Evaluation of three Brettanomyces qPCR commercial kits : results from an interlaboratory study

a Univ. Bourgogne Franche-Comté, AgroSup Dijon, PAM UMR A 02.102, F-21000 Dijon, France Institut Universitaire de la Vigne et du Vin, Equipe VAlMiS, rue Claude Ladrey, BP 27877, 21078 Dijon, France b Inter-Rhône, service technique, 2260 route de grès, 84100 Orange, France c Microflora – Institut des Sciences de la Vigne et du Vin, Unité de Recherche Œnologique EA 4577, Association pour le Développement de l’Enseignement et de la Recherche en Aquitaine (ADERA), 210 chemin de Leysotte, CS 50008, 33882 Villenave d’Ornon cedex, France


Introduction
The yeast Brettanomyces bruxellensis is a source of wine spoilage, especially in red wines.Compounds naturally present in grape juice and wine that originate from the grapes are generally esterified (p-coutaric, fertaric and caftaric acids) (Dugelay et al., 1993).Enzymes with an esterase activity can release the free form of the acids (p-coumaric, ferulic and caffeic acids) (Gerbaux et al., 2002).These acids are decarboxylated to vinylphenols by yeast, fungi and lactic acid bacteria.B. bruxellensis is able to reduce these vinylphenols to ethylphenols leading to an unpleasant taste (Chatonnet et al., 1997(Chatonnet et al., , 1995)).
In this study, we report on the variability of three qPCR kits designed to quantify B. bruxellensis in red wine based on data generated from three laboratories specialized in wine analyses.Each laboratory followed the commercial protocols to generate comparable data.The laboratories used standardized protocols and the same batch of DNA isolation and amplification reagents to limit variability.

Participants
Three laboratories were selected for participation: the VAlMiS lab (Dijon, FR), the Microflora-ISVV lab (Bordeaux, FR) and the Inter-Rhône lab (Orange, FR).The names of the laboratories were randomly codified (lab 1, lab 2 and lab 3).

Conditions
Each laboratory used a different strain of B. bruxellensis to artificially contaminate red wine.The Inter-Rhône, Microflora, and VAlMiS labs used the strains GSLEV17, CRBO LO417 (Centre de Ressources Biologiques Oenologiques, ISVV, Villenave d'Ornon, France), and LO2E6, respectively, and red wine of each region was inoculated at four different population levels: 10 2 , 10 3 , 10 4 , and 10 5 cells/mL, referred to as levels 1, 2, 3, and 4, respectively.The cells were adapted to ethanol by growing them on YPD agar (10 g/L yeast extract, 20 g/L bacto-peptone, 20 g/L glucose, 20 g/L agar) supplemented with 5% (v/v) ethanol at 28°C for 5 days.Stationary phase cells were used to inoculate diluted red wine (50% red wine: 50% physiological saline water) and incubated for one week at 28°C.The adapted cells were then used to inoculate filter sterilized red wine from each region.
Moreover, five naturally contaminated wines from different wineries were also analyzed by the three laboratories.

b. Alternative method
The alternative method consisted of qPCR performed with commercial kits.Three different commercial kits (arbitrarily named Kit 1, Kit 2, and Kit 3) were tested for B. bruxellensis DNA extraction and amplification.DNA extraction and amplification protocols were performed according to the manufacturers' instructions.Standard curves were used in two kits, whereas one kit allowed direct quantification based on the amplification of reference DNA.Each laboratory used two kits, i.e. each sample was analyzed by the same kit in triplicate by two different laboratories.Amplification reactions were performed on a CFX96 real-time PCR system (Bio-Rad) for two laboratories and on an iCycler IQ5 system (Bio-Rad) for the third.Results were analyzed using Bio-Rad CFX Manager® software.The PCR cycle where fluorescence first occurred (quantification cycle: C q ) was determined automatically after setting the regression method.
Red wines from Côtes du Rhône, Burgundy, and Bordeaux were supplemented with a high level of molecular sulfite (approximately 2 mg/L mSO 2 ) to determine whether the kits quantify dead B. bruxellensis.Cell quantifications using the three kits were performed after two weeks.The total B. bruxellensis population and culturable cells present in the red wines were determined by flow cytometry (FCM) or microscopy methods and plate counting, respectively.Viability was determined by FCM.Solutions containing cells used to determine total populations were stained with dyes (propidium iodide and fluorescein diacetate) according to the protocol described in the study of Salma et al. (2013).

Construction of the accuracy profiles and statistical processing
The construction of the accuracy profile was performed as described by Boubetra et al. (2011).The acceptability criterion was defined as ± 0.3 log units/mL for the alternative method in our study.Target values, based on the median values obtained using the reference method, were determined.The results were generated using the alternative method, and the reproducibility standard deviation (SD) (s R ), the limits of acceptability (λ), the proportion of βexpectation tolerance interval (β), and the difference between the level determined by qPCR and the target value (Bias) were determined for each inoculation concentration.The accuracy profiles were constructed using these results.
The reproducibility standard deviation (SD) (s R ) is calculated based on the SD between triplicates (s r ) and SD between labs (s L ): A β expectation tolerance interval (β-ETI) is defined as an interval that covers an average percentage of a For example, a β-ETI can claim to contain 80% of future measurements, on average.A β-ETI can be expressed as: where k M is the coverage factor, given by the equation: where s r is the repeatability standard deviation, Qt the percentile of a Student t test distribution, β the chosen probability (80% in this study), I the number of laboratories, J the number of replicates, v the number of degrees of freedom, and G given by the equation: where 2 is the reproducibility variance, and s r 2 the repeatability variance.The number of degrees of freedom, v, is given by the equation: where i is the number of laboratories performing the analysis (i < I).In our study, i was equal to 2.

Reference results
We calculated the reference values, also called target values, for each level of contamination from the median values obtained using the reference method (plate counting).Table 1 shows the theoretical values (10 2 , 10 3 , 10 4 , and 10 5 cells/mL).B. bruxellensis populations counted by plating were very close to the expected cell population except for one wine.The Bordeaux wine had a population that was lower than  the theoretical values due to the inability to obtain a high cell concentration for this B. bruxellensis strain in this wine and a likely decrease in viability after inoculation.

Linearity
Linearity of the results for each wine and each kit was determined by plotting the logarithmic results obtained by plate counting (mean of the three labs) against the values determined using the qPCR commercial kits.The correlation coefficient (r²) values are shown in Table 2.The mean r² values were 0.9558 ± 0.0471 for Kit 1, 0.8934 ± 0.0820 for Kit 2, and 0.9094 ± 0.0670 for Kit 3.With four population levels, the degree of freedom is equal to 2 for this statistical analysis.With a risk of error (α) of 10%, the critical r² value is 0.9.Eight of 18 results were not valid (r² < 0.9) (Table 2).

Validation criteria and statistical results
Counts obtained using the alternative method are presented in log 10 units.The validation criteria and statistical results for the Côtes du Rhône, Burgundy, and Bordeaux wines are shown in Supplementary data 1, Supplementary data 2 and Supplementary data 3, respectively.
Repeatability was calculated for each wine according to population level and kit.The mean repeatability was 0.257, 0.183, and 0.390 log 10 cells/mL for Kits 1, 2, and 3, respectively.Kit 1 underestimated the four population levels in Burgundy wine by a mean of 1.2 log 10 cells/mL.The population levels determined by Kit 1 red wines were overestimated by 0.5 and 1.1 log 10 cells/mL relative to the reference method.
Kit 2 underestimated all population levels in the Burgundy and Côtes du Rhône wines (bias of -2.3 and -0.9 log 10 cells/mL, respectively).This kit also underestimated the lowest three population levels in the Bordeaux wine, with a bias of -0.9 log 10 cells/mL, whereas the highest population level had a bias of approximately 0.6 log 10 cells/mL.Kit 3 also led to an underestimation of all population levels in the Burgundy wine (mean bias of 0.8 log 10 cells/mL).This kit also underestimated two population levels in the Bordeaux and Côtes du Rhône wines, with a bias of -0.2 and -1.8 log 10 cells/mL, respectively, whereas two others were overestimated by a mean of 1.1 and 0.9 log 10 cells/mL.
In summary, we could not establish any relation between the population level and the reproducibility or bias.
We compared the reproducibility standard deviation (SD) and absolute bias between the kits (Table 3).All absolute values for the reproducibility SD were high (from 0.4 to 1.1 log 10 cells/mL).The best bias was 0.5 and the highest was 2.3 log 10 cells/mL.Based on these results, no kit precisely quantified B. bruxellensis levels in red wine because the reproducibility SD and bias exceeded the acceptability limits.B. bruxellensis levels were frequently underestimated, highlighting the imprecision of this contamination measure and the risk of obtaining false negative results.

Accuracy profiles
We generated accuracy profiles to visualize the level of imprecision in quantifying B. bruxellensis levels in red wine.Examples of the accuracy profiles for the Côtes du Rhône wine calculated from the results obtained using Kits 1, 2, and 3 are presented in Fig. 1a, 1b, and 1c, respectively.The acceptability limit for this study was ± 0.3 log 10 cells/mL.This value was the maximum acceptable limit and the performance of each kit was tested to determine whether the alternative method is at least as good as the reference method.For this wine, only two values obtained using Kit 1 (level 1) and Kit 2 (level 4) were within the acceptable limits.However, the β-ETIs  were not within these limits, meaning that there is an 80% probability that future analyses will be outside these limits.
The accuracy profiles for the Burgundy red wine are presented in Supplementary data 4.Only two values were within the acceptable limits (levels 3 and 4 with Kit 3) but the bias (β-ETIs) was outside the acceptable limits.
The accuracy profiles for the Bordeaux wine are presented in Supplementary data 5. Two values obtained with Kit 2 were within acceptable limits (levels 1 and 2) as were two values obtained with Kit 3 (levels 2 and 3), but, as above, the β-ETIs were not within these limits.

Analysis of dead yeast
We performed trials with the commercial kits on cells subjected to sulfite treatment to test whether overestimation of B. bruxellensis populations may be due to the quantification of dead yeast.The results are shown in Table 4. None of the three wines contained culturable B. bruxellensis.Only the Côtes du Rhône wine contained a viable population of B. bruxellensis, whereas the red wines from Burgundy and Bordeaux did not, validating the cell death caused by the sulfite treatment.
The quantification results using the kits were precise for the same kit and sample (repeatability) for the three red wines.
The viable population in the Côtes du Rhône wine was higher than the culturable population, probably due to viable but non-culturable (VBNC) cells.Kits 1 and 3 led to an overestimation of 1.7 and 0.9 log 10 cells/mL, respectively, relative to the viable population determined by FCM combined with viability staining.Such overestimation may come from the quantification of dead cells.
The population levels determined for the Burgundy wine from Kits 2 and 3 were approximately identical to the total population (dead cells).
The quantification of B. bruxellensis in the Bordeaux wine by Kit 1 largely underestimated the population (-3.7 log 10 cells/mL), whereas Kit 2 led to an overestimation of the population (+0.6 log 10 cells/mL).Only two results for Wine 1 were not significantly different from the enumeration results by plate counting (from Kit 1 performed by lab 3 and Kit 3 performed by lab 1) (Table 5).No culturable cells were detected in Wine 2, whereas there were two positive results from the kit quantifications.The results for Wine 3 were significantly identical to plate counting when the quantifications were performed by lab 3 with Kit 1 and lab 1 with Kit 3. Two quantifications were significantly different from the populations determined by plate counting for Wine 4 (Table 5).Wine 5 did not contain culturable B. bruxellensis, whereas four kit-based quantifications were positive.

Quantification of B. bruxellensis in five
The results of B. bruxellensis quantification of naturally contaminated red wines validate the previous results performed in artificially contaminated red wines.The were similar or the population was underestimated when the yeast was detected by plate counting.No significant overestimation was made for these red wines.

Discussion
Accuracy profiling was applied to analyze an alternative method against the reference method.In our study, the reference method chosen was plate counting on selective medium as this approach is the most widely used by enological laboratories to study the culturability of this yeast.Three commercial kits that quantify B. bruxellensis in red wine were used as the alternative method.
Using a β of 80% and a λ equal to ± 0.3 log 10 cells/mL, none of the kits were validated because the level of B. bruxellensis determined by these kits was under or overestimated with a bias that was generally higher than the acceptable limit.Moreover, the predicted results resulted in a large discrepancy, leading to a large incertitude of future quantifications.However, the quantification results were precise for the same kit and sample (repeatability).
Using the results of the accuracy profile, a correction factor can be applied if, for example, all results are     slightly and repeatedly overestimated according to the population levels.However, in our study, the kits sometimes overestimated the population at one level and underestimated it at another.We observed no continuous error between the alternative and reference methods, making it impossible to apply a correction factor to the results.Moreover, as the tolerance intervals were higher than the acceptability limits, no quantification limit could be determined.Overestimation of B. bruxellensis using the qPCR kits may be due to the presence of VBNC cells in the wine which may not be detectable by plate counting (Du Toit et al., 2005;Millet and Lonvaud-Funel, 2000;Serpaggi et al., 2012).As we show here, it may also be due to the fact that the kits do not discriminate live from dead or VBNC cells, confirming previous studies (Andorrà et al., 2010;Vendrame et al., 2014;Willenburg and Divol, 2012).Propidium monoazide (PMA) and ethidium monoazide bromide can discriminate between live and dead microorganisms (Andorrà et al., 2010;Rizzotti et al., 2015;Vendrame et al., 2014) and could be used in this context.
The best solution to prevent underestimation is the use of an internal control.The internal control is often a microorganism not found in wine (Longin et al., 2016;Tessonnière et al., 2009) and added to the samples at a known concentration before DNA extraction.If, for example, Yarrowia lipolytica is added as an internal control (Tessonnière et al., 2009)  .Although longer than commercial kit protocols, it has been shown to be sensitive and efficient.However, in this protocol, the target DNA corresponds to the RAD4 gene.Thus, the amplification of this gene after cell death needs to be assessed to prevent overestimation.Alternatively, the commercial kits could be improved by using both a microbiological internal control and PMA.
potentially contaminated red wines B. bruxellensis populations present in five red wines (Wine 1, Wine 2, Wine 3, Wine 4, and Wine 5) from different wineries were determined by plate counting and quantification using the commercial kits.The results are shown in

Figure 1 -
Figure 1 -Accuracy profiles of the alternative method based on the results of Kit 1 (a), Kit 2 (b), and Kit 3 (c) for the Côtes du Rhône red wine with a β equal to 80% and a λ of ± 0.3 log 10 unit.

Table 4 -Analyses of red wines containing B. bruxellensis killed by a high sulfite dose (2 mg/L molecular SO 2 ).
experiments were performed on Côtes du Rhône, Burgundy and Bordeaux red wines.The results are expressed in log 10 .Total B. bruxellensis populations were determined by flow cytometry or microscopy, the culturable populations by plate counting (reference method), and quantification by commercial kits (Kit 1, Kit 2 and Kit 3; alternative method), in triplicate.
These results highlight the poor quantification by the commercial kits, given the experiments were performed by three laboratories specialized in the wine field and in the use of qPCR technics.It is necessary for all winemakers to use the same quantification methods to monitor B. bruxellensis populations.The reference method based on plate counts provides reliable results.Nutritive media have different selectivity and it is essential to always use the same nutritive media to monitor yeast from the same tank throughout vinification and aging.It is also essential to have knowledge and know how in molecular biology and qPCR analysis, because of the sensitivity of the method.
Tessonnière et al. (2009)f this yeast must be performed to validate the quantification.Similar values for the quantification of the internal control and the initial added population that are not significantly different indicate that the DNA extraction yield is acceptable.B. bruxellensis quantification is feasible under these conditions.To conclude, our study highlights that commercial kits for the quantification of B. bruxellensis have different extraction yields leading to different quantification results.The drawbacks of the methods described above could negatively affect a winemaker's decision and lead to wine spoilage due to over or underestimation.It is thus necessary to add a standardized qPCR protocol for B. bruxellensis quantification in wines.One such standardized protocol based on the work ofTessonnière et al. (2009)which includes a microbial internal control is already available in the OIV methods (OIV-OENO