The value of AFLP
One of the arguments against the use of AFLP is the possible bias caused by homoplasy [17, 19, 20]. Non-identical co-migrating bands in the AFLP fingerprints can contribute noise instead of signal to the dataset without being detected. However, it is not likely that in the tuber-bearing wild potatoes homoplasy will cause many problems because the species are all very closely related and homoplasy becomes a problem when distantly related species are involved. Koopman  showed that in a set of closely related Lactuca species, sufficient phylogenetic signal was present and concluded that in practice the influence of possible limitations of AFLP, such as co-migration of nonhomologous fragments is limited. However, he stresses that the conclusion only applies to datasets with closely related species. Moreover, Kardolus  concludes from his AFLP results that in Solanum section Petota the AFLP technique is suitable up to the species level. The AFLP method has since then successfully been used in more studies on potato taxonomy [21–24].
Status of groups within section Petota
Not all the groups found in this study have the same level of cohesion or have the same level of demarcation. Some groups have clear borders, while from others we can only vaguely recognize the contours. First, there is a number of groups that are always well supported, whether the analysis is done in a phenetic or phylogenetic way, see Figure 2 and 3. This is the case for the group of Mexican diploid species, the group of Mexican tetraploids, the group of S. demissum and S. acaule, the group of S. circaeifolium, the group of S. commersonii and the group of S. schenckii and S. hougasii. Then there are groups that are not supported in the MP jackknife tree (Figure 2) but that can be found in both the original MP trees and NJ trees (not shown) and are supported in the NJ jackknife tree (Figure 3). This applies to the group with Mexican hexaploid species, the group containing polyploid species belonging to series Conicibaccata, the group containing diploid Piurana species, and the small groups of S. huancabambense, S. kurtzianum, S. medians, S. mochiquense, S. hannemanii, S. buesii, and S. paucijugum.
The largest part of the jackknife trees consists of a polytomy of species that does not seem to contain structure at all. If one was only to consider the structure shown in the jackknife trees, the conclusion would have to be that according to the results of the present AFLP analyses the largest part of section Petota is without any taxonomic structure.
However, it is possible to identify additional groups that are present in many of the original NJ and MP trees, but do not have enough support to be shown in the jackknife trees. For example, in the 4929 dataset NJ tree a cluster represents the group of cultivated potatoes together with species of series Tuberosa from Peru. The groups that are found in both the phenetic and phylogenetic analysis are strong groups with clear borders. The exchange of genetic material is most likely restricted to the members of the group. The groups with only low support in the MP alone or in both trees are groups that probably share a considerable amount of genetic material with genotypes outside the group. In a study of Jacobs, van den Berg and Vosman: Comparison of Chloroplast DNA and AFLP data from Solanum section Petota reveals incongruencies between the datasets, submitted, the incongruencies found between the chloroplast data and the AFLP data suggest that hybridization occurs between species of different series in section Petota. For example, the composition of species of the clade representing the series Piurana in the chloroplast tree is different from that of the clade representing the Piurana series in the AFLP tree.
The resulting groups also have implications for the theory on EBN of Hawkes and Jackson . EBN stands for Endosperm Balance Number and refers to a hypothetical genetic factor that would explain the success or failure of crosses due to the functioning or breakdown of the endosperm after fertilization. Crosses between species with the same EBN are generally successful and crosses between species with different EBN generally are not, independent of ploidy levels. Hawkes and Jackson  claim that there is a correlation between the EBN hypothesis and the evolution of the group of tuber-bearing Solanum species. EBN 1 is found mainly in species that are considered to be close to the ancestors of the group: Mexican series Morelliformia, Bulbocastana, Pinnatisecta, and Polyadenia. The EBN 2 condition would have arisen as an isolating mechanism when potato species moved southwards. The EBN 4 condition occurs in hexaploids which are allopolyploids.
From the present results it is clear that there is no absolute relationship between EBNs and the groups found. In the group which contains S. acaule, S. demissum, S. semidemissum and S. edinense, different ploidy levels and different EBNs occur. This mixture of ploidy and EBN levels also occurs in the group with representatives of series Conicibaccata. The species S. moscopanum and S. tundalomense both are hexaploid and have EBN 4 and they form a group or cluster together with other series Conicibaccata species which are known to be tetraploid and have EBN 2. Although these tetraploid and hexaploid species from series Conicibaccata are mixed, the diploid series Conicibaccata (EBN 2) species do form a separate cluster.
With regard to the overall structure of the section as found in this study two main observations can be made. There seems to be a lack of supported structure, especially in the South American part of section Petota. Furthermore, there is a lack of support for the relationships between the different groups that were found in the NJ and MP trees. It is important to differentiate between these two phenomena because the causes underlying both cases could be different.
Lack of structure in South American part of section Petota
The AFLP jackknife NJ tree and the jackknife MP tree in this study shows a lack of structure or rather, an unresolved structure for the part of the tree which contains South American species while the other part of the tree shows several well supported groups.
Kardolus et al.  mentioned that within series Tuberosa different genotypes of the same species are not always grouped together and are scattered among genotypes from other species. He claims that the cause of this phenomenon is not the lack of resolution of AFLP, but the overclassification of a group of species, the so-called brevicaule-complex. The cpDNA RFLP studies of Spooner and Sytsma , and Spooner and Castillo  also showed a lack of support for a resolved structure within the group of South American species, and the branch uniting all these species had a bootstrap support value of only 67.
Volkov et al.  compared the ETS region of rDNA for 30 species of Solanum section Petota and found high bootstrap values for the branch uniting all the South American species in three different types of dendrogram (Maximum parsimony, Bayesian statistics and Neigbour Joining). However, the two subgroups within the South American clade that they distinghuished (variants C1 and C2) often show polytomies and resolution within the groups is mostly lacking.
Outside the field of potato taxonomy, researchers have reported similar patterns. Hughes and Eastwood  report a low sequence divergence and lack of resolution in the large Andean clade of the genus Lupinus. This would point at a rapid and recent diversification in the Andes. The authors also suggest that Lupinus is probably only one example of many plant radiations that followed the final uplift of the Andes. They assume that many of these plant radiations are yet unknown. It is possible that the factors underlying the Lupinus diversification are also responsible for the Solanum section Petota diversification. According to Hughes and Eastwood  these factors would be the large scale of the area over which the radiation extends, repeated fragmentation of high altitude habitats due to quaternary climate fluctuations, the extremely dissected topography, and the habitat heterogeneity.
Lack of support for relationships between different groups
Except for the outgroup consisting of S. etuberosum, S. palustre and S. fernandezianum which connects to the main branch of the NJ jackknife and MP jackknife tree with respectively 100 or 98 support value, none of the branches connecting two or more groups have jackknife support of 69 or higher. That is the reason why in the schematized jackknife NJ and jackknife MP trees these branches collapse in a polytomy. Contrastingly, the branches of the groups that can be recognised within the polytomy do have jackknife support, although not all species can be put in groups as discussed previously.
In the first study on the use of AFLP in Petota taxonomy by Kardolus et al. , it proved also difficult to find bootstrap support for branches connecting the different groups in section Petota. Bootstrap support above 70 were given for a NJ tree branch connecting the outgroup of S. etuberosum and S. brevidens, for a branch connecting the outgroups, and for the Mexican diploids and S. circaeifolium and S. circaeifolium subspecies quimense with the other part of the tree. In the cpDNA RFLP studies on the South American part of section Petota  only a few branches connecting the larger groups showed bootstrap support above 70. Clade 1, consisting of Mexican diploids (except S. cardiophyllum and S. bulbocastanum) is connected to the other clades with a bootstrap value of 87, and Clade 3 (mainly accessions belonging to series Piurana) and Clade 4 (the rest of section Petota) are connected to each other with a branch with 96 bootstrap support.
We can conclude from these previous results that it is indeed difficult to find good support for the backbone structure of section Petota in general. This indicates that our and previous results represent the real biological situation in Solanum section Petota. Since the phylogenetic signal is clearly present in our data as shown in the well-supported groups in the present study, the lack of structure in parts of the tree is not caused by the lack of phylogenetic signal in AFLP markers.
New informal species groups for Solanum section Petota
As outlined in this paper and in other earlier studies, there are no results that support the classification of section Petota in 21 series. Although a few of the series seem to form natural groups, the majority of the series as proposed by Hawkes  could not be found as separate clusters or clades. Our goal is to use the found structure in the present study at maximum for classifying the section Petota.
We propose to divide section Petota in informal species groups, following the approach of Spooner et al.  who constructed 11 informal species groups for the North and Central American species. They followed the approach of Whalen  and Knapp [29, 30] who applied a similar informal species group classification. We will use the names already used by Spooner et al.  if applicable, and add new groups that were not treated in their study. We chose to base the informal group classification on the groups that are supported in the NJ jackknife tree. The NJ jackknife tree shows more resolution relative to the MP. However, it would not be useful to consider every small group that appears in the schematized tree as a biologically meaningful group. Therefore, the choice for species groups is restricted to groups of species that make sense in the light of former studies and contain at least 3 species. We maintain the species group Verrucosa which contains only one species, because this species group is already designated by Spooner et al .
In total, the NJ jackknife tree can be partitioned into 10 species groups. It would be possible to construct more species groups based on the structure shown in the various trees made in the present study, but these groups would then not be supported by bootstrap or jackknife supports.
Although a closed classification following the rules of the Botanical Code is desirable, it seems in this case difficult to apply. In the present study, many species cannot be accommodated in groups. These species do not automatically form a group themselves, but are intentionally left unclassified.
We suggest recognizing the following informal species groups as shown in the NJ jackknife tree (Figure 3):
Diploid Mexican group
This group contains the species groups of Spooner et al. : Pinnatisecta, Stenophyllidia, Trifida, Polyadenia, Morelliforme, and Bulbocastana. These species groups can be recognized in the present study as separate branches within the NJ cluster which represents this species group. In the present study we recognize a higher level of group structure which contains all the mentioned species groups, because the detailed contents of each subgroup in our study (Figure 3) differs from the contents from the species groups from Spooner et al. .
In our study this group contains 2 supported subgroups, one branch with jackknife support of 96 containing the species S. semidemissum, S. demissum and S. x edinense. The other group shows a jackknife support of 98 and contains S. juzepczukii, S. albicans and the three subspecies S. acaule subsp. acaule, S. acaule subsp. aemulans, S. acaule subsp. punae.
This group contains the species S. schenckii, S. hougasii, that form a strongly supported cluster together (jackknife support 100) and a cluster containing the species S. iopetalum, S. brachycarpum, S. guerreroense (jackknife support 90). All species were formerly designated by Hawkes  to series Demissa which also included the species S. demissum and closely related species. The species in our group are the same as in the species group Iopetala designated by Spooner et al. . They reduced the species S. brachycarpum as a synonym of S. iopetalum.
As the name does suggest, this group contains species that were formerly placed by Hawkes  in the series of Longipedicellata. The species included in this group are S. fendleri including S. fendleri subsp. arizonicum, S. stoloniferum, S. hjertingii,. S. papita, S. polytrichon, S. leptosepalum, and S. matehualae. The species S. leptosepalum, S. fendleri, S. papita, and S. polytrichon have been reduced as synonyms of S. stoloniferum . The species S. matehualae is reduced as synonym of S. hjertingii .
Polyploid Conicibaccata group
This group contains species placed there by Spooner et al. , complemented with South American species. The species in this species group are mainly the same as Hawkes  placed in series Conicibaccata. According to the present study the group consists of S. flahaultii, S. moscopanum, S. orocense, S. sucubunense, S. tundalomense, S. oxycarpum, S. longiconicum, S. garcia-barrigae, S. otites, S. oxycarpum, S. agrimonifolium, S. moscopanum, S. subspanduratum, S. paramoense, and S. colombianum.
Diploid Conicibaccata group
Although most of the series Conicibaccata can be put in the species group Conicibaccata there are a few species that form a separate group. This group consists of the diploid species S. buesii, S. sandemanii, and S. laxissimum.
Diploid Piurana group
This species group was not designated by Spooner et al. . The name refers to the former series Piurana as the contents of the group are roughly similar: S. piurae, S. acroglossum, S. blanco-galdosii, S. irosinum, S. chomatophilum, and S. paucissectum from series Piurana and S. chiquidenum from series Tuberosa.
Tetraploid Piurana group
The situation as described before for the Conicibaccata group also applies partly for the Piurana group. There are a few species from the formerly designated Piurana series  that form their own species group. This species group contains the tetraploid species S. paucijugum, S. tuquerrense, and S. solisii.
This group consists of S. circaeifolium, S. soestii, S. capsicumbaccatum and S. circaeifolium subsp. quimense. The contents is conform Hawkes' series Circaeifolia.
This group contains only 2 species; S. macropilosum and S. verrucosum. The species S. macropilosum was reduced to a synonym of S. verrucosum by Spooner et al. .