Parasitic Stress Dynamics in a Novel Host: Responses of Arctium lappa L. to Multiple Races of Orobanche cumana Wallr.

Abstract

Sunflower broomrape (Orobanche cumana Wallr. var. helianthi) is traditionally regarded as strictly host-specific, parasitizing cultivated sunflower (Helianthus annuus L.). Arctium lappa L. (burdock) was recently identified as a novel susceptible host, but whether this susceptibility is uniform across different sunflower O. cumana races remained unknown. Here, three naturally occurring A. lappa populations from north-eastern Bulgaria and a susceptible sunflower inbred line (AD-66) were challenged under controlled glasshouse conditions with three O. cumana races of increasing virulence (F, G, and G+/H), and tubercle number was recorded 45 days after inoculation (n = 50 plants per host × race combination). Race F was largely unable to establish infection on A. lappa (0–4% infected plants) while remaining highly infective on sunflower (96%), whereas races G and G+/H retained substantial infectivity on A. lappa (68–96%), though with up to two-to-three-fold lower mean tubercle numbers than on sunflower. A negative binomial generalized linear model showed that race accounted for the largest share of explained variation in tubercle number, with weaker and less robust host and host × race effects. Additive Main Effects and Multiplicative Interaction (AMMI) and genotype plus genotype-by-environment (GGE) biplot analyses further resolved this interaction into a single dominant axis, along which A. lappa was disproportionately susceptible to race G+/H and disproportionately resistant to race F. These results demonstrate that, under controlled glasshouse conditions, A. lappa is a race-dependent, alternative host for O. cumana. Whether this translates into a field-relevant reservoir capacity beyond the sunflower cropping cycle remains to be established.

1. Introduction

Orobanche cumana Wallr. (sunflower broomrape) is an obligate root holoparasite of the family Orobanchaceae and one of the most damaging biotic constraints to sunflower (Helianthus annuus L.) production across Europe and Asia [1]. Within its natural range, extending from Central Asia to south-eastern Europe, the species is primarily associated with wild Asteraceae, chiefly Artemisia spp. [2,3]. Parasitism of cultivated sunflower was first recorded in Russia in the 1890s and in Bulgaria in 1935, and the crop-adapted lineage is now formally distinguished as O. cumana var. helianthi (Tzvelev, 2015, as cited in Sharma et al. [4]), a taxon separate from the wild-type populations. The parasite’s economic and geographic significance continues to grow, with recent first reports extending its range into new sunflower-producing regions [5,6]. The sunflower-O. cumana pathosystem follows a well-characterized gene-for-gene model, in which dominant resistance genes (Or1-Or7) in the host correspond to matching avirulence loci in the parasite [2,7]. Widespread deployment of monogenic resistant cultivars has repeatedly selected for new physiological races, historically designated A through E and later followed by the more aggressive F, G, and G+/H races that have successively overcome resistance genes in most sunflower-growing regions, including Bulgaria [2,8,9,10,11]. Race evolution remains an active process, arising both through point mutation [12,13,14,15] and through recombination between distinct gene pools [16], and has recently been compounded by the first documented case of herbicide resistance in this parasite [14], underscoring the adaptive capacity of O. cumana under sustained selection pressure and the continuing need for new resistance sources [13,17].

Molecular marker studies (SSR and SNP) have shown that weedy O. cumana populations parasitizing sunflower are organized into distinct, genetically depauperate gene pools that are geographically structured, such as the Guadalquivir Valley and Cuenca gene pools in Spain or the Eastern European gene pool that includes Bulgarian populations [2,3,12]. Along the Bulgarian Black Sea coast, wild populations parasitizing Artemisia maritima display clear signatures of introgression from weedy, sunflower-adapted populations, indicating active, bidirectional gene flow between the two ecological forms that may accelerate the emergence of new virulent races in cultivated fields [3]. Morphometric comparisons of broomrape seeds collected from H. annuus, Artemisia spp., and Matricaria sp. further indicate that host identity is the principal driver of phenotypic divergence among O. cumana seed populations [11].

Although O. cumana has long been regarded as a narrowly host-specific parasite, confined to sunflower among cultivated crops and to Artemisia spp. in the wild [1], accumulating evidence indicates that its host range is broader and more dynamic than previously assumed. In the Western Himalaya of India, wild-type O. cumana was recently confirmed for the first time parasitizing Artemisia japonica, a species not previously reported as a host for any Orobanche species [4]. In Xinjiang, China, where the parasite causes substantial losses in both sunflower and tomato (Solanum lycopersicum) fields, a systematic comparison of host-derived populations demonstrated that tomato-infesting and sunflower-infesting broomrape belong to the same species and gene pool, with tomato acting as an opportunistic rather than a preferred host [18]. Independent surveys from the same region have documented further, previously unreported hosts of sunflower O. cumana, including the desert shrub Karelinia caspia [19], the noxious weed Cirsium arvense (Canada thistle) [20], and the ornamental genus Coleus [21], each representing an expansion of the known host range of the parasite beyond Helianthus and Artemisia. Most recently, controlled inoculation trials in Bulgaria demonstrated that sunflower-O. cumana can also complete its parasitic cycle on Arctium lappa L. (burdock), forming viable tubercles and emerging shoots on this species, whereas eight other tested Asteraceae taxa remained fully resistant [22]. This finding is particularly notable because A. lappa belongs to a different tribe of the Asteraceae than either Helianthus or Artemisia, suggesting that host specificity in O. cumana may be less evolutionarily constrained than previously believed and that additional, currently unrecognized alternative hosts could act as reservoirs of infection and genetic diversity within agricultural landscapes.

The discovery of A. lappa as a susceptible host raises the question of whether its response to O. cumana mirrors that of sunflower or instead represents a qualitatively different, less compatible interaction that can nonetheless still support parasite development under field conditions. Because sunflower resistance to O. cumana is governed by physiological races of differing virulence [2,10], it is essential to determine whether the ability of the parasite to infect A. lappa is uniform across races or, as in sunflower, race-dependent, and whether the intensity of parasitism, measured as the number of tubercles produced per plant, differs quantitatively between this novel host and a fully susceptible sunflower genotype. Clarifying these host × race interactions is of direct practical relevance, since perennial, widely distributed weeds such as A. lappa growing at field margins could act as an alternative green bridge, sustaining O. cumana populations and their genetic diversity independently of the sunflower cropping cycle [3,18]. We hypothesized that the susceptibility of A. lappa to O. cumana would not be uniform but would instead depend on the virulence of the challenging parasite race, mirroring the race-specific compatibility patterns already established for sunflower.

The objective of the present study was therefore to characterize, under controlled glasshouse conditions, the parasitic stress dynamics of three O. cumana races differing in virulence (F, G, and G+/H) on three natural populations of A. lappa in comparison with a fully susceptible sunflower inbred line, using tubercle number as a quantitative measure of infection intensity, and to evaluate host × race interaction patterns through negative binomial modelling, Additive Main Effects and Multiplicative Interaction (AMMI) analysis, and genotype plus genotype-by-environment (GGE) biplot analysis.

2. Results

2.1. Infection Frequency and Tubercle Production Across Hosts and Races

Across all twelve host × race combinations, tubercle counts ranged from complete absence of infection to a maximum of 20 tubercles per plant, and every combination showed pronounced overdispersion, with standard deviations frequently exceeding half of the corresponding mean (Table 1, Appendix A, Figure A1).

Table 1. Descriptive statistics of tubercle number and infection frequency of three Orobanche cumana races on sunflower and A. lappa hosts.

Host	Race	Tubercules Number					Infected Plants
		Mean	SD	Med	Min	Max	Number	I%
AD-66	G+/H	6.58	3.39	5.0	1	16	50	100.00
AD-66	G	5.38	2.55	5.0	1	12	50	100.00
AD-66	F	3.72	1.73	4.0	0	8	48	96.00
AL-2	G+/H	3.52	2.55	3.0	0	12	47	94.00
AL-2	G	2.24	2.02	1.5	0	7	41	82.00
AL-2	F	0.00	0.00	0.0	0	0	0	0.00
AL-3	G+/H	4.32	3.25	3.0	0	15	48	96.00
AL-3	G	2.36	1.96	2.0	0	6	40	80.00
AL-3	F	0.02	0.14	0.0	0	1	1	2.00
AL-4	G+/H	4.66	4.26	4.0	0	20	48	96.00
AL-4	G	1.94	2.00	1.0	0	8	34	68.00
AL-4	F	0.12	0.72	0.0	0	5	2	4.00

AD-66—susceptible sunflower (Helianthus annuus L.) inbred line; AL-2, AL-3, AL-4—Arctium lappa L. populations; F, G, G+/H—Orobanche cumana races.

The susceptible sunflower inbred line AD-66 supported the highest overall level of parasitism among the four hosts tested. Mean tubercle numbers on AD-66 were 6.58 ± 3.39 for race G+/H, 5.38 ± 2.55 for race G, and 3.72 ± 1.73 for race F, and infection frequency remained at or near 100% irrespective of race (100%, 100%, and 96% for G+/H, G, and F, respectively; Table 1). Statistical comparison among races within this host (likelihood-ratio-based pairwise contrasts; Table 2 and Table 3) indicated that G+/H and G did not differ significantly from one another (both assigned letter “a”), whereas F produced significantly fewer tubercles (letter “b”), pointing to a two-tier virulence hierarchy on sunflower in which the two most aggressive races behave similarly.

Table 2. Likelihood ratio tests from the negative binomial generalized linear model evaluating the effects of host, race and their interaction on Orobanche cumana tubercle number.

Effect	All Hosts			A. lappa Hosts
	LR χ2	df	p	LR χ2	df	p
Host	200.0865	3	<0.001	1.7918	2	0.408
Race	296.1922	2	<0.001	370.7101	2	<0.001
Host × Race	209.6176	6	<0.001	11.9862	4	0.018

Table 3. Model-estimated marginal means, standard errors, and 95% confidence intervals for Orobanche cumana tubercle number across host × race combinations (negative binomial GLM), with compact letter displays comparing races within each host and hosts within each race under both Wald- and likelihood-ratio-based inference.

Host	Race	Marginal Mean	SE	Lower 95% CI	Upper 95% CI	Wald Test		LR Test
						Host Letters Within Race	Race Letters Within Host	Host Letters Within Race	Race Letters Within Host
AD-66	G+/H	6.580	0.569	5.302	8.165	a	a	a	a
AD-66	G	5.380	0.486	4.294	6.741	a	a	a	a
AD-66	F	3.720	0.368	2.905	4.764	a	b	a	b
AL-2	G+/H	3.520	0.354	2.738	4.525	b	a	b	a
AL-2	G	2.240	0.259	1.678	2.990	b	b	b	b
AL-2	F	0.000	0.000	0.000	-	ab	ab	b	c
AL-3	G+/H	4.320	0.411	3.406	5.479	b	a	b	a
AL-3	G	2.360	0.268	1.777	3.134	b	b	b	b
AL-3	F	0.020	0.020	0.002	0.244	b	c	bc	c
AL-4	G+/H	4.660	0.435	3.690	5.884	b	a	ab	a
AL-4	G	1.940	0.236	1.433	2.627	b	b	b	b
AL-4	F	0.120	0.050	0.043	0.337	b	c	c	c

Race letters within host: races sharing a letter within the same host do not differ significantly (Holm-adjusted, p < 0.05). Host letters within race: hosts sharing a letter within the same race do not differ significantly. Wald- and LR-based letters are reported separately because the AL-2 × race F combination (0/50 infected) constitutes a complete-separation case for which the Wald-based upper confidence bound is not estimable; the LR-based comparison (Supplementary Tables S10 and S11) is unaffected by this artefact and should be treated as the more reliable of the two for this combination.

A markedly different pattern emerged on the three Arctium lappa populations. Averaged across AL-2, AL-3, and AL-4, model-estimated mean tubercle numbers were 4.17 for race G+/H and 2.18 for race G—consistently lower than the corresponding sunflower values (Table 1 and Table 3). Unlike the two-tier response observed in AD-66, all three A. lappa populations displayed a fully resolved three-tier virulence hierarchy under likelihood-ratio-based pairwise comparisons, with G+/H, G, and F assigned to three statistically distinct groups (“a”, “b”, “c”) in every population (Table 3); under the Wald-based test, this distinction is less sharp for AL-2 × race F specifically, reflecting the complete-separation issue noted for this near-zero-infection combination. This indicates that, on A. lappa, the parasite races are more clearly discriminated from one another than on the fully susceptible sunflower host, where the two most virulent races reach a similar performance ceiling.

Race F was essentially unable to infect A. lappa, and no tubercles were observed on the AL-2 plants; only 2% of AL-3 and 4% of AL-4 seedlings were infected. In contrast, race F resulted in 96% infection of sunflower. By comparison, races G and G+/H displayed substantial infectivity on all three A. lappa populations, with infection frequencies ranging from 68% (AL-4, race G) to 96% (AL-3 and AL-4, race G+/H). This demonstrates that the ability of O. cumana to parasitize this novel host is race-dependent.

Among the three A. lappa populations, subtle differences in mean tubercle number and infection frequency were also apparent, although these population-level differences were not statistically significant when formally tested. AL-3 and AL-4 supported slightly higher mean tubercle numbers under race G+/H (4.32 ± 3.25 and 4.66 ± 4.26, respectively) than AL-2 (3.52 ± 2.55), and a comparable ranking was reflected, in reverse, in the population-level infection frequency for race G, which was highest in AL-2 (82%) and lowest in AL-4 (68%). Notably, the single highest tubercle count recorded under race G+/H occurred on an AL-4 plant (20 tubercles), exceeding the sunflower maximum (16 tubercles). This single observation should not be read as evidence that A. lappa can match sunflower in infection severity: it represents one plant among 50 in this combination, within a distribution whose mean (4.66 ± 4.26) remained well below the corresponding sunflower value (6.58 ± 3.39), and its interpretation is further limited by the single-container design (Section 4.1). It does, however, indicate that under the most virulent race tested, individual A. lappa plants are occasionally capable of supporting heavy parasite loads, even though this is not representative of the population as a whole.

The violin plots (Figure 1) and heatmap (Figure 2) illustrate these patterns, with tubercle distributions for race F on A. lappa reduced almost to zero in contrast to the broader, right-skewed distributions displayed by G and G+/H.

Figure 1. Distribution of raw per-plant Orobanche cumana tubercle counts across all 600 plants in infested soil (violin plots), grouped by host × race combination. Each combination reflects 50 plants grown within a single shared container.

Figure 2. Heatmap of mean Orobanche cumana tubercle numbers across host–race combinations.

2.2. Effects of Host, Race, and Their Interaction on the Level of Parasitism

Table 2 presents the results of the negative binomial generalized linear model used to formally partition the effects of host genotype, Orobanche cumana race, and their interaction on tubercle number. When all four host genotypes were considered jointly, the plant-level negative binomial model detected statistically significant effects of host, race, and host × race interaction (likelihood ratio χ² = 200.09, df = 3; χ² = 296.19, df = 2; and χ² = 209.62, df = 6, respectively; all p < 0.001), consistent with the patterns apparent in the descriptive statistics and graphical summaries (Table 1, Figure 1 and Figure 2).

To assess the relative magnitude, rather than merely the statistical significance, of each effect, McFadden’s pseudo-R² was calculated for each model term (Supplementary Methods, Section S1.4). Race alone accounted for a larger share of explained deviance than host alone (pseudo-R² = 0.066 vs. 0.030), and this ranking was consistent with the AMMI variance partitioning reported below (Section 2.3), in which race explained a larger percentage of total sum of squares than host (33.42% vs. 22.53%; Table 4). Race was therefore the larger source of variation by these model-based effect-size measures, not solely by virtue of its likelihood-ratio chi-square value, which is not directly comparable across terms with unequal degrees of freedom.

Table 4. AMMI analysis of variance for the interaction between host and Orobanche cumana race based on square-root transformed tubercle numbers √(x + 0.5).

Source	SS	df	MS	F	Sig	SS%
Host	10.775	3	3.592	109.681	<0.001	22.53
Race	15.984	2	7.992	244.051	<0.001	33.42
Host × Race	1.815	6	0.302	9.237	<0.001	3.79
IPC1	1.681	4	0.42	12.833	<0.001	3.51
IPC2	0.134	2	0.067	2.044	0.1305	0.28
Error	19.255	588	0.033	-	-	-
Total	47.83	599	-	-	-	-

Because the AMMI error term (df = 588) treats all 600 plants as independent observations, the F- and p-values reported above are subject to the same container-level pseudoreplication risk quantified for the negative binomial model. These statistics should be read as descriptive rather than confirmatory.

The host effect, while highly significant, was smaller in magnitude than the race effect and was driven predominantly by the contrast between the fully susceptible sunflower line AD-66 and the three Arctium lappa populations, which as a group supported consistently lower tubercle numbers (Table 1). This is in line with the reduction in mean tubercle production observed on A. lappa relative to sunflower for both race G and race G+/H, and the absence of infection recorded for race F on A. lappa.

The plant-level negative binomial model also detected a statistically significant host × race interaction, indicating that the observed race-response pattern was not identical across host genotypes. However, this interaction was substantially less robust than the host and race main effects under the pseudoreplication diagnostics and alternative two-part models described below. We therefore used AMMI and GGE analyses primarily to explore and visualize the structure of the observed interaction pattern rather than to provide independent inferential confirmation.

To evaluate whether the interaction effect was driven solely by the contrast between sunflower and Arctium, or whether it also reflected true differential responses among the three A. lappa populations, the negative binomial model was refitted using only the A. lappa subset of the data. In this restricted analysis, race remained highly significant (χ² = 370.71, df = 2, p < 0.001), and the host × race interaction remained statistically significant but comparatively modest (χ² = 11.99, df = 4, p = 0.018; Table 2). However, the host effect itself, reflecting differences among the three A. lappa populations alone, was not statistically significant once sunflower was excluded from the analysis (χ² = 1.79, df = 2, p = 0.408). This indicates that, while race remains an overwhelming determinant of tubercle production on A. lappa and some degree of race-dependent variation exists among the three populations, the three A. lappa populations do not differ significantly in their overall level of susceptibility once the dominant sunflower-versus-Arctium contrast is removed. This qualifies our earlier suggestion that the host × race interaction reflects a strong, independent population-level effect. The modest but significant interaction more plausibly reflects subtle, race-specific differences among populations superimposed on an otherwise uniform overall susceptibility level.

Pairwise comparisons of estimated marginal means, adjusted for multiple testing using the Holm procedure, corroborated these overall model results and are summarized by the compact letter displays reported in Table 3.

Because tubercle counts contained a substantial proportion of zeros, the standard NB-GLM was compared against hurdle-NB and zero-inflated NB alternatives (Table S5). AIC favoured the hurdle-NB model (2095.8) over the standard NB-GLM (2104.8) and zero-inflated NB (2100.6), whereas BIC favoured the standard NB-GLM (2162.0) over both zero-augmented alternatives (2205.7 and 2210.5), reflecting their substantially larger parameter count (25 vs. 13). A two-part model separating infection incidence from severity among infected plants showed host and race remained highly significant determinants of both components (all p < 0.001), but the host × race interaction was not significant in either component (infection incidence: χ² = 5.11, df = 6, p = 0.530; severity given infection: χ² = 4.73, df = 5, p = 0.449), consistent with the reduced robustness of this term already noted.

2.3. AMMI Analysis of the Host × Race Interaction

The AMMI analysis of variance (Table 4), based on square-root transformed tubercle counts, provided a complementary partitioning of the sources of variation identified by the generalized linear model and allowed the host × race interaction to be decomposed into interpretable components. As in the negative binomial model, host and race main effects were both highly significant and accounted for the largest shares of total variation (host: SS = 10.775, F = 109.681, p < 0.001, 22.53% of total sum of squares; race: SS = 15.984, F = 244.051, p < 0.001, 33.42% of total sum of squares). Together, the two main effects accounted for more than half of the total variance in the dataset, confirming that host identity and, even more so, parasite race were the dominant drivers of tubercle production.

The host × race interaction, although comparatively modest in relative terms (SS = 1.815, 3.79% of total sum of squares), was nonetheless statistically significant (F = 9.237, p < 0.001), yielding a pattern consistent with the interaction detected by the negative binomial model, although subject to the same container-level pseudoreplication constraint. Partitioning of this interaction sum of squares into interaction principal component axes showed that the first axis, IPC1, was highly significant (F = 12.833, p < 0.001) and captured the great majority of the interaction variance, accounting for 3.51 of the 3.79 percentage points of total sum of squares attributable to the interaction, equivalent to approximately 93% of the interaction sum of squares. The second axis, IPC2, was not statistically significant (F = 2.044, p = 0.1305) and explained only the residual 0.28% of total variance. The predominance of IPC1 therefore motivated the use of the AMMI1 biplot as a descriptive visualization of the observed interaction structure.

It is worth emphasizing that, although statistically significant, the host × race interaction accounted for only 3.79% of total sum of squares—several-fold smaller than either main effect (host: 22.53%; race: 33.42%; Table 4). Thus, the low p-value should not be interpreted as evidence that the interaction is practically large: its contribution to total variation was modest relative to either main effect. This distinction is consistent with the reduced robustness of the interaction term already noted under the pseudoreplication diagnostics (Section 4.2) and the two-part incidence/severity model (Section 2.2), and it should temper any interpretation of host × race compatibility patterns as a dominant feature of this pathosystem.

In the AMMI1 biplot (Figure 3), along the main-effect axis, races were clearly separated according to their overall virulence, with G+/H positioned furthest to the right, G in an intermediate position, and F furthest to the left. Among hosts, AD-66 occupied the most positive position on the main-effect axis, consistent with its higher overall tubercle production relative to the three Arctium lappa populations, which clustered together at lower, more similar main-effect values.

Figure 3. AMMI1 biplot of the host × race interaction for Orobanche cumana tubercle development. Horizontal axis (OCTN): additive main effect of each host or race on square-root-transformed tubercle number (√(x + 0.5)). Vertical axis (IPC1): first interaction principal component score. Blue circles = hosts; green diamonds = races.

The distribution of IPC1 scores revealed a clear pattern of differential interaction between host groups and races. Race G+/H and the three A. lappa populations shared IPC1 scores of the same, positive sign, indicating a positive interaction between this host group and this race: A. lappa plants performed relatively better against G+/H than would be predicted from the main effects alone, consistent with G+/H remaining the only race capable of producing very high tubercle numbers on individual A. lappa plants (Table 1). Conversely, race F and the three A. lappa populations displayed IPC1 scores of opposite sign, indicating a negative interaction that reinforced the already very low main effect of race F on this host group, consistent with the near-complete absence of infection recorded for race F on A. lappa (Table 1).

The sunflower line AD-66 showed the opposite pattern to that of A. lappa. Its IPC1 score was strongly negative, placing it on the same side of the biplot as race F. This positive interaction term accounts, in the AMMI decomposition, for the fact that F, despite being the least virulent race overall, achieved a comparatively high infection frequency (96%) and low tubercle production on sunflower, exceeding what would be expected from the race main effect alone. Conversely, the opposite signs of the IPC1 scores for AD-66 (negative) and G+/H (positive) indicate a mild negative interaction between this host and its most virulent challenger, tempering, without reversing, the already high tubercle numbers produced by G+/H on sunflower.

2.4. GGE Biplot Analysis: Race Discrimination and Host Ranking

The genotype plus genotype-by-environment (GGE) biplot analysis, performed on the same square-root transformed tubercle data used for the AMMI analysis, provided two complementary perspectives on the host × race interaction: the discriminating ability and representativeness of individual races (Figure 4), and the ranking of hosts according to their overall response across races (Figure 5). In both biplots, the first principal component explained 98.26% of the total variation among host × race means, while the second principal component accounted for only 1.38%, indicating that the observed pattern was predominantly structured along a single axis. In the discriminativeness-versus-representativeness view (Figure 4), each O. cumana race is represented as a vector radiating from the biplot origin, with vector length indicating the discriminating ability of that race, that is, its capacity to differentiate among host genotypes, and the angle between each vector and the average-race axis indicating how representative that race is of the overall pattern of host discrimination.

Figure 4. Discriminating ability and representativeness of Orobanche cumana races (GGE biplot; Scaling = 0, Centering = 2, SVP = 2). PC1 and PC2 explain the percentages of host × race mean variation indicated on each axis. Vector length from the origin reflects a race’s discriminating ability among hosts; the angle between a race vector and the average-race axis (heavy line) reflects how representative that race is of the overall discrimination pattern. Vector length and angle are descriptive ordination properties, not evidence of a physiological or molecular mechanism.

Figure 5. GGE biplot ranking of hosts by response to Orobanche cumana races. Hosts are ranked by projection onto the average-race axis (heavy line), approximating mean performance across races.

Race F produced the longest vector among the three races tested, identifying it as the most discriminating environment in the GGE sense, despite being the least virulent race overall. This apparent paradox is readily explained by the extreme contrast in host response already documented for race F: while it retained a high infection frequency on sunflower (96%), it was almost entirely unable to infect any of the three A. lappa populations (0–4% infection frequency; Table 1), producing the largest relative spread among host responses of any race tested. Races G and G+/H produced comparatively shorter vectors, reflecting their more homogeneous performance across the four hosts: both races achieved moderate-to-high infection frequencies on all four genotypes (68–100%), and therefore discriminated less sharply between sunflower and A. lappa than race F did.

The angular relationships among the three race vectors in Figure 4 were relatively small, indicating that G, G+/H, and F, despite differing in discriminating power, tended to rank hosts in a broadly similar order rather than producing contradictory patterns of host response, a result consistent with the absence of any host reversing its relative ranking, that is, no crossover interaction, across races in the descriptive data (Table 1). This pattern is consistent with the AMMI representation and suggests that the observed host × race interaction is better characterized as quantitative and non-crossover, reflecting differences in the magnitude of response rather than changes in host ranking across races.

The ranking-hosts biplot (Figure 5) uses the same underlying principal component scores but adds an average-race coordinate axis, allowing hosts to be ranked according to their projection onto this axis, which approximates their overall mean performance across the three races tested, while their perpendicular distance from the axis reflects the degree to which their response departed from this average, race-independent expectation.

Sunflower (AD-66) was clearly separated from the cluster formed by the three A. lappa populations along this ranking axis, consistent with its higher tubercle production documented throughout the descriptive, GLM, and AMMI analyses. Among the three A. lappa populations, AL-2, AL-3, and AL-4 occupied closely neighbouring, though not identical, positions in the biplot, consistent with the absence of a significant overall population effect in the A. lappa-only negative binomial model (p = 0.408), while allowing for the subtle race-specific differences reflected in the modest host × race interaction (p = 0.018; Table 2).

Taken as a whole, the GGE biplot analyses provide a descriptive visualization of the host × race patterns identified in the preceding analyses. Together with the results reported in the preceding sections, these patterns indicate that, under the glasshouse conditions tested, A. lappa exhibited race-dependent compatibility with O. cumana, with susceptibility to the more virulent races approaching, but not matching, that observed in sunflower.

3. Discussion

The present results extend the initial report of Arctium lappa as a novel host of O. cumana [22] by demonstrating that its susceptibility is not a fixed property of the host but a race-dependent outcome, ranging from near-complete resistance to race F to substantial, though consistently lower than sunflower-level, susceptibility to races G and G+/H. This race-specificity confirms that the host range of O. cumana is wider than previously thought and not confined to Artemisia in the wild and cultivated Helianthus [1,4,18,19,20,21].

The observation that races differed sharply in their ability to parasitize A. lappa, while none of them showed a corresponding qualitative shift on sunflower, parallels findings in other O. cumana populations in which changes in the specificity of seed germination in response to host-derived stimulants have been directly linked to host-range expansion. Dor et al. [23] described a novel O. cumana race able to parasitize tomato because its seeds, unlike typical O. cumana, responded to strigolactones in addition to dehydrocostus lactone, the principal germination stimulant exuded by sunflower roots [24]. A comparable, though not yet characterized, shift in stimulant perception may explain the ability of races G and G+/H to establish tubercles on A. lappa, a host phylogenetically distant from both Helianthus and Artemisia. It is possible that such shifts are mediated through changes in the KAI2d receptor family, which has recently been shown to govern recognition of strigolactones and sesquiterpene lactones in O. cumana seeds more generally [25]. Comparative genomic analyses of Orobanche species with contrasting host ranges have similarly identified lineage-specific expansions in gene families associated with host recognition [26], offering a molecular framework for testing—rather than explaining—the race-dependent compatibility pattern reported here in future work.

The clear separation of races according to virulence on both hosts is consistent with the gene-for-gene framework long established for the sunflower-O. cumana pathosystem [2,10], recently refined through direct mapping of an avirulence locus segregating between race F and race G populations [27], and with accumulating evidence that new, more aggressive races arise both through point mutation [12] and through recombination between distinct gene pools, a process associated with elevated intrapopulation genetic diversity in highly virulent Spanish populations [28]. Comparable population-level surveys from Serbia [29] and Spain [30] similarly show that gains in virulence against sunflower are frequently accompanied by measurable shifts in gene-pool composition; whether the capacity to parasitize A. lappa is associated with a specific genetic background within the G and G+/H race complex used here, or is instead a more general property shared by highly virulent isolates regardless of geographic origin, remains an open question that will require genotyping of these Bulgarian races alongside reference populations such as those recently characterized in Serbia [29] and Spain [27,30].

The pattern uncovered here, in which host compatibility is race-, rather than species-dependent, has clear parallels elsewhere in the Orobanchaceae. Cryptic, host-associated genetic divergence has been documented among broomrape lineages parasitizing different host genera in the Western Hemisphere [31], and Moroccan populations of Orobanche foetida have been shown to be evolving from parasitism of wild legumes toward compatibility with cultivated vetch [32], illustrating that host shifts driven by within-species genetic variation, rather than by speciation, may be a general feature of broomrape biology rather than a peculiarity of O. cumana. Because host specificity in Orobanchaceae is primarily governed by the chemical dialogue between host root exudates and parasite seed receptors [24,33], future work should determine whether A. lappa roots release germination stimulants recognized preferentially by races G and G+/H, which would provide a mechanistic explanation for the AMMI and GGE results reported here.

From a methodological standpoint, AMMI and GGE biplot analysis, tools originally developed for genotype-by-environment interaction in multi-environment yield trials with independently replicated genotype × environment combinations [34,35], were applied here to a host × parasite-race dataset in which each combination was represented by a single, non-independently replicated container. We therefore present the resulting decomposition of the host × race interaction as an exploratory visualization of the observed pattern rather than as independently validated statistical evidence of interaction structure, consistent with the reduced robustness of the interaction term already noted in Section 2.2. Applying this framework to independently replicated host × race data in future work would allow the single-axis pattern identified here to be formally tested, and may usefully complement conventional differential-host virulence assays [2,10] in future characterizations of O. cumana race structure.

From an applied perspective, the demonstration that A. lappa can sustain substantial O. cumana infection under the most virulent races evaluated reinforces concerns, already raised for tomato in China [18] and for wild Artemisia populations along the Bulgarian Black Sea coast [3], that alternative hosts growing at field margins could possibly function as a green bridge, maintaining inoculum and potentially facilitating gene flow between wild and weedy O. cumana populations independently of the sunflower cropping cycle, with implications for the broader threat this parasite poses to food security [6]. These conclusions should be interpreted with the caveat that the present experiment was conducted under controlled glasshouse conditions with only three races and three A. lappa populations from a single geographic region; field validation, a broader race and population panel, and molecular characterization of the A. lappa-infecting isolates, following approaches recently applied to wild-derived resistance sources in sunflower [36], represent important next steps toward establishing the epidemiological significance of this novel host relationship.

4. Materials and Methods

4.1. Plant Material, Experimental Design and Inoculation

Three naturally occurring populations of Arctium lappa L. collected from northeastern Bulgaria and one susceptible sunflower (Helianthus annuus L.) inbred line were evaluated for their response to Orobanche cumana Wallr. infestation. The A. lappa populations originated from Dobrich (AL-2; 43.591299° N, 27.857399° E), Varna (AL-3; 43.268915° N, 27.870598° E), and Debrene (AL-4; 43.388185° N, 27.855436° E). Each population sample was bulk-collected from multiple individual plants within the respective wild stand and is genetically heterogeneous. Original seed samples were cleaned of broken, damaged, and non-seed material. No further pre-sowing treatment was applied. The sunflower genotype (AD-66) is an inbred line, lacking Or resistance genes against O. cumana [37]. The three A. lappa populations and the susceptible sunflower line were treated as four independent host genotypes throughout the study.

Three O. cumana races, designated G+/H, G and F, were used for artificial inoculation. These races are established reference populations of documented field origin, previously identified through standard differential host assays (according to methodology described in [38]) and routinely used in the sunflower breeding programme at the Dobrudzha Agricultural Institute for the screening of breeding material. Seed stocks of each race were maintained in sealed glass containers under standard storage conditions between experiments.

The experiment was conducted in a regulated glasshouse with controlled temperature and ventilation. Relative humidity was not independently controlled or recorded. Plants were grown in rectangular metal containers measuring 30 × 30 × 7 cm (length × width × height), without drainage holes. Prior to the experiment, this soil source was screened for the absence of a native O. cumana seed bank by growing the highly susceptible sunflower genotype AD-66 in the unamended substrate for 60 days. No tubercle formation was observed, confirming the substrate to be free of viable O. cumana propagules before intentional inoculation. Irrigation was applied manually as needed, based on visual assessment of substrate moisture, without a fixed schedule or recorded volume. Within each container, seeds were sown in a 5-row × 10-column arrangement (n = 50 sowing positions per container). Given the high germination rate of the seed lots used (100%), no post-emergence thinning was required.

Each container represented one host × race combination and contained 50 plants arranged in a 5-row × 10-column grid. A total of twelve host × race combinations were evaluated, each in a single container, resulting in 600 plants in total. Because each host × race combination was represented by a single container, the container constitutes the true experimental unit for this design, and the 50 plants within it are correlated subsamples rather than independent replicates. The implications of this hierarchical structure for statistical inference, and the diagnostic analyses undertaken to quantify and address it, are described below.

The growth substrate consisted of a 1:1 (v/v) mixture of soil and sand. O. cumana seeds were thoroughly mixed with the substrate before sowing to ensure uniform parasite distribution throughout the rooting zone. The amount of inoculum applied followed the procedure described by Panchenko [39], which was applied without modification.

Plants were maintained under identical environmental conditions throughout the experiment. Container positions within the glasshouse were assigned completely at random, with no formal blocking structure. Containers remained in their assigned positions for the duration of the experiment and were not rotated. Glasshouse temperature was maintained at 22–25 °C during the day and 18–20 °C at night under natural photoperiod and light conditions; microenvironmental gradients were not separately monitored. Forty-five days after sowing, plants were carefully removed from the substrate, and the root systems were washed to expose attached parasite structures. Only visible O. cumana tubercles were counted. The number of tubercles per plant was recorded individually, resulting in fifty plant-level observations for each host × race combination.

4.2. Statistical Analysis

The single-container-per-treatment design reflects a well-documented constraint of controlled-environment experiments, in which the growth chamber or glasshouse unit—rather than the individual plants within it—frequently constitutes the true experimental unit [40,41]. Because each host × race combination in this experiment was represented by a single, non-independently replicated container, the diagnostic and modelling strategy described below was designed to quantify, rather than eliminate, the statistical consequences of this design. These diagnostics do not substitute for independent container replication. Results should accordingly be interpreted as provisional pending confirmatory replication.

To assess whether the fixed-effects results reported below are robust to the non-independence of plants within a container, three complementary diagnostics were performed and are reported in full in the Supplementary Methods (Sections S1.1–S1.3; Tables S1–S4): estimation of the intraclass correlation and design effect for plants sharing a row within a container; a mixed model testing for a row-within-container random effect; and a Monte Carlo simulation comparing plant-level and container-level inference across a range of intraclass correlations spanning the estimated value for this dataset.

The number of O. cumana tubercles (OCTN) per plant was used as the primary response variable. Descriptive statistics, including mean, standard deviation, median, minimum, maximum, and the percentage of infected and non-infected plants, were calculated for each host × race combination.

Because tubercle counts represent discrete count data with numerous zero observations and evident overdispersion, treatment effects were analyzed using a negative binomial generalized linear model (NB-GLM). The model included host genotype, broomrape race, and their interaction as fixed effects:

$$\mathrm{Tubercles}\sim \mathrm{Host}+\mathrm{Race}+\mathrm{Host}\times \mathrm{Race}$$

The significance of model terms was evaluated using likelihood ratio chi-square (LR χ²) tests. Pairwise comparisons among estimated marginal means were performed using Holm-adjusted multiple comparisons.

Because tubercle counts contain a substantial proportion of zero observations, the standard NB-GLM was additionally compared against hurdle-NB and zero-inflated NB alternatives using AIC and BIC, and a two-part model separating infection incidence (binomial) from tubercle number among infected plants (negative binomial) was fitted to test whether these represent distinct biological processes. Full specifications and results are reported in the Supplementary Methods (Section S1.4, Table S5).

The primary negative binomial model converged without warnings (θ = 4.51, SE = 0.71, 95% CI: 3.12–5.91); DHARMa diagnostics showed no evidence of residual misspecification (dispersion test p = 0.936; KS test p = 0.992; outlier test p = 0.337), and Cook’s distance identified no observation with undue influence (max = 0.335; Table S13).

To further investigate the host × race interaction pattern, an Additive Main Effects and Multiplicative Interaction (AMMI) analysis was performed. Prior to AMMI analysis, tubercle counts were square root transformed according to √(x + 0.5) in order to stabilize variance and improve compliance with ANOVA assumptions for count data. The interaction sum of squares was partitioned into Interaction Principal Component Axes (IPCAs), and the significance of each component was assessed using F-tests. The first interaction principal component (IPC1), which explained the majority of the interaction variation, was subsequently used for biological interpretation.

Genotype plus genotype-by-environment (GGE) biplot analysis was conducted using the transformed data to visualize host responses across broomrape races and to evaluate the discriminative ability and representativeness of individual races. Mean tubercle numbers were additionally illustrated using heatmaps, whereas the distribution of observations within each host × race combination was visualized using violin plots.

Because AMMI and GGE require independently replicated genotype × environment combinations, which this single-container design does not provide, the analyses are presented as an exploratory decomposition of the host × race means rather than as independently validated inferential results.

Data management, descriptive statistics, the violin plots, and heatmap were produced in Python (version 3.14.0) using pandas, NumPy, and Matplotlib. The AMMI analysis of variance was likewise implemented in Python (NumPy, SciPy), with the host × race interaction matrix decomposed by singular value decomposition into interaction principal component axes; the corresponding AMMI and GGE biplots were generated separately in R (in RStudio, version 2025.05.0+496) using the metan package [42]. All remaining inferential statistical modelling was performed in R. Negative binomial generalized linear models were fitted using the MASS package, and the significance of host, race, and host × race effects was evaluated through likelihood ratio chi-square tests. Pairwise comparisons were derived from a saturated cell-means negative binomial model; compact letter displays based on Holm-adjusted likelihood-ratio tests were produced using the multcompView package. Comparison of the standard NB-GLM against hurdle-NB and zero-inflated NB alternatives, and the two-part incidence/severity model, were implemented using the pscl package together with MASS. Diagnostics for the pseudoreplication structure—the row-within-container mixed model, residual checks, and spatial autocorrelation testing—were conducted using glmmTMB, DHARMa, and ape, and the Monte Carlo comparison of plant-level versus container-level inference was implemented as a custom parallelized simulation using the future and future.apply packages. Diagnostics for the main negative binomial model (dispersion parameter, simulated residual checks, and identification of influential observations) were conducted using DHARMa and base R functions. Statistical significance was declared at p < 0.05.

5. Limitations

Several aspects of the experimental design constrain the interpretation of these results. First, each host × race combination was represented by a single container housing 50 plants—the container, not the individual plant, is the true experimental unit. Although the host and race main effects were robust to the estimated degree of within-container clustering, the host × race interaction was less robust, and these diagnostic analyses cannot substitute for independently replicated containers. The results should therefore be regarded as provisional pending such replication. Second, the experiment was conducted as a single glasshouse run, so day-to-day or seasonal environmental variation cannot be distinguished from treatment effects. Third, relative humidity was not independently controlled or recorded, and irrigation was applied manually without a fixed schedule. Fourth, the three A. lappa populations originated from a single geographic region of north-eastern Bulgaria, and the race panel was limited to three of the several physiological races known in Bulgarian O. cumana populations—neither sampling captures the full genetic diversity of either species. Fifth, only one sunflower genotype (AD-66) was included as a reference host, so the sunflower comparison reflects a single point on the range of cultivated susceptibility rather than its full diversity. Finally, the experiment measured only tubercle formation 45 days after artificial inoculation; parasite emergence, flowering, viable seed production, inoculum persistence, and field transmission were not assessed, all of which are required to establish A. lappa as an epidemiologically meaningful reservoir. These limitations constrain the inferential strength and generalizability of the observed race-dependent compatibility pattern, define the scope within which it should be interpreted, and motivate the priorities identified in the Section 3 and Section 6.

6. Conclusions

This study provides the first quantitative, multi-race characterization of Orobanche cumana parasitism on Arctium lappa. Under the glasshouse conditions tested here, tubercle formation on three A. lappa populations was consistently lower than on a fully susceptible sunflower line, and this difference was strongly race-dependent: race F almost failed to parasitize A. lappa, while races G and G+/H produced tubercle numbers approaching, though still below, those recorded on sunflower. Negative binomial modelling identified strong host and race effects, whereas evidence for the host × race interaction was less robust across alternative analyses. AMMI and GGE analyses suggested a coherent, predominantly single-axis structure in the observed interaction pattern, but, given the single-container design, these results should be regarded as exploratory rather than independently validated evidence. These results demonstrate that A. lappa’s compatibility with O. cumana, under the controlled glasshouse conditions tested, is race-dependent rather than a fixed trait of the host. This conclusion rests on a single-container-per-treatment design and awaits confirmation through independently replicated trials. Whether A. lappa can sustain broomrape populations independently of the sunflower cropping cycle under field conditions remains an open epidemiological question, one that lies beyond the scope of the present study and warrants dedicated field-based and molecular investigation.