What is the fuss about in determining the number of colony forming units? After all, the CFU is only an estimate of the number of cells present. It is a skewed estimate at best as the only cells able to form colonies are those that can grow under the conditions of the test (incubation media, temperature, time, oxygen conditions, etc). Even among that group of microorganisms a colony does not represent a single cell, but rather cells that happened to be well separated on the plate and so can be distinguished after growth. A colony could arise from one cell, or several thousand. So why the fuss?
One reason for concern is that microbiology has a well-deserved reputation for being highly variable. Our lax attention to precision and accuracy in our measurements helps further this perception. We have allowed specifications for environmental monitoring, raw material bioburden, in-process bioburden and finished product bioburden to be imposed by regulation without regard for the ability of the method to support those specifications.
A second reason for concern is that now we are trying to introduce alternate microbiological methods into the lab. Being obsessive by training, we are trying to exceed measures of accuracy and precision in this exercise that the traditional methods cannot come close to matching. A good example of this is the Pharm Eur “Precision” requirement for an alternate method (quantification) to have a Relative Standard Deviation (RSD) in the range of 10-15% (1). While you might get lucky and hit this with dilutions whose plate counts are in the 150-250 CFU/plate range, – at lower plate counts this target value imposed by regulation will virtually guarantee a long, difficult and quite possibly unsuccessful, validation exercise.
The general ranges in common acceptance for countable numbers of colonies on a plate are 30 – 300 and 25 – 250. The origin of those ranges is worth examination. Breed and Dotterrer published a seminal paper on this topic in 1916 (2). They set out to determine the “limit in the number of colonies that may be allowed to grow on a plate without introducing serious errors…in connection with the proposed revisions of standard methods of milk analysis.” They note that “the kind of bacteria in the material under examination will have an influence on the size of the colonies, and consequently, on the number that can develop on a plate.” They also note that food supply can be an issue, that colonies close to each other on the plate may merge, and that neighbor colonies may inhibit growth or conversely stimulate growth. “Because of these and other difficulties certain plates in any series made from a given sample are more satisfactory for use in computing a total than are others. The matter of selecting plates to be used in computing a count becomes therefore a matter requiring considerable judgment.”
Breed and Dotterrer chose their countable plates from triplicate platings of each dilution, requiring acceptable plates to be within 20% of the average. On this analysis, plates with more than 400 CFU were unsatisfactory, as were those of less than 30 CFU, with best results in the range of 50-200 CFU/plate.
The major paper from Tomasiewicz et al (3) provides an excellent review of the continued evolution of the appropriate number of CFU per plate from milk. They took data from colony counts of raw milk from three different experiments (each dilution plated in triplicate) and used to determine a mean-squared-error of the estimate for all plates. Their recommendation at the end of the study was for a countable range of 25-250 CFU/plate in triplicate. It is interesting to note that although the authors note that CFU follow a Poisson distribution, no mention is made of any data transformation used to approximate a normal distribution prior to the use of normal statistical analytical tools. Tomasiewicz et al provide excellent cautionary advice:
“The data presented herein are not necessarily applicable to other systems. For automated equipment, the optimum range may well vary with the instrument…Furthermore, even if automation is not used appropriate numbers of colonies that should be on a countable plate can very widely, depending on many other variables. With soil fungi for example…”
The compendia have recently harmonized a microbial enumeration test (4), and in this test recommend that the technician “Select the plates corresponding to a given dilution and showing the highest number of colonies less than 250 for TAMC and 50 for TYMC.” In determination of the resistance of biological indicators, USP recommends a range of “20 to 300 colonies, but not less than 6” (5). However, the most complete description of the countable range is found in the informational chapter <1227> (6):
“The accepted range for countable colonies on a standard agar plate is between 25 and 250 for most bacteria and Candida albicans. This range was established in the food industry for counting coliform bacteria in milk. The range is acceptable for compendial organisms, except for fungi. It is not optimal for counting all environmental monitoring isolates. The recommended range for Aspergillus niger is between 8 to 80 cfu per plate. The use of membrane filtration to recover challenge organisms, or the use of environmental isolates as challenge organisms in the antimicrobial effectiveness testing, requires validation of the countable range.”
ASTM provides countable ranges of 20-80 CFU/membrane, 20-200 for spread plates and 30-300 for pour plates (7). The FDA Bacterial Analytical Manual (BAM) recommends 25-250 CFU/plate as a countable range (8).
The upper limit of plate counts is dependent on a number of factors, as described previously. The major issues include the colony size and behavior (swarming?), and the surface area of the plate. The size particularly comes into play with plating a membrane for determination of CFU as the surface area of that membrane is so much smaller than that of a standard plate.
TNTC can be reported out several ways. ASTM (7) recommends reporting this out as >”upper limit”. For example, a 1:10 dilution with more than 200 CFU on a spread plate would be reported as “>2,000 CFU/mL (or gram). FDA’s BAM recommends counting the colonies from the dilution with plates giving counts closest to 250, counting a portion of the plate, estimating the total number and then using that number as the Estimated Aerobic Count. It is not clear to the author how this is greatly superior to guessing. In my opinion this is an invalid plating and needs to be done correctly at a later date (note I am strenuously avoiding the use of the word retest. This result invalidates the plating and therefore the test was not performed correctly.) I know this is a hardship to the lab, who were trying to reduce the plating load initially by not plating out sufficient dilutions. However, making a mistake initially is not a reasonable excuse to avoid doing it correctly after the mistake is recognized. If the lab wishes to use this “estimated count” it should, at a minimum, have it clearly described in their “counting CFU” SOP with a rationale as to when the plate counts are not critical and can be estimated in this fashion.
There are methods available if you should want to accurately determine the upper limit for a unique plating surface or a unique colony type. One is presented in the USP informational chapter <1227> (5) which is based on a pair-wise comparison of counts from a dilution series. This is based on the assumption that at the upper limit the observed numbers of CFU will fall off the expected numbers at some point (see Figure 1). This divergence will become significant at some point – that defines the upper limit of CFU/plate.
A central concern in this determination is the reporting of the Limit of Quantification (which is what we are really interested in reporting) against the Limit of Detection (1 CFU). This is an important distinction if we are being held to specifications in the lower range.
ASTM recommendations focus on the LOD, and urge the user to report that answer out if no colonies are recovered (ie <10 CFU/mL for a 1:10 dilution) (7). If countable colonies are present, but below the countable range, count them anyway and report an estimated count.
USP (6) does not have a specific recommendation on how to report out these low numbers, but does note “Lower counting thresholds for the greatest dilution plating in series must be justified.”
FDA BAM (8) recommends a different reporting format. In the FDA BAM method, all counts are recorded in the raw data, but the information is reported out as <2,500. This is, in my opinion, the prudent course. The crux of the argument is that experiment studies have shown very poor accuracy in plate counts below 25 (see above). Theoretically we can argue that since the CFU follow the Poisson distribution, the error of the estimate is the square root of the average (USP <1227>). This leads to graphs such as in Figure 2 which shows us that as the CFU/plate drops below the countable range, the error as a percent of the mean increases rapidly. This confusion between the Limit of Detection and the Limit of Quantification for plate counts has led to some very difficult situations (as discussed below).
Ideally you would never see two separate dilutions with counts in the countable range, as the countable ranges cover a ten-fold range of CFU. However, this is microbiology. ASTM recommendations (7) urge you to take both dilutions into account, determining the CFU/mL (or gram) separately for each, then averaging the results for the final result. Breed and Dotterrer (2) also used several dilutions if the numbers fit the QC requirements (see below). FDA BAM has no recommendations in this situation.
While the argument can be made to use all counts, this is a stronger argument if triplicate plates are used and QC limits are in place to discard erroneous plates.
A strong argument can also be made to take the dilution providing the larger number of CFU in the countable range. This approach minimizes two concerns, that the errors in the estimates increase with increasing serial dilutions, and that the error in the estimate increases with decreasing plate counts. Use of the smaller dilution (eg 1:10 vs 1:100) could be justified from this perspective.
Whichever method used should be documented and justified in the “Counting CFU” SOP.
Periodically there are recommendations to establish Quality Control limits on replicate plate counts. Breed and Dotterrer in their 1916 paper (2) required valid plate counts from triplicate plates to provide estimates of CFU/mL within 20% of the mean. In other words, all plates were counted, each plate’s CFU count was used to estimate the original CFU/mL, then each estimate was evaluated. If the individual plate’s estimate was within 20% of the mean, it was deemed acceptable. This method is not practical in the QC lab.
Establishment of QC limits for plate counts works best if you have at least three replicate plates for each dilution. The average of the dilution replicates can be determined, variant counts (hopefully no more than one plate per triplicate plating) discarded and the final average determined. If you try this with duplicate plates you frequently end up with trying to average the results of one plate. While this method looks good on paper, the prudent lab manager will evaluate some historical data before instituting it as a practice.
The method used to QC individual plate counts, if used, should be documented and justified in SOP, along with the response to finding variant counts.
There have been suggestions that a larger volume of material may be plated across several plates, and the results reported out for the larger volume. For example, plating 10 1 mL samples on 10 different plates, and then reporting it as if a 10 mL sample was plated. This approach is flawed in that it ignores several sources of variability in plating including sampling error, plating errors, growth/incubation error and counting errors (9, 10). The correct interpretation for this situation that you have just plated 1 mL ten times, not 10 mL once. The numbers might be averaged, they cannot be added.
To discuss this we need to determine what the significant figures might be in the measure. For raw colony counts, common practice determines that the CFU observed determine the significant figure, and that the average is one decimal to the right of that number (sticklers for accuracy will report the geometric mean rather than the arithmetic mean given the Poisson distribution followed by CFU). In reporting, it is common practice to report out as scientific notation using two significant figures. This requires rounding.
USP (11) and ASTM (7) both round up at five if 5 is the number to the right of the last significant figure. FDA BAM has a more elaborate scheme, rounding up if the number is 6 or higher, down if 4 or lower. If the number is 5, BAM looks to the next number to the right and rounds up if it is odd, down if it is even.
This is one of those situations where you want every-one to do the calculations the same way (I am hard pressed to come up with a situation in a lab where you want everyone to do it differently). Be sure to include direction and its justification in the “Counting CFU” SOP if it does not already exist in a separate SOP.
We are back to the question of WHO CARES?
If you are faced with a finished product bioburden of NMT (Not More Than) 100 CFU/gram, and your method suitability study requires a 1:100 dilution of the product to overcome any antimicrobial effects, then how are you to test it? Common practice is to perform the 1:100 dilution, perform a pour plate of 1 mL in duplicate and if 2 colonies grow on each plate, the product fails specification. This common practice is scientifically unsupportable – it confused the Limit of Detection with the Limit of Quantification for the plate count method.
Let’s take a look at environmental monitoring alert and action levels for aseptically produced products. Hussong and Madsen (12) recently published a thoughtful review of this topic where they argue that the levels of acceptable CFU for many room classifications are below the noise level of plate count technology (eg in the range of 1-2 CFU/m3). In addition, environmental data can be extremely variable, much more so than controlled lab studies as the numbers of microorganisms, the physiological state of the isolates, even the species are completely out of the control of the investigator. In addition the numbers do not conform to a normal distribution as there are sporadic counts with a count of “zero” CFU predominating. They conclude that since the numbers are unreliable, the trend in the data is the only important consideration, and that EM counts cannot be used for product release criteria. A separate treatment of this subject was presented by Farrington (13) who argues that the relationship between EM data and finished product quality is a widely held, but unproven belief, compounded by the problems in accuracy with the low counts generated by plate count methodology.
In conclusion, all methods have limitations. One of the major limitations to the plate count method is the relatively narrow countable range (generally considered to be 25-250 CFU bacteria on a standard petri dish). The currently prevailing confusion between the Limit of Detection (1 CFU) and Limit of Quantification (25 CFU) for the plate count method creates a larger degree of variability in microbiology data than is necessary. An unfortunate regulatory trend in recent years is to establish expectations (specifications, limits, levels) for data generated by the plate count method that the accuracy of the method cannot support. This is a real opportunity for modification of current practice to approach the goal of “science-based regulations”.
Experts at the Microbiology Network are ready to assist with consulting or training to meet your needs. Have customized, in-house assistance with your questions from our recognized experts at your schedule either through consulting agreements, in-house training or customized webinars – contact us using the “Let’s Talk” communication found in the upper right of this page.