High-throughput long paired-end sequencing of a Fosmid library by PacBio

Background Large insert paired-end sequencing technologies are important tools for assembling genomes, delineating associated breakpoints and detecting structural rearrangements. To facilitate the comprehensive detection of inter- and intra-chromosomal structural rearrangements or variants (SVs) and complex genome assembly with long repeats and segmental duplications, we developed a new method based on single-molecule real-time synthesis sequencing technology for generating long paired-end sequences of large insert DNA libraries. Results A Fosmid vector, pHZAUFOS3, was developed with the following new features: (1) two 18-bp non-palindromic I-SceI sites flank the cloning site, and another two sites are present in the skeleton of the vector, allowing long DNA inserts (and the long paired-ends in this paper) to be recovered as single fragments and the vector (~ 8 kb) to be fragmented into 2–3 kb fragments by I-SceI digestion and therefore was effectively removed from the long paired-ends (5–10 kb); (2) the chloramphenicol (Cm) resistance gene and replicon (oriV), necessary for colony growth, are located near the two sides of the cloning site, helping to increase the proportion of the paired-end fragments to single-end fragments in the paired-end libraries. Paired-end libraries were constructed by ligating the size-selected, mechanically sheared pooled Fosmid DNA fragments to the Ampicillin (Amp) resistance gene fragment and screening the colonies with Cm and Amp. We tested this method on yeast and Setaria italica Yugu1. Fosmid-size paired-ends with an average length longer than 2 kb for each end were generated. The N50 scaffold lengths of the de novo assemblies of the yeast and S. italica Yugu1 genomes were significantly improved. Five large and five small structural rearrangements or assembly errors spanning tens of bp to tens of kb were identified in S. italica Yugu1 including deletions, inversions, duplications and translocations. Conclusions We developed a new method for long paired-end sequencing of large insert libraries, which can efficiently improve the quality of de novo genome assembly and identify large and small structural rearrangements or assembly errors.


Background
The development of DNA sequencing technology has a short and rich history, and there have been many advancements in just over 40 years [1]. With Sanger's electrophoresis (the first generation) sequencing technology [2], the door to DNA sequencing was opened with its long read length and high precision, but its high cost and low throughput limits its development [3,4]. Massively parallel genome-sequencing technologies [4], with their low cost, high throughput, high accuracy and other characteristics, have become the mainstay of biological sequencing, except that short read lengths seriously hinder the study of large and complex genomes containing long repeats [5]. Single-molecule real-time synthesis and sequencing technology such as PacBio [6,7] and Oxford Nanopore Technologies [8][9][10] are new leading technologies with

Open Access
Plant Methods *Correspondence: mzluo@mail.hzau.edu.cn 1 College of Life Science and Technology, Huazhong Agricultural University, Wuhan 430070, China Full list of author information is available at the end of the article high throughput, long read length and other advantages, that create a new era of biological sequencing, although their disadvantages, such as a high error rate, can not be ignored. Currently, these DNA sequencing technologies are being rapidly developed and updated, and are widely used in de novo assembly [3,4], individual genome resequencing [11][12][13][14], clinical applications such as non-invasive prenatal testing [15,16], and counting devices for a wide range of biochemical or analytical phenomena [1].
Genomic libraries are collections of genomic DNA from a certain species that has been fragmented into specific sizes by biological, chemical or physical disruption. They are important tools and materials for molecular cloning, genomic structure and functional characteristic research [17]. Among genomic libraries, large-insert genomic libraries, such as Fosmid libraries (average insert approximately 40 kb) [18] and BAC library (average insert > 100 kb) [19][20][21], are widely used in physical map construction, genomewide sequencing, comparative genomics research, and genomic resource conservation due to their capacity for long lengths of foreign DNA fragments.
Paired-end (or mate-pair) sequencing technology using genomic libraries with different inserts to obtain pairedend sequences through different sequencing technologies-plays an important role in the field of biological sequencing. For example, the BAC library clones' end sequences are generated through Sanger sequencing technology to construct physical maps that help resolve long repeats and segmental duplications and provide long-range connectivity in shotgun assemblies of complex genomes [22][23][24]. Fosmids are shorter than BACs but much easier to generate. Therefore, mate-pair Fosmid library clones' end sequences [25,26] based on the Illumina sequencing platform enable the detection of structural variation predominantly mediated by repetitive elements such as insertions, deletions, and inversions [4,[27][28][29], which are commonly larger than 1 kb and are difficult to identify using conventional small insert paired-end libraries (300-500 bp) [30][31][32]. This method also enables the identification of unique sequences in the flanking regions of repetitive elements that potentially reveal precise structural variants breakpoint(s). In addition, data generated by paired-end libraries facilitates clinical application and shows that when the physical coverage increases, the required minimum read depth decreases [26,32]. Moreover, paired-end sequences of Fosmid and BAC libraries have made significant contributions in identifying long range structural variations in inter-or intra-chromosomes and in assessing the quality of whole genome assemblies, even correcting misassemblies and reducing contig numbers [33][34][35].
However, the first and second generation sequencing platforms can not generate DNA sequences longer than 1 kb, and the cost of the first generation sequencing platform is very high. Thus, the short read pairs (< 1 kb) generated by these paired-end sequencing technologies are limited in the assembly of complex genomes, and repetitive regions (> 1 kb) are usually missing or misassembled, leading to fragmented and incomplete genomes. Therefore, longer paired-end reads are required.
Recently, new technologies that can also be used for genome assembly such as 10× Genomics, Hi-C and Bio-Nano are being developed. They have their own characteristics and applications. Data from 10× Genomics are widely used in de novo whole genome assembly [36,37], assisting genome assembly [38] and detecting structural variants [39,40] because of large spans (> 50 kb) and a low cost. Hi-C related articles such as identifying target genes [41], revealing structural remodeling [42] and analyzing enhancer expression [43], have risen exponentially since 2017. The Hi-C technology also has been widely used in assisting genome assembly [44,45]. BioNano improves genome assembly [46] and detects genomewide SVs [47] based on single-molecule optical mapping technologies with its long connective data. Single molecule sequencing technologies have become routine in genomics. However, the paired-end sequencing of fosmid and BAC clones, 10× genomics, Hi-C, and Bionano optical mapping provide long connective data that are necessary for genome assembly and regularly used across the plant tree of life.
Although many methods have been developed as described above and applied in the study of genomic sequencing, the biological genome is difficult to explore clearly with just one or a few methods, especially for large animal and plant genomes with a high GC content and long repeat sequences. Therefore, the combination of different methods and mutual verification has become the mainstay of current genome sequencing. Hence, we developed a new method for genome sequencing to break the limitation that traditional jumping libraries can not generate reads with an average length longer than 1 kb. Our method provides an alternative way to assist genome assembly and has an advantage that the interested large fragment clones can be screened out by their corresponding end sequences. The utilities of the method in de novo assembly and structural rearrangement detection were tested on the yeast and S. italica Yugu1 genomes.

The pipeline of high-throughput long paired-end sequencing of a Fosmid library
To enrich the approaches of genome sequencing, we developed a new method to generate high-throughput long paired-end fragments of a Fosmid library. Figure 1 shows the pipeline of the method. A Fosmid library was constructed. Pooled Fosmid DNA was sheared into 13-18 kb fragments and separated by pulse field gel electrophoresis (PFGE). Size selected DNA fragments were recovered by electroelution, end-repaired and ligated to the Ampicillin resistance gene label. Colonies transformed with the paired-end fragments containing the vector and the Amp tag were screened by chloramphenicol and ampicillin. Then, the vector was removed by I-SceI, and the paired-end fragments containing Amp were recovered and sequenced on the PacBio Sequel platform.

The first modification of the Fosmid vector based on pcc2FOS
In our new method, the recovery of complete long paired ends as single fragments from the paired-end library was critical. Therefore, we replaced the two 8-bp NotI restriction sites flanking the LacZ fragment harboring the cloning sites in pcc2FOS ( Fig. 2a) with the 18-bp homing endonuclease I-SceI sites by PCR using the primers P1 (5′-attacc-ctgttatccctaGTC GGG GCT GGC TTA ACT AT-3′) and P2 (5′-attaccctgttatccctaTTC GCG TTG GCC GAT TCA TT-3′) containing the I-SceI sites at the 5′ ends, resulting in the fragment named A (Additional file 1: Figure S1).
In the pipeline, mechanical interruption was adopted to break the pooled Fosmid DNA. This resulted in 3 main types of fragments: (1) Fragments containing the entire vector sequence and the paired-end insert sequence (2) fragments containing part of or the entire vector sequence and single-end insert sequence, and (3) fragments containing only the insert sequence without the vector sequence. Only the fragments containing both the replicon (oriV) and Chloramphenicol resistant gene (CmR) in vector as in (1) and (2) could be screened out by transformation (Additional file 1: Figure S2). However, oriV and CmR were both on the same side of the multiple cloning sites in pcc2FOS, which resulted in a high proportion of single ends in our prediction. To improve efficiency and reduce the cost of sequencing, the proportion of (1) must be increased. Thus, we moved the LacZ fragment containing multiple cloning sites to the position between the oriV and CmR. The pcc2FOS vector was digested by NotI, and the pcc2FOS backbone without LacZ was recovered, self-ligated and propagated in E. coli EPI300.-T1R. Then, new PCR primers, P3 (5′-ATT CAA ATC GTT TTC GTT ACCGC-3′) and P4 (5′-ATG CCT TCA GGA ACA ATA GAA ATC T-3′), with sequences complementary to the area between oriV and CmR were used to generate the skeleton of the vector pcc2FOS, named B (Additional file 1: Figure S1). The PCR products A and B were ligated, resulting in pHZAUFOS2 (Additional file 1: Figure S3).

Preliminary test of the method for Fosmid long paired-end sequencing
To test the new Fosmid paired-end sequencing strategy, we used pHZAUFOS2 to construct two Fosmid libraries: Y1 for Saccharomyces cerevisiae S288C and S1 for Setaria italica Yugu1. The library sizes were estimated to be 1.2 Fig. 1 The pipeline of Fosmid-size long paired-end library construction. The red area represents the vector, the blue area represents the large inserted genomic fragment, and the yellow area represents the Ampicillin resistance gene tag. The Fosmid clones were pooled together, and DNA was extracted for paired-end library construction. Pooled Fosmid plasmid DNA was sheared into ~ 15 kb fragments by g-TUBE (Covaris). It generated insert only, vector with single-ends and vector with paired-ends. Then, these DNA fragments were end repaired and gel purified for ligation with the Ampicillin resistance gene tag. Although all fragments could be ligated to the Ampicillin resistance gene tag, only those containing the chloramphenicol resistant gene and oriV ligated to an Amp tag were screened out with double resistance to chloramphenicol and ampicillin after transformation. Finally, the vector was removed by I-SceI and the paired-end fragments with the Amp tag were sequenced on PacBio million colony-forming units (cfu) and 90 thousand colony-forming units (cfu), corresponding to 15× physical genome coverage and 10× physical genome coverage for Y1 and S1, respectively. Fosmid clones of each library were amplified in bulk by overnight liquid culture at 37 °C, and pooled Fosmid DNA was prepared. A pairedend library was constructed with pooled Fosmid DNA. Again, pooled paired-end library DNA was extracted, digested with I-SceI and size-selected on PFGE gels. Paired ends were recovered and sequenced on Frasergene's PacBio RSII platform. The reads were aligned to the reference genomes of the S. cerevisiae S288C and S. italica Yugu1 (Additional file 2: Table S1).
We obtained a total of 35,510 clean end subreads from library Y1 after removing reads shorter than 50 bp. The N50 of each end was almost 3 kb, and the longest subread was 15 kb (Table 1, library Y1). These clean end reads were used for alignment with the reference genome S. cerevisiae S288C. After removing those unaligned reads, single-end aligned reads, chimaeras and reads aligned to multiple places, 25,812 reads (73%) were obtained as unambiguously placed paired ends. A total of 22,192 (86%) of 25,812 reads were unambiguously mapped in the expected spacing (20-50 kb) and correct orientation (convergent) on the reference genome. On average, these correct Fosmid jumps were 38 kb in length with a standard deviation of 2.2 kb. After deduplication, we recovered a total of 3067 unique Fosmid-size jumps, covering approximately tenfold of the S. cerevisiae S288C genome.
We also obtained a total of 67,220 clean subreads from library S1. The N50 of each end was 2.8 kb ( Table 1, library S1). These clean end reads were used for alignment with the reference genome S. italica Yugu1. After removing those unaligned reads, singleend aligned reads, chimaeras and reads aligned to multiple places, 41,998 (63%) reads were obtained as unambiguously placed paired ends. A total of 36,969 (88%) of 41,998 reads had correct Fosmid jumps (20-50 kb). After deduplication, we recovered a total of 13,334 unique Fosmid-sized jumps, covering approximately 1.3-fold of the S. italica Yugu1 genome.
Those paired ends located in unexpected spacing or orientation, e.g., spacing < 20 kb, > 50 kb, inverted orientation, tandem orientation and linking 2 reference contigs, were identified as chimaeras and counted (Additional file 2: Table S1). The chimaeric rate of unique read pairs (1157) in the nonredundant set of Y1 was 27.1% (Fig. 3a), and the chimaeric rate of unique read pairs (2663) in the nonredundant set of S1 was 16.6% (Fig. 3b).

Further modification of the Fosmid vector based on pHZAUFOS2
In the pHZAUFOS2 -based method above, the two I-SceI sites were used to release the complete paired ends. However, the resulting complete pHZAUFOS2 vector band was ~ 8 kb (Additional file 1: Figure S4), Fig. 2 The maps of the vectors pcc2FOS and pHZAUFOS3.A is the map of pcc2FOS. NotI was used to release the insert and the lacZ fragment was outside of CmR and oriV. B is the map of pHZAUFOS3. The LacZ fragment was moved between CmR and oriV; the two I-SceI sites adjacent to LacZ were used to release the insert, and another two I-SceI sites were used to break the vector skeleton into small fragments (2-3 kb) which was just within the 5-10 kb range of the pairedend DNA fragments we recovered (Additional file 1: Figure S5A). This is why we had high vector contamination rates in the datasets of Y1 and S1. Therefore, to reduce the vector contamination rate and increase the effective paired-end data, we introduced another two I-SceI sites into the skeleton of pHZAUFOS2 without affecting its function. This was accomplished with two pairs of PCR primers P5 and P6 and P7 and P8 (Additional file 1: Figure S1). The new version of the vector was named pHZAUFOS3 (Fig. 2a). Then, we constructed the libraries Y2 (10× physical genome coverage) and S2 (20× physical genome coverage) in the pHZAUFOS3 vector. Digestion of the pHZAU-FOS3 libraries with I-SceI resulted in complete inserts and 2-3 kb of vector pieces (Additional file 1: Figure  S5B).

Optimization of the method for Fosmid long paired-end sequencing
Our preliminary test data showed that too many chimaeras were introduced during Fosmid and/or pairedend library constructions. For large-insert library construction, the trapped small DNA fragments in the size-selected large fragment fractions used for library construction were usually the main cause of chimaeras. The higher the DNA fragment concentration loaded on the PFGE gel, the more the small DNA fragments were trapped.
To reduce chimaeras as much as possible, we took several measures for the construction of another two Fosmid libraries/paired-end libraries series: Y2 for S. cerevisiae S288C and S2 for S. italica Yugu1. First, we screened DNA fragments twice on PFGE gels in both the Fosmid library and paired-end library constructions to reduce the trapping of small fragments. In contrast to the paired-end library constructions of Y1 and S1, we  italica Yugu1 library S1 (grey) and S2 (black) (B) aganist their respective reference genomes. The y-axis represents percentage of all unique read pairs that fall in the 1-kb bin. The x-axis represents the distance between read pairs dephosphorylated the paired-end fragments and ligated them to the phosphorylated Amp tag to reduce the ligation of the unrelated small DNA fragments. As a result, the chimaeric rates of Y2 and S2 were reduced to 10.6% and 4.2% compared to 27.1% and 16.6% of Y1 and S1, respectively (Fig. 3). The numbers of nonredundant 20to 50-kb jumps from Y2 and S2 were 1518 (88.7%) and 9363 (95.3%), respectively. Moreover, we sought to generate more effective paired-end data at lower sequencing costs by increasing the physical coverage of the Fosmid library clones in each pool. Therefore, a total of ~ 0.2 million clones of S2 (20× physical genome coverage) were used to construct a paired-end library and sequenced in one PacBio flow cell, which generated 9363 unique Fosmid-size jumps, approximately the same as the number of S1 generated from six PacBio flow cells (Additional file 2: Table S1). A detailed breakdown of the sequencing reads from all four test libraries is available in Additional file 2: Table S1.

Impact on de novo genome assemblies of whole genome PacBio reads
We tested the effect of Fosmid long paired-end sequences with long-range connectivities on de novo genome assemblies of whole-genome PacBio reads. First, we tested the effect of simulated Fosmid long paired-end data on de novo genome assemblies of simulated whole-genome PacBio subreads. We simulated the sequencing data of the S. cerevisiae S288C strain on the PacBio Sequal platform based on the reference genome of the S. cerevisiae S288C strain from NCBI (GCF_000263155.2) at different sequencing depths, 10×, 20×, 30×, 40× and 50×, and assembled five draft yeast genomes, Pb10, Pb20, Pb30, Pb40 and Pb50, respectively (Additional file 2: Table S2). Additionally, we simulated five yeast Fosmid libraries with insert sizes of 38 kb and a standard deviation of 2.2 kb at different genome physical coverages (10×, 20×, 30×, 40×, 50×) and correspondingly simulated five Fosimd long pairedend sequence sets (Fos10, Fos20, Fos30, Fos40, Fos50) generated by PacBio, with read lengths of 7 kb (paired ends) and a standard deviation of 2 kb (Additional file 2: Table S3). We reassembled Pb10, Pb20, Pb30, Pb40 and Pb50 by adding the simulated paired-end data of Fos10, Fos20, Fos30, Fos40 and Fos50, respectively. The results showed that the assembly quality improved as the sequencing depth of the genome increased and the physical genome coverage of the Fosmid library increased. Notably, when the sequencing depth of the genome reached 20× and the physical genome coverage of the Fosmid library reached 10× , the assembly quality significantly improved. All chromosomes were assembled completely and covered by one scaffold except chromosome 12 (Additional file 1: Figure S6A). Moreover, the assembly result reached chromosome level when the sequencing depth reached 30× and the physical genome coverage of the Fosmid library reached 20× (Additional file 1: Figure S6B). Then, we tested the effect of our real Fosmid long paired-end data on the de novo yeast genome assembly with the simulated whole-genome PacBio subreads. Of the five draft yeast genomes that were de novo assembled only by simulated PacBio whole-genome sequencing data, Pb30 had the average assembly quality. However, it only had an N50 scaffold length of 568 kb. When we added our real long paired-end data, Y1 (tenfold physical subread coverage) and Y2 (fivefold physical subread coverage), to improve the qualities of the draft yeast genome Pb30 (details see additional file 1: Table S4), the N50 of the assembled scaffold improved to 935 kb (Fig. 4a) and 786 kb (Fig. 4b), respectively.
Finally, we tested the effect of our real S. italica Yugu1 Fosmid long paired-end data on the de novo assembly of real whole genome PacBio subreads of the S. italica genome. Although the S. italica Yugu1 genome was published as the foxtail millet reference genome, it was assembled with Sanger reads. Therefore, we made use of the PacBio contigs of S. italica Yugu18, which has a genome sequence that is almost the same as that of S. italica Yugu1 (GWHABGJ00000000). The contig number of Yugu18 was 383, and the N50 length was 3.75 Mb. After adding our long paired-end sequences of S1 and S2 together (~ tenfold Fosmid physical subread coverage and ~ 1.5-fold whole-genome subread coverage), the scaffold number of S. italica Yugu18 was reduced to 330, and the N50 length was increased to 5.2 Mb, which was a 1.5-fold improvement to the assembly of the WGS only ( Table 2).

Detection of structural rearrangements or assembly errors
One important application of jumping libraries is the comprehensive detection of chromosomal structural rearrangements/variants or assembly errors. Large fragment sizes enable the identification of uniquely aligned reads in both ends, particularly when the chromosomal structural variants or assembly errors are likely mediated by repetitive elements [30]. We used the S2 data of S. italica Yugu1 to detect the structural variants or assembly errors of the published S. italica Yugu1 genome sequence, which is used as the foxtail millet reference genome sequence. After filtering out the low-quality sequences, almost 50,000 unambiguous Fosmid-size subread paired ends were obtained that covered the genome sequence of approximately 0.75×. These paired ends with a left-end length of 2.85 kb and a right-end length of 2.85 kb on average were mapped to the Yugu1 genome sequence. Five distinct large rearrangements or assembly errors of dozens of kb were identified with 9 or more independent supporting subread pairs for each. These rearrangements or assembly errors included the three most frequently observed events-deletion, duplication and translocation (Table 3). A large deletion (~ 58 kb) was detected in chromosome VIII and framed by 12 unique subread pairs. This deletion was located at the end of the chromosome and without any annotation.
Our approach generated long paired ends and single ends. The length reached up to 2-3 kb for each end of paired-ends and 5 kb for single ends on average. Therefore, we took advantage of the long ends to detect small structural rearrangements or assembly errors. Five different rearrangements or assembly errors of several kb, including inversion, deletion and duplication, were detected with 7 or more unique supporting reads for each (Table 4).
Large-insert paired ends also play an important role in comparative genomics studies [24,48]. We aligned the long paired-ends of S2 to the S. italica Yugu1 and S. italica Yugu18 genome assemblies. The rate of nonredundant jumps located in the range of 20-50 kb was 96.0% in Yugu1 and 93.0% in Yugu18 genome (Additional file 2: Table S5).

Discussion
We developed a new method for long paired-end sequencing of large DNA fragment libraries that could be complimentary to other methods, such as Fosill, pBA-Code, 10× Genomics, Hi-C and BioNano, to improve de novo genome assembly and detect structural arrangements and assembly errors.  Paired-ends from large DNA fragment libraries, such as Fosmid and BAC library, are usually used for detecting structural rearrangements/variants and assembly errors, delineating associated breakpoints and assisting de novo genome assembly. Their large spans help to resolve long repeats and segmental duplications and provide longrange connectivity to shotgun assemblies of complex genomes [22,49,50]. Several high-throughput paired-end sequencing approaches using large-insert genomic libraries, such as the Fosmid library called Fosill (Fosmid libraries by Illumina) [51] and the BAC library called pBACode [52], were developed and used for the de novo assembly and SV detection of several genomes [52,53]. Also, large insert size paired-ends methods that do not depend on large-insert genomic libraries have been created for large and complex genomes, especially those rich in repeats, such as 10× Genomics [38], Hi-C [54], and BioNano [11,46]. They make a significant contribution to the assembly of complex genomes [4,55,56], closing gaps [57,58] and detecting structural variations [59] or large scale errors, such as those in pseudomolecules spanning chromosomes [60], including insertions, deletions, duplications and inversions spanning tens to hundreds of kb. However, these strategies based on massively parallel genomesequencing technologies can not produce end sequences much longer than 1 kb. Therefore, the paired-ends generated by these methods are usually too short and require much higher physical coverages for partial compensation. Single-molecule real-time synthesis and sequencing technologies such as PacBio [6,7] and Nanopore [8][9][10] are leading to a new era of biological sequencing. It is suitable for assisting de novo genome assembly via overlapconsensus methods, especially for large and complex genomes. Recently, the single-molecule real-time synthesis and sequencing technology is significantly improved and the error rate of it can be reduced to the level as NGS [61]. Our method applied the characteristics of large inserts of genomic libraries and long subreads of the PacBio platform to generate DNA calipers with long spans of 20-50 kb and long paired ends of up to 2-3 kb each end on average. These paired ends are much longer than those generated by other methods, and would become longer as the average subread length increases in the single-molecule real-time synthesis and sequencing technology. Since these long paired ends better tolerate sequencing errors, the positioning of sequences can be more precise, and the connection error of contigs can be reduced. Besides, the long-distance ends can be used to correct assembly errors of complex genomes [33][34][35]. The longer DNA read lengths can significantly increase the detection rate of structural rearrangement events and reduce the rate of mismatching with low physical coverage, especially for genomes containing high-repeat regions [62]. Moreover, our method results in a certain proportion of single ends; these long single ends (average > 5 kb) can be used as whole-genome sequences to detect small structural variants of tens to thousands of bp. In the application of our long Fosmid-size paired-end method with only ~ tenfold Fosmid physical subread coverage and ~ 1.5-fold wholegenome sequence subread coverage, five distinct large rearrangements or assembly errors of dozens of kb were identified with 9 or more independent supporting subread pairs for each (Table 2), and five small different rearrangements or assembly errors of several kb were detected with 7 or more unique supporting long single subread ends for each ( Table 3). All of these large and small rearrangements of S. italica Yugu1 may imply misassemblies in the Yugu1 reference genome.
It has been shown that the rate of concordant jumps in which two ends were aligned to the same scaffold with correct spacing and orientation is an important parameter for the quality of paired-end methods. This parameter was 95.3% in our optimized method (Fig. 3). It is almost the same as the previously reported 96% of Fosill [51] and better than the 90.2% of pBACode [52]. Chimaeras were the main factor affecting this parameter and are usually an obstacle in the application of paired-end technology, which could result in misassemblies. In our study, we performed DNA fragment size selection twice on pulse field gel both in constructing the Fosmid library and pairedend library and ligated dephosphorylated paired-end fragments to phosphorylated Amp tag. By this measure, the chimaera rate of S2 significantly decreased to 4.2% ( Fig. 3; Additional file 2: Table S2). The chimaeric rates of Y1 and Y2 were higher than those of S1 and S2 (Fig. 3). This phenomenon is most likely because the DNA concentration of the S. cerevisiae S288C loaded on the pulsed field electrophoresis gel was too high (much higher than that of S. italica Yugu1) to separate well (not shown) and can be avoided in future practice. The conventional 40 kb mate-pair library was constructed by enzyme digestion [63]. The uneven distribution of the restriction sites might produce cloning bias. In Fosill method, Fosmid-size paired-end library was constructed with nick translation that could reduce the cloning bias [51]. However, this method depends on the delicate concentration of DNA polymerase I and dNTPs and has a limit in generating long paired ends. In pBA-Code method, a random-barcode-based high-throughput approach with ultrasonic interruption was used for BAC paired-end sequencing [52]. This approach can generate single ends of up to 800 bp long and pair them with the same barcode. All above three methods are based on Illumina technology that generate short end reads and are incompatible with emerging long-read high-throughput technologies [64,65]. They usually use biotin labels [66] for recovering paired ends and/or use enzyme sites [67] to screen the positive paired ends. Paired-end sequencing samples are prepared by inverse PCR. However, the rate of base errors introduced by PCR will increase as amplification and insert size increase. This is incompatible with long-read technologies (> 10 kb). We instead adopt mechanical randomly interrupted DNA to effectively reduce bias and obtain uniform long ends. Our method is straightforward and easy to manipulate. In our study, paired-end samples were prepared through cloning and vector removal instead of PCR, and additional base errors and bias can be avoided. The ampicillin resistance gene was used both as a marker to screen positive long paired-end clones together with the vector chloramphenicol resistant gene (CmR) and as a tag to distinguish the left and right ends. The latter is highly important in paired-end sequencing methods, especially those generating long reads based on PacBio or Nanopore sequencing technologies. In addition, there are many options for tags used in this method, such as different antibiotic genes or one antibiotic gene with a random sequence of several bp as indexes. The indexes are very important in pooling samples of different libraries. Moreover, if random-barcode pairs such as pBACode [52] are introduced into our vector, pHZAUFOS3, our method can also distinguish different clones in pools to construct high-quality physical maps.
To adapt long-read single molecule sequencing technologies and generate long paired ends, we modified the vector based on pcc2FOS. Previously available Fosmid vectors usually use NotI digestion for insert sizing and release. For large DNA-insert clones from high GC content organisms or monocotyledonous plant genomes, digestion with NotI would cut each insert into several to many fragments, which makes insert sizing difficult and the release of intact inserts almost impossible [21]. In our new vector, pHZAUFOS3, four I-SceI sites were introduced, and the chloramphenicol resistant gene (CmR) and replicon (oriV) necessary for colony growth were located near to the two sides of the cloning site. Since I-SceI is a rare-cut restriction enzyme that recognizes an 18-bp sequence, the recognition sequence was not found in most species when searching the genome sequence database. Two I-SceI sites that flank the cloning site in the vector can be used to release complete large DNA inserts. Another two I-SceI sites located in the skeleton of the vector can fragment the vectors into pieces with lengths that are much shorter than those of the paired ends, ans so can effectively reduce the vector contamination rate and increase the effective paired-end data (Additional file 1: Figure S5). Adjusting the positions of the CmR and oriV can help to increase the proportion of the paired-end fragments to single-end fragments in the paired-end libraries.
It is well known that single molecule sequencing technologies such as PacBio and Oxford Nanopore Technologies can produce long read length sequences with an average length of more than 10 kb, but have a reduced accuracy (75-90%) due to their dependence on singlemolecules detection [50,68]. As the high error rate, the long-read technologies are rarely used to detect SNVs or indels. In these technologies, CCS derives a consensus sequence from noisy individual subreads [69,70]. Recently, a long high-fidelity (HiFi) technology has been used to produce highly accurate (99.8%) HiFi reads of 13.5 kb in average length and applied for variant detection [61]. However, this technology is limited by the number of passes required to achieve the desired accuracy and the polymerase read length of the sequencing platform. Thus, the insert of the CCS library can't be too long. The paired-ends generated by our method were shorter than 15 kb, which is in the range of the insert of the CCS library for HiFi sequencing. If our method is applied to the HiFi technology, it might generate highly accurate (99%) fosmid paired-ends that could be used to provide validation to structural variant calls. Moreover, in order to obtain longer connective information, we are attempting to apply our method to BAC pairedends production. In fact, our vector pHZAUFOS3 can be used to construct both Fosmid and BAC libraries (our unpublished data). We believe that the highly accurate long BAC paired-ends could be used to further improve the quality of genome assembly and make the detection of large-scale structural variations more accurate and efficient.

Conclusion
We developed a new method for obtaining long spanning long paired ends. This method is straightforward and enables DNA manipulation to be performed easily. It can be applied complimentary to other methods in assembling complex genomes, detecting structural variations and assembly errors, and assessing assembly qualities.

Construction and preparation of the pHZAUFOS2 and pHZAUFOS3 vectors based on pcc2FOS
PCR primers (P1: 5-attaccctgttatccctaGTC GGG GCT GGC TTA ACT AT-3′ and P2: 5-attaccctgttatccctaTTC GCG TTG GCC GAT TCA TT-3′) containing the I-SceI sites were used to amplify the LacZ fragment based on the pcc2FOS vector. The resulting fragment was named the A fragment. The pcc2FOS vector was NotI digested, and then the pcc2FOS skeleton without LacZ was recovered, self-ligated and propagated in E. coli EPI300.-T1R (Epicentre). The new PCR primers (P3: 5′-ATT CAA ATC GTT TTC GTT ACCGC-3′ and P4: 5′-ATG CCT TCA GGA ACA ATA GAA ATC T-3′) complementary to the area between oriV and CmR were used to generate the new skeleton of the vector pcc2FOS, named B. These two PCR products, A and B, were ligated and then transformed into E. coli strain EPI300.-T1R (Epicentre), resulting pHZAUFOS2. Transformants were cultured on LB plates with 12.5 μg/mL chloramphenicol, 80 μg/mL X-gal and 100 μg/mL IPTG overnight before counting and collecting.
pHZAUFOS2 and pHZAUFOS3 plasmid DNA were propagated in E. coli strain EPI300.-T1R (Epicentre) grown at 37 °C in LB broth with shaking (225-250 rpm), 12.5 μg/mL chloramphenicol and the 500× Copy Control Fosmid Autoinduction Solution overnight (16-20 h). Plasmid DNA was prepared using the plasmid midi kit (Qiagen) according to the manufacturer's instructions. Vectors were prepared as described by Shi et al. [21] for BAC library construction. Plasmid DNA (40 µg) was linearized using 200 units Eco72I restriction endonucleases from Fermentas at 37 °C for 2 h, dephosphorylated by a two-step incubation (1 h at 37 °C and 1 h at 55 °C) with 2 × 25 units calf intestine alkaline phosphatase (NEB), self-ligated at 16 °C overnight, separated on a CHEF agarose gel. The linear vector fragments were recovered by electroelution. The undigested circular plasmid DNA and/or re-ligated non-dephosphorylated vector DNA will be removed in this process. Ultra-0.5 centrifugal filter devices (Amicon) were used to concentrate the linear vectors up to a final concentration of 0.5 mg/μL.

Fosmid library construction
Fosmid libraries were constructed using the method modified from the protocol of Copy Control ™ HTP Fosmid Library Production Kit with pCC2FOS ™ Vector (Epicentre). High molecular weight genomic DNA was prepared as described by Shi et al. [21] for BAC library construction. Liquid culture and young leaves were used for yeast S. cerevisiae strain S288C and S. italica Yugu1, respectively. Yeast protoplasts and S. italica Yugu1 nuclei were evenly embedded in the gel plugs of low melting temperature agarose. The gel plugs were then treated with proteinase K for 48 h at 50 °C and partially sheared by freezing and thawing (20 s freeze and 45 s thaw). The DNA fragments were size-selected twice by pulsed-field gel electrophoresis. The DNA fragments of 30-45 kb were recovered, end repaired, ligated to the vector and then packaged with the MaxPlax Packaging Extract [20]. The packaged products were used to infect EPI300-T1R cells (Epicentre) and then the transfected cells were spread on LB plates with 12.5 μg/mL chloramphenicol, 80 μg/mL X-gal and 100 μg/mL IPTG. After incubation at 37 °C overnight, the clones were washed off plates using liquid LB, pooled together and then stored at -80 °C.
A total of 400 μg of pooled Fosmid DNA was sheared into fragments by g-TUBE (Covaris), with a mean size ranging from 6 to 20 kb. All DNA samples were loaded into a united single well in the middle of the gel and the markers on the wells of the two sides, and separated twice on CHEF apparatus at 0.5-1.5 s linear ramp, 9 V/ cm, 14 °C in 0.5× TBE buffer for 15-17 h. The gel fraction of 12-18 kb was recovered from the unstained center part of the gel. DNA fragments were electroeluted at 4 °C in 1× TAE buffer, concentrated by Ultra-0.5 centrifugal filter devices (Amicon) and dephosphorylated by a two-step incubation (1 h at 37 °C and 1 h at 55 °C) with 2 × 25 units calf intestine alkaline phosphatase (NEB). Phenol:chloroform:isoamyl alcohol (25:24:1) was used to remove the calf intestine alkaline phosphatase (NEB). The supernatant was concentrated and purified by Ultra-0.5 centrifugal filter devices (Amicon). A total of 200 μL DNA was end repaired with 50 units of T4 DNA polymerase (ThermoFisher), 100 units of Klenow Fragment (ThermoFisher) in 500 μL reaction mixture [10× Klenow Fragment buffer, 200 μM of dNTP Mix] and incubated at 37 °C for 1 h. The reaction mix was incubated at 65 °C for 15 min to terminate the end repairing and treated with phenol:chloroform:isoamyl alcohol (25:24:1). The supernatant was concentrated and purified by Ultra-0.5 centrifugal filter devices (Amicon) again.
The Amp resistance gene fragment was amplified by PCR from the plasmid puc19 with the phosphorylated primers (5-AAA CGC GCG AGA CGA AAG GG-3′ and 5-GGG GTC TGA CGC TCA GTG GA-3′). The PCR products were purified through glue recycling and concentrated by Amicon ® Ultra-0.5 centrifugal filter devices to a final concentration of 0.5 mg/µL. The end repaired DNA fragments (30 μL) were ligated with the Amp resistance gene fragments (1:10) with 10 units T4 DNA ligase (ThermoFisher) in a final volume of 50 μL at 16 °C for 16-18 h. After incubated at 65 °C for 15 min to terminate the reaction, the ligation mix was used to transform TransforMax ™ EPI300 ™ Electrocompetent E. coli (Epicentre) by electroporation. The tranformants were spread on LB plates with 12.5 μg/mL chloramphenicol, 50 μg/ mL carbenicillin, 80 μg/mL X-gal and 100 μg/mL IPTG. After incubation at 37 °C overnight, the clones were washed off plates using liquid LB, pooled together and then stored at -80 °C.

Preparation of Fosmid long paired-ends for sequencing
The pooled clones of Fosmid paired-end sequencing library were cultured and induced to a high copy number by the 500× Copy Control Fosmid Autoinduction Solution (Epicentre) at 37 °C overnight (16-20 h) with 12.5 μg/mL chloramphenicol and 50 μg/mL carbenicillin and shaking (225-250 rpm). Plasmid DNA was extracted using the plasmid large constructed kit (Qiagen) according to the manufacturer's instructions, digested with I-SceI, and separated on agarose gel. The paired-end fragment fractions of 5-10 kb were recovered, electroeluted and purified. The final samples were concentrated by Ultra-0.5 centrifugal filter devices (Amicon) to a final amount of 30 μg and sent to Frasergen Company for sequencing on the PacBio Sequel platform.
The Fosmid long paired-end sequences were aligned to the simulated contigs by minimap2 (v2.11) (-a -x mappb) [78]; low-quality reads were removed, and chimaeras alignment results were generated by samtools (v1.3) (view -h -q 60 -F 2048) [79]. Paired-end sequences with a mass alignment value of 60 without chimaericas were retained. The software bamToBed (v2.27.0) [80] was used to obtain the alignment coordinate information, and the longest paired-end was retained after calculating the total length of the multiple paired-end sequences from one clone. Then, the retained paired end sequences were combined with the simulated contigs to assemble the scaffolds by SSPACE (v3.0) (-k 2 -p 1) [81]. The order and direction of the assembled scaffolds were adjusted, and the scaffolds were aligned to the S. cerevisiae strain S288C (GCF_000146045.2) or S. italica Yugu1 (GCF_000263155.2) reference genome sequences to assess the assembly quality.

SV detection
After the low-quality data was filtered out by samtools (v1.8), the long paired ends were aligned to the reference genome sequences by bwa, and then, the data were transferred from bam file to deduplication by sambamba (0.6.7). Large structual arrangements were detected by Delly (0.8.1). Small SVs were detected by sniffles (v1.0.10) using the long single ends including those split from the paired ends as PacBio whole-genome sequencing subreads.
Additional file 1: Figure S1. The modification process of pHZAUFOS2 and pHZAUFOS3. Figure S2. The putative clone types of the pcc2FOS paired-end library. Figure S3. The map of the vector pHZAUFOS2. Figure  S4. Sequence read length distribution of preprocessed PacBio sequencing data. Figure S5. I-SceI digestion of the random clones from the pHZAUFOS2 (A) and pHZAUFOS3 (B) paired-end libraries DNA. Figure S6. Simulated yeast genome alignments between the scaffolds and reference.
Additional file 2: Table S1. Detailed breakdown of sequencing reads. Table S2. Statistics of yeast contig assembly by simulation. Table S3. Statistics of simulated yeast Fosmid libraries and paired-end reads. Table S4. Statistics of scaffold assembly by simulated PacBio data and our Y1 and Y2 long paired ends. Table S5. Assessment of genome assemblies.