This article appeared in a journal published by Elsevier. The attached

copy is furnished to the author for internal non-commercial research

and education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling or

licensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of the

article (e.g. in Word or Tex form) to their personal website or

institutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies are

encouraged to visit:

http://www.elsevier.com/copyright

Author's personal copy

Multi-centre testing and validation of current protocols for the identificationof Gyrodactylus salaris (Monogenea)

A.P. Shinn a,*, C. Collins b, A. García-Vásquez a, M. Snow b, I. Matejusová b, G. Paladini a, M. Longshaw c,T. Lindenstrøm d, D.M. Stone c, J.F. Turnbull a, S.M. Picon-Camacho a, C. Vázquez Rivera a, R.A. Duguid a,T.A. Mo e, H. Hansen e, K. Olstad f, J. Cable g, P.D. Harris f, R. Kerr c, D. Graham h, S.J. Monaghan a, G.H. Yoon a,K. Buchmann i, N.G.H. Taylor c, T.A. Bakke f, R. Raynard b, S. Irving c, J.E. Bron a

a Institute of Aquaculture, University of Stirling, Stirling FK9 4LA, UKbMarine Scotland – Science, Marine Laboratory, 375 Victoria Road, Aberdeen AB11 9DB, UKcCefas Weymouth Laboratory, Barrack Road, Weymouth DT4 8UB, UKdAdjuvant Research, Dept. of Infectious Disease Immunology, Division of Vaccine, Statens Serum Institut, 5 Artillerivej, 81/306, 2300 Copenhagen S, DenmarkeNational Veterinary Institute, Section for Parasitology, P.O. Box 750 Sentrum, NO-0106 Oslo, NorwayfNatural History Museum, Dept. of Zoology, University of Oslo, P.O. Box 1172, NO-0318 Oslo, Norwayg School of Biosciences, Cardiff University, Cardiff CF10 3AX, UKhDisease Surveillance & Investigation Dept., Veterinary Sciences Division, Agri-food & Biosciences Institute, Stoney Road, Stormont, Belfast BT4 3SD, UKiUniversity of Copenhagen, Faculty of Life Sciences, Dept. of Veterinary Pathobiology, Section of Fish Diseases, Stigbøjlen 7, DK-1870 Frederiksberg C, Denmark

a r t i c l e i n f o

Article history:

Received 21 January 2010

Received in revised form 29 April 2010

Accepted 30 April 2010

Keywords:

Gyrodactylus salaris

Contingency planning

Pathogen introduction

Validation

Identification

Monogenea

Protocol

a b s t r a c t

Despite routine screening requirements for the notifiable fish pathogen Gyrodactylus salaris, no standard

operating procedure exists for its rapid identification and discrimination from other species of Gyrodacty-

lus. This study assessed screening and identification efficiencies under real-world conditions for the most

commonly employed identification methodologies: visual, morphometric and molecular analyses.

Obtained data were used to design a best-practice processing and decision-making protocol allowing

rapid specimen throughput and maximal classification accuracy. True specimen identities were estab-

lished using a consensus from all three identification methods, coupled with the use of host and location

information. The most experienced salmonid gyrodactylid expert correctly identified 95.1% of G. salaris

specimens. Statistical methods of classification identified 66.7% of the G. salaris, demonstrating the need

for much wider training. Molecular techniques (internal transcribed spacer region-restriction fragment

length polymorphism (ITS-RFLP)/cytochrome c oxidase I (COI) sequencing) conducted in the diagnostic

laboratory most experienced in the analysis of gyrodactylid material, identified 100% of the true G. salaris

specimens. Taking into account causes of potential specimen loss, the probabilities of a specimen being

accurately identified were 95%, 87% and 92% for visual, morphometric and molecular techniques, respec-

tively, and the probabilities of correctly identifying a specimen of G. salaris by each method were 81%, 58%

and 92%. Inter-analyst agreement for 189 gyrodactylids assessed by all three methods using Fleiss’ Kappa

suggested substantial agreement in identification between the methods. During routine surveillance

periods when low numbers of specimens are analysed, we recommend that specimens be analysed using

the ITS-RFLP approach followed by sequencing of specimens with a ‘‘G. salaris-like” (i.e. G. salaris, Gyro-

dactylus thymalli) banding pattern. During periods of suspected outbreaks, where a high volume of spec-

imens is expected, we recommended that specimens be identified using visual identification, as the

fastest processing method, to select ‘‘G. salaris-like” specimens, which are subsequently identified by

molecular-based techniques.

� 2010 Published by Elsevier Ltd. on behalf of Australian Society for Parasitology Inc.

1. Introduction

Identification of parasite pathogens to species level can be a

complex and time consuming task. In the face of a potential disease

outbreak, the need for identification may exceed the capacity to

deal with the number of specimens entering the identification

pipeline. Recent evidence demonstrates that the translocation of

fish across national borders has increased the rate of introduction

of exotic pathogens into indigenous fish stocks with serious eco-

nomic consequences. Of 14 fish metazoan parasites recently re-

ported to have been introduced into the United Kingdom (UK),

0020-7519/$36.00 � 2010 Published by Elsevier Ltd. on behalf of Australian Society for Parasitology Inc.

doi:10.1016/j.ijpara.2010.04.016

* Corresponding author. Tel.: +44 1786 473171; fax: +44 1786 472133.

E-mail address: aps1@stir.ac.uk (A.P. Shinn).

International Journal for Parasitology 40 (2010) 1455–1467

Contents lists available at ScienceDirect

International Journal for Parasitology

journal homepage: www.elsevier .com/locate / i jpara

Author's personal copy

10 are already fully established, and this pattern looks set to con-

tinue (Gibson, 1993; Kennedy, 1993; Yeomans et al., 1997). Some

of these introduced parasites are known to be serious pathogens,

which may have wide-ranging repercussions for conservation

and fisheries management as well as for aquaculture (e.g. Bothrio-

cephalus acheilognathi Yamaguti, 1934; see Williams, C.F., 2007.

Impact assessment of non-native parasites in freshwater fisheries

in England and Wales. Ph.D. thesis, University of Stirling, UK).

Of particular concern to fisheries is the ectoparasite Gyrodacty-

lus salaris Malmberg, 1957 which can be highly pathogenic to

Atlantic salmon, Salmo salar Linnaeus, 1758. This monogenean

has caused a catastrophic decline of salmon stocks in Norway, dec-

imating stocks in 46 rivers (see Table 2 of Bakke et al. (2007)) and

leading to near extermination in five of these rivers (Mo, 1994).

Gyrodactylid surveys conducted in the 1990s (Platten et al.,

1994; Shinn et al., 1995) and on-going government-based surveil-

lance programmes indicate that the UK is currently free of G. salar-

is. Nevertheless, there is considerable concern about the accidental

introduction of this species, particularly since experimental expo-

sure of native British salmon stocks to G. salaris in Norway demon-

strated their susceptibility (Bakke and MacKenzie, 1993;

MacKenzie and Bakke, 1994). Furthermore, G. salaris is now re-

corded from 13 neighbouring European countries, the most recent

records originating from Poland (Rokicka et al., 2007) and Italy

(Paladini et al., 2009b).

Comprehensive screening for G. salaris as a part of national

monitoring programmes would generate huge numbers of sam-

ples, particularly in the event of a suspected outbreak. In order

for monitoring to be effective, screening must rapidly identify

specimens to the species level. However, G. salaris is notoriously

difficult to discriminate from closely related and morphologically

similar species present on European salmonids. If government pol-

icy seeks to maintain high standards of fish health and welfare in

the UK, it is vital to have validated standard operating procedures

(SOPs) for the efficient processing of specimens, while maintaining

the highest possible likelihood of correctly identifying G. salaris.

Several approaches have been used in the identification of G. salar-

is, besides the classical method of morphological examination of its

haptor (attachment organ) under light microscopy. Morphometric

analyses, which rely upon a range of statistical classifiers (Kay

et al., 1999) and specific molecular techniques (Cunningham

et al., 1995; Cunningham, 1997; Meinilä et al., 2002) have been

developed to discriminate this pathogenic species from benign

species also associated with salmonid hosts.

Morphometric discrimination of Gyrodactylus spp. can be diffi-

cult, due to the small size of taxonomically important structures

(i.e. haptoral attachment hooks) and the importance of relatively

small variations in diagnostic characters. Difficulty of identification

is compounded by the extreme plasticity of these hooks and the

potential for closely related species to hybridise (Bakke et al.,

2007). Current estimates of identification performance have been

made under artificially simplified conditions (Kay et al., 1999;

McHugh et al., 2000; Shinn et al., 2000), with estimates based on

controlled, small sets of specimens constructed from a limited

number of populations analysed under ideal conditions. Standard-

ised test sets used for determining classification efficiency nor-

mally preclude those specimens lost in preparation, difficult to

measure, of poor quality for molecular analysis, or discarded be-

cause their identity was not confirmed. In addition, studies carried

out to estimate classification efficiency for different techniques are

often conducted in isolation using different individual specimens

such that no direct comparisons can be made between methodol-

ogies. Under conditions of a suspected outbreak, numerous sam-

ples of Gyrodactylus requiring specific identification might be

sent to a designated expert laboratory with a requirement for rapid

and accurate identification.

The purpose of this study, therefore, was to conduct a double-

blind trial to assess the classification performance of each of these

morphometric and molecular methods, both singly and in combi-

nation, against a panel of gyrodactylid experts (i.e. visual identifi-

cation) as techniques for the rapid identification of G. salaris and

its accurate discrimination from other species of Gyrodactylus

found on British salmonids, under conditions of a simulated out-

break. An improved understanding of the performance of these

techniques under real-world conditions and of the parameters that

affect their performance allows improvement of identification pro-

tocols and technologies and assists the development of robust

SOPs. Such data can improve the monitoring and control of serious

pathogens by British and Irish Fish Health Inspectorates and helps

policy formulation, implementation and adherence.

2. Materials and methods

2.1. Origin of material

A total of 28 UK sites with salmonid populations were sampled

for Gyrodactylus during routine electrofishing surveys and fish farm

visits during February to May 2007. The following hosts were sam-

pled: Esox lucius Linnaeus, 1758, Oncorhynchus mykiss (Walbaum,

1792), S. salar, Salmo trutta fario Linnaeus, 1758 and Thymallus thy-

mallus (Linnaeus, 1758). Additional material included Gyrodactylus

specimens collected from six salmonid populations in mainland

Europe and from gyrodactylid populations maintained in research

aquaria (Table 1). Where possible, entire fish (n = 10 per site;

approximately 5 g in body weight) were taken and fixed in 96% re-

search grade ethanol. For larger fish (i.e. >10 g in body weight),

only the fins were removed and preserved. For each site, a random

selection of gyrodactylids (usually 10–20) were picked off either

the body or fins using triangular, mounted surgical needles and

were placed in individually-labelled 1.5 ml Eppendorf tubes con-

taining 96% ethanol. The time taken to screen each fish, and there-

fore the time taken to collect all of the specimens for the study, was

recorded; the time taken to screen the small number of detached

fin-only samples (Table 1), was not included in the time calcula-

tions. For each individual host fish, all Gyrodactylus present were

removed and placed into two tubes, one for fins and one for those

parasitising the body. In total, 620 gyrodactylids were harvested.

Anonymised recoded tubes were then passed to another research-

er. Variable numbers of Gyrodactylus were randomly taken from

every tube. The randomly selected individuals were then prepared,

as described in the following section, to provide 443 gyrodactylids

for submission to the subsequent identification methodologies.

2.2. Sample preparation

Individual worms had their posterior attachment organ (i.e.

haptor) excised using a scalpel under a dissecting microscope. Fol-

lowing excision, the body was transferred to a new 1.5 ml Eppen-

dorf tube containing 96% ethanol while the haptor was subjected

to proteolytic digestion to remove the tegument and musculature

enclosing the haptoral armature. For the digestion step, the pro-

teinase K-based method (Paladini et al., 2009a) (i.e. 100 lg/ml pro-

teinase K (Cat. No. 4031-1, Clontech UK Ltd., Basingstoke, UK),

75 mM Tris–HCl, pH 8, 10 mM EDTA, 5% SDS) was used and the

digestion of each gyrodactylid haptor was carefully monitored at

3�magnification on an Olympus SZ40 dissecting microscope. Once

the tissues enclosing the haptoral hooks had been removed, diges-

tion was arrested by the addition of 3 ll 50:50 formaldehyde:glyc-

erine solution. A coverslip was added to the preparations (n = 443;

239 Gyrodactylus derjavinoides Malmberg, Collins, Cunningham &

Jalali, 2007, 20 Gyrodactylus lucii Kulakovskaya, 1952, 41 G. salaris,

1456 A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467

Author's personal copy

four Gyrodactylus teuchis Lautraite, Blanc, Thiery, Daniel & Vig-

neulle, 1999, 31 Gyrodactylus thymalli Zitnan, 1960 and 108 Gyro-

dactylus truttae Gläser, 1974) and the edges were sealed with nail

varnish to make a permanent mount. All of the Eppendorf tubes

containing the body of a gyrodactylid (confirmed by the visual

inspection of each tube at 4� magnification using an Olympus

SZ40 dissecting microscope) and their corresponding digested hap-

tors prepared on glass slides were passed to a third researcher for

recoding to allow blind analysis. The parasite bodies in Eppendorf

tubes were then divided into two sets and sent blind to two inde-

pendent laboratories to be identified using molecular methods

(internal transcribed spacer region-restriction fragment length

polymorphism (ITS-RFLP), ITS or cytochrome c oxidase I (COI)

sequencing). The structured splitting of samples to two or more

laboratories also provides protection against catastrophic protocol

failures which may lose data for entire sample collections.

2.3. Visual identification of gyrodactylids from digital images

Following proteolytic digestion, images of the hooks of each

specimen were captured using an AxioCam MRC (Zeiss) 1.5 mega-

pixel camera fitted with a MicroCam Olympus LB Neoplan D-V C

mount 0.75� interfacing lens. These were attached to an Olympus

BX51 compound microscope and specimens were viewed under a

100� oil immersion objective using MRGrab v. 1.0.0.0.4 (Carl Zeiss

Vision GmbH, München, Germany) software. Where required, mul-

tiple images of a specimen, at different focal depths, were taken to

provide a comprehensive presentation of the hooks. The digital

images of the hooks (n = 1371) were then sent out to six gyrodacty-

lid experts, three who were nominally salmonid Gyrodactylus ex-

perts (SGEs) and three who were non-salmonid Gyrodactylus

experts (NSGEs). Experts were asked to identify each specimen to

species level or, where this was not possible, they were requested

Table 1

Specimen details relating to the 443 specimens of Gyrodactylus that were collected and used within this study. For sites, the number of infected hosts from

which Gyrodactylus specimens were removed is shown in parentheses after the site name. For hosts with mixed infections, the number of each species on

either the body (B) or the fins (F) is indicated under the species present heading.

Site (hosts infected) Longitude/Latitude Host Microhabitat

(No. of specimens)

Species present

Scotland

River Almond (2) 56�240520 0N; 3�300450 0W Salmo salar Body (6); Fins (1) Gd

North Esk River (8) 56�440500 0N; 2�270170 0W S. salar Body (8); Fins (11) Gd

River Ordie (1) 56�29070 0N; 3�300370 0W Salmo trutta Body (5); Fins (5) 5Gd (B); 5Gtr (F)

River Tweed site 1 (4) 55�320430 0N; 2�590320 0W S. salar Body (7); Fins (8) Gd

(2) 55�320430 0N; 2�590320 0W S. trutta Body (3); Fins (7) 3Gtr (B); 2Gd/5Gtr (F)

site 2 (1) 55�28040 0N; 2�54060 0W S. trutta Body (5); Fins (3) 1Gd/4Gtr (B); 1Gd/9Gtr (F)

Shockie Burn (1) 56�27010 0N; 3�300560 0W S. trutta Body (10); Fins (10) Gtr

FSR1 Borders (a) Confidential location Oncorhynchus mykiss Fins (9) Gd

FSR2 Strathclyde (a) Confidential location O. mykiss Fins (18) Gd

FRS3A Strathclyde (a) Confidential location S. salar Fins (6) Gd

FRS3B Strathclyde (a) Confidential location S. salar Fins (6) Gd

FRS4 Highlands (a) Confidential location S. salar Fins (10) Gd

FRS5 Highlands (a) Confidential location S. salar Fins (20) Gd

Northern Ireland

Ballinderry River (1) 54�380250 0N; 6�460300 0W S. salar Fins (3) 2Gd/1Gtr

(1) S. trutta Body (10); Fins (10) 1Gd/9Gtr (B); 10Gtr (F)

Broughderg River (1) 54�410500 0N; 7�020000 0W S. salar Body (10); Fins (9) Gd

(2) S. trutta Body (11); Fins (4) Gd

Colebrooke River (4) 54�220400 0N; 7�160000 0W S. salar Body (15); Fins (2) Gd

(1) S. trutta Body (10); Fins (10) 10Gd (B); 9Gd/1Gtr (F)

Enler River (1) 54�200300 0N; 5�470100 0W S. trutta Body (9); Fins (5) Gtr

Glasswater (1) 54�240550 0N; 5�460300 0W S. trutta Body (5); Fins (5) 5Gd (B); 3Gd/2Gtr (F)

Glenariff River (1) 55�010150 0N; 6�060400 0W S. trutta Body (5); Fins (5) 2Gd/3Gtr (B); 1Gd/4Gtr (F)

River Bann (3) 54�200300 0N; 6�130300 0W S. salar Fins (2) Gd

(1) S. trutta Body (9); Fins (11) 9Gd (B); 8Gd/3Gtr (F)

River Lagan (1) 54�230300 0N; 6�030200 0W S. trutta Body (5); Fins (5) 2Gd/3Gtr (B); 1Gd/4Gtr (F)

Six Mile Water (2) 54�460250 0N; 5�570000 0W S. trutta Body (6); Fins (5) Gtr

England

River Aire (1) 53�430370 0N; 1�7090 0W Esox lucius Fins (10) Gl

River Frome (2) 50�400420 0N; 2�100370 0W S. salar Fins (6) Gd

River Itchen (1) 51�50320 0N; 1�130490 0W Thymallus thymallus Fins (11) Gth

River Nidd (a) 54�30540 0N; 1�420240 0W S. trutta Fins (5) 3Gd/2Gtr

(1) T. thymallus Fins (3) Gth

River Ouse (1) 53�530190 0N; 1�50300 0W E. lucius Fins (10) Gl

River Test (4) 51�110480 0N; 1�220280 0W T. thymallus Fins (10) Gth

River Wylye (2) 51�70120 0N; 1�53030 0W T. thymallus Fins (5) Gth

Continental Europe

Laerdalselva, Norway (a) 61�60110 0N; 7�280120 0E S. salar Fins (15) Gs

Mosbjerg, Denmark (a) 57�300100 0N; 10�16050 0E O. mykiss Fins (15) Gs

Refsgaard nr Vejle, N. Jutland, Denmark (a) 55�360570 0N; 9�430370 0E O. mykiss Fins (10) Gd

Jyvaskyla, Finland (a) Confidential farm location O. mykiss Fins (10) Gs

River Nera, Italy (a) 42�530N; 13�020E O. mykiss Mucus scrape (5) 1Gd/4Gte

River Sérchio, Italy (a) 44�040N; 10�200E O. mykiss Mucus scrape (2) Gs

B, body; Gd, Gyrodactylus derjavinoides; F, fins; Gl, Gyrodactylus lucii; Gs, Gyrodactylus salaris; Gte, Gyrodactylus teuchis; Gth, Gyrodactylus thymalli; Gtr,

Gyrodactylus truttae.a The number of infected hosts is unknown as the Gyrodactylus specimens were taken from either a mucus scrape or from fins removed from multiple

hosts.

A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467 1457

Author's personal copy

to specify whether the specimen was ‘‘G. salaris-like” or ‘‘not G.

salaris-like”. As with the specimens sent for molecular analysis,

no supporting information regarding host, likely gyrodactylid spe-

cies or origin was provided. Neither were they supplied with any

guidance regarding what constituted ‘‘G. salaris-like” or ‘‘not G.

salaris-like”.

2.4. Morphometric analysis and statistical classification

Morphometric measurements of the haptoral hooks and bars

were taken from the images of each gyrodactylid specimen by

uploading images into the dedicated Point-R software (ver. 1.0 �

University of Stirling, 2003) running within the KS300 v3.0 image

analysis environment (Carl Zeiss Vision GmbH, München, Ger-

many). Normal laboratory practice would be to work on the origi-

nal slide preparations to allow specimens to be analysed at

numerous focal depths, but for the purposes of this study to allow

for a direct comparison between the visual identification approach

and the statistical classification approach, measurements were

made on the same digital images that were passed to the visual ex-

perts. A total of 25 point-to-point measurements were made on

each specimen (11 from the hamuli, six from the ventral bar and

eight from the marginal hooks) which follow those described in

Shinn et al. (2004). Once measured, five different linear discrimi-

nant analysis-based classifiers were used to provide the likely

identity of each specimen. The five classifiers that were used were

constructed on data sets derived from an earlier study of gyro-

dactylids parasitising salmonids (Shinn et al., 2000). The data used

for the original training of the classifiers contained some but not all

of the species considered in this study. The species G. teuchis and G.

lucii, for example, were not included in the data sets used to train

the classifiers constructed by Shinn et al. (2000). For the purposes

of the original classifier training, the identity of each specimen

used for the training exercise was based upon the visual examina-

tion of the haptoral hooks of each specimen by a panel of experts

and their consensus agreement on its identity. Each statistical clas-

sifier, as part of its output and in addition to providing an identity

for each specimen, also provided a posterior probability value for

each specimen to indicate a level of confidence with regard to

the identification.

The five classifiers that were constructed for the study of Shinn

et al. (2000) were used for the classification of species in this study

and were based on the following combination of hook measure-

ments. The classifier ‘‘All” used all 25 point-to-point measure-

ments; classifier ‘‘H” used only the 11 measurements made on

the hamuli; classifier ‘‘M” used only the eight measurements de-

rived from the marginal hooks; classifier ‘‘HM” used all the mea-

surements made on the hamuli and marginal hooks; and, the

classifier ‘‘8” used a combination of the eight best measurements

suggested by a forward stepwise discriminant analysis applied to

all of the haptoral hook measurement variables taken from the

specimens used for training the classifiers. From a forward step-

wise discriminant analysis, the total length of the marginal hook,

the hamulus point length, the marginal hook sickle distal width,

the ventral bar process length, the marginal hook sickle length,

the hamulus aperture length, the total length of the ventral bar

and the total width of the ventral bar were regarded as the best

eight measurements. The classifiers were run using the statistical

package S-PLUS 4 (Data Analysis Products Division, Mathsoft, Seat-

tle, USA; Venables and Ripley, 1997). Classifier decisions depended

upon whether the measurements fell within the classifier’s previ-

ous experience for a species and the posterior probability of the

specimen belonging to a particular species class.

Only 217 specimens were assessed using the statistical classifi-

cation methodology to permit a direct comparison with the identi-

fication of all the specimens determined by the molecular

approaches. The full complement of 25 measurements, however,

was not possible on 18 (8.3%) specimens, either because the spec-

imens were damaged or poorly mounted, and these were rejected

and regarded as ‘‘unmeasurable” in the summary statistics.

2.5. Molecular analysis

The parasite bodies in Eppendorf tubes were sent blind to two

independent laboratories to be identified using molecular ap-

proaches (ITS-RFLP, ITS or COI sequencing). The 222 samples sent

to one of the laboratories, however, were lost during DNA

extraction.

2.6. DNA extraction

The body of each gyrodactylid was removed from the 96% eth-

anol in which it was stored and transferred to a 0.5-ml tube con-

taining 7.5 ll lysis buffer (proteinase K 20 lg/ml, 0.45% Tween

20, 0.45% Igepal CA630 in TE buffer (10 mM Tris, 1 mM EDTA, pH

8.0)). Tubes were incubated at 55 �C for 1 h and then for 10 min

at 95 �C to inactivate the proteinase K. Lysates were stored at

�20 �C prior to further analysis.

2.7. ITS rRNA PCR

The entire region spanning the ITS1, ITS2 and the 5.8S gene was

amplified using primers ITS1 (50-TTT CCG TAG GTG AAC CT-30) and

ITS2 (50-TCC TCC GCT TAG TGA TA-30) as described in Cunningham

(1997). Each amplification reaction contained 1.8 ll Gyrodactylus

lysate, 1� NH4 PCR buffer (Bioline), 1.5 mM MgCl2, 250 lM dNTPs,

1.4 lM of each primer and 5 U BioTaq polymerase (Bioline) in a to-

tal volume of 20 ll. The cycling conditions were 1 cycle of 5 min at

94 �C, followed by 35 cycles of 1 min at 94 �C, 1 min at 51 �C and

2 min at 72 �C, followed by an extended elongation at 72 �C for

5 min. Positive PCR control (diluted plasmid containing ITS PCR

product from G. derjavinoides) as target and a negative control were

run alongside the experimental samples. To visualise the products,

3 ll of each PCR product was run on a 1.5% Tris–acetate–EDTA

(TAE) (0.04 M Tris, 1.14% v/v, glacial acetic acid, 1 mM EDTA) buf-

fered ethidium bromide-stained agarose gel under UV illumina-

tion. The remaining products were stored at 4 �C prior to further

analysis.

2.8. ITS rDNA RFLP analysis

RFLP analysis of the Gyrodactylus ITS rDNA PCR product was car-

ried out as described by Cunningham (1997) with minor modifica-

tions. Five microlitres of aliquots of PCR product were digested

with restriction enzyme HaeIII (Invitrogen). For each digestion

reaction the following was used: 1� React 2 buffer (Invitrogen),

2 U HaeIII, 5 ll PCR product in a total volume of 10 ll. Positive con-

trol RFLP reactions were prepared as above using ITS PCR products

generated previously from known G. derjavinoides, G. salaris, G. teu-

chis and G. truttae DNA. These were run alongside experimental

samples and 100 bp size markers (Invitrogen). Restriction patterns

were analysed following electrophoresis of 10 ll of digested PCR

product on a 2% TAE ethidium bromide-stained agarose gel.

2.9. ITS sequencing

The ITS PCR products were excised from the gels and purified

using MinElute gel purification kits (Qiagen Ltd., Crawley, UK)

according to the manufacturer’s instructions. Concentrations of

the purified ITS PCR products were estimated following agarose

gel electrophoresis alongside mass markers (Low DNA Mass

Ladder, Invitrogen). Sequencing was carried out using the same

1458 A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467

Author's personal copy

primers as used for ITS rDNA PCR amplification (10 lM primer per

reaction) and the GenomeLab DTCS–Quick Start Kit (Beckman

Coulter) according to the manufacturer’s protocol. Reactions con-

tained approximately 60 ng of purified PCR product (MinElute, Qia-

gen). Forward and reverse sequencing reactions were performed.

Sequencing reaction products were resolved on a Beckman Coulter

CEQ 8800 DNA Analysis system (Beckman Coulter). Sequences

were examined using Sequencher software (ver. 4.5) (Intelligenet-

ics/Gene Codes Corporation).

2.10. COI PCR amplification and sequencing

Partial sequence of the mitochondrial COI gene was obtained

from gyrodactylids presenting a ‘‘G. salaris-like” pattern upon RFLP

analysis. COI PCR amplification was carried out as detailed in Mei-

nilä et al. (2002) and Hansen et al. (2003) with minor modifica-

tions. One microlitre of Gyrodactylus lysate was added to a

reaction mix containing the following: 1� NH4 PCR buffer (Bioline),

2.0 mM MgCl2, 200 lM dNTPs, 0.5 lM of each primer LA (50-TAA

TCG GCG GGT TCG GTA A-30) and HA (50-GAA CCA TGT ATC GTG

TAG CA-30), 1.5 U BioTaq (Bioline) enzyme and DNAse/RNAse free

water (Sigma) to a total reaction volume of 20 ll. Where the prim-

ers LA and HA failed to amplify a product, they were replaced in the

above reaction by the primers LB (50-TAA TTG GTG GGT TTG GTA A-

30) and HB (50-AGC TAC CAC GAA CCA TGT AT-30). Thermocycling

was carried out using the following conditions: an initial denatur-

ation at 95 �C for 5 min followed by 35 cycles of 95 �C for 1 min,

50 �C for 1 min and 72 �C for 1.5 min, and an extension step at

72 �C for 7 min. COI PCR products were then visualised on 1.5%

TAE ethidium bromide-stained agarose gels under UV illumination.

The PCR products were excised from the gels and purified using

MinElute gel purification kits according to the manufacturer’s

instructions. Concentrations of the purified PCR products were

estimated following agarose gel electrophoresis alongside mass

ladders (Low DNAMass Ladder, Invitrogen). Purified products were

sequenced directly using approximately 60–80 ng of product and

the same primers as were used in COI PCR amplification, employ-

ing the sequencing method described above for ITS sequencing.

2.11. Phylogenetic analysis of COI sequences

Available COI sequences from G. salaris and G. thymalli isolates

were retrieved from GenBank and aligned using CLUSTALW (Hig-

gins et al., 1994) with sequences obtained from gyrodactylids dur-

ing this study. The alignment was manually edited using the

BioEdit program (Hall, 1999) and identical sequences removed.

The final alignment consisted of 742 nucleotides of which 127

were parsimony-informative, 270 were variable and 472 were con-

servative. The alignment file was used in phylogenetic analysis fol-

lowing the approach used in Hansen et al. (2003) using the

software MEGA 4.0 (Tamura et al., 2007). Phylogenetic trees were

inferred by Neighbour-Joining (NJ) and Maximum Parsimony

(MP) (unweighted), using Kimura’s 2-parameter method to calcu-

late genetic distances (Kimura, 1980). Bootstrap values were ob-

tained using 1000 replicates. Gyrodactylus mariannae Winger,

Hansen, Bachmann & Bakke, 2008 from alpine bullhead, Cottus poe-

cilopus Heckel, 1837 (AY258375) and G. lavareti Malmberg, 1957

(AY225306) from common whitefish Coregonus lavaretus

(Linnaeus, 1758) were used as outgroups.

2.12. Task duration

The identification component of the study was limited to

6 weeks (from the time the prepared specimens were passed to

the relevant researcher for identification). To establish the relative

speeds and efficiencies of each step for the different methods, the

time taken to complete each task was recorded. The morphometric

and statistical classification data from each researcher was as-

sessed to determine whether they were more likely to make errors

in the early or late stages of the measurement exercise or the time

spent in a single measuring session (�2 h).

2.13. Statistical analysis

A Fleiss’ Kappa test (Fleiss, 1971) was used to assess inter-ana-

lyst agreement between the three methods examined (visual

assessment, morphometric and molecular analyses). Data for the

visual assessment were based on consensus agreement between

the three G. salaris experts (93.8–98.8% agreement), while the data

for the morphometric and molecular analyses were based on the

results generated from a single researcher in each laboratory.

3. Results

3.1. Specimen preparation

From the samples presented in Table 1, mixed infections of G.

derjavinoides and G. truttae were commonly found on S. trutta fario

at the UK sites and a mixed infection of G. derjavinoides and G. teu-

chis was found in the mucus scrapes taken from captive O. mykiss

in the River Nera, Italy. A total of 88 whole fish were screened dur-

ing the study, 47 of which were infected. These fish took a total of

21.8 h (1310 min) to screen and remove the gyrodactylids (mean

14.9 min/fish; 2.1 min/gyrodactylid). From 47 Gyrodactylus-posi-

tive fish processed in this study, 89.4% of fish had gyrodactylids

on the fins, 66% had gyrodactylids on the body and 55.3% of fish

had parasites on their fins and body. Sampling whole fish as op-

posed to only the fins provided a 10.6% increase in the detection

of infected hosts. The subsequent processing steps of removing

and digesting the haptors of 443 specimens, making permanent

preparations, photographing them and then transferring the bodies

to individual tubes for molecular identification took 70.8 h (mean

9.5 min/gyrodactylid).

From the 443 specimens processed, 10 specimens (2.26%) were

lost during the initial preparation stage (i.e. their removal from the

Eppendorf tubes in which they were stored). A further 11 haptors

(2.48%) were lost as they were separated from their respective

bodies and prepared for proteolytic digestion, so 433 gyrodactylid

bodies were used for molecular analysis. The ‘‘true” identity of the

specimens was established by consensus opinion from all the iden-

tification approaches coupled with the use of host/location infor-

mation. This approach gave a unanimous identity for each

specimen with no specimens of questionable identity. Using the

189 specimens which were identified by all three techniques, for

example, gave agreements of 99.47% (visual identification and

molecular) and 83.07% (statistical classification) with the consen-

sus identity. Host data can be useful in prioritising material from

different hosts for processing and for assisting in the identification

of some species, notably G. salaris and G. thymalli, and occasionally,

the discrimination of some specimens of G. derjavinoides from G.

truttae, when using the visual method.

3.2. Visual identification

A total of 1371 images of the 422 digested gyrodactylid haptors

(�3.2 photos/specimen), were sent to six gyrodactylids experts for

identification. The results of three experts who routinely identify

species of Gyrodactylus from salmonids and pike are shown in

Table 2A, the results from the three experts who routinely identify

gyrodactylids from other hosts are presented in Table 2B and

the combined results are presented in Table 2C. Classification

A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467 1459

Au

tho

r's p

ers

on

al c

op

y

Table 2

The visual identification of 422 Gyrodactylus specimens parasitising European salmonids and pike. Identification was based on the examination of images of proteolytic digested haptors only and in the absence of host/locality

information. The average percentage of classifications were calculated using (A) three gyrodactylid experts who routinely identify species of Gyrodactylus that parasitise salmonids and pike; (B) three gyrodactylid experts who routinely

identify gyrodactylids but not necessarily those that parasitise salmonids and pike; and, (C) the combined data of all six experts. Correct identification of the Gyrodactylus specimens is given as a percentage of the total number of

specimens representing that species included in the study; for example, 82.11% of the Gyrodactylus salaris specimens were correctly identified by the three experts as G. salaris, 9.76% of the G. salaris specimens were misclassified as

Gyrodactylus thymalli, 0.81% misclassified as Gyrodactylus teuchis, 6.50% misclassified as either Gyrodactylus derjavinoides or Gyrodactylus truttae, 0.81% of the specimens could not be specifically allocated to a species but were recorded as

being ‘‘G. salaris-like”. The final column ‘‘All G. sal-like” provides a summary – 93.50% of the G. salaris specimens were classified to a species resembling G. salaris, i.e. either G. salaris, G. thymalli, Gyrodactylus lucii, G. teuchis or the ‘‘G.

salaris-like” group. The figures shown in bold highlight the percentage of specimens correctly allocated to their species class while the figures in parentheses represent the range of classification efficiencies across the experts.

True class Predicted class

Gyrodactylus salaris Gyrodactylus thymalli Gyrodactylus lucii Gyrodactylus teuchis Gyrodactylus

derjavinoides/truttae

‘‘G. salaris-like” All ‘‘G. salaris-like”

(A) The average percentage of classifications using three experts who routinely identify gyrodactylids collected from salmonids and pike

G. salaris (n = 41) 82.11 (70.73–95.12) 9.76 (7.32–14.63) 0.00 (0.00–0.00) 0.81 (0.00–2.44) 6.50 (0.00–19.51) 0.81 (0.00–2.44) 93.50 (80.49–100.00)

G. thymalli (n = 29) 4.60 (0.00–10.34) 91.95 (79.31–100.0) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 3.45 (0.00–10.34) 0.00 (0.00–0.00) 96.55 (89.66–100.00)

G. lucii (n = 18) 0.00 (0.00–0.00) 1.85 (0.00–5.56) 87.04 (72.22–100.0) 0.00 (0.00–0.00) 11.11 (0.00–27.78) 0.00 (0.00–0.00) 88.88 (72.22–100.00)

G. teuchis (n = 4) 0.00 (0.00–0.00) 8.33 (0.00–25.00) 0.00 (0.00–0.00) 58.33 (0.00–100.0) 33.33 (0.00–75.00) 0.00 (0.00–0.00) 66.67 (25.00–100.00)

G. derj./trut. (n = 330) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 100.00 (100.0–100.0) 0.00 (0.00–0.00) 0.00 (0.00–0.00)

(B) The average percentage of classifications using three experts who routinely identify species of Gyrodactylus but not necessarily from salmonids and pike

G. salaris (n = 41) 8.94 (0.00–26.83) 20.33 (9.76–39.02) 1.63 (0.00–2.44) 0.00 (0.00–0.00) 59.35 (31.71–87.80) 9.76 (0.00–29.27) 40.65 (12.20–68.29)

G. thymalli (n = 29) 4.60 (0.00–13.79) 71.26 (51.72–89.66) 11.49 (0.00–31.03) 0.00 (0.00–0.00) 10.34 (3.45–17.24) 1.15 (0.00–3.45) 89.66 (82.76–96.55)

G. lucii (n = 18) 5.56 (0.00–16.67) 11.11 (5.56–22.22) 51.85 (38.89–72.22) 0.00 (0.00–0.00) 25.93 (5.56–50.00) 7.41 (0.00–22.22) 75.93 (50.00–94.44)

G. teuchis (n = 4) 16.67 (0.00–50.00) 0.00 (0.00–0.00) 33.33 (25.00–50.00) 0.00 (0.00–0.00) 41.67 (0.00–75.00) 8.33 (0.00–25.00) 58.33 (25.00–100.00)

G. derj./trut. (n = 330) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 100.00 (100.0–100.0) 0.00 (0.00–0.00) 0.00 (0.00–0.00)

(C) The average percentage of classifications across all six Gyrodactylus experts

G. salaris (n = 41) 45.53 (0.00–95.12) 14.63 (7.32–39.02) 0.81 (0.00–2.44) 0.41 (0.00–2.44) 32.52 (0.00–87.80) 6.10 (0.00–29.27) 67.48 (12.20–100.00)

G. thymalli (n = 29) 4.60 (0.00–13.79) 81.61 (51.72–100.0) 5.75 (0.00–31.03) 0.00 (0.00–0.00) 6.90 (0.00–17.24) 1.15 (0.00–3.45) 93.10 (82.76–100.00)

G. lucii (n = 18) 2.78 (0.00–16.67) 6.48 (0.00–22.22) 69.44 (38.89–100.00) 0.00 (0.00–0.00) 17.59 (0.00–50.00) 3.70 (0.00–22.22) 82.41 (50.00–100.00)

G. teuchis (n = 4) 8.33 (0.00–50.00) 4.17 (0.00–25.00) 16.67 (0.00–50.00) 29.17 (0.00–100.00) 37.50 (0.00–75.00) 4.17 (0.00–25.00) 62.50 (25.00–100.00)

G. derj./trut. (n = 330) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 0.00 (0.00–0.00) 100.00 (100.0–100.0) 0.00 (0.00–0.00) 0.00 (0.00–0.00)

1460

A.P.Shinnet

al./In

ternatio

nalJournalforParasito

logy40(2010)1455–1467

Author's personal copy

efficiencies for the correct identification of the G. salaris specimens

only, ranged from an average 8.9% (0–26.8%) for the NSGEs to

82.1% (70.7–95.1%) for the SGEs. The expert with the longest expe-

rience of identifying salmonid gyrodactylids correctly identified

95.1% of the G. salaris specimens (i.e. 39/41 of the true G. salaris

specimens). For the SGEs, most of the misclassifications occurred

between G. salaris and G. thymalli (9.8% across three experts). Of

these an average 17.1% (i.e. average 7/41 specimens classified by

the three SGEs) were type I errors (i.e. G. salaris being assigned to

other species other than G. thymalli), whilst only an average 4.6%

(i.e. average 1.34/29 specimens classified by the three SGEs) were

type II errors (i.e. species other than G. salaris being misclassified as

G. salaris) (Table 2A). The 6.5% of G. salaris specimens that were

classified as either ‘‘G. derjavinoides-like” or ‘‘G. truttae-like”, how-

ever, is of concern, but reflects the results of a single SGE. For the

NSGEs, the results are more serious with 59.3% of the G. salaris

specimens being ascribed to the G. derjavinoides–G. truttae group.

Surprisingly, all of the 330 G. derjavinoides–G. truttae specimens

were correctly assigned to their true class.

The visual identification exercise took the six experts

14.5 ± 12.1 h (4–32 h; �2.1 min/specimen); the three SGEs were

quicker at 9.0 ± 7.8 h (4–18 h; �1.2 min/specimen). This time does

not include that required to screen fish and prepare the specimens;

if included, the time for the whole exercise (i.e. screening, collect-

ing, processing and identifying each gyrodactylid) would be

13.7 min/gyrodactylid. Much of the time allocated to the task

was spent in assessing the general variability in hook morphology

among the images, selecting the best images to represent reference

types and the measurement of some of the key diagnostic features

(i.e. total length of hamulus and marginal hook) on the ‘‘reference

images” before confirming that the measurements were within the

range of those provided in published taxonomic descriptions.

3.3. Statistical morphometric classification

Five linear discriminant analysis-based statistical classifiers,

using different combinations of the 25 point-to-point measure-

ments made on the haptoral hooks and bars of Gyrodactylus, were

tested for their ability to classify specimens. The 217 haptors cor-

responding to the bodies processed by one of the molecular labo-

ratories were prepared for the statistical classification approach.

Only 190 haptors were measured; nine of the haptors were lost

during preparation and 18 were subsequently not measurable.

The results of each classifier are presented in Table 3 showing clas-

sification efficiencies between 8.3% (classifiers ‘‘All” and ‘‘8”) and

66.7% (classifier ‘‘M”) for G. salaris. Despite the apparent poor per-

formance of classifier ‘‘M” for many species, it successfully discrim-

inated G. salaris/G. thymalli specimens from other species. Whilst

66.7% of the G. salaris specimens were correctly identified using

classifier ‘‘M”, a proportion of those were misclassified as G. thym-

alli, which suggests that this classifier, with its current training

state, is not suitable alone for identification of G. salaris. Classifiers

‘‘HM” and ‘‘8” performed similarly with misclassifications of G.

salaris being assigned to G. thymalli. None of these three classifiers

had type I errors, which were seen in the use of the ‘‘All” (one type I

error) and ‘‘H” (four type I errors) classifiers. The ‘‘M” classifier had

the highest number of type II errors with eight type II errors. When

both types of errors are considered, the classifier giving the best

overall performance was classifier ‘‘HM” with no type I errors

and one type II error. A summary of the classifiers’ performances

is provided in Table 4.

The average time to measure a specimen was 6.6 ± 1.6 min

(n = 190 haptors prepared for morphometric assessment), approx-

imately 16.1 min if the time required to digest the haptor and

make a slide mount was also included and 18.2 min/specimen if

the process of parasite collection from the fish is included.

3.4. Statistical classification errors

To determine the possible basis of the errors in the statistical

classification approach, the type of error, the pattern of errors

and the time taken to measure difficult preparations were investi-

gated. From the results generated by the ‘‘All” classifier, there were

35 (18.4%) misidentifications out of 190 specimens. Of these, all se-

ven (3.7%, i.e. 7/190) specimens of G. lucii and the two (1.0%, i.e. 2/

190) specimens of G. teuchis were classified as G. thymalli with

posterior probabilities of 0.999 ± 0.002 and 0.999 ± 0.0007,

respectively. The ‘‘All” classifier, however, was not trained using

specimens of either species and therefore had no a priori experi-

ence of these taxa. Eleven of the 12 G. salaris specimens were mis-

classified as G. thymalli with posterior probabilities of 0.923 ±

0.126. Although higher correct classification efficiencies of G. salar-

iswere obtained for other classifiers, notably the classifier based on

marginal hook measurements, the results demonstrate that the

larger size of the G. thymalli specimens used for training the origi-

nal classifier predisposes all gyrodactylids subsequently exposed to

the classifier, to fall inside its previous experience with regard to

the size range and pattern of measurements. The remaining mis-

classifications were between G. derjavinoides (nine misclassified

as G. truttae) and G. truttae (five misclassified as G. derjavinoides).

In each case, the posterior probability of correctly assigning the

species to its true class was less than 0.9 (i.e. the values for the nine

misclassified specimens of G. derjavinoides were 0.869 ± 0.150

while those of the 94 specimens correctly assigned were

0.968 ± 0.085). Similarly, the five misclassified specimens of G.

truttae had posterior probability values of 0.824 ± 0.104 while the

values of the 46 correctly assigned specimens were 0.964 ± 0.085.

We tested whether researchers were more likely to make errors

in the early or late stages of measuring large numbers of speci-

mens. An analysis of the proportion of errors that were made

against the time spent in a measuring session (maximum

140 min) found no relationship (R2 = 0.0004) between session

length and classification error (P = 0.947). Equally, there was no

significant relationship between the number of errors made and

the time from a rest break (Mann–Whitney, P = 0.584) and no evi-

dence (P = 0.424) to suggest that specimens which took longer to

measure were more likely to be amongst the group of specimens

which were misidentified.

3.5. Molecular identification

ITS-RFLP successfully distinguished all G. teuchis and G. lucii

specimens from the control RFLP patterns for G. salaris, G. derjavi-

noides and G. truttae. Subsequent sequencing of the ITS region for

those specimens giving an ambiguous or unknown ITS restriction

pattern, correctly identified all the G. lucii and G. teuchis specimens.

Only one specimen of G. derjavinoides was misclassified as G. trut-

tae, and one specimen of G. derjavinoides misclassified as G. salaris.

The latter, however, was subsequently determined to be due to

analysis of the wrong parasite lysate. All other specimens were cor-

rectly assigned to their true identity. Only one (0.48%) specimen

was not classified and represents a failed ITS PCR amplification.

Sequencing of the COI successfully differentiated G. salaris (i.e.

specimens from salmon) from G. thymalli (i.e. specimens from

grayling). The COI analysis was further able to assign the G. thymalli

specimens to a clade currently represented by G. thymalli speci-

mens from the River Test, UK (Hansen et al., 2007a), and the

G. salaris specimens to clade III (Hansen et al., 2003), a clade repre-

sented by G. salaris specimens from a number of different rivers

and from both salmon and rainbow trout. The origin of the

G. thymalli (UK rivers) and the G. salaris (S. salar from Lærdalselva,

Norway; O. mykiss from Mosbjerg, N. Jutland, Denmark) specimens

A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467 1461

Au

tho

r's p

ers

on

al c

op

y

Table 3

Classification of 190 specimens of Gyrodactylus based on different combinations of the 25 point-to-point measurements made on the proteolytic digested haptors. The classifications are based on the posterior probabilities (i.e. 1 = most

probable; 0 = improbable) generated by a trained linear discriminant analysis-based statistical classifier. The classifier had no a priori experience of Gyrodactylus lucii or Gyrodactylus teuchis. The column ‘‘G. salaris-like” provides a

summary of the specimens that were classified to a species resembling Gyrodactylus salaris, i.e. either G. salaris, Gyrodactylus thymalli, G. lucii or G. teuchis. The figures shown in bold highlight the percentage of specimens correctly

allocated to their species class presented as the actual number of specimens followed by a percentage in parentheses.

True class Predicted class

Gyrodactylus

derjavinoides

Gyrodactylus lucii Gyrodactylus

salaris

Gyrodactylus

teuchis

Gyrodactylus

thymalli

Gyrodactylus

truttae

Gyrodactylus

sp. (other)

‘‘G. salaris-like”

(A) Correct classification of specimens using a classifier (‘‘All”) that uses all 25 point-to-point measurements made on the haptoral armature of each Gyrodactylus specimen

G. derj. (n = 105) 95/105 (90.5) 0/105 (0.0) 1/105 (1.0) 0/105 (0.0) 0/105 (0.0) 9/105 (8.6) 0/105(0.0) 1/105 (1.0)

G. lucii (n = 7) 0/7 (0.0) 0/7 (0.0) 0/7 (0.0) 0/7 (0.0) 7/7 (100.0) 0/7 (0.0) 0/7 (0.0) 7/7 (100.0)

G. salaris (n = 12) 0/12 (0.0) 0/12 (0.0) 1/12 (8.3) 0/12 (0.0) 10/12 (83.3) 0/12 (0.0) 1/12 (8.3) 11/12 (91.7)

G. teuchis (n = 2) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0)

G. thymalli (n = 13) 0/13 (0.0) 0/13 (0.0) 0/13 (0.0) 0/13 (0.0) 13/13 (100.0) 0/13 (0.0) 0/13 (0.0) 13/13 (100.0)

G. truttae (n = 51) 5/51 (9.8) 0/51 (0.0) 0/51 (0.0) 0/51 (0.0) 0/51 (0.0) 46/51 (90.2) 0/51 (0.0) 0/51 (0.0)

(B) Correct classification of specimens using a classifier (‘‘H”) based on the 11 variables measured on the hamuli of each Gyrodactylus specimen

G. derj. (n = 105) 98/105 (93.3) 0/105 (0.0) 1/105 (1.0) 0/105 (0.0) 0/105 (0.0) 4/105 (3.8) 2/105 (1.9) 1/105 (1.0)

G. lucii (n = 7) 0/7 (0.0) 0/7 (0.0) 0/7 (0.0) 0/7 (0.0) 6/7 (85.7) 0/7 (0.0) 1/7 (14.3) 6/7 (85.7)

G. salaris (n = 12) 0/12 (0.0) 0/12 (0.0) 2/12 (16.7) 0/12 (0.0) 6/12 (50.0) 0/12 (0.0) 4/12 (33.3) 8/12 (66.7)

G. teuchis (n = 2) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 1/2 (50.0) 0/2 (0.0) 1/2 (50.0) 1/2 (50.0)

G. thymalli (n = 13) 0/13 (0.0) 0/13 (0.0) 0/13 (0.0) 0/13 (0.0) 13/13 (100.0) 0/13 (0.0) 0/13 (0.0) 13/13 (100.0)

G. truttae (n = 51) 5/51 (9.8) 0/51 (0.0) 1/51 (2.0) 0/51 (0.0) 0/51 (0.0) 45/51 (88.2) 0/51 (0.0) 1/51 (2.0)

(C) Correct classification of specimens using a classifier (‘‘M”) based on the eight variables measured on the marginal hooks of each Gyrodactylus specimen

G. derj. (n = 105) 54/105 (51.4) 0/105 (0.0) 2/105 (1.9) 0/105 (0.0) 0/105 (0.0) 44/105 (41.9) 5/105 (4.7) 2/105 (1.9)

G. lucii (n = 7) 0/7 (0.0) 0/7 (0.0) 1/7 (14.3) 0/7 (0.0) 6/7 (85.7) 0/7 (0.0) 0/7 (0.0) 7/7 (100.0)

G. salaris (n = 12) 0/12 (0.0) 0/12 (0.0) 8/12 (66.7) 0/12 (0.0) 4/12 (33.3) 0/12 (0.0) 0/12 (0.0) 12/12 (100.0)

G. teuchis (n = 2) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0)

G. thymalli (n = 13) 0/13 (0.0) 0/13 (0.0) 2/13 (15.4) 0/13 (0.0) 11/13 (84.6) 0/13 (0.0) 0/13 (0.0) 13/13 (100.0)

G. truttae (n = 51) 2/51 (3.9) 0/51 (0.0) 3/51 (5.9) 0/51 (0.0) 0/51 (0.0) 45/51 (88.2) 1/51 (2.0) 3/51 (5.9)

(D) Correct classification of specimens using a classifier (‘‘HM”) that uses all nineteen measured variables taken from the hamuli and marginal hooks of each specimen of Gyrodactylus

G. derj. (n = 105) 96/105 91.4) 0/105 (0.0) 1/105 (1.0) 0/105 (0.0) 0/105 (0.0) 8/105 (7.6) 0/105 (0.0) 1/105 (1.0)

G. lucii (n = 7) 0/7 (0.0) 0/7 (0.0) 0/7 (0.0) 0/7 (0.0) 7/7 (100.0) 0/7 (0.0) 0/7 (0.0) 7/7 (100.0)

G. salaris (n = 12) 0/12 (0.0) 0/12 (0.0) 3/12 (25.0) 0/12 (0.0) 9/12 (75.0) 0/12 (0.0) 0/12 (0.0) 12/12 (100.0)

G. teuchis (n = 2) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0)

G. thymalli (n = 13) 0/13 (0.0) 0/13 (0.0) 0/13 (0.0) 0/13 (0.0) 13/13 (100.0) 0/13 (0.0) 0/13 (0.0) 13/13 (100.0)

G. truttae (n = 51) 4/51 (7.8) 0/51 (0.0) 0/51 (0.0) 0/51 (0.0) 0/51 (0.0) 47/51 (92.2) 0/51 (0.0) 0/51 (0.0)

(E) Correct classification of Gyrodactylus specimens using a classifier (‘‘8”) based on the best eight point-to-point measurements suggested by a forward stepwise discriminate analysis applied to all the haptoral hook

measurement variables. The eight best measurements used in this study include the total length of the marginal hook, the hamulus point length, the marginal hook sickle distal width, the ventral bar process length, the

marginal hook sickle length, the hamulus aperture length, the total length of the ventral bar and the total width of the ventral bar

G. derj. (n = 105) 96/105 (91.4) 0/105 (0.0) 1/105 (1.0) 0/105 (0.0) 0/105 (0.0) 8/105 (7.6) 0/105 (0.0) 1/105 (1.0)

G. lucii (n = 7) 0/7 (0.0) 0/7 (0.0) 1/7 (14.3) 0/7 (0.0) 6/7 (85.7) 0/7 (0.0) 0/7 (0.0) 7/7 (100.0)

G. salaris (n = 12) 0/12 (0.0) 0/12 (0.0) 1/12 (8.3) 0/12 (0.0) 11/12 (91.7) 0/12 (0.0) 0/12 (0.0) 12/12 (100.0)

G. teuchis (n = 2) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0)

G. thymalli (n = 13) 0/13 (0.0) 0/13 (0.0) 1/13 (7.7) 0/13 (0.0) 12/13 (92.3) 0/13 (0.0) 0/13 (0.0) 13/13 (100.0)

G. truttae (n = 51) 6/51 (11.7) 0/51 (0.0) 0/51 (0.0) 0/51 (0.0) 0/51 (0.0) 45/51 (88.2) 0/51 (0.0) 0/51 (0.0)

1462

A.P.Shinnet

al./In

ternatio

nalJournalforParasito

logy40(2010)1455–1467

Author's personal copy

used in the comparative analyses supports the predicted COI-based

clades.

Three analysts, in one laboratory, worked simultaneously pro-

cessing 73–74 gyrodactylid specimens, each analyst taking an

average of 5.3 ± 1.65 h (n = 221; total time = 16.1 h; average

4.37 min/specimen) to pick the specimens out and to place them

into lysate. A gyrodactylid body could not be found in 11 of the

Eppendorf tubes and so 210 specimens were prepared for ITS-RFLP,

each analyst taking 17.5 ± 0.42 h (total time = 52.5 h) to prepare

their batch of specimens for assessment by ITS-RFLP. Specimens

with a ‘‘G. salaris/G. thymalli-like” banding pattern, were then

passed directly to COI sequencing. Sequencing of the ITS regions

of the 14 specimens giving ambiguous or novel banding patterns

took a further 6.17 h while the COI sequencing and phylogenetic

analysis of the 28 specimens identified as G. salaris/G. thymalli by

the initial ITS-RFLP step took 32.6 h. The total hands-on time was

72.6 h which included the extra time required to re-analyse spec-

imens giving unique (i.e. G. teuchis and G. lucii) or ambiguous

ITS-RFLP patterns (n = 2) and COI sequences (n = 3). The total time

to process the 210 specimens was 127.1 h, a time which can be sig-

nificantly reduced if the focus remains on the identification of G.

salaris rather than on the specific identification of each specimen

within the analysis.

The loss of specimens from one of the molecular laboratories is

considered atypical but nevertheless highlights the fact that extra

precautions need to be taken to protect against large scale sample

losses, including structured division of samples between laborato-

ries and processing runs.

3.6. Comparison of the three identification approaches

All three identification approaches (visual identification, mor-

phometric and molecular methods) are compared in Table 6 and

Supplementary Fig. S1. Using the combined molecular approach

(ITS-RFLP and COI sequencing) offers the greatest likelihood of G.

salaris being detected and identified and of all other specimens

being correctly assigned to their true species (Table 6). Supplemen-

tary Fig. S1 also includes estimates of the time required to under-

take each step and the probability of either a G. salaris or a ‘‘G.

salaris-like” specimen being correctly identified. The visual identi-

fication approach was the fastest method (13.7 min/specimen), fol-

lowed by the statistical classifier (18.2 min/specimen) and then the

molecular approach (38.4 min/specimen (although in reality arriv-

ing at an identification for any single specimen will only be

achieved at the end of the complete ITS-RFLP process). Each meth-

od includes the time taken to collect a gyrodactylid from a fish,

perform the appropriate method and to obtain an identification.

The timings for the molecular method, however, were based on

simultaneously processing 74 specimens per run. The visual iden-

tification process involving six experts took an average of

14.5 ± 12.2 h, a figure which would undoubtedly decrease with

successive rounds, as much of the time used for this exercise was

spent in assessing the morphological variation among the speci-

mens and, in the absence of host details, in determining how many

species were present and then selecting reference specimens for

each species from among all the images/specimens that were pro-

vided. Although host details can be used to prioritise the order in

which samples are analysed, there must be a limit to the emphasis

placed upon these as G. salaris has the ability to accidentally trans-

fer to and survive for relatively long periods on non-preferred

hosts. While the time taken to visually identify the 422 gyrodacty-

lids can be improved upon, the time required for the two analytical

methodologies, the morphological and the molecular based ap-

proaches, are unlikely to change dramatically.

Inter-analyst agreement for 189 worms (following removal of

missing values and only using SGE visual results and ‘‘HM” classi-

fier results) had an overall value of 0.8 (Fleiss’ Kappa, z = 28.9;

P < 0.001) for all three techniques. Such agreement is rated ‘‘sub-

stantial agreement” (Landis and Koch, 1977) according to usual

Kappa assessment protocols. However, results varied according

to species (Table 7), with the methods identifying G. derjavinoides

and G. truttae with near perfect agreement, but G. salaris with only

moderate agreement.

As the morphometric classifier employed here relied upon a

training set that did not include some of the species used in the

current trial (i.e. G. teuchis), it was predicted to perform sub-opti-

mally, and this was clearly apparent from these results. If the anal-

ysis of agreement is carried out using only the results from the

molecular and visualisation analyses (SGE only), a far higher agree-

ment betweenmethods is observed (Table 8). In this case, the over-

all value of Kappa for 189 worms is 0.983 (z = 20.8; P < 0.001)

which can be described as a near perfect agreement. Although

there is some variation in accuracy of identification between spe-

cies, all are nonetheless accurately assigned with near perfect

agreement (Kappa > 0.93 for all species, P < 0.001), with the critical

species G. salaris being identified with perfect agreement between

methods (Kappa = 1; z = 13.75; P < 0.001).

4. Discussion

This study investigated the analytical process for identification

of Gyrodactylus spp. from salmonids, following the entire process

from the receipt of Gyrodactylus-infected host specimens to final

species determination using three selected methodologies (visual

identification, statistical classification, molecular characterisation).

As part of this investigation, this study documented processing

Table 4

Summary of classifier performance. The classifiers giving the highest correct classification of each true class and the lowest number of misclassifications are given as a percentage

followed by the number of specimens allocated to that class in parentheses. As the classifiers had no a priori experience of Gyrodactylus lucii or Gyrodactylus teuchis, their

allocation to the Gyrodactylu thymalli class is scored (a).

True class Predicted class

Gyrodactylus derjavinoides Gyrodactylus

lucii

Gyrodactylus salaris Gyrodactylus

teuchis

Gyrodactylus thymalli Gyrodactylus truttae

G. derj. (n = 105) 93.3 (98/105) (H) na 1.0 (1/105) (All, H, M, 8) na 0.0 (0/105) (All, H, M, HM, 8) 3.8 (4/105) (H)

G. lucii (n = 7) 0.0 (0/7) (All, H, M, HM, 8) na 0.0 (0/7) (All, H, HM) na 100.0* (7/7) (All, HM) 0.0 (0/7) (All, H, M, HM, 8)

G. salaris (n = 12) 0.0 (0/12) (All, H, M, HM, 8) na 66.7 (8/12) (M) na 33.3 (4/12) (M) 0.0 (0/12) (All, H, M, HM, 8)

G. teuchis (n = 2) 0.0 (0/2) (All, H, M, HM, 8) na 0.0 (0/2) (All, H, M, HM, 8) na 100.0a (2/2) (All, M, HM, 8) 0.0 (0/2) (All, H, M, HM, 8)

G. thymalli (n = 13) 0.0 (0/13) (All, H, M, HM, 8) na 0.0 (0/13) (All, H, HM) na 100.0 (13/13) (All, H, HM) 0.0 (0/13) (All, H, M, HM, 8)

G. truttae (n = 51) 3.9 (2/51) (M) na 0.0 (0/51) (All, HM, 8) na 0.0 (0/51) (All, H, M, HM, 8) 92.2 (47/51) (HM)

All, classification based on the use of all 25 point-to-point measurements; H, classification based only on hamuli variables; HM, classification is based on the use of hamuli

and marginal hook variables; M, classification based only on marginal hook variables; na, classifications not provided as the classifiers have no a priori experience of either G.

lucii or G. teuchis; 8, classification based on the use of eight selected variables measured on the haptoral armature. The figures shown in bold highlight the percentage of

specimens correctly allocated to their species class.

A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467 1463

Author's personal copy

bottlenecks, loss of critical specimens and experimental errors. The

study also set out to investigate possible trade-offs that could oc-

cur between speed, capacity and accuracy, and to assess how these

parameters might be affected by the method selected. The impor-

tance of each of these parameters is clearly affected by the purpose

of the sampling in question, i.e. slow turn-around, low-volume

routine surveillance versus rapid turn-around, high volume

event-driven scenarios during an expected outbreak (Supplemen-

tary Figs. S2 and S3). The purpose of this study was therefore to ob-

tain accurate estimates of performance under realistic conditions

in order to assist in the design of a robust SOP, which maximises

the number of specimens that can be handled (throughput) whilst

ensuring the highest probability of correct species identification.

This study does not, however, make comment on the field col-

lection of salmonids or on the number of gyrodactylids that should

be sampled from each fish. With reference to the latter, however, it

is recommended that the surveillance guidelines on detecting lev-

els of disease within a population of a given size are followed (e.g.

Ossiander and Wedermeyer, 1973; Cameron, 2002).

This study suggests that, when a small number of specimens

need to be identified, all three methods, when performed by ade-

quately trained personnel, are effective in discriminating ‘‘G. salar-

is-like” specimens, but only the combined molecular approach

provides a definitive identity. The best method for the accurate

screening of G. salaris, in descending order, is molecular (100%, Ta-

bles 5 and 6), visual identification (82.1–85.3% by SGEs, 8.9–29.4%

by NSGEs, Tables 2 and 6) and statistical classification (17.7–66.7%,

Tables 3, 4 and 6). The separation of the haptor and body is a nec-

essary step, permitting a second identification method to be ap-

plied for confirmation if the first method indicates that G. salaris

is suspected. Further, the separation of each gyrodactylid into

two parts insures against, to a certain degree, inherent losses. If

whole specimens analysed by the molecular approach are found

to have degraded or fail to produce an amplified PCR product (as

happened for this study in one laboratory), then determining the

identity of the specimen is no longer possible. If the gyrodactylid

haptor has been retained as a slide preparation, then identification

by a second technique, i.e. visual identification or morphometrics,

is still possible (see feedback loops identified in Supplementary

Fig. S2). The loss of material by one of the laboratories conducting

the molecular analysis is regarded as an informative but atypical

error. Such errors do, however, highlight potential sources of loss

within the identification process and argue for the inclusion of

additional control steps to make provision for this.

The morphometric approach, however, can also be improved

through the further training of the statistical classifiers to take ac-

count of the species, those that have not been encountered before,

and to modify the architecture of the classifiers such that they are

multi-staged and able to remove particular species at each internal

layer (McHugh et al., 2000).

While the molecular method outperforms the other two meth-

ods with regard to the correct identification of G. salaris, during

periods of crisis, when a large number of specimens require species

identification, time becomes the limiting factor. Under such situa-

tions it is suggested that gyrodactylids are first prioritised by host,

with those from salmon being analysed first, followed by, in rank

order, rainbow trout, grayling, charr and then brown trout (see

Bakke et al. (2007) for a review of host susceptibility to G. salaris;

Supplementary Fig. S2), and, by using the visual method of identifi-

cation, assuming adequately trained personnel. The suggested or-

der of specimen priority, however, relies upon the assumption

that populations of UK brown trout are as resistant to G. salaris

as Scandinavian brown trout populations.

When the timing results are considered, a combination of three

experts, without the need for consensus identification, might be

considered as the best way forward to pre-screen specimens in

outbreak situations. In this instance, it would not be necessary

for the experts to agree by consensus as to the identity of every

specimen but rather for any specimen suspected of being ‘‘G. salar-

is-like” to be submitted for validation by a second method. This ap-

proach would allow for up to ca. 400 specimens per day to be

visually assessed from images by each assessor, the fastest of

whom took 4 h for the 433 specimens, and would provide a signif-

icant reduction in the number of specimens put forward for valida-

tion by a second identification method. Use of physical slides

instead of images would reduce the number of samples that could

be screened per day. Use of visual screening can help to reduce the

number of molecular tests needed, by flagging priority specimens

and exempting those that are definitely not G. salaris or related

species. The time taken to prepare specimens in this study was

�9.6 min/specimen and included the time to completely remove

the tissues surrounding the hooks enabling the expert performing

the visual identification to have an unobscured view. While this is

not essential for identification, it decreases the probability of mak-

ing an incorrect identification. Assuming that slide preparation and

body lysis is performed for each individual parasite as recom-

mended above, then the 9.6 min preparation time is common to

all methodologies.

Where an outbreak is suspected, all collected specimens may be

considered critical and therefore extra precautions need to be ta-

ken to avoid loss of data. Large sample sets should therefore be di-

vided in a structured way between diagnostic facilities, and within

each facility, critical steps, e.g. DNA extraction, should not be car-

ried out on all specimens simultaneously. Whilst three sites were

unrepresented in the molecular analysis, none of which contained

specimens of G. salaris, the fact that visual methods were used

simultaneously, ensured that they were screened nevertheless.

This provides a strong argument for obtaining material suitable

for both molecular and morphological analyses, e.g. DNA and hook

preparations, during periods of a suspected outbreak.

Table 5

Classification of 210 specimens of Gyrodactylus based on restriction fragment length polymorphism (RFLP) banding patterns and derived internal transcribed spacer (ITS) and

cytochrome c oxidase I (COI) sequences from one of the two laboratories performing the molecular analyses. All figures represent classification efficiencies based on molecular

identifications compared to the true identity of specimens. The figures shown in bold highlight the actual number and percentage of specimens correctly allocated to their species

class. Annotations: 1Includes one specimen classified as belonging to Gyrodactylus salaris/Gyrodactylus thymalli based on its RFLP pattern but subsequently not assigned to either

species class when its COI sequence was determined – subsequently it was discovered that the wrong lysate had been used in the ITS-RFLP analyses and a re-analysis gave a

Gyrodactylus derjavinoides RFLP pattern.

True class Predicted class

Gyrodactylus

derjavinoides

Gyrodactylus lucii Gyrodactylus salaris Gyrodactylus

teuchis

Gyrodactylus

thymalli

Gyrodactylus

truttae

Failed PCR

G. derjavinoides (n = 114) 112/114 (98.2)1 0/114 (0.0) 0/114 (0.0) 0/114 (0.0) 0/114 (0.0) 1/114 (0.9) 1/114 (0.9)

G. lucii (n = 11) 0/11 (0.0) 11/11 (100.0) 0/11 (0.0) 0/11 (0.0) 0/11 (0.0) 0/11 (0.0) 0/11 (0.0)

G. salaris (n = 14) 0/14 (0.0) 0/14 (0.0) 14/14 (100.0) 0/14 (0.0) 0/14 (0.0) 0/14 (0.0) 0/14 (0.0)

G. teuchis (n = 2) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0) 2/2 (100.0) 0/2 (0.0) 0/2 (0.0) 0/2 (0.0)

G. thymalli (n = 16) 0/16 (0.0) 0/16 (0.0) 0/16 (0.0) 0/16 (0.0) 16/16 (100.0) 0/16 (0.0) 0/16 (0.0)

G. truttae (n = 53) 0/53 (0.0) 0/53 (0.0) 0/53 (0.0) 0/53 (0.0) 0/53 (0.0) 53/53 (100.0) 0/56 (0.0)

1464 A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467

Author's personal copy

Under conditions of a suspected G. salaris outbreak when the

number of specimens entering a diagnostic laboratory is expected

to increase, it is recommended that there is a switch from using the

ITS-RFLP method of identification as the initial screen to that of

using visual determination (Supplementary Figs. S2 and S3). Simi-

larly, there is also a change in the rigour of providing a precise spe-

cies identity from ‘‘G. salaris” or ‘‘another determined species not G.

salaris” to ‘‘G. salaris-like” or ‘‘not G. salaris-like”. This approach al-

lows for a reduction in the number of specimens that then pass for-

ward to further methods of identification, i.e. simultaneous

assessment by ITS-RFLP-COI sequencing and visual confirmation

by a panel of world experts (Supplementary Fig. S3). This approach

is important as it allows for a degree of subjectivity in the visual

identification method by NSGEs. This makes allowances for experi-

ence which is key to providing a precise species identification com-

pared with the time required to train personnel to perform the

molecular analyses. The 6.5% misidentification at the ‘‘G. salaris-

like level” derived from one SGE as presented in Table 2 appears

worrying, given that the ‘‘expertise” was assumed, but in this exer-

cise those undertaking the visual identification exercise were spe-

cifically asked to provide an identity for each specimen where

possible. Changing the grading criteria to ‘‘G. salaris-like” or ‘‘not

G. salaris-like”, allows for NSGEs to undertake the work and for

any specimen resembling G. salaris or those of questionable

identity to be put forward for assessment by a second method. This

rapid screening approach increases the speed of specimen

throughput and reduces the number of specimens to be identified

by the molecular approach.

The combined inclusion of molecular and visual-morphometric

based methods as options in both analysis pipelines (Supplemen-

Table 6

A summary and comparison of all identification approaches commonly used for the identification of Gyrodactylus salaris and congeners. The correct classification efficiencies

determined by the three methods for 218 specimens of Gyrodactylus are given. Specimens that were correctly allocated to their proper species class are shown in a bold font.

True class Predicted class within each identification method

Visual identificationa experts Visual identificationb

non-experts

Statistical classifierc RFLP-ITS-COI sequencingd

Gyrodactylus derjavinoides (n = 116) Der as Der (109.5/116; 94.4%) Der as Der (103/116; 88.8%) Der as Der (54/105; 51.4%) Der as Der (112/114; 98.2%)

Der as Trut (1.5/116; 1.3%) Der as Trut (5/116; 4.3%) Der as Trut (44/105; 41.9%) Der as Sal (1/114; 0.9%)

Der as NS (5/116; 4.3%) Der as NS (7/116; 6.0%) Der as Sal (2/105; 1.9%) Der as NS (1/114; 0.9%)

Der as Sal (1/116; 0.8%) Der as G. sal-like (2/105; 1.9%)

Der as NS (5/105; 4.7%)

Gyrodactylus lucii (n = 11) Luc as Luc (8.5/11; 77.3%) Luc as Luc (5/11; 45.5%) Luc as Luc (0/7; 0.0%) Luc as Luc (11/11; 100.0%)

Luc as Thy (0.5/11; 4.6%) Luc as Thy (3/11; 27.3%) Luc as Thy (6/7; 85.7%)

Luc as NS (2/11; 18.2%) Luc as Sal (3/11; 27.3%) Luc as G. sal-like (7/7; 100.0%)

Gyrodactylus salaris (n = 17) Sal as Sal (14.5/17; 85.3%) Sal as Sal (5/17; 29.4%) Sal as Sal (8/12; 66.7%) Sal as Sal (14/14; 100.0%)

Sal as Teu (0.5/17; 2.9%) Sal as Trut (2/17; 11.8%) Sal as Thy (4/12; 33.3%)

Sal as Thy (1/17; 5.9%) Sal as Thy (8/17; 47.1%) Sal as G. sal-like (12/12; 100.0%)

Sal as NS (1/17; 5.9%) Sal as Luc (2/17; 11.8%)

Gyrodactylus teuchis (n = 2) Teu as Teu (1/2; 50.0%) Teu as Teu (0/2; 0.0%) Teu as Teu (0/2; 0.0%) Teu as Teu (2/2; 100.0%)

Teu as Trut (1/2; 50.0%) Teu as Luc (1/2; 50.0%) Teu as Thy (2/2; 100.0%)

Teu as Sal (1/2; 50.0%) Teu as G. sal-like (2/2; 100.0%)

Gyrodactylus thymalli (n = 16) Thy as Thy (15.5/16; 96.9%) Thy as Thy (6/16; 37.5%) Thy as Thy (11/13; 84.6%) Thy as Thy (16/16; 100.0%)

Thy as Sal (0.5/16; 3.1%) Thy as Sal (3/16; 18.7%) Thy as Sal (2/13; 15.4%)

Thy as Luc (6/16; 37.5%) Thy as G. sal-like (13/13; 100.0%)

Thy as NS (1/16; 6.3%)

Gyrodactylus truttae (n = 56) Trut as Trut (49/56; 87.5%) Trut as Trut (42/56; 75.0%) Trut as Trut (45/51; 88.2%) Trut as Trut (53/53; 100.0%)

Trut as Der (4/56; 7.1%) Trut as Der (3/56; 5.4%) Trut as Der (2/51; 3.9%)

Trut as NS (3/56; 5.4%) Trut as NS (2/56; 3.6%) Trut as Sal (3/51; 5.9%)

Trut as Sal (8/56; 14.2%) Trut as NS (1/51; 2.0%)

Trut as Thy (1/56; 1.8%)

Der, Gyrodactylus derjavinoides; Luc, Gyrodactylus lucii; NS, not scoreable; Sal, Gyrodactylus salaris; Som, Gyrodactylus sommervillae; Teu, Gyrodactylus teuchis; Thy, Gyrodactylus

thymalli; Trut, Gyrodactylus truttae.a Results represent the average score of the top two performing salmonid Gyrodactylus experts.b Results are based on the average score of two non-salmonid Gyrodactylus experts.c Results of the classifier ‘‘All” which uses all 25 point-to-point measurements made on the haptoral hooks.d Results are based on the 210 specimens available for analysis, the remaining specimens were lost during the specimen preparation stage.

Table 7

Fleiss’ Kappa scores for individual Gyrodactylus sp. assessed using visual assessment,

morphometric analysis and molecular techniques (n = 189 worms, missing values

removed).

Gyrodctylus spp. Kappa z P value Scoring

Gyrodactylus

derjavinoides

0.89 21.10 <0.001 Near perfect

agreement

Gyrodactylus lucii 0.45 10.77 <0.001 Moderate agreement

Gyrodactylus salaris 0.57 13.63 <0.001 Moderate agreement

Gyrodactylus teuchis 0.5.0 11.86 <0.001 Moderate agreement

Gyrodactylus thymalli 0.66 15.62 <0.001 Substantial

agreement

Gyrodactylus truttae 0.87 20.82 <0.001 Near perfect

agreement

Table 8

Fleiss’ Kappa scores for individual Gyrodactylus sp. assessed using morphometric

analysis and molecular techniques only (n = 189 worms, missing values removed).

Gyrodactylus spp. Kappa z P value Scoring

Gyrodactylus

derjavinoides

0.99 13.60 <0.001 Near perfect

agreement

Gyrodactylus lucii 0.93 12.79 <0.001 Near perfect

agreement

Gyrodactylus salaris 1.00 13.75 <0.001 Near perfect

agreement

Gyrodactylus teuchis 1.00 13.75 <0.001 Near perfect

agreement

Gyrodactylus thymalli 0.96 13.23 <0.001 Near perfect

agreement

Gyrodactylus truttae 0.99 13.56 <0.001 Near perfect

agreement

A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467 1465

Author's personal copy

tary Figs. S2 and S3) provides an optimal diagnostic approach as

recommended by the Office International des Epizooties (OIE,

2009) for the identification and confirmation of G. salaris.

It should be mentioned that a number of G. salaris strains which

are not pathogenic to S. salar have been described in recent years

(Jørgensen et al., 2007; Lindenstrøm et al., 2003; Olstad et al.,

2007; Robertsen et al., 2007). These strains have a low number of

nucleotide differences in their ITS rDNA which can be used to de-

velop specific RFLP-based diagnostic tests (Kania et al., 2007). Hae-

III restriction will not differentiate the majority of these non-

pathogenic strains from the pathogenic G. salaris and neither will

COI sequencing (Hansen et al., 2007b; Jørgensen et al., 2007; Rob-

ertsen et al., 2007). Morphometrical analysis may be suitable to

distinguish between some (Robertsen et al., 2007) but not all

(Jørgensen et al., 2007) non-pathogenic strains although this has

not been tested using the statistical classifiers used here.

Undertaking this exercise has also identified a number of prac-

tical bottlenecks notably with regard to the manpower available

that could be recruited to assist in processing specimens at times

of high specimen throughput. To maintain the availability of, for

example, three experts for routine screening in non-outbreak situ-

ations would be a drain on laboratory resources and/or the good-

will of the experts depending on the agreement for their services.

At current staffing levels in the molecular laboratories participat-

ing in this exercise, the ITS-RFLP method represents the most reli-

able method for screening (but with the incorporation of a step

which preserves the haptor). The switch to the visual method dur-

ing times of a suspected outbreak, however, does not specifically

require experts as the analysis pipeline allows for the grading of

specimens by researchers with lower levels of experience, i.e. into

‘‘G. salaris-like” and ‘‘not G. salaris-like”. Event-driven situations,

however, also pose further questions of training needs and

whether certain tasks should be sub-divided, i.e. ‘‘fish screeners”,

‘‘haptor preparers”, ‘‘visual identifiers”, etc. They also raise ques-

tions concerning the logistics/feasibility of recruiting external ex-

perts during an outbreak as they are likely to be engaged in

other activities. The answers to these questions, however, depend

to a certain extent upon internal management policies and avail-

able resources.

In this study, morphometric measurements were made on the

digital images that were also sent to the experts for visual identi-

fication. While it would be normal practice to make measurements

directly from the slide mounted specimens, the decision to take

measurements from images allowed a three-way cross-validation

between researchers independently measuring the same speci-

mens. Given the limited time allocated for the study to simulate

a crisis situation, it was necessary to perform several identification

methods simultaneously. While it was anticipated that this would

provide a direct comparison between the identifications made by

visual inspection and the statistical classification approach, this ap-

proach restricted the researchers to extracting data from images

taken at a limited number of focal depths. Three researchers were

used for the morphometric measurement exercise, one was a gyro-

dactylid expert who routinely made measurements, the other two

were researchers who were recruited to undertake the study. Each

researcher was given 2 days training, where necessary, and then a

further 2 weeks (80 h) to measure as many of the 190 gyrodacty-

lids as possible. Within this period, the researchers measured

130, 140 and 190 specimens, respectively, the latter, the expert,

taking only 20.8 h. The number of specimens measured by the first

and second researchers was not sufficient, however, to permit a ro-

bust comparison with the truncated number of results from the

molecular approach.

The current sample of gyrodactylids for identification included

specimens of G. teuchis and G. lucii, both of which were not used

in the construction and training of the classifiers used in this study.

Of the species used for the training of the classifiers (Gyrodactylus

arcuatus Bychowsky, 1933, G. derjavinoides, Gyrodactylus gasterostei

Gläser, 1974, Gyrodactylus kherulensis Ergens, 1974, G. salaris, Gyro-

dactylus sommervillae Turgut, Shinn, Yeomans & Wootten, 1999, G.

thymalli and G. truttae) the haptoral hooks of G. thymalli were the

largest (83.59 ± 1.83 total hamulus length; 41.00 ± 2.45 total mar-

ginal hook length). As the classifiers had no a priori experience of

either G. lucii or G. teuchis, the specimens were assigned to the larg-

est species class, i.e. G. thymalli, into which their measurements

would fit. In the current architectural structure of the classifiers,

it is likely, therefore, that a disproportionate number of specimens

are likely to be assigned to the G. thymalli class. Further evidence

for this is provided by the classifications of the G. salaris and G.

thymalli specimens using the classifier ‘‘H” (Table 3B). Here, it

can be seen that all of the G. thymalli specimens are correctly as-

signed to their true class but the majority of G. salaris specimens

were misclassified and placed into the biggest species class into

which they would fit, i.e. G. thymalli. This may be partly due to

the fact that the classifier was trained using a far more constrained

set of G. salaris derived largely from populations that were not

those sampled for the present study. The study of McHugh et al.

(2000) suggested that the use of a two stage classifier based on lin-

ear discriminant analysis (LDA) and k-nearest neighbour (KNN)

methods were effective in separating species into either G. salar-

is–G. thymalli types or into G. derjavinoides–G. truttae types in a first

round of classification. In a second round of classification, speci-

mens allocated to each species pairing were then further separated

into one of the two species classes. Stratifying the classifier to pro-

duce a multi-stage classifier and supplementing the existing data-

base with additional specimens and species, notably G. lucii and G.

teuchis, will improve upon existing classification abilities for not

only G. salaris but for all the salmonid-infecting gyrodactylids.

For all of the foregoing descriptions of classification efficiency,

for the various methodologies employed in this study, it is impor-

tant to keep in mind that accuracy of identification is likely to be

higher in real-world diagnostic situations. For the current blind

study, all researchers were deliberately deprived of key informa-

tion that could improve identification probabilities. No host or cap-

ture location details were revealed nor was any indication

provided as to which worms came from the same hosts, host pop-

ulations or environments. Such information is a key to real-world

diagnoses, however this study provides evidence that accurate

identifications can be made even without such data. In addition, vi-

sual observation normally employs physical slides rather than pho-

tographs and it is likely that access to such specimens would also

improve researchers’ ability to discriminate and identify species.

Finally, use of real specimens would also improve measurements

for use in morphometric statistical classification and, given appro-

priately trained classifiers, it is likely that this group of techniques

would also produce more accurate results under field-test

conditions.

If the agreement between methods, with respect to the correct

identification of G. salaris, is considered, then the Fleiss’ Kappa va-

lue is 1 (z = 13.7, P < 0.001) suggesting perfect agreement between

the visualisation and molecular techniques. It is, however, worth

noting that this is, in some respects, an overestimate, since all of

the foregoing statistical tests do not include specimens for which

there is a missing value for one or more of the examination tech-

niques. The results can nevertheless be taken to indicate that, using

the best of currently available techniques, properly applied by

trained personnel, G. salaris can be reliably discriminated from

other species.

Whilst this study demonstrates that high classification efficien-

cies for G. salaris are possible using the three identification ap-

proaches, it highlights potential weaknesses in the current

methodologies and allows for realistic estimates of risk, e.g. losing

1466 A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467

Author's personal copy

or misidentifying specimens, to be made. These risks can then be

factored into field collection and specimen sampling protocols to

maximise the likelihood of G. salaris identification. In the absence

of supporting host data, this study allows for estimates of a single

G. salaris being correctly identified by each of the identification

methods. This is important when sampling populations that have

been recently infected/exposed to G. salaris as methods must be

sufficiently rigorous to detect a single G. salaris in a given sample.

For established infections of G. salaris, however, the implications of

loss of specimens and/or low levels of misclassification diminish as

this is compensated for by multiple representatives within a

sample.

Acknowledgements

The authors gratefully acknowledge the financial support pro-

vided by the diagnostic services of both Cefas Weymouth Labora-

tory, UK and the Marine Laboratory, Marine Science Scotland,

Aberdeen. The authors also thank the staff of the freshwater fisher-

ies services operating in each country notably Kevin Denham and

Richard Gardiner (Cefas Weymouth Laboratory), Helen Rowley

and Amanda Brown (Disease Surveillance & Investigation Dept.,

Veterinary Sciences Division, Agri-food & Biosciences Institute, Bel-

fast, Northern Ireland) for the provision of Gyrodactylus infected

salmonid and pike material for study.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in

the online version, at doi:10.1016/j.ijpara.2010.04.016.

References

Bakke, T.A., Cable, J., Harris, P.D., 2007. The biology of gyrodactylid monogeneans:the ‘‘Russian-doll killers”. Adv. Parasitol. 64, 161–378.

Bakke, T.A., MacKenzie, K., 1993. Comparative susceptibility of native Scottish andNorwegian stocks of Atlantic salmon, Salmo salar L., to Gyrodactylus salaris

Malmberg: laboratory experiments. Fish. Res. 17 (1–2), 69–85.Cameron, A., 2002. Survey toolbox for aquatic animal diseases: a practical manual

and software package. Australian Centre for International Agricultural Research,Canberra, Monograph No. 94, 375 pp.

Cunningham, C.O., 1997. Species variation within the internal transcribed spacer(ITS) region of Gyrodactylus (Monogenea; Gyrodactylidae) ribosomal RNA genes.J. Parasitol. 83, 215–219.

Cunningham, C.O., McGillivray, D.M., MacKenzie, K., Melvin, W.T., 1995.Discrimination between Gyrodactylus salaris, G. derjavini and G. truttae(Platyhelminthes: Monogenea) using restriction fragment lengthpolymorphisms and an oligonucleotide probe within the small subunitribosomal RNA gene. Parasitology 111, 87–94.

Fleiss, J.L., 1971. Measuring nominal scale agreement among many raters. Psychol.Bull. 76 (5), 378–382.

Gibson, D.I., 1993. Monobothrium wageneri: another imported tapewormestablished in wild British freshwater fishes? J. Fish Biol. 43, 281–285.

Hall, T.A., 1999. BioEdit: a user-friendly biological sequence alignment editor andanalysis program for Windows 95/98/NT. Nucleic Acids Symp. Ser. 41, 95–98.

Hansen, H., Bachmann, L., Bakke, T.A., 2003. Mitochondrial DNA variation ofGyrodactylus spp. (Monogenea, Gyrodactylidae) populations infecting Atlanticsalmon, grayling, and rainbow trout in Norway and Sweden. Int. J. Parasitol. 33,1471–1478.

Hansen, H., Bakke, T.A., Bachmann, L., 2007a. Mitochondrial haplotype diversity ofGyrodactylus thymalli (Platyhelminthes; Monogenea): extended geographicsampling in United Kingdom, Poland, and Norway reveals further lineages.Parasitol. Res. 100, 1389–1394.

Hansen, H., Bakke, T.A., Bachmann, L., 2007b. DNA taxonomy and barcoding ofmonogenean parasites: lessons from Gyrodactylus. Trends Parasitol. 23, 363–367.

Higgins, D., Thompson, J., Gibson, T., Thompson, J.D., Higgins, D.G., Gibson, T.J., 1994.CLUSTAL W: improving the sensitivity of progressive multiple sequence

alignment through sequence weighting, position-specific gap penalties andweight matrix choice. Nucleic Acids Res. 22, 4673–4680.

Jørgensen, T.R., Larsen, T.B., Jørgensen, L.G., Bresciani, J., Kania, P.W., Buchmann, K.,2007. Characterisation of a low pathogenic form of Gyrodactylus salaris fromrainbow trout. Dis. Aquat. Org. 73, 235–244.

Kania, P.W., Jørgensen, T.R., Buchmann, K., 2007. Differentiation between apathogenic and a non-pathogenic form of Gyrodactylus salaris using PCR-RFLP.J. Fish Dis. 30, 123–126.

Kay, J.W., Shinn, A.P., Sommerville, C., 1999. Towards an automated system for theidentification of notifiable pathogens. Parasitol. Today 15 (5), 201–206.

Kennedy, C.R., 1993. Introductions, spread and colonization of new localities by fishhelminth and crustacean parasites in the British Isles: a perspective andappraisal. J. Fish Biol. 43, 287–301.

Kimura, M., 1980. A simple method for estimating evolutionary rate of basesubstitutions through comparative studies of nucleotide sequences. J. Mol. Evol.16, 111–120.

Landis, J.R., Koch, G.G., 1977. The measurement of observer agreement forcategorical data. Biometrics 33, 159–174.

Lindenstrøm, T., Collins, C.M., Bresciani, J., Cunningham, C.O., Buchmann, K., 2003.Characterisation of a Gyrodactylus salaris variant: infection biology, morphologyand molecular genetics. Parasitology 127, 165–178.

MacKenzie, K., Bakke, T.A., 1994. Comparative susceptibility of native Scottish andNorwegian stocks of Atlantic salmon, Salmo salar L., to Gyrodactylus salaris

Malmberg: laboratory experiments. In: Pike, A.W., Lewis, J.W. (Eds.), ParasiticDiseases of Fish. Samara Publishing Ltd., Dyfed, pp. 57–58.

McHugh, E.S., Shinn, A.P., Kay, J.W., 2000. Discrimination of the notifiable pathogenGyrodactylus salaris from G. Thymalli using statistical classifiers applied tomorphometric data. Parasitology 121, 315–323.

Meinilä, M., Kuusela, J., Zietara, M.S., Lumme, J., 2002. Primers for amplifying�820 bp of highly polymorphic mitochondrial COI gene of Gyrodactylus salaris.Hereditas 137, 72–74.

Mo, T.A., 1994. Status of Gyrodactylus salaris problems and research in Norway. In:Pike, A.W., Lewis, J.W. (Eds.), Parasitic Diseases of Fish. Samara Publishing Ltd.,Dyfed, pp. 43–56.

OIE, 2009. Manual of diagnostic tests for aquatic animals 2009 (Chapter 2.3.3).Gyrodactylosis (Gyrodactylus salaris). Available from: <http://www.oie.int/eng/normes/fmanual/2.3.03_Gyrodactylosis.pdf>.

Olstad, K., Robertsen, G., Bachmann, L., Bakke, T.A., 2007. Variation in hostpreference within Gyrodactylus salaris (Monogenea): an experimentalapproach. Parasitology 134, 589–597.

Ossiander, F.J., Wedermeyer, G., 1973. Computer program for sample sizes requiredto determine disease incidence on fish populations. J. Fish. Res. Board Can. 30,1383–1384.

Paladini, G., Cable, J., Fioravanti, M.L., Faria, P.J., Di Cave, D., Shinn, A.P., 2009a.Gyrodactylus orecchiae n. sp. (Gyrodactylidae: Monogenea) from farmedpopulations of gilthead seabream (Sparus aurata L.) in the Adriatic Sea. FoliaParasitol. 56 (1), 21–28.

Paladini, G., Gustinelli, A., Fioravanti, M.L., Hansen, H., Shinn, A.P., 2009b. The firstreport of Gyrodactylus salarisMalmberg, 1957 (Platyhelminthes, Monogenea) onItalian cultured stocks of rainbow trout (Oncorhynchus mykiss Walbaum). Vet.Parasitol. 165, 290–297.

Platten, M., McLoughlin, M., Shinn, A.P., 1994. The distribution and identification ofgyrodactylid species in fish farms and rivers of Northern Ireland. Vet. Rec. 135,411–412.

Robertsen, G., Hansen, H., Bachmann, L., Bakke, T.A., 2007. Arctic charr (Salvelinusalpinus) is a suitable host for Gyrodactylus salaris (Monogenea, Gyrodactylidae)in Norway. Parasitology 134, 257–267.

Rokicka, M., Lumme, J., Zietara, M.S., 2007. Identification of Gyrodactylus

ectoparasites in Polish salmonid farms by PCR-RFLP of the nuclear ITSsegment of ribosomal DNA (Monogenea, Gyrodactylidae). Acta Parasitol. 52(3), 185–195.

Shinn, A.P., Hansen, H., Olstad, K., Bachmann, L., Bakke, T.A., 2004. The use ofmorphometric characters to discriminate species laboratory-reared and wildpopulations of Gyrodactylus salaris and G. Thymalli (Monogenea). Folia Parasitol.51, 239–252.

Shinn, A.P., Kay, J.W., Sommerville, C., 2000. The use of statistical classifiers for thediscrimination of species of the genus Gyrodactylus (Monogenea) parasitisingsalmonids. Parasitology 120 (3), 261–269.

Shinn, A.P., Sommerville, C., Gibson, D.I., 1995. Distribution, characterization ofspecies of Gyrodactylus Nordmann, 1832 (Monogenea) parasitizing salmonids inthe UK, their discrimination from G. salaris Malmberg, 1957. J. Nat. Hist. 29,1383–1402.

Tamura, K., Dudley, J., Nei, M., Kumar, S., 2007. MEGA4: Molecular EvolutionaryGenetics Analysis (MEGA) software version 4.0.. Mol. Biol. Evol. 24, 1596–1599.

Venables, W.N., Ripley, B.D., 1997. Modern Applied Statistics with S-Plus, second ed.Springer-Verlag, Berlin.

Yeomans, W.E., Chubb, J.C., Sweeting, R.A., 1997. Khawia sinensis (Cestoda:Caryophyllidea) – indicator of legislative failure to protect freshwater habitatsin the British Isles? J. Fish Biol. 51, 880–885.

A.P. Shinn et al. / International Journal for Parasitology 40 (2010) 1455–1467 1467

Supplement to Shinn et al. (2010) International Journal of Parasitology, 40, 1455-1467

Fig. S1. A schematic diagram showing the number of Gyrodactylus specimens passing into the analysis pipeline, the time

taken for each diagnostic methodology to be completed and the sample proportion (SP) making it through to the next

analytical step. 1The classification results of two groups of gyrodactylid experts. 2Total loss of samples by one molecular

laboratory is thought to be an informative but an atypical event hence further calculations are made assuming that losses

in the molecular process will reflect those found in molecular laboratory 1. 3Laboratory 1 was sent 221 tubes; 11 of these

were found to be empty and, therefore, subsequent calculations are based on 210 specimens. Losses, therefore, are

calculated on the 221 specimens and a proportion of those lost in the steps prior to the application of the molecular

methods. 4SP value takes account of the combined losses incurred from both laboratories.

Fig. S2. Decision tree for the processing of Gyrodactylus specimens during periods of routine surveillance and low

sample volume. Time estimates for handling a batch of 96 specimens (i.e. a full PCR run) are provided. 1This step and

time estimation includes the proteolytic digestion of the haptor. If haptors are air dried only, then the time could be cut to

~250 min.

Fig. S3. Decision tree for the processing of Gyrodactylus specimens during periods of a suspected outbreak when

diagnostic laboratories might be expected to process a large number of samples.