Molecular phylogenetic approach for studying life-history evolution: the ambiguous example of the genus L.

Gilles Bena1,2,3*, Bernard Lejeune2, Jean-Marie Prosperi3 and Isabelle Olivieri1,3 1Institut des sciences de l'Evolution, Place Eugene Bataillon, Universite¨ Montpellier II, cc065, 34095 Montpellier cedex 05, France 2Insitut de Biotechnologie des Plantes, Universite¨ Paris-Sud, Baª timent 630, F-91405 Orsay Cedex, France 3Laboratoire de Ressources Ge¨ ne¨ tiques et d'Ame¨ lioration des Luzernes Me¨ diterrane¨ ennes, INRA Montpellier-Domaine de Melgueil, 34130 Mauguio, France

We present a molecular phylogeny including most species of the genus Medicago L. (). Based on the consensus of the 48 most parsimonious trees, life-history and mating-system characters are mapped, and a putative history of the genus is suggested. The most parsimonious reconstruction suggests an ancestral annual and sel¢ng state, and recurrent evolution towards perenniality and outcrossing. Based on theore- tical predictions and classical hypotheses of the history of the genus, di¡erent assumptions about the ancestral state and di¡erent weighting schemes of evolution between the character states are made. Assuming an outcrossing, perennial ancestral state (partly supported by morphological features) does not fundamentally change the reconstruction. To meet theoretical expectations, various weighting schemes favouring evolution towards annuality and sel¢ng are applied. In£uence and validity of such weighting schemes are discussed with regard to other studies. Keywords: molecular phylogeny; life-history evolution; weighting schemes; Medicago

The goal of the present paper is to examine how mating 1. INTRODUCTION system and life history may have evolved in a group of Several theoretical studies have addressed the evolution of related species. We chose to study the genus Medicago. This mating systems (e.g. Lande & Schemske 1985; appears particularly well suited for our goal; there are 55^ Uyenoyama 1986; Ronfort & Couvet 1995; Morgan & 61 species in the genus and most have been, at least partly, Schoen 1997) and the evolution of plant life histories studied with regard to mating system and life history (Stearns 1976; Fox 1992; Ronce & Olivieri 1997), whereas (Heyn 1963; Lesins & Lesins 1979; Small & Jomphe empirical studies have concentrated on characterizing 1988). This is most likely because Medicago sativa (alfalfa) plant mating systems (Schemske & Lande 1985; Barrett is an important crop species worldwide (Hanson et al. & Harder 1996, and references therein), studying the 1988). The following three issues will be addressed: consequences of inbreeding depression (Jarne & Charles- 1. Has sel¢ng in this genus evolved from outcrossing (as worth 1993), and measuring the cost of reproduction expected), and, if so, how many times? (Primack 1979). One general conclusion of such studies is 2. Has annuality in Medicago evolved from perenniality that evolution from outcrossing towards sel¢ng is more and, conversely, could perenniality have evolved from likely than the reverse (but see Ronfort & Couvet 1995), annuality? and that evolution towards annuality is easier in sel¢ng 3. Is there any evidence for the coevolution of a given life species than in outcrossing species (Charnov et al. 1981). history with a given mating system? One question that has been rarely addressed is the joint evolution of life-history tactics and mating systems To answer these questions, we built a molecular phylogeny (Olivieri et al. 1994; Zhang & Wang 1994). Although based on the combined sequences of the external (ETS) empirical investigations based on molecular phylogenetic and internal (ITS) transcribed spacer region of the reconstruction should be very useful and allow theoretical nuclear ribosomal DNA (Bena et al. 1998b). predictions to be tested, relatively few studies have yet been carried out (Weller et al. 1995; Barrett et al. 1996; Desfeux et al. 1996; Kohn et al. 1996; Schoen et al. 1997). 2. MATERIALS AND METHODS One reason could be that the mating system is not always (a) Study species well known in groups of species large enough to study such We included 44 of the 55 Medicago species described by Lesins coevolution, but small enough to construct a robust phylo- & Lesins (1979) plus three other species (M. rigiduloides, geny. M. sphaerocarpos and M. syriaca) described by Small (1990), Small et al. (1990) and Gillespie & McComb (1991). At least one species *Author for correspondence ([email protected]). from each of the nine sections de¢ned by Lesins & Lesins (1979)

Proc. R. Soc. Lond. B (1998) 265, 1141^1151 1141 & 1998 The Royal Society Received 11 February 1998 Accepted 26 February 1998 1142 G. Bena and others Life-history evolution in Medicago

Table 1. Description of species studied

number of mating life populations speciesa systemb historyc studiedd accession origine herbariumf

M. carstiensis Wulf Out. Per. 2 2595 Bot. Gard. Brussel U 2440 Ottawa Res. Stat. U M. cretacea M. Bieb. Out. Per. 2 2975 Museum Paris U 2739 Coll. Kerguelen U M. falcata L. Out. Per. 2 22382 Russia L 2362 Gatersleben U M. sativa L. Out. Per. 3 DZA059 Algeria L Gabes Tunisia L Cuf101 Cult. L M. glomerata Balbis Out. Per. 1 83001 France L M. cancellata M. Bieb. Out. Per. 2 2713 Gatersleben U 2731 Former USSR U M. pironae Visiani Out. Per. 2 2729 Yugoslavia U 2843 Yugoslavia U M. arborea L. Out. Per. 2 Helios Greece F Melagon Algeria F M. lupulina L. Sel. Per. 3 F20018 France L ES66 Spain L GR66 Greece L M. marina L. Sel. Per. 2 GR00X Greece L Carnon France L M. su¡ruticosa Ramond Sel. Per. 2 FR11008 France L 66003 France L M. hybrida Trautverr. Sel. Per. 1 2446 Coll. Kerguelen U M. secundi£ora Durieu Sel. Ann. 2 DZA311 Algeria L PI537238 France W M. orbicularis L. Sel. Ann. 2 ES009 Spain L ES041 Spain L M. radiata L. Sel. Ann. 1 2589 Spain U M. heyniana Greuter Sel. Ann. 1 12027 Greece A M. rotata Boissier Sel. Ann. 2 SA14095 ? A IC350 ? I M. bonarotiana Arcangeli Sel. Ann. 2 2591 ? U 115216 Israel A M. noeana Boissier Sel. Ann. 2 early Iraq 15487 Iran A M. shepardii Post Sel. Ann. 2 26682 Turkey A 16402 Turkey A M. rugosa Desrousseaux Sel. Ann. 1 GR026 Greece L M. scutellata L. Sel. Ann. 3 GR045 Greece L ES103 Spain L F34044 France L M. soleirolii Duby Sel. Ann. 2 8615 Tunisia A 1334 Malta A M. italica Miller Sel. Ann. 3 ES050 Spain L Murray. Cult. A Torna¢eld Cult. A M. littoralis Rodhe Sel. Ann. 1 ES167 Spain L M. truncatula Gaertn Sel. Ann. 3 Jemalong Cult. A DZA246 Algeria L DZA201 Algeria L M. rigiduloides Small Sel. Ann. 3 IC632 Iraq I IC716 Iraq I 128275 Iraq A M. rigidula L. Sel. Ann. 2 ES024 Spain L ES016 Spain L M. murex Willd. Sel. Ann. 3 FR83009 France L DZA042 Algeria L 23199 Greece A M. sphaerocarpos Bertol Sel. Ann. 2 127474 Greece A 131585 Romania A (Continued)

Proc. R. Soc. Lond. B (1998) Life-history evolution in Medicago G. Bena and others 1143

Table 1. (Continued)

number of mating life populations speciesa systemb historyc studiedd accession origine herbariumf

M. constricta Durieu Sel. Ann. 3 117743 Syria A 116288 Malta A GR068 Greece L M. turbinata L. Sel. Ann. 2 GR020 Greece L GR035 Greece L M. doliata Carmign. Sel. Ann. 2 ES047 Spain L DZA231 Algeria L M. sauvagei Ne© gre Sel. Ann. 1 11476 Morocco A M. laciniata L. Sel. Ann. 1 DZA239 Algeria L M. minima L. Sel. Ann. 1 DZA013 Algeria L M. praecox de Candolle Sel. Ann. 2 GR042 Greece L F20059 France L M. coronata L. Sel. Ann. 1 GR062 Greece L M. syriaca Small Sel. Ann. 2 124962 Turkey A 128437 Syria A M. polymorpha L. Sel. Ann. 3 ES103 Spain L CircleV. Cult. A F34003 France L M. arabica L. Sel. Ann. 2 ES057 Spain L GR067 Greece L M. disciformis de Candolle Sel. Ann. 1 GR023 Greece L M. tenoreana Seringe Sel. Ann. 1 4639 France A M. intertexta L. Sel. Ann. 1 DZA027 Algeria L M. ciliaris L. Sel. Ann. 1 ES101 Spain L M. muricoleptis Tineo Sel. Ann. 1 GR021 Greece L M. granadensis Willd. Sel. Ann. 1 2600 Coll. Lesins U Mel. o¤cinalis L. Out. Bia. 1 443 France L Mel. italica L. ? Ann. 1 FR0300 France L Trig. foenum-graecum L. Sel. Ann. 1 Nimes France L Trig. cretica L. ? Ann. 1 CR007 Crete L Trif. pratense L. Out. Per. 1 Alpilles France L Trif. subterraneum L. Sel. Ann. 1 Argeles France L a M., Medicago; Mel., Melilotus;Trig.,Trigonella;Trif.,Trifolium. Authorities are given according to Lesins & Lesins (1979), Small & Jomphe (1988), Small (1989,1990) and Gillespie & McComb (1991). b Out.ˆoutcrosser, Sel.ˆselfer, ?ˆunknown. c Per.ˆperennial, Bia.ˆbiannual, Ann.ˆannual. Life-history and mating-system characters were obtained from literature (Heyn 1963; Lesins & Lesins 1979; Small 1986b) and from personal observations. d One or two individuals per population included in the study were sequenced and analysed in the phylogenetic reconstruction, and at least two individuals per species. e Cult., cultivated varieties; Coll., collection. f L, Laboratoire de Ressources Ge¨ ne¨ tiques et d'Ame¨ lioration des Luzernes Me¨ diterrane¨ ennes (sampling byJ. M. Prosperi in natural popu- lations); A, Australian Medicago Genetic Resource Centre;W,Washington Regional Plant Introduction Station; U, Unite¨ d'Ame¨ lioration des Plantes Fourrage© res de Lusignan; F, Forest Institut Research of Thessalonique; I, International Centre for Agriculture Research in DryAreas. was included in the study. Thirty-¢ve of the 47 species studied (see Baldwin et al. (1995) for a review for the use of ITS in phylo- have been described as annual selfers, four as perennial selfers genetic reconstruction, and Bena et al. (1998a) for the use of and eight as perennial outcrossers (Lesins & Lesins 1979; G. ETS). Phylogenetic trees were constructed by Fitch parsimony Bena, personal observations) (table 1). Two species from each of methods (Fitch 1971). Heuristic searches for the most parsimo- the three genera Melilotus, Trigonella and Trifolium were used as nious trees were carried out using the program PAUP (Swo¡ord outgroups (see Bena et al. 1998b). These genera are generally 1993). recognized as the closest genera related to Medicago (Heyn 1981). At least two individuals were studied for each species, taken (c) Estimation of mating-system and life-history whenever possible from di¡erent populations. traits Mating-system and life-history traits of all the species included (b) Phylogenetic analysis in our study were recorded from the literature (Heyn 1963; DNA polymerase chain reaction (PCR) ampli¢cation, sequen- Lesins & Lesins 1979; Small 1986b) and from our own observa- cing, alignment and phylogenetic analysis were performed as tions. Outcrossing and sel¢ng rate have been proved to be highly described in Bena et al. (1998b). ETS and ITS sequences of variable in time and space in many species (Richards 1997). One nuclear rDNA were analysed by combining the two data sets study in the genus Medicago (Gjuric & Smith 1996), using RAPD

Proc. R. Soc. Lond. B (1998) 1144 G. Bena and others Life-history evolution in Medicago markers, has estimated a 27% sel¢ng rate in autotetraploid- could not be resolved with the present molecular data. The Medicago sativa. However, Brouwer & Osborn (1997), using topology of the tree does not coincide with classical taxo- RAPD markers, have shown that the lowest observed hetero- nomic descriptions of the genus. Such incongruence zygosity in diploid Medicago sativa after four selfed generations between classi¢cation based on morphological character- was 30%, despite expectations as low as 6%. They suggested istics and molecular phylogenetic relationships are that inbreeding depression may strongly limit the level of homo- discussed for annual species in Bena et al. (1998b). zygosity obtainable. None of our own observations has shown any evidence of a mixed mating system in Medicago species, which (b) Mating-system and life-history evolution in either have a normal seed set or do not produce any seed in the Medicago absence of a pollinator.We did not take into account the potential Figure 2a shows the most parsimonious reconstruction e¡ect of mating system polymorphism in the phylogenetic of the evolution of the mating system in Medicago. This reconstruction. However, the possible in£uence of a residual tree suggests that sel¢ng was the primitive state. The most outcrossing in sel¢ng species on our hypotheses is considered in parsimonious reconstruction involves two separate losses the last part of the discussion. of sel¢ng (i.e. towards outcrossing) and one reversal towards sel¢ng, leading to M. marina. A single species in (d) Reconstruction of the evolution of characters the basal polytomy, M. carstiensis (clade 3), exhibits a The evolution of life history and mating system in Medicago di¡erent mating system to the others (i.e. outcrossing as was studied by mapping the characters of each species onto the opposed to sel¢ng); mapping the breeding system was not consensus tree using the program MacClade (Maddison & su¤cient for resolving the polytomy. Maddison 1992). Because we had no reason for choosing any Figure 2a,b shows two of the most parsimonious recon- particular one of the 48 most parsimonious trees, we decided structions of the evolution of the life history of Medicago. It to map character onto the strict consensus tree (see Schoen et undoubtedly suggests that annuality is the ancestral char- al. (1997) for a similar analysis). In the ¢rst instance, no acter state. When life history is treated as an unordered, assumptions were made about ancestral states and character equally weighted character, three alternate reconstruc- evolution. Ancestral states were estimated using a parsimonious tions involving four steps are found. The ¢rst one (¢gure approach. The simplest assumption of character evolution was to 2a) suggests four di¡erent gains of perenniality, with no estimate that changes in either direction between the di¡erent example of reversal towards annuality. The second recon- states of a character were equally likely. We considered that struction (¢gure 2b) could not ¢rmly resolve the basal evolution from annuality to perenniality was as likely as the polytomy, suggesting that M. carstiensis clustered within reverse, and that evolution from sel¢ng to outcrossing was also the M. su¡ruticosa^M. disciformis clade (clade 2), with a as likely as the reverse. The results were then reanalysed by ¢rst evolution towards perenniality and two reversals making assumptions that were more consistent with theoretical towards annuality. The last reconstruction (not shown) expectations. also suggested that M. carstiensis was in the same clade as the M. su¡ruticosa^M. disciformis clade, with evolution at (e) Resolution of polytomies ¢rst towards perenniality, a later reversal towards The consensus tree contains a number of polytomies (¢gure annuality in the M. lupulina^M. disciformis clade (clade 1), 1). We interpreted these as being soft rather than hard and evolution once more towards perenniality leading to polytomies (i.e. polytomies were considered to indicate M. lupulina. uncertainties about the evolutionary relationships, rather than Figure 2c,d shows the evolution of mating system and multiple-speciation events (Maddison 1989)). These polytomies life cycle under two alternative weighting schemes. When were resolved, whenever possible, under the assumption of the mating system is treated as a weighted character, and maximum parsimony when mapping characters onto the considering that the evolution from outcrossing towards cladogram. Trees requiring the minimum number of steps for sel¢ng is six times more likely than the reverse, the most each character were retained for the study and compared with parsimonious reconstruction of the mating system implies each other (Schoen et al. 1997). that sel¢ng has evolved independently nine times (¢gure 2c). An intermediate reconstruction, with a weighting scheme involving a 1:3 ratio, involves four gains of sel¢ng 3. RESULTS and one reversal towards outcrossing (¢gure 2d). When life (a) Phylogenetic analysis history is treated as a weighted character, and assuming a All sequences have been deposited in the EMBL nucleo- 1:6 ratio favouring evolution towards annuality, the most tide database (accession numbers Z99207^Z99254, parsimonious reconstruction implies only gains towards Z97655^Z97729 and Z92912^Z92950). The data for most annuality (¢gure 2c). An intermediate ratio (1:3) leads to annual Medicago species are being published and discussed an intermediate reconstruction (¢ve evolutions towards elsewhere (Bena et al. 1998b). Using only informative sites annuality and one reversal towards perenniality, ¢gure and including the six outgroups, this analysis resulted in 2d). the production of 48 equally parsimonious trees, each of 780 steps in length, with a consistency index of 0.650 and a retention index of 0.798. The strict consensus is shown in 4. DISCUSSION ¢gure 1. This tree is well resolved and several of the clades (a) Parsimonious reconstruction and ancestral state within it are supported by high bootstrap values. However, Mapping mating system and life history onto the a basal polytomy involving four clades (i.e. M. granadensis^ consensus tree with an assumption of unordered character M. orbicularis, M. su¡ruticosa^M. disciformis, M. arborea^M. states leads to an acceptance, under a hypothesis of parsi- secundi£ora and M. carstiensis alone; clades 1^4, see ¢gure 1) mony, of an ancestral, annual, sel¢ng species. This result

Proc. R. Soc. Lond. B (1998) Life-history evolution in Medicago G. Bena and others 1145

Figure 1. The strict consensus tree of 48 equally parsimonious trees, based on ITS and ETS sequence data, reconstructing phylo- genetic relationships within Medicago. Numbers along branches represent the percentage of times a monophyletic group occurred in 1000 bootstrap replicates. Stars ( ) indicate a 100% value. Numbers in front of some clades indicate the four clades involved in the basal polytomy (see text). disagrees with all classical hypotheses on the evolution of gical argument supports the latter hypothesis: the tripping the genus, according to which the most recent common mechanism, the obvious purpose of which is to ensure cross- ancestor to all species in the genus Medicago was perennial pollination, which is characteristic of the entire genus Medi- and outcrossing (Lesins & Lesins1979). A major morpholo- cago (Lesins & Lesins 1979). When an insect tries to reach

Proc. R. Soc. Lond. B (1998) 1146 G. Bena and others Life-history evolution in Medicago

Figure 2. Reconstructions of mating-system and life-history evolution using various ancestral states and weighting scheme hypotheses. Reconstruction of character evolution was performed using MacClade (Maddison & Maddison 1992). (a) and (b) use unordered and equally weighted optimization schemes and show two of the three most parsimonious trees; (c) uses an unordered `6:1' optimization scheme for both characters, whereas (d) uses an unordered `3:1' optimization scheme. Both (c) and (d) favour evolution towards annuality and sel¢ng. A solid square indicates a sel¢ng species, whereas an open square indicates an outcrossing species. An open circle indicates a perennial species, and a solid circle indicates an annual species. A star across a branch indicates evolution towards annuality from perenniality, and a cross indicates that perenniality has evolved towards annuality. Branch labels indicate mating-system evolution, with solid branches indicating sel¢ng, and dashed branches indicating outcrossing. A change along a branch indicates a change between mating systems.

Proc. R. Soc. Lond. B (1998) Life-history evolution in Medicago G. Bena and others 1147

Figure 2. the nectaries at the base of the staminal column, which mechanism is also present in three sections of the genus comprises nine fused stamens, the column shoots out so Trigonella, although Small (1986a), based on several charac- that the anthers and stigma touch the insect's body, strewn teristics (mainly leaf and fruit morphological features), with pollen from visits to previous £owers, thereby accom- transferred these Trigonella sections to Medicago. The trip- plishing cross-pollination (Lesins & Lesins 1979). This ping mechanism is absent from the genera Melilotus and

Proc. R. Soc. Lond. B (1998) 1148 G. Bena and others Life-history evolution in Medicago

Trifolium. Thus, it can then be treated not only as ancestral (Lande & Schemske 1985). Indeed, once a species is but also derived in the genus Medicago (i.e. apomorphic for inbred, there is no longer any inbreeding depression, the tribeTrifolieae but plesiomorphic for the genus Medicago). because of the purge of deleterious alleles, which would Sel¢ng Medicago species have the same tripping mechanism compensate for the twofold cost of meiosis in outcrossing as outcrossers, but it is super£uous as they self-pollinate and species. Inbreeding depression is known to be very high sel¢ng occurs before the tripping mechanism is operational. in M. sativa (the cultivated alfalfa). Therefore, evolution For some species, this mechanism has degenerated, but from sel¢ng towards outcrossing, although possible morphological evidence remains. As a sel¢ng species under some particular demographic conditions (see would have no reason for developing such structures, their Ronfort & Couvet 1995), would be less likely than the presence argues strongly in favour of an ancestral reverse. Another point supporting preferential evolution outcrossing state for the genus. towards sel¢ng is the presence of a self-incompatibility Similarly, mapping life history onto the consensus tree mechanism in Medicago sativa. One study has dealt with leads to an acceptance, under a hypothesis of parsimony, of this problem in the genus Medicago. Bauchan et al. (1991) ancestral annuality rather than perenniality. Unlike the have shown that there was partial self-incompatibility in mating system, we have no morphological evidence to indi- Medicago sativa, in that self-pollen tubes grow more slowly cate the ancestral state of the genus, and the only way to do than those from outcrossing fertilizations, and many fail this is via parsimonious reconstruction. As both Trigonella to reach the ovule. It has often been assumed that the and Melilotus contain perennial and annual species, neither breakdown of self-incompatibility was more straightfor- can we suggest a hypothesis based on the outgroups. ward than the reverse, given the evidence for repeated However, according to many authors (e.g. Davis & origin of self-compatibility from self-incompatibility Heywood 1963; Quiros & Bauchan 1988), the ancestor of within some angiosperm families (Weller et al. 1995). Medicago was perennial.This hypothesis is based mainly on This suggests that it is easier to lose than to gain self- a strong association between life history and mating system incompatibility. Even if we have no evidence of self- in the genus Medicago, as all annuals are selfers, whereas incompatibility in other predominantly outcrossing perennials can be selfers or outcrossers. The presence of the species, this result again supports a weighting scheme tripping mechanism leads to an assumption of an ancestral favouring evolution towards sel¢ng. Similarly, one might outcrossing state. The assumption of an ancestral annual consider that evolution from perenniality to annuality state would imply that all ancestral, annual, outcrossing would be more likely through increased resource alloca- species have gone extinct as all of the annual species of the tion to reproduction and loss of survival function, rather genus are selfers. Under the assumption of ancestral peren- than the reverse, because the plant would need to `learn' niality, however, M. carstiensis, which is present in the basal how to allocate resource for summer survival. part of the tree and which has often been considered as an These arguments indicate that an unordered and ancestral species (Quiros & Bauchan 1988), could be one of equally likely optimization scheme is unlikely to re£ect the ¢rst-emerging species. However, it has to be assumed the relative likelihood of the gain versus loss of characters that most of the ¢rst-emerging, perennial, outcrossing such as sel¢ng and annuality. We will, therefore, now species have either become extinct or are not included in explore the evolution of the mating system and life cycle our study. under a variety of weighting schemes.

(b) Reconstruction under alternate ancestral state (c) Reconstructions under alternate weighting The history of the genus was reconstructed using several schemes new assumptions of ancestral state. Assuming that peren- A model of evolution that is more restrictive than the niality and outcrossing are the ancestral states, the most simple one (i.e. unordered parsimony) always reconstruction leads only to one gain of sel¢ng and requires more evolutionary steps (Meyer 1996). One annuality at the base of the tree and implies recurrent alternative to unordered parsimony is to consider a reversals towards perenniality and outcrossing (recon- weighting scheme in which the evolution of one character struction not shown). This second reconstruction, under towards another is less likely than the reverse. However, alternative ancestral assumptions, does not modify the we have no a priori idea about the weighting value to be history of the genus, as it only adds one step at the base of used in the evolution of mating-system and life-history the tree for each character. However, it does suggest that traits in the genus Medicago, apart from using the theore- complementary data, such as morphological characters or tical expectations described above. Therefore, we palaeontological data, would be necessary to determine increased the weight favouring evolution towards ancestral states in the genus Medicago. annuality and sel¢ng until we observed a switch in the Both reconstructions (i.e. with and without assump- parsimonious reconstruction of the evolution of each tions concerning ancestral state) show a recurrent character. We increased these weightings until no more appearance of perenniality and outcrossing from change could be observed (i.e. the reconstruction of annuality and sel¢ng. However, these hypotheses evolution no longer presented any reversal towards peren- disagree with general theoretical expectations concerning niality and outcrossing). A 6:1 weighting scheme the evolution of mating systems and life histories. These favouring evolution towards sel¢ng and annuality studies have shown that many factors promote sel¢ng implies only gain towards those two characters, with no under a great variety of circumstances, such as reproduc- example of reversal. A 3:1 intermediate weighting tive insurance or the maintenance of good genotypes (see scheme involves both gain and loss of annuality and Richards 1997). Absence of inbreeding depression also sel¢ng (i.e. reconstruction also implies reversals towards favours the evolution from outcrossing towards sel¢ng perenniality and outcrossing).

Proc. R. Soc. Lond. B (1998) Life-history evolution in Medicago G. Bena and others 1149

(d) In search of a relevant weighting scheme reconstruction, under an assumption of equally weighted Assuming such weighting schemes allows one to meet optimization, agreed with theoretical expectations. theoretical predictions. However, the methodological and According to Barrett et al. (1996), virtually all phyloge- biological signi¢cance of such weighting is still unclear. netic studies on the evolution of plant reproductive traits Considering that changes between the two states are have treated character shift as equally likely, presumably unordered and weighted is a ¢rst step that is classically because the procedure is simpler. However, they also used in such studies (Weller et al. 1995; Kohn et al. 1996; recognized that, when biological evidence indicates an Schoen et al. 1997), but estimating the weighting to be used unequal probability of possible changes, workers should at is much more di¤cult. Many factors may have played a role least explore how the character optimization scheme in the evolution of the genus and cannot be easily quanti¢ed. employed in£uences historical reconstruction (Barrett et Weller et al. (1995) faced the same problem when studying al. 1996). It seems quite unrealistic to give a single value the evolution of self-compatibility in £owering . In for di¡erent studies, or even for an entire phylogeny in their analysis, weighting against the independent loss of which variable factors may have come into play. We think self-incompatibility would tend to favour placing self- that neither empirical nor theoretical studies can suggest incompatibility at the base of the angiosperms, whereas any weighting value let alone the problem of the existence results based on equal weighting imply that evolution of such a value. between the two breeding systems is similar in both directions and that it is more parsimonious to suppose that (e) Alternative factors which may in£uence these self-compatibility is ancestral in angiosperms. They raised results the question of how one could determine the appropriate Unless we use an unequal weighting scheme our conclu- weighting scheme, butconcludedthat other lines of evidence sions disagree with theoretical predictions.We now explore and studies should be used to provide a relevant answer. alternative reasons for such incongruences. Ambiguities in One major problem is that few comparable studies have our results could result from three main factors. been conducted, and not all of them have shown the predominance of evolution towards autogamy and (i) Inferring species phylogeny from gene phylogeny annuality or the need of a weighting scheme favouring We must keep in mind that the tree we produced is a loss of outcrossing and perenniality. Based on an unor- gene phylogeny, from which we infer a species phylogeny. dered, equally likely scheme, Barrett et al. (1996) This gene tree could be fundamentally incongruent with concluded that autogamy in Polemoniaceae evolved at the true species phylogeny owing to, for example, intro- least 14 times from outcrossing, with no evidence of gression and horizontal genetic transfer (see Doyle 1992; reversal. They also concluded that an annual life cycle Lawrence & Hartl 1992). A gene tree could also be incon- had evolved at least seven times, and that reversals from gruent with the true species phylogeny because of the an annual to a perennial habit had apparently occurred parsimony method used in the study. However, alternative at least three times. They concluded that the recurrent methods, such as maximum likelihood (Felsenstein 1981), transitions towards annuality and perenniality suggest a could not be performed, owing to the large number of considerable instability of life cycle, at least among herbac- taxa included in the study. The in£uence of heterogeneity eous species. Armbruster (1993) investigated the evolution upon the rate of evolution in the reconstruction was not of mating systems in Dalechampia. An unordered, equally explored, but could have played a role (i.e. `positively likely scheme suggests that autogamy and allogamy have misleading' (Felsenstein 1978), or `long-branch attraction' originated independently several times. Schoen et al. phenomenon (Hendy & Penny 1989)). However, in this (1997), using two di¡erent weighting schemes (1:1 and 2:1 case, the inconsistency would be due to strongly unequal weighting) and di¡erent ancestral states, found alternate rates of change along di¡erent branches, which does not reconstructions involving gains or losses of distyly, which seem to be the case in our study. is closely associated with outcrossing, whereas homostyly is closely associated with sel¢ng. Kohn et al. (1996) (ii) Extinction of basal outcrossing and perennial lineages studied the evolution of reproductive characters in Ponte- Random extinction of some ancestral perennial deriaceae. They concluded that the class of optimization outcrossing species could explain why reconstruction of scheme used was the most critical factor in reconstructing the phylogenetic relationships among surviving species character evolution. With unordered, unequally weighted led to a most parsimonious criterion that suggested an schemes, they suggested that tristyly originated once or ancestral, sel¢ng, annual state. However, such a problem twice, that tristyly had repeatedly broken down leading would seem impossible to resolve. to self-fertilizing, homostylous taxa and that self-incom- patibility probably arose after the origin of £oral (iii) Theoretical expectations might be wrong trimorphism. Use of an unordered, equally weighted We based our study mainly on theoretical models that scheme was also explored, but not used, as several lines of suggest that reversal from sel¢ng to outcrossing is very evidence indicate that this optimization scheme is unlikely unlikely. Indeed, sel¢ng has often been described as an to re£ect the likelihood of the gain versus the loss of a evolutionary dead-end (Stebbins 1974; Barrett et al. 1996). complex polymorphism such as tristyly (Kohn et al. 1996). Lynch et al. (1995) and Schoen et al. (1997) argued that a These examples clearly indicate that empirical investi- greater mutation load in sel¢ng species would decrease gations cannot be used to suggest any ordering or population growth rates and increase the probability of weighting scheme for use in our phylogeny. Some of them extinction of sel¢ng lineages, through mutational melt- did not investigate the in£uence of various weighting down. However, a greater ability of selfers to colonize schemes on the reconstruction, especially when the new sites may also play a great role, and balance the

Proc. R. Soc. Lond. B (1998) 1150 G. Bena and others Life-history evolution in Medicago expected decrease of ¢tness in sel¢ng lineages. Schoen et al. Baldwin, B. G., Sanderson, M. J., Porter, J. M.,Wojciechowski, M. (1997) also suggested that some residual outcrossing in a F., Campbell, C. S. & Donoghue, M. J. 1995 The ITS region of predominantly sel¢ng species could greatly lower the nuclear ribosomal DNA: a valuable source of evidence on mutation load and help explain the maintenance of ances- angiosperm phylogeny. Ann. Missouri Bot. Gard. 82, 247^277. tral sel¢ng lineages (see also Charlesworth et al. 1993). In Barrett, S. C. H. & Harder, L. D. 1996 Ecology and evolution of plant mating.Trends Ecol. Evol. 11, 73^79. M. truncatula, for example, estimates of sel¢ng rates vary Barrett, S. C. H., Harder, L. D. & Worley, A. C. 1996 The from 97% to 99% (Bonnin et al. 1996b); this very low comparative biology of pollination and mating in £owering frequency of outcrossing might be su¤cient to generate plants. Phil.Trans. R. Soc. Lond. B 351, 1271^1280. the very large genetic variability that was observed in Bauchan, G. R., Campbell,T. A., O'Neill, N. R. & Elgin, J. H. Jr this species (Bonnin et al. 1996a,b). However, whether this 1991 Self-incompatibility in two alfalfa populations. Crop Sci. level of outcrossing is large enough for maintaining a 30, 1205^1210. potential evolution towards allogamy remains an open Bena, G., Jubier, M. F., Olivieri, I. & Lejeune, B. 1998a question (as pointed out by Husband & Schemske 1996). Ribosomal external and internal transcribed spacers: On the other hand, we have no idea about the achieved combined use in the phylogenetic analysis of Medicago sel¢ng rate in outcrossing species and its in£uence on our (Leguminosae). J. Molec. Evol. 46, 299^306. Bena, G., Prosperi, J. M., Lejeune, B. & Olivieri, I. 1998b conclusions. In some of these species, £owers occur in a Evolution of annual species of the genus Medicago: a molecular raceme and insects may transfer pollen between £owers phylogenetic approach. Molec. Phyl. Evol. (In the press.) of the same plant. Equilibrium between partial self-incom- Bonnin, I., Huguet, T., Gherardi, M., Prosperi, J. M. & Olivieri, patibility, geitonogamous sel¢ng and inbreeding I. 1996a High level of polymorphism and spatial structure in a depression remains unresolved. sel¢ng plant species, Medicago trunculata (Leguminosae), using RAPDs markers. Am. J. Bot. 83, 843^855. (f) Conclusion Bonnin, I., Prosperi, J.-M. & Olivieri, I. 1996b Genetic markers More and more studies involving character history and quantitative genetic variation in Medicago trunculata reconstruction based on molecular phylogenies are being (Leguminosae): a comparative analysis of population struc- published (Harvey et al. 1996). In most cases, and in ture. Genetics 143, 1795^1805. particular in the present paper, authors have empirical Brouwer, D. J. & Osborn, T. C. 1997 Molecular marker analysis of the approach to homozygosity by sel¢ng in diploid alfalfa. and theoretical expectations about the results of their Crop Sci. 37, 1326^1330. studies. This clearly in£uences the results obtained and Charlesworth, D., Morgan, M. T. & Charlesworth, B. 1993 the methods of analysis used. An exploration of several Mutation accumulation in ¢nite outbreeding and inbreeding possibilities, and in particular many weighting schemes, population. Genet. Res. 61, 39^56. seem necessary. Theoretical studies can suggest tendency, Charnov, E. L., Bull, J. J. & Mitchell-Olds, S. T. 1981 A note on but no more. In particular, the joint evolution of mating sex and life histories. Am. Nat. 117, 814^818. system and life history should be theoretically explored. Davis, P. H. & Heywood, V. H. 1963 Principles of angiosperm But the evolution of complex characters, such as the repro- . Edinburgh: Oliver & Boyd. ductive system, depends on so many parameters (genetic, Desfeux, C., Maurice, S., Henry, J.-P., Lejeune, B. & Gouyon, ecological, etc.) that it may be hopeless to weight the P.-H. 1996 The evolution of reproductive system in the genus Silene. Proc. R. Soc. Lond. B 263, 409^414. changes in a phylogenetic reconstruction, even at the Doyle, J. J. 1992 Gene trees and species trees: molecular systema- genus level. Theoretical expectations would support the tics as one-character taxonomy. Syst. Bot. 17, 144^163. notion that sel¢ng and annuality have arisen repeatedly. Felsenstein, J. 1978 Cases in which parsimony or compatibility In the genus Medicago, phylogenetic reconstruction, with method will be positively misleading. Syst. Zool. 27, 401^410. no a priori expectation, suggests the reverse. Final recon- Felsenstein, J. 1981 Evolutionary trees from DNA sequences: a structions are highly dependent of the weighting scheme maximum likelihood approach. J. Molec. Evol. 17, 368^376. and ancestral state chosen, which would best be viewed as Fitch,W. M.1971Toward de¢ning the course of evolution: minimal working hypotheses rather than as ¢rmly based hypoth- change for a speci¢c tree topology. Syst. Zool. 20, 406^416. eses. It seems impossible to solve the problem if one is Fox, G. A. 1992 Annual plant life histories and the paradigm of willing to avoid any tautology in the approach. Our study resource allocation. Evol. Ecol. 6, 482^499. thus points to the limit in the usefulness of phylogenies to Gillespie, D. J. & McComb, J. A. 1991 Morphology and distribu- tion of species in the Medicago murex complex. Can. J. Bot. 69, study the evolution of complex characters such as mating 2655^2662. systems and life histories. Gjuric, R. & Smith, S. R. 1996 Identi¢cation of cross-pollinated and self-pollinated progeny in alfalfa through RAPD nulliplex We thank M. Kirkpatrick, D. Schoen, T. Bataillon, J. Ronfort, E. loci analysis. Crop Sci. 36, 389^393. Jenczewski, O. Ronce and two anonymous reviewers for helpful Hanson, A. A., Barnes, D. K. & Hill, R. R. (eds) 1988 Alfalfa and discussions and comments on the manuscript.We are very grateful alfalfa improvement (Agronomy Monograph no 29). Madison, to E. Besin, M. F. Jubier and I. Gy for technical assistance. This is WI: American Society of Agronomy. contribution 98-048 of the Institut des Sciences de l'Evolution de Harvey, P. H., Leigh Brown, A. J., Maynard Smith, J. & Nee, S. Montpellier (UMR 5554 CNRS). This work was supported by (eds) 1996 New uses for new phylogenies. New York: Oxford the GREG (no. 42-95). G.B. was supported by a grant from the University Press. French Ministry of `Enseignement Supe¨ rieur et Recherche'. Hendy, M. D. & Penny, D. 1989 A framework for the quantitative study of evolutionary trees. Syst. Zool. 38, 297^309. Heyn, C. C. 1963 The annual species of Medicago. Scr. REFERENCES Hierosolymitana 12, 1^154. Armbruster, W. S. 1993 Evolution of plant pollination systems: Heyn, C. C. 1981 Trifoliae. In Advances in legume systematics, part 1 hypotheses and tests with the neotropical vine Dalechampia. (ed. R. M. Polhill & P. H. Raven), pp. 383^385. Kew: Royal Evolution 47, 1480^1505. Botanic Gardens.

Proc. R. Soc. Lond. B (1998) Life-history evolution in Medicago G. Bena and others 1151

Husband, B. C. & Schemske, D. W. 1996 Evolution of the magni- Ronce, O. & Olivieri, I. 1997 Evolution of reproductive e¡ort in a tude and timing of inbreeding depression in plants. Evolution metapopulation with local extinctions and ecological succes- 50, 54^70. sion. Am. Nat. 150, 220^249. Jarne, P. & Charlesworth, D. 1993 The evolution of the sel¢ng Ronfort, J. & Couvet, D. 1995 A stochastic model of selection rate in functionally hermaphrodite plants and animals. A. Rev. on sel¢ng rates in structured populations. Genet. Res. 65, Ecol. Syst. 24, 441^466. 209^222. Kohn, J. R., Graham, S.W., Morton, B., Doyle, J. J. & Barrett, S. C. Schemske, D.W. & Lande, R. 1985 The evolution of self-fertiliza- H.1996 Reconstruction of the evolution of reproductive characters tion and inbreeding depression in plants. II. Empirical in Pontederiaceae using phylogenetic evidence from chloroplast observations. Evolution 39, 41^52. DNA restriction-site variation. Evolution 50,1454^1469. Schoen, D., Johnston, M. & L'Heureux, A. 1997 Evolutionary Lande, R. & Schemske, D.W. 1985 The evolution of self-fertiliza- history of the mating system in Amsinckia (Boraginaceae). tion and inbreeding depression in plants. I. Genetic models. Evolution 51, 1090^1099. Evolution 39, 24^40. Small, E. 1986a A taxonomic study of the `medicagoid' Trigonella Lawrence, J. G. & Hartl, D. L. 1992 Inference of horizontal (Leguminose). Can. J. Bot. 65, 1199^1211. genetic transfer from molecular data: an approach using the Small, E. 1986b Pollen^ovule pattern in tribe Trifoliae bootstrap. Genetics 131, 755^760. (Leduminosae). Pl. Syst. Evol. 160, 195^205. Lesins, K. A. & Lesins, I. 1979 Genus Medicago (Leguminosae), a Small, E. 1989 Medicago rigiduloides, a new species segregated taxogenetic study. The Hague: W. Junk. from M. rigidula. Can. J. Bot. 68, 2614^2617. Lynch, M., Conery, J. & Bu« rger, R. 1995 Mutational meltdown Small, E. 1990 Medicago syriaca, a new species. Can. J. Bot. 68, in sexual populations. Evolution 49, 1067^1080. 1473^1478. Maddison, W. P. 1989 Reconstructing character evolution on Small, E. & Jomphe, M. 1988 A synopsis of the genus Medicago polytomous cladograms. Cladistics 5, 365^377. (Leguminosae). Can. J. Bot. 67, 3260^3294. Maddison, W. P. & Maddison, D. R. 1992 MacClade, version 3.0. Small, E., Brookes, B. & Crawford, E. J. 1990 Intercontinental Sunderland, MA: Sinauer. di¡erentiation in Medicago rigidula. Can. J. Bot. 68, 2607^2613. Meyer, A. 1996 The evolution of body plans: HOM/Hox cluster Stearns, S. C. 1976 Life-history tactics: a review of the ideas. Q. evolution, model systems, and the importance of phylogeny. In Rev. Biol. 51, 3^47. New uses for new phylogenies (ed. P. H. Harvey, A. J. Leigh Stebbins, G. L. 1974 Flowering plants. Evolution above the species level. Brown, J. Maynard Smith & S. Nee), pp. 322^340. New York: Cambridge: Belknap Press. Oxford University Press. Swo¡ord, D. L. 1993 PAUP*: Phylogenetic Analysis Using Parsimony, Morgan, M. T. & Schoen, D. J. 1997 The role of theory in an version 3.1. Formerly distributed by Illinois Natural History emerging new plant reproductive biology.Trends Ecol. Evol. 12, Survey, Champaign, IL. 231^234. Uyenoyama, M. K. 1986 Inbreeding and the cost of meiosis: the Olivieri, I., Couvet, D. & Slatkin, M. 1994 Allocation of repro- evolution of sel¢ng in populations practising biparental ductive e¡ort in perennial plants under pollen limitation. Am. inbreeding. Evolution 40, 388^404. Nat. 144, 373^394. Weller, S. G., Wagner, W. L. & Sakai, A. K. 1995 A phylogenetic Primack, R. B. 1979 Reproductive e¡ort in annual and perennial analysis of Schiedea and Alsinidendron (Caryophyllaceae: species of Plantago (Plantaginaceae). Am. Nat. 114, 51^62. Alsinoideae): implications for the evolution of breeding Quiros, C. F. & Bauchan, G. R. 1988 The genus Medicago and the systems. Syst. Bot. 20, 315^337. origin of the Medicago sativa complex. In Alfalfa and alfalfa Zhang, D.-Y. & Wang, G. 1994 Evolutionarily stable reproductive improvement (Agronomy Monograph no. 29) (ed. A. A. strategies in sexual organisms: an integrated approach to life- Hanson, D. K. Barnes & R. R. Hill), pp. 93^124. Madison, history evolution and sex allocation. Am. Nat. 144, 65^75. WI: American Society of Agronomy. Richards, A. J. 1997 Plant breeding systems, 2nd edn. London: As this paper exceeds the maximum length normally permitted, Chapman & Hall. the authors have agreed to contribute to production costs.

Proc. R. Soc. Lond. B (1998)