If You Plant It, Will They Come? Flowering Phenology, Breeding System, and Pollination Ecology of the Threatened Texas Endemic Hibiscus dasycalyx in Natural and Experimental Populations

Abstract

The federally threatened Neches River rose-mallow (Hibiscus dasycalyx) is a rare wetland species restricted to Eastern Texas, where population recovery depends on successful pollination and reproduction. The reproductive ecology was investigated to evaluate how floral phenology, pollinator activity, and breeding system influence reproductive success in natural versus created populations. Observations and controlled breeding system trials were conducted during 2024–2025 in a natural population (Houston County) and a created population (Walker County). Flowering was extended and synchronous in the natural population but shorter and less synchronized in the restored population. Breeding system trials confirmed self-compatibility without autogamy: fruit set was highest in outcrossed flowers, moderate in hand-selfed flowers, and absent in bagged flowers. Seed set, seed weight, and germination did not differ significantly between treatments. Pollinator visitation was high in the natural population, dominated by the specialist bee Ptilothrix bombiformis (>98% of visits), resulting in high fruit set. In contrast, low visitation in the restored population was associated with reduced fruit production. Pollen receipt varied significantly between sites and was strongly associated with P. bombiformis activity. These results demonstrate that the successful restoration of H. dasycalyx requires not only suitable habitat and planting efforts but also the presence of effective pollinators to ensure reproductive success.

1. Introduction

Understanding flowering phenology, floral visitation, and breeding systems is essential for conserving endangered plants, particularly those reliant on pollinators for reproduction. Phenological traits—such as timing, duration, and synchrony of flowering—strongly influence reproductive success by shaping pollinator visitation and conspecific pollen transfer [1,2,3,4,5]. These traits can vary between natural and created populations due to differences in environmental conditions, demographic maturity, or genetic diversity [6,7,8]. In obligately outcrossing species, reduced flowering synchrony or altered phenology may limit reproductive success and compromise long-term viability [9]. Restoration efforts often focus on habitat suitability and plant survival, but reproductive outcomes depend equally on pollinator availability and effectiveness [10,11,12,13]. Restoration sites may also differ in pollinator abundance, visitation rates, or network structure [12,13]. Small or recently established populations may experience pollinator-mediated Allee effects, where low floral density reduces visitation and pollen transfer [14]. Moreover, shared pollinators among congeners can facilitate hybridization, posing genetic risks to rare species [15]. Numerous studies have documented that artificially created or restored populations often experience reduced reproductive output compared to natural populations due to pollinator limitation, genetic bottlenecks, and disrupted ecological interactions [16,17,18,19,20,21,22]. Such failures underscore the complexity of restoring self-sustaining populations, where environmental suitability alone does not guarantee successful reproduction.

These challenges are especially pertinent for Hibiscus dasycalyx (Neches River rose-mallow), a federally threatened, state-endangered wetland species endemic to eastern Texas [23]. It is restricted to seasonally inundated riparian habitats and faces threats from habitat loss, hydrological alteration, invasive species, and genetic introgression [24,25]. Hybridization with Hibiscus laevis and Hibiscus moscheutos has been documented, with H. laevis posing the greater risk due to closer genetic affinity and overlapping ranges [26,27,28,29,30,31,32,33]. Recent research has focused on spatial relationships with hydrology and soils [34]. However, there are no studies of pollination biology and reproductive success. The specialist bee Ptilothrix bombiformis, an oligolectic pollinator of Malvaceae, is the primary pollinator of multiple Hibiscus species and a potential vector for interspecific pollen transfer [35,36,37,38,39,40]. In the most recent recovery plan, the need to document insect pollinators and understand the potential for gene flow and hybridization with congeners was noted as a top research priority [25].

The species’ dependence on pollinators suggests that reproductive processes may differ substantially between natural and experimentally established populations. Despite its conservation status and intensive efforts to create new populations that now outnumber the remaining natural populations [25], no studies have compared reproductive ecology between natural and created populations of H. dasycalyx. Investigating these differences provides a critical opportunity to identify the ecological and genetic mechanisms influencing reproductive performance in H. dasycalyx, thereby informing more effective conservation and restoration strategies for this and similar narrowly distributed wetland taxa. This study addresses that gap by evaluating flowering phenology, floral visitor composition, and breeding system, using a comparative approach, between a natural population and a recently created experimental population. By examining how differences in flowering synchrony and pollinator activity influence fruit and seed set, this research contributes to understanding the ecological dynamics that underpin successful restoration and long-term population viability—key considerations in sustainable biodiversity management.

2. Materials and Methods

2.1. Study Species, Study Sites, and Plant Morphology

The Neches River rose-mallow (Hibiscus dasycalyx) is a federally threatened, state-endangered perennial wetland plant endemic to four counties (Cherokee, Harrison, Houston, and Trinity) in southeastern Texas [23,24,25,26,27]. Two congeners, the swamp rose-mallow (Hibiscus moscheutos) and halberd-leaf rose-mallow (Hibiscus laevis), are widespread across the eastern United States and frequently co-occur with H. dasycalyx in its native range [23]. Although H. dasycalyx had been considered a potential subspecies of H. laevis [28], molecular evidence supports its recognition as a distinct species [29,30,31,32,33].

The natural population (LL) was located on a conservation property near Lovelady, Houston County, Texas (67 m elevation; coordinates restricted for conservation purposes) (Figure S1). The experimental population was established at the Sam Houston State University Piney Woods Environmental Research Laboratory (PERL), Walker County, Texas (30.75° N, 95.48° W; 76 m elevation), approximately 39.6 km southwest of LL and outside the natural range of H. dasycalyx. Seeds collected from LL in August 2019 were germinated in the PERL greenhouse in 2020 and planted in spring 2021 at the Historical Pond Site 14 (HPS 14), an abandoned fishpond within the Piney Woods Environmental Research Laboratory (PERL) (Figure S2). At the LL site, a stratified random sample of 30 plants was obtained by selecting 10 individuals from each of three subareas, whereas at PERL, the entire population of nine plants were sampled. For each plant, the number of stems, stem diameter at soil level, and stem height were recorded for all stems exceeding 1 m in height. Plants over 1 stem have summed diameter and average height values. Environmental variables recorded at each site 15 June 2024 were soil temperature, soil pH, soil moisture, and light level using a Four-in-One digital soil thermometer (AgraTronix, Streetsboro, OH, USA), measurements made at 20 cm depth) and canopy openness measured with a canopy densiometer (Forest Densiometers, Marianna, FL, USA). Soil pH and soil temperature were quantitative variables while soil moisture and light levels were quantitative scores on a nine-point scale ranging from low, normal and high (with −, 0, and + adjustments). Two-sample t-tests were used to evaluate site differences.

2.2. Flowering Phenology

At LL, censuses were conducted approximately every two weeks from June through August in 2024; at PERL, more frequent censuses were conducted due to easier access in 2024 and 2025. For each plant, flower buds, open flowers, young fruits (green, indehiscent), and mature fruits (brown, dehiscent) were counted. Natural fruit set was calculated as the number of mature fruits divided by the number of flowers. Population-level flowering duration was estimated from the first and last flowering dates among all plants at each site and year. Mean flowering synchrony within populations was quantified using Augspurger’s index [41,42], which measures the degree of overlap in flowering periods among individuals. Reproductive effort was expressed as the number of flowers produced per day of flowering for each plant.

Differences in flowering duration and total flower production between years at PERL were tested using two-sample t-tests, and 95% confidence intervals were generated by bootstrapping with 10,000 iterations. Comparisons between LL and PERL were made using site-level means, and reproductive effort (flowers per day) was evaluated using an independent-samples t-test. All analyses were conducted in R (v4.3.2) [43], and results were reported as means ± standard error (SE) with associated p-values and 95% confidence intervals.

Differences in flowering synchrony among sites and years were evaluated using a bootstrap resampling approach. Individual flowering curves were resampled with replacement 10,000 times, and the synchrony index was recalculated for each iteration. Mean differences and 95% confidence intervals were derived from the bootstrap distributions. Synchrony values were considered significantly different when the 95% confidence interval of the difference excluded zero. This method allowed nonparametric comparison of synchrony while avoiding assumptions of normality.

2.3. Floral Visitors and Pollen Loads

2.3.1. Floral Visitors

At both sites, floral visitors were observed during 10-min periods on randomly selected flowering Hibiscus plants, supplemented with digital camera video recordings. The identity and frequency of floral visitors were recorded, and visitation rates were calculated as the number of visits per flower per minute for each day of observation. Visitor behavior was recorded as fly-by, nectaring, or pollen-collecting activity (Figures S3–S5).

To assess similarity in floral visitor assemblages among sites and years, pairwise niche overlap analyses were performed using Pianka’s niche overlap index [44] (O_jk). The index quantifies the degree of proportional overlap in resource use—in this case, the relative frequency of each visitor taxon—between two communities. Values range from 0 (no overlap) to 1 (complete overlap).

Relative abundances of floral visitors were calculated for each site–year combination based on direct observation data. Pianka’s index was then computed for all pairwise comparisons (LL 2024 vs. PERL 2024; LL 2024 vs. PERL 2025; PERL 2024 vs. PERL 2025) using the formula:

$$O_{jk}={\displaystyle \frac{\sum _i{p}_{ij}{p}_{ik}}{\sqrt{\left(\sum _i{p}_{ij}^2\right)\left(\sum _i{p}_{ik}^2\right)}}}$$

where p_ij and p_ik represent the proportional abundances of visitor species i at sites j and k, respectively. Analyses were performed in R (v4.3.2) [43] using custom scripts based on the spaa package.

Bootstrap resampling with 10,000 iterations was used to generate 95% confidence intervals for each overlap value, allowing for nonparametric assessment of differences in community composition. Higher overlap values indicated similar visitor assemblages, whereas low values reflected distinct communities or strong interannual turnover.

2.3.2. Pollen Receipt and Removal

At LL in 2024, 10 flowers were randomly harvested on 15 June after dehiscence, and stigma pollen loads were counted directly using a Keyence digital microscope (Keyence Corporation of America, Plano, TX, USA). In 2025 at PERL, 54 flowers from selfed, crossed, and open treatments were categorized for pollen deposition on the stigmas and pollen removal from the anthers. For each flower, pollen deposition was recorded separately on all five stigma lobes. At PERL (2025), stigma lobes were scored categorically (

Table 1. Categorical scoring system for stigma lobe pollen grain counts at PERL (2025).

Score	Pollen Grain Count
0	No grains
1	<10 grains
2	10–50 grains
3	50–100 grains
4	>100 grains

):

For analysis, these categorical scores were converted into estimated pollen grain counts using midpoints: 0 = 0, 1 = 5, 2 = 30, 3 = 75, 4 = 150. Scores across all five lobes were summed to produce an estimated total pollen deposition per flower. This approach provides a conservative estimate and enabled comparisons between PERL and LL populations.

To quantify pollen removal, each flower’s anthers (typically >30 per flower) were scored individually along the staminal column: 1 = full (pollen abundant), 2 = sparse (partially depleted), or 3 = empty (depleted. Anthers were grouped into three zones: basal (positions 1–2), middle (3–8), and apical (9–10). For analysis, scores were converted to estimated removal percentages (full = 0%, sparse = 50%, empty = 100%) and averaged across all anthers per flower to obtain a whole-flower pollen removal index.

Pollen deposition and removal were summarized as medians, means, and ranges. Non-parametric Mann–Whitney U tests were used to compare distributions between populations (LL vs. PERL), pollination treatments, and bee-active vs. post-bee periods at PERL in 2025. Correlations between stigma pollen deposition and anther removal were assessed using Pearson’s correlation coefficient. For biological interpretation, a threshold of 60 grains per flower was used, corresponding to the approximate number of ovules per fruit in H. dasycalyx.

2.4. Breeding System, Fruit Set, Seed Set, and Seed Germination

2.4.1. Breeding System Study Design, Fruit Set, and Seed Set

Breeding system experiments were conducted on seven flowering plants at PERL in 2024 and five flowering plants in 2025. Data was summed across both years. All flower buds were bagged with a fine mesh pollination bag and then assigned randomly to four treatments: (1) bagged to test autogamy, (2) open to test natural pollination, (3) hand-pollinated with self-pollen from the same flower, or (4) hand-pollinated with cross-pollen from another plant. Standardized pollen loads were applied across all five stigma lobes using toothpicks. Self- and cross-pollinated flowers were bagged immediately after pollination to prevent visitation. Fruits were harvested when the ovary and calyx turned brown. All seeds were counted and weighed and scored as good, bad, or ovules not fertilized. Any fruits attacked by herbivores were scored as successful if they contained seeds. Seed set was not calculated for herbivore attacked fruits due to consumption of seeds. At LL in 2024, flowers were sampled only under open pollination, reflecting natural conditions. Fruit set among pollination treatments (Autogamy, Hand Self-Pollination, Hand Outcross-Pollination, and Open Pollination) was analyzed using pairwise two × two contingency tables. For each treatment comparison in the combined dataset (2024 and 2025), odds ratios (ORs), 95% confidence intervals (CIs) based on the Wilson score interval, and p-values were computed using Fisher’s exact test in R (v4.3.2) [43], with the binom package used to estimate proportions and their error bars. Differences in seed set (mean number of viable seeds per fruit) and mean seed weight per fruit were assessed across the Selfed, Outcrossed, and Open Pollination treatments using one-way ANOVA.

2.4.2. Seed Germination

In 2025, germination outcomes for seeds derived from three breeding treatments—Control, Selfed, and Outcrossed—were evaluated in a randomized blocked greenhouse experiment. Each replicate consisted of one fruit containing 20 seeds, and trays served as blocks in the experimental design. Using fruits from one maternal plant with >40 seeds from each treatment, 10 fruits were randomly picked and used to test for germination. Twenty randomly picked seeds were nicked with a scalpel, soaked in distilled water for 24 h and planted in a randomized block design in four trays in Miracle Gro Potting Soil in the Sam Houston State University Life Sciences Building greenhouse (temperature range 21.1–35.6 °C). Seeds were misted twice daily, and germination was monitored daily until no germination was noted for seven days. Cumulative germination and germination speed were analyzed using ANOVA and mixed-effects models. Treatment was included as a fixed effect, and tray was modeled as a random effect to account for block-level variation. Germination speed was assessed using the day of first germination per seed.

2.5. Statistical Analysis and Visualization

All analyses were conducted in R (v4.3.2) [43] and Minitab 21.4.3 [45]. Data were tested for normality and homogeneity of variances prior to analysis. T-tests and ANOVA used Tukey’s HSD post hoc tests to compare means at α = 0.05. During the preparation of this manuscript/study, AI tools (ChatGPT v5 and Microsoft Copilot GPT-5) were used for figure generation under author supervision.

3. Results

3.1. Site and Plant Characteristics

There were significant differences in soil pH, soil temperature, and light levels between LL and PERL (

Table 2. Comparison of environmental variables between LL (natural population) and PERL (experimental population) sites, reported as mean ± standard error (SE). Two-sample t-tests were used to evaluate site differences. Soil moisture scores were identical between sites and could not be tested. The PERL and LL variables with no SE had no variation among samples.

Variable	LL Mean ± SE	PERL Mean ± SE	t-Statistic	p-Value
Soil pH	5.92 ± 0.09	5.50 ± 0.00	4.499	0.0003
Soil Temperature (°C)	24.78 ± 0.13	24.00 ± 0.00	6.018	<0.0001
Soil Moisture (score)	4.00 ± 0.00	4.00 ± 0.00	—	—
Light (score)	5.20 ± 0.20	6.00 ± 0.00	4.000	0.0161
Canopy Openness (%)	88.19 ± 3.29	96.17 ± 3.83	1.578	0.1308

) but no difference in soil moisture or canopy openness. PERL was more acidic, had cooler soil temperatures, and higher categorical light levels, although the quantitative canopy openness score was higher but not significantly different.

The number of stems per plant did not differ significantly between sites (F = 3.27, p = 0.08) (Figure 1A). Plants at LL were taller (F = 9.65, p = 0.004) (Figure 1B) and had greater stem diameters (F = 7.19, p = 0.012) than those at PERL (Figure 1C).

3.2. Flowering Phenology

Flowering phenology varied across sites and years (Table S1). At LL in 2024, 20 of the 30 monitored plants (66%) flowered, producing a total of 335 flowers. At PERL, seven of nine plants (78%) flowered in 2024, producing only 35 flowers, whereas in 2025, five of seven plants (71%) produced 150 flowers. At PERL in 2024, buds were first noted on 29 May, flowering began in early June, peaked in mid-June, and ceased by 27 June. In 2025, flowering began at a similar date but extended longer, ending on 23 July. In contrast, LL exhibited an earlier and much longer flowering season in 2024, beginning in mid-May and continuing into early August (Figures S6 and S7).

Flowering duration and total flower production at PERL were significantly higher in 2025 than in 2024 (duration: +15.86 days, 95% CI [0.86, 31.29], p = 0.036; flower output: +24.94, 95% CI [1.31, 66.06], p = 0.015). LL in 2024 showed moderate output (mean = 16.9 flowers) and a longer phenological spread. Reproductive effort (flowers/day) did not differ significantly between LL and PERL (p = 0.923), although the most productive PERL plants in 2025 exceeded those at LL. Flowering synchrony—defined as the overlap in flowering among individuals—was consistently higher at LL than at PERL (Table S2). LL exhibited high synchrony in 2024 (0.650), whereas synchrony at PERL was substantially lower in 2024 (0.375) and improved only modestly in 2025 (0.419). Bootstrapped comparisons confirmed significantly greater synchrony at LL relative to PERL in both years (mean differences = +0.275 and +0.231, p < 0.001), while the slight increase at PERL between years was not significant (p = 0.132).

3.3. Floral Visitors and Pollen Receipt and Removal

3.3.1. Floral Visitors

Floral Visitors to Natural Population LL (2024)

Floral visitors included bees, butterflies, flies, and beetles. At the natural population (LL) in 2024, floral visitation was abundant throughout the flowering season except in mid–late August as flowering declined (Table S2). Across four census dates totaling 1285 min of observation, 509 visits were recorded to 151 flowers, averaging 0.40 visits per flower per minute. Ptilothrix bombiformis dominated the community, accounting for 97.8% of all visits (Figure 2A). On three of the four census dates, over 95% of visits were made by P. bombiformis, and on 13 July all visits were by this species. Species richness of floral visitors ranged from one to eight species per census, with lower diversity later in the season. Visits per flower per minute peaked on August 5, but few flowers were open on this date. During the core flowering period, peak rates were noted on 29 June just past the median flowering date (Figure S8).

Visitor behavior of P. bombiformis shifted over time (Table S3). Early-season visits consisted primarily of nectaring or brief flybys, whereas by mid- and late-season most visits involved pollen collection (Figure S4A–C). Female bees exhibited stereotyped foraging behavior: landing at the base of the staminal column, spiraling upward while gleaning pollen, and making repeated contact with stigmas (Figure S5A,B, Video S1). Visits for nectar took only a few seconds, while pollen-collecting bouts lasted longer. Other occasional visitors included Bombus pennsylvanicus, Xylocopa micans, Strymon melinus, skipper butterflies, and several pollen-feeding beetles, although these accounted for <3% of total visits and were unlikely to provide substantial pollination services.

Floral Visitors to Created Population PERL (2024 and 2025)

In contrast to the natural site LL, floral visitation was nearly absent at PERL in 2024. Despite 515 min of observation across 12 flowers, only two visits were recorded, both by Melissodes bimaculatus (Table S4). No P. bombiformis were observed visiting flowers, although they were collected in passive traps within 600 m of the site, suggesting that the species was present in the broader landscape but did not utilize the restored population (but see Moskowitz [37]) in 2024.

In 2025, however, visitation at PERL increased substantially (Table S5). During 3726 min of observation across 48 flowers, 271 visits were recorded. P. bombiformis dominated early in the flowering season (10–14 June), while secondary visitors included Melissodes communis, Agapostemon splendens, and several skipper butterflies. Later in the season, floral visitation declined sharply, with only butterflies recorded after 15 June. As a result, visitation rates were strongly concentrated in early June, coinciding with peak activity of P. bombiformis, and declined thereafter (Figure 2B).

Time spent on flowers varied significantly among the three most common bee visitors (Agapostemon splendens, Melissodes communis, and Ptilothrix bombiformis) with ≥4 observations (H = 16.17, p < 0.001). Agapostemon splendens spent the longest time per visit, followed by Melissodes communis, and Ptilothrix bombiformis (

Table 3. Handling time (seconds) of 139 floral visits to Hibiscus dasycalyx at PERL in 2025 for species with at least ≥4 observations. Values are means ± SE, with ranges in parentheses. A Kruskal–Wallis test across species found significant differences in foraging time (H = 16.17, p < 0.001).

Species	n	Handling Time (s) (Mean ± SE, Range)
Agapostemon splendens	8	73.5 ± 25.6 (16–229)
Melissodes communis	9	26.2 ± 7.26 (1–56)
Ptilothrix bombiformis	122	14.1 ± 1.68 (1–134)

). Post hoc tests indicated that Agapostemon splendens visits were significantly longer than those of Ptilothrix bombiformis (p < 0.001), whereas differences involving Melissodes communis were not significant.

Within Ptilothrix, handling time differed strongly by foraging behavior (

Table 4. Handling time (seconds) of Ptilothrix bombiformis by behavior. Values are means ± SE, with ranges in parentheses. Welch’s t-test (unequal variance) was highly significant (t = 3.89, p < 0.001). Not recorded are mixed foraging bouts and fly-bys.

Behavior	n	Handling Time (s) (Mean ± SE, Range)
Nectaring	7	5 (2–14)
Pollen gathering	107	14.4 ± 1.81 (1–134)

). Pollen-collecting visits (mean = 14.4 s, n = 107) were substantially longer than nectaring visits (mean = 5.0 s, n = 7). Statistical tests confirmed this difference (Welch’s t = 3.89, p < 0.001). Thus, pollen foraging is associated with considerably longer floral handling times compared to nectar foraging (Video S1). Note that some visits in Table 3 were not included in Table 4 as they were either not visiting flowers (fly-bys) or had combined pollen/nectar foraging.

Community overlap analyses confirmed these contrasts between LL and PERL as well as between 2024 and 2025 at PERL. The natural LL population in 2024 and PERL in 2025 were highly similar in floral visitor composition (Pianka’s index = 0.9915), reflecting the dominance of P. bombiformis in both. By contrast, LL 2024 and PERL 2024 exhibited almost no overlap (Pianka’s index = 0.0097), and PERL 2024 and 2025 also showed low overlap (Pianka’s index = 0.1385).

3.3.2. Pollen Receipt and Removal

Pollen Receipt

Stigma pollen loads differed strongly among controlled treatments (Figure 3A). Autogamy flowers (n = 4) received no pollen grains (all stigma lobe scores = 0), confirming that Hibiscus dasycalyx is not autogamous. In contrast, hand self-pollinations (n = 1) and hand cross-pollinations (n = 7) each saturated stigmas with more than 500 grains per flower, equivalent to all five lobes receiving the maximum score.

Open-pollinated flowers in the PERL experimental population (n = 43) exhibited highly variable stigma pollen loads (Figure 3B). Median deposition was only 10 grains, with a range from 0 to 750 grains. Approximately 26% of flowers received no pollen, and fewer than one-third (28%) exceeded the 60-grain threshold required to fertilize the 60 ovules typically present per fruit. Comparison with the natural source population at LL in 2024 revealed a striking difference. Flowers at LL (n = 10) had a median stigma load of 91 grains (range: 9–268), and 80% of flowers exceeded the 60-grain threshold. A Mann–Whitney U test confirmed that deposition in the source population was significantly higher than in the experimental population (U = 314.0, p = 0.024) (Figure 3C).

Pollen Removal

In the experimental PERL population (2025), open-pollinated flowers showed a pronounced mismatch between pollen removal from anthers and pollen receipt on stigmas (Figure S10). On average, only ~29% of anthers per flower were depleted, indicating that pollen remained abundant and largely unexported. Yet stigma deposition was highly variable and generally low (median = 10 grains, range = 0–750), with most flowers receiving fewer than the 60 grains required to fertilize all ovules. Across flowers, pollen removal and receipt were only weakly correlated (r ≈ 0.25), demonstrating that pollen export from anthers did not consistently translate into deposition on conspecific stigmas.

Field observations at PERL in 2025 indicated that the specialist Hibiscus bee Ptilothrix bombiformis was abundant from 10 to 14 June and present through 18 June but disappeared thereafter. This temporal shift in the pollinator community corresponded closely with patterns of pollen receipt and removal (Figure 4). Flowers observed before and during Ptilothrix activity (≤18 June, n = 20) received significantly more pollen (median = 65 grains per flower) than those sampled afterward (>18 June, n = 23; median = 5 grains; Mann–Whitney U = 339.0, p = 0.007). In addition, anther pollen removal was greater during the period of bee activity (median = 50%) than after (median = 10%), although this difference was not statistically significant (U = 286.5, p = 0.155). These results demonstrate that high pollen deposition and anther depletion were concentrated during the brief window of Ptilothrix abundance, whereas flowers sampled later, when only butterflies and other bees were present, showed little pollen removal and negligible stigma deposition.

3.4. Breeding System Study

3.4.1. Fruit Set

Breeding experiments demonstrated that Hibiscus dasycalyx is self-compatible but not autogamous (Table S6). Bagged flowers, which excluded pollinators, produced no fruits (0% fruit set) (Figure 5). Hand-pollinated flowers set the highest numbers of fruits when outcrossed (76.9%), followed by selfed flowers (57.1%) (Figure 5). Open-pollinated flowers at the experimental site set very few fruits (3.5%) in 2024, reflecting the absence of effective floral visitors. Open-pollinated flowers in 2025 had higher fruit set than in 2024 (Figure 5).

Because of small sample sizes in 2024, the data was combined for statistical analysis. In the combined data, fruit set differed significantly among pollination treatments (Figure 5 and Figure S10). Autogamy resulted in no fruit set and was significantly less effective than hand self-pollination (odds ratio = 0.01, 95% CI: [0.00–0.17], p = 0.0016), hand cross-pollination (OR = 0.00, 95% CI: [0.00–0.05], p = 0.00007), and open pollination (OR = 0.04, 95% CI: [0.00–0.65], p = 0.0239). Hand self-pollination produced significantly higher fruit set than open pollination (OR = 3.94, 95% CI: [1.92–8.06], p = 0.0011) but was less effective than hand cross-pollination (OR = 0.26, 95% CI: [0.08–0.79], p = 0.0176). Hand cross-pollination was the most effective treatment overall, with fruit set odds over 15 times greater than under open pollination (OR = 15.40, 95% CI: [5.62–42.18], p < 0.0001). In contrast to fruit set, the mean number of good seeds per fruit and mean seed weight per fruit did not differ significantly between selfed, outcrossed and open-pollinated treatments.

3.4.2. Seed Germination

Total Germination Percentage

By day nine after planting, all treatments approached saturation, with estimated asymptotic germination near 100% (Figure 6A) (Control = 96%, Selfed = 95.5%, Outcross = 91.5%, total of 566 seeds germinated). A one-way ANOVA revealed no statistically significant differences in cumulative germination among treatments (F (2, 597) = 1.327, p = 0.266). A two-way ANOVA including Tray as a blocking factor confirmed that neither treatment nor tray significantly influenced total germination (Treatment: F (2, 597) = 1.343, p = 0.262; Tray: F (3, 597) = 0.317, p = 0.813). A mixed-effects model with Tray as a random effect and Treatment as a fixed effect further supported these findings, with negligible block-level variation and no significant treatment effects.

Germination Speed

A one-way ANOVA comparing mean germination day across treatments showed no significant differences (F (2, 597) = 1.474, p = 0.230) (Figure 6B). This indicates that the timing of germination onset was consistent across breeding treatments.

4. Discussion

4.1. Differences Between Natural and Created Sites

The selection of sites for creating new populations balances ecological comparability with long-term protection [16,17,18,19,20,24,25,26,27]. The two sites studied—a natural stand at LL and a created stand at PERL—were broadly similar in environmental conditions, though PERL is an artificial wetland (a former fish hatchery pond). Environmental and structural contrasts reflect their histories: LL plants experienced slightly higher soil pH and warmer soils, while the created population at PERL, by contrast, had lower pH, slightly cooler soils, and higher light availability, features typical of a more open canopy in an artificial pond setting.

The greater stature and thicker stems at LL likely reflect both plant age and site stability, with individuals benefitting from years of favorable hydrological conditions. At PERL, established only in 2021, plants remain in earlier developmental stages. Their higher stem counts may represent vegetative proliferation rather than true clonal spread. These structural contrasts shaped phenological patterns: LL exhibited an extended, evenly distributed flowering season with high synchrony, maximizing pollinator visitation and conspecific pollen exchange. In contrast, PERL in 2024 flowered over a short, highly concentrated, and poorly synchronized period, limiting overlap among individuals and reducing opportunities for cross-pollination. By 2025, flowering duration and output increased at PERL, but synchrony remained lower than LL, suggesting continued constraints for a preferentially outcrossing species such as H. dasycalyx [1,2,3,4,5]. These results indicate that restored populations may experience shorter and less coordinated flowering periods compared to natural populations, even as overall flowering duration and output increase with time since establishment.

While this study provides insight into the reproductive ecology of Hibiscus dasycalyx, interpretation is limited using a single natural and a single created population. Differences in flowering phenology, pollinator activity, and reproductive success between sites may reflect interacting effects of environmental conditions and demographic maturity, which cannot be disentangled with the current design. Acknowledging this limitation, future research will expand sampling to include three to five natural and created populations each across East Texas. Several of the created populations to be sampled are older and exhibit more complex demographic structure, with over 200 planted individuals, some of which are successfully reproducing. This expanded dataset will allow more rigorous evaluation of how environmental variation and population age structure influence reproductive outcomes in created populations of H. dasycalyx.

4.2. Floral Visitors, Behavior, and Pollination Efficiency

Handling time varied strongly among floral visitors. Generalist bees (Agapostemon splendens) and skipper butterflies often remained for extended periods while Melissodes communis showed intermediate and variable visit lengths. By contrast, the specialist Ptilothrix bombiformis [35,36,37,38,39,40,46] typically exhibited short, stereotyped visits. Within Ptilothrix, visit length depended on behavior: nectaring bouts lasted only a few seconds, whereas pollen-collecting bouts ranged from 10 to 90 s. These pollen-collecting visits consistently involved stigma contact and efficient pollen removal, confirming Ptilothrix as the most effective pollinator. The extended visits of generalists and butterflies occasionally resulted in deposition but contributed less consistently to reproduction. Thus, pollinator identity and behavior strongly influence reproductive success, with H. dasycalyx disproportionately dependent on its specialist. The community overlap (Pianka index) results indicate that PERL only intermittently supports a pollinator community resembling natural populations and that pollinator presence, particularly of P. bombiformis, may vary strongly across years.

At PERL in 2025, only five of seven plants flowered, and most open-pollinated flowers received insufficient pollen. This reflects pollinator-mediated Allee effects typical of sparse populations, where small floral displays attract fewer visitors [14]. In contrast, LL flowers in 2024 received significantly more pollen, consistent with higher success in large, dense populations. Recognizing this limitation, additional individuals have now been planted at PERL, boosting the population from <10 to >200 plants.

Patterns of pollen removal and receipt further explain seed set outcomes. Although anthers were rarely depleted, stigmas usually received fewer grains than needed to fertilize 60 ovules. This mismatch explains the low and variable seed set in open-pollinated flowers, where many fruits aborted or contained few seeds. In contrast, controlled pollinations deposited >500 grains per stigma, saturating ovules and producing full capsules. At LL, most flowers naturally exceeded the 60-grain threshold, explaining higher seed set. These results indicate that the larger, denser floral display in the LL population attracted pollinators and ensured higher pollen transfer, while the small and sparse PERL stand experienced severe pollinator limitation.

Thus, reproduction in Hibiscus dasycalyx is limited not by pollen availability but by transfer efficiency. Although pollen was abundant, low visitation and inefficient deposition left many ovules unfertilized. This mismatch highlights a key conservation challenge: small, recently established populations are especially prone to reproductive failure unless floral density is sufficient to attract and retain effective pollinators [12,13,15]. The weak correlation between pollen removal and stigma deposition suggests substantial pollen use for provisioning brood cells or loss via transfer to heterospecific flowers [35,36,37].

At PERL, successful pollination was tightly linked to Ptilothrix presence. Before 18 June, stigmas received significantly more pollen and anthers were depleted, reflecting effective transfer. After Ptilothrix disappeared, stigmas received negligible pollen despite visits from butterflies and other bees. As with the floral visitor and behavior data, these pollen deposition results confirm Ptilothrix as a keystone pollinator of H. dasycalyx [25,35,36,37,38,39,40]. With the recent expansion of additional plants at PERL, nesting by Ptilothrix near the host plants may increase, potentially restoring pollination services in future years.

4.3. Breeding System

These findings confirm that H. dasycalyx is self-compatible but not autogamous, consistent with prior greenhouse studies [28]. Hand pollinations (self- and cross-) saturated stigmas with pollen and produced well-seeded fruits, showing that ovules are readily fertilized once pollen is delivered. However, autogamy produced no fruit in either year, confirming reliance on pollinators. The breeding system closely parallels that of H. laevis, that is also incapable of autonomous self-pollination in areas of adequate pollinator visitation [28,47,48] and is capable of using self-pollen but favors outcrossing [28]. By contrast, H. moscheutos is more self-compatible and capable of autogamy [49,50]. The complete absence of autogamous fruit set in H. dasycalyx highlights divergence within the genus, possibly shaped by habitat context: H. moscheutos occupies wetter, disturbed sites where reproductive assurance is advantageous, while H. dasycalyx persists in isolated, hydrologically variable habitats where outcrossing may be favored. These findings emphasize the species’ vulnerability to pollinator limitation and the importance of maintaining effective pollinator communities for conservation.

4.4. Seed Germination

Breeding system variation can affect early seed performance, with implications for restoration. In this study, seeds from selfed, outcrossed, and control pollinations showed no significant differences in total germination or germination speed under greenhouse conditions. This contrasts with patterns reported in many other taxa, where selfing often reduces germination through inbreeding depression, and outcrossing among genetically divergent individuals can result in outbreeding depression [51]. The absence of such effects here may reflect limited genetic divergence or strong maternal provisioning.

Encouragingly, this suggests that even populations experiencing limited pollinator services can produce viable seed cohorts. High germination percentages ensure strong recruitment potential, buffering against demographic stochasticity. As selfed seeds germinate as reliably as outcrossed seeds, they may still contribute meaningfully to persistence, particularly in small or isolated populations. However, long-term monitoring of survival, growth, and reproduction remains essential to determine whether differences emerge later in the life cycle. Ongoing tracking of these seedlings will help clarify whether early germination success translates into sustainable population growth.

4.5. Implications for Conservation and Restoration

Pollinator limitation poses a significant challenge to the reproductive success of Hibiscus dasycalyx, a wetland-dependent, threatened species endemic to the Neches River basin in eastern Texas [23,24,25]. Although its large, hermaphroditic flowers are adapted to bee pollination, fragmented habitats and reduced pollinator diversity can severely constrain pollen transfer [52]. Field observations confirm that the specialist bee Ptilothrix bombiformis is the primary and most effective pollinator [35,36,37,38,39,40], with generalist bees and butterflies contributing minimally. The disparity between hand-pollinated and naturally pollinated flowers underscores the importance of pollinator quality and visitation frequency as noted in a similar study of Hibiscus mutabilis [53]. Autogamy yields no fruit, and while self-pollination is possible, cross-pollination consistently results in higher fruit set [28,47,48,49,50]. These findings highlight the need for restoration strategies that enhance pollinator habitat, such as buffer zones and continuous floral resources [54,55], to support effective pollination and genetic exchange.

Beyond habitat suitability, successful restoration of H. dasycalyx requires integrated management that addresses ecological, genetic, and biotic factors [6,7,8,16,17,18,19,20,21,22]. Landscape-scale studies show that pollinator services improve with habitat heterogeneity and sequential flowering, reducing isolation and enhancing reproductive outcomes [54,55]. Similar reproductive constraints have been documented in other rare plants, where disrupted pollinator networks limit fruit and seed viability [56,57,58]. Conservation efforts must therefore include pollinator support, genetic monitoring, and long-term demographic assessments [12,13,14,15,30]. Establishing populations with sufficient size and floral density is essential to attract and retain Ptilothrix bombiformis [16,50], ensuring viable reproduction and long-term persistence. Ultimately, sustaining functional pollinator interactions is as critical as protecting habitat in the conservation of this rare Texas endemic [51,53,56,57,58].

5. Conclusions

This study demonstrates that restored populations of Hibiscus dasycalyx can survive and reproduce under novel environmental conditions, but their reproductive success remains constrained by limited flowering synchrony, small population size, and dependence on a specialized pollinator. Natural populations benefit from consistent pollination by Ptilothrix bombiformis, while experimental populations experience reduced visitation and pollen transfer, especially during periods of low floral density or asynchronous bloom.

Although seeds from selfed and outcrossed pollinations germinated equally well under greenhouse conditions, long-term population viability will depend on maintaining effective pollinator interactions and sufficient floral display to support cross-pollination. These findings highlight the importance of integrating pollinator ecology into restoration planning for rare and threatened plant species. Additional planned studies will examine the impact of hybridization on gene flow between H. dasycalyx and co-occurring congeners.

To support the sustainable recovery of H. dasycalyx, conservation strategies should prioritize:

• Establishing populations with adequate size and synchrony,
• Enhancing habitat features that attract and retain specialist pollinators,
• Monitoring reproductive metrics such as pollen deposition, fruit set, and seed viability.

By addressing both ecological and reproductive constraints, restoration efforts can better ensure the persistence of this rare Texas endemic in both natural and managed landscapes.