Study System

Fieldwork was conducted on Mertensia ciliata (James ex Torr.) G. Don (Boraginaceae) in a subalpine meadow along Rustler Gulch in Gunnison National Forest (38°59′32.68″ N, 107°00′23.16″ W; 3,009 m.a.s.l.) located 4.3 km from the Rocky Mountain Biological Laboratory (RMBL) in Gothic, Gunnison County, Colorado, United States. Between 1973 and 2006, mean spring (April–June) temperatures at RMBL have increased by 2.0°C, and the average date of spring snowmelt has advanced by nearly 2 weeks (Miller-Rushing and Inouye, 2009). In many subalpine systems, both flowering phenology and summer water availability are largely driven by spring temperatures, snowpack depth, and snowmelt timing (Wielgolaski and Inouye, 2013). In the southern Rocky Mountains, over the next century, temperatures are expected to continue to increase, while both winter snow fall and total precipitation are expected to decrease (Overpeck and Udall, 2010; Pederson et al., 2011), resulting in earlier snowmelt timing (Saunders et al., 2008) and earlier, longer dry seasons prior to mid-summer thundershowers.

Mertensia ciliata, the tall-fringed bluebell, is an herbaceous, rhizomatous perennial of the subalpine and lower alpine zones of the Rocky and Sierra Nevada Mountains. Plants form compact clones of a few to several hundred flowering ramets and are commonly found along streams and in wet meadows (Pelton, 1961). The flowers are pendant and tubular, expanding to a wider, lobed mouth, and are borne in dense clusters of cymes along leafy stems. Flowers are typically open for 6 days, with receptive stigmas throughout flowering and can produce a maximum of four one-seeded nutlets (hereafter seeds). The seeds have elaisomes and so are likely ant-dispersed.

Pollen is usually removed within 24–48 h of anther dehiscence by medium and long-tongued bumblebees, including Bombus balteatus (Dahlbom), B. bifarius (Cresson), B. flavifrons (Cresson), and B. frigidus (Smith) (Geber, 1985; Suzuki, 1994; Gallagher and Campbell, 2017). Mertensia ciliata flowers are also visited by flies (Bombyliidae, Muscoidea, and Syrphidae) and solitary bees (Colletidae: Colletes p. paniscus Vier. and Megachilidae: Osmia spp.) (Pelton, 1961; Gallagher and Campbell, 2017). Although M. ciliata is self-compatible, seed set is dependent on insect pollination, with flowers prevented from receiving an insect visit rarely producing seeds and averaging only 8% as many seeds as those open to insect pollination (Geber, 1985).

Mertensia ciliata flowers from late June through early August in the subalpine meadows around RMBL. An individual plant flowers for 3–5 weeks (Geber, 1985). As with many species in this region, M. ciliata’s flowering phenology is strongly correlated with the timing of snowmelt (Miller-Rushing and Inouye, 2009; Wielgolaski and Inouye, 2013). In fact, between 1973 and 2006, the average date of first bloom (DFB) has advanced by 3.3 days per decade (Miller-Rushing and Inouye, 2009). Over the same 34-year period, M. ciliata has become less common at lower elevations (≤2,900 m a.s.l.), and observed declines in peak floral abundance (15 fewer flowers per decade) correlate with earlier snowmelt timing (1.6 fewer flowers per day earlier snowmelt) (Miller-Rushing and Inouye, 2009). In the plant communities surrounding RMBL, phenological responses to warmer spring temperatures and early snowmelt are producing a longer mid-season dip in floral abundance (Aldridge et al., 2011), which may impact pollinator abundance and pollination success of summer-blooming plants like M. ciliata. Over a somewhat longer record of 39 years, its date of peak bloom advanced by an amount (1.6 days per decade) fairly similar to the average advance in peak bloom near RMBL (2.5 days per decade; CaraDonna et al., 2014). With some species advancing quickly and others less so, M. ciliata will overlap more with some species and less with others in the future. Overall, these patterns suggest that changes in spring temperatures and snowmelt timing may alter M. ciliata pollination and reproductive success through changes in water availability and flowering phenology.

Previous field experiments with M. ciliata revealed that both changes in water availability and changes in flowering phenology can independently affect pollination success in this species. Floral responses to experimental changes in water availability altered pollinator visitation rates, but the effects were non-linear, with visitation peaking at intermediate water levels (Gallagher and Campbell, 2017). In a separate experiment, shifts in timing of flowering onset affected the pollination of plants, such that early-flowering plants received a higher frequency and diversity of pollinator visitors than did late-flowering plants. Those pollinators that visited late-flowering plants were more effective pollinators than their early-season counterparts, resulting in no net difference in seed set between early- and late-flowering plants (Gallagher and Campbell, 2020). In this system we expected that water availability would have less effect on pollinator visitation early in the season, when pollinators are abundant and diverse, allowing visitation to all size flowers and perhaps less choosy visitors due to competition for nectar. But later in the season, when visits are dominated by worker bumblebees, including B. flavifrons which shows preferences based on flower size (Campbell et al., 2014), we predicted that differences in water-mediated floral traits (e.g., size, nectar) would more likely impact pollinator visitation rates.

Experimental Design

To test the extent to which effects of phenology on pollination and reproductive success vary with co-occurring changes in water availability, we manipulated both flowering onset and water availability in a factorial experiment using potted M. ciliata plants. Between 2012 and 2015, 120 plants were collected from a large M. ciliata population in Rustler Gulch and potted using native soil in 2-gallon plastic pots (actual volume: 1.593 Gallons, Nursery Supplies, Inc.). A portion of the plants were used for other studies in 2013 and 2015, but otherwise remained in trenches and overwintered in the ground under snow at RMBL until their use for this experiment. In 2016, potted plants were randomly assigned to one of three water treatments, dry, average, and wet. To inhibit flowering, potted plants were moved to Schofield Pass (39°00′54.98″ N, 107° 2′49.40″ W; 3,263 m.a.s.l.) in early June, where they were placed in a shaded snowbank under a mesh shade-shelter. Each week, 30 randomly selected plants, 10 per water treatment, were moved back to RMBL, where the higher light and warmer temperatures at low elevation induced them to flower at the experimentally chosen time (typically within 5–7 days after being moved to RMBL).

Each week, 30 plants in their first week of flowering (10 per water treatment) were moved from RMBL to a meadow near the original source population in Rustler Gulch. Plants were arranged 30 cm apart into five randomized arrays of six plants, set 30 cm apart in a single line, with 2 m between arrays. Each array included two plants of each water treatment. In week 4, 12 plants stopped flowering mid-week and therefore we re-randomized the plants that had flowers remaining into three new arrays with six plants each. Where appropriate, we calculated the mean values per array of week four plants before and after the plants were rearranged, and then averaged those two values for each plant. To create distinct experimental populations, arrays were located 50 meters away from unmanipulated M. ciliata populations. Plants remained in the arrays for 1 week before being replaced by the next group. This period was the only time when their flowers were available for pollination. A total of 114 plants flowered and were included in the experiment, for a total of four phenology treatment groups spanning 4 weeks (June 20–July 17).

The water manipulations were maintained through the growing season (June 10–August 1) and discontinued once seeds were collected. We watered pots manually with watering cans slowly and evenly to avoid pooling, in the mid to late afternoon to coincide with the timing of July thundershowers. Throughout the experiment, we measured soil moisture as volumetric water content (VWC) every third day using a 12 cm Campbell Scientific “HydroSense” probe inserted into the center of each pot (if done on the day of watering, always before applying water). We used these VWC measurements to maintain soil moistures within the pots at levels that correspond with VWC levels in a previous water manipulation experiment with M. ciliata (Gallagher and Campbell, 2017) performed at a nearby site with similar soil. In that study, plants within naturally occurring populations received either 50% reduction in precipitation (hereafter “dry”), additional rainfall equal to the historic average rainfall during July from 1990 to 2009 (hereafter “wet”; based on long-term data reported in Campbell and Wendlandt, 2013), or ambient conditions (“control”). In that previous field study, VWC had averaged 9.4, 11.7, and 15.4%, respectively. In the end, we watered wet pots daily, control pots every other day, and dry pots every third day, achieving 10.4 ± 0.4%, 12.7 ± 0.4%, and 17.5 ± 0.6% average VWC in the soil for dry, control, and wet pots respectively (Mean ± SEM). Whereas these values for VWC resembled those for in situ plants in our earlier study, the measured values may slightly underestimate actual average soil moisture depending on the relationship between soil moisture and time between watering and measurement, as well as the impact of variable rain events. Average VWC values for plots were analyzed with a linear mixed model with the main and interactive effects of water treatment and phenology week as fixed effects, and array nested in phenology week as a random effect. The resultant gradient in soil moisture did not vary significantly among phenology weeks [Water: χ²₍₂₎ = 106.3, P < 0.0001, Phenology: χ²₍₃₎ = 5.2, P = 0.2, Water × Phenology: χ²₍₆₎ = 1.44, P = 0.96, Figure 1A]. Note that we manipulated both factors within a realistic range of natural variation. The four phenology weeks corresponded closely with the range of flowering time onset in nearby natural populations (Gallagher and Campbell, 2020). The water treatments of adding 100% or subtracting 50% of average rainfall fell within the range of summer precipitation over the past few decades (Campbell and Wendlandt, 2013).

Measurements of Reproductive Traits, Pollinator Visitation, and Reproduction

For each phenology week, we measured reproductive traits, including total abundance of flowers open during the phenology week, corolla size, and nectar volume and sugar concentration. We measured corolla width at the opening of the tube and corolla length from the base of the calyx to a randomly chosen corolla lobe for an average of 4.4 ± 0.4 flowers per plant. Corollas were measured on the second or third day of each phenology week. All flowers within an array were measured at the same time, and the order in which arrays were measured was randomized each week. At the end of each week, after pollination observations were complete, individual flowers in each phenology group were labeled and all flowering stems were bagged with fine mesh jewelry bags (Uline, Pleasant Prairie, WI, United States) to prevent further pollination and loss of seeds, and to provide a count of the total number of flowers open during that phenology week. For plants with flowers remaining at the end of each week, we measured nectar volume and percent sugar concentration 48 h after plants were bagged (N = 71 plants). For an average of 2.6 ± 0.2 flowers per plant, we measured nectar volume using 5 μl microcapillary tubes (Kearns and Inouye, 1993) and percent sugar concentration using a handheld nectar refractometer (Bellingham + Stanley Ltd., Basingstoke, Hants, United Kingdom). No flowers remained for nectar measurements in week 4; therefore, we only include nectar data from phenology weeks 1–3 in our analyses. To select which flowers to measure, we marked the calyx of buds to track when flowers opened and then chose at random among available flowers of the same age. For each floral trait, we calculated the mean trait value of each potted plant, to be used as the response variable in our analyses.

Plants in each phenology group were open to pollination for 1 week. During that time, we conducted pollinator observations and tracked pollinator identity and the number of flowers visited during multiple 30 min observation periods between the hours of 9:00 and 16:00. At the beginning of each observation period, we counted the number of open flowers per potted plant. We calculated mean pollinator visitation rate per plant as (total number of flowers visited)/(number of flowers available per hour of observations) averaged across the phenology week. Visitors were counted as pollinators if they crawled inside the flower corolla. For each 150 min round of observations to the five arrays we randomized the order of observations among arrays. In total, we completed 25 h of pollinator observations per phenology group.

During 100 h of pollinator observations to six plants at a time, we observed 340 floral visitors to experimental plants. The most common pollinators, bumblebees (Bombus spp.) and solitary bees (Osmia spp.), accounted for 92.6% of floral visitors, with flies (Muscoidea and Syrphidae 7.1%) and a moth (0.3%) making up the rest. We excluded the moth from our analyses. For a metric of pollinator type, for each potted plant we calculated mean percent of visitors that were bumblebees as (100% × number of visitors that were bumblebees)/(total number of flower visitors per hour of observation) averaged across the phenology week.

All plants remained in the field until seeds were collected to standardize conditions after pollination exposure. We counted the total number of seeds produced per marked flower (as described by Forrest and Thomson, 2010). We calculated the average seeds per flower for each potted plant as (number of mature seeds/number of flowers). Mature seeds from tagged flowers were collected in coin envelopes and transported to the University of California, Irvine to be weighed. We calculated mean seed mass for each plant as (mass of collected seeds/number of collected seeds). Seed mass included the mass of the elaiosome. All field procedures followed RMBL permitting guidelines. 

For details, please see "gallagher&campbell_frontiersinecology&evolution_metaDataFile.docx"