Overcoming the Inflexibility of Most-Probable-Number Procedures Paul Woomer,* James Bennett and Russell Yost ABSTRACT cation compared to the enumeration of microbial populations in diagnostic culture media. In general, plant infection procedures are replicated between two and four times and with base dilution ratios of 2, 4, 5, and 10 (Brockwell, 1963; Brockwell et al., 1975; Fisher and Yates, 1943; Vincent, 1970). Experiments using MPN methods are currently restricted to published MPN tables (Table 1), These tables form an incomplete matrix of base dilution ratio and replicate number that do not allow for variations such as increasing accuracy through increasing the replication number (i.e., from 2 to 3 or from 5 to 6) or by decreasing the base dilution ratios (i.e., from 4 to 3 or from 2 to 1.5). The original tables of Fisher and Yates (1943) from which the tables of Vincent (1970) were derived, offer greater variety in replicate numbers but the solutions require additional calculations in order to yield an MPN estimate. In fact, tables which present the population estimates for MPN data, based on the dilution ratio and the replicate number, are less available than are the published descriptions of either the confidence factors (Cochran, 1950) or the probability of the experimental results (Scott and Porter, 1986; Stevens, 1958). In order to improve versatility of the MPN technique, we have developed a computer program which generates MPN tables for a wide range of dilution ratios at selected replicate levels, and which solves for an MPN estimate for many base dilution ratios and replicate numbers. The design of most-probable-number (MPN) experiments is restricted by the availability and completeness of tables for certain dilution ratio and replicate number combinations. The results presented in this paper are from a computer program, the Most Probable Number Enumeration System (MPNES), which generates solutions and confidence limits (P = 0.05) for population estimates of MPN data. Tables generated by MPNES agreed with existing tables yet MPNES was able to generate other tables and discrete solutions for design combinations that have not been published. The program was also used to generate population estimates from fractional base dilution ratios. The MPNES increases the accuracy of MPN estimates by decreasing the base dilution ratio, by increasing the number of replicates per dilution or by correcting for constantly inaccurate diluent volumes. The MPNES program also adjusts for inoculation volume and initial dilution ratios. In this way, MPN experiments can be designed to better measure the organism of interest rather than have the design dictated by published tables. THE MPN TECHNIQUE is widely used to enumerate microorganisms in foods, water and soil (Brockwell, 1963; deMan, 1975; Halvorson and Ziegler, 1933; West and Coleman, 1986; Woodward, 1957). This method is most important when a single type of microorganism must be measured within a heterogenous microbial population or when direct enumeration or plate counting are difficult. The MPN method does not rely on direct counts of single cells or colonies, but on the statistical probability of diluting to extinction (Alexander, 1982). Standard methods have been developed for some applications of the MPN procedure (American Public Health Assoc., 1985). In general, for microbiological applications, serial dilutions are applied to 5 or 10 replicates of some selective medium (American Public Health Assoc., 1985; deMan, 1975) because increases in replicate number result in narrower confidence intervals than less replicated MPN procedures (Cochran, 1950). In the case of the plant-infection technique for counting rhizobia, serial dilutions are applied to the root systems of plant units. There are no established standards for the number of replicates used. The range of microbial densities varies greatly between samples, preventing standard base dilution ratios. The need for uniformity of the plant host results in reduced repliP. Woomer, NifTAL Project, University Hawaii, 1000 Holomua Avenue, Paia, Hawaii, USA 96779; J. Bennett, Dep. Mathematics, Univ. Hawaii, Honolulu, Hawaii, USA 96817; R. Yost, Dep. Agronomy and Soil Science, Univ. Hawaii, Honolulu, Hawaii, USA 96822. This work was supported by the NSF Grant BSR-8516822 and the U.S. AID Cooperative Agreement D.AN-4177-A-00-603500 (NifTAL Project). Journal Series no. 3320 of the Hawaii Inst. Tropical Agric. and Human Resour. Received 1 I July 1988. *Corresponding author. Published in Agron. J. 82:349-353 (1990). 349 350 AGRONOMY JOURNAL, VOL. 82, MARCH-APRIL 1990 found such that f [(X 1 + X2)/2] - K = 0, and hence a solution has been found. Implementation The above algorithm was used to develop a program that provides two options for generating MPN estimates. One which generates tables of MPN data (positive tubes at various dilutions) and the corresponding MPN estimate, and the other which, given specific conditions, generates a specific MPN estimate. The confidence limits (P = 0.05) of MPN estimates were first calculated by Cochran (1950). The MPNES program calculates the confidence factor (CF) as where a = the dilution ratio and n = the number of replicates of the MPN equation. The upper and lower confidence limits (P = 0.05) are obtained from the confidence factor through multiplication with and division into the population estimate, respectively. The MPNES program presents the upper and lower confidence limits (P = 0.05) for every population estimate. The algorithm was initially written in IBM Basic and complied with Microsoft Basic Compiler. The program will run on IBM-compatible microcomputers with one floppy disk. The operating system requires MS DOS 2.0 or above and does not require additional software. Memory requirements are minimal (less than 256K RAM). Other formats may be accommodated on a case-by-case basis. RESULTS AND DISCUSSION Using the program Algorithm The computer solution follows the arguments above and is based on the following algorithm. First, the computer finds a number XI such that f(Xl) - K > 0. It then finds a number X2 such that f(X2) - K < 0. Now f[(X1 + X2)/2] - K is computed. If it equals 0 the algorithm is finished and a solution (X 1 + X2)/2 has been found. If it is greater than 0, then (X1 + X2)/2 is substituted for X1 above, X2 remains X2, and f [(X1 + X 2)/2] - K is again computed. If it is less than 0, (X1 + X 2)/2 is substituted for X 2, X 1 remains the same, and f [(XI + X2)/2] - K is computed. This cyclic process starts over again and continues until values of X1 and X 2 are The program is composed of two parts; one for generating MPN tables, and the other for computing individual MPNs. For a table, the user is asked to provide R (the base dilution ratio) and N (the number of tubes per dilution) via input statements. The combinations of positive tubes (P1-P6) in the table are provided by the program. The combinations of positive tubes do not include data statements in which the lowest serial dilution is not entirely positive (P1 < N). Only the most probable combinations are provided (deMan, 1975; Scott and Porter, 1986). While the base dilution ratio (R) may be any number, the number of tubes per dilution (N) is limited to 2, 3, 4, and 5. The individual MPN format can be used to obtain MPN estimates for MPN data not included among the combinations printed in the table. For an individual MPN, the base dilution ratio (R), number of tubes per dilution (N), and the numbers of positives per dilution (P1-P6) are provided by the user in response to prompts (Table 2). Options are available for correction in inoculation volume (when other than 1 mL) and/or initial dilution (when different from the dilution ratio). When all the necessary information is provided, an MPN for the combination of positive tubes for each dilution is generated. Accuracy of the Program The accuracy of the MPNES program is apparent when the MPN estimates generated by this program WOOMER ET AL.: OVERCOMING THE INFLEXIBILITY OF MPN PROCEDURES are compared to the estimates of other authors (Table 3). Note that the MPN estimates yielded through the MPNES program agrees least with the tables in Vincent (1970) and is more similar to the solutions offered by Brockwell (1963), Alexander (1982) and deMan (1975). The tables in Vincent (1970) are constructed through the mathematics of Fisher and Yates (1943) in which the sum of the positive experimental units is used to derive a population estimate regardless of the pattern (the number of positive results at each dilution step) of the MPN data. The other authors (Brockwell, 1963; Alexander, 1982; deMan, 1975) as well as the authors of the MPNES program rely on the mathematics of Halvorson and Ziegler (1933) in which a unique population estimate is assignable to each pattern of MPN data. Improvements Offered by the Program The MPNES program offered several advantages over existing MPN tables. Researchers were able to adjust base dilution ratios and replicate combinations to meet specific degrees of accuracy (Tables 4A and 4B). Infrequent codes which are not found in existing tables were assigned a population estimate (Table 4C). Through the ability to generate tables and estimates from individual code data of fractional base dilution ratios, adjustments of the base dilution ratio to actual diluent volumes were conveniently obtained (Table 4D). Cochran (1950) demonstrates that the confidence interval of a given MPN procedure is defined by the base dilution ratio and the replicate number through their influence on the standard error of the mean. Thus, certain combinations of base dilution ratio-replicate numbers allow for greater accuracy through reduction of the confidence factor. An important ex- 351 ample of increased accuracy is demonstrated in studies of relatively low numbers of microorganisms in agricultural (Singleton and Tavares, 1986) or public health (West and Coleman, 1986) applications using MPN procedures. More accurate measurement of microbial populations may lead to improved management of these microorganisms. Increasing the Range or Accuracy The flexibility of the MPNES table format is presented in Tables 5 and 6. Data in these tables illustrate how MPN estimates of increased accuracy (Table 5) or increased range of detection (Table 6) were generated. These experiments require fewer total growth units when compared to other tables (Vincent, 1970). The MPN procedure represented by Table 5 requires 18 experimental units yet has a confidence interval similar to a fourfold dilution table with four replicates (Vincent, 1970). The MPNES program produced unique MPN tables with extremely wide ranges that required few growth units (Table 6). This table has a base ratio of 12.5 with two replicates for six serial dilutions (total of 12 growth units) and is able to detect between 6.0 and 2.4 X 106 organisms per gram of sample. Nonstandard Dilution Ratios The ability to generate MPN estimates for nonstandard dilution ratios is important in three regards: it can be used to further adjust the range of the MPN to the population densities expected by the researcher (i.e., Table 5), fractional decreases in the base dilution ratio result in increased accuracy of MPN procedures through reduction of the confidence interval as in Table 413, and finally, should a researcher discover that the diluent blanks prepared for a given dilution series 352 AGRONOMY JOURNAL, VOL. 82, MARCH-APRIL 1990 are of uniform but incorrect volume, the MPNES program may be used to generate corrected population estimates for the nonwhole number base dilution ratio. The importance of this feature is demonstrated in Table 4D where instead of a fivefold dilution series, a four-and-one-half-fold series is initiated due to insufficient volume of diluent. In the example given, using the fivefold estimate results in an overestimation of the population by 70%. As situations require, researchers may choose to alter the inoculation volume or to include an initial dilution that is different from the base dilution ratio. For the individual MPN format, these modifications to the MPN procedure may be entered into the calculation of the MPN estimate. Increases in the inoculation volume and decreases in initial dilutions (lower than the base dilution ratio) serve to detect lower microbial populations, although not with greater accuracy, since neither the base dilution ratio nor the replicate number are being altered. Initial dilutions also serve to center the range of transition of the MPN data within the serial dilution series, allowing the microorganisms in the sample to be diluted to extinction [when all replicates of the final dilution(s) are entirely negative]. It is important that the sample be diluted to extinction since these treatments are the controls for the MPN experiment. The MPNES program generates MPN estimates for nonstandard initial dilutions without additional calculations. The accuracy of this MPNES option is compared to other sources (Table 7) where changes in the MPN estimate due to an initial dilution or inoculation volume are either incorporated into the MPN table (Brockwell, et al., 1975) or require additional calculations (Vincent, 1970). Limitations of the Program The MPNES table format was able to generate MPN tables for two, three, four, and five replicates. The range of base dilution ratios for which the tables accurately generate MPN estimates was between 1 and 14.5. The MPNES individual MPN format could generate population estimates for any positive whole number of replicates. Population estimates were generated by entering nonwbole numbers of replicates but these estimates are meaningless. Criteria to accept or reject specific MPN data are described by Stevens (1958) and Scott and Porter (1986) and applied by MPN researchers (Brockwell, 1963; Woomer, et al., 1988). The range of transition (Stevens, 1958) is the number of dilution steps between the first, not entirely positive and the last, not entirely negative dilution level. Extended ranges of WOOMER ET AL.: OVERCOMING THE INFLEXIBILITY OF MPN PROCEDURES transition indicate lower probability data or failure of technique. Other authors (deMan, 1975; Woodward, 1957) have assigned probabilities to the occurrence of specific patterns of MPN data. Certain data of low probability are acceptable when these occur infrequently and may be demonstrated as acceptable by the range of transition. Under these circumstances, the MPNES program is useful in assigning population estimates to codes which are not included in published MPN tables (Brockwell, 1963; Brockwell et al., 1975; deMan, 1975; Woodward, 1957). This feature must be used with caution. For example, the MPN data given in Table 4C does not warrant an MPN population estimate because the extended range of transition of 4 is extremely improbable (P = 0.0036). This is indicative of a failure in experimental technique (Stevens, 1958). In conclusion, the MPNES program was shown to be an improvement from existing MPN tables because the options available through this program allowed researchers greater flexibility. Restraint must be exercised in application of this program because MPN estimates can now be derived from MPN data that are very improbable (deMan, 1975; Halvorson and Ziegler, 1933) and may be the result of failure in meeting the principle assumptions underlying the MPN procedure (Stevens, 1958). Researchers must be aware that MPNES is not a solution to poor experimental results, but a tool in designing improved MPN experiments. ACKNOWLEDGMENTS The authors appreciably acknowledge technical assistance by Mr. Joe Rourke and Mr. Zhi-Cheng Li in development of the MPNES program and thank Ms. Susan Hiraoka for manuscript preparation. 353 REFERENCES I. Alexander, M. 1982. Most probable number method for microbial populations. In A.L. Page et al. (ed.) Methods of Soil Analysis. Part 2. 2nd ed. Agronomy 9:815-820. 2. American Public Health Assoc. 1985. Estimation of bacterial density. p. 802-804. In M. Franson (ed.) Standard Methods for the Examination of Water and Wastewater. 16th ed. American Public Health Assoc., Washington, DC. 3. Brockwell, J. 1963. Accuracy of a plant-infection technique for counting populations of Rhizobium trifohi. Appl. Microbiol. 1 1:377-383. 4. Brockwell, J., A. Diatloff, A. L. Garcia, and A.C. Robinson. 1975. Use of wild soybean (Glvcine ussuriensis Regel and Maack) as a test plant in dilution-modulation frequency tests for counting Rhizobium japonicum. Soil Biol. Biochem. 7:305-311. 5. Cochran, W.G. 1950. Estimation of bacterial densities by means of the most probable number. Biometrics 6:105-116. 6. deMan, J.C. 1975. The probability of most probable numbers. European J. Appl. Microbiol. 1:67-78. 7. Fisher, R.A., and F. Yates. 1943. Statistical tables for biological, agricultural and medical research. p. 6-8, 45. Oliver and Boyd Ltd., London. 8. Halvorson, H.O., and N.R. Ziegler. 1933. Applications of statistics to problems in bacteriology. I. .A means of determining bacterial population by the dilution method. J. Bacteriol. 25:101-121. 9. Scott, J.M., and F.E. Porter. 1986. An analysis of the accuracy of a plant infection technique for counting rhizobia. Soil Biol. Biochem. 18:355-362. 10. Singleton. P.W., and J.W. Tavares. 1986. Inoculation response of legumes in relation to the number and effectiveness of indigenous Rhizobium populations. Appl. Environ. Microbiol. 51:1013-1018. 11. Stevens, W.L. 1958. Dilution series: A statistical test of tech nique. J. R. Statis. Soc. Ser. B. 20:205-214. 12. Vincent, J.M. 1970. A manual for the practical study of the root nodule bacteria. Blackwell Scientific Publ., Oxford and Edinburgh. 13. West, P.A., and M.R. Coleman. 1986. A tentative national reference procedure for isolation and enumeration of Escherichia coli for bivalve mulluscan shellfish by most probable number method. J. Appl. Bacteriol. 1:505-516. 14. Woodward, R.L. 1957. How probable is the most probable num ber. J. .AWWA 49:1060-1068. 15. Woomer, P., P.W. Singleton, and B.B. Bohlool. 1988. Reliability of the most-probable-number technique for enumerating rhizobia in tropical soils. Appl. Environ. Microbiol. 54:1494-1497.