Caldarone et al Nonlethal techniques for estimating responses of postsmolt Sa/mo salar to food availability 
261 
closely approximate the real electrical values of biologi- 
cal tissue (RJL Systems, http://www.rjlsystems.com/ 
docs/bia_info/principles/, accessed April 2008), only 
the parallel forms of the BIA measures are discussed 
and reported. 
Serial R and Xc values were transformed to their 
parallel equivalents with the following formulas: 
R par (Q) = R + Xc 2 / R, (2) 
Xc p JQ) = Xc + R 2 / Xc . (3) 
Because R and Xc are dependent upon the distance the 
current must travel ( D , distance between the electrodes 
in cm), the BIA instrument manufacturer advises that 
when these variables are used in prediction equations, 
the effect of this distance must be accounted for (RJL 
Systems, http://www.rilsvstems.com/docs/bia info/prin- 
ciples/ . accessed April 2008). We therefore also calcu- 
lated standardized R par and Xc par values by dividing 
K par ™*Xc par by D(R p JD,Xc p JD). 
Conductor volumes were calculated by using the fol- 
lowing formulas: 
R„ ar conductor volume = D 2 /R„„ r , (4) 
pdi pdi 
Xc conductor volume = Z) 2 /Xc .. . (5) 
pdr r 
Capacitance (a measure of the electrical storage ca- 
pacity) and impedance (a measure of the opposition to 
the flow of electrical current) were calculated using the 
following formulas: 
capacitance (pF) = 1 x 10 12 /(2 tt •50000»Xc par ), (6) 
impedance (Q) = sqrt((R par 2 ) + ( Xc par 2 )) , (7) 
where 50,000 is the frequency applied by the BIA instru- 
ment in Hertz. 
In order to conform to values previously reported in 
the literature, phase angles were calculated with Xc 
and R in their series form: 
phase angle (°) = arctan (Xc //? ) • 180/zr, (8) 
where Xc and R are the vertical and horizontal axis, 
respectively. The arctangent of the ratio will yield the 
angle of the impedance vector in radians, which is then 
converted to degrees by multiplying by 180/7T. A series- 
based phase angle is equal to 90° minus the parallel- 
based phase angle. 
Impedance measurements are negatively related to 
temperature (van Marken Lichtenbelt, 2001). Water 
temperature, room temperature, and internal fish tem- 
peratures (mean muscle temperature=13.4°C, SD = 0.9; 
mean stomach temperature = 12.6°C, SD = 1.0) were con- 
stant in our study and therefore not included as fac- 
tors in our analyses. Unless otherwise noted, the term 
“BIA measures” refers collectively to i? par , Xc par , R p .JD , 
Xc par /D, phase angle, /? par and Xc paT conductor volumes , 
capacitance, and impedance. 
Fulton's K 
Fulton’s condition factor (K) was calculated with the 
following formula (Ricker, 1975): 
K = 100 • WW/FL 3 , (9) 
where WW (in g) and FL (in cm) are values from the day 
the fish was sacrificed. 
Data analysis 
Within treatments A Dunnett two-tailed /-test with 
final FL as a covariate was used to detect changes 
in body composition (% WW), BIA measures, and K, 
within the fed and fasted treatments and the refed 
group. Baseline values (day 0) were specified as the 
control for all variables except BIA measures, where 
day-3 fed values were the specified controls. 
Between treatments A two-way multivariate analysis 
of covariance (MANCOVA) for unbalanced design 
was used to compare body composition (%WW) (total 
fat concentration, TF%; total water concentration, 
TWa%\ carcass protein concentration, CP%), BIA 
measures, growth rate, and K between the three 
feeding treatments (fed; fasted; fasted, then refed) 
and sampling times, with final FL as the covariate. 
When interactions were significant, feeding treat- 
ment was nested in day and follow-up comparisons 
were examined by using Tukey’s HSD multiple range 
tests. Because we found no significant differences 
in percent liver fat between any of the treatments 
or days, and liver fats comprised <1% of total fats 
(range: 0.39-0.99%), only total fat values were used 
in all analyses. 
Prediction models Prediction models for body compo- 
sition expressed as both content (total fat, TF\ TWci; 
carcass protein, CP), and concentration ( CP% , TF%, 
TWa%), and growth rate were developed. We used an 
information-theoretic approach for small sample sizes 
(Akaike’s information criterion, AICc) to select the 
“best-fit” models (Wagenmakers and Farrell, 2004). 
Because we had no prior knowledge of the variables 
or combination of variables that would be the best 
predictors of the dependent variables, all nine BIA 
measures plus the interaction of R par , and Xc with 
D were tested, along with WW, FL, and K. Testing 14 
independent variables generated a large number of 
models for each dependent variable, with many models 
significant at the PcO.0001 level. For brevity, only the 
top three most parsimonious models (as indicated by 
the smallest AICc values) are reported and discussed. 
Correlations Pearson product-moment correlations were 
used to investigate the relations between body compo- 
sition (both g and %WW), and WW, BIA measures, K, 
and growth rate. All statistical analyses were carried 
out with SAS software vers. 9.1 (SAS Inst., Inc., Cary, 
