Microorganisms Detection on Substrates using QCL Spectroscopy Amira C. Padilla-Jiméneza, William Ortiz-Riveraa, John R. Castro-Suareza, Carlos Ríos-Velázquezb, Iris Vázquez-Ayalaa, Samuel P. Hernández-Riveraa,∗ ALERT DHS Center of Excellence for Explosives Dept. of Chemistry, University of Puerto Rico, Mayagüez, PR b Dept. of Biology, University of Puerto Rico, Mayagüez, PR
ABSTRACT Recent investigations have focused on the improvement of rapid and accurate methods to develop spectroscopic markers of compounds constituting microorganisms that are considered biological threats. Quantum cascade lasers (QCL) systems have revolutionized many areas of research and development in defense and security applications, including his area of research. Infrared spectroscopy detection based on QCL was employed to acquire mid infrared (MIR) spectral signatures of Bacillus thuringiensis (Bt), Escherichia coli (Ec) and Staphylococcus epidermidis (Se), which were used as biological agent simulants of biothreats. The experiments were carried out in reflection mode on various substrates such as cardboard, glass, travel baggage, wood and stainless steel. Chemometrics statistical routines such as principal component analysis (PCA) regression and partial least squares-discriminant analysis (PLS-DA) were applied to the recorded MIR spectra. The results show that the infrared vibrational techniques investigated are useful for classification/detection of the target microorganisms on the types of substrates studied. Keywords: microorganisms, MIR-spectroscopy, quantum cascade lasers (QCLs), PCA, PLS-DA
1. INTRODUCTION Defense and security agencies, as well as the private sector, are highly interested in finding new ways for the detection and identification of unknown chemical and biological threats. By developing new capabilities and expanding on current experiences, food industries, environment protection agencies, pharmaceutical and biotechnology industries will also benefit using infrared sensing applications. Spectroscopic signatures obtained from the vibrational frequencies of most molecules correspond to the frequencies of mid-infrared (MIR) light and contain direct molecular specific information. In MIR spectroscopy, IR radiation in the desired wavelength range is directed through the sample to be analyzed, and the intensity of the transmitted/reflected radiation is measured as a function of wavenumber. Typically, the technique is used to study organic compounds and biological samples using spectral region spanned by range 4000–400 cm−1. This radiation is absorbed by molecular vibrations, where the various atoms that comprise molecules vibrate about their equilibrium positions. The salient feature of MIR spectroscopy, as well as Raman spectroscopy, is that the MIR spectroscopic analysis is equivalent to functional group analysis. The spectroscopic technique is not only capable of assisting in the identification of the kinds of functional groups present in the molecule, such as C=O and C=N, but also gives information on the environment in which they are placed. "Fingerprints" of molecules are obtained, since the MIR absorption spectrum of a compound is characteristic for that compound. MIR spectroscopy has a much stronger absorption probabilities than corresponding near-infrared spectroscopy ones1,2. Recent research has focused on the development of rapid and accurate methods to recognize agents representative of microorganisms such as, disposable electrochemical immunosensor3, immobilized probes for the detection of E. coli6 and solid phase microextration/gas chromatography/mass spectrometry of bacteria.7 However, these techniques are time consuming, expensive and involve many preparation steps and necessary selective pre-enrichment. Although identification and discrimination of bacterial spores using mid-infrared technologies has already been reported4,5, a ∗
Authors for correspondence: [email protected]
, [email protected]
, [email protected]
Chemical, Biological, Radiological, Nuclear, and Explosives (CBRNE) Sensing XIV, edited by Augustus Way Fountain, Proc. of SPIE Vol. 8710, 871019 · © 2013 SPIE · CCC code: 0277-786X/13/$18 · doi: 10.1117/12.2016099 Proc. of SPIE Vol. 8710 871019-1
quantum cascade laser (QCL) based spectrometer that operates in the MIR region was selected for this work.6 QCLs have revolutionized many areas of research and development in defense and security applications.7 QCL sources are unipolar semiconductor injection lasers based on intersubband transitions in a multiple quantum-well heterostructure.8 They operate at wavelengths in the MIR starting at about 3 µm, which matches well with the fundamental vibrational absorption bands of many chemical species in comparison with conventional diode sources where the laser emission generally matches the weaker overtone. The emission wavelength of a QCL depends on the thickness of the quantum well and barrier layers of the active region rather than the band gap of diode lasers. QCLs operate at near room temperature, produce milliwatts of radiation, and offer the possibility of tailoring the emission wavelength within a broad range of frequencies.9,12 Using reflection mode, where the same optical device is used both to project the beam onto the specimen and to collect the reflected radiation, the QCL system was tested to determine if this newly designed system has better spectral signals MIR for bacteria as Bacillus thuringiensis (Bt), a Gram-positive bacteria with the ability to form spores, a latent state that is highly resistant to both chemical and thermal extremes13. As a bacillus species, it has a bacterial life cycle in which it grows as vegetative cell forming endospores as defense mechanism. Endospores are highly resistant to environmental stresses such as high temperature, irradiation, strong acids, disinfectants, etc. They are able to tolerate extreme environments thus making them suitable for transporting before or during a biological attack14. Bt, is a nonharmful microorganism to humans and has been chosen as a model because of its similarities with Bacillus anthracis (Anthrax) which has a great potential of being used for terrorist attacks. Likewise, Escherichia coli (Ec) is a member of the Enterobacteriaceae family of bacteria Gram-negative. It is a thermo-tolerant coliform present in the intestinal flora of warm blood animals, thus used as an indicator of fecal contamination in food15. Staphylococcus epidermidis (Se), also a sphere forming Gram-positive bacterium, is also part of our normal flora, usually the skin.16 These bacteria were deposited on different substrates such as cardboard, glass, travel baggage, wood, and stainless steel. As vibrational spectra of bacterial cells consist of signal contributions of all components in the cells, they reflect their overall molecular composition. The MIR studies of Naumann and other workers were extended in order to identify and discriminate vegetative bacteria17-19 by applying established chemometrics methods to our spectra, the data could be reduced to demonstrate the capability of this spectroscopy technique in the identification and discrimination of bacterial cells between three types of species considered as biological agent simulants.20
2. METHODS 2.1. Preparation of Bacterial Samples The bacterial strains Bt (ATCC #35646), Ec (ATCC #8789), and Se (ATCC #2228), were provided by the Microbial Biotechnology and Bioprospecting Lab, Biology Department, University of Puerto Rico-Mayagüez Campus. The selection of these biological agents for testing was based on previous work in IR region and availability of the organisms. Pure cultures were grown using Miller modified Luria-Bertani (LB) agar and broth (Fisher Scientific International; Thermo Fisher Scientific, Waltham, MA) and later stored at -800C in micro vials containing 20% glycerol (cryoprotectant) until analyzed. After a satisfactory growth was achieved on agar colonies of Bt, Ec and Se then they were isolated on LB plates and inoculated into 5.0 mL LB broth or tryptic soy broth (TSB) and left to grow overnight. Se and Bt were placed in an orbital shaker at 32°C (∼120 rpm) for 24 hours and cultured for 72 hours. Ec was placed in an orbital shaker at 37°C until growth of 5 h post inoculation. Sub-cultures were diluted 1:50 in appropriate media and centrifuged at 5K x g for 5 min at 4°C in order to form pellets. Pellets of bacteria were washed once with 20.0 mL of 1% phosphate buffered saline (PBS) to remove growth media. Final pellets were re-suspended in 4.0 mL of PBS (Ec and Se) or distilled water (Bt) that continued to grow for 72 h at 32°C and 250 rpm. Harvested spores were stored at 4°C in distilled, deionized H2O until the experiment was performed. Serial dilution followed by plating on LB agar in order to enumerate bacteria suspended in PBS. Plates were incubated at 37°C for > 24 h in order to form colonies. These were counted on all plates, and the concentration of the stock suspension of bacteria was determined by back calculating the dilution series. The concentrations obtained were 5.9x1010/5.0x103, 3.5x109 and 9.5x109 colony forming units per milliliter (CFU/mL) for Bt vegetative cell/endospores, Ec and Se respectively. The OD600 (Biophotometer, Eppendorf) was measured before and after centrifugation, because it
was considerred that the inccubation time of o the organism m must be deteermined from iits growth curvve starting at ann optical density of 0.0 025 at 600 nm wavelength. mentation 2.2. Instrum SEM imagess were obtain ned on a JEO OL-JSM 6500 instrument X XL series 30S--Philips/FEI too verify the ssize and morphology of bacterial ceells (Fig.1). Th his figure show ws Bt. and Ec. with rod shappe and an approximate size between 0.5-1.0 x 1.4-3.0μm. Se. ap ppears sphericaally-shaped witth a size of 0.55x1.5 μm.
Figure 1: SEM M images of threee types bacteriaa used in this reseearch: a Bt, b. Ecc and c. Se. Opeerating voltage: 115 kV.
A LaserScan n™, model 71 12, was used for f QCL data acquisition inn the MIR ~8 30-1430 cm-1 of the three bacterial suspension samples. The baacterial samplees were deposited on the surffaces under stuudy, which inclluded cardboarrd, glass, stainless steeel, wood and trravel baggage, in order to dettermine the efffectiveness of tthis instrumentt to detect the ppresence of bacteria on n the substrates of interest (Fig. 2).
Figure 2. Exp perimental set-u up for QCL MIR R spectroscopic system used foor detection of B Bt, Ec and Se oon real substratee such as stainless steel,, glass, cardboard d, travel baggage and wood.
The QCL speectrometer useed was a diffusse reflectance system s that opeerated at non-ccontact distancces of 6 in. Thee system was sensitivee to the way th he material is deposited d with h about 1% effficiency in cappturing light froom the ideal substrate. For this reason, the placem ment of the sam mple at the corrrect angle wass critical. The ssource light coould also be lost due to absorption orr multiple refleections on the material of in nterest, especiaally on thick suurfaces. Visible pointers (HeeNe laser beams) align ned with the no on-visible MIR R radiation illu uminate the lasser spot (approoximately 2 × 4 mm) on the surface, and built-in algorithms a enaabled near reall-time detection n of samples. Both trace andd bulk amountts of substancees can be detected to leevels as low ass 0.5 μg/cm2 ou ut in the field.200
2.3. Sample Preparation and Data analysis The detection of suspension bacteria present as traces on different substrates required a simple sample preparation methodology that would be able to deposit 10 μL of each bacterial sample on the selected solid substrate and designed for in situ measurements.21, 22 Aluminum plates were used as material support for placing substrates which contained 1015 samples of each of the 3 types of bacteria analyzed deposited on 1 cm × 1 cm substrates (Fig.2). OPUS 6.0 Software (Bruker Optics, Billerica, MA, USA) was used to analyze the data obtained. During the analysis performed, the spectra were normalized in order to allow a proper comparison of the spectra. For multivariate data analysis, a matrix interlinking the individual measurements as rows and the selected wavenumber values as columns was generated. The spectral intensity for each measurement-wavenumber combination comprised the matrix elements. Principal component analysis (PCA) is a well-known way to reduce the number of variables, in which a data set with many variables can be simplified by performing data reduction which makes it more easily interpretable. Partial least squares-discriminant analysis (PLS-DA) is a multivariate method used for classification of samples when it is necessary to reduce the number of variables and it is tentative whether the differences between groups will dominate the total variability of the samples. This algorithm was applied in MATLAB™ computational environment (The MathWorks, Inc., Natick, MA, 01760 USA), using the PLS-Toolbox v. 7.0.3 (Eigenvector Research, Inc., Wenatchee, WA, USA).
3. RESULTS MIR spectra of biomolecules allow the measurement of molecular vibrational modes that contain valuable information on biochemical make up.27, 28 Multivariate analysis assists in the process of handing large data sets, making spectral analysis viable. The spectral analysis performed included tentative assignments of bands, or peaks, according to reported characteristic MIR absorption frequencies for agents representative of the microorganisms tested: Bt, Ec, and Se.29, 30 A total of 836 different MIR experiments were carried out, but only 245 are reported for simplicity, each consisting of fifteen replicate spectral measurements of each bacterium on a given substrate. Previous reports have made use of the concept of using a single spectrum to represent a dispersion of replicate spectra, which is not unprecedented, because this concept has yielded meaningful and concrete results31, 32. In Figure 3 the MIR spectra of bacteria suspensions from the three strains (Bt, Ec, and Se) are shown. It was difficult to difference in the raw MIR spectra between the different classes on the various surfaces tested due to the high degree of band overlap. To solve this, some pre-treatments were applied. The first used pre-treatment was spectral normalization.
Av. Bt Cardboard Av. Ec Cardboard Av. Se Cardboard Av. Bt Glass Av. Ec Glass Av. Se Glass Av. Bt Travel bag Av. Ec Travel bag
Av. Se Travel bag Av. Bt Stainless steel Av. Ec Stainless steel
MIR S pectra
0.8 20 0.6
Av. Bt Wood Av. Ec Wood
Av. Se Wood Cardboard
Glass Travel bag
Av. Se Stainless steel
Figure 3. Normalized QCL spectra of bacterial suspensions.
Each bacterial species has a unique IR fingerprint f duee to the stretchhing and bendiing vibrations of molecular bbonds or functional grroups present in its proteins, nucleic acid ds, lipids, suggars, and lipoppolysaccharidees as illustrated in the reference speectra shown in n Figure 4.30 The cell surfacee characteristiccs vary betweeen Gram-positiive and Gram--negative bacteria. Graam-positive baacteria have a thicker t and rig gid layer of peeptidoglycan of 40-80% by w weight of the ccell wall than Gram-n negative bacteriia; also contain n teichoic acidss that are covaalently bound to the peptidogglycan, whereass, Gramnegative cellls do not contain teichoic accids, and havee lipoproteins covalently bouund to the pepptidoglycan in the cell walls. Gram--negative bacteeria have an ou uter membrane outside of thee peptidoglycann layer which ccontains phosppholipids in the inner layer and lipo opolysaccharid des in the outeer layer.31 The amino acid ccomposition off peptide chainns in the Gram-negativ ve bacterium Ec E consists of D-alanine, D-g glutamic acid, and meso-diaaminopimelic aacid, on the othher hand in the Gram--positive Se co onsists of L-alaanine, D-glutam mine, L-lysine,, and D-alaninne.32 MIR specttra measured ffor intact cells of bactteria are usuaally complex and a the peaks are broad duue to superpossition of contrributions from m all the biomoleculess present in a bacterial b cell (F Fig. 4).30
Bt Ref 0.1is
Wavenumber ,/ Cm-i
Figure 4. Refeerence MIR specctra of Bt, Ec and d Se in the specttral region studieed. 1 32, 34 Following Naumann’s N reco ommendations17, IR specctra should be analyzed for identification oof bacteria in thhe 1200-1 1500 cm reegion for bend ding vibrationss of fatty acid, proteins, andd phosphate-car arrying compouunds. The regiion from 900-1200 cm m-1 contains carrbohydrates baands of the miccrobial cell walll and the regioon of 700-900 cm-1 is the ‘finngerprint region’ that contains c weak but very uniqu ue vibrations characteristic c oof specific bactteria34. Given tthese regions, tentative assignments of functional groups g associaated with majorr vibration bannds in MIR speectra of bacteriia are showed in Table 1. Figure 5 shows s represen ntative MIR-QCL spectra (reeflection mode)) of each bacteerium analyzedd after depositting onto stainless steeel surface.
Ta able 1. Tentativee band assignmeents used for baccterial identificattion.17, 32, 34-37 Wavenumber W (ccm-1) 1400 1310-1240 1240 1200-900 1085 900-700
Molecular vibrations fu unctional group ps/ biomolecule C= =O symmetric sttretching of COO O- group in aminno acids, fatty accids. Am mide III band co omponents of prooteins. P= =O asymmetric stretching s of phoosphodiesters in phospholipids. C--O-C, C-O domiinated by ring viibrations in varioous polysacchariides. P= =O symmetric sttretching in DNA A, RNA and phoospholipids. "F Fingerprint region"
ú 0.7 C L) cl)
0.5 0.4 0.3
Ec on SS
Bt on SS Se on SS
avenumber / cifll-1
Figure 5. Mid d-infrared spectraa of Bt, Ec and Se S at room tempeerature depositedd on stainless steeel substrate. 3.1 PCA regressions
The data sett of 245 MIR spectra that contain c spectraal information of bacterial ssuspension waas subjected too several preprocessing g steps to inco orporate unwan nted variability y into the PCA A models. It waas observed thaat the model built with scores plot itt not have welll defined separration of classees between grooups of bacteriia on different substrates, altthough it represented the t main part of o the data variaance. These ressults are presennted in Table 22. Table 2. Classification between n groups of bacteeria on differentt substrates Acctual discrimina ation
Gro oup size
Bt 45 (60.0%) 3 (3.89%) 14 (18.7%) 3 (16.7%)
Predicted disccrimination Ec Se 18 8 (24.0%) (10.7%) 64 9 (11.7%) (83.1%) 3 50 (4.0%) (66.7%) 4 0 (22.2%) (0.00%)
Substrates 4 (5.3%) 1 (1.3%) 8 (10.7%) 11 (61.1%)
Perrcentage of casees correctly classsified: 69.4%
Another PCA A model was built b by selectting ten to fifteen MIR specctra of sampless of each bacterium on eachh surface (cardboard, travel t bag, wo ood, glass and stainless steel), for a total oof 225 MIR sppectra. They w were then preprocessed employing 1st derivative (order: 2, win ndow: 15pt.) and mean ceentered on alll substrates. A Additionally thhe SNV preprocessing g was applied only to data acquired on wo ood substrate fo for obtaining beest results. Thee data from setts Bt, Ec and Se, weree ran together in the PCA model m and the variance v descriibed by each P PC was examiined on each ssubstrate. Figure 6A, show s the scorees plot (PC3 vs. v PC1) on caardboard, whichh indicates theere are not good separation between datasets acco ording to bacteerium type. Ev vidently, PC1 (52.84% ( variannce) tends to rrelate to the diifferences betw ween the three bacteriaa. The 60.0% of o Bt samples were w classified d as Ec, while tthat Ec and Se were classifieed correctly in 93.0 and 100% respecctively. As it can be seen in Figure F 6B, thee scores plot (P PC2 vs.PC1) oon glass substraate shown a veery good separation off Bt samples, while w it was ob bserved that forr Ec and Se sam mples the 30.00% are closest together since they are the most sim milar spectra on glass. This treend was also in nvestigated in tthe scores plotss for PCs on trravel bag, as shhowed in Figure 7. It was w found thatt (PC2 vs.PC1, Fig. 7A) scores plot do nott show class separation betw ween the three types of bacteria on trravel bag substtrates, although h it represents part p of the dataa variance (14..2 and 52.7%, rrespectively), bbut (PC3 vs. PC2, Fig.. 7B) scores plo ot accounted fo or minimum vaariance: 10.2% % and 14.2%, reespectively.
:ore on PC 3 (10.£
Ec C Bt Cal
H.001 & Rt f
)K Test Ec
X Test Bt
-0.002 L -0.0 1
'e on PC 1 (52.84°r
Scor e on PC 1 (96.00 °l
Figure 6. PCA A for QCL specttra of Bt, Ec and d Se deposited on n: (A) cardboardd; (B) glass. 3 (10.33 %)
Bt C Bt Cal
O ó A
X Test Bt ff-
ó ó N
ó 0 A
on PC 2 (14.17%)
+ÑA N A 0Of 0 °
X Test Ec
3 1 -0.09 -0.06
-0. 15 -0.12 -0.09 -0.06 -0.03 0.09
Score on PC 2 (14.1nx)
Scor"e on PC 1 (52.70%;
A for QCL specttra of Bt, Ec and d Se deposited on n travel bag: (A)) PC2 vs. PC1; (B B) PC3 vs. PC2. Figure 7. PCA
Bt Cal. Ec Cal.
* Test Ec
X lest Eft
+ Test Se 7
Scoire on PC 1 (44.7( ro)
Score on PC 1 (621.37"/x)
Figure 8. PCA A for QCL specttra of Bt, Ec and d Se deposited on n: (A) stainless ssteel; (B) wood.
Substrates where discrimination between the three bacteria was best were made of stainless steel. However, scores plots (e.g.: PC2 13.9% vs. PC1 44.7%) show that a significant variance is not represented by the three types of bacteria. Se samples (40.0%) were classified close together as Bt and Ec samples at 94% on substrates. These results indicate that the bacteria were discriminated according the cell wall characteristics between Gram (+) and Gram (-), (see Fig.8B). Scores plots of bacteria deposited on wood substrates required higher PCs, (PC3 5.65%, PC4 3.17% and PC5 2.93%). In addition it was observed that Ec samples were well classified at 100%, as is displayed (PC2 10.70% vs. PC1 63.37%) in Figure 8A, while Se samples (53.33%) were incorrectly characterized as Bt. Samples of Bt were then separated as Ec species (74.0%). These results demonstrate that PCA does result in good discrimination between the three types of bacteria. PCA, which is frequently used as an unsupervised classification method, might have not been powerful enough to achieve class separation, as it is no longer effective when within-group variations are larger than between group variations. In such situations use of supervised classification methods such as PLS-DA have to be considered. 3.1 PLS-DA Partial least squares discriminant analysis (PLS-DA) was employed as a chemometrics tool and classification method to differentiate bacterial species studied (Bt, Ec and Se) on five surfaces (cardboard, travel bag, wood, glass and stainless steel). Fifteen MIR spectra of samples of each bacterium on each surface were used for a total of 225 MIR spectra. Vibrational information was organized into two groups. About 75% of the samples spectra were randomly selected as training set for the calibration and cross-validation models. The other 25% of the spectra were used as external test set. In order to build a robust model, the spectral data was limited to the spectral regions of 848-1012 cm-1, 1022-1170 cm-1, and 1173-1400 cm-1. They were then pretreated employing smoothing (order 0: 15 pts.) and then 1st derivative (order: 2, window: 15 pts.) was employed as preprocessing technique. The cross validation procedure was performed using venetian blinds with 10 splits. This procedure builds a classification model with 90% of the spectra and then tests the remaining 10% to assess classification accuracy. The MIR spectra and the preprocessing of smooth and 1st derivative clearly improved the visualization of spectra of bacteria. The discrimination model for each bacteria and its respective surface are shown in Figures 9-12. These figures illustrate the cross-validated predicted classes versus samples, as shown in the PLS-DA plot. 1.4
PLS-DA Bt on Cardboard 1.1
CV Class Predicted
CV Class Predicted
Test set Bt
Thresold 0.6 0.4 0.2 0 0
Se Se test set
Ec 0.5 0.3 0.1
-0.1 0 -0.3
PLS-DA Ec on Cardboard
Test set Ec
CV Class Predicted
PLS-DA Se on Cardboard
0.2 0 0
Figure 9. PLS-DA plots for the discrimination of bacteria on cardboard surface.
X Bt Test S