How much energetic trade‐offs limit selection? Insights from livestock and related laboratory model species

Abstract Trade‐offs between life history traits are expected to occur due to the limited amount of resources that organisms can obtain and share among biological functions, but are of least concern for selection responses in nutrient‐rich or benign environments. In domestic animals, selection limits have not yet been reached despite strong selection for higher meat, milk or egg yields. Yet, negative genetic correlations between productivity traits and health or fertility traits have often been reported, supporting the view that trade‐offs do occur in the context of nonlimiting resources. The importance of allocation mechanisms in limiting genetic changes can thus be questioned when animals are mostly constrained by their time to acquire and process energy rather than by feed availability. Selection for high productivity traits early in life should promote a fast metabolism with less energy allocated to self‐maintenance (contributing to soma preservation and repair). Consequently, the capacity to breed shortly after an intensive period of production or to remain healthy should be compromised. We assessed those predictions in mammalian and avian livestock and related laboratory model species. First, we surveyed studies that compared energy allocation to maintenance between breeds or lines of contrasting productivity but found little support for the occurrence of an energy allocation trade‐off. Second, selection experiments for lower feed intake per unit of product (i.e. higher feed efficiency) generally resulted in reduced allocation to maintenance, but this did not entail fitness costs in terms of survival or future reproduction. These findings indicate that the consequences of a particular selection in domestic animals are much more difficult to predict than one could anticipate from the energy allocation framework alone. Future developments to predict the contribution of time constraints and trade‐offs to selection limits will be insightful to breed livestock in increasingly challenging environments.

tive genetic correlations between productivity traits and health or fertility traits have often been reported, supporting the view that trade-offs do occur in the context of nonlimiting resources. The importance of allocation mechanisms in limiting genetic changes can thus be questioned when animals are mostly constrained by their time to acquire and process energy rather than by feed availability. Selection for high productivity traits early in life should promote a fast metabolism with less energy allocated to self-maintenance (contributing to soma preservation and repair). Consequently, the capacity to breed shortly after an intensive period of production or to remain healthy should be compromised. We assessed those predictions in mammalian and avian livestock and related laboratory model species. First, we surveyed studies that compared energy allocation to maintenance between breeds or lines of contrasting productivity but found little support for the occurrence of an energy allocation tradeoff. Second, selection experiments for lower feed intake per unit of product (i.e. higher feed efficiency) generally resulted in reduced allocation to maintenance, but this did not entail fitness costs in terms of survival or future reproduction. These findings indicate that the consequences of a particular selection in domestic animals are much more difficult to predict than one could anticipate from the energy allocation framework alone. Future developments to predict the contribution of time constraints and tradeoffs to selection limits will be insightful to breed livestock in increasingly challenging environments.

| INTRODUC TI ON
Most livestock breeding programmes currently generate hyperproductive animals: in less than half a century, the chicken body growth rate has increased almost fivefold (Collins et al., 2014;Havenstein et al., 2003;Zuidhof et al., 2014), annual egg production of layer hens has doubled (Preisinger & Flock, 2000), pig litter size at birth has increased since the nineties by one piglet every 5 years (Merks et al., 2012), and annual milk yield of dairy cows has increased by 140 kg every year from the seventies. As selection for higher productivity (see glossary in Box 1 for a definition) generally accelerates the conversion of feed supply into increasing amounts of animal products (Vandehaar, 1998), it has been a successful economic strategy to make nutrient-rich animal source food increasingly affordable to a growing human population (Garnett et al., 2013). Besides the major sustainability concerns including environmental, health and welfare issues, the large and continuous increase in livestock productivity falls into the lively debate regarding biological limits to energy expenditure (Denny, 2008;Piersma, 2011;Careau et al., 2013; see Marck et al., 2017 for a review).
In agricultural sciences, the extent to which further genetic gains (Box 1) can be sustained is a long-standing issue (Collier et al., 2005;Fredeen, 1984;Hammond, 1947;Hunton, 1984;Kennedy, 1984;Legates, 1967;reviewed in Hill & Bünger, 2004) that continues to be puzzling (Hill, 2016;Tallentire et al., 2018). Evolutionary theory predicts the depletion of additive genetic variance as a consequence of selection (Hill, 2016;Merilä et al., 2001). Thus, a strong and sustained directional selection on productivity traits should lead to the fixation of favourable alleles and elimination of disfavoured alleles, and thereby should stop the evolution of more productive genotypes (Robertson, 1960). However, most long-term (i.e. >30 accompanied by negative side effects on health and fertility traits Julian, 1998;Nebel & McGilliard, 1993;Rauw et al., 1998;Roberts, 1979). To counterbalance these negative side effects, several affected traits are now part of the breeding objectives (Box 1) as their continuous degradation would otherwise carry economic losses Miglior et al., 2005;Neetesonvan Nieuwenhoven et al., 2013). However, it remains unclear whether selection adjustments in the breeding objectives are high enough to mitigate the degradation in health, survivorship or fertility  or to prevent the occurrence of new detrimental effects (e.g. increasing incidence of growth-related breast meat abnormalities in chicken selected for fast growth; Soglia et al., 2018).
Neither do we know whether those negative side effects are linked to physiological constraints (Box 1), which may ultimately impose limits to selection (Blows & Hoffmann, 2005).
In all organisms, productivity measures the output of an energy conversion process governed by general physical and chemical principles and sustained by a metabolic engine that is also energetically costly to maintain (Biro & Stamps, 2010). Accordingly, the traditional view that the energy available to individuals is limited so that any energy allocation to one function reduces the allocation to another function (Cody, 1966) has also been considered in the context of livestock and laboratory animals (Beilharz et al., 1993;Rauw, 2009). The resource allocation theory became popular within livestock sciences after Rauw et al. (1998) conjectured that it offers a biological explanation for the widespread occurrence of negative side effects of selection for high productivity. This view received some support from experimental comparisons of selected lines (e.g. Rauw et al., 1999;Savietto et al., 2014;Schütz & Jensen, 2001;Theilgaard et al., 2007) and was further established through reports on farm animal health and welfare (e.g. Oltenacu & Broom, 2010;Prunier et al., 2010).
Although developments of a resource allocation framework in the livestock context have then focused on predicting the consequences of animal breeding (Doeschl-Wilson et al., 2009;Friggens & Newbold, 2007;Van Der Waaij, 2004), the applicability of the allocation principle to such context has rarely been questioned (Gilbert et al., 2017).
This may seem confusing as environments optimized for production such as farms or laboratories are unlikely to primarily set limits on the amount of energy that animals can obtain. Without such limits and given the persistent genetic variation in productivity traits, livestock selection should simply generate animals that escape resource allocation trade-offs by compensating the higher nutrient allocation to biosynthesis with higher rates of feed acquisition (van Noordwijk & de Jong, 1986;Reznick et al., 2000). Yet, in practice, nutrient intake may be frequently limited by the economic dependence of intensive production systems on expensive nutrient-rich feed inputs such as grain (Garnett et al., 2013). Nutritional limitations may also be imposed by farmers to avoid health problems of excess nutrient consumption (D'Eath et al., 2009) or may occur incidentally (e.g. as a result of competitive group-feeding; Grant & Albright, 2001).
Most obviously, time must be limiting and time constraints remain independently of the amount and quality of resources available , leading to the existence of potential limits in the rate at which animal bodies metabolize and process energy (Hammond & Diamond, 1997). Animals selected for high productivity may increase their time spent foraging but such increase is inevitably limited. For instance, dairy cows selected for high milk yield spend an increasing proportion of the day eating at the expense of lying time which, in turn, may approach welfare limits (Løvendahl & Munksgaard, 2016). There are also inherent physiological limits on the time to achieve certain production processes. For instance, layer hens cannot produce more than one egg per day under a normal light-dark cycle of 24 h, so selection on egg production basically seeks to increase the laying frequency over the annual production cycle (Hunton, 1984). Such time constraints are exceptionally challenged in livestock breeding given F I G U R E 1 Several paths of change in energy acquisition and allocation may be expected from applying the Y model of energy allocation to selection for livestock productivity. MEI allocated to productivity (energy deposited in new biomass during growth or reproduction plus the energy of biosynthesis; in red) versus maintenance and activity (once the covariation with body size has been accounted for; in yellow). Each point represents an individual, and paths 1 to 3 predict how a same selective increase in a productivity trait (e.g. growth rate) would be mediated through energy acquisition and allocation depending on the energy available. Grey shade reflects the trade-off intensity that feed availability is relatively nonlimiting but that strong selection is exerted to quickly convert feed intake into animal products. However, in those conditions where animals are mostly constrained by their time to metabolize energy rather than by the amount of energy, we still have a limited knowledge of how do selective increments in productivity translate in terms of energy allocation trade-offs (Jackson & Diamond, 1996;Konarzewski et al., 2000). Although the role of time constraints and subsequent energy allocation in life history evolution is well established (Arnold, 1992;Garland & Carter, 1994;Ricklefs & Wilkelski, 2002), their link to the limits of selection for higher rates of energy expenditure in the form of animal products is mostly unexplored, despite their strong biological and socio-economic implications.
Here, we propose that the apparent absence of selection limits for increased productivity despite the existence of physiological trade-offs illustrates a conundrum in our understanding of selection responses. Did selection for ever-increasing rates of biosynthesis in nutrient-rich environments lead us to overlook the existence of physiological limits to acquire and process energy, which can then favour the occurrence of energy allocation trade-offs? To address this issue, data from livestock and related laboratory models represent a relevant starting point. Indeed, a large body of research on feed efficiency (Box 1) of agricultural production exists that is primarily motivated by economic implications as feed accounts for most of the production costs on farms (Johnson et al., 2003). Greater feed conversion rates or feed efficiency is limited by the energy required for animal maintenance, which goes from 10% in pigs and poultry to more than 50% in ruminants, whereas this energy allocated to self-maintenance (Box 1) is thought to be key to sustain production processes over lifetime (Kirkwood, 2017). This research on livestock feed efficiency has a clear general biological basis, which, for instance, allowed inference from laboratory species such as mice or Japanese quails (Pitchford, 2004). However, its evolutionary implications have been seldom considered (Swallow et al., 2009). Though, data available so far represent an untapped opportunity to highlight the role of energetic trade-offs in selection responses across diverse vertebrate species (Swallow et al., 2009). Here, we predict that if the energy allocation framework strictly applies, then selection for high productivity early in life should concurrently reduce the relative energy allocation to self-maintenance (prediction 1).
Those changes in the energy allocation to productivity relative to the allocation to maintenance are more direct when selecting for productivity outputs while controlling for feed inputs. In this last case, we thus expect a reduced energy allocation to maintenance and repairing processes that will result in impaired health or future reproduction (prediction 2). After examining those predictions with the data currently available in the literature, we argue that the mechanisms through which productivity is selectively increased should be addressed in a more complete energy allocation framework than the current one. This appears as a key issue if we are to predict the consequences of selection for livestock bred in increasingly diverse and challenging environments.

| TR ADE-OFFS DEFINED IN THE ENERGY ALLO C ATI ON FR AME WORK
The principle of energy allocation between self-maintenance, growth and reproduction is a cornerstone of life history theory on which the expectation of evolutionary trade-offs is based (Cody, 1966;Williams, 1966). Accordingly, several studies performed in the wild have revealed negative associations between the allocation to growth or reproduction and the subsequent survival or reproductive performance (Hamel et al., 2010;Lemaître et al., 2015). However, trade-offs are far from being consistently detected and positive associations are commonly reported (Lemaître et al., 2015. Van Noordwijk and de Jong (1986) proposed that energy allocation trade-offs should only occur when individual variation in resource allocation is larger than the variation in resource acquisition (a model called the 'Y model'). However, the model assumption that the relative resource allocation is independent from resource acquisition is too restrictive to cover the diversity of resource allocation strategies observed among taxa (Descamps et al., 2016). Moreover, the association between resource acquisition and resource availability is complex and prevents any reliable prediction of whether a tradeoff occurs or not in a particular context (Boggs, 2009;Glazier, 1999;Roff & Fairbairn, 2007). In practice, energy acquisition and allocation are difficult to measure and are most commonly inferred from a limited number of phenotypic measurements (Ng'oma et al., 2017).
Experimental manipulations of the quantity or quality of feed supply or of the reproductive effort have shown that physiological tradeoffs are more acute under stressful environments (Boggs, 2009).
Likewise, several studies in the wild have detected fitness cost of reproduction only when resources are limited during specific years and/or in specific areas (Cohen et al., 2019). For example, in Soay sheep (Ovis aries), young breeding males suffer from an impaired survival only in years of high population density (Stevenson & Bancroft, 1995), likely as a result from an earlier allocation to reproduction.
The costs of reproduction are indeed particularly high in this species, probably because Soay sheep bear the hallmarks of selection for increased productivity during their initial domestication (Clutton-Brock & Pemberton, 2004). Still, in the wild, a proper test of the Y model is largely hindered by practical difficulties to get reliable estimates of individual variation in resource acquisition. This is less of an issue in high-input livestock production systems where the availability of feed intake data enables large-scale genetic evaluations.
In contrast to life history evolution that primarily views the principle of allocation as a powerful theoretical framework to interpret evolutionary outcomes, this principle has been more narrowly applied to livestock breeding as an organismal constraint on the fluxes of energy and nutrients between functions that governs responses to artificial selection (Beilharz et al., 1993;Rauw, 2009). Under the Y model, MEI (Box 1) can be split on a daily basis by individuals into one part allocated to productivity (energy retained as new biomass such as muscle, fat, milk, egg or wool plus the associated energy of biosynthesis dissipated as heat; Figure 1), and the other part allocated to activity and self-maintenance (with the latter maximizing the chance of survival and future reproduction; Kirkwood, 2017; Box 1). Most negative side effects of selection for high productivity are thus thought to stem in the extent to which energy allocation to maintenance is reduced in the total energy budget MEI (Beilharz et al., 1993;Rauw et al., 1998) once its covariation with body size has been accounted for (Konarzewski & Książek, 2013). However, under the assumption of independence between resource acquisition and relative allocation (Descamps et al., 2016), a same increase in productivity over generations of selected animals could be mediated through an increase in MEI, or through an increase in relative allocation (i.e. energy allocation to productivity relative to MEI), or through a mixture of both changes possibly independently from each other (Figure 1).
Further, if we consider that energy allocation to maintenance mostly reflects the minimal energetic cost to maintain the tissues and essential life functions or to support the 'metabolic engine' (Biro & Stamps, 2010), then the different paths of Figure 1 match hypotheses developed by physiologists to explain intra-specific variation in BMR (Box 1) in an ecological context (Burton et al., 2011;Careau & Garland, 2012). Similar to the 'independent hypothesis' that predicts variation in metabolic rates independently from daily energy expenditure (Konarzewski, 1995), livestock nutrition generally considers that energy allocation to productivity is independent from the allocation to maintenance (path 2 in Figure 1). A key implication of this assumption is a dilution of the maintenance energy requirements (Bauman et al., 1985;Vandehaar, 1998) so that gains in livestock productivity are associated with gains in feed efficiency (Capper & Bauman, 2013). In other words, the energy costs of maintenance are assumed to be fixed, leading the maintenance cost per unit of animal production to be reduced. In contrast, a greater productivity might involve higher maintenance costs (the 'increased-intake hypothesis', Biro & Stamps, 2010;Burton et al., 2011;Careau & Garland, 2012;Koteja, 2000;Nilsson, 2002; path 1 in Figure 1). Studies in livestock have implicitly assumed that the amount of feed available in the environment (often categorized as 'limiting' vs.
'nonlimiting') determines which of the previous paths is undertaken during selection for increased productivity. In limiting environments (path 3 in Figure 1), trade-offs show up because any increase in productivity should result in reduced energy allocation to maintenance during the productive phases of animal life (cf. prediction 1). If this reduction is associated with insufficient maintenance and repairing processes, then the capacity to start over a new productive cycle or to remain healthy should be impaired (cf. prediction 2). In contrast, energy allocation trade-offs are unlikely to occur as far as the feed supply keeps up with genetic gains in productivity (paths 1 and 2 in

| SURVE Y OF B REED COMPARISONS AND S ELEC TI ON E XPERIMENTS
To assess prediction 1 (i.e. selection for productivity results in reduced allocation to maintenance) and prediction 2 (i.e. reduced energy allocation to maintenance results in impaired health or future reproduction), we surveyed studies in livestock species (mammals and birds) that have investigated the genetic relationships between productivity and energy allocation to self-maintenance.
We thus performed a systemic search in Web of Science by using the following string of search terms: ('metabolic rate' OR 'maintenance' OR 'heat production') AND ('line' OR 'breed' OR 'strain') in the Web of Science categories ('Agriculture Dairy Animal Science' OR 'Genetics Heredity' OR 'Veterinary Sciences' OR 'Ecology' OR 'Multidisciplinary Sciences'), and using citation chaining ( Figure S1 and the PRISMA diagram in Supplementary Material). Surprisingly, eligible studies were quite rare. We looked chiefly for selection experiments as most of the phenotypic differences among lines can be attributed to genetic differences generated during the selection process. Moreover, selection experiments are particularly relevant to investigate physiological constraints in action during evolutionary changes (Garland, 2003;Box 2). Unfortunately, only a few of them have actually measured RMR in livestock. Data on RMR are more frequent for a particular breed or commercial line to estimate its nutritional requirements. In this case, the genetic effect related to productivity was less obvious, but our survey also included those studies based on breed comparisons to further discuss the effect of different levels of productivity on RMR (Tables 1 and 2). To gain complementary insights, we also considered laboratory species that have been commonly used in livestock research as relevant biological models. Thus, results from previous syntheses within particular taxa (e.g. Konarzewski & Książek, 2013) could be compared with those available for other mammals and birds.

| Prediction 1: selection for productivity results in reduced allocation to maintenance
We distinguished two main types of productivity traits: those related to body growth (Table 1; studies 'Gi'), and those related to maternal reproductive outputs such as litter size or milk yield in mammals (Table 2; studies 'Ri'). As for energy allocation to self-maintenance, it cannot be approximated by BMR when animals are in a production phase. We considered instead any measure of RMR (Box 1) as previously done (Biro & Stamps, 2010;Burton et al., 2011). In livestock studies, related metrics include heat production after fasting (using indirect calorimetry) or estimates of energy intake by extrapolation at various feeding levels towards zero energy retention (using comparative slaughter technique to estimate the retained energy (RE) in body mass gain). None of the methods that were developed to determine energy maintenance requirements during productive phases have gained broad acceptance due to the difficulty to tease apart the energetic costs related to productivity from the minimal energetic cost of living (Knap, 2009). Moreover, previous metrics are not homogeneously adjusted for the effect of metabolic allometry that stands within species (Konarzewski & Książek, 2013). Most studies in livestock research divide them by the so-called metabolic body mass (i.e. m 0.75 ), but we also found studies that simply reported mass-specific estimates and others that used covariance analysis

BOX 2 When selection experiments reveal the importance of physiological constraints in selection responses
Identifying physiological constraints at the level of genes is challenging. Focal traits under selection, such as productivity traits, are often the end products of complex genetic, physiological, developmental and endocrine processes. The biological complexity in the mapping of genotype to phenotype can be reduced by focusing on mechanistic traits, which usually refer to a small number of physiological traits that determine focal traits. To assess the extent to which those mechanistic traits constrain or enable selection responses, laboratory selection experiments exploring multiple directions of selection are relevant tools (Garland, 2003). If in all directions of selection, observed direct responses of focal traits are consistent with predicted responses based on mechanistic traits, this would support strong physiological constraints. This is well illustrated by Davidowitz et al. (2016)'s study case on the tobacco hornworm (Manduca sexta; Figure 2).
Two focal traits of larva development to pupation were artificially selected: development time and pupal weight. Those traits can be predicted from their functional relationships with three mechanistic traits (i.e. critical weight (CW), length of the growth period (ICG) and growth rate (GR)) that characterize the last stage of larva development into pupae: where the attainment of the CW marks the transition from the juvenile growth to the reproductive adult phase (initial adult weight), ICG corresponds to the duration of the period of growth from CW to pupal weight, and GR refers to the linear rate of change in body weight during this stage.
Both CW and ICG have positive effects on development time and pupal weight, whereas GR has a positive effect on pupal weight and a negative effect on development time. Pupal weight and development time were simultaneously selected to explore four directions (i.e. same or opposite directions). Regression lines in Figure 2 show that whatever the direction of selection, responses are highly predictable from the correlated changes in the three mechanistic traits. In other words, those physiological determinants of development at the individual level strongly constrain responses to selection in multiple directions.   : Kuenzel and Kuenzel (1977); G10: Jackson and Diamond (1996); G11: Fan et al. (1997); G12: Sundstøl et al. (1979); G13: Kolstad and Vangen (1996) (re-evaluated in Knap (2009)   with body mass as covariate, as recommended by integrative physiologists (Konarzewski & Książek, 2013). All studies from our survey used nutrient-rich diets that were provided ad libitum, and thus, MEI was not limited by the environment.
Overall, about half cases (10 out of 21) did not detect any difference in maintenance energy metabolism between animal groups with contrasting growth rates (Table 1). However, results varied across species. In mice, no association was reported between productivity and RMR in 80% of the cases. In poultry, a negative association occurred in two thirds of the cases. In pigs, a positive association was detected in all the four studies, regardless of a selection on low fatness in addition to high productivity (studies G12 and G15). Finally, there was no clear trend in ruminants. Interestingly, the relationship between maintenance and fatness (studies G12, G15 and G18) was reciprocally observed in some cases (i.e. a higher fatness associated with a lower maintenance; studies G3, G4, G8 and G19), and none of the 21 cases indicated a positive covariation between fatness and

RMR.
All six studies in reproducing females (Table 2) found a positive effect of maternal productivity on RMR, in support of the 'increasedintake hypothesis' (Figure 1; path 1).
Associations between the genetic potential for growth and maintenance requirements were all measured in growing animals, sometimes in several growth phases (e.g. studies G8 and G14).
However, there was little information on the persistence of those associations beyond the growth period in these determinate growers (i.e. few studies on mature animals such as G19). In contrast, evidence for positive associations between maternal productivity and RMR was also observed when adult females were not reproducing (e.g. studies R3-R5). Still, the detection of associations is inevitably hindered by the limited data relative to the large number of factors, notably caused by the use of various metrics to measure RMR.

| Prediction 2: reduced energy allocation to maintenance results in impaired health or future reproduction
To study the consequences of genetic changes in the energy allocation to maintenance on components, or physiological correlates, of health or future reproduction, we surveyed selection experiments directly targeting energy metabolism (e.g. Konarzewski et al., 2005) or indirectly through feed efficiency (Table 3, studies 'Mi'). Indeed, in the livestock context, feed efficiency and related metrics such as residual feed intake (Box 1) are more commonly studied than RMR because they have higher practical relevance and become increasingly available in commercial populations with the development of individual feed intake monitoring. In our survey, all the selection experiments on feed efficiency where a RMR-related metric was measured showed that feed efficiency and RMR were consistently and positively related (Table 3). Selection for feed efficiency also affects components other than energy metabolism (e.g. digestive efficiency, especially in ruminants). However, experiments in mice show that, in turn, selection for RMR (e.g. heat loss/m 0.75 (Exp. M7) or massindependent BMR (Exp. M8)) has a positive selection response on feed intake (Table 3). Thus, we assumed that results from selection for high feed efficiency and for low RMR were comparable. Overall, feed efficiency metrics have a moderate heritability (e.g. mean of 0.25 from 35 estimates in seven species as reported by Pitchford, 2004), close to that reported for RMR (Konarzewski & Książek, 2013;Pettersen et al., 2018). Both feed efficiency and RMR metrics thus well respond to selection.
Although selection was supposed to be mass-independent, about one third of the cases had a positive effect on body mass and several experiments showed an increasing mass of some digestive organs (e.g. liver mass; exps. M2, M6-8 and M10; Table 3).
In addition, experiments that measured physical activity all found it positively associated with feed intake. Altogether, these results support the 'increased-intake hypothesis' (Biro & Stamps, 2010;Koteja, 2000;Nilsson, 2002), even when variation in body mass is accounted for.
We found equivocal evidence for a relationship between RMR metrics and components or physiological correlates of reproduction or survival (Table 3). Six of the 11 cases investigating those aspects detected a positive association between RMR and reproductive output, in agreement with previous results (  (Burton et al., 2011;Pettersen et al., 2018). Finally, it is noteworthy that the set of physiological traits commonly used to date to assess health and survival in that context (see Table 3) does not include accurate markers of survival prospects (Kennedy et al., 2014;López-Otín et al., 2013), especially when they are investigated separately. For instance, field evidence from Soay sheep has revealed that multiple markers of oxidative stress are required to assess reliably the relationship between oxidative damage and survival (Christensen et al., 2016).

| S HORTCOMING S OF A RE SOURCE SUPPLY-DEMAND APPROACH
Trade-offs in life history evolution are generally viewed as a result of competing allocation between functions for a fixed amount of resources. Although resource allocation trade-offs have also been considered to play a key role in the context of livestock selection (Beilharz et al., 1993;Rauw, 2009)

| Energy allocation to maintenance is difficult both to estimate and to interpret
In comparison with the energy expended in biomass production (e.g. milk, egg, meat, wool), the energy used for self-maintenance and dissipated as heat is more challenging to estimate and to interpret functionally, particularly when animals are in a productive phase ( Figure 3b). Just as RMR (Pettersen et al., 2018), feed efficiency is likely to involve more than one trait. Thus, using those metrics to measure energy allocation to maintenance as presented in Figure 1 is undoubtedly an oversimplification. For instance, in meattype chicken, a positive relationship between growth rate and RMR may translate into an increase in heat dissipated due to biosynthesis (proportional to growth rate) and, at the same time, a decrease in the energy needed for other functions, such as the expression of energetically expensive behaviours (Tallentire et al., 2016). The magnitude of the energy costs of biosynthesis is sometimes so high that it can flip the sign of the relationship between productivity and RMR if not discounted from total heat production (Konarzewski, 1995). For instance, besides changes in energy acquisition and allocation, the efficiency of use in metabolizable energy in mitochondria is also subject to selection for high feed efficiency as observed in chicken (Bottje & Carstens, 2009), pigs (Gilbert et al., 2017) and cattle (Cantalapiedra-Hijar et al., 2018). If high mitochondrial efficiency incurs fitness costs, notably through higher generation of reactive oxygen species (ROS; Salin et al., 2015), this would imply that fitness costs do not necessarily result from a low energy allocation to somatic maintenance per se. Surprisingly, studies in livestock found that ROS was low in lines selected for high feed efficiency (Table 3) and that energetically repairing processes are actually enhanced in efficient animals with a relatively low RMR ( Even in the absence of direct selection for growth, several anabolic pathways were upregulated in low-fear chickens, just as observed in the modern chicken (Jackson & Diamond, 1996). Reciprocally, our results suggest that a reduction in activity is probably the most obvious side effect of selection for high feed efficiency (Table 3), in line with Rauw et al. (2017). In livestock, many selection experiments have been performed on traits not directly related to feed efficiency (e.g. behavioural traits or resistance to specific diseases).

TA B L E 3 (Continued)
Unfortunately, correlated responses on whole-organism energy metabolism and allocation have too rarely been assessed. Although such responses are challenging to measure, they would be insightful to discriminate between the role of correlational selection and constrained resource supply in causing trade-offs (Figure 3d).

| Constraints on energy supply and expenditure are elusive
If there is a constraint on feed intake (as in Figure 3a), should selection for higher productivity effectively result in a reduced amount of energy allocated to other traits? Support for this has been provided by earlier experiments in mice, where selection for growth under restricted feed conditions has resulted in a decrease in RMR (Falconer, 1960;McPhee et al., 1987;Nielsen & Andersen, 1987). In contrast, livestock selected for high productivity are often fed with abundant nutrient-dense feed. Moreover, physiological limits to feed intake have rarely been of concern, except in the growing chicken (Jackson & Diamond, 1996;Konarzewski et al., 2000;Ricklefs, 1985).
Physiological limits to sustained energy intake and expenditure have been extensively studied in rodents (reviewed in Speakman & Król, 2011). The focus of these studies has shifted from central limitations-where energy is assimilated-to limitations at the tissue level-where energy is used (Hammond et al., 1996)-or at the whole-organism level-where it is dissipated as heat (Speakman & Król, 2010). However, the role of those limits acting within individual is rarely considered in the context of individual variation although this is a critical issue for the manifestation of trade-offs (Careau & Wilson, 2017;van Noordwijk & de Jong, 1986). In particular, few studies have addressed whether potential limits are absolute or can be overcome by selection (Garland & Carter, 1994). To this end, the development of mechanistic approaches to link physiological processes to selection responses (e.g. Box 2) is a promising avenue for future research, with key applications for livestock breeding.
An absolute limit on daily feed intake is currently suspected to cause a plateau in commercial selection for fast-growing chicken (Tallentire et al., 2018). However, the underlying mechanisms of this limit remain elusive. For long, the primary focus has been on limits in the developmental capacity of central organs involved in energy supply to keep up with the selective increase in tissue demand (e.g. Dror et al., 1977). A nutritional imbalance is associated with the development of various metabolic troubles (e.g. cardiovascular diseases and musculoskeletal disorders (Julian, 2005) or muscle abnormalities (Soglia et al., 2018) in poultry; but see also Prunier et al. (2010) in pigs or Ingvartsen et al. (2003) in dairy cows). However, the general picture that has emerged from decades of research on the limits to sustained energy intake suggests that multiple constraints are involved, to different extents throughout development, and both from the supply side through the digestive and circulatory systems and from the demand side through peripheral tissues (Ricklefs, 2003;Speakman & Król, 2011). Moreover, ontogenetic processes governing nutrient supply and demand are amenable to selection as demonstrated by selection on the shape of growth curves in chicken (Mignon-Grasteau et al., 1999). Hence, the incidence of several metabolic diseases related to an imbalance between supply and demand has been reduced through selection (e.g. notably in chicken with tibial dyschondroplasia (Wong-valle et al., 1992) or pulmonary F I G U R E 3 Pathways mediating the responses in energy allocation to maintenance to selection for increased productivity. (a) Allocation constraint under limited energy supply; (b) multifunctionality of maintenance, including foraging behaviour; (c) contextdependent selection acting separately (e.g. selection for decreased antipredator response during domestication); and (d) correlated selection response in maintenance due to genetic association with the selected productivity trait. Pathways (a) and (b) involve physiological constraints on energy allocation (dark arrows), contrarily to (c) and (d) (grey arrows). Pathways are nonexhaustive and not mutually exclusive. Solid arrows indicate energy flows, dotted lines indicate functional relationships, dashed double arrows indicate genetic association, and triangles denote selective pressures hypertension (Pavlidis et al., 1997)). In these cases, selection for reduced disease incidence also led to a decrease in growth that may well reflect that facing the boundaries of the physiological space is ineluctable. However, in several cases, the limits that trade-offs impose on the simultaneous improvement of antagonistically related traits have been overcome by selecting for an index that combines those traits (e.g. Index = b 1 X 1 + b 2 X 2 , where b 1 and b 2 are the selection weights applied to traits X 1 and X 2 , respectively) or by selecting for a composite trait (Neeteson-van Nieuwenhoven et al., 2013). For instance, Nielsen et al. (2013) showed that in sows, selection for high litter size after the early nursing period simultaneously increased litter size at birth and neonatal survival, whereas a trade-off exists between these two latter traits. However, another trait may then become first limiting (e.g. number of sow teats) and then a new target of selection. Knowing to what extent simultaneous selection for increased demand and lower supply can be sustained and how this translates morphologically would be useful to clarify the importance of physiological constraints at an among-individual level.
So far, the hypothesis that morphological constraints on organs involved in energy supply would limit selection for productivity has received little support (Schmidt et al., 2009;Zuidhof et al., 2014).
Indirectly, supply organs such as the heart, the liver or the intestine seem to respond to selection for increased nutrient demand.
Moreover, clear evidence supports the view that selection for increased productivity in nutrient-rich environments has mostly been achieved through higher feed intake. This has been the key response in selection experiments for increased growth rate (e.g. reviewed in mice and rats (Roberts, 1979), in rabbits (Feki et al., 1996), in Japanese quails (Marks, 1978), in chickens (Dunnington et al., 2013), in turkeys (Nestor et al., 2008), or in sheep (Thompson et al., 1985)), as well as for increased litter size at birth in mice (Rauw et al., 2000) or in rabbits (Quevedo et al., 2006). The same phenomenon has likely occurred in commercial populations too. For instance, data from digestibility trials indicate that a doubling of milk yield in US dairy cows between 1970 and 2014 was accompanied by a 70% increase in dry matter intake (Potts et al., 2017).
Overall, selection responses are also consistent with responses occurring within individuals as the achievement of increased growth rate through increased intake is also supported during ontogeny. For instance, in chicken, the ontogenetic scaling of MEI tends to be simply proportional to the increasing body mass (i.e. isometry) when chickens are increasingly fast-growers (Figure 4). These changes in the scaling exponent thus suggest a continuous morphological adaptation when individuals with higher energy requirements are selected.
This is in line with evidence showing that allometric relationships do not always resist selection (White et al., 2019). Nevertheless, the room for adaptation may be limited as proposed by the 'metaboliclevel boundary hypothesis' of Glazier (2010; Figure 4). Either way, those observations should warn against the application of interspecific rules of metabolic scaling within species and regardless of ontogeny in livestock nutrition .
In our survey, support for the 'independent' or 'increased-intake hypothesis' (paths 1 and 2, respectively, in Figure 1) suggests that despite intensive selection, resulting in a marked increase in nutritional requirements, the total energy budget is not constrained by an absolute limit on the rate of energy assimilation (as presented in Figure 3a). In the allocation framework, such physiological constraint would be a prerequisite for a negative association between productivity and maintenance (Konarzewski et al., 2000). Our results are thus in line with the rationale of most feeding recommendation systems that consider that more productive animals will simply eat more to meet their higher nutritional requirements (Emmans & F I G U R E 4 Selection for fast-growing chicken has substantially changed growth curves and the allometric relationships between body mass and energy intake. Changes in chicken body mass during the first 8 weeks of age (a) are associated with changes in the scaling of energy intake with body mass (b). In inset, scaling exponents of the relations are reported according to the maximum growth rate of each curve shown in (a). When the metabolic costs of growth increase, the scaling exponent of energy intake converges towards unity. In other words, the growth of new tissue becomes proportional to existing tissue mass, as predicted by the metabolic-level boundary hypothesis (Glazier, 2010). Data were compiled from Jackson and Diamond (1996) and Zuidhof et al. (2014) and from nutritional recommendations of a high-yielding broiler chicken (Ross 308, Aviagen) Kyriazakis, 2001). However, livestock nutritionists generally consider that maintenance nutrient requirements are independent from the level of productivity (Johnson et al., 2003). Others have recommended to revise those requirements upwards to account for an additional demand accompanying increased intake. For instance, the high contribution of visceral organs to RMR is clear according to relationships between RMR and the mass of organs in cattle or in pigs (Ferrell, 1988;van Milgen et al., 1998). However, the mass of digestive organs is not always a good predictor of the energy costs of feed intake (Sadowska et al., 2013;Selman et al., 2001;Speakman et al., 2004).

| Fitness costs of intake have been largely overlooked to date
Although physiological constraints on energy intake and expenditure remain elusive, this does not preclude a physiological restraint that is evolutionary shaped due to the fitness costs of a high intake (Speakman & Król, 2005). Such genetically determined restraint would thus represent a limit that can be overcome by artificial selection for high energy expenditure. Owing to the allometric relationship between feed intake and RMR, it is not surprising that fitness costs of high productivity have also been considered in the evolution of upper limits to metabolic rate (Drent & Daan, 1980;Peterson et al., 1990;Piersma, 2011). In particular, long-term energy expenditures may be restrained by an 'optimal working capacity', that is a maximum level of metabolism that can be sustained and beyond which the mortality risk would become too high due to body reserve exhaustion, organ failure or wear-and-tear processes. As maximum levels of energy expenditure inevitably decline with the duration of the effort (Peterson et al., 1990) and since productivity traits are by definition constrained by time, then we could similarly assume optimal levels of livestock productivity for particular environments.
In livestock, evolutionary hypotheses on the control of feed intake have received some attention (Illius et al., 2002;Tolkamp & Ketelaars, 1992) but they had relatively little impact on the research done on energetic efficiency (Johnson et al., 2003). Two main reasons can be the limited scope for the expression of fitness costs of feed intake late in life in livestock (cf. Results Section 3.2), and the weak empirical and mechanistic evidence of a negative association between high productivity early in adulthood and late-life reproduction or survival traits (e.g. Douhard et al., 2016;Theilgaard et al., 2007). Senescence (i.e. age-specific decline in survival and/ or reproductive performance; see ) seems ubiquitous in contexts as different as those of populations in the wild (Nussey et al., 2013) and highly productive livestock (Brody et al., 1926;Theilgaard et al., 2007). Yet, the trade-off observed in wild populations between high reproduction in early adulthood and performance later in life (Lemaître et al., 2015) is challenging to assess in livestock given that most of those animals are rapidly slaughtered or replaced (Hoffman & Valencak, 2020). For example, dairy cattle, sheep, pigs or goats are commonly slaughtered before reaching 30% of their possible lifespan (Hoffman & Valencak, 2020).
Accordingly, genetic correlations between productivity traits and longevity are often found inconsistent (Essl, 1998). However, longevity is not a proper metric of senescence, which rather corresponds to the declining performance with increasing age that can be analysed well before individuals have reached old ages (Gaillard & Lemaître, 2017;Kinzina et al., 2019). Moreover, the currently quite short lifespan of reproducing adults on farm is often economically suboptimal when the different costs of maintaining herd structure are correctly accounted for. For instance, in high-producing dairy farms, cows should be kept on average 2 years longer than at present (i.e. five lactations instead of three) to compensate for the farm costs of breeding females before they reach sexual maturity and then develop their milk yield potential (which peaks at around four annual lactations; De Vries, 2020). Animals able to sustain their production over the long-term are also increasingly relevant to address welfare and ethical concerns of breeding intensively and remove prematurely too many young animals (Oltenacu & Broom, 2010;Prunier et al., 2010). Improved opportunity to assess senescence patterns in livestock should thus take place in the near future. If early-late life trade-offs also hold in livestock, then strong selection for early productivity (e.g. early age at first reproduction and high litter size) may have incidentally favoured rapidly ageing individuals (Mysterud et al., 2002, on domestic sheep, Grange et al., 2009, which improves the power of studying senescence prior to old ages. In the context of improving livestock welfare and the lifetime production of reproducing adults through lifespan extension, biomarkers involved in health and senescence such as the telomere dynamics or the age-specific changes in the DNA methylation profiles are coming under scrutiny (Monaghan et al., 2018;Simpson & Chandra, 2021). As those biomarkers of biological age become increasingly accurate, they may have a strong interest in the early detection of animals with a high potential for long-life performance.
For instance, the relative leucocyte telomere length quickly declines after birth and is positively correlated to lifespan in high-producing dairy cows (Seeker et al., 2018). As this trait is moderately heritable, it could thus be considered in future breeding objective to extend cow lifespan. Such biomarkers (see also Simpson and Chandra (2021) for a discussion on the use of the epigenetic clocks in that context) may benefit to a broader understanding of early-late life trade-offs (Lemaître et al., 2015). So far, little is known about the respective fitness consequences of energy intake and its allocation to somatic maintenance in livestock.
Viewing the physiological control of feed intake as a target of selection is also in line with incentives to consider more explicitly signalling pathways in the resource allocation framework (Glazier, 2015;Ng'oma et al., 2017) and in behavioural evolution (Garland et al., 2016). Although the genetic basis of the control and orchestration of short-term and long-term regulations of energy metabolism has long been acknowledged by scientists working on livestock (Bauman & Currie, 1980), the associated endocrine pathways and their genetic basis are still largely unravelled. In some cases, the effects of artificial selection on the hormonal control of nutrient allocation during productive phases or of behavioural traits have been extensively studied, but their implications in terms of senescence are rarely considered. For instance, major changes in the somatotropic axis of dairy cows have been associated with selection for high milk yield (Lucy et al., 2009), whereas those hormones also have an evolutionary conserved role in modulating senescence (Bartke et al., 2013). In chicken, selection for growth has substantially altered neuroendocrine pathways, leading to hyperphagia and obesity in adults (Decuypere et al., 2010), which raises major issues in terms of management and animal welfare (D'Eath et al., 2009).

| CON CLUS I ON S AND FUTURE DIREC TIONS
Observations in livestock do not conflict with the compelling evidence that supports resource allocation in the broad sense-that is the fact that changes in energy and material apportioning are part of selection responses. As observed in our study, genetic changes in energy allocation to maintenance are clearer when feed intake is part of the selection objective than when selection is simply based on a productivity trait. However, livestock data question the nature and the importance of resource allocation constraints in the narrow sense (Glazier, 2009)-that is the assumption that resource limitation primarily causes trade-offs that limit selection responses. In particular, we found little evidence that a reduction in the energy allocation to maintenance is at the basis of genetic antagonisms observed in livestock populations between productivity traits and traits related to health or future reproduction. As we discussed, such lack of evidence may well reflect limitations in the measurement of self-maintenance or in the data available so far in livestock to assess costs of high productivity on fitness-related traits. However, it becomes increasingly clear that the resource allocation framework narrowly defined in terms of energy and nutrient supply and expenditure is insufficient to explain trade-offs so that a more complete picture requires the integration of time constraints  and alternative physiological causes (Harshman & Zera, 2007;Leroi, 2001;Ricklefs & Wilkelski, 2002;Speakman, 2008;Zera & Harshman, 2001 and fertility . As shown by Pryce et al. (2014), this antagonism increases in the most favourable environments. Such relationship might be thus consistent with physiological constraints on high metabolic intensity ( Figure 5). Would there be a link between such potential stabilizing selection in livestock populations and the concept of optimal working level of energy metabolism? Answers to this question will be favoured by framing the energy allocation to self-maintenance as a multivariate trait in a microevolutionary F I G U R E 5 Genetic antagonism between dairy cow milk yield and fertility (calving interval, the time elapsed between two successive calving events) according to phenotypic average of milk yield (an indicator of energy availability in the production environment). Here, trade-off intensity (grey shade) increases in less limiting environments. The polynomial trend is shown (dotted line). Figure adapted from Pryce et al. (2014) framework so that genetic covariances among others traits can be measured and potential constraints identified (Konarzewski & Książek, 2013;Pettersen et al., 2018).
Finally, although continuous improvement in livestock productivity seems feasible from the perspective of quantitative genetics (Hill, 2016), there is now alarming evidence that industrial food systems that rely on this strategy are unsustainable given their contribution to unhealthy diets and environmental degradation (Willett et al., 2019). In other words, further increase in productivity may first be limited as a result of constraints at the level of livestock production systems rather at the level of animal biology. Some advocate further gains in animal productivity to reduce feed use and environmental impacts per unit of animal product (e.g. Capper & Bauman, 2013;Garnett et al., 2013). Yet, further increase in the genetic potential for productivity fuels the intensification and specialization of industrialized farming systems. In those systems, heavy reliance on external inputs (e.g. grains, freshwater, antibiotics) raise strong concerns in terms of global (e.g. greenhouse gas emissions, land conversion off-farm) and local (e.g. soil and water pollution) footprints (Willett et al., 2019). Moreover, ethical debates exist on the acceptability of creating ever more productive livestock animals (Hötzel, 2014;Sandøe et al., 1999). Although industrial production systems have provided ideal conditions to foster continuous genetic gains in high-yielding livestock (i.e. controlled and homogeneous environments), livestock breeding now needs to emphasize production in challenging environments (low levels of nutrition, heat stress; Hayes et al., 2013). In this context, the development of farming systems designed from ecological concepts and principles, as proposed by the agroecological paradigm (Altieri, 2018), provides a timely opportunity to renew the historical partnership between evolutionary biology and livestock breeding and genetics (Hill & Kirkpatrick, 2010).
Agroecological systems seek to achieve whole-system productivity by minimizing their reliance on external inputs, by using genetic diversity within farms and by promoting animal capacities to sustain lifetime production in varying conditions. In this respect, research on breeding strategies to improve animal capacities to recover from various environmental challenges (i.e. resilience) is at the forefront (Berghof et al., 2019), just like research on the flexibility of resource allocation and the plasticity of metabolic rate surges with climate change (Ng'oma et al., 2017;Norin & Metcalfe, 2019). Developing connections between those areas has great promise to address their integration within the resource allocation framework.

K E Y WO R DS
livestock breeding, metabolic rate, pleiotropy, senescence, trade-offs

ACK N OWLED G EM ENTS
We are grateful to the Associate Editor, Frédéric Thomas and the three reviewers of this paper, including Marco Festa-Bianchet and Peter Biro, for their thorough revision and their insightful comments on a first draft. Their feedback helped us a lot to improve the final version.

CO N FLI C T S O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were created or analysed in this study.