quantitative fatty acid signature analysis

25 Pages·2004·0.84 MB·English

EcologicalMonographs,74(2),2004,pp.211–235
(cid:1)2004bytheEcologicalSocietyofAmerica
QUANTITATIVE FATTY ACID SIGNATURE ANALYSIS: A NEW METHOD
OF ESTIMATING PREDATOR DIETS
SARA J. IVERSON,1,4CHRIS FIELD,2 W. DON BOWEN,3 AND WADE BLANCHARD2
1Departmentof Biology,DalhousieUniversity,Halifax,NovaScotia,Canada B3H4J1
2Departmentof Mathematicsand Statistics,DalhousieUniversity,Halifax,NovaScotia,Canada B3H3J5
3MarineFishDivision,BedfordInstituteof Oceanography,Departmentof Fisheriesand Oceans,Dartmouth,
Nova Scotia,Canada B2Y4A2
Abstract. Accurate estimates of the diets of predators are required in many areas of
ecology, but for many species current methods are imprecise, limited to thelast meal,and
often biased. The diversity of fatty acids and their patterns in organisms,coupledwiththe
narrow limitations on their biosynthesis, properties of digestion in monogastric animals,
andtheprevalenceoflargestoragereservoirsoflipidinmanypredators,ledustopropose
the use of quantitative fatty acid signature analysis (QFASA) to study predator diets. We
present a statistical model that provides quantitative estimates of the proportions of prey
species in the diets of individual predators using fatty acid signatures. We conducted sim-
ulation studies using a database of 28 prey species (n(cid:1) 954 individuals)fromtheScotian
ShelfoffeasternCanadatoinvestigatepropertiesofthemodelandtoevaluatethereliability
with which prey could be distinguished in the model. We then conducted experiments on
grey seals (Halichoerus grypus, n (cid:1) 25) and harp seals (Phoca groenlandica, n (cid:1) 5) to
assess quantitative characteristics of fatty acid deposition and to develop calibration co-
efficientsforindividualfattyacidstoaccountforpredatorlipidmetabolism.Wethentested
themodelandcalibrationcoefficientsbyestimatingthedietsofexperimentallyfedcaptive
greyseals(n(cid:1)6,switchedfromherringtoamackerel/capelindiet)andminkkits(Mustela
vison, n(cid:1) 46, switched frommilk to one ofthreeoil-supplementeddiets).Thedietsofall
experimentallyfedanimalsweregenerallywellestimatedusingQFASAandwereconsistent
with qualitative and quantitative expectations, provided that appropriate calibration coef-
ficients were used. In a final case, we compared video data of foraging by individualfree-
rangingharborseals(Phocavitulina,n(cid:1)23)fittedwithCrittercamsandQFASAestimates
of thedietofthosesamesealsusingacomplexecosystem-widepreydatabase.Amongthe
28 prey species in the database, QFASA estimated sandlance to be the dominant prey
species in the diet of all seals (averaging 62% of diet), followed primarily by flounders,
but also capelin and minor amounts of other species, although therewasalsoconsiderable
individualvariabilityamongseals.Theseestimateswereconsistentwithvideodatashowing
sandlance to be the predominant prey, followed by flatfish. We conclude that QFASA
provides estimates of diets forindividuals at time scalesthatarerelevanttotheecological
processes affecting survival, and can be used to study diet variability within individuals
over time, which will provide important opportunities rarely possible with other indirect
methods.WeproposethattheQFASAmodelwehavesetforthwillbeapplicabletoawide
range of predators and ecosystems.
Keywords: feedingecology;foodwebs;marinecarnivores;pinnipeds;predatordiets;predator–
preyrelationships;preyfattyacidcompositionand signatures;statisticalmodel.
INTRODUCTION direct observation of feeding can be used to estimate
diet. However, for many carnivores, including ceta-
The dynamics of predator–prey relationships, the
ceans, pinnipeds, mustelids, and ursids, as well as for
structure of food webs, and the foraging behavior of
nonbreeding seabirds, direct observation of feeding is
individualsarecentralthemesinecology(e.g.,Schoe-
rarely possible and indirect methods must be used to
ner1971,Paine1980,StephensandKrebs1986,Pimm
reconstruct the diet. These indirect methods are based
et al. 1991, Sih et al. 1998). Accurate estimates of
on the recovery of digestion-resistant prey structures
predatordietsarerequiredtounderstandtheseareasof
from feces, stomach contents, or from spewings such
ecology. For some carnivores (e.g., lions [Panthera
as owl pellets (Gaston and Noble 1985, Pierce and
leo]; wolves[Canis lupis];seaotters[Enhydralutris])
Boyle 1991). While there are some differences in the
way such methods are used across taxa, the principles
Manuscriptreceived9December2002;revised25July2003;
are the same (e.g., Carss and Parkinson 1996).
accepted27July2003;finalversionreceived17September2003.
Althoughmuchofourcurrentunderstandingofpred-
CorrespondingEditor:J.T.Wooten.
4E-mail:[email protected] atordietsisderivedfromthesemethods,suchestimates
211
212 SARA J. IVERSONET AL. EcologicalMonographs
Vol.74,No.2
can be biased (e.g., Jobling and Breiby 1986, Jobling dicated the presence of fish or other prey in the diets
1987, Carss and Parkinson 1996). Most soft-bodied of terrestrialandaquaticcarnivores(e.g.,Johnsonand
preyaredifficulttoidentifygiventheirrapiddigestion. West1973,RouvinenandKiiskinen1989,Wamberget
The diagnostic hard parts of some prey (e.g., shellsof al. 1992, Colby et al. 1993, Pond et al. 1995, Raclot
crustaceans,headsoflargefish)maynotbeconsumed etal.1998),thedegreetowhichplantshavebeencon-
bythepredatorormaybeerodedduringdigestion,such sumed by terrestrial carnivores (Iverson and Oftedal
thatthe sizeofpreyconsumedmaybeunderestimated 1992, Iverson et al. 2001b), and changes in the diets
or the identification of prey may not be possible. Fur- ofpinnipeds(Iverson1993,Iversonetal.1997a,Kirsch
thermore,thedegreeoferosionofhardpartsisspecies- et al. 2000).
specificandoftenafunctionofpreysizewithinspecies To date, fatty acid signatures have been used qual-
(Bowen 2000). Thus, differential rates of digestion itativelytoinfertrophiclevelsandspatialandtemporal
among prey species may seriously bias estimates in differences in diets both within and among species
favor of species with large and robust hard parts. Fi- (e.g., Kakela et al. 1993, R. J. Smith et al. 1996, S.
nally, these methods provide only a snapshot of the Smith et al. 1997, Iverson et al. 1997a,b). However,
most recent meal and may not berepresentativeofthe since the pattern of fatty acids found in some plants
longer term diet. and in many fish and invertebrates can be used to ac-
These limitations have led to the development of curately identify individual species (Iverson et al.
techniques that do not depend on the recovery of di- 1997b,2001b,2002,Budgeetal.2002),preyfattyacid
gestion-resistant hard parts (e.g., antisera to Atlantic signatures might provide quantitative estimates of
salmon,Salmosalar,withlimitedsuccess[Boyleetal. predator diets. To do thisrequires anunderstandingof
1990];stableisotoperatiosofcarbonandnitrogen[Rau thecharacteristicsofpreyfattyacidsignaturesandthe
etal.1992,Gannesetal.1997,Kelly1999]).Although extent to which they differ in a given ecosystem, an
stableisotoperatiosareusefulinestimatingthetrophic understanding of how ingested fatty acids are metab-
level of a predator, they usually cannot determine the olizedanddepositedinvarioustissuesofthepredator,
species composition of the diet (e.g., Hobson 1993, appropriatesamplingofpredatortissue,andastatistical
Gilmore et al. 1995, Koch et al. 1995). model that relates the predator signature to a mixture
A third method involves the use of fatty acid sig- ofpossiblepreysignatures.Herewepresentastatistical
natures (Iverson 1993). Fatty acids are the main con- model that provides quantitative estimates of the pro-
stituentofmostlipids,andunlikeothernutrients,such portionsofpreyspeciesinthedietsofindividualpred-
as proteins that are readily brokendownduringdiges- ators using fatty acid signatures. We use simulation
tion, fatty acids are released from ingested lipid mol- studies to investigate the properties of the model, and
ecules (e.g., triacylglycerols)during digestion,butare controlled feeding studies of grey seals (Halichoerus
not degraded. The fatty acids of carbon chain-length grypus)andharpseals(Phocagroenlandica)toassess
14 or greater pass into the circulation intact and are quantitativecharacteristicsoffattyaciddeposition.We
generally taken up by tissues the same way. Since a then test the model by estimating the diets of experi-
relativelylimitednumberoffattyacidscanbebiosyn- mentally fed captive grey seals and mink (Mustelavi-
thesized by animals (Cook 1991), it is possible todis- son), and the diets of individual free-ranging harbor
tinguish dietaryvs.nondietarycomponents.Oncetak- seals (Phoca vitulina) filmed during natural feeding
en up by tissues, fatty acids are eitherusedforenergy events. We used each of these systems to represent
orre-esterified,primarilytotriacylglycerols,andstored increasing complexity of diet estimation.
in adipose tissue. Although some metabolism of fatty
acids occurs within the predator, such that the com- METHODS
position of predator tissue will not exactly match that
The model
of their prey, fatty acids can be deposited in adipose
tissuewithlittlemodificationandinapredictableway. We refer to the quantitative distribution of all fatty
Fatty acids in marine organisms are extremely di- acidsmeasuredinapredatororpreysampleasitsfatty
verse and have high levels of long-chain, polyunsat- acidsignature.Toestimatethecompositionofthepred-
urated fatty acids that originate from various unicel- ator’sdietbasedonthesesignatures,wetakeaweight-
lularphytoplanktonandseaweeds(Ackman1980).Nu- ed mixture of the fatty acid signatures ofthepotential
merous studies have demonstrated that specific fatty prey types and choose the weighting that minimizes a
acidpatternsarepassedfrompreytopredatornearthe statisticaldistancefromthatofthepredator.Eachprey
bottomofthefoodweb(e.g.,Sargentetal.1988,Fraser type (typically species, but potentially subsets of spe-
et al. 1989, Graeve et al. 1994, Navarro et al. 1995, ciesorgroupingsofsimilarspecies;e.g.,Iversonetal.
St. John and Lund 1996, Kirsch et al. 1998) and that 2002) is summarized by its mean fatty acid signature,
the fatty acid composition of zooplankton directly in- and we estimate its proportional contribution to the
fluencesthefattyacidcompositionofblubberlipidsof predator’s diet.
baleen whales (e.g., Klem 1935, Ackman and Eaton Weproceedbyfirstdefininghowclosethepredicted
1966, Hooper et al. 1973). Fatty acids have also in- diet(i.e.,thequantitativemixtureofsignatures)isfrom
May2004 QUANTITATIVEFATTYACID SIGNATUREANALYSIS 213
the true diet. We then develop the concept of ‘‘cali- three distances, which give more weight to the differ-
brationcoefficients,’’whicharerequiredtoaccountfor ences in the rare fatty acids, are preferable; of these
predator lipid metabolism and the fact that the fatty three distances, the KL distance does so most conser-
acidsignatureofthepreywillnotbelaiddownexactly vatively and proportionately.
in the predator (i.e., for some fatty acids the values To then estimate the p, we carried out an optimi-
k
observed in the predator may be always higher, or al- zation over the number of prey types, k, with the p’s
k
ways lower, than that found in the diet; e.g., Kirschet constrained to be positive and sum to 1. The starting
al. 2000). Related to the concept of calibration, is valuesfortheoptimizationhavethep’sallequal.The
k
whether to estimate the diet using all fatty acids iden- optimization was carried out in S-Plus (S-Plus 2000)
tified or a subset that might better reflect diet. Lastly, using the function nlminb, which is a local minimizer
theestimatedsignaturecontributionfrompreymustbe for smooth nonlinear functions subject to bound-con-
correctedtoaccountfordifferencesinfatcontent(and strained parameters,andusesaquasi-Newtonmethod.
thusfattyacidcontribution)amongpreytypes.Allelse However, to efficiently conduct the simulations on
beingequal,specieswithahigherfatcontentwillcon- large, complex data sets, we used a FORTRAN opti-
tribute proportionately more to the predator signature mizer from Netlib.
than those with a lower fat content. However, given Standard errors of estimates.—A major source of
thatweknowthefatcontentofeachprey,itisstraight- variabilitycomesfromvariationinfattyacidsignatures
forward to translate the estimated signature contribu- amongindividualsofaparticularpreytype(e.g.,Iver-
tion to the proportion of each prey type eaten. son et al. 1997b, 2002, Budge et al. 2002). To capture
Model notation.—To set the basic model notation, thisvariability,wecarriedoutthefollowingbootstrap-
let y denote the proportion of the jth fatty acid of the pingprocedureinwhichwerepeatedlycreatenewprey
ij
ith predator. The i notation will be dropped when it is means by sampling with replacement from the prey
clearwearereferringtoasinglepredator.Letx denote database.
klj
theproportionofthejthfattyacidfromthelthpreyof For b (cid:1) 1, ..., B, steps 1 and 2 below are carried
thekthpreytype(inthiscasespecies)andn thenum- out:
k
ber of individual prey of type k. The mean x¯ is the
kj
meanofthepreyoftypekforfattyacidj.Theproblem 1)For each prey type k, randomly select n individ-
k
is to estimate (cid:2), the true proportion of the kth prey uals with replacement and create a new mean
k
type found in the predator’s diet with the estimatede- x¯*b.
k
noted by p. The estimated proportion of each prey in 2)Carry out the estimation procedure for the boot-
k
the diet, yˆ, over all fatty acids, is formed as follows: strap prey means and compute p*b. The estimate
(cid:1) k
yˆ (cid:1) p x¯ . of the standard(cid:2)e(cid:1)rror (SE) is computed as
k k
k [p*b (cid:3) mean(p*b)]2
k k
Distance measures and estimation of (cid:2)k.—The es- SE(pk) (cid:1) b B (cid:3) 1 .
timationproblemistochoosep suchthatyˆ is‘‘close’’
k
to y. Both y and yˆ sum to 1 and can be thought of as
distributions over the fatty acids. In this context, the Calibration coefficients.—Calibration coefficients,
Kulback-Liebler (KL) distance (Encyclopedia of Sta- cj,werecomputedasfollows:foraparticularfattyacid,
tistics 1983), defined(cid:1)as cj is computed as the 10% trimmed mean of the fol-
lowing rj’s:
KL (cid:1) (y (cid:3) yˆ )log(y /yˆ ) li
j j j j j rj (cid:1) seal /diet
li ij lj
is a natural choice, as it was developed to compare for all l and i. For example, to estimate the ‘‘greyseal’’
distributions.Thereareseveralotherpossibledistances calibrationcoefficients,wehadeightsealsand30herring.
including the more usual squared error (SQ) distance, Since we could not analyze the actual herring that indi-
(cid:4)j(yj(cid:3) yˆj)2, thesquared relative error(REL),(cid:4)((yj(cid:3) vidual seals ate, i (1 to 8) indexes the seals and j (1 to
yˆj)/yj)2andthesquarederrordistanceofthelogs(LSQ), 30) indexes the herring. This gives 240 calibration co-
(cid:4)j (log(yj) (cid:3) log(yˆj))2. To understand the relative be- efficientsforeachfattyacid,forwhichthe10%trimmed
havior of these distances, we considered an absolute mean is then computed. These coefficients are then in-
difference of 0.01 between the true (y) and predicted cluded in the distance measures by replacing the preda-
(yˆ) proportion for a common, an intermediate, and a tor’s observed proportion of fatty acid of type j by
rarer fatty acid, respectively (i.e., true proportions:
(cid:1)y /c
0.20,0.05,and0.01;predictedproportions:0.21,0.06, z (cid:1) j j .
and 0.02, respectively). TheSQdistanceattributesthe j y /c
s s
same weight for all true values. However, an absolute s
error of 0.01 should be more serious in the rare as Although we used the trimmed mean across all indi-
opposed to the common fatty acid. Hence, the other viduals in modeling, we also estimated the 10%
214 SARA J. IVERSONET AL. EcologicalMonographs
Vol.74,No.2
trimmed mean within each individual to estimate a Conversion fromproportionsinfattyacidsignature
within-study SE for coefficients. to those in diet.—Given the estimated proportions of
Fatty acid subsets.—We refer to fatty acids by the each prey type in the predator’s fatty acid signature,
standard nomenclature of carbon chain length:number the p’s, and theaverage fatcontentofeachpreytype,
k
of double bonds, and the location (n-x) of the double the f’s, one can then express the proportion of the
k
bondnearesttheterminalmethylgroup.Inanalysesof actual diet derived from the kth prey type, denoted by
marinelipids,over70fattyacidscanbeidentifiedand a, as follows:
k
quantified, depending on the analytical methods and
(cid:1)p /f
gas chromatograph (GC) column used (Fig. 1). How- a (cid:1) k k .
k p /f
ever,notallfattyacidsprovideequalinformationabout k k
k
diet due to predator metabolism (Iverson 1993). For
instance, if short- or medium-chain fatty acids (i.e.,
(cid:5)14carbons;alsoincludingiso5:0insomecetaceans) The data
are found in predator adipose tissue, these could arise Thedatausedinthepresentstudyrepresenthundreds
only from biosynthesis, since any of these consumed of samples analyzed and 67 fatty acids identified per
in the diet would be immediately oxidized (Jackson sample,andcannotbepresentedindetail.Thus,where
1974). In contrast, fatty acids with n-6 or n-3 double possible we show representative examples.
bonds or components such as 22:1n-11generallyarise Prey fatty acid signatures.—Simulation studies of
onlyfromdiet;however,22:1n-11mayexhibitreduced theestimationmodelwerebasedonapreydatabaseof
deposition(BremerandNorum1982).Otherfattyacids 954fattyacidsignatures(e.g.,Fig.1)of28marinefish
arise from a combination ofdietandbiosynthesis.For andinvertebratespeciescollectedontheScotianShelf
instance, although both arefoundinprey,inpredators off eastern Canada (from Budge et al. 2002).
14:1n-5 is produced predominantly frombiosynthesis, Calibrationcoefficients.—Todeterminetheextentto
whilesome22:5n-3arisesfrommodification(Ackman whichspecificfattyacidsundergoselectivedeposition
et al. 1988, Iverson 1993, Iverson et al. 1995). Fatty or metabolism, we conducted three feeding experi-
acids such as 16:0, 16:1n-7, 18:0 and 18:1n-9, may ments. The aim of these experiments was to develop
arise to some extent frombiosynthesisinthepredator, calibrationcoefficientstoweightindividualfattyacids
but are also highly indicativeofdifferencesinvarious according to how directly they were deposited from
prey (e.g., Fig. 1; Iverson 1993, Iverson et al.2001b). diet. The first two studies used eight juvenile (2–3 yr
Thus, for both of these latter types of fatty acids (i.e., old) grey seals (‘‘grey calibration’’) and five juvenile
those that always occur at predictably higher or lower harp seals (‘‘harp calibration’’), which were housed
levels in the predator than in prey due to some bio- temporarilyinlargeindoorseawatertanksatDalhousie
synthesis or some reduced deposition, respectively), University’s Aquatron facilities. The grey seals were
calibration coefficients can be used to reduce the in- maintainedforatleastfivemonthsonadietconsisting
fluence of systematic deviations on diet estimation. solely of Atlantic herring (Clupea harengus, 6.2 (cid:6)
Finally,somefattyacidsfoundatlowortracelevels 0.30% fat). The harp seals were maintained for up to
may not be correctly identified and separated from fivemonthsonthesameherring,buttheseanimalshad
abundant nearby peaks (e.g., 18:1n-11 from 18:1n-9; been in captivity for less time than the grey seals. All
Fig. 1) depending upon the nature of the chromato- herringfedduringthefive-monthperiodhadbeencol-
graphic equipment used. Therefore their detection in lected from a single lot and, although variable in fat
chromatograms can be problematic or inconsistent. and fatty acid composition, were considered to be the
Since most such fatty acids occur at low levels incar- most uniform diet we could feed. At the end of the
nivore tissue, these can beremoved fromfurtheranal- five-monthperiod,afull-depth((cid:7)5cm)blubberbiopsy
ysis if necessary. was taken from the pelvic region of each seal using a
In the present study, we did not use the fatty acids sterile biopsy punch according to Kirsch et al.(2000).
that could only be present in the predator primarily Theblubberbiopsywasplacedinaglassvialcontain-
from biosynthesis, nor any fatty acids that were in- ingchloroformwith0.01%BHTandstoredfrozenuntil
consistentlyidentified(AppendixA).Oftheremaining analysis. Thirty herring were randomly collected
fattyacids,weusedtwosubsetsformodeling:(1)‘‘di- throughout the feeding period and kept frozen until
etary,’’ which includes only those 33 fatty acids that analysis ((cid:5)six months). In these two studies, we used
could arise from dietary origin, and (2) ‘‘extended- the initial assumption that in the approximate five-
dietary’’(41fattyacids),whichincludesall‘‘dietary’’ month period, the fatty acid composition of blubber
fatty acids as well as eight fatty acids that could be would resemble that of the seal’s diet as much as it
biosynthesizedbypredators,butwhoselevelsinapred- ever would.
atorarealsoinfluencedbyconsumptionofspecificprey In the third calibration study, we examined the de-
(Appendix A). The subsets of fatty acids used were greetowhichblubberfattyacidcompositionresembled
renormalized to sum to 1 (after application of calibra- the diet after a period of complete and rapid fattening
tion coefficients if used) prior to modeling. on a high-fat diet. Grey seal pups are born with neg-
May2004 QUANTITATIVEFATTYACID SIGNATUREANALYSIS 215
FIG. 1. Fatty acid chromatogram of one individual of each of two prey species, (a) pollock (Pollachius virens)and (b)
sandlance(Ammodytesdubius),fromtheScotianShelf,illustratingrelativedifferencesbetweenspecies.Here67fattyacids
are identified and quantified in each chromatogram; however, only selected peaks are labeled on this plot. Fatty acids are
eluted(‘‘retentiontime’’)inorderofcarbonchainlength,numberofdoublebonds,andpositionofdoublebondsonapolar
capillary column (see Methods). The integrated area under each peak represents the relativemasspercentageofeachcom-
ponent.
216 SARA J. IVERSONET AL. EcologicalMonographs
Vol.74,No.2
ligible blubber, but at weaning (about 16 days post- experimentaldiet,sealswereagainweighedandablub-
partum [dpp]) they have deposited (cid:7)24 kg of fat in ber biopsy was taken as described above; on day 20,
blubber from a milk-only diet, which is in turn pro- bodycompositionwasagainmeasured.Throughoutthe
ducedcompletelyfromtheblubberstoresofthefasting experiment, individual herring (n (cid:1) 15), mackerel (n
mother(Iversonetal.1993).Thus,virtuallyallblubber (cid:1) 25), and capelin (n (cid:1) 25) were randomly collected
fatty acids in suckling pups arise from milk intake, andstoredfrozeninairtightcontainersforanalysis((cid:5)6
permittingaccurateestimationofcalibrationfactorsfor months).
individual fatty acids from a completely homogenous In the second study, we used fatty acid data from
diet. Full-depth blubber biopsieswerecollectedasde-
fattening mink kits as an example of a terrestrial car-
scribed above from 17 grey seal pups at 15 dpp (i.e.,
nivore(Layton1998).Briefly,until21dpp,17lactating
immediately prior to weaning) on Sable Island, Nova
femaleswerefedprimarilyawetdiet(6.6%fat)along
Scotia, Canada (43(cid:8)55(cid:9) N, 60(cid:8)00(cid:9) W). Milk samples
with some pelleted diet (17.3% fat), while kits con-
(40–60% fat; Iverson et al. 1993) werecollectedfrom
sumed solely their mother’s milk. Both the wet and
each of these pups’ mothers (n (cid:1) 17) at 0, 5, 10, and
pelleteddietsconsistedofprimarilypoultryoffal(Lay-
15 dpp, and the average milk fatty acid signature for
ton1998).Priortofeedingtheexperimentaldietsat21
each mother (i.e., here used as the ‘‘prey’’) was com-
dpp, perirenal adipose tissue was sampled from 10
pared with that of her single pup (‘‘pup calibration’’).
mink kits, euthanized in the course of other studies.
Allsampleswerestoredfrozeninglassvialscontaining
The remainder of kits and their mothers werethenfed
chloroform with 0.01% BHT until analysis.
one of three experimental wet diets. Each diet (6.6%
Experimentaldietstudies.—Weinvestigatedtheper-
fat) was composed of primarily poultry offal and fish
formance of the model using data from two captive
feedingexperiments(Kirsch1997,Layton1998).Both meal,supplementedwitheitherpoultryfat,aquaculture
ofthesestudiesweredesignedtoevaluatetheeffectof herring oil, or seal oil (purchased from commercial
a known change in diet on the fatty acid signature of sources) as 70% of the dietary fat source. Perirenal
a predator. In one study, a second group of juvenile adipose tissue was sampled from six kits on each of
grey seals (n (cid:1) 6, age 1–3 yr), housed temporarily in the three wet diets at both 28 and 42 dpp (i.e., n(cid:1) 36
a seawater tank at the Aquatron facilities, had previ- total).Sincedietswerecompletelyhomogenous,asin-
ously been maintained on a diet of Atlantic herring gle sample of each was analyzed in duplicate for fat
(averaging 5.1 (cid:6) 0.46% fat, from various lots) for up content and fatty acid composition. We were not able
to five months. At the start of the diet trial, each seal to obtain milk samples from the mothers. All samples
was weighed, body composition was measured using were stored as described above.
isotopedilution(OftedalandIverson1987,Bowenand Free-rangingharborsealsfilmedduringforaging.—
Iverson1998)andafull-depthblubberbiopsywastak- In a final case, we studied 23 free-ranging adult male
en and stored as described above. Seals were then fed harbor seals during the breeding season of May–June
an experimental diet, consisting of Atlantic mackerel 1997 on Sable Island. Throughout this period, males
(Scomberscombrus)andcapelin(Mallotusvillosus)for makeroutineforagingtripsontheScotianShelfinthe
a period of 20 days. Atlantic herring, mackerel, and
vicinityofSableandreliablyreturntotheislandevery
capelin share some similaritiesinfattyacidsignatures
few days (Walker and Bowen 1993, Coltman et al.
(e.g., Budge et al. 2002), thus allowing evaluation of
1997).Eachmalewasfittedwithananimal-bornevideo
model performance when species in the diet do not
system (‘‘Crittercam,’’ [National Geographic Televi-
differmarkedlyfromoneanother.Duetothelargesize sion, Washington, D.C., USA] Marshall 1998) for (cid:1)3
of the mackerel (averaging 38.1 cm, 0.5 kg), we re-
d. The camera was positioned such that the animal’s
movedtheheadsandcuttheremainderofeachinto5-
headwasvisibleinthecamera’sfieldofviewandpro-
cmthickcross-sections(i.e.,includingtheviscera)for
grammedtofilm10-minsegmentsevery45minduring
feeding. Seals were fed to satiation (or until they lost
daylight, thus permitting the prey species that were
interest) twice daily; however, due to the constraints
eaten to be recorded (Bowen et al. 2002). At each de-
of this captive situation, it was not possible to deter-
ployment/recapture, a full-depth blubber biopsy was
mine individual intakes. As a result, some individuals
takenanddietswereestimatedusingthemodelandthe
undoubtedly consumed more and also different pro-
Scotian Shelf prey database. Since adultmalesremain
portions of the prey species than others. Capelin (av-
eraging1.8(cid:6)0.23%fat)wasofferedonlyinthemorn- in the vicinity of Sable for several months prior to
ings and mackerel (averaging 18.3 (cid:6) 0.56% fat) only reproduction,weassumedthatpreyeatenduringthese
in the afternoons, in an attempt to get seals to eat the short-term studies would reflect the somewhat longer
less-preferredcapelin.Theapproximatedailyrationof- termdietinferredthroughblubberfattyacids.Thus,in
fered averaged 5.4 kg·d(cid:3)1·seal(cid:3)1, comprising about onesensethiswasavalidationexperimentunderfree-
three parts capelin to one part mackerel. At this daily rangingconditions,whichemployedamuchmorecom-
ration, approximate fat intake would be 0.32 plexpreybasethanwouldhavebeenpossibleincaptive
kg·d(cid:3)1·seal(cid:3)1 (Kirsch 1997). On days 12 and 20 of the experiments.
May2004 QUANTITATIVEFATTYACID SIGNATUREANALYSIS 217
FIG.2. Hierarchicalclusteranalysisonthemeanfattyacidsignatures(extendeddietarysubset)of28preyspecies(n(cid:1)
954 individuals) from the Scotian Shelf(Budge etal. 2002).Scientificnamesofallspeciesnotpreviouslydescribedinthe
textareasfollows(inalphabeticorderofteleosts,crustaceans):argentine(Argentinasilus),butterfish(Peprilustriacanthus),
gaspereau (Alosa pseudoharengus), halibut (Hippoglossus hippoglossus), ocean pout (Macrozoarces americanus), red hake
(Urophycischuss),redfish(Sebastessp.),sculpin(Myoxocephalusoctodecemspinosus),searaven(Hemitripterusamericanus),
smooth skate (Raja senta), thorny skate (Raja radiata), white hake (Urophycistenuis),winterskate(Rajaocellata),lobster
(Homarus americanus), red crab (Geryon quinquedens), rock crab (Cancer irroratus), shrimp (Pandalus borealis). The
Kulback-Liebler(KL)distancemeasurewasusedtodeterminehowsimilaranytwotaxawerewithrespecttotheirfattyacid
signatures.The averagelinkagemethodwas used, whichtends to identifysphericalclusters.
Laboratory analyses methods: known standard mixtures (Nu Check Prep.,
Elysian, Minnesota, USA), silver-nitrate (argentation)
Lipid was quantitatively extracted from all samples
chromatography, and GC-mass spectrometry (Hewlett
(Folch et al. 1957, Iverson et al. 2001a). Each whole
Packard6890GC,1:20splitinjection,MicromassAu-
prey was individually ground and homogenized prior
tospec oa-TOF mass spectrometer, operated at 1000
to extraction. Milk and blubber samples werealsoho-
resolution,scanningmasses120to450[HewlettPack-
mogenizedpriortoextraction.Fattyacidmethylesters
ard, Palo Alto, California, USA]). Fatty acid identifi-
werepreparedusing1.5mLof8%borontrifluoridein
cations on all chromatograms were checked, and cor-
methanol (Iverson et al. 1997b); this method in our
rected and reintegrated as necessary. Fatty acids are
laboratory produces identical resultstothatusingHil-
expressed as mass percent of total fatty acids.
ditch reagent (0.25 mol/L H SO in methanol). Dupli-
2 4
cateanalysesoffattyacidcompositionwereperformed
Simulation studies
on all samples using temperature-programmed gas
chromatographyasdescribedpreviously(Iversonetal. Simulation with no calibration coefficients.—To in-
1992, 1997b, Budge et al. 2002), on a Perkin Elmer vestigate the properties of the estimation procedures
Autosystem II Capillary FID (Perkin Elmer, Boston, andtherobustnessofthemodelindeterminingagiven
Massachusetts, USA) gas chromatograph (GC) fitted diet, we performed a numberofsimulationstudiesus-
with a 30 m (cid:10) 0.25 mm ID column coated with 50% ing the Scotian Shelf prey database. The first simula-
cyanopropyl polysiloxane (0.25 (cid:11)m film thickness; tions were performed without calibration coefficients
J&W DB-23; Folsom, California, USA) and linked to to assess the ability to estimate true diet based solely
acomputerizedintegrationsystem(Turbochrom4soft- ondifferentiatingandquantifyingpreyspeciesbytheir
ware, PE Nelson, San Jose, California, USA). Fatty fattyacidsignatures.Weusedhierarchicalclusteranal-
acids and isomers were identified using the following ysistodeterminetherelativesimilarityofpreyspecies’
218 SARA J. IVERSONET AL. EcologicalMonographs
Vol.74,No.2
TABLE 1. Speciescompositionof dietsconstructedforsimulationstudies.
Nonzeroelementsof the compositionvector,(cid:2) (proportionof diet)
Silver Winter Yellowtail
Diet Cod Haddock Pollock hake Plaice flounder flounder Sandlance
1 0.333 0.333 0.167 0.167
2 0.200 0.800
3 0.200 0.800
4 0.100 0.100 0.100 0.100 0.100 0.500
Notes: Prey species used were based on 954 fatty acid signatures of 28 marine fish and
invertebrate species collected on the Scotian Shelf off eastern Canada (Budge et al. 2002).
Samplesizesoftheabovepreyspecieswereasfollows:cod(Gadusmorhua;n(cid:1)84),haddock
(Melanogrammus aeglefinus; n (cid:1) 54), pollock (Pollachius virens;n(cid:1) 25), silverhake (Mer-
luccius bilinearis; n (cid:1) 38), plaice (Hippoglossoides platessoides; n (cid:1) 99), winter flounder
(Pseudopleuronectesamericanus; n(cid:1) 25),yellowtailflounder(Limandafurruginea;n(cid:1)92),
and sandlance(Ammodytesdubius;n (cid:1) 71).
signatures(Fig.2).Wethenconstructedfourdiets(Ta- sets, as these should be applicable to the other exper-
ble 1): Diets 1–3 eachcontainedtwoorfourpreyspe- imentalsealdietstudiesandtothefree-rangingharbor
cies that were more similar to one another than to all sealsandarosefromthelongerofthetwosealfeeding
other species in the fatty acid database. These three trials. The procedures for these simulations are de-
diets represented difficult or, in some sense, ‘‘worst scribedinAppendixB.WeusedthesumoftheRMSEs
case’’ estimation scenarios. Diet 4 contained six spe- of predicted diet from true diet (i.e., Table 1) of each
cies, some of which again were similar in fatty acid pseudo-seal for the 1000 simulation runs and for the
composition,andwasconstructedtorepresentthediet two fatty acid subsets to evaluate performance. These
of a free-ranging grey seal based on results of fecal RMSEswerethencomparedtotheRMSEsofpredicted
analysis (Bowen and Harrison 1994). dietfromtruedietofthesamepseudo(grey)sealusing
Simulationswereusedtoevaluatehowtheaccuracy nocalibrationcoefficients,andusingharpandpupcal-
ofourestimateswasaffectedbyfivefactors:diet(four ibration coefficients in the fitting process.
diets),fattyacidsubset(dietaryandextended-dietary),
distance measure (KL, LSQ, SQ, REL), amount of RESULTS
‘‘noise’’ in the simulated seal (0, 10%, 20%), and the
Calibration coefficients
number of individual prey (n (cid:1) 30, 60, or 90) used in
constructing the ‘‘pseudo-seal’’ fatty acid signature. Despite large differences in fat content and homo-
Noisewasmeanttorepresenttheproportionofthediet geneityofthedietfed,intheknowndietaryhistoryof
madeupofincidentalconsumptionofpreyspeciesthat the animals, and in the degree to which they fattened
were not included in the assumed diet. The pseudo- duringthestudy,overalltherewasareasonabledegree
seal fatty acid signature was constructed by sampling ofcorrespondencebetweenthethreesetsofcalibration
the prey database in the proportions specified by our coefficients and low within-study variability (Fig. 3).
simulated diet, with additional random prey added in Calibrationcoefficientsformostfattyacidswereclose
tocreatethenoise.Detailsofthesimulationprocedures to one, particularly in the case of suckling pups;how-
are provided in Appendix B. ever,therewerenotableexceptions.Ingeneral,theco-
We calculated the relative mean squared error efficients for the grey and harp seals fed herring were
(RMSE) to measure how well simulations estimated more similar to one another and deviated more from
theassumeddiet.TheRMSEwasconstructedbysum- 1.0 than did the pup coefficients, but the pattern of
ming the relative squared deviations of the true diet deviations(Fig.3)wassimilarinallthreestudies,sug-
from the estimated diet, ([true (cid:3) estimate]/true)2, for gesting that the underlying metabolic processes were
each simulation run and then averaging over the 1000 commonamonganimalsanddiets.Thefattyacidswith
simulation runs for each factor setting. the10highestand10lowestcalibrationcoefficientsin
Simulation with calibration coefficients.—To esti- both grey and harp seals, were also mostly among the
mate the diet of a real predator, the effect of predator highest and lowest in pups, although again the mag-
lipid metabolism on the deposition of dietaryfattyac- nitude of deviation from 1.0 was smallerinpups(Fig.
ids must be included. Therefore, we also performed 3, Appendix A). Fatty acids such as 14:1n-5,
simulations using the three sets of calibration coeffi- 16:1n-11, 16:1n-9, 17:1 and 18:1n-11, with generally
cients to examine how model estimates of diets were highcoefficients,arepredominantlybiosynthesizedby
affected by the use of calibration coefficients and to the predator and/or occur at low levels (generally oc-
test whether all sets of coefficients produced similar curring at (cid:5)0.8% of total fatty acids in seals and/or
results. We used the grey seal calibration coefficients prey).Becausesmallerrorsinminorortracefattyacids
as the standard with which to compare the other two withlargecalibrationcoefficientsmighthavelargeef-
May2004 QUANTITATIVEFATTYACID SIGNATUREANALYSIS 219
FIG.3. Calibrationcoefficients(mean(cid:6)1SE)ofthe10%trimmedmeanscalculatedwithineachindividual(notethatinmost
casesthestandarderroristoosmalltosee)estimatedforall67fattyacidsquantified,usingthreedifferentfeedingstudies:juvenile
grey and harp seals maintained for five months on a diet of herring (6.2 (cid:6) 0.30% fat), and suckling grey sealpups at weaning
havingconsumedonlytheirmothers’milk(40–60%fat)andinwhichvirtuallyallblubberfattyacidshavearisenfrommilkintake.
The1:1lineispresented,whichdenotesthedeviationofagivenfattyacidinapredatorfromthatconsumedinitsdiet.Stars((cid:2))
indicate examples offattyacidswithlargedeviationsfrom1:1butwhichusuallyoccuratminoramounts((cid:5)0.5%)insealsand
theirprey.Arrowsindicatecommonfattyacidsthatwouldbeexpectedtohaveadditionalcontributionfrombiosynthesisinpredators,
especiallyifonlowerfatdiets.SeeAppendixAforfattyacidsusedinmodelingsets.
fects on estimates from the model, we removed these measure,amountof‘‘noise’’inthesimulatedseal,and
fatty acids from modeling subsets at the outset (see prey sample size in minimizing the RMSE of the es-
AppendixA).Relativelyhighcoefficientsofotherfatty timated diet. Variation in RMSEduetosamplesizeof
acids,suchas16:1n-7and18:1n-9or22:5n-3,arealso individual prey (30, 60 or 90) was obtained by aver-
consistentwiththeexpectedcontributionfrombiosyn- aging over all the other factors. The RMSE decreased
thesis or metabolic modification, respectively, in the with increasing sample size by (cid:7)20% and 5% for the
predator. However, these major fatty acids are good extended-dietary and dietary fatty acid subsets, re-
indicatorsofpreyspecies(e.g.,Fig.1),andcalibration spectively, indicating that asamplesizeof30individ-
coefficientsprovideameansofusingtheminthemod- ualpreywouldproviderepresentativeresults.Variation
el. intheaverageRMSEduetothelevelof‘‘noise’’used
In all three studies, some of the lowest calibration (0%,10%,or20%)didnotexceed10%.Thus,toassess
coefficients were found for 20:0 (except in pups), theeffectoftheotherthreefactorsontheperformance
22:1n-11,22:1n-9,22:1n-7,and24:1(Fig.3).Ofthese, of the estimation model, we used a sample size of 30
20:0 and 24:1 are either rareandnotindicativeofdiet and 10% noise in the other simulations.
or inconsistently detectable (Appendix A) and thus We next considered the effects of the distance mea-
were eliminated from use in the model at the outset. sure, fatty acid subset, and the complexity of the sim-
In contrast, 22:1n-11, 22:1n-9 and 22:1n-7 are impor- ulated diet on model performance. Significant effects
tantdietaryindicators(e.g.,Fig.1;Iverson1993,Iver- were found for fatty acid subset, and diet, with a dis-
sonetal.1997b).Again,fortheseandmostotherfatty tancemeasurebydietinteraction(P(cid:5)0.05,three-way
acidswithdeviationsfrom1.0,calibrationcoefficients ANOVAonthemediansacrossthe1000simulations),
allow their use in the model. butnotfordistancemeasurealone.Forthedietaryfatty
acid subset, SQ tended to perform somewhat worse
Simulations with no calibration coefficients
than the other distance measures, whereas for the ex-
Our aim here was to determine the relative impor- tended-dietary fatty acid subset, the KL distance gen-
tance of diet complexity, fatty acid subset, distance erallyperformedbest.Overall,theRMSEswerelowest
220 SARA J. IVERSONET AL. EcologicalMonographs
Vol.74,No.2
FIG. 4. Results of the simulation study for Diet 1 as defined in Table 1 with 10% error (noise) added, using the28 Scotian
Shelfpreyspecies(n(cid:1)954),theextended-dietaryfattyacidsubset,andnocalibrationcoefficients,andwith30individualprey
usedinconstructingthepseudo-seal.Speciesarelistedinalphabeticorder(teleosts,crustaceans).Inplots,‘‘a’’denotesthevalue
(proportion)specifiedforeachofthefourpreyspecieschosenforthediet.Thesimulationwasrun1000times,andestimateddiet
resultsarerepresentedinboxplots,asthemedian(middlehorizontalbar),the25thpercentile(lowerbar),andthe75thpercentile
(topbar)ofthedatadistribution(i.e.,theboxcontains50%ofthedata).Dotsrepresentoutliersdefinedasbeinganyvaluegreater
(orless)than1.5timestheinterquartilerange(75thpercentile–25thpercentile)abovethe75th(orbelowthe25th)percentile.
for the extended-dietary subset and KL distance, and overestimate of the other prey category. In contrast,
highest for SQ. On the basis of these results, we con- using the extended-dietary subset, estimates of indi-
cluded that any of the three distances, KL, REL, or vidual species within each diet were generally closer
LSQ, would generally give reasonable results. How- to the true values, but the other prey category still
ever, we have chosen to use the KL distance as this is tended to be somewhat overestimated. When simula-
anaturaldistancebetweentwodistributions,andarises tions of the same four diets were performed with no
in a number of statistical settings including the boot- noiseincluded,inallcasescomponentsofthedietwere
strap (DiCiccio and Romano 1989). more accurately predicted and a lower proportion of
Next we examined how well the model estimated the diet was attributed to other prey.
each component of the simulated diets. As the noise Patterns of values across these simulations provide
was set at 10% for these simulations, accurate esti- insight into how the model performed within each diet
mation would give a total of 10% other prey. Hence (Fig. 5). For Diet 1, while the best fits corresponded
for Diet 1, we should estimate 30% each of cod and closely to the specified diet, as the fit worsened the es-
haddockand15%eachofpollockandsilverhake,ob- timates became low for cod and high for pollock, sug-
tained by multiplying Diet 1 levels in Table 1 by 0.9. gesting that these two species may be difficult to dis-
Usingtheextended-dietaryfattyacidsubset,themodel tinguish. We also underestimated silver hake as the es-
estimated the true diet rather well (Fig. 4), with the timates deteriorated. In Diet 2, theestimatesofpollock
major species in the diet distinguished from others in decreased as RMSE increased, with the balance going
the prey database. Nevertheless, there was some mis- either to cod or other prey. In Diet 3, the estimates of
identification(7%)ofthedietcompositiontootherprey haddock and silver hake decreased as the fit worsened,
typesabovetheaddednoise.Theresultsofsimulations with the proportion attributed to other prey becoming
for all four diets and both fatty acid subsets are sum- verylargeintheworstfits.Incontrasttotheotherdiets,
marizedinTable2.Usingthedietaryfattyacidsubset, estimates did not change notably for Diet 4 as the fit
although some species in each diet were reasonably worsened,exceptthatyellowtailflounderbecamesome-
estimated, others were not, resulting in a consistent what overestimated. In summary, especially for diets