Christophe Octave S, Grelet Clément, Simon Huzaïfa, Leblois Julie, Baeten Vincent, Dehareng Frédéric
Walloon Agricultural Research Center (CRA-W), Chaussée de Namur, 24, 5030, Gembloux, Belgium.
Elevéo asbl, AWE groupe, rue des Champs Elysées, 4, 5590, Ciney, Belgium.
Sci Rep. 2025 Jul 25;15(1):27083. doi: 10.1038/s41598-025-12332-9.
Chronic stress in dairy cows could adversely affect their emotional well-being, immune function, reproductive capability and milk yield. Effective measurement and assessment of chronic stress in herds is crucial for maintaining welfare and addressing issues. Hair cortisol concentration is a promising biomarker of chronic stress that is typically measured using enzyme linked immunosorbent assay (ELISA), which is reliable but costly and time-consuming. This study explores the potential of three vibrational spectroscopic techniques near-infrared (NIR), mid-infrared (MIR), and Raman spectroscopy to predict cortisol content in dairy cow hair samples as a more efficient alternative. 134 hair samples from 30 cows were analysed using bench-top spectrometers to obtain NIR, MIR and Raman spectra. Hair cortisol levels, determined by ELISA, ranged from 8.3 to 91.5 pg/mg, with an average of 27.1 pg/mg. The partial least squares (PLS) model yielded R values of 0.63, 0.62, and 0.52 and RMSE values of 9.5, 9.2, and 9.1 for NIR, MIR, and Raman respectively. MIR and Raman spectroscopy showed better performance than NIR, with MIR chosen for further study. Subsequently, 1183 additional samples were collected from five countries (Belgium, Luxembourg, France, Germany and Austria) for a robust prediction model using MIR. The full dataset now included 1316 samples. Various machine learning techniques: principal component regression (PCR), partial least squares (PLS), elastic net regression (ENR), and support vector machine regression (SVM-R) were employed, with tenfold cross-validation and random external validation (excluding 20% of samples). RMSEv values were 6.65, 6.05, 5.94, and 6.24 pg/mg for PCR, PLS, ENR, and SVM-R respectively.
奶牛的慢性应激会对其情绪健康、免疫功能、繁殖能力和产奶量产生不利影响。有效测量和评估牛群中的慢性应激对于维持福利和解决问题至关重要。毛发皮质醇浓度是慢性应激的一个有前景的生物标志物，通常使用酶联免疫吸附测定（ELISA）进行测量，该方法可靠但成本高且耗时。本研究探索了近红外（NIR）、中红外（MIR）和拉曼光谱这三种振动光谱技术预测奶牛毛发样本中皮质醇含量的潜力，作为一种更有效的替代方法。使用台式光谱仪对来自30头奶牛的134份毛发样本进行分析，以获得近红外、中红外和拉曼光谱。通过ELISA测定的毛发皮质醇水平在8.3至91.5 pg/mg之间，平均为27.1 pg/mg。偏最小二乘法（PLS）模型对近红外、中红外和拉曼的R值分别为0.63、0.62和0.52，RMSE值分别为9.5、9.2和9.1。中红外和拉曼光谱表现优于近红外，因此选择中红外进行进一步研究。随后，从五个国家（比利时、卢森堡、法国、德国和奥地利）收集了另外1183份样本，用于使用中红外建立强大的预测模型。完整数据集现在包括1316份样本。采用了各种机器学习技术：主成分回归（PCR）、偏最小二乘法（PLS）、弹性网络回归（ENR）和支持向量机回归（SVM-R），进行十倍交叉验证和随机外部验证（排除20%的样本）。PCR、PLS、ENR和SVM-R的RMSEv值分别为6.65、6.05、5.94和6.24 pg/mg。
