A team of researchers led by Seren S. Villwock of Cornell University, working with colleagues at the Boyce Thompson Institute and the USDA-Agricultural Research Service, set out to determine what actually drives the trade-off between total carotenoid content and dry matter in cassava storage roots. Their investigation, which combined measurements of metabolites with gene expression profiling across a diverse panel of twenty-four cassava genotypes from Africa, Latin America, and hybrid backgrounds, produced a surprising verdict: the trade-off is real and physiological, but it is not the product of the two biochemical pathways being switched on and off together by a shared transcriptional program.

The experimental design was notable for its spatial and temporal depth. Rather than treating each root as a uniform object, the researchers sampled both inner and outer layers of the storage roots, tissues that differ naturally in how much carotenoid they accumulate. They also collected samples at multiple timepoints across the growing season, allowing them to track how the relationship between carotenoids and dry matter changed as the roots developed. This within-root and across-time approach gave the team far more statistical power to separate genuine biological associations from the confounding effects of genotype, since each plant served in part as its own comparison.

The phenotypic results confirmed the trade-off with unusual clarity. Total carotenoid content was negatively associated with dry matter even after statistically accounting for genotype, and the association was particularly pronounced for cis-carotenes, compounds that sit early in the carotenoid biosynthetic pathway. Intriguingly, the strength of the negative relationship increased across developmental timepoints, meaning that as roots matured, the inverse coupling between carotenoid accumulation and starch content grew tighter. That developmental pattern is consistent with a physiological component to the trade-off: the two traits appear to compete for resources or share sensitivity to the same metabolic conditions within the growing root, rather than simply being inherited together by chance.

To probe the mechanism, the researchers turned to gene-metabolite co-expression networks, constructed using Weighted Gene Co-expression Network Analysis, a widely used computational framework that groups genes into modules based on correlated expression patterns and then asks whether those modules relate to measured traits. The team built networks from genetic, non-genetic, and overall phenotypic trait components, a strategy designed to distinguish variation that is stable across genotypes from variation that arises within a plant’s own tissues and developmental stages. The central question was whether carotenoid-related genes and starch-related genes would show significant topological overlap, meaning they would occupy the same network neighborhoods and be co-regulated at the transcriptional level.

The answer was no. Across the network analyses, carotenoid and starch-related gene sets showed no significant topological overlap, indicating that the trade-off between total carotenoids and dry matter is not driven by transcriptional co-regulation of the two main biosynthetic pathways. In other words, the genes that build carotenoids and the genes that build starch are not being controlled by the same master switches in a way that would lock the two traits together. This finding challenges a common assumption in biofortification genetics, namely that a negative correlation between two traits implies a shared regulatory architecture. The cassava data suggest instead that the correlation emerges from physiology, from the metabolic realities of the root itself, rather than from a hard-wired genetic linkage of pathway regulation.

That said, the networks were not entirely silent on connections between the two trait systems. Individual associations between carotenoid metabolites and carbohydrate-related genes did appear, hinting at possible indirect metabolic links between carotenoid accumulation and starch metabolism. These scattered connections may reflect shared precursors, competing carbon flux, or signaling interactions that operate at the metabolite level rather than at the level of coordinated transcription. The authors are careful to frame these as suggestive rather than definitive, but they provide a starting point for future experiments aimed at understanding how carbon and isoprenoid metabolism interact inside the storage root.

Perhaps the most actionable result of the study concerns what carotenoid accumulation was actually associated with. Rather than the canonical carotenoid biosynthesis genes, total carotenoid content tracked most strongly with genes involved in regulation, stress responses, redox biology, and plastid function, and the relevant gene sets were enriched for heme-binding and oxidoreductase functions. This pattern makes biological sense: carotenoids are synthesized inside plastids and are highly susceptible to oxidative degradation, so the cellular environment, including its redox state and stress status, may matter as much as the supply of biosynthetic enzymes. Genes such as those encoding cytochrome P450 enzymes, early light-induced proteins, and enzymes of the methylerythritol phosphate pathway that feeds carotenoid precursors all belong to the broader regulatory and metabolic context the study highlights.

The practical implications for breeding are significant. If the trade-off were an immutable transcriptional constraint, breeders would face a genuine dilemma: every gain in provitamin A content would come at the cost of the dry matter that determines yield, cooking quality, and farmer acceptance. But because the study finds no evidence of transcriptional co-regulation and instead points to physiological and environmental factors, the trade-off looks more like a malleable relationship than a fixed law. The researchers explicitly conclude that the trade-off is not a fixed constraint, and their identification of candidate regulatory genes, particularly those tied to redox and plastid biology, offers concrete targets for functional validation. Genes such as phytoene synthase 2 and 9-cis-epoxycarotenoid dioxygenase, which connect carotenoid metabolism to abscisic acid signaling, sit within this regulatory landscape and represent plausible levers for decoupling the two traits.

For the global effort to biofortify cassava with provitamin A, a crop on which vitamin A deficiency takes a heavy toll in many rural communities, the study offers both a caution and an encouragement. The caution is that the trade-off has a genuine physiological dimension that will not disappear simply by stacking carotenoid biosynthesis genes; managing the root’s internal metabolic environment may be just as important. The encouragement is that the mechanism is not the one breeders feared. By mapping gene-metabolite networks across genotypes, tissues, and developmental stages, the Cornell-led team has replaced a black box with a hypothesis-generating map, one that points toward stress, redox, and plastid biology as the arenas where the next generation of nutritious, high-yielding cassava varieties may be won.

Subject of Research: The physiological and genetic basis of the trade-off between carotenoid and dry matter accumulation in cassava storage roots

Article Title: Gene-metabolite networks reveal physiological trade-offs but not transcriptional co-regulation between carotenoid and dry matter accumulation in cassava (Manihot esculenta) roots

Article References: Villwock, S. S., Gómez, K. M., Fish, T., Lee, J., Doherty, A., White, A., Thannhauser, T., Gore, M. A., & Jannink, J.-L. (2026). Gene-metabolite networks reveal physiological trade-offs but not transcriptional co-regulation between carotenoid and dry matter accumulation in cassava (Manihot esculenta) roots. BMC Plant Biology. https://doi.org/10.1186/s12870-026-10012-y

Image Credits: AI Generated

DOI: 10.1186/s12870-026-10012-y

Keywords: cassava, carotenoids, dry matter, biofortification, provitamin A, gene-metabolite networks, WGCNA, cis-carotenes, starch metabolism, transcriptomics, Manihot esculenta, plant breeding

The study, led by Reisane Teles Santiago of the Federal University of Recôncavo da Bahia and Embrapa Mandioca e Fruticultura, together with colleagues Massaine Bandeira e Souza, Diego Fernando Marmolejo Cortes and Eder Jorge de Oliveira, carried out one of the first genome-wide association studies (GWAS) of herbicide tolerance in cassava. The researchers evaluated 194 genotypes from the Embrapa Cassava and Fruits breeding program, exposing them to three post-emergence herbicides with very different modes of action: mesotrione, which blocks carotenoid biosynthesis; S-metolachlor, which inhibits very-long-chain fatty acid synthesis; and chloransulam-methyl, an inhibitor of acetolactate synthase.

Timing proved to be everything in the experiment. The team scored phytotoxicity symptoms—leaf burn, chlorosis and wilting—at 3, 6, 9, 15 and 30 days after application, on a five-point scale converted to percentage injury values. Segmented regression analyses showed that the window around 9 to 11 days after application captured both the maximum symptom expression and the greatest differentiation among genotypes. That time point, dubbed PhytoX9DAA, became the phenotype of choice for the genetic mapping. Visual scores were converted to percentage injury, then distilled into genotypic best linear unbiased predictors, or BLUPs, using linear mixed models. After deregression, these continuous genetic values served as the input for association testing.

Genotyping was performed at Cornell University using a genotyping-by-sequencing approach, in which DNA is digested with the ApeKI enzyme, barcoded and sequenced on an Illumina platform. After stringent quality control—removing markers with call rates below 80 percent and minor allele frequencies under 5 percent—and imputation with Beagle, the team retained 23,047 single-nucleotide polymorphisms spread across all 18 cassava chromosomes. The panel, dominated by Brazilian landraces, showed healthy diversity: expected and observed heterozygosity were nearly identical at around 0.24, and linkage disequilibrium decayed rapidly, reaching background levels at roughly 4 to 6 megabases.

Association analyses were run with three complementary statistical models—MLM, MLMM and BLINK—each incorporating both a kinship matrix and population structure covariates derived from principal component analysis to guard against spurious associations. The results were strikingly herbicide-specific. For S-metolachlor, a single SNP on chromosome 9, chr9:1858429, emerged consistently in the BLINK model across 2023, 2024 and the combined analysis. For mesotrione, the picture was far richer: 68 significant SNP associations were detected, distributed across chromosomes 1, 4, 5, 7, 11, 14, 15 and 18, with the strongest signals clustering on chromosome 5, where four adjacent markers hinted at a haplotype block tied to tolerance. Notably, no significant associations surfaced for chloransulam-methyl, a silence the authors attribute cautiously to a possibly highly polygenic architecture, limited statistical power, or the relatively modest marker density of the GBS platform.

Mining the genomic neighborhoods of the significant markers within a 240-kilobase window, the researchers cataloged 59 candidate genes whose annotations read like a molecular toolkit for surviving chemical stress. On chromosome 7, genes annotated with antiporter activity and xenobiotic transmembrane transporter activity evoke the ABC and MATE transporter families, well-known agents of herbicide sequestration and efflux in other plants. On chromosome 1, candidates linked to RNA binding and RNA methyltransferase activity suggest post-transcriptional regulation at play, while chromosome 15 genes carry annotations for transcription factors and transmembrane transport. Key standouts include Manes.02G151900 and Manes.02G152700 on chromosome 2, tied to channel activity and signal peptide processing; Manes.09G060900 on chromosome 9, annotated with protein kinase and ATP-binding functions; and Manes.15G083800 and Manes.15G084000 on chromosome 15, linked to S-adenosylmethionine-dependent methyltransferase activity and phosphorylation.

The authors are careful to stress that these functional interpretations rest on annotation and comparative evidence from other species rather than direct experimental validation in cassava. Still, the emerging picture fits neatly with established mechanisms of metabolic herbicide resistance documented in crops and weeds alike: cytochrome P450-mediated oxidation, glutathione S-transferase conjugation, and vacuolar sequestration via ABC transporters. Comparable GWAS findings in rice, soybean, faba bean and wheat have similarly implicated transporters, protein kinases and detoxification enzymes in herbicide responses, suggesting that cassava may share evolutionarily conserved pathways for coping with xenobiotic stress.

The study also confirmed that substantial heritable variation for herbicide tolerance exists within the cassava gene pool. Variance component analysis showed a highly significant genotypic effect for phytotoxicity at 9 days after application for all three herbicides, meaning breeders have real material to select from. A significant genotype-by-year interaction was also detected across compounds, a reminder that temperature, humidity and growing conditions shape how herbicide injury unfolds in the field. Genomic inflation factors close to 1.0 for the BLINK and MLMM models, together with well-behaved quantile-quantile plots, indicate the association results are statistically sound.

For a crop whose expansion is increasingly squeezed by the rising cost of hand weeding, the practical implications are considerable. The chromosome 9 locus for S-metolachlor tolerance and the robust mesotrione signals on chromosomes 4 and 5 could, after independent validation and functional characterization, feed directly into marker-assisted selection and haplotype-based breeding. Given the polygenic nature of the trait, the authors also point to genomic selection as a complementary route, with GWAS-derived markers potentially boosting prediction accuracy when integrated into training populations. For smallholder farmers in Brazil, Nigeria and beyond, tolerant varieties could translate into cheaper, more reliable weed control and more stable yields.

Scientifically, the message is equally clear: herbicide tolerance in cassava is a quantitative, polygenic trait governed by many small-effect loci rather than a single master switch. The integration of membrane transport, cellular signaling and metabolic regulation highlighted by the candidate genes underscores the physiological resilience that has made cassava such a successful tropical staple. What this study delivers is a genomic map—a first, detailed chart of where that resilience resides in the cassava genome. Fine mapping, gene editing and validation experiments will now be needed to convert candidate loci into tools, but the first genomic dissection of herbicide tolerance in this vital crop has firmly been delivered, opening a path toward cassava varieties bred as much for chemical resilience as for the drought tolerance and low-input hardiness for which the species is already famous.

Subject of Research: Genome-wide association mapping of herbicide tolerance loci in cassava (Manihot esculenta)

Article Title: Pinpointing genomic regions conferring herbicide tolerance in cassava via genome-wide association mapping

Article References: Santiago, R. T., Souza, M. B. E., Cortes, D. F. M., & de Oliveira, E. J. (2026). Pinpointing genomic regions conferring herbicide tolerance in cassava via genome-wide association mapping. Plant Molecular Biology, 116(5), Article 95. https://doi.org/10.1007/s11103-026-01761-3

Image Credits: AI Generated

DOI: 10.1007/s11103-026-01761-3

Keywords: cassava, GWAS, herbicide tolerance, Manihot esculenta, SNP markers, mesotrione, S-metolachlor, candidate genes, marker-assisted selection, plant breeding, genomic selection, detoxification pathways