https://orcid.org/0000-0003-0564-7068
Figures
Abstract
Random regression models (RRM) are a powerful tool to evaluate genotypic plasticity over time. However, to date, RRM remains unexplored for the analysis of repeated measures in Jatropha curcas breeding. Thus, the present work aimed to apply the random regression technique and study its possibilities for the analysis of repeated measures in Jatropha curcas breeding. To this end, the grain yield (GY) trait of 730 individuals of 73 half-sib families was evaluated over six years. Variance components were estimated by restricted maximum likelihood, genetic values were predicted by best linear unbiased prediction and RRM were fitted through Legendre polynomials. The best RRM was selected by Bayesian information criterion. According to the likelihood ratio test, there was genetic variability among the Jatropha curcas progenies; also, the plot and permanent environmental effects were statistically significant. The variance components and heritability estimates increased over time. Non-uniform trajectories were estimated for each progeny throughout the measures, and the area under the trajectories distinguished the progenies with higher performance. High accuracies were found for GY in all harvests, which indicates the high reliability of the results. Moderate to strong genetic correlation was observed across pairs of harvests. The genetic trajectories indicated the existence of genotype × measurement interaction, once the trajectories crossed, which implies a different ranking in each year. Our results suggest that RRM can be efficiently applied for genetic selection in Jatropha curcas breeding programs.
Citation: Peixoto MA, Alves RS, Coelho IF, Evangelista JSPC, de Resende MDV, Rocha JRdASdC, et al. (2020) Random regression for modeling yield genetic trajectories in Jatropha curcas breeding. PLoS ONE 15(12): e0244021. https://doi.org/10.1371/journal.pone.0244021
Editor: Evangelia V. Avramidou, Institute of Mediterranean Forest Ecosystems of Athens, GREECE
Received: August 14, 2020; Accepted: December 1, 2020; Published: December 23, 2020
Copyright: © 2020 Peixoto et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the manuscript and its Supporting Information files.
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Jatropha curcas L. ranks among the most relevant crops for biofuel production [1,2]. Its high adaptability, which allows its cultivation under different environmental conditions, tolerance to drought, longevity and high oil quality are some desirable characteristics which highlight Jatropha curcas as an excellent alternative for renewable energy production [1,3,4].
It has a high oil content in its seeds (up to 35%), with a high oil-to-biofuel conversion efficiency, compared to other species [4,5]. After extracted, the oil is composed by 47% crude fat and 25% crude protein; and, when compared to other vegetable oils, it presents better oxidation stability and lower viscosity [6].
Jatropha curcas is a perennial plant that produces during several years, and in its breeding process, genetic selection has been carried out considering only one harvest [7], several harvests independently [8] or several harvests simultaneously [9,10].
It is worth to highlight some specific issues in perennial plant breeding that affect genetic selection. Once repeated measures are taken in the same individuals over time, analyses must consider: the permanent environmental effects and genetic and residual covariance among harvests. Thus, for quantitative traits, genetic selection must consider several harvests to maximize the selective accuracy [11].
In this context, random regression models (RRM) can be a very efficient alternative for repeated measures analyses in Jatropha curcas breeding because they consider the genetic and residual covariance among harvests and allows fitting the genetic and non-genetic trajectories (plot and permanent environment) over time [11,12]. Besides that, RRM allow assessing genotype persistence, i.e., the capacity to maintain the yield performance over the years, which can be affected by biotic and abiotic effects [13–15]. Thus, this work aimed to apply the random regression technique and study the possibilities it offers for the analysis of repeated measures in Jatropha curcas breeding.
Material and methods
Experimental data
Seven hundred and thirty individuals from 73 half-sib families of Jatropha curcas were evaluated for grain yield (GY) trait (kg plant-1), in six harvests (2010 to 2015) (Supplementary material–S3 Table). The experiment was carried out in a randomized complete block design with two replications, five plants per plot, and spacing of 4 m between rows and 2 m between trees. The experiment was conducted in the experimental field of Embrapa Cerrados, located in Planaltina, Distrito Federal—Brazil (15°35’30” S and 47°42’30” W; 1007 m asl). All management practices were based on [16].
Statistical analyses
The time of the harvests must be scaled to range from -1 to +1 in order to use Legendre polynomials. The scaling formula is given below [12]:
where hx refers to the time of the harvest x; hmin is the time of the first harvest (2010); and, hmax is time of the last harvest (2015).
Variance components were estimated by restricted maximum likelihood [17], and genetic values were predicted by best linear unbiased prediction [18], according to [19]. Random regression models were fitted through Legendre polynomials, considering all possible degrees of fit for each random effect, using the following general model:
where Yijkl is the ith individual (i = 1, 2, …, 730) on the jth harvest (j = 1, 2, …, 6) on the kth replication (k = 1, 2) on the lth plot (l = 1, 2, …, 146); Rk is the fixed effect of replication;. bM is the fixed regression coefficient fitted through the fifth degree of Legendre polynomials to the common average trajectory of progenies. The random effects, gim, pikm, and sik are the random regression coefficients for the Legendre polynomials of degree m for the genetic, permanent environment, and plot effects, respectively. ϕijm is the mth Legendre polynomial for the jth harvest from the ith individual; m is the fit degree, ranging from M = 0 to M = 5, of the Legendre polynomial for the genetic, permanent environment, and plot effects, respectively; and εijkl is the residual random effect associated with Yijkl.
In the matrix notation, the above model is described as follows:
where y is the phenotypic data vector; b is the vector of the effects of measurement-replication combinations (assumed to be fixed) added to the overall mean; g is the vector of the individual genetic effects (assumed as random); p is the vector of the permanent environmental effects (assumed as random); s is the vector of the plot effects (assumed as random); and e is the vector of residuals (random). X, Z, W and Q refer to the incidence matrices for these effects.
In this model, g~N(0, Kg⊗I), p~N(0, Kp⊗I), s~N(0, Ks⊗I), and e~N(0, R); where Kg, Kp, and Ks are the covariance matrices for genetic, permanent environment, and plot effects, respectively; ⊗ denotes the Kronecker product; I is an identity matrix with appropriate order to the respective random effect; and R refers to the matrix of residual covariances. Different residual covariance structures (homogeneous, diagonal, and unstructured) were tested.
The polynomial order in random regression models was selected using the Bayesian information criterion (BIC) [20], as follows:
where LogL is the logarithm of the maximum (L) of the residual likelihood function, p is the number of estimated parameters, n is the number of observations, and r(x) is the rank of the incidence matrix of fixed effect.
The significance of the genetic, permanent environment and plot effects was tested using the likelihood ratio test (LRT) [21], as follows:
where LogLR is the logarithm of the maximum (LR) of the residual likelihood function of the reduced model (without genetic or permanent environmental or plot effects).
Variance components estimates () and the predicted genetic values (
), on the original scale, were obtained by the following expressions [22]:
where x refers to the genetic or permanent environmental or plot covariance matrices.
Phenotypic variance (), individual heritability between progenies (
) and selective accuracy (
) were obtained by the following expressions [23]:
where
is the residual variance and PEV is the prediction error variance extracted from the diagonal of generalized inverse of the coefficient matrix of the mixed model equations.
The eigenfunction (Ψf) of the genetic coefficient covariance matrix, aiming to evaluated the genotype x measurement interaction, was calculated by the following expression [22]:
where
is the mth element of the fth eigenvector of Kg, and Φm is the normalized value of the mth Legendre polynomial.
The areas under the genetic trajectories (A), aiming to rank the clones, were obtained by the following expression [24,25]:
where μ is the phenotypic mean and where xm is the is the harvest scaled. Genetic correlations (ρg) between each pair of harvests were obtained by the following expression:
where
, is the genetic covariance between progenies for the pair of harvests i and j;
and
are the genetic variance between progenies for the harvests i and j, respectively.
Statistical analyses were performed using the ASReml 4.1 [19] and R [26] software.
Results
According to the BIC, the best RRM was the one of order two for the genetic effects, order five for the plot effects, and order one for the permanent environment effects; with diagonal residual variance structure (Table 1 and Supplementary material—S1 Table). Thus, this model was adopted to estimate the variance components and predict the genetic values. When the models without genetic, plot or permanent environmental effects were tested by the LRT, the significance of genetic, plot, and permanent environmental effects (p-value < 0.01) was detected (LRT equal to -43.19, -1450.28, and -361.98, respectively).
The genetic variance was not constant through the harvests (Table 2). The estimates increased from the first harvest (0.0034) to the last harvest (0.3232). Similar patterns were found for permanent environmental variance (0.0035 to 0.1064) and plot variance (0.0042 to 0.2165). In addition, the residual variance rate in the first harvest was 0.0020. After a steady increase, it reached 0.2154 in the last harvest. Consequently, the phenotypic variance presented a steady increase from the first harvest (0.0133) to the sixth harvest (0.8616). In addition, the individual heritability estimates ranged from 0.05 (2012) to 0.37 (2015) (Table 1). The selective accuracy presented high magnitudes in all harvests, and demonstrated an upward trend over time, ranging from 0.79 in the first harvest to 0.86 in the last harvest (Table 2).
The genetic trajectories exhibited a non-linear form and genotypic plasticity for the 73 half-sib Jatropha curcas progenies (Fig 1). They presented a continuous deviation in the first three harvests and then, different forms of deviation until the sixth harvest.
The first eigenfunction presented a concave crescent trajectory and accounted for more than 99% of the genetic variance (eigenvalue = 0.1460) (Fig 2). The second and third eigenfunctions explain only 0.6% (eigenvalue = 0.0008) and 0.4% (eigenvalue = 0.0005) of the genetic variability, respectively. The trajectory of the second eigenfunction was continuously over the third harvest and then decreased, whereas the third eigenfunction showed a concave deviation, with decreasing values until the third harvest and rising values until the last harvest (Fig 2).
Their proportional eigenvalues for the genetic covariance function are in parentheses.
The area under the genetic trajectories was calculated for genotype ranking, and the highest values represent those progenies with the best overall performance over time. In this case, different values of areas were found for the 73 half-sib Jatropha curcas progenies. They ranged from -0.5879 to 2.0520, and the top ten families (those with larger area under the genetic trajectories, from the first to the tenth families selected) were: 6, 70, 48, 16, 10, 1, 34, 39, 15, 29, and 54. The complete rank are presented in the supplementary material (S2 Table). In addition, the genetic correlations between pairs of harvests presented moderate magnitudes (0.33 < ρg < 0.66) in the 2010–2012, 2010–2013, 2010–2014, 2010–2015; high magnitudes (0.67 < ρg < 0.89) in the 2010–2011, 2011–2014, and 2011–2015; and very high magnitudes (ρg > 0.90) in the remaining pairs of harvests (Fig 3).
Discussion
Among the various criteria for selection of models, the BIC is prominent, because it is a consistent criterion. The selected model fit diagonal residual variance structure (i.e., one residual variance for each harvest). Considering the other random effects of the model (genetic, permanent, and plot), a total of 36 covariance components were estimated. The RRM are equivalent to the covariance functions and can be considered a reduced and simplified multiple-trait model, which allows the same parameters of interest (heritability and genetic correlation among all pairs of harvests) to be obtained, but with lower parameterization and with less computational effort [23,27].
Since there are reliable estimates of variance components, they allow the prediction of genetic values of individuals evaluated at different ages (and with different numbers of ages evaluated) and the projection of these genetic values for a common age, for ordering and selection purposes. Besides that, Legendre polynomials have been used to model growth curves in perennial plant breeding [14,28].
According to the LRT, there was genetic variability among the Jatropha curcas families. Besides that, the plot and permanent environmental effects are statistically significant (p-value < 0.01), i.e., they differ from zero. The significance of genetic effects shows the potential of this population and allow the selection of superior families, even with a restricted genetic basis explored in Jatropha curcas breeding in Brazil [6,29]. It is worth mentioning that the RRM allows fitting the permanent environmental effect. On the other hand, the multiple-trait model does not allow fitting this effect, since it assumes that the same trait in different harvest is a different trait [10].
The RRM is mainly focused on visualizing trajectories and covariance functions over time. Therefore, the variance components and genetic parameters can vary over the harvests. In this study, it is generally observed an increasing trend in the variance components and genetic parameters over time. Such pattern was also observed by [1,10,30].
The individual heritability estimates decreased from the first to the third harvest. Then, there was an upward trend until the sixth harvest, which indicates that the third harvest is more affected by the environmental conditions. According to [31], the heritabilities are classified as low (, except for the last one, which was classified as moderate (
). Similar results were reported in other studies with Jatropha curcas [10,30]. Besides that, the variation in the individual heritability estimates was like the variation in the variance components estimates.
Upward trend in genetic variances and heritabilities along the time, are biologically consistent. Such trends can be related to the fact that Jatropha curcas genotypes are more sensitive to environmental variations in the early stages of growth [1,30], due to the higher genotype × measurement interaction. Indeed, in this work, the genotype × measurement interaction effect was significant. Further, it is related with the fact that the metabolism of young perennials often privilege vegetative rather than reproductive growth [32], which leads to an uneven production in the early harvests. Then, given the temporal trend of the genetic parameters and genetic values, the selection for the GY trait should consider several harvests (three to six, according to [11] for an accurate genetic selection).
According to [33], the selective accuracies were classified as high in all harvests (0.70 < < 0.90). In fact, when compared with other models, including the multiple-trait models, the selective accuracy by RRM has presented values considered higher [34]. This is probably because RRM does not require pre-adjustment of weights at standard ages, which may provide gains in selective accuracy [35].
The genetic trajectories describe the genetic values of each family over time and encompass the six harvests evaluated in this study. The RRM can predict the genetic value for any family in any time (between the first and sixth harvest). The trajectories demonstrated that the families presented similar performance in the first harvests, which reveals that a precocious selection tends to be less efficient. The genetic correlations reinforce the inefficiency in earlier selection. The genetic correlations between the pairs of harvest presented very high values (> 0.90) only between the last harvests. In addition [9], showed that ten harvests are necessary for an accurate genetic selection for the GY trait in Jatropha curcas.
The RRM fitted through Legendre polynomials allows obtain the eigenfunctions and eigenvalues [22]. According to those authors, the eigenfunctions are similar to the eigenvectors of the principal component analyses and can be interpreted as proportional to the amount of genetic variation in the population corresponding to that eigenfunction. The first eigenfunction clustered general adaptability genes equally expressed in all harvests [14]. This can be interpreted as the genetic correlation among the harvests. The second and third eigenfunctions showed small eigenvalues and represent deformations for which there is little (or no) genetic variation [22]. Perennial plants usually present great variations in productivity in the initial harvests, since many genes expressed in that period are associated with the formation of vegetative organs [32]. This fact is typical of perennials and is indicated to occur in Jatropha curcas [30].
The genetic trajectories of the 73 Jatropha curcas progenies reinforced the presence of genotype × measurement interaction, once their trajectories are non-linear and intersect each other, which implies a different ranking in each harvest. In addition, the trajectories could also be interpreted as genetic variability. The more distant the genetic trajectories from each other, the more genetically distinct are the progenies [23].
In this study, genotype ranking was performed based on the areas under the genetic trajectories. Therefore, genetic selection was based on the higher area under the genetic trajectories. The advantage of this strategy is that selection response can be predicted not only in the genotypic expression in any harvest but also in quantifying the environmental sensitivity of the trait through the genetic trajectories (robustness or responsiveness to changes in the environment). Besides that, it can be used for any number of harvests [36].
The RRM can be used to help describing the observed phenotypic over time efficiently and allows genetic selection based on adaptability, stability and yield performance [14,36–38]. The main advantage of the RRM is the fact that it is biologically realistic through their emphasis on dynamic aspects of the phenotype and for allowing breeding questions on plasticity, adaptability, stability and yield performance to be answered [36]. Thus, our results suggest that RRM fitted through Legendre polynomials can be efficiently used in Jatropha curcas breeding programs.
Supporting information
S1 Table. ASReml output for the models that converged for the grain yield trait evaluated in 73 half-sib Jatropha curcas progenies.
P.D. Gen: polynomial degree for the genetic effect; P.D. Plot: polynomial degree for the plot effect; P.D. Perm: polynomial degree for the permanent environment effect; DF: degrees of freedom; LogL: logarithm of the maximum of the restricted likelihood function; Gen. P.: total number of genetic covariances estimated; Plot P.: total number of plot covariances estimated; Perm. P.: total number of permanent environment covariance estimated; Res. P: total number of residual covariances estimated; and BIC: Bayesian information criterion. The selected model by BIC was indicated in bold.
https://doi.org/10.1371/journal.pone.0244021.s001
(DOCX)
S2 Table. Ranking of the 73 half-sib Jatropha curcas progenies for the grain yield trait based on the areas under the trajectories (A).
https://doi.org/10.1371/journal.pone.0244021.s002
(DOCX)
S3 Table. Data of the 73 half-sib Jatropha curcas progenies for the grain yield trait.
https://doi.org/10.1371/journal.pone.0244021.s003
(DOCX)
References
- 1. Laviola BG, Rodrigues EV. Pinhão-manso: pesquisas, conhecimentos e práticas. Embrapa Agroenergia-Livro técnico (INFOTECA-E). 2019;1: 420.
- 2. Peixoto LA, Laviola BG, Bhering LL, Mendonça S, Costa TSA, Antoniassi R. Oil content increase and toxicity reduction in jatropha seeds through family selection. Ind Crops Prod. 2016;80: 70–76.
- 3. Nithiyanantham S, Siddhuraju P, Francis G. Potential of Jatropha curcas as a Biofuel, Animal Feed and Health Products. J Am Oil Chem Soc. 2012;89: 961–972.
- 4. Pu Y, Treasure T, Gonzalez RW, Venditti R, Jameel H. Autohydrolysis pretreatment of mixed hardwoods to extract value prior to combustion. BioResources. 2011;6: 4856–4870.
- 5. Berchmans HJ, Hirata S. Biodiesel production from crude Jatropha curcas L. seed oil with a high content of free fatty acids. Bioresour Technol. 2008;99: 1716–1721. pmid:17531473
- 6. Laviola BG, Alves AA, Rosado TB, Bhering LL, Formighieri EF, Peixoto L de A. Establishment of new strategies to quantify and increase the variability in the Brazilian Jatropha genotypes. Ind Crops Prod. 2018;117: 216–223.
- 7. Junqueira VS, Peixoto LA, Galvêas Laviola B, Lopes Bhering L, Mendonça S, Costa TSA, et al. Bayesian multi-trait analysis reveals a useful tool to increase oil concentration and to decrease toxicity in Jatropha curcas L. PLoS One. 2016;11: 1–14. pmid:27281340
- 8. Bhering LL, Barrera CF, Ortega D, Laviola BG, Alves AA, Rosado TB, et al. Differential response of Jatropha genotypes to different selection methods indicates that combined selection is more suited than other methods for rapid improvement of the species. Ind Crops Prod. 2013;41: 260–265.
- 9. Alves RS, Peixoto LA, Teodoro PE, Silva LA, Rodrigues EV, Resende MDV, et al. Selection of Jatropha curcas families based on temporal stability and adaptability of genetic values. Ind Crops Prod. 2018;119: 290–293.
- 10. Alves RS, Teodoro PE, Peixoto LA, Rocha JR, Silva LA, Laviola B, et al. Multiple-trait BLUP in repeated measures analysis on Jatropha curcas breeding for bioenergy. Ind Crops Prod. 2019;130: 558–561.
- 11.
Resende MDV. Genética biométrica e estatística no melhoramento de plantas perenes. 1st ed. Embrapa, editor. Colombo; 2002.
- 12. Schaeffer LR. Random Regression Models. 2016 p. 171. Available: http://animalbiosciences.uoguelph.ca/~lrs/BOOKS/rrmbook.pdf.
- 13. Riday H, Brummer EC. Persistence and Yield Stability of Intersubspecific Alfalfa Hybrids. Crop Sci. 2006;46: 1058–1063.
- 14. Rocha JRAS, Marçal T, Salvador FV, Silva AC, Machado JC, Carneiro PCS. Genetic insights into elephantgrass persistence for bioenergy purpose. PLoS One. 2018;13: 1–16. pmid:30212554
- 15. Wilkins PW, Humphreys MO. Progress in breeding perennial forage grasses for temperate agriculture. J Agric Sci. 2003;140: 129–150.
- 16.
Carels N. Towards the domestication of Jatropha: the integration of sciences. Jatropha, challenges for a new energy crop. Springer; 2013. pp. 263–299.
- 17. Patterson HD, Thompson R. Recovery of inter-block information when block sizes are unequal. Biometrika. 1971;58: 545–554.
- 18. Henderson CR. Best linear unbiased estimation and prediction under a selection model. Biometrics. 1975; 423–447. pmid:1174616
- 19.
Gilmour AR, Gogel BJ, Cullis BR, Welham Sj, Thompson R. ASReml user guide release 4.1 structural specification. Hemel hempstead VSN Int ltd. 2015.
- 20. Schwarz G. Estimating the dimension of a model. Ann Stat. 1978;6: 461–464.
- 21. Wilks SS. The Large-Sample Distribution of the Likelihood Ratio for Testing Composite Hypotheses. Ann Math Stat. 1938;9: 60–62.
- 22. Kirkpatrick M, Lofsvold D, Bulmer M. Analysis of the inheritance, selection and evolution of growth trajectories. Genetics. 1990;124: 979–993. pmid:2323560
- 23.
Resende MDV, Silva FF, Azevedo CF. Estatística matemática, biométrica e computacional: modelos mistos, multivariados, categóricos e generalizados (REML/BLUP), inferência bayesiana, regressão aleatória, seleção genômica, QTL-QWAS, estatística espacial e temporal, competição, sobrevivência. Viçosa: UFV; 2014.
- 24. Alves RS, Resende MDV, Carvalho JRASC, Peixoto MA, et al. Quantifying individual variation in reaction norms using random regression models fitted through Legendre polynomials: application in eucalyptus breeding. Bragantia. 2020.
- 25. Peixoto MA, Coelho IF, Evangelista JSPC, Alves RS, Rocha JRASC, Farias FJC, et al. Reaction norms‐based approach applied to optimizing recommendations of cotton genotypes. Agron J. 2020; agj2.20433. pmid:33303994
- 26.
R Development Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2020. Available: https://www.r-project.org/.
- 27. Resende MDV, Furlani-Júnior E, Moraes MLT, Fazuoli LC. Estimativas de parâmetros genéticos e predição de valores genotípicos no melhoramento do cafeeiro pelo procedimento REML/BLUP. 2001.
- 28. Marchal A, Schlichting CD, Gobin R, Balandier P, Millier F, Muñoz F, et al. Genotype by environment interactions in forest tree breeding: review of methodology and perspectives on research and application. Plant Genome. 2019;13: 281–288.
- 29. Rosado TB, Laviola BG, Faria DA, Pappas MR, Bhering LL, Quirino B, et al. Molecular markers reveal limited genetic diversity in a large germplasm collection of the biofuel crop Jatropha curcas L. in Brazil. Crop Sci. 2010;50: 2372–2382.
- 30. Laviola BG, Oliveira AMC e, Bhering LL, Alves AA, Rocha RB, Gomes BEL, et al. Estimates of repeatability coefficients and selection gains in Jatropha indicate that higher cumulative genetic gains can be obtained by relaxing the degree of certainty in predicting the best families. Ind Crops Prod. 2013;51: 70–76.
- 31.
Resende MDV. Genética quantitativa e de populações. Suprema, Visconde do Rio Branco. 1st ed. 2015; 463.
- 32. Iglesias DJ, Cercós M, Colmenero-Flores JM, Naranjo MA, Ríos G, Carrera E, et al. Physiology of citrus fruiting. Brazilian J Plant Physiol. 2007;19: 333–362.
- 33. Resende MDV, Duarte JB. Precisão e controle de qualidade em experimentos de avaliação de cultivares. Pesqui Agropecuária Trop. 2007;37: 182–194.
- 34. Mota RR, Tempelman RJ, Lopes PS, Aguilar I, Silva FF, Cardoso FF. Genotype by environment interaction for tick resistance of Hereford and Braford beef cattle using reaction norm models. Genet Sel Evol. 2016;48: 3. pmid:26767704
- 35. Meyer K. Scope for a random regression model in genetic evaluation of beef cattle for growth. Livest Prod Sci. 2004;86: 69–83.
- 36. Alves RS, Resende MDV, Azevedo CF, Silva FF, Rocha JRASC, Nunes ACP, et al. Optimization of Eucalyptus breeding through random regression models allowing for reaction norms in response to environmental gradients. Tree Genet Genomes. 2020;16: 38.
- 37. Morrissey MB, Liefting M. Variation in reaction norms: Statistical considerations and biological interpretation. Evolution (N Y). 2016;70: 1944–1959. pmid:27431762
- 38. Marchal A, Schlichting CD, Gobin R, Balandier P, Millier F, Muñoz F, et al. Deciphering hybrid larch reaction norms using random regression. G3 Genes, Genomes, Genet. 2019;9: 21–32. pmid:30333192