Cannabis is becoming an ever-increasing topic of discussion, so it is important that scientists and the public can discuss Cannabis in a similar manner. Currently, not only are Sativa and Indica types disputed (Emboden 1974; Hillig 2005; Russo 2007; Clarke and Merlin 2013; Clarke et al. 2015; Clarke and Merlin 2016; McPartland 2017; Piomelli and Russo 2016; Small 2015b; De Meijer and Keizer 1996), but experts also are at odds about nomenclature for Cannabis (Emboden 1974; Hillig 2005; Russo 2007; Clarke and Merlin 2013; Clarke et al. 2015; Clarke and Merlin 2016; McPartland 2017; Piomelli and Russo 2016; Small 2015b; De Meijer and Keizer 1996). We postulated that genetic profiles from samples with the same strain identifying name should have identical, or at least, highly similar genotypes no matter the source of origin. The multiple genetic analyses used here address paramount questions for the medical Cannabis community and bring empirical evidence to support claims that inconsistent products are being distributed. An important element for this study is that samples were acquired from multiple locations to maximize the potential for variation among samples. Maintenance of the genetic integrity through genotyping is possible only following evaluation of genetic consistency and continuing to overlook this aspect will promote genetic variability and phenotypic variation within Cannabis. Addressing strain variability at the molecular level is of the utmost importance while the industry is still relatively new.

Genetic analyses have consistently found genetic distinction between hemp and marijuana, but no clear distinction has been shown between the common description of Sativa and Indica types (Lynch et al. 2016; Soler et al. 2017; Sawler et al. 2015; Dufresnes et al. 2017; De Meijer and Keizer 1996). We found high support for two genetic groups in the data (Fig. 1) but no discernable distinction or pattern between the described Sativa and Indica strains. The color-coding of strains in the PCoA for all 122 samples allows for visualization of clustering among similar phenotypes by color: Sativa (red/orange), Indica (blue/purple) and Hybrid (green) type strains (Fig. 2). If genetic differentiation of the commonly perceived Sativa and Indica types previously existed, it is no longer detectable in the neutral genetic markers used here. Extensive hybridization and selection have presumably created a homogenizing effect and erased evidence of potentially divergent historical genotypes.

Wikileaf maintains that the proportions of Sativa and Indica reported for strains are largely based on genetics and lineage (Nelson 2016), although online databases do not give scientific evidence for their categorization other than parentage information from breeders and expert opinions. This has seemingly become convoluted over time (Russo 2007; Clarke and Merlin 2013; Small 2015a; Small 2016). Our results show that commonly reported levels of Sativa, Indica and Hybrid type strains are often not reflected in the average genotype. For example, two described Sativa type strains “Durban Poison” and “Sour Diesel”, have contradicting genetic assignments (Fig. 1, Table 2). This analysis indicates strains with similar reported proportions of Sativa or Indica may have differing genetic assignments. Further illustrating this point is that “Bruce Banner”, “Flo”, “Jillybean”, “Pineapple Express”, “Purple Haze”, and “Tangerine” are all reported to be 60/40 Hybrid type strains, but they clearly have differing levels of admixture both within and among these reportedly similar strains (Table 2, Fig. 1). From these results, we can conclude that reported ratios or differences between Sativa and Indica phenotypes are not discernable using these genetic markers. Given the lack of genetic distinction between Indica and Sativa types, it is not surprising that reported ancestry proportions are also not supported.

To accurately address reported variation within strains, samples were purchased from various locations, as a customer, with no information of strains other than publicly available online information. Evidence for genetic inconsistencies is apparent within many strains and supported by multiple genetic analyses. Soler et al. (2017) found genetic variability among seeds from the same strain supplied from a single source, indicating genotypes within strains are variable. When examining the STRUCTURE genotype assignments, it is clear that many strains contained one or more divergent samples with a difference of > 0.10 genotype assignment (e.g. “Durban Poison” – Denver 1; Figs. 1, 3a). Of the 30 strains examined, only four strains had consistent STRUCTURE genotype assignment and admixture among all samples. The number of strains with consistent STRUCTURE assignments increased to 11 and 15 when one or two samples were ignored, respectively. These results indicate that half of the included strains showed relatively stable genetic identity among most samples. Six strains had only two samples, both of which were different (e.g., “Trainwreck” and “Headband”). The remaining nine strains in the analysis had more than one divergent sample (e.g., “Sour Diesel”) or had no consistent genetic pattern among the samples within the strain (e.g., “Girl Scout Cookies”; Table 3, Figs. 1, 2, Additional file 3: Figure S2). It is noteworthy that many of the strains used here fell into a range of genetic relatedness indicative of first order siblings (see Lynch & Ritland analysis below) when samples with high genetic divergence were removed from the data set (Table 3; Figs. 3, 4). Eight of the 30 strains examined are identified as clone only (Table 2). All eight of the strains described as clone only show differentiation of at least one sample within the strain (Fig. 1). For example, one sample of “Blue Dream” is clearly differentiated from the remaining eight, and “Girl Scout Cookies” has little genetic cohesiveness among the eight samples (Figs. 1, 2). Other genetic studies have similarly found genetic inconsistencies across samples within the same strain (Lynch et al. 2016; Soler et al. 2017; Sawler et al. 2015). These results lend support to the idea that unstable genetic lines are being used to produce seed.

A pairwise genetic heat map based on Lynch & Ritland (1999) pairwise genetic relatedness (r-values) was generated to visualize genetic relatedness throughout the data set (Additional file 4: Figure S3). Values of 1.00 (or close to) are assumed to be clones or plants from self-fertilized seed. Six examples of within-strain pairwise comparison heat maps were examined to illustrate common patterns (Fig. 3). The heat map shows that many strains contain samples that are first order relatives or higher (r-value > 0.49). For example, “Sour Diesel” (Fig. 3) has 12 comparisons of first order or above, and six have low/no relationship. There are also values that could be indicative of clones or plants from a stable seed source such as “Blue Dream” (Fig. 3), which has 10 nearly identical comparisons (r-value 0.90–1.00), and no comparisons in “Blue Dream” have negative values. While “Blue Dream” has an initial overall r-mean indicating first order relatedness within the samples (Table 3, Fig. 4), it still contains more variation than would be expected from a clone only strain (Clone Only Strains n.d.). Other clone-only strains (Clone Only Strains n.d.) e.g. “Girl Scout Cookies” (Table 3, Fig. 3) and “Golden Goat” (Table 3, Fig. 3), have a high degree of genetic variation resulting in low overall relatedness values. Outliers were calculated and removed iteratively to demonstrate how they affected the overall r- mean within the 12 popular strains (Table 3, Fig. 4). In all cases, removing outliers increased the mean r-value, as illustrated by “Bruce Banner”, which increased substantially, from 0.3 to 0.9 when samples with two outlying genotypes were removed. There are unexpected areas in the entire dataset heat map that indicate high degrees of relatedness between different strains (Additional file 4: Figure S3). For example, comparisons between “Golden Goat” and “Island Sweet Skunk” (overall r- mean 0.37) are higher than within samples of “Sour Diesel”. Interestingly, “Golden Goat” is reported to be a hybrid descendant of “Island Sweet Skunk” (Leafly 2018a; Wikileaf 2018; NCSM 2018; PotGuide.com 2018; Seedfinder 2018) which could explain the high genetic relatedness between these strains. However, most of the between strain overall r- mean are negative (e.g., “Golden Goat” to “Durban Poison” -0.03 and “Chemdawg” to “Durban Poison” -0.22; Additional file 4: Figure S3), indicative of limited recent genetic relationship.

While collecting samples from various dispensaries, it was noted that strains of “Chemdawg” had various different spellings of the strain name, as well as numbers and/or letters attached to the name. Without knowledge of the history of “Chemdawg”, the assumption was that these were local variations. These were acquired to include in the study to determine if and how these variants were related. Upon investigation of possible origins of “Chemdawg”, an interesting history was uncovered, especially in light of the results. Legend has it that someone named “Chemdog” (a person) grew the variations (“Chem Dog”, “Chem Dog D”, “Chem Dog 4”) from seeds he found in a single bag of Cannabis purchased at a Grateful Dead concert (Danko 2016). However, sampling suggests dispensaries use variations of the name, and more often the “Chemdawg” form of the name is used, albeit incorrectly (Danko 2016). The STRUCTURE analysis indicates only one “Chemdawg” individual has > 0.10 genetic divergence compared to the other six samples (Fig. 1, Additional file 3: Figure S2). Five of seven “Chemdawg” samples cluster in the PCoA (Fig. 2), and six of seven “Chemdawg” samples are first order relatives (r-value > 0.50; Table 3, Fig. 3). The history of “Chem Dog” is currently unverifiable, but the analysis supports that these variations could be from seeds of the same plant. This illustrates how Cannabis strains may have come to market in a non-traditional manner. Genetic analyses can add scientific support to the stories behind vintage strains and possibly help clarify the history of specific strains.

Genetic inconsistencies may come from both suppliers and growers of Cannabis clones and stable seed, because currently they can only assume the strains they possess are true to name. There is a chain of events from seed to sale that relies heavily on the supplier, grower, and dispensary to provide the correct product, but there is currently no reliable way to verify Cannabis strains. The possibility exists for errors in plant labeling, misplacement, misspelling (e.g. “Chem Dog” vs. “Chemdawg”), and/or relabeling along the entire chain of production. Although the expectation is that plants are labeled carefully and not re-labeled with a more desirable name for a quick sale, these misgivings must be considered. Identification by genetic markers has largely eliminated these types of mistakes in other widely cultivated crops such as grapes, olives and apples. Modern genetic applications can accurately identify varieties and can clarify ambiguity in closely related and hybrid species (Guilford et al. 1997; Hokanson et al. 1998; Sarri et al. 2006; Costantini et al. 2005; United States Department of Agriculture 2014).

Matching genotypes within the same strains were expected, but highly similar genotypes between samples of different strains could be the result of mislabeling or misidentification, especially when acquired from the same source. The pairwise genetic relatedness r-values were examined for incidence of possible mislabeling or re-labeling. There were instances in which different strains had r-values = 1.0 (Additional file 4: Figure S3), indicating clonal genetic relationships. Two samples with matching genotypes were obtained from the same location (“Larry OG” and “Tahoe OG” from San Luis Obispo 3). This could be evidence for mislabeling or misidentification because these two samples have similar names. It is unlikely that these samples from reportedly different strains have identical genotypes, and more likely that these samples were mislabeled at some point. Misspelling may also be a source of error, especially when facilities are handwriting labels. An example of possible misspelling may have occurred in the sample labeled “Chemdog 1” from Garden City 1. “Chemdawg 1”, a described strain, could have easily been misspelled, but it is unclear whether this instance is evidence for mislabeling or renaming a local variant. Inadvertent mistakes may carry through to scientific investigation where strains are spelled or labeled incorrectly. For example, Vergara et al. (2016) reports genome assemblies for “Chemdog” and “Chemdog 91” as they are reported in GenBank (GCA_001509995.1), but neither of these labels are recognized strain names. “Chemdawg” and “Chemdawg 91” are recognized strains (Leafly 2018a; Wikileaf 2018; cannabis.info 2018; NCSM 2018; PotGuide.com 2018; Seedfinder 2018), but according to the original source, the strain name “Chemdawg” is incorrect, and it should be “Chem Dog” (Danko 2016), but the name has clearly evolved among growers since it emerged in 1991 (Danko 2016). Another example that may lead to confusion is how information is reported in public databases. For example, data is available for the reported monoisolate of “Pineapple Banana Bubba Kush” in GenBank (SAMN06546749), and while “Pineapple Kush”, “Banana Kush” and “Bubba Kush” are known strains (Leafly 2018a; Wikileaf 2018; cannabis.info 2018; NCSM 2018; PotGuide.com 2018; Seedfinder 2018), the only record we found of “Pineapple Banana Bubba Kush” is in GenBank. This study has highlighted several possible sources of error and how genotyping can serve to uncover sources of variation. Although this study was unable to confirm sources of error, it is important that producers, growers and consumers are aware that there are errors and they should be documented and corrected whenever possible.