Intro:
Electronic Supplemental Material (ESM):
Journal: Oecologia
Title: Thermal and maternal environments shape the value of
early hatching in a natural population of a strongly cannibalistic
freshwater fish
Authors: Thilo Pagel1,2, Dorte Bekkevold3, Stefan Pohlmeier1,
Christian Wolter1 and Robert Arlinghaus1,2
1Department of Biology and Ecology of Fishes, Leibniz-Institute
of Freshwater Ecology and Inland Fisheries, Mggelseedamm 310, 12587
Berlin, Germany
2Division of Integrative Fisheries Management,
Albrecht-Daniel-Thaer Institute of Crop and Agricultural Sciences,
Faculty of Life Sciences, Humboldt-Universitt zu Berlin,
Philippstrae 13, 10155 Berlin, Germany
3National Institute of Aquatic Resources, Technical University
of Denmark, Vejlsvej 39, 8600 Silkeborg, Denmark
Thilo Pagel
Tel.:+49(0)30 64181 724
Fax:+49(0)30 64181 750
e-mail: [email protected]
This supplement consists of five parts:
Online resource 1: Optimal air temperature averaging period
Online resource 2: Age validation
Online resource 3: Parentage assignment method
Online resource 4: Model results and parameter estimates
Online resource 5: Supplementary references
Online resource 1: Optimal air temperature averaging period
A linear regression model based on the method described by
Matuszek and Shuter (1996) was developed to calculate missing daily
average water temperatures for 2008. The analysis was based on
daily air and water temperature measurements between 04 April and
19 June for all three sampling years. Air temperature data were
obtained from a weather station located 25 km from Kleiner Dllnsee.
Daily air temperatures were calculated as the mean of daily minimum
and maximum temperatures. Water temperature in 2008, as mentioned
in the main text, was measured using YSI-Multi-Parameter-Sensor
(YSI 6600, Yellow Springs, Ohio). In the two subsequent years,
water temperature was measured using 11 (2009) or 5 (2010)
temperature loggers (Hobo StowAway TidbiT v2). Independent
variables used to predict mean daily water temperature included
mean air temperature (T) for 0, 5, 10, 15, 20, 25 and 30-day
periods (each period extending back in time from the day the water
temperature was measured). In addition, day of the year (YDAY) and
its transformations (square, cube and logarithm) was included in
the model (as a time function). The optimal air temperature
averaging period for predicting water temperature was then
estimated based on maximum r2 (adjusted) and different measures of
the goodness of fit (AICc and AICc). The best model was used to
impute missing values.
Table 1a Model summary of linear regression models used to
determine the optimal air temperature averaging period for Kleiner
Dllnsee in the three sampling years.
Model: Mean daily water temperature
adj r2
N
K
AICc
AICc
1. 0 + 1T10 + 2YDAY + 3YDAY2 + i
0.928
231
4
692.116
0
2. 0 + 1T10 + 2YDAY + 3 logYDAY + i
0.922
231
4
692.135
0.019
3. 0 + 1T10 + 2YDAY + 3 YDAY3 + i
0.928
231
4
692.641
0.525
4. 0 + 1T10 + 2YDAY + i
0.910
231
3
744.796
52.681
5. 0 + 1T5 + 2YDAY + i
0.899
231
3
769.763
77.647
6. 0 + 1T15 + 2YDAY + i
0.899
231
3
771.844
79.728
7. 0 + 1T20 + 2YDAY + i
0.875
231
3
819.815
127.699
8. 0 + 1T25 + 2YDAY + i
0.867
231
3
833.984
141.868
9. 0 + 1T30 + 2YDAY + i
0.859
231
3
848.117
156.001
10. 0 + 1T0 + 2YDAY + i
0.829
231
3
891.862
199.746
11. 0 + 1T0 + i
0.620
231
2
1075.847
383.731
T = air temperature; YDAY = day of the year; 0 = intercept; i =
error term; N = total number of observations; K = number of
parameters; AICc = corrected Akaike`s information criterion; AICc =
delta AICc
Online resource 2: Parentage assignment method
DNA was extracted from caudal fin clips of all potential
spawners and age-0 pike using the E.Z.N.A.TM tissue DNA kit (Omega
Bio-Tek, Inc.) following the manufacturers guidelines. Polymerase
chain reaction (PCR) was used to amplify 16 microsatellite loci,
which were visualized and size fractioned using a BaseStation and
an ABI 3139 Genetic Analyser (Applied Biosystems, Forster City,
USA). Maternity was determined using the approach implemented in
CERVUS 3.0 (Kalinowsky et al. 2007). CERVUS was first used to
estimate the statistical power for assigning maternity to
offspring. A large number of offspring (10,000) were simulated
based on allele frequency estimates for 16 microsatellite loci in
all parental candidates collected across all three years (N =
1,130). Then, the statistical power to correctly assign age-0 pike
to a sampled female was estimated based on assigning the simulated
offspring, assuming that 85% of all spawning females in the lake
had been sampled. This estimate was based on the average proportion
of sampled mature females in relation to the estimated total mature
female population size (Pagel 2009). Based on the assignments of
simulated offspring, the critical delta associated with 95% correct
assignment was estimated, following Kalinowski et al. (2007). The
power to identify the correct mother was compared with the power to
simultaneously identify both mother and father, where all sampled
mature males and pike of unknown sex, were used as paternal
candidates. Numbers of sampled maternal and paternal candidates
varied over the three years (2008 to 2010) at respectively 338, 439
and 520 candidate mothers and 392, 473 and 584 candidate fathers.
Using that approach, some fraction of offspring could in theory
have been erroneously assigned paternity to a mother who could not
be sexed on collection. However this was not expected to lead to
bias in the current analysis, where only offspring that could be
assigned to a specific maternal candidate were used on subsequent
analyses. The probability of identity, defined as the probability
of two randomly sampled individuals from our data set having the
same genotype, was also estimated with CERVUS.
Sixteen microsatellite loci were typed in a total of 1,130
parental candidates and in 66, 104 and 134 age-0 pike from the
respective collection years 2008, 2009 and 2010. Loci exhibited
from 4 to 19 alleles, scoring success was high at 99.95% across
loci and individuals, and none of the sixteen loci exhibited
statistically significant deviation from Hardy-Weinberg proportions
(Table 1a). The Pid was estimated at 0.016. Simulation analyses
showed that applying critical delta for the three analysis years of
respectively 3.67, 3.47 and 3.53 would lead to 95% of all
assignments being to correct mothers. In comparison, critical delta
for correct assignment of fathers were somewhat higher (3.65, 4.03,
4.15), due to the assumed lower sampling efficiency on mature
males.
Table 2a Summary data for microsatellite marker types in all
candidate parent individuals collected across the three years.
Listed for each locus is the observed number of alleles (NA), the
expected (HE) and observed (HO) heterozygosity, the polymorphic
information content (PIC) together with tests for deviation from
Hardy-Weinberg expectations (P) and the original source. No locus
retained significance following correction for multiple testing
Locus
NA
HE
HO
PIC
P
Source
B24
11
0.803
0.793
0.775
NS
Aguilar et al. 2005
B117
6
0.099
0.095
0.096
NS
Aguilar et al. 2005
B259
10
0.817
0.803
0.792
NS
Aguilar et al. 2005
B281
6
0.700
0.693
0.649
NS
Aguilar et al. 2005
B422
9
0.472
0.475
0.450
NS
Aguilar et al. 2005
B451
19
0.897
0.897
0.888
NS
Aguilar et al. 2005
B457
18
0.852
0.857
0.836
NS
Aguilar et al. 2005
Elu2
5
0.183
0.171
0.175
NS
Hansen et al. 1999
EluBe
10
0.538
0.548
0.457
P < 0.05
Launey et al. 2003
EluB38
7
0.326
0.312
0.335
P < 0.05
Launey et al. 2003
EluB108
9
0.328
0.304
0.311
NS
Launey et al. 2003
EluB118
5
0.675
0.660
0.614
NS
Launey et al. 2003
Elu51
4
0.276
0.272
0.238
NS
Miller and Kapuscinski 1996
Elu64
4
0.369
0.359
0.315
NS
Miller and Kapuscinski 1996
Elu37
17
0.732
0.695
0.708
P < 0.05
Miller and Kapuscinski 1997
Elu76
19
0.816
0.807
0.793
NS
Miller and Kapuscinski 1997
NS = non-significant locus specific test
Online resource 3: Age validation
Age data from scales notoriously underestimate fish age and thus
need to be calibrated before it can be accepted as valid and
reliable method to age a given fish species (Campana 2001). Age
estimates of pike were validated by three different approaches.
Firstly, we compared the scale-read age of fish with the true age
obtained from tag-recapture data. In total, 208 pike were tagged
and recaptured in the period between 2007 and 2011. Ideally, first
tagging takes place very early in life where age estimates are
pretty certain (e.g., age-1). Accordingly, we only used pike of
age-1 to age-3 at first capture for tagging, assuming that the
initial aging error was negligible for these young fish. Using this
approach, a high correspondence between true age (y) and scale-read
age (x) was found (linear regression without intercept: y = 1.007x,
r = 0.990, P < 0.001, N = 133). Age estimates at first tagging
for all pike age-4 to age-6 were corrected using the parameters of
this model. This allowed us to include more and also older
individuals in the final analysis (all pike age-1 to age-6 at first
tagging). As shown in Figure 2a, a high correspondence was observed
between true age and scale-read age (linear regression without
intercept: y = 1.014x, r = 0.994, P < 0.001, N = 198),
indicating that our age estimates were reasonable and reliable.
Secondly, for some pike caught in the study lake on 13 April in
2005, age estimates by one reader were cross-checked with those
obtained by the same reader from cleithra. According to Laine et
al. (1991), cleithra yield more accurate age estimates for pike
especially for old individuals. Therefore, it was assumed that
cleithra-based estimates reflect the true age of pike (Babaluk and
Craig 1990; Casselman 1996). Total length of pike investigated
ranged between 14.7 and 74.5 cm, and age estimates varied between 0
to 7 years. A high agreement between age estimates by both scales
und cleithra (linear regression without intercept: y = 1.016x, r =
0.985, P < 0.001, N = 49) was obtained as shown in Figure 2b.
However, age estimates using scales tended to underestimate the
true (cleithrum) age slightly. Finally, regression analysis was
used to compare age estimates by scales from two different readers
using the same pike. Again, high agreement was observed (linear
regression without intercept: y = 1.011x, r = 0.960, P < 0.001,
N = 48; not shown). Based on these three lines of evidence, it was
assumed that the age estimates in our study and back-calculated
data such as juvenile growth by mature females reflected the true
values well, acknowledging a tendency for underaging old fish.
Fig. 3a Relation between true age (years) and scale-read age
(years) of pike from Kleiner Dllnsse (r = 0.994, P < 0.001, N =
198)
Fig. 3b Relation between cleithrum age (years) and scale age
(years) of pike from Kleiner Dllnssee (r = 0.985, P < 0.001, N =
49)
Online resource 4: Model results and parameter estimates
Table 4a General linear model (GLM) with total length of age-0
pike in early summer as dependent variable, year as a fixed factor,
hatch date and age as covariate
Source
Sum of Squares
df
F
P
Corrected model
94192.345a
6
205.549