| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
2 Molecular Systematics Laboratory, Swedish Museum of Natural History, P.O. Box 50007, SE-104 05 Stockholm, Sweden; and 3 Department of Phanerogamic Botany, Swedish Museum of Natural History, P.O. Box 50007, SE-104 05 Stockholm, Sweden
Received for publication May 25, 1999. Accepted for publication December 21, 1999.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
The results show that Maesa is sister to all other taxa from the three families. Theophrastaceae are well supported, but both Myrsinaceae and Primulaceae are paraphyletic. We conclude that four families should be recognized, Maesaceae, Theophrastaceae, Primulaceae, and Myrsinaceae. For all families to be monophyletic, Samolus was transferred to Theophrastaceae, and Lysimachia, Anagallis, Trientalis, Glaux, Asterolinon, and Pelletiera were moved to the Myrsinaceae together with the genera Coris, Ardisiandra, and Cyclamen.
Key Words: atpB • DNA • Maesaceae • Myrsinaceae • ndhF • phylogeny • Primulaceae • rbcL • Theophrastaceae
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
), represent a small monophyletic group within the reclassified order Ericales s.l., one of the major clades of the subclass Asteridae (Angiosperm Phylogeny Group, 1998
). The Ericales s.l. comprise 24 families, including families such as Theaceae, Lecythidaceae, Sarraceniaceae, Ericaceae, Balsaminaceae, and Polemoniaceae. The variation in floral morphology is considerable, and representatives may have polystemonous, diplostemonous, or haplostemonous flowers, with regular or zygomorphic, choripetalous or sympetalous corolla, and one or two integuments, but always tenuinucellate ovules.
Theophrastaceae, Myrsinaceae, and Primulaceae share characters such as haplostemonous flowers with sympetalous corolla, stamens opposite the petals, free central placentation, bitegmic tenuinucellate ovules, and nuclear endosperm formation. They have consistently been found to constitute a monophyletic group in cladistic analyses of morphological as well as molecular data (Anderberg and Ståhl, 1995
; Morton et al., 1996
; Anderberg, Ståhl, and Källersjö, 1998
; Källersjö et al., 1998
), but it has been indicated several times in recent years that there is a need for a taxonomic realignment. In particular, the position of Maesa in Myrsinaceae has been questioned. Anderberg and Ståhl (1995)
, in their morphological study, concluded that Maesa does not belong in the Myrsinaceae and could merit recognition as a separate family. Later studies based on rbcL sequences (Morton et al., 1996
; Anderberg, Ståhl, and Källersjö, 1998
) also showed that Maesa did not group with other Myrsinaceae.
The rbcL studies also indicated that Myrsinaceae (excluding Maesa) appeared to be nested inside Primulaceae, although its exact position could not be identified. Anderberg, Ståhl, and Källersjö (1998)
, who in their study included a more extensive taxon sampling than Morton et al. (1996)
, found Myrsinaceae in a trichotomy with Cyclamen and Primulaceae–Lysimachieae. They concluded that dramatic changes in circumscriptions would be necessary to make the families represent monophyletic groups. However, neither the study of Morton et al. (1996)
, nor that of Anderberg, Ståhl, and Källersjö (1998)
was able to resolve the basal relationships in the "Primulales" clade. Morton et al. suggested that Theophrastaceae form the basal group of the clade, although this was not supported by their jackknife analysis, whereas Anderberg, Ståhl, and Källersjö found a basal trichotomy of Maesa, Theophrastaceae and a group consisting of Primulaceae and the remaining Myrsinaceae.
In order to clarify the phylogenetic relationships between the three families of the "Primulales," we sequenced rbcL from several additional taxa, and we also sequenced two other chloroplast genes, ndhF and atpB. Data from the three genes were analyzed together and also in combination with morphology to give the best estimate of the phylogeny. To explore the data each gene was analyzed separately using different sets of characters: all nucleotide positions, first and second vs. third codon positions, and finally, including transversions only.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
. Theophrastaceae are represented by three of its six genera. In Myrsinaceae, representatives of the two tribes Ardisieae and Myrsineae, as well as Aegiceras (sometimes recognized as a separate family) and the distinctive genus Maesa, were examined. Representatives of all recognized tribes of Primulaceae, i.e., Primuleae, Androsaceae, Ardisiandreae, Lysimachieae, Glauceae, Anagallideae, Corideae, Cyclamineae, and Samoleae (e.g., Pax and Knuth, 1905; Takhtajan, 1997
) have been included. From the large and heterogeneous genera Primula and Lysimachia a few species of each were sequenced. As outgroups we used Diospyros (Ebenaceae) and Manilkara (Sapotaceae), both members of Ericales s.l., a choice based on earlier studies (Chase et al., 1993
; Anderberg and Ståhl, 1995
; Morton et al., 1996
). A complete list of investigated taxa, including GenBank accession numbers and voucher information, is presented in Table 1.
|
. A nested PCR (Polymerase Chain Reaction) was performed by first amplifying an 
2000-base long fragment, which subsequently was used for two separate reactions, one amplifying 800 bases of the 5' end, the other 1200 bases of the 3' end. For PCR reactions, "Ready-to-go PCR beads" from Pharmacia Biotech (Amersham Pharmacia Biotech, SE-751 84 Uppsala, Sweden) were used, following the manufacturer's standard protocol and suggested thermal cycling profile. Amersham's "ThermoSequenase Flourescent labelled primer sequencing kit" (Amersham Pharmacia Biotech, SE-751 84 Uppsala, Sweden) was used for sequencing reactions, and fragments were separated on an AlfExpress automated sequencer from the same company. Primers used for PCR and sequencing are listed in Table 2. The ndhF sequences obtained were aligned using the AssemblyLign software (Oxford Molecular Group Inc., 2105 SO Bascom Avenue, Campbell, California 95008 USA). Alignments were unproblematic since only a few insertions/deletions were required (Table 3). The aligned data set comprised 1956 positions (corresponding to positions 111–2048 of the tobacco sequence), of which 420 were informative.
|
|
were used for amplification and sequencing: RBCL1, S2, S385, S611, S766R, and S1494R. For sequencing reactions we used the "Big Dye Terminator Sequencing kit" (PE Applied Biosystems, Warrington, WA1 4SR, UK), and fragments were separated on an ABI377 from PE Applied Biosystems. Alignment could be performed by eye, as no insertion/deletions were necessary. The aligned sequences included 1461 positions (corresponding to positions 1–1461 of the spinach sequence; Zurawski, Bottomley, and Whitfield, 1982
), and 179 of these were informative.
An rbcL data set was produced by combining 16 GenBank sequences with 17 new sequences. Extraction, sequencing, and alignment followed the procedure described in Anderberg, Ståhl, and Källersjö (1998)
, except that PCR was performed as above, using "Ready-to-go PCR beads" from Pharmacia Biotech. Alignment was straightforward since no assumptions of insertions/deletions were required. The final data set comprised 33 taxa including outgroups and 1408 positions (corresponding to positions 21–1428 of the tobacco sequence), of which 208 were informative.
A second rbcL data set was compiled by adding all available sequences of rbcL from "Primulales" including taxa for which the ndhF and atpB genes have not yet been sequenced. This second data set comprised 51 taxa including outgroups and 1408 positions, of which 256 were informative. Compared to the 33-taxon data set it has a more extensive sampling from the Myrsinaceae and from large genera such as Clavija, Jacquinia, and Primula.
A combined molecular data set was constructed by adding the corresponding atpB and rbcL sequences to the 33 taxa of the ndhF data set. The same specimens were used for the three gene sequences, except in a few cases where material was not available or amplification unsuccessful. The rbcL sequence of Diospyros virginiana was combined with an ndhF and atpB sequence of D. digyna, and, likewise, Clavija domingensis was combined with C. euerganea. The rbcL and ndhF sequences of Samolus valerandi were combined with an atpB sequence of Samolus repens. For Anagallis arvensis, the three sequences represent the same species, but have two different vouchers. For Glaux, Lysimachia nemorum, Primula edelbergii, and Primula sikkimensis the entire atpB gene was coded as question marks because sequence data could not be obtained (see above). For detailed information on voucher specimens, see Table 1. The combined three-gene data set comprised a total of 4825 positions, 807 of which were informative.
Morphology
A combined morphological and molecular data set was produced by adding 57 morphological characters published by Anderberg and Ståhl (1995)
to the combined three-gene data set mentioned above. Several of the characters listed in Anderberg and Ståhl's Appendix 1 (characters 8, 10, 12, 14, 16, 19, 20, 24, 27–31, 41, 43, 46, 55, 58, 59, 61, 64, 71, 76, 81, and 82) turned out to be uninformative with our selection of taxa, and hence they were excluded from the analysis. A detailed discussion on morphology and the choice of characters is provided by Anderberg and Ståhl (1995)
.
Phylogenetic analyses
All six data sets mentioned above, i.e., the combined morphological and molecular data, the three-gene data, the ndhF, the atpB, the 33-taxon rbcL and the 51-taxon rbcL, were analyzed by parsimony jackknifing (Farris et al., 1996
), using the computer software "Xac" (discussed by Källersjö et al., 1998
) with the following settings: 10 000 replications, each with branch-swapping and 100 random addition sequences. Gaps in the ndhF sequences were treated as missing information (but see Results).
To explore the data, several additional analyses were performed. Subsets of the original data sets were created as follows: (1) including first and second codon positions only, (2) including third codon positions only, and (3) including transversions only. All subsets were analyzed using parsimony jackknifing with the same settings as above.
The choice of measures to evaluate performance of characters is not trivial. Commonly used indices, e.g., the consistency index or derivations thereof, can be positively misleading as shown in Källersjö, Albert, and Farris (1999)
. In the present study, performance of sets of characters is measured by number of supported groups and their average support. These measures are admittedly crude, groups with a support frequency close to 50% tend to come and go, and their presence will increase the number of groups found, while lowering the average support. Still, they do provide useful and reasonably consistent information as pointed out by Källersjö, Albert, and Farris (1999)
.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
|
Maximum support (100%) was found for the monophyly of the group including Theophrastaceae, Myrsinaceae, and Primulaceae. Maesa appears as the sister to all other taxa, a relationship also supported at 100%. On the next higher level, a small group comprising Theophrastaceae together with Samolus from the Primulaceae is the sister group to the remaining taxa. Theophrastaceae are well supported at 100%, but the support for a relationship between Samolus and this family is considerably lower, only 60%. Within the Theophrastaceae, Theophrasta is sister to a group formed by Clavija and Jacquinia (62%). The remaining Primulaceae–Myrsinaceae form a well-supported (100%) monophyletic group composed of two major clades, which in the following will be referred to as Clade I and Clade II.
Clade I, supported at 100%, is formed by the genera of Primulaceae–Primuleae, i.e., Androsace, Douglasia, Omphalogramma, Soldanella, Dodecatheon, Cortusa, and Primula. Androsace and Douglasia form a well-supported (100%) group, sister to the remaining taxa (supported at 99%). At the next level, Omphalogramma and Soldanella (100%) constitute the sister group to the Dodecatheon + Primula + Cortusa group (97%). In this latter group a clade (100%) is formed by two sister pairs of Primula, P. sikkimensis + P. veitchiana (79%), and P. edelbergii + P. gaubeana (100%). This clade is sister to a group supported at 61%, consisting of two groups, one being Dodecatheon + Primula palinuri (100%), another (supported at 100%) includes Cortusa + Primula sieboldii (86%) as well as P. cortusioides.
Clade II, supported at 100%, includes taxa from Primulaceae–Lysimachieae (Trientalis, Anagallis, Glaux and Lysimachia), Myrsinaceae s.s. (sensu stricto) (Hymenandra, Aegiceras, Grammadenia, Myrsine), together with the three primulaceous genera Coris, Ardisiandra, and Cyclamen. In this clade Coris is the sister to all other taxa (100%), followed by Ardisiandra as sister to the remaining taxa (93%), which in turn form a group with a basal trichotomy. The first branch in the trichotomy is the genus Cyclamen. The second consists of the Myrsinaceae genera forming a well-supported group (100%), and among them Hymenandra is sister to the other taxa (76%). The third branch is Primulaceae–Lysimachieae, supported at 100%, with Trientalis as a basal to the remaining taxa (supported at 99%), where Lysimachia nemorum groups with Anagallis (100%), and Glaux appears as the sister (95%) to L. maxima + L. menoricensis (100%).
In the ndhF alignment there are three putatively informative insertions/deletions (Table 3). Indels can be coded in different ways for phylogenetic analyses. We have treated gaps as missing data, but a different coding would not have changed the tree topology. If the informative insertions/deletions are plotted onto the tree, we find that one insertion (three bases) is a synapomorphy for the Theophrastaceae, a group already supported at 100%. The two deletions (six and 12 bases, respectively) characterize the large group, also with maximum support, including all Primulaceae and Myrsinaceae, except Maesa and Samolus.
Combined analysis of ndhF, atpB, and rbcL
The combined molecular data set included 33 taxa and 4825 nucleotide positions, of which 807 were informative. First and second codon positions contributed 37% of the informative sites. The tree produced by parsimony jackknifing (Fig. 2) recognized the same 28 supported groups as found in the tree based on a combination of molecular data and morphology, differing only in support values for some groups. The difference in support is small, with one exception: the three-gene tree shows a considerably higher support (90%) for grouping Samolus with the Theophrastaceae. The average group support for the three-gene tree is also somewhat higher (95%).
|
Analysis of ndhF
The ndhF data set comprised 1956 nucleotide positions. Of the 420 informative sites 45% were first or second codon positions. Analysis of the ndhF data alone produced a tree very similar to that of the combined data, with 27 supported groups, but with a slightly lower average group support, 91% (Fig. 3). The recognized groups are the same as in the combined tree, except that there is <50% support for joining Dodecatheon + Primula palinuri as sister to the group consisting of Cortusa + Primula sieboldii + P. cortusioides. The position of Maesa as the sister to a clade formed by all the other genera of Theophrastaceae, Primulaceae, and Myrsinaceae has maximum support. As in the combined analysis, Samolus groups with Theophrastaceae, but with low support (53%).
|
|
|
|
Analyses of subsets show that most phylogenetic structure is provided by third positions, whereas the tree based on first and second positions is poorly resolved (eight groups) with low average support (69%). The sister-group relation of Maesa to the remaining ingroup taxa is supported at 60%, and Theophrastaceae have a jackknife frequency of 66%. The remaining resolution consists of small groups in Primulaceae–Primuleae. Dodecatheon groups together with Omphalogramma, a relationship not found in any of the other trees, but this group has only a weak support of 57%. In the tree based on third positions, 19 groups are identified, and they have a higher average support of 86%; all but one are the same as in the all-positions atpB tree. Clade I has a support of <50% in the third position tree, and the sister-group relation of Androsace and Douglasia is not found. Instead Douglasia groups with Omphalogramma and Soldanella, albeit with low support (52%). The tree based on transversions only is similar to that resulting from analysis of first and second positions, but the average support is higher (79%), and Dodecatheon does not group with Omphalogramma but with Primula palinuri.
Analysis of rbcL (33 taxa)
The rbcL data set with 33 taxa included 1408 positions with 208 informative sites (35% from first and second codon positions). The parsimony jackknife tree (Fig. 5) is less resolved (24 groups) than that from the combined analysis, and the average support is lower (83%). The ingroup (Maesa, Theophrastaceae, Primulaceae, Myrsinaceae) is monophyletic with maximum support, but the sister-group relation of Maesa to the remaining taxa is not supported. Instead, there is a basal trichotomy of Maesa, Theophrastaceae, and the remaining Primulaceae–Myrsinaceae taxa. Theophrastaceae are well supported (100%), but they do not group with Samolus. In the rbcL tree, unlike the other analyses, Samolus is sister to the well-supported (98%) Primulaceae–Myrsinaceae group (Clades I + II), although the support for this placement of Samolus is low, only 56%. As in the result from the combined analysis, the large Primulaceae–Myrsinaceae group is divided into two clades. Clade I is supported at 77% and includes the same taxa as in the combined analysis, but has a somewhat different topology. Androsace and Douglasia group together and form the sister group of the remaining taxa, but unlike the combined analysis where Omphalogramma together with Soldanella appears at the next higher level, the remaining taxa form a trichotomy. The first group of this trichotomy is Dodecatheon + Primula palinuri, the second is P. cortusioides + P. sieboldii + Cortusa, and the third group has Omphalogramma + Soldanella as sister group to the remaining Primula species (P. sikkimensis, P. veitchiana, P.edelbergii + P.gaubeana). Clade II has a support of 88%, and as in the combined analysis Coris is sister to the rest, followed by Ardisiandra at the next higher level. The internal structure of the group containing the remaining taxa is only slightly different from the combined analysis. As in the atpB tree Hymenandra groups with Aegiceras within Myrsinaceae, although with low support (54%).
|
Analysis of rbcL (51 taxa)
The rbcL data set with 51 taxa included 1408 positions, 256 of them informative. Of the informative sites 38% were first or second codon positions. The parsimony jackknife tree (Fig. 6) includes 33 supported groups, with an average support of 83%. Maximum support (100%) was found for the ingroup. The two species of Maesa appear together as sister to the remaining taxa, but this relationship has very low support, only 54%. The three Theophrastaceae genera constitute a monophyletic group with maximum support. Clavija and Jacquinia are each represented by more than one species, and each forms a monophyletic group, but the generic interrelationships in the family are unresolved. Theophrastaceae are part of an unresolved trichotomy with Samolus of the Primulaceae, and a clade composed of the remaining Primulaceae–Myrsinaceae.
|
As in the 33-taxon rbcL analysis above, analyses restricted to first and second codon positions, or to transversions, resulted in major loss of resolution, as well as lower average support. The first and second position tree found only one of the major groups, the Theophrastaceae (53%). The tree based on transversions found support for two of the major groups, the ingroup (72%), and the Theophrastaceae (97%). Analysis of third positions resulted in a well-resolved tree, with 31 groups and an average support of 83%, recognizing all major groups, except Clade I. Unlike the tree based on all positions of rbcL, this tree shows Samolus grouping with the Theophrastaceae (57%).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
. Their study, which included 45 taxa and 82 characters, was a first attempt to examine the circumscription of the primuloid families through a cladistic study of morphological characters. In the present study the choice of taxa is somewhat different, rendering several of Anderberg and Ståhl's characters uninformative and consequently they were excluded (see Materials and Methods). A parsimony jackknife analysis of the reduced morphological matrix resulted in a tree (not shown) with few supported groups: Theophrastaceae (81%), Androsace + Douglasia (88%), Primula cortusioides + P. sieboldii + P. sikkimensis (65%), and Cortusa + Soldanella + Cyclamen + Dodecatheon (68%). The latter two groups do not appear in any of the molecular analyses. The tree of Anderberg and Ståhl agrees with our present study in some respects, e.g., Maesa does not group with other Myrsinaceae, Theophrastaceae are monophyletic, and Lysimachia is paraphyletic. There is also some incongruence, particularly concerning the relationships among the major groups. In the morphological study Myrsinaceae are basal to, and not nested within, Primulaceae. Furthermore, Samolus belongs to the Primulaceae clade.
The tree obtained from a combination of morphological and molecular data (Fig. 1) has a topology identical to that of the three-gene tree (Fig. 2), the topology of the former being influenced mostly by the highly structured molecular information. Addition of morphological information has led to a slight increase, or decrease, in support for some groups, but generally it has had little effect. However, there is one notable exception. The support for Samolus as sister to the Theophrastaceae is 90% when the combined molecular data are analyzed, but the jackknife frequency drops to 60% when the morphological characters are added. There are several morphological characters supporting a close relation of Samolus to Primulaceae (particularly Primulaceae–Primuleae; see Anderberg and Ståhl, 1995
), but when morphological data are analyzed separately the support for that group is <50% due to high levels of homoplasy. Even so, in the combined data set the contradictory information from morphology strongly influences the jackknife frequency, probably because it adds to the already existing and conflicting information present in rbcL where there is evidence for a placement of Samolus close to the entire Primulaceae–Myrsinace clade (see Table 5). This will be further discussed below.
A detailed discussion on the conflict in the "Primulales" between molecular data and some of the morphological characters was presented in Anderberg et al. (1998)
.
Comparing ndhF, atpB and rbcL
There is little conflict in topology between the trees from the three genes, which differ mainly in degree of resolution. The rbcL and the atpB data sets, analyzed separately, produce trees that are unresolved at several nodes where the ndhF data provide resolution. The combination of the three genes gives a tree that much resembles that from ndhF alone, but with higher support for most nodes. Apparently there is information in rbcL and atpB that is insufficient to diagnose groups when analyzed separately, but adds support when analyzed in combination with ndhF.
There are, however, some conflicts among the six trees. One concerns the position of the Soldanella–Omphalogramma group within the Primulaceae–Primuleae clade (Clade I, see above). In the atpB tree this group is part of a basal polytomy in the clade. The ndhF data place it as sister to Dodecatheon, Cortusa, and all Primula, whereas rbcL data (both data sets) show it to be most closely related to a group of four Primula species. The support for this latter grouping is low (65%), however, and the tree from the combined data reflects the strong support found in ndhF for a more basal position of the Soldanella–Omphalogramma group in the Primulaceae–Primuleae clade.
The atpB gene disagrees with the other two genes in two respects. In the ndhF and rbcL trees Primula sieboldii groups with Cortusa, with P. cortusioides as their sister. In the atpB tree the order is different, with Cortusa sister to the two Primula species. Furthermore, Hymenandra is sister to other Myrsinaceae in both the rbcL and the ndhF trees, but in the atpB tree it groups with Aegiceras. In both cases the tree from the combined data reflects the agreement of rbcL and ndhF.
The most interesting conflict among the trees is the position of Samolus (see above). There are three alternative positions for this genus, as sister to the Theophrastaceae (ndhF and atpB), as sister to the large Primulaceae–Myrsinaceae clade (33-taxon rbcL), or finally unresolved in relation to these two groups (51-taxon rbcL). Samolus and its placement will be discussed further below.
It is often suggested that some molecular changes are less reliable for phylogenetic reconstruction than others. Changes that are thought to occur at high frequency, such as transitions, or substitutions in third codon positions, are expected to show more homoplasy, and therefore less phylogenetic structure than those occurring at a slower rate (e.g., Stewart, 1993
; Swofford et al., 1996
). The purpose of the various subset analyses was to compare the behavior of the different genes, to evaluate which parts of the data contributed to the overall phylogenetic structure, and to investigate whether removing allegedly "noisy" characters would produce more robust tree topologies. Comparisons of the three chloroplast genes show that ndhF provides twice as many informative characters as rbcL, which in turn provides almost twice as many as atpB (although it should be noted that only 29 taxa were sequenced for the latter; see also Table 4). To some extent this can be explained by the difference in length; the ndhF data set incorporated 1956 sites compared to the 1408 for rbcL or 1461 for atpB, but it also indicates a higher substitution rate in ndhF. This is not unexpected, since other studies have show a higher rate of change in ndhF (e.g., Kim and Jansen, 1995
; Olmstead and Reeves, 1995
). The subset analyses reveal additional differences between the three genes (Tables 4 and 5).
For ndhF, 27 groups were found in the tree resulting from analysis of all positions. Most of those groups (23–25) were also found in the analyses of the subsets, showing that transversions, as well as all three codon positions, contributed to the tree in Fig. 3. The rbcL gene behaved differently. For both rbcL data sets (33 and 51 taxa, respectively) analyses of the first two codon positions by themselves, of transversions only, resulted in massive loss of resolution. When first and second positions were analyzed, the major groups listed in Table 5 were not found, not even the ingroup. In contrast, analysis of third positions produced a tree similar to that from all positions, although somewhat less resolved. It is interesting to note that the position of Samolus as sister to the Theophrastaceae, found in the ndhF and atpB analyses, also has some support when third positions of rbcL are analyzed separately. This congruent information results in an increased support for this relationship in the combined analysis. For the atpB gene 20 groups were found when all positions were analyzed. Analysis of the third position recovered 19 groups (18 of them identical to the ones in the all-position tree), whereas first and second positions found only eight. Analysis of transversions also resulted in loss of resolution, recognizing only nine groups.
It may be argued that in relatively slowly evolving genes like rbcL or atpB, changes in third positions should provide more structure as they are more common and need not have accumulated multiple hits. In our 33-taxon rbcL data set third positions contribute 65% of the informative sites, so it seems reasonable that they contribute more structure to the tree. What is surprising is the poor performance of the first two codon positions, where 73 informative sites are able to recognize only seven supported groups. Moreover, these groups are at the tips of the tree and include only 2–3 terminals each. First and second positions in atpB appear to perform better; 38 informative characters find eight groups, recognizing the sister relation of Maesa to the remaining taxa, but the average group support is the lowest found in any of the analyses.
In all three chloroplast genes third positions provide well-structured topologies, recognizing older as well as younger groups, with high average support. The difference in the performance of the first two positions compared to the third positions cannot simply be explained by differences in substitution rates. Even if we compensate for the larger proportion of informative first and second positions in the ndhF data, there is still a considerable difference in resolving power, especially compared to rbcL. First and second positions consistently show lower average support compared to third positions. In the ndhF data transversions and third positions both produce well-resolved trees with overall high supports. For atpB and rbcL transversions do not give better support for basal relationships than do transitions, and they show lower average support compared to third positions.
Our results differ from the widely accepted predictions that first and second codon positions and transversions are more reliable estimators of phylogeny than third codon positions and transition, especially for finding older groups. It is in agreement, however, with recently published results from large-scale analyses of rbcL and other genes (Källersjö et al., 1998
; Källersjö, Albert, and Farris, 1999
).
Phylogenetic relationships in the "Primulales" group
The phylogenetic trees from all data sets, i.e., a combination of molecular and morphological data (Fig. 1), the three chloroplast genes analyzed in combination (Fig. 2), or separately (Figs. 3–6), all have strong support for the paraphyly of both Primulaceae and Myrsinaceae, as they are currently circumscribed. The discussion will focus on the tree obtained from the largest data set, i.e., the combined molecular and morphological data, which in our view presents the best phylogenetic hypothesis. Where appropriate, we will compare the results with those from morphology, or from each of the genes, respectively. All jackknife frequencies mentioned below refer to the combined morphological and molecular analysis, unless otherwise stated. Seven evolutionary lineages will be discussed, they are (I) Maesa, (II) Samolus and Theophrastaceae, (III) Coris, (IV) Ardisiandra, (V) Cyclamen, (VI) Primulaceae–Lysimachieae, and (VII) Primulaceae–Primuleae.
Lineage I. Maesa is a genus of trees with small, shortly tubular flowers and a semi-inferior ovary. The flowers are arranged in axillary or terminal racemes or panicles. The genus has 100 species occurring in tropical and subtropical areas of the Old World, from East Africa to Japan. Maesa was once placed in the Primulaceae tribe Samoleae (Reichenbach, 1828
), but has more often been treated as a tribe or subfamily of the Myrsinaceae (de Candolle, 1844
; Pax, 1889
). It is, however, evident that the Myrsinaceae as presently circumscribed (e.g., Cronquist, 1981
; Takhtajan, 1987
; Brummit, 1992
; Mabberley, 1997
) is paraphyletic, since all genera except Maesa are more closely related to the tribe Lysimachieae of Primulaceae. Maesa has a chromosome number based on x = 10, that distinguishes the genus from other Myrsinaceae. Janssonius (1920)
pointed out differences in wood anatomy between Maesa and other Myrsinaceae, and suggested that Maesa does not belong in that family. Anderberg and Ståhl (1995)
also found Maesa to be separate from the Myrsinaceae and argued for raising Maesa to family level. This possibility was also discussed in two studies of molecular data (Morton et al., 1996
; Anderberg, Ståhl, and Källersjö, 1998
), but neither study was conclusive as to the exact systematic position of the genus.
Our study clearly shows that Maesa is the sister group of all other taxa of the former "Primulales" and not closely related to Myrsinaceae s. s. Presence of schizogenous cavities is one of the characters traditionally used to place Maesa in the Myrsinaceae, but given the present results these structures may have evolved in parallel four times, in Maesa and Myrsinaceae, and also in Samolus and Coris.
Lineage II. Samolus is a small genus of perennial or suffrutescent herbs with rather small staminodial flowers and a semi-inferior ovary. The genus has only 15 species, and unlike the other members of the Primulaceae it has its main distribution in the Southern Hemisphere. The species of this genus often grow in saline habitats, such as seashores and salt lakes. Within Primulaceae, Samolus has generally been placed in its own tribe, the Samoleae, diagnosed by a semi-inferior ovary.
The Theophrastaceae are a small group of flowering plants with robust staminodial flowers and fleshy, berry-like fruits. The ovary is superior and the corollas often have a distinct corolla-tube. The family comprises only six genera confined to South and Central America, i.e., Theophrasta L., Clavija Ruiz & Pavón, Jacquinia L., Deherainia Decne., Neomezia Votsch, and Votschia Ståhl. Many of the genera have been monographed or treated in regional surveys (e.g., Ståhl, 1987, 1989, 1991, 1993, 1995
). The Theophrastaceae were once included in Myrsinaceae (Pax, 1889
), but since the beginning of this century, they have been treated as a separate family (Mez, 1902, 1903
), which by some (e.g., Melchior, 1964
) has been considered as the most primitive of the three families of the "Primulales." The family has several synapomorphies, pointed out by Anderberg and Ståhl (1995)
, e.g., pseudoverticillate leaves with extraxylary fibers in bundles, anthers initially forming a cone but later becoming spreading, conspicuous clusters of calcium oxalate crystals in upper and lower part of anther thecae, ovules in more than two series, and endosperm with pitted cell-walls. Extrorse anthers are typical of Theophrastaceae, and in our analysis they constitute a synapomorphy for the family in the strict sense since both Maesa and Samolus have introrse anthers (Anderberg and Ståhl, 1995
).
Our analyses show the Theophrastaceae as a well-supported group, further characterized by a three-base pair insertion in the ndhF gene (see Table 3). There is little resolution at the intergeneric level; the three included genera form a basal trichotomy. To elucidate relationships in Theophrastaceae more data are required. A morphological cladistic analysis of the family by Bertil Ståhl (Göteborg) and a more extensive molecular study (Källersjö, unpublished data) are in progress.
The combined analysis (Fig. 1) in the present study places Samolus of the Primulaceae as sister group to the family Theophrastaceae. Anderberg and Ståhl (1995)
suggested that Theophrastaceae were the sister to the two other families, Primulaceae and Myrsinaceae, but did not identify the close phylogenetic relationship between Samolus and Theophrastaceae. In Anderberg, Ståhl, and Källersjö (1998)
a relationship between Samolus and Theophrastaceae was not supported because the relationships with the large Primulaceae–Myrsinaceae clade was unresolved.
Support for a Samolus + Theophrastaceae relationship can be found in all three genes, in atpB, in the first and second codon positions of ndhF, and in the third positions of rbcL. In spite of the conflicting information provided by first and second codon positions of rbcL, the group is well supported (90%) in the combined molecular tree (Fig. 2). It is worth noting that Samolus does not have the two deletions in ndhF that characterize all other Primulaceae and Myrsinaceae (except Maesa). When all information is combined (Fig. 1) Samolus is still found as a sister to Theophrastaceae, but due to conflicting morphological characters the support is reduced to 60%. In the morphological data one character, presence of staminodial flowers, supports the relationship of Samolus to Theophrastaceae, whereas four characters are shared between Samolus and Primulaceae (excluding Myrsinaceae), viz. herbaceous habit, thin corolla texture, capsular fruit, and capsule opening with teeth. Another character, perigynous flowers, is shared by Samolus and Maesa. However, the morphological characters shared by Samolus and Primulaceae are homoplasious and jackknife analysis of the morphological data alone leaves Samolus unresolved in relation to all other groups. An herbaceous growth habit is known to evolve independently in various groups, and in our analyses, as well as in that of Andeberg and Ståhl (1995)
, herbaceous and frutescent taxa are mixed. Maesa is woody, Samolus and the Primulaceae are herbaceous, Coris is woody, and most Lysimachia are herbs, whereas L. maxima is frutescent. The thin corolla texture of Samolus occurs also in other herbaceous taxa and may have to do with a general reduction in flower size or a short flowering period in temperate biota. The character is relative rather than qualitative and is often difficult to interpret. The fruit in Samolus is a capsule opening with teeth, but the flowers are perigynous and the fruit semi-inferior with a wall partly formed by surrounding tissue.
There is strong support for a close relationship between Samolus and the Theophrastaceae in the three-gene analysis. Based on the presently available information we suggest that Samolus should be treated as part of Theophrastaceae. It may be considered morphologically aberrant in that family, but even when included in Primulaceae it was considered aberrant.
Lineage III. Coris is a genus of Primulaceae with one or two species distributed in the Mediterranean region and Somalia. The species are subshrubs with ericoid leaves and flowers with an epicalyx, schizogeneous cavities on the calyx lobes, a zygomorphic corolla, and ovules in one row, immersed in the placenta. The autapomorphies of this genus have made it difficult to place in relation to other genera of the family and have also led to it being placed in a separate tribe, Corideae. It has also been suggested that Coris is misplaced in Primulaceae (e.g., Sattler, 1962
; Dahlgren, 1989
), but this has been shown not to be the case (Ronse Decraene, Smets, and Clinckemailie, 1995
; Anderberg, Trift, and Källersjö, 1998
). In Anderberg, Ståhl, and Källersjö (1998)
, Coris grouped with the Cyclamen + Lysimachieae + Myrsinaceae clade, and this is confirmed by our present study. The same topology is found in all gene trees, and in the combined morphological and molecular analysis (Fig. 1) it has a support of 100%. It is interesting to note that in a detailed study of vegetative anatomy in Primulaceae, Decrock (1901) concluded that Coris is a xeromorphic representative closely related to Lysimachia. This conclusion is supported by observations in pollen morphology where Coris has been shown to have colpi surrounded by a prominent margo, as in some genera of the Lysimachieae (Carrion, Delgado, and Garcia, 1993
). The corolla aestivation in Coris is imbricate, as in Primulaceae proper and in Ardisiandra. However, imbricate corolla aestivation is apparently a symplesiomorphy, whereas twisted corolla aestivation, which is predominant in Myrsinaceae s. s., Cyclamen, and most taxa of the Primulaceae–Lysimachieae, is a derived feature.
Lineage IV. Ardisiandra is a small genus of three species occurring in the mountains of East Africa. They are perennial herbs with more or less trailing stems, alternate petiolate leaves with coarsely dentate lamina, and small white or yellow campanulate flowers with stamens forming a cone. The flowers are arranged in small umbel-shaped racemes in the leaf axils. The corolla aestivation is imbricate (see discussion above). Initially, the ovules are not immersed in the placenta, but towards maturity the placenta becomes enlarged into a spongy tissue that encloses the seeds. The fruit opens with valves or disintegrates irregularly and the fruitwall is thin and sometimes conspicuously transparent. The pollen of Ardisiandra is tricolpate, which is common in Primulaceae, but has no margo. Ardisiandra has most often been considered to be close to Cortusa of Primulaceae–Primuleae, but this idea has been rejected by several authors (e.g., Wendelbo, 1961
; Spanowsky, 1962
; Røsvik, 1969
) and it has been proposed that it constitutes a tribe of its own, the Ardisiandreae (Schwarz, 1963
; Røsvik, 1968
). It has also been suggested that Ardisiandra is closest to Stimpsonia, but Røsvik (1969)
concluded that no support could be found for this view. Material of Stimpsonia has unfortunately not been available for the present study, and its systematic position awaits evaluation.
The association of Ardisiandra with the Cyclamen + Lysimachieae + Myrsinaceae clade is unexpected, but it is supported by information from all three chloroplast genes and receives maximum support in the combined analysis (Fig. 1). A feature shared by Ardisiandra, Coris, Cyclamen, and the genera of Lysimachieae is the shape of the corolla epidermal cells, which are elongated and not isodiametric as in the Primulaceae–Primuleae. The style in Ardisiandra is punctate, another character state shared with Cyclamen and the Lysimachieae, but not with the Primuleae.
Lineage V. Cyclamen is a small genus of 20 species occurring in the Mediterranean region. They are geophytic perennials with a tuberous hypocotyle. The flowers have conspicuously reflexed corolla lobes, and the fruiting pedicel generally coils at maturity. The highly apomorphic genus Cyclamen has been difficult to place within Primulaceae and has always been treated as a separate tribe, Cyclamineae, with unclear affinities (e.g., Røsvik, 1966
). At one point, Pax (1889)
included Dodecatheon in this tribe, but later authors have not followed this suggestion, although Anderberg and Ståhl (1995)
found a close relationship between these two genera in their analysis of morphological data.
Our analyses do not show Cyclamen and Dodecatheon to be closely related; instead, they belong to different clades (Fig. 1). Dodecatheon is nested among Primula of the Primulaceae–Primuleae, while Cyclamen is a single branch in a trichotomy supported at 93%, with Myrsinaceae s. s. and Primulaceae–Lysimachieae being the other two groups. The similarities between Cyclamen and Dodecatheon thus seem to be parallelisms that presumably have evolved in response to similar pollination strategies. There are similarities in floral structure and development that support a position of Cyclamen in the Coris + Ardisiandra + Lysimachieae + Myrsinaceae clade. All except Ardisiandra have ovules immersed in the placenta. A twisted corolla aestivation, elongated corolla epidermal cells, entire leaf margins, and lack of articulated trichomes are other character states of Cyclamen that are also typical of Lysimachieae and many Myrsinaceae s. s., and together all these features diagnose them as a group separate from other groups, e.g., the Primulaceae–Primuleae. Two character states in particular link Cyclamen to Myrsinaceae, the late petal development from the stamen-petal primordium (Sundberg, 1982
) and the presence of pitted endosperm cell walls (Woodcock, 1933
; Anderberg and Ståhl, 1995
).
Within the Myrsinaceae clade there is little resolution in the combined analysis, and the same is true for the rbcL analysis, which includes a larger sample of taxa from the Myrsinaceae. Still, the resolution is sufficient for rejecting the idea that the apomorphic mangrove genus Aegiceras can be separated as a family of its own (Aegicerataceae), as suggested by several workers over the years (de Candolle, 1844
; Hutchinson, 1969
; Takhtajan, 1987
; Dahlgren, 1989
). A more detailed study is needed to elucidate the phylogenetic relationships within Myrsinaceae.
Lineage VI. The tribe Lysimachieae of Primulaceae is characterized by entire leaves, an almost rotate, deeply lobed corolla, with twisted corolla aestivation, elongated corolla epidermal cells, and ovules immersed in the placenta. The tribe comprises six genera and 180 species, most of which belong to the genus Lysimachia and occur in China. In our analysis the tribe Lysimachieae is monophyletic (100%) and includes Trientalis, Glaux, Anagallis, and Lysimachia. Two genera, Asterolinon and Pelletiera, were not available for analysis, but they are very similar to each other and share a number of derived features with other Lysimachieae. They were sister groups within the Lysimachia complex in Anderberg and Ståhl (1995)
, and Asterolinon is sometimes included in Lysimachia (e.g., Leblebici, 1978
), which is a genus with much morphological variation. Within the lysimachioid clade, Trientalis is sister to the remaining taxa, which form a group, supported by 100%. Few Lysimachia species were studied but Lysimachia nemorum joins Anagallis in a group with maximum support, indicating that Lysimachia is not monophyletic as presently circumscribed. In general habit, Lysimachia nemorum and a few other species are very similar to Anagallis, being distinguished merely by yellow flowers and capsules opening with valves. Character states such as trailing growth habit, sessile ovate leaves, solitary flowers in the leaf axils, and curving of the fruiting pedicel are all shared with Anagallis. The salient feature of Anagallis is its circumscissile capsule; it is variable in habit. Glaux is monotypic and differs from all other genera of Primulaceae by its apetalous flowers with a pink, petaloid calyx. Apart from its apetalous flowers, the morphological traits of Glaux fit well among species of Lysimachia. Lysimachia is paraphyletic and its species are part of at least two clades, with both Glaux and Anagallis having their closest relatives within Lysimachia. The phylogeny of the variable and widely distributed genus Lysimachia with >100 species, including its relationships to the other genera of Primulaceae–Lysimachieae, i.e., Trientalis, Glaux, Asterolinon, Pelletiera, and Anagallis, is one of our ongoing projects.
Lineage VII. The tribe Primuleae of Primulaceae is characterized by scapose inflorescences, distinctly tubular flowers with campanulate or hypocrateriform corolla, imbricate corolla aestivation, isodiametric corolla epidermal cells, leaves almost always in a basal rosette, and ovules rarely immersed in the placenta. They often have syncolpate or sometimes polycolpate pollen, without margo. The tribe has 13 genera and 600 species, most of which belong to Primula and Androsace and grow in the mountains of Europe and Asia, mainly China. In our study, the tribe Primuleae is monophyletic (100%) and comprises the genera Androsace, Douglasia, Omphalogramma, Soldanella, Dodecatheon, Cortusa, and Primula. Other genera generally classified in Primuleae, but not included in the present investigation are Dionysia, Vitaliana, Hottonia, Bryocarpum, and Pomatosace, which all share features typical of genera in the Primulaceae–Primuleae. The relationships of these genera to other members of the tribe were discussed by Anderberg and Ståhl (1995)
.
In the Primuleae clade, Androsace and Douglasia together constitute the sister group to a large clade (99%) which includes Primula, Dodecatheon, Cortusa, Omphalogramma, and Soldanella. The two latter together form the sister to the remaining taxa, which in turn, are divided into two groups. The genera Dodecatheon and Cortusa both have their closest relatives in Primula, which implies that the latter is paraphyletic. The heterogeneity of Primula was earlier evident in studies of leaf vernation, floral morphology (Richards, 1993), pollen morphology (Wendelbo, 1961
; Spanowsky, 1962
), and morphology of epicuticular waxes (Ditsch and Barthlott, 1997
). A more detailed investigation of generic delimitations and major clades in the Primula complex is being carried out by Ida Trift at the University of Stockholm.
Classification
Several recent investigations provide examples of well-known families being nested within other, apparently paraphyletic "families." Subsequent reclassification will in many cases lead to new family circumscriptions with monophyletic, but often morphologically more heterogeneous families. One such example is Ericaceae in which both Empetraceae and Epacridaceae are small monophyletic subgroups (Anderberg, 1993
; Kron and Chase, 1993
). The small Theligonaceae is part of Rubiaceae (e.g., Bremer, Andreasen, and Olsson, 1995
), and the highly specialized asclepiads have been shown to be a derived group within Apocynaceae (e.g., Sennblad and Bremer, 1996
). Following the principles of cladistic classification, retaining the apomorphic ingroup as families would lead to considerable splitting of paraphyletic groups into several new families. By merging the derived ingroups with their paraphyletic counterparts taxonomic stability is maximized.
There are also examples of small groups being recognized as families to avoid losing well-known families. Examples are the small families Diervillaceae (two genera) and Linnaeaceae (five genera), which were described to avoid merging Morinaceae, Dipsacaceae and Valerianaceae with the paraphyletic Caprifoliaceae (Backlund and Pyck, 1998
).
Our present investigation shows that the circumscriptions of the three families Theophrastaceae, Myrsinaceae, and Primulaceae have to be changed if they are to be continued to be recognized as monophyletic families. In addition, Maesa and Samolus belong neither in Myrsinaceae, nor in Primulaceae. This latter problem is easily solved. Maesa can be raised to family level and Samolus can be included in Theophrastaceae. The main problem is, however, the treatment of the highly paraphyletic Primulaceae, a family well known to all. To merge a paraphyletic Primulaceae with Myrsinaceae s. s. would be feasible, but not recommendable since the latter have been recognized consistently since the beginning of the 19th century. If Myrsinaceae is reduced to synonymy under Primulaceae, it is difficult to find arguments for maintaining Theophrastaceae and Samolus as a separate family, and if all three families are merged into a heterogeneous Primulaceae sensu latissimo, there is no reason at all to separate Maesa.
A better alternative is a classification with four separate families, corresponding to the four major clades of the phylogenetic tree, Primulaceae Vent., Myrsinaceae R. Br., Theophrastaceae Link, and "Maesaceae." Three of the four family names are already available and have been in consistent use for more than a century. The idea of recognizing "Maesaceae" was first suggested by de Candolle (1841)
and later also proposed by Anderberg and Ståhl (1995)
. The new family "Maesaceae" will formally be described in a separate paper (Anderberg, Ståhl, and Källersjö, in press
). Recognition of four families is the best way of preserving taxonomic stability, but requires some generic realignments. Samolus is hereby transferred to Theophrastaceae, and Lysimachia, Anagallis, Trientalis, Glaux, Asterolinon, Pelletiera, Coris, Ardisiandra, and Cyclamen are included in Myrsinaceae. A synopsis of the new classification is presented in Table 6 together with the classifications of some earlier workers. Characters for each of the four families is presented in the Appendix.
|
Since classification is traditionally based on visual features, it is inevitable that some will argue that reclassifications and circumscriptions of taxa at higher rank should include at least some visible morphological characters to support it. The dilemma of phylogenetic analyses including molecular data is that although the results may be robust, and hence our best estimate of the evolutionary relationships, they may indicate a close relationship between taxa which may be very different morphologically. From an evolutionary standpoint this is not surprising because every organ, floral morphology, and vegetative structures alike, are susceptible to transformation with time in response to selective pressure. This means that actual close evolutionary relationships will be obscured by differences in aspect. Traditional taxonomy allows for considerable morphological variation in many groups, but tend to emphasize shared presence of certain characters that are assumed to be more "important," particularly floral structures.
The monophyly of the four families Maesaceae, Theophrastaceae, Primulaceae, and Myrsinaceae, as circumscribed here, is well supported, and we find that there are both morphological and molecular data to diagnose them. The variation in morphology between the four families is an aesthetic, rather than scientific problem.
|
FOOTNOTES |
|---|
4 Author for correspondence (FAX: 00946-8-5195 42 21, e-mail: arne.anderberg{at}nrm.se ).
