2.1 Collection and maintenance of leaf tissue of black pepper genotype

Black pepper genotype (P. nigrum accession 4226) was cultivated at ICAR-Indian Institute of Spices Research, Calicut, India (Latitude: 11° 15′ 0.00″ N; Longitude: 75° 46’ 12.00″ E). This genotype was previously identified as drought tolerant (Krishnamurthy et al., 2016). The plants were maintained as rooted cuttings under greenhouse conditions with temperature above 30°C, relative humidity >75 and day length of 12 h. The plants were grown in pot (one plant per pot) using soil, cow dung, and sand in the ratio of 1:1:1 and watered once a day for 100 days. Abiotic water stress was induced by withdrawal of irrigation on the 15th day, before the 4–5 nodes stage, mimicking drought conditions (Vijayakumari & Puthur, 2014). The control plants were watered regularly.

2.2 RNA isolation and sequencing

The RNA was extracted by pooling leaf tissues from uniform-sized leaves randomly collected from 10 different plants just before wilting, both for control and drought-affected samples. To minimize the across-sample variability, pooling approach of 10 biological replicates was employed (Zou et al., 2016). Total RNA was isolated from the Liquid N₂ frozen leaves of drought-induced and control plants using Spectrum Plant Total RNA Kit (Sigma-Aldrich) and On–column DNAse I digestion set (Merck) was used for elimination of DNA from the total RNA preparations, following manufacturer's protocol. The RNA quality was checked by both gel electrophoresis (1% agarose gel) and Bioanalyzer (Agilent 2100, Average RIN 6.4). Paired-end sequencing from the drought-induced and control RNA samples was done using Illumina HiSeq 2000 platform and utilising TruSeq RNA Sample Prep Kit v2 for the library preparation to get a read length of 101 bp. The paired reads fastq files from all the samples used in transcriptome analysis are available in the Sequence Retrieval Archive (SRA) at National Center for Biotechnology Information (NCBI) with BioProject: PRJNA515366 and BioSamples: (SAMN10754251, SAMN10754252).

2.3 Pre-processing and de novo assembly

The obtained 17.5 GB was pre-processed before assembly. The visualization for accessing the raw reads was done using FASTQC. Trimmomatic (v0.36) was used for the removal of low-quality reads, adaptors, and overrepresented sequences (Bolger et al., 2014). Reads with phred-score ≥ 30 and read length ≥ 36 bp were taken for downstream analysis. Trinity (Haas et al., 2013) and SOAPdenovo-Trans (Xie et al., 2014) assembler was used for de novo transcriptome assembly of control and drought samples of P. nigrum. Based on the N50 value, trinity assembly was used for further analysis. CAP3 assembler was employed to remove redundant reads from the trinity assembly (Huang & Madan, 1999).

2.4 Abundance estimation and identification of differentially expressed genes

The contigs generated by de novo assembly of P. nigrum transcriptome were considered as reference on which the trimmed reads of control and drought-affected samples were aligned using Bowtie tool (Langmead et al., 2009). The calculation of expression values in the form of fragments per kilo base of exon per million mapped reads (FPKM) was performed using RNA-Seq by Expectation–Maximization (RSEM) tool (Li & Dewey, 2011). For the identification of differential expressed genes, EdgeR (Empirical analysis of Digital Gene Expression in R; Robinson et al., 2010) of the R Bioconductor and NOIseq (Tarazona et al., 2011) tools were used to reduce the noise and enhanced the computational accuracy. EdgeR package was employed for the identification of the differentially expressed genes (DEGs) in drought-stressed samples while comparing it with control samples using stringent parameters of FDR 0.01 and log2 fold change (FC) value as five. A total of 4914 differentially expressed genes were found, out of which 2862 and 2052 were upregulated and downregulated, respectively. The results of edgeR tool were further validated by NOIseq tool, which is comparatively better as it can handle non-replicate data with its data-adaptive and non-parametric approaches. For the identification of significant DEGs, the threshold P-value and variance were set to 0.05 (Jaiswal et al., 2018). To check the reliability of results, a comparative analysis of the EdgeR results with q = 0.9 and q = 0.95 was performed.

Our de novo assembly of black pepper leaf transcriptome was mapped on the reference genome of Hu et al., 2019 (available at NCBI: BioProject accession PRJNA529758, PRJNA529760 and genome assembly at http://cotton.hzau.edu.cn/EN/download.php). This was compared at three different thresholds of DEGs (75, 90, 95 of % similarity) to have an idea of accuracy and uniformity of our transcriptome assembly.

2.5 Homology search, annotation, and functional characterization

The homology search of DEGs from P. nigrum L. transcriptome assembly was performed with NCBI non-redundant database (ftp://ftp.ncbi.nlm.nih.gov/blast/db/) as reference using Blastx algorithm. The standalone local ncbi-blast-2.2.31+ with threshold E-value 1e-3 (Altschul et al., 1990) was considered for the same. For further analyzes, Blast2GO Pro tool (https://www.blast2go.com/; Conesa et al., 2005) was used for mapping and annotation. The functional characterization of the identified candidate genes along with the involved pathways and the interproscan of DEGs were identified. The functional classification was done using the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases for a broader overview of the crop species and the pathways involved. These genes were functionally categorized into three sub-categories: biological processes, cellular components, and molecular functions. Further, Blastx tool was performed against PlantTFDB 4.0 (http://planttfdb.cbi.pku.edu.cn/download.php; Jin et al., 2016) for finding out transcriptional factors in the differentially expressed genes. The KEGG pathways were also identified using Blast2GO tool.

2.6 Putative molecular markers identification

Mining putative SSR markers was done from de novo transcriptome assembly of P. nigrum L. using perl script of MISA (Micro SAtellite identification tool; http://pgrc.ipk-gatersleben.de/misa/; Thiel et al., 2003). For identification of SSRs, default parameters were considered, i.e., 10 repeating units in case of mononucleotides, six repeating units in case of dinucleotides, and five repeating units in case of trinucleotides, tetranucleotides, pentanucleotides, and hexanucleotides. The maximum difference between two SSRs markers was kept as 100 bp. Primers were generated for the desired marker using PRIMER3 tool (http://sourceforge.net/projects/primer3/; Untergasser et al., 2012).

For variant detection, alignment and mapping were done, where the reads were mapped over P. nigrum L de novo transcriptome assembly. This was performed using Burrows-Wheeler Aligner (BWA) tool (http://bio-bwa.sourceforge.net/; Li & Durbin, 2009). For findins SNPs (Single Nucleotide Polymorphism) and INDELs (INnsertion and DELetion), the SAM tools package was used (http://samtools.sourceforge.net/; Li, 2011). Filtering of SNPs was done using SNP-EFF tool. The critera appliers were: read depth ≥ 4, quality score ≥ 20, and flanking regions of 50 on both the sides.

2.7 Identification of long non-coding RNAs and their annotation

Long non-coding RNAs (lncRNAs), usually >200 nucleotides, are involved in various developmental process and stress response. For the identification of lncRNAs in the present study, assembled contigs were filtered based on transcript length (> 200 nt). The transcripts with open reading frame (ORF) length longer than 100 amino acids were discarded using ORF Predictor (Min et al., 2005) (https://bioinformatics.ysu.edu/tools/OrfPredictor.html). The remaining transcripts were filtered based on the coding potential using CPC (Kong et al., 2007; http://cpc.gao-lab.org/) and PLEK (Li et al., 2014). Transcripts with coding-potential score ≥ 0.5 were discarded; PLEK is based on k-mer and SVM-based approach identifying lncRNAs from mRNAs sequences. The remaining transcripts were searched with the help of BLASTx against the NCBI-nr (E-value 1e-03), Swissprot database (with e-value: 1e-4, coverage: 35%, alignment length: 40 aa, % sequence identity: 35%), and HMMER search against the Pfam database (Finn et al., 2016) with default parameters. This was done to filter the transcripts that matched to protein coding regions or to protein family domains. To remove the possibility of getting housekeeping genes (including tRNA, rRNA, snRNAs, and snoRNAs etc.), the transcripts were also cross-checked against the Rfam database (http://rfam.xfam.org/: using INFERNAL) and rRNA database (https://www.arb-silva.de/; Kang & Liu, 2015, Kumar et al., 2019).

2.8 Development of transcriptome-based web resource of black pepper

The Black Pepper Drought Transcriptome Database (BPDRTDb) is based on “three-tier architecture,” also called as client–server architecture, where functional and logical processes, data access, user interface, and computer data storage were developed as independent modules and maintained on separate platforms. The client–server architecture comprises of client tier or presentation tier, middle tier or logical and database tier or physical tier (Figure 1). The top most level (i.e., client tier) is the user interface for which web pages were developed using HTML (Hypertext markup language) and Javascript for defining the queries and browsing. It translates tasks resulting to something that user can recognize. In the middle tier or logical tier, PHP (Hypertext Pre-processor) language has been used for writing codes for the server to process, define queries, fetch data, and create and maintain database connectivity. The last tier (i.e., database tier) was developed using MySQL, the information is stored and retrieved from the database for storing various data in the form of tables such as contigs ID, sequence, length, assembly, blast result file, DEGs, expression values, markers like SSRs, SNPs, INDELs, transcription factors, pathways. All the tables are interlinked among themselves and to contigs ID. It also provides provision for users to blast transcript IDs against NCBI non-redundant database.

3.1 Pre-processing and de novo assembly

The reads' quality assessment was done using FASTQC tool based on several parameters, namely basic statistics, per sequence quality scores, per base sequence content, adapter content, per sequence GC content, sequence length distribution, per base N content, sequence duplication levels, overrepresented sequences, per tile sequence quality, and k-mer content. The higher the score, the better is the base call. The median score was found to be approximately 41, which indicates good quality of reads. Further, quality trimming was performed using Trimmomatic, where reads were truncated on the basis of their average quality.

A total of 16,396,607 and 19,883,219 paired-end reads in control and drought-stressed samples, respectively, were generated with reads length of 101 bp. After removal of 22,587 and 28,596 low-quality reads from the control and drought stress drought-stressed samples, respectively, the remaining reads were further used for downstream analysis. These cleaned and high-quality reads were further used for de novo transcriptome assembly of P. nigrum L. We got a total of 149,678 transcripts generated by Trinity with 42.75% GC content. The total Trinity genes were 90,064 with N50 value of 1409. The average contig length was 843.72 bp (Table 1). Further, the redundant sequences were removed using Cap3 assembly. A total of 114,598 transcripts were generated showing 42.49% GC content and N50 value as 1481 bp. The range of transcripts varied from 201 bp to 12,166 bp. The minimum length of transcripts was 201 bp, while the largest transcript of the length was 12,166 bp. The maximum numbers of transcripts (i.e. 31,719 transcripts) ranged between 200 to 299, followed by 16,632 and 9317 transcripts for the lengths ranging between 300 to 399 and 400 to 499, respectively. Also, the comparison of three different thresholds of reads (75, 90, 95% similarity) showed a variation < 5% when mapped to the reference genome (Table 2). This low variation is indicative of a high accuracy of the transcriptome assembly, uniformity of coverage, and read depth.

3.2 Abundance estimation and identification of differentially expressed genes

The paired-end reads of both control and drought-affected samples were mapped and aligned to the de novo transcriptome assembly of P. nigrum L. This was done to calculate the expression values in the form of Fragments Per Kilo base of exon per Million mapped reads (FPKM) values. These expression values were used for getting the differentially expressed genes (DEGs). A total of 4914 DEGs were found, out of which 2862 and 2052 were upregulated and downregulated, respectively. MA plot visualizes the differences between the data in two samples, i.e., control and drought affected. In MA plot, the horizontal x-axis represents A, which is the log2 transformed mean expression level and the vertical y-axis represents the transformed data onto M (log 2 transformed fold change). The horizontal axis in the Volcano plot is the fold change (FC) on log scale so that upregulated and downregulated genes appear symmetric. The vertical axis represents the mean expression value of −log10 (False discovery rate). The light black dots indicate the up and downregulated DEGs, while the pink dots represent non DEGs (Figure 2).

A total of 4914 DEGs were obtained from edgeR tool, out of which 2862 and 2052 were upregulated and downregulated, respectively. Whereas we found 7139 DEGs in q = 0.9 (4670 up and 2469 down [± 7 fold change value]) and 4990 DEGs in q = 0.95 (3130 up and 1860 downregulated [± 8 fold change value]). Out of the total 4914 DEGs from edgeR, 4496 and 4473 DEGs were commonly found with NOIseq at q = 0.9 and q = 0.95, respectively, as evident in the Venn diagram in Figure 3.

Also, the comparison of three different thresholds of DEGs (75, 90, 95% similarity) mapped to the reference genome showed < 5% variation, indicating the accuracy of transcriptome assembly, uniformity of coverage and read depth (Table 3).

3.3 Homology search, annotation, and functional characterization

The homology search of DEGs was performed using BlastX to identify similar genes present in the database. Out of 4914 transcripts, we found that 4161 transcripts showed similarity with other genes present in the database, while 529 transcripts were novel as they did not show any similarity or were without hits. A total of 46 transcripts were used in mapping and 148 transcripts for the GO annotation process. The top hit species distribution revealed that maximum hits were found with Nelumbo nucifera (i.e. 766 transcripts), followed by Macleaya cordata and Elaeis guineensis in 489 and 225 transcripts, respectively (Figure 4).

Blast2GO annotation results revealed that these differentially expressed genes are into three sub-categories: biological process, molecular functions, and cellular components. In biological process, molecular function and cellular component, a total of 1064, 231, and 613 transcripts were found, respectively. Under biological process, there were 141 transcripts involved in cellular process, 107 in metabolic process, 103 in development process, 75 in biological regulation, 66 in signalling, and 22 in growth. In molecular functions, 70 transcripts were involved in catalytic activity, 85 in binding, 5 in structural molecule activity, and 6 in molecular function regulation. In cellular component, 132 transcripts were involved in cell, 129 in cell part, 106 in organelle, 62 in membrane, and 29 in extracellular part (Figure 5).

To study the pathways involved in stress mechanism, KEGG pathway analysis was done. A total of 4914 transcripts were involved in 187 pathways. The highest number of transcripts (i.e. 25) was involved in purine metabolism. This was followed by 16 transcripts in thiamine metabolism and 10 in biosynthesis of antibiotics (Figure 6).

The interproscan was used to identify domains and families in DEGs. A total of 786 domains in 2298 transcripts were observed. The highest number of transcripts belonged to the protein kinase domain (IPR000719; 107 transcripts) followed by the WD40-repeat-containing domain (IPR017986; 38 transcripts) (Figure 7).

In InterProScan family search, a total of 3999 transcripts were found to be involved in 1137 families. Many transcripts were present in the family of P-loop containing nucleoside triphosphate hydrolase ((IPR027417; 186 transcripts) followed by the protein kinase-like domain superfamily (IPR011009; 122 transcripts) (Figure 8).

The transcription factors (TF) were identified by performing blast in the PlantTFDB 4.0 using BlastX tool. A total of 2110 transcription factors among the 4914 transcripts with an expected e-value 1e-3 were observed. The transcripts corresponding to bHLH, MYB, and NAC transcription factors occurred a maximum number of times, i.e., 228, 192 and 168, respectively (Figure 9).

3.4 Identification of putative molecular markers

3.5 Identification of lncRNAs and their annotation

A total of 114,598 assembled transcripts had their lengths greater than 200 bp and they were subsequently used for downstream analysis. We retained 70,608 transcripts (~61.61%) with ORF length less than 100 amino acids as predicted by ORF Predictor. After filtering mRNAs using PLEK, a total of 70,364 sequences were found to have a coding-potential score less than 0.5 using CPC (Coding Potential Calculator). A total of 46,702 transcripts were removed on the account of similarity with known proteins or protein domains when BLAST against NCBI-nr, Swissprot, Pfam and housekeeping genes (tRNAs, snRNAs, and snoRNAs). Finally, a total of 23,662 transcripts were found as high confidence lncRNAs in the black pepper transcriptome. It was observed that the maximum number of sequence (17,006 corresponding to 71.87%) had a length ranging between 200 and 400 bp, followed by 4177 (17.65%) of the range 400–600, and only 0.74% of sequence with a size above 1200 bp (Table 5).

3.6 Black Pepper Transcriptome Database: BPDRTDb

The Black Pepper Drought Transcriptome Database based on three-tier architecture contains the information of candidate genes, putative genic region markers (SSRs, SNPs, Indels), transcriptional factors, pathways, domains, families, etc. It houses 114,598 transcripts, 4914 differential expressed genes, 20,124 putative markers, 14,742 primers, and 259,236 variants, which contains 246,458 SNPs and 12,778 Indels identified from de novo assembly. A total of 2110 transcriptional factors, 786 domains and 1137 families are also catalogued in this resource. It contains five tabs: Home, Transcripts, Markers, Candidate genes, and Supplements (Figure 10).

The first page, namely “Home page,” shows general information regarding the crop (P. nigrum L.) data and database. The “Transcripts” tab is further categorized into Expression profile, Transcription factor families, Domain and family, and Pathways. The expression values are expressed in the form of FPKM values of control and drought-affected samples with BLAST results under the Expression profile tab. The transcriptional factors identified in DEGs have hyperlinks to PlantTFDB, which was used as a reference to BLAST DEGs under the Transcription Factor families tab. Under the tab, Domain and family search, one can access the domain and families and hyperlinks are provided for direct link to Interpro database website. The “Pathways” tab has the details of pathways identified from DEGs directly linked to KEGG database.

The “Markers” tab has detailed information on the molecular markers identified from the black pepper transcriptome analyzed in this study. This tab contains three sub-tabs, namely SSRs, SNPs, and Indels. SSRs tab provides the information of repeats detailing the three sets of primers, while the SNPs and Indels tabs provide the variants, insertion and deletion with 50 bp flanking regions on both sides. For SSRs, three sets of primers will be generated for each particular marker, searchable using a transcript ID. The result displays the start and end positions of primers and primer length.

The “candidate genes” tab results into transcript IDs, log Fold Change value, log Counts Per Million reads value, log P-value, False Discovery Rate value. It also allows user to blast transcript IDs against the NCBI non-redundant database.