Carotene

All assays were made using 96-well microplates (Nunclon, Nunc, Roskilde, Denmark) and were measured in an ELISA Reader Infinite Pro 200F (Tecan Group Ltd., Männedorf, Switzerland).
The DPPH^• (2,2-diphenyl-1-picryl-hydrazyl) radical scavenging activity was measured by a previously described method [57 (link)]. DPPH solution (180 μL of freshly prepared 0.07 mg/mL solution) was mixed with 20 μL of the examined extract in various concentrations in microplates. The absorbance at 517 nm was monitored after 30 min incubation at 28 °C and the results were expressed as an EC₅₀ value. Ascorbic acid was used as the control. The antiradical potential was also analyzed using the previously described ABTS^•+ (2,2′-azinobis[3-ethylbenzthiazoline]-6-sulfonic acid) assay [57 (link)]. The absorbance was measured at 734 nm and the results were expressed as millimoles of Trolox equivalents per g of dry extract (TEAC).
The metal chelating activity was determined by the method described by Guo et al. [58 ] with some modifications. In this assay, 0.2 mM aqueous solution of ferric chloride and 0.5 mM aqueous solution of ferrozine were used. Twenty microliters of the 0.2 mM aqueous solution of ferric chloride (II) was mixed with 100 μL of extract at different concentrations. Next, 40 μL of 0.5 mM aqueous solution of ferrozine was added and microplates were shaken and incubated for 10 min in 24 °C. The absorbance was measured at 562 nm and the percentage of inhibition of ferrozine–Fe²⁺ complex formation was calculated using the following formula:
where A_c is the absorbance of the control (water instead of the extract), and A_s is the absorbance of the extract.
The results were presented as the concentration of the extract that causes metal chelating in 50% (EC₅₀) calculated on the basis of the linear correlation between the inhibition of ferrozine–Fe²⁺ complex formation and the concentrations of the extract. EDTA was used as a positive control.
Antioxidant activity was also assayed using the β-carotene bleaching method described and modified by Deba and co-authors [51 (link)]. Twenty microliters of extract at different concentrations were mixed with freshly prepared β-carotene-linoleic acid emulsion, and incubated for 20 min at 40 °C. The absorbance was measured at 470 nm. BHT was used as a positive control.

Over the course of the experiment, one sample from each of the 2018 grass silage bales used was taken, four were obtained from the 2019 grass silage silo, five from the corn silage silo, and two each from 2018 and 2019 hay and one from concentrate. Grass and corn silage samples were lyophilized, and all feed samples were milled to 1 mm particulate size using a centrifugal mill (model ZM1, Retsch GmbH, Haan, Germany) before analysis. All feed analyses were performed in duplicate.
Contents of DM of feed items were assessed by a thermogravimetric device (TGA-701, Leco St. Joseph, MI, USA), and ether extract (EE) contents were determined using a Soxhlet extractor (B-811, Büchi, Flawil, Switzerland). Fatty acids were extracted with a solvent extractor using a hexane: propane-2-ol mixture (3:2 v/v). They were then transformed to FA methyl esters (FAME) [17 ] and cleaned following Wettstein et al. [18 (link)]. The FAME analysis was performed with a GC equipped with a CP7421 column using split injection (1:5). Internal and external standards applied were C11:0 and commercially available sunflower oil (response factor). The initial temperature was set to 170°C and held for 1 h. This was then increased by 5°C /min to 230°C and held for 32 min, then increased by another 5°C /min to 250°C and held for 15 min. The FA peaks were identified by their retention times using a standard mixture of 37 FA, C4-C24 (Supelco 37-Component FAME Mix CRM47885, Merck, Switzerland) and the Matreya Cis-Trans Isomer standard (MAYA1131; Avantor, Switzerland).
Feed β-carotene content was analyzed with a normal-phase UV/VIS HPLC according to the European Standard method 12823–2 [19 ].

To aid in the evaluation and quantification of suitable milk biomarkers for %GB a multi-step analysis approach was taken. An initial one-way analysis of variance (using PROC ANOVA of SAS) was performed across both dietary-intake and milk-related variables comparing the years of harvest 2018 and 2019. This allowed the assessment of the stability of potential biomarkers to feed storage duration (year of harvest).
Next, the best model for each response variable was selected based on the lowest corrected Akaike Information Criterion for small samples sizes [24 (link)] by stepwise selection of the initial model, depicted as:
$${\text{Y}}_{\text{ij}}={\text{b}}_0+{\text{b}}_1*\%{\text{GB}}_{\text{ij}}+{\text{b}}_2*\%{\text{GBsq}}_{\text{ij}}+{\text{b}}_3*{\text{DIM}}_{\text{ij}}+{\text{b}}_4*1_{\{\text{yearij}=2018\}}+{\text{b}}_5*\%{\text{GB}}_{\text{ij}}*1_{\{\text{year}=2018\}}+{\text{e}}_{\text{ij}}$$
where Y = the response variable; i = 1, 2 (for years 2018 and 2019) and j = 1,2, …, n_i, where n_i = 12 (or 11) as equally as many cows received 2018 as 2019 harvests. Let 1_{{year_ij = 2018}} be an indicator function which is 1 if year_ij = 2018 and 0 if year_ij = 2019; b₀ = overall mean; b₁,₂,…₅ = regression coefficients of the observed effects of the grass-based diet (%GB, linear and %GBsq, quadratic), of days in milk (DIM) and of year of harvest (l_{{year = 2018}}) and e_ij = random error. From the initial mode, a stepwise model selection procedure (both forward selection and backward elimination) was applied (using PROC GLMSELECT of SAS). The lower range of the search scope included %GB or %GB squared. Including %GB as both a linear and quadratic term allowed the identification of linear and non-linear (associative) relationships of variables with %GB. To ensure that the dietary effects were corrected for different lactation stages, DIM was included in the initial model as a continuous explanatory variable.
The number of independent datasets (24 for intake data and 22 for milk related traits, the latter excluding the two cows supplemented with β-carotene) approached the threshold given by Jenkins and Quintana-Ascencio [25 (link)] for regression analysis with high data variance (≥ 25) and by far exceeded the threshold of ≥ 8 given for low data variance.
Figs 1 and 2 depict individual cow data. Harvest years are presented separately (2018, filled data points; 2019, hollow data points) and the slope of the regression across all data is depicted as a solid line (the individual data points of the two cows receiving the β-carotene supplement and excluded from the regression analysis are depicted as triangles in the figures). Any time DIM was included in the regression model, it was held constant at the average DIM (321 days). Milk FA regression values are given in Tables 2, 4, 6 and S5 Table.
Lastly, as the first step towards the simplest determination of the six most promising milk biomarker thresholds for a given %GB, DIM and Year were omitted from the model and a simplified regression equation was applied:
$${\text{Y}}_{\text{ij}}={\text{b}}_0+{\text{b}}_1*\%{\text{GB}}_{\text{ij}}+{\text{b}}_2*\%{\text{GBsq}}_{\text{ij}}$$
where b₀ = the intercept, b₁ = the coefficient for %GB_ij and b₂ = the coefficient for %GBsq_ij, based on the previously selected best fit regression model. This allowed for the development of a preliminary set of biomarker prediction values to determine the proportion of dietary grassland-based feeds when the more detailed information of DIM and year of harvest are unavailable. Pearson correlation coefficients were computed with the PROC CORR function of SAS to determine the association between GC and MIR FA data. Effects were considered as statistically significant at P < 0.05. All statistical analyses were performed with the statistical software SAS, version 9.4 (SAS Institute, Cary, NC, USA) and figures were generated with SigmaPlot (v. 13; Systat Software, San Jose, CA).