From an historical perspective, Hooker 7 allied Cynomorium and Balanophoraceae with Haloragaceae (the latter is a family in Saxifragales, see Figure 3) because both groups have epigynous flowers with stamens opposite the valvate perianth lobes. He also noted the striking similarities between the female flowers of Gunnera and Lophophytum (Balanophoraceae). Our analyses that included Gunnera (data not shown) indicate it is not closely related to either Cynomoriaceae or Balanophoraceae, a result in agreement with other work that placed Gunnera as sister to all core eudicots 1718. Other 19^th century workers such as Hoffmeister proposed a relationships between Cynomorium and Hippuris (Plantaginaceae, an asterid) because both have a single, epigynous stamen attached to the top of a unicarpellate ovary. This relationship was not confirmed following a detailed floral morphological study 19 nor was Cynomorium shown to be related to asterids following molecular analyses reported here.

The exact position of Saxifragales within the core eudicot clade remains uncertain because various molecular phylogenetic analyses are in conflict (c. f. 17182021). These studies have proposed relationships with rosids, asterids, caryophyllids, and Santalales, i.e. nearly all members of the core tricolpate clade (see review by Judd and Olmstead 22). The results reported here also demonstrate uncertainty with regard to the topology of the above clades (Figure 2). Given the variety of ways these clades could resolve, the possibility exists that Balanophoraceae and Cynomorium are more closely related than graphically depicted in Figure 2. Although future molecular phylogenetic studies may support this, it is unlikely that these two holoparasite groups will become sister given that the Balanophoraceae + Santalales clade has BS support of 94% and Cynomorium is sister to Peridiscus with 98% BS support. Moreover, the latter two genera are sister to Hamamelis with a 98% BS value.

Long-branch attraction has long been suggested as a confounding factor when conducting phylogenetic analyses of organisms with marked among-lineage rate heterogeneity 23. Indeed, our group has shown that Rafflesiales, another holoparasitic angiosperm group, is susceptible to long-branch attraction, particularly when analyzed with MP 3. The results reported here differ from that study in several respects. First, MP analysis of the global data set does not indicate that Balanophoraceae and Cynomorium exhibit particularly long branches with respect to other angiosperms, nor are these holoparasite clades "attracted" to each other (Additional File 8). This result differs from that obtained when the Rafflesiales molecular data are analyzed with MP where the four distinct lineages (Apodanthaceae, Cytinaceae, Mitrastemonaceae and Rafflesiaceae) appear monophyletic 3. When the global data set is analyzed with BI, some degree of branch length heterogeneity is seen (Additional File 9), however, Balanophoraceae and Cynomorium still remain on distinct clades. It should be pointed out that for the global data set, branch length comparisons between holoparasites and photosynthetic plants are somewhat biased given that the chloroplast genes were coded as missing for the holoparasites. Taken together, we feel our results are not being influence by long-branch attraction and that they provide strong evidence for the independent evolution of these two holoparasite lineages.

Attempts to find morphological features that link Cynomorium with Saxifragales are complicated, first because the latter order is morphologically heterogeneous and second because of morphological reductions in the holoparasite (see below). The heterogeneity of Saxifragales is evidenced by the fact that pre-molecular classifications placed many of its component families in distinct subclasses 8. Although a close relationship between Cynomorium and Crassulaceae was not seen on the BI tree, the shortest MP tree showed these genera as sister taxa, thus this possible affinity will here be considered. Tricolporate pollen (Figure 1e), stem succulence, and proanthocyanidins (red pigments, see 24) are possible characters that may reflect a relationship between Cynomorium and Crassulaceae, but these are all general features found in many groups. Other than such facies, there are few floral characters that could be considered synapomorphies between the extremely reduced flowers of Cynomorium and those of Crassulaceae. The staminate flower of Cynomorium is composed of four to six distinct perianth parts forming a whorl below the single stamen and a wedge-shaped structure interpreted as a pistillode 25. The apical portion of the pistillode is concave and accepts the base of an anther theca prior to filament elongation (Figure 1b) and its internal lateral surface has a groove that accepts the filament. Stamen morphology is not unlike many tricolpate angiosperms: the anthers are dorsifixed, tetrasporangiate, dithecal and open by longitudinal slits. The carpellate flower (Figure 1c) is composed of a single carpel with an elongated, grooved style. The perianth (or perigonial scales) are much smaller than in the staminate flower, reduced to small papillae at the summit of the ovary or along its sides. The bisexual flowers are similar to carpellate flowers except for the addition of a stamen at the base of the ovary (Figure 1d). As with many holoparasitic flowering plants, the reduction or loss of morphological features interferes with the comparative morphological approach when searching for relatives among less modified photosynthetic plants.

Although precedents exists for placing Balanophoraceae with Santalales 826272829, recovering this relationship with both nuclear and mitochondrial genes is surprising because any morphological similarities (e.g., reduction in the gynoecium) have usually been assumed to be cases of convergence 30. Moreover, BI of the Saxifragales data set indicates a derivation of Balanophoraceae from within Santalales, not as its sister (Figure 3 and Additional File 8). Although a multigene analysis of Santalales with robust taxon sampling exists 31, it was generated with nuclear ribosomal DNA and chloroplast genes. To conduct analyses with Balanophoraceae and Santalales, mitochondrial genes from the latter will be needed to test relationships.

Cronquist 8 viewed the sandalwoods as the best candidate ancestral group for Balanophoraceae, however, he admitted any similarities might reflect convergent adaptations to parasitism. Of the 17 genera in the family, only Dactylanthus, Hachettea and Mystropetalon were used in this study because they are sister (and basal) to the remaining genera and have lower substitution rates in SSU rDNA, thus avoiding potential long-branch artifacts. The floral morphology of Mystropetalon is less reduced than other genera in the family. Given that its staminate flowers have pistillodes (vestigial carpels) and carpellate flowers have staminodes (vestigial stamens), it is assumed this plesiomorphic condition reflects evolution from an ancestor with bisexual flowers. Floral features in common with Santalales include a valvate perianth and stamens with "typical" morphology (i.e. anther sacs and filaments as opposed to synangia as seen in Balanophora). When examining the haustorial structure of Mystropetalon, Weber 32 noted the presence of a collapsed zone, graniferous tracheary elements, and "runners," which he associated with Santalales. Similarly, reduction in ovules is seen in several santalalean families, particularly the mistletoes in Viscaceae where no true ovule exists. Either an embryo sac is embedded within central placenta called a "mamelon" or the archesporium develops hypodermally in the ovary, functioning directly as megaspore mother cells 33. Similarly, ovules have been lost in some Santalaceae (Exocarpos) and Loranthaceae. During the course of floral reduction, the loss of ovules most influenced Fagerlind 27 who drew a direct ancestor-descendant relationship between Santalaceae and Balanophoraceae.

As mentioned above, such apparent similarities between Balanophoraceae and Santalales have also been interpreted as convergent. Kuijt 10 suggests that the reduction of santalalean ovules may permit retention of meristematic activity, a requirement for the formation of complex fruits in epiparasites. For root parasites such as Balanophoraceae, he suggests ovular reduction is related to selection for tiny seeds that require host stimulants for germination, as is seen in Orobanchaceae. In addition to ovules, the androecia of both Viscaceae and Balanophoraceae are diverse with both showing trends toward the formation of synandria and porose dehiscence. Whether these trends are convergent or follow from homologous suites of genes may yield to direct testing using floral homeotic mutants.

Although herbaceous root parasites have evolved in Santalaceae (e.g. Thesium, Quinchamalium), most members of Santalales are woody. Completely holoparasitic species do not occur in this order, with the possible exception of Daenikera, an endemic monotypic genus of New Caledonia whose photosynthetic capabilities remain to be determined. As adults, dwarf mistletoes (Arceuthobium, Viscaceae) fix only about 30% of the carbon needed for growth 3435 but prior to host attachment seedlings actively photosynthesize. For this reason these mistletoes must be considered advanced obligate hemiparasites, not holoparasites. Thus, to include with Santalales the entirely holoparasitic group Balanophoraceae requires further investigation to identify the exact sister group.

Cynomorium coccineum, known to the Muslim world as "tarthuth," has been harvested from the deserts of north Africa and the Middle East for thousands of years. Arabs and Bedouins eat the interior portions of fresh young stems, prepare infusions of older stems to treat colic or stomach ulcers, or dry and pulverize the plant for use as a spice or condiment with meat dishes 36. Medicinal uses of tarthuth can be traced to Al-Kindi, Al-Razi (Rhazes), Ibn Masawayh, Ibn Wahshiya, and Maimonides but the plant became known to Europeans only in the 16^th century. A group called the Knights Hospitaller of St. John operated a hospital in Jerusalem and learned of the medicinal qualities of tarthuth from local Muslim physicians. When the Crusaders lost Jerusalem to the Muslims, they moved to the island of Malta where Cynomorium was also native. The site where the "Maltese Mushroom" grew (Fungus Rock) was thereafter vigorously guarded and thieves were imprisoned or made galley slaves. The "treasure of drugs," as the Arabs called it, was used for a variety of purposes, including treating apoplexy, venereal disease, high blood pressure, vomiting, irregular menstrual periods and as a contraceptive and toothpaste.

Modern biomedical and phytochemical research on Cynomorium coccineum has demonstrated a variety of activities from plant extracts. Effects of Cynomorium extracts on mammalian reproductive cells modulation of pituitary gonadotrophins 37 as well as changes in testicular development 38 and epididymal sperm patterns 39 in rats. Cynomorium songaricum, known in Chinese medicine as "suo yang," has been shown to contain triterpenes with HIV protease inhibitory activity 4041. Interest in herbal medicines is growing at a rapid pace, and attention has been focused upon Cynomorium as evidenced by its availability via hundreds of distributors advertising (via the internet) herbal remedies for kidney and intestinal ailments as well as for impotence. Although it is not at present clear how many original distributors of the plant exist, and whether plant material being marketed as Cynomorium is actually this plant, what is clear is that it is not being cultivated, thus authentic herbal preparations must be obtained from wild populations. Very little information exists on the cultivation of Cynomorium 25 and certainly commercial suppliers are not practicing sustainable harvest by cultivating this obligate holoparasite. There is evidence that overexploitation of this plant has resulted in localized extinction 42. For these reasons, we here raise concerns for the conservation of both species of Cynomorium and strongly voice the need to develop cultivation methodologies. Given the potential and actual biomedical applications of extracts from this plant, and conservation concerns given extensive harvesting from wild populations, information on its phylogenetic position within angiosperms is timely. Further molecular work will likely illuminate its closest relatives within Saxifragales. These taxa should then become the subject of phytochemical analyses to determine whether they also contain compounds of biomedical interest. Cultivation of photosynthetic plants would be more straightforward than the holoparasite, thus possibly relieving some of the pressure to harvest this more sensitive species from the wild.

Voucher information for all newly sequenced taxa is given in Additional Files 1 and 2. DNA was extracted, PCR amplified, cloned, and sequenced following reported methods 343. The nuclear and mitochondrial sequences were amplified using primers reported elsewhere 144445. Sequencing was conducted using an ABI Prism^® 377 automated DNA sequencer (Applied Biosystems) according to the manufacturer's protocols. Two multiple sequence alignments were constructed. The purpose of the global data set was to test the position of Cynomorium and the three Balanophoraceae genera among angiosperm orders. This matrix included nuclear small-subunit (SSU) rDNA, chloroplast rbcL and atpB, and mitochondrial matR for 67 taxa with Laurus and Cinnamomum as outgroups. Balanophoraceae and Cynomorium lack the chloroplast genes rbcL and atpB, but they were included in the global data set to stabilize the overall tree topology as previously demonstrated 23. Given the strong support for a relationship between Cynomorium and Saxifragales obtained from this global analysis, a Saxifragales data set was constructed to further resolve the position of the parasites within this clade. Here the genes used were nuclear SSU and LSU rDNA, chloroplast rbcL, atpB, and matK for 62 taxa with Lea and Vitis as outgroups. In both alignments, chloroplast genes were coded as missing for the holoparasites. Accession numbers for newly sequenced genes are: AY957440 – AY957454.

The global and Saxifragales data sets were analyzed using maximum parsimony (MP) in PAUP* 4.0b10 46 and Bayesian inference (BI) methods in MrBayes 3.0b4 47. MP searches were performed using 100 random addition sequence replicates with tree-bisection-reconnection (TBR) branch-swapping, holding ten trees at each addition step, with all sites equally weighted. Fully partitioned Bayesian analyses were performed by partitioning each data set by gene and, for the protein-coding genes, by codon position. This resulted in a total of 10 partitions for the full data set and 11 partitions for the Saxifragales data set (Table 1). MP trees were constructed for each gene (following the protocol described above), and these trees were used in PAUP* to evaluate 24 nucleotide substitution models for each data partition. For example, models were evaluated for the "rbcL Pos1" partition (the first codon position of the rbcL data set) on the MP tree for the rbcL data set. MrModelTest 2.0 48 was used to select an appropriate model from the PAUP* output via a second-order version of the Akaike Information Criterion (AICc) that takes sample size into account, as recommended by Posada and Buckley 49. AICc values were computed using both the total number of characters and the number of variable characters per partition. For nearly all partitions, the Akaike weight for the chosen model was much higher than the Akaike weight of the next best model, so model-averaged analyses were not performed. The best-fitting models and their Akaike weights for each data partition are listed in Supplementary Data. Partitioned Bayesian analyses were performed with all model parameters unlinked (i.e., model parameters for each data partition were estimated separately from each partition), and with topology and branch lengths linked. Two separate analyses (with different MCMC seeds, random starting trees and default uniform priors for all parameters) were run for each data set for 15 million generations, with trees sampled every 500 generations. Trees recovered during the first 2.5 million generations (the first 5000 trees) in both runs were discarded as burn-in, leaving a total of 25,000 trees which were used to construct majority-rule consensus trees.