Direct and correlated responses to long-term residual feed consumption–based selection in Brown Tsaiya ducks

Abstract

Improving feed efficiency (FE) in poultry, particularly in laying ducks, has become increasingly essential for reducing production costs and environmental impact. This study investigated the long-term effects of residual feed consumption (RFC)-based selection on FE and production traits in Brown Tsaiya ducks over nine generations. Two lines were maintained: a selected line (S line), which comprised ducks bred for low RFC over nine generations (G0–G9) of selection, and an unselected control line (C line), which comprised ducks bred randomly under identical management conditions. Selection for low RFC led to substantial reductions in RFC, feed consumption, and feed conversion ratio in the S line, confirming consistent genetic progress in FE compared with that in the C line. Phenotypic differences between the two lines emerged from the fourth generation and became more pronounced in subsequent generations. Although RFC exhibited moderate heritability and strong positive genetic correlations with feed consumption and feed conversion ratio, its relationships with reproductive traits were generally weak. However, the long-term selection response revealed an unfavorable correlated decline in egg mass and egg weight, reflecting a genetic antagonism, confirmed by a highly positive genetic correlation (r_g = 0.566) between RFC and egg mass found in the selected line. Furthermore, the S line exhibited reduced additive genetic variance and increased inbreeding over generations, suggesting the accumulation of selection pressure and narrowing of genetic diversity. Despite these challenges, the S line continued to improve in FE without reaching a genetic plateau, highlighting the potential for further genetic gain. These results establish RFC as a valuable selection criterion for enhancing FE in laying ducks but underscore the need for balanced breeding approaches. Incorporating multitrait or restricted selection indices may help sustain both FE gains and egg production performance, ensuring the long-term genetic and economic sustainability of Taiwan’s Brown Tsaiya duck population.

Introduction

Global climate change poses major challenges to agricultural production, necessitating optimization of animal feed efficiency (FE) to reduce production costs, ensure food security, conserve resources, and mitigate carbon emissions. In the poultry industry, feed costs typically account for ≥60% of total production expenses (NAIF, 2024). Therefore, improving FE is essential for sustainable development of the industry.

The Brown Tsaiya duck (Anas platyrhynchos) is Taiwan’s only indigenous laying breed. It has a small body size, high egg production, large egg weight, and strong eggshells. It is Taiwan’s primary source of processed egg products, including century eggs and salted eggs. Notably, feed costs constitute >80% of total duck egg production costs. Considering this, Liu et al. (2012) developed a reliable method for measuring residual feed consumption (RFC) using the Brown Tsaiya LRI 1 line as the base population, which had been selected primarily for egg production. From this population, a line selected for low RFC (selected [S] line) and a contemporaneous control line (C line) maintained without RFC selection were established. The researchers indicated that RFC measured over 4 weeks (34–37 weeks of age) was genetically correlated (correlation coefficient: 0.93–1.00) with that measured over the entire observation period (22–52 weeks of age); heritability estimates for RFC during this period ranged from 0.30 to 0.43. The Ilan Branch of the Livestock Research Institute, Council of Agriculture, Executive Yuan, Taiwan (now the Eastern Branch of the Livestock Research Institute, Ministry of Agriculture) continued selection based on phenotypic values or estimated breeding values (EBVs) for RFC and officially named the S line high feed efficiency Brown Tsaiya in 2018. By the sixth generation (G6), the S line consumed approximately 10% less feed than did the C line.

The concept of RFC was first proposed by Byerly in 1941. RFC is defined as the difference between an animal’s actual feed consumption (FC) and predicted FC (PFC) for maintenance and production (e.g., egg production and body weight [BW] gain). Low-RFC (R−) animals can achieve egg production levels similar to those of high-RFC (R+) animals while consuming less feed, exhibiting superior FE. Compared with other ratio-based traits used to evaluate production efficiency—such as the feed conversion ratio (FCR) and FE, RFC offers statistical advantages as a selection criterion (Chapuis et al., 2017). Because RFC is often genetically weakly correlated with or nearly independent of its component traits, it allows for better selection responses. Numerous studies (Luiting and Urff, 1991c; Bordas et al., 1992; Basso et al., 2012; Zeng et al., 2018) have explored RFC-based selection in poultry and have highlighted RFC as a trait with moderate to high heritability that effectively reduces feed intake.

However, the underlying biological mechanisms and correlated responses to RFC selection can be complex and variable. A recent review by Ramankevich et al. (2025) emphasized that RFC is influenced by diverse factors, including physical activity, behavior, and metabolic rate, suggesting that selection outcomes may vary depending on the specific genetic and environmental context. For example, while they noted that low-RFC birds often exhibit reduced activity to conserve energy, Gao et al. (2025) found that low-RFC laying hens displayed enhanced immune and metabolic functions without compromising egg production. Given these varied physiological drivers and the lack of long-term data in indigenous laying ducks, it remains unclear whether continuous selection for RFC will eventually lead to trade-offs. Specifically, we question whether the accumulation of selection pressure might force a genetic antagonism between maintenance efficiency and laying performance.

In the present study, we investigated the direct and correlated responses to long-term (G0 to G9) RFC-based selection in Brown Tsaiya ducks. We hypothesized that while selection for low RFC would effectively improve FE traits, prolonged selection pressure might eventually induce unfavorable correlated responses in egg production traits due to physiological limits or genetic antagonisms. Our findings aim to clarify these long-term dynamics and inform future sustainable breeding strategies.

Materials and methods

Experimental birds and management

The Brown Tsaiya duck populations used in this study were the low RFC-selected line (high feed efficiency Brown Tsaiya, S line) and the Brown Tsaiya LRI 1 line maintained by randomly selecting breeder ducks (C line). These lines were established in 2009 at the Ilan Branch of the Taiwan Livestock Research Institute, located in Wujie Township, Yilan County, Taiwan (${24}^{\circ }{40}^{\prime }22.6"N,{121}^{\circ }{49}^{\prime }56.2"E$). G0 of the S line comprised individuals selected for low RFC from the 16th generation of Brown Tsaiya LRI 1 ducks, with the RFC measurement method developed by Liu et al. (2012). The remaining individuals were assigned to the C line. Details regarding the tested duck count, selected breeder ducks, selection percentage, and inbreeding coefficient for each generation and line are presented in Table 1. The animals, management methods, and experimental protocols used in this study were approved by the Institutional Animal Care and Use Committee (IACUC) of the Ilan Branch of the Livestock Research Institute (case numbers: LRI IACUC 104-012, LRIIL IACUC107004, LRIIL IACUC108003, and LRIIL IACUC109010). All experimental ducks were managed as per standard operating procedures.

Table 1.

Number of individuals tested, breeders selected, selection rate (%), and mean inbreeding coefficient per generation for the RFC selected line (S) and the control line (C) of Brown Tsaiya ducks.

Generat-ion.	Line	No. of individuals	No. of breeders	Selection rate (%)	Inbreeding coefficient
0		M = 157 F = 195	S: M = 10 F = 42	6.4 21.5	0 0
			C: M = 10 F = 40	6.8 26.1	0 0
1	S	M = 66 F = 119	M = 10 F = 42	15.2 35.3	0 0
	C	M = 78 F = 120	M = 11 F = 46	14.1 38.3	0 0
2	S	M = 70 F = 125	M = 12 F = 26	17.1 20.8	0 0
	C	M = 83 F = 123	M = 12 F = 61	14.5 49.6	0 0
3	S	M = 42 F = 46	M = 12 F = 38	28.6 82.6	0.033 0.037
	C	M = 84 F = 89	M = 12 F = 46	14.3 51.7	0.007 0.007
4	S	M = 65 F = 104	M = 11 F = 42	16.9 40.4	0.056 0.063
	C	M = 81 F = 118	M = 13 F = 46	16.0 39.0	0.008 0.011
5	S	M = 71 F = 136	M = 12 F = 40	16.9 29.4	0.070 0.069
	C	M = 82 F = 144	M = 12 F = 40	14.6 27.8	0.015 0.013
6	S	M = 78 F = 148	M = 12 F = 45	15.4 30.4	0.086 0.085
	C	M = 64 F = 94	M = 12 F = 46	18.8 48.9	0.025 0.023
7	S	M = 86 F = 126	M = 12 F = 45	14.0 35.7	0.098 0.098
	C	M = 74 F = 94	M = 12 F = 43	16.2 45.7	0.026 0.026
8	S	M = 80 F = 152	M = 12 F = 45	15.0 29.6	0.109 0.108
	C	M = 76 F = 113	M = 12 F = 43	15.8 38.1	0.038 0.041
9	S	M = 70 F = 151	M = 12 F = 43	17.1 28.5	0.139 0.138
	C	M = 86 F = 129	M = 12 F = 41	14.0 31.8	0.054 0.054

S, selected line; C, control line.

The S line was established through selection for low residual feed consumption.

To minimize the confounding effects of heat stress on RFC estimation, the breeding schedule was strategically managed to align the phenotyping period with cooler or moderate seasons. The founder population (G0) and G1 were hatched in November and April, respectively. For subsequent generations, a consistent seasonal schedule was adopted: G2 to G6 birds were hatched between June and July, placing their RFC testing period (34 to 37 wk of age) in the following spring (February to April). G7 to G9 birds were hatched in September, with the RFC test completed by May. Consequently, the critical data collection periods for the selected generations avoided the peak summer months (June to August). Environmental conditions were characterized using the Temperature-Humidity Index (THI), calculated as $THI=\left(1.8\times T_{db}+32\right)-\left(0.55-0.0055\times RH\right)\times \left(1.8\times T_{db}-26\right)$, where $T_{db}$ is the dry-bulb temperature (°C) and RH is the relative humidity (%). A typical annual THI profile for the experimental location, derived from the nearest weather station data (2019–2026), is presented in Supplementary Fig. S1 to illustrate the environmental context of the testing windows.

From hatching to 4 weeks of age, ducklings were reared in a brooding house with elevated stainless steel mesh floors and heating lamps, with a density of approximately 15 ducklings/m². After 4 weeks, they were moved to a rearing house with the same flooring and a 10/14 (light/dark)-h photoperiod, where the density was approximately 2 ducks/m². Regarding health management, birds were vaccinated against fowl cholera using a subunit vaccine (Formosa Biomedical Inc., Taiwan) at 28 and 56 days of age. No prophylactic antibiotics or other medications were administered throughout the experiment. At 12 weeks of age, they were transferred to a steel-structured duck house equipped with evaporative cooling pads and fans, housed in individual cages, and maintained under a 14/10 (light/dark)-h photoperiod. Each cage measured 30 cm in length, 33 cm in width, and 45 cm in height, providing a floor area of 990 cm² per duck. Nipple drinkers were installed, with two ducks sharing one nipple. Female ducks generally entered the laying period at approximately 16 weeks of age.

During the brooding period (0–4 weeks) and early rearing period (5–8 weeks), a starter diet containing 19.5% crude protein and 2,909 kcal/kg metabolizable energy was provided. During the late rearing period (8 weeks until 5% of all individuals began laying) and for mature drakes, a grower diet containing 13.5% crude protein and 2,660 kcal/kg metabolizable energy was provided. During the laying period (from when >5% of all individuals began laying until 52 weeks of age), a layer diet containing 20% crude protein and 2,712 kcal/kg metabolizable energy was provided. Feed and water were supplied ad libitum throughout the experimental period.

Trait testing and selection plan

We followed the test procedure reported in Liu et al. (2012). Individual RFC was measured over a 4-week period, when female ducks were 34 to 37 weeks old (232–259 days of age). During this period, 700 g of layer feed was provided every 3 to 4 days by using custom-made acrylic feeders designed to prevent FC by neighboring ducks. Individual actual FC was recorded every 3 or 4 days, and the cumulative amount over the 4-week period was calculated as total FC (g). Eggs were collected and weighed daily, and total EM (g) for the same period was obtained by summing daily values. BW was recorded at the beginning and end of the RFC test period to evaluate average BW (ABW, g) and change in BW (CBW, g). PFC was estimated using a multiple linear regression model (Bordas et al., 1992):

$$\mathbf{PF}{\mathbf{C}}_{\mathbf{i}}={\mathit{\beta }}_0+{\mathit{\beta }}_1\times {\mathbf{ABW}}_{\mathbf{i}}^{\mathbf{0}.\mathbf{5}}+{\mathit{\beta }}_2\times \mathbf{CB}{\mathbf{W}}_{\mathbf{i}}+{\mathit{\beta }}_3\times \mathbf{E}{\mathbf{M}}_{\mathbf{i}}$$

where PFC_i is the PFC for the i-th duck during the test period; β₀ is the intercept; and β₁, β₂, and β₃ are the regression coefficients for the square roots of ABW (ABW^0.5), CBW, and EM, respectively. Because the present study was not a divergent selection experiment and to avoid bias in the regression coefficients due to selection in the experimental line, the multiple regression equation and its coefficients were estimated annually by using data from C line ducks only. Subsequently, the resulting coefficients were applied to all individuals (both S and C lines) to calculate corresponding PFC values. Next, individual RFC was calculated as follows:

$$\mathbf{RF}{\mathbf{C}}_{\mathbf{i}}=\mathbf{F}{\mathbf{C}}_{\mathbf{i}}-\mathbf{PF}{\mathbf{C}}_{\mathbf{i}}$$

where RFC_i is the RFC (g) for the i-th duck during the test period, and FC_i is the actual total FC (g) during the same period. FCR was calculated as total FC (g) divided by total EM (g) for the same period: FCR = FC/EM (g/g).

In addition to RFC-related traits, other traits were recorded—for example, age at first egg (AFE, days), BW at 40 weeks of age (BW40, g), and numbers of eggs laid up to 40 (EN40) and 52 (EN52) weeks of age. Average egg weight at 40 weeks of age (EW40, g) and eggshell strength at 40 weeks of age (ES40, kg/cm²) were calculated as the mean of measurements obtained over 3 consecutive days.

From line establishment to G9, breeder ducks in the S line were primarily selected on the basis of their EBVs for RFC, with the aim being to reduce RFC. One drake was mated with approximately 3 to 5 hens to produce offspring for the subsequent generation. Matings between full sibs and half sibs were avoided. In the C line, breeder ducks were randomly selected each generation with avoidance of full-sib and half-sib matings.

Data preprocessing and statistical analysis

All collected trait data were first subjected to pedigree verification and correction. Specifically, the entire pedigree was traced to identify inconsistencies, which were then resolved by cross-referencing original paper breeding lists and hatching records. For traits with skewed distributions, to approximate normality, data shifting and Box–Cox transformation were performed when the absolute value of the Box–Cox parameter (λ) exceeded 0.5. To preserve the original biological properties of certain traits, ratio traits such as FCR were not transformed. In the transformed data sets, observations falling outside the range of the mean ± 3 standard deviations (SDs) or those deemed physiologically unreasonable (e.g., BW change exceeding 50% of the original BW) were considered outliers and treated as missing values. Phenotypic data were analyzed using a general linear model (GLM):

$${\mathbf{Y}}_{\mathit{ijk}}=\mathit{\mu }+{\mathbf{L}}_{\mathit{i}}+{\mathbf{G}}_{\mathit{j}}+\mathbf{L}{\mathbf{G}}_{\mathit{ij}}+{\mathbf{e}}_{\mathit{ijk}}$$

where Y_ijk is the observation of a trait for an individual in the i-th line and j-th generation, μ is the overall mean, L_i is the fixed effect of the i-th line (S or C), G_j is the fixed effect of the j-th generation, ${\left(LG\right)}_{ij}$ is the fixed interaction effect between line and generation, and e_ijk is the random residual error.

Genetic parameters (heritabilities, genetic variances, genetic correlations) and phenotypic correlations were estimated using the VCE software package (version 4.2.5) (Groeneveld and García-Cortés, 1998), with a multitrait animal model and the restricted maximum likelihood (REML) method applied. The animal model was as follows:

y = Xb + Za + e

where y is the vector of observations; b is the vector of fixed effects, primarily representing the fixed effect of generation in this model; a is the vector of random additive genetic effects for individuals, assuming a ∼ N(0, A${\sigma }_a^2$​), where A is the matrix of additive genetic relationships among individuals and ${\sigma }_a^2$​ is the additive genetic variance; and e is the vector of random residual effects, assuming e ∼ N(0, I${\sigma }_e^2$​), where I is an identity matrix and ${\sigma }_e^2$​ is the residual variance. X and Z are the corresponding design matrices for fixed and additive genetic effects, respectively. Because of pedigree connectivity between the two lines and contamination of S line individuals into the C line during G3 and G4, the fixed effect of line was excluded from the animal model to ensure convergence stability.

The EBV for each trait was calculated using the PEST program (version 4.2.3) (Groeneveld, 1990) with best linear unbiased prediction. To enable comparison of relative genetic improvements across traits, EBVs were standardized by dividing them by their respective genetic SDs (square root of the genetic variance). Genetic and phenotypic trends were assessed using mean EBVs and least squares mean of the line-generation interaction (reflecting the actual phenotypic performance in each generation), respectively. Selection response was evaluated as the between-line difference in EBVs in each generation. To explore the dynamic evolution of genetic parameters in the S line during long-term selection, we performed a sliding window analysis adapted from the study of Chen and Tixier-Boichard (2003). This analysis focused exclusively on S line data. A consecutive five-generation interval was used as the analytical window (e.g., G0–G4 and G1–G5), which was progressively advanced by one generation. This strategy was specifically implemented to increase the sample size to approximately 500 records per analysis, thereby mitigating the instability associated with small datasets in individual generations. Within each window, the same multitrait animal model and REML method were used to re-estimate heritabilities, additive genetic variances, and genetic correlations between RFC and other key traits. This method enabled detailed visualization of the progressive changes in genetic parameters as selection continued across generations. Sensitivity analyses using alternative windows (G6–G10 and G7–G11), and non-overlapping generation blocks (G5–G8 and G9–G12) are summarized in Supplementary Table S1. Because selection criteria were expanded after G10, the main manuscript reports results only through G9; estimates spanning G10 are provided for robustness only.

Results

Population overview and selection summary

From G0 to G9, 10–12 drakes were selected in each generation as breeders for both lines. For each selected sire, a full-sib or half-sib brother was retained as a spare sire. With the exceptions of the S line in G2 and G3, which comprised only 26 and 38 dams, respectively, and the C line in G2, which comprised 61 dams, the number of dams selected per generation in both lines ranged from approximately 37 to 46. When G0, G2, and G3 data were excluded, the selection rate for drakes in the S line generally ranged from 14% to 17%, whereas that for dams ranged from 26% to 44%. In G3 of the S line, the numbers of female ducks tested (46) and selected (38; selection rate: 82.6%) were lower than those in other generations, likely due to feed quality–related problems reducing the population size that year. For the C line, the selection rate for drakes ranged from 13% to 17%, whereas that for dams ranged from 32% to 41%. Population sizes for the S and C lines were maintained at approximately 150–250 and 150–200 individuals per generation, respectively.

Because pedigrees before G0 could not be verified, G0 individuals were assumed to be unrelated, with an inbreeding coefficient of 0. By G9, the mean inbreeding coefficient in the S line was 0.139 for drakes and 0.138 for hens, whereas that in the C line was 0.054 for both sexes. These results indicate that inbreeding increased more rapidly in the S line than in the C line.

Phenotypic trends

Fig. 1A–K presents the least square means of 11 traits for the S and C lines from G1 to G9, using a GLM including the fixed effects of line, generation, and their interaction (line × generation). The FC of the S line (Fig. 1A) was significantly lower than that of the C line in G6, G8, and G9. The RFC of the S line (Fig. 1B) was also significantly lower than that of the C line G4 onward, indicating that the between-line differences in feed intake emerged early in selection and persisted through G9.

Fig. 1.

Least square mean values of 11 traits in the selected and control lines from G1 to G9. (A–D) RFC-related traits. (E and F) Weight traits. (G–K) Egg traits. The least square mean ± standard error of the mean values of 11 traits in the selected line (■) and the control line (□) are presented. The asterisk indicates significant differences (P < 0.05) between two lines within the same generation. RFC: residual feed consumption.

The EM of the S line (Fig. 1C) did not differ significantly or was slightly higher than that of the C line before G4. However, it exhibited a decreasing trend from G5 to G8, becoming significantly lower than that of the C line in G8 and G9. The FCR (Fig. 1D) of the S line was consistently lower than that of the C line. The ABW (Fig. 1E) of the S line was generally higher than that of the C line before G6, but this difference became less pronounced thereafter. CBW (Fig. 1F) exhibited greater fluctuations between the two lines, with no significant differences in most generations. AFE (Fig. 1G), EN40 (Fig. 1H), and EN52 (Fig. 1I) exhibited no significant between-line differences across generations. The EW40 (Fig. 1J) of the S line was slightly higher than that of the C line before G4 but became lower in G8 and G9, with significant differences observed in G9. ES40 (Fig. 1K) exhibited no clear consistent trend between the two lines.

Estimates of genetic parameters from G0 to G9 generations

The heritabilities (h²), genetic correlations (r_g), and phenotypic correlations (r_p) for 11 traits, estimated from the combined data of both lines from G0 to G9, are presented in Fig. 2. Heritability estimates were categorized as high (h² > 0.4), moderate (0.2 ≤ h² ≤ 0.4), or low (h² < 0.2) (Bourdon, 2014). Traits in decreasing order of heritability estimates were ABW (h² = 0.706), AFE (h² = 0.524), FC (h² = 0.481), EW40 (h² = 0.471), RFC (h² = 0.340), EN40 (h² = 0.333), EN52 (h² = 0.281), EM (h² = 0.260), ES40 (h² = 0.246), FCR (h² = 0.192), and CBW (h² = 0.050), respectively (Fig. 2).

Fig. 2.

Heatmap depicting genetic correlations (upper-right triangle), heritabilities (diagonal), and phenotypic correlations (lower-left triangle) between 11 traits estimated from G0–G9 of the selected and control lines of Brown Tsaiya ducks. Values in the second row represent corresponding standard errors, with asterisks indicating significant phenotypic correlations (*: P < 0.05; **: P < 0.01; ***: P < 0.001). The color scale indicates the level and direction of correlations. FC: feed Consumption; RFC: residual feed consumption; EM: egg mass; FCR: feed conversion ratio; ABW: average body weight; CBW: change in body weight; AFE: age at first egg; EN40 and EN52: numbers of eggs laid up to 40 and 52 weeks of age, respectively; EW40: egg weight at 40 weeks of age; ES40: eggshell strength at 40 weeks of age.

Phenotypic and genetic correlations between production traits varied in magnitude (Fig. 2). Correlations were considered high when the absolute value exceeded 0.5, moderate when it was between 0.3 and 0.5, low when it was between 0.1 and 0.3, and negligible when it was less than 0.1. Regarding RFC-related traits, highly positive genetic and phenotypic correlations were observed between RFC and FC (r_g = 0.760, r_p = 0.683) and between RFC and FCR (r_g = 0.763, r_p = 0.487). In contrast, RFC exhibited no significant correlations with most egg production traits (EM, AFE and EN) in overall population (r_g ranging from -0.049 to 0.089). However, a low positive genetic correlation was noted between RFC and EW40 (r_g = 0.210). Furthermore, genetic analysis revealed that EM was more strongly correlated with EW40 (r_g = 0.768) than with egg number (EN40: r_g = 0.148; EN52: r_g = 0.316), suggesting that the observed variance in EM is primarily driven by egg weight.

Notable discrepancies were observed between the genetic and phenotypic correlations of RFC with weight traits. For example, RFC exhibited a moderate negative genetic correlation with CBW (r_g = −0.521) despite a negligible phenotypic correlation (r_p = 0.005). Similarly, RFC showed a low positive genetic correlation with ABW (r_g = 0.121) but a near-zero phenotypic correlation (r_p = −0.037). Detailed correlation estimates for all trait pairs are provided in Fig. 2.

Genetic trends and selection responses from G0 to G9

Genetic parameters were compared across selection stages and lines. Fig. 3 presents heatmaps depicting heritabilities and genetic correlations for four traits (RFC, FC, EM, and FCR), estimated from data of the overall population, the S line, and the C line during G0–G8. The additive genetic variance (${\sigma }_G^2$) for RFC in each group is also presented. During G0–G8, data from the overall population and the C line indicated that RFC had high positive genetic correlations with FC (r_g = 0.734–0.779) and FCR (r_g = 0.822–0.945) but a low positive or slightly negative genetic correlation with EM (r_g = −0.205 to 0.113). This trend was consistent with the combined G0–G9 data (Fig. 2), suggesting that when selection pressure was weak or in early generations, the indirect genetic effect of improved RFC on EM was small. In the C line, selection for low RFC might have exerted a slightly positive genetic effect on EM. However, in G0–G8 of the S line (Fig. 3), the genetic correlation between RFC and EM was highly positive (r_g = 0.566), and the additive genetic variance for RFC (${\sigma }_G^2$ = 71,942) was markedly lower than those of the overall population (${\sigma }_G^2$ = 214,085) and the C line (${\sigma }_G^2$ = 253,791).

Fig. 3.

Comparative heatmaps of heritabilities (diagonal) and genetic correlations (upper-diagonal) among four RFC-related traits from G0-G8 for different data groups (the overall population, selected line [S], and control line [C]) in Brown Tsaiya ducks. The additive genetic variance for RFC is presented under each graph. Values in the second row represent corresponding standard errors. The color scale indicates the level and direction of correlations. FC: feed Consumption; RFC: residual feed consumption; EM: egg mass; FCR: feed conversion ratio.

To assess relative genetic changes in various traits during selection, we calculated mean EBVs for each line and generation and standardized these values by dividing them by the genetic SD of the respective trait. The standardized genetic trends are presented in Fig. 4, with dashed lines representing the C line and solid lines representing the S line. Standardized genetic trends in the C line suggested that all traits fluctuated minimally around zero from G0 to G9, with variation generally within 0.5 genetic SDs. This finding indicates that the C line remained genetically stable, serving as a reliable reference for assessing genetic progress in the S line. By contrast, the S line exhibited significant genetic improvements in FE-related traits. The standardized EBVs for RFC, FC, and FCR in the S line exhibited a distinct downward trend (Fig. 4), indicating that selection for low RFC effectively reduced RFC and FC while improving FCR at the genetic level. Furthermore, the divergence between the S and C lines widened across generations, suggesting that the selection response had not yet plateaued.

Fig. 4.

Standardized genetic trends (mean estimated breeding value/genetic standard deviation of the trait) for 11 traits in the selected line (solid lines) and the control line (dashed lines) of Brown Tsaiya ducks from G0 to G9. FC: feed Consumption; RFC: residual feed consumption; EM: egg mass; FCR: feed conversion ratio; ABW: average body weight; CBW: change in body weight; AFE: age at first egg; EN40 and EN52: numbers of eggs laid up to 40 and 52 weeks of age, respectively; EW40: egg weight at 40 weeks of age; ES40: eggshell strength at 40 weeks of age.

Standardized genetic trends in the S line suggested that most traits exhibited changes similar to those in the C line, fluctuating within approximately 0.5 SDs. However, some traits exhibited correlated genetic changes. The standardized genetic trend for CBW in the S line exhibited considerable fluctuations across generations, with a pronounced positive shift after G6, compared with the relatively stable trends observed in other traits. However, the standardized EBV for EM in the S line exhibited a continuous downward trend after G8. In G9, the standardized EBV for RFC in the S line decreased by approximately −2.0 SD units compared with its G0 value, whereas that for EM decreased by approximately −0.52 SD units. Therefore, on average, every 1 SD unit reduction in RFC in the S line corresponded to a 0.26 SD unit reduction in EM.

Further analysis of RFC-related genetic parameters in the S line by using a five-generation sliding window (Table 2) unveiled strong selection effects on RFC in the early stages (e.g., G0–G4), with marked declines in its heritability and genetic variance. During G2–G6, RFC in the S line exhibited highly positive genetic correlations with FC (r_g = 0.972) and EM (r_g = 0.797) and a highly negative correlation with FCR (r_g = −0.516). Only sliding windows with estimable standard errors are reported in Table 2; additional sensitivity analyses are summarized in Supplementary Table S1. Across these sensitivity checks, RFC heritability and additive genetic variance showed a consistent tendency to be higher in later windows/blocks; estimates spanning G10 are not used for primary inference.

Table 2.

Estimates of h²; ${\mathit{\sigma }}_{\mathit{G}}^2$ for RFC; and RFC’s genetic correlations with FC, EM, and FCR in a selected line of Brown Tsaiya ducks, obtained using a five-generation sliding window.

Generation	RFC h2	SE h2*	${\sigma }_G^2$	${\mathrm{r}}_g$ of FC	${\mathrm{r}}_g$ of EM	${\mathrm{r}}_g$ of FCR	No. records
G0-4	0.120	0.062	71,766	0.436	0.312	-0.253	470
G1-5	0.129	0.068	74,467	0.769	0.860	-0.928	486
G2-6	0.109	0.042	74,779	0.972	0.797	-0.516	528
G3-7	0.119	0.041	79,224	0.956	0.681	0.055	519

h², heritability; SE h², standard error of heritability; ${\sigma }_G^2$, additive genetic variance; RFC, residual feed consumption; FC, feed consumption; ${\mathrm{r}}_g$, genetic correlation of RFC with FC; EM, egg mass; FCR, feed conversion ratio; No. records, number of records used for estimation.

Above findings indicate that RFC-based selection was effective and that continued selection for low RFC in the S line led to a decline in EM at the genetic level, consistent with the G0–G9 genetic trends.

Comparative Phenotypic and Genetic Trends

Fig. 5 presents both phenotypic (solid lines) and genetic (dotted lines) trends in true units of all 11 traits in the S line (red lines) and C line (blue lines) of Brown Tsaiya ducks from G0 to G9, comprehensively depicting selection progress across generations.

Fig. 5.

Phenotypic trends (solid lines) and genetic trends (dotted lines) for 11 traits in the selected line (S, red lines) and the control line (C, blue lines) of Brown Tsaiya ducks from G0 to G9.

The phenotypic values of RFC in the S line fluctuated during early generations but became significantly lower than those in the C line from G4 onward (Fig. 5B), and this difference persisted through G9. EBVs also exhibited a continuous downward trend, indicating a significant genetic improvement in RFC. Similar trends were observed for FC (Fig. 5A) and FCR (Fig. 5D), with both phenotypic values and EBVs of the S line being significantly lower than those of the C line, confirming effective improvement in FE.

The phenotypic values of EM in the S line fluctuated during early generations but declined after G5, becoming lower than those of the C line in G8 and G9 (Fig. 5C). Concurrently, EBVs exhibited a downward trend, particularly after G8. Therefore, single-trait selection for low RFC induced an unfavorable correlated genetic response in EM.

The phenotypic values of ABW in the S line were slightly higher than those in the C line before G6 (Fig. 5E), with less distinct differences thereafter; EBVs remained relatively stable between the two lines. CBW exhibited large between-line differences in both phenotypic and genetic trends (Fig. 5F), with no clear consistent trends.

For AFE (Fig. 5G), EN40 (Fig. 5H), and EN52 (Fig. 5I), neither phenotypic nor genetic trends exhibited significant or consistent between-line differences across generations, indicating limited effects of single-trait selection for low RFC on these traits.

The phenotypic values of EW40 were slightly higher in the S line before G4 but became significantly lower than those of the C line in G8 and G9 (Fig. 5J). The EBVs of EW40 exhibited a decreasing trend similar to that of EM. ES40 (Fig. 5K) exhibited no clear or consistent phenotypic or genetic differences between the two lines.

Discussion

We comprehensively investigated the direct and correlated responses to long-term selection for low RFC on key production traits in the Brown Tsaiya duck. Our results indicate that single-trait selection for low RFC from G0 to G9 effectively reduced FC and improved FCR.

Effectiveness of RFC-based selection

For RFC estimation, different formulas and populations are often used depending on the species, line, and purpose. In the present study, PFC was estimated using a multiple regression model incorporating ABW^0.5, CBW, and EM. This method is similar to that proposed by Byerly in 1941 for restricting hen FC, as cited by Basso et al. (2012). It is also similar to the method developed by Bordas et al. (1992). To estimate maintenance requirements, most studies use a specific power of BW. Luiting and Urff (1991a) used a power of 0.75, whereas Byerly et al. (1980) used both 0.653 and 0.75 and identified no significant difference between them. Bordas and Mérat (1984) used a power of 0.5 and reported similar outcomes across the range of 0.5–1. Therefore, following the study of Liu et al. (2012), we used a power of 0.5 to estimate maintenance requirements.

Regarding the population used for estimation, in a divergent RFC-based selection experiment by Bordas et al. (1992), the R+ and R− lines of the same year were used for a combined multiple regression analysis, with different formulas for drakes and hens. In a selection experiment by Schulman et al. (1994), which included an S line and a C line, a single multiple regression equation covering the entire laying period (16–42 weeks of age) was used for the base population before separation for PFC estimation. In subsequent generations, the laying period was divided into seven subperiods, and regression coefficients for each period were calculated from the base population and applied to the next four generations. Although the present study also included an S line and a C line, regression coefficients were initially estimated from both lines annually, following Bordas et al. (1992). However, inclusion of the S line might have introduced bias to the regression coefficients due to selection responses. Therefore, to ensure unbiased estimation, we established the multiple regression equation and estimated the partial regression coefficients annually by using data solely from the C line. These coefficients were then applied to both S and C lines to evaluate PFC and RFC. Given the stability of genetic trends for RFC and related traits in the C line, this approach seemed suitable for our experiment. However, it implies that the phenotypic independence between RFC and its component traits is mathematically guaranteed only within the C line, which likely accounts for the non-zero correlations observed in the S line.

To measure RFC, a 4-week test period from 34 to 37 weeks of age (232–259 days of age; total of 28 days) was adopted. Liu et al. (2012) reported that RFC for this period exhibited a high genetic correlation with that for the entire period (r_g = 0.93–1.00). Basso et al. (2012) indicated that to ensure reliability, RFC should be measured for >2 weeks, particularly during a stable laying period. Luiting and Urff (1991b) suggested measuring RFC between 32 and 56 weeks of age with at least a 4-week recording period and highlighted that genetic factors influencing RFC differ between early and late laying periods. Bordas and Mérat (1975) compared long-term (3 months) and short-term (16 days) measurements and discovered that after adjustment for BW and egg production, approximately 2 weeks of adjusted FC provided reliable estimates. Sabri et al. (1991) compared test durations of 2, 4, and 8 weeks and discovered that although the 8-week period yielded the highest heritability, the 4-week period yielded the narrowest range of heritability estimates. Therefore, the use of a 4-week test period in this study is reasonable.

We acknowledge that unintentional contamination of S-line individuals into the C line occurred during G3 and G4. As a result, the fixed line effect was excluded from the animal model. This contamination may have artificially lowered the C line's average RFC, potentially leading to an underestimation of the true selection response (S–C difference) in our phenotypic trends (Fig. 1, 5). However, this issue does not invalidate our central finding of genetic antagonism. Our sliding-window analysis (Table 2), which used exclusively S-line data and was independent of the C-line, confirmed a strong positive genetic correlation (e.g., r_g = 0.860 in G1–G5) between RFC and EM. This strongly supports that the antagonism is a genuine biological response to selection, not a statistical artifact from pedigree mixing.

Direct response

As an indicator of FE, RFC has gained increasing attention in the literature because it is phenotypically independent of its component production traits. In this study, the heritability of RFC estimated from the combined data of G0–G9 was 0.340 (Fig. 2). This moderate heritability provides the genetic basis for selection, indicating that a substantial portion of the variation in feed efficiency is additive and can be effectively exploited. This estimate aligns with those reported previously: 0.30–0.43 in G0 of Brown Tsaiya ducks (Liu et al., 2012), 0.42–0.62 in laying hens (Luiting and Urff, 1991c), 0.24 in laying Pekin ducks (Basso et al., 2012), 0.41 in Pekin ducks (Zhang et al., 2017), 0.27 in Rhode Island Red chickens (Tixier-Boichard et al., 1995), and 0.46 in laying hens (Schulman et al., 1994). These consistent findings across species support the moderate heritability of RFC and its suitability as a selection criterion.

Regarding the selection response, in our study, RFC differed consistently and significantly between the S and C lines after G4. By G9, the standardized EBV of RFC in the S line exceeded (by >2.0 SDs) that of RFC in the C line, indicating a significant selection effect with no evidence of a plateau (Fig. 4). Biologically, this continuous improvement in feed efficiency implies a successful reduction in the energy required for maintenance processes. Recent physiological studies in laying hens by Gao et al. (2025) suggest that such improvements are likely driven by enhanced hepatic metabolic efficiency and optimized immune resource allocation, rather than merely passive changes in body composition. This direct selection response exhibited a trend similar to the long-term responses reported by Bordas et al. (1992), where the phenotypic difference exceeded 2 SDs after 14 generations, and Zerjal et al. (2021), who reported a difference of approximately 5 SDs after 40 generations. Our results, supported by these studies, demonstrate the strong and sustained potential of RFC-based selection.

Correlated responses

In the present study, the genetic evaluation from G0–G9 revealed highly positive genetic correlations between RFC and both FC (r_g = 0.760) and FCR (r_g = 0.763) (Fig. 2). Furthermore, standardized genetic trends (Fig. 4) indicated that both RFC and total FC in the S line exhibited significant gradual declines at the genetic level, confirming the effectiveness of selection for low RFC. These strong associations confirm that RFC constitutes a substantial component of the variation in total feed intake. The results align with prior findings in other Poultry. For example, in Pekin ducks, Basso et al. (2012) found a very high positive genetic correlation between RFC and FC (r_g = 0.89). In Shaoxing and Jinyun ducks, Zeng et al. (2018) noted genetic correlations between RFC and FC (r_g = 0.79 and 0.86, respectively) and between RFC and FCR (r_g = 0.47 and 0.63, respectively). Long-term selection responses reported by Katle and Kolstad (1991), Bordas and Mérat (1984), and Zerjal et al. (2021) further support this trend, showing that R− lines consumed less feed (e.g., 6%–8% lower or even half the consumption after 40-year selection) than control or the R+ lines. Collectively, our findings and the literature confirm that selection for low RFC is a robust strategy to improve overall FCR and reduce FC.

In this study, the S and C lines did not differ significantly in AFE, egg number, or ES40, consistent with the negligible genetic correlations between RFC and these egg production traits (Fig. 1, Fig. 2) and with the literature. However, EM was lower in the S line than in the C line in G8 and G9, and EW40 was lower in G9. When we analyzed combined data from both lines across G0–G9, the genetic correlation between RFC and EM was close to zero (r_g = 0.089; Fig. 2). However, when a separate genetic analysis was performed for the S line alone over G0–G8, a moderate to high positive genetic correlation was noted between RFC and EM (r_g = 0.566; Fig. 3). Furthermore, the sliding-window analysis (Table 2) indicates that this antagonism persisted across overlapping converged windows, with consistently positive rg between RFC and EM (e.g., r_g = 0.860 in G1–G5 and r_g = 0.681 in G3–G7). Sensitivity analyses (Supplementary Table S1) provide converged estimates in later windows/blocks and support the robustness of the observed patterns; however, results spanning G10 are presented for robustness only because selection criteria changed after G10.

By definition, RFC is considered independent of other production traits. Most studies have reported low genetic correlations between RFC and laying traits such as EM, egg weight, and egg number. For example, (Schulman et al., 1994) found no genetic change in EM, AFE, egg number, egg weight, or CBW with altered RFC. Tixier-Boichard et al. (1995) reported low genetic correlations between RFC and egg number (r_g = 0.11), egg weight (r_g = −0.03), and AFE (r_g = −0.21). Luiting and Urff (1991c) estimated that the genetic correlations of RFC with metabolic BW, EM, and CBW were not significantly different from zero. In an experiment involving F2 cross between White Leghorn and a local chicken breed, Yuan et al. (2015) noted that both phenotypic and genetic correlations between RFC and EM were close to zero. Similarly, in an F2 population derived from Shaoxing and Jinyun ducks, Zeng et al. (2018) discovered that the phenotypic and genetic correlations of RFC with BW, CBW, and EM were close to zero. These findings support the prevailing view that selection for low RFC does not adversely affect egg production traits.

However, some findings of long-term selection differ from those of prior studies. For example, Bordas and Mérat (1984) found that the R− line had significantly thicker eggshells and higher eggshell strength but lower egg weight than did the R+ line in five of six generations after 1978; moreover, EM exhibited a decreasing trend compared with the result in the R+ line. Bordas et al. (1992) reported that after 14 generations of divergent selection, EM and egg weight were significantly higher in the R+ line than in the R− line; the authors attributed these inconsistent changes across generations partly to random genetic drift accumulation. In brown-egg layers, Hagger (1994) noted moderately positive genetic correlations between RFC and both EM (r_g = 0.306) and egg number (r_g = 0.276). In Pekin ducks, Basso et al. (2012) noted a highly positive genetic correlation between RFC and daily EM (r_g = 0.57). These findings indicate that selection for low RFC genetically reduces EM or egg number. Therefore, long-term selection for low RFC adversely affected EM in the Brown Tsaiya duck at both the phenotypic and genetic levels. This persistent positive correlation underscores the challenge of long-term single-trait selection.

In our study, RFC exhibited nearly no phenotypic correlation with ABW and CBW but a low positive genetic correlation with ABW (r_g = 0.121). In the S line, RFC exhibited a highly negative genetic correlation with CBW (r_g = −0.521), suggesting that the individuals genetically tend to lose less weight or gain more weight. However, given the very low heritability of CBW (h² = 0.050), no significant phenotypic difference was observed between our S and C lines.

Several studies (Bordas et al., 1992; Luiting and Urff, 1991c; Schulman et al., 1994; Tixier-Boichard et al., 1995) have indicated that selection for low RFC did not lead to significant phenotypic or genetic changes in BW or CBW. However, Basso et al. (2012) reported that in laying Pekin ducks, RFC exhibited a low positive genetic correlation with BW (r_g = 0.12) but a high negative genetic correlation with CBW (r_g = −0.65), and CBW exhibited low heritability (h² = 0.09). In Shaoxing and Jinyun laying ducks, Zeng et al. (2018) reported the coefficients of genetic correlations between RFC and BW to be −0.09 and −0.11 and those of genetic correlations between RFC and CBW to be −0.22 and −0.13, respectively; CBW heritabilities were 0.03 and −0.02, respectively. Our study is consistent with these two studies.

To elucidate the mechanisms underlying differences in RFC, many studies have compared physiological factors and energy partitioning between R+ and R− lines. Differences in thermogenesis may partially explain this trend. Gabarrou et al. (1997) suggested that R+ chickens dissipate excess ingested energy through high diet-induced thermogenesis, which results in a leaner phenotype. Morphologically, R+ lines often have larger combs and wattles and higher shank temperatures than do R− lines (Bordas and Mérat, 1981). Luiting et al. (1991d) noted that compared with R− chickens, R+ chickens have poor feather quality, which promotes heat dissipation. Conversely, R− individuals produce less heat, which allows surplus energy to be stored as fat, consistent with the finding of El-Kazzi et al. (1995), who reported greater fat deposition in the R− line than in the R+ line. However, some studies involving meat-type chickens and ducks have demonstrated that selection for low RFC results in a small or no significant difference in abdominal fat content (Bai et al., 2022; Emamgholi Begli et al., 2017).

Activity level is another major determinant of energy expenditure. Our findings from the Brown Tsaiya duck experiment, along with those of other studies, indicate that individuals from RFC-selected line tend to be less active, calmer, and less responsive to stress than the ones from the control line. For example, Braastad and Katle (1989) found that R− chickens spent more time resting, whereas R+ chickens exhibited more pacing and pecking behaviors. Bezerra et al. (2013) reported that R− animals are less sensitive to stress than are R+ animals. Luiting et al. (1991d) revealed that activity-related heat production accounted for 29%–54% of the difference in total heat production between the R+ and R- chickens, suggesting that behavioral activity exerts a large effect on RFC.

The aforementioned physiological and behavioral differences may be traced to endocrine and blood biochemical regulation. Gabarrou et al. (2000) reported that although R+ roosters had a relatively large appetite, they exhibited lower plasma insulin levels. Additionally, plasma triglyceride level was significantly lower in the R+ line than in the R− line, whereas that of triiodothyronine (T3), which is related to thermogenesis, was higher in the R+ line, facilitating the dissipation of excess ingested energy as heat. Yang et al. (2020) reported that R+ chickens had lower plasma levels of growth-promoting insulin-like growth factor-1 but higher plasma levels of T3 and cortisol, which are associated with metabolism and stress, than did R− chickens. These findings suggest that R− chickens conserve energy by reducing T3 production and thermogenesis. Therefore, the low RFC individuals exhibit distinct patterns of energy acquisition and allocation (thermogenesis, fat storage, and activity level) than the high RFC or control line.

Nevertheless, findings regarding EM, abdominal fat content, blood biochemical parameters, and immune function remain inconsistent, indicating that the correlations between RFC and these traits are complex and influenced by factors such as species, genetic line, and selection duration.

Strategic reflection on selection outcomes and future directions

In our study, selection from G0 to G9 successfully reduced RFC. However, RFC exhibited an antagonistic genetic correlation with EM. For the Brown Tsaiya duck, maintaining strong laying performance is as crucial as improving FE. Therefore, adjusting the selection strategy to mitigate this antagonism is imperative.

Previous breeding programs for the Brown Tsaiya LRI 1 line, as reported by Cheng et al. (1996), have faced challenges in achieving a balanced improvement across traits and successfully adopted a restricted genetic selection index to meet breeding objectives (Chen et al., 2003).

Findings from G0 to G9 indicated that continued single-trait selection for low RFC adversely affected the overall production performance of Brown Tsaiya ducks. Therefore, future breeding strategies for this breed may adopt a multitrait selection approach, such as a restricted genetic selection index, to achieve a balanced improvement between FE and laying traits. This approach would support comprehensive improvement and the sustainable use of the Brown Tsaiya duck genetic resource. Finally, although the Brown Tsaiya duck is a regional breed, these findings have broader implications for the global poultry industry. Our results demonstrate that while RFC selection is effective, it carries the risk of unfavorable genetic correlations with reproductive traits over the long term. This underscores the necessity of regular genetic evaluation in any breeding program to timely identify and address such antagonisms, ensuring balanced improvements in both feed efficiency and production performance.

Parallel to optimizing selection goals, maintaining genetic diversity is equally vital. We acknowledge the concern regarding the accumulation of inbreeding. Given practical constraints on facility capacity that limit population expansion, we have implemented cross-generation genetic monitoring (Chang et al., 2023) to check the fluctuation of the genetic diversities . To ensure long-term sustainability, we plan to implement Optimum Contribution Selection (OCS) in future generations to optimally balance genetic gain with the restriction of inbreeding rates within the existing population structure.

Conclusion

Our long-term selection experiment demonstrated that nine generations of single-trait selection for low RFC effectively improved FE in Brown Tsaiya ducks, as indicated by consistent reductions in RFC, FC, and FCR at both phenotypic and genetic levels. The absence of a plateau in the genetic trend suggests that further improvement is achievable. However, the selection also resulted in an unfavorable correlated response in EM, highlighting a trade-off between FE and reproductive performance. Therefore, future breeding programs for Brown Tsaiya ducks should integrate multitrait or restricted selection indices to balance FE with key reproductive and production traits, ensuring sustainable genetic progress and long-term productivity of this vital indigenous breed.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this work the author used Gemini (an AI-assisted large language model by Google) in order to review and refine the manuscript for clarity, consistency, and scientific rigor. After using this tool/service, the author reviewed and edited the content as needed and takes full responsibility for the content of the published article.

CRediT authorship contribution statement

Yi-Ying Chang: Writing – original draft, Visualization, Resources, Methodology, Investigation, Formal analysis, Data curation. Chih-Feng Chen: Writing – review & editing, Supervision, Project administration, Methodology, Funding acquisition, Conceptualization.

Disclosures

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.