YEAST METABOLIC STATE IDENTIFICATION BY FIBER OPTICS

SPECTROSCOPY

C. C. Castro

1

, J. S. Silva

1

, V. V. Lopes

2

, R. C. Martins

3 ∗

1

IBB - Institute for Biotechnology and BioEngineering, Universidade do Minho

Campus de Gualtar, 4710-057 Braga, Portugal

2

INETI - Instituto Nacional de Engenharia Tecnologia e Inovac¸˜ao

Estrada do Pac¸o do Lumiar, 22, 1649-038 Lisboa

3

BioInformatics - Molecular and Environmental Biology Research Center, Universidade do Minho

Campus de Gualtar, 4710-057 Braga, Portugal

Keywords:

Saccharomyces cerevisiae, Morphology, LWUV-VIS-SWNIRreﬂectance spectroscopy, Singular value decom-

position, Classiﬁcation.

Abstract:

In this manuscript we explore the feasibility of using LWUV-VIS-SWNIR (340 - 1100 nm) spectroscopy

to classify Saccharomyces cerevisiae colony structures in YP agar and YPD agar, under different growth

conditions, such as: i) no alcohol; ii) 1 % (v/v) Ethanol; iii) 1 % (v/v) 1-Propanol; iv) 1 % (v/v) 1- butanol;

v) 1 % (v/v) Isopropanol; vi) 1 % (v/v) (±)-1-Phenylethanol; vii) 1 % (v/v) Isoamyl alcohol; viii) 1 % (v/v)

tert-Amyl alcohol (2-Methyl-2-butanol); and ix) 1 % (v/v) Amyl alcohol. Results show that LWUV-VIS-

SWNIR spectroscopy has the potential for yeasts metabolic state identiﬁcation once the spectral signatures

of colonies differs from each others, being possible to acheive 100% of classiﬁcation in UV-VIS and VIS-

SWNIR. The UV-VIS region present high discriminant information (350-450 nm), and different responses to

UV excitation were obtained. Therefore, high precision is obtained because UV-VIS and VIS-NIR exhibit

different kinds of information. In the future, high precision analytical chemistry techniques such as mass

spectroscopy and molecular biology transcriptomic studies should be performed in order to understand the

detailed cell metabolism and genomic phenomena that characterize the yeast colony state.

1 INTRODUCTION

Recent studies show that S. cerevisiae can form com-

plex colony structures with an apparent cell special-

ization. Colonies of wild yeasts can contain all the

varieties of cells, from which the mostly known are

the diploid, haploid, hyphae form (diploid or haploid)

and ascus (spore); opposing to the most well known

yeast cell cycle - the budding yeast. Furthermore, it

is known that S. cerevisiae can undergo changes in

their replicative patterns and morphologies, accord-

ing to environmental conditions (i.e., deleterious), to

produce elongated cells joined-together in ﬁlaments

(Dickinson, 2008) and colonies can signal each other

(Palkova and Vachova, 2003).

The yeast-form and ﬁlamentous-form cell cycles

are similar but, according to (Kron and Gow, 1995), in

yeast-form growth, daughter cells are smaller than its

mother and must undergo a period of further growth

(in phase G1) before starting a new cell cycle (asym-

∗

Corresponding author: rui.martins@bio.uminho.pt

metric cell division). On the other hand, ﬁlament-

form cells have a symmetric cell division, once after

mitosis and cell division, both mother and daughter

cells are equal-sized and bud emerge starts in both

cells. Furthermore mitochondrial mass and chitin

deposition increases in ﬁlament form. The ﬁlament

form cellular walls have a greater strength and rigid-

ity than most of the other yeast forms, which has been

suggested as a mean of penetration on solid media

because of the yeast lack of natural mobility. The

transcription of all of genes also decreases in ﬁlament

forming yeast, and therefore has been proven difﬁcult

to ﬁnd a direct transcriptomic relationship (Dickin-

son, 2008). Filament formation can be induced by

nitrogen starvation or limitation (Rua et al., 2001),

growth on a poor nitrogen source (Dickinson, 1994)

or growth in the presence of low concentrations of

fusel alcohols (end-products of cells catabolism). In

the case of nitrogen starvation or limitation, ﬁlamen-

tation can be explained as a foraging response be-

cause yeast is non-motile and cannot move to search

169

C. Castro C., S. Silva J., V. Lopes V. and C. Martins R. (2009).

YEAST METABOLIC STATE IDENTIFICATION BY FIBER OPTICS SPECTROSCOPY.

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing, pages 169-178

DOI: 10.5220/0001551201690178

Copyright

c

 SciTePress

Figure 1: Spectroscopy signalling of different yeast structures.

for a richer supply of nutrients, it can only grow to

explore its surroundings (Gimeno et al., 1992). Fil-

amentation has also been induced by superior alco-

hols and AMPc, and has been argued that these may

act as communication molecules between different

yeast, allowing for the colony to synchronize its de-

velopment, a phenomena known as quorum-sensing.

Therefore, the yeast colony state and its dynamics is

not yet explained. In order to understand this phe-

nomena, an non-targeted, holistic and high-output ap-

proach is needed in order to gather the maximum in-

formation as possible to understand colony dynamics

and yeast communication. In this sense, the use of

spectroscopy, in conjunction with other techniques,

allows to implement real-time and non-destructive

methodologies that can explain the transcriptomics

and metabolomics processes happening in yeast cel-

lular communities.

Microorganismstraditional identiﬁcation methods

are supported by morphological and growth capac-

ity in selective media (Gerard et al., 2006). The use

of high-output methodologies to increase analysis ca-

pacity, such as mass spectroscopy, PCR and spec-

troscopy are becoming popular, not only because of

the time needed for an effective identiﬁcation, but

more importantly because these methods are multi-

variate, which allows to obtain vast amounts of infor-

mation in one measurement (Rosah et al., 2005).

Spectroscopy is a simple, precise, rapid, multi-

variate and non-destructive technique. Spectra is pro-

portional to the chemical composition of the analite,

acting as a non-destructive methodology capable of

both ﬁngerprint and quantiﬁcations. In this case,

spectroscopy is may be able to classify the invari-

able structure of yeast, the metabolism and cell com-

munication. Cells morphology is a visible expres-

sion of microorganisms physiology and metabolism

(Treskatis et al., 1997). Different morphologies char-

acterize different proteomic composition and different

metabolism states that can be differentiated by UV-

VIS-SWNIR spectroscopy.

Many spectroscopy techniques have been used for

microorganisms identiﬁcation, where NIR and Ra-

man spectroscopy are the most popular (Stuart, 2004;

Dziuba et al., 2007; Bhatta et al., 2005). Recently, a

previous study revealed that UV-VIS-SWNIR is also

a highly accurate spectroscopy method for microor-

ganisms identiﬁcation (Silva et al., 2008). UV spec-

troscopy records electronic transitions between elec-

tron energy levels from molecular levels in the UV-

VIS region depend upon the energy involved. For

any molecular bound (sharing a pair of electrons),

orbitals are a mixture of two contributing orbitals σ

and π, with corresponding anti-bounding orbitals σ

∗

and π

∗

, respectively. Some chemical bounds present

characteristic orbital conditions, ordered by higher to

lower order energy transitions: i) alkanes (σ → σ

∗

;

150nm); ii) carbonyls (σ → π

∗

; 170nm); iii) unsat-

urated compounds (π → π

∗

; 180nm); iv) molecular

bounds to O, N, S and halogens (n → σ

∗

; 190nm);

and v) carbonyls (n → π

∗

; 300nm). As most UV-

VIS spectrometers yield a minimum wavelength of

200nm, this technique has been considered to pro-

vide lower information in terms of functional groups

when compared to IR, because spectral differences

mostly attributed to conjugated π → π

∗

and n → π

∗

transitions (Levine, 1975; Denney and Sinclair, 1987;

Perkauparus et al., 1994).

Many organic molecules present conjugated

unsaturated and carbonyls bounds, such as

aminoacids, phospholipids, free fatty acids, phe-

nols and ﬂavonoids, peroxides, peptides and proteins,

BIOSIGNALS 2009 - International Conference on Bio-inspired Systems and Signal Processing

170

sugars and their polymers absorb in these bands. UV-

VIS not only records the effect of electron excitation,

but also the effect of return to lower orbitals, which

result in vibrational and rotational modes, increasing

the characteristic spectra of biological materials.

This effect enhances photochemical reactions and

ﬂuorescence which are important features for micro-

biological identiﬁcation (Levine, 1975; Coyle, 1989;

Klessinger and Michl, 1995) and help to identify

metabolic states of yeast. Many biological molecules

also present chromophore groups, which increase

the absorption in the UV-VIS region, such as: nitro,

nitroso, azo, azo-amino, azoxy, carbonyl and thio-

carbonyl (Coyle, 1989; Klessinger and Michl, 1995).

Moreover the sensitivity of today’s spectrometers

has highly increased, being possible to obtain low

noise to signal ratios which expands the detection

limits (Optics, 2006). LWUV-VIS-SWNIR has as

main advantages the minimization of liquid water

absorbance and effect of temperature. Furthemore,

as state of the art spectrometers also include high

frequency vibrational infrared (SWNIR), it is also

possible to obtain important information on water,

fats and proteins (Burns and Ciurczak, 2001; Devices,

2005).

The main objective of this exploratory work is to

determine if UV-VIS-SWNIR is a suitable methodol-

ogy that may be used to recognize the state of S. cere-

visiae colonies by spectral signal processing to obtain

discrimination among different induced metabolism,

cellular communication, morphology and growth me-

dia.

2 MATERIALS AND METHODS

2.1 Sample Preparation

Saccharomyces cerevisiae wild type was obtained

from the microbiological collection of the IBB - In-

stitute for Biotechnology and BioEngineering at the

University of Minho.

The incubation was performed in YPD broth

medium (Sigma Aldrich - ref. Y1375) during 12

hours at 25

o

C under constant agitation (250 rpm).

Wild type yeast (20 µl) was inoculated on the surface

of YP and YPD agar mediums using different growth

conditions, such as: without alcohol and with 1 %

(v/v) of an alcohol and was incubated at 25 C dur-

ing 144 h. Studied alcohols were: Ethanol (Riedel-

de Han - ref. 32221), 1-Propanol (Sigma Aldrich

- ref. 538000), 1- Butanol (Sigma Aldrich - ref.

BT105), Isopropanol (Sigma Aldrich - ref. 190764),

(±)-1-Phenylethanol (Fluka - ref 09449), Isoamyl al-

(a) (b)

(i) (j)

(c) (d)

(e) (f) (g) (h)

(m) (n)

(k) (l)

(o) (p)

Figure 2: (a) YPD tert-amylOH; (b) YP tert-Amyl; (c) YPD

Ethanol; (d) YP Ethanol; (e) YPD without alcohol; (f) YP

without alcohol; (g) YPD Isopropanol; (h) YP Isopropanol;

(i) YPD Propanol; (j) YP Propanol; (k) YPD Phenylethanol;

(l) YP Phenylethanol; (m) YPD Butanol; (n) YP Butanol;

(o) YPD Isoamyl; (p) YP Isoamyl.

cohol (SAFC - ref. W205710), tert-Amyl alcohol

(2-Methyl-2-butanol) (Sigma Aldrich - ref. 152463)

and Amyl alcohol (SAFC - ref. 205605) (Sigma-

Aldrich Quimica, 2008).

Growth medium present the following consti-

tutions: YPD broth medium (Sigma-Aldrich ref.

Y1357): 10 g.l

−1

Yeast extract , 20 g.l

−1

Peptone and

20 g.l

−1

Glucose (Sigma-Aldrich Quimica, 2008);

YP agar: 10 g.l

−1

Yeast extract (Fluka - ref. 70161),

20 g.l

−1

Peptone (Bacto

TM

- ref. 211677) and 15

g.l

−1

Agar (Fluka ref. 05039) and YPD agar medium

(Sigma Aldrich - ref. Y1500): 10 g.l

−1

Yeast extract ,

20 g.l

−1

Peptone, 15 g.l

−1

Agar and 20 g.l

−1

Glucose

(Sigma-Aldrich Quimica, 2008).

Both agar medium were prepared according to the

indications of the manufacturer: i) suspension of the

dehydrated media in puriﬁed water (amounts deﬁned

by the manufacturer); ii) heating of the media, with

frequent agitation, until complete dilution; iii) auto-

clave of the mixture at 121

o

C for 15 minutes; and iv)

shed in petri plate (Sigma-Aldrich Quimica, 2008).

2.2 Spectroscopy

Saccharomyces cerevisiae UV-VIS-SWNIR spec-

troscopy was performed with: i) Avantes multi-

channel ﬁber optic spectrometer AvaSpec-2048-4-

DT (200 to 1100 nm; 2048 pixel) (Avantes, 2007);

ii) reﬂection UV-VIS and VIS-SWNIR probes,

YEAST METABOLIC STATE IDENTIFICATION BY FIBER OPTICS SPECTROSCOPY

171

300 350 400 450 500 550

0 1 2 3 4

(a)

Wavelength (nm)

Absorbance

300 350 400 450 500 550

−0.04 −0.02 0.00 0.02 0.04 0.06

(b)

Wavelength (nm)

First Derivative

500 600 700 800 900 1000 1100

0 1 2 3 4

(c)

Wavelength (nm)

Absorbance

500 600 700 800 900 1000 1100

−0.010 −0.005 0.000 0.005 0.010

(d)

Wavelength (nm)

First Derivative

Figure 3: Microorganisms spectra: (a) Absorbance spectrum LWUV-VIS; (b) First derivative spectrum LWUV-VIS; (c)

Absorbance spectrum VIS-SWNIR; (d) First derivative spectrum VIS-SWNIR.

models FCR-7UV200-2ME and FCR-7IR200-2-ME

(Avantes, 2007); and iii) a balanced deuterium-

tungsten halogen light source, model DH-2000-BAL

(Micropack, 2008). AvaSoft 6.0 was used to con-

trol the spectrometer and data acquisition (Avantes,

2007).

Spectra were obtained at the room temperature of

18 ±2

o

C and the light source in (a) UV-VIS: the deu-

terium lamp was let to stabilize during 20 min; and

(b) VIS-NIR: the tungsten lamp lamp was let to sta-

bilize during 15 min. The dark spectra was recorded

and measurements were taken with linear and elec-

tric dark correction. Both light spectra were moni-

tored by statistically assessing the reproducibility of

the light source with measurements of light during

the several days of the experiment. Twenty spectra

replicates were recorded of UV-VIS and VIS-SWNIR

measurement of both plate count agar and microor-

ganisms colonies to study scattering effects. Further-

more, spectra were obtained inside a box designed to

isolate the environmental light and maintain the probe

horizontally.

2.3 Spectral Analysis

2.3.1 Robust Mean Scattering Correction

The collected spectrum were smoothed by using a

Savisky-Golay ﬁlter (length = 4, Order= 2) (Sav-

itzky and Golay, 1964) and afterwards, was pre-

processed using a modiﬁed robust multiplicative scat-

ter correction algorithm (RMSC) (Gallager et al.,

2005; Martens and Stark, 1991; Martens et al., 2003):

x

corr

= xb + a. The a and b are computed by mini-

mizing the following error: e

j

= bx

j

+a− x

ref

; where

the x

j

is the j sample spectrum and x

ref

is the growth

media spectrum.

The RMSC algorithm is based on the application

of the robust least squares method to determine the

a and b matrices, ensuring that spectral areas that do

not correspond to scattering artifacts are not taken into

consideration. The robust least squares algorithm is

implemented by the re-weighted least squares with

the weights computed using the Huber function. The

algorithm high breakdownpoint (50%) means that ex-

istent outliers will not distort the model ﬁtting (eq. ??)

and thus, the a and b scatter correction parameters are

determined using only the consistent spectral areas.

The iterative algorithm can be described, brieﬂy as

follow: 1) set the reference spectrum (x

ref

) equal to

the sample spectrum closest to the median spectrum;

2) correct the remaining sample spectrum by applying

the above described robust least squares procedure;

and 3) recompute the median spectrum and iterate un-

til convergence.

2.3.2 Singular Value Decomposition

Singular value decomposition (SVD) is a blind signal

decomposition technique widely used in spectroscopy

data, where the corrected spectrum (x

corr

) is decom-

BIOSIGNALS 2009 - International Conference on Bio-inspired Systems and Signal Processing

172

−0.2 −0.1 0.0 0.1 0.2 0.3 0.4

−0.06 −0.04 −0.02 0.00 0.02 0.04

−0.15

−0.10

−0.05

0.00

0.05

0.10

PC1 (53.7 %)

(a)

PC2 (32.6 %)

PC3 (3.8 %)

$

(1)

(1)−

(2)

(2)−

YP Butanol

YP Control

YPD Butanol

YPD Control

YPD Ethanol

YPD Phenylethanol

YPD Isopropanol

YPD Propanol

YPD /Alcohol

YPD tert−Amyl

YP Ethanol

YP Phenylethanol

YP Isoamyl

YP Isopropanol

YP Propanol

YP /Alcohol

YP tert−Amyl

−0.02 0.00 0.02 0.04 0.06

−0.020−0.015−0.010−0.005 0.000 0.005 0.010

−0.020

−0.015

−0.010

−0.005

0.000

0.005

0.010

0.015

(b)

PC1 (63.8 %)

PC2 (14.0 %)

PC3 (2.5 %)

$

(1)

(1)−

(2)

(2)− YP Butanol

YP Control

YPD Butanol

YPD Control

YPD Ethanol

YPD Phenylethanol

YPD Isopropanol

YPD Propanol

YPD /Alcohol

YPD tert−Amyl

YP Ethanol

YP Phenylethanol

YP Isoamyl

YP Isopropanol

YP Propanol

YP /Alcohol

YP tert−Amyl

Figure 4: First derivative spectra PCA analysis: (a) LWUV-VIS Gabriel Plot (PC1 (78,40%), PC2 (8,03%); Symbols: i)

YPD terta-Amyl (⊕); ii) YP terta-Amyl ($); iii) YPD Ethanol (); iv) YP Ethanol (♦); v) YPD without alcohol (•); vi) YP

without alcohol (◦); vii) YPD Isopropanol (×); viii) YP Isopropanol (⊠); ix) YPD Propanol (N); x) YP Propanol (△); xi)

YPD Phenylethanol (⊞); xii) Phenylethanol (+); xiii) YPD Butanol (); xiv) YP Butanol (); xv) YP Isoamyl (▽); xvi) YPD

control (∗); xvii) YP control (⊗).

posed in order of magnitude of variation directions

in the variable space (wavelengths). Generally, most

variability is captured in the ﬁrst principal compo-

nents (PC), where as, in good signal to noise spec-

tral data, noise is captured in the last decomposi-

tions. Therefore a spectrum can be decomposed as:

x

corr

=

b

x+ ε(x); where

b

x is the signal and ε(x) is the

estimated noise of x. This decomposition is possi-

ble to be performed by singular value decomposition

(SVD):

X = USV

T

(1)

where X is a matrix that contains all the corrected

spectra, US are the scores, V

T

the loadings and the

S singular values (Jolliffe, 1986; Krzanowski, 1998;

Baig and Rehman, 2006).

To distinguish between the number of relevant de-

compositions, a randomization test is performed to

the original matrix (x) to determine the number of

relevant singular values (Manly, 1998). In this re-

search, 500 randomizations were performed by per-

mutation the spectral scope value for the same wave-

lengths among the different samples, to do not vio-

late the spectral continuity. By comparing the singu-

lar values of randomized spectrum with the original

spectrum, the number of independent singular values

and decompositionsthat discriminate the differentmi-

croorganisms spectrum are obtained, so that:

b

X = US

relv

V

T

relv

(2)

where US

relv

and V

T

relv

are the statistically rele-

vant scores and loading of X. To further discriminate

between the microorganisms spectrum, the relevant

PC’s scores(US

relv

) were subjected to hierarchical

clustering analysis using the eucledian distance. Fur-

ther class identiﬁcation was performed using soft in-

dependent class analogy (SIMCA) (Doytchinova and

Flower, 2006).

Not all features in the spectrum ﬁngerprint pre-

serve the same quality after signal decomposition into

relevant principal components. In these cases, their

reconstruction is statistically impossible. In practical

terms, features that are not compressed, cannot be an-

alyzed in the score plot. Feature extraction quality can

be assessed by the Q-statistic (square prediction error)

of the relevant decomposition developed by (Jackson

and Mudholkar, 1979):

Q

i

= e

i

e

T

i

(3)

where e

i

= x

i

−

b

x

i

is the reconstruction error for

the i

th

spectrum present in the data matrix (the i

th

row

of X). An accepted way of computing the Q statistical

conﬁdence interval (Q

α

) is deﬁned as:

Q

α

= θ

1

·



1+

Z

α/2

θ

1

q

2θ

2

h

2

0

+

θ

2

h

0

(h

0

− 1)

θ

2

1



1

h

0

(4)

where Z

α

is the inverse normal distribution value

for the signiﬁcance level (α/2), θ

j

=

∑

(S

i

)

j

and

h

0

= 1 −

2

3

θ

1

θ

3

/θ

2

(Choi et al., 2005). The sam-

ples with a Q-statistic above Q

α

are outside the PCA

model reconstruction and contain non-common fea-

tures(Conlin et al., 2000).

In these cases, the contribution plot is estimated

to determine which variables are affecting the Q-

statistics (Miller et al., 2003; Dunia et al., 1996), and

YEAST METABOLIC STATE IDENTIFICATION BY FIBER OPTICS SPECTROSCOPY

173

diagnostic why features are not captured. The recon-

structed sample

b

x

i

, the variable contribution for the

reconstruction error is estimated by the square error

of each variable E

2

ij

(Miller et al., 2003; Dunia et al.,

1996).

Another well known statistic is the Hotelling T

2

.

In SVD, this is used as a measure of the distance to

the center of data, being computed by:

T

2

i

= x

T

i

VA

−1

V

T

i

x (5)

and A =

1

n−1

TT

T

, where T = (US)

rel

and n is the

number of wavelength in the spectra (X columns).

The upper conﬁdence interval for the Hoteling T

2

is

estimated by: T

2

α

=

l(n−1)

n−1

F

l,n−1;α

; where l is the num-

ber of relevant singular values and F

l,n−1;α

the F dis-

tribution value with l and n− 1 degrees of freedom at

α = 0.05 level of signiﬁcance. Samples with a T

2

-

statistic above T

2

α

are considered to have signiﬁcantly

different features (Qin, 2003).

3 RESULTS AND DISCUSSION

3.1 Spectral Absorbance

Figure 3 presents LWUV-VIS and VIS-SWNIR yeast

spectrum for different growth medium conditions. It

is possible to assess in the absorbance spectrum (Fig-

ure 3 (a) and (c)) that colonies are directly distin-

guishable by the intensity and spectral shape. The

ﬁrst derivate spectrum (Figure 3 (b) and (d)) was cal-

culated to eliminate background and baseline effects.

In this signal colonies can be distinguishable in the

wavelength interval of 350-500 nm and 600-900 nm

in the LWUV-VIS and VIS-SWNIR, respectively.

Spectrum may contain information of the growth

media. This was minimized by maximizing the con-

trast between the growth media and colonies. It is

possible to observe that spectrum signatures for each

growth medium conditions are different and well dis-

tinguished, and therefore it is reasonable to assume

that most of the information obtained in the spectra is

coherent with the colony metabolomic state.

3.2 Singular Value Decomposition

Analysis

Figure 4 (a) present relevant scores plot in the 3 PC’s

for LWUV-VIS ﬁrst derivative of absorbance, total-

izing 90.1 % of spectral variance with discriminant

power (PC1 (53.7 %), PC2 (32.6 %), PC3 (3.8 %)).

PC1 (53.7 %) separates the samples by spectral

intensity into four groups: i) growth media control

(YPD and YP) and colony growth in YP without alco-

hol; ii) colonies growth in: YP Phenylethanol and YP

Isomyl; iii) colonies growth in: YPD tert-Amyl, YPD

Butanol, YP Isopropanol, YP Ethanol, YP Butanol,

YPD Isopropanol, YP Propanol, YPD Ethanol, YPD

w/o alcohol and YPD Phenylethanol; and iv) colonies

growth in YPD Propanol and YPD tert-Amyl.

PC2 (32.6 %) distinguishes the growth medium

control (without colonies) from the other samples,

with the exception of the colony growth in YP me-

dia without alcohol. This similarity may be due to the

small sized colony, which causes the passage of light

through the media.

PC3 (3.8 %) captures a small variance in the spec-

trum. Neverthenless, it is also signiﬁcant for discrimi-

nation of the colonies of YPD Propanol and YPD tert-

Amyl.

Figure 4 (b) shows the ﬁrst derivative spectra PCA

analysis for the VIS-SWNIR light. It is also de-

composed into 3 PC’S, (80.3 % of total variance) in

the VIS-SWNIR region. PC1 (63.8.0%) also segre-

gates colonies by spectral intensity, and YPD con-

trol is completely distinguished from other samples

. PC2 (14.0 %) segregates colonies into different

groups, where YPD Phenylethanol, YPD w/o alcohol

and YPD Ethanol, are completely distinguish. PC3

(2.5%) segregates differences between colonies in the

same groups.

Figure 5 (a) and (b) presents the diagnostic plot

(Q-T

2

h

plot) for the two light sources. In these two ﬁg-

ures, YPD control and YP control samples are above

the Q or T

2

h

limits, which means that the growth me-

dias are dissociated from the rest of the samples, be-

ing possible to afﬁrm that their spectral features are

signiﬁcantly different from the colonies, and statisti-

cally guaranteeing that the information contained in

the collected spectra is mostly independent of the

growth media, measuring the metabolomic state of

each colony.

The analysis of the diagnostic plots also allows to

conclude that colonies in: YP w/o alcohol (Figure 5

(a)) and YPD phenylethanol (Figure 5 (b)) are above

the Q limit, which means that the features captured in

the 3 ﬁrst components for these medias, are not suf-

ﬁcient to reconstruct the original spectral data, due to

large reconstruction errors (Qin, 2003). This allows

us to conclude that these colonies are in completely

different metabolic state than the rest of the studied

growth conditions.

In Figure 5 (b), some spectra replicates of colonies

in YPD Phenylethanol media are above the T

2

h

limit,

which means that these spectra colony is statisti-

BIOSIGNALS 2009 - International Conference on Bio-inspired Systems and Signal Processing

174

Table 1: Integration time, morphology and classiﬁcation probabilities results.

Classiﬁcation probabilities

Integration Time (ms) ABS + HCA (%) Derv + HCA (%)

Growth Media UV-VIS VIS-NIR Morphology UV-VIS VIS-NIR UV-VIS VIS-NIR

YPD without alcohol 524 439 w/o hyphae 100 100 100 100

YP without alcohol 80 94 w/o hyphae 100 100 100 100

YPD ethanol 639 497 w/o hyphae 100 100 100 100

YP ethanol 1906 850 w/o hyphae 100 100 100 100

YPD butanol 455 373 w/o hyphae 100 100 100 100

YP butanol 529 351 hyphae 100 100 100 100

YPD propanol 1546 625 w/o hyphae 100 100 100 100

YP propanol 856 740 w/o hyphae 100 100 100 100

YPD isopropanol 490 378 hyphae 100 100 100 100

YP isopropanol 632 288 w/o hyphae 100 100 100 100

YP isoamyl 328 92 hyphae 100 100 100 100

YPD phenylethanol 620 240 w/o hyphae 100 100 100 100

YP phenylethanol 177 81 w/o hyphae 100 100 100 100

YPD terta-Amyl 1365 571 hyphae 100 100 100 100

YP terta-Amyl 400 379 hyphae 100 100 100 100

0 5 10 15

0.001 0.002 0.003 0.004 0.005 0.006 0.007

(a)

T2−statistic

Q−statistic

YP Butanol

YP Control

YPD Butanol

YPD Control

YPD Etanol

YPD Phenylethanol

YPD Isopropanol

YPD Propanol

YPD /alcohol

YPD tert−Amyl

YP Etanol

YP Phenylethanol

YP Isoamyl

YP Isopropanol

YP Propanol

YP /alcohol

YP tert−Amyl

0 10 20 30 40 50

4e−05 6e−05 8e−05 1e−04

(b)

T2−statistic

Q−statistic

YP Butanol

YP Control

YPD Butanol

YPD Control

YPD Ethanol

YPD Phenylethanol

YPD IsopropanolL

YPD PropanolL

YPD /alcohol

YPD tert−Amyl

YP Ethanol

YP Phenylethanol

YP Isoamyl

YP IsopropanolL

YP PropanolL

YP /alcohol

YP tert−Amyl

282 305 328 351 374 396 417 438 461 484 505 527 550 573

(c)

Wavelength(nm)

SPE Contribution

0e+00 1e−05 2e−05 3e−05 4e−05 5e−05 6e−05 7e−05

282 305 328 351 374 396 417 438 461 484 505 527 550 573

(d)

Wavelength(nm)

SPE Contribution

0.0e+00 5.0e−06 1.0e−05 1.5e−05 2.0e−05 2.5e−05 3.0e−05

440 469 497 527 556 584 608 637 666 695 722 751 779 807

(e)

Wavelength(nm)

SPE Contribution

0.0e+00 5.0e−07 1.0e−06 1.5e−06

Figure 5: Diagnostic and Contribution plots: (a) LWUV-VIS Diagnostic plot; (b) VIS-SWNIR Diagnostic plot; (c) YP without

alcohol LWUV-VIS (d) YP ethanol LWUV-VIS and (e) YPD phenylethanol VIS-SWNIR.

YEAST METABOLIC STATE IDENTIFICATION BY FIBER OPTICS SPECTROSCOPY

175

cally different from the average spectral features com-

pressed by the 3 PC’s. Therefore, this colony can be

directly classiﬁed from the global SVD model (see

Figure 4 (b)).

The diagnostic plots allowed to understand that

the 3 PC’s model is capable of discriminating the ma-

jor spectral differences between colonies, but it can-

not compress all the spectral features in the relevant

PC’S. Such may leads to errors in distinction and

spectral features interpretation.

Contribution plots allow to interpret why colonies

of YPD w/o alcohol, YP Ethanol and YPD

Phenylethanol (Figure 5 (c), (d) and (e), respectively),

are distinguishable from the rest of the samples.

YPD w/o has higher reconstruction errors in the

region of 300-350 nm, which are linked to chro-

mophoric groups of C=C and -N=N-, respectively.

YP ethanol colonies has high contribution in the in-

tervals of 280-300 nm, which is dominated by C=C

chromophoric group. The colonies which grown in

the YPD phenylethanolhas high contribution errors in

the intervals of 440-520 nm and 750-800 nm, which

are linked to C=S chromophoric group and OH over-

tones, respectively.

After SIMCA analysis, HCA was performed tak-

ing into consideration the euclidean distance between

the center of the scores of each yeast spectra. HCA is

presented in Figure 6 (a), (b) for LWUV-VIS and (c),

(d) for VIS-SWNIR wavelengths, respectively.

Hierarchical clustering analysis differs in both

light sources, and then the relative positions of

colonies are different. In both hierarchical trees,

yeasts structures that were grown under different con-

ditions are well discriminated from each one. It is

possible to observe a good discrimination between

control mediums YP and YPD in LWUV-VIS and

VIS-SWNIR trees, but YP ethanol, YP without alco-

hol and YP isoamyl are more similar to the control

mediums. This phenomena can be explained because

of the low density of cells and the translucency of the

colony that allows light to cross the colony and then

incorporate signiﬁcant amount of growth media spec-

tral information, leading to lower contrast between

growth media and microorganisms in this cases. Such

is especially problematic, if the colony is small sized

or when the probe is not properly placed.

Comparing LWUV-VIS and VIS-SWNIR trees

with Table 1, we can relate colonies aggroupment

with its morphologies. In LWUV-VIS tree it is pos-

sible to distinguish some colonies with the same

morphology, such as YPD phenyl-ethanol, YP 1-

propanol, YP ethanol and YPD without alcohol

(which do not form ﬁlament forms) from the other

groups. S. cerevisiae that has grown in YP tert-Amyl

YP Phenylethanol

YP Control

YPD Control

YP /Alcohol

YP Etanol

YPD Propanol

YPD tert-Amyl

YP Propanol

YP Isopropanol

YPD Phenylethanol

YPD Etanol

YP Isoamyl

YPD Butanol

YP tert-Amyl

YP Butanol

YPD /Alcohol

YPD Isopropanol

(a)

YP Phenylethanol

YP Control

YPD Control

YP /Alcohol

YP Etanol

YPD Propanol

YPD tert-Amyl

YP Propanol

YP Isopropanol

YPD Phenylethanol

YPD Etanol

YP Isoamyl

YPD Butanol

YP tert-Amyl

YP Butanol

YPD /Alcohol

YPD Isopropanol

(b)

YP Phenylethanol

YP Control

YPD Control

YP /Alcohol

YP Etanol

YPD Propanol

YPD tert-Amyl

YP Propanol

YP Isopropanol

YPD Phenylethanol

YPD Etanol

YP Isoamyl

YPD Butanol

YP tert-Amyl

YP Butanol

YPD /Alcohol

YPD Isopropanol

(c)

YP Phenylethanol

YP Control

YPD Control

YP /Alcohol

YP Etanol

YPD Propanol

YPD tert-Amyl

YP Propanol

YP Isopropanol

YPD Phenylethanol

YPD Etanol

YP Isoamyl

YPD Butanol

YP tert-Amyl

YP Butanol

YPD /Alcohol

YPD Isopropanol

(d)

Figure 6: Hierarchical cluster Analysis: (a) Absorbance

- LWUV-VIS; (b) First derivative - LWUV-VIS; (c) Ab-

sorbance - VIS-SWNIR and (d) First derivative VIS-

SWNIR.

medium is well separated from the other. groups

in the LWUV-VIS, but is integrated in YP-butanol

group in the VIS-SWNIR wavelengths. Some groups

with yeast-form can be distinguish in VIS-SWNIR

wavelengths such as YPD ethanol, YPD 1-propanol

and YP ethanol and YP isopropanol and YPD bu-

tanol. Furthermore, YPD without alcohol and YPD

isopropanol are completely separated from all groups.

Table 1 presents a 100 % classiﬁcation probabil-

ities for all colonies. Such means that all colonies

are completely differentiated from from each others.

However, this classiﬁcation does not depends only

colony morphology, but it mostly depends on the

chemical composition and metabolism of colonies.

Comparing LWUV-VIS and VIS-SWNIR informa-

tion (Figure Spectra), it is possible to conclude that

there are considerable differences that indiciate the

presence of different chemical compounds that re-

spond differently to the UV excitation. It would be

expectable that absorvance would show the same pat-

tern in both light sources, but results show that the

BIOSIGNALS 2009 - International Conference on Bio-inspired Systems and Signal Processing

176

yeast spectra has completely different features when

responding to UV excitation, especially in the region

of 350-450 nm. Such high resolution and the dif-

ferences between the two light sources spectra, al-

lows us to conclude that UV-VIS-SWNIR is capa-

ble of high performance discrimination of the yeast

metabolic states.

However, as spectroscopy is a non-target ap-

proach, it is not possible identify directly the chem-

ical compounds and the transcripted genes that dif-

ferentiate the colonies. In order to understand deeper

the potential of UV-VIS-SWINR spectroscopy, it is

necessary to correlate spectroscopy data against high-

precision analytical techniques, such as mass spec-

troscopy (e.g. LC-MS/MS, GC-MS or Maldi-TOF),

NMR and transcriptomics (e.g. DNA/RNA Microar-

rays).

4 CONCLUSIONS

This work has shown that after appropriate prepro-

cessing and signal classiﬁcation, UV-VIS-SWNIR

spectroscopy is a high resolution technique capable

of attaining interesting possibilities in non-destructive

metabolomics in the near future. Further insights will

be gained when spectral information is deeper under-

stood, not only by correlating with other high resolu-

tion analytical chemistry and molecular biology tech-

niques, but also in understanding of the spectra shape

of the differents microorganisms.

ACKNOWLEDGEMENTS

Part of this work was supported by the project Open-

MicroBio (PTDC/BIO/69310/2006) - ’A Framework

for Computational Simulation of Cellular Commu-

nities during BioProcess Engineering’; and partially

supported by CBMA, IBB/CEB and ISR/IST pluri-

anual funds through the POS-Conhecimento Program

that includes FEDER funds.

REFERENCES

Avantes, I. (2007). Users manual.

Baig, S. and Rehman, F. (2006). Signal modeling using sin-

gular value decomposition. In Advances in Computer,

Information, and Systems Sciences, and Engineering.

Springer Netherlands.

Bhatta, H., Goldys, E., and Learmonth, R. (2005). Rapid

identiﬁcation of microorganisms by intrinsicﬂuores-

cence. In Imaging, Manipulation, and Analysis of

Biomolecules and Cells: Fundamentals and Applica-

tions III,. SPIE.

Burns, D. and Ciurczak, E. (2001). Handbook of near in-

frared analysisI. Marcel Dekker, Inc, New York, 2nd

edition edition.

Choi, S., Lee, C., Lee, J., Park, J., and Lee, I. (2005).

Fault detection and identiﬁcation of non-linear pro-

cesses based on kernel pca. Chemometrics and intel-

ligent laboratory systems, 75:55–67.

Conlin, A., Martin, E., and Morris, A. (2000). Conﬁdence

limits for contribution plots. Journal of Chemomet-

rics, 14:725–736.

Coyle, J. (1989). Introduction to Organic PhotoChemistry.

John Wiley & Sons, London.

Denney, R. and Sinclair, R. (1987). Visible and ultraviolet

spectroscopy. John Wiley & Sons, London.

Devices, A. S. (2005). Near-ir absorption bands.

Dickinson, J. (1994). Irreversible formation of pseudohy-

phae by haploid Saccharomyces cerevisiae. FEMS Mi-

crobiol. Lett., 119:99–104.

Dickinson, J. (2008). Filament formation in Saccharomyces

cerevisiae - a review. Folia Microbiol., 53(1):3–14.

Doytchinova, F. and Flower, D. (2006). Modeling the

peptide t-cell receptor interaction by the compara-

tive molecular similarity indices analysis-soft inde-

pendent modeling of class analogy technique. Journal

of Medicinal Chemistry, 49(7):2193–2199.

Dunia, R., Qin, S., Edgar, T., and T.J., M. (1996). denti-

ﬁcation of faulty sensors using principal component

analysis. American Institute of Chemical Engineers,

42:2797–2812.

Dziuba, B., Babuchowski, A., Naleczb, D., and Niklewicz,

M. (2007). Identiﬁcation of lactic acid bacteria using

ftir spectroscopy and cluster analysis. In International

Dairy Journal 17: 183189. Elsevier.

Gallager, N. B., Blake, T., and Gassman, P. (2005). Ap-

plication of extended inverse scattering correction to

mid-infrared reﬂectance of soil. Journal of Chemo-

metrics, 19:271–281.

Gerard, J., Berdell, R., and Christine, L. (2006). Microbi-

ology: An Introduction. Benjamin Cummingsc, New

York, 2th edition edition.

Gimeno, C. J., Ljungdahl, P., Styles, C., and Fink, G.

(1992). Unipolar cell divisions in yeast Saccha-

romyces cerevisiae lead to ﬁlamentous growth: reg-

ulation by starvation and ras. Cell, 68:1077–1090.

Jackson, J. and Mudholkar, G. (1979). Control proce-

dures for residuals associated with principal compo-

nent analysis. Technometrics, 21:341–349.

Jolliffe, I. (1986). Principal Component Analysis. Springer,

New York, USA.

Klessinger, M. and Michl, J. (1995). Exited states and pho-

tochemistry of organic molecules. VCH Publishers,

New York.

Kron, S. and Gow, N. (1995). Budding yeast mor-

phogenesis: gignaling, cytoskeleton and cell cycle.

Curr.Opin.Cell Biol., 7:845–855.

YEAST METABOLIC STATE IDENTIFICATION BY FIBER OPTICS SPECTROSCOPY

177

Krzanowski, W. J. (1998). Principles of Multivariate Data

Analysis. Oxford University Press, Oxford, UK.

Levine, I. (1975). Molecular Spectroscopy. John Wiley &

Sons, New York.

Manly, B. F. (1998). Randomization, Bootstrap and Monte

Carlo Methods in Biology. Chapman and Hall, Lon-

don, UK, 2nd edition.

Martens, H., Nielsen, J. P., and Engelsen, S. B. (2003).

Light scattering and light absorbance separated by ex-

tended multiplicative signal correction. application to

near-infrared transmission analysis of powder mix-

tures. In Analytical Chemistry 75(9): 394-404. Amer-

ican Chemical Society.

Martens, H. and Stark, E. (1991). Extended multiplicative

signal correction and spectral interference subtraction:

new preprocessing methods for near infrared spec-

troscopy. In Journal of Pharmaceutical and Biomedi-

cal Analysis 9: 625-635. American Chemical Society.

Micropack (2008). DH2000 BAL: Installation and operat-

ing manual. Ocean Optics, Ostﬁlden, Germany.

Miller, P., Swanson, R., and Heckler, C. (2003). Contribu-

tion plots: the missing link in multivariate quality con-

trol. International Journal of Production Economics,

9:775–792.

Optics, O. (2006). HR4000 - High resolution ﬁber op-

tic spectrometers: instalation and operation manual.

Ocean Optics, Dundelin, FL USA.

Palkova, Z. and Vachova, L. Ammonia signaling in yeast

colony formation.

Perkauparus, H., Grinter, H., and Therfall, T. (1994). Uv-Vis

spectroscopy and its applications. Springer-Verlag,

New York.

Qin, S. (2003). Statistical process monitoring: basics and

beyond. Journal of Chemometrics, 17:480–502.

Rosah, P., Hard, M., Schmitt, M., Peschke, K. D.,

Ronneberger, O., Burkhart, H., Mutzkus, H. W.,

Laukers, M., Hofer, S., Thiele, H., and Popp, J.

(2005). Chemotaxonomic identiﬁcation of sim-

ple bacteria by micro-raman spectroscopy: applica-

tion to clean-room-relevant biological contamination.

A&EnvMicro, 71(3):1626–1637.

Rua, D., Tobe, B., and Kron, S. (2001). Cell cycle control

of yeast ﬁlamentous growth. Curr.Opin.Microbiol.,

4:720–727.

Savitzky, A. and Golay, M. (1964). Smoothing and differen-

tiation of data by simpliﬁed least squares procedures.

Analytical Chemistry, 36:1627–1639.

Sigma-Aldrich Quimica, S. (2008). Sigma, Life Science:

Produtos para Investigao em Ciłncias da Vida, 2008-

2009. Sigma-Aldrich, Portugal, 1nd edition.

Silva, J., Martins, R. C., Vicente, A., and Teixeira, J.

(2008). Feasibility of yeast and bacteria identiﬁca-

tion using lwuv-vis-swnir diffusive reﬂectance spec-

troscopy. volume 1.

Stuart, B. (2004). Infrared Spectroscopy: Fundamentals

and Applications. John Wiley & Sons, Ltd, London,

1nd edition.

Treskatis, S., Orgeldinger, V., Wolf, H., and Gilles, E. D.

(1997). Morphological characterization of ﬁlamen-

tous microorganisms in submerged cultures by on-line

digital image analysis and pattern recognition. volume

53(2), pages 191–201.

BIOSIGNALS 2009 - International Conference on Bio-inspired Systems and Signal Processing

178