Animals and herds

The study was conducted between 03 February 2021 and 09 June 2021 in a total of 6 herds and 290 cows in Finland, Spain and Italy. Sampling was carried out in three different countries, since the trial was performed as an activity of the ClearFarm project, in which institutions from these three countries collaborated. ClearFarm aimed to co-design, develop and validate a software platform using smart farming technology to provide animal welfare information. Hair cortisol was one of several welfare indicators tested in the project. Sampling protocol was discussed and agreed prior to experiment to assure that samples were collected consistently in all countries. General information on the study herds and sample collection is presented in Table 1. The cows were of the breeds Holstein and Nordic Red. All cows were fed a total mixed ration and had ad libitum access to water.

Table 1.

Overview of herds and animals participating in the study

¹ DIM – days in milk.

² T1- represents days 0-45 and T2- represents days 46-90 of the study period.

³ all samples were collected in year 2021.

Data collection

The study lasted 90 days. Data collection was divided into two study periods, T1 and T2, each lasting 45 days.

Data relevant to animal welfare (animal-based assessment according to the Welfare Quality (WQ®) protocol, veterinary treatments, meteorological data) were collected during the study period. A detailed methodology of animal welfare data collection is presented in Stygar et al. (Reference Stygar, Frondelius, Berteselli, Gómez, Canali, JK, Llonch and Pastell2023). Here, the methods are briefly presented to allow the reader to understand how the average welfare class was calculated for all cows participating in the study.

Trained scorers visited the farms three times during the study period, on days 0, 45 and 90. The assessments were carried out by the same assessor in each country. During these visits, on-farm animal welfare data were collected according to WQ® (Welfare Quality®, 2009) guidelines. Only animal-related measures that could be assessed at an individual level were included. These measures covered three of the four WQ® principles: good nutrition (body condition score), good housing (udder, leg, hind quarter cleanliness) and good health (skin changes, locomotion score, mastitis, nasal discharge, eye discharge, vulvar discharge, diarrhea). The WQ® measures were supplemented with a daily temperature-humidity index to account for heat stress (detailed methodology in (Stygar et al., Reference Stygar, Frondelius, Berteselli, Gómez, Canali, JK, Llonch and Pastell2023). To account for any health problems that might develop between farm visits, farm records of animal health were used to complement the data from the on-farm welfare assessment. Each farmer provided farm records of disease diagnosis and treatment (including clinical mastitis, dystocia, respiratory disease, reproductive disease and metabolic disease) for all animals enrolled in the study for the duration of the study. The summary of veterinary treatment records is presented by Stygar et al. (Reference Stygar, Frondelius, Berteselli, Gómez, Canali, JK, Llonch and Pastell2023, Supplementary material).

A daily welfare index was created by combining the on-farm welfare measures, the on-farm veterinary records and the meteorological data. The severity (0 = none, 1 = mild to moderate and 2 = severe) and duration (from 1 to 45 days) of the different welfare problems were estimated by the expert judgement of the authors and based on the relevant scientific literature. For example, if a cow suffered from clinical mastitis, it was assigned a severity score of 2 for 21 days from the day of treatment. Conversely, if a cow had a moderate lameness problem, it was assigned a severity score of 1 for 45 days from the date of assessment. The final welfare index for each cow was obtained by summing the severity scores on a daily basis. A cow could receive a daily welfare index ranging from a minimum of 0 to a maximum of 23. For each cow, an average welfare index was calculated for each 45-day study period (T1 and T2). Based on these 45-day average indices, each cow was classified into one of three welfare classes for both T1 and T2: class 1 (if welfare index < 2), class 2 (with index > 2 but ≤ 3), class 3 (index > 3), representing animals with good, medium and poor welfare, respectively. The mean and distribution of welfare classes for each farm are shown in Fig. 1.

Figure 1.

Boxplot representing the distribution of an individual welfare class obtained in six herds during the study period. Dashed line represents mean value calculated from all herds, red dots represent mean value of a welfare class in each herd.

Hair sampling

Hair samples were collected between May and June 2021, depending on the herd (Table 1). For sample collection, a segment of approximately 10 × 5 cm of hair was shaved from the dorsal region at the initial segment of the tail switch, adjacent to the skin, using electric clippers and/or scissors. This sampling area was chosen because hair growth is faster than at other sites and therefore sensitive enough to capture changes in cortisol over short intervals (Burnett et al., Reference Burnett, Madureira, Silper, Nadalin, Tahmasbi, DM and RL2014). Hair samples were collected on day 104 of the study (14 days to allow for hair growth + first 45-day period + second 45-day period). The waiting period of 14 days for hair to appear on the upper skin surface was based on the available literature (Meyer and Novak, Reference Meyer and Novak2012; Russell et al., Reference Russell, Koren, Rieder and Van Uum2012; Vesel et al., Reference Vesel, Pavič, Ježek, Snoj and Starič2020). When collecting hair samples, excessively contaminated hair was avoided as much as possible and the base of the hair was marked with a coloured elastic band to facilitate identification of the direction of hair growth during sample processing. Collected hair samples were identified with a unique cow and farm identification number and dried at room temperature (RT) for 24 hours.

Given that the growth rate of a tail switch hair was estimated to be 0.51 ± 0.05 mm/day (Burnett et al., Reference Burnett, Madureira, Silper, Nadalin, Tahmasbi, DM and RL2014), it was assumed that a length of 21 mm represented the cumulative cortisol concentrations over a 45-day period. Hair samples of individual cows were therefore cut into two separate samples representing two distinct study periods, namely 0-45 days (sample T1 – cut between 22 mm and 43 mm from the skin surface) and 46-90 days (sample T2 – cut between 0 and 21 mm from the skin surface). Subsequently, each period sample (T1 and T2) was placed separately in small sealable plastic bags. An animal and a sample identification number were attached to the bag and sent to the laboratory of the Faculty of Veterinary Medicine of the University of Murcia (30,001 Espinardo, Murcia, Spain), where they were stored at RT until further analysis.

Hair cortisol extraction

Hair cortisol extraction was performed according to the protocol of Davenport et al. (Reference Davenport, Tiefenbacher and Lutz2006) modified by López-Arjona et al. (Reference López-Arjona, Tecles, Mateo, Contreras-Aguilar, Martínez-Miró, Cerón and Martínez-Subiela2020a).

From each hair sample, 250 mg was weighed and placed in a standard round bottomed polystyrene microcentrifuge tube (12 × 75 mm) covered with 2.5 mL of isopropanol (isopropyl alcohol, Macron Fine Chemicals, Avantor Performance Materials, Center Valley, PA, USA). After shaking for 3 min (100 rpm, VXR basic VIBRAX, IKA, Staufen, Germany), the tube was centrifuged at 1,500 g × 1 min at RT, and the isopropanol was discarded. This washing procedure was repeated twice. The hair sample was then left at RT for at least 2 h until completely dry. Next, 60 mg of each hair sample was cut into small pieces and placed in tubes with balls (Precellys Lysing kit, hard tissue grinding MK28, Precellys Bertin Technologies, France) and then pulverized (6,800 rpm × 20 sg, 3 cycles, twice) into a fine powder using a homogenizer (Precellys Tissue homogenizer, Bertin Technologies, France). Once pulverized, the hair was incubated with 1 mL methanol (Methanol 361,091, ITW Reagents, Monza, Italy) for 18 h at RT with continuous gentle agitation (100 rpm, VXR basic VIBRAX, IKA, Staufen, Germany) for steroid extraction. The samples were centrifuged (2,000 g × 5 min) and 0.6 mL of each methanol extract was aliquoted into a new Eppendorf tube. The liquid portion of the samples was evaporated using a Speed Vac Concentrator (Centrifugal Vacuum Concentrator 5301, Eppendorf, Hamburg, Germany). The dry extracts were reconstituted with 0.1 mL phosphate buffer saline (PBS) and stored at − 80ºC until cortisol analysis.

Analytical validation of automated assays for cortisol measurement

A sub-sample of hairs collected during the project was used for analytical validation. The analytical validation was performed on samples collected from two Spanish herds, from 20 randomly selected animals and independent of the hair segment (T1 or T2). The samples were first analysed independently and after which they were mixed in order to make pools of different concentrations.

The precision of the method was evaluated by the intra- and inter-assay coefficient of variation (CV) as described by López-Martínez et al. (Reference López-Martínez, Escribano, Martínez-Miró, Ramis, Manzanilla, Tecles, Martínez-Subiela and Cerón2022). For intra-assay precision, two pools of hair extracts with high and low cortisol concentrations (9.2 and 4.4 pg/mg, respectively) were measured 5 times in a single analysis. Inter-assay precision was assessed by analysing the same samples as in the intra-assay test on 5 different days over a period of 15 days. To avoid freeze-thaw cycles that could interfere with the results, separate hair samples were aliquoted and stored at − 80°C so that they were only thawed once for analysis. To determine the sensitivity of the method the limit of detection (LOD) and the lower limit of quantification (LLQ) were determined. LOD was understood as the lowest cortisol concentration that the method can distinguish from the zero value. It was obtained by calculating the mean of 5 replicate measurements of the zero standard (phosphate buffered saline) plus two standard deviations. The lowest cortisol concentration that was measured with a precision of less than 20% was adopted as LLQ (López-Arjona et al., Reference López-Arjona, Mateo, Manteca, Escribano, Cerón and Martínez-Subiela2020b). To calculate it, five replicates of a pool of hair extract, serially diluted with assay buffer, were measured in the same run.

To test the accuracy of the method, a recovery test and a linearity after dilution assay were performed. To assess the linearity of the assay, two hair extracts with high and low cortisol concentrations (13.2 and 4.45 pg/mg) were serially diluted from 1:2 to 1:64 with the assay buffer as described by Escribano et al. (Reference Escribano, Fuentes-Rubio and Cerón2012). Each dilution was assayed in duplicate. The measured HCC was then plotted against the expected HCC. The average of the two initial measurements (1:2 dilution) was used to calculate the expected value for each dilution. For the recovery test, a high cortisol sample was mixed with a low cortisol sample at different dilutions. The high cortisol sample was diluted to 50%, 25% and 10% and the low cortisol sample was diluted to 50%, 75% and 90%. In addition, the low cortisol sample was diluted to 25% with a high cortisol sample at 75%. The resulting curve represents the measured HCC versus the expected HCC.

To determine the stability of the sample at −20ºC and −80ºC, two pools of hair extracts with different concentrations (9.05 and 6.6 pg/mg) were measured immediately after processing, before first freezing, and 30 days, 90 days and 1 year after freezing. In addition, aliquots were taken and refrigerated at 4°C and measured after 5, 10 and 30 days. Cross-reactivity was assessed by performing a linearity of corticosterone and cortisone.

HCC measurement

HCC was measured on samples from individual cows for both study periods, T1 and T2 separately, to investigate the relationship between HCC and welfare class. A solid-phase competitive chemiluminescent enzyme immunoassay (Immulite/Immulite 1000 cortisol, Siemens Medical Solutions Diagnostics, Los Angeles, CA) was used for HCC determination. The immunoassay uses a rabbit polyclonal anti-cortisol in addition to cortisol-conjugated alkaline phosphatase as a single reagent. According to the manufacturer, the cross-reactivity with prednisolone is 49% and the detection limit of the assay is 0.05 μg/dl. HCC results are expressed as pg/mg hair.

Associations between HCC and level of animal welfare

A linear mixed model was fitted to explore the associations between HCC (response variable) and welfare class, study period and cow factors (explanatory variables). HCC was log-transformed to obtain a normal distribution. The effect of the explanatory variables on HCC was modelled as follows:

\begin{align*}{Y_{i,j}} =& \,{\beta _0} + {\beta _1}Parity + {\beta _2}DIM + {\beta _3}T + {\beta _4}WelfareClass \\&+ {\beta _5}Herd + { }{A_{ij}} + {\varepsilon _{ij}}\end{align*}

where ${Y_{i,j}}$ was log transformed cortisol measurement for cow i = 1,…,n in time j = 1,2, ${\beta _0}$ is overall mean of log cortisol, ${\beta _1},\,{\beta _{\,2}},\,{\beta _3},\,{\beta _4}$ are regression coefficients of parity (continuous), DIM (continuous), study period (T1 and T2), welfare class (three categories: 1,2 and 3) and herd (six categories: 1,2,3,4,5 and 6), respectively. ${A_{ij}}$ was a random effect of a cow, ${\varepsilon _{ij}}$ is a measurement error assumed to follow a multivariate normal distribution with zero mean. Observations were collected from animals in six herds and observations from cows in the same herd were correlated. Due to the small number of herds observed, herd was considered as a fixed effect (rather than a random effect). HCC originated from the same animals and therefore samples T1 and T2 collected within a cow were also correlated. This was accounted for in the model as a random effect of cow. Hair color was not included as an explanatory variable due to the small sample size of non-white hair. Model control was performed using graphical and numerical summaries (Q-Q plot, residual vs. fitted plot) as recommended by Pinheiro and Bates (Reference Pinheiro and Bates2014). A significance level of α = 0.05 was used for all tests. The proportion of variance in HCC explained by the models was expressed by the marginal and conditional coefficients of determination (R2M and R2C, respectively), as defined by Nakagawa and Schielzeth (Reference Nakagawa and Schielzeth2013).

Statistical analyses and plots were performed using spreadsheets Excel (2019, Microsoft) and R (2021) with the following packages: nlme (Pinheiro et al., Reference Bates, Douglas and Pinheiro, Jose2014), MuMIn (Barton, Reference Bartoń2016) and ggplot2 (Wickham, Reference Wickham2016).