We determined 18 partial rbcL sequences and these were 929–1322 bp ( = 1293 bp) in length. Four additional gerardioid, Rhinantheae, and outgroup rbcL sequences were obtained from GenBank 30 (Table 3) yielding a 22 taxon data set with an aligned length of 1322 bp. We determined 9 partial ndhF sequences and these were 2096–2122 bp ( = 2115 bp) in length. We were unable to amplify ndhF from nine Agalinis species (A. auriculata, A. divaricata, A. fasciculata, A. filicaulis, A. plukenetii, A. pulchella, A. setacea, A. strictifolia, and A. tenella). We were unable to amplify any part of ndhF using numerous internal and external primer combinations under a variety of PCR conditions, indicating that the lack of amplification was not due simply to modification of one priming site. Difficulty in amplifying ndhF sequences in some Scrophulariaceae genera has been attributed to absence or divergence 27. This phenomenon has not been previously reported among species within a single genus. Three additional sequences for ndhF were obtained from GenBank (Table 3) resulting in sequences for 12 taxa with an aligned length of 2132 bp. We determined 18 matK sequences and these were 2715–3740 bp ( = 2953 bp) in length. Four additional sequences were obtained from GenBank (Table 3), yielding 22 taxa with an aligned length of 3869 bp. The combined data set included 4042–7155 bp ( = 5186 bp) of newly determined sequences for each of 18 taxa. Together with sequences of Seymeria macrophylla and three additional taxa, the combined data set of 22 taxa had an aligned length of 7323 bp.

Sampling from multiple regions of a genome and sampling moderately long total sequence length (approximately 5000 bp or more) have both been shown to significantly improve resolution and support phylogenetic analyses 3839404142. The sequence length used in this study is substantially greater than that included in most single studies in molecular systematics. This amount of sequence data was necessary to meet our challenging objectives of elucidating relationships at multiple levels including potentially relatively recently diverged taxa. In general, longer sequences are more likely to provide more robust estimation of phylogenetic relationships 383941, and improve computational efficiency by increasing differentiation among alternative topologies 4142. Because differences in rate variation among different regions 4344, and differences in rate variation among different codon positions within genes is higher than overall rate variation among genes, all regions sequenced provided some phylogenetic information across all levels of investigation.

Although the greatest rate difference is among the three codon positions within each gene region, the pairwise distances among taxa based on the three gene regions differed (Figure 1), as might be expected. Using the slope of the relationships of the pairwise distances between each pair of taxa as estimated for each pair of gene regions based on linear models we have estimated that the matK gene region evolves 1.615 × faster than rbcL, and ndhF evolves 2.184 × faster than rbcL in these taxa. A simple generalization was not possible for comparing matK and ndhF because the estimated linear model fit to the pairwise distance values for these two gene regions crossed null model expectation of rate equality (Figure 1).

Further analysis suggests that there is rate heterogeneity across lineages among the taxa we examined when Lindenbergia is used as the outgroup for rooting. A test of the molecular clock hypothesis based on the likelihood ratio of the topology with the highest log likelihood (Figure 2) with (-18764.414) and without (-18429.128) assuming a molecular clock significantly rejected the molecular clock hypothesis (P < 0.001).

The best fit likelihood model based on both likelihood ratio tests and Akaike Information Criterion included six nucleotide substitution rate parameters, gamma distributed substitution rates and some invariant sites (i.e., GTR + G + I model). Successive heuristic searching and parameter value estimation yielded a tree of highest log likelihood of -18429.128. The principal implications of this tree are discussed below.

Agalinis, as represented by the North American taxa we sampled and including Tomanthera, is monophyletic with a bootstrap value of 0.96 (Figure 2). We also show that Agalinis is sister to a group composed of the gerardioid genera with yellow or red corollas: Aureolaria, Brachystigma, Dasistoma, and Seymeria with a bootstrap value of 0.99 (Figure 2). These taxa as a whole form a monophyletic group within Rhinantheae as previously suggested based on sequences from Agalinis and Seymeria 24.

Our results suggest that the current section-level classification within the genus needs revision. While a number of sectional and subsectional classifications appear to be natural (i.e., taxa within some sections and subsections form monophyletic groups), there are numerous cases of polyphyly. Sections Erectae and Tenuifoliae as defined by Canne-Hilliker (Table 1) appear to be the only proposed sections comprising more than one taxon that are monophyletic (Figure 2). In contrast, sections Purpureae and Heterophyllae are clearly polyphyletic. Section Chytra proposed by Pennell 4 roughly in place of section Purpureae also is not monophyletic. Section Purpureae is the largest proposed section in Agalinis and is usually treated as being composed of multiple subsections (Table 1). While the section is not monophyletic, each of the subsections as they are represented in our sample except subsection Purpureae appears to be monophyletic (Figure 2). One member of subsection Purpureae, A. fasciculata, appears to be most closely related to A. linifolia although the bootstrap support is low. Agalinis linifolia has always been considered sufficiently distinct from the rest of the members of the genus due to its perennial life history as well as anatomical characters of the stems, roots, and leaves, to be placed in its own section (Table 1 and citations therein). While nodes in the vicinity of A. linifolia do not have high bootstrap values, it does not appear that this species is basal or that it forms a single-species sister group to all other species in the genus (Figure 2). The placement of the other member of subsection Purpureae, A. tenuifolia, is also not well supported as indicated by low bootstrap values in our analysis. However, our data do support removing A. tenuifolia from section Tenuifoliae, as has been suggested based on stem anatomy 23.

Relationships of the other subsections within Purpureae are also weakly supported. The maximum likelihood tree indicates that Subsection Pedunculares is a sister group to section Erectae and together they appear to form a monophyletic clade, although the bootstrap value was low (0.60). This clade unites the two groups within the genus that have a chromosome number of n = 13. This result is surprising because subsection Pedunculares and section Erectae have been considered only distantly related due to differences in floral and vegetative characters 52122. This finding indicates that Pennell's submersion of subsection Pedunculares into subsection Setaceae (Table 1) 4 was in error. While the results support the resurrection of subsection Pedunculares proposed by Canne-Hilliker (Table 1), they do not support inclusion of subsection Pedunculares in section Purpureae. It appears that the inferred ancestral chromosome number for Agalinis is n = 14, and that there was a single chromosome number reduction within the genus.

The two species in section Tenuifoliae in our sample form a strongly supported monophyletic group (bootstrap value = 0.99); a result supporting the naturalness of this section. Section Tenuifoliae appears as a sister group to the n = 13 taxa (section Erectae and subsection Pedunculares of section Purpureae); however, this relationship has low bootstrap support (< 0.50). Section Tenuifoliae was originally considered to be more closely related to section Purpureae 34. Section Erectae was originally considered only distantly related to other sections based on its lighter, yellow-green foliage, lack of tannins and numerous inflorescence and floral characteristics 4. More recently detailed analysis has found many similarities between Tenuifoliae and Erectae 22 that are corroborated by our data.

Although the three groups Tenuifoliae, Pedunculares, and Erectae form a clade; the bootstrap values uniting these groups and uniting the Pedunculares and Erectae to the exclusion of the Tenuifolieae are low to moderate. This clade is interesting because as discussed above it unites two groups that have been considered closely related based on a number of morphological characters (Erectae and Tenuifoliae) and two groups that are morphologically distinct but that share chromosome number (Erectae and Pedunculares). While bootstrap support uniting Erectae and Pedunculares as sister groups is only 0.60, the alternative grouping of Erectae and Tenuifoliae as sister groups never appeared in any of the 2000 bootstrap replicates. Additional molecular data are necessary to provide statistically significant support for relationships among these taxa.

Our results also clarify placement of certain taxa whose relationships have been debated such as placement of A. aphylla in section Erectae 22. Pennell originally considered this taxon to be in its own subsection (Aphyllae) within section Purpureae based on its minute, scale-like, appressed leaves 3. Later he placed A. aphylla in subsection Setaceae of section Chytra 4. Canne-Hilliker moved A. aphylla to section Erectae based on chromosome number (n = 13) 22, seed characteristics 5, and stem anatomy 23. That move is strongly supported by our data.

In another case, A. auriculata, which was placed in the genus Tomanthera 34, clearly falls within the genus Agalinis as has been proposed 23. The genus Tomanthera was distinguished from Agalinis by lack of the yellow guide lines on the lower corolla lip and by having large, lobed leaves; foliaceous calyx lobes; retrorsely hispid stems; raised seed reticulations; and reduced anther cells on the posterior stamens 3. Whereas we demonstrate that A. auriculata is part of Agalinis it is not sister taxon to the other species in section Heterophyllae we sampled (A. heterophylla; Figure 2), making the section Heterophyllae polyphyletic. Although A. auriculata falls well within Agalinis (i.e., it shares a number of inferred common ancestors with other Agalinis species), A. heterophylla, another species placed within section Heterophyllae, is basal to the rest of the species in the genus.

One of our objectives was to evaluate the evolutionary distinctiveness of A. tenella from A. obtusifolia, and A. acuta from A. tenella. We were able to determine that A. tenella and A. obtusifolia are not synonymous as has been suggested 35, and thus submerging A. tenella is not warranted. These taxa are closely related, but are evolutionarily distinct (Figure 2). The branch length from the inferred most recent common ancestor to A. obtusifolia is significant (0.00738, S.E. = 0.00112, P < 0.001), as is the branch length to A. tenella (0.00026, S.E. = 0.00026, P = 0.003). The number of pairwise differences between A. obtusifolia and A. tenella over 3657 aligned nucleotides positions includes 66 substitutions (0.018) and 5 indels involving 7 positions. The differentiation of these taxa is important because A. tenella has been considered to be imperiled in at least two states but merging these taxa would eliminate A. tenella for consideration for protection.

In contrast, we found very little divergence between A. acuta and A. tenella, indeed the smallest amount of divergence among any of the taxa we examined (Figure 2); one substitution over 4048 aligned nucleotide positions (0.0002). Unfortunately, we were unable to obtain ndhF sequence from A. tenella, thus limiting our ability to detect divergence. It is critical to more thoroughly examine this issue using molecular markers that will provide sufficient resolution to determine if these two taxa are really on independent evolutionary trajectories. Given the apparent lack of differentiation in chloroplast genes, yet recognized morphological differentiation, it appears be that nuclear genome markers (e.g., expressed sequence tags [ESTs] and/or microsatellites) sampled from multiple populations combined with coalescent-based analysis and detailed morphological measurement would be useful for studying the relationships of these taxa.

Beyond the genus Agalinis we were able to provide some insights into relationships among the gerardioid genera. Previous thoughts regarding evolution of the these genera that Aureolaria is the most primitive genus and closely resembles the common ancestor of the group 14 are clearly incorrect. Aureolaria is among the most derived genera in our sample (Figure 2). Further, there is a close relationship between Aureolaria and Dasistoma. Divergence between D. macrophylla and A. pedicularia is less than divergence between all pairs of Agalinis species except A. acuta and A. tenella. Similarities in vegetative parts in Aureolaria and Dasistoma have been noted, but the two genera have been considered distinct based on floral morphology 14. On the whole the differentiation among the gerardioid genera (as indicated by relatively short branch lengths) is modest compared to that among species of Agalinis and among other Rhinantheae genera (Figure 2).

A total of 22 taxa were included in our study (Tables 2 and 3). We sampled 15 taxa representing all North American sections of the genus Agalinis as well as Aureolaria pedicularia (L) Raf., Brachystigma wrightii (A. Gray) Pennell, and Dasistoma macrophylla (Nutt.) Raf. (Tables 2 and 3). Sequences from four additional taxa (Seymeria pectinata Pursh, Castilleja linariifolia Benth., Pedicularis foliosa L, and Lindenbergia philippensis (Cham.) Benth.) were obtained from GenBank 30 (Table 3) to help elucidate phylogenetic relationships among Agalinis and the other gerardioid genera.

DNA was isolated from fresh or frozen leaves and flower buds by grinding 50–75 mg of tissue to powder in liquid nitrogen with a mortar and pestle, and then using GenElute Plant Genomic DNA Kits (Sigma Chemical Company, St. Louis, Missouri, USA) following manufacturer's instructions.

We sampled sequences of the following chloroplast gene regions: rbcL, which encodes the large subunit of ribulose-1,5-bisphosphate carboxylase/oxygenase; matK, which includes partial sequence of the gene for ribosomal protein S16 (rps16), an intergenic spacer, the 5' exon of lysine tRNA (trnK), the gene for maturase K (matK), an intron sequence and the 3' exon of lysine tRNA (trnK), and a partial gene for photosystem II D1 protein (psbA); and ndhF which encodes NADH dehydrogenase subunit F.

We selected these gene regions for sequencing because previous studies indicated that ndhF and matK are among the most rapidly evolving protein coding genes in the chloroplast genome 4445, and thus have been firmly established as useful regions at the levels of divergence that are the primary focus of this research. rbcL evolves more slowly, and collectively the three regions provide information across the range of divergence levels we examined (from among species within Agalinis to among genera within Orobanchaceae). Further, the majority of these regions are protein coding thus minimizing insertions and deletions and allowing unambiguous alignments. Finally, the large amount of comparative data from other studies of these gene regions in angiosperms 4546 and particularly in the Scrophulariaceae sensu lato 25262728 allowed us to incorporate existing data for other taxa, provided a strong comparative framework for our results, and allowed us to take advantage of primer sequences designed and tested by others.

Polymerase chain reactions (PCR) were based on Eppendorf MasterTaq PCR kits (Brinkman, Westbury, New York, USA) run on an MJ Research PTC-200 Thermal Cycler. For rbcL we used primer sequences Z-1 and Z-1375 from Zurawski et al. 47 to amplify the whole target region. We also used their internal primer Z-1204R and five internal primers of our own design for sequencing. For the matK region we used primers rps16-4547F and psbA-R from Johnson and Soltis 45 to amplify the whole region. We used seven of their internal sequencing primers and six primers of our own design. For ndhF we used primers 1 and 2110R from Olmstead and Sweere 41 to amplify the entire ndhF region, six of their internal sequencing primers, and two primers of our own design (Additional file 1). Specific amplification conditions for each primer combination in each of gene region varied (Additional file 2). In general, the PCR temperature profile was 30 cycles of 94°C for 60 s, annealing temperature set approximately 5°C below the lower of the two primer melting temperatures for 90 s, 72°C for 150 s, and a final 15 min elongation period at 72°C. PCR products were separated by agarose gel electrophoresis and DNA from fragments of the expected size was extracted and purified using the QIAQuick DNA cleanup system according to manufacturer's instructions (Qiagen Inc., Valencia, California, USA).

Direct sequencing of PCR-generated templates was done using reactions based on the chemistry of BigDye Terminator v3.0 Cycle Sequencing Ready Reaction Kits (Applied Biosystems, Foster City, California, USA) with reactions set up in 96-well microtiter plates using a robotic workstation. Cycle sequencing was performed on MJ Research PTC-200 Thermal Cyclers. Sequencing reactions were cleaned by isopropanol precipitation to remove unincorporated labeled terminators prior to running the samples on an Applied Biosystems 3700 DNA Analyzer. We conducted bi-directional sequencing with ≥ fourfold coverage to ensure high accuracy of the sequence data.

Sequence trace curves were collected on the computer controlling the sequencer. After completion of each set of sequencing runs the trace curves were transferred as SCF-formatted files to a Linux workstation for all subsequent processing and analysis. Base calling and quality assignments were made using the program phred 4849. Data from individual sequencing runs were assembled into final complete sequence using the program phrap 50. Contigs were evaluated with the help of the program consed 51.

Multiple alignments for each gene region were performed using ClustalW 52 and edited if deemed appropriate. These multiple alignments were simple and results were unambiguous because the majority of the sequences coded for proteins and showed relatively few insertion or deletion events. Multiple alignments for each gene region were concatenated to form a single combined data set.

Phylogenetic relationships were determined by maximum likelihood analysis 53 of the aligned nucleotide sequences. Data exploration was done to determine the most appropriate model using preliminary phylogenetic trees and the programs Modeltest 54 and PAUP* 55. Successive heuristic searching (with multiple random taxon addition and tree bisection-reconnection branch swapping) and model parameter value estimation was done to find the highest likelihood tree using PAUP*. Support for specific relationships was assessed with the bootstrap 56 using PAUP*. For the bootstrap analysis, the parameters of the likelihood model were set to those of the highest likelihood tree based on the original data. For each of 2000 replicates a single simple addition heuristic search was conducted with tree bisection-reconnection branch swapping. Estimating likelihood parameter values and applying them to bootstrap replicates is more efficient then re-estimating the values for each bootstrap replicate 57. As a supplementary evaluation, we estimated branch lengths with standard errors, and tested their significance with likelihood ratio tests using PAUP*.

To compare the rates of evolution of the three gene regions, we estimated pairwise distances among all pairs of taxa using the same likelihood model described above using PAUP* 55. We used linear models as implemented by the R system for statistical computing 58 to describe the relationships of distances among these taxon pairs as estimated by each of the three gene regions. We also tested the molecular clock hypothesis (i.e., that all lineages evolved at the same rate) based on the likelihood ratio of the topology with the highest log likelihood with and without assuming a molecular clock.