Bioinformatics and analysis
Demultiplexing resulted in 1,536 individual fastq files (192 samples across four cycle treatments with forward and reverse reads). We used zgrep in bash to identify sequence motifs unique to each of the four species, combining forward and reverse read counts (Supplemental information).
We generated a standard curve by combining results from across the two sets (Figure S2). As discussed, the critical number to determine relative abundance in qPCR is the Ct value. Because qAMPseq data do not directly yield amplicon counts at the end of every cycle, we did not have direct knowledge of the exact shape of the PCR curve – the important step in determining relative abundance. To determine the cycle when samples crossed a Ct value required using a different approach: first, we log-transformed read counts associated with each of the four-cycle points for which we quantified amplicons. Log-transformation of a PCR sinusoidal curve theoretically results in a linear relationship of cycle and amplicon number. We took advantage of this by determining the slope of the amplification line, i.e.\(\frac{\ log(amplicon\ count)}{\ \text{cycle\ number}}\). We set our Ct number as log(10,000 reads), and used it in a simple equation to determine the cycle that corresponded to Ct for each amplification curve of each species in each sample. We henceforth call this number the cycle count. If after 35 cycles a species in a sample had fewer than 10,000 reads, it was assigned a cycle count value of zero. We calculated cycle numbers and performed all of the following analyses in R version 3.5.2 (R Core Team, 2018).
Ecologically speaking, our goals were to quantify the amount of eachClarkia species’ pollen in each pollen ball to determine if bees (1) were inconstant pollen foragers (2) exhibited preference for certain species of Clarkia and (3) used Clarkia species pollen in ways that were not apparent from observations of floral visitation. We also wanted to compare results yielded by the new approach, qAMPseq, to the results from relative read abundance from the plateau phase of PCR.
To determine if bees were inconstant while foraging for pollen, we asked which Clarkia species were present or absent in pollen balls using both the qAMPseq method as well as the RRA method. Inconstant pollinators will have more than one species in their pollen balls. In qAMPseq, we determined presence/absence of Clarkia in our samples by asking simply if the cycle values were nonzero (present) or zero (absent) for each species of Clarkia . In contrast, RRA does not yield a cycle count but instead yields relative reads after full amplification (35 cycles). To determine presence/absence ofClarkia with RRA, we used three different sample proportion cutoffs to determine the presence of a species in samples: 0%, where presence was defined as any nonzero read count; and 5% and 10 % cutoffs, where presence was defined as anything above 5% or 10%, respectively. We counted the proportion of pollen balls that had more than one species of Clarkia using our four different metrics (raw relative read abundance (RRA), and relative read abundance with 5% and 10% sample proportion cutoffs (RRA5 and RRA10)).
To further understand constancy, we compared the number of species in bee pollen balls to the number of flowering species where they were captured. When we sampled bees in Clarkia communities, the communities contained one to four species of flowering Clarkia . If bees are completely inconstant, then we expect their pollen balls to contain the same number of species as the communities they were captured in. To test this, we tallied the number of bees caught in communities with one, two, three, and four species of flowering, as well as the number of Clarkia in each of their pollen balls. We ran a Pearson’s Chi-squared test to determine if the proportion of samples containing one to four species of Clarkia pollen matched the proportion of bees caught in communities with one to four species of flowering Clarkia. If bees are inconstant when pollen foraging, these proportions would be the same, and the test would return a non-significant result.
Preference for different Clarkia species was estimated as the difference between the relative amount of a species’ pollen in a sample and the relative amount of that species’ floral abundance in a surveyedClarkia community where the bee was captured (as in James 2020). This measure of preference can only be calculated for communities with more than one Clarkia species, because there is not an available ‘choice’ to make between plants in single-species communities; as such, we only calculate preference using the pollinator visits/pollen balls from communities with more than one Clarkia species. The calculation yields a value between -1 and 1 for each Clarkiaspecies in each pollen ball. Negative values indicate avoidance, positive values indicate preference, and values of zero indicate that bees do not preferentially forage for any species.
We calculated preference using values generated by both RRA with a 0% sample proportion cutoff and qAMPseq. To calculate preference using qAMPseq values, we used the cycle count for a given species divided by the sum of cycle counts in a sample to estimate relative abundance of each species in each sample. We used a paired t-test to determine if there was a significant difference in estimates of preference using qAMPseq versus RRA. We then ran an ANOVA, using Tukey’s honest significant difference test to determine if pollinator preference forClarkia species were significantly different, and t-tests to determine if pollinator preferences were significantly different from zero.
For a complete picture of pollen use by pollinators, we also incorporated Clarkia abundance in communities and in pollen balls into our analyses. First, we compared the average flowering abundance ofClarkia species to compare the amount of floral resources the different Clarkia provided to pollinators when flowering. To do so, we log-transformed all non-zero values of flowering abundance and ran an ANOVA with Tukey’s honest significant difference test toClarkia species. There was one ‘zero’ value that we retained in the analysis: the four-Clarkia community Kingsnake only had three flowering Clarkia species during one sampling period, and as such, the species with no open flowers (C. xantiana ) was assigned a zero. The second way we incorporated abundance into our analyses was to use the data we generated with the qAMPseq method. For each bee taxon X Clarkia species combination, we summed the number of bees carrying pollen from that Clarkia species, weighted by the proportion the Clarkia species was represented in the pollen ball. This weighted value tells us not only the presence/absence ofClarkia pollen on the bee, but the extent to which the bee species used that particular pollen resource.
Finally, we compared pollinator visitation to Clarkia andClarkia pollen use by constructing two networks of plants and pollinators: one network with observations of the Clarkia species bees were caught on, and the other with the Clarkia pollen that we identified in bees’ pollen balls. In the case of the visitation network, the dataset consists of the number of times each pollinator was caught visiting each of the Clarkia species. The pollen-use dataset consists of proportions of Clarkia pollens in each sample, rather than a single plant-pollinator connection or the presence/absence of Clarkia species in a pollen ball. To address this and build the dataset for the pollen network, we multiplied the proportion of each Clarkia species in each pollen sample by 100, and rounded to the nearest whole number.
We measure and compare networks’ network-level specialization, H2’, to understand if pollinators use pollen in ways similar to their floral visitation. We use H2’ because it is robust to differences in the number of interactions (Blüthgen, Menzel & Blüthgen 2006). Values of network specialization, H2’, are between 0 and 1, where higher H2’ values indicate that a network is comprised of more specialized relationships between plants and pollinators, and lower values indicate the network has more generalized relationships. Network specialization should be the same between the two networks if pollinators carry Clarkia pollen at the same rates that they visit Clarkia . Bipartite networks were built and H2’ was calculated using the package bipartite (Dorrman et al. 2020).