Length of lags in responses of milk yield and somatic cell score on test day to heat stress in Holsteins

Abstract We used daily records from provincial Japanese weather stations and monthly test‐day records of milk production to investigate the length of the lags in the responses of cows’ milk yield and somatic cell score (SCS) to heat stress (HS). We also investigated the HS thresholds in milk yield and SCS. Data were a total of 17,245,709 test‐day records for milk and SCS in Holstein cows that had calved for the first time between 2000 and 2015, along with weather records from 60 weather stations. Temperature–humidity index (THI) values were estimated by using average daily temperature and average daily relative humidity. Adjusted THI values were calculated by using temperature, relative humidity, wind speed, and solar radiation. The model contained herd, calving year, month of test day, age group, days in milk, and THI as a fixed effect. THIs for each day from 14 days before the test day until the test day were used to represent the HS effects. The HS occurring 3 days, and between 8 and 10 days, before the test day had the greatest effect on the milk yield and SCS, respectively. The threshold THI values for the HS effect were about 60–65 for both traits.

In previous studies, a delayed response of test-day yield to HS was reported. Hammami et al. (2015) assigned a THI averaged over the 3 days before the test day for yield traits and SCS in their analysis of HS effects in Luxembourg and Germany. Bernabucci et al. (2014) reported that the greatest negative effect on yield traits (milk, fat, and protein yields and fat and protein percentages) in Italian Holsteins was observed when the HS occurred 4 days before the test day. Carabaño et al. (2016) used the average THI of the test day and the two preceding days in their analysis of HS, to take into account the length of the lag in the response to HS in Holsteins in Belgium, Luxembourg, Slovenia, and Spain; the results varied across countries.
The number of studies of SCS response to HS is limited compared with those of milk yield and components. Moreover, to our knowledge, no published reports have used THI to investigate the length of the lag in the response of SCS to HS in Japan or northeast Asia.
Here, our main aim was to elucidate the lengths of the lags in responses of milk yield and SCS to HS in cows by using test-day records and daily weather records from provincial weather stations in Japan. We also investigated the threshold values of THI in milk yield and SCS.

| Data
Test-day records of milk and somatic cell count (SCC) at 6 through 305 days in milk in Holstein cows that had calved for the first time between 2000 and 2015 were provided by the Livestock Improvement Association of Japan (Tokyo, Japan).
Records were collected through the Dairy Herd Improvement Program. The data included records of test-day milk yield and SCC for first-lactation cows from all over Japan. The SCCs were log-transformed into SCSs by using the following formula (Ali & Shook, 1980): Weather records from 60 provincial weather stations for the period from 2000 to 2015 were obtained from the website of MeteoCrop DB (an agro-meteorological database coupled with crop models; Institute for Agro-Environmental Science, 2017).  reported that maximum daily temperature and minimum daily relative humidity were the most critical variables in calculating the THI to quantify the HS. However, in our preliminary study using weather records in Japan, we found that the THI based on average temperature and average relative humidity in a day was more effective than that calculated by using maximum temperature and lowest relative humidity in a day. Therefore, THI values were estimated by using average daily temperature and average daily relative humidity.
First, the THI was estimated by using the following formula (NRC, 1971): where t is the temperature in degrees Celsius and rh is the relative humidity as a percentage. In the preliminary study, we estimated the HS effect as the THI adjusted for a daily average wind speed and solar radiation in a day (Mader et al., 2006). However, the AICs with adjusted THI were similar to those with THI. That is, no superiority of adjusted THI over THI was observed under Japanese weather conditions.
A total of 17,245,709 test-day records from 2,018,406 cows were used. Five subsets (divided randomly by herd because of computing memory limitations) were analyzed separately in the cases of milk yield and SCS. Mean daily milk yield was 27.1 kg and mean SCS was 2.33 (Table 1). Mean THI ranged from 50.7 to 51.0 ( Table 2). The minimum THI was 4; the respective maximum value was 84.

| Model
Test-day records were linked to the data from provincial weather stations in the 14 branches in Hokkaido, which is a northern island in Japan, and in the other 46 prefectures. The effects of HS were estimated by using a statistical model, as follows: (1) Milk yield (kg/day)  Assuming that the effect of HS was linear, the breakpoint of THI was estimated by segmented-regression analysis by using the Segmented package (Muggeo, 2003(Muggeo, , 2008 of (R Core Team, 2014).
In a way similar to the method of Carabaño et al. (2016), the LSMs of THI effects in Equation (2) were used as the dependent variables and the slope of the segmented linear regression at lower than the breakpoint was assumed to be 0, as follows: where y * i is the LSM of the THI effect estimated by using 3 (or 8) days before test day in milk yield (or SCS), c is a constant, a is an intercept, b is a regression coefficient on THI x i , and e i is the random residual term. BP is the breakpoint, defined as the appropriate threshold value of THI when the linear regression was applied to the HS effect. The number of records on THI is shown in Figure 1.
We used THI classes with more than 20,000 records for segmentedregression analyses owing to the stability of LSM estimates (THI, ranging from 18 to 82).

| Length of lags in response to HS
When we used THI to model the effect of HS, the estimated AIC for milk yield decreased sharply from test day to 3 days before the test day and then increased toward 14 days before the test day ( Figure 2). For SCS, the estimated AIC with HS using THI decreased gradually from the test day to 7-10 days before the test day and then increased a little toward 14 days before the test day ( Figure 3).

| Least-squares means for the effect of heat stress
When the effect of HS was assumed to be linear, the estimates of   Bohmanova, Misztal, Tsuruta, Norman, and Lawlor (2008) found in their preliminary study that the weather data 3 days before the test day explained more of the variability in milk yield than the data on the 2 days before the test day or on the test day itself. Bernabucci et al. (2014) reported that the greatest negative effect was observed when the HS occurred 4 days before the test day. West et al. (2003) similarly reported that milk yield was affected by the THI as recorded 2 days before the test day. Hayes et al. (2003), in contrast, found in preliminary investigations of their data that the THI on the test day and 1, 2, 3, and 4 days before the test day had significant effects on test-day yield. Here, we found no significant differences between the estimated AIC for milk yield using the THI 3 days before the test day and those from 2 to 6 days before the test day. Our finding here of a critical effect of the THI 3 days before the test day on milk yield was thus generally in line with these previous results.
In their analysis of HS effects, Hammami et al. (2015) and Santana, Bignardi, Pereira, Stefani, and El Faro (2016) used the THI averaged over the 3 days before the test day to examine the effects on yield traits and SCS. Smith, Smith, Rude, and Ward (2013) also used the averaged THI for the 3 days before the test day to estimate the effect of HS on SCS in Holstein and Jersey cows. Here, we found no significant differences between the AIC estimated for SCS by using the THI 10 days before the test day and those from 5 to 14 days before the test day. However, our results showed a longer lag time for SCS than milk yield, with the greatest response to HS between 7 and 10 days before the test day.
Dry matter intake (DMI) is sensitive to the mean air temperature 2 days earlier than the test day (West et al., 2003), and it causes a decrease in the availability of nutrients used for milk synthesis (Polsky & von Keyserlingk, 2017). Maintenance energy requirements are increased during heat stress because of the high energy levels required for heat loss from the body during heat stress (Atrian & Shahryar, 2012). These effects lead to a negative energy balance, resulting in a decrease in milk.
A negative effect of HS on yield traits is found when the THI reaches 72 . The HS thresholds for a THI over reported that the THI threshold value was 62 for milk yield. By using the method of Hayes, Bowman, Chamberlain, Verbyla, andGoddard (2009), Nguyen, Bowman, Haile-Mariam, Nieuwhof, andHayes (2017) estimated a genetic evaluation of heat tolerance with a threshold of THI 60. Bernabucci et al. (2014) reported that the threshold varied among studies; they concluded that this variation may have occurred because of variations in the methods used to detect the THI thresholds. They also suggested that threshold values were affected by the type of herd cooling system, such as fans or sprinklers. Carabaño et al. (2017) concluded that estimates of thresholds can vary with climatic regions and milk production levels. In recent reports, relatively low THI thresholds, ranging from 60 to 65, have been reported for HS effects on yield traits (e.g., Ammer, Lambertz, & Gauly, 2016;Nguyen, Bowman, Haile-Mariam, Pryce, & Hayes, 2016). Our THI thresholds seemed to be about 60 and 65 for both traits-similar to the values in these recent reports. However, further studies on HS thresholds are needed to elucidate quadratic changes in the effect of HS on milk yield. For SCS, when the THI values were less than 45, the estimated effects of THI appeared higher than when the THI values were around 50 (see Figure 5); the optimal THI for Holstein cows seemed to be around 50. However, the differences between the effects of low THIs and moderate THIs were quite small. For SCS, Bertocchi et al. (2014) reported that the slope (0.0003 × THI) on THI was small before the breakpoint. Our results were in line with their trends, and the breakpoints we obtained seemed to be somewhat higher than the THI thresholds. However, the breakpoints estimated here may be applicable when the genetic merit of animals for heat tolerance is estimated by using a random regression model (Atagi et al., 2018;Santana et al., 2016).
We investigated the lengths of the lags in responses of milk yield and SCS to HS in cows by using test-day records and daily records from provincial weather stations in Japan. The HS occurring 3 days before the test day had the greatest effect on milk yield. For the effect on SCS, there was a longer lag in the response to HS: the peak response was when the HS occurred between 7 and 10 days before the test day. The threshold THI effect values of thresholds were about 60-65 for both traits.