Flexible feeding: Dairy cow personality affects changes in feeding behavior and milk production under feed competition conditions

The objective of this study was to determine the effect of individual cow personality traits on feeding behavior and production under low levels of feeding competition, and to determine whether personality traits influence how feeding behavior changes in response to greater feeding competition. Forty-two Holstein cows were assigned to 1 automated feed bin per cow (low competition condition) from 15 to 28 d in milk (DIM; period 1, P1), and 2 feed bins per 3 cows (higher competition condition) from 63 to 76 DIM (period 2, P2). A total mixed ration (TMR) was fed into the automated feed bins which recorded each feed bin visit time, duration, and intake. Cow personality traits were assessed at 21 DIM during P1 and at 70 DIM during P2 using a combined arena test, measuring behavioral responses to a novel environment, novel object, and novel human. Principal components analysis of behaviors observed during the P1 combined arena test revealed 1 factor (interpreted as active-explorative) from the novel environment test explaining 51% of the variance, and 3 factors (interpreted as fearfulness, active-explorative, and sociability toward conspecifics) from each of the novel object (76% cumulative variance) and human (75% cumulative variance) tests. The principal components analysis of behaviors observed during the P2 combined arena test revealed 2 factors jointly from the environment, object, and human tests (interpreted as fearfulness and active-explorative) that together explained 68% of the variance. Fearfulness and active-explorative trait scores were correlated across P1 and P2, indicating stability of personality over a challenging period and advancing DIM. In P2 when competition for feed was increased at greater stage of lactation, the more active-explorative cows appeared to make few alterations to their feeding behavior, yet still maintained their milk yield, compared with lower competition in P1. In contrast, cows who were more fearful increased their feed bin visits from P1 to P2, and less fearful cows


INTRODUCTION
There is increasing evidence across many species that individual animals differ in their behavioral expression.This behavioral variation between individuals can be explained by animal personalities, with specific aspects of an animal's behavioral repertoire that are consistent across contexts referred to as personality traits (Koolhaas et al., 1999;Réale et al., 2007;Carter et al., 2013).However, intra-individual consistency in behavior does not mean that behavior cannot change with physiological, social or environmental conditions, but rather that the degree of flexibility in behavior within individuals generally remains consistent over time (Finkemeier et al., 2018).
The expression of personality traits is often observed when animals are faced with challenging situations.Dairy cattle will experience changes in physical and social aspects of their feeding environment, such as feeding space and social regrouping changes (DeVries, 2019).How individuals respond to these situations, in addition to other stressful management practices, may contribute to variability in feeding behavior of grouphoused cattle (Melin et al., 2005;Meagher et al., 2017;Llonch et al., 2018).Understanding why this variability occurs is important in informing management practices that may be adjusted to meet the needs of different individuals, thus improving animal welfare and efficiency of the production system.
For some individual animals, adjusting their feeding behavior may be a coping mechanism in response to changes in their environment (Neave et al., 2018b).The ability to alter behavior to cope adaptively with environmental change is referred to as behavioral plasticity, whereby some individuals are more flexible or "plastic" in their reaction to environmental change and adjust their behavior accordingly (Dingemanse and Wolf, 2013).There is limited research in production animals that has investigated how individual personalities respond to social changes, particularly in the feeding environment.Research in pigs has demonstrated that variation in meal patterns could be explained by individual behavioral strategies in response to altered competition at the feeder (Boumans et al., 2018).Furthermore, dairy cows have been reported to increase variability in their feeding patterns when competition to access feed was increased (Crossley et al., 2017(Crossley et al., , 2018)).This research suggests that individuals may be relatively flexible or plastic in their feeding behavior when faced with a change in their feeding environment.
In addition, our previous research has demonstrated that more fearful cows have greater daily variation in concentrate intake in an automated milking system, and are less likely to consume greater allocations of concentrate (Schwanke et al., 2022).Thus, it is possible that individuals may vary in the degree of plasticity they demonstrate based on their personality traits.
The objectives of this study were (1) to determine the effect of dairy cow personality traits on feeding behavior in early lactation in a situation with low feeding competition, and (2) to determine whether personality traits influence how feeding behavior changes in response to greater feeding competition later in lactation.It was predicted that (1) the effects of personality traits on feeding behavior would be less pronounced in a low competitive situation, (2) personality traits would be consistent as stage of lactation and feeding competition increased, and (3) less fearful cows would have more consistent feeding behavior (i.e., inflexible) compared with more fearful cows across the 2 periods, while more explorative cows would adjust their feeding behavior (i.e., flexible), in response to increased feeding competition, whereas less explorative cows would be inflexible.

Animals and Housing
Forty-two lactating Holstein cows (16 primiparous and 26 multiparous; mean ± SD parity = 2.4 ± 1.5) were enrolled in the study at the University of Guelph, Elora Research Station, Ontario Dairy Research Centre (Elora, Canada).Cows had to be free from any health concerns to be eligible for inclusion in the study.Cows were enrolled at 7 DIM and their feeding behavior and milk production was observed for the first 14-d observation period (P1) starting at 15 DIM, and for a second 14-d observation period (P2) starting at 63 DIM.For the duration of the study, all cows were milked 2×/d on a rotary milking parlor (DeLaval, Tumba, Sweden).Cows were housed within 1 of 4 pens; in those pens cows had access to 30 lying stalls, 1 cow brush, and were offered ad libitum access to water from 2 water troughs located at each end of the pen.The stalls were laid out tail-to-tail in 2 rows of 15.Each stall measured 295 cm in total length and 127 cm in width, with the neck rail positioned 188 cm from the rear curb and 125 cm above the stall mattress base (Pasture Mat; Pro-Mat; Woodstock, Canada).Stalls had chopped straw bedding, which was added once weekly, and stalls were cleaned and groomed 2×/d.
The use of cows and all experimental procedures complied with the guidelines of the Canadian Council on Animal Care (2009) and were approved by the University of Guelph Animal Care Committee (Animal Use Protocol #4109).

Study Design and Dietary Treatments
Sample size and power analyses were used to calculate (as per Morris, 1999) the minimum number of replicates needed (n = 42) to detect a 7% level of observed mean difference between observation periods for the primary outcome variables: feeding time, feeding rate, DMI, and milk production.Estimates of variation (average CV = 16%) for these variables were based on previously reported values (Johnston and DeVries, 2018).
The study was implemented during the months of January to June 2020 as an uncontrolled longitudinal design with 2 data collection periods.At 8 DIM (±2 d), cows were assigned an individual automated feed bin (Insentec, B. V., Marknesse, the Netherlands) in one pen and were trained to eat from their assigned bin by being guided to their bin with feed on the day that bins were assigned.Cows were then observed during the 30 min after feed delivery over the next 3 d to discourage them from eating out of other cows' bins.Cows that were observed attempting to eat out of other cows' bins were redirected to their own bin.At 36 ± 6 DIM (dependent on space in pens), cows were moved to 1 of 3 other pens where they were assigned to share 2 adjacent feed bins with 2 other cows (i.e., a stocking rate of 2:3), and trained in the same way to access the assigned bins.Each cow was observed from 15 to 28 DIM (low feeding competition; P1) and then from 63 to 76 DIM (higher feeding competition; P2).At all points of the study, the pens had a dynamic composition and also contained cows who were not enrolled in the study; those additional cows were trained to access feed bins Schwanke et al.: COW PERSONALITY TRAITS AND FEEDING BEHAVIOR using the same methods as the study cows.The experimental cows shared feed bins with these cows during P2.The P1 pen was maintained at a constant stocking density of 15 cows and the P2 pens were maintained at a constant stocking density of 24 cows per pen, but were not balanced for parity.All cows experienced the treatments in the same order (i.e., low feeding competition in P1 from 15 to 28 DIM followed by higher feeding competition in P2 from 63 to 76 DIM).

TMR Feeding Procedure
The TMR (Table 1) was formulated to meet the nutritional requirements for a 680 kg cow with an expected milk yield of 36 kg/d with 4.1% milk fat and 3.1% true protein (Cattle Professional, Agricultural Modeling & Training Systems, Groton, NY).The TMR was prepared in a mixer wagon (Jaylor Model 5572, Jaylor Fabricating, Orton, Canada) twice daily and transferred into a feed cart (Super Data Ranger; American Calan, Northwood, NH) before being fed into the automated feed bins between 1000 and 1100 h and between 1500 and 1600 h.The TMR was available ad libitum and refusals of 5% to 10% were targeted relative to the amount of as-fed feed offered.Actual refusal rate during the study was 9.7 ± 8.3% (mean ± SD).Feed within the bins was not remixed during the day.At 0930 h, before each daily AM feeding, the amounts of refusals were recorded and removed from the bins.The quantity of TMR offered in each bin (or pair of bins in P2) was adjusted daily based on the average intake over the previous 3 d.

Feeding Behavior and Rumination
The automated feed bins recorded the weight of TMR in the bin at the start and end of each bin visit (as validated by Chapinal et al., 2007), such that the number of feed bin visits, feeding time, eating rate, DMI, and meal parameters for each cow could be calculated.The DMI of each bin visit was calculated by multiplying the as-fed intake by the weekly average DM% of the diet, as determined by DM analysis of fresh TMR feed samples.Feeding rate at each bin visit was calculated as DMI (kg DM) divided by the time spent feeding (min).These data were then summarized to determine total daily DMI (kg/d), total daily feeding time (min/d), and average feeding rate (kg/min).The percentage of feed bin visits in which cows attempted to eat from bins they were not assigned to (i.e., stealing attempts) was computed.
Meal criteria (the minimum time interval between feed bin visits to be considered a new meal) were determined for each cow in each treatment period, using the MIX 3.1.3software package (MacDonald and Green, 1988) to fit normal distributions to the frequency of log 10 transformed time intervals between bin visits.These criteria were used to combine individual bin visits into meals for each cow, from which meal patterning was then analyzed (as described by DeVries et al., 2003).If the time interval between 2 bin visits exceeded the determined meal criteria, this indicated a different meal.The number of different meals was classified as meal frequency (meals/d).Meal duration (min/meal) was calculated as the time from the start of the first feeding bout until the end of the last feeding bout, at which time the meal criterion was exceeded.Meal size (kg/d) was calculated as DMI divided by the meal frequency.An electronic monitoring system (HR-TAG-LD, SCR Engineers Ltd., Netanya, Israel; validated by Schirmann et al., 2009) was used to continuously monitor rumination activity.A nylon collar equipped with a data logger was fitted to each cow before the start of P1 and removed at the end of P2.Data were downloaded from the system once weekly during the study and were used to determine total daily time spent ruminating (min/d) for each cow.

Milk Production and Components
During each treatment period, milking production data were automatically recorded by the milking parlor software (DeLaval, Tumba, Sweden).Milk samples from each cow were collected once weekly over 2 consecutive days at every milking during each treatment period.These samples were sent to a DHI testing laboratory (CanWest DHI, Guelph, Canada) for analysis of fat and true protein, using a Fourier Transform Infrared full spectrum analyzer (Milkoscan FT+ and Milkoscan 6000;Foss,Hillerød,Denmark).Fat (%) and true protein (%) values per cow on each sampling day were obtained by calculating the average across milkings, weighted by the milk yield of the sampled milkings.FCM was calculated as: 0.4 × milk yield (kg/d) + 15.0 × fat yield (kg/d) (NRC, 2001) and ECM was calculated as: (0.327 × kg of milk) + (12.95 × kg of fat) + (7.2 × kg of protein) (Tyrrell and Reid, 1965).

Feed Sampling and Analysis
Throughout the duration of the study, samples of the TMR were collected at the time of feed delivery 3x/wk.Samples of each of the TMR ingredient components were taken every 2 wk for DM and nutrient analysis.After collection, feed samples were frozen at −20°C until further analysis, at which point samples were thawed in a refrigerator for at least 24 h before processing.Feed samples were dried at 55°C in an oven for a minimum of 48 h to determine DM and chemical composition of the TMR (Table 1).After being dried, feed samples were ground through a 1-mm sieve (Model 4 Wiley Laboratory Mill, Thomas Scientific, Swedesboro, NJ).Ground samples TMR were pooled by week and ground component samples were pooled by month.Pooled feed samples were shipped to A&L Canada Laboratories Inc. (London, ON, Canada) for analysis of DM (60°C; AOAC International, 2000, method 934.01),CP (combustion;AOAC International, 2000, method 990.03;Leco FP-628 Nitrogen Analyzer, Leco, St. Joseph, MI), OM (550°C; AOAC International, 2000, method 942.05;Blue M Electric programmable asher, Blue M, Watertown, WI), ADF (AOAC International, 2000, method 973.18;Ankom 200 Fiber Analyzer, Ankom Technology, Macedon, NY), NDF with amylase and sodium sulfite (AOAC International, 2000, method 2002.04;Ankom 200, Ankom Technology), and starch with heat stable amylase and amyloglucosidase (AOAC International, 2000, method 996.11, K-TSTA Total Starch Assay Procedure, Megazyme, Wicklow, Ireland).

Assessing Cow Personality
The personality of each cow was assessed once at 21 ± 3 DIM (mean ± SD; P1) and once at 70 ± 3 DIM (P2) using a combined arena test following methodology of Schwanke et al. (2022).Up to 3 cows were assessed on each day and each cow was assessed using the same 3-step procedure consisting of: (1) a novel environment test; (2) a novel object test; and (3) a novel human test.These tests were chosen to demonstrate consistency across contexts, representing common sources of novelty that dairy cows may encounter and are in line with those tests used in previous literature (see Van Reenen et al., 2004;Lauber et al., 2006;Neave et al., 2018a).The assessment was conducted in a 6 × 11 m observational arena that was unfamiliar to the cows.The arena contained one water trough and was located such that the cow being assessed had limited visual contact with conspecifics.The floor consisted of rubber mats with the center of the arena marked with chalk, divided into quadrants, and concentric 2 × 2 m and 4 × 4 m squares drawn onto the floor at the start of each test day to determine when the test cow was within 1 or 2 m of the novel object or novel human (Figure 1).At the start of each assessment, a familiar researcher calmly moved the cow from the home pen to the arena by walking behind the cow.The test started when the gate was closed behind the cow, after she had willingly entered the arena.In the first part of the assessment (novel environment test), the cow spent 10 min alone in the arena.The same familiar researcher then entered the arena and placed either a pink yoga ball (used during the P1 test; diameter = 0.65 m) or an inflatable brown pool ring (used during the P2 test; diameter = 0.60 m) in the center of the arena and exited the arena.After 10 min, the familiar researcher entered the arena and removed the object, after which an unfamiliar male human wearing unfamiliar bright blue coveralls with high visibility yellow striping entered the arena and stood at the center of the arena without making eye contact for 10 min.The height of the unfamiliar human varied, but was representative of the heights and builds of the facility staff.After a visual signal that the test was concluded, the novel human exited the arena and the familiar researcher led the cow back to the home pen.The unfamiliar human was different in P1 and P2, but the coveralls were the same.The novel human was instructed to face the secondary camera (Figure 1) regardless of the location of the cow or what she was doing, but they were told they could step away at any point from the cow or exit the arena if they were uncomfortable with the cow's behavior or felt threatened.Researchers observed the entirety of the novel person test through a window in the nearby room, such that they could signal the novel person to exit the arena if necessary.Behaviors exhibited by the cows toward the novel person consisted of sniffing, licking, and head rubbing.No cows exhibited aggressive behavior such as kicking or charging, and none of the novel people felt the need to exit the arena before the conclusion of the test.All assessments were conducted between 1130 and 1400 h and movement of farm staff and other researchers in the vicinity of the arena was restricted during personality assessment to minimize external stimuli not part of the combined arena test.Each 30min personality assessment session was video recorded (using 2 video cameras placed such that all parts of the arena would be visible on at least one recording; GoPro HD Hero, GoPro Inc., San Mateo, CA, and Handycam Camcorder, Sony Corporation, Tokyo, Japan).The arena was scraped clean of manure between each cow and washed with a high-pressure water hose at the end of each assessment day.Any cow suspected or confirmed in heat was not eligible for assessment on that day, but was assessed 3 d later.
Video recordings of the personality assessments were later used to determine frequencies, durations, and latencies of behaviors displayed by the cows during each stage of the combined arena test (novel environment, novel object, and novel human; using the ethogram described in Table 2).Video analysis was performed by a single researcher, after establishing inter-observer reliability of κ > 0.71 for each test between 2 observers, and intra-observer reliability of κ > 0.88.Training for behavior scoring was provided through a detailed explanation of the ethogram in Table 2.In any cases where one camera did not record (through either malfunction or human error) and the cow was subsequently not visible on any video recording during the assessment, the time the cow was not visible was excluded from analyses.All time duration variables were converted to a percentage of test time, calculated as time spent performing the behavior, divided by total test time.If applicable, the time the cow was not visible during the assessment was subtracted from the total assessment time before calculation.

Statistical Analyses
All statistical analyses were conducted using SAS 9.4 software (SAS Institute Inc., 2013).All values reported are least squares means.Significance was considered at P ≤ 0.05 and tendencies at 0.05 < P ≤ 0.10.Before analyses, all data were screened for normality using the UNIVARIATE procedure of SAS.The assumptions of normality were met for all the outcome variables of DMI, feeding behavior, rumination, milk production, and milk components.Data were summarized by period for each cow.Due to equipment failures, rumination data were only available for 35 cows in P1 and 34 cows in P2.Due to pandemic-related shutdowns, milk component data could only be analyzed for 38 cows in P1 and 31 cows in P2.Due to missing video footage of portions of 2 cows' P2 personality assessments, not all behaviors could be recorded, and these cows had to be excluded from the PCA, and subsequent analyses of P2 data.
The FACTOR procedure of SAS was used to perform principal components analysis (PCA) with varimax rotation to identify personality traits by condensing correlated behavioral measures from the combined arena tests.Criteria for PCA analysis of animal behavior data followed Budaev (2010).For behaviors analyzed from the P1 personality assessments, the variables of % time spent exploring, % time spent attempting to escape, % time attracted to the home pen, % time spent in contact and % time spent within 1 m of the novel object or human were log 10 transformed, and the number of vocalizations and latency to contact the novel human were square root transformed to meet assumptions of normality.From the P2 personality assessments, the % time spent within 1 m of the novel object or human were log 10 transformed.Due to few behavioral variables being correlated across tests (and thus not able to be averaged), as well as to achieve sampling adequacy (in which <0.50 is considered inappropriate to perform PCA), behaviors from the novel environment, novel object, and novel human tests during the P1 personality assessment were analyzed in separate PCAs (novel environment test: 6 variables retained, sampling adequacy = 0.62; novel object test: 9 variables retained, sampling adequacy = 0.57; novel human test: 10 variables retained, sampling adequacy = 0.68).Behaviors from the P2 personality assessment (except vigilance) were able to be averaged across tests and analyzed in one PCA (10 variables retained; sampling adequacy = 0.73).After examination of scree plots, factors with eigenvalues greater than 1.0 were retained, and behavioral variables with low communality estimates (<0.30) were excluded.This resulted in 1 factor from the novel environment test PCA, and 3 factors each from the novel object and human test PCAs from the P1 personality assessment (51%, 75% and 75% cumulative variance, respectively), as well as 2 factors from the P2 personality assessment PCA (68% cumulative variance).High loadings of behaviors from the tests on each factor were considered equal to or greater than ± 0.63.Each factor was interpreted as a personality trait; for each factor, the score for each cow was extracted using the regression procedure, indicating where each cow lies along an axis from highly negative to highly positive for that factor.
To verify the stability of the PCA, cows were classified as low (lactation 1), medium (lactation 2 and 3), or high (lactation 4 or greater) parity and the residuals of each test behavior were obtained from a generalized linear model with parity class as the fixed effect.The PCA was then repeated using the residuals from each linear model as input variables.The results produced a similar factor loading pattern as the PCA on the original observations, verifying that the observed factors reported herein could be interpreted as putative personality traits.
First, we verified the consistency of personality traits between P1 and P2 by analyzing associations between corresponding factor scores (interpreted as personality traits) from each P1 and P2 personality assessment using the REG procedure of SAS.To address the first study objective to determine the effect of personality traits on feeding behavior and production in early lactation, associations of those outcomes with the cow factor scores of each trait identified from the P1 personality assessment (when feeding competition was low) were analyzed using the MIXED procedure.The model included the fixed effects of trait score and parity, and the random effect of cow.To address the second study objective to determine the effect of personality traits on changes to feeding behavior and production under greater competition, the difference (P2 − P1) in each outcome variable of DMI, feeding behavior, rumination, milk production, and milk components was computed for each cow; associations of these changes with the traits identified in the P2 personality assessment were analyzed using the same MIXED procedure as above.The P2 personality assessment was used for this analysis because P1 and P2 personality assessments were consistent (see Results), and P2 was the more concise analysis (only 2 factors, while P1 yielded 7 factors across 3 PCAs, see Results).

Principal Component Analyses
The loadings for each factor extracted from the P1 (low competition) and P2 (higher competition) PCAs are reported in Table 3.In the P1 novel environment PCA, a single factor was retained that was interpreted as the "active-explorative" trait (explained 50.7% of the total variance; high positive loadings: locomotion, number of quadrant switches, exploring, and escape attempt; high negative loading: vigilance).
The P1 novel object and novel human PCAs each produced 3 factors that were interpreted as reflecting the same 3 personality traits.In P1 novel object PCA, factor 1 was interpreted as the "fearful" trait (explained 34.9% of the total variance; high positive loadings: vigilance and time spent more than 2 m away from the object; high negative loadings: time spent in contact and within 1 m of the object), factor 2 was interpreted as the active-explorative trait (explained 26.7% of the total variance; high positive loadings: locomotion, number of quadrant switches, and exploring), and factor 3 was interpreted as the "social" trait (directed toward conspecifics; explained 14.1% of the total variance; high positive loadings: attraction toward the home pen and number of vocalizations).Similarly, in the P1 novel human PCA, factor 1 was interpreted as the fearful trait (explained 35.6% of the variation; high positive loadings: vigilance, latency to contact, and time spent more than 2 m from the novel human; high negative loading: time spent in contact with the novel human), factor 2 was interpreted as the active-explorative trait (explained 24.2% of the total variance; high positive loadings for locomotion, number of quadrant switches, and exploring), and factor 3 was interpreted as the social trait (directed toward conspecifics; explained 14.9% of the total variance; high positive loadings: attraction toward the home pen and time spent behind the novel human).The time spent behind the novel human was chosen to represent avoidance behavior.
In the P2 PCA, 2 factors were retained that were similar to those interpreted in P1.Factor 1 was interpreted as the fearful trait (explained 39.1% of the total variance; high positive loadings: vigilance in the novel object and novel human tests, latency to contact, and time spent more than 2 m from the object or human; high negative loadings: time spent in contact and time spent within 1 m of the novel object or human), and factor 2 was interpreted as the active-explorative trait (explained 28.5% of the total variance, high positive loadings: locomotion, number of quadrant switches, and exploring, high negative loadings: vigilance in the novel environment test).

Consistency of Personality Traits
The personality traits identified in P1 and P2 assessment periods were positively associated; thus, greater DIM and treatment (change in stocking density resulting in increased competition) did not affect the behavioral responses of cows in the combined arena test.The P2 the fearful trait scores were positively associated with the fearful trait scores identified in the P1 novel object PCA (R 2 = 0.16; P = 0.01; Figure 2A) and novel human PCA (R 2 = 0.68; P < 0.01; Figure 2B).The P2 active-explorative trait scores were positively associated with the active-explorative trait scores identified in the P1 novel environment PCA (R 2 = 0.37; P < 0.01; Figure 3A), and novel object PCA (R 2 = 0.17; P = 0.01; Figure 3B) and tended to be positively associated with the "active-explorative" trait scores identified in the P1 novel human PCA (R 2 = 0.08; P = 0.08; Figure 3C).High loadings (≥|0.63|) are shown in bold, indicating behavior variables that were highly correlated within each Factor. 3 Variable was log 10 transformed to meet assumption of normality. 4 Variable was square root transformed to meet assumption of normality.

5
Personality traits were interpreted based on behaviors exhibited during the combined arena test that loaded highly (either positively or negatively) onto each factor.

Association of Personality Traits with Feeding Behavior and Production Under Low Feeding Competition
Feeding behavior and production data are presented in electronic supplementary material, Supplemental Table S1 (https: / / doi .org/ 10 .5683/SP3/ VMAZID; Schwanke et al., 2023).Cows who were more activeexplorative in the P1 novel environment test had lesser FCM yield, ECM yield, and milk fat and protein yield, and tended to have lesser milk yield (Table 4).Parity was positively associated with feeding rate, meal size, milk yield, FCM yield, ECM yield, milk fat yield, and milk protein yield.
Cows who were more fearful during the P1 novel object test had greater rumination time (Table 5).No associations between the active-explorative and social trait scores from the P1 novel object test, or the fearful trait from the P1 novel human test, and the behavior and production outcomes were detected (Supplemental Tables S2, S3, and S4; https: / / doi .org/ 10 .5683/SP3/ VMAZID; Schwanke et al., 2023).
Cows who were more active-explorative (identified during the P1 novel human test) tended to have lesser milk fat content and feeding time (Table 6).No associations of this trait with DMI were detected, thus, feeding rate tended to be greater in these cows.Cows who were more social (identified during the P1 novel human test) spent more time ruminating and tended to have greater milk yield (Table 7).Across all traits identified in the P1 novel object and novel human tests, parity was generally positively associated with DMI (tendency for factor 3 of the novel object test), feeding rate (tendency for factor 3 of the novel object test), meal size, milk yield, FCM yield, ECM yield, milk fat content, milk fat yield, and milk protein yield, as well as generally (factors 1 and 2 only of the novel object and novel human tests) tended to be negatively associated with meal frequency (Tables 5, 6, and 7, Supplemental Tables S2, S3, and S4).

Association of Personality Traits with Changes in Feeding Behavior and Production Under Higher Feeding Competition
Cows who were more fearful during the P2 personality assessment increased their feed bin visits in P2 after feeding competition was increased, and tended to be more consistent in their eating rate across the 2 observation periods, compared with cows who were less fearful who increased their eating rate in P2 (Table 8).Cows who were more active-explorative during the P2 personality assessment tended to have more consistent milk yield and greater FCM yield across the 2 observation periods, whereas less active-explorative cows had lesser milk yield and FCM yield in P2 compared with P1 (Table 9).Parity was negatively associated with FCM yield, ECM yield, and milk fat yield in P2 (Table 8).

DISCUSSION
Personality traits in animals are often displayed more prominently when they are faced with a challenging situation (Finkemeier et al., 2018), such as competition in their feeding environment (Boumans et al., 2018).However, few researchers have documented how such challenges affect dairy cow feeding behavior and production, and how cows of different personalities react to and cope with such challenges.This study aimed to address that knowledge gap by investigating the effect of personality traits on feeding behavior and how those behaviors change when feed competition increases; we also aimed to document the consistency of personality traits over this challenge.
The consistency of the fearful and active-explorative traits observed in this study under the lower and higher feeding competition indicates that the challenge of increased feed competition with advanced DIM may not be sufficient to initiate a significant change in personality.Even though, by definition, personality traits are fairly consistent over time (Finkemeier et al., 2018), they may be plastic and researchers have previously demonstrated that major physiological and management changes, such as those occurring around puberty, induce significant shifts in personality traits (Neave et al., 2020).
There is a possible role of sex or physical size of the novel (unfamiliar) human in terms of the cows' responses to that person; for example, cows have been  Total number of feed bin visits over 14-d observation period.demonstrated to have a greater approach distance to both familiar and unfamiliar male humans compared with familiar and unfamiliar female humans (Arave et al., 1992).Thus, to minimize variation, only male unfamiliar humans were used in the current study.
It should be noted, though, that repeating the study using a female unfamiliar human may yield different results.Given that the male humans used in this study had heights and builds representative of the facility staff, it can be concluded that this study truly assessed the "novelty" of the unfamiliar human, rather than the cows' response to their physical attributes or sex.In early lactation with low feeding competition, cows who were more active-explorative in the novel environment test had lower milk and component production compared with cows that were less active-explorative.In a previous study, we also observed that more active cows (measured using the same novel environment and object tests) had lower milk yield (Schwanke et al., 2022).Interestingly, there was no association of the active-explorative trait with DMI, suggesting that other mechanisms were responsible for the difference in milk production in these cows.If these cows were more active in their home pen, a greater proportion of nutrients may have gone to maintenance rather than milk production, but this relationship needs further investigation.Cows who were more active-explorative in the novel human test under low competition also tended to have lesser feeding time, possibly due to more time in their day that was spent walking around the pen.An alternative explanation for the lesser milk production and feeding time is that cows who were more activeexplorative during the novel environment test in P1 may have had more issues coping with the transition period compared with cows who scored lower on this trait, as such issues could lead to suboptimal metabolic and physiological functioning (Proudfoot, 2023).
Except for an association with greater rumination time, cows that were fearful (in either the novel object or novel human tests) showed limited associations with feeding behavior or milk production measures when under low competition conditions.This finding is in line with previous work that did not detect any associations between cows that were more fearful and their milk yield (Hedlund and Løvlie, 2015;Marçal-Pedroza et al., 2020).In addition, Neave et al. (2022)   Total number of feed bin visits over 14-d observation period.detect an association between fearful cows and grazing time.It should be noted that the above studies as well as the present study were conducted in groups of cows milked on herringbone and rotary parlors.In contrast, in our previous study (Schwanke et al., 2022) which assessed personality using the same tests in a group of cows milked using an automated milking system, we demonstrated that more fearful cows were less likely to consume high allocations of concentrate in the milking unit, and had greater daily variation in that concentrate intake (Schwanke et al., 2022).Thus, it may be that the effects of certain personality traits such as fearfulness on behavior and production are more apparent in certain milking or management systems.
Cows who were more social toward conspecifics (during the novel human test) when assessed in early lactation under low feeding competition spent more time ruminating and tended to have greater milk yield, compared with less social cows.As there was no association of the social trait with DMI, it would suggest that other mechanisms were responsible for the difference in rumination and milk yield in these cows.Other researchers have reported energy-corrected milk yield to be lower in cows that vocalized more and spent more time facing the herd during a social isolation test (similar to 'social' cows in our novel object and human tests; Hedlund and Løvlie, 2015), while in another study it was reported that milk ejection was improved in heifers that vocalized more during social isolation (interpreted as greater sociability; Van Reenen et al., 2013).The present study lends support to this pattern of behavioral responses to social conditions (e.g., greater vocalizations and attraction to home pen in response to isolation) having associations with productivity; an association which has also been previously documented in pigs (Kavlak et al., 2021).The underlying mechanism of why cows that are more social may differ in milk yield and components remains to be understood, but we suggest this may relate to differences in the way cows interact with their peers in their feeding environment.As no differences between more and less social cows were detected during the higher competition period, the differences in rumination time and milk production observed in early lactation may also be due to how well these cows coped with the transition period.
When competition for feed increased at greater DIM, the more active-explorative cows were able to maintain their milk yield compared with P1, compared with less active-explorative cows whose milk production decreased in P2.We speculate that if the more active-explorative cows were frequently walking around the pen, they would be more likely to encounter an unoccupied feed bin and were thus able to maintain their DMI to support their milk production.This would suggest that these cows were better able to cope with greater competition for feed.This is in line with previous work showing that more active and explorative animals tend to perform better under variable or unpredictable conditions (Carere et al., 2010;Coppens et al., 2010;Koolhaas et al., 2010).The fearful cows in our study responded to the change in feeding competition by more frequently visiting both their assigned and unassigned feed bins.This strategy apparently enabled these cows to maintain their DMI.In contrast, the less fearful cows tended to increase their feeding rate in P2 under greater feeding competition, whereas the more fearful cows had a consistent eating rate across P1 and P2.Cows are known to increase their feeding rate with greater competition at the feed bunk (Proudfoot et al., 2009), so this result suggests fearful cows may have avoided the feeding area when other cows were eating, allowing them to eat at a slower rate during less crowded times.This type of behavioral pattern has been observed in beef heifers responding to competitive feedlot conditions (Zobel et al., 2011).Together our results demonstrate that cows of different personalities adopt different feeding behavior strategies to adapt to a change in their environment.Cows could therefore benefit from tailored management strategies that support more successful adaptation to changes in their feeding environment, such as providing dedicated access to feed areas, opportunities to feed at less crowded times, or more targeted feed supplementation (as possible in automated milking systems).The associations of parity with greater milk and component yields that were observed are to be expected based on previous established knowledge of lactation trends (Rogers and Stewart, 1982) and changes in eating behavior with age (Beauchemin et al., 2002;Azizi et al., 2009;DeVries et al., 2011).
To our knowledge, this work is the first to describe how individual dairy cow personality traits may be associated with changes in feeding behavior in response to greater feeding competition.The increase in competi- tion for feed was confounded with the increase in DIM.Thus, there is the possibility that cows were altering their feeding behavior in response to other factors, such as altered nutrient demands with greater milk production (DeVries et al., 2003;Johnston and DeVries, 2018) and changes in rumen fill capacity (Allen et al., 2009).However, our study has demonstrated that cows differ in how they respond to that challenge, based on their personality.We believe that complementing the personality assessments conducted herein with measures of social dominance index and home pen behaviors may further characterize the personality of the cow to reveal other associations with feeding behavior and production, as well as lead to the development of more automated measures of personality that would be efficient to conduct and use on commercial dairy operations.Due to the paucity of research in this area, we strongly recommend that future research investigates relationships between personality traits, social interactions, exploratory behaviors within the home pen, and the plasticity of feeding behavior to better understand how dairy cows cope with challenges in their feeding environment or changes in physiological demands, and to develop strategies to support individual animals who struggle to cope with these challenges.

CONCLUSIONS
The cow scores for the personality traits of fearful and active-explorative that were identified in P2 under higher feeding competition were consistent with the scores for the corresponding traits identified in P1 under lesser feeding competition, indicating that personality was not affected by a change in environment and advancing DIM.Under low competition conditions, more active-explorative cows had lesser milk and component yields, as well as lesser feeding time, while more social cows had greater milk and component yields.When competition increased (in combination with greater DIM), cows of different personalities adopted different feeding strategies.Active-explorative cows may have naturally encountered unoccupied feed bins more often, allowing them to maintain their DMI and milk yield.In contrast, fearful cows increased their feed bin 1 P2 = period 2; 63-76 DIM in which cows were assigned 2 adjacent feed bins per 3 cows (high feeding competition).
3 P1 = period 1; 15-28 DIM in which cows were assigned 1 feed bin per cow (low feeding competition). 4+ indicates positive association; -indicates negative association; no symbol indicates no detected statistical association. 5 Data were collected over 14 d for 42 cows. 6Total number of feed bin visits over 14 d observation period. 7Data were collected over 14 d for 35 cows. 8Data were collected over 8 milkings for 31 cows.
visits, whereas less fearful cows increased their eating rate.This study builds on a growing body of research exploring the influence of dairy cow personality traits on behavior, and demonstrates that cows of different personalities use different strategies to adapt to challenges in their feeding environment. 1P2 = period 2; 63-76 DIM in which cows were assigned 2 adjacent feed bins per 3 cows (high feeding competition).
3 P1; 15 to 28 DIM in which cows were assigned 1 feed bin per cow (low feeding competition). 4+ indicates positive association; -indicates negative association; no symbol indicates no detected statistical association. 5 Data were collected over 14 d for 42 cows. 6Total number of feed bin visits over 14 d observation period. 7Data were collected over 14 d for 35 cows. 8Data were collected over 8 milkings for 31 cows.
Figure1.Schematic of the arena used for personality assessment, with arena center, quadrants, 1-m and 2-m distances from the center of the arena marked with chalk (dashed lines).The cow entered the arena from the long side of quadrant 1, which was the quadrant closest to (although not in view of) the home pen.The novel human faced the secondary camera for the duration of the novel human test.
Schwanke et al.: COW PERSONALITY TRAITS AND FEEDING BEHAVIOR

Figure 2 .Figure 3 .
Figure 2. Associations of P2 (period 2; 63-76 DIM in which cows were assigned 2 feed bins per 3 cows) principal components analysis (PCA) fearful trait scores with (A) P1 (period 1; 15-28 DIM in which cows were assigned 1 feed bin per cow; n = 40) novel object PCA fearful trait scores, and (B) P1 novel human PCA fearful trait scores.The dotted line in each panel indicates the line of best fit. 4

Table 1 .
Ingredient and chemical composition (mean ± SD) of the formulated TMR Schwanke et al.: COW PERSONALITY TRAITS AND FEEDING BEHAVIOR

Table 2 .
Schwanke et al.:COW PERSONALITY TRAITS AND FEEDING BEHAVIOR Definition of behavioral measures and events recorded during the combined arena test for assessment of personality traits of 42 cows 1 Non-normally distributed, log 10 or square root transformed data used in later analysis.

Table 3 .
Coefficients (loadings)of the eigenvectors for the factors extracted from each principal components analysis of behaviors recorded when cows (P1: n = 42; P2: n = 40) were subjected to a combined arena test1,2 1Combined arena test consisted of successive exposure to a novel environment (10 min), novel object (10 min), and novel human (10 min). 2

Table 4 .
Schwanke et al.:COW PERSONALITY TRAITS AND FEEDING BEHAVIOR Relationships between P1 1 novel environment test principal component analysis factor 1 scores and parity with feeding behavior, rumination, and milk production 2+ indicates positive association; -indicates negative association; no symbol indicates no detected statistical association.3 Data were collected over 14 d for 42 cows.

Table 5 .
did notSchwanke et al.:COW PERSONALITY TRAITS AND FEEDING BEHAVIOR Relationships between P1 1 novel object test principal component analysis (PCA) factor 1 scores and parity with feeding behavior, rumination, and milk production 2 3Data were collected over 14 d for 42 cows.

Table 6 .
Schwanke et al.:COW PERSONALITY TRAITS AND FEEDING BEHAVIOR Relationships between P1 1 novel human test principal component analysis factor 2 scores and parity with feeding behavior, rumination, and milk production

Table 7 .
Schwanke et al.:COW PERSONALITY TRAITS AND FEEDING BEHAVIOR Relationships between P1 1 novel human test principal component analysis (PCA) factor 3 scores and parity with feeding behavior, rumination, and milk production 2

Table 8 .
Schwanke et al.:COW PERSONALITY TRAITS AND FEEDING BEHAVIOR Relationships between P2 1 principal component analysis (PCA) factor 1 scores and the change 2 in feeding behavior, rumination, and milk production between P1 3 and P2

Table 9 .
Relationships between P2 1 principal component analysis (PCA) factor 2 scores and the change 2 in feeding behavior, rumination, and milk production between P1 3 and P2