The experiment was carried out during the late spring season of 2020 on one commercial farm affiliated with the University of Perugia (Italy) for the project. Chickens were reared according to EU Regulation 834/07, EU Regulation 889/2008, and the Italian directives (European Parliament and Council of the European Union, 2013) on animal welfare for experimental and other scientific purposes. The experimental protocol was positively evaluated and approved by the Ethical Committee of the Istituto Zooprofilattico dell’Umbria e delle Marche (ID number: 62705 of 07/15/2020).

A total of 200 chickens belonging to four different slow-growing genotypes (SGs; 25 chickens x 2 replicates x genotype), widely diffused in Europe for meat production, were used. In particular, the genotypes were Red JA57 (RJ), Naked Neck (NN), Lohmann Dual meat-type (LD), and an Italian crossbreed (Robusta Maculata x Sassò, CB). The birds were provided by three commercial breeding companies: RJ by Aviagen (Cocconato, AT, Italy), NN by Hubbard (Le Foeil-Quintin, France), and LD by Lohmann Breeders (Avizoo Lohmann, Italy); CB was directly developed by the Department of Agricultural, Environmental and Food Science of the University of Perugia (Italy).

Two replications of each chicken strain were reared in different outdoor pens, which were also equipped with a shelter. The indoor (0.10 m²/bird) and outdoor (4 m²/bird) densities were specified according to organic regulations (EC Regulation No. 834/2007 and 889/2008). From 1 to 15 d, birds were housed in an environmentally controlled poultry house with a temperature range of 30–32°C (with the presence of heat lamps) and relative humidity of 65-70%; subsequently, the lamps were removed and the temperature was maintained at approximately 20-22°C with the same relative humidity. At 36 d, the chickens were provided with free access to the outdoor space. The mean outdoor temperature and humidity were 19 ± 7°C and 50.1 ± 12.5%, respectively. As required, the pasture was never treated with pesticides or herbicides.

The animals were fed ad libitum with the same diet (starter feed for 1–21 d, grower feed from 22 d to slaughter; Table 1 ); the diets were formulated to satisfy the requirements recommended by the breeding companies. Water was always available, and the birds were kept in shelters only during the night to protect them from predators. Once a week, 15 chickens per replicate from all genotypes were weighed to evaluate the daily weight gain (DWG, g/d/chicken). Mortality was recorded daily.

Behavioral observations were performed in all pens using a computerized system (Noldus Technology, Wageningen, Netherlands) comprising two different software programs, Media Recorder and Observer XT, to record and analyze the videos. All the pens were inspected to identify the best position for the camera and the best time of day for video recording. As the birds were allowed access to the outdoor area at 36 d, video acquisition started at 42 d. The period from 36 to 42 d was considered an adaptation period for the birds to the outdoor area.

Two videos/week of 2 hours (h) length (9.00–11.00 AM) were recorded in each pen. Two expert observers analyzed all the videos in poultry behavior using the reported ethogram ( Table 2 ). Each video was analyzed by two expert observers in poultry behavior using a 10 min scan interval (6 scans per hour; 12 scans per video) and the reported ethogram ( Table 2 ), following the method of Cartoni Mancinelli et al. (2023).

At 81 d, 15 animals/genotype were subjected to plumage evaluation. The observation focused on five parts of the body: neck, breast, back, wings, and tail. According to the Tauson method (Tauson, 1984), a range of scores between 0 (no feathering) and 4 (intact and perfect plumage) was used. The presence of sternal hock burn and footpad dermatitis was also assessed using a scale from 0 to 2, where 0 indicated absence, 1 indicated light lesions, and 2 indicated deep lesions (Berg, 1998).

At 81 d, 15 chickens/genotype (total n=60) were randomly selected and slaughtered in a commercial slaughterhouse after feed withdrawal. The animals were electrically stunned (110 V; 350 Hz) before being killed. A 2 mL blood sample was collected in heparinized tubes for plasma extraction to assess in vivo oxidative status, while another 2 mL blood sample was collected in plain tubes for serum extraction to determine fatty acid and immunity traits.

After collection, the blood samples were immediately sent to the laboratory of the Department of Agricultural, Food and Environmental Science, University of Perugia, where the plasma tubes were centrifuged at 1,500 × g for 10 min at +4°C, and the serum tubes were left to separate for 2 h at room temperature. Finally, the plasma and serum samples were frozen at −80°C until analysis.

After bleeding, the carcasses were placed in hot water (56.5°C for 1 min), then plucked and eviscerated (non-edible viscera: intestines, proventriculus, gall bladder, spleen, esophagus, and full crop), and stored for 24 h at 4°C to obtain the cold carcass. The head and feet yield (% cold carcass weight including head, neck, and feet relative to live weight), carcass weight, and yield (% cold carcass without head, neck, and feet relative to live weight), as well as the breast and drumstick weights, were calculated. The breast and drumstick were sampled and stored at −20°C for 1 week. The drumstick meat was entirely removed from the bone.

The extent of plasma lipid peroxidation was evaluated using the method detailed in Mattioli et al. (2019). A spectrophotometer (set at 532 nm; Shimadzu Corporation UV-2550, Kyoto, Japan) measured the absorbance of thiobarbituric acid reactive substances (TBARS), and a tetraethoxypropane calibration curve in sodium acetate buffer (pH=3.5) was used. The results were expressed as nmol of malondialdehyde (MDA)/mL of plasma.

The detection of protein carbonyl groups followed the method of Dalle-Donne et al. (2003), using 2,4-dinitrofenilhidrazina (DNPH) as the reagent. The serum was diluted 1:40 with phosphate-buffered saline (PBS) before analysis. Carbonyl content was determined from the absorbance at 366 nm using a molar absorption coefficient of 22,000 M^-1/cm, with the same spectrophotometer previously described. The results were expressed as nmol/mg of protein. Protein was quantified using Coomassie Brilliant Blue R-250 staining solution (Bio-Rad Laboratories, Inc., Milan, Italy), following the Bradford method (1976).

The tocols (α-tocopherol and its isoforms γ and δ, and α and γ-tocotrienol) and retinol levels were measured according to Schüep and Rettenmaier (1994). Briefly, 0.2 mL of plasma was mixed with 1 mL of water and 4 mL of an ethanol solution of 0.06% butylated hydroxytoluene (BHT). The mixture was saponified with water/potassium hydroxide (KOH, 60%) at 70°C for 30 min and extracted with hexane/ethyl acetate (9/1, v/v). After centrifugation, 2 mL of the supernatant was transferred into a glass tube, dried under N₂, and re-suspended in 200 μL of acetonitrile. The pellet was re-extracted twice using the same method. A 50 μL volume of filtrate was then injected into the HPLC/FD system (pump model Perkin Elmer Series 200, equipped with an autosampler system, model AS 950-10, Jasco, Tokyo, Japan) on a Sinergy Hydro-RP column (4 µm, 4.6 × 100 mm; Phenomenex, Bologna, Italy) with a flow rate of 2 mL/min.

All tocopherols and tocotrienols were identified using an FD detector (model Jasco FP-1525; excitation and emission wavelengths of 295 and 328 nm, respectively) and quantified using external calibration curves prepared with increasing amounts of pure standard solutions (Sigma-Aldrich, Bornem, Belgium) in ethanol. The tocols sum was used for statistical analysis.

Retinol was analyzed using the same HPLC system with a UV-VIS spectrophotometer detector (Jasco UV2075 Plus) set at λ 325 nm. Retinol was identified and quantified by comparison with a pure commercial standard in chloroform (Sigma-Aldrich, Steinheim, Germany; Extrasynthese, Genay, France).

Reactive oxygen molecules (ROMs) in plasma were evaluated using a commercial kit (Diacron, Grosseto, Italy) following the manufacturer’s protocol and expressed as mmol H₂O₂.

The biological antioxidant potential (BAP) was measured with a commercial kit (Diacron, Grosseto, Italy), which evaluated the ability of the plasma to counteract the oxidative action of a hypochlorous acid (HClO) solution. BAP levels for each sample were expressed as μmol of neutralized HClO.

The moisture, ash, and total nitrogen content of the meat were assessed using AOAC methods (AOAC, 1995–N. 950.46B, 920.153, and 928.08, respectively). Total protein was calculated using the Kjeldahl method with a 6.25 conversion factor. Total lipids were extracted in duplicate from 5 g of each homogenized sample and calculated gravimetrically (Folch et al., 1957).

The ultimate pH (24 h) was measured with a Knick digital pH meter (Broadly Corp., Santa Ana, CA, USA) after homogenization of 1 g of raw muscle for 30 s in 10 mL of 5 M iodoacetate (Korkeala et al., 1986).

The water-holding capacity (WHC) was estimated by placing 1 g of whole muscle on tissue paper inside a tube and centrifuging for 4 min at 1,500 × g. The water remaining after centrifugation was quantified by drying the samples at 70°C overnight. WHC was calculated as follows: (weight after centrifugation − weight after drying)/initial weight × 100.

At 24 h post-mortem, the color values (L*-lightness, a*-red index, b*-yellow index) of the cut surface of fillets were measured using a tristimulus analyzer (Minolta Chroma meter CR-200, Osaka, Japan), following the CIELab color system (Robertson, 1977).

The oxidative parameters and fatty acid profile of meat were analyzed in duplicate. The α-, γ-, and δ-tocopherol, α- and γ-tocotrienol, and retinol content of the meat were quantified using the HPLC system described above, according to Hewavitharana et al. (2004). Five mL of distilled water and 4 mL of ethanol were added to 2 g of sample and vortexed for 10 s. After mixing, 4 mL of hexane containing BHT (200 mg/L) was added, and the mixture was carefully shaken and centrifuged at 8,000 × g for 10 min. An aliquot of the supernatant (3 mL) was dried under a stream of nitrogen and dissolved in 200 μL of acetonitrile; 50 μL was then injected into the HPLC as previously described.

Lipid oxidation was evaluated using a spectrophotometer set at 532 nm (Shimadzu Corporation UV-2550, Kyoto, Japan), which measured the absorbance of TBARS and a 1,1,3,3-tetraethoxypropane calibration curve (Ke et al., 1977). Oxidation products were quantified as malondialdehyde equivalents (μg MDA/g).

Carbonyl derivatives of proteins were detected according to the method of Mattioli et al. (2018). Briefly, the pellets from trichloroacetic acid (TCA) extracts were mixed with 1 mL of 10 mM DNPH in 2 M HCl. Samples were incubated for 1 h at room temperature and then centrifuged at 13,000 × g for 5 min. Supernatants were discarded, and the pellets were washed three times with 1 mL of ethanol–ethylacetate (1:1, v/v) to remove unreacted DNPH. The pellets were then dissolved in 1.5 mL of 6 M guanidine-HCl and centrifuged again to pellet insoluble particles. The carbonyl content of the resulting supernatants was evaluated spectrophotometrically at 370 nm using a molar extinction coefficient of 22000 1/M*cm; values were expressed as nmol of carbonyl/mg of protein in the guanidine chloride solution. Protein concentrations were measured using the Bradford method with Coomassie Brilliant Blue G-250 (Bradford, 1976), using bovine serum albumin as the standard. The same TCA extract was also used to evaluate thiol groups based on the 5,5’-dithio-bis-2-nitrobenzoic acid assay, with an extinction coefficient of 13,600 1/M*cm and expressed as µmoL SH – group per g.

Total lipids were extracted from 10 g of each homogenized sample, according to Folch et al. (1957), and esterified according to Christie (1982). One mL of each solution containing fatty acid esters was transferred into vials for gas chromatographic analysis. The separation of fatty acid methyl esters (FAME) was performed using a Varian gas chromatograph (CP-3800) equipped with a flame ionization detector (FID) and an Agilent capillary column (100 m x 0.25 mm, CPS Analitica, Milan, Italy) coated with a DB-Wax stationary phase (film thickness of 0.25 µm). Injector and detector temperatures were set at 270°C and 300°C, respectively. The carrier gas was helium at a flow rate of 0.6 mL/min. The oven temperature was programmed as follows: from 40°C (1 min hold) to 163°C (10 min hold) at 2°C/min ramp, to 180°C (7 min hold) at 1.5°C/min, to 187°C (2 min hold) at 2°C/min, and to 230°C (25 min hold) at 3°C/min.

Methyl esters of individual fatty acids were identified by comparing their retention times with those of a commercially available FAME standard mixture (FAME Mix Supelco 2560, Sigma-Aldrich, Germany). C21:0 methyl ester (CAS number 2363-71-5; Merck H5149, Germany), eluted under the same conditions as the samples, was used as an internal standard (1 mg/100 µL of added solution). The area of each peak was used to calculate the fatty acid proportion. For the quantitative analysis (mg/100g of meat), the calculation method reported by Vahmani et al. (2017) was applied.

Total saturated fatty acids (SFA), total monounsaturated fatty acids (MUFA), total polyunsaturated fatty acids (PUFA), total n-6 and n-3 fatty acids, as well as the n-6/n-3 ratio, were also calculated. The healthy fatty acid index (HFI) was also calculated according to Bosco et al. (2022) applying the following equation: HFI = ((mg/100 g of MUFA × 2) + (mg/100 g of n-6 × 4) + (mg/100 g of n-3 × 8) + (mg/100 g n-3/mg/100 g n-6))/((mg/100 g of SFA) + (mg/100 g of MUFA × 0.5) + (mg/100 g of n-6 × 0.25) + (mg/100 g of n-3 × 0.125) + (mg/100 g n-6/mg/100 g n-3)).

Statistical analyses were performed using SPSS Statistics version 23 (IBM, SPSS Inc., Chicago, IL, USA; Figure 1 ). The nearly 200 variables collected were classified into four main pillars (i.e., behaviors, body conditions, meat quality, in vivo health) and reduced to 121 ( Supplementary Table S1 ) based on ease of assessment and use, cost-effectiveness, and scientific soundness. Furthermore, to avoid multicollinearity, the variables were first selected using correlation and pooled within-groups correlation matrices by eliminating those with a coefficient ≥ │0.8│ (Garson, 2012a; Frizzarin et al., 2023). In the “behavior” pillar, variables with a mean occurrence< 1 were not included (i.e., rare behaviors), in order to obtain a relatively small number of variables (n=35; Table 3 ).

Then, the variables were subjected to

(i) discriminant analysis (DA) to select the variables for each pillar characterizing the four genotypes,

(ii) principal component analysis (PCA) to create a composite index including all the pillars,

(iii) Cronbach’s alpha to refine the index and assess its internal consistency and reliability.

Thus, each pillar was initially subjected to DA with the stepwise method, using an F value of 0.02 as the criterion for entering variables. The DA produced the most parsimonious linear combinations of indicators describing among-genotype differences for each pillar (discriminant function; DF). The standardized DF coefficients express the relative discrimination power of each variable. The performance of the final DFs was estimated by a prediction matrix table, which calculated the probability for each sample to be accurately classified in the correct genotype through the DF (Garson, 2012b; Agradi et al., 2023).

For each pillar, the variables to be included were selected based on their contribution to discrimination. In particular, for each DA, only the DFs explaining at least 90% of the variance among genotypes were considered: DF1 for the “behavior” and “body condition” pillars; DF1 and DF2 for the “meat quality” and “in vivo health” pillars. Within these DFs, only the variables with coefficients >0.3 were chosen (Pituch and Stevens, 2016). At this step, the number of variables included was 27.

Subsequently, the pool of variables selected by these DFs was processed using the principal component analysis (PCA) procedure to create a unidimensional measure representing the four pillars (Menchetti et al., 2019; Cartoni Mancinelli et al., 2021). The first principal component (PC) extracted by PCA was considered as an index of the adaptive response of animals, as it condenses the consistent information of the original measurements into a single variable. The selected variables were then standardized (to a mean of zero and a standard deviation of one) and included in a correlation matrix. Variables that were highly (r>│0.8│) or poorly (r<│0.3│) correlated were excluded to optimize the PCA performance (Statistics, 1986; Vyas and Kumaranayake, 2006). Subsequently, these variables were submitted to several rounds of PCA to eliminate those with low loadings. Variables sharing less than 10% of their variance with the PC (i.e., loading ≥0.32) were removed. Through an iterative process, PCAs were repeated until all factors met these criteria (Vyas and Kumaranayake, 2006; Statistics, 2011; Pituch and Stevens, 2016; Abunab et al., 2017). The cumulative variance explained by the final PCA—reflecting the total proportion of variation captured by the selected principal components—is reported in the Results to illustrate how effectively the reduced dimension represents the original dataset.

The final list of variables was refined and validated using Cronbach’s α. The function “Cronbach’s Alpha if Item Deleted” (i.e., the change in Cronbach’s α that would occur if a particular item were deleted) was used for further variable selection. In particular, a variable was removed when its elimination increased the Cronbach’s α value and improved the reliability of the index (Ortolani et al., 2021; Field, 2024). Through an iterative process, PCAs were repeated until the included variables ensured the best reliability. The Cronbach’s α value (α >0.7 indicating good reliability; Field, 2009; Menchetti et al., 2018; Pérez-gálvez et al., 2020) of the final PC was used to test the internal consistency reliability of the index.

The index score was calculated as the sum of the variables weighted by the loading obtained by the PCA:

Where,

I=score of the index

X_i = value of the i^th variable

W_i =weight (i.e., loading) of the i^th variable

After checking for multicollinearity, the following eight variables with mean occurrence > 1 were included in the DA for the “behavior” pillar (DA_behaviour): Walking, Resting, Scratching, Running, Roosting, Grass pecking, Attacking, and Grooming. The stepwise procedure selected six variables, and the first DF explained 98.8% of the variance between genotypes ( Table 3 ). The DA_behaviour allowed 100% of the animals to be correctly classified ( Figure 2A ).

Among the 15 variables entered for the “body conditions” pillar (DA_{body condition}), six were selected by the stepwise procedure ( Table 3 ). The first DF extracted explained 84.0% of the variance and correctly classified 100% of the cases ( Figure 2B ). Among these six variables, live weight, DFI, breast yield, and breast score had a loading greater than 0.3.

Thirty-eight variables were entered into the DA_{meat quality,} and 14 were finally selected ( Table 3 ). The DA_{meat quality} correctly classified 97.1% of animals. However, to achieve at least 80% (i.e., 83.3%) of the explained variability ( Figure 2C ), only the first two DFs were considered. Within these two DFs, the variables having a coefficient > 0.3 were breast HFI, drumstick carbonyls, drumstick drip loss, breast MDA, breast n-3 PUFA, drumstick SFA, drumstick n-6 PUFA, breast SFA, breast retinol and breast lipids ( Table 3 ).

Similarly, the DA _{In vivo health} produced two DFs explaining at least 80% of the variance (i.e., 93.7%; Table 3 ). Among the 12 variables included in the DA _{In vivo health}, eight were selected, all with a loading greater than 0.3. This DA allowed 77% of samples to be correctly classified ( Figure 2D ).

Overall, a total of 34 variables were selected, and 26 variables were included in the PCA.

According to the correlations and loadings from the PCA, ten variables were selected for the creation of the index ( Table 4 ). Cronbach’s alpha value for this composite index was 0.780. However, the reliability analysis indicated that the trait “Running” did not contribute significantly to the index and was eliminated. Thus, a final PCA was conducted, including the nine variables listed in Table 4 , while Supplementary Table S3 presents the descriptive statistics for the nine selected variables across the four genotypes. This PCA explained 38.6% of the variance, demonstrating good internal reliability (Cronbach’s alpha = 0.783).

Figure 3 and Supplementary Table S4 report the final classification based on the estimated indices obtained for the different genotypes. In detail, CB and LD genotypes showed the worst indices, whereas NN and RJ showed the best ones.