Introduction

This is the third study describing the results of a backcross-self-approach to map and introgress quantitative trait loci (QTLs) for fiber quality traits from Gossypium mustelinum into Upland cotton (G. hirsutum). In the previous two studies, we described the interspecific genetic linkage map and 24 QTLs for fiber elongation (15 detected in BC3F2 and 14 in BC3F2:3 and BC3F2:4 but five are common; Wang et al. 2016a, b). For one-third of these QTLs, the G. mustelinum alleles improved fiber elongation, indicating the potential benefit of introgressing genes from this species in breeding for higher quality Upland cotton.

In this study, we report the results of interspecific introgression and QTL analysis for fiber length, which is considered by some to be the most important property of cotton in raw material marketing and yarn processing (Kuang and Yu 2015). The most widely used parameters for fiber length include the average length of the longer half of the fiber span length distribution (i.e., upper-half mean length, UHM); length uniformity or uniformity ratio or fiber uniformity index (UI) which is defined as “a ratio between the mean length and the upper-half mean length of the fibers and is expressed as a percentage” (Testore and Minero 1988); and short fiber content (SFC), which is currently specified as the percentage of fibers shorter than 10, 11, 12, or 12.5 mm in the great majority of cases. Rotor yarn strength initially increases with staple length (fiber length) up to 40 mm and then remains constant, although with coarse fibers there is a slight increase up to 60 mm (Lawrence 2003). Longer fibers can be processed at greater efficiencies and produce finer yarns with higher quality than short fibers, since short fibers require increased twisting during spinning, reducing production speed, and causing changes to other yarn properties that impair fabric properties (Lawrence 2003). Low uniformity index is often accompanied by a high content of short fibers, which reduces the quality of the textile product.

The objectives of the investigation reported in this study were to determine the number, chromosomal locations, and phenotypic effects of QTLs for fiber length in an interspecific G. hirsutum × G. mustelinum backcross-self-population and to investigate the relationship between QTLs associated with UHM, UI, and SFC. The QTLs identified in this study add a new level of information to our knowledge of the genetic basis of fiber quality, also laying the foundation for marker-assisted introgression of G. mustelinum alleles to improve fiber quality of Upland cotton.

Materials and methods

Population development and field evaluation

A modified backcross-self-mating design was used in this study. Advanced-generation backcross populations were developed by first crossing G. hirsutum acc. PD94042 and G. mustelinum (AD4-8), then independently backcrossing F1 plants to the G. hirsutum parent for three cycles. A total of 21 lineages led to the production of at least one BC3F1 plant, each of which was self-pollinated to generate 21 BC3F2 families ranging in size from 127 to 160 plants per family (totally 3203 BC3F2 progeny) that were planted in 2006 in Tifton Georgia. In addition, 12 BC3F2:3 and BC3F2:4 families of 130–160 lines per family (totally 1826 lines) were derived from 12 of the 21 BC3F2 families with enough seeds, and were planted in completely randomized designs with two replicate plots for each line in 2008 and 2009 in Tifton, Georgia. All cultural practices followed standard recommendations for Georgia cotton production as described (Wang et al. 2016a). Seed cotton was hand-harvested from all bolls of each plant for BC3F2 family, and from 50 bolls from the middle of each plot (to avoid edge effects) for BC3F2:3 and BC3F2:4 families, and ginned on a saw gin. For phenotyping, the fiber quality data of each individual were collected for BC3F2, whereas the fiber quality data of two replicated lines were collected for each line in BC3F2:3 and BC3F2:4 generation. The family mean values of 21 BC3F2 families and 12 BC3F2:3 and BC3F2:4 families were used to demonstrate the phenotypic distribution of fiber length traits measured with family mean deviation from the recurrent Gossypium hirsutum parent. Three traits reflecting fiber length parameters, specifically UHM, UI, and SFC were determined by the Cotton Incorporated Textile Services Laboratory (Cotton Incorporated, Cary, N.C.) using the High-Volume Precision Instrument (HVI; Zellweger-Uster, Knoxville, Tenn.).

Genotyping and data analysis

A total of 218 SSR markers that were approximately evenly distributed on our G. hirsutum by G. mustelinum map constructed from an F2 population of the same two parents comprising 1055 loci (Wang et al. 2016b) were used for genotyping. The markers that detected introgression of G. mustelinum alleles in the BC3F1 were used to screen every individual in each BC3F2 family, which constitutes the genotype of each BC3F2 individual and also the corresponding BC3F2:3 and BC3F2:4 lines.

Since many populations segregated for G. mustelinum alleles in only small segments of the genome, we tested associations between fiber length phenotypes and marker genotypes for statistical significance by one-way variance analyses for every marker locus segregating within each BC3 family. The GLM procedure of the SAS ver.8 software package (SAS Institute 1999) was used to perform the analyses, with a significance threshold set at P < 0.001 (F-test). The appropriate linear combination of the estimated model parameters was used to evaluate the modes of gene action (additive, a; and dominant, d) for individual QTLs, and their significance levels were assessed with corresponding contrasts as described (Paterson et al. 1990). QTLs were considered to be over- or under-dominant if the absolute value of the dominance/additive (d/a) ratio exceeded 3 (Chee et al. 2005a). The R 2 of variance analysis at a marker locus estimated the proportion of phenotypic variance accounted for by a nearby QTL.

In order to confirm the reliability of the QTLs detected by one-way variance analyses and also to map epistatic QTLs, QTLNetwork V2.1 (Yang et al. 2008) was also used to analyze QTLs for fiber length traits in each of the 21 BC3F2 families and 12 BC3F2:3 and BC3F2:4 families. The mixed model-based composite interval mapping (MCIM) method was used with the critical F value calculated based on 1000 permutation tests, with window size and walk speed set at 10 and 1 cM, respectively. The significance threshold was set at P = 0.001 to claim a putative main-effect or epistatic QTL. QTLs for fiber length traits were mapped in each of the 21 BC3F2 families and 12 BC3F2:3 and BC3F2:4 families. QTLs were also detected in joint analysis of the 12 families grown in three generations/environments (BC3F2, BC3F2:3, and BC3F2:4), also considering environmental effects. QTLs detected by QTLNetwork 2.1 sharing a common marker with those detected by one-way variance analyses were considered to be the same QTLs. QTL nomenclature was as described (McCouch et al. 1997), with the QTL name beginning with “q” indicating a QTL, followed by an abbreviation of the trait name, the name of chromosome (Wang et al. 2016b; Zhang et al. 2015b), and consecutive numbers when more than one QTL was detected on the same chromosome for the same trait. The sequences and genome positions of all the markers that linked to QTLs are available from CottonGen (Yu et al. 2014a; Zhang et al. 2015b).

For loci segregating in two or more families, the MIXED procedure of the SAS ver. 8 package was used to perform two-way mixed model variance analyses, including genotype (G) as a fixed factor and family (F) and genotype × family (G × F) interaction as random factors. The residual maximum likelihood (REML) method was used to estimate model parameters, and an F statistic using a general Satterthwaite approximation for the denominator degrees of freedom (SAS Institute 1999) was applied to test marker-trait association (genotype factor). A likelihood-ratio (ChiSq) test was performed for the G × F interaction (Chee et al. 2005a; Self and Liang 1987). A significance level of P < 0.001 was set for G and G × F effects.

Results

Phenotypic distribution and correlations

The distributions of fiber quality traits of the BC3 progenies are shown in Table 1, Supplementary Fig. 1. All traits expressed transgressive segregation in the three generations, with distribution ranges being wider in BC3F2 than BC3F2:3 and BC3F2:4 (Table 1). Although G. mustelinum does not produce spinnable fiber, many BC3 progenies have higher UHM and UI, and lower SFC (usually lower SFC is preferred) than the G. hirsutum parent (Table 1). For UHM, six of 21 BC3F2, four of 12 BC3F2:3, and three of 12 BC3F2:4 families had mean values exceeding that of the G. hirsutum parent; three families, namely POP10, POP11, and POP27 showed higher mean values than the G. hirsutum parent in all generations. For UI, 15 of 21 BC3F2, no BC3F2:3, and only 1 of 12 BC3F2:4 families had mean values exceeding that of the G. hirsutum parent. For SFC, 16 of 21 BC3F2, but only 2 of 12 families in both BC3F2:3 and BC3F2:4 had lower mean values than the G. hirsutum parent (Supplementary Fig. 1). Correlations were calculated to detect relationships between the three fiber length traits. There was significant positive correlation between UHM and UI (P < 0.01), and significant negative correlation of SFC with both UHM and UI (P < 0.01) in all three generations (Table 2).

Table 1 Summary statistics of fiber quality traits measured on the G. hirsutum parent and BC3 progenies
Full size table
Table 2 Correlation coefficients between different fiber quality traits based on mean values of three generations
Full size table

Main effect QTLs detected for each trait

A total of 65 non-overlapping QTLs were estimated to be segregating in the three generations (Table 3). Blocks of linked markers within a family that showed significant marker-trait association (P < 0.001) were inferred to represent single QTLs. These QTLs were mapped to 19 chromosomes with 26 on eight A-subgenome chromosomes, and 39 on 11 D-subgenome chromosomes. In 19 of the 21 families we detected one or more QTLs for fiber length traits, with a maximum of eight QTLs in family POP02. Twenty-three of the 65 QTLs could be detected at least twice near linked markers in the same family or near the same markers in different families/generations. For example, as shown in Table 3, the QTL qUHM-5-1 was detected near two markers (BNL2656 and NAU3498) in the same family, POP15; another QTL (qUHM-24-1) was detected near the same marker, DPL0068, in two different families, POP16 and POP20.

Table 3 Biometrical parameters of QTLs affecting fiber length parameters
Full size table

Twelve QTLs explained more than 20% of PV, and four could be detected at least twice near linked markers in the same family or near the same markers in different families/generations. Thirty-two of the 65 QTLs were also detected by the MCIM method of QTLNetwork. The biometrical parameters of all significant marker-trait associations are listed in Table 3. A summary of the QTLs detected for each trait follows.

QTLs for UHM

A total of 19 non-overlapping QTLs were detected on 14 chromosomes for UHM (Table 3), with five in the A-subgenome and 14 in the D-subgenome. Eleven were found at least twice near linked markers in the same family or near the same markers in different families/generations, namely qUHM-3-1, qUHM-5-1, qUHM-7-1, qUHM-8-1, qUHM-14-1, qUHM-19-2, qUHM-21-1, qUHM-24-1, qUHM-24-2, qUHM-25-1, and qUHM-26-1. The PVE per individual association ranged from 9.05% (qUHM-19-2) to 32.01% (qUHM-24-2), with an average of 16.06%; and five QTLs explained more than 20% of PV (qUHM-8-1, qUHM-11-1, qUHM-23-1, qUHM-24-2, qUHM-26-1). For 14 of the 19 QTLs, alleles from G. hirsutum increased UHM, which was consistent with the parental phenotypes. For five of the 19 QTLs, alleles from G. mustelinum increased UHM (Table 3). Twelve of the 19 QTLs were also detected by the MCIM method of QTLNetwork (Table 3).

QTLs for UI

A total of 20 non-overlapping QTLs for UI were mapped to 14 chromosomes with ten mapped to seven A-subgenome chromosomes and ten to seven D-subgenome chromosomes. Six QTLs were found at least twice near linked markers in the same family or near the same markers in different families/generations. The PVE per individual locus ranged from 8.52% (qUI-5-1) to 26.48% (qUI-24-1), with an average of 14.66%; and five QTLs explained more than 20% of PV (qUI-11-1, qUI-12-2, qUI-12-3, qUI-24-1, qUI-26-1). For 19 of the 20 QTLs, G. hirsutum alleles increased UI, consistent with the parental phenotypes. For one of the 20 QTLs (qUI-14-1), the G. mustelinum allele increased UI (Table 3). Ten of the 20 QTLs were also detected by the MCIM method of QTLNetwork (Table 3). Among these 10 stringent QTLs, four explained more than 20% of PV; one (qUI-24-1) was found in two generations in the same family.

QTLs for SFC

A total of 26 non-overlapping QTLs of SFC were mapped to 13 chromosomes; eleven to six A-subgenome chromosomes and 15 to seven D-subgenome chromosomes. Six QTLs were detected at least twice near linked markers in the same family or near the same markers in different families/generations. The PVE per individual locus ranged from 8.56% (qSFC-7-2) to 21.26% (qSFC-24-3), with an average of 13.77%; and two QTLs explained more than 20% of PV. For 25 QTLs, G. mustelinum alleles increased SFC, consistent with the parental phenotypes. For one of the 26 QTLs (qSFC-17-2), the G. mustelinum allele decreased SFC (Table 3). Ten of the 26 QTLs were also detected by the MCIM method of QTLNetwork (Table 3).

Co-location of main-effect QTLs for different traits

Sixteen cases of co-location of QTLs for different traits suggest that marker-assisted selection based on these QTLs may improve each of three measures of fiber length simultaneously (Table 3). Co-location of QTLs for UHM and UI were found in four regions, namely qUHM-5-1 and qUI-5-1 near BNL2656 and NAU3498; qUHM-8-1 and qUI-8-1 near CIR354b and NAU4900; qUHM-23-3 and qUI-23-2 near BNL3511; and qUHM-23-1 and qUI-23-1 near DPL0378. Two cases of QTL co-location were detected for UHM and SFC, namely qUHM-17-1 and qSFC17-1 near BNL2496A; and qUHM-19-3 and qSFC-19-2 near DPL0140. Five cases of QTL co-location were detected for UI and SFC, namely qUI-10-1 and qSFC-10-2 near JESPR6; qUI-12-2 and qSFC-12-3 near CIR293; qUI-12-3 and qSFC-12-2 near DPL0866; qUI-12-1 and qSFC-12-1 near JESPR300; and qUI-26-1 and qSFC-26-1 near NAU3862. Five cases of co-location of QTLs for all three traits simultaneously were found, namely qUHM-7-1, qUI-7-1, and qSFC-7-2 near MUCS616 and NAU2002; qUHM-19-2, qUI-19-1, and qSFC-19-3 near NAU5489; qUHM-24-2, qUI-24-2, and qSFC-24-2 near NAU3605; qUHM-24-2, qUI-24-1, and qSFC-24-1 near TMHB1; and qUHM-26-1, qUI-26-2, and qSFC-26-2 near STV122. The phenomenon of QTL clusters has been found in many previous studies of cotton (Chee et al. 2005b; Tang et al. 2015; Yu et al. 2013b; Zhang et al. 2011, 2015a). The nature and direction of the correlation between the traits suggest that these co-located QTLs may be conferred by pleiotropic effects of the underlying genes; however, recent evidence that cotton fiber quality QTLs may involve large numbers of co-expressed genes serving different functions offers an intriguing alternative hypothesis (Paterson et al. 2012). One way to test these hypotheses (without making transgenics) would be to develop a segregating population with recombinants in these regions of co-located QTLs (Kalladan et al. 2013; Paterson et al. 1990). In our research, QTLs for SFC showed opposite allelic effects to those of UHM or UI in most cases, with the only exception being qUHM-19-3 and qSFC-19-2 near DPL0140. UHM and UI always showed allelic effects in the same direction (Table 3), consistent with the correlations between them (Table 2).

Consistency of QTLs across families

Among the 218 SSR marker loci, 211 were segregating in two or more families; therefore, we performed two-way ANOVA to test for marker-trait associations and assess their consistency among families. Significant (P < 0.001) among-family G effects were detected at 18, 16, and 16 loci for UHM, UI, and SFC, respectively (Supplementary Table 1, Table S1). For UHM, the 18 loci appeared to represent only 11 non-overlapping genomic regions on 11 different chromosomes, for which QTLs were detected near six loci in within-family analysis (qHUM-3-1, qUHM-7-1, qUHM-21-1, qUHM-23-2, qUHM-24-1, qUHM-24-2). For UI, the 16 loci represent 13 non-overlapping genomic regions on 13 different chromosomes, for which QTLs were detected for six loci in within-family analysis (qUI-2-1, qUI-12-1, qUI-12-2, qUI-18-1, qUI-23-1, qUI-24-1). For SFC, the 16 loci represent 13 non-overlapping genomic regions, for which QTLs were detected for four loci in within-family analysis (qSFC-12-1, qSFC-12-3, qSFC-16-2, qSFC-24-1).

A total of six loci were significant (P < 0.001) for G × F interactions for UHM (Supplementary Table 2, Table S2). Four of these loci detected three QTLs (qUHM-21-1, qUHM-24-2, qUHM-26-1; Table 3, Table S2) in different segregating families, while the fifth locus (BNL2725) just missed significance (P < 0.002) for POP16. Two loci showed G × F interactions for UI (Supplementary Table 2, Table S2). The locus BNL1151 also detected a QTL (qUI-11-1; Table 3, Table S2) in POP31, while the other locus (NAU915) just missed significance (P < 0.003) for POP32. Two loci were significant (NAU915 and NAU4045, P < 0.001) for G × F interactions (Supplementary Table 2, Table S2) for SFC; and both loci just missed significance (P < 0.005 in POP04 for NAU4045, P < 0.002 in POP32 for NAU915) for declaring QTLs.

Epistatic QTLs and their interactions with environments

A total of 17 epistatic QTLs with significant additive × additive (AA) effects (P < 0.001) were identified, with the majority (14/17) involving loci not linked to any main-effect QTLs (Table 4). Two interactions were able to be detected in a single environment and also by joint analysis: for UHM, the interaction between a region on Chr4 and a region on Chr9 was detected in both BC3F2:3 and joint analysis in POP10 simultaneously, with alleles from G. mustelinum increasing UHM; for SFC, the interaction between a region on Chr6 and a region on Chr12 was identified in both BC3F2 and joint analysis in POP32 simultaneously.

Table 4 Estimated epistasis and epistasis × environment interaction effects of QTLs for fiber length traits
Full size table

Discussion

The cotton genus Gossypium L. comprises seven tetraploid species—while G. hirsutum provides most of the world’s cotton production, its narrow genetic diversity forms the bottle neck restricting its further improvement and leading to increase of genetic vulnerability (Paterson et al. 2004). G. barbadense and the five wild tetraploid cotton species [G. tomentosum Nutall ex Seemann, G. mustelinum Miers ex Watt, G. darwinii Watt, G. ekmanianum Wittmack, and Gossypium sp. nov.(Wendel and Grover 2015)] provide resources that can be used to improve G. hirsutum traits, including fiber quality. Many studies have been performed to exploit elite genes from G. barbadense to improve fiber quality in G. hirsutum (Chee et al. 2005a, b; Draye et al. 2005; Lacape et al. 2010; Rong et al. 2004; Said et al. 2015; Shi et al. 2015; Wang et al. 2011, 2013; Yu et al. 2013a, b, 2014b). Exploratory efforts have also been made to construct genetic linkage maps and/or identify fiber quality QTLs from G. tomentosum (Hou et al. 2013; Khan et al. 2016; Waghmare et al. 2005; Zhang et al. 2011) and G. darwinii (Chen et al. 2015; Wang et al. 2012).

The results of this study build on a rich and growing body of evidence that DNA marker-assisted approaches offer the potential to extract from wild tetraploid cottons alleles of value in improvement of elite cultivated cottons. G. mustelinum, the basal species in the tetraploid Gossypium clade, is a wild tetraploid cotton endemic to the semi-arid region of northeastern Brazil and is sexually compatible with cultivated cotton (G. hirsutum or G. barbadense). It is not known to have yet been bred or commercially exploited (Alves et al. 2013; Borém et al. 2003), or subjected to domestic use by indigenous people in its range that may have contributed to the improvement of important traits such as fiber quality (Alves et al. 2013).

In this study, AB-QTL analysis was performed based on SSR markers and phenotypic data from three generations of BC3-derived families. Phenotypic evaluation of the mapping population showed transgressive segregation in the progenies for all fiber length traits in each of the three generations, indicating the presence of both positive and negative alleles for each trait in each parent (Supplementary Fig. 1; Table 1). Since G. mustelinum has inferior fiber length traits relative to the G. hirsutum parent, it is not a surprise to find that most BC3 families (15 of 21 in BC3F2, eight of 12 in BC3F2:3, and nine of 12 in BC3F2:4) have lower mean UHM values than the recurrent G. hirsutum parent, PD94042. However, it is interesting that for UI and SFC, most BC3F2 families (15 of 21 for UI, and 16 of 21 for SFC) outperformed the recurrent parent (Supplementary Fig. 1). Swamy et al. (2012) reported transgressive segregation for most grain quality traits of rice with 75% of families showing at least 5% increase over the cultivated parent for several traits. Wickneswari et al. (2012) also found that 62.8 and 74.8% of BC2F2 families outperformed the recurrent parent for grain yield and 1000 grain weight. In our research, even though G. mustelinum has inferior fiber length, it contributed positively to UI and SFC in most BC3F2 families and also to UHM in some families. In these families, many individual plants have better fiber length traits than the recurrent parent (Table 1). The transgressive segregation suggests that many new gene combinations formed by interspecific hybridization are desirable, although some are clearly not; the accumulation and interaction of superior alleles from the two divergent parents may have played an important role in phenotype expression. Such ‘positive’ transgression is of particular importance in view of the genetic bottlenecks that have constrained variation in the elite Upland cotton gene pool (Paterson et al. 2004).

Advanced backcross mapping populations have more uniform genetic background than early generation crosses, which increases the ability to detect even QTLs with small effects (Nagata et al. 2015; Tanksley and Nelson 1996). Advanced backcross populations have been created in many crops and made noteworthy contributions in revealing favorable alleles from wild species (Wang and Chee 2010), such as tomato (Fulop et al. 2016; Kinkade and Foolad 2013), barley (Haas et al. 2016; Kalladan et al. 2013), rice (Kim et al. 2015; Nagata et al. 2015), maize (Trachsel et al. 2016), peanut (Burow et al. 2014), and wheat (Naz et al. 2015). Some of these QTLs were further fine-mapped and cloned, e.g., six genes controlling fruit weight and shape variation in tomato played important roles in tomato breeding, and may also aid fundamental and applied research in other plants (van der Knaap et al. 2014). However, linkage of QTL alleles in repulsion may decrease the ability to detect QTLs and result in underestimation of the additive effect when the two QTLs are segregating (Kroymann and Mitchell-Olds 2005; Nagata et al. 2015). In addition, problems may exist in QTL mapping for families only in the BC3F2 generation, since genotype by environment interactions affect phenotype evaluation (Table 4)—phenotype of only a single individual was measured for each genotype in all BC3F2 families, which might also lead to escape of some QTLs from detection. By also evaluating BC3F2:3 and BC3F2:4 generations for 12 of the 21 families with enough seeds, using multiple individuals of each genotype, QTLs mapped in the BC3F2 generation can be validated, also detecting additional ones missed in the BC3F2.

From the three generations studied, one-way analysis of variance detected 19, 20, and 26 non-overlapping QTLs with PVE of 16.06%, 14.66%, and 13.77% on average for UHM, UI, and SFC, respectively; twelve of the QTLs explained more than 20% of PV (Table 3). As expected, G. hirsutum alleles increased UHM and UI and reduced SFC for most QTLs, consistent with the parental fiber phenotypes. The effects of many QTLs showed good reproducibility, with 23 detected at least twice near adjacent markers in the same family or near the same markers across different families/generations. In addition, 32 of the 65 QTLs were detected in both one-way variance analyses and the MCIM method of QTLNetwork (Table 3). Although the G. mustelinum parent does not produce spinnable fiber, G. mustelinum alleles increased UHM for five of the 19 QTLs, also increasing UI for qUI-14-1 and decreasing SFC for qSFC-17-2, which can be detected in two generations in POP20—these QTLs are of great interest to be further exploited by transfer into Upland cotton. Efforts are now underway to construct near-isogenic introgression lines that will allow these G. mustelinum alleles to be more readily accessible in breeding programs for improving fiber quality.

Using data accumulated over three generations, epistatic QTLs were detected in addition to main-effect QTLs for fiber length (Table 4). Both main-effect and epistatic QTLs play important genetic roles in controlling cotton fiber length, with environment also playing a role but to a lesser degree than additive and epistatic effects (Table 4). It is interesting that most of the epistatic interactions (14/17) were detected between genetic background loci not linked to any QTL (Table 4), indicating the complexity of fiber length inheritance. Naturally, introgressing and testing epistatic alleles involves at least twice the difficulty associated with utilizing main-effect alleles from G. mustelinum (Chee et al. 2005b; Tang et al. 2015; Yu et al. 2013b; Zhang et al. 2011, 2015a). More than one G. mustelinum BC3 family often segregated for the same chromosomal segment in our research, making it possible to investigate the effects of genetic background on introgressed chromatin. Significant (P < 0.001) among-family G effects were detected at 18, 16, and 16 loci for UHM, UI, and SFC, respectively, with several of these loci revealing QTLs in within-family analysis (Supplementary Table 1, Table S1). Some among-family G effects showed good reproducibility, with six of 18 for UHM, two of 16 for UI, and one of 16 for SFC detected in different generations (Table S1). In addition, six, two, and two QTL regions for UHM, UI, and SFC present in multiple families showed G × F interactions, indicating that the introgressed region may have a strong positive effect on these traits in one family but a lesser effect in other families (Table 3, Table S2). Similar interactions between QTL and genetic backgrounds were also observed in Upland cotton populations introgressed with G. barbadense (Chee et al. 2005a, b; Draye et al. 2005) and G. tomentosum chromosome segments (Zhang et al. 2011).

This study also add further to widespread reports of the large contribution to the genetic control of fiber quality traits of the tetraploid D-subgenome, derived from a diploid progenitor that does not produce spinnable fiber (Chee et al. 2005b; Jiang et al. 1998; Zhang et al. 2008, 2011). In the present study, among the 65 QTLs that affect fiber length properties, the D-subgenome accounted for 39, much more than the 26 in the A-subgenome. The use of interspecific gene combinations appears to offer an important means by which to improve fiber length traits in Upland cotton. Although barriers to gene introgression from G. mustelinum to G. hirsutum may exist, the availability of DNA markers linked to G. mustelinum QTLs identified in this research could assist breeders in transferring and maintaining these traits during cultivar development. Further, since many QTLs from G. mustelinum are now in a near-isogenic state in the AB-QTL population, the phenotypic effect evaluated for each QTL measured here is likely to better predict its eventual effect when transferred to other cultivated backgrounds.

Author contribution statement

AHP conceived and designed the experiments; PWC and OLM oversaw crossing, phenotyping and genotyping, which was performed by BW, ZZ, ZZ, and ELL; DJ oversaw fiber analysis; BW, XD, and ML analyzed the data; BW, AHP, and PWC wrote the paper.