Supporting Information (SI) Molecular Ecology
Fine-grained adaptive divergence in an amphibian: genetic
basis of phenotypic divergence and the role of non-random
gene flow in restricting effective migration among wetlands
Alex Richter-Boix; María Quintela; MarcinKierczak; Marc Franch; AnssiLaurila
MATERIAL INCLUDED
Section 1.PREDATOR ABUNDANCE ESTIMATION DETAILS AND PCR INFORMATION.
Table S1. Environmental description of localities
Figure S1. Map and localization of the study sites
Section 2.COMPONENT ANALYSES OF ENVIRONMENTAL VARIABLES AND SPATIAL DISTRIBUTION
Table S2.Component analysis for mixtures of quantitative and qualitative variables.
Table S3.Correlograms results from habitat parameters (CS1 and CS2)
Figure S2.Direction of environmental traits across CS1 and CS2.
Figure S3.Correlograms results from habitat parameters (CS1 and CS2)
Section 3.LANDSCAPE RESISTANCE MODELS: TESTING ISOLATION-BY-RESISTANCE
Table S4. Conductance values for each land-cover type
Table S5. Results with the correlation coefficients of the different IBR models.
Figure S4. Map with the study area showing the basic cover types landscapes and sampling sites.
Figure S5. Resistance map with the best IBR model.
Figure S6.Mean ±2SE of (a) mass at metamorphosis, (b) larval period, and (c) growth rate
Section 4.TRβ HAPLOTYPES DESCRIPTION AND GENOTYPIC INFORMATION PER POPULATION
Table S6. Base composition of TRβ haplotypes detected and frequencies
Table S7. Summary statistics for TRβ gene variation in the 17 Ranaarvalispopulations
Table S8. FST TRβ genetic differentiation between pairs of populations
Table S9.Results from the machine learning approach for phenotypic-TRβ gene relationship.
SECTION 1.PREDATOR ABUNDANCE ESTIMATION DETAILS AND PCR AMPLIFICATION
DETAILS.
Predator abundance estimation details.Macroinvertebrate predators and newts were sampled in mid June
and early July. Sweep net (diameter dimension: 32 cm, mesh size: 0.08 cm) samples were taken by
holding the net in a vertical position and moving from the water surface through the vegetation and
water column five times over a distance of 3-5 m (Cheal et al. 1993). Five sites were selected in each
wetland, resulting in 25 samples for each pond. Sampling was restricted to the littoral region and to
depths of 0.5-1.5 m. Invertebrates and newts were identified, classified as potential predators,
counted and then returned to the water. Three types of insects were considered as predators: large
dragonflies (aeshnid and libellulid naiads), aquatic heteroptera (notonectids) and diving beetles
(larvae or adults). Predator densities were determined based on the water volume sampled,
calculated by using the dimensions of the net frame x distance walked.
PCR information. Multiplexed PCR amplifications were performed in a final volume of 10 μl containing 50 ng
DNA template, 5 µL PCR mix (QIAGEN Multiplex PCR Kit), 1 µL 10xprimer mix, and 3 µL RNAsefree water in an Applied Biosystems Gene Amp PCR Systems 2700 thermal cycler. PCR profiles
consisted of 3 min denaturation at 94°C followed by 23-25 cycles of 30s denaturation at 94°C, 30 s
annealing at 50-55°C and 30 s extension at 72°C with two final steps: 2 min 48 °C and 5 min at 72ºC
respectively. PCR products were genotyped on a MegaBACE 1000 automatic sequencer (Amersham
Biosciences, Buckinghamshire, UK) and scoring was performed in a blind manner using the software
Fragment Profiler (Fragment Profiler 1.2, Amersham Biosciences, 2003).
Figure S1. Map with the spatial distribution of ponds and potential connectivity among ponds within the study
area. Red dots represent the 17 ponds used in the present study; small blue circles are other wetlands where
reproduction of Rana arvalis has been observed. X and Y axis are expressed in UTM coordinates. Each tick
corresponds to five kilometres.
Table S1.- Ecological parameters per sites: Geographical coordinates (GPS), breeding time (Day) respect
first of January, habitat category (Hab: M=marshes, P=ponds), water permanency category (Per:
T=temporary, P=permanent), mean of water temperature, percentage of forest canopy cover, relative density
of potential predators, percentage of emergent vegetation, number of clutches used in genetic analyses.
Site
d1
d2
d3
d4
d5
d6
d7
d9
d10
d11
d12
d13
d14
d15
d16
d17
d18
GPS
59°43'54.12"N
16°59'9.24"E
59°43'39.78"N
16°49'52.14"E
59°44'28.92"N
16°50'22.32"E
59°45'17.34"N
17°2'6.72"E
59°51'10.17"N
17°28'21.31"E
59°37'28.55"N
17°13'3.27"E
59°45'14.04"N
17°24'37.50"E
59°38'24.60"N
16°54'16.20"E
59°36'57.30"N
16°55'35.64"E
59°42'53.76"N
17°4'15.60"E
59°46'44.88"N
17°1'24.90"E
59°47'56.46"N
16°52'12.90"E
59°50'29.46"N
17°21'48.18"E
59°43'41.52"N
16°53'58.92"E
59°45'42.84"N
16°54'8.94"E
59°52'40.98"N
17°12'8.40"E
59°55'53.64"N
17°17'49.74"E
Day
Hab
Per
WTemp
Canopy
Predators
EmVeg
N clutches
92
M
T
11.15
60
0.88
83
27
92
M
T
13.16
60
1.31
24
26
87
P
P
13.94
0
3.12
29
30
95
P
P
12.24
40
1.54
20
30
94
P
P
12.86
0
2.38
46
25
99
M
T
12.86
0
3.81
65
25
100
M
T
13.62
0
1.44
22
30
96
P
T
12.81
100
2.27
27
7
101
P
P
11.35
80
3.09
42
30
102
M
P
10.82
40
3.83
43
22
103
M
T
11.89
40
0.44
61
30
103
M
T
11.16
20
0.75
85
20
107
M
T
9.49
80
0.86
59
24
108
M
T
11.27
40
1.42
78
28
108
P
P
14.03
20
1.78
38
18
109
M
T
8.94
100
1.03
77
26
105
P
P
13.45
60
1.23
39
29
Section 2. COMPONENT ANALYSES OF ENVIRONMENTAL VARIABLES AND SPATIAL
DISTRIBUTION
Table S2. Results of the analysis of a mixture of numeric variables and factors using “dudi.mix” function in
ade4 R library. temp = temperature; pred = predator abundance; eveg = emergent vegetation; cate =
category (marsh or pond); cano = forest canopy; perm = permanency (temporary or permanent pond); dept =
water depth.
CS1
CS2
CS3
Temperature
0.405
0.339
-0.006
Forest Canopy
-0.281
-0.555
-0.013
Depth
0.336
-0.467
-0.047
Predator abundance
0.452
0.141
-0.020
Emergent vegetation
-0.450
-0.155
0.019
Category (M vsPo)
0.344
-0.3856
0.748
Permanency (T vs P)
0.341
-0.405
-0.660
Eigenvalues
4.6702
2.0211
0.2345
Proportion Variance
0.6671
0.2887
0.0335
Cumulative Variance
0.6671
0.9559
0.9894
0.8
0.6
temp
CS2 (28.87%)
0.4
0.2
pred
0.0
-0.2
eveg
cate
-0.4
perm
dept
-0.6
-0.8
-0.8
cano
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
CS1 (66.71%)
Figure S2. Analysis of a mixture of numeric variables and factors using “dudi.mix” function in ade4 R library
showing the weight and direction of the parameters across the two principal components (0.95% explained).
temp = temperature; pred = predator abundance; eveg = emergent vegetation; cate = category (marsh or
pond); cano = forest canopy; perm = permanency (temporary or permanent pond); dept = water depth.
Table S3.Correlograms results from habitat parameters (CS1 and CS2), assuming randomly distributed data,
and inverse distance weighting and with permutation tests based on 2000 permutations.
CS2
Min
Max
Pairs
I
Prob
RandProb
I
Prob
RandProb
0.00000
5771.17
14
-0.066
0.976
0.780
0.013
0.777
0.956
5771.17
9809.48
16
0.002
0.788
0.989
0.086
0.521
0.701
9809.48
11863.11
15
0.295
0.134
0.221
-0.005
0.821
0.983
11863.11
14895.78
15
0.162
0.313
0.451
-0.139
0.714
0.509
14895.78
18097.95
15
-0.042
0.941
0.838
0.173
0.299
0.426
18097.95
21223.42
16
-0.171
0.607
0.440
-0.819
0.0005
0.002
21223.42
24950.84
15
-0.151
0.690
0.488
-0.104
0.843
0.671
24950.84
28245.22
15
-0.255
0.395
0.259
0.111
0.460
0.640
28245.22
32700.17
16
-0.157
0.656
0.451
-0.221
0.459
0.331
32700.17
40860.96
16
-0.247
0.365
0.220
0.281
0.102
0.188
1.0
1.0
0.5
0.5
Moran's I
PCA2_pond
Moran's I
PCA1_pond
CS1
0.0
-0.5
-1.0
0
10000
20000
30000
40000
50000
0.0
-0.5
-1.0
0
10000
20000
30000
40000
50000
Figure S3.Correlograms results from habitat parameters (CS1 and CS2), assuming randomly distributed
data, and inverse distance weighting and with permutation tests based on 2000 permutations.
Section 3. LANDSCAPE RESISTANCE MODELS: TESTING ISOLATION-BY-RESISTANCE
Landscape traits included in the resistance models.The input for Circuitscape is a raster map in which
each cell is assigned a conductance value corresponding to the relative probability of R. arvalis moving
through the habitat type encoded by the cell. To derivate the grids, we used landscape variables that we
predicted would have possible relevance to habitat selection, vagility and gene flow based on relevant
literature and expert opinion. The value of resistance assigned to each habitat type was based on previous
studies on this species (Arens et al. 2007; Loman and Lardner 2009), which demonstrated the negative
impact of crops and human infrastructures (roads and urban settlement) on frog movements and gene flow,
and empirical studies with other forest amphibian species which showed the negative effects of forest
clearcuts and open canopy areas on species vagility (e.g. Semlitsch et al. 2008; Popescu and Hunter 2011).
Because the absence of hills or mountains in our study area, we did not consider slopes and ridges in our
models. Soil texture has been considered an important variable for landscape resistance because it reflects
the presence of subterranean refugia and the risk of dehydration for amphibians (Cosentino et al. 2011). As
we did not have information about soil texture, this factor was not included in the models.
To generate the land cover data sets, we acquired a 5-m-resolution digital land cover map of the study area
(http://www.metria.se/), which contains hundreds of detailed land classes. We simplified these classes into
eight general categories representing basic cover types that previous studies have demonstrated to influence
R. arvalis dispersal (Figure S2 and Table S2). To generate the land cover input file, we assigned different
conductance values to each land cover category. Low conductance value indicates a high resistance to frog
movement: big water bodies = 1; urban and industrial areas = 1; low urban areas = 3; grassland and
agriculture = 5; agriculture-forest transition = 6; recently clearcut forest = 6; forest = 8; wetlands = 10. These
ranks were chosen to reflect the relative order of conductance of these landscape features based on previous
empirical and non-empirical studies on amphibians. The magnitudes of these ranks, although reflecting
expert opinion, were arbitrary, and assigning resistance values to different land cover categories can be
problematic (Spear et al. 2010). Subsequently we created different conductance maps and models, to
evaluate the extent to which the results were influenced by the magnitude of these rankings (Table S4). Grids
were exported from ArcGIS for Circuitscape analysis using the “Export to Circuitscape” tool
(http://www.circuitscape.org/Circuit- scape/ArcGIS.html).
Table S4. Conductance values for each land-cover type for calculating resistance distances between
breeding ponds. We created four resistance models: IBD is the control Euclidean distance model, where
conductance for each land-cover is maximum and equal; this model did not consider landscape
heterogeneity. IBRlow, IBRm and IBR high are three resistance models considering landscape heterogeneity,
from low resistance to high resistance model.
Landclasses
Conductancevalues
IBD
IBRlow
IBRm
IBRhigh
Big waterbodies
10
1
1
1
Urban and industrial areas
10
2
1
1
Lowdensityurbanareas
10
5
3
2
Grassland and agriculture
10
6
5
3
Agriculture-foresttransition
10
7
6
4
Forestrecentlyclearcut
10
8
6
5
Forest
10
9
8
8
Wetlands
10
10
10
10
Figure S4. The study area in Sweden showing basic cover types landscapes and breeding sampling sites.
Yellow: crops and agricultural areas. Light green: forest. Dark green: recently clearcut forest. Red: semiurban
areas. Grey: urban and industrial areas. Blue: big water bodies.
The relative importance of various landscape features in predicting population genetic structure across the
study area was analyzed by conducting a series of Mantel tests examining the correlation between pairwise
genetic structure (FST/(1-FST)) and models of pairwise landscape resistance distance: isolation-by-resistance
(McRay 2006). We used Circuitscape 2.2 to generate pairwise matrices of landscape resistance based on the
landscape. We used simple Mantel tests and significance testing to compare the models with the software
PASSaGE 2 (Rosenberg and Anderson 2011). Significance of Mantel correlations was assessed based on
200000 random permutations of the data. Results obtained of the Mantel correlations for each model are
summarized in Table S5. All resistance models considering landscape heterogeneity showed a higher
correlation coefficient than the simple Euclidean model (IBD). As IBRm showed the highest r-value, it was the
model used in the main text and the other analyses. Figure S5 illustrates the resistance map between the
populations within the area of study created with the IBRm land conductance values.
Table S5. Correlation coefficient and corresponding significance for Mantel test comparing neutral genetic
distance (FSTN) and geographical distance assuming, either the Euclidean model (IBD not considering
landscape heterogeneity), or one of the three resistance distance models. Bold letters highlight the model
with the highest correlation coefficient, and thereby the one used in the other analyses included in the main
text.
Matrix
Mantel r
P-value
FSTN-IBD
0.0790
0.2364
FSTN-IBRlow
0.0704
0.2445
FSTN-IBRm
0.0946
0.2017
FSTN-IBRhigh
0.0847
0.2388
Figure S5. Resistance map between the sampling breeding ponds within the area of study. Blue color shows
areas of low conductance, which are expected to have low densities of dispersing frogs; yellow and red color
show well connected areas with low resistance to animal movements.
REFERENCES









Arens P, van der Sluis T, van’t Westende W, et al.(2007) Genetic population differentiation and connectivity
among fragmented Moor frog (Ranaarvalis) populations in The Netherlands. Landscape Ecology22, 1489-1500.
Cosentino B J,Schooley RL, Phillips CA (2011) Connectivity of agroecosystems: dispersal costs can vary among
crops. Landscape Ecology26, 371–379.
Loman J, Lardner B (2009) Does landscape and habitat limit the frogs Ranaarvalis and Ranatemporaria in
agricultural landscapes? A field experiment. Applied Herpetology6, 227-236.
Murphy MA, Dezzani R, Pilliod DS, Storfer A (2010) Landscape genetics of high mountain frog metapopulations.
Molecular Ecology19, 3634-3649.
Popescu VD, Hunter ML (2011) Clear-cutting affects habitat connectivity for a forest amphibian by decreasing
permeability to juvenile movements. Ecological Applications21, 1283-1295.
Rosenberg MS, Anderson CD (2011) PASSaGE: Pattern Analysis, Spatial Statistics and Geographic Exegesis.
Version 2. Methods in Ecology and Evolution2, 229-232.
Semlitsch, RD, Conner CA, Hocking DJ, Rittenhouse TAG, Harper EB (2008) Effects of timber harvesting on
pond-breeding amphibian persistence: testing the evacuation hypothesis. Ecological Applications18, 283–289.
Spear SF, Balkenhol N, Fortin M-J, McRae BH, Scribner KIM (2010) Use of resistance surfaces for landscape
genetic studies: considerations for parameterization and analysis. Molecular Ecology19, 3576-3591.
Spear SF, Peterson CR, Matocq MD, Storfer A (2005) Landscape genetics of the blotched tiger salamander
(Ambystomatigrinummelanostictum). Molecular Ecology14, 2553-2564.
Figure S6.- Mean ±2SE of (a) mass at metamorphosis, (b) larval period, and (c) growth rate from each
locality (d1, d2, […], d17) in the common-garden experiment. Localities are ordered by breeding time, with
earlier localities in the beginning of the x-axis, and later breeding localities at the end of the axis.
Section 4. TRβ HAPLOTYPES DESCRIPTION AND GENOTYPIC INFORMATION PER POPULATION
Number of Loci detected: 13
Position of loci: 63 71 79 121 124 160 183 230 321 537 693 723 759
Synonymous changes (7 of 13): 63 183 321 537 693 723 759
Non-synonymous changes (6 of 13): 71 79 121 124 160 230
Haplotypes parsimony informative sites > 1%
Table S6.Base composition of haplotypes detected in the families analysed.
No. haplotype
Base composition
Frequency
HTRβ 1
GCTAGGAGGTGGC
5%
HTRβ 2
GCTAGGGGGTGGC
1%
HTRβ 3
GCTAGGGGGTGAC
3%
HTRβ 4
GCTAGGGGACGGC
1%
HTRβ 5
GCTTGGAGGTGGC
40%
HTRβ 6
GCTTGGAGGTGAC
1%
HTRβ 7
GCTTGGAGGCGGC
2%
HTRβ 8
GCTTGGAGATGGC
1%
HTRβ 9
GCTTGGGGGTGGC
42%
Fast development = short larval period Loci 4 and 7 are different from HTRβ 1
HTRβ 10
GCTTGGGGGTGAC
10%
Slow development = long larval period
Loci 12 are different from HTRβ 1, 9 and 13
HTRβ 11
GCTTGGGGGCGGC
7%
HTRβ 12
GCTTGGGGGCGAC
>1%
HTRβ 13
GCTTGGGGATGGC
8%
Slow development = long larval period
Loci 9 are different from HTRβ 1, 9 and 10
HTRβ 14
GCTTGGGGATGAC
3%
HTRβ 15
GCTTGGGGACGGC
27%
HTRβ 16
GCTTGGGGACGAC
1%
Fast development = short larval period Loci 4 and 7 are different from HTRβ 9, 10 and 13
HTRβ 17
GCTTGGGCGCCGA
1%
HTRβ 18
GCTTGAAGGTGGC
>1%
HTRβ 19
GCTTGAGGGTGAC
1%
HTRβ 20
GCTTGAGGATGGC
>1%
HTRβ 21
GCTTGAGGACGGC
>1%
HTRβ 22
GCTTAGGGGTGGC
>1%
HTRβ 23
GCTTAGGGGTGAC
1%
HTRβ 24
GCTTAGGGGCGGC
1%
HTRβ 25
GCTTAGGGATGGC
>1%
HTRβ 26
GCTTAGGGACGGC
1%
HTRβ 27
GCATGGAGGTGGC
1%
HTRβ 28
GCATGGGGGTGGC
1%
HTRβ 29
GCATGGGGGCGGC
>1%
HTRβ 30
GCATGAGGGTGGC
>1%
HTRβ 31
GATTGGGGATGGC
1%
HTRβ 32
GATTGGGGACGGC
1%
HTRβ 33
GATTAGGGACGGC
1%
HTRβ 34
ACTTGGAGGTGGC
5%
HTRβ 35
ACTTGGAGACGGC
>1%
HTRβ 36
ACTTGGGGGTGGC
1%
Table S7. Summary statistics for TRβ gene variation in the 17 Rana arvalis wetlands analyzed.
d1
d2
d3
d4
d5
d6
d7
d9
d10
d11
d12
d13
d14
d15
d16
d17
d18
No of bp
882
882
882
882
882
882
882
882
882
882
882
882
882
882
882
882
882
Singletonsites
5
1
5
4
0
1
4
1
3
2
4
4
3
2
1
0
1
Parsimonyinformativesites
7
9
5
9
6
5
9
3
8
4
6
5
4
12
4
4
9
No of mutations
12
10
10
11
6
6
8
4
11
6
10
9
7
13
5
4
10
Syn
5
8
5
5
4
4
5
3
6
4
7
7
4
10
4
4
7
Non-syn
7
2
5
6
2
2
3
1
5
2
3
2
3
3
1
0
3
No haplotypes
15
13
15
18
14
11
12
5
21
7
22
14
8
14
8
8
12
Haplotypediversity
0.831
0.879
0.875
0.895
0.877
0.816
0.883
0.758
0.845
0.775
0.877
0.813
0.717
0.904
0.782
0.828
0.828
Table S8. FST genetic distances between pairs of wetlands for the TRβ gene. Bold indicate significant FST values for q-values<0.05 using a FDR
approach.
d1
d2
d3
d4
d5
d6
d7
d9
d10
d11
d12
d13
d14
d15
d16
d17
d1
--
d2
0.071
--
d3
0.026
0.015
--
d4
0.051
0.027
0.013
--
d5
0.042
0.021
0.015
0.001
--
d6
0.016
0.038
0.020
0.018
-0.003
--
d7
0.045
0.033
-0.006
0.026
0.032
0.047
--
d9
-0.001
0.022
0.000
-0.005
-0.014
-0.040
0.032
--
d10
0.037
0.106
0.047
0.033
0.042
0.043
0.058
0.016
--
d11
0.008
0.043
-0.006
0.035
0.024
0.014
0.017
-0.014
0.039
--
d12
0.025
0.019
-0.005
0.004
0.005
0.007
0.017
-0.010
0.024
0.008
--
d13
0.000
0.103
0.057
0.081
0.065
0.046
0.062
0.014
0.054
0.023
0.062
--
d14
0.025
0.062
0.012
0.025
0.008
-0.010
0.040
-0.037
0.019
-0.005
0.007
0.028
--
d15
0.084
0.045
0.026
0.037
0.049
0.061
0.047
0.037
0.046
0.037
0.022
0.114
0.050
--
d16
-0.011
0.036
0.008
0.034
0.011
-0.004
0.028
-0.034
0.044
-0.012
0.014
-0.007
-0.008
0.065
--
d17
0.095
-0.004
0.005
0.022
0.023
0.066
0.007
0.060
0.099
0.045
0.022
0.124
0.079
0.041
0.057
--
d18
0.015
0.031
0.013
0.022
0.003
0.011
0.020
-0.012
0.045
0.006
0.021
0.011
0.009
0.052
-0.012
0.031
d18
-
Table S9. Detailed results obtained from the machine learning approach to establish a relationship between phenotypic traits and TRβ gene
sequence.
Larval period
Correctly Classified Instances
Incorrectly Classified Instances
Growth rate
Mass at metamorphosis
120
64.86%
107
57.83%
77
41.62%
65
35.13%
78
42.16%
108
58.37%
Kappa statistic
0.1745
0.2891
0.2176
Mean absolute error
0.1679
0.1772
0.1954
Root mean squared error
0.3145
0.3278
0.3459
Relative absolute error
82.63%
82.53%
88.71%
Root relative squared error
99.46%
100.55%
104.46%
Coverage of cases (0.95 level)
87.56%
89.72%
82.16%
Larval period
Class
ROC Area
-inf - 37
0.626
37 – 38.4
0.818
38.4 – 39.8
0.589
39.8 – 41.2
0.601
41.2 - inf
0.773
Growth rate
Class
ROC Area
-inf – 0.0096
0.185
0.0096 – 0.0103
0.942
0.0103 – 0.0109
0.769
0.0109 – 0.0116
0.651
0.0116 – 0.0122
0.454
0.0122 – inf
0.775
Weighted average
Weighted
average
0.723
0.717
Mass at metamorphosis
Class
ROC Area
-inf – 0.416
0.418
0.416 – 0.442
0.781
0.442 – 0.468
0.606
0.468 – 0.494
0.641
0.494 – 0.520
0.549
0.520 – 0,546
0.516
0.546 - inf
0.748
Weighted
0.652
average