Introduction

Despite many efforts, it has remained very difficult to obtain accurate risk assessment of both development and progression of head and neck squamous cell carcinoma (HNSCC). This is most evident in HNSCC patients who are at high risk of local tumor relapse or development of a second primary tumor which both have a strong negative impact on patient survival. The failure of preventing and controlling the disease has been linked to the concept of field cancerization developed by Slaughter et al. (1953), which implies that the mucosa of the upper aerodigestive tract is potentially prone to undergo malignant transformation or further progression at multiple sites.

However, it is now clear from recent studies that molecular changes underlying field cancerization are not confined to areas with altered histology as initially assumed by Slaughter et al. (1953) It was frequently observed that the genetic damage persisted beyond the histological border of precancerous lesions, and tumors often developed away from the precancerous site (Tabor et al., 2001; Braakhuis et al., 2003). Thus, a large fraction of a carcinogen-exposed field or even the whole of it may not present with clinical or morphological symptoms. Similarly, mucosal fields of genetically altered cells without histological changes have also been found in tumor patients (reviewed in Braakhuis et al. (2005)). Therefore, field cancerization is currently viewed as a field of cells in which a single stem cell has acquired initiating alterations caused by carcinogens, allowing the daughter cells to grow out to form a patch and then a field of premalignant cells (Garcia et al., 1999; Braakhuis et al., 2005). At some degree of preneoplastic progression, these changes may become histologically apparent as mild, moderate or severely dysplastic areas, but as a rule remain unrecognizable.

Evidence for the presence of genetic lesions was mostly based on genetic and cytogenetic analyses of tumor-adjacent and tumor-distant mucosae from HNSCC patients. For example, patches of cells harboring mutations in the p53 tumor suppressor gene and overexpressing the p53 protein have been detected in very early lesions (Nees et al., 1993; Waridel et al., 1997; Prevo et al., 1999). These early p53 alterations have been more frequently observed in patients suffering from multiple cancers compared to those with only one tumor (Waridel et al., 1997; Homann et al., 2001), in agreement with the field model (Garcia et al., 1999). In cytogenetic studies, allelic imbalances/loss of heterozygosity (LOH) and chromosomal aneuploidy have been detected in tumors and mucosal biopsies of tumor patients (Soder et al., 1995; Bedi et al., 1996; Califano et al., 1996; Ai et al., 1999; Partridge et al., 2000; Braakhuis et al., 2002; Tabor et al., 2004). Chromosomal aneuploidy appears to precede malignant transformation as indicated by findings of monosomy and trisomy in histologically normal squamous mucosa (Ai et al., 2001; Wolf et al., 2004).

These molecular findings support the model in which fields of genetically altered cells play a central role and bear important clinical consequences. For instance, such fields may persist after resection of the primary tumor and may result in either local recurrency (minimal residual disease) or, if located at distance to the primary tumor, in second primary cancer (Braakhuis et al., 2002). Accordingly, diagnosis, treatment and clinical monitoring should focus on the presence of genetically altered fields. Rapid and cost-effective diagnostic methods are required which can reliably detect such fields, without negatively impacting on the life quality of the patients.

Surface-enhanced laser-desorption-ionization time-of-flight mass spectrometry (SELDI-TOF-MS) combines chromatographic surfaces with solid-state TOF-MS for the high through-put analysis of proteins in diverse clinical samples (serum, urine, cerebrospinal fluids, saliva, surgical biopsies) and has been widely used for biomarker discovery (Wright, 2002; Wiesner, 2004). In comparison to other proteomics technologies such as liquid chromatography LC-MS/MS or two-dimensional polyacrylamide gel electrophoresis (2D PAGE), SELDI-TOF-MS is more versatile and easy to set up in a clinical environment: it is rapid, reproducible, and adaptable to diagnostic high-throughput formats. In this study, SELDI-TOF-MS was applied to a selection of 303 clinical biopsies (tumors and mucosae from tumor patients and healthy donors). We aimed at establishing a robust experimental procedure for the identification of premalignant changes. Our findings indicate and support the existence of genetically altered fields, both in close vicinity to and at considerable distance from the primary tumors. The procedure may be clinically useful to rapidly identify patients at a high risk of local recurrence or second primary tumor development.

Results

Protein expression patterns in normal and pathologic tissues

A total of 303 biopsies (113 HNSCCs, 73 healthy, 99 tumor-distants, and 18 tumor-adjacent squamous mucosae) were analysed on IMAC30 ProteinChip arrays, preactivated with copper. From all protein peaks detected in at least 8% of all samples, 48 were differentially expressed between healthy mucosa and HNSCC. Some of the peaks had very high P-values in different statistical tests (e.g. P<10^-20 in Student's t-test or ANOVA), indicating strong statistical significance of differential expression. In comparison, a peak at 11.3 kDa showed no significant difference between HNSCC and mucosae (P=0.127), and the spiked-in aprotinin control peak scored P=0.775.

Special focus was on mucosa samples that were excised at the same time the patients were undergoing tumor operation and were confirmed as free of tumor by histological examination using hematoxylin and eosin (H&E) stained sections. Eighteen samples were tumor-adjacent (within the resection margin), 99 were tumor-distant (>2 cm distance to the resection margin). The analysis revealed striking expression patterns in a subset of the mucosae from tumor patients which resembled patterns seen in HNSCC, as exemplified in Figure 1 for a set of prominent peaks in the mass range between 3.3 and 3.8 kDa. With very few exceptions, none of the normal control samples showed significant expression of these proteins (Figure 1, 'Patient 1', left panel), while three out of five tumor-distant mucosa samples of Patient 2 showed high expression levels (Figure 1, 'Patient 2', right panel).

Figure 1.

Representative SELDI-TOF MS expression patterns for the low molecular weight range between 3000 and 4000 kDa in five mucosal biopsies from one healthy donor (left panel) and five tumor-distant mucosal biopsies from one HNSCC patient (right panel). The three most prominent peaks around 3500 kDa were identified as neutrophil -defensins 1, 2 and 3 (encircled) and the peak at 3700 Da as the C-terminal fragment of hemoglobin chain (^*). Expression of these peaks was hardly detectably in most normal control samples, but high in most of the tumors.

Full figure and legend (135K)

Figure 2 shows a heat map generated from all biopsies analysed. Protein peaks in the mass range between 10.8 and 13.0 kDa revealed very homogeneous expression patterns in the normal control mucosa samples including a subset of the tumor-distant mucosae (Figure 2, cluster 1). From this group, all but one peak (11.3 kDa) were significantly reduced in the HNSCC samples (cluster 3). They were also highly variable and often strongly reduced in tumor-adjacent mucosae, and in some tumor-distant mucosae located in clusters 2 and 3 as compared to cluster 1. A peak at 38.6 kDa showed strongly reduced expression in tumor samples (Figure 2, cluster 3), in tumor-adjacent mucosae and even in some tumor-distant mucosae (also in clusters 2 and 3). These protein peaks showed a strong negative correlation to the small peptides described above in the mass range of 3.3–3.7 kDa. The diagnostic value for all of these peaks was demonstrated in highly significant receiver operated characteristic (ROC) plots showing areas under the curves (AUC) well above 0.9 (data not shown, see (Roesch Ely et al., 2005)).

Figure 2.

Unsupervised hierarchical cluster analysis of SELDI-TOF MS protein expression data (average linkage clustering, Spearman rank correlation). The clustering algorithm groups the biopsies in three main clusters. Cluster 1 (left) contains most of the healthy control mucosae, and only one tumor sample (HNSCC 37). Cluster 1 shows the most homogeneous expression pattern with few outliers. Cluster 2 (middle) consists mainly of tumor-adjacent and tumor-distant mucosae. Cluster 3 (right panel) contains all the HNSCC samples (with one exception), and none of the healthy control mucosa samples. The tumor-adjacent mucosa samples are divided exclusively between Clusters 2 and 3, while the tumor-distant mucosa samples can be found in all three, including the HNSCC cluster. Red bar: m/z of DEF 1–3; small red square, -hemoglobin fragment; blue bar and blue dots, S100 group/Cystatins, Histone H4; black dot, annexin 1.

Full figure and legend (299K)

Identification of differentially expressed proteins

To purify and identify differentially expressed proteins from biopsies of healthy mucosa or HNSCC, lysates were selected in which the protein peaks of interest were present in high abundance. These samples were fractionated by IMAC and/or reverse phase chromatography as described in the 'Materials and methods' section. The fractions containing the enriched proteins were loaded on a SDS–PAGE. Here, it cannot be taken for granted that a protein band cut out from the gel and subjected to tryptic digestion and mass spectrometry will actually contain the candidate biomarker. Therefore, passive elution of intact proteins from gel bands was employed as described previously (Le Bihan et al., 2004; Hegedus et al., 2005). We have employed this strategy to the proteins in the mass region between 8 and 14 kDa, and an example is shown in Figure 3 (a larger m/z range is shown in Supplementary Figure 2). As indicated in A, two bands (B and B') were cut out from the gel. The analysis of the eluted proteins showed that from each band only one major protein had eluted (C and C'), yielding masses of 11.312 and 11.043 Da, respectively. Thus, we had purified the proteins of interest. The identification of these two proteins peaks was achieved by tryptic digestion in solution (D) followed by mass spectrometry. The peak with 11.312 Da which was constant in both HNSCC and mucosa was identified as Histone H4 by MALDI-MS and ESI–MS/MS (Table 1). The MALDI-MS derived peptide fingerprint is shown in Figure 3e. The observed mass of 11.312 Da deviates from the average calculated mass of 11.236 Da listed in the database. However, a mass of 11.315 Da for Histone H4 was also reported in a study on transitional cell carcinoma of the bladder (Tolson et al., 2006). These authors also found that the Histone H4 peak was of similar abundance in tumor samples and normal controls. The protein with mass of 11.043 Da was identified as Cystatin A by ESI-MS/MS (Table 1). Whether the deviation from the calculated mass of Cystatin A (11.006 Da) is due solely to a inaccuracy of the SELDI-equipment or whether some post-translational modifications are involved, is not clear at present. The downregulation of Cystatin A in many tumors was confirmed by immunohistochemistry on tissue microarrays (data not shown). The protein peaks at 9.973 and 11.744 Da were processed in similar fashion and identified as Acyl-CoA binding protein and Calgizzarin (S100A11), respectively. Again, some deviations from the calculated masses were observed.

Figure 3.

From the gel in A, bands B and B' were excised and submitted to passive elution before tryptic digest and protein identification. B and B' were analysed on a normal phase (NP20) array by SELDI-TOF-MS. The peaks in C and C', showing masses of 11.313 and 11.043 Da, were identical to the original protein peaks of interest in SELDI-TOF-MS. The tryptic digest D from peak 11.312 Da generated the fingerprint in E, identifying the protein as Histone H4.

Full figure and legend (94K)

Table 1 - Summary of proteins identified by mass spectrometry.

Full table

A peptide peak at 3.700 Da, upregulated in HNSCC, was enriched by reverse phase chromatography. In this special case, direct post source decay (PSD) MALDI-MS identified this peak as a C-terminal fragment of hemoglobin . Therefore, we also aimed at the identification of the intact - and -hemoglobin subunits which we assumed to be represented by two prominent peaks with m/z of 15.150 and 15.885 in the SELDI-spectra. Indeed, the two hemoglobin subunits were identified by gel-digests from bands running at approximately 15 kDa. The differential expression of the different hemoglobin peaks are presented in Table 2. The upregulation of the C-terminal fragment of hemoglobin correlates well with the downregulation of the intact -chain, whereas the -chain of hemoglobin remained constant. Protein identification of a peak with a m/z of 27.805 downregulated in HNSCC, employed enrichment by metal affinity chromatography through an imidazole gradient from both tumor and mucosa samples, as described previously (Roesch Ely et al., 2005). A gel band at 28 kDa showed reduced intensity in tumor samples as compared to mucosal samples. In gel-trypsin digestion followed by SELDI-MS/MS identified this peak unambiguously as Stratifin (14-3-3 sigma).

Table 2 - Differential regulation of Hemoglobin related peaks.

Full table

Three peaks around 3.500 Da (m/z values of 3.442, 3.371 and 3.486) showed strongly increased expression in many HNSCC samples but also in a number of tumor-adjacent and -distant biopsies (Figure 1). It was reasonable to assume that these were the -defensins 1, 2 and 3 which had already been described in several proteomic studies (Vlahou et al., 2001; Albrethsen et al., 2005; Buhimschi et al., 2005). The assignment of this peak triplet to -defensins was confirmed by analysis of the tryptic digest of the gel bands taken from the low mass range followed by MALDI-MS (Table 1). Table 1 summarises the MALDI-MS, ESI-MS/MS and SELDI-MS/MS results for the proteins identified in this study.

In Figure 4, the changes in the expression of the proteins identified in this study from healthy mucosa to HNSCC are shown. Stratifin (14-3-3-sigma), Cystatin A (CYTA), Calgizzarin (S10AB) showed a major drop, and the -defensins a major increase in their respective protein levels from tumor-distant to tumor-adjacent mucosa. On the other hand, the Acyl-CoA-binding protein (ACBP) and the intact -hemoglobin (HBB) dropped in intensity in the late step between tumor-adjacent mucosa and HNSCC, and the concomitant increase in the C-terminal fragment (frag.) of HBB also occurred at this late stage.

Figure 4.

Changes in the expression of the proteins identified in this study from healthy mucosa to HNSCC. Y axis refers to mean intensities as calculated in the Biomarker Wizard software for each peak, X axis lists the proteins by their Swiss-Prot entry names. Note that 1433S (Stratifin, 14-3-3-sigma) is of low intensity and abundance and therefore given at different scale.

Full figure and legend (57K)

The successive alterations in expression patterns from healthy normal mucosa to HNSCC as seen in the entire set of peaks are illustrated in Supplementary Figure 1. Overall, a gradual reduction or increase of mean expression in the four different sample groups is apparent along the hypothetical axis normal/tumor-distant/tumor-adjacent mucosa/HNSCC.

Supervised class comparison and class prediction analysis

The optimized predictor built from the healthy mucosae and HNSCC (the training set) was applied to the entire set of 303 samples including the tumor-distant and tumor-adjacent mucosae (=test set). If any mucosa received >14 votes of a maximum of 20 as being a mucosa, or any HNSCC received >14 votes as being a HNSCC, this was taken as correct classification. 13–8 votes were taken as a failure to classify ('not predicted'), and <8 votes were taken as classified to the opposite sample group. Hence, failure or opposite classification of a mucosa (distant or adjacent to a tumor) was taken to indicate an aberrant premalignant expression pattern significantly deviating from the normal expression pattern. The results are summarized in Table 3 which presents the mean and median numbers of 'mucosa votes' or 'HNSCC votes' assigned to the four sample groups. The predictor allowed excellent classification of both healthy mucosa and HNSCC samples (94.5 and 92.9% cases correctly classified, respectively). The four normal control mucosae voted to show aberrant expression patterns were examined by histology and displayed signs of inflammation, whereas the single tumor voted to be a mucosa was a typical tumor histologically. In the tumor-adjacent and tumor-distant mucosa groups, dramatic changes were observed: only 59.6% of the tumor-distant samples were classified as normal, 27.3% as HNSCC. Strikingly, 72.2% of the tumor-adjacent mucosae were predicted as tumors, 16.7% were not predicted and only 11.1% as normal. Comparison of the protein profiles in the tumor-distant samples with clinical outcome of 32 patients revealed that the aberrant profiles (not predicted and voted as tumor combined) were significantly associated with tumor relapse events (P=0.018; Fisher's exact test, two-tailed; Table 4).

Table 3 - Summary of results from supervised classification and prediction analysis.

Full table

Table 4 - Correlation of relapse events with proteomic votes for tumor-distant mucosal biopsies.

Full table

Validation of differential -defensin expression by immunohistochemistry on tissue microarrays

IHC using the avidin–biotin–horseradish peroxidase complex was performed on formalin-fixed paraffin-embedded tissues, using a series of conventional sections and a tissue microarray containing a total of 180 HNSCC tissue cores and some healthy mucosae. Immunohistochemistry revealed a similar variability as was seen in the proteomic profiles. Examples of the diverse staining patterns observed are shown in Figure 5. In A, one of the rare tumors with little defensins in both SELDI-TOF MS and IHC is shown, whereas the tumor in B displayed strong staining in vessels and in the tumor cells. -Defensins were frequently found in endothelial cells and in stromal neutrophilic granulocytic cells as indicated by the granular nucleus displayed by these cells (Figure 5c). The consecutive section was stained with von Willebrand factor, confirming defensin expression in endothelial cells in small capillaries (Figure 5d).

Figure 5.

Immunohistochemical analysis of -defensin protein expression in HNSCC. Weak staining in tumor region (a) compared with strong staining inside vessels and tumor cells (b). Positive staining is frequently found in stromal neutrophilic cells and in endothelial cells (c) which are also stained with von Willenbrandt factor (d).

Full figure and legend (336K)

Discussion

In this study, we were able to document dramatic tumor-related expression changes in clinical mucosa samples, although these biopsies did not show abnormal histology. Surprisingly, most of the tumor-adjacent mucosae (72%) were classified as tumors due to highly aberrant expression patterns. Such severe changes were also observed in tumor-distant mucosae, but at lower frequency (27% were classified as tumors). We have previously described the calgranulins A and B (S100A8 and S100A9) and annexin 1 as being very strongly downregulated in HNSCC. We have now identified several more proteins which were also very significantly downregulated: Stratifin (14-3-3 sigma), the Cysteine proteinase inhibitor Cystatin A, the hemoglobin subunit, ACBP, and Calgizzarin/S100A11 (Table 1). This member of the S100 protein family was reduced with much lower significance in HNSCC and had appeared relatively stable in another recent study (Melle et al., 2004). Therefore, its significance awaits further investigation. Downregulated expression of the other proteins identified was associated with extremely high P-values (P<10^-20), similar to the Calgranulins A and B (Roesch Ely et al., 2005).

The human neutrophil -defensins 1–3 and a C-terminal fragment of -hemoglobin were identified as being strongly upregulated in HNSCC (Tables 1 and 2), again affording high statistical significance (ROC, AUC above 0.9; data not shown).

Histone H4 and the subunit of hemoglobin revealed only insignificant alterations in protein levels from healthy mucosa to HNSCC. These proteins can be used as markers to monitor the integrity of tissue lysates derived from squamous epithelia and carcinomas.

Figure 4 demonstrates that the proteins identified should be very useful biomarkers also for the analysis of premalignant changes in mucosal samples from cancer patients. Stratifin (1433S), Cystatin A (CYTA), Calgizzarin (S10AB) as well as the previously identified Calgranulins A and B (see Supplementary Figure 3) all dropped in their respective protein levels from tumor-distant to tumor-adjacent, that is, they represent early changes occurring during field cancerization. Similarly, the upregulated -defensins 1–3 changed most dramatically from tumor-distant to tumor-adjacent.

In contrast, the C-terminal fragment of -hemoglobin specifically increased late in progression from tumor-adjacent to HNSCC, concomitantly to the sharp drop of the intact -hemoglobin at the same stage, while the -hemoglobin showed a remarkably stable expression during progression. Thus, -hemoglobin and its C-terminal fragment may provide an excellent novel tumor marker.

The candidate biomarkers described here relate to various biochemical aspects of the multistep carcinogenic process. Stratifin is a protein kinase C inhibitor and important protein in the p53-regulated cell cycle check point at the G2/M phase transition (Dellambra et al., 2000). It has been shown to regulate senescence of, and its downregulation by promoter hypermethylation to contribute to the immortalization of keratinocytes (Hermeking et al., 1997). Interestingly, silencing of stratifin in oral squamous cell carcinomas occurred at very similar frequency as that of p16^INK4a and was significantly less frequent in HPV positive oral tumors (Gasco et al., 2002), which are also characterized by p16^INK4a positivity (Andl et al., 1998).

The cysteine proteinase inhibitors Cystatin A and B are likely relevant for intracellular protein degradation as well as degradation of the extracellular matrix by cathepsins (for review, see Abrahamson et al. (2003)). Cystatin B has also been found downregulated in esophageal squamous cell carcinoma, consistent with our results (Shiraishi et al., 1998). We are continuing our efforts to identify further members of the cystatin family of cathepsin inhibitors and also the cathepsins, and to examine whether they are also differentially regulated in HNSCC and in field cancerization. Likely, the altered balance between the whole complement of cystatins and cathepsins present, will be more informative for the understanding of tumor cell invasion and metastasis, than altered expression levels by individual members. This was shown in a study of prostatic carcinoma where the ratio of cathepsin B to cystatin A (stefin A) was predictive for metastasis if shifted in favor of the cathepsin B (Shiraishi et al., 1998; Sinha et al., 2002).

The ACBP, a member of one of several fatty ACBP families, is a highly conserved polypeptide with hallmarks of a housekeeping gene. It is expressed in various organs and implicated in the regulation of multiple processes. ACBP expression was found to be altered in brain tumors, ovarian tumors and in colonic and prostatic cell lines, but has not been studied in squamous cell carcinomas, and the significance of our observation is unclear at present. However, ACBP could be of great interest as a biomarker because of its involvement in both lipid and glucose metabolism. Reduced expression of ACBP might contribute to the high glucose uptake and energy consumption of squamous cell carcinomas (Alho et al., 1995; Petrescu et al., 2003).

Finally, we want to discuss the group of small peptides identified as the -defensins 1–3 (HNP1–3), together with the previously identified calgranulins A and B and annexin 1. The -defensins 1–3 are major constituents of the azurophilic granules of neutrophils. They are normally not expressed in epithelia. It is believed that they play an important role in the mucosal innate immune defense against a number of infectious diseases (reviewed in Ganz (2003)), and they are regarded as general markers for inflammation-related diseases, which includes epithelial cancers. Accordingly, overexpression of -defensins in epithelial cancers is not a novel observation (Vlahou et al., 2001; Muller et al., 2002; Albrethsen et al., 2005). In HNSCC, increased expression of -defensins 1–3 has been described in neutrophils that infiltrate HNSCCs (Lundy et al., 2004). We have confirmed this patchy expression pattern in the majority of the tumors, but also in a limited number of tumor-distant mucosae. Strong expression was frequently also seen in capillaries directly underneath the mucosa, as confirmed by the co-expression of von Willebrandt factor (vWF) which is an established vascularity marker. However, we also observed novel intriguing expression patterns in this study. Tumor cells, and in rare cases even some tumor-distant mucosa cells in the basal and parabasal layers, were detected to display de novo defensin expression themselves (Figure 5). Healthy mucosa showed only very little defensin expression which was strictly restricted to submucosal neutrophil granulocytes. It is likely that the defensins support angiogenesis already during premalignancy, in agreement with similar findings in experimental premalignancy (Smith-McCune, 1997). Thus, the defensins represent an important link between inflammation, angiogenesis and cancer (Chavakis et al., 2004). Our findings on the calgranulins A and B further underline this concept: in contrast to the defensins, the calgranulins are expressed in both epithelia and neutrophils but likely with different activities. In squamous cell carcinomas, these closely interacting proteins may be viewed primarily in the context of altered Ca-dependent gene expression resulting in inhibited or retarded differentiation and, in particular, in loss of proapoptotic activity. In the neutrophils, the abundantly expressed calgranulins show strong proinflammatory activity similar to the defensins (Kerkhoff et al., 1998). Intriguingly, our IHC studies confirmed the extensive reduction in calgranulin levels in the tumors and surrounding epithelia, but at the same time showed retained or even enhanced calgranulin levels in stromal neutrophils (Roesch Ely et al., 2005).

Importance for clinical screening and monitoring

For clinical diagnosis and prevention of primary and recurrent disease, it is highly relevant to detect and characterize genetically altered fields of premalignant cells. However, diagnostic protocols that allow their detection in daily routine practice do not yet exist, and clinical examination alone is insufficient to predict which lesions might progress. Histologically recognizable premalignant lesions do not regularly appear, and most secondary tumors in HNSCC patients seem to arise in inconspicuous mucosa that shows no signs for a tumorigenic risk potential. Thus, it would be very helpful to identify biomarkers (or sets of markers) that could be used for a more reliable cancer-risk assessment. Individuals recognized to be at risk could benefit from regular monitoring, dietary advice and recruitment to chemoprevention studies which have already demonstrated to significantly lower the incidence of second primaries in HNSCC patients. In this study, the retrospective comparison of the protein profiles of tumor-distant mucosae with the documented clinical outcome for 32 patients revealed that the aberrant profiles were significantly associated with tumor relapse (P=0.018; Fisher's exact test, two-tailed).

The SELDI-TOF MS procedure to analyse clinical biopsies described in this paper represents a useful novel diagnostic tool: it requires only little biopsy material and only little sample processing time. It allows the reproducible analysis of a large number of samples in a short period of time and is suitable to support and complement large-scale clinical studies. Simple standard operating procedures can be defined for routine application. The data sets generated can be directly used to build mathematical algorithms for the prediction and classification of samples and for the comparison with clinical parameters. The approach is comparable to cDNA microarray analysis, but yields results faster and with less expense. The same statistical approaches and software packages can be used to analyse both kinds of data sets. Although each individual analysis generates only a limited number of reproducibly detectable peaks, this is fully sufficient for defining a mathematical model. Last not least, the resulting spectra are greatly influenced by the use of different methods used for the analyses. This includes, for example, different ProteinChip arrays or alternative lysis buffers used to prepare the samples for analysis. This will allow the introduction of even greater experimental variability and expands the number of potentially detectable biomarkers.

We conclude that proteomic profiling by SELDI-TOF-MS outperforms and supplements histopathological diagnosis. It allows the identification of mucosae with premalignant alterations despite inconspicuous appearance, independent of single molecular markers such as p53 mutations. We believe that the large-scale analysis of clinical biopsies by this proteomic-based method has much potential for further development into routine application in diagnosis, risk assessment, individualized therapy and prognosis.

Materials and methods

Patients and tissue samples

A total of 303 clinical biopsies were analysed, consisting of 113 HNSCC samples (obtained from 98 patients), 73 healthy, 99 tumor-distant and 18 tumor-adjacent squamous mucosae. Comprehensive clinical data were available for all patients. Untreated primary tumors were from the oropharynx (n=42), larynx (n=16), hypopharynx (n=27), oral cavity (n=7) and nasopharynx (n=2). The remainder were lymph node metastases (n=8) and recurrences (n=11). HNSCC were selected on the basis of the percentage of tumor cells present in H&E-stained sections. Only tumor samples containing more than 40% of tumor cells (range 40–90%) were used and no microdissection was carried out. Normal healthy control mucosae from nontumor patients were collected from individuals undergoing reconstructive operation in the upper aerodigestive tract for reasons other than HNSCC (e.g. sleep apnea syndrome, septo-rhinoplastic surgery). Tissue lysis in the presence of a protease inhibitor cocktail containing the peptide aprotinin, was performed as recently described in detail (Roesch Ely et al., 2005).

Protein profiling by SELDI-TOF MS

After lysis, samples were processed and analysed on IMAC30 ProteinChip arrays in the ProteinChip Reader (PBSIIc) with version 3.2 ProteinChip Software (Ciphergen Biosystems, Fremont, CA, USA) as described (Roesch Ely et al., 2005). All spectra were combined to a single experiment set and analysed using the Biomarker Wizard Program (Version 3.0, Ciphergen Biosystems Inc.). For sample group statistics and for identifying differentially expressed peaks/proteins, a modified nonparametric Mann–Whitney U-Test was applied. Heat maps were generated from normalized data by unsupervised hierarchical clustering and unweighted Spearman log rank test, using the GeneSpring 6.1 software package (Silicon Genetics Inc., San Carlos, CA, USA).

Quality control of protein profiling

The overall strategy for obtaining a high-quality data set was to analyse all samples of the study within the shortest possible period of time, using the same batch of reagents, ProteinChip arrays and using the same instrument setting (chip- and spot protocols) throughout. The reproducibility of the profiling data was monitored twofold: for part of the cohort, tissue lysates were loaded in duplicates, using neighboring as well as distant spot positions. After completion of profiling, all samples were rerun on different chip positions. All tumors and tumor-adjacent, and all tumor-distant and normal mucosal biopsies exceeding approximately 10 mg of tissue mass were histologically examined by H&E staining of frozen sections, and inappropriate biopsies were precluded from the analysis. Upon normalization, a further part of the spectra was also eliminated, leaving a total of 303 samples. This study followed criteria from guidelines in publication of peptide and protein identification (Carr et al., 2004; Wilkins et al., 2006).

Supervised class comparison and class prediction analysis

These analyses followed and took into account recently published recommendations and notes of caution (Simon et al., 2003). The 303 individual spectra were divided into four sample groups ('classes'): HNSCC and healthy control mucosae were used as the training set; the additional tumor-adjacent mucosae and tumor-distant mucosae were used as the test set. In order to construct a predictor (a k-nearest-neighbors classifier algorithm) to optimally distinguish between the healthy mucosa and HNSCC samples, all differentially expressed peaks were individually examined and ranked on their power to discriminate between both classes. The optimal number of peaks to be employed in the predictor was reached at approximately 40 top-ranked protein peaks, using a decision threshold of P<0.2 and a 20-fold cross-validation procedure at each expansion round. This predictor and validated classifier was then used to classify the remaining 'test' set of tumor-adjacent (n=18) and tumor-distant mucosae (n=99), and the entire experiment set (n=303). The number of votes for either mucosa or HNSCC was taken as a measure of how well a sample was assigned. Misclassification or failure to classify was interpreted as aberrant expression patterns that indicate premalignant change.

Protein purification strategy

Proteins of interest for identification were enriched either by IMAC chromatography as described (Roesch Ely et al., 2005) or by a combination of reverse phase chromatography and size fractionation. For reverse phase chromatography 50 l of reverse phase RPC Poly-beads (BioSepra, Paris, France) were equilibrated with 10% ACN/0.1% TFA. The tissue lysates were adjusted to a final concentration of 10% ACN/0.5% TFA and incubated with RPC beads for 45 min (volume ratio of beads/sample 1:4). The beads were collected by centrifugation and the proteins were eluted in 80 l aliquots using an acetonitrile gradient (six increment steps from 10 to 60% of ACN in 0.1% TFA). The proteins were further enriched using Microcon filtration cutoff membranes (YM30, YM10, Millipore). The fractions obtained were analysed on IMAC30 as well as on NP20 (hydrophilic normal phase) arrays to monitor in which fractions the proteins of interests were enriched. As NP20 is a nonselective array, it shows a general profile of the proteins in a sample and thus gives information on the relative enrichment upon fractionation. As final purification step SDS–PAGE was applied. The fractions were loaded on precast 12 and 16% Tris-tricine polyacrylamide gels or 12% NuPAGE gels (Novex Invitrogen, Paisley, UK). Gels were stained overnight using Colloidal Coomassie blue (Novex Invitrogen) and destained with water. Proteins in the low molecular weight range were passively eluted from the gel bands as described by Le Bihan et al. (2004) using 30 l of a solution of 50% formic acid, 25% ACN, 15% isopropanol, 10% H₂0. A small amount of the passively eluted proteins (2 l) were analysed on a NP20 array. This allowed the reanalysis of an eluted protein on a NP20 chip and thus confirmation its m/z identity with the protein peak in the original SELDI-TOF MS profile. The remaining 28 l were used for tryptic digestion and identification by peptide mass fingerprinting and MS-MS analysis.

The dried protein eluates were resolved in 15 l of 10 ng/l sequencing grade trypsin (Sigma, Deisenhofen, Germany) in 50 mM ammonium bicarbonate and subjected to tryptic digest for 3 h at 37°C. In-gel digestion of proteins was performed overnight with 200 ng trypsin in 20 l of 50 mM ammonium bicarbonate.

Protein identification

The tryptic digests were analysed either by SELDI Q-TOF tandem mass spectrometry on a QStar Pulsar instrument (ABI Applied Biosystems, Darmstadt, Germany) equipped with a PCI 1000 ProteinChip Interface (Ciphergen Biosystems Inc.) or by peptide mass fingerprinting on a MALDI-TOF mass spectrometer (Reflex II, Bruker-Daltonik GmbH, Germany) as decribed (Roesch Ely et al., 2005).

nanoLC-ESI MS/MS

When more than two proteins were detected in the peptide mass fingerprint the digest was further analysed by nanoLC-ESI MS/MS on a quadrupole time of flight mass spectrometer (QTof Ultima, Waters GmbH, Germany) equipped with a nano liquid chromatography system (CapLC, Waters GmbH, Germany).

Tissue microarray IHC (TMA-IHC)

Tissue microarrays were prepared as described (Freier et al., 2003). IHC using the mouse monoclonal antibody against neutrophil -defensins 1–3 (Hycult Biotechnology, Germany) was carried out on the TMAs as described recently (Roesch Ely et al., 2005).

References

1. Abrahamson M, Alvarez-Fernandez M, Nathanson CM. (2003). Biochem Soc Symp 70: 179–199. | PubMed | ChemPort |
2. Ai H, Barrera JE, Meyers AD, Shroyer KR, Varella-Garcia M. (2001). Laryngoscope 111: 1853–1858. | Article | PubMed | ChemPort |
3. Ai H, Barrera JE, Pan Z, Meyers AD, Varella-Garcia M. (1999). Mutat Res 439: 223–232. | PubMed | ChemPort |
4. Albrethsen J, Bogebo R, Gammeltoft S, Olsen J, Winther B, Raskov H. (2005). BMC Cancer 5: 8. | Article | PubMed | ChemPort |
5. Alho H, Kolmer M, Harjuntausta T, Helen P. (1995). Cell Growth Differ 6: 309–314. | PubMed | ChemPort |
6. Andl T, Kahn T, Pfuhl A, Nicola T, Erber R, Conradt C et al. (1998). Cancer Res 58: 5–13. | PubMed | ISI | ChemPort |
7. Bedi GC, Westra WH, Gabrielson E, Koch W, Sidransky D. (1996). Cancer Res 56: 2484–2487. | PubMed | ISI | ChemPort |
8. Braakhuis BJ, Leemans CR, Brakenhoff RH. (2005). Semin Cancer Biol 15: 113–120. | Article | PubMed |
9. Braakhuis BJ, Tabor MP, Kummer JA, Leemans CR, Brakenhoff RH. (2003). Cancer Res 63: 1727–1730. | PubMed | ISI | ChemPort |
10. Braakhuis BJ, Tabor MP, Leemans CR, van der Waal I, Snow GB, Brakenhoff RH. (2002). Head Neck 24: 198–206. | Article | PubMed | ISI |
11. Buhimschi IA, Christner R, Buhimschi CS. (2005). Bjog 112: 173–181. | PubMed | ChemPort |
12. Califano J, van der Riet P, Westra W, Nawroz H, Clayman G, Piantadosi S et al. (1996). Cancer Res 56: 2488–2492. | PubMed | ISI | ChemPort |
13. Carr S, Aebersold R, Baldwin M, Burlingame A, Clauser K, Nesvizhskii A. (2004). Mol Cell Proteomics 3: 531–533. | Article | PubMed | ISI | ChemPort |
14. Chavakis T, Cines DB, Rhee JS, Liang OD, Schubert U, Hammes HP et al. (2004). FASEB J 18: 1306–1308. | PubMed | ChemPort |
15. Dellambra E, Golisano O, Bondanza S, Siviero E, Lacal P, Molinari M et al. (2000). J Cell Biol 149: 1117–1130. | Article | PubMed | ISI | ChemPort |
16. Freier K, Joos S, Flechtenmacher C, Devens F, Benner A, Bosch FX et al. (2003). Cancer Res 63: 1179–1182. | PubMed | ChemPort |
17. Ganz T. (2003). Nat Rev Immunol 3: 710–720. | Article | PubMed | ISI | ChemPort |
18. Garcia SB, Park HS, Novelli M, Wright NA. (1999). J Pathol 187: 61–81. | Article | PubMed | ISI | ChemPort |
19. Gasco M, Bell AK, Heath V, Sullivan A, Smith P, Hiller L et al. (2002). Cancer Res 62: 2072–2076. | PubMed | ISI | ChemPort |
20. Hegedus CM, Gunn L, Skibola CF, Zhang L, Shiao R, Fu S et al. (2005). Leukemia 19: 1713–1718. | Article | PubMed | ChemPort |
21. Hermeking H, Lengauer C, Polyak K, He TC, Zhang L, Thiagalingam S et al. (1997). Mol Cell 1: 3–11. | Article | PubMed | ISI | ChemPort |
22. Homann N, Nees M, Conradt C, Dietz A, Weidauer H, Maier H et al. (2001). Clin Cancer Res 7: 290–296. | PubMed | ChemPort |
23. Kerkhoff C, Klempt M, Sorg C. (1998). Biochim Biophys Acta 1448: 200–211. | Article | PubMed | ISI | ChemPort |
24. Le Bihan MC, Tarelli E, Coulton GR. (2004). Proteomics 4: 2739–2753. | Article | PubMed | ChemPort |
25. Lundy FT, Orr DF, Gallagher JR, Maxwell P, Shaw C, Napier SS et al. (2004). Oral Oncol 40: 139–144. | Article | PubMed | ChemPort |
26. Melle C, Ernst G, Schimmel B, Bleul A, Koscielny S, Wiesner A et al. (2004). Cancer Res 64: 4099–4104. | Article | PubMed | ChemPort |
27. Muller CA, Markovic-Lipkovski J, Klatt T, Gamper J, Schwarz G, Beck H et al. (2002). Am J Pathol 160: 1311–1324. | PubMed | ISI | ChemPort |
28. Nees M, Homann N, Discher H, Andl T, Enders C, Herold-Mende C et al. (1993). Cancer Res 53: 4189–4196. | PubMed | ChemPort |
29. Partridge M, Pateromichelakis S, Phillips E, Emilion GG, A'Hern RP, Langdon JD. (2000). Cancer Res 60: 3893–3898. | PubMed | ISI | ChemPort |
30. Petrescu AD, Payne HR, Boedecker A, Chao H, Hertz R, Bar-Tana J et al. (2003). J Biol Chem 278: 51813–51824. | Article | PubMed | ChemPort |
31. Prevo LJ, Sanchez CA, Galipeau PC, Reid BJ. (1999). Cancer Res 59: 4784–4787. | PubMed | ISI | ChemPort |
32. Roesch Ely M, Nees M, Karsai S, Magele I, Bogumil R, Vorderwulbecke S et al. (2005). Eur J Cell Biol 84: 431–444. | PubMed |
33. Shiraishi T, Mori M, Tanaka S, Sugimachi K, Akiyoshi T. (1998). Int J Cancer 79: 175–178. | Article | PubMed | ChemPort |
34. Simon R, Radmacher MD, Dobbin K, McShane LM. (2003). J Natl Cancer Inst 95: 14–18. | Article | PubMed | ChemPort |
35. Sinha AA, Quast BJ, Wilson MJ, Fernandes ET, Reddy PK, Ewing SL et al. (2002). Cancer 94: 3141–3149. | Article | PubMed | ChemPort |
36. Slaughter DP, Southwick HW, Smejkal W. (1953). Cancer 6: 963–968. | Article | PubMed | ISI | ChemPort |
37. Smith-McCune K. (1997). Obstet Gynecol 89: 482–483. | Article | PubMed | ChemPort |
38. Soder AI, Hopman AH, Ramaekers FC, Conradt C, Bosch FX. (1995). Cancer Res 55: 5030–5037. | PubMed | ISI | ChemPort |
39. Tabor MP, Brakenhoff RH, Ruijter-Schippers HJ, Kummer JA, Leemans CR, Braakhuis BJ. (2004). Clin Cancer Res 10: 3607–3613. | Article | PubMed | ISI | ChemPort |
40. Tabor MP, Brakenhoff RH, van Houten VM, Kummer JA, Snel MH, Snijders PJ et al. (2001). Clin Cancer Res 7: 1523–1532. | PubMed | ISI | ChemPort |
41. Tolson JP, Flad T, Gnau V, Dihazi H, Hennenlotter J, Beck A et al. (2006). Proteomics 6: 697–708. | Article | PubMed | ChemPort |
42. Vlahou A, Schellhammer PF, Mendrinos S, Patel K, Kondylis FI, Gong L et al. (2001). Am J Pathol 158: 1491–1502. | PubMed | ISI | ChemPort |
43. Waridel F, Estreicher A, Bron L, Flaman JM, Fontolliet C, Monnier P et al. (1997). Oncogene 14: 163–169. | Article | PubMed | ChemPort |
44. Wiesner A. (2004). Curr Pharm Biotechnol 5: 45–67. | Article | PubMed | ChemPort |
45. Wilkins MR, Appel RD, Van Eyk JE, Chung MC, Gorg A, Hecker M et al. (2006). Proteomics 6: 4–8. | Article | PubMed | ChemPort |
46. Wolf C, Flechtenmacher C, Dietz A, Weidauer H, Abel U, Maier H et al. (2004). Laryngoscope 114: 698–704. | Article | PubMed |
47. Wright Jr GL. (2002). Expert Rev Mol Diagn 2: 549–563. | Article | PubMed | ChemPort |

Acknowledgements

Grant support: We specially thank the CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior) for financial support to M Roesch-Ely. We greatly appreciate the help from our medical staff in collecting and processing the tissue specimens, and Dr Christa Flechtenmacher, Pathologisches Institut der Universität Heidelberg, for histopathological assessment and fruitful discusssions. We thank Antje Schuhmann and Nataly Henfling for excellent technical support. Equipment was financed via the 'Hochschulbau-Förderungsgesetz'. This study was in part supported by the 'Forschungsförderungsprogramm der Medizinischen Fakultät Heidelberg', Grant No. 007 and a grant from the National Genome Research Network to MS (Förderkennzeichen 01GS0460).

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).