Introduction

Cotton is the world’s leading natural fiber crop and is an important oil and protein crop (Cherry and Leffler 1984; Lusas and Jividen 1987; Alford et al. 1996; Chen et al. 2007). The cotton genus (Gossypium) comprises approximately 50 species, including 45 diploids and 5 allotetraploids (Fryxell 1979; Fryxell et al. 1992; Percival et al. 1999). The allotetraploid species arose about 1–2 million years ago through the hybridization of an A-genome taxon related to the species G. herbaceum L. (2n = 2x = 26), with a D-genome taxon related to the species G. raimondii Ulbrich and G. gossypioides L. (2n = 2x = 26) (Beasley 1940, 1942; Wendel et al. 1992). The ‘AD’ allotetraploid (2n = 2x = 52) group consists of the species G. barbadense, G. darwinii Watt, G. hirsutum, G. tomentosum Nuttall, and G. mustelinum Miers ex Watt (Percival et al. 1999).

Four Gossypium species are cultivated, including diploids G. arboreum L. and G. herbaceum L., and tetraploids, G. hirsutum L. and G. barbadense L. ‘Upland’ cotton (G. hirsutum L.) accounts for about 95% of the world’s total production (Chen et al. 2007). However, genetic diversity among modern Upland cotton cultivars is narrow, as revealed by isozyme analysis (Wendel et al. 1992) and various DNA markers including RFLP (Brubaker et al. 1993; Brubaker and Wendel 1994), RAPD (Multani and Lyon 1995; Guo et al. 1997; Iqbal et al. 1997; Xu et al. 2001; Linos et al. 2002; Lu and Myers 2002; Zhu et al. 2003), AFLP (Abdalla et al. 2001; Iqbal et al. 2001), and SSR (Zhu et al. 2003; Rungis et al. 2005; Lacape et al. 2007). Increasing Upland cotton diversity is essential for genetic improvement. Extensive genetic variation is available among members of the genus Gossypium (Percival and Kohel 1990; Khan et al. 2000; Lacape et al. 2007), and efforts to increase the genetic base of G. hirsutum can draw upon a host of cultivated varieties, as well as primitive domesticated landraces and the other allotetraploid species.

Interspecific germplasm introgression can increase genetic variation (Percy and Wendel 1990) and has been attempted in transferring specific cotton genes and useful traits including high fiber quality (Culp et al. 1979; Cantrell and Davis 1993) and low-gossypol seeds with high-gossypol plants (Vroh Bi et al. 1998, 1999a, 1999b). However, interspecific introgression often encounters problems such as segregation distortion (Jiang et al. 2000), suppression of recombination (Paterson et al. 1990), and linkage drag (Young and Tanksley 1989), and the use of interspecific germplasm has been limited in the breeding of G. hirsutum and G. barbadense.

DNA markers provide a useful tool for detecting and resolving complications such as segregation distortion or linkage drag encountered in interspecific gene introgression (Chee et al. 2005a). In particular, the advanced backcross approach facilitates the detection and integration of beneficial quantitative trait loci (QTL) from secondary gene pools into elite breeding lines (Tanksley and Nelson 1996). By using this approach, favorable alleles for agronomically important traits have been introgressed from G. barbadense into G. hirsutum (Chee et al. 2005a, b; Draye et al. 2005; Lacape et al. 2005; Saha et al. 2004, 2006).

The objectives of the present study were to (1) identify and characterize QTLs for fiber quality traits using advanced backcross self-populations segregating for G. tomentosum introgressed chromatin segments, (2) analyze the interaction between G. tomentosum introgressed chromatin segments and genetic background, and (3) analyze the interaction between G. tomentosum introgressed chromatin segments and the environment.

Materials and methods

Population development and phenotyping

An F2 population from a cross between G. hirsutum acc. TMS-22 and G. tomentosum acc. WT936 was used to develop the advanced backcross population. About 200 F2 seeds were grown in Lubbock (TX) in 1998, and only about 20 plants flowered. Eleven F2 plants were crossed successfully to G. hirsutum cultivar CA3084, and another six F2 plants were crossed successfully to G. hirsutum cultivar CA3093. From 1999 to 2001, the 17 F2BC1 families were backcrossed to CA3084 or CA3093, respectively, obtaining 17 BC3F1 families. Seeds of all BC3F1 families were planted in peat pellets and germinated in the greenhouse, and seedlings were hand-planted in the field (Lubbock, TX) during early April of 2002. Plants were spaced 30 cm apart within rows and rows were 152 cm apart. A total of 634 BC3F2 plants from two background populations were used to extract DNA and also were self-pollinated to produce BC3F3 populations in 2002. Eleven BC3F2 families in the CA3084 background included 319 plants ranging from 2 to 94 plants per BC3F1-derived family, and six BC3F2 families in the CA3093 background included 315 plants ranging from 16 to 93 lines per BC3F1-derived family. In 2003, the BC3F3 populations were grown in the field in Lubbock (TX) and Tifton (GA), respectively. Except for spacing (which was as described for the BC3F2 plants), cultural practices were normal for cotton production in Texas and Georgia. Seed cotton from mature bolls of BC3F2 and BC3F3 plants were hand-harvested and ginned on a saw gin. Some plants were sterile or produced insufficient lint for fiber analysis. Fiber samples from 218 progenies in the CA3084 background (ranging from 1 to 75 plants per family), and 243 progenies in the CA3093 background (16–73 plants per family) were sent to the Cotton Incorporated Textile Services Laboratory (Cotton Incorporated, Cary, NC) where fiber quality traits were determined by a High-Volume Precision Instrument (HVI; Zellweger-Uster, Knoxville, Tenn.). Lint percentage and lint yield of BC3F3 populations were only determined in Tifton (GA) in 2003.

Genotyping and data analysis

The genome composition of the introgression populations was obtained by genotyping BC3F2 plants with 448 RFLP markers from a previously constructed genetic map (TH map) containing 589 loci and covering 4,259.4 cM (Waghmare et al. 2005). A total of 97 and 62 informative RFLP markers were identified and genotyped for CA3084 and CA3093 background populations, respectively.

Associations between marker genotypes and fiber quality and/or yield traits were tested for statistical significance by one-way variance analyses for each marker locus segregating in the BC3F2/BC3F3 populations, using the GLM procedure of SAS 9.2 (SAS Institute Inc. 2008) and an F test significance threshold of P < 0.001. The R 2 of each variance analysis provided an estimate of the proportion of phenotypic variance explained by the corresponding marker locus. For significant marker–trait associations, the phenotypic effect of individual QTLs was estimated by MapQTL 6.0 (Van Ooijen 2009). The extreme paucity of homozygotes made it impossible to estimate gene action (additivity and dominance) at most loci, so the analysis of QTL effects in this population was essentially limited to estimating phenotypic effects of a single allele substitution. The QTL map was presented by MapChart 2.2 (Voorrips 2006). QTL nomenclature follows a method used in rice (McCouch et al. 1997), starting with ‘q’, followed by an abbreviation of the trait name [fiber elongation (FE), fiber fineness (FF) (Micronaire), fiber length (FL), fiber strength (FS), fiber length uniformity (FU), boll weight (BW), 25 boll numbers (BN), lint percent (LP), Lint yield (LY)] and the name of chromosome, then followed by a number. While most cotton fiber traits have intuitive definitions, we note that ‘fiber elongation’ refers to the maximum extension at which the yarn (composed of individual fibers) breaks.

For loci segregating in two or more families of one background, or in both backgrounds, two-way mixed model variance analyses were performed using the MIXED procedure of SAS 9.2 (SAS Institute Inc. 2008). The analytical model included genotype (G), family (F) and genotype × family (G × F) interaction, or background (B) and genotype × background (G × B) interaction as fixed factors. The interaction between genotype (G) and environment (E) was also analyzed by the MIXED procedure. Model parameters were estimated using the residual maximum likelihood (REML) method. Marker–trait association (genotype factor) was tested with an F statistic using a general Satterthwaite approximation for the denominator degrees of freedom (SAS Institute Inc. 2008). A likelihood-ratio statistic (ChiSq) was performed for the G × F, G × B and G × E interaction (Self and Liang 1987). Genotypic effects and interaction effects were considered significant if P < 0.001. This stringent threshold, similar to an LOD score of 3.0, is used to achieve an experiment-wise false positive rate no higher than 0.05.

Results

Population structure

Out of 448 informative RFLP markers in the TH genetic map, 129 showed segregation in at least one of the two background populations. Among these markers, 97 (average 8.8 per family) segregated among 11 BC3F1 families in the CA3084 background population, whereas 62 (average 10.3 per family) segregated among six BC3F1 families in the CA3093 background population. The 17 BC3F1 families included 461 BC3F2 individuals with phenotypic data (218 for CA3084 background, 243 for CA3093) ranging from 1 to 75 plants per family. The average allele introgression ratio is 4.32% (4.49% in CA3084, 4.06% in CA3093), much lower than the Mendelian expectation (12.5%). The introgressed chromosome segments covered a total of 1,010.0 cM (719.7 cM in CA3084, 546.3 cM in CA3093, 256 cM overlapping) and accounted for 23.7% (16.9% in CA3084, 12.8% in CA3093) of the TH genetic map. The total length of introgressed G. tomentosum segments per chromosome ranged from 7.9 cM (Chr. 2) to 131.6 cM (Chr. 19) with an average of 47.8 cM. A total of 28 overlapping introgressions were found in the two backgrounds.

Phenotypic performance

The statistical parameters reflecting fiber quality and yield traits are listed in Suppl Table s1 and the distributions of traits are presented in Fig. 1. The fiber quality and yield traits are normally distributed except for lint percentage of the CA3093 population and lint yield of both populations. For fiber length, fiber length uniformity, fiber strength and lint yield, trait means of the CA3084 background population were substantially and significantly higher (P < 0.0001) than those of the CA3093 background population, except that the higher fiber length uniformity of the CA3084 BC3F3 population in one location (Tifton, GA) was not statistically significant. Fiber fineness (Micronaire) is significantly lower for the CA3084 background population than the CA3093 population, another desirable attribute of CA3084. However, for fiber elongation and lint percentage, the trait means of the CA3093 background population are substantially and significantly higher (P < 0.0001) than those of the CA3084 population, except that the advantage of CA3093 for fiber elongation of the BC3F3 in one location (Tifton, GA) only reaches P < 0.005.

Fig. 1
figure 1figure 1

Frequency distribution of fiber quality and yield traits in the BC3F2/BC3F3 populations. Plain, hashed and empty bars indicate BC3F2 at Lubbock (TX), BC3F3 at Lubbock (TX), and BC3F3 at Tifton (GA), respectively

Full size image

Analyses of variance of fiber quality traits are presented in Table 1. Fiber quality traits showed highly significant genetic and environment effects (P < 0.0001), except that fiber uniformity of the CA3093 background population had a less significant genetic effect (P < 0.001) and non-significant environment effect. Among the five fiber quality traits, fiber elongation is the most affected by environment. For example, the minimum value in the CA3084 BC3F3 background population in one environment (Tifton, GA) is larger than the maximum in the other environment (Lubbock, TX) (Table s1; Fig. 1).

Table 1 Variance analysis of fiber quality and yield traits in two background populations
Full size table

Correlations among the five fiber quality traits generally followed patterns that are characteristic of cotton germplasm (Table 2). Fiber length, strength, and uniformity are significantly positively related to one another and significantly negatively related to fiber fineness (where lower values represent higher quality) and elongation. High lint percentage was weakly associated (r = 0.18, p < 0.01) with increased (undesirable) micronaire in CA3084, and strongly associated (r = −0.63, p < 0.001) with reduced fiber length in CA3093. Boll weight was strongly associated with reduced lint yield in both CA3084 and CA3093. High lint yield was weakly associated with both increased fiber length (r = 0.23, p < 0.001) and increased lint percentage (r = 0.20, p < 0.01) in CA3084.

Table 2 Correlation coefficients between different fiber quality and yield traits
Full size table

Correlations of fiber quality and yield traits across years and environments are generally high (Table 3), with only fiber length uniformity failing to be statistically significant between virtually all years and backgrounds. This result showed that fiber quality traits had relatively high heritability, except for fiber length uniformity.

Table 3 Correlation coefficients of fiber quality traits across years and environments
Full size table

Main-effect QTL

A total of 3,180 marker–trait associations (1,940 in the CA3084 background population, 1,240 in CA3093) were tested, and 73 reached statistical significance. These marker–trait associations are presented in a supplemental document (Suppl Table s2). Thirty-one significant marker–trait associations were on 11 A-genome chromosomes and 42 were on 11 D-genome chromosomes.

However, segregation distortion was extreme in this population, with some loci containing only one G. tomentosum allele in the entire population. While it is possible to statistically exclude the phenotype of one individual from being part of the same population as the remaining segregants (and all such statistically significant comparisons are shown in Suppl Table s2), it is not possible to ascribe a phenotypic effect to the unusual allele because a single individual is clearly insufficient to provide a sample that is ‘homogenized’ for the remainder of the genome. For example, we found statistically significant associations of fiber elongation with pGH276 on Chr. 6, and pAR873 on Chr. 9 that were due to the presence of G. tomentosum alleles at these two loci in the same individual. Further data would be necessary to distinguish which of these loci, or perhaps others, are responsible for the phenotypic difference. Among those codominant loci that remain heterozygous in BC3F1, heterozygotes occurred at an average frequency of 28.9% whereas G. tomentosum homozygotes occurred at 1.8% in BC3F2—in other words, only about one individual in three harbor at least one G. tomentosum allele at a locus that remains heterozygous in the BC3F1. To be 99% confident that individuals that all share one or more G. tomentosum alleles at one locus do not also all share G. tomentosum alleles at another unlinked locus, about 15 individuals must be examined. Accordingly, while we have listed all statistically significant marker–trait associations in the supplemental document (Suppl Table s1), we have only considered those cases in which the smaller genotypic class included at least 15 individuals as representing QTLs (Table 4).

Table 4 QTL for fiber-related traits in G. tomentosum chromosome introgressed populations
Full size table

Fiber elongation

Four QTLs were identified, including one in the CA3084 background population and three in CA3093 (Table 4; Fig. 2). Two of the four QTLs were identified in each of 2 years, and one in each of two locations. The phenotypic effects of these QTLs ranged from −0.72 to −0.14, explaining from 4.5 to 8.9% of phenotypic variance. The favorable alleles of all four QTLs originated from G. tomentosum. One QTL was located on an A-subgenome chromosome (11) and three on D-subgenome chromosomes (14, 19, 21).

Fig. 2
figure 2figure 2

QTLs for fiber quality and yield traits identified in G. tomentosum introgression Upland cotton populations. Red, green, and black bars indicate the G. tomentosum chromosome segments in CA3084, CA3093, and both CA3084 and CA3093 background populations, respectively

Full size image

Fiber fineness

Eight QTLs were identified in the BC3F3 generation (none in BC3F2), two in GA and six in TX, including three in CA3084 background and five in CA3093 (Table 4; Fig. 2). The phenotypic effects of these QTLs ranged from −0.11 to 0.42 micronaire units, explaining from 5.1 to 11.9% of phenotypic variance. The favorable alleles of seven QTLs (87.5%) originated from G. tomentosum and one (12.5%) from G. hirsutum. Four QTLs were located on A-subgenome chromosomes and four on D-subgenome chromosomes.

Fiber length

Four QTLs were identified on three chromosomes in the CA3084 background (Table 4; Fig. 2). The phenotypic effects of these QTLs ranged from 0.03 to 0.05 inches, explaining from 6.3 to 8.3% of phenotypic variance. One QTL (FL04.1) was identified in two environments. The favorable alleles of all four QTLs originated from G. hirsutum. Three QTLs were located on A-subgenome chromosomes and one on a D-subgenome chromosome.

Fiber strength

Four QTLs were identified, including two in the CA3084 background and two in CA3093 (Table 4; Fig. 2). One (qFS02.1) was identified in two environments. The phenotypic effects of these QTLs ranged from −1.13 to 0.89 cN/tex, explaining from 4.7 to 13.1% of phenotypic variance. The favorable alleles of three QTLs originated from G. tomentosum and one from G. hirsutum. One QTL was located on an A-subgenome chromosome and three QTLs were on D-subgenome chromosomes. One additional QTL, qFS05.1 shows no main effect but significant interaction between family lines.

Fiber uniformity

Six QTLs were identified on five chromosomes, including four in CA3084 background population and two in CA3093 (Table 4; Fig. 2). Each QTL was detected in only one environment. The phenotypic effects of these QTLs ranged from −0.61 to +0.76, explaining from 5.4 to 8.3% of phenotypic variance. The favorable alleles of four QTLs originated from G. tomentosum and two from G. hirsutum. All QTLs were located on D-subgenome chromosomes.

Boll number

One QTL was identified on Chr. 21 of the D-subgenome in the CA3093 background (Table 4; Fig. 2) in the BC3F3 generation only. This QTL explained 6.2% of phenotypic variance. The favorable allele originated from G. hirsutum.

Weight per boll

One QTL was identified on Chr. 4 of the A-subgenome in the CA3084 background (Table 4; Fig. 2). The phenotypic effect was 0.87, explaining 9.9% of phenotypic variance. The favorable allele originated from G. hirsutum.

Lint yield

Three statistically significant marker–trait associations were identified on three D-genome chromosomes in CA3084 background population (Table 4; Fig. 2). The phenotypic effects of the QTLs ranged from −0.28 to 0.23, explaining from 5.2 to 8.9% of phenotypic variance. The favorable alleles at two loci originated from G. tomentosum and one from G. hirsutum. Because lint yield is a complex trait of low heritability and we have data from only a single environment (GA BC3F3), we have not formally claimed these associations to be QTLs, but note them here to facilitate future meta-analyses that may provide corroborative data.

Interaction between genotype and background

Only one genotype × family interaction was found for the 39 markers segregating in two or more family lines in the CA3084 background population, whereas no genotype × family interaction was found for the 25 markers segregating in two or more family lines in the CA3093 background population. However, for the significant genotype × family interaction locus, FS05.1, the QTL main effect, was not significant. For the 28 markers segregating in both CA3084 and CA3093 background populations, seven significant genotype × background interactions were found but only two, qFF14.1 and qFU14.1, each diagnosed by the same RFLP marker (pAR815E3C) (Table 4), showed significant main effects.

Interaction between genotype and environment

For the five fiber quality traits, a total of 97 marker–environment associations were tested in the CA3084 background population, and only one genotype × environment interaction was found, involving qFS15.1 (Table 4). No interaction was found among the 62 marker–environment combinations in the CA3093 background population.

Discussion

Favorable QTL alleles from G. tomentosum, a wild cotton

Among the QTLs identified, favorable alleles originated from G. tomentosum for 18 (64.3%) and G. hirsutum for 10 (35.7%). G. hirsutum contributed all four favorable alleles for fiber length (all in the CA3084 background), and G. tomentosum contributed most favorable alleles for fiber elongation and fiber fineness. This result is consistent with the finding that favorable QTL alleles may be recovered from an apparently unfavorable parent (Tanksley and Nelson 1996; Xiao et al. 1996), of particular importance to cotton in view of the series of genetic bottlenecks that have constrained variation in its present gene pool.

It may be noteworthy that the traits, for which G. tomentosum contributed the preponderance of favorable alleles, fiber fineness and elongation, have a relatively short history of selection in scientific cotton breeding. Differences between the CA3084 and 3093 backgrounds show that there is heterogeneity for favorable traits in the G. hirsutum gene pool. For traits that have been under selection in G. hirsutum for a long time, repeatedly crossing the best with the best would ‘concentrate’ favorable alleles. The cases in which G. tomentosum conferred most of the favorable alleles may suggest that (for reasons we do not understand) the favorable alleles for those traits are at higher frequency in G. tomentosum than G. hirsutum. So, it may be more likely that traits for which G. tomentosum had many favorable alleles are those that had not been under long or strong selection in G. hirsutum.

Relatively more QTLs for fiber fineness (Micronaire) than other traits were identified in the present study, similar to an Upland cotton intraspecific population (Shappley et al. 1998) and a G. hirsutum × G. barbadense BC1/BC2 population (Lacape et al. 2005), but different from another G. barbadense introgression population which showed more QTLs for fiber length (Chee et al. 2005b). Rong et al. (2007) summarized fiber quality QTLs from ten (G. hirsutum × G. barbadense) populations and found more QTLs for fiber fineness than any other trait, although fiber elongation and fiber length showed more QTLs than the remaining traits.

Common QTLs shared by different populations

Among the 28 QTLs detected in the present study, eight (28.6%) controlling fiber quality traits were also found in the same chromosome regions (either associated with a same RFLP marker or overlapped by common marker) in previous studies with different populations (Paterson et al. 2003; Draye et al. 2005). These common QTLs included seven (qFF05.1, qFF13.1, qFF14.1, qFL08.1, qFL08.2, qFS22.1 and qFU14.1) identified in G. hirsutum × G. barbadense F2 populations (Paterson et al. 2003) and one (qFF05.1) in a G. barbadense introgression BC3F2 population (Draye et al. 2005). These common QTLs explained less than 10% of the phenotypic variance in the present study, reiterating prior findings that the advanced-backcross population may be particularly effective for mapping small-effect QTLs that escape detection in other studies (Chee et al. 2005a).

Relationships between QTLs for fiber quality traits

Previous studies showed that many QTLs controlling fiber quality traits often co-located on some chromosomal regions in allotetraploid cotton (Saranga et al. 2002; Paterson et al. 2003; Chee et al. 2005b; Lacape et al. 2005; Rong et al. 2007; Wan et al. 2007; Zhang et al. 2009). In the present G. tomentosum introgression populations, QTLs for fiber elongation corresponded with three QTLs for fiber fineness (qFE14.1 = qFF14.1, qFE19.2 = qFF19.1, qFE21.2 = qFF21.1) and one for fiber length uniformity (qFE14.1 = qFU14.1). QTLs for fiber fineness corresponded with one QTL for fiber length (qFF04.1 = qFL04.1) and one for fiber length uniformity (qFF14.1 = qFU14.1). QTLs for fiber strength corresponded with two QTLs for fiber length uniformity (qFS15.1 = qFU15.1, qFS22.1 = qFU22.1). These co-locating QTLs were partially responsible for the correlations among traits. For example, the two favorable alleles of fiber strength and fiber length uniformity present on the same two chromosomal regions contributed to the significant positive relationship between fiber strength and fiber length uniformity. Similar relationships were also found among fiber length-related traits in a G. barbadense introgression population (Chee et al. 2005b).

QTL interaction between genotype and background

In the present study, one G. tomentosum introgression region on Chr. 5, which segregated among more than two family lines in CA3084 background population, showed a significant genotype × family interaction for fiber strength, but the QTL main effect was not significant. Similar interactions between QTL and genetic backgrounds were observed in Upland cotton population with G. barbadense chromosome segments (Chee et al. 2005a, b; Draye et al. 2005) and in other crops (Bernacchi et al. 1998; Lecomte et al. 2004). A possible explanation for this observation is that some chromosomal regions may harbor linked alleles with opposite effects (Bernacchi et al. 1998). When the linkages were broken in some of the BC3F2 families, the QTLs may show either positive or negative effects depending on which allele is present. Alternatively, the effects of these QTLs may depend on the presence of genetic loci from other donor chromosome segments (Chee et al. 2005a). For the 28 common markers segregating in both CA3084 and CA3093 background population, two significant genotype × background interactions were identified. This result showed that the different G. hirsutum cultivars carried different alleles for fiber quality and yield traits, or different alleles at trans-acting loci influencing these QTLs.

QTL interaction between genotype and environment

The extent to which the performance of complex traits varies among environments reflects the complexity of genotype × environment interactions (Saranga et al. 2002; Paterson et al. 2003). In the present study, significant environmental variances were found for all fiber quality traits, and many significant environment effects were identified in the two populations. Only one fiber strength locus, on Chr. 15 in the CA3084 background population, showed significant genotype × environment interaction. This result showed that although many QTLs controlling fiber quality traits had significant environment effects, only a few QTLs had significant genotype × environment interactions.

QTL distribution across subgenomes

This study adds further evidence to prior studies, indicating that the majority of genetic variation associated with fiber quality in tetraploid cotton traces to the D-subgenome from a diploid ancestor that does not produce spinnable fiber (Jiang et al. 1998; Rong et al. 2007). Of the 28 QTLs, 18 (64.3%) claimed were on D-subgenome chromosomes, albeit with individual traits varying in their distribution from 100% (of six, fiber uniformity) D-subgenome to 100% (of one, weight per boll) A-subgenome. Favorable alleles from G. tomentosum were found for six of ten A-subgenome and 12 of 18 D-subgenome loci, suggesting that the evolutionary forces resulting in these favorable alleles apply similarly to the two subgenomes.

Synthesis

The present study showed that G. tomentosum contains many favorable alleles for fiber quality traits, offering a potentially important means to improve Upland cotton. Especially attractive are some favorable alleles that are linked together and might be co-introgressed. Although barriers to gene introgression from other allotetraploid species may exist (Stephens 1949; Jiang et al. 2000), the availability of DNA markers linked to QTLs identified in this and other studies (Paterson et al. 2003; Chee et al. 2005a, b; Draye et al. 2005; Lacape et al. 2005) may assist breeders in transferring and maintaining these traits during Upland cultivar development. Furthermore, because many QTLs from interspecific introgression are in a near-isogenic state in the advanced backcross population, the phenotypic effect measured for each QTL is likely to be a better predictor of its ultimate effect when transferred to other cultivated backgrounds (Chee et al. 2005b). The present QTL mapping offers an additional source of allelic variation to Upland cotton germplasm.

The levels and patterns of segregation distortion in these populations, briefly noted herein and further investigated in a companion study (Waghmare et al., submitted), have consequences for how G. tomentosum introgressions might best be deployed in crop improvement. As noted above, among those codominant loci that remain heterozygous in BC3F1, in BC3F2 heterozygotes occurred at an average frequency of 28.9% whereas G. tomentosum homozygotes occurred at 1.8%. This suggests that G. tomentosum introgressions might be especially attractive for use in F1 hybrid cottons, if genotypes homozygous for the introgressions can be stabilized. The use of F1 hybrids often mitigates ‘linkage drag’ associated with introgression, by virtue of the presence of adapted alleles from one parent. Natural selection against G. tomentosum introgressions, once they are allowed to segregate, provides a strong incentive to use F1 seed each year in that the benefits of the introgressions will quickly be lost in F2 seed from the F1 plants.