Rapid Methods for Testing the Efficacy of Sterilization-Grade Filter Membranes

Abstract

The validation of sterilization-grade membranes is integral to ensuring the efficient and safe use of microfiltration systems. Here validation refers to the production of sterile filtrate for sterilizing-grade membranes under challenge test conditions. Current validation methods require 48 h of culture for results to become available, which creates time delays within the manufacturing process and quality control (QC) backlogs. This work compares four methods for the production of filter challenge test data, to the desired test sensitivity, within 24 h using bioluminescent and fluorescent recombinant strains of the test organism Brevundimonas diminuta. These methods should provide a way to implement more rapid QC test regimens for filters.

Product sterility for pharmaceuticals is crucial to guarantee their safe use. However, when heat-labile products, e.g., insulin, cannot be terminally sterilized by autoclaving, microfiltration provides a good noninvasive alternative. Filter failure can result not only in unsafe pharmaceutical and food products but also in lost revenue in industry. The infusion of contaminants, e.g., bacteria and pyrogens, into patients from parenteral products can cause fatalities. Hence, efficient filter validation methods are needed to ensure sterilization efficacy. The aim of this project was to reduce the bacterial challenge integrity test time.

Currently, filters are challenge tested with one of the smallest known bacteria, Brevundimonas diminuta ATCC 19146. The test allows the detection of 1 CFU per filtrate, which may be a large volume. For 0.2-μm-pore-size sterilizing-grade membranes, the filtrate should contain no challenge test organisms, according to regulatory guidelines (9). Challenge testing requires 48 h for colony development, and the time delay creates problems for filter manufacturers. Molecular DNA tests, e.g., PCR, and probe hybridization methods, are useful and rapid for bacterial enumeration, but these methods do not necessarily confirm that the organisms detected are viable, and they are expensive and technically demanding.

Here we report the construction of recombinant B. diminuta carrying genes encoding bacterial luciferase (13) or green fluorescent protein (GFP) (5). These recombinants were tested, in conjunction with various detection systems, for their suitability to improve rapid filter integrity assays.

MATERIALS AND METHODS

Bacteria and plasmids.

Bacteria and plasmids used are shown in Table 1. For cloning experiments all bacteria were grown using Luria broth and Luria agar (16), and all antibiotics were supplied by Sigma (Poole, United Kingdom). The enzymes were supplied by Life Technologies (Gibco BRL).

TABLE 1.

Bacteria and plasmids used in this study

Strain or plasmid	Source or reference	Comments
E. coli DH5α	American Type Culture Collection	supE44 ΔlacU169 (φ80 lacZ ΔM15)hsd R17 recA1 gyrA96 thi-1 relA1 (16)
E. coli S17-1λpir	American Type Culture Collection	RP4; pir lysogen (15)
B. diminuta ATCC 19146	Pall Europe Limited	Filter challenge test strain (2)
pSfi390	S. Swift	pUC18 (16) containing lux genes from P. luminescens ATCC 29999 (13)
pUTLUXAB	G. Stewart	RP4 oriT, R6K plasmid, Tn5::tetracycline resistance cassette containing luxAB genes from Vibrio harveyi (6)
pVAGFP	S. Taylor	Plasmid pVAGFP containing mutant gene for GFP (5)
pBSLLUX2	This study	pBSL204 with luxABCDE genes (13)
pBSLGFP1	This study	pBSL204 with gfp gene (5)
pBSL204	Department of Molecular Cell Biology, Utrecht, The Netherlands	RP4 oriT, R6K plasmid, Tn5::tetracycline resistance cassette (1)

Construction of plasmids.

pBSLLUX2 has the luxABCDE genes (13) from Photorhabdus luminescens from pSfi390 (S. Swift [Nottingham University], unpublished data) cloned into the EcoRI site of pBSL204 (1). pBSLGFP1 has gfp (5) from pVAgfp (S. Taylor [Leicester University], unpublished data) cloned into the BamHI site of pBSL204. All DNA was prepared by the alkaline lysis method (16), using Escherichia coli S17-1λpir as the host for pBSL204 and E. coli DH5α as the host for pSfi390 and pVAgfp. Recombinant plasmids were transformed into E. coli by electroporation (16). Recombinants were selected on Luria agar containing tetracycline (5 μg/ml) and ampicillin (50 μg/ml).

Bioluminescent recombinants were identified with X-ray film (Bio-Rad), for 10 min and fluorescent recombinants were detected on selection plates by visual inspection using epifluorescent microscopy (Zeiss) at ×25 magnification.

The orientation of gene insertion in recombinant plasmids was determined by BstXI restriction digestion and NotI/NdeI dual digestion for pBSLLUX2 and pBSLGFP1, respectively. The BstXI sites in pBSLLUX2 were identified using data from Alexeyev et al. (1) and Meighen and Szittner (13) (GenBank accession number M90092). The NotI and NdeI sites in pBSLGFP1 were identified using data from Alexeyev et al. (1) and Cormack et al. (5) (GenBank accession number U73901). The orientation of the cloned genes was shown to be in the 5′ to 3′ direction with respect to the distal end of the transposon, Tn5. This orientation allows transcription of the cloned genes to be initiated from a B. diminuta promoter after insertion into the chromosome.

Transformation of B. diminuta.

Plasmids pUTLUXAB (6), pBSLLUX2, and pBSLGFP1 were introduced into B. diminuta by filter matings between the donor (E. coli S17-1λpir) and recipient (6). Recombinants were selected on Luria agar containing tetracycline (5 μg/ml) and rifampin (50 μg/ml), and incubation was carried out for 72 h at 30°C. Bioluminescent and fluorescent recombinant colonies were identified as described above. Bioluminescence from recombinants derived from pUTLUXAB matings was induced by addition of 10 μl of 1% (vol/vol) decanal (Sigma) in ethanol to the petri dish lid.

Bioluminescence from recombinants producing the strongest signals on X-ray film was confirmed and quantified by a microtiter plate assay using a Luminoskan RS luminometer (Labsystems, Basingstoke, United Kingdom) in which 10 μl of 1% (vol/vol) decanal in ethanol was added to each well prior to measurement. Fluorescence from derived B. diminuta::pBSLGFP1 recombinants was quantitated using a fluorospectrophotometer (model RF1501; Shimadzu) set to low gain.

Characterization of B. diminuta recombinants.

Biochemical tests on B. diminuta recombinant strains using API 20 NE and API ZYM test kits (Biomerieux) were carried out according to the manufacturer's instructions. Catalase and oxidase tests, Gram staining, and motility tests using the hanging drop method were done (4). All tests were conducted in parallel with B. diminuta ATCC 19146 as the wild-type control.

Cell size was determined by transmission electron microscopy, using the uranyl acetate fixing method (14). Size also was indirectly measured by use of the standard filter challenge protocol (2). For this, all bacterial strains were grown in saline lactose broth (SLB) (sodium chloride [7.6 g/liter] and lactose [0.39 g/liter]) statically for 24 h at 30°C and used to challenge filter membranes (pore sizes, 0.2 and 0.45 μm; Pall) in a sterile Sterilfil filter holder (Millipore). The filtrates were passed through a sterile prevalidated 0.1-μm-pore-size analysis membrane which was incubated on TSA for 48 h at 30°C, as previously outlined (2).

Growth rates of recombinant and wild-type B. diminuta were determined in tryptic soy broth (Oxoid) at 30°C with shaking (200 rpm). Viable counts were determined by serial dilution in 0.9% (wt/vol) saline and plating on TSA.

The bacterial adherence to hydrocarbons hydrophobicity assay (7) was done by estimating the partition of B. diminuta into n-octane. Relative cell net charge was determined according to the electrostatic interaction chromatography method (7) using a CM-S CL-6B cation exchanger and DEAE-S CL-6B anion exchange resin (Pharmacia).

Bioluminescent detection with Nightowl charge-coupled device (CCD) camera.

The B. diminuta recombinant strains (pUTLUXAB and pBSLLUX2) were grown in tryptic soy broth at 30°C for 24 h with shaking (200 rpm). The culture was diluted in saline to 50 bacteria per ml, and various volumes were pipetted into a Sterilfil filter holder (Millipore) containing 10 ml of saline and filtered through a 0.22-μm-pore-size rated Durapore membrane (Millipore), using a vacuum (0.7 × 10⁵ Pa). Each membrane was placed onto TSA, and bioluminescence measurements were taken at various time intervals.

The membranes were scanned for bioluminescence using four combinations of scan settings with a CCD camera (EG&G Wallac Berthold, Wildbad, Germany). In all cases the focusing distance was set to a distance of 77 mm from the membrane and low resolution was selected.

All membranes were scanned for either 1 or 60 min (with and without background subtraction) after 6 to 48 h of incubation on TSA at 30°C. For all detection methods with B. diminuta(pUTLUXAB) light production was induced by colony exposure to 10 μl of 10% (vol/vol) decanal in ethanol added to the petri dish lid. The visible colony counts per test membrane were determined using the naked eye after 48 h of incubation. The detection sensitivity for each membrane was calculated, according to the method outlined by Linardakis and Khatchatryan (11). A true positive was a light signal detected at 24 h and with growth visible by eye after 48 h. A false negative was a colony with no detectable luminescence at 24 h but which was visible at 48 h.

Bioluminescent detection with Nucleovision workstation CCD camera.

B. diminuta(pUTLUXAB) was used to evaluate microcolony detection sensitivity using the Nucleovision workstation (Nucleotech Ltd.). All membrane preparations were done as described for the Nightowl system, except Isopore 0.2-μm-pore-size black membranes (Millipore) were used. Test membranes were incubated on TSA at 30°C for 48 h and were scanned for 1, 5, and 20 min after incubation for 12 to 48 h. All of the scan parameters were those preset for the machine. Blank control membranes were scanned after 12 to 48 h of incubation. Sensitivity was calculated as for the Nightowl system.

Bioluminescent measurement by Bioprobe luminometer.

Test membranes were prepared as described for the Nightowl system. At least 10 membrane samples were aseptically placed on a methanol-sterilized test plate, (Hughes Whitlock, Monmouth, United Kingdom), and readings were taken using a light collection time of 30 s. The number of visible colonies per test membrane was determined after 48 h of incubation. Ten control blank membranes were tested to calculate the background noise limits for the luminometer. Background limits were calculated as the mean luminescence plus two standard deviations (SD).

Epifluorescent detection of B. diminuta(pBSLGFP1) microcolonies.

Test cultures and membranes were prepared as described for the Nightowl, using Isopore 0.2-μm-pore-size black membranes. The membranes were inspected for fluorescent microcolonies using epifluorescent microscopy after incubation at 30°C on TSA for 0, 12, 24, and 48 h. A total magnification of ×25, a mercury vapor light source (Zeiss), and a fluorescein light filter (Zeiss) were used. Sensitivity was calculated as for the Nightowl, except a true positive was a fluorescent microcolony visible after 24 h.

Statistical analysis.

All statistical tests (one-way analysis of variance tests) were performed using Graphpad Instat (8).

RESULTS

Construction of bioluminescent and fluorescent B. diminuta strains.

Plasmids pBSLGFP1 and pBSLLUX2 were constructed, as outlined in Materials and Methods, to enable insertion of gfp and luxABCDE into the B. diminuta chromosome, following filter mating. Ten of each set of B. diminuta recombinants producing the strongest signals on X-ray film were selected for further assay of bioluminescence, as outlined in Materials and Methods. From these experiments two strains that produced the strongest bioluminescence after 24 and 48 h of incubation were chosen for rapid filter challenge testing and were denoted ab5 (pUTLUXAB) and ae3 (pBSLLUX2). Strain ab5 was found to be about 200 times more bioluminescent than strain ae3 after 48 h of incubation.

A pBSLGFP1-derived recombinant was chosen using epifluorescence microscopy. Not all recombinants produced the same level of fluorescence, and therefore the 10 most-fluorescent recombinants were selected for further assay, as outlined in Materials and Methods. The strain producing the strongest fluorescence after 24 and 48 h of incubation was strain gf3. Hence, this strain was chosen for rapid filter challenge testing.

Characterization of B. diminuta recombinant strains.

The recombinant B. diminuta strains were compared to the B. diminuta wild-type strain ATCC 19146 currently used for challenge testing. The recombinant strains ab5, ae3, and gf3 were identical to the wild type in the 43 diagnostic tests described in Materials and Methods. Cell size was determined using transmission electron microscopy. There was no significant difference (P > 0.05) in the sizes of the recombinant and wild-type organisms. Table 2 shows the mean cell dimensions for all B. diminuta strains.

TABLE 2.

Summary of characterization tests^a

B. diminuta strain	Cell length (μm)	Cell width (μm)	Generation time (h)	% Adherence to:
				Octane	DEAE-Sepharose	CM-Sepharose
Wild type	1.05 (0.05)	0.52 (0.09)	0.99 (0.01)	16.7 (2.2)	96.8 (1.5)	32.1 (3.2)
ae3	1.04 (0.05)	0.51 (0.09)	1.01 (0.04)	18.0 (1.4)	90.7 (13.2)	30.9 (1.7)
ab5	1.03 (0.07)	0.42 (0.09)	0.97 (0.06)	17.8 (1.4)	95.9 (1.1)	31.5 (2.2)
gf3	1.02 (0.09)	0.43 (0.05)	0.95 (0.07)	18.1 (2.5)	96.6 (1.6)	30.1 (1.9)

^aData shown are means (standard deviations are given parenthetically). The number of replicates were as follows: 5 for cell length, width, and surface charge assays, 12 for hydrophobicity assays, and 3 for generation time. 

Because cell size can be influenced by the growth rate, the mean generation time was calculated. There was no significant difference (P > 0.05) in the mean generation times of the recombinant and wild-type organisms (Table 2). Net charge and hydrophobicity can affect the adsorption of bacterial cells to filter media, but there was no significant difference (P > 0.05) in cell net charge or hydrophobicity of the recombinant and wild-type organisms (Table 2).

Filter challenge tests.

To demonstrate the suitability of the recombinant strains for filter challenge testing, they were compared under filter challenge test conditions. Pall rated filters (0.2- and 0.45-μm pore size) were challenged with cell suspensions of recombinant and wild-type B. diminuta. The retention efficiency of the strains was compared using the log reduction value. None of the challenges of 0.2-μm-pore-size sterilizing-grade filters yielded bacteria in challenge test filtrates.

When 0.45-μm-pore-size filters were challenged there was no significant difference (P > 0.05) in retention of the different B. diminuta strains. The mean log reduction values from at least 12 experiments were 5.54 ± 1.99, 6.58 ± 4.03, 6.04 ± 2.17, and 6.80 ± 4.76 for strains ab5, ae3, and gf3 and the wild-type, respectively.

Bioluminescence detection methods. (i) Nightowl CCD camera.

With B. diminuta strains ae3 and ab5 and the wild-type B. diminuta, acting as a negative control, four combinations of CCD camera scan parameters were tested, as outlined in Materials and Methods. Bioluminescent microcolonies were displayed by the camera as white dots on a black background (Fig. 1), and the position of these signals was subsequently compared to the position of colonies detectable by eye after 48 h of growth. No bioluminescence signals were detected on the membranes after 6 and 18 h of incubation, whatever the scan parameters. After 48 h of incubation, all colonies visible to the naked eye were detected by bioluminescence; i.e., the sensitivity was 100% for each strain, whatever the scan parameter. Table 3 shows the detection sensitivity for the six test membranes after 24 h of incubation. As can be seen, the detection sensitivity was highest when a scan time of 1 min was used. Other scan parameters did not affect sensitivity. Bioluminescence was not detected on any membranes with the B. diminuta wild-type organism after 48 h of incubation: these two membranes had 6 and 82 colonies.

FIG. 1.

Nightowl scans taken from the same test membrane after 24 (a) and 48 (b) h of incubation, using strain ab5. After 24 h bioluminescent microcolonies were visible as white dots on a black background. In this example 12 signals were detected after 48 h of incubation (b), but after 24 h of incubation only 9 bioluminescent signals were detected (a).

TABLE 3.

Nightowl system detection sensitivity at various scan parameters^a

Background subtraction	Scan time (min)	No. of microcolonies detected after 24-h incubationb	Sensitivity (%)
No	1	3, 4, 7, 8, 9, 18	71.9
Yes	1	3, 4, 7, 8, 9, 18	71.9
Yes	60	3, 3, 4, 7, 5, 12	54.3
No	60	3, 3, 4, 7, 5, 12	54.3

^aThe colony counts after 24 h of incubation as detected by the Nightowl system (pixel setting, 9 by 9) are shown. The colony counts after 48 h of incubation for each of the six membrane samples were 4, 4, 8, 10, 26, and 33 CFU. 

^bA result is given for each of the six membranes. 

Because the B. diminuta strain producing decanal, ae3, was about 200 times less bioluminescent than the strain requiring exogenous decanal, ab5, the detection sensitivity with the two strains was compared using optimal scan parameters. Twelve membrane samples were scanned for bioluminescent microcolonies of strains ae3 and ab5. After 48 h of incubation the sensitivity was 100%. After 24 h, the sensitivity fell below 100%. As can be seen in Table 4, after 24 h of incubation the mean sensitivities SDs were 83.9 ± 18.0% and 77.0 ± 19.3% for strains ab5 and ae3, respectively. There was no significant difference in these means (P > 0.05). From the data in Table 4 it may be concluded that a filter failure resulting in two colonies being detected by growth at 48 h would be detected within 24 h using strain ab5 and the Nightowl system.

TABLE 4.

Detection sensitivities for all detection methods used

Method	No. of visible colonies at 48 h	% Sensitivity at 24 h
Epifluorescence microscopya	76	94.7
	58	100
	43	93.0
	39	94.9
	35	97.1
	29	96.6
	23	100
	21	100
	19	100
	18	100
	15	100
	8	100
	7	85.7
	5	100
	4	100
	3	100
	2	100
	2	100
	1	100
	1	100
Nucleovisionb	66	53.0
	49	65.3
	40	92.5
	35	85.7
	3	42.4
	24	45.8
	21	57.1
	16	87.5
	15	80.0
	14	92.9
	11	90.9
	10	30.0
	10	100
	10	90.0
	7	100
	6	100
	5	80.0
	4	100
	2	50.0
	2	100
	2	100
	1	100
	1	100
Nightowl (ae3)c	66	77.3
	57	73.7
	46	84.8
	26	34.6
	22	72.7
	15	100
	10	50.0
	10	90.0
	9	88.9
	8	87.5
	4	100
	4	75.0
Nightowl (ab5)d	71	81.7
	64	81.3
	33	51.5
	25	84.0
	25	100
	19	100
	12	75.0
	12	100
	6	83.3
	4	100
	2	100
	2	50

^aMean sensitivity, 98.1%. 

^bNucleovision CCD camera (strain ab5). Mean sensitivity, 80.1%. 

^cNightowl CCD camera (strain ae3). Mean sensitivity, 77.0%. 

^dNightowl CCD camera (strain ab5). Mean sensitivity, 83.9%. 

(ii) Nucleovision CCD camera.

To test the performance of the bioluminescent strains with an alternative CCD camera system, a Nucleovision CCD camera was used. No bioluminescence was detected after membranes had been incubated for 12 and 18 h, with scan times of 1, 5, or 20 min. After membranes had been incubated for 24 h, only weak bioluminescence was detected after a scan time of 20 min, due to the accumulation of background noise, and only faint signals were detected using a scan time of 1 min. However with a scan time of 5 min strong bioluminescence signals were detectable. Bioluminescence signals were not detected on any negative control membrane using the B. diminuta wild-type organism after 24 or 48 h of incubation: these two membranes had 10 and 63 colonies.

After 48 h of incubation of membranes, detection sensitivity was 100% when scanned for 5 min. As can be seen in Table 4, within 24 h the required sensitivity to determine a filter challenge test failure was obtained with the two membranes with one microcolony. An example of a scan map obtained after 24 h of incubation is shown in Fig. 2. Table 4 shows that after 24 h of incubation the mean sensitivity ± SD was 80.1 ± 22.7% for all membrane samples. There was no significant difference (P > 0.05) in the sensitivity of detection of strain ab5 by the Nucleovision CCD camera and Nightowl system and of strain ae3 by the Nightowl system.

FIG. 2.

Example of a Nucleovision scan taken after 24 h of incubation of strain ab5. All four colonies detected using the naked eye after 48 h of incubation were detectable at 24 h using the CCD camera.

(iii) Bioprobe luminometer.

A Bioprobe luminometer was also tested. This system was chosen because it has a different principle of measurement for bioluminescence compared to a CCD camera. Here, the bioluminescence from the filter is calculated as a single value rather than as spots of luminescence. The nonbioluminescent B. diminuta wild-type strain gave a luminescence (mean ± SD) of 13.0 ± 5.6 relative light units (RLU), of which was below the background noise limits calculated using the blank membranes after 24 h of incubation on TSA (24.2 RLU). There was a significant correlation (P < 0.0001) between log₁₀ RLU and log₁₀ CFU for both test strains and both incubation times (Fig. 3). After 6 h of incubation, the detection limits were about 1,500 and 2,000 CFU per membrane for strains ab5 and ae3, respectively. Hence, the detection sensitivity needed for filter testing was not achieved after 6 h of incubation. It was possible to detect fewer colonies per membrane after 24 h. The detection limits by measurement of bioluminescence were calculated as about 1 and 4 CFU per membrane for strains ab5 and ae3, respectively. Thus after 24 h of incubation the detection limits for both strains were close to the ideal detection limit for filter challenge testing (one colony per membrane).

FIG. 3.

Example calibration graph for B. diminuta(pUTLUXAB) after 24 h of incubation. The background noise limits are shown as a dotted line on the graph.

(iv) Epifluorescence detection.

Epifluorescence microscopy was used to assess the potential of strain gf3 for microcolony detection. Incubation times of 0 and 12 h were insufficient for microcolony detection at ×25, ×160, and ×400 magnification. All colonies (100%) were detectable at ×25 magnification on membranes after 48 h of incubation. Table 4 shows that after 24 h of incubation the mean sensitivity ± SD was 98.1 ± 3.7%. This was a significantly greater sensitivity than was obtained with strain ae3 with the Nightowl system (P < 0.01) or with strain ab5 with the Nucleovision system (P < 0.05). It was not significantly different (P > 0.05), however, from the sensitivity of detection for strain ab5 by the Nightowl system.

With each of two membranes on which a single colony grew after 48 h of incubation, one microcolony was detected by epifluorescence after 24 h of incubation (Table 4). Thus, it was possible to reach the required detection sensitivity needed to determine filter challenge test failures.

DISCUSSION

These experiments showed that the mini-Tn5 system (1) was suitable for the cloning and expression of genes for bacterial bioluminescence and gfp in B. diminuta. A range of bioluminescence and fluorescence intensities was observed for the recombinant strains, presumably reflecting promoter strength at the insertion site in the B. diminuta genome. The strains producing the strongest intensity of light enabled single-microcolony detection on filter membranes after 24 h of incubation on TSA.

Because phenotypes such as cell size, surface hydrophobicity, and surface charges affect bacterial retention by filters, tests were done to compare these properties in recombinants and wild-type B. diminuta. No significant difference for these phenotypes was observed for the recombinant and wild-type strains, suggesting that all the strains would be comparably retained by filter membranes. This was confirmed as there was no significant difference in the retention of the recombinant and wild-type strains under standard filter challenge test conditions with rated membranes (0.2- and 0.45-μm pore size). These results confirmed that the recombinant strains would be suitable for filter challenge testing.

The aim of the work was to devise a method to detect filter challenge test failures with greater rapidity than the 48 h currently needed using the existing methodology (2). Success would alleviate the time delay imposed on quality control testing and the filter manufacturing process. Previously, Waterhouse (18) achieved the detection of >100 CFU per filtrate within 2 h, by cell staining and epifluorescence microscopy, but greater sensitivity (10 to 100 CFU per filtrate) after 5 h was achieved by ATP luminescence. Although these methods were rapid, the desired sensitivity of 1 CFU per filtrate was not obtained and there was no rapid way to confirm the identity of the test bacterium to detect false positives. In contrast, in the experiments described in this work the desired sensitivity of 1 CFU per challenge test filtrate was obtained using the bioluminescent and fluorescent strains of B. diminuta, using various detection techniques, and furthermore the recombinant B. diminuta strains provided a built-in identification system for the test bacterium. This is because most contaminating bacteria from the environment do not express bioluminescence or are not strongly fluorescent. Furthermore, microcolony detection confirmed that the bacteria detected were culturable and viable, as required for the test (2).

Three commercially available systems were assessed for their suitability for use in filter challenge testing with bioluminescent B. diminuta. There was little to choose between the systems in terms of their sensitivity and rapidity of detection but each system had advantages and disadvantages in their use for these tests. All of the systems were able to detect one to two microcolonies after 24 h, which is a significant improvement in rapidity compared with the standard method. None however achieved satisfactory detection performance at earlier times.

Using the Bioprobe luminometer it was possible only to detect >10³ test bacteria per filtrate on membranes within 6 h, but the desired sensitivity was achieved using strain ab5 only after 24 h of incubation. The Bioprobe luminometer is normally used to detect bacteria by ATP luminescence; however, for these experiments it was applied to the detection of bacterial bioluminescence on membranes and it produced the desired sensitivity after 24 h. The Bioprobe luminometer was not optimized for measuring bacterial bioluminescence; however, its electronics can be adjusted and this may increase the detection sensitivity. One disadvantage of the Bioprobe luminometer was that microcolony enumeration on membranes depended on interpolation from a standard curve. Thus, each area of light detected could not be assigned to a particular microcolony and it was not possible to precisely determine the number of false-positive results; therefore, one is reliant on statistical means to evaluate false positives. This is less of a problem when detection of bioluminescence is by CCD camera. Here the ability to localize a spot of luminescence on scan images allows subsequent assignment of microcolonies to that location, thus enabling precise determination of the number of false positives, albeit only at 48 h posttest. However, this advantage must be balanced against the large difference in the cost of the Bioprobe system and the CCD camera systems.

Two CCD camera systems were evaluated, and there was no statistical difference in their sensitivity of detection of either bioluminescent strain after 24 h of incubation. The Nucleovision system enabled the detection of one organism per filtrate after 24 h compared to two per filtrate using the Nightowl system.

Masuko was able to detect 70% of bioluminescent Photobacterium sp. on membranes after 1 h, using a Hamamatsu Argus 100 VM-3 CCD camera (12). This rapidity was not achieved using the cameras tested here. It is not clear if this was a consequence of comparing naturally bioluminescent bacteria with recombinant bacteria or because of camera sensitivity. The manufacturers of the cameras each use different criteria to test sensitivity, making absolute comparisons difficult. It may be possible in the future to improve rapidity using a CCD camera linked to a microscope. Single cells of bioluminescent bacteria can be detected directly using such a system (17).

As an alternative to bioluminescence, GFP fluorescence was assessed as a means of rapid microcolony detection. The notable feature of this system was that higher detection sensitivity (98.1%) was achieved by epifluorescence assessment after 24 h than with the CCD camera system. The higher sensitivity of the epifluorescence method seems to be a consequence of the capacity to magnify and directly visualize fluorescent microcolonies by the operator, allowing more accurate assignment of a light signal as being derived from a colony. Epifluorescence required no interpretation of scan maps and no setting of scan parameters and made use of an epifluorescence microscope, which is commonly found in microbiology laboratories. The disadvantages of this method over the bioluminescence method were the longer times required to acquire the data and consequent operator fatigue. In the future, it may be possible to increase the rapidity of detection by using alternative methods, for example, flow cytometry, which has been used for fluorescent cell detection to a sensitivity of 100 gram-negative cells per ml in 30 min (3), or confoccal microscopy (10).

In conclusion use of any of the recombinants produced significant savings, in the time required to obtain an indication of filter failure, over the standard method. On balance use of the B. diminuta strain gf3 combined with fluorescence microscopy was considered the best method for rapid filter retention testing. The method was rapid, accurate, and inexpensive. Each of the bioluminescence systems produced acceptable performance, but the Nightowl system was much more expensive than the other two, with no apparent performance gain. Of the bioluminescence systems the Nucleovision system is favored as a laboratory system on the basis of sensitivity and ease of use for the operator. Finally, one advantage of the Bioprobe system that should be considered is its portability, which enables it to be used outside the laboratory—a feature that none of the others offers.