C HLOROPLAST DNA SEQUENCE UTILITY FOR THE LOWEST

Transcription

1 American Journal of Botany 101 (11 ): , C HLOROPLAST DNA SEQUENCE UTILITY FOR THE LOWEST PHYLOGENETIC AND PHYLOGEOGRAPHIC INFERENCES IN ANGIOSPERMS: THE TORTOISE AND THE HARE IV 1 J OEY S HAW 2,3,5, H AYDEN L. SHAFER 2, O. RAYNE L EONARD 4, M ARGARET J. KOVACH 2, M ARK S CHORR 2, AND A SHLEY B. MORRIS 4 2 Department of Biological and Environmental Sciences, University of Tennessee at Chattanooga, Chattanooga, Tennessee USA; 3 Botanical Research Institute of Texas, Fort Worth, Texas USA; and 4 Department of Biology, Middle Tennessee State University, Murfreesboro, Tennessee USA Premise of the study: Noncoding chloroplast DNA (NC-cpDNA) sequences are the staple data source of low-level phylogeographic and phylogenetic studies of angiosperms. We followed up on previous papers (tortoise and hare II and III) that sought to identify the most consistently variable regions of NC-cpDNA. We used an exhaustive literature review and newly available whole plastome data to assess applicability of previous conclusions at low taxonomic levels. Methods: We aligned complete plastomes of 25 species pairs from across angiosperms, comparing the number of genetic differences found in 107 NC-cpDNA regions and matk. We surveyed Web of Science for the plant phylogeographic literature between 2007 and 2013 to assess how NC-cpDNA has been used at the intraspecific level. Key results: Several regions are consistently the most variable across angiosperm lineages: ndhf-rpl32, rpl32-trnl (UAG), ndhctrnv (UAC), 5 rps16-trnq (UUG), psbe-petl, trnt (GGU) -psbd, peta-psbj, and rpl16 intron. However, there is no universally best region. The average number of regions applied to low-level studies is ~2.5, which may be too little to access the full discriminating power of this genome. Conclusions: Plastome sequences have been used successfully at lower and lower taxonomic levels. Our findings corroborate earlier works, suggesting that there are regions that are most likely to be the most variable. However, while NC-cpDNA sequences are commonly used in plant phylogeographic studies, few of the most variable regions are applied in that context. Furthermore, it appears that in most studies too few NC-cpDNAs are used to access the discriminating power of the cpdna genome. Key words: chloroplast; DNA barcode; intergenic spacer; intraspecific; intron; noncoding cpdna; phylogeography; plastid region; plastome; tortoise and hare. The chloroplast genome has long been recognized as the workhorse for testing relationships between biological and geographical phenomena in angiosperms (Palmer, 1987 ; Palmer et al., 1988 ; Olmstead and Palmer, 1994 ; Soltis et al., 1997 ; Graham and Olmstead, 2000 ; Kelchner, 2000 ). Chloroplast sequences have been used successfully to infer relationships at all taxonomic levels, from the deepest-level relationships of land plants and angiosperms ( Chase et al., 1993 ; Borsch et al., 2003 ; Hilu et al., 2003 ; Moore et al., 2010 ), through intermediate taxonomic levels of orders and families ( Downie et al., 2000 ; Potter et al., 2007 ; Chin et al., 2014 ), to relationships among closely related species or populations ( Soltis et al., 1997, 2006 ; Schaal et al., 1998 ; Shaw and Small, 2004 ; Morris et al., 2008 ). 1 Manuscript received 9 September 2014; revision accepted 17 September The authors thank Brad Ruhfel, Stephen Downie, an anonymous reviewer, and the Associate Editor for their careful consideration of this article. The authors thank Ed Schilling (University of Tennessee) and James Beck (Wichita State University) for reviewing early drafts of this manuscript. They also extend special thanks to R. Small, E. Schilling, E. Lickey, J. Beck, K. Grubbs, S. Farmer, W. Liu, J. Miller, and C. Winder for their work and important contributions to the earlier chapters of this research and the Hesler Foundation at the University of Tennessee for financially supporting the earlier chapters. 5 Author for correspondence ( joey-shaw@utc.edu) doi: /ajb It is at this shallowest taxonomic level, (inter- and intraspecific studies) that researchers are often challenged with finding sufficient genetic variability to address their questions ( Schaal et al., 1998 ; Schaal and Olsen, 2000 ; Holderegger and Abbott, 2003 ; Petit and Vendramin, 2007 ). Shaw et al. (2005, 2007 ) provided guidance in the search for the most consistently variable noncoding chloroplast regions (NC-cpDNA). In The Tortoise and the Hare II (TH2) ( Shaw et al., 2005 ), the focus was to compare the relative number of genetic differences typically found in NC-cpDNA regions that were prevalent in the literature at that time. In that study, 21 NC-cpDNA regions were assessed for utility within each of 10 seed plant lineages. Each lineage comprised three congeneric species. The results were surprising in that the most commonly used regions at that time (e.g., trnl intron, trnl-trnf ), were among the least informative, while some uncommonly used regions ( trnd-trnt, trns-trng, and rpob-trnc ) appeared to be much more informative. In The Tortoise and the Hare III (TH3) ( Shaw et al., 2007 ), the focus was to expand genomic sampling to all single-copy NC-cpDNA regions. Initially, paired plastomes from three angiosperm lineages ( Atropa vs. Nicotiana [Solanaceae]: asterid; Lotus vs. Medicago [Fabaceae]: rosid; and Saccharum vs. Oryza [Poacae]: monocot) were used to screen all single-copy NC-cpDNA regions to search for any that might have higher variability than the best regions identified in TH2. The outcome highlighted 13 regions, and these were sequenced across the same three-species groups as the TH2 study. In the end, nine rarely or never before American Journal of Botany 101 (11 ): , 2014 ; Botanical Society of America 1987

2 1988 AMERICAN JOURNAL OF BOTANY [Vol. 101 used NC-cpDNA regions were identified as being, on average across angiosperms, the most informative for low-level studies ( rpl32-trnl (UAG), trnq (UUG) -5 rps16, 3 trnv (UAC) -ndhc, ndhfrpl32, psbd-trnt (GGU), psbj-peta, 3 rps16 5 trnk (UUU), atpiatph, and petl-psbe ). Before the publication of TH3, very few congeneric species pairs of whole plastomes were available in GenBank, and we were fortunate to find species pairs within the same family (asterids: Solanaceae; rosids: Fabaceae; and monocots: Poaceae). There are now ~340 angiosperm chloroplast genomes in Gen- Bank (January 2014), many of which represent congeneric accessions. This wealth of recent data allows us the new opportunity to more thoroughly screen the utility of noncoding regions of the plastome across a wider spectrum of the angiosperm phylogeny to evaluate inferences made in TH3. As a consequence of the rapidly increasing number of whole plastomes now available in public databases, we have an opportunity to learn from the growing availability of such data for the benefit of the larger community of researchers not yet sequencing chloroplast genomes. With the rapid advancement of next-generation sequencing (NGS) approaches, we have entered the Golden Age of Molecular Ecology ( Paliy, 2013 ), and sequencing whole plastomes, or even hundreds of nuclear loci, for low-level inquiry is certainly the future ( Soltis et al., 2013 ). It is becoming increasingly feasible to sequence whole or nearly whole plastomes using highthroughput methods ( Cronn et al., 2012 ; De Wit et al., 2012 ; Shi et al., 2012 ; Straub et al., 2012 ; Stull et al., 2013 ; Uribe-Convers et al., 2014 ), and the cost savings over Sanger sequencing can be significant for large-scale studies involving many taxa ( Godden et al., 2013 ) or for laboratories that are already established with the needed laboratory and computational infrastructure. Core sequencing facilities now offer opportunities for multiple laboratories to share the cost of an NGS run, further lowering the price per sample. While there are a few notable cases in which researchers have published phylogenies based on whole plastome data, most are targeted at resolving deep relationships ( Ruhfel et al., 2014 ) or tackling questions in model organisms ( Eserman et al., 2014 ) or economically important groups ( Nikiforova et al., 2013 ; Njuguna et al., 2013 ), but this is rapidly changing. At present, the majority of whole plastome publications are limited to one or a few accessions per study (e.g., Martin et al., 2013 ; Walker et al., 2014 ), and these typically come from the larger research laboratories, reflecting the financial and computational costs associated with this kind of data generation ( Rocha et al., 2013 ; Soltis et al., 2013 ). The start-up costs for hardware and software alone may prohibit smaller, lesser-funded laboratories around the world from moving from sequencing a few NCcpDNA regions to generating massive quantities of genomic data. For smaller-scale studies, or studies coming from lesser-funded or equipped laboratories, the cost of NGS and the complexity of bioinformatics analyses required for such large data sets may still be prohibitive ( Doyle, 2013 ; Rocha et al., 2013 ; Bowen et al., 2014 ; but see Straub et al., 2012 for an alternative viewpoint). Bowen et al. (2014), in summarizing results from a meeting at the National Evolutionary Synthesis Center (NESCent), suggested that (1) single mtdna locus studies (for plants, we would modify this phrase to one to several cpdna loci studies ) provide powerful first assessments of patterns; (2) genome-wide analyses are warranted if results from standard markers are discordant with other aspects of the organisms biology; (3) at this stage of software development and technology, a judicious rather than wholesale application of genomics is prudent. Given all of this, we anticipate that Sanger or NGS of a smaller number of individual NC-cpDNA regions will continue to be an important approach for many exploratory inquiries, low-level systematics and phylogeographic studies, and for barcoding studies, at least for the near future. Therefore, there is still a need to be able to select from a few, presumably the most potentially informative, NCcpDNA regions. In the present study, we asked: Which, and how many, of the most informative regions of the TH series have been useful at the lowest phylogenetic and phylogeographic levels in angiosperms? While we are particularly interested in intraspecific data, there are currently no publicly available whole plastome phylogeography data sets, nor are there many publications for such an assessment (but see Whittall et al., 2010 and Doorduin et al., 2011 ). Therefore, we addressed our question by analyzing a large set of whole plastomes, reflecting as many low-level taxonomic comparisons as are currently available through the NCBI Organelle Genome Resources (GenBank). We used 25 closely related species pairs congeners when possible (20/25) across 10 major lineages of angiosperms (Nymphaeales, magnoliids, monocots, Proteales, eurosid I, eurosid II, Caryophyllales, Ericales, euasterid I, euasterid II) ( Fig. 1 ) to compare all single-copy NCcpDNA regions and matk because it is among the most variable coding regions and has been a popular region since 1997 ( Hilu and Liang, 1997 ). We also were able to compare the results within the major clades to overall results. In addition to this new data set, we searched the literature covering the period since the last TH paper ( ). To determine whether NC-cpDNA sequences are an important tool for low-level studies, we asked: What percentage of intraspecific studies of plants used cpdna sequence data? Beyond this initial question, we also asked: Which NC-cpDNA regions were chosen and sequenced? How many NC-cpDNAs were needed to generate sufficient data? How many papers explicitly reported screening NC-cpDNAs, and how many were typically screened? All results are discussed in the context of the trends in the current literature and findings of our previous TH studies. In the end, we propose a strategy for investigators of low-level taxonomic or phylogeographic studies using NC-cpDNA sequence data. MATERIALS AND METHODS Taxon selection for genomic comparisons Initially, we set out to use only congeneric species pairs in angiosperms; however, doing so, left us with phylogenetic gaps in our sampling. To fill in these phylogenetic gaps, we used the most closely related species pairs available in a few lineages, resulting in an initial list of 34 angiosperm species pairs for comparison. After we had compiled all of the data, four species pairs were removed ( Acorus americanus and A. calamus, Nicotiana sylvestris and N. tabacum, Oryza nivara and O. sativa, and Phyllostachys edulis and P. nigra var. henonis ) because they contained too few genetic differences between them, as we later explain in detail. We also had to remove five other species pairs ( Wolffi a australiana and Wolffi ella lingulata, Colocosia esculenta and Lemna minor, Ranunculus macranthus and Megaleranthus saniculifolia, Cuscuta exaltata and C. obtusifl ora, and Erodium carvifolium and E. texanum ) because they were too variable to be confidently aligned across the entire genome (the large indels and genomic rearrangements in these species pairs that made our analyses difficult could certainly provide workers of these groups with important information in other contexts). The remaining 25 species pairs (including 20 congeneric pairs) used in our analyses are shown in Fig. 1. We made a strong effort to sample across the phylogenetic breadth of angiosperms, using nomenclature and topology from the Angiosperm Phylogeny Group ( ). We made an effort to sample about the same number of species pairs per major lineage. In the end, we were able to sample one species pair from Nymphaeales, two from

3 November 2014] SHAW ET AL. TORTOISE AND HARE IV 1989 Fig. 1. Phylogenetic breadth of sampling for plastome comparisons. Species pairs (mostly congeners) are listed beside family names and next to major lineages. Tree topology follows Angiosperm Phylogeny Group III (APG III) ( ). magnoliids, five from monocots, one from Proteales, four from eurosid I, four from eurosid II, one from Caryophyllales, one from Ericales, three from euasterid I, and three from euasterid II ( Fig. 1 ). NC-cpDNA selection Rather than try (1) to determine exactly where the large single copy (LSC), small single copy (SSC), and inverted repeat (IR) regions occur in each of the 25 lineages and (2) expand the data set to include uncommon intergenic spacers (i.e., intergenic spacers unique to a given taxon because of genomic rearrangements unique to that taxon), we used the Gossypium hirsutum genome ( Lee et al., 2006 ) as a model. It was used because it is a typical chloroplast genome to determine which regions to include or exclude as well as the starting and stopping points of the LSC, SSC, and IRs. It is well accepted that the most variable portions of the cpdna genome are not in the IR regions ( Clegg et al., 1984 ; Wolfe et al., 1987 ). The IR regions have been shown to contain low levels of variability ( Maier et al., 1995 ), and Presting (2006) confirmed and quantified earlier reports that the IR was significantly more resistant to genetic changes compared to the single-copy regions. We therefore concentrated on the single-copy portions of the genome. The two single-copy regions consist of the LSC, which is typically about 80 kb long, and the SSC, which is typically 20 kb. These regions contain a combination of rrna and trna genes, as well as protein-encoding genes. It is common knowledge that gene-encoding regions will accumulate genetic differences more slowly than noncoding regions, the exceptions being matk (Steele and Vilgalys, 1994 ; Johnson and Soltis, 1995 ; Fazekas et al., 2008 ; Lahaye et al., 2008 ) and perhaps ycf1 ( Dong et al., 2012 ), which seem to accumulate mutations at a fast pace for coding regions. Because matk has historically been an important region ( Hilu and Liang, 1997 ) and is commonly recognized as being as variable as many noncoding regions of the plastome, our research effort was focused on matk and all single-copy, noncoding regions of the plastome. In total, 107 noncoding regions and matk were compared across 25 species pairs spanning the breadth of angiosperms. Sequence alignment The GenBank nucleotide BLAST function was used for initial alignment. BLAST aligns local regions of two separate sequences based on nucleotide similarity. In initial BLAST searches, we found that some of the more genetically dissimilar species pairs (e.g., Aethionema ) often contained too many gaps to ensure reliable alignments. Thus, we altered the default BLAST algorithm parameters to increase the standard linear gap cost, as a measure to reduce the presence of misaligned sequences. BLAST parameters were as follows: Max target sequences = 100, expected threshold = 10, Max matches in a query range = 0, Match/Mismatch scores = 1-2, and Gap Costs = Existence: 5; Extension: 2. After the sequences were aligned with BLAST, the alignments were visualized in Portable Document Format, and the gene-encoding regions were unequivocally denoted using the annotations in GenBank. Alignments were manually scored for genetic differences. In a few cases where small sections of the alignments were too variable to be confidently aligned (e.g., poly A/T regions), gaps were opened up, and these were conservatively scored as a single genetic difference (an indel). Data analysis All noncoding portions of the genome, including intergenic spacers (IGS) and introns, were analyzed for genetic differences between the

4 1990 AMERICAN JOURNAL OF BOTANY [Vol. 101 species pairs, and these genetic differences were counted as potentially informative characters (PICs) following Shaw et al. (2005, 2007 ). (PICs have been shown to be a good predictor of parsimoniously informative characters [ Fior et al., 2013 ].) The PICs included indels, nucleotide substitutions, and inversions. Indels and inversions were scored as binary (presence/absence) characters following Simmons and Ochoterena (2000). The process of alignment, annotation, and analysis was repeated for each of the 25 species pairs. The length of the noncoding region was recorded, and the number of PICs tallied for each noncoding cpdna region and matk in each of the 25 lineages. All genetic differences were scored as independent, single characters. Three types of calculations were performed to evaluate NC-cpDNA variability. First, we estimated the proportion of observed genetic differences for each NC-cpDNA using a modified version of the formula used in O Donnell (1992), Gielly and Taberlet (1994), and Shaw et al. (2005, 2007 ). The proportion of genetic differences (or % variability) = [(NS + ID + IV) / L] 100, where NS = the number of nucleotide substitutions, ID = the number of indels, IV = the number of inversions, and L = the aligned sequence length. Second, we averaged the number of PICs found within each noncoding chloroplast region across the 25 lineages that contained those NC-cpDNA regions (if the region was missing in a lineage, the total was divided by 24 rather than 25). Third, to ensure that lineages containing a greater number of genetic differences between the species pair (greater evolutionary distance) were not overrepresented (weighted) in global comparisons, we determined the percentage contribution of each noncoding region to the overall PIC total within a lineage (what was called normalized PICs in earlier TH papers). In effect, we calculated the percentage contribution of a NC-cpDNA to the number of genetic differences observed in a species pair. These values were then used to generate an average value for each noncoding region so that the regions could be directly compared. We argue that the percentage contribution values are the most accurate for comparisons of NC-cpDNA variability across lineages because they reduce the overrepresentation of average PIC values from the species pairs that evolved at faster rates or are evolutionarily further apart. For example, Silene species accumulated 1483 PICs across all regions, while Olea species only had 321 PICs. Without this normalization of the data, the species pair with the greatest overall number of genetic differences (e.g., Silene ) would more strongly influence the analysis compared with other lineages that had many fewer genetic differences between species. A downside to the percentage contribution analysis is that species pairs with a very low number of genetic differences between them tend to have overly high scores in this analysis. For example, the Nicotiana species pair was omitted because there were only two nucleotide substitutions across the entire alignment, and therefore, each NC-cpDNA containing them would have the very large percentage contribution PIC score of For this reason, we omitted species pairs with fewer than 100 PICs in the alignment (discussed earlier in Taxon selection for genomic comparisons ). Adjusted mean PICs A mixed-model analysis of covariance (ANCOVA) was used to look for statistical separation between the 10 NC-cpDNAs that ranked highest in the overall percentage contribution analysis. We chose to analyze only 10 regions because statistical inference would be weakened with all 108 regions. Furthermore, we argue that if statistical separation among the top 10 is present, then it is sure to exist between these and the rest. The mean PICs/ region was treated as the response variable and the fixed-effects factor was the NC-cpDNAs; lineage was treated as a block (random-effects factor) and region length (number of base pairs per region) as a covariate. Data were log-transformed [log 10 (PIC + 1)]; [log 10 (length + 1)] to satisfy the parametric test (ANCOVA) assumptions of normality, variance homogeneity, and linearity. Region-specific PIC estimates were reported as geometric means ± SE. A preliminary ANCOVA indicated that there was a direct linear relationship between the number of PICs/region and length of the region ( P < ) and that the slopes of PIC-length relationships were similar ( P > 0.10), which allowed us to compare slope-adjusted mean PICs/region. Region-specific estimates of adjusted mean PICs ( y ), which control for the effect of region length ( x, covariate), were determined using linear regression models [log ( y + 1) = a + b( x + 1)] with a common slope ( b = ) and length value ( x = mean length = ~1073 bp). Scheffé s method was used to make pairwise comparisons (contrasts) between region-specific mean PICs and/or groups of mean PICs, based on patterns exhibited by the data. Scheffé s procedure, a highly conservative post hoc test that maintains a familywise type I error rate at or below the alpha level, is recommended for multiple contrasts between groups of sample means (Zar, 2010 ). Data were analyzed using the UNIVARIATE, REG, GLM, MIXED, and LSMEANS procedures of the Statistical Analysis System ( SAS Institute, 2011 ). Statistical significance was declared at an alpha level of Literature review To answer the question What proportion of intraspecific plant studies used cpdna sequence data?, we performed a Web of Science search in May 2014 by confining the dates to and searching on the terms phylogeography or phylogeographic, refined to articles, then further refined by *aceae. Subsequent refinements were performed using either nuclear or ndna or mitochondria or mitochondrial or mtdna or chloroplast or plastid or cpdna. To address the other questions posited in the Introduction section, we performed a search on Web of Science in January We first used the topic search terms phylogeography or phylogeographic limited to the years We refined this number using the topic search terms chloroplast or cp- DNA to focus specifically on plant phylogeographic studies. While this may exclude papers that exclusively use nuclear markers, we determined this to be the most appropriate search strategy for the present survey. The top five source titles for these plant phylogeographic studies during the selected 7 years were Molecular Ecology (99), Journal of Biogeography (79), Molecular Phylogenetics and Evolution (45), American Journal of Botany (32), and Plant Systematics and Evolution (28), representing approximately 41% of total publications under our search criteria. Because our emphasis here is on low taxonomic level and population studies, we excluded papers in Molecular Phylogenetics and Evolution and Plant Systematics and Evolution due to the more traditional systematic nature of publications therein. Of the total of 210 plant phylogeographic studies in the remaining three journals ( Molecular Ecology, Journal of Biogeography, and American Journal of Botany ), 33 were excluded following our review for either being a review or focused on an animal, more than three species, or a nonvascular plant. This exclusion resulted in 178 publications being included in our review. We built a database containing the following information for each of these 178 studies: author, year of publication, source title, taxon, family, markers tested (when available), markers used, and length and PICs for markers used (when available). We used these data to answer the questions posited in the introduction. We were also interested in comparing relative utility of regions within studies to determine whether patterns predicted by TH2 and TH3 were supported by the literature. For this comparison, we could only include studies that explicitly stated the number of PICs/region and used at least two or more regions. Of the 178 studies included in our review, 45 met these criteria. The remaining studies either reported PICs across all regions combined, or many did not report PICs at all. For these 45 studies, NC-cpDNAs were ranked by decreasing number of PICs observed, and trends were qualitatively assessed across studies. RESULTS Overview of molecular data set In all, we manually examined over 2.5 million base pairs of aligned data from 25 species pairs, resulting in a data set of 107 single-copy NC-cpDNA regions including 15 introns, 92 IGS, and matk for 25 species pairs (Appendix S1, see Supplemental Data accompanying the online version of this article). Taxon selection, omission, and results on omitted taxa At the outset, we identified 34 congeneric or fairly closely related species pairs; however, five of the pairs were too variable to be confidently aligned, and four were too invariable to yield reliable information. There were only two genetic differences between Nicotiana sylvestris (NC_007500) and Nicotiana tabacum (NC_001879); one difference was in the psbm-trnd (GUC) spacer, and the other was in the petd intron. There were only 22 genetic differences between Acorus americanus (NC_010093) and Acorus calamus (NC_007407). These differences were found in ndhf-rpl32 (1), ccsa-ndhd (1), ndhg-ndhi (1), 5 trnk (UUU) - 3 rps16 (2), rps16 intron (1), trnq (UUG) -psbk (1), trnc (GCA) -petn (2), trnd (GUC) -trny (GUA) (1), trnt (GGU) -psbd (3), trnt (UGU) - trnl (UAA) (1), trnm (CAU) -atpe (1), atpb-rbcl (1), ycf4-cema (1), peta-psbj (1), psbe-petl (1), psaj-rpl33 (1), rps18-rpl20 (1), and rpl16 intron (1). We only observed 26 genetic differences between Oryza nivara (NC_005973) and Oryza sativa (NC_ ), and these were found in ndhf-rpl32 (1), ccsa-ndhd

5 November 2014] SHAW ET AL. TORTOISE AND HARE IV 1991 (1), ndha intron (1), psba-3 trnk (UUU) (1), rps16 intron (2), trnd (GUC) -trny (GUA) (2), trne (UUC) -trnt (GGU) (2), psbz-trng (GCC) (2), ycf3-trns (GGA) (2), ndhc-trnv (UAC) (3), trnm (CAU) -atpe (1 ), accd-psai (1), petl-petg (1), petd-rpoa (1), rps11-rpl36 (1 ), rpl14-rpl16 (2), rpl16 intron (1), and rpl22-rps19 (1). Twentyfour genetic differences were observed between Phyllostachys edulis (NC_015817) and Phyllostachys nigra (NC_015826). In Phyllostachys, differences were found in rpl32-trnl (UAG) (5), ccsa-ndhd (2), psac-ndhe (2), matk-5 trnk (UUU) (1), 5 rps16- trnq (UUG) (2), rpob-trnc (GCA) (1), petm-psbm (1), psbm-trnd (GUC) (1), ycf3 intron1 (1), ycf3-trns (GGA) (1), trnt (UGU) -trnl (UAA) (2), ndhc-trnv (UAC) (1), rpl33-rps18 (1), petb intron (1), and rpl14- rpl16 (2). We also had to remove species pairs because they were too variable to be confidently aligned; that is, between these species pairs, GenBank BLAST opened up large gaps with scattered unaligned nucleotides, and we could not manually improve on these alignments. Species pairs in which this was the case include: Erodium carvifolium (NC_015083) and Erodium texanum (NC_014569); Cuscuta exaltata (NC_009963) and Cuscuta gronovii (NC_009765); Wollfia australiana (NC_015899) and Wollfiella lingulata (NC_015894); Colocasia esculenta (NC_016753) and Lemna minor (NC_010109); and Ranunculus macranthus (NC_008796) and Megaleranthus saniculifolia (NC_012615). In a few cases, one or two NC-cpDNAs had to be omitted from an otherwise neatly aligned genomic comparison because these NC-cpDNAs were too variable to be confidently aligned. This was the case in Nuphar/Nymphaea for petn-psbm, Phoenix/Pseudophoenix for trnc-petn and petn-psbm, Gossypium for psbe-petl, Silene for trnh (GUG) -psba and clpp-psbb, and Anthriscus /Daucus for ndhf-rpl32, rpl32-trnl (UAG), 5 trnk- 3 rps16, 5 rps16-trnq (UUG), and atph-atpi (Appendix S1). The most informative noncoding cpdna regions Figure 2A shows the number of genetic differences observed within every single-copy NC-cpDNA and matk, averaged across 25 species pairs. In this analysis, the top 10 regions were: 5 rps16-trnq (UUG), ndhc-trnv (UAC), ndhf-rpl32, trnt (GGU) -psbd, psbe-petl, petapsbj, rpl32-trnl (UAG), rpl16 intron, ndha intron, and rpobtrnc (GCA) (underlines highlight the regions in common between this analysis and the analysis of average percentage contribution to total PICs, below). The matk gene was ranked 12th. Figure 2B summarizes the PIC data where the value for each NC-cpDNA represents the average percentage contribution of the total genetic differences observed in pairwise species comparisons averaged across the 25 lineages. In this analysis, the top 10 regions were: ndhf-rpl32, ndhc-trnv (UAC), rpl16 intron, rpl32-trnl (UAG), 5 rps16-trnq (UUG), 5 trnk (UUU) -3 rps16, psbe-petl, trnt (GGU)- psbd, peta-psbj, and psbm-trnd (GUC) (underlines highlight the regions in common between this analysis and the analysis of overall average PICs, above). The matk gene was again ranked 12th. While the relative positions of the most variable regions may change between the averaged PIC and percentage contribution of averaged PIC analyses (or even lineage to lineage, see below), there are eight NC-cpDNAs in the top 10 of both of these analyses and these are underlined above and throughout the rest of this paper. These eight regions account for roughly 21% of the PICs in a given lineage (based on percentage contribution of each region averaged across the 25 species pairs). The top 12 regions account for >33% of the total PICs. Percentage variability within each region is shown in Fig. 2C. Eight of the top 11 most variable regions in this analysis had an average total length of about 250 bp and 9/11 were 350 bp, making the total PICs offered by these regions relatively low even though their percentage variability was high (compare Fig. 2C with Fig. 2A and 2B ). The rpl32-trnl (UAG) and rps15-ycf1 regions were the only two of the top regions that were over 500 bp and only rpl32-trnl (UAG) was over 750 bp. Because choosing the most informative NC-cpDNAs requires combining the effects of NCcpDNA length and percentage variability, Fig. 2D is positioned next to Fig. 2C to highlight the fact that several NC-cpDNAs that are highly variable by percentage are also very short. Because we had more than one species pair within most major lineages, we compared the trends in the major clades of angiosperms with the overall trends described above ( Fig. 3A F ). While the ranking of the top regions is not perfectly consistent across the major lineages, there are some NC-cpDNAs that consistently rank among the top 13 (top 13 shown to incorporate matk because it has been a popular and informative marker since the beginning of sequencing cpdna). Either ndhf-rpl32 or rpl32-trnl (UAG) was the highest ranked region in four of six major lineages, in one lineage rpl32-trnl (UAG) was ranked second behind 5 rps16-trnq (UUG ), and both regions were too variable to be included in one species comparison ( Anthriscus/Daucus ). The 5 rps16-trnq (UUG) region ranked in the top two in two of six major lineages and in the top 13 in four of six lineages, but was excluded from one lineage for being too variable ( Daucus/Anthriscus ). It should be noted that eurosid I was the only major lineage in which neither ndhf-rpl32 nor rpl32- trnl (UAG) ranked in the top 2; however, these NC-cpDNA regions are missing from Populus and unusually small in Vigna, perhaps skewing the data because these two lineages account for two of four species pairs of the eurosid I group. The rpl32-trnl (UAG) region did rank in the top five in the other eurosid I, Cucumis. Another high-ranking region, ndhc-trnv (UAC), ranked in the top five in four of six of the major lineages. Analysis of the top eight most variable regions in the 25 individual species pair comparisons (Fig. 4 ) shows that ndhctrnv (UAC), rpl32- trnl (UAG), and ndhf-rpl32 all ranked in the top 10 in 16 of 25 lineages, psbe-petl in 15 of 25 lineages (and it was too variable to be included in Gossypium ), trnt (GGU)-psbD in 14 of 25 lineages, 5 rps16-trnq (UUG) and peta-psbj in 11 of 25 lineages, and rpl16 intron in 10 of 25 lineages. Adjusted mean PICs Results of the mixed-model AN- COVA (which accounted for variation across lineages and controlled for the effect of region length) indicate that the mean log-transformed number of PICs/region differed across the 10 NC-cpDNAs that ranked the highest in the overall percentage contribution analysis ( F 9,194 = 3.09, P = ). Following the ANCOVA, we performed pairwise contrasts of adjusted means and/or groups of means using Scheffé s procedure; selected comparisons involving combined DNA regions (combined vs. combined; single vs. combined) revealed significant differences in the mean PICs/region between certain groups of the 10 regions. Scheffé s contrasts revealed significant differences in adjusted mean PICs/region among three groups of combined DNA regions ( P < 0.05; Fig. 5 ). Mean PICs/region (compared as groups based on visual breaks in the data) was highest for Group 1 (geometric mean = 19.2 PICs; regions ndhf-rpl32, rpl32-trnl (UAG) ), intermediate for Group 2 (12.7 PICs; regions ndhc-trnv (UAC), rpl16 intron, peta-psbj ), and lowest for Group 3 (10.2 PICs; regions 5 rps16-trnq (UUG), psbe-petl, trnt (GGU) - psbd, ndha intron, rpob-trnc (GCA) ). Pairwise contrasts within the three groups yielded no differences in mean PICs/region ( P > 0.05).

6 1992 AMERICAN JOURNAL OF BOTANY [Vol. 101 Fig. 2. Expected number of genetic differences within the noncoding single-copy portions of the plastome (NC-cpDNA). Gene order is preserved beginning at the top with the intersection of inverted repeat b (IRb) and the large single-copy (LSC) region; the small single-copy (SSC) region is at the bottom and begins with rps15-ycf1. The vertical black lines and the four-point stars highlight the 10 regions that on average contain the greatest number of potentially informative characters (PICs) in those analyses (A, B). (A) Number of PICs averaged across the 25 species pairs. (B) Average percentage of potentially informative characters (PICs) that each noncoding region contributes to the total PICs observed between species pairs. (C) Percentage variability averaged across 25 species pairs. The top 11 most variable regions by percentage are cutoff from the rest by the vertical black line and the four-point stars (there was a tie for 10th place). Arrows from NC-cpDNAs in (C) point to the same regions in (D) to highlight the fact that 8/11 of the most variable regions in terms of percentage variability are too short to be of great value. (D) Length of noncoding regions averaged across 25 species pairs. The black line in (D) marks 250 bp to highlight those regions that may be potentially highly variable in terms of percentage but are less than 250 bp long.

7 November 2014] SHAW ET AL. TORTOISE AND HARE IV 1993 Fig. 2. Continued.

8 1994 AMERICAN JOURNAL OF BOTANY [Vol. 101 Fig. 3. Clade-specific ranks of single-copy noncoding NC-cpDNAs (top 13). Percentage contribution of total potentially informative characters (PICs) ( y -axis) for each NC-cpDNA ( x -axis) averaged across the species pairs in the clade. The eight highest-ranking regions in the analysis of all 25 lineages combined are marked with four-pointed stars ( ndhf-rpl32, rpl32-trnl, rps16-trnq, ndhc-trnv, trnt-psbd, psbe-petl, peta-psbj, rpl16 intron ). In another contrast (excluding Group 1), the mean PICs/region for the highest-ranking Group 2 region (geometric mean = 14.7 PICs; region ndhc-trnv (UAC ) was higher than that for other regions in Groups 2 and 3 (10.6 PICs; regions rpl16 intron, petapsbj, 5 rps16-trnq (UUG), psbe-petl, trnt (GGU) -psbd, ndha intron, rpob-trnc (GCA) ; P < 0.05). In summary, the ANCOVA and multiple contrasts of adjusted mean PICs/region within the top 10 regions (averaged across lineages) indicated that ndhfrpl32, rpl32-trnl (UAG), and ndhc-trnv (UAC) were the most variable frames, respectively.

9 November 2014] SHAW ET AL. TORTOISE AND HARE IV 1995 Literature review The total number of phylogeography or phylogeographic papers published between 1987 and 2013 was We refined this number by limiting the selections to articles only (9286) and refining by the search term *aceae to capture any mention of a plant family in the text, resulting in a total of 1089 papers. Further refining by the search terms chloroplast or plastid or cpdna resulted in 699 papers; refining by mitochondria or mitochondrial or mtdna resulted in 351 papers; refining by nuclear or ndna resulted in 253 papers ( Fig. 6 ). For the literature review, 178 papers published between 2007 and 2013 were reviewed for this study. Of those 178, 123 (69%) used cpdna sequence data ( Fig. 7 ). The percentage of papers using cpdna sequences varied by year, but ranged between 50% (2007) and 81% (2012). The average number of cpdna regions used was two to three, with a minimum of one to a maximum of six. Only 28 of the 178 papers explicitly reported screening additional loci, although it was not always intimated what loci were tested. When this information was provided, we determined that the mean number of loci screened by year ranged from four to 15, with a few papers indicating several regions were screened. Comparing the noncoding regions within studies showed that in 11 comparisons, trnh (GUG) -psba provided more PICs than other regions in the same study. However, in six other studies, trnh (GUG) -psba was not the best performer. The region with the second highest observed PICs was trns (GCU) -5 trng (GCC), based on nine comparisons. There were only two studies (in which it was included) that it was not the top performer. Of the top regions reported in the present study, two ( ndhc-trnv (UAC) and psbm-trnd (GUC) were not used in any of the studies reviewed for this comparison. The remaining top regions identified here were used infrequently in the surveyed papers: rpl32-trnl (UAG) (best in three of four studies); ndhf-rpl32 (best in the one study in which it was used); psbe-petl (best in the one study in which it was used); trnt (UGU) -trnl (UAA) (best in one of two studies); trnt (GGU)-psbD, rpob-trnc (GCA), and peta-psbj were not best in any of the studies in which they were used (2, 1, and 2 studies, respectively; Appendix S2, see online Supplemental Data). DISCUSSION NC-cpDNAs are important for low-level studies Today, data from NC-cpDNA is the most commonly used tool for phylogeographic and low-level phylogenetic studies of plants ( Fig. 6 ). Our literature review highlighted the fact that researchers working at inter- and intraspecific levels still rely heavily on NC-cpDNA sequence data from one to a few loci ( Figs. 6, 7 ). Of the more recently published plant phylogeography papers, those published between 2007 and 2013, nearly 70% used cp- DNA sequence data ( Fig. 7 ). There is a body of literature dating back to the 1990s in which botanists have been searching for the most appropriate NC-cpDNAs for low-level phylogenetic and phylogeographic studies ( Taberlet et al., 1991 ; Demesure et al., 1995 ; Dumolin- Lapegue et al., 1997b ; Small et al., 1998 ; Hamilton, 1999 ; Saltonstall, 2001 ; Shaw et al., 2005, 2007 ). In the mid-2000s, a few studies compared numerous NC-cpDNA regions across a broad enough taxonomic scope such that general trends about NC-cpDNA region relative variability could be reported ( Small et al., 1998 ; Bastia et al., 2001 ; Grivet et al., 2001 ; Aoki et al., 2003 ; Dhingra and Folta, 2005 ; Shaw et al., 2005, 2007 ; Ebert and Peakall, 2009 ). Reports of these general trends may have allowed workers to begin utilizing the most informative NC-cpDNA regions and push the use of the plastome further toward the shallowest taxonomic levels. The most informative noncoding cpdna regions A few NC-cpDNA regions stand out as being the most likely to contain high levels of variability at the inter- and intraspecific levels, even though this study ( Figs. 2 4 ) and a body of literature ( Mort et al., 2007 ; Shaw et al., 2007 ; Särkinen and George, 2013 ) reinforces the observation that there is no universally best NC-cpDNA. That said, screening what are known to be, on average, the fastest evolving regions is a good starting point for molecular investigations of new groups ( Cires et al., 2012 ; Fior et al., 2013 ). In the best-case scenario, researchers would have, or could generate, at least two complete plastomes within their study genus (or study section of large genera) to screen for the most informative regions, before beginning a phylogeographic or low-level phylogenetic study ( Doorduin et al., 2011 ; Fehrmann et al., 2012 ; Li et al., 2013 ; Särkinen and George, 2013 ). Since there may be as many as genera of angiosperms ( org ) and GenBank presently (August 2014) has only 374 angiosperms plastomes representing 205 genera and a fair number of these are either species poor, endemic, or genera of parasitic plants and not likely to be useful predictors of variability across large numbers of species it is unlikely that most researchers will have this advantage for several more years. While it is true that the cost of NGS approaches are rapidly declining, and novel and simpler methods for undertaking such work are being rapidly published, many researchers (particularly at primarily undergraduate institutions) are still limited by funding and computational capacity. For these researchers, as well as others, there still remains a trade-off between collecting big data for a small number of taxa or individuals at the same cost of collecting small data for a larger number of taxa or individuals. The cost of software (e.g., Geneious) may be inexpensive to some, yet to other researchers this cost alone may be prohibitive. Additionally, for phylogeographic studies, software has not yet fully developed to be able to handle the computational loads associated with NGS big data. Therefore, the choice between NGS and Sanger approaches, while not necessarily mutually exclusive ( Fior et al., 2013 ), is going to be an independent one based on funding (both short-term and long-term potential), laboratory infrastructure, and computational support in a given laboratory. As such, there is still a valid need for information such as that presented here, providing insight into which NC-cpDNA regions or portions of the plastome are likely to be most informative (Fig. 8 ). That is, the results of our work can guide researchers who want to use the most variable portions of the plastome, whether they are amplified independently, in larger clusters, multiplexed, or extracted from whole plastome data sets. That said, three regions of the plastome stand out to us as hotspots of variability ( Fig. 8 ). First is the area in the SSC that runs from ccsa to ndhf. This portion of the plastome contains two of the most variable single regions ( ndhf-rpl32 and rpl32- trnl (UAG) ). The second hotspot, is a larger area of the genome from matk to 3 trng (GCC), which contains several fairly variable regions including matk, 5 trnk (UUU) -3 rps16, 5 rps16- trnq (UUG), and trns (GCU) -5 trng (GCC), among others. The third hotspot is from rpob to psbd, and this larger portion of the genome contains several high-ranking regions all clustered together. Interestingly, there are a fair number of rearrangements in these regions of the plastome as well (Appendix S1). Based on the extensive plastome survey included in the present study,

10 1996 AMERICAN JOURNAL OF BOTANY [Vol. 101 Fig. 4. Potentially informative characters ( y -axis) for each NC-cpDNA ( x -axis) within each species pair. The eight highest-ranking regions in the analysis of all 25 lineages combined are marked with four-pointed stars ( ndhf-rpl32, rpl32-trnl, rps16-trnq, ndhc-trnv, trnt-psbd, psbe-petl, peta-psbj, rpl16 intron ). Mag. = Magnolia, Cymb. = Cymbidium, Phalae. = Phalaenopsis, Aeth. = Aethionema, Sol. = Solanum, Chrys. = Chrysanthemum.

11 November 2014] SHAW ET AL. TORTOISE AND HARE IV 1997 Fig. 4. we suggest that is it possible to make good predictions based on either the overall trends or based on the trends observed within a clade of interest ( Figs. 2 4 ). Continued. We used 25 closely related species pairs (most were congeners) from 10 major lineages of angiosperms, with multiple species pairs from six of these clades, to test all NC-cpDNA regions

12 1998 AMERICAN JOURNAL OF BOTANY [Vol. 101 Fig. 5. Geometric mean potentially informative characters (PICs) ± SE per region for the top 10 NC-cpDNA regions ( n = lineages per region). Regression-based estimates [ y = log 10 (total PICs per region +1), x = log 10 (length of region + 1)] were calculated for a common x value. Regions are coded 1 to 10 (highest to lowest) and arranged in decreasing order of mean values. Mean PICs/region are reported for three statistical groups; group means followed by different letters are significantly different ( P < 0.05; ANCOVA, Scheffé contrasts). plus matk. Our data predicts that the regions most likely to contain the greatest number of variable sites in angiosperms are 5 rps16-trnq (UUG), ndhc-trnv (UAC), ndhf-rpl32, trnt (GGU) -psbd, psbe-petl, peta-psbj, rpl32-trnl (UAG), rpl16 intron, ndha intron, rpob-trnc (GCA), trns (GCU) -5 trng (GCC), 5 trnk (UUU) -3 rps16, psbmtrnd (GUC), and matk ( Fig. 2A, B ). Primers have been developed and are universal for most of these regions (see Shaw et al., 2007 ). However, the coding regions of ndhf-rpl32 and rpl32-trnl (UAG) Fig. 6. Intraspecific studies of plants from 1987 to 2013 showing the contributions of the separate plant genomes. We performed a search in Web of Science (May 2014) to determine which plant genome(s) was most frequently used in the phylogeographic literature. Search terms included phylogeography or phylogeographic refined by articles only and the term *aceae to identify studies including a plant family, with subsequent independent refinements using the terms chloroplast or plastid or cpdna or mitochondria or mitochondrial or mtdna or nuclear or ndna. Fig. 7. Summary of cpdna sequence used in 178 plant phylogeographic studies published between 2007 and Of 178 studies surveyed over 7 years, 69% used cpdna sequence data. This percentage varied by year, ranging between 50% (2007) and 81% (2012), and it has not dropped below 68% since (specifically, ndhf and rpl32 ) are highly variable, making universal primer design difficult for these two NC-cpDNAs. Appendix S3 contains a brief discussion on the ndhf-rpl32 and rpl32-trnl (UAG) primers and a table that will aid in lineagespecific primer design for future projects. The most variable NC-cpDNA regions identified in this research agree with the previous findings by Shaw et al. (2007) and others ( Byrne and Hankinson, 2012 ; Dong et al., 2012 ). It is interesting that these studies are highly corroborative regarding the best NC-cpDNAs for low-level studies, even though they sampled completely different species from many different plant families. That is, none of the 25 species pairs sampled here were the same as any species pair from TH2 or TH3. Both the present study and TH3 converge on ndhf-rpl32, rpl32-trnl (UAG), 5 rps16-trnq (UUG), ndhc-trnv (UAC), psbj-peta, trnt (GGU) -psbd, and psbe-psbl being among the most informative regions of the plastome for most groups. Even analyzed in a different way and with different species, Dong et al. (2012) and Byrne and Hankinson (2012) showed rpl32-trnl (UAG), 5 rps16-trnq (UUG), ndhc-trnv (UAC), trnt (GGU) -psbd, and rpl16 intron to be among the most variable. Additionally, new comparative plastomic data continue to be consistent with our findings at multiple taxonomic levels, including Illicium (O. R. Leonard and A. B. Morris, unpublished data), Solanaceae ( Särkinen and George, 2013 ), and Apiales ( Downie and Jansen, in press ). Finally, numerous other studies with narrower taxonomic scopes have also shown these same regions to be the most variable of the angiosperm plastome (see discussion, below). Like TH3, we show that when making finer level comparisons, there is a lesser degree of predictability among the top NCcpDNAs. For example, trns (GCU) -5 trng (GCC) ranked just outside of the top 10 regions in the overall comparison, and it ranked in the top 12 in 3/6 of the comparisons of major lineages, but it ranked first place in the Phalaenopsis (monocot) and Populus (eurosid I) comparisons and third place in Olea (euasterid I) ( Fig. 4 ). Interestingly,

13 November 2014] SHAW ET AL. TORTOISE AND HARE IV 1999 Fig. 8. Map of the potentially informative characters (PICs) expected to be found around the plastome. Gene content, order, and mapping, was based on the map of Lee et al. (2006) for Gossypium hirsutum. Bars radiating in from NC-cpDNA regions show the average percentage contribution to total PICs for each NC-cpDNA and matk. The blue, orange, red, and pink circles indicate increasing predicted PIC values for each NC-cpDNA, as demonstrated by the scale bar. The inverted repeats are indicated by bold black lines on the genome and on the blue, orange, red, and pink circles. Note that three areas highlighted in light green standout as hotspots or clusters of highly variable regions of the plastome: ccsa-ndhf, matk-3 trng, and rpob-psbd. both Phalaenopsis and Populus have plastome rearrangements that result in the partial loss of the ndhf-rpl32-trnl (UAG) region (actually, further supporting our data that this is a highly variable region). In another example, matk, which ranked 12th place overall ( Fig. 2A, 2B ) and in the top 12 positions in 3/6 comparisons of major lineages ( Fig. 3B D ), was ranked first in Aethionema