Research Reports

Experimental Design: A Useful Tool for peR Optimization

Biolechniques 21: 134-140 (July 1996)

M.D. Boleda, P. Briones, J. Farres', L. Tyfield2 and R. PP Institut de Bioqufrnica ClfnicaCorporaci6 Sanitaria and CSIC, Cerdanyola, Spain; 'Universitat Autonorna de Barcelona, Bellaterra, Spain; 2Southmead Hospital, Bristol, UK; and 3Henkel Iberica S.A, Barcelona, Spain

ABSTRACT

Because of complex interactions among the components of Pt.R and the wide and increasing variety of applications in which this technique is used. optimization is necessary for every reaction. Here we describe the use of experimental design techniques (2k fractional factorial design and central composite design) to attain easier, quicker and cheaper peR optimization of DNA from blood spots. By determining the factors affecting the product yield first (factors screening), the quantity of template DNA needed for peR and the Mg2+ concentration are easily optimized (factors optimization).

134 BioTechniques

INTRODUCTION

In the brief period since its inception (8,20), the polymerase chain reaction (PCR) has become a widely used research technique. PCR has facilitated molecular analysis in many different fields of biological research including the diagnosis and characterization of human genetic disease, infectious disease and neoplasia (26,30,31), as well as the study of somatic mutations, evolution and forensic science (14,21,27). New applications for this powerful technology will undoubtedly continue to be found.

Because of complex interactions among the components of PCR and the wide variety of its application, it is very unlikely that one set of amplification conditions would be optimal for all situations (12,29). Thus, each application of the technique requires optimization, which frequently can be complex (as in the case of multiplex PCR) and timeconsuming. Here we suggest the use of experimental design techniques to achieve faster, easier and more cost-effective PCR optimization.

The experimental design technique was first described by Sir Ronald Fisher in 1920 (9) in an agricultural application. Later, Plackett and Burman (24) contributed to its development with the introduction of screening designs, and, in 1949, Taguchi incorporated the orthogonal matrix. Currently there are large numbers of experimental design techniques in use including complete factorial designs, 2k factorial designs, 2k fractional factorial designs, central

composite, latin squares, grecolatin squares, Plackett-Burrnann designs, Taguchi designs, mixtures, etc. (2). The wide variety of available techniques enables an investigator to choose the ones that are most appropriate in solving a specific problem. Many of these techniques have been used successfully in the development of paints (23), surfactants (16,22), cosmetics (4) and adhesives (19), and in the application of organic syntheses (7), enzymology (3), clinical chemistry (15,25) and protein chemistry (5,28). More recently, application has also been found in molecular biology, where the Taguchi methods were used in optimizing PCR (6).

Statistically designed experiments can be applied both during an initial screening or the developmental stage, or ultimately in experimental optimization. Provided the most appropriate experimental designs are chosen, an investigator is able to obtain information about the behavior of the system, to evaluate experimental error and to distinguish between correlation and causality with a minimum number of assays and minimal cost.

In this report we describe the application of two experimental design techniques (2k fractional factorial design and central composite design) to optimize PCR conditions for studying the most common mutation of the GALT gene, which encodes the enzyme galactose-I-phosphate uridyl transferase (GPUT) (E.C. 2.7.7.12). This mutation, Q 188R, results in a substitution of arginine for glutamine at amino acid position 188 of the GPUT enzyme and, in

Vol. 21, No.1 (1996)

the homozygous state, results in classical galactosaemia, an inborn error of carbohydrate metabolism. This is the most common disease-causing mutation in North America and accounts for over 60% of galactosaemia alleles (17).

min) and centrifugation (13); and B) soaking in 25 ul, of 1 x PCR buffer from Life Technologies (Gaithersburg, MD, USA) and boiling for 15 min. After cooling the sample on ice, 25 IlL of 1 x PCR buffer were added and the sample was centrifuged at 10 OOOx g for 15 min. The supernatant (containing the DNA) was then collected.

Chemicals

The mononucleotides dATP, dCTP, dGTP, and dTTP and primers for amplification were from Pharmacia Biotech (Piscataway, NJ, USA). Taq DNA polymerase was from Life Tech-

MATERIALS AND METHODS

Source of Template DNA

Human genomic DNA was obtained from dried blood spots (2 x 2 mm) according to two methods: A) fixation with methanol, evaporation, addition of distilled water (61 ul.), boiling (15

A)

Ml+
11+1.41
+1
- .II l& -
- ~ +1 +i:"41
-1.41 -1
-1
II -1.41 B)

I 116.4
7.4 15.1
4~3 16.:. )..5.7 15.1
-
- "14.3 -
8.1) '11.5
113.7 Figure 1. Central composite design to study the response surface for two quantitative factors (DNA volume and Mg2+ concentration). (.) Factorial points; (.) axial points; (.A.) center points. A) Coded level. B) Results.

Vol. 21, No.1 (1996)

nologies. Agarose was from Sigma Chemical (St. Louis, MO, USA). Molecular biology grade chemicals and sterilized Milli-Q® Plus water (Millipore, Molsheim, France) were used.

PCR

PCR was performed in a thermal cy-

cler (PTC-lOOTM; MJ Research, Water-

town, MA, USA). Primers for the Q188R mutation of the galactose-lphosphate uridyl transferase gene were

as follows: LH: 5'-ACC CCT GGT

CGG ATG TAA CG 3' and RH: 5'-

TCC CTG AGT AGC TCC TGG CG

3'. These provided a product of 607

bases. Hot start was performed (5 min/94°C), and after the Taq DNA ~ polymerase was added to the reaction

tubes, thirty cycles of amplification

were carried out with the following thermal profile: 1 minl94°C, 2 minl65°C and 1 minln°e. Total vol-

ume of the PCR reaction was 100 ul.; Reagent concentrations were as fol-

lows: 0.8 mM dNTPs, 0.25 mM primers and 2.5 U Taq DNA polymerase (standard conditions). The remaining conditions were tested as out-

lined in this paper. PCR products (10

ul.) were visualized in 1 % agarose gel electrophoresis stained with ethidium bromide. Restriction enzyme digestion

with Hpall cuts the product at an inter-

nal control cutting site into two frag-

ments of 559 and 48 bases. The smaller fragment is not usually visible on agarose gel electrophoresis. The Ql88R mutation creates a second ~ Hpall cutting site, thereby cutting the fragment of 559 bases into two frag-

ments of 285 and 274 bases. Thus, an individual heterozygous for the Ql88R mutation would have fragments of 559,

285 and 274 bases.

EXPERIMENTAL DESIGN

In designing a series of experiments to investigate the influence of several factors on a variable response, there are essentially two kinds of experimental approach. The more traditional method is the so-called "one factor each time" approach in which a single experiment is designed and only one factor is tested in each trial. The remaining factors are fixed and changed in successive

BioTechniques 135

Research Reports

experiments. A more effective and efficient approach is the experimental design strategy, which modifies several factors each time.

With experimental design it is possible to detect the interaction among various parameters. Even though the number of variables is high, a model predicting the responses of the individual factors can be obtained with a limited number of experiments. Because several statistical models of experimental design are available, it is essential that the most appropriate one be chosen. This would depend upon the objectives of the experiment, the types and numbers of factors, the number of runs

W needed and the statistical properties to be determine (18).

Factorial designs are used to determine the influence of several factors (variables, parameters) on several responses at different levels. In the complete factorial design, all of the possible combinations of factors and levels are carried out. When a plan of experiments is designed with k factors and n levels for each factor, there are nk experiments to perform. When the number of levels of each factor is two, then it is called a 2k factorial design. Thus, 23 means that there are three factors at two levels to study. Both designs (complete and 2k factorial) enable the determination of the main effects and the interactions among factors, but the former implies that a larger number of trials is needed to meet the objectives.

In the initial screening phase of setting up an experiment, the fractional factorial design is most useful for determining the most significant factors that influence the system. This consists of a fraction of the 2k factorial design in which there are only a limited number of combinations of factor levels. The combinations are selected to obtain the maximum information about the system with a limited number of trials. There will be some loss of information about the system (2), and interactions among the various components of the system will not be obtained. Nevertheless, this approach can be used in the first step of research development and can be followed subsequently with other designs that provide more complete information about individual factors. In this paper we have used the fractional

136 BioTechniques

Table 1. Characteristics of Experimental Designs

Number of
Design Scope Factors Information
2k Factorial Screening 3-15 Main effects, Interactions
2k Fractional
Factorial' Influence of factors 2-8 Main effects
Composite
Design" Optimization 2-4 Response surface , 2k Fractional factorial designs are not suitable when the factors are not independent because interactions remain hidden.

"Central composite designs are not suitable when the number of factors is large because the number of trials becomes extremely large.

peR Yield

a.a

10

15
2.4 ,,;
%
2
.
....
"
13 eo

/

Figure 2. Contour plot of PCR yield as function of applied template DNA and Mg2+ concentration. The optimum conditions are those encircled in the shaded area.

1.6

1.2

20

40

so

DNA (pL) "~I .....

Vol. 21, No. I (1996)

Table 2. Factors and Levels Assayed in Fractional Factorial Design: Factors Screening

Factor Level + Level-
1. Autoclaved blood spots Yes No
2. Method of DNA extraction A B
3. Number of blood spots (2 x 2 mm) 12 4
4. Quantity of template DNA 40 I-lL 20 I-lL
5. Mg2+ concentration 4mM 2mM
Table 3. Summary of the Experiments
a: Matrix for Fractional Factorial Screening Design
Factors
2 3 4 5
DNA No. of Blood Template Mg2+
Trial Autoclave Extraction Spots DNA Cone. Score .,
+ + 8
2 + 2
3 + 3
4 + + + + 8
5 + + 0
6 + + + 2
7 + + + 4
8 + + + + 0 b: Effects of the Factors on the PCR Yield

Factor Effect
-3.75
2 +0.75
3 -0.75
4 +4.25
5 +1.25 factorial design techniques in the first stage of PCR optimization.

Central composite design is used in the final stage of optimization to obtain the equation that best describes the behavior of the system inside the limits of the region studied (response surface). It is possible to obtain an equation composed of the main effects of the system, interactions of the factors and square effects. This design is built from a 2k factorial design and also includes some points on the axes and at the intersection of the axes (see Figure I).

In order to optimize the concentration of' magnesium chloride and the quantity of DNA for the PCR to screen

Vol. 21, No.1 (1996)

samples for the Q 188R mutation, we chose the central composite design (2). All data analyses were performed using Statgraphics 5.0 software (Graphic Software System, Rockville, MD, USA). Table 1 summarizes the characteristics of the aforementioned experimental design systems.

RESULTS AND DISCUSSION

PCR Yield: Factors Screening Using 2k Fractional Factorial Design (FFD)

The system under investigation is the amplification of a region of the

Research Reports

Table 4. Matrix for Central Composite Design
Factors Densitometry
Coded Level Actual Level (Arbitrary
Template Mg2+ Template Mg2+ Units
Trial DNA (ILL) (mM) DNA (ILL) (mM) x 10-3)
II 1. -1 -1 20 1.5 8.0
2. +1 ·1 60 1.5 11.5
3. -1 +1 20 2.5 7.4
4. +1 +1 60 2.5 15.1
5. -1.41 a 11.7 2.0 4.03
6. +1.41 0 68.2 2.0 15.1
7. 0 -1.41 40 1.3 13.7
8. 0 + 1.41 40 2.7 16.4
9. 0 0 40 2.0 16.9
10. 0 0 40 2.0 15.7
11. 0 0 40 2.0 14.3 GALT gene by PCR in order to characterize the c mmon mutation (Q 188R). Thi mutation results. in the hornozy-

1 2 3

607 bp

Figure . Agarose gel electrophoresis with standard and optimized PCR conditions. Lane I: Optimized conditions (50 ul, DNA. 2.2 111M Mg2+). Lane 2: Standard conditions (20 ~L DNA. 1.5 mM Mg2+). Lane 3: I kb DNA ladder (DNA molecular ize standard).

138 Biofecbniques

gous state, in classical galactosaemia (11) through a loss in activity of the enzymeGPUT.

FFD was applied to determine which factors within two procedures, DNA extraction and PCR conditions, influenced the PCR yield. Five variables were assayed and included the following: 1) the importance of autoclaving the blood spots prior to DNA extraction in order to inactivate DNases; 2) the method of extraction of human genomic DNA from dried blood spots; 3) the number of blood spots (2 x 2 mm) in each DNA extraction lube; 4) the volume of template DNA used in the PCR: and 5) Mg2+ concentration. The number of cycles and the length and temperature of each cycle were fixed. AU factors were studied at two levels (Table 2).

If a full factorial experiment were to be undertaken (5 factors at 2 levels), 32 trials would be required. In order to determine the main effects of the five factors, a quarter-fraction of the full factorial design should be sufficient to determine the most important factors that would optimize the PCR yield. Thus, the fractional factorial design 2{S- 2), which requires only 8 experiments, was selected. A summary of the experiments is given in Table 3a.

We evaluated the intensity of the

electrophoretic bands obtained from the 8 ex peri ments and scored a val ue rangi ng from 0 to 10. The effect of each factor is estimated by taking the sum of the experimental result for aU experiments, with its appropriate ign, and dividing the total by 4. The same result can be obtained by subtracting the average experimental value of the 4 (-) trials from that of the 4 (+) trials. The values of the effects are given in Table 3b. For example, the effect of autoclaving the blood spots before PCR on the subsequent ampl ification is calculated in the following way:

Average of Average of Autoclave Autoclave

Level (+) Level (-)

Autoclave Effect ==

0+2+4+0 - 8+2+3+8 = -3.75

4 4

It can be concluded from Table 3b that the number of blood spots and the DNA extraction method have little effect on PCR yield, whereas the other three factors (1,4 and 5) have a major influence. Of these, autoclaving has a considerable effect in reducing the PCR yield; whereas, both increasing the concentration of magnesium from 2 to 4 mM and increasing the quantity of template DNA applied have beneficial effects. Thus, the blood spots were not autoclaved before use, five blood spots were used in the PCR and method B (see Materials and Methods) was cho en for extraction.

Mg2+ and Quantity of Template DNA Optimization: Central Composite Design (CCD)

CCD was used to optimize the two continuous variables that have the greatest positive impact on amplification yield in PCR: Mg2+ concentration and volume of template D A.

Central composite designs are response surface designs. These are used to characterize a response that is approximated by a polynomial (usually low order) of quantitative factors (1). When a region in the factor space is found where the response is highest, response surface designs can outline an experiment that maps the response over that region. The central composite design. suitable for investigating the te-

Vol. 21, o. J (1996)

F

sponse surface for two quantitative factors, is a factorial design 22 with additional experimental points on the axes and at the point of intersection of the axes of the design (composite designs). This design is illustrated in Figure lA. The response was evaluated according to the intensity of the bands when the PCR products were separated by electrophoresis in a 1 % agarose gel. The intensity of the bands was determined by using a Shimadzu CS9000 densitometer (Columbia, MD, USA). The experimental matrix, as applied to this investigation, and the results are shown in Figure lB and Table 4.

By means of least squares fitting, the equation can be calculated as follows, which describes the behavior of the system in the region studied:

Yield = 17.93 + 0.72 + 14.73 B + 0.1 AB - 9.55 x 10-3 A2 - 4.31 B2

where A is the volume of template DNA and B is the Mg2+ concentration.

The contour plot of this equation (Figure 2) allows localization of the optimal working zone which will enable the best amplification for the DNA fragment studied. Thus, the Mg2+ concentration must be between 2.2 and 2.4 mM, and the optimum volume of template DNA must be approximately 50 ut,

After optimization of the PCR system, an experiment was performed in which the volume of template DNA was increased to 71.5 ul, in a total volume of 100 ul.. The above equation would predict a reduction in the yield of amplified product since the coefficient for A2 is negative. Such a reduction in yield was observed (results not shown) and is compatible with reports of others (10).

We have illustrated in this paper that discrete, as well as continuous, variables affecting PCR performance can be conveniently evaluated by using the technique of experimental design. Figure 3 shows the results obtained in agarose gel electrophoresis using standard and optimized PCR conditions. Experimental design can also be a valuable tool for optimizing other PCR processes, such as multiplex PCR, random amplified polymorphic DNA (RAPD)-PCR, long PCR and for developing new PCR applications.

Vol. 21, No.1 (1996)

[

Research Reports

ACKNOWLEDGMENTS

This work was supported in part by a grant from the Fondo de Investigaciones Sanitarias de la Seguridad Social (Grant No. 91100110201).

REFERENCES

•

I.Bayne, C.K. and I.B. Rubin. 1986. Practical Experimental Designs and Optimization Methods for Chemists. VCH Publishers, Deerfield Beach, Florida.

2.Box, G.E.P., W.G. Hunter and J.S. Hunter. 1978. Statistics for Experimenters. John Wiley & Sons, New York.

3.Burtis, C.A., W.D. Bostick, J.B. Overton and J. E. Mrochek. 1981. Optimization of a kinetic method by response surface methodology and centrifugal analysis and application to the enzymatic measurement of ethanol. Anal. Chern. 53:1154-1 159.

4.Carlotti, M.E., R. Carpignano, M.R. Gasco and M. Trotta. 1991. Optimization of emulsions. International Journal of Cosmetic Science 13:209-219.

5.Carter Jr., C.W. 1990. Efficient factorial designs and the analysis of macromolecular crystal growth conditions. Methods 1:12-24.

6.Cobb, B.D. and M. Clarkson. 1994. A simple procedure for optimizing the polymerase chain reaction (PCR) using modified Taguchi methods. Nucleic Acids Res. 22:3801-3805.

7.Elguero, J., M. Espada, D. Mathien and R.

Phan-Tan-Luu. 1979. Criblage de nombreux facteurs. Analyse d'un cas d' alcoylation dans les conditions de la catalyse par transfert de phase. An. Quimica 75:729-732.

8.Faloona, F. and K. Mullis. 1987. Specific synthesis of DNA in vitro via a polymerase catalysed chain reaction. Methods Enzymol. 155:335-350.

9.Fisher, R.A. 1935. The Design of Experiments. Oliver and Boyd, London.

10.Gregersen, N., A.I.F. Blakemore, V. Winter, B. Andersen, S. Kolvraa, L. Bolund, D. Curtis and P.c. Engel. 1991. Specific diagnosis of medium-chain acyl-CoA dehydrogenase (MCAD) deficiency in dried blood spots by a polymerase chain reaction (PCR) assay detecting a point-mutation (G985) in the MCAD gene. Clin. Chim. Acta 203:23-34.

1 I.Holton, J.B. 1990. Galactose disorders: an overview. J. Inherit. Metab. Dis. 13:476-522.

12.Innis, M.A., D.H. Gelfand, J.J. Sninsky and T.J. White (Eds.). 1990. PCR Protocols: A Guide to Methods and Applications. Academic Press, San Diego.

I3.Jinks, D.C., M. Minter, D.A. Tarver, M.

Vanderford, J.F. Hejtmancik and E.R.B. McCabe. 1989. Molecular genetic diagnosis of sickle cell disease using dried blood specimens on blotters used for newborn screening. Hum. Genet. 81:363-366.

14.Kim, H.S. and O. Smithies. 1988. Recombinant fragment assay for gene targetting based on the polymerase chain reaction. Nucleic Acids Res. 16:8887-8903.

15.Krause, R.D. and J.A. Lott. 1974. Use of the

•

140 BioTechniques

simplex method to optimize analytical conditions in clinical chemistry. Clin. Chern. 20:775-782.

16.Lalonde, J., E. J. Witte, M.L. O'Connell and L. Holliday. 1995. Protease stabilization by highly concentrated anionic surfactant mixtures. J. Am. Oil Chern. Soc. 72:53-59.

17.Leslie, N.D., E.B. Immerman, J.E. Flach, M. Florez, J.L. Fridovich-Keil and L.J. EIsas. 1992. The human galactose-I-phosphate uridyltransferase gene. Genomics 14:474-480.

18.Morgan, E. 1991. Chemometrics: Experimental Design. John Wiley & Sons, London.

19.Morphy, J.S., T.M. Santouso and D.J. Zimmer. 1987. Polymer emulsion blends as adhesives for flexible packaging. TAPPI J. 70 :144- 147.

20.Mullis, K.B., F.A. Faloona, S.J. Scharf, R.K.

Saiki, G.T. Horn and H.A. Erlich. 1986. Specific enzymatic amplification of DNA in vitro via a polymerase catalysed chain reaction. Cold Spring Harbor Symp. Quant. BioI. 51:263-273.

2l.Plilibo, S., J.A. Gifford and A.C. Wilson. 1988. Mitochondrial DNA sequences from a 7000-year old brain. Nucleic Acids Res. 16: 9775-9787.

22.Pi, R. 1994. Solid esterquats and experimental design, p. 267-279. Proceedings of XXV Comite Espafiol de la Detergencia Congress, Barcelona.

23.Pi, R. and A. Munoz. 1992. Experimental design: a useful tool. European Coatings Journal 437-443.

24.Plackett, R.L. and J.P. Burman. 1946. Design of optimum multifactorial experiments. Biometrika 3:305-325.

25.Rautela, G.S., R.D. Snee and W.K. Miller. 1979. Response surface co-optimization for reaction conditions in clinical chemical methods. Clin. Chern. 25:1954-1964.

26.Reiss, J. and D.N. Cooper. 1990. Application of the polymerase chain reaction to the diagnosis of human genetic disease. Hum. Genet. 85:1-8.

27.Reynolds, R. and G. Sensabaugh. 1991.

Analysis of genetic markers in forensic DNA samples using the polymerase chain reaction. Anal. Chern. 63:2-15.

28.Rubin, I.B., T.J. Mitchell and G. Goldstein. 1971. A program of statistical designs for optimizing specific transfer ribonucleic acid assay conditions. Anal. Chern. 43:717-721.

29.Saiki, R.K. 1989. The Design and optimization of the PCR, p. 7-16. In H.A. Erlich (Ed.), PCR Technology: Principles and Applications for DNA Amplification. Stockton Press, New York.

30.Williams, S.D. and S. Kwok. 1991. Polymerase chain reaction: applications for viral detection, p. 147-173. In E.H. Lennette (Ed.), Laboratory Diagnosis of Infections. Dekker, New York.

3 I .Zhu, X., J.M. Dunn, R.A. Phillips, A.D.

Goddard, K.E. Paton, A. Becker and B.L. Gallie. 1989. Preferential germ line mutations of the paternal allele in retinoblastoma. Nature 340:312-313.

Received 30 June 1995; accepted 5 January 1996.

Address correspondence to:

M. Dolors Boleda

Institut de Bioquimica Cl£nica

Corporacio Sanitaria. Avda Flor de Maig sin Apartat de Correus 137

08290-Cerdanyola del Valles

Barcelona. Spain

Vol. 21, No. I (1996)