Fig 2. Variation among individuals in food preferences,
considering energetic and nutritional implications. The plots show
individual-level random effects of Bayesian Mixed Models, estimated
using four replicates per individual. The red line represents the
average value of the population. The multidimensional niche centre is
represented by PCoA 1 and PCoA 2.
If energy value were the only driver of preferences, we would expect
that pigeons mostly consumed sunflower seeds, which are richer in fat
than any other tested food (i.e. they should show higher values of PCoA
1; see Table S2 ). However, only some individuals showed strong
preferences for sunflower; others instead preferred cereal, notably
wheat (Figs. 1A, 2, S1 ). The energy density of cereal is lower,
due to the reduced fat contents, but they are richer in carbohydrates.
The breakdown and absorption of carbohydrates is much faster than fat,
providing energy that can rapidly be used by the animal. A bi-variate
Bayesian Mixed Model revealed that preferences for carbohydrates and fat
were negatively correlated among individuals, a correlation that was
independent of the amount of food consumed during the trial (MOIA:
Correlation = -0.95, 95% CI=-0.98 to -0.88; BCN: Correlation = -0.99,
95% CI=-0.98 to -0.99). While energy has traditionally been considered
the primary currency influencing food choice, our results suggest that
the relative amounts of various nutrients is also relevant (Fig.
2 ).
Besides nutrient contents, food choice also appeared to be influenced by
morphological characteristics of the food. Of the three cereal grains,
wheat was generally preferred over popcorn and oats. The main difference
among these food types was their size. Pigeons use both size and number
of items in assessing the amount of food, but they weight number more
heavily than size (Shettleworth 1987b). It follows that when provided in
the same amount, they should prefer the smaller seed —i.e. wheat.
Again, however, some individuals consistently preferred corn over wheat
during the four trials, reinforcing the view that individuals remain
selective even when the searching and handling costs of using
alternative resources is negligible.
Having established consistent individual differences in food
preferences, we estimated the contribution of individual niche breadth
(WIC) to the total niche breadth of the population (TNW). We used Rao’s
quadratic entropy to estimate WIC and TNW while considering nutritional
and morphological differences between the resources. We found that the
niches of individuals represented, on average, 59% of the niche breadth
of the population (mean ± WIC/TNW = 0.68 ± 0.51 for MOIA and 0.51 ± 0.35
for BCN). Thus, individual niche widths were considerably narrower than
the niches of their populations, a pattern that was consistent despite
striking ecological differences among populations (Sol 2008). However,
we also detected substantial niche overlap between each individual and
the remaining individuals of the population (75.4% ± 24.7 for BCN and
88.9 ± 15.8 for MOIA). This overlap indicates that within a population
the majority of individuals shared similar food preferences
(Fig. 2 ). It also reflects that the studied populations
included both specialists and generalists (Table 1, Fig S2 ).
Variation among individuals in food preferences may be ephemeral if they
reflect differences in the physiological condition of the individual,
for example regarding its nutritional state. Indeed, the amount of food
consumed was highly repeatable across individuals (Fig. 2 ).
This suggests that some individuals were more motivated than others to
eat. However, we found no evidence that the niche centre of individuals
was influenced by their body condition (Table S4 ) or by the
total amount of food consumed during an assay (effect size
> 0.01 [± 0.1] in all cases). We neither found that
fasting makes individuals either more discriminant or more willing to
accept a wider variety of foods (Moon & Zeigler 1979). Indeed, choices
in the first trial of the day, when the individual had not consumed any
food in the previous 12h, were highly consistent with choices in the
second trial—when individuals have had that chance to already eat for
20 minutes (Table S3 ). These findings do not necessarily
contradict the view that hungry animals make sub-optimal food choices
(Moon & Zeigler 1979), as before the experiments food was providedad libitum . Therefore, it was unlikely that any individual was in
bad body condition.
Long-term consistencies in food preferences among individuals are
expected if these are either heritable or learned individually or in a
social context. To disentangle genetic and environmental effects, we
examined the preferences of birds raised in captivity. Like in the case
of wild pigeons, captive-bred individuals also exhibited consistent
variation in food preferences despite being raised under similar
environmental conditions (Fig. 2, Table 1 ). However,
the cross-fostering experiment revealed that the additive genetic
variance component of the niche centre was low (PCoA 1:h2 = 0.082, CI95% = 0.000 to
0.375; PCoA 2: h2 = 0.041,
CI95% = 0.011 to 0.256). The niche centre was indeed
little affected by more stable state variables like body size and beak
shape (Tables S6, S7 ), which have high genetic influences
(h2 > 0.7, Authors unpublished).
Likewise, we found that the effect of the common rearing environment was
also low (PCoA 1: e2 = 0.001,
CI95% = 0.021 to 0.214; PCoA 2:e2 = 0.047, CI95% = 0.015 to
0.153), suggesting that diet was little influenced by the vertical
transmission of information from the parents. Together, additive genetic
and common environmental variances accounted for >14% of
all phenotypic variance, and hence only marginally contributed to
variation in niche centre (Table S5 ).
Not surprisingly, the food preferences of individuals were not conserved
after one year (Fig 3 ). During the long-term assays
individuals still showed consistency in their food preferences across
trials (Table 1 ), yet these preferences were different from
those observed during the short-term assays (Figs. 3, 4 ). These
changes in preferences did not reflect heritable variation in diet
plasticity (Table S8 ). However, a role for learning is
suggested in the tendency of individuals to converge over time toward a
diet richer in fat (i.e. more caloric) and proteins (Fig. 4 ).
If niche shifts were driven by random processes, we would not expect
such a convergence. Thus, the extent to which the niche centre changed
over time did not seem to reflect variation in responsiveness (i.e.
ability to accommodate the behaviour to current conditions). Rather, it
appears to be mainly determined by the distance to the “optimum” to
which individuals tended to converge, changing on average more for wild
pigeons than for those raised in captivity.