The majority of studies on the genetic stability of cannabis in the context of micropropagation have not conducted long-term trials following multiple subcultures (Monthony et al. 2021), a gap we aimed to address in this study. Indeed, industrial scale micropropagation by legal producers is likely to imply long multiplication cycles, which we emulated by maintaining multiple cannabis cultivars in the vegetative phase for 60 weeks with 3-weeks-subclutures, allowing plantlets to grow multiple new shoots without outgrowing their vessels. Additionally, most studies assessing the genetic stability of in vitro cannabis have relied on multilocus marker-based genetic sequencing methods, which only detect mutations at specific sites and are therefore less representative of the genetic modifications throughout the whole genome (Lata et al. 2011; Ioannidis et al. 2022; Borin et al. 2025; Cull and Joly 2025). As a result, the true extent of in vitro mutation accumulation in cannabis remains unclear, despite its potential for significant impact on the desired phenotypes of regenerated plants. Therefore, we proposed a different experimental approach taking advantage of the genotyping by sequencing (GBS) technology, more specifically the 3D-GBS method to assess genome wide mutation accumulation since it allows an evenly distributed coverage of the genome compared to other marker-based methods while retaining the ultra-high-throughput genome-wide genotyping capabilities at a more affordable cost than conventional GBS or whole genome sequencing (de Ronne et al. 2023; Aina et al. 2025). Additionally, implementation of methylome sequencing via EM-seq method provided for the first time in the context of cannabis micropropagation, insights into DMPs and their long-term modifications. Among other things, we quantified mutation levels at various stages of micropropagation. Our results show that the mother plants initially had 12,943 (CPK), 12,460 (GC), and 13,145 (GEL) SNPs (Fig. 1). It is also noteworthy to mention that the sequencing reads were mapped to the CS10 v2.0 cannabis reference genome, which is inherently different from our studied cultivars and is thus responsible for part of the initial SNP counts (Grassa et al. 2021). Most interestingly, we observed a rise in mutation count between culture initiation (T = 0) and the 5th transplant (T = 5), with CPK, GC and GEL accumulating 1,587, 1,284 and 1,743 SNPs, respectively, over a four-month period. This mutation rate was unmatched throughout the duration of this experiment and was remarkably uniform across all three genotypes. In fact, only a slight increase in mutations were observed from the 5th (T = 5) to the 15th (T = 15) transplant, accumulating only 127, 176 and 111 additional mutations for CPK, GC and GEL, respectively, probably indicating that in vitro plantlets have acclimated to their environment by that point. This result is particularly important for growers considering micropropagation for industrial purposes. Indeed, our findings suggest that the 5th transplant mark could be regarded as the optimal period to assess various parameters such as plant regeneration, vigor, morphology and yield as part of a phenotyping program. First, this would allow growers to eliminate plantlets harboring deleterious mutations and could even lead to the development of new elite genotypes harboring advantageous mutations or, at least, neutral ones (non-deleterious). Second, our results indicate that the vegetative state could probably be maintained in vitro for at least 15 additional cycles (approximately one year) with minor risk to genetic stability and ultimately to flower phenotype reproducibility. Additionally, between the 15th (T = 15) and the 20th (T = 20) subculture cycles, we observed a slight decrease in the number of mutations in two out of three cultivars with CPK and GEL losing 15 and 158 SNPs, respectively, while GC barely gained 7 additional mutations. These very minor changes are likely a result of our bulk sampling method; however, they suggest that the genetic stability of in vitro plantlets can be maintained well over a year. Obviously, further research is needed to confirm these results over a longer period, but overall, our data suggest that micropropagation is a well-suited method for long-term industrial cannabis production.

In this study, we identified a substantial number of genetic variations sites during 60 weeks of in vitro cultivation. These variations were similarly distributed across the three cultivars: CPK (16,169), GC (15,472) and Gel (16,605), (Suppl. Table S1). In a similar study (Adamek et al. 2022) identified 592,000 unique variants in the top leaves of a 1.5-year-old mother plant of the cannabis cultivar Honey Banana using whole-genome sequencing (WGS). However, our results are not directly comparable to the former study since the WGS approach is expected to reveal more SNPs than GBS. In addition, their experimental design does not account for the temporal factor in the same way as ours, since their identified mutations are relative to the reference genome at a single time point, whereas we also considered de novo mutations that emerged throughout the multiple in vitro subcultures (Fig. 1).

Interestingly, our results seem to some extent contradict the genetic mosaicism hypothesis brought forward by Adamek et al. (2022), as we observed a sudden increase in mutations followed by a stationary phase from the 5th subculture onwards, suggesting that in vitro propagated plants, once they have acclimated to their new conditions, are more genetically stable than those obtained through cuttings. Otherwise, extrapolation of our results could provide an estimated number of mutations across the whole genome, which would most likely be around the same order of magnitude. Indeed, the expected fraction of the genome covered is roughly 1.2% with 3D-GBS (de Ronne et al. 2023). We measured the mean nucleotide diversity (π) of our 60-week-old samples to be around 2 × 10^–4 for all cultivars (Suppl. Table S2). This value is of the same order of magnitude as that reported by Adamek et al. (2022), who found a mean nucleotide diversity of 6 × 10^–4. The linkage disequilibrium (LD) decay curve shows a rapid decline of r² within the first few kilobases (Suppl. Fig. S10). A similar pattern of rapid LD decay was also reported by Adamek et al. 2022), whereas extremely fast LD decay as been reported in wild cannabis populations by Aina et al. (2025).

To better understand the potential biological impact of these variants, the GO enrichment terms of SNP-associated genes were determined and show a partial overlap with those found in the DMP-associated genes, pointing to the processes being most likely influenced by both somatic and epigenetic mutations, especially in molecular function categories involving ATP-driven catalytic processes, lyase activity, ion and phospholipid binding and active transmembrane transport, highlighting a potential convergence in regulatory and metabolic pathways potentially shaped by both genetic variation and epigenetic modifications in the context of cannabis micropropagation.

Although we report significant over-representation for these terms, we must keep in mind that only a minority of the identified mutations are classified as high impact (0.1%). Therefore, the GO results serve more as an approximation of genetic targets that are more susceptible to being affected by nucleotide variations than other regions. On that note, the generation of precursor metabolites is probably one of the most important terms to consider in future studies due to its significant role in cannabinoid biosynthesis. Other terms mentioned above are not to be taken lightly either, as the biological machinery allowing cannabis plants to thrive and produce high-value compounds is highly complex and involves a plethora of underlying mechanisms. Our results highlight the importance of monitoring both genetic and epigenetic stability for long-term cannabis cultivation.

The principal component analysis (PCA) plots highlight temporal shifts in mutational patterns across cultivars, with certain time points (e.g., T5 or T10 depending on the cultivar) showing distinct profiles that may reflect biological or culture-induced mutation spikes. In contrast, CpG methylation patterns were more stable throughout this study. The PCA of the methylation data revealed a lack of clustering between CPK samples, although T = 0 and T = 5 are grouped together, as well as T = 15 and T = 20, suggesting a slower but steady change in methylation patterns over time (Fig. 3A). Molecular functions of DNA methylation in particular regions are still not clearly understood. For example, promoter DNA methylation, which is usually linked with gene transcription inhibition, can in some cases promote gene transcription instead (Zang et al. 2018a). We identified 79 shared CpG DMPs between all sampling time points for CPK, reinforcing the idea that DNA methylation changes may occur less frequently than mutations (Fig. 3C). However, methylation changes may be involved in regulating the expression of genes affected by mutations to minimize their impact on the plant’s overall health in the case of deleterious mutations. Furthermore, the shared CpG DMPs between the different time points of the same cultivar identified in this study support the results of other studies suggesting that epigenetic changes, such as DNA methylation, can persist across multiple generations, reinforcing the heritable nature of these modifications (Johannes et al. 2009; Cortijo et al. 2014; Virgen et al. 2025). A recent paper on methylation in micro-propagated cannabis using the novel comparative restriction enzyme analysis of methylation (CREAM) method laid some foundational work regarding specific methylation patterns (Boissinot et al. 2023). Despite the different sequencing methods used, one of the key limitations they encountered was the lack of different sampling time points, which they identified as a crucial component for future research, especially regarding the initiation of culture as plants are transitioning from regular growth conditions to in vitro conditions. Since we observed a rapid plateau in genetic mutations after the fifth generation, but a sustained accumulation of DMPs over time, we suggest that the decline in genetic stability of cannabis during micropropagation may be associated primarily with alterations in DNA methylation patterns. Our study provides valuable insight and complementary data on this subject, as we found that differential methylation levels do not spike between T = 0 and T = 5 as observed with SNPs. This suggests that the initiation of culture may have less impact on DNA methylation than previously thought. Instead, we noticed slight increases and decreases in differential methylation levels over time for all three cultivars studied. Thus, DNA methylation does not accumulate in a strictly progressive manner but instead appears to vary over time, potentially reflecting adaptative responses to growth conditions and, to some extent, could possibly be linked to the observed increase in mutational variants. Indeed, DNA methylation can be altered at individual loci or across the entire genome under environmental stress conditions (Zhang et al. 2018b). Although some CHG and CHH methylation were detected, we focused our attention on CpG methylation since the EM-seq method is better at detecting this type of modification (Vaisvila et al. 2021). Moreover, a notable finding result from our dataset is the difference between cultivars at the level of CpG methylation. For instance, we observed that CPK, GC and GEL respectively had 22%, 9% and 13% for promoter regions, 13%, 5% and 6% for exon regions, 4%, 4% and 3% for intron regions and 61%, 82% and 78% for intergenic regions, indicating a strong cultivar dependency in the region of accumulation of these types of epimutations.

Overall, DNA methylation in cannabis micropropagation is still new from a scientific standpoint, and further research should absolutely take it into account, especially in relation to mutational variants. Our results also demonstrate that both mutations and methylation changes seem to appear in specific genomic hotspots, with certain regions being clearly more prone to mutation accumulation while others remain largely unaffected. In this study, we were able to detect mutational variability during micropropagation, confirming our initial hypothesis that 3D-GBS is a cost-effective and reliable tool for monitoring genetic stability in in vitro cannabis culture. While our findings provide concrete, interpretable data on mutation rates during long-term cannabis micropropagation, we also uncovered evidence indicating that mutations do not accumulate indefinitely as cultivation progresses, instead they appear to plateau after five transplanting cycles, which is a highly valuable insight for growers investing in micropropagation systems.

In line with our second hypothesis, EM-seq proved to be an effective tool for methylation monitoring, which revealed that overall DNA methylation levels remain relatively steady throughout the different stages of micropropagation, although their nature seems to vary over time. Nevertheless, our results show that both mutation and methylation profiles differ across cultivars, implying that cultivar-dependent optimization and monitoring strategies are essential for the successful implementation of an industrial-scale micropropagation production systems.