US hemp breeding programs have been legislatively impeded for nearly a century due to federal prohibition under the 1937 Marihuana Tax Act and the Controlled Substances Act of 1970. Federal drug law policies prohibited the cultivation and use of Cannabis in any form, crippling potential industrial and medicinal use. Interests in industrial hemp production increased with the passage and expansion of the 2014¹² and 2018 Farm Bill¹³. Still, the limited availability of complaint and diverse Cannabis germplasm represents a major bottleneck for breeding and research efforts.

Although the last legal US hemp crops were planted in 1958^37,38, remnants of domesticated industrial hemp have escaped to form naturalized, thriving feral populations across regions of the US where it had been previously cultivated. Additional dispersal of seed by birds, animals, and anthropogenic activities might have resulted in further establishment in regions where hemp had not previously been cultivated. Sites suggestive of such secondary range expansion include roadsides, railroad rights-of-way, livestock feedlots, and agricultural boundaries especially as associated with trees and utility lines. We relied on university collaborators, farmers, and citizen scientists to identify and/or collect feral Cannabis plants across the US based on sites recorded in iNaturalist or received by word of mouth. Whereas site identification was essentially straightforward, collection of samples was occasionally challenging because of residual stigma associated with cannabis. Additionally, occurrences within interstate highway corridors, on private property, and accompanied by dense vegetation made collection difficult. Wider identification of feral Cannabis occurrence and collection is particularly needed in the West, South, and Northeast geographical regions of the US. Meeting this need will require effective communication about the potential economic benefits of adopting industrial hemp production among existing agricultural crops.

Five subpopulations were identified among US feral Cannabis collections. Generally, samples clustered according to state of origin showing that geographical location explains genetic variation. Most notable were two clusters, one containing accessions primarily from Indiana and the other containing accessions from New York. The remaining three clusters were less variable and contained samples from states along the Mississippi River basin and within West North Central states.

Pairwise comparison of fixation indices (F_ST) between subpopulations found genetic distance to be in accordance with geographical location with the most divergent subpopulations from Indiana (group 5) and New York (group 3). The F_ST values from this study are higher than those previously recorded in numerous studies focused on drug-type Cannabis^26,27,29,34, which is likely attributed the broader genetic background of hemp as compared to drug-type Cannabis. The least divergent subpopulations (F_ST = 0.07 - 0.09) may have resulted from a single introgression to the Upper Midwest during the WWII “hemp for victory” era followed by dispersal by humans or animals across the sampled regions. Interestingly, many of the collection sites in these regions were either along railroad rights-of-way or livestock facilities. High F_ST between the Indiana and New York groups indicate separate introgressions. Indeed, hemp arrived at different geographical regions at distinct periods throughout history with samples arriving in the Northeast as early as the 16^th century, Kentucky in the 18^th century and Wisconsin during the 20^th century^19,39,40. The three primary subpopulations identified in the current study may have originated from distinct introgression and feralization events and therefore harbor unique alleles based on origin.

High pairwise F_ST scores were found for all comparisons indicating regions of the genome that may be under selective pressures across different regions. The highest F_ST values were associated with genomic regions on chromosomes 4 and X (Figure 4; Supplementary Figure 4). Previous work identifies similar peaks and hypothesizes that the regions may be the result of chromosomal inversions maintained at high frequency in the feral US populations³⁴. The Cannabis pangenome has confirmed a large-scale inversion on chromosome 4 in a feral collection from Boone County, Iowa as compared to high cannabinoid cultivars⁴¹. We find this same elevated F_ST primarily in samples from the West North Central-a and Indiana groups suggesting it is segregating within US feral populations. Low global F_ST observed between West North Central-a and -b groups, indicates these accessions are mostly differentiated by this inversion. Analysis of the 15 Mb region surrounding the chromosome 4 inversion explains 65% of the variation among samples in the PCA (Supplementary Figure 4). Because this inversion is found in some frequency across all subpopulations, it is likely this inversion occurred prior to the introgression of hemp to the United States. Fixation in several populations suggests that it may confer a selective advantage in certain environments.

The expected heterozygosity level (0.20) for the accessions in our study was similar to the observed heterozygosity (0.20) indicating the populations are not experiencing significant selective forces such as inbreeding or genetic draft. Our observed heterozygosity is more similar to what has been previously described in hemp (0.27-0.16) as compared to drug-type cannabis (0.12-0.14)^26,28. A lower observed heterozygosity was observed in the New York subpopulation which may be attributed to its small population size and geographic isolation. The negative value of inbreeding coefficients recorded in the MS-River, WNC-a and WNC-b subpopulations, imply increased heterozygosity and higher genetic diversity when compared to the Indiana and New York feral values reflecting random mating and genetic stability among the populations. Polymorphism Information Content (PIC) is a measurement of how well a genetic marker can distinguish different individuals in a population. It is a key indicator of marker quality in genetic studies and is used to predict the likelihood of a marker’s heterozygous genotype being passed on to offspring⁴². The PIC marker values per population ranged between 0.15 – 0.30, indicating they are very informative at discriminatory analysis between feral populations.

High genetic variability from AMOVA as observed within accessions followed by between accessions within groups and lowest between groups suggests potential for trait discovery in breeding programs. The moderate differentiation between groups reiterates the high heterozygosity often recorded in Cannabis associated with wind pollination. Within accession AMOVA values among US feral collections was higher than that of the Spanish and Moroccan Cannabis germplasm studies^34,43. Discrepancies in variation contributed by these studies are consistent with other studies and may be due to differences in sample size, sample composition and types of markers used.

Linkage disequilibrium decayed rapidly (6.7kb) in the feral germplasm. This was similar to a recent report using a mixture of hemp and drug-type ferals (LD = 6.0 kb)²⁹ and faster than in a recent study of drug-type Cannabis averaging between 22.6 to 89.0 kb across chromosomes⁴⁴. Historic breeding efforts during the 18^th and 19^th centuries that entailed intentional hybridization as well as wind dispersal of hemp pollen over great distances may account for the plunge in LD decay of feral Cannabis plants. This finding will be important for finding marker-trait associations and developing genotyping resources to adequately cover haplotype blocks.

Total cannabinoid content in Midwestern feral Cannabis was rather low compared to drug cultivars although somewhat higher than hemp cultivars⁴⁵ with population averages (±SE) ranging from 0.62 (0.05) to 4.04 (0.26) (Supplementary Table 8). Most populations contained at least some individual plants with THC content above the US statutory threshold of 0.3%. Such plants were either of intermediate (C^FC^X) or THC-type (C^XC^X) genotypes as indicated by the presence of non-functional CBDAS alleles at low frequency. Approximately one third of 53 populations with at least 10 resolved accession genotypes were fixed for functional CBDAS, with mixed populations varying in non-functional allele frequency from >0.1 to 0.6. These observations are consistent with evidence from Minnesota³⁵ suggesting that midwestern “feral hemp” does not neatly fit the current statutory definition of hemp but is more accurately described as feral Cannabis. Genotype frequencies in mixed populations were generally skewed in favor of CBD-type, with a lower frequency of intermediates and the occasional THC-type plant. There is little reason to expect that the original Cannabis introductions to the region in the 19th century for fiber production were pure hemp, considering that the chemical nature of intoxicating versus non-intoxicating Cannabis (THC versus CBD) was not discovered until decades later^46,47. Populations at or near Hardy-Weinberg equilibrium with relatively low yet varying non-functional CBDAS allele frequencies might be interpreted as subject to local genetic drift with little or no selection on cannabinoid content and/or little immigration of THC-type genetics from elsewhere. However, several populations deviated from this pattern with a high proportion of intermediate-type plants or a non-equilibrium excess of THC-type plants, which might be interpreted as the result of introgression from more recent drug-type cultivation. Genotyping of historical specimens from herbarium collections might shed light on how and whether cannabinoid genetics has changed over time⁴⁸.

CBDAS genotypes as measured by the PACE assay⁴⁹ accurately predicted each THC:CBD ratio class in most cases (87%). In three cases, cannabinoid content was so close to the limit of detection by gas chromatography (0.01%) that the THC:CBD ratio lacked meaning, and we do not consider them to challenge the accuracy of the genetic assay. However, 33 of 465 accessions (13%) had CBDAS genotypes that did not correlate with the THC:CBD ratio even after independent, repeated measures. Such plants merit further investigation of the possibility that allelic variation at the primer binding sites could skew the assay. This seems more likely than a departure from the simple genetic mechanism of codominant inheritance¹¹ as others have suggested⁵⁰. Another limitation of the PACE assay that we observed when applied to feral Cannabis was a relatively high failure rate (20%) in which some accessions failed to amplify (18%) while others (2%) yielded ambiguous fluorescence patterns where it was not possible to distinguish between the intermediate genotype and a homozygous state.

Feral accessions are important genetic reservoirs of useful alleles and superior traits for the improvement of Cannabis plants for fiber, grain or cannabinoids^51,52. The persistence of these feral Cannabis populations in selected habitations and even spread to new localities after decades of naturalization attest to their resilience and importance as new tools for breeding improved cultivars with increased adaptation to biotic and abiotic stress factors. The conservation of these Cannabis collections in the USDA-ARS Hemp Germplasm Repository will ensure the long-term security of these genetic resources against any future political or societal pressures and their genetic integrity will be continually maintained through thorough regeneration practices. Conservation of these diverse feral Cannabis materials in the genebank promises in contrast to historic absence under decades-long prohibition, to provide access to diverse and compliant Cannabis germplasm.