The sugarcane accessions reported in this study are part of the WCSRG, which is curated by the USDA-ARS, SHRS, in Miami, FL. Figure 1 provides a summary of the geographical distribution of the germplasm accessions. A detailed list of all accessions is provided in Table S1 . The S. spontaneum accessions are maintained in 7-gallon pots on a concrete pad and not allowed to flower as they are considered invasive. The rest of the accessions are planted in the field and rotated to new field plots every 4 years. The mature plants are cut to the ground every year in the early spring until replanting. The species name of each accession in the WCSRG was defined based on the curator’s naming system.

From each sugarcane plant, one fully expanded young leaf was collected into labeled paper envelopes. A total of eight leaf disks were collected using the BioArk Leaf kit provided by LGC, Biosearch Technologies (https://www.biosearchtech.com/). The prepared BioArk Leaf kits were then submitted to LGC Genomics (Middleton, WI) for DNA extraction and subsequent genotyping.

The single-dose markers were initially selected from the Axiom Sugarcane 100K SNP array, which was developed based on five Saccharum species (S. sinense, S. barberi, S. robustum, S. officinarum, and S. spontaneum) and 37 sugarcane hybrids (You et al., 2019). A set of two thousand bi-allelic and single-dose SNPs were randomly selected from the Axiom Sugarcane 100K SNP array. Probes targeting these SNPs were designed by Biosearch Technologies (https://www.biosearchtech.com) and used to amplify sugarcane genome DNA libraries, which were then sequenced using the 1x 75 bp Illumina sequencing platform. Reads were trimmed by removing the first 40 bases and quality-checked with a Q value >20. The 2000 candidate SNPs were first evaluated in a pilot study using 196 sugarcane germplasm accessions ( Table S1 ). Based on call rate, Minor Allele Frequency (MAF), and Linkage Disequilibrium (LD), we then selected a low-density genotyping panel including 400 SNPs and used it to genotype all the Saccharum clones including six Saccharum species and inter-specific hybrids.

The genotyping was performed using a targeted genotyping-by-sequencing approach called SeqSNP, which has been successfully used in several crops (Zhang et al., 2020; Jo et al., 2021; Ziarsolo et al., 2021). Sequence reads were aligned to the sequences (300bp) flanking the SNP markers, using Bowtie2 (version 2.4.5) (Langmead and Salzberg, 2012). The subsequent alignments (BAM files) were used to call SNP variants by the freebayes program (version 1.1.0) (Garrison and Marth, 2012) with the following parameters: min-quality 20, –min-supporting-allele-qsum 10; –read-mismatch-limit 4; –mismatch-base-quality-threshold 10; –exclude-unobserved-genotypes; –no-mnps; –no-complex; –ploidy 4; –min-alternate-fraction.08333; –legacy-gls. The called SNP variants (VCF files) were further filtered using the following criteria: 1) SNPs must be biallelic; 2) SNPs must be supported by at least 20 reads, otherwise, they were marked as missing genotypes; 3) both alternative and reference alleles must each be supported by at least 2 reads. After the initial filtering, informative SNPs were selected by excluding SNPs with a missing rate of 10% and a minor allele frequency of <5%. Additionally, samples with 10% or more missing SNP genotyping were also excluded. These filtering processes resulted in a dataset of 357 SNPs and 901 samples for downstream analyses. To assess the degree of variation among SNP markers, genetic parameters such as minor allele frequency (MAF), expected heterozygosity (H_exp), and observed heterozygosity (H_obs) for each SNP marker were measured using the R-package snpReady (version 0.9.6) (Granato et al., 2018).

For this study, we defined three types of problems related to genetic integrity of sugarcane germplasm. The first type was synonymous mislabeling or “duplicate error,” meaning that sugarcane clones had different names but shared the same SNP genotype. The second was homonymous mislabeling, meaning that individual clones had the same name in this collection, but they had different SNP genotypes. The third type was mistakes, inaccuracies, or a lack of information in species classification, where the species membership of a given clone was wrongfully recorded (Brown et al., 2007; Yang et al., 2019).

To identify synonymous mislabeling among sugarcane clones, the allele difference between each pair of individuals was computed using the R-package poppr (Kamvar et al., 2014). Individuals with zero allele difference at all loci were considered duplicates. The groups of duplicates were visually inspected by constructing a network graph using the R-packages of network and ggplot2. Since genotyping errors are not uncommon, a pair of clones that differ by one or two loci could be the same clone (Kalinowski et al., 2006; Zhang et al., 2006). To assess potential genotyping error, we included 12 sugarcane samples in genotyping. These 12 samples were propagated from a single clone “P-Mag-84-2” ( Table S1 ) and served as an internal control. Mismatched SNP loci among these 12 samples were calculated and used as a baseline to determine the “mismatch threshold” for clone identification. Any pair of samples that had mismatched loci below the threshold (near-identical genotypes) were considered as putative duplicates (Zhang et al., 2006; Akpertey et al., 2021).

The statistical rigor of duplicate identification was assessed using the probability of identity that two individuals may share the same multilocus genotype by chance (Waits et al., 2001). The computer program GenAlEx 6.5 (Peakall and Smouse, 2006; Peakall and Smouse, 2012) was used to calculate the probability of identity among siblings (PIDsib). PIDsib is defined as the probability that two sibling individuals drawn at random from a population have the same mutilocus genotype (Waits et al., 2001).

To identify homonymous mislabeling, SNP genotypes of the clones with the same name were manually grouped and compared using multi-locus matching. If the clones differ by more than two loci, then these clones were considered to have different SNP genotypes thus were claimed as homonymous mislabeling.

To identify clones with mistake in species membership, assignment test based on Bayesian clustering analysis of population structure (Pritchard et al., 2000) was used. Clones with wrongfully assigned species membership were detected and corrected based on the assignment result (see the next section).

To assess population structure in the collection, we only used accessions with explicit species names recorded in the passport data. Clones recorded as ‘Hybrid’ or ‘Unknown’ were excluded in this stage, which led to the retention of 591clones for the population structure analysis. The computer program STRUCTURE ver. 2.3.4 (Pritchard et al., 2000) was used. The program was run at 10 independent repetitions for population numbers ranging from K = 1 to K = 10, with a burn-in period of 50,000 and 100,000 Markov chain Monte Carlo (MCMC). The optimal number of model components (K) was determined based on delta K (Evanno et al., 2005). Based on the result, iterative runs were performed on each partitioned cluster to explore the sub-structures within each cluster, as recommended by Evanno et al. (2005). Ancestry and admixture proportions were visualized using computer program CLUMPAK (Kopelman et al., 2015).

Based on the result of the STRUCTURE analysis, clones with the assignment coefficient above 0.80 (Q value >0.80) were considered core members of each cluster and retained for subsequent genetic diversity analysis, including computation of F statistics, Analysis of Molecular Variance (AMOVA), Principal Coordinate Analysis and Neighbor-Joining Clustering Analysis.

AMOVA was performed using the program GenAlex 6.5 (Peakall and Smouse, 2006; Peakall and Smouse, 2012). The significance of fixation index (F_ST) was tested using 999 random permutations. In addition, the F_ST for each pair of core germplasm groups was calculated and the statistical significance was tested using permutations with the program GenAlEx 6.5.

Key summary statistics including gene diversity (H_exp) and observed heterozygosity (H_obs) were calculated for each species using the program GenAlEx 6.5 (Peakall and Smouse, 2006; Peakall and Smouse, 2012). To illustrate genetic relationships among the species, a distance-based multivariate analysis was performed. Pairwise genetic distances were computed using the Distance option, and Principal Coordinates Analysis (PCoA) within the GenAlEx 6.5 program. The PCoA results are presented as two-axis PCO plots, and both plots axis 1 vs 2 and axis 1 vs 3 are presented separately.

To further examine the genetic relationship among different species, a neighbor-joining (NJ) clustering analysis was performed. The NJ tree was constructed based on the SNP genotype data using the R-package poppr (Kamvar et al., 2014). Pairwise genetic distances between the sugarcane clones were estimated and the neighbor-joining method was used to construct the tree. The final tree was visualized using FigTree version1.4.4 (http://tree.bio.ed.ac.uk/software/figtree).

To assess species membership and parentage for clones recorded as ‘hybrid’ or ‘unknown’, we used the core members of the five Saccharum species as references and increased their samples size to 500 for each species, using the Simulation procedure implemented in the computer program ONCOR (https://www.montana.edu/kalinowski/software/oncor.html). The simulated populations were then analyzed together with the 298 clones with unclarified species membership (clones labeled as ‘unknown’ or ‘hybrid’) using STRUCTURE 2.3.4. An admixed model was selected, and the number of clusters (K value) was set to five, corresponding to the five Saccharum species. Ten independent runs were conducted at K = 5, each consisting of 100,000 iterations after a burn-in period of 50,000 iterations. From the 10 independent runs, the mean membership score was presented as the inferred species/parentage or species membership.

After applying an initial filtering process to exclude SNPs and samples with a missing rate of 10% or greater, a total of 751 SNP markers and 901 clones were obtained. Among the Saccharum clones, there were 286 S. spontaneum, 175 unknown, 158 S. officinarum, 123 hybrids, 76 S. robustum, 46 S. sinense, 32 S. barberi, 3 S. edule, 1 S. narenga, 1 S. arundinaceum, as recorded in passport data ( Table S1 ). Of the SNPs, 217 (29%) were found to be monomorphic. To obtain informative SNP markers, we further filtered out SNPs with a minor allele frequency of less than 5%, resulting in a final set of 357 markers. The final genotype data for these markers showed an average missing rate of 0.15%, ranging from 0 to 7.6%. The sugarcane clones, on the other hand, showed an average missing rate of 0.15%, ranging from 0 to 6.2%. Among the 357 SNP markers, the expected heterozygosity ranged from 0.09 to 0.5, with an average of 0.23. The observed heterozygosity ranged from 0.09 to 0.99 with an average of 0.3. Additionally, the minor allele frequency ranged from 0.05 to 0.5 with an average of 0.153 ( Table S2 ).

Clonal propagation and field maintenance of sugarcane germplasm plants can often lead to mislabeling and name loss. The WCSRG, which houses collections from diverse locations worldwide, often encounters duplicated accessions with different regional names but identical clones. To estimate mislabeling and clonal redundancy, we measured the allele difference distance between each pair of sugarcane accessions at all SNP loci. Synonymous groups (duplicates) were determined when clones shared the same alleles at all SNP loci. Our analysis revealed 86 groups of duplicates comprising 211 accessions (23% of the examined Saccharum clones), demonstrating a high rate of synonymous mislabeling and genetic redundancy within the collection ( Figure 2 ). The number of duplicated clones within each group ranged from two to nine clones, with 67 groups consisting of two clones ( Table 1 ; Table S3 ).

To estimate genotyping error, twelve clonal samples propagated from a single clone of ‘P-Mag-84-2’ were included as an internal control. Of the 12 samples, however, only eight samples (Group1 in the Table S3 ) were identified as duplicates with zero allele difference, while three samples showed one allele difference, and one sample showed two allele differences. These differences were attributed to likely genotyping errors at four loci. Assuming no mutation in the clonal plants, these results indicated an error rate of 0.093% in our genotyping, as four loci were called wrongly out of a total of 4,284 loci in the 12 clonal samples. Based on the observed error rate, we relaxed the threshold of detecting duplicates. Any pair of samples that had up to two allele differences were considered putative duplicates. Based on this relaxed threshold, we identified 135 groups consisting of 471 clones (52% of the total clones) as putative duplicates. The number of duplicated clones within each group ranged from 2 to 38 ( Table S4 ).

The result of duplicate identification was supported by the probability of identity among siblings (PID-sib). The cumulative PID-sib of the first 48 SNPs ranged from 3.85E-04 (S. robustum) to 6.73E-07 (S. officinarum), which demonstrated that a high level of statistical power can be achieved in sugarcane germplasm analysis using only a small fraction of the 357 SNP markers ( Table S5 ; Figure S1 ). When all 357 SNP loci were included, the cumulative PID-sib ranged from 2.4E-20 (S. robustum) to 7.6E-42 (S. officinarum), which indicates that there is almost a null probability of finding two individual clones with the same genotype within any of the five Saccharum species.

To identify homonymous mislabeling, clones with the same name were compared for their SNP genotypes using multi-locus matching. A total of 27 homonymous mislabeling groups were detected in all the studied Saccharum species, except S. edule ( Table 2 ). Most of the identified homonymous groups were collected from the same country and geographical region, indicating that mislabeling occurred before the germplasm were introduced into WCSRG.

Of the 901 Saccharum (sensu stricto) accessions maintained in this sugarcane germplasm collection, there were 175 clones that do not have clear passport data for their species membership. These clones were recorded as ‘unknown’ in the collection. In addition, there were 123 clones recorded as ‘Hybrid’, but their parentage information was lacking. Using the selected core members of the five Saccharum species as references, we were able to assign the species membership and parentage for all the clones that have ambiguous passport records. Our result showed that most of these clones have a species membership of S. officinarum ( Table S8 ). Among the 175 clones recorded as ‘unknown’, two were assigned to S. barberi, 12 to S. sinense, 12 to S. spontaneum, 10 to S. robustum, 91 to S. officinarum, and 40 to inter-specific hybrids. Similarly, among the 123 clones that were recorded as ‘Hybrid’, two were assigned to S. spontaneum, one to S. robustum, 96 to S. officinarum, and 24 to inter-specific hybrid ( Table S9 ).

Using the selected SNP markers, we genotyped 901 Saccharum clones from six species including hybrids. Based on the SNP genotypes, we discovered a high rate of clone mislabeling and genetic redundancy within the studied sugarcane collection. Because genotyping errors frequently occur, a pair of clones with a small number of mismatched loci could be duplicates as well (Zhang et al., 2006). Therefore, a threshold of mismatches to determine duplicates needs to be established. To evaluate the genotyping error rate, we included 12 samples propagated from the same clone as an internal control. Among the 12 samples, eight samples exhibited zero allele difference, while three samples showed one allele difference, and one sample showed two allele differences, suggesting that a two-allele difference could be used as the threshold for duplicate identification. Based on this threshold, we found that that half of the clones had at least one other clone with up to a two-allele difference ( Table S4 ). We assessed how many SNP markers are needed to provide sufficient statistical power for sugarcane duplicate identification. Based on the cumulative PID-sib values for each species, we demonstrated that when utilizing 48 SNPs, the probability that two sibling individuals may share the same multilocus genotype by chance (Waits et al., 2001) was smaller than 0.001 (PIDsib <0.001) ( Table S8 ; Figure S1 ). Therefore, the panel of 357 SNPs is far more than sufficient to identify synonymously mislabeled clones in each species ( Table S5 ; Figure S1 ).

In addition to the detection of synonymous groups, the single-dose SNP genotyping enabled the identification of 27 homonymous mislabeling groups ( Table 2 ), where clones with the same name had different SNP genotypes. These homonymous mislabeling groups were detected in all five Saccharum species, often in germplasm accessions collected from the same geographical region. For example, the three clones of “AGOULE” were all collected from Tamil Nadu, India. However, they exhibited two different SNP genotypes. In another case, two clones were labeled as “Chino”, both from Hawaii, but their SNP genotypes were different. A more noteworthy example is the two “Uba” clones from India. Although they were both classified correctly as S. sinense, they had different SNP genotypes. Since “Uba” has been widely used as an important source of disease resistance in sugarcane breeding, the identified homonymous mislabeling has significant implications on sugarcane genetic studies and new variety development.

Furthermore, a high rate of mislabeling and inaccuracies was also detected in recorded species membership. Most of the mislabeling and inaccuracies occurred in species pairs that shared morphological similarities, such as S. officinarum vs S. robustum and S. barberi vs S. sinense. In contrast, there was almost no mislabeling of species membership between S. spontaneum and the rest of species. The high rate of mislabeling revealed in the present study was likely due to sugarcane’s feature as a clonally propagated crop, which has allowed the exchange of sugarcane germplasm as clones among regions, countries, and continents. However, passport data, such as records and labels of the germplasm have not always followed the same naming conventions, leading to limited information about their correct identity. In fact, the majority of mislabeling and redundancy were observed in cultivated species, indicating a more intensive exchange of cultivated germplasm than wild species. Additional efforts of characterization are needed to fully resolve the mislabeling problem. SNP profiles for reference sugarcane clones need to be established through international collaboration. The putative mislabeled clones need to be compared with established references to correct the mislabeling errors. For the putative duplicate groups with near-identical genotypes, SNP genotyping will need to be repeated to confirm their clone identity. Moreover, since somaclonal mutation can occur in sugarcane, phenotypic examination remains essential to complement the result of molecular characterization.

The sugarcane research community usually regarded Saccharum (sensu stricto) as containing six species, including two wild species (S. spontaneum and S. robustum), and four cultivated species - S. officinarum, S. edule, S. barberi, and S. sinense (Purseglove, 1972; Daniels and Roach, 1987; Grivet et al., 2006; Hemaprabha et al., 2022). The present study, using both model-based clustering and multivariant analysis based on 357 single dose SNP makers, supported the current classification of Saccharum germplasm. The only exception is S. edule, which could not be differentiated from S. robustum. S. edule is cultivated in New Guinea and nearby islands for its aborted edible inflorescences. Our result is consistent with the hypothesis that S. edule is a small group of sterile mutants that originated from S. robustum (Purseglove, 1972; Grivet et al., 2004; Grivet et al., 2006).

The present study also showed that S. spontaneum is a well-differentiated wild species, as shown by the analytical results of STRUCTURE ( Figure 3 ), PCoA ( Figure 5 ) and the NJ-tree ( Figure 6 ). This observation is consistent with previous reports based on SSR markers (Brown et al., 2007), genome re-sequencing data (Yang et al., 2019; Fickett et al., 2020), and plastid genome sequences (Evans and Joshi, 2016). In the present study, we found very few mislabeling between S. spontaneum and the other four species. Nonetheless, the result of the hierarchical STRUCTURE analysis on S. spontaneum showed that there were two sub-groups within this species ( Figure 4A, B ). The first sub-group was mainly from India and south Asia, whereas the second sub-group was dominantly from Southeast Asia countries ( Table S6 ). This result agrees with the recent report of Pompidor et al. (2021) and further indicates the importance of maintaining differentiated populations based on broader geographical regions.

A close genetic relationship was observed between S. robustum and S. officinarum. This observation is consistent with the recent finding of Pompidor et al. (2021), which suggested that both S. officinarum and S. robustum were derived from the same two ancestral genomes (A and B genomes), indicating a common origin of both species. Nonetheless, our result showed that the two species could be clearly differentiated at the molecular level, as demonstrated by the results of STRUCTURE ( Figure 4B ), PCoA ( Figure 5 ) and NJ tree ( Figure 6 ).

S. barberi and S. sinense are two cultivated species that are closely related (Purseglove, 1972; Lu et al., 1994; Hemaprabha et al., 2022). However, the taxonomy status, as well as the relationship between these two species, has been a subject of debate. It was proposed that S. officinarum hybridized with S. spontaneum in Asia continental and the hybrid progenies developed into S. barberi in India and S. sinense in China (Brandes, 1956; Daniels and Roach, 1987; D’Hont et al., 2002; Li et al., 2022). The geographical barrier (Southern China for C. sinensis vs. Northern India for C. barberi) could played significant role in the genetic differentiation of these two species. Using target enrichment sequencing of 307 germplasm accessions from WCSRG, Yang et al. (2019) showed that S. sinense and S. barberi were different in terms of genome compositions and potential ancestor accessions. Our result showed that S. barberi had the closest relationship with S. officinarum, which supported the proposal that S. barberi is a hybrid of S. officinarum and S. spontaneum. However, the present result also showed that relative to S. barberi, S. sinense had a larger genetic differentiation from S. officinarum and S. spontaneum ( Table 3 ). Nonetheless, the number of samples of S. barberi and S. sinense used in the present study is relatively small. A systematic collection of samples representing the full geographical range of these two species is needed for a comprehensive analysis of population structure and genetic diversity in these two species.

To elucidate the underlying patterns of genetic variation in the sugarcane population, we conducted AMOVA to partition the observed variation among sugarcane clones. According to the AMOVA results, a significant proportion of the observed genetic diversity was attributed to variation among individual clones, accounting for 85% of the total variation, while the remaining 15% of the variation was attributed to variation between species ( Table 4 ). These results are consistent with previous studies (Nayak et al., 2014; Manechini et al., 2018; Fickett et al., 2020; Singh et al., 2020), indicating that the primary source of genetic diversity resides within species rather than between different species. The relatively low percentage of genetic variation between species suggests that there may be a considerable level of introgression and systematic crossing occurring between different sugarcane species, reflecting intercrossing nature among sugarcane species (Moore et al., 2013). In all species, the observed heterozygosity was higher than the expected heterozygosity, which suggests more intercrosses between isolated populations than the founders. The inter-specific gene flow in sugarcane is well-supported by historical accounts that, throughout the seventeenth, eighteenth, and nineteen centuries, there was an extensive exchange of varieties among the sugarcane planters worldwide (Warner, 1962). The development of improved cultivars involved frequent intercrossing between the species, which also likely contributed to the low genetic variation between species. This is supported by the observed heterozygosity exceeding Hardy-Weinberg expectations, suggesting a greater number of intercrosses than expected from the founders alone.

In conclusion, accurate germplasm identity is critical for efficient management of sugarcane germplasm. Using single-dose SNP markers, we genotyped all the Saccharum clones in WCSRG, maintained at USDA-ARS. The single-dose SNP genotypes enabled us to detect a high rate of mislabeling and genetic redundancy in this collection. In addition, an analysis of population structure using both ordination and model-based clustering, revealed five genetic groups in the Saccharum germplasm, corresponding to S. barberi, S. robustum, S. officinarum, S. sinense, and S. spontaneum. The pattern of genetic structure in the Saccharum gene pool suggested a high level of gene flow among sugarcane species and across different geographical regions, likely facilitated by human intervention, as evident from the lower genetic variation observed between species. This extensive germplasm exchange, predominantly as clonal material, may contribute to the mislabeling and redundancy observed within the sugarcane collections. Through comprehensive analyses of genetic identity, we were able to detect genetic redundancy in the collection. Furthermore, we assessed the ancestral species/populations among the Saccharum germplasm clones and ascertained the presence of core members in each species. Using these core members as references, we were able to correct mistakes and/or inaccuracy in species membership, as well as clarify the parentage for hybrid clones. The corrected mislabeling in species membership needs to be validated based on phenotypic characteristics. Our results ensured that the preserved clones in the WCSRG have distinct genetic makeup. This is the first time that a large germplasm collection of sugarcane—a complicated polyploidy crop - was systematically characterized in terms of clone integrity and genetic redundancy. The single-dose SNP markers developed by this study offer a powerful tool for characterizing sugarcane germplasm worldwide. These markers can also be potentially used for identifying chromosomes. The reported findings have important implications for sugarcane genebank management, germplasm exchange, and crop genetic improvement.