PAPER Ital. J. Food Sci., vol. 27 - 2015 191 - Keywords: Bactoscan FC, conversion line, cow milk, total bacterial count - NEW NATIONAL CONVERSION LINE FOR BACTOSCAN FC IN ITALY: A STEP FORWARD G. BOLZONI1*, A. MARCOLINI1, G. DELLE DONNE1, B. APPICCIAFUOCO† and A.M. FERRINI † 1National Reference Centre for Bovine Milk Quality, IZSLER, Brescia, Italy †National Reference Laboratory for Milk and Milk Products, Istituto Superiore di Sanità, Roma, Italy *Corresponding author: Tel. +39 030 2290541, Fax +39 030 2290537, email: giuseppe.bolzoni@izsler.it ABSTRACT To improve the reproducibility of flow cytometry technique for total bacterial count in milk, a conversion from instrumental results (impulses/μL) to the reference method resultes (cfu/mL) is needed. In 2008 in Italy, a project for a common conversion line for Bactoscan FC was initiated. In this paper we report on the second phase of the project focusing on the statistical procedure used to evaluate the validity of the data. The new conversion line, representative of national milk (2,732 valid samples from 29 labs) obtained from both rounds of the study is: Log 10 (cfu mL-1) = Log 10 (IBC µL-1) x 0.939 + 2.559, with S y:x = 0.282 with an application range up to 70,000 IBC µL-1. mailto:giuseppe.bolzoni%40izsler.it?subject= 192 Ital. J. Food Sci., vol. 27 - 2015 INTRODUCTION Regulation (EC) 1664 (EC 1664:2006) estab- lished that the reference method for determin- ing total bacterial count at 30°C in raw milk is EN ISO 4833 (ISO 4833:2003), however the use of alternative methods is acceptable when they are validated against the reference method in accordance with the protocol set out in EN/ISO standard 16140 (ISO 16140:2003) or other sim- ilar internationally-accepted protocols. In the case of milk, ISO 21187 (ISO 21187:2004) and ISO 16297 (ISO 16297:2013) are examples of other such protocols. EN ISO 4833 instructs that colonies grown in defined conditions must be counted after 72 h of incubation at 30°C whereas flow cytome- try instruments count free cells independent- ly from their physiological status or their ca- pability to develop into a colony. The counts are obtained from electrical impulses (derived by the fluorescence of bacterial DNA and RNA stained by fluorochrome ethidium bromide) and must be converted into cfu mL-1 equiva- lents, as this is the regulatory unit of measure. This conversion (when calculated by a single laboratory) is the main reason for the low re- producibility of the alternative method in spite of its otherwise better repeatability, rapidity and cost effectiveness compared to the refer- ence method and this could have major con- sequences both from economic and food safe- ty points of view. Currently the flow-cell auto- matic instruments for total bacterial count are indispensable to the centralized and special- ized laboratories in charge of large numbers of milk samples per day. For this reason, at the end of 2008, the Reference Centre for Bovine Milk Quality of IZSLER launched a project for a “common conversion line” for Bactoscan FC (Foss, DK), the most commonly used instru- ment in Italy. The result of that study (BOL- ZONI and MARCOLINI, 2010) was adopted on a voluntary basis by several laboratories in our country in the last few years. In 2012, with the coordination of the Italian National Reference Laboratory (NRL) for milk (Istituto Superiore di Sanità), a second round of the project was developed with the objectives to: verify the re- sults of the first round of the project; study a wider range of milk contamination levels; de- rive a conversion formula that is more tailored to Italian milk, meaning a single, mandatory conversion formula to be applied at the nation- al level; propose a statistical model to evalu- ate the reliability of the raw data. MATERIALS AND METHODS The study involved 29 laboratories from all over Italy. The number of samples analyzed from each laboratory and for the different levels of contamination was determined on the basis of their previous participation or not in the first round of the project in 2008 (Table 1). The protocol adopted in 2008 (BOLZONI and MARCOLINI, 2010) was adopted again in 2012 with the intention of producing compa- rable data. Participating laboratories, during the period from January to June 2012, se- lected samples of cow bulk tank milk (refrig- erated and without preservatives) from those submitted for daily analytical activity. The in- strument’s calibration status was checked through an inter-laboratory trial using lyophi- lized milk samples at 3 different contamina- tion levels that were shipped to participants (data not shown). Considering that ISO/TS 19036 (ISO 19036, 2006) estimates that the standard deviation for aerobic mesophilic flo- ra in milk (S R = 0.12) is affected more by oper- ative conditions (S cond = 0.09) than by the ini- tial suspension (S IS = 0.04), it was decided that the reference method would be performed us- ing 2 plates per dilution with one series of di- lutions. In each laboratory, immediately before analysis, each sample was mixed as stated in ISO 6887-5 (ISO 6887-5, 2010), tested in du- plicate by the Bactoscan FC and immediate- ly analyzed by the reference method. A single series of at least 3 decimal dilutions was pre- Table 1 - Selection of samples – percentage of samples and respective ranges of impulses required from each lab. Range Impulses % samples analyzed % samples analyzed (IBC µL-1) (from 10 to 50 samples)A (from 50 to 100 samples)B 0-20 3 3 21-100 30 10 100-1,000 30 10 1,000-5,000 25 10 5,000-10,000 4 30 10,000-50,000 4 27 50,000-99,999 4 10 A: Laboratories WITH participation in the project prior to 2009. B: Laboratories WITHOUT participation in the project prior to 2009. Ital. J. Food Sci., vol. 27 - 2015 193 pared with quarter-strength Ringer’s solution (the level of dilution was established on the basis of the previous instrumental results); 1 mL of each dilution was dispensed in each of 2 plates of milk-PCA medium and then incu- bated at 30°C ± 1°C for 3 days. Each partici- pating lab contributed their data on the Bac- toscan FC double counts in “impulses” (IBC µL-1) and colonies counts from the two plates of each dilution to a database. After the rele- vant controls of raw data (see: point “d” in the “selection of results” section below) as indicat- ed by ISO 7218 (ISO 7218:2010; ISO 14461- 2:2005) and the additional controls (see: points “e” and “f”), the linear mixed effect model (LME) was applied to produce the regression line of the data from the “valid samples”. The statis- tical evaluation of the results is described in the following section. The software “Procedure R 2.15” and Excel 2010 (Microsoft Corp., Red- mond, WA) were used. RESULTS AND DISCUSSION Range of measurement and linearity The ratio between observed values (O.V.) and expected values (E.V.) in impulses µL-1 (IBC µL-1) from serial dilutions of ad hoc heavily contami- nated milk samples was taken as an indicator of linearity of the instrumental signal response. The ratio O.V./E.V. ~ 1 (Fig. 1) suggests the ac- ceptable instrumental linearity continues up to 50,000 IBC µL-1, which is well above the produc- er’s declared limit of 30,000 IBC µL-1 and con- firms our previous evaluation (BOLZONI et al. 2000, BOLZONI et al., 2001). Since one of the aims of the work was to eval- uate whether a broader range of instrumental measures could be accepted without affecting the conversion line, values > 30,000 IBC µL-1 were also considered. Ratio O.V./E.V. = 0.9 was adopted as an arbitrary lower limit of accepta- bility of the linearity indicator (equivalent to 3 standard deviations from the mean of the ra- tios obtained). These considerations allowed us to accept 70,000 IBC µL-1 as the upper limit for the range of application of the conversion line. We would like to note that samples with IBC µl-1 > 30,000 (approximately > 4,000,000 cfu mL-1) are rather unusual in Italy. Selection of results Of the 1,827 total milk samples analyzed, which is equivalent to more than 10,000 ana- lytical results produced by 29 participating lab- oratories, the selection process for valid data led to the rejection of 499 (27%) samples due to the following factors: a) Unreliability – 19 samples were eliminat- ed for absence of correspondence between the instrumental results and the reference meth- od results or errors in the report transmission results. b) Out of range of measurement – 65 samples were eliminated because their values were out- side the established range of linearity (12 sam- ples lower than 10 IBC µL-1 l and 53 higher than 70,000 IBC µL-1). c) Instrumental repeatability – 31 samples were eliminated because the difference between repli- cates exceeded the repeatability limit of the Bac- toscan FC: Critical Log Difference between rep- licates > 2.83 S r (P 95%). Additionally 12 sam- ples were eliminated because they exceeded the instrumental reproducibility limit (S R ). d) Maximum - minimum numbers of colonies on the plates and proportionality between dilutions – plates outside the range 10 - 324 colonies were not considered for the count (ISO 7218:2007). The G2 factor test, which compares the relation- ship between pairs of plates and dilutions, led to the elimination of 179 samples. e) Sub-dispersion of reference method results - no laboratories were eliminated on this basis (which compares the relationship between ob- served and expected values on plates) but the frequency of sub-dispersed samples was one cri- terion used for the selection of laboratories de- scribed in point f. f) Single laboratory performance evaluation – the effect of each individual laboratory on the extrapolation of the final regression line was considered on the basis of the following factors: – excessive or insufficient dispersion of the in- dividual lab’s regression line; – high frequency of sub-dispersed results from the reference method; – high frequency of eliminated results from the G2 factor test. The dispersion of data around single-lab re- gression lines is reported in Table 2 as S y:x . Giv- Fig. 1 - Bactoscan FC linearity: the relationship between the observed values and the ratio of the observed value to the expected value. 194 Ital. J. Food Sci., vol. 27 - 2015 Table 2 - Dispersion of the conversion line for individual laboratories (S y:x). Lab Code Samples (n) Intercept Slope Sy:x 40 50 2.1184 1.0309 0.0139 27 98 2.9025 0.7797 0.0930 14 42 2.4432 0.9911 0.1455 38 36 2.2563 1.0279 0.1577 31 93 2.3363 1.0408 0.2517 35 52 2.1976 1.0711 0.2556 41 16 2.1718 1.0859 0.2594 1 40 2.1280 0.9966 0.2676 39 88 2.6219 0.8914 0.2766 15 26 2.5538 1.0119 0.3086 11 26 2.5408 0.9711 0.3118 23 98 2.6394 0.9257 0.3223 24 50 2.4829 0.9593 0.3291 6 68 2.2774 1.0927 0.3365 28 54 3.5260 0.6413 0.3546 37 79 3.6620 0.5508 0.3707 22 76 2.7561 0.8592 0.3756 26 55 2.1806 0.9484 0.3766 7 22 2.4747 1.0033 0.3830 29 24 2.7238 0.9293 0.3893 33 89 3.0733 0.6950 0.4104 34 103 2.1782 1.2029 0.4145 25 97 2.8959 0.7690 0.4286 30 36 2.8759 0.7964 0.4379 8 34 2.6250 0.9531 0.4410 32 29 3.1796 0.6974 0.4567 9 30 3.0099 0.8643 0.6225 36 32 2.5325 0.6897 0.6386 21 110 3.1939 0.8881 0.8504 en S y:x < 0.40 is a criterion for acceptability (listed as a “tentative value” in ISO 16297:2013), nine of twenty-nine labs were over range. Of the nine, six were considered borderline and only labora- tories 21, 36 and 9 were eliminated for over-dis- persion. Furthermore laboratory 40 was elimi- nated for sub-dispersion, which suggested their results were not completely reliable. Two labo- ratories exhibited a high frequency of eliminat- ed samples by the G2 factor test (> 50% of sam- ples); in the first case we decided to eliminate all results (Lab 36, which had already been elimi- nated for high dispersion as mentioned above), whereas in the second case (Lab 28) we decided to preserve the remaining “valid results” consid- ering the very low value of dispersion of its re- gression line (0.3546 S y:x ). Evaluation of the regression line The LME model was applied to produce the regression line of the selected 1,388 valid sam- ples. Multi-step selection of outliers (residu- al standard deviation > 2.58) was preliminar- ily applied (ISO 21187:2004). In synthesis, af- ter a 3-step sequential elaboration, 65 outliers were eliminated, narrowing the number of val- id results to 1,323 and improving the S y:x value from 0.3547 to 0.2781. After the third step, no significant improvement in the level of estima- tion could be obtained so no further elimination of data was considered appropriate. The following conversion equation was calcu- lated from the 1,323 residual samples (charac- teristics of the conversion equation are report- ed in Table 3): Log 10 (cfu mL-1) = Log 10 (IBC µL-1) x 0.946 + 2.569 Fig. 2 shows the conversion line from 2012 alongside the conversion line from 2009 (black dashed line) (6), calculated by: Log 10 (cfu mL-1) = Log 10 (IBC µL-1) x 0.911 + 2.599 The conversion line from 2012 is very sim- ilar to the line from 2009 although differenc- es are seen at high and very high contamina- Table 3 - Characterization of the conversion line from 2012. Parameters Coefficient St. error T Sig Low High Intercept 2.569 0.038 67.57 0.000 2.493 2.645 Slope 0.946 0.009 106.91 0.000 0.928 0.964 Number of samples = 1,323; S y:x = 0.278. Fig. 2 - Distribution of data from the 2012 conversion line compared with the 2009 conversion line. Ital. J. Food Sci., vol. 27 - 2015 195 tion levels as a consequence of the extension of the measurement field in the second round of the project. In Figs. 3 and 4, the distribution of random effects in the LME for data from individual lab- oratories is presented. Statistically four labs were found to be apparently different from the others: numbers 6 and 34 overestimated their counts while numbers 1 and 26 underestimated their counts. No factors affecting this distribu- tion could be identified (e.g. bacterial flora, sam- ple characteristics, or systematic bias in refer- ence method execution), so the data from these labs were kept in the regression line calculation. New national conversion line Considering that the same procedure and the same statistical evaluation were used in both rounds of the project, we considered it not only possible but also appropriate to pool the valid results from 2009 and 2012 and to run a new mixed statistical evaluation. Taking a step back before the respective outliers were excluded, a new multi-step selection was performed on the 1,474 valid results from 2009 combined with the 1,388 from 2012. The total elimination of 130 samples at the third step of selection led to no further increase in estimation (Table 4). The final regression line was computed from 2,732 samples and it is represented by the equa- tion: Log 10 (cfu mL-1) = Log 10 (IBC µL-1) x 0.939 + 2.559 The characteristics of the combined regression line are reported in Table 5 and Fig. 5. Fig. 4 - Distribution of random effect coefficients from the labs compared with a normal distribution (2012 conver- sion line). Fig. 3 - Q-Q plot of random effects from each laboratory in the Linear Mixed Effect Model Table 4 - Multi-step selection of outliers on 2009 and 2012 aggregated data. Step No. Samples (n) S y:x Intercept Slope Min Std Max Std Residual Residual 1 2,862 0.3533 2.591 0.921 4.503 -5.798 2 2,793 0.3048 2.575 0.931 2.989 -3.103 3 2,752 0.2886 2.565 0.937 2.707 -2.691 4 2,732 0.2821 2.559 0.939 2.651 -2.645 5 2,724 0.2796 2.558 0.939 2.660 -2.597 6 2,718 0.2778 2.557 0.939 2.620 -2.590 Table 5 - Characterization of the new national conversion line (2009 and 2012 pooled results). Parameters Coefficient St. error T Sig Low High Intercept 2.559 0.032 80.77 0.000 2.496 2.622 Slope 0.939 0.006 150.38 0.000 0.927 0.952 Number of samples = 2,732; S y:x = 0.282 196 Ital. J. Food Sci., vol. 27 - 2015 alyzed and their results should be entered into the geometric mean of the last three months, as per the calculation system defined by Reg. EC 853:2004. The present project led to the creation of a conversion relationship between impulse µl-1 and cfu mL-1for the enumeration of the total bacte- rial counts in Italian raw cow milk using a Bac- toscan FC. In summary the conversion line in- corporates the following points: – the conversion relationship was constructed according to ISO 21187:2004; – the level of accuracy obtained was satisfacto- ry (S y:x = 0.282 log 10 ); – the number of samples was representative of Italian milk production variability; – 80% of all Italian laboratories involved in milk control by routine method joined the project. The new conversion line appeared robust and representative of milk quality and variety in It- aly, with a range of application up to 70,000 IBC µL-1. It was ultimately validated and adopted as the national conversion line in Italy. This is an im- portant advance for both the industry and pub- lic hygiene because the use of a unique conver- sion line should significantly improve the repro- ducibility of the bacterial count results obtained by Bactoscan FC in Italy. In addition, the use of the conversion line for highly-contaminated samples is a further contribution to improve an- alytical harmonization. Data quality control was focused on the evaluation of data entry quality and consequently the accuracy and robustness of the elaborated conversion line. This was done by checking the raw data (agreement between pairs of plates, and proportionality between suc- cessive dilutions). REFERENCES Bolzoni G., Marcolini A. and Varisco G. 2000. Evaluation of the Bactoscan FC. 1. Accuracy, comparison with Bactos- can 8000 and somatic cells effect. Milchwissenschaft(2), 67-70. Bolzoni G., Marcolini A. and Varisco G. 2001. Evaluation of Bactoscan FC. Second Part: Stability,Linearity, Repetea- bility and Carry-over. Milchwissenschaft, 56(6), 318-321. Bolzoni G. and Marcolini A. 2010. Bactoscan FC - project for unified conversion line in Italy. Milchwissenshaft, 65(3), 309-310. International Organization for Standardization. 2003. EN ISO 4833:2003 Microbiology of food and animal feeding stuffs -- Horizontal method for the enumeration of mi- croorganisms -- Colony-count technique at 30 degrees C. International Organization for Standardization. 2003. ISO 16140: 2003 Microbiology of food and animal feeding stuffs -- Protocol for the validation of alternative methods. International Organization for Standardization. 2005. ISO 14461-2:2005 (IDF 169-2: 2005) Milk and milk products -- Quality control in microbiological laboratories -- Part 2: Determination of the reliability of colony counts of paral- lel plates and subsequent dilution steps. International Organization for Standardization. 2006. ISO/ TS 19036:2006 Microbiology of food and animal feeding stuffs - Guidelines for the estimation of measurement un- certainty for quantitative determinations. Fig. 6 - Distribution of random effect coefficients from labs compared with a normal distribution (new conversion line). Fig. 5 - The new national conversion line (computed from 2009 and 2012 aggregated data). The distribution of random effects in the LME for individual laboratory data is presented in Fig. 6. CONCLUSIONS During the first round, in 2008, the main fo- cus was on milk samples with total bacterial counts around 100,000 cfu mL-1, the European Legal Limit for compliance (Reg. EC 853:2004). Contrastingly, during the second round, labo- ratories were invited to include milk samples with high to very high levels of bacterial con- tamination in order to test the instrumental re- sponse to bacteria levels outside of the linear range indicated by the Bactoscan FC produc- ers. In routine situations the submission of very highly contaminated samples is a rare occur- rence, however they should nonetheless be an- Ital. J. Food Sci., vol. 27 - 2015 197 International Organization for Standardization. 2006. UNI EN ISO 21187:2004 Milk -- Quantitative determination of bacteriological quality -- Guidance for establishing and verifying a conversion relationship between routine meth- od results and anchor method results. International Organization for Standardization. 2007. Uni EN ISO 7218: 2007 microbiology of food and animal feed- ing stuffs – general requirements and guidance for mi- crobiological examination. International Organization for Standardization. 2010. ISO 6887-5: 2010 Microbiology of food and animal feeding stuffs -- Preparation of test samples, initial suspension and decimal dilutions for microbiological examination - Part 5: Specific rules for the preparation of milk and milk products. Paper Received April 10, 2014 Accepted August 6, 2014 International Organization for Standardization. 2013. ISO 16297/ IDF 161: 2013 Milk - Bacterial count - Protocol for the evaluation of alternative methods. The European Parliament and the Council. 2004. Regu- lation (EC) No 853/2004 of the European Parliament and of the Council of 29 April 2004 laying down spe- cific hygiene rules for food of animal origin. OJ L 139, 30/04/2004, 55-205. The European Parliament and the Council. 2006. Commis- sion Regulation (EC) No 1664/2006 of 6 November 2006 amending Regulation (EC) No 2074/2005 as regards im- plementing measures for certain products of animal ori- gin intended for human consumption and repealing cer- tain implementing measures. OJ L 320, 18/11/2006, p. 13-45.