Are melanized feather barbs stronger?
Biology Department, 6500 College Station, Bowdoin College, Brunswick, ME 04011, USA
* Author for correspondence (e-mail: ajohnson{at}bowdoin.edu)
Accepted 8 October 2003
![]() |
Summary |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Key words: barb, feather, strength, biomechanics, melanin, color, material properties
![]() |
Introduction |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Few studies, however, have tested for mechanical differences between
melanized and unmelanized keratin. Several studies have shown a correlation
between increased melanin and abrasion resistance (Burtt,
1979,
1986
;
Barrowclough and Sibley, 1980
;
Lee and Grant, 1986
;
Kose and Møller, 1999
).
Bonser and Witter (1993
) found
that the keratin of melanized European starling (Sturnus vulgaris)
bills had a significantly higher Vickers hardness than did unmelanized bill
keratin. Similarly, Bonser
(1995
) found that melanized
feather keratin of the willow ptarmigan (Lagopus lagopus race
scoticus) had a higher Vickers hardness than did unmelanized feather
keratin. In the behavioral literature, strength and hardness have been equated
(Fitzpatrick, 1998
), despite
any direct experimental evidence linking the two. Perhaps they are related
because in the center of feather keratin is a large crystalline region (Greg
and Rogers 1984, cited in Vincent,
1990
), and in crystals tensile strength is a maximum of one-third
its Vickers hardness (Vincent,
1990
). Thus, greater hardness sometimes implies greater tensile
strength. Although a direct linkage between hardness and breaking has not been
established for feather barbs, Burtt
(1986
) quantified a direct
melanin-related effect on breakage. In that study, abrasion was simulated by
small glass beads blown at feathers by an air stream; a smaller fraction of
melanized barbs than unmelanized barbs were broken.
Differences in melanin, however, might be associated with other factors of
biomechanical importance, confounding the comparisons cited above. It has been
suggested, for example, that melanin is associated with thickening of the
structure of the outer, cortical layer of the feather keratin via
deposition of melanoprotein granules (see
Burtt, 1986 for a review).
Such thickening affects the cross-sectional morphology, which will affect
derived mechanical parameters such as breaking stress. Indeed, cross-sectional
morphology was found to affect flexural stiffness of the rachises of eight
species (Bonser and Purslow,
1995
). In addition, it is possible, indeed likely, that color will
be non-randomly distributed with respect to both cross-sectional area and
position on feathers.
To assess the contribution of barb morphology and position to mechanical performance, we quantify cross-sectional area, breaking force, breaking stress, breaking strain and toughness of melanized and unmelanized barbs along the entire rachis of a primary feather from an osprey (Pandion haliaetus).
![]() |
Materials and methods |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Mechanical tests
Each feather barb was mounted in a materials testing device such that the
initial length between the screw-and-nut grips was 9-15 mm. The specimen was
extended at 8 mm min-1 until breakage. Force was measured using a
strain-gage-based force beam (error, less than ±0.02 N), and extension
was measured using a linear variable differential transformer or LVDT (error,
less than ±0.004 mm). Data were digitized (12-bit) at 100 Hz. Relevant
mechanical variables were calculated from the force extension curve, the
initial length and the cross-sectional area. The initial length was measured
using calipers (error, less than ±0.01 mm) as the distance between the
grips at zero load.
Grips at each end consisted of a screw with two nuts. The feather barb was inserted through one nut; the screw, which already had one nut threaded onto it, was then screwed onto the nut while the feather barb was in the nut. The nut already on the screw was tightened against the nut containing the feather barb. In this way, the barb was held in place by the corrugations of the screw and nut. By microscopic examination, it was possible to determine that there was no slippage at the grips because the screw threads caused permanent crimps of the barb inside the nut. The crimps corresponded to the screw threads. Breakage usually did not occur at the grips, indicating that the grips did not act to concentrate force at the ends of the test length.
We used tensile rather than bending tests because of the simplicity of
determining breaking stresses with such tests. Previous research had shown
that the rachis of pigeon flight feathers failed by buckling during four-point
bending (Corning and Biewener,
1998). We initially tried bending 10 mm lengths of feather barbs
but, because they are relatively slender (40-400 µm diameter) compared with
pigeon feather rachis, our specimens bent into U-shapes before buckling. Such
extreme bends made four-point bending tests impractical.
Cross-sectional area and wall thickness
The cross-section, height and width of each barb were measured at the
broken end via scanning electron microscopy (SEM) and NIH image
analysis software. To do this, the 3 mm closest to the broken end of each barb
was snipped off, mounted on an SEM stub and carbon-coated. Digital
photographs, including scales superimposed by the SEM software, were taken of
the snipped end of each barb. Each cross-section was typically oval to
rectangular and consisted of a solid cortex surrounding a foam-like medullary
space (Fig. 1). To obtain barb
cross-sectional areas, the outer and inner boundaries of the cortex were
traced, yielding the cross-sectional area of the entire barb,
so, and the cross-sectional area of the medullary space,
sm. By subtracting the area of the space from the area of
the entire barb, we calculated the cross-sectional area of the cortex wall,
sc.
|
Although feather barbs are almost never cylindrical (see
Fig. 1), for comparison with
previous work (Corning and Biewener,
1998; Brazier,
1927
; Alexander,
1996
) we calculated the ratio of wall thickness to mean radius,
t/
, assuming a cylindrical
shape. For a hollow cylinder with outer radius ro and
inner radius ri,
t=ro-ri and
=(ro-ri)/2.
In terms of area,
where ao is the cross-sectional area inside the outer edge
of the cylinder. Similarly,
where ai is the cross-sectional area of the hollow region.
Combining the equations immediately above, the ratio of thickness to mean
radius is:
![]() | (1) |
Calculated mechanical variables
Breaking stress, brk, was defined as breaking force,
Fbrk, divided by sc. Breaking strain,
brk, was calculated by dividing the breaking extension,
lbrk, by the original length, lo, of
the barb test section. Work to break, Wbrk, was determined
by integrating the area under the force-extension curve. Toughness,
T, or work per volume, was calculated by
Wbrk/(sclo).
Statistical tests
Factorial analysis of variance (ANOVA) was used to test for differences
between breaking force and breaking stress of melanized and unmelanized barbs.
Analyses of covariance (ANCOVA) were used with either d or
brk as covariates. ANCOVA formulas were from Zar
(1996
) and were used to test
whether standard least squares linear regression lines for the classified
variables were parallel (i.e. do the variables covary at the same rate). If
significant differences among slopes were not found, then lines were tested to
seek differences in elevations. When there were no differences in either
slopes or intercepts, the common slope and common intercept were used to plot
the least squares regression lines.
![]() |
Results |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
|
The breaking force of barbs increased as a function of increasing fractional distance from the proximal end of the feather, with a force plateau between fractional distances of 0.51 and 0.85 (Fig. 3A). Within this plateau, and in the three color bands just proximal to it, adjacent bands of different color typically did not differ in breaking force. The two most proximal bands and the most distal band all showed a significantly lower breaking force than all other bands. Similarly, cross-sectional area increased initially as a function of increasing distance and was least for the most proximal and distal bands (Fig. 3B). However, the cross-sectional area plateau was narrower and shifted more distally than the breaking force plateau.
|
ANOVA indicated overall that the breaking force of melanized barbs (1.38±0.06 N, mean ± S.E.M.) was greater than that of unmelanized barbs (1.00±0.05 N; P1,174<0.001). Similarly, overall the cross-sectional area of melanized barbs (5.70x10-3±0.26x10-3 mm2) was greater than that of unmelanized barbs (3.56x10-3±0.23x10-3 mm2; P1,154<0.001).
However, when breaking force was normalized for cross-sectional area (i.e. breaking stress was calculated), ANOVA indicated that the breaking stress of unmelanized barbs (292.7±7.9 MN m-2) was greater than that of melanized barbs (249.4±8.9 MN m-2; P1,154<0.001). When we took into account the position of the barbs on the feathers, regressions of breaking stress as a function of fractional distance along the feather were best fit by a one-slope, one-intercept model (Fig. 4; statistics reported in figure legend). Thus, strength was independent of barb color when position along the rachis was taken into account. Breaking stress of barbs decreased from the proximal to the distal end of the feather.
|
To assess the causes of this pattern of breaking stress, one needs to consider the possible contribution of the medullary foam. The relevant variable is the ratio of the medullary cross-sectional area to cortex cross-sectional area, sm/sc. This ratio increases linearly from 0.48 at the most proximal end to 1.2 at a fractional distance of 0.88 and then decreases to 0.43 for barbs at the most distal end of the rachis (mean ± S.E.M., 0.87±0.014; Table 2). In the Discussion, we consider whether the medullary foam contributes to the observed patterns of breaking stress.
|
The breaking stress did increase as a function of increasing strain, with no significant difference in this relationship between the two barb colors (Fig. 5; statistics reported in the figure legend).
|
Both toughness (Fig. 6) and breaking strain (Fig. 7) decreased as a function of fractional distance, with no difference between the slope or intercept for melanized and unmelanized barbs (statistics reported in the figure legends).
|
|
For comparison with data available in the literature on feather rachis, and because the behavior of beams is affected by their cross-sectional shape, a summary of measured morphological variables is given in Table 2.
![]() |
Discussion |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
We did find that if only color was taken into account then the breaking force of melanized barbs was significantly greater than that of unmelanized barbs, which is consistent with reports from the literature (see Introduction). However, this difference was less compelling when mean breaking force for each color band was plotted as a function of fractional distance from the distal end of the feather (Fig. 3A). Adjacent bands in the middle of the feather tended to have the same breaking force, independent of the color of the band. Similarly, bands of opposite color on either end of the feather were more similar in breaking force to each other than they were to bands in the middle of the feather.
This positional pattern of breaking force was associated with a similar positional pattern in barb cross-sectional area (Fig. 3B). Cross-sectional area largely determined differences in breaking force between barbs. Thus, the plateaus of breaking force and cross-sectional area both occur at similar points along the feather, with lower values at the more extreme ends of the feather. When we normalized breaking force for cross-sectional area (i.e. calculated breaking stress), without taking into account position, however, the breaking stress of unmelanized barbs was actually significantly greater than that of melanized barbs. When position of the barbs along the feather was taken into account as well (Fig. 4), there was no longer any significant difference in the strength of melanized versus unmelanized barbs. Thus, the strength of unmelanized barbs was higher not because of intrinsic strength differences associated with melanin but because there were more of them located in a stronger (proximal) location. Position entirely explained the strength differences among barbs.
The absence of material property differences that could be attributed to melanization is also apparent when breaking stress is analyzed as a function of breaking strain (Fig. 5). Similarly, toughness (Fig. 6) and breaking strain (Fig. 7) of barbs was not different for melanized versus unmelanized barbs when these variables were considered as a function of fractional distance. Thus, material properties of barbs did not depend on melanin.
It has been suggested that melanization serves to increase hardness by inducing thickening of the tegument (see Introduction), thereby increasing cross-sectional area. In Fig. 3B, we can compare cross-sectional area of adjacent differently colored bands. In five of nine such comparisons, there were no differences in cross-sectional area. In the four comparisons where significant differences were detected, the melanized barbs always had higher cross-sectional area. This lends at least speculative support to the possibility that melanin might slightly increase cross-sectional area and hardness. Mainly, however, the pattern of variation indicates the necessity of careful sampling design to tease apart positional effects from possible melanin-induced effects.
Does the strength of barb cortex decrease distally along the
rachis?
Our estimate of barb cortical strength assumed no contribution from the
medullary material; could a systematically changing contribution of the
medullary material explain the apparent distal decrease in cortical strength?
It can be shown, as follows, that if the relative stiffness of the medulla is
1% of that of the cortex (Bonser,
1996) then the overestimate of strength is between 0.4% and 1.2%,
depending on the proportion of total cross-sectional area that consists of
medulla. To see this, consider that a barb is a structure with two materials
that contribute to the total tensile breaking force,
F=Fc+Fm, where
Fc and Fm are the breaking force of
the cortical and medullary materials, respectively. The structure fails at a
strain,
, and the forces due to the cortex and medulla are
Fc=Ecsc
and
Fm=Emsm
,
respectively, where Ec and Em are the
Young's moduli of the cortex and medulla, respectively. Let the modulus of one
material be a multiple of the modulus of the other material,
Em=kEEc, with a constant
kE, and let the cross-sectional area of one material be a multiple
of the other area, sm=kasc,
with a constant ka. We estimated cortical strength by dividing the
total force by the cortical area. To the degree that the medullary material
contributes to the total force, cortical strength will be overestimated. We
can calculate the factor by which the cortical strength is overestimated by
dividing the estimated strength by the hypothetical strength, combining the
equations above:
![]() | (2) |
For the feather barbs in this study, ka averages 0.87
(Table 2) and ranges between
0.43 and 1.2; Bonser (1996)
quantified the stiffness of rachis medulla as 1% of the stiffness of the
cortex such that kE=0.01. Thus, the range of values for
(F/sc)/(Fc/sc)
are 1.0043 to 1.012, which indicates that the variability in estimated
cortical strength due to the contribution of the medullary material is
1%. Such a small contribution of the medulla is unlikely to account for
the observed 25% decrease in estimated cortical strength at the most distal
positions along the rachis. Furthermore, the ratio of medullary
cross-sectional area to total cross-sectional area of barbs increased with
distance along the rachis up to a fractional distance of 0.88, which is in the
opposite direction needed to explain a decrease in estimated cortical strength
due to differential contribution of the medulla. So, changes in the
contribution of the medullary material cannot explain the decrease in cortical
strength that we observed. The observed decrease in strength of barbs towards
the distal end of the rachis must be due to changes in the material properties
of the cortex keratin.
Material property changes with position along the rachis have been
previously observed for feather cortex keratin. For example, the stiffness of
the cortex keratin from the rachis of mute swan (Cygnus olor) flight
feathers has been shown to increase by 100% from the proximal to the distal
end of the rachis (Bonser and Purslow,
1995). If we calculate a nominal breaking stiffness as breaking
stress over breaking strain, we find that there is a 37% increase in nominal
breaking stiffness of barbs located towards the distal end of the rachis.
Thus, the cortex of both rachis and barbs increases in stiffness towards the
distal end of the rachis. Such location-dependent material property changes in
stiffness, strength, breaking strain and toughness should be incorporated into
studies of feather structure and function.
Do material property changes of the more distal barbs affect flight
performance?
A small decrease in material strength of the barbs towards the distal end
of the rachis is unlikely to contribute much to differential function of barbs
during flight. This is because barbs are normally loaded in bending during
flight. The bending performance of barbs will be controlled by their flexural
stiffness, which is the product of stiffness (a material property) and the
second moment of area (a measure of the distribution of cortex material around
the neutral axis in the plane of bending). Bonser and Purslow
(1995) concluded that the
flexural stiffness of the primary feathers of the mute swan was principally
controlled by the second moment of area, despite a 100% increase in stiffness
along the rachis. Similarly, the consequences to bending performance of
changes in the second moment of area that occur in barbs along the length of
the rachis are likely to overwhelm the consequences of small changes in
material properties of those barbs.
Does bending performance of barbs differ from that of the
rachis?
When bent to failure (buckling), the deflection of barbs is relatively much
greater than that of the rachis. This phenomenon is perhaps best understood in
terms of the radius of curvature at failure. The radius of curvature ()
in a bent beam is:
![]() | (3) |
![]() | (4) |
We use the tensile rupture stress as a reference point even though both rachis and barbs fail by buckling. We do this because the buckling stress is not known and changes systematically with the wall thickness. As cylindrical beams become thicker walled, the buckling moment becomes equal to the moment at tensile rupture. We are interested here in evaluating mainly the effect of the relative slenderness of the beam and the relative curvature at failure. Thus, we use as a single reference point the tensile failure stress and acknowledge that in both barbs and feathers the buckling failure will occur at a somewhat variable lower value of stress and curvature.
We define relative curvature as the radius of curvature at failure divided
by the length of the beam, /L, and we define slenderness as the
ratio, L/h, of a beam's length to its effective height
(where effective height is the dimension perpendicular to the length but in
the plane of bending). These two dimensionless numbers can be plotted against
each other (Fig. 8), such that
the typical range of slenderness values is shown for the feather rachis and
barbs used in this study. At failure, barbs are relatively much more curved
than the rachis. Thus, barbs tend to avoid buckling failure by bending out of
the way of high forces, whereas the rachis is less able to bend sufficiently
to avoid high forces.
|
This effect of slenderness on flexibility is enhanced in the barbs because
they twist as they bend. Barbs are typically taller than wide
(Fig. 1;
Table 2), but when they twist,
the smaller dimension, the width, becomes the effective height, thus lowering
the second moment of area and allowing a smaller radius of curvature before
reaching a critical buckling stress. The tendency to twist has also been
observed to a lesser degree in the rachis of pigeon flight feathers
(Corning and Biewener, 1998).
In barbs, twisting persists even when groups of barbs are tested, despite the
increase in lateral stability provided by such groups (M.B. and A.S.J.,
personal observation).
Thus, if you assume that cortex keratin of barbs and rachis buckles at the
same stress, a smaller radius of curvature will be observed for the barbs at
that stress than will be observed for the rachis. For feather rachis, the
stress for buckling is less than the tensile rupture stress
(Corning and Biewener, 1998),
and buckling can generally be expected in thin-walled beams
(Brazier, 1927
). But as the
wall gets thicker, the buckling and rupture moments converge. Whereas the
pigeon rachis was thin walled
(t/
=0.081±0.0078;
Corning and Biewener, 1998
),
the osprey barbs were relatively thicker walled
(t/
=0.39±0.0046;
Table 2). The barbs will
therefore be likely to have a higher buckling stress, a stress approaching
that predicted by the tensile rupture of feather keratin.
One can estimate the relative wall thickness at which a beam should fail in
tensile rupture rather than in buckling. This calculation is done using the
equation for buckling moment (equation 4 in
Corning and Biewener, 1998),
the equation for the rupture moment (a rearrangement of equation 6 in
Corning and Biewener, 1998
;
not their equation 3, which contains a typographical error) and the definition
ri=
rro, where
r is the ratio of inner to outer radii. Then, using the same
values for Young's modulus, tensile failure stress, constant K and Poisson's
ratio as did Corning and Biewener
(1998
), the ratio of inner to
outer radii at which the buckling moment equals the rupture moment is
r=0.81. The ratio
r is related to the
t/
ratio by the following
formula:
![]() | (5) |
In contrast to this prediction, we observed that bent barbs failed in buckling. Buckling presumably occurred because the barbs twist during bending and the wall thickness in the plane of bending after this twisting is considerably less than the mean wall thickness of the barb. [See Fig. 1 to understand a change in the plane of bending due to twisting; bending dorso-ventrally (up-down in Fig. 1) becomes lateral bending after twisting.] It may be that twisting partly functions to prevent failure by tensile rupture, which would probably be more catastrophic to barb function than is buckling. Buckling usually leaves an intact but weakened barb and therefore a barb that still functions almost as well as before buckling. We propose the following design principle for barbs. By being thicker-walled dorso-ventrally (see Fig. 1), their flexural stiffness is increased during flight; but by allowing for twisting when loaded with dangerously high forces they firstly avoid failure by bending and secondly avoid complete failure by buckling rather than rupturing.
High flexibility and deformability may function to prevent breakage in
barbs, as it does in other systems. For instance, daffodil stems, which have
low torsional stiffness, twist to reduce drag in wind
(Etnier and Vogel, 2000).
Similarly, flexibility in terebellid polychaete tentacles (which deform;
Johnson, 1993
) and stipes of
kelp (which bend; Johnson and Koehl,
1994
) also functions to reduce drag. Finally, extreme
extensibility in viscid spider silk
(Denny, 1976
) and mussel
byssal threads (Bell and Gosline,
1996
) functions to avoid high forces by allowing deformation.
The role of melanin in signaling feather quality
Theories about the role of bird plumage in signaling feather quality
(Fitzpatrick, 1998) rest
critically on whether feather coloration accurately reflects mechanical
properties of the feather. To the extent that feather coloration is
unimportant to the mechanical function of feathers, it suggests that patterns
of melanized and unmelanized feather coloration evolved under selective
pressures, such as communication, counter-shading or thermoregulation,
different from those involved in the mechanical function of feathers. It may
be that the preference of feather-eating lice for unmelanized regions of
feathers compromises feather strength in some species of birds
(Kose and Møller,
1999
), and, perhaps, unbroken feathers with large white spots
signal the absence of feather-eating lice. However, while dark and light bar
patterns may indeed aid the perception of the extent of wear and damage, the
corollary that the absence of melanin facilitates feather wear needs to be
examined more rigorously.
Summary
Whereas our measurements of material properties on one feather from one
species cannot determine whether melanization contributes to strength or
hardness in bird species in general, our discovery of the importance of
position to material properties calls into question previous results in which
positional effects were not considered. Sampling of the tissues to be tested
must involve careful experimental design of the sampling scheme to adequately
account for effects of location and cross-sectional area. Such sampling must
be undertaken before it can be concluded that melanization functions to
increase hardness, toughness or strength of keratin-based structures in
birds.
![]() |
Acknowledgments |
---|
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Alexander, R. McN. (1996). Optima for animals. Revised edition. Princeton: Princeton University Press.
Barrowclough, G. F. and Sibley, F. C. (1980). Feather pigmentation and abrasion: test of a hypothesis. Auk 97,881 -883.
Bell, E. C. and Gosline, J. M. (1996).
Mechanical design of mussel byssus: material yield enhances attachment
strength. J. Exp. Biol.
199,1005
-1017.
Bonser, R. H. C. (1995). Melanin and the abrasion resistance of feathers. Condor 97,590 -591.
Bonser, R. H. C. (1996). The mechanical properties of feather keratin. J. Zool. Lond. 239,477 -484.
Bonser, R. H. C. and Purslow, P. P. (1995). The Young's modulus of feather keratin. J. Exp. Biol. 198,1029 -1033.[Medline]
Bonser, R. H. C. and Witter, M. S. (1993). Indentation hardness of the bill keratin of the European starling. Condor 95,736 -738.
Brazier, L. G. (1927). On the flexure of thin cylindrical shells and other "thin" sections. Proc. R. Soc. Lond. Ser. A 116,104 -114.
Burtt, E. H., Jr (1979). Tips on wings and other things. In The Behavioral Significance of Colour (ed. E. H. Burtt, Jr), pp. 75-125. New York: Garland STPM Press.
Burtt, E. H., Jr (1986). An analysis of physical, physiological and optical aspects of avian colouration with emphasis on wood-warblers. Ornithol. Monogr. 38, 1-126.
Corning, W. R. and Biewener, A. A. (1998).
In vivo strains in pigeon flight feather shafts: implications for
structural design. J. Exp. Biol.
201,3057
-3065.
Denny, M. W. (1976). The physical properties of spider's silk and their role in the design of orb webs. J. Exp. Biol. 65,483 -506.
Denny, M. W. (1988). Biology and the Mechanics of the Wave-Swept Environment. Princeton: Princeton University Press.
Etnier, S. A. and Vogel, S. (2000).
Reorientation of daffodil (Narcissus: Amaryllidaceae) flowers in wind: drag
reduction and torsional flexibility. Am. J. Bot.
87, 29-32.
Fitzpatrick, S. (1998). Birds' tails as signaling devices: markings, shape, length, and feather quality. Am. Nat. 151,157 -173.[CrossRef]
Gill, F. B. (1995). Feathers. In Ornithology (ed. F. B. Gill), pp.65 -92. New York: W. H. Freeman and Co.
Johnson, A. S. (1993). Sag-mediated modulated
tension in terebellid tentacles exposed to flow. Biol.
Bull. 185,10
-19.
Johnson, A. S. and Koehl, M. A. R. (1994).
Maintenance of environmental stress factor in different flow habitats: thallus
allometry and material properties of a giant kelp. J. Exp.
Biol. 195,381
-410.
Kose, M. and Møller, A. P. (1999). Sexual selection, feather breakage and parasites: the importance of white spots in the tail of the barn swallow (Hirundo rustica). Behav. Ecol. Sociobiol. 45,430 -436.[CrossRef]
Lee, D. S. and Grant, G. S. (1986). An albino greater shearwater: feather abrasion and flight energetics. Wilson Bull. 98,488 -490.
Vincent, J. (1990). Structural Biomaterials. Princeton: Princeton University Press.
Wainwright, S. A., Biggs, W. D., Currey, J. D. and Gosline, J. M. (1976). Mechanical Design in Organisms. Princeton: Princeton University Press.
Zar, J. H. (1996). Biostatistical Analysis. Third edition. New Jersey: Prentice-Hall.
Related articles in JEB: