equation alog = 0.55/n- developed from Hal- vorson and Ziegler's data (11). ..... LITERATURE CITED. 1. Ashby, R. E., and M. E. Rhodes-Roberts. 1976. The use ...
APPLIED AND ENVIRONMENTAL MICROBIOLOGY, Nov. 1983, p. 1032-1037 0099-2240/83/111032-06$02.00/0 Copyright C 1983, American Society for Microbiology
Vol. 46, No. 5
Microplate Fecal Coliform Method to Monitor Stream Water Pollution A. MAUL AND J. C. BLOCK* de 1'Environnement, 57000 Metz, France des Sciences Centre
Received 5 April 1983/Accepted 12 August 1983
A study has been camred out on the Moselle River by means of a microtechnique based on the most-probable-number method for fecal coliform enumeration. This microtechnique, in which each serial dilution of a sample is inoculated into all 96 wells of a microplate, was compared with the standard membrane filter method. It showed a marked overestimation of about 14% due, probably, to the lack of absolute specificity of the method. The high precision of the microtechnique (13%, in terms of the coefficient of variation for log most probable number) and its relative independence from the influence of bacterial density allowed the use of analysis of variance to investigate the effects of spatial and temporal bacterial heterogeneity on the estimation of coliforms. Variability among replicate samples, subsamples, handling, and analytical errors were considered as the major sources of variation in bacterial titration. Variances associated with individual components of the sampling procedure were isolated, and optimal replications of each step were determined. Temporal variation was shown to be more influential than the other three components (most probable number, subsample, sample to sample), which were approximately equal in effect. However, the incidence of sample-to-sample variability (16%, in terms of the coefficient of variation for log most probable number) caused by spatial heterogeneity of bacterial populations in the Moselle River is shown and emphasized. Consequently, we recommend that replicate samples be taken on each occasion when conducting a sampling program for a stream pollution survey.
Obtaining precise and accurate estimates of bacterial populations in river water is complicated by the deficiencies of enumeration methods, such as their relative lack of precision and specificity (10, 27, 28), and by the highly nonrandom distribution of the bacteria. The scale of bacterial aggregations can be in the range of micrometers (when cells are sticking to detritus particles) (3, 8, 14, 23, 26, 29), centimeters (1, 18), or even kilometers (9, 19). The variability due to patchiness creates serious problems by hindering extrapolation of results from a single sample, which is often assumed to be representative, to a large body of water. Thus, to define a sampling design for fecal coliform analysis of surface water it is necessary to have both a good method of titration and an accurate knowledge of the different sources of titration errors, particularly those caused by the temporal and spatial heterogeneity of bacterial populations. The objectives of this study were as follows: (i) to compare the well-known membrane filtration (MF) technique for fecal coliform enumeration with a microplate most-probable-number method (FC-96) by using 96 inocula at each
serial dilution (method adapted from Prost and Hugues [21]); (ii) to study the relative importance of several aspects of the spatial and temporal variations upon fecal coliform density in the Moselle River; and (iii) to determine the contribution of each stage of the sampling to the increase of variance, to develop an optimal sampling design that minimizes the total variance. MATERIALS AND METHODS Sampling stations. Water samples were collected at different stations (A, B, and C, or C only) spaced at 200-m intervals along a canal supplying a thermal power station with Moselle River water. Samples were taken with sterile 500-ml bottles in the flow about 6 m away from the bank. Bottle samples held at ambient temperature were transported to the laboratory within 30 min after collection and analyzed immediately, in random order, for fecal coliform content. MF technique for fecal coliform enumeration. After the bottles were shaken manually, serial 10-fold dilutions of the original samples were made in sterile 0.8% NaCl solution. One milliliter of each dilution was transferred into 9 ml of diluent and immediately filtered through a sterile membrane (Millipore Corp.; HAWG, 0.45-jLm pore size). The membranes were
1032
VOL . 46, 1983
MONITORING STREAM WATER POLLUTION
deposited on lactose agar medium (Institut Pasteur, I.P. 64041) supplemented with triphenyltetrazolium chloride (I.P. 62635) and tergitol 7 (I.P. 62655). After 24 h of incubation at 44°C, the yellow-orange colonies with lactose fermentation into the agar were counted and expressed as fecal coliforms (CFU) in 1 ml of initial sample. These colonies were not submitted to biochemical tests for confirmation of their identity. FC-96. After the bottles were shaken manually, the original samples were diluted as previously described from one or two subsamples (1 or 10 ml), according to the experimental scheme. A 4.8-ml portion of each dilution was then distributed into the 96 wells (0.05 ml per well) of a microplate (Microtest II). Each well next received 0.2 ml of lactose broth plus bromocresol purple (I.P. 64041) as an indicator. After 24 h of incubation at 44°C, yellow-colored wells were scored as positive. The most probable number (MPN) of fecal coliform organisms per 1 ml of initial sample was calculated from the combination of positive readings in three successive dilutions by using a programmable calculator (Hewlett-Packard 41 C) for solving the wellknown MPN equation of Cochran (7). The yellow positive wells considered to contain fecal coliforms were not submitted to a confirmation step (indole production and growth at 44°C in brilliant green bile lactose broth). However, a previous study (4) showed that the proportion of false-positive wells was around 17.5% for the kind of water analyzed. Comparison of MF and FC-96 methods. Over a 1month period (April 1981), samples taken at the same station on the canal on 8 randomly selected days (one sample per day) were numbered serially and analyzed in sextuplicate for fecal coliforms. The six independent estimates obtained for each sample and for each analytical method were submitted to a standard twoway analysis of variance after testing for homogeneity of variance by Bartlett's test on log-transformed data. Characterizing spatial and temporal variations of fecal coliform density. Over a 1-month period (June 1981), 30 samples were taken at the three stations, A, B, and C (from downstream to upstream), on the canal on five randomly selected days at 9:00 a.m. and 3:00 p.m. (one sample per station per selected hour per selected day). Each sample was analyzed for fecal coliforms by the FC-96 method. An analysis of variance was performed on the log-transformed data after testing for homogeneity of the variances by using the Fm._, test (25). To assess the relative magnitudes of the several sources of variation, the total variance was partitioned into components associated with: the analytical variance (precision of the FC-96 method), the
1033
LSD at the
5% level
E
CL
see
E
4
tee.
C0 E-
1
2
3
4
5
6
7
8
sample FIG. 1. Comparison of FC-96 method (U) with the standard MF method (E) for fecal coliform determination. LSD, Least significant difference.
subsampling (variability between subsamples), the sampling (variability between bottle samples), diurnal variation, and daily fluctuations. RESULTS
Comparison of the MF technique and FC-96 method. Coliform counts obtained from both enumeration techniques are illustrated in Fig. 1. Each value represents the mean of six replicate assays performed on the same sample. The FC96 method, with 96 inocula per dilution, gives higher results than the MF technique in seven cases out of eight; the least significant difference between any two values calculated at the 5% level shows that the difference was significant for the samples 1, 3, 5, 7, and 8. The data taken as a whole were analyzed by using a two-way analysis of variance (Table 1). The effect of the sample-method interaction (SM) is significant at the 0.5% level. This means that the difference between coliform numbers given by the two methods varies with the analyzed sample. The F ratio calculated for factor M (method) is significant at the 6% level. Note that comparing the mean square associated with factor M to the
TABLE 1. Comparison of FC-96 and MF enumeration; analysis of variance of log-transformed FC-96 and MF results of Sums Degrees Source of variation F ratio Mean squares P Expected mean squares freedomof squares 7 Factor S (sample) 5.05039 0.72148 or + 12cr2 272.26 0.0000a Factor M (method) 1 0.07393 5.40 0.0531b 0.07393 aE + 6USM + 48aM oj + 6EKSM 5.17 0.0000a 0.09583 7 Interaction SM 0.01369 oj 0.21232 80 0.00265 Error E a
b
Significant at the 0.5% level. Not significant at the 5% level.
1034
APPL. ENVIRON. MICROBIOL.
MAUL AND BLOCK
interaction variance instead of the error variance increases the weight of the conclusion which is then extended beyond the eight analyzed samples, that is, to all of the possible collections from which they issue. A one-sided test performed on the arithmetic means of all of the log-transformed data obtained by the two enumeration techniques indicates that, on the average, the FC-96 method gives higher results than does the MF procedure by a factor of about 14 6%. The high falsepositive rate (17.5%) can probably account for this positive bias of FC-96 relative to that of MF. The adequacy of the Poisson distribution to fit the data can be examined by calculating Fisher's index of dispersion (20) for each set of six counts obtained by the MF procedure. None of the eight values of D2 (5.91, 1.46, 3.00, 3.99, 1.57, 1.87, 5.26, 0.95) was significantly higher than 1 (null value) at the 5% level. Consistently the same outcome occurred when comparing the sum of the eight D2 values to the chi-square distribution involving 40 degrees of freedom. This indicates that the Poisson distribution is an appropriate model to describe bacterial dispersion in small-volume, well-stirred samples. Last, the average standard deviation calculated for the entire body of data obtained in these studies by the MF technique is 0.0487, which amounts to 11.2% when expressed in terms of the coefficient of variation of the transformed bacterial density estimates. The reproducibility of the FC96 method expressed in terms of the coefficient of variation amounts to 12.5% (SMPN = 0.0542). The 95% confidence limits of this coefficient of variation calculated for 40 degrees of freedom were 10.3 and 16.1%. As additional support for the Poisson model assumption, the assessment of SMPN is in full agreement with the theoretical standard deviation of the log transformed MPN (alog) given by equation alog = 0.55/n- developed from Halvorson and Ziegler's data (11). When putting n equal to 96, alOg is 0.0561, which corresponds to a coefficient of variation of 13%. Consequently, all uncontrolled factors (i.e., handling errors, variability of the volume of drops, etc.) do not seem to affect the results in an appreciable way. Thus, the FC-96 method is characterized by a specific precision (=13%) that is, moreover, equivalent to that expected for the MF technique. Spatial and temporal variations of fecal coliform density in water. The results of the experiment devised to study spatial and temporal variation in fecal coliform density are given in Table 2. All reported values fell in the range of 517 to 2,004 organisms per ml. The outcome of the analysis of variance for the three-factor factorial experiment is present-
TABLE 2. Variation in fecal coliform densities of river water samples from three sampling stations on S sampling days at 9:00 a.m. and 3:00 p.m. SamDate
18 May
pling tion A B
±
C
26 May
A B
C 29 May
A B
C 1 June
A B
C
5 June
Sampling hour
No. of coliforms
Confidence interval
per mla
(95%)
9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m.
648 798 517 702 532 555
506-830 623-1,022 403-662 548-899 415-681 433-711
9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m.
1,421 1,388 1,883 1,855 1,724 1,769
1,110-1,821 1,084-1,778 1,470-2,411 1,448-2,376 1,346-2,208 1,381-2,265
9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m.
1,523 759 1,361 603 2,004
1,189-1,951 593-972 1,063-1,743
541
1,565-2,567 423-693
9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m.
1,987 1,056 1,796 1,579 1,221 1,223
1,551-2,544 824-1,352 1,402-2,300 1,233-2,022 953-1,564 955-1,566
471-772
870 679-1,114 858-1,407 1,099 B 920 718-1,178 951 743-1,219 C 926 723-1,185 887 692-1,136 a Geometric mean of results obtained from two 1-ml subsamples independently taken in each 500-ml bottle A
9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m. 9:00 a.m. 3:00 p.m.
sample. ed in Table 3. The technique yields a variance or mean square for each source of variation, i.e., the sampling day (factor D, with random levels), the sampling station (factor S, with fixed levels), and the sampling hour (factor H, with fixed levels). When a factor had no significant effect, the corresponding mean square and the next appropriate mean square of the table were pooled. The cUDSH component that is associated with patchiness interdependent with turbulence and water movements of an unpredictable nature is not significant at the 5% level. This may be attributed to the insufficient number of degrees of freedom involved in the test in view of the relative smallness of the corresponding coefficient of variation, which amounted to 12.1%.
VOL. 46, 1983
MONITORING STREAM WATER POLLUTION
1035
TABLE 3. Results of analysis of variance designed to separate spatial and temporal components of variance (analysis of log MPN) Source of variation
squares
Factor D (sampling day) 1.36988 Factor S (sampling 0.11730 station) + DS Factor H (sampling 0.55896 hour) + DH Interaction SH + 0.09704 interaction DSH Error E 0.16609 a Significant at the 0.5% level. b Not significant at the 5% level.
freedom
Mean squares
4 10
Expected mean squares
F ratio
P
0.34247 0.01173
o2 + 12cd2 o2 + 4(uj2s + as2)
61.82 2.12
0.0000a
5
0.11179
or2 + 6(L2H + uC4)
20.19
0.O000a
10
0.00970
or +
2(u2SH
1.75
0.1148b
30
0.00554
(J2
The results are undoubtedly affected by the hour of sampling, which indicated diurnal variation in coliform density. Particularly, differences between values obtained at 9:00 a.m. and 3:00 p.m. vary very significantly from one sampling day to another (day-hour interaction effect). It is interesting to note that the sampling station does not have a statistically significant effect on the results. This means that even iffixed and persistent spatial patchiness between stations exists, this does not, at any rate, preclude considering the sampling stations as equivalent in any further investigations. Hence, the section studied may be considered as homogeneous when taken as a whole. The latter statement is not surprising, since there is no water outlet along the canal; furthermore, the section is not long enough to allow substantial self-purification. The day effect was statistically significant at the 0.5% probability level, and the ratio between extreme daily means was approximately 3. Still, referring to other experiments conducted under identical conditions on the same site, daily fluctuations may vary over a wider range (i.e., as much as 25). Seen another way, the component of variance, cU2, mentioned in Table 3 is in fact the sum of two components, 44PN and rs2UB, which are, respectively, the analytical error variance characteristic of the FC-96 technique and the variance between repeated subsamples taken in the same sample, caused by patchiness within the bottle samples. The effects of this small-scale heterogeneity can be checked by comparing the ratio FSUB = S/S2 N to the appropriate F distribution. Putting SMPN equal to 0.00294 (estimation of i2MPN with 40 degrees of freedom) in the equation, FsUB becomes 1.884, which is higher than the F value given in tables with 30 and 40 degrees of freedom at the 95% level (P = 0.0305). Thereby, the component of variance,
C2UB, expressing variability among subsamples
has been estimated to SUB = 0.00260, which yields a coefficient of variation of 11.7%.
+ CrSH)
0.0544b
DISCUSSION MF technique and FC-96 method. The MPN method has often been criticized because of the poor precision permitted by too low a number of replicates per dilution (3, 5, or 10 replicates). However, using microplates with 96 wells inoculated at every serial dilution (21) results in some valuable properties: (i) the logarithm of the MPN is normally distributed, (ii) the coefficient of variation is practically independent of the bacterial density, and (iii) the precision is about 13%. The first two properties are true given the condition that bacterial density lies in an interval excluding extreme values (i.e., between 0.85 and 158 CFU per inoculum at the lowest dilution). Note that the range of this interval is convenient for practical use of this method. Furthermore, the properties mentioned are those required for the use of the analysis of variance. It seemed interesting to compare the results given by the FC-96 method with those obtained by the standard MF method in estimating the number of fecal coliforms at 44°C in river water. Our results, in agreement with those of other authors (10, 16, 27, 28), showed a tendency of the MPN to overestimate the bacterial density as determined by the plate count to an extent that depends on the origin and the nature of the water examined. Such an observation can be explained by a better growth of bacteria in liquid medium or by the sample-method interaction. As a matter of fact, the variations in the profile of relative frequencies of bacterial species (for instance, in consideration of climatic conditions as well as physicochemical characteristics of the water) undoubtedly influence the selectivity patterns of the two procedures. The results obtained by either fecal coliform method will be greatly affected by the incidence of false-positive or false-negative findings (12, 22, 27). Spatial and temporal variations of fecal coliform density. Environmental microbiology is devoted in large part to the study of spatial and
1036
MAUL AND BLOCK
APPL. ENVIRON. MICROBIOL.
temporal variation of bacterial density. Field sis; because of spatial heterogeneity, the results studies most often emphasized one or the other cannot be extrapolated to a volume of water of these types of variation, which not only greater than that which was analyzed. Thus, if depend on the physical or chemical characteris- the weight of any estimate is to be enhanced by tics of the water, but are also influenced by taking into account the effect of patchiness on climatic, biogeographic, and hydrographic fac- fecal coliform enumeration, a three-stage (samtors (5, 6). ple, subsample, assay) sampling process involvThe relative magnitudes of several sources of ing analysis of several samples should be devariation observed in our study and expressed as signed. Furthermore, with variance components in hand and the above equation giving the total a percentage of the total variance are 51.9% for daily fluctuations, 32.8% for diurnal variation, variance, it is possible to predict the confidence 5.4% for the FC-96 method, 5.1% for sample interval for the mean produced by any experivariation, and 4.8% for subsample variation. mental design. Moreover the optimal allocation Although these estimates are somewhat rough of sampling effort to each level for decreasing and therefore subject to appreciable distortion, the total variance can be determined. The results they show the preponderance of temporal varia- in Table 4 were calculated on the basis of the tion over the other three components, which are above-mentioned estimates of Oj4PN, USUB, and approximately equal. It must be emphasized that a2s. These results emphasize the effects of varisuch investigations are made easier by the math- ous sampling designs on the precision of the ematical properties of the FC-96 estimator (es- estimated mean. The effectiveness of different pecially the high level of its specific precision). replication schemes, involving four assays each, Note, for instance, that the variance of 5-5-5- may be compared with the first three rows of tube MPN would be very high and would then Table 4. Examination of the range of confidence represent 54% of the total variance. This points intervals shows the advantage of increasing repout the lack of accuracy to be expected in any lication at the upper levels of the sampling microbiological study conducted by such meth- design. Assessment of the error of the mean due to ods. In another study (15) we showed significant analytical error alone may be obtained by comdifferences between the replicate subsamples paring the schemes in the fourth and the fifth and among samples taken successively at the rows of Table 4. This shows how limited are the same hour. This indicates, on the one hand, a gains in precision that can be attained even with small-scale heterogeneity within the 500-ml sam- a titration technique that is assumed to be perples whereas on the other hand, variability be- fect, and it indicates that research devoted to tween repeated bottle samples reflects a larger improving the precision of the enumeration techscale patchiness. The estimated variance com- nique, without taking account of heterogeneity, ponents derived from each hierarchical level can accomplish very little. were S2PN = 0.00294, SSUB = 0.00270, and S2 = The last row of Table 4 refers to a single 0.00479 (i.e., method, subsample, and sample, assay, assuming incorrectly that coliform bacterespectively), with associated coefficients of ria are randomly distributed in the entire water variation amounting to 12.5, 12, and 16%, re- mass. Note that precision to this last case spectively. Although these figures derive from a (12.5%) is attained only by the first design, single experiment, similar results were obtained on other occasions. Optimal sampling design. The components of TABLE 4. Comparison of several sampling designs variance cannot be considered constant and Coeffi- Confidence SubAssays specific to a given environment, since the effect per Variance cient of limits at the samples of spatial heterogeneity on their magnitude un- Sampr variasubSI pies 95% level' doubtedly confers a certain variability to them, tion (M% sample sample even within a single body of water and all the 4 1 1 0.00261 11.8 79.4-125.9 more when environments are different. In any 2 2 1 0.00381 14.3 75.7-132.1 case each level of such a multistage, sampling 4 1 1 0.00823 21.1 66.4-150.6 process contributes to an increase in the total 1 1 1 0.01043 23.8 63.1-158.6 variance for the mean of all results. According to 00b 0.00749 1 1 20.1 67.7-147.8 Snedecor and Cochran (24), in a sampling design 1C 1 1 0.00294 12.5 78.3-127.7 consisting of p samples, q subsamples, and r a In each case the estimated mean has been stanassays per subsample, the total variance (oi) is dardized as 100%. b equal to (ops/p) + (crsuSIpq) + (OAipNlpqr). The symbol Xo indicates that no error must be Estimates of fecal coliform density that are attributed to the enumeration technique. c Assessment obtained by assuming a perfect homobased on the assay of a single sample or even a single subsample lack weight in statistical analy- geneity of the water mass.
VOL. 46, 1983
MONITORING STREAM WATER POLLUTION
which involved four samples. Based on the previous statements, and in agreement with other authors (2, 13, 17, 24), the most efficient and the least costly experimental design would consist of taking several samples and one subsample per sample to be analyzed only once. It is worth noting that the proposed scheme should be suitable for most cases in spite of the peculiar assumptions on which these findings are based. The number of samples to be collected must be chosen to satisfy a constraint on total variance or cost. Yet, if three samples appear to be a minimum, it should not be necessary to take more than six samples to meet requirements of most microbiological studies. If p is the number of analyzed samples, the logarithm of the bacterial density is given as X ± S/p-) t, where X and S are the arithmetic mean and the sample standard deviation of the p log-transformed estimations, respectively, when t is from Student's t distribution and has p - 1 degrees of freedom.
10.
11.
12.
13.
14.
15. 16.
17.
ACKNOWLEDGMENTS We have appreciated M. A. Dollard for her invaluable assistance and J. Nikes for typing this paper. We thank D. 0. Cliver and A. H. El-Shaarawi for discussion and criticisms.
18.
19.
LITERATURE CITED 1. Ashby, R. E., and M. E. Rhodes-Roberts. 1976. The use of analysis of variance to examine the variations between samples of marine bacterial populations. J. Appl. Bacteriol. 41:439-451. 2. Baleux, B., and M. Trousseller. 1982. Distribution spatiale et echantillonnage des bacteries heterotrophes dans les sediments lagunaires superficiels. J. Fr. Hydrol. 13:125139. 3. Bell, C. R., and L. J. Albright. 1981. Attached and freefloating bacteria in the Fraser River estuary, British Columbia, Canada. Mar. Ecol. Prog. Ser. 6:317-327. 4. Block, J. C., A. Maul, and M. A. Dollard. 1982. Estimation du nombre le plus probable (N.P.P.) des coliformes en riviere en realisant 96 inoculums par niveau de dilution. Sciences de l'Eau 4:387-399. 5. Breittmayer, J. P., and M. J. Gauthier. 1979. Denombrement des bacteries en milieu marin: facteurs de variation et d'incertitude. Ann. Microbiol. 130A:245-256. 6. Churchland, L. M., G. Kan, and A. Ages. 1982. Variation in fecal pollution indicators through tidal cycles in the Fraser River estuary. Can. J. Microbiol. 28:239-247. 7. Cochran, W. G. 1950. Estimation of bacterial densities by means of the "most probable number." Biometrics 6:105116. 8. Ducklow, H. W., D. L. Kirchman, and G. T. Rowe. 1982. Production and vertical flux of attached bacteria in the Hudson River plume of the New York Bight as studied with floating sediment traps. Appl. Environ. Microbiol. 43:769-776. 9. El-Shaarawi, A. H., S. R. Esterby, and B. J. Dutka. 1981. Bacterial density in water determined by Poisson or
20. 21.
22.
23. 24. 25. 26. 27.
28. 29.
1037
negative binomial distributions. Appl. Environ. Microbiol. 41:107-116. Green, B. L., and W. Litsky. 1980. Evaluation of membrane filter methods for enumeration of fecal coliforms from marine water. Marine Environ. Res. 3:267-276. Halvorson, H. O., and N. R. Ziegler. 1933. Application of statistics to problems in bacteriology. III. A consideration of the accuracy of dilution data obtained by using several dilutions. J. Bacteriol. 26:559-567. Hussong, D., J. M. Damare, R. M. Weiner, and R. R. Colweli. 1981. Bacteria associated with false-positive most-probable-number coliform test results for shellfish and estuaries. Appl. Environ. Microbiol. 41:35-45. Kirchman, D., J. Sigda, R. Kapuscinski, and R. Mitchell. 1982. Statistical analysis of the direct count method for enumerating bacteria. Appl. Environ. Microbiol. 44:376382. Kogure, K., V. Simidu, and N. Taga. 1982. Bacterial attachment to phytoplancton in sea water. J. Exp. Mar. Biol. Ecol. 56:197-204. Maul, A., M. A. Dolward, and J. C. Block. 1982. Etude de l'het6rog6n6ite spatio-temporelle des bacteries coliformes en rivitre. J. Fr. Hydrol. 13:141-156. McCarthy, J. A., J. E. Delaney, and R. J. Grasso. 1961. Measuring coliforms in water. Water Sew. Works 108:238-245. Montagna, P. A. 1982. Sampling design and enumeration statistics for bacteria extracted from marine sediments. Appl. Environ. Microbiol. 43:1366-1372. Palmer, F. E., R. D. Method, and J. T. Staley. 1976. Patchiness in the distribution of planktonic heterotrophic bacteria in lakes. Appl. Environ. Microbiol. 31:10031005. Platt, T., L. M. Dickie, and R. W. Trites. 1970. Spatial heterogeneity of phytoplancton in a near-shore environment. J. Fish. Rev. Board Can. 27:1453-1473. Poole, R. W. 1974. An introduction to quantitative ecology. McGraw-Hill Book Co., New York. Prost, H., and B. Hugues. 1982. Denombrement des microorganismes par la technique du nombre le plus probable (indice N.P.P.): emploi statistique de cet indice. Ann. Falsif. Expert. Chim. 75:185-207. Public Health Laboratory Service. 1972. Comparison of membrane filtration and multiple tube methods for the enumeration of coliform organisms in water. J. Hyg. 70:691-705. Seki, H. 1968. Some characteristics of marine microorganisms in the pacific central water. Rec. Oceanogr. Works Jpn. 9:247-255. Snedecor, G. W., and W. G. Cochran. 1957. Statistical methods, 6th ed. Iowa State University Press, Ames. Sokal, R. R., and F. J. Rohlf. 1969. Biometry. W. H. Freeman and Co., San Francisco. Sorokin, Y. I. 1970. On the degree of aggregation of marine bacterio-plankton. Dokl. Akad. Nauk. SSSR 192:905-907. Thomas, H. A., and R. L. Woodward. 1955. Estimation of coliform density by the membrane filter and the fermentation tube methods. J. Public Health 45:1431-1437. Velz, J. C. 1970. Bacterial enumeration, p. 572-593. In Applied stream sanitation. Wiley Interscience Publishers, Inc., New York. Williams, P. J. 1970. Heterotrophic utilization of dissolved organic compounds in the sea. I. Size distribution of population and relationship between respiration and incorporation of growth substrates. J. Marine Biol. Assoc. 50:859-870.
