Optical diagnosis of peritoneal metastases by infrared microscopic imaging PDF

Preview

Summary

Anal Bioanal Chem (2009) 393:1619–1627 DOI 10.1007/s00216-009-2630-2 ORIGINAL PAPER Optical diagnosis of peritoneal metastases by infrared microscopic imaging Valérie Untereiner & Olivier Piot & Marie-Danielle Diebold & Olivier Bouché & Elodie Scaglia & Michel Manfait Received: 1...

Description

Anal Bioanal Chem (2009) 393:1619–1627 DOI 10.1007/s00216-009-2630-2

ORIGINAL PAPER

Optical diagnosis of peritoneal metastases by infrared microscopic imaging Valérie Untereiner & Olivier Piot & Marie-Danielle Diebold & Olivier Bouché & Elodie Scaglia & Michel Manfait

Received: 14 November 2008 / Revised: 14 January 2009 / Accepted: 15 January 2009 / Published online: 14 February 2009 # Springer-Verlag 2009

Abstract Fourier transform infrared (FTIR) spectroscopy is nowadays widely accepted as a technique with high potential for diagnosis of cancerous tissues. This study presents an example of the investigation of peritoneal metastases by FTIR microimaging. Peritoneal malignancies are generally secondary localizations of primary visceral cancers such as ovarian, stomach or colon cancers. By analysing simultaneously both formalin-fixed paraffinembedded and frozen specimens, we examined malignant and non-malignant (i.e. fibrotic and cicatricial) peritoneal lesions. Paraffin-embedded tissues were analysed without any previous dewaxing. Multivariate statistical approaches, based on the classification of infrared data by hierarchical cluster analysis, allowed the discrimination of these various samples. Microimaging also permits the revelation of the heterogeneity of the tissue: it was possible to localize precisely the cancerous areas, and to distinguish, on the basis of their spectral signatures, the peritumoral neighbouring connective tissue close to the carcinomatous areas from the connective V. Untereiner : O. Piot (*) : M.-D. Diebold : O. Bouché : M. Manfait Unité MéDIAN, CNRS UMR 6237 MEDyC, IFR53, UFR Pharmacie, Université de Reims Champagne-Ardenne, 51 rue Cognacq Jay, 51096 Reims Cedex, France e-mail: [email protected] M.-D. Diebold Laboratoire d’Anatomie et Cytologie Pathologiques, CHU Robert Debré, Avenue du Général Koenig, 51092 Reims Cedex, France O. Bouché : E. Scaglia Service d’Hépato-Gastroentérologie, CHU Robert Debré, Avenue du Général Koenig, 51092 Reims Cedex, France

tissue distant from the cancerous areas. These spectral differences could be useful as complementary information to study molecular changes associated with the malignancy. Keywords Micro Fourier transform infrared analysis . Peritoneal metastases . Optical diagnosis . Multivariate statistical treatment Abbreviations FFPE formalin-fixed paraffin-embedded FTIR Fourier transform infrared HCA hierarchical cluster analysis HE haematoxylin–eosin

Introduction Fourier transform infrared (FTIR) microimaging permits the characterization of complex biological systems in terms of molecular and structural information. The non-destructive spectral data that can be collected at a spatial resolution of tens of micrometres allow the assessment of the molecular features of a tissular specimen, without any particular preparation of the tissue section. The spectral signature of a tissue, which is highly representative of its histopathological state, can be used not only to discriminate healthy and pathological specimens, but also to differentiate different types or grades of tumour lesions. Various solid cancers have been analysed by vibrational spectroscopy. Breast, cervix, skin, ovarian and colon tissues [1–11] are some examples of organs in which malignant lesions have been examined by mid-infrared (FTIR) spectroscopy. Recently, potential clinical developments involving highthroughput infrared imaging for tissular characterization of tumoral specimens have also been published [12, 13]. The

1620

development of spectral imaging as a routine analytical technique for biomedical purposes requires, on one hand, the construction of a database of spectroscopic markers highly specific for the different histopathological states encountered in a given disease and, on the other hand, the validation of the method for a large number of samples. These steps can be achieved by analysing histologically well characterized specimens stored in either frozen or formalin-fixed paraffin-embedded (FFPE) form in tumour libraries. Usually, studies of tissues by FTIR spectroscopy are performed on thin tissue sections obtained from either frozen specimens (not always available in tumour libraries) or FFPE specimens but after chemical dewaxing to avoid the parasitic signal of the paraffin. However, Tfayli et al. [14] have demonstrated that infrared microspectroscopy is efficient in discriminating malignant melanomas from melanocytic benign tumours in FFPE sections without previous dewaxing, by exploiting a spectral region devoid of the paraffin infrared signal but that contains spectral markers associated with the malignancy. In this study, we present a mid-infrared spectroscopic investigation of peritoneal tissues. Peritoneal malignancies (except in rare cases of mesothelioma) occur as a result of secondary localizations of primary visceral cancers such as ovarian, stomach or colon cancers. In certain situations, the detection of cancer cells within the peritoneal cavity represents a real challenge particularly in the case of per-operative histological examination of tissue sections likely to contain scattered malignant cells as in some independent gastric carcinoma cells. The potential of FTIR microspectroscopy for the discrimination between malignant and non-malignant peritoneal specimens was assessed by analysing both frozen and FFPE biopsies of peritoneum. Statistical multivariate analysis (hierarchical cluster analysis, HCA) was performed firstly on paraffin-embedded specimens and, in a second step, on both frozen and paraffin-embedded samples processed simultaneously. Within malignant specimens, the spectral differences between the cancerous regions and the connective tissues, peritumoral and distant from the tumoral zones, were outlined. The samples were previously subjected to histological examination for diagnosis to verify the extent of visceral malignancy at the level of the peritoneum.

Materials and methods Tissue samples Twelve peritoneal samples, coming from 11 different patients, were collected during the surgical resection of visceral cancers. The specimens were selected by a pathologist (M.-D.D.) as being representative of patholog-

V. Untereiner et al.

ical cases observed in clinics (Table 1). Note that the incidence of the peritoneal metastases was independent of the sex. Two (samples 1 and 7) of them were available in both frozen and FFPE forms. One sample (sample 4) was available only in frozen form. All other samples were FFPE. In all cases, the specimens came from abnormal peritoneal tissues: the malignant samples correspond to cancerous tissues, while the benign ones were cicatricial or fibrotic tissues. Samples 5 and 6 came from the same patient: sample 6 corresponds to a stomach adenocarcinoma which is the primary cancer of the peritoneal sample 5. Tenmicron-thick microtomed sections were deposited on ZnSe windows, transparent to the mid-infrared electromagnetic radiation. The FFPE sections were analysed without any dewaxing before spectroscopic measurements. In each case, an adjacent section was stained with haematoxylin–eosin (HE) for conventional histopathological examination. In sample 1f (Table 1), the section was stained directly on the ZnSe window after FTIR analysis. Immunohistochemistry with KL1 (monoclonal mouse antibody of cytokeratin, Immunotech, France) and vimentin (monoclonal mouse antibody of vimentin, Dako Cytomation, Denmark) was performed on the benign sample 7 to confirm the absence of isolated malignant cells, as a complement to the HE staining [15]. FTIR microspectroscopic imaging FTIR images were recorded using a Spectrum Spotlight 300 FTIR imaging system coupled to a Spectrum One FTIR spectrometer (both from PerkinElmer Life Sciences, France). For imaging, the device is equipped with an XY motorized plate and a nitrogen-cooled mercury cadmium telluride 16-pixel-line detector which can be operated either at 6.25 or at 25 μm per pixel resolution. In our study, we used the 25 μm spatial resolution: each spectrum corresponds to a 25 μm × 25 μm pixel. Spectral resolution was set to 4 cm−1. Each absorption spectrum composing the infrared images, and resulting from 32 scans, was recorded for each pixel in the transmission mode over the 720– 4,000 cm−1 (13.88−2.5 μm) spectral range using the Spotlight software (PerkinElmer). Prior to each data acquisition, a reference spectrum from the ZnSe window and the environment was recorded at 4 cm−1 spectral resolution with 240 accumulations. This reference spectrum was subsequently subtracted from the dataset automatically. A visible image of the tissue was also recorded to delineate the zone to be analysed. Data preprocessing and analysis Spectral images, converted from transmittance into absorbance, were first corrected for the contribution of atmo-

Optical diagnosis of peritoneal metastases by infrared microscopic imaging

1621

Table 1 Description of the specimens analysed Patient

Conditioning

Pathological data

Sample number Sex Frozen FFPE

Final diagnosis

1

M

1f

1p

Peritoneal metastasis

2

M

–

2p

3

M

–

3p

4

M

4f

–

5

F

–

5p

6

F

–

6p

7

F

7f

7p

8

F

–

8p

9

F

–

9p

10

F

–

10p

11

F

–

11p

12

M

–

12p

Specimen type Suspected diagnosis

Peritoneal invasion Omentum metastasis Peritoneal invasion Peritoneal metastasis Peritoneal invasion Peritoneal metastasis Peritoneal invasion Peritoneal metastasis Peritoneal invasion Stomach adenocarcinoma Stomach invasion Cicatricial tissue Peritoneal invasion Hypodermic endometriosis Peritoneal invasion Dense fibrosis Peritoneal invasion Fibrosis Peritoneal invasion Absence of malignancy Peritoneal invasion Absence of malignancy Peritoneal invasion

Medical history

Colon adenocarcinoma

Malignant

Colon adenocarcinoma

Malignant

Stomach adenocarcinoma

Malignant

Pancreas adenocarcinoma

Malignant

Stomach adenocarcinoma

Malignant

Stomach adenocarcinoma

Malignant

Gall bladder adenocarcinoma

Non-malignant

Paraumbilical nodule (undetermined cause) Non-malignant Pancreas adenocarcinoma

Non-malignant

Herniated omentum

Non-malignant

Pancreas adenocarcinoma

Non-malignant

Colon adenocarcinoma

Non-malignant

M male, F female, FFPE formalin-fixed paraffin-embedded

spheric water vapour and CO2 absorption bands by a builtin function in the PerkinElmer Spotlight software. They were baseline-corrected using an elastic scattering correction algorithm and normalized to the amide I band intensity. The second derivative was then calculated for each dataset. HCA was applied to the spectral images using Cytospec version 1.2 (http://www.cytospec.com). Spectral data were clustered using Ward’s algorithm after calculation of the Euclidean interspectral distances. Colour-coded images, mapping the distribution of HCA clusters within the tissue slices analysed, were then constructed on the basis of three, six and nine clusters. Histological correspondence of the HCA clusters was carried out by comparing colour-coded spectral images with the HE-stained sections. For each image, a substantial number of 30 spectra were extracted in a manner which covers the whole expanse of the image. The extracted spectra were processed by HCA using the Opus software suite (Bruker, Reinstetten, Germany), with Euclidean distances and Ward’s algorithm as parameters. The merged result can be visualized in a treelike diagram, which is called a “dendrogram”, presenting the regrouping of the spectra in clusters according to a heterogeneity scale.

Results To describe the methods used in this work, we first present the processing of the first sample (sample 1) named “mixed sample” since it contained both cancerous and nonmalignant regions. This sample was available in both frozen and FFPE forms. Figure 1 shows typical infrared spectra of cancerous and non-malignant areas from both frozen (sample 1f) and paraffin-embedded (sample 1p) sections; the tissue areas were localized by means of the HE-stained sections. The main infrared vibrations of the paraffin were clearly identified in 1440–1485 and 2,820– 2,995 cm−1 spectral windows; and also less intensely at 1,378 cm−1. Consequently, spectral ranges with no paraffin contribution could be selected for the infrared characterization of the FFPE tissues. In particular, the more distinctive spectral differences between malignant and non-malignant tissues were located in the 905–1,300 cm−1 spectral window, which is devoid of paraffin-associated signal (Fig. 1b). From this point, all of the analyses were conducted on this selected 905–1,300 cm−1 range whatever the conditioning of the sample. To reveal the main tissue

1622

V. Untereiner et al.

Fig. 1 Typical spectra (baseline-corrected, normalized on the amide I absorption band) of malignant and non-malignant areas from both frozen and formalin-fixed paraffin-embedded (FFPE) peritoneal tissues (samples 1f and 1p). a Mean spectra in the 830–4,000 cm−1 spectral

range. b Enlargement of the mean spectra in the 905–1300 cm−1 spectral range. FFPE malignant-tissue (red), FFPE non-malignant tissue (black), frozen malignant tissue (green), frozen non-malignant tissue (blue)

structures, FTIR microimaging was performed on thin sections of the specimens. For a frozen section of sample 1f, Fig. 2 depicts an example of a raw spectral image (Fig. 2a) and of a colour-coded image (Fig. 2b) based on three clusters determined by HCA, together with the corresponding HE-stained section (Fig. 2c). This “mixed tissue” contained a malignant region comprising carcinomatous areas (colon cancer metastases in this sample) surrounded by neighbouring connective tissue and a nonmalignant region consisting of healthy connective tissue

distant from the cancerous nests. The reconstruction of the colour-coded image permitted us to enhance the image contrast by clustering the spectral pixels; that allowed the main tissular structures to be outlined on the basis of their specific molecular content. The comparison of the clusterbased image with the HE staining of the same section allowed assignment of the three clusters to carcinomatous areas, the peritumoral connective tissue and the distant connective tissue, respectively. Data classification based on six and nine clusters was also carried out but did not lead to

Fig. 2 Fourier transform infrared imaging of malignant peritoneal specimen (sample 1f). a Spectral image reconstructed from the total absorbance (720– 4,000 cm−1). b Pseudo-colour images reconstructed on the basis of three clusters determined by hierarchical cluster analysis (HCA) on the 905–1,300 cm−1 spectral range. c Corresponding haematoxylin–eosin stained section on a ZnSe window

Optical diagnosis of peritoneal metastases by infrared microscopic imaging

1623

Fig. 3 HCA (905–1,300 cm−1 spectral range, second derivatives) of spectra belonging to the three clusters, identified in the image in Fig. 2b and associated with malignant and non-malignant areas. The spectral distance between malignant and non-malignant regions is

indicated by the circle, the distance between carcinomatous areas and peritumoral connective tissue is indicated by the square and the intracluster distance within the non-malignant region is indicated by the triangle

better distinction between these three tissular structures (data not shown). To assess the spectral differences between these structures, HCA was performed, over the 905– 1,300 cm−1 window, on spectra belonging to the three different clusters, extracted from the raw spectral image. The dendrogram in Fig. 3 indicates a clear distinction between the non-malignant part of the tissue, corresponding to the connective tissue distant from tumoral nests, and the malignant part consisting of both the tumour areas and the peritumoral connective tissue. The discrimination potential of infrared spectroscopy can be assessed by comparing the

intercluster distance (see the corresponding circle on the dendrogram) with the intracluster distance for the nonmalignant cluster (see the triangle). The dendrogram also indicates, by a low value of the spectral distance (see the square), that the spectroscopic features of the peritumoral tissue were similar to those of the cancerous areas. This result could support the idea that the peritumoral connective tissue contains spectroscopic markers associated with the malignancy. To investigate the molecular basis of this spectral classification, the spectral signatures (mean spectra of the clusters) specific for the three structures of this

Fig. 4 Mean spectra [second derivative, (x-1)] of the clusters identified from the dendrogram in Fig. 3. Cancerous tissue (red), peritumoral tissue (blue), nontumoral tissue (green)

1624

V. Untereiner et al.

Table 2 Assignment of main infrared spectral differences between the carcinomatous areas, the peritumoral connective tissue and the nonmalignant connective tissue Wavenumber (cm−1)

Assignments

References

[16–18] [4, 5, 17, 19, 20–24] [4, 5, 18, 21] [23, 26–29] [16–18, 27, 30] [22, 28]

Connective tissue distant from the tumour

Connective tissue close to the tumour

Carcinomatous areas

1,283 –

1,283 –

– 1,241

1,238 – 1,204 1,172

1,238 – 1,204 1,172

– 1,223 – 1,172

1,157

1,157

–

1,126

1,126

1,126

Collagen Nucleic acids (DNA, RNA) and phosphorylated proteins: νasym(PO2−) Collagen Nucleic acids: νasym(PO2−) Collagen Cell proteins: ν(C–O) of C–OH side chain groups of proteins, more intense in malignant tissue Cell proteins and carbohydrate: ν(C–O) of C–OH side chain groups of proteins + ν(C–O) of carbohydrates such as collagen or glycogen RNA: ribose vibration

–

–

1,088

Nucleic acids (DNA, RNA): νsym(PO2−)

1,084

1,084

–

+ Glycoprotein such as collagen or glycogen, νC–C

1,062 –

1,062 –

– 1,054

1,048

1,048

–

Glycoprotein such as collagen or glycogen: ν(C–O) + δ(C–O) Superimposition of 1,062- and 1,048-cm−1 vibrations + Nucleic acid (DNA, RNA): weak intensity Glycoprotein such as collagen or glycogen: ν(C–O) + δ(C–O)

1,032

1,032

–

– 971

995 970

995

–

–

967

Glycoprotein such as collagen or glycogen: −CH2OH vibrations coupled with δ(C–O) RNA: νs(PO42−) Glycoprotein such as collagen or glycogen? DNA: νs(PO42−)

[5, 19, 28, 31, 32] [20, 23, 24, 26, 32, 33] [4, 5, 16, 17, 19, 22, 23, 25–29, 34] [19, 25, 31, 32, 35] [22, 27, 36] [23] [19, 22, 24, 31, 32] [4, 17, 22, 24, 26, 35, 36] [20, 22, 23, 37] No reference found [18, 20, 26]

The assignment is done for the spectra presented in Fig. 4

malignant specimen were compared...