodolski population (P04) is of particular interest because our data indicate a certain gene flow between this population and the populations in the continental eco-geographical region. The Northern Adriatic region, stretching from Novi Vinodolski to Senj, has been linked to Central Croatia throughout history by trade and the exchange of goods. It is, therefore, safe to assume human-mediated transplanting and exchange of plant material happened between the continental population of Tounj and the Mediterranean population Novi Vinodolski, which would explain the mixed origin of the Novi Vinodolski population.
The tests for isolation-by-distance were not significant for the studied populations, and the morphological structure between the Mediterranean and continental populations was largely explained by the environmental conditions and fits the isolation-by-environment pattern. In addition, we revealed that water, crude fat, and total phenolic contents of the fruits were associated with environmental variables. Overall, our results suggest that populations from the continental and Mediterranean eco-geographical regions, despite their admixture in some cases, are isolated by the environment. It is well known that environmental heterogeneity can have a fundamental impact on phenotypic diversity [91][92][93]. In particular, different environments can result in spatially variable selection pressures, thereby contributing to phenotypic divergence among populations via phenotypic plasticity or local adaptation [92,94]. Lower annual mean temperatures accompanied by high precipitation during the warmest quarter in the continental eco-geographical region resulted in larger fruits with higher water content. On the other hand, higher annual temperatures, especially during the warmest quarter, resulted in smaller fruits with lower water content and higher fat content from cuticular wax. Overall, our results indicate that multiple morpho-and chemo-types, each having higher fitness in its native habitat than the others, are probably the result of local adaptation [95].

Plant Material and Study Area
The study encompassed two populations from the continental eco-geographical region (P01-Psunj, P02-Tounj), and five populations from the Mediterranean eco-geographical region (P03-Istria, P04-Novi Vinodolski; P05-Split; P06-Brač; P07-Konavle). In each population, samples were collected from seven trees, which were then used in morphometric and proximate analyses as well as studies of total phenolics and antioxidant activity. The small number of sampled trees is a direct result of the small density of populations [6,8,9,29], i.e., a small number of mature trees spread across large areas in the habitat. This fragmentation made collecting challenging and caused a smaller overall sample size.
To describe the environmental differences of the studied populations, for the principal component (PC) analysis and the calculation of environmental distances, 19 bioclimatic variables were obtained for each collection site.

Morphometric Analysis
In order to conduct the morphological analysis, 50 fruits were collected from each tree. Firstly, fruit mass (m) was determined with 0.1 g precision weighing. After weighing, fruits were cut longitudinally, and the dimensions were measured automatically using the WinFolia software [96]. The accuracy of measurements was 0.1 mm, and the following morphological characteristics were measured: fruit length (FL); fruit length, measured from the fruit base to the point of maximum fruit width (PMFW); maximum fruit width (MFW); and fruit width at 15% (FW1) and 85% (FW2) of fruit length. After the fruits were measured, seeds were extracted, the pulp removed, and the seed dimensions, i.e., length (SL) and width (SW), were measured automatically using the WinSEEDLE software [97]. In addition, the number of seeds per fruit (NS) was counted manually. Finally, ten morphological characters were examined in order to assess the variation within and between populations.

Proximate Analysis
Water, crude protein, crude fat, ash, cellulose, and sugar contents were determined according to the procedures established by the Association of Official Analytical Chemists (AOAC). All analyses were performed in duplicate. Water content in the samples was determined by a physical, indirect method, in which a sample of known mass was dried in an air dryer (Instrumentaria, Zagreb, Croatia) at 105 • C until a constant weight was achieved (4 h) [98]. Total mineral content was measured as ash content, which represented inorganic fruit compounds left after the organic matter was combusted [99]. The 4 g fruit samples, previously carbonized on open gas burners, were ignited in a muffle furnace (Nabertherm GmbH, Lilienthal, Germany) at 580 • C, until a constant weight was achieved, i.e., until uniform, light-grey ash with no black admixtures formed. The Kjeldahl method was employed to determine the total nitrogen content, in combination with a copper catalyst using the block digestion system Foss Tecator 6-1007 Digestor (Foss Tecator, Höganäs, Sweden) and the Foss Kjeltec™ 8100 Auto Distillation unit (Foss Tecator, Höganäs, Sweden). Crude protein content was obtained by multiplying total nitrogen by a conversion factor of 6.25 [100]. Total crude fat extraction was performed by the Soxhlet apparatus (Inkolab d.o.o., Zagreb, Croatia); medical-grade petroleum was used for extraction during 5 h [101]. Crude cellulose was determined by the method of Kürschner and Hanak [102]. Sugar content in the fruits was determined according to the AOAC 925.35 [103] method. Reducing sugars, glucose and fructose were determined using Fehling solution A, which was reduced to copper (I) oxide by the sugars, under specific conditions. Non-reducing sugar, e.g., sucrose, was firstly hydrolyzed down to reducing sugars, whose content was determined using Fehling solutions and used to calculate the total sugar content (total inverted sugars).

Fruit Extraction Procedure
Extraction was conducted according to Benvenuti et al. [104]. After the seeds were removed, fruits were finely chopped and 6 g (±0,01) were placed in a 250 mL Erlenmeyer flask. 20 mL of methanol (MeOH)/2% hydrochloric acid (HCl) (95:5) extraction solvent was added. The flask was sealed, wrapped in aluminum foil, and left for 1 h on the shaker. After 1 h on the shaker, extracts were filtered in a Büchner flask, using vacuum, and transferred into 50 mL volumetric flasks. The remaining extract in the Büchner flask was rinsed with a few mL of extraction solvent. Residue from the Büchner funnel was transferred into the same 250 mL Erlenmeyer flask and the procedure was repeated, resulting in a double volume of extract in the 50 mL volumetric flask. Flask was filled with extraction solvent up to the 50 mL mark and left refrigerated, at +4 • C, until further analysis. These extracts were used in determining the total phenolic content, the antioxidant capacity assay by DPPH method, as well as the reducing power assay using the FRAP method.

Estimation of Total Phenolic Content Using the Folin-Ciocalteu Method
Total phenolic content in the samples was determined according to the Folin-Ciocalteau method [105]. This colorimetric method is based on the reduction of the phosphotungsticphosphomolybdenum acid complex, creating a blue chromophore. Maximum absorption of the chromophores depends on the alkaline solution and the concentration of phenolic compounds, with a more intense blue color indicating higher absorption and concentration of antioxidants [106].
In a tube, 40 µL of diluted extract sample, 3.16 mL of distilled water, and 0.2 mL FC reagent were added and vortexed. After 3 min, 0.6 mL of aqueous sodium carbonate (20%) were added, the sample was vortexed again and placed in a water bath at 40 • C. After 30 min, the absorbance was read, at 765 nm. Total phenol concentration was calculated from the calibration curve, where gallic acid was the standard. Results were expressed as mg of gallic acid per g of dried fruit matter (seeds excluded) [24].

Evaluation of Antioxidant Activity Using the DPPH Method
DPPH method is based on the reduction of stable radical 2,2-Diphenyil-picrylhydrazyl (DPPH*) in the presence of antioxidants. Due to its unpaired electron, DPPH* strongly absorbs in the spectrum of 517 nm [107]. Reduction causes stable, purple-colored DPPH* to change into yellow-colored DPPH-H [108][109][110]. Phenol compounds act as hydrogen donors, thus making this method suitable as an antioxidant assay.
A volume of 2 mL of diluted extract, 2 mL of methanol, and 1 mL of 0.5 mM DPPH methanolic solution were placed into a tube and shaken on a shaker. The blank was methanol without DPPH. Closed tubes were then kept in the dark, at room temperature, for 20 min. Absorbance was measured at a wavelength of 517 nm, against the blank of DPPH-free methanol. The DPPH* concentration in the reaction medium was calculated from the calibration curve for Trolox. Results were expressed as a percentage (%) of residual DPPH.

Determination of FRAP
Ferric reducing antioxidant power (FRAP) assay is a method in which yellow-colored Fe 3+ -2,4,6-tripyridyl-s-triazine (TPTZ) complex is reduced to the blue-colored ferrous form, in a low pH-value medium [109]. The FRAP reagent was prepared by mixing 50 mL of acetate buffer with 5 mL of TPTZ and 5 mL of FeCl 3 (10:1:1). For sample analysis, 240 µL of distilled water was mixed with 80 µL of appropriately diluted sample and 80 µL of FRAP reagent. Blank was prepared in the same manner, substituting sample with extraction solution (methanol/HCl 2%). The mixture was left to stand for 5 min in a water bath, at 37 • C. Absorbance was measured at 595 nm. A calibration curve for FeSO 4 ·7H 2 O was used to calculate FRAP values of the measured absorbances. Results were expressed as FRAP values, e.g., mmol Fe 2+ .

Statistical Analysis
Descriptive statistical parameters, arithmetic means, and coefficient of variation were calculated for the particular fruit morphometric characters and chemical composition for the studied populations. To assess the possibility of conducting multivariate statistical analyses and parametric tests, the symmetry, unimodality, and homoscedasticity of data were verified [111]. Assumptions of normality were checked using the Shapiro-Wilk test, and the assumption of homogeneity of variance using Levene's test. A hierarchical analysis of variance (ANOVA) was performed to examine the partition of phenotypic variation between the studied populations and within populations. In addition, one-way ANOVA was used to test the differences in chemical composition between the studied populations. For each analysis, statistically significant differences between means were identified using the Fisher's LSD multiple comparison test, at p ≤ 0.05. Descriptive statistics and ANOVA analyses were carried out using the program package STATISTICA [112].
To evaluate the correlation between multicharacter differences among populations, a Mantel test [113] was performed on the matrices of Euclidean distances. To assess isolation-by-distance and isolation-by-environment, morphological and chemical distance matrices were compared to the geographic and environmental distance matrices using the simple Mantel test [114,115]. To calculate the environmental distance matrix, climate data were obtained from the WorldClim 2 database with a spatial resolution close to a square km [116,117]. In addition, a simple Mantel test was performed between morphological and chemical distance matrices. The significance level was assessed after 10,000 permutations, and the Mantel test was performed with the R package "Vegan" [118].
To identify the divergence and structure of the studied populations, multivariate statistical methods were used [119]. Agglomerative hierarchical clustering algorithms were used to construct a tree diagram. Pairwise Euclidean distances were calculated, and cluster analysis was performed using the unweighted pair-group method with arithmetic mean (UPGMA). The K-means method was applied to detect structure and define the number of K-groups that best explained the morphological and chemical variation of populations (e.g., [49,[120][121][122]). The eco-geographical structure of the studied populations was further analyzed using the Monmonier' maximum difference algorithm, implemented in Barrier 2.2 software [123]. In addition, principal component (PC) analysis was used to calculate principal components across all individuals and all studied morphometric and chemical traits. The biplots were constructed by two principal components showing analyzed individuals and traits. To calculate the discriminatory power of characters among three groups of service tree populations established by the K-means clustering method, a canonical discriminant analysis was performed. The proportion of individuals correctly classified into the above-mentioned groups was determined using classificatory discriminant analyses [118,124]. The input data in multivariate statistical methods were previously standardized, i.e., standardization of characters to zero mean and unit standard deviation was performed prior to each multivariate analysis. The above statistical analyses were conducted using the statistical program R v.3.2.2 [125].

Conclusions
Our study showed high levels of phenotypic and chemical diversity for service tree populations despite its scattered nature and low-density populations. We revealed the existence of three morphologically and chemically distinct and well-defined groups of service tree populations. We suggest that those differences are the result of both neutral and adaptive microevolution processes. With regard to the island population Brač, non-effective pollen and seed dispersal, along with genetic drift and specific environmental factors, resulted in a distinct phenotype with specific chemical characteristics. On the other hand, environmental heterogeneity, i.e., differences between the Mediterranean and continental eco-geographical regions, resulted in an adaptive phenotypic and chemical evolution. Overall, the patterns found in this study confirmed that multiple evolutionary processes influence the morphological and chemical diversity and structure of the populations.  Table S1, Table S2: Pearson's correlation coefficients between 19 bioclimatic variables and scores of the first three principal components. Acronyms for environmental variables as in Table S1, Figure S2: Sorbus domestica fruit samples from seven studied populations, Table S3: Pearson's correlation coefficients between ten morphological traits and scores of the first three principal components, Table S4: Pearson's correlation coefficients between ten chemical traits and scores of the first three principal components, Table S5: Results of the stepwise discriminant analyses for morphometric traits. p(λ), significance of Wilks'λ: *** significant at p < 0.001, ** significant at 0.001 < p < 0.01, * significant at 0.01 < p < 0.05, ns non-significant values (p > 0.05), Table S6: Results of the stepwise discriminant analyses for chemical traits. p(λ), significance of Wilks'λ: *** significant at p < 0.001, ** significant at 0.001 < p < 0.01, * significant at 0.01 < p < 0.05, ns non-significant values (p > 0.05).