To evaluate the genetic structure of commonly cultivated Cannabis, we genotyped 81 marijuana and 43 hemp samples using genotyping-by-sequencing (GBS) [5]. The marijuana samples represent a broad cross section of modern commercial strains and landraces, while the hemp samples include diverse European and Asian accessions and modern varieties. Although we sampled a diverse collection of cannabis types in our study, access to samples is complicated by the fact that marijuana is an illicit drug and there are limited repositories of hemp germplasm in international seedbanks. In total, 14,031 SNPs were identified after applying quality and missingness filters. Principal components analysis (PCA) of both marijuana and hemp (Fig 1a) revealed clear genetic structure separating marijuana and hemp along the first principal component (PC1). This distinction was further supported using the fastSTRUCTURE algorithm [6] assuming K = 2 ancestral populations (Fig 1c). PCA and fastSTRUCTURE produced highly similar results: a sample’s position along PC1 was strongly correlated with its group membership according to fastSTRUCTURE at K = 2 (r² = 0.98; p-value < 1 x 10⁻¹⁵).

We observe a putative C. indica marijuana strain from Pakistan that is genetically more similar to hemp than it is to other marijuana strains (Fig 1a). Similarly, the hemp sample CAN 37/97 clusters more closely with marijuana strains (Fig 1a). These outliers may be due to sample mix-up or their classification as hemp or marijuana may be incorrect. The sample of CAN 37/97 that we genotyped was from a Canadian hemp germplasm collection, which obtained this accession from the IPK Genebank (Gatersleben, Germany). The original source country is France but there is limited information to indicate the cultivation of CAN 37/97 as hemp. Alternatively, these samples may be true outliers and represent exceptional strains that are genetically unlike others in their group. Using the current data set, the unambiguous identification of a sample as either hemp or marijuana would be possible in the former case, but not in the latter. In any case, we find that the primary axis of genetic variation in Cannabis differentiates hemp from marijuana.

These results significantly expand our understanding of the evolution of marijuana and hemp lineages in Cannabis. Previous analyses have shown that marijuana and hemp differ in their capacity for cannabinoid biosynthesis, with marijuana possessing the B _T allele coding for tetrahydrocannabinolic acid synthase and hemp typically possessing the B _D allele for cannabidiolic acid synthase [7]. As well, transcriptome analysis of female flowers showed that cannabinoid pathway genes are significantly upregulated in marijuana compared to hemp, as expected from the very high THC levels in the former compared to the latter [3]. Our results indicate that the genetic differences between the two are distributed across the genome and are not restricted to loci involved in cannabinoid production. In addition, we find that levels of heterozygosity are higher in hemp than in marijuana (Fig 1b; Mann-Whitney U-test, p-value = 8.64 x 10⁻¹⁴), which suggests that hemp cultivars are derived from a broader genetic base than that of marijuana strains and/or that breeding among close relatives is more common in marijuana than in hemp.

The difference between marijuana and hemp plants has considerable legal implications in many countries, and to date forensic applications have largely focused on determining whether a plant should be classified as drug or non-drug [8]. EU and Canadian regulations only permit hemp cultivars containing less than 0.3% THC to be grown. While hemp and marijuana appear relatively well separated along PC1 (Fig 1a), we found no SNPs with fixed differences between these two groups: the highest F_ST value between hemp and marijuana among all 14,031 SNPs was 0.87 for a SNP with an allele frequency of 0.82 in hemp and 0 in marijuana (S1 Table). The average F_ST between hemp and marijuana is 0.156 (S1 Fig), which is similar to the degree of genetic differentiation in humans between Europeans and East Asians [9]. Thus, while cannabis breeding has resulted in a clear genetic differentiation according to use, hemp and marijuana still largely share a common pool of genetic variation.

Although the taxonomic separation of the putative taxa C. sativa and C. indica remains controversial, a vernacular taxonomy that distinguishes between “Sativa” and “Indica” strains is widespread in the marijuana community. Sativa-type plants, tall with narrow leaves, are widely believed to produce marijuana with a stimulating, cerebral psychoactive effect while Indica-type plants, short with wide leaves, are reported to produce marijuana that is sedative and relaxing. We find that the genetic structure of marijuana is in partial agreement with strain-specific ancestry estimates obtained from various online sources (Fig 2, S2 Table). We observe a moderate correlation between the positions of marijuana strains along the first principal component (PC1) of Fig 2a and reported estimates of C. sativa ancestry (Fig 2c)(r² = 0.36; p-value = 2.62 x 10⁻⁹). This relationship is also observed for the second principal component (PC2) of Fig 1a (r² = 0.34; p-value = 1.21 x 10⁻⁸). This observation suggests that C. sativa and C. indica may represent distinguishable pools of genetic diversity [1] but that breeding has resulted in considerable admixture between the two. While there appears to be a genetic basis for the reported ancestry of many marijuana strains, in some cases the assignment of ancestry strongly disagrees with our genotype data. For example we found that Jamaican Lambs Bread (100% reported C. sativa) was nearly identical (IBS = 0.98) to a reported 100% C. indica strain from Afghanistan. Sample mix-up cannot be excluded as a potential reason for these discrepancies, but a similar level of misclassification was found in strains obtained from Dutch coffee shops based on chemical composition [10]. The inaccuracy of reported ancestry in marijuana likely stems from the predominantly clandestine nature of Cannabis growing and breeding over the past century. Recognizing this, marijuana strains sold for medical use are often referred to as Sativa or Indica “dominant” to describe their morphological characteristics and therapeutic effects [10]. Our results suggest that the reported ancestry of some of the most common marijuana strains only partially captures their true ancestry.

As a wind-pollinated dioecious plant (though monoecious forms exist), Cannabis is highly heterozygous and many marijuana strains are clonally propagated in order to retain their genetic identity. In contrast to other clonally propagated crops like apples and grapes, however, strain names are assigned to marijuana plants even if grown from seed. Thus, a marijuana strain name does not necessarily represent a genetically unique variety. To investigate the genetic identity of named marijuana strains at the genetic level, we compared samples with identical names to each other and to all other genotyped samples. We found that in 6 of 17 comparisons (35%), samples were more genetically similar to samples with different names than to samples with identical names. We conclude that the genetic identity of a marijuana strain cannot be reliably inferred by its name or by its reported ancestry.

Hemp is consistently classified as C. sativa in previously published literature [11, 12], and the prevailing assumption has been that varieties used for fibre and seed production are derived from C. sativa [1]. Our results are incompatible with this proposition: a marijuana strain’s genetic distance to hemp is negatively correlated with its reported C. sativa ancestry (r² = 0.17; p-value = 0.0002) and is negatively correlated with its position along PC1 of Fig 2a (r² = 0.43; p-value = 4.63 x 10⁻¹¹). Moreover, we find that F_ST is higher between hemp and marijuana strains with 100% reported C. sativa ancestry (F_ST = 0.161) than between hemp and strains with 100% reported C. indica ancestry (F_ST = 0.136). Hillig (2005) challenged the C. sativa origin of hemp, and noted that fibre/seed strains from Asia cluster with C. indica [1]. Similarly, a recent study using Random Amplified Polymorphic DNA (RAPD) markers found that C. indica clustered more closely with hemp than with C. sativa or hybrid marijuana strains [8]. Consistent with these studies, our findings suggest that hemp has a greater proportion of alleles in common with C. indica than with C. sativa.

There is a paucity of public repositories for hemp germplasm and only a patchwork of private collections of marijuana strains worldwide. These factors, and the low viability of Cannabis seed after prolonged storage, cause concern that the genomic variation that we describe here is in danger of being lost. After decades of restrictive regulations and the replacement of hemp fibre with synthetic products, Cannabis cultivation is now undergoing a resurgence in many parts of the world. For example, Canadian hemp acreage reached 27,000 ha in 2013 [13] and support for hemp research was included in the recent US Farm Bill [14]. The present study provides clarity on the genetic structure of marijuana and hemp and highlights the severe challenges associated with marijuana germplasm curation due to its clandestine past. Achieving a practical, accurate and reliable classification system for Cannabis, including a variety registration system for marijuana-type plants, will require significant scientific investment and a legal framework that accepts both licit and illicit forms of this plant. Such a system is essential in order to realize the enormous potential of Cannabis as a multi-use crop (hemp) and as a medicinal plant (marijuana).