Report to the Stapledon Memorial Trust
The effect of dietary factors on nitrogen use efficiency
and the relationship with the efficiency of feed
utilisation within dairy production systems.
Dr David Barber
Agri-science Queensland, Department of Agriculture, Fisheries and Forestry,
University of Queensland Gatton Campus, John Mahon Building 8105, Lawes,
Queensland, 4343, Australia
[email protected]
UK Contacts:
Professor Chris Reynolds
School of Agriculture, Policy and Development, University of Reading, PO Box
237, Earley Gate, Reading, RG6 6AR, UK
Email: [email protected]
Fellowship period: 30 September to 29 November 2013
Fellowship Objectives
The objective of the 2 month fellowship was to investigate the dietary factors
that determine nitrogen use efficiency within a range of dietary crude protein
levels to improve overall efficiency of both nitrogen and feed utilization.
Background
In dairy cows, nitrogen use efficiency and feed efficiency are strongly influenced
by diet composition, with nitrogen use efficiency declining as the crude protein
(CP) content of the diet increases (Castillo et al. 2000; Broderick, 2003;
Ipharraguerre and Clark, 2005; Huhtanen et al., 2008; Huhtanen and Hristov,
2009). The effect of crude protein content on feed efficiency and the relationship
between milk nitrogen efficiency (MNE) and feed efficiency (FE) is not well
defined in the literature. Research shows that increased dry matter intake (DMI)
and improved nutrition in general has a positive impact on FE (Beever and
Doyle, 2007), however Ipharraguerre and Clark (2005) reported no significant
effect of increasing dietary CP content on feed efficiency.
Anecdotal and preliminary scientific evidence (Wales et al.; unpublished data) in
Australia shows a significant response in milk yield and milk protein yield and
concentration when lactating cows are fed a protein supplement (true protein
source) while grazing pastures high in CP (>…%). Diet formulation using the
chemical analysis of feeds in nutrition models show that forage diets high in CP,
particularly ryegrass based, are often in excess of requirements of both CP and
RDP according to the national feeding standards and nutritional targets for
lactating dairy cows. These models indicate that there is no need to supplement
with additional CP sources, however there are a number of examples seen on
farm where an increase in milk yield of up to 5 L/cow/day is seen in addition to
an increase in milk protein % of 0.5 to 1.0% units with the supplementation of 1
to 1.5 kg of protein meal (soybean meal, canola meal), therefore increasing
feed efficiency. Other groups internationally are investigating the alternative
Stapledon Memorial Trust Report - Dr David Barber
January 2014
1
strategy, where CP levels in the diet are reduced to improve the nitrogen use
efficiency within the high producing dairy cow; however what is the resultant
effect on feed efficiency. The underpinning question is, can the CP content of
the diet be decreased to reduce N losses and improve MNE, whilst also
improving FE and profitability?
Research Approach
Historical data was compiled from applied research studies conducted at the
Centre for Dairy Research (CEDAR) at the University of Reading (England, UK)
between 1995 and 2009. Data from 14 long-term continuous (randomised block
design) feeding study reports was collated into an applied study database using
electronic and printed sources and included the following data:
• Trial data – cow number, block number, treatment, week of lactation and
week of study.
• Animal data – milk (kg/d), fat, protein and lactose yield (g/d); fat protein
and lactose concentration (%); fat and energy corrected milk yield (kg/d);
liveweight (kg) and body condition score (BCS).
• Diet data – dry matter (DM) intake (kg DM/d), metabolisable energy (ME;
MJ/kg DM), crude protein (CP), neutral detergent fibre (NDF), acid
detergent fibre (ADF), starch, sugar and fat concentration (% DM).
• Efficiency data – milk nitrogen efficiency (%), gross feed efficiency (kg
milk/ kg DM) and energy corrected feed efficiency (kg ECM/kg DM).
The trial, animal and efficiency data was reported on a weekly basis and
compiled in the database on an individual cow basis with a total of 765
individual cows used (Table 1). Diet composition information within the reports
was variable and was generally reported on an average basis within treatments
and individual feed ingredients.
With the lack of individual cow dietary data to adequately assess the effect of
dietary components on MNE and FE, two additional datasets were compiled
using cow mean production data and treatment mean data averaged from week
8 of lactation until the end of each study. The average number of weeks used to
calculate the mean cow and treatment data was 13 weeks and ranged from 8 to
25 weeks, therefore the length of the studies ranged from 15 to 32 weeks of
lactation. Individual cow and treatment mean data was used to run individual
simulations within the Diet Check (v7.2, Hampshire, England) feeding software
to generate the diet composition and nutrient intakes of individual cows and
treatment means of each study. Additional diet composition information
captured from Diet Check included predicted DM intake, forage intake,
estimated milk from ME, predicted microbial protein synthesised from the
nitrogen and energy supplied in the diet, predicted metabolisable protein (MP)
supply from the nitrogen and energy supplied in the diet and an estimate of
rumen stability (rumen stability value; RSV). A total of 751 individual cows and
52 treatment means averaged from week 8 of lactation until the end of each
study were individually simulated within Diet Check. Fourteen cows were
removed from the individual cow and treatment mean datasets due to lost or
erroneous data as reported in individual study reports.
Stapledon Memorial Trust Report - Dr David Barber
January 2014
2
Table 1. Summary of information on the report data compiled in the CEDAR
applied study database.
Study No.
Study
Date
Experimental
Treatment Type
Design
33
1995
Randomised
Block
56
1996
Continuous
57
1996
62
1995
83
1997
84
1997
100
1997
104
1998
Continuous
136
1998
Continuous
174
2001
Continuous
180 (Study 1)
1997
Continuous
180 (Study 2/3)
1998
Continuous
268
2008
Continuous
275
2009
Continuous
Randomised
Block
Randomised
Block
Randomised
Block
Randomised
Block
Randomised
Block
Number of
treatments
Experiment
length (weeks)
No of
Cows
Feed Additive
4
21
35
Concentrate
Source
4
15
19
Feed Additive
4
22
42
Feed Additive
2
32
94
Forage Source
4
14
40
Forage Source
5
19
51
Feed Additive
4
16
116
5
15
50
4
15
40
3
20
47
3
20
31
2
20
72
Feed Additive
4
16
60
Feed Additive
2
15
68
Concentrate
Source
Concentrate
Source
Concentrate
Source
Concentrate
Source
Concentrate
Source
Regression analysis was conducted on the individual cow and treatment mean
datasets to investigate the individual effects of dietary components on MNE and
FE (gross FE and energy corrected FE). Regression tree analysis was also
conducted using Genstat (v14; VSN International Ltd, UK) to identify the main
dietary factors affecting MNE and FE and to assess the interactions between
dietary variables on MNE and FE. A comprehensive meta-analysis is currently
being conducted to account for the effect of study, time and increased milk
production at CEDAR over time into account.
Research Outcomes
Regression Analysis
Regression analysis of the cow and treatment mean data resulted in the
following outcomes:
• MNE and FE were positively correlated with approximately 50% of the
variation in feed efficiency accounted for by milk N efficiency (Figure 1).
• MNE declined as dietary CP content increased (Figure 2); however there
was no major effect of CP content or intake on FE. The CP:ME ratio in
the diet slightly improved the relationship between CP and MNE.
• FE decreased as acid detergent fibre (ADF) intake increased, however
there was some divergence in the data evident which may be due to
missing ADF values for some studies (Figure 3).
• Milk yield per cow at CEDAR increased over the 15 years of the studies
used in this analysis (Figure 4), which is potentially having a large effect
on MNE and FE within the dataset and will need to be taken into account
with the meta-analysis.
Stapledon Memorial Trust Report - Dr David Barber
January 2014
3
Energy Corrected Feed Efficiency (kg ECM/kg
DM)
2.50
2.00
1.50
1.00
0.50
0.00
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0
Milk Nitrogen Efficiency (%)
Figure 1. The correlation between milk nitrogen efficiency (%, MNE) and
energy corrected feed efficiency (kg ECM/kg DM) for individual cow (green dots
(●) and black linear trendline (▬); R2 = 0.5107) and treatment (blue square (■)
and red linear trendline (▬); R2 = 0.5334) mean data.
45.0
40.0
Milk Nitrogen Efficiency (%)
35.0
30.0
25.0
20.0
15.0
10.0
5.0
0.0
150
160
170
180
190
200
210
220
230
Crude Protein content (g/kg)
Figure 2. The effect of diet crude protein content (g/kg; CP) on milk nitrogen
efficiency (%, MNE) for individual cow (green dots (●) and black linear trendline
(▬); R2 = 0.2155) and treatment (blue square (■) and red linear trendline (▬);
R2 = 0.1481) mean data.
Stapledon Memorial Trust Report - Dr David Barber
January 2014
4
1.90
1.80
Feed Efficiency (L/kg DM)
1.70
1.60
1.50
1.40
1.30
1.20
1.10
1.00
2000
2500
3000
3500
4000
4500
5000
5500
6000
ADF Intake (g/d)
Figure 3. The effect of diet acid detergent fibre intake (g/d) on energy corrected
feed efficiency (kg ECM/kg DM, ECFE) for treatment (blue square (■) and red
linear trendline (▬); R2 = 0.1632) mean data.
45.0
Milk Yield (L/cow/day)
40.0
35.0
30.0
25.0
20.0
15.0
1994
1996
1998
2000
2002
2004
2006
2008
2010
Year
Figure 4. The mean milk yield per cow (kg/d) over 15 years at CEDAR using
mean data for each year (blue circle (●) and black polynomial trendline (▬); R2
= 0.8593).
Stapledon Memorial Trust Report - Dr David Barber
January 2014
5
Regression Tree Analysis
Regression tree analysis is a data mining procedure that allows all factors to be
considered at the one time by including all relevant x variables (factors) and a y
variate (response variate), therefore accounting for all the interactions that
occur between the factors. In the development of a regression tree, nodes of
the tree (branch point) split to form two branches with a new node formed at the
end of a new branch. The point at which a node splits (splitting point) is an
average value of the x variable used in the calculation of that node. Once a tree
has finished growing and no more branches can be formed at a node, this node
becomes a terminal node and a value is assigned, which is a predicted value of
all the y variables used in branches that lead to the terminal node. Trees are
then pruned to a desired size or accuracy. The regression tree analysis starts
with the most influential factor and proceeds until all variation is accounted for.
The regression trees developed in this report have been pruned to a maximum
of twenty (20) terminal nodes. Regression trees were pruned to 20 terminal
nodes in this study as the amount of variation accounted for by trees with
greater than 20 terminal nodes was minimal within this dataset, seen as an
increase in the value of R2. Trees were also pruned at 20 terminal nodes to
reduce the overall size of the tree for reporting purposes. The coefficient of
determination (R2) displayed on each regression tree figure is a measure of the
accuracy of the tree displayed and of the amount of variation accounted for by
the variables used. The coefficient of determination (R2) is defined as the
proportion of the variation in the dependant variable that is explained by the
regression tree and was calculated according to the equation;
R2= 100*(1 – (RSSQ/TSSQ));
where RSSQ = Residual Sum of Squares and
TSSQ = Total Sum of Squares (De'ath and Fabricius 2000).
To test the accuracy of the regression trees as predictive models, other
datasets with known x and y values can be analysed and a prediction of the y
values is given, then a comparison between predicted and actual y values can
be made. Once validated, these trees can be used to predict a y value (e.g.
MNE or FE) from a known set of x values, thus providing a model for the
prediction of milk nitrogen and feed efficiency on-farm. This validation process
was not undertaken as part of this fellowship.
The following factors were included in the regression tree model to test their
effect on MNE and ECFE:
• DMI – dry matter intake (kg DM/d);
• ForDMI – forage dry matter intake (kg DM/d);
• MPN - predicted metabolisable protein supplied from dietary nitrogen (g/d);
• MPE - predicted metabolisable protein supplied from dietary energy (g/d);
• MCPN - predicted microbial crude protein supplied from the dietary nitrogen
(g/d);
• MCPE - predicted microbial crude protein supplied from the dietary energy
(g/d);
• DUP – dietary undegraded protein (g/d);
• RumStch - rumen starch supplied from the diet (g/d)
• BypStch - bypass starch supplied from the diet (g/d);
• ME – diet metabolisable energy content (MJ/d)
• CP – diet crude protein content (g/kg DM)
Stapledon Memorial Trust Report - Dr David Barber
January 2014
6
•
•
•
•
•
•
•
•
•
•
•
•
Starch – diet starch content (g/kg DM)
Sugar – diet sugar content (g/kg DM)
NDF – diet neutral detergent fibre content (g/kg DM)
ADF – diet acid detergent fibre content (g/kg DM)
MSilDMI – diet maize silage DM intake (kg DM/d)
MEI - metabolisable energy intake (MJ/d)
CPI – crude protein intake (g/d)
StchI - starch intake (g/d)
SgrI - sugar intake (g/d)
NDFI – neutral detergent fibre intake (g/d)
ADFI – acid detergent fibre intake (g/d)
Nint - nitrogen intake (g/d)
The main outcomes of the regression tree analysis resulted in the following
outcomes:
• The main dietary factors affecting MNE were diet maize silage DM intake
(MSilDMI, kg DM/d), diet ME content (ME, MJ/kg DM), diet starch
content (Starch, g/kg DM), diet crude protein content (CP, g/kg DM),
bypass starch supplied from the diet (BypStch, g/d), predicted
metabolisable protein supplied from dietary energy (MPE, g/d), forage
DM intake (ForDMI, kg DM/d), rumen starch supplied from the diet
(RumStch, g/d), diet sugar intake (SgrI, g/d) and diet acid detergent fibre
intake (ADFI, g/d). These factors accounted for 74.4% of the variation
seen in MNE (Figure 6).
• These results identify an interaction between starch and ME intake from
corn silage and CP content of the diet, which suggests that the
protein:energy ratio within the diet is important for improving MNE.
• The main dietary factors affecting ECFE were diet ME content (ME,
MJ/kg DM), diet starch content (Starch, g/kg DM), diet starch intake
(StchI, g/d), predicted microbial crude protein supplied from the dietary
nitrogen (MCPN, g/d), diet maize silage DM intake (MSilDMI, kg DM/d),
diet crude protein content (CP, g/kg DM), predicted metabolisable protein
supplied from dietary energy (MPE, g/d), metabolisable energy intake
(MEI, MJ/d), bypass starch supplied from the diet (BypStch, g/d), rumen
starch supplied from the diet (RumStch, g/d) and predicted microbial
crude protein supplied from the dietary energy (MCPE, g/d). These
factors accounted for 43.6% of the variation seen in ECFE (Figure 7).
• Interactions between ME, Starch and Protein supply were also identified
for ECFE.
• Balancing the supply of protein and energy ratio in the diet will have
potential benefits for improving milk nitrogen efficiency and feed
efficiency.
• Regression trees were pruned to 20 terminal nodes for ease of
interpreting and graphical presentation, however if the whole tree was
used as the model they would account for 86.2 (99 terminal nodes) and
90.8% (219 terminal nodes) of the variation in MNE and ECFE
respectively.
Stapledon Memorial Trust Report - Dr David Barber
January 2014
7
MSilDMI<1.84
(<1.84)
(>1.84)
ME<11.96
Starch<153.0
ME<12.03
24.2
CP<183.0
17.9
CP<177.0
26.8
23.6
16.4
23.5
BypStch<265.4
RumStch
<1945
21.2
30.3
MPE<2018
DMI<13.96
34. 1
ForDMI<9.95
RumStch<3989
MSilDMI
<4.7
BypStch
<837.7
28.4
SgrI<1477
ME<11.97
MPE<2130
MSilDMI<7.68
26.6
ADFI<2953
30.5
29.1
27.9
25.6
27.7
25.4
26.4
34.1
28.0
Figure 6. Regression tree analysis of the dietary factors affecting the milk
nitrogen efficiency (MNE) from the individual cow mean data averaged from
week 8 of lactation until the end of each study. Values expressed in blue text
(terminal nodes) are average MNE (%) and factors identified in the regression
tree analysis include diet maize silage DM intake (MSilDMI, kg DM/d), diet ME
content (ME, MJ/d), diet starch content (Starch, g/kg DM), diet crude protein
content (CP, g/kg DM), bypass starch supplied from the diet (BypStch, g/d),
predicted metabolisable protein supplied from dietary energy (MPE, g/d), forage
DM intake (ForDMI, kg DM/d), rumen starch supplied from the diet (RumStch,
g/d), diet sugar intake (SgrI, g/d) and diet acid detergent fibre intake (ADFI, g/d).
R2 (coefficient of determination) = 73.4%. Trees branch to the left for < variable
and to the right for > variable. Trees were pruned to 20 terminal nodes for ease
of interpretation and presenting graphically, hence the reported R2 is lower than
the potential using the whole tree as the model.
Stapledon Memorial Trust Report - Dr David Barber
January 2014
8
ME<12.25
(<12.25)
(>12.25)
Starch
<155.0
StchI<4948
MCPN
<2351
CP
<216
MSilDMI
<3.04
Starch<163.0
MPE
<2280
MSilDMI<2.80
1.47
MEI
<275
1.18
1.79
MPE<2127
1.47
BypStch
<310.9
MSilDMI
<7.42
1.15
1.74
1.56
1.32
MCPE<1735
1.78
1.67
1.79
RumStch<2899
MCPN
<2001
1.66
1.60
1.49
1.88
ME<12.06
1.14
RumStch
<5444
MCPE
<2066
MCPE
<2023
1.58
1.69
1.44
1.61
1.40
Figure 7. Regression tree analysis of the dietary factors affecting the energy
corrected feed efficiency (ECFE) from the individual cow mean data averaged
from week 8 of lactation until the end of each study. Values expressed in blue
text (terminal nodes) are average MNE (%) and factors identified in the
regression tree analysis include diet ME content (ME, MJ/d), diet starch content
(Starch, g/kg DM), diet starch intake (StchI, g/d), predicted microbial crude
protein supplied from the dietary nitrogen (MCPN, g/d), diet maize silage DM
intake (MSilDMI, kg DM/d), diet crude protein content (CP, g/kg DM), predicted
metabolisable protein supplied from dietary energy (MPE, g/d), metabolisable
energy intake (MEI, MJ/d), bypass starch supplied from the diet (BypStch, g/d),
rumen starch supplied from the diet (RumStch, g/d) and predicted microbial
crude protein supplied from the dietary energy (MCPE, g/d). R2 (coefficient of
determination) = 43.6%. Trees branch to the left for < variable and to the right
for > variable. Trees were pruned to 20 terminal nodes for ease of interpretation
and presenting graphically, hence the reported R2 is lower than the potential
using the whole tree as the model.
Stapledon Memorial Trust Report - Dr David Barber
January 2014
9
Acknowledgements
I would like to thank the Stapledon Memorial Trust fund and Dairy Australia for
their financial support with airfares and operating expenses respectively. I
sincerely thank Professor Chris Reynolds, Dr Les Crompton and Dr Jonathon
Mills at the University of Reading for their time and scientific input, which made
this trip a significant part of my professional development and an extremely
enjoyable time. I look forward to continued collaboration into the future between
DAFFQ and the University of Reading. My sincere thanks to Dr Jon Moorby, Dr
Conrad Ferris and Dr Trish Lewis for hosting my visit to each of their research
facilities.
References
Beever, D.E. and Doyle, P. T. (2007). Feed conversion efficiency as a key
determinant of dairy herd performance: a review. Australian Journal of
Experimental Agriculture 47, 645-657.
Broderick, G.A. (2003). Effects of varying dietary protein and energy levels on
the production of lactating dairy cows. Journal of Dairy Science. 86 (4), 13701381.
Castillo, A.R., Kebreab, E., Beever, D.E. and France, J. (2000). A review of the
efficiency of nitrogen utilisation in lactating dairy cows and its relationship with
environmental pollution. Journal of Animal and Feed Sciences. 9, 1-32.
De'ath, G. and Fabricius, K. E. (2000) Classification and regression trees: A
powerful yet simple technique for ecological data analysis. Ecology. 81, 31783192.
Huhtanen, P., Nousiainen, J.I., Rinne, M., Kytola, K. And Khalili, H. (2008).
Utilisation and partition of dietary nitrogen in dairy cows fed grass silage-based
diets. Journal of Dairy Science. 91, 3589-3599.
Huhtanen, P. and Hristov, A.N. (2009). A meta-analysis of the effects of dietary
protein concentration and degradability on milk protein yield and milk N
efficiency in dairy cows. Journal of Dairy Science. 92, 3222-3232.
Ipharraguerre I. R. and Clark, J.H. (2005). Varying protein and starch in the diet
of dairy cows. II. Effects on performance and nitrogen utilisation for milk
production. Journal of Dairy Science. 88, 2556-2570.
Stapledon Memorial Trust Report - Dr David Barber
January 2014
10