I. Plant materials, water deficit stress treatment.

The experimental design was completely randomized with three replicates, carried out under greenhouse conditions at Embrapa Mandioca e Fruticultura (12°40’47.8"S 39°05’20.7"W) comprising two contrasting cassava varieties (the bitter BRS Formosa and the sweet BRS Dourada), two treatments [non-stressed plants (negative control) and stressed plants (water deficit)] and plots with five plants per variety. BRS Dourada is a local variety with an unknown pedigree, specifically developed for fresh consumption. Its roots exhibit yellow color, and it contains less than 100 ppm of cyanogenic compounds, making it ideal for sweet cassava usages. In contrast, BRS Formosa is a hybrid of industrial cassava, known for its higher concentration of cyanogenic compounds (above 100 ppm), which imparts a bitter taste. The female parent of BRS Formosa is the BGM-0361 accession, while the male parent is unknown. These two varieties display significant genetic differences, as confirmed by molecular analysis through genotyping-by-sequencing techniques [17].

The average temperature inside the greenhouse was regulated by airflow and maintained at maximum temperature at 28°C ± 3°C and minimum at 23°C± 1, with average air relative humidity of 80%. To conduct the experiment, 30-liter plastic pots were utilized filled with a substrate blend consisting of 70% yellow latosol, 20% coconut fiber, and 10% vermiculite. Slow-release fertilization was implemented using 25 grams of specially formulated fertilizer. This fertilizer was encapsulated with elemental sulfur and coated with non-water-soluble organic polymers, containing 11% nitrogen (N), 22% phosphorus pentoxide (P₂O₅), 11% potassium oxide (K₂O), 12.29% sulfur (SO₄), 0.35% boron (B), 0.30% copper (Cu), and 0.30% zinc (Zn).

Throughout the preliminary phase leading up to the water deficit imposition, samples from all treatments were regularly irrigated in order to maintain soil moisture levels close to field capacity. In order to prevent water evaporation from the soil the pots were sealed with aluminum foil. The stress treatment consisted of suspension of irrigation of plants after 50 days after planting. Plants of the negative control maintained the irrigation during the assay, keeping the soil moisture content close to field capacity. The soil moisture content was monitored daily (at 6 am and 6 pm) using a TDR device (Time Domain Reflectometry), with water replacement of the control plants carried out up to field capacity (FC), based on TDR readings. The Fraction of Transpirable Soil Water (FTSW) was calculated by the methodology according to Sinclair and Ludlow (1986) [18]: FTSW = . Leaf gas exchange analysis was carried out in mature leaves – third leaf, between 9:00-10:00am, throughout the experiment using a portable photosynthesis meter ADC BioScientific (model LCpro-SD, Ltd., UK).

When plants presented wilted leaves and with 10% soil moisture (40% of FC), root samples were collected, frozen in liquid nitrogen, and stored in an ultrafreezer until total DNA extraction (CTAB methodology [19]). Total DNA quality and quantity was checked in 1% agarose gel electrophoresis (w/v) and spectrophotometer analysis (NanoVue Plus, GE Healthcare, Waukesha, WI, United States).

II. MethylRAD-Seq libraries.

After DNA quantification, the samples (two cassava varieties, two treatments and three replicates – composed by a pool of five plants per variety) were prepared for the construction of twelve MethylRAD-Seq libraries (CD Genomics, USA).

The quality of the raw reads considered the Phred values of the bases, where the filtration step (PEAR software v0.9.6) excluded reads showing more than 15% of bases with a Phred value < 30 or with N content greater than 8% [20]. Reliable methylation sites (CCGG and CCNGG sites) considered sequencing depths greater than 3. The analysis excluded reads without the expected restriction site and Enzyme Reads (those showing unexpected size) of each library. The predicted size tags containing the reference sequences CCGG (32 bp) and CCNGG (31 bp) were mapped (SOAP software v.2.21; parameters: -M 4-v 2 -r 0) in the reference genome of M. esculenta v6.1 (Phytozome, https://phytozome-next.jgi.doe.gov/; NCBI RefSeq assembly GCF_001659605.1) disregarding the repeatedly mapped reads.

III. Mapping of reads in the reference genome.

The pair-end reads were aligned in the reference genome of M. esculenta v6.1 (Phytozome), using BWA aligner v0.7.17-r1188 (BWA-MEM) [21] with the standard parameters. The Samtools v1.7 tool [22] was used to convert SAM files into BAM, draw, remove duplicates, and index data. Next, Picard Tools [23] was used to assign all readings a single new reading group and finally, GATK v4 [24] was used to perform the call for variants to obtain the VCFs files containing the information of the variants by genome.

I. Filtering of variants (SNPs/INDELs) and annotation.

The filtering of the variants was done by the bcftools tool [25] with the parameters QUAL>30 and MQ>30 to generate the filtered VCFs files. For the annotation of the variants regarding the region and the prediction of the effects (impact), the SNPeff [26] was used and the database was created for the reference genome used in the assemblies.

II. Distribution of methylation sites in structural and functional gene elements.

Identification of structural gene elements of M. esculenta, including introns, exons, UTRs, and regions (2,000 bp) upstream of the TSS (Transcription Start Site), considered the cassava Phytozome dataset (version 12.1; https://phytozome.jgi.doe.gov/pz/portal.html, ID: LTYI01000000). The methylation sites covering the UTRs were counted performing the SNPeff software [26]; and for the other genetic elements, the bedtools software (v2.25.0 https://bedtools.readthedocs.io/en /latest/), was used.

III. Relative quantification of DNA methylation and analysis of the distribution of methylation sites in gene regions.

The methylation levels (CCGG and CCNGG target sites) reflected the consistency of the amplification efficiency of isometric sequences and the sequencing depth of the methylated tags. Thus, the methylation level, expressed in RPMs (reads per million), considered the number of reads covering the site by the total of good quality reads in the library × 1,000,000. The RPM calculations also considered three biological replicates (R1, R2, and R3) for each variety and each treatment.

The analysis covering the distribution of methylation sites in genes comprised regions of CDS and 2 kb upstream and downstream of the TSS (Transcription Start Site), and 2 kb upstream and downstream of the TTS (Transcription Terminator Site). The methylation level of two samples (libraries) were analyzed by Pearson’s correlation coefficient. Scatterplots of sequencing depths of all methylation sites concerning the pairs of comparisons allowed to verify the consistency of the data of the biological replicates, since values close to 1.00 reflected the best correlations between the two sets of data [27].

IV. Differential methylation profiles.

The methylation levels of each cassava variety, when comparing the treatments with and without stress, allowed us to classify those differentially methylated sites as hypermethylated or hypomethylated by performing the edgeR software. The adjusted p-values considered the FDR (False Discovery Rate) technique applied to correct type I errors. The methylation levels comparing the two contrasts [FST (BRS Formosa Stressed Treatment) vs FNC (BRS Formosa Negative Control) and DST (BRS Dourada Stressed Treatment) vs DNC (BRS Dourada Negative Control)] considered for each treatment a pooled sample composed by the three biological replicates showing sequencing depths greater than three. Differently methylated sites were associated with previously annotated genes, whose expressions (Log₂ RPM values) were illustrated in heatmaps generated by the Tbtools v1.09 software. Venn diagrams (https://bioinfogp.cnb.csic.es/tools/venny/) contemplating the sets of differentially methylated sites identified those common or exclusive identifiers from the BRS Formosa or BRS Dourada drought-stress responses.

V. Analysis of gene ontology and prediction of interactomes.

The sets of genes showing differential methylation levels (hyper and hypomethylated for CCGG and CCNGG sites) in structural gene elements of BRS Formosa and BRS Dourada were characterized in terms of gene ontology (GO), applying a PlantRegMap tool (http://plantregmap.cbi.pku.edu.cn/). The analysis also identified those enriched terms (Fisher’s tests, p-value < 0.01), comparing the input sets with the global M. esculenta data available.

Proteins encoded by hyper- or hypomethylated genes of each variety helped us to predict potential protein-protein interaction (PPI) networks performing the STRING tool (v11.0b; https://string-db.org/), using A. thaliana orthologs as a reference model. The interactions showed a minimum score of 0.7 (high confidence) and maximum interactors of 20 (1^st shell) and 10 (2^nd shell). Also, only interactions derived from experimental data or co-expression/co-occurrence of results were reported. A functional enrichment analysis using the Aggregate Fold Change method [28] identified the main biological processes, molecular functions, and cellular components concerning the PPI members. Additional information involved hierarchical local clusters, metabolic pathways, and annotations derived from the STRING analysis.

I. The PPI networks based on BRS Formosa’s hypomethylated genes.

The PPI networks involving proteins encoded by hypomethylated genes of the bitter BRS Formosa after the stress, pointed a mannose-1-phosphate guanylyltransferase 1 (CYT1, Manes.10G025000) interacting with the ADP-glucose pyrophosphorylase protein, an enzyme acting in biosynthetic processes. The CYT1 protein plays a key role in plant cell wall development since A. thaliana loss-of-function mutant plants show irregular cell walls, affecting the cell turgor [36]. Turgor pressure is critical to cell wall growth and expansion, and physiological changes in plants under drought showed loss of turgor, with low turgor pressure affecting the adaptation of the plant’s hydraulic system, reducing plant growth [37]. Comparing with the reference M. esculenta genome (v.6.1), the CYT1 gene presented SNPs/INDELs polymorphisms (six of low impact and two of moderate effects). One of moderate impact (a missense alteration) was in heterozygous condition in BRS Formosa (non-polymorphic in BRS Dourada; S2 Table). This SNP/INDEL deserves to be explored if such condition provides a selective advantage for a drought-tolerant variety. The FASTA sequence is shown in S2 Table.

A potential interaction of cullin-like protein1 (Manes.02G033400) and auxin response factor 6 (ARF6; Manes.07G102300) highlighted the plant’s hormonal response. Plants under abiotic stress usually trigger hormonal signaling pathways. Abscisic acid (ABA) and ethylene, the most frequent stimulants of abiotic responses, also interact with other phytohormones [salicylic acid (SA), JA, and auxin], altering gene expression and adaptive stress responses [38]. Auxin is distributed in plant tissues helping plants to adapt to environmental conditions by controlling biological processes [39].

In addition, many developmental processes and physiological responses, including cell division, floral development, and hormonal signaling, are influenced by proteins after post-translational control, even protein turnover by ubiquitination [40]. Concerning the ubiquitination process, the substrate specificity derives from the E3 ubiquitin-ligase enzymes. The Cullin-RING Ligases (CRLs) are E3 Ligases binding to proteins to be ubiquitinated. The gene encoding the cullin-like protein1 (Manes.02G033400) presented six SNPs/INDELs of low impact (S2 Table). Furthermore, the F-box proteins confer the substrate specificity of the SCF complex, and the CUL1 association involves the SKP1 adapter protein. The SCF^TIR1/AFB complex acts as an auxin receptor, and binding with auxin increases the affinity of the complex for substrates, such as Aux/IAA proteins, increasing the ubiquitination and protein turnover. As a result, auxin response factors (ARFs), normally repressed by dimerization with Aux/IAAs, are released, acting as transcriptional regulators of auxin-responsive genes [41]. Besides, the signalosome COP9 (Constitutive Photomorphogenesis 9), known as CSN, cleaves RUB/NEDD8 from Cullins [42]. Thus, CSN acts as a promoter of E3 Ligase activity by protecting the substrate adapters of CRLs from autocatalytic degradation [43]. From the predicted PPI network, the components RBX1 (from the SCF ubiquitination complex for AUX response), CAND1 (Cullin-associated and neddylation-dissociated protein 1), and Cullin 1 (Manes.02G033400) have been reported for interacting with COP9 and the CSN complex [44]. Also, the role of CSN in the JA biosynthesis pathway has already been reported [45]. Another group directly related to the COP9 complex involved FUS12 (CSN2 - subunit 2 of the signalosome COP9 complex), a recognized repressor of plant development [46], and ARF6 protein (auxin response factor 6; Manes. 07G102300), also interacting with SHY2 (Short hypocotyl 2, an Aux-IAA protein).

Since environmental abiotic stresses also induce alternative transcripts [47], a potential mRNA splicing factor (Manes.01G115300) was linked with proteins including MEE (nuclear component ribonucleoprotein CLO-5), MAC (pre-mRNA processing factor 19 homologous 1), SKIP (protein-SNW/SKI interaction), PRP (protein pre-mRNA splicing processing), CDC (cell division cycle protein), and ESP (pre-mRNA-splicing factor RNA helicase DEAH1). All these proteins characterized by GO terms highlighted alternative splicing (AS) of RNAs. Polymorphisms covering the gene encoding that putative mRNA splicing factor (Manes.01G115300) comprised three SNPs/INDELs; one of moderate effect (missense variant) would be heterozygous in bitter BRS Formosa and homozygous in BRS Dourada (S2 Table).

Additional interactions involved the protein PAP15 (purple acid phosphatase 15, Manes.15G054300), already reported to enhance plant tolerance to osmotic and salinity stresses by modulation of ascorbic acid (AsA) levels, an antioxidant compound effective against the overproduced ROS [48]. The related gene presented six SNPs/INDELs of low impact and two of moderate effects (S2 Table). One missense variant of moderate effect was heterozygous in the BRS Formosa and non-polymorphic in BRS Dourada (S2 Table). Again, the heterozygous condition was evidenced in the bitter (BRS Formosa) variety, to the detriment of the homozygous or non-polymorphic status in the sweet BRS Dourada.

II. The PPI network based on BRS Formosa’s hypermethylated genes.

The hypermethylated genes of the bitter BRS Formosa potentially silenced three PPI networks, one of them linked several proteasome-related proteins. As proteins control cellular activities and physiological processes through the regulation of specific biochemical or metabolic pathways, in this regulation, protein degradation is as relevant as protein synthesis. Proteases and the ubiquitin-proteasome system (UPS) are responsible for the most protein degradation (80 - 90%), and the UPS system is an efficient and fast strategy to control cellular processes by selectively removing regulatory proteins [49]. Also, genes encoding proteases are induced in plants under abiotic stress, altering proteolytic activity levels of different specificities [50].

Other network members potentially silenced, but not expected to be, were transcription initiation factor TFIID subunits. These proteins are part of the RNA polymerase II pre-initiation complex. The promoter recognition by the basal transcription factor TFIID includes the TATA-binding protein (TBP) in a complex with TBP-associated factors (TAFs). The TAFs are transcriptional co-activators, assembling transcriptional complexes and recognizing promoters [51]. Some TFIID-related TFs were reported to be ethylene-responsive and relevant in drought-stress responses [52]. Since the related genes were hypermethylated, the bitter BRS Formosa is not adequately exploring this alternative.

The adequate importation of proteins to the nucleus is relevant to gene expression reprogramming, leading to plant drought tolerance. In this way, a network member was the KPNB1 protein (importin beta-1 subunit, ARM-repeat family, Manes.01G040400). The inactivation of the AtKPNB1 gene from A. thaliana increased stomatal closure (in response to ABA influence), reduced water loss, and increased water deficit tolerance [53]; such results reinforced our finding of the hypermethylated gene.

Another potentially silenced PPI network stood out RNA regulation, epigenetic regulation, and gene silencing. Predicted interactions linked a member of the 3’-5’- exoribonuclease family (Manes.01G181600) and proteins with RNA degradation activity (RRP). A related gene (AtRRP6L1) promoting a loss-of-function mutation decreased DNA methylation, probably retaining RNAs and accumulating RNAs in association with chromatin [54]. Additional predicted proteins included RRP45b (subunit of the exosome) and CCR4-1, a relevant protein of the 3’-5’ repression complex for RNA degradation [55].

I. The PPI networks based on BRS Dourada’s hypomethylated genes.

Concerning the proteins encoded by hypomethylated genes of the sweet BRS Dourada, small PPI networks stood out. A network linked a transporter of the ABC family (ABCF1, Manes.02G005700), a Lysine -tRNA ligase (Manes.10G055800), and two 40S ribosomal proteins (S8-2, Manes.03G134200; S9-2, Manes.05G099700) with proteins/ribosomal subunits. In plants under (a)biotic stress, genes encoding ribosomal proteins (RPs) are differentially regulated, affecting ribosomal biogenesis and plant growth [56]. Mutations in RP-related genes have regulatory roles in plant development processes, showing the mutants some abnormalities and reduced cell growth/proliferation [57]. The analysis of the related genes, besides showing several polymorphisms, also detected the heterozygous condition of the sweet BRS Dourada variety in detriment of the homozygous or non-polymorphic status of BRS Formosa, as observed for the genes encoding Lysine -tRNA ligase (Manes.10G055800) and 40S ribosomal protein S8-2 (Manes.03G134200), both showing one SNP/INDEL of moderate impact (a missense variant) (S2 Table).

Regarding plant water use efficiency, some RPL proteins (60S ribosomal proteins) were already reported as relevant [58], and such proteins (AT2G17360, AT5G58420, AT5G67510, AT2G20450, and AT3G49910) appeared In the PPI network. Mutant plants for the RPL11 gene regulated by salt stress showed defective growth [59], while mutant plants for the AtRPL24 gene (A. thaliana) had their development affected, as well as their restart of mRNA translation of bZIP11 and ARF, both TFs relevant in abiotic stress responses [60]. In addition to the RPs, some interactions linked the PWP protein [periodic tryptophan (W) protein], the EMB complex [nucleolar associated protein complex (NOC4), and the 40S ribosomal protein S6-2 (RPSb)]. PWP proteins participate in ribosome biogenesis (18S ribosomal pre-RNA processing), while NOC4 proteins contribute to nucleolar processing of the 40S/90S rRNAs, maturation of 5,8S rRNA, and ribosome assembly [61].

Another component already reported in plants responding to (a)biotic stress [62] and predicted network member was the ABCF1 transporter (Manes.02G005700), which related gene (LrABCF1, Lilium regale) reinforced its involvement in addition to transmembrane transport inducing transcription in plants under abiotic stimuli (cold, salinity, and wounds) after SA and ethylene (ET) treatment.

Additional interactions highlighted the DNA mismatch repair protein MSH2 (Manes.01G020600) with other proteins acting in DNA damage repair (MSH, MLH, and PMS1). Components of the MMR (mismatch repair) system form MutS alpha heterodimers (MSH2-MSH6 heterodimer) that bind to the DNA mismatches, initiating the repair. These proteins correct base-base mismatches and insertion-deletion loops from eventual DNA damages [63]. The DNA repair keeps the plant metabolism working properly [64].

Also, interactions linked proteins from the SCAR and the Arp2/3 complexes. The ARP proteins play a key role in actin polymerization, mediating the formation of branched networks of actin. Furthermore, the Arp2/3 complex activates the SCAR complex, staying directly involved in plant growth [65].

II. The PPI networks based on BRS Dourada’s hypermethylated genes.

Proteins probably silenced by the hypermethylated genes from the sweet BRS Dourada presumed five networks. One network linked the 2-oxoglutarate dehydrogenase (component E1; Manes.10G034400) with the domain protein RING associated with BRCA1 (Manes.12G035400), a member of the mini-chromosome maintenance (MCM) family protein (Manes.17G068400), and an unknown protein (Manes.05G206400). Such proteins interact in origin of replication complexes (ORC) and involve cell division cycle (CDC) proteins. The MCM proteins participate in the initiation of DNA replication (transition from G1 to S phase). Such a phase requires the formation/activation of the pre-replicative complex, which begins with the binding of CDC6 to ORCs during the G1 phase, allowing the recruitment of the MCM protein complex. In turn, the S phase is triggered by the activation of this complex by cyclin-dependent kinases, which leads to a change from the pre-replication to a post-replication complex [66]. The hypermethylation of the related genes will reduce the cell proliferation after the stress.

Other interactions linked the DEAD-box ATP-dependent RNA helicase 47 (mitochondrial, Manes.02G025500) with translation initiation factor 3 (Manes.02G056600) and members of the eIF family. The eIFs (families eIF1 to eIF6) act in translation initiation and protein synthesis [67]. The eIF1, eIF1A, and eIF3 form the 43S pre-initiation complex (PIC) through the binding of ribosomal subunits and in turn, eIF4 promotes the correct combination of PIC with the CAP-5′ end [68]. Some eIF genes have already been reported in plants responding to abiotic stresses. Genes such as MieIF1A-a, MieIF5, and MieIF3sB were strongly expressed in plants under salinity, drought (PEG), and cold stress, respectively [68]. Since some of the mentioned proteins involved hypermethylated genes, the sweet BRS Dourada probable does not explore efficiently that alternative.

Additional interaction linked a member of the CPSF complex (cleavage specification factor and polyadenylation, Manes.10G123600) with the PAP (nuclear poly(A)-polymerase) protein. The CPSF proteins play a key role in 3’-end pre-mRNA formation by interacting with poly(A)-polymerase and protein factors to cleave the transcript and add the poly(A) tail. Such proteins could be involved in the post-transcriptional control of plants responding to stress as highlighted during oxidative stress [69]. Another interaction linked protein of the RNA exosome complex (RRP45B, Manes.01G181600) to a member of the ARM repeat superfamily protein (Manes.15G056200). The RNA exosome complex is a multiprotein complex that targets RNAs acting from maturation to quality control and final turnover. Additional interaction involved a PWI domain splicing factor (Manes.01G169300) with the 4-a subunit of the THO complex (Manes.04G032200). The PWI domain in splicing factors is the RNA/DNA binding domain and it has multiple functions in pre-mRNAs processing [70]. In turn, the THO complex, subunit 4, participates in the biogenesis of the ribonucleoprotein particle (mRNP) that aids in the export of mRNAs out of the nucleus [71]. Therefore, the RNA regulation could be affected in the BRS Dourada stress response. Thus, the potential hypermethylation of those related genes could compromised the RNA regulation and also the plant stress-response.