You are on page 1of 5

Detection of antibiotics in muscle tissue with microbiological

inhibition tests: effects of the matrix

Lieve Okerman,* Katia De Wasch and Jan Van Hoof
Department of Veterinary Food Inspection, Faculty of Veterinary Medicine, University of
Ghent, Salisburylaan 133, B-9820 Merelbeke, Belgium
Received 26th June 1998, Accepted 26th August 1998

The effects of the tissue matrix on detection limits of antibiotics with microbiological inhibition tests, intended for
muscle tissue, were measured. Pieces of frozen meat were laid directly on top of paper disks impregnated with
aqueous antibiotic solutions. Inhibition zones were compared with those obtained by the same standard solution
without tissue. Only tetracyclines were detected as efficiently with as without muscle tissue. Inhibition zones of
the beta-lactam antibiotics ampicillin and penicillin G, and the fluoroquinolone antibiotics enrofloxacin and
ciprofloxacin were smaller when muscle tissue was added to low levels of standard solution. At higher levels the
differences were not substantial. Inhibition zones of tylosin were smaller and irregular or had disappeared
completely, while ceftiofur, sulfadimidine, erythromycin, lincomycin, and streptomycin were not detected in
spiked muscle tissue at concentrations fivefold higher than the detection limits without tissue. These results
indicate that ceftiofur, sulfonamides, streptomycin and some macrolide antibiotics cannot be detected in intact
meat with the plates and bacterial strains prescribed in the European Four Plate Test, a test which was initially
intended as a multi-residue method for muscle tissue. Two plates of this system are not suitable for screening
purposes; a third one detects tetracyclines and beta-lactam antibiotics in spiked tissue; the fourth one is sensitive
for beta-lactam antibiotics and for some but not all macrolides. Samples spiked with the fluoroquinolones
enrofloxacin and ciprofloxacin can be detected with an additional plate, not included in the Four Plate Test.

Microbiological inhibition tests are considered as multi-residue
screening tests for antibiotics in milk, meat or other animal
tissues. Several methods have been described to investigate
disks of frozen tissue, which are laid directly on one or more
agar media, each seeded with a susceptible bacterial strain.15
An inhibition test method is useful for detection of an
antibiotic or a group of antibiotics, if the detection limits of
these antibiotics are below safe levels or maximal residue limits
(MRL). With agar diffusion methods used for animal tissues,
such as the Four Plate Test (FPT), detection limits of antibiotics
have most often been determined using aqueous solutions of
analytical standards.1,4 However, residues are detected directly
in undiluted meat with the FPT and other comparable tests, and
effects of this matrix on detection limits never have been
Residues of substances other than antibacterials are nowadays most often detected with immunological or chemical
methods. To determine the recovery of the analyte or analytes,
a spiked sample is prepared by adding a known amount of
analytical standard to the sample before it is mixed with
extraction fluid. The extraction is then followed by a clean-up
procedure and a detection step. Inhibition tests such as the FPT
detect residues in intact meat, without any extraction or cleanup procedures; determination of recoveries is not necessary.
During routine testing, spiked samples are not prepared and this
is another reason why matrix effects have never been measured.
The composition and the properties of the medium, used in a
microbiological inhibition test, influence the detection limits of
antibiotics.3,57 For example, the pH of the test medium is an
important factor influencing the detection limits of most
antibiotics,3,5 and inhibitory zones produced by 500 ng sulfadimidin may vary from 0 to more than 10 mm, depending on the
origin of the peptone in the medium.7 It is therefore possible that

tissue components, such as proteins, change the composition of

the medium and influence the inhibitory zone produced by an
antibiotic residue present in the sample.
In the present paper, we describe a method to investigate
possible influences of animal tissues on the detection limits of
commonly used antibiotics. A piece of blank chicken, pork or
beef meat was applied directly on top of a paper disk
impregnated with an antibiotic standard solution. Inhibition
zones of standards with meat and standards without meat were
compared. The significance of our findings to routine antibiotic
residue screening is discussed.

Materials and methods

The FPT described in the Manual of Reference Materials and
Methods to Detect Veterinary Drug Residues, is intended to
detect residues of beta-lactam antibiotics, tetracyclines, sulfonamides, aminoglycosides and macrolides in muscle tissue of
slaughter animals.1 A fifth plate, with low detection limits of
quinolones and fluoroquinolones,2,5 was tested with three
antibiotics belonging to this group.

1. Media used for maintainance of Escherichia coli and

Micrococcus luteus strains and for preparation of inocula
Tryptone soya agar (TSA) (CM131, Oxoid, Basingstoke,
England) and Tryptone soya broth (TSB) (Oxoid CM 129) were
prepared and autoclaved as indicated by the manufacturers.
Mist. desiccans (MD) was used as a medium intended for
conservation of bacterial strains at 220 C and was prepared as
follows: 6 g of dextrose was added to 20 ml Brain Heart
Infusion Broth (Difco 0037-17-08; Detroit, USA), prepared
according to manufacturers instructions but not autoclaved.
Analyst, 1998, 123, 23612365


This solution was sterilized by filtration and added to 60 ml

inactivated horse serum. The medium was kept frozen for
maximum 12 months before use.

2. Bacterial suspensions
Bacillus subtilis spore suspension: Merck (Darmstadt, Germany) No. 10649 is a ready-to-use suspension.
M. luteus bacterial suspension: the ATCC9341 strain was
prepared as described in the Manual of Reference Materials and
Methods to Detect Veterinary Drug Residues,1 but MD was
used instead of culture broth to maintain the stock inoculum. A
few colonies were suspended in 0.5 ml MD in sterile Eppendorf
tubes; these tubes were kept frozen until needed.
E. coli suspension: A freeze-dried strain of E. coli
ATCC11303 was reconstituted and inoculated onto TSA in a
Petri dish. The plate was incubated for 24 h at 37 C and
inspected for purity. Sterile Eppendorf tubes with 0.5 ml MD
were inoculated with several colonies of the E. coli strain. This
stock inoculum was kept at 220 C for maximally two years.
When needed, the stock inoculum was thawed and inoculated
onto a TSA plate. After overnight incubation at 37 C the plate
was inspected for purity. Ten ml of TSB were inoculated with
several colonies obtained on the TSA plate and incubated
overnight. The TSB culture, which contained at least 5 3 108
colony forming units of the E. coli strain per ml, was diluted 1
+ 9 in sterile TSB and added to the prepared medium cooled to
45-50 C.

4. Influence of the matrix

Three plates, from one batch of medium, were used for each
concentration tested. Four paper disks (diameter 6 mm) were
laid on each plate, at a distance of 10 mm from the edge of the
plate, and 0.01 ml of an appropriate standard solution was
added. As soon as possible, and always within 15 min, pieces of
frozen chicken, pork and beef muscle tissue were each laid upon
two paper disks, on opposite sides of the plate. Each meat
species was tested twice on one plate while the other two paper
disks on the plate served as controls. The muscle tissue had been
bored with a cork borer of 8 mm diameter and cut into disks
approximately 2 mm thick, as prescribed for the FPT. The
volume of such tissue disks is 0.4 3 0.4 3 3.14 3 0.2 = 0.1
mm3 and the weight can be estimated about 0.1 g. Plates
containing media I, II, III and V were incubated overnight at
30 C, while plates with medium IV were incubated for 24 h at
37 C. Diameters of inhibition zones of standard with muscle
tissue were measured and compared with zones observed
around standard disks without tissue.

All results are summarized in Table 1. Six observations were
obtained with each concentration of antibiotic without meat,
and six with meat. No obvious differences were seen between
the different meat species (data not shown in the table). The
ranges of zones obtained with aqueous solutions of antibiotics
and the ranges of zones obtained with meat spiked with the same
level can be found in Table 1.

3. Media for residue testing

1. Beta-lactam antibiotics

Test agar pH 6 (Merck; dehydrated medium 10 663), test agar

pH 7.2 (Merck; dehydrated medium 15 787) and test agar pH 8
(Merck; dehydrated medium 10 664) were prepared and
autoclaved. After cooling to 4550 C, the bacterial suspension
and the supplement (if necessary) were added.
Five different inoculated media were used for antibiotic
detection: medium I, test agar pH 6, seeded with B. subtilis;
medium II, test agar pH 7.2, with trimethoprim (Sigma, St.
Louis, MO, USA; no. T-7883) added to a final concentration of
0.05 mg l21, and seeded with B. subtilis; medium III, test agar
pH 8, seeded with B. subtilis; medium IV, test agar pH 8, seeded
with M. luteus; medium V, test agar pH 6, seeded with E. coli
(0.1 ml of the diluted suspension).
Sterile Petri dishes (diameter 90 mm) were filled with 5 ml of
the prepared and seeded media.
Antibiotic standards were all purchased from Sigma, with the
exception of enrofloxacin, ciprofloxacin and ceftiofur, which
were kindly provided by the pharmaceutical companies Bayer
(Leverkusen, Germany) and Pharmacia and Upjohn (Puurs,
Belgium). Detection limits of antibiotics frequently used in
veterinary medicine were determined as follows: two-fold
dilutions were prepared and 0.01 ml of these dilutions were
applied to 6 mm diameter paper disks. Each dilution was tested
four times on the medium considered to be most sensitive for the
antibiotic tested. Concentrations producing a 12 mm diameter
inhibition zone were calculated. Detection limits were as
follows: on medium I: penicillin G (sodium salt), 0.4 ng;
ceftiofur (sodium salt), 9 ng; Ampicillin trihydrate, 3 ng;
oxytetracycline, 8 ng; tetracycline, 5 ng; chlortetracycline,
0.5 ng; doxycycline, 1 ng; on medium II: sulfadimidin, approx.
50 ng; on medium III: streptomycin sulfate, 20 ng; on medium
IV: erythromycin, 1 ng; tylosin, 10 ng; lincomycin, 20 ng;
penicillin G, 0.5 ng; ampicillin, 0.8 ng; ceftiofur, 2 ng; on
medium V: enrofloxacin, 2 ng; ciprofloxacin, 1 ng; flumequin,
10 ng.

Beta-lactam antibiotics were detected on two plates, but

detection limits on plate IV were lower than on plate I. Meat
spiked with levels slightly higher than the respective detection
limits did not always produce inhibition zones. Meat spiked
with 3 ng penicillin G (0.03 mg kg21) was detected on both
plates. Plate I was less sensitive for ampicillin: meat spiked with
4 ng per disk was not always detected on plate I, but zones were
always wider than 12 mm on plate IV.


Analyst, 1998, 123, 23612365

2. Cephalosporins
Ceftiofur was detected on plate I and on plate IV. Addition of
meat affected the detection of ceftiofur substantially on both
Results on plate I were as follows: at 20 ng no zones occurred
around the disks with meat. At 50 ng, which is five times higher
than the detection limit, inhibition zones larger than 12 mm
were only observed in two of six cases, while the zones
surrounding the six control disks were larger than 23 mm. The
highest level, 100 ng per disk, was only tested on plate I;
inhibition zones were observed around the six meat disks, but
usually they were small and in one case less than 12 mm.
Detection limits of ceftiofur were lower on plate IV. Two of
six meat samples spiked with 20 ng ceftiofur and all six meat
samples spiked with 50 ng ceftiofur were detected on this plate.
Large variations of inhibition zones of samples spiked with the
same level were seen.
3. Tetracyclines
Low quantities, approximately equal to the detection limits,
were not always detected in spiked meat. With higher
quantities, the zones were almost equal around paper disks with

or without meat. Chlortetracycline and doxycycline were

detected at concentrations far below 0.1 mg kg21; the detection
limits of oxytetracycline and tetracycline in spiked meat were
approximately 0.1 mg kg21.
4. Aminoglycosides
Only streptomycin was tested. No inhibition zone was observed
around the meat pieces spiked with 100 ng streptomycin (1
mg kg21), which is five times higher than the detection limit of
an aqueous solution.

Erythromycin and tylosin were tested with meat at concentrations five-fold higher than the detection limits, 1 ng and 10 ng,
respectively. No inhibition was seen with meat spiked with 5 ng
erythromycin (0.05 mg kg21), while the zones of meat spiked
with 50 ng tylosin (0.5 mg kg21) were small and irregular. Meat
spiked with 40 ng (0.4 mg kg21) erythromycin yielded positive
results; diameters of zones ranged from 17 to > 30 mm.
Erythromycin and tylosin were also detectable on plate III,
but the detection limits were 2 and 20 ng, respectively. The
effects of samples spiked with 5 ng erythromycin (0.05
mg kg21) and 50 ng tylosin (0.5 mg kg21) were analoguous to
those observed on plate IV. Meat spiked with 40 ng (0.4
mg kg21) erythromycin was not always detected.

5. Macrolides and lincosamides

6. Sulfonamides
Two macrolides and one lincosamide were tested on plate IV.
Meat spiked with 80 ng lincomycin, corresponding with 0.8
mg kg21 tissue and four times higher than the detection limit,
was not detected (Fig. 1).

Only sulfadimidin was tested. No zones were observed round

meat disks spiked with 200 ng and 400 ng sulfadimidin
(corresponding to 2 and 4 mg kg21).

Table 1 Comparison of inhibition zones observed with paper disks impregnated with aqueous antibiotic solutions and identical disks layered with pork,
beef or chicken meat


Antibiotic (and detection limit)

Penicillin G (0.5 ng)

Ceftiofur (9 ng)
Ampicillin (3 ng)

Doxycycline (1 ng)
Chlortetracycline (1 ng)
Tetracycline (5 ng)
Oxytetracycline (8 ng)

Sulfadimidin (50 ng)


Streptomycin (20 ng)

Tylosin (20 ng)
Erythromycin (2 ng)


Penicillin G (0.5 ng)

Ampicillin (0.8 ng)
Ceftiofur (3 ng)
Tylosin (10 ng)
Erythromycin (1 ng)

Lincomycin (20 ng)

Flumequin (10 ng)

Enrofloxacin (0.5 ng)

Ciprofloxacin (0.3 ng)

Paper disk

Diameters of zones
without tissue (range of
six observations)/mm

Diameters of zones with muscle tissue

(range of six observations)/mm


> 30
28 > 30
> 30
> 30

No inhibition
No inhibition
< 815
Absent or small and asymmetric
Absent or asymmetric
< 812
No inhibition
No inhibition
No inhibition
Absent or small and asymmetric
No inhibition
No inhibition
Mostly absent
Absent or small and asymmetric
No inhibition
17 > 30
No inhibition
No inhibition
No inhibition
< 811

Analyst, 1998, 123, 23612365


7. Quinolones and fluoroquinolones

Low concentrations of flumequin, enrofloxacin and ciprofloxacin (10, 1 and 1 ng per disk, respectively), which were
detected in aqueous solutions, did not produce an inhibitory
zone of 12 mm when meat was laid upon the disks. Higher
concentrations of enrofloxacin and ciprofloxacin (5 ng per disk)
gave nearly equal zones with and without meat. The zones of
50 ng flumequin were smaller with meat than without meat.

Microbiological inhibition tests are intended for screening of
foods for residues of antibiotics. The purpose of such tests is to

Fig. 1 Inhibitory zones of 80 ng lincomycin on pH 8 medium seeded with

M. luteus: 0.01 ml of an aqueous solution, containing 0.008 mg ml21
lincomycin, cause a zone of 2829 mm diameter, but not when a disk of
frozen meat is laid upon it. Growth is even more abundant around the spiked
tissue, compared to the edges of the plate.

select samples which probably contain one or more analytes and

which should be investigated with more sophisticated immunochemical and/or chromatographic methods. Screening
tests should be simple, cheap, easy and fast. Multi-residue
screening methods are preferred to methods detecting only one
analyte. Inhibition tests such as the FPT fit into that profile.3
In routine residue testing, the great majority of samples are
evaluated on the basis of a screening test. Nevertheless, most
laboratories still give too little attention to the reliability of their
screening methods. Of course, the validation characteristics
differ between screening and confirmation methods. Analysts
prefer confirmation methods that do not give false positive
results, while a limited number of false positives is accepted in
the case of a screening test. On the other hand, the number of
false negative screening results should be as low as possible,
because negative samples are accepted without further analysis.
Our simplified spiking method demonstrated that the FPT is
not suited for detection of many antibiotics in muscle tissue. A
false negative result can be defined as follows: a negative result
from a sample spiked with a specified level of analyte,
preferentially corresponding with the MRL or safe level. The
number of false negative results was 100% for several
antibiotics when meat was added to the standard compound,
although the sensitivity of the plates was optimal (Table 2).
Sulfadimidin will not be detected with the FPT. The detection
limit of sulfadimidin is approximately 50 ng; this corresponds
with 0.5 mg kg21 tissue, which is far above the MRL.
Moreover, 400 ng are not detected when combined with 0.1 g of
frozen meat; this corresponds with 4 mg kg21 tissue. It is clear
that samples with sulfadimidin levels higher than the MRL
cannot be detected with this method. This finding corresponds
with unpublished results of routine tests in 375 pork meat
samples, obtained in our laboratory. Not one of 21 samples
containing sulfadimidin levels above the MRL was detected
with the FPT. Two samples with more than 1mg kg21
sulfadimidin did not cause inhibition on the plate supplemented
with trimethoprim.
The same applies for lincomycin. Lincomycin is only
detected on the pH 8 medium seeded with M. luteus, but the
detection limit is not sufficiently low to meet the MRL. Even
levels much higher than the MRL will not be detected in meat.
The FPT should not be used for detection of this antibiotic.
Furthermore, the aminoglycoside streptomycin, which is still
often used in combination with penicillin via the parenteral way
in large farm animals, is not detected in spiked meat with the
FPT. The method is also not suited for detection of tylosin. On
the other hand, plates III and IV are very sensitive for

Table 2 Maximal residue limits of antibiotics in muscle tissue, and usefulness of the FPT or a fifth medium intended for detection of quinolones for each
antibiotic or antibiotic group


MRL/mg kg21


Penicillin G

0.2 (B)
0.5 (P)
0.05 (P)




* B = beef; P = pork.


Analyst, 1998, 123, 23612365

Detection of MRL in
aqueous solution?

Detection of MRL in spiked

muscle tissue?

Yes, on plates I and IV

Yes, on plates I and IV
Yes, on plates I and IV

Not always on plate I
Most often not found
Not always on plate III

Yes, on plate I
Yes, on plate I
Yes, on plate I
Yes, on plate I
Yes, on plate III
Yes, on plates III and IV
Yes, on Plate IV
Yes, on plate V
Yes, on plate V

erythromycin. Although the detection was affected substantially

by the addition of a meat disk, all samples spiked with 40 ng
erythromycin, corresponding with the MRL of this antibiotic,
produced a zone wider than 12 mm on plate IV.
Only tetracyclines and quinolones are detected nearly equally
well with as without meat. Probably all samples with levels of
oxytetracycline, doxycycline, tetracycline, chlortetracycline,
enrofloxacin or ciprofloxacin equal to or higher than the MRL,
will be detected. This applies only to spiked samples; it is still
possible that residues of one or more antibiotics belonging to
these groups do not diffuse completely into the medium, when
they are present in naturally contaminated samples.
However, our experience with routine samples indicates that
tetracyclines are easily found with the FPT. A large survey on
antibiotic residues in poultry, beef, veal and pork meat obtained
from retail outlets revealed that 2% of the samples inhibited
B. subtilis. The majority of these positives contained a tetracycline.8 Similar observations were made with other routine
meat samples (unpublished data). In 1997, we tested 1768
poultry and 173 pork samples for presence of inhibiting
substances on plate I of the FPT. We found, respectively, 103
(5.8%) positive poultry samples, and 20 (8.3%) positive pork
meat samples. A limited number of these positive samples were
confirmed using an enzyme immunoassay (EIA) and with a
liquid chromatography-mass spectrometry (LC-MS) method,
and we found a fair correlation between the surface of the
inhibition zones and the actual amount of doxycycline measured
with LC-MS.9 In this series, doxycycline levels higher than the
MRL were not found in samples with inhibition rings smaller
than 5 mm (diameter of the zone 18 mm). This approaches
the zone produced by the same level of doxycycline in an
aqueous solution. On the basis of these data, it can be assumed
that the method is suitable for detection of tetracyclines unless
other observations contradict this statement.
Low levels of the beta-lactam antibiotics penicillin G and
ampicillin are detected more readily in aqueous solutions than
in spiked samples. Detection of MRL levels is, however, still
possible for both antibiotics. The detection limit of ampicillin is
lower on the pH 8 medium seeded with M. luteus than on the
pH 6 medium seeded with B. subtilis. Penicilline G levels equal
to the MRL can be found in spiked meat on both media. MRL
levels of ceftiofur are not detected in spiked meat on the pH 6
plate with B. subtilis, and not always on the plate seeded with M.
In routine practice we have confirmed the presence of betalactam with EIA in samples which had reacted positively on one
or more of the plates of the system (unpublished observations).
High levels of the fluoroquinolones enrofloxacin and ciprofloxacin were found with LC-MS in two FPT positive samples
of chicken meat.8 Up till now, we do not have sufficient data to
compare the actual levels with inhibition zones, and it cannot be
ascertained that all naturally contaminated samples with levels
above the MRL will cause inhibition zones on the appropriate
plates. This should preferably be confirmed with incurred
The causes of the differences between detection limits of
antibiotics in aqueous solution and in spiked muscle tissue were
not investigated. With some antibiotics the difference was only
observed at low levels. This can be due to variation in diffusion
into the agar layer. The phenomenon was not observed with
high levels of tetracyclines and the fluoroquinolones ciprofloxacin and enrofloxacin. It was seen, but to a lesser extent,
with beta-lactam antibiotics and the quinolone flumequin. A
possible explanation for the nearly total absence of inhibitory
zones with high levels of sulfadimidin, ceftiofur and macrolides

is a change in composition of the medium. The modified FPT

prescribes the use of a thin layer of agar medium; only 5 ml are
poured into 90 mm plates, which is just enough to cover the
surface. Thin agar layers cause higher sensitivities for aqueous
solutions of antibiotics. On these thin layers 2 mm thick pieces
of frozen meat are laid. Soluble cell contents as well as the
antibiotics diffuse into the agar. Normal muscle tissue has a pH
lower than 6. It is clear that the pH of the medium is influenced
by the matrix. Furthermore, the diffusion of proteins or other
nutrients into the medium may influence the detection of
antibiotics too. The definition of matrix effect, as used in
analytical chemistry, is an influence of the matrix on the
sensitivity of the sensor.10 Obviously, this is the case with the
inhibition test: very often test strains grow more abundantly
around blank meat disks, and this phenomenon was even
observed around the spiked tissues, as can be seen in Fig. 1.
When considering the FPT, it was concluded that the plates
with media at pH 7.2 and pH 8, both seeded with B. subtilis and
intended for the detection of sulfonamides and aminoglycosides, are not of any interest. They do not permit the detection
of muscle tissue spiked with sulfadimidin or streptomycin levels
far above the MRL. Only tetracyclines and beta-lactam
antibiotics can be detected up to the MRL level in spiked muscle
tissue. One or two plates are sufficient for that purpose. The
plate seeded with M. luteus detects penicillin and ampicillin up
to the MRL level, but is not suited to detect all macrolides and
lincosamides in spiked tissue. The addition of a plate seeded
with E. coli to the system will facilitate the detection of
quinolones and fluoroquinolones. These statements are based
upon the demonstration of an effect of the tissue matrix, added
to the antibiotic standards. Further evidence is needed that the
majority of naturally contaminated or incurred samples, with
residue levels equal to the MRL, will also react positive.

We thank Martine Boonaert and Ghislaine Vermassen for
technical help, and Chris Puttevils for the photograph. We
appreciated helpful discussion with Rgine Fuselier.



Veterinary Drug ResiduesResidues in Food Producing Animals

and their Products: Reference Materials and Methods, ed. R. J.
Heitzman, Blackwell Scientific Publications, Oxford, 1994, pp. Sg3/
L. Ellerbroek, Fleischwirtschaft, 1991, 71, 187.
G. O. Korsrud, J. O. Boison, J. F. M. Nouws and J. D. MacNeil,
J. AOAC Int., 1998, 81, 21.
A. M. Ferrini, V. Mannoni and P. Aureli, Arch. Lebensmittelhyg,
1997, 48, 133.
L. Ellerbroek, C. Schwarz, G. Hildebrandt, E. Weise, E.-M. Bernoth,
H.-J. Pluta and G. Arndt, Arch. Lebensmittelhyg, 1997, 48, 3.
J. F. M. Nouws, M. van Schothorst and G. Ziv, Arch. Lebensmittelhyg., 1979, 30, 4.
L. Ellerbroek, Arch. Lebensmittelhyg., 1998, 39, 7.
L. Okerman, J. Van Hoof and W. Debeuckelaere, J. AOAC Int., 1998,
81, 51.
K. De Wasch, L. Okerman, C. Croubels, H. De Brabander, J. Van
Hoof and P. De Backer, Analyst, 1998, in the press.
Chemometrics: a Textbook, ed. D. L. Massart, B. G. M. Vandeginste,
S. N. Deming, Y. Michotte and L. Kaufman, Elsevier, Amsterdam,
1988, ch. 8, pp. 115117.

Paper 8/04903C

Analyst, 1998, 123, 23612365