bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023.

The copyright holder for this preprint (which

was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

1 Dissecting the genetic architecture of quantitative traits using genome-wide identity-by-descent sharing
2
3 Antoine Fraimout*1, Frédéric Guillaume1, Zitong Li1, Mikko J. Sillanpää2, Pasi Rastas1,3 & Juha Merilä1,4
4
5 1
Organismal and Evolutionary Biology Research Programme, Faculty of Biological and Environmental Sciences,
6 FI-00014 University of Helsinki, Finland
7 2
Research Unit of Mathematical Sciences, FI-90014 University of Oulu, Finland
8 3
Institute of Biotechnology, FI-00014 University of Helsinki
9 4
Area of Ecology and Biodiversity, School of Biological Sciences, The University of Hong Kong, Hong Kong
10 SAR
11 *Corresponding author: [email protected]
12
13 Abstract

14 Additive and dominance genetic variances underlying the expression of quantitative traits are important

15 quantities for predicting short-term responses to selection, but they are notoriously challenging to estimate in

16 most non-model wild populations. Specifically, large-sized or panmictic populations may be characterized by

17 low variance in genetic relatedness among individuals which in turn, can prevent accurate estimation of

18 quantitative genetic parameters. We used estimates of genome-wide identity-by-descent (IBD) sharing from

19 autosomal SNP loci to estimate quantitative genetic parameters for ecologically important traits in nine-spined

20 sticklebacks (Pungitius pungitius) from a large, outbred population. Using empirical and simulated datasets, with

21 varying sample sizes and pedigree complexity, we assessed the performance of different crossing schemes in

22 estimating additive genetic variance and heritability for all traits. We found that low variance in relatedness

23 characteristic of wild outbred populations with high migration rate can impair the estimation of quantitative

24 genetic parameters and bias heritability estimates downwards. On the other hand, the use of a half-sib/full-sib

25 design allowed precise estimation of genetic variance components, and revealed significant additive variance and

26 heritability for all measured traits, with negligible dominance contributions. Genome-partitioning and QTL

27 mapping analyses revealed that most traits had a polygenic basis and were controlled by genes at multiple

28 chromosomes. Furthermore, different QTL contributed to variation in the same traits in different populations

29 suggesting heterogenous underpinnings of parallel evolution at the phenotypic level. Our results provide

30 important guidelines for future studies aimed at estimating adaptive potential in the wild, particularly for those

31 conducted in outbred large-sized populations.

1
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

32
33 Keywords: quantitative genomics, heritability, identity-by-descent, Pungitius, relatedness, dominance

2
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

34 Introduction

35 Given the novel selection pressures stemming from global environmental change, our ability to predict short-

36 term adaptive responses in ecologically important traits in natural populations is becoming ever more important

37 (Waldvogel et al. 2020). Through the estimation of quantitative genetic parameters underlying phenotypic traits

38 (e.g., additive genetic variance VA and narrow-sense heritability h², Lynch & Walsh 1998) a population’s ability

39 to respond to selection can be estimated using various metrics and equations based on these parameters (e.g.,

40 evolvability, Hansen & Pelabon 2021; multivariate breeder’s equation, Lande 1979; Robertson-Price identity,

41 Robertson 1966, Price 1970). Therefore, estimation of quantitative genetic parameters in the wild is an important,

42 but at the same time, difficult challenge given the logistic demands associated with obtaining the required data.

43
44 Traditionally, estimation of VA and h² requires measures of continuous traits among individuals of known

45 relatedness, using specific crossing designs in the laboratory, or pedigree information gathered from natural

46 populations (Lynch & Walsh, 1998). In both cases, establishing the degree of relatedness between individuals is

47 logistically challenging because most non-model organisms are not amenable to laboratory rearing or because

48 building sufficiently large pedigrees requires extensive, and often long-term, work in the wild. However, the

49 genomic era has provided researchers with an increasing number of molecular markers allowing estimation of

50 the actual proportion of genome that is shared identically-by-descent (IBD) between individuals and by directly

51 building the Genomic Relationship Matrices (GRM) from wild individuals genotyped at many marker loci. The

52 GRM can subsequently be included into quantitative genetic models of variance partitioning (e.g., animal

53 models, Kruuk 2004) to evaluate the covariance between phenotypic and genetic resemblance. These advances

54 in quantitative genomics have been well explored in the fields of human and medical genetics (Visscher et al.,

55 2006, 2007; Manolio et al.,, 2009; Visscher, 2009; Powell et al.,, 2010; Speed & Balding, 2015), animal and

56 plant breeding (e.g. Edwards & Batley, 2010; Hayes & Goddard 2010; Sinclair-Waters et al., 2020), and more

57 recently also in evolutionary ecology (e.g., Sillanpää, 2011; Thompson, 2013; Robinson et al. 2013; Perrier et al.,

58 2018; Gienapp et al., 2019; Gervais et al. 2020; Duntsch et al. 2020; De la Cruz et al. 2020; Fraimout et al.

59 2022a). This ‘pedigree-free’ approach has proven particularly useful in getting accurate estimates of additive

3
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

60 genetic variance, heritability and genetic correlations underlying the expression of quantitative traits in several

61 wild (or semi-wild) populations including the Soay sheep (Bérénos et al. 2014, 2015), Roe deer (Gervais et al.

62 2019) and passerine birds (Robinson et al. 2013, Santure et al. 2013, Perrier et al. 2018).

63
64 Nonetheless, and despite having greatly advanced the field of evolutionary genetics, IBD-based approaches to

65 quantitative genetics in the wild carry drawbacks inherent to the need to sample wild individuals. They might

66 also not be suitable for all study systems. The main limitation is that random sampling of wild individuals from

67 populations with high levels of genetic variation increases the probability of sampling unrelated (i.e., genetically

68 distant) individuals. If most sampled individuals are unrelated, the statistical power of GRM-based models will

69 be low and heritabilities of focal traits can become underestimated (Ritland 1996, Ritland 2000, Ødergård &

70 Meuwissen 2012, Jensen et al. 2014, Gienapp et al. 2017). Although this might not be of concern for some

71 moderately closed or philopatric mammalian or avian study systems, or when deep pedigree information is

72 available, studies of species characterized by high fecundity, large effective population sizes or high levels of

73 gene flow could be impaired by the sampling of mostly unrelated individuals.

74
75 Here, we explore the utility of IBD-based quantitative genetic models in estimating genetic variance components

76 from large-sized populations with low relatedness among individuals. We hypothesize that the level of

77 relatedness in the population sample and its variance will determine the accuracy of the genetic variance

78 component estimates and that they will be biased downwards if level of relatedness in a population is low. To

79 this end, we estimated quantitative genetic parameters underlying phenotypic variation in a marine outbred

80 population of the nine-spined stickleback (Pungitius pungitius). We used empirical data on traits known to be

81 under selection in P. pungitius (Karhunen et al. 2014), and calculated heritability, additive and dominance

82 genetic variances in controlled crosses of laboratory-raised P. pungitius and compared these estimates to those

83 obtained from wild collected samples. Furthermore, we used computer simulations to verify the robustness of

84 our results and further test for the effect of migration rate on the estimation of quantitative genetics parameters.

85 Following our hypothesis, we expect that: i) data from controlled crosses should provide better estimates than

4
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

86 wild-collected data and that ii) the accuracy of quantitative genetics parameters should scale negatively with

87 migration rate among populations. In addition, by estimating the contribution of different chromosomes and

88 quantitative trait loci (QTL) to the phenotypic variance of all traits (Visscher et al., 2007; Yang et al., 2011) we

89 investigated the genetic architecture of the focal traits. We did this to look for further evidence (cf. Kemppainen

90 et al. 2021, Fang et al. 2021) that phenotypic parallelism across different P. pungitius populations is underlined

91 by lack of similar parallelism in genetic underpinnings of these phenotypic traits.

92
93 Our findings suggest that quantitative genomic approaches in the wild should be carried with caution in species

94 where relatedness mean and variance among individuals is expected to be low, that is, in species exhibiting low

95 degree of population structuring and/or high rates of migration.

96
97 Material and methods

98
99 Empirical datasets

100 Our first objective was to use different data structures and statistical approaches (and see, Quantitative genetics

101 section below) to assess their efficiency in obtaining estimates of variance components and heritability of

102 quantitative traits. To this end, we used three different datasets corresponding to three pedigree structures; i) a

103 population sample of wild-collected nine-spined sticklebacks from the Baltic Sea in Helsinki (60°13 N, 25°11 E) ′ ′

104 brought to the University of Helsinki aquarium facility. This “Helsinki-wild” dataset corresponds to a standard

105 quantitative genomic approach of building GRM from wild-collected individuals of unknown relatedness.

106 ii) Individuals from the Helsinki-wild dataset were used as founders to produce F1 generation offspring

107 following standard in vitro fertilization procedure for sticklebacks (Divino and Shultz, 2014). F1 generation Full-

108 sib/Half-sib families were generated by mating one female to two different males (i.e., maternal half-sib design).

109 Once the eggs hatched and larvae started feeding, the families were thinned to approximately 25 offspring per

110 family and moved to two large aquaria so that half of each family were placed in each aquaria unit. The larvae

111 were mass-reared in these aquaria, and their family identity was later identified from the genotype data (see

5
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

112 below). After eleven months (May 2013 to April 2014), the experiment was terminated and F1 offspring

113 distributed across 87 families were used. This dataset (“Helsinki-HS”) constitutes a classical common garden

114 Full-sib/Half-sib breeding design maximizing the number of different families and relatedness variation in the

115 sample. iii) The third dataset corresponded to a subset of the Helsinki-HS dataset retaining only 46 F1 full-sib

116 families (“Helsinki-FS” dataset). The rationale behind generating this dataset was to investigate the impact of

117 half-sib relationships on the estimated parameters (see also Simulation Study below).

118
119 Empirical data: Phenotypes

120 We obtained phenotypic data for all individuals of the three datasets on three morphological traits: body length,

121 body depth and length of the pelvic spines, all known to have heritable basis (Shimada et al., 2011; Kemppainen

122 et al., 2020) and be involved in adaptive differentiation between freshwater and marine populations (Karhunen et

123 al., 2014). Individuals were photographed next to a millimeter scale with a digital camera, and images were used

124 to measure body length and body depth using the tpsDig software (v.2.10; Rohlf, 2006). Body length was

125 measured as the distance between the tip of the snout and the base of the posterior end of the hypural plate. Body

126 depth corresponded to the distance between landmarks 3 and 12 in Herczeg et al. (2010). Length of the pelvic

127 spines were measured using digital calipers to the nearest 0.01 mm. We did not focus on the asymmetry of pelvic

128 spines (but see: Blouw & Boyd, 1992; Bell et al., 2007; Coyle et al., 2007) but rather decided to use the mean

129 values of the left- and right-side measurements as the focal trait. All measurements were made by the same

130 person twice, and the repeatabilities were > 0.9 (p < 0.001) for all traits (see also Yang et al., 2016, Kemppainen

131 et al., 2021). Sex information for all samples was obtained first by genotyping all individuals at a sex-specific

132 microsatellite locus (Stn 19; Shikano et al., 2011), and subsequently confirmed with sex-specific SNPs. After

133 correcting for missing values, the final datasets contained full phenotypic record for 133 wild individuals

134 (Helsinki-wild, body length only), 936 F1 individuals (Helsinki-HS) and 482 F1 individuals (Helsinki-FS).

135
136 Empirical data: SNP genotyping

6
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

137 SNP data was the same as in Kivikoski et al. (2021). In summary, 49 511 SNPs were obtained from

138 sequencing/genotyping by Diversity Arrays Technology (Pty Ltd, Australia), using their DarTseq technology.

139 Prior to all analyses, we pruned the datasets from sex-linked markers and markers with a minimum allele

140 frequency (MAF) lower than 0.01 using the raw.data function of the snpReady R package (Granato et al. 2018)

141 and included only markers with a maximum of 10% missing data. We further used the imput option of the

142 raw.data function to allow for data imputation of missing genotypes by using the mean genotypic value at each

143 SNP (Granato et al. 2018). This resulted in a total of 15 147 SNPs used for the construction of the GRM (see

144 below).

145
146 Quantitative genetics: Construction of the Genomic Relationship Matrices

147 We estimated two types of matrices describing the relationship among individuals in each empirical dataset: i)

148 pedigree-based relationship matrices (PRM), corresponding to the theoretical expectation of sibling relatedness

149 based on their pedigree and ii) SNP-based GRM corresponding to the estimated fraction of the genome shared

150 IBD between siblings. The PRMs were constructed from the pedigree structure describing each dataset using the

151 nadiv R package (Wolak, 2012) which allowed us to compute both additive (PRMADD) and dominance

152 (PRMDOM) matrices using the makeA() and makeD() functions, respectively.

153 For the construction of the GRM, we used a modified construct of the additive GRM originally proposed by

154 VanRaden (2008):

155

156     

 ᇲ ⁄
[1]

157
158 where W is the marker matrix of additive coefficients and n the number of individuals, as implemented in the

159 snpReady R package (Granato et al., 2018) for each dataset to construct GRMs from autosomal loci. This

160 allowed us to produce for each dataset both additive and dominance variance matrices (GRMADD & GRMDOM;

7
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

161 Granato et al., 2018). The construction of GRMDOM follows the same formula as in equation [1] replacing the

162 matrix W by the matrix S corresponding to the dominance deviation coefficients (see p. 5 in Granato et al. 2018).

163
164 Quantitative genetics: Estimation of variance components and heritabilities

165 Three quantitative genetic parameters underlying trait variability were estimated: additive genetic variance (VA),

166 dominance variance (VD) and heritability (h2). We employed the ‘animal model’ approach (Kruuk, 2004), a

167 random effect model using the PRM or GRM as the covariance of the random effect linking to the phenotypes,

168 to estimate these parameters using the relatedness among individuals in each dataset characterized in each

169 relationship matrix. The Bayesian approach implemented in the MCMCglmm R package (Hadfield, 2010) was

170 used to fit the following model for each trait and each dataset separately:

171

172        [2]

173
174 where y is the vector of phenotypic values for each trait, β is the vector of fixed effects, a and d are the vectors of

175 random additive and dominance effects, respectively. ε is the vector of residual errors and X, ZA and ZD the

176 design matrices relating to the fixed effects and the additive and dominance random effects, respectively. We

177 added the sex of the individuals as a fixed effect and implemented additive and dominance effect from either

178 type of relationship matrix (PRM or GRM) as random effects. Model convergence was checked visually from

179 the mixing of MCMC chains by inspecting trace plots of many different model parameters, using the

180 Heidelberger and Welch convergence test (Plummer et al. 2006). We assessed whether models had been run for

181 enough iterations by inspecting the effective sample sizes of the MCMC posterior distributions of each variance

182 components. We calculated heritability as the ratio of VA to total phenotypic variance (VP) for a given trait.

183 For each model, we evaluated the uncertainty in the estimation of variances and heritability based on the 95%

184 Highest Posterior Density (HPD) intervals constructed from the posterior distribution of each variance

185 component using the HPDinterval() function (Plummer et al. 2006). We then compared the 95% HPD intervals

186 for all parameters (VA, VD, h²) between models to assess the precision of each approach in estimating quantitative

8
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

187 genetic parameters. As the pedigree for the Helsinki-wild data contains no information (i.e., all individuals are

188 considered unrelated) we only ran models based on the GRM for this dataset. All variance components obtained

189 with MCMCglmm using GRMs and PRMs are reported as the median of the posterior distributions for each

190 model.

191 Finally, meaningful comparisons of variance components estimated using different relationship matrices (here,

192 GRM & PRM) must refer to a common reference population wherein genetic variance is estimated (Speed &

193 Balding 2015, Legarra 2016). Specifically, models using PRM will estimate variance in the base population of

194 the pedigree (i.e., the unrelated founders) while models using GRM will do so within the set of genotyped

195 individuals (Powell et al. 2010, Legarra 2016; Joshi et al. 2020).

196 Here, we followed the approach of Legarra (2016) and adjusted all variance components as follows:

197

198   .       . 
  [3]

199
200 where ² is the variance component estimated from the animal model and K the corresponding relationship
σ

201 matrix (Legarra 2016; Joshi et al., 2020). By doing so, we set a reference population to provide meaningful

202 comparison between the PRM and GRM-based models.

203
204 Genetic architecture: Chromosome partitioning of the genetic variance and QTL mapping

205 To investigate the genetic architecture of the three quantitative traits, we estimated the proportion of phenotypic

206 variance explained by SNPs from each separate chromosome. To this end, we used the Genomic Best Linear

207 Unbiased Prediction (G-BLUP) approach from the GCTA software (Yang et al. 2011). We calculated a GRM for

208 each chromosome and a GRM for all but the focal chromosome and estimated genetic variance parameters with

209 the GREML approach implemented in GCTA. Models were run independently for each trait and each

210 chromosome for all crosses.

9
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

211 Furthermore, we performed QTL mapping using three inter-cross F2 datasets (see Supplementary methods). We

212 used a single-locus mapping approach (Li et al. 2017, 2018) to identify the QTL explaining variation in body

213 size, body depth and pelvic spine length. The total phenotypic effect of each SNP can be obtained from the

214 regression coefficient of the Gaussian regression model:

. .
215         ,  0,  , [4]

216 where yi is the vector of phenotypes for individuals i; β0 is the phenotypic mean; xij is the genotypic value of

217 individual i and marker j coded as -1, 0 and 1 for the three genotypes AA, AB and BB, respectively; βj is the

218 effect of the SNP j, and εi is the residual error assumed to be independent and identically distributed (i.i.d.) under

219 a normal distribution with zero mean and unknown variance  . In our inter-cross F2 design, up to four possible

220 segregating alleles can be found: two alleles A1 and A2 from the dam, and two alleles B1 and B2 from the sire

221 (Xu, 1996, 2013). Thus, model [4] can be reformulated as:

. .
222               ,  0,  , [5]

223 where the substitution effects βdj and βsj correspond to the alleles A1 and A2 and B1 and B2 at locus j for the

224 dam d and sire s, respectively, and where  is the dominance effect. In model [5], the genotype matrix coding

225 system for  ,  and  (Xu, 2013) is specified as:

1 1 1 1!1
1
1
1
1
1
1
1!2
2!1
1 1 1 2!2

226 To obtain additional information on the origin of the QTL effects and distinguish between the dam and sire allele

227 effects, parental phasing information needs to be incorporated in the model [5]. To this end, we used the data

228 produced in Kemppainen et al. (2021). Briefly, parental and grandparental phases were obtained from a dense

10
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

229 SNP panel (see detailed description in Kemppainen et al. 2021 and Li et al. 2018) using the LEP-MAP3

230 software (Rastas, 2017). Data redundancy due to linkage was reduced using a linkage disequilibrium (LD)

231 network approach implemented in the LDna R package (Kemppainen et al. 2015). Each LD-cluster comprising

232 set of highly correlated SNPs was subjected to a principal component analysis (PCA) and the PC-coordinates

233 from the first axis explaining the largest proportion of variation was used for QTL mapping. For each of the

234 three F2 crosses, we applied the model from equation [5] separately to each phenotypic trait using the

235 complexity-reduced SNP panel and sex as covariate. A permutation procedure (10,000 permutations) was used

236 in each QTL model to control for false positives (Li et al., 2017; Westfall & Young, 1993).

237
238 Simulation study

239 We ran forward-in-time, Wright-Fisher simulations with Nemo (v.2.3.56; Guillaume & Rougemont 2006) to

240 generate genomic and phenotypic data for datasets varying in their pedigree structures. From a base population

241 of N = 10,000 individuals, three categories of datasets matching our empirical data were generated after 10N

242 generations: i) wild samples, ii) full-sib/half-sib crosses and iii) full-sib crosses. For each cross type we

243 generated different sample sizes to evaluate the number of individuals needed to obtain precise variance

244 component estimates. For the full-sib crosses, we simulated a dataset with three independent (i.e., unrelated

245 founders) full-sib families each consisting of 200 F1 offspring (“FS_3x200”, N = 600) and a dataset with five

246 independent full-sib families each with 200 F1 offspring (“FS_5x200”, N = 1000) and finally, a dataset

247 equivalent to our empirical dataset consisting of 50 families each containing 20 offspring (“FS_50x20”, N =

248 1000). For the full-sib/half-sib design we simulated a design consisting of 50 dams each mated to two sires and

249 each family consisting of either five F1 offspring per sire (“HS_50x10”, N = 500) or 10 F1 offspring per sire

250 (“HS_50x20”, N = 1000) and an additional design with 150 dams each mated to two sires and each family

251 consisting of three F1 offspring (“HS_150x6”). Finally, we simulated two datasets of wild unrelated individuals

252 consisting of 500 (“Wild_500”) or 1000 individuals (“Wild_1000”) randomly sampled from the base population.

11
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

253 Moreover, we explored the effect of variation in relatedness on the estimation of variance components by

254 simulating additional ‘wild’ datasets with varying levels of IBD among individuals. To this end, we simulated a

255 structured population (N = 10,000) of 5 and 10 sub-populations connected by migration in a Stepping-Stone

256 model. Migration rates varied between low (Nem = 1.6, 3.2) and high (Nem = 16, 32) (Fig. S1) so that relatedness

257 among all individuals would scale negatively with migration. In other words, under a low-migration regime,

258 individuals within sub-populations are more related to one-another (Fig. S1a, c), while high migration rates

259 homogenize genetic variation in the meta-population and consequently, decreases average relatedness (Fig. S1b,

260 d). After 10N generations of random mating (Wright-Fisher) and migration, we generated samples of 500

261 individuals equally sampled among sub-populations (100 or 50 individuals per sub-population).

262 For each dataset, a polygenic quantitative trait underlined by 100 biallelic QTL was simulated and 10,000 neutral

263 biallelic SNP markers were generated on the same genetic map for estimation of genomic relatedness. Loci were

264 placed on 20 chromosomes with total map length 190cM. Neutral markers were placed equidistantly every

265 0.5cM, while QTL locations were randomly set. QTL allelic effects were ±0.4 with mutation rate 10-4. Mutation

266 rate for neutral markers was 10-5. Narrow-sense heritability h2 was maintained constant across simulations by

267 adjusting the environmental variance of the trait VE to the within population additive genetic variance VA

268 stemming from mutation-drift(-migration) balance. GRMADD for the simulated genotypes were constructed using

269 the same approach as described above after pruning for markers with a MAF lower than 0.01. We used the same

270 modeling approach to estimate all variance components from each simulated dataset with MCMCglmm as

271 described above. For each dataset, 10 replicates were generated and the mean values of all replicates along with

272 their standard deviations are reported.

273
274 Results

275 Distribution of IBD-sharing values

276 The GRM estimated from the Helsinki-wild cross data was well representative of the full-sib/half-sib structure of

277 the data (Fig. S2) with a high frequency of full-sibs sharing half of their genome IBD (0.5; Fig. S2) and half-sibs

12
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

278 showing an average relatedness coefficient of 0.25. Most founders were ‘unrelated’ (IBD sharing < 0.1; Fig. S2),

279 but we did find some degree of relatedness among parents despite the random nature of the sampling (Fig. S2).

280
281 Quantitative genetics parameters

282 Overall, more precise estimates of quantitative genetic parameters were recovered from the half-sib/full-sib data

283 (as represented by the width of the 95% HPD intervals) compared to those obtained using either unrelated

284 individuals from the wild, or pedigree information (Fig. 1, Table S1). In the Helsinki-wild dataset, estimates of

285 quantitative genetic parameters for body size could not be recovered using the GRM, and genetic variance

286 component estimates, as well as heritability estimates, had wide 95% HPD intervals (Fig. 1, Table S1). We

287 found similar moderate heritability for the three quantitative traits (Fig. 1) but different levels of underlying

288 additive genetic variance, with SL showing the highest and PL the lowest VA (Fig. 1). Overall, the contribution

289 of dominance genetic variance (VD) in all traits for these datasets was small (Fig. 1). Accuracy of the parameter

290 estimation using the Helsinki-FS data was on par with the Helsinki-HS dataset, but the latter tended to be more

291 precise (i.e., with lower HPD intervals) for all variance component estimates (Fig. 1).

292 Results of the simulation study showed that VA and h² were highly underestimated in all but the half-sib/full-sib

293 crosses (Fig. 2). Accuracy of the estimation (i.e., distance of the point estimates from the true simulated value)

294 increased with increasing family numbers, while precision (i.e., width of the confidence intervals) increased with

295 increasing sample size (Fig. 2). As predicted, and despite the relatively large sample sizes, wild samples did not

296 allow for meaningful estimation of VA or h².

297 Further investigations of the wild datasets showed a clear effect of migration rate and relatedness levels on the

298 statistical power to estimate VA and h² (Fig. 3). Migration rates among the subpopulations in each simulated

299 meta-population scaled negatively with the accuracy of VA and h² parameters and the most accurate estimates

300 were obtained using the samples with the lowest migration rate and therefore, the highest level of relatedness

301 (Fig. 3, Fig. S1).

302
303 Genetic architecture: Chromosome partitioning

13
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

304 Overall, the proportion of variance explained by each individual chromosome for each trait was low to moderate

305 in all crosses (Fig. 4, 5), except for pelvic spine length (PL) in the HEL x RYT cross (Fig. 5) where

306 Chromosome 7 explained 66% (SE = 0.086) of the total phenotypic variance. In the Helsinki-HS, we found

307 significant regional heritability for body depth and pelvic length in Chromosomes 6 and 5, respectively (Fig. 4).

308 In the F2 crosses, different chromosomes contributed to the total phenotypic variance of all traits in the three

309 datasets (Fig. 5).

310 The proportion of variance explained by each chromosome was significantly correlated with chromosome length

311 for body size only in the HEL x PYO cross (r = 0.619, p = 0.003; see Fig. S3 and Table S2 for details on

312 chromosome length).

313
314 Genetic architecture: QTL mapping

315 The genetic architecture of the three focal traits differed between the three different F2 crosses (Fig. 6). In the

316 HEL x BYN cross, we found no significant QTL underlying the variation in standard length (Fig. 6A), one

317 significant QTL of parental origin (i.e., from the F0 pond sire) for body depth on Chromosome 16, and three

318 significant QTL for pelvic spine length on Chromosomes 15, 16 and 21 also of a paternal origin (Fig. 6A). In

319 addition, there was one significant QTL of maternal origin (i.e., F0 marine female) on Chromosome 6 (Fig. 6A).

320 In the HEL x RYT cross, a single significant QTL underlying the variation in pelvic spine length on

321 Chromosome 7 (Fig. 6B) inherited from both F0 parents was found. This QTL expressed also dominance on this

322 trait (Fig. 6B). In the HEL x PYO cross, there was a single paternal QTL for standard length located on

323 Chromosome 1 and one significant QTL for the dominance effect on Chromosome 9. A large effect QTL of

324 shared paternal, maternal and dominance origin located on Chromosome 9 was found to explain variation in

325 body depth (Fig. 6C). Finally, a single significant QTL on Chromosome 9 explained the phenotypic variation in

326 pelvic spine length in this cross (Fig. 6C).

327
328 Discussion

14
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

329 The results show that genome-wide IBD information derived from SNP markers using controlled crosses such as

330 half-sib/full-sib family can yield meaningful estimates of quantitative genetic parameters for ecologically

331 important traits, and greatly improve the precision of estimates compared to traditional pedigree-based

332 approaches and to estimates obtained from wild unrelated individuals. Moreover, implementation of the GRM in

333 the animal model allowed for greater model complexity, particularly by adding a random effect term for the

334 estimation of dominance genetic variance. In line with the results of some earlier studies, which have estimated

335 dominance variance components for morphological traits (Roff & Emerson 2007, reviewed in Wolak & Keller

336 2014), the dominance contributions for all traits were negligible in our data. In the following, we discuss the

337 usefulness and limitations of our results, and their potential utility for evolutionary studies of wild populations.

338
339 Quantitative genomics in the wild

340 Despite the appealing prospect of applying GRM-based animal models directly to samples originating from the

341 wild rather than from complex laboratory crosses, our results show, in line with early predictions (Ritland 1990,

342 Ritland et al. 1996), that such approach should not be undertaken in populations with low variance in relatedness.

343 This was particularly evident from our simulation study showing that estimates of heritability and additive

344 genetic variance were severely underestimated when using unrelated individuals with low variance in relatedness

345 compared to half-sib/full-sib data. Furthermore, the results show that in a simulated meta-population

346 representative of a wild outbred species, migration rate (or inversely, population structure) would have to be very

347 low, and variance in relatedness among individuals to be substantial in order to obtain accurate estimates of VA

348 and h². Of course, the used simulation scheme makes several assumptions (e.g., random mating and absence of

349 mate choice) that are not necessarily met in all species with more complex mating strategies and population

350 structures. Furthermore, sweepstake reproductive success typical of marine species (Vendrami et al. 2021) can

351 disrupt the genetic homogeneity of populations otherwise characterized by large census population size, and,

352 therefore, increase the mean and variance in relatedness at local geographic scales. Nonetheless, the simulations

353 were parameterized with results from real genomic data derived from the study of Baltic Sea nine-spined

354 sticklebacks (Fang et al. 2021, Feng et al. 2022). Hence, our results can be considered representative of at least

15
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

355 this outbred fish species, and likely other marine organisms with large population sizes. As such, our results

356 provide important guidelines for future studies aiming at estimating quantitative genetic parameters such as

357 marine species and more generally in species characterized by large population sizes.

358
359 Distribution of IBD values

360 In humans, relatedness among full-sibs can vary substantially around the 0.5 expectation: Visscher et al. (2006)

361 reported IBD sharing among full-sibs to range from 0.317 to 0.617. Here, we found the average IBD sharing to

362 span much wider range than in humans (mean IBD sharing: 0.509, min: 0.259; max: 0.737; Fig. S2). In other

363 words: some full-sib pairs did not share more of their genome than half-sibs, while others shared more than

364 expected from brother-sister mating. Variance in IBD sharing between sibs is inherently linked to structural

365 properties of the genome (i.e., number of chromosomes, chromosome length) and the recombination process and

366 number of crossovers occurring during meiosis: higher number of crossovers will lead to lower variance in IBD

367 (Risch & Lange, 1979; Kivikoski et al., 2021). Linkage maps constructed from human data are on average twice

368 the length as that from P. pungitius (Broman et al., 1998; Kivikoski et al., 2023) indicating more crossovers in

369 the human genome than in that of the nine-spined stickleback. Therefore, it is possible that this difference in the

370 recombination landscape between the two species along with the smaller size of P. pungitius’ genome (Xu 2006)

371 is responsible for the different range of IBD proportions in sticklebacks. Regardless of the mechanism behind the

372 higher variance in the nine-spined stickleback IBD, our results confirm that this species, and likely also its close

373 relative, the three-spined stickleback (Gasterosteus aculeatus), sharing similar genetic architecture and cross-

374 over pattern (Kivikoski et al., 2023), are particularly suitable models for quantitative genomic studies of wild

375 populations (Merilä, 2013). This view is further reinforced by the discovery that the two species differ

376 fundamentally in the way standing genetic variation is distributed among local populations, and therefore in the

377 way local adaptation is expected to proceed in response to similar selection pressures in different populations

378 (Fang et al. 2021).

379
380 Dominance variance underlying quantitative traits

16
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

381 The results further demonstrate that use of the GRMs in our models allowed for greater model complexity by

382 implementation of the dominance matrix (GRMDOM) as an additional random term to estimate VD. To date,

383 dominance genetic variance has received less attention in the evolutionary genetic literature (Wolak & Keller,

384 2014). This is in part due to the fact that founding theoretical work predicted that non-additive genetic effects

385 should contribute little to quantitative trait variance (Fisher, 1958), a prediction later verified in some empirical

386 studies (e.g., Hill et al., 2008; Zhu et al., 2015; Class & Brommer, 2020; but see: Merilä et al. 2004).

387 Nevertheless, a body of theoretical and empirical work suggest that dominance variance can indeed account for a

388 significant proportion of phenotypic variance in the wild when populations are finite and genetically structured

389 (e.g., Wright, 1931; Crnokrak & Roff, 1995; Kosova et al., 2010). Hence, the importance (or lack thereof) of

390 dominance genetic variance cannot be taken for granted and requires empirical data. Two results of particular

391 importance regarding dominance variance from our study are worth highlighting. First, we found that dominance

392 genetic variance accounted for a very low proportion of the total phenotypic variance for all traits in all datasets.

393 On average, it accounted for 0.19% of VP across datasets compared to VA, which explained on average 21.2% of

394 VP. Thus, our results are in line with some earlier studies suggesting that dominance variance in morphological

395 traits can be negligible. Second, the results further demonstrate the difficulty in obtaining estimates of VD from

396 family data. Here, even when using an appropriate crossing design (i.e., maternal half-sib; Wolak & Keller,

397 2014) with relatively large number of families and individuals, we found that estimates of VD obtained from

398 pedigree-based analyses tended to be overestimated as compared to the ones obtained with GRM. As a result,

399 uncertainty around the estimation of the other components of variance in animal model was affected, and

400 heritability estimates became biased downwards. All this said, we acknowledge that there are more advanced

401 breeding strategies combining paternal half-sibs and double-first-cousins that are tailored for dominance

402 variance estimation (Sztepanacz & Blows, 2015; Koch et al. 2020) and that larger sample sizes than those used

403 here might be required to obtain reliable estimates of VD. While it is possible that we did not have enough

404 statistical power to accurately estimate dominance variance in our data, the range of possible values for our focal

405 traits should nonetheless be captured by the HPD intervals of our model estimates.

406
17
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

407 The genetic architecture of parallel evolution in P. pungitius

408 The pace of adaptive evolution is determined by the amount of heritable variation in traits under selection, but

409 the heritability of a given trait may be specific to a given environment, or sex (Price & Schluter, 1991; Roff,

410 1997; Hoffmann & Merilä, 1999; Wilson et al., 2010). Here we studied three morphological traits previously

411 shown to be involved in the phenotypic differentiation between different ecotypes of P. pungitius. Following the

412 colonization of lakes and ponds from the marine environment, freshwater populations of P. pungitius have

413 repeatedly evolved distinct phenotypes such as increased body size (i.e., gigantism; Herczeg et al. 2009),

414 changes in behavior (Herczeg et al. 2009, Fraimout et al. 2022b) and pelvic reduction (Kemppainen et al. 2021).

415 A reasonable assumption is therefore that sufficient additive genetic variance was available in the marine

416 ancestral population to respond to the selection pressures associated with the colonization of the freshwater

417 habitat. In the current study, the Helsinki population may be considered as representative – i.e., genetically and

418 morphologically similar (Shikano et al., 2010, Karhunen et al., 2014) – of an ancestral marine population of P.

419 pungitius, and in line with the previous assumption, we found significant VA underlying all three traits in this

420 population. Further dissection of the genetic architecture of body size revealed that heritability of this trait was

421 partitioned across the genome, with several chromosomes accounting for small proportions of total phenotypic

422 variance. Thus, these results confirm the polygenic nature of body size in P. pungitius (Laine et al., 2013).

423 However, our analysis of pelvic spine length and body depth revealed a different picture: we found little

424 phenotypic and additive genetic variance underlying these two traits despite moderate heritabilities. Regressive

425 evolution of the pelvic apparatus (hereafter pelvic reduction) has been much studied in stickleback fishes (Bell et

426 al.,, 1993; Gibson, 2005; Chan et al., 2010; Xie et al., 2019; Kemppainen et al., 2021). Based on the results of

427 Kemppainen et al. (2021) and the current study, it is clear that the pelvic reduction is a heterogeneous process in

428 P. pungitius, and that the heritability of this trait varies from one population to another. Therefore, it is possible

429 that the alleles (or combination of alleles) underlying pelvic variation are not segregating in our Helsinki

430 population, in turn explaining the low additive variance of this trait. Alternatively, it is possible that the pelvic

431 phenotype in this population is close to an adaptive optimum where stabilizing selection has depleted most of the

432 additive genetic variance underlying this trait (Fisher, 1958; Karavolias et al., 2020). Whether body depth is

18
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

433 under a similar selective scenario in P. pungitius is however unknown, and additional studies of selection

434 intensity and in-depth analyses of the genetic architecture of these two traits (e.g., with genome wide association

435 study) may shed more light on their genetic bases in the Baltic Sea population of P. pungitius.

436 Interestingly, our chromosome partitioning analyses revealed heterogeneous chromosome contributions to the

437 overall phenotypic variance for all three morphological traits, between all crosses. This result was further

438 supported by different QTL underlying phenotypic variation between crosses, for all three traits. Differences in

439 the genetic architecture of the same trait between these crosses indicate that different alleles govern phenotypic

440 variation in the three different pond populations despite their adaptation to the same environment. As recently

441 demonstrated by Kemppainen et al. (2021) and Fang et al. (2021), marine populations of P. pungitius display

442 relatively high level of genetic isolation-by-distance resulting in substantial population structure in the sea.

443 Consequently, and conversely to the classical model of parallel evolution described in the three-spined

444 stickleback (Gasterosteus aculeatus; Colosimo et al., 2005), repeated colonization of similar freshwater habitats

445 by P. pungitius was likely initiated by populations with different allele frequencies at causal loci underlying

446 traits under selection. The likelihood of parallel evolution is therefore expected to be low among freshwater

447 populations of P. pungitius, as was demonstrated for pelvic phenotypes by Kemppainen et al., (2021) and further

448 suggested by the current study along with evidence for genetic non-parallelism in the two additional

449 morphological traits. This important result further reinforces the role of non-parallelism in polygenic adaptation

450 among populations evolving in similar habitats (Barghi et al. 2020).

451
452 Possible caveats

453 We found that genetic variances in our simulation study were encompassed in the HPD intervals of the animal

454 models, but was also slightly underestimated. Although we do not have a clear explanation for the observed

455 underestimation of VA, this result should not stem from technical issues associated with our animal models (i.e.,

456 convergence of MCMC chains, Fig. S4) and rather suggests a potential lack of statistical power to estimate VA

457 from the proposed designs. Nevertheless, our results further highlight the importance of variance in genetic

458 relatedness in the estimation of variance parameters.

19
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

459 Other sources of variance such as common environmental and parental effects may increase phenotypic

460 similarity within full-sib families and attenuate between-family differences (Lynch & Walsh 1998). Although

461 our data did not allow us to specifically account for parental effects, previous experimental findings of Ab Ghani

462 et al. (2012) found that maternal effects on body size in both marine and pond populations of P. pungitius were

463 negligible. Nonetheless, other type of experimental designs, such as paternal half-sib design (Lynch & Walsh

464 1998, Falconer & McKay 1996), able to yield estimates free from maternal effects would be preferable. As for

465 the common environmental effects, estimation of genetic variance from within-family variation using the actual

466 IBD relationships between full-sibs is not expected to be affected by such effects (Visscher et al. 2006).

467 Although the current manuscript did not aim at comparing different methods or software capable of IBD

468 estimation, it is worth noting that discrepancies may exist between different relatedness estimators (Ackerman et

469 al. 2017, Weir & Goudet 2017) and that the estimation of quantitative genetic parameters from GRM-based

470 models may require the use of several relatedness coefficients. Furthermore, the computing cost of Bayesian

471 sampling associated with large matrices such as the GRMs used in the present paper could quickly become

472 prohibitive with large sample sizes. Although providing a formal comparison of different R packages/software in

473 handling such matrices was beyond the scope of this manuscript, we refer interested readers to other statistical

474 tools such as BGLR (Perez & de los Campos, 2014) or brms (Bürkner 2017) and the recent work by de

475 Villemereuil (2018) on the matter.

476
477 Conclusions and prospects

478 In conclusion, we have explored the utility of a quantitative genomics approach in obtaining estimates of

479 quantitative genetic parameters for non-model organisms and demonstrated that this approach should be

480 undertaken carefully when sampling wild individuals, particularly when low variance in relatedness is expected

481 in the study populations. Rather than maximizing sample size only, we suggest that such studies should consider

482 sampling multiple local demes or multiple family groups to increase the variance in relatedness among

483 individuals. While the applicability of this approach may be limited for organisms that are not amenable to be

484 raised in laboratory/mesocosm conditions, it could be readily used for instance in many plant, insect, and fish

20
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

485 systems to obtain estimates of quantitative genetic parameters from multiple populations simultaneously.

486 However, since the accuracy of genomic heritability method depends on the variance in genome wide IBD (Xu,

487 2006; Visscher, 2009), the genomic characteristics of the model system will influence the utility and required

488 sampling design in obtaining parameter estimates with sufficient accuracy and precision.

489
490 Acknowledgements

491 We thank Federico Calboli and Ari Löytynoja for constructive comments on the earlier version of the manuscript.

492 We thank Gabor Herczeg, Abigel Gonda, Yukinori Shimada, Mirva Turtiainen, Chris Eberlein, Takahito

493 Shikano, Laura Hänninen, Kirsi Kähkönen, Miinastiina Issakainen and Sami Karja for help in fish breeding and

494 DNA extractions, and Jing Yang for help with fish phenotyping. We thank Mikko Kivikoski for helpful

495 discussion for estimation of IBD proportions. The computing resource support from CSC - the Finnish IT Center

496 for Science Ltd administered by the Ministry of Education and Culture, Finland is gratefully acknowledged. Our

497 study was supported by the Academy of Finland (grant nos. 129662, 134728 and 218343 to J.M.). Partial

498 support was also received from the NSFC/RGC Joint Research Scheme sponsored by the Research Grants

499 Council of the Hong Kong Special Administrative Region, China and the National Natural Science Foundation

500 of China (Project No. N_HKU763/21).

501
502 Ethical note

503 All experimental protocols were approved by permission (ESLHSTSTH223A) form the National Animal

504 Experiment Board, Finland.

505
506 Data availability

507 Raw sequence reads for the F2 crosses have been submitted to NCBIs short read archives with accession nos.:


508 PRJNA673430 and PRJNA672863. All data and code necessary to replicate the analyses presented in the

509 manuscript will been deposited in the Dryad repository and R scripts are currently available at:

510 https://github.com/afraimout/Relatedness/

21
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

511
512 Author contributions

513 A.F, J.M. Z.L M.S. and P.R. conceived the study; P.R. contributed to pre-processing and getting the genotype

514 data; A.F. analyzed the data; F.G. performed all computer simulations; J.M. and A.F. led the writing of the

515 manuscript. All authors contributed critically to the drafts and gave final approval for publication.

516 References
517 Ab Ghani, N. I., Herczeg, G., & Merilä, J. (2012). Body size divergence in nine-spined sticklebacks:
518 disentangling additive genetic and maternal effects. Biological Journal of the Linnean Society, 107(3),
519 521–528.
520 Ackerman, M. S., Johri, P., Spitze, K., Xu, S., Doak, T. G., Young, K., & Lynch, M. (2017). Estimating seven
521 coefficients of pairwise relatedness using population-genomic data. Genetics, 206(1), 105-118.
522 Andrews S. (2010). FastQC: a quality control tool for high throughput sequence data. Available online at:
523 http:/www.bioinformatics.babraham.ac.uk/projects/fastqc
524 Bell, M. A., Khalef, V., & Travis, M. P. (2007). Directional asymmetry of pelvic vestiges in threespine
525 stickleback. Journal of Experimental Zoology Part B: Molecular and Developmental Evolution, 308(2),
526 189–199.
527 Bell, M. A., Ortí, G., Walker, J. A., & Koenings, J. P. (1993). Evolution of pelvic reduction in threespine
528 stickleback fish: a test of competing hypotheses. Evolution, 47(3), 906–914.
529 Bennewitz, J., & Meuwissen, T. H. E. (2010). The distribution of QTL additive and dominance effects
530 in porcine F2 crosses. Journal of Animal Breeding and Genetics, 127(3), 171-179.
531 Bérénos, C., Ellis, P. A., Pilkington, J. G., & Pemberton, J. M. (2014). Estimating quantitative genetic
532 parameters in wild populations: a comparison of pedigree and genomic approaches. Molecular Ecology,
533 23(14), 3434–3451.
534 Blouw, D. M., & Boyd, G. J. (1992). Inheritance of reduction, loss, and asymmetry of the pelvis in Pungitius
535 pungitius (ninespine stickleback). Heredity, 68(1), 33–42.
536 Broman, K. W., Murray, J. C., Sheffield, V. C., White, R. L., & Weber, J. L. (1998). Comprehensive human
537 genetic maps: individual and sex-specific variation in recombination. American Journal of Human Genetics,
538 63(3), 861–869.
539 Bruneaux, M., Johnston, S. E., Herczeg, G., Merila, J., Primmer, C. R., & Vasemagi, A. (2013). Molecular
540 evolutionary and population genomic analysis of the nine-spined stickleback using a modified restriction-
541 site-associated DNA tag approach. Molecular Ecology, 22(3), 565–582.
542 Bürkner, P. C. (2017). brms: An R package for Bayesian multilevel models using Stan. Journal of statistical
543 software, 80(1), 1-28.
544 Chan, Y. F., Marks, M. E., Jones, F. C., Villarreal, G., Shapiro, M. D., Brady, S. D., Southwick, A. M., Absher,
545 D. M., Grimwood, J., & Schmutz, J. (2010). Adaptive evolution of pelvic reduction in sticklebacks by
546 recurrent deletion of a Pitx1 enhancer. Science, 327(5963), 302–305.
547 Charmantier, A., & Garant, D. (2005). Environmental quality and evolutionary potential: lessons from wild
548 populations. Proceedings of the Royal Society B: Biological Sciences, 272(1571), 1415-1425.
549 Charmantier, A., Garant, D., & Kruuk, L. E. B. (2014). Quantitative genetics in the wild. Oxford University
550 Publishing.
551 Class, B., & Brommer, J. E. (2020). Can dominance genetic variance be ignored in evolutionary quantitative
552 genetic analyses of wild populations? Evolution, 74(7), 1540–1550.
553 Colosimo, P. F., Hosemann, K. E., Balabhadra, S., Villarreal, G., Dickson, M., Grimwood, J., Schmutz, J.,
554 Myers, R. M., Schluter, D., & Kingsley, D. M. (2005). Widespread parallel evolution in sticklebacks by
555 repeated fixation of ectodysplasin alleles. Science, 307(5717), 1928–1933.

22
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

556 Coyle, S. M., Huntingford, F. A., & Peichel, C. L. (2007). Parallel evolution of Pitx1 underlies pelvic reduction
557 in Scottish threespine stickleback (Gasterosteus aculeatus). Journal of Heredity, 98(6), 581–586.
558 Crnokrak, P., & Roff, D. A. (1995). Dominance variance: associations with selection and fitness. Heredity, 75(5),
559 530–540.
560 De la Cruz, I. M., Merilä, J., Valverde, P. L., Flores Ortiz, C. M., & Núñez Farfán, J. (2020). Genomic and
   

561 chemical evidence for local adaptation in resistance to different herbivores in Datura
562 stramonium. Evolution, 74(12), 2629-2643.
563 P de Villemereuil (2018). Quantitative genetic methods depending on the nature of the phe-
564 notypic trait. Annals of the New York Academy of Sciences. The Year in Evolutionary Biology
565 1422, 29–47
566 Duntsch, L., Tomotani, B. M., de Villemereuil, P., Brekke, P., Lee, K. D., Ewen, J. G., & Santure, A. W. (2020).
567 Polygenic basis for adaptive morphological variation in a threatened Aotearoa| New Zealand bird, the hihi
568 (Notiomystis cincta). Proceedings of the Royal Society B, 287(1933).
569 Edwards, D., & Batley, J. (2010). Plant genome sequencing: applications for crop improvement. Plant
570 biotechnology Journal, 8(1), 2-9.
571 Falconer, D. S., & Mackay, T. F. C. (1996). Introduction into quantitative genetics. Essex: Prentice Hall.
572 Fang, B., Kemppainen, P., Momigliano, P., & Merilä, J. (2021). Population structure limits parallel evolution in
573 sticklebacks. Molecular Biology and Evolution.
574 Fisher, R. A. (1918). The correlation between relatives on the supposition of Mendelian inheritance.
575 Transactions of the Royal Society of Edinburgh, 52(2), 399–433.
576 Fisher, R. A. (1958). The Genetical Theory of Natural Selection. The Clarandon Press.
577 Gervais, L., Hewison, A. J., Morellet, N., Bernard, M., Merlet, J., Cargnelutti, B., et al. (2020). Pedigree free 

578 quantitative genetic approach provides evidence for heritability of movement tactics in wild roe
579 deer. Journal of Evolutionary Biology, 33(5), 595-607.
580 Gibson, G. (2005). The synthesis and evolution of a supermodel. Science, 307(5717), 1890–1891.
581 Gienapp, P., Fior, S., Guillaume, F., Lasky, J. R., Sork, V. L., Csilléry, K. (2017). Genomic quantitative genetics
582 to study evolution in the wild. Trends in Ecology & Evolution, 32(12), 897–908.
583 Gienapp, P., Calus, M. P., Laine, V. N., & Visser, M. E. (2019). Genomic selection on breeding time in a wild
584 bird population. Evolution Letters, 3(2), 142-151.
585 Granato, I. S. C., Galli, G., de Oliveira Couto, E. G., Bandeira e Souza, M., Mendonça, L. F., & Fritsche-Neto, R.
586 (2018). snpReady: a tool to assist breeders in genomic analysis. Molecular Breeding, 38(8), 1–7.
587 Guillaume, F., & Rougemont, J. (2006). Nemo: an evolutionary and population genetics programming
588 framework. Bioinformatics, 22(20), 2556-2557.
589 Hadfield, J. D. (2010). MCMC methods for multi-response generalized linear mixed models: the MCMCglmm R
590 package. Journal of Statistical Software, 33(2), 1–22.
591 Hayes, B., & Goddard, M. (2010). Genome-wide association and genomic selection in animal
592 breeding. Genome, 53(11), 876-883.
593 Herczeg, G, Gonda, A., & Merilä, J. (2009a). Evolution of gigantism in nine-spined sticklebacks. Evolution,
594 63(12), 3190–3200.
595 Herczeg, G, Gonda, A., & Merilä, J. (2009b). Predation mediated population divergence in complex behaviour
596 of nine-spined stickleback (Pungitius pungitius). Journal of Evolutionary Biology, 22(3), 544–552.
597 Herczeg, Gábor, Gonda, A., Kuparinen, A., & Merilä, J. (2012). Contrasting growth strategies of pond versus
598 marine populations of nine-spined stickleback (Pungitius pungitius): a combined effect of predation and
599 competition? Evolutionary Ecology, 26(1), 109–122.
600 Herczeg, Gábor, Turtiainen, M., & Merilä, J. (2010). Morphological divergence of North-European nine-spined
601 sticklebacks (Pungitius pungitius): signatures of parallel evolution. Biological Journal of the Linnean
602 Society, 101(2), 403–416.
603 Hill, W. G., Goddard, M. E., & Visscher, P. M. (2008). Data and theory point to mainly additive genetic
604 variance for complex traits. PLoS Genet, 4(2), e1000008.
605 Hoffmann, A. A., & Merilä, J. (1999). Heritable variation and evolution under favourable and unfavourable
606 conditions. Trends in Ecology & Evolution, 14(3), 96–101.

23
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

607 Jensen, H., Szulkin, M., & Slate, J. (2014). Molecular quantitative genetics. Quantitative genetics in the
608 wild, 209-227.
609 Joshi, R., Meuwissen, T. H., Woolliams, J. A., & Gjøen, H. M. (2020). Genomic dissection of maternal, additive
610 and non-additive genetic effects for growth and carcass traits in Nile tilapia. Genetics Selection
611 Evolution, 52(1), 1-13.
612 Karavolias, N. G., Greenberg, A. J., Barrero, L. S., Maron, L. G., Shi, Y., Monteverde, E., Piñeros, M. A., &
613 McCouch, S. R. (2020). Low additive genetic variation in a trait under selection in domesticated rice. G3:
614 Genes, Genomes, Genetics, 10(7), 2435–2443.
615 Karhunen, M., Ovaskainen, O., Herczeg, G., & Merilä, J. (2014). Bringing habitat information into statistical
616 tests of local adaptation in quantitative traits: A case study of nine-spined sticklebacks. Evolution, 68(2),
617 559–568.
618 Kemppainen, P., Li, Z., Rastas, P., Löytynoja, A., Fang, B., Yang, J., Guo, B., Shikano, T., & Merilä, J. (2021).
619 Genetic population structure constrains local adaptation in sticklebacks. Molecular Ecology. Accepted
620 Author Manuscript. https://doi.org/10.1111/mec.15808
621 Kemppainen, P., Li, Z., Rastas, P., Löytynoja, A., Fang, B., Yang, J., Guo, B., Shikano, T., & Merilä, J. (2021).
622 SNP and phenotype data for F2 crosses of Pungitius pungitius. https://doi.org/10.5061/dryad.76hdr7str
623
‐
Kivikoski, M., Rastas, P., Löytynoja, A., & Merilä, J. (2021). Automated improvement of stickleback reference
624 genome assemblies with Lep Anchor software. Molecular Ecology Resources, 21(6), 2166-2176.
625 https://doi.org/10.1111/1755-0998.13404
626 Kivikoski, M., Rastas, P., Löytynoja, A., & Merilä, J. (2023). Predicting recombination frequency from map
627 distance. Heredity, 130 (3), 114-121. https://doi.org/10.1038/s41437-022-00585-3
628 Koch EL, Sbilordo SH, and F Guillaume. (2020). Genetic variance in fitness and its cross-sex covariance predict
629 adaptation during experimental evolution. Evolution 74:2725-2740; doi:10.1111/evo.14119
630 Kosova, G., Abney, M., & Ober, C. (2010). Heritability of reproductive fitness traits in a human population.
631 Proceedings of the National Academy of Sciences USA, 107(suppl 1), 1772–1778.
632 Kruuk, L. (2004). Estimating genetic parameters in natural populations using the ‘animal model’. Philosophical
633 Transactions of the Royal Society of London. Series B: Biological Sciences, 359(1446), 873–890.
634 Laine, V. N., Shikano, T., Herczeg, G., Vilkki, J., & Merilä, J. (2013). Quantitative trait loci for growth and
635 body size in the nine spined stickleback Pungitius pungitius L. Molecular Ecology, 22(23), 5861–5876.


636 Legarra, A. (2016). Comparing estimates of genetic variance across different relationship models. Theoretical
637 Population Biology, 107, 26-30.
638 Leinonen, T., McCairns, R. S., O'hara, R. B., & Merilä, J. (2013). QST–FST comparisons: evolutionary and
639 ecological insights from genomic heterogeneity. Nature Reviews Genetics, 14(3), 179-190.
640 Li, Z., Guo, B., Yang, J., Herczeg, G., Gonda, A., Balázs, G., Shikano, T., Calboli, F. C. F., & Merilä, J. (2017).
641 Deciphering the genomic architecture of the stickleback brain with a novel multilocus gene mapping 

642 approach. Molecular Ecology, 26(6), 1557–1575.

643 Li, Z., Kemppainen, P., Rastas, P., & Merilä, J. (2018). Linkage disequilibrium clustering based approach for 

644 association mapping with tightly linked genomewide data. Molecular Ecology Resources, 18(4), 809–824.
645 Li, Z., Löytynoja, A., Fraimout, A., & Merilä, J. (2019). Effects of marker type and filtering criteria on QST-
646 FST comparisons. Royal Society Open Science, 6(11), 190666.
647 Lush, J. L. (1943). Animal Breeding Plans. Second Edition. Iowa State Press.
648 Lynch, M, & Walsh, B. (1998). Genetics and Analysis of Quantitative Traits. Sinauer.
649 Lynch, M, & Ritland, K. (1999). Estimation of pairwise relatedness with molecular markers. Genetics, 152(4),
650 1753–1766.
651 Malécot, G. (1948). Les Mathématiques de l'hérédité. Masson.
652 Manichaikul, A., Mychaleckyj, J. C., Rich, S. S., Daly, K., Sale, M., & Chen, W.-M. (2010). Robust relationship
653 inference in genome-wide association studies. Bioinformatics, 26(22), 2867–2873.
654 Manolio, T. A., Collins, F. S., Cox, N. J., Goldstein, D. B., Hindorff, L. A., Hunter, D. J., McCarthy, M. I.,
655 Ramos, E. M., Cardon, L. R., & Chakravarti, A. (2009). Finding the missing heritability of complex
656 diseases. Nature, 461(7265), 747–753.
657 Merilä, J. (2013). Nine-spined stickleback (Pungitius pungitius): an emerging model for evolutionary biology

24
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

658 research. Annals of the New York Academy of Sciences, 1289, 18–35.
659 Merilä, J., & Sheldon, B. C. (1999). Genetic architecture of fitness and nonfitness traits: empirical patterns and
660 development of ideas. Heredity, 83(2), 103–109.
661 Merilä, J., Söderman, F., O'hara, R., Räsänen, K., & Laurila, A. (2004). Local adaptation and genetics of acid-
662 stress tolerance in the moor frog, Rana arvalis. Conservation Genetics, 5(4), 513-527.
663 Mousseau, T. A., Ritland, K., & Heath, D. D. (1998). A novel method for estimating heritability using molecular
664 markers. Heredity, 80(2), 218-224.
665 Nilforooshan, M.A. mbend: an R package for bending non-positive-definite symmetric matrices to positive-
666 definite. BMC Genet 21, 97 (2020). https://doi.org/10.1186/s12863-020-00881-z
667 Nishio, M. & Satoh, M. (2014) Including dominance effects in the genomic BLUP method for genomic
668 evaluation. PLoS One. 2014; 9(1): e85792.
669 Ødegård, J., & Meuwissen, T. H. (2012). Estimation of heritability from limited family data using
670 genome-wide identity-by-descent sharing. Genetics Selection Evolution, 44(1), 1-10.
671 Pérez, P., & de Los Campos, G. (2014). Genome-wide regression and prediction with the BGLR
672 statistical package. Genetics, 198(2), 483-495.
673 Perrier, C., Delahaie, B., & Charmantier, A. (2018). Heritability estimates from genomewide relatedness
674 matrices in wild populations: Application to a passerine, using a small sample size. Molecular Ecology
675 Resources, 18(4), 838–853.
676 Plummer M, Best N, Cowles K, Vines K. (2006). CODA: Convergence diagnosis and output analysis for
677 MCMC. R News. 6(1), 7-11.
678 Powell, J. E., Visscher, P. M., & Goddard, M. E. (2010). Reconciling the analysis of IBD and IBS in complex
679 trait studies. Nature Reviews Genetics, 11(11), 800–805.
680 Price, T., & Schluter, D. (1991). On the low heritability of life history traits. Evolution, 45(4), 853–861.


681 Rastas, P., Calboli, F. C., Guo, B., Shikano, T., & Merilä, J. (2016). Construction of ultradense linkage maps
682 with Lep-MAP2: stickleback F 2 recombinant crosses as an example. Genome biology and evolution, 8(1),
683 78-93.
684 Risch, N., & Lange, K. (1979). Application of a recombination model in calculating the variance of sib pair
685 genetic identity. Annals of Human Genetics, 43(2), 177–186.
686 Ritland, K. (1996). A marker based method for inferences about quantitative inheritance in natural populations.


687 Evolution, 50(3), 1062–1073.

688 Ritland, K. (2000). Marker inferred relatedness as a tool for detecting heritability in nature. Molecular Ecology,


689 9(9), 1195–1204.

690 Robinson, M. R., Santure, A. W., DeCauwer, I., Sheldon, B. C., & Slate, J. (2013). Partitioning of genetic
691 variation across the genome using multimarker methods in a wild bird population. Molecular Ecology,
692 22(15), 3963–3980.
693 Roff, D. A. (1997). Heritability. In Evolutionary quantitative genetics (pp. 24–72). Springer.
694 Roff, D. A., & Emerson, K. (2006). Epistasis and dominance: evidence for differential effects in life history 

695 versus morphological traits. Evolution, 60(10), 1981-1990.

696 Rohlf, F. J. (2006). tpsDig, version 2.10. http://Life. Bio. Sunysb. Edu/Morph/Index. Html.
697 Sargolzaei, M., & Schenkel, F. S. (2009). QMSim: a large-scale genome simulator for livestock.
698 Bioinformatics, 25(5), 680-681.
699 Shikano, T, Shimada, Y., Herczeg, G., & Merila, J. (2010). History vs. habitat type: explaining the genetic
700 structure of European nine-spined stickleback (Pungitius pungitius) populations. Molecular Ecology, 19(6),
701 1147–1161.
702 Shikano, Takahito, Herczeg, G., & Merilä, J. (2011). Molecular sexing and population genetic inference using a
703 sex-linked microsatellite marker in the nine-spined stickleback (Pungitius pungitius). BMC Research Notes,
704 4(1), 1–6.
705 Shikano, Takahito, Laine, V. N., Herczeg, G., Vilkki, J., & Merilä, J. (2013). Genetic architecture of parallel
706 pelvic reduction in ninespine sticklebacks. G3: Genes, Genomes, Genetics, 3(10), 1833–1842.
707 Shimada, Y., Shikano, T., Kuparinen, A., Gonda, A., Leinonen, T., & Merilä, J. (2011). Quantitative genetics of

25
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

708 body size and timing of maturation in two nine-spined stickleback (Pungitius pungitius) populations. PLoS
709 One, 6(12), e28859.
710 Sillanpää, M. J. (2011). On statistical methods for estimating heritability in wild populations. Molecular Ecology,
711 20(7), 1324–1332.
712 Sinclair-Waters, M., Ødegård, J., Korsvoll, S. A., Moen, T., Lien, S., Primmer, C. R., & Barson, N. J. (2020).
713 Beyond large-effect loci: large-scale GWAS reveals a mixed large-effect and polygenic architecture for age
714 at maturity of Atlantic salmon. Genetics Selection Evolution, 52(1), 1-11.
715 Sztepanacz, J. L., & Blows, M. W. (2015). Dominance genetic variance for traits under directional selection in
716 Drosophila serrata. Genetics, 200(1), 371-384.
717 Speed, D., & Balding, D. J. (2015). Relatedness in the post-genomic era: is it still useful? Nature Reviews
718 Genetics, 16(1), 33–44.
719 Teacher, A. G. F., Shikano, T., Karjalainen, M. E., & Merila, J. (2011). Phylogeography and genetic structuring
720 of European nine-spined sticklebacks (Pungitius pungitius)-Mitochondrial DNA Evidence. Plos One, 6(5).
721 Tenesa, A., & Haley, C. S. (2013). The heritability of human disease: estimation, uses and abuses. Nature
722 Reviews Genetics, 14(2), 139.
723 Thompson, E. A. (2013). Identity by descent: variation in meiosis, across genomes, and in populations. Genetics,
724 194(2), 301–326.
725 VanRaden, P.M. (2008). Efficient methods to compute genomic predictions. Journal of Dairy Science, 91(11):
726 4414–23.
727 Varadharajan, S., Rastas, P., Löytynoja, A., Matschiner, M., Calboli, F. C. F., Guo, B., Nederbragt, A. J.,
728 Jakobsen, K. S., & Merilä, J. (2019). Genome sequencing of the nine-spined stickleback (Pungitius
729 pungitius) provides insights into chromosome evolution. bioRxiv. https://doi.org/10.1101/741751
730 Vendrami, D. L., Peck, L. S., Clark, M. S., Eldon, B., Meredith, M., & Hoffman, J. I. (2021). Sweepstake
731 reproductive success and collective dispersal produce chaotic genetic patchiness in a broadcast spawner.
732 Science advances, 7(37), eabj4713.
733 Visscher, P. M. (2009). Whole genome approaches to quantitative genetics. Genetica, 136(2), 351.
734 Visscher, P. M., Macgregor, S., Benyamin, B., Zhu, G., Gordon, S., Medland, S., Hill, W. G., Hottenga, J.-J.,
735 Willemsen, G., & Boomsma, D. I. (2007). Genome partitioning of genetic variation for height from 11,214
736 sibling pairs. The American Journal of Human Genetics, 81(5), 1104–1110.
737 Visscher, P. M., Medland, S. E., Ferreira, M. A. R., Morley, K. I., Zhu, G., Cornes, B. K., Montgomery, G. W.,
738 & Martin, N. G. (2006). Assumption-free estimation of heritability from genome-wide identity-by-descent
739 sharing between full siblings. PLoS Genetics, 2(3), e41.
740 Walsh, B., & Lynch, M. (2018). Evolution and Selection of Quantitative Traits. Oxford University Press.
741 Weir, B. S., & Goudet, J. (2017). A unified characterization of population structure and relatedness. Genetics,
742 206(4), 2085-2103.
743 Wilson, A. J., Reale, D., Clements, M. N., Morrissey, M. M., Postma, E., Walling, C. A., Kruuk, L. E. B., &
744 Nussey, D. H. (2010). An ecologist’s guide to the animal model. Journal of Animal Ecology, 79(1), 13–26.
745 Wolak, M. E. (2012). nadiv: an R package to create relatedness matrices for estimating non additive genetic 

746 variances in animal models. Methods in Ecology and Evolution, 3(5), 792–796.
747 Wolak, M. E., & Keller, L. F. (2014). Dominance genetic variance and inbreeding in natural populations.
748 Quantitative Genetics in the Wild, 104–127.
749 Wootton, R. J. (1976). Biology of the Sticklebacks. Academic Press.
750 Wright, S. (1931). Evolution in Mendelian populations. Genetics, 16(2), 97.
751 Xiang, T., Christensen, O. F., Vitezica, Z. G., & Legarra, A. (2018). Genomic model with correlation
752 between additive and dominance effects. Genetics, 209(3), 711-723.
753 Xie, K. T., Wang, G., Thompson, A. C., Wucherpfennig, J. I., Reimchen, T. E., MacColl, A. D. C., Schluter, D.,
754 Bell, M. A., Vasquez, K. M., & Kingsley, D. M. (2019). DNA fragility in the parallel evolution of pelvic
755 reduction in stickleback fish. Science, 363(6422), 81–84.
756 Xu, S. (2006). Population genetics: separating nurture from nature in estimating heritability. Heredity, 97, 256–
757 257.
758 Yang, J., Lee, S. H., Goddard, M. E., & Visscher, P. M. (2011). GCTA: a tool for genome-wide complex trait

26
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

759 analysis. American Journal of Human Genetics, 88(1), 76–82.

760 Zhao, J. H., Luan, J., & Congdon, P. (2018). Bayesian linear mixed models with polygenic effects. Journal of
761 Statistical Software, 85(6), 1–27
762 Zhu, Z., Bakshi, A., Vinkhuyzen, A. A. E., Hemani, G., Lee, S. H., Nolte, I. M., van Vliet-Ostaptchouk, J. V,
763 Snieder, H., Esko, T., & Milani, L. (2015). Dominance genetic variation contributes little to the missing
764 heritability for human complex traits. American Journal of Human Genetics, 96(3), 377–385.
765

27
bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

Figures

Figure 1. Quantitative genetic parameters estimated from the empirical datasets. The narrow-
sense heritability (h²), additive genetic variance (VA), dominance variance (VD), phenotypic variance
(VP) and residual variance (VR) estimated for the empirical datasets are shown. For each dataset and
each variance component, the median of the posterior distribution is shown. The 95% HPD interval is
represented by vertical solid lines. For each trait (SL: standard length; BD: body depth; PL: pelvic
length) each variance component is shown as estimated from the half-sib/full-sib design (HS, filled
circle), the full-sib design (FS, filled square) and from the wild population sample (Wild, filled
triangle).
bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

Figure 2. Quantitative genetic parameters estimated from the simulated datasets. The narrow-
sense heritability (h²) and additive genetic variance (VA) estimated for the simulated datasets are
shown. For each dataset and each variance component, the mean estimate over all replicates is shown
(filled circle) along with the 95% HPD interval (vertical bars) for the true simulated values (blue) and
the estimates recovered from the MCMCglmm models (red). Codes for the datasets on the x-axis are
explained in Methods.
bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

Figure 3. Effect of migration rate on the quantitative genetic parameters estimated from the
wild simulated datasets. The narrow-sense heritability (h²) and additive genetic variance (VA)
estimated for the wild simulated datasets are shown. For each dataset and each variance component,
the mean estimate over all replicates is shown (filled circle) along with the standard deviation (vertical
bars) for the true simulated values (blue) and the estimates recovered from the MCMCglmm models
(red). Codes for the datasets on the x-axis correspond to the type of metapopulation used (P05, P10; 5
or 10 subpopulations, see Methods) and the level of migration rate among subpopulations (M).
bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

Figure 4. Chromosome partitioning of the genetic variance (Helsinki). The phenotypic variance in
body depth (BD), pelvic spine length (PL) and standard length (SL) explained by the SNPs are shown
of each chromosome for the Helsinki crosses. Black stars correspond to non-zero estimates (+/-SE) for
each chromosome.
bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

Figure 5. Chromosome partitioning of the genetic variance (QTL crosses). The phenotypic
variance in body depth (BD; left column), pelvic spine length (PL; center column) and standard length
(SL; right column) explained by the SNPs are shown of each chromosome for the HB cross (red; top
row), HP cross (green; middle row) and HR (blue; bottom row). Black stars correspond to non-zero
estimates (+/-SE) for each chromosome.
bioRxiv preprint doi: https://doi.org/10.1101/2021.03.01.432833; this version posted May 24, 2023. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
available under aCC-BY-NC-ND 4.0 International license.

Figure. 6. QTL-mapping results for the three phenotypic traits. Results from the four-way QTL-
mapping are shown for each cross (A, B, C) and for each trait (SL: standard length; BD: body depth;
PL: pelvic spine length). For each cross and trait, panels show whether the QTL is inherited from the
sire (M) or the dam (F) along with the dominance effect (D) estimated from model [5] in the main text.
Results are based on permutation and the significance threshold (dashed horizontal line) is shown on
the logarithm scale (p = 0.05). Colors represent different chromosomes.