Journal of Microbiological Methods 98 (2014) 35–40

Contents lists available at ScienceDirect

Journal of Microbiological Methods journal homepage: www.elsevier.com/locate/jmicmeth

Bacterial and Archaeal direct counts: A faster method of enumeration, for enrichment cultures and aqueous environmental samples Barry A. Cragg ⁎, R. John Parkes School of Earth and Ocean Sciences, Cardiff University, Main Building, Park Place, Cardiff, CF10 3AT, UK

a r t i c l e

i n f o

Article history: Received 23 August 2013 Received in revised form 29 November 2013 Accepted 1 December 2013 Available online 17 December 2013 Keywords: Bacteria Archaea Cell counts AODC Presence/Absence Poisson distribution

a b s t r a c t A new presence/absence method has been developed to count fluorochrome-stained bacterial and archaeal cells on membrane filters using epifluorescence microscopy. This approach was derived from the random distribution of cells on membranes that allowed the use of the Poisson distribution to estimate total cell densities. Comparison with the standard Acridine Orange Direct Count (AODC) technique shows no significant difference in the estimation of total cell populations, or any reduction in the precision of these estimations. The new method offers advantages over the standard AODC in considerably faster counting, as there is no need to discriminate between every potential cell visible on a field and fluorescent detritus, it is only necessary to confirm the presence of one cell. Additionally, the new method requires less skill, so has less reliance on expert counters, and that should reduce inter-counter variability. Although this work used the fluorochrome Acridine Orange, clearly the results are applicable to any fluorochrome used to count bacterial and archaeal cells. This method was developed using enrichment cultures for use with enrichment cultures and aqueous environmental samples where interfering detrital and mineral particles are minimal e.g., freshwater/seawater, therefore, it is not suitable for estimating total cells from sediment samples. This method has the potential for use in any situation where counts of randomly distributed items are made using a grid or quadrat system. © 2013 Elsevier B.V. All rights reserved.

1. Introduction Accurate estimation of the total number of cells in enrichment cultures or environmental water samples requires the use of a direct microscopic count method, and not a method that needs individual cells to demonstrate growth or colony formation as viable counts considerably underestimate total cell populations (Daley, 1979). Early versions of direct counting used either Coulter counters or light microscopy with counting chambers or, from 1933, membrane filters (Ehrlich, 1955; Francisco et al., 1973). Coulter counters are useful where samples are uniform and particle-free but where particulates exist miscounting can be a problem, and counting chambers are limited to cell concentrations greater than 106 – 107/mL, so generally requiring some sort of concentration technique for cells in natural waters (Collins, 1957). Early use of fluorescent dyes attempted to measure total fluorescence of a viewed microscope field and relates this to the cell abundance in that field. This proved problematic as each species of bacteria displayed a different fluorescence/cell count relationship, making the use of any such relationship with mixed environmental samples impossible (Ehrlich and Ehrmantraut, 1955). It also failed to differentiate stained cells from detritus and auto-fluorescence. The use of the fluorescent dye Acridine Orange (AO) with epifluorescence microscopy to just visualize cells was developed by Strugger (1948), and this was adapted for ⁎ Corresponding author. Tel.: +44 29 20874928. E-mail address: [email protected] (B.A. Cragg). 0167-7012/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.mimet.2013.12.006

counting cells by Francisco et al. (1973). Daley and Hobbie (1975), further modified the method and Hobbie et al. (1977), published detailed methodology in what has become the prime cited reference for the Acridine Orange Direct Count (AODC) technique. Since 1977 a number of articles have been published that seek to advance the technique by; investigating alternative fluorochromes, improving cell visualization on membrane filters, reducing the time taken to process samples and increasing the statistical precision of the counts. A variety of new fluorochromes have been tested against AO and, except for specific applications e.g., live/dead and active/inactive staining; only one, (DAPI) has, in the past, been adopted for general use (Maki and Remsen, 1981; Kepner and Pratt, 1994, and references therein). Until recently AO remained the fluorochrome of choice where samples contain particulates, and specifically for sediment samples, where the staining contrast between cells (blue/green) and background detritus (dull orange) is far superior to DAPI. Additionally, even with water samples, DAPI may give reduced counts when compared with AO (Suzuki et al., 1993). However, in recent years there has been increasing use of Sybr green (Shibata et al., 2006; Morono et al., 2009; Schippers et al., 2010), and this has now joined DAPI and AO as being in common usage for cell visualization and enumeration. Methods to improve cell visualization on the membrane filters have included moving from self staining the polycarbonate membrane using Irgalan Black (Kirchman et al., 1982) or Sudan Black (Pedersen and Ekendahl, 1990) to the adoption of commercially produced black polycarbonate membranes (Lee and Deininger, 1999), ensuring the

36

B.A. Cragg, R.J. Parkes / Journal of Microbiological Methods 98 (2014) 35–40

membrane is mounted damp and not dried out (von Münch and Pollard, 1997) and using slide mountants such as Citifluor AF2 to reduce the rate at which AO stained cells fade under UV light (Wynn-Williams, 1985). Increasing the statistical precision of the data has generally conflicted with the aim of reducing counting time as increasing precision has usually meant increased counting. Hobbie et al. (1977) initially suggested that 10 fields containing, in total, at least 200 cells should be counted on a filter. However, the number of cells that are counted has tended to increase with von Münch and Pollard (1997) counting approximately 30 cells over 1–6 fields and then repeating this at least 12 times on the same filter membrane. Konda et al. (1994) prepared duplicate filters and counted at least 500 cells/filter and Pedersen and Ekendahl (1990) counted up to 600 cells/filter on duplicate filters. Given the time taken to count this number of cells tedium and operator fatigue are recognized problems (Kirchman et al., 1982; Roser et al., 1984: Martens-Habbena and Sass, 2006). Additionally, a degree of expertise is required to accurately identify large numbers of fluorescing particles as bacterial or archaeal cells rather than stained detritus or auto-fluorescing minerals. Statistical analysis of the various components of cell counting by Kirchman et al. (1982), determined that increasing the number of fields counted beyond seven did very little to decrease sample variance and that increasing the number of filters that were counted was more useful, or if more than one subsample was available then counting only one filter from each of the subsamples was even better. However, these are based on 30 – 50 cells per field. They demonstrated that if the number of cells drops below 20 per field then the coefficient of variance rapidly increases, and here it makes sense to increase the number of fields counted (Montagna, 1982). Our approach to direct counting, based on that of Hobbie et al. (1977), consists of single samples where, from aqueous environmental samples, the number of cells per field is often considerably fewer than 20. Thus we count three filters each up to 200 cells, given a minimum viewing of 20 fields per filter, or 200 fields per filter, whichever occurs first. This is nevertheless still a time-consuming and tedious procedure requiring expertise and taking experienced counters up to one hour to process a sample and inexperienced individuals substantially longer. The aim of this work is to develop a method to rapidly enumerate cells in more or less particle-free aqueous samples that requires reduced counting, less expertise in identifying bacterial and archaeal cells and saves time.

The Poisson equation expands into an infinite series of discrete probabilities for all possible numbers of events that can occur, and using the example above, and a standard count of 200 fields, then predicted values can be calculated for the different numbers of cells/field that can occur (Eq. (2)). If it is known that 200 fields were counted and also known how many fields contained 0 cells, 1 cell, 2 cells, etc. then it is possible to back-calculate to obtain the mean number of cells/field (λ) and thus make an estimation of the population density. P¼ Probability of; at 200 fields counted ðprobability x 200Þ Number of fields with;

3:450 e−3:45 3:451 e−3:45 3:452 e−3:45 3:453 e−3:45 3:454 e−3:45 þ þ þ þ þ 0! 1! 2! 3! 4! 0 cells=field 1 cell=field 2 cells=field 3 cells=field 4 cells=field ¼ 0:0317 ¼ 0:1095 ¼ 0:1889 ¼ 0:2173 ¼ 0:1874 0 cells ¼6

1 cell ¼ 22

2 cells ¼ 38

3 cells ¼ 43

4 cells ¼ 37

ð2Þ Any single term in the Poisson expansion would serve this purpose, and this neatly separates the first term of the series (0 cells/field) from all of the others (1 or more cells/field) and allows a presence/absence approach to counting. Additionally, this first term further simplifies to e−λ as both λ0 and 0! are equal to 1. An equation can then be derived that covers any size field area, any size filterable membrane area and any number of fields counted to produce a population estimate from presence/absence counts. Müller et al. (2011) have already demonstrated, with some complex mathematics, that it is theoretically possible to estimate abundance from such presence–absence maps. 3. Calculation In the Cardiff laboratory setup for direct counting the total filterable area of the polycarbonate membrane is calculated to be 326 851 300 μm 2 and the sampling field is nominally an indexed grid of 10 × 10 squares measuring 95 μm × 95 μm overall on each side, equaling 9025 μm2. Thus there are 36 216 possible views per filter. Back-calculating from the Poisson term for numbers of occurrences of 0 cells/field the predicted number of cells/filter can be obtained from 36216 × (Ln 200 − Ln [number of 0 cells/field]) when 200 fields/filter are viewed. This can be simplified for general usage by converting natural logs to base 10 logs by multiplying by 2.3026 (Eq. (3)).     A F Cells=filter ¼ 2:3026 T Log10 Z AF

ð3Þ

2. Theory Where; Many authors have noticed, and demonstrated, that the distribution of cells on a filter is random or approaching random and therefore follows, or approximates to, a Poisson distribution (Jones, 1974; Kirchman et al., 1982; Roser et al., 1984; Jones et al., 1989; Fry, 1990; Kepner and Pratt, 1994; Fischer and Velimirov, 2000). Naturally this is only true where cells are individual rather than clumped or filamentous, and this is one of the conditions of this approach to counting. The Poisson distribution (Eq. (1)) is a special case of the binomial family of distributions, and occurs when the mean equals the variance. The mean usually

AT AF F Z

Total filterable area Area of a field Total number of fields viewed Number of fields containing 0 cells/field

After applying this equation the result is corrected for the volume of sample that was stained and any relevant dilution factor to produce an estimate of total cells/mL of the sample tested. 4. Method

P¼

n λ

λ e n!

where λ ¼ mean

ð1Þ

has a low numerical value and this distribution is used when positive events are rare and random. Using an example from our laboratory, if 0.25 mL of a freshwater sample was prepared for counting, containing 0.5 × 106 cells/mL, then given our laboratory setup where there are a possible 36 216 fields to view over the total filterable membrane area then the mean number of cells per field is only 3.45.

4.1. Sample preparation Two sets of experimental samples were used to test this procedure. In the first an enrichment pure culture of the Archaean Methanococoides sp. was used as the test cell. AODC of this culture gave the population size as (Log) 8.475 or 2.98 × 108 cells/mL. A dilution series was constructed of 9 × 1 mL volumes (M1 to M9) in plastic sterile (soaked in ethanol and dried in a laminar flow cabinet under UV light) Eppendorf vials (Fisher Scientific, Loughborough, UK). Each member of the dilution

B.A. Cragg, R.J. Parkes / Journal of Microbiological Methods 98 (2014) 35–40

series was individually constructed from stock culture and not serially diluted to ensure their statistical independence, and consisted of the relevant volume of stock culture plus 150 μL of 13.3% filter sterilized (0.1 μm) formaldehyde (to give a final concentration of 2%) and made up to 1 mL with filter sterilized (0.1 μm) Milli-Q water. These vials contained a range of calculated cell concentrations from (Log) 8.372 or 2.35 × 108 cells/mL to (Log) 6.168 or 1.47 × 106 cells/mL. In the second set of experimental samples an enrichment culture of the rod-shaped bacterium Desulfotomaculum geothermicum was used as the test cell. Again nine samples were constructed this time in parallel with one set, total volume 1 mL/vial, in hydrophobic Eppendorf vials (No-Stick microtubes, Alpha Laboratories, Eastleigh, UK) treated as above, and one set, total volume 5 mL/vial, in 10 mL Wheaton glass vials (furnaced at 400 °C for 2 h) with butyl rubber crimp seals (soaked in ethanol and dried in a laminar flow cabinet under UV light). Again samples were separately produced using the appropriate volume of stock culture, 150 μL/mL of 13.3% filter sterilized (0.1 μm) formaldehyde and made up to volume with filter sterilized (0.1 μm) Milli-Q water. Samples 1 to 5 (D1 to D5) contained from (Log) 7.255 or 1.8 × 107 cells/mL to (Log) 5.051 or 1.12 × 105 cells/mL. Low cell concentration samples were made by diluting the stock culture (50 μL in 5 mL) and dispensing as described above Samples 6 to 9 (D6 to D9) containing from (Log) 4.896 or 7.86 × 104 cells/mL to (Log) 3.197 or 1.57 × 103 cells/mL. AODC of triplicate filters from sample D1 from both vial types gave 1.793 × 107 (Eppendorf) and 1.801 × 107 cells/mL (Wheaton glass vial). A t-test on the log-transformed data showed no significant difference (t = 0.039; d.f. = 4; not sig.) between the population densities in the two vial types at this level, and the mean of these two values was taken as the level 1 cell concentration of the dilution series. 4.2. Filter preparation Prior to sub-sampling the vials were vigorously vortex mixed to ensure the cells were in suspension. Aliquots of the samples (4 – 1500 μL) were added to 10 mL volumes of filter sterilized (0.1 μm, Vacucap filter units, Gelman Sciences, Northampton, UK) 2% formaldehyde, made up in a solution containing 35 g/L NaCl for osmotic balance, in a 25 mL Sterilin bottle (Fisher Scientific). Filter sterilized (0.1 μm) AO at 50 μL of a 1 g/L solution was added, the sample was vortex mixed and left to stain for 3 min. The stained sample was vacuum filtered through a filter funnel (Sartorius, Epsom, UK) containing a pre-wetted black polycarbonate membrane (Fileder Filter Systems Ltd. Maidstone, UK). A further 10 mL volume of filter-sterilized 2% formaldehyde containing 35 g/L NaCl was filtered to flush excess stain from the filter. The filter was removed from the Sartorius funnel whilst still damp, but with no free water droplets, and mounted on a glass slide on a minimum amount of paraffin oil (Spectrograde, Fisher Scientific). More paraffin oil was added to the filter surface and a glass cover slip placed over the filter and firmly pressed down. A drop of immersion oil (Imersol 518 N, Zeiss, Cambridge, UK) was placed on top of the cover slip and the slide transferred to the microscope stage for counting. 4.3. Microscope setup Slides were viewed with a Zeiss Axioskop setup for epifluorescence, using a 50 W UV lamp and Zeiss filter set 9 for acridine orange (Excitation filter BP450-490, Chromatic beam splitter FT510, Barrier filter LP520). The objective was a Plan Neofluor 100× 1.3 oil (Zeiss) and ×10 eyepieces, one containing an eyepiece graticule indexed 10 × 10 grid. Having the microscope's transmission lamp on at its minimum level provided sufficient backlight to make observing the black indexed 10 × 10 grid against the black filter membrane background possible, and maintaining the focal plane on the filter surface easy, without compromising the counting of cells.

37

4.4. Counting procedure Samples M1 to M9 were counted on quadruplicate filters. For each filter 200 cells or 200 fields were counted (whichever came first) so long as a minimum of 20 fields had been counted. Field selection was performed by small random rotations of the horizontal stage control knob, so moving in a linear transect across the filter. Generally three such transects across each filter membrane were made to complete a count. Initially, (Samples M1 to M4), only the standard AODC method was used as cell densities were too high to use the new presence/ absence method. From Samples M5 to M7 both methods of counting were performed on exactly the same fields viewed. For samples containing low cell concentrations staining a larger subsample volume was not always sufficient and use of the indexed 10 × 10 grid was abandoned in favour of the entire circular field of view. This had been measured with a stage micrometer and was 29 865 μm2 some 3.3 times larger than the area of the indexed 10 × 10 grid, so increasing counting sensitivity. Samples M1 to M7 were counted twice more, using 200 fields/filter, after adjusting the sub-sample volume stained for each run such that filters firstly had high cell densities, and hence few fields containing 0 cells/field, and secondly had low cell densities and hence the majority of fields contained 0 cells/field. Recounting of samples M8 and M9 was not possible as cell densities were so low that the majority of the total sample volume (1 mL) had already been used in the first run of counts. Counts were recorded on an electronic counter (Diff-Count 10-308, Burkard Scientific, Uxbridge, UK). For AODC this involved registering each field with a single keystroke and registering all cells in that field by multiple strokes of a second key. For the presence/absence method each field was registered with a single keystroke and then if the field contained one or more cells a second key was pressed once. Although the new technique is based on determining the number of fields containing zero cells (cell absence) it is easier to record cell presence and adjust subsequent calculations accordingly. The two parallel sets of vials making up the second set of samples (D1 to D9) were counted once only on triplicate filters with both count methods always used on exactly the same views, although not necessarily the same size field areas. Where cell densities were high, for example, then AODC was usually based on the full indexed grid of 10 × 10 squares, whereas for the presence/absence method the counts may have been based on 3 × 3 or 4 × 4 or 5 × 5 of the small squares making up the indexed grid. The greater use of only part of the counting grid, particularly when small such as 3 × 3 and 2 × 2 squares has the potential to increase variability as with low numbers of cells on the grid the problem of what to do with cells that straddle the grid edge arises. Our approach has been to count all cells that straddle two of the perimeter sides and ignore those that straddle the other two sides. Subsequently triplicate filters from each of three of the Wheaton vial samples (samples D2, D3 & D4), covering a range of cell densities, were counted by both methods using 200, 100, 50 and 25 fields, to determine the effect of reducing the numbers of fields counted on the precision of the result. Additionally, the same three samples (D2, D3 & D4) were counted again at one filter/sample, and using 200 fields, where the numbers of cells in each field were separately registered. Calculations were then performed to confirm cell distributions on filters were indeed Poisson. 5. Results and discussion 5.1. Distribution testing Data from the samples counted to assess the distribution of the data (D2, D3 & D4) were plotted as frequency distributions (Fig. 1) and Chisquared tests were performed for each data set comparing the observed

B.A. Cragg, R.J. Parkes / Journal of Microbiological Methods 98 (2014) 35–40

5.2. Comparison of AODC and presence/absence methods

120 100

n = 203 = 1.773

80 60 40 20

B

120

Number of fields

A Number of fields

All data where both the AODC and presence/absence methods were performed on the same sub-sample were calculated as mean and 95% confidence limits (of both methods) and plotted as presence/absence results against standard AODC results (Fig. 2). The data were linear and a regression line was fitted (R2 = 0.9986). This line, with a slope of 0.993 was close to the line of equality and a comparison of slopes test comparing the regression line with the line y = x showed no significant difference (t = 1.92; d.f. = 32; n. sig.) indicating that both methods produced the same calculated cell concentrations. The most time consuming part of counting is the identification of fluorescent particles as cells given that organic detritus may take up AO, that some particles may be very small and that some mineral particles auto-fluoresce under UV. Therefore on many occasions an assessment may have to be made of a particle's shape and size. In any particular field there may be, for example, eight fluorescent particles of which five are clearly cells and the remaining three require assessment, and this requires a level of expertise in the technique. Using the presence/absence method the confirmation of only one real cell in the field is sufficient to generate a positive data value. This removal of the need to interpret the entire visual field considerably speeds up filter reading. It has the additional benefits of reducing variability between different people counting the same samples and, in requiring less expertise, individuals new to counting can be both as fast and directly comparable to more experienced counters. Additionally, as considerably fewer keystrokes are required for a presence/absence count, data registration is much faster. In our experience well-trained individuals can deal with a sample by the standard AODC method (using triplicate filters) at the rate of approximately one sample/h. This involves five minutes to prepare a filter and 15 min to count it. Under the new method it takes five minutes to prepare a filter and around two to three seconds to assess each view, giving a total count time of around 23 min/sample almost three times faster. The reduction in counting time is even greater for individuals with less expertise in the technique. The only situation where this time advantage starts to be lost is when cell concentrations are very low such that many fields contain only one cell. In this instance

100

0 1 2 3 4 5 6 7

Cells/field

8

y = 0.993x -0.0284 R² = 0.99862

7

6

5

4

3

2 2

4

6

8

Cell density by AODC (Log10 cells/mL) Fig. 2. Comparison of mean counts obtained from presence/absence and AODC methods on identical samples of two different cell cultures. Open squares using quadruplicate filters, and closed circles using duplicate filters with Desulfotomaculum geothermicum. Open circles using triplicate filters with Methanococcoides sp. Horizontal and vertical error bars are 95% confidence limits. Dotted line at 45° is the line of equality, solid line in the data is a regression line of best fit.

the need for interpretation of what is being viewed may return and slow counting down. These speeds of counting are comparable with those using image analysis techniques. Sieracki et al. (1985) digitized fluorescence images and used a modified Artek 810 image analyzer to acquire total counts. They estimated five minutes to prepare a filter and 6 s to count each field. With counting 15 fields/filter and three filters/sample this comes to 19.5 min/sample. This compared to visual counting times of 40 s/field (at around 50 cells/field) totaling to 45 min/sample by Sieracki et al. (1985), and 30 s/field (at 10 – 70 cells/field) by Kirchman et al., resulting in 37.5 min/sample under the same count conditions. Our samples usually have lower cell densities/field than these hence why we count up to 200 fields and take up to one hour/sample. An alternative approach proposed by Roser et al. (1984) adapted a plotless nearest-neighbour technique for counting cells on filters and this involved measuring the distance between the centre of a concentricallyringed graticule and the 3rd-nearest-cell in each of four quadrants of a field. This was done on the basis that cell densities on the membrane were so high that the nearest-neighbour method was faster than counting all the cells in a field. Again this is not applicable to our method, as with lower cell densities the nearest-neighbour method would

n = 201

= 0.577

80 60 40 20 0

0

9

C 120 Number of fields

distribution to a theoretical one obtained from assuming a Poisson distribution on the means obtained. Samples (D2) and (D3) showed no significant difference from a Poisson distribution (χ2 = 9.23; d.f. = 5; and χ2 = 6.93; d.f. = 3 respectively), however sample (D4) was significantly different (χ2 = 11.79; d.f. = 5; P b 0.05). The critical χ2 for this test is 11.07 thus the level of significance was not strong. This is comparable with earlier workers who have demonstrated that the distribution of cells on a filter is Poisson, or approximates to a Poisson distribution (Jones, 1974; Kirchman et al., 1982), and therefore we have assumed the Poisson distribution to be an acceptable model for the new method.

Cell density by presence/absence (Log10 cells/mL)

38

100

n = 200

= 1.995

80 60 40 20 0

0 1 2 3 4

Cells/field

0 1 2 3 4 5 6 7

Cells/field

Fig. 1. Distributions of cells/field for three different cell density samples A. Sample D2, B. Sample D3 & C. Sample D4 of a culture of Desulfotomaculum geothermicum. Text on the graphs indicates numbers of fields counted (n) and the mean cell density/field (λ).

B.A. Cragg, R.J. Parkes / Journal of Microbiological Methods 98 (2014) 35–40

take considerably longer. Where there are high cell densities it would be faster to stain and filter a smaller sub-sample and use the presence/ absence method.

Table 1 Comparison of confidence intervals obtained from presence/absence and AODC methods. Filters

Method

Mean

St. dev.

n

4 replicate filters 3 replicate filters 2 replicate filters

AODC Pres/Abs AODC Pres/Abs AODC Pres/Abs

±0.099 ±0.115 ±0.197 ±0.204 ±0.610 ±0.519

0.074 0.076 0.109 0.105 0.391 0.282

20 15 14 15 3 3

t value

Sig.

0.635

n. sig.

Confidence interval (Log10 units)

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 1

2

3

4

5

Number of filters Fig. 3. Confidence intervals (±) obtained from counting cells using 2, 3 or 4 filter replicates. Data from AODC and presence absence data combined as they do not significantly differ from each other (see Table 1).

are approximately ±0.305 Although a confidence interval of ±0.305 is not overly large it does demonstrate that as cell density on the filters decreases then count variability increases. However when the same analysis was applied to data obtained from quadruplicate filters no such significant relationship was found (t = 0.67; d.f. = 29; not significant), and confidence intervals were randomly distributed with respect to calculated cell concentrations.

A ± confidence intervals

Although the presence/absence method has been shown to be equally accurate to the standard AODC technique and much faster to use, it is necessary to demonstrate that precision hasn't been lost. The effect of the number of filters used is clearly seen (Fig. 2). The smallest confidence intervals were generally with sub-samples using quadruplicate filters (open squares), the intervals generally increased for sub-samples using triplicate filters (open circles) and were largest for sub-samples using duplicate filters (closed circles). Duplicate filters were forced by circumstance on three occasions, as the cell concentrations were sometimes so low that sub-sample size had to be increased to count the cells. In the Eppendorf vials with a total sample volume of 1 mL this meant two sub-samples of 450 μL in preference to three smaller sub-samples. Analysis of the confidence interval data using two-sample t-tests on both count methods under the three different replicate filter conditions indicated no significant differences between the count methods (Table 1) and thus no precision is lost using the presence/absence method. Our arbitrary maximum allowable confidence interval of ±0.5 Log10 units on our counts was chosen as it easily translates into finding two counts are likely to be significantly different from each other if they differ by one Log10 unit. Routinely, confidence limits for bacterial counts are acceptable at “twice as much” and “half as much”, equivalent to ±0.3 Log10 units. In practice, however, our confidence intervals obtained using the AODC technique are rarely greater than ±0.25 Log10 units, and frequently much lower. Table 1 indicates the preferred choice of triplicate filters/sample as reducing the unacceptably high variability of using two filters and not increasing the time of counting four filters. If the confidence interval data for both methods are plotted against the number of filters used per sample (Fig. 3) an exponential line fit is obtained with a strong fit (R2 = 0.990). From this relationship using triplicate filters reduces the confidence interval from using duplicate filters by 62%, and using quadruplicate filters reduces the confidence interval a further 19%. By extrapolation using five filters/sample was calculated to reduce the confidence interval by another 9% as the law of diminishing returns comes into play. As Kirchman et al. (1982) suggest that data variance is best minimized by increasing replication at the highest possible level it would appear that counts can be improved by decreasing the number of fields viewed to less than up-to-200 and increasing the number of filters prepared from three to four, or perhaps five, where necessary. This is supported by a different analysis of the same data set (Fig. 4). If confidence intervals for both methods, from triplicate filter data only, are independently plotted against their calculated cell concentrations/mL a negative slope regression line is obtained for the combined data with a significant slope (t = −4.19; d.f. = 29; P b 0.001). Interpolation indicated cell concentrations of 107/mL have confidence intervals of approximately ±0.12 whereas at 103 cells/mL the confidence intervals

y = 2.9387x-2.408 R² = 0.98961

0.45

y = -0.0462x + 0.4437 R² = 0.38553

0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 2.0

4.0

6.0

8.0

Cell density (Log10 cells/mL)

B ± confidence intervals

5.3. Counting precision

39

0.45 0.40

y = 0.0073x + 0.0601 R² = 0.01594

0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00

0.181

n. sig.

−0.327

n. sig.

2.0

4.0

6.0

8.0

10.0

Cell density (Log10 cells/mL) Fig. 4. Confidence intervals (± values) against cell density when samples are counted using A; three membrane filters, B: four membrane filters.

40

B.A. Cragg, R.J. Parkes / Journal of Microbiological Methods 98 (2014) 35–40

5.4. Reducing the number of fields counted

Acknowledgments

Where samples were counted on triplicate filters then counting 200, 100, 50 or 25 fields made no difference to the calculated mean cell concentration/mL of the samples, accuracy was maintained (Anova; F = 0.068, 0.123 & 0.796; df = 3,8; no significance, for samples D2, D3 & D4 respectively). However, precision declined with confidence intervals, increasing as the number of fields counted decreased (Table 2), and the increase was greater the lower the cell concentration/mL. Once again a paired sample t-test between the data for both methods within each sample (D2 to D4) showed no significant differences in the confidence intervals between the two methods. Counting only 25 fields produced large confidence intervals from all samples. Counting 50 fields, however, gave confidence intervals that were not significantly different from the confidence intervals obtained by counting 200 fields (t = −1.81; d.f. = 5; n. sig) for samples D2 and D3, indicating that no extra precision was gained by counting 200 fields compared to stopping at 50 fields. This was not true for sample D4 that, with a smaller cell concentration/mL, had greater variability, and therefore larger confidence intervals. It is probable that this would be rectified by counting an additional filter (a fourth), rather than increasing the number of fields counted (Kirchman et al., 1982).

We thank X. Tang for assistance in the cell counting and E. Roussel for helpful comments on this paper. This work was conducted whilst B. Cragg was in receipt of support from the Vice-Chancellor's Fund, Cardiff University.

5.5. Conclusion Cell counts using the new presence/absence method were consistently equivalent to cell counts made using the standard AODC method. The new method is considerably faster requiring considerably less actual counting and, as not every fluorescent particle has to be differentiated from detritus and absolutely identified as a definite bacterial or archaeal cell, it requires less skill. This opens the method to less experienced counters and should reduce inter-counter bias. Analysis of the variability of the data obtained by the new method indicated that there was no loss of precision in adopting presence/absence counting, and also suggested that counting up to 200 fields was unnecessary and that in most situations 50 fields was sufficient. Where this was not the case counting an additional filter was preferable to counting more fields. The results presented here were obtained only with AO; nevertheless, the usefulness of this protocol would apply equally well to DAPI and Sybr green, and indeed any other suitable fluorochrome if used under the same conditions. Additionally, this approach would seem to have wider applications where any rarely encountered item is being counted using a grid system, ranging from the microscopic e.g., pollen grains on a gridded slide to macroflora distribution in a field assessed with random quadrats.

Table 2 Confidence intervals (±) obtained from using the AODC and Presence/Absence methods to count three separate bacterial culture samples (D2, D3 & D4) using 25, 50, 100 or 200 fields counted per sample. A t-value, and its significance, is given for a paired-sample-ttest for each sample comparing confidence intervals from AODC and Presence/Absence methods of counting. The total cell concentration/mL is given in the final column. Fields counted

Mean Log10

Method

Sample

200

100

50

25

AODC Pres/Abs AODC Pres/Abs AODC Pres/Abs

D2 D2 D3 D3 D4 D4

0.097 0.076 0.046 0.101 0.177 0.172

0.051 0.111 0.117 0.217 0.344a 0.25

0.084 0.067 0.196 0.07 0.382a 0.361a

0.257 0.446a 0.482a 0.447a 0.613b 0.643b

a

t-value

Sig.

cells/mL

−1.075

n. sig.

6.983

0.03

n. sig.

6.453

0.862

n. sig.

5.825

indicates a confidence interval exceeding ±0.3 Log 10 units, and confidence interval exceeding ±0.5 Log 10 units.

b

indicates a

References Collins, V.C., 1957. The enumeration of waterborne bacteria by a new direct count method. J. Appl. Bacteriol. 20, 257–264. Daley, R.J., Hobbie, J.E., 1975. Direct counts of aquatic bacteria by a modified epifluorescence technique. Limnol. Oceanogr. 20, 875–882. Daley, R.J., 1979. Direct epifluorescence enumeration of native aquatic bacteria: uses, limitations and comparative accuracy. In: Costerton, J.W., Colwell, R.R. (Eds.), Native Aquatic Bacteria: Enumeration, Activity and Ecology. ASTM STP 605, Philadelphia, pp. 29–45. Ehrlich, R., 1955. Technique for microscopic count of microorganisms directly on membrane filters. J. Bacteriol. 70 (3), 265–268. Ehrlich, R., Ehrmantraut, H.C., 1955. Instrumental estimation of bacterial population by fluorescence microscopy. Appl. Microbiol. 3 (4), 231–234. Fischer, U.R., Velimirov, B., 2000. Comparative study of the abundance of various bacterial morphophytes in an eutrophic freshwater environment determined by AODC and TEM. J. Microbiol. Meth. 39, 213–224. Francisco, D.E., Mah, R.A., Rabin, A.C., 1973. Acridine orange-epifluorescence technique for counting bacteria in natural waters. Trans. Am. Microsc. Soc. 92 (3), 416–421. Fry, J.C., 1990. Direct methods and biomass estimation. In: Grigorova, R., Norris, J.R. (Eds.), Methods in Microbiology, 22. Academic Press, London, pp. 41–85. Hobbie, J.E., Daley, R.J., Jasper, S., 1977. Use of Nuclepore filters for counting bacteria by fluorescence microscopy. Appl. Environ. Microbiol. 33 (5), 1225–1228. Jones, J.G., 1974. Some observations on direct counts of freshwater bacteria obtained with a fluorescence microscope. Limnol. Oceanogr. 19, 540–543. Jones, S.E., Ditner, S.A., Freeman, C., Whitaker, C.J., Lock, M.A., 1989. Comparison of a new inorganic membrane filter (Anopore) with track-etched polycarbonate membrane filter (Nuclepore) for direct counting of bacteria. Appl. Environ. Microbiol. 55, 529–530. Kepner Jr., R.L., Pratt, J.R., 1994. Use of fluorochromes for direct enumeration of total bacteria in environmental samples: Past and present. Microbiol. Rev. 58 (4), 603–615. Kirchman, D., Sigda, J., Kapuscinski, R., Mitchell, R., 1982. Statistical analysis of the direct count method for enumerating bacteria. Appl. Environ. Microbiol. 44 (2), 376–382. Konda, T., Takii, S., Fukui, M., Kusuoka, Y., Matsumoto, G.I., Torii, T., 1994. Vertical distribution of bacterial population in Lake Fryxell, an Antarctic lake. Jpn. J. Limnol. 55 (3), 185–192. Lee, J., Deininger, R.A., 1999. A rapid method for detecting bacteria in drinking water. J. Rapid Methods Autom. Microbiol. 7, 135–145. Maki, J.S., Remsen, C.C., 1981. Comparison of two direct-count methods for determining metabolizing bacteria in freshwater. Appl. Environ. Microbiol. 41 (5), 1132–1138. Martens-Habbena, W., Sass, H., 2006. Sensitive determination of microbial growth by nucleic acid staining in aqueous suspension. Appl. Environ. Microbiol. 72 (1), 87–95. Montagna, P.A., 1982. Sampling design and enumeration statistics for bacteria extracted from marine sediments. Appl. Environ. Microbiol. 43 (6), 1366–1372. Morono, Y., Terada, T., Masui, N., Inagaki, F., 2009. Discriminative detection and enumeration of microbial life in marine subsurface sediments. ISME J. 3 (5), 503–511. Müller, C.H., Huggins, R., Hwang, W.-H., 2011. Consistent estimation of species abundance from a presence-absence map. Stat. Probabil. Lett. 81, 1449–1457. von Münch, E., Pollard, P.C., 1997. Measuring bacterial biomass-COD in wastewater containing particulate matter. Water Res. 31 (10), 2550–2556. Pedersen, K., Ekendahl, S., 1990. Distribution of bacteria in deep granitic groundwaters of southeastern Sweden. Microb. Ecol. 20, 37–52. Roser, D., Nedwell, D.B., Gordon, A., 1984. A note on ‘plotless’ methods for estimating bacterial cell densities. J. Appl. Bacteriol. 56, 343–347. Schippers, A., Köwerker, G., Höft, C., Teichert, B.M.A., 2010. Quantification of microbial communities in forearc sediment basins off Sumatra. Geomicrobiol J. 27, 170–182. Shibata, A., Goto, Y., Saito, H., Kikuchi, T., Toda, T., Taguchi, S., 2006. Comparison of SYBR green I and SYBR gold stains for enumerating bacteria and viruses by epifluorescence microscopy. Aquat. Microb. Ecol. 43, 223–231. Sieracki, M.E., Johnson, P.W., Sieburth, J. McN, 1985. Detection, enumeration, and sizing of planktonic bacteria by image-analysed epifluorescence microscopy. Appl. Environ. Microbiol. 49 (4), 799–810. Strugger, S., 1948. Fluorescence microscope examination of bacteria in soil. Can. J. Res. 26c (2), 188–193. Suzuki, M.T., Sherr, E.B., Sherr, B.F., 1993. DAPI direct counting underestimates bacterial abundances and average cell size compared to AO direct counting. Limnol. Oceanogr. 38 (7), 1566–1570. Wynn-Williams, D.D., 1985. Photofading retardant for epifluorescence microscopy in soil micro-ecological studies. Soil Biol. Biochem. 17 (6), 739–746.