Contrasting reproductive strategies and their functional consequences in two populations of Opuntia robusta (Cactaceae) Authors: Citlalli H S Mendoza-Reyes, Adán Miranda-Pérez, Juan Núñez-Farfán, Mariusz Krzysztof Janczur Published: 2026 ABSTRACT AbstractHermaphroditism is the predominant sexual condition in angiosperms, but populations of a single species can diverge markedly in their reproductive allocation and in the expression of their mating system. Opuntia robusta provides a suitable system to examine this divergence, as its populations differ in floral morphology, sexual expression, and reproductive performance. We compared pollen–ovule ratios, early seed performance, and floral visitors in two populations from central Mexico that occur along an altitudinal gradient and differ in their mating‑system classification. Reproductive allocation varied between populations, with differences in ovule production and in the relative investment in male and female functions. The populations also differed in their mating system: the hermaphroditic population expressed functional autogamy, whereas the trioecious population exhibited facultative autogamy. Germination assays revealed consistent differences between populations, and pollination treatments confirmed that autonomous self‑pollination can produce viable seeds in both. Floral‑visitor communities were contrasting: one population was visited by a diverse assemblage of bees, whereas the other was dominated by a single genus. Statistical analyses revealed functional associations between visitors and germination, highlighting the consistent role of Macrotera sp. and the possible mechanical facilitation of autogamy by Nitidulidae in the hermaphroditic population. Taken together, these results demonstrate that O. robusta populations differ in their reproductive allocation, early offspring performance, and visitor assemblages, indicating functional divergence in mating‑system expression at short geographic scales. This population‑level variation may influence the stability of hermaphroditism and the expression or functional erosion of trioecy within the species.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12870-026-09625-0.Keywords: Opuntia robusta, Mating system variation, Pollen–ovule ratio, Germination, Reproductive allocation, Pollinator communities, Population divergence Introduction In angiosperms, hermaphroditism is considered the ancestral sexual condition, and from it a wide diversity of reproductive systems has evolved from it [1–3]. This diversity reflects how different mating systems shape the transmission of genetic variation and population responses to ecological and evolutionary pressures [4, 5]. Understanding the mechanisms that generate this variation, particularly within a single species, is essential for interpreting how reproductive strategies diversify across contrasting ecological contexts. Differences in mating systems not only determine the genetic combinations received by seeds but also influence their early performance, including germination and establishment. Seed germination represents a critical stage in the plant life cycle, as it determines seedling establishment and influences population dynamics. Germination outcomes may vary due to genetic differences among populations, environmental conditions during seed development, and the pollination mode that produced them [6–8]. Although self‑pollination may reduce germination through inbreeding depression [1, 6], cross‑pollination often enhances early performance [9]. However, these patterns are not universal: natural self‑pollination can produce high‑quality seeds in some species, whereas manual pollinations may reduce seed performance due to floral damage or shifts in resource allocation [10]. Differences in germination among populations have also been documented, underscoring the importance of considering intraspecific variation when evaluating reproductive success [8]. Despite the relevance of these factors, few studies have jointly examined how population origin and the conditions under which pollination occurs influence germination using approaches that account for among‑plant variation. Mixed‑effects models provide an appropriate framework for this purpose, as they allow the evaluation of fixed effects while incorporating heterogeneity among individuals, a component often overlooked in germination studies [11], underscoring the importance of considering intraspecific variation when evaluating reproductive success, as documented in other systems where seed traits and population context strongly influence early performance [12, 13]. These considerations highlight the importance of studying species that exhibit population‑level variation in reproductive traits. Opuntia robusta is a prickly pear cactus widely distributed in central Mexico that reproduces both clonally and sexually and exhibits notable variation in reproductive traits [14–16]. Previous studies have documented differences in sexual expression among populations, including the presence of trioecy in some localities [17, 18]. The species also shows variation in growth form that may be associated with differences in ploidy level, although the cytotypes present have not yet been identified with certainty [19, 20]. Together, these features make O. robusta a suitable system for examining population divergence in reproductive strategies, integrating sexual, morphological, and potential cytotypic variation within a single ecological framework. Sex determination in Opuntia robusta is based on easily distinguishable floral traits. Functionally male flowers exhibit abundant pollen production and a reduced or sterile ovary, whereas female flowers show a well-developed ovary and receptive stigmas but have atrophied or non-dehiscent anthers. Hermaphroditic flowers display fully functional male and female structures [17]. Pollination in cacti is typically mediated by insects, particularly bees (Insecta: Apoidea) [21, 22], and beetles (Insecta: Coleoptera), additional pollinator classes include birds, mainly hummingbirds (Aves: Trochilidae), and mammals, such as nectar‑feeding bats (Mammalia: Chiroptera) may also contribute in some taxa [17, 23, 24]. In O. robusta, floral visitor assemblages vary among populations: some are visited by a balanced community of bees, whereas others are dominated by a single pollinator group [25–27]. In our study populations, we recorded floral visitors that have not been reported in other localities. In both San Nicolás Tecoaco (SNT) and Matías Rodríguez (MR), flowers were visited primarily by native bees (Insecta: Apoidea), with occasional visits by beetles (Insecta: Coleoptera) [25]. Such differences in visitor composition may influence mating‑system expression and reproductive allocation [25]. Indicators such as pollen and ovule production [28], fruit set [29], and seed number per fruit [30] provide insight into how plants invest resources in reproduction and how these investments respond to variation in floral visitors and sexual composition [31]. In Opuntia robusta, flowers open during the morning and remain receptive for a single day. Field observations indicated that hermaphroditic flowers exhibit protandry, as pollen release precedes stigma receptivity [25]. Similar patterns have been documented in other species of the genus, such as O. lindheimeri [32] and O. polyacantha [33], suggesting that the sequence of sexual maturation may influence both the likelihood of self-pollination and the dependence on floral visitors. In this study, we investigated two O. robusta populations in Hidalgo, Mexico, that differ in sexual composition and ecological context. San Nicolás Tecoaco (SNT) has been described as trioecious [18], whereas Matías Rodríguez (MR) consists exclusively of hermaphroditic individuals [25]. Our objectives were to (1) document floral visitors in each population, (2) quantify pollen and ovule production to estimate Cruden’s pollen–ovule ratio [28], and (3) evaluate early seed performance through controlled pollination experiments and germination assays. By integrating these components, we assess whether populations differ in reproductive allocation and seed quality and explore how such divergence may reflect functional variation in mating‑system expression within the species. Materials and methods Species description Opuntia robusta (Opuntioideae: Cactaceae) is a species endemic to central Mexico that reproduces both clonally and sexually [14]. In the literature, it is typically described as a shrubby plant with large, thick, glaucous-green cladodes and yellow spines grouped in the areoles [14]. However, individuals in the two populations studied here exhibited a prostrate growth habit [25]. The flowers are diurnal, large, and yellow, with green stigma lobes [14]. The species presents three floral types: male flowers, characterized by a long style and an atrophied ovary; female flowers, with reduced anthers and a unilocular ovary filled with ovules; and hermaphroditic flowers, which have an intermediate-length style, a functional ovary, and well-developed anthers [17]. These morphological differences are relevant for interpreting the expression of the mating system in the populations examined. Study area and study sites Fieldwork was conducted in the municipality of Singuilucan, Hidalgo, Mexico, a region characterized by heterogeneous vegetation that includes mixed pine–oak forest, grasslands, shrublands, nopaleras, and chaparral. This environmental mosaic reflects variation in topography and climate, which range from semiarid temperate to temperate subhumid and cool subhumid, with summer‑dominated rainfall [25, 26]. We examined two of populations O. robusta: San Nicolás Tecoaco (SNT), a trioecious population located in a semiarid environment dominated by xerophytic scrub, where Mammillaria sp., Tillandsia, and several Opuntia species co‑occur [27], and Matías Rodríguez (MR), a hermaphrodite population situated in a pine–oak forest, where O. robusta was the only cactus species observed. Geographic coordinates, environmental characteristics, and reproductive systems of both populations are summarized in Table 1. Table 1Environmental and geographic characteristics of the two Opuntia robusta populations studiedReproductive systemLatitudeLongitudeAltitude (m)Precipitation (mm)Temperature (°C)VegetationHermaphroditic20.0348− 98.5319257350–15018–22Pine–Oak forestTrioecious20.044− 98.5877260050–15018–22Xerophytic scrubThe hermaphroditic and trioecious populations are located in Hidalgo, Mexico, under similar precipitation and temperature regimes but differ slightly in elevation and vegetation type. The hermaphroditic population occurs at 2573 m in pine–oak forest, whereas the trioecious population is found at 2600 m in xerophytic scrub Environmental and geographic characteristics of the two Opuntia robusta populations studied The hermaphroditic and trioecious populations are located in Hidalgo, Mexico, under similar precipitation and temperature regimes but differ slightly in elevation and vegetation type. The hermaphroditic population occurs at 2573 m in pine–oak forest, whereas the trioecious population is found at 2600 m in xerophytic scrub Plant identification and voucher information No voucher specimen was deposited because the study used non-destructive sampling; however, the plant material was formally identified by Dr. Salvador Arias (Instituto de Biología, UNAM). Selection of individuals During the 2015 flowering season (mid‑April to late June), we marked plants bearing visible reproductive buds with permanent ink and georeferenced using a Garmin eTrex 10 GPS. To minimize sampling of ramets belonging to the same genet, individuals were selected with a minimum spacing of 5–10 m between them. To document the geographic context of the study, we generated a scaled location map based on these coordinates. The map was produced in R using elevation layers to depict topographic relief, along with state boundaries, and is presented in color to facilitate interpretation. It also includes a scale bar, a north arrow, and distinct symbols and colors for each population. This map (Figure S1) illustrates the spatial distribution of the sampled populations and the elevational separation between them. The geographic distance between populations (~ 6 km) was estimated using the Haversine formula based on the GPS coordinates recorded in the field. Sex determination Sex determination in O. robusta was conducted visually in the field using the diagnostic floral traits described for the species. Male flowers were identified by their reduced ovary and abundant pollen release, female flowers by their functional ovary and reduced or sterile stamens, and hermaphroditic flowers by the simultaneous presence of fertile stamens and a fully developed stigma. These criteria allowed us to assign sexual morphs with high confidence. Based on these traits, we assessed the sexual expression of each flowering individual during anthesis. All sampled plants exhibited fully developed yellow anthers and stigmas with more than ten lobes and were therefore classified as hermaphroditic. To confirm these field assignments, we collected only a subset of floral buds, which were preserved in 70% ethanol and later examined under a stereomicroscope. Not all observed buds were collected, as the purpose of this sampling was solely to verify sexual expression rather than to obtain a complete floral inventory. In addition, one anther per plant was collected and stained with lactophenol blue to verify the presence of functional pollen, which corroborated the hermaphroditic condition of all individuals included in the experiments. Although no functionally male or female flowers were observed during the 2015 sampling period, previous and subsequent studies have documented the presence of these morphs in the trioecious population. This population was initially described as gynodioecious [34], and later work demonstrated that it is trioecious, with males and females occurring at low frequency [35, 36]. However, during our study year only hermaphroditic individuals produced flowers. As a result, all morphological measurements, reproductive analyses, and germination assays for this population are based exclusively on hermaphroditic flowers, and sex could not be included as a factor in the statistical models. Reproductive system characterization We longitudinally sectioned mature floral buds collected prior to anthesis (hermaphrodite population: n = 68; trioecious population: n = 20) and examined them under a stereomicroscope (Leica MZ125, 3 ×) to count anthers and ovules. These buds were later preserved in 70% ethanol, but anthers were removed first for pollen analyses, and the remaining tissue was subsequently fixed. Each bud was collected from a different plant to ensure sample independence. We dried two non‑dehiscent anthers per bud with silica gel and released pollen grains by pressing or vortexing in distilled water. Pollen was counted on a 1 × 1 mm grid, and total pollen production per flower was estimated by multiplying the mean pollen count by the total number of anthers. From this sampling, 39 flowers (20 from the hermaphrodite population and 19 from the trioecious population) were also measured morphologically. In the trioecious population, three of these flowers did not receive floral visits due to the low visitor activity recorded during the observation period. For these flowers, and for part of the sampling in both populations, we used an alternative procedure: anthers were collected before dehiscence, when pollen was mature, and ovules were later counted in the laboratory from fixed flowers. Floral buds were collected after visitor observations and, once wilted, preserved in 70% ethanol solely for morphological measurements (e.g., ovary size, ovule number, stigma size). This procedure allowed us to obtain complete morphological data even when fresh flowers were not available. In the hermaphrodite population, whole flowers were collected in ethanol to confirm sexual condition and avoid misclassification, as this population had not been previously sampled. In both populations, all examined flowers, whether used for morphological measurements or reproductive counts, showed hermaphroditic morphology. Floral visitor observations We used an ad libitum sampling approach [37, 38], similar to other pollination studies where continuous standardized censuses are not feasible [39]. Each plant was observed for ~ 10 min to record visitor identity, foraging behavior, and contact with reproductive structures; afterward, we inspected the bagged flowers used in the controlled pollination treatments. Although sampling effort differed slightly between populations due to logistical constraints, the same protocol was applied at both sites. We recorded visitor groups, behaviors, and visit counts per flower when visits occurred. Some insects were collected in 70% ethanol for identification; bees were photographed with an AxioCam ICc5 and identified by I. A. Hinojosa-Díaz (Instituto de Biología, UNAM). Because observations were opportunistic, we did not quantify pollinator effectiveness (e.g., pollen deposition or visitation rates), and all behavioral descriptions are qualitative. Nectar production phenology We recorded nectar production daily in a subset of flowers to document its temporal pattern relative to floral development. Each flower was evaluated multiple times per day from the day prior to anthesis until senescence. At each observation, we recorded nectar presence or absence by visually inspecting the nectary chamber. Because nectar volume was analyzed in a companion study currently in preparation [40], only qualitative presence/absence data were collected for the present work. These observations were used to characterize the timing of nectar production and its correspondence with other phenological events. Floral opening phenology We recorded floral opening phenology through hourly observations of a subset of flowers from the day before anthesis until senescence, noting when the corolla began to open, reached full opening, and initiated wilting. These data were used to characterize variation in corolla-opening timing within and between populations. Sampling effort differed slightly between populations due to logistical constraints. In the hermaphrodite population, we located and marked new plants and monitored them for a longer period while establishing the study plots. In the trioecious population, focal plants had been marked in previous studies, and we monitored only the individuals that remained identifiable during the flowering season. Consequently, the hermaphrodite population yielded a greater number of observed flowers and phenological records. Despite these differences in sampling duration, the same hourly protocol was applied at both sites, and all flowers were followed continuously throughout their anthesis period. The resulting data were analyzed statistically as described below. Controlled pollination experiments To evaluate the mating system, we estimated the pollen–ovule ratio [28] and conducted controlled pollinations from April to July 2015. We applied five pollination treatments to hermaphroditic flowers (Figure S2): 1. Automatic self‑pollination (AS): flowers bagged before anthesis, no manipulation, 2. Manual self‑pollination (MS): flowers bagged, emasculated, and hand‑pollinated with their own pollen, 3. Manual cross‑pollination (CR): flowers emasculated and pollinated with mixed pollen from ≥ 5 plants located ≥ 500 m away, 4. Pollen supplementation (SP): flowers exposed to visitors and supplemented with mixed pollen from ≥ 5 distant plants, 5. Open pollination (CT): unmanipulated flowers, exposed to visitors [38]. We harvested mature fruits from August to November 2015 and stored them for seed collection. Immediately after harvest, the pulp was removed and fruits were left to dry on kraft paper. The cleaned fruits were then stored in kraft paper bags under dry room conditions from late November until early February, when germination experiments began. This procedure ensured consistent seed processing prior to germination trials. Fruit set, seed number, and fruit size We calculated fruit set as the proportion of flowers included in the pollination experiments that developed a mature fruit. For each harvested fruit, we recorded the total number of seeds (seed set). Fruit size was estimated using an estimated fruit volume, derived from two perpendicular diameters and fruit length assuming an ellipsoid shape. This estimated volume was used as a proxy for fruit size because fresh weight was not recorded. Seed germination To evaluate the effect of the controlled pollination treatments on early seed performance, we conducted a germination experiment using all seeds obtained from fruits produced under the five reproductive treatments. We then followed the cactus germination protocol described in [41]. Seeds were chemically scarified through sequential immersion in concentrated H₂SO₄, 70% ethanol, and 60% NaClO, with sterile distilled‑water rinses between steps. After scarification, seeds were sown on MS medium at 0.2 × strength supplemented with sucrose, MES, and bacteriological agar. We adjusted the medium to pH 5.6, sterilized by autoclaving, and poured into sterile Petri dishes. In total, we germinated 108 fruits (22 auto‑self, 18 manual self, 24 cross, 18 supplemental pollen, and 26 control). Because seed number varied widely among fruits, we sowed all available seeds per fruit, resulting in 6097 seeds across treatments (1782 control, 1277 auto‑self, 1038 manual self, 1088 cross, and 912 supplemental pollen). Each fruit was treated as an independent replicate. Petri dishes were maintained at 28 °C under a 12 h light/12 h dark cycle and monitored daily for 90 days, recording the germination day of each seed. Statistical analyses Nectar production and protandry Nectar presence was analyzed as a binary variable (presence/absence) across floral developmental days. We initially fitted a binomial generalized linear model to test for a temporal trend; however, the model exhibited complete separation due to the low number of positive events. To formally explore the presence of a trend, we aggregated data by day and applied a Cochran–Armitage trend test. The lack of statistical evidence for a temporal trend reflects the low frequency of nectar production and the strong protandry observed in the flowers. Floral opening dynamics Corolla opening was analyzed as a continuous variable as a function of time of day. We fitted GAMs with corolla opening as the response and decimal time as a smooth predictor, using population-specific smoothing functions and REML estimation. We first fitted an inter-population model including population as a fixed effect and a population-specific smooth term for time. We then fitted separate models for each population to describe intra-population opening curves. Residual normality was evaluated using Shapiro–Wilk tests. We also generated LOESS curves with bootstrap resampling to provide descriptive visualizations of floral opening dynamics and their confidence intervals. To compare corolla opening between populations, we used a Mann–Whitney test and estimated the rank-biserial correlation and its confidence intervals via bootstrap resampling. These non-parametric analyses were used for descriptive purposes and complemented the inference obtained from the GAM models. Reproductive traits (ovules, pollen, and P/O ratio) We analyzed continuous reproductive traits (number of ovules, number of pollen grains, and Cruden’s P/O ratio) using a candidate set of generalized linear models that included Poisson, quasi-Poisson, negative binomial, and Gamma GLMs with log link, as well as linear models applied to log-transformed data. For each trait, we evaluated overdispersion and compared candidate models using AICc. Model selection identified a negative binomial GLM as the best model for ovule number, a negative binomial GLMM with plant identity as a random effect as the best model for pollen grain number, and a Gamma GLM with log link as the best model for the P/O ratio. Model diagnostics were performed using DHARMa residual simulations, including uniform Q–Q plots and residual-versus-predicted plots, to assess model fit and detect potential deviations such as overdispersion, zero-inflation, or misspecification. These diagnostics were used to validate the adequacy of the selected GLMs and GLMMs. As complementary descriptive statistics, we compared populations using Wilcoxon rank–sum tests for ovule number, pollen grain number, and P/O ratio, and reported median estimates with confidence intervals. We additionally applied Shapiro–Wilk tests to raw residuals as descriptive assessments of normality, as requested by reviewers; however, these tests were not used for model selection and did not replace inference derived from the selected GLMs. Fruit set, seed set, and fruit volume Fruit formation was classified as a binary variable based on estimated fruit volume (fruit formed = 1 when volume > 0). Fruit set was calculated as the proportion of flowers that formed fruit within each population × treatment combination. All fruits were cut open and examined to verify seed development. Seed set was quantified using only fruits that, upon verification, contained developed seeds, and was summarized using the mean and standard deviation of seed number per fruit. Estimated fruit volume was analyzed descriptively for fruits with recorded measurements. These analyses were descriptive and are summarized in a master table provided in the supplementary material. To complement these summaries, we generated graphical comparisons using ggplot2. Because these visualizations do not include built-in statistical tests, we computed all non-parametric analyses separately and incorporated the results into the figures. Specifically, for seed set and fruit volume we first performed Kruskal–Wallis tests within each population to evaluate overall treatment effects. When significant, we conducted Dunn post hoc tests with Holm correction for multiple comparisons. These non-parametric analyses were used as secondary, exploratory tools to aid interpretation of treatment-level differences and did not replace the primary inference derived from the generalized linear models applied to the other reproductive traits. Germination Germination was analyzed as the proportion of germinated seeds per fruit. For each plant × fruit × population × treatment combination, we constructed a dataset including germinated seeds, total seeds, and failures. We initially fitted binomial GLMs and binomial mixed models with plant identity as a random effect. Due to overdispersion and residual patterns detected with DHARMa diagnostics, we compared alternative error structures including quasi-binomial models, zero-inflated models, and beta-binomial models fitted with glmmTMB. AIC comparisons consistently identified beta-binomial GLMMs as the best-supported models. We evaluated treatment effects at two levels. First, we fitted a model including a population × treatment interaction and a random effect of plant nested within population. Second, we conducted within-population analyses by fitting independent beta-binomial GLMMs with plant as a random effect. Treatment comparisons were performed using estimated marginal means (EMMs) with Tukey adjustments. As a visual summary of these pairwise contrasts, we generated a heatmap of treatment-level comparisons within each population. Model assumptions were evaluated using tests for uniformity, overdispersion, zero-inflation, and Q–Q plots of simulated residuals. We also performed descriptive comparisons between populations using Wilcoxon rank–sum tests, reporting median estimates as complementary descriptive statistics. These non-parametric comparisons were exploratory and did not replace the inference obtained from the beta-binomial GLMMs. Finally, for descriptive comparisons among treatments within each population, we used Kruskal–Wallis tests and reported median estimates as complementary descriptive statistics. These analyses were exploratory and did not replace the inference obtained from the GLMMs. Relationship between floral visitors and germination To evaluate whether floral visitation was associated with reproductive success, we constructed a plant-level dataset combining the number of germinated seeds from control fruits only with the total number of floral visits recorded for each visitor group. We modeled the proportion of germinated seeds using binomial GLMs and beta-binomial models fitted with glmmTMB, with the number of visits per visitor group as predictors. Binomial and beta-binomial models were compared using AICc, and the beta-binomial models were selected as the best supported. We fitted a global model including all plants and separate models for each population. Because exploratory patterns suggested that visitor groups differed in the strength and direction of their association with germination, we generated a visualization based on the hermaphrodite population, where sample sizes were highest. For this subset, we plotted germination proportion against the number of visits for each visitor group and overlaid beta-binomial model predictions and their confidence intervals derived from the population-specific model. This allowed us to illustrate the modeled relationships for all visitor groups under a consistent analytical framework. Model assumptions were evaluated using DHARMa residual simulations, including uniform Q–Q plots and residual-versus-predicted plots. All analyses were conducted in R version 4.5.3 [42]. Results Nectar production phenology Nectar production exhibited a pattern fully consistent with the species’ protandry. In all monitored flowers, nectar was absent on day 0 (closed bud) and appeared only on day 1, coinciding with the late male phase. On day 2, nectar presence varied among individuals, and by day 3 (senescence) it disappeared again. Although each flower was evaluated multiple times per day, nectar was detected only once per day, with no evidence of additional production within the same day. This discrete pattern, nectar presence restricted to a very narrow window within the male phase, was consistent across all monitored buds. Because of this data structure, statistical analyses of temporal trends could not be fitted adequately. The binomial GLM exhibited complete separation, preventing convergence, and the Cochran–Armitage trend test could not be estimated (Z = NaN, p = NA) due to insufficient variation across days. This lack of statistical trend does not indicate absence of a biological pattern; rather, it reflects the discrete, phase-dependent nature of nectar production in O. robusta, characteristic of a protandrous species. Overall, these results confirm that nectar production is strictly associated with the late male phase and absent during the female phase, reinforcing the evidence for a strongly protandrous reproductive phenology. Floral opening phenology LOESS curves revealed clear differences in the daily dynamics of floral opening between populations (Fig. 1). The hermaphrodite population showed a broader and more curved trajectory, whereas the trioecious population exhibited a narrower and more stable pattern. Although median opening times were similar, variability was greater in the hermaphrodite population, consistent with the absence of significant differences in the global comparison (Mann–Whitney test: W = 6607, p = 0.19). Floral opening phenology and magnitude in two Opuntia robusta populations. A Bootstrap LOESS curves showing the daily dynamics of floral opening (corolla opening, mm) in the hermaphrodite population and the trioecious population. Points represent individual measurements, and shaded areas indicate 95% confidence intervals. B Comparison of floral opening between populations using violin plots. A Mann–Whitney test detected no significant difference between populations The GAM models supported this pattern. In the inter-population model, the smooth term for the hermaphrodite population was highly significant (edf = 4.00, F = 7.83, p = 1.2 × 10⁻⁶), while the trioecious population showed no detectable curvature (edf = 1.00, F = 0.87, p = 0.352; Table S1). Because within-population models yielded the same qualitative results, only the inter-population model is presented. In this context, effective degrees of freedom (edf) quantify curve complexity: values near 1 indicate an approximately linear trend, whereas higher values reflect increasing curvature. Thus, edf ≈ 4 in the hermaphrodite population indicate a clearly nonlinear opening dynamic, while edf ≈ 1 in the trioecious population confirm a simple, inflection-free pattern. Normality diagnostics (Table S2) showed stronger deviations in the trioecious population (W = 0.889, p = 2.5 × 10⁻4), consistent with the slight asymmetry observed in its distribution despite its simpler phenological profile. Hermaphroditic individuals in both populations exhibited an extended flowering period, from April to July, exceeding previously reported durations by more than a month. Reproductive traits Reproductive traits differed markedly between populations, with generalized linear models providing the primary inference. Consistent with the visual patterns shown in Fig. 2, the negative binomial GLM for ovule number showed a strong population effect, with flowers from the trioecious population producing significantly fewer ovules than those from the hermaphrodite population (β = –0.402, SE = 0.098, p = 3.9 × 10⁻5). Back-transformed estimates indicated an average of 670 ovules per flower in the hermaphrodite population (95% CI: 611–736) and 448 in the trioecious population (95% CI: 383–514), representing an approximate 33% reduction (Table 2). Comparison of ovule number and pollen grain number per flower between two Opuntia robusta populations. Distributions of ovules per flower (left) and pollen grains per flower (right) in the hermaphrodite population (n = 65) and the trioecious population (n = 20). Median values were higher in the hermaphrodite population for both traits. A Wilcoxon test indicated a significant difference in ovule number between populations, whereas no significant difference was detected in pollen grain number. The panel shows the statistical results Table 2Mean (± SE) ovule number, pollen number, and pollen–ovule ratio (P/O) for hermaphroditic individuals of Opuntia robusta in two populations in Hidalgo, MexicoTraitHermaphrodite (mean ± SE)Trioecious (mean ± SE)Ovules per flower619 ± 31.3423 ± 38.5Pollen grains per flower145,030 ± 17,900162,785 ± 36,200P/O ratio5.46 ± 0.275.95 ± 0.48log(P/O)2.442.62Log(P/O) values are derived quantities without SE Mean (± SE) ovule number, pollen number, and pollen–ovule ratio (P/O) for hermaphroditic individuals of Opuntia robusta in two populations in Hidalgo, Mexico Log(P/O) values are derived quantities without SE For pollen production, the negative binomial GLM indicated no detectable population effect (β = –0.027, SE = 0.180, p = 0.878), matching the overlapping distributions shown in Fig. 2 and suggesting comparable levels of absolute male investment. The Gamma GLM for Cruden’s pollen–ovule ratio (P/O) revealed a marginal trend toward higher P/O values in the trioecious population (β = 0.317, SE = 0.174, p = 0.072), consistent with the pattern illustrated in Fig. 3A. Back-transformed means were 349 for the hermaphrodite population (95% CI: 296–412) and 479 for the trioecious population (95% CI: 360–645). Pollen–ovule ratio and fruit set across populations and pollination treatments. A Comparison of the pollen–ovule ratio (P/O) between the hermaphrodite and trioecious populations of Opuntia robusta. Distributions show that median P/O values differed between populations, with the trioecious population exhibiting a marginally higher ratio. A Wilcoxon rank-sum test indicated a non-significant trend toward higher P/O in the trioecious population. The panel includes the corresponding statistical results. B Fruit set across pollination treatments in the hermaphrodite and trioecious populations. Bars represent the percentage of flowers that developed mature fruits under each treatment: MS (manual self-pollination), AS (autonomous self-pollination), CT (control), CR (cross-pollination), and SP (supplemented pollination). In both populations, the control treatment (CT) produced the highest fruit set, with the remaining treatments contributing smaller proportions To complement these model-based analyses, we also compared populations using Wilcoxon rank-sum tests. Ovule number differed significantly between populations (median hermaphrodite population = 652, median trioecious population = 470; p = 0.000184), consistent with the GLM results. Pollen production did not differ significantly (p = 0.637), and P/O showed a marginal trend toward higher values in the trioecious population (median hermaphrodite population = 341, median trioecious population = 464; p = 0.069), again matching Fig. 3A. Goodness-of-fit diagnostics (Figures S2–S4) and normality tests (Table S3) indicated deviations from symmetry for pollen and P/O, supporting the use of non-parametric tests as descriptive complements to the GLMs. Although this difference was not statistically significant, it suggests a modest increase in relative male investment in the trioecious population. According to Cruden’s classification, both populations fall within the range characteristic of facultative autogamy (100–1000), although the trioecious population approaches the upper limit associated with facultative xenogamy. Reproductive performance: fruit set, seed set, and fruit volume Reproductive parameters associated with fruit and seed formation showed clear differences between populations (Table S4). In terms of fruit set, the trioecious population exhibited higher raw percentages than the hermaphrodite population, consistent with the patterns shown in Fig. 3B. In the hermaphrodite population, the control treatment accounted for the largest proportion of fruits formed (approximately 62–65% of the total), followed by manual self-pollination and supplemental pollination. In the trioecious population, the control treatment also dominated fruit set (57–80%), but manual self-pollination and supplemental pollination contributed proportionally more than in the hermaphrodite population. These patterns reflect differences in reproductive efficiency between populations, with the trioecious population showing higher fruit set and a more evenly distributed contribution across treatments. Regarding the number of seeds per fruit, the trioecious population consistently produced more seeds than the hermaphrodite population, with means ranging from 128 ± 157 to 286 ± 188 seeds, compared with 83 ± 130 to 178 ± 148 seeds in the hermaphrodite population. These differences align with the distributions shown in Fig. 4A. Dunn’s post-hoc tests revealed significant differences among treatments within each population. In the hermaphrodite population, cross-pollination produced significantly fewer seeds than control, manual self-pollination, and supplemental pollination, and autonomous self-pollination produced fewer seeds than control. Together, these results indicate that cross-pollination was the treatment with the lowest seed production in this population. In the trioecious population, significant differences were more limited: the control treatment produced fewer seeds than manual self-pollination and supplemental pollination, suggesting that these treatments promote higher seed production in this population. No other pairwise differences were significant. Seed set and fruit volume across pollination treatments in two Opuntia robusta populations. Pollination treatments include MS (manual self-pollination), AS (autonomous self-pollination), CT (control), CR (cross-pollination), and SP (supplemental pollination). A Seed set across treatments in the hermaphrodite and trioecious populations. Violin plots show the distribution of seed number per fruit, with medians indicated. Kruskal–Wallis tests detected significant treatment effects within each population; statistically significant pairwise differences from Dunn’s post hoc tests (Holm correction) are indicated by asterisks. B Fruit volume across the same pollination treatments. Violin plots show the distribution of fruit volume, with medians indicated. Kruskal–Wallis tests detected significant treatment effects within each population; statistically significant pairwise differences from Dunn’s post hoc tests (Holm correction) are indicated by asterisks Fruit volume showed a similar pattern (Fig. 4B). In the hermaphrodite population, values ranged from 184,705 ± 268,296 to 340,746 ± 281,546 mm3, whereas in the trioecious population they ranged from 211,412 ± 242,127 to 464,746 ± 313,954 mm3. Pairwise comparisons indicated that in the hermaphrodite population, cross-pollination produced significantly smaller fruits than control and manual self-pollination, reinforcing its lower reproductive performance under cross-pollination. In the trioecious population, the control treatment produced significantly smaller fruits than autonomous self-pollination, manual self-pollination, and supplemental pollination, identifying it as the treatment with the lowest fruit volume in this population. No other significant differences were detected. Taken together, these results show that the trioecious population exhibits higher reproductive performance across most treatments, both in fruit set and in seed number and fruit volume, suggesting greater reproductive efficiency in this population. Seed germination between populations Seed germination differed between populations (Fig. 5A). The trioecious population showed a higher median germination proportion (0.29) than the hermaphrodite population (0.20), although this difference was not significant in the non-parametric comparison (Wilcoxon test: p = 0.118). Germination probability and germination proportion across populations and pollination treatments. A Germination probability in the hermaphrodite and trioecious populations of Opuntia robusta. Distributions show that median germination was higher in the trioecious population, although a Wilcoxon rank-sum test indicated no significant difference. B Germination proportion across five pollination treatments in the hermaphrodite and trioecious populations. Treatments include MS (manual self-pollination), AS (autonomous self-pollination), CT (control), CR (cross-pollination), and SP (supplemented pollination). Violin plots show the distribution of germination proportions, with median values indicated. A Kruskal–Wallis test detected no significant differences among treatments in the hermaphrodite population (P = 0.782) and a near-significant trend in the trioecious population (P = 0.0579). In the trioecious population, pairwise contrasts from estimated marginal means (EMMeans) identified a significant difference between AS and CT (Holm-adjusted P ≈ 0.02), marked with an asterisk in the figure In contrast, the beta-binomial mixed model detected a significant population effect, indicating that seeds from the trioecious population germinated more frequently than those from the hermaphrodite population (β = 0.466, SE = 0.189, z = 2.46, p = 0.0137). This corresponds to a 59% increase in germination probability (OR = 1.59). Estimated marginal means were 0.213 for the hermaphrodite population (95% CI: 0.169–0.265) and 0.302 for the trioecious population (95% CI: 0.245–0.366; Fig. 5A). Residual diagnostics confirmed appropriate model fit and no evidence of overdispersion (Figure S6). The model also revealed substantial among-plant variation (Var = 0.642), suggesting individual-level differences in seed viability or maternal effects. Seed germination across pollination treatments Seed germination varied among pollination treatments (Fig. 5B). In the hermaphrodite population, germination probabilities were similar across treatments (0.19–0.26), with broad overlap in confidence intervals, indicating that germination is relatively homogeneous and insensitive to pollination type. In the trioecious population, however, clear treatment differences emerged. Germination was highest in autonomous self-pollination (AS = 0.44, 95% CI: 0.29–0.60) and manual self-pollination (MS = 0.37, 95% CI: 0.23–0.54), whereas the control (CT = 0.24, 95% CI: 0.15–0.35) and cross-pollination (CR = 0.22, 95% CI: 0.12–0.38) showed lower values. Supplemental pollination (SP = 0.27, 95% CI: 0.15–0.44) occupied an intermediate position. The beta-binomial model (Table 3) confirmed these patterns. Autonomous self-pollination resulted in significantly higher germination than the control (β = 0.221, SE = 0.100, z = 2.21, p = 0.027), indicating that autonomous selfing produces seeds of higher viability than natural pollination. This is consistent with the estimated marginal means (AS: 0.4386 ± 0.0799 SE; CT: 0.2354 ± 0.0503 SE). Table 3Coefficients of the beta-binomial germination model including pollination treatment, population, and their interactionTermEstimateSEzp(Intercept: Hermaphrodite–CT)− 1.0360.224− 4.633.7 × 10⁻⁶AS− 0.0500.293− 0.170.865CR− 0.1380.307− 0.450.653MS0.0850.3210.260.792SP− 0.2430.343− 0.710.480Population (trioecious vs hermaphrodite)− 0.2710.335− 0.810.418Trioecious × AS1.1280.4692.400.016Trioecious × CR0.1090.5290.210.837Trioecious × MS0.6200.4931.260.209Trioecious × SP0.5010.5330.940.347Treatments include MS (manual self-pollination), AS (autonomous self-pollination), CT (control), CR (cross-pollination), and SP (supplemented pollination). The hermaphrodite population and the control treatment (CT) serve as baselines. The table reports estimated coefficients, standard errors, z-values, and p-values for all main effects and interaction terms. A significant interaction (p < 0.05) indicates that the effect of a specific pollination treatment differs between populations Coefficients of the beta-binomial germination model including pollination treatment, population, and their interaction Treatments include MS (manual self-pollination), AS (autonomous self-pollination), CT (control), CR (cross-pollination), and SP (supplemented pollination). The hermaphrodite population and the control treatment (CT) serve as baselines. The table reports estimated coefficients, standard errors, z-values, and p-values for all main effects and interaction terms. A significant interaction (p < 0.05) indicates that the effect of a specific pollination treatment differs between populations Cross-pollination (β = –0.192, p = 0.097) did not increase germination relative to the control. Supplemental pollination (β = –0.241, p = 0.046) resulted in significantly lower germination than the control, which does not indicate pollen limitation. Germination distributions showed greater variability in CT and CR, and more consistently high values in AS and MS (Fig. 5B). Residual diagnostics confirmed appropriate model fit and revealed greater heterogeneity in the trioecious population (Figure S7), consistent with its broader phenological and reproductive variation. These treatment-level patterns are also summarized in Supplementary Figure S8, which provides a visual representation of the pairwise contrasts from the interaction model. The heatmap reflects the same result reported above, including the significant difference between AS and CT in the trioecious population. Floral visitor community The floral visitor community was dominated by solitary bees and beetles (Figure S9A). The most frequent visitors included Anthophora sp., Macrotera sp., Tetraloniella sp., and Nitidulidae. Figure S9B shows diagnostic images of Macrotera sp. and Tetraloniella sp. used for taxonomic confirmation in the laboratory. Effect of floral visitors on germination The effect of floral visitors on germination probability showed contrasting patterns between populations. In the global model, Anthophora sp. was the only visitor associated with a significant increase in germination (β = 0.737 ± 0.336 SE, p = 0.028; Table 4). However, this pattern did not persist when populations were analyzed separately, indicating that the global effect reflects between-population differences rather than a consistent effect of this visitor species. Table 4Effects of the main floral visitors on seed germination across the global model and within each populationVisitorGlobal modelHermaphroditeTrioeciousMacrotera sp.0.256 ± 0.141, p = 0.0690.488 ± 0.174, p = 0.004960.53 ± 0.27, p = 0.052Nitidulidae–0.137 ± 0.177, p = 0.4390.337 ± 0.176, p = 0.055NAAnthophora sp.0.737 ± 0.336, p = 0.028–2.37 ± 1.14, p = 0.0380.565 ± 0.417, p = 0.175Tetraloniella sp.–0.265 ± 0.175, p = 0.129–0.887 ± 0.264, p = 0.00077–0.188 ± 0.233, p = 0.420Cecropterus sp.0.212 ± 0.471, p = 0.6520.378 ± 0.331, p = 0.254NACoefficients (β ± SE) are from beta‑binomial models fitted separately for the global dataset and for each population. Significant effects (p < 0.05) are shown in bold. Visitor observations were opportunistic and non‑systematic; coefficients represent ecological associations rather than pollinator effectiveness Effects of the main floral visitors on seed germination across the global model and within each population Coefficients (β ± SE) are from beta‑binomial models fitted separately for the global dataset and for each population. Significant effects (p < 0.05) are shown in bold. Visitor observations were opportunistic and non‑systematic; coefficients represent ecological associations rather than pollinator effectiveness It is important to note that floral visitor observations were opportunistic and non‑systematic; therefore, these models do not allow inference of pollinator effectiveness. Results should be interpreted solely as associations between visitor presence and germination. Within-population effects Within hermaphrodite population (Table 4), floral visitor identity is strongly and directionally modulated germination. Macrotera sp. increased germination (β = 0.488 ± 0.174, p = 0.00496), a pattern also visible in the LOESS curve (Fig. 6), which shows higher germination proportions at higher visitation rates. Tetraloniella sp. significantly reduced germination (β = − 0.887 ± 0.264, p = 0.00077), and Anthophora sp. showed a similar negative effect (β = − 2.37 ± 1.14, p = 0.038). Nitidulidae exhibited a marginally positive effect (β = 0.539, p = 0.069). These patterns are illustrated in Fig. 6, which shows beta-binomial model predictions and confidence intervals for each visitor group based on the hermaphrodite-population model. Relationship between germination proportion and floral visitation in the hermaphrodite population. Germination proportion is shown as a function of visits per plant for each floral visitor group in the hermaphrodite population. Points represent individual plants, and curves show beta-binomial model predictions from the population-specific model, with shaded 95% confidence intervals. Each panel displays the estimated regression coefficient (β) and associated p-value for the effect of visitation by that group on germination. The figure illustrates that visitor groups differ in both the strength and direction of their associations with germination, with some taxa showing positive relationships and others negative or weak effects In trioecious population (Table 4), effects were weaker and less consistent. Macrotera sp. showed a marginal positive effect (β = 0.53 ± 0.27, p = 0.052), whereas Anthophora sp. and Tetraloniella sp. showed no detectable effects. This suggests that in trioecious population, germination is less dependent on visitor identity, possibly due to a more homogeneous or stable visitor community. Together, these results show that although the global model suggested a positive effect of Anthophora sp., the true ecological patterns emerge only at the within‑population level. Hermaphrodite population exhibits strong and contrasting visitor‑specific effects, whereas trioecious population shows weaker and largely non‑significant patterns. Given the non‑systematic nature of visitor observations, these associations should not be interpreted as evidence of pollinator effectiveness. Residual diagnostics confirmed appropriate model fit (Figure S10). Discussion Autogamy appears to be favored in Opuntia robusta Hermaphroditic individuals in both populations exhibited spontaneous self‑pollination and protandry, a pattern also reported for O. lindheimeri [32]. Rapid maturation of the stamens facilitates self-pollination, because the anthers release pollen early enough that it frequently coats the style and stigma at anthesis. Although protandry can reduce selfing in some Opuntia species, such as O. polyacantha, where pollen viability declines before stigma receptivity [33], our results indicate that in O. robusta, protandry does not prevent self‑fertilization. The extremely narrow nectar‑production window, restricted to the late male phase, reinforces this interpretation. Pollen production did not differ between populations, but ovule number did: the hermaphrodite population produced nearly 70% more ovules than the trioecious population. In contrast, the trioecious population produced fewer ovules but exhibited higher germination, consistent with greater per‑seed investment and similar to patterns observed in another trioecious population [43]. This combination of lower ovule number and higher seed viability suggests a reproductive strategy consistent with reproductive‑assurance mechanisms and with the advantages of selfing under conditions where the availability of floral visitors or cross‑pollen receipt is insufficient [44, 45]. Flowering phenology varied between populations Within the sampled period, the daily rhythms of floral opening were similar between populations and showed no significant differences. However, sampling times did not coincide between sites, which prevented us from determining whether both populations share the same initial opening time. Although the hermaphrodite population exhibited a more curved daily trajectory, reflected in the LOESS fit, and the trioecious population showed a simpler, linear pattern, these differences in curve shape did not translate into statistical differences in floral opening distributions. The greater phenological stability observed in the trioecious population may enhance reproductive efficiency by synchronizing sexual phases with more predictable environmental windows, consistent with patterns described in other systems [46, 47]. In trioecious populations of Opuntia robusta, flowering phenology differs among sexual morphs. Del Castillo and González-Espinoza [17] reported that in the Santa Rosa population (San Luis Potosí), male individuals begin flowering in February, females in March, and hermaphrodites in April, with all morphs finishing by late May or early June. In the dioecious population of the Jardín Botánico Regional de Cadereyta, flowering begins in mid-March for males and shortly thereafter for females, with both morphs ending around mid-April [25]. During 2015, we monitored the flowering of the 15 identified females in the trioecious population during several sampling periods between February and July; however, no female flowering was recorded that year. Given that females typically flower from March to mid-April in nearby populations, it is likely that their flowering period occurred before we completed the identification and marking of individuals in hermaphrodite population. In contrast, hermaphrodite individuals may continue flowering until June or July, so their presence during our sampling period is consistent with the expected phenological sequence. This phenological pattern, together with the fact that focal plants in trioecious population had been marked in previous studies and some individuals were no longer identifiable, explains why only hermaphroditic individuals were found flowering during our sampling period. The absence of specialized visitors such as Diadasia sp, highly effective pollinators in other cacti [48], may reduce synchronizing pressure and contribute to the extended flowering period observed in both populations, consistent with studies showing that pollinator availability influences flowering duration and phenological synchrony [46, 47]. Moreover, the phenological range documented here aligns with the diversity reported for other cacti, from the brief flowering of Opuntia tomentosa [49] to the prolonged and repeated flowering of Astrophytum ornatum [50]. A mixed mating system in Opuntia robusta Mixed mating systems are common in plants, even though classical theory predicts that pure selfing or pure outcrossing should be evolutionarily stable [1, 4]. Approximately 63% of plant species exhibit mixed mating [3, 8], and ecological factors such as pollinator availability and partial self-incompatibility have been proposed to explain this prevalence [51]. Variation in selfing rates among populations of the same species may reflect differences in the biotic pollination environment, which in turn can shape evolutionary responses in mating-system traits [52]. In this context, the facultative autogamy observed in the hermaphrodite population may function as a reproductive assurance mechanism under low pollinator availability, whereas the facultative xenogamy in the trioecious population is consistent with a system dependent on cross-pollination. Although we did not record fresh fruit weight, fruit volume remained a reliable indicator of reproductive output. Despite some fruit predation by phytophagous insects [18], fruit volume showed patterns parallel to seed number across all treatments. This pattern is consistent with the positive correlation observed between fruit volume and plant biomass, suggesting that variation in vegetative vigor contributes to differences in reproductive output [35]. Beyond fruit volume, previous field studies in trioecious populations of O. robusta revealed marked differences between sexual morphs: hermaphrodites produced more seeds than females, which is consistent with the fact that females depend obligatorily on cross-pollination to produce viable seeds [34, 43]. In contrast, hermaphrodites can maintain relatively high reproductive output through autogamy, particularly under conditions of reduced pollinator availability. In both populations, manual self-pollination and supplemental pollination produced the largest fruits with the highest seed numbers, and germination was not lower in self-pollination treatments in either population. Taken together, these results indicate that no inbreeding depression was detected in the reproductive traits evaluated (fruit volume, seed number, and germination). Moreover, the trioecious population consistently produced larger fruits with more seeds than the hermaphrodite population, suggesting generally higher reproductive_efficiency. This pattern aligns with observations in Polaskia chende [53], where reproductive output is also strongly associated with seed production, reinforcing the idea that differences in mating system and pollination environment can scale up to population-level variation in reproductive performance. Germination patterns and the efficiency of autogamy Germination varied significantly among populations and treatments. Seeds from the trioecious population germinated more frequently than those from the hermaphrodite population, possibly due to genetic divergence, environmental differences, or maternal effects, which are well known to influence early seed performance [54]. Autonomous self-pollination produced higher germination than manual self-pollination, whereas control, cross-pollination, and supplemental treatments tended to reduce it. Floral manipulation may affect seed quality through mechanical damage, altered nutrient flow, or physiological stress. The lack of benefits from cross-pollination may reflect incompatibility, suboptimal pollen quality, or physiological costs [6, 9]. The substantial random effect of plant identity indicates intrinsic variation among individuals. Overall germination (29%) exceeded previous reports for O. robusta [43]. Germination exhibited contrasting patterns between populations, with direct implications for the mating system. In the hermaphrodite population, germination was low across treatments but not reduced in selfing treatments, indicating that self-fertilization does not diminish early seed success consistently with studies showing that inbreeding depression is often weaker at early life stages [6, 55]. In the trioecious population, germination was equal to or higher in selfing treatments than in cross-pollination, a result that contradicts classical expectations of inbreeding depression and suggests that selfing is not only viable but may be reproductively efficient in this population. Taken together, these results reinforce the absence of detectable inbreeding depression in the early fitness traits evaluated and support the presence of a flexible mixed mating system in O. robusta, consistent with patterns described for other species with mixed mating systems [3, 56]. Floral visitors: ecological associations and contrasting functional assemblages Associations between floral visitors and seed germination revealed contrasting functional patterns across species and populations. Although field observations were not systematic and therefore do not allow assessment of pollinator effectiveness, the statistical models do capture associations between visitor presence and early seed performance, providing a robust ecological framework for interpreting the mating system. At the global level, no visitor species showed a significant association with germination, suggesting that the ecological patterns are population-specific and only become evident in the within-population analyses. In contrast, Macrotera sp. was the second most frequent visitor in both populations and emerged as the taxon with the most consistent positive association at the within-population level: it showed a significant effect in the hermaphrodite population and a positive trend in the trioecious population. This pattern indicates that, regardless of its relative frequency, Macrotera sp. is statistically associated with higher early seed performance, suggesting a potentially important role within the visitor assemblage. This interpretation is consistent with studies documenting the presence of Macrotera sp. associated with Opuntia species in the United States, including M. opuntiae in Montana and Colorado [57], as well as with evidence of strong genetic structure and low gene flow in oligolectic species of the genus [58], which may favor strong local associations with host plants. Recent work also show that Macrotera species can be frequent visitors of Opuntia flowers, such as M. sinaloana in O. cantabrigiensis [59], where their foraging behavior is influenced by florivory. In the hermaphrodite population, visitor identity strongly modulated germination. Tetraloniella sp. and Anthophora sp. showed significant negative associations with germination, whereas Macrotera sp. had a significant positive effect. Nitidulidae also showed a positive association, although only marginally so. The trend observed for nitidulid beetles is noteworthy: these small, opportunistic visitors often move within the flower and generate mechanical contact between anthers and stigma, potentially facilitating autogamy, a mechanism consistent with the importance of selfing in this population. Nitidulid beetles are also common opportunistic visitors in Opuntia flowers, where they enter the stamen mass and can mechanically transport pollen [60], although they are not considered effective pollinators. Their frequent presence in our observations is therefore consistent with their generalist foraging behavior and their tendency to exploit exposed floral resources. In the trioecious population, Macrotera sp. again showed the strongest positive association, although only marginally, while the remaining visitors exhibited no clear effects. This suggests that in the trioecious population, germination is less dependent on visitor identity, possibly due to a more homogeneous or stable visitor community. These patterns are consistent with studies of plant–pollinator networks showing that visitation frequency is a poor predictor of the functional contribution to plant reproductive success, as mutualistic networks are typically structured in ways that decouple interaction frequency from ecological importance [61–63]. The importance of Macrotera sp. in this study, despite not being the dominant visitor, fits this expectation, while the absence of positive effects from Tetraloniella sp., the dominant visitor, reflects the decoupling between frequency and functional importance characteristic of mutualistic networks. Moreover, differences between populations in the associations of visitors with germination suggest contrasting functional assemblages, a common pattern in pollination networks where populations with similar species composition may differ in functional structure and in the contribution of each visitor to reproductive output [64, 65]. Although these associations do not allow inference of pollinator effectiveness, they do reveal that visitors differ functionally between populations and that not all contribute equally to reproductive success. This pattern aligns with the variation documented in cacti bee systems [60, 66] and suggests that O. robusta operates under a generalist pollination regime in which the composition and functional structure of the visitor assemblage provide an ecological context compatible with the flexibility of its mating system. Evolutionary implications of the visitor assemblage and the mating system Differences in the associations between floral visitors and germination have evolutionary implications for understanding the dynamics of the mating system in O. robusta. Although we did not estimate explicit selection gradients, the observed patterns allow us to identify differences in the opportunities for reproductive success faced by the populations. In the hermaphroditic population, the presence of visitors positively associated with germination, particularly Macrotera sp. and Nitidulidae, suggests that the interaction between autogamy and opportunistic visits may generate an environment in which variation in reproductive performance is partially structured by visitor identity. The potential facilitation of self-pollination by Nitidulidae is consistent with their floral behavior: by moving within the staminal chamber and among reproductive structures, these beetles can generate mechanical contact with pollen and, occasionally, with the stigma [60]. In contrast, the trioecious population showed more homogeneous associations, dominated only by a positive trend of Macrotera sp., with no evidence of visitors contributing differentially to performance. This pattern suggests a more uniform reproductive context, where the contribution of visitors to reproductive success is more limited and less variable. Thus, the trioecious population, with lower contributions of autogamy and visitors with weaker effects, may experience an environment in which the influence of visitors on performance is less pronounced. This scenario is consistent with studies showing that the magnitude of visitor-mediated evolutionary change depends on the strength and consistency of their effects on reproductive success [67, 68]. These differences between populations align with patterns of the mating system. The hermaphroditic population, where autogamy is more frequent and visitors show contrasting effects, appears to operate under a regime in which the combination of functional autogamy and opportunistic visitors generates enough variation in performance to influence the relative reproductive success of plants. In contrast, the trioecious population, with lower contributions of autogamy and visitors with weaker effects, may experience an environment where the influence of visitors on performance is less marked, consistent with greater flexibility or functional erosion in the mating system. Taken together, the results suggest that the structure of the visitor assemblage not only modulates the ecological context of the mating system but also shapes the differences in reproductive performance available in each population. This interpretation aligns with theoretical expectations that differences in the structure and magnitude of fitness variation among populations can generate divergent evolutionary trajectories, even in the absence of strong or consistent selection [69, 70]. Under this framework, the contrasting contributions of autogamy and visitor-mediated performance in the two populations of O. robusta represent distinct selective environments that may promote population-level divergence. The combination of autogamy, opportunistic visitors, and contrasting functional assemblages can generate variation in reproductive success that contributes to population divergence, even without explicit selection gradients or marked floral specialization [62, 64]. Differences in the strength and direction of evolutionary processes among populations are common in floral evolution and often arise when visitor assemblages and mating systems differ in their composition or in their degree of dependence on visitors, generating contrasting reproductive contexts among populations [71]. Additional evidence from a companion study supports this interpretation: in a broader comparison including a dioecious population, floral traits in the trioecious population show no detectable evolutionary change, the hermaphroditic population exhibits only weak or inconsistent effects, and the dioecious population displays strong patterns of evolutionary change, including a clear trade-off between nectar production and female function [40]. This pattern is consistent with well-established theoretical expectations: efficient autogamy reduces dependence on visitors and, consequently, opportunities for evolutionary change; partial dependence on visitors maintains moderate opportunities; and strong reliance on specialized visitors increases the sensitivity of floral traits to biotic interactions [68, 71]. Overall, these results suggest that variation in the mating system modulates the opportunities and directions of evolutionary change among populations of O. robusta. Supplementary Information Supplementary Material 1. Acknowledgements We extend our appreciation to the Municipio de Singuilucan for granting us permission to conduct this study in its jurisdiction, and we thank the Pérez Juárez and Samperio Pérez families for allowing us to work on their lands at San Nicolás Tecoaco, as well as Sra. Lucía and her husband, Justino Butrón, for their support and hospitality during our stay in Matías Rodríguez. We thank Dr. Ismael A. Hinojosa‑Díaz (Instituto de Biología, UNAM) for his assistance with the identification of bee specimens using photographs and non‑preserved material. We thank Mario Sandoval-Molina from Cornell University for his assistance in identifying the lepidopteran genus Cecropterus. Authors’ contributions C.M.R. and A.M.P. designed the study and developed the field and laboratory protocols. C.M.R., A.M.P., and M.K.J. conducted fieldwork and performed pollination experiments. C.M.R. and A.M.P. carried out the germination assays. A.M.P. processed and analyzed the data, generated all statistical models, and produced the figures and tables. J.N.F. and M.K.J. provided conceptual guidance, supervised the development of the study, and contributed to the interpretation of results. All authors contributed to the writing and revision of the manuscript and approved the final version. Funding This study was supported by CONACYT grant 180694/2012 awarded to Mariusz K. Janczur. Citlalli Mendoza Reyes received support from CONACYT through a Master’s scholarship (No. 565089). Data availability All datasets generated and analyzed during this study are publicly available in Zenodo at: https://zenodo.org/records/20452221. The analytical workflow was implemented in R version 4.3.1 using reproducible scripts that include data processing, model fitting, model selection, and figure generation. R scripts are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate This research did not involve experiments on humans or animals. Permission to conduct fieldwork was obtained from the landowners. Legal authorization for plant collection and field activities was granted by the Mexican Ministry of Environment and Natural Resources (SEMARNAT) under permit SGPA/DGGFS/712/1230/18. The study site is not located within a protected area, and no species listed as endangered under Mexican regulations (NOM‑059‑SEMARNAT‑2010) were collected. Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Authors’ information Not applicable. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Citlalli H. S. Mendoza-Reyes, Email: citlalli.mendoza@iecologia.unam.mx Mariusz Krzysztof Janczur, Email: majmx@interia.pl. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement All datasets generated and analyzed during this study are publicly available in Zenodo at: https://zenodo.org/records/20452221. The analytical workflow was implemented in R version 4.3.1 using reproducible scripts that include data processing, model fitting, model selection, and figure generation. R scripts are available from the corresponding author upon reasonable request.