Milkalyser: Science

The link between hormones in milk and ovulation was discovered in the 1970s

Since artificial insemination was introduced the biggest problem facing the  dairy farmer is the problem of identifying ovulation. This is traditionally done by detecting the change in behaviour known as oestrus (estrus) which usually precedes ovulation. The figure shows how the hormone level in milk rises and falls through the 21 day ovulation cycle and how a drop precedes ovulation.  Cows should calve approximately annually but worldwide the average calving index is now about 440 days.  The outcome of our inability to get cows pregnant ia a loss of milk productive capacity and large numbers of healthy fertile animals culled as infertile.  Since the 1970s we have known that we can map the ovulation and oestrus cycles accurately by measuring progesterone in milk. Other hormones are also present but at low concentrations and for short durations that could fall between milkings.

Schematic graph of hormones in milk

The bovine ovulation cycle is 21 days and starts about 15 days after calving the best marker for it is progesterone

Milkalyser directly measures progesterone in milk and maps the ovulation cycle.  We know that 97% of dairy cows have functioning ovaries but only 70% show behavioural oestrus.   The gold standard test for ovulation is measuring the drop in progesterone that occurs before oestrus/ovulation.  Milkalyser measures progesterone directly and so ovulation can be predicted allowing optimal timing of insemination.


History and details of the science

Laboratory analysis of progesterone became common from the 1980s after the pioneering biological science in the 1970s at Nottingham and other universities. The first engineers to demonstrate an automated  system were Rod Claycomb and Michael Delwiche in Claycomb_etal1995 but it has been superseded by new developments in biosensing.

In the late 1990s I and my team developed  a progesterone analyser that took only a few minutes and then deployed it in a milking parlour. A sensor based method of detecting ovulation, 2001 The system took a milk sample about 90 seconds after the start of milking and passed it to a sensing chamber where a signal from the electrochemical sensor relating to the concentration of progesterone was measured. The numbers were then sent to a database of previous values. The drop in progesterone concentration below 5ng/ml was the signal that ovulation was imminent. We characterised the flow mechanics of the sensor Quantitative analysis of the response of an electrochemical biosensor of the milk droplets onto the sensor to improve the performance. This technique could have been developed into a comprehensive online analyser Experiments in Automatic Ovulation Prediction 2002 but circumstances prevent me taking it forward then.   Milkalyser is the result of waiting for the right combination of sensors and electronics to become available.

I wrote about the latest efforts to solve the problem in an invited review for Animal published in 2015.   An extract of the relevant sections is below or the full paper is available on my pages.  Every system that has been tried to detect oestrus is always compared to the gold standard of progesterone assay and to me the obvious solution to this problem is to automate the inline assaying of milk for progesterone concentration.


Oestrus detection

The calving index or numbers of days between calving has been steadily rising since the introduction of artificial insemination at about 1-2 days per year and in 2009 it was 420 days in the UK ( A number of studies summarised by Boyd (1992) showed 95% of cows had active ovaries. As milk yields have risen and body condition declined with more extreme Holstein breeds the number may now be lower, 71% was measured Lopez et al. (2005) so failure to observe oestrus is the most important factor preventing higher conception rates. Claus et al., (1983) measured progesterone in milk fat in 123 cows from a number of farms to determine factors influencing fertility. The greatest influence on the cow fertility was management, with 32% of cycles detectable by progesterone not being recognised by the herdsman. Between 5% and 21% of cows were inseminated at the incorrect time in the cycle.

A wide variety of techniques and devices have been proposed and tested to improve oestrus detection. However, this review assumed that there could only be a limited role for devices that need to be inserted into the cow either surgically or into the vagina; these devices were reviewed by Senger (1994) and they have little market presence.

The oestrus cycle The sequence of events associated with the ovarian cycle in dairy cows is well known and described in a number of standard texts such as Peters & Ball (1995). For the purposes of detection of the stage of the ovarian cycle the cow may emit a number of hormonal and behavioural signals which may be measured externally. Progesterone levels are elevated from 4 – 16 days post ovulation then fall to a nadir. The hormone oestradiol has an inverse relationship to progesterone so that as progesterone levels fall that of oestradiol rises. The oestradiol stimulates the oestrus behaviour which is most commonly used to select cows for insemination. The optimum time to inseminate the cow is 6- 12 hours after the peak in oestradiol concentration. Schofield (1988) describes diurnal peaks of oestrus behaviour early in the morning and late at night. However, he showed that standing to be mounted was not a reliable indicator of a cow being suitable to be served, with as many as 21% of ridden cows were pregnant. Warren (1984) surveyed 35000 cows in 255 herds and showed that 26% of interservice intervals were greater than 48 days and that 19% of cows were served at an interval of 1-17 and 25-35 days suggesting that one or other of the observed oestrus events was not accompanied by ovulation whose periodicity has been established as very consistent. Frequently the performance of a device is compared with the ability of the herdsman to observe behavioural indications of oestrus. A more satisfactory reference to use is milk or blood progesterone level, which gives a reliable indication of ovulation.

Oestrus: Mount Detectors Stevenson et al., (1996) report the use of an electronic, radiotelemetric, pressure-sensitive, rump-mounted device, which was designed to be activated by the weight of a mounting animal. The sensor sent a signal by radio to a computer which recorded the mounted animal’s identification, date, time and duration of the mounting. An experiment was carried using 49 peripubertal, crossbred yearling beef heifers with synchronised oestrus. The heifers were inseminated after oestrus was detected, or after a fixed interval if oestrus was not detected. The performance of the sensor system was compared to conventional observation (for 45 minutes at 7.30 and 16.30). Observation detected 30 heifers, all of which were also detected by the device. The device detected an additional 11 heifers which were not detected by observation. Neither method detected oestrus for the remaining eight heifers although two of these conceived. One development of the historic dye patch system is to use a camera system to automatically measure whether the heat patch has been triggered Hempstalk et al. 2013 and despite some initial problems ensuring that cameras were correctly set up achieved an acceptable correlation to herdsmans observations mapped to progesteron analysis. The high level of success of the herdsman in detecting oestrus in this experiment with 797 cows (over 80%) indicates a commercial herd with very good results without automation.

Oestrus: Pedometers Pedometers are electronic devices which are strapped to a cow’s leg to count steps. The older designs contain an electrical switch which opens or closes when the device is moved. Each activation of the switch adds to a count in the pedometer. The total count is transferred to a base station by telemetry when the cow is in the vicinity of a receiver unit. Since the number of movements counted by a pedometer has no absolute significance, it is necessary to establish a comparative technique in which a count is compared to a baseline count which would be expected from the cow if it were not exhibiting oestrus. If a count exceeds the baseline by some predetermined multiplier, the cow is deemed to be in oestrus. The main variables involved in this process are therefore how frequently the total counts are transferred to the base station (this is often twice a day at milking), and the multiplier that is used to set the oestrus alert threshold. The Alpro from de Laval sends a signal to the base station whenever it detects a rise above threshold.

There have been many studies aimed at assessing the effectiveness of pedometers. For example Peter & Bosu (1986) carried out a trial with 47 cows at pasture.  Pedometer readings were taken twice a day at milking. No details are given of the alert threshold that was used, but 76% of ovulations were detected by pedometer,  compared to 35% by herdsman observation (30 minutes twice a day). All of the cases that were detected by the herdsman were also detected by the pedometers. The reference method for ovulation detection was measurement of blood progesterone concentration.

Koelsch et al., (1994) used pedometers on 21 cows. The reference method was milk progesterone level. Various methods of analysing the pedometer data were tried. The most successful method, which involved several stages of data processing, provided a specificity (SP) of 99% but a sensitivity (SN) of only 69%. The highest sensitivity of any of the attempted methods was 76%, and this was at the expense of a reduced specificity of 94%. It is important to note that the procedure was tested on the same data that were used to develop it, so the performance of the method on a fresh set of cows (or even a fresh set of data from the same set of cows) is unknown. All of the above studies support the general conclusion that pedometers are not capable of providing completely reliable ovulation detection. If a specificity of close to 100% is required (i.e. a minimal number of false positives) the associated sensitivity (the proportion of ovulating cows that are detected) is generally less than 70%. There are at least two possible reasons for this less than ideal performance. One is that activity is not a totally reliable indicator of ovulation, and the second is that the pedometer is not totally reliable method of registering the relevant activities. Van Vliet & Van Eerdenburg (1996) investigated these two possibilities. They fitted pedometers to 37 non- pregnant cows among the herds on two dairy farms. The total number of cows on the two farms was about 100. They also observed the behaviour of the 37 cows for periods lasting 30 minutes at two hour intervals for six weeks. This very careful observation resulted in ovulation detection with a sensitivity of 74% and a specificity of 100%. Step counting is not reliable as an indicator of ovulation as about 20% of cows do not exhibit the behaviours.

Oestrus: Detection with Collars Collars for detecting oestrus have been available since the early eighties. The early versions used simple mercury tilt switches or rolling balls to count the number of head movements. These were relatively expensive devices until the integrated circuits with tri-axial accelerometers became available in the 1990s. Since the advent of cheap tri-axial accelerometers and digital signal processor chips the collar has been shown as a tool that can not only detect oestrus but also lying behaviour, lameness, location, and with the collection of audio data rumination and eating activities. These collars have been used widely and models are slowly emerging as to how the data can be integrated into a wholistic management system. Commercial activity has probably been more advanced than research reports. The Voronin et al., (2011) invention provided a method and device for detecting oestrus in animal by sensing along time the motion of the animal and identifying when the sensed motion is not related to eating periods of the animal. Kamphuis et al., (2012) had similar results to those with pedometers and achieved 76.9% SN, 99.4% SP and 82.4% PPV. While activity only collars achieved 62.4% SN, 99.3 SP, 76.6% PPV all in comparison to progesterone analysis of milk as the gold standard. This and other patents in this area shows over 50 listed each attempting improvements on the basic concept.

Oestrus: Milk temperature A cow’s body temperature rises at oestrus. A non- invasive method of oestrus detection is to measure the milk temperature in the claw piece or short milk tube of the milking system. Maatje et al., (1987) carried out two experiments; one with 28 housed cows, and one with 20 cows that were grazing during the day and housed at night. Oestrus was assumed to be associated with a significant (twice the standard deviation of the temperature during the previous five days) rise in milk temperature, compared to the average temperature over the five previous days. This produced an oestrus detection sensitivity of 74%, with a false rate of 8%, using milk progesterone as the reference technique. McArthur et al., (1992) have also examined the reliability of this method. They made measurements under experimental, controlled conditions, and under commercial conditions. The milk temperature of the two cows studied under controlled conditions rose by about 0.4° C on the day when behavioural oestrus was observed. On a commercial farm milk temperatures were measured for 18 cows which exhibited a total of 34 periods of oestrus. Setting a threshold of 0.3° C elevation in temperature over the average for the previous 5 days resulted in a oestrus detection sensitivity of 50% with an associated false positive rate of 81%. Increasing the threshold elevation to 0.6° C resulted in a reduction in the false rate to 65%, but a reduction in sensitivity to 32%. McArthur et al., compared their results with those from other studies which show wide ranges of sensitivity and false rates. They observe that, although some of these have reported better results, the only other study which was carried out, like theirs, in a commercial herd, produced similar results (sensitivity about 40% with 70% False Positives). McArthur et al. also made measurements of vaginal temperature which showed that the temperature increase associated with oestrus only lasts about 9 hours. They concluded that the shortness of this period, which can be less that the interval between milkings, contributes towards the uncertainty of oestrus detection. The conclusion from McArthur et al. is that the detection of oestrus based on twice daily measurement of milk temperature is not reliable and no further reports have been found.

Oestrus: Milk yield Schofield et al., (1991), Blanchard et al., (1987) have suggested that the continuous monitoring of milk yield can indicate oestrus. However, these patterns have not proved sufficiently specific and this approach has been abandoned except in robotic milking where it can be combined with other data (number and timing of visits).

Oestrus: Skin temperature Hurnik et al., (1985) investigated the possibility of using thermal infrared scanning of the body surfaces of a cow to detect temperature changes related to ovulation. The study was conducted using 27 cows, housed in tie stalls. A thermal imaging device, comprising a temperature sensing camera and a video display, was used to take images of the gluteal region of the cow, including the anal and vulval areas, the posterior zone of the udder attachment and the two posterior lobes of the udder. According to the specification of the imaging system, it was possible to resolve differences of 0.2° C. The images were analysed by measuring the total area of the cow that was enclosed in a given 37° C isotherm.

Using a criterion based on a given percentage increase in this area, the ovulation detection sensitivity was 80%, but this was associated with a false rate of 33%. The conclusion was that high frequency of false positives and false negatives meant that the technique was not suitable for routine oestrus detection. Since this paper was written there have been improvements in thermal imaging equipment, but the difficulty of eliminating or accounting for non-oestrus related temperature variation caused by factors such as environmental temperature variation, and moisture on the skin, remain. Considering also the inherent complexity of the system it seems very unlikely that the thermal imaging approach will yield a reliable, practical automatic oestrus detection system although Future Dairy (2012) investigated the technique.

Oestrus: Combined Measures Maatje et al., (1997) report a multivariate oestrus detection model which bases detection on a combination of activity (measured by pedometer), milk temperature, yield and feed intake. Using data from two experimental farms, which included over 500 cases of oestrus, they achieved a detection sensitivity of 87% with a specificity of 97%. This represented an improvement in sensitivity, with equal specificity compared to results obtained using activity alone. Mitchell et al., (1996) have carried out a preliminary investigation of the possibility of combining milk yield and data on milking order to detect oestrus.

These variables were chosen because of appropriateness for New Zealand dairy herds. The proposition was that, since at oestrus milk volumes sometimes fall and then rise at the next milking and the order in which cows present themselves to be milked changes, it may be possible to use a computer to recognise characteristic patterns in the data. Two different machine learning procedures (C4.5 and FOIL) were tried on a year’s data from a herd of 130 cows. The best result that was achieved was a sensitivity of 69%, with an associated false positive rate of 74%.

Fundamental questions regarding the nature of cow performance variations at and around oestrus remained to be answered. It was suggested that performance could be improved by including more monitored variables. Intuitively one might expect results to improve as the number of variables that are included increases. However, the increased complexity of the system particularly if it involves adding extra sensors, and the difficulty of fusing the various sets of data has to be overcome. This effectively requires an appropriate weighting to be given to the data from each source. For example, one set of data, say pedometer readings, might, if taken alone, indicate oestrus, whereas another set of data, say milk temperature, does not indicate oestrus. The combined system would be required to attach relative levels of confidence to the two indications to produce a decision. There are established techniques available for data fusion, but whether any of them would be suitable for this application has yet to be investigated.

Oestrus: Electronic Nose The natural method of oestrus and relies on a combination of senses, olfactory, visual and auditory. The existing methods reviewed in the previous sections all rely on visual signs but there now exists the potential to detect olfactory signals electronically. Kiddy et al., (1984) trained dogs to identify different bottles containing suitable body compounds of dairy cows and concluded that odours specific to oestrus were distributed throughout the body. Blazquez et al., (1988) showed that pheromonal odour secreted from the perineal glands near the vagina was the determinant of bull behaviour towards the cows. However, they could not identify the compounds responsible or whether it was an increased rate of secretion that was important. Klemm et al., (1987, 1995) steadily developed an understanding of the odours secreted in vaginal mucus. They identified acetaldehyde as a compound associated with oestrous although this not a specific marker for oestrous and the identity of the pheromone was not determined. Llobet et al., (1999) indicated that an array of tin oxide sensors could discriminate between oestrus and di-estrus from the odour of vaginal swabs but not from air samples taken from the surface of the cow. A subsequent study did not produce encouraging results (Mottram et al., 2000).

Oestrus: Progesterone Assay Most experiments to measure the efficacy of oestrus detection systems use progesterone assay as the standard calibration tool. A sample of milk is taken and analysed once per day firstly to identify that oestrus cycling has begun and then to identify the drop in progesterone which precedes ovulation by approximately 48 hours. Nebel (1988) reviewed the development of the immuno-sensing tests for progesterone. The principle of these tests is that an antibody to progesterone is attached to a plastic surface during manufacture. The farmer or veterinarian then adds milk. Progesterone in the milk then attaches to the antibody coated on the surface. A reagent is added and rinsed out and a colour change is observed to identify the amount of progesterone present in the milk. Analysis of progesterone levels in milk can not only be used to monitor the stage of the oestrous cycle but also to detect pregnancy and to identify early ovarian disorders (Macleod & Williams 1991, Darwash & Lamming 1996). Taking and analysing milk samples away from the milking parlour is labour intensive, strategies have been needed to minimise the number of samples needed to improve fertility. Experimental work during the eighties established a sampling protocol and showed that inseminating cows on the basis of progesterone profiles using laboratory analysis of samples could achieve significant improvements in fertility. The protocol used by McLeod et al., (1991) showed that 99% of 88 ovulations were correctly identified using on-farm progesterone kits compared with 78% in a control group monitored conventionally.

Milk samples were taken three times a week starting 25 d. post partum. Once an ovulation had been detected by a fall in progesterone concentration to below 4 ng/ml and a subsequent rise to more than 7 ng/ml sampling was suspended for 15 days. Sampling resumed on alternate days until a fall in progesterone indicated the onset of oestrus. The cows were then inseminated 48 hours following the fall in progesterone. Sampling continued so as to determine whether oestrus had been correctly identified or whether the cow had conceived.

The protocol used by McLeod & Williams (1991) was also used to detect ovarian malfunction. A study of over 500 cows in a controlled trial suggested that alternate day progesterone profiles were a better method of analysing ovarian malfunction than rectal palpation. This study confirmed other results and showed that there was little incidence of ovarian dysfunction and that the principal cause of extended calving intervals was a failure to detect oestrus. Cows were diagnosed as anoestrus – when progesterone was below 4 ng/ml for 30 d post partum. Ovulation was deemed to have occurred if progesterone was below 4 ng/ml followed by 5 days of progesterone rising above 4 ng/ml, with at least one sample greater than 7 ng/ml. Normal cycling was detected by an increase of progesterone which remained high for more than 5 days and less than 18 days. Insemination was assumed to be correctly timed if progesterone increased in the 2-6 days following. Conception was identified if ovulation and correctly timed insemination coincided and that progesterone remained above 4 ng/ml for more than 20 days. If progesterone was greater than 4 ng/ml for more than 30 days after conception then pregnancy was assumed to be established. The progesterone diagnosis proved to be more accurate than the rectal palpation and indicated that 36.5 % of clinical diagnoses were incorrect. At 42 days post partum 92 % of cows were cycling normally which reinforces the vie w that identification of the oestrus cycle is the problem. Williams & Esslemont (1993) reported ovulation detection rates of 98% using progesterone assay.

Oestrus: Biosensors for progesterone The term biosensor is loosely used to describe a number of different devices, the objectives of which are to identify specific complex biological molecules by a change in electrical or opto-electronic signal. A biosensor system consists of a sensor, a system to interrogate it and a microcomputer to convert the electrical signal into a format to be displayed to an operator or computer program. The sensing methods are usually based on a monoclonal antibody which is specific to the compound being detected. A review of systems suitable for agricultural applications was conducted by Velasco-Garcia & Mottram (2003).

Koelsch et al., (1994) reported a method of detecting progesterone with a quartz crystal micro-balance device. A crystal with a known natural frequency of oscillation was coated with an antibody which bound to progesterone when exposed to it (QCM or Quartz Crystal Microbalance). Since the mass of the progesterone binding to the antibody would change the frequency at which the crystal oscillated, the degree of binding could be determined by measuring a change in oscillation. The device was dipped into a solution of progesterone and then exposed to air and the change in oscillation monitored. However, there have been no further reports of work on this device. Claycomb et al., (1995) showed that realistic levels of progesterone would only produce a change in mass of 0.4% on a QCM very close to the noise level of the microelectronics they proposed an alternative method by automating an ELISA test for on-line measurement of progesterone in bovine milk and detection of oestrus. The biosensor used an enzyme immunoassay format for molecular recognition, which was developed to run in approximately eight minutes. The sensor was designed to operate on-line in a dairy parlour using microinjection pumps and valves for fluid transport, fibre optics and photodiodes for light measurement, and a control computer for sequencing. Calibration showed a dynamic response between 0.1 and 5ng/ml progesterone in milk.

Pemberton et al., (1998) reported a device was based on a disposable screen- printed amperometric progesterone biosensor, operated in a competitive immunoassay. The biosensor comprised a monoclonal anti-progesterone antibody (mAb) immobilised on the working area of a screen-printed carbon electrode (SPCE). It relied upon a reduction in the binding of alkaline phosphatase-labelled progesterone in the presence of endogenous milk progesterone. The enzyme substrate was naphthyl phosphate and the 1-naphthol generated in the enzymatic reaction was electrochemically oxidised, producing a signal inversely proportional to the concentration of unlabelled progesterone in milk. This SPCE-based immunosensor for progesterone was incorporated into a thin-layer flow cell offering advantages such as on-line analysis and improved fluid handling with the possibility of future automation (Pemberton et al., 2001). Between 1999 and 2005 built an online progesterone monitoring system. (Velasco-Garcia & Mottram 2001). The system was based on applying a milk sample to an electrochemical biosensor and reading the electrical response.

A commercial venture between Foss Electric and de Laval developed the Herd Navigator system which has been on sale since 2008. It combines automated sampling and five sensing systems including progesterone in milk. Scientific reports are limited as yet. Blom & Ridder (2010) reported that heat detection rates of 95%- 97% were being achieved on three farms in Denmark with days open reduced by 20 days in the first year of operation. Pregnancy rates also increased significantly in a very short time: up to 50%. Herd Navigator automatically measures the level of progesterone in milk, software indicates insemination time, lists animals for final pregnancy confirmation, indicates early abortion and lists the cows with risk for cysts and prolonged anoestrus. Vreeburg (2010) reported heat detection rates of two farms in the Netherlands using Herd Navigator was 94 & 99%. Pregnancy rates of 42 & 46% were reported. Mazeris (2010) claim that a number of Herd Navigator farmers have stopped performing manual pregnancy tests and saved between €250 and €350 per cow per year. Herd Navigator consists of a large analyser boxed with controlled environment and complex plumbing system to bring the milk samples to the analyser from the short milk tubes. It is designed to fit into parlours with up to 8 milking points when newly installed and would be difficult to retrofit into existing parlours, particularly large systems that are working continuously.