Gene dispersal in Heliconia 1 2 Côrtes et al. Supporting information: Data S1. Study site 3 4 Figure S1- Map of the study area with location of continuous forest sites (CF1 and CF2) and 5 three 1-ha fragments (F1, F2 and F3). Areas in white are forested areas and in gray are areas 6 that were clear-cut (today these areas encompass secondary growth vegetation and abandoned 7 pastures). Black lines are access roads. 8 9 10 Table S1. Pairwise distances in meters among populations and between edges of fragments to nearest old-growth continuous forest. 11 12 13 14 1 Gene dispersal in Heliconia Côrtes et al. CF1 CF2 F1 F2 F3 Edge to nearest forest CF1 - 41,400 24,300 39,400 40,200 - CF2 - - 17,200 2,100 1,300 - F1 - - - 15,100 15,900 100 F2 - - - - 840 100 F3 - - - - - 100 15 16 Data S2. Implementation of the Bayesian model 17 18 19 a) Implementation: In the genetic component of the model, the probability of observing the seedling’s 20 genotype, given its parental genotypes, P(GkO|GiO,Gi’O) is calculated by producing an (na +1) x 21 (na+1) matrix of parentage probabilities for each seedling, with rows representing the identified 22 potential mothers within the plot, plus a hypothetical mother from outside the plot, and 23 columns representing the identified potential fathers within the plot, plus a hypothetical father 24 from outside the plot. The matrix represents the probability that a seedling k is produced by 25 mother i’ and father i, relative to all potential parental combinations, given genotyping errors 26 and Mendelian inheritance. This step is conducted separately and is independent of ecological 27 variables and dispersal parameters. 28 29 Priors were defined based on previous information about pollen and seed vector movements. A prior of up = 2000 (standard deviation 1500) was used for the pollen dispersal 2 Gene dispersal in Heliconia Côrtes et al. 30 kernel, corresponding to a conservative modal pollination distance of 36 m. Although there are 31 no data on hummingbird movement in the area, these pollinators can move across the 32 heterogeneous landscape at the study site and elsewhere, covering distances of 1000 m in less 33 than four hours (Stouffer & Bierregaard 1995; Hadley & Betts 2009). A prior of us = 500 34 (standard deviation 1000) was used for the seed dispersal kernel, representing a conservative 35 modal dispersal distance of 18 m, which is the average seed dispersal distance found for the 36 main frugivores of Heliconia in the study area (Uriarte et al. 2011). Parameters up and us were 37 drawn from a truncated normal distribution, with lower and upper bounds of 10 and 15,000 for 38 up and 10 and 3,000 for us. The assignment of maternity and paternity is defined from drawing 39 the parameters from either the pollination truncated normal distribution (which will define the 40 pollination kernel) or the seed dispersal truncated normal distribution (which will define the 41 seed dispersal kernel). Therefore, the plant is defined as a mother or father based on the 42 distribution the parameter is drawn from. Instead of assigning the nearest parent as the mother 43 (on the assumption that seed dispersal is more limited than pollination), the seed dispersal 44 distribution has a lower mean and narrower standard deviation, which will most likely result in 45 lower distances of seed dispersal compared to pollination, based on prior information we have 46 on the system. This method allows room for uncertainty in the assignment of maternity and 47 paternity, which makes the model more realistic. We included more details in the supporting 48 information. We used different priors to evaluate the effects of changing initial conditions on 49 the recovery of the dispersal kernel parameters (Supporting information S2.b). 50 51 The contributions of adult plants from outside the plot can be incorporated in the model by assigning hypothetical plants at 10 interval distances of 10 m, going outwards from the 3 Gene dispersal in Heliconia Côrtes et al. 52 sampled plot. Pollen and seed production outside the plots was calculated as the mean 53 production within plots, multiplied by the plant density within each of the rectangular zones 54 delimited by these hypothetical plants. For populations in continuous forest sites (CF1 and CF2), 55 density of flowering plants outside plots was set as the density of flowering plants within plots, 56 under the assumption that density, although variable in space, is homogeneous within a 57 continuous forest of similar environmental conditions. By contrast, areas outside fragment 58 plots encompass secondary growth vegetation, where density of Heliconia acuminata is 59 generally lower than in forested areas (Bruna & Ribeiro 2005). We set the density of flowering 60 plants to 0.0008 plants/m2 outside the Dimona plots (F2 and F3) and 0.0012 plants/m2 outside 61 F1, based on the results of Bruna & Ribeiro (2005). Genotypes of hypothetical plants, as well as 62 of plants with missing genotypes at locus l, were set as the product of the allele frequencies of 63 the respective populations (Moran & Clark 2011). Results are sensitive to the density of 64 hypothetical parents outside the sampled plots. It is likely that density of flowering H. 65 acuminata will increase as secondary forest regenerates in the matrix, which may lead to 66 changes in animal space use and movement. Thus we repeated the analysis by setting the 67 density of flowering plants outside plots equal to density of plots inside the fragments 68 (Supporting information S2.c), to determine whether the change would matter for the 69 estimates. As expected, increasing density of flowering plants outside plots consistently 70 decreased estimates of pollen and seed dispersal distances in fragments (Supporting 71 information S2.c). 72 73 Implementation of Gibbs sampler: 4 Gene dispersal in Heliconia Côrtes et al. 74 1. The iterative chain is initialized by generating for each seedling a random 75 pedigree Pk = (mk,fk) following a multinomial distribution of the parentage probability 76 matrix. For each step of the Gibbs sampler, 77 2. 78 The conditional distribution is: Parameters us and up are sampled from a truncated normal distribution, given Pk. 79 80 π( π’π , π’π |π) = 81 82 = ∏π [(∑ ππ′ π π′π π(ππ′π |π’π )ππ πππ π(πππ |π’π ) )] π(π’π |ππ , π π )π(π’π |ππ , π π ) π,π′ ππ′ π π′π π(ππ′π |π’π )ππ πππ π(πππ |π’π ) (Eqn. S2.1) 83 84 Where i and i’ are the inputted father and mother, mp and ms are the prior mean values of the 85 truncated normal distribution, and sp and ss are the a priori standard deviations. If the 86 conditional probability, based on the new values (pnew), is greater than the conditional 87 probability of the current values (pnow), pnew is accepted. If pnew < pnow then the new values are 88 accepted with probability a = pnew / pnow. 89 90 1- Pedigree Pk is sampled, conditioned on the us and up parameters. New mothers and 91 fathers are proposed for each seedling by sampling from the parentage probability 92 matrix. Now, any parental pair for which parentage probability for seedling k is 0 is 93 excluded from the pool of possible combinations. This step avoids sampling parent 94 combinations with incompatible genotypes. The conditional probability of the proposed 95 pedigree, given proposed parameters is: 5 Gene dispersal in Heliconia Côrtes et al. 96 π(ππ = (π, π ′ )|π’π , π’π ) = π(πππ ′ , ππ ′ π |π’π , π’π , ππ )π(πΊππ |πΊππ , πΊππ′ ) 97 =∑ 98 ππ′ π π′π π(ππ′π |π’π )ππ πππ π(πππ |π’π ) π,π′ ππ′ π π′π π(ππ′π |π’π )ππ πππ π(πππ |π’π ) π(πΊππ |πΊππ ,πΊ π′ ) π ∑π,π′ π(πΊππ |πΊππ ,πΊ π′ ) (Eqn. S2.2) π 99 100 The new pedigree is accepted or rejected using the same approach as described in the 101 previous step. 102 4. 103 iterations. For each chain, we ran 100,000 iterations with a discarded burn in of 50,000 104 steps. Steps 2-3 are repeated until the iterative chain reaches the desired number of 105 106 107 b) Effect of varying priors on posteriors Results may be sensitive to the chosen priors, so to evaluate the effect of different initial 108 conditions on the recovery of the parameters, we tested different prior means for the pollen 109 (800, 2000, 8000) and seed dispersal (100, 500, 2000) kernels and standard deviations for 110 pollen (500, 1500, 4000) and seed dispersal (500, 1000, 1500). Because the routine is time 111 expensive, we ran different priors on only one population (F2). 112 In general, mean posteriors were more sensitive to standard deviations than to mean 113 priors, and variation was higher for pollination parameters, probably because the range of 114 values were wider than for seed dispersal. Mean posteriors of pollen dispersal varied from 115 2,400 to 14,000, while seed dispersal varied from 1,800 to 2,800. Lower mean posteriors for 116 seed dispersal (almost half of the current value) were found at low mean priors (100 and 500) 117 and standard deviation (500). Under all other parametric choices, mean posteriors did not 6 Gene dispersal in Heliconia Côrtes et al. 118 change substantially (Table S2.1). Mean posteriors of pollen dispersal also increased for low 119 standard deviation values, increasing 2.5 times at standard deviations of 500 and only 1.4 times 120 at standard deviation of 4000 (Table S2.1). Doubling the standard deviation increased the mean 121 posterior of seed dispersal parameter by less than 50% at low priors, but no differences were 122 found among the posteriors by increasing standard deviations of high mean priors (Table S2.1). 123 For pollen dispersal parameters, tripling the standard deviations doubled the mean posterior on 124 average, with decreased difference as the mean prior increased (Table S2.1). 125 The results were not highly sensitive to changes in prior means, but were more sensitive 126 to standard deviations. Influence of wider standard deviations to parameters and pedigree 127 recovery have been associated with low numbers of zero genetic mismatches inside the 128 sampled plot, an effect that is pronounced with high rates of genotyping error and small 129 numbers of parents within plots (Moran & Clark 2011). Also, higher uncertainty around the 130 parameter values were found for populations with lower sample size (e.g. CF2), which was 131 further aggravated by the high immigration rate and a concomitantly reduced number of 132 parent pairs within sampled plot (Chybicki & Burczyk 2010). 133 134 Table S2.1 - Summary of posterior means (and support intervals) obtained for pollen and seed 135 dispersal, based on different priors (mean and standard deviations) for population F2. 136 Underlined numbers highlighted in bold face indicate the parameters used in the present study. Pollen dispersal parameters Prior mean Seed dispersal parameters Posterior mean and Prior Credible interval mean SD Posterior mean and SD Credible interval 7 Gene dispersal in Heliconia Côrtes et al. 800 500 2400 [1700-3100] 100 500 1800 [1200-2500] 800 1500 4800 [2100-7600] 100 1000 2700 [2300-3000] 800 4000 11000 [8900-14000] 100 1500 2800 [2400-3000] 2000 500 3100 [2400-3900] 500 500 2100 [1500-5800] 2000 1500 6100 [4300-7600] 500 1000 2700 [2300-3000] 2000 4000 10000 [7300-12000] 500 1500 2800 [2500-3000] 8000 500 8200 [7400-9200] 2000 500 2700 [2200-3000] 8000 1500 2000 1000 2700 [2200-3000] 8000 4000 2000 1500 2700 [2200-3000] 14000 [11000-15000] 137 138 139 140 c) Effect of density of hypothetical parents on posteriors Fragments may be less isolated if there are large quantities of resources for the vectors 141 in the surrounding areas. To test the effect of higher density of plants in the matrix, we 142 increased the density of hypothetical parents outside the plots in the gene dispersal model. 143 Setting the flowering density equal to flowering density inside forest fragments reduced 144 distances of pollination and seed dispersal in all three fragments (Table S2.2). The posterior 145 mean of pollen dispersal was about 45% lower in F1 and F2 and 20% lower in F3. The decrease 146 in posterior mean of seed dispersal was more subtle in F1 and F2, decreasing by less than 10% 147 (Table S2.2). The seed dispersal parameter of population F3, however, was about 55% times 148 lower than in the original scenario (Table S2.2). 8 Gene dispersal in Heliconia 149 Côrtes et al. Changing the initial conditions in the gene dispersal model altered the pollen and seed 150 dispersal distances of H. acuminata. Assuming a greater number of flowering plants outside 151 sampled plots in fragmented sites decreased gene dispersal distances in all three fragments. 152 This is expected given the potential role that density of flowering plants has on the behavior 153 and patterns of space use by foraging birds. That is, we created a scenario of low patch isolation 154 so animals have access to a continuum of food resources throughout the environment. Given 155 the availability of resources in the neighborhood, animals do not need to move further 156 distances to find food, and therefore, do not disperse propagules to greater distances. 157 158 Table S2.2- Average distance values based on the posterior means (and 95% support intervals 159 within brackets) of the modal pollen dispersal parameter (up) and seed dispersal parameter (us) 160 for the current scenario, in which density of flowering plants outside plots is lower than that 161 within fragment, relative to a scenario in which the density of flowering plants outside plots is 162 equal to that inside fragments. Values highlighted in bold face indicate non-overlapping support 163 intervals between current and new scenarios. 164 Fragment Modal pollination distance (meters) Modal seed dispersal distance (meters) Current New Current New F1 58 [47-64] 36 [27-44] 43 [39-45] 39 [27-44] F2 66 [55-77] 34 [20-46] 40 [30-43] 39 [32-44] F3 64 [54-71] 50 [41-63] 42 [39-45] 19 [12-28] 165 9 Gene dispersal in Heliconia Côrtes et al. 166 167 Data S3. Probability of parental identity (PPaI) 168 Probability of parental identity was calculated for each Heliconia acuminata population, 169 using the r - estimator, as follows: 170 171 π (π −1) π π π 0 = ∑πΎ π=1 (2π(2π−1)) (Eqn. S3) 172 173 Where, Xk is the number of seedlings that plant k either fathered or mothered, and N is the 174 total number of offspring recruited for which either father or mother are inside the sampled 175 plot. The two sets of chromosomes, one coming from each parent, are multiplied by the 176 number of offspring. The total number of chromosome sets is used, instead of the number of 177 offspring, because we are interested in the contribution of both father and mother to the 178 seedling genotype, meaning that one seedling can have up to two inside plants contributing to 179 its whole genotype. 180 181 Data S4. Flowering and seedling phenology 182 The proportion of flowering plants was very low in all populations but was higher for 183 CF1 and F3, with around 15% of all recruited plants flowering at least once during the 11 year- 184 study period. In contrast, only about 7% of the plants flowered in CF2, F1 and F2, with some 185 years lacking any flowering events (Table 1, Fig. S4). Higher flowering probability indicates that 186 more plants have the potential to contribute genes to next generations within the plots, 10 Gene dispersal in Heliconia Côrtes et al. 187 whereas populations with lower flowering probability depend on immigration of seeds from 188 other locations to secure plant recruitment. 189 190 191 Figure S4- Heliconia acuminata phenology at the study site: A. Number of flowering plants, and 192 B. Number of seedlings recruited along the 11-yr study period. B. depicts the total number of 193 seedlings ever recruited, including the ones that were not included in the genetic analysis. 194 195 Data S5. Inbreeding coefficient and fine-scale spatial genetic structure 11 Gene dispersal in Heliconia Côrtes et al. - - - - - Inbreeding - Reproductive plants - 0.00 0.05 0.10 0.15 0.20 Inbreeding 0.00 0.05 0.10 0.15 0.20 Seedlings - - - F1 F2 F3 CF1 - - 0.04 0.02 - sp - -0.02 - -0.04 - - - -0.02 - - CF1 CF2 F1 F2 F3 CF2 - F1 F2 F3 - - - - - 0.00 0.02 - -0.04 sp - -0.10 -0.10 CF2 0.04 CF1 0.00 - - - 196 - - - CF1 CF2 F1 F2 F3 197 Fig. S5- Actual inbreeding coefficients and sp-values and 95% confidence intervals computed 198 from 2,000 bootstrapped samples of size equal the least dense population (CF2) of seedlings (N 199 = 52) and reproductive plants (N = 20) across the five populations of Heliconia acuminata at the 200 BDFFP. 201 202 203 204 205 206 12 Gene dispersal in Heliconia Côrtes et al. 207 References 208 Bruna EM, Ribeiro MBN (2005) Regeneration and population structure of Heliconia acuminata 209 in Amazonian secondary forests with contrasting land-use histories. Journal of Tropical 210 Ecology 21, 127-131. 211 Chybicki IJ, Burczyk J (2010) Realized gene flow within mixed stands of Quercus robur L. and Q. 212 petraea (Matt.) L. revealed at the stage of naturally established seedling. Molecular 213 Ecology 19, 2137-2151. 214 215 Hadley AS, Betts MG (2009) Tropical deforestation alters hummingbird movement patterns. Biology Letters 5, 207-210. 216 Moran EV, Clark JS (2011) Estimating seed and pollen movement in a monoecious plant: a 217 hierarchical Bayesian approach integrating genetic and ecological data. Molecular 218 Ecology 20, 1248-1262. 219 220 221 Stouffer PC, Bierregaard RO (1995) Effects of forest fragmentation on understory hummingbirds in Amazonian Brazil Conservation Biology 9, 1085-1094. Uriarte M, Anciães M, da Silva MTB, et al. (2011) Disentangling the drivers of reduced long- 222 distance seed dispersal by birds in an experimentally fragmented landscape. Ecology 92, 223 924-937. 224 225 13