Genetic testing shows cultivar names are unreliable

Order Durban Poison flower from three different dispensaries and, in one genetic-marker study, more than a third of the sample pairs tested came back with no measurable relationship to each other at all ^[1]. That is not one contaminated batch. It is the pattern across three independent genetic studies, run on different continents with different lab methods, on dozens of cultivar names: a name on a jar or a seed pack is a claim about the plant inside, not a genetic guarantee.

• Marker position where the samples in a panel disagree
• Marker position the samples share

Schematic: bar heights stand in for genetic-marker results. No real samples, names or data are shown.

What three genetic-marker studies actually measured

Nobody has tested every cultivar name in circulation. What exists is three independent studies, in three countries, using three different genotyping technologies, each checking whether a shared name predicts a shared genotype.

Three labs, three marker technologies, six years apart, sourcing from three different legal supply chains on two continents. That's the part that makes the agreement between them worth taking seriously: it isn't one dataset being read three ways, it's three separate experiments landing on the same practical answer.

Same name, different genotype: the core result

Schwabe and McGlaughlin bought or were given 122 cannabis flower samples from around 20 dispensaries across Colorado, California and Washington, covering 30 cultivar names, and genotyped them at 10 microsatellite markers developed from the Cannabis sativa Purple Kush genome ^[1]. Only four of those 30 names, Chemdawg (7 samples), Island Sweet Skunk (3), Larry OG (3) and Jack Flash (2), returned a consistent genotype across every sample carrying that name. The other 26 did not.

The mismatch ran in the other direction too. One sample labelled Larry OG had a genotype identical to a sample labelled Tahoe OG, both bought from the same San Luis Obispo dispensary. The authors read this as a likely case of mislabelling, not helped by how close the two names sound ^[1].

Sawler and colleagues found the same pattern with entirely different methods and a different supply chain. Working with 81 drug-type cannabis samples from licensed Canadian producers and breeders, genotyped at 14,031 SNP markers, they found 17 cases where two or more samples in the collection shared a name, then tested whether each sample sat closer, genetically, to its own namesake or to some other, differently named sample. In 6 of those 17 comparisons (35%), the sample turned out to be a closer genetic match to a stranger than to the plant sharing its own label ^[2]. Working from opposite ends of the legal cannabis industry, on different marker technology, two labs reached the same conclusion: a name is not a reliable stand-in for a genotype.

A third study, six years later and on a different continent, tested this again with far more genetic resolution. Watts and colleagues collected 297 cannabis samples through a licensed Dutch producer, a hemp company and Dutch coffee shops, and had enough DNA from 137 of them to genotype at 116,296 SNPs, roughly eight times the marker density of the Canadian study. Pairs of samples carrying the same cultivar name, OG Kush among them, were often no closer to each other genetically than either was to a sample sold under a completely unrelated name ^[3]. One partial exception is worth keeping: a small number of terpene-synthesis genes, not the genome as a whole, did track reasonably well with whether a sample was labelled indica- or sativa-leaning, with myrcene content the single best predictor of the two. A name is a workable guess at aroma. It is not proof of lineage.

One popular name, 36 comparisons: Durban Poison up close

Averages hide what actually happens when a grower buys one named cultivar from more than one source, so it helps to look at a single case in detail. Schwabe and McGlaughlin's Durban Poison samples, pooled across the study's dispensaries, produced 36 separate pairwise comparisons. Here is how those 36 pairs sorted by genetic relatedness ^[1]:

A third of the pairs were close relatives: nearly identical or first-order siblings, the kind of relatedness you would expect from cuttings off the same mother plant, or a parent and its direct offspring. Another third sat at second-order or low relatedness, plausibly the same breeding line a generation or two apart, or a cross sharing one parent. The remaining 13 pairs, more than a third of the total, showed no measurable relatedness at all: two plants, one name, no detectable family connection.

That spread, not a single bad actor, is the more honest way to read the whole body of evidence. A cultivar name usually captures something real, some of the time, for some fraction of the samples carrying it. It doesn't capture it reliably enough to build a breeding programme or a production contract on the label alone.

Why cultivar names drift: no registry, an undocumented chain of cuttings

None of the three studies could prove exactly where a mismatch enters the chain, and none of them pretend to. Schwabe and McGlaughlin list the plausible failure points: a cutting mislabelled at the moment its name changed hands, a tag swapped in transit, a name misspelled once and then copied forward for years, or a grower relabelling a plant because the new name sells better ^[1]. Any one of these, repeated across hundreds of growers passing cuttings hand to hand over decades, produces exactly the pattern the data shows: some names holding together, others drifting apart into several unrelated plants sharing one label.

The reason this happens so easily is structural, not just carelessness. Sawler and colleagues point out that, unlike many other clonally propagated crops, cannabis cultivar names carry no organised horticultural registration system behind them ^[2]: no seed lot deposited against the name, no reference genotype held anywhere a buyer, a breeder or a lab could check a sample against. Formal variety-registration systems exist for other crops, seed-propagated varieties in the United States can be registered under the Plant Variety Protection Act of 1970 with a certified reference sample on file, but cannabis's long decades as an illegal or semi-legal crop, and the patchwork of countries where growing it still is illegal (see the law section), have kept it outside that kind of system almost everywhere. Schwabe and McGlaughlin argue the industry needs the equivalent: a baseline genotype filed against a name, the way other crops already do ^[1]. What that would look like in practice, and how far today's plant-variety-rights options actually reach for cannabis, is covered in plant variety rights and cannabis cultivars.

What these studies don't show

Three studies agreeing is meaningful. It's still worth being precise about what that agreement does and doesn't cover.

The sample sizes, large enough to make the pattern statistically convincing within each study, are still modest against the many hundreds of cultivar names in commercial circulation: 30 names in the US microsatellite study, a smaller set of directly name-matched comparisons in the Canadian SNP study. A finding that a third of one cultivar's samples were unrelated doesn't mean a third of every cultivar's samples are. Some names, Chemdawg among them in the US study, held together perfectly across every sample tested ^[1].

Genetic similarity between two differently named samples is also not automatic evidence of a mistake. Golden Goat and Island Sweet Skunk samples showed real, moderate genetic relatedness in the US study, but that's consistent with Golden Goat's reported parentage as a cross involving Island Sweet Skunk, a parent-offspring relationship, not a mislabelled clone ^[1]. A naming error and an honest lineage can leave the same genetic signature; only breeding records, where they exist, tell the two apart.

These three papers are, as far as this article can confirm, still the only peer-reviewed genetic-marker studies to test cultivar-name consistency directly. This page should be revisited if that changes.

What to do with a name, at home and at scale

None of this makes a cultivar name worthless. It usually points to a real lineage, a real breeder, a real set of traits somebody once selected for. It just isn't proof, and the studies above show how often the gap between name and genotype shows up in practice.

Two practical rules follow directly from the numbers above. Don't pay a premium for a cultivar purely because the name is well known: fame is not evidence of genetic consistency, and by the studies above it often isn't even evidence of one consistent genotype behind the name. Do reach for genetic or lab testing when consistency actually matters for the job, a breeding programme, a production line that has to repeat, a claim made to a buyer, rather than trusting the label on its own. What a genetic or sex-testing service can and can't tell you, and how to use that lab data in your own selection work, are both covered elsewhere in this cluster. When you're buying from a breeder's listing rather than a tested clone, reading the listing critically is the cheaper first check.

The record nobody else is keeping for you

No supplier, dispensary or seed bank is obliged to hold a reference genotype against the name on their label, and outside a handful of licensed producers running their own clean-stock programmes, none of them do. If a specific plant's identity needs to survive being passed to another grower, a new facility or the next generation of mother stock, the only record that reliably follows it is the one kept directly: where the cutting came from, when, from whom, and what was tested. A name travels for free. A genotype only travels with the paperwork behind it.