A longitudinal prospective cohort study investigating the association of premilking stimulation and teat-end shape on milking characteristics and teat tissue condition in dairy cows

Background Premilking udder preparation is essential for harvesting high-quality milk as gently, completely, and quickly as possible. The associations between characteristics such as teat-end shape and premilking stimulation on milking characteristics and machine milking-induced changes to the teat tissue condition have not been rigorously investigated. The primary objective was to investigate the interactive effects of manual premilking stimulation (i.e., preparation lag time) and teat-end shape on total milk yield, two-minute milk yield, milking unit-on time, and time in low milk flow rate. Our secondary objective was to study the association of manual premilking stimulation and changes to the teat tissue condition after machine milking (i.e., short-term changes). In a longitudinal prospective cohort study, 384 milking observations from 129 cows were analysed. Holstein cows were housed in sand-bedded free-stall pens, fed a total mixed ration, and milked 3 times a day. Cows were classified by teat-end shape into 1 of 3 categories: pointed, flat, or round. Individual cow milking characteristics were recorded with electronic on-farm milk meters. The duration of manual stimulation, preparation lag time, and presence of short-term changes were documented for each milking observation. General linear mixed models were used to study the interactive effects of preparation lag time and teat-end shape on milking characteristics. Results There was an interaction between preparation lag time and teat-end shape for two-minute milk yield and time in low milk flow rate. The preparation lag time effect was modified by teat-end shape, while no interaction was observed for total milk yield or milking unit-on time. A generalized linear mixed model revealed that preparation lag time was associated with short-term changes in teat tissue condition, where the odds of short-term changes decreased as preparation lag time increased. Conclusions In summary, cows with different teat-end shapes may require different premilking stimulation regimens. Increasing preparation lag time benefits teat tissue condition during machine milking. Further research is warranted to optimize individual premilking stimulation in dairy cows. Electronic supplementary material The online version of this article (10.1186/s12917-019-1803-2) contains supplementary material, which is available to authorized users.


Background
The ultimate objective of premilking udder preparation is to harvest high-quality milk from cows as gently, completely, and quickly as possible. In addition, it is valuable to monitor mammary health and thus animal well-being. Current recommendations for premilking udder preparation include teat sanitization (i.e., premilking teat disinfection, cleaning, and drying teats) and premilking stimulation [1]. Premilking stimulation is composed of some form of tactile stimulus (i.e., manual or mechanical stimulation) and milking unit attachment timing (i.e., preparation lag time). Premilking stimulation is essential for activating the milk-ejection reflex to harvest the alveolar milk, which represents approximately 80% of the udder's milk volume [2][3][4]. Improper premilking stimulation reduces milk flow, which can result in increased vacuum load on the teat tissue [5,6]. The premilking stimulation extent needed to achieve optimal milk flow is affected by stimulation technique [7,8] and cow characteristics such as breed [9], stage of lactation [3,9,10], relative degree of udder filling, and milking interval [5,11]. Although teat characteristics such as teat-end shape have been investigated in several studies on milk flow [12][13][14], their interactive effects with preparation lag time on milking characteristics have not been rigorously investigated.
Historically, most researchers have described the effects of premilking stimulation on milk yield, milking unit-on time, average milk flow rate, and peak milk flow rate [2,7,15]. Due to recent developments in on-farm milk meter technology, the ability to accurately measure milking characteristics has improved [10]. Among the many milking characteristics recorded by different milk meters from various dairy equipment manufacturers, two-minute milk yield and time in low milk flow rate are reported to be valuable measures of parlour performance including milking routine [16,17]. Our study investigated the interactive effects of preparation lag time and teat-end shape on total milk yield, two-minute milk yield, milking unit-on time, and time in low milk flow rate in dairy cows milked 3 times daily. We hypothesized that the effect of preparation lag time on milking characteristics was modified by the effect of teat-end shape.
Most studies investigating premilking stimulation have concentrated on milking characteristics to improve parlour efficiency and profitability [18][19][20]. In contrast, information on the relationship between premilking stimulation and teat tissue condition after milking is scarce with no consensus [7,21]. Albeit, the degree of machine milking-induced changes on teat tissue condition may be associated with the risk of new intramammary infections [22,23]. Thus, reducing the risk and the degree of machine milking-induced changes in teat tissue condition may improve udder health and cow well-being. Our second objective, therefore, was to investigate the association between premilking stimulation and machine milking-induced changes to the teat tissue condition. We hypothesized that premilking stimulation was associated with machine milking-induced changes to the teat tissue condition.

Animals and housing
This study was conducted during October and November 2015 at the Teaching Dairy Barn of the College of Veterinary Medicine, Cornell University (Ithaca, NY). Holstein cows were housed in 2 free-stall pens with sand bedding and fed a total mixed ration consistent with National Research Council [24] requirements. Cow characteristics such as parity, stage of lactation (DIM, days in milk), and previous lactation 305-d mature equivalent milk yield were obtained from a dairy management software program (Dairy Comp 305, Valley Agricultural Software, Tulare, CA).

Enrolment
The study population consisted of 150 lactating Holstein cows at the Teaching Dairy Barn of Cornell University. Cows (n = 129) without clinical mastitis for the previous 4 weeks and no udder abnormalities, such as nonlactating quarters or teat injuries, were eligible for enrolment. Data collection was subdivided into 3 intermittent days (3 consecutive Saturdays), evaluating a subset of cows to minimize the impact of data collection on milking routine. Data from each subset of cows were collected over 3 consecutive milking sessions.

Teat-end shape
Teat-end shape on the quarter level was assessed prior to data collection with a digital voice recorder (ICD-UX533BLK, Sony, San Diego, CA) and classified into 1 of 3 categories: pointed, flat, or round. To investigate teat-end shape on cow level (TES), cows were classified into 3 groups based on the teat-end shape frequency of all 4 quarters as follows: cows with 2 or more pointed teats were classified as pointed, cows with 2 or more flat teats as flat, and cows with 3 or 4 round teats were classified as round. No cows were described to have 2 pointed and 2 flat teats or 2 round, 1 pointed, and 1 flat teat.

Milking and premilking udder preparation
Cows were milked 3 times daily at 0400 (milking #1), 1100 (milking #2), and 1900 (milking #3) h in a 2 × 10 parallel milking parlour (P2100, DeLaval International AB, Tumba, Sweden). The system vacuum was 46 kPa with an average claw vacuum of 39 kPa during the peak milk flow period. The pulsator (EP100, 96,679,084, DeLaval International AB) was set at a pulsation rate of 60 cycles/min, a pulsation ratio of 65:35, and a side-to-side alternating pulsation (duration of pulsation phases: a-phase, 120; b-phase, 530; c-phase, 90; and d-phase, 260 ms). Milking clusters were removed automatically when milk flow decreased to below 1.3 kg/min (delay time: 0 s). The milking unit consisted of the milking cluster MC70, weighing 2.1 kg (DeLaval International AB), and the milking liner LS01 (LS01 SR 12 mm, 834,115,001, DeLaval International AB). The milking liner characteristics were as follows: mouthpiece bore diameter, 20.3 mm; barrel shape, round; barrel length, 113 mm; and barrel diameter at 75 mm, 21.8 mm. The long milk tube was silicone (90,843,501, DeLaval International AB) with an inner diameter of 19 mm. The milk line was installed 75 cm below cow standing level.
Milking routine was performed by 2 operators per session. Operators were composed of farm staff (n = 3) and students (n = 6) of the College of Veterinary Medicine, Cornell University. Premilking udder preparation was performed in sets of 5 cows and performed in 3 steps as follows: step 1 was to preclean the teats with a dry cloth towel and dip with an iodine-based teat dip (Udder-dine110, Boumatic, Madison, WI, USA); step 2 was to forestrip all 4 teats then dry the teats with a separate clean cloth towel; and step 3 was to attach the milking unit. The time per task (precleaning, dipping, forestripping, drying time, and unit attachment) was recorded for each cow milking observation with a digital voice recorder (ICD-UX533BLK, Sony) by 2 investigators. Recorded time points (start and stop) per task were transferred to a spreadsheet (Microsoft Excel, 2013) using a comprehensive audio application program (Sound Organizer 4.1, Sony). Manual stimulation and preparation lag time durations were calculated. Manual stimulation time (STIM) was defined as the duration of forestripping and subsequent teat drying with a cloth towel, and preparation lag time (LAG) was defined as the time spent from the first teat stripping until the milking unit was attached.

Milking characteristics
Total milk yield (kg; TMY), two-minute milk (kg; 2MIN), milking unit-on time (s; DUR), time in low milk flow rate (s; LMF), and milking mode (manual versus automatic take-off) were assessed at each milking with electronic milk meters (MM27, DeLaval International AB) and recorded using the herd management system ALPRO (DeLaval International AB). Additional file 1: Table S1 provides definitions for reported milking characteristics.
Machine milking-induced short-term changes to the teat tissue condition Machine milking-induced short-term changes to the individual teat conditions were visually assessed by the same investigator within 60 s after unit detachment and based on the scoring system described by Hillerton et al. [25]. Key characteristics included skin colour changes, teat base condition, teat-end firmness, and teat orifice openness. Teat skin colour was evaluated as normal (score 0), red (score 1), or blue (score 2). Teat base condition showed no visible mark (score 0), visible mark present (score 1), or significant swelling (score 2). Teat-end consistency was scored soft (score 0), firm (score 1), or wedging present (score 2). The teat orifice was scored into 1 of 2 categories: opening < 2 mm (score 0) or opening ≥2 mm (score 1). All scores were documented with a digital voice recorder (ICD-UX533BLK, Sony) and subsequently transferred to a spreadsheet (Microsoft Excel, 2013). For subsequent analysis, machine milking-induced short-term changes to the teat condition were dichotomized as follows. A short-term change was present if the teat base score condition was 2, or the teat-end consistency score was ≥1, or the teat orifice score was 1. Short-term changes were considered absent otherwise. Pigmentation was noted in 84/516 (16.3%) of the teats. Because this impeded consistent evaluation, teat skin colour was excluded from subsequent analyses.

Analytical approach
Prior to statistical analyses, the data from 387 cow-milking observations of 129 cows were investigated for missing and erroneous values, and 1 milking observation each from 3 cows were excluded because of erroneous values (average milk flow rate = 0 kg/min). Statistical analyses were performed using R Statistical Software (R Core Team, [26]).

Baseline characteristics
Descriptive statistics of the previous lactation 305-d mature equivalent milk yield, TMY, 2MIN, DUR, LMF, STIM, and LAG are presented as the mean and standard deviation, median, minimum value, and maximum value. Stage of lactation (DIM) was stratified into 3 categories (early lactation, ≤ 100; mid-lactation, 101-200; and late lactation, > 200 DIM), and parity was classified into 3 categories (1st, 2nd, and ≥ 3rd lactation). Frequency distributions for parity, DIM, and TES are presented as absolute values and percentage.

Milking characteristics
Cow milking observation was the unit of analysis, and the continuous variables, TMY, 2MIN, DUR, and LMF, were the outcomes of interest. A separate general linear mixed model was fitted for each dependent variable (TMY, 2MIN, DUR, and LMF) using the 'nlme' package [27]. To account for clustering of milking session within cow, cow was included as a random effect. To model the within-cow covariance of repeated measurements, 3 covariance structures (compound symmetry, autoregressive order 1, and unstructured) were tested for each dependent variable investigated. The covariance structure with the smallest Akaike's information criterion was selected. F-tests for Type-III analysis of variance were calculated using the 'car' package [28]. The initial model included the independent variables: milking session, parity, DIM, TES, STIM, LAG, and the interaction term between TES and LAG. For the dependent variables 2MIN, DUR, and LMF, TMY was also included as an independent variable. Collinearity among eligible variables was assessed by calculating the variance inflation factor from the initial models. A variance inflation factor > 5 was considered to indicate multicollinearity. No collinearity was observed among the variables: milking session, parity, DIM, TMY, TES, STIM, or LAG. Backward stepwise selection was performed until each independent variable had a p-value of ≤ 0.05. Confounding effects were monitored by observing regression coefficient changes. Variables that modified regression coefficients by > 20% were considered confounding factors. No confounding was observed. For the final models, homoscedasticity and residual normality assumptions were assessed by plotting residuals versus corresponding predicted values and examining residual quantile-quantile plots. To satisfy these assumptions, data for the dependent variable LMF were log transformed. Resulting coefficient and least squares estimates were consequently back-transformed and presented as the geometric mean and 95% CI. Least squares means (LSM) ± SE and LSM (95% CI) were calculated using the 'lsmeans' package [29]. Results were averaged over all variable levels included in the final model. Tukey-Kramer's post hoc test was used to control for experiment-wise error rate for comparing means across different categorical variable levels. Least squares means ± SE and LSM (95% CI) were calculated with the 'at' function in the 'lsmeans' package [29] as follows: 2MIN for a hypothetical milking observation during milking #1 of a cow with a 15-kg TMY; pointed, flat, and round TES; and a LAG of 60 and 90 s, respectively; LMF for a hypothetical milking observation during milking #1 of a cow in 2nd lactation with a 15-kg TMY; pointed, flat, and round TES; and a LAG of 60 and 90 s, respectively.
Machine milking-induced short-term changes to the teat tissue condition Cow milking observation was the unit of analysis, and the dichotomous variable, machine milking-induced short-term changes to the teat condition on cow level (STC), was the outcome of interest. Machine milkinginduced short-term changes to the teat condition on cow level were considered present if 1 or more teats showed short-term changes as defined above, and absent otherwise. A generalized linear mixed model with a logit link and a binomial distribution was fitted with the 'lme4' package to determine the association between the dependent variable, STC, and the independent variables, milking session, parity, DIM, TES, TMY, STIM, and LAG [30]. To account for within-cow clustering between milking sessions, cow was included as random effect. Univariable associations were tested between the dependent and all independent variables. All variables with a p-value ≤ 0.20 in this step were considered for inclusion in the initial model. Collinearity among eligible variables was assessed by Pearson's correlation for continuous variables and Spearman's correlation for categorical variables. No collinearity was observed among the eligible variables, milking session, parity, DIM, TMY, TES, and LAG (correlation coefficients ≤ |0.41|). Backward stepwise selection was performed until each independent variable had a p-value of ≤ 0.05. Type-III Wald chi-square tests were calculated using the 'car' package [28]. Two-way interactions between remaining variables were investigated and retained in the model if p-value ≤ 0.05. Pearson goodness-of-fit statistic of a logistic regression model, including the same fixed but not random effects, was used to indirectly assess the final model fit. The adjusted probabilities (95% CI) of STC for a hypothetical milking observation during milking #1 of a late lactation cow, and a LAG of 60 and 90 s was calculated using the ' AICcmodavg' package [31].

Milking characteristics
The interactive effects of LAG and TES on milking characteristics were evaluated based on the 4 milking characteristics, TMY, 2MIN, DUR, and LMF. General linear mixed model results are provided in Additional file 3: Tables S3-S6 and described separately below.
Machine milking-induced short-term changes to the teat tissue condition Table 2 depicts unadjusted regression coefficients, SE, unadjusted odds ratios (OR), and 95% CI for independent variables tested for the association with STC from univariable generalized linear mixed models. The final  Table 3). The adjusted probability (95% CI) of STC for a hypothetical milking observation during milking #1 of a late lactation cow and a LAG of 60 and 90 s was 60% (37-80%) and 40% (20-64%), respectively.

Discussion
This study investigated the interactive effects of LAG and TES on milking characteristics (TMY, 2MIN, DUR, and LMF). We observed an interaction between TES and LAG for 2MIN and LMF, but no interaction for TMY and DUR. The LAG effect was modified by TES, as an increased LAG yielded an increased 2MIN in cows with flat TES, while it had no meaningful effect in cows with round TES, and it decreased the 2MIN in cows with pointed TES. In addition, increased LAG yielded higher LMF in cows with pointed and flat TES, but resulted in a slightly decreased LMF in cows with round TES; thus, cows with different TES responded differently to LAG. This suggests that cows with different TES may require different premilking stimulation regimens to achieve optimal milk flow. Our study adds to the existing literature indicating that cows with different characteristics (i.e., udder filling and stage of lactation) require different preparation lag times to achieve optimal milk let-down [5,10,18]. Addressing these requirements in premilking stimulation may improve parlour efficiency and alleviate the potentially harmful effects of machine milking on teat tissue condition in dairy cows.
The lack of association between premilking stimulation and TMY is consistent with some studies [15,32,33], but contrasts with other studies that found a positive association between premilking stimulation and milk production [8,9,34].
Teat-end shape was associated with DUR, as cows with flat TES milked fastest, while those with pointed TES milked slowest. Our findings indicated that milkability differed among cows with different TES, supporting descriptions by other investigators that cows with flat TES had the largest 2MIN [14,35]. Increased milking speed has been a breeding objective for increased milking efficiency and parlour throughput in recent years [36,37]. As suggested by Guarín and Ruegg [38], this may have influenced teat characteristics such as TES. Possible factors associated with TES that may explain differences in cow milkability are teat canal length [12] and diameter [39]. In contrast to previous studies   [2,33,40], premilking stimulation did not influence DUR in this study. Discrepancies in milking frequency and automatic cluster removal thresholds could account for the differences observed between those studies and this one. Differences in milking characteristics among milking sessions were likely attributed to differences in milking intervals between sessions, as well as differences in milking routine including cow handling and premilking stimulation (type of forestripping) among different operators.
Several factors may explain the interactive effects of TES and LAG on milking characteristics. Possible explanations include different dimensions in anatomical structures such as teat canal and gland cistern, which may act in concert with milking machine settings and equipment (i.e., milking liner) yielding different responses to premilking stimulation among cows with different TES. Another variable that may explain this is the adrenoceptor pattern difference (beta-2:alpha-2-adrenoceptor ratio), which may have led to differences in the premilking stimulation response, and thus milkability among cows with different TES. However, although these anatomical and physiological characteristics have been reported to be associated with milkability by previous investigators [12,41,42], these hypotheses remain to be tested. Therefore, this work must be extended to further understand the differences in milkability among cows with different characteristics and determine the potential of the individual premilking stimulation regimen as a possible method of improving milking efficiency, teat tissue condition, and cow well-being.
Our study also investigated the association between premilking stimulation and STC. Our results show that increased LAG decreased STC probability, while no association was seen between STIM and STC. This contrasts with results reported by Vetter et al. [7] who found no differences in teat tissue condition as assessed by ultrasonography when comparing 2 different premilking udder stimulation regimens. Several reasons may explain the benefits of LAG on teat tissue condition after machine milking.
One reason could be that LAG may affect teat diameter. Teat diameter has been shown to increase before milking due to milk accumulating in the teat cistern [43]. To form a seal between the teat and milking liner barrel, teats must be a few mm wider than the liner barrel diameter [44]. Therefore, the increased LAG in our study may have resulted in an increased teat diameter, yielding a better seal between the teat and the liner barrel. This could have lowered the mouthpiece chamber vacuum [45], thereby decreasing teat congestion and oedema, as outlined by Penry et al. [46]. Another possible explanation could have been that LAG affected the milk flow rate. Multiple studies demonstrated that preparation lag time positively affects continuous milk flow and decreases bimodal milk let-down [8,10,18], which is reported to increase vacuum load resulting in teat congestion and oedema [2,5].
Although we intended to best represent current milking practices, our study had some limitations that must be considered. To reduce inadvertent variability, we excluded cows with a history of mastitis during the 4 weeks prior to enrolment as well as cows with a nonlactating quarter. This likely introduced a selection bias and decreased the generalizability of our results. Another limitation may have been the subjective assessment of TES and STC, which may have yielded information bias. Another limitation is that the study was observational and has been conducted on a single farm during one season (i.e., fall); thus, the results should be interpreted cautiously. These results should be confirmed in a controlled experiment investigating the effect of different premilking stimulation regimens in cows with different TES from several dairy farms considering the effect of climate and season across different regions, as well as different management systems (e.g., housing and milking routine).

Conclusions
In conclusion, LAG and TES had an interactive effect on 2MIN and LMF, indicating that cows with different TES respond differently to premilking stimulation. In contrast, premilking stimulation was not associated with TMY or DUR. Preparation lag time was associated with STC probability, which decreased as the LAG increased. Further studies investigating individual milk flow curves among cows with different TES are warranted to better