RAPID COMMUNICATIONS IN MASS SPECTROMETRY

Rapid Commun. Mass Spectrom. 2008; 22: 2555–2560

) DOI: 10.1002/rcm.3649

Published online in Wiley InterScience (www.interscience.wiley.com

Discrimination of Penicillium isolates by matrix-assisted

laser desorption/ionization time-of-flight mass

spectrometry fingerprintingy

Justin M. Hettick*, Brett J. Green, Amanda D. Buskirk, Michael L. Kashon,

James E. Slaven, Erika Janotka, Francoise M. Blachere, Detlef Schmechel

and Donald H. BeezholdCenters for Disease Control and Prevention, National Institute for Occupational Safety and Health, Health Effects Laboratory Division, 1095

Willowdale Road, Morgantown, WV 26505, USA

Received 7 March 2008; Revised 21 April 2008; Accepted 17 June 2008

*CorrespoWillowdaE-mail: jhyThis artidomain iContract/grant num

Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) was

used to generate highly reproducible mass spectral ‘fingerprints’ for twelve Penicillium species. Prior

to MALDI-TOF MS analysis, eight replicate cultures of each Penicillium species were subjected to

three one-minute bead-beating cycles in an acetonitrile/trifluoroacetic acid solvent. The mass spectra

contained abundant peaks in the range ofm/z 5000–20 000, and allowed unambiguous discrimination

between species. In addition, a biomarker common to all Penicillium mass spectra was observed at

m/z 13 900. Discriminant analysis using theMALDI-TOFMS data yielded classification error rates of

0% (i.e. 100% correct identification), indicating that MALDI-TOFMS data may be a useful diagnostic

tool for the objective identification of Penicillium species of environmental and clinical importance.

Published in 2008 by John Wiley & Sons, Ltd.

The hyphomycete genus Penicillium is a large, ubiquitous

group that has a wide geographical distribution and is

economically important for various industrial, biodeteriora-

tion, and health applications.1 However, the genus has been

difficult to classify using traditional taxonomic methods due

to lack of information on the teleomorphic states and the

similarities of morphological criteria between anamorphic

states, in particular, the terverticillate species within the

subgenus Penicillium.2,3 Recent advances in molecular and

biochemical techniques have enabled systematic studies

based on the analysis of metabolites as well as ribosomal,

mitochondrial, b-tubulin, C01 and calmodulin gene

sequences.2,4–14 These studies have provided detailed insight

into the taxonomy of Penicillium and its teleomorphs.15

The detection of Penicillium species using viable and non-

viablemonitoring techniques is routine in the fields of indoor

air hygiene, agriculture, and food microbiology. Non-viable

or direct microscopic techniques are often subjective and

unicellular conidia are often grouped in a combined

ndence to: J. M. Hettick, NIOSH/HELD/ACIB, 1095le Rd., MS L-2040, Morgantown, WV 26505, USA.ettick@cdc.govcle is a U.S. Government work and is in the publicn the U.S.A.grant sponsor: Inter-Agency Agreement; contract/ber: NIEHS Y1-ES0001-06.

Aspergillus/Penicillium classification. Viable or culture-based

techniques can enable species differentiation; however, this

is subjectively based on the taxonomic judgment of a trained

mycologist. Several studies have shown that morphological

similarities often exist between various Penicillium species,

whereas variations in morphological and physiological

characteristics have also been identified between individual

strains.2,3,16 As a consequence, the difficulties associatedwith

identifying Penicillium species have resulted in numerous

misidentifications.17,18

Microbiological fingerprinting based on matrix-assisted

laser desorption/ionization (MALDI)19,20 time-of-flight mass

spectrometry (TOF MS)21 is a powerful tool for the analysis

and discrimination of microorganisms such as bacteria.22,23

Although MALDI-TOF MS fingerprinting techniques have

been developed for a number of bacterial species such as

E. coli 24,25 and M. tuberculosis,26,27 the utility of MALDI-TOF

MS to discriminate between various fungi including

Penicillium has only recently been explored.28–32 In a 2005

study, Chen and Chen reported the ability to discriminate

between the fungal spores of six Penicillium species on

the basis of their MALDI-TOF mass spectra.31 In this study,

we explore the utility of MALDI-TOF MS using bead

disruption during sample preparation, combined with

biostatistical tools to identify and discriminate between

twelve Penicillium species.

Published in 2008 by John Wiley & Sons, Ltd.

2556 J. M. Hettick et al.

EXPERIMENTAL

ReagentsAngiotensin II (human), insulin oxidized B chain (bovine),

cytochrome C (equine), albumin (bovine serum), a-cyano-4-

hydroxycinnamic acid (CHCA), and trifluoroacetic acid

(TFA) were purchased from Sigma-Aldrich (St. Louis, MO,

USA). Acetonitrile (HPLC grade) was purchased from Fisher

Scientific (Fairlawn, NJ, USA). Malt extract agar (MEA) was

purchased from Difco (Sparks, MD, USA). Distilled deio-

nized water was produced by a Synthesis A-10 (Millipore,

Billerica, MA, USA).

Fungal cultureTwelve fungal species belonging to the genus Penicillium

(Table 1) were sub-cultured from NIOSH (National Institute

for Occupational Safety andHealth, Morgantown,WV, USA)

or NRRL (National Center for Agricultural Utilization

Research, Peoria, IL, USA) stock sources and grown for 20

days on MEA. Penicillium variabile was provided by Dr. John

I. Pitt, formerly of CSIRO, North Ryde, Australia. In addition,

Paecilomyces variotii Bainier (ATCC #66705) and Aspergillus

versicolor (Vuillemin) Tiraboschi (ATCC #44408) were sub-

cultured from stock sources and used to compare the process

of cell disruption via bead beating (100mL/mL 0.1mm dia.

zirconium beads; Biospec, Bartlesville, OK, USA) against

simple chemical extraction. To ensure reproducibility, eight

independent replicates of each Penicillium species were

cultured, resulting in a total of 96 individual fungal cultures.

Following 20 days of growth, conidia and hyphae were

scraped from the surface of an agar plate and collected into

1mL 50:50 acetonitrile/4% TFA containing zirconium beads.

The samples were then subjected to three 1-min bead-beating

cycles. The resulting extract was centrifuged at 14 500 rpm for

10min to remove cellular debris and an aliquot of the

supernatant was transferred for MALDI-TOF MS analysis.

Mass spectrometrySamples for MALDI-TOF MS were prepared by mixing

sample supernatant 1:1 with 10mg/mL CHCA. A 1mL

aliquot of the resulting solution was deposited on a gold

sample stage (Bio-Rad, Hercules, CA, USA) and allowed to

Table 1. Fungi of the genus Penicillium utilized in this study

SpeciesNRRL

Catalog No.NIOSH

Catalog No.

A Penicillium aurantiogriseum 971 197708B Penicillium brevicompactum � 324604C Penicillium citrinum 1843 176211D Penicillium chrysogenum 1951 176206E Penicillium expansum 973 176202F Penicillium fellutanum 746 197709G Penicillium jensenii � 263822H Penicillium melinii � 324601I Penicillium purpurogenum 1062 197706J Penicillium roqueforti 844 176209K Penicillium simplicissimum 902 176201L Penicillium variabile �� 271114

�NIOSH environmental isolate.��Received from Dr. John I. Pitt.

Published in 2008 by John Wiley & Sons, Ltd.

air dry. Each sample was analyzed in duplicate, for a total of

sixteen composite MALDI-TOF mass spectra per species.

MALDI-TOF mass spectra were acquired using a Ciphergen

PBS-IIc linear time-of-flight mass spectrometer (Bio-Rad).

Spectra were acquired over the m/z range 0–100 000, with

the delayed extraction parameters set to optimally focus the

m/z 10 000–20 000 range. Composite mass spectra are the

average of 100 laser shots taken from 20 distinct positions

across the sample deposit. These positions were held

constant for all samples utilized in this study. The 337-

nm N2 laser intensity was maintained just above the

threshold for ion production (laser intensity set to 140–170

in control software). Mass spectra were externally calibrated

using a set of peptide and protein calibrants that covered the

range of m/z 1000–66 000. Duplicate spectra for each culture

replicate were averaged prior to statistical analysis, resulting

in eight mass spectra per species.

Data analysisInitial data analysis was performed using the Biomarker

Wizard (Bio-Rad) software suite. Spectra were baseline

corrected and normalized to total ion current. ‘Clusters’ of

peaks common to a given isolate were generated by selecting

all peaks with signal-to-noise (S/N) ratio greater than 5 that

occurred in each spectrum from that isolate. Linear

discriminant functions were analyzed using SAS/STAT

software, version 9.1 of the SAS system for Windows (SAS

Institute, Cary, NC, USA). The intensity values were first

tested for their distribution, and were found to be log-

normal, so a natural log transformation was utilized on the

intensity values. A stepwise variable selection method using

the ‘PROC STEPDISC’ procedure, which selects a subset of

the variables of interest using a stepwise discriminant

analysis, keeping themost significant variables from iterative

F-tests, was then performed to select a subset of variables that

could serve as predictor variables for class membership.

Using this new subset of significant peaks, ‘PROC DISCRIM’

was utilized to determine the classification error rate

between the various Penicillium species. This was done by

calculating a discriminant function that allows each data

point to be compared with all others for cross-validation and

resubstitution classification. This process was done itera-

tively for each individual data point. This procedure was

followed by the ‘PROC CANDISC’ procedure, which

reduces the number of dimensions to find linear combi-

nations of the variable set that also summarizes between-

class variation, to perform a canonical discriminant analysis

on that subset of variables. This creates new variables by

taking linear combinations of the original variables and aids

in determining the true underlying dimension of the data

space. The canonical functions generated allow the calcu-

lation of canonical scores, which can be used to discriminate

among the various strains. Using the same subsets of

variables, cluster analysis was then performed using the

‘PROC CLUSTER’ procedure, which uses distances between

data points to form hierarchical clusters, and a dendrogram

was generated using the ‘PROC TREE’ procedure, a

procedure that uses the data set from ‘PROC CLUSTER’ to

produce the dendrogram.

Rapid Commun. Mass Spectrom. 2008; 22: 2555–2560

DOI: 10.1002/rcm

Figure 1. Comparison of MALDI-TOF fingerprint mass spec-

tra sample preparation methodologies. (A) Paecilomyces

variotii chemical extraction; (B) Aspergillus versicolor chemi-

cal extraction; (C) Paecilomyces variotii bead beating; and

(D) Aspergillus versicolor bead beating

Figure 2. MALDI-TOF MS fingerprint spectra of four inde-

pendent cultures of P. jensenii.

Penicillium isolates by MALDI-TOF MS fingerprinting 2557

RESULTS AND DISCUSSION

Initial studies of undisrupted, whole cell extracts of P. variotii

and A. versicolor using MALDI-TOF MS demonstrated mass

spectral profiles with substantially poorer quality than those

derived from bacteria (Figs. 1(A) and 1(B)). This observation

is in good agreement with the prior studies.26,27,31,32 To

overcome this methodological limitation, cellular disruption

of fungal cells using bead beating was explored as a method

of increasing the number and relative abundance of peaks

observed in the fingerprint MALDI-TOF mass spectra.

Following cellular disruption, the MALDI-TOF fingerprint

mass spectra of both fungal isolates exhibited significantly

more peaks with greater relative abundance than the

untreated samples (Figs. 1(C) and 1(D)). The MALDI-TOF

MS fingerprint mass spectra thus obtained are rich in

information content, in stark contrast to the spectra obtained

by simple chemical extraction or whole cell MALDI-TOFMS.

The spectra in Figs. 1(A) and 1(B) show no peaks in the mass

range of interest (m/z 5000–20 000). Although not all fungi

show such poor spectra upon simple chemical extraction,

this figure demonstrates the possible ‘worst case’ scenario

and the dramatic improvement observed when utilizing the

bead-beating extraction method. As a result, pretreatment

with bead beating during extraction was incorporated for all

Penicillium species analyzed.

Previous studies with bacteria have demonstrated that

whole-cell MALDI-TOF MS data is highly dependent on the

culture conditions and sample preparation procedures.33,34

In order to ensure the reproducibility of the method, each

species was cultured eight different times, followed by

Published in 2008 by John Wiley & Sons, Ltd.

identical bead disruption and MALDI-TOF sample prep-

aration procedures. To demonstrate the reproducibility of

themethod, theMALDI-TOFmass spectra of four of the eight

independent cultures of P. jensenii are presented in Fig. 2.

Because the MALDI-TOFMS fingerprint spectra of Penicillia

are highly reproducible, this method could be a valuable tool

for the chemotaxonomy of fungi. However, it should be

noted that variations in morphological and biochemical

characteristics within Penicillium species,16,35 as well as the

lack of consensus on Penicillium taxonomy, may currently

limit identification capabilities. Future studies should focus

on the characterization of mass spectral fingerprints derived

from ex-type strains to enable the correct identification of

unknown Penicillium species.

Representative MALDI-TOF MS fingerprint mass spectra

from twelve Penicillium species are presented in Figs. 3–5.

The mass spectra are characterized by abundant peaks

within the range of m/z 5000–20 000, including a common

peak at m/z 13 900. This biomarker is observed in the

fingerprint mass spectrum of each species of Penicillium

examined here. Although a more exhaustive examination of

Penicillia and other fungal generawill be necessary, this peak

may serve as a potential Penicillium genus specific biomarker.

Chen and coworkers31 have previously published MALDI-

TOF mass spectra from spores of P. citrinum Thom,

P. chrysogenum Thom, and P. expansum. Their fingerprint

mass spectra differ significantly from those presented here

(Figs. 3(C), 3(D) and 4(E), respectively). Specifically, the

fingerprint mass spectra of Chen are dominated by

biomarker peaks at approximately m/z 3000, with ions of

significantly reduced intensity in the range ofm/z 5000–8000.

In contrast, our spectra are characterized by a larger number

of peaks in the range of m/z 10 000–20 000. The differences in

the appearance of the fingerprint mass spectra between the

Rapid Commun. Mass Spectrom. 2008; 22: 2555–2560

DOI: 10.1002/rcm

Figure 5. MALDI-TOF MS fingerprint spectra of

(I) P. purpurogenum, (J) P. roqueforti, (K)

P. simplicissimum, and (L) P. variable.

Figure 3. MALDI-TOF MS fingerprint spectra of

(A) P. aurantiogriseum, (B) P. brevicompactum,

(C) P. citrinum, and (D) P. chrysogenum.

2558 J. M. Hettick et al.

two laboratories are attributable to several factors, including

instrumentation (reflectron TOF vs. linear TOF), culture

conditions (potato dextrose agar vs. MEA), sample prep-

aration (whole cell MALDI vs. bead beating), and MALDI

Figure 4. MALDI-TOF MS fingerprint spectra of

(E) P. expansum, (F) P. fellutanum, (G) P. jensenii, and

(H) P. melinii.

Published in 2008 by John Wiley & Sons, Ltd.

matrix (sinapinic acid vs. CHCA). These differences

underscore the importance of standardized methodology

for microbiological analyses utilizing MALDI-TOF MS,

whether with fungi or bacteria. Moreover, these studies

demonstrate that a standard reference library of biological

mass spectra will be difficult to implement, given the wide

variety of commercially available mass spectrometers in use

in laboratories worldwide.

The database of 96 total fingerprint mass spectra was

further analyzed using canonical discriminant analysis. This

analysis produces canonical variables that are combinations

of variables in the dataset that can be used to discriminate

between species. Figure 6 presents a plot of the first two

canonical variables, accounting for most of the variation in

the dataset. Inclusion of more canonical variables resulted in

multi-dimensional plots that are difficult to illustrate;

however, the two-dimensional plot presented in Fig. 6

demonstrates the power of this statistical tool for fungal

discrimination. All eight spectra for each fungal species are

tightly clustered, indicating a high degree of homogeneity

within the mass spectra from one species and heterogeneity

between spectra from other species. Interestingly, outlier

clusters such as P. purpurogenum Stoll and P. variabile Sopp

(Figs. 6(I) and 6(L)) belong to the Penicillium subgenus

Biverticillium.3 The remaining Penicillium species belong to

the subgenus Penicillium, which in general are clustered

together. These data demonstrate the potential utility of

chemotaxonomic interpretations of MALDI-TOF MS data,

whereas previous studies have demonstrated that metabolite

profiles alone only support broader taxonomic concepts.35

The error rates for species prediction based on theMALDI-

TOF MS dataset were calculated using both resubstitution

Rapid Commun. Mass Spectrom. 2008; 22: 2555–2560

DOI: 10.1002/rcm

Figure 6. Canonical discriminant analysis of MALDI-TOF MS data for 12 species of Penicillium as listed

in Table 1.

Penicillium isolates by MALDI-TOF MS fingerprinting 2559

and cross-validation methods. In resubstitution, the dis-

criminant function is fitted to the dataset and then applied to

each observation. In contrast, cross-validation deletes the

observation, fits the discriminant function to the remaining

dataset, and then applies the function to the deleted

observation. Resubstitution tends to underestimate classifi-

Figure 7. Cluster analysis dendrogram of MALDI-TOF

Table 1.

Published in 2008 by John Wiley & Sons, Ltd.

cation error, whereas cross-validation is unbiased and is

preferred for large datasets.36 For our dataset, resubstitution

and cross-validation correctly classified each fungal finger-

print spectrum with a 0% error rate. Similarly, after

application of the STEPDISC procedure in SAS/STAT to

select a subset of variables with which to perform the

MS data for 12 species of Penicillium as listed in

Rapid Commun. Mass Spectrom. 2008; 22: 2555–2560

DOI: 10.1002/rcm

2560 J. M. Hettick et al.

discrimination, the resubstitution and cross-validation

methods both result in 0% error rate (100% correct

identification).

Cluster analysis was also used to compare the similarity

between the various Penicillium MALDI-TOF MS datasets

(Fig. 7). Cluster analysis seeks to discover structure and/or

similarity in datasets. The data is presented in a dendrogram,

which groups together the two observations that are most

similar. In each case, the eight replicate cultures for each

species clustered together, indicating reproducibility

between mass spectra. Similarly, species belonging to the

subgenus Biverticillium and Furcatum were also clustered

together, which suggests that these species are closely

related. It is important to note that, although the cluster

analysis suggests similarities between spectra, the discrimi-

nant analysis was able to correctly classify each of these

Penicillium species. In light of the well-documented sub-

jectivity and the essential mycological training associated

with the accurate identification of environmental and

clinically derived Penicillia in viable and non-viable

samples,17 the results of the current experiments are exciting

and clearly demonstrate the utility of MALDI-TOF MS to

objectively differentiate and identify various Penicillium

species isolated from various environmental and clinical

samples.

CONCLUSIONS

MALDI-TOF MS of fungal isolates is demonstrated to be a

valuable tool for discriminating between species of the genus

Penicillium. The information content of the MALDI-TOF

fingerprint mass spectra is dramatically improved by using a

bead-beating procedure in conjunction with chemical

extraction in acetonitrile/TFA. The Penicillium species can

be correctly identified with 100% accuracy using canonical

discriminant analysis with stepwise variable selection.

Resubstitution and cross-validation error estimates both

result in a 0% error rate for the dataset of 96 fungal

fingerprint mass spectra. Although the fungal fingerprint

mass spectra produced by the described methodology are

highly reproducible, the spectra of Penicillium species

obtained in our laboratory differ from others published

previously, reiterating the importance of standardized

culture, sample preparation, and mass spectrometry con-

ditions for MS-based identification of microbiological

samples.

AcknowledgementsThis work was supported in part by the Inter-Agency Agree-

ment NIEHS Y1-ES0001-06. The findings and conclusions in

this report are those of the author(s) and do not necessarily

represent the views of theNational Institute for Occupational

Safety and Health.

Published in 2008 by John Wiley & Sons, Ltd.

REFERENCES

1. Paterson RR, Venancio A, Lima N. Res. Microbiol. 2004; 155:507.

2. Skouboe P, Frisvad JC, Taylor JW, Lauritsen D, Boysen M,Roseen L. Mycol. Res. 1999; 7: 873.

3. Pitt JI. The Genus Penicillium and Its Teleomorphic States Eupe-nicillium and Taraomyces. Academic Press: London, 1979.

4. Larsen TO, Frisvad JC. Mycol. Res. 1995; 99: 1167.5. Smedsgaard J, Frisvad JC. J. Microbiol. Methods 1996; 25: 5.6. Smedsgaard J, Frisvad JC. Biochem. Syst. Ecol. 1997; 25: 51.7. Smedsgaard J, Hansen ME, Frisvad JC. Stud. Mycol. 2004; 49:

243.8. Hong SG, Park YD, JeongW, Bae KS. J. Microbiol. 2000; 38: 62.9. Wang L, Zhuang WY. Biosystems 2007; 88: 113.10. Peterson SW. In Integration of Modern Taxonomic Methods

for Penicillium and Aspergillus Classification, Samson RA,Pitt JI (eds). Harwood Academic Publishers: Amsterdam,2000; 163.

11. Verweij PE, Meis JF, van den Hurk P, Zoll J, Samson RA,Melchers WJ. J. Med. Vet. Mycol. 1995; 33: 185.

12. Samson RA, Seifert KA, Kuijpers FA, Houbraken JAMP,Frisvad JC. Stud. Mycol. 2004; 49: 175.

13. Frisvad JC, Filtenborg O, Samson RA, Stolk AC. Antonie VanLeeuwenhoek 1990; 57: 179.

14. Seifert KA, Samson RA, Dewaard JR, Houbraken J, LevesqueCA, Moncalvo JM, Louis-Seize G, Hebert PD. Proc. Natl.Acad. Sci. USA 2007; 104: 3901.

15. Pitt JI. Can. J. Bot. 1995; 73: S768.16. La Guerche S, Garcia C, Darriet P, Dubourdieu D, Labarere J.

Curr. Microbiol. 2004; 48: 405.17. Fischer G, Braun S, Dott W. Int. J. Hyg. Environ. Health 2003;

206: 65.18. Frisvad JC, Nielsen KF, Samson RA. Adv. Exp. Med. Biol.

2005; 571: 33.19. Karas M, Hillenkamp F. Anal. Chem. 1988; 60: 2299.20. Karas M, Bachmann D, Bahr U, Hillenkamp F. Int. J. Mass

Spectrom. 1987; 78: 53.21. Cotter RJ. Anal. Chem. 1999; 71: 445A.22. Fenselau C, Demirev PA. Mass Spectrom. Rev. 2001; 20: 157.23. Lay JO Jr. Mass Spectrom. Rev. 2001; 20: 172.24. Lynn EC, Chung M, Tsai W, Han C. Rapid Commun. Mass

Spectrom. 1999; 13: 2022.25. Siegrist TJ, Anderson PD, HuenWH, Kleinheinz GT,McDer-

mott CM, Sandrin TR. J. Microbiol. Methods 2007; 68: 554.26. Hettick JM, Kashon ML, Simpson JP, Siegel PD, Mazurek

GH, Weissman DN. Anal. Chem. 2004; 76: 5769.27. Hettick JM, Kashon ML, Slaven JE, Ma Y, Simpson JP, Siegel

PD, Mazurek GN, Weissman DN. Proteomics 2006; 6: 6416.28. Amiri-Eliasi B, Fenselau C. Anal. Chem. 2001; 73: 5228.29. Welham KJ, Domin MA, Johnson K, Jones L, Ashton DS.

Rapid Commun. Mass Spectrom. 2000; 14: 307.30. Valentine NB, Wahl JH, Kingsley MT, Wahl KL. Rapid

Commun. Mass Spectrom. 2002; 16: 1352.31. Chen HY, Chen YC. Rapid Commun. Mass Spectrom. 2005; 19:

3564.32. Li T, Liu B, Chen Y. Rapid Commun. Mass Spectrom. 2000; 14:

2393.33. Saenz AJ, Petersen CE, Valentine NB, Gantt SL, Jarman KH,

Kingsley MT, Wahl KL. Rapid Commun. Mass Spectrom. 1999;13: 1580.

34. Arnold RJ, Karty JA, Ellington AD, Reilly JP. Anal. Chem.1999; 71: 1990.

35. Samson RA, Gams W. A. van Leeew. J. Microb. 1984; 50: 815.36. Braga-Neto U, Hashimoto R, Dougherty ER, Nguyen DV,

Carroll RJ. Bioinformatics 2004; 20: 253.

Rapid Commun. Mass Spectrom. 2008; 22: 2555–2560

DOI: 10.1002/rcm