C4 photosynthesis is a series of biochemical and structural modifications to C3 photosynthesis that has evolved numerous times in flowering plants, despite requiring modification of up to hundreds of genes. To study the origin of C4 photosynthesis, we reconstructed and dated the phylogeny of Molluginaceae, and identified C4 taxa in the family. Two C4 species, and three clades with traits intermediate between C3 and C4 plants were observed in Molluginaceae. C3–C4 intermediacy evolved at least twice, and in at least one lineage was maintained for several million years. Analyses of the genes for phosphoenolpyruvate carboxylase, a key C4 enzyme, indicate two independent origins of fully developed C4 photosynthesis in the past 10 million years, both within what was previously classified as a single species, Mollugo cerviana. The propensity of Molluginaceae to evolve C3–C4 and C4 photosynthesis is likely due to several traits that acted as developmental enablers. Enlarged bundle sheath cells predisposed some lineages for the evolution of C3–C4 intermediacy and the C4 biochemistry emerged via co-option of photorespiratory recycling in C3–C4 intermediates. These evolutionarily stable transitional stages likely increased the evolvability of C4 photosynthesis under selection environments brought on by climate and atmospheric change in recent geological time.

One aim of evolutionary biology is to understand how complex traits emerge during the diversification of organisms. The development of complex traits involves modification of multiple genes, a process thought to occur gradually. Successive steps in an evolutionary transition are difficult to reconstruct, but extant taxa with intermediate characters can help (Combes 2001; Lamb et al. 2007; Herron and Michod 2008; Ogawa et al. 2009). However, the evolutionary significance of these intermediate taxa must be evaluated in an appropriate phylogenetic framework to inform us about the evolution of a particular trait (Adoutte et al. 1999; Herron and Michod 2008).

C4 photosynthesis is one of the best systems in which to study complex trait evolution. It has evolved at least 50 times in a wide range of flowering plants (Muhaidat et al. 2007), making it one of the most convergent of evolutionary phenomena. The function of C4 photosynthesis is to enhance the efficiency of Rubisco, the primary CO2 fixing enzyme in C3 photosynthesis (Fig. 1). At current atmospheric conditions, Rubisco is significantly inhibited in warm climates by its ability to fix O2 instead of CO2, an inhibitory process termed photorespiration. C4 plants overcome photorespiration by metabolically concentrating CO2 into an inner cellular compartment where Rubisco is localized (Fig. 1). The C4 concentrating mechanism arises from both morphological and biochemical innovations that function in unison to first fix CO2 into organic compounds in the mesophyll tissue, and then to transport these compounds and release CO2 into the chloroplasts of the cells that surround the vascular tissue (Fig. 1). These bundle sheath cells (BSC) are mainly involved in exchanges between veins and mesophyll in C3 plants (Leegood 2008), but are responsible for CO2 assimilation by Rubisco in C4 plants (Fig. 1). Compared to C3 plants, the typical C4 foliar anatomy is characterized by large BSC surrounded by a low number of mesophyll cells, a reduction of the interveinal distance and an aggregation of chloroplasts in BSC (Fig. 1). These modifications allow a rapid exchange of metabolites between mesophyll cells and BSC and an efficient concentration of CO2 from mesophyll to BSC. The kinetics and regulation of the enzymes used in the C3 and C4 cycles are also modified from ancestral forms, leading to a close coordination of the enzymes of each cycle (Leegood and Walker 1999; Engelmann et al. 2003; Hibberd and Covshoff 2010; Chastain 2011). Overall, dozens if not hundreds of genes have been modified during the evolution of C4 plants from C3 ancestors (Monson 1999; Sawers et al. 2007; Hibberd and Covshoff 2010).

Figure 1.

Anatomical and biochemical differences between C3 and C4 photosynthesis. These simplified diagrams present (A) the carbon assimilation in a C3 leaf and (B) details of the biochemical steps in a mesophyll cell of a C3 plant. By comparison, (C) carbon assimilation in a C4 leaf involves important anatomical modifications and the addition of a biochemical pathway (black arrows), with (D) a close coordination of the biochemical steps between mesophyll and bundle sheath cells. Numbers in brackets represent the principal biochemical steps; 1 = fixation of CO2 by Rubisco, which starts the Calvin cycle, 2 = fixation of CO2 into an organic compound by a coupling of carbonic anhydrase and phosphoenolpyruvate carboxylase, 3 = transformation and transport of the organic compounds, 4 = release of CO2 by one of the decarbxylating enzymes, 5 = regeneration of the carbon acceptors.

How the C4 pathway was repeatedly assembled in so many groups of flowering plants remains an open question. Hypotheses have focused on the successive acquisition of increasingly C4-like characters in harsh environments induced by global climate change and reductions in atmospheric CO2 content over the past 35 million years (Ehleringer et al. 1997; Sage 2004). The development of these hypotheses has been assisted by the study of species that exhibit characteristics intermediate between C3 and C4 photosynthesis (Hattersley et al. 1986; Rajendru et al. 1986; Griffiths 1989; Monson and Moore 1989; Sage et al. 1999; McKown et al. 2005; Vogan et al. 2007). These C3–C4 intermediates are known from multiple independent plant lineages (Sage et al. 1999). In these plants, the degree of cellular and enzymatic rearrangements varies from being close to C3 species to being similar to fully developed C4 plants. The Asteraceae genus Flaveria stands out as having more C3–C4 species than any other genus, with approximately 12 intermediate species (Ku et al. 1991, 1996; McKown et al. 2005). Flaveria has thus become the principle model for inferring past transitional stages in the origin of both C4 anatomy (McKown and Dengler 2007) and C4 biochemistry (e.g., Nakamoto et al. 1983; Engelman et al. 2003; Svensson et al. 2003). Phylogenetic evaluation of the relationships between Flaveria species confirmed that C3 photosynthesis is the ancestral condition to C4 photosynthesis, and that C4 characters were successively acquired until C4 species emerged (McKown et al. 2005). However, Flaveria is but one of the numerous plant lineages where C4 photosynthesis independently evolved. Multiple C3 to C4 transitions should be evaluated to assess the generality of patterns, and to identify characters that might predispose certain C3 taxa to evolve the C4 pathway (Marshall et al. 2007; Vogan et al. 2007). Most C3–C4 intermediate species are relatively restricted in terms of geographic distribution and floristic importance (Sage et al. 1999). However, two species, Mollugo verticillata (carpet weed) and Mollugo nudicaulis (John's Folly) are successful cosmopolitan weeds of disturbed areas in warm climates (Vincent 2003). Mollugo verticillata was the first discovered C3–C4 species (Kennedy and Laetsch 1974), and for many years, it has been assumed that this species was a close relative of the only C4 species known in the Molluginaceae, Mollugo cerviana. However, there has been no systematic survey of photosynthetic pathways in this family and a lack of sufficient phylogenetic information has prevented any analysis of the relationships between its C3, C3–C4 and C4 species. The few phylogenetic studies incorporating representatives of Molluginaceae indicate the group as traditionally circumscribed is likely polyphyletic (Cuénoud et al. 2002; Brockington et al. 2009).

In this study, we identify the photosynthetic pathway of over 100 species in the Molluginaceae sensu lato, and then address C4 evolution in the family by examining leaf anatomy and reconstructing the evolutionary relationships between C3 and C4 species and taxa that exhibit intermediate traits between the two pathways. We adopted a dense, family-wide sampling, including multiple accessions from diverse geographic origins for several species, notably the C4 and C3–C4 taxa. Phylogenetic hypotheses using plastid markers were integrated in a broader analysis of eudicots and time calibrated with information from numerous fossils. We also sequenced nuclear genes encoding phosphoenolpyruvate carboxylase (PEPC), a key enzyme of the C4 pathway, to gain insights into the evolutionary optimization of C4 biochemistry in the Molluginaceae. The combination of photosynthetic, anatomical, and molecular datasets enabled us to isolate some of the steps in C4 evolution, and provides fertile new ground for developing hypotheses about anatomical and ecological conditions that promote the evolution of this complex trait.

Material and Methods


We sampled extensively from as many species of the Molluginaceae sensu lato as possible, following the classification of Endress and Bittrich (1993). Dried samples from herbarium specimens were obtained from numerous botanical gardens and herbaria (Table S1). When available, multiple accessions per species were analyzed.

For each sample, approximately 2 mg of plant tissue (stem, roots or leaves) was assayed for carbon isotope ratio using an Integra mass spectrometer with a Pee Dee belamnite standard. Carbon isotope ratios were determined by the University of California stable isotope facility (

Most herbarium specimens were too degraded for DNA extraction and only a subset of the samples used for carbon isotope ratios were included in the phylogenetic analyses (Table S2).


Genomic DNA (gDNA) was extracted using the DNeasy Plant Mini Kit (Qiagen, GmbH, Germany) following the provider recommendations. Two plastid markers were selected for phylogenetic reconstruction, the coding gene rbcL and the region encompassing trnK introns and matK coding sequence. For accessions that yielded good quality DNA, each of these markers was amplified in a single polymerase chain reaction (PCR), using primers designed in this study based on sequences available in GenBank. Primers for rbcL were rbcL_4_ForTCACCACAAACAGARACTAAAGC and rbcL_1353_RevGCAGCNGCTAGTTCAGGACTC. They amplify a 1326-bp fragment, which represents 93% of the whole coding sequence. For trnK-matK, the primers were trnKmatK_ForAGTTTRTMAGACCACGACTG and trnKmatK_RevGCACACGGCTTTCCCTATG. They amplify a 2260–2420 segment, depending upon the species that comprises the whole coding sequence of matK and intron regions of trnK. PCRs were carried out in a total volume of 50 μl, including about 100 ng of gDNA template, 10 μl of 5× GoTaq Reaction Buffer, 0.15 mM dNTPs, 0.2 μM of each primer, 2 mM of MgCl2, and 1 unit of Taq polymerase (GoTaq DNA Polymerase, Promega, Madison, WI). For rbcL, the PCR mixtures were incubated in a thermocycler for 3 min at 94°C followed by 37 cycles consisting of 1 min at 94°C, 30 sec at 48°C and 90 sec at 72°C. This was followed by 10 min at 72°C. The PCR conditions were similar for trnK-matK except that annealing temperature was set to 51°C and the extension time to 2 min 30. Successful amplifications were purified using the QIAquick PCR Purification Kit (Qiagen) and sequenced with the Big Dye 3.1 Terminator Cycle Sequencing Kit (Applied Biosystems, Foster City, CA), following the provider instructions, and separated on an ABI Prism 3100 genetic analyzer (Applied Biosystems). For rbcL, two internal primers were used; rbcL_629_ForCRTTTATGCGTTGGAGAGACC and rbcL_760_revCAAYTCTCTRGCAAATACAGC. Sequencing of trnK-matK was first performed with the trnKmatK_For primer and internal primers were then designed based on the partial sequences obtained (Fig. S1).

For most samples, the gDNA obtained were too degraded to amplify long fragments of DNA. The gDNA of species that failed the amplification of full rbcL or trnK-matK were used as a template to amplify short overlapping fragments, with a battery of internal primers designed for this study (Fig. S1). The size of targeted fragments was reduced, until PCR succeeded, to 200 bp for some gDNA. PCR reactions were run as described above for rbcL except that the extension time was lowered to 45 sec. Purifications and sequencing were performed as described above, but one of the PCR primers was used for sequencing reaction.


The GenBank database was screened and Caryophyllales species for which both rbcL and matK or trnK-matK were available were added to the dataset. In addition, species from other eudicot lineages, one taxon from the eudicot sister group (Ceratophyllales) and one monocot (Acorus americanus; used as outgroup) were added from GenBank to allow more calibration points to be used in the molecular dating analyses (see below; Table S3). Sequences were aligned using ClustalW (Thompson et al. 1994) and the alignment was manually refined. Phylogenetic trees were then inferred using a Bayesian procedure implemented in MrBayes 3.1 (Ronquist and Huelsenbeck 2003). The noncoding introns of trnK were very difficult to align between species from different families. Therefore, only rbcL and matK were considered for species outside Molluginaceae sensu stricto (hereafter referred to as Molluginaceae), these coding genes being unambiguously aligned. These two markers were analyzed separately and in combination, to check for congruence. The noncoding trnK introns were included only for Molluginaceae, after the alignment was manually refined. These species being more closely related, the alignment of trnK was not problematic.

The substitution model was set to a general time reversible model with a gamma shape distribution and a proportion of invariant sites (GTR + G + I), as identified as the best-fit model by hierarchical likelihood ratio tests (hLRT). Two Bayesian analyses, each of four parallel chains, were run for 10,000,000 generations. A tree was sampled each 1,000 generations after a burn-in period of 3,000,000 generations. A consensus tree was computed from the 14,000 sampled trees.

The obtained phylogeny was used for molecular dating using a Bayesian method that accounts for changes in rates of evolution among branches, following the recommendations of Rutschmann (2006). The trnK marker was not used in the dating analyses. Model parameters were estimated with baseml (Yang 2007) for the two genes separately. Branch lengths and the variance–covariance matrix were then optimized using estbranches (Thorne et al. 1998). A Bayesian Markov chain Monte Carlo (MCMC) procedure implemented in multidivtime (Kishino et al. 2001; Thorne and Kishino 2002) approximated the posterior distributions of substitution rates and divergence times, given a set of time constraints. The MCMC procedure was run for 1,000,000 generations after a burn-in of 100,000 generations, with a sampling frequency of 100 generations. The outgroup (A. americanus) was removed during the analysis. The maximal age for the root of the tree was set to 160 million years ago (Mya), a time that generally exceeds estimates of monocot–eudicot divergence (Magallon and Sanderson 2001; Friis et al. 2006), and nine different constraints were set on internal nodes. The first evidence of eudicots in the fossil record comes from tricolpate pollen, which appeared during the late Barremian and early Aptian (Magallon and Sanderson 2001; Friis et al. 2006). This was used to set a lower bound of 120 Mya to the stem group node and an upper bound of 130 Mya to the crown group node of eudicots. This upper bound assumes that the time span between the emergence of eudicots and their fingerprint in the fossil records does not exceed a few million years. Lower bounds were set to the stem group nodes of several eudicot orders following Magallon and Sanderson (2001): 102.2 Mya for Buxales, 91.2 for Malpighiales, 59.9 for Fabales, 69.7 for Malvales, 88.2 for Myrtales and 91.2 for Ericales. In addition, a lower bound of 34 Mya was set to the divergence of Polycarpon and the higher Caryophyllaceae (here represented by Silene, Schiedea and Scleranthus), according to the phylogenetic position of a fossil reported by Jordan and Mcphail (2003). The same analysis was rerun excluding successively each calibration point to check for major incongruence among constraints, and also rerun without the upper bound on the crown of eudicots and using a maximum age of the root of 200 Mya, to take into account recent analyses suggesting that Angiosperms may be older than previously thought (Smith et al. 2010).


Genes encoding phosphoenolpyruvate carboxylase (ppc) of eudicot species were retrieved from GenBank. These were aligned and used to design a pair of primers theoretically able to amplify all Caryophyllales ppc; ppc-1294-ForGCNGATGGAAGYCTTCTTG and ppc-2890-RevGCTGGNATGCAGAACACYG. The forward primer is located in exon 8, whereas the reverse primer extends to the stop codon in exon 10. The amplified region is homologous to the gene portion previously studied in grasses (Christin et al. 2007) and sedges (Besnard et al. 2009) and which contains major determinants of the C4 function (Bläsing et al. 2000; Jacobs et al. 2008). The studied fragment includes more than 1500 bp of coding sequence, which represents more than half of the full coding sequence.

The designed primers were used to PCR-amplify ppc genes from a subsample of the gDNA used for plastid markers and that were of good quality. About 100 ng of gDNA were mixed with 5 μl of 10× AccuPrime PCR Buffer, 0.2 mM of each dNTP, 0.2 μM of each primer, 3 mM of MgSO4, 2.5 μl of DMSO and 1 unit of a proof-reading Taq polymerase (AccuPrime Taq DNA Polymerase High Fidelity, Invitrogen, Carlsbad, CA) in a total volume of 50 μl. The PCR mixtures were incubated at 94°C for 2 min, followed by 35 cycles consisting of 30 sec at 94°C, 30 sec at 51°C and 3 min at 68°C. The last cycle was followed by 20 min at 68°C. PCR products were run on a 0.8% agarose gel and purified with the QIAquick Gel Extraction Kit (Qiagen). Purified products were cloned into the pTZ57R/T vector using the InsT/Aclone PCR Product Cloning Kit (Fermentas, Vilnius, Lithuania). Up to 20 clones for each PCR product were amplified with ppc-1294-For and ppc-2890-Rev primers. The PCR products were restricted with the TaqI restriction enzyme (Invitrogen) and insert of each clone with a distinct restriction pattern was purified and sequenced as described for plastid markers. Sequencing reactions were performed first with the ppc-1294-For primer and then with internal primers.

Exons were identified through homology with the available ppc sequences and following the GT–AG rule. Coding sequences were translated into amino acids and aligned using ClustalW (Thompson et al. 1994). Once translated back into nucleotide, the alignment was manually refined and used to infer a phylogenetic tree using MrBayes 3.1 (Ronquist and Huelsenbeck 2003). Eudicots and a sample of monocots ppc genes available in GenBank were added to the dataset (Table S4). The best-fit substitution model was determined through hLRT as the GTR + G + I. Bayesian analysis was run as described for plastid markers, but model parameters were estimated independently for first, second, and third positions of codons. Amino acids changes similar to those shown to be under positive selection for the C4 function in Poales (Christin et al. 2007; Besnard et al. 2009) were reported on the phylogenetic tree.


Leaf sections from living specimens of M. pentaphylla, M. verticillata, M. nudicaulis, and M. cerviana were fixed in glutaraldehyde, postfixed in osmium, and embedded in Spurr's resin (Spurr 1969) as described by Sage and Williams (1995). In addition, because access to living species is often difficult, and the use of herbarium material to produce microimages of leaves would assist phylogenetic-based studies of photosynthetic pathway evolution, 18 accessions from Molluginaceae were sampled to assess variation in leaf anatomy (Table S2). Approximately 5 mm2 of herbarium leaf samples were rehydrated in ddH2O over night and subsequently fixed in 2% glutaraldehyde buffered with 0.05 M sodium cacodylate buffer (pH 6.9) for 24 h. Fixed samples were dehydrated in 10% ethanol increments and also embedded in Spurr's resin. All embedded leaf samples were sectioned at 1.5 μm, stained with toluidine blue-O in 0.2% benzoiate buffer (pH 4.4; O’Brien and McCully 1981), and imaged with a Zeiss Axioplan microscope (Carl Zeiss, Göttingen, Germany) equipped with an Olympus DP71 digital camera and imaging system (Olympus Canada, Markham Ontario).



A total of 314 accessions classified into 116 species from the Molluginaceae and affiliated taxa was typed for carbon isotope ratios. Values between −21‰ and −32‰ are indicative of C3 species; values between −9‰ and −16‰ indicate C4 species, whereas −16‰ to −19‰ indicate C4-like species (von Caemmerer 1992). C3–C4 intermediate species normally have a C3 isotopic ratio unless there is significant engagement of PEPC and a C4 metabolic cycle as occurs in C4-like species (von Caemmerer 1992). Only accessions from two Molluginaceae species, M. cerviana and Mollugo fragilis, exhibited C4 carbon isotope ratios typical of C4 plants (Table 1; Table S1). Although M. cerviana is known to be C4 (Kennedy and Laetsch 1974), M. fragilis represents a newly discovered C4 taxon. All other taxa had C3 isotope ratios, including two taxa (M. verticillata and M. nudicaulis) previously demonstrated to be C3–C4 intermediates (Sayre and Kennedy 1979; Kennedy et al. 1980; Fig. S2).

Table 1.  Carbon isotope ratios of species sampled for this study. C4 and C3–C4 species are highlighted in bold. Numbers in parenthesis are sample size if greater than 1. See Table S1 for herbarium vouchers and raw isotope values.
  δ13C (sample size)Geographic region
Adenogramma Confined to South Africa
 A. capillaris−30.7South Africa
 A. glomerata−28.2 (4)South Africa
 A. lichtensteiniana−26.8 (2)South Africa
 A. mullugo−28.1 (3)South Africa
 A. rigida−27.3South Africa
 A. sylvatica−28.9 (3)South Africa
Coelanthum South Africa, SW Africa
 C. grandiflorum−25.8 (2)South Africa, Namibia
 C. parviflorum−27.0South Africa
 C. semiquinquefidum−24.7 (2)South Africa Namibia
Corbichonia southern Africa and Saudi Arabia
 C. decumbens−27.2 (3)arid & semiarid Africa to SW Asia
 C. rubriviolacea−25.6 (2)SW Africa
 C. andina−30.4(sub)tropical South and central America
 C. capensis−28.2 (3)tropical east Africa
 C. drymaroides−26.6 (2)Malawi. Mozambique, Zimbabwe
 C. litoralis−28.1 (3)temperate zones to subtropics, old world
 C. paniculata−27.0Tanzania, Zambia
 C. squamosa−28.2(sub)tropical South America
Glinus pantropical to temperate zones
 G. bainesii−26.4 (3)Zambia, Zimbabwe
 G. herniarioides−25.3SE Asia
 G. lotoides−27.1 (10)pantropical to temperate weed
 G. oppositifolius−28.4 (14)sub-Saharan Africa to Australia
 G. orygioides−23.3NE Australia
 G. radiatus−28.8 (7)Americas, introduced to Africa
 G. setiflorus−26.2 (3)Arabia, Kenya, Somalia, Tanzania
Glischrothamnus ulei−28.0Brazil
Hypertelis southern Africa, Madagascar, St. Helena
 H. angrae-pequenae−25.5 (2)South Africa, Namibia
 H. arenicola−26.6South Africa
 H. bowkeriana−26.0 (5)southern Africa, Kenya
 H. ceaspitosa−24.0Namibia, South Africa
 H. salsoloides−25.1 (3)southern Africa
 H. spergulacea−26.7 (2)Namibia, South Africa
 H. trachysperma−25.6South Africa
Limeum southern Africa to south Asia
 L. aethiopicum−25.6 (9)South Africa, Zimbabwe
 L. africanum−26.7 (2)southern Africa
 L. angustifolium−27.7Somalia
 L. arabicum−25.5 (2)Arabian peninsula
 L. arenicolum−24.3 (2)southern Africa
 L. argute-carinatum−25.0southern Africa, Angola
 L. deserticolum−26.8southern Africa
 L. dinteri−26.6 (3)southern Africa
 L. fenestratum−27.2 (4)southern subtropical Africa
 L. glaberrimum−26.2Mozambique
 L. humifusum−27.0South Africa
 L. linifolium−24.9South Africa
 L. myosotis−25.5 (2)southern Africa, Angola
 L. obovatum−26.6 (2)Sahara region to India
 L. praetermissum−28.3 (2)Kenya, Somalia
 L. pterocarpum−28.1 (5)subtropical Africa
 L. rhombifolium−25.7 (2)Namibia, South Africa
 L. sulcatum−25.5 (4)southern Africa
 L. telephioides−25.2South Africa
 L. viscosum−26.4 (4)subtropical, east-tropical Africa
Macarthuria Australia
 M. apetala−26.5 (2)Australia
 M. australis−27.0 (2)Australia
 M. complanata−29.4NE Australia
 M. ephedroides−28.5NE Australia
 M. neocambrica−27.6 (2)Australia
Mollugo pantropical to temperate zones
 M. angustifolia−29.3Somalia
 M. cerviana13.3 (19)hot arid regions, pantropics to temperate
 M. crockeri−26.5Galapagos
 M. decandra−29.4 (3)Madagascar
 M. enneandra−28.2 (2)Cuba
 M. flavescens−25.5 (6)Galapagos
 M. f. subsp. gracillima−27.0 (3)Galapagos
 M. f. subsp. insularis−27.2Galapagos
 M. floriana−25.5 (2)Galapagos
 M. fragilis13.2 (3)Angola (coastal sand)
 M. molluginis−29.0 (6)Australia
 M. namaquensis−27.4 (3)South Africa
 M. nudicaulis (C3–C4)28.4 (10)pantropical and subtropical
 M. nudicaulis var. navassensis−29.7 (2)Caribbean
 M. pentaphylla−28.8 (13)pantropics and subtropics
 M. pusilla−28.5Namibia, South Africa
 M. racemosa−27.4 (4)India, Sri Lanka
 M. snodgrassii−25.3 (4)Galapagos
 M. tenella−28.0 (5)Namibia, South Africa
 M. verticillata (C3–C4)27.6 (12)cosmopolitan weed
Pharnaceum southern Africa
 P. albens−26.9 (2)Namibia, South Africa
 P. brevicaule−26.9southern Africa
 P. confertum−27.4South Africa
 P. cordifolium−27.8South Africa
 P. croceum−25.2 (2)Namibia, South Africa
 P. detonsum−26.5 (2)South Africa
 P. dichotomum−27.3 (2)South Africa
 P. elongatum−29.1 (2)South Africa
 P. exiguum−27.9South Africa
 P. fluviale−26.7South Africa
 P. incanum−28.1 (3)South Africa
 P. lanatum−28.2 (3)South Africa
 P. lineare−25.5 (2)South Africa
 P. microphyllum−28.5South Africa
 P. reflexum−27.8South Africa
 P. subtile−27.2South Africa
 P. thunbergii−32.0 (2)South Africa
 P. trigonum−29.4South Africa
 P. verrucosum−24.0 (4)southern Africa
 P. viride−26.5South Africa
 P. zeyheri−28.9South Africa
Polpoda South Africa
 P. capensis−24.9 (3)South Africa
 P. stipulacea−25.2South Africa
Psammotropha southern to tropical Africa
 P. alternifolia−26.6South Africa
 P. anguina−30.1South Africa
 P. frigida−26.1South Africa
 P. mucronata−27.4 (4)Mozambique, South Africa
 P. myriantha−28.2 (2)southern Africa
 P. obovata−24.3 (2)South Africa
 P. obtusa−26.6South Africa
 P. quadrangularis−28.1 (2)South Africa
 Suessenguthiella scleranthoides−24.9 (2)Namibia, South Africa
Telephium Mediterranean basin and Arabian peninsula
 T. barbeyanum−28.2Libya
 T. imperati−26.6 (2)North Africa, Spain
 T. orientale−26.4Turkey
 T. oligopsermum−27.2 (2)Iraq
 T. sphaerospermum−27.6 (2)Yemen
 T. portulacastrum14.0 
 Z. pentandra12.4 
 Mollugo minuta−24.0 (2)South Africa
 Pharnaceum cupitatum−25.7South Africa
 Pharnaceum spergulacea−24.9South Africa
 Psammotropha eriantha−29.2South Africa
 Psammotropha rosularis−28.0southern Africa


Plastid markers were obtained for a total of 94 accessions, including 73 Molluginaceae and 46 Mollugo. Both rbcL and trnK-matK markers were completed for all but 15 accessions (Table S2). With 80 additional accessions retrieved from GenBank, the phylogenetic dataset contained 174 accessions labeled as 144 different species (Table 1). The phylogenies inferred separately from rbcL and matK were fully congruent with each other (data not shown). These coding markers were thus combined (Fig. 2). The family Molluginaceae as originally circumscribed (Endress and Bittrich, 1993) is polyphyletic, confirming results found with a limited species sampling (Cuénoud et al. 2002; Brockington et al. 2009). Limeum is sister to a large clade containing Molluginaceae and other families, justifying its treatment as a separate family (Limeaceae; APG III 2009). The Australian genus Macarthuria appears as the sister group of all other core Caryophyllales. Corbichonia and most Hypertelis are sister to a clade comprising Aizoaceae and Nyctaginaceae. Notably, one species of Hypertelis (H. spergulacea) falls within Molluginaceae (Fig. 3). The close relationship between H. spergulacea and M. cerviana inferred from plastid markers was also highly supported by two different nuclear genes (ppc-1 and ppc-2; Fig. S3). The synonym Mollugo linearis auct. Non Ser. Em Dc. (Tropicos 2010) should possibly be resurrected for H. spergulacea. Molluginaceae are highly supported as the sister group of the Portulacineae clade (Nyffeler et al. 2008; Nyffeler and Eggli, 2010) and contain the genera Mollugo, Adenogramma, Coelanthum, Glinus, Glischtrothamnus, Pharnaceum, Polpoda, Psammotropha, and Suessenguthiella, and H. spergulacea (Figs. 2 and 3).

Figure 2.

Dated phylogeny and position of Molluginaceae within eudicots. The topology was obtained through Bayesian inference of coding sequences from the plastid markers rbcL and matK, and trnK for Molluginaceae. Bayesian support values are indicated near branches when greater than 90. Branch lengths are proportional to estimated divergence times, in million years ago (Mya). This tree was rooted on monocots, which were removed during the dating analysis. Families of Caryophyllales and other orders are compressed, except for taxa with problematic taxonomic position, for which genus or species names are indicated. The part of the tree corresponding to Molluginaceae is detailed in Figure 3. Major groups are indicated on the right.

Figure 3.

Dated plastid phylogeny of Molluginaceae and photosynthetic types. This figure details the Molluginaceae part of Figure 2. Bayesian support values are indicated near branches when greater than 90. Branch lengths are proportional to estimated divergence times, in million years ago (Mya). Each accession name is followed by its geographical origin, in square brackets. Numbering is used to differentiate accessions from the same species and geographical origin. It corresponds to that in Table S2. Groups of accessions described in the text are delimited on the right. Squares represent indicators of photosynthetic types; (A) size of bundle sheath cells (BSC; white = typical C3, black = enlarged), (B) localization of chloroplasts (white = typical of C3 plants, black = aggregation in BSC) and (C) carbon isotopes ratio (white = non-C4, black = C4). Branches where PEPC was optimized for the C4 function, as deduced from analyses of genes encoding PEPC (Fig. 5), are indicated by vertical bars.

Within Molluginaceae, the phylogeny inferred from rbcL and trnK-matK is very well resolved, with most branches having a posterior probability of 0.99 or 1.0 (Fig. 3), and fully congruent with relationships deduced from the nuclear ppc genes (Fig. S3). Mollugo is not monophyletic, being largely mixed with the other Molluginaceae genera. The clade with the two C4 species (M. cerviana and M. fragilis) and H. spergulacea is sister to a clade composed of diverse genera (Adenogramma, Coelanthum, Mollugo, Pharnaceum, Polypoda, Psammotropha, Suessenguthiella) originating mainly from southern Africa (Fig. 3). All members of this South African clade have C3 isotopic ratios. The rate of molecular evolution of plastid markers was strongly accelerated in this region of the tree when compared to other eudicots, a pattern observed with both rbcL and matK (data not shown), but not with nuclear markers (Fig. S3). This phenomenon, which also occurred in the mitochondrial genome of some Geraniaceae and Plantaginaceae (Cho et al. 2004; Parkinson et al. 2005), suggests a higher mutation rate, potentially due to the decrease of the quality of DNA replication and repair in the chloroplasts of some Molluginaceae. However, the grouping of these species cannot be attributed to long-branch attraction because they also share numerous unique insertions that represent synapomorphies (in particular, a 33 bp insertion in the coding sequence of matK).

The sampling of multiple accessions per species also revealed several Mollugo species that are not monophyletic—M. cerviana, M. nudicaulis, and M. verticillata. Mollugo cerviana is paraphyletic with respect to M. fragilis and H. spergulacea (Fig. 3). Two clades of M. cerviana include accessions from overlapping geographical ranges and are highly divergent in both plastid and nuclear markers (Fig. 3 and Fig. S3), suggesting they correspond to distinct species. The two M. cerviana clades will hereafter be referred to as the cerviana group (which includes old world and Australian M. cerviana) and the fragilis group (which includes old and new world M. cerviana as well as M. fragilis). This complex (M. fragilis, M. cerviana and H. spergulacea) requires renewed taxonomic attention. Biogeographic and population genetics approaches should be adopted to determine the number of true biological species and subsequent systematic studies should aim at identifying synapomorphies of these biological species and proposing new binomial names. Mollugo nudicaulis is similarly paraphyletic. In addition to a clade composed of accessions collected worldwide, two accessions from the Caribbean (Navassa Island and British Virgin Islands) formed a distinct clade separated from the other M. nudicaulis by several species (Fig. 3). Ekman determined the M. nudicaulis he collected on Navassa Island as M. nudicaulis var. navassensis (inscription on the herbarium sheet, Erik L. Ekman—10810, MO), a variety that might be raised to the species level in light of our results.

Finally, M. verticillata is also not monophyletic, with a clade sister to a Cuban species (M. enneandra) and a Bolivian accession sister to several species of the Galapagos islands, which present almost no variation in plastid markers (M. crockerii, M. flavescens, M. flavescens subsp. insularis, M. floriana, and M. snodgrassii). This complex, which will be referred to as the verticillata group, also requires fine-scale biogeographic and population genetics studies to determine species boundaries. Our initial results suggest that the cosmopolitan weed M. verticillata has produced divergent morphotypes; one in Cuba (M. enneandra), and several others after colonizing the Galapagos Islands. Wallace already hypothesized this evolutionary scenario for Galapagos Mollugo based on seed characters (Wallace 1987). The generation of significant morphological disparity without much apparent genetic divergence mirrors the diversification patterns of Opuntia in the Galapagos archipelago (Helsen et al. 2009).

The incongruence between molecular phylogenetics and taxonomy in the Molluginaceae questions the validity of using currently recognized species as evolutionarily relevant units in this group. We will consequently consider accessions as separate entities and use species names only as labels to describe groups of accessions.


The split of Molluginaceae from their sister-group (Portulacineae) was estimated at 51.9 (±4.7) Mya and the first divergence within Molluginaceae at 46.7 (±4.8) Mya. The stem and crown nodes of the verticillata group were estimated at 13.4 (±4.2) Mya and 8.6 (±3.5) Mya, respectively. For the Galapagos endemics, stem and crown nodes were optimized at 4.3 (±2.1) and 3.0 (±1.6) Mya, respectively, which is congruent with the recent emergence of the Galapagos archipelago. The common ancestor of the cerviana and fragilis groups and H. spergulacea is estimated to have diverged from the Pharnaceum/Adenogramma clade 20.9 (±3.7) Mya. For the cerviana group, the stem and crown group ages were estimated at 9.6 (±2.7) and 2.0 (±1.1) Mya, respectively and for fragilis, at 7.2 (±2.2) and 1.9 (±0.9) Mya. The estimated 9.6 My of divergence between cerviana and fragilis groups supports the hypothesis that these represent different species. Similarly, M. nudicaulis var. navassensis and the other M. nudicaulis diverged 19.8 (±4.7) Mya, and likely do not belong to the same biological species.

The dating analyses produced very similar results when constraints were successively removed (Fig. S4), indicating no major conflict between the different calibration points. Removing the upper bound of 130 Mya on the crown of the eudicots and changing the maximal age for the root of the tree from 160 to 200 Mya strongly affected the estimate of the age of the root, as expected. However, estimates for Molluginaceae remained largely unchanged. In this analysis with relaxed upper limits, stem and crown of Molluginaceae were brought back to 56.1 (±5.8) and 50.3 (±5.8) Mya, respectively and the stem group nodes of cerviana and fragilis groups to 10.2 (±2.9) and 7.6 (±2.4) Mya, respectively. Thus, our conclusions remain largely unaffected by any uncertainty regarding the age of angiosperms.


Although most of the dried leaf samples show some damage when fixed, it was still feasible to make out critical features pertaining to C4 trait evolution, in particular, enlarged BSC, increased organelle density in BSC, aggregation of organelles along the inner BSC wall, and increased vein density such that the distance between mesophyll cells and BSC is reduced. Together, these features indicate C3–C4 intermediacy in species with a C3 isotope ratio. Mollugo pentaphylla had a clear C3 anatomy (Fig. 4A), whereas M. verticillata (C3–C4) exhibited a dense aggregation of plastids in slightly enlarged BSC (Fig. 4B), confirming its prior determination as a C3–C4 species (Kennedy and Laetsch 1974). Mollugo nudicaulis had large numbers of chloroplasts in BSC relative to typical C3 plants (Fig. 4C), and has been previously physiologically shown to be a weak C3–C4 intermediate (Kennedy et al. 1980). Concentration of chloroplasts in BSC seem to be also present in its closest relatives (M. decandra and M. nudicaulis var. navassensis), although whether this is linked to C3–C4 physiology in these taxa remains to be assessed. In the clade of Adenograma/Pharnaceum that is immediately sister to the M. cerviana complex, BSC of three accessions that branch from the basal portion of the clade (Suessenguthiella scleranthoides, Adenogramma sylvatica and M. tenella) were enlarged compared to a typical C3 pattern, but lacked any obvious enhancement of chloroplasts (Fig. S5). A species with narrow, linear leaves, Pharnaceum detonsum, did not have BSC that were enlarged from the C3 condition (Fig. S5).

Figure 4.

Light micrographs illustrating leaf anatomy of select Molluginaceae. (A) Mollugo pentaphylla, (B) M. verticillata, (C) M. nudicaulis, (D) Hypertelis spergulacea, (E) M. cerviana, and (F) M. fragilis. Mollugo cerviana and M. fragilis are C4 species. Arrows highlight chloroplasts in the bundle sheath cells. B = bundle sheath; M = mesophyll. Bars = 50 μm.

The leaf anatomy of H. spergulacea was similar to C3–C4 intermediates (Fig. 4D; McKown and Dengler 2007). Bundle sheaths were enlarged compared to M. pentaphylla, and showed a distinct aggregation of multiple organelles on the inner bundle sheath wall, as is widely observed in many C3–C4 species (Sage 2004). Hypertelis spergulacea also showed an increase in vein density to the degree that veins and BSC cells were separated by one to two mesophyll cells only (Fig. 4D). The accessions labeled M. fragilis and M. cerviana have a clearly C4 leaf anatomy, assignable to the “Atriplicoid” type (Fig. 4E–F).


Two major ppc gene lineages exist in the core eudicots, ppc-1 and ppc-2 (Fig. S3). Eudicot ppc were highly supported as monophyletic and were sister to all monocot ppc genes, confirming that the diversification of ppc occurred after the eudicot–monocot split (Christin and Besnard 2009). The species relationships deduced from each of these gene lineages are congruent with the angiosperm phylogeny (APG III 2009).

Relationships among Molluginaceae taxa deduced from both ppc-1 and ppc-2 are compatible with plastid phylogenies (Fig. 5, Fig. S3). Genes encoding C4-optimized PEPC are recognizable by their Ser at position 780, a residue that characterizes the sequenced C4ppc, but not non-C4ppc, in Flaveria, Alternanthera (Amaranthaceae), grasses, and sedges (Engelmann et al. 2003; Gowik et al. 2006; Christin et al. 2007; Besnard et al. 2009; Christin and Besnard 2009). C4ppc were only found in cerviana and fragilis groups and all belonged to the ppc-1 gene lineage (Fig. 5). However, no ppc gene with the C4-specific Ser780 were isolated from the Australian accessions of the cerviana group. Our PCR-mediated ppc survey could have missed the C4-specific genes in these accessions because of mutations in the primer binding sites. More likely, these accessions use a PEPC that is competent but not optimized for the C4 pathway, or is optimized by amino acids that differ from those currently known in other plant groups, such as Ser780. Detailed transcriptome analyses are needed to resolve this issue.

Figure 5.

The evolutionary history of genes encoding PEPC in core Caryophyllales. This figure details the evolutionary history of Caryophyllales ppc-1 lineage. Branches leading to C4-specific genes, identified by a Serine at position 780, are in bold. Species names are followed by accession numbers for sequences retrieved from GenBank and geographic identifications for Molluginaceae. The numbering follows Figure 3 and Table S2. On the right, amino acid at positions shown to be under positive selection for the C4 function in grasses (Christin et al. 2007) and which underwent similar changes in Caryophyllales are indicated. The amino acids at these positions are unknown for M. decandra, for which only a partial sequence was successfully isolated. Amino acids numbering is based on Zea mays PEPC (CAA33317). Residues that differ from the C3 type are highlighted in gray.

In addition to the Ser780, the Mollugo genes for C4 PEPC present several amino acid changes compared to the non-C4 sister genes that also appeared through independent adaptive changes in monocots and other eudicot C4 lineages (Christin et al. 2007; Besnard et al. 2009). These amino acid substitutions are absent from related ppc genes from analyzed C3 and C3–C4 intermediates in the Molluginaceae (Fig. 5). This reinforces the putative C4 function of the genes with the Ser780, and confirms the occurrence of genetic convergence on wide taxonomic scales during C4 PEPC evolution (Besnard et al. 2009). In the fragilis group, the ppc-1 gene was duplicated after the divergence of this group from the cerviana group and only one of the duplicates possesses putatively C4-adaptive amino acids (Fig. 5). In the cerviana group, many of putative C4-adaptive residues are present only in the accessions from South Africa and Spain (Fig. 5).



In our survey, we found evidence of C4 photosynthesis in only two genetic clusters of Mollugo accessions; the cerviana and fragilis groups (Fig. 3). Analyses of genes encoding PEPC showed that the optimization of this C4 key enzyme of the cerviana group followed a gene duplication event that occurred after the divergence from the fragilis group, which demonstrates two independent optimizations of C4 biochemistry in these closely related taxa during the last 10 Mya (Fig. 6). Moreover, C4-characteristic amino acids were acquired in some accessions of the fragilis groups after the divergence of the Australian line (Fig. 5), and a few variations exist at these sites between South African and Spanish cerviana accessions. This indicates that fine tuning of C4 biochemistry took place in this group within the last million years, after the emergence of a functional C4 pathway (Fig. 6). This also shows that a relatively high number of adaptive amino acid replacements can be fixed in a short period of time.

Figure 6.

Gradual acquisitions of C3–C4 and C4 photosynthesis. The major transitions that finally led to the emergence of C4 photosynthesis are indicated by arrows on the calibrated phylogeny of Molluginaceae (see Fig. 3 for details). Dashed arrows indicate alternative scenarios. Diagrams of leaf anatomy (excluding epidermis) on the right are based on Figure 4 and Figure S5, except for Glinus, which is based on Kennedy et al. (1980). They are at the same scale, represented by the red bar at the bottom.

Hypertelis spergulacea, which separates the two C4 groups in plastid and nuclear phylogenies (Figs. 3 and 5), is C3–C4 based on structural traits that characterize well-developed C3–C4 species (Fig. 4D; Monson and Rawsthorne 2000; Sage 2004; McKown and Dengler 2007; Voznesenskaya et al. 2007). The common ancestor of the two C4 groups and H. spergulacea thus likely also exhibited structural characters intermediate between the C3 and C4 conditions and used an intermediate photosynthetic pathway. These morphological and physiological attributes were independently co-opted in the evolution of full C4 syndromes in the cerviana and fragilis lineages. Similar co-options of C3–C4 traits are thought to have occurred during C4 evolution within other C4 groups (Monson and Rawsthorne 2000; Marshall et al. 2007; McKown and Dengler 2007). In C3–C4 intermediates, CO2 released in photorespiration is localized to the BSC due to a loss of glycine decarboxylase (GDC) expression in the mesophyll tissue (Hylton et al. 1988; Monson and Moore 1989; Monson and Rawsthorne 2000). This concentrates CO2 around Rubisco in the BSC and suppresses photorespiration (von Caemmerer 1992). Models of C4 evolution developed from Flaveria predict that exploitation of CO2 release by BSC-localized GDC promotes an expansion of the BSC and an increase of the number of chloroplasts and mitochondria in the BSC, and leads to the establishment of efficient flux networks to rapidly move photorespiratory metabolites between BSC and mesophyll tissues (Monson and Rawsthorne 2000; Sage 2004). Thus, many of the C4 traits are preestablished in the C3–C4 intermediates, and this may facilitate multiple spin-offs of C4 species.


Several species from the sister group to the cerviana/fragilis/H. spergulacea complex have enlarged BSC and reduced interveinal distance compared to C3 taxa (Fig. S5). The distribution of enlarged BSC in the phylogeny (Figs. 3 and 6) suggests that this character appeared more than 20 Mya at the base of the clade containing the C4 groups, H. spergulacea and the Adenogramma/Pharnaceum group (Fig. 6). This could be linked with C3–C4 photosynthesis, but the absence of organelle clusters in the BSC of Suessenguthiella, Adenogramma, Pharnaceum and M. tenella does not support this hypothesis. Enlarged BSC probably evolved in a C3 context, perhaps to act as hydraulic capacitors to buffer sudden surges in transpiration in a hot, windy environment (Sage 2001, 2004). This character was likely co-opted to evolve the C3–C4 type that was further used for the C4 pathway of the cerviana and fragilis groups. Thus, enlarged BSC probably represent developmental enablers sensu Donoghue (2005). The presence of enlarged BSC and close vein spacing reduces the chance that a knockout of mesophyll GDC expression is lethal, simply by reducing the distance between mesophyll and BSC and thus allowing for sufficiently rapid transport of photorespiratory metabolites to the BSC. Similar anatomical traits may have increased the probability of transition to C3–C4 photosynthesis in other lineages where the C4 pathway evolved (McKown and Dengler 2007; Marshall et al. 2007; Muhaidat 2007; Muhaidat et al. 2007).

The distribution of anatomical characters, adaptive amino acid mutations, and functional characters in Molluginaceae phylogeny presents a scenario in which characteristics associated with C4 photosynthesis, such as increased BSC, emerged in a C3 or a moderate C3–C4 context. Successive speciation and migration events presented the plants with new environmental pressures that promoted the emergence of C4-like anatomy in some lineages, probably linked with the development of a C3–C4 biochemistry. These successive co-options of preconditioning characters followed by fine-tuning of the C4 biochemistry, spread over several million years, finally gave rise to widespread, ecologically successful C4 taxa.


Characters of C3–C4 intermediacy appeared at least twice in the Molluginaceae. Mollugo verticillata is positioned within a clade of C3 species, and this taxon is not closely related to any C4 taxon or C3–C4 intermediates, having diverged from the M. cerviana and M. nudicaulis groups more than 40 Mya. C3–C4 photosynthesis also evolved in the clade encompassing M. nudicaulis and H. spergulacea, but it is difficult to determine the exact number of C3–C4 origins in this group. This would require alternative approaches such as comparing the genes controlling specific C3–C4 characteristics (Christin et al. 2010). A C3–C4 type without enlarged BSC could have appeared in their common ancestors and a more pronounced C4-like leaf anatomy may have subsequently evolved in the lineage leading to the cerviana/fragilis groups and H. spergulacea. An alternative scenario would imply two independent origins of C3–C4 intermediacy, one in the nudicaulis group and one in the lineage leading to cerviana-fragilis-H. spergulacea group. This would explain why H. spergulacea and M. nudicaulis have C3–C4 characteristics that are anatomically distinct.

C3–C4 photosynthesis is believed to be a relatively rare condition in plants, with only a few dozen identified species, many of which belong to Flaveria (Sage et al. 1999). Most C3–C4 lineages have only one or two species, and these tend to have restricted distributions, both geographically and in terms of habitat (Frohlich 1978; Powell 1978; Sage et al. 1999). Of all the C3–C4 intermediates, M. nudicaulis and M. verticillata are the most widespread and abundant. Both are found in hot, ruderal habitats where competition is low and the potential for photorespiration is high. Their ability to survive on such sites is likely due to their C3–C4 pathway, which improves carbon gain in the reduced atmospheric CO2 levels that were prevalent in recent geological time (Vogan et al. 2007). The ecological success of these C3–C4Mollugo demonstrates that C3–C4 intermediacy has to be considered as a successful photosynthetic pathway in its own right and not merely a transitional phase to C4 photosynthesis. If all members of the verticillata and nudicaulis groups are C3–C4, then our results indicate C3–C4 intermediacy is older than 8.6 and 19.8 My in these two groups, respectively. Intermediate traits between C3 and C4 were also estimated to be older than 10 Mya in the lineage that gave birth to H. spergulacea (Fig. 6). This evolutionary stability increases the probability that some descendants would fix mutations toward more C4-like characteristics and thus increases the chances of C4 photosynthesis evolving in certain clades.


The earliest C4 origins are estimated to have occurred in the grasses about the time atmospheric CO2 levels fell to near current levels (32–25 Mya), an event that created the primary environmental requirement for C4 evolution in terrestrial plants (Christin et al. 2008). Other C4 clades in grasses, sedges and Amaranthaceae appeared over the subsequent 20 My (Kadereit et al. 2003; Christin et al. 2008; Besnard et al. 2009). Our dating results indicate C3–C4 and C4 photosynthesis in the Molluginaceae also appeared during this period. Although it is currently difficult to place the geographic location for the origins of C4 photosynthesis in the cerviana and fragilis clades, it is plausible to hypothesize that in both the cerviana and fragilis clades, the C4 pathway arose in southwestern Africa. This hypothesis is supported by the restriction of H. spergulacea and the Adenogramma/Pharnaceum clade to this area (Riley 1963; African Plant Database 2010). Many Molluginaceae of this region grow in sandy soils, in climates characterized by very hot summers with some monsoon precipitation (African Plant Database 2010). Because of the episodic monsoon rains, summer photosynthesis by ephemeral species is possible on sandy soils where vegetation cover is sparse and soil moisture is readily available for brief periods. However, temperatures near the soil surface are very hot (>40°C typically) and photorespiration rates in C3 species would be extreme, particularly in the low CO2 atmospheres of the past 30 Mya. It seems likely that the sandy soils of southern Africa, coupled with extreme conditions of very high temperatures and low atmospheric CO2, provided a strong selection pressure for carbon conservation mechanisms that, through successive transitional stages, ultimately led to two C4 lineages in the Molluginaceae.


In this study, we have shown that C3–C4 photosynthesis evolved at least twice and possibly three times in the Molluginaceae, with two subsequent transitions to fully developed C4 photosynthesis. This multiplicity of photosynthetic transitions is remarkable given that both C3–C4 intermediacy and C4 photosynthesis are complex traits that require important genetic modifications. Initial transitions to C3–C4 were facilitated by anatomical traits present in the C3 ancestors, whereas the C4 pathway in two closely related lineages emerged through co-option of C3–C4 characters present in their common ancestor. The cerviana and fragilis C4 groups likely made the transition to a fully functional C4 pathway independently, but they acquired their C4 anatomy from their common ancestor. Similarly, the optimization of the C4 biochemistry occurred independently in geographically isolated members of the cerviana group, although they likely inherited a functional C4 trait from the same ancestor. Thus, although the optimization of the C4 biochemistry occurred within the last million years in the cerviana group, the full transition from a typical C3 ancestor to a fully C4 plant was probably spread across more than 20 million years (Fig. 6). Acknowledging that many apparent “multiple origins” of the C4 trait are in reality only partially independent can contribute to understanding the apparent paradox between the high number of C4 lineages and the genetic complexity of the C4 trait. Rather than considering C3 and C4 photosynthesis as discrete, binary traits, evolutionary studies should consider the probability of developing anatomical characters suitable for a more C4-like condition and then, given the presence of such facilitators, the probability of transition toward a more C4-like state. Such considerations would likely confirm that C4 photosynthesis, as a complex trait, is unlikely to evolve in a randomly selected plant lineage. Instead, the possibility of co-opting anatomical or biochemical traits preexisting in certain groups, and the evolutionary stability of C3–C4 intermediates, can strongly increase the probability of some lineages acquiring C4 photosynthesis. Given this understanding, it is apparent that the study of C4 evolution should examine in detail the traits of the C3 plants related to C4 taxa, to identify when and why critical developmental enablers for C4 first appeared.

Associate Editor: J. Vamosi


This study was partially funded by the Swiss National Science Foundation grant PBLAP3-129423 to PAC, NSERC Discovery grants to RFS and TLS, and National Science Foundation grants IOS-0843231 and DEB-1026611 to EJE. The authors are extremely thankful to the New York Botanical Garden, the Royal Botanical gardens of Kew, the Australian National Herbarium (CANB), the herbarium of the Royal Botanical gardens at Sydney, New South Wales (NSW), the herbarium of the University of Sao Paulo, Brazil (SPF), the Herbarium of the conservatory and botanical garden of the City of Geneva and the Herbarium Senckenbergianum at Frankfurt for providing the necessary plant samples and providing their expertise. We are particularly thankful for the assistance of M. Thulin of the University of Uppsala, M. Chase and E. Kapinos at Kew gardens, L. Gautier and N. Fumeaux at Geneva, and J. Palmer at CANB who provided herbarium samples. We also thank the Compton Herbarium (NBG) at Kirstenbosch for providing material support during a collecting trip in South Africa. We thank D. Hansen for assistance with the fixation of herbarium specimens. We also thank M. Arakaki, G. Besnard, L. Buchi, D. Chatelet, N. Salamin, E. Samaritani, K. Schmandt, and S. Schmerler for helpful discussions and comments on earlier versions of the manuscript.