European Journal of Clinical Nutrition (1999) 53, 13±21
ß 1999 Stockton Press. All rights reserved 0954±3007/99 $12.00
http://www.stockton-press.co.uk/ejcn
Description and evaluation of an experimental model to examine
changes in selection between high-protein, high-carbohydrate
and high-fat foods in humans
RJ Stubbs1*, LM O'Reilly1, AM Johnstone1, CLS Harrison1, H Clark1, MF Franklin1 and CA Reid2
1
The Rowett Research Institute, Bucksburn, Aberdeen; and 2Biomathematics and Statistics, Scotland, The Rowett Research Institute,
Bucksburn, Aberdeen, UK
Objective: To develop and test an experimental model designed to detect changes in selection between foods
individually enriched in protein, carbohydrate and fat in human subjects.
Design: Randomised counterbalanced (Latin square) design.
Setting: The metabolic suite at the Rowett Research Institute's Human Nutrition Unit.
Subjects: 16 normal-weight men (mean BMI ˆ 23.5).
Interventions: Subjects were each studied 4 times in a 2-day protocol. On day 1 subjects received a ®xed
maintenance diet; on day 2 they received a mandatory intake as breakfast (08.30) plus a drink at 10.30. This
comprised 80% of resting energy requirements as high-protein (HP), high-carbohydrate (HC) or high-fat (HF)
foods (60% of energy in each case) or an equal mixture (M) of macronutrients, 33% by energy. All mandatory
treatments contained the same energy content and density. From 12.30 onwards, subjects had ad libitum access to
a counter-balanced selection of three groups of familiar foods (10 HP, 10 HC and 10 HF; 30 foods in total). Most
energy in each food was derived from one macronutrient ( 60%), the remainder being equally split between the
other two macronutrients.
Results: Subjects were signi®cantly less hungry before lunch on the HP and M (33% protein) treatments
(F3;44 ˆ 7:35; P < 0.001). At lunch, they ate more energy after the HF treatment than after any of the other
treatment (F1;38 ˆ 9:00; P ˆ 0.005). This was largely in the form of fat and protein, and to a lesser extent
carbohydrate. Subsequent energy intake (EI) were lower on the HF treatment, largely through selection of less fat
in the afternoon (F1;42 ˆ 6:90; P ˆ 0.012). Daily EIs were similar across treatments.
Conclusion: This design appears sensitive meal-to-meal to changes in both nutrient and EIs.
Sponsorship: This work was supported by the Scottish Of®ce, Agriculture, Environment and Fisheries
Department.
Descriptors: appetite; carbohydrate; diet composition; fat; food intake; humans; hunger; macronutrient
selection; protein
Introduction
Understanding the determinants of quantitative and qualitative feeding in humans is of relevance to basic knowledge, public health policy, consumer health and well-being
and strategies of nutritional support in the clinical setting.
Here quantitative intake is de®ned as the amount of food
and energy ingested by subjects on a daily basis, and
qualitative intake is de®ned as the types of foods and
hence nutrient composition of the diet selected on a daily
basis. This paper considers macronutrient selection in
humans (viz. how ingesting foods of a certain macronutrient composition may in¯uence subsequent voluntary
selection of foods with differing compositions) and how
to detect it experimentally.
A number of models have been employed to assess
human feeding behaviour. These range from self-reported
dietary intakes of free-living subjects (e.g. DeCastro, 1987)
*Correspondence: RJ Stubbs, The Rowett Research Institute, Greenburn
Road, Bucksburn, Aberdeen AB21 9SB, UK. Part of this work was
presented as an abstract at the Eighth European Congress on Obesity,
Trinity College, Dublin, 18 ± 21 June 1997.
Received 22 August 1997; revised 6 July 1998; accepted 4 August 1988
to a variety of laboratory models. Methods of directly
measuring food, energy and nutrient (FEN) intake in the
laboratory include continuous weighing of foods (Hill et al,
1995), use of food dispensing machines (Wurtman &
Wurtman, 1982; Silverstone & Goodall, 1986; Rising et
al, 1992), liquid diets and laboratory weighing of food prior
to and after consumption by volunteers (Cotton et al, 1994;
Poppitt et al, 1995; Stubbs et al, 1995). The nature,
composition and degree of variety of foods made available
to subjects often vary considerably from laboratory to
laboratory. The reason why a certain selection of foods is
offered to human subjects is not often stated, beyond the
general notion that foods comprise an ad libitum diet from
which subjects can select, usually in response to some prior
manipulation. The relative advantages and disadvantages of
these approaches for measuring the amount of food and
energy eaten by subjects in response to various experimental manipulations has been reviewed elsewhere (Hill et
al, 1995; Stubbs et al, 1998).
A number of models have been developed that allow
human subjects to select between different foods, such as a
selection of high- and low-fat foods (King, 1995), a variety
of familiar foods given largely as snack items (Foltin et al,
1988, 1990) or a large buffet (de Graaf & Hulshof, 1995).
Model for detecting food selection changes
RJ Stubbs et al
14
Figure 1 Experimental design to detect changes in amount and composition of foods selected. (a) The model was based on rodent studies that provide a
selection of pure macronutrients to animals. (b) The present design for humans used common familiar foods.
While some of these experimental models allow changes in
food selection to be monitored, none of them appears to
have been speci®cally designed to detect changes in the
selection between the three macronutrients protein, carbohydrate and fat.
Leibowitz has developed a model for studying the effects
of a variety of compounds on macronutrient choice in
rodents (Leibowitz, 1992). The animals are given access
to sources of pure macronutrients as depicted in Figure 1(a).
Representing the macronutrient composition of the diet
available in terms of the percentage of protein, carbohydrate
or fat delineates an equilateral triangle that de®nes the
nutrient space in which the subject can select nutrients.
The animal with ad libitum access to three pure macronutrients is able to move anywhere within the space (i.e. to
select any possible ratio of protein : fat : carbohydrate). This
model is particularly powerful in demonstrating the effects
of a variety of compounds thought to be important in
in¯uencing macronutrient selection in rodents (Leibowitz,
1992). However, the experiments that are conducted use
rodents familiarised with such unusual laboratory diets.
Clearly such a design would require substantial modi®cation for use in humans.
In order to facilitate and thereby detect macronutrient
selection in humans, it is necessary to provide a counterbalanced selection of foods individually enriched in each of
these macronutrients. There is therefore a need to develop a
model for human studies that can be statistically and
experimentally demonstrated to be capable of detecting
changes in both food and energy intake on the one hand
and selection of protein-, carbohydrate-, and fat-rich foods
respectively, on the other. This creates a logistical problem
since, in the laboratory, it is impossible to simulate the
degree of food choice that is often available in real life.
Indeed, a number of studies have speci®cally precluded
nutrient selection as a component of subject-response (e.g.
Stubbs et al, 1995).
The purpose of this work was therefore (i) to derive an
experimental design based on existing animal selection
models that uses a counterbalanced arrangement of macronutrient-rich foods that are familiar and acceptable to
human subjects, (ii) to ensure the model's design allows
subjects to alter quantitative and qualitative feeding behaviour and (iii) to conduct a study examining whether this
experimental model was suf®ciently sensitive to detect
changes in quantitative and qualitative intake in response
to prior nutrient manipulations of the diet.
Materials and Methods
Derivation of the dietary design for the study of
macronutrient selection in humans
In order to detect possible changes in selection between
protein-, carbohydrate- and fat-rich foods it was necessary
to provide subjects with access to a counterbalanced range
of macronutrient sources as in the Leibowitz model for
rodents (Leibowitz, 1992). This design is depicted in Figure
1(b). Since pure macronutrient sources are likely to prove
aversive to human subjects, it was decided to provide a
counterbalanced range of foods, each rich in one of the
macronutrientsÐprotein, carbohydrate or fat. Since the
design of the current model was concerned with selection
between protein-rich, carbohydrate-rich and fat-rich foods,
it was possible to adapt the triangular `nutrient space' model
used by Leibowitz (1992). The experimental design
required that throughout the day subjects had ad libitum
access to 10 protein-rich, 10 carbohydrate-rich and 10 fatrich foods, giving a total of 30 foods during the study day.
These foods are described in Table 1. Each of the 30 foods
was rich ( 60% by energy) in an individual macronutrient
and the remaining energy in the food was split evenly (as far
as possible) between the other two macronutrients. This was
necessary to avoid covariance in the selection of nutrients.
For instance if a subject preferentially selected high-protein
foods, but these tended to be higher in fat than carbohydrate, then fat would covary with protein, giving the
erroneous impression that subjects were selecting both
protein and fat.
Model for detecting food selection changes
RJ Stubbs et al
15
Table 1
High-protein, high-carbohydrate and high-fat foods available ad libitum to subjects
High protein
High carbohydrate
High fat
Salad
Shrimp sandwich
Salmon sandwich
Tuna sandwich
Corned beef sandwich
Turkey breast
Chicken breast
Fromage frais
Crab sticks
Crab pate cracker
Cottage cheese cracker
Lettuce and grape sandwich
Banana sandwich
Jam sandwich
Honey sandwich
Savoury ricea
Pot noodleb
Pizza topping crackerc
Raisins
Swiss rolld
Twigletse
Tuna and mayonnaise sandwich
Egg and cheese sandwich
Cheese sandwich
Salami sandwich
Pork pie
Sausage rolls
Philadelphia cheese and crispbreadf
Pistachio nuts
Nut and rasin mix
Pro®teroles and creamg
Celery
Cucumber
Lettuce
Sweetcorn
Cress
Tea, coffee and water were available ad libitum. The values for the composition of each food used and (where appropriate) recipes for foods can be
obtained by contacting the corresponding author. a Savoury Rice, Batchelors Foods, Brooke House, Crawley RH10 2RQ, UK; b Pot Noodle, Golden Wonder
Ltd, Knorr-Naehrmittel AG, CPC (UK) Ltd, Surrey; c Pizza Topping, Tesco Stores Ltd, Cheshunt, Herts EN8 9SL, UK; d Swiss Roll, Tesco Stores Ltd;
e
Twiglets, The Jacob's Bakery Ltd, Liverpool L9 7JX, UK; f Philadelphia, Kraft Jacobs Suchard, Cheltenham GL50 3AE, UK; g Pro®teroles and cream,
Tesco Stores Ltd.
Changes in diet selection in animals generally require a
period of conditioning whereby the animal learns to associate the sensory characteristics of a food with the postingestive consequences of having eaten it (Forbes, 1995).
Without such learning (i.e. in naive animals) selective
responses are far less reliable (Forbes, 1995). Indeed, it
has been argued by Blundell and Stubbs (1997) that many
humans studies that use covertly manipulated diets are
likely to produce less precise compensatory responses
because it is more dif®cult to associate the post-ingestive
consequences of having eaten foods of a certain composition with their sensory characteristics. For this reason, the
present design used common, familiar foods.
The purpose of providing subjects with a counterbalanced selection of foods rich in particular macronutrients
is based on the statistical assumption thatÐall things being
equalÐsubjects will select among the available foods with
an equal probability of selecting HP, HC or HF foods.
However, not all in¯uences on feeding are equal and
subjects are unlikely to select randomly between these
foods. Subjects tend to eat what they like and avoid what
they do not like. Furthermore, individuals vary considerably in their preferences for certain foods. It would not
therefore be possible to tailor all food items to the preference pro®les of each individual subject, since preparing
such a potentially vast variety of foods would prove
impractical in the laboratory. Ten common food items of
each macronutrient category were therefore chosen to
provide suffcient variety for subjects to be able to select
foods from each food category (HP, HC or HF) without
their choice being heavily constrained by the avoidance of
most foods within any one category simply because they
did not like those foods. It was also decided to have
subjects rate how much they actually liked each food,
both prior to its ingestion and subsequent to its ingestion
in a tasting session conducted before the actual experiment
began. Thus, while it was impossible to provide a selection
of foods that were liked to the same extent by each subject,
it was possible to quantify subjects' own expression of their
perceived pleasantness of the foods used in the study.
In addition to the macronutrient-rich foods, a number of
salad vegetables and non-caloric drinks were made available to the subjects (Table 1). Figure 2 illustrates the
composition of the foods used in the dietary design.
Figure 2 Energy density and macronutrient composition of the HP, HC
and HF foods used in the study.
Within each category, foods are ranked by energy density.
HP foods were the least energy dense and the HF foods the
most energy dense, HC foods being intermediate. There
was approximately 50% overlap in the energy density
between food categories. The bottom panel of Figure 2
illustrates the percentage of energy in each of the foods that
was derived from each macronutrient. Virtually all of the
HC and HF foods met the requirements of the experimental
design in that most of their energy content was derived
from one macronutrient with the remaining energy evenly
split between the other two macronutrients. Six of the HP
foods also satis®ed this criterion. Two of the remaining
foods contained more fat than carbohydrate but these were
balanced against the remaining two foods which contained
more carbohydrate than fat.
Having constructed the experimental model as a counterbalanced arrangement of HP, HC and HF foods, it was
necessary to ascertain whether the model was capable of
allowing subjects to vary energy and nutrient intakes in a
manner that allowed changes in quantitative and qualitative
feeding to be detected. This was achieved in two ways: (i)
by statistical simulations (see Results section) and (ii) using
Model for detecting food selection changes
RJ Stubbs et al
16
Figure 3
The experimental protocol followed by the 16 men for each dietary treatment during the test day.
the dietary design in an experimental protocol that examined feeding patterns in response to prior manipulation of
the nutrient content of the diet. Since alterations (especially
decrements) in prior food and energy intakes appear to
elicit changes in quantitative intake (Caputo and Mattes,
1992; Heavey et al, 1995), it was decided to alter the
nutrient composition but not the energy content or energy
density of manipulated meals that the subjects would be
obliged to ingest. Although this manipulation is relatively
weak in terms of eliciting changes in feeding behaviour, it
enabled subsequent responses to alterations in the composition of the diet, uncontaminated by changes in prior energy
intake, to be assessed.
We have previously detected (relatively subtle) differences in hunger in response to isoenergetic macronutrient
manipulations of a mandatory diet (Stubbs et al, 1996;
Johnstone et al, 1996).
Study protocol
Figure 3 describes the experimental protocol. Sixteen subjects were each studied four times in a 2-day protocol,
using a randomised Latin-square design. On day 1, subjects
were fed a ®xed diet designed to maintain energy balance,
estimated at 1.6 6 resting metabolic rate (RMR). This diet
comprised 47% fat, 40% carbohydrate and 13% protein as a
proportion of energy. On day 2, subjects received a ®xed
mandatory intake during the morning designed to match
80% of RMR (Manipulation) and were given ad libitum
access to the selection diet from 12.30 for the remainder of
the day until bed time (Outcome). The mandatory intake
comprised breakfast at 08.30 (50% of RMR) and a drink at
10.30 (30% of RMR). Both breakfast and drink were of the
same nutrient composition. On each occasion subjects
received one of the four isoenergetic and energetically
iso-dense macronutrient manipulations corresponding to
HP, HC, HF and mixed regimes. This took the form of a
tuna lasagne and drink for breakfast (at 08.30) and a midmorning drink. From 12.30 onwards subjects had ad libitum
access to the selection of 10 high-fat (HF), 10 highcarbohydrate (HC) or 10 high-protein (HP) foods. These
foods were the same on each treatment. During the ad
libitum period they determined their own meal times
(including lunch). Subjects had continuous ad libitum
access to these foods throughout the remainder of the day
and could come to the Human Nutrition Unit (HNU) to
consume these foods as and when they required. Additionally, from 12.30 onwards all subjects could also take food
items to other parts of the campus in a large food bag (size
of a picnic hamper) in which were placed the HF, HC or HP
foods to which they had ad libitum access. If subjects
consumed all or part of one food, more items of that food
were provided. Most subjects were members of staff at the
Rowett (but not the Human Nutrition Unit) and all subjects
were able to come to the Unit or phone to request any
additional foods they required. All subjects were in the Unit
or on the Rowett campus from before 08.30, until after
18.00. It was thus ensured that subjects had access to these
foods throughout the day and a member of staff was
continually on call to deal with any requests or problems
in relation to the protocol. Prior to leaving the campus at
18.00, subjects met the duty member of staff who replenished the selection of foods in the food bag, which they
could then take home. All remaining food items and leftover packaging and food remains were returned to the HNU
the following morning. Subjects had ad libitum access to
the same 30 foods on all runs. The experimental design thus
enabled an examination of feeding responses (throughout
the afternoon and evening, until bed time) to a gross
alteration in nutrient but not energy intake during the
morning.
Prior to the commencement of the study, subjects were
required to taste all 30 ad libitum foods. Subjects were
fasted for 3 hours prior to the tasting session. They
completed one questionnaire rating their perceived pleasantness of each named food without actually being
exposed to that food. They were then given each food to
taste and again asked the same question. This was done to
establish subjects' perceived pleasantness of the generic
food type that they were familiar with and also their
perceived pleasantness of the actual foods used in the
study. The tasting session also enabled the subjects to
taste all the foods that they would receive and may have
helped avoid subjects attempting to taste numerous foods
on the ®rst exposure to them during the experiment.
Formulation and preparation of the diets
The mandatory intakes were designed to contain approximately 60% by energy of the main macronutrient with the
remaining 40% of energy evenly split between the remaining two macronutrients (Table 2). The mixed mandatory
intake was designed to contain 33% of its energy from each
macronutrient. The compositions of the breakfast and midmorning drinks were calculated from current UK food
tables and supplemented updates (Holland et al, 1991).
These foods were not part of the ad libitum design, and
contained rather unusual, covert modi®cations of a basic
recipe. This may have led to a less pronounced behavioural
response in terms of subsequent macronutrient selection,
since subjects appear to respond less to covert than overt
dietary manipulations (see Blundell and Stubbs, 1997).
Thus subjects would not be responding to cognitive cues
associated with prior experience of known food items
Model for detecting food selection changes
RJ Stubbs et al
Table 2 Mean energy and nutrient intakes for the four manipulated
breakfasts
Diet
Mixed
High-protein
High-carbohydrate
High-fat
Percent
protein
Percent
CHO
Percent
fat
Energy
(MJ)
Weight
(kg)
33
60
19
20
33
20
61
20
34
20
20
60
5.81
5.77
5.84
5.81
1504
1504
1504
1504
contained in the mandatory intakes given at breakfast and
mid-morning. It should be noted that such cognitive beliefs
may have formed part of the behavioural response when
subjects consumed the ad libitum macronutrient-selection
diet, since this comprised common, familiar foods. Thus
the design requirements of the mandatory and ad libitum
foods were very different. The ad libitum diets had to
contain, normal, recognisable everyday food. The breakfast
and the mid-morning drink on the other hand, had to be
similar in taste, texture and appearance but all foods had to
be palatable and appealing to the subjects. The breakfast
and mid-morning drink were pilot tested and altered
accordingly to meet these design criteria. The water loss
on the cooking of the lasagnes was measured and replaced
after cooking. The lasagnes were made prior to the start of
each run and were frozen. All the ad libitum foods were
tasted by the unit staff (n ˆ 16) for acceptability and
comments prior to inclusion in the experimental diets.
The drinks and milkshakes were made up fresh on day 2.
During the ad libitum period all food was available in
excess and the same amounts of foods were given on each
run.
Presentation of the diets and measurements of food intake
On day 1 (maintenance day) diets were packed into a large
coolbag for subjects to take away with them. They were
instructed to consume meals at ®xed times. On day 2,
subjects attended the HNU for breakfast (08.30) and their
mid-morning drink (10.30) and from 12.30 onwards subjects returned to the HNU where they ate and=or received a
coolbag with 30 different foods to which they had ad
libitum access. They could eat in the Unit or from the
foods in their coolbag throughout the day and could request
extra foods throughout the day. All volunteers returned the
coolbags the following day. All foods were weighed before
and the remainders after consumption. The investigator
noted, to the nearest gram, the weight of the food recovered
and therefore the weight of food consumed could be
calculated. Since the subjects recorded what they ate and
when they ate, it was possible to build up a pro®le of meal
size, frequency and composition for each subject throughout the ad libitum feeding period.
Subject recruitment and physical measurements
Sixteen healthy non-obese men, who were not consuming
any specialised diet and who were not taking any medication, were recruited by advertisement. Their mean (s.d.)
weight was 73.56 (8.24) kg, height was 1.77 (0.07) m and
age was 27.50 (9.42) years. They were not informed that
the true purpose of the study was to measure changes in
food selection as a main outcome variable. Height, weight
and resting metabolic rate were measured as detailed elsewhere (Johnstone et al, 1996).
Questionnaires
A number of questionnaires were administered to subjects
at various points in the protocol. These are detailed below.
Assessment of restraint emotionality and externality:
Subjects were required to complete a Dutch Eating Behaviour Questionnaire (van Strien et al, 1986) that attempts to
dissociate restraint, externality and emotionality.
Psychometric assessments of hunger and appetite:
Visual analogue scales were completed every waking hour
throughout the study on hand-held computers with a penbased graphical interface (Apple Newton Messagepad).
These assessed changes in subjective appetite, hunger and
satiety, based on the methodology of Blundell and colleagues (Blundell, 1979; Hill and Blundell, 1982). The Electronic Appetite Rating System has recently been described
and validated (Stratton et al, in press).
The computer also contained a diary facility in which
subjects recorded what they ate and when they ate it. Once
the diary was completed, another two questions appeared
on the screen in the form of visual analogue scales asking
subjects to rate how pleasant and how satisfying the meal
was.
End-of-day questionnaire:
Subjects also had to complete end-of-day questionnaires
that assessed general mood over the whole day. The
questionnaire took the form of visual analogue scales in
relation to alertness, liveliness, hunger, thirst, preoccupation with thoughts of food, cravings for foods, depression,
boredom and anxiety.
The study was approved by the Joint Ethical
Committee of Grampian Health Board and the University
of Aberdeen.
Statistical analysis
The visual analogue scales were analysed by splitting the
day into three periods and, for each subject on each diet,
calculating a mean rating in each period. The three periods
were pre-lunch (and post-breakfast), post-lunch (and presupper) and throughout the whole day. Because subjects ate
meals at times determined by themselves, the mean intermeal value was derived from the ®rst value after the last
meal until the last value before the next meal. Analysis of
variance was performed on the average ratings in each
period, with diet as a factor and subject as a blocking
factor. Because subjects may have eaten at different times,
the latency to eat after the last mandatory intake (10.30 on
day 2) was also examined by ANOVA, using diet as factor
and subject as blocking factor. The blocking factor operates
as follows: in an experiment comparing a set of treatments
on several subjects, if the experimental units (i.e. individual
subjects on a particular treatment) can be placed into
groups in which the variation is less than over the entire
experiment (e.g. an individual subject's behaviour on all
the treatments is likely to be less variable than that of all the
subjects on all treatments) and treatments are assigned
within groups (i.e. each subject receives each treatment),
comparisons between treatments can be made with greater
accuracy. These groups (in our case subjects) are referred
17
Model for detecting food selection changes
RJ Stubbs et al
18
Table 3 Correlation matrix depicting the relationship between
macronutrients in random simulations of food selection from the
experimental diets in 100 subjects
Protein
Fat
Carbohydrate
Protein
Fat
Carbohydrate
1.00
7 0.06
7 0.326
1.00
7 0.74
1.0
to as blocks. In the analysis of variance the variation
between the blocks (subjects) can be separated from the
remaining variation (between treatments), and differences
between treatments can be estimated within the blocks
(subjects), often with a marked improvement in accuracy.
Lunch and post-lunch intake of food, energy and macronutrients was analysed by analysis of variance with diet as a
factor and subject a blocking factor. All analysis was
performed using the GENSTAT 5 statistical package (Genstat 5 Committee of the Statistics Department, AFRC, UK).
Food selection in relation to preference was analysed using
chi-squared tests of independence. End-of-day questionnaires were analysed by ANOVA using subject as blocking
factor and sex and treatment as factors.
Results
Statistical simulations
Statistical simulations were conducted based on an
assumed 100 subjects each randomly selecting eight
foods. Based on these simulations, a correlation matrix
(Table 3) was constructed to test for covariance between
macronutrients (expressed as a percentage of energy).
Simulations suggested a strong negative correlation
between carbohydrate and fat and between carbohydrate
and protein, indicating good separation during random
selection. The lack of relationship between fat and protein
suggested slightly less good separation. There was no
positive covariance between any macronutrients, indicating
that the model allowed independent selection of each
macronutrient.
Food, energy and nutrient intakes
The mean energy and nutrient intakes for the day (including
the mandatory breakfast intakes), on each of the dietary
manipulations are shown in Table 4 together with the Fratios and probability statistics for the main effects. As can
be seen from the table, there were no signi®cant differences
in the amount of food and drink consumed during the whole
day. Subjects ate 3.53, 3.55, 3.51 and 3.50 kg=day on the
mixed, HP, HF and HC treatments, respectively
(s.e.d. ˆ 0.18). Subjects differed from each other in the
amount of food and drink consumed (F15;42 ˆ 3:49,
P < 0.01). This effect was apparent for most parameters
and is typical of feeding responses in the laboratory and in
real life, and will not therefore be discussed further.
Subjects also ate a similar amount of energy subsequent
to each treatment, giving total values of 15.94, 17.38, 17.64
and 17.18 MJ=day (inclusive of mandatory intakes) on the
mixed, HP, HF and HC treatments, respectively
(s.e.d. ˆ 1.17). Since subjects were relatively sedentary
throughout the day they were in a positive energy balance
with intakes at 1.41, 1.54, 1.55 and 1.51 times the estimated
daily energy requirements, respectively (at 1.6 6 BMR).
By the end of the day, subjects had consumed signi®cantly more protein on the HP treatment, signi®cantly more
fat on the HF treatment and signi®cantly more carbohydrate
on the HC treatment than on any of the other diets,
re¯ecting the composition of the breakfasts eaten. Analysis
of variance conducted on the ad libitum intakes alone
con®rmed that there were no signi®cant differences in the
weight of food, energy or nutrient intakes during this
period. These values can be obtained by subtracting the
average mandatory intakes in Table 2 from the total 24hour intakes in Table 4.
Breaking the intakes down further into lunch and postlunch intakes of food, energy and nutrients revealed interesting patterns within the day. Subjects ate signi®cantly
more food, energy, protein and fat and to a lesser degree
carbohydrate at lunch time on the HF treatment than on any
other diet. The intakes on the other diets were not signi®cantly different from each other. Average values, Fratios and probability statistics for the lunch time intakes
are given in Table 5. During the remaining ad libitum
period, subjects showed a tendency to reverse this pattern
on the HF-breakfast treatment. Thus energy intake tended
to be lower than on the other three diets (F…3;42 † ˆ 2:96,
P ˆ 0.09). This was speci®cally attributable to a decrease in
fat intake on the HF-breakfast treatment, throughout the
post-lunch ad libitum period (F…3;42 † ˆ 6:90, P ˆ 0.012),
leading to the similar intakes of food and energy on all
treatments over the 24 hours. These effects were not
signi®cant for protein or carbohydrate. Thus only on the
HF-breakfast treatment was food, energy and particularly
fat intake higher at lunch, and fat intake (and hence EI) was
lower during the remainder of the ad libitum period, than
on any other treatment.
Table 4 Average 24-hour food, energy and nutrient intakes (including the manipulated breakfasts) for the four dietary treatments, together with the
variance ratios and statistical probability for the main treatment effects
Wet weight (kg)
Energy (MJ)
Protein (MJ)
Fat (MJ)
Carbohydrate (MJ)
M
HP
HF
HC
3.53
(0.14)
15.94
(0.68)
3.32
(0.14)
6.92
(0.41)
5.70
(0.31)
3.55
(0.19)
17.38
(1.39)
4.98
(0.25)
6.76
(0.68)
5.64
(0.36)
3.51
(0.15)
17.64
(0.88)
2.99
(0.17)
9.08
(0.52)
5.57
(0.36)
3.50
(0.18)
17.18
(1.21)
2.77
(0.20)
6.80
(0.71)
7.61
(0.45)
Variance
ratio
(F…3; 42†)
s.e.d.
P
0.03
0.18
0.993
0.85
1.17
0.476
46.02
0.21
< 0.001
9.55
0.51
< 0.001
6.80
0.54
< 0.001
Model for detecting food selection changes
RJ Stubbs et al
19
Table 5 Average food, energy and nutrient intakes for the four dietary treatments, together with the variance ratios and statistical probability for the main
treatment effects, during the lunch period
Wet weight (kg)
Energy (MJ)
Protein (MJ)
Fat (MJ)
Carbohydrate (MJ)
M
HP
HF
HC
0.81
(0.15)
3.65
(0.51)
0.56
(0.08)
1.79
(0.27)
1.30
(0.21)
0.85
(0.18)
3.58
(0.63)
0.49
(0.11)
1.58
(0.33)
1.51
(0.25)
0.98
(0.16)
5.73
(0.74)
0.84
(0.13)
2.78
(0.42)
2.11
(0.28)
0.70
(0.16)
3.87
(0.69)
0.55
(0.11)
1.74
(0.34)
1.58
(0.34)
Latency to eat
There was no signi®cant difference between treatments in
the time before the onset of the next ingestive event (lunch)
after completion of the mandatory intakes (F…3;42 † ˆ 0:71,
P ˆ 0.553). The average time between the end of the 10.30
drink and the onset of lunch was 208, 256, 208 and 222 min
on the mixed, HP, HF and HC treatments respectively
(s.e.d. ˆ 38.5 min).
Food preference
Subjects tended to select foods they liked and avoid foods
they disliked. Pro®teroles and cream were selected most
often (47 times), closely followed by sausage rolls (42
times). The favourite foods from the tasting tended to be
selected in each run of the experiment. Foods that subjects
claimed to dislike in the tasting session, such as cottage
cheese crackers, crab pate crackers and pizza crackers were
not selected at all. Surprisingly, `Twiglets' (a savoury
marmite corn snack), which many subjects claimed were
their least favourite food, were selected 16 times. There
was no evidence of selection of different food groups in
response to the breakfast treatments (X2 ˆ 34.34, 87 df,
P ˆ 1.0) and there was no evidence of selecting different
foods on different runs (X2 ˆ 44.77, 87 df, P ˆ 1.0).
Subjective hunger and fullness
Table 6 gives the average values for subjective hunger
between breakfast and lunch, between lunch and supper,
and during the whole day, together with the F-ratios and
probability values for the main effects. Average subjective
hunger was signi®cantly different between breakfast and
lunch (F…3;44 † ˆ 7:35, P < 0.001). Average values were 21,
16, 31 and 29 mm on the mixed, HP, HF and HC treatments, respectively (s.e.d. ˆ 4 mm). These could be
accounted for by a contrast between the mixed and HP
Table 6 Average subjective hunger for the 16 subjects on each dietary
treatment before and after lunch
Subjective
hunger
Before lunch
After lunch
Whole day
Mixed
HP
HF
HC
s.e.d.
Variance
ratio
(F(2,44))
21
31
27
16
28
26
31
31
31
29
31
30
4
4
3
7.35
0.35
1.66
Probability
< 0.001
0.788
0.185
Variance
ratio
(F(4,41))
s.e.d.
Probability
0.83
0.18
0.485
3.37
0.75
0.028
3.13
0.13
0.039
3.82
0.40
0.017
2.25
0.32
0.098
treatments on the one hand and the HF and HC treatments
on the other (P < 0.001).
Subjective fullness was also signi®cantly in¯uenced by
the dietary treatment, in the period between breakfast and
lunch (F…3; 44† ˆ 5:93, P < 0.002). Average values were
61, 70, 54 and 63 mm on the mixed, HP, HF and HC
treatments, respectively (s.e.d. ˆ 3.5 mm). Thus subjects
were signi®cantly more full on the HP treatment relative
to all other treatments and signi®cantly less full on the HF
treatment relative to all other treatments, during the period
between breakfast and lunch. There were no signi®cant
treatment effects on desire to eat. Prospective consumption,
urge to eat and preoccupation with thoughts of food all
followed broadly similar trends to those for subjective
hunger. There were no signi®cant treatment effects for
hunger or any other subjective sensation either throughout
the afternoon or over the whole day.
End-of-day questionnaires
The end-of-day questionnaire produced no signi®cant treatment effect for hunger, alertness, liveliness, preoccupation
with thoughts of food, depression, boredom, anxiety or
cravings. Subjects recalled being signi®cantly more thirsty
throughout the day on the HF treatment relative to other
treatments (F…1; 37† ˆ 6:46, P ˆ 0.015).
Dutch Eating Behaviour Questionnaires
Subjects' restraint varied with a range between 1.00 and
3.60 with a mean of 2.20. The CV% was 40.61. Thus on
average these subjects were unrestrained eaters.
Discussion
Testing the ability of the model to detect changes in
macronutrient selection
Statistical modelling: The simulation based on 100 people,
each randomly selecting an average of eight foods, suggested a strong negative correlation between carbohydrate
and fat and between carbohydrate and protein and no
relationship between fat and protein. Since there was no
positive covariance between any macronutrients, the
simulations suggested that the model allowed independent
selection of each macronutrient. This suggests that subjects
would be able to use the foods available in the experimental
design to directionally alter their nutrient and energy intake
if they chose to do so.
Model for detecting food selection changes
RJ Stubbs et al
20
Table 7 Mean 24-hour and range (inclusive of mandatory intakes) of food (and drink), energy and nutrient intakes for each dietary treatment. The range
re¯ects the extent of interindividual variation
Mixed
Weight (kg)
Energy (MJ)
Protein (MJ)
Fat (MJ)
Carbohydrate (MJ)
HP
HF
HC
Overall
Mean
Range
Mean
Range
Mean
Range
Mean
Range
Mean
Range
3.53
15.94
3.32
6.92
5.70
2.20
11.01
2.20
5.66
4.32
3.55
17.38
4.98
6.76
5.64
2.90
18.35
3.64
9.92
7.53
3.51
17.64
2.99
9.08
5.57
2.15
13.71
2.29
6.78
6.20
3.50
17.18
2.77
6.80
7.61
2.59
16.86
2.78
10.50
6.01
3.52
17.04
3.51
7.39
6.13
3.00
18.96
5.53
11.27
8.88
Interindividual variation: Subjects showed signi®cant
interindividual variation in food, energy and nutrient
intakes throughout the study (Table 7). This showed that
differences in food, energy and macronutrient intake
existed between subjects as one would expect if the dietary
design facilitated signi®cant differences in quantitative and
qualitative feeding behaviour between subjects.
Experimental test: The primary objective of this experiment was to design a model that is capable of detecting
changes in both quantitative and qualitative feeding behaviour. From the results of the total ad libitum intakes, it
initially appears that the model did not detect any signi®cant changes in energy and nutrient intakes in spite of prior
manipulations of the nutrient composition of mandatory
intakes during the morning. However, closer scrutiny of
feeding patterns within the day revealed interesting trends
regarding the ability of the model to detect changes in
energy and nutrient intake. The breakdown of intakes into
lunch and post-lunch intakes of food, energy and nutrients
revealed that subjects ate signi®cantly more energy, protein
and fat and to a lesser degree carbohydrate at lunch time, on
the HF-breakfast treatment than on any other treatment.
Since weight, volume appearance and energy density of all
the breakfast treatments were similar, it appears that
differences in feeding behaviour at lunch were due to the
composition of the mandatory intakes consumed during the
morning. This difference cannot be accounted for by a
preference effect, since the same selection of foods was
used in a counterbalanced, repeated-measures design.
Energy and fat intakes were speci®cally decreased in the
afternoon on the HF-breakfast treatment compared to other
dietary treatments. This model was able to detect signi®cant changes in energy and nutrient intake in relation to a
relatively weak (isoenergetic macronutrient manipulation)
alteration of diet composition during the morning. These
data support the statistical simulations in suggesting that
the model is sensitive to changes in energy and nutrient
intake.
The combined data derived from the statistical simulations, analysis of interindividual variation of energy and
nutrient intakes, and the experimental results demonstrate
that this new model appears capable of detecting meal-tomeal changes in energy and nutrient intakes. However,
further re®nements of the model are still required.
Subjects did not show any detectable difference in the
selection of food groups (HP, HC or HF) either at lunch or
during the rest of the day. It can be appreciated that
sizeable changes in energy and nutrient intakes will occur
before signi®cant changes in the selection of food groups
will be detected.
Ability of the model to distinguish between the effects of
protein, carbohydrate and fat on aspects of appetite control
Protein is widely recognised as the most satiating macronutrient (Stubbs, 1995). HP foods exert a potent in¯uence on
satiety relative to isoenergetically dense HC or HF foods
(Johnstone et al, 1996; Stubbs et al, 1996). In the present study
there was a signi®cant difference in hunger, during the
mandatory intake period, between the HP and mixed mandatory intakes and the HC and HF treatments. Furthermore, all
macronutrients were distinguished from each other on the
basis of subjectively expressed fullness. The fat treatment was
distinguished further from other treatments through its effects
on feeding behaviour in the ad libitum period and expressed
thirst (end of day questionnaires).
Apparent compensation but not regulation of energy intake
Subjects did not appear to regulate energy balance over the
course of this study. Indeed, they consumed on average
1.50 times their daily energy requirements estimated at 1.6
times their basal metabolic rate. It is likely that a large
mandatory intake during the earlier part of the day did not
precipitate a large degree of compensation in the second
half of the day. This study could not discern whether this
was the case since there was no zero-breakfast controlÐthe
main objective of the study being to assess subjects'
response to isoenergetic macronutrient manipulations.
Furthermore, there was no apparent tendency to autoregulate macronutrient intakes during the ad libitum period in
response to the gross manipulation of macronutrient intake
during the mandatory morning feeding period. It is likely
that large, directional changes in macronutrient selection
will take a minimum of several days to develop. However,
on the HF-breakfast treatment alone, subjects did not
appear to compensate during the afternoon for the higher
intakes (compared with other treatments) at lunch. Why did
the HF-treatment produce an apparent positive feedback on
EI at lunch, which was compensated for later? Ramirez et
al (1989) noted that part of the hyperphagic effect of HF
diets may be due to the fact that they exert a weaker
osmotic effect in the gastrointestinal tract. Subjects were
indeed more hungry and less full before the ad libitum
period on the HF-breakfast treatment. The end-of-day
questionnaire also showed that they recalled being signi®cantly less thirsty over the course of the day on this
treatment. The exact way in which ¯uids interact with the
food matrix in the gut, to in¯uence overall feeding patterns
is presently unclear. It may be hypothesised that the
Model for detecting food selection changes
RJ Stubbs et al
initially larger intake on the HF-breakfast treatment was
partially due to this osmotic effect between dietary treatments. We have previously described how a HF diet
produced more delayed effects on satiety than did an
energetically iso-dense HC diet (Johnstone et al, 1996).
This may account for the apparent compensation on the
HF-breakfast treatment during the remainder of the ad
libitum period in the present study.
Relationship between preference and food selection
The major determinant of food selection appears to have
been subjects' habitual preference for foods. This is
because habitual preference correlated well with actual
preference when rated for the tasted food, and these
preference pro®les predicted reasonably well the qualitative
food selection. However, given the size of the effect of
apparent habitual preference pro®les in determining patterns of intake, it is reasonable to suggest that this model
should be continued over at least several days per treatment
in order to increase the probability of detecting macronutrient-speci®c effects. Such studies are currently under way
as an extension to the present study.
Conclusions
This experimental model appears sensitive in that it was able
to detect signi®cant changes in meal-to-meal energy and
nutrient intake on the HF treatment relative to other treatments. The model was also able to distinguish between the
effects of different macronutrients through psychometric
responses relating to hunger, fullness and thirst. The statistical
simulations suggested that the model provided foods that gave
good separation of macronutrients to an extent that subjects
were able to achieve marked interindividual variations in
nutrient and energy intakes in the actual experiment (Table 7).
The major determinant of macronutrient and food selection in
this particular study appeared to be the preference pro®les of
the subjects. There are a number of re®nements to be made to
this model to better achieve the ability to detect qualitative
and quantitative feeding in humans. These re®nements are
currently being incorporated into additional studies. The
model appears useful for detecting changes in selection
between foods enriched in protein, carbohydrate and fat in
laboratory studies using within-subject, repeated-measures
designs.
References
Blundell JE (1979): Hunger appetite and satietyÐconstructs in search of
identities. In: Nutrition and Lifestyles, ed. M Turner, pp 21 ± 42.
London: Applied Science Publishers.
Blundell JE & Stubbs RJ (1997): Diet and food intake in humans. In
International Handbook of Obesity, ed. GA Bray, C Bouchard & WPT
James. New York, Marcel Dekker.
Caputo FA & Mattes RD (1992): Human dietary responses to covert
manipulations of energy, fat and carbohydrate in a midday meal. Am. J.
Clin. Nutr. 56, 36 ± 43.
Cotton J, Burley V, Westrate J & Blundell J (1994): Dietary fat and appetite:
similarities and differences in the satiating effect of meals supplemented
with either fat or carbohydrate. J. Hum. Nutr. Diet. 7, 11 ± 24.
de Graaf C & Hulshof T (1995): The effect of non-absorbable fat on
energy and fat intake. Eur. J. Med. Res. 1, 72 ± 77.
DeCastro JM (1987): Macronutrient relationships with meal patterns and
mood in the spontaneous feeding behaviour of humans. Physiol. Behav.
39, 561 ± 569.
Foltin R, Fischman M, Emurian C & Rachlinski J (1988): Compensation
for caloric dilution in humans given unrestricted access to food in a
residential laboratory. Appetite 10, 13 ± 24.
Foltin R, Fischman M, Moran T, Rolls B & Kely T (1990):
Caloric compensation for lunches varying in fat and carbohydrate
content by humans in a residential laboratory. Am. J. Clin. Nutr. 52,
969 ± 980.
Forbes JM (1995): Voluntary Food Intake and Diet Selection in Farm
Animals. CAB International.
Heavey PM, McKenna APM, Goldberg PG, Murgatroyd PR &
Prentice AM (1995): Underfeeding by reduction in fat or carbohydrate
intake: effects on energy expenditure, macronutrient oxidation and
subsequent food intake in lean men. Nutr. Soc. 54, 160A (abstract).
Hill AJ & Blundell JE (1982): Nutrients and behaviour: research
strategies for the investigation of taste characteristics food preferences
hunger sensations and eating patterns in man. J. Psychol. Res. 17,
203 ± 212.
Hill AJ, Rogers PJ & Blundell JE (1995): Techniques for the experimental
measurement of human eating behaviour and food intake: a practical
guide. Int. J. Obesity 19(6), 361 ± 375.
Holland B, Welch AA, Unwin I, Buss DH, Paul AA & Southgate DAT
(1991): McCance and Widdowson's The Composition of Foods, 5th
revised edn. London: HMSO.
Johnstone AM, Stubbs RJ & Harbron CG (1996): Effect of overfeeding
macronutrients on day-to-day food intake in man. Eur. J. Clin. Nutr. 50,
418 ± 430.
King NA (1995): PhD Thesis. Leeds University, Department of Psychology.
Leibowitz SF (1992): Neurochemical ± neurendocrine systems in the brain
controlling macronutrient intake and metabolism. Trends Neurosci. 15,
491 ± 497.
Poppitt SD, Swann D, Black AE & Prentice AM (1995): Is under-reporting
of energy intake in obese women macronutrient speci®c? Covert
measurements in a metabolic facility. Int. J. Obesity 20, 27 (abstract).
Ramirez I, Tordoff G & Friedman MI (1989): Dietary hyperphagia and
obesity: what causes them? J. Physiol. Behav. 45, 163 ± 168.
Rising R, Alger S, Boyce V, Seagle H, Ferraro R, Fontvieille A &
Ravussin E (1992): Food intake measured by an automated foodselection system: relationship to energy expenditure. Am. J. Clin.
Nutr. 55, 343 ± 349.
Silverstone T & Goodall E (1986): Measurement of hunger and food
intake. In Disorders of Eating Behaviour, a Psychoneurendocrine
Approach, ed. E. Ferrari, pp 128 ± 134. Oxford: Pergamon Press.
Stratton RJ, Stubbs RJ, Hughes DA, King N, Blundell JE & Elia M (1997):
Comparison of the traditional paper visual analogue scale questionnaire
with an Apple Newton electronic appetite rating system (EARS) in free
living subjects feeding ad libitum. Proc. Nutr. Soc. 56, 230A
Stubbs RJ (1995): Macronutrient effects on appetite. Int. J. Obesity 19(5),
S11 ± S19.
Stubbs RJ, Harbron C, Murgatroyd PR & Prentice AM (1995):
Covert manipulation of dietary fat and energy density: effect
substrate ¯ux and food intake in men eating ad libitum. Am. J. Clin.
Nutr. 62, 1 ± 14.
Stubbs RJ, Van Wyk MCW, Johnstone AM & Harbron C (1996): Breakfasts high in protein fat or carbohydrate: effect on within-day appetite
and energy balance. Eur. J. Clin. Nutr. 50, 409 ± 417.
Stubbs RJ, O'Reilly LM, Johnstone AM & Poppit SA (1998):
Methodological issues relating to the measurement of food, energy
and nutrient intake in human laboratory-based studies. Proc. Nutr. Soc.
57, 1 ± 16.
van Strien, T. Frijiters JER, Berges GPA & Defares PB (1986): The Dutch
eating behaviour questionnaire (DEBQ) for assessment of restrained
emotional and external eating behaviour. Int. J. Eat. Disord. 5, 295 ±
315.
Wurtman JJ & Wurtman RJ (1982): Studies on the appetite for carbohydrates in rats and humans. J. Psychiatr. Res. 17(2), 213 ± 221.
21