Volume 56, Number 12, 2002 APPLIED SPECTROSCOPY 15450003-7028 / 02 / 5612-1545$2.00 / 0q 2002 Society for Applied Spectroscopy

Fourier Transform Infrared Microspectroscopy as a Tool toDifferentiate Nitzschia closterium and Nitzschia longissima

SUZANNE VARDY* and PHILLIPA UWINSDepartment of Chemistry, University of Queensland, St Lucia, QLD, 4072 (S.V.); and CMM, University of Queensland, St Lucia,QLD, 4072 (P.U.)

Nitzschia closterium and Nitzschia longissima are two species ofdiatom that are extremely dif� cult to differentiate using lightmicroscopy. This paper details an investigation into the use ofFT-IR microscopy combined with discriminant analysis to dif-ferentiate between these species. Spectra were taken from un-identi� ed samples and classi� ed against a training set using ei-ther Mahalanobis distances or principal component analysiscombined with canonical discriminant analysis. Unidenti� edsamples were classi� ed with a 100% accuracy using both math-ematical techniques.

Index Headings: Diatoms; FT-IR microspectroscopy; Identi� cation.

INTRODUCTION

Diatoms (Bacillariophyta) are single-celled phyto-plankton that are encased in a siliceous frustule and areoften associated with toxic blooms known as red tides.Nitzschia closterium (5 Cylindrotheca closterium ) andNitzschia longissima are two species of diatom that arenotoriously dif� cult to differentiate using light micros-copy. Light microscopy is the most common methodof identifying algae in the laboratory as it is cheap andan easily accessible method, although a taxonomic ex-pert is usually needed to identify the organisms to thespecies level. Hasle and Syversten1 state that ‘‘the iden-ti� cation of C. closterium may have caused more prob-lems and more confusion than the identi� cation of anyother diatom encountered in marine plankton. Whenworking with coastal material the most common prob-lem is distinguishing between C. closterium and thecoarser N. longissima.’’ The two species are clearlydistinguishable using scanning electron microscopy(SEM); however, SEM is expensive, involves time-con-suming sample preparation, and can be hard to access.It is important to be able to differentiate between thesespecies as N. closterium has been associated with themortality of prawn larvae in aquaculture hatcheries dueto tissue damage of the host caused by the spines ofthe diatom, 2 whereas there have been no reports of tox-icity associated with N. longissima.

Recently, a number of automated methods of analysishave been investigated in an attempt to develop a rapidmethod of identifying bloom algal species. Flow cytom-etry,3–6 automated image analysis,7,8 and resonance Ra-man spectroscopy9 have all been investigated, yet thesemethods have several limitations and have yet to providean all encompassing method of identifying unknown phy-

Received 2 June 2002; accepted 8 August 2002.* Author to whom correspondence should be sent. Current address:

SERG, Gatty Marine Laboratory, University of St. Andrews, St. An-drews, Fife, KY168LB Scotland.

toplankton samples to the species level. FT-IR micros-copy has been successfully investigated as a tool for clas-sifying bacteria10,11 and cyanobacteria.12

The use of FT-IR spectroscopy in this application isrelatively simple. Essentially, the spectrum of whole cellsis obtained, each spectrum being a summation of all theinfrared-active biochemical components of the cell. Aseach species contains slightly different combinations andamounts of these components, small variations in thespectra are used to differentiate samples. The use of FT-IR to differentiate between N. longissima and N. closter-ium is a simple example of the usefulness of the tool.

MATERIALS AND METHODS

Cultures and Samples. Nitzschia closterium was pur-chased from the Hobart CSIRO culture collection, codeCS-5, isolated from Port Hacking, Australia. Cultureswere maintained in an F2 medium (recipe supplied byHobart CSIRO), at 22 8C with a 12 h light/12 h darkcycle. Nitzschia longissima was collected during a bloomevent in the Brisbane River, Queensland, Australia, andwas identi� ed using SEM. The bloom was located ap-proximately 26 km upstream of the river mouth.

Electron Microscopy and Sample Preparation. Twomilliliters of either N. closterium or N. longissima samplewas removed and centrifuged at 1000 rpm for 5 min.Supernatant liquid was removed. The sample was thentreated using concentrated nitric acid for 2 h to removeextracellular material that obscures frustule details. Adrop of sample was then secured to a microscope coverslip using poly-l-lysine and platinum coated using anElko IB5 ion coater. Images used to independently iden-tify the species were collected on a JEOL 6400F Scan-ning Electron Microscope.

Sample Preparation for FT-IR. Two milliliters ofculture was removed and centrifuged at 1000 rpm for 20min. Samples were washed with millipore water and cen-trifuged to remove growth media. This was repeated threetimes. Approximately 1.8 mL of supernatant liquid wasremoved and the pellet was resuspended in the remainingwater. The suspension was then pippetted onto a 3M IRCard (polyethylene). Samples were placed in a vacuumdessicator overnight to dehydrate.

Spectra Collection. Spectra were obtained using aPerkin Elmer System 2000 FT-IR spectrometer, with amercury cadmium telluride (MCT) detector cooled withliquid nitrogen. The FT-IR is attached to an infrared mi-croscope, also equipped with a small MCT detector. Res-olution was set at 8, and 50 scans were taken and aver-aged for each spectra. The aperture of the microscopewas set at 25 mm 2, and the spectra of different clusters

1546 Volume 56, Number 12, 2002

TABLE I. Summary of taxonomic descriptions of Nitzschia closterium (nc) and Nitzschia longissima (nl).

Descriptiona

nc nl

Descriptionb

nc nl

Descrip tionc

nc nl

Samples

nc nl

Length (mm)Width (mm)Fibulae/10 mmStriae/10 mmSiliceous

30–4002.5–810–1270–100weak

125–4506–76–14

52–60strong

25–1801.5–812–25

· · ·weak

70–2004–8

10–1750–60strong

15–59· · ·· · ·· · ·

weak

270–5005–86–7

16–18strong

321.3

20140weak

883

1560

strong

a See Ref. 1.b See Ref. 13.c See Ref. 14.

FIG. 1. The normalized spectra of N. closterium (� lled line) and N.longissima (dashed line).

TABLE II. Tentative assignments of peaks in the FT-IR spectraof micro-organisms (see Refs. 10 and 12).

cm21 Strength Assignations and comments

1745–1735 S C5O stretch of saturated esters (lipids andfatty acids)

1720–1700 S C5O stretch of –COOH, aliphatic carbox-ylic acid (dimer)

1690–1620 S C5O secondary amide stretch, amide(I)band (proteins and peptides)

1625–1585 M C5C sk, i–p aromaticø1500 var C5C sk, i–p aromatic1570–1515 s N–H def. secondary amide (II) band (pro-

teins and peptides)ø1460 m C–H2 bend (asym)–CH 3 bend (asym)1460–1400 s C–O sym str–COO2 carboxylate1385–1375 m C–H2 bend (sym)–CH 3 bend (sym) (pro-

teins)1340 w C–H bend alkane1305–1200 m N–H def. secondary amide: amide (III)

band1270–1150 s C–O str –(O5)C–O–R in esters1250–1150 vs P5O str phosphoric ester H bonded P5O1200–1040 C–O str C–O–C–O–C, cyclic acetal (4–5

bands)1200–1170 s C–O str propionic and higher esters1200–1000 s C–OH str carbohydrates1090–1030 vs P–O–C (phosphodiester backbone of nu-

cleic acids, polysaccharides)1090–102015 vs, br Si–O–Si open-chain siloxanes; Si–O–C str970–940 br P–O–P pyrophosphate770–730 s C–H o.o.p bend unsubstituted phenyl710–690720 m (br) N–H def. secondary amide, bonded amide

V band

of cells was collected by moving the stage via computercontrol. An aperture size of less than 25 mm 2 producedpoor spectra. To test for reproducibility, spectra for Nitz-schia closterium were collected monthly over a period ofa year from different subcultures of the same initial cul-ture. Cells in the stationary growth phase were used. Itis not possible to tell whether the Nitzschia longissimasample was from an exponential or a stationary growthphase, although the crust on the surface of the bloom areaindicates that the bloom was probably in a stationarygrowth phase. Due to the different levels of biochemicalcomponents present in the exponential and stationaryphases, it is important that this is taken into account dur-ing analysis.

Data Analysis. Data were normalized using the mostintense peak in the spectra, at around 1080 cm21, andbaseline-corrected using an interactive polynomial func-tion of the form:

baseline 5 A 1 D*X3

Two methods of analysis were used to identify ‘‘un-knowns’’: (1) discriminant analysis based on Mahalan-obis distances using the Grams PLS software, and (2)PCA using Grams PLS software combined with canoni-cal discriminant analysis using Statistics for Social Sci-entists (SPSS). Scores from the PCA were transferred toSPSS using Microsoft Excel software. Both these statis-tical methods rely on a ‘‘training set’’ of data. The train-ing samples for N. closterium were taken from samplescollected on 3 days, for a total of 23 spectra. The ‘‘un-knowns’’ (13 spectra) were taken from samples collectedon different days and from different subcultures. The N.longissima samples were taken from the same bloom, but

the spectra of the training set (20 spectra) and the ‘‘un-known’’ set (18 spectra) were taken from different sam-ples of the bloom.

RESULTS AND DISCUSSION

The main taxonomic features that are used to classifythe N. longissima and N. closterium species are summa-rized in Table I. It can be seen that neither the numbernor the size of the reported features are characteristicenough to adequately distinguish the species as the rangefor all values overlap. Hence, the taxonomic features oflength, width, � bulae/10 mm, and striae/10 mm do notnecessarily place a species into a particular group. Theonly feature that clearly distinguishes the species is theamount of silici� cation—a rather arbitrary feature whenviewed under the light microscope.

In the complicated mixture of chemicals that are con-tained in a cell, absolute peak assignment is extremelydif� cult although some authors have tentatively assigned

APPLIED SPECTROSCOPY 1547

TABLE III. Ratio of siloxane peak height (at 1068 cm21) to am-ide(I) peak height (at 1655 cm21) for N. closterium and N. longissi-ma.

N. closterium N. longissima

MeanStandard deviation

0.990.11

3.800.25

TABLE IV. Mahalanobis distances between training sets and un-knowns.

N. closteriumtraining set

N. longissimatraining set

N. closterium ‘unidenti� ed’N. longissima ‘unidenti� ed’

0.79 6 0.18908.72 6 300.26

39.62 6 11.342.12 6 1.23

TABLE V. Classi� cation of unknowns using PCA and canonicaldiscriminant analysis.

Training set

NC NL

Unknowns NCNL

130

018

FIG. 2. PC1 scores vs. PC4 scores ( m : N. closterium; m : T. closterium ‘unknown’; 3: N. longissima; l : N. longissima ‘unknown’).

characteristic peaks in a range of micro-organisms.10,12

Table II details possible assignments of peaks arisingfrom the vibrations of commonly occurring cellular con-stituents such as proteins, carbohydrates, amino acids,polysaccharides, and nucleic acids. Diatoms have an ad-ditional feature that is evident in the IR spectra, and thatis the presence of siliceous components in the cell wall.Although the structure of the diatom wall and its inter-action with outer membrane proteins is yet to be fullyelucidated,16 it is known that the wall consists of a poly-meric siloxane. The extremely large peak at around 1075cm21, present in the spectra of both species, is possiblycontributed to by a Si–O–Si stretch. As this region alsocovers the P–O–C ring vibrations of polysaccharides andperhaps the Si–O–C stretch of the cell wall bonded toproteins (a theorized structure), the region 1100–1000cm21 will contain a combination of these bands.

The normalized FT-IR spectra obtained for cultured N.closterium and wild N. longissima are compared in Fig.1. The differences between the spectra are clear fromvisual examination. It was noted previously that the ma-jor visual differentiating feature in microscopic analysisbetween N. closterium and N. longissima was the degreeof silici� cation. The degree of silici� cation is also thepredominant distinguishing feature in the spectra, withthe ratio of the height of the 1075 cm21 peak to the heightof the amide(I) peak being around 3.8 times greater in N.longissima than in N. closterium (see Table III). Hence,the spectral information is corroborated by known de-scriptive information.

To con� rm that the apparent spectral differences be-tween the species were valid over a range of samples, thespectra were further investigated using traditional clas-si� cation techniques. Two mathematical methods wereused to compare spectra from unidenti� ed samples withspectra from identi� ed (or training) samples: (1) discrim-inant analysis using Mahalanobis distances, and (2) prin-cipal component analysis followed by canonical discrim-inant analysis. Discriminant analysis using Mahalanobisdistance is a quick and easy method to use and comes inthe software package Grams PLS. However, it is a mod-eling method that is very sensitive to baseline � uctuationsand outliers. The results for discriminant analysis usingthe Mahalanobis distances are summarized in Table IV.It has been suggested that a good match occurs when theunidenti� ed sample is less than three Mahalanobis dis-tances from the training set.17 For both species, the un-identi� ed samples had less than three Mahalanobis dis-tances with the correct training sets. Hence, all of theunidenti� ed samples were correctly matched with theirtraining sets, whereas the Mahalanobis distances between

1548 Volume 56, Number 12, 2002

the unidenti� ed samples and the incorrect training setswere of a much greater magnitude.

Principle components analysis combined with a clas-si� cation technique is more robust than discriminantanalysis using Mahalanobis distances. The canonical dis-criminant function calculated from scores obtained froma PCA on both species correctly classi� ed all ‘unknowns’(see Table V). The � rst four PCs described 99% of thevariance, so only the scores from these PCs were used inorder to avoid over� tting. The canonical discriminantfunction coef� cients used for classi� cation were calcu-lated from PC1 and PC4 by the program. Figure 2 showsthe wide separation obtained between the two groups andillustrates how unambiguously the ‘unknown’ was as-signed to the correct training set.

CONCLUSION

Fourier transform infrared microscopy has proved tobe an easy and accurate method of differentiating N. clos-terium and N. longissima based on known biochemicaldifferences (particularly silici� cation). Further applica-tion of this method with other red tide organisms maylead to the development of a much-needed method forrapidly identifying hazardous species.

1. G. R. Hasle and E. E. Syversten, ‘‘Marine Diatoms’’, in IdentifyingMarine Diatoms and Dino� agellates, C. R. Tomas, Ed. (AcademicPress, San Diego, 1996), Chap. 2, p. 294.

2. A. S. Sahul Hameed, J. Aquacult. Trop. 10, 337 (1995).3. W. K. W. Li, Cytometry 10, 564 (1989).4. R. J. Olson, E. R. Zettler, and O. K. Anderson, Cytometry 10, 636

(1989).5. L. Boddy, C. W. Morris, M. F. Wilkins, G. A. Tarran, and P. H.

Burkhill, Cytometry 15, 283 (1994).6. J. W. Hofstraat, W. J. M. van Zeijl, M. E. J. de Vreeze, J. C. H.

Peeters, L. Peperzak, F. Colijn, and T. W. M. Rademaker, J. Plank-ton Res. 16, 1197 (1994).

7. G. Gorsky, P. Guilbert, and E. Valenta, Mar. Ecol. Prog. Ser. 58,133 (1989).

8. S. U. Thiel, R. J. Wiltshire, and L. J. Davies, Water Res. 29, 2398(1995).

9. Q. Wu, W. H. Nelson, P. Hargraves, J. Zhang, C. W. Brown, and J.Selenbinder, Anal. Chem. 70, 1782 (1998).

10. D. Helm, H. Labischinski, and D. J. Naumann, Microbiol. Methods14, 127 (1991).

11. D. Naumann, C. P. Schultz, and D. Helm, ‘‘What can infrared spec-troscopy tell us about the structure and composition of intact bac-terial cells?’’, in Infrared Spectroscopy of Biomolecule, H. H.Mantsch and D. Chapman, Eds. (Wiley Liss, New York, 1996),Chap. 10, p. 279.

12. M. Kansiz, P. Heraud, B. Wood, F. Burden, J. Beardall, and D.McNaughton, Phytochemistry 52, 407 (1999).

13. Y. Fukuyo, H. Takano, M. Chihara, and K. Matsuoka, Eds., RedTide Organisms in Japan—An Illustrated Taxonomic Guide (Uch-ida Rokakuho Co. Ltd., Tokyo, 1990).

14. J. John, The Diatom Flora of the Swan Estuary, Western Australia(J. Kramer, Vaduz, 1983), Bibliotheca phycologica, Bd.64.

15. L. J. Bellamy, The Infrared Spectra of Complex Molecules (Me-thuen and Co. Ltd, London, 1958), p. 334.

16. K. D. Lobel, J. K. West, and L. L. Hench, Mar. Biol. 126, 353(1996).

17. H. L. Mark and D. Tunnell, Anal. Chem. 57, 1449 (1985).