Physiological significance of the cytometric distribution of fluorescent yeasts after viability staining

Physiological Significance of the CytometricDistribution of Fluorescent Yeasts AfterViability Staining

J.C. Bouchez,1 M. Cornu,2 M. Danzart,1 J.Y. Leveau,1 F. Duchiron,3 M. Bouix1

1Ecole Nationale Superieure des Industries Agricoles et Alimentaires, 1 avenuedes Olympiades, 91744 Massy Cedex, France; telephone: 69-93-51-43;fax: 69-93-50-84; e-mail: [email protected] FranBaise de Securite Sanitaire des Aliments, 22 rue Pierre Curie,BP332-94709 Maisons Alfort Cedex, France3Laboratoire de Microbiologie Industrielle, Universite de Reims, BP 1039 Moulinde la Housse, 51687 Reims Cedex 2, France

Received 15 April 2003; accepted 15 December 2003

Published online 8 April 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/bit.20054

Abstract: This article describes a new method for the earlydetection of alcoholic fermentation arrest. This methodo-logy is based on the flow cytometric assessment ofSaccharomyces cerevisiae yeasts stained with a carboxy-fluorescein diacetate fluorescent viability probe. Multi-component analysis of viable cell distribution constitutesa promising new tool to describe physiological and dynamicchanges to heterogeneous viable populations duringalcoholic fermentation, through its ability to discriminatebetween successful processes and those ending prema-turely. This framework, which is based on the comparisonof cytometric histogram descriptors’ combinations that canbe related to simple physiological significance comparison,quickly and simply, allows testing yeasts for their fermen-tation ability and can be used to detect any kind of viabilityloss so that fermentation arrest can be avoided. B 2004Wiley Periodicals, Inc.

Keywords: Saccharomyces cerevisiae; alcoholic fermenta-tion; yeast physiology; fluorescent viability probe; flowcytometric histogram modelisation

INTRODUCTION

Viable preparations of Saccharomyces yeasts are widely

employed in the wine industry, particularly for the pro-

duction of Champagne. In addition to the specificities of

yeast strains, the consistency and quality of Champagne

may be directly influenced by their physiological state or

fermentation operating conditions. Because the process is

often well controlled, producers are frequently facedwith the

problem of must composition, which is certainly the most

variable parameter responsible for fermentation arrest prior

to consumption of the entire carbon source. This is usually

caused by a nitrogen deficiency or the persistence of pesti-

cide residues in the must which hinder satisfactory fermen-

tation because they affect the viability or vitality of yeasts.

Yeast performance in alcoholic fermentation depends

directly on yeast activity, which can be seen as a function of

cell viability as well as the physiological state of viable cells.

If it were possible to measure yeast viability and/or vitality

rapidly and accurately, producers would then be able to take

corrective action at the beginning of fermentation.

Over the years, several methods have been developed to

measure viability and vitality (Attfield et al., 2000; Boulton,

1996; Edwards et al., 1997; Lentini, 1993; Lloyd and

Hayes, 1995; McFeters et al., 1995; Porter et al., 1996):

plating, slide culture, vital stains, metabolic activity or other

methods to assess viability, and metabolic activity, cell

components, fermentation capacity, acidification potential

or oxygen uptake ability to assess vitality.

Among vital stains, fluorescent dyes are amenable to

extremely rapid analysis, particularly when used in con-

junction with a flow cytometer. Staining with carboxy-

fluorescein diacetate is based on the assumption that only

cells with an intact membrane and esterase activity are able

to accumulate the fluorescent agent. This concept has been

applied to the counting of viable yeasts and bacteria in

a variety of food products (Laplace-Builhe et al., 1993).

Flow cytometric alternatives have been compared with

standard methods for the assessment of yeast cultures in

baking, wine-making, cider manufacture, and brewing. In

all cases, the results of flow cytometry closely correlated

with the standards.

However, although knowledge of whether cells are alive

or dead is important, it is inadequate because cells may be

viable but weakly active and incapable of achieving

fermentation. Recently, Bouix and Leveau (2001) used the

carboxyfluorescein diacetate efflux phenomenon described

by Breeuwer et al. (1994) to determine yeast vitality by

calculating the relative decrease in the median fluorescent

index of loaded viable cells after probe excretion for 15 min.

The evolution of flow cytometric distribution during

fermentation reveals dynamic changes in heterogeneous

B 2004 Wiley Periodicals, Inc.

Correspondence to: J.C. Bouchez

populations of viable cells which can be explained by

cellular physiological diversity occurring throughout the

process (Campbell and Yentsch, 1989a,b; Campbell et al.,

1989; Jespersen and Jakobsen, 1994; Muller et al., 1997).

An assessment of yeast vitality based on analysis of a multi-

component cytometric histogram in terms of physiological

significance (or, at least, explicable biological phenomena)

would be of considerable value to understanding the rea-

sons for loss of viability with respect to these physiological

parameters and allow a distinction between normal fermen-

tations and those which can be expected to end prematurely.

A variety of methods have been devised to analyze

cytometric histograms, especially for the rapid determina-

tion of DNA content in cell populations, but some have only

been applied successfully to certain classes of histograms,

depending on their distribution shape (Bagwell, 1979;

Baisch et al., 1982; Brown et al., 1996; Dean and Jett,

1974; Fox, 1980). In order to evaluate and compare the

distribution of a large number of viable yeasts, this dis-

tribution should be reduced to its typical mathematical

features, following data fit using parametric or nonpara-

metric methods.

Parametric methods are of interest because they summa-

rize distribution using only a few parameters. However,

they do not provide accurate estimates when the actual data

do not fit the assumed theoretical distribution, with the

result that the first conventional moments are insufficient to

define complex cytometric histogram distributions, apart

from those in the Pearson (including uniform,Gaussian, beta,

Fischer, Student, andgamma)orZipf (Christodoulakis, 1983;

Cooper, 1963) families.

When the histogram is composed of several cohorts

(gaussian or nongaussian components), a statistician adopts

an approach involving sophisticated mathematical proce-

dures that require numerical processing on a computer. The

best-known examples are the multigaussian deconvolution

technique (Fried, 1976, 1979), Bhattacharya logarithmic dif-

ferences (Lepetit et al., 1991), the polynomial curve-fitting

technique (Dean et al., 1984), multiharmonic analysis

(Sonneveld et al., 1988), or neural network analysis (Radvin

et al., 1993). For all these techniques, a theoretical his-

togram is constructed and considered a reliable reconstruc-

tion of the true histogram.

When the law is unknown, nonparametric methods can be

used that do not make a priori assumptions about dis-

tribution. These are subdivided into two classes: first, scale-

based methods for which equal width (Chu, 1991; Mostardi,

1990), equal-height (Liou and Yao, 1977; Muralikrishna

and DeWitt, 1988; Piatetsky-Shapiro and Connell, 1984)

and variable-width (Merrett and Otoo, 1979; Muthuswamy

and Kerschberg, 1985; Whang et al., 1994) histograms are

well-known examples, and second, analytic methods which

approximate the actual data distribution using a finite series

of analytic functions (Lefons, 1988; Lefons et al., 1983,

1995; Sun et al., 1993).

Guillaud and Chassery (1992) used a pattern recognition

approach to interpret and analyze histograms which were

defined in a new vectorial space determined by Legendre

polynomials. According to these authors and to Hosking

(1990) before them, linear combinations of an ordered

dataset, called L-moments, have the theoretical advantage

over conventional statistical moments in that they character-

ize a broader range of distributions.

Inspired by their work, histogram descriptors were

chosen for their likeness to these moments, but particularly

(to a certain extent), for their physiological significance.

The histogram descriptor method was developed in order to

extract relevant parameters from the fluorescence distribu-

tions measured in viable yeasts, the aim being to associate

each histogram shape signature with a particular fermenta-

tion behavior: normal, arrested, or sluggish (the latter was

not tested during this study). An effective tool should be

able, first of all, to identify fluorescence distribution dif-

ferences resulting from well-described, underlying, bio-

logical phenomena related to contrasting fermentation

behaviors (normal and arrested), and then be tested when

more complicated biological phenomena are involved

(sluggish fermentation).

In this article, we assumed that multicomponent analysis

of cytometric histograms could provide information on dy-

namic changes to the number of viable cells during normal

fermentations carried out under different pH and temper-

ature conditions, and therefore could enable the detection of

drastic viability loss in the event of fermentation arrest.

In the first part, we describe how a limited set of relevant

descriptors could reliably explain physiological signifi-

cance. Indeed, our approach was to start with an assumed

pattern of distribution position and shape and a set of rules

governing its shifting and skewing as a function of phys-

iological parameters related to the fluorescence response

of cells (mainly esterase activity and pHi). The latter part

demonstrates the efficiency of the cytometric histogram

method for early discrimination between normal and

induced fermentation arrest.

MATERIALS AND METHODS

Yeast

The yeast was a French Champagne yeast strain, Saccha-

romyces cerevisiae (S005), obtained from Moet and Chan-

don (Epernay, France).

Fermentations

All fermentations were carried out in a 2-L fermenting

vessel (Set 2M, Inceltech, Toulouse, France) with contin-

uous stirring (140 rpm); temperature and pH were regulated

(adjustment of pH by the use of 2N NaOH). The fer-

mentation medium was a synthetic substance containing:

dextrose (150 g.L�1), yeast extract (2.5 g.L�1), (NH4)2SO4

(2.5 g.L�1), KH2PO4 (2.5 g.L�1), and MgSO4 (0.5 g.L�1).

BOUCHEZ ET AL.: CYTOMETRIC DISTRIBUTION OF FLUORESCENT YEASTS 521

The pH was adjusted to 4.0 or 3.2. The dextrose solution and

other ingredients were sterilized independently by heat

(20 min, 120jC). The reactor was inoculated with 100 mL

of yeast culture grown for 24 h at 28jC in Sabouraud broth.

During fermentations, samples were collected every 2 h for

further analysis.

Two types of alcoholic fermentations were performed:

normal fermentations where glucose consumption was com-

plete, and arrested fermentations where the process stopped

before the carbon source was entirely consumed. Normal

fermentations were carried out under four conditions: tem-

peratures of 18jC, 22jC, or 28jC at pH 3.2, and 28jC at

pH 4.0. Arrested fermentations were carried out at pH 4.0

and a temperature of 40jC.In a previous, unpublished study using the same yeast

strain, we saw arrested fermentations at 40jC that still had

a high concentration of sugar. The effect of higher tem-

peratures was a premature end to fermentation with small

cell size, low pHi, and esterase activity; the latter two

factors were closely linked to a reduction in the fluores-

cence response.

During this high-temperature fermentation, inactivation

reactions were expected to be accelerated. The impact of a

process at 40jC on physiological status is discussed in Re-

sults and Discussion: Arrested Fermentations. Each fer-

mentation process was performed in triplicate.

Fluorescent Staining

For staining, 105 harvested cells were resuspended in 1 mL

McIlvaine buffer, which was composed of citric acid

(100 mM) and disodium hydrogen phosphate dihydrate

(Na2HPO4, 200 mM) adjusted to pH 4.0, with no energy

supply to prevent the efflux of fluorescent compounds into

cells. Only 20 AL of carboxyfluorescein diacetate (cFDA,

10 mg per mL in acetone) was added to 0.9 mL of the S.

cerevisiae suspension.

The cFDA concentration, duration, and temperature of

incubation for staining were chosen according to Breeuwer

et al. (1994) in order to permit the optimum passive uptake

rate of the prefluorochrome, with a permeability coefficient

of 1.3 10�8 m.s�1, which was assumed to remain constant

whatever the changes to membrane composition.

cFDA assays were read immediately after the 10-min

incubation (40jC) period before the occurrence of any

significant, energy-dependent leakage of the fluorescent

product from cells.

Flow Cytometry Analysis

Analysis of the fluorescence intensity of individual cells

was achieved using a flow cytometer (ChemFlow III flow

cytometer; Chemunex, Ivry sur Seine, France). A logarith-

mic amplification of the incoming signal was applied to

measuring a broader dynamic range of signals in one histo-

gram. Overall fluorescence intensity values (log value) were

recorded by the D-Pack software in ASCII-format files

(Chemunex) for further running after loading in a personal

computing MatLab 5.1 program (MathWorks, Natick, MA)

(available on request from the first author).

Total/Viable Cells and Determinationof Ethanol Levels

Total yeast cell counts were performed using a classic mi-

croscopic Malassez cell counter, while viable cell levels

were calculated from the cytometric count and associated

dilution factor. In a previous work (Bouix and Leveau,

2001), the plate count method and the number of fluorescent

cells determined by flow cytometry were compared. Be-

cause a good correlation was found between the two

viability assessments, the cytometric count alone was used

to determine viable cells during the present study. After

sample centrifugation, ethanol levels in the supernatant

were determined using a gas chromatograph (Autosystem,

Perkin Elmer, Courtaboeuf, France) equipped with a head-

space analyzer (HS 40XL, Perkin Elmer).

Computational Methods

Estimate of Growth Kinetic Parameters

The Baranyi and Roberts (1994) growth model was used to

analyze yeast growth kinetics. The MicroFit software

(Institute of Food Research, UK) enabled the extraction

of Baranyi’s growth parameters from the experimental

dataset: maximum growth rate, lag time, initial and final

cell counts.

Calculation of Specific Ethanol Production Rates andActivity Classification

Estimated Baranyi parameters were loaded into a personal

computing MatLab 5.1 program, which allowed calculation

of the specific ethanol production rate curve based on the

fitted growth and ethanol curves used in the differen-

tial equation:

r ¼ 1

X� dYdt

As fermentation proceeded and stress arose (notably due to

a rise in the ethanol concentration), the calculated high or

rising rates associated with an upper activity class grad-

ually decline to become constant from about 30 g.L�1

ethanol in all normal fermentations. Beyond this ethanol

concentration, all rates were associated with a lower activ-

ity class.

Algorithm of Histogram Descriptor Set Calculation

An algorithm was used to calculate a combination of a few

descriptors from each smoothed, normalized, gated, sample

histogram. The most prominent features of the histogram

522 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 86, NO. 5, JUNE 5, 2004

were extracted from this algorithm. The exponential fit of

the histogram preserved maximum frequency and quartile

index values which differed from the expected outcomes

obtained using other fitting methods:

8je�0;256�;aj ¼

Pj

i¼1

Ai:exp½�a:ðj� iÞ�

Pj

i¼1

exp½�a:ðj� iÞ�

The histogram gating avoided taking account the slight

fluorescence background noise (left cursor). Gated histo-

gram normalization enabled the comparison of dif-

ferent histograms.

Xy

j¼x

ðajÞN ¼ 100 with ðajÞN ¼ 100:aj

Py

j¼x

aj

and x < y < j

All descriptors were calculated on the normalized gated

histogram after fittingwith the exponential coefficient a = 0.3.

This coefficient was a compromise value, chosen to preserve

the global shape of the histogram and avoid local disconti-

nuities in distribution which could have distorted calculation

of the descriptors.

Index channels were converted from a log to a linear

scale using the formula:

j0 ¼ 0:0386exp½0; 0187��

Choice of Histogram Descriptors

Descriptors were histogram-extracted variables that could

be very strongly collinear, and therefore condensed (prior to

classification) into a smaller number using principal

component analysis. Of the 14 original descriptors, only

three were chosen for their nonredundancy, physiological

significance, and ability to distinguish between normal and

arrested fermentations. The choice of histogram descriptors

was based on their ability to describe the position of the

histogram on cytometer channels and the overall shape of

the distribution, without presuming the likely law of

distribution (Fig. 1).

1) The position of the histogram could be determined from

the median of distribution: Median index, M.

2) Equivalence of kurtosis (measurement of whether the

data were peaked or flat when compared with a normal

distribution) could be assessed from the width of the

histogram width at the middle of the modal value: Width

index, W.

3) Equivalence of skewness (measurement of the lack of

symmetry in a distribution) could be assessed from the

ratio between the left and right widths at the middle of the

modal value of the distribution: Dissymmetry index, D.

It should be noted that in many studies the median index

was the only parameter taken into account, while histogram

width was rarely considered, although it was used to cal-

culate a variation coefficient using the formula: EW/mean.

Because the distributions were always fitted by a gaussian

law, dissymmetry (linked to kurtosis) was never assessed.

RESULTS AND DISCUSSION

Normal Fermentations

Biochemical Results

Normal fermentations performed at pH 3.2 at 18jC, 22jC,or 28jC, and pH 4.0 at 28jC, consumed all the dextrose

available (150 g.L�1), and attained a level of nearly 70 g.L�1

ethanol (vs. the theoretical level of 75.6 g.L�1 ethanol

according to the Pasteur rate). Total and viable cells profiles

for normal and arrested fermentations are shown in

Figure 2a, ethanol production profiles in Figure 2b, and

specific ethanol production rate curves in Figure 2c.

Fermentations carried out at higher temperatures and higher

pH values were more rapid in terms of growth and ethanol

production than those performed at low pH and temperature

values. As shown in Figure 2c, the specific ethanol

production rate as a function of ethanol concentration

differed markedly between the normal fermentations at

28jC (pH 3.2, 4.0) and those performed at 18jC or 22jC.Different trends could be seen for the arrested fermenta-

tions: production rates were high but dropped directly.

Cytometric Results

Each descriptor profile was plotted as a function of the

ethanol concentration. This avoided the problem of kinetic

shifts resulting from pH and temperature differences during

normal fermentations. As shown in Figure 3a–c, median

fluorescence, width, and dissymmetry index profiles could

Figure 1. Calculation of descriptors: median (M), width (W), and dis-

symmetry (D) indices were calculated on normalized gated histogram after

exponential fit.


be reduced to a single trend which was common to all

normal fermentations whatever the pH or temperature. As a

result, all descriptor trends were calculated to take account

of the mean and associated confidence interval (95%) of all

descriptor values found at the same ethanol concentration

in all normal fermentations (Fig. 4).

The median fluorescence index (M) rose slightly before

the level of 5 g ethanol per liter was reached. Shortly after,

this index fell significantly, just prior to 10 g.L�1 ethanol.

This was followed by a sharp increase to a high level and a

final, sudden drop. These results agree with those obtained

Figure 3. Cytometric results for two normal fermentations carried out at

pH 3.2 T22jC (w ) and pH 4.0 T28jC (E). Mean and confidence interval

(95%) plots for the three descriptors: (a) median index, (b) width index,

(c) dissymmetry index.

Figure 2. Biochemical results for normal and arrested fermentations:

Total cells:

pH 3.2 T18jCpH 3.2 T22jCpH 3.2 T28jCpH 4.0 T28jCpH 4.0 T40jC


by Jespersen and Jakobsen (1994), who studied brewingwort

and yeasts using the same fluorescence staining procedure:

the FI-value rose slightly at pitching and then fell signifi-

cantly shortly afterwards, reaching a minimum after about

24 h. This was followed by an increase to a final level, after

about 3–4 days, which was close to that seen after pitching.

Even though the profiles obtained could be transposed

from one pH and temperature condition to another, their

interpretation were difficult because fluorescence staining

depended on intracellular events or properties (intracellu-

lar pH, esterase activity, probe efflux) that could not be

attributed solely to vitality, except during the early stage

of fermentation.

According to Jespersen and Jakobsen (1994), the increase

in the fluorescence intensity of cFDA-stained cells observed

during the early exponential growth phase was probably

indicative of high levels of esterases. The subsequent de-

crease in fluorescence intensity has also been observed for

brewing yeasts where the increase of yeast sterols is sug-

gested to inhibit esterase activity.

By the end of fermentation, the principal reason for

increases in the median fluorescence index was probably

stronger esterase activity, according to our unpublished

study using h-naphtyl caprylate as a substrate. However,

there may be no link between cFDA conversion (in vivo

substrate) and h-naphtyl caprylate conversion (in vitro

substrate) due to intracellular esterases in yeast cells.

Furthermore, according to Breeuwer et al. (1995), the

esterase activity determined by measuring the hydrolysis of

cFDA in permeabilized cells was f15 nmol.min�1.mg

protein�1 and remained constant during growth. Consid-

erable caution should therefore be used regarding the

meaning and relevance of potential evolutions in esterase

activity in S. cerevisiae during various stages of growth and

its link to fluorescence response.

Another, less hypothetical, explanation may be that, as

the percentage of viable cells fell, the few fluorescent cells

assessed could present the same strong fluorescence

properties as those encountered after pitching (but not nec-

essarily the same distribution shape) because of their ability

to overcome stress and/or because of other phenomena

related to the staining response (mainly pH, esterase

activity, cytoplasm volume, vacuolization, dye interaction

with cellular compounds, etc.). In practice, if percentage

fluorescence could not be evaluated and no assumption

could be made concerning the ethanol produced, the

median index alone did not allow determination of the

physiological state of cells. The introduction of additional

descriptors would assist this diagnosis, because it could

then be assumed that each physiological state has its own

distribution signature.

Indeed, width and dissymmetry indices, which have

never been described before, demonstrated reproducible

profiles throughout the process, showing that their values

were probably related to the physiological state of yeasts.

For these two descriptors, index values rose slightly

between 15 and 25 g.L�1 ethanol, beyond the values ex-

pected under a gaussian distribution (reference = 1) in

terms of width and reaching about 2.5 regarding dissym-

metry, demonstrating that the distributions were clearly

skewed to the left.

It is clear that histograms became much thinner and taller

as yeast homogeneity increased (which means that different

Figure 4. Discrimination between normal and aborted fermentations.

Mean and confidence interval (95%) plots for the three descriptors: (a)

median indice (normal –n– and aborted –5– fermentations); (b) width

index (normal –x – and aborted – w – fermentations); (c) dissymmetry

index (normal –E– and aborted –4– fermentations).


staining properties among the total viable population were

not markedly different). At the initiation of fermentation,

high width values could be explained by heterogeneity of

the inoculum. Thereafter, width values rapidly fell, reach-

ing a minimum at about 10 g.L�1 ethanol. At this stage of

fermentation the specifically high ethanol production rate

resulted from the contribution of an increasing proportion

of viable cells in the same healthy physiological state

(marked homogeneity). During the process, differences

among viable cells were revealed by flattening of the

histogram and greater cell heterogeneity, because of the

combined effect of yeast aging (disparity in budding states)

and increased ethanol production (disparity of stress sus-

ceptibility related to yeast aging and its effect on the

evolution of membrane composition).

As from 20 g.L�1 ethanol, cell homogeneity was likely

to reappear gradually, for two main reasons:

1) The oldest yeasts, which had previously contributed to a

fraction of less fluorescent cells and were incapable of

overcoming the combined stress of the production of

ethanol and other toxic substances, gradually disap-

peared because of lysis.

2) Those cells capable of recovery contributed to a more

homogeneous fraction of cells (viable, surviving cells),

although their number only represented a small

percentage of total cells.

As for the dissymmetry index profile, it is suggested that

their peak heterogeneity was due to an increase in the

number of less viable cells in the total viable population

due to cell death and loss of vitality, resulting in a drop in

the proportion of viable cells together with a decline in the

specific rate of ethanol production.

Arrested Fermentations

During alcoholic fermentation in batches, the increasing

concentration of ethanol adversely affected the state and

the activities of the yeast population, including its specific

growth rate, specific rate of fermentation, and viability.

Under an increased process temperature, some of these

effects may have become more severe (Beney et al., 2000).

The underlying mechanisms are numerous and include

the irreversible denaturation and hyperbolic noncompeti-

tive inhibition of glycolytic enzymes, the exponential

noncompetitive inhibition of the glucose transport system,

depression of the optimum and maximum temperatures for

growth, and the enhancement of thermal death.

Biochemical Results

For this 40jC high-temperature fermentation, high concen-

trations of ethanol were formed and this occurred most

rapidly during the early part of the fermentation, during

active yeast growth and multiplication. At this stage the

intracellular concentration of ethanol may be high and par-

ticularly toxic, as it may act from within the cell to interact

with the cell membrane, leading to a loss of viability in the

overall population. These results are in line with previous

reports that yeast viability decreased as the temperature in-

creased. This decrease was thought to be due to a greater

accumulation of intracellular ethanol at higher temper-

atures, which would produce cell toxicity and alter the

structure of the membrane, thus reducing its functionality.

The usual growth curve, with a series of short-lag,

exponential, stationary, and decline phases, was observed,

but at this high, 40jC temperature, a large quantity of yeast

died earlier during the process. This high yeast mortality

may have induced fermentation to arrest with a high sug-

ar content.

Cytometric Results

The median fluorescence index (M) profile was quite

similar to that found during normal fermentations up to

about 15 g.L�1 ethanol, with a slight shift towards low

values which could be explained by one (or more) hy-

pothetical phenomena: a lower intracellular pH, a higher

probe efflux or fluorescence quenching resulting from a

higher intracellular probe concentration, because the

volume of stressed yeasts was lower (data not shown).

Because protection against extreme conditions such as

heat and high ethanol concentrations is dependent to a

large extent on a high membrane sterol content (because of

its protective function in modulating the fluidity of the

membrane phospholipid ‘‘bulk membrane function’’, and

also its role in the initiation of the cell cycle, or ‘‘sparking

function’’), it could be suggested that the median value

was low because of the adaptative increase of yeast sterols

inhibiting esterase activity (Jespersen and Jakobsen, 1994).

Once median index values had spiked, they were not

followed by the increase that would have been expected

because of a loss of viability in total cells (no cells capable

of recovery). Thus, in the case of arrested fermentations

due to a high temperature, the reasonable assumptions

which can be made are that more and more cells may have

severely damaged membranes, there is low esterase ac-

tivity, and low pHi. The pHi is very important, principally

because the fluorescence of fluorescein and cF is high-

ly pH-dependent, and also because pHi affects en-

zyme activity.

The width index (W) profile exhibited almost gaussian

homogeneity throughout fermentation. This was probably

due to the combined effect of yeast aging and increased

ethanol production, which appeared earlier in this case

because of the synergistic effect of high temperature

exposure (Piper, 1995) that led to more sustained hetero-

geneity than in the normal fermentations.

The dissymmetry index (D) also demonstrated an almost

normal profile up to an ethanol concentration of 15 g.L�1,

after which the histograms became skewed to the right.

Thus, fermentation arrest was consistent with loss of vi-


ability: fewer and less viable cells with greater hetero-

geneity and a lower level of fluorescence.

The primary aim of this work was to demonstrate that

hitherto unused cytometric histogram descriptors related to

simple physiological parameters such as the median

fluorescence intensity, heterogeneity, and dissymmetry

of viable cells might characterize dynamic changes to

yeast vitality during normal fermentation. Differences

were observed throughout the process in all descriptors,

revealing that the fermentation stage, which was linked to a

particular physiological state of viable cells, could be

associated with range values for each descriptor. In some

cases, however, the separate consideration of descriptors

independently did not enable the differentiation of phy-

siological status, because descriptor values could be

common to quite different stages of fermentation.

The second aim of this work was to use cytometric

histogram descriptors to detect any loss of viability and

thereby to distinguish, early in the process, between normal

and arrested fermentations. For most descriptors, differ-

ences in trends between normal and arrested fermentations

occurred throughout the process. In some cases, however,

the separate consideration of descriptors would not have

been sufficient because some descriptor values were

common to both arrested and normal fermentations, es-

pecially after ‘‘yeast pitching’’ with respect to median

fluorescence and dissymmetry indices and between 15 and

25 g.L�1 ethanol for the width index.

For the above reasons, a combination of descriptors was

the onlymeans of collating all data on viable cell distribution

and associating each histogram shape to a particular

physiological state, and thus anticipating the satisfactory or

poor progress of fermentation.

Early Discrimination Between Normaland Arrested Fermentations

Fermentations were divided into six phases based on

ethanol scale values. For each phase the means of des-

criptors were calculated taking into account all quantitative

values associated with ethanol concentrations within this

range. The means and their associated confidence

intervals (95%) are shown in Figure 5 for both normal

and arrested fermentations. For each histogram descrip-

tor, quantitative values were associated with one of three

possible qualitative indexes defined with respect to the

corresponding profile, with a view to permitting further

discrimination between normal and arrested fermentations

(Table I). The cutoffs were chosen visually after the

descriptor profiles had been established. Typical combi-

Figure 5. Discrimination between normal (n) and aborted (5)

fermentations for six ethanol concentration ranges. Mean and confidence

interval (95%) histograms for the three descriptors: (a) median index, (b)

width index, (c) dissymmetry index.

Table I. Classes of descriptors.

1 2 3

Median (M) M1 if < 1 M2 if between

1 and 1.5

M3 if > 1.5

Width (W) W1 if < 0.6 W2 if between

0.6 and 0.8

W3 if > 0.8

Dissymmetry (D) D1 if < 1 D2 if between

1 and 2

D3 if > 2

Quantitative values were associated with one of three possible

qualitative indexes. For instance, a width quantitative value of 0.85 would

be associated with qualitative index W3.


nations of normal and arrested fermentations are shown

in Table II.

Although within the 0–5 g.L�1 ethanol concentration

range typical combinations could be sketched for the two

fermentations, if the normal and arrested mean standard

deviations for each descriptor were taken independently,

they overlapped, not allowing accurate discrimination. As

from 5 g.L�1 ethanol, some combinations began to provide

very specific information on likely fermentation behavior.

If the width index (W) alone was considered at between

5–10 g.L�1 ethanol, it was possible to forecast the outcome

of fermentation: the thinner the histogram, the more likely it

was that fermentation would be completed. Inversely, viable

cell heterogeneity appeared as a result of cell aging and also

the combined effect of a rising ethanol concentration and a

high temperature. At lower temperatures (normal condi-

tions), cell division and ethanol effects appeared later

because of longer latency and lower ethanol toxicity.

It is likely that other, more subtle differences also

pertain, giving rise to differential staining, depending on

the cell cycle. All cells in a culture are not identical; in a

batch culture, each individual had its own characteristics

and in some sense may be unique. All stages of the growth

and division cycles are represented. It is therefore perhaps

readily understandable that the mean cell age of a yeast

population may be an important determinant of the

differences in the width index between fermentations

during the early stage of growth.

As from 10 g.L�1 ethanol, the width index was less and

less discriminating because the cell heterogeneities of

normal and arrested fermentations no longer differed

significantly. However, the median and dissymmetric

indices were of particular interest: during normal fermen-

tations, less fluorescent subpopulations began to merge

from overall viable cells (the mean D value increased),

whereas during arrested processes the histograms were

skewed to the right and the median value fell.

In addition, the combinations associated with arrested

fermentations were never those encountered at the

completion of normal fermentation. As from 30 g.L�1

ethanol (when the specific rate of ethanol production was

low or decreasing), viability losses in these two cases were

related to two distinct mechanisms, resulting in separate

physiological responses.

CONCLUSIONS

Two different approaches were applied to making an early

distinction between normal and arrested fermentations.

Under the first approach, cytometric histograms could be

recorded at random time points during the process. The

combinations of descriptors resulting from this experi-

mental distribution could be compared with a theoretical

database and suggest a probability of whether the fer-

mentation would be normal or not. Further, other histo-

grams could easily be recorded to confirm or invalidate the

first assessment.

Second, cytometric histograms could also be recorded

at key stages of the process, within a known ethanol

concentration range. It was therefore possible to determine

within which minimum ethanol concentration range a

distinction could be made between normal and ‘‘heat-

arrested’’ fermentations. As soon as a level of 5 g.L�1

ethanol was attained, it was possible to forecast whether a

fermentation was likely to abort, because at this concen-

tration the viable population was more heterogeneous,

according to its highest width index value. As from

15 g.L�1 ethanol, the median and dissymmetry indices were

the only useful descriptors for discrimination.

These interesting, reproducible differences in descriptor

profiles and combinations highlight the complex nature of

yeast physiology and the changes which may occur during

normal anaerobic and arrested fermentations. However,

even though flow cytometry may reveal some subpopula-

tions which merge according to the shape of the histogram

and thus provide a more precise picture of the overall

fluorescence response of the yeasts under test, the questions

raised by the physiological significance of descriptors

cannot be entirely answered because some necessary, ac-

companying biological measurements were not made at the

time of the experiment.

The biological interpretation of the cytometric variability

observed, that mainly relied on esterase activity and

membrane integrity, should be verified for different strains

because their enzyme pool and ethanol tolerance may vary.

Taking account of the fact that our study was carried out in

a synthetic medium using a selected Champagne strain, the

median profiles for normal fermentations were quite similar

to those obtained by Jespersen and Jakobsen (1994) for a

brewing strain fermenting wort, and also quite similar in

must with the same strain and the same synthetic medium

for another, confidential Champagne strain (unpubl. study).

We were therefore encouraged to think that the only

differences may occur in terms of quantitative index values,

and that the combinations of descriptors may be the same

Table II. Combination of descriptors for six ethanol concentration ranges

that permitted discrimination between normal and aborted fermentations.

Combination of descriptorsEthanol

concentration range Median Width Dissymmetry Fermentation

0–5 M2 W2 D2 Normal

M1 W3 D1 Aborted


M2 W3 D1 Aborted


M1 W3 D1 Aborted


M1 W3 D1 Aborted


M1 W3 D2 Aborted

>30 M3 W1 D2 Normal

(low activity)


with respect to the cutoffs chosen. Nevertheless, our

purpose of presenting the applications described here was

to illustrate how the theoretical framework could be

applied. We did not present these results as definitive

proof that all arrested or sluggish fermentations lead to the

same shapes and patterns.

Ideally, a study of this nature should teach us how to

proceed in the reverse direction, i.e., beginning with the

histogram and using it to infer a unique type of fermentation

behavior. That goal still eludes us in many situations.

In this article, we have demonstrated the efficiency of

using combinations of cytometric histogram descriptors to

discriminate at an early stage between normal and arrested

fermentations. Because the alcoholic fermentation process

for Champagne wine is often well controlled, exposure to

high temperature, which is the easiest way to induce

fermentation arrest, is probably a theoretical rather than a

realistic cause of arrest. On the other hand, as producers are

mainly faced with problems of must composition, it was

encouraging to think, therefore, that the technique presented

here, enabling characterization of the inherent variability of

viable cell distributions, could provide a basis for the study

of cytometric histogram signatures for viable yeasts in the

case of slow and sluggish fermentations due to nitrogen

deficiency or the persistence of pesticide residues in the

must, because these two factors are known to hinder

satisfactory fermentation.

NOMENCLATURE

Y Ethanol concentration (g.L�1)X Total biomass concentration (g.L�1)i and j Channel numbers of cytometer in log scalejV Channel number of cytometer in linear scaleAi Number of events on channel ia Exponential fit coefficientaj Calculated number of events on channel j after exponential fitr Ethanol production specific rate

The authors thank Moet & Chandon Research Laboratories, the

Champagne Wine Trade Committee, and Europol’Agro for their

contributions to this work.

References

Attfield PV, Kletsas S, Veal DA, Van Rooijen R, Bell PJL. 2000. Use of

flow cytometry to monitor cell damage and predict fermentation

activity of dried yeasts. J Appl Microbiol 89:207–214.

Bagwell CB. 1979. Theory and application of DNA histogram analysis.

Thesis, University of Miami.

Baisch H, Beck HP, Christensen IJ, Hartman NR, Fried J, Dean PN, Gray

JW, Jett JH, Johnston DA, White RA, Nicolini C, Zietz S, Watson JV.

1982. A comparison of mathematical methods for the analysis of DNA

histograms obtained by flow cytometry. Cell Tissue Kinet 15:235–249.

Baranyi J, Roberts TA. 1994. A dynamical approach to predicting bacterial

growth in food. Int J Food Microbiol 23:277–294.

Beney L, Martinez de Maranon I, Marechal PA, Gervais P. 2000. Influence

of thermal and osmotic stresses on the viability of the yeast Saccha-

romyces cerevisiae. Int J Food Microbiol 55:275–279.

Bouix M, Leveau JY. 2001. Rapid assessment of yeast viability and yeast

vitality during alcoholic fermentation. J Inst Brew 107:217–225.

Boulton CA. 1996. A critical assessment of yeast vitality testing. Ferment

9:222–226.

Breeuwer P, Drocourt J-L, Rombouts FM. 1994. Energy-dependent,

carrier-mediated extrusion of carboxyfluorescein from Saccharomyces

cerevisiae allows rapid assessment of cell viability by flow cytometry.

Appl Environ Microbiol 60:1467–1472.

Breeuwer P, Drocourt J-L, Bunschoten N, Zwietering M-H, Rombouts

F-M, Abee T. 1995. Characterization of uptake and hydrolysis of

fluorecein diacetate and carboxyfluorescein diacetate by intracellular

esterases in Saccharomyces cerevisiae, which result in accumulation of

fluorescent product. Appl Environ Microbiol 61:1614–1619.

Brown RD, Linden MD, Mackowiak P, Kubus JJ, Zarbo RJ, Rabinovitch

PS. 1996. The effect of number of histogram events on reproduc-

tibility and variation of flow cytometric proliferation measurement.

Am J Clin Pathol 105:696–704.

Campbell JW, Yentsch CM. 1989a. Variance within homogeneous

phytoplankton populations. I. Theoretical framework for interpreting

histograms. Cytometry 10:587–595.

Campbell JW, Yentsch CM. 1989b. Variance within homogeneous phy-

toplankton populations. II. Analysis of clonal cultures. Cytometry 10:

596–604.

Campbell JW, Yentsch CM, Cucci TL. 1989. Variance within homoge-

neous phytoplankton populations. III. Analysis of natural populations.

Cytometry 10:605–611.

Christodoulakis S. 1983. Estimating record selectivities. Inf Syst 8:

105–115.

Chu PC. 1991. A contingency approach to estimating record selectivities.

IEEE Trans Software Eng 17:544–552.

Cooper PW. 1963. Statistical classification with quadratic forms. Bio-

metrika 50:439–448.

Dean PN, Jett JH. 1974. Mathematical analyis of DNA distributions

derived from flow microfluorimetry. J Cell Biol 60:523–527.

Dean PN, Dolbeare F, Gratzner H, Rice GC, Gray JW. 1984. Cell-cycle

analysis using a monoclonal antibody to BrdUrd. Cell Tissue Kinet

17:427–436.

Edwards C, Porter J, West M. 1997. Fluorescent probes for measuring

physiological fitness of yeast. Ferment 9:288–293.

Fox MH. 1980. A model for computer analysis of synchronous DNA

distributions obtained by flow cytometry. Cytometry 1:71–77.

Fried J. 1976. Method for the quantitative evaluation of data from flow

cytometry. Comp Biomed Res 9:263–276.

Fried J. 1979. Multi-user system for analysis of data from flow cytometry.

Comput Programs Biomed 10:218–230.

Guillaud M, Chassery J-M. 1992. Histogram analysis by use of L-moments,

linear functions of order statistics. Stat Anal Donnees 16:85–106.

Hosking JRM. 1990. L-moments: analysis and estimation of distributions

using Linear combinations of order statistics. JR Stat Soc B 52:

105–124.

Jespersen L, Jakobsen M. 1994. Use of flow cytometry for rapid estimation

of intracellular events in brewing yeasts. J Inst Brew 100:399–403.

Laplace-Builhe C, Hahane K, Hunger W, Tirilly Y, Drocourt JL. 1993.

Application of flow cytometry to rapid microbial analysis in food and

drinks industries. Biol Cell 78:123–128.

Lefons E. 1988. The functional object model for analytic databases. Riv

Inf 18:305–339.

Lefons E, Silvestri A, Tangorra F. 1983. An analytical approach to sta-

tistical databases. Proc 9th Int Conf Very Large Data Bases; Florence,

Italy. p 260–274.

Lefons E, Merico A, Tangorra F. 1995. Analytical profile estimation in

database systems. Inf Syst 20:1–20.

Lentini A. 1993. A review of various methods available for monitoring the

physiological status of yeast: yeast viability and vitality. Ferment

6:321–327.

LepetitM, Loranchet S, Gillet P,Marion JM. 1991. Un logiciel de traitement

des histogrammes de structure de population par la methode des

differences logarithmiques de Bhattacharya. Vie Milieu 41:127–131.

Liou JH, Yao SB. 1977. Multidimensional clustering for data base orga-

nizations. Inf Syst 2:187–197.


Lloyd D, Hayes AJ. 1995. Vigour, vitality and viability of microorga-

nisms. FEMS Microbiol Lett 133:1–7.

McFeters GA, Yu FP, Pyle BH, Stewart PS. 1995. Physiological as-

sessment of bacteria using fluorochromes. J Microbiol Meth 21:

1–13.

Merrett TH, Otoo E. 1979. Distribution models of relations. Proc 5th Int

Conf Very Large Data Bases. Rio de Janeiro, Brazil. p 418–425.

Mostardi T. 1990. Estimating the size of relational SPqJ operation results:

an analytical approach. Inf Syst 15:591–601.

Muller S, Hutter KJ, Bley T, Petzold L, Babel W. 1997. Dynamics of

yeast cell states during proliferation and non proliferation periods in

a brewing reactor monitored by multidimensional flow cytometry.

Bioproc Eng 17:288–293.

Muralikrishna M, DeWitt DJ. 1988. Equi-depth histograms for estimating

selectivity factors for multi-dimensional queries. Proc ACM SIGMOD

Int Conf Management of Data. Chicago. p 28–36.

Muthuswamy B, Kerschberg G. 1985. A DDSM for relational query

optimization. Proc ACM Annu Conf. Denver. p 439–448.

Piatetsky-Shapiro G, Connell C. 1984. Accurate estimation of the number

of tuples satisfying a condition. Proc ACM SIGMOD Int Conf Man-

agement of Data. Boston. p 256–276.

Piper PW. 1995. The heat shock and ethanol stress responses of yeast

exhibit extensive similarity and functional overlap. FEMS Microbiol

Lett 134:121–127.

Porter J, Deere D, Pickup R, Edwards C. 1996. Fluorescent probes and flow

cytometry: new insights into environemental bacteriology. Cytometry

23:91–96.

Radvin PM, Clark GM, Hough JJ, Owens MA, McGuire WL. 1993. Neural

network analysis of DNA flow cytometry histograms. Cytometry 14:

74–80.

Sonneveld P, Schultz FW, Mulder JA. 1988. New method for the analysis

of flow cytometric data. Anal Quant Cytol Histol 10:261–268.

Sun W, Ling Y, Rishe N, Deng Y. 1993. An instant and accurate size

estimation method for joins and selection in retrieval-insensitive en-

vironment. Proc ACM SIGMOD Int Conf Management of Data.

Washington, DC. p 79–88.

Whang KY, Kim SW, Wiederhold G. 1994. Dynamic maintenance of data

distribution for selectivity estimation. VLDB J 3:29–51.


Physiological significance of the cytometric distribution of fluorescent yeasts after viability staining

Documents