All samples were collected from shoots of Calligonum individuals from Xinjiang, Qinghai, Inner Mongolia, Gansu, and Ningxia across the northwest of China during summer from 2006 to 2015 (Table 1).

The classical identification key was used to differentiate these species mainly based on fruit characteristics and geographic locations, and the C. mongolicum complex has been identified by its fruit characteristics (Mao 1992; Bao and Grabovskaya-Borodina 2003), primarily based on quantifiable differences in fruit and bristle size, such as fruit length (LF), fruit width (WF), bristle length (BS), bristle distance (BD), rib distance (RD), achene length (AL), achene width (AW), and fruit shape (FF) (Shi et al. 2012, 2016; Fig. 2A). The same fruit indices have been used to compare C. jeminaicum with C. mongolicum. The flower traits for differentiating between the two species were selected based on the identification key in “Flora of China” (Bao and Grabovskaya-Borodina 2003), including the shape of perianth segments (PS, broadly elliptic or ovate Fig. 2B), pedicel length (1–2 cm in C. jeminaicum and 2–4 cm in C. mongolicum: Fig. 2C), spreading or reflexed in fruit (PSF, Fig. 2D), and pedicel joint position (below or middle). The shape of perianth segments (Fig. 2B) and pedicel length (Fig. 2C) were used to make quantitative distinctions between C. jeminaicum and C. mongolicum.

Figure 2. 

Measurements of fruit characters (A) and flower traits (B form of perianth segments C pedicel), and the distinction of the form of perianth segments in Calligonum jeminaicum and C. mongolicum fruits (D).

Some species with distinctive fruit characters were used as references in the DNA data analysis: Calligonum calliphysa Bunge, which was previously named Calligonum junceum (Fisch. & C. A. Mey.) Litv. (Bao and Grabovskaya-Borodina 2003), is the only species in sect. Calliphysa, was selected as a representative species; Calligonum arborescens Litv. and Calligonum ebinuricum Ivanova ex Y. D. Soskov (sect. Medusa) were used for comparison because they are regarded as distinct from the C. mongolicum complex. The number of individuals used for morphological analysis and DNA extraction in each population and the accession numbers of some ITS and plastid marker sequences obtained from GenBank are given in Table 1.

For all the newly collected samples, total genomic DNA was extracted from fresh or silica gel dried leaves according to the protocol of Doyle and Doyle (1990) or the CTAB method of Doyle and Doyle (1990). The ribosomal DNA regions are known to be potentially problematic when inferring phylogeny (Alvarez and Wendel 2003). In this study, we followed the guidelines for obtaining reliable ITS sequences in plants proposed by Feliner and Rossello (2007). The ITS regions were amplified and sequenced using the previously described primers “ITS5a” and “ITS4” (Stanford et al. 2000). The ETS primers were designed by Shi et al. (2016): the forward primer ETScalli1: 5'-GTTACTTACACTCCCCACAACCCC-3' and the reverse primer 18SIGS: 5'-GAGACAAGCATATGACTACTGGCAGGATCAACCAG-3'. Primers and polymerase chain reaction (PCR) protocols used for the amplification of chloroplast psbA-trnH, ycf6-psbM, rpl32-trnL, trnL-F, and rbcL (the first part of the entire rbcL gene) were described in previous studies (Demesure et al. 1995; Small et al. 1998; Shaw et al. 2005, 2007; Falchi et al. 2009).

The specific Sanger sequencing studies of the Calligonum mongolicum complex and other species were divided into two parts, with most experiments completed at the Smithsonian Institution in 2014, and additional data, particularly those concerning C. jeminaicum, being supplied by the Key Laboratory of Biogeography and Bioresource in Arid Land (KLBB), Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences. At the Smithsonian Institution, PCR amplification of DNA was performed using 10 ng of genomic DNA, 4 pmol of each primer, 0.5 U Taq polymerase (Bioline, Randolph, MA, USA), and 2.5 mM MgCl₂ in a volume of 25 µL using a PTC-225 Peltier thermal cycler. The PCR cycling parameters were as follows: a 95 °C initial hot start for 5 min, 32 cycles of 94 °C for 30 s, primer-specific annealing (ITS and ETS: 55 °C for 60 s; the five cpDNA primers: 53 °C for 40 s), and 72 °C for 60 s, and a final extension of 72 °C for 10 min. At the Smithsonian Institute, the PCR products were isolated and purified using ExoSAP-IT (US Biological, Swampscott, MA, USA) and sequenced in both directions using the PCR primers. Cycle sequencing was carried out using an ABI Prism Big Dye Terminator Cycle Sequencing Ready Reaction kit (Applied Biosystems, Foster City, CA, USA) with 5 ng of each primer, 1.5 µL of sequencing dilution buffer, and 1 µL of cycle sequencing mix in a 10 µL reaction volume. Cycle sequencing conditions comprised 30 cycles of 30 s denaturation (96 °C), 30 s annealing (50 °C), and 4 min elongation (60 °C). The sequencing products were separated on an ABI 3730xl DNA analyzer (Applied Biosystems, Foster City, CA, USA). At KLBB, the amplified products were purified using a PCR Product Purification Kit (Shanghai SBS, Biotech Ltd., China). Sequencing reactions were conducted with the forward and reverse PCR primers using the DYEnamic ET Terminator Cycle Sequencing Kit (Amersham Biosciences, Little Chalfont, Buckinghamshire, U.K.) with an ABI PRISM 3730 automatic DNA sequencer (Shanghai Sangon Biological Engineering Technology & Services Co., Ltd., Shanghai, China). Both strands of the DNA were sequenced with overlapping regions to ensure that each base was unambiguous. Electropherograms were assembled and consensus sequences were generated with Sequencher 4.5 (Gene Codes, Ann Arbor, MI, USA).

Multiple sequence alignments were performed using MUSCLE in the Geneious v.10.0.6 platform (Kearse et al. 2012) using the default settings and manual adjustments. The phylogenetic tree reconstruction of the nrITS and ETS sequence alignment included 44 accessions: 35 newly generated nrITS sequences, 24 new ETS sequences, and nine ITS and 20 ETS sequences from GenBank (Table 1).

Phylogenetic analyses were conducted on both the nuclear and combined plastid datasets. The best-fit nucleotide substitution models for the ITS1, 5.8S, ITS2, ETS, psbA-trnH, ycf6-psbM, rpl32-trnL, trnL-F, and rbcL regions were determined separately using jModelTest (Darriba et al. 2012) and the Akaike information criterion (AIC) were used to rank the best-fit model for the Bayesian analyses.

Phylogenetic relationships were inferred using Bayesian inference (BI) as implemented in MrBayes v.3.2.5 (Ronquist and Huelsenbeck 2003) and the maximum likelihood (ML) analyses were accomplished with RAxML v.8.2 (Stamatakis 2014). Partitioned analyses of both the nuclear and plastid datasets were implemented by applying the previously determined models to each data partition (Brown and Lemmon 2007). The nuclear ITS dataset was partitioned into ITS1, 5.8S, and ITS2 partitions. For the concatenated plastid dataset, separate partitions were used for the psbA-trnH, ycf6-psbM, rpl32-trnL, trnL-F, and rbcL regions. 51 samples in Calligonum were selected as the ingroup and Pteroxygonum giraldii Dammer & Diels was selected as the outgroup. Two independent BI analyses with one cold and three incrementally heated Markov chain Monte Carlo (MCMC) chains were run for 10,000,000 generations, with trees sampled every 1,000 generations. All Bayesian analyses produced split frequencies of less than 0.01, indicating convergence between the paired runs. The first 2,500 trees were discarded as burn-in, and the remaining trees were used to construct a 50% majority-rule consensus tree and posterior probabilities (PP). In the ML analyses, rapid bootstrap analysis was performed with a random seed, 1,000 alternative runs, and the same partition scheme as was used in the Bayesian analysis. The model parameters for each partition of the dataset were optimized by RAxML with the GTRCAT command. Trees were visualized in FigTree v1.4.3 (http://tree.bio.ed.ac.uk/software/figtree/). The ML bootstrap support values (BS) were labeled on the corresponding branches of the BI trees.

A network analysis was carried out with SplitsTree 4.13.1 (Huson and Bryant 2006) using the uncorrected p-distances between the C. mongolicum complex and C. jeminaicum species from the Bayesian analyses. Branch support was estimated using bootstrapping with 1,000 replicates (Felsenstein 1985).

The descriptions of the shape of perianth segments in fruit (PSF) and the pedicel joint position (below or middle) used to distinguish between the two species were qualitatively compared. The shape of perianth segments in fruit differs between the two species: spreading in the fruit of C. mongolicum, but reflexed in that of C. jeminaicum (Fig. 2D).

The morphological differences between C. mongolicum and C. jeminaicum focus primarily on their fruit and flower characteristics. Compared with the ambiguous characters in C. mongolicum, these taxonomical characters of C. jeminaicum were clearer and more stable. Quantitative comparisons of the fruit traits (Fig. 2A), the perianth segment shape (broadly elliptic or ovate, identified by the value of the length of the perianth segments/width of the perianth segments: Fig. 2B), and the pedicel length (Fig. 2C) were made between the two species (Fig. 3). Although some fruit characters appeared simultaneously in the two species and led to difficulty in distinguishing C. jeminaicum from C. mongolicum, the shape of perianth segments in fruit could be regarded as an effective character for their identification (Fig. 2D).

The quantifiable morphological characters in both fruits and flowers were compared between the two species. The fruit of C. mongolicum (0.106–1.880 cm; 1.134 ± 0.284 cm) was significantly (P = 0.026) longer than that of C. jeminaicum (0.415–0.649 cm; 0.432 ± 0.44 mm). Additionally, the fruit width (FW) for C. mongolicum (0.226–1.742 cm; 0.923 ± 0.347 cm) was larger than that of C. jeminaicum (0.348–0.508 cm; 0.428 ± 0.113 cm; P = 0.017). The bristle length of C. jeminaicum (0.372 ± 0.020 cm) was significantly shorter (P = 0.06) than that of C. mongolicum (0.312 ± 0.121 cm), and the bristle distance (0.077 ± 0.006 cm) and rib distance (0.087 ± 0.004 cm) of C. jeminaicum were significantly smaller than those of C. mongolicum (bristle distance 0.131 ± 0.032 cm, P = 0.01; rib distance 0.105 ± 0.032 cm, P = 0.02). Significant differences were also detected in achene length (0.823 ± 0.146 cm in C. mongolicum and 0.195 ± 0.105 cm in C. jeminaicum, P = 0.00) and achene width (0.359 ± 0.089 cm in C. mongolicum and 0.333 ± 0.004 cm in C. jeminaicum, P = 0.00) (Fig. 3), although the difference in achene width was small. The fruit shape, as the key character, was substantially different between the two species (P = 0.000), with the subglobose fruit of C. jeminaicum (1.048 ± 0.467 cm/cm) being much more rounded than the broadly ellipsoid fruit of C. mongolicum (1.357 ± 0.442 cm/cm). Thus, the fruit characteristics could be used to distinguish between the two species (Fig. 3). Both the pedicel length (P = 0.00) and the form of perianth segments (P = 0.01) of the two species showed significant differences. The pedicel length of C. jeminaicum (0.313 ± 0.004 cm) was much longer than that of C. mongolicum (0.219 ± 0.03 cm). The shape of perianth segments for C. jeminaicum (1.222 ± 0.167 cm/cm) was broader than that of C. mongolicum (2.544 ± 1.799 cm/cm) (Fig. 3).

Figure 3. 

Quantitative comparisons of fruit and flower characters in Calligonum mongolicum and C. jeminaicum.

The aligned matrix of 44 accessions of the combined nrITS and ETS sequences comprised 807 bp that did not include any abnormal SNPs or unreasonable sequences according to the Phi test (P = 0.0321). The best-fit substitution models were GTR+G for ETS (nucleotide frequencies A: 0.200803 C: 0.329510 G: 0.295074 T: 0.174613) and GTR+I+G for nrITS (nucleotide frequencies A: 0.163227 C: 0.337699 G: 0.352720 T: 0.146353) based on the jModelTest (Darriba et al. 2012) results. The GTR+G model was selected for the ML analyses of the aligned matrix of nrDNA.

The two phylogenetic tree reconstruction methods, BI and ML, produced consistent topologies. However, the nuclear and the chloroplast data were analyzed separately to reconstruct the phylogenetic relationships among C. jeminaicum, the C. mongolicum complex, and other species in Calligonum because obviously different topologies based on the nuclear (Fig. 4, 5) and the chloroplast (Fig. 6) data were found. In the nrDNA data, no single nucleotide polymorphism (SNP) was identified among the C. jeminaicum samples, but the species from the C. mongolicum complex showed heterogeneity and did not form a single clade (Fig. 4). The populations of the C. mongolicum complex, C. arborescens, C. calliphysa, and C. jeminaicum, were distributed within the same broad geographic region. The three individuals of C. ebinuricum, which had specific fruit characteristics that were different from the C. mongolicum complex, formed an independent clade (Fig. 4). Interestingly, the p-distance among the Calligonum taxa for the ITS and ETS regions reached 11.364% between species C. arborescens and C. calliphysa. The p-distance was as high as 22.54% between C. ebinuricum and the C. mongolicum complex group, which reflects their interspecific differentiation. Consistent results were obtained in the ML analysis in the same phylogenetic tree for nrDNA, conforming the C. mongolicum complex and C. jeminaicum independently (Fig. 4, PP = 1, BS = 98%).

Figure 4. 

Bayesian inference tree of the concatenated nuclear ITS and ETS sequence data showing Calligonum jeminaicum and its congeners. Bayesian posterior probabilities and maximum likelihood bootstrap support values are given above the branches.

Figure 5. 

Neighbor-net analyses based on uncorrected p-distances of the nuclear ITS and ETS sequence data. Numbers indicate bootstrap values over 1,000 replicates.

The neighbor-net constructed for the C. mongolicum complex and C. jeminaicum using the ITS and ETS sequences (Fig. 5) also did not support a single clade for the C. mongolicum complex. The three C. jeminaicum samples formed a separate branch from other groups, which is distant from the entire C. mongolicum complex, with a bootstrap support value of 94.9%.

Independent phylogenetic trees were reconstructed based on the concatenated plastid dataset, including the psbA-trnH, ycf6-psbM, rpl32-trnL, trnL-F, and rbcL regions, using the BI and ML methods. The tree topologies of the BI and ML trees were identical, and only the BI tree is shown (Fig. 6). A new haplotype (X), which occurred in all the C. jeminaicum individuals, was identified in the combined cpDNA dataset. The distribution of the C. mongolicum complex within the cpDNA tree could be separated into five to six regions that appear to reflect their geographical distribution. The first branch included sequences from six populations of the C. mongolicum complex (3, 4, 5, 9, 10, and 11) that were distributed in the west and northeastern regions of the Tengger Desert, where C. arborescens and C. calliphysa occurred sympatrically with these six populations. The second independent branch included sequences from four populations (14, 15, 16 and 17) from the Taklimakan Desert. The third independent branch included sequences from three populations (6, 7, and 8) from the Gurbantunggut Desert in the east of Xinjiang. Populations 12 and 13 comprised C. mongolicum complex samples from the Qaidam Desert that were distributed sympatrically with C. ebinuricum. Population 1 was the most phylogenetically distant from other populations, perhaps owing to its geographic isolation in the extreme north of Inner Mongolia. However, the new haplotype X of C. jeminaicum was separated from the above-mentioned branches of the C. mongolicum complex with strong support (Fig. 6, PP = 1, BS = 100%). Meanwhile, the other reference species of Calligonum (C. ebinuricum, C. arborescens, and C. calliphysa) did not form their own separate branches, but were interspersed within branches of the C. mongolicum complex (Fig. 6).

Figure 6. 

Bayesian inference tree of the concatenated plastid DNA sequence data (psbA-trnH, ycf6-psbM, rpl32-trnL, rbcL, and trnL-F) showing Calligonum jeminaicum and its congeners. Bayesian posterior probabilities and maximum likelihood bootstrap support values are given above the branches.