ABSTRACT
The relationship between behavior and low-level, subclinical systemic inflammation was investigated in a group of matched-pair (sub-clinical mastitis, SCM, versus clinically healthy control, CTRL) intensively housed dairy cows (n = 34) over short (24h) distinct periods. We report, for the first time, that an increase in an inflammatory marker (salivary serum amylase-A, SAA) occurs during the early stages of a bovine disease. SAA was higher in SCM cows, and positively correlated with somatic cell count, the defining parameter of mastitis. SCM cows were observed to display reduced activity (behavioral transitions and distance moved), and reductions in several measures of social behavior including: social exploration, social reactivity (following the receipt of agonistic behavior), performance of social grooming and head butts, and the receipt of challenges. In addition, SCM cows received more head swipes, and spent a greater proportion of time lying with their head on their flank than CTRL cows. SCM cows also demonstrated a preference for lower-risk ‘within-herd feeding’; a greater proportion of time feeding was spent in direct contact with herd-members, and a lower proportion of time feeding was spent at self-locking feed barriers, than the CTRL cows. We also present evidence for diurnal differences in the daily behavioral routine between the two groups: SCM cows appear to shift their activity (social and otherwise) to quieter times of the day, a tactic that could actively avoid agonism. Many behavioral measures were found to correlate with SAA in a direction consistent with predictions for sickness behavior. We conclude that salivary SAA, social behavior and activity changes offer potential for use in the detection and monitoring of pre-clinical inflammatory disease states in cows.
1. INTRODUCTION
Infectious disease is highly detrimental to animal welfare and can negatively impact dairy herd profitability via production losses and expensive treatment costs (Grohn et al., 2003; González et al., 2008). It is imperative that emerging health problems are identified and treated as soon as possible; however, low level inflammatory processes and early symptoms of sickness and disease are not easily recognised in any animal (Weary et al., 2006). Although no abnormalities in the gland or milk are visible during subclinical mastitis in cows, milk production drops, milk somatic cell counts (SCC) elevate, and inflammation (with or without an intramammary pathogen) occurs (Sordillo et al., 1997). In cows, effective health monitoring is further hampered by logistical difficulties associated with direct animal observations within large open barn systems, reductions in human interaction linked with the substantial uptake in robotic milking units, and the tendency of cattle (as a prey species) to display only subtle indicators of pain/weakness (Gleerup et al., 2015). Advances in image analysis now allow automated recognition of individual cows within a herd (Andrew et al., 2020), accurate identification of some health-related abnormal behaviors, e.g. foot disease (Gu et al., 2017), and detection of social interactions (Guzhva et al., 2016). The identification of behaviors associated with subclinical disease may therefore find application in future diagnostic software algorithms targeted at early disease monitoring in dairy cows (see e.g. Wagner et al. 2020).
Sickness is an adaptive response to infection and inflammation, characterized by endocrine, autonomic, and behavioral change. Infection triggers the immune system to initiate a febrile response and reprioritise behavior (Hart, 1988; Dantzer et al., 2008). This process is mediated by pro-inflammatory cytokines acting upon the CNS (Dantzer and Kelley, 2007). ‘Sickness behavior’ is one such strategy to facilitate recovery, achieved via a reduction in physical activity (energy conservation) and the minimisation of heat loss. Sickness behaviors are useful for early disease detection (see reviews: Weary et al., 2009; von Keyserlingk and Weary, 2010). They include anorexia and withdrawal from the physical and social environment (Dantzer and Kelley, 2007; Tizard, 2008; Hart, 2010). Behavior in healthy animals can be thought of as being of two types in relation to fitness: ‘core maintenance’ (or ‘high-resilience’) with immediate, short-term benefits (e.g. sleep, feed, drink) and ‘luxury’ (or ‘low-resilience’) with delayed, longer-term benefits (e.g. play, exploration, grooming) (Dawkins, 1990; Špinka et al., 2001; Weary et al. 2009). Due to the relative ease with which core behavior can be automatically monitored in dairy cows this has received most attention to date. Changes in feeding behavior associated with bovine clinical disease, for example, have been used to predict ketosis (Gonzales et al., 2008), respiratory disease (BRD) (Toa◻-Rosenstein and Tucker, 2018), clinical lameness (Gonzalez et al., 2008), metritis (Schirmann et al., 2016; Neave et al., 2018) and mastitis (Fogsgaard et al., 2012; Sepúlveda-Varas et al., 2014). However, because core behaviors are essential for short-term survival, they will only start to decline at relatively late disease stages (e.g. Littin et al., 2008) and have low sensitivity during the early stages (Sepúlveda-Varas et al., 2014). A change in low-resilience behaviors such as social exploration and interaction, however, could potentially flag disease sooner because their immediate importance is further diminished when the animal’s energy resources are diverted to fight infection (Dawkins, 1990). We, therefore, here investigate whether any such changes are detectable in subclinical mastitis.
In cows, serum amyloid A (SAA), a non-specific inflammatory marker, primarily synthesised in the liver and readily measured in serum, is a key acute phase protein (APP) (Murata et al., 2004). APPs have demonstrated great utility in the identification of infectious disease. Increased SAA levels have been reported in cows with BRD (Joshi et al., 2018), reticuloperitonitis, metritis (Nazifi et al., 2008), and subclinical mastitis (Kovac et al., 2011; Nazifi et al., 2011; Kovacevi-Filipovic et al., 2012). A mammary isoform of SAA, synthesised by infected glands within the udder, is also upregulated in milk during sub-clinical mastitis (Eckersall et al., 2006; Kovac et al., 2011; Kovacevi-Filipovic et al., 2012). Although the main veterinary focus for salivary bioscience has been directed at companion animals (Prickett and Zimmerman, 2010; Cerón, 2019), enormous potential exists for non-invasive APP measurement in farm animal saliva; the presence of SAA in bovine saliva has been confirmed (Lecchi et al., 2012; Rahman et al., 2013). To date, very few studies have linked sickness behavior with physiological measures of inflammation in cattle. Des Roches et al. (2017) report correlations between pain indices (including attitude and attentiveness) and serum SAA prior to, and following, intra-mammary challenge with E. coli. A later study, utilising a similar mastitis model, identified lower mood (greater lethargy, dejection and su◻ering) in mastitic cows, and this was associated with clinical indicators and high SAA levels (des Roches et al., 2018).
The key aim of the current study was to identify any changes in SAA, social behavior and other behavioral signs of sickness associated with subclinical disease states (here mastitis) in dairy cows; this with a view to supporting early detection of disease and identification of more chronic, subclinical, inflammation states. To this end we: (a) compared the behavior of cows with subclinical mastitis with matched healthy cows and, (b) correlated key behavioral measures with both milk SCC (as a measure of local infection severity) and salivary SAA (as a measure of systemic inflammation). We thus investigate for the first time whether SAA measurement in bovine saliva has the potential to detect systemic inflammation associated with a sub-clinical condition.
2. MATERIALS AND METHODS
2.1. Ethics Statement
The study was conducted between October 2017 and February 2018 at Bristol Veterinary School dairy farm. The experimental procedures were approved by the Animal Welfare and Ethical Review Board at the University of Bristol.
2.2. Animals
Focal cows (n = 34) were part of an indoor commercial Holstein-Friesian dairy herd (n = 200) and resided within the low milk-yield group (approx. n = 80 cows) at the time of the study; having been part of the group for at least one month prior to data collection they were well-established within the social dominance hierarchy. Cows were housed within a free-stall barn; their section contained: 93 lying stalls with sand bedding, three drinking troughs, a swinging brush (DeLaval), automatic floor scrapers and vulcanized rubber floors in the front (feeding) alley. Cows were milked three times daily (the low yielders at 06:00, 14:00 and 22:00h) and fed a total mixed ration once daily (06:00h). The feed passage was accessible via self-locking yokes (n = 42) and an open section of post-and-rail feed barrier. Only clinically healthy non-lame cows (mobility score ≤1; AHDB, 2015) were selected for inclusion within the study. To control for confounding effects (air temperature, reproductive status, parity, and stage of pregnancy) data were collected from cows in matched pairs, whereby each pair comprised one cow with subclinical mastitis (SCM) and a clinically healthy control (CTRL). Cows with a SCC of >200 (x1000 cells/ml) were classified as SCM (Madouasse et al., 2010), while cows with a SCC of <100 (x1000 cells/ml) were classified as CTRL. The focal group comprised pregnant (n = 24) and nonpregnant (n = 10) cows, primiparous (n = 20) and multiparous (parity of 2: n = 2; 3: n = 2; 4: n = 6; 5: n = 4) individuals, and the time to expected calving date ranged from 55 - 236 days.
2.3. Somatic Cell Count
Composite quarter milk samples were collected at the second (afternoon) milking on the day prior to behavioral observations (Day 1). Somatic cells were manually counted using a standard direct microscopic methodology (ISO 13366-1, 1997) following staining with Newman-Lampert stain solution: Levowitz-Weber modification (Newman’s Stain Solution: modified, 01375, Sigma-Aldrich).
2.4. Behavioral Measures
Each focal cow was fitted with a coloured collar to facilitate recognition (Day 1). Two CCTV systems (N441L1T, Annke®, CA 91748, US), including six cameras were used to record video footage from the entire low-yield pen. Continuous behavioral data were scored retrospectively from video for each focal cow for 24h starting from 00:00h (Day 2); all measures are described in Table 1. The following behaviors were not analysed individually but included within the measure of total behavioral transitions (‘Trans’): eat sand, lick salt, paw sand, run, shake head, stand, and walk. Rumination was not logged as it proved difficult to definitively identify from the video footage. Tendencies for/against social proximity were investigated using nearest neighbour scores. When the focal cow was located at the feed barrier or resting within a cubicle the number of other cows (0, 1 or 2) in proximity were recorded. At the feed barrier proximity was defined as physical flank-to-flank contact. In a cubicle it was the occupation of (i.e. another cow recumbent within) an adjoining cubicle. To enable the calculation of distance moved (‘Dist’) the pen floorspace was hypothetically subdivided into 29 units (4.8 m wide); front and back passages were each divided into 13 units (F1-13 and B1-13, respectively), in addition to three raised trough areas (M1, M7, M13, Figure 1). The location of each cow was recorded at 5 min scan intervals and ‘Dist’ was calculated as the number of units crossed within a specific time-period.
Several different data sets were analysed. The main data set (24h) comprised all data for the 24h observational period. Behavior associated with fresh feed delivery was examined using 60 min of data collected from each cow immediately following their individual return to the home pen after first milking (1hPostM1). Diurnal differences in behavior between the two groups were assessed using hourly blocks of data for specific ‘key’ measures of interest (Diurnal). Data was not available for 06:00, 14:00 or 22:00h as the cows were in the milking parlour or collecting yard during these periods.
To account for differences in total time visible (i.e. due to variations in time spent within the parlour) data in the ‘24h’ and ‘Diurnal’ data sets were standardised to either: number per hour visible (behavioral events) or seconds per hour visible (behavioral states). Outliers (± 2SD) were removed prior to data analysis.
2.5. Saliva Collection and SAA
Saliva was collected (Day 3) using a cotton swab (SalivaBio Children’s Swab, Item No. 5001.06, Salimetrics) and then immediately stored at −80°C prior to analysis. SAA was measured in saliva from 31 cows, diluted 1:2, using a commercially available kit (Bovine Serum amyloid A protein ELISA Kit, EB0015, Finetest®, Wuhan Fine Biotech Co. Ltd.). To assess the suitability of the kit for use with saliva an assay validation was performed. To determine parallelism (linearity) a displacement curve, produced by double-diluting a pooled saliva sample with assay buffer, was compared to a standard curve. Percentage binding (as a percentage of that recorded for the zero standard) was calculated, in addition to the Log of the standard concentration (SAA standard) and the Log of the inverse of the dilution factor (saliva sample), e.g. 1:4 was transformed to Log(1/4). Parallelism was confirmed using a statistical test for the analysis of covariance (ANCOVA, SPSS). To measure assay accuracy the percentage recovery of exogenous SAA was calculated following the addition of 300 ng/ml SAA standard to a pooled saliva sample. Precision was assessed via intra- and inter-assay coefficients of variation (CV); the former was determined following the repeated measurement of aliquots of pooled saliva containing either high (quality control: QChigh) or low (QClow) endogenous SAA within the same plate, while the later was determined following the assay of QChigh and QClow samples in different plates.
2.6. Statistics
Following tests for normality (Shapiro-Wilk analysis), comparisons between the CTRL and SCM groups were made for all behavioral and physiological measures (Paired samples t-test or Wilcoxon signed-rank tests, IBM SPSS Statistics 24.0). Since the experimental design required the performance of multiple comparisons between measures there was an increased associated risk of Type I errors. Use of Bonferroni correction procedures has been highlighted as problematic (especially for animal behavioral studies, where sample sizes are often small) due to their tendency to increase Type II errors (Nakagawa, 2004). As an alternative to standard correction procedures we, therefore, calculated measures of observed (standardised) effect size in addition to p-values. Effect size measures the strength/magnitude of a relationship and, thereby, helps us to determine the strength of a statistical claim and whether a difference is real (i.e. it enables us to judge biological importance). Hedges’ g-value (Equations 1 and 2), also termed ‘Cohen’s d-value for paired samples’ (Hedges, 1981; Cohen, 1988; Nakagawa and Cuthill, 2007) and 95% confidence intervals (CI) for effect size (Equations 3 and 4), were calculated for all measures that met the assumptions of normality.
where x◻1 and x◻2 are the means of the two groups, σpaired is the pooled standard deviation, n is the number of data points, and s2 is the sample variance.
where seg is the asymptotic standard error for the effect size, n = n1 = n2, and r1,2 is the correlation coefficient between the two groups. For all behavioral measures that failed to meet the assumptions of normality, bootstrap effect size values (Hedges’ g-value with 95%CI, R = 2000) were computed using the software package ‘bootES’ (Gerlanc and Kirby, 2012; Kirby and Gerlanc, 2013) and R (Version 3.2.2., www.r-project.org/). Effect size statistics were interpreted as follows: (a) the size of the effect (based upon the estimated g-values: ≤0.39 = small, 0.40 - 0.79 = medium, ≥0.80 = large); (b) statistical significance (attributed to all measures where the associated 95%CI did not contain ‘0’) (Lee, 2016).
Interpretation of statistically non-significant p-values is possible using effect size confidence intervals in combination with the effect size (see Nakagawa and Foster, 2004). To identify measures that failed to reach statistical significance in the current study (24h data set) but that could potentially still be biologically important, we used information from pre-existing literature to set accepted relative difference levels (RDL%, Table 2).
For those measures where no relevant literature was available, an RDL% of 20% was employed; this was the average of our other RDL% levels. For each measure, relative difference values (RDV) were then calculated using the RDL% and the respective mean value from the CTRL group. 95%CIRDV were calculated using the confidence intervals from the effect size statistics and the (between-group) difference in means. In those cases where the 95%CIRDV did not include the RDV, we conclude (with 95% confidence) that the current study showed no important biological effect for that measure; we refer to these as ‘biologically unimportant’. In cases where the 95%CIRDV did include the RDV, we conclude that a difference was inconclusive but plausible; we refer to these as ‘biologically inconclusive’. For example, if CTRL cows performed more body pushes than the SCM cows, yet this difference failed to reach statistical significance (P≥0.10), using p-value alone we would dismiss this behavior as being unaffected by subclinical inflammation. However, if the RDV for this behavior was within the 95%CIRDV range (e.g. RDV = 0.08, 95%CIRDV = −0.08 to 0.27) we can conclude that, although this effect is biologically inconclusive based upon our evidence, the difference may become significant given a larger sample size. Alternatively, if the RDV, in the above example, was 0.3, then we would conclude that the effect was biologically unimportant. To identify correlations between physiological (SCC and SAA) and behavioral measures (24h data set) we performed curve estimation regression statistics (SPSS: ANOVA, coefficient of determination).
3. RESULTS
3.1. Behavioral Differences
3.1.1. Core behavior
Paired t-tests suggested no significant differences in time spent feeding (‘Feed’), drinking (‘Drink’), or lying (‘Lie’) between SCM and CTRL groups (Tables 3 and 4), nor was subclinical mastitis considered to have biologically significant effects on these core maintenance behaviors (‘Feed’: RDV = 82.39 s, 95%CIRDV = −0.03 to 0.03 s; ‘Drink’: RDV = 6.86 s, 95%CIRDV = −1.33 to 3.99 s; ‘Lie’: RDV = 200.19 s, 95%CIRDV = −23.54 to 79.67 s). Over the 24h period SCM cows spent more time lying with their head on their flank than CTRL cows (‘HOF’, Paired t-test: p = 0.046; Hedges’ g: significant medium effect). CTRL cows were more active, performing more behavioral transitions (‘Trans’ 24h, p = 0.040; large; 1hPostM1, p = 0.041; large) and moving over a greater distance than the SCM cows (‘Dist’: 24h, p = 0.032, large; 1hPostM1, p = 0.098; medium).
Although SCM cows tended only to perform statistically more ‘Comfort’ behavior following morning milking (1hPostM1: p = 0.073, medium), confidence intervals for effect size differences classified both measures of self-grooming to be biologically inconclusive over 24h (‘Brush’: RDV = 8.70 s, 95%CIRDV = −2.74 to 16.45 s; ‘Comfort’: RDV = 4.23 s, 95%CIRDV = −2.81 to 5.42 s). This indicates that, given a larger sample size, significant differences may have become apparent (i.e. higher brush use by CTRL cows and more comfort behavior by SCM cows). No difference in environmental exploration was evident between the groups over the 24h period, nor was this deemed to be biologically important in the context of this study (‘ExpEnv’: RDV = 11.08 s, 95%CIRDV = −1.30 to 4.66 s); however, CTRL cows did explore their environment more than SCM cows 1hPostM1 (‘ExpEnv’: p = 0.031, large).
3.1.2. Social behavior and pen-mate proximity
No differences were observed in the overall performance (24h), or receipt, of social behavior between groups (Tables 5 and 6). Overall performance of cumulative social behavior was classified as biologically unimportant in the context of this dataset, while receipt of social behavior was biologically inconclusive (‘SocG’: RDV = 0.50, 95%CIRDV = −0.26 to 0.43; ‘SocR’: RDV = 0.44, 95%CIRDV = −0.21 to 0.64). CTRL cows performed more social exploration (‘ExpSoc’: 24h, p = 0.009, large; 1hPostM1, p = 0.026, medium), more head butts (‘HBG’: 24h, p = 0.043; 1hPostM1, p = 0.055, medium) and more head pushes (‘HPG’: 1hPostM1, p = 0.027, large) than the SCM cows. Confidence intervals classified all other agonistic measures which failed to reach statistical significance as being biologically inconclusive (‘FocSocG’: RDV = 0.40, 95%CIRDV = −0.31 to 0.73; ‘BPG’: RDV = 0.08, 95%CIRDV = −0.08 to 0.27; ‘CG’: RDV = 0.02, 95%CIRDV = 0.00 to 0.15; ‘HPG’: RDV = 0.03, 95%CIRDV = −0.03 to 0.18; ‘HSG’: RDV = 0.17, 95%CIRDV = −0.12 to 0.63).
In the hour after first milking (1hPostM1) CTRL cows received significantly more social behavior (‘SocR’: p = 0.038, large), agonistic behavior (‘FocSocR’: p = 0.049, large), body pushes (‘BPR’: p = 0.022, large) and head butts (‘HBR’: p = 0.080, medium) than SCM cows. Overall (24h), CTRL cows also received more challenges (‘CR’: p = 0.050, large), while SCM cows received more head swipes (‘HSR’: p = 0.070, medium). With the exception of ‘BPR’, which was classified as biologically unimportant (‘BPR’: RDV = 0.08, 95%CIRDV = −0.01 - 0.02), all other measures that failed to reach statistical significance were deemed to be biologically inconclusive (‘HBR’: RDV = 0.07, 95%CIRDV = −0.05 to 0.13; ‘HPR’: RDV = 0.02, 95%CIRDV = −0.00 to 0.14; ‘FocSocR’: RDV = 0.34, 95%CIRDV = −0.13 to 0.40; ‘MutHdButt’: RDV = 1.13 s, 95%CIRDV = −0.19 to 8.03 s).
CTRL cows were more reactive, i.e. more likely to be displaced, than SCM cows following the receipt of agonistic behavior cumulatively (‘%FocSocR+D’: p = 0.010, large), a head butt (‘%HBR+D’: p <0.001, large), head push (‘%HPR+D’: p = 0.037, large), or a head swipe (‘%HSR+D’: p = 0.048, medium). All reactivity measures that failed to reach statistical significance were classified as biologically unimportant in the context of this study (‘%BPR+D’: RDV = 5.15%, 95%CIRDV = −1.13 to 1.82%; ‘%CR+D’: RDV = 8.44%, 95%CIRDV = −2.43 to 8.35%). Although CTRL cows were observed to allogroom significantly more than SCM cows during the 24h period (‘AlloG’: p = 0.047, medium), no overall difference in the receipt of allogrooming was evident, and this measure was classified as biologically unimportant (‘AlloR’: RDV = 11.02 s, 95%CIRDV = −1.14 to 1.67 s). CTRL cows were allogroomed more than SCM cows 1hPostM1 (p = 0.012, medium).
SCM cows spent a significantly greater proportion of their time at the feed passage flanked by two neighbours (‘%FB_2NN’: p = 0.019, medium), and a significantly greater proportion of their time at the open section of the feed barrier (‘%FB_Open’: p = 0.032, large), than did the CTRL cows. All other measures of social proximity were classified as biologically unimportant within the context of this study (‘%FB_0NN’: RDV = 8.28%, 95%CIRDV = −0.11 to 4.75%; ‘%C_0NN’: RDV = 7.75%, 95%CIRDV = −0.06 to 2.58%; ‘%C_2NN’: RDV: 4.38%, 95%CIRDV = −2.69 to 1.93%).
3.2. Differences in Diurnal Behavior Patterns
A combination of paired-sample and post-hoc (effect size statistic) testing between the two groups revealed differences in diurnal patterns of activity and social behavior (Figure 2). The SCM cows were more active between: (a) 00:00 and 01:00h, when they performed more exploratory behavior and moved over a greater distance than the CTRL cows, and (b) 13:00 and 14:00h, when they performed more exploratory and social behavior. CTRL cows were more active than the SCM cows during three periods: (a) between 02:00 and 03:00h they performed more social exploration, received more social behavior, and walked a greater distance, (b) during 05:00 to 06:00h they lay less, moved further, and performed more behavioral transitions, exploratory, and agonistic behavior, and (c) between 16:00 and 17:00h they performed more behavioral transitions and performed/received more agonistic behavior.
3.3. Correlations Between Physiology and Behavior
3.3.1. Assay validation
Parallelism (F1,9 = 3.46, p >0.05) was confirmed between serial dilutions of saliva (range: 1:4 to 1:64) and SAA standards (range: 0, 9.38, 18.75, 37.5, 75, 150, 300 ng/ml), indicating that the ELISA kit was suitable for use with bovine saliva. Recovery of 300 ng/ml SAA from a spiked saliva sample was 93.76 ± 4.63% (n = 10). The intra-assay CV was 3.09% (250.87 ± 7.75 ng/ml, n = 10) for QClow and 4.68% (1360.33 ± 63.70 ng/ml, n = 10) for QChigh. The inter-assay CV was 2.77% (246.06 ± 6.81 ng/ml, n = 2) for QClow and 3.89% (1323.96 ± 51.43 ng/ml, n = 2) for QChigh.
3.3.2. SAA and SCC
The average SCC per group was: CTRL = 48.29 ± 28.33 (x1000 cells/ml); SCM = 351.12 ± 176.73 (x1000 cells/ml). A trend was evident for a higher concentration of salivary SAA in the SCM cows (CTRL = 343.42 ± 269.60 ng/ml, SCM = 519.59 ± 315.43 ng/ml; t1,12 = 1.93, p = 0.076), and a weak positive logarithmic relationship was evident between SCC and SAA (F(1,26) = 6.26, p = 0.019; R2 = 0.194; y = 113.99ln(x) − 81.384, Figure 3).
3.3.3. SAA, SCC and behavior
Of the 34 behavioral measures to have had correlation analyses calculated against SAA and SCC, 24 were significant; most relationships identified were weak (Table 7). No correlation (SAA or SCC) was evident for: ‘HOF’, ‘BPG’, ‘CG’, ‘HBR’, ‘HPG’, ‘%BPR+D’, ‘AlloR’, ‘%FB_2NN’, ‘Brush’ or ‘Comfort’. A weak positive correlation between SAA and ‘Lie’, and moderate negative correlations between SAA and both ‘Feed’ and ‘Drink’, indicate that as systemic inflammation rose consumption dropped and lying increased. Positive correlations between SAA and both ‘ExpEnv’ and ‘ExpSoc’, suggest that cows with higher inflammation levels were more explorative. However, a negative correlation between SCC and ‘ExpSoc’ was also observed. Quadratic relationships were evident between SCC and both ‘Trans’ and ‘Dist’; these described an initial drop in activity, as SCC increased to approximately 300 (x1000 cells/ml), followed by an increase as SCC continued to rise (Figure 4). Positive correlations between SAA and ‘SocR’, ‘FocSocR’, ‘CR’, and ‘HPR’, indicate that cows with higher levels of systemic inflammation were receiving more socially agonistic behaviors.
Negative correlations between SCC and both ‘BPR’ and ‘HPR’ imply that certain agonistic behaviors were primarily directed at cows without intra-mammary inflammation. Negative correlations between SAA and ‘SocG’, ‘FocSocG’ and ‘HSG’ indicate that the performance of social behavior decreased with systemic inflammation. Increasing SCC levels were also associated with the performance of fewer head butts and the receipt of more head swipes.
Negative correlations were observed between SAA and ‘%FocSocR+D’, ‘%CR+D’, ‘%HBR+D’ and ‘%HSR+D’, and between SCC and both ‘%FocSocR+D’ and ‘%HBR+D’. This indicates that cows with greater inflammation were less likely to move away (i.e. be displaced) after receiving agonistic behavior, suggesting lower social reactivity. However, a positive correlation between SAA and ‘%HPR+D’ suggests that, following receipt of a head push, cows with greater systemic inflammation were displaced more frequently. A negative correlation between SCC and ‘AlloG’ indicates that the performance of allogrooming decreased with increased mammary inflammation. Finally, positive correlations between both inflammatory markers and ‘%FB_Open’ reveal that the self-locking feed barriers were used less as inflammation increased.
4. DISCUSSION
The main purpose of this study was to identify salivary SAA, social and other behavioral changes associated with subclinical inflammation (mastitis) in cows. Salivary SAA was shown to have potential as a marker of low-level systemic inflammation because levels were found to be higher in cows with subclinical mastitis, and higher levels were associated with several key sickness behaviors. Furthermore, SCM cows displayed lower ‘activity’, ‘sociality’ (including the performance and receipt of multiple social behaviors) and ‘social reactivity’, and demonstrated a shift in activity peaks for several behaviors to quieter times of the day.
4.1. Salivary SAA
Positive correlations between SCC and non-salivary SAA have often been reported from cows with clinical and sub-clinical mastitis (serum: des Roches et al., 2017; milk: O’Mahony et al., 2006; Akerstedt et al., 2007; Pyörälä et al., 2011) and we report here, for the first time, the same for salivary SAA. SAA in saliva thus appears to offer potential as a non-invasive means of detecting subclinical infection. During field conditions, several bacterial strains can cause mastitis of varying duration and degree (Verbeke et al., 2014), and it is likely that the concentration of SAA in saliva will vary accordingly. Pyörälä et al. (2011) detected significant differences in SAA (milk) collected from cows with spontaneous mastitis caused by different pathogens; low SAA was associated with A. pyogenes, while high concentrations were associated with E. coli.
4.2. Social Behavior
Although no differences in cumulative social behavior given or received were evident between the two groups overall (24h), the CTRL cows received significantly more social behavior than the SCM cows following morning milking (1hPostM1); this provides evidence for diurnal differences in behavior. By dividing social behavior into the broad categories of socio-negative (i.e. agonistic competitive) and socio-positive (affiliative) we were able to identify specific disparities.
4.2.1. Agonistic competitive behavior
Sick cows are often reported to perform fewer agonistic interactions and competitive displacements from the feed-bunk (bacterial lameness: Galindo and Broom, 2000; 2002; sub-clinical metritis: Huzzey et al., 2007; Patbandba et al., 2012; clinical and sub-clinical mastitis: Sepúlveda-Varas et al., 2014; 2016), and from cubicles (Jensen and Proudfoot, 2017), than healthy individuals. In addition, sick cows often receive more agonism, and are displaced more frequently, than healthy cows (bacterial lameness: Galindo and Broom, 2002; metritis: Patbandba et al., 2012; Neave et al., 2018; Lomb et al., 2018; metritis and sub-clinical ketosis: Schirmann et al., 2016). On this basis we predicted our SCM cows to also perform less and receive more agonistic behavior than the CTRL cows. Counter to these expectations the CTRL group received more challenges (24h), and more head butts, body pushes and total aggression (1hPostM1), than the SCM group, supported by negative correlations between SCC and the receipt of both head and body pushes. Presumably, the healthy animals partook in more physical contact and jostling as part of actively re-establishing dominance, since aggressive competitive interactions are key to establishing and maintaining social order within dynamic groups (Val-Laillet et al., 2008).
The receipt of head swipes was the one agonistic measure that was significantly higher in our SCM group (24h). Since this is a common social behavior, occurring almost exclusively at the feed barrier as part of feed competition (i.e. a means of displacing immediate neighbours), this finding does correspond with predictions of ‘sickness’ and the wider literature. Interestingly, we also observed positive correlations between SAA and several measures of social and agonistic receipt (‘SocR’, ‘FocSocR’, ‘CR’, ‘HPR’), indicating that cows with higher levels of systemic inflammation also were more frequently the recipients of agonistic behavior. We cannot discount the possibility that SAA upregulation occurred within our CTRL group due to early undiagnosed non-mastitic infection or following exposure to social stress. Upregulation of C-Reactive Protein (an APP known to increase during illness and stress) has been reported in zoo-housed gorillas following an aggressive encounter (Fuller and Allard, 2018). In the current study saliva samples were collected the day after behavioral observations were made, therefore an elevation in SAA may also occur as a consequence of agonistic encounters experienced during the previous day.
In line with the hypothesis that social behavior should decrease with inflammation/sickness, and our observation that CTRL cows performed more agonistic behavior, SAA was negatively correlated with ‘SocG’, ‘FocSocG’ and ‘HSG’, and SCC was negatively correlated to ‘HBG’. Although social rankings were not calculated in the current study it is possible that the social rank of an individual could be influenced by the effects of disease due to a loss of competitive vigour. Dominant animals frequently displace subordinate cows from the feed barrier (DeVries et al., 2004; Huzzey et al., 2006), and subordinate animals adjust their eating patterns accordingly (DeVries et al., 2004). In our study CTRL cows performed more agonistic behavior immediately before the first milking (05:00h) and mid-afternoon (16:00h), while SCM cows performed more prior to the second milking (13:00h). Focusing activities outside of peak times may be a means of avoiding agonistic interactions with socially dominant individuals but, due to high stocking densities, there will always be an immediate social environment to manage. It is conceivable, but beyond the reach of our data, that agonistic behavior performed by the SCM cows was aimed at other sick or low-ranking individuals employing similar competitive tactics.
4.2.2. Social reactivity
Llonch et al. (2018) identified a group of cows that were more reactive to the presence of conspecifics at the feed barrier; frequent feeding interruptions lead to shorter, but more frequent, visits to the feeder. Since subordinate or sick cows are more likely to engage in avoidance behavior in response to social confrontation (Huzzey et al., 2006; Goldhawk et al., 2009; Proudfoot et al., 2009) we may anticipate such individuals to also display greater reactivity (i.e. be more likely to move away when challenged at the feeder). However, in the current study the opposite was true; CTRL cows were more likely to be displaced than SCM cows following the receipt of agonistic behavior. Not only were CTRL cows more reactive than SCM cows, but reactivity appeared to decrease with increasing inflammation (negative correlations were observed between SAA and ‘%FocSocR+D’, ‘%CR+D’, ‘%HBR+D’ and ‘%HSR+D’, and between SCC and both ‘%FocSocR+D’ and ‘%HBR+D’). This observation could be explained by social environment. Choice of feeding position has been shown to be affected by dominance relationships; dissimilar neighbours (low/high rank) are known to maintain a greater distance of separation than individuals of similar rank (Manson and Appleby, 1990). If the CTRL cows were less discriminatory, regarding their social environment, then they may have been more likely to receive aggressive encounters from dominant individuals and reacted accordingly. Conversely, if the SCM cows proactively avoided dominants, and preferentially selected the company of other sick or lower ranking cows, then moderate competitive aggression received from an individual of similar standing may have been tolerable (i.e. not elicited displacement).
4.2.3. Affiliative behavior
Very little research has been conducted on allogrooming and illness in cows. Galindo and Broom (2002) observed lame cows to be allogroomed more than non-lame cows, and this was interpreted as a self-instigated coping strategy triggered by pain/discomfort. Cows appear to find comfort in being licked; individuals who solicit more licking are licked more frequently (Benham, 1984). We predicted that allogrooming would be lower in the SCM group since subordinate individuals are licked, and lick, less frequently than high ranking individuals (Napolitano et al., 2007), allogrooming decreases more in low-ranking individuals under conditions of increased competition (Val-Laillet et al., 2008), and mild ‘sickness’ is likely to reduce motivation for luxury behavior. As hypothesised, CTRL cows performed more allogrooming than SCM cows overall (24h), confirmed by a negative correlation between SCC and ‘AlloG’; in addition, they were also allogroomed more than SCM cows in the hour following morning milking. Since social grooming serves a variety of functions in cattle, including roles in hygiene, the provision of pleasure, the maintenance of social bonds and in lowering social tension (Sato et al., 1991; 1993), it is possible that prolonged suppression of this behavior could have negative implications for welfare and fitness.
4.2.4. Social avoidance
Sickness-driven social avoidance is well documented in lab-animals and humans, and can be predicted by the action of pro-inflammatory cytokines on the CNS (Kent et al., 1992; Bluthe et al., 1996; Dantzer and Kelley, 2007; Arakawa et al., 2010). Due to the limited opportunity for social avoidance within intensive systems this behavior has been relatively understudied in sick farm animals. In the current study CTRL cows performed more social exploration than the SCM cows (and a negative correlation was evident between SCC and ‘ExpSoc’), potentially due to the SCM cows actively avoiding social interaction. The unexpected weak positive correlation between SAA and ‘ExpSoc’ may be, at least partially, explained by the presence of both pre-clinical and post-clinical cows within our focal group; i.e. early-stage mastitic cows (low SAA), demonstrating sickness-driven reductions in ‘ExpSoc’, in combination with individuals in remission (high SAA), demonstrating normal baseline ‘ExpSoc’ (see Section 4.4.4).
The prevalence of agonistic behavior at the feed barrier is known to be influenced by barrier design; self-locking yokes have vertical bars which separate the necks of adjacent cows, and these are better at reducing competitive interactions (displacements) compared to open post- and-rail barriers (Endres et al., 2005; Huzzey et al., 2006). In the current study SCM cows spent a significantly greater proportion of their time feeding at the open section of the barrier than did the CTRL cows, and a positive correlation between SAA and ‘%FB_Open’ indicates this preference to increase with rising systemic inflammation. Although this appears to be counter-intuitive, other factors are likely to contribute to the choice of feeding location; e.g., in the open section cows have better visibility and are more quickly able to withdraw from potential agonistic interactions.
4.3. Activity
The observation that our SCM cows made fewer behavioral transitions and moved over a shorter distance than CTRL cows is in agreement with other studies that describe reduced activity prior to the clinical diagnosis of mastitis (Fogsgaard et al., 2012; Kester et al., 2015; Stangaferro et al., 2016; Veissier et al., 2017; King et al., 2018). The quadratic relationships between SCC and both ‘Trans’ and ‘Dist’ described in the current study are of interest because mastitic cows have also been reported to display increased activity (Siivonen et al., 2011; Medrano-Galarza et al., 2012), presumably due to udder discomfort and an associated reduction in lying time. Jadhav et al. (2018) argue that the threshold SCC value to delineate subclinical mastitis from normal should be 310, rather than 200 (x1000 cells/ml), as conventionally judged (e.g. Madouasse et al., 2010). This higher value closely corresponds with the parabola vertex in both quadratic plots (Figure 4), of approx. 300 (x1000 cells/ml); i.e. the point at which activity once again begins to rise.
The circadian rhythm of cow activity is known to become disrupted during disease (Veissier et al., 1989; 2017; Kauppi, 2014). Veissier et al. (2017) observed that diseased cows may not consistently decrease their activity, but instead focus their activities within specific time periods; mastitic cows were observed to be hyperactive throughout the day, whereas lame cows were hyperactive at night. We identified two periods during which our SCM cows were more active than the CTRL animals (00:00 to 01:00h and 13:00 to 14:00h); presumably these represented quieter periods, when a proportion of the herd, including socially dominant individuals, were resting.
4.4. Core and Non-Social Behavior
4.4.1. Ingestion
Changes in feeding behavior have long been used to diagnose the onset of illness (Weary et al., 2009). Although we observed a negative correlation between SAA and feeding duration, as would be hypothesised to occur with sickness, the average inflammatory response within our SCM group overall was not sufficiently pronounced to trigger obvious anorexia, as compared to healthy controls. Gonzalez et al. (2008) report variability in feeding behavior relating to naturally occurring udder disorders; some cows demonstrated a decrease in feeding duration with the onset of mastitis, while others showed no change. It is possible that aspects of feeding behavior, other than duration, may have been altered. Barn-housed cattle demonstrate highly synchronised feeding activity, with large peaks in both feeding and social competition coinciding with fresh food delivery, and smaller peaks following milking (DeVries and von Keyserlingk, 2005; Dollinger and Kaufmann, 2013). Mastitic cows, presumably to avoid adverse social interactions, have been shown to feed at less popular times such as early afternoon (Schirmann et al., 2016). Our study identified such a period, between 13:00 and 14:00h (immediately prior to second milking) as one in which the SCM cows fed for longer than the CTRL cows. Sepúlveda-Varas et al. (2016) observed a decrease in feed intake (but not duration) prior to the diagnosis of clinical mastitis which may be attributed to underlying malaise.
Water and feed intake are positively related in cattle (Kume et al., 2010); however, drinking tends to be less affected by health than feeding (Hart, 1988). Water is more immediately vital for maintaining bodily functions (Kyriazakis and Tolkamp, 2011), and since drinking takes less time than food consumption it is less at risk from social competition at the trough (Huzzey et al., 2007). Although a reduction in water consumption has been reported in cows with mastitis (Lukas et al., 2008; Siivonen et al., 2011), and we observed a negative correlation between SAA and drinking duration, the level of systemic inflammation within our SCM group may have been too low, and/or our sample size too small, to induce a group difference.
4.4.2. Lying
Lying is a highly prioritised behavior in cattle due to its importance in rumination (Jensen et al., 2004; 2005; Munksgaard et al., 2005); dairy cows spend approximately 11h/day recumbent (Ito et al., 2009; 2010). Increased lying duration, as a means of conserving energy and facilitating recovery, is a key adaption for sickness, and a positive correlation between SAA and ‘Lie’ was evident within our test population. Although extended lying duration has been frequently reported during cattle illness (e.g. BRD: Toaff Rosenstein et al., 2016; moderate lameness: Weigele et al., 2017; metritis: Huzzey et al., 2007; Sepúlveda-Varas et al., 2014; Barragan et al., 2018), lying may decrease during mastitis (Yeiser et al., 2012; Medrano-Galarza et al. 2012; Fogsgaard et al., 2012; 2015), presumably due to udder pain (Cyples et al., 2012). Although we observed no difference in ‘Lie’ between our two groups, the SCM cows were observed to lie with their heads held against their flank more than the CTRL cows. This posture is primarily associated with rapid eye movement (REM) sleep; however, cows are also known to display non-rapid eye movement (NREM) sleep and drowsing in this position (Ternman et al., 2013), and NREM (deep) sleep often increases during infection (Bryant et al., 2004; Opp, 2005).
4.4.3. Self-grooming
Although grooming is a comfort activity that cows are highly motivated to perform (McConnachie et al., 2018), brush use is a luxury activity, characterised by low behavioral resilience (Dawkins, 1990), and has been shown to decrease during disease (sub-clinical metritis: Mandel et al., 2017; lameness: Weigele et al., 2017; BRD: Toaff Rosenstein et al., 2016). A trade-off between brush location and the sensitivity of brush use for detecting stress and morbidity (Mandel et al., 2013; 2017) may help to explain why we failed to observe a significant difference between our two groups, or indeed a correlation between SAA and ‘Brush’. The brush in the current study was readily accessible to all cows in the group with minimal effort, being located central to many resources including the feed barrier, a water trough and cubicles. Much variation exists in the reporting of self-grooming (licking) following illness and/or immune challenge in cows; studies report a decrease (LPS: Borderas et al., 2008; mastitis: Fogsgaard et al., 2012; BRD: Toaff-Rosenstein and Tucker, 2018), an increase (lameness: Almeida et al., 2008), and no di◻erence (mastitis: Siivonen et al., 2011).
In the current study SCM cows performed more comfort behavior (including self-licking) than CTRL cows immediately following morning milking. This may be a response to mild udder discomfort or as a substitute for allogrooming (see section 4.2.3).
4.4.4. Environmental exploration
Des Roches et al. (2017; 2018) report behavioral changes (including reduced attentiveness) during the pre-clinical phase of an experimental mastitis model (prior to the upregulation of SCC and serum SAA), and during the acute phase (coinciding with raised levels of SCC and SAA), but not during the remission phase, even although high levels of SCC/SAA were still evident. This suggests that a peak in serum SAA corresponds with the remission, rather than the acute, phase of inflammation. The weak positive correlation between SAA and environmental exploration (‘ExpEnv’) described in the current study was unexpected since exploratory behavior would be predicted to decline with sickness/inflammation. If our test cohort contained a proportion of individuals in the pre-clinical phase (low salivary SAA but demonstrating sickness-lowered social exploration) and a proportion of individuals in the remission phase (high SAA with healthy baseline levels of social exploration) then this may offer an explanation. One of the behaviors included within the ‘ExpEnv’ measure was ‘explore sand’. Cows can spend a long-time sniffing sand prior to selection of a cubicle to lie down in. Although our SCM cows did take longer to lie down than the CTRL cows this difference failed to reach significance (data not shown). Ruminants generally display low baseline levels of exploratory behavior when maintained in intensive housing, since it is a largely unstimulating environment (De Rosa et al., 2009). Although a reduction in ‘ExpEnv’ was not evident in our SCM group overall (24h), the CTRL cows did display more interest in their surroundings than SCM cows during the hour following the morning milking, which coincided with the provision of the single daily meal. CTRL cows also displayed more exploratory behavior between 05:00 – 06:00h (the hour prior to morning milking/feed delivery), presumably in anticipation of what was to come. SCM cows, conversely, dominated exploratory behavior at quieter times of the day (00:00 – 01:00h and 13:00 – 14:00h), providing further evidence for a diurnal shift in activity.
4.5 Future Research
Using effect size statistics on the 24h data set, we classified the between-group differences in several measures as being biologically ‘inconclusive’; i.e., effect size differences between the treatment groups remain plausible but were not conclusive given our sample size, and these, therefore, provide a promising focus for future research into behavioral correlates of subclinical disease states in cows. They included brush use, body push given, challenge given, head swipe given, head butt received and mutual head butt. On the basis of several unexpected correlations within the current study (significant, but occurring in the opposite direction than predicted), and that disease models have demonstrated that peak immunity (APP levels) occurs during remission and persist after the recovery of sickness behavior (des Roches et al., 2017, 2018), we recommend further studies to investigate the association between inflammatory markers and behavior over natural disease progression. In humans, sickness is characteristically accompanied by a feeling of ‘malaise’, an affective state that involves the negative subjective experience of depression, lethargy and anhedonia, and is induced by pro-inflammatory cytokines as part of the body’s sickness response (Dantzer et al., 2004). Whether subclinically-mastitic cows similarly experience malaise or a ‘malaise-like’ affective state is not known (see Weary et al., 2009). However, it does seem possible based on the evidence of pro-inflammatory cytokine-induced anhedonia and depression-like states in rats (Dantzer et al., 2008), and therefore also merits further investigation.
5. CONCLUSIONS
By studying the behavior of a group of matched-pair cows over short distinct time periods (1h and 24h) we identified that sub-clinical mastitis (SCM) was associated with a reduction in activity, social exploration, the receipt of affiliative behavior, social reactivity (following the receipt of agonistic behavior), and an increase in the receipt of head swipes, compared to clinically healthy control (CTRL) cows. Several of these measures are low-resilience behaviors, which have previously been highlighted as having potential for early illness detection since they are expected to decrease earlier than core activities (Weary et al., 2009). Although no difference in any core maintenance behavior (feeding, drinking, lying duration) was detected, the SCM cows did demonstrate a preference for risk-adverse ‘within-herd feeding’, spending a greater proportion of time feeding in direct contact with two neighbours, and spending a lower proportion of their time feeding at the self-locking feed barriers than the CTRL cows. We present evidence for diurnal differences in the daily behavioral routine between the two groups, which indicates that SCM cows shift their activity to quieter times of the day. It seems likely that this is a tactic employed by the SCM cows to actively avoid agonistic encounters since the CTRL cows were more likely to perform and receive agonistic behavior. A positive relationship between SCC and SAA was observed, indicating salivary SAA (a marker of systemic inflammation) to be a potential physiological marker of subclinical mastitis. The majority of our behavioral measures was also found to correlate with salivary SAA in a direction consistent with sickness behavior. Taken together, these findings demonstrate that physiological and behavioral changes associated with subclinical mastitis in cows are consistent with predictions for low-level sickness responses.
COMPETING INTERESTS STATEMENT
The authors have no competing interests to declare.
ACKNOWLEDGEMENTS
This study was funded by The John Oldacre Foundation. We would also like to thank the staff at Wyndhurst Farm for facilitating our research.
Footnotes
gina.caplen{at}bristol.ac.uk; suzanne.held{at}bristol.ac.uk