We enlarged the DNA data matrix of Bellot and Renner (in review) by extracting DNA from additional specimens representing either unusual individuals or potential new species. No DNA sequences could be obtained from Pilostyles stawiarskii, known only from two collections in R, and Pilostyles holtzii, the only collection of which was destroyed in World War II. Suppl. material 1 shows species names and their authors, herbarium vouchers, and GenBank accession numbers. In total, 10 sequences (3 of 18S and 7 of matR) were newly generated for this study.

Total genomic DNA was extracted from herbarium specimens using the commercial plant DNA extraction Invisorb® Spin Plant Mini Kit (Stratec molecular, Berlin, Germany). The mitochondrial matR and the nuclear 18S genes were amplified using the primers listed in Bellot and Renner (in review) . PCR products were purified with the ExoSAP or FastAP clean-up kits (Fermentas Life Sciences, St. Leon-Rot, Germany), and sequencing relied on the Big Dye Terminator v. 3.1 cycle sequencing kit (Applied Biosystems, Foster City, CA, USA) and an ABI 3130-4 automated capillary sequencer. Chromatograms were checked and sequences were edited using Geneious R7 (Biomatters, available from http://www.geneious.com), and contigs were then blasted against GenBank to rule out contamination. Alignments of the clean sequences were performed using the program MAFFT v. 7 ( Katoh 2013 ) resulting in matrices of 1626 and 1727 aligned nucleotides for matR and 18S, respectively. We failed to amplify the gene matR from the African Pilostyles aethiopica and from the Iranian Pilostyles haussknechtii. Phylogenetic reconstructions relied on maximum likelihood (ML) as implemented in RAxML-7.2.8-ALPHA ( Stamatakis 2006 ), using the GTR + G model of nucleotide substitution with 100 bootstrap replicates under the same model. Trees were rooted on Corynocarpus laevigatus (Corynocarpaceae; Cucurbitales), based on Filipowicz and Renner (2010) .

We geo-referenced locality data from 785 herbarium collections on loan from the herbaria B, G, C, GH, K, M, MO, MSB, W, NA, PMA, and SI and added data from the Global Biodiversity Information Facility (GBIF Backbone Taxonomy, 2013-07-01, http://www.gbif.org/species/7279680). We also recorded host names, up-dating their taxonomy as relevant. All label information was compiled in a database using the Botanical Research and Herbarium Management System (BRAHMS, http://herbaria.plants.ox.ac.uk/), and maps were produced using DIVA-GIS 7.5 (http://www.diva-gis.org). Collections were sorted by geography, flowering specimens were sexed to evaluate sexual dimorphism, and a representative number of flowers were then dissected under a stereoscope. For each dissected flower, the first author recorded the number, arrangement and size of the tepals, shape and ornament of the pistil/central column, number of pollen sacs, presence of hairs and presence of a nectary at the base of the flower. Pictures of representative organs were taken using a Dino-Lite USB microscope model AM413ZT (Dino-Lite Europe) and the DinoCapture Imaging software version 2.0 of the same company.

The dissections showed that species have characteristic flower sizes, number of tepals, tepal cilia, and number of anthers rings. For the American species, we use these differences in the key (below). Suppl. material 2 shows measurements and counts from the 123 dissected flowers. Six collections could not reliably be assigned to these groups because their flowers were slightly unusual: R. Callejas et al. 8062, a male plant from Colombia identified as Apodanthes caseariae by A. Idarraga in 2002; Y. Mexia 4540, a female plant from Brazil that is the type of the name Apodanthes minarum; H. S. Irwin et al. 20350, a female plant from Brazil identified as Pilostyles ulei by Ida de Vattimo in 1975; H.S. Irwin 31560, a male plant identified as Pilostyles blanchetii by the first author but parasitizing an uncommon host (Dioclea, Fabaceae); J. Rzedowski 11303, a female plant from Mexico identified by the collector as Pilostyles thurberi; and F. Chiang 9034, a female plant from Mexico identified as Pilostyles thurberi by J. Henrickson in 1972.

The 18S and matR molecular trees show the Pilostyles collections that we wanted to identify (in red on Fig. 1) grouped with Pilostyles thurberi or Pilostyles blanchetii. The collections R. Callejas et al. 8062 and Y. Mexia 4540 grouped with two undoubted representatives of Apodanthes caseariae. R. Callejas et al. 8062 is a male plant and comes from the border with Panama, a country where Apodanthes caseariae has been repeatedly collected. The host of R. Callejas et al. 8062 was originally identified as Trema (Cannabaceae), but a partial matR sequence of this host BLASTed to Casearia nitida, making it likely that the host was in fact a Casearia. If that is the case, this would suggest that the collection represents an Apodanthes. The few male flowers of Apodanthes caseariae that have so far been dissected (Suppl. material 2) do not allow assessing the full morphological variability of the male flowers of this species. Therefore we had to rely on DNA for identification. In terms of its matR (Fig. 1A) R. Callejas et al. 8062 was embedded among other sequences of Apodanthes caseariae, while in terms of its 18S (Fig. 1B), it was sister to them. We identified the specimen as Apodanthes caseariae. Other matR and 18S sequences in the Apodanthes caseariae clade are from the type of the name Apodanthes minarum (Mexia 4540) from Brazil. Its host was a Casearia and its (female) flowers match those of Apodanthes caseariae (Suppl. material 2). We therefore synonymize Apodanthes minarum under Apodanthes caseariae (an action carried out below).

In combination, the present morphological and molecular results show that Apodanthaceae comprise at least ten biological species that can be allocated to two mutually monophyletic genera. In the Americas, these are Pilostyles thurberi in the southern United States of America and Mexico, Pilostyles mexicana in Mexico, Guatemala and Honduras, the widely distributed Pilostyles blanchetii from Panama to Jamaica to Brazil and Uruguay, and Pilostyles berteroi in Chile and Argentina. The Americas also harbor Apodanthes caseariae from Guatemala to Brazil (Fig. 2). Australia has three species, Pilostyles coccoidea, Pilostyles collina, and Pilostyles hamiltonii; Iran has Pilostyles haussknechtii, and Africa has Pilostyles aethiopica. The second African species, Pilostyles holtzii has not been recollected since 1907, when its type collection was made. Another species, the southern Brazilian Pilostyles stawiarskii, is only known from two specimens (one of them the type) collected at the same locality in Jan./Feb. 1948 and Dec. 1949; morphologically it resembles Pilostyles blanchetii (Vattimo, 1950). The host ranges of our accepted genera and species do not overlap. Apodanthes parasitizes only Salicaceae, whereas Pilostyles parasitizes only Fabaceae. As seen on Figure 3, there is a correspondence, although not perfect, between the phylogenies of host genera and parasitic species, and host specialization may have played a role in speciation of Apodanthaceae. At the species level, Table 1 shows that species of Apodanthaceae can grow on one or up to thirteen host species. As seen in Figures 2 and 3, our species concepts are corroborated by geographic and host ranges, except in the case of Apodanthes caseariae and Pilostyles blanchetii, both widespread in Brazil. These two species have different sized flowers (see below), and parasitize phylogenetically distantly related hosts (Fig. 3).

Harms (1935) tried to place the two African names, Pilostyles aethiopica Welw. and Pilostyles holtzii Engl., in a separate section, Pilostyles section Berlinianche, named for their legume host species in the genus Berlinia, but failed to include a Latin diagnosis for the new section. Later, Vattimo-Gil (1955, 1971) decided to rank this section as a separate genus because of the hair cushions on the inner perianth whorl and strictly tri- and hexamerous flowers compared to the tetramerous flowers of the American species of Pilostyles. This assessment, however, could only have been based on specimens of Pilostyles aethiopica, since the only collection of Pilostyles holtzii burnt in World War II. Unfortunately, Vattimo-Gil also neglected to provide a Latin diagnosis, and the genus name is therefore not valid. Based on our results (Fig. 1), Pilostyles aethiopica does not deserve generic status because it is embedded among the other species of Pilostyles.

Flavio González and Natalia Pabón-Mora, at the university of Antioquia in Colombia, are studying the ecology and morphology of Apodanthaceae in Colombia ( González and Pabón-Mora accepted a ) and are describing a new species of Pilostyles ( González and Pabón-Mora accepted b ). This species is the first Pilostyles parasitizing the legume genus Dalea in South America and occurs in dry valleys of the Colombian Eastern Cordillera at altitudes above 2000 m. Morphologically, the new species is most similar to Pilostyles berteroi, which grows in the Chilean and Peruvian Andes at up to 3000 m of altitude (Fig. 2) and parasitizes Adesmia (closely related to Dalea, see Fig. 3).