**Background:** Thoroughbred racehorses undergo intense training from a young age, requiring rapid adjustment to novel environments. The hypothalamic-pituitary-adrenal (HPA) axis releases cortisol in response to stress, and chronic over-activation can have deleterious health and behavioural effects. Identifying horses with more sensitive temperaments via objective measures could enable proactive management. Salivary cortisol is a less-invasive proxy for serum cortisol and has been used to evaluate stress responses in horses.
**Methods:** The study cohort comprised n=96 yearling Thoroughbreds (46 female, 50 male) born in 2017, progeny of 18 sires, all housed and trained at a single yard. Saliva samples were collected using a dental sponge and stored in 1.5ml tubes with a plastic straw segment to prevent reabsorption. Samples were refrigerated within 30 minutes, centrifuged at 1500 × g for 10 minutes, and cortisol concentration measured using the Salimetrics™ Salivary ELISA Kit. Resting samples were collected between 14:00–15:30 at three timepoints: at the nursery yard before moving to the main yard (RN, n=66), within three days of arrival at the main yard (RY_T1, n=67), and after 2–3 weeks in the main yard (RY_T2, n=50). A timecourse experiment sampled n=5 female horses every two hours from 08:00–16:00. Pre- and post-event samples were collected for three novel training events: first time being driven with long reins (FD, n=6 paired samples), first time being backed by a jockey (FR, n=34 paired samples), and first time ridden on the gallops (FG, n=10 paired samples). Post-event samples were taken ~30 minutes after the event. Statistical analyses used ANOVA and paired t-tests in RStudio (version 3.6.0), with P<0.05 considered significant.
**Key Results:** The inter-assay CV was 0.96%; mean intra-assay CV was 0.83% (SD±0.69). Across all samples, cortisol concentrations ranged from 1.03–36.18 nmol/L. In the timecourse experiment, mean cortisol at 08:00 (T1) was 3.33 nmol/L (SD±0.63), significantly higher than later timepoints (Paired t-test, P<0.05). No significant difference was found among the later four timepoints (10:00–16:00). For resting samples, there was no significant difference in mean cortisol across RN (mean 2.69 nmol/L, SD±1.08), RY_T1 (mean 2.30 nmol/L, SD±0.92), and RY_T2 (mean 2.36 nmol/L, SD±1.19) (ANOVA, F=2.98, df=2, P>0.05). For the training events: first driving showed a significant increase from pre (mean 3.09 nmol/L, SD±1.42) to post (mean 9.73 nmol/L, SD±5.66) (paired t-test, t=-4.493, df=5, P<0.005). First backing showed a significant increase from pre (mean 3.17 nmol/L, SD±1.48) to post (mean 11.09 nmol/L, SD±7.21) (paired t-test, t=-10.48, df=33, P=4.928×10⁻¹²). First gallops showed a significant increase from pre (mean 2.67 nmol/L, SD±0.56) to post (mean 12.69 nmol/L, SD±5.77) (paired t-test, t=-11.22, df=9, P=1.364×10⁻⁶). Post-event ranges were wide: driving 3.88–19.62 nmol/L, backing 2.71–36.18 nmol/L, gallops 4.40–24.33 nmol/L.
**Clinical Implications:** The significant cortisol increases after all three novel training events confirm these milestones are acutely stressful for yearling Thoroughbreds. The wide individual variation in post-event cortisol, particularly after first backing (range 2.71–36.18 nmol/L), suggests that some horses mount a substantially greater HPA axis response than others. This variation is not explained by the new training yard environment itself, as resting cortisol did not differ across locations. The findings support using salivary cortisol as an objective phenotype for stress reactivity, which could enable trainers to identify horses requiring more gradual or tailored introduction to training. Such proactive management may reduce chronic stress, improve welfare, and potentially decrease the >6% of Thoroughbreds that discontinue racing due to 'unsuitable temperament/behaviour'. Limitations include convenience sampling, variation in timing of training events, and the timecourse experiment using only female horses. Nonetheless, the uniform husbandry and single-yard design are strengths that reduce confounding variables.