Soil moisture affects plant–pollinator interactions in an annual flowering plant

(a) Study sites and species

Study sites are located in the Zoige alpine wetland in the eastern Qinghai-Tibetan Plateau, Hongyuan County (32°48′-32°52′ N, 102°01′-102°33′ E), Southwest China, with altitudes ranging from 3400 to 3600 m. We selected three study sites in Hongyuan County (electronic supplementary material, figure S4). In our study sites, pastures are grazed by livestock only during the winter but are otherwise undisturbed communities. Gentiana aristata, grows across this region but has populations that vary in colour (electronic supplementary material, figure S1). The detailed information of climate, soil conditions and vegetation, evaluating the correlation between soil moisture and G. aristata population composition is provided in the electronic supplementary material, Methods S1, figures S1,S5, S6 and S7.

(b) Measurements of reproductive and vegetative traits in different habitats

In studies conducted in 2012, we monitored the reproductive and vegetative traits of G. aristata. In each of our three sites; we selected three adjacent pastures (ranging from 5–10 ha) with dry, intermediate and wet habitats, respectively. We randomly selected ten 2 m × 2 m plots in each of the pastures (details regarding the location of the sites and plots are provided in the electronic supplementary material, figure S1). We monitored soil moisture (0–20 cm probes) in each plot to confirm that habitat designations (see the electronic supplementary material, Methods S1 for the detailed methods of soil moisture measurement). In each plot, we randomly selected five G. aristata individuals to measure the floral and vegetative traits. Note that each plot included only one floral type of G. aristata. In total, we measured the floral and vegetative traits of 150 plants for each population during the peak of the flowering season. For detailed methods of the measurements of floral and vegetative traits, biomass allocation and total reproductive output of the plant see the electronic supplementary material, Methods S1.

After the seeds were collected, composite soil samples consisting of five soil cores, 7.5 cm in diameter and 0–20 cm depth were taken in each plot for soil nutrition analysis. Samples were manually smashed and sieved, plant roots were picked out. The soil was mixed and sampled (more than 500 g) and transported to the laboratory. In total, 90 soil samples were analysed. Total soil nitrogen and total phosphorus were analysed by the Kjeldahl method [24] and spectrophotometric colorimetry [25].

(c) Assessment of plant–pollinator interactions in wet, intermediate and dry habitats

We selected 30 individual plants for each population for measurement (30 individual plants × 3 populations × 3 sites = 270 individual plants; electronic supplementary material, figure S4). Observation periods were uniformly distributed between 9.00 and 17.00 each day. Observers monitored pollinator visitation to the flower from a distance of 3 m. Each individual plant was viewed by one observer for 2 min during each hour period (30 individual plants × 2 min per plant). There were two observers and thus each individual plant was viewed twice per hour in each site. The observers recorded the flower number of each plant. We also used sweep netting to capture the visiting insects during the observation time (voucher specimens are preserved at the Specimens Museum at Mianyang Normal University). The observations were conducted only on sunny days. A total of 28 800 min (480 h) of field observations were accumulated, with 320 min per plant. Visitation rates (R) per flower per minute were calculated as the total number of visits (N_v) divided by the flower number of per plant (N_c), i.e. R = N_v/N_c [26]. At the end of flowering season, five ripe fruits per plant were harvested from each plot. A fruit was considered to be mature if its pod emerged from its closed corolla (pods enclosed by corollas emerge as soon as their seeds mature). For the detailed methods of seed traits see the electronic supplementary material, Methods S1.

(d) Common garden experiment

To rule out the effects of background insect and plant community composition on insect visitation rate of G. aristata flowers, we placed adult plants from each population into the same plant community background in 2015. In late July, we randomly selected 300 plants of G. aristata from site 1 (each population represented by 100 plants) and transplanted them into 150 pots (21 cm × 14 cm, 50 plots were filled with dry, intermediate and wet habitat soil, respectively), each pot had two plants from the same population. We randomly divided all the pots into five groups, each group had 30 pots and 60 plants (20 plants from the pink-flowered, blue-flowered and white-flowered populations, respectively). All pots were placed in an intermediate habitat. See the electronic supplementary material, Methods S1 for detailed methods of monitoring visitation rates. A total of 14 400 min (240 h) of field observations were accumulated, with a total 320 min per pot.

(e) Reciprocal transplant experiment

To determine if the populations of G. aristata in each habitat were ecologically and phenotypically differentiated and potentially selected for higher offspring production within their respective habitats, we conducted reciprocal transplant experiments for all the three groups. We monitored the time of first flowering, corolla size, pollen grain numbers per flower, flower number per plant and plant height, visiting pollinator species and visitation rate from each population in each habitat. For detailed methods of reciprocal transplant experiment see the electronic supplementary material, Methods S1.

(f) Data analysis

We first averaged the soil moisture for each plot. Subsequently, Poisson regression analyses were used to determine the relationships between volumetric soil water content and the plant number for each population per plot. We used linear mixed models or generalized linear mixed models to analyse the effects of habitat on soil nitrogen, soil phosphorus and, vegetative and floral traits, visitation rates and community composition. We analysed which factors are determined to the floral and vegetative traits as well as pollinator visitation rate by a model selection approach. Based on a priori hypotheses summarized in our conceptual framework (electronic supplementary material, figure S3) and the result of best model selection, we built a path model to assess the drivers of bumblebee and fly visitation rates. For detailed information of data analysis see the electronic supplementary material, Methods S1.

(a) Does the flower colour of the Gentiana aristata population vary across habitats?

Plants with pink flowers were only found within the dry habitats, where soil water content ranged from 25% to 35% vwc. Plants with blue flowers were only found within intermediate habitats, where soil water content ranged from 35 to 45% vwc. Plants with white flowers were only found in wet habitats, where soil water content ranged from 50% to 60% vwc (electronic supplementary material, figure S8).

(b) Do floral traits and community composition vary across habitats?

There were significant statistical differences in floral and vegetative traits among populations from dry, intermediate and wet habitats (figure 1; electronic supplementary material, table S1 and figure S9). Plants from intermediate habitats had blue flowers, more flowers, larger flowers, higher pollen production and exhibited a relatively early time of first flowering. Plants from wet habitats produced white flowers, significantly fewer and smaller flowers, fewer pollen grains, taller plants and exhibited a late time of first flowering. Plants from dry habitats produced pink flowers, smaller and more flowers, more pollen, shorter plants and exhibited an earlier time of first flowering.

• Download figure
• Open in new tab
• Download PowerPoint

Plant biomass allocation varied significantly across habitats that varied in soil moisture (electronic supplementary material, figure S7). Plants from intermediate habitats exhibited higher flower biomass fraction (mean ± s.e.: 31.1 ± 1.8%) relative to plants from dry and wet habitats (21.7 ± 2.0%, 23.8 ± 1.8%). In addition, plants from wet habitats have high stem biomass fraction (42.0 ± 0.7%), but plants from the dry habitat have high root biomass fraction (20.8 ± 1.0%).

(c) Do pollinator visitation rates and seed set vary across habitats?

Pollinator preferences varied across all habitats in our study (figures 2 and 3, electronic supplementary material, table S1). Bumblebees (Bombus filchnerae, B. humilis and B. supremus) exhibited higher visitation rates to flowers in dry and intermediate habitats (pink or blue flowers), while flies (Calliphora vicina) had higher visitation rates to flowers in wet habitats (white flowers). Seed set also varied with habitats (figure 1f). Plants from intermediate and dry habitats have a high seed set (0.62 ± 0.02 and 0.54 ± 0.02). By contrast, plants from wet habitats have low seed set (0.43 ± 0.01).

• Download figure
• Open in new tab
• Download PowerPoint

• Download figure
• Open in new tab
• Download PowerPoint

(d) Are plant and environmental traits correlated with pollinator visitation rates?

Bumblebee visitation rate was positively influenced by pollen grain number, plant height and flower colour (figure 3). Fly visitation rate was positively influenced by corolla size and flower colour (figure 3).

Vegetative and reproductive traits (e.g. plant height, flower colour, pollen grain number, flower size and number) were determined by soil water content and soil total nitrogen (electronic supplementary material, table S1), whereas community composition and other neighbour species did not influence the visitation rate of bumblebees and flies (electronic supplementary material, figure S10), suggesting that pollinator visitation rates are indirectly driven by soil moisture and soil nutrition.

(e) Common garden experiment: are differences in visitation rates owing to variation in the pollinator and plant community?

We placed potted flowering plants from the three populations into a common intermediate habitat. Plants with pink and blue flowers attracted only bumblebee spp. (0.60 ± 0.01 visitors flower⁻¹ min⁻¹ × 10^–2 in pink flowers; 0.75 ± 0.02 visitors flower⁻¹ min⁻¹ × 10^–2 in blue flowers, respectively), while white flowers are visited by only fly species (0.40 ± 0.01 visitors flower⁻¹ min⁻¹ × 10⁻²) (electronic supplementary material, figure S10).

(f) Reciprocal transplant experiment: are the populations of Gentiana aristata in the three habitats genetically differentiated?

When plants from dry (pink), intermediate (blue) and wet (white) habitats were transplanted into endemic and nonendemic habitats, there was an interaction between population origin and habitat (electronic supplementary material, figure S11). Plants from wet habitats (white populations) could not grow in dry habitats, even when transplanted as seedlings. Plants from intermediate habitats (blue) produced significantly larger flowers (12.5 ± 0.10) than other populations in intermediate habitats (plants from dry habitats had 9.4 ± 0.12 and plants from wet habitats had 8.4 ± 0.15), while plants from wet habitats (white) produced significantly larger flowers (9.7 ± 0.14) than other populations in wet habitats (plants from dry habitats had 7.2 ± 0.15 and plants from wet habitats had 8.7 ± 0.15; electronic supplementary material, figure S12A). Plants from intermediate habitats produced more pollen grains (11054.2 ± 335.27) than the other population in intermediate (plants from dry habitats had 9903.5 ± 339.71 and plants from wet habitats had 3936.3 ± 294.75) and wet habitats (plants from dry habitats had 7026.1 ± 405.18 and plants from wet habitats had 5390.4 ± 398.21; electronic supplementary material, figure S12B). Plants from intermediate habitats produced more flowers (13.7 ± 0.11) than the other populations in intermediate (plants from dry habitats had 10.7 ± 0.21 and plants from wet habitats had 4.9 ± 0.29) and wet habitats (plants from dry habitats had 6.1 ± 0.15 and plants from wet habitats had 7.0 ± 0.24; electronic supplementary material, figure S12C). However, habitat also influenced growth of plants from dry habitats, with populations of plants from dry habitats producing more flowers in intermediate habitats (electronic supplementary material, figure S12C). There was no variation in plant height or time of flowering among plants from the different populations or grown in different habitats (electronic supplementary material, figure S12D,E).

Plants had higher seed set within their native habitats relative to plants from other populations (electronic supplementary material, figure S12F). Flowers have higher visitation rate within their native habitats relative to flowers from other populations (electronic supplementary material, figure S13). Thus, this suggests that plants are adapted to attract pollinator populations best in their native habitats, which may be a result of colour preferences of the pollinators.

(a) Influence of soil moisture on plant traits

Variation in floral and vegetative traits, including variation in biomass allocation, associated with the three plant morphotypes and habitats probably reflects a trade-off of resource allocation between vegetation and reproduction to deal with water stress [6]. With sufficient soil moisture, plants invest more resources into flower and seed production, while plants exposed to dry or wet stress tend to invest more resources in roots and stems [4–6]. Plants grown under optimal moisture conditions show larger flowers, more flowers, and a higher flower biomass ratio, while plants experiencing conditions of too high or low moisture have lower numbers of flowers and a lower floral biomass fraction, and increased root or stem biomass [5,6,18]. In our study, blue/intermediate population plants had the highest floral biomass, white/wet population plants had the highest stem biomass, while pink/dry population plants had the highest root biomass (electronic supplementary material, figure S9). Intermediate levels of soil moisture were associated with the largest flowers (corolla diameter) and most number of flowers per plant (electronic supplementary material, figure S6). Our reciprocal transplant experiments further demonstrated this trade-off, and indicated there may have been selection on the growth characteristics of the different plant populations. Blue plants that typically grow in intermediate habitats tended to produce more flowers in both intermediate and wet habitats, compared to the pink/white population plants (electronic supplementary material, figure S12).

Pollen grain number is determined by the number of pollen mother cells, which is susceptible to resource and environmental stresses during early flower development [5,11,27]. More pollen grains could be produced once the plant has enough soil moisture and nitrogen availability to mature all pollen mother cells [28,29]. Our results show the high pollen production at intermediate levels of soil moisture and soil nitrogen (electronic supplementary material, figure S6). Again, the reciprocal transplant experiments suggest there may have been selection on the growth characteristics of the different plant populations. Blue plants that typically grow in intermediate habitats produced more pollen grains in intermediate and wet habitats, compared to the pink/white population plants (electronic supplementary material, figure S12).

Soil moisture can influence flower phenology by mediating the time of vegetative growth [30]. Stress from dry and low soil nitrogen conditions can result in abbreviated vegetation growing time, and advanced the time of first flowering [31]. High soil moisture and nitrogen can extend the vegetation growing season and postpone onset of flowering [6,32]. In our field populations, increased soil moisture was positively associated with the time to first flowering and plant height; thus, plants growing in more wet habitats flowered later and had more vegetative growth (electronic supplementary material, figure S6). In the reciprocal transplant experiments, there was no difference among the populations, but there was a tendency for all populations to reach larger plant heights and flower later when grown in wet habitats.

The different populations are probably adapted to the soil conditions from the habitats in which they are typically found. Indeed, previous studies found blue-flowered morphs outperformed white-flowered morphs during years of drought, whereas white morphs survive better during years of high spring precipitation [32,33]. Interestingly, in our study, it was not possible to germinate or transplant seeds or plants from the white-flowering population into dry habitats, suggesting that they could not tolerate these dry conditions.

(b) Variation in pollinator preferences

Plants with pink and blue flowers attracted only bumblebees, while plants with white flowers attracted only flies. This visitation pattern was observed in plants growing in their endemic habitats (figure 3), and in the common garden and the reciprocal transplant experiments (electronic supplementary material, figure S10 and S13). This suggests that preferences are probably driven by floral traits, and not by variation in the background community of plants or pollinators, and this preference is consistent across both specialized pollinators in this system (flies) and more generalized pollinators (bees), though it would be anticipated that background plant community can influence pollinator behaviour [34]. Comparable results were found in the orchid Spiranthes sinensis in the Chinese Himlayas, with a white-flowered form on wet soils, and a pink-flowered form on dry soils. These two forms also had distinct pollinator communities, with white flowers attracting large-bodied bees (Apis cerana & Bombus spp.) and pink flowers attracting smaller bees (Ceratina sp. and Halictidae) [35].

Different traits were associated with bee versus fly preferences in this system. Pollen number, plant height and flower colour were important variables determining bee visitation rate to pink and blue flowering populations, while fly visitation to white populations was influenced by pollen number, corolla size and flower number. Notably, many of these traits are also influenced by the plant genotype and the soil moisture and soil nitrogen conditions. Pollen is the primary source of proteins and lipids for bees for both individual nutrition and for rearing their offspring [36]. Flower-visiting bees often prefer flowers with high pollen production [37], and this may underlie the preference of bumblebees for plants with greater amounts of pollen grains. Interestingly, visitation rates to blue flowers was greater in intermediate habitats, while visitation rates to pink flowers was greater in dry habitats. In the reciprocal transplant experiment, in dry habitats, the plants from the blue flowering morphotype had more pollen and more flowers, and thus should have been more attractive to the bumblebees in these locations. Since the surrounding population of G. aristata is pink in these habitats, and the blue population flowered significantly later, this preference may be the result of learning and experience. Indeed, previous studies demonstrated that when bumblebees have individual experience foraging on a colour with high-quality rewards, they preferred to forage on that colour [38].

Fly visitation patterns were associated with flower size and number of flowers. Flies typically collect nectar and pollen to support flight and reproduction [39]. Though nectar volume was very low in flowers in our system, flies were observed to be foraging for nectar rather than pollen grains (J. Mu 2013, personal observation). Generally, more and larger flowers associated with more nectar [7,38], and this may have driven this preference. Furthermore, the reciprocal transplant experiments demonstrated the increasing of flower number and corolla size could attract flies to visit white flowers in intermediate habitats.

(c) Effect of habitat and plant community composition on plant–pollinator interactions

The three habitats (dry, intermediate and wet meadows) supported distinct plant communities (electronic supplementary material, figure S7). The proportion of sedges was higher in wet habitats, while the proportion of insect pollinated forbs was higher in dry and intermediate habitats. Typically, the primary source of nutrition of bumblebees and flies is from insect pollinated forbs [2,15]. However, there was generally lower visitation rates of both bees and flies in wet habitats versus intermediate habitats (electronic supplementary material, figure S13; note that flies were not observed in the dry habitats, but that may have been because they were only attracted to white flowers and the white flowering population of plants could not grow in dry habitats). Thus, the community of bumblebees may have been lower in wet habitats. Most bumblebees nest in underground cavities, wet sites may have had reduced nesting opportunities [40]. Alternatively, the bees and flies may be living in the wet habitat but flying to the intermediate habitats to forage.

(d) Evidence for adaptation to local habitat and plant–pollinator community conditions

Plants and their pollinators are engaged in a mutualism where pollinators gain food resources in the form of pollen and nectar and in the process plants efficiently outbreed by pollinators vectoring their gametes [41]. Interestingly, seed set was highest in the endemic habitat of each colormorph (electronic supplementary material, figure S12). This is despite the fact that some plant traits preferred by pollinators (such as high pollen number) might be higher in a transplanted population: for example, the pink populations transplanted to intermediate habitat produced higher pollen number than in their native habitat, but attracted fewer bumblebees and had lower seed set. In wet habitats, white flowering plants attracted only flies and exhibited equivalent visitation as blue/pink plants, which attracted bumblebees. While bumblebees should carry and transfer more pollen [42], white flowering plants still had higher rates of seed set. Thus, these plant populations seem to have adapted to their habitats, including soil conditions, plant communities and available pollinators, to effectively set seed.

Clearly, the major difference between the populations of G. aristata growing in these different habitats is flower colour. This variation in colour is probably associated with changes in the anthocyanin pathway [43]: loss of function of this pathway results in white flowers versus pink flowers in petunia species, which then results in altered pollinator attraction [42,44]. Stress, including from variation in soil moisture, can affect the activity of the anthocyanin pathway, which may have results in variation in colour across habitat types. Subsequent selection by a bumblebee versus fly-dominated pollinator community may have results in selection for a mutation in this pathway, which resulted in white flowers in wet habitats which are attractive to flies. While the blue and pink colour morphs appear to be equally attractive to bumblebees, it may be that there was selection for a dominant morph in a particular location. Selection for different plant growth characteristics (biomass allocation, number of flowers and time to flowering) may have limited the ability of these populations to invade other habitats. Flower colour polymorphism is associated with pH value. However, in the reciprocal transplant experiment, we did grow the populations in the different soils and the colour differences remained, so it does not seem that pH influences flower colour in this system.