Plastome comparative genomics in maples resolves the infrageneric backbone relationships

Taxon sampling

Leaf samples of 16 species of maples were obtained from cultivated trees in private and state gardens of Europe and China (Table S1). The identity of all samples was verified with experts from these institutions. Herbarium vouchers were deposited in the BGT herbarium, Guangxi University, China.

Each of the 16 species selected belongs to a different section in the genus (sensu De Jong, 2004), and 13 are the type of their respective section. Section Hyptiocarpa was recently merged with section Rubra (Harris et al., 2017) and was not considered in this study. Acer oblongum, placed in section Pentaphylla series Trifida by De Jong (2004), was included as a representative of section Oblonga, which was recognized by Xu et al. (2008) in Flora of China. Only sections Spicata, Wardiana, and Macrophylla could not be sampled.

For the phylogenomic reconstruction, plastome sequences from six additional species of maples were obtained from GenBank (see accessions in Table S1). We also retrieved sequences of the two species of Dipteronia (the sister genus of Acer), Litchi chinensis J. F. Gmel (Sapindaceae) and Spondias mombin L. (Anacardiaceae), which were incorporated as outgroups.

DNA sequencing and plastome assembly and annotation

DNA extraction, library construction, and sequencing were performed by Annoroad Gene Technology (Beijing, PR China) Co., Ltd. Genomic DNA was isolated from frozen leaves using the DNAquick Plant system (TianGen Biotech, Beijing) following the manufacturer’s protocol. DNA degradation was assessed on 1% agarose gels. Purity and concentration were determined with a NanoPhotometer (Implen, USA) and a Qubit 2.0 fluorometer (Thermo Fisher Scientific, Massachusetts). Total DNA was fragmented to approximately 350 bp on an Ultrasonic Processor. Libraries were constructed using the NEBNext Ultra II DNA Library Prep Kit (Ipswich, Massachusetts) according to the manufacturer’s protocol, and subsequently diluted to 1 ng/µL. The final concentration and fragment sizes were verified on an Agilent 2100 Bioanalyzer (Agilent Technologies, California). Sequencing was performed on an Illumina HiSeq X Ten System (San Diego, California). Approximately one GB of 150 bp paired-end reads was generated for each sample.

Plastomes were assembled from cleaned reads using NOVOPlasty v. 2.7.2 (Dierckxsens, Mardulyn & Smits, 2017). When multiple contigs were obtained (rather than a single circularized assembly), we mapped them contigs against one of several Acer plastomes from GenBank (see accessions in Table S1) using Geneious v. 11.0.4 and merged the ones that overlapped. Because in several instances the assemblies resulted in short non-overlapping contigs, we mapped our reads against the contigs to extend their ends until the gap was closed. Mapping was performed with medium-low sensitivity for 100 iterations.

The annotation of the assemblies was performed with GeSeq (Tillich et al., 2017). We selected ARAGORN as third party tRNA annotator and the plastome sequence of Acer miyabei Maxim. subsp. miaotaiense (P.C. Tsoong) A.E. Murray (NC_030343) as the reference genome. The annotated sequences were aligned and verified with published Acer plastomes using MAFFT v. 7.450 (Katoh & Standley, 2013) in Geneious and submitted to GenBank (Table S1).

Plastome comparison

The boundaries between the four plastome regions were inspected with the online tool IRscope (Amiryousefi, Hyvonen & Poczai, 2018), which allows visualizing the position of genes in the vicinity of these sites across species. We compared sequences and identified regions of variability with mVISTA (Frazer et al., 2004) using the annotated plastome of Acer acuminatum Wall. ex D. Don as a reference. Additionally, we performed a sliding window analysis as implemented in DnaSP v. 6.12.03 (Rozas et al., 2017) to locate genomic regions with high levels of variation. The alignment of the 16 Acer plastomes obtained with MAFFT (with default settings) was used as input file. The window length and step size were set to 600 bp and 100 bp, respectively. Using window’s position, we identified regions with 60 or more variable (polymorphic) sites (S). These regions were extracted from the alignment and analyzed individually to estimate their number of variable sites and parsimony-informative sites.

A codon usage analysis for protein-coding genes was also performed in DnaSP. We extracted the CDSs using Geneious and calculated the codon frequency and the relative synonymous codon usage (RSCU) values as a measure of codon usage bias (Sharp, Tuohy & Mosurski, 1986).

REPuter (Kurtz et al., 2001) was used to detect various types of repeated sequences (forward, reverse, complement, and palindromic) with a minimum size of 25 bp and sequence identity greater than 90%. Simple sequence repeats (SSRs) were identified with MISA-web (Beier et al., 2017). We used the default values of 10, 6, 5, 5, 5, and 5 to set the minimum number of repetitions for mono, di-, tri-, tetra-, penta-, and hexanucleotide repeats, respectively.

Phylogenomic reconstruction

Phylogenomic reconstruction was performed on a dataset consisting of 22 Acer plastomes (six retrieved from GenBank), and four outgroup species (Table S1). We aligned the annotated sequences using MAFFT with default parameters and then removed one inverted repeat. Maximum likelihood and Bayesian analyses were performed on this data set using two partitioning strategies to explore the effect of partitioning on tree topology. In the first one, the alignment was fully partitioned into coding and non-coding regions with protein-coding genes further divided by codon position. We used Geneious to extract the protein-coding sequences to define the codon positions in the data blocks and then concatenated these with the non-protein-coding regions. In the second strategy, the whole plastome was treated as a single partition (i.e., unpartitioned).

We used PartitionFinder2 (Lanfear et al., 2016) to select the best partitioning scheme and best-fit substitution models for the partitioned dataset. Branch lengths were set as ‘linked’ and the AICc was used for model selection. The search was performed using ‘rcluster’, a fast algorithm recommended for large datasets with a high number of partitions (Lanfear et al., 2014). We defined 324 data blocks which were reduced to 93 subsets in the best-fit scheme. These subsets and their corresponding substitution models were specified in both ML and Bayesian partitioned analyses. The unpartitioned analysis was run using GTR+I+G as a substitution model, which was selected using both PartitionFinder2 and ModelTest-NG (Darriba et al., 2019). The maximum likelihood analysis was performed in RaxML-NG (Kozlov et al., 2019) with 1,000 bootstrap replicates. The Bayesian analysis was conducted in MrBayes v. 3.2.7 (Ronquist et al., 2012) on the CIPRES Science Gateway (Miller, Pfeiffer & Schwartz, 2010). We ran two runs of 100 million generations and four chains, sampling trees every 4,000 generations and discarding the 20% as burn-in. Tracer v. 1.6.0 (Rambaut et al., 2018) was used to verify that both runs reached stationarity and converged on the same distribution.

Size and composition of Acer plastomes

The 16 plastomes generated in this study varied in length from 155,212 bp in Acer carpinifolium Siebold & Zucc. to 157,023 bp in Acer palmatum Thunb.var. palmatum (Table 1). All exhibited the typical tetrapartite organization, with the large single copy (LSC), small single copy (SSC), and inverted repeat (IR) regions ranging from 85,313 to 86,147, 17,724 to 18,232, and 26,020 to 26,757 bp, respectively. The largest genome size of A. palmatum is due to its expanded IRs relative to the other species.

The GC content was similar for all species (37.8–38%). As is common in plastomes of seed plants, the IRs had the highest GC content (42.8–43%) due to the presence of the GC-rich rRNA genes, whereas the SSC region had the lowest values (35.9–36.2%, Table 1).

All plastomes contained 117 different genes, of which 19 are duplicated for a total of 136 genes (Table 2). Acer palmatum is the only species with 137 genes due to an entire additional copy of the ribosomal rps19. Eighty-one of the 117 genes are protein-coding genes, 31 are transfer RNA genes, and four are ribosomal RNA genes. One gene, infA, appears as a pseudogene and is likely to be non-functional in all species. Sixteen genes contain a single intron, and two have two introns (Table 2).

The size of homologous genes was very similar across species except for a few cases. For example, the rps2 gene was found to be truncated at different positions in 11 species due to mutations leading to premature stop codons. It was significantly shorter (72–264 bp) and perhaps non-functional in A. acuminatum, A. carpinifolium, A. micranthum, A. negundo, A. nipponicum, A. palmatum, A. pilosum, A. pseudoplatanus, A. rubrum, and A. tataricum. The other six species had lengths of 588 and 633 bp. The rpl22 gene was also notably shorter (183–225 bp) in two species (A. sterculiaceum and A. pentaphyllum) compared to the rest whose length ranged from 480 and 498 bp.

In all species, with the exception of A. maximowiczianum and A. palmatum, the boundary between the LSC and IRb regions is located in the 3′ end of the rps19 gene (Fig. 1). In A. maximowiczianum, the entirety of rps19 lies in the LSC region, whereas in Acer palmatum, it is found within the inverted repeats. This species is atypical in that the boundary LSC-IRb is situated in the rpl22 gene, which results in a duplicated fragment of this gene in the IRa adjacent to the junction with the LSC region (Fig. 1).

A truncated but seemingly functional copy of the ycf1 gene is present in the IRb of all plastomes extending to various lengths into the SSC region. The ndhF gene also spans the boundary SSC-IRb in the opposite direction, partially overlapping with the 3′ end of ycf1. The only exception is Acer nipponicum, in which the entire ndhF sequence lies within the SSC region (Fig. 1).

Sequence variation

The plastomes of Acer are very similar and largely conserved as shown in the mVISTA alignment, with most of the variation found in the non-coding sequences of the LSC and SSC regions (Fig. 2). Fifteen most variable regions (S ≥ 60) were identified in the sliding window analysis. All but one, the ycf1 gene, were or contained intergenic spacers. Three compound regions, formed by one protein-coding gene (rps2, rpl32 and ccsA) plus one or both flanking spacers, were detected (Table 3, Fig. 3). When these regions were analyzed separately, the number of polymorphic sites varied from 36 (ccsA–ndhD) to 438 (ycf1). The percent of variability and parsimony-informative sites ranged from 5.3% (rps16–trnQ-UUG) to 16.7% (ccsA–ndhD), and from 1.0% (accD–psaI and rps16–trnQ-UUG) to 8.8% (ccsA–ndhD), respectively (Table 3). The five most variable regions, both in percent of variability and total number of parsimony-informative sites were the spacers ccsA–ndhD, psbZ–trnG-GCC, ndhC–trnV-UAC, trnE-UUC–trnT-GGU, and the ycf1 gene. This gene exhibited greater variation in the SSC portion of the genome (9.7%) compared to the IR portion (1.1%).

Codon usage

The total number of codons (including stop codons) of the protein-coding regions of the plastomes ranged from 26,678 in A. psudoplatanus to 26,865 in A. glabrum. Codon frequency and RSCU values were very similar across species (Table 4, Table S2). The most frequent codons were AUU-Ile (1084-1112), AAA-Lys (1,060–1,079), and GAA-Glu (1,011–1,024), the three accounting for about 11% of the total number of codons in all species. The three least frequent were UGC-Cys (81–86), AGC-Ser (124–131), and CGC-Arg (127–136). The most commonly specified amino acids were leucine and isoleucine, encoded by about 10% and 8% of codons, respectively (Table 4). Codons that have T or A in their third position had RSCU >1, whereas codons ending in C or G had RSCU <1 indicating a strong bias in favor of codons ending with T and A. The only exceptions to this pattern were UUG-Leu with RSCU values between 1.20 and 1.23, CUA-Leu with 0.83–0.85, and AUA-Ile with 0.92–0.94. The two codons with the highest RSCU values (both with 1.78–1.81) were UUA-Leu and AGA-Arg, and the two with the lowest were AGC-Ser (0.36–0.38) and UAC-Tyr (0.37–0.39) (Table 4, Table S2).

Repeat sequence analysis

The analysis with REPuter detected between four and 18 repeats with a length ≥25 bp in all plastomes (Fig. 4, Table S3). Acer platanoides L. had the highest number while all other species ranged between four and eight repeats. The most common type were forward sequences in the range of 40–49 bp followed by palindromes between 30–39 bp (Fig. 4). No complement repeats were found in any of the plastomes analyzed.

The majority (about 67%) of these repeated sequences were identified in the intergenic spacers of the LSC and SSC regions; nearly a third were detected in the most variable spacers (Table 3). Only two were located in intergenic spacers of the IRs in one species (Acer glabrum Torr.). The remaining repeats were detected in five protein-coding genes (psaA, psaB, rpoC1, rpl22, rpl32) and one transfer RNA gene (trnS-GGA). The repeats found in the rpoC1 gene were located in its intron sequence (Table S3).

The number of SSR loci ranged from 60 in the plastomes of A. acuminatum and Acer pilosum Maxim. to 92 in A. glabrum. Ninety-seven percent were mononucleotide repeats of up to 28 bp long. A few dinucleotide repeats were found in 13 of the 16 species, whereas tri- and tetranucleotide repeats were detected in only five and one species, respectively (Fig. 5). The most frequent motifs were (A/T)10 and (A/T)11, which collectively accounted for up to 78–79% of all SSRs in some species. C and G repeats were rare, with only one to three per species (Table S4).

The percentage of the cp genome containing SSRs varied from 0.48 to 0.94% (average = 0.7%). Most loci (90%) were identified in the LSC and SSC regions (Fig. 6). About 58% were located in the intergenic spacers of the genome, whereas 42% were detected in 25 different genes. The ycf1 gene had the highest number of SSRs with six to ten per species (Table S5).

Phylogenomic reconstruction

The plastome data set (with one inverted repeat removed) consisted of 153,383 aligned sites, of which 20,478 (13.4%) were variable, and 5,774 were parsimony-informative. Pairwise percent identity for all species is provided in Table S6. Both ML and Bayesian analyses on the fully partitioned and the unpartitioned dataset produced trees with identical topologies, very short internal branches, and similar support values (Fig. 7).

All analyses provided maximum support for the monophyly of Acer and its sister relationship to Dipteronia. Relationships within the genus were fully resolved with strong support; only two branches were weakly to moderately supported in the ML analysis with BS values of 67 and 83%, respectively (Fig. 7). The earliest diverging lineages within the genus included species of sections Acer (Acer pseudoplatanus L.), Glabra (A. glabrum), Pubescentia (A. pilosum), Negundo (Acer negundo L.) and Rubra (Acer rubrum L.). The remaining species were grouped into three mutually exclusive clades, one consisting of A. nipponicum and A. carpinifolium (sections Parviflora and Indivisa, respectively), another formed by species of sections Palmata and Lithocarpa, and the other composed by members of the remaining sections. Within this clade, A. tataricum subsp. ginnala is sister to all other species. Section Platanoidea was recovered as sister to section Macrantha with A. acuminatum (section Arguta) as sister to this clade. The species of sections Trifoliata, Oblonga, and Pentaphylla formed another clade that is sister to the Platanoidea-Macrantha-Arguta clade. All sections represented by more than one species were recovered as monophyletic with high support (Fig. 7).

Plastome composition and variable regions

The plastomes of Acer generated in this study are very homogeneous in size and structure as well as gene content. Of the 16 species sampled, A. palmatum (section Palmata) is the only one that differed from the rest by having much longer IRs and an extra copy of the rps19 gene (Fig. 1, Table 1). A survey of plastomes of other members of section Palmata, including Acer wilsonii Rehder (MG012225) and 16 additional species assembled by the authors (Data S2) shows that all share these IRs of about 700 bp longer. Expanded IRs containing whole copies of rps19 are also present in Dipteronia (NC_029338, NC_031899), Dodonaea viscosa Jacq. (NC_036099), Dimocarpus longan Lour. (NC_037447), Koelreuteria paniculata Laxm. (NC_037176), Litchi sinensis (NC_035238), Sapindus mukorossi Gaertn. (NC_025554), and Xanthoceras (NC_037448), but are absent in Aesculus wangii Hu (NC_035955). They have also been detected in other families of Sapindales, including Meliaceae, Rutaceae, Nitrariaceae, and Simaroubaceae (Lu et al., 2017; Mader et al., 2018; Saina et al., 2018). In the case of Acer, this feature appears to be exclusive to species of section Palmata.

In the 16 plastomes studied, the infA gene was found as a pseudogene, with no start codon and the open reading frame disrupted by internal stop codons. In all other Sapindaceae surveyed (listed above), infA also appears as a pseudogene suggesting that it became obsolete in the ancestor of this family or perhaps of Sapindales. It has been reported as missing or as a pseudogene, in several genera of Meliaceae (Mader et al., 2018), Simaroubaceae (Saina et al., 2018), Nitrariaceae (Lu et al., 2017), and Anacardiaceae (Wang et al., 2020). The loss of this gene from cpDNA has been documented in many other angiosperm lineages, and there is even indication of its transfer to the nuclear genome in a number of species (Daniell et al., 2016; Millen et al., 2001).

The ribosomal protein genes rps2 and rpl22 exhibited considerable variation in length and might be non-functional in several species. In A. acuminatum, A. carpinifolium, A. micranthum, A. negundo, A. nipponicum, A. palmatum, A. pilosum, A. pseudoplatanus, A. rubrum, and A. tataricum, rps2 was 60–80% shorter compared to the other Acer species and genera of Sapindales. In the case of rpl22, it was 50 and 60% shorter in A. sterculiaceum and A. pentaphyllum, respectively. The likely loss of function in these species is notable because both genes are considered essential for plant survival (Daniell et al., 2016; Tiller & Bock, 2014). A possible explanation is that they have been transferred to the nucleus, as documented for rpl22 in Fabaceae and Fagaceae (Jansen et al., 2011). Another possibility is their substitution by a nuclear gene that encodes chloroplast (and mitochondrion) targeted proteins as is the case of the rps16 gene in some plant lineages (Keller et al., 2017; Ueda et al., 2008). A search for these genes in the nuclear and mitochondrial genomes will reveal insights into their evolutionary fate in Acer.

In addition to providing insight into the structure and gene organization, genome comparisons are useful for identifying variable regions suitable for the development of molecular markers. In Acer, only a few cpDNA loci (atpB-rbcL, ndhF, psbA-trnH, psbM-trnD, rbcL, rpl16, trnD-trnT, trnL and trnL-trnF) have been employed in phylogenetic (Li, 2011; Li, Yue & Shoup, 2006; Pfosser et al., 2002; Renner et al., 2007; Renner et al., 2008; Tian, Guo & Li, 2002; Zhang, Li & Li, 2010) and phylogeographic studies (Guo et al., 2014; Saeki et al., 2011). The maximum number included in a single dataset is six, but the resulting trees failed to provide adequate resolution and support for many major clades (Renner et al., 2007; Renner et al., 2008). Two additional loci, matK and trnS-trnG, were explored along with rbcL as barcodes to distinguish among 85 Acer species, but they also showed low discrimination power even when used in combination (Han et al., 2016). Only one of these commonly used markers, psbA-trnH, was identified in this study among the most variable in the genus. We found that 14 other loci (Table 3) exhibited greater variation and are thus potentially more informative than the previously used sequences. This supports the notion that universal markers may not be variable enough for many groups and underscores the importance of identifying specific loci for phylogenetic studies at the genus and family level (Cvetkovic, Hinsinger & Strijk, 2019).

Codon usage and repeat sequence analysis

The pattern of codon usage in Acer plastomes is very similar to that of other members of Sapindales. The three most frequent codons (AUU, AAA, and GAA) and the least frequent (UGC, AGC, and CGC) are shared, for example, with Ailanthus (Simaroubaceae), Nitraria (Nitrariaceae) and Toxicodendron (Anacardiaceae) (Lu et al., 2017; Saina et al., 2018; Wang et al., 2020). The most commonly specified amino acids, leucine and isoleucine, are also the most commonly specified in these and many other plant genera (Li et al., 2018; Silva et al., 2018; Yang et al., 2018).

Codon usage was strongly biased toward codons ending in A or T, i.e., there are more codons ending in A or T (68–69% of all codons), and these are used more often (RSCU>1) than the ones ending in C or G (Table 4). This pattern has been observed in most chloroplast genomes of plants (Morton, 2000; Shimada & Sugiura, 1991), and has been related to the high AT composition bias of the protein-coding genes (Morton, 2000). However, selection for translation efficiency is likely the most important factor driving the codon usage of plastid genes (Suzuki & Morton, 2016).

Repetitive elements are a common feature of plastomes, although the amount and distribution can vary widely among plant groups (Sveinsson & Cronk, 2014); (Wicke et al., 2011). In the Acer plastomes analyzed, dispersed repeats of 25 bp and longer were few, ranging between four and eight in most species (Fig. 4). Except for the IRs, all had less than 60 bp in length and were more often located in the non-coding regions of the plastome conforming to the general pattern in angiosperms (Wicke et al., 2011). In the case of SSRs, T and A mononucleotide repeats were more frequent than C/G repeats and than more complex motifs, and were more abundant in non-coding DNA, also conforming to the trend observed in many plant groups (Ebert & Peakall, 2009).

SSRs are among the most variable components of the genome, and constitute an invaluable source of information for population genetic studies, DNA fingerprinting, and plant breeding programs (Nybom, Weising & Rotter, 2014; Wheeler et al., 2014). They have been used to characterize the genetic diversity in Acer species of economic and conservation interest (e.g., Acer campestre L., Acer capillipes Maxim., Acer mono Maxim. Acer opalus Mill., A. pseudoplatanus, A. saccharum and Acer yangbiense Y. S. Chen & Q. E. Yang). However, most studies have employed nuclear (nSSRs) microsatellite loci (Chybicki, Waldon-Rudzionek & Meyza, 2014; Graignic, Tremblay & Bergeron, 2013; Kikuchi & Shibata, 2008; Liu et al., 2014; Neophytou, Konnert & Fussi, 2019; Pandey et al., 2012; Segarra-Moragues, Gleiser & González-Candelas, 2008; Terui et al., 2006; Zhao, Sun & Yang, 2011), and very few have used cpSSRs (Petit et al., 2003; Saeki et al., 2011; Neophytou, Konnert & Fussi, 2019). Our SSR analysis revealed a large number of these repeats distributed across non-coding and protein-coding regions of the Acer plastomes (Table S5) providing an opportunity for the development of new cpSSRs markers for the genus. Because plastids are maternally inherited in Acer (Corriveau & Coleman, 1988), cpSSRs can be particularly useful for studies of gene flow through seed dispersal, and for tracing the maternal lineage in space and time (Nybom, Weising & Rotter, 2014; Petit et al., 2003; Provan, Powell & Hollingsworth, 2001).

The locus with the highest number of SSRs was ycf1. This gene was also identified among the most variable in Acer. It has been reported as among the most variable of the plastome in many plant groups (Jiang, Hinsinger & Strijk, 2016; Kumar et al., 2009; Neubig et al., 2009; Silva et al., 2018; Thomson, Vargas & Dick, 2018), and has even been recommended as a barcode marker (Dong et al., 2015). However, it has been rarely used for phylogenetic inference and barcoding compared to other loci. Future studies should investigate the potential of this gene as a phylogenetic marker in Acer and other plant groups.

Phylogenomic reconstruction

We conducted ML and Bayesian analyses with 22 species of Acer using two different partitioning strategies (i.e., a fully partitioned and an unpartitioned dataset) to explore the effect of partitioning on tree inference. We found that both methods produced identical trees with similar support values regardless of the partitioning strategy used.

It is widely accepted that partitioning is important to account for rate heterogeneity and patterns of substitution among sites and that the choice of partitioning scheme can affect the phylogenetic inference (Brown & Lemmon, 2007; Kainer & Lanfear, 2015; Lanfear et al., 2014). A number of studies that have investigated the effect of partitioning have reported improvement in tree topology, branch lengths, and branch support when the data is appropriately partitioned. In contrast, unpartitioned or poorly partitioned analyses may lead to well-supported but incorrect relationships (Brandley, Schmitz & Reeder, 2005; Kainer & Lanfear, 2015; Tao, Mayden & He, 2013; Ward et al., 2010). In our case, partitioning the cp dataset into 93 subsets (based on spacers, genes, introns, exons, and codon position) did not have an effect on the phylogenetic reconstruction. A similar outcome was reported by Fu et al. (2017) using plastid genomes of Cornales and by Dong et al. (2018) with Saxifragales. The results of these studies, including our own, suggest that the effect of partitioning might be less important in large datasets, presumably due to the increase in phylogenetic signal. Kainer & Lanfear (2015) examined the impact of partitioning scheme choice by analyzing 34 datasets, the largest of which had over 25,000 sites. They found that “the longer the alignment, the less the results depend on the partitioning scheme”. With typically more than 130 kb in length, plastid genome data sets appear to be large enough (and to contain enough phylogenetic signal) to converge on the correct tree irrespective of the partitioning scheme used. It would be interesting, nonetheless, to explore further the effect of partitioning on tree inference using a diverse set of complete plastid genome data.

Our phylogenetic analyses confirmed many of the relationships inferred in previous studies using cp markers (Renner et al., 2007; Renner et al., 2008). These include the earliest-diverging position of the clade formed by A. glabrum and A. pseudoplatanus, the close relationship between A. nipponicum and A. carpinifolium (sections Parviflora and Indivisa respectively), the sister relationship of section Pentaphylla to the Trifoliata-Oblonga clade, and the sister relationship of sections Macrantha and Platanoidea (Fig. 7). Previously unresolved or unsupported relationships in the cp tree were also clarified in our analysis. For example, section Lithocarpa, represented here by Acer sterculiaceum Wall. was resolved as sister to the clade comprising members of section Palmata with maximum support. Section Rubra, A. negundo, and A. pilosum, whose positions were also unresolved, were placed among the early-diverging lineages of the genus (Fig. 7). The placement of A. tataricum, the only species of sect. Ginnala, differed from previous studies based on cp data. This species was recovered by Renner et al. (2007) and Renner et al. (2008) as sister to the clade comprising members of section Platanoidea, Macrantha, Arguta, and part of Negundo, although with no support. In our study, A. tataricum was placed as sister to a major clade formed by species of sections Platanoidea, Macrantha, Arguta, Trifoliata, Oblonga, and Pentaphylla with maximum support (Fig. 7). Overall, these results represent a significant improvement over previous studies based on cp markers, which have been characterized by poor resolution and statistical support (Harris et al., 2017; Li, Yue & Shoup, 2006; Renner et al., 2007; Renner et al., 2008).

A comparison with the phylogenomic analysis by Li et al. (2019) based on nuclear sequences reveals very different topologies between the plastid and nuclear trees. In the nuclear tree, Acer species were grouped into two main lineages, one comprising members of sections Spicata, Palmata, Negundo and Arguta, and the other including the remaining sections. Relationships among sections within each of these two clades also differed markedly between the two trees. For example, sect. Platanoidea and Macrantha were not sister groups in the nuclear tree as was inferred in our cp tree (Fig. 7). Section Macrantha was recovered as sister to a clade comprising species of sect. Ginnala, Indivisa, Lithocarpa, Platanoidea, Rubra, Acer, Trifoliata, and Pentaphylla, whereas monotypic sect. Glabra was sister to sect. Parviflora (Li et al., 2019).

Unlike nuclear trees, which typically tend to be consistent with taxonomy (as is the case in Acer), cp-based trees often correlate with geographic patterns, which can be explained by the occurrence of hybridization and introgression (Albaladejo et al., 2005; McKinnon et al., 1999; Rautenberg et al., 2010). Natural hybridization has been documented in Acer (e.g., Liao et al., 2010). However, few studies have employed molecular tools to study its contribution to the evolution of the genus. For example, Grimm, Denk & Hemleben (2007), using nuclear markers, found evidence of ancient hybridization and introgression in Acer section Acer, a group exhibiting high morphological variability. Also, polyploidization has been suggested to have played a role in the diversification of section Rubra, which includes diploid, tetraploid, hexaploid, and octoploid species (Harris et al., 2017), although no study has been conducted to test this hypothesis. Unfortunately, the limited sampling of cp genomes in this study does not allow detecting any apparent geographic pattern, except perhaps for the position of the three North American species (A. glabrum, A. negundo, and A. rubrum) which were placed among the earliest-diverging lineages (Fig. 7).

There are three polyploid species in our study, and their discordant positions in the plastid and nuclear trees (Li et al., 2019) might indicate past hybridization and introgression. Acer rubrum is hexaploid and octoploid, whereas Acer carpinifolium and A. pseudoplatanus are both tetraploids (Contreras & Shearer, 2018). Acer pseudoplatanus (section Acer) is an autotetraploid (Grimm, Denk & Hemleben, 2007; Pandey et al., 2012) i.e., its polyploidy did not result from hybrid speciation. Yet, its different placement in the plastid and nuclear trees suggests a hybrid origin for this species. In the cp tree, this species was placed as sister to A. glabrum (section Glabra) in the earliest diverging lineage of Acer with maximum support (Fig. 7). It was also recovered as sister to A. glabrum in the analyses of Renner et al. (2007) and Renner et al. (2008) and outside the clade comprising most species of section Acer. However, in the nuclear tree of Li et al. (2019), A. pseudoplatanus was placed in a monophyletic Acer section, also with strong support. This incongruence might be explained by chloroplast capture via hybridization between these two lineages (but also by the retention of ancestral polymorphism, i.e., incomplete lineage sorting). Further analyses with an increased sampling within sections (and populations) will likely reveal important clues on the causes of discordance between plastid and nuclear trees, and on the role of hybridization in the diversification of the genus.

The very short internal branches in our trees suggest a rapid differentiation of the main lineages of Acer within a short period of time. This is consistent with earlier studies that have analyzed the distribution of the rich fossil record (Boulter et al., 1996; Manchester, 1999; Wolfe & Tanai, 1987), which have suggested a burst of diversification during the second half of the Eocene. Based on nuclear data, Li et al. (2019) estimated that most sections in Acer had originated by the late Eocene (33–38 Mya). However, their biogeographic analysis was limited by a narrow sampling (30 taxa) and only two calibration points. Dating analyses incorporating a denser sampling of both plastid and nuclear genomic data are needed to improve our understanding of the diversification and biogeographic history of this diverse genus.