Acta Tropica 185 (2018) 1–12

Contents lists available at ScienceDirect

Acta Tropica
journal homepage: www.elsevier.com/locate/actatropica

A novel use of infra-red spectroscopy (NIRS and ATR-FTIR) coupled with T

variable selection algorithms for the identification of insect species (Diptera:
Sarcophagidae) of medico-legal relevance
Taciano M. Barbosaa,1, Leomir A.S. de Limab,1, Marfran C.D. dos Santosa,b,1,
⁎
Simão D. Vasconcelosa,1, Renata A. Gamac,1, Kássio M.G. Limab,1,
a
Insects of Forensic Importance Research Group, Department of Zoology, Federal University of Pernambuco, Av. Prof. Moraes Rego, Recife, PE, 50.670-420, Brazil
b
Biological Chemistry and Chemometrics, Institute of Chemistry, Federal University of Rio Grande do Norte, Natal, RN, 59.072-970, Brazil
c
Laboratory of Insect and Vectors, Department of Microbiology and Parasitology, Federal University of Rio Grande do Norte, Natal, RN, 59.072-970, Brazil

A R T I C LE I N FO A B S T R A C T

Keywords: Unequivocal identification of fly specimens is an essential requirement in forensic entomology. Herein, a simple,
NIRS non-destructive and rapid method based on two vibrational spectroscopy techniques [Near-Infrared
Forensic entomology Spectroscopy (NIRS) and attenuated total reflection Fourier-transform infrared (ATR-FTIR) spectroscopy] cou-
ATR-FTIR pled with variable selection techniques such as genetic algorithm-linear discriminant analysis (GA-LDA) and
SPA-LDA
successive projection algorithm-linear discriminant analysis (SPA-LDA) were applied for identifying and dis-
GA-LDA
PCA
criminating six species of flesh flies (Diptera: Sarcophagidae) native to Neotropical regions. This novel approach
is based on the unique spectral “fingerprints” of their biochemical composition. One hundred sixty (160) NIRS
and FT-IR specimens (120 male, 40 female) were acquired; different pre-processing methods such as baseline
correction, derivative and Savitzky-Golay smoothing were also performed. In addition, the multivariate classi-
fication accuracy results were tested based on sensitivity, specificity, positive (or precision) and negative pre-
dictive values, Youden index, positive and negative likelihood ratios. Principal components analysis (PCA) was
employed for male vs. female category using NIRS, strongly showing the separation between the classes with
only three principal components and 99% explained variance. Differentiation between the genera
Oxysarcodexia, Peckia and Ravinia was efficiently confirmed by both techniques. In comparison with other
biological methods, this approach represents an effective choice for fast and non-destructive identification in
forensic entomology.

1. Introduction environments (Barbosa et al., 2017).

Taxonomical identification of Sarcophagidae is hindered by their
Accurate identification of necrophagous insect species is a major similar morphology − which relies mostly on laborious examination of
obstacle for strengthening forensic entomology worldwide. This is best male genitalia − and inadequate documentation of their thermal bio-
illustrated in the case of flesh flies (Diptera: Sarcophagidae), a taxon logical histories (Silva and Mello-Patiu, 2010). Since cadavers pre-
that has been frequently found on cadavers resulting from homicides dominantly attract females for feeding and larviposition, identification
and natural death (Oliveira and Vasconcelos, 2010). Sarcophagidae is a is left incomplete in most forensic cases. To overcome such obstacles,
highly diverse family with over 3200 described species (Pape et al., DNA-based techniques for barcoding (Madeira et al., 2016) and ana-
2011), of which some have medical and veterinary relevance as causal lytical methods (e.g., characterization of cuticular hydrocarbons)
myiasis agents and as vectors of microorganisms which are pathogenic (Braga et al., 2013) based on gas and liquid chromatography-mass
to humans and other animals (Greenberg, 1971). Recently, sarcopha- spectrometry (Carvalho et al., 2001; Gosselin et al., 2010) have been
gids have also been used as indicators of environmental quality due to developed.
their fast response to habitat modification, such as deforestation in the Although these techniques present high sensitivity, they have sev-
Amazon (De Sousa et al., 2014) and anthropogenic action in coastal eral drawbacks in the analysis of insect tissues such as being an invasive

⁎
Corresponding author at: Institute of Chemistry, Biological Chemistry and Chemometrics, UFRN, Natal, RN, 59.072-970, Brazil.
E-mail address: [email protected] (K.M.G. Lima).
1
These authors contributed equally to this work.

https://doi.org/10.1016/j.actatropica.2018.04.025
Received 22 February 2018; Received in revised form 14 April 2018; Accepted 22 April 2018
Available online 24 April 2018
0001-706X/ © 2018 Elsevier B.V. All rights reserved.
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

and destructive procedure and the use of bulky instrumentation that Table 1a
impairs in-field monitoring. A rapid, inexpensive and non-destructive Number of training, validation and prediction specimens in each category for
method for species identification with the potential for high throughput three classes classification.
would thus be desirable as an alternative to morphology-based Category Set training Set Validation Set Prediction
methods. However, most techniques require sophisticated equipment
and expensive reagents, and frequently demand a large number of in- PC 14 3 3
PS 14 3 3
dividuals − a major hindrance given the fact that sarcophagids are
RF 14 3 3
usually found in high species richness and low abundance in nature
(Sousa et al., 2014).
In this scenario, vibrational spectroscopies such as near infra-red Table 1b
(NIR) and attenuated total reflection Fourier-transform infrared (FTIR) Number of training, validation and prediction specimens in each category for
are label-free, rapid, non-destructive techniques that are cost effective all classes classification.
and require little sample preparation. They can be used to determine
Category Set training Set Validation Set Prediction
the insect metabolic identity (lipids, proteins, cellular processes) and to
differentiate between species based on their absorbance characteristics OTI 14 3 3
because the cuticle of each species may have a unique chemical com- OT 14 3 3
PC 14 3 3
position (Lima et al., 2014). Thus, it can be as specific as barcoding,
PS 14 3 3
without the need for time-consuming and expensive DNA extraction RF 14 3 3
and analysis techniques. Because absorption is determined by the in- TA 14 3 3
ternal and external biochemical composition of the organism, a species
will have a “fingerprint” based on its particular absorption spectrum
(Rodríguez-Fernández et al., 2011). sample preparation, spectroscopic measurement, data preprocessing,
NIR applications in entomology have varied so far. It has been feature selection and analytical validation were developed. Specifically,
useful to determine the age and species of Anopheles gambiae sensu lato we aimed to: i) test the validity of NIR and ATR-FTIR for identifying
complex (Sikulu et al., 2010), to detect flunitrazepan in larvae, puparia flesh fly species native to Neotropical regions; ii) compare the effec-
and adults of necrophagous blow flies (Chrysomya megacephala, Chry- tiveness of the two methods for taxonomical purposes; and iii) infer on
somya albiceps and Cochliomyia macellaria) (Oliveira et al., 2014; Baia the applicability of NIR as an accessible tool for species identification in
et al., 2016), to discriminate live individuals of two Drosophila species comparison to morphological methods. To our knowledge, this is the
(D. subobscura and D. obscura) (Fischnaller et al., 2012), to identify first application of PCA-LDA, SPA-LDA and GA-LDA to differentiate
stored grain beetles (Jia et al., 2007), to differentiate species of Lepi- insect samples based on spectral data.
doptera (Dowell et al., 2005), and to determine gender in fly pupae
(Dowell et al., 1999). 2. Material and methods
However, this technique generates several hundreds or even thou-
sands of variables that exist in the near infra-red/infra-red spectra. In 2.1. Insects
addition, redundancy and collinearity are widespread phenomena
among these variables, since they contain interference coming from Specimens used in this study were collected from different locations
background, noise and overlapping bands, challenging a high-quality in Pernambuco State, Northeastern Brazil, between October/2012 and
calibration model for unknown sample prediction. Therefore, the use of August/2013. Collecting was done using traps baited with decomposing
appropriate chemometrics tools for multivariate calibration and clas- chicken liver and fish, previously exposed for 48 h at 24 °C. The field-
sification is largely responsible for advancing spectroscopic techniques. caught flies were killed with ethyl acetate for morphology-based species
These include partial least squares (PLS) (Dupuy et al., 2010), principal identification. Ten male specimens of each species were identified −
component regression (PCR) (Xie and Kalivas, 1997) artificial neural Oxysarcodexia timida (OTI), Oxysarcodexia thornax (OT), Peckia chry-
networks (ANN) (Makino et al., 2010) and least squares-vector support sostoma (PC), Peckia lambens (PS), Ravinia belforti (RF) and Tricharaea
machine (LS-SVM) (Shao et al., 2012). Further, there is still the prin- occidua (TA) and 40 unidentified female individuals were selected for
cipal component analysis (PCA) for initial data reduction (Marques the analyses. These species were chosen due to their forensic and/or
et al., 2013), hierarchical cluster analysis (HCA) to analyze groups in a medical relevance and ubiquitous presence in several environments in
set of data on the basis of spectral similarities (Martin et al., 2011) and the Neotropical Region (Vasconcelos et al., 2015). Prior to the analysis,
linear discriminant analysis (LDA) to classify unknown samples into specimens were removed from the alcohol and left to dry on absorbent
predetermined groups (Cheung et al., 2011). Finally, a well-succeeded tissue paper for at least 10 min, to allow for alcohol evaporation.
approach to overcome problems with redundancy or collinearity is the
successive projections algorithm (SPA) (Pontes et al., 2005) in con- 2.2. NIR spectroscopy
junction with linear discriminant analysis (LDA) and genetic algorithm
(GA) (Tapp et al., 2003). NIR spectral [n = 60, 20 Peckia chrysostoma, 20 Peckia lambens, 20
The choice and development of the multivariate classification ap- Ravinia belforti] measurements were performed using an Antaris MX FT-
proaches ensure reliable insect identification using NIR/IR spectro- NIR spectrophotometer (Thermo Fisher Scientific Inc., USA) equipped
scopy. For instance, multivariate classification quality features such as with a transflectance fiber optic probe. The NIR spectra were obtained
sensitivity, specificity, positive and negative predictive values, Youden over a range of 10,000–4000 cm−1, or 1000–2500 nm, and were re-
index, and positive and negative likelihood ratios should be calculated corded with a spectral resolution of 32 cm−1, with 32 co-added scans.
to ensure the validity of the results in accordance with International The time measurement was 26 s (32 scans) per spectrum. The trans-
guidelines (Costa et al., 2016). flectance probe was washed with ethanol (70% v/v) and dried using
Herein, we have evaluated a simple, non-destructive and rapid tissue paper after each sample. Cleanliness of the transflectance probe
method based on two vibrational spectroscopy techniques (NIR and was verified by collecting an absorbance spectrum of the probe using
ATR-FTIR) coupled with variable selection techniques such as genetic the most recently collected background as a reference. Spectral mea-
algorithm-linear discriminant analysis (GA-LDA) and successive pro- surements were done in an acclimatized room under controlled tem-
jection algorithm-linear discriminant analysis (SPA-LDA) for the iden- perature of 22 °C and 60% relative air humidity. Samples were allowed
tification and discrimination of Sarcophagidae species. In our study, to equilibrate to this temperature before the analysis.

2
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Fig. 1. (a) Average spectra acquired from three categories. The spectra from Peckia chrysostoma (PC) are shown in black; those with Peckia lambens (PS) are shown
in green; those with Ravinia belforti (RF) are shown in pink. (b) Pre-processing spectra from three categories. The spectra from PC are shown in black; those with PS
are shown in green; those with RF are shown in pink. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of
this article.)

2.3. ATR-FTIR spectroscopy the projection. The variables created through LDA (factors) are linear
combinations of the wavenumber-absorbance intensity values (Martin
IR spectra [n = 160, 60 male (2 spectra), 40 female] were collected et al., 2007). Thus, the use of LDA for identification or classification of
from each individual insect using the Bruker Lumus FTIR spectrometer spectral data generally requires appropriate variable selection proce-
with motorized ATR crystal (Bruker Optics Ltd, Coventry, U.K.). Prior to dures (Silva et al., 2013). In the present study, the PCA, SPA and GA
analyzing each specimen, the diamond crystal within the spectrometer were adapted for this function. In the PCA-LDA, SPA-LDA and GA-LDA
was washed and a background spectrum was obtained to account for models, the validation set was used to guide the variable selection, a
atmospheric composition. strategy to avoid overfitting. The optimum number of variables for SPA-
LDA and GA-LDA was determined from the minimum of the cost
function G calculated for a given validation data set as:
2.4. Chemometrics methods: PCA-LDA, SPA-LDA, GA-LDA and PCA
NV
LDA is a supervised linear transformation that projects the variables 1
G=
NV
∑ gn
(wavenumbers, for example) into a variable-reduced space, which is n=1
optimal for discrimination between treatment classes. An LDA seeks for (1)
a projection matrix such that Fisher criterion (i.e. the ratio of the be-
where gn is defined as
tween-variance scatter to the within-class variance) is maximized after

3
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Fig. 2. (a) Variables selected by SPA-LDA in PC, PS and RF classification. (b) Variables selected by GA-LDA in PC, PS and RF classification. (c) Variables selected by
SPA-LDA in OTI, OT, PC, PS, RF and TA classification. (d) Variables selected by SPA-LDA in Oxysarcodexia timida (OTI), Oxysarcodexia thornax (OT), Peckia
chrysostoma (PC), Peckia lambens (PS), Ravinia belforti (RF) and Tricharaea occidua (TA) classification.

r 2 (x n , mI (n) ) cutting 1000–1800 nm, and a Savitzky-Golay derivative (first order, 15

gn =
minI (m) ≠ l (n) r 2 (x n , mI (m) ) (2) points) was applied. ATR-FTIR raw spectra were pre-processed cutting
900–1600 cm−1 in the biological fingerprint region; baseline and
th
where I(n) is the index of the true class for the n validation object xn. Savitzky-Golay smoothing filter (15 points) were applied. Mean cen-
In the GA-LDA model, the mutation and reproduction probabilities tering was applied to all spectra before performing variable subset se-
were kept constant at 10 and 80%, respectively. The initial population lection and calibration. For PCA-LDA, SPA-LDA and GA-LDA models,
was 120 individuals, with 60 generations. The best solution resulting each class was treated separately. The samples were divided into
from the three GA realizations was kept. training, validation and test sets by applying the classic Kennard-Stone
For this study, LDA scores, loading and discriminant function (DF) (KS) uniform sampling algorithm (Kennard and Stone, 1969) to the
values were derived for the biochemical-insect fingerprint region. The NIR/IR spectra. The number of samples in each set is presented in
first LDA factor (LD1) was used to visualize the alterations of the insect Tables 1a and 1b, respectively.
sample in 1-dimensional (D) scores plots that represented the main
chemical alterations. 3. Results

2.5. Software 3.1. Discrimination of species using NIR (males)

The import and pre-treatment data, as well as the chemometric For the NIR method, the raw and pre-treated spectra can be visua-
model constructions were implemented into MATLAB R2014a software lized (Fig. 1a, b). When considering all classes, the discriminant func-
(Mathworks Inc, Natick, MA, USA). NIR raw spectra were pre-processed tions did not present a very clear segregation for the PCA-LDA, SPA-

4
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Fig. 2. (continued)

LDA and GA-LDA, respectively. When the classes were reduced to the LDA (Fig. 5b and c). Regardless of the algorithm used, it was observed
most discriminating, the number of informational variables decreased that the results were similar, with a clear separation of the species
from 78 to 39 (SPA-LDA), and from 48 to 24 in the GA-LDA method, Tricharaea occidua (TA) and Ravinia belforti (RF) being clearly set in
where both decreases were statistically significant (χ2Yates = 13; specific clusters in relation to the other clusters, which allowed for
P = .0004, and χ2Yates = 8, P = .0067, respectively) and can be visua- distinct genera: a) a group of the genus Peckia [P. chrysostoma (PC) near
lized in Fig. 2a and b. The discriminant function showed clearer seg- P. lambens (PS)] and b) a group of the genus Oxysarcodexia [O. timida
regations for the species Peckia chrysostoma (PC), Peckia lambens (PS) (OTI) approach to O. thornax (OT)].
and Ravinia belforti (RF) (Fig. 3a–c).
3.3. Identification of females from male profiles
3.2. Species segregation (Males) using FT-IR
The set of raw spectra and cut spectra in the biological fingerprint
The raw and preprocessed spectra for the FT-IR method can be vi- region for male specimens of the [O. thornax (OT) and P. chrysostoma
sualized in Fig. 4a and b, respectively. The biological fingerprint region (PC)] and unidentified females are available in Fig. 6a and b. The best
was used for this method for elaborating the classification models in separation between males and females using the PCA technique was
applying the same PCA-LDA, SPA-LDA and GA-LDA algorithms. Despite achieved using three main components (PCs), which together re-
showing a tendency to segregate between classes, PCA-LDA formed presented more than 99.5% of the total data variance. That is, most of
only one group (Fig. 5a). In the case of SPA-LDA, the number of vari- the significant values were reached to discriminate both sex and species
ables selected was similar to that of the NIR method (78 variables) among Sarcophagidae adults.
(Fig. 2c), whereas the number was lower in GA-LDA (31 variables) Based on the analysis of the components, it is observed that PC1 and
(Fig. 2d). In addition, FT-IR was more sensitive in segregating pre-de- PC2 were efficient in allowing segregation between sex and species, so
fined classes, with clearer group formation in both SPA-LDA and GA- that clusters above the PC2 axis mean different species, while the right

5
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Fig. 3. (a) DF1x DF2 discriminant function values calculated by using principal component analysis (PCA) – linear discriminant analysis (LDA) results from three
categories. (b) DF1x DF2 discriminant function values calculated by using the variables selected by successive projection algorithm (SPA) – linear discriminant
analysis (LDA) results from three categories. (c) DF1x DF2 discriminant function values calculated by using the variables selected by genetic algorithm (GA) – linear
discriminant analysis (LDA) results from three categories.

6
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Fig. 4. (a) Average spectra acquired from all categories. The spectra from Oxysarcodexia timida (OTI) are shown in blue; those with Oxysarcodexia thornax (OT) are
shown in red; those with Peckia chrysostoma (PC) are shown in cyan; those with Peckia lambens (PS) are shown in green; those with Ravinia belforti (RF) are shown
in purple; those with Tricharaea occidua (TA) are shown in black. (b) Pre-processing spectra from all categories. The spectra from OTI are shown in blue; those with
OT are shown in red; those with PC are shown in cyan; those with PS are shown in green; those with RF are shown in purple; those with TA are shown in black. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

PC1 clusters correspond to the male two species, P. chrysostoma and O. and 2b, it is possible to see that these rates varied according to the
thornax (Fig. 7a). The PC2 components confirm the segregation be- model used; meaning PCA-LDA, SPA-LDA and GA-LDA. Furthermore, as
tween clusters corresponding to the species, also pointing out the ex- shown in Table 2b, the specificity for all categories suggests that PCA-
istence of unidentified specimens of the genus Peckia which were con- LDA as well as GA-LDA presented improved accuracy in comparison
figured as outliers. PC3 clearly demonstrates the separation between with SPA-LDA.
male and female specimens (Fig. 7b). That is, three components are PCA-LDA still achieved sensitivity and specificity scores of 100% for
sufficient to identify female specimens from the comparison of the all the species categories, thus showing that the species can be rela-
profiles obtained for males of P. chrysostoma and O. thornax. tively well classified by these methods. However, in general the other
tested methods also showed high sensitivity and specificity (Table 2b).
These results still show that NIR and ATR-FTIR microspectroscopy in
3.4. Performance of methods conjunction with powerful chemometric approaches has the potential
to identify and differentiate species of necrophagous flies captured in a
Classification rates were determined by using the best models. corpse.
Tables 2a and 2b present the performance features results for the op-
timized models (PCA-LDA, SPA-LDA and GA-LDA) of each category.
According to the sensitivity and specificity results shown in Tables 2a

7
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Fig. 5. (a) DF1x DF2 discriminant function values calculated by using principal component analysis (PCA) – linear discriminant analysis (LDA) results from all
categories. (b) DF1x DF2 discriminant function values calculated by using the variables selected by successive projection algorithm (SPA) – linear discriminant
analysis (LDA) results from all categories. (c) DF1x DF2 discriminant function values calculated by using the variables selected by genetic algorithm (GA) – linear
discriminant analysis (LDA) results from all categories.

8
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Fig. 6. (a) Average spectra acquired from three categories. The spectra from Oxysarcodexia thornax (OT) are shown in red; those with Peckia chrysostoma (PC) are
shown in green; those with Female are shown in gray. (b) Pre-processing spectra from three categories. The spectra from OT are shown in red; those with PC are
shown in green; those with Female are shown in gray. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of
this article.)

4. Discussion identification of plants, bacteria, fungi and insects (Rodríguez-

Fernández et al., 2011; Luginbühl et al., 2006; Morais et al., 2017).
The need for rapid identification of insect species requires com- Although biospectroscopy has been revealed as a promising tool in
plementary tools for morphology-based taxonomical keys. In this study, the discrimination and identification of Muscidae species (Rodríguez-
we have demonstrated that NIR and/or FT-IR associated with proper Fernández et al., 2011), and corroborated for Sarcophagidae in this
techniques for variable selection (PCA-LDA, SPA-LDA and GA-LDA) study, its use in forensic entomology is embryonic. NIR applications
enable the segregation of male specimens of flesh flies (Diptera: include the detection of insect fragments in stored products (Jia et al.,
Sarcophagidae) at a specific level based on biological spectral “finger- 2007) and of illicit substances in necrophagous species (Baia et al.,
prints” derived from the biochemical composition of their cuticles. 2016), as well as rapidly discriminating and determining the age of
These tools can help to elucidate divergences in species identification immature blow fly stages (Pickering et al., 2015). Additionally, the
rapidly and accurately without the need for previous treatment of the geographical origin of the fungus Ganoderma lucidum was determined
biological material (Rodríguez-Fernández et al., 2011; Kim et al., by NIR (Chen et al., 2008), which opens possibilities for its use in drug
2004). Accordingly, the efficiency of infrared spectroscopy has been tracking investigations.
demonstrated in diverse areas of knowledge such as medicine In the case of Sarcophagidae flies, ATR-FTIR proved to be more
(Theophilou et al., 2016), quality control of commercial products sensitive to the spectra of all species studied, regardless of the algorithm
(Costa et al., 2015), the design of landscape ecology theoretical models tested (PCA-LDA, SPA-LDA and GA-LDA). This may be associated with
(Youngentob et al., 2012), vector control (Sikulu et al., 2010) and the the fact that FT-IR spectroscopy is more efficient in discriminating

9
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Fig. 7. (a) Biplot graph of PC1xPC2 scores calculated by using principal component analysis (PCA); (b) Biplot graph of PC2xPC3 scores calculated by using principal
component analysis.

species and the varieties of this taxon (Kim et al., 2004) because the scarcity of bionomical and behavioural studies limits the potential of
technique is based on vibrations correlated to functional groups present Sarcophagidae species in estimating the PMI (Vairo et al., 2014), ac-
in biomolecules present on the exoskeleton, such as carbohydrates, curate identification of the target species of this study is of utmost
proteins, cuticular lipids, DNA and RNA, whereas in the case of FT-NIR importance. From a medical standpoint, identification of larvae and
overtones of these functional groups may overlap, making it difficult to adults is crucial for developing the monitoring and control of flesh fly
visualize possible spectral differences. On the other hand, the apparatus species that act as vectors of pathogens and causal myiasis agents
needed for the use of NIR is more easily manipulated in the field when (Greenberg, 1971; Vairo et al., 2014).
compared to the equipment used in the FT-IR analysis. On the other In this study, it was possible to identify females at a specific level
hand, it can be argued that both FT-NIR and FT-IR can be used to based on the spectral signatures of adult males of the same species,
discriminate closely-related species (Maree and Viljoen, 2011). This being able to explain 99.5% of the variance between classes, and re-
was demonstrated in a study that applied NIR for the identification of quiring only three PCs. This contribution expands the usefulness of
species and subspecies of Zootermopsis (Isoptera: Termopsidae), with females in practical cases of criminal investigations, since ephemeral
over 95% and 80% precision, respectively (Aldrich et al., 2007). substrates (carcasses, cadavers) are mostly visited by females for
In this study, NIR and ATR-FTIR were successful in discriminating feeding and immature deposition; the female-driven sex ratio fre-
the genera and species of sarcophagids, such as Peckia chrysostoma and quently exceeds 90% in field studies (Barbosa et al., 2017). Tax-
Ravinia belforti, previously registered on cadavers (Oliveira and onomical keys currently available prioritize morphological characters
Vasconcelos, 2010; Vasconcelos et al., 2014). Given the fact that the of the male (e.g. shape of the aedeagus), so that tools based on female

10
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

Table 2a and FT-IR were calculated and can be visualized in Tables 2a and 2b,
Values of quality performance features from three classification methods (PCA- respectively. FT-NIR spectroscopy showed sensitivity and specificity
LDA, SPA-LDA and GA-LDA) by FTIR microspectroscopy for three category. close to the values predicted by the morphological methodology,
PC PS RF around 60% accuracy for the PS class and superior for the other classes.
On the other hand, FT-IR spectroscopy presented values of sensitivity
PCA-LDA sens 66.67 0 66.67 and specificity superior to the morphological methodology, obtaining
spec 66.67 0 66.67
100% correctness. In addition, this technique has other positive points
ppv 66.67 0 66.67
npv 66.67 0 66.67 because it does not require sample preparation, it does not generate
you 33.33 −100 33.33 residue, and is fast, non-destructive and low cost (Rodríguez-Fernández
LR(+) 2 0 2 et al., 2011).
LR(−) 0.5 0 0.5
The results show that the high efficiency of FT-NIR, revealed in the
SPA-LDA sens 66.67 66.67 100 high potential for identifying females, can be extrapolated to dis-
spec 66.67 66.67 100 criminate the young stages (larvae and pupae) of the species of re-
ppv 66.67 66.67 100
npv 66.67 66.67 100
cognized medical and forensic potential, as already seen for the
you 33.33 33.33 100 Calliphoridae family (Pickering et al., 2015). From the forensic science
LR(+) 2 2 0 perspective, our study shows that the taxonomy tool resulting from
LR(−) 0.5 0.5 0 infrared spectroscopy (FT-NIR and FT-IR) acts as a starting point for
GA-LDA sens 100 66.67 66.67 developing a spectral database library covering different necrophagous
spec 100 66.67 33.33 species, allowing for practical use by forensic investigators. The busy
ppv 100 66.67 50
routine of a criminal expert in cities exposed to high homicide rates, for
npv 100 66.67 50
you 100 33.33 0
example, in Northeast Brazil, makes it impossible to collect, create and
LR(+) 0 2 1 identify necrophagous specimens under controlled conditions, which is
LR(−) 0 0.5 1 a necessary procedure to obtain reliable entomological evidence. Co-
operation with academic institutions would amplify the realism of in-
vestigations by elucidating the identity of the entomological agent in-
Table 2b volved. However, it is extremely important to substantiate the
Values of quality performance features from three classification methods (PCA- identification through reliable methods in order to guarantee the
LDA, SPA-LDA and GA-LDA) by FTIR microspectroscopy for all category.
quality of the spectral bank.
OTI OT PC PS RF TA
Acknowledgments
PCA-LDA sens 100 100 100 100 100 100
spec 100 100 100 100 100 100
ppv 100 100 100 100 100 100 We thank Coordenaçao de Aperfeiçoamento de Pessoal de Nível
npv 100 100 100 100 100 100 Superior (CAPES) and Conselho Nacional de Desenvolvimento
you 100 100 100 100 100 100 Científico e Tecnológico (CNPq) for financial support, the team of the
LR(+) 0 0 0 0 0 0
Insects of Forensic Importance Research Group for their invaluable help
LR(−) 0 0 0 0 0 0
with insect collection, and the Instituto Chico Mendes de Conservação
SPA-LDA sens 66.66 33.33 66.66 33.33 100 100
da Biodiversidade (ICMBio) for sampling authorization. K.M.G. Lima
spec 66.66 33.33 66.66 33.33 100 100
ppv 66.66 33.33 66.66 33.33 100 100
acknowledges the CNPq/Capes project (Grant 070/2012 and 442087/
npv 66.66 33.33 66.66 33.33 100 100 2014-4) for financial support.
you 33.33 −33.33 33.33 −33.33 100 100
LR(+) 2 0.5 2 0.5 0 0 References
LR(−) 0.5 2 0.5 2 0 0

GA-LDA sens 100 100 33.33 100 100 100 Aldrich, B.T., Maghirang, E.B., Dowell, F.E., Kambhampati, S., 2007. Identification of
spec 100 100 33.33 100 100 100 termite species and subspecies of the genus Zootermopsis using near-infrared re-
ppv 100 100 33.33 100 100 100 flectance spectroscopy. J. Insect Sci. 7, 18.
npv 100 100 33.33 100 100 100 Baia, T.C., Gama, R.A., Lima, L.A.S., Lima, K.M.G., 2016. FTIR microspectroscopy coupled
you 100 100 −33.33 100 100 100 with variable selection methods for the identification of flunitrazepam in necropha-
LR(+) 0 0 0.5 0 0 0 gous flies. Anal. Method 8, 968–972.
LR(−) 0 0 2 0 0 0 Barbosa, T.M., Carmo, R.F.R., Silva, L.P., Sales, R.G., Vasconcelos, S.D., 2017. Diversity of
sarcosaprophagous calyptratae (diptera) on sandy beaches exposed to increasing le-
vels of urbanization in Brazil. Environ. Entomol. 46, 460–469.
Bickford, D., Lohman, D.J., Sodhi, N.S., Ng, P.K.L., Meier, R., Winker, K., Ingram, K.K.,
specimen morphology are still scarce and limited to a few species (Vairo Das, I., 2007. Cryptic species as a window on diversity and conservation. Trends Ecol.
et al., 2014). Evol. 22, 148–155.
Braga, M.V., Pinto, Z.T., de Carvalho Queiroz, M.M., Matsumoto, N., Blomquist, G.J.,
In the last decades, several techniques have been applied to identify
2013. Cuticular hydrocarbons as a tool for the identification of insect species: pu-
Diptera species of forensic importance such as molecular identification parial cases from Sarcophagidae. Acta Trop. 128, 479–485.
(Madeira et al., 2016), due to their high specificity and sensitivity, and Carvalho, L.M., Linhares, a X., Trigo, J.R., 2001. Determination of drug levels and the
effect of diazepam on the growth of necrophagous flies of forensic importance in
not having the need to wait until adult emergence, using larvae
southeastern Brazil. Forensic Sci. Int. 120, 140–144.
(GilArriortua et al., 2015) and empty puparia (Bickford et al., 2007). Chen, Y., Xie, M.Y., Yan, Y., Zhu, S., Bin Nie, S.P., Li, C., Wang, Y.X., Gong, X.F., 2008.
However, molecular technique applications (DNA) have disadvantages Discrimination of Ganoderma lucidum according to geographical origin with near
such as high cost, the need for fresh material that motivates collection infrared diffuse reflectance spectroscopy and pattern recognition techniques. Anal.
Chim. Acta 618, 121–130.
of new specimens (Bickford et al., 2007), in addition to being a de- Cheung, K.T., Trevisan, J., Kelly, J.G., Ashton, K.M., Stringfellow, H.F., Taylor, S.E.,
structive technique that consumes considerable effort and time. Singh, M.N., Martin-Hirsch, P.L., Martin, F.L., 2011. Fourier-transform infrared
On the other hand, NIR and ATR-FTIR associated with variable se- spectroscopy discriminates a spectral signature of endometriosis independent of
inter-individual variation. Analyst 136, 2047–2055.
lection techniques (GA-LDA and SPA-LDA) proved to be an efficient Costa, F.S.L., Pedroza, R.H.P., Porto, D.L., Lima, K.M.G., 2015. Multivariate Control
alternative in identification, since it presented high specificity and Charts for Simultaneous Quality Monitoring of Isoniazid and Rifampicin in a
sensitivity both in sex segregation (male and female), as well as the Pharmaceutical Formulation Using a Portable Near Infrared Spectrometer, vol. 26.
pp. 64–73.
taxonomic level of genus and species. The merit figures for both FT-NIR Costa, F.S.L., Silva, P.P., Morais, C.L.M., Arantes, T.D., Lima, M.G., 2016. Attenuated total

11
T.M. Barbosa et al. Acta Tropica 185 (2018) 1–12

reflection Fourier transform- infrared (ATR-FTIR) spectroscopy as a new technology resolution-alternating least squares. Analyst 136, 4950–4959.
for discrimination between Cryptococcus neoformans and Cryptococcus gattii. Anal. Morais, C.L.M., Costa, F.S.L., Lima, K.M.G., 2017. Variable selection with a support vector
Methods 8, 7107–7115. machine for discriminating Cryptococcus fungal species based on ATR-FTIR spec-
De Sousa, J.R.P., Esposito, M.C., Carvalho Filho, F.D.S., Juen, L., 2014. The potential uses troscopy. Anal. Methods 9, 2964–2970.
of sarcosaprophagous flesh flies and blowflies for the evaluation of the regeneration Oliveira, T.C., Vasconcelos, S.D., 2010. Insects (Diptera) associated with cadavers at the
and conservation of forest clearings: a case study in the Amazon forest. J. Insect Sci. Institute of Legal Medicine in Pernambuco, Brazil: implications for forensic en-
14, 1–5. tomology. Forensic Sci. Int. 198, 97–102.
Dowell, F.E., Throne, J.E., Wang, D., Baker, J.E., 1999. Identifying stored-grain insects Oliveira, J.S., Baia, T.C., Gama, R.A., Lima, K.M.G., 2014. Development of a novel non-
using near-infrared spectroscopy. J. Econ. Entomol. 92, 165–169. destructive method based on spectral fingerprint for determination of abused drug in
Dowell, F.E., Parker, A.G., Benedict, M.Q., Robinson, A.S., Broce, A.B., Wirtz, R.A., 2005. insects: an alternative entomotoxicology approach. Microchem. J. 115, 39–46.
Sex separation of tsetse fly pupae using near-infrared spectroscopy. Bull. Entomol. Pape, T., Blagoderov, V., Mostovski, M., 2011. Order Diptera Linnaeus, 1758. In Z. Q.
Res. 95, 249–257. Zhang (Ed.), Animal biodiversity: an outline of higher-level classification and survey
Dupuy, N., Galtier, O., Ollivier, D., Vanloot, P., Artaud, J., 2010. Comparison between of taxonomic richness. Zootaxa 3148, 222–229.
NIR, MIR, concatenated NIR and MIR analysis and hierarchical PLS model. Pickering, C.L., Hands, J.R., Fullwood, L.M., Smith, J.A., Baker, M.J., 2015. Rapid dis-
Application to virgin olive oil analysis. Anal. Chim. Acta 666, 23–31. crimination of maggots utilising ATR-FTIR spectroscopy. Forensic Sci. Int. 249,
Fischnaller, S., Dowell, F.E., Lusser, A., Schlick-Steiner, B.C., Steiner, F.M., 2012. Non- 189–196.
destructive species identification of Drosophila obscura and D: subobscura (Diptera) Pontes, M.J.C., Galvão, R.K.H., Araújo, M.C.U., Moreira, P.N.T., Neto, O.D.P., José, G.E.,
using near-infrared spectroscopy. Fly (Austin) 6, 284–289. Saldanha, T.C.B., 2005. The successive projections algorithm for spectral variable
GilArriortua, M., Saloña Bordas, M.I., de Pancorbo, M.M., 2015. Uso de vestigios mole- selection in classification problems. Chemom. Intell. Lab. Syst. 78, 11–18.
culares en entomología forense. Cienc. Forense 12, 35–78. Rodríguez-Fernández, J.I., Carvalho, C.J.B., Pasquini, C., Lima, K.M.G., Moura, M.O.,
Gosselin, M., Fernandez, M. del M.R., Wille, S.M.R., Samyn, N., Boeck, G., De Bourel, B., Arízaga, G.G.C., 2011. Barcoding without DNA? Species identification using near
2010. Quantification of methadone and its metabolite third instar larvae of Lucilia infrared spectroscopy. Zootaxa 2933, 1–9.
sericata (Diptera: Calliphoridae) using liquid chromatography–tandem mass spec- Shao, Y., Zhao, C., Bao, Y., He, Y., 2012. Quantification of nitrogen status in rice by least
trometry. J. Anal. Toxicol. 34, 374–380. squares support vector machines and reflectance spectroscopy. Food Bioprocess
Greenberg, B., 1971. Ecology, classification and biotic associations. Flies and Disease. Technol. 5, 100–107.
Princeton University Press, Princeton. Sikulu, M., Killeen, G.F., Hugo, L.E., Ryan, P.A., Dowell, K.M., Wirtz, R.A., Moore, S.J.,
Jia, F., Maghirang, E., Dowell, F., Abel, C., Ramaswamy, S., 2007. Differentiating tobacco Dowell, F.E., 2010. Near-infrared spectroscopy as a complementary age grading and
budworm and corn earworm using near-infrared spectroscopy. J. Econ. Entomol. 100, species identification tool for African malaria vectors. Parasite Vectors 3, 49.
759–764. Silva, K.P., Mello-Patiu, C.A., 2010. New species of Dexosarcophaga Townsend from
Kennard, R.W., Stone, L.A., 1969. Computer aided design of experiments. Technometrics Panama with an illustrated key to species of the subgenus Bezzisca (Diptera:
11, 137–148. Sarcophagidae). J. Nat. Hist 44, 89–106.
Kim, S.W., Ban, S.H., Chung, H., Cho, S., Chung, H.J., Choi, P.S., Yoo, O.J., Liu, J.R., Silva, C.S., Borba, F.D.S.L., Pimentel, M.F., Pontes, M.J.C., Honorato, R.S., Pasquini, C.,
2004. Taxonomic discrimination of flowering plants by multivariate analysis of 2013. Classification of blue pen ink using infrared spectroscopy and linear dis-
Fourier transform infrared spectroscopy data. Plant Cell Rep. 23, 246–250. criminant analysis. Microchem. J. 109, 122–127.
Lima, L.A.S., Baia, T.C., Gama, R.A., Lima, K.M.G., 2014. Near Infrared Spectroscopy as Sousa, J., Esposito, M., Filho, F., Juen, L., 2014. The potential uses of sarcosaprophagous
an Emerging Tool for Forensic Entomotoxicology, vol. 25. pp. 5–7. flesh flies and blowflies for the evaluation of the regeneration and conservation of
Luginbühl, W., Jimeno, J., Zehntner, U., 2006. Identification of seven species of the forest clearings: a case study in the amazon forest. J. Insect Sci. 14, 1–5.
Lactobacillus acidophilus group by FT-IR spectroscopy. LWT – Food Sci. Technol. 39, Tapp, H.S., Defernez, M., Kemsley, E.K., 2003. FTIR spectroscopy and multivariate ana-
152–158. lysis can distinguish the geographic origin of extra virgin olive oils. J. Agric. Food
Madeira, T., Souza, C.M., Cordeiro, J., Thyssen, P.J., 2016. The use of DNA barcode for Chem. 51, 6110–6115.
identifying species of Oxysarcodexia Townsend (Diptera: Sarcophagidae): a pre- Theophilou, G., Lima, K.M.G., Martin-Hirsch, P.L., Stringfellow, H.F., Martin, F.L., 2016.
liminary survey. Acta Trop. 161, 73–78. ATR-FTIR spectroscopy coupled with chemometric analysis discriminates normal,
Makino, Y., Ichimura, M., Oshita, S., Kawagoe, Y., Yamanaka, H., 2010. Estimation of borderline and malignant ovarian tissue: classifying subtypes of human cancer.
oxygen uptake rate of tomato (Lycopersicon esculentum Mill.) fruits by artificial Analyst 141, 585–594.
neural networks modelled using near-infrared spectral absorbance and fruit mass. Vairo, K.P.E., Ururahy-Rodrigues, A., Osvaldo Moura, M., Antunes De Mello-Patiu, C.,
Food Chem. 121, 533–539. 2014. Sarcophagidae (Diptera) with forensic potential in Amazonas: a pictorial key.
Maree, J.E., Viljoen, A.M., 2011. Fourier transform near- and mid-infrared spectroscopy Trop. Zool. 27, 140–152.
can distinguish between the commercially important Pelargonium sidoides and its Vasconcelos, S.D., Soares, T.F., Costa, D.L., 2014. Multiple colonization of a cadaver by
close taxonomic ally P reniforme. Vib. Spectrosc. 55, 146–152. insects in an indoor environment: first record of Fannia trimaculata (Diptera:
Marques, A.S., de Melo, M.C.N., Cidral, T.A., de Lima, K.M.G., 2013. Feature selection Fanniidae) and Peckia (Peckia) chrysostoma (Sarcophagidae) as colonizers of a
strategies for identification of Staphylococcus aureus recovered in blood cultures human corpse. Int. J. Legal Med. 128, 229–233.
using FT-IR spectroscopy successive projections algorithm for variable selection: a Vasconcelos, S.D., Barbosa, T.M., Oliveira, T.P.B., 2015. Diversity of forensically-im-
case study. J. Microbiol. Methods 98, 26–30. portant dipteran species in different environments in northeastern Brazil, with notes
Martin, F.L., German, M.J., Wit, E., Fearn, T., Ragavan, N., Pollock, H.M., 2007. on the attractiveness of animal baits. Florida Entomol. 98, 770–775.
Identifying variables responsible for clustering in discriminant analysis of data from Xie, Y.-L., Kalivas, J.H., 1997. Local prediction models by principal component regres-
infrared microspectroscopy of a biological sample. J. Comput. Biol. 14, 1176–1184. sion. Anal. Chim. Acta 348, 29–38.
Martin, F.L., Llabjani, V., Evans, G., Trevisan, J., Martin-Hirsch, P.L., Patel, I.I., Youngentob, K.N., Renzullo, L.J., Held, A.A., Jia, X., Lindenmayer, D.B., Foley, W.J.,
Stringfellow, H.F., 2011. High contrast images of uterine tissue derived using Raman 2012. Using imaging spectroscopy to estimate integrated measures of foliage nutri-
microspectroscopy with the empty modelling approach of multivariate curve tional quality. Methods Ecol. Evol. 3, 416–426.

12