Genetic Mapping, Identification, and Characterization of a Candidate Susceptibility Gene for Powdery Mildew in Cannabis sativa

Interest in commercial cultivation of Cannabis sativa is expanding as regulatory agencies ease restrictions on the production of hemp and high-tetrahydrocannabinol (THC) C. sativa. Among the greatest challenges for producers of these emerging crops is the management of pests and diseases. Powdery mildew (PM) is one of the most prevalent diseases affecting the production of high-cannabinoid C. sativa. The disease is frequently observed under controlled environment cultivation (Scott and Punja 2021) and is common in outdoor cultivation of high-cannabinoid cultivars when environmental conditions are favorable for disease development (Stack et al. 2021). The most common pathogen causing PM in C. sativa is the biotrophic ascomycete, Golovinomyces ambrosiae (formerly G. spadiceus) (Cala 2022; Szarka et al. 2019; Wiseman et al. 2021). Infection of susceptible host genotypes by G. ambrosiae results in the production of dense fungal hyphae and chains of asexual conidia, which appear as white, powdery patches on adaxial leaf surfaces and occasionally on stems and inflorescences (Wiseman et al. 2021). While the economic impact of PM on C. sativa has not been rigorously assessed, PM can reduce photosynthesis and yield potential by damaging foliar tissues and blocking light from reaching the leaf surface, resulting in premature senescence and defoliation. Further, the presence of residual fungal biomass reduces the postharvest marketability of dried inflorescences (Thompson et al. 2017).

One existing management strategy employed by growers to control PM is the use of fungicides. There are approved and reasonably efficacious fungicides available for controlling PM in C. sativa (Akinrinlola and Hansen 2023; Cala 2022; Scott and Punja 2021). However, there is grower concern about the potential for residual biological fungicides (e.g., Bacillus spp.) to increase microbial abundance in harvested biomass, which could result in failed regulatory tests. Another management strategy employed in many crops to reduce PM is the manipulation of environmental conditions to suppress pathogen establishment and proliferation, including relative humidity, temperature, and lighting (Amsalem et al. 2006). One particularly successful strategy is the application of supplemental ultraviolet (UV) radiation at night (Suthaparan et al. 2014). While these methods are effective at controlling PM in other pathosystems, studies concerning their efficacy for C. sativa are limited. A third strategy to manage PM is the deployment of resistant cultivars. However, there is a lack of knowledge of G. ambrosiae genetic diversity and the number of resistance alleles present in existing C. sativa germplasm. To date, only one PM resistance/susceptibility locus in C. sativa has been characterized (Mihalyov and Garfinkel 2021). Thus, the identification and introgression of durable PM resistance into elite germplasm is imperative for growers to implement robust integrated pest management programs.

Genetic resistance to PM has been documented in many plant species (Asgarinia et al. 2013; Cui et al. 2022; Feechan et al. 2011; Fondevilla and Rubiales 2012; Kasettranan et al. 2010; Olczak-Woltman et al. 2011). There are two primary mechanisms for qualitative resistance against PM. The first is based on gene-for-gene resistance, in which a pathogen-secreted effector protein is recognized by the compatible protein product of a host resistance (R) gene. R genes are often characterized by conserved nucleotide-binding site (NBS) and/or leucine-rich repeat (LRR) domains, and upon effector binding, function to induce a hypersensitive response in the host plant (Bent and Mackey 2007; Kang et al. 2020; Schulze-Lefert and Vogel 2000). The second resistance mechanism involves the loss-of-function mutation of a susceptibility (S) gene. The Mildew Locus O (MLO) family of S genes encode transmembrane proteins in host plants that can facilitate infection when they interact with the fungal pathogen (Jacott et al. 2021; Kusch and Panstruga 2017). Repression of MLO protein activity through loss-of-function mutations (mlo) has been shown to confer broad-spectrum PM resistance in numerous taxa (Elliott et al. 2002). Because mlo-based resistance is not race specific, it has greater potential for durable PM resistance compared with R-gene resistance, which can be more easily overcome. In addition to qualitative resistance resulting from dominant R genes or homozygous S gene knockouts, small-effect QTL may also be important contributors to quantitative PM resistance (He et al. 2013).

There are very few studies that have investigated potential sources of genetic resistance to PM in C. sativa. Recently, Mihalyov and Garfinkel (2021) identified PM1, a dominantly inherited locus from the cultivar ‘PNW39’. Proximal to PM1, 10 annotated genes in a 1.5-Mb interval flanking the peak marker have conserved N-terminal coiled-coil, NBS, or LRR domains, which are consistent with plant R genes. However, the specific gene or causal polymorphism(s) at the PM1 locus was not reported. Recent curation of the MLO gene family in C. sativa by Pépin et al. (2021) identified 15 MLO (CsMLO) genes from five publicly available C. sativa genome assemblies and found that two clade V MLO genes, CsMLO1 and CsMLO4, were upregulated in response to G. ambrosiae infection. This is consistent with the behavior of clade V MLO genes, which are reported to act as S genes in other dicot species (Consonni et al. 2006).

In this study, we characterize a new source of PM resistance in C. sativa. The clonal cultivar ‘FL 58’ (U.S. National Plant Germplasm System [NPGS] accession G 33236) was first identified as a potential source of resistance to PM as reported in Stack et al. (2021). FL 58 was included in field trials subject to natural and controlled inoculation for 3 consecutive years and no significant PM infection was observed, while adjacent plants were heavily infected with the pathogen. Two F₂ populations derived from the biparental cross of FL 58 with the PM-susceptible cultivar ‘TJ's CBD’ (U.S. NPGS accession G 33580) were leveraged to identify the genetic basis of PM resistance (Figs. 1 and 2A). We discuss several QTL associated with partial or complete resistance to PM and introduce new molecular tools to facilitate the breeding of resistant C. sativa cultivars.

Download as PowerPoint

Download as PowerPoint

Results

Distributions of PM susceptibility in F₁ and F₂ populations

Eight F₁ progeny from the cross of the resistant genotype FL 58 with the susceptible genotype TJ's CBD were evaluated in replicated field and growth chamber trials. A range of susceptibility to PM was observed in F₁ progeny (Fig. 2C). While the level of PM disease severity for any F₁ progeny was not as low as in the resistant parent, FL 58, there was a range of susceptibility, with many having greater disease severity than the susceptible parent, TJ's CBD. The rank order of disease severity between the field evaluation and the growth chamber inoculation were similar.

Two individual F₁ progeny were self-pollinated to generate two F₂ families that were planted in the field, inoculated with PM, and rated weekly over the course of 5 weeks. Similar to F₁ progeny, the F₂ populations displayed a range of susceptibility to PM, with many being more susceptible than the susceptible parent, TJ's CBD (Fig. 2D). Approximately 25% of individuals in both F₂ populations appeared qualitatively resistant with a PM rating of 0% disease, consistent with the phenotype of the resistant parent. This resistance observed in the F₂ progeny was not observed in any F₁ progeny, albeit only a small number were surveyed. The proportions of susceptible and resistant individuals in the F₂ populations were not significantly different from a 3:1 ratio expected for a single recessive susceptibility (S) gene (Table 1).

Table 1. Chi-square tests for goodness-of-fit for each of the two Cannabis sativa ‘FL 58’ × ‘TJ's CBD’ F₂ mapping populations

		Resistant		Susceptible		Test for 3:1 segregationa
Population	n	Obs.	Exp.	Obs.	Exp.	χ2	df	P
GVA-H-21-1004	378	99	94.5	279	283.5	0.085	1	0.77
GVA-H-21-1005	305	78	76.25	227	228.75	0.005	1	0.94

^a To satisfy the null hypothesis, a 3:1 segregation is expected of susceptible (AUDPC > 0) to resistant (AUDPC = 0) individuals. AUDPC, area under the disease progress curve.

View as image HTML

Identification of QTL and hot spots

Among the eight measured and derived traits used in linkage mapping, 29 QTL were identified across five chromosomes, many of which were colocated with other QTL (i.e., hot spots) (Table 2). One example hot spot was at 23.3 centimorgans (cM) on Chr01 (Chr01.23), harboring QTL for seven of the eight traits. Considering QTL with overlapping log of odds (LOD) support intervals, five significant hot spots were identified that potentially contribute to resistance/susceptibility to PM: Chr01.23, Chr02.56, Chr04.45, Chr08.58, and Chr09.27 (Table 2). For area under the disease progress curve (AUDPC), all hot spots were represented (Fig. 3).

Table 2. Quantitative trait loci (QTL) mapping of powdery mildew (PM) traits in two segregating Cannabis sativa F₂ populations evaluated in Geneva, NY^a

Trait	ST	QTL	HSPOT	LOD	CHR	POS (cM)	INT (cM)	INT (Mb)	PVE	|d/a|	CLASS
16 dpi	4.62	pm_16dpi.1	Chr01.23	31.8	1	23.1	(22.1–24.6)	(0.00–59.37)	4.52	0.62	PD – Sus.
		pm_16dpi.2	Chr02.56	10.9	2	63.7	(43.1–69.9)	(18.27–95.16)	2.58	0.34	PD – Res.
		pm_16dpi.4	Chr04.45	5.3	4	45.2	(39.2–99.9)	(2.92–85.04)	2.89	0.03	A
		pm_16dpi.8	Chr08.58	6.2	8	61.4	(47.1–66.2)	(57.16–60.41)	3.17	0.47	PD – Res.
23 dpi	4.49	pm_23dpi.1	Chr01.23	85.7	1	23.3	(23.1–23.7)	(19.60–22.99)	9.84	0.71	PD – Sus.
		pm_23dpi.2	Chr02.56	7.2	2	53.6	(47.7–62.6)	(30.30–95.16)	2.37	1.46	OD – Sus.
		pm_23dpi.4	Chr04.45	17.4	4	45.2	(39.2–50.1)	(2.61–21.42)	5.19	0.02	A
		pm_23dpi.8	Chr08.58	6.6	8	55.4	(53.1–74.3)	(57.49–61.78)	3.76	0.45	PD – Res.
31 dpi	4.58	pm_31dpi.1	Chr01.23	95.1	1	23.3	(23.1–23.7)	(19.60–22.99)	22.31	0.86	D – Sus.
		pm_31dpi.2	Chr02.56	10.3	2	53.6	(47.7–62.6)	(30.30–95.16)	3.61	1.63	OD – Sus.
		pm_31dpi.4	Chr04.45	17.2	4	45.2	(39.2–50.2)	(2.61–21.42)	4.27	0.49	PD – Sus.
		pm_31dpi.8	Chr08.58	8.9	8	61.4	(54.2–63.9)	(57.49–60.06)	3.65	0.07	A
39 dpi	4.64	pm_39dpi.1	Chr01.23	99.5	1	23.3	(23.1–23.7)	(19.60–22.99)	33.44	0.84	D – Sus.
		pm_39dpi.2	Chr02.56	9.1	2	53.6	(47.7–62.6)	(30.30–95.16)	2.6	1.62	OD – Sus.
		pm_39dpi.4	Chr04.45	11.9	4	45.2	(39.2–50.2)	(2.61–21.42)	2.42	1.54	OD – Sus.
		pm_39dpi.8	Chr08.58	7.4	8	55.4	(53.1–71.7)	(57.49–60.97)	2.91	0.01	A
		pm_39dpi.9	Chr09.27	5.6	9	26.9	(21.3–32.6)	(2.07–4.10)	2.03	0.6	PD – Res.
44 dpi	4.63	pm_44dpi.1	Chr01.23	98	1	23.3	(23.1–23.7)	(19.60–22.99)	38.96	0.95	D – Sus.
		pm_44dpi.2	Chr02.56	5.4	2	53.6	(33.3–69.3)	(13.98–95.16)	1.53	1.64	OD – Sus.
		pm_44dpi.4	Chr04.45	6.9	4	45.2	(39.2–50.2)	(2.61–21.42)	1.72	5.44	OD – Sus.
		pm_44dpi.8	Chr08.58	7.3	8	61.4	(54.2–66.2)	(57.49–60.41)	2.17	0.03	A
AUDPC	4.58	pm_AUDPC.1	Chr01.23	93.1	1	23.3	(23.1–23.7)	(19.60–22.99)	29.27	0.86	D – Sus.
		pm_AUDPC.2	Chr02.56	9.4	2	53.6	(47.7–62.6)	(30.30–95.16)	2.92	1.59	OD – Sus.
		pm_AUDPC.4	Chr04.45	12.5	4	45.2	(39.2–50.2)	(2.61–21.42)	3.4	0.87	D – Sus.
		pm_AUDPC.8	Chr08.58	8.4	8	61.4	(54.2–65.5)	(57.48–60.32)	3.16	0.07	A
		pm_AUDPC.9	Chr09.27	5.1	9	26.9	(13.1–30.8)	(1.33–3.76)	2.07	0.65	PD – Res.
Days to first PM	3.56	pm_daystoPM.2	Chr02.56	8.3	2	63.7	(33.0–71.2)	(7.58–95.16)	7.33	0.98	D – Sus.
		pm_daystoPM.8	Chr08.58	4.6	8	49.4	(3.4–61.9)	(0.35–62.51)	2.88	0.54	PD – Sus.
Res./Sus.	3.5	pm_R/S.1	Chr01.23	89.9	1	23.3	(23.1–23.7)	(19.60–22.99)	46.4	1.08	D – Sus.

^a For each trait, the 95% LOD significance threshold (ST) based on 1,000 permutations is reported. For each quantitative trait locus, the hot spot (HSPOT), QTL LOD score (LOD), chromosome (CHR), position of the peak marker (POS) in centimorgans (cM), 1.5 LOD support interval in cM and Mb based on alignment to the CBDRx genome (INT), percent variance explained (PVE), dominance-to-additive effect ratio (|d/a|), and classification based on |d/a| (CLASS) are reported. A, additive; PD, partially dominant; D, dominant; OD, overdominant; Res., resistant; Sus., susceptible.

View as image HTML

Download as PowerPoint

Percent variance explained (PVE) values ranged from 1.5 to 46.4% and the dominance-to-additive effect (|d/a|) ratios from 0.02 to 5.44 (Table 2). For some colocated QTL, |d/a| values varied based on the trait used for mapping. At later days postinoculation (dpi), |d/a| increased for QTL at the Chr04.45 hot spot, decreased for those at Chr08.58, but remained relatively consistent for those at Chr02.56. For all reported QTL, the allele contributed from FL 58 at the peak marker was the beneficial allele with respect to reduced levels of susceptibility to PM.

QTL within the Chr01.23 hot spot explained a majority of the phenotypic variation for all traits analyzed and harbored the QTL for the binary ‘Resistant/Susceptible’ trait. Importantly, |d/a| and AUDPC data for the QTL at Chr01.23 (Table 2; Fig. 3G) support a single recessive S-gene model, initially predicted by the 3:1 segregation of susceptible to resistant individuals. Epistasis was detected between pm_AUDPC.1 and pm_AUDPC.8 (Supplementary Fig. S1). When an individual was AA or AB at the pm_AUDPC.8 locus, the pm_AUDPC.1 locus was partially dominant or dominant with respect to AUDPC, but when an individual was BB at the pm_AUDPC.8 locus, the pm_AUDPC.1 locus was overdominant, whereby heterozygotes were much more susceptible than either homozygote.

Identification of candidate S genes for the chr01.23 hot spot

The region surrounding the peak marker for the Chr01.23 hot spot was searched for candidate S genes. Of the seven traits with QTL in this region, six had the same peak marker at 23.35 cM. Similarly, LOD support intervals were most frequent between 23.13 and 23.74 cM. This 0.61 cM interval represents a 3.4-Mb interval on CBDRx Chr01, spanning 19.60 to 22.99 Mb (Fig. 4A). Within this interval, there are 158 annotated gene models, of which, 40 were not protein coding and an additional 33 were uncharacterized (Supplementary Table S1). Starting closest to the peak marker, genes were examined to identify candidate S genes. Approximately 93.5 kb from the peak marker alignment at 21.54 Mb, the clade V MLO gene CsMLO1 (LOC115705883) was identified as the primary candidate S gene.

Download as PowerPoint

Candidate S gene CsMLO1 has a large insertion in FL 58

Further investigation of CsMLO1 sequences in PacBio HiFi data generated from the parents revealed that the resistant parent, FL 58, has a 6.8-kb insertion between exons 10 and 11 that was not present in either the susceptible parent TJ's CBD or the CBDRx reference sequence (GenBank accession no. OR352188; Fig. 4B). This insertion has nucleotide and amino acid sequence homology with transposable elements in other plant species, including copia retrotransposon Aesp8 in Aegilops speltoides, a putative gag-pol polyprotein in Cucumis melo and Citrus sinensis, a putative Ty1-copia subclass retrotransposon protein in C. melo, a copia retrotransposon rider in Trifolium pratense, a putative RNA-directed DNA polymerase in Helianthus annuus, and retrovirus-related pol polyprotein from transposon TNT in Melia azedarach and Vitis vinifera (Supplementary Tables S2 and S3). The nucleotide sequence also aligned with LOC115695807 and LOC115725460 at 18.3 and 70.9 Mb of CBDRx Chr06, respectively (Supplementary Table S2).

Insertion in FL 58-type CsMLO1 disrupts mRNA splicing and translation

Amplification and sequencing of cDNA from the resistant and susceptible parents indicated that the CsMLO1 present in TJ's CBD produced a single spliced transcript that contained all 15 exons and encoded a full-length open reading frame (ORF) (Fig. 4C). In contrast, the CsMLO1 with the 6.8-kb insertion present in FL 58 produced two transcripts, one with sequence similarity to that found in TJ's CBD and another with an 846-bp fragment of the insertion incorporated into the final transcript (Fig. 4C). This extra sequence included in the spliced transcript encodes a premature stop codon, truncating the protein before the 11th exon (Fig. 4C).

To investigate the relative abundance of the transcripts encoding full-length ORFs and those encoding premature stops, as well as the impact of inoculation with G. ambrosiae on transcript abundance, detached branches of four genotypes were inoculated or mock-inoculated with G. ambrosiae. Leaf samples were collected 72 h after inoculation, and a quantitative polymerase chain reaction (qPCR) was conducted using transcript-specific qPCR primers. With respect to inoculation, there was a numerical increase in relative expression of CsMLO1 for all genotypes, but only FL 58 (χ²(1) = 3.86, P = 0.0495) and GVA-H-19-1067-001 (χ²(1) = 3.86, P = 0.0495) were significantly different by treatment (Fig. 4D). In TJ's CBD, only transcripts with full-length ORFs were detected, while in FL 58, both transcripts were detected, but the transcript encoding a premature stop was more abundant (Fig. 4D). In the heterozygous individuals, GVA-H-19-1166-005 and GVA-19-1067-001, both transcripts were present at ratios between 1:1 and 3:1 (Fig. 4D).

CsMLO1_FL58-1 and CsMLO1_FL58-2 are strong predictors of PM resistance

The CsMLO1_FL58-1 fluorescence-based assay and CsMLO1_FL58-2 gel-based assay were able to clearly resolve three genotypic groups (Fig. 5A and C) that correspond nearly perfectly with the AUDPC phenotype from progeny across both F₂ populations (Fig. 5B). CsMLO1_FL58-1 and CsMLO1_FL58-2 were able to resolve resistant from susceptible individuals in 99.5 (n = 192) and 100% (n = 42) of cases, respectively, and there was a significant difference in AUDPC by allelic group using both CsMLO1_FL58-1 (χ²(2) = 96.42, P < 0.001) and the CsMLO1_FL58-2 (χ²(2) = 28.59, P < 0.001) (Fig. 5B). These phenotypes match a single-gene recessive model, wherein AA homozygotes were resistant to PM and the other two genotype groups were susceptible.

Download as PowerPoint

The FL 58-type allele of CsMLO1 is uncommon in available genome sequences

Sequence queries using NCBI databases, five additional genotypes with PacBio HiFi data, and a C. sativa pangenome being produced by the Michael lab determined the FL 58-type allele of CsMLO1 is relatively uncommon in available genome sequences. Of the genomes available on NCBI, the FL 58-type allele only aligned to Cannbio-2 (GCA_016165845.1). Similarly, the sequence was only identified in one of the genotypes sequenced with PacBio HiFi, GVA-H-19-1067-001, and was later determined to be heterozygous in that genotype using the previously described molecular markers. The sequence was identified in 29 of the 193 genomes in the C. sativa pangenome (Supplementary Table S4), many of which are directly related to one another. In all cases, the sequence was >99% identical to the FL 58 sequence.

Discussion

This study reports at least five unique and previously unidentified loci that contribute to variation in resistance or susceptibility to PM in C. sativa. The marker–trait relationships for the QTL at the Chr01.23 hot spot are consistent with a single-locus, recessive S-gene model that confers resistance to PM. Genomic sequence, expression analysis, and cDNA sequencing support the hypothesis that the resistance is a result of an insertion in the FL 58 CsMLO1 sequence that causes improper mRNA splicing resulting in a premature stop codon.

Candidate gene CsMLO1 colocates with the chr01.23 hot spot

Seven PM traits that had overlapping QTL at the Chr01.23 hot spot shared a peak marker at 23.35 cM. The associated single-nucleotide polymorphism (SNP) array probe sequences for this marker aligned to Chr01 in the CBDRx genome assembly at 21.54 Mb, less than 100 kb from CsMLO1, one of the two candidate MLO S genes identified by Pépin et al. (2021). Further investigation of CsMLO1 in the parents revealed a 6.8-kb insertion in the sequence of the resistant parent when compared with the susceptible parent. The sequence homology of this insertion to characterized transposable element features, coding sequences, and other regions of the CBDRx genome assembly suggests that the insertion may be a remnant of transposition from another region of the genome and could interfere with CsMLO1 transcription, translation, or protein function.

Expression data confirm that an 846-bp piece of this genomic insertion is routinely spliced into the CsMLO1 mRNA transcript, resulting in a premature stop codon that truncates the protein. While there were CsMLO1 transcripts with full-length ORFs detected in FL 58, transcripts encoding a premature stop were 36 to 65 times as abundant. The dramatic reduction in functional CsMLO1 proteins could explain the resistant phenotype observed in genotypes homozygous for the FL 58-type CsMLO1. Additionally, there is evidence in other species that MLO proteins dimerize/oligomerize (Elliott et al. 2005). The overwhelming number of truncated relative to functional CsMLO1 proteins in FL 58 could further reduce the number of functional proteins if dimerization/oligomerization of a functional protein with a truncated protein disrupts protein function.

In many species, all of the clade V MLO genes must be knocked out to confer complete resistance to PM (Consonni et al. 2006), while in others, knockout of a subset of the clade V MLO genes is sufficient for conferring resistance (Bai et al. 2008; Pavan et al. 2011). In this study, no QTL or epistasis was detected near CsMLO4, the other known clade V CsMLO gene. Further, the CsMLO4 sequence from FL 58 was found to be 100% identical to the CBDRx CsMLO4 sequence. Upregulation of host gene expression following pathogen infection is a strong predictor of clade V MLO genes that contribute to host susceptibility (Feechan et al. 2011; Pessina et al. 2014). Pépin et al. (2021) observed upregulation of both CsMLO1 and CsMLO4 after infection, suggesting that both would behave as functional S genes and a double-knockout would be necessary to confer mlo-based resistance.

Assuming the insertion found in the FL 58 CsMLO1 confers the recessive resistance observed, there are several potential explanations for the complete resistance we observed with only a knockout of CsMLO1 and a predicted multigenic model involving both CsMLO1 and CsMLO4 based on gene expression data. These include: (i) CsMLO4 inactivation is not necessary for mlo-based resistance in C. sativa; (ii) CsMLO4 is functional in FL 58 and TJ's CBD but only contributes to PM susceptibility in some C. sativa genotypes; (iii) CsMLO4 is functional in FL 58 and TJ's CBD, but only contributes to PM susceptibility for some G. ambrosiae genotypes; and (iv) CsMLO4 is nonfunctional in both FL 58 and TJ's CBD, thus epistasis between CsMLO1 and CsMLO4 was not observed in these F₂ progeny, but could be in other genetic backgrounds. Further, some C. sativa genomes appear to have multiple copies of CsMLO1 (Pépin et al. 2021), so there is potential that multiple genes would need to be knocked out to confer resistance. More comprehensive analysis of CsMLO1 and CsMLO4 homologs in diverse germplasm and potential epistasis between clade V MLO genes is necessary to build a comprehensive understanding of mlo-based resistance in C. sativa.

Nonadditive genetic effects contribute to quantitative variation in PM susceptibility

The dramatic range of PM phenotypes in qualitatively susceptible genotypes highlights the prevalence and importance of loci that confer intermediate quantitative resistance to PM. With respect to AUDPC and many of the other PM traits, the F₂ populations had substantial transgressive segregation, with numerous progeny that were more susceptible than the susceptible parent. One prediction from the phenotype data alone would be that both FL 58 and TJ's CBD contributed complementary beneficial or deleterious alleles, such that some F₂ progeny would inherit deleterious alleles from both parents and be more susceptible than the susceptible parent. However, when the marker data are considered, the pyramiding of homozygous deleterious alleles does not seem to be a major contributor to the transgressive segregation. For all QTL identified, FL 58 contributed the beneficial allele.

Nonadditive genetic effects, including overdominance and epistasis, explain some of the observed transgressive segregation. Seven QTL had a |d/a| > 1.2, indicating overdominance at those loci, where the heterozygous group was on average more susceptible than the susceptible homozygote group. This also could explain why many of the F₁ progeny were more susceptible than the susceptible parent, as they were all heterozygous for the overdominant QTL. Overdominance is not uncommon for disease resistance QTL and has been identified in other PM pathosystems (He et al. 2013; Maroof et al. 1994). Beyond overdominance, epistasis was detected between pm_AUDPC.1 at the Chr01.23 hot spot and pm_AUDPC.8 at the Chr08.58 hot spot. This interaction resulted in a group, AB at pm_AUDPC.1 and BB at pm_AUDPC.8, that was on average more PM susceptible than the other eight allele group combinations and the susceptible parent. In addition to the QTL identified in this study, there must be other loci that contribute to quantitative susceptibility in broader germplasm, as many cultivars are significantly more PM susceptible than TJ's CBD (Stack et al. 2021). Development and evaluation of additional populations and diversity panels will certainly reveal more loci associated with PM resistance and susceptibility.

Considerations for QTL-by-pathogen interactions

Very little is known about the diversity and population structure of G. ambrosiae, specifically, whether there are races that display differential virulence based on host genotype. One key feature in pathogen evolution to overcome or avoid host resistance is recombination facilitated by sexual reproduction. Generally in PM fungi, two mating types are necessary for sexual recombination and the subsequent production of chasmothecia (Brewer et al. 2011). Chasmothecia production has been reported for G. ambrosiae in Kentucky, U.S.A. (Szarka et al. 2019), suggesting that both mating types exist in Kentucky with capacity for sexual recombination, but this has not yet been reported in other regions. Controlled inoculation challenges of host genotypes representing different sources of resistance against characterized diverse pathogen isolates will be necessary to optimally stack and select for resistance loci in improved cultivars and then deploy those into production systems.

While the majority of the PM on C. sativa results from infection by G. ambrosiae, other fungal species that cause PM have been demonstrated to infect C. sativa, including Podosphaera macularis (Bates et al. 2021; Punja 2022). Weldon et al. (2020) inoculated three hemp cultivars with an isolate of P. macularis and observed successful infection and disease development on two of them. Interestingly, those authors observed an incompatible host-resistance interaction resulting in a localized hypersensitive response when inoculating TJ's CBD, the susceptible parent in this study. This suggests that TJ's CBD has an R gene that confers resistance to P. macularis. Other F₂ progeny from these populations, or other populations derived from TJ's CBD, could be leveraged to investigate that prediction. A related question is whether the mlo-based resistance in C. sativa is functional against P. macularis. In Solanum lycopersicum, knockout of SlMlo1 confers resistance to both PM pathogens Oidium neolycopersici (Bai et al. 2008) and Leveillula taurica (Zheng et al. 2013). However, more studies are needed to determine whether different MLO knockouts, and the relative resistance conferred by each, may be pathogen specific.

Development of durable PM-resistant cultivars

Introgression and pyramiding of both qualitative disease resistance and susceptibility loci in elite germplasm have long been strategies for building durable disease resistance (Nelson et al. 2018). High-throughput molecular marker assays like those presented here for CsMLO1 and by Mihalyov and Garfinkel (2021) for PM1, facilitate rapid selection and backcrossing without costly recurrent phenotypic selection, especially for recessive genes. The Chr01.23 hot spot/CsMLO1 is not linked to PM1 on Chr02, nor to other minor QTL identified in this study, which facilitates gene pyramiding without breaking physical linkage. However, it will require considerable effort to maintain numerous unlinked loci in future improvement of elite germplasm. In addition, some loci that confer resistance may also result in a yield penalty when incorporated alone or in combination with other genes (Brown and Rant 2013; McGrann et al. 2014).

While the FL 58-type allele of CsMLO1 does not appear to be common in available germplasm, it does appear to be present in individuals other than FL 58. This will certainly aid in the introgression of the allele into elite breeding lines. Further characterization of this allele in diverse backgrounds will be necessary to improve knowledge of its function and interaction with other loci. Both GVA-H-19-1067-001 and ‘White CBG’, which are heterozygous for the FL 58-type allele, are susceptible to PM (Cala 2022).

One challenge for introgression of the Chr01.23 hot spot/CsMLO1 is the proximity to the major-effect photoperiod insensitivity gene Autoflower1. Importantly, these two loci are linked in repulsion phase, so very large population sizes would be required to break the linkage. However, once in coupling phase, linkage should be relatively easy to maintain in a breeding program. This is particularly relevant for hybrid breeding of high-cannabinoid cultivars, where day-neutral Autoflower1 homozygous parents are often used as pollen donors (Smart et al. 2022). Besides traditional breeding methodologies, gene editing is a promising technology that could be used to precisely knock out S genes but otherwise maintain an existing genetic background (Zaidi et al. 2018). Once efficient gene-editing systems are deployed in C. sativa, this will likely be the most efficient method of introducing S-gene knockouts, especially in hybrid cultivar development, in which both inbred parents must be fixed homozygous to produce resistant progeny.

Materials and Methods

Population development

As described in Carlson et al. (2021), two clonal cultivars, FL 58 (generously contributed by Sunrise Genetics, Fort Collins, CO) and TJ's CBD (generously contributed by Stem Holdings Agri, Eugene, OR), were crossed to generate the F₁ family designated GVA-H-19-1166 (Fig. 2A). Eight of the F₁ progeny were evaluated, and two were selected as parents of F₂ populations. For the two F₁ genotypes, GVA-H-19-1166-002 and GVA-H-19-1166-008, five cuttings were taken from a single source plant and moved to separate pollen-isolated growing spaces with 16-h photoperiod. After the plants had grown to approximately 1 m in height, the lighting schedule was changed to a 12-h photoperiod to induce inflorescence development. Two of the five cuttings for each genotype were induced to produce pollen using three weekly foliar spray applications of a 8 mM silver thiosulfate solution (Lubell and Brand 2018). Seed was collected, cleaned, and pooled from all five cuttings. The F₂ population produced by self-pollinating GVA-H-19-1166-002 was designated as GVA-H-21-1004, and the F₂ population produced by self-pollinating GVA-H-19-1166-008 was designated as GVA-H-21-1005.

Phenotyping for PM susceptibility

In 2020, four cuttings of each of the F₁ progeny and of the two parents were rooted in LM-111 potting mix with Clonex (Hydrodynamics International, Lansing, MI), transplanted into a field in Geneva, NY, and maintained following the protocols described by Stack et al. (2021). The field was naturally inoculated with G. ambrosiae, and disease severity was rated (0 to 100%) on August 14, September 2, and September 17. In addition to the field evaluation, six cuttings of each F₁ genotype were rooted and transplanted into 10.2-cm² pots, moved to a growth chamber, and arranged in a randomized complete block design. Liquid inoculum of the G. ambrosiae isolate 19002 (Weldon et al. 2020) was prepared by washing infected hemp leaves in a solution of 100 µl of Tween 20 per 1 liter of distilled water. Plants were spray-inoculated to runoff with a conidial suspension of 2 × 10⁵ spores/ml and evaluated for disease severity at 7, 12, 14, 19, 21, and 25 dpi.

In 2021, the individuals from the GVA-H-21-1004 (n = 383) and GVA-H-21-1005 (n = 323) F₂ populations were seeded in a greenhouse on June 7 and then transplanted into a field in Geneva, NY, on July 6. Ten replicated single-plant plots of each of the parental genotypes FL 58 and TJ's CBD, rooted using the protocol above, were also included throughout the field. All of the plants in the field were sprayed with liquid inoculum from G. ambrosiae isolate 20001 (Cala 2022) prepared in the same manner described above with a concentration of 1.1 × 10⁴ spores/ml on August 16, 2021. Disease severity was rated for each of the plants in the field at 16, 23, 31, 39, and 44 dpi.

For all evaluations, AUDPC was calculated using the audpc function from the R (R Core Team 2022) package agricolae (De Mendiburu and Simon 2015). For the evaluation of the F₁ progeny, AUDPC values were normalized by trial to correct for the variation in rating frequency and timing among trials. If PM was observed on an individual, the number of days between inoculation and observation of PM disease was calculated. Additionally, F₂ individuals were classified as either resistant (AUDPC = 0) or susceptible (AUDPC > 0).

Chi-square goodness-of-fit tests were used to determine whether the observed proportions of susceptible and resistant individuals in the two F₂ populations statistically deviated from a 3:1 ratio expected for a single recessive S gene.

Genotyping

Leaf tissue was collected from all plants in the field (n = 730) on September 2, 2021, and immediately lyophilized. DNA was isolated from all F₂ genotypes and three replicates of each parent genotype using the DNeasy 96 Plant Kit (Qiagen, Germantown, MD). Samples were genotyped using an Illumina (San Diego, CA) SNP array developed by Phylos Bioscience (Portland, OR). Of the 54,883 SNPs called, 14,267 were polymorphic in the F₂ populations.

Linkage mapping

The R packages ASMap (Taylor and Butler 2017) and qtl (Broman et al. 2003) were used to construct the linkage map and conduct QTL analyses, respectively. Biallelic F₂ marker data from the Illumina array were coded such that allele A was contributed by FL 58 and allele B was contributed by TJ's CBD. Using the pullCross function, markers were removed that were in complete linkage disequilibrium with other markers, had significant segregation distortion (threshold = 1 × 10⁻¹⁵), or had a significant proportion of missing data (missingness > 0.35). Genotype statistics were calculated, and individuals with greater than 750 crossovers were removed. The mstmap function was used to construct linkage maps. Linkage groups containing fewer than 10 markers were removed. Three pairs of linkage groups were merged because of significant associations among markers across linkage groups. Marker order within each linkage group was optimized after merging and filtering. The final linkage map included 670 F₂ individuals and contained 2,703 markers across 10 linkage groups representing the 10 chromosomes of C. sativa (Supplementary Dataset S1). The Illumina array probe sequences (Supplementary Dataset S2) were aligned to the CBDRx v. 2.0 reference genome (cs10, NCBI accession GCA_900626175.2) (Grassa et al. 2021) to order and orient the linkage groups relative to the reference. Each linkage group aligned to a single CBDRx chromosome.

QTL analysis

Genotype probabilities at each marker were calculated using the calc.genoprob function, and pseudomarkers were inserted in the map at 5-cM intervals. For each trait, the scanone function was used to calculate marker LOD scores and to determine a 95% LOD significance threshold based on the results of permutation tests (n = 1,000). The models used for each trait were: normal (model = ‘normal’) for ‘Days to PM’, binary (model = ‘binary’) for ‘Resistant/Susceptible’, and two-part (model = ‘2part’) for all other traits. The two-part model was selected to account for both the binary (resistant/susceptible) and continuous components of the distribution. For the two-part models, the lod.p.mu LOD scores were used. For each quantitative trait locus that exceeded the significance threshold, the maximum LOD and peak marker position were reported. Based on the peak markers, 1.5 LOD support intervals were calculated using the lodint function. When QTL from different traits had overlapping LOD support intervals, QTL hot spots were identified and named using the mean genetic position of the peak QTL markers in the linkage map. The function fitqtl was used to construct an additive-effects model for each trait and calculate the PVE for each quantitative trait locus. The fitqtl function was used to test pairwise epistasis between QTL. The model was reduced in a reverse stepwise fashion until all nonsignificant terms (P > 0.05) were removed from the model. Mean phenotypic values for each of the three possible genotypes of a peak marker were used to calculate the dominance-to-additive effect ratio (|d/a|) (Edwards et al. 1987).

Candidate gene identification

The 0.61-cM LOD support interval for the PM resistance quantitative trait locus on Chr01 was aligned to a 3.4-Mb physical interval (19.60 to 22.99 Mb) on CBDRx Chr01. Within that interval, there were 158 annotated gene models, of which 87 were characterized and protein coding. Among those, the clade V MLO gene CsMLO1, located 93.5 kb from the peak marker, was identified as a strong candidate for further investigation. PacBio HiFi whole-genome sequencing of FL 58 and TJ's CBD DNA was performed at the University of Wisconsin Biotechnology Center (Madison, WI) to compare CsMLO1 and other candidate regions. Pairwise alignment of CsMLO1 in CBDRx reference and the parents was performed to identify polymorphisms between parental alleles. An insertion identified in CsMLO1 of FL 58 was used as the query in BLASTN and BLASTX searches of the NCBI sequence database.

Identification of FL 58-type CsMLO1 alleles’ available genome sequences

To investigate the prevalence of the FL 58-type CsMLO1 allele in available germplasm, the 12-kb FL 58 gene sequence was aligned using BLASTN to available C. sativa genome sequences available on NCBI, PacBio HiFi reads of five additional C. sativa genotypes (GVA-H-19-1067-001, GVA-H-19-1186-059, GVA-H-21-1003-002, GVA-H-22-1061-002, and ‘Panakeia v2.0’) generated using the same method described for FL 58 and TJ's CBD, and the C. sativa pangenome being assembled by the Michael lab.

Sequencing of CsMLO1 transcripts in FL 58 and TJ's CBD

RNA was isolated from leaves of TJ's CBD and FL 58 using a Spectrum Plant Total RNA Kit (Sigma-Aldrich, St. Louis, MO, U.S.A.) and cDNA was synthesized using a ProtoScript II First Strand cDNA Synthesis Kit (New England Biolabs, Ipswich, MA, U.S.A.) with Oligo-dT [d(T)₂₃VN] primers, both following the standard protocols outlined in the kits. Primers were designed in the annotated 5′ and 3′ untranslated regions (UTRs) of CsMLO1 to amplify the full-length ORF (Supplementary Table S5). PCR products were run on a 1% agarose gel at 125 V for 45 min with a 1-kb ladder to estimate amplicon length. Three PCR products, one from TJ's CBD and two from FL 58, were cloned into Escherichia coli using a NEB PCR Cloning Kit with a pMiniT 2.0 vector backbone (New England Biolabs). Three colonies containing one of each of the PCR products were confirmed using colony PCR following the insert screening protocol in the cloning kit. Plasmids were purified using a QIAprep Spin Miniprep Kit (Qiagen) and then Sanger sequenced at the Cornell Sequencing Core facility using the 5′ and 3′ UTR primers as well as four internal primers (Supplementary Table S5).

Expression of CsMLO1 in response to inoculation with G. ambrosiae

To determine whether there were changes in expression of CsMLO1 in response to inoculation with G. ambrosiae, six detached branches of FL 58, TJ's CBD, GVA-H-19-1166-005, and GVA-H-19-1067-001 were split into two treatment groups, each with three branches. One treatment group was inoculated with a suspension of G. ambrosiae isolate 19002 (Weldon et al. 2020) conidia suspended in water, and the other was mock-inoculated with water. Samples were collected and flash frozen at 72 h postinoculation and stored at –80°C until RNA was isolated.

RNA was extracted and cDNA was synthesized using the same protocols as for sequencing the CsMLO1 transcripts. To verify that there was no genomic DNA contamination in any of the RNA samples, primers were designed to span a conserved intron (Supplementary Table S5). PCR was performed on all cDNA samples as well as genomic DNA controls of FL 58 and TJ's CBD using Q5 2X Master Mix (New England Biolabs) following the standard protocol with a 65℃ annealing temperature. All of the samples, both cDNA and control samples with genomic DNA, were run on a 1% agarose gel and produced a single band of the expected size, 510 and 2,324 bp, respectively. Based on the cDNA sequences, qPCR primers were designed to target transcripts with a full-length ORF and those with a premature stop (Supplementary Table S5). Additionally, primers for a housekeeping gene, ubiquitin (DSK2b/LOC115702664), were used as described by Guo et al. (2018) (Supplementary Table S5).

All qPCR reactions were performed using the Luna Universal qPCR Master Mix (New England Biolabs) according to the protocol described in the kit using 1 μl of standardized 10 ng/μl template cDNA, apart from the serial dilution, which had a range of concentrations. After performing qPCR on one sample of inoculated FL 58 for all primer sets with a temperature gradient, 57°C was selected as the optimal melting temperature. A 1:2 serial dilution (20 ng/μl to 0.625 ng/μl) of the same sample was prepared, and qPCR was run for all primer sets. All primer sets produced a single product based on the melt curves and behaved linearly within the range having R² between 0.987 and 0.995 and efficiencies between 93.6 and 108.1%. All three primer sets were used for qPCR on all RNA samples and a no-template control, and each primer set–sample combination was run in triplicate. C_t values were determined for all samples, and the 2^–∆∆Ct method was used to determine relative expression with the uninoculated samples of TJ's CBD as the control group. A series of Kruskal-Wallis tests were used to determine whether there was a statistically significant difference between the inoculated and uninoculated treatment groups for each genotype.

Molecular marker assay development

A high-throughput PCR allele competitive extension (PACE) assay named CsMLO1_FL58-1 was developed based on a C to G transversion identified in the predicted exon 8 (CBDRx-cs10 Chr01:21634336) sequences of the FL 58 and TJ's CBD alleles of CsMLO1 (Supplementary Table S5). PACE assays were conducted following the methods described by Toth et al. (2020). Additionally, a three-primer gel-based assay named CsMLO1_FL58-2 was developed based on the 6.8-kb insertion in FL 58 and flanking sequences. A common reverse primer is used with two forward primers, one in the insertion sequence that will only amplify when the insertion in present in CsMLO1, producing a 537-bp product, and one on the 5′ side of the insertion that will only amplify if the insertion is absent, producing a 1,000-bp product. For this assay, DNA was extracted with a DNeasy Plant Mini Kit (Qiagen). PCR was conducted using OneTaq 2X Master Mix (New England Biolabs) following the standard protocol with a 55℃ annealing temperature (Supplementary Table S5). A Kruskal-Wallis test was used to determine whether there was a statistically significant difference between the three allelic groups resolved by the CsMLO1_FL58-1 and CsMLO1_FL58-2 assays. When the result of the Kruskal-Wallis test was significant, a post hoc Dunn's test was used to determine pairwise differences between the allelic groups.

Acknowledgments

We are grateful to the research teams from the L. Smart and C. Smart labs for their excellent technical assistance on this project, especially Alexander Wares, McKenzie Schessl, Colin Day, Taylere Herrmann, Skylar Bedell, Natalie McFadden, Joel DeVries, Emily McFadden, Aaron Brooks, Erica Miller, and the farm crew at Cornell AgriTech. We are also grateful to the U.S. Department of Agriculture–Agricultural Research Service Plant Genetic Resources Unit hemp team, especially Nick Genna and Tony Barraco, and to Heather Grab, Maylin Murdock, and Savanna Shelnutt for assistance with data collection. We are also grateful to Phylos Bioscience, especially Alisha Holloway and Courtland Fowler, for assistance with genotyping and bioinformatics.