Rapporto SCAHAW sulla Brucella melitensis



Bichat guidelines

Bichat guidelines for the clinical management of brucellosis and bioterrorism-related brucellosis


EFSA's First Community Summary Report 2004

EFSA's First Community Summary Report on Trends and Sources of Zoonoses, Zoonotic Agents and Antimicrobial resistance in the European Union in 2004 - Brucella


Validation of FPA and cELISA for the detection of antibodies to Brucella abortus in cattle sera and comparison to SAT, CFT, and iELISA

J.A. McGiven*, J.D. Tucker, L.L. Perrett, J.A. Stack, S.D. Brew, A.P. MacMillan

rological techniques are the mainstay of diagnosis

and mass testing programmes. The most successful

of the serological diagnostic tests for
B. abortus, B. melitensis, and B. suis are based on the detection of

antibodies to the LPS antigen of smooth Brucella

strains. The immunodominant epitope of the LPS is

the O-chain which is a homopolymer of 1,2-linked

4-amino-4, 6-dideoxy-a-D-mannopyranosylresidues

(Caroff et al., 1984). Traditional and welldocumented

techniques for serological diagnosis

include the Rose Bengal plate agglutination test

(RBPAT), serum agglutination test (SAT), complement

fixation test (CFT), and, more recently, the indirect

ELISA (iELISA) and competitive ELISA (cELISA)

being put into more regular use. The fluorescence

polarisation assay (FPA)
(Nielsen et al., 1996) has a

shorter history of use and has yet to become established

within the routine testing procedures of most

National Brucellosis Reference Laboratories.

Fluorescence polarisation measures the excitation

by plane polarised light of a fluorescent molecule
(Perrin,1926). Measurement of returned photons in the

planes parallel and perpendicular to the excitation

plane allows for the assessment of the rotation of the

fluorophore. Other factors being constant, then the rate

of rotation of this molecule is inversely proportional to

its size
(Nasir and Jolley, 1999). Thus, the rotation of a

fluorescent molecule (fluorophore) conjugated to, in

this case, Brucella O-chain, will slow if bound by anti-

Brucella LPS antibodies. The FPA is rapid and requires

no solid phase bound reagent or removal of excess

reagents. It is host species-independent and can also be

conducted on whole blood
(Nielsen et al., 2001a) andmilk (Nielsen et al., 2001b). High throughput serological

testing for brucellosis by FPA has been enabled by

the development of 96-well plate readers. The use of

this equipment for FPA has not previously been validated.

The FPA shows great potential as a diagnostic test

due to its ease of use and potentially wide application.

However, its performance must withstand close scrutiny

against the more conventional tests, and the results

of such examinations are described below. The bovine

populations tested in these validation studies were

derived from Northern Europe, including Germany,

The Netherlands, Eire, and Britain. The diagnostic

sensitivity (DSn) and diagnostic specificity (DSp) data

can be extrapolated to a target population of such


2. Materials and methods

2.1. FPA method

The FPA was conducted using a 96-well microtitre

plate format. Test buffer was prepared by the addition

of 0.836 g of sodium monophosphate monohydrate

2PO4H2O), 1.49 g of sodium triphosphate

dodecahydrate (Na3PO412H2O), 9 g of sodium chloride(NaCl), and 0.5 g of lauryl sulphate (C


per litre of distilled water, with a final pH of 7.5. The

first step of the test required the addition of 180
Al ofbuffer to 20 Al of test serum within each test well.

Control samples were added to each plate, a minimum

of 4 wells for the negative control [equilibrated to the

Office International Des Epizooties (OIE) ELISA

International Standard anti-
Brucella Serum: negative]

and also 4 high positive control wells. Buffer and test

sample were mixed during addition by repeated pipette

action, and the test plates were incubated for 2 min

at room temperature (18–24
jC) on a rotary shaker

(125 rpm).

After initial incubation, the test plate was read on a

Tecan Polarion Fluorescence Polarisation microplate

reader (reader set to gain of 115, number of flashes per

well at 20, and filters of 485 nm for excitation and 535

nm for emission) to obtain a background reading, in

J.A. McGiven et al. / Journal of Immunological Methods 172 278 (2003) 171–178

relative fluorescent units (RFUs), for each sample.

Subsequently, 10
Al of antigen (Brucella O-polysaccharide

conjugated isothiocyanate fluorophore: supplied

by Diachemix, Whitefish Bay, WI, USA) was

added to each well and mixed by pipette action

followed by a further 2-min incubation as above.

The plate was read again as before to obtain the raw

parallel and perpendicular data for each sample. This

data was converted to millipolarisation units (mP) by

the formula: mP=((IvIh)/(Iv + Ih))
1000, where

Iv = the intensity of parallel light and Ih = intensity of

perpendicular light (Perrin, 1926).

2.2. SAT, CFT, and ELISA methods

SAT, CFT, and iELISA were standardised and conducted

according to the OIE Manual of Standards for

Diagnostic Tests and Vaccines
(Corbel and MacMillan,1996). Additionally, the iELISA conforms to the standards

set out in the annex of the current European

Directive 64/432 in that it detects a 1/16 dilution

(further diluted by 1/200 as per test protocol) of the

OIE ELISA International Standard anti-Brucella

serum: strong positive, while finding the negative

serum negative. All iELISA reagents are produced at

the VLA, Weybridge, UK. The cELISA is conducted

using a monoclonal antibody specific to the O-chain

polysaccharide portion of
Brucella LPS (Stack et al.,

1999) as recommended by the OIE Manual.

2.3. Serum samples

For the validation of the FPA, 1947 blood samples

Brucella negative animals (by virtue of their presence

in Britain: officially brucellosis-free since 1985)

collected from August to October 2001 were tested.

Due to the difficulty of acquisition, only 146 serum

samples from
Brucella culture positive (B. abortus)

animals were tested. These samples came from Germany,

France, Eire, and from Britain pre-1985. All

serum samples from animals confirmed positive by

culture (herein described as ‘positive reference serum’)

were tested by CFT, SAT, iELISA, cELISA, and FPA.

A further 1440 negative samples (from Britain in 2002)

were tested by cELISA for purposes of validation. A

total of 6957 negative sera from Britain, collected

during August to October 2001, was used to assess

the performance of the iELISA.

2.4. Test validation

The most appropriate cutoff was selected for the

FPA and cELISA by using a two-way receiver operating

characteristic (TW-ROC) analysis
(Greiner et al.,1995) that plots the DSn (true positives/true positives

+ false negatives) and DSp (true negatives/true

negatives + false positives) as a function of cutoff. This

is often the test value which produces the highest

combined DSn and DSp, as has been chosen here.

Fig. 1. TG-ROC analysis of FPA results from 1947 Brucella uninfected cattle serum samples and 146 serum samples from Brucella culture

positive animals. Solid line shows DSn; broken line shows DSp.

J.A. McGiven et al. / Journal of Immunological Methods 278 (2003) 171–178 173

The intersection of the DSn and DSp line is the test

value where the two parameters are equal; however,

this is rarely the same value as the point where the

combined values are highest. Confidence intervals for

DSn and DSp can be determined
(Jacobson, 1998) to

show the range within which the parameters are likely

to be when the test is used in the target population.

Further statistical analysis was conducted on the

results for the independent samples (negative reference

sera) using two sample
t-tests, and on the paired samples

(positive reference sera) using a two-way binomial

test, to look for differences in DSp and DSn,


As a field trial, 259 samples from Irish herds

infected with
B. abortus were tested by FPA, iELISA,

and cELISA to compare the relative performance of

each of the tests. Analysis of theses samples was performed

by determining the percentage of results that

were in agreement between tests and testing those in

disagreement by McNemars test for paired data (with

Yates’ correction) to investigate significant differences.

3. Results

To assess the data from the FPA method, it was

necessary to obtain results that were as precise as possible.

To ensure high-quality results, test plates were

accepted and rejected on the basis of the absolute

values of the control serum used and the degree of

variation between multiple wells of the same control


A cutoff for the FPA was evaluated by subtracting

the average plate negative control result from each test

sample result and by analysing this data from known

noninfected cattle sera and positive reference sera

using TW-ROC analysis
(Fig. 1). The cutoff which

provided the highest combined DSn and DSp values

was chosen. This was a value of 15.5 mP above the

negative control.

The DSp data for the cELISAwas also evaluated by

TG-ROC analysis (Fig. 2). The cutoff that maximised

Fig. 2. TG-ROC analysis of cELISA results from 1440 Brucella uninfected cattle serum samples and 146 serum samples from Brucella culture

positive animals. Solid line shows DSn; broken line shows DSp.

Table 1

Diagnostic specificity and sensitivity values for each method


Test cutoff 20 IUs 30 IUs 10% 70% 15.5mP

DSp (%) 99.9










Total no. of


a 995a 6957 1440 1947

DSn (%) 91.8










Total no. of


146 146 146 146 146

DSp + DSn 191.7










Values in parentheses indicate 95% confidence interval.

a Data from Emmerzaal et al. (2002).

J.A. McGiven et al. / Journal of Immunological Methods 174 278 (2003) 171–178

the combined DSp and DSn was at 70% of the

conjugate control (positive sera give lower values).

Table 1 shows the number of samples tested by each

method, the cutoff for each method, and the DSn and

DSp for each—including the 95% confidence intervals.

DSp data for the SAT and CFT are from a recent

validation trial of an iELISA to be used in the Netherlands

(Emmerzaal et al., 2002) where the CFTand SAT

method conformed to the European Union (EU) (directive

64/432) and OIE requirements as is also the case

in Britain. The DSp values for the iELISA method are

calculated from British surveillance data where the

British herd is screened by this method before any

positive samples are subsequently tested by SAT and

CFT for confirmation. This table clearly shows that the

performance of the SAT is comparatively substandard,

while the CFT is also inferior compared with the

ELISAs and the FPA. The results for the iELISA,

cELISA, and FPA showed the FPA to have the highest

combined DSp and DSn. The confidence limits in

Table 1 for the DSp + DSn revealed some overlap for

all tests but the SAT. Testing for significant differences

between these results is restricted by the combination

of data from independent (negative reference sera) and

dependent (positive reference sera) samples. Testing

for significant differences (
P < 0.05) between the DSp

for each test, used with the cutoff as described in Table1

, revealed the following magnitude of DSp: (CFT =

cELISA)>(FPA= SAT)>iELISA. Significant differences

between the tests, based on the results from the

positive reference sera alone and using a binomial

distribution (
P= 0.5), were found as described in Table2. These results show that the iELISA and FPA are

significantly more sensitive than the CFT and SAT, but

not the cELISA. The SAT was significantly less

sensitive than all other tests.

The Pearson product-moment correlation coefficient

(Fig. 3) of the continuous FPA and iELISA

results from the positive reference sera shows a highly

significant (
P < 0.01, n = 146) positive correlation(r = 0.683). Of the 146 positive serum samples, 140

tested positive for both tests, 3 were negative for both,

with 2 iELISA positive but FPA negative and only 1

iELISA negative but FPA positive.

Fig. 3. FPA test results against iELISA results (expressed as percentage of positive control) for the 146 serum samples from culture positive


Table 2

Comparison of differences between test diagnostic sensitivities

P values


SAT 0.0007* 0.0000* 0.0000* 0.0000*

CFT 0.0625 0.0156* 0.0078*

cELISA 0.5000 0.2500

FPA 1.0000


Results from each test for 146 positive reference sera were compared.

The distribution of the samples differentially diagnosed when two

tests were compared was analysed using a binomial distribution. The

null hypothesis states that a sample differentially diagnosed by two

tests is equally likely to be either positive for test A and negative for

test B, or negative for test A and positive for test B ( P= 0.5). The P

values in the table show the probability (to 4 decimal places) of the

observed test results occurring through chance if the null hypothesis

is correct.

* Results are significantly different ( P < 0.05).

J.A. McGiven et al. / Journal of Immunological Methods 278 (2003) 171–178 175

The samples from infected herds in Eire were tested

by iELISA, cELISA, and FPA. The results (Table 3)

show that the iELISA method finds the highest number

of samples positive, with the cELISA finding the least

number of samples positive. When the tests were

compared to each other, the results for each sample

were in agreement over 90% of occasions. Although

the cELISA and the iELISA showed the highest degree

of agreement, the samples in disagreement were heavily

weighted towards iELISA positive and cELISA

negative results. This split was less profound in the

other two test comparisons. McNemar’s test with

paired samples (with Yates’ correction) reflects this

and revealed that there was no significant difference in

the proportion of positive samples between the iELISA

and FPA, or the FPA and cELISA methods (

However, there was a significant difference between

the iELISA and cELISA results (P < 0.05).

4. Discussion

Diagnostic procedures for brucellosis should be

specific, sensitive, and detect all stages of the infection.

Currently, no such test exists. Yet, serological testing

and notification in cases of abortion with subsequent

serological and cultural examination, with an identification

system in accordance with EU regulations, are

the main requirements to be fulfilled in order to retain

official brucellosis-free status as a Member State.

The current British brucellosis surveillance strategy

uses the described iELISA method to screen nondairy

herds (dairy herds are monitored by the bulk milk

iELISA which is a minor adaptation of the serum

iELISA used). ELISAs are easy to perform, to automate,

produce objective results, rapidly lend themselves

to Quality Assurance programmes, and can cut

the cost of staff training due to the universality of the

ELISA method for different applications. Samples that

are positive by iELISA are subjected to further testing

by both CFT and SAT for confirmation.

The SAT is recognised as inferior to other tests

(Corbel and MacMillan, 1996). The CFT is recognised

as a highly sensitive and specific test when correctly

performed, yet its comparative underperformance in

this study questions these properties. The CFT also has

many practical drawbacks, not least of all which is its

relative technical complexity. The test is also subject to

anticomplementary reactions and will not work on

haemolysed samples. Furthermore, prozone reactions

caused by competition between C
Vactivating and non-

CVactivating IgG isotypes (McGuire et al., 1979), andpossibly IgM

(Nielsen and Duncan, 1988), require

serum to be tested at multiple dilutions. Yet, it must

be recognised that these problems did not prevent the

successful eradication of brucellosis (as defined by EU

directive 64/432) from many countries using these

tests (e.g. Britain and The Netherlands).

The cutoff of the FPA was determined using TGROC.

The cutoff value of 15.5 mP was selected to

maximise total DSp and DSp. The cELISA, although a

more established method, was also evaluated by TGROC

to obtain a cutoff of 70% of the conjugate control.

This maximised the total DSp and DSn. Data was not

available to perform the same analysis on the SAT,

CFT, and iELISA. As these methods have strictly

defined cutoff criteria, this method of evaluation was

not necessary. Although cutoffs can be chosen to fall at

the test value that provides the highest combined DSn

and DSp, in practice, other factors also need to be

considered when selecting a cutoff. The relative costs

of false negative and false positive samples should be

evaluated. If it will be more costly to get false negatives,

then the test cutoff can reflect this by being

moved in favour of increasing sensitivity. This might

be the case in serious conditions where the false

positives can be screened using a second, possibly

more expensive test, but the consequences of missing

a true positive will be severe.

Table 1 shows the DSp and DSn results for each

test. For DSp + DSn, the results are in the following

order of magnitude: FPA>iELISA>cELISA>CFT>

Table 3

Test results from
B. abortus-infected herds from Eire


Number positive 166 168 158

% Pos. 64.09% 64.86% 61.00%

Number negative 93 91 101

% Neg. 35.91% 35.14% 39.00%

Test comparison FPA vs.


iELISA vs.


cELISA vs.


Result agreement 90.7% 93.1% 91.5%

P>0.1 P < 0.05 P>0.1Significance: P < 0.05 = tests are significantly different at the 95%

confidence level.

*McNemars paired test With Yates’ correction.

J.A. McGiven et al. / Journal of Immunological Methods 176 278 (2003) 171–178

SAT. The conjecture that these relative performances

can be extrapolated to the target population is difficult

to prove statistically where the confidence limits for

the result overlap those for other tests. This is because

the DSn + DSp results are from both dependent samples

(positive reference sera) and independent samples

(negative reference sera). At the test cutoffs shown in

Table 1, statistical analysis relating to the DSp alone

shows the following significant differences: (CFT =

cELISA)>(FPA= SAT)>iELISA. The results from the

146 positive reference samples for each of the tests

were analysed using a two-tailed binomial test. The

results from two tests were compared, and only the

samples which were diagnosed differently in each test

were considered. The null hypothesis was that these

differentially diagnosed samples had an equal chance

P= 0.5) of being positive for test A and negative for

test B, or negative for test A and positive for test B. A

significant rejection of this hypothesis, that
P p 0,

concludes that one test is more sensitive at detecting

positives than the other. These results
(Table 2) show

that the iELISA and FPA are significantly more sensitive

than the CFT and SAT, but not the cELISA. The

SAT is significantly less sensitive than all other tests.

The sampling methods for validation trials are of the

utmost importance. Reliable confidence intervals for

the test parameters DSn and DSp can only be achieved

when the samples are selected at random from a

selection of the population which has a proportionate

representation (to the potential target population) of all

parameters that may affect test outcome, for example,

age of animal, sex, diet, breed, location, accommodation,

etc. Serum samples from animals confirmed as

Brucella positive by cultural identification are in short

supply due to the nature of the disease and its distribution.

Therefore, it is very difficult to select for such

factors and obtain sufficient samples. However, a

relatively large number of positive animals were

sampled in this study. In previous validation studies

of the FPA, despite the number of samples being

indicated, it has not always been clear how many

animals were sampled to obtain the positive reference

sera set. As such, the confidence intervals are included

to serve as only a guide to the uncertainty of the results,

but in this instance have added validity over previous


Of equal, if not more, importance than the absolute

results is the relation of the tests to one another in

terms of their diagnostic performance. It should also be

noted that the ‘Gold Standard’ of positive culture data

is only a gold standard specificity. Far less is known

about the sensitivity of culture. As a result, even the

most stratified culture positive serum samples may be

biased towards overstating a tests DSn, as it is by

serology that most animals are identified and subsequently

slaughtered for cultural examination. Although

the same negative samples were not used to

determine DSp for each of the tests, the selection was

far larger than for the culture positive samples. As a

result, the confidence intervals are smaller, yet these

too must be interpreted with a degree of caution due to

potential sampling bias such as the time of year the

samples were collected.

There is a significant positive correlation between

the continuous data from the iELISA and FPA results

on the culture positive serum samples. Similarly, it was

possible to discriminate between samples of high titre

and lower titre positives. The strong correlation was

also evident in the data from the tests on samples from

Brucella-infected herds where the FPA and the

iELISA results were insufficiently different to be

significant. The data from the cELISA test was sufficiently

different from the iELISA to be able to say with

95% confidence that the two tests possessed different

diagnostic properties. Without knowing the infection

status of each animal sampled, it was not possible to

say if each test was more sensitive or less specific, but

judging from the validation data, the pattern of the

field trial data was likely to be a result of both.

The data presented here suggests that the FPA is a

significantly superior diagnostic test to the SAT, but

despite having the highest combined DSp and DSn

total, it was not conclusively superior in sensitivity and

specificity over the CFT or the ELISAs. The advantages

of the FPA in terms of the simplicity of the

method compared to the other tests are clear. As such,

further validation work on the FPA is warranted in

order to evaluate its performance against known false

positive reactors and during different stages of the


It may not be possible to get real DSp and DSn

values due to the bias of culture positive serum samples

used as a ‘Gold Standard’, unless validation is

performed using methods of analysis such as maximum

(Hui and Walter, 1980). This method

requires at least two diagnostic tests which are condi-

J.A. McGiven et al. / Journal of Immunological Methods 278 (2003) 171–178 177

tionally independent such as one of the described

serological tests and a diagnostic PCR. A satisfactory

diagnostic PCR method, one that does not require preculture,

is a current research goal
(Nakkas et al., 2002).

Maximum likelihood analysis would be an ideal method

to validate such a test. These problems are true for

all diagnostic tests as well as for the FPA. It has also

been claimed that the FPA can discriminate between

serum from vaccinated and nonvaccinated animals

(Neilsen et al., 1996), and further investigation here

is also warranted. It is clear that the methodology of

the FPA offers clear advantages due to its ease of use.

Full implementation and acceptance of FPA methods

for the diagnosis of brucellosis will likely necessitate

the use of an International Standard Serum panel containing

at least a low titre positive sample and a

negative. This would be similar to those used to standardise

the ELISA which allowed it to be formally

introduced into EU legislation.


The authors would like to thank Kevin Kenny (Key

Laboratories, Dublin, Ireland) for his technical expertise

and for his help in the preparation of this



Caroff, M., Bundle, D.R., Perry, M.B., Cherwonogrodzky, J.W.,

Duncan, J.R., 1984. Antigenic S-type lipopolysaccharide of
Brucellaabortus 1119-3. Infect. Immun. 46, 384.

Corbel, M.J., MacMillan, A.P., 1996. Bovine brucellosis. In: Reichard,

R. (Ed.), Office International Des Epizooties Manual of

Standards for Diagnostic Tests and Vaccines, Paris. Office International

Des Epizooties, 12 rue de Prony, 75017 Paris,

France, p. 248.

Dalrymple-Champneys, W., 1960.
Brucella Infection and Undulant

fever in Man, Oxford Univ. Press, London.

Emmerzaal, A., de Wit, J.J., Dijkstra, Th., Bakker, D., van Zijderveld,

F.G., 2002. The Dutch
Brucella abortus monitoring programme

for cattle: the impact of false-positive serological

reactions and comparison of serological tests. Vet. Q. 24, 40.

Fekete, A., Bantle, J.A., Halling, S.M., Sanborn, M.R., 1990. Preliminary

development of a diagnostic test for
Brucella using

polymerase chain reaction. J. Appl. Bacteriol. 69, 216.

Greiner, A., Dorit, S., Go¨bel, P., 1995. A modified ROC analysis

for the selection of cut-off values and the definition of intermediate

results of serodiagnostic tests. J. Immunol. Methods

185, 123.

Hui, S.L., Walter, S.D., 1980. Estimating the error rates of diagnostic

tests. Biometrics 36, 167.

Jacobson, R.H., 1998. Validation of serological assays for diagnosis

of infectious diseases. Rev. Sci. Tech.-Off. Int. Epizoot. 17, 469.

Manthei, C., Carter, R., 1950. Persistence of
Brucella abortus infection

in cattle. Am. J. Vet. Res. 11, 173– 180.

McGuire, T.C., Musoke, A.J., Kurtti, T., 1979. Functional properties

of bovine IgG1 and IgG2: interaction with complement,

macrophages, neutrophils and skin. Immunology 38, 249.

Morgan, W.J.B., 1977. The national brucellosis programme of Britain.

In: Crawford, R.P., Hidalgo, R.J. (Eds.), Brucellosis: an

International Symposium. A and M Univ. Press, Austin, TX,

p. 378.

Nakkas, A.F.,Wright, S.G., Mustafa, A.S.,Wilson, S., 2002. Singletube,

nested PCR for the diagnosis of human brucellosis in Kuwait.

Ann. Trop. Med. Parasitol. 96, 397.

Nasir, M.S., Jolley, E., 1999. Fluorescence polarisation: an analytical

tool for immunoassay and drug discovery. Comb. Chem.

High Throughput Screen. 2, 177.

Nielsen, K., Duncan, J.R., 1988. Further evidence that bovine IgM

does not fix guinea pig complement. Vet. Immunol. Immunopathol.

19, 197.

Nielsen, K., Gall, D., Jolley, M., Leishman, G., Balsevicius, S.,

Smith, P., Nicoletti, P., Thomas, F., 1996. A homogeneous fluorescence

polarisation assay for detection of antibody to
Brucellaabortus. J. Immunol. Methods 195, 161.

Nielsen, K., Gall, D., Smith, P., Kelly, W., Yeo, J., Kenny, K.,

Heneghan, T., McNamara, S., Maher, P., O’Connor, J., Walsh,

B., Carrol, J., Rojas, X., Rojas, F., Perez, B., Wulff, O., Buffoni,

L., Salustio, E., Gregoret, R., Samartino, L., Dajer, A., Luna-

Martinez, E., 2001a. Fluorescence polarisation assay for the

diagnosis of bovine brucellosis: adaptation to field use. Vet.

Microbiol. 80, 163.

Nielsen, K., Smith, P., Gall, D., Perez, B., Samartino, L., Nicoletti,

P., Dajer, A., Rojas, X., Keyy, W., 2001b. Validation of the

fluorescence polarisation assay for detection of milk antibody

Brucella abortus. J. Immunoassay. Immunochem. 22, 203.

Perrin, M.F., 1926. Polarization de la lumiere de fluorescence. Vie

moyenne de molcules dans l’etat excite. J. Phys. Radium 7, 390.

Stack, J.A., Perrett, L.L., Brew, S.D., MacMillan, A.P., 1999. Competitive

ELISA for bovine brucellosis suitable for testing poor

quality samples. Vet. Rec. 145, 735.

J.A. McGiven et al. / Journal of Immunological Methods 178 278 (2003) 171–178