INTRODUCTION

Herbaceous cotton (Gossypium hirsutum L.) is among the three most important crops in Brazilian agribusiness. Brazil stands out as the world’s fourth-largest producer and third largest exporter of cotton (^{FREIRE et al., 2015}; ^{ABRAPA, 2017}; ^{FAO, 2018}; ^{USDA-FAS, 2018}). The Brazilian Midwest is the largest producer, followed by the Northeast region. For both regions in the 2017/2018 growing season, production estimates were over 200 thousand tons (^{CONAB, 2018}).

In Brazil, fiber production has always been based on the cultivation of white cotton fiber, with a large number of cultivars with great productive potential and excellent fiber quality. Cotton breeding programs in the twentieth century were directed at white cotton fiber, while those of colored fiber have been used only by hand or as an ornamental plant (^{CARVALHO et al., 2015a}).

The interest in colored cotton fiber cultivation in the Northeast region by family farming has increased in the current century, mainly because farmers sell colored fibers at a better price when compared to white fiber. To meet this demand, Embrapa Cotton has been carrying out breeding programs since the mid-1990s to select colored fiber cultivars with good yield associated with fiber traits within the standards of the textile industry (^{CARVALHO; ANDRADE; SILVA FILHO, 2011}; ^{CARVALHO, 2016}). However, climate irregularities in the semiarid of the Northeast region led to the need to develop and select new cotton cultivars that, in addition to being productive and having superior fiber quality, are adapted or tolerant to water stress (^{CARVALHO et al., 2015a}; ^{DEEBA et al., 2012}; ^{DHIVYA et al., 2014}).

Therefore, colored cotton fiber cultivars have become an important alternative for producers, as they have a high economic value and profitability, adding value to northeastern agriculture and changing the perspective of family farming, which needs a new dynamic to become competitive in the current market.

Several studies have been developed to select colored cotton fiber genotypes for the semiarid region. ^{Carvalho et al. (2005)} observed the effect of mass selection on colored fiber genotypes to obtain genetic gains for the percentage of lint, while ^{Carvalho et al. (2015a)} verified the presence of interaction genotype x environment (GxE) and determined the adaptability and phenotypic stability of eleven brown-colored cotton fiber genotypes under irrigated and rainfed conditions.

Thus, this study aimed to select superior genotypes of colored cotton fiber for agronomic and technological traits of fibers with wide adaptability to the Brazilian semiarid conditions.

MATERIAL AND METHODS

RESULTS AND DISCUSSIONS

The summary of joint variance analysis (Table 2) showed significant differences (p<0.01) for all traits. As such, the genotypes have genetic variability and hence favorable responses to selection. Regarding the environmental effect, the genotypes showed significant differences (p<0.01) only for traits LP and LEN. Yet for GxE interaction, significant differences were observed only for SCY (p<0.01) and MIC (p<0.05), that is, the genotypes have different behaviors in each environment for these traits. Similar results were found by ^{Carvalho et al. (2015a)}, who observed different behaviors of colored cotton fiber genotypes as a function of the environment, indicating the need for further studies of adaptability and phenotypic stability.

All evaluated genotypes showed trait averages (Table 2) in accordance with mean standards for colored herbaceous cotton medium fiber, except for LP (35.91%). This parameter has a recommendation of above 40% (^{VIDAL NETO; FREIRE, 2013}).

CVe (%) presented changes (Table 2). The highest CVe was found for SCY (12.13%) and the lowest for LP (3.71%), indicating a good experimental precision. Similar results for experiments with cotton cultivation in semiarid regions were found by ^{Cordão Sobrinho et al. (2015)}, ^{Queiroz et al. (2017)}, and ^{Vasconcelos et al. (2018)}.

Regarding genetic parameters from joint variance analysis (Table 2), coefficient of genotypic determination (CGD) ranged from 85.73% (SCY) to 97.60% (LP), showing a significant presence of genotypic variability regarding total phenotypic variability for all traits. This indicates the possibility of significant gains in the selection process. According to ^{Bonifácio, Mundim, and Sousa (2015)}, CGD above 70% can be considered high and desirable; however, it varies according to species and assessed trait. ^{Cruz, Regazzi, and Carneiro (2012)} reported that heritability coefficient estimates, which corresponds to CGD for fixed models, are important for choosing an effective selection strategy.

Our CGD estimates were higher than those found in other studies (^{KHAN et al., 2010}; ^{RESENDE et al., 2014}; and ^{KOTHARI el al., 2016}). These authors obtained discrepant estimates, ranging from 77% to 87% for LEN and from 0% to 81.10% for MIC. Such differences are due to genotype and environmental conditions.

According to ^{Carvalho et al. (2016)}, a coefficient of genetic variation (CVg) (Table 2) can be interpreted as the magnitude of the genetic variation available for selection, among which the highest are mostly desired. The highest CVg estimates were found for traits SCY (10.51%), MIC (10.29%), and STR (10.05%), while LEN (8.01%) presented the lowest value. ^{Resende et al. (2014)} obtained a higher CVg for LP (14.010%) when studying cotton.

The CVg/CVe ratio quantifies the proportion of genetic in relation to environmental variance (BORÉM; MIRANDA; FRITSCHE-NETO, 2013) and expresses the degree of selection difficulty. In our study, CVg/CVe estimates were higher than the unit in all traits (Table 2), except for SCY (0.86), i.e., offering advantages for selection. According to ^{Vencovsky and Barriga (1992)}, values similar to or higher than one are favorable to selection.

Table 3 shows the grouping of means for the studied parameters from the ^{Scott and Knott (1974)} test at 0.05 probability (mean of both studied environments). Several groups were formed for all the traits, thus confirming the genetic variability of these parameters among genotypes.

Two groups were formed for SCY (Table 3), standing out the genotypes CNPA 2002 10087 (3618.83 kg ha^-1) and CNPA 2002 10398 (3485.42 kg ha^-1), which exceeded the controls. This finding corroborates the results found by ^{Carvalho et al. (2015b)}. ^{Zonta et al. (2015a)} reported water deficit influence on yield of cotton cultivars under varied irrigation depths, with values lower than those verified in our study.

Five groups were formed for LP, standing out the control BRS Topázio (41.96%), followed by the genotypes: CNPA 2002 10327, CNPA 2002 10398, CNPA 2001 5823, and CNPA 98 1034. The control BRS Verde was classified in the last group, with the lowest LP. A similar result was found by ^{Carvalho et al. (2015b)}, who reported an overall mean of 36.02% for LP. ^{Cordão Sobrinho et al. (2015)} used five water regime conditions and observed no influence of water depths on LP, finding higher values for LP for the cultivars BRS Aroeira (44.25%) and BRS Araripe (42.77%). ^{Zonta et al. (2015a)} also observed results similar to ours using a 360 mm water depth for cotton irrigation.

Three groups were formed for LEN, standing out the genotypes CNPA 2002 10575, CNPA 2002 10576, CNPA 2002 10245, and CNPA 2002 10327, which exceeded the controls. ^{Carvalho et al. (2015b)} used the irrigation regime recommended for cotton cultivation and observed similar means for length compared to our results. ^{Zonta et al. (2015a)} found similar means for length using different irrigation depths, while ^{Cordão Sobrinho et al. (2015)} reported that the amount of water made available to cotton caused a direct effect on length.

The group formed by genotypes CNPA 2002 10575, CNPA 2002 10576, CNPA 2006 3667, CNPA 2006 3334, and CNPA 2002 10327 stood out for STR, surpassing all the other genotypes and controls. Similar results were found by ^{Carvalho et al. (2015b)}, ^{Zonta et al. (2015b)}, and Cordão Sobrinho et al. (2015), using irrigation depths similar to the ones we assessed.

Three groups of genotypes were formed for MIC. The lowest MIC was expressed by the control BRS Verde (2.95 µg in^-1), followed by the genotype CNPA 2002 10245 (4.12 µg in^-1). These results exceeded those found by Cordão Sobrinho et al. (2015) and ^{Zonta et al. (2015b)}, while ^{Carvalho et al. (2015b)} observed micronaire estimates ranging from 2.70 to 4.25 µg in^-1 when using the irrigation regime recommended for cotton.

CONCLUSION

Significant genetic variability was found among the studied colored cotton genotypes, allowing significant gains in selection for the semiarid conditions of northeastern Brazil.

The best performing genotypes were BRS Topázio, CNPA 2002 10245, and CNPA 2002 10327, as they achieved high cotton yields associated with superior fiber characteristics, with satisfactory tolerance to water stress. Thus, these genotypes can be used in breeding programs for colored cotton fiber quality in the semi-arid region of northeastern Brazil.