Statistical analysis was conducted using REML in the MIXED procedure in SAS (SAS SAS, 2002). Standard model for a repeated measurements experiment is:
𝑌𝑖𝑗𝑘=µ+ 𝑔𝑟𝑜𝑢𝑝𝑖+ 𝑑𝑘+ cow𝑗(𝑔𝑟𝑜𝑢𝑝𝑖) + (𝑔𝑟𝑜𝑢𝑝𝑖× 𝑑𝑘) +𝑒𝑖𝑗𝑘
Yijk: parameter of cow j in group i at time k.
µ: overall mean
groupi: fixed effect of group i
dk: fixed effect of day in lactation k cowj(groupi): random effect of cow j in group i
groupi× dk: fixed interaction effect of group i with time k eijk: random error at time k on cow j in group i
As measurements of different animals were independent, covariance structure referred to variances at different time points and to correlation between measurements of the same animal. This
correlation consists firstly of the fact that two measurements of the same animal are correlated just because it is the same animal (variation between animals, RANDOM statement) and secondly it consists of covariation within the same animal (measurements close in time tend to be more highly correlated than those far apart, REPEATED statement). For every parameter, three covariance structures were evaluated: unstructured, compound symmetry or autoregressive order one. Used was the covariance structure that was closest to zero in the Akaike information criterion or Sawa's Bayesian information criterion (Littell et al., 1998). Day of lactation and group as well as their interaction (day × group) were used as fixed effects, whereas cow within treatment was the random effect. Furthermore, cow was the repeated subject. PDIFF function was used to determine the differences between treatments. Data was considered to differ significantly at P < 0.05.
Specific calculations or statistic calculations differing from above mentioned approach are explained below.
3.1.
Milk parameters during experimental period
Energy balance (EB) was calculated using the formula 𝐸𝐵= (𝐷𝑀𝐼𝑑𝑖𝑒𝑡×𝑁𝐸𝐿𝑑𝑖𝑒𝑡) + (𝐷𝑀𝐼𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑒𝑠 × 𝑁𝐸𝐿𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑒𝑠)− (0.293 ×𝑏𝑜𝑑𝑦𝑤𝑒𝑖𝑔ℎ𝑡0.75− [(0.38 ×
𝑚𝑖𝑙𝑘𝑓𝑎𝑡𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛)−(0.21 ×𝑚𝑖𝑙𝑘𝑝𝑟𝑜𝑡𝑒𝑖𝑛𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛) + 0.95] ×𝑚𝑖𝑙𝑘𝑦𝑖𝑒𝑙𝑑 as described by Kamphues et al. (2004). ECM was calculated with the formula 𝐸𝐶𝑀= (𝑚𝑖𝑙𝑘𝑦𝑖𝑒𝑙𝑑 × 0.327) + (𝑓𝑎𝑡𝑦𝑖𝑒𝑙𝑑 × 12.86) + (𝑝𝑟𝑜𝑡𝑒𝑖𝑛𝑦𝑖𝑒𝑙𝑑× 7.65) and FCM with 𝐹𝐶𝑀= (𝑚𝑖𝑙𝑘𝑦𝑖𝑒𝑙𝑑 × 0.4) + (𝑓𝑎𝑡𝑦𝑖𝑒𝑙𝑑 × 15).
Prior to statistical analysis, repeated end point measurements (daily milk yield, milk composition and hydrocortisone) were pooled to weekly means. Days 21, 23 to 31, 127 and 138 to 146 pp were excluded from calculation of weekly means because FR and ivGTT were conducted at these days. For better illustration, milk parameters were subsumed to 6 time periods: d 1 to 22 pp (period 1), d 33 to 56 pp (period 2), d 57 to 84 pp (period 3), d 85 to 112 pp (period 4), d 113 to 134 pp (period 5) and d 148 to 155 pp (period 6). Calculation of statistical differences were performed using REML in the MIXED procedure in SAS. To conserve the effect of different intervals between measurements, SAS received the first day of each period as day of interest (e.g. 1 for period 1 or 85 for period 4).
3.2.
Feed restrictions in early and mid-lactation
Repeated end point measurements of the last day before, the last day during and the last day after restricted feeding were compared within groups for milk parameters including hydrocortisone and β-hydroxybutyrate (d 25, 28 and 31 pp as well as d 140, 143 and 146 pp), blood parameters (d 26, 29 and 32 pp as well as d 141, 144 and 147 pp) and for feed intake and EB (d 25, 28 and 31 pp as well as d 140, 143 and 146 pp) using REML in the mixed procedure in SAS, like mentioned above. Differences between groups within each FR and differences between similar days of early and mid-lactation FR were evaluated correspondingly.
3.3.
Intravenous glucose tolerance tests
Area under the curve for insulin (AUCI) and glucose (AUCG) from -20 to 120 minutes after infusion were calculated with the trapezoid-rule:
𝐴𝑈𝐶 ≈ � �𝑡𝑖2− 𝑡𝑖1�× (𝐶𝑖1+𝐶𝑖2) × 1 2 𝑖𝑛=120 𝑖1=−20 t: minute of sampling
C: concentration of glucose [mmol/L] or insulin [µU/mL] i: sampling time point
III. Materials
and methods
31 Values for basal insulin and basal glucose were obtained by calculating arithmetic means of the four measurements before glucose infusion (-20, -15, -10 and immediately before infusion). Afterwards, basal AUCI and AUCG (bAUCI and bAUCG) were calculated as 𝑏𝐴𝑈𝐶 ≈ 𝑏𝑎𝑠𝑎𝑙𝑖𝑛𝑠𝑢𝑙𝑖𝑛or𝑔𝑙𝑢𝑐𝑜𝑠𝑒 × 140, assuming that the base is a rectangle with height of basal insulin or glucose concentrations and length of 140 minutes. Then net AUCI and AUCG (nAUCI and nAUCG) were calculated by subtraction of bAUCI from AUCI and bAUCG from AUCG, respectively.According to Kerestes et al. (2009), clearance of glucose was calculated using the formula 𝐶𝑅= ln 𝑔𝑙𝑢𝑐𝑜𝑠𝑒𝑖5−ln 𝑔𝑙𝑢𝑐𝑜𝑠𝑒𝑖60
60−5 × 100. Consistent with Radziuk (2000), homeostatic model
assessment of insulin was calculated as 𝐻𝑂𝑀𝐴 − 𝐼𝑅 = 𝑏𝑎𝑠𝑎𝑙𝑔𝑙𝑢𝑐𝑜𝑠𝑒22 ×.5𝑏𝑎𝑠𝑎𝑙𝑖𝑛𝑠𝑢𝑙𝑖𝑛.
Statistical differences were calculated using REML in the MIXED procedure in SAS, like mentioned before.
3.4.
Major milk proteins in skim milk
Data was provided by the software as percentage amount of total protein concentration. Due to technical reasons, the absolute amount of individual proteins were constantly over- or underestimated, so all following calculations were conducted with the percentage amount of protein concentration in the sample. To obtain comparable results between chips, every protein fraction was corrected by multiplication with a correction factor on the basis of the results of the respective proteins in the protein mix sample. Correction factor (CF) was calculated as 𝐶𝐹= % 𝑜𝑓𝑝𝑟𝑜𝑡𝑒𝑖𝑛𝑓𝑟𝑎𝑐𝑡𝑖𝑜𝑛20% 𝑖𝑛𝑝𝑟𝑜𝑡𝑒𝑖𝑛𝑚𝑖𝑥. After correction, percentage amount of each protein fraction within the sum of all protein fractions was determined. Results were subsumed to six time periods like done before with milk parameters. Differences between groups and time points were estimated with the above described model in SAS.
IV.
R
ESULTS1.
Animal experiment
1.1.
Cows and classification
Approximately four weeks before parturition 26 cows were transported from a farm in Saxony to the research farm Veitshof. One cow died during the course of parturition due to severe calving difficulties caused by an oversized calf. Another cow had to be euthanized shortly after parturition due to downer-cow-syndrome and slipping injuries. Additionally, all data of one animal had to be omitted because of recurring inflammations which were caused by a perforating foreign body in the reticulum.
Classification of remaining 23 cows was done based on mean FCM yield and mean milk protein content during d 23, 24 and 25 pp and resulted in four groups: 6 cows with high FCM (52.66 ± 2.91 kg/d) and high protein content (3.28 ± 0.07%; MP-cows), 5 cows showing low FCM (40.49 ± 1.15 kg/d) and low protein content (2.84 ± 0.06%; mp-cows), 7 cows with high FCM (48.98 ± 2.12 kg/d) and low protein content (2.90 ± 0.06%; Mp-cows) and 5 cows showing low FCM (39.08 ± 0.60 kg/d) and high protein content (3.40 ± 0.05%; mP-cows; table 4). Furthermore MP- and Mp-cows showed higher FCM compared to mp- and mP-cows (P < 0.01), whereas mp- and Mp-cows showed lower milk protein content compared to MP- and mP-cows (P < 0.001). Parity amongst all groups was comparable (MP-cows 2.8 ± 0.3, mp-cows 2.4 ± 0.2, Mp-cows 2.3 ± 0.3, and mP-cows 2.8 ± 0.4; P = 0.35).
Table 4: Parameters of classification (mean FCM yield, kg/d and mean milk protein content, %) of 23 cows during d 23, 24 and 25 pp.
cow parity group FCM, milk kg/d
mean FCM, kg/d
milk protein
content, % mean protein content, %
14024 57758 4 MP 58.73 52.66 ± 2.91a 3.07 3.28 ± 0.07a 14027 34346 3 MP 51.04 3.06 14027 34439 3 MP 60.16 3.50 14030 03463 3 MP 55.78 3.42 14030 03870 2 MP 40.97 3.29 14031 15582 2 MP 49.25 3.33 14026 25564 3 mp 44.25 40.49 ± 1.15b 3.02 2.84 ± 0.06b 14027 34303 3 mp 37.70 2.70 14031 15366 2 mp 38.64 2.86 14032 20073 2 mp 41.59 2.73 14032 20330 2 mp 40.24 2.89 14026 25242 4 Mp 54.13 48.98 ± 2.12a 3.00 2.90 ± 0.06b 14028 63689 2 Mp 46.72 3.00 14030 03642 2 Mp 55.40 2.71 14030 03827 2 Mp 41.13 2.95 14030 03863 2 Mp 44.10 3.02 14031 15265 2 Mp 54.23 2.95 14031 15625 2 Mp 47.17 2.67