Feeding behavior parameters and temporal patterns in mid-lactation Holstein cows across a range of residual feed intake values

Residual feed intake (RFI) is a measurement of the difference between actual and predicted feed intake when adjusted for energy sinks; more efficient cows eat less than predicted (low RFI) and inefficient cows eat more than predicted (high RFI). Data evaluating the relationship between RFI and feeding behaviors (FB) are limited in dairy cattle; therefore, the objective of this study was to determine daily and temporal FB in mid-lactation Holstein cows across a range of RFI values. Mid-lactation Holstein cows (n = 592 multiparous; 304 primiparous) were enrolled in 17 cohorts at 97 ± 26 d in milk (± standard deviation), and all cows within a cohort were fed a common diet using automated feeding bins. Cow RFI was calculated as the difference between predicted and observed dry matter intake (DMI) after accounting for parity, days in milk, milk energy, metabolic body weight and change, and experiment. The associations between RFI and FB at the level of meals and daily totals were evaluated us-ing mixed models with the fixed effect of RFI and the random effects of cow and cohort. Daily temporal FB analyses were conducted using 2-h blocks and analyzed using mixed models with the fixed effects of RFI, time, RFI × time, and cohort, and the random effect of cow (cohort). There was a positive linear association be-tween RFI and DMI in multiparous cows and a positive quadratic relationship in primiparous cows, where the rate of increase in DMI was less at higher RFI. Eating rate, DMI per meal, and size of the largest daily meal were positively associated with RFI. Daily temporal analysis of FB revealed an interaction between RFI and time for eating rate in multiparous and primiparous cows. The eating rate increased with greater RFI at 11 of 12 time points throughout the day, and eating rate differed across RFI between multiple time points. There tended to be an interaction between RFI and time for eating time and bin visits in multiparous cows but not primiparous cows. Overall, there was a time effect for all FB variables, where DMI, eating time and rate, and bin visits were greatest after the initial daily feeding at 1200 h, increased slightly after each milking, and reached a nadir at 0600 h (6 h before feeding). Considering the relationship between RFI and eating rate, additional efforts to determine cost-effective methods of quantifying eating rate in group-housed dairy cows is warranted. Further investigation is also warranted to determine if management strategies to alter FB, especially eating rate, can be effective in increasing feed efficiency in lactating dairy cattle.


INTRODUCTION
In lactating dairy cows, residual feed intake (RFI) is a measurement of the difference between actual and predicted feed intake when adjusted for BW, BW change, and milk energy output (VandeHaar et al., 2016).For 2 animals with the same predicted energy needs, cows with low RFI consume less feed compared with those with high RFI and are thus more efficient.Given that feed is the primary expense on dairy farms, increasing feed efficiency of cows by reducing feed consumed per equivalent unit of output can enhance dairy farm profitability.
Determining phenotypic RFI currently requires individual feeding systems where total daily DMI can be quantified; however, the cost and labor-intensive nature of individual feeding systems makes on-farm, cow-level DMI quantification prohibitive for most US dairy cows.As such, there is a need to identify metrics related to RFI that can be used to estimate a cow's RFI status.For example, individual cow feeding behaviors could provide a potential marker of RFI, which could theoretically be determined through wearable sensors and smart technologies cheaper than with individual feeding bins.Evaluation of the relationship between RFI and feeding behaviors (FB) in lactating cows is limited, but eating time, meal size, and eating rate differ between high-and low-RFI cows (Connor et al., 2013;Xi et al., 2016).It is unknown whether these FB are expressed differently throughout the day between efficient and inefficient lactating dairy cows, but data in beef calves, dairy heifers, and pigs suggest interactions of FB and time are possible (Golden et al., 2008;Young et al., 2011;Green et al., 2013).
Evaluation of whether the relationship between FB and RFI changes throughout the day may present an opportunity to identify potential management strategies to enhance feed efficiency.In beef cattle, standing time, standing duration, and feeder visit frequency and duration are significant predictors of RFI (Haskell et al., 2019), and these types of behaviors can be influenced by management factors such as stocking density, feed barrier design, and consistency of feed delivery to name a few (Huzzey et al., 2006;Proudfoot et al., 2009).Feeding rate increases with greater feed competition in lactating cows (Proudfoot et al., 2009), and feeding rate has an inverse relationship with feed conversion ratio (Krpálková et al., 2021).Additionally, interactions exist between nutrient intake and FB, which could be further manipulated.For example, researchers have attempted to increase feed efficiency through feed restriction, which reduced eating time, meal frequency, and bin visit frequency (Ben Meir et al., 2019).If a temporal relationship exists between FB and RFI, then perhaps inefficient cows can be managed in a manner to alter daily and temporal FB patterns and increase efficiency.Therefore, the objective of this study was to evaluate the overall and daily temporal relationship between FB and RFI in a large data set of mid-lactation Holstein cows using automated feed bins.

MATERIALS AND METHODS
Animal handling and sampling procedures were approved by the University of Wisconsin-Madison College of Agricultural and Life Sciences Animal Care and Use Committee.Data were collected from mid-lactation Holstein cows (n = 896) in 17 cohorts between 2012 and 2019 at the University of Wisconsin-Madison as part of a national feed efficiency project (Tempelman et al., 2015;Li et al., 2019).Cows were housed in a sand-bedded freestall facility at the Emmons Blaine Dairy Cattle Research Center (Arlington, WI), and cow descriptive statistics by parity group are outlined in Table 1.Data observation periods were 37 to 64 d in length.Cows were provided ab libitum access to feed and water, with diets formulated to meet or exceed nutrient requirements.All cows within a cohort were fed a common diet, and no dietary treatment was implemented.Composition of the diets and parity profile are outlined in Supplemental Table S1 (https: / / data .mendeley.com/datasets/ bn4mtg5hxk/ 1; Brown, 2022).
Diets were fed using a roughage intake control system (RIC; Hokofarm Group) and access to the RIC feeders was by electronic radiofrequency ID (RFID) tag.Cows had access to all feeders (n = 32) for cohorts included in this analysis, with 2:1 ratio of cows for every bin.All cows within and across cohorts were housed in the same pen.The RIC system allowed cow access to each feeder using an RFID tag and recorded the start and end time of each visit to the bin and the total weight of feed disappearance during each bin visit to the nearest 0.1 kg as-fed.Cows were locked away from the bins for approximately 1 h each day (1100 h) to discard orts and distribute fresh feed and for approximately 15 min at the end of each calendar day.Cows were fed at 1200 ± 1 h and 1700 ± 1 h, and milked 2×/d at 0400 and 1500 with milk weights recorded electronically.Milk samples were obtained weekly at 4 consecutive milkings for determination of milk composition (AgSource).Body weights were obtained on 3 consecutive days at the beginning, middle, and end of the cohort, and BCS was determined by 3 trained investigators thrice at the same intervals as BW collection.The BCS was determined on a 5-point scale where 1 = thin and 5 = obese, with increments of 0.25 (Wildman et al., 1982).The RFI phenotype for each cow was calculated by the Council of Dairy Cattle Breeding according to the equation of Gaddis et al. (2021) as follows: where parity × Σ(DIM) is the fixed effect of parity class (primiparous vs. multiparous) and the fifth order polynomial regression of DMI on DIM; b 1 MilkE is the partial regression of DMI on milk energy output (Mcal/d); b 2 MBW is the partial regression of DMI on metabolic BW; b 3 ΔBW is the partial regression of DMI on the change in BW; cohort is the random effect of experiment; test week is the random effect of mid-point test week within cohort, and ε is the residual, which is considered as the RFI phenotype.

Data Cleaning and Behavioral Variable Calculation
Bin visits that resulted in 0 kg as-fed feed disappearance were removed, which represented approximately 7.75% of all observations in the data set.Total DMI for a given visit was calculated by multiplying the quantity of feed disappearance by the calculated diet DM.
The FB variables were determined on an average daily, average meal, and temporal basis, and were calculated by determination of a meal criterion using the log 10 frequency distribution of the interval length Brown et al.: FEED EFFICIENCY AND TEMPORAL FEEDING BEHAVIORS between bins visits, as described previously (DeVries et al., 2003;Horvath and Miller-Cushon, 2019).Briefly, the meal criterion was determined by fitting the mixed probability density function of the log 10 -transformed intervals between bin visits using the maximum likelihood method (mixdist procedure; Macdonald and Du, 2018; R version 4.0.3,https: / / www .r-project .org/).Visual inspection showed 2 distinct distributions corresponding to the short and long intervals between bin visits, respectively (Supplemental Figure S1, https: / / data .mendeley.com/datasets/ bn4mtg5hxk/ 1; Brown, 2022).The intersection of the distribution curves between the transformed short and long interval visits was deemed the meal criterion.Any intervals less than the defined meal criterion were defined as feeding activity within the same meal, and those that exceeded the meal criterion represented a new meal event.The meal criterion was calculated for the entire data set and by parity group (primiparous vs. multiparous); the criterion was determined to be 22.2 min for primiparous cows and 27.9 min for multiparous cows.A paired t-test of the of daily primiparous meal frequency using either the meal criterion for only primiparous cows or the entire data set (26.4 min) was significant (P ≤ 0.001); therefore, the FB were calculated separately for primiparous and multiparous cows using the corresponding meal criterion for each parity group.
The FB data were calculated on a daily, per meal, and temporal basis.A summary of calculated FB, their abbreviations, and definitions are outlined in Table 2.After calculation of the daily FB for each cow, the overall mean and SD for each cow within cohort was determined, and any daily value that exceeded 3 SD of the mean was eliminated from the data set to ensure biologically relevant values for healthy mid-lactation Holstein cows.The percentage of data removed was minimal and ranged from a 0.03% for bin visits per meal to 1.05% for largest daily meal.

Statistical Analysis
The association between nontemporal daily and meal FB and RFI by parity group (primiparous and multiparous) was analyzed in JMP (version 14.0, SAS Inc.) using a model accounting for the linear and quadratic fixed effects of RFI, and the random effects of cow and cohort.The quadratic effect of RFI was removed from the model when P ≥ 0.10.Model residuals were assessed for normality and reasonably met model assumptions in all scenarios.Daily and meal FB variables analyzed are outlined in Table 2.
The temporal FB analysis was conducted for each cow using 2-h time blocks throughout the day, such that there were 12 total time blocks for each cow.Variables included in temporal analysis are outlined in Table 2. Data were analyzed using repeated measures in SAS (version 9.4, SAS Inc.) with the fixed effects of linear and quadratic RFI (continuous variables), time, RFI × time, and cohort, and the random effect of cow (cohort).Observations recorded over consecutive 2-h blocks were considered as repeated measures, and an autoregressive covariance structure was used.For interactions between RFI and time, the slope of RFI at each time point was tested against the null hypothesis that it was equal to 0, and the slope for RFI between different time points was compared.The model was reduced when the quadratic effect of RFI was P ≥ 0.10.Model residuals were assessed for normality, and data were transformed to meet model expectations, when appropriate.

RESULTS AND DISCUSSION
The size of the data set used in the current study represented the largest known analysis of FB for lactating dairy cows.Dado and Allen (1994) determined the estimated cow numbers required to detect significant differences in FB for lactating cows for various study designs, with several hundred cows required per treatment (depending on study design) with 5 d of data collection.With 35 d or more of data collection for approximately 900 cows, this data set is uniquely positioned to accurately provide insight into feeding behaviors and the potential relationship with feed efficiency in mid-lactation dairy cows.
Primiparous and multiparous cows were not compared directly in this analysis because the calculation of RFI in dairy cows accounts for parity, and FB were determined for each parity group using a separate meal criterion; therefore, comparing parities directly would be confounded.There is no commonly accepted definition of a meal criterion in dairy cattle, and meal criterion have been arbitrarily established to range from 5 to 12 min (Bingham et al., 2009;Mullins et al., 2012;Connor et al., 2013).The use of the log 10 frequency of the interval between bin visits establishes parameters for bin visits that fall within or between a meal.This approach provides the most relevant meal criterion for the behavior expressed (Tolkamp and Kyriazakis, 1999).Log-transformation of the intervals between bin visits typically results in meal criterion of 20 to 35 min in lactating dairy cows (Tolkamp and Kyriazakis, 1999;DeVries et al., 2003;DeVries and Chevaux, 2014), consistent with the meal criteria for primiparous and multiparous cows in the present study.
In this study, we opted to model FB using RFI as the independent variable, as has been done previously (Bingham et al., 2009;Durunna et al., 2011;Connor et al., 2019).It is yet unclear if FB affects RFI, or vice versa.Our research objective was to identify markers that are easily measurable and can help identify efficient cows (low RFI) given that RFI is difficult to assess on dairy farms.By constructing our models with RFI as the independent variable, we could evaluate whether more efficient cows exhibited certain FB that could be more readily measured.Therefore, to maximize interpretability across studies previously published in the literature, and to identify markers that are indicative of efficient cows, we opted to model FB as a function of RFI.

Daily Association Between FB and RFI
Descriptive statistics of the daily and meal FB by parity group are in Table 3.As anticipated, less efficient cows consumed more feed as there was a positive linear association between daily DMI and RFI for multiparous cows (P < 0.001; Table 4) and a positive quadratic association for primiparous cows (P = 0.04; Table 4; Figure 1).There was no detected evidence of association between daily eating time and RFI (P > 0.21; Table 4).Less efficient cows ate at a faster rate as daily eating rate was positively associated with RFI (P < 0.001; Table 4) for both parity groups.Similar to our data, Connor et al. (2013) did not detect an effect of RFI on eating time in early-lactation cows using GrowSafe bins, but they observed an increase in feeding rate with inefficient cows.In another study, inefficient cows had longer eating time when quantified with visual observation in tie stalls (Xi et al., 2016).In Australian dairy heifers fed with electronic feeding stations (20 heifers per pen; 2 bins/pen), there was no detected evidence of difference in feeder occupation time between the most and least efficient heifers, but eating rate was greater for inefficient heifers (Williams et al., 2011).In general, data in beef cattle show that eating rate is greater for inefficient animals (Kelly et al., 2010;Durunna et al., 2011;Fitzsimons et al., 2014).However, eating time is greater for inefficient beef cattle in some studies (Nkrumah et al., 2006;Durunna et al., 2011;Fitzsimons et al., 2014) but not in others (Kelly et al., 2010;McGee et al., 2014).Because eating rate is Brown et al.: FEED EFFICIENCY AND TEMPORAL FEEDING BEHAVIORS a function of DMI and eating time, the eating rate in these studies may have been influenced by the definition of eating time.For example, in our study, daily eating time was calculated as the time with the head in the feed bin, whereas other researchers may use the total meal time as eating time.These disparities make interpretation of the literature difficult for eating time.
On a meal basis, there was no detected evidence of an association between daily meal frequency (M d ) and RFI for either parity group (P ≥ 0.14; Table 4).With M d being a function of meal length and intermeal interval, there similarly was no detected evidence of an association between intermeal interval and RFI (P > 0.43).Early-lactation dairy cows, beef steers, and beef heifers had a similar meal frequency across RFI categories (Bingham et al., 2009;Connor et al., 2013;McGee et al., 2014).Conversely, using electronic feeders with a high feed bin stocking density of 8 animals per bin, inef- ficient dairy heifers and growing Angus bulls had more daily meals than their efficient counterparts (Lancaster et al., 2009;Green et al., 2013), raising the question of whether social dominance has a role in meal disruption and feed efficiency when access to feed is restricted.Furthermore, in lactating cows, increasing the cow: bin ratio up to 3:1 increases feed bin displacements, increases the time spent standing idle at feeding (Huzzey et al., 2006), decreases the lying time, and increases the eating rate (Crossley et al., 2017).In our study, where primiparous and multiparous animals were mixed with an animal-to-bin ratio of 2:1 and cows had access to all bins, social dominance interactions could occur at peak feeding times as has been observed in ongoing research (Baier et al., 2021).Previous analysis of bin occupancy rate from 6 studies within the same housing system and study design (which included 4 studies from the current data set) indicated that bin occupancy rate >90% is limited to 33 min/d, and overall bin occupancy rate across the entire day was only 32% (Brown et al., 2022).Increasing the bin density to the degree the variable meal criterion between their treatment groups makes interpretation difficult.Overall, although social interactions are not a focus within this study, their exploration in controlled studies may provide insights into the relationship of certain FB and RFI.Within a meal, less efficient cows consumed feed at a faster rate, demonstrated by a positive association between meal eating rate (ER m ) and RFI.The ER m increased linearly with greater RFI for multiparous cows (P < 0.001; Table 4) and quadratically for primiparous cows (P = 0.03; Figure 2).The increase in meal ER m was likely explained by the positive association between meal DMI and RFI (P ≤ 0.001; Table 4) and no detected evidence of association between meal eating time and RFI (P ≥ 0.12; Table 4).These results mirrored what we observed on a daily basis for DMI, eating time, and eating rate, likely because of the lack of association between RFI and M d .Connor et al. (2013) also observed an increase in meal size for inefficient early-lactation Holstein cows, but they did not report the meal eating time or rate.Alternatively, there was no detected evidence of a change in meal size between RFI groups in dairy heifers, beef steers, or beef heifers (Green et al., 2013;McGee et al., 2014).In our study, even if the meal frequency changed between efficient and inefficient cows, it was unclear if there would be a concomitant change in eating rate.For example, despite a correlation between RFI and meal frequency in beef bulls, there was no correlation between RFI and eating rate (Lancaster et al., 2009).In general, few data are available in the literature evaluating within-meal FB parameters among divergent RFI categories, and thus it remains an area of opportunity for further exploration.
There was no detected evidence of association between the number of daily bin visits or meal bin visits with RFI for either parity group (P ≥ 0.26; Table 4).The use of bin visits as a parameter to evaluate feeding behavior is novel in dairy cattle but common in beef cattle (Bingham et al., 2009;Durunna et al., 2011;Fitzsimons et al., 2014).In our study, a bin visit was defined as any instance the cow entered the bin and consumed at least 0.1 kg of feed as-fed, and cows could have multiple bin visits within a meal if they temporarily left the bin or accessed a different bin within the meal criterion.Unrewarded bin visits (a visit that resulted in 0.0 kg as-fed feed disappearance and was removed from analysis), accounted for approximately 7.5% of all visits, similar to previous reports in beef studies (Kelly et al., 2010;Fitzsimons et al., 2014).With the unique nature of the RIC system used herein and its potential effects on feeding behavior compared with a continuous feed bin in a typical dairy farm, it is unclear if the visits to automated feed bins would translate similarly to cows accessing an open feed bin.Tolkamp et al. (2000) suggested that meal frequency, not bin visits, is the biologically relevant unit of feeding behavior; however, bin visits could serve a useful role in research farms for analyzing social interactions between animals (Huzzey et al., 2014).For example, in our data set, primiparous cows numerically had 10 more daily bin visits (41 vs. 31; Table 3), potentially indicating more frequent displacements by dominant multiparous cows.
It is unclear what other behaviors may be expressed within bin visits that could be associated with the increased DMI and eating rate observed for inefficient cows.The eating rate could potentially be influenced by feed sorting, pre-ingestion mastication time, and time spent actively ingesting feed.For example, there is great variation in the degree of sorting between cows on the same diet (Leonardi and Armentano, 2003), and sorting against longer particles decreases feeding rate and total DMI (Greter and DeVries, 2011).Whether cows that sort are likely to have high or low phenotypic RFI has not yet been evaluated.One potentially useful method to further interrogate these global DMI and eating rate differences is the evaluation of the time the animal spends with its head down in the bin.Less efficient beef steers spent more time with their heads down during bin visits (Durunna et al., 2011), but the opposite was true for growing Brangus heifers (Bingham et al., 2009).The use of different methodologies, diets, stage of production, ratios of animals to feeders, and behavior definitions to quantify FB data likely contribute to the contradictory results.Nonetheless, evaluation of head-down time in the bins in the current data set would have been useful in further explaining the behavior of these cows (i.e., feed sorting vs. mastication) and represents an area of opportunity for future studies.

Temporal Association Between FB and RFI
The current paper represents one of the first analyses of the association between daily temporal feeding behaviors and RFI in lactating dairy cows, and one of very few involving temporal analysis of RFI and FB in any livestock species (Golden et al., 2008;Young et al., 2011;Green et al., 2013).Although temporal patterns were not statistically analyzed, a previous lactating cow study demonstrated that there may be numerical differences between high-and low-efficient cows at the peak feeding times throughout the day (Ben Meir et al., 2018) and supports the need to further investigate.Evaluation of temporal FB is important because it may help clarify whether FB is being expressed differently throughout the day among animals that differ in RFI.Within the current study, there was no time and RFI interaction for temporal DMI (DMI t ) within either parity group (P ≥ 0.11; Table 5), which is similar to results in pigs (Young et al., 2011).Conversely, dairy heifers that were stocked with 8 animals per feeding station displayed marked differences in temporal feed intake (Green et al., 2013).The least efficient heifers primarily consumed feed during the afternoon hours directly after feed delivery, whereas the efficient animals consumed feed more evenly throughout the day.This pattern was also reflected in each group's respective meal frequency and duration (Green et al., 2013).It is not clear whether their bin stocking density (8:1) altered these temporal feeding behaviors, but perhaps similar patterns would have been observed in our study if cows were managed with greater bin-stocking density.At stocking densities of 3 or fewer cows per bin, it appears that lactating cows maintain similar feeding patterns throughout the day and are stimulated by feeding, feed pushup, and milking (Huzzey et al., 2006;Crossley et al., 2017).Therefore, in our study, we expect that there would have been a similar lack of association between RFI and DMI t if there was a 1:1 ratio of cows to bins.
There was an interaction between RFI and temporal eating rate (ER t ) in both multiparous and primiparous cows (P = 0.01; Table 5).For multiparous cows, ER t was positively associated with RFI (i.e., eating rate was greater for inefficient cows) at all time points except 0400 h (P ≤ 0.03; Table 6); for primiparous cows, ER t was associated with RFI at all time points except 0600 and 1000 h (P ≤ 0.02; Table 6).The rate of increase for ER t from low to high RFI differed between multiple time points for multiparous (P ≤ 0.05; Table 7) and primiparous cows (P ≤ 0.05; Table 8).In general, for multiparous cows, the slope of ER t across RFI was greater at 1000, 1400, 1800, and 2000 h compared with time points earlier in the day (Table 7).For primiparous cows, the slope of ER t across RFI was greater at 1800 h compared with early morning hours, and the slope of ER t was less at 0600 and 1000 h compared with most afternoon and evening hours (Table 8).In general, the differences in slopes of eating rates between daily time points mostly correspond to periods of relatively greater feeding activity in the afternoon, after feed delivery, compared with diminished feeding activity in the morning.Golden et al. (2008) evaluated eating rate in finishing beef steers over 2 experiments using 12 to 16 animals per experiment.A significant difference in eating rate was observed in only 1 experiment, where efficient steers fed a no-roughage diet had a greater ER several hours before feeding compared with their inefficient counterparts, but no differences were observed throughout the rest of the day (Golden et al., 2008).
There tended to be an interaction between RFI and time for temporal eating time (ET t ) in multiparous cows (P ≤ 0.07; Table 5) but not primiparous cows.The ET t was associated with RFI (ie, eating time was greater for less efficient cows) at 2000 h (P = 0.01; Slope of eating rate across the range of RFI ≠ 0 when P ≤ 0.05.
3 SE for the LSM = 1.16;SE for the slope = 0.92.
Table 9), but there was no detected evidence of an association between ET t and RFI at any other time point.The slope of ET t across RFI was greater at 2000 h than at 0000, 0400, 1200, 1400, and 1600 h (P ≤ 0.05; Supplemental Table S2; https: / / data .mendeley.com/datasets/ bn4mtg5hxk/ 1; Brown, 2022).Feeder occupation time (which was calculated similarly to ET t in our study) was assessed by Young et al. (2011) in pigs that were selected for low RFI over several generations.Feeder occupation time differed between the low-RFI and randomly selected control line throughout the day, depending on generation and dam parity.Low-RFI animals generally had more feeder occupation time during daylight hours, but the difference was less evident or nonexistent during nighttime hours (Young et al., 2011).
Clearly, feeding behaviors between growing swine and lactating dairy cattle are expressed in different manners due to diet and physiology differences between species.Nonetheless, the temporal results for eating time in our study and Young et al. (2011) demonstrated that eating time differs throughout the day depending on RFI, and this warrants further evaluation.
There tended to be an interaction between time and RFI for temporal bin visits (BV t ) in multiparous cows (P = 0.09; Table 5).There also tended to be an association between BV t and RFI (ie, bin visits were greater for inefficient cows) at 1000 and 1600 h (P ≤ 0.09; Table 9) but not at other time points.There were differences between slopes for BV t across RFI at different time points (P ≤ 0.05; Supplemental Table S3; https: / / data .mendeley.com/datasets/ bn4mtg5hxk/ 1; Brown, 2022); primarily, the slope of BV t at 1000 h was different than 0400, 1400, and 1600 h.As with the ET t results, the paper by Young et al. (2011)  the only other temporal bin visit analysis that we are aware of in production livestock.In pigs, the number of feeding visits was greater for animals bred for low RFI at several time points throughout the day, compared with the randomly selected control line (Young et al., 2011).The limited data on temporal feeding behaviors across a range of RFI in livestock species underscores the potential for future analysis across species and management scenarios.Additionally, there is opportunity for a focus on the potential interaction of RFI with management, diet composition, and energy demands of the animal.
There was an overall time effect for DMI t in both parity groups, regardless of RFI (P < 0.001; Table 5).The DMI t nadir was observed at 0600 h and was greatest at the time of feed delivery (1200 h; Figure 3).There were also smaller peaks in these FB following milking (0400 and 1500; Figure 3) and the second daily feeding (1700 h).Other researchers have observed peaks in feed intake following feeding and milking events in group-housed dairy cows (Ben Meir et al., 2019).In our data set, the greatest FB activity occurred with approximately 15 to 20% of feed disappearance taking place in the 2-h block after the initial feeding.Similarly, Heinrichs and Conrad (1987) estimated that 18 to 35% of feed is consumed within the first hour after feeding when cows are fed once daily.In primiparous cows, ET t and BV t temporal patterns were expressed similarly to DMI (P < 0.001; Figure 3); however, there was no detected evidence of an interaction between time and RFI.Temporal changes in FB should be taken in context of the daily milking and feeding patterns on any given farm and likely vary between farms.

Other Considerations
From this data set, it would appear that a slower daily or temporal eating rate would be the most compelling FB to pursue for identification of efficient animals; however, although highly associated with RFI, determination of eating rate still necessitates quantification of DMI.Perhaps DMI prediction models combined with the use of smart technologies to quantify FB related to eating rate could be applied in the future.Researchers have used electronic systems to quantify feed alley attendance patterns (DeVries et al., 2003) and to predict DMI using cow sensors equipped to quantify rumination time, lying time, and activity levels (Martin et al., 2021); therefore, accurate methods to quantify animal behaviors is attainable.Further work to ascertain whether specific behaviors (i.e., head down or feed contact time) are related to DMI and eating rate could also be useful, especially if they can be quantified using smart technologies.Additionally, if eating rate or other measurable traits can be used as an indicator of feed efficiency, it could become a variable considered in context of feed efficiency genetic selection efforts.There is an estimated genetic correlation, albeit weak (0.10 to 0.13), between RFI and feeding rate in beef and dairy cattle (Kelly et al., 2021;Cavani et al., 2022), but genetic selection is limited by the pace at which RFI phenotypes can be determined on research farms.The beef industry commonly measures RFI for bull marketing purposes using specialized feed efficiency centers.Assuming a similar process could be implemented on private dairy farms interested in adding value to their genetics, it would greatly enhance the geno-and phenotype data available to incorporate into genetic databases.Ultimately, although exploration of eating rate as a potential indicator of feed efficiency in lactating dairy cows is of interest based on our data, continued work should be conducted to understand the potential correlation with other traits.Selection for eating rate would have to be conducted carefully and in concert with multiple trait indexes to avoid indirectly selecting for undesirable traits.
Another approach to improving feed efficiency may be to leverage dietary factors known to influence eating rate, such as concentrate type and physical form (Kertz et al., 1981;Spörndly andÅsberg, 2006), diet sortability (DeVries et al., 2007), and forage particle length (Beauchemin and Yang, 2005;Leonardi et al., 2005;Spörndly and Åsberg, 2006).Certain management factors such as bin stocking density and feed delivery timing also affect eating rate (King et al., 2016;Crossley et al., 2017) and could potentially be used to reduce eating rate in inefficient cows.Whether reducing eating rate through these management strategies corresponds to increased feed efficiency remains unclear and warrants consideration.Considering that most research has explored the association of RFI and FB using RFI as the explanatory variable, an opportunity still exists to determine if FB mechanistically influences RFI.Ultimately, evaluating strategies such as these to increase efficiency of dairy cows will require more research and development of integrative technology.Creative approaches to determine cow-level eating rate on a large scale at the same time as leveraging existing factors influencing eating rate have the potential to improve feed efficiency in lactating dairy cows.

CONCLUSIONS
Our results revealed that efficient cows had a slower eating rate on a daily, meal, and temporal basis for both parity groups.Because there was no detected difference in the time spent eating between efficient and inefficient cows, the slower eating rate for efficient cows was associated with a lower DMI.Furthermore, temporal analysis revealed that the eating time and bin visits differed between efficient and inefficient cows at certain times of the day.Overall, further investigation is warranted for more specific behaviors within a meal that could influence eating rate (head down time, mastication time, sorting), including behaviors that could be more easily quantified on-farm.Development of creative solutions to ascertain individual cow eating rate through smart technology and prediction models could prove useful in a quest to manage and select for more efficient dairy cows in the future.
(8:1) executed inLancaster et al. (2009; growing bulls)   andGreen et al. (2013; dairy heifers)  certainly may have increased meal frequency for inefficient animals, although a stocking density of that level is not realistic on dairy farms.Conversely, reducing the bin density in our study to 1:1 may have increased meal frequency for all cows as observed inCrossley et al. (2017), although

Figure 1 .
Figure 1.Plot of residual feed intake (RFI) versus DMI for primiparous mid-lactation Holstein cows.Blue line is the line of fit for the quadratic response.

Figure 2 .
Figure 2. Plot of residual feed intake (RFI) versus meal eating rate for primiparous mid-lactation Holstein cows.Blue line is the line of fit for the quadratic response.
Brown et al.: FEED EFFICIENCY AND TEMPORAL FEEDING BEHAVIORS

Figure 3 .
Figure 3. Temporal changes in DMI (primiparous and multiparous; panel A), eating time (primiparous; panel B), and bin visits (primiparous; panel C) in mid-lactation Holstein cows (time P < 0.001).Data are presented with the 95% confidence interval for each time point.

Table 1 .
Descriptive statistics for primiparous and multiparous midlactation Holstein cows included in feeding behavior analysis

Table 2 .
Feeding behavior abbreviations and definitions t Number of times per 2-h time block the cow gained access to the feed bin and resulted in at least 0.1 kg of as-fed feed disappearance, n

Table 3 .
Brown et al.:FEED EFFICIENCY AND TEMPORAL FEEDING BEHAVIORS Descriptive statistics of feeding behaviors for primiparous and multiparous mid-lactation Holstein cows

Table 4 .
Association between residual feed intake (RFI) and feeding behavior characteristics in mid-lactation

Table 5 .
Brown et al.:FEED EFFICIENCY AND TEMPORAL FEEDING BEHAVIORS Temporal analysis (2-h time blocks) of the association between residual feed intake (RFI) and feeding behavior (FB) variables in midlactation Holstein cows Data for response variables were square root transformed for variance stabilization.Model fitting details are reported on the transformed scale.

Table 6 .
Least squares means and slope of eating rate (g/min) for residual feed intake (RFI) during 2-h time blocks throughout the day in multiparous and primiparous mid-lactation Holstein cows 1 SE for the LSM = 1.11;SE for the slope = 0.74. 2

represents
Brown et al.: FEED EFFICIENCY AND TEMPORAL FEEDING BEHAVIORS

Table 7 .
Differences between slopes for eating rate across residual feed intake (RFI) between 2-h time blocks throughout the day (above diagonal) and P-value (below diagonal) for multiparous mid-lactation Holstein cows 1

Table 8 .
Differences between slopes for eating rate across residual feed intake (RFI) between 2-h time blocks throughout the day (above diagonal) and P-value (below diagonal) for primiparous mid-lactation Holstein cows 1

Table 9 .
Slope of eating time and bin visits across a range of residual feed intake (RFI) during 2-h time blocks throughout the day in multiparous mid-lactation Holstein cows 2Slope of eating time or bin visits across the range of RFI ≠ 0 when P ≤ 0.05.