Comparative analysis of complete plastid genome reveals powerful barcode regions for identifying wood of Dalbergia odorifera and D. tonkinensis (Leguminosae)

Dalbergia odorifera T. C. Chen (Leguminosae), a rare and endangered tree species endemic to Hainan Island of China, produces the most expensive and rarest wood in China. The wood characteristics of D. odorifera are remarkably similar to those of D. tonkinensis (a much less sought‐after species from Vietnam), and the DNA from wood is often highly degraded, making it very difficult to identify the two species using anatomical features or DNA barcoding based on regular DNA markers. To solve the confusion of identifying wood reliably from the two species, we built and analyzed the plastome library of 26 samples from 18 Dalbergia species, of which 12 samples from eight closely related species of D. odorifera are newly sequenced in this study. Phylogenomic analysis suggested that the relationships among the 26 samples are mostly well resolved, and conspecific individuals from different populations of D. odorifera and D. tonkinensis clustered together. Between the plastid genomes of the two species, we identified 129 indels and 114 single nucleotide polymorphisms. By assessing a subset of 20 nucleotide polymorphisms and 10 indels using 37 population‐level samples (20 samples of D. odorifera and 17 samples of D. tonkinensis), we recovered eight species‐specific barcode regions that could be suitable for identifying the wood D. odorifera and D. tonkinensis. To examine their utility in wood identification, we amplified the eight DNA barcodes using six wood samples and recovered an amplification success rate of 83.3%, demonstrating a reliable method for precise wood identification of the two species.


Introduction
Dalbergia L.f., a genus of Leguminosae (alternatively Fabaceae), comprises ca. 250 species of trees, shrubs, and lianas mainly in the pantropic region (Klitgård & Lavin, 2005;Vatanparast et al., 2013;Bolson et al., 2015;Li, 2017). Many species within this genus are of major economic importance for their high quality wood (Bhagwat et al., 2015), such as the richly hued and costly woods of the Brazilian rosewood D. nigra (Vell.) Allemão ex Benth., the Indian rosewood D. latifolia Roxb., and the are from Pterocarpus Jacq., two are Millettia Wight & Arn. species, and one is a species of Senna Miller Gard. Five species of Diospyros L. belonging to the family Ebenaceae have also been recognized as Hongmu. Dalbergia odorifera, a rare Hongmu species endemic to Hainan Island of southern China, produces excellent quality wood with characteristic aroma, which is the most expensive found in the Chinese wood market. Since the 10th century, the wood of D. odorifera has been used in furniture manufacturing, and during the Ming and Qing dynasties (ca. 1600-1700) it was widely used by the nobility and royal families of China (Xu & Li, 2003). Due to its smooth, beautiful grain, and the lasting pleasant aroma it produces, there is high demand for this wood, and as such products made from D. odorifera have become a symbol of wealth and a lavish display of status. The heartwood of D. odorifera has also been used as medicine for blood disorders, ischemia, swelling, and epigastric pain in China (Chan et al., 1998;Lee et al., 2014). However, the resources of D. odorifera are extremely limited because of restricted distribution and the fact that it takes more than 100 years to grow to maturity. This species is nearly extinct in the wild due to deforestation and illegal logging. It was last assessed in 1998 for the IUCN Red List of Threatened Species as of "Vulnerable" status (Ban, 1998), and was ranked in the second-class category of the National Protected Plants list, issued by the Chinese government in 1999 (Yu et al., 2017). Based on the comprehensive survey of wild populations of D. odorifera, Yang et al. (2012) assessed D. odorifera as "Critically Endangered" according to the IUCN endangered level assessment standard. Only seven individuals were found in their study. According to our follow-up survey in 2016, only three wild individuals of the species were found in the National Natural Reserve of Bawangling of Hainan Island, China.
The high-quality wood and the minimal distributions as well as survival of individuals of the species have resulted in the wood price of D. odorifera being $1000-1500 per kilogram (http://www.chinatimber.org/price/60290. html; 2019.01-2019.06). Consequently, this could cause traffickers to sell cheaper wood from similar species fraudulently (e.g., D. tonkinensis Prain, D. hainanensis Merr. & Chun, Pterocarpus erinaceus Poir, Streblus asper Lour., and Spirostachys africana Sond.) as D. odorifera (Cheng et al., 2014). Most of the wood from these counterfeit species can be distinguished from D. odorifera by congeneric wood anatomy. However, wood anatomy cannot distinguish the most closely related species, D. tonkinensis (northern to central Vietnam and Laos; Yu et al., 2016;Li, 2017), from D. odorifera. Recently, studies concerning wood identification of D. odorifera and D. tonkinensis have attracted the attention of several research projects in China (Guo et al., 2011;Wang et al., 2016;Ma et al., 2018;Zhang et al., 2018). Unfortunately, however, wood identification of these two species is still far from being resolved. The wood color, density, and structure of the two species are very similar Yu et al., 2016), making it impossible to identify the wood of the two species by anatomical means. Moreover, it is not applicable to discriminate between woods of the two species using chemical methods or isotopic fingerprinting, due to very similar chemical compounds of these congeneric species (Marco et al., 1994;Deguilloux et al., 2002;Dev et al., 2014).
The chemical compounds found in many plants could be unstable, depending on the age of the trees or the habitat (Deguilloux et al., 2002). Yang et al. (2016) examined the volatile and fat-soluble components from D. odorifera and D. tonkinensis, respectively, by gas chromatography-mass spectrometry and found the two species can be differentiated by fat-soluble components based on their composition. However, when including more samples of different ages and places of origin, these results proved less accurate (Zhang DY, 2018, unpublished data).
The difficulties in identifying wood of D. odorifera and D. tonkinensis using anatomical and chemical methods call for new and effective techniques in order to facilitate their identification. One means by which this could be accomplished is via DNA barcoding. For example, Ohyama et al. (2001) explored the identification of six species of Cyclobalanopsis Oerst. (Fagaceae) using the DNA barcoding approach and examined the feasibility of the method in identifying the woods of this genus. Such identification based on DNA barcoding was adopted by many researchers due to its utility and relative accuracy. Within Dalbergia, in particular, there have been multiple studies examining the efficacy of DNA barcoding. For example, Hartvig et al. (2015) sequenced rbcL, matK, and internal transcribed spacer (ITS) of 95 samples covering 31 Dalbergia species and tested their discriminatory ability with both distance-based and modelbased methods. They suggested rbcL + matK as the potential combined barcodes to identify woods within Dalbergia. Furthermore, Yu et al. (2017) sequenced eight DNA regions of nine endangered Dalbergia species and suggested ITS2 + trnH-psbA as the best combination barcode. These studies strongly suggested that DNA barcoding is robust in identifying distantly related species within Dalbergia. However, these DNA barcodes might not identify closely related species, such as the species pair D. odorifera and D. tonkinensis. Yu et al. (2016) advocated the intergenic spacer of trnH-psbA that contains seven single nucleotide polymorphism (SNP) sites as a barcode region for the identification of D. odorifera and D. tonkinensis. We carefully analyzed the trnH-psbA region of the two species and found that the seven SNP sites are locating in a 47-bp palindromic region, which is equal to a reverse complement of itself (Figs. 1A,1B,1D). This means that the alignment concerning the 47-bp palindromic region and the seven SNPs in Yu et al.'s (2016) study is not correct. Furthermore, the palindromic sequences can form stem-and-loop structures and are often unconserved among different individuals or even among different cells of the same plant (Cavalier, 1974). This is the case in our study because we detected both the forward strand and the reverse complementary strand of the 47-bp palindromic region in the raw reads of the same individual for both D. odorifera and D. tonkinensis (Fig. 1C). Additionally, previous studies only covered a few samples of D. odorifera and D. tonkinensis, which cannot answer whether the diagnostic sites in those studies are species-specific.
Genomic data have been used by many researchers to resolve the phylogenetic relationships of either deep or shallow recalcitrant nodes (Kumar et al., 2009;Nock et al., 2011;Yang et al., 2013;Song et al., 2015;Gamboa-Tuz et al., 2018;Wu et al., 2018;Zhao et al., 2018). It is useful for resolving relationships between closely related species. Liu et al. (2019aLiu et al. ( , 2019b developed 19 single sequence repeat markers for D. odorifera and D. tonkinensis based on transcriptome data, and they recovered genetic diversity and population structure of the species. These studies also showed the potential power of genomic data to differentiate D. odorifera from D. tonkinensis using fresh or silicon-dried plant tissues. In the case of identifying woods of D. odorifera and D. tonkinensis, however, the utility of nuclear genome data could be restricted because DNA from aged or processed wood was often severely degraded and the concentration could be too low to be enriched for sequencing. Genome data derived from the plastid could be a better choice for wood identification because most plant cells contain many more copies of the plastid genome than the nuclear genome (Fazekas et al., 2009). Song et al. (2019) identified eight mutation hotspot regions as candidate DNA barcodes by comparing the chloroplast genomes of nine species of Dalbergia. The length of these eight fragments ranges from 700 to 1500 bp. However, its feasibility in wood identification might not be guaranteed because it lacks of species-specific test and it is unlikely to obtain such long sequences from severely degraded wood products through Sanger sequencing. In general, shorter DNA fragments are more likely to be successfully amplified (Rachmayanti et al., 2009;Jiao et al., 2019).
An accurate identification of the two species is urgent, as currently D. odorifera is widely cultivated in southern China as a measure to upgrade the forestry production in the country. In Guangdong, Guangxi, and Hainan alone the total cultivation area could exceed 30 000 ha and new plantations are increasing rapidly (Xu DP, 2020, pers. comm.). The confusion in the germplasm sources could cause heavy loss to the farmers and forestry sections.
In this study, our main objectives were to: (i) screen potential barcoding regions for D. odorifera and D. tonkinensis through complete plastid genome comparative analysis; (ii) explore the most powerful and easily-amplified speciesspecific barcodes for the two species using population-level sampling; and (iii) verify the utility of these barcodes by testing their efficacy on wood samples.

Sampling, DNA extraction, and sequencing
For plastid genome sequencing, we sampled fresh leaves from two individuals each of Dalbergia odorifera, D. tonkinensis, and D. yunnanensis Franch, in addition to other eight Dalbergia species. For Sanger sequencing, we used silica gel-dried leaves of 20 and 17 samples of D. odorifera and D. tonkinensis, respectively (covering their distributional range in China and Vietnam). Six wood samples of D. odorifera and D. tonkinensis were either bought from a local seller in Hainan, China or through online shopping as a test case. The voucher information is presented in Table S1, and the specimens were deposited in the herbarium of South China Botanical Garden (IBSC).
Total DNA from leaf samples was extracted using the modified CTAB method (Doyle & Doyle, 1990). DNA from wood samples was extracted from 0.5-1 g wood powder using the Qiagen DNeasy Plant Mini Kit (Hilden, Germany), combined with N-phenacylthiazolium bromide .
Primers (Table S2) were designed in Primer3 version 4.1.0 (Koressaar & Remm, 2007;Untergasser et al., 2012). Due to the poor quality and severe degradation of wood DNA, we aimed at short DNA fragments with 100-250 bp in length for Sanger sequencing. The high-throughput sequencing of the plastid genomes was carried out using the Illumina Hiseq 2000 (San Diego, CA) at the Beijing Genomics Institute (Beijing, China). Sanger sequencing was performed using the ABI 3730 sequencer at the Shanghai Majorbio Bio-pharm Technology Co. (Shanghai, China).

Plastid genome assembly and annotation
The clean data of the Illumina sequencing received from Beijing Genomics Institute were directly assembled using the GetOrganelle pipeline (Bankevich et al., 2012;Langmead & Salzberg, 2012;Jin et al., 2018). Bandage version 5.6.0 (Wick et al., 2015) was used to visualize and manually correct the assembly results. We checked the sequence directions and verified the accuracy of the assemblies in Geneious version 9.0.5 (Kearse et al., 2012). The annotation of the chloroplast genomes was undertaken in Plastid Genome Annotator (Qu et al., 2019). Manual correction of start/stop codons and intron/exon boundaries was carried out in Geneious. To further verify the identified transfer RNA (tRNA) genes, the tRNAscan-SE version 1.21 program with default parameters was used to predict their corresponding structures (Schattner et al., 2005). All genome maps were drawn by OrganellarGenomeDRAW version 1.3.1 (Greiner et al., 2019). The annotated chloroplast genomes were deposited in GenBank (accession numbers see Table 1).

Phylogenetic analysis
We built a reference library using 14 plastid genome sequences generated in this study, in addition to 12 plastome sequences of Dalbergia and three outgroup plastome sequences of Pterocarpus downloaded from GenBank (Table  S1). The sequence of D. hainanensis was not included (MF926268, Deng et al., 2018), due to the possible misidentification of the sample (see also Song et al., 2019). Based on the reference library, we undertook sequence alignment using the MAFFT version 7.0.17 (Katoh & Standley, 2013) plugin in Geneious by concatenating 77 coding genes, 4 ribosomal RNA (rRNA) genes, and 26 Fig. 1. The 47-bp palindromic region between intergenic region trnH-psbA of the genus of Dalbergia. A, Seven high-lighting nucleotide mutation sites reveal that this 47-bp palindromic region (the region inside the red square) is unconserved among different species or even among different individuals of the same species. B, The forward sequences are identical with the reverse complement of the 47-bp palindromic regions. C, Two types of reads detected in the raw next generation sequencing data of the same sample (Dalbergia tonkinensis, voucher NZ1600). D, The palindromic region is capable of forming stem-andloop structures, which is lethal to the carrier and replicon (inviability), and unstable among different individuals or even among different cells of the same plant (instability) (Zuker, 2003). intergenic spacers. Glaring alignment errors were adjusted by hand. We inferred phylogenetic relationships for species within the reference library by constructing a maximum likelihood tree using IQ-TREE version 1.6.11 (Trifinopoulos et al., 2016). Branch supports were estimated using standard bootstraps with 100 iterations, and a minimum correlation coefficient of 0.99.
2.4 Species-specific population test and the utility of selected barcodes in wood product identification Based on the alignment of complete plastomes from the reference library, potential insertion-deletion (indel) and SNP sites, which were flanked by regions suitable for designing polymerase chain reaction (PCR) primers and were effective in the identification of D. odorifera and D. tonkinensis, were selected as barcode regions. The primers were designed using Primer3. We sequenced 37 population-level samples (20 samples of D. odorifera and 17 samples of D. tonkinensis) to test whether the selected SNP and indel sites were species-specific. The species-specific variable sites were considered to be powerful barcodes for identifying D. odorifera and D. tonkinensis. We built the maximum likelihood tree using IQ-TREE version 1.6.11 (Trifinopoulos et al., 2016) based on the concatenated alignment of the species-specific variable sites of the 37 population-level samples. Branch supports were estimated using standard bootstraps with 100 iterations, and a minimum correlation coefficient of 0.99. Additionally, we used six wood product samples of D. odorifera and D. tonkinensis to test the validity of identification of these species-specific variable sites. The sequences of the 37 population-level samples are provided in Doc. S1.

Features of the Dalbergia plastid genomes
The 14 complete plastid genome sequences newly generated in this study range from 155 823 to 156 314 bp in length. The total GC-content was 36.1% for all plastomes except D. cearensis Ducke (36%). The 14 plastid genomes each contained 77 coding genes, 30 tRNA genes, and 4 rRNA genes. In general, the plastid genomes of the Dalbergia species analyzed in this study were similar in terms of genome size, gene content, gene order, introns, intergenic spacers, and GC-content. A summary of the structure of the plastid genomes is shown in Table 1. The gene map is shown in Fig. 2. 3.2 Single nucleotide polymorphism and indel mutations between plastid genomes of D. odorifera and D. tonkinensis We detected 129 indels ranging from 1 to 22 bp in size (105 in intergenic regions, 22 in introns, and 2 in coding regions) among the four plastid genomes generated in this study of D. odorifera and D. tonkinensis, of which 66% were single base-pair. The longest indel was 22 bp in length, which was in the intergenic region of psbA-trnK-UUU. In the coding region of ndhF, we found an 8-bp insertion in the ndhF gene of D. tonkinensis (voucher: Lishijin 3616), an advanced stop codon, and a five-amino difference when compared to the homologous region of D. odorifera (vouchers: scbg1 and zhucj062). A 10-bp insertion in the ndhF gene of D. tonkinensis (voucher: Lishijin 3616) resulted in a delayed stop codon and a loss of five amino acids when compared to the same region of D. tonkinensis (voucher: scbg4).
We found 114 SNPs between D. odorifera and D. tonkinensis, with a transversion to transition ratio of 2.56:1. Of these, 69 SNPs (58 transversions and 11 transitions) were in the noncoding regions and 45 SNPs (24 transversions and 21 transitions) were from coding regions. We found 28 synonymous and 17 non-synonymous mutations in the coding regions between the two species.

Phylogenetic relationships and species-specific barcode sites
We built a reference library based on 26 plastid genomes of Dalbergia and three outgroup plastid genomes of Pterocarpus. The phylogenetic relationships among all species were well resolved (Fig. 3) in the maximum likelihood tree. The sister relationship between D. odorifera and D. tonkinensis was guaranteed in both the phylogenomic analysis of 26 Dalbergia samples at species level (Fig. 3) and the analysis based on eight barcoding regions using the populationlevel sampling (Fig. 4). A sister relationship between the clade of D. odorifera-D. tonkinensis and D. yunnanensis-D. vietnamensis P.H.Hô and Niyomdham is recovered with a bootstrap value of 100 (Fig. 3). Morphologically, the latter two species are most comparable to D. odorifera-D. tonkinensis in the genus. The intraspecific relationships within D. odorifera-D. tonkinensis were poorly resolved and polytomies were found for both species. Dalbergia sissoo Roxb. ex DC. newly sequenced in this study does not cluster with the one published by Song et al.  candidate DNA barcodes for identifying D. odorifera and D. tonkinensis. By evaluating the identification capability of these variations among the 37 population-level samples, we found that two indels (ndhK-ndhJ and trnG_UCC) and six SNPs (psbA, atpI, ndhA, ycf1-4, ycf1-5, and ycf1-7) have stable interspecific specificity and could robustly distinguish D. odorifera from D. tonkinensis (Fig. 5 shows different loci of the eight barcode fragments). Accordingly, we amplified the regions that contain these eight variable sites (100-250 bp in length) as DNA barcode markers to test the feasibility of amplification and sequencing for the six wood products of D. odorifera from D. tonkinensis. Five of the six wood product samples can be successfully amplified and identified, with a success rate of 83.3%.

Discussion
Combined indel and SNP DNA barcodes using comparative plastid genome analysis has proved to have significant potential for species identification in other taxa. For example, Huang et al. (2014) found 15 highly variable noncoding regions with more than 1.5% sequence divergence, which led to successful phylogenetic reconstruction and identification of species in the genus Camellia L. Similar work was carried out on the medicinal herb Panax L. (Dong et al., 2014), which revealed rps16, ycf1a, and ycf1b as the best mini-barcodes for identification of Panax ginseng C. A. Mey. and P. notoginseng (Burkill) F. H. Chen ex C. Y. Wu & K. M. Feng by comparing the complete plastid genomes of the two species. In this study, we analyzed 26 entire plastid genomes of Dalbergia and identified multiple variable sites that could be used as valuable plastid markers for the discrimination of the closely related species Dalbergia odorifera and D. tonkinensis.
Among the 129 indels found within the plastid genomes of D. odorifera and D. tonkinensis, 105 indels were found in intergenic spacers, 22 were from introns, and 2 in coding regions. Some highly variable regions, such as the intergenic spacers trnP_UGG-psaJ, rps11-rpl36, trnN_GUU-ycf1, ndhG-ndhI, psbA-trnK_UUU, trnV_UAC-ndhC, trnG_GCC-psbZ, trnG_UCC-trnS_UCC, contain multiple indel sites. Compared to gene regions that show biological functions, the intergenic spacers and gene introns appear to be more variable among and within species, as has been reported by many authors (Timme et al., 2007;Liu et al., 2017;Wu et al., 2018). In this study, we found the intergenic spacer ndhK-ndhJ and the intron region trnG_UCC are the best indel fragments to distinguish D. odorifera and D. tonkinensis. These two regions have also been shown to be very efficient at distinguishing between different, closely related species belonging to multiple taxonomic groups, including bacteria and species from several different plant families (Kaneko et al., 2000;Iwasaki et al., 2012;Panero et al., 2014). The mutations in these two regions are species-specific and conserved among  Song et al., 2020, voucher unavailable.) different individuals within D. odorifera and D. tonkinensis, as shown in our phylogenetic result using 37 population-level samples (Fig. 4).
The 114 nucleotide substitutions occurring in the plastid genomes of D. odorifera and D. tonkinensis suggest that such SNP events are much less common in this genus than in other genera such as Panax, Machilus Rumph. ex Nees and Citrus L. (Dong et al., 2014;Su et al., 2014;Song et al., 2015).
Several SNP sites were proved to be species-specific between D. odorifera and D. tonkinensis in this study. The promising gene, ycf1, which has shown great potential as a plastid DNA barcode for land plants (Cai et al., 2012;Dong et al., 2015), contained three species-specific SNP sites between D. odorifera and D. tonkinensis. In the atpI gene, we found one species-specific SNP which resulted in an amino acid change from valine into isoleucine between the two species. The psbA and ndhA genes were also highly effective in distinguishing D. odorifera from D. tonkinensis. Even though there is only one single mutation in each gene, they prove to be highly stable within species and with no interspecies overlap among the 37 population-level samples.
To find a single-barcoding gene for all plants is a challenging task due to the inherent inaccuracies of many barcode regions (Rubinoff et al., 2006). For our study, we combined two indel regions (ndhK-ndhJ and trnG_UCC) and six SNP regions (psbA, atpI, ndhA, ycf1-4, ycf1-5, and ycf1-7) to distinguish D. odorifera from D. tonkinensis robustly. Single-copy nuclear genes were excluded from the present study because it is particularly difficult to obtain nuclear DNA from timbers or wood products, from which DNA quality is very poor and the content is often extremely low (Fazekas et al., 2009).
The major obstacle for DNA barcoding of wood could be the difficulties in attaining high-quality DNA from commercial woods and wood products (Yu et al., 2017). Therefore, we have avoided amplifying long sequence regions and designed primers to amplify eight smaller regions in 100-250 bp for all the 37 population-level samples. From our results (Fig. 4), it is evident that these eight regions are highly discriminatory and that the mutations in these eight regions are species-specific as, in this plot, D. tonkinensis and D. odorifera accessions fell into distinctly separated groups. Furthermore, the eight barcode regions performed well in the wood product test as well. Five of the six wood samples were successfully amplified and identified, with a success rate of 83.3%.
To conclude, this study sequenced 14 complete plastid genomes for the species of Dalbergia using Illumina next generation sequencing and built a reference library of Dalbergia using 26 complete plastid genomes. Based on the reference library, 10 indels and 20 SNP sites that were flanked by regions suitable for designing PCR primers were selected to test whether they were species-specific among 37 population-level samples of D. odorifera and D. tonkinensis. Our work showed that two indel regions (ndhK-ndhJ and trnG_UCC) and six SNP regions (psbA, atpI, ndhA, ycf1-4, ycf1-5, and ycf1-7) are species-specific. We also found that these eight species-specific barcodes are effective at species identification even when using DNA extracted from wood products. Five of the six wood samples were successfully amplified and identified, with a success rate of 83.3%. The eight barcodes proved to be very successful in the identification of D. tonkinensis and D. odorifera. We recommend the use of these eight barcodes for species identification and wood product identification of these two species.
A significant breakthrough could be made for many existing wood identification problems through the rapid development of high-throughput sequencing technology. DNA extracted from wood is often serially degraded and fragmented, and analysis rendering PCR amplification from such samples is very difficult, which has been a pervasive problem in the identification of timber species. By sequencing herbarium specimens up to 80 years old from various flowering plant families, Zeng et al. (2018) reported that genome skimming could be used to generate genomic information using as little as 500 pg of degraded starting DNA. We can enrich the degraded DNA extracted from wood by using commercial kits despite the low concentration and poor quality of wood-derived DNA. These enriched DNA libraries can then be sequenced using a genome-skimming approach. Beyond D. odorifera, another 14 Hongmu species of Dalbergia meet the weak identification problem as well. Based on the reference library of 26 Dalbergia plastid genomes, we will extend current research by including more Dalbergia species collected from different areas throughout the distributional range of the genus.
Our results, based on multiple samples, clearly indicate that D. odorifera and D. tonkinensis represent two independent evolutionary lineages. In order to reveal their taxonomic relationship, it is necessary to study the differences in morphological characters between D. odorifera and D. tonkinensis in a large population scale. Whether the distinction between them is at the level of species or subspecies is worth discussing. However, it would be less wise to merge the two independent evolutionary lineages, whose wood products possesses huge differences in price, as one taxonomic entity, as the practice could cause damage to the forestry session, and bring about confusion in forming conservation strategies.