Construction of a high density genetic map of an interspecific cross of Capsicum chinense and Capsicum annuum and QTL analysis of floral traits

The yield of pepper plants (Capsicum spp.) is their most important trait and is affected by the flower number and flowering time. Capsicum annuum produces a single flower per node and has an early flowering habit. By contrast, Capsicum chinense yields multiple flowers per node and has a late flowering character. However, the genetic mechanism underlying the control of these floral traits remains largely unknown. In this study, 150 F2 populations from an interspecific cross between the inbred lines 740 (C. chinense) and CA1 (C. annuum) and their parents were used to construct a molecular genetic linkage map using the specific length amplified fragment sequencing (SLAF-seq) technique. This linkage map, spanning 1,586.78 cM in length, contained 9,038 markers on 12 chromosomes, with a mean marker distance of 0.18 cM. Phenotypic data on the flowering time and flower number per node were collected over multiple years, and QTL analysis identified 6 QTLs for the flowering time and flower number per node by composite interval mapping (CIM) and genome-wide composite interval mapping (GCIM) methods at least in two environments. The candidate genes within the major QTL were predicted. In the major flowering time QTL, the candidate gene Capana02g000700, which encodes the homeotic protein APETALA2, was identified. Quantitative reverse-transcription PCR (qRT-PCR) analysis indicated that its expression level in 740 was higher than that in CA1. Gene expression analysis indicated that the expression of Capana02g000700 was significantly upregulated in flowers, and many floral development-related genes were found to be coexpressed with Capana02g000700, supporting the function of this gene in association with flowering time in C. chinense and C. annuum species.

Scientific RepoRts | (2019) 9:1054 | https://doi.org/10.1038/s41598-018-38370-0 one flower per node and early maturity, whereas C. chinense plants have multiple flowers (always two to four) per node and a late flowering time habit. Therefore, interspecific hybridization of C. annuum with C. chinense, with the multiple flower trait being transferred into C. annuum, may be potentially useful to increase the yield and enhance uniform maturity, which may make mechanical harvesting feasible 9,10 . However, transferring superior traits from C. chinense into C. annuum to develop viable commercial varieties is very time-consuming and expensive because of the number of backcrosses required. Since most of the traits mentioned above are quantitatively inherited or controlled by multiple major genes 1,9,11 , the discovery of the QTLs or major genes that govern these traits in various backgrounds is imperative; furthermore, the use of molecular-assisted selection can shorten breeding cycle and accelerate the breeding process of new varieties of pepper. During the past few decades, many genetic maps, including integrated maps, have been constructed for peppers using either intraspecific 12,13 or interspecific 8,14,15 populations to identify the QTLs of horticultural traits 16 . However, most of these genetic maps were low-density, and many identified QTLs covering large region lead hard to use for molecular assistant selection. A genetic map, especially a high-density genetic map, provides an important foundation for QTL mapping and major QTL cloning. In Capsicum species, a number of interspecific and intraspecific genetic maps have been constructed by using conventional methods in previous studies 1,15 . However, the current number of markers is too small to build a high-density genetic map, which limits the efficiency and accuracy of QTL mapping. Next-generation sequencing (NGS) technologies can be used to detect large quantities of SNP markers in the whole genome for high-resolution genetic map construction. Several methods combine NGS with restriction enzyme digestion to reduce the complexity of the target genomes, including genotyping-by-sequencing (GBS-seq) 17 and restriction site-associated DNA sequencing (RAD-seq) 18 . The selection of digested DNA fragment sizes is critically important to improve the efficiency of tag utilization. Unlike GBS-seq, which does not select the size of the digested fragment before PCR amplification, the RAD-seq conducts the size-selection step of the digested fragment before PCR amplification 19 . However, traditional RAD-seq technology has several shortcomings, such as more operation steps and shorter read length. By combining bioinformatics and RAD-seq technology, specific-locus amplified fragment sequencing (SLAF-seq) is developed 20 . SLAF-seq applies a bioinformatics approach to simulate the results of enzyme digestion, selects the most suitable restriction enzymes for double digestion, and then sequences the PCR-amplified fragments on an Illumina sequencer. This approach can effectively avoid repetitive sequences in the genome, develop SNP markers with uniform distribution in the genome, increase the effective reads obtained by sequencing, and improve the efficiency of molecular marker development 20 .
To date, many pepper intraspecific high-density genetic maps have been reported [21][22][23] . However, interspecific high-resolution genetic maps have rarely been reported in peppers 24 . Due to an abundance of polymorphic DNA sequences in interspecific individuals, the construction of a high-density genetic map based on SNPs markers of Capsicum species is possible. In this study, based on SLAF-seq, we constructed a high-density genetic map of an interspecific cross of C. annuum and C. chinense. Furthermore, we used a high-density genetic linkage map to detect QTLs for certain traits of peppers: flowering time and flower number per node.

Results
SLAF sequencing and genotyping of the interspecific cross F 2 population. In this study, the 150 interspecific cross F 2 populations and their parents were genotyped using SLAF-seq technology. Based on the results of the SLAF pilot experiment, HaeIII was used for SLAF library construction. The library comprised SLAF fragments ranging from 414 to 514 bp in length. After high-throughput sequencing of the SLAF library, 183.83 Gb of raw data was generated. In total, 749.26 M pair-end reads were obtained for both parents and 150 progenies, with an average of 4.92 M reads for each individual line. For quality control processing, reads of low-quality were discarded during quality checks in each cycle. This dynamic process was repeated until the average genotype quality score of all SLAFs reached the cut-off value of 30 (quality scores of at least 30, indicating a 1% likelihood of an error and thus 99.99% confidence). On average, Q30 was 92.78%, and the GC content was 38.21%. After reads clustering, 406,563 high-quality SLAFs were detected. The average depths of these SLAFs were 90.34 (male parent) and 58.61 (female parent) for parents and 11.74 for each individual progeny (Table 1).
Based on the results of SLAF positioning on the Zunla-1 (Capsicum annuum) genome, the SLAFs on each chromosome were calculated (Table 2), and a distribution diagram of SLAFs on each chromosome is shown in Fig. 1a,b. The SLAFs were distributed equally on each chromosome, and the pepper genome has been successfully simplified. Among the 406,563 high-quality SLAFs, 171,413 were polymorphic according to an analysis of allele numbers and the differences between gene sequences, with a polymorphism rate 42.16%. Of the 171,413 polymorphic SLAFs, 94,733 were classified into eight segregation patterns (ab × cd, ef × eg, hk × hk, lm × ll, nn × np, aa × bb, ab × cc, and cc × ab) (Fig. 1c). Because the F 2 population was obtained from a cross of two diverse pepper inbred lines with the genotype aa or bb, only the SLAF markers that had segregation patterns of aa × bb were used in map construction.   The x-axis indicates eight segregation patterns of polymorphic SLAF markers; the y-axis indicates the number of markers. F 2 population is obtained from a cross of two pepper inbred lines with the genotype aa or bb; therefore, only the SLAF markers, which had segregation patterns of aa × bb, were used in map construction. segregation ratios (χ 2 test, p < 0.05) were also maintained for genetic map construction, and (4) the markers demonstrating less than 85% integrity were discarded. Finally, 13,472 high-quality makers were distributed into 12 chromosomes according to their physical locations on the pepper reference genome and the MLOD scores >3 with other markers. As a result, 9,038 markers were designated for use in the final linkage map construction (Supplementary Tables 1 and 2). These markers are homozygous in the two parents, with a sequence depth 150-fold for the male parent and 101.4-fold for the female parent, respectively, and have more than 99% integrity of SLAFs for individuals. The polymorphic SLAFs used for the map construction ranged from 0.73% to 11.50% among the 12 linkage groups (LGs) ( Table 2). Finally, the map contained 12 LGs and spanned a genetic distance of 1,586.78 cM in total, with an average distance of 0.18 cM between adjacent markers ( Table 2). The distribution of the SLAF markers on each LG is shown in Fig. 2. On average, each LG contained 753 markers that covered an average of 132.23 cM. As shown in Table 2, the largest LG was Chr2, and it spanned a length of 188.3 cM, with 1,730 markers and an average distance of 0.11 cM between adjacent markers. In contrast, the smallest LG was Chr6, which harbored 1,355 makers, covered a length of 75.14 cM, and had a 0.06 cM average intermarker distance. The largest gap of 12 LGs ranged from 1.6 (Chr2) to 17.2 cM (Chr5). This genetic map included 28,741 SNPs ( Table 2).

Evaluation of the Capsicum Genetic
Map. The quality of the genetic map was evaluated by using HighMap to construct haplotype map and heat map. The haplotype maps, which reflect the double recombination and deletion of the population, were generated for the parental controls and 150 offspring using 9,038 SLAF markers. In this study, most of the recombination blocks were distinctly defined. The missing data for each LG ranged from 0.76% to 1.14% (Supplementary Table 3). Most of the LGs were uniformly distributed, suggesting that the genetic maps were of high quality. The heat maps showed the relationships of the recombination between markers from each LG. The comparisons between markers were used to assign recombination scores to the 9,038 SLAF markers, after which the heat maps were constructed. The resulting maps showed that the order of the SLAF markers in most of the LGs have been correctly ordered ( Supplementary Fig. S1). In total, 803 distorted SLAF markers were integrated into the map (Supplementary Table 4). They were noted in all LGs except Chr2 and Chr3, and most of the distorted markers were skewed toward the male parent. The frequencies of the distorted markers in Chr4 and Chr12 were much higher than those of the other LGs.
To evaluate the collinearity between the genetic map and the Zunla-1 reference genome, all SLAF markers were anchored on the Capsicum reference genome. As the results presented in Fig. 3, a sufficient genome coverage and the accurate genetic location of the markers was revealed by the consecutive curves except for Chr1 and Chr8. Nevertheless, the Spearman rank correlation test of the genetic map and the physical map revealed that the correlations were significant (p < 0.001) among the 12 chromosomes, indicating that the 9,038 SLAF markers were accurately positioned on 12 chromosomes, and the Capsicum genome was sufficiently covered with these SLAF markers. The genetic arrangements of most markers were also considered to coincide with their physical direction based on the falling trend of the curve.  Tables 5 and 6). As a result, the fruit in the parental and offspring populations mainly ranged from 1 to 3 per node ( Fig. 4e-h).
Phenotypic data of flowering time and flower number per node of the two parents, and the F 1 and F 2:3 families were collected from three environments across three years. In all experiments, the phenotypic data flowering time of the two parents (740 and CA1) were located at the extreme ends of the largely normally distributed family means, and those for the F 1 and F 2:3 families were close to the mid-parent values, suggesting the quantitative genetics of these traits (Supplementary Fig. S2; Supplementary Tables 5 and 6). For the multiple flower trait,   being collected over different years, data for these traits were highly consistent and of good quality, which provided a solid foundation for the QTL analysis.

QTL analysis of flowering time and flower number per node traits.
We conducted QTL analysis using the composite interval mapping (CIM) approach with data for each year in R/QTL. The LOD threshold for declaring significance of a QTL for flowering time and flower number per node was determined with 1000 permutations (p = 0.05). Details of each detected QTL, including the map location, LOD value, percentages of total phenotypic variances explained (R 2 ), and 1.5-LOD support interval are presented in Table 3 and Supplementary  in floral development (Fig. 5b). Analysis of the nucleotide sequence of Capana02g000700, 8 SNPs and a 6 bp deletion were detected among the 740 and CA1 ORF regions ( Supplementary Fig. S3), which lead to predicting that the protein from CA1 has two amino acid deletions and five amino acid changes, including an amino acid in the AP2 domain ( Supplementary Fig. S4). The qRT-PCR analysis of Capana02g000700 expression indicated the expression was significantly more upregulated in the floral than other detected tissues (Fig. 5c). A comparison of the 740 and CA1 mRNA amount in the different tissues indicates its expression in 740 was significantly higher than in the corresponding CA1 (Fig. 5c). The RNA-Seq expression data in flower, leaf and fruit differential developmental stages were retrieved from the pepper inbred line 6421 (C. annuum) 29 , and Capana02g000700 was found to have a higher expression level in pepper flower than in other tissues (Fig. 5d), which indicates that it plays an important role in flower development, strongly suggesting that Capana02g000700 is candidate gene for transcriptional repression in the control of flowering time of pepper. With regard to the Mf2.1, the QTL region was mapped to 0.382 cM, and the interval physically represents approximately 1400 kb in the Zunla-1 reference genome, and 98 putative predicted genes were found to reside in this region (Supplementary Fig. S5a; Supplementary Table 9). Inflorescence architecture is based on changes in the activity of the meristems, small groups of stem cells located at the tips of shoots 30,31 . Studies illustrated that, during the vegetative transition to flowering, the dynamically expressed genes were found enriched for transcription factors 32 . In the Mf2.1 region, GO and COG analysis indicate that the genes involved in the transcription regulation process were enriched (Supplementary Fig. S5b). We retrieved the pepper transcriptome data of the middle vegetative meristem (MVM), transition meristem (TM), and floral meristem (FM) tissues reported previously 32 . Expression analysis revealed that a large number of genes were upregulated in TM, including TF from different family such as GATA (Capana02g002708, Capana02g002714), NAC (Capana02g002682), Zinc-finger homeodomain protein (Capana02g002717), and bHLH (Capana02g002736) (Supplementary Fig. S5c). These upregulated TFs within the Mf2.1 region may be the candidates involved in control of transition meristem to form of floral meristem and then determine flower numbers.

Coexpression analysis of genes expression.
We adopt the WGCNA to identify genes with differential expression at distinct stages of flower development. We identified a flower-development-specific module, which contained 107 genes ( Fig. 6a; Supplementary Table 10). Of these genes, ten transcription factors (TFs), including Capana02g000700, were identified, indicating that these TFs may be involved in the transcriptional control of flower development. Strikingly, among these genes, nine were found highly co-expressed with Capana02g000700, and three TF Capana11g000298 (bZIP family), Capana08g000623 (MADS-box family), Capana05g001110 (bHLH family) were detected as coming from different families (Fig. 6a). Capana08g000623, an orthologous gene from A. thaliana PISTILLATA, was reported as being involved in flower development and its expression regulated by the AtAPETALA2 33 . In tobacco plants, reduced amounts of TGA2.1 (Capana11g000298 orthologous gene) from tobacco resulted in the development of petal-like stamens, indicating a regulatory role of TGA2.1 in defining organ identity in tobacco flowers 34 . In addition, Capana05g001110 encoding brassinosteroid enhanced expression 3 (BEE3) homologue positively regulated brassinosteroid signaling and required flower normal growth in Arabidopsis 35 . A heat map shows the expression (FPKM) of 65 genes selected from Fig. 6a and found most of the expression of these genes was flower-development regulated (Fig. 6b). These results strongly support that Capana02g000700 may repress target genes and/or TFs expression to regulate flower development and then affect flowering time.

Discussion
High-density genetic map constructed with C. chinense and C. annuum interspecies hybridization F 2 population. The pepper genome sequence has been completed, and the size of the pepper genome is large, estimated to be 3.48 Gb 3,36 . Whole-genome deep resequencing or low-coverage sequencing is relatively costly for large genomes and usually unnecessary for gene/QTL mapping 11,37 . The SLAF-seq, which was developed based on high-throughput sequencing, is an effective strategy for large-scale SNP discovery and genotyping 20 .
In contrast to conventional methods, which are inefficient, expensive, and time-consuming 20,38 , SLAF sequencing can generate large amounts of sequence information and handle whole genome density distributions, which ensures density, uniformity, and efficiency of marker development 20,39,40 . In this study, we constructed a high-density genetic map using the SLAF-seq technology with an interspecies F 2 population consisting of 150 individuals. This map, which included 9,038 high-quality SLAF markers on 12 LGs, covered a genetic distance of 1,586.78 cM in total, with an average distance of 0.18 cM between adjacent markers, and showed 99% integrity for individuals (Fig. 3, Table 2). To date, compared with other genetic maps of Capsicum species 16,24,41,42 , the genetic map reported in this study is the highest density map and had the smallest average distance between interval markers for Capsicum genus plants. With this high-density map combination with high-quality genome sequences, the candidate genes within a narrow region interval between adjacent markers can be predicted directly. However, as reflect in our results we also need to keep in mind that success QTL mapping depend on the chromosome region the QTL falls into. Overall, SLAF-seq technology is ideal for population genotyping and for high-resolution linkage map construction because of its high success rates, specificity, and stability 20 . Accordingly, the genetic map could be used for detecting QTL for important horticultural traits, and the narrowed QTLs also provided several promising candidate genes for further functional identification. Segregation distortion is a commonly observed in both interspecific and intraspecific cross populations 43 . In this study, to avoid the losing information, the 803 SLAF markers that were also utilized in map construction. In total, almost all the segregation makers were skewed toward the male parent. The distorted segregation could be caused by gametophytic factors that affect female gametes 43,44 , but this distortion needs further study. We found that the rate of the polymorphic SLAFs in the Chr1 and Chr8 being used for map construction were significantly lower than other LGs. Possibly, chromosome translocation between chromosome 1 and chromosome 8 in the inbred pepper lines 740 and CA1 and the many markers had to be filtered before being used for genetic map construction. Indeed, the translocation between chromosome 1 and chromosome 8 in C. frutescens, C. chinense, C. baccatum and wild C. annuum compared to cultivated C. annuum was previously reported and has been well characterized 45 . In addition, only C. annuum-originated sequences were used for developing the SLAF markers, whereas some C. chinense specific markers were unable to be detected. Therefore, the alignment of the SLAFs to the C. chinense genome needs to be developed for a more complete genetic map in the future.

QTL identification of floral-related traits.
The yield-related traits are important for pepper production, and high yield and high disease resistance always are the most important objectives for pepper breeding. Therefore, the detection of QTLs or genes for these traits should be important for pepper genetic improvement. Hitherto, QTLs studies on yield related traits have been widely documented in intraspecific and interspecific crosses of Capsicum species populations 8,12,46,47 . Previous studies have strongly indicated that selection for early flowering can enhance yields in various horticultural types of pepper. Flowering time is a fully quantitative complex trait, and researchers always record the number of days between sowing and anthesis or at certain day after sowing to score the flower or fruit developmental status of the third node to evaluate flowering times 8,15 . In addition, some others also used the leaves numbers on primary stem to evaluate flowering time 28 . After comprehensively comparing the characteristic of parents and progenies used in this and previous studies, we selected the score criterion for evaluating the flowering time, and three QTLs were detected in the study ( in our populations across the three years. The major QTL related to flowering time in the Chr2 was also detected by using the RILs derived from the interspecific cross of the C. frutescens × C. annuum 15 . A comparison of the physical position reported previously, adjacent to the Ft2.1 QTL interval, revealed a gene previously reported and mapped on the Chr2 from different studies 14,28 , which promotes the phase transition from inflorescence meristem to floral specification in pepper CaAP2 (Capana02g003062). In addition, in the C. annuum intraspecific F 2 populations, using an SLAF-seq and BSA analysis, a candidate region was mapped on Chr12 that controls the first flower node and determines the flowering time 47 . However, the Ft2.1 not located in a previously reported physical position, indicated the Ft2.1 is a new locus that plays an important function in controlling flower development in the Capsicum genus among plants of different genetic backgrounds. In addition, among the two QTLs detected on Chr6, the novel QTLs Ft6.1 and Ft6.2 detected in this study differ from previous studies, suggesting that perhaps the population used for QTL identification came from different species, as previously reported. However, the three detected QTLs, which accounted for approximately of 50% of the total phenotypic variance, were detected with two methods, which may have been because the flowering time was scored based on a visual scale of 1-6, which limits the resolution of mapping a fully quantitative trait. Alternatively, markers with significant segregation distortion were used for genetic map construction, and these distorted markers may have affected the QTL analysis 43,44,48 . A more detailed scoring criterion may contribute to detecting more QTLs for the flowering time.
Possibly, the gene for the multiple flower trait could be transferred from the C. chinense to C. annuum varieties with a more concentrated fruit set, thereby contributing to a potentially higher yield 10 . In this study, we found the hybrid progenies with multiple flowers plants can produce an average of approximately 1.5 flowers per node (Supplementary Table 6). In addition, we observed that progeny with multiple flowers are always accompanied with more than one fruit setting in a node (Fig. 4), and this result was consistent with results reported previously by others 9,10 . Previous studies have proposed hypotheses for different genetic control mechanisms of multiple flowers: (i) a three-gene dominance model with epistasis 49 , (ii) at least five independently segregating chromosome segments involved in the multiple-flower habit 9 , (iii) seven semi-recessive additive genes from C. chinense 50 , and (iv) three major dominant genes from C. chinense 10 . Possibly, the populations were delivered from differential accessions of C. annuum and C. chinense, and the interrelationships among these models are unclear. In this study, with the CIM and GCIM QTL analysis, the detected QTLs Mf2.1, Mf7.1 and Mf10.1, with a positive additive effect (increased flower number), could exhibit greater phenotypic variation of the multiple flower trait. However, the GCIM method was also able to detect another two QTLs Mf8.1 and Mf11.1. In addition, the QTLs detected by GCIM had higher LOD peak scores, had more narrowed interval and were greater in number, indicating that GCIM mapping approach is more effective and reliable for detect more QTLs. Over all, the result seems consistent with the hypothesis of five-gene model of genetic control of multiple flowers in Capsicum 9,10 , but this needs further study. Nevertheless, we provided more detailed loci information, and this could contribute to uncovering the underlying genetic and molecular mechanism of multiple flowers traits in C. chinense. In addition, three QTLs related to multiple flowers Mf2.1, Mf7.1, Mf8.1, Mf10.1 and Mf11.1 were mapped in a narrow interval, which means that molecular markers closely linked to multiple-flower-per-node locus can be more effectively applied in early selection. Within the Mf2.1 region, even though we identified many TFs and found the expression of these TFs were upregulated in the transition meristem, these TFs may play an important role in control of vegetative meristem transition to floral meristem, which determines inflorescence architectures (flower number determinants). However, we found ANANTHA (AN) and COMPOUND INFLORESCENCE (S, the homolog of WUSCHEL-RELATED HOMEOBOX 9, WOX9) and their orthologs genes have conserved functions in the control of inflorescence architecture among Solanaceae species 31,51 . Given that the QTL can vary according to various factors such as mapping populations, genotyping method, detection method and environmental factor 43,44,52 . When considering these factors, we cannot rule out the possibility that these genes were involved in the control of the trait of flower number per node between C. annuum and C. chinense species. Further study is needed to finely map and identify the candidate genes underlying molecular mechanism control of multiple flowers and for the development of reliable makers for marker-assisted selection to pyramid the genes that control multiple-flower into commercial cultivars.
Multiple strategies to prediction of flowering time candidate genes. Within the flower major region Ft2.1, we were able to identify a candidate gene for the regulation of flowering time and flower development. After annotation of the 25 genes resides in genomic regions, we found Capana02g000700 encoding a floral APETALA2 protein, whose homologs from AtAPETALA2 and AtFLO2 act as flowering suppressors (Fig. 5b). In addition, we found Capana02g000700 expression significantly upregulated in flowers, and we also determined its expression in late flowering 740, where the expression was significantly higher than observed for the early flowering CA1 (Fig. 5c). Furthermore, we found many flower developmental-stage-specific genes were highly coexpressed with Capana02g000700 (Fig. 6). In a sequence comparison of CDS between the parents used for the QTL mapping, we detected 8 SNPs and a 6 bp deletion in CA1, but none of these changes result in a premature stop codon (Supplementary Figs S3, S4). We cannot rule out the possibility that the sequence variations between 740 and CA1 may have caused change in the gene activity. However, after comparing the Capana02g000700 gene from the C. annuum and C. chinense reference genomes reported before, we found the SNPs were highly conserved among the same species ( Supplementary Figs S3 and S4), and we assumed that no major changes in the activity of the protein could be expected. In Arabidopsis, one of actions of AtAPETALA2 was to control flower development by repressing the flowering-promoter MADS-box transcription factor such as PISTILLATA 33 , AGAMOUS 40 and SOC1 41 . So, the results presented in this study show that the higher expression of Capana02g000700 in late flowering 740 results in enhanced repression of target flowering-promoter genes expression, which finally leads to late flowering. Although we could illustrate significant expression differences between the early flowering and late flowering parents, the underlying specific sequence variation in the promoter region associated with this difference is still unknown. We retrieved the C. chinense and C. annuum genome sequence reported previously 3,36,53 , and we found that the nucleotide sequence within the Ft2.1 region was misassembled or that the assembly quality was quite low for many of the gaps presented. We tried to elucidate the DNA variations between CA1 and 740, but we failed to amplify of Capana02g000700 promoter using the primers designed for the C. annuum genome sequence (Zunla-1 and CM334) 3 gene detected in this study was different from the previously reported CaAP2, we assume the Capana02g000700 detected in this study is CaAP2 and Capana02g000700, which evolved from the same ancestor gene but evolved after the two AP2/ERFs (i.e., CaAP2 and Capana02g000700). These genes, which evolved independently, control flowering time in a different genetic background. We hypothesized that the Capana02g000700 mainly exerts control on the flowering time between C. annuum and C. chinense species, but further research is needed. SLAF library construction and high-throughput sequencing. The SLAF-seq strategy was used to analyse the genotype of two parents and 150 F 2 offspring as described previously 20 . Briefly, the genomic DNA of the two parents and 150 F 2 populations was digested using the HaeIII restriction enzyme (New England Biolabs, USA). Subsequently, a signal nucleotide overhang was added to the digested fragments along with Klenow fragments (New England Biolabs, USA) and dATP at 37 °C. Then, PAGE-purified Duplex Tag-labelled sequencing adapters (Life Technologies, USA) were ligated to the A-tailed DNA with T4 DNA ligase (New England Biolabs, USA). After incubation, the reaction products were pooled and purified using a Quick Spin column (Qiagen, Germany). The purified products were electrophoresed on a 2% agarose gel, and fragments with sizes ranging from 414 to 514 bp were collected and purified using a gel extraction kit (QIAGEN, Germany). The purified product was sequenced on an Illumina HiSeq. 2500 system (Illumina, USA) according to the manufacturer's instructions.

Materials and Methods
Sequence data analysis and genotyping. The SLAF marker grouping and genotyping were performed using procedures as described previous 20 . Briefly, after filtering out the low-quality reads (quality score < 30e), the remaining cleaned SLAF pair-end reads were clustered based on sequence similarity as alignment with BLAT (-tilesize = 10 -stepsize = 5). Subsequently, pair-end clean reads were mapped onto the reference genome of C. annuum. var Zunla-1 36 and reads with over 90% similarity sequences were grouped into one SLAF locus. Minor allele frequency evaluation was used to define alleles in each SLAF locus. C. annuum and C. chinense are diploid species, one locus could harbor at most four SLAF tags, locus containing more than four tags were filtered out as repetitive SLAFs, and those with two, three, and four tags were identified as polymorphic SLAFs. Then polymorphic SLAFs were classified into eight segregation patterns (aa × bb, ab × cc, ab × cd, cc × ab, ef × eg, hk × hk, lm × ll and nn × np). Because the F 2 population is obtained from a cross of two diverse pepper inbred line with the genotype aa or bb, therefore only the SLAF markers which had segregation patterns of aa × bb were used in map construction.
Sequence data analysis and genotyping. The SLAF marker grouping and genotyping were performed using procedures as described previous 20 .
Briefly, after filtering out the low-quality reads (quality score < 30e), the remaining cleaned SLAF pair-end reads were clustered based on sequence similarity as alignment with BLAT (-tilesize = 10 -stepsize = 5). Subsequently, pair-end clean reads were mapped onto the reference genome of C. annuum. var Zunla-1 36 and reads with over 90% similarity sequences were grouped into one SLAF locus. Minor allele frequency evaluation was used to define alleles in each SLAF locus. C. annuum and C. chinense are diploid species, one locus could harbor at most four SLAF tags, locus containing more than four tags were filtered out as repetitive SLAFs, and those with two, three, and four tags were identified as polymorphic SLAFs. Then polymorphic SLAFs were classified into eight segregation patterns (aa × bb, ab × cc, ab × cd, cc × ab, ef × eg, hk × hk, lm × ll and nn × np). Because the F 2 population is obtained from a cross of two diverse pepper inbred line with the genotype aa or bb, therefore only the SLAF markers which had segregation patterns of aa × bb were used in map construction. Genetic map construction. In order to ensure the quality of genetic map, high-quality SLAF markers for the genetic map construction were filtered by the following criterions: (1) SLAF makers with parents sequence depth less than 18 were filtered out; (2) a SLAF which had less than five SNPs and average depth of each sample above four, was defined as a high quality SLAF marker; (3) markers with complete less than 85% were filtered; (4) Since distortedly segregated markers are ubiquitous and would affect the mapping construction and QTL analysis 43,44,48 , partial distorted polymorphism markers showing significance (p < 0.05) were maintained to construct the map. Subsequently, by using the HighMap strategy, the SLAF markers were assigned into chromosomes based on the pepper genome, and 12 linkage groups (LGs) were obtained. The modified logarithm of odds (MLOD) value was calculated between two adjacent makers, the SLAFs with MLOD values less than three were excluded. In addition, using HighMap software to analyze the linear array of markers in each LG, and estimate the genetic distances between two adjacent markers.
QtL/Gene mapping. Plant trait QTLs were identified by different methods. QTLs were detected by CIM methods with the R/QTL package methods using R/QTL v3.1.1 55 . The significance thresholds were determined using 1,000 permutations (p < 0.05). The results from the CIM analysis were used to construct the QTLs, and their positions were used in a default model. In addition, multilocus QTL mapping was performed by the software of QTL.gCIMapping.GUI (https://cran.r-project.org/web/packages/QTL.gCIMapping/index.html) according to the user manual 52,56 .
Candidate gene selection and annotation. The predicted genes in the target QTL region were analysed according to the annotation of the pepper Zunla-1 and CM334 reference genomes 3,31 . The function of genes detected in the candidate region was manually confirmed using protein BLAST. In addition, GO enrichment and KEGG pathway analyses were performed with default settings. Multiple sequence alignments were performed with ClustalX, and a phylogenetic tree was calculated by the neighbour-joining method and bootstrap analysis with 1000 replicates via MEGA7 software 57 .
qRT-PCR analysis candidate gene expression level. Floral buds, fruit at developmental stage of 16 day post anthesis, leaves at 20 day after emergence, stem and root were collected from 740 and CA1, respectively. Samples were ground into fine powder in liquid nitrogen, and then the RNA was isolated from all samples using HiPure Plant RNA Mini Kit (Magen, China). Subsequently, the RNA from each sample was used for the reverse transcription reaction using a HiScript Q RT SuperMix for qPCR reagent kit with gDNA wiper (Vazyme, China). Quantitative real-time PCR analysis was performed on a LightCycler 480 Real-Time PCR System according to the manufacturer's instructions; the qPCR program was according to described previously 2 and primers used for analysis were listed in Supplementary Table 11. The reported values represent the mean of three biological replicates.
Candidate gene cloning and sequence analysis. When the candidate gene sequence was cloned, the total RNA was extracted from the bud of the CA1 and 740 by using a HiPure Plant RNA Mini Kit (Magen, China). The RNA from each sample was used for the reverse transcription reaction using an ImProm-II Reverse Transcription System (Promega, USA). The cDNA samples served as the template for amplification with the LA Taq DNA polymerase (TaKaRa, Japan) for the gene-specific marker (Supplementary Table 11), and the PCR products were cloned into a pMD19-T vector (TaKaRa, Japan). Positive clones were picked to culture for plasmid extraction and sequencing.
Co-expression analysis of gene expression. Co-expression network of gene expression was constructed with the weighted gene coexpression network analysis (WGCNA) package 58 using gene expression data of different tissue samples from different developmental stages 29 . The modules were obtained using the automatic network construction function blockwiseModules with default settings. The genes co-expressed network was visualized by the Cytoscape 3.0 59 .