Chen et al.

BMC Biology (2025) 23:327 BMC Biology

[Link]

RESEARCH Open Access

A novel method for effectively selecting

fragments not associated with restriction sites
for whole‑genome genotyping
Peng Chen1†, Bipei Zhang2†, Sheng Zhao1,3, Zhenghang Zhu1, Yiming Yan1, Haotian Chen1, Hong Lu1,
Yong Xiang1, Yongquan Li2* and Yuxiao Chang1*

Abstract
Background Reduced representation sequencing (RRS) using Restriction site-associated DNA sequencing (RAD-seq)
has become a widely adopted method for whole-genome genotyping, owing to its cost-effectiveness and applicabil-
ity across species. However, traditional RAD-seq approaches face challenges, including complex workflows and high
labor demands. Existing RAD-seq methods rely primarily on the strategy of “selecting fragments first, then preparing
a library,” which involves targeting fragments associated with restriction enzyme cut sites.
Results In contrast, we developed inverse RAD-seq (iRAD-seq), a novel method that employs a “prepare library first,
then select” strategy to efficiently capture fragments not associated with restriction sites for whole-genome genotyp-
ing. This innovative approach utilizes Tn5 transposase to simultaneously fragment DNA and ligate adapters, followed
by pooled processing of hundreds of libraries for batch restriction digestion, thereby significantly streamlining RAD-
seq library preparation. This simplified workflow makes iRAD-seq highly compatible with liquid handling automation,
enhancing its throughput. We validated iRAD-seq through both in silico analyses and wet-lab experiments in maize
and rice. The results demonstrate that iRAD-seq provides consistent genome-wide Single-Nucleotide Polymorphism
(SNP) distributions. In maize germplasm, iRAD-seq successfully enables effective genetic diversity analysis. Further-
more, genetic mapping of maize populations has confirmed its utility for identifying quantitative trait loci (QTL).
Conclusions Our findings suggest that iRAD-seq offers a more streamlined, efficient, and flexible approach to geno-
typing and marker development than traditional RAD-seq methods. Owing to its potential for high-throughput
applications and cost reductions, iRAD-seq is a valuable tool for genomic research and molecular breeding.
Keywords Reduced representation sequencing, Whole-genome genotyping, Molecular breeding, Restriction site-
associated DNA sequencing, Tn5

†
Peng Chen and Bipei Zhang contributed equally to this work.
*Correspondence:
Yongquan Li
yongquanli@[Link]
Yuxiao Chang
changyuxiao@[Link]
Full list of author information is available at the end of the article

© The Author(s) 2025. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit [Link]
Chen et al. BMC Biology (2025) 23:327 Page 2 of 16

Background The RRS strategy contains two essential steps, as dem-

Genome-wide single nucleotide polymorphism (SNP) onstrated in traditional RAD-seq and its derivatives: one
discovery and genotyping are essential for breeding is how to “select” the fractional representative DNA frag-
research [1–3]. However, for many projects, genotyping ments, and the other is how to add adapters to the frag-
via whole genome sequencing is expensive and gener- ments for sequencing. Although these RAD-seq methods
ates more SNPs than needed. Therefore, the strategy differ in the type of REs and ligated adapters used, they
of reduced representation sequencing (RRS), which all prioritize selecting regions adjacent to the RE recog-
selects only the partial genome (usually 1–10%) for nition site before preparing the sequence library. The
sequencing [4–6], is more popular for genome-wide step of adapter ligation by T4 ligase during the library
molecular marker detection in breeding projects. preparation process is time-consuming, labor-intensive,
Restriction site-associated DNA sequencing (RAD- and inefficient. As a result, RAD-seq methods also suffer
seq) is a classical RRS method. In the original RAD- from these drawbacks.
seq protocol, genomic DNA is digested with a RE and An alternative method for adapter ligation that is
ligated with the P1 adapter containing a barcode; then, simple and highly efficient involves the use of the Tn5
the barcoded fragments are pooled, randomly sheared, transposase, which randomly cleaves genomic DNA and
and size-selected; after that, the P2 adapters are ligated ligates adapters [13–15]. Previously, we reported the
to the pooled fragments. Finally, the fragments with All-in-one sequencing (AIO-seq), an improved library
both P1 and P2 adapters are amplified and sequenced preparation method with Tn5. With this method, hun-
[7]. Although this RAD-seq method has proven to be dreds of samples can be pooled after PCR for size selec-
an effective method for selecting partial genomes and tion and quantification [16]. On this basis, we propose a
has been used for genetics and breeding research in novel RRS strategy called Inverse Restriction site Asso-
various species, its library preparation process has been ciated DNA sequencing (iRAD-seq). This represents an
too tedious, and efforts to improve it never stopped [7, improvement over current RAD-seq methods. Unlike the
8]. conventional RAD-seq strategy, which selects fragments
For example, double digest-RAD (dd-RAD) elaborates first and then prepares the library, our iRAD-seq follows
the original RAD-seq by eliminating random shearing an inverse approach: preparing the library first and then
and explicitly using size selection, hence only regions selecting the fragments. The introduction of Tn5 enables
adjacent to restriction enzyme (RE) recognition sites are one-step fragmentation and adapter addition, simplifying
chosen for sequencing [9]. More than 90% of REs belong the library construction process and saving time [13–15].
to the type IIP subtype, which recognizes and cleaves Our iRAD-seq method, which is based on this strategy
symmetric (or “palindromic”) DNA sequences 4 to 8 base and combined with our AIO-seq method, offers a more
pairs in length. In contrast, Type IIB REs cleave DNA effective and easier method for RRS library preparation.
upstream and downstream ~ 7–15 bp away from their
specific recognition site and generate ~ 30 bp DNA frag- Results
ments with recognition sites in the central region [10]. The iRAD‑seq workflow
2b-RAD uses the type IIB restriction endonucleases to Our goal was to develop a novel RAD-seq method capa-
digest genomic DNA and recover fragments of uniform ble of identifying genome-wide SNP markers, utilizing
length for library preparation and sequencing [11]. Simi- the AIO-seq method [16] and RE reduced representa-
lar improvements in RAD-seq also include ezRAD [12]. tion sequencing. Our iRAD-seq protocol involves the
However, the original RAD-seq method and its related following steps: initially, a whole-genome resequencing
derivatives are still based on a tedious multistep separa- library with a unique dual-index was prepared for each
tion and purification process, which usually takes a long sample using Tn5 transposase, following the AIO-seq
time and has high labor costs when dealing with large protocol; Subsequently, the dual-indexed libraries were
cohorts of samples. In addition, the ezRAD and ddRAD pooled and subjected to enzymatic digestion with a panel
techniques incorporated size selection after restriction of REs; Fragments containing the respective restriction
enzyme digestion. As designed, these methods exclu- sites were cleaved into smaller pieces and filtered by size
sively perform size selection at the single-sample level, selection. Following digestion, fragments ranging from
which may introduce inter-sample variability in fragment approximately 430 bp to 780 bp (including the insertion
recovery efficiency, potentially leading to a significant and 136 bp adaptors) were size-selected for sequenc-
loss of markers. As this selection is random, experimen- ing (Fig. 1A and B). Despite some fragments fell within
tal repeatability is poor, making it difficult to control the this length range after enzyme cleavage, they could still
number of makers [8]. Therefore, fast, cost-effective and not be sequenced due to the absence of the necessary P5
high-throughput genotyping techniques are needed. and P7 adapters required for sequencing. By employing
Chen et al. BMC Biology (2025) 23:327 Page 3 of 16

this design, genomic fragments devoid of the utilized and 1.8%, respectively (Fig. 1F). However, the addition of
REs’recognized sites, with lengths ranging from approxi- HinP1I (GCGC), DpnII (GATC), and AluI (AGCT) sig-
mately 300 bp to 650 bp, were selected and sequenced to nificantly affected genome reduction, particularly DpnII
represent the genotyping of the entire genome (Fig. 1B). and AluI, which reduced the final genome coverage to
15.0% and 13.8%, respectively (Fig. 1F). The patterns of
The selection of panel of REs and fragment lengths soybean, maize, and wheat were similar (Fig. 1C, E and
impacts the efficient of reduced representation Additional file 1: Figs. S2A-S2D). These findings suggest
The degree of genome representation (OR simplification) that our iRAD-seq method is effective, and that various
of iRAD-seq can be estimated based on in silico digestion panels of REs can generate different degrees of genome
of the reference genome. We selected six REs (MseI, MspI, simplification that can meet diverse requirements for
AluI, DpnII, HindIII, and HinP1I) to form five panels marker density.
(MseI + MspI, MseI + MspI + AluI, MseI + MspI + DpnII, We then evaluated the uniform distribution of the
MseI + MspI + HindIII, and MseI + MspI + HinP1I), which selected restriction fragments across the genome via
underwent in silico digestions of various species includ- in silico digestion. We first calculated the rice genomic
ing maize (Zea mays L.), rice (Oryza sativa L.), soybean fragment distribution selected by different panels of REs
(Glycine max L.), wheat (Triticum aestivum L.), pig (Sus across the genome in a 10-kb sliding window and found
scrofa) and cow (Bos taurus) (Additional file 8: Table S1). that the distribution of selected restriction fragments
We anticipate that the iRAD-seq method, which uti- throughout the genome was widespread (Fig. 1G-K
lizes restriction enzymes recognizing 4–6 nucleotide and Additional file 1: Fig. S3). We subsequently calcu-
sequences, will be employed. Following this, DNA frag- lated the gap length between two fragments (> 300 bp)
ments sized 300–650 bp will be selected through sorting. for the genomes of rice, maize, pig and cow digested by
Among these, fragments containing intact paired-end MseI, MspI and AluI. The results revealed that gaps were
sequencing adapters (uncut by the enzyme) will be mainly concentrated in regions below 50 kb (Fig. 1L-O);
sequenced. These fragments are suitable for paired-end for rice, 22 gaps longer than 50 kb were detected, and the
150 bp sequencing, covering around 10%−20% of the total length of the gaps was 7,952,540 kb, which occupied
genome, thus can reduce the genomic data by 80%−90%, 2.0% of the total genome (Fig. 1L). Compared with those
significantly streamlining the sequencing process and in rice, the number of gaps longer than 50 kb and the
increasing the efficiency of downstream functional total length of the gaps were lower in maize, pig and cow
genomics and breeding research. (Fig. 1M-O). To some extent, these findings reveal that
As anticipated, different panels of REs can yield varying the distribution of fragments in the genome is relatively
degrees of representation (Fig. 1C-F and Additional file 1: widespread and that iRAD-seq can meet the require-
Fig. S1A). Specifically, for rice, digestion by the panels of ments of genotype imputation analysis. Taken together,
MseI (recognizing TTAA) and MspI (CCGG) produced our novel strategy of selecting fragments not associated
282,747 and 145,146 fragments longer than 300 and 400 with restriction sites can be used as an alternative for
bp, respectively (Fig. 1D), corresponding to 33.1% and reduced representation.
21.0% genome coverage (Fig. 1F). In addition to MseI
and MspI, the inclusion of HindIII (AAG​CTT​) decreased Proof‑of‑concept of iRAD‑seq demonstrated with maize
the number of fragments longer than 300 and 400 bp and rice
by only 4.3% and 7.0%, respectively (Fig. 1D), resulting To test the feasibility of our iRAD-seq method in a wet
in corresponding decreases in genome coverage of 2.0% lab, we used rice (Nip) and maize (B73) as tests. We

(See figure on next page.)

Fig. 1 Overview and in silico digestion evaluation of iRAD-seq. A Schematic of the iRAD-seq method. After preparation of the AIO-seq library
(I), the fragments were digested using REs, and the digestion products were sorted (II) and then sequenced on the next-generation sequencing
platform (III, IV). The colors in I, II, and III are in a corresponding relationship. B The detailed experimental process of iRAD-seq. Extraction of DNA
from samples (I). Genomic DNA was digested with Tn5 transposase and the sequencing adapter was added to the fragments by PCR (II).
Fragments were digested with 2–3 REs, and the digested fragments could not be sequenced (III). The digestion products were sorted and then
sequenced on the next-generation sequencing platform (IV). Steps I-IV represent identical experimental workflow in both A and B. C-D The
number of fragments of different lengths in maize and rice. E–F Genome coverage analysis of fragments of different lengths in maize and rice.
G-K Chromosome distribution of fragments (length > 300 bp). The rice genome was digested by different panels of REs. The size of sliding window
was 10 kb. Red, high density; green, low density. L-O Histogram of the number of gaps in rice, maize, pig (susScr11), and cow (bosTau9). MseI,
MspI, and AluI were selected for in silico digestion of the genome. The gap length in the genome is defined as the distance between fragments
(length > 300 bp)
Chen et al. BMC Biology (2025) 23:327 Page 4 of 16

Fig. 1 (See legend on previous page.)

initially prepared the AIO-seq library, which was sub- for ​30%−50% genome reduction (Fig. 1E-F). Then, we
sequently subjected to RE digestion using two dis- selected fragments range from 430 to 780 bp for sequenc-
tinct panels: Panel_A (AluI, MseI, MspI) for ​10%−20% ing (Fig. 1B). Four technical replicates were carried out
genome reduction and Panel_B (HindIII, MseI, MspI) for each panel of REs. Sequencing data of 4.2–4.8 Gb
Chen et al. BMC Biology (2025) 23:327 Page 5 of 16

were obtained for the maize library, and 2.5–3.0 Gb for majority of the reads were located in the region without
the rice library (Additional file 8: Table S2). For the clean the REs recognition site (Fig. 2A-C); occasionally, we
data, we first calculated the ratio of reads that still con- could find high-depth reads in the region containing the
tained the REs’ recognition sites used and found that recognition site, but further analysis revealed that SNPs
approximately 0.8%−7.0% of the reads contained these occurred in the test variety compared with the reference,
recognition sites. Four technical replicates showed simi- thus losing this recognition site (Fig. 2D). These results
lar result, indicated it maybe the intrinsic property of the showed that the iRAD-seq method effectively reduces
REs (Additional file 2: Figs. S4A-S4H). Then, we mapped genome representation.
the reads to the reference and determined the read dis- To assess the coverage and reproducibility of the tar-
tribution via the IGV genome browser. As expected, the get genomic regions of iRAD-seq, we removed reads

Fig. 2 Relationship between sequencing reads and REs recognition sites in the genome. A-C The reads of iRAD-seq are primarily enriched
in non-REs recognition regions (> 300 bp). D Reads with mutations in RE recognition sites are enriched in non-RE recognition regions (> 300 bp). The
data from rice by iRAD-seq with MseI, MspI and HindIII. The black vertical lines indicate the amplified regions. Red boxes indicate bases with mutated
of REs recognition sites in reads
Chen et al. BMC Biology (2025) 23:327 Page 6 of 16

containing recognition sites of REs and fragments shorter lines), “Second Cycle inbred lines from USA” (38 lines)
than 300 bp (noise background). We counted the genome and “Sweet corn” (55 lines) (Fig. 3M). Moreover, we per-
coverage using different amounts of data. The wet-lab formed principal component analysis (PCA) base on
sequencing data of iRAD-seq exhibit the coverage that is SNP. The samples were also divided into three clusters,
comparable to the results of in silico digestion (Figs. 1E-F exhibiting a clustering pattern identical to the three main
and 3A-D). For the data from Panel_A in maize, with 2.0 clusters found in the phylogenetic analysis. (Fig. 3M-
Gb of sequence data, the genome coverage was 11.3%, N). Overall, the clear classification of 139 maize lines
increasing the sequencing data to 4.0 Gb and increasing revealed the ability of iRAD-seq to be used in germplasm
the genome coverage to only 14.3%, which was compara- genetic background investigations.
ble to in silico digestion prediction of coverage of 18.8%
(Figs. 1E and 3A). The digestion by Panel_B in maize and
rice followed a similar pattern (Figs. 1E-F, 3B and D). The application of iRAD‑seq in maize genetics research
After digestion by Panel_A in rice, the actual genomic To test the ability of iRAD-seq for genetics research, we
coverage at a data volume of 4.0 Gb was 16%, whereas perform iRAD-seq with Panel_A and Panel_B for the
the result of in silico digestion was 14% (Figs. 1F and 3C). maize ­BC1F4 population of CIMBL83/GEMS41 and the
We speculate that this discrepancy may be due to muta- ­BC2F4 population of CML496/GEMS41. Since the previ-
tions occurring at the recognition sites of REs within the ous results have showed that approximately 4.6%−7.0% of
sample (Fig. 2D). In summary, the wet-lab data confirmed the reads still contained the RE recognition sites (Addi-
that the iRAD-seq method effectively selects fragments tional file 2: Figs. S4A and S4B), thus we performed
devoid of recognition sites of REs for genome reduced two rounds of digestion for the Panel_A. For each line,
representation sequencing. Moreover, in silico diges- we obtained approximately 2.2 ± 0.8 Gb (Panel_A) and
tion can be used to guide the selection of RE panels for 2.3 ± 0.8 Gb (Panel_B) of sequencing data (Additional
iRAD-seq. file 3: Figs. S8A, S8B; Additional file 8: Table S5). We
Finally, we evaluated the reproducibility among the also sequenced the two populations with low coverage
four technical replicates. For maize (2.0 Gb) and rice whole genome resequencing by AIO-seq and obtained
(2.5 Mb), the Spearman correlation coefficients ranged approximately 2.8 ± 1.2 Gb of sequencing data for
from 0.88 to 0.95 (Fig. 3E-3 and Additional file 2: Fig. each line (Additional file 3: Figs. S8A, S8B; Additional
S5). Additionally, SNP overlap across the replicates was file 8: Table S5). We also sequenced the three parents
78.1% (Panel A) and 77.6% (Panel B) for maize, and 64.7% (CIMBL83, CML496, and GEMS41) and obtained 33.2
(Panel A) and 69.1% (Panel B) for rice. The SNP overlap Gb, 24.1 Gb and 26.3 Gb raw bases respectively.
was positively correlated with the data volume (Fig. 3I-L For these data, we first analyzed the RE digestion effi-
and Additional file 2: Fig. S6). Collectively, these results ciency. On average, across all lines of CIMBL83/GEMS41
highlight the consistent stability and high reproducibility and CML496/GEMS41, 93.5% ± 0.7% of the reads in
of the iRAD-seq method. Panel_A and 24.8% ± 0.4% of the reads in AIO-seq did
not contain the recognition sites of the REs of Panel_A
The application of iRAD‑seq in maize germplasm (Figs. 4A and B). Similarly, 89.6% ± 1.4% of the reads in
for phylogenetic analysis Panel_B and 40.9% ± 0.4% of the reads in AIO-seq did
In plant breeding, assessing the genetic background of not contain the recognition sites of the REs of Panel_B
germplasms helps understanding genetic diversity and (Fig. 4C and D). Panel_A and Panel_B showed reduc-
population structure, providing guidance for subsequent tions of 68.7% ± 0.6% and 48.7% ± 1.4%, respectively,
hybridization and selection strategies. To test the per- in the number of reads containing recognition sites of
formance of iRAD-seq in plant germplasm phylogenetic REs. Furthermore, we analyzed the digestion efficiency
analysis, we investigated the population genetic struc- of individual REs. CIMBL83/GEMS41, with averages of
ture of 139 maize germplasms by iRAD-seq with Panel_A 98.7% ± 0.09%, 94.2% ± 0.9%, and 67.0% ± 1.3% of the reads
(AluI, MseI, MspI). We obtained 106.3 Gb (0.6 ± 0.2 Gb in Panel_A, Panel_B, and AIO-seq, respectively, did not
per line) of clean data (Additional file 2: Fig. S7A; Addi- contain MseI recognition sites (Fig. 4E). A similar pattern
tional file 8: Table S3). From this dataset, we identified was observed for MspI, AluI, and HindIII (Fig. 4F and
a total of 1812 high-quality SNPs with a missing rate of Additional file 3: Figs. S9A-S9D). These results showed
less than 20% and a minor allele frequency (MAF) greater that iRAD-seq had the remarkable ability to significantly
than 0.05 (Additional file 8: Table S4). These SNPs are increase the proportion of reads that did not contain rec-
widespread in the genome (Additional file 2: Fig. S7B). ognition sites (P < 0.001). Furthermore, the percentage
An unroot neighbor joining tree showed that the 139 of target reads (without recognition sites of REs) from
samples were divided into three clusters: “Iodent” (46 Panel_A was significantly greater than that from Panel_B
Chen et al. BMC Biology (2025) 23:327 Page 7 of 16

Fig. 3 Experiment of iRAD-seq and its utility for maize population structure analysis. A-D Genome coverage of maize and rice by varying
sequencing data sizes. Panel_A (AluI, MseI and MspI) and Panel_B (HindIII, MseI and MspI) were selected for iRAD-seq. E–H Consistency of coverage
among 4 repeats when using varying sequencing data sizes. Panel_A was used for iRAD-seq. Pairwise Spearman correlation of the mean coverage
between replicates with 100 kb windows. I-L Number of shared SNPs among 4 repeats when using varying sequencing data sizes. Panel_A
was used for iRAD-seq. M Neighbor-joining tree of the 139 maize lines based on high-quality SNPs. N Principal component analysis of 139 maize
lines
Chen et al. BMC Biology (2025) 23:327 Page 8 of 16

Fig. 4 Efficiency of restriction endonucleases and simplified efficiency of iRAD-seq. A-B Proportion of reads without RE recognition sites
in the CIMBL83/GEMS41 population (A) and CML496/GEMS41 population (B) with Panel_A. C-D Proportion of reads without RE recognition sites
in the CIMBL83/GEMS41 population (C) and CML496/GEMS41 population (D) with Panel_B. E–F Proportion of reads without MseI recognition
sites in the CIMBL83/GEMS41 population (E) and CML496/GEMS41 population (F) with Panel_A, Panel_B and AIO-seq. G-H Proportion of reads
without MspI recognition sites in the CIMBL83/GEMS41 population (G) and CML496/GEMS41 population (H) with Panel_A, Panel_B and AIO-seq. I-L
Genome coverage of iRAD-seq data when using varying sequencing data sizes. M The number of SNPs using varying sequencing data sizes. N SNP
number at different depths. Log10(n + 1): n is the number of SNPs. For boxplots, the center line is the median; box limits represent upper and lower
quartiles. Employ the t-test to assess differences in different method, p < 0.0001. Different methods use different REs (Panel_A: MseI, MspI and AluI;
Panel_B: MseI, MspI and HindIII; AIO-seq: none)
Chen et al. BMC Biology (2025) 23:327 Page 9 of 16

(P < 0.001, Fig. 4E-H), indicating that the addition of the better than that in AIO-seq. This result indicates that
second round of RE digestion was effective. iRAD-seq exhibits better stability with a low dataset. The
To evaluate the efficiency of reduced representation SNPs identified from the three methods were evenly dis-
using iRAD-seq, we randomly extracted 0.5, 1.0, 1.5, and tributed throughout the genome (Additional file 4: Fig.
2.0 Gb from the datasets across three strategies in eight S11, Additional file 5: Fig. S12 and Additional file 6: Fig.
individuals. We then analyzed whether these reads were S13). Collectively, these results suggest that iRAD-seq
enriched in specific regions, as indicated by read cover- can effectively reduce genome representation, and that
age along the chromosomes. With the 0.5 Gb dataset, different REs can lead to varying degrees of reduction.
the genome coverage with a read depth of at least 1 × was The efficiency of reduced representation appears to cor-
16.6% ± 0.6% for AIO-seq, decreasing to 6.5% ± 0.2% for relate positively with the results of in silico digestion.
Panel A and 11.7% ± 0.2% for Panel B (Fig. 4I). In contrast, Across all lines of CIMBL83/GEMS41 and CML496/
for regions covered with at least 3 × depth, the percent- GEMS41, we identified 6,582,817 SNPs between the
age for iRAD-seq was greater than that for AIO-seq. This parents of CIMBL83/GEMS41 and 6,662,089 SNPs
trend was consistent across different data volumes from between the parents of CML496/GEMS41. On average,
0.5–2.0 Gb, with the differences between iRAD-seq and iRAD-seq with Panel_A, Panel_B, and AIO-seq yields
AIO-seq becoming more pronounced as the data volume 154,901 ± 32,878, 289,627 ± 67,365 and 386,207 ± 173,087
increased (Fig. 4I-L). Between Panel_A and Panel_B, the SNPs per line in CIMBL83/GEMS4 and 127,874 ± 32,039,
efficiency of the reduced representation of Panel_A is 223,823 ± 41,585 and 260,258 ± 84,135 SNPs in CML496/
greater than that of Panel_B (Fig. 4I-L). This result is con- GEMS41 (Fig. 5A, B and Additional file 8: Table S5).
sistent with the results of the digestion efficiency analysis After genotyping, with these SNPs, we obtain 4465
(Fig. 4E-H). These results indicated that iRAD-seq exhib- (per lines 23), 4794 (25), and 4616 (24) breakpoints in
ited a significant enrichment effect compared with whole CIMBL83/GEMS41 and 1137 (17), 1272 (19), and 1278
genome resequencing, and that different RE panels can (19) in CML496/GEMS41 (Fig. 5C and Additional file 7:
achieve a controllable and predictable level of genome Figs. S14A-S14E). We detected 2261, 2785, and 3193 bin
representation. markers in CIMBL83/GEMS41 and 773, 958, and 1131
In line with this, the number of SNPs increased with in CML496/GEMS41 (Fig. 5D, E and Additional file 8:
more sequencing data, and as the efficiency of reduced Table S6). Using bin markers, we constructed linkage
representation improved, the rate of increase decreased maps. We obtained 1027.5 cM, 1315.8 cM and 1314.8
(Fig. 4M). This is because increasing amounts of data cM genetic maps for CIMBL83/GEMS41 and 1299.1 cM,
are more easily enriched in simplified genomic regions. 1505.7 cM and 1682.1 cM genetic maps for CML496/
Consequently, iRAD-seq can obtain more high-quality GEMS41 (Fig. 5F, Additional file 7: Figs. S14F-S14J and
SNPs with a low data volume. Furthermore, analysis of Additional file 8: Table S6). Finally, we conducted quan-
the depth of the SNP loci revealed a similar distribu- titative trait locus (QTLs) mapping in combination with
tion pattern to that of coverage (Fig. 4I-L, N and Addi- phenotypic (Additional file 7: Figs. S15A and S15B), and
tional file 3: Figs. S10A-S10C). With the 1.0 Gb dataset, the results could be used to map the same main-effect
iRAD-seq showing significantly greater number of SNPs QTLs (Fig. 5G-L and Additional file 8: Table S7). For
after > 2 × and Panel_A significantly surpassed Panel_B example, in the CIMBL83/GEMS41 population, we suc-
after > 4 × (Fig. 4N). Both the 0.5 Gb、1.5 Gb and 2.0 Gb cessfully mapped a major QTL on chromosome 3, which
datasets showed similar patterns (Fig. 4N and Additional included the previously identified lg2 gene (Fig. 5G-I)
file 3: Figs. S9B and S9C). This result suggests that iRAD, [17]. In CML496/GEMS41, we mapped two major QTLs
relying on high-depth SNPs, can provide more accurate on chromosome 2, which included the previously identi-
genetic analysis. In addition, with the 0.5 Gb dataset, fied ibh1 and lg1 genes (Fig. 5J-L) [18, 19]. Additionally,
the stability of the SNP quantity in iRAD is significantly iRAD-seq identified a QTL that was not identified by

(See figure on next page.)

Fig. 5 Application of iRAD-seq for maize genetics research. A-B Number of SNPs identified by different sequencing strategies in the CIMBL83/
GEMS41 population (A) and CML496/GEMS41 population (B). For boxplots, the center line is the median; box limits represent upper and lower
quartiles. C Construction of a recombination bin map for the CIMBL83/GEMS41 population with Panel_A. D-E Number of bin markers in CIMBL83/
GEMS41 (D) and CML496/GEMS41 (E) by different sequencing strategies. F A high-density genetic map of CIMBL83/GEMS41 population
with Panel_A. G-L LOD values of the quantitative trait locus (QTL) results for leaf angle (LOD value > 3) by different sequencing strategies
in the CIMBL83/GEMS41 population (G-I) and CML496/GEMS41 population (J-L). Different methods use different REs (Panel_A: MseI, MspI and AluI;
Panel_B: MseI, MspI and HindIII; AIO-seq: none)
Chen et al. BMC Biology (2025) 23:327 Page 10 of 16

Fig. 5 (See legend on previous page.)

Chen et al. BMC Biology (2025) 23:327 Page 11 of 16

AIO-seq, which included the previously identified na1 96 libraries. These pooled libraries are then digested with
gene in CML496/GEMS41 [20]. In general, the iRAD-seq restriction enzymes for the “fragment selection” step. In
method can perform accurate genotyping, high-density comparison, other RAD-seq methods can typically pool
genetic mapping and major QTL mapping. no more than 12 samples, or may not allow pooling at
all. This gives iRAD-seq a much higher throughput than
Discussion traditional RAD-seq methods. Third, although iRAD-seq
Genome-wide marker-assisted selection (MAS) breed- is fundamentally a resequencing library, it selects only
ing can significantly increase and accelerate the breed- approximately 10% of the whole genome. Consequently,
ing process. Among various whole-genome genotyping the bioinformatics analysis follows the same process as
methods, RAD-seq stands out as one of the most widely for whole-genome sequencing libraries, with the only dif-
used techniques due to its versatility. It is applicable to ference being that the data volume is reduced to around
a broad range of species, including non-model organ- 10%. This reduction simplifies the analysis process and
isms, without the upfront investment required for probe/ results in significant savings in computational resources.
primer design and synthesis in multiple PCR and tar- The improvements and innovations introduced by the
get enrichment methods [8, 21, 22]. RAD-seq, a form of iRAD-seq protocol have yielded significant benefits. Not
RRS, uses REs to reduce the complexity of the genome, only has the protocol reduced the consumption of rea-
resulting in cost-effective sequencing while still provid- gents, but its streamlined workflow also makes it suitable
ing high-quality data on genomic variation. However, for automation using liquid handling stations, thereby
for practical use in breeding programs, particularly in increasing throughput and reducing labor costs.
developing countries and for small- and medium-sized We conducted a rough comparison of the reagent costs
breeding companies, reducing per-sample costs remains for processing 96 samples using iRAD-seq versus tradi-
a key objective for the development of WGG methods. tional RAD-seq methods (Additional file 2: Table S8). For
RAD-seq and its derivatives employ various REs to digest traditional RAD-seq, the primary costs are associated
the genome, followed by sequencing library preparation with restriction enzyme digestion and adaptor ligation.
using adaptor ligation PCR to capture genomic sequences In contrast, the main cost for iRAD-seq comes from the
near RE sites [5, 11, 23]. Tn5 transposase-based library preparation. However, the
The RAD-seq protocol fundamentally involves select- pooling of samples prior to restriction enzyme digestion
ing a portion of the genome for library preparation. significantly reduces the need for restriction enzymes,
Typically, existing RAD-seq methods follow a two-step DNA purification, and library quantification in the sub-
process: “select fragments first, then prepare library.” sequent steps. Overall, while iRAD-seq relies on com-
In these methods, the focus is on the positive selection mercially available Tn5 transposase kits, which can be
of genomic fragments flanking the recognition sites of more expensive, the cost advantage over other RAD-seq
restriction enzymes for sequencing. In contrast, the methods may not be substantial for most labs. On the
iRAD-seq method we introduce here also selects a por- other hand, iRAD-seq uses only a single enzyme, Tn5
tion of the genome for library preparation, but differs in transposase, for both DNA fragmentation and adaptor
two important aspects. First, our approach involves “pre- ligation. Given the widespread use of Tn5 in the devel-
paring library first, then selecting”. Second, instead of opment of novel genomic research methods, several
positive selection, we employ a negative selection strat- protocols for expressing homemade Tn5 transposase are
egy that targets fragments not flanking the recognition already available [24–27]. Utilizing a self-made version of
sites of the restriction enzymes. Tn5 can reduce the cost of library construction by over
Our innovative iRAD-seq method offers two key 80%, significantly enhancing the cost-effectiveness of
advantages over traditional approaches, distinguished iRAD-seq, especially for laboratories handling large sam-
by its simple protocol and high throughput for library ple volumes (Additional file 8: Table S8).
preparation. First, iRAD-seq leverages the advantages The simplified experimental workflow of iRAD-seq
of the Tn5 transposase for genome fragmentation and makes it highly suitable for automation using a liquid
adaptor ligation, significantly reducing the process to handling workstation. We have established the complete
just a 10-min reaction with a single enzyme. This con- iRAD-seq protocol on a liquid handling system (Addi-
trasts with other RAD-seq methods that typically require tional file 1: Fig. S16). With our setup, 960 samples can
restriction enzymes for genome fragmentation and T4 be processed in two days by one person, from the leaf
ligase for adaptor ligation, which increases both hands- sample to the sequence-ready iRAD-seq library. When
on time and total experimental time. Second, following preparing NGS libraries using transposases, the distri-
Tn5-based library construction, each sample is tagged bution of DNA fragment sizes is sensitive to the ratio of
with a unique barcode, allowing for the pooling of up to transposase to input DNA. To achieve optimal library
Chen et al. BMC Biology (2025) 23:327 Page 12 of 16

fragmentation, several strategies have been proposed. representation, balancing marker density for diverse
The first and most straightforward strategy involves applications.
measuring the DNA concentration of the extracted iRAD-seq demonstrated practical utility in resolving
samples and then adding varying volumes of water to maize germplasm population structure and mapping
normalize the concentration across all samples. This agronomically relevant QTLs, underscoring its reliabil-
process can be cumbersome when handling large num- ity through high technical reproducibility. By integrating
bers of samples manually. A liquid handler with either cost-effective Tn5 protocols and scalable automation, it
individual channels or an independent 8-channel system addresses bottlenecks in large-scale genotyping, par-
can automate this step, but throughput remains limited, ticularly for resource-limited breeding initiatives. Col-
and some manual intervention is still required. During lectively, iRAD-seq bridges high-throughput sequencing
our iRAD-seq experiments, we employed an automated demands and agricultural genomics, offering a stream-
nucleic acid concentration analysis and normalization lined, efficient platform for molecular breeding and func-
workstation (Automatic Nucleic Acid Concentration tional genomics.
Analysis and Normalization Workstation, HC Scientific,
Chengdu, China) for DNA concentration normaliza- Methods
tion. This instrument is specifically designed for DNA Plant materials and DNA extraction
normalization during NGS library construction, using a The samples of maize (Zea mays L. cv. B73) and rice
stacker system that can complete concentration meas- (Oryza sativa ssp. japonica cv. Nipponbare) were
urements and normalization for 10 96-well plates in 260 obtained from the Agricultural Genomic Institute at
min, with no manual intervention required. This signifi- ShenZhen, Chinese Academy of Agricultural Sciences
cantly improved the efficiency of iRAD-seq library con- (AGIS, CAAS). A panel consisting 139 natural maize
struction. The second strategy involves engineering the lines, including sweet corn (55) and field corn from China
Tn5 transposase to enhance its tolerance to varying ratios and USA (84). These inbred lines were received as a gift
of input DNA and enzymes during tagmentation, thereby from Dr. Lu Hong (AGIS, CAAS). Two maize (Zea mays)
reducing the stringent need for DNA concentration nor- populations were obtained from a previous study. The
malization [26, 27]. The third strategy involves conjugat- first population, B
­ C1F4 (193 lines), was derived from a
ing Tn5 transposase directly to magnetic beads, enabling cross between GEMS41 and CIMBL83, followed by back-
it to bind a fixed amount of DNA and produce a normal- crossing to GEMS41 and selfing for four generations. The
ized, sequencing-ready library [28]. This approach is suit- second population, B ­ C2F4 (68 lines), was derived from a
able for a wide range of DNA input quantities and largely cross between GEMS41 and CML496, followed by back-
eliminates the need for the most cumbersome step of crossed to GEMS41, and selfed for four generations [29].
DNA quantification and normalization. We believe Genomic DNA was extracted from the leaves of each

oDrop2000 (Thermo Fisher Scientific) and Qubit™ 4.0

this approach is the most effective way to minimize the accession using the modified CTAB method [30]. Nan-
impact of the Tn5 to input DNA ratio on the final DNA
fragment distribution. If the iRAD-seq method we devel- Fluorometer (Invitrogen, NY, USA) were used to quantify
oped can eventually incorporate bead-linked trans- the DNA. DNA samples were inspected on 1% agarose
posomes, it will further decrease the dependence on gels for the presence of lower molecular weight DNA.
costly liquid handling workstations and streamline the DNA normalization utilizing MetricAnalyzer 3430S
experimental workflow. (Automatic Nucleic Acid Concentration Analysis and
Normalization Workstation, HC Scientific, Chengdu,
Conclusions China).
The development of iRAD-seq represents a significant
advancement in RRS methods for whole-genome geno- In silico digestions of Genome
typing. Leveraging Tn5 transposase for simultaneous The reference genomes of maize (B73_v4) [31], rice
DNA fragmentation and adapter ligation, this method (ZhenShan97) [32], soybean (Glycine_v2.0) [33], wheat
drastically simplifies library construction, reduces hands- (CS_v1.0) [34], cow (bosTau_v9) [35] and pig (susScr_
on time, and enhances automation compatibility. Vali- v11.1) [36] were downloaded from the NCBI database
dated in maize and rice, iRAD-seq achieves consistent (Additional file 2: Table S1). The Perl script fa2cmap_
genome-wide SNP distributions, with in silico predic- multi_color.pl from Solve-V3.5 [37] was used to find the
tions aligning closely with empirical data, confirming its recognition site of REs in the genomes. A self-written
precision in excluding restriction-associated fragments. Perl script find_fragments_length.pl was used to calcu-
Customizable enzyme panels enable flexible genome late the positions and lengths of fragments (a sequence
Chen et al. BMC Biology (2025) 23:327 Page 13 of 16

between two recognition sites of REs) in the genome that Reproducibility and reliability evaluation
exceed a threshold length. To ensure the reproducibility and reliability of iRAD-seq
method, we performed iRAD-seq on maize (B73) and
Library preparation of AIO‑seq and sequencing rice (Nip) using Panel_A (AluI, MseI, MspI) and Panel_B
(HindIII, MseI, MspI), with four biological replicates.
from the TruePrep® DNA Library Prep Kit V2 for Illu-
50 ng gDNA was digested with 2.5 μL Tn5 transposase
Using the sample module in seqkit [44], we randomly
mina® (Vazyme Biotech, Nanjing, China, Cat. TD501- subsampled sequencing data at different scales (maize:
02) for 10 min at 55 ℃ in an 8 μL reaction volume. These 0.5 Gb, 1.0 Gb, 2.0 Gb, and 4.0 Gb; rice: 0.25 Gb, 0.5 Gb,
fragment products were subjected to PCR amplification 1.0 Gb, and 2.0 Gb) for downstream analysis. We used
with 5 μL 10 × TAB buffer, 1 μL TAE, 3 μL PPM, 0.5 mM the bamqc tool in QualiMap [45] to analyze the distribu-
P5 and P7 primers in the same tube, TAB buffer and TAE tion of genome coverage. We performed sliding window
from the TruePrep Amplify Enzyme Kit (Vazyme Bio- statistics with a window size of 100 kb to calculate the
tech, Nanjing, China, Cat. TD601-01). The thermocy- average depth across each window using bedtools [46].
cling program was as follows: chain displacement at 72 Then, we calculated Spearman’s rank correlation coeffi-
°C; pre-denaturation at 98 °C for 2 min; five cycles of 98 cient (SRCC) between all pairs of duplicates and plotted
°C for 30 s, 60 °C for 30 s, and 72 °C for 3 min; and a final the correlation coefficient graph using R package corrplot
extension at 72 °C for 5 min. The amplification products [47]. Besides, we obtain high-quality SNP information
were purified with 1.2 × beads (Vazyme Biotech, Nanjing, through the GATK pipeline. The R package VennDiagram
China, Cat. N411-02). All libraries were pooled according [48] was used to draw Venn diagrams among repeats.
to equal amounts of products in one tube using the AIO-
seq method [16]. The quality inspection of the library
was performed on BiOptic Qsep100 Bio-Fragment Ana- Phylogenetic relationship and population structure
lyzer (Bioptic, C100100). Fragments from 430 to 780 bp analysis
was obtained using PippinHT (Sage) and sequenced on VCFtools (V0.1.13) [49] was used to select a subset of
NovaSeq 6000 platform (Illumina, San Diego, CA, USA) high-quality SNPs for subsequent analyses for maize
by the Berry Genomics Company (Beijing, China). natural lines using the following options: ‘–max-missing
0.8 –maf 0.05’. The high-quality SNP files for population
Library preparation of iRAD‑seq and sequencing structure analysis were transformed into plink files using
The library preparation is identical to AIO-seq. How- VCFtools (V0.1.13) [49]. Then, we perform principal
ever, to delete the sequence containing the RE recogni- component analysis (PCA) using Plink (V1.9) [50], and
tion site, the pooled library was digested by REs before PCA diagram was drawn using the R package ggplot2
sorting. The reaction was kept at 37℃ with 1 μg DNA [51]. We used VCF2Dis ([Link]
from the pooled library, 5 μL 10 × Cutsmart buffer and hen/​VCF2D​is) to calculate a P distance matrix based on
10 U of each RE in a 50 μL reaction system. The digested all the SNPs. The distance matrix was further calculated
products were subsequently purified using Cycle-Pure by FastME (V2.0) [52] to generate an unroot neighbor-
kit (OMEGA). After quality inspection and sorting, the joining (NJ) tree file with BIONJ method [53], and unroot
library was sequenced on the sequencer. tree diagram was drawn using R package ggtree [54].

Sequence quality checking and SNPs calling

The quality of the data was initially assessed using Population genetic analysis with BCF lines
FastQC (V0.11.8) [38]. The data were trimmed using A sliding window approach was used to find recombina-
Trimmomatic (V0.39) [39] to remove adapter sequences tion breakpoints of the B­ C1F4 and B
­ C2F4 lines. We used
and low-quality terminals. Duplicated data were then the Python 1 script to select homozygous SNP sites with
identified and removed using FastUniq (V1.1) [40]. The polymorphism between parents. The Python 2 script
high-quality data were mapped to the reference genome can select SNP sites in progeny that are consistent with
(B73_V4 [31] or ZhenShan97 [32]) using BWA-MEM the SNP sites from the results of Python 1, and can also
[41]. We used Samtools [42] and GATK (V4.1.8.1) [43] to genotype SNPs. The R language and the python 3 and
sort BAM files and mark duplicated reads. We performed the python 4 scripts were used to find the recombina-
variant calling using the GATK with the HaplotypeCaller tion breakpoints, after which bin markers were obtained.
command and default settings. We filtered the SNP infor- Bin map plots were drawn using RectChr [55]. The cal-
mation based on the modified parameters (QD < 2.0, culation of genetic distance was performed using QTL
QUAL < 30.0, SOR > 3.0, FS > 60.0, MQ < 40.0, MQRank- IciMapping [56], and the QTL were statistic with Win-
Sum < −12.5, ReadPosRankSum < −8.0) by GATK. QTLCart (v2.5) [57]. Genetic map and LOD diagrams
Chen et al. BMC Biology (2025) 23:327 Page 14 of 16

were drawn by the R package LinkageMapView and Additional file 5: Figure S12. Chromosome distribution of SNPs when
ggplot2, respectively. Python and R scripts are available using varying sequencing data sizes with panel_B. The size of the sliding
at Zenodo [58]. window is 1 Mb. Red, high density; green, low density.
Additional file 6: Figure S13. Chromosome distribution of SNPs when
Abbreviations using varying sequencing data sizes with AIO-seq. The size of the sliding
RRS Reduced representation sequencing window is 1 Mb. Red, high density; green, low density.
RAD-seq Restriction site-associated DNA sequencing
iRAD-seq Inverse Restriction site-associated DNA sequencing Additional file 7: Figure S14. Bin maps and genetic maps of [Link]-
SNP Single-Nucleotide Polymorphism struction of recombination bin maps illustrating recombination events
QTLs Quantitative trait loci in the CIMBL83/GEMS41and CML496/GEMS41 populationswith different
dd-RAD Double digest-RAD sequencing [Link]-density genetic maps of the CIMBL83/GEM-
RE Restriction enzyme S41and CML496/GEMS41 populationwith different sequencing strategies.
AIO-seq All-in-one sequencing Different methods use different REs. Figure S15. Phenotypic distribution
MAS Marker-assisted selection of the leaf [Link] of 193 maize lines from CIMBL83/GEMS41.
SRCC​ Spearman’s rank correlation coefficient The distribution of 68 maize lines from CML496/GEMS41. Figure S16. iRAD-
PCA Principal component analysis seq workflow integrated with a liquid handling automation system. The
NJ Neighbor-joining estimated time required is based on the processing of 960 samples.
SRA Sequence Read Archive Additional file 8: Table S1. Genome information of different species for in
silico digestion. Table S2. Sequencing information and statistics for maize-
and rice. Table S3. Sequencing information and statistics for a germplasm
Supplementary Information population of 132 maize lines. Table S4. Information of genotype from
The online version contains supplementary material available at [Link] maize germplasm for phylogenetic analysis. Table S5. Sequencing informa-
org/​10.​1186/​s12915-​025-​02330-8. tion and statistics for two BCF4 populations of maize. Table S6. Summary
of the genetic map for the maize BCF4 population. Table S7. Summary of
Additional file 1: Figure S1. The percentage of the genome covered by all quantitative trait lociidentified. Table S8. Comparison of performance
fragments larger than 300 bp after in silico digestion. Figure S2. Distribu- between iRAD-seq and other RAD methods.
tion of fragments after enzymatic [Link] coverage analysis
of fragments in soybean and [Link] number of fragments in soybean
and wheat. Five different panel of REs were selected for in silico digestion Acknowledgements
of the genome. Figure S3. Chromosome distribution of fragments after Special thanks go to Wang Yue for their invaluable experimental assistance
enzymatic digestion. Different genomes were digested by five different during the early stages of this research.
panel of REs. The size of sliding window is 10 kb. Red, high density; green,
low density. Authors’ contributions
Y.C., Y.L., and B.Z. discussed and designed the experiments; P.C., B.Z., S.Z., Z.Z.,
Additional file 2: Figure S4. Evaluation of RE [Link] of reads and Y.Y. performed the experiments; P.C., B.Z., Y.L., and Y.C. wrote the manu-
without RE recognition sites with Panel_A and Panel_B in maize and rice. script with contributions from all authors; H.L., and Y.X. contributed the maize
Proportion of reads without each RE recognition sites with Panel_A and population and rice line; P.C., B.Z., and H.C. participated in the data analysis
Panel_B in maize and rice. Panel_A: MseI, MspI and AluI; Panel_B: MseI, and prepared the figures; Y.C., and Y.L. supervised the project. All authors have
MspI and HindIII. Figure S5. Consistency of coverage among 4 repeats read and approved of the manuscript.
when using varying sequencing data sizes. Pairwise Spearman correlation
of the mean coverage between replicates with 100 kb windows. Panel_A: Funding
MseI, MspI and AluI; Panel_B: MseI, MspI and HindIII. Figure S6. The overlap This work was funded by the Biological Breeding-National Science and
of SNPs among 4 repeats when using varying sequencing data sizes. Technology Major Project (2023ZD0406107), the R&D program of Shen-
Panel_A: MseI, MspI and AluI; Panel_B: MseI, MspI and HindIII. Figure S7. zhen (KCXFZ20211020164207012), and the R&D program in key areas of
Application of iRAD-seq in maize [Link] yields of 132 Guangdong Province (2021B0707010006). The funders had no role in study
maize germplasms.1812 high-quality SNPs of 139 maize germplasms design, data collection and analysis, decision to publish, or preparation of the
widespread in the genome. Panel_A: MseI, MspI and AluI. manuscript.
Additional file 3: Figure S8. Sequencing yields of the maize genetic
[Link] yields of different sequencing strategies in the Data availability
CIMBL83/GEMS41 [Link] yields of different sequencing The raw sequence data reported in this paper have been deposited in the
strategies in the CML496/GEMS41 population. For boxplots, center line Genome Sequence Archive [59] in National Genomics Data Center [60], China
is the median; box limits represent upper and lower quartiles. Different National Center for Bioinformation/Beijing Institute of Genomics, Chinese
methods use different REs. Figure S9. iRAD significantly reduces the num- Academy of Sciences (GSA: CRA024789) that are publicly accessible at GSA
ber of reads containing restriction enzyme recognition [Link] of [61], and are also available in the NCBI Sequence Read Archive (SRA) under
reads without AluI recognition sites in the CIMBL83/GEMS41 population- accession number PRJNA1216733 [62]. The custom scripts and codes used in
and CML496/GEMS41 populationwith Panel_A, Panel_B and [Link]- this study are available at Zenodo [58].
portion of reads without HindIII recognition sites in the CIMBL83/GEMS41
populationand CML496/GEMS41 populationwith Panel_A, Panel_B and
AIO-seq. Different methods use different REs. Figure S10. iRAD-seq can
Declarations
enhance the depth of the obtained [Link] number of SNPs at different
Ethics approval and consent to participate
depths. The data used in the experiment was 0.5 Gb, 1.5 Gb, and 2.0 Gb.
Not applicable.
SNP number at different depths. Log10: n is the number of SNPs. For
boxplots, center line is the median; box limits represent upper and lower
Consent for publication
quartiles. Employ the t-test to assess differences in different method, p <
Not applicable.
0.0001. Different methods use different REs.
Additional file 4: Figure S11. Chromosome distribution of SNPs when Competing interests
using varying sequencing data sizes with Panel_A. The size of the sliding J.R., P.C., Y.C., and J.G. have filed Chinese patent applications 202211356595.5
window is 1 Mb. Red, high density; green, low density. based on this work. The authors declare that they have no conflict of interest.
Chen et al. BMC Biology (2025) 23:327 Page 15 of 16

Author details 20. Hartwig T, Chuck GS, Fujioka S, Klempien A, Weizbauer R, Potluri DP, et al.
1
Guangdong Laboratory of Lingnan Modern Agriculture, Genome Analy- Brassinosteroid control of sex determination in maize. Proc Natl Acad Sci
sis Laboratory of the Ministry of Agriculture and Rural Affairs Agricultural U S A. 2011;108(49):19814–9.
Genomics Institute at Shenzhen, Chinese Academy of Agricultural Sciences, 21. Kozarewa I, Armisen J, Gardner AF, Slatko BE, Hendrickson CL.
Shenzhen 518120, China. 2 College of Horticulture and Landscape Architec- Overview of Target Enrichment Strategies. Curr Protoc Mol Biol.
ture, Zhongkai University of Agriculture and Engineering, Guangzhou 510225, 2015;112:72121–272123.
China. 3 College of Agronomy, Anhui Agricultural University, Hefei 230036, 22. Onda Y, Takahagi K, Shimizu M, Inoue K, Mochida K. Multiplex PCR
China. Targeted Amplicon Sequencing (MTA-Seq): Simple, Flexible, and Versatile
SNP Genotyping by Highly Multiplexed PCR Amplicon Sequencing. Front
Received: 17 February 2025 Accepted: 9 July 2025 Plant Sci. 2018;9:201.
23. Yang GQ, Chen YM, Wang JP, Guo C, Zhao L, Wang XY, et al. Develop-
ment of a universal and simplified ddRAD library preparation approach
for SNP discovery and genotyping in angiosperm plants. Plant Methods.
2016;12:39.
References 24. Xu T, Xiao M, Yu L. Method for efficient soluble expression and purifica-
1. Xu Z, Zhou Z, Cheng Z, Zhou Y, Wang F, Li M, et al. A transcription factor tion of recombinant hyperactive Tn5 transposase. Protein Expr Purif.
ZmGLK36 confers broad resistance to maize rough dwarf disease in 2021;183: 105866.
cereal crops. Nat Plants. 2023;9(10):1720–33. 25. Penkov D, Zubkova E, Parfyonova Y. Tn5 DNA Transposase in Multi-Omics
2. Chen W, Chen L, Zhang X, Yang N, Guo J, Wang M, et al. Convergent Research. Methods Protoc. 2023;6(2):24.
selection of a WD40 protein that enhances grain yield in maize and rice. 26. Kia A, Gloeckner C, Osothprarop T, Gormley N, Bomati E, Stephenson M,
Science. 2022;375(6587):7985. et al. Improved genome sequencing using an engineered transposase.
3. Huang X, Huang S, Han B, Li J. The integrated genomics of crop domesti- BMC Biotechnol. 2017;17(1):6.
cation and breeding. Cell. 2022;185(15):2828–39. 27. Hennig BP, Velten L, Racke I, Tu CS, Thoms M, Rybin V, et al. Large-Scale
4. Hirsch CD, Evans J, Buell CR, Hirsch CN. Reduced representation Low-Cost NGS Library Preparation Using a Robust Tn5 Purification and
approaches to interrogate genome diversity in large repetitive plant Tagmentation Protocol. G3 (Bethesda). 2018;8(1):79–89.
genomes. Brief Funct Genomics. 2014;13(4):257–67. 28. Bruinsma S, Burgess J, Schlingman D, Czyz A, Morrell N, Ballenger C, et al.
5. Karbstein K, Tomasello S, Hodac L, Wagner N, Marincek P, Barke BH, et al. Bead-linked transposomes enable a normalization-free workflow for NGS
Untying Gordian knots: unraveling reticulate polyploid plant evolution by library preparation. BMC Genomics. 2018;19(1):722.
genomic data using the large Ranunculus auricomus species complex. 29. Zhao S, Li X, Song J, Li H, Zhao X, Zhang P, et al. Genetic dissection of
New Phytol. 2022;235(5):2081–98. maize plant architecture using a novel nested association mapping
6. Slovak M, Melicharkova A, Stubnova EG, Kucera J, Mandakova T, Smycka J, population. Plant Genome. 2022;15(1): e20179.
et al. Pervasive Introgression During Rapid Diversification of the European 30. Murray MG, Thompson WF. Rapid isolation of high molecular weight
Mountain Genus Soldanella. Syst Biol. 2023;72(3):491–504. plant DNA. Nucleic Acids Res. 1980;8(19):4321–5.
7. Baird NA, Etter PD, Atwood TS, Currey MC, Shiver AL, Lewis ZA, et al. Rapid 31. Jiao Y, Peluso P, Shi J, Liang T, Stitzer MC, Wang B, et al. Improved
SNP discovery and genetic mapping using sequenced RAD markers. PLoS maize reference genome with single-molecule technologies. Nature.
ONE. 2008;3(10): e3376. 2017;546(7659):524–7.
8. Andrews KR, Good JM, Miller MR, Luikart G, Hohenlohe PA. Harnessing 32. Zhang J, Chen LL, Xing F, Kudrna DA, Yao W, Copetti D, et al. Extensive
the power of RADseq for ecological and evolutionary genomics. Nat Rev sequence divergence between the reference genomes of two elite indica
Genet. 2016;17(2):81–92. rice varieties Zhenshan 97 and Minghui 63. Proc Natl Acad Sci U S A.
9. Peterson BK, Weber JN, Kay EH, Fisher HS, Hoekstra HE. Double digest 2016;113(35):E5163-5171.
RADseq: an inexpensive method for de novo SNP discovery and geno- 33. Schmutz J, Cannon SB, Schlueter J, Ma J, Mitros T, Nelson W, et al.
typing in model and non-model species. PLoS ONE. 2012;7(5): e37135. Genome sequence of the palaeopolyploid soybean. Nature.
10. Roberts RJ, Belfort M, Bestor T, Bhagwat AS, Bickle TA, Bitinaite J, et al. A 2010;463(7278):178–83.
nomenclature for restriction enzymes, DNA methyltransferases, homing 34. International Wheat Genome Sequencing Consortium (IWGSC). Shifting
endonucleases and their genes. Nucleic Acids Res. 2003;31(7):1805–12. the limits in wheat research and breeding using a fully annotated refer-
11. Wang S, Meyer E, McKay JK, Matz MV. 2b-RAD: a simple and flexible ence genome. Science. 2018;361(6403):eaar7191.
method for genome-wide genotyping. Nat Methods. 2012;9(8):808–10. 35. Rosen BD, Bickhart DM, Schnabel RD, Koren S, Elsik CG, Tseng E, et al. De
12. Toonen RJ, Puritz JB, Forsman ZH, Whitney JL, Fernandez-Silva I, Andrews novo assembly of the cattle reference genome with single-molecule
KR, et al. ezRAD: a simplified method for genomic genotyping in non- sequencing. Gigascience. 2020;9(3):giaa021.
model organisms. PeerJ. 2013;1: e203. 36. Groenen MA, Archibald AL, Uenishi H, Tuggle CK, Takeuchi Y, Rothschild
13. Adey A, Morrison HG, Asan, Xun X, Kitzman JO, Turner EH, et al. Rapid, MF, et al. Analyses of pig genomes provide insight into porcine demogra-
low-input, low-bias construction of shotgun fragment libraries by high- phy and evolution. Nature. 2012;491(7424):393–8.
density in vitro transposition. Genome Biol. 2010;11(12):R119. 37. Bionano. Software Downloads. [Link]
14. Cheng S, Miao B, Li T, Zhao G, Zhang B. Review and Evaluate the Bioin- oads/ (2019). Accessed 17 Dec 2019.
formatics Analysis Strategies of ATAC-seq and CUT&Tag Data. Genom 38. Andrews S. FastQC: a quality control tool for high throughput sequence
Proteom Bioinform. 2024;22(3):qzea054. data. [Link]
15. Chang L, Xie Y, Taylor B, Wang Z, Sun J, Armand EJ, et al. Droplet Hi-C 27816​42 (2010). Accessed 4 Oct 2018.
enables scalable, single-cell profiling of chromatin architecture in hetero- 39. Bolger AM, Lohse M, Usadel B. Trimmomatic: a flexible trimmer for Illu-
geneous tissues. Nat Biotechnol. 2024. mina sequence data. Bioinformatics. 2014;30(15):2114–20.
16. Zhao S, Zhang C, Mu J, Zhang H, Yao W, Ding X, et al. All-in-one sequenc- 40. Xu H, Luo X, Qian J, Pang X, Song J, Qian G, et al. FastUniq: a fast de novo
ing: an improved library preparation method for cost-effective and high- duplicates removal tool for paired short reads. PLoS ONE. 2012;7(12):
throughput next-generation sequencing. Plant Methods. 2020;16:74. e52249.
17. Walsh J, Waters CA, Freeling M. The maize gene liguleless2 encodes a 41. Li H, Durbin R. Fast and accurate short read alignment with Burrows-
basic leucine zipper protein involved in the establishment of the leaf Wheeler transform. Bioinformatics. 2009;25(14):1754–60.
blade-sheath boundary. Genes Dev. 1998;12(2):208–18. 42. Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, et al. The
18. Cao Y, Zeng H, Ku L, Ren Z, Han Y, Su H, et al. ZmIBH1-1 regulates plant Sequence Alignment/Map format and SAMtools. Bioinformatics.
architecture in maize. J Exp Bot. 2020;71(10):2943–55. 2009;25(16):2078–9.
19. Moreno MA, Harper LC, Krueger RW, Dellaporta SL, Freeling M. liguleless1 43. DePristo MA, Banks E, Poplin R, Garimella KV, Maguire JR, Hartl C, et al. A
encodes a nuclear-localized protein required for induction of ligules and framework for variation discovery and genotyping using next-generation
auricles during maize leaf organogenesis. Genes Dev. 1997;11(5):616–28. DNA sequencing data. Nat Genet. 2011;43(5):491–8.
Chen et al. BMC Biology (2025) 23:327 Page 16 of 16

44. Shen W, Le S, Li Y, Hu F. SeqKit: A Cross-Platform and Ultrafast Toolkit for

FASTA/Q File Manipulation. PLoS ONE. 2016;11(10): e0163962.
45. Okonechnikov K, Conesa A, Garcia-Alcalde F. Qualimap 2: advanced
multi-sample quality control for high-throughput sequencing data.
Bioinformatics. 2016;32(2):292–4.
46. Quinlan AR, Hall IM. BEDTools: a flexible suite of utilities for comparing
genomic features. Bioinformatics. 2010;26(6):841–2.
47. Wei T, Simko V. R package ’corrplot’: Visualization of a Correlation Matrix.
[Link] (2021). Accessed 8 Mar 2024.
48. Chen H, Boutros PC. VennDiagram: a package for the generation of
highly-customizable Venn and Euler diagrams in R. BMC Bioinformatics.
2011;12:35.
49. Danecek P, Auton A, Abecasis G, Albers CA, Banks E, DePristo MA, et al.
The variant call format and VCFtools. Bioinformatics. 2011;27(15):2156–8.
50. Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. Second-
generation PLINK: rising to the challenge of larger and richer datasets.
Gigascience. 2015;4:7.
51. Wickham H. ggplot2: Elegant Graphics for Data Analysis. [Link]
tidyv​erse.​org (2016). Accessed 15 Jul 2024.
52. Lefort V, Desper R, Gascuel O. FastME 2.0: A Comprehensive, Accurate,
and Fast Distance-Based Phylogeny Inference Program. Mol Biol Evol.
2015;32(10):2798–2800.
53. Gascuel O. BIONJ: an improved version of the NJ algorithm based on a
simple model of sequence data. Mol Biol Evol. 1997;14(7):685–95.
54. Yu G, Smith DK, Zhu H, Guan Y, Lam TT-Y. ggtree: an r package for
visualization and annotation of phylogenetic trees with their covari-
ates and other associated data. Methods in Ecology and Evolution.
2017;8(1):28–36.
55. He WM. RectChr: Multi-level visualization of genomic statistical variables
on rectangular chromosomes. [Link]
(2024). Accessed 18 Oct 2024.
56. Meng L, Li H, Zhang L, Wang J. QTL IciMapping: Integrated software for
genetic linkage map construction and quantitative trait locus mapping in
biparental populations. The Crop Journal. 2015;3(3):269–83.
57. Wang S, Basten CJ, Zeng Z. Windows QTL Cartographer 2.5. Department
of Statistics, North Carolina State University, Raleigh, NC. [Link]
ncsu.​edu/​qtlca​rt/​WQTLC​art.​htm (2012). Accessed 9 Aug 2021.
58. Chen P. chenpeng-allen/iRAD-script: Script for iRAD-seq method (V1.0.0).
Zenodo. 2025, [Link]
59. Chen T, Chen X, Zhang S, Zhu J, Tang B, Wang A, et al. The Genome
Sequence Archive Family: Toward Explosive Data Growth and Diverse
Data Types. Genomics Proteomics Bioinformatics. 2021;19(4):578–83.
60. Members C-N, Partners. Database Resources of the National Genomics
Data Center, China National Center for Bioinformation in 2022. Nucleic
Acids Res. 2022;50(D1):D27-D38.
61. Chen P. Testing and application of iRAD-seq. GSA. [Link]
cn/​gsa/​search?​searc​hTerm=​CRA02​4789 (2025).
62. Chen P. Testing and application of iRAD-seq. GenBank. [Link]
nlm.​nih.​gov/​biopr​oject/​PRJNA​12167​33 (2025).

Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.