Breeding 4.0

From the 1800s, starting with the Catholic monk Gregor Mendel, people began grappling with the statistics of heritability, and breeding grew into a science. The innovations of Breeding 2.0 included controlled crossings, field trials, statistical analyses, and other methods that have since become standard tools of quantitative genetics.

James Watson and Francis Crick’s 1953 publication of the structure of DNA gave us our first glimpse under the hood of natural (and human-mediated) selection. It became possible to evaluate an organism’s genes directly rather than inferring its genetic makeup from observable characteristics. In 2007, the grapevine became the first fruit crop to have its genome sequenced (the second was papaya, in 2008). By 2024, in addition to the initial sequence of a Pinot Noir clone, about 20 grapevine sequences had been published, nearly all of them from wine grapes.

Recent decades have seen scientists unravel the genetic basis of traits such as seedlessness, berry colour, and muscat flavour, and breeders honing their selection skills with the power of molecular markers and genomic data. The integration of phenotypic (observable characteristics) and genotypic data from many individuals to inform breeders’ decisions is the basis of Breeding 3.0.

Although new technologies have transformed cultivar development, Breeding 3.0 remains similar to Breeding 1.0 and natural selection in one critical aspect: outcomes are influenced by the inherent randomness of genetic variation.

Random variation

Genetic variation is the raw material of natural selection and cultivar development. Variation can arise through multiple mechanisms, but the two discussed below, mutation and recombination, are the most relevant to breeders.

The genome of any plant consists of DNA, which in turn contains strings of nucleotides that act like letters to spell out text. There are only four kinds of nucleotides, but cells read them in groups, making DNA more akin to Morse code than the alphabet (except that DNA is always read in groups of three bases).

The grapevine genome has about 500 million base pairs and is estimated to contain 35 000-40 000 genes. Each cell has a copy that is subject to considerable wear and tear during daily use and to errors during cell division. Cells do have DNA repair mechanisms, but mistakes occur naturally. If these mistakes become long-term changes, we call them mutations.

Like most typos, most mutations are inconsequential. Very few mutations are harmful, and even fewer are beneficial. The size of the sequence change doesn’t necessarily correlate with its impact. For example, a single base change in the VvDXS1 gene is responsible for muscat flavour in grapes.

The DNA in grapevine cells is packaged into 19 pairs of chromosomes, one set inherited from each parent. Each set of chromosomes carries the same genes, but sometimes in different forms (alleles), such as the two forms of the VvDXS1 gene mentioned above.

During normal cell division (mitosis), each new cell receives all 38 chromosomes. In contrast, during division to create sex cells (meiosis), each new cell only receives one set of 19 chromosomes. Furthermore, these chromosomes represent a different mix of alleles from the plant’s original chromosomes, due to recombination (the shuffling of large sections of DNA during meiosis).

Mutation and recombination provide variation from which breeders can select, but breeders can’t control the outcomes of either. Despite advances in screening potential parents and their progeny, Breeding 3.0 still involves a degree of unpredictability.

Designer genomes

In 2024, a group of researchers from the Grape Genetics Unit at the USDA Agricultural Research Service in Geneva, New York State, edited the genome of Scarlet Royal, converting it from a neutral-tasting to a muscat table grape. The team used prime editing, a novel technique that allowed them to change a single base in the VvDXS1 gene without inducing any unintended mutations.

Gene editing’s precision could help to improve the predictability of outcomes in breeding. Currently, if breeders want to create new cultivars, they cross parents with desirable traits or genes. But there is no guarantee that the offspring will have the wished-for combination of traits, and the process is labour-intensive and time-consuming.

For example, in a breeding programme partly funded by the Italian Variety Club, researchers aimed to develop seedless table-grape cultivars with resistance to powdery and downy mildew. They made 109 crosses, hand-pollinated 1,612 bunches, obtained 32,638 seedlings, and eventually patented 10 selections.

The Italian group focused on simple traits where a few known alleles have a big impact. But most traits are complex, resulting from the interaction of many genes, each contributing a small effect. So, breeding for complex traits is slow, with incremental improvements.

Each breeding cycle may also introduce unwanted alleles, which can be challenging to breed out again.

Gene editing is one aspect of Breeding 4.0. Instead of relying on serendipity, breeders could edit a grapevine’s genome to match their target cultivar, thereby developing allele combinations more efficiently than conventional breeding allows.

Genomics to phenomics

DNA may be the blueprint of life, but it’s written in a foreign language, one that scientists are only slowly deciphering. We know that cells transcribe DNA into RNA to make proteins, that proteins drive cellular metabolism, and that cellular metabolism underlies traits. But we need to figure out the details to unlock the full potential of Breeding 4.0.

For example, if we return to muscat, scientists have established that the VvDXS1 gene encodes the first enzyme (a protein that accelerates a chemical reaction) in the pathway forming terpenes. In muscat grapes, this enzyme is more efficient, resulting in increased terpene synthesis. The terpenes are converted into aromatic compounds, so more terpenes equal more flavourful grapes.

Very rarely are the effects of alleles as straightforward as in VvDXS1. So, to understand how we get from DNA sequences to grapevine characteristics, researchers collect enormous data sets on populations, individuals, and even single cells.

At the population scale, scientists analyse the genomes of many individuals (genome-wide association studies) to zero in on areas which may contain genes of interest. In contrast, at the cellular level, they measure which genes are transcribed (transcriptomics), which proteins are made (proteomics), and which pathways are active (metabolomics) under particular conditions.

As data sets expand, artificial intelligence and machine learning become indispensable for making sense of them. For example, precision breeding directly targets alleles based on their known effects. Up to now, those effects have been determined experimentally, but models are now being developed to predict the function and expression of DNA sequences, and even of mutations in those sequences.

Whereas predictive models may still be some years off, high-throughput phenotyping (phenomics) is more immediate. For example, a Chilean team published a study last year that validated automated phenotyping of bunch architecture in table grapes, based on 3D scanning.

The increased adoption of artificial intelligence and machine learning will eventually shift our paradigm from interpreting to predicting the results of breeding interventions. By encompassing a combination of advanced techniques, including genome sequencing and editing, artificial intelligence-based analysis, and computationally planned crosses, Breeding 4.0 promises breeders unprecedented control over cultivar design.

Further reading

“Genome editing targets drought tolerance” SAFJ Feb|Mar 2026.

“Table-grape breeding in the genomic era”. SAFJ Oct|Nov 2024. Slewinski TL, S Turner-Hissong, T Paciorek, B Brower-Toland and C Shyu. 2025. “Beautiful and delicious mutants: The origins, fates, and benefits of molecular sequence variation in plant evolution and breeding”. Plant Physiology 199(1) pkiaf378.

Wallace JG, E Rodgers-Melnick and ES Buckler. 2018. “On the road to breeding 4.0: unraveling the good, the bad, and the boring of crop quantitative genomics”. Annual Review of Genetics 52(1) pp421-444.