The starchy root crop cassava (Manihot esculenta Crantz, also known as tapioca, yuca, or manioc) is an important staple for over a billion people in tropical and subtropical regions of the world, including roughly 40% of Africans (Nweke, 2004; Lebot, 2019). It is an excellent food security crop due to its tolerance for drought and marginal soils, and because its tuberous roots can remain in the ground until needed (Howeler et al., 2013). A major challenge, however, is the presence of toxic cyanogenic compounds (e.g., cyanogenic glucosides) in cassava, which must be removed by post-harvest processing to prevent cyanide exposure and illness. Cassava root processing can be laborious, results in nutrient loss, and in Africa falls disproportionately on women and girls (Chiwona‐Karltun et al., 1998; Curran et al., 2009; Maziya-Dixon et al., 2009; Montagnac et al., 2009; Boakye Peprah et al., 2020). Troublingly, cyanogen levels in cassava increase under drought stress (El-Sharkawy, 1993; Okogbenin et al., 2003; Vandegeer et al., 2013; Brown et al., 2016). As drought frequency, duration, and severity are projected to increase due to climate change (Ayugi et al., 2022), cassava consumers’ risk of cyanide exposure may increase as well.

Following cellular disruption (e.g., during ingestion), cyanogenic glucosides are broken down to release the toxin cyanide. Distributed throughout the body via the bloodstream, cyanide halts mitochondrial electron transport, thereby preventing cells from using oxygen to produce energy and causing cell death (Dobbs, 2009). The central nervous system is particularly impacted by this toxin due to its substantial oxygen demand. The risks of insufficient cassava processing include acute cyanide poisoning which can be fatal. Chronic cyanide exposure from dietary intake induces the paralytic disease konzo, is associated with neurodevelopmental deficits, and exacerbates tropical ataxic neuropathy and goiter (Nhassico et al., 2008; Nzwalo and Cliff, 2011; Tshala-Katumbay et al., 2016; Kashala-Abotnes et al., 2018). Sulfur-containing amino acids are required to detoxify cyanide in the body; thus, those with a protein-poor diet heavily reliant on cassava are particularly at risk for adverse effects from cyanide exposure (Nzwalo and Cliff, 2011). Konzo is more likely to occur in women of childbearing age and children (Baguma et al., 2021).

Processing to remove cyanogenic content from tuberous roots can be achieved by chipping and air drying, grinding, mashing and steeping, and/or fermentation. All require 24 hours to several days to complete. Though premature consumption exposes consumers to risk, shortcuts are sometimes taken during processing, especially when food is in short supply (Banea et al., 1992; Essers et al., 1992; McKey et al., 2010; Fitzpatrick et al., 2021). Processing approaches vary by region and specific cultivated variety (cultivar) used. There are cultural preferences for growing high cyanogenic (known as “bitter”) cultivars in some contexts, for example to deter theft (Chiwona‐Karltun et al., 1998). A mismatch between expected and actual cyanide levels (due to use of a different cultivar or environmental factors) may render the usual processing insufficient. Industrial scale processing of cassava poses risks to the environment and to workers through cyanide release into wastewater and the air, respectively (Adewoye et al., 2005; Ehiagbonare et al., 2009; Dhas et al., 2011). Cyanide levels above WHO recommendations have been found in commercial cassava products as well as household flour (Burns et al., 2012; Kashala-Abotnes et al., 2018).

The biosynthetic pathway for cyanogenic glucosides requires cytochrome P450 (CYP) enzymes of the CYP79 family (Luck et al., 2016). In cassava, the enzymes CYP79D1 and CYP79D2 catalyze the first, limiting step of cyanogen biosynthesis (Andersen et al., 2000) ( Figure 1A ). The genes CYP79D1 and CYP79D2 are paralogous, having arisen through the whole-genome duplication found in this lineage (Bredeson et al., 2016). Cassava’s principal cyanogens, linamarin and lotaustralin, derived from valine and isoleucine, respectively, are synthesized in the canopy and transported to the storage roots (Nartey, 1968; Jørgensen et al., 2005). Linamarin accounts for greater than 90% of cassava cyanogens (Nartey, 1968).

Cyanogens play multiple roles in plants including defense and metabolism. Although cyanogens can deter herbivores (Bernays et al., 1977; Rajamma and Premkumar, 1994; Gleadow and Møller, 2014), this is not the case for cassava in all contexts or against all herbivores, possibly due to coevolution (Riis et al., 2003; Pinto-Zevallos et al., 2016). For example, the whitefly Bemisia tabaci detoxifies cyanogenic glucosides by enzymatic conversion to inert derivatives (Easson et al., 2021). It has been proposed that cyanogens shuttle reduced nitrogen to cassava roots for protein synthesis; increased nitrate reductase activity in roots, however, may compensate for reduced cyanogen availability (Siritunga and Sayre, 2004; Jørgensen et al., 2005; Narayanan et al., 2011; Zidenga et al., 2017). Cyanogens are also hypothesized to play a role in initiating postharvest physiological deterioration of the roots by triggering reactive oxygen species production (Zidenga et al., 2012). Modulation of cyanide levels may, therefore, bolster the longevity of harvested roots. Generation of acyanogenic cassava will facilitate further investigation of cyanogenic potential in these roles.

Cyanogen production varies naturally among cultivars (Whankaew et al., 2011; Ogbonna et al., 2021; Ospina et al., 2021). RNAi knockdown of the CYP79D genes reduced cyanogen levels in cassava; knockdown plants displayed wildtype morphology in soil (Siritunga and Sayre, 2003; Jørgensen et al., 2005; Piero, 2015). These observations indicate that cyanogen levels can be modulated without disrupting other desirable plant properties.

Here, we show that cassava cyanogenesis can be prevented via genome editing. We used CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats)-Cas9 (CRISPR-associated protein 9) mutagenesis to knock out the CYP79D genes in the model variety 60444; the popular West African landrace TME 419; and the improved variety TMS 91/02324, which retains robust resistance to cassava mosaic disease following regeneration through somatic embryogenesis (Chauhan et al., 2018). Agrobacterium-mediated CRISPR-Cas9 editing is efficient in cassava (Odipio et al., 2017; Bull et al., 2018; Hummel et al., 2018; Gomez et al., 2019; Veley et al., 2021). In contrast to RNAi knockdown, our targeted genome editing approach provides a precise, complete, and permanent loss of function, not requiring the ongoing expression of a transgene. The efficiency and precision of CRISPR-Cas9 editing are advantageous in this vegetatively propagated crop for which conventional breeding is laborious. We find that dual knockouts eliminate cyanogenic potential in all three cassava accessions. Single gene knockout lines reveal differential contribution of the two CYP79D genes to cassava cyanogenesis. The knockout lines described here facilitate further research into the role of cyanogens in cassava, and chart a course toward the development of acyanogenic planting materials.

We disabled CYP79D1 and CYP79D2 using CRISPR-Cas9 constructs with guide RNAs (gRNAs) targeting the two genes, both singly and in combination ( Table 1 ; Figure 1B ; Supplementary Figure 1 , Methods). For each gene, we selected two gRNAs with minimal off-target potential that were ~500 bp apart. We assembled these gRNAs into CRISPR constructs, and confirmed construct functionality in planta by adapting a geminivirus system in the surrogate model Nicotiana benthamiana (Baltes et al., 2014) ( Supplementary Note 1 ; Supplementary Figure 2 , Methods). These constructs were transformed using Agrobacterium into friable embryogenic calli (FEC) from the three cassava accessions. For each construct-accession pair, we recovered multiple independent T0 transgenic plant lines and characterized the induced mutations using Sanger sequencing ( Figures 1C, D ; Supplementary Figures 3 , 4 ; Supplementary Data File 1 ). We found mutagenesis of targets 1A and 2B occurred at a higher frequency than of 1B and 2A, which may be due to differences in the gRNA sequences and their respective binding efficiencies. Rarely was the region between the two target sites within a gene deleted. This result was unexpected since this excision, by all CRISPR constructs, was easily detectable in the N. benthamiana surrogate assay. Simultaneous cleavage of the target sites and excision may have occurred more frequently in this assay due to the great number of target DNA copies delivered into the surrogate plant.

We obtained four classes of CRISPR-induced mutations for each of the target loci: bi-allelic (carrying two different mutations, one for each copy of the targeted gene); homozygous (having two identical mutations of their alleles); heterozygous (carrying one mutagenized allele and one wildtype allele); and complex (carrying more than two sequence patterns, indicating genetic mosaicism or chimerism; Frank and Chitwood, 2016). For further analysis we selected mutant lines showing bi-allelic or homozygous frameshift mutations leading to premature stop codons ( Supplementary Data File 2 ), and confirmed their genotypes using Illumina amplicon sequencing (Methods). This amplicon sequencing revealed one putative dual knockout mutant as complex, bearing some wildtype alleles as well ( Supplementary Figure 5 ; Supplementary Note 2 ). This result highlights the importance of thorough sequence analysis. This line was excluded from further analysis. We found no evidence of off-target mutagenesis in 60444-derived edited lines based on the sequencing of candidate off-target sites for our gRNAs ( Supplementary Tables 1 , 2 ; Methods). Furthermore, cDNA of CYP79D1 and CYP79D2 transcripts in 60444-derived lines confirmed the expected sequences.

To test the impact of CYP79D edits on cyanogen levels, we measured linamarin and lotaustralin in leaves of edited 60444 and TME 419 in vitro plantlets using liquid chromatography-mass spectrometry (LC-MS). Linamarin was not detected in dual knockout lines ( Supplementary Note 2 ; Supplementary Figures 6 , 7 ). We also measured cyanide levels in leaves and tuberous roots of adult wildtype and mutant 60444, TME 419, and TMS 91/02324 plants, using a picrate assay (Bradbury et al., 1999). Assays were performed on greenhouse grown synchronous cohorts of plants 6–11 months after transfer to soil. Edited plants were morphologically indistinguishable from wildtypes ( Supplementary Figure 8 ). Up to nine root samples from at least three plants per line were analyzed. As observed in in vitro plantlets, dual knockout lines showed no cyanogenic potential, and CYP79D2 knockouts showed a more drastic reduction in cyanogenic potential relative to wildtype than did CYP79D1 knockouts ( Figure 2 ; Supplementary Figures 9 , 10 ). As cyanogens have been implicated in nitrogen storage and transport in cassava, we tested the ability of our acyanogenic plants to grow in nitrogen limited media. In this context, dual knockout plantlets displayed a morphology typical of, and indistinguishable from, wildtypes ( Supplementary Figure 11 ; Supplementary Table 3 ).

We found significant differences in cyanide content between edited and unedited lines despite the well-known variability of cyanide levels between roots of the same plant and plants of the same cultivar (Cooke et al., 1978). To account for observed variability we conducted pairwise comparisons of cyanide levels using Wilcoxon Rank Sum tests ( Figure 2 ; Supplementary Figure 9 ). In dual knockout lines generated from the three accessions, the zero (or very near zero) assayed cyanide levels were distinct from those of wildtype and single knockout lines. In each of the three accessions, one CYP79D1 knockout line had cyanide levels distinct from wildtype and the other did not. This is consistent with our observation that knocking out CYP79D1 alone does not reliably reduce cyanogenic potential below wildtype levels. All CYP79D2 knockout lines had cyanide levels distinct from corresponding wildtype, CYP79D1 knockouts, and dual knockouts. This is consistent with our assessment that knocking out CYP79D2 alone provides a near complete, but not total, reduction in cyanogenic potential.

To identify cultivars that could be prospective targets for CYP79D gene modification, we measured leaf and storage root cyanide content in accessions that have established transformation protocols: 60444, TME 419, TMS 91/02324, TMS 98/0505, TME 3, MCol 22, and MCol 2215 ( Figure 3 ) (Li et al., 1996; Siritunga and Sayre, 2003; Taylor et al., 2012; Zainuddin et al., 2012; Chauhan et al., 2018). Genotype and environment both impact cassava root cyanide levels (Ogbonna et al., 2021). In our controlled environment, the ranges of root cyanide values largely overlapped; 60444 and TME 419 were significantly lower, but all others were not distinguishable from each other. Relative cyanide content in leaves was not predictive of relative cyanide content in roots across accessions. This result is consistent with previous work that showed weak correlation between leaf and storage root cyanide content, using a different cyanide assay method and field-grown landraces (Ospina et al., 2021).

This study marks the first report of engineering acyanogenic plants via the CRISPR-Cas system. Genome editing is a powerful and heritable method to disable genes of interest for functional assessment and crop improvement. Here, we targeted cyanogenesis genes CYP79D1 and CYP79D2 to achieve reduction in cassava’s cyanogenic potential. We demonstrated the elimination of cyanogenic potential in eight dual knockout cassava lines. We discovered differing contributions to cyanogenesis by the two CYP79D genes, and observed that loss of CYP79D2 alone is sufficient for dramatic and stable reduction in cyanide levels. These results were consistent across three different cassava accessions, in in vitro plantlets as well as adult plant leaves and storage roots, and via measurement of cyanogenic glucosides or evolved cyanide, respectively. We also assayed cyanide levels among a group of cassava lines that have elucidated transformation protocols, and are thus poised for genome engineering. The observed incongruity between relative root and leaf cyanide content indicates that these lines differ in terms of cyanogen biosynthesis, transport and/or metabolism. Understanding the mechanism and regulation of these differences may be useful for future modulation of this pathway.

The paralogous CYP79D genes were duplicated during the ancient paleotetraploidy in the cassava lineage (Bredeson et al., 2016). Gene duplication provides an evolutionary substrate for functional diversification (Otto and Yong, 2002), since initially redundant genes can then accumulate novel and/or complementary mutations, including variation in substrate specificity and gene expression, and may also alter the regulation of biochemical networks. Our data indicate that CYP79D2 is likely responsible for a greater proportion of CYP79D enzymatic function than is CYP79D1. Differences in gene expression and/or protein sequence may explain this disparity: CYP79D2 shows higher transcriptional activity than CYP79D1 (Wilson et al., 2017) ( Supplementary Figure 12 ), and lies in the cyanide biosynthetic gene cluster, whereas CYP79D1 does not (Takos et al., 2011; Bredeson et al., 2021; Ogbonna et al., 2021). The 1000-bp regions upstream of the CYP79D1 and CYP79D2 transcript sequences have 50.17% identity; the amino acid sequences of CYP79D1 and CYP79D2 have 85.74% identity (Bredeson et al., 2021), with multiple mismatches in the transmembrane and P450 domains. There may be differences in spatial and/or temporal expression between these genes. It is also possible that genetic and metabolic feedback loops are influencing cyanogenic output of the intact CYP79D gene in single knockout lines. Thus, regulation and expression of the CYP79D and associated genes merit further study. CYP79D2 knockouts showed a three- to ten-fold reduction of cyanide content, when comparing the mean values of these edited lines to mean values from the corresponding wildtypes. Hence, knockout of CYP79D2 alone provides a straightforward mechanism for generating stably low-cyanide plants, if so desired.

A recent article reported a significant reduction of cyanide in the leaves of cassava plants via CRISPR-Cas9 targeting of the CYP79D1 gene alone (Juma et al., 2022). The target sequence used, however, also matches a site in the CYP79D2 gene (Bredeson et al., 2021). Thus, perhaps the degree of cyanide reduction observed in the edited lines resulted from mutations in CYP79D2, in addition to the reported mutations in CYP79D1.

Field testing of our edited lines with their corresponding wildtypes will allow well-controlled interrogation of the roles of cyanogens in cassava, including herbivory defense, stress response, nitrogen metabolism, and postharvest physiological deterioration. In addition to the CYP79D genes, the CRISPR-Cas9 system can be applied to modulate cyanogenic potential through modification of other genes of interest. A recent genome-wide association study, for example, identified two proteins that regulate cassava cyanide levels in storage roots (Ogbonna et al., 2021).

For all of our mutant lines, both copies of the target gene(s) are mutated, and we demonstrated that these changes to the genomic DNA are stably inherited via clonal propagation, which is typically used in cassava cultivation. We thus anticipate that the mutant alleles would be stably inherited via sexual propagation, and if crossed with other cassava, would segregate in a typical Mendelian fashion. The potential for changing a particular trait in outbred cassava varieties, without disrupting the complement of other traits for which they are preferred, is an alluring aspect of this precision breeding method, and can play a role in the maintenance of genetic diversity across the global population of cassava cultivars. For example, the editing approach demonstrated here can be applied to cassava varieties popular in regions of Africa projected to experience increased drought as a result of climate change, and hence higher cyanide risk (Nhassico et al., 2016).

With the increasing severity and frequency of drought, the ability to modulate cyanide levels in preferred cassava cultivars will increase in importance. TME 419, popular in Nigeria, is known as a low cyanide “boil and eat” cultivar. Under environmental conditions that would increase cyanogenesis, farmers and consumers using acyanogenic or reduced cyanide TME 419 would not have to alter farming and preparation practices intended for low cyanide roots. Previous work indicates that the disruption of cyanogenesis can impede cassava growth in the absence of sufficient reduced nitrogen (Siritunga and Sayre, 2004; Jørgensen et al., 2005). We observed no impact on morphology on plants in soil in glasshouse, potentially due to the provision of fertilizer containing ammoniacal nitrogen ( Supplementary Figure 8 , Methods). We also did not observe any differences between wildtype and acyanogenic plantlets grown on media with limited nitrogen ( Supplementary Figure 11 ). If the complete absence of cyanogens proves deleterious to the crop in nitrogen-limited soils in the field, releasing low-cyanide CYP79D2 knockout versions of farmer preferred cultivars may serve as a suitable approach for the reduction of cyanide risk.

Reduction of cyanogen content in cassava has the potential for broad socioeconomic benefits for cassava producers and consumers as well as positive effects on the environment. As detoxification of cassava can take days (Tewe, 1992), acyanogenic cassava can reduce processing time and labor. Women and girls, who disproportionately bear the burden of this labor, may then be at greater liberty to pursue other forms of work and education. If reduction of cyanide levels delays postharvest physiological deterioration of storage roots, the increased shelf life could economically benefit farmers and other stakeholders in the value chain (Zainuddin et al., 2018). At the industrial scale, processing of acyanogenic cassava would not release cyanide into wastewater, thereby reducing the labor and cost of wastewater treatment and/or the toxicity to local terrestrial and aquatic life (Adewoye et al., 2005; Silva et al., 2017). Moreover, acyanogenic cassava cultivars would be a boon for food safety and consumer wellness. As described above, excessive consumption of cyanide with a protein-poor diet can lead to neurological harm including decline in motor proficiency and cognitive performance, and, in severe cases, paralysis (Kashala-Abotnes et al., 2019). Acyanogenic cassava could preclude these debilitating conditions and open at-risk consumers and their would-be caretakers to other pursuits.

The data produced in this study are available in the article and Supplementary Material . The raw Illumina sequence data generated for this study were deposited in the NCBI Sequence Read Archive (SRA), under BioProject PRJNA906627. Vectors are available upon request to BS stask@berkeley.edu. Plants are available upon request to M-JC mjcho1223@berkeley.edu.

Designed the project: MG, NT, BS, M-JC, DR, JL. Molecular biology lead: MG. Gemini-vector transient assay: MG, AS. Cassava transformation: BG, RC. Genotyping of on- and off-target sites: MG, KB, SL, JM. Extractions for LC-MS: MG, BG, JL. LC-MS: AI. LC-MS plots: JL. Picrate assays: MG, KB, SL, NK, JL. Picrate assay plots and statistical analysis: KB, NK. Illumina amplicon data analysis: SW. Project leadership: JL, with co-PIs BS, M-JC, DR; Danforth Center lead NT. Wrote the paper: MG and JL; with contributions by KB, BG, AI, SL, SW, NT, M-JC; and edited by NT, M-JC, and DR. All authors contributed to the article and approved the submitted version.