We retrieved RNA-seq data for 35 BioProjects from National Center for Biotechnology Information (NCBI) through literature mining, encompassing 603 samples across 13 tissues (Supplementary Table S1). Genomic data pertinent to tobacco were obtained from the Sol Genomics Network database (https://solgenomics.net/). The genome published by Edwards et al. [13] was used as a reference.

The original data were converted to fastq format using fastq-dump from the SRA Toolkit v3.0.5 (https://github.com/ncbi/sra-tools). Subsequently, adapter sequences and low-quality sequences in the fastq were removed using fastp with default parameters [14]. After obtaining clean data, the transcriptome data were mapped to the reference genome using STAR (v2.7.10a) [15]. Subsequently, StringTie (v2.1.7) [16] with default parameters was used in conjunction with the annotation of the reference genome to quantify gene expression levels. The gene expression levels were subjected to t-distributed stochastic neighbour embedding (tSNE) clustering using the R package Rtsne (v0.16) [17], and the umap (v0.2.10.0) package was utilized for uniform manifold approximation and projection (UMAP) clustering of gene expression [18].

We utilized the median transcripts per million (TPM) values of genes in each tissue (calculated with R function media) for calculation of the tissue-specific index (TSI). Genes with a median expression level less than 0.1 TPM were assigned a value of 0. TPM values were then log2 transformed (TPM + 1). Subsequently, all genes or transcripts with accumulated expression values less than 0.1 were excluded from the analysis to filter out genes with low expression or transcripts with low expression [19].

The formula for calculating the TSI is as follows: T S I = ∑ i = 1 n 1 − x i max 1 ≤ i ≤ n x i n − 1 where n is the number of tissues, x is the median gene expression in the sample group, and i is the sample group. This study identified tissue-specific genes for which the TSI was greater than 0.85. The functional annotation of protein sequences was performed using eggNOG-mapper software [20]. Enrichment analysis of tissue-specific genes was conducted using the R package clusterProfiler (v4.6.2) [21]. The enrichment results of tissue-specific genes in sepal tissue were visualized using CirGO (v1.0) [22]. Transcription Factor (TF) identification of tobacco was carried out using the Plant Transcription Factor Database (PlantTFDB) database [23] (https://planttfdb.gao-lab.org/).

From 603 transcriptome datasets, 29 stress groups including roots, leaves, axillary shoots, and other tissues, were selected. We used featureCounts (v2.0.1) [24] to calculate the read counts of genes in the samples. Gene differential expression analysis was conducted using the R package DESeq2 (v1.38.3) [25]. Genes with a |log2FoldChange| > 1 and an adjusted p value < 0.05 were identified as differentially expressed genes (DEGs). Genes with accumulated TPM expression values less than 0.1 in all samples were filtered out, and the remaining genes were considered to have stable expression. The R package corrplot (v0.92) was used to create the correlograms with log2FoldChange in each group [26]. Functional enrichment analysis of the DEGs was conducted using the R package clusterProfiler [27].

To construct the regulatory network of TFs and genes, we first downloaded the plant TF protein files from PlantTFDB (https://planttfdb.gao-lab.org/). Next, we used BLASTP to align the tobacco proteins with the plant TF protein sequences, using an E-value cut-off of 1E−5. Genes that passed the threshold were identified as tobacco TFs. Subsequently, we calculated the Pearson correlation coefficients (PCCs) between the tobacco TFs and the DEGs shared across all the stress treatments. TF and DEG pairs with a PCC greater than 0.9 and q-values less than 0.01 were considered coexpressed. We utilized HOMER (v4.11) (http://homer.ucsd.edu/homer/) [28] to identify common stable motifs in the upstream 2 kb of DEGs shared across all the stress treatments. First, we used the loadGenome.pl subroutine of HOMER to load the tobacco genome file; next, we used the findMotifsGenome.pl subroutine to identify motifs in the 2 kb upstream promoter regions of the DEGs. As a background, we randomly selected the 2 kb upstream regions of 5000 genes across the entire tobacco genome, repeated the sampling 50 times, and selected motifs that consistently appeared in each sampling as the final identified motifs. We constructed a coexpression network of TFs that could bind to these motifs and were coexpressed with the DEGs shared across all the stress treatments with the coexpression PCC threshold set to 0.8. The coexpression network was visualized using Cytoscape (v3.10.2) software [29].

RNA-seq alignment results were sorted using SAMtools (v1.8) software with default parameters [30], and duplicated reads within the BAM files were marked using Picard (https://github.com/broadinstitute/picard). The GATK tool SplitNCigarReads was used to filter the BAM files obtained in the previous step [31]. Subsequently, SNP calling and variant filtration were performed using the HaplotypeCaller tool from the GATK toolkit, filtering candidate SNPs with Fisher strand (FS) values < 30.0 and quality depth (QD) values > 2.0. Further filtering was conducted to retain only biallelic SNPs. VCFtools (0.1.16) software [32] was then utilized with the parameters—max-missing 0.9 and—maf 0.05 to filter SNPs in the population with missing rates exceeding 10% and minor allele frequencies greater than 5%, respectively.

Gene expression data were filtered first, and genes with a median TPM greater than 0 were selected for eQTL analysis. The filtered expression data were subjected to confounding factor identification using the PEER (v1.0) package, and the top 10 confounding factors for covariance in eQTL analysis were selected. FastQTL (v2.0) [33] was used to identify cis-eQTLs in tobacco leaves. The size of the cis window was set to 1000000, and the ‘normal’ parameter was used for the transformation of phenotypes. EMMAX [34] was used to identify SNPs associated with the expression of each gene. The pvalues obtained from FastQTL were adjusted using the Bonferroni method to derive adjusted pvalues (p.adjust), with a significance threshold set at p.adjust < 0.05 in cis-eQTL analysis. Genetic Type I error calculator (GEC) software [32] was utilized to determine the effective number of SNPs (Me), and a threshold of 1.43 × 10⁻⁷ (0.05/Me) was adopted as the threshold for determining the significance of trans-eQTLs, adjusted by the Bonferroni method. PLINK (1.90b6.21) software [35] was used with the ‘clump’ parameter to merge linkage blocks of related SNPs, with a maximum block length of 500 kb. The SNP with the smallest pvalue in each linkage block containing at least three significant SNPs was selected as the lead SNP. Enrichment analysis was performed for cis- and trans-eGenes, and visualization was carried out using CirGO software.

Tobacco is an important economic crop, and transcriptomic studies of this species have generated a large amount of sequencing data. In this study, we collected 603 available tobacco transcriptome datasets from 13 tissues or organs, including leaves, roots, seedlings, stems, trichomes, petals, and sepals, from the NCBI database and published research (Fig. 1A). Notably, the samples from leaf and root tissues were the most numerous, with 329 and 146 samples, respectively. These RNA-seq samples were derived from 35 distinct NCBI BioProjects, which is indicative of their diverse origins. This aggregation reflects the inclusion of samples from various laboratories potentially harbouring rich genetic polymorphisms, thereby laying a foundation for the exploration of regulatory loci influencing tobacco gene expression. Among these studies, 10 Bioprojects were related to stress studies in tobacco, providing a foundation for cross-stress analysis of tobacco stress response mechanisms.

After RNA-seq data were downloaded from the NCBI SRA database, a unified analysis pipeline was used to perform data cleaning, alignment, and quantification on all datasets. After quantifying the expression levels of all genes in the samples (Supplementary File https://doi.org/10.6084/m9.figshare.29666060), clustering was performed using tSNE. The results revealed that some samples clustered together based on tissue relationships (Fig. 1B). However, the clustering analysis could not completely differentiate certain tissues, possibly due to anatomical proximity. For example, petals and sepals are parts of flowers, whereas the blade, lamina, and midrib are constituents of leaves. However, flowers are considered a morphological modification of leaves, leading to an inability for complete discrimination. Notably, there was a clear separation between leaves from the seedling and nonseedling stages in the clustering results, indicating differences in gene expression patterns between tobacco seedling leaves and mature leaves and validating the feasibility of categorizing them into two distinct tissues. From the perspective of treatments, samples under different treatments showed a clear clustering trend, indicating that both treatment conditions and growth environments can influence the clustering results (Supplementary Fig. S1). Similar patterns have been observed in transcriptome clustering analyses of poplar and soybean, suggesting that tissue type and environmental factors are major determinants of clustering [36,37].

The above results indicate that tSNE analysis can effectively distinguish most tissues, suggesting relatively high expression similarity among samples within the same tissue. This finding implies that identifying genes with tissue-specific expression can further elucidate the relationship between gene expression and tissue function. The samples collected in this study covered a total of 13 tissues, and the TSI of each gene was calculated as the median expression level across tissues (Supplementary Table S2). Additionally, genes were categorized into four groups (null, weak, broad, and tissue-specific) based on their median expression across tissues (Supplementary Table S3) [7]. The results revealed that 28.9% of the genes were classified as null genes (TPM <1 in all tissues; n = 10,272), 9.9% were classified as weakly expressed genes (TPM <5 in all tissues; n = 3529), over 57% were classified as broadly expressed genes (TPM ≥ 5 in all tissues, TSI < 0.85; n = 20,313), and only a small fraction were classified as tissue-specific genes (TPM ≥ 5 in all tissues, TSI ≥ 0.85; n = 1405) (Supplementary Table S4) (Fig. 2A).

The number of tissue-specific genes was the greatest in the immature flower, petal, sepal, and root tissues, with 308, 642, 63, and 303 genes, respectively (Fig. 2B). Although most tissue-specific genes are unique to individual tissues, there are also a few genes that exhibit specific expression in multiple tissues. Notably, a subset of genes exhibited very high tissue-specific expression in both tobacco flowers and petals (Fig. 2C). The greater number of tissue-specific genes and high expression levels in tissues or organs related to reproduction may be associated with their unique functions. To further understand the functions of these tissue-specific genes, this study conducted gene ontology (GO) enrichment analysis using all genes as background. The genes that were specifically expressed in nine tissues—axillary shoot, blade, young leaf, midrib, petal, sepal, immature flower, root, and stem—were collectively enriched in 341 GO terms (Supplementary Table S5, Supplementary Fig. S2). For example, genes specifically expressed in the axillary shoot were enriched in genes related to anatomical structure formation involved in morphogenesis, plant organ formation, regulation of development, heterochrony, anatomical structure formation involved in morphogenesis, regulation of plant organ formation, monooxygenase activity, and other genes related to plant organ formation. Genes specifically expressed in petals and flowers were enriched in a series of GO terms related to floral organ development such as floral whorl development, floral organ development, androecium development, gynoecium development, and carpel development. Particularly in sepals, a large number of tissue-specific genes related to terpenoid biosynthesis processes, sesquiterpenoid biosynthesis processes, terpene synthase activity, and other terpenoid synthesis-related processes were identified (Fig. 2D). The top four enriched terms of tissue-specific genes in sepals were sesquiterpene biosynthetic process, sesquiterpene metabolic process, terpene biosynthetic process, and terpene metabolic process, and other biological processes were also predominantly associated with terpenoid synthesis. Several studies have revealed the involvement of terpenoids in processes such as attracting insects for pollination [38,39], indicating that the secondary metabolites synthesized in tobacco sepals also play similar important roles.

Given the crucial regulatory roles of TFs, this study aimed to identify the TFs contained within the abovementioned tissue-specific genes. Among the genes showing tissue-specific expression in petals, flowers, and roots, there was a relatively greater abundance of TFs such as MYB and MYB_related (Fig. 2E). Furthermore, ERFs were exclusively found in reproductive-related tissues such as petals, sepals, and flowers. These TFs most likely play important regulatory roles in growth, development, and secondary metabolite synthesis in their corresponding tissues.

As sessile organisms, plants face numerous biotic and abiotic stresses, so they have evolved a wide range of mechanisms to respond to and minimize these stresses. The transcriptome data collected in this study, including data from 29 biotic and abiotic stresses in roots, leaves, and axillary shoots, were manually curated (Supplementary Table S6). This provides an opportunity to identify various stress-responsive genes and outline relationships between these genes that single-stress studies cannot reveal. The stress studies conducted on leaves primarily involved investigations into resistance mechanisms against drought stress induced by salicylic acid (SA), melatonin (MEL), and abscisic acid (ABA) as well as resistance to cold and salt stress. The stress studies conducted on root tissues mainly involved investigations into resistance to Phytophthora nicotianae and high and low concentrations of nitrate (N), whereas axillary shoot studies have primarily focused on stresses associated with topping treatments.

Transcriptomic analysis comparing various stress treatment groups with the control group revealed significant differences in the number of genes responsive to different external stresses in tobacco (Supplementary Table S7). The quantity of DEGs shows a gradual increase in response gene numbers with increasing duration or intensity of the same type of stress. In particular, when tobacco leaves were subjected to 4°C environmental treatment for 0.5, 1, 2, 4, or 8 h, the number of DEGs gradually increased (Fig. 3A). In this study, the number of upregulated and downregulated DEGs between the stress treatment groups and control group samples generally ranged from 1 to 7-fold. Taking Phytophthora nicotianae infection as an example, as the stress increased, the log2FoldChange also increased accordingly (Fig. 3B). GO enrichment analysis was performed separately for genes under various stresses (Supplementary Figs. S3 and S4, Supplementary Table S8), revealing that many upregulated genes under adverse stress conditions were enriched in stress-related GO terms such as those related to drought response and auxin response. For example, under drought stress, the differentially expressed genes were enriched in the response to water category, including a histone deacetylase (Nitab4.5_0001922g0030), a gene widely known to participate in drought tolerance across plant species [40]. Under biotic stress treatments, the differentially expressed genes were enriched in GO terms such as response to bacterium, defense response to bacterium, and response to karrikin. Notably, these categories contained a shared gene, NB-LRR (Nitab4.5_0000068g0290), which encodes a key immune protein broadly involved in plant defense [41].

Genes with TPM expression levels less than 0.1 in all samples were filtered out, and the similarity of gene response patterns between different treatments was assessed based on the log2FoldChange. In the roots, regardless of the N concentration, there was a strong positive correlation between the different treatments based on the log2FoldChange (Fig. 3C). Based on the number of DEGs and their log2FoldChange, although there were differences in the number of DEGs under varying concentrations and durations of nitrate treatment, the direction of the gene response (upregulation or downregulation) and the log2FoldChange were similar (Fig. 3A,B). Moreover, in tobacco roots infected by Phytophthora nicotianae, the direction and magnitude of gene responses were strongly correlated with those after nitrate treatment. This suggests that although nitrate is a nonbiological stress and Phytophthora nicotianae infection is a biotic stress, they may stimulate some common response mechanisms. However, in the case of the root treatments under the shoot topping treatment, the direction and log2FoldChange of stress-responsive genes were negatively correlated with those under the N and Phytophthora nicotianae treatments, reflecting differences in gene response patterns between the different stresses.

The response to shoot topping treatment in roots was not only negatively correlated with the response of genes under other stresses in roots but also negatively correlated with the response of genes under various stresses in tobacco leaves, such as those induced by cold stress. Additionally, in leaves, there was a correlation between samples subjected to drought stress alone and those treated with MEL and SA, but the correlation between gene expression under drought stress after ABA treatment and drought stress alone was weaker, possibly because ABA can enhance the drought resistance of plants. In summary, understanding the response mechanisms of genes to different stresses plays a crucial role in enhancing the stress tolerance of tobacco plants.

TFs play pivotal roles in the regulation processes of tobacco. In this study, we analysed tobacco gene expression under various stress conditions. Initially, 1865 TFs were identified in the tobacco genome using the PlantTFDBdatabase. The number of differentially expressed TFs varied with the type of stress treatment. There were more differentially expressed TFs in the roots under N treatment and topping treatment, and in the leaves under drought treatment and 4-h and 8-h freezing treatments (Fig. 4A,B). The distribution of differentially expressed TFs was similar to the overall distribution trends of DEGs, with an increase in the number of differentially expressed TFs with increasing stress (Fig. 3A and Fig. 4A). Notably, in the drought stress treatment of leaves, the application of hormones resulted in a decrease in the number of differentially expressed TFs compared with those in the drought treatment without hormones, suggesting that these TFs with altered expression levels may be involved in the hormone response. Additionally, both the roots and leaves exhibited abundant common responsive genes among the different stress treatments (Supplementary Figs. S5 and S6).

Based on the aforementioned research findings, our study focused on genes that exhibit shared responses to stress in both roots and leaves. In leaves, the ABA_D1 and S treatments resulted in fewer DEGs (376 and 493, respectively), indicating that these stresses were not sufficiently intense to fully activate defence mechanisms. Furthermore, MEL is not a stress. Similarly, the T1 and T2 treatments resulted in a low number of DEGs in the roots, and these treatments resulted in root stress. Therefore, data from these conditions were excluded. There were 338 and 24 DEGs shared among the root and leaf tissues under the various stress treatments. To understand the regulatory mechanisms of genes responding to multiple stresses, this study calculated the PCC of TFs and the DEGs shared by roots and leaves under different treatments. Gene pairs with a PCC greater than or equal to 0.9 were considered coexpressed; i.e., the TFs within the gene pairs may regulate the DEGs (Supplementary Tables S9 and S10). In the roots, we identified 285 TFs regulating 132 DEGs, including more than 40 different families of TFs such as WRKY, MYB, and bHLH (Fig. 4C). This suggests the presence of a complex stress resistance regulatory network in tobacco roots. Numerous response genes are shared between biotic and abiotic stressors, and these genes are also commonly regulated by various TFs. In the leaves, 25 TFs, including WRKY, MYB, and ERF, regulated the expression of 4 stress-responsive genes (Fig. 4D). These genes encode proteins such as F-box PP2-A13-like and G-type lectin S-receptor-like serine threonine-protein kinase, which are considered important stress resistance candidate genes in plant leaves, according to many studies [42,43]. Additionally, we predicted common stable motifs in the 2 kb upstream region of the DEGs. Due to the stringent coexpression PCC threshold of 0.9, which resulted in a very small number of TFs that could bind to these common stable motifs in the DEGs, we relaxed the threshold to 0.8. Combined with the coexpression results, we identified 26 TFs in leaves and 21 in roots that can bind to these common stable motifs among the DEGs (Supplementary Figs. S7 and S8, Supplementary Table S11). This result provides valuable resources for analysing gene regulation in response to stress. In summary, these results provide an opportunity for a more comprehensive understanding of the functions of TFs in stress responses.

Tobacco leaves are key organs used in cigarette processing, and the availability of 379 leaf samples enabled a population-level analysis of the relationship between gene expression and genetic polymorphisms. From the RNA-seq data, a total of 598,293 high-quality SNPs were identified across the tobacco genome. Their chromosomal distribution revealed clear regional enrichment on chromosomes 2, 4, 12, 14, and 19 (Fig. 5A), suggesting that SNP density varies among chromosomes, likely reflecting differences in structural features and evolutionary histories.

To dissect how these genetic variants influence transcriptional regulation, we performed eQTL mapping using leaf tissues and identified 28,396 eQTLs (Supplementary Table S12). Based on their physical proximity to associated genes, these loci were classified as cis- or trans-eQTLs. The cis-eQTLs displayed a strong diagonal pattern, whereas trans-eQTLs were widely dispersed and formed several distinct regulatory bands (Fig. 5B), highlighting clear differences in regulatory architecture. In total, 14,559 genes (41% of all expressed genes) were regulated by at least one eQTL, indicating that genetic variation extensively shapes gene expression in tobacco leaves. Among these eGenes, 53% were regulated exclusively by cis-eQTLs, 21% only by trans-eQTLs, and 26% by both (Fig. 5C). On average, each gene was associated with 4.6 eQTLs, and only 5771 genes were influenced by a single locus (Fig. 5D), underscoring the complexity of regulatory interactions involving both local and distal genetic effects.

To further characterize the biological roles of these eGenes, we conducted functional enrichment analyses. Trans-regulated eGenes showed strong overrepresentation in biological process categories, with limited enrichment in molecular function and cellular component terms (Supplementary Fig. S9). Many trans-eGenes were involved in photosynthesis-related pathways, including photosynthesis, light reaction, photosynthesis, light harvesting, tetrapyrrole metabolic process, and chloroplast organization. Additionally, they were enriched in hormone signaling and stress-responsive processes such as response to fatty acid, response to jasmonic acid, response to herbicide, intracellular monoatomic cation homeostasis, and protein K63-linked ubiquitination (Supplementary Fig. S9). Several eGenes were also associated with nicotine biosynthesis (Supplementary Table S13), providing candidate genes relevant to tobacco trait improvement.

We further identified 1669 cis-eQTL hotspots and 1858 trans-eQTL hotspots across the genome (Fig. 5E). Notably, the trans-eQTL hotspots contained 68 transcription factors, including members of the WRKY, bHLH, and MYB families (Supplementary Table S14), suggesting that these regions may exert broad regulatory influence through transcription factor–mediated cascades. For example, within the trans-eQTL hotspot I47_C2, the locus eQTL (chr04:1227307) cis-regulated the transcription factor MYB (Nitab4.5_0001158g0050), a TF related to drought responses, while simultaneously trans-regulating a drought-induced protein (Nitab4.5_0001808g0020). This illustrates how single regulatory variants can coordinate hierarchical control of stress-responsive pathways. These findings provide a comprehensive view of the genetic regulation of gene expression in tobacco leaves.