All head and neck tumor samples (n = 14) and normal controls (n = 14) were obtained after surgical resection at the ENT (Ear, Nose, Throat) Department of the University Hospital Jena; they had been collected fresh, snap frozen in liquid nitrogen, and were stored at -80°C. Tumor specimen were categorized to the WHO classification criteria 10 . Ethical approval was obtained from the local Research Ethic Committee.

From these samples 12 μm cryostat sections were prepared. One section was stained with hematoxylin-eosin (HE) and examined microscopically in order to detect tissue areas of interest for microdissection (see 11 ). A corresponding unstained tissue section was mounted on a microscope slide coated with a 1.35 μm membrane (polyethylene naphtalate (PEN) Zeiss/Palm, Bernried, Germany). Tissue areas from normal and tumor stroma (approx. size 300 × 300 μm) containing approximately 100 to 500 cells were cut out and moved by a laser microdissection and pressure catapulting microscope (LMPC; Zeiss/Palm, Bernried, Germany) or a fine needle directly on ProteinChip arrays (Fig. 1). For catapulting, a microplasma is induced under the dissected tissue area. This plasma lifts the piece of tissue to a reaction cup or to a ChipArray fixed by a special mount, each. For regular formed tissue pieces with more than 100 cells we found that it is more secure to attach the dissected area to a fine needle and deposit it elsewhere under microscopically control.

A Q10 ProteinChip array (strong anion exchanger; BioRad) was activated (see 11 ) and wetted with 0.5 μl lysis buffer (100 mM Na-phosphate (pH 7.5), 5 mM EDTA, 2 mM MgCl₂, 3 mM 2-β-mercaptoethanol, 0.1% CHAPS, 500 μM leupeptine, and 0.1 mM PMSF). Under a stereo microscope (Stemi 2000c, Zeiss) the tissue section was placed on the spot of the ProteinChip array. Tissue lysis on spot was performed for 1.5 h at 4°C in a humidity chamber. After lysis and incubation the spots were washed three times with 5 μl of a washing/binding buffer (100 mM Tris-buffer, pH 8.5 with 0.02% Triton X-100) and rinsed 2 times with water. 2 × 0.5 μl sinapinic acid (saturated solution in 0.5% TFA/50% acetonitrile) was applied as matrix on the dried spots. The matrix which co-crystallizes with proteins absorbs the laser energy and transfers part of its charge to the proteins. Mass analysis was performed in a ProteinChip Reader (PCS 4000, Ciphergen Biosystems Inc, Fremont, CA) with a manual data collection protocol.

Because cells were microdissected, placed and lysed directly on the spot of the ProteinChip array under control of a stereo microscope, we named this technique 'tissue on chip based mass spectrometry' (toc-MS). Areas of different size and cell number were tested. At least 100 cells were needed for an adequate spectrum. For the analysis of the normal and tumor samples 300 cells were dissected for more robust results. Compared to the SELDI standard procedure the sensitivity is increased at least tenfold and, because no protein lysis and extraction is needed, time of analysis is shorter by half. In contrast to MALDI imaging, which allows to analyse spatial resolved protein spectra over tissue sections and other mass spectrometry techniques, the SELDI characteristic affinity chromatograhic chip surfaces allow a more quantitative analysis of proteins.

The resulting protein profiles between 2 kDa to 20 kDa (low range) and 20 kDa to 200 kDa (high range) were subjected to CiphergenExpress™ Client 3.0 software (CE) and a cluster and rule-based data mining algorithm (XLminer 3.0, BioControl Jena GmbH). The CE software was used for the processing of raw spectra and the calculation of P-values and cluster plots. In the low range we found 8 peaks with a P-value lower 0.05. In the high range 5 peaks with this characteristic could be found. The two most significant proteins for the low and high range are displayed in box plots in Figure 2.

The 7,477 Da peak is significantly higher expressed (P = 0.0003) in tumor stroma, while the 80,044 Da peak (P = 0.0009) is reduced in tumor stroma. An initial data base search according molecular size offered for the 7,477 Da mass the beta defensin 119 (UniProtKB/Swiss-Prot Q8N690 Chain: 22-84: 7493 Da). Human beta-defensines (HBD) are cationic, antimicrobial peptides produced by epithelial cells and show altered inconsistent expression in cancers 12 13 . Analyses of their expression in tumor stroma are not published yet. For the fibroblast growth factor 23 only a role in phosphate homeostasis and related disorders is known 14 . The protein with a molecular mass of 80,044 Da is equivalent in size to the unphosphorylated ski oncogene (UniProtKB/Swiss-Prot P12755, 80,005 Da) which was discovered as oncogene by its ability to transform chicken embryo fibroblasts upon overexpression. But in newer studies also anti-oncogenic activities are discussed (for review see 15 ).

The subsequent modified fuzzy c-means data analysis algorithm underlying the XLminer software 5 consists of three steps in particular allowing adequate analysis of small sample groups. The clustering step, the rule extraction and rating step, and the rule-base construction step finally result in a heat-map and in values for sensitivity and specificity separating both groups. The analysis of all tumor and normal samples with XLminer resulted in a sensitivity of up to 92.8% and a specificity of 100% (Fig. 3).

In conclusion, we applied toc-MS successfully to analyse a few hundred stromal cells quantitatively and to differentiate between those stromal areas near to tumor and to normal epithelium. An exact identification of these proteins with tryptic digestion and tandem MS is in progress. Ongoing research focuses on down-scaling the procedure to a higher sensitivity.