a subset of 42 most-informative SNPs, however, clearly segregated three clusters:
“Sativa”(n = 17)“Indica”(n = 9), and twofiber-type strains. These results were
confirmed with a neighbor-joining algorithm.
Lynch et al. ( 2015 ) sequenced 60 accessions using WGS, and added to this
dataset seven previous WGS reads (Van Bakel et al. 2011 , Medicinal Genomics
Corporation 2011 ). For SNP-calling they aligned sequences with the draft genome
(Van Bakel et al. 2011 ). Then they sequenced 182 accessions using GBS, with
ECoRI andMseI restriction enzymes, for SNP-calling. A subset of 195 accessions
from WGS and GBS shared 2894 SNPs for analysis.
Two algorithms were used to K-partition the 195 accessions. FLOCK recognized
K = 3 groups, and fast STRUCTURE optimized the data at K = 2. The authors
went with FLOCK, because of perceived shortcomings in fast STRUCTURE,
although these perceived differences are contentious (Anderson and Barry 2015 ).
The K = 3 groups were recognized as WLD biotypes (e.g., “Afghan Kush,”
“Chemdawg”), NLD biotypes (e.g., “Durban Poison,”“Easy Sativa”), and a
polyphyletic“hemp”group (e.g.,‘Finola,’“AC/DC,” Chinese hemp, Dagestan
plants).
Lynch and colleagues foundno evidencefor admixture (hybridization) in these
populations, based on results with the f 3 statistic and TreeMix. This seems unlikely,
given historical evidence of hybridization going back to the 1970s. TreeMix and the
f 3 statistic were developed with animal models; they may fall short with plants
having complicated histories of hybridization. TreeMix analyzes data with a
maximum of only 10 admixture (“migration”) events. The f 3 statistic must compare
three populations, so it was applied to the disputed FLOCK results.
They used vcftools to calculate FST between each FLOCK population.
FST= 0.099 between “hemp” and combined NLD + WLD, and FST= 0.036
between WLDs and NLDs. More genetic heterozygosity existed within drug-type
plants (0.31%) than withinfiber-type plants (0.22%, significantp< 0.001), which
they attributed to widespread hybridization of drug strains—an incongruous
hypothesis, given the previous paragraph.
Phylogenetic relationships between the 195 accessions were visualized in an
unrooted neighbor-joiningnetwork—a phylogenetic tree with reticulation (diver-
gence and hybridization among ancestral lineages). The network revealed aspects of
ancestry not captured by a simple bifurcating tree, such as genetic admixtures
between Chinese hemp and feral hemp plants in the USA.
Next they pooled WGS data with GBS data from Sawler et al. ( 2015 ), with 4105
SNPs in common, and generated a neighbor-joining network with 210 accessions.
These data revealed a second NLD biotype clade, consisting of Indian, Southeast
Asian, and South African populations, along with various“Haze”hybrids. This
clade may represent accessions of Indian heritage with minimal admixture from
WLD biotypes. Lastly they pooled WGS data with both GBS datasets, a total of
289 accessions,filtered for overlapping SNPs (only 45 SNPs in common—the two
GBS datasets were generated with different restriction enzymes), and used MEGA6
to generate a neighbor-joining tree.
116 J.M. McPartland