MATERIALS AND METHODS

Propagation

Accessions (population level seed samples) of O. eriopoda were sourced from the National Seed Bank of the Australian National Botanic Gardens Canberra, The Australian Botanic Gardens Mount Annan, The Royal Tasmanian Botanical Gardens, and Royal Botanic Gardens Victoria and germinated as described in Satyanti et al. (2019). We included seedlings from 16 populations (Figure S1 and Table S1) that represented the range of germination strategies expressed in this species. We aimed for a balanced representation of germination strategies based on early assay results, however there were many populations with postponed germination so we included more of those to be representative. Consequently, we had three populations with an immediate germination strategy, three with a staggered strategy, seven postponed and three with a postponed-deep germination strategy (Table S1). Seeds from each population were germinated in two batches so that germination of early (autumn) and late (spring) seedlings coincided and seedling age was consistent regardless of germination strategy (Appendix S1). This design addressed the potentially confounding factors of age, size and starting time of the experiment and allowed us to manage these factors within the logistical constraints of a blocked design (Lu et al., 2016). Seeds germinating on agar were transplanted within 1-2 weeks to soil media in pots and grown under temperatures conducive to growth. Seedlings were grown until they were 18 – 20 weeks old when they had 8 ± 3.6 (SD) leaves and we inferred that the establishment stage was complete (Appendix S1). AsO. eriopoda is slow growing, this establishment step was important to ensure that plants across all populations were large enough to withstand transplant stress and thus to avoid confounding soil warming effects.

Experimental design

Soil temperatures are particularly important for alpine species because given their low stature, soil temperatures determine the microclimate to which plants are exposed more than the air temperatures do (Körner, 2003; Reinhardt & Odland, 2012; Scherrer, Schmid, & Körner, 2011; Wolkovich et al., 2012). We therefore focused our study on the effect of soil warming on plant traits.
For each population, ten individuals were randomly allocated to each of the two soil warming treatments. We assessed initial seedling size and confirmed that there were no a-priori differences in the size of plants between ambient and warm soil treatments (linear mixed model with germination strategy as a random term; p = 0.286 and 0.52 for leaf number and leaf length, respectively). For the staggered germination strategy, both early (autumn) and late (spring) seedlings, were placed in the respective block, thus, these populations had 20 individuals/soil warming treatments. One representative population of each putative germination strategy was assigned to each of the four blocks. The imbalance of populations for each germination strategy was not an impediment for the analyses. There were five populations for which we did not have 20 seedlings and for these we assigned half of the available number to each warming temperature. Extra plants were placed in the empty spaces on the bench to maintain homogeneous spacing across the experiment but were not included in analyses (number in grey, Figure S1).
Warm and ambient benches were set up in a glasshouse with air temperature set to follow the seasonal changes and natural photoperiod (Fig. S2). The targeted air temperature day/night sequence was 20/10 °C (autumn), 5/5 °C (winter), 20/10 °C (spring), 25/15 °C (summer), and finally 20/10 °C (autumn). Soil warming was achieved by placing a heating mat (Electronic Foil Panel with Thermostat, ADLOHEAT, Victoria) on a given bench set to be continuously ~ 5 °C warmer than the set glasshouse air temperature throughout the experimental period, including winter. The 5 °C soil temperature increase is based on Australian alpine mean air temperature predictions for 2050, i.e. an increase of +0.6 to +2.9 °C (Hennessy et al., 2003) and that the maximum soil temperatures in Australia are to increase by almost double that of air temperature by the end of 21st century (Ooi, Auld, & Denham, 2009). The 5 °C, thus, falls between the predicted +1.2 to 5.8 °C soil temperature increase by 2050. The ambient treatment was located on benches that had mats but no heat, paired one each with the four heated benches. A frame of 6 mm-thick PVC sheet was placed around each bench, 17 cm above the mat, and a 5 cm thick sheet of polystyrene foam was placed on the top of this to insulate the soil. Square openings in the polystyrene matched the pot size and held these in place in the frame. Temperature at plant level (15 cm above bench; Fig. 1) and soil (2, 8 and 14 cm below surface) was monitored during the course of the experiment with i-Button data-loggers (Thermochron DS1921G, Temperature Technology, Adelaide) in each block (32 in total, Fig 1). We analysed the temperatures at plant and soil level to determine efficacy of the design using ANOVA and found significant warming differences for each of the soil and air depths/height. The temperature difference between ambient and warm soil at 8-cm below the surface (where most roots were located) was approximately 6 °C during the day and 9 °C during the night (Fig. 1), in agreement with climate patterns which show that night-time temperatures have increased more than day-time temperatures (Donat & Alexander, 2012; Easterling et al., 1997). With warmer air temperature and reduced snow cover, the soil in the Australian Alps becomes warmer, evidenced by a snow removal experiment (Slatyer, 2016). Thus, we also increased the soil temperature during winter for the warmer soil scenario. In summer soil was warmed to ~35 °C which is realistic for the Australian mountains where bare dark soils can easily exceed 45 °C on sunny days and the difference to the nearby vegetated soils can be over 30 °C (Slatyer, 2016).
The irrigation system was set to keep plants adequately-watered. We used an automatic Water-Pro vapour pressure deficit (VPD) to provide an automatic watering system (MicroGrow Green House Systems, Temecula, California), with each plant being watered individually by a dripper at soil level. Drippers were calibrated to a standard flow rate that was checked at the beginning of the experiment. Watering events were triggered when pre-set VPD targets were reached. Plants received ~ 90 ml per watering which was enough to saturate the soil at the start of the experiment. As the plants grew, we adjusted the VPD point, based on the two sensors for ambient and warm soils, so that the drippers delivered water sufficiently more frequently to the plants in the warmed treatment to keep all plants healthy.

Trait measurement

Leaf number and the length of the longest leaf were recorded for all plants at the start of the experiment, and at the end of autumn (day 38). Those traits were recorded for a subset of plants (4 individuals per soil temperature and population) in early spring (day 124). Leaf increment rate at the early vegetative stage was calculated from the difference of leaf number at the end of autumn from the start divided by 38 (number of days from planting). Leaf increment at the transition to reproductive stage was the difference between total leaf number at day 38 and day 124, divided by 86 (number of days between measurement).
Specific Leaf Area (SLA) is an indicator of resource allocation and ecological strategy. We expected that plants growing faster in warmed conditions would have higher SLA. SLA was measured for every individual at the start of winter (day 52). SLA was measured by acquiring the youngest fully expanded leaf for each individual. The leaf was then scanned on a flatbed scanner, dried at 60 °C for 72 hrs and weighed. The SLA was calculated as area/weight (cm2g-1).
In early spring, a random sample (four plants, even numbered individuals, per population per soil treatment) was photographed with a reference scale over a white Styrofoam board to determine total canopy area. Each plant image was then converted to an 8-bit graphic file. The threshold was adjusted so that only the actual canopy area was detected for selection and using the known distance of the reference scale we calculated the plant area. The image analysis of the canopy area was performed in ImageJ (Schneider, Rasband, & Eliceiri, 2012).
Plants were monitored every 1 – 2 days for phenology. The date at which the first inflorescence with closed buds emerged from the plant base was recorded as first flowering. All infructescences were collected at the point of natural seed dispersal when all seeds were brown or purple-brown and easily dislodged, and the date of collection was recorded for each infructescence. Infructescences were stored at 15% RH and 15 °C. Total number of inflorescences and infructescence were determined for each plant. Further, we measured individual seed mass for five, even numbered, individuals per population per soil treatment by weighing three replicates of 25 seeds (Figure S1). The date of collection of the last infructescence was recorded as the end of seed dispersal, and the time between the first and the last seed harvest was defined as the duration of seed production. Some plants naturally senesced over the course of the experiment and the date of senescence (all leaves browned, no new leaves emerging) was recorded. At the end of the experiment, aboveground biomass was harvested for all plants, dried at 60 °C to constant weight, and weighed. Plants that died before harvest were sampled for biomass within a week of senescence.

Transgenerational effect on germination traits

To minimise potential genetic influences that may have occurred from cross-pollination of populations, populations were separated by organza fabric sheeting supported by a plastic frame. Seed was collected from plants at the time of natural dispersal and stored separately at the plant level.
Seeds from parent plants of three germination strategies: immediate, staggered, and postponed were used to test whether there was an effect of soil temperature during development on the germination strategy and response to germination temperature of progeny. Ten populations were selected to represent the three germination strategies, three immediate, three staggered and four postponed strategies (Figure S1). For each population and soil temperature we selected 50 seeds produced in the peak seed production period from each of five fruiting individuals per warming treatment. Twenty-five seeds each were sown on each of two petri dishes, and randomly allocated to one of two germination chambers set at 25/15 °C and 30/20 °C 12/12 hours photoperiod for 9 weeks. Subsequently the seeds were transferred to 5 °C for 8 weeks and then returned for 17 weeks to the same temperature regimes they were initially allocated to (25/15 °C or 30/20 °C). Germination was scored weekly as the seeds moved through the temperature regimes (beginning week 1 until week 34), to develop germination curves for each population and warming treatment. At the end of the experiment a cut-test was performed to determine whether ungerminated seeds were empty, dormant, or dead. Each chamber consisted of five blocks (shelves). For a given population, one maternal plant was represented in each block.

Statistical analyses

Mixed models were selected for the analysis of plant traits. Models included terms for germination strategy and soil temperature and the interaction thereof as fixed factors, and populations, nested within blocks, were assigned as random factors. Some exceptions were made in the random model where either block or population was used as the random factor because inclusion of population nested in block resulted in convergence failure (see Table S2). Vegetative and reproductive traits that were discreet (number of leaves, number of inflorescences, number of inflorescences, day to flower, day to seed dispersal, seeding duration, and day to senesce) were analysed using Generalized Linear Mixed Models (GLMM), setting the distribution family as Poisson and the link function as natural logarithm (Bolker et al., 2009). Leaf number (but not length of the longest leaf) at the start of the experiment significantly varied across germination strategies (Table S2). Hence, we used leaf number at the start of the experiment as a covariate for corresponding traits, i.e. leaf number, leaf increment rate, SLA, plant area, aboveground biomass. For proportion data (survival and proportion of plants producing seed), GLMM were used with the distribution family as binomial, the link function as logit, and the dispersion parameter set to be estimate. Responses that were continuous (longest leaf, leaf increment rate, individual seed mass, SLA, and aboveground biomass) were analysed with Linear Mixed Models (Restricted Maximum Likelihood, REML. Leaf increment rates, plant area, seed mass, and aboveground biomass were log-transformed prior to fitting to REML (Table S2).
Repeated measures analyses were run for the leaf number and leaf length using germination strategy, soil temperature and measurement time as fixed factors. Population nested in block was used as the random model and leaf number at the start of the experiment as covariates for the analysis of the leaf number. However, the results were the same as when we performed the analyses for each measurement time point and thus, we present the results from the two measurement points as they provide clearer visual inference.
To assess transgenerational effects on germination strategy, we analysed the final germination, non-dormant seed fraction (germination before spring), and time to reach 50% germination of the F1 seeds. The final germination and non-dormant seed fraction were analysed using GLMM, assigning germination temperature, soil temperature, and germination strategy as the fixed factors, population and individual plants nested on the incubator shelf were set as the random factor, and we set the distribution family as binomial and the link function as logit. Time to reach 50% germination was derived by examination of cumulative germination for each dish to the closest 0.25 week and treated as a continuous variable and analysed using Linear Mixed Models (REML) with fixed and random factors as for the non-dormant fraction analyses. GLMM and REML were performed in Genstat 19th Edition.