Salvia umbratica Hance. (Lamiaceae) is an annual herb with erect stems of 1–1.2 m in height. It grows between 600 and 2,000 m a.s.l. on hillsides, and in valleys or roadsides in northern China where it is native to Anhui, Gansu, Hebei, Hubei, Shaanxi, and Shanxi provinces. The type specimen of the species was collected from Beijing (Wu and Li, 1977; Li and Hedge, 1994). The flower of S. umbratica is typical of Salvia being composed of a bilateral symmetrical corolla (Figure 1), the upper lip (hood) of which is formed from two petals, while the lower lip (palate) develops from the remaining three petals. The two stamens are each composed of two anther thecae and two connective arms attached to a filament by a joint (Huang et al., 2014). The upper anther theca (upper theca) of each stamen is positioned in the hood of the corolla just below the style of the pistil and is connected to the lower anther theca (lower theca) via the connective arm. A much larger quantity of pollen is produced in the upper than the lower thecae. In fully developed flowers, the stigma is bifid and protrudes slightly from the upper lip with its receptive surface facing outwards. At the base of the corolla tube is a disc containing a nectary, above this is an ovary containing four ovules (Strelin et al., 2017).

Field studies were conducted in 2018 and 2020 at the Beijing Forest Ecosystem Research Station (BFERS) located on Dongling Mountain (115°25′33^′′ E, 39°57′29^′′ N, elevation 1,200 m) where a semi-humid and monsoon climate prevails, with an average annual temperature of 5–10°C and annual precipitation of 500–650 mm (Xu et al., 2018). The studies included observations on insect pollinators and amount of pollen deposited during a single pollinator visit, the process of DAS in the absence and presence of pollinators, and seed set following various pollination treatments.

To confirm further the occurrence and prevention of DAS and the reproductive fitness of S. umbratica, greenhouse studies were conducted at the Shanghai Chenshan Botanical Garden (SCBG) on plants transplanted in May 2020 from the wild at the BFERS. These studies investigated the amount of pollen required for sufficient pollination and high seed set, the prevention of DAS by prior sufficient pollination, and levels of seed set resulting from DAS in the absence of pollinators. Both field and greenhouse studies were conducted in August and September during the natural blooming period of S. umbratica.

Pollinator visits to S. umbratica flowers were recorded in the field during the natural flowering period on sunny days on the 27, 28, and 31 of August, 2018. For this purpose, we established two 1 ×1 m quadrats at the field site, each of which contained four or five individual plants of S. umbratica and an average of 26 and 28 blooming flowers. Two SONY HDR-CX510E cameras were used to record the types, numbers, number of visits, and visit duration of flower-visiting insects in each quadrat between 08:00 and 17:00 each day. We noted the behavior of insect visitors within the quadrats, recording legitimate visits to flowers, i.e., where a pollinator contacted the stigma and stamens while searching for nectar (see Li and Huang, 2009; Ma et al., 2014), and illegitimate visits, i.e., where a pollinator robs nectar and pollen without pollinating the flower. Specimens of pollinators were captured with an insect-net and stored in ethanol (70%) for accurate identification. Vouchers were deposited at the entomological collection at the Horticulture Department of SCBG.

To detect the amount of pollen deposited on a stigma during a single pollinator visit, 45 flower buds were randomly selected from 15 S. umbratica plants (3 flowers per plant) that grew naturally at the field site outside the quadrats. Each bud was enclosed in a nylon mesh bag which was removed immediately after the bud opened allowing pollinators to visit the flower produced. After a single visit by a pollinator, the pistil was immediately removed from the flower and pollen deposited on the stigma was counted using a STV-120m portable microscope (Japan, Kenko).

The process and timing of DAS in flowers of S. umbratica were investigated first in the absence of pollinators. Two flower buds from each of 10 plants of S. umbratica growing naturally at the field site were placed in nylon mesh bags. Photographs of the development of the stigma and upper thecae of each flower were taken using a Nikon D300s digital camera, and records of the distance, D (see Figure 1B), between the stigma and upper thecae (measured using Photoshop CS3, Adobe, USA), were made when each flower first opened (0 h) and thereafter at 24 h intervals until DAS was completed, i.e., when D = 0. Flowers were deemed to have first opened when the upper and lower lips of the corolla mouth had just separated. This experiment was then repeated with minor modifications in the presence of pollinators. Thus, another two flower buds from each of 10 different plants of S. umbratica were placed in nylon mesh bags with a record taken of distance between the upper thecae and the stigma when each flower opened (0 h). Bags were then removed from flowers to allow pollinators to visit them. After a pollinator had entered the corolla tube and successfully pollinated a selected flower, stigma-upper thecae distance was measured again and at intervals of 24 h thereafter.

To confirm the breeding and mating system characteristics of S. umbratica, flowers on a number of different plants in the field were subjected to five different pollination treatments after which fruit and seed set was recorded per flower. These treatments were: (1) autogamy, involving bagging flowers to allow autonomous selfing to occur; (2) simulated geitonogamy, involving hand-pollination of emasculated and bagged flowers using pollen from another flower of the same plant; (3) simulated xenogamy, as in (2) above but using pollen from a different plant; (4) detection of apomixis, involving emasculation and bagging of flowers to determine if seed were produced; and (5) open-pollination, to determine seed set in flowers left to be naturally pollinated. Each of the five treatments was repeated on three flowers on each of 30 plants (90 flowers per treatment), with fruit production and seed set (number of seeds/4 ovules) per flower recorded 3–4 weeks after pollination had occurred.

To determine the amount of pollen required for sufficient pollination and fertilization of ovules in a flower, flowers on a number of greenhouse cultivated plants were subjected to five different pollination treatments using a mixture of pollen grains. Pollen was obtained from an equal number of newly dehiscent anthers from a flower on the same plant and a different plant, mixed on parchment and deposited on stigmas using a dissecting needle. The number of pollen grains deposited was determined using a STV-120m portable microscope (Japan, Kenko). The five treatments involved depositing on the stigma of a flower 1–3, 4, 5–10, 11–20, or >20 pollen grains from the mixture. Each treatment was applied to two flowers on each of 10 plants, i.e., 20 flowers per treatment, with fruit production and seed set per flower recorded 3–4 weeks after pollination had occurred.

To test the hypothesis that sufficient prior pollination by a pollinator likely prevents DAS from occurring, 40 flowers across 10 greenhouse cultivated plants were hand-pollinated immediately after they opened in either of two ways using a mixture of pollen grains. In 20 flowers, 1–3 pollen grains were deposited on their stigmas, whereas in the other 20 flowers >20 grains were deposited. Stigma-upper thecae distance in flowers (D) was recorded at the stage they were pollinated and at 24 h intervals thereafter. Because Salvia produces four ovules per flower, for 20 flowers the number of pollen grains deposited on a stigma was less than the number required to fertilize all ovules (i.e., there was pollen-limitation), whereas for the other 20 flowers the number of pollen grains deposited on the stigma was more than required to fertilize all ovules (i.e., no pollen-limitation; Knight et al., 2005; Jorge et al., 2014).

In a follow-up study, the effect of DAS on seed set was determined on five plants cultivated in a greenhouse in the absence of pollinators. Flower number per plant averaged 174 and ranged between 48 and 244 across the five plants. During the period from the first flower opening to the last flower dropping, we recorded at 24 h intervals the proportion of flowers on a plant that had completed the process of DAS (when stigma-upper thecae distance was zero), and subsequently the fruit and seed set of these flowers.

All statistical analyses were performed using SPSS 22.0 (Chicago, IL, USA). The analysis of differences in mean stigma-upper thecae distances at different times in flowers pollinated with 1–3 pollen grains was conducted using a generalized linear mixed model (GLMM), with distances as dependent variable, time as fixed effect, and plant as a random factor. A t-test was performed to analyze differences of stigma-upper thecae distances at the time of pollination and 24 h later in pollination treatment with >20 pollen grains. Differences in mean fruit set and seed set among different pollination treatments were analyzed by one-way ANOVA followed by a post-hoc Tukey's test after excluding treatment 4 (emasculation and bagging of flowers produced no seed indicating an absence of apomixis).

Over the 3 days that observations were made, a total of 174 legitimate pollinator visits to flowers in quadrats were recorded. These were by the bumblebees Bombus opulentus, B. longipes, B. consobrinus, and B. hedini. On entering the corolla tube, bumblebees searched for and fed on nectar. In doing so, pollen on their backs was deposited on the stigma. Furthermore, they pushed against the lower connective arms of the thecae (Figure 1), causing the upper connective arms to bend downwards, resulting in pollen from the upper thecae being deposited on their backs (Figure 2E). On average, the number of visits per flower was 1.23 ± 0.31 per day with visits per flower lasting 2.19 ± 0.09 s.

Counts of pollen grains deposited on the stigma of a flower during a single pollinator visit after the flower opened showed that, on average, 26.31 ± 2.07 (11–65 pollen grains per stigma, n = 45) pollen grains were deposited, which is more than enough to result in a high level of fruit and seed set (see below).

Based on observations made on bagged flowers (Figures 2A–D) and measurements of distance between the stigma and upper anther thecae from the time a flower first opens until DAS is completed (D = 0) (Figure 3A), it was evident that DAS occurs in S. umbratica when flowers are not pollinated by insects. ANOVA showed that stigma-upper thecae distance decreased significantly with time [F_(5,74) = 8.808, p < 0.001]. At 24 h after a flower first opened the style had begun to recurve causing the receptive stigma surface to become angled toward the upper thecae (Figure 2C). At this stage, mean stigma-anther thecae distance was reduced from the initial mean of 3.17 mm at the time of flower opening (Figure 3A). Recurvature of the stigma toward the thecae continued at different rates in different flowers. In 70% of flowers (14 out of 20), DAS (D = 0) was completed by 72 h after the flower first opened. However, in three flowers it took 120 h to complete. On average, DAS took 66.7 h to complete across the 20 flowers examined. Corollas began to fade in color and dry out after styles started to recurve (Figures 2C,D). The entire sequence of events leading to DAS are shown in Supplementary Video 1.

For flowers that were accessible to pollinators, mean stigma-upper thecae distance was 3.11 ± 0.80 mm at time of flower opening (0 h) (Figure 3A). All marked flowers were visited and pollinated by insects (Figure 2E) during the following 24 h (mean time to successful pollination equaled 19.7 h) and by 24 h mean stigma-upper thecae distance was 3.52 ± 0.59 mm (Figure 3A). The two means were not significantly different from each other (t = −1.796; p = 0.081). The receptive surface of the stigma remained facing outwards from the corolla with large amounts of pollen deposited on it (Figure 2F). The corolla quickly dropped from a flower after a pollinator visited it (Figure 2G) and the floral processes leading to DAS were terminated.

No fruits and seeds were produced by emasculated and bagged flowers, and it was concluded, therefore, that apomixis did not occur in the species. Fruit set for all other treatments was 100% (Figure 4A). High levels of seed set were recorded across these other treatments (Figure 4A); however, a one-way ANOVA showed that differences were significant [F_(3,324) = 7.163, p < 0.001]. Post-hoc pairwise comparisons of means using Tukey's test indicated that mean seed set of bagged flowers (90.38%, resulting from DAS) was significantly lower than mean seed sets of flowers subjected to either cross-pollination by flowers of the same plant (98.03%, geitonogamy, p = 0.002) or different plants (98.29%, xenogamy, p = 0.001). However, mean seed set of bagged flowers was not significantly different from that of open-pollinated flowers (94.88%, p = 0.098), which, in turn, was not significantly different from the means of the two other treatments. Flowers subjected to the geitonogamy and xenogamy pollination treatments had equivalent high seed set (p = 0.999).

High levels of fruit set (100%) and seed set (>80%) resulted when the number of pollen grains deposited by hand on a stigma of a bagged flower was 5–10 or greater, but were reduced to 45.00 ± 11.12% and 22.50 ± 6.59%, respectively, when 1–3 grains were deposited [Figure 4B; One-way ANOVA, Fruit set, F_(4,95)= 12.579, p < 0.001; Seed set, F_(4,95) = 20.432, p < 0.001]. There was no significant difference for fruit set when 4 or more pollen grains were deposited on a stigma (80 vs. 100%, p = 0.235), however seed set increased when >4 grains were deposited (Figure 4B).

In flowers in which 1–3 pollen grains were deposited on the stigma, mean stigma-upper thecae distance decreased gradually over time (Figure 3B) in a similar, though more rapid, way to when pollinators were excluded from flowers (and presumably zero pollen grains were deposited on stigmas) (Figure 3A). Thus, DAS was completed in 14 of the 20 flowers by 48 h after they opened, and in all flowers by 96 h, and on average took 55.2 h to complete. This average time to completion was not significantly different to that (66.7 h) estimated for bagged flowers where access by pollinators was prevented (t = 1.807, df = 38, p = 0.079). In marked contrast, DAS did not occur in flowers in which >20 pollen grains were deposited on stigmas. Mean stigma-upper thecae distance in these flowers remained >1 mm at 24 h after they were pollinated, and was not significantly different from D at the time stigmas were pollinated (Figure 3B; t = 2.00; p = 0.053). After 24 h, the corollas of these flowers quickly dropped, indicating that sufficient prior pollination and fertilization had occurred preventing DAS from occurring.

Records made on bagged flowers of five plants showed that DAS was completed after flower opening in ~17% of flowers by 24 h, a further 52% of flowers by 48 h, and another 24% by 72 h (Figure 4C). Thus, DAS was completed in ~94% of flowers by 72 h after opening, with only a small proportion of flowers taking longer to complete this process. There was no significant difference in fruit set (all 100%) and seed set (always >80%) of flowers in which DAS was completed at different times over the 24–120 h period that records were taken [Figure 4C; One-way ANOVA, seed set, F_(4,765) = 0.813, p = 0.517].