Simultaneous CRISPR/Cas9-mediated editing of cassava eIF4E isoforms nCBP-1 and nCBP-2 confers elevated resistance to cassava brown streak disease

Cassava brown streak disease (CBSD) is a major constraint on cassava yields in East and Central Africa and threatens production in West Africa. CBSD is caused by two species of positive sense RNA viruses belonging to the family Potiviridae, genus Ipomovirus: Cassava brown streak virus (CBSV) and Ugandan cassava brown streak virus (UCBSV). Diseases caused by the family Potyviridae require the interaction of viral genome-linked protein (VPg) and host eukaryotic translation initiation factor 4E (eIF4E) isoforms. Cassava encodes five eIF4E isoforms: eIF4E, eIF(iso)4E-1, eIF(iso)4E-2, novel cap-binding protein-1 (nCBP-1), and nCBP-2. Yeast two-hybrid analysis detected interactions between both CBSV and UCBSV VPg proteins and cassava nCBP-1 and nCBP-2. CRISPR/Cas9-mediated genome editing was employed to generate eif4e, ncbp-1, ncbp-2, and ncbp-1/ncbp-2 mutants in cassava cultivar 60444. Challenge with CBSV showed that ncbp-1/ncbp-2 mutants displayed delayed and attenuated CBSD aerial symptoms, as well as reduced severity and incidence of storage root necrosis. Suppressed disease symptoms were correlated with reduced virus titer in storage roots relative to wild-type controls. However, full resistance to CBSD was not achieved, suggesting that remaining functional eIF4E isoforms may be compensating for the targeted mutagenesis of nCBP-1 and nCBP-2. Future studies will investigate the contribution of these other isoforms to development of CBSD.

eIF(iso)4E as a loss of susceptibility locus (Duprat et al., 2002;Lellis et al., 2002). More broadly, polymorphisms in eIF4E isoforms of pepper, tomato, lettuce, pea, and other crops confer resistance to numerous potyviruses (Robaglia and Caranta, 2006). The direct physical interaction between potyvirus VPg and specific host eIF4E isoforms is well supported through in vitro and in vivo binding assays (Kang et al., 2005;Leonard et al., 2000;Schaad et al., 2000;Wittmann et al., 1997;Yeam et al., 2007). In most of these cases, amino acid substitutions within the interaction domains on either VPg or eIF4E isoforms abolished infection, highlighting the necessity of eIF4E isoform interaction.
The eIF4E protein family plays an essential role in the initiation of cap-dependent mRNA translation. eIF4E isoforms interact with the 5'-7mGpppN-cap of mRNA and subsequently recruit a complex of initiation factors for ribosomal translation. eIF4E and its different isoforms, eIF(iso)4E and novel cap-binding protein (nCBP), vary in degrees of functional redundancy and may have undergone neo-or subfunctionalization (Browning and Bailey-Serres, 2015). Little is known regarding nCBPs, in particular. Studies in A. thaliana have shown that nCBP exhibits weak cap-binding, similar to eIF(iso)4E, and increased levels in cap-binding complexes at early stages of cell growth (Kropiwnicka et al., 2015, Bush et al., 2009. However, the specialized function of nCBPs remains unknown. Potyviruses hijack the eIF4E protein family via their VPg for translation initiation, genome stability, and/or viral movement ( Fig S1) (Contreras-Paredes et al., 2013;Eskelin et al., 2011;Gao et al., 2004;Miras et al., 2017;Zhang et al., 2006). Transgenic approaches leveraging amino acid changes that abolish interaction with VPg or loss of the VPg-associated eIF4E protein have previously been implemented as a form of potyviral disease control (Cui and Wang, 2017;Piron et al., 2010;Wang, 2015).
Targeted genome editing techniques have emerged as alternatives to classical plant breeding and transgenic methods (Belhaj et al., 2015). The CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats)/Cas9 (CRISPR associated protein 9) system has rapidly become a favored tool for biotechnology because of its simple design and easy construction of reagents. The Cas9 nuclease is recruited to a specific site within the genome via a guide RNA (gRNA) (Jinek et al., 2012). Upon binding, Cas9 induces a double strand break (DSB) at the target site (Belhaj et al., 2015). Repair of the DSB by error-prone non-homologous end joining (NHEJ) can generate insertion or deletion (INDEL) mutations that disrupt gene function by altering the reading frame and/or generate a premature stop codon (Britt, 1999;Gorbunova and Levy, 1999). We aimed to apply the CRISPR/Cas9 technology to knockout the VPg-associated cassava eIF4E isoform(s). This approach to engineering potyvirus resistance has been successfully demonstrated in A. thaliana and cucumber (Chandrasekaran et al., 2016;Pyott et al., 2016). Here, we show that targeted mutagenesis of specific cassava eIF4E isoforms nCBP-1 and nCBP-2 by the CRISPR/Cas9 system reduces levels of CBSD associated disease symptoms and CBSV accumulation in storage roots. Simultaneous disruption of both nCBP isoforms resulted in a larger decrease in disease symptoms than disruption of either isoform individually.

Identification and sequence comparison of eIF4E isoforms in cassava varieties
To identify the eIF4E family protein(s), a BLAST search of the AM560-2 cassava cultivar genome (assembly version 6.1) was done via Phytozome using A. thaliana eIF4E family proteins as the queries (Bredeson et al., 2016;Goodstein et al., 2012). Five cassava proteins were found that phylogenetically branched with the eIF4E, eIF(iso)4E, and nCBP sub-groups ( Fig 1A). Two of the cassava eIF4E family proteins joined within the eIF(iso)4E sub-group, and another two joined within the nCBP sub-group. This is in agreement with findings by Shi et al. (2017). Percent identity analysis further supported this grouping as the eIF(iso)4E-and nCBPsimilar proteins had high amino acid identity (Fig 1B). Based upon this phylogenetic analysis, one eIF4E, two eIF(iso)4E, and two nCBP genes cassava genes were re-named according to their sub-groups, as described in Fig 1C.

nCBP-1 and nCBP-2 isoforms interact with CBSV and UCBSV VPg in yeast
To identify the interaction partner(s) for the CBSV and UCBSV VPgs, a yeast two-hybrid system was used to assess the VPg-eIF4E isoform interactions. The VPg proteins from CBSV Naliendele isolate TZ:Nal3-1:07 (CBSV-Nal) and UCBSV isolate UG:T04-42:04 (UCBSV-T04) were fused to the B42 activation domain and transformed into yeast strain EGY48. All five capbinding proteins were fused to the LexA DNA-binding domain and transformed into VPg yeast lines. Likewise fused, TuMV VPg and A. thaliana eIF(iso)4E were transformed into yeast as a positive control, and empty vectors were transformed as negative controls. Five colonies from each transformation were plated on selective media supplemented with X-gal. In this assay, a blue color is indicative of protein-protein interaction dependent activity of the β -galactosidase reporter. Based upon high amino acid sequence identity within eIF4E-family subgroups, we hypothesized both members of a sub-group would interact with CBSV and UCBSV VPgs. Both nCBP-1 and nCBP-2 showed strong interactions with the VPgs, visually comparable to the positive control (Fig. 2). nCBP-1 and nCBP-2 were selected for CRISPR/Cas9-mediated editing to abolish the critical VPg-eIF4E family protein interaction.

Site-specific mutation of eIF4E isoforms by transgenic expression of sgRNA-guided Cas9
CRISPR/Cas9 was employed to generate mutant alleles of cassava eIF4E isoforms. Seven constructs were assembled to target various sites in nCBP-1, nCBP-2, and eIF4E (Table 1). Agrobacterium carrying these constructs were then used to transform friable embryogenic calli (FEC) derived from cassava cultivar 60444 (Fig. 3). Transgenic T0 plants were selected in tissue culture using the nptII selectable marker in order to recover plants in which the CRISPR/Cas9 reagents had been integrated into the plant genome. Multiple independent T0 transgenic plant lines were recovered for each construct (Table S1). Sites in each nCBP gene were targeted to disrupt restriction enzyme recognition sequences (Fig. S2). Restriction digestion done on a PCR product from T0 plants using restriction enzyme SmlI indicated successful mutagenesis of both nCBP genes (Fig. S2). Cassava is diploid, carrying two copies of each nCBP gene. Absence of the wild-type digested product indicates that both alleles were successfully mutagenized.
The range of mutations generated in each transgenic plant was analyzed by subcloning and sequence analysis, revealing an array of homozygous, bi-allelic, heterozygous, complex, and wild-type genotypes (Table S1). Bi-allelic mutations contained different mutations on the two alleles. Heterozygous plants carried one mutagenized allele and one wild-type allele. Plants were considered complex if they carried more than two sequence patterns, strongly suggesting chimerism (Odipio et al., 2017;Zhang et al., 2016). The genotypes of edited plants had Cas9induced INDELs ranging from insertions of 1 to 16 bp and deletions as large as 127 bp (Table  S1). Review of all genotyped plants revealed that 13/55 (24%) carried homozygous mutations, 31/55 (56%) carried bi-allelic mutations, 1/55 (2%) were heterozygous, 5/55 (10%) were complex, and another 5/55 (10%) were wild-type genotypes (Table 1). In total, 80% of plants contained either homozygous or bi-allelic mutations, and CRISPR/Cas9 activity was observed in 91% of the plants studied.

Sequence analysis of INDEL-induced frameshifts in nCBPs identifies unpredicted ncbp-1 splice variants
Given the yeast two-hybrid interaction between the nCBPs and viral VPg proteins, we chose to test the effects of mutations in nCBP-1 and nCBP-2 individually, as well as both nCBPs in tandem, in CBSV and UCBSV disease trials in a greenhouse. Lines with homozygous mutations in exon 1 were prioritized (Fig. 3). The mutant lines chosen for these trials, ncbp-1, ncbp-2, ncbp-1/2 #2, and ncbp-1/2 #8, each had an INDEL at the 3' end of the first exon of each targeted gene (Fig. 4). The INDELs either directly resulted in a frameshift or disrupted the exonintron junction so that an out of frame splice variant was predicted to be produced (Fig. S3). To further characterize these mutations, cDNA clone sequencing (clone-seq) was done (Fig. 3, 5). The homozygous ncbp-1 allele from ncbp-1/2 #2 was also analyzed for comparison. Of nine ncbp-1 clones from ncbp-1/2 #2, eight displayed the wild-type splicing pattern (referred to as type 1) with the A-insertion predicted from genomic DNA sequence results. This generates a frameshift. One splice variant (referred to as type 2) was also observed (Fig. 5). This alternative splice form results in an insertion of 35 nucleotides but does not shift the reading frame. Thus, this splice variant encodes a full protein with a 12 amino acid internal insertion. This splicing pattern was not observed in any wild-type nCBP-1 clones, however, may occur at low frequency.
Clone-seq analysis of seven ncbp-1 clones from mutant line ncbp-1/2 #8 cDNA similarly found predicted INDELs. Two clones displayed the wild-type (type 1) splicing pattern and the predicted A-deletion that alters the reading frame. Four clones showed a sequence pattern that suggests a third splicing variant (type 3) at an upstream alternative splice site (Fig. 5) (Reddy et al., 2007). Both observed cDNA sequence patterns are frameshifted.
ncbp-1/ncbp-2 double mutants exhibited reduced UCBSV symptom incidence and slowed CBSV symptom onset CBSV has been described as being more virulent than UCBSV (Kaweesi et al., 2014;Mohammed et al., 2012;Ogwok et al., 2015). Challenge with a stronger pathogen may mask subtle phenotypes that could be presented during challenge with the weaker pathogen. As such, three disease trials for each virus species were carried out. The ncbp-1, ncbp-2, ncbp-1/2 #2, ncbp-1/2 #8, and wild-type 60444 plants were chip-bud graft inoculated with either CBSV-Nal or UCBSV-T04 (Wagaba et. al, 2013). Aerial disease incidence was scored every week for 12 to 14 weeks after grafting. This analysis describes the percentage of plants showing any level of foliar or stem symptoms at each time point. At least five replicate plant clones were included for each genotype (n≥5). Inoculation with UCBSV-T04 did not produce stem symptoms. Consequently, only foliar disease incidence was recorded for those trials. Fluctuations in the percentage of plants that exhibited symptoms at each time point (% incidence) results from the shedding of symptomatic leaves throughout the experiment. Disease incidence for each genotype varied across experimental replicates, possibly due to variance in viral load of the chip-bud donor or a change in environmental conditions affecting disease pressure. However, a consistent relationship between genotypes was observed. Across all three experimental replicates, ncbp-1/ncbp-2 double mutants exhibited reduced symptom incidence relative to wildtype plants and ncbp-1. The ncbp-2 phenotype was intermediate between the double mutant and wild-type incidence rates (Fig. 6b, S4). Aerial UCBSV virus titer was measured for one experiment, but proved to be highly variable across biological replicates (Fig. S9).
In challenges with CBSV-Nal, wild-type plants produced strong foliar and stem symptoms in contrast to the UCBSV trials. Across all three experimental replicates, ncbp-1/ncbp-2 double mutants exhibited delayed symptom development relative to wild type and ncbp-1 (Fig. 6a, S4). In two experiments the double mutant lines reached 100% incidence at a markedly reduced rate relative to wild type and ncbp-1; in the remaining experiment, the same lines never rose above 43% incidence. ncbp-2 exhibited symptom incidence development similar to wild type and ncbp-1 in two experiments and displayed an intermediate phenotype in the third experiment (Fig 6b,  S4a, S4b).

ncbp-1/ncbp-2 lines exhibit reduced aerial symptom severity after challenge with CBSV
For the described CBSV challenges, combined leaf-stem scores were also used to track aggregate aerial CBSD severity for each genotype over time (Table 2, Fig. 6c, S5). Wild-type, ncbp-1, and ncbp-2 plants displayed similar levels of disease, although in one of three experiments ncbp-2 developed statistically significantly less severe symptoms than wild type or ncbp-1 (Fig. S5a). This experiment was the same one in which ncbp-2 symptom incidence was intermediate between wild type/ncbp-1 and ncbp double mutant levels (Fig. S4c). The ncbp-1/ncbp-2 double mutants had greatly reduced CBSD severity in all three trials. Area under the disease progression curve (AUDPC) analysis revealed the reduced aerial symptom severity in ncbp-1/ncbp-2 double mutants to be statistically significant in all three experimental replicates (Fig. 6d, S5). While ncbp-1/ncbp-2 stem symptom severity was reduced in all three experiments, average leaf symptom severity tracked closely with wild type in one experiment (Fig. 7, S6, S7). Despite this, it is clear that mutating both nCBP isoforms had an effect on CBSD disease development.

ncbp-1/ncbp-2 double mutant storage roots are less symptomatic and accumulate less virus
At 12 to 14 weeks after graft inoculation, storage roots were excavated and assessed for root necrosis. Only inoculation with CBSV-Nal produced storage root symptoms. Each storage root of a plant was divided into five sections and each section scored on a 1-5 scale for CBSD symptom severity (Fig. 8a). Average symptom scores for each genotype were compared. ncbp-2 and ncbp-1/ncbp-2 mutant lines all exhibited significantly reduced symptom scores relative to wild type and ncbp-1 (Fig. 8b). Reverse transcription-quantitative polymerase chain reaction (qPCR) was used to measure CBSV-Nal RNA levels in ncbp-1/ncbp-2 double mutants. Viral RNA levels in ncbp-1/ncbp-2 roots were reduced 43-45% compared to wild-type roots (Fig. 8c).

Discussion
The CRISPR/Cas9 system has emerged as a powerful tool for plant genome editing and rapid crop improvement. In the context of disease resistance in crop species, this system has been employed to target mildew-resistance locus O (MLO) in wheat, and generate broad potyvirus resistance in cucumber by disrupting function of the eIF4E gene (Chandrasekaran et al., 2016;Wang et al., 2014). In the present study, we targeted the nCBPs to assess their putative function as CBSD susceptibility factors in cassava.
Previous studies have shown that host eIF4E and viral VPg interaction is necessary for potyviral infection (Ashby et al., 2011;Charron et al., 2008;Kang et al., 2005;Leonard et al., 2000;Yeam et al., 2007). We identified five eIF4E family members in cassava, corroborating a recent analysis by Shi et al. (2017). Cassava is thought to be an ancestral allopolyploid, likely yielding the two eIF(iso)4E and nCBP genes (Fregene et al., 1997). The presence of multiple eIF4E isoforms may indicate sub-functionalization and specialization in translational control of differently methylated mRNA cap structures, or confer some functional redundancy that eases constraints on eIF4E evolution for potyvirus resistance (Carberry et al., 1991;Charron et al., 2008;Moury et al., 2014). Attempts to identify markers associated with CBSD resistance indicate that multiple loci are involved, and transcriptional analyses suggest the contribution of hormone signaling pathways (Maruthi et al., 2014;Masumba et al., 2017). Examination of CBSD-resistant, -tolerant, and -susceptible cultivars by Shi and colleagues also found that these categories are not associated with eIF4E family single nucleotide polymorphisms (Shi et al., 2017). As such, a biochemical study of the VPg and eIF4E family interaction was essential to identify a potential susceptibility gene(s).
Yeast two-hybrid analysis showed strong interactions between the nCBPs and the CBSV-Naliendele and UCBSV-T04 VPg proteins, to levels visually equivalent to the positive control TuMV VPg-A. thaliana eIF(iso)4E interaction (Fig. 2). First identified in A. thaliana, nCBP has a distinct amino acid sequence and exhibits methylated-cap-binding property (Kropiwnicka et al., 2015;Ruud et al., 1998). To date, there is no precedent for recruitment of nCBPs by VPg proteins belonging to the family Potyviridae. However, this isoform has been identified as a novel recessive resistance gene toward viruses in the Alphaflexiviridae and Betaflexiviridae families (Keima et al., 2017). In the case of potexvirus Plantago asiatica mosaic virus (PlAMV), nCBP loss in A. thaliana impaired viral cell-to-cell movement by inhibiting accumulation of viral movement proteins from a subgenomic RNA. It is unclear if A. thaliana nCBP is either required for subgenomic RNA stability or translation of PlAMV movement proteins. In contrast, there is evidence that many members of Potyviridae produce the potyvirus P3N-PIPO movement protein through RNA polymerase slippage (Hagiwara-Komoda et al., 2016;Olspert et al., 2015;Rodamilans et al., 2015). As such, while nCBP may similarly play a critical role in the accumulation of the CBSV movement protein, the underlying mechanism is likely to be different from those used during Alpha-and Betaflexavirdae infection. It has also been found that distantly related potyviruses that infect a common host may utilize different eIF4E isoforms for movement (Contreras-Paredes et al., 2013;Eskelin et al., 2011;Gao et al., 2004;Miras et al., 2017). Furthermore, evidence suggests that some potyviruses may utilize one specific isoform for translation and another distinct isoform for movement (Contreras-Paredes et al., 2013;Gao et al., 2004). This complexity makes it difficult to predict what roles cassava nCBPs may have in the CBSV life cycle. Further study is required to characterize the role of nCBP in translation, genome stability, and viral movement processes.
Five CRISPR/Cas9 expression constructs were designed and transgenically integrated into the cassava genome to target the nCBP genes individually and simultaneously. In transgenic plant lines, mutations were detected by restriction enzyme site loss analysis and Sanger sequencing. We observed homozygous, bi-allelic, heterozygous, complex, and wild-type genotypes. Homozygous mutations may have been generated by identical NHEJ repair, or homologous recombination-based repair from the opposite allele. Considering the low incidence of the latter in plants, identical NHEJ repair may be more likely (Peng et al., 2016). While transgenic plants derived from FECs are thought to be of single cell origin (Fig. 3a), reducing the likelihood of transgenic chimeras (Schreuder et al., 2001;Taylor et al., 1996), Odipo et al. (2017) have reported the production of chimeric plants via CRISPR/Cas9-mediated gene editing of phytoene desaturase in cassava. In depth analyses of lines with complex genotypes were not pursued, but they are likely chimeras resulting from Cas9/sgRNA activity being delayed until after embryogenic units began to replicate (Odipio et al., 2017;Zhang et al., 2014). Integrating CRISPR/Cas9 constructs into the cassava genome proved to be efficient for achieving gene editing as 91% of the transformed plant lines carried INDELs at the target sites, and desired homozygous and bi-allelic mutations were observed in 80% of plant lines. These frequencies compare favorably to previous CRISPR/Cas9-mediated mutagenesis studies in cassava, rice and tomato (Ma et al., 2015;Odipio et al., 2017;Pan et al., 2016). Homozygous and bi-allelic genotype frequencies in rice and tomato were approximately 80% and 19%, respectively.
CBSD inoculation experiments were limited to plant lines carrying homozygous and bi-allelic mutations that resulted in a frameshift or disrupted the exon-intron junction, thus resulting in the production of frameshifted splice variants. Single-nCBP and double-nCBP mutant lines were challenged with isolates CBSV-Naliendele and UCBSV-T04 (Beyene et al., 2017;Wagaba et al., 2013). Levels of resistance to CBSD were strongly correlated with disrupting function of both nCBP genes. Over the course of 12 to 14 weeks, double-nCBP mutant lines exhibited delayed CBSD aerial (combined leaf and stem) symptom onset and reduced severity. Full resistance to CBSD was not achieved as some brown streaking of the stem occurred in double mutants, and leaf symptom severity tracked closely with wild type in our last experiment. Furthermore, toward the end of each challenge, aerial symptom incidence in the double-nCBP mutants approached wild-type levels (Fig. 6, Fig. S4). Single-nCBP mutants were generally not significantly different from the susceptible wild-type plants in response to CBSV-Nal challenges, but symptom incidence for ncbp-2 fluctuated between wild-type and double mutant levels across UCBSV-T04 challenges. These results could be due to UCBSV being less virulent than CBSV (Kaweesi et al., 2014;Mohammed et al., 2012;Ogwok et al., 2015). nCBP-2 may be more important for viral accumulation, and it could be that the mutant phenotype is masked by challenge with a more virulent pathogen or conditions conducive to high disease pressure. The latter may also influence inoculum concentrations in donor plants and result in the observed experiment-toexperiment variation in disease severity (Fig. S4, S5, S6, S7). This is consistent with observations that increases in ncbp-2 symptom incidence to wild-type levels occurred when symptom incidence in ncbp-1/ncbp-2 plants was elevated relative to other experiments. Variation in disease pressure may also explain the inconsistent leaf phenotype in the double-nCBP mutant plants. The mechanisms underlying CBSD leaf and stem symptom development are unknown and it is possible that symptoms in different tissue types can be unequally influenced by varying levels of disease pressure.
At our challenge endpoints, symptom development and virus accumulation in the agronomically important tuberous roots were analyzed. Consistent with observations of aerial tissues, symptom severity in the roots was significantly lower in the double-nCBP mutants than in wildtype plants. CBSV titers in roots were significantly reduced in the double-nCBP mutants. Interestingly, the mutagenesis of nCBP-2 resulted in reduced symptom severity as compared to wild-type plants and ncbp-1 mutant lines. This result may be explained by the 10 fold higher expression of nCBP-2 in the roots, but also highlights the possibility that nCBP-1 and nCBP-2 are not fully redundant (Fig. S8) (Wilson et al., 2017). Assuming that the effects of nCBP mutations are due to the disruption of VPg-nCBP interactions, it is also possible that CBSV VPg may have a higher dependence for nCBP-2 than nCBP-1. Forcing CBSV to utilize less abundant isoforms, or those with suboptimal binding affinities, could attenuate CBSD progression. Additional transcriptional and biochemical studies will be needed to investigate these hypotheses.
Several explanations may account for the incomplete CBSD resistance of double-nCBP mutant cassava plants. First, unpredicted splice variants may have coded for proteins that were biologically functional for viral infection, at least in part (Fig. 5). The activation of normally silent, cryptic, splice sites is consistent with the intron definition of splicing (Lal et al., 1999). Under this model, disruption of the wild-type splice site motifs, typically dinucleotides GU and AG at the 5' and 3' termini of introns, respectively, can activate cryptic splice sites that redefine intron boundaries and consequently frameshifts the mature transcript (Reddy et al., 2007). This is consistent with our cDNA clone-seq analysis identifying a type 3 splice variant of ncbp-1 from line ncbp-1/2 #8. However, the type 2 ncbp-1 variant of ncbp-1/2 #2 does not appear to be the result of splice site disruptions. Furthermore, it codes for full length nCBP-1 with a 12 amino acid extension. It is possible that similar unpredicted splice variants exist at low abundance in ncbp-1/2 #8. Complementation assays will need to be performed to determine whether such putative splice variants can be utilized by the viruses. The level of these transcripts and/or their encoded protein's affinity for CBSV and UCBSV VPgs are likely low considering the the clear impact on CBSD development. Second, CBSV and UCBSV VPgs may have some inherent, low-level affinity for the other eIF4E isoforms. Co-expression of the cassava eIF(iso)4E-1 and -2 with VPg from both species showed weak reporter activation that could be interpreted as weak interaction or reporter auto-activation as seen in the TuMV VPg plus empty vector control (Fig.  2). VPg is an intrinsically disordered protein, which could enable it to bind several different proteins (Jiang and Laliberté, 2011). The ability to use multiple eIF4E isoforms has precedence, such as in Pepper veinal mottle potyvirus for which simultaneous mutations of both eIF4E and eIF(iso)4E is required to restrict infection (Gauffier et al., 2016;Ruffel et al., 2006). Recruitment of eIF4E or the two eIF(iso)4E isoforms by CBSV/UCBSV could result in sub-optimal viral replication or movement, resulting in lower symptom severity and incidence. This has previously been hypothesized by Chandrasekaran et al. (2016) for breaking of eif4e-mediated resistance in cucumber. Further investigation will be required to test this hypothesis in cassava.
CBSD remains a major threat to food security in sub-Saharan Africa. Mitigation of crop losses is imperative to sustaining Africa's rapidly growing population. Due to the challenges of breeding cassava, genetic editing strategies provide an attractive means to engineer disease resistance. In this study, we show that simultaneous CRISPR/Cas9-mediated editing of the nCBP-1 and nCBP-2 genes confers statistically significantly elevated resistance to CBSD. Editing of these host translation factors significantly hampers CBSV accumulation in the plant. By stacking this approach with other forms of resistance such as RNAi, potential exists to provide improved cassava varieties with robust and durable resistance to CBSD.

Production of plants and growth conditions
Transgenic cassava lines of cultivar 60444 were generated and maintained in vitro as described previously (Taylor et al., 2012). In vitro plantlets were propagated, established in soil, and transferred to the greenhouse (Taylor et al., 2012;Wagaba et al., 2013). Throughout the course of a disease trial, all plants were treated bi-weekly for pest control by gently spraying the undersides of all leaves with water.

Identification and phylogenetic analysis of eIF4E isoforms
BLAST search of the AM560-2 cassava cultivar genome was done via Phytozome V10 using A. thaliana eIF4E family proteins as the queries. The coding sequences of each isoform were verified by comparison to RNA-seq data (Cohn et al., 2014). Clustal Omega (EMBL-EBI) was used to generate the percent identity matrix of all eIF4E isoform amino acid sequences (Goujon et al., 2010;Sievers et al., 2014). MEGA 6 software was used to generate a phylogenetic tree of the cassava and A. thaliana eIF4E isoforms (Tamura et al., 2013). The evolutionary history was inferred by using the Maximum Likelihood method based on the Le_Gascuel_2008 model (Le and Gascuel, 1993). This amino acid substitution model was determined as best fit using the MEGA 6 model test. The tree with the highest log likelihood (-2025.7966) is shown. Initial tree(s) for the heuristic search were obtained automatically by applying Neighbor-Join and BioNJ algorithms to a matrix of pairwise distances estimated using a JTT model, and then selecting the topology with superior log likelihood value. A discrete Gamma distribution was used to model evolutionary rate differences among sites (5 categories (+G, parameter = 1.9218)). The analysis involved 9 amino acid sequences. All positions containing gaps and missing data were eliminated.

CRISPR/Cas9 binary construct design
CRISPR/Cas9 construct design and assembly of entry clone pCR3 were conducted as described by Paula de Toledo Thomazella et al. (2016). CRISPR/Cas9 constructs targeting two sites were assembled via Gibson Assembly of the other U6-26/gRNA into the SacII site of the entry clone. Flanked by the attL1 and attL2 recombination sequences, the cassette carrying the Cas9/gRNA expression system was Gateway cloned into the binary destination vector pCAMBIA2300 (Hajdukiewicz et al., 1994).

gRNA design and cloning
Target sequences were identified in nCBP-1 and nCBP-2 genes of cassava using the online CRISPR-P software (Lei et al., 2014). This tool was used to select targets with predicted cut sites within exons, minimal off-target potential, and overlapping restriction enzyme recognition sites.
gRNA forward and reverse primers were designed with overhangs compatible with the BsaI-site described above. The Golden Gate (GG) cloning method was used to BsaI digest the pCR3 vector and ligate in the gRNA. In the case of the dual targeting CRISPR/Cas9 construct, the pCR3 vector bearing gRNA1 was digested with SacII, a site within the LR clonase attL sequences. The A. thaliana U6-26 promoter and gRNA2 were PCR amplified using primers suitable for Gibson Assembly into the SacII cut site of the digested pCR3-gRNA1 vector. For Gibson Assembly, 100 ng of SacII-digested vector was incubated with 200 ng of U6-26p-gRNA2 PCR amplicon and Gibson Assembly Master Mix for one hour and transformed into E. coli (NEB5a). Sequences of cloned CRISPR constructs were verified via Sanger sequencing.

Yeast two-hybrid
The eIF4E isoforms were amplified by PCR using primers suitable for Gibson Assembly into the BamHI site of pEG202. Yeast codon optimized coding sequences of the CBSV and UCBSV VPg were synthesized through Genewiz, Inc (South Plainfield, NJ, USA). The VPg coding sequences were amplified using primers suitable for Gibson Assembly into the EcoRI site of pEG201. Yeast two-hybrid analyses were carried out as described previously (Kim et al., 2014).

Genotyping and mutant verification
100 mg of leaf tissue was collected from T0 transgenic cassava in vitro plantlets and genomic DNA extracted using the CTAB extraction procedure (Murray and Thompson, 1980). Transgenic plants were genotyped for Cas9-induced mutagenesis via restriction enzyme site loss (RESL) and Sanger sequencing (Voytas, 2013). Initially, 100 ng of genomic DNA was PCR amplified using primers encompassing the Cas9 target sites. PCR amplicons were gel purified on 1.5% agarose gel and purified with the QIAquick Gel Extraction Kit. For RESL analysis, 50 ng of PCR amplicon were digested with restriction enzyme SmlI for 12 hours, then run and visualized on a 1.5% agarose gel. For genomic and cDNA sequence analysis, the amplicons were subcloned and Sanger sequenced through the UC Berkeley DNA Sequencing Facility. Between six to eight clones were sequenced to discriminate INDEL polymorphisms and sequences were aligned to the intact nCBP-1 and nCBP-2 using SnapGene software (from GSL Biotech; available at snapgene.com).

CBSV and UCBSV inoculation and disease scoring
Prior to virus challenge, micropropagated cassava plantlets were transplanted to soil, allowed to acclimate for six to eight weeks, and chip-bud graft inoculation performed as described previously (Wagaba et al., 2013). Briefly, one plant of each genotype received an axillary bud from a single previously infected wild type plant, resulting in one inoculation cohort. Multiple cohorts were used in a single experiment to control for donor plants with varying viral concentrations.
Shoot tissues were scored two to three times a week for 12 to 14 weeks. Leaves and stems were each scored on separate 0-4 scales ( Table 2). Leaf and stem scores were then summed to generate an overall aerial severity score for a particular time point. These data were used to calculate the area under the disease progression curve (Simko and Piepho, 2012). To assess symptom severity in storage roots, each storage root was evenly divided into five pieces along its length. Each storage root piece was then sectioned into one-centimeter slices and the maximum observed severity was used to assign a symptom severity score to that storage root piece. The scores for all storage root pieces of a given plant were then averaged to determine the overall severity score.

Storage root viral titer quantification
Five to ten representative storage root slices per plant were collected for viral titer quantification. Samples were flash frozen in liquid nitrogen and lyophilized for two days. Lyophilized storage roots were pulverized in 50 mL conical tubes with a FastPrepTM-24 instrument (MP Biomedicals) and 75 mg of pulverized tissue was aliquoted into Safe-Lock microcentrifuge tubes (Eppendorf) pre-loaded with two mm zirconia beads (BioSpec Products). Samples were flash frozen in liquid nitrogen, further homogenized to a finer consistency, and one mL of Fruit-mate (Takara) added to each sample. Samples were homogenized and subsequently centrifuged to remove debris. The supernatant was removed, mixed with an equal volume of TRIzol LS (Thermo Fisher), and the resulting mixture processed with the Direct-zol RNA MiniPrep kit (Zymo Research). Resulting total RNA was normalized to a standard concentration and used for cDNA synthesis with SuperScript III reverse transcriptase (Thermo Fisher).
Quantitative PCR was done with SYBR Select Master Mix (Thermo Fisher) on a CFX384 Touch Real-Time PCR Detection System (Bio-Rad). Primers specific for CBSV-Nal HAM1-LIKE and cassava PP2A4 were used for relative quantitation (Table S2). Normalized relative quantities were calculated using formulas described by Hellemans et al. (2007). For combined analysis of all experimental replicates, normalized relative quantities for all samples were further normalized as a ratio to the geomean of wild type for their respective experiments. Data were then pooled and a Mann-Whitney U test was used to assess statistical differences. Figure S1. Roles of host eIF4E-potyvirus VPg interaction and sources of recessive resistance. (a) Linkage of potyvirus VPg to its binding site on eIF4E can provide translation initiation via recruitment of necessary factors and ribosomal subunits, genomic stability via protection from host-encoded exonucleases, and intracellular trafficking via eIF4G microtubule binding activity. (b) Non-conservative amino acid changes and gene deletions that abolish VPg-eIF4E binding removes above described roles, therefore conferring recessive resistance Figure S2. CRISPR/Cas9-induced mutagenesis evident in nCBP-1 (a) and nCBP-2 (b) via restriction enzyme site loss (RESL). PCR amplicons of targeted regions were digested with SmlI. Map of amplicons with nCBP exon (purple), protospacer adjacent motif (red), gRNA spacer (green), predicted Cas9 cut site (black arrow), and overlapping SmlI restriction enzyme recognition site (bold, red). Bands are measured relative to O'Gene Ruler 1 kb Plus Ladder. Experimental banding pattern is consistent with predicted RESL. Figure S3. CRISPR/Cas9 -induced mutagenesis creates out of frame alternate splice variants. Exon 1 and exon 2 splice junction of nCBP-1 (a) and nCBP-2 (b) were examined via sequence analysis of cDNA. Predicted Cas9 cut site is shown as a black arrow. STOP codon is shown as starred, red box. Figure S4. ncbp-1/ncbp-2 double mutants exhibit reduced UCBSV symptom incidence and delayed CBSV symptom onset (a), (b), aerial symptom incidence reported as percent of wild type, ncbp-1, ncbp-2, or ncbp-1/ncbp-2 plants bud-graft inoculated with UCBSV T04 (n≥10). (c), (d), aerial symptom incidence as previously described in plants inoculated with CBSV Naliendele (n≥7). Figure S5. ncbp-1/ncbp-2 double mutants exhibit reduced aerial CBSV symptom severity (a), (b), disease progression curves of wild type, ncbp-1, ncbp-2, or ncbp-1/ncbp-2 plants budgraft inoculated with CBSV Naliendele (n≥7). Leaf and stem symptoms were each scored on a 0-4 scale and summed to obtain an aggregate aerial score. (c), (d), average area under the disease progression curve (AUDPC) derived from data plotted in A and B. Error bars indicate standard error of the mean. Statistical differences were detected by Welch's t-test, α =0.05, *≤0.05, **≤0.01, ****≤0.0001. Figure S6. ncbp-1/ncbp-2 double mutant stem symptom severity is consistently reduced across all experiments Separate leaf and stem disease progression curves for wild type, ncbp-1, ncbp-2, or ncbp-1/ncbp-2 plants bud-graft inoculated with CBSV Naliendele (n≥7). Leaf and stem symptoms were each scored on a 0-4 scale. (a), (c), and (e) represent leaf disease progression curves from three different experiments while (b), (d), and (f) represent corresponding stem disease progression curves. Error bars represent standard error of the mean. Figure S7. 12-2016 CBSV challenge leaf symptom severity is similar across all genotypes Wild type, ncbp-1, ncbp-2, or ncbp-1/2 plants bud-graft inoculated with CBSV Naliendele isolate all develop widespread chlorotic leaf symptoms. Leaf images were taken near 12-2016 challenge endpoint. Scale bar denotes 1 cm. Figure S8. nCBP-2 is highly expressed in storage roots Heat map describing tissue specific expression of cassava eIF4E isoforms. Data was extracted from the Bart Lab Cassava Atlas (http://shiny.danforthcenter.org/cassava_atlas/). Expression values are defined as fragments per kilobase of transcript per million mapped reads (FPKM).       (a) Schematic of canonical and alternative nCBP-1 splice sites. Boxed region of the nCBP-1 gene model is enlarged below. Exon and intron sequences are given in capital and small letters, respectively. Green and red boxes highlight splice motifs at the 5' and 3' end of introns, respectively. Type 1 splicing produces the predicted wild type nCBP-1 cDNA sequence. Type 2 and 3 splicing are observed in ncbp-1/2 lines #2 & #8, respectively. (b) cDNA sequences detected in clone-seq experiments. Red boxes denote INDELs resulting from both CRISPR/Cas9-mediated edits and alternative splicing. In ncbp-1/2 #2, type 2 splicing results in retention of 3' sequence from intron 1 of one ncbp-1 allele (1 of 9 clones sequenced). In ncbp-1/2 #8, an INDEL disrupting the canonical splice motif between exon 1 and intron 1 of ncbp-1 results in a type 3 splice variant (4 of 6 clones sequenced).          ATG-118nt-CTGGTTTGCATCCTCTCAAGgtt-----1190nt-----tagaagtgcttattgggcttttgcatatctgtgatcagCAC Intron 1 Type 3 Type 2 Type 1 Figure 5. Alternative splicing of ncbp-1 alleles is detected in ncbp-1/2 double mutants (a) Schematic of canonical and alternative nCBP-1 splice sites. Boxed region of the nCBP-1 gene model is enlarged below. Exon and intron sequences are given in capital and small letters, respectively. Green and red boxes highlight splice motifs at the 5' and 3' end of introns, respectively. Type 1 splicing produces the predicted wild type nCBP-1 cDNA sequence. Type 2 and 3 splicing are observed in ncbp-1/2 lines #2 & #8, respectively. (b) cDNA sequences detected in clone-seq experiments. Red boxes denote INDELs resulting from both CRISPR/Cas9-mediated edits and alternative splicing. In ncbp-1/2 #2, type 2 splicing results in retention of 3' sequence from intron 1 of one ncbp-1 allele (1 of 9 clones sequenced). In ncbp-1/2 #8, an INDEL disrupting the canonical splice motif between exon 1 and intron 1 of ncbp-1 results in a type 3 splice variant (4 of 6 clones sequenced).   c. d. Figure 6. ncbp-1 ncbp-2 double mutants exhibit delayed CBSV symptom onset and reduced symptom severity (a), (b), aerial symptom incidence reported as percent of wild type, ncbp-1, ncbp-2, or ncbp-1 ncbp-2 plants bud-graft inoculated with either CBSV Naliendele or UCBSV T04 (n≥5) isolates, respectively. ncbp-1 ncbp-2 double mutant lines #2 and #8 are the product of independent transgenic events. (c), disease progression curves for previously described CBSV inoculated plants. Leaf and stem symptoms were each scored on a 0-4 scale and summed to obtain an aggregate aerial score. (d), average area under the disease progression curve (AUDPC) derived from data plotted in C. Error bars in C and D indicate standard error of the mean.

c.
Root disease scoring scale Non-conservative amino acid changes and gene deletions that abolish VPg-eIF4E binding removes above described roles, therefore conferring recessive resistance. a. b. a.
WT BS02-6 BS05-2 BS05-8 . CRISPR/Cas9 -induced mutagenesis creates out of frame alternate splice variants. Exon 1 and exon 2 splice junction of nCBP-1 (a) and nCBP-2 (b) were examined via sequence analysis of cDNA. Predicted Cas9 cut site is shown as a black arrow. STOP codon is shown as starred, red box. a. b.
a. b.