Introduction

In the genus Arracacia Bancroft, the number of species growing from Mexico to Bolivia is not definitive according to the results reported by several authors: 24 species are recorded by Constance (1949), 28 species by Hiroe (1979), 55 species by Pimenov and Leonov (1993), and 30 species by Constance (1997). The “World Umbelliferae Database” in 2005 indicates 72 species (http://www.rbge.org.uk/data/URC/Nomenclature). Most of these species names are provisional (Blas 2000). Arracacia genus has been studied mainly in Central and North America (Mexico) (Mathias and Constance 1944, 1968, 1973; Constance and Affolter 1995a, b). In South America, Knudsen (2003) mentioned 10 Arracacia species, distributed in the mountainous Andean region of Venezuela, Colombia, Ecuador, Peru and Bolivia. In this region, the last descriptions of new Arracacia species were reported in Venezuela (A. tilletti Constance and Affolter) and in Colombia (A. colombiana Constance and Affolter) by Constance and Affolter (1995a, b). In Peru, no taxonomical treatment of Arracacia specimens was developed for more than half a century, although six species were reported from this country: A. xanthorrhiza Bancroft, A. peruviana (Wolff) Const., A. andina Britt, A. equatorialis Const., A. incisa Wolff, and A. elata Wolff (Mathias and Constance 1962; Brako and Zarucchi 1993). Of these species, A. xanthorrhiza is the only cultivated arracacha of northern South America. However, at present, no reliable classification of the Arracacia genus from the Andean region is possible due to the poor representation of the above-mentioned species in different herbaria. This is the case for the four taxa: A. xanthorrhiza, A. peruviana, A. andina and A. equatorialis for which the botanical descriptions are incomplete and need further investigations.

This work is aimed at the study of Arracacia genus in Peru with two specific objectives in mind: to select useful characteristics for the identification of Andean species, and to estimate intraspecific and interspecific diversity. Prior to this study, we separated A. equatorialis from A. xanthorrhiza by analyzing original descriptions (Constance 1949; Mathias and Constance 1962) and looking at some representations from different herbaria. Distinction between the two taxa was based upon leaflets incision, shape and margin of involucel. On the other side, the species A. andina and A. peruviana could not be differentiated from A. xanthorrhiza due to the fact that they share similar morphological characteristics. The two species A. elata and A. incisa were clearly differentiated from the other four Peruvian Arracacia species (Blas 2005). A. elata was distinguished by its sprawling habit, spinulose-serrate leaflet margin, and absence of storage roots. A. incisa was identified by its smaller size and ovate to lanceolate, scarious and denticulate-margined involucel, which exceeds the flowers but is shorter than the fruit. On the basis of this preliminary study, four species from Peru were therefore considered in our analysis: A. elata, A. equatorialis, A. incisa and A. xanthorrhiza.

Materials and methods

Plant materials

Natural populations of the four wild species, A. elata, A. incisa, A. equatorialis and A. xanthorrhiza from eight Peruvian Departments, representative of wild arracacha in Peru, were collected and analyzed (Table 1). Dried herbarium were deposited at the “Herbario MOL” of the Universidad Nacional Agraria La Molina (UNALM) and true seeds, dried with silica gel, were deposited at the “Instituto de Biotecnologia” (IBT-UNALM). Seeds with silica gel were placed into sealed plastic bags and maintained at −20°C. For each species, three accessions and 20 plants per accession were sampled from three localities. The number of 20 plants per accession was determined according to a previous study on sample size estimation in Arracacia natural populations (Blas 2005). Taxonomic identification was checked according to the provisional artificial key for Arracacia species, proposed by Constance (1949) and to the comparison of our samples with herbarium material.

Table 1 Arracacia accessions used for morphological and molecular characterization
Full size table

Morphological data

Morphological data of Arracacia species were taken on each area where the accessions were collected (Table 1). The observations on each plant covered 100 (54 qualitative and 46 quantitative) characters previously established by Blas (2005). These characters are indicated in Table 2 and cover the four distinctive morphological parts: the storage roots, the central rootstock, the aerial stems and the leaves (Fig. 1). The morphological descriptors and their states were established according to the provisional key of Arracacia species in South America proposed by Constance (1949), the descriptors proposed by Blas (1998), and additional field observations (Blas 2005). Characters of the aboveground parts were recorded at flowering time while those of the underground parts were recorded immediately after harvest. Color of some plant organs was described according to the color chart of the Royal Horticultural Society (1995).

Table 2 Morphological descriptors used in the diversity analysis of Arracacia wild species
Full size table
Fig. 1
figure 1

Plant of cultivated Arracacia

Full size image

Data analysis

For statistical analysis, data from 20 plants/accession and three accessions/species were used and considered as the Operational Taxonomic Unit (OTU). From data compiled according to the states of descriptors (Table 2), a matrix of 240 plants or OTUs × 100 characters was constructed.

To evaluate the inter- and intra-species variances, descriptors for quantitative traits were analyzed using basic descriptive statistics and variance (ANOVA), with the help of Minitab 13.31 software (2000).

For multivariate analysis, each descriptor was standardized by subtracting the mean value from the observed value and dividing by standard deviation, in order to minimize the scale effects (Crisci and López 1983). Then, an exploratory analysis was elaborated with Principal Component Analysis (PCA) (Dagnelie 1986). A correlation symmetric matrix (100 variables × 100 variables) was calculated in order to understand how the characters were associated and grouped. Eigenvalues and Eigenvectors were determined from this matrix, using the NTSYS 2.1 (Rohlf 2000) and Minitab 13.31 softwares. From these values, the accessions were plotted in order to identify population groups. Canonical Discriminant Analysis (CDA) was made using the procedure CANDISC of SAS system v8 (Palm 2000). The objective was to identify characters with high discrimination capacity for genetic diversity evaluation and to allow identification of Peruvian arracacha species.

AFLP marker analysis

DNA was extracted by the cetyltrimethylammonium bromide (CTAB) method of Doyle and Doyle (1990), adapted and standardized by CIP (1998). For DNA isolation, healthy young leaves were collected from natural populations and dried over silica gel.

The AFLP analysis was carried out according to Vos et al. (1995) and following manufacturer’s instructions with minor modification (CIP 1998). Commercial AFLP kits were purchased from InvitrogenTM Life technologies (Carlsbad, California, USA). A sample of 450 ng genomic DNA was digested in 12.5 μM of 5× reaction buffer [50 mM Tris–HCl (pH 7.5), 50 mM Mg-acetate, 250 mM K-acetate] with 1 μl EcoRI/MseI enzyme combination [1.25 units/μl each in 10 mM Tris–HCl (pH 7.4), 50 mM NaCl, 0.1 mM EDTA, 1 mM DTT, 0.1 mg/ml BSA, 50% glycerol (v/v), 0.1% Triton® X-100]. Following digestion, 12.5 μl of adapter/ligation solution containing 12 μl EcoR I/Mse I adapters, 0.4 mM ATP, 10 mM Tris–HCl (pH 7.5), 50 mM K-acetate and 0.5 μl T4 DNA Ligase (1 unit/μl) was added directly to the DNA digest, incubated 2 h at 20°C, and subsequently diluted 10-fold in 1× TE buffer (10 mM Tris–HCl (pH 8.0), 0.1 mM EDTA).

Preamplification was performed in a total volume of 12.75 μl, containing 1.25 μl of diluted digestion DNA template, 0.125 μM of EcoRI + A and MseI + C primers. PCR was performed in a MJ Research PTC-100 thermocycler (Global Medical Instrumentation Inc., Minnesota, USA) using the following temperature profile: 20 cycles of 30 s at 94°C, 30 s at 56°C and 60 s at 72°C. The PCR products were diluted 10 times with water and used as templates for the selective amplification.

Five AFLP primer combinations were identified as best for selective amplification: E-ACT/M-CAC, E-AAC/M-CTT, E-ACC/M-CTC, E-ACT/M-CTC and E-ACG/M-CTC. Selective amplification was carried out in a volumen of 10 μl containing 3 μl of diluted preselective PCR product, 0.05 μM of EcoRI + ANN and MseI + CNN primers. PCR was performed with the following temperature profile: 12 cycles of 30 s at 94°C, 30 s at 65°C, decreasing −0.7°C/cycle to 56, 60 s at 72°C; followed by 25 cycles of 30 s at 94°C, 30 s at 56°C and 60 s at 72°C. Following amplification, 5 μl of formamide loading buffer (98% formamide, 10 mM EDTA, 0.1% each of xylene cyanol and bromophenol blue) was added, and the samples were denatured at 95°C for 5 min. Amplified fragments were separated on 6% polyacrylamide denaturing gels with 29:1 Acrylamide/Bis solution. The gels were pre-run at 110 W for 20 min in 1× TBE (Tris base 1.8%, Boric acid 0.6% and 0.5 M EDTA 0.2%), prior to the loading of 5 μl of sample and, thereafter, run at 110 W for approximately 3.5 h. Following electrophoresis, DNA was detected with silver staining (CIP 1998).

Gel scoring and data analysis

Each polymorphic fragment produced by each AFLP primer combination was treated as a unit character and numbered sequentially on the APC film or directly analyzed using Paint Shop Pro 5.01 software (Jasc Corporate, MN, USA). Genotypes were scored for the presence (1) or absence (0) of each fragment, only those fragments with medium or high intensity were analyzed. Fragments with the same mobility on the gel, but with different intensities, were not distinguished from each other when OTUs were compared, so the AFLPs were treated as dominant markers. In order to characterize the capacity of each primer combination, and to reveal polymorphic loci in the germplasm, the polymorphic index content (PIC) = 1 − (faa2 + fan2) was calculated, where faa2 is the frequency of the amplified allele and fan2 is the frequency of the nonamplified allele. The highest value of PIC indicated the most informative AFLP marker. The PIC values for the AFLP markers generated by the same primer combination were cumulated and named AFLP marker index (Ghislain et al. 1997). This index indicates the information content of the AFLP primer by assay. We used Analysis of Molecular Variance (AMOVA) procedure (Excoffier et al. 1992) to estimate variance components at different levels, partitioning the variation among individuals/within population, among populations/within species and among species. This analysis was undertaken with Arlequin ver. 2000, a software for population genetics data analysis, provided by Laurent Excoffier (http://anthropologie.unige.ch/arlequin/; Department of Anthropology and Ecology, University of Geneva, Switzerland).

The genetic analysis was performed using the NTSYSpc 2.1 software (Rohlf 2000). Similarities between OTUs were calculated with the Simple Matching Coefficient (SMC) = (a + d)/(a + b + c + d); where a = (1,1) is the number of common fragments between two accessions, d = (0,0) is the number of fragments absent in both accessions, c = (0,1) and b = (1,0) are the numbers of mismatches (Jaccard 1908, cited by Hancock 2004). According to this distance, the OTUs were plotted using nonmetric multidimensional scaling analysis (Rohlf 2000), in order to identify population groups.

Results and discussion

Morphological analysis

The univariate analysis of variance (ANOVA) for each of the 46 quantitative variables is detailed in Blas (2005). Data show highly significant differences (P < 0.001) between species for 43 characters, significant differences (P < 0.05) for one character [pinnas’ number of leaves (14)] and no significant differences for two characters [involucre number (50) and vittae number on the commissure (82)]. The ANOVA between populations in each species show similar tendencies. Populations of A. xanthorrhiza show highly significant differences for 40 characters, no significant differences for five characters [rootstock length (10), rootstock width (11), involucel number (45), peduncle width (55) and number of hermaphrodite flowers/umbellule (70)]; one character [petiolule length at upper level of leaf (40)] does not show variation. Populations of A. equatorialis show highly significant differences for 26 characters, and no significant differences for 20 characters; all characters show variation in this taxon. Populations of A. incisa show highly significant differences for 18 characters, no significant differences for 27 characters; only one character does not show variation [petiolule length at upper level of leaf (40)]. Populations of A. elata show highly significant differences for eight characters, no significant differences for 28 characters, four characters [petiolule length at lower level of leaf (39), petiolule length at upper level of leaf (40), branches number of generative shoot (63), branch number of root (98)] do not show variation. The following six characters [cormels number (4), cormels length (7), cormels width (8), rootstock length (10), rootstock width (11), number of storage roots per plant (85)] are absent in the populations of A. elata.

With this analysis, we observe a high morphological diversity among the material analyzed at both between-species (A. elata, A. equatorialis, A. incisa and A. xanthorrhiza) and within-species (between populations or accessions) levels. This morphological diversity is higher in populations of A. xanthorrhiza, A. equatorialis and A. incisa than in populations of A. elata.

In general, in these species, a high variability is found for most of the morphological descriptors. However, this variability could have been inflated by environmental conditions as the data were recorded in different regions and ecosystems, at the time of our collecting missions.

Principal Component Analysis (PCA)

In the Principal Component Analysis, the first three principal components (PC) show respectively 35.3%, 10.9% and 8.1% of the total variability. These low values mean that the populations and characters analyzed have high variability.

According to the first two PC, three groups of distribution are observed in plotting the 240 genotypes (Fig. 2). Group I represents the three populations of the species A. elata (accessions: RBS-130, RBS-134 and RBS-136); this group is positively related to the second component. A. elata is morphologically easy to identify by the absence of storage roots, rootstocks and cormels, by its plant size (more than 3 m), its ovoid seed and its involucel shape and size.

Fig. 2
figure 2

Distribution of the 240 plants belonging to the four Arracacia species according to the first two PC

Full size image

Group II includes only a single population of the species A. xanthorrhiza (accession RBS-138); this population is isolated from the other A. xanthorrhiza populations; this group is positively related to the first component and negatively to the second component. This population is a wild form accession, morphologically very closely related to the cultivated ones and characterized by a high quantity of seeds and well-developed storage roots.

In Group III, eight populations are positively related to the first component, covering the three species: A. equatorialis (RBS-93, RBS-96 and RBS-141), A. incisa (RBS-94, RBS-95 and RBS-129), and A. xanthorrhiza (RBS-124 and RBS-144). All genotypes of this group have storage roots, rootstocks and cormels. The RBS-95 and RBS-129 accessions of A. incisa are shown to be related to other populations corresponding to A. equatorialis, e.g. individuals corresponding to RBS-129 accession appear to be more related to accession RBS-141 of A. equatorialis. Concerning the three populations of A. equatorialis, two accessions are grouped together, e.g. RBS-93 and RBS-141; the third population (accession RBS-96) is more related to A. xanthorrhiza populations.

The two populations corresponding to A. xanthorrhiza are grouped with the two different related species, i.e. A. incisa and A. equatorialis. For example, individuals corresponding to RBS-124 accession are clustered in different groups; some individuals being more related to A. equatorialis and others to A. incisa. However, inside Group III, differentiation between individuals is not easy, rendering necessary the definition of more discriminant characters. Such characters would be useful to better define groups and to obtain a reliable identification of the species. As a result, this analysis does not enable to differentiate clearly between populations corresponding to the three species A. equatorialis, A. incisa and A. xanthorrhiza.

Canonical Discriminant Analysis (CDA)

Three variables (of the 100 used) are eliminated due to a very low variability in the basic descriptive statistics and PC analyses previously made. These characters are: petiole length at upper level of the leaf (40), secondary storage root flesh color (94) and distribution of secondary storage root flesh color (95). In the CDA, the first three components explain all the variability (Table 3); the first component explains more than 92% of the total variation while the second component explains more than 5% of the total variation. Consequently, characters with great discriminant capacity for the genetic diversity analysis in arracacha are those that display high vector with the first and the second component of the canonical variables.

Table 3 Eigenvalue of the first three canonical variables and their proportion of variations
Full size table

The criterion applied to identify discriminant characters is the absolute value of the Eigenvector coefficient (r) between the character and the canonical variable (Table 4). This coefficient should be r ≥ 0.5. Considering this criterion, descriptors are reduced to fifty discriminant characters. Among the latter, the most important are in decreasing order the leaflet adaxial surface (31), stylopodium shape (67), number of rays per umbel (56), leaflet margin (36), storage root shape (84), rootstock width (11), vittae breadth (83), number of flowers per umbellule (69), cormel width (8), generative shoot length (61), plant conformation (1), petiolule length at lower level of leaf (39), cormels shape (6), and mericarp transection shape (78).

Table 4 The 50 screened characters and the Eigenvector coefficient (r) between the character and the canonical variable
Full size table

In the CDA, some characters used in the original descriptions to differentiate the Arracacia species do not have high discriminant values. This is the case for fruit length, fruit width, fruit shape, leaf length, and leaf width. On the other side, other characters in our analysis appear to have more discriminant power. These characters correspond to vegetative parts (cormels, rootstock and storage root) and generative parts (generative shoots, flowers and fruits). The characters corresponding to the underground vegetative parts were not taken into consideration in the original descriptions of the Andean Arracacia species, partly because most samples in the herbarium lack representations of these organs.

Considering the 50 selected descriptors, the 12 populations were analyzed in order to test the discriminant power of characters and to determine cluster of individuals.

These selected descriptors allow us to obtain a good clustering among the 240 plants of the 12 analyzed populations (Fig. 3). These populations are clearly separated into three groups: Group I includes three populations of A. elata, Group II includes three populations of A. incisa and Group III includes six populations corresponding to A. equatorialis and A. xanthorrhiza.

Fig. 3
figure 3

Plotting of 240 plants with 50 selected descriptors according to the first two canonical variables (1 = A. xanthorrhiza, 2 = A. elata, 3 = A. equatorialis, and 4 = A. incisa)

Full size image

The three populations identified in A. equatorialis are not clearly separated from A. xanthorrhiza, indicating that the former taxon could not deserve the status of a species. Accordingly, populations of A. equatorialis could be classified together with populations of A. xanthorrhiza, because the discriminant descriptors overlap between both taxa. This overlap in characters expression could be due to the natural gene flow between the two taxa and the production of viable hybrid seeds. To support this hypothesis, it is interesting to note that Knudsen (2003) obtained viable hybrid seeds in artificial crosses between cultivated and wild forms of A. xanthorrhiza in Cajamarca (Peru). Crosses could have occurred spontaneously between A. xanthorrhiza and A. equatorialis as the two taxa have the same geographical distribution. As A. xanthorrhiza was botanically described first, its status as species has to be preserved, should one decide to merge the two taxa in a single species. On the other part, populations of A. incisa and A. elata are clearly separated allowing easily their identification with the selected descriptors. The CDA shows that 48 characters out of the 50 present an eigenvector coefficient r ≥ 0.5 with the first three canonical variables. The exception is for two characters: cormels number (4) and distribution of secondary root surface color (91). In spite of their high vector coefficients, we decided to discard the following characters: plant conformation (1), cormels length (7), cormels width (8), rootstock length (10), rootstock width (11), leaves shape (15), petiole width (22), main leaflet adaxial color (25), main leaflet abaxial color (26), lateral leaflet length (37), lateral leaflet width (38), petiolule length at lower level of leaf (39), involucel margin (47), involucre form (51), peduncle length (54), number of fertile rays (57), umbelulle conformation (60), pollen grain color (66), number of flower/umbellule (69), ribs of fruit (80), vittae number in the intervals (81), vittae breadth (83), root shape (84), root length (86), skin thickness (88), root oil (96), branch number of root (98), and root odor (99).

This is justified by the fact that we do not know the age of the evaluated material. As natural populations of arracacha are perennial plants, samples coming directly from natural areas do not represent a homogenized material. For instance, the length and the width of cormels, rootstocks and storage roots are “strongly” influenced by the environmental conditions, the soils where such material is found and the plant age. In addition, the color of the petioles and leaflets are not stable characters. However, these characters could be useful in some diversity analysis, such as at within population level.

On the basis of this consideration (CDA, Eigenvector coefficient), we selected 28 characters for identification of the Arracacia species from Andean region (Table 5). We also decided to include nine characters although they had in the CDA Eigenvector coefficient r < 0.5: life form (2), leaflet division (35), number of umbels/GS (68), number of grains/GS (72), carpophore insertion (75), seed length (76), seed width (77), fruit shape (79), and predominant storage root flesh color (93). We consider these characters as very discriminant within each species due to their stability. Then, the discriminant capacity of these selected descriptors will be useful in further Arracacia species identification from the Andean region and the analysis of natural populations.

Table 5 The 28 selected morphological descriptors for the identification of Arracacia wild species
Full size table

Differences among Arracacia species of Peru

Figure 4 illustrates the differences in vegetative and generative structures of these Peruvian species, considering stylopodium, carpophore, fruit, umbel, leaflet acumen, involucel, cormel and rootstock.

Fig. 4
figure 4

Some morphological descriptors for the identification of Arracacia species: (A) Stylopodium shape; (B) Carpophore insertion; (C) Fruit shape; (D) Umbel form; (E) Acumen shape of leaflet; (F)involucel form; (G) Cormels presence and (H) Rootstock shape

Full size image

Arracacia elata is not likely to be confused with any of the Peruvian Arracacia species. In general, A. elata displays spinulose-serrate leaflet margin, ovoid fruit with acute apex, and ovate-lanceolate involucel.

A. incisa can be distinguished by the length of the involucel which exceeds the flower, the conspicuously scarious margins, and the semi-rounded seeds. In addition, A. incisa is characterized by a short plant size, combining short petioles and small and incised leaflets. Additionally, the taxon is geographically isolated, being present in hilly and dry areas of the Peruvian western and central highlands of Quechua region. A. incisa is closely related to A. xanthorrhiza through the shared compressed basal stem structures (cormels), the ramification of the flowering stem and the perennial growth habit.

A. xanthorrhiza includes cultivated form, as well as monocarpic and polycarpic wild forms. The A. xanthorrhiza cultivated form has many storage roots and well developed cormels, two properties distinguishing this form from other Arracacia specimens belonging to the group with the storage root. The A. xanthorrhiza monocarpic form differs from the A. xanthorrhiza polycarpic form and A. xanthorrhiza cultivated form by the monocarpic behavior (the root dying away after the end of flowering), the presence of one (or rarely 2) cormel and one (rarely 2) reproductive shoot per plant. In addition, this reproductive shoot is very vigorous and shows many ramifications that carry up to 1,000 seeds (Blas 2005).

A. xanthorrhiza polycarpic form differs from A. xanthorrhiza cultivated form, in having few and small cormels, and often a splitting fibrous root. A. xanthorrhiza polycarpic form is related to A. xanthorrhiza monocarpic form, but differ in some characters: the presence of cormels with several shoots, the few umbels or low amount of seeds/generative shoot, the short plant size and the perennial behavior. Apparently, the A. xanthorrhiza polycarpic form appears to be the most closely related to A. xanthorrhiza cultivated form, sharing a perennial habit and similar ramifications of the flowering shoot, leaves, involucel and fruit shape.

Molecular analysis

The five most informative screened primer combinations (E-ACT/M-CAC, E-AAC/M-CTT, E-ACC/M-CTC, E-ACT/M-CTC and E-ACG/M-CTC) generate 202 AFLP reliable and reproducible markers. The highest values of AFLP marker index, showing the information content of the combinations of AFLP primers by assay, concern E-ACC/M-CTC, followed by E-ACT/M-CTC (Table 6). These two primers will be used in subsequent fingerprinting research in arracacha. Within species, very few unique polymorphic fragments are observed. A. elata shows three unique polymorphic fragments, representing only 1.5% of the total polymorphic fragments. No distinctive fragments are found for the species with storage roots, previously identified as A. incisa, A. equatorialis and A. xanthorrhiza (Table 6).

Table 6 Primer combinations, marker index and polymorphic bands in each studied species
Full size table

In general, the number of polymorphic bands is higher for A. elata than for the other species (A. incisa, A. equatorialis and A. xanthorrhiza). A. elata presents 78.5% polymorphic bands of the total polymorphic fragments. A. incisa, A. equatorialis and A. xanthorrhiza show 37.6%, 44.1% and 47.6% polymorphic bands respectively. Polymorphism among populations within species presents the same tendency in the four analyzed species, varying from 36.5 to 60.2%, 24.8 to 30.5%, 13.3 to 28.7% and 23.2 to 26.7%, respectively for A. elata, A. incisa, A. equatorialis and A. xanthorrhiza (Table 7).

Table 7 Polymorphic bands for each population and species
Full size table

The AMOVA analysis shows highly significant differences (p = 0.000, determined from a 10,000 replication bootstrap) among the four species, among populations within species and within populations. The analysis shows that 49.5% of the total variation is attributed to differences between species, while variation among populations within species is 25.8% of the total variation and the variation within populations accounts for 24.6% of the total variation (Table 8). This relatively high degree of genetic differentiation recorded between species, could be attributed to the geographic boundaries among species that do not facilitate gene flow among them. The high degree of variation within the populations could be explained by the widespread occurrence of wind pollination and breeding systems that promote outcrossing. Indeed, in arracacha, the presence of protogyny hinders self-fertilization and promotes outcrossing (Hermann 1997). Arracacha appears most likely to be a facultative outbreeder, giving progenies of genotypes with new genetic combinations. Another possible source for such pronounced genetic variation within populations might also result from hybridisations between wild and cultivated forms of A. xanthorrhiza or between A. xanthorrhiza and closely related species.

Table 8 AMOVA results for variation among species, among populations within species and within populations
Full size table

Three main groups are found after plotting the 12 populations with 202 AFLP markers according to nonmetric multidimensional scaling analysis (Fig. 5). The first group corresponds to the three populations of A. elata (RBS-130, RBS-134 and RBS-136). Individuals from each population are grouped closely, confirming previous results in the morphological analysis.

Fig. 5
figure 5

Plotting 240 plants of 4 Arracacia species according to nonmetric multidimensional scaling analysis with 202 AFLP markers

Full size image

The second group corresponds to the three populations of A. incisa (RBS-94, RBS-95 and RBS-129); individuals from each population are grouped closely.

The third group includes six populations corresponding to the two species previously identified as A. xanthorrhiza (RBS-124, RBS-138, RBS-144) and A. equatorialis (RBS-93, RBS-96 and RBS-141). A population of A. equatorialis (RBS-96) is closely related to A. xanthorrhiza. The same tendency is reported from the morphological analysis.

On the basis of morphological and molecular markers analysis, it is difficult to separate A. xanthorrhiza and A. equatorialis, indicating the closely related connection between them. Consequently, the identified taxon A. equatorialis could be considered as a different form within the A. xanthorrhiza species.

On the other side, A. incisa is a taxon deserving the status of species, in concordance with Constance’s (1949) study; this conclusion is drawn from our morphological and molecular analysis. Therefore two species are recognized in the Andean region: A. incisa and A. xanthorrhiza; both of them have storage roots. One of them, A. xanthorrhiza, includes cultivated and wild forms; while the second, A. incisa, includes only wild forms. The A. xanthorrhiza cultivated form has many storage roots and well developed cormels, two properties distinguishing this form from other Arracacia specimens belonging to the group with the storage root (Blas 2005).

Essential factors to clarify Arracacia taxonomy

The clarification of the wild Andean Arracacia taxonomy is not an easy task due to the influence of various parameters, such as introgressive hybridisation between distinct species and hybrid speciation, phenotypic plasticity within species, morphological convergence among species and presence of the same ploidy level in the Andean Arracacia species (Blas et al. 1997; Blas 2005).

In particular, the extent and effect of hybridisation and introgression in wild Arracacia do not facilitate the botanical separation between some taxa and the phyletic understanding of the genus. The taxonomical differentiation among the wild Andean species is even more complicated when cross-compatible cultivated arracacha varieties occur in the vicinity of these wild populations: as those cultivated varieties are more variable phenotypically due to both anthropical and environmental influences, gene flow between the cultivated and wild forms hinders seriously the taxonomical identification of the Andean wild Arracacia. Hybridisation between Andean wild and cultivated arracacha genotypes have produced extensive hybrid swarms; some of them have been recognized as taxa while others are still considered as intermediate forms. This explains why some species associated to the Andean region like A. incisa and A. xanthorrhiza, show a complex pattern of variation with not well defined boundaries between genotypes of this group. This complex pattern might also result from the two following factors:

  • Phenotypic plasticity which induces polymorphism in some morphological traits influenced by environmental components; this is illustrated by the high variation in leaf shape observed in several of the concerned taxa, depending on their ecological habitats;

  • Weediness which is common in many representatives of this group displaying a preference for disturbed habitats (such as roadsides, boundaries of cultivated fields, or ancient ruins of the Peruvian civilization).

Domestication process of Arracacia

Andean South America seems to be the place of domestication of arracacha, although the genetic diversity of the genus Arracacia in other regions like Mexico is particularly high. According to Bukasov (1930), arracacha was domesticated in the southern region of Colombia where he found different varieties of arracacha. At present, in South America, the highest diversity of Arracacia wild species was found in Colombia and Venezuela. Nevertheless, these wild species are not closely related to A. xanthorrhiza, due to the lack of storage root. The wild species A. incisa and the wild forms of A. xanthorrhiza, bearing storage roots and cormels and resembling more closely the cultivated arracacha (A. xanthorrhiza), present high genetic diversity in the Andean region, from where they probably originated. Arracacia xanthorrhiza is widely distributed through Bolivia, Peru and Ecuador from 2,000 to 4,000 m; whereas A. incisa is confined to the highland areas between the central and southern regions of Peru (Blas et al. 2007). A. incisa is the species more closely related to A. xanthorrhiza (Blas 2005).

So far it is not possible to elucidate clearly the origin of cultivated arracacha. Chromosome or gene mutations, selection in the wild forms, as well as hybridization between wild forms or between A. incisa and A. xanthorrhiza are all hypothetical ways to explain the origin of the cultivated arracacha. One of the farmers we met during our field survey in the region of Cajamarca considered that a selection of wild forms during three consecutive cycle at small farm level could lead to the development of domesticated form.

Evidences of Andean domestication are indicated by the uses of wild arracacha in medicinal or religious activities and as plants providing animal and human food since pre-Inca’s time (Blas 2005). There are some historical references about pre-Columbian cultivation of arracacha in Ecuador, Colombia and Peru, which were documented by Spaniard chroniclers during the conquest of South America (Cobo 1956). In addition, there is some archeological evidence, such as Nazca pottery designs more or less 2,000 years old (Yacovleff and Herrera 1934). This pre-Inca culture developed in the Central and South Peru, covering coastal and highland parts. Only, in this region, the species A. incisa considered as the most closely related to A. xanthorrhiza is found. The precise site of arracacha origin is difficult to establish in this context, but undoubtedly arracacha was domesticated in the Andean region, perhaps in the Southern region of Peru. This area corresponds to Ayacucho, Apurimac, Huancavelica, Cuzco Departments, the cradle of ancient cultures of the Inca Empire like Nazca and Chancas.

Conclusions

The 28 most discriminant morphological characters were selected to identify Arracacia species. These descriptors contributed to the distinction between the three species known from Peru: A. elata, A. incisa and A. xanthorrhiza. Each species was also well differentiated along its evolutionary process. For example, A. elata grows in eastern humid highlands, while A. incisa is geographically confined to the hilly and dry area of Peruvian central and western highlands (no rain for 7 months). The species A. equatorialis and A. xanthorrhiza share the same ecological and geographical distribution throughout all Peruvian Inter-Andean valleys (no rain for 5 months), which favors constant gene flow between them (Blas 2005) and even the likely production of viable hybrid seeds. Differentiation between these two taxa is not easy due to overlapping of morphological characters. The high morphological diversity among the species (A. elata, A. equatorialis, A. incisa and A. xanthorrhiza) and between the different populations analyzed in each of them could also be the result of environmental conditions, as samples were collected and evaluated in different regions. Molecular characterization (with 202 AFLP markers) in this material also showed a high diversity among species and inside populations. The magnitude of genetic differentiation between and within species suggests that the population size must be large enough to maintain such levels of genetic variation over time. The genetic analyses presented here could be used to develop conservation strategies for the species, for example through the definition of appropriate units for plant genetic resources management. In order to help explain the role of such units of conservation, the nature of (seed, pollen) dispersal mechanism, the pattern of gene flow within and between populations or species and their effects on reproductive and demographic processes should be further investigated. Patterns of variation in quantitative genetic traits should also be clarified in the building of conservation plans.

Morphological and molecular marker analysis do not separate obviously A. xanthorrhiza and A. equatorialis, indicating the close relation between them. For this reason, the populations corresponding to A. equatorialis could be considered as populations of A. xanthorrhiza and therefore the taxon might not deserve the status of a species. Consequently, our morphological and molecular analysis identified clearly at this stage three wild species of Arracacia genus: A. elata, A. incisa and A. xanthorrhiza. However, to clarify taxonomical differentiation, particularly between A. xanthorrhiza and A. equatorialis, it will be useful to validate the most discriminant characters among the studied populations, by comparing their performance and stability in multilocation trials.

Analysis of the wild arracacha genetic diversity for agronomical traits such as resistance to drought and decay of storage roots is also important to improve the cultivated arracacha. Incorporation of these desirable characteristics into existing cultivars will be useful to overcome the limiting agricultural factors of this Andean neglected crop.