RNA-Seq raw data were retrieved from NCBI (National Center for Biotechnology Information) SRA database, code ERA047687 [27]. We used the transcriptome data of developing male flowers (accession: ERX021379), leaf (accession: ERX021378), germinating seed (accession: ERX021377), developing endosperm V/VI (accession: ERX021376), and developing endosperm II/III (accession: ERX021375).

For reads mapping to the reference genome (R. communis, http://castorbean.jcvi.org/index.php) was used HISAT [28]. Annotation of the coding genes was made by comparison into the TIGR database [29]. After mapping of reads in the gene models of R. communis genome, were obtained seven files with the read counts per gene using Cufflinks version 0.9.3 [30]. Finally, differential expression ratio for each gene was calculated contrasting different stages (II to V) of endosperm development using Cuffdiff in [30]. Read counts from germinating seed, leaf and developing male flowers were pooled and taken as control.

We conducted a combined analysis of PPI networks (interactome) for R. communis, which was inferred by peer-to-peer orthology analysis with the proteins reported from version TAIR10. The reconstruction of the interaction network was completed by mapping orthologous proteins against the network reported for A. thaliana (ftp://ftp.arabidopsis.org/home/tair/Proteins/) using the software OrtoMCL [31]. The interaction network was enriched with the differential expression ratios obtained in Section 2.1, having fold changes in different developmental stages of the endosperm. Finally, nodes analysis corresponding to the metabolic pathways involved in fatty acid synthesis and accumulation was performed, displaying nodes (enzymes) that exhibited overexpression in the endosperm compared to the control.

For this research, we used cultivars of castor bean (R. communis) classified as experimental cultivar Ricinus Corpoica 03 (VERC03, Colombian collection) and experimental cultivar Ricinus Corpoica 12 (VERC12, a result of genetic improvement from Brazil). Considered as promassing cultivars after having been studied for more than a decade and that are reported as suitable to be grown in dry agoclimatic areas [10]. These cultivars were obtained from the germplasm collection at the La Selva AGROSAVIA Research Center (Rionegro, Colombia) through a Material Transfer Agreement 20162103285. The two cultivars were cultivated in an experimental field plot in the Bajo Cauca Antioqueño, Colombia (Latitude: 7˚56'29''N, Longitude: 75˚8'43''W), located at 150 meters above sea level, with 85% of relative humidity and 28˚ Celsius of average temperature during the day, the experimental area was uniformly fertilized according to [10] and irrigation at 10-day intervals. Plant material was collected for four different stages of seed development (Figure1 and FigureS1) under the collection framework permit granted by the National Environmental Licensing Authority (in Spanish, Autoridad Nacional de Licencias Ambientales or ANLA) under Resolution 1516 of 2014 (modified by Resolution 1312 of 2015). Fruits with different development stage, coloration and size were collected from VERC03 and VERC12 plants located in the furrows or edges of the growing fields (Figure1). All samples collected were stored at −80˚C in liquid nitrogen.

Total saturated, monounsaturated, and polyunsaturated fats in castor bean seeds during different development stages were determined using gas chromatography (GC).

Extraction Method: In total, 2 ± 0.0020 g of each sample was weighed in a glass capsule with an addition of 100 mg of butylated hydroxytoluene (BHT) and 2.0 mL of internal standard (5.00 mg/mL of triundecanoate triglyceride). This was subjected to acid hydrolysis for 1 h with HCl 4 N in SOXCAP 2047. Then, the samples were washed with distilled water to remove acid residue and then dried in a microwave oven for 45 min. The glass capsules with the dry samples were exposed to the SOXTEC 2050 equipment to extract the fat with 50 mL of hexane. The fat extract was incubated at room temperature in a desiccator and then methylated with boron trifluoride in methanol at 14%.

Analytical Equipment: The fatty acids were identified using an Agilent 7890B gas chromatograph system with a flame ionization detector (GC-FID), equipped with 7963A autosampler, 100:1 split injection port, and 100% biscianopropil polysiloxane TR-CN100 (60.0 m × 0.25 mm i.d. × 0.20 µm) capillary column (Teknokroma, Spain). Temperature program conditions were: the initial temperature 100˚C, ramp to 14,235˚C at 83˚C/min, then by a rate of 2˚C/min to 220˚C, the injector temperature was 250˚C. Helium was used as a drag gas at a flow rate of 1.0 mL/min.

Method of identification and quantification of compounds: Signals were recorded and manipulated using the software OpenLab CDS ChemStation v A.01.05. Fatty acid identification was performed by comparison with the retention times of the Restek brand Food Industry FAME Mix standard (37 components). The quantification of fatty acids was performed against an internal standard, triundecanoic triglyceride (C11:0), before the calculation of total fat. These were converted to the respective triglycerides and fatty acids; subsequently, the total fats were calculated as the sum of the individual fatty acids expressed as triglyceride equivalents and saturated, monounsaturated, and polyunsaturated fats.

According to Addendum #1 (June 16, 2017) to the Framework Contract for Access to Genetic Resources and Their Derivative Products (in Spanish, Contrato Marco de Acceso a Recursos Genéticos y Sus Productos Derivados) #127 of May 13, 2016, between the Ministry of the Environment and Sustainable Development and EAFIT University, the total RNA was extracted from different tissues of castor bean cultivars. Equal amounts of embryo and endosperm tissue were removed from seeds collected from fruits. RNA extraction was performed using the reagent Ribozol^TM (Amresco, USA), then purified with the commercial kit GeneJET Plant RNA Purification (Thermo Fisher Scientific, USA) according to the manufacturer’s instructions and quantified by spectrophotometry (260/280 nm) in Nanodrop ND-2000 UV-V is Spectrophotometer (Nanodrop Technologies, USA). RNA Integrity was analyzed by electrophoresis in a 1% agarose gelcDNA synthesis was performed using the Maxima First Strand cDNA Synthesis Kit for RT-qPCR with dsDNAsa (Thermo Fisher Scientific) and cDNA products obtained were stored at −80˚C until the evaluation process.

Primer design was made using the Primer 3 program following the parameters established by [32] seeking to amplify fragments of approximately 150 bp (Table 1) [33]. The specificity of the primers was verified using the Primer-BLAST program (NCBI).

cDNA synthesized was used to perform the PCR technique in real-time using the Maxima SYBR Green qPCR Master Mix (2×) kit (Thermo Scientific, USA), following the instructions of the manufacturer, in a CFX96 Touch Real-Time PCR (Bio-Rad, Foster City, CA). PCR conditions were denaturation at 95˚C every 20 s, hybridization at 60˚C every 30 s, and elongation at 72˚C every 30 s for 39 cycles. Melting curve was also determined for each PCR product. Difference significance in the gene expression was estimated using the relative expression for each gene in the three developing stages for the two cultivars, considering the differences (ΔΔ) in the crossing threshold (Ct) [34] and, assuming an efficiency of 2. All data were analyzed using the REST program, version 2009 (Relative Expression Software Tool, Gene Quantification, Munich, Germany) [35], at 5% (p < 0.05) level of significance. As housekeeping gene, the ubiquitin was used as a reference, as previously described for castor bean [27].

Based on the analysis of PPI networks and validation by qPCR, four genes

coding enzymes involved in the biosynthesis and accumulation of fatty acids in castor bean were selected. Primer designing for the amplification of these genes was made with Primer 3 software [34] using the default parameters (Table 2).

The pDrive vector (QIAGEN, USA) was used to clone the four genes of interest, following the parameters set by the manufacturers. Subsequently, as a verification mechanism, a plasmid DNA extraction was made with the QIAprep Spin Miniprep kit (QIAGEN Sample and Assay Technologies) and a polymerase chain reaction for amplification of cloned genes was performed. PCR products were sequenced in triplicate by Sanger method the sequences were manually edited using Geneious 5.3.6 created by Biomatters [36] and aligned with the algorithm Clustal W [37] in the same software.

Phylogenetic reconstruction of the desaturated/monooxygenase protein family was performed using the cloned sequence of the FAH12 gene from castor bean VERC03 to corroborate its identity. FAH12 and FAD2 gene sequences of R. communis, A. thaliana, Glycine max, and Jatropha curcas among other phylogenetically close plant species were retrieved from GenBank database (see the access numbers of the sequences in Figure 5).

Firstly, the FAH12 sequence of castor bean VERC03 was translated with Virtual Ribosome 1.1 [38] using the standard genetic code. It was added to the genes dataset in amino acids and then aligned using MAFFT v7.271 [39]. Secondly, nucleotide sequences were aligned with RevTrans2 [40]. Thirdly, these alignments were repartitioned using GBLOCKS 1 [34], and the suggested partitions were compared to the Pfam [41] annotations for the location of the active protein site. Fourthly, the best partitioning scheme was selected according to the evolution models using Partition Finder Protein 1.0.1 [42]. Finally, three topologies were reconstructed: 1) maximum likelihood with Garli (10 independent replicates supported with 1000 bootstrap pseudoreplicates), 2) maximum likelihood with RA × ML (10 independent replicates supported with 1000 bootstrap pseudoreplicates), and 3) Bayesian inference with MrBayes (2 MCMC analyses

with 10,000,000 generations with 4 chains each). All topologies were contrasted using nucleotides and aminoacid seeking to determine inconsistencies.

A combined analysis of R. communis PPI networks (interactome) was conducted using an A. thaliana network, as a reference, composed by 11,706 proteins of (TAIR), with 24,417 interactions between these proteins. Overall, it was identified 11,192 proteins orthologous to R. communis, with significant homology to A. thaliana proteins with 23,777 interactions between them. From the protein network obtained for the castor bean plant, FAH12 protein (oleate 12-hydroxylase, genome ID 28035.m000362, E.C. 1.14.18.4) was overexpressed 10.77 times in advanced seed development stages compared with the initial development stages and other tissues (Figure 2). Furthermore, it was evident that FAH12 also interacts with other proteins related to the production of fatty acids, establishing a possible role of this gene in the production of fatty acids in castor bean (Figure 2(b)). In addition to this gene, FAD2, DGAT2, and PDAT1-2, genes were identified, such as possible genes involved in two physiological processes: biosynthesis and fatty acid accumulation.

The analytes detected in the chromatogram of the samples were mainly classified into 3 groups of lipids: total saturated (SL), monounsaturated (ML), and polyunsaturated (PL). The quantity of lipids detected for each of the groups was dependent on the stage of development of the seeds, showing a better resolution in the initial stages of development, whereas, in the more advanced stages of development, peaks was hindered by the high content of carbohydrates, leading to a loss in the stability of the baseline at the end of the run (FigureS2).

The highest amounts of SL, ML, and PL (2.1226 g/100 g, 2.1342 g/100 g, and 2.6517 g/100 g, respectively) were detected and quantified in the fatty acids of the fruits at developmental stage II (S2) of the tropical VERC03 cultivar, revealing that the production and accumulation of triacylglycerols starts at an early stage during the formation of the fruit (Figure 3(a)). Likewise, a direct proportional relationship was determined between the quantities of fatty acids and the size of the fruit until developmental stage III (S3). Analyses of acid types 18:1-OH and 18:2-OH in the tropical VERC03 and VERC12 cultivars showed an increase in oleic and linoleic acids (Figure 3(b)).

The results of gene expression obtained with the RT-qPCR technique for the key biosynthetic FAH12, FAD2, and DGAT genes, in three different fruit developmental stages of the VERC03 and VERC12 cultivars, showed to FAH12 and FAD2 transcripts being up-regulated more than 3-folds and 1-fold respectively,

in the stages S2 and S3 for only VERC03 cultivar (Figure 4). Expression of these genes in cultivar VERC12 (improved commercial cultivar from Brazil-Embrapa) showed to be statistically invariable during the three developmental stages with

respect to the UBI housekeeping gene and S1 stage as a reference. DGAT transcript was invariable in all stages and cultivars (Figure 4).

To validate the identity of FAH12 and FAD2 genes, two fragments of 1166 bp and 1151 bp respectively, were obtained from amplification using cDNA from VERC03 cultivar, and then sequenced. The sequences obtained were aligned and comparative with sequences reported in the GenBank. Differences were found in the amino acids A242V and Q319H in FAH12 and F88V in FAD2 (FigureS3). The sequences were reported under the GenBank code MH990329 (FAH12) and MK033959 (FAD2).

A phylogenetic reconstruction was made for the desaturase/monooxygenase protein family using sequences cloned from FAH12 and FAD2 of the VERC03 castor bean cultivar and accessions retrieved from GenBank. Two groups were formed with a significant value of maximum likelihood, one with sequences of FAH12 genes and another derived from this, with FAD2 group in the top. We corroborated the identity of the clones sequenced and they were located in the same clade with each reference sequence for R. communis (Figure 5).