Proceedings of the Royal Society B: Biological Sciences
You have accessResearch article

Balanced genetic diversity improves population fitness

Yuma Takahashi

Yuma Takahashi

Frontier Research Institute for Interdisciplinary Sciences, Tohoku University, Miyagi, Japan

Graduate School of Life Sciences, Tohoku University, Miyagi, Japan

[email protected]

Google Scholar

Find this author on PubMed

,
Ryoya Tanaka

Ryoya Tanaka

Graduate School of Life Sciences, Tohoku University, Miyagi, Japan

Google Scholar

Find this author on PubMed

,
Daisuke Yamamoto

Daisuke Yamamoto

Graduate School of Life Sciences, Tohoku University, Miyagi, Japan

Google Scholar

Find this author on PubMed

,
Suzuki Noriyuki

Suzuki Noriyuki

Department of Environmental Science, Policy, and Management, University of California Berkeley, Berkeley, CA, USA

Center for Geo-Environmental Science, Rissho University, Saitama, Japan

Google Scholar

Find this author on PubMed

and

    Abstract

    Although genetic diversity within a population is suggested to improve population-level fitness and productivity, the existence of these effects is controversial because empirical evidence for an ecological effect of genetic diversity and the underlying mechanisms is scarce and incomplete. Here, we show that the natural single-gene behavioural polymorphism (Rover and sitter) in Drosophila melanogaster has a positive effect on population fitness. Our simple numerical model predicted that the fitness of a polymorphic population would be higher than that expected with two monomorphic populations, but only under balancing selection. Moreover, this positive diversity effect of genetic polymorphism was attributable to a complementarity effect, rather than to a selection effect. Our empirical tests using the behavioural polymorphism in D. melanogaster clearly supported the model predictions. These results provide direct evidence for an ecological effect of genetic diversity on population fitness and its condition dependence.

    1. Introduction

    Species diversity contributes to increases in invasion resistance, productivity and stability of natural communities or ecosystems—an effect known as the diversity effect [15]. The observation that the productivity of a functionally diverged community is higher than that expected based on the sum of the productivities of the individual species is usually explained by two mechanisms: (i) complementarity, that is, resource partitioning or positive interaction among species and (ii) a selection effect, because in a diverse community, the probability that a highly productive species will be included is increased [2,5,6]. Analogously, it has been suggested that intraspecific diversity also improves population fitness [7,8]. Diversity within a population or species due to the presence of polymorphisms and personalities is hypothesized to enhance population fitness, as indicated by productivity, growth rate, carrying capacity and stability, by reducing resource competition within the population and by dispersion of risks due to predation, feeding damage, harmful effects from conspecific individuals and extinction under fluctuating environmental conditions [7,911]. However, there have been surprisingly few attempts to evaluate how the evolution of polymorphism affects population fitness, partly, perhaps, because of misunderstanding and critiques of group selection [12].

    Genetic polymorphism in a population is a widespread phenomenon in nature that is potentially maintained by balancing selection or by a selection–migration balance [13,14]. Polymorphisms maintained by balancing selection are predicted to, at least, enhance population fitness [7,9,10,12] because all individuals in a polymorphic population more or less enjoy the rare morph advantage, irrespective of the forces driving the balancing selection, whereas no individuals in a monomorphic population enjoy the rare morph advantage [10]. The demographic consequences of the evolution of polymorphism may vary, however, depending on the ecological and evolutionary context [12,1518] and because empirical support for the context dependence of the effects of genetic polymorphism on population fitness is largely lacking, the existence of a diversity effect of genetic polymorphism is controversial [7,9,10,15,19].

    In Drosophila melanogaster, genetic polymorphism in larval foraging behaviour, which is observed in natural populations, is greatly influenced by allelic variation in the foraging (for) gene [20]; larvae with the dominant Rover (forR) allele move more than those homozygous for the recessive sitter (fors) allele when foraging for food [21]. When reared together, fitness (larval survivorship to pupation) of either morph (forR and fors) decreases when it is common in the population, indicating that negative frequency-dependent selection is acting on the population [22]. Interestingly, negative frequency-dependent selection is observed when the food is nutrient limited, causing larval resource competition to be severe, but it disappears when the food is relatively nutritious, suggesting that the maintenance of foraging behavioural polymorphism by negative frequency-dependent selection is due to frequency-dependent intra-morph resource competition [22]. To investigate whether the co-occurrence of Rover/sitter polymorphism enhances the productivity of a D. melanogaster population, we first developed a simple numerical model predicting the relationship between polymorphism and population fitness under three selection scenarios. Then we empirically examined whether the evolution of genetic polymorphism had a demographic effect on D. melanogaster population in populations exposed to three different nutrient levels.

    2. Material and methods

    We first developed a simple game theory model to predict population fitness over a range of morph frequencies under frequency-dependent selections. We calculated the mean absolute fitness of individuals in a population under negative frequency-dependent selection, frequency-independent selection or positive frequency-dependent selection by changing pay-offs of the game (figure 1). In the latter two scenarios, the polymorphism is passive, meaning that it is maintained by a migration-selection, mutation-selection or mutation-drift balance, rather than by balancing selection. Consider player X with occurrence probability p and player Y with occurrence probability q (= 1−p). Mean absolute fitness of individuals is calculated as [(p × fX) + (q × fY)], where fX is the absolute fitness of X and fY is the absolute fitness of Y when the frequency of X is p. The individual contributions of complementarity and the selection effect to the net diversity effect were separated by additive partitioning [23].

    Figure 1.

    Figure 1. Predicted population absolute fitness and the partitioning of the diversity effect as a function of morph frequency under negative frequency-dependent selection (a,b), frequency-independent selection (c,d) and positive frequency-dependent selection (e,f). Potential fitness (i.e. fitness when the population is monomorphic) is assumed to be symmetric (a,c,e), or asymmetric (b,d,f), as shown by the pay-off matrix (inset in each panel). In the top graph of each panel, solid, dashed and orange lines denote the mean fitness of morph X, morph Y and the population, respectively, and the dotted line indicates the weighted mean of the fitness of the two monomorphic populations (i.e. expected population fitness when there is no diversity effect). A higher (lower) realized population fitness (bold orange line) than expected population fitness (dotted line) indicates a positive (negative) diversity effect. In the bottom graph of each panel, solid, dashed and bold blue lines denote the magnitudes of the complementarity, selection and net diversity effects, respectively. (Online version in colour.)

    For the empirical tests, we obtained two strain sets of the Rover/sitter polymorphism encoded by the foraging (for) gene from the Sokolowski laboratory at the University of Toronto. One strain set, forR and fors, carries the wild-type polymorphism, on which negative frequency-dependent selection has been demonstrated to act [22]. The two morphs were reared separately; each was homozygous for the forR or the fors allele, but shared the same first and third chromosomes. The other strain set consisted of recently developed isogenic strains derived from forR or fors (forNR or forNs, respectively) with a similar genetic background [24].

    To analyse the relationship between morph diversity and population fitness, we reared the larvae in monomorphic or dimorphic populations at one of three nutrient levels and then assayed population productivity. The high-nutrient (a 75% reduction of the yeast + sugar content relative to that of the standard rearing medium) and the low-nutrient (85% reduction) treatments are the same as those used by Fitzpatrick et al. [22], who observed negative frequency-dependent selection under the low-nutrient condition but not under the high-nutrient condition. We added a third nutrient level (50% reduction: very-high nutrient), and we expected that in this treatment, negative frequency-dependent selection would not occur because of reduced competition for resources. To collect fertilized eggs, approximately 100 adults were transferred into collection chambers containing Petri dishes (φ 60 mm) containing grape agar (50 ml grape juice, 50 ml water, 2 g agar, 1.5 ml ethanol and 1.5 ml acetic acid). Then, 32 randomly chosen eggs laid on the grape agar were carefully transferred onto a medium in an experimental vial (50 ml, 28 mm in diameter) containing Rover or sitter morphs or both (Rover : sitter = 0 : 32, 16 : 16 or 32 : 0), and raised to emergence at room temperature. Density (32 individuals per vial) was held constant according to the previous study testing the negative frequency-dependent selection [22]. The dimorphic populations consisted of four strain combinations (two intra-strain combinations, forRfors and forNRforNs, and two inter-strain combinations, forRforNs and forNRfors). Ten replicates of each treatment (three nutrient conditions, three morph frequency and four strain combinations) were performed.

    Population productivity was assessed on the basis of population biomass, BP, and survival rate [22]. To estimate BP, the proportion of the population that survived to pupation was calculated as S/32, where S is the number of pupae. The population biomass at emergence was calculated as (S × RM × WM) + (S × RF × WF), where RM and RF are the proportions of males and females, respectively, among adults collected at the end of the experiment, and WM and WF are the dry weights of an individual male and female, respectively. WM and WF were determined as the mean of five intact individuals of each sex (if the number of intact adults of each sex was less than five, one to four individuals were used).

    The improvement of population productivity in polymorphic populations was assessed by two indicators: a positive deviation of the fitness of the polymorphic population from (i) the average productivity of the two monomorphic populations and (ii) the productivity of the pooled monomorphic population. The former was tested by a one-sample t-test, and the latter was by using a generalized linear model (GLM) and a generalized linear mixed model (GLMM) with frequency (polymorphism or monomorphism) as a fixed factor for survival rate (binomial distribution) and population biomass (Gaussian distribution), respectively. In the GLMM, sex ratio (male ratio) was used as a random factor because body size is quite different between the sexes in this species.

    The contribution of rate of survival to pupation and of sex ratio (male ratio) and mean body weight (mean of two sexes) to population biomass was analysed by GLMM assuming a Gaussian error distribution. Rate of survival to pupation, sex ratio and mean body weight were standardized and used as independent variables, and population biomass was used as a dependent variable. Frequency treatment, nutrient treatment and the combination of two morphs were used as random variables.

    To partition the complementarity and selection effects in the low-nutrient treatments, in which a positive diversity effect was found, the survival rate of each morph in the polymorphic population was estimated. Sixteen adults in each treatment were genotyped by the RFLP method (I Anreiter 2016, personal communication) [25]. The survival rate of a given morph in a polymorphic population was calculated as the number of survivors × the frequency of the morph among the genotyped survivors. Polymerase chain reaction (PCR) analyses were run using whole adult flies with MightyAmp DNA polymerase (Takara Bio), the forward primer TTGATGACTATCCTCCCGATCCT and the reverse primer AAGGCAACCCGATTTGTATGC as follows: 98°C for 10 s; 40 cycles of 98°C for 20 s, 59°C for 30 s and 72°C for 45 s, followed by a final step of 72°C for 7 min. The amplification product was 578 bp long. Pst1 was used to digest the PCR product (for 8 h at 37°C). Gel electrophoresis showed that the restriction enzyme reaction produced two fragments of sitter (123 and 455 bp) and one fragment of Rover (578 bp).

    The diversity effect, Inline Formula, was partitioned into complementarity and selection effects as follows [23]:

    Display Formula
    3.1
    where Mi is the population productivity (survival rate) for a monomorphic population of morph i, PO,i is the observed population productivity of morph i in a polymorphic population, PO is the total observed polymorphic population productivity (Inline Formula), RPE,i is the expected relative productivity (RPE) of morph i in the polymorphic population (the original ratio of eggs of the two morphs), RPO,i is the observed relative productivity of morph i in the polymorphic population (PO,i/Mi), PE,i is the expected productivity of morph i in the polymorphic population (RPE,iMi), PE is the total expected productivity of the polymorphic population, ΔP is the deviation of the total observed productivity in the polymorphic population from the expected productivity (PO − PO), ΔRPi is the deviation of observed relative productivity of morph i in the polymorphic population from the expected relative productivity (RPO,i RPE,i) and N is the number of morphs in the polymorphic population. In this equation, the term Inline Formula represents the contribution of the complementarity effect, andInline Formula represents the contribution of the selection effect.

    To predict the effect of phenotypic diversity on the population stability, the geometric mean (GM), the geometric standard deviation (GSD) and the coefficient of variation (CV, calculated as GSD divided by GM) of population biomass over three different nutrient conditions were calculated and were compared among three frequency treatments.

    3. Results

    Under symmetric negative frequency-dependent selection, where the fitness of X and Y symmetrically decreased with its own frequency and the mixed evolutionarily stable state is p = 0.5 (figure 1a upper graph, solid and dashed black lines), the average fitness of the polymorphic population (figure 1a upper graph, orange line) is higher than that of either monomorphic population (p = 0 and p = 1) and also higher than would be expected from the weighted average fitness of the two monomorphic populations (dotted line). Likewise, under asymmetric negative frequency-dependent selection (e.g. as in a hawk–dove game or with frequency-dependent sexual selection [26]), where the fitness function of a morph in relation to its own frequency differs between the morphs, and the fitness of the two morphs in monomorphic populations is not the same, the fitness of the polymorphic population is also higher than would be expected on the basis of the weighted average of the two monomorphic populations (figure 1b).

    By contrast, under both the symmetric and the asymmetric frequency-independent selection scenario, no increase in average fitness is found in the polymorphic population. Both when the two morphs are selectively neutral and when one morph is more advantageous than the other, the average fitness of the polymorphic population is the same as would be expected on the basis of the weighted average fitness of the two monomorphic populations (figure 1c,d). Moreover, under the positive frequency-dependent selection scenario, the fitness of the polymorphic population is lower than would be expected from the weighted average fitness of the two monomorphic populations, irrespective of whether selection is symmetric or asymmetric (figure 1e,f). These results suggest that the co-occurrence of multiple morphs in a population has a positive effect on population fitness only when the polymorphism is maintained in the population through negative frequency-dependent selection.

    In the negative frequency-dependent selection, frequency-independent selection and positive frequency-dependent selection scenarios, the net diversity effect was positive, absent and negative, respectively, and it was completely or largely attributable to the complementarity effect (lower graphs in figure 1). The selection effect was either absent or slightly negative under all of the scenarios.

    We next examined the relationship between morph diversity and population fitness (productivity) experimentally by using D. melanogaster. When we pooled the results from all four strain combinations (i.e. forRfors, forNRforNs, forRforNs and forNRfors), we found a positive diversity effect D on population biomass in both the low-nutrient and high-nutrient treatments (figure 2), but not in the very high-nutrient treatment. In the low-nutrient treatments, the fitness of the polymorphic population as indicated by the population biomass (BP) was significantly greater compared with the fitness predicted from the weighted sum of the two monomorphic populations (left side of figure 2). Both frequency (F) and the interaction of frequency and strain combination (C) on population biomass were significant (F: F2,70 = 1.204, p = 0.306; C: F3,70 = 4.48, p < 0.006; F × C: F2,70 = 4.93, p = 0.01), and in all four strain combinations, the fitness of the polymorphic population was always greater than that expected on the basis of the weighted sum of the fitness of the monomorphic populations in the low-nutrient treatments (figure 2). In high- and very high-nutrient conditions, weak or no deviation of population biomass of polymorphic population from the fitness was predicted from the weighted sum of the two monomorphic populations, respectively.

    Figure 2.

    Figure 2. Population biomass in a polymorphic (poly, P) and two monomorphic populations (sitter, s, and Rover, R) in low-, high- and very high-nutrient treatments. On the left, the pooled results of all four Roversitter strain combinations are shown. Asterisks indicate a significant deviation of the observed population biomass of the polymorphic population from the weighted sum of the biomass of the two monomorphic populations (population biomass predicted by the null hypothesis; one-sample t-test: *0.01 < p < 0.05; **0.001 < p < 0.01; ***p < 0.001). Plus symbols indicate a significant difference in fitness between polymorphic and monomorphic populations (GLM: +, 0.01 < p < 0.05; ++ , 0.001 < p < 0.01; +++ , p < 0.001). See electronic supplementary material, table S1 for details. Error bar is standard error of the mean (SEM). (Online version in colour.)

    In the low-nutrient treatments of the intra-strain combinations, population biomass of the monomorphic Rover population (BR) differed considerably from that of the monomorphic sitter populations (Bs), probably owing to a difference in genetic background or a pleiotropic effect of the for gene (figure 2; electronic supplementary material, table S1). In these combinations, the biomass of the polymorphic population was lower than that of the monomorphic population with the highest biomass, but, importantly, the biomass of the polymorphic population was higher than average biomass of the two monomorphic populations (figure 2), suggesting a positive diversity effect. This pattern is similar to the pattern predicted under the asymmetric negative frequency-dependent selection scenario (figure 1b). In the low-nutrient treatments of the inter-strain combinations, the biomasses of the monomorphic Rover and sitter populations (BR and Bs, respectively) were approximately equivalent (figure 2), and the population biomass of the polymorphic population was higher than either BR or Bs.

    By contrast, in the high-nutrient treatments, in which negative frequency-dependent selection did not occur, at least in the intra-strain combinations, the biomass of the polymorphic population was increased only in the inter-strain combinations (figure 2). In the very high-nutrient treatments, in which negative frequency-dependent selection is not expected to occur [21,22], the biomass of the polymorphic population was not higher than the mean biomass of the two monomorphic populations (figure 2). These patterns are consistent with the model predictions for the frequency-independent selection or positive frequency-dependent selection scenarios, respectively. The diversity effect (log[BP/BE], where BE is (BR + Bs)/2) on population productivity increased as the nutrient content of the food medium decreased (nutrient content: F = 13.95, p = 0.02; strain combination: F = 1.93, p = 0.267; interaction term: F = 1.35, p = 0.378) (figure 3).

    Figure 3.

    Figure 3. Condition dependence of the net diversity effect. The diversity effect was calculated as loge(observed population productivity/expected population productivity). A positive diversity effect was detected in the low-nutrient treatments, in which negative frequency-dependent selection occurred.

    The survival rate patterns were largely the same as the population biomass patterns (see electronic supplementary material, figure S1 and table S2). The weights of the adult males and females did not show a consistent pattern in relation to frequency and nutrient level, but higher body weights tended to be observed in the polymorphic population in almost all nutrient treatments (electronic supplementary material, figures S2, S3 and tables S3, S4). Over all combinations of nutrient levels (three levels), occurrence frequencies (two monomorphic and one polymorphic) and strains (four combinations), population biomass was positively affected by the rate of survival to pupation (effect = 0.78, p < 0.0001) and mean body weight (mean of both sexes) (effect = 1.22, p < 0.0001), but slightly and negatively affected by sex ratio (proportion of males; effect = −0.05, p < 0.001). In the low-nutrient treatments, the contribution of complementarity to the net diversity effect, calculated from the survival rate, was larger than the contribution of the selection effect (category (complementarity or selection effect): F = 6.446, p = 0.013; strain combination: F = 0.142, p = 0.935; interaction: F = 0.696, p = 0.558), and the selection effect was roughly zero across all four strain combinations, as predicted by our model (figure 4).

    Figure 4.

    Figure 4. Complementarity and selection effects in the low-nutrient treatments. The complementarity effect was higher than the selection effect for all four strain combinations. Error bar is standard error of the mean (SEM). N = 10 for each, excepting forRfors (N = 8). (Online version in colour.)

    The GM of the biomass of the polymorphic population over three different nutrient conditions was higher than that expected from two monomorphic populations in three of the four strain combinations (electronic supplementary material, figure S4). Likewise, the GSD and the coefficient of variation of the biomass of the polymorphic population over three nutrient conditions was lower than those expected from the two monomorphic populations in all of the four strain combinations (electronic supplementary material, figure S4). These suggest high and stable population productivity of polymorphic populations.

    4. Discussion

    In this study, a positive diversity effect on population was observed in the low-nutrient treatments in which negative frequency-dependent selection occurred [22]. Our empirical results thus suggest that genetic diversity in a population improves population fitness when the genetic polymorphism is maintained by balancing selection. This pattern is consistent with our model predictions and suggests that all individuals in a population more or less enjoy a rare morph advantage under negative frequency-dependent selection. In the high- and very high-nutrient treatments, in which negative frequency-dependent selection is not expected to occur, diversity in a population has no or a harmful effect on population performance with a few exceptions, probably because individuals are unlikely to enjoy a rare morph advantage even when population is polymorphic. A harmful effect of diversity was predicted by our model to occur under the positive frequency-dependent selection scenario, but the underlying mechanisms of positive frequency-dependent selection and the presence of positive frequency-dependent selection itself are unclear at this time. When polymorphism is established and maintained by persistent mutation and migration without balancing selection, diversity in a population is predicted to not have a positive (constructive) effect on the population.

    In actuality, improvement of population under polymorphic condition has been reported in various organisms under balancing selection, including damselflies [10] and grasshoppers [27], whereas when polymorphism is maintained by a migration-selection balance, it has a harmful effect on population dynamics [18]. In bacterial populations, genetic diversity improves the carrying capacity in a complex resource environment, where balancing selection may occur, but not in a simple resource environment, where balancing selection apparently does not occur [28]. Likewise, in polymorphic grasshopper populations, a positive diversity effect is found only when the population density is high, where a balancing selection is prone to occur [29]. Importantly, the ecological consequences of genetic diversity on a population's performance may vary depending on the mechanism through which the evolution of genetic diversity is established. As a result, the consequences and the by-product effects of the evolution of genetic polymorphism remain under debate [15,17,19]. Similar variations in ecological consequences depending on the process by which diversity is established and maintained might be found in the relationship between species diversity and ecosystem functioning. Coexistence of diversity and its ecological function may be two sides of the same coin.

    The ecological functions of genetic diversity are interpreted by analogy with the functions of species diversity in ecological communities and ecosystems. In polymorphic populations, complementarity or selection effects, or both, can potentially lead to higher population fitness. However, according to our conceptual model, only complementarity contributes to an increase in population fitness in a dimorphic system under balancing selection. Moreover, the empirical data for Drosophila confirmed that complementarity of genetic polymorphism is the more important mechanism leading to enhanced population fitness.

    At this time, the exact behavioural mechanism or inter-morph interaction that leads to negative frequency-dependent selection and ecological improvement of polymorphic Rover/sitter populations of D. melanogaster is unknown. Intra-morph resource competition may lead to a rare morph advantage. Biased encounters with the same morph might result in costly interactions (e.g. intense resource competition) among individuals with the same morph in a polymorphic population and in inefficient food utilization in a monomorphic population. To deduce the processes underlying negative frequency-dependent selection and the resulting improvement in population fitness, direct behavioural observations are needed. Because genome-wide analyses of European and African populations of D. melanogaster did not detect for genes under balancing selection, the balancing selection acting on for may be relatively weak in natural settings [30].

    For further research, it is important to consider the possibility that increased biomass and survival rate due to the evolution of polymorphism affect long-term population performance and persistence [31]. Although previous studies suggested that polymorphism enhances growth rate, carrying capacity [10,28], establishment success [27,32] and long-term persistence of populations or species [33,34], the long-term ecological effects of behavioural polymorphism in D. melanogaster are currently unclear because we quantified only the survival rate, body size and population biomass in the focal generation. However, our results may provide insight into the long-term demographic effect of the polymorphism in a population. Polymorphic conditions may increase the growth rate of the population on a timescale longer than one generation because both the number and the size of females can increase the number of progenies and biomass of subsequent generations [35,36]. In fact, polymorphic populations showed higher population density and productivity through increased fecundity and oviposition rate of individual females in a damselfly species [10]. Moreover, because the increased survival rate in polymorphic populations of D. melanogaster might be caused by efficient utilization of resources, polymorphic conditions are suggested to increase the carrying capacity of a population. Furthermore, higher GM and lower variation of the biomass of this polymorphic population suggested that polymorphism could improve population persistence in a temporally fluctuating environment. To confirm the demographic consequences of the evolution of behavioural polymorphism in D. melanogaster, long-term population dynamics of monomorphic and polymorphic populations should be directly assessed under both stable and fluctuating environments in future studies. In addition, the long-term impact of polymorphism on high-level ecological processes such as community structure and ecosystem functioning should also be prioritized in biodiversity management as well as our understanding of the further ecological role of evolutionary diversification within a population [3,37,38].

    Ethics

    This research was performed in accordance with the laws, guidelines and ethical standards of Japan, where the research was performed.

    Data accessibility

    The data that support the findings of this study are available in Dryad with the identifier http://dx.doi.org/10.5061/dryad.kb015 [39].

    Authors' contributions

    Y.T. and R.T. designed and performed the experiments. Y.T. conducted data analysis. Y.T. and S.N. wrote the first draft of the manuscript. Y.T., S.N., R.T., D.Y. and M.K. edited the manuscript. D.Y. gave technical support.

    Competing interests

    The authors declare no competing financial interest.

    Funding

     This work was supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI grant no. 26650154 to Y.T. and a JSPS Overseas Research Fellowship to S.N.

    Acknowledgements

    We thank I. Anreiter and A. Allen for providing information on molecular techniques to distinguish the two morphs, forNR and forNs, used in the present study. We thank M. Sokolowski for kindly providing the fly strains and for suggestions that helped to improve the manuscript.

    Footnotes

    Electronic supplementary material is available online at https://dx.doi.org/10.6084/m9.figshare.c.3965412.

    Published by the Royal Society. All rights reserved.

    References