APPENDIX 1: Data Collection, Identification and Quantification Methods (From Eklöf Et

APPENDIX 1: Data collection, identification and quantification methods (from Eklöf et al., 2013 SI, p31).

Pollen grains were collected from the undersides of the trapped pollinator species, and observed under 100x power.The number of pollen grains on seven equally spaced straight-line transects across the square coverslip (22x22mm) was recorded. Whenpollen grains were encountered, we used high power magnification (400X) to ensure appropriate identification. We summed the pollen grain counts from all seven transects as an estimate of pollen carried per individual(Kearns & Inouye 1993). This technique measures pollen on the underside of insects and does not include pollen contained within corbiculae which may contribute to total pollen load but are unlikely to be involved in pollination events (Thorp, 2000).

Table S1: Pollinator species list

Order / Family / Species
Hymenoptera / Apidae / Apis mellifera (Linnaeus, 1758)
Bombus terrestris (Linnaeus, 1758)
Netelia producta (Brullé, 1846)
Halictidae / Lasioglossum sordidum (Smith, 1853)
Ichneumonidae / Ichneumon promissorius(Erichson, 1842)
Diptera / Anthomyiidae / Delia platura (Meigen, 1826)
Calliphoridae / Calliphora stygia (Fabricius, 1794)
Calliphora vicina (Robineau-Desvoidy, 1830)
Pollenia pseudorudis (Rognes, 1985)
Stratiomyidae / Odontomyia atrovirens (Bigot, 1879)
Syrphidae / Eristalis tenax (Linnaeus, 1758)
Melangyna novae-zelandiae (Macquart, 1855)
Helophilus hochstetteri (Nowicki, 1875)
Sarcophagidae / Oxysarcodexia varia (Walker, 1836)
Tachinidae / Pales marginata (Hutton, 1901)
Staphylinidae / Thyreocephalus orthodoxus(Olliff, 1887)

Table S2: Plant species list

Family / Species
Apicaeae / Daucus carotaL.
Asparagaceae / Cordyline sp.
Asteraceae / Cirsium vulgare(Savi) Ten.
Taraxacum officinaleWeber
Betulaceae / Alnus glutinosa Gaertn.
Alnus serrulata(Aiton) Willd.
Brassicaceae / Raphanus sp.
Chenopodiaceae / Chenopodium albumL.
Fabaceae / Trifolium sp.
Malvaceae / Acea sp.
Sp. (other)
Myrtaceae / Eucalyptus globulus
Eucalyptus nitens H.Deane & Maiden
Poaceae / Lolium sp.
Rosaceae / Rubus fruticosus

Table S3: Insect functional traits used in this study and data sources. Data were collected in Field Based Observation Studies (FBOS) conducted by Rader et al., 2014, or from other Literature (LIT): Colles & McAlpine 1991; Howlett et al. 2005; Donovan 2007; Michener 2007; Rader et al. 2009.

# represents number of individuals sampled.

Pollinator traits
(and abbreviations) / Mean / Range / Unit / Data source / Reference / #
Average body length (“b.size”) / 10.5 / 15.1 / mm / FBOS / This study / 10
Average body width between the base of the wings (“b.size”) / 3.3 / 5.6 / mm / FBOS / This study / 10
Pollen carrying structure
(“carrying_structure”) / - / - / - / LIT / 3-5 / -
Mean duration of flower visitation (“vis_time”) / 8.8 / 14.1 / seconds / FBOS / 1 / 30
Foraging preference for nectar or pollen (“nec.poll”) / - / - / % / FBOS / 1 / 20
Social or solitary (“soc_sol”) / - / - / - / LIT / 3,4 / -
Feeding preference of larvae (“larval_diet_pref”) / - / - / - / LIT / 3,4 / -
Hour of peak daily activity based on abundances (“daily”) / 257.3 / 5683 / Insects / FBOS / This study / -
Month of peak seasonal activity based on abundances (“season”) / 89.1 / 3273 / Insects / FBOS / This study

Table S4: Plant functional traits used in this study and category description. All traits were measured for the study of Rader et al., 2014 and provided by R. Rader.

Plant trait / Description
growth_form /

1. grass
2. herb
3. shrub
4. tree

longitivity /

1. annual
2. biennial
3. perennial

type_infl (type of inflorescence) /

1. spike, raceme, panicle/cyme/thyrse, umbel/corymb
2. captitulum/head
3. catkin

pollnec_acces (ease of acces to pollen/nectar) /

1. open or easy access to nectar and pollen
2. tubular or hidden resources or closed/restrictive access

single_infl (clustering of inflorescence) /

1. solitary flower or small clusters of single flowers
2. inflorescence

flo_per_infl (number of flowers per inflorescence) /

1. tens of flowers
2. hundreds
3. thousands

fragrance /

1. none
2. some
3. yes

nec_per_pol_per_day (amount of nectar per flower unit per day) /

1. < 1 μ
2. 1-5 μl
3. > 6 μl

flo_symmetry (symmetry of flower) /

1. actinomorphic
2. zygomorphic

infl_symmetry (symmetry of whole inflorescence) /

1. actinomorphic
2. zygomorphic

season / Presence/absence of flowers across the 4 seasons
flo_sex /

1. monoecious
2. dioecious

Figure S1: Accumulation curve of interaction pairs. We calculated the accumulation of interaction pairs (instead of individual species used in species accumulation curves) by pooling all interaction pairs from the different sites in the rarefaction calculation. ‘Interaction richness’ on the y-axis thus corresponds to the number of unique pairs of plant-pollinator partners, and the x axis is the number of observations of a pair during the sampling process.

As the accumulation curve does not reach saturation, we conclude that the interactions that were sampled were incomplete. This is a common issue, well known for pollination networks, and can be due to insufficient sampling effort; however the high number of rare species, common in pollination networks, also increases the expected richness and hence the recorded interactions appear undersampled (Chacoff et al. 2012). Furthermore, this issue is corrected for by including network size throughout our analyses.Importantly, sampling effort is consistent across replicate networks to allow comparison. Furthermore, our main focus is on comparisons of species roles within, not across networks, such that equivalence of sampling effort does not affect our main results.
APPENDIX 2: Additional analyses

1. Analysis for the pollinator community without excluding the outlier.

The abundance of the pollinator Lasioglossum sordidumin the first sitegave this interaction a very high leverage (calculated as in Crawley 2007, p.354), such that we excluded it from the main analysis. To do so, we set its abundance to 0. However, that species had recorded interactions with several plant species of that community, which we left unchanged. Therefore, normalized degree (ND) and specificity (HS) were calculated including the interactions of L.sordidum. We simply removed the measures of ND and HS for that species in the analysis, such that the degree and specificity of the other species of the community remained unchanged (see Appendix 2 code).

Figure S2: Leverage of the pollinator species based on their abundances. The dotted line corresponds to an approximate threshold above which a data point can be considered as a statistical outlier (Crawley 2007, p.354).

In our main analysis, we removed this outlier, which corresponds to the abundance of Lasioglassum sordidum in the first site. We nevertheless ran the models including all data points, and present the results in Table S5. The leverage of the outlier influenced the results of the abundance covariate, but did not qualitatively change those pertaining to originality.

Table S5: Results for the models containing all data points.In the case of weighted pollinator originality predicting HS (b), no model averaging was needed, such that the degrees of freedom (DF) and t-value are presented instead of the Adjusted Standard Error and z-value presented otherwise.

a: Weighted pollinator originality predicting ND
(conditional average) / Estimate / Std. Error / Adjusted SE / z value / Pr(>|z|)
(Intercept) / 0.760 / 0.099 / 0.100 / 7.581 / < 0.0001
pol.abun1 / 0.001 / 0.001 / 0.001 / 2.655 / 0.008 / **
size / -0.006 / 0.004 / 0.004 / 1.626 / 0.104
w. pol. orig2 / -0.381 / 0.141 / 0.143 / 2.670 / 0.008 / **
abun :w.pol.orig / 0.760 / 0.099 / 0.100 / 7.581 / < 0.0001
size :w.pol.orig / <0.001 / 0.001 / 0.001 / 2.655 / 0.008 / **
b: Weighted pollinator originality predicting HS
Value / Std.Error / DF / t-value / p-value
(Intercept) / 0.408 / 0.102 / 66.000 / 4.018 / 0.000 / ***
pol.abun / 0.000 / 0.000 / 66.000 / 2.380 / 0.020 / *
w.orig / 0.314 / 0.128 / 66.000 / 2.445 / 0.017 / *
pol.abun :w.orig / -0.002 / 0.001 / 66.000 / -3.082 / 0.003 / **
c: Pollinator uniqueness predicting ND
(conditional average) / Estimate / Std. Error / Adjusted SE / z value / Pr(>|z|)
(Intercept) / -0.056 / 0.162 / 0.165 / 0.337 / 0.736
uniq3 / 0.606 / 0.175 / 0.178 / 3.403 / 0.001 / ***
size / -0.001 / 0.001 / 0.001 / 1.187 / 0.235
pol.abun / 0.001 / 0.001 / 0.001 / 0.497 / 0.619
d: Pollinator uniqueness predicting HS
(conditional average) / Estimate / Std. Error / Adjusted SE / z value / Pr(>|z|)
(Intercept) / 1.152 / 0.235 / 0.239 / 4.825 / 0.000
uniq / -0.647 / 0.244 / 0.248 / 2.612 / 0.009 / **
size / -0.002 / 0.001 / 0.001 / 1.278 / 0.201

1: pollinator abundance; 2: weighted pollinator originality; 3: pollinator uniqueness

2. Results for the models containing unweighted versions of originality

Table S1: Linear Mixed-effects Model partial coefficient estimates from the averaged best-fitting models predicting pollinator normalised degree (ND) or specificity (HS) and containing pollinator originality not weighted by pollinator relative abundances (i.e., the distances to the community centroid are based on pollinator presence-absence data). Non-significant results are shown only if they were retained in the model; significant results are shown in bold. ‘pol.abund’ stands for pollinator abundance, ‘pol.orig’ for pollinator originality. (:) represents an interaction effect.

a: Pollinator unweighted originality predicting ND
(conditional average) / Estimate / Std. Error / Adjusted SE / z value / Pr(>|z|)
(Intercept)
pol.abun
pol.orig
size
pol.abun:pol.orig
pol.orig:size / 0.3517531
-0.0009139
0.3369720
-0.0056374
0.0035279
0.0086938 / 0.2168130
0.0015150
0.3761163
0.0036848
0.0020789
0.0064090 / 0.2197549
0.0015283
0.3812035
0.0038143
0.0021175
0.0065265 / 1.601
0.598
0.884
1.478
1.666
1.332 / 0.1095
0.5499
0.3767
0.1394
0.0957
0.1828
b: Pollinator unweighted originality predicting HS
(conditional average) / Estimate / Std. Error / Adjusted SE / z value / Pr(>|z|)
(Intercept)
pol.abun
pol.orig
pol.abun:pol.orig
size / 1.0349833
-0.0001914
-0.5225217
-0.0016586
0.0005421 / 0.1503902
0.0008402
0.2583113
0.0021309
0.0009303 / 0.1530945
0.0008514
0.2629079
0.0021713
0.0009934 / 6.760
0.225
1.987
0.764
0.546 / <2e-16 ***
0.8221
0.0469 *
0.4449
0.5853

These results show that there were no significant relationships between pollinator normalized degree and the unweighted calculation of originality; however, unweighted originality was negatively correlated with pollinator specificity (HS). This indicates that when using only the presence/absence of pollinators, species with a distinct trait profile from the community average were less specific in their interactions among their partners (i.e., they pollinated plants more evenly), which is the opposite finding from the equivalent model using weighted originality (Table 1b).

The choice between weighted and unweighted versions of functional originality is therefore of major importance, and should be chosen carefully depending on whether a trait-centered (reflected by weighing by relative abundances) or a species-centered approach is preferred, as they assess different ecological questions. In the present study, this seems to show that on the surface, pollinators do not preferentially interact with plants based on their trait profiles. However, in comparison with the analysis comprising weighted originality, this suggests that certain pollinator traits and their underlying functions are nonetheless more original (and others more common), which influences the interaction patterns relative to trait occurrence rather than species identity. This also illustrates how the pollinator community is functionally more homogenous when considering traits qualitatively, and this masks the underlying processes defining how species interact.

APPENDIX 3: Determining which traits were most important: jack-knife analysis for individual traits

The traits measured can have an important influence on the outcome of the study, as they define the nature of the functional diversity of the community. We chose our traits to measure general features associated with pollination fulfilment. However, in a context where one would be interested in the relative importance of traits related to several different functions (for example, if traits pertaining to plant growth on one hand versus investment in defenses against herbivores on the other had been measured) in determining the strength of the relationship between species’ functional and network roles, it could be measured through a jack-knife approach (Gagic et al. 2015), i.e. the successive removal of one trait at a time from the raw trait data.

Negativedifferences in model fit (AICs) between the original model and the ones calculated after each trait was removed revealed that the corresponding traits improved the model fit when included in the calculation of pollinator originalityor uniqueness, and thus contributed to itsrelationship with network role. Conversely, positive AICs indicated that those traits worsened the model fit, suggesting that they went against the relationships between functional and network roles detected by the models.

Results showed that the jack-knife removal of successive traits produced better or worse model fits depending on the identity of the trait removed (Fig. S1a). Despite some variation in the contributions of each trait, depending on which model was jack-knifed, the removal of body size and daily peak abundance improved the predictive power of the models pertaining to pollinator’s normalised degree and specificity predicted by their originality or uniqueness by reducing their AIC scores. The removal of social vs. solitairy nesting behavior, pollen carrying structure, larval diet preference and peak seasonal abundance were more contrasted in their contributions to the model fits. Conversely, traits pertaining to adult diet preferences (nectar or pollen) and flower visiting time reduced the fit of all models, suggesting that these traits either did not affect the position of the pollinators in the functional niche space in relation to their network roles, or even added noise that masked any relationship.

This jack-knife analysis allowed us to directly link specific traits to their impact on network structure: in addition to determining each species’ contribution to functional diversity, we were able to identify which specific traits were responsible for the relative positioning of species in the functional niche space, and predict these positions using metrics that define the arrangement of species interactions within a network. This can inform the selection of traits to be used when predicting that two given species will interact according to trait matching (Dehling et al. 2014; Dray et al. 2014). However, it should be noted that individual traits are not necessarily independent from each other, such that the presence of one trait could be associated with that of another (Fenster et al. 2004). When quantifiable, assigning weights to these trait complexes (e.g. as we did for the traits relating to body size) when estimating functional diversity is an important step. Using a jack-knife approach may nonetheless not reflect the importance of traits in determining interaction partners when they are involved in such complexes.

Figure S3: Results of the jack-knife removal of species traits from the calculation of pollinator functional originality and uniqueness on their predictive power of normalised degree (ND) and specificity (HS). Negative values show a reduced fit of the model when the corresponding trait is removed, indicating that the corresponding trait is important for relating functional and network roles; positive values suggest the corresponding traits are degrading the model fit. b.size = body size, daily = hour of peak daily activity based on abundances, soc.sol = social vs. solitary, carrying.structure = the presence of a pollen carrying structure, such as corbiculae, larv.diet = larval diet, season = month of peak seasonal activity based on abundances, nec.poll = foraging preference for nectar or pollen, vis.time = mean duration of flower visitation.

APPENDIX 4: R code used for analyses.

library(FD)

library(ESM)

library(bipartite)

library(magrittr)

#################################################################################

# code by Camille Coux

# last modified 20 Jan 2016

# This is code for the analysis ran to obtain results in our manuscript. We

# provide it as a courtesy and in an effort of transparency and reproducibility.

################################################################################

# 2 functions to calculate species' functional originality and uniqueness:

# - species_coords calls the dbFD function from FD package to calculate species'

# coordinates in the trait space, and rearranges the outputs

# - FD_measures calculates originality and uniqueness for species present in

# a given community.

#

# Inputs are the same as those needed for dbFD; row/column name-matching applies.

# Output: a dataframe with species as rows and FD measures as columns, in addition

# to abundances and farm (site name).

library(FD)

# The coordinates are calculated for the whole community, it's the centroid

# that's going to change coordinates. And then I'll just need to adapt the nearest

# neighbour part such that only the distances to *present* neighbours are calculated.

species_coords <- function(traitmat, abund, weights, ...){

# calculating species' coordinates and the centroid

coords <- dbFD(traitmat, abund, w=weights, corr="cailliez", print.pco=T)$x.axes

# now first calculating the centroid of the pollis present for each site

centr <- list(NULL)

# centr = averaged trait values for pollis present. 14 dimensions for 21 sites.

for(i in 1:length(abund[,1])){

pres <- which(abund[i,]>0)

vec <- coords[as.logical(abund[i,]),]

w <- abund[i,pres]

centr[[i]] <- apply(vec, 2, weighted.mean, w = w)

}

centr <- do.call(rbind, centr)

rownames(centr) <- rownames(abund)

return(list(coords=coords, centr=centr))

}

# now to calculate originality and uniqueness

FD_measures <- function(coords, centr, abund){

# adds a line for the centroid coodinates in each community

coords2 <- list(NULL)

for (i in 1:length(abund[,1])){

coords2[[i]] <- data.frame(rbind(coords, centr[i,]))

rownames(coords2[[i]])[dim(coords)[1]+1] <- "centr"

}

dists_centr <- lapply(coords2, function(x){

d.poll<- as.matrix(dist(x, diag=TRUE, upper=TRUE))

for (i in 1:dim(d.poll)[1]) {d.poll[i,i] <- NA}

return(d.poll)

})

# Originality: Distance to centroid of the species present

originality<- unlist(lapply(1:length(abund[,1]), function(i){

dists <- dists_centr[[i]]

pres <- which(abund[i,]>0)

dists[pres,dim(dists)[1]]

}))

# Uniqueness: nearest neighbour among present species

uniqueness <- unlist(lapply(1:length(abund[,1]), function(i){

pres <- which(abund[i,]>0)

dists <- dists_centr[[i]][pres,pres]

uniq <- NULL

for (j in 1:length(pres)){uniq <- c(uniq, min(dists[j,], na.rm=T))}

return (uniq)

}))

# also need to keep track of farm, species names and their abundances

measures <- do.call(rbind, lapply(1:length(abund[,1]), function(i){

pres <- which(abund[i,]>0)

abundance <- abund[i, pres]

sp <- names(abundance)

site <- rownames(abund)[i]

farm <- rep(site, length(pres))

tab <- data.frame(farm=farm, sp=sp, abundance=as.numeric(abundance))

return (tab)

}))

# merge all

measures$orig <- originality

measures$uniq <- uniqueness

return(as.data.frame(measures))

}

#################################################################################

# FD and bipartite metric calculation with ntw weighed by the number of insect

# individuals from a same species carrying pollen of 1 plant.

#################################################################################

library(ESM)

library(bipartite)

library(magrittr)

### 1. Import data:

### ###############

## pollinator traits

t3 <- read.table("pollinator_traits.csv", sep=",", header=T, row.names = 1 )

# assign pollinator trait weights

weight <- c(0.5, 0.5,1,0.5,0.5,1, rep(1/6, 6), rep(1,3))

## plant traits

pl_t <- read.table("plant_traits.csv", sep=",", header=T, row.names = 1 )

# assign plant trait weights

pl.weight <- c(rep(1, 8), rep(0.25, 4), rep(1, 3))

## import abudances

a2 <- read.table("pollinator_abundances.csv", sep=",", header=T, row.names = 1 )

a <- as.matrix(a2)

a[which(a>0)] <- 1

p <- read.table("plant_abundances_bin.csv", sep=",", header=T, row.names = 1 )

# outlier check (methods from the R Book, Crawley M.J., 2007, p363)

leverage<-function(x){1/length(x)+(x-mean(x))^2/sum((x-mean(x))^2)}

x1 <- a2[a2>0]

# pdf("C:/Users/camille/Desktop/PhD/Australian data/mes trucs/FD/fdisp/proper intnat/scaled_strength/revisions/figs/final_figs/S2.pdf")

plot(leverage(x1),type="h", ylab="Pollinator abundances")

points(leverage(x1))

# A general rule is that a point is highly influential if its leverage is

# greater than 2P/N, where p is the number of parameters in the model and

# N is the number of data points.

P <- 5; N <- length(x1)

abline(((2*P)/N),0,lty=2)

# dev.off()

# point 47 is clearly an outlier; it corresponds to the L_sordidum abundance in

# the first site. We replace its abundance by 0 in the abundance matrix, such that

# the FD space is calculated without it. The analysis is repeated without changing

# the

# abundance of this outlier and presented in Appendix 3.

a2[1,"L_sordidum"] <- 0

## arranging interactions into a list adjacency matrices (one for each site)

interactions <- read.table("interactions.csv", sep=",", header=T, row.names = 1 )

interactions %>%

split(., .$Site) %>%

lapply(., function(x){