Introduction

Cotton is an important economic crop worldwide and represents the most important natural fiber in the textile industry. China is the leading producer of cotton, particularly in the North China Plain, the Yangtze River Valley, the Yellow River Valley, and the Xinjiang Uygur Autonomous (Northwest) Region. In recent years, the arable land per capita has rapidly decreased, reflecting the increasing human population and decreasing the amount of arable land due to urbanization. Therefore, enhancing fiber yield through breeding has become a major goal for most cotton improvement programs.

The genus Gossypium comprises 46 diploid (2n = 2x = 26) and 5 allotetraploid (2n = 4x = 52) species (Wendel and Albert 1992). The allotetraploid cotton species, including two commercially important cultivated species, G. hirsutum (Upland cotton) and G. barbadense (Sea-Island, or Pima, or Egyptian cotton), originated from a natural hybridization of an extant A and an extant D genome followed by chromosome doubling 1–2 million years ago (Wendel et al. 2009). Upland cotton produces more than 95 % of the world cotton fibers due to its high lint yield and wide adaptations, while G. barbadense is the only cultivated tetraploid relative of G. hirsutum and is valued for its premium fiber quality attributes. Combining the yield from G. hirsutum and the fiber quality from G. barbadense, has been an ongoing goal of cotton breeders and geneticists, with limited success (Zhang et al. 2014).

Cotton fibers are single-celled trichomes that terminally differentiate from single cells located in the epidermis of the ovule. Although all epidermal cells are potential fibers, only approximately 30 % of these cells actually differentiate into fibers (Basra and Malik 1984; Tiwari and Wilkins 1995). Morphologically, the initiation and elongation of each fiber cell is associated with the spherical expansion and protrusion of an epidermal cell above the ovular surface during anthesis. Accordingly, increasing fiber initials and, therefore, lint percentage (i.e., percent of lint weight in seedcotton weight) will contribute to high lint yield. Therefore, understanding the mechanisms that determine which epidermal cells differentiate into fibers is essential for increasing fiber productivity through genetic engineering and molecular breeding. The identification and characterization of quantitative trait loci (QTL) for yield and yield component traits in cotton have been a research focus for more than a decade. In cotton, QTL for agronomic traits, including yield and yield components, have been identified using F2 (Guo et al. 2007), F2/F2:3 (Zhang et al. 2005; Ulloa et al. 2005; He et al. 2005; Yu et al. 2013b), recombinant inbred lines (RILs) (Shen et al. 2007; Wang et al. 2007; Wu et al. 2009; Zhang et al. 2008; Lacape et al. 2010; Yu et al. 2012), or backcross inbred lines (BILs) (Yu et al. 2012, 2013a). Yu et al. (2013a) used a BIL population in G. hirsutum × G. barbadense and identified 39 QTL affecting yield and yield component traits, including lint yield, seedcotton yield, lint percentage, and boll weight. Said et al. (2013, 2015a, b) recently performed the most comprehensive meta-analyses of QTL reported to date. RNA-Seq has been recently used to study the transcriptomes of the two cotton species, i.e., G. barbadense (Pima cotton) and G. hirsutum (Upland including Acala cotton), during boll development (Pang et al. 2012; Lacape et al. 2012; Fang et al. 2014; Paterson et al. 2012). However, it is unknown whether these genes are yield-related or species-specific, and genetic and physical map-based cloning of lint yield-related genes has not been reported. In fact, no genome-wide analyses of differentially expressed genes (DEGs) have been performed to identify genes associated with lint yield and its component traits using an appropriate genetic population in cotton.

In recent years, microarray analysis and RNA-Seq have been employed to identify DEGs in cotton (e.g., Hinchliffe et al. 2010; Curtiss et al. 2011; Gilbert et al. 2013). Compared to a deep RNA-Seq, although the number of unigenes represented in the Affymetrix GeneChips was limited, the number of DEGs identified was similar to that between two NILs differing in fiber length from a deep genome-wide RNA-Seq (Jinfa Zhang, unpublished). Therefore, it is likely that each unigene in the GeneChips may represent several homologous genes when it was designed before the Upland cotton genome TM-1 was recently sequenced (Li et al. 2015; Zhang et al. 2015). Furthermore, the predicted number of genes in the tetraploid Upland cotton is known only until its genome was fully sequenced in 2015. However, there have been no significant reports to date on the use of the microarray array or RNA-Seq technology to identify lint-yield-related transcripts in cotton.

In the present study, RNA from developing fibers at 10 DPA (days post-anthesis) was extracted from both a high-yielding (HY) and a low-yielding (LY) groups each with three BILs and compared for a transcriptome analysis. Lint yield in cotton is the product of boll number, boll weight and lint percentage. Because developing bolls after 10 DPA usually does not abscise, boll number, as one of the important yield components, is determined at this time. Furthermore, 10 DPA fibers are the fastest-growing fibers after fiber initiation, so lint fiber numbers and part of the fiber length are also determined for another yield component-lint percentage. Of course, boll size at 10 DPA also contributes to the final boll weight at maturity. Therefore, we chose 10 DPA fibers to identify differentially expressed genes (DEGs) that may be involved in cotton yield determination during this important fiber developmental stage. However, it should be pointed out that many genes associated with fiber initiation at −3 to 1 DPA (the stage for lint fiber determination) and secondary wall synthesis after 15 DPA (another important stage for fiber weight determination) may be undetected using 10 DPA fibers.

The objective of the present study was to identify DEGs that are associated with lint yield by employing a microarray analysis using Affymetrix GeneChips and to obtain insight into the underlying molecular genetic differences of fiber development in cotton between two groups of backcross inbred lines (BILs) differing in lint yield.

Materials and methods

Plant materials

An interspecific backcross inbred line (BIL) population of 146 lines developed at New Mexico State University was used for the current study. The BIL population was developed from a cross between the high-yield (HY) parent (HYP) Upland cotton SureGrow 747 (SG 747) and the low-yield (LY) parent (LYP) Egyptian cotton (G. barbadense) Giza 75 through two generations of backcrossing using SG 747 as the recurrent parent followed by four generations of selfing. The 146 BILs with a designation of NMGA and their two parents were planted in five environments in three locations: Anyang, Henan province in 2006, 2007 and 2008; Wangjing, Anhui province in 2007; and Aksu, Xinjiang Uyghur Autonomous Region in 2007. The three locations represent the major cotton production regions with three different cultivation systems in China-Yellow River valley (Henan province), Yangtze River valley (Anhui province) and Northwest (Xinjiang Uygur Autonomous Region). The 148 entries were arranged in a randomized complete block design with two replications and single row plots in each environment. The growth conditions, experimental design, and trait including lint-yield evaluation were previously described (Yu et al. 2013a). To represent the BIL population, 17 BILs were selected from the BIL population based on lint yield, boll size, lint percentage and fiber quality traits and analyzed using least significant difference (LSD) tests for significant differences with the SAS software (SAS Institute Inc., USA).

For the current study on lint yield, 6 BILs were then chosen from the 17 BILs based on differences in lint yield and similarities in fiber quality traits. Three BILs (i.e., NMGA-026, NMGA-017, and NMGA-144, designated HY1, HY2, and HY3, respectively) were high-yielding (HY) with an average lint yield of 60.46 kg ha−1, fiber length of 29.32 mm, strength of 28.54 cN tex−1 and micronaire of 4.50; and three other BILs (i.e., NMGA-066, NMGA-095 and NMGA-023, designated LY1, LY2, and LY3, respectively) were low-yielding (LY) with 32.57 kg ha−1, 29.17 mm, 28.76 cN tex−1 and 4.55 for above traits, respectively. The two parents, i.e., SG 747 and Giza 75, differed significantly in lint yield (41.50 vs. 25.53 kg ha−1), fiber length (29.44 vs. 34.02 mm), strength (27.61 vs. 40.23 cN tex−1) and micronaire (5.55 vs. 4.55). The 6 lines, together with their two parents (HYP and LYP), were planted in the Experimental Farm at the Institute of Cotton Research (ICR), Chinese Academy of Agricultural Sciences (CAAS), Anyang, Henan Province, China (E 114°48′, N 36°06′) in the summer of 2009. The field plots were arranged using a randomized complete block design with three replications. Flowers at the day of flowering (i.e., 0 DPA) in each plot were tagged, and cotton boll samples were harvested at 5, 10, 15, 20, and 25 days post anthesis (DPA) before 10:00 a.m. and immediately placed on ice. To minimize experimental errors, all samples within each developmental stage were collected in the same day. Fibers on the ovules of 5–10 developing bolls per replication were dissected, frozen in liquid nitrogen, and stored at −80 °C until further processing.

RNA isolation from cotton fibers

Tissue samples for total RNA isolation were prepared from 1 g of cotton fiber harvested from ovules of each genotype at 5, 10, 15, 20, and 25 DPA on a replication basis. Total RNA was extracted using a Sigma Spectrum™ Plant Total RNA kit (Sigma–Aldrich, St. Louis, MO, USA) with an on-column DNaseI digestion according to the manufacturer’s instructions. An agarose gel electrophoresis was performed and the results showed high RNA integrity, as both 28S and 18S rRNAs in the total RNA were predominantly present (data not shown). The RNA quantity and quality were then determined based on the 260/280 nm ratio and 260/230 nm ratio using a DU®800 spectrophotometer (Beckman Coulter, Brea, CA, USA).

Microarray hybridization and data analysis

For the microarray experiments using the GeneChip® Cotton Genome Array (Affymetrix, Santa Clara, CA, USA), 2 mg of purified total RNA was extracted from 10 DPA fibers from the three biological replications of each HY and LY genotype. The GeneChip® Cotton Genome Array is a single GeneChip brand array comprised of 239,777 probe sets representing 21,854 cotton transcripts, and is based on EST sequences obtained from four cotton species: G. hirsutum, G. raimondii, G. arboreum, and G. barbadense (http://www.affymetrix.com/catalog/131430/AFFY/Cotton+Genome+Array#1_1). The extracted RNA was used for cRNA synthesis and hybridization to the Cotton GeneChip Array based on a standard protocol (Rodriguez-Uribe et al. 2014). The default Affymetrix method MAS 5.0, RMA was used for a primary normalization. The results were submitted to NCBI as a GEO record under the accession number GSE81134 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE81134).

A pairwise comparison between the HY and LY fibers was conducted to identify differentially expressed genes (DEGs). Using the Affymetrix GeneChip Operating System v1.4 (GCOS; Affymetrix), the relative mean signal, detection calls, signal-log ratios and change calls were independently calculated using four different algorithms for each probe set. Fold changes in expression (e.g., >twofold), the Microarray Suite (MAS) 5.0 probe set detection p values, and/or p values (e.g., Bonferroni-corrected p-value threshold <0.05) from the standard statistical tests (e.g., independent t test, Welch’s t test, Wilcoxon’s rank sum test, ANOVA, etc.) were used to identify genes with significant changes in expression. The DEGs were further screened using the Partek software (http://www.partek.com). The Basic Local Alignment Search Tool (BLAST) was used to identify a significant homology between a gene and known gene products. The BLAST search results were subsequently sorted based on gene function to identify DEGs between the HY and LY groups. The GO annotations for the DEGs were obtained using the Blast2GO program (https://www.blast2go.com/). For a pathway enrichment analysis, the DEGs were mapped to the terms in the KEGG (Kyoto Encyclopedia of Genes and Genomes) database using KOBAS 2.0 (KEGG Orthology-Based Annotation System) (Mao et al. 2005). KEGG pathways with corrected p values ≤0.05 were considered statistically significant.

RT-PCR and quantitative reverse transcription PCR

To perform an RT-PCR analysis, the first-strand cDNA was synthesized from 5 µg of total RNA from each tissue sample using the Superscript First-Strand Synthesis System (Invitrogen, Carlsbad, CA, USA). The cDNA fragments were then PCR amplified using gene-specific primers designed from selected DEGs and commercially synthesized (Invitrogen, USA) as listed in Supplementary Data 1 online. The PCR amplification was performed in a PTC-200 thermocycler (Scientific Support, USA). Gel images were then acquired with a gel-scanning imager (Gel Doc-2000, Bio-Rad, USA) and ethidium bromide staining. Based on RT-PCR results, four DEGs were selected for a further real-time quantitative RT-PCR (qRT-PCR) analysis. The qRT-PCR reactions were performed using iTaq™ SYBR Green Super Mix (Bio-Rad Laboratories, USA) via a Bio-Rad CFX96 real-time PCR detection system. The relative quantification (RQ) of the two targets in the treated samples was calculated by the formula \( RQ = E^{{(Ct,{\text{ctrl}} - Ct,{\text{sample}})}} \) , where E is the experimental efficiency of the qPCRs (Bennett et al. 2015). The 18S ribosomal RNA (rRNA) was used as the internal control. Significant differences in gene expression were determined using the t test.

Chromosome location of DEGs in yield trait QTL and QTL hotspots

To locate DEGs with yield and yield trait QTL as anchored by simple sequence repeat (SSR) markers, we collected the marker names in the map of [“Guazuncho2” (G. hirsutum) × “VH8-4602” (G. barbadense)] as published (Said et al. 2013, 2015a). The marker sequence data were downloaded from the CottonGen database (https://www.cottongen.org/data/download/ marker) (Yu et al. 2014). Then, the chromosome locations of markers and DEGs were identified using the BLAST program with the marker sequences and DEG sequences as the query and the genome sequence (Zhang et al. 2015) as the subject. The positions of yield trait QTL hotspot regions as reported by Said et al. (2013, 2015a, b), and the 24 QTL loci for yield as reported by Yu et al. (2013a) on chromosomes were identified using anchoring markers’ locations in the map. The candidate genes in the regions of QTL hotspots or loci were then identified and used for further studies.

Single nucleotide polymorphism (SNP) identification and confirmation using single strand conformation polymorphism (SSCP) analysis

Sequence alignment was conducted using BLAST. Potential SNPs were identified via visual inspection based on a comparative alignment between the 106 DEG sequences obtained from the Affymetrix GeneChip Cotton Genome Array and the 183,035 sequences with SNPs obtained from CottonGen (http://www.cottongen.org/data/download/marker) (E value ≤ 1×10−20). For SNP confirmation, primers annealing immediately upstream or downstream of an SNP site were designed as the forward or reverse primers, respectively (Supplementary Data 2). The primers were designed using the Primer Premier 6.0 software (http://www.premierbiosoft.com) based on the following parameters: 18–26 nucleotides (nt) (20 nt optimum) in primer length, annealing at 55 °C, 35–60 % for GC content, and 100–300 bp for PCR product size. Briefly, each 20 µL PCR reaction contained 10 ng of DNA, 10 µL of TaKaRa Premix Ex Tag, and 0.4 µL of each 20 µM primer. A touchdown PCR was performed in a Bio-Rad C1000 Thermal Cycler with the following conditions: denaturation at 94 °C for 3 min, followed by 29 cycles at 94 °C for 30 s, touchdown primer annealing beginning at 56 °C for 30 s and extension at 72 °C for 60 s, and a final extension at 72 °C for 10 min. The PCR products of SNP markers were resolved using the polyacrylamide gel electrophoresis according to the SSCP method of Lu et al. (2009). Polymorphic markers were scored as 0 for the absence and 1 for the presence of an SSCP fragment in the 17 BILs (from which the 6 BILs were selected for the current study) and were used for a correlation analysis with lint yield, yield traits and fiber quality using SAS (SAS Institute Inc., USA).

Results

Microarray and data analysis

Using the Affymetrix GeneChip Cotton Genome Array with multiple probes for each gene, a comparison between the HY and LY BIL groups further minimized complications due to background differences resulting from the use of only two NILs. The six BILs were selected based on lint yield, boll size, and lint percentage to form a high-yield (HY) group (including three BILs designated HY1, HY2, and HY3) and a low-lint yield (LY) group (with three BILs designated LY1, LY2, and LY3). The two groups of BILs had significant differences in seedcotton yield, lint yield, lint percent, and boll weight, but no significant differences in fiber length and elongation (see "Materials and methods"). Although genetically, the BILs showed 87.5 % recovery of the recurrent parent used in developing the BIL population from which the six BILs were selected, our simple sequence repeat (SSR) and SSCP-SNP analysis of 2349 markers showed that the genetic similarity between the HY group and the LY group was as high as 98.5 % (unpublished), indicating a highly similar background between the lines used in the study. Therefore, although two backcrosses were used in the development of the BIL population, the BILs selected in this study may be considered NILs.

A total of 1486 genes were differentially expressed (DE) between the two groups (≥twofold change; Bonferroni-corrected P value threshold <0.05) (Supplementary Data 3). Compared to the LY group, 278 (19 % of 1486) of the DEGs were up-regulated, and 1208 genes (81 % of 1486) were down-regulated in the HY group. A cluster analysis of the gene expression data from the 1486 DE genes (DEGs) separated all the three HY lines from two of the three LY BILs (Fig. 1). Specifically, HY2 and HY3 had the highest similarity in gene expression patterns, followed by the high similarity of gene expression between LY1 and LY2. The result demonstrated the reliability of the selection of BILs and grouping method used in the present study, providing a great confidence in the obtained results. However, one of the three LY lines, i.e., LY3 was grouped together with the HY group, indicating that high levels of similarity in phenotypes (here lint yield, yield traits and fiber quality) may not always be reflected by high levels of similarity in gene expressions. The DEGs identified between the two BIL groups may indicate a common gene regulation mechanism for fiber development-related (more specifically, yield-related) genes in cotton, but these genes need to be further identified through a co-localization analysis with lint yield QTL.

Fig. 1
figure 1

Clustering of the differentially expressed genes identified in developing fibers at 10 days post-anthesis (DPA) between three high-yielding (HY) and three low-yielding (LY) backcross inbred lines (BILs). HY1 NMGA-026, HY2, NMGA-017, HY3, NMGA-144, LY1, NMGA-066, LY2, NMGA-095, LY3, NMGA-023

Full size image

To identify the developmental processes that differed between the two groups of BILs, the 1486 DEGs were categorized into 38 GO terms based on biological process, cellular component and molecular function, respectively (Fig. 2). Under the cellular component category, DEGs were further categorized as cell (521 genes, 35.06 % of the 1486 DEGs) and organelle (373 genes, 25.10 %). For the biological processes, metabolic process (649 genes, 43.67 %) and cellular process (544 genes, 36.61 %) represented a majority of the DEGs identified. Under the molecular function category, binding (525 genes, 35.33 %) and catalytic activity (476 genes, 32.03 %) were the two most abundant subcategories followed by transporter activity, nucleic acid binding, transcription factor activity, structural molecule activity, and molecular transducer activity. Interestingly, binding, catalytic, and transporter activity were associated with hormone binding, hydrolase activity, transferase activity, and carbohydrate transporter activity.

Fig. 2
figure 2

Distribution of 1486 differentially expressed genes between three high-yielding (HY) and three low-yielding (LY) backcross inbred lines (BILs) based on Gene ontology functional classifications

Full size image

A gene expression enrichment analysis was further performed on the 278 genes upregulated in the HY group and 1,208 genes upregulated in the LY group. The 278 upregulated DEGs in the HY group were categorized into 37 GO terms based on the biological process, cellular component, and molecular function, respectively (Fig. 3). For the biological processes, macromolecule metabolic process (29 genes, 10.43 % of the upregulated 278 DEGs), and organic substance biosynthetic process (22 genes, 7.91 %) had the highest numbers of the DEGs identified among the GO terms. Under the molecular function category, ion binding (33 genes, 11.87 %) and transferase activity (22 genes, 7.91 %) were the two most abundant subcategories, followed by hydrolase activity, organic cyclic compound binding, oxidoreductase activity, transmembrane transporter activity, and small molecule binding. Under the cellular component category, the DEGs were further categorized as intracellular (54 genes, 19.42 %) and intracellular organelle (43 genes, 15.47 %). The 1208 upregulated DEGs in the LY group were also categorized into 39 GO terms based on the biological process, cellular component, and molecular function (Fig. 4). For the biological processes, organic substance metabolic process (443 genes, 36.67 % of the 1208 DEGs) and cellular metabolic process (407 genes, 33.39 %) represented a majority of the DEGs identified. Under the molecular function category, organic cyclic compound binding (346 genes, 28.64 %) and heterocyclic compound binding (346 genes, 28.64 %) were the two most abundant subcategories, followed by ion binding, small molecule binding, hydrolase activity, transferase activity, carbohydrate derivative binding, oxidoreductase activity, and protein binding. Under the cellular component category, DEGs were further categorized as intracellular (433 genes, 35.84 %) and intracellular part (427 genes, 35.35 %).

Fig. 3
figure 3

Distribution of 278 up-regulated differentially expressed genes in the high-yielding (HY) group as compared to the low-yielding (LY) group in developing fibers at 10 days post-anthesis (DPA) based on Gene Ontology functional classifications

Full size image
Fig. 4
figure 4

Distribution of 1208 up-regulated differentially expressed genes in the low-yielding group (LY) as compared to the high-yielding group (HY) in developing fibers at 10 days post-anthesis (DPA) based on Gene Ontology functional classifications

Full size image

The 1486 DEGs identified from the microarray analysis were further analyzed using KOBAS (i.e., KEGG Orthology-Based Annotation System) to identify the metabolic pathways in which they function. KOBAS mapped 467 of the 1486 DEGs to 102 KEGG pathways. The pathways with the highest unigene representation were starch and sucrose metabolism (32, 6.85 %), phenylpropanoid biosynthesis (24, 5.14 %), and phenylalanine metabolism (23, 4.93 %). As expected (Gilbert et al. 2013), starch and sucrose metabolism was the most abundant pathway in fiber development at 10 DPA (Table 1).

Table 1 Significantly enriched pathways of differentially expressed genes (DEGs) identified using KOBAS
Full size table

Corroboration of the microarray data

To validate the results obtained from the microarray analysis, a RT-PCR analysis of eight genes was first performed on developing fibers at 15, 20, and 25 DPA from the six BILs and their parents. The eight genes included ribosomal protein L16 (designated GhRPL16), kinesin-related 11 (designated GhKIF11), filament-like plant protein 4 (designated GhFPP4), a gene of unknown function (designated Unknown), alpha-expansin 4 (designated GhEXPA4), ABC transporter G family member 10-like (designated GhABC-10), bacterial-induced peroxidase (designated GhPOX), and glucose-1-phosphate adenylyltransferase large subunit 3 (designated GhAPL3). Once different expression levels were found among the eight genotypes, 4 of the 8 genes were chosen for a further quantitative RT-PCR (qRT-PCR) analysis on developing fibers at 10 DPA. These four genes included two up-regulated and two down-regulated DEGs in the HY group identified by the microarray analysis (Fig. 5). The selected genes were GhEXPA4 (GhiAffx.26845.1.S1_at), GhRPL16 (Ghi.10647.1.S1_s_at), GhAPL3 (Gra.1083.1.A1_at), and Unknown (GhiAffx.32232.1.S1_at). Based on the results from the six BILs and the two parents, the coefficients of correlation between microarray and qRT-PCR were positive for two genes and significantly positive for another gene, indicating a relative congruence between the microarray and qRT-PCR results. The microarray analysis of the alpha-expansin 4 gene showed a reduction in gene expression in LY compared to HY, and similar results were obtained from the qRT-PCR analysis (Fig. 5). One of the four probe sets targeting the Unknown gene (GhiAffx.32232.1.S1_at) exhibited a reduction in expression in the LY group as compared with the HY group, which was consistent with the qRT-PCR analysis (Fig. 5). Two of the four probe sets targeting the genes encoding for ribosomal protein GhPRL16 (Ghi.10647.1.S1_s_at) and glucose-1-phosphate adenylyltransferase large subunit 3 (Gra.1083.1.A1_at) exhibited increased expression in the LY group compared to the HY group, congruent with the results of qRT-PCR analysis (Fig. 5). The results indicated that the differential expression of the four genes between the HY and LY groups was consistent between both the qRT-PCR and microarray analyses.

Fig. 5
figure 5

Quantitative RT-PCR analysis of GhEXPA4, an unknown gene, GhRPL16 and GhAPL3, and its comparison with the results from a microarray analysis in developing fibers at 10 days post-anthesis (DPA). LYP, low-yielding parent Giza 75, LY1, NMGA-066, LY2, NMGA-095, LY3, NMGA-023, HY1, NMGA026, HY2, NMGA-17, HY3, NMGA-144, HYP, high-yielding (HY) parent SG747

Full size image

Two other genes, i.e., the gene coding for bacterial-induced peroxidase (GhPOX) (Ghi.3212.1.A1_at) and the gene coding for kinesin-related 11 (GhK1F11) (Ghi.10183.1.S1_at), were finally chosen for a dynamic qRT-PCR analysis of fibers at various fiber development stages (i.e., 5, 10, 15, 20, and 25 DPA). The expression of GhPOX was the highest in fibers at 10 DPA than in the other fiber developmental stages, and the expression of GhPOX was up-regulated in the LY group than in the HY group in 5-15 DPA, especially 10 DPA fibers (Fig. 6). However, for the GhKIF11 gene, its expression at the mRNA level was the highest at the 5 DPA fibers and then linearly decreased until 20–25 DPA at a negligent level. Furthermore, its expression in fibers at 5 and 10 DPA was higher in the LY lines than in the HY lines. These results were again consistent with the microarray analysis for the two genes in fibers at 10 DPA.

Fig. 6
figure 6

Expression profile of GhPOX and GhKIF11 in developing fibers at 5–25 days post-anthesis (DPA). LY1, NMGA-066, LY2, NMGA-095, LY3, NMGA-023, HY1, NMGA026, HY2, NMGA-17, HY3, NMGA-144

Full size image

Candidate DEGs in lint-yield QTL regions identified in the BIL population

The completion of the G. hirsutum (AD) genome sequencing (Li et al. 2015; Zhang et al. 2015) has made it possible to determine the chromosome distribution of the DEGs identified in this study. Based on published data for the cotton genome, out of the 1486 DEGs, 536 were mapped onto chromosome locations from At_chr1 to At_chr13 of the At subgenome, 698 to Dt_chr1 to Dt_chr13 of the Dt subgenome, and the remaining DEGs (i.e., 125 genes) were on unmapped scaffolds. Interestingly, 115 genes were homologous to genes on both the At subgenome and the Dt subgenome at the sequence identify of 79.89–99.45 %.

Co-localization of quantitative trait loci (QTL) with the identified 1486 DEGs will facilitate the identification of candidate genes for lint yield (LY) and its component traits (Supplementary Data 4). We mapped 212 DEGs with the 24 yield and yield trait QTL (Yu et al. 2013a), and 12 yield trait QTL hotspots per a previous meta-analysis (Said et al. 2013) (Supplementary Data 5). These 212 genes were distributed in A01 (4 DEGs), A03 (2), A05 (1), A07 (26), A09 (6), A10 (4), A11 (2), A12 (1), A13 (8), D02 (38), D03 (8), D04 (9), D06 (20), D07 (9), D11 (8), D12 (31), and D13 (25). Of the 212 DEGs, 156 were categorized into 47 GO terms based on the biological process, cellular component, and molecular function (Fig. 7). Under the cellular component category, the DEGs were further categorized as cell part (89 genes, 42.58 % of 212 DEGs) and membrane-bounded organelle (67 genes, 32.06 %). For the biological process, organic substance metabolic process (55 gene, 26.32 %), and primary metabolic process (55 gene, 26.32 %) had more DEGs identified. Under the molecular function category, ion binding (51 genes, 24.40 %), heterocyclic compound binding (30 genes, 14.35 %), and organic cyclic compound binding (30 genes, 14.35 %) were the three most abundant subcategories, followed by hydrolase activity, oxidoreductase activity, transferase activity, lyase activity, and isomerase activity.

Fig. 7
figure 7

Distribution of 212 differentially expressed genes between the high-yielding group and the low-yielding group in developing fibers at 10 days post-anthesis (DPA) that were co-localized with lint yield and yield trait quantitative trait loci (QTL), based on Gene Ontology functional classifications

Full size image

Since the six BILs were selected from a G. hirsutum × G. barbadense backcross inbred line (BIL) population analysis (Yu et al. 2013a), we focused on the 11 lint-yield QTL identified in the BILs to identify lint-yield-related DEGs. Of the 11 yield QTL, 7 were found to co-localize with 81 DEGs, as shown in Supplementary Data 6. The 7 QTL for lint yield were qLY-06A-c1-1 on At_chr1, qLY-06A-c7-1 on At_chr7, qLY-07A-c9-1 on At_chr9, qLY-07A-c13-1 on At_chr13, qLY-07X-c14-1 on Dt_chr2, and qLY-07X-c18-1 and qLY-06A-c18-1 in different regions on Dt_chr13.

The At subgenome carried more LY QTL with DEGs. The qLY-06A-c1-1 on At-chr1 had three co-localized DEGs encoding for a blue copper protein, a lysine histidine transporter-like 8, and a dva-1 polyprotein. The qLY-06A-c7-1 on At_chr07 had 2 co-localized DEGs encoding for an RNA-binding-protein 38 and an NAD-binding rossmann-fold superfamily protein. The qLY-07A-c9-1 locus had one co-localized DEG encoding for a mitotic checkpoint serine threonine-protein kinase isoform 2. The qLY-07A-c13-1 on At_chr13 carried a histone gene.

On the Dt subgenome, the qLY-07X-c14-1 locus had 30 co-localized DEGs encoding for a transcription factor 21, a calcium-dependent lipid-binding (domain) family protein isoform 3, protein far-red elongated hypocotyl 3, beta-glucosidase family protein, among others. The qLY-07X-c18-1 locus had 21 co-localized DEGs encoding for a DNA replication licensing factor mcm 4, branched-chain-amino-acid aminotransferase-like protein 2, protein root initiation defective 3-like, microtubule-associated protein 65-1 isoform 1, among others. The qLY-06A-c18-1 locus had one co-localized DEG encoding for a calmodulin-binding-protein isoform 1.

Mapping more DEGs with previously reported yield trait QTL hotspots on the G. hirsutum genome

Since the BIL population, including the six lines used in this study only mapped 7 QTL for lint yield, explaining 6.88–16.77 % of the phenotypic variation, we also used the 20 yield trait QTL hotspots reported from a meta-analysis (Said et al. 2013) to identify more lint-yield-related genes. A total of 70 DEGs were found to co-localize with 12 yield trait QTL hotspots for boll weight (BW), lint percent (LP), seedcotton yield (SCY), lint yield (LY), and boll number (BN), while other yield trait QTL hotspots carried no DEGs (Fig. 8). The 12 yield trait QTL hotspots with DEGs were mapped onto 11 chromosomes, including 4 At and 7 Dt chromosomes, i.e., At_chr3, At_chr9, At_chr12, At_chr13, Dt_chr2, Dt_chr3, Dt_chr4, Dt_chr6, Dt_chr7, Dt_chr12, and Dt_chr13 (Fig. 8).

Fig. 8
figure 8figure 8figure 8

Mapping differentially expressed genes (DEGs) with lint yield and yield trait quantitative trait loci (QTL) hotspots on the G. hirsutum genome

Full size image

Of these 12 yield trait QTL hotspots, 6 were mapped to 3 pairs of homoelogous chromosomes (i.e., At_chr3 vs. Dt_chr3, At_chr12 vs. Dt_chr12, and At_chr13 vs. Dt_chr13). The c3_BN/LP_mQTL_Hotspots(5) was identified at 1.76–2.82 cM, and carried 5 QTL on At_chr3, and contained 2 DEGs, encoding for an isoform 3 and a pectin lyase-like superfamily protein isoform 1. The c17_LP/BW/_mQTL_Hotspots(4) with 4 QTL residing at 22.25–39.58 cM on Dt_chr3 contained 8 DEGs, including genes encoding for proteins, such as ubiquitin carboxyl-terminal hydrolase isozyme l5, RNA polymerase II transcriptional coactivator kelp, luminal-binding-protein 5, activating signal cointegrator 1, DNAJ heat shock N-terminal domain-containing, RNA pseudouridine synthase 7 isoform × 2, copper transporter 1-like, and scarecrow-like protein 9. Since the two hotspots were not located in the same regions of the two homeologous chromosomes (At_chr3 vs. Dt_chr3), no common DEGs were identified.

Similarly, no common DEGs were identified for QTL on other two pairs of homeologous chromosomes. The c12_LP/YLD/LP_mQTL_Hotspots(5) was identified at 76.44–77.19 cM on At_chr12 and carried 5 QTL, and it contained only 1 DEG encoding for a serine threonine-protein kinase. As a comparison, the c26_BW/LY/LP/YLD_mQTL_Hotspots(4) was identified at 50.67–59.04 on Dt_chr12 and carried 4 QTL, and it contained 16 DEGs, including genes encoding for an acid beta-fructofuranosidase-like, protein cornichon homolog 4-like, protein iq-domain 1, among others. Another QTL hotspot on At_chr13 named c13_LY/BW/SCY_mQTL_Hotspots(3) was identified at 79.04-79.57 cM and carried 3 QTL, containing 5 DEGs coding for alpha-glucan isozyme, histone h4, histone h2b, probable serine threonine-protein kinase, and serine threonine-protein phosphatase 6 regulatory subunit 3. Only one DEG was co-localized with the c18_LP/BW_mQTL_Hotspots(4) carrying 4 QTL at 59.98–60.35 cM on Dt_chr13, which encodes for a rubisco subunit binding-protein alpha subunit.

In addition to the above three At-subgenome chromosomes, the c9_LP/SCY/LY/BN_mQTL_Hotspots(10) with 10 QTL ranged from 1.12 to 6.98 cM on At_chr9 and contained 5 genes, including genes encoding for a calcium-activated outward-rectifying potassium isoform 1, alpha- glucan phosphorylase l chloroplastic amyloplastic for 2 genes, chaperone protein chloroplastic, and e3 ubiquitin-protein ligase ring1-like.

Four additional Dt subgenome chromosomes carried yield trait QTL hotspots. A yield trait QTL hotspot, i.e., c14_BW/LP/LY_mQTL_Hotspots(9) carrying 9 QTL, was located at 0–5.13 cM on Dt_chr2, and it contained 8 DEGs, encoding for proteins, such as calreticulin 3 isoform 1, selenium-binding-protein 2-like, cytochrome p450, chalcone synthase, dihydroxy-acid chloroplastic, glutamine synthetase leaf chloroplastic, non-specific lipid-transfer protein at2g13820-like, and transcription factor 21. The c22_LP/BW/SCY_mQTL_Hotspots(12) with 12 QTL ranged from 46.37 to 50.57 cM in Dt_chr4, and contained 6 DEGs encoding for a magnesium transporter mrs2-4-like, organ-specific protein s2-like, a squalene monooxygenase, a polygalacturonase at1g48100, an eukaryotic translation initiation factor 5b, and an unknown gene. The c25_BW_mQTL_Hotspots(13) ranged from 0.69 to 2.82 cM with 13 QTL on Dt_chr6, and it contained 6 DEGs, including genes encoding for a c3hc4 transcription factor, an allene oxide synthase, an ent-kaurene chloroplastic, a hypothetical protein F383_19939, a probable isoaspartyl peptidase l-asparaginase 2, and a proteasome subunit alpha type-3. The c16_LY/YLD/LP/SCY/LY_mQTL_Hotspots(6) with 6 QTL ranged from 18.34 to 29.32 cM on Dt_chr7, and contained 9 co-localized DEGs coding for a ubiquitin-conjugating enzyme19 isoform 1, a sulfite reductase, a hydrophobic lti6a, a transcription factor myb86-like protein, a uncharacterized calcium-binding-protein at1g02270-like isoform x1, an auxin response factor 18, an aspartic proteinase nepenthesin-1, an o-acyltransferase wsd1-like, and a golgi snap receptor complex member 1–2.

Single nucleotide polymorphic (SNP) locus identification and association with agronomic traits

To illustrate the utility of the DEGs identified in the present study, a sequence alignment was conducted via BLAST with 106 gene sequences from DEGs against 183,035 sequences in a reference database downloaded from the CottonGen database (http://www.cottongen.org/data/download/marker). The results showed that 56 gene sequences did not contain SNPs, whereas 45 gene sequences contained 209 putative SNPs (E value ≤ 1 × 10−20). An example of the results of the PCR amplification using the SSCP method for genomic DNAs from the six NILs used in this study is shown in Fig. 9. Among the 45 primer pairs, 24 pairs (53 %) yielded polymorphic SSCP markers, revealing a total of 81 SSCP-SNP loci (Supplementary Data 7). One gene (gb|DW225231.1) was identified to encode for a protein kinase and it carried 6 SNP loci which was located in c25_BW_mQTL_Hotspots(13). Another gene (gb|L04497.1) was identified to encode for myb-like transcription factor 1 and contained 1 SNP locus which was located in the same region for QTL qLY-07A-c13-1 and qSCY-07A-c13-1. A further correlation analysis between SSCP markers and 10 agronomic traits in the 17 BILs revealed that 26 markers were significantly (at P < 0.05 for 22 markers) or highly significantly (at P < 0.01 for 4 markers) associated with one or more agronomic traits, including 7 markers for yield component traits, and 3 makers for lint yield, 17 markers for fiber quality, and 1 marker for cottonseed oil content (Supplementary Data 8). Interestingly, repulsion linkage phases were found for several polymorphic SSCP markers amplified using the same primer pairs designed for 3 genes, resulting in opposite correlations. This may reflect the nature of co-dominance of these markers amplified by the same primer pairs designed from the same genes.

Fig. 9
figure 9

Electrophoretic analysis of single strand conformation polymorphism (SSCP)-based single nucleotide polymorphic (SNP) markers developed for the genes encoding for MYB (a), RDL (b), and MADS-3 (c). 1 LY1, 2 NMGA-060, 3 HY3, 4 NMGA-101, 5 HY1, 6 NMGA-062, 7 NMGA-092, 8 NMGA-043, 9 NMGA-140, 10 NMGA-089, 11 LY3, 12 LY2, 13 NMGA-070, 14 NMGA-051, 15 NMGA-105, 16 NMGA-012, and 17 HY2

Full size image

Discussion

The present study performed a microarray-based comparative transcriptome analysis on developing fibers at 10 days post-anthesis (DPA) between the two groups (high yield—HY vs. low yield—LY) of BILs. A total of 1486 differentially expressed genes (DEGs) were identified, providing the first glimpse of DEGs into the relationship to genetic variation of fiber yield in cotton. Several fiber development-related genes that exhibited altered expression profiles in HY group and LY group were validated through qRT-PCR. Specifically, gene expression levels of GhKIF11 and GhPOX genes were relatively higher at 15 and 20 DPA in the LY group than in the HY group. We further mapped 212 DEGs with the 24 yield QTL and 11 yield trait QTL hotspots as reported by others, and 81 DEGs with the 7 lint-yield QTL mapped in the BIL population from which the two groups of BILs were chosen for this study. Molecular markers from the DEGs closely associated with the lint-yield-related QTL were developed. The results will provide a good foundation for fine mapping and map-based cloning of lint-yield genes and facilitate targeted gene transfer from G. barbadense to Upland cotton to further improve cotton yield.

Mapped DEGs with the yield QTL and yield trait QTL hotspots as reported by others, and the SSCP markers associated with important agronomic traits from the association analysis provide a good foundation for map-based cloning and MAS of lint-yield QTL. Thus, this study took one of the first comprehensive genomic approaches to genotype and profile the interspecific Pima × Upland backcross inbred lines during fiber development to associate DNA markers and gene expression levels with yield traits (Michaelson et al. 2009; Pang et al. 2012). In cotton, several microarray platforms have been developed and used to study the expression of genes associated with fiber development (Arpat et al. 2004; Shi et al. 2006; Wu et al. 2005, 2007; Lee et al. 2007), fiber traits or cotton evolution via comparisons of genotypes with differing fiber qualities (Hinchliffe et al. 2005, 2010; Wu et al. 2008; Li et al. 2010), or normal genotypes vs. mutants (Lee et al. 2006; Shi et al. 2006; Yang et al. 2006; Bolton et al. 2009; Gilbert et al. 2013). However, a comprehensive comparison between the results obtained in the present study and previously published results is difficult to obtain due to the use of different genetic systems. To our knowledge, this fiber yield DEG study is the first to use genotyping and expression profiling to examine BILs from an interspecific Egyptian cotton × Upland cotton BIL population during fiber development to associate DNA markers and gene expression levels with yield traits.

Lint yield is the product of boll number, boll weight and lint percentage. Boll weight is the sum of lint weight (lint fiber number × fiber length × fiber fineness) and seed weight, and lint percentage is the ratio of lint weight by boll weight. Fiber fineness can be measured by the diameter of the mature fiber, but it is usually determined by micronaire (i.e., the fiber weight per-unit fiber length). Therefore, it should be recognized that the number of mature fibers developed from fiber initials is also a contributing factor to lint yield. To identify DEGs for the understanding of its genetic and genomic basis, it will require a similar comparative transcriptome study among BILs or NILs differing in fiber initials using ovules at the fiber initiation stage, i.e., −3 to 1 DPA. Furthermore, because fiber fineness (usually measured based on micronaire), as an important contributing factor to fiber weight, is majorly determined during secondary wall biosynthesis stage after 15 DPA, many genes associated with fiber fineness are unlikely detected using 10 DPA fibers. Therefore, the current study is not intended to detect most DEGs that are associated with lint yield and its component traits due to the use of 10 DPA fibers that did not cover the entire spectrum of fiber initiation and development and the genetic variation in yield limited by the BILs used in this study.

Since the microarray analysis was performed in 2009–2010 prior to the publication of the tetraploid cotton genome sequences (Li et al. 2015; Zhang et al. 2015), two DEGs genes, i.e., GhKIF11 and GhPOD, were chosen for a dynamic analysis of gene expression during fiber development. GhKIF11 expression levels were relatively higher at 15 and 20 DPA in the LY group compared to the HY group, but it was not located in any of the QTL regions for lint yield and yield traits. Previous studies have identified more than a dozen of kinesins that are abundantly expressed during fiber development in Upland cotton. Kong et al. (2015) reported a member of the kinesin-4 subfamily named GhKINESIN-4A that shares the highest homology to the FRA1 (fragile fiber 1) protein previously reported in Arabidopsis thaliana. FRA1 or KINESIN-4A has been implicated in the deposition and orientation of cellulose microfibrils during secondary cell wall synthesis in fiber cells in A. thaliana (Zhou et al. 2007). A kinesin-like calmodulin-binding-protein involved in long-chain fatty acid biosynthesis showed greater than 50-fold accumulation in 10 DPA fiber cells compared with 0 DPA ovules (Ji et al. 2003). Preuss et al. (2003, 2004) proposed that MT-based motor kinesins actively participate in fiber growth and identified several kinesin-encoding genes that are abundantly expressed in developing cotton fibers. Tiwari and Wilkins (1995) reported diffuse cell expansion of fiber development for more than 10 days prior to the initiation of secondary cell wall deposition. Although these results demonstrate that the GhKIF11 gene is closely associated with fiber cell wall biosynthesis, the differential expression of GhKIF11 between the LY and HY groups was not one of the causal genes underlying lint-yield QTL in our study. Its expression may be directly or indirectly regulated by one of the lint-yield QTL, and therefore, it is indirectly related to lint yield.

The GhPOX (encoding for peroxidase) expression levels in the LY group were also relatively higher at 10 and 15 DPA, compared with the HY group. The transcript level of GhPOX increased >5.0-fold at 10 DPA (i.e., the period of fast fiber cell elongation) in the LY group. Most plant POX isozymes catalyze oxidoreductions between H2O2 and variable reductants and have been implicated in many physiological processes, such as the polymerization of cell wall compounds, defense against pathogen attacks, salt tolerance, oxidative stress, and the metabolism of phytohormones and alkaloids (Hiraga et al. 2001; Gabaldón et al. 2005; Passardi et al. 2005; Costa et al. 2008). A total of 73 AtPOX and 138 OsPOX genes were identified within the A. thaliana and Oryza sativa genomes, respectively (Tognolli et al. 2002; Welinder et al. 2002; Duroux and Welinder 2003; Passardi et al. 2004). A microarray analysis of G. hirsutum gene expression revealed that the Class III peroxidase genes were up-regulated in fast elongating fiber cells (Shi et al. 2006). Another microarray analysis indicated that GhPOX1 plays an important role during fiber cell elongation through the production of reactive oxygen species (Mei et al. 2009). Similar to GhKIF11, this gene was also not located in any of the lint-yield QTL regions, and therefore, is unlikely one of the candidate genes for lint-yield QTL.

However, through a comparative genome-wide gene expression study between two groups of BILs contrasting in lint yield and co-localization of DEGs with lint yield and yield trait QTL, 1486 DEGs were narrowed to 212 genes in the QTL regions, providing an important list of genes for further studies. An in silico analysis of the QTL-co-localized DEGs based on the sequenced TM-1 genome allowed identification of putative single nucleotide polymorphisms (SNPs) for the development of SSCP-SNP markers for an association analysis with the lint yield and yield traits, resulting in the identification of several yield and yield trait-associated DEG-based SNP markers. Although more studies are needed, this positional (physical map) candidate gene approach will assist in the fast and targeted identification of candidate genes for lint yield and other traits to facilitate molecular breeding and map-based cloning of QTL in cotton.