UNIVERSITY OF HAWAl'I LIB UTILIZATION OF SHORT-WAVELENGTH NEAR INFRARED REFLECTANCE SPECTROSCOPY, MORPHOMETRIC MEASUREMENTS, SEX AND LOCATION OF CATCH TO PREDICT NUTRIENT CONTENT OF WHOLE HOMOGENIZED PACIFIC HERRING (CLUPEA PALLAS!) A THESIS SUBMITTED TO THE GRADUATE DIVISION OF THE UNIVERSITY OF HAWAI'I IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE IN ANIMAL SCIENCES MAY 2004 By Carey L. Morishige Thesis Committee: James R. Carpenter, Chairperson Yong Soo Kim Barbara Rasco
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
UNIVERSITY OF HAWAl'I LIB
UTILIZATION OF SHORT-WAVELENGTH NEAR INFRARED REFLECTANCE
SPECTROSCOPY, MORPHOMETRIC MEASUREMENTS, SEX AND LOCATION
OF CATCH TO PREDICT NUTRIENT CONTENT OF WHOLE HOMOGENIZED
PACIFIC HERRING (CLUPEA PALLAS!)
A THESIS SUBMITTED TO THE GRADUATE DIVISION OF THE UNIVERSITY OFHAWAI'I IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE
DEGREE OF
MASTER OF SCIENCE
IN
ANIMAL SCIENCES
MAY 2004
ByCarey L. Morishige
Thesis Committee:
James R. Carpenter, ChairpersonYong Soo KimBarbara Rasco
ACKNOWLEDGEMENTS
I would first like to thank the members ofmy thesis committee James R.
Carpenter, Yong Soo Kim and Barbara Rasco for their expertise and assistance in the
completion, compilation and editing ofthis thesis project. More specifically, I would like
to thank James R. Carpenter, PhD ofthe Department ofHuman Nutrition, Food and
Animal Sciences at the University ofHawaii at Manoa for his guidance and mentoring
throughout this project. Within the same department I would also like to thank Yong Soo
Kim, PhD for his willingness to aid in this project and especially for his expertise in
working with meat products. The willingness and expertise ofBarbara Rasco, PhD ofthe
Department ofFood Science and Human Nutrition at Washington State University, was
an invaluable asset to this thesis project. Aside from my committee, I would also like to
thank the Marine Mammal Research Program ofthe University ofHawaii, for their
gracious donation ofthree lots ofPacific herring, as well as the use oftheir fish
preparation facility on the Kaneohe marine base, Hawaii. For the fourth lot ofherring, I
would like to thank the Alaska SeaLife Center and the marine mammal staffand trainers
in Seward, Alaska for their generous donation. From Washington State University
(WSU), I would like to thank Mengshi Lin, PhD for his unfailing support, assistance and
expertise in Near Infrared Reflectance (NIR) spectroscopy. Also from WSU, I would
like to thank Dewi Setiady who aided in the scanning, analysis and NIR calibration ofmy
samples. For his constant encouragement, as well as expertise in marine mammal
husbandry, I would like to thank Jeffrey Pawloski ofthe Laboratory Animal Services at
the University ofHawaii. Finally, I would like to thank Kathryn Stanberry for her
constant support and assistance in the proximate analysis ofmy samples.
iii
ABSTRACT
In marine mammal husbandry, diet analysis is an integral part ofmaintaining
optimal health ofan animal. At many marine parks in the United States, Pacific herring
(Clupea pallasi) makes up a component ofthese animals' diets. Previous studies on
Pacific herring have found correlations between morphometric measurements, sex and
location ofcatch with the nutrient content ofherring. These variables were analyzed to
determine the effects of source and sex on nutrient composition ofherring and to examine
the possibility ofusing morphometric measurements ofPacific herring including length
(em), weight (g), volume (ml) and/or their ratios to predict its proximate composition.
Results showed significant lot and location differences. Ninety-nine samples were
measured, homogenized, the analyzed with standard methods ofproximate analyses.
Results showed that no significant correlations existed between morphometric
measurements or sex and nutrient composition. However, there were significant
differences (P<O.OOI) between all analytes (except carbohydrate) and location of catch.
Traditional methods ofdiet analysis, in the marine mammal field, are time
consuming and costly. Short-Wave Near Infrared Reflectance (SW-NIR) Spectroscopy
was suggested for its speed, efficiency and cost effectiveness. The objective ofthis
project was to develop SW-NIR spectroscopy calibration equations for moisture, fat,
protein and mineral content for whole homogenized Pacific herring. The prediction
equations (R2) for the calibrations for lots ranged from .27-.88, .66-.93, .48-.87, and .24
.91 for moisture, fat, protein and ash, respectively. Calibration models for each location
were fairly comparable: Alaska (Lot D) R2=O.66-0.91 and Canada (Lots A through C)
R2=0.47-0.81.
IV
TABLE OF CONTENTS
Acknowledgments iii
Abstract iv
List ofTables vii
List ofFigures ix
Chapter 1: Introduction 1
1.1 Marine Mammal Diet Analysis 1
1.2 Pacific Herring (Clupea pallasi) 1
1.3 Morphometric Measurements 2
1.4 Traditional Methods ofProximate Analysis 2
1.5 Near Infrared Reflectance Spectroscopy 4
1.6 Objectives 6
Chapter 2: Literature Review 7
2.1 Marine Mammal Field 7
2.2 Pacific herring (Clupea pallasi) 7
2.3 Effects on Nutrient Content ofPacific Herring 8
2.4 Laboratory Analyses 12
2.5 Near Infrared Reflectance Spectroscopy 13
Chapter 3: Utilization ofmorphometric measurements, sex and/or location ofcatch topredict the nutrient content ofwhole, homogenized Pacific herring (Clupea pallasi) 19
3.1 Introduction 19
3.2 Materials and Methods 21
3.3 Results 25
v
3.4 Discussion 30
3.5 Conclusion 34
Chapter 4: Utilization of Short Wave Near Infrared Reflectance Spectroscopy to predictthe nutrient content ofwhole, homogenized Pacific herring (Clupea pallasi) 51
4.1 Introduction 51
4.2 Materials and Methods 54
4.3 Results 58
4.4 Discussion 61
4.5 Conclusion 66
Chapter 5: Overall Summary and Conclusions 79
Appendix Table 1 87
Appendix Table 2 88
Appendix Table 3 89
Appendix Table 4 90
Appendix Table 5 91
Appendix Table 6 92
References 93
VI
LIST OF TABLES
3.1 Average, standard deviation and minimum and maximum values formorphometric measurement results by lot and location for Pacific herring 36
3.2 Linear regression relationships between morphometric measurements forPacific herring, including the R2 value and Root Mean Square Error (RMSE) 37
3.3 Average, standard deviation and minimum and maximum values for nutrientcomponent results ofPacific herring - overall and by lots and locations 38
3.4 Average, standard deviation and minimum and maximum values for nutrientcomponents ofPacific herring by size and sex (P >0.05 for all) 39
3.5 Linear regression equations and R2 values (>60%) nutrient componentrelationships in Pacific herring analyte correlations (n=99, P<O.OOI) 40
3.6 Sequence oflinear regression equations used to predict energy density (GE(kcaVg», lipid (%), ash (%), and protein (%) content ofPacific herring starting frommoisture content 40
3.7 Calculated proximate composition of samples from a previous study (usinglinear regression equations to predict analyte content) compared to known chemicalreference values 41
3.8 Linear regression R2 values, means and standard deviations (SD) forcalculated (using linear regression equations to predict analyte content) and knownchemical reference results 42
4.1 Linear regression equations and R2 values (>60%) nutrient componentrelationships in Pacific herring analyte correlations (n=99, P<O.OOI) 67
4.2 Sequence oflinear regression equations used to predict energy density (GE(kcaVg», lipid (%), ash (%), and protein (%) content ofPacific herring starting frommoisture content 67
4.3 Number oflatent variables and standard error ofprediction (SEP) for eachcalibration model chosen per component for each lot and location ofPacific herring.... 68
4.4 Actual chemical and SW-NIR predicted results for feed components ofPacificherring 69
vii
4.5 R2 value comparing SW-NIR predicted and calculated (using regressionequations) values with those from chemical analyses ofPacific herring 70
vm
LIST OF FIGURES
Figure: Page:
3.1 Map ofBritish Columbia, Canada. Pacific herring harvest area is circled 43
3.2 Map ofAlaska, United States. Pacific herring harvest area is circled 43
3.3 Quadratic regression fitted line plot for volume (ml) versus length (cm) ofPacific herring 44
3.4 Quadratic regression fitted line plot for weight (g) versus length (cm) ofPacific herring 44
3.5 Quadratic regression fitted line plot for weight (g) versus volume (ml) ofPacific herring 45
3.6 Quadratic regression fitted line plot for weight per unit length (g/cm) versusvolume (ml) ofPacific herring 45
3.7 Quadratic regression fitted line plot for volume per unit length (ml/cm) versusweight (g) ofPacific herring 46
3.8 Quadratic regression fitted line plot for weight per unit volume (g/ml) versuslength (cm) ofPacific herring 46
3.9 Bar graph showing mean length, volume and weight for Pacific herring sexcategories offemale, male and female with eggs 47
3.10 Bar graph showing mean volume per unit length (ml/cm), weight per unitlength (g/cm) and density (g/ml) for Pacific herring sex categories -offemale, maleand female with eggs 47
3.11 Bar graph showing mean length, volume and weight for Pacific herring sizecategories of small, medium and large 48
3.12 Bar graph showing volume per unit length (ml/cm), weight per unit length(g/cm) and density (g/ml) for Pacific herring size categories of small, medium andlarge 48
3.13 Linear regression plot ofmoisture versus gross energy (GE) content on anas fed basis (R2=O.96) for Pacific herring 49
3.14 Linear regression plot ofgross energy (GE) versus crude lipid (EE%)content on an as fed basis (R2=O.97) for Pacific herring 49
IX
3.15 Linear regression plot ofgross energy (GE) versus ash content on a drymatter basis (R2=0.72) for Pacific herring 50
3.16 Linear regression plot ofcrude lipid (EE) versus protein content on a drymatter basis (R2=0.68) for Pacific herring 50
4.1 Short-Wavelength Near Infrared (SW-NIR) spectra for lot A ofPacificherring 71
4.2 Second derivative transformation ofSW-NIR spectra for lot A ofPacificherring 71
4.3 Average chemical reference results for components in Pacific herring byharvest location 72
4.4 Average chemical reference results for components in Pacific herring by lot..... 72
4.5 Linear regression plot ofgross energy (GE) versus moisture content on an asfed basis (R2=O.96) ofPacific herring 73
4.6 Linear regression plot ofgross energy (GE) versus lipid (EE) content on anas fed basis (R2=o.97) ofPacific herring 73
4.7 Linear regression plot ofgross energy (GE) versus ash content on an as fedbasis (R2=0.72) ofPacific herring 74
4.8 Linear regression plot of lipid (EE) versus protein content on a dry matterbasis (R2=0.68) ofPacific herring 74
4.9 Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for protein in Pacific herring lots A (LV=5), B (LV=7), and C (LV=6) 75
4.10 Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for lipid in Pacific herring lots A (LV=8), B (LV=6), and C (LV=6) 75
4.11 Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for protein in Pacific herring lots D (LV=5) 76
4.12 Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for lipid in Pacific herring lots D (LV=5) 76
4.13 Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for protein in Pacific herring lots A, B, & C combined (LV=6) '" '" 77
x
4.14 Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for lipid in Pacific herring lots A, B, & C combined (LV=7). 77
4.15 Pacific herring lot C linear regression plot oflipid (EE) content by chemicalanalysis versus EE predicted by SW-NIR (R2=O.93) using six latent variables 78
4.16 Pacific herring lot D linear regression plot ofash content by chemicalanalysis versus ash predicted by SW-NIR (R2=O.91) using five latent variables 78
..
Xl
CHAPTER 1. INTRODUCTION
1.1 MARINE MAMMAL DIET ANALYSIS
Numerous marine parks in the United States house and exhibit marine mammals
such as Steller sea lions, Hawaiian monk seals and Harbor seals. One ofthe main
components to these animals' diets is herring. For those marine parks located on the
West Coast (including Hawaii and Alaska), the main species ofherring fed is Pacific
herring (Clupea pallasi).
In most marine mammal facilities, diet assessment is a key part ofdaily
husbandry. In a facility that houses an endangered or threatened species ofmarine
mammal, diet analysis is more critical. Diet analysis becomes even more important if
key research is being performed at the facility. Having the animals in captivity facilitates
needed research on the behavior, physiology and nutrition ofmarine mammals. Many
times, one or more components ofa study will rely upon accurate diet analysis.
1.2 PACIFIC HERRING (Clupeapallasi)
Pacific herring, a member ofthe Clupeidae family, is a widely used commercial
species. In the United States, during 2001, 91.3 million pounds ofPacific herring were
caught, equaling approximately 13.2 million dollars (Holliday and O'Bannon, 2001).
Consumed by humans and used as a component offeed for animals, the nutrient
content ofherring is important to know. According to the U.S. Department of
Agriculture's (USDA) National Nutrient Database for Standard Reference (2002), the
1
proximate composition of 100 g of(edible) raw Pacific herring is: protein 16.39 g, total
lipid 13.88 g, ash 2.37 g, and water 71.52 g (as-fed basis).
There are numerous factors influencing the proximate composition ofthis species.
Both season and location ofcatch contribute to the nutrient variation ofPacific herring,
especially in lipid content. Also, the age ofthe fish, sex and/or stage ofthe reproductive
cycle influence nutrient content. Some herring are "wild caught," which means that there
is minimal fishery preparation ofthe fish before freezing. In other fisheries, herring are
corralled, starved (to clear gut contents) and then collected before being frozen. These
different methods offishery processing can also affect nutrient content of the fish.
1.3 MORPHOMETRIC MEASUREMENTS
Because correlations between morphometric characteristics ofherring and its
proximate content have been found in previous studies (Castellini et al., 2001, Anthonyet
al., 2000), there exists a possibility ofutilizing these measurements to predict nutrient
content ofherring. Ifa correlation ofthis kind does exist, a marine mammal facility
would be able to simply measure the length, volume, or weight ofa herring and predict
its nutrient content. The ability to do this would greatly facilitate diet analysis. This
would be the simplest and quickest method ofgaining proximate content ofa herring.
There would be very minimal sample preparation and no laboratory analysis needed.
1.4 TRADITIONAL METHODS OF PROXIMATE ANALYSIS
In the marine mammal husbandry field, diet analysis is an integral part of
maintaining optimal health ofan animal. The amount ofa species offish fed to an
2
animal is based not only upon the age, weight and energy requirement ofthe animal, but
also upon the proximate content of the fish. In order to estimate how much ofa certain
species offish should be fed, accurate nutrient analyses (to determine percentages offat,
protein, moisture and ash and its energy density, kcaVg) must be readily available.
In most facilities, when a new lot of fish comes in, samples ofthat fish are sent to
outside laboratories for proximate analysis prior to diet formulation. In some cases, the
new lot of fish may need to be used quickly as the current lot is running low. In these
instances, accurate proximate analyses are needed quickly.
These laboratories can be far away and the lab analyses are usually quite
expensive and time consuming. Ofthe many chemical analyses that can be run, the most
common for proximate analysis are Goldfisch or soxhlet methods using ethyl ether for
lipid extraction (crude fat content), Kjeldahl procedure for nitrogen determination
(Nitrogen x 6.25 = crude protein content), freeze-dried or oven-dried dry matter (moisture
content) and ash via combustion furnace (mineral content). Before the analyses can be
done the sample needs to be prepared by homogenization, drying, freeze-drying or a
combination oftwo ofthese methods. This sample preparation can be time consuming,
taking up to a day or more to complete depending upon the size ofthe fish or sample and
method used. The optimal sample consistency is entirely homogenous. One ofthe
challenges in achieving this with fish is the grinding ofthe bones, scales and skin in order
to get a completely homogenous sample mixture. From here, the analyses are run. Both
the ethyl ether extraction and the Kjeldahl procedure also include the use ofhazardous
chemicals such as ethyl ether (C.JIlOO), sodium hydroxide (NaOH), and sulfuric acid
(H2S04). They also require large volumes ofwater, mainly to run through the equipment.
3
All analyses rely heavily upon human accuracy in measuring and weighing. Therefore,
there is a greater possibility for human error in determining the proximate content of the
sample. These analyses also rely on the availability ofproper equipment and materials,
and a trained lab technician. These materials are costly, and cleaning them after every
analysis takes time. Another drawback ofthese chemical analyses is that most times the
sample is destroyed during the processing and/or analysis, and therefore unavailable for
possible fatty acid, amino acid or individual mineral element analysis later on.
Due to the length oftime these analyses take, many times a marine mammal diet
is formulated without the needed information for fat, protein, moisture, mineral and gross
energy content. In these cases, the pounds ofthat species offish to be fed are estimated
based on previous proximate data and curator judgment. In some cases, the animals are
fed simply a percentage oftheir bodyweight. This leads to an erroneous diet and possible
nutrition and diet related behavioral problems in the animal being fed (e.g., cessation of
eating or "playing" with food). Because ofthis, a more efficient and less costly
alternative method ofnutrient analysis is needed. Near Infrared Reflectance
Spectroscopy has been suggested as that alternative.
1.5 NEAR INFRARED REFLECTANCE SPECTROSCOPY
Near Infrared Reflectance (NIR) spectroscopy is an expanding technology which
has recently grown in popularity. In the livestock feed and forage industry, it has been
widely used and well accepted for the prediction ofnutrient composition used in ration
(diet) formulation. This technology has the ability to provide both quantitative and
qualitative information on a wide array ofproducts (especially raw materials) in a matter
4
ofminutes (Burns and Ciurczak, 1992). A researcher, for example, is able to
simultaneously obtain the amounts of several organic analytes (e.g., protein, fat, fiber,
etc.) in a sample. NIR spectroscopy is a cost effective and accurate method ofanalysis
where no hazardous chemicals are needed. It is also non-destructive to samples. As
indicated previously, this technology often requires minimal sample preparation and
provides the capability to analyze multiple nutrients simultaneously in a very short period
of time. Most NIR units normally scan samples for approximately 30 seconds and take
an average of20-30 scans in that time period.
NIR spectroscopy can be used to detect virtually all organic compounds. Easiest
to detect are those with functional groups such as hydroxyl, carboxyl, amine and carbon
hydrogen (Burns and Ciurczak, 1992). NIR spectroscopy works by measuring the
amount of light reflected from a thoroughly homogenized sample. The light reflected is a
function ofthe light absorbed by the chemical components (i.e., functional groups)
making up the sample. The light absorbed by a specific chemical component is related to
the amount of that component present in the sample. Basically, a NIR spectrophotometer
will shine a light source ofvarying wavelengths, normally between 600-11OOnm
(depending upon the type ofsample) onto the sample. This will cause the molecules in
the sample to vibrate. These vibrations are caused by the bending and stretching of
hydrogen bonds with carbon, nitrogen and oxygen. The spectrophotometer detects the
signals reflected and amplifies those signals. These spectra then go to a computer
(containing a software program that usually comes with the NIR spectrophotometer)
where they are correlated, by comparison to an internal reference standard (calibration),
to determine the amount ofthat particular component present in the sample. There is also
5
a form ofspectroscopy that detects the signals that travel through the sample. This is
known as Near Infrared Transmission spectroscopy.
The accuracy ofNIR spectroscopy depends greatly upon an accurate independent
method ofmeasuring the properties within a sample of interest. Accuracy also depends
highly on sample preparation. Samples must be uniform in consistency and particle size
to reduce error and increase both the accuracy and repeatability ofthe analysis. The
results ofthis independent method ofanalysis (such as chemical analysis) will be used to
build a calibration model, which the spectrophotometer will use when analyzing that
particular sample. Although NIR spectroscopy is an extremely rapid method ofanalysis,
the first stage ofdata analysis (i.e.• chemical and statistical analyses) is time-consuming
(Murray and Cowe, 1992) and sometimes quite costly.
1.6 OBJECTIVES
The objectives of this project are: 1) to determine the effects ofsource and sex on
nutrient composition ofherring, 2) to examine the possibility ofusing morphometric
their ratios to predict its proximate composition, and 3) to develop and compare NIRS
calibration equations and nutrient regressions for fat, protein, moisture and mineral
content for whole homogenized Pacific herring (C/upea pa//asi).
6
CHAPTER 2. LITERATURE REVIEW
2.1 MARINE MAMMAL FIELD
Diet analysis is an integral component ofgood marine mammal husbandry.
Marine parks across the United States house and exhibit various species of marine
mammals, some of them endangered. In cases like these, where an endangered or
threatened species, such as the Hawaiian monk seal (Monachus schauinsiandi), is housed,
diet analysis becomes increasingly more important. Having marine mammals in captivity
facilitates needed research on the behavior, physiology and nutrition of different species.
Many times, one or more components ofa study will rely upon accurate diet analysis.
Due to the known variation in the nutrient composition offish, along with their variation
in seasonal availability, routine sampling and proximate analysis is necessary (Bernard et
ai., 1997).
Several marine mammal facilities formulate diet based upon a percentage of the
animal's weight. Others formulate diet based upon the gross energy content (kcal/kg per
day) of the feed. Either way, a nutritionally balanced diet, which takes into consideration
the physiological health and characteristics of the specific animal, is the primary goal.
2.2 PACIFIC HERRING (Ciupeapallasi)
Pacific herring (Ciupea harengus pallasi Valenciennes 1847) is a widely and fully
used commercial species. This schooling specie is distributed from northern Baja
California to Toyama Bay, Japan, continuing to the shores ofKorea and the Yellow Sea
(Miller and Lea, 1972). This species of herring is one of the main components to captive
7
marine mammal diets in most U.S. marine parks located on the west coast, Hawaii and
Alaska. Pacific herring is also a major prey item for marine mammals in the Pacific,
including the endangered North Pacific population ofhumpback whales.
Since it is consumed by humans and used as a component offeed for animals,
knowing the nutrient content ofa herring is important. Several studies have looked at the
proximate composition ofPacific herring. In a study done by Castellini et al. (2001)
Pacific herring was 16.8±2.2% lipid (as-fed basis) and had an average energy density of
9.3±0.8kJ/g (as-fed basis). In another study conducted by Anthony et al. (2000), Pacific
herring was 77.6±0.20% moisture (as-fed basis) and 1O.5±0.80% lipid (dry matter basis).
According to the U.S. Department ofAgriculture's (USDA) National Nutrient Database
for Standard Reference (2002), the proximate composition of 100 g of (edible) raw
Pacific herring is: protein 16.39 g, total lipid 13.88 g, ash 2.37 g, and water 71.52 g (as
fed basis).
2.3 EFFECTS ON NUTRIENT CONTENT OF PACIFIC HERRING
There are numerous factors influencing the chemical make-up of this species.
Among these factors are harvest location and/or season, sex and/or reproductive status
and methods of processing and storage.
2.3.1 Size and Morphometric Characteristics
Having the ability to predict proximate content ofherring based upon
morphometric characteristics, such as length, weight and/or volume, would be extremely
helpful to the field of marine mammal husbandry. In an emergency case, where a box of
8
herring needs to be fed and its proximate analysis was unavailable, the ability to estimate
(predict) chemical composition from moisture content would prove invaluable.
Lipid content was found to be correlated with the size of a herring. In a study by
Anthony et al. in 2000, large herring (14-30 cm, 38 ± 1.6%) had, on average, twice the
lipid content of medium sized herring (10-14 cm, 25 ± 0.7%) and three times that of the
small sized herring «10 cm, 10 ± 0.5%). Other studies have found that the size ofa
herring is related to age. Lipid content was found to be positively correlated with the age
ofa herring (Payne et al., 1999; Paul et al., 1998; Harris et al., 1986; Stansby, 1976).
Therefore, it can be inferred that the size of a herring is correlated with lipid content.
These size-related differences may result from changes in the allocation ofenergy with
stage ofgrowth and sexual maturity (Calow and Townsend, 1981). In younger fish, more
energy is allotted to somatic growth. This is due to an increase in locomotive efficiency,
predator evasion and food procurement (Calow and Townsend, 1981). During this stage,
growth requires more protein turnover than does weight gain and maintenance in adults.
Therefore, adults store more energy as lipid (Anthony et al., 2000; Harris et al., 1986).
2.3.2 Location ofCatch
A study by Anthony et al. (2000) found that location of catch also affects the
proximate content of herring. Researchers from this study found that there was a great
variation in lipid content (21-29%) in herring between different areas ofPrince William
Sound, Alaska (i.e., northeast versus central and southwest). A significant difference
(P<0.01) in lipid content ofherring was also seen between Prince William Sound (41%)
and the Lower Cook Inlet (32%), Alaska. The differences seen between areas within
9
Prince William Sound were attributed to availability of prey (i.e., zooplankton and
calanoid copepods). Seasonal differences in prey availability have also been cited as a
possible reason for differences in body growth (Ware, 1985) and lipid reserves (Anthony
et al., 2000). Similar regional differences were found in a study done by Paul and Paul
(1999) in Prince William Sound, Alaska.
2.3.3 Sex
Reproductive cycle stage, sex, and fecundity have also been found to influence
nutrient content, especially for lipids (Castellini et al., 2001). Spawning occurs at
different times depending upon the population's geographic location. In Alaska, for
example, herring are spring spawners (ADF&G, 1978). In Pacific herring, when the
ovary weights exceed five percent of the total body weight, sexual maturation has begun
(Hay and McCarter, 1999). Sexual maturity has been reported between the ages of two
(Rumyantsev and Darda, 1970) and three years (Hay and Cater, 1999). Most herring do
not spawn until at least three or four years ofage. By the age offive, approximately 95%
ofthe population has reached sexual maturity (Rumyantsev and Darda, 1970). A study
done by Paul et al. (1996) found that whole ovarian energy content was positively
correlated with body weight (R2=0.76).
In Atlantic herring (CIupea harengus), a species very similar to Pacific herring, it
was found that female gonads were higher in lipid content than male gonads. It could be
assumed then that females may be higher in lipid content than males.
Fecundity in herring has been found to increase with body length and width
(Nagasaki, 1958) and varies depending upon age (Rumyantsev and Darda, 1970).
10
2.3.4 Lot or Season of Catch
When frozen or processed herring are purchased, they are purchased in lots. Each
lot ofherring has a specific date ofcatch; therefore, lot is related to the date or season of
catch. Depending upon the season of catch, lipid content in Pacific herring varies (Paul
et al., 1998; Payne et al., 1999). This is thought to be due to the seasonal variation in
availability of food in different locations. This variation has been directly related to lipid
content (Blaxter and Holiday, 1963; Stansby and Hall, 1967). In Alaskan herring, energy
intake increases during the spring, summer and early fall months in preparation for the
winter when prey availability and feeding conditions are poor (Blaxter and Holiday,
1963; Paul et al., 1996; Paul and Paul, 1998). This has resulted in higher lipid content in
herring caught in fall and early winter (Castellini et al., 2001). Lipid content and energy
density of herring caught in the same season, during different years, will also vary. In a
study done by Anthony et al. (2000), herring of the same size caught in 1995 had a higher
lipid content and energy density than those caught in the same region in 1996. This was
thought to be due to the availability and/or quality oftheir food being better in 1995.
Similar interannual variations in lipid content were found in the study done by Paul and
Paul (1999).
2.3.5 Processing and Storage
Methods used in the harvest (catch), shipping, and handling ofPacific herring,
along with the storage temperature of the fish affect their resulting nutrient content. The
methods used in freezing the fish prior to shipping is also very important. Fish come
both block-frozen, as well as individually quick frozen (IQF). Ofthe two, IQF fish are
11
usually of higher quality due to the shorter time it takes for the fish to freeze. With this
method offreezing fish exposure to oxygen is minimized. This decreases the likelihood
of lipid oxidation, which oxygen exposure is thought to increase (Hultin, 1992).
Factors related to storage, such as temperature, may also affect the likelihood of
lipid oxidation. At higher temperatures, degradation of proteins and oxidation oflipids
occur (Hultin, 1992; Bernard et al., 1997). The suggested optimal storage temperature is
between -18C and -30C (Bernard et al., 1997). This is especially key with high fat fish
such as herring since fattier fish are more likely to undergo lipid oxidation (Hultin, 1992).
2.4 LABORATORY ANALYSES
2.4.1 Methods ofChemical Analyses
Most times, feed samples are sent to laboratories outside of the marine mammal
facility for proximate analysis. These laboratories can be far away and the lab analyses
are usually quite expensive and time consuming. Ofthe many chemical analyses that can
be run, commonly used methods for nutrient analysis are Goldfisch and Soxhlet methods
using ethyl ether extraction (crude lipid content) and Kjeldahl procedure (nitrogen
determination for crude protein content). Common methods for determining moisture
and ash content are freeze-drying and oven-drying (moisture content), and using an
electric muftle furnace (ash or total mineral content).
There exist a variety of laboratories and methods for determining the lipid content
of a fish. This is due to the tremendous variation in the methods used for processing and
preparation of herring for nutrient analysis.
12
2.4.2 Drawbacks ofLaboratory Analyses
Before the analyses can be done the sample needs to be prepared by
homogenization, drying, freeze-drying or a combination of two of these methods. This
sample preparation can be very time consuming, and often taking up to a day or more to
complete depending upon the method used. Once uniform particle size and texture is
gained, the samples are prepared and the nutrient analyses can be run.
The Goldfisch or Soxhlet lipid extractions and the Kjeldahl procedure to
determine nitrogen content both include the use ofhazardous chemicals such as ethyl
ether (CJflOO), sodium hydroxide (NaOH), and sulfuric acid (H2S04). All ofthese
analyses rely heavily upon human accuracy in measuring and weighing. Therefore, there
is a greater possibility for human error in determining the nutrient content of the sample.
These analyses also rely on the availability of proper equipment and materials and a
trained lab technician. These required materials (e.g., equipment, glassware and
reagents) are costly, chemicals are difficult to dispose ofand cleaning them after every
analysis takes time. Other drawbacks of these chemical analyses are that most times the
sample is destroyed during the process and large quantities ofwater are used.
2.5 NEAR INFRARED REFLECTANCE SPECTROSCOPY
2.5.1 Background
Near Infrared (NIR) Reflectance spectroscopy is an expanding technology which
has recently grown in popularity. In the livestock feed and forage industry, it has been
widely used and well accepted for the prediction ofnutrient composition used in diet
analysis (Williams and Norris, 1987). This technology has the ability to provide both
13
quantitative and qualitative information on a wide array ofproducts (especially raw
materials) in a matter ofminutes (Bums and Ciurczak, 1992). This enables a researcher,
for example, to obtain the amount of an organic analyte ofinterest (e.g., protein) in a
sample very quickly. NIR spectroscopy is a cost effective and accurate method of
analysis where no hazardous chemicals are needed. It is also non-destructive to samples.
This technology provides the capability to analyze multiple nutrients simultaneously.
Most NIR units normally scan samples for approximately 30 seconds, take an average of
20-30 scans during that time period and conveniently determines the average spectra and
analyte/nutrient content for that sample.
NIR spectroscopy can be used to detect virtually all organic compounds. Easiest
to detect are those with functional groups such as hydroxyl, carboxyl, amine and carbon
hydrogen bonds (Bums and Ciurczak, 1992). NIR spectroscopy works by measuring the
amount oflight reflected from a thoroughly homogenized sample. The light reflected is a
function ofthe light absorbed by the chemical components (i.e. functional groups) that
make up the sample. The light absorbed by a specific chemical component is related to
the amount of that component present in the sample (Osborne, 2001). Basically, a NIR
spectrophotometer will shine a light source ofvarying wavelengths onto the sample. This
will cause the molecules in the sample to vibrate. The bending and stretching of
hydrogen bonds with carbon, nitrogen and oxygen cause these vibrations. For NIR
reflectance spectroscopy the spectrophotometer detects the signals reflected. This
spectral data is analyzed using linear regression based statistical models where they are
quantified, by comparison to a series ofreferences with known composition (calibration),
to determine the amount ofthat particular component present in the sample. There is also
14
a form of spectroscopy that detects the signal that travels through, or is absorbed by the
sample. This is known as Near Infrared Transmission (NIT) spectroscopy.
2.5.2 Calibration Model
Once the primary data analyses are completed and the amounts or properties of
the sample are collected, a calibration model is built. In order to build the calibration
model, the NIR spectra ofthe sample are needed. To obtain this, these samples must be
run on an NIR spectrophotometer. Regression equations are then calculated based on the
NIR spectra obtained and the known analyte information. The model is used to predict
future unknowns. Partial Least Squares regression statistics are most commonly used for
linear calibration models. This method takes into account all wavelengths used.
Depending on the make of the NIR unit, or company from which equipment or software
are purchased, there can be considerable variation in models employed. Based upon
these statistics, the model with the lowest standard error ofprediction is used as the
calibration model (Murray and Cowe, 1992). The NIR computer software does much of
the analyses and generally aids developing this calibration model.
The next step is to test the validity and the prediction accuracy of the calibration
model chosen using cross validation. To do this, the samples are divided into two sets: 1)
calibration set and 2) prediction or validation set. Here, the analyte data in the calibration
set is used, with the calibration model chosen, to predict the amount ofan analyte in the
prediction set. More recently, "leave-one-out" cross validation is used. With this
method, one sample is left out while the rest are used to predict the analyte ofinterest
within that sample. An advantage ofcross validation is that the sample is not included in
15
the calibration model and therefore, the model can be tested independently (Murray and
Cowe, 1992).
Once the calibration model is chosen and validated, this calibration can be used to
predict future unknowns. These unknowns, however, must be within similar sample
populations as those used the calibration set. For example, the calibration for fish is
thought to be species specific (Rasco, personal communication).
The accuracy ofany type ofNIR spectroscopy depends greatly on an accurate,
independent method ofmeasuring the properties within a sample of interest. Increasing
the number of samples used to create the calibration is thought to increase the reliability
of the calibration. However, errors can occur and are usually associated with the NIR
prediction accuracy. They arise due to several factors such as, previous errors in the
reference methods (i.e. chemical analysis), instability with the NIR spectrophotometer,
and inappropriate choice ofthe calibration model (Burns and Ciurczak, 1992),
inconsistent sample preparation and particle size. Samples must be uniform in
consistency to reduce error and increase the accuracy ofthe analysis. They should also
be ofthe same constant temperature (Bechmann and Jorgensen, 1998). The results of this
reference analysis (such as laboratory chemical analysis) will be used to build the
calibration models, which the spectrophotometer will use when analyzing that particular
sample. Although NIR spectroscopy is an extremely rapid method ofanalysis, this first
stage ofdata analysis (i.e. chemical and statistical analyses) is time-consuming (Murray
and Cowe, 1992).
16
2.5.1 NIR Spectroscopy and Fish
The use ofNIR spectroscopy in the fish industry has grown in recent years. This
technology is used mainly for quality control purposes (i.e. in aquacultured fish) leading
to a higher quality offish in the market. Recent studies looked at minimizing sample
preparation time by scanning whole, intact fish. In a study done by Downey (1996) NIR
spectroscopy (700-1100nm) was used to predict the moisture and oil content offarmed
salmon. Scanning was done through the skin on intact fish. R2 values in this study
ranged from 0.69-0.77 for moisture and 0.70-0.74 for lipid content. Intact fish were also
scanned in studies done by Lee et al. (1992), Isaksson et al. (1995) and Wold and
Isaksson (1997). In the study by Lee et at. (1992), whole rainbow trout were scanned
intact and calibration models were created for lipid content. Isaksson et at. (1995),
scanned intact salmon fillets for fat, protein and moisture content. Wold and Isaksson
(1997), used a fiber optic probe to scan 49 intact salmon for fat and moisture content. In
some studies, sample preparation was minimized, however not removed completely. In
one study by Rasco et al. (1991), cross sections offrozen plugs taken from cultured
rainbow trout were scanned. This method was able to predict, with fairly good accuracy,
fat, moisture and protein content. In another study done by Wold et al. (1996), cylinders
of intact tissue were taken from various sites along and within ten farmed salmon. The
R2 results for fat prediction were greater than 0.90 for all sites except one (0.44 for the
plug containing skin and dark muscle). Other studies went further in preparing samples
by homogenizing or grinding them prior to spectra collection. One such study was done
by Sollid and Solberg (1992) where they ground cultured Atlantic salmon fillets before
scanning. This study used NIR spectroscopy (850-1050nm) to predict the fat content of
17
the ground salmon. Prediction results were exceptional with an R2 value of0.98. Ground
fish were also scanned in studies done by Isaksson et al. (1995), Valdes et al. (1997), and
Zhang and Lee (1997). In a study done by Darwish et al. (1989), fish were processed and
ground to milk-like consistency in order to be analyzed via mid-infrared transmittance
spectroscopy. The method of sample preparation used in this study was extensive and
time consuming. There is a range in sample preparation techniques when utilizing this
technology, however current research tends to focus on minimizing sample preparation as
much as possible.
18
CHAPTER 3. UTILIZATION OF MORPHOMETRIC MEASUREMENTS,SEX AND/OR LOCATION OF CATCH TO PREDICT THE NUTRIENTCONTENT OF WHOLE, HOMOGENIZED PACIFIC HERRING (Clupea
pallasi)
3.1 INTRODUCTION
Numerous marine parks in the United States house and exhibit marine mammals
such as Steller sea lions, Hawaiian monk seals and Harbor seals. One of the main
components to these animals' diets is herring. The main species of herring fed in marine
parks located on the west coast (including Hawaii and Alaska) is Pacific herring (Clupea
pallasi).
In most marine mammal facilities, diet assessment is a key part of daily
husbandry. In a facility that houses an endangered or threatened species ofmarine
mammal, diet analysis is more critical. Diet analysis becomes increasingly more
important ifvital research is being performed using the animals at the facility. Having
marine mammals in captivity facilitates research on the behavior, physiology and
nutrition of marine mammals. Many times, one or more components ofa study will rely
upon accurate diet analysis.
3.1.1 Morphometric Measurements ofPacific Herring
Lipid content in many species offish depends upon the size of the fish. This
relationship has been reported in Pacific herring (paul et al., 1998; Payne et al., 1999).
Since lipid is related to the size ofa fish and size was found to be closely related to age, it
could be stated that the lipid content ofa fish is related to size. Studies have found a
positive correlation has been found between standard size and lipid content in herring
19
(Castellini et al., 2001). However, another study, done by Paul and Paul (1998), found
no predictable association between standard length and whole body energy content,
which is made up primarily of lipid. Their linear regression resulted in an R2 value of
0.008. This could be due to the fact that growth in length and mass are both influenced
by different factors so there is notable variability in mass-at-age (Tanasichuk, 1997).
There seems to be a general tendency for a correlation between lipid content and size,
however other factors such as location and season ofcatch, and reproductive status may
also affect this correlation.
Since some correlations have been found between morphometric characteristics of
herring and its nutrient content in previous studies, there exists a possibility ofutilizing
these measurements to predict nutrient content ofherring. If a correlation does exist
between these parameters, a marine mammal facility would be able to simply measure the
length, volume, or weight of a herring, then calculate its volume per unit length, or
weight per unit length, and estimate its nutrient content. The ability to do this would
greatly facilitate diet analysis because there would be minimal sample preparation and no
laboratory analysis would be needed.
3.1.2 Sex and Location Effects on Nutrient Content
Reproductive cycle stage, sex, and fecundity have also been found to influence
nutrient content, especially lipids (Castellini et al., 2001). Location also affects the
nutrient content ofherring as was found in a study by Anthony et al. (2000) where a great
variation was observed in lipid content of herring within Prince William Sound, Alaska.
20
They also reported a significant difference (P<0.01) in lipid content ofherring between
Prince William Sound and the Lower Cook Inlet, Alaska.
Studies have also found that the quantities of several other sample constituents are
highly correlated. This method of predicting the proximate content of a herring was
however, mentioned in a study done by Anthony et al. (2000), where they suggested the
use of the linear regression formula for predicting gross energy based on moisture
content. The use of this method of calculating analyte concentrations was not found in
previous literature.
Therefore, the objectives of this study were to determine the impact of source
(location ofcatch) and sex ofPacific herring on nutrient composition, and to assess the
possibility ofusing morphometric measurements ofherring including length (cm), weight
(g), volume (ml) and/or their ratios to predict its nutrient (fat, protein, and carbohydrate),
total mineral or ash and gross energy (GE) content.
3.2 MATERIALS AND METHODS
3.2.1 Sample Collection
The Pacific herring samples (n = 99) used were from two fisheries - North Bay
Meat Company which sells herring caught in British Columbia, Canada (lots A, B, and
C) and Petersburg Fisheries in Alaska (lot D) (Figures 3.1 and 3.2). The Marine
Mammal Research Program ofthe University ofHawaii donated the North Bay herring.
Samples for those lots were prepared at their fish preparation facility at Kaneohe Marine
Base Headquarters, Hawaii. The Petersburg herring was donated by the Alaska SeaLife
Center in Seward, Alaska. Each lot (A, B, C, and D) was made up of herring caught on a
21
specific date. Catch dates for these lots were not known since dates were not recorded.
Herring samples within these lots were individually quick-frozen and randomly selected
using the methods described below. Samples were then shipped, packed in blue ice and
still in the frozen state, to Hawaii via overnight FedEx.
Herring samples were collected from several boxes from each lot, and stored at
-80°C. Four to five fish were collected once a week from a 13.6 kg box ofherring.
These herring were being sampled as boxes were being opened and used to feed
bottlenose dolphins, a false killer whale, Harbor seals and/or Steller sea lions. Due to
sample collection within such a dynamic setting, the number ofboxes sampled from each
lot could not be controlled. Approximately 30 fish were collected per lot. Herring were
randomly selected based upon estimated size categories within each box - small, medium
and large. Sizes were determined by comparison to other herring in the box that could be
seen at the time ofcollection. Three size categories were collected not only to randomize
sample collection based upon the size distribution within each lot, but also to randomize
possible nutrient range in each lot. Ten fish of each size were collected per lot. Handling
ofthe fish was kept to a minimum for quality control purposes, especially because the
unsampled herring would later be fed.
3.2.1 Morphometric Measurement and Sample Preparation
To prepare tissue samples for nutrient analysis, the herring were first thawed in a
cool, fresh water bath. Fish were then put into a clean stainless steel pan. Standard
length was measured, with a ruler, in centimeters (to the nearest 0.25 em) from the tip of
the mouth to the end of the caudal peduncle. The herring was then weighed to the nearest
22
tenth gram, using an Ohaus model DS10 digital scale. Volume was measured by
displacement ofwater in a 1000 ml cylindrical glass graduated cylinder with 10 ml
demarcations. The beaker was filled with 500-600 ml offresh water. The fish was then
dropped in gently and the water displacement was measured to the nearest 5 ml and
recorded. Prior to homogenization, the herring were cut sagittally into 4 pieces and the
sex (male, female or female with eggs) was determined and noted. Herring were
homogenized in a glass container with pouring lip, using a 12-speed Oster blender with a
450-watt motor. The blending process for each fish took approximately 10min. since
time was spent scrapping down the sides of the container as blender speed was increased
from "chop," to blend and finally to "mix." The thoroughly mixed samples were then
scooped out with a stainless steel spoon and distributed into two 50 m1 disposable sterile
centrifuge tubes. Whole herring were homogenized because the nutrient make-up of the
entire fish was needed since marine mammals are fed the entire fish. The sample tubes
were then stored at -80°C until analysis.
3.2.2 Proximate Analysis
Chemical proximate analysis was conducted at the University ofHawaii Animal
Sciences nutrition laboratory. Standard methods of the Association of Official Analytical
Chemists (1990) for meat (Ellis, 1984) were used. Duplicate analyses were run for each
component. To determine dry matter content, 3-5 g ofwet sample were placed in
ceramic crucibles and dried for 24 hours in a 100°C drying oven. Moisture was
determined by calculating the difference between the initial wet weight and final dry
weight. Ash content was determined on dried samples by combustion in an electric
23
muflle furnace for 8 hours at 650°C. Crude protein content was determined using a
modified Kjeldahl nitrogen analysis. Approximately 1-3 g ofwet sample was first
digested in a Tecatur 1015 block digester programmed with a 4.5 hour ramp up and
digestion time. Samples were then transferred into Kjeldahl flasks for distillation and
titration using normal Kjeldahl units, and nitrogen content was determined by titration.
Crude fat content was determined using the Goldfisch ether extract procedure.
Approximately 4-9 g of fresh wet sample were placed in thimbles and dried at 50°C for
48 hours prior to analysis. The lower temperature ensured that fat from the herring did
not seep through the thimble during the drying process. The dried samples in the
thimbles were then analyzed. For all data, when errors between duplicates were greater
than 5%, they were rerun, with the exception ofash. Due to a lack of sufficient sample
material to rerun analyses, the means of the ash duplicates were used. Ash results with
errors greater than 15%, if it was determined to be a definite outlier, were removed.
Considerable diversity in ash weights has been found and may be due to individual
differences in robustness of the skeleton (Paul et ai., 2001). Also, due to the high fat
content of herring, there is the possibility of some splattering as samples are heated
quickly during the ashing process, thus there is a potential for sample loss.
To calculate crude protein and fat content of the herring samples on dry matter
basis, the dry matter percentage was used to estimate dry sample weight. Carbohydrate
(CRO) content was determined by difference using the equation: CRO (dry matter basis)
= 100 - (%Ash + % Crude Protein + % Crude Fat). Gross energy was calculated using
the protein, fat and CRO energy values ofPond et al (1995) in the following equation:
Table 3.2 Linear regression relationships between morphometric measurements forPacific herring, including the R2 value and Root Mean Square Error (RMSE).
Morphometric measurement n Regression equation R2 RMSE
Lot A Length (em) vs. Vol (ml) L=12.22+0.07V 0.95 0.62Length (em) vs. Wt (g) L=12.63+0.07W 0.93 0.68Vol (ml) vs. Wt (g) 29 V=6.37+0.94W 0.98 5.71Wt (g) vs. VlL (ml/em) W=-36.03+26.80(V/L) 0.94 8.65Vol (ml) vs. W/L (g/em) V=-18.70+23.50(W/L) 0.94 9.49Length (em) vs. WN (g/ml) L=11.54+6.94(W1V) 0.04 2.59
Lot B Length (em) vs. Vol (ml) L=12.75+0.07V 0.93 0.74Length (em) vs. Wt (g) L=12.95+0.06W 0.94 0.68Vol (ml) vs. Wt (g)
28V=4.83+0.95W 0.97 6.83
Wt (g) vs. V/L (ml/em) W=-33.46+26.77(V/L) 0.96 9.20Vol (ml) vs. W/L (g/em) V=-18.93+23.66(W/L) 0.95 8.31Length (em) vs. WN (g/ml) L=6.25+12.69(W1V) 0.17 2.51
LotC Length (em) vs. Vol (ml) L=13.05+0.07V 0.76 0.57Length (em) vs. Wt (g) L=13.32+0.06W 0.77 0.56Vol (ml) vs. Wt (g)
22V=9.13+0.85W 0.87 5.32
Wt (g) vs. VlL (mllem) W=-25.12+25.25(V/L) 0.78 6.92Vol (ml) vs. W/L (g/em) V=-11.01 +20.22(W/L) 0.79 7.74Length (em) vs. WN (g/ml) L=12.43+1.33(W1V) 0.57 0.76
Lot 0 Length (em) vs. Vol (m1) L=12.24+0.07V 0.86 0.64(Alaska) Length (em) vs. Wt (g) L=11.98+0.08W 0.93 0.44
Vol (ml) vs. Wt (g)20
V=2.31+0.95W 0.92 6.03Wt (g) vs. VlL (ml/em) W=-19.15+22.94(V/L) 0.87 7.26Vol (ml) vs. W/L (g/em) V=-19.73+22.55(W/L) 0.88 7.77Length (em) vs. WN (g/ml) L=12.51 +5.09(W1V) 0.03 1.66
Lots ABC Length (em) vs. Vol (ml) L=12.71+0.07V 0.92 0.68(Canada) Length (em) vs. Wt (g) L=12.90+0.06W 0.93 0.65
Vol (ml) vs. Wt (g)79
V=4.65+0.94W 0.97 6.26Wt (g) vs. VlL (ml/em) W=-33.30+26.64(V/L) 0.94 8.96Vol (ml) vs. W/L (g/em) V=-20.21 +23.42(W/L) 0.93 8.58Length (em) vs. WN (g/ml) L=1 0.87+7.67(W1V) 0.09 2.24
Overall Length (em) vs. Vol (ml) L=12.70+0.07V 0.91 0.67Length (em) vs. Wt (g) L=12.82+0.06W 0.93 0.62Vol (ml) vs. Wt (g)
99V=4.12+0.94W 0.96 0.17
Wt (g) vs. VlL (ml/em) W=-32.14+26.34(V/L) 0.94 8.45Vol (ml) vs. W/L (g/em) V=-21.20+23.51 (W/L) 0.93 8.67Length (em) vs. WN (g/ml) L=11.29+7.04(W1V) 8.10 2.16
37
Table 3.3 Average, standard deviation and minimum and maximum values for nutrientcomponent results ofPacific herring - overall and by lots and locations.
LOT
Protein % EE % Ash % CHO% Moisture % GE(kcal/g)
A Avg 52.33 a 31.23 c 8.44 a 7.18 a 72.40 a 6.25 c
(n =29) StDev 4.99 5.51 1.13 2.55 2.84 0.28
Min 39.01 20.66 6.05 2.71 67.94 5.48
Max 64.08 40.32 10.88 11.71 80.26 6.71
B Avg 54.52 a 35.53 b 7.71 b 2.49 c 71.03 a 6.55 b
(n =28) StDev 4.64 4.54 0.99 2.38 1.95 0.22
Min 45.89 24.10 5.92 0.00 68.32 5.98
Max 63.83 42.56 10.02 8.96 75.72 6.93
C Avg 44.43 b 43.39 a 6.92 c 5.26 b 68.24 b 6.83 a
(n =22) StDev 5.21 6.36 1.05 1.29 2.67 0.30
Min 37.02 27.50 5.47 3.06 63.29 6.08
Max 59.11 51.85 9.18 8.56 74.60 7.220 Avg 42.80 b 45.73 a 6.57 c 5.11 cfb 66.26 b 6.95 a
(n =20) StDev 3.38 5.57 0.70 3.37 3.26 0.30
Min 35.90 35.55 5.40 0.00 59.88 6.43
Max 48.59 56.81 8.08 11.55 71.93 7.60
LOCATION
Alaska Avg 42.80 b 45.73 a 6.57 b 5.11 66.26 b 6.95 a
Max 64.08 56.81 10.88 11.71 80.26 7.52a, , C indicate significant differences (P<O.001) between means within lots and within locations for eachnutrient.
38
Table 3.4 Average, standard deviation and minimum and maximum values for nutrientcomponents ofPacific herring by size and sex (P >0.05 for all).
Table 3.5 Linear regression equations and R2 values (>6QO.Io) nutrient componentrelationships in Pacific herring analyte correlations (n=99, P<O.OOI).
Nutrient content Regression equation R2
Dry Matter Basis
Ash vs GE (k Ash =25.55 - 2.73 * GE 0.72
P ein (EE) Prot =75.93 - 0.70 * EE 0.68
Protein vs. Moisture Prot =-58.42 + 1.54 * Moist 0.65
lipid (EE) VS. Ash EE =78.32- 5.35 * Ash 0.68
lipid (EE) vs. Moisture EE =173.29 - 1.94 * Moist 0.74
lipid (EE) VS. GE (kcaVg) EE =-98.11 + 20.64 * GE 0.97
Ash vs. Moisture Ash =-11.51 + 0.27 * Moist 0.62
Moisture vs. GE (kcaVg) Moist =120.93 -7.74 * GE 0.69
As-Fed Basis
Moisture ) Moist = 90.23 -10.18 *GE 0.96
lipid (EE) g} EE = - 9.82 + 10.76 * GE 0.97
lipid (EE) ¥s. Moisture EE = 80.69 - 0.99 * Moist 0.88Shaded areas indicate "best fit" regression equation to use for predicting nutrient content,based on r-squared value.
Table 3.6 Sequence oflinear regression equations used to predict energy density (GE(kcallg», lipid (%), ash (%), and protein (%) content ofPacific herring starting frommoisture content.
Nutrient contents Basis Fonnula
to calc. GE
to calc. lipid
convert to lipid (Dftf)
convert to GE (Dftf)
moisture
GE
lipid (as fed)
GE(asfed)
GE
lipid
to calc.
to calc.
ash
protein
8S fed GE =8.59 - 0.09 * Moist
as fed EE =-9.82 + 10.76 * GE
OM lipid (OM) = Hpid (as fed) / ((1~Moist)/1()())
OM GE (OM) =GE (as fed) / ((1~st)/1()())
OM Ash =25.55 - 2.73 * GE
OM Prot =75.93 - 0.70 * EE
40
Table 3.7 Calculated proximate composition of samples from a previous study (usinglinear regression equations to predict analyte content) compared to known chemicalreference values.
PacificHerring
AtlanticHerring
AVG Overall
SO +\-
AVG Pacific
SO +\-
AVG Atlantic
SO+\-
% Prot %EE %
41
% Ash GE(kalllg) (kcaUg)
(Chem)
6.65
6.66
6.63
6.66
6.66
6.46
6.99
6.90
7.12
5.59
5.59
5.42
5.99
5.81
5.89
5.56
5.61
5.92
6.23
0.57
6.75
0.21
5.71
0.20
Table 3.8 Linear regression R2 values, means and standard deviations (SD) forcalculated (using linear regression equations to predict analyte content) and knownchemical reference results.
Iii! Female o Male ID Female wi eggs I• Denotes significant difference ofP<O.05.
Figure 3.9 Bar graph showing mean length, volume and weight for Pacific herring sexcategories of female, male and female with eggs.
Morphometric Averages by Sex
6.00
5.00
*4.00
•iI 3.00or:(
2.00
1.00
0.00VII.. (mlIcm) Wll.~)
Morphometric rneuurement
WN(~
III Female o Male IIFemale wi eggs I• Denotes significant difference ofP<O.05.
Figure 3.10 Bar graph showing mean volume per unit length (mVcm), weight per unitlength (glcm) and density (glml) for Pacific herring sex categories -offemale, male andfemale with eggs.
47
Morphometric Averages by Size
•120 t---------------"-.-------.......----.--1
100 +--------------t•
t 80
~ 60-f---------...........-f
40+----------
20
Length (an) VoIl.me (mI)
MorphometrIc measurement
II Small 0 Medium
• Denotes significant difference ofP<O.05.
Weight (g)
IDLarge
Figure 3.11 Bar graph showing mean length, volume and weight for Pacific herring sizecategories ofsmall, medium and large.
Morphometric Averages by Size
7
6
5IIal4l!•~ 3
2
o
......-:E- r--W-. .
~ ---:-.. ..
r----~ ~-,--.",
~
== - -'-, ~
-----;;;;;;;
r---- := ~
~ I r= -~....z;
VIL (milan) WIL~) WN~
Morphometrtc meuurwnent
B Small 0 Medium II Large
• Denotes significant difference ofP<O.05.
Figure 3.12 Bar graph showing volume per unit length (mVcm), weight per unit length(g1cm) and density (g1ml) for Pacific herring size categories ofsmall, medium and large.
Figure 3.13 Linear regression plot ofmoisture versus gross energy (GE) content on an asfed basis (R2=O.96) for Pacific herring.
Regression PlotEE .... -9.82{5l + 10.7!i94 GE (kca'g)
S .. 0.637754 R-Sq" 97.1 .. R-Sq(8lI .. 97.0 ..
25
"*' 15WW
5
2
GE (kcallg)3
Figure 3.14 Linear regression plot ofgross energy (GE) versus crude lipid (EE%)content on an as fed basis (R2=0.97) for Pacific herring.
49
Regression Plot/WI ..=25.5511 - 2 nr:o (£ (lafg)
S =QeB2303 R-Sq =71.8" R-Sltact =71.5"
11
10
9
;j.~
8enc(
7
6
5
55
••. .
• •.. .••• ••. .. , . .. .
6.5
GE (kcallg)75
Figure 3.15 Linear regression plot ofgross energy (GE) versus ash content on a drymatter basis (R2=O.72) for Pacific herring.
Regression PlotProBn"" • 75.9329· 0.7OO28ll EE ..
S:z: 3.84511) RSq .. 67.9 "" R-S«-e .. 67.5 ""
1lO40
EE%3020
35 ~---_,~--_,._---~---r'
65 -r----------------,
55'#.c:jeQ.. 45
Figure 3.16 Linear regression plot ofcrude lipid (EE) versus protein content on a drymatter basis (R2=O.68) for Pacific herring.
50
CHAPTER 4. UTILIZATION OF VISIBLE AND SHORT-WAVELENGTH NEARINFRARED REFLECTANCE SPECTROSCOPY TO PREDICT THE PROXIMATECONTENT OF WHOLE, HOMOGENIZED PACIFIC HERRING (Clupea pallasi)
4.1 INTRODUCTION
4.1.1 Diet Assessment in the Marine Mammal Field
Numerous marine parks in the United States house and exhibit marine mammals
such as Steller sea lions, Hawaiian monk seals and Harbor seals. One of the main
components of the diets of these animals is herring. The quantity ofherring fed at any
given time depends upon the proximate analysis results ofthe herring being fed. Other
commonly fed species offish include capelin (Mallotus villosus) and pollock (Theragra
chalcogramma). The main species ofherring fed in marine parks located on the West
Coast (including Hawaii and Alaska) is Pacific herring (Clupea pallasi).
In most marine mammal facilities, diet assessment is a key part of daily
husbandry. In a facility that houses an endangered or threatened species of marine
mammal, diet analysis is more critical. Diet analysis becomes increasingly more
important ifvital research is being performed using the animals at the facility. Having
marine mammals in captivity facilitates needed research on the behavior, physiology and
nutrition of marine mammals. Many times, one or more components of a study will rely
upon accurate diet analysis.
For this analysis, samples are usually sent to an outside laboratory. Nutrient
analyses are run to determine crude protein, crude lipid, total mineral and moisture
content. These analyses can be costly and results can take a week or more to obtain.
51
4.1.2 Near Infrared Reflectance Spectroscopy
Near Infrared (NIR) Reflectance spectroscopy is an expanding technology which
has recently grown in popularity. In the livestock feed and forage industry, it has been
widely used and well accepted for the prediction ofnutrient composition used in diet
analysis (Williams and Norris, 1987). This technology has the ability to provide both
quantitative and qualitative information on a wide array of products (especially raw
materials) in a matter of minutes (Burns and Ciurczak, 1992).
In this study visible and short-wavelength near-infrared (SW-NIR) reflectance
was used. This spectroscopy uses wavelengths from 600 to 1100nm. SW-NIR
spectroscopy yields good quantitative results, especially for higWy scattering samples
(Phelan et ai., 1989; Cavinato et al., 1990). This technology also permits the remote
collection ofa full spectrum with a fiber optic probe (Lee et ai., 1992; Huang et ai.,
2000). There are, however, some limitations ofSW-NIR. SW-NIR should be used to
predict analytes that are at concentrations greater than one percent, due to lower
sensitivity. This type of spectroscopy also has a limited scope of functional groups as
well as a tendency for absorption bands to overlap. When this happens, one or more
interfering peaks may overlap a peak of interest. Because of this, complicated calibration
models are sometimes required (Lee et ai., 1992; Huang et ai., 2000).
4.1.3 NIR Spectroscopy and the Fish Industry
The use of NIR spectroscopy in the fish industry has grown in recent years. This
technology is used mainly for quality control purposes (i.e., in cultured salmonids)
leading to a higher quality of fish in the market. Recent studies looked at minimizing
52
sample preparation time by scanning whole, intact fish. In a study done by Downey
(1996), NIR spectroscopy (700-11 OOnm) was used to predict the moisture and oil content
offarmed salmon. Scanning was done through the skin on intact fish. R2 values in this
study were fairly similar to those seen herein. They ranged from 0.69-0.77 for moisture
and 0.70-0.74 for lipid content. Intact fish were also scanned in studies done by Lee et
al. (1992), Isaksson et al. (1995) and Wold and Isaksson (1997). In some studies, sample
preparation was minimized, but not removed completely. In one study by Rasco et al.
(1991), cross sections of frozen plugs taken from cultured rainbow trout were scanned.
This method was able to predict, with fairly good accuracy, fat, moisture and protein
content. In another study done by Wold et al. (1996), cylinders of intact tissue were
taken from various sites on the bodies of ten farmed salmon. The R2 results for fat
prediction were greater than 0.90 for all sites except one (0.44 for the plug containing
skin and dark muscle). Other studies went further in preparing samples, homogenizing or
grinding them prior to spectra collection. One such study was done by Sollid and Solberg
(1992), who ground cultured Atlantic salmon fillets before scanning. This study used
NIR spectroscopy (850-1050 nm) to predict the fat content ofground salmon. Prediction
results were exceptional with an R2 value of0.98. Ground fish were also scanned in
studies done by Isaksson et al. (1995), Valdes et al. (1997), and Zhang and Lee (1997).
In a study done by Darwish et al. (1989), fish were processed and ground to milk-like
consistency prior to analysis via mid-infrared transmittance spectroscopy. The method of
sample preparation used in this study was extensive and time consuming.
The objective of this project was to use visible and short-wavelength near-infrared
(SW-NIR) reflectance spectroscopy to develop chemometric calibration models for the
53
prediction of fat, protein, moisture and mineral content ofwhole homogenized Pacific
herring (Clupea pallasi).
4.2 MATERIALS AND METHODS
4.2.1 Sample Preparation
The Pacific herring samples (n = 86) used were from two fisheries - North Bay
Meat Company which sells herring caught in British Columbia, Canada (lots A, B, and
C), and Petersburg Fisheries in Alaska (lot D). Lots represent separate seasonal catch
dates within a three-year span (1999-2001). Specific year and/or season data were not
available. The Marine Mammal Research Program ofthe University ofHawaii donated
the North Bay herring. Samples for those lots were prepared at their fish preparation
facility at Kaneohe Marine Base Headquarters, Hawaii. The Alaska herring was donated
by the Alaska SeaLife Center in Seward, Alaska. All samples from the fisheries came
individually quick-frozen. Samples were randomly selected using the methods described
in the previous chapter, and shipped, still in the frozen state, to Hawaii via overnight
FedEx. Lots were stored at -80°C. Samples were prepared using the methodology
described in chapter three.
To prepare samples, the herring were first thawed in a cool fresh water bath, then
put into a clean stainless steel pan. Herring were homogenized in a glass container with a
pouring lip, using a 12-speed Oster blender with a 450-watt motor. The blending process
for each fish took approximately 10min. since time was spent scrapping down the sides
of the container as blender speed was increased from "chop," to blend and finally to
"mix." The thorougWy mixed samples were then scooped out with a stainless steel spoon
54
and placed into 50 ml disposable sterile centrifuge tubes. Whole herring were
homogenized because the nutrient make-up of the entire fish was needed. Since marine
mammals are fed the entire fish. The sample tubes were stored at -80°C.
4.2.2 Reference Analysis
Chemical proximate analysis was conducted at the University ofHawaii Animal
Science nutrition laboratory. Tests performed were those to determine moisture, ash,
crude protein and crude fat content of the herring. Standard methods of the Association
ofOfficial Analytical Chemists (1990) for meat were used (Ellis, 1984). Duplicate
analyses were run for each component. To determine moisture content, samples were
placed in ceramic crucibles and dried for 24 hours in a 100°C drying oven. Ash content
was determined on dried samples by combustion in an electric muffle furnace for 8 hours
at 650°C. Crude protein content was determined using a modified Kjeldahl nitrogen
analysis (see chapter three for more specifics). Crude fat content was determined using
samples dried at 50°C for 48 hours prior to analysis using the Goldfisch ether extract
analysis method.
4.2.3 Spectra Collection
All homogenized herring samples were thawed to room temperature on a
laboratory bench with an ambient temperature ofapproximately 21°C. The temperature
ofall samples was kept constant allowing for optimal NIR scanning. Samples were
thorougWy mixed, and approximately 5 g sub-samples were taken from each sample tube.
This sub-sample was then placed in direct contact with the detector on the NIR probe.
55
This amount of sample covered the entire detection window ofthe detector. Once placed
on the detector, the top of each sample was smoothed over to ensure that no air pockets
existed between the detector and sample. Visible and short-wavelength near-infrared
reflectance (SW-NIR) spectra were recorded in diffuse reflectance mode with a GMS
spectrophotometer (Textron Systems, Wilmington, MA). Spectra were collected from
600 to 1100 nm at 0.5 nm intervals. Each spectrum was the average of20 scans with
30ms exposure time for each scan. The detector was wiped and cleaned thoroughly with
distilled water between samples. Three replicate scans were carried out for each sample.
4.2.4 Statistical Analysis
Spectra were scanned using the D2TSC 3.2 software (Figure 4.1). The spectral
data was analyzed using DeLight 3.2 software (DSquared Development, Inc., La Grande,
OR). This software is designed for chemometric analyses. The spectra were transformed
using the pre-processing algorithms ofbinning, smoothing, and second derivative
transformation (Figure 4.2). Binning is the averaging of a number of points into one.
This increases the signal to noise ratio by a factor of~(n). Smoothing takes the value of
each point and replaces it by the mean of the values in the interval surrounding it.
Smoothing enhances the signal to noise ratio without decreasing the spectral resolution.
Second derivative transformation removes baseline shifts and is able to separate out
overlapping absorption bands. It also minimizes scatter effects due to differences in
particle size (Valdes et al., 1997). First, spectral data were binned by 2 nm and smoothed
with a Gaussian function over 12 nm. Then a second derivative transformation with a 12
nm gap was calculated (Lin et al., 2003 a, b).
56
Regression equations (calibration models) are then calculated based on the NIR
spectra obtained and the known analyte information. The model was used to predict
nutrient composition ofunknown samples. Multiple Linear Regression (MLR) and
Partial Least Squares (PLS) statistics are most commonly used for linear calibration
models. In this study, calibrations were developed using the full NIR spectrum by
applying a PLS regression method to statistically analyze the data. The PLS method uses
full spectrum analysis (Haaland and Thomas, 1988). Initially, the calibration model was
built using one to ten latent variables for each sample component. This was done for
each lot (A-D), and then by location (Canada and Alaska). The software calculated the
R2 value and the standard error ofprediction (SEP) for each regression equation. The
standard error of prediction is indicative of the predictive performance of calibration
models (Lin et at., 2003). The best-fit calibration model was then determined by
graphing the number of latent variables by the SEP using Microsoft Excel 2000. The
next step was to test the validity and the prediction accuracy of the calibration models
using cross validation. In this study, leave-one-out cross-validation was run, with the
software, using the best-fit calibration model to predict each nutrient. In this type of
cross validation, one sample is left out while the rest are used to predict the analyte of
interest within that sample. An advantage ofcross validation is that the sample is not
included in the calibration model and therefore, the model can be tested independently
(Murray and Cowe, 1992). This method ofvalidation also allows for the maximum use
ofall samples. Instead ofdividing the available samples into calibration and validation
sets, all samples are used in both sets with leave-one-out cross validation. This is
especially helpful when sample numbers are limited.
57
4.3 RESULTS
4.3. 1 Reference Results
Calibration models were created for each nutrient or chemical component in five
categories: lots A through D separately and lots A, B, and C combined. The combination
oflots A through C represent fish from the North Bay Meat Company (Canada) and lot D
represents fish from Petersburg Fisheries (Alaska).
First, location was analyzed to determine whether sample composition between
locations were different enough to warrant separate calibration models. Mean values
were compared between the Canada and Alaska herring. Values for each location were
found to be statistically different with p-values less than or equal to 0.001 (Figure 4.3).
For example, lipid content between locations showed a variation in range with herring
from Canada with 23.3-51.9% lipid and those from Alaska, 35.6-56.8% lipid.
Lots were also analyzed to determine whether nutrient contents between them
were statistically significant. Mean nutrient values between lots A through D were
significantly different with p-values less than 0.001 (Figure 4.4). Differences between
lots A through C and lot D were not seen. Therefore, it can be inferred that although lots
are somewhat nested within location, the effects of each on sample composition are
separate. For calibration modeling purposes, the more feasible of the two to use would
be location, a category with a broader range than lot.
Linear regressions were run on all the chemical constituents to determine whether
or not correlations existed between them (Table 4.1). Only adjusted R2 values are cited
herein. These values represent R2 values adjusted for degrees of freedom. There was a
negative correlation found between moisture and gross energy (GE) on a dry matter basis
58
(R2=0.69). When the as-fed data were used, the resulting R2 value increased to 0.96
(Figure 4.5). There was a highly significant positive correlation (R2=0.97) between lipid
and gross energy content on a dry matter basis. When the regression was run using an as
fed basis, the correlation remained the same with an R2 value of 0.97 (Figure 4.6). On
the dry matter basis, ash was negatively correlated with gross energy, with an R2 value of
0.72 (Figure 4.7). Based on an R2 value of0.68, protein was best correlated with lipid on
a dry matter basis (Figure 4.8). These linear regression equations could be used to later
calculate gross energy (GE (kcal/g)), lipid, ash, and protein content based on a fairly easy
to obtain moisture content (Table 4.2).
4.3.2 SW-NIR Spectral Results
Calibration models were established and selected with the optimal number of
latent variables. The latent variables (x-axis) were graphed against the Standard Error of
Prediction (SEP) (y-axis) for each component. Graphs for protein and lipid can be seen
in Figures 4.9-4.14. These graphs were used to select the best-fit calibration model for
each component. Usually, the lowest point on the line was chosen as the best-fit
calibration model. This point indicated the optimal number of latent variables included
in the calibration equation, as well as the lowest SEP (in most cases). The number of
latent variables chosen for sample constituents in this study ranged from five to eight.
Table 4.3 shows the number oflatent variables and the SEP for the calibration models
chosen for each chemical component for alilotsllocations.
The mean nutrient content predicted using the SW-NIR calibration models chosen
(predicted) can be seen in Table 4.4 along with the mean results obtained with the
59
reference method (chemical/analytical). When linear regression analyses were done for
all components comparing the predicted and analytical results by both lot and location, R2
values ranged from 0.24-0.93. When looking at lots A through D, lot A had, on average,
the lowest R2 values for protein, lipid, ash, dry matter and moisture. However, lot Chad,
overall, the highest R2 values for components. The best positive correlation was found in
lipid content in lot C with an R2 value of0.93 (Figure 4.15). Similarly, the correlation in
ash content between the chemical and NIR predicted data in lot D, or Petersburg (Alaska)
herring, was fairly high, with an R2 value of0.91 (Figure 4.16).
When the R2 values for locations were compared, the calibration for Alaska
herring was, on average, better than that ofthe British Columbia herring. For Alaska
herring, R2 values ranged from 0.66-0.91, for British Columbia, from 0.47-0.81. These
low R2 values were most likely due to the inclusion ofthe highly variable data results
from lot A.
4.3.3 Calculated versus SW-NIR Predicted Results
Nutrient content was also calculated using the linear regression equations listed in
Table 4.2. These calculated values for various chemical components were compared to
the results predicted using the SW-NIR calibration equations. Gross energy content
(kcallg) for those NIR predicted results was calculated using the equation (Pond, et ai.,
Linear regressions were run comparing both the calculated and SW-NIR predicted
results to the original chemical analysis results. This was done for all components
(protein, lipid, ash, dry matter, moisture and gross energy) for both locations, as well as
overall. Table 4.5 shows the results of these linear regressions. All comparisons were
significantly different with p-values less than 0.001. When looking at all ofthe overall
data, R2 values for SW-NIR predictions ofprotein (R2=0.75), lipid (R2=0.86) and ash
(R2=0.81) were higher than those of the calculated results (R2=0.64, 0.79, and 0.68,
respectively). SW-NIR calibration models predicted protein content consistently better
(overall R2=0.75; Canada R2=0.67; Alaska R2=0.70) than did linear regression
calculations. However, the calculation method seemed to predict ash content better than
SW-NIR for the overall data (R2=0.69), as well as by location (Canada R2=0.66; Alaska
R2=0.70). When R2 values were compared, SW-NIR yielded higher R2 values overall.
SW-NIR calibration models, on average, predicted content for herring with a slightly
higher degree of accuracy than did the calculation method.
4.4 DISCUSSION
4.4. 1 Reference Results
Reference results for both locations and lots were similar to those seen in the
previous chapters. Virtually the same data were used for both studies, however several
data points were left out for the NIR calibrations. This was because there was not enough
sample volume left to scan on the SW-NIR measurement. Calibrations for chemical
components varied by location and were significantly different for all components tested.
61
When reference results for nutrients were compared to one another, several
interesting correlations were found. The highly significant positive correlation (R2=0.97)
between lipid content and gross energy was similar to that found in other studies
(Castellini et at., 200 I). A significant negative correlation on both dry matter and as-fed
basis (R2=0.74 and 0.88, respectively) was found between moisture and lipid content.
The same type of correlation was also found between gross energy and moisture content
on a dry matter and as-fed basis (R2=0.69 and 0.96, respectively). The same two
correlations were found in a study done by Anthony et at. (2000). This study also found
slightly significant correlations between ash content and gross energy, and between
protein and lipid content (R2= 0.72 and 0.68, respectively).
4.4.2 Spectral Analysis
Partial least squares (PLS) regression method was used to establish the SW-NIR
calibration models. With PLS, the entire spectrum is used (600-1100 nm); therefore,
variations between the samples will be built into the calibration model (SoUid and
Solberg, 1992). The establishment of calibration model depends on the selection of
optimal number oflatent variables. To avoid over-fitting, 5-8 latent variables were
selected for prediction for different models. This resulted in an average standard error of
prediction (SEP) of2.27 for nutrient analysis. Over-fitting is counter-productive fitting
by the regression to the noise in the data (Murray and Cowe, 1992).
62
4.4.3 SW-NIR Calibration Models
The calibration models that would be applicable to the marine mammal field are
those for each location. Lots are specific to not only a location, but also a catch date.
Because a separate calibration model cannot feasibly be created for each lot for each
location, the closest calibration would therefore be location. Also, this study, as well as
previous studies, has found significant differences in nutrient content in feeds between
locations. One study done by Anthony et al., (2000) showed that Pacific herring caught
in Prince William Sound (PWS) and Lower Cook Inlet, Alaska were significantly
different in lipid content (41±1.9% vs. 32±2.7%). Their study also showed that herring
within PWS were significantly different in their lipid contents (29±0.1% vs. 21±1.3%),
depending upon where in PWS they were located (e.g., northeast versus southwest). The
calibration model for combined lots A, B, and C (Canada herring) can be used to analyze
Pacific herring caught in the area ofBritish Columbia, Canada. The model created for lot
D (Alaska herring) could be used to analyze Pacific herring caught in southeast Alaska,
where Petersburg Fisheries catch their fish.
Using the R2 values from the linear regressions, the calibration for southeast
Alaskan Pacific herring would better predict the nutrient content of herring than the
calibration for British Columbian Pacific herring. The Alaskan calibration model could
best predict protein (R2=0.86) and ash (R2=0.91). However, the levels oflipid and
moisture could also be predicted fairly accurately (R2=0.66 and 0.70, respectively). This
model could predict the nutrient content ofPacific herring caught in southeast Alaska
with a fairly high degree ofaccuracy. Overall, the calibration model for the Canada
herring is fairly good. The best prediction with this model would be oflipid content
63
(R2=0.81), then moisture and ash (R2=0.63 and 0.72 respectively). The lowest accuracy
in prediction would be for protein, with an R2 value of0.47. This model could be used to
predict the nutrient content of fish if approximations were acceptable. These correlations
are similar to those found in a study done by Lee et al., (1992). In this study, SW-NIR
was used to predict crude lipid content in intact rainbow trout. Using three latent
variables in the PLS method, the resulting R2 value was 0.81.
There are only a handful of studies involving the homogenization of fish prior to
NIR scanning. In one study done by SoUid and Solberg in 1992, homogenized Atlantic
salmon fillets were used. Samples in this study were scanned with near infrared
transmission spectroscopy in diffuse mode, and fat was measured. Prediction accuracy
was good with an R2 value of 0.99. Valdes et al. (1997) conducted another study using
homogenized samples. In their study they were looking at the utilization ofNiR
spectroscopy for the rapid analysis of fish being used as feed, however individual
calibrations for each species of fish were not created; instead, nine species offish and
cephalopod were combined and homogenized. Herring was one ofthose species and
made up 24% ofthe samples that went into the final homogenous mixture. The final
calibration models for fat and protein in this study were good with R2 values greater than
0.90. Researchers from this study suggested that NIR should be investigated further for
analysis of nutrient contents ofzoo feeds.
With the methodology used herein, whole herring could be sampled, processed
(homogenized) and scanned quickly yielding a fairly accurate result. Errors in
calibration, and the resulting lower R2 values may be the result ofnon-homogeneity
within the samples. Because of the use ofNIR spectroscopy (as opposed to NIT (near
64
infrared transmittance) spectroscopy), samples need to be uniformly homogenized (Sollid
and Solberg, 1992). Whole fish were ground and blended to a fairly homogenous
consistency. Homogenization was not easy as fish have high moisture content, and too
much homogenizing could result in fat and water separation (Sollid and Solberg, 1992).
Once samples were homogenized however, tiny pieces of skin, bones and scales were
still evident. Error in these models could also be due to the small number of samples
making up each calibration model, especially for lots (n=20-24). Outliers were apparent
in all calibrations. These outliers indicate that further research should be performed on
these samples. Additionally, due to slightly high SEP's, more samples should be
included in the calibration sets to account for greater variability in chemical composition
of the herring samples and to improve the precision of this method (Lee et al., 1992).
4.4.4 Calculated Nutrient Content
Although the R2 values for the linear regression equations for component-to
component prediction were fairly high (0.68-0.97), this method was not as accurate as
SW-NIR spectroscopy. Linear regressions were run to compare NIR predicted and
calculated results with the chemical reference results. All R2 values were compared.
There were instances where calculated nutrient contents were better predicted than those
by NIR. Overall, however, NIR spectroscopy was better at predicting the nutrient content
ofwhole, homogenized Pacific herring.
65
4.5 CONCLUSION
Based on the results of this study, SW-NIR spectroscopy could be used very
effectively to predict the nutrient content ofwhole, homogenized Pacific herring.
Resulting calibration models for both Alaska and Canada herring could be used to predict
nutrient content of southeast Alaskan and British Columbian herring. R2 values were not
as high as in other studies; however, this method could still be used where approximate
results are needed. It should be noted that the strengths of SW-NIR spectroscopy lie in
its ability to predict nutrient content quickly and easily. With the repeated scannings
during analysis and the fact that there are 20-30 scans per sample, the accuracy and
repeatability of analysis would be much greater.
66
Table 4.1 Linear regression equations and R2 values (>600./0) nutrient componentrelationships in Pacific herring analyte correlations (n=99, P<O.OOl).
Nutrient content Regression equation R2
Dry Matter Basis
h vs. GE (kcaU ) Ash =25.55 - 2.73 * GE 0.72
Prot n . Lipid (EE) Prot =75.93 - 0.70 * EE 0.68
Protein vs. Moisture Prot =-58.42 + 1.54 * Moist 0.65
Lipid (EE) vs. Ash EE =78.32- 5.35 * Ash 0.68
Lipid (EE) vs. Moisture EE =173.29 - 1.94 * Moist 0.74
Lipid (EE) vs. GE (kcallg) EE =-98.11 + 20.64 * GE 0.97
Ash vs. Moisture Ash =-11.51+ 0.27 * Moist 0.62
Moisture vs. GE (kcallg) Moist =120.93 - 7.74 * GE 0.69
As-Fed Basis
M Moist = 90.23 -10.18 *GE 0.96
Lipid (EE) EE =-9.82 + 10.76 * GE 0.97
Lipid (EE) vs. Moisture EE =80.69 - 0.99 * Moist 0.88Shaded areas indicate "best fit" regression equation to use for predicting nutrient content, based onr-squared value.
Table 4.2 Sequence of linear regression equations used to predict energy density (GE(kcal/g», lipid (%), ash (%), and protein (%) content ofPacific herring starting frommoisture content.
GE (DAf) OM GE (OM) =GE (as fed) / ((1QO-Moist)/100)
ash OM Ash =25.55 - 2.73 * GE
protein OM Prot = 75.93 - 0.70 * EE
67
Table 4.3. Number oflatent variables and standard error of prediction (SEP) for each1"1 • .. 1 1 1 • t'" .. ... 1 .. • .. t'" T" .. ~ 1 ..
calloranon moael cnosen per component ror eacn lOt ana IOcanon Ot yacmc nemng.
Latent Variables SEP R2
Lot A (n =20)Protein 5 9.91 0.48
Lipid (EE) 8 3.07 0.74Ash 8 0.84 0.24
Moisture 8 1.67 0.27Dry Matter (OM) 8 1.67 0.27
Lot B (n =24)Protein 7 3.08 0.79
Lipid (EE) 6 3.22 0.71Ash 8 0.72 0.64
Moisture 8 1.08 0.82Dry Matter (OM) 8 1.08 0.82
Lot C (0 =22)Protein 6 3.63 0.87
Lipid (EE) 6 2.79 0.93Ash 6 0.63 0.72
Moisture 6 1.3 0.88Dry Matter (OM) 6 1.3 0.88
Lot D or ALASKA (n =20)Protein 5 2.17 0.86
Lipid (EE) 5 2.85 0.66Ash 5 0.5 0.91
Moisture 5 1.57 0.7Dry Matter (OM) 5 1.57 0.7
Lots A-C or CANADA (0 = 66)Protein 6 6.8 0.47
Lipid (EE) 7 2.89 0.81Ash 7 0.7 0.63
Moisture 7 1.39 0.72Dry Matter (OM) 7 1.39 0.72
68
Table 4.4. Actual chemical and SW-NIR predicted results for feed components ofPacific herring.
LatA Avg 52.13
n=20 StDev 3.81
Min 45.03
Max 58.46
Lat B Avg 54.08
n=24 StDev 4.79
Min 45.89
Max 63.83
LatC Avg 44.43
n=22 StDev 5.21
Min 37.0
Max 59.11
LatD Avg 42.80
Alaska StDev 3.38
n=2O Min 35.90
Max 48.59
ABC Avg 49.5
Canada StDev 8.73
n=66 Min 37.03
69
Table 4.5. R2 value comparing SW-NIR predicted and calculated (using regressionequations) values with those from chemical analyses ofPacific herring.
LOTS A, B, & C - Canada (n =66)ProteinLipid (EE)AshMoistureDry Matter (OM)
Protein
LOT 0 - Alaska (0 =20)ProteinLipid (EE)AshMoisture
Dry Matter (OM)
Protein
OVERALL (n =86)ProteinLipid (EE)Ash
MoistureDry Matter (OM)
Protein
NIR PredictedR2
67.4081.6062.7071.9071.9084.60
69.6085.9061.5091.4091.4080.40
74.8086.2067.8083.7083.7081.00
70
CalculatedR2
58.9070.3065.60
77.10
42.6091.2069.70
87.60
64.4079.4070.10
67.60
02
015
0.1
• o £15uIii-e0... 0~
-005
·0.1
-015
600 700 eoo 900Wav length (ron)
1000 1100
Figure 4.1. Short-wavelength Near Infrared (SW-NIR) spectra for lot A ofPacificherring.
Lipid
.········•
. .·••··•···· .... ...-
Water
Protein
-00001
n rTfll
0.00014
Ot.U.1J6
0.0000El
O.CKXlO2
-000012
600 700 1llOO 1100
Figure 4.2. Second derivative transformation ofSW-NIR spectra for lot A ofPacificherring. Boxed areas indicate most common regions for analyte measurements (analytesas indicated).
71
Average Proxi..... Results by LoeBtion
80
70
60
50
)40(J3O
20
10
o
*l1li
*I- f.I- •~I- ,.----
~ -I- ~ * *
I ~ I ~I ~}J1
Ash %
Analyte
GE(kcallg)
o Canada mAlaska
* indicates significant differences (P~ 0.(01) between locations for each nutrient
Figure 4.3. Average chemical reference results for components in Pacific herring byharvest location.
Average Proximate Results by Lot
80
70
60
50
J~30
20
10
o
.!.~b
'a~~
a a ~,-r-- ~-
'i~ ~~ I- 'g
I c..:..i- - -'"
- ~ -
f--.:I b ;a-I I [fl n:.rn ..... ~-'"-
[I ~ IAsh %
Analyte
Mlisue% GE(kcaIIg)
o Lot A o Lot B elate .LotD
.. b. C indicate significant differences between lots for each nutrient
Figure 4.4. Average chemical reference results for components in Pacific herring by lot.
Figure 4.7. Linear regression plot ofgross energy (GE) versus ash content on an as fedbasis (R2=O.72) ofPacific herring.
Regression PlotP1dein '" .. 75.9329- 0.7t1J288 EE '"
S .. 3.845l!IO RoSq .. 61.9 '" ~ .. 61.5 '"
65
55
'#.c:!e0. 45
35 ~----...,.------.------r-----~
20 30 40
EE %50 60
Figure 4.8. Linear regression plot oflipid (EE) versus protein content on a dry matterbasis (R2=O.68) ofPacific herring.
74
Protein: Latent Variables va. Standard Error of Prediction(SEP)
14.00
12.00
10.00
Q. 8.00wfI) 6.00
4.00
2.00
•..~ .L~
.....
-
-----------...~ ~ • !I I •..
1210842 6
Latent Variables
r ----+- Lot A • Lot B ----A- Lot c=-J
0.00o
Figure 4.9. Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for protein in Pacific herring lots A (LV=5), B (LV=7), and C (LV=6).
Lipid: Latent Variables vs. Standard Error of Prediction(SEP)
7.00
6.00
5.00
Q. 4.00wf/) 3.00
2.00
1.00
-
~
=--------",. &
...~~ -.-- -' • .....- -
0.00o 2 468
Latent Variables
10 12
----+- Lot A • LotB ----A- Lot C
Figure 4.10. Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for lipid in Pacific herring lots A (LV=8), B (LV=6), and C (LV=6).
75
Lot D Protein: Latent Variables vs. Standard Error ofPrediction (SEP)
\\\~
.\" .. ~ -....
3.40
3.20
3.00
D. 2.80wrn 2.60
2.40
2.20
2.00
o 2 4 6
Latent Variables
8 10 12
Figure 4.11. Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for protein in Pacific herring lots D (LV=5).
Lot 0 Upid: Latent Variables va. Standard Error of Prediction(SEP)
5.50
5.00
4.50
D. 4.00wrn 3.50
3.00
~
~~
~ .......... ..---2.50
2.00o 2 4 6
Latent Variables
8 10 12
Figure 4.12. Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for lipid in Pacific herring lots D (LV=5).
76
Lot A, B & C Cormined Protein: Latent Variables vs. StandardError of Prediction (SEP)
\
\~ ......,
~ /~-.r
~
9.00
8.50
8.00C1.wUJ
7.50
7.00
6.50o 2 4 6
Latent Variables
8 10 12
Figure 4.13. Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for protein in Pacific herring lots A, B, & C combined (LV=6).
Lot A, B & C Contined Upld: Latent Variables va. StandardError of Prediction (SEP)
.....~
•~~ -
.~
...... • • •
6.00
5.50
5.00
C1. 4.50wUJ 4.00
3.50
3.00
2.50o 2 4 6
L8tent Variables
8 10 12
Figure 4.14. Graph oflatent variables (LV) and SEP's to choose the best-fit calibrationmodel for lipid in Pacific herring lots A, B, & C combined (LV=7).
77
Regression PlotEE (Predic) =7.11750 + 0.834250 EE (Chern)
S =1.48al9 R-Sq =93.1 % R--Sq(adj) =92.8 %
50
--uis~e:, 40
WW
30
30 40
EE (Chern)50
Figure 4.15. Pacific herring lot C linear regression plot oflipid (EE) content by chemicalanalysis versus EE predicted by SW-NIR (R2=O.93) using six latent variables.
Figure 4.16. Pacific herring lot D linear regression plot of ash content by chemicalanalysis versus ash predicted by SW-NIR (R2=O.91) using five latent variables.
78
CHAPTER 5. OVERALL SUMMARY AND CONCLUSIONS
5.1 PROXIMATE ANALYSIS
In this study, samples were oven dried. Errors between duplicate samples were
below five percent with this method, however for future studies, freeze drying should
also be used and compared to oven drying for accuracy of predictability.
In this study, higher errors were seen between duplicates ofash content. This
may have been due to the high fat content of herring. Because of this there is the
possibility of some splattering as samples are heated quickly during the ashing process,
thus there is a potential for sample loss. Future research should be done to test whether
fat really does boil and cause spattering during the ashing process.
5.2 MORPHOMETRIC MEASUREMENTS
When statistical analyses were run on morphometric measurements (length (cm),
volume (ml), weight (g), volume per unit length (ml/cm), and weight per unit length
(g/cm)) results, suggested that length is related to both volume and weight since volume
and weight were also highly positively correlated.
5.3 MORPHOMETRICS AND MAIN VARIABLES
Data for the main variables oflot, location, and sex were analyzed to determine if
nutrient content of herring varied in this study. Size was statistically analyzed to confirm
that sample size classes (small, medium and large), as assigned by visual appraisal,
significantly differed from one another. There was no significant effect oflot on
79
morphometric measurements. When the effects of location were looked at, similar results
were seen for both locations. When graphed, however, herring from Alaska were slightly
smaller in comparison to those from British Columbia.
The main variable of sex was also analyzed to assess differences in morphometric
measurements. When all of the data were analyzed, there was a very significant
difference between males and females without eggs for volume (ml), volume per unit
length (ml/cm) and weight per unit length (glcm). Because some of the fish within the
small size category that were characterized as "females" may not have been sexually
mature, the analyses were rerun with the small size category data removed. The results
were similar with the difference being between males and females without eggs in only
volume (ml) and volume per unit length (ml/cm). Morphometric measurements were
very similar for males and females with eggs. The analysis of these data was made
difficult because the reproductive stages ofthe Pacific herring are fairly complex.
Because some of the herring in this study were fairly small in size, they might not have
been accurately identified and/or assigned to the three sex categories correctly.
5.4 MORPHOMETRICS AND PROXIMATE CONTENT
In this study, no correlation was found between morphometric measurements and
the various nutrient concentrations of the homogenized whole Pacific herring. In contrast
to other studies which found significant correlations between standard length (cm) and
crude lipid content, this study showed a very low correlation (R2<0.01). Further research
should be conducted on herring from various locations in order to ensure that a
representative range ofboth nutrient content and size are included in the database. In
80
addition, morphometric measurements ofvolume (ml), weight (g), volume per unit length
(ml/cm) and weight per unit length (g/cm) should be recorded along with the common
measurement of standard length (cm). Statistical analyses should also be done to discern
whether or not a true correlation exists between the various morphometric measurements
and each nutrient concentration (moisture, fat, protein, ash, carbohydrates and gross
energy).
5.5 MAIN VARIABLES AND PROXIMATE CONTENT
Correlations were also determined between the main variables of lot, location and
sex and the concentration of the various nutrients (protein, lipid, ash, moisture,
carbohydrate (CHO) and gross energy (kcal/g)) in the herring. Both lot and location were
found to have an effect on nutrient content (P<0.001). Lot most likely had an effect on
the nutrient content since each lot was caught in a different season and year. Specific
years and seasons of catch were not used or compared because data were obtained from
random sampling, not programmed sampling from lots for specific harvest or catch dates,
or even catch method. Previous studies have found that both season and year have an
effect on the nutrient content of herring (Blaxter and Holiday, 1963; Paul and Paul, 1999;
Anthony et al., 2000; Castellini et aI., 2001). Studies have also found that herring from
different locations have dissimilar nutrient content (Anthony et al., 2000). Sex was found
to have no impact on the nutrient content ofthe Pacific herring in this study. Previous
studies have suggested a relationship between lipid and energy content to sex and
reproductive status (Henderson and Almatar, 1989; Paul and Brown, 1998). Pacific
herring store fat during the spring and summer to survive the winter's decline in prey
81
availability. In the winter these fish also allocate stored energy to gonadal products for
spring spawning (Hay et al., 1988). These relationships have also been found in Atlantic
herring (Clupea harengus). In this species, a depletion of lipid corresponds with the
development ofthe gonads in the winter season and the lowest levels were found in
March, after the fish have spawned (Anokhina, 1959). Lipid content has also been shown
to be directly related to fecundity (Anokhina, 1959). The results from the study herein
warrant further research based on what has been found in previous literature. Most
likely, a significant correlation was not seen in this study because each sex category
lacked sufficient sample numbers and ranges in size and maturity to get accurate
representation.
5.6 PROXIMATE CONTENT
Several studies have found correlations between different nutrients. One
relationship that has been cited is between moisture content and energy density. This
correlation, between moisture content and calculated GE, was also seen in this study
(R2=O.96). This correlation may exist because of the close relationship between lipid
content and gross energy, and the inverse relationship between moisture and lipid
content. Because lipid is so high in energy (9.8 kcallg) it makes up a very large portion
of a diet's gross energy, therefore, it would make sense that the two would be strongly
correlated. This high correlation (R2=O.97) between lipid and GE in this study
substantiates this very closer relationship.
Because of the extremely high correlation coefficient ofthese nutrient
relationships, the hypothesis that the linear regression formulas being created could be
82
used to calculate the entire nutrient content ofa Pacific herring was tested. Indeed, when
analyses were run, the prediction ability of these regression equations (n=99) was fairly
reliable. This reliability however, depends upon the R2 of the regression equation. The
lower the R2, the lower the predictability of the formula. With added proximate data on
herring, these linear regression equations could be improved upon to increase
predictability. Further research should be done to continue to study the possibility of
utilizing simple regression equations to predict the nutrient content of herring, and
potentially other fish species. The sample preparation required for this method would be
very minimal. Calculations would use the moisture content of a herring as a starting
point. This would be the easiest to measure, requiring the least amount ofequipment and
training. This method is by far the simplest, most cost effective and quickest method of
predicting the nutrient content ofa herring. This method would only require a fairly
accurate digital scale and a drying oven or microwave, and it would not require a trained
laboratory technician or special analytical equipment. Therefore, this method could be
implemented in most any situation, especially when timely decisions need to be made at
marine mammal facilities using large quantities ofherring. This method might also be
applicable for other species of fish that are commonly fed to marine mammals or other
aquatic animals. Further research should be conducted to validate these possibilities.
5.7 SW-NIR SPECTROSCOPY
From the results of this study, calibration models for lots Band C could be used
to predict to a fairly accurate degree, the proximate composition ofwhole Pacific herring.
The calibration model created for lot A however, was not as accurate, yielding R2 values
83
ranging from 0.24 through 0.74. This is most likely due to the variability in reference
results from this lot. A calibration model ofthis sort would not be applicable to the field
as a feasible method ofproximate analysis because a high degree of accuracy is required.
Because lots A through D differed by catch date, without a new calibration model
representing each new catch date, these models would not be practical. Given this,
calibration models were also created by location ofcatch. The calibration model for lot
D was also the model for herring caught off the southeastern coast ofAlaska. The model
created for lots A through C combined could be used to predict the nutrient content of
Pacific herring caught offthe coast ofBritish Columbia, Canada. These location based
calibrations models are more practicable. The model for the British Columbian herring
(R2=0.66-0.91) was fairly comparable to that of the Alaskan herring (R2=0.47-0.81).
This technology has widespread application and potential in the field ofmarine
mammal husbandry. The strengths of this technology lie in its ability to predict nutrient
content quickly and easily. This technology, once set up, is cost effective and non
destructive to samples. A portable NIR spectrophotometer with a hand-held fiber-optic
probe is now available. This expands the potential for this technology, making it easier
to analyze components of interest outside of the laboratory. One of the drawbacks of this
technology is the start-up cost ofequipment. Also, in order to scan a sample, a
calibration model must be created and installed. Finally, a trained laboratory technician
would be required.
A component of current NIR research focuses on minimizing sample preparation.
Increasingly, a hand-held fiber optic NIR probe is used to scan a section of an intact fish
to determine its nutrient composition. Usually, the portion of the fish scanned has been
84
found, in previous studies, to be representative of the nutrient content in the entire fish.
A fiber optic probe was not available for use in this study; therefore, the next simplest
method of preparation was used. Herring were homogenized using a store-bought Oster
brand blender. Because an NIR spectrophotometer is very sensitive to the particle size
within a sample, homogeneity must be carefully considered. In this study, methodology
used for sample preparation was created through trial and error. Problems of incomplete
homogenization were encountered with the skin, scales and small bones in the herring.
Precaution should be taken with the increasing size ofa fish due to thicker skin and larger
bones and vertebrae. These factors may increase the difficulty in homogenizing samples.
It has been suggested that a commercial meat grinder could be used on partially thawed
fish samples. Further testing on proper particle size is still needed to obtain optimal
reproducibility.
Further research should be done looking at increasing the predictability of these
calibration models. Additional nutrient data for samples would need to be added to the
calibrations created herein. More research is needed to look at the effects ofother
variables, such as season and prey availability, on the nutrient content ofPacific herring.
Once a better understanding of these variables is gained, a more comprehensive
calibration model could possibly be created.
This technology and the principle of nutrient analysis prediction could be applied
to other fish species used in marine mammal facilities. Research should be conducted to
look at the use of SW-NIR spectroscopy to predict the nutrient content of Atlantic herring
(Clupea harengus). That study should look at the possibility ofusing or adding to the
Pacific herring calibration to predict the proximate content ofAtlantic herring. These
85
nutrient calculation regression equations should also be evaluated for their applicability
in other industries such as the aquaculture and fish-farming industries.
86
Appendix Table 1. Average morphometric measurements, standard deviation, minimumand maximum values for lots and locations.
Max 64.08 51.85 10.88 31.30 80.26 7.22 16.05 22.58 2.46 3.65 71.93Shaded areas indicate proximate results on an as fed basis: unshaded areas indicate proximate results on a dry matter basis.
Appendix Table 4. Average proximate results, standard deviation, minimum and maximum values for size and sex cateQories.Size IProtein EE % Ash % CHO% Mol.t % OE Pratefn EE
wI eggs StDev 4.20 5.25 0.98 1.68 2.21 0.25~ 0.52 2.81n=12 Min 37.02 34.19 5.47 1.88 63.29 6.39 13.55 9.89
Appendix Table 5 Linear regression equations and R2 results for analyte correlations.
:;c
0.98
Analyte Regression equation R2
VS. CHO Ash =-7.32 + 0.04 (CHO) 0.01
VS. Moisture Ash =-11.51 + 0.27 (Moist) 0.62
VS. Gross energy (GE) Ash =25.55 - 2.73 (GE) 0,72
VS. Moisture CHO = -1.47 + 0.10 (Moist) 0
VS. Gross energy (GE) CHO =36.50 - 4.74 (GE) 0.19 bHO.
vs. Ash EE = 78.32 - 5.35 (Ash) 0.68
vs. CHO EE =41.42 - 0.64 (CHO) 0.09
VS. Moisture EE =173.29 - 1.94 (Moist) 0.74
vs. Gross energy (GE) EE = -98.11 • 20.64 (GE) 0.97
vs. Lipid (EE) Prot =75.93 - 0.70 (EE) 0.68
vs. Ash Prot =20.11 + 3.88 (Ash) 0.49
vs. CHO Prot =51.58 - 0.44 (CHO) 0.06
vs. Moisture Prot =-58.42 + 1.54 (Moist) 0.65
vs. Gross energy (GE) Prot =134.25 - 12.88 (GE) 0.52
Shaded areas indicate results on an as fed basis; unshaded areas indicate results on a dry matter basis.
Ash
CHO
Moisture VS. Gross energy (GE) Moist = 120.93 - 7.74 (GE) 0.69 'Moist = 90.23
Lipid (EE)
Protein
"0-
Appendix Table 6 Linear regression equations and R2 results for 8W-NIR predicted andcalculated (using linear regression equations to predict analyte content) values comparedto chemical reference results.
Analyte Comparison Regression equation
SW-NIR predicted vs. Pred =10.02 + 0.78 (Chern) 0.75Protein
Chemical reference
Calculated vs. ChemicalCalc =19.79 + 0.59 (Chern) 0.64
reference
SW-NIR predicted VS. Pred =5.93 + 0.85 (Chern) 0.86Chemical reference
Lipid (EE)Calculated VS. Chemicalreference
Calc =7.73 + 0.81 (Chern) 0.79
SW-NIR predicted vs. Pred =2.46 + 0.66 (Chern) 0.68Ash
Chemical reference
Calculated vs. ChemicalCalc =2.05 + 0.73 (Chern) 0.70