early discrimination of atlantic salmon smolt age: time course of the relative effectiveness of body...

This article was downloaded by: [McGill University Library]On: 28 November 2014, At: 08:07Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American FisheriesSocietyPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/utaf20

Early Discrimination of Atlantic SalmonSmolt Age: Time Course of the RelativeEffectiveness of Body Size and ShapeJamie H. Pearlstein a b , Benjamin H. Letcher c & MariskaObedzinski a b da U.S. Geological Survey , Leetown Science Center, S.O. ConteAnadromous Fish Research Center , Post Office Box 796, OneMigratory Way, Turners Falls, Massachusetts, 01376, USAb Department of Natural Resources Conservation , University ofMassachusetts , Amherst, Massachusetts, 01003, USAc U.S. Geological Survey, Leetown Science Center, S.O. ConteAnadromous Fish Research Center , Post Office Box 796, OneMigratory Way, Turners Falls, Massachusetts, 01376, USAd University of California Cooperative Extension , 133 AviationBoulevard, Suite 109, Santa Rosa, California, 95403, USAPublished online: 09 Jan 2011.

To cite this article: Jamie H. Pearlstein , Benjamin H. Letcher & Mariska Obedzinski (2007) EarlyDiscrimination of Atlantic Salmon Smolt Age: Time Course of the Relative Effectiveness of BodySize and Shape, Transactions of the American Fisheries Society, 136:6, 1622-1632, DOI: 10.1577/T07-010.1

To link to this article: http://dx.doi.org/10.1577/T07-010.1

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or

http://www.tandfonline.com/loi/utaf20

http://www.tandfonline.com/action/showCitFormats?doi=10.1577/T07-010.1

http://www.tandfonline.com/action/showCitFormats?doi=10.1577/T07-010.1

http://dx.doi.org/10.1577/T07-010.1

howsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/page/terms-and-conditions

Early Discrimination of Atlantic Salmon Smolt Age:Time Course of the Relative Effectiveness of Body Size and Shape

JAMIE H. PEARLSTEIN

U.S. Geological Survey, Leetown Science Center, S. O. Conte Anadromous Fish Research Center,Post Office Box 796, One Migratory Way, Turners Falls, Massachusetts 01376, USA; and Department of

Natural Resources Conservation, University of Massachusetts, Amherst, Massachusetts 01003, USA

BENJAMIN H. LETCHER*U.S. Geological Survey, Leetown Science Center, S. O. Conte Anadromous Fish Research Center,

Post Office Box 796, One Migratory Way, Turners Falls, Massachusetts 01376, USA

MARISKA OBEDZINSKI

U.S. Geological Survey, Leetown Science Center, S. O. Conte Anadromous Fish Research Center,Post Office Box 796, One Migratory Way, Turners Falls, Massachusetts 01376, USA; Department of NaturalResources Conservation, University of Massachusetts, Amherst, Massachusetts 01003, USA; and Universityof California Cooperative Extension, 133 Aviation Boulevard, Suite 109, Santa Rosa, California 95403, USA

Abstract.—The goal of this study was to test the relative effectiveness of morphological measurements and

body size in predicting the smolt age of Atlantic salmon Salmo salar and to determine the time course of body

size and shape differences between smolt ages. Analyses were conducted on age-0 to age-2 fish that were

stocked as fry in the West Brook, Massachusetts and on laboratory-raised age-0 to age-1 fish. Using both

body size and shape, we could partition the age-0 fish collected during fall into future early or late smolts,

although the predictive ability of body shape was somewhat weaker than that of body size, especially in the

laboratory. Classification success averaged 81% (size) and 79% (shape) in the field and 85% (size) and 73%

(shape) in the laboratory. Despite differences in smolt age between the field and the laboratory, the relative

timing of growth rate differences between future early and late smolts was similar in the field and the

laboratory and peaked at 50–60% of development from fry to smolt. While body shape differed between early

and late smolts well before smoltification, it did not improve classification based on size alone.

Smoltification in salmon is a seasonal process

marked by changes in physiology, biochemistry,

morphology, and behavior that prepare anadromous

fish for migration and saltwater residence (Hoar 1976;

McCormick and Saunders 1987). These changes

include spring peaks in salinity tolerance, silvering of

the body, fin darkening, decreased condition factor

(Vanstone and Markert 1968), increased sensitivity to

external stimuli (Saunders and Henderson 1970),

changes in swimming, agonistic behaviors, and social

structure (Schreck 1981), and increased plasma growth

hormone and gill Naþ,Kþ-ATPase (enzyme number

3.6.1.36; IUBMB 1992) activity (McCormick et al.

1995).

Although the changes that occur during the parr–

smolt transformation are well documented, the specific

timing of the developmental decision to smolt remains

unclear (Nicieza et al. 1991). This timing is difficult to

determine because it is influenced both by endogenous

and environmental factors that direct future smolts to

adopt the smolt developmental trajectory well in

advance of the life history event (Thorpe et al. 1998).

There is evidence that to migrate in the spring, juvenile

salmon must meet population-specific thresholds (size

or nutritional requirements), and a fish’s ability to meet

such thresholds depends on the environmental (e.g.,

temperature and feeding conditions) and genetic

(inherited standard metabolic rate) conditions that

determine its opportunity for resource acquisition and

growth (Thorpe et al. 1998). Thus, these factors and

their interaction cause variation within and among

populations in the timing of the developmental decision

to smolt and, consequently, the age at smoltification

(Metcalfe and Thorpe 1990; Metcalfe 1998).

The number and size of fish smolting in a particular

year can be related to the number of returning adults

and therefore can be used in understanding the long-

term survival of local populations (Lundqvist et al.

1994). Smolt size is also positively correlated with

smolt age (Økland et al. 1993) and survival during

migration (Marschall et al. 1998). In addition, the

differential survival and reproduction of early smolts

* Corresponding author: [email protected]

Received January 11, 2007; accepted June 6, 2007Published online November 15, 2007

1622

Transactions of the American Fisheries Society 136:1622–1632, 2007American Fisheries Society 2007DOI: 10.1577/T07-010.1

[Article]

Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

(in most streams, those that smolt in spring at age 2)

and late smolts (those that smolt in spring at age 3)

affect the ability to predict adult returns to natal streams

and identify sources of freshwater and marine mortal-

ity. A technique that could be implemented earlier in

the year (e.g., the fall) to indicate an individual’s future

smolt status would refine smolt estimates and decrease

dependence on spring sampling, when water levels are

high and sampling methods are inefficient (Johnsson et

al. 1996). In addition, the timing of smolting is a vital

aspect of the study of the evolution of conditional

strategies and the means of strategy choice in salmon

(Bohlin et al. 1994).

Numerous studies have focused on predicting smolt

age, but predictive methods are inconsistent among

systems, unsuitable for field studies, or unable to

predict the timing of smolting and the number of

smolts to a level of accuracy useful for management

purposes. One common method for predicting smolt-

ing, for instance, is an individual’s position in the

autumn bimodal size frequency distribution, which

usually is initiated by late summer, about 9 months

prior to smolting. Although bimodality has been shown

to be distinct 6 months prior to smolting (Thorpe

1977), other data show that entrance into either mode

can occur as late as winter or can be a continuous

process (Kristinsson et al. 1985). Bimodality also may

be unclear, especially in rivers where fish grow slower

and smolt at an older age (Heggenes and Metcalfe

1991), or may be obscured by overlapping sizes of

upper- and lower-mode fish (Saunders et al. 1994) or

by other biotic and abiotic factors influencing growth

(Heggenes and Metcalfe 1991).

Physiological and endocrinological changes that

occur during smolt transformation, such as increased

gill Naþ,Kþ-ATPase activity (McCormick 1993),

plasma thyroxine levels, salinity tolerance, and

decreased lipid content (Fessler and Wagner 1969)

also have been used as markers for smolting.

Although not all of these methods require sacrificing

the animal to obtain samples, they do involve complex

laboratory assays or experiments and typically are not

diagnostic until late March or early April, just prior to

smolting.

Some investigators have used behavioral means to

predict future smolting (Metcalfe et al. 1989; Thorpe et

al. 1992). Social status, along with length (Metcalfe et

al. 1989) or growth rate (Thorpe et al. 1992), has been

used to predict smolt age. It also has been suggested

that decreased swimming proficiency (as measured by

increased tail beat frequency) could be a predictor of

smolt status (Flagg and Smith 1982). Behavioral

measurements, however, are better suited for laboratory

than field settings.

Morphological techniques, which are well suited for

field research and studies with threatened or endan-

gered species (because sacrificing specimens is not

necessary) are a potential alternative. Numerous studies

have measured the morphological changes correlated

with smolting (Fessler and Wagner 1969; Riddell and

Leggett 1981; Winans 1984; Beeman et al. 1995) and

found that smolts develop longer caudal regions,

shallower bodies, and larger heads than fish that do

not emigrate (Beeman et al. 1995; Nicieza 1995).

Morphological differences between salmon of hatchery

and natural origin have also been observed (von

Cramon-Taubadel et al. 2005). Other studies identified

morphological changes in early development before the

perceived onset of the parr–smolt transformation. A

study of the early morphometric development in

hatchery-reared Chinook salmon (Oncorhynchus tsha-wytscha) detected changes in body slenderness during

smoltification and suggested the use of these changes

for predicting the release time of hatchery-reared fish

and maximizing survival and return (Winans 1984).

Similarly, morphometric variables had an average

success rate of 67% in classifying Atlantic salmon

Salmo salar as smolts from their age-1 summer until

smolting at age 2 (Letcher 2003), but the morphometric

approach was not compared to predictions based on

size alone.

Although shape variation has the potential to help

discriminate between early and late smolts, the relative

effectiveness of shape and the traditionally used

variable—body size—has not been evaluated. The

present analysis examines the use of morphometric

measurements created from a truss network to predict

timing of smolting. Body measurements were collected

from digital photographs taken seasonally of tagged

Atlantic salmon reared in field and laboratory settings,

and their ability to predict smolt age at various times

throughout ontogeny was tested. The predictive ability

of these morphometric variables was then compared

with that of fork length.

Methods

Field and laboratory studies were used to compare

the ability of shape and size to predict Atlantic salmon

smolt age (age of smolting in the spring). Both groups

were derived from adult sea-run Atlantic salmon in the

Connecticut River. Owing to differing growth condi-

tions in the field and laboratory, fish in the field usually

first smolt at age 2, while those in the laboratory first

develop smolt characteristics at age 1. Early smolts are

defined here as age-2 smolts (field) or age-1 smolts

(laboratory), and late smolts are defined as fish that did

not smolt at age 2 (field) or age 1 (laboratory).

Different criteria were used to assign smolt age in the

ATLANTIC SALMON SMOLT AGE PREDICTION 1623

Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

field and laboratory. In the field, fish that were caught

in the smolt trap were considered early smolts and

those captured in the study area after the smolt run

were coded as late smolts. In the laboratory, smolt age

was assessed via a series of factors (see below), one of

which was fork length at the final sampling occasion.

Because the difference in fork length between early and

late smolts is well established at the time of smolting,

we assumed that the use of fork length to assess smolt

status at the last sampling occasion did not conflict

with the use of fork length in the present analysis as a

predictor of future smolt age.

Study area.—This study was conducted as part of a

long-term mark–recapture study of a small third-order

stream (West Brook) in western Massachusetts within

the Connecticut River basin (428250N, 728390W). The

1-km long study reach is stocked with Atlantic salmon

fry (26–28 mm fork length) each spring at a density of

50 fish/100 m2 and sampled four times per year,

beginning in autumn. Stocked fry were F1

offspring of

adult returns to the Connecticut River. Individual

tagging combined with multiple sampling provided a

means to follow a fish throughout its freshwater

residence and to retrospectively classify it as a parr

or smolt for each sampling occasion.

Atlantic salmon were collected via electroshocking

(400-V unpulsed DC). Fish were also sampled via a

smolt trap (picket weir) assembled 3 km downstream of

the study site in the spring of 2001. Each fish was

anesthetized with NaHCO3-buffered tricaine methane-

sulfonate (MS-222; 100 mg/L) and measured for fork

length (61 mm) and wet mass (60.1 mg). All

untagged fish greater than 60 mm fork length and 2 g

were tagged with passive integrated transponder tags

(12 mm; Destron-Fearing Corp.) through a small

incision between the pectoral fins (Gries and Letcher

2002), allowing for individual identification. Each fish

was then aligned on a straight line, and a digital

photograph of the left side was taken at a fixed

distance. After recovery, each fish was returned to its

location of capture. Further details of the study stream

and fish capture are described by Letcher et al. (2002)

and Letcher and Gries (2003).

The 573 field photographs used in this study were of

Atlantic salmon from the 1999 stocking year and were

taken on eight sampling occasions between September

1999 (age 0) and May 2001 (age 2; Table 1). For

analyses, the data from the 2001 smolt trap sampling

occasion (containing only early smolts) and the

subsequent sampling occasion that occurred 1 month

later (containing only future late smolts) were com-

bined.

Laboratory methods.—The photographs of labora-

tory fish were taken during a study conducted by

Obedzinski and Letcher (2004) in which Atlantic

salmon from the Connecticut River stock were raised

from eyed eggs to age-1 smolts in a controlled

environment. Eggs were obtained from the U.S. Fish

and Wildlife Service’s White River National Fish

Hatchery (White River, Vermont). Broodstock were F1

offspring of sea-run adult returns to the Connecticut

River. Eggs were kept in darkness except during

sampling periods, and artificial light was timed based

on natural photoperiod after hatching. Once hatched,

fish were fed a combination of commercial dry feed

and nauplii of the brine shrimp Artemia franciscana.

Fish were sampled monthly from October 2002 (age

0) to June 2003 (age 1; Table 1). On every sampling

occasion, each fish was anesthetized with MS-222,

measured for fork length (6 1 mm) and wet mass (6 1

mg), and photographed on its left side at a fixed

distance. Individuals were assigned to life history

group (early or late smolt) based on size, condition

factor, and a morphological rating of silvering and

darkening of fin margins at the last sample. We

analyzed 953 pictures taken over the nine sampling

occasions (Table 1).

Morphometric and statistical analyses.—Linear

distance measurements taken from digital photographs

served as variables in two multivariate analyses,

principal components analysis (PCA) and discriminant

analysis (DA). The aim of PCA was to identify and

TABLE 1.—Number of Atlantic salmon parr and smolts

photographed digitally for morphometric analysis to predict

early or late smoltification. Sampling environments were West

Brook, Massachusetts (field) and the laboratory.

Sample date Age Smolts Parr Total

Field

1999Sep 0 13 58 71

2000Mar 1 20 57 77May 1 24 75 99Sep 1 24 74 98Dec 1 21 76 55

2001Mar 2 19 39 58May 2 42 52 94

Laboratory

2002Oct 0 71 48 119Nov 0 70 49 119Dec 0 71 49 120

2003Jan 1 71 49 120Feb 1 71 49 120Mar 1 71 49 120Apr 1 71 49 120May 1 71 49 120Jun 1 66 49 115

1624 PEARLSTEIN ET AL.

Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

describe patterns of variation between early and late

smolts present at each sampling occasion. To determine

the influence of size and shape on defining the

principal components (PCs), we calculated Pearson’s

product-moment correlations between the first three

PCs and fork length or wet mass at each sampling

occasion. The PCs that had a strong correlation with

length, mass, or both were characterized by size,

whereas those that lacked a correlation were considered

nearly free of its influence and were characterized by

shape (Haas and McPhail 2001).

The purpose of DA was to identify the variables that

best described the differences between early and late

smolts at each sampling occasion and to test the

variables’ ability to predict group membership (early or

late smolt). Three different DAs (described below)

were conducted to determine the relative discriminato-

ry power of size and shape.

Morphometric measurements.—To analyze morpho-

metric variation, 20 linear distance measurements were

used based on the box truss protocol (Bookstein et al.

1985; Figure 1). These network truss distances

consisted of a series of measurements computed

between 10 landmarks (Figure 1) that form an array

of contiguous quadrilaterals across the body form

(Strauss and Bookstein 1982). The landmarks were

digitized on each photograph using tpsDig software

(Rohlf 2001). To reduce skewness and kurtosis, loge

transformation was applied to all 20 distance measure-

ments. The distance measurements were divided into

head, body, and tail regions (Figure 1).

Fork length and growth rate analyses.—The mean

fork lengths and mean growth rates of early and late

smolts at each sampling occasion were calculated to aid

in explaining the trends in the discriminatory power of

shape and size over time. To determine the growth

rates for early and late smolts, we first performed a

linear regression to obtain the residual length between

time intervals for every fish caught on consecutive

sampling occasions. The mean residual of early and

late smolts for each time interval was then calculated

and plotted. We also calculated standardized lengths

(observed length minus average length at each

sampling occasion) to allow comparison between

laboratory and field studies.

Paired t-tests and the nonparametric alternative, the

Mann–Whitney U-test, were used to evaluate the

differences in early and late smolt mean standardized

fork lengths and mean growth rate residuals for each

sampling occasion. The normality and homogeneity of

variance assumptions of the t-test were evaluated via

the Shapiro–Wilk W statistic and Levene’s test

(Statistica 6.1).

Principal components analysis.—Body form vari-

ability between early and late smolts was assessed

using 16 (7 field and 9 laboratory) pooled-group PCAs,

one per sampling occasion. The PCA was performed

on the variance–covariance matrix of the 20 log-

transformed truss distances. The significance of the

eigenvalues and eigenvector coefficients was evaluated

using the jackknifed standard errors and coefficient

error ratios (Gibson et al. 1984).

Discriminant analysis.—Discriminant analysis was

used to classify fish as either early or late smolts based

on morphometric variables, fork length, growth rate, or

a combination of these. The null hypothesis was that

the truss measurements or fork length for early and late

smolts at each sampling occasion were equal (i.e., the

mean discriminant scores did not differ between groups

within any sampling occasion). All DAs were per-

formed using the Statistical Analysis System version

8.02. Prior probabilities were set proportional to group

sample sizes, and morphometric variables were select-

ed at each sampling occasion by the forward stepwise

procedure. The cross-validation (jackknife) procedure

was used to test the stability of the discrimination

functions.

To compare the relative importance of size and

shape variables as discriminators between life histories,

classification success was compared among three

different DAs at each sampling occasion. The first set

of DAs classified fish at each sampling occasion based

on their shape only. This was achieved by applying

Burnaby’s (1966) size adjustment procedure to the log-

transformed truss variables, which allowed for dis-

crimination based on size- or age-related shape

differences alone (Klingenberg 1996). The second set

of DAs was performed using only the fork length of

each individual at each sampling occasion. Fork length

is used here as a measure of size. The third set of DAs

was conducted using the log-transformed truss vari-

ables without Burnaby’s size-adjustment; this allowed

a comparison of discrimination based on both shape

and size.

For each sampling occasion, a one-way analysis of

variance (ANOVA) was used to test for differences in

classification success rate among the three sets of DAs.

FIGURE 1.—Landmark locations and morphometric distanc-

es (lines) used in shape analysis for the purpose of predicting

smolt age among Atlantic salmon.


Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

To test for the violation of assumptions of parametric

analysis, the Shapiro–Wilk W-test (normality) and

Levene’s test (homogeneity of variances) were used.

Contrasts for least-squares means were used as a post

hoc test. Due to the violation of assumptions, the

Kruskal–Wallis test was used as a nonparametric

alternative to the one-way ANOVA. Differences with

P-values less than 0.05 were considered significant.

Analyses were performed using Statistica version 6.1.

Results

Fork Lengths and Growth Rates

The average fork lengths of early smolts were

significantly larger (P , 0.001) than those of late

smolts in the laboratory and field on all but the first

field sampling occasion (September 1999; P ¼ 0.469;

Figure 2a, b). Growth rates were significantly higher (P, 0.05) for early smolts than for late smolts at all field

sampling intervals except for March–May 2000 and

2001 (Figure 3a). Early smolts also grew significantly

faster (P , 0.05) than late smolts in the laboratory at

all but the last two intervals (April–May and May–

June; Figure 3b).

In the field, growth rates started to diverge between

early and late smolts in April 2000 (age 1; Figure 3a),

11 months before the largest difference in lengths

(March 2001; age 2; Figure 2a) and 13 months prior to

smolt migration (May 2001). In the laboratory, growth

rates started to diverge in October 2002 (age 0; Figure

3b), 7 months before the largest difference in lengths

(May 2003; age 1; Figure 2b) and 8 months prior to

smolting (June 2003). Although the peak difference in

growth rates occurred in November for laboratory fish

(Figure 3b) and midsummer in the field, the peak

occurred at about the same percentage of days to

smolting (51% in laboratory, 61% in the field) as did

the largest difference in size (85% in the laboratory,

90.5% in the field; Figure 4).

FIGURE 2.—Mean fork lengths for early (solid lines) and late

(dashed lines) Atlantic salmon smolts (a) collected in the field

(West Brook, Massachusetts; 1999–2001) and (b) held in a

laboratory (2002–2003). The vertical lines indicate 95%confidence intervals.

FIGURE 3.—Mean growth rate residuals of early (solid lines)

and late (dashed lines) Atlantic salmon smolts (a) collected in

the field (1999–2001) and (b) held in a laboratory (2002–

2003). The vertical lines indicate 95% confidence intervals.


Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

Principal Components Analysis

Because size effects were not removed prior to

conducting the PCAs, the first PC (PC1) contained a

combination of size and size-related shape variation, as

reflected by the relatively large and positive loadings

and large eigenvalues. The identification of the PC1

values as size components and PC2 and PC3 as shape

components was reinforced by the highly significant (P

, 0.01) correlations of Atlantic salmon fork length

(Table 2) and wet mass (data not shown, but similar to

length data) with PC1 and their lack of correlation with

PC2 and PC3.

Field.—Averaged over the seven sampling occa-

sions, the first three PCs accounted for 79% of the total

variation in the data set. Principal component 1

explained the maximum amount of the variation

(64.0%), and PC2 (8.7%) and PC3 (6.3%) explained

considerably less of the remaining variation (Table 3).

The large amount of variation accounted for by PC1

and the lack of variation accounted for by PC2 are

reflected in the average PC scores for early and late

smolts (Figures 5a, 6a). Among these three PCs, size

(64.0%) accounted for four times more variation than

did shape (14.9%).

The results of the jackknife procedure for PC1

showed that, on average, 19.8 (99%) of the 20

coefficients of PC1 had significant coefficient error

ratios (using the conservative ratio of 3.0; 0.01 , a ,

0.001). For PC2, an average of only 3.8 (19%) of the

coefficients had coefficient error ratios greater than 3.0.

The characterization of PC2 in general was based on

the number and type of significant characters for PC2

at all sampling occasions combined. The number of

significant characters was fairly balanced among head

(10, or 20%), body (10, or 24%), and tail (7, or 14%)

characters, but almost all significant characters varied

over time. The three earliest sampling occasions

contained 74% of the significant characters and all of

the significant body characters. For PC3, an average of

only two (11%) of the coefficients had coefficient error

ratios greater than 3.0. These results indicate that most

of the shape variation was described by PC2 and that

shape change occurred in all three body regions and

occurred relatively early in development (but see

comment below on PC stability).

The jackknife estimates of the eigenvalues and their

errors indicated that PC1 was a stable vector but that

FIGURE 4.—Mean growth rate residual and standardized

fork length ratios in relation to the percent time to smolting of

Atlantic salmon collected in the field (1999–2001) and held in

a laboratory (2002–2003).

TABLE 2.—Correlations between Atlantic salmon fork

length and principal components (PCs) 1–3, representing size

and shape variation. All correlations for PC1 were significant

(P , 0.01); none were significant PC2 or PC3 (P . 0.05).

Sample date Age PC1 PC2 PC3

Field

1999Sep 0 0.94 �0.13 0.01

2000Mar 1 0.93 0.13 �0.02May 1 0.96 �0.05 �0.06Sep 1 0.94 �0.10 �0.05Dec 1 0.97 �0.07 �0.03

2001Mar 2 0.76 �0.19 �0.04May 2 0.93 �0.13 �0.04

Laboratory

2002Oct 0 0.88 0.15 0.04Nov 0 0.93 �0.06 0.02Dec 0 0.94 0.06 �0.08

2003Jan 1 0.99 0.03 �0.04Feb 1 0.98 �0.01 �0.01Mar 1 0.99 �0.02 �0.04Apr 1 0.98 �0.02 0.03May 1 0.98 0.06 �0.01Jun 1 0.96 0.01 �0.07

TABLE 3.—Proportion of variance between Atlantic salmon

exhibiting early and late smoltification in the field (West

Brook, Massachusetts), as explained by principal components

(PCs) 1–3, which represent size and shape variation.


1999Sep 0 0.474 0.125 0.087

2000Mar 1 0.573 0.093 0.084May 1 0.675 0.085 0.051Sep 1 0.690 0.051 0.048Dec 1 0.710 0.068 0.053

2001Mar 2 0.740 0.072 0.046May 2 0.621 0.115 0.069

Average 0.640 0.087 0.063


Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

PC2 and PC3 were not. The first eigenvalue was

separate from the second and third at all sampling

occasions; except for September 1999 and May 2000,

there was overlap between the second and third

eigenvalues, which suggests that PC2 and PC3 were

unstable (Gibson et al. 1984) and thus should be

interpreted cautiously.

Laboratory.—As with the field samples, the major-

ity of the variation in the laboratory data set was

explained by PC1 (73.5%) and less by PC2 (7.5%) and

PC3 (5.0%; Table 4). The large amount of variance

accounted for by PC1 and the lack of variation

accounted for by PC2 are reflected in the average PC

scores for early and late smolts (Figures 5b, 6b).

Among these three PCs, size (73.5%) accounted for

four times more variation than did shape (12.5%).

The results of the jackknife procedure showed that,

on average, 19.8 (99%) of the 20 coefficients of PC1

had significant coefficient error ratios (using the

conservative ratio of 3.0; 0.01 , a , 0.001). For

PC2, 30% of the jackknifed coefficients were signif-

icantly different from zero and 81% of those were tail

characters. The pattern of significance was almost the

same at all times except for the second sampling

occasion, when only head characters were significant.

The jackknife procedure showed that the average

number of significant coefficients found in PC3 was

half that of PC2 (15.6%) and that significance was split

between head (50%) and tail (43%) characters.

The jackknife estimates of eigenvalues and their

errors indicated that PC1 was a stable vector but that

PC2 and PC3 were not. The first eigenvalue was

separate from the second and third at all samples and,

except in March and May of 2003, there was overlap

between the second and third eigenvalues.

Discriminant Analysis

Field.—The average classification success rates of

the three predictors in the field were 81% for fork

length, 79% for shape, and 84% for shape and size.

FIGURE 5.—Mean principal component 1 (PC1) scores for

early (solid lines) and late (dashed lines) Atlantic salmon

smolts (a) collected in the field (1999–2001) and (b) held in a


FIGURE 6.—Mean principal component 2 (PC2) scores for

early (solid lines) and late (dashed lines) Atlantic salmon

smolts (a) collected in the field (1999–2001) and (b) held in a



Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

Length’s predictive ability decreased on the last

sampling occasion from 91% to 73% classification

success (Figure 7a). This is probably because the final

sampling occasion contained fish that were captured on

two separate occasions and combined for the DA.

Future (age 3 in the subsequent spring) late smolts were

sampled in June, 1 month after sampling of the early

age-2 smolts and therefore had more opportunity to

grow, thus artificially decreasing the difference in size

between early and late smolts. In contrast, shape

variables performed well on the last sampling occasion

(96% classification success) because morphological

differences between early and late smolts were greatest

at this time and were apparent by visual inspection.

Fork length (82% classification success), shape (80%),

and shape and size (82%) could all be used to

distinguish between early and late smolts early in

development (September 1999; age 0; Figure 7a).

The prior probabilities were set proportional (rather

than equal) to group sample sizes so that classification

would be weighted by early and late smolt numbers in

the field and laboratory. However, because late smolts,

on average, were three times more numerous than early

smolts in the field (except for in May 2001),

classification error rates were much lower for late

smolts (4.5%) than for early smolts (62%).

The ANOVAs revealed that there were no signifi-

cant differences among classification success rates until

the last sampling occasion (P , 0.00001; age 2), when

the discrimination by fork length was significantly

lower (73%) than that by either shape (96%) or the

combination of shape and size (97%; Figure 7a).

Laboratory.—The average classification success

rates of the three predictors in the laboratory were

85% for fork length, 73% for shape, and 86% for shape

and size. As with the field data, all three DAs could

discriminate between early and late smolts early in

development (October 2002; age 0; Figure 7b).

Using ANOVA, significant differences in classifica-

tion rates were found at all but the first two sampling

occasions. The planned comparisons revealed that from

December 2002 (age 1) to May 2003 (age 2), the

classification success rates were significantly lower

when shape was used as a predictor then when either

fork length or shape and size were used as predictors.

In June 2003 (age 2), however, fork length resulted in a

significantly lower classification success rate than the

other two discriminant functions (Figure 7b).

Discussion

In this study, we have shown that both body size and

shape can be used to discriminate early from late

smolts in the fall at age 0. Size and shape together

performed as well as size alone, indicating that a large

portion of the early differences between early and late

smolts could be attributed to size differences, but shape

TABLE 4.—Proportion of variance between Atlantic salmon

exhibiting early and late smoltification in the laboratory, as

explained by principal components (PCs) 1–3, which

represent size and shape variation.


2002Oct 0 0.381 0.178 0.118Nov 0 0.496 0.143 0.112Dec 0 0.695 0.086 0.055

2003Jan 1 0.842 0.052 0.033Feb 1 0.854 0.039 0.024Mar 1 0.877 0.036 0.019Apr 1 0.884 0.033 0.022May 1 0.846 0.040 0.024Jun 1 0.738 0.068 0.045

Average 0.735 0.075 0.050

FIGURE 7.—Rates of jackknife classification success in

predicting early or late smoltification based on size, shape, or

both in Atlantic salmon (a) collected in the field (1999–2001)

and (b) held in a laboratory (2002–2003). Asterisks denote

significant differences within a sampling occasion (0.001 , P, 0.05*; P , 0.0001**).


Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

by itself also allowed correct classification. In fact, in

the field there was no difference in classification

success between shape alone and size alone. In the

laboratory, however, size alone consistently outper-

formed shape alone. Interestingly, the relative timing of

early differences in size and growth were very similar

between early and late smolts for both the field and

laboratory, despite different smolt ages (age 1 in the

laboratory; age 2 in the field). Overall, our results

indicate that size and shape differences are expressed

long before smoltification and that shape accounted for

relatively more of the difference between early and late

smolts in the field than in the laboratory.

The timing of the growth rate differences between

early and late smolts in this study suggests that some

fish increased their energy intake in spring, which may

have resulted in their attainment of a summer

emigration threshold (Metcalfe and Thorpe 1990;

Thorpe et al. 1998). Differences in growth rate among

field fish increased in April (age 1) and peaked in July,

perhaps because appetite was declining in some

individuals (parr that would not smolt at age 2) but

was sustained in others (age-2 smolts; Metcalfe 1998).

Laboratory fish also may have met an early emigration

threshold because the differences in growth rate peaked

at a similar percentage of time to smolting for

laboratory fish and field fish, despite faster average

growth rates in laboratory fish and a different smolt

age. The similarity in the relative timing of growth rate

and size patterns suggests that fish in the laboratory and

field are following similar developmental courses.

Although the timing of the peaks in growth rate in

both settings points to a summer emigration threshold,

the significant size differences that occurred early in

development in the field (age 1 in March, at 42% of

time to smolting) and laboratory (age 0 in October, at

40% of time to smolting) suggest that a developmental

decision was made prior to midsummer. Similarly,

Letcher and Gries (2003) found early differences in

mass (at age 0 in December onward) between West

Brook parr that smolted at age 2 and those that did not.

Likewise, the early size differences in the present study

imply that fish entered the early smolt trajectory before

the summer prior to smolting. If an emigration switch

does occur earlier in development, then perhaps a

second threshold during the summer before smolting

either prevents or promotes the loss of freshwater

adaptation (Thorpe et al. 1988).

In contrast to the relatively early peaks in growth

rate that resulted in size differences early on,

differences in shape did not peak until spring. One

reason shape differences were not as evident until the

parr–smolt transformation, rather than earlier in stream

residence, may be the influence of seasonal physiolog-

ical changes on growth and, hence, shape. After an

increase in day length, circulating salmon growth

hormone (GH) usually increases in smolting salmon

(McCormick et al. 1995), resulting in an increase in

metabolism and skeletal (length) growth (for review of

salmon GH, see Bjornsson 1997). The decrease in total

lipids (Vanstone and Markert 1968; Fessler and

Wagner 1969) and increase in skeletal growth probably

results in a leaner and more streamlined body, which is

reflected in the decrease in condition factor (100 3

[weight/length3]; Vanstone and Markert 1968; Fessler

and Wagner 1969; Saunders and Henderson 1970) and

disproportionately fast growth in caudal peduncle

length (Winans 1984; Winans and Nishioka 1987)

during the parr–smolt transformation. Hence, changes

in body morphology may be tightly linked to the

hormonal changes affecting growth during the parr–

smolt transformation.

There is also evidence that the initiation of some

smolt characteristics (weight–length relationships, sil-

vering, and salinity tolerance) is size dependent, which

may influence survival probabilities. Treatment with

GH, for instance, has been shown to reduce antipred-

ator behavior in brown trout S. trutta (Johnsson et al.

1996); thus, the requirement that fish reach a certain

size before GH levels increase would not only limit the

amount of time they might be behaviorally compro-

mised but also restrict the release of GH to a period

when the fish are larger and less susceptible to

predation. Furthermore, the GH-induced increase in

skeletal growth that occurs during smolting may be of

benefit later in life, since vertebrae are an important

source of calcium during the spawning migration

(Kacem et al. 1998). Larger smolts have greater

osmoregulatory ability in saltwater (reviewed in Mc-

Cormick and Saunders 1987), experience less preda-

tion, and have higher return rates than smaller smolts

(reviewed in Marschall et al. 1998).

The presence of size-dependent and seasonal

physiological changes helps explain why shape

changes during smolt metamorphosis, but it does not

explain the shape differences that occur earlier in

development. Even though the shape variables had the

most predictive success during smolt metamorphosis,

they had an average of 76% (field) and 70%(laboratory) classification success prior to peak dis-

crimination. Based on our results, it appears that shape

changes occur throughout freshwater residence, but

these are smaller in magnitude than those present

during smoltification.

We observed that the strength of shape differences

between early and late smolts throughout the course of

our tagging studies was stronger in the field than in the

laboratory. The primary differences between the field


Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

and laboratory studies were habitat, food source, and

developmental rate (age-2 versus age-1 smolts). The

similarity of differences in time-adjusted growth

between field and laboratory suggests that the com-

pressed smolt development cannot account for weaker

shape differences in the laboratory. It is possible that

food differences (stream drift and benthos versus dry

pellet) could influence shape variation. Fish fed

pelleted food tend to have a higher fat content (Fleming

et al. 1994), but it is unknown whether this could

mediate shape differences between smolt ages. Habitat

seems the most likely explanation for stronger smolt

age differences in the field. Shape differences are

common between hatchery and wild populations

(Swain et al. 1991; Fleming et al. 1994; von Cramon-

Taubadel et al. 2005). Based on common environment

studies, shape differences between habitats seem to

arise mainly from phenotypic plasticity (e.g., Swain et

al. 1991; von Cramon-Taubadel et al. 2005).

From a practical standpoint, size is much easier to

measure than shape. Our results suggest that shape

adds little to the discriminatory power of size alone;

therefore, shape measurements will not add to

predictive ability except at the time of smolting.

Despite this, it is intriguing that shape differences

were expressed well before smoltification. It will be

interesting to determine with future studies whether

early shape differences reflect variation resulting from

growth rate differences alone or from early expression

of future life history differences. Finally, if early shape

differences are adaptive in the field, the less-distinct

shape differences in the laboratory may argue against

smolt stocking and for stocking at an earlier life stage.

Acknowledgments

We thank Steven Cadrin of the National Marine

Fisheries Service, whose generosity with his time and

expertise helped to move this project forward. We wish

to extend thanks to all the people who worked many

hours in the laboratory and field to collect these data,

especially Aimee Varady, Todd Dubreuil, Gabe Gries,

and Matt O’Donnell.

References

Beeman, J. W., D. W. Rondorf, M. E. Tilson, and D. A.

Venditti. 1995. A nonlethal measure of smolt status of

juvenile steelhead based on body morphology. Transac-

tions of the American Fisheries Society 124:764–769.

Bjornsson, B. T. 1997. The biology of salmon growth

hormone: from daylight to dominance. Fish Physiology

and Biochemistry 17:9–17.

Bohlin, T., C. Dellefors, and U. Faremo. 1994. Probability of

first sexual maturation of male parr in wild sea-run brown

trout (Salmo trutta) depends on condition factor 1 year in

advance. Canadian Journal of Fisheries and Aquatic

Sciences 51:1920–1926.

Bookstein, F. L., B. Chernoff, R. L. Elder, J. M. Humphries,

Jr., J. R. Smith, and R. E. Strauss. 1985. Morphometrics

in evolutionary biology. Academy of Natural Sciences of

Philadelphia, Special Publication 15, Philadelphia.

Burnaby, T. P. 1966. Growth-invariant discriminant functions

and generalized distances. Biometrics 22:96–110.

Fessler, J. L., and H. H. Wagner. 1969. Some morphological

and biochemical changes in steelhead trout during parr–

smolt transformation. Journal of the Fisheries Research

Board of Canada 26:2823–2841.

Flagg, T. A., and L. S. Smith. 1982. Changes in swimming

behavior and stamina during the smolting of coho

salmon. Pages 191–195 in E. L. Brannon and E. O. Salo,

editors. Salmon and trout migratory behavior symposium.

School of Fisheries, University of Washington, Seattle.

Fleming, I. A., B. Jonsson, and M. R. Gross. 1994. Phenotypic

divergence of sea-ranched, farmed, and wild salmon.

Canadian Journal of Fisheries and Aquatic Sciences

51:2808–2824.

Gibson, A. R., A. J. Baker, and A. Moeed. 1984.

Morphometric variation in introduced populations of

the common myna (Acridotheres tristis): an application

of the jackknife to the principal component analysis.

Systematic Zoology 33:408–421.

Gries, G., and B. H. Letcher. 2002. Tag retention and survival

of age-0 Atlantic salmon following surgical implantation

with passive integrated transponder tags. North American

Journal of Fisheries Management 22:219–222.

Haas, G. R., and J. D. McPhail. 2001. The post-Wisconsinan

glacial biogeography of bull trout (Salvelinus confluen-tus): a multivariate morphometric approach for conser-

vation biology and management. Canadian Journal of

Fisheries and Aquatic Sciences 58:2189–2203.

Heggenes, J., and N. B. Metcalfe. 1991. Bimodal size

distributions in wild juvenile Atlantic salmon populations

and their relationship with age at smolt migration.

Journal of Fish Biology 39:905–907.

Hoar, W. S. 1976. Smolt transformation: evolution, behavior,

and physiology. Journal of the Fisheries Research Board

of Canada 33:1234–1252.

IUBMB (International Union of Biochemistry and Molecular

Biology). 1992. Enzyme nomenclature 1992. Academic

Press, San Diego, California.

Johnsson, J. I., E. Petersson, E. Jonsson, B. T. Bjornsson, and

T. Jarvi. 1996. Domestication and growth hormone alter

antipredator behaviour and growth patterns in juvenile

brown trout, Salmo trutta. Canadian Journal of Fisheries

and Aquatic Sciences 53:1546–1554.

Kacem, A., F. J. Meunier, and J. L. Bagliniere. 1998. A

quantitative study of morphological and histological

changes in the skeleton of Salmo salar during its

anadromous migration. Journal of Fish Biology

53:1096–1109.

Klingenberg, C. P. 1996. Multivariate allometry. Pages 23–49

in L. F. Marcus, M. Corti, G. J. P. Naylor, and D. E.

Slice, editors. Advances in morphometrics. Plenum, New

York.

Kristinsson, J. B., R. L. Saunders, and A. J. Wiggs. 1985.

Growth dynamics during the development of bimodal


Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

length-frequency distribution in juvenile Atlantic salmon

(Salmo salar L.). Aquaculture 45:1–20.

Letcher, B. H. 2003. Life-history-dependent morphometric

variation in stream-dwelling Atlantic salmon. Oecologia

442:533–540.

Letcher, B. H., and G. Gries. 2003. Effects of life history

variation on size and growth in stream-dwelling Atlantic

salmon. Journal of Fish Biology 62:97–114.

Letcher, B. H., G. Gries, and F. Juanes. 2002. Survival of

stream-dwelling Atlantic salmon: effects of life history

variation, season, and age. Transactions of the American

Fisheries Society 121:838–854.

Lundqvist, H., S. McKinnell, H. Faengstam, and I. Berglund.

1994. The effect of time, size, and sex on recapture rates

and yield after river releases of Salmo salar smolts.

Aquaculture 121:245–257.

Marschall, E. A., T. P. Quinn, D. A. Roff, J. A. Hutchings,

N. B. Metcalfe, T. A. Bakke, R. L. Saunders, and N. L.

Poff. 1998. A framework for understanding Atlantic

salmon (Salmo salar) life history. Canadian Journal of


McCormick, S. D. 1993. Methods for nonlethal gill biopsy

and measurement of Naþ,Kþ-ATPase activity. Canadian

Journal of Fisheries and Aquatic Sciences 50:656–658.

McCormick, S. D., B. T. Bjornsson, M. Sheridan, C.

Eilertson, J. B. Carey, and M. O’Dea. 1995. Increased

day length stimulates plasma growth hormone and gill

Naþ,Kþ-ATPase in Atlantic salmon (Salmo salar).

Journal of Comparative Physiology B 165:245–254.

McCormick, S. D., and R. L. Saunders. 1987. Preparatory

physiological adaptations for marine life of salmonids:

osmoregulation, growth, and metabolism. Pages 211–229

in American Fisheries Society, Symposium 1, Bethesda,

Maryland.

Metcalfe, N. B. 1998. The interaction between behavior and

physiology in determining life history patterns in Atlantic

salmon (Salmo salar). Canadian Journal of Fisheries and

Aquatic Sciences 55:93–103.

Metcalfe, N. B., F. A. Huntingford, W. D. Graham, and J. E.

Thorpe. 1989. Early social status and the development of

life history strategies in Atlantic salmon. Proceedings of

the Royal Society of London B 236:7–19.

Metcalfe, N. B., and J. E. Thorpe. 1990. Determinants of

geographical variation in the age of seaward-migrating

salmon, Salmo salar. Journal of Animal Ecology

59:135–145.

Nicieza, A. G. 1995. Morphological variation between

geographically disjunct populations of Atlantic salmon:

the effects of ontogeny and habitat shift. Functional

Ecology 9:448–456.

Nicieza, A. G., F. Brana, and M. M. Toledo. 1991.

Development of length–bimodality and smolting in wild

stocks of Atlantic salmon, Salmo salar L., under different

growth conditions. Journal of Fish Biology 38:509–523.

Obedzinski, M., and B. H. Letcher. 2004. Variation in

freshwater growth and development among five New

England Atlantic salmon (Salmo salar) populations

reared in a common environment. Canadian Journal of


Økland, F., B. Jonsson, A. J. Jensen, and L. P. Hansen. 1993.

Is there a threshold size regulating seaward migration of

brown trout and Atlantic salmon? Journal of Fish

Biology 42:541–550.

Riddell, B. E., and W. C. Leggett. 1981. Evidence of an

adaptive basis for geographic variation in body mor-

phology and time of downstream migration of juvenile

Atlantic salmon (Salmo salar). Canadian Journal of


Rohlf, F. J. 2001. tpsDig. Department of Ecology and

Evolution, State University of New York, Stony Brook.

Saunders, R. L., J. Duston, and T. J. Benfey. 1994.

Environmental and biological factors affecting growth

dynamics in relation to smolting of Atlantic salmon,

Salmo salar L. Aquaculture and Fisheries Management

25:9–20.

Saunders, R. L., and E. B. Henderson. 1970. Influence of

photoperiod on smolt development and growth of

Atlantic salmon (Salmo salar). Journal of the Fisheries

Research Board of Canada 27:1295–1311.

Schreck, C. B. 1981. Parr–smolt transformation and behavior.

Pages 164–171 in E. L. Brannon and E. O. Salo, editors.

Salmon and trout migratory behavior symposium. School

of Fisheries, University of Washington, Seattle.

Strauss, R. E., and F. L. Bookstein. 1982. The truss–body

form reconstructions in morphometrics. Systematic

Zoology 31:113–135.

Swain, D. P., B. E. Riddell, and C. B. Murray. 1991.

Morphological differences between hatchery and wild

populations of coho salmon (Oncorhynchus kisutch):

environmental versus genetic origin. Canadian Journal of


Thorpe, J. E. 1977. Bimodal distribution of length of juvenile

Atlantic salmon (Salmo salar L.) under artificial rearing

conditions. Journal of Fish Biology 11:175–184.

Thorpe, J. E., M. Mangel, N. B. Metcalfe, and F. A.

Huntingford. 1998. Modelling the proximate basis of

salmonid life history variation, with application to

Atlantic salmon, Salmo salar L. Evolutionary Ecology

122:581–599.

Thorpe, J. E., N. B. Metcalfe, and F. A. Huntingford. 1992.

Behavioral influences on life history variation in juvenile

Atlantic salmon, Salmo salar. Environmental Biology of

Fishes 33:331–340.

Vanstone, W. E., and J. R. Markert. 1968. Some morpholog-

ical and biochemical changes in coho salmon, Onco-rhynchus kisutch, during parr–smolt transformation.

Journal of the Fisheries Research Board of Canada

25:2403–2418.

von Cramon-Taubadel, N., E. N. Ling, D. Cotter, and N. P.

Wilkins. 2005. Determination of body shape variation in

Irish hatchery-reared and wild Atlantic salmon. Journal

of Fish Biology 66:1471–1482.

Winans, G. A. 1984. Multivariate morphometric variability in

Pacific salmon: technical demonstration. Canadian Jour-

nal of Fisheries and Aquatic Sciences 41:1150–1159.

Winans, G. A., and R. S. Nishioka. 1987. A multivariate

description of change in body shape of coho salmon

(Oncorhynchus kisutch) during smoltification. Aquacul-

ture 66:235–245.


Dow

nloa

ded

by [

McG

ill U

nive

rsity

Lib

rary

] at

08:

07 2

8 N

ovem

ber

2014

early discrimination of atlantic salmon smolt age: time course of the relative effectiveness of body...

Documents