Material and Methods
Study species
We used a total of eight species (see Table S 1) with different
distributions and IUCN red list status within Switzerland. All species
are co-occurring in the “arable vegetation of calcareous soils”
habitat (“Caucalidion” according to the classification by Delarze,
Gonseth, Eggenberg, & Vust, 2015), insect-pollinated and have
overlapping flowering times in nature (Landolt et al., 2010; Lauber,
Wagner, & Gygax, 2018). We classified species as “common” if their
IUCN status in Switzerland was “least concern”, and as rare otherwise.
IUCN status in Switzerland correlated strongly with number of
observations within Switzerland pooled between the years 2000 and 2020
(cor: 0.86, p-value <0.001). Breeding system (self-compatible vs
self-incompatible) was extracted from the BioFlor database (Kuehn,
Durka, and Klotz 2004). To test the relationship between recipient-donor
relatedness and heterospecific pollen interference, we constructed a
phylogenetic tree for our species (see Figure S 1) by pruning a modified
version (Malecore et al. 2018) of the dated DaPhnE supertree of Central
European plant species (Durka and Michalski 2012) and then calculated
the phylogenetic distance using the cophenetic function of the “ape”
package (Paradis, Claude and Strimmer 2004) in R (R Core Team 2023).
Experimental set-up
We sowed all species into 12 cm x 17 cm trays filled with
“Seedlingsubstrat” potting soil (Klasmann-Deilmann GmbH, 49741 Geeste,
Germany) and put them into the dark cold storage room at -4°C for
stratification between 5 and 8 weeks. Once seeds started to germinate,
we moved the trays to a greenhouse compartment. We transplanted
seedlings into 11 cm x 11 cm x 12 cm pots filled with “Selmaterra”
(fertilized heavy soil with 30% volume peat, see Table S 2). We
randomized pots on tables of a single pollinator free greenhouse
compartment (14-hour light cycle, heating starting at 10°, cooling
starting at 20°C) and watered as well as fertilized regularly. We
treated aphids and fungi whenever necessary.
All species flowered between May 2021 and October 2021. To ensure
continuing of flowering, we regularly dead-headed the plants. To assess
the effect size of heterospecific pollen interference, we performed hand
pollinations between and within all species and measured seed set
(yes/no) and seed number. For the heterospecific pollen treatment, we
prepared a saturated mix of conspecific pollen and heterospecific pollen
and applied it to the stigma of the recipient flower. For each flower
treated with heterospecific pollen mixture, we treated a second flower
on the same individual on the same day with conspecific pollen only as a
control, using the same conspecific pollen donor that we used for the
heterospecific pollen mix. Conspecific pollen and Heterospecific pollen
were from a single individual respectively. The two flowers with
heterospecific and conspecific treatment would constitute a pair with
the same “pair ID” (see Figure 1). Pollen grain number per anther
differed greatly depending on individual and on anther ripeness
(personal observation from preliminary pollen counts with a Neubauer
counting chamber) thus we standardized treatment by always applying a
pollen amount above saturation level. We extracted the pollen for the
treatments from the anthers by tapping them on a glass slide and with
the help of tweezers, and then mixed it for the heterospecific
treatment. We then applied the pollen mixture (HP) or the conspecific
pollen (CP) to the open stigma of the recipient flower using tweezers.
To avoid selfing, we emasculated recipient flowers by removing the
anthers some days before treatment. For some species (Bupleurum
rotundifolium, Fallopia convolvulus, Myosotis arvensis ), anther removal
would cause too much flower damage due to the small size, thus anthers
were not removed. For these species, selfing could not be completely
excluded.
For each donor-recipient combination, we treated between 2 and 16 flower
pairs (HP and CP; median: 12 flower pairs per donor-recipient
combination; see Table S 3 in supporting information). We collected
seeds after ripening, and counted them either by hand or by using an
imaging method with imageJ (Abramoff, Magalhaes, and Ram 2004) (forPapaver rhoeas , see “Protocol seed counting in ImageJ” in
Supporting Information).
Statistical analysis
Do pollen type and recipient status affect seed set and
seed
number?
To test whether seed set and seed number are affected by pollen type
(conspecific vs heterospecific) and whether the effect size depends on
recipient status (common vs rare), and to account for the high
proportion of zeroes (~24%), we ran a hurdle model using
the function glmmTMB of the homonymous package (Brooks et al.
2017). In a hurdle model, zero counts and non-zero counts are treated as
two separate categories, meaning that a binomial model is fitted for
zeroes vs non-zeroes (the “zero-inflated” model), and a separate model
for the non-zero counts only (“conditional model”). For the
conditional model, we used a truncated negative binomial error
distribution (“truncated nbinom2” in glmmTMB ). We implemented
the same formula for both the zero-inflated and the conditional model,
with pollen type (conspecific=0, heterospecific=1), recipient status
(common=0, rare=1) and their interaction as fixed effects. To account
for non-independence, we included pair ID, treatment date, recipient
species, recipient individual ID and donor individual ID as random
factors.
After running the models, we used the functions emmeans andpairs of the emmeans package (Lenth, 2023) to calculate
the 95% confidence intervals of the estimated marginal mean for each
group (conspecific and heterospecific treatments for common and rare
recipients) and to test for significance of the comparisons of interest
(conspecific vs heterospecific treatment for common recipients,
conspecific vs heterospecific treatment for rare recipients).
Does donor status affect seed set and seed
number?
To explore in more detail the effects of donor and recipient status on
seed set and seed number, we separately analyzed a subset including only
the HP treatment. We ran a hurdle model using the functionglmmTMB of the homonymous package with recipient status
(common=0, rare=1), a dummy factor indicating whether the heterospecific
pollen donor was of the same or the opposite status (same=0,
opposite=1), as well as their interaction, as fixed effects. We included
the same random factors as in the previous model (pair ID, treatment
date, recipient species, recipient individual ID and donor individual
ID).
After running the models, we used the functions emmeans andpairs of the emmeans package to calculate the 95%
confidence intervals of the estimated marginal mean for each group
(heterospecific treatment for common and rare recipients with common and
rare donors) and to test for significance of the comparisons of interest
(heterospecific treatment: common donors vs rare donors on common
recipients, common donors vs rare donors on rare recipients).
Do recipient and donor self-compatibility affect seed set
and seed
number?
To test whether seed set and seed number are affected by the
self-compatibility of recipient and donor species in interaction with
heterospecific pollen deposition, we repeated the same analyses as
above, replacing recipient and donor status with recipient and donor
self-compatibility.
Does recipient-donor relatedness affect seed
number?
To test whether the phylogenetic distance between recipients and donors
affects seed set and seed number, we calculated for all non-zero counts
the log-response ratio of seed-number with HP treatment on seed number
with CP treatment, within each pair (same pair ID).We then fitted a
Gaussian glmmTMB model including recipient-donor phylogenetic
distance, recipient status (common=0, rare=1), a dummy factor indicating
whether the heterospecific pollen donor was of the same or the opposite
status (same=0, common=1), as well as all their interactions, as fixed
effects. To account for non-independence, we included treatment date,
recipient species, donor species, recipient individual ID and donor
individual ID as random factors.
After running the model, we used the functions emmtrends of theemmeans package to calculate the estimated trends with their 95%
confidence intervals for the relationship between log-response ratio and
recipient-donor relatedness for each group (common recipient with common
donor, common recipient with rare donor, rare recipient with rare donor,
rare recipient with common donor).
For all models, we inspected Pearson residuals for homogeneity of
variance against all grouping variables.