## Doi:10.1016/j.envexpbot.2004.01.004

Environmental and Experimental Botany xxx (2004) xxx–xxx
Bacterial populations on the leaves of Mediterranean plants:
quantitative features and testing of distribution models
R.K.P. Yadav , J.M. Halley , K. Karamanoli , H.-I. Constantinidou , D. Vokou
a

*Department of Ecology, School of Biology, Aristotle University, GR-541 24 Thessaloniki, Greece*
b

*Laboratory of Agricultural Chemistry, School of Agriculture, Aristotle University, GR-541 24 Thessaloniki, Greece*
**Abstract**
We studied the bacterial colonization of the phyllosphere of eight perennial species occurring in a Mediterranean ecosystem
in Sithonia (Halkidiki), northern Greece, over a period of 2 years. The plant species were

*Arbutus unedo*,

*Quercus coccifera*,

*Pistacia lentiscus*,

*Myrtus communis*,

*Lavandula stoechas*,

*Calamintha nepeta*,

*Melissa officinalis *and

*Cistus incanus*. They differin a number of morphological features, mainly in habit and leaf trichome. The bacterial colonization of their leaves is highlyvariable. Over all species and sampling dates, observed values ranged from non-detectable to a maximum of 1

*.*4 × 107 CFU g−1in

*C. nepeta*. The average size of the microbial community varied among species by a factor of about 10, from 1

*.*3 × 104 CFU g−1in

*A. unedo *and

*L. stoechas *to 1

*.*3 × 105 CFU g−1 in

*M. officinalis *and

*C. nepeta*. Within species variability was far larger thanthat among species or among seasons. Apart from the fact that low values were recorded in summer, the marked seasonality ofthe Mediterranean climate was not reflected in the phyllosphere bacterial populations. Essential-oil producing species were notless colonized than the others. The hemicryptrophytes,

*M. officinalis *and

*C. nepeta*, shorter than all other species and equippedwith both glandular and non-glandular trichome, consistently sustained high bacterial populations on their leaves. Ice-nucleationactive (INA) bacteria were absent from all species, except for

*C. nepeta*, on which a very low population was recorded in winter.

Given these results and additional literature information, we argue that perennial species of Mediterranean-climate areas are notsystematic hosts of INA bacteria. We tested the statistical distribution of the phyllosphere bacterial populations of these speciesfor lognormality using the Kolmogorov–Smirnov (K–S) test. Using the Akaike information criterion (AIC), we compared thelognormal, gamma and Weibull distributions, as well as their compound (Poisson-sampled) versions. The results clearly favourthe lognormal hypothesis for these data. This may have important implications for our understanding and interpretation ofbacterial population dynamics.

2004 Elsevier B.V. All rights reserved.

*Keywords: *Phyllosphere bacteria; INA bacteria; Mediterranean shrubs; Aromatic plants; Trichome; Lognormal distribution

**1. Introduction**
The total leaf area of terrestrial vegetation amounts
Corresponding author. Tel.: +30-2310-998323;
Nevertheless, interest in leaves as a distinct ecologi-

*E-mail address: *[email protected] (D. Vokou).

cal milieu and potential habitat of microbes has only
0098-8472/$ – see front matter 2004 Elsevier B.V. All rights reserved.

doi:10.1016/j.envexpbot.2004.01.004

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
This finding provided a basis for the development of
The leaf is unique as a habitat for epiphytic microbes.

appropriate sampling techniques for epiphytic bacteria
It is ephemeral, lasting only few weeks for many an-
and also important insights into their epidemiology.

nuals up to a few years for woody perennial species;
it is highly dynamic, exposed to pronounced cyclic
and non-cyclic variation of environmental conditions
rial populations of the rhizosphere also approximated
a lognormal distribution. However, this is not univer-
edge is more or less arbitrarily defined, the leaf mar-
gin clearly delineates the boundary of the microbial
found that the Weibull distribution gave better agree-
community. Epiphytic communities are subjected to
ment with their observations for the bacterial popula-
frequent and rapid alternation between highly varying
tions on the leaves of

*Phaseolus vulgaris*. Departures
states of environmental factors, such as temperature,
from lognormality had also been observed by
relative humidity, wind speed, radiation, any one of
which may be considered stressful to at least some
for using lognormal distribution in this context has
High variability in space and time is a typical fea-
not been formally developed beyond a hypothesis
ture of epiphytic bacterial populations. Population
that bacterial counts are the results of unidentified
sizes can vary enormously even among adjacent, visu-
ally identical leaves of the same individual
argued that reasons governing frequency distributions
in nature usually favour the lognormal distribution.

tions of

*P. syringae *pathovar

*mors-prunorum *on indi-
Representation of large bacterial populations by a
vidual cherry leaves ranged from the non-detectable
continuous distribution as the lognormal, in spite of
to 1

*.*3 × 105 CFU per leaf within a set of 60 individ-
the discrete nature of bacterial cells, is a convenient
ual leaves. As a consequence of this high variability,
approximation that facilitates mathematical and sta-
he suggested the testing of large numbers of leaves if
tistical analysis. The advantage of this distribution is
bulked samples are being used. Apart from problems
that by making a simple logarithmic transformation of
arising from our limited capacity in elucidating the
bacterial counts obtained from a leaf sample or bulked
sources of such variability, there are also method-
samples, individual data points will follow a normal
ological problems associated with it, regarding both
collecting and analysing relevant data.

prevalent but inconsistent reporting of lognormality
Accurate estimates of bacterial population sizes are
mentioned above, the current state of our knowledge
needed if we are to safely predict phenomena related
to them, including, for example, disease or frost dam-
Because of their impact on world-wide agriculture,
age. Because of the variability of epiphytic bacterial
most studies of the microbiology of the phyllosphere
populations in space and time and the procedural com-
concern phytopathogens, and concomitantly cultivated
plexity of their estimation, quantitative studies of phyl-
plants, mostly with short-lived leaves, which account
losphere bacterial populations must address the statis-
tical distribution (i.e. probability density function) of
these populations. This is very important because this
may offer new insight as to the importance of the
distribution dictates the statistical methods that can be
phyllosphere as a habitat for microorganisms. Further-
employed. Apart from the statistical issue, knowledge
more, studies of the microbiology of leaves of peren-
of the distribution may cast light upon the type of
nial or non-deciduous plants may reveal patterns in
dynamics controlling the population. This is because
the ecology of the phyllosphere microbial community
different types of distributions are associated with dif-
Mediterranean ecosystems have not been consis-
tently investigated as natural habitats for epiphytic
time that epiphytic bacterial populations of various
microorganisms in general, and bacteria in particu-
crop plants approximated a lognormal distribution.

lar. There is only scarce information regarding fungi

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
a Mediterranean ecosystem, to examine whether the
features differentiating them play a decisive role in
determining epiphytic bacterial colonization, and fur-
a few species found in them. The Mediterranean en-
ther to assess the lognormal model and compare it
vironment is unique in that factors favouring growth
with a number of other models so as to have an under-
do not coincide for a considerable part of the year. In
standing of the distribution of phyllosphere bacterial
summer, when the prevailing high temperatures can
populations on Mediterranean perennial plants.

enhance growth, water is scarce. When water is avail-able, temperatures are not always favourable. Maquis(syn. choresh in Israel, monte bajo in Spain, macchia

**2. Materials and methods**
in Italy, chaparral in California) and phrygana (syn.

batha in Israel, tomillares in Spain, gariga in Italy,
coastal sage in California) are major ecosystem typesin areas with Mediterranean-type climate. They are
Our study site is a typical Mediterranean ecosystem,
dominated by woody species with different prominent
in Sithonia peninsula, Halkidiki (northern Greece).

features: evergreen and sclerophyll in maquis, season-
Eight perennial plant species of different life form,
ally dimorphic in phrygana. A number of features of
major components of this ecosystem, were chosen as
their component species can be considered of adap-
our experimental material Four species (

*A.*
tive nature to face the shortage of water during the

*unedo*,

*Q. coccifera*,

*P. lentiscus*,

*M. communis*) make
the group of evergreen-sclerophyll species and are
of the species are also aromatic; in fact, 49% of the
main components of maquis. Another four produce es-
genera with representatives falling within the group
sential oils (

*C. nepeta*,

*L. stoechas*,

*M. communis*, and
of aromatic plants are found in Mediterranean-climate
We made a total of 10 samplings during a 2-year
world where this climatic type prevails.

period (November 2000–November 2002). Sampling
In this paper we study the size of the phyllosphere
always took place early in the morning. Samples were
bacterial population of eight co-occurring perennial
placed in sterile plastic bags, were transported to the
Mediterranean species that differ in a number of
laboratory in an icebox, and were analysed within
morphological features, mainly in plant habit and
24 hrs. The sampling protocol was the following:
leaf trichome, glandular or non-glandular. We fur-
Three individuals from each species were marked.

ther examine the statistical distribution of bacteria
Mature leaves were collected at random from each
on their leaves. Our objective is to estimate the level
marked individual. Leaf samples consisted of one leaf
of bacterial colonization of species participating in
for

*A. unedo*, which has the largest leaves, while for all
Table 1Maximum, median (log CFU) and average values [log(CFU + 1)] over all sampling dates of total bacterial populations on the leaves ofeight Mediterranean perennial species, their life form and essential oil content of their leaves (average ± standard error)
The minimum values are not given as for all species, there were samples with non-detectable bacterial populations.

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
other species, they consisted of more, approximating
the area of one

*A. unedo *leaf; three or five leaf sam-
sprayed with 3% paraffin in xylene, was folded into
ples were collected per individual. Thus, there were
a flat-bottomed boat, which was floated on polyethy-
always 9 or 15 samples from each species. An excep-
lene glycol solution maintained at −5 ◦C in a refrig-
tion was for April 2002; for this date, there were no
erated constant temperature bath. Discrete fluorescent
samples of

*C. incanus*,

*C. nepeta*,

*L. stoechas *and

*M.*
colonies from KB plates were removed with sterile

*officinalis*, whereas for the rest, four leaf samples were
inoculation loop and suspended in 0.25 ml phosphate
collected from the marked three plus an additional
buffer to yield a turbid suspension. Twenty 10 l
individual (in total 16 leaf samples). In consequence,
droplets of suspension from each colony were placed
there are in total 76 data sets (8 plant species × 10
on the −5 ◦C test surface. A colony was considered
sampling times − 4 species in one sampling time).

to contain nucleus activity at −5 ◦C, if half or more
A meteorological station close to our study site had
been operating only for the years 1968–1975. To havean estimate of the climatic variability during the study
period, we used data from the meteorological stationof Thessaloniki, the closest major city. We concluded
Air-dried leaves (30–50 g) of

*C. nepeta*,

*L. stoechas*,
that data from this station adequately reflect the cli-

*M. communis*, and

*M. officinalis *were water-distilled
matic situation in the study area because comparison
in a Clevenger apparatus for three hours. Essential
of temperature data from the two stations over the pe-
oil content was expressed in ml per 100 g of dry leaf
riod 1968–1975 showed a very high degree of correla-
tion (

*R *= 0

*.*99,

*P < *0

*.*001), while rainfall data werealso considerably correlated (

*R *= 0

*.*54,

*P < *0

*.*001).

CFU data were log transformed (log10) and sub-
jected to the Kolmogorov–Smirnov (K–S) test for nor-
mality for each species and for each season. Since
no bacterial population was detected in several sam-
placed in 100-ml Erlenmeyer flasks containing 25 ml
ples, integer 1 was added to all CFU counts before
sterile phosphate buffer (0.01 M, pH 7.3) supple-
log transformation. These data were also standardized
mented with 0.1% peptone, and washed for 10 min
to a mean of zero and a standard deviation of unity.

on an ultrasonic cleaner; the temperature of water did
In order to improve the power of our tests, the stan-
not exceed 20 ◦C. Portions (100 l) from the original
dardised data sets of each species were pooled over all
wash and appropriate serial dilutions thereof, prepared
seasons, and those of each season were pooled over
in 0.01 M phosphate buffer (pH 7.3), were plated onto
all species. These were then also checked for normal-
a) nutrient agar medium supplemented with 2.5%
ity using the K–S test When the pooled set
(v/v) glycerol (NAG), b) Kings B medium (KB), both
of data is found to deviate from normality, this im-
amended with natamycin (30 g ml−1) to prevent
plies that one or more of the individual sets are not
fungal contamination. The dilutions examined were
1:1, 1:10 and 1:100. Total bacterial populations were
In order to compare the goodness of fit of the lognor-
enumerated from the NAG plates, and ice-nucleation
mal to our CFU data with that of other distributions,
active (INA) bacteria from the KB plates, after incu-
we used the Akaike information criterion (AIC). This
bation for 2–5 days at 24 ◦C; they were expressed as
is a likelihood-based comparison and is widely used as
colony forming units (CFU) per gram fresh weight.

an alternative to hypothesis-testing (The AIC works

*2.3. Testing for ice-nucleation active bacteria*
by comparing the data with a probability model: thelower the value of the AIC, the better the model fit.

Ice-nucleation activity of bacterial colonies, ob-
The models with which we compared the lognormal
tained from KB plates, was tested following the
were the Normal, gamma and Weibull distributions.

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
All of these are encountered commonly in biological
this procedure is bias. Specifically, it might happen that
situations. In addition, the lognormal and Weibull have
in small datasets like ours, the AIC shows a preference
been used in studies of populations of epiphytic bacte-
for the “wrong” distribution. For example, datasets
{

*k*1

*, k*2

*, . . . , kn*} generated from a gamma distribution
normal and Weibull, respectively). While the gamma
may on average yield a better fit (lower AIC) for a log-
distribution has not been used in the latter context, it
that the AIC could be biased towards one of the dis-
tributions, we also generated sets of simulated data,
of CFU = 0 is zero, the AIC cannot be computed
which act as “controls”. For each of the 76 data sets,
for datasets that include zeroes, and therefore we re-
10 replicates were generated using the estimated pa-
placed CFU by CFU + 1, as in the earlier analysis.

rameters associated with each of the six models. Thus,
Given that the number of zeroes in our data is substan-
there was one simulated group of data sets, containing
tial, such a transformation is open to legitimate criti-
760 simulated data sets, associated with each distribu-
cism. Also, these distributions, which are continuous,
tion. Each of these groups of data sets was subjected
neglect the basic fact that in practice we measure an
to the same testing procedure as the real data.

integer number of colonies. The best way of dealing
This maximum-likelihood fit provides an alterna-
with these concerns is to use the associated compound
tive means of estimating the average population values
models, Poisson-lognormal, Poisson-gamma and the
without recourse to the devices of adding 1 to CFU
before log-transforming (or ignoring 0 values entirely,
as many researchers do). It was, therefore, used as
set of distributions, as they are intrinsically integer
a check on the averages computed by our CFU + 1
valued, and in addition account for zero naturally. For
these reasons we used them in spite of the technicaldifficulties in their calculation, which have led to theirnot being so widely applied.

**3. Results**
The AIC takes into account the number of parame-
ters in each distribution. For all the distributions con-
The bacterial colonization of the phyllosphere of
sidered in this paper, the AIC has exactly two param-
the species examined was highly variable. Taking
all samples together (of all species and samplingtimes), the size of the phyllosphere microbial popu-
lation ranged from non-detectable up to a maximumof 1

*.*4 × 107 CFU g−1 in

*C. nepeta *For
where ˆ is the maximum value of the log-likelihood
all species, there were samples in which no bacterial
population was detected (

*A. unedo *contained
the largest proportion of samples with non-detectable
1

*)f(k*2

*)f(k*3

*) *· · ·

*f(kn)*]
bacterial populations (13.5%), whereas

*C. nepeta *and
for the given data set of CFU counts {

*k*1

*, k*2

*, . . . , kn*}

*M. officinalis *contained the smallest (3.9 and 1.5%,
and probability function,

*f*(

*k*), the calculation of which
respectively). Leaf samples with non-detectable bacte-
is explained in For most distributions of
rial populations were most frequent in summer; in four
interest, there is no formula that gives the parameters
species, for more than 30% of the August samples,
that maximize ˆ, so we used a Monte-Carlo search
there was no detectable population Given the
to find the best parameters. We first parameterize the
high number of zero values encountered in our data,
distribution crudely, then we search for the parameter
transformations of the form log(CFU) will cause seri-
values giving a higher ˆ, using the method described
ous concern. If we exclude zero values and transform
the rest, we ignore a considerable and valid segment
each of the 76 data sets for each of the models tested.

of our data. For this reason, we adopted the alternative
For each data set, the model with the lowest AIC was
solution of using the log(CFU + 1) transformation in
declared the ‘winner’. A possible complication with
order to calculate averages and standard errors.

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
Table 2Proportion of leaf samples (%) of the plant species studied with non-detectable bacterial populations, at each sampling date
The average total bacterial population of the phyl-
Analysis of variance showed that the species and
losphere (over all sampling times) ranged between
season factors accounted for only 27% of the to-
1

*.*3 × 104 CFU g−1 in

*A. unedo *and

*L. stoechas *to
tal variance of the phyllosphere bacterial population
1

*.*3 × 105 CFU g−1 in

*M. officinalis *and

*C. nepeta*,
[log

*(*CFU + 1

*)*]. This means that within-species vari-
varying by a factor of 10. Regarding the highest values
ability is far larger than that among species or among
recorded in leaves of each species, they ranged from
106 to 1

*.*4 × 107 CFU g−1 leaf (The leaves
Ice-nucleation active bacteria were practically ab-
of

*M. officinalis *and

*C. nepeta *consistently sustained
sent in this Mediterranean ecosystem. Only in one
high total bacterial populations throughout the study
case, in the February 2001 sampling, a low popula-
tion of INA bacteria was recorded on the leaves of

*C.*
hemicryptophytes in nature, are the most heavily col-

*nepeta *(0

*.*93 ± 0

*.*6 log CFU g−1).

onized, followed by

*Q. coccifera *and

*P. lentiscus*, two
ws the results of testing the null hypothe-
sis of lognormality (i.e. the normality of log(CFU+1)),
The average size of the phyllosphere microbial pop-
using the Kolmogorov–Smirnov goodness-of-fit test.

ulations of essential-oil producing plants ranged be-
For all the individual datasets, the null hypothesis was
tween 1

*.*3 × 104 and 1

*.*3 × 105 CFU g−1. The highest
accepted. However, when the standardized data were
value was recorded in

*M. officinalis*, which is the poor-
pooled by species, the hypothesis was rejected for
est in essential oil of the species examined (

*A. unedo*,

*M. communis *and

*L. stoechas*, and when
whereas the lowest in

*L. stoechas*, with an essential oil
they were pooled by season, it was rejected for Au-
concentration in its leaves more than 10 times higher
noting that species exhibiting the largest deviations
There was no obvious seasonal pattern of change
from lognormality were those for which the median
in the population of phyllosphere bacteria. Neverthe-
and the mean of log(CFU + 1) differed substantially
less, for most species, the summer values were among
(Given these deviations from lognormality,
the lowest (The climatic variability during the
there was a need to examine more closely the relation-
study period is shown in Analysis of data of
ship between the data and the models being used to
a 60-year period (1931–1990) show that August is
fit them. ws the results of the model com-
the driest month of the year with a mean rainfall of
parisons using the Akaike information criterion. The
14.3 mm; 2002 was an exceptional year in that rainfall
was unusually high during summer. Comparison of
tributions. We can see that the lognormal-generated
the two figures show that the bacterial populations on
data most closely approximate the results for the ob-
the phyllosphere do not follow the seasonal climatic
served data. Note that the AIC tends to be biased; it
discriminates in favour of the lognormal, even if the

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
Fig. 1. Total bacterial population [log(CFU + 1)] of the leaves of the Mediterranean species examined during the period November2000–November 2002. Error bars represent standard errors.

*A. unedo*,

*Q. coccifera*,

*P. lentiscus *and

*M. communis *are evergreen-sclerophyllspecies;

*L. stoechas*,

*C. nepeta*,

*M. officinalis *and

*M. communis *are aromatic plants.

data themselves are not lognormally distributed. If we
idence that the population itself is best described by
the lognormal distribution. Further evidence is sup-
clearer picture, since the AIC is not so biased. Here,
plied by the fact that the number of zeroes predicted
for the simulated data, the best fitting models corre-
by this model is consistent with the number observed.

spond to the models used to generate the data. The
There were a total of 93 zeros in the observed 952
Poisson-lognormally generated data are more similar
samples, and over the 10 simulated replicates an av-
to the real data than either those Poisson-gamma or
erage of 101.5 (ranging between 89 and 110). There-
Poisson–Weibull generated. This provides strong ev-
fore, the observed number of zeroes is well within the

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
Fig. 2. Climatic variability during the study period; data from the meteorological station of Thessaloniki.

typical range of behaviour for a sampled lognormal
an evergreen oak also participating in maquis. They
reported an average size of 3

*.*13 × 104 CFU cm−2.

This value is much higher than the respective one
adding 1 to CFU prior to taking logarithms, we tested
(after transforming it on a per-area basis) for

*Q. coc-*
the effects of these potentially distorting transforma-

*cifera *and may be due to the presence of tufted hairs
tion by comparison with results derived from the supe-
on the abaxial surface of

*Q. ilex*. However, results for
rior maximum-likelihood method. We found that the

*Q. ilex *are not in fact comparable to our results for
average error for each of the 76 population datasets

*Q. coccifera*, as the authors used an entirely different
was about 4% in excess of the value calculated by
technique to recover bacteria from the leaves.
maximum likelihood (and never exceeded 10% in any
a minimum of 1

*.*7 × 104 CFU cm−2 and
dataset). Thus, we concluded that results in
a maximum of 3

*.*4 × 105 CFU cm−2 on the leaves of

*Olea europaea *L. ‘coratina’. The olive tree, in its wildform, is also a typical component of maquis. Thesevalues are much higher than those of the four related

**4. Discussion**
species that we examined. There are two possiblereasons explaining this higher colonization of the
In the present study we examine the bacterial col-
olive leaves. (a) Our study was carried out in a natu-
onization of the phyllosphere of eight Mediterranean
ral ecosystem without major disturbances. Ercolani’s
species, components of a Mediterranean ecosys-
study area was an olive grove subjected to cultivation
tem, that differ in a number of features. For the
techniques which may affect the concentration of air-
evergreen-sclerophyll species, the average size of
the phyllosphere bacterial community ranged from
(b) Olive leaves bear non-glandular scales on
1

*.*3 × 104 to 4

*.*6 × 104 CFU g−1, with

*A. unedo *and

*M. communis *being less colonized than

*Q. coccifera*
chomes have been repeatedly suggested to consist
appropriate sites for bacterial colonization (
surfaces of

*Quercus ilex*, which, like

*Q. coccifera*, is

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
of 3.8–4.7 log CFU g−1. For

*Q. coccifera*, it fluc-
Normality of total bacterial population [log(CFU + 1)] of the
tuated between a minimum of 2.5 to a maximum
phyllosphere of (a) each species pooled for all seasons and(b) each season pooled for all species, as determined by the
of 5.2 log CFU g−1, thus extending even further of
the lower end of the range reported for the genus,possibly reflecting the more arid conditions of the
Of all species, the hemicryptophytes

*C. nepeta*
and

*M. officinalis *were the most heavily colonized
(105 CFU g−1), sustaining high bacterial populations

**0.007 **(127)

**0.014**
throughout the study period. According to
ground usually have bacterial populations larger and

**0.003 **(127)

**0.006**
**0.02 **(111)

**0.04**
less variable in size than taller ones. This is because
leaves closest to the soil surface are likely to expe-
rience more uniform physical conditions and higher
relative humidity than are leaves located at a range
becomes a more direct source of epiphytic bacteria.

Because of their life form,

*C. nepeta *and

*M. officinalis*
are shorter than all other species. In addition, they are
equipped with leaf hairs, glandular and non-glandular.

The combination of these factors could explain the
fact that they consistently sustain high bacterial pop-

**0.007 **(120)

**0.014**
Essential-oil producing plants differed in the size of
the microbial populations on their phyllosphere. Re-

**0.001 **(120)

**0.002**
garding the essential oil in their leaves,

*L. stoechas*
and

*C. nepeta *had on average similar concentration,
1.5 ml 100 g−1 d.w., whereas

*M. officinalis *and

*M.*

communis had a much lower concentration, of 0.12
Given are the probability values obtained from the K–S test. Valuesin bold italics are significant at

*P < *0

*.*05. The number of samples
and 0.28 ml 100 g−1 d.w., respectively. In spite of their
after they were pooled is given in parenthesis. Since each datum
resemblance in terms of the essential oil concentra-
participates in two tests, a BF correction requires

*P*-values to be
tion in their leaves,

*L. stoechas *and

*C. nepeta *differed
increased by a factor of 2. We show this for significant values, all
markedly in the bacterial colonization of their phyllo-
sphere;

*C. nepeta *(along with

*M. officinalis*) was mostheavily colonized and

*L. stoechas *least colonized.

of the evergreen-sclerophyll species that we studied
In another study of four Mediterranean aromatic
are waxy and devoid of hairs (personal observations).

plants, different to the ones that we examined
plants tend to have lower bacterial population than
tions on their phyllosphere ranged between 2.4 and
5.4 log CFU g−1. This range is comparable to the one
Various other

*Quercus *species, which unlike

*Q.*
we found here, yet extended at both ends. Also, like

*coccifera *are deciduous, have been studied regard-
in this study, the richer in essential oil species were
ing the bacterial colonization of their phyllosphere.

Essential oils are known to possess antimicro-
reported a total bacterial population for

*Q. virgini-*
*ana *of 5.7 log CFU g−1, for

*Q. macrocarpa *a range
of 3.8–6.4 log CFU g−1, and for

*Q. rubra *a range
phyllosphere microbial population, this activity is

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
Table 4Goodness of fit (with AIC) to observed and simulated data for different (a) simple distribution models and (b) compound distribution models
The values (given as percentages) are the proportion of data sets, for which the corresponding model fitted best. Numbers in parenthesesare the numbers of data sets.

not clearly expressed. Essential-oil producing species
and

*C. incanus*. Apart from these, the marked season-
were not on average less colonized than non-producing
ality of the Mediterranean climate was not reflected in
species. Also, in some cases, species richer in es-
the phyllosphere bacterial populations.

sential oil were more heavily colonized than poorer
Reported average population densities of epi-
ones. These findings show that the nature of the re-
phytic bacteria cover a huge range, from 104 to
lationship between essential oils and microbes is not
a simple one. The qualitative and quantitative com-
are often encountered in sizes averaging 106 to
position of the essential oil may be very important,
107 CFU cm−2 (up to 108 CFU g−1) of leaf (
as some compounds have much higher antimicrobial
also found on the leaves of the species that we
may be equally or even more important in determin-
examined but only as maximum values; averages
ing the level of microbial colonization. In addition,
never attained these levels. According to
although the antimicrobial activity of essential oils
leaves in the field are subject to bacterial
is very well known, it is not universal against all
immigration of about 104 cells per month, and there-
microbes. A number of them can use isoprenoid com-
fore a substantial proportion of epiphytic bacteria are
pounds making the mixture of essential oils as carbon
just temporary immigrants and not real colonizers.

Thus, he argued, plant species with epiphytic bacterial
populations less than 104 CFU per average sized leaf
some phyllosphere bacteria also have such a property.

should be only incidental hosts. All eight species that
The lowest phyllosphere bacterial populations were
we examined supported population sizes higher than
recorded for most species in summer. Others have
this threshold, but there are numerous instances where
observed this pattern for some but not all perennial
the population falls well below this level, not only for
individual leaves but also for monthly averages.

example, no significant differences were found in the
As for the ice-nucleation active bacteria, their pres-
size of bacterial populations on the mango phyllo-
ence was incidental; only on one species and only on
sphere, in South Africa, among the different seasons
one sampling date did we find any of them, and this
in very low numbers. Various authors suggest a broad
with non-detectable bacterial populations were most
distribution of INA bacteria on leaf surfaces (
frequent in summer, accounting even for 40% or more
of the samples of the species

*A. unedo*,

*Q. coccifera*,

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
on the leaves of woody species belonging to the genus
of the AIC (in favour of the lognormal) at low sam-

*Quercus*. For example,

*Q. macrocarpa *had INA bacte-
ple sizes. A second possible cause of departures from
rial populations ranging from 2.1 to 5.9 log CFU g−1.

lognormality is due to the fact that our data come
However, in another study of four woody species oc-
in the form of integers (including a large number of
curring in Mediterranean-type ecosystems (all four be-
zeroes), whereas the simple distributions examined
ing representatives of aromatic plants), INA bacteria
are continuous and do not allow zeroes. The only
were found in half of them, and again in very low num-
reliable way to deal with this was to upgrade the
bers (

*<*0.5 log CFU g−1) ().

simple models (lognormal, gamma, Weibull) to com-
These results combined provide substantial evidence
pound models (Poisson-lognormal, Poisson-gamma,
that perennial species of Mediterranean climate areas
Poisson–Weibull), where the sampling process is ac-
are not systematic hosts for INA bacteria.

tually included in the distribution. Using the com-
Authors have been divided on whether the lognor-
pound distributions, a very clear picture in favour of
mal hypothesis can be accepted as a general rule for
a lognormally-distributed population emerged on the
epiphytic bacterial populations. Since most of our data
basis of the AIC. The number of zeroes was also con-
sets are small, departures from lognormality were not
sistent with the picture that would be expected with a
always easy to detect. They showed up only when we
lognormal distribution of the epiphytic bacterial pop-
standardised and pooled our data, either by species
ulation. Thus, we believe that the zeroes arise mostly
or by season. Two possible causes could be respon-
from the sampling process. They are more a conse-
sible for departures from lognormality. Firstly, there
quence of the failure of the measurement process to
might be a systematic departure from lognormality to-
detect very low populations rather than an indication
wards an alternative distribution (gamma or Weibull),
of true absence of bacteria on leaves.

which suggested to us to use the AIC for compar-
The incidence of lognormality is a matter of
ing different distributions. Using this technique, we
importance, because it provides clues as to the
found that the lognormal was somewhat better fit than
population-dynamic processes that drive the sys-
the other plausible alternatives. However, the picture
was not a clear one, mainly due to the strong bias
(1988) explained how population models can give
Fig. 3. A model of bacterial population growth in a limited environment, subject to normally-distributed environmental variability. Theheavy line is the average population size for 1000 simulated population trajectories using the logistic approximation (while the inset panels are histograms of population size, for three different times. Notice that although the variability is normal, in therapid-growth phase the distribution is lognormal.

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
rise to lognormal distributions. An example of this is
fully understood yet, in semi-arid environments as the
shown in which is a model of a population
that grows in size from a small colony to a popu-lation that fully exploits the available resources, ac-cording to the well-known logistic model

**Acknowledgements**
The expected probability distributions associ-ated with population growth are also shown. The de-
The authors wish to thank Dr. Pablo Inchausti for
mographic rates in this model are normally-distributed
helpful suggestions. R.K.P. Yadav was supported by a
grant from the State Scholarships Foundation (IKY),
growing slowly at first, its distribution is almost nor-
Greece. He is currently on study leave from Tribhuvan
mal. Once it enters the phase of rapid exponential
University, Nepal. This project was also supported by
growth, the distribution becomes skewed and typically
the Greek Ministry of Environment (01 ED 317).

lognormal. However, when it finally saturates at thecarrying capacity of the environment, it returns to thesymmetric normal form once again. The lognormal

**Appendix A. Calculation of compound**
distribution, in most models, tends to be associated

**distributions**
with periods of temporary exponential growth, whilepopulations that have reached “equilibrium” with
Various models useful for describing population
their environment tend to have a more symmetric
sizes, such as the lognormal, gamma and Weibull dis-
distribution. In a large study of 544 ecological time
tributions, run into the problem that often the estimates
of population size appear as a series of low-valued
ecological distributions were less skewed than log-
integers, even though the population itself may be
normal and that the gamma distribution was at least
very large. This is the case for the data appearing in
as good a fit as the lognormal. This is in contrast to
this paper. To estimate a bacterial population size

*X*,
the distributions found for bacteria on leaves, which
we measure a number

*k *of colony-forming units. In
conform more closely to the lognormal distribution.

effect, we are measuring a small proportion

*δ *of the
This may reflect a more ephemeral “boom and bust”
total population. Thus

*k *is approximately

*δX*, rounded
type of population-dynamics than for macroscopic
to the nearest integer. So, if

*k *is simply re-scaled by
organisms. Bacterial populations, dominated by ex-
the factor 1/

*δ*, we have an estimate of population

*X*,
ponential growth at first, tend to be wiped out before
namely

*X *≈

*k/δ*. But information is lost, especially
when the population is low, because different pop-
In concluding, the perennial species that we ex-
ulations are represented by the same integers. For
amined, commonly found in Mediterranean ecosys-
example, if

*δ *= 10−3, then all populations between
tems, are not systematic hosts for INA bacteria; often
500 and 1500 are represented by

*k *= 1.

they are not systematic hosts for epiphytic bacteria
However, the really serious problem is that all pop-
either. The total bacterial population on their phyllo-
ulations less than 500 are undetectable, because the
sphere is low during summer. A number of plant fea-
nearest integer is zero. In such cases,

*X *is estimated to
tures are involved in determining the size of epiphytic
be zero. This creates big problems, when testing for
bacteria on their leaves, as is the leaf trichome. Yet,
distributions such as the lognormal, for which we need
plants with waxy cuticled leaves (e.g.

*Q. coccifera*) are
to calculate log(

*X*). The usual way of getting around
not always characterized by lower phyllosphere bac-
this problem, is by replacing

*k *with

*k *+ 1. This dis-
terial population than rough, trichomatic leaved plants
torts the distribution to an extent which is acceptable
(e.g.

*C. incanus*) that even produce essential oil in
only when the number of zeroes in the data is small
fair quantities (e.g.

*L. stoechas*). Regarding the sta-
tistical distribution of phyllosphere bacteria, our re-
When the aim is to compare different distribution
sults clearly support the lognormal hypothesis. This
models or to test whether a sample is of a given distri-
may have important implications about the underly-
bution, the best solution is to estimate the compound
ing population dynamic processes that are acting, not
distribution. This is the distribution of

*k *that arises

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
when small proportion of the population

*X *(with a
values (

*k > *20) we used the perturbation approxima-
given distribution) is used as a sample (
If the population size itself can be described by theprobability density

*g*(

*x*) (e.g. the lognormal), then the
probability that a randomly chosen leaf will be repre-
sented by a count

*k *in the data is given by a probability
e−

*λ g(λ)*d

*λ, k *= 0

*,*1

*,*2

*,*3

*. *(A.1)

*A.2. Gamma and Poisson-gamma (negative binomial)*
The distributions thus formed are referred to as the
The gamma random variable, with shape parameter
corresponding Poisson-sampled distributions. Finding

*c *and scale parameter

*b *has the probability density
the values

*f(*0

*), f(*1

*), f(*2

*), f(*3

*)*, etc. requires calcu-
lating the integral in (A.1), which often is not possible
in closed form. It is usually technically difficult and
most of the rest of this appendix is devoted to speci-
The parameters

*c *and

*b *are estimated from the mean

*M *and variance

*V *of the sample data by the formulas

*A.1. Lognormal and Poisson-lognormal*
*M *=

*bc *and

*V *=

*b*2

*c*, respectively.

For the Poisson-gamma, we use the formula (A.1)
If a random variable

*X *has a lognormal distribution,
with

*g(x) *=

*G(x)*. The Poisson-gamma is, in fact,
with parameters

*µ *and

*σ*, then its probability density
The parameters

*µ *and

*σ *are the mean and standard
deviation of the corresponding normal distribution,which is obeyed by the random variable

*Y *= ln

*k*.

which means that the values

*f*(

*k*) for the observed sam-
1,

*k*2,

*. . . *,

*kN *} of a lognormal
distribution,

*µ *and

*σ *can be found simply by takingthe natural logarithm of all the observations {ln

*k*
*A.3. Weibull and Poisson–Weibull distribution*
ln

*k*2,

*. . . *, ln

*kN*} and getting the mean and standarddeviation of these.

The Weibull random variable, with shape parameter
For the Poisson-lognormal, we use the formula

*c *and scale parameter

*b *has the cumulative distribution
(A.1) with

*g(x) *=

*L(x)*. However, for programming
function (i.e. the probability of a value less than or
purposes, it is best to transform the Poisson-lognormal
integral (using

*φ *= log

*λ*) to the form (

*k *=

*F(k) *= 1 − exp −

*b*
The probability density function is found by differen-
tiating (A.7). A crude estimate of the parameters

*c *and
− e

*ϕ *− ln

*(k*!

*) *d

*ϕ*
*b *is found first by rearranging (A.7):
ln

*k *= − 1

*c*ln[ln

*(*1 −

*yk)*] + ln

*b*
For small values of

*k*, this integral can be evaluatednumerically (We carried out the
We can find 1

*/c *and ln

*b *by regressing ln

*k *against
integration between the limits −5 and +5. For larger

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
For the Poisson–Weibull, we use the formula (A.1)

**References**
with

*g*(

*x*) the function found by differentiating (A.7).

The Poisson–Weibull is thus given by
Anderson, D.R., Burnham, K.P., Thompson, W.L., 2000. Null
hypothesis testing: problems, prevalence and an alternative. J.

exp

*(*−

*(*

*/b)c) *d

*λ*
Andrews, J.H., 1996. Phyllosphere ecology: past, present and
future. In: Morris, C.E., Nicot, P.C., Nguyen, C. (Eds.),Aerial Plant Surface Microbiology. Plenum Press, New York,
Andrews, J.H., Harris, R.F., 2000. The ecology and biogeography
As in the case of the Poisson-lognormal, for small
of microorganisms on plant surfaces. Annu. Rev. Phytopathol.

values of

*k*, we must use numerical integration. The
best results were found by integrating (A.8) by parts
Beattie, G.A., Lindow, S.E., 1995. The secret life of foliar bacterial
and as before using a transformation of the variable
pathogens on leaves. Annu. Rev. Phytopathol. 33, 145–172.

Belaˆıche, P., 1979. Traité de phytothérapie et d’ aromathérapie.

L’ aromatogramme. Saint-Just-la-Pendue pour les Editions

*F(*e

*ϕ)*e

*ϕ *−

*k*
Bosabalidis, A.M., Kofidis, G., 2002. Comparative effects of

*k *exp[

*kϕ *− e

*ϕ *− ln

*Γ(k)*] d

*ϕ,*
drought stress on leaf anatomy of two olive cultivars. Plant
Bulmer, M.G., 1974. On fitting the Poisson-lognormal distribution

*F(*e

*ϕ)*exp[

*ϕ *− e

*ϕ*] d

*ϕ*
to species-abundance data. Biometrics 30, 101–110.

Burrage, S.W., 1976. Aerial microclimate around plant surfaces.

In: Dikinson, C.H., Preece, T.F. (Eds.), Microbiology of Aerial
For larger values of

*k*, the perturbation procedure
Plant Surfaces. Academic Press, New York, pp. 173–185.

does not work as well for this distribution as it does
Crosse, J.E., 1959. Bacterial canker of stone fruits. IV. Investigation
for the lognormal. However, for values greater than
of a method for measuring the inoculum potential of cherry
trees. Ann. Appl. Biol. 47, 306–317.

30, the error incurred by approximating the
Daniel, J., Boher, B., 1978. Ecology of cassava bacterial
Poisson–Weibull directly by the Weibull distribution
blight epiphytic survival of

*Xanthomonas manihotis *on
itself is very small, and this was the approach we used.

aerial parts of the cassava plant. In: Station de PathologieVégétale et Phytobactériologie, Proceedings of the InternationalConference on Plant Pathology and Bacteria, fourth ed. Instituteof National Research on Agriculture, Beaucouzé, pp. 763–771.

**Appendix B. Using the Akaike information**
de Jager, E.S., Wehner, F.C., Korsten, L., 2001. Microbial ecology

**criterion (AIC)**
of the mango phylloplane. Microb. Ecol. 42, 201–207.

Dennis, B., Patil, G.P., 1988. Applications in ecology. In: Crow,
The AIC involves finding maximum-likelihood dis-
E.L., Shimizu, K. (Eds.), Lognormal Distributions: Theory andApplications. Marcel Dekker, New York, pp. 303–330.

tributions. For the continuous distributions, this may
Diserud, O.H., Engen, S., 2000. A general and dynamic species
abundance model, embracing the lognormal and the gamma
above have two parameters. We first estimate their val-
Ercolani, G.L., 1991. Distribution of epiphytic bacteria on olive
ues crudely, using the methods described above, and
leaves and the influence of leaf age and sampling time. Microb.

then carry out a Monte-Carlo search of the neighbor-
Fisher, R.A., Corbet, A.S., Williams, C.B., 1943. The relation
hood of the values until we find those that minimize
between the number of species and the number of individuals
in a random sample from an animal population. J. Anim. Ecol.

The compound distributions, although more com-
plex, also have only two parameters, which are
estimated in the same way. Thus, for the Poisson-
Sissions, D.J., Trudgill, P.W., 1987. Bacterial metabolism of
lognormal, as starting values for parameters

*µ *and
alpha-pinene: pathway from alpha-pinene oxide to acyclic
metabolites in

*Nocardia *sp. strain P18.3. J. Bacteriol. 169,
found to minimize the AIC for the simple lognormal
Halley, J., Inchausti, P., 2002. Lognormality in ecological time

*R.K.P. Yadav et al. / Environmental and Experimental Botany xxx (2004) xxx–xxx*
Hirano, S.S., Nordheim, E.V., Arny, D.C., Upper, C.D., 1982.

Margaris, N.S., 1981. Adaptive strategies in plants dominating
Lognormal distribution of epiphytic bacterial populations on
Mediterranean-type ecosystems. In: di Castri, F., Goodall,
leaf surfaces. Appl. Environ. Microbiol. 44, 695–700.

D.W., Specht, R.L. (Eds.), Mediterranean-type Shrublands.

Hirano, S.S., Upper, C.D., 2000. Bacteria in the leaf ecosystem
with emphasis on

*Pseudomonas syringae*—a pathogen, ice
Mariano, R.L.R., McCarter, S.M., 1993. Epiphytic survival of
nucleus and epiphyte. Microbiol. Mol. Biol. Rev. 64, 624–653.

*Pseudomonas viridiflava *on tomato and selected weed species.

Amaral-Collaco, M.T., Spencer-Martins, I., 2002. Estimation
Mew, T.W., Vera Cruz, C.M., 1986. Epiphytic colonization of
and diversity of phylloplane mycobiota on selected plants in a
host and non-host plants by phytopathogenic bacteria. In:
Mediterranean type ecosystem in Portugal. Microb. Ecol. 44,
Fokkema, N.J., van den Heuvel, J. (Eds.), Microbiology of
the Phyllosphere. Cambridge University Press, New York,
Ishimaru, C., Eskridge, K.M., Vidaver, A.K., 1991. Distribution
analyses of naturally occurring epiphytic populations of
Misra, G., Pavlostathis, S.G., Perdue, E.M., Araujo, R., 1996.

Aerobic biodegradation of selected monoterpenes. Appl.

Microbiol. Biotechnol. 45, 831–838.

Janssen, A.M., Scheffer, J.J.C., Baerheim-Svendsen, A., 1987.

Moleyar, V., Narasimbam, P., 1992. Antibacterial activity of
Antimicrobial activity of essential oils: a 1976–1986 literature
essential oil components. Int. J. Food Microbiol. 16, 337–342.

review: aspects of the test methods. Planta Med. 53, 395–
Morris, C.E., Kinkel, L.L., 2002. Fifty years of phyllosphere
microbiology: significant contributions to research in related
Karamanoli, K., Vokou, D., Menkissoglu, U., Constantinidou, H.-I.,
fields. In: Lindow, S.E., Hecht-Poinar, E.I., Elliott, V.J.

2000. Bacterial colonization of phyllosphere of Mediterranean
(Eds.), Phyllosphere Microbiology. APS Press, Minnesota,
aromatic plants. J. Chem. Ecol. 26, 2035–2048.

Kinkel, L.L., Wilson, M., Lindow, S.E., 2000. Plant species and
O’Brien, R.D., Lindow, S.E., 1989. Effect of plant species and
plant incubation conditions influence variability in epiphytic
environmental conditions on epiphytic population sizes of
bacterial population size. Microb. Ecol. 39, 1–11.

*Pseudomonas syringae *and other bacteria. Phytopathology 79,
distributions across the sciences: keys and clues. BioScience
Pereira, P.T., De Carvalho, M.M., Girio, F.M., Carlos Roseiro,
J., Amaral-Collaco, T., 2002. Diversity of microfungi in the
Lindemann, J., Constantinidou, H.A., Barchet, W.R., Upper, C.D.,
phylloplane of plants growing in a Mediterranean ecosystem.

1982. Plants as sources of airborne bacteria, including ice
nucleation-active bacteria. Appl. Environ. Microbiol. 44, 1059–
Périssol, C., Roux, M., Le Petit, J., 1993. Succession of bacteria
attached to evergreen oak leaf surfaces. Eur. J. Soil Biol. 29,
Lindow, S.E., 1986. Specificity of epiphytic interactions of

*Pseudomonas syringae *stranis on leaves. Phytopathology 76,
Press, W.H., Teukolsky, S.A., Vetterling, W.T., Flannery, B.P.,
1992. Numerical Recipes in C, second ed. Cambridge
Lindow, S.E., 1996. Role of immigration and other processes in
University Press, Cambridge, 129 pp.

determining epiphytic bacterial populations: implications for
Renshaw, E., 1991. Modelling Biological Populations in Space
disease management. In: Morris, C.E., Nicot, P.C., Nguyen, C.

and Time. Cambridge University Press, Cambridge, 50 pp.

(Eds.), Aerial Plant Surface Microbiology. Plenum Press, New
Ross, J.D., Sombrero, C., 1991. Environmental control of
essential oil production in Mediterranean plants. In: Harborne,
Lindow, S.E., Arny, D.C., Upper, C.D., 1978a. Distribution of ice
J.B., Tomas-Barberan, F.A. (Eds.), Ecological Chemistry and
nucleation-active bacteria on plants in nature. Appl. Environ.

Biochemistry of Plant Terpenoids. Clarendon Press, Oxford,
Lindow, S.E., Arny, D.C., Upper, C.D., 1978b.

*Erwinia herbicola*:
Rouse, D.I., Nordheim, E.V., Hirano, S.S., Upper, C.D., 1985. A
a bacterial ice nucleus active in increasing frost injury to corn.

model relating the foliar disease incidence to the population
frequencies of bacterial plant pathogens. Phytopathology 75,
Loper, J.E., Suslow, T.V., Schroth, M.N., 1984. Lognormal
distribution of bacterial populations in the rhizosphere.

Vardavakis, E., 1988. Seasonal fluctuation of non-parasitic
mycoflora associated with living leaves of

*Cistus incanus*,
McArdle, B., Gaston, K., Lawton, J., 1990. Variation in the size of

*Arbutus unedo *and

*Quercus coccifera*. Mycologia 80, 200–210.

animal populations: patterns, problems and artefacts. J. Anim.

Vokou, D., Margaris, N.S., 1988. Decomposition of terpenes by
soil microorganisms. Pedobiologia 31, 413–419.

Vokou, D., 1999. Essential oils as allelochemicals: research
microscopy of colonization of pear leaves by

*Pseudomonas*
advances in Greece. In: Narwal, S.S. (Ed.), Allelopathy Update.

*syringae *pv.

*syringae*. Can. J. Bot. 65, 2517–2522.

Science Publishers, Enfield, USA, pp. 47–63.

Source: http://www.jmax.gr/images/stories/pubs/33_JMH_2004b_LeafBacteria.pdf

School of Chemical and Biomedical Engineering Web: http://www.ntu.edu.sg/home/arvindDr. Sci. Tech. Process Engg. ETH Zurich, SwitzerlandM.Engg. Chem. Engg. National University of Singapore, SingaporeB.E Chem. Engg. Annamalai University, IndiaNanyang Technological University, SingaporeAssistant Professor, School of chemical and biomolecular engineeringSPIC Fertilizer Complex, Naptha base

DERMaTOlOGIa piante e sole: un mix E SE FOSSE allERGIa? In alcuni casi, meno frequenti, le stesse piante che provocano (a volte) pericoloso reazioni fototossiche possono indurre anche una reazione al- lergica. Ma che differenza c'è? a cura di Grazia Manfredi • La reazione fototossica dipen- de esclusivamente dal contatto con la sostanz