1.

Introduction

The head and neck region is a complex anatomical area with a great variety of lesions, from inflammatory pathologies to benign and malignant neoplasia. Many of these lesions are infrequently observed. Therefore, pathologists do not have the opportunity of becoming familiar with them.¹ For this reason, a correct diagnosis should not be merely restricted to morphological details investigated through histopathological, immunohistochemical, and radiological techniques.²

Fourier transform infrared imaging (FTIRI), together with conventional biomedical assays,³ is an objective and well-established analytical technique for the unambiguous diagnosis of cancerous and precancerous lesions. By acquiring IR images of biological samples (fluids, cells, and tissues), it is possible to obtain reliable and reproducible information on their biochemical composition. In addition, it is possible to monitor with high accuracy the compositional and structural changes caused by tumoral pathologies by analyzing functional groups, bonding types, and molecular conformations of the most relevant biomolecules.^4–7 Many contributions describe the use of infrared spectroscopy to detect cancerous lesions in a very early stage of the pathology, even before any morphological evidence can be detected.^8,9

In the past decade, we have been studying from a vibrational point of view various epithelial pathologies that may affect the oral compartment.^10–13 In most cases, the IR imaging approach played a key role in detecting early stages of inflammatory, benign, and malignant diseases, helping to gain a deeper understanding of the etiologic nature of different pathologies.

Sinonasal inflammatory polyps (SNPs) are common non-neoplastic lesions of the sinonasal mucosa. They frequently originate from allergic rhinitis and may cause the onset of some biochemical alterations in the epithelial layer, not always easily detectable by routine immunohistochemical assays. They show an intact respiratory epithelium and may present polymorphic stromal cells, which can be confused with malignant ones. The presence of some inflammatory cell aggregations is demonstrated by patches of stromal nonedematous collagen.² Schneiderian papillomas (BSPs) are sinonasal benign neoplasms arising from the Schneiderian mucosa. They are classified into three different varieties: fungiform, oncocytic, and inverted. The latter, considered the most common type, is characterized by a transitional epithelium with squamous, transitional, and columnar cells.^1,2 These papillomas have the tendency to recur, destroying neighboring tissues, and to progress into more aggressive squamous cell carcinomas. For these reasons, they are primarily treated with a rather invasive surgical resection; when necessary, radiation therapy follows and a close follow-up of the patient is recommended. Sinonasal undifferentiated carcinomas (SNUCs) are high-grade malignant epithelial tumors, arising both ex novo and from Schneiderian papillomas. They are characterized by hypercellular proliferation and by the presence of squamous foci with high amounts of keratin.²

The association of Schneiderian papillomas with malignant carcinomas and their tendency to be confused with more common benign lesions, such as SNPs, make it necessary to explore new methods for their reliable diagnosis.¹⁴ To improve the knowledge in this field, we performed FTIRI on different inflammatory and neoplastic sinonasal lesions, such as SNPs, BSPs, and SNUCs. Visible and infrared images of each section were acquired and the biomolecular features of the epithelial and connective components were evaluated, pinpointing specific spectral markers of each pathology.

2.

Experimental Section

2.2.

FTIR Measurements and Data Analysis

FTIR measurements were carried out at the IR beamline SISSI, ELETTRA—Sincrotrone Trieste, using a Bruker VERTEX 70 interferometer coupled with a Hyperion 3000 Vis–IR microscope and equipped with a liquid nitrogen–cooled bidimensional FPA detector (detector area size $64\times 64\text{pixels}$; Bruker Optics GmbH, Germany). Using a $15\times$ condenser/objective, the visible image of each section was achieved, whereas IR maps of specific areas ($\approx 160\times 490\mu {\mathrm{m}}^2$ with a pixel resolution of about $2.56\mu \mathrm{m}$), selected according to the pathologist’s suggestions, were acquired in transmission mode (12,288 spectra, 256 scans, spectral resolution $4{\mathrm{cm}}^{-1}$). Background spectra were acquired on a clean portion of the ${\mathrm{CaF}}_2$ windows. Raw spectra were pretreated by simply applying the atmospheric compensation routine of OPUS 7.1 for atmospheric ${\mathrm{CO}}_2$ and water vapor correction.

To evaluate the spatial distribution of the principal biocomponents (lipids, proteins, and nucleic acids), false color images were calculated by integrating IR maps under the following spectral regions: CH stretching region (3050 to $2800{\mathrm{cm}}^{-1}$), amide I and II bands (1720 to $1480{\mathrm{cm}}^{-1}$), and phosphates and carbohydrates zone (1350 to $900{\mathrm{cm}}^{-1}$). For this purpose, R package HyperSpec was used.¹⁵ Before integration, spectra were vector normalized for avoiding artifacts induced by local thickness variations.

To discriminate the spectra corresponding to epithelial and connective components, for each IR map, unsupervised hierarchical cluster analysis (UHCA) was applied to preprocessed spectra in the 1800 to $900{\mathrm{cm}}^{-1}$ spectral range upon two-points baseline linear fit and vector normalization. The Euclidean distance and Ward’s algorithm were used in order to group spectra according to their similarity. The obtained clusters were localized on the sections using a topographical method, and their position was cross-validated by the histological examination of the corresponding H&E sections. The spectra of all epithelia-belonging pixels were resubmitted to principal component analysis (PCA). The same procedure was followed for the spectra of connective clusters.

For each section, average spectra of the epithelium layer and connective tissue were analyzed in absorbance and second derivative (DII, nine-points smoothing) mode. In addition, to calculate the areas of the component bands, a peak-fitting (Gaussian algorithm) procedure was applied to average absorbance spectra in the following spectral ranges: 3050 to $2800{\mathrm{cm}}^{-1}$ (lipids), 1800 to $1480{\mathrm{cm}}^{-1}$ (proteins), and 1480 to $920{\mathrm{cm}}^{-1}$ (carbohydrates and phosphates). Both the number and center of the subcomponent bands were fixed according to DII results. Mean values of area and wavelength were evaluated for each component band and assigned according to previous studies.¹¹

The whole methodological process of the study is sketched in Fig. 1.

Fig. 1

Methodological process. Each sample was submitted both to (a) histological and (b) FTIR Imaging analysis. On specific areas, the (c) IR maps were acquired, on which (d) false color images, corresponding to the distribution of lipids, proteins, and nucleic acids, were calculated. IR data were then submitted to multivariate analysis, (e) UHCA and (f) PCA, from which (g) absorbance average spectra were achieved both for epithelia and connectives.

For data handling, the following software packages were used: OPUS 7.1 (Bruker Optics GmbH, Germany), HyperSpec, Pirouette 4.5 (Infometrix Inc.), and GRAMS/AI 9.1 (Galactic Industries, Inc., Salem, New Hampshire).

3.

Results and Discussion

All the samples were first submitted to histological examination to detect meaningful structural differences, both in the epithelial layers and in the connective tissues. As an example, in Fig. 2, three sections representative of SNPs, BSPs, and SNUCs were reported. In the section of the SNP, a thin cellular layer of respiratory epithelium (rE) was detected surrounding the whole sample; beneath the epithelial membrane, two zones of loose connective tissue (Co), consisting of amorphous ground substance with loose extracellular fibers and different types of connective tissue cells, were highlighted, together with a large area of inflammatory infiltrate (iI), characterized by an inflammatory granulation tissue. In the section of the BSP, a trE internally growing into the stroma by forming some crypts, was found in the outer limit of the sample; the inner part was composed of loose connective tissue (Co). In the SNUC, two large areas of tumoral epithelium (tE) spaced apart by a small zone of residual Co, not yet infiltrated, were detected.

Fig. 2

Visible images of the following H&E tissue sections: inflammatory sinonasal polyp (SNP), benign inverted Schneiderian papilloma (BSP), and squamous undifferentiated carcinoma (SNUC). Various tissue components were identified: rE, respiratory epithelium; trE, transitional epithelium; tE, tumoral epithelium; Co, connective tissue; and iI, inflammatory infiltrate.

For each section, according to the pathologist’s suggestions, IR chemical images were acquired on specific areas, containing both the epithelial layer and the neighboring connective tissue. In Fig. 3(a), the microphotographs of three acquired areas, representative of SNP, BSP, and SNUC sections, were reported. To localize the distribution of lipids, proteins, and nucleic acids inside the investigated zones, false color images of each section were calculated by integrating under the following spectral regions: 3050 to $2800{\mathrm{cm}}^{-1}$ [lipids, Fig. 3(b)], 1720 to $1480{\mathrm{cm}}^{-1}$ [proteins, Fig. 3(c)], and 1350 to $900{\mathrm{cm}}^{-1}$ [nucleic acids, Fig. 3(d)]. By analyzing false color images, a relatively lower content of lipids and nucleic acids with respect to proteins was evident for all the experimental groups, as expected due to the nature of these tissues. Moreover, for all the sections, a clear colocalization of lipids and proteins was found. The same trend was less evident for the nucleic acids.

Fig. 3

(a) Visible images of SNP, BSP, and SNUC tissue sections. The areas acquired by FTIRI spectroscopy (red squares) included various tissue components: rE, respiratory epithelium; trE, transitional epithelium; tE, tumoral epithelium; and Co, connective tissue. For each section, false color images corresponding to the distribution of lipids, proteins, and nucleic acids were reported, obtained by integrating the IR maps under the following spectral regions: (b) 3050 to $2800{\mathrm{cm}}^{-1}$, lipids (color scale 0.000 to 0.010 a.u.); (c) 1800 to $1480{\mathrm{cm}}^{-1}$, proteins (color scale 0.000 to 0.0450 a.u.); and (d) 1350 to $920{\mathrm{cm}}^{-1}$, nucleic acids (color scale 0.000 to 0.012 a.u.). Each acquired area corresponded to $492\times 164\mu {\mathrm{m}}^2$.

By means of UHCA on the raw spectra of each section, the ones belonging to the clusters of the epithelial and connective components were isolated, then independently resubmitted to PCA. As shown in the PCA score plot of epithelia [Fig. 4(a)], three well-segregated clusters were detected in the PC1–PC2 space, where the first and second principal components account for the total spectral variability in the amount of 57.3% and 19.2%, respectively. This result confirmed the histological suggestion of the occurrence of different epithelia in SNP, BSP, and SNUC lesions. In the PC1 (69.9%) versus PC2 (10.1%) score plot of connectives [Fig. 4(b)], only the cluster corresponding to the tumoral connective was fully segregated, in agreement with the presence, in the neighboring of tE, of an altered connective tissue produced by the tumor itself.^12,16 The other two clusters corresponding to SNP and BSP specimens were partially overlapped, pinpointing that connective inflammation was a common feature of both SNPs and BSPs, and unveiling one of the reasons it is hard to diagnose these two pathologies.

Fig. 4

PCA scores plots (PC1 versus PC2) of IR spectra corresponding to: (a) epithelia and (b) connectives found in SNP, BSP, and SNUC samples. The percentages in brackets represent the spectral variability of each component.

3.1.

Vibrational Analysis of Epithelia

To highlight specific spectral markers of each pathology, the average spectra of the different epithelia found in SNP, BSP, and SNUC lesions (Fig. 5) were analyzed using the peak-fitting procedure in the following spectral regions: 3050 to $2800{\mathrm{cm}}^{-1}$ (alkyl stretching modes), 1800 to $1480{\mathrm{cm}}^{-1}$ (amide I and II), and 1350 to $920{\mathrm{cm}}^{-1}$ (nucleic acid; data not shown).

Fig. 5

Absorbance average spectra of the various epithelia found in SNP, BSP, and SNUC sections. The spectra were vector normalized and two-points baseline linear fitted in the range 3800 to $900{\mathrm{cm}}^{-1}$. For reasons of clarity, BSP and SNUC spectra were shifted along the $y$-axis (absorbance was expressed in arbitrary units).

The IR region 3050 to $2800{\mathrm{cm}}^{-1}$, usually referred to as the lipid component, includes the symmetric and asymmetric stretching modes of ${\mathrm{CH}}_2$ and ${\mathrm{CH}}_3$ alkyl groups and the stretching vibration of $═\mathrm{CH}$ moiety. In all average spectra of the epithelia, the following absorptions were detected: $3013\pm 2$ ($\nu ═\mathrm{CH}$), $2959\pm 1$ (${\nu }_{\mathrm{asym}}{\mathrm{CH}}_3$), $2924\pm 1$ (${\nu }_{\mathrm{asym}}{\mathrm{CH}}_2$), $2870\pm 1$ (${\nu }_{\mathrm{sym}}{\mathrm{CH}}_3$), and $2850\pm 1$ (${\nu }_{\mathrm{sym}}{\mathrm{CH}}_2$) ${\mathrm{cm}}^{-1}$. The resulting band-area ratios were calculated: $3013/2959$ (unsaturation level) and $2924/2959$ (alkyl chains’ length and saturation level) (Table 1). With respect to the respiratory epithelium, the transitional and tumoral ones showed lower values of the ratio $2924/2959$, suggesting a decrease in ${\mathrm{CH}}_2$ groups, which could correspond to shorter lipid alkyl chains in BSP and SNUC epithelia.^17,18 In all the samples, a very weak band at $3013{\mathrm{cm}}^{-1}$, better detected in DII mode, was found, corresponding to $═\mathrm{CH}$ moieties. The trend of the band-area ratio $3013/2959$ shows a higher degree of unsaturation in the tE.^18,19

Table 1

Numerical variations of meaningful band-area ratios for the epithelia found in SNP, BSP, and SNUC lesions. Values indicate mean±SD. Significant differences between experimental groups were determined as detailed in Sec. 2 and are indicated by asterisks: #P nonsignificant; *P<0.05; **P<0.01; and ***P<0.001.

Band-area ratios	SNP	BSP	SNUC	Biochemical features	P values
3013/2959 ($\nu ═\mathrm{CH}/{\nu }_{\mathrm{asym}}{\mathrm{CH}}_3$)	$0.0063\pm 0.0005$	$0.0060\pm 0.0006$	$0.0104\pm 0.0008$	Unsaturation level	SNP versus BSP#
					BSP versus SNUC***
					SNP versus SNUC***
$2924/2959$ (${\nu }_{\mathrm{asym}}{\mathrm{CH}}_2/{\nu }_{\mathrm{asym}}{\mathrm{CH}}_3$)	$1.89\pm 0.02$	$1.19\pm 0.03$	$1.32\pm 0.02$	Saturation level	SNP versus BSP***
					BSP versus SNUC#
					SNP versus SNUC***
Helix/AI	$0.53\pm 0.05$	$0.51\pm 0.08$	$0.60\pm 0.06$	Proteins secondary structure	SNP versus BSP*
					BSP versus SNUC**
					SNP versus SNUC***
$\beta /\mathrm{AI}$	$0.29\pm 0.02$	$0.25\pm 0.04$	$0.21\pm 0.02$		SNP versus BSP*
					BSP versus SNUC*
					SNP versus SNUC***
Random/AI	$0.18\pm 0.02$	$0.24\pm 0.03$	$0.24\pm 0.01$		SNP versus BSP***
					BSP versus SNUC*
					SNP versus SNUC*
$1745/\mathrm{AI}$ ($\nu \mathrm{C}═\mathrm{O}$ ester/AI)	$0.0133\pm 0.001$	$0.0083\pm 0.0007$	$0.0103\pm 0.001$	Lipid amount	SNP versus BSP***
					BSP versus SNUC***
					SNP versus SNUC***
$1310/\mathrm{AI}$ (keratin/AI)	$0.0098\pm 0.001$	$0.0169\pm 0.002$	$0.0207\pm 0.002$	Keratin amount	SNP versus BSP***
					BSP versus SNUC#
					SNP versus SNUC***
$1120/1021$ (RNA/DNA)	$2.23\pm 0.2$	$1.81\pm 0.3$	$9.42\pm 0.1$	Cellular proliferation	SNP versus BSP#
					BSP versus SNUC***
					SNP versus SNUC***
$1050/1086$ ($\text{glycogen}/{\nu }_{\mathrm{sym}}{\mathrm{PO}}_{2^{-}}$)	$0.71\pm 0.08$	$0.40\pm 0.03$	$0.48\pm 0.02$	Glycogen amount	SNP versus BSP**
					BSP versus SNUC***
					SNP versus SNUC***
$1154/1171$ ($\nu \mathrm{COH}/\nu \mathrm{COP}$)	$0.87\pm 0.05$	$0.71\pm 0.04$	$0.48\pm 0.04$	Phosphorilative processes	SNP versus BSP*
					BSP versus SNUC**
					SNP versus SNUC**

Amide I and II bands elucidate proteins secondary structures. In our case, the following absorption bands were found: $1697\pm 2$, $1680\pm 1$, and $1621\pm 3{\mathrm{cm}}^{-1}$ ($\beta$ structures); $1668\pm 1$ and $1657\pm 1{\mathrm{cm}}^{-1}$ (helix components); $1639\pm 1{\mathrm{cm}}^{-1}$ (random coil structures); $1555\pm 1{\mathrm{cm}}^{-1}$ (helix and random coil structures); and $1538\pm 3{\mathrm{cm}}^{-1}$ ($\beta$ structures). According to amide I, meaningful band-area ratios were calculated: helix/AI, $\beta /\mathrm{AI}$, and random/AI, where AI is amide I total area (Table 1). In particular, a decrease in $\beta$-sheet components was found in the tE (SNUC), together with a small increase in helix and random coil structures.^20,21

A weak band at $1745{\mathrm{cm}}^{-1}$ (usually attributed to the stretching vibration of ester carbonyls of triglycerides) was found in all epithelia but was weaker in BSP and SNUC samples (Fig. 5). By analyzing the trend of the band ratio $1745/\mathrm{AI}$, a higher consumption of lipids could be conceivable in the tumoral lesions.¹⁷

The increase in the band ratio $1310/\mathrm{AI}$, both in BSP and SNUC epithelia, confirms the presence of keratin and, as a consequence, the squamous nature of these two lesions (Table 1).²²

The spectral region 1350 to $920{\mathrm{cm}}^{-1}$ includes the main vibrations of nucleic acid functional groups. In the tE (SNUC), an increase in the band-area ratio $1120/1021$ (RNA/DNA), together with a decrease in $1050/1086$ ($\text{glycogen}/{\nu }_{\mathrm{sym}}{\mathrm{PO}}_{2^{-}}$), was found, confirming the high degree of malignancy with a hypercellular proliferation and an enhanced consumption of glycogen.^11,23,24 In SNUC, the presence of phosphorilative processes was also pinpointed by the trend of the band-area ratio $1154/1171$ ($\nu \mathrm{COH}/\nu \mathrm{COP}$) (Table 1).²⁵

3.2.

Vibrational Analysis of Connective Tissues

As mentioned above, some epithelial tumors may infiltrate and modify the neighboring connective tissue, inducing evident morphological and structural modifications in it.^12,16 To better highlight this aspect, absorbance average spectra of connective tissues in SNP, BSP, and SNUC samples were analyzed by the peak-fitting procedure in the spectral range 1480 to $920{\mathrm{cm}}^{-1}$, in which the most significant bands attributable to collagen are present. The following trends were found (Fig. 6): (i) an increase in the band-area ratio $1397/1450$ (${\nu }_{\mathrm{sym}}\mathrm{COO}─/\delta {\mathrm{CH}}_{2/3}$), suggesting more evident peroxidative processes in SNUC;^16,26 (ii) an increase in the band-area ratio $1082/1238$, typical of tumoral tissues;^27,28 and (iii) a decrease in the side bands in the characteristic tricuspid shape of collagen (1281, 1238, and $1202{\mathrm{cm}}^{-1}$).²⁹

Fig. 6

Numerical variations of meaningful band-area ratios for the connective tissues found in SNP, BSP, and SNUC lesions: $1397/1450$ (${\nu }_{\mathrm{sym}}\mathrm{COO}─/\delta {\mathrm{CH}}_{2/3}$), $1082/1238$ (${\nu }_{\mathrm{sym}}{\mathrm{PO}}_{2^{-}}/{\nu }_{\mathrm{asym}}{\mathrm{PO}}_{2^{-}}$), $1281/1238$ ($\text{collagen}/{\nu }_{\mathrm{asym}}{\mathrm{PO}}_{2^{-}}$), and $1202/1238$ ($\text{collagen}/{\nu }_{\mathrm{asym}}{\mathrm{PO}}_{2^{-}}$). Values indicate mean ± SD. Significant differences between experimental groups were determined as detailed in Sec. 2 and are indicated as follows: #$P$ nonsignificant; *$P<0.05$; **$P<0.01$; and ***$P<0.001$.

4.

Conclusions

FTIRI might represent a valid complementary tool in clinical analysis and prognosis, especially when it is difficult to make an indubitable diagnosis by exploiting only conventional methodologies. In fact, it has been well ascertained that FTIRI could detect at a molecular level the biological modifications in pathological tissues, even before they can be discovered using routine diagnostic assays.

In this light, this contribution represents a starting point in the assessment of FTIRI to achieve an objective diagnosis of some recurring sinonasal lesions, such as SNPs, BSPs, and SNUCs. In particular, specific spectral features of the various epithelium layers and, to a further extent, of the neighboring connective tissues were defined to discriminate malignant lesions from benign and inflammatory ones.

The tE found in the undifferentiated sinonasal carcinoma shows, with respect to the respiratory one, the occurrence of peroxidative ($3013/2959$, $\nu ═\mathrm{CH}/{\nu }_{\mathrm{asym}}{\mathrm{CH}}_3$) and phosphorilative ($1154/1171$, $\nu \mathrm{COH}/\nu \mathrm{COP}$) processes, a modified proteins secondary structure (with a decrease in $\beta$-sheets components and a small increase in helix and random coil structures), a major consumption of lipids ($1745/\mathrm{AI}$, $\nu \mathrm{C}═\mathrm{O}$ ester/AI) and glycogen ($1050/1086$, $\text{glycogen}/{\nu }_{\mathrm{sym}}{\mathrm{PO}}_{2^{-}}$), and a higher cellular proliferation ($1120/1021$, RNA/DNA). The squamous nature of this lesion is confirmed by the presence of keratin ($1310/\mathrm{AI}$, keratin/AI) and also of a modified connective tissue. The trE of the Schneiderian papilloma, which shows some features similar to the tumoral one (a higher consumption of lipids and glycogen and the presence of keratin), is characterized by shorter alkyl chains ($2924/2959$, ${\nu }_{\mathrm{asym}}{\mathrm{CH}}_2/{\nu }_{\mathrm{asym}}{\mathrm{CH}}_3$) and unordered proteins (an increase in random coil structures).

The application of this spectroscopic method in common clinical practice demands a larger cohort of patients and for the extension of the study to an increasing number of oral pathologies. However, this paper once more highlights the full potential of FTIRI in medical diagnosis.