Introduction

Head and neck squamous cell carcinomas (HNSCCs) comprise a group of neoplasms developing in the nose, oral cavity, pharynx, and larynx [1] and make up more than 90% of neoplasms of the head and neck [2]. The long-term survival of patients with HNSCC treated with curative intent has not improved significantly over recent decades [2] due to lymph node metastasis, local tumour recurrence, and mainly because of distant metastasis (met-HNSCC), for example to the lung [3].

The differentiation between a pulmonary metastasis of a primary known HNSCC and a secondary primary tumour (SPT) of the lung after a prior HNSCC is difficult. More than two-thirds of SPTs are squamous cell carcinomas [4, 5] and both tumours have similar patterns of genetic mutations and pathological molecular alterations [5-8]. No biomarker has yet been established for their differential diagnosis. Therefore, the differentiation is mainly based on radiological and clinical criteria [5, 9, 10].

However, the distinction between met-HNSCC and SPT is crucially important for further prognosis and the decision between a curative and palliative treatment regimen. We therefore aimed to identify, by high-resolution mass spectrometry (MS)-based proteomics, differentially expressed proteins that can be used as immunohistochemical markers to distinguish between squamous cell lung carcinoma (SQCLC) and HNSCC.

Methods Proteomic analysis of squamous cell carcinoma cell lines

The human tumour cell line NCI-H2170 (CRL-5928; American Type Culture Collection, Manassas, VA, USA) originates from a squamous cell carcinoma of the lung. The PCI-13.1 cell line (Department of Pathology, UPMC, PA, USA) derives from HPV-negative squamous cell carcinoma of the oropharynx. Labelling with stable isotopes in cell culture and proteomic comparison were conducted as described previously [8, 11, 12]. In brief, NCI-H2170 and PCI3.1 cells were cultured in RPMI 1640 medium supplemented with 10% dialysed foetal calf serum (Invitrogen, Waltham, MA, USA), 4 mm glutamine, antibiotics, 0.115 mm l-arginine-13C6 and 0.275 mm l-lysine-4,4,5,5-D4 (Eurisotop, St-Aubin, France), or equimolar levels of the corresponding non-labelled (light) amino acids (Merck Millipore, Burlington, MA, USA) for at least 10 cell cycles. Labelled cells were lysed in 0.5% Nonidet P-40 buffer containing 50 mm Tris/HCl, pH 7.8, 150 mm NaCl, 1 mm Na3VO4, 1 mm NaF, 0.2% lauryl maltoside, and protease inhibitors (Complete, Roche, Basel, Switzerland). Equal amounts of proteins of light-labelled NCI-H2170 were mixed with heavy-labelled PCI-13.1 and vice versa to obtain two biological replicates with different stable isotope labelling by amino acids in cell culture (SILAC). Afterwards, they were separated by 1D-PAGE (4–12% NuPAGE Bis-Tris Gel; Invitrogen) and stained with Coomassie brilliant blue. Next, the stained gel was separated in 23 slices and each one was reduced with 10 mm DTT for 55 min at 56 °C, alkylated with 55 mm iodoacetamide (IAA) for 20 min at 26 °C, and digested with modified trypsin (Promega, Madison, WI, USA) overnight at 37 °C. Resulting peptides were separated by a C18 precolumn (2.5 cm, 360 μm o.d., 100 μm i.d., Reprosil-Pur 120 Å, 5 μm, C18-AQ; Dr. Maisch GmbH, Ammerbuch, Germany) at a flow rate of 10 μl/min and a C18 capillary column (20 cm, 360 μm o.d., 75 μm i.d., Reprosil-Pur 120 Å, 3 μm, C18-AQ; Dr. Maisch GmbH) at a flow rate of 300 nl/min, with a gradient of acetonitrile ranging from 5 to 35% in 0.1% formic acid for 90 min using an Proxeon nano LC coupled to an Q Exactive mass spectrometer (Thermo Electron, Waltham, MA, USA). MS conditions were as follows: spray voltage, 1.8 kV; heated capillary temperature, 270 °C; and normalised collision energy, 28. The mass spectrometer automatically switched between MS and MS/MS acquisitions (data-dependent mode). Survey MS spectra were acquired in the Orbitrap (m/z 350–1,600) with the resolution set to 70,000 at m/z 200. The 15 most intense ions were sequentially isolated for higher-energy collisional dissociation (HCD) MS/MS fragmentation and detection. Raw data were analysed with MaxQuant (version 1.3.0.5) using Uniprot human as a sequence database. Up to two missed cleavages of trypsin were allowed. Oxidised methionine was searched as variable modification and cysteine carbamidomethylation as fixed modification. The modifications corresponding to arginine and lysine labelled with heavy stable isotopes were handled as fixed modifications. The false positive rate was set to 1% at the peptide level, the false discovery rate was set to 1% at the protein level, and the minimum required peptide length was set to six amino acids.

Resulting data from MaxQuant analysis of the raw data were further analysed by Perseus (version 1.5.2.6; Max Planck Institute for Biochemistry, Martinsried, Germany). Ratios of intensity of heavy- and light-labelled proteins corresponding to either of the two cell lines were log2-transformed and the medians of the two biological replicates were calculated. Intensity values were log10-transformed. Next, an outlier significance score for SILAC ratios depending on intensity values (significance B in Perseus, see citation for more details [13]) for every protein was calculated and fold discovery rate was corrected by Benjamini–Hochberg correction. The complete results of the statistical analysis with Perseus including the number of unique peptides can be found in supplementary material, Table S1.

Tissue samples

The patient samples were collected at the University Medical Centre in Göttingen, Germany. In total, 98 SQCLC, 96 HNSCC, and 12 lung tumours with squamous cell carcinoma histology of unknown origin after primary HNSCC were included in this study. The samples of the main cohort (98 SQCLC, 96 HNSCC) derived from oncological resections. Approval for using the human patient material in this study was obtained from the Ethics Committee of the University Medical Centre Göttingen (vote no. 07/06/09, updated in April 2018). All procedures were conducted in accordance with the Declaration of Helsinki and institutional, state, and federal guidelines.

Immunohistochemistry

Immunohistochemical reactions were performed on 2-μm formalin-fixed and paraffin-embedded tissue sections, as described previously [14]. Antigen retrieval was carried out at 97 °C in citrate buffer (pH 6) or EDTA buffer (pH 9). The antibodies and dilutions used are listed in supplementary material, Table S2. The sections were incubated with a ready-to-use horseradish peroxidase-labelled secondary antibody at room temperature for 25 min (anti-rabbit/mouse, produced in goat; REAL EnVision Detection System; Dako, Agilent Technologies, Waldbronn, Germany). The substrate DAB + Chromogen system produces a brown end product, and is applied to visualise the target antigen (REAL DAB + Chromogen; Dako, Agilent Technologies). Tissue samples were counterstained with Mayer's haematoxylin (Dako, Agilent Technologies) for 8 min and analysed using light microscopy.

Two independent investigators evaluated all stained tissue sections by using an immunoreactivity staining score (IRS) as described previously [12]. The percentage of positively stained cells was first classified using a 0–4 scoring system: score 0 = 0% positive cells, score 1 = less than 10% positive cells, score 2 = 10–50% positive cells, score 3 = 51–80% positive cells, and score 4 = >80% positive cells. The intensity of staining was evaluated on a four-tiered scale (0 = negative, 1 = weak, 2 = intermediate, and 3 = strong). Afterwards, the scores of intensity and staining were multiplied and the mean value per patient was calculated, where 0–1 point was interpreted as negative, 2–3 as weakly positive, 4–6 as moderately positive, and 8–12 points as strongly positive. Therefore, ‘immunohistochemically positive’ tumours have an IRS ≥ 2. Examples of the various IRSs are shown in supplementary material, Figure S1.

Statistical analysis

Differences of the obtained IRS between the different subtypes of squamous cell carcinomas were statistically evaluated using the Mann–Whitney U-test (GraphPad Software, San Diego, CA, USA). A P value of <0.05 was considered significant. Using receiver operating characteristic (ROC) curve analysis, we were able to determine the area under the curve (AUC) value and sensitivity and specificity of the antibodies. The cut-off of the ROC analyses was set at IRS ≥ 2 as this value was considered immunohistochemically positive.

Results Quantitative proteomic comparison of pulmonary and HNSCC cell lines

The differential diagnosis of metachronous primary and metastatic squamous cell carcinoma in patients with HNSCC is crucial for clinical treatment decisions. However, diagnostic biomarkers are still not established in routine diagnostic pathology, partially due to overlapping genetic and morphological features. We therefore aimed to identify new proteomic biomarkers that can be translated to immunohistochemical support for the pathological diagnosis. We therefore compared the protein expression profile of the two human cell lines NCI-H2170 and PCI-13.1. NCI-H2170 originates from a squamous cell carcinoma of the lung and PCI-13.1 from an HPV-negative squamous cell carcinoma from the oropharynx. By using high-resolution MS in combination with SILAC, we were able to quantify 4,606 proteins and 379 of those were significantly differently expressed (Figure 1 and supplementary material, Tables S1 and S3).

Analysis of the significance of the SILAC-based MS. Distribution of the log2 SILAC ratios (PCI-13.1/H2170) of all quantified proteins based on their relative expression in H2170 (SQCLC) and PCI-13.1 (HNSCC) depending on the respective log10 intensity values. Significantly differentially expressed proteins are shown in green according to their significance. Proteins selected for immunohistochemistry are labelled.

We compared those proteins with the expression data of the Human Protein Atlas and chose six markers (two upregulated in the lung cancer cell line H2170 and four upregulated in the HNSCC cell line PCI-13.1) for immunohistochemical validation (Table 1). Staining for these six markers was examined in a small training cohort of HNSCC (n = 6) and SQCLC (n = 6) (Table 2).

Table 1. Proteins selected for immunohistochemistry with their log2 SILAC ratios (PCI-13.1/H2170), log10-transformed intensity values, and calculated P values. The original data are in supplementary material, Table S3. Protein Gene Log2 SILAC ratio (PCI-13.1/H2170) Log10- intensity P value Caveolin-1 CAV1 2.30 10.247 4.50E-05 Caveolin-2 CAV2 1.32 9.311 3.19E-02 Galectin-1 LGALS1 2.32 10.726 4.14E-11 Galectin-7 LGALS7 2.20 9.624 6.00E-04 Cytokeratin-19 KRT19 −2.64 9.33638 2.12E-05 UDP-glucose 6-dehydrogenase UGDH −1.80 11.2308 6.63E-03 Table 2. Results of the marker candidates in the HPA and cohort 1. Marker SILAC-MS (significantly stronger in) IHC (HPA) (n) positive IHC (cohort 1) (n) positive CAV1 PCI-13.1 (HNSCC) 3/3 HNSCC 6/6 HNSCC 1/5 SQCLC 0/6 SQCLC CAV2 PCI-13.1 (HNSCC) 3/3 HNSCC 6/6 HNSCC 1/4 SQCLC 1/6 SQCLC LGALS1 PCI-13.1 (HNSCC) 3/4 HNSCC 5/6 HNSCC 1/5 SQCLC 1/6 SQCLC LGALS7 PCI-13.1 (HNSCC) 3/3 HNSCC 6/6 HNSCC 1/4 SQCLC 1/6 SQCLC CK19 H2170 (SQCLC) 1/4 HNSCC 2/6 HNSCC 3/4 SQCLC 6/6 SQCLC UGDH H2170 (SQCLC) 1/3 HNSCC 2/6 HNSCC 5/6 SQCLC 5/6 SQCLC The table shows the results of the SILAC-MS as well as those of the IHC in the HPA and cohort 1. The absolute number of positive cases and the total number of cases are given for each case. The proteins that were upregulated in HNSCC are highlighted in blue and the proteins that were downregulated are highlighted in grey. HPA, Human Protein Atlas; IHC, immunohistochemistry; SILAC-MS, SILAC-based mass spectrometry. Patient collection for immunohistochemical validation

The clinical and pathological data of 98 patients with SQCLC and 96 patients with HNSCC were evaluated. Of the 96 HNSCC examined, 12 tumours were p16 positive (12.5%); of these, six tumours were located in the oropharynx and six in the pharynx or larynx. In contrast, 84 tumours were p16 negative (87.5%). The SQCLCs were UICC stage I–III and the HNSCCs ranged from stage I to IV (Table 3). All patients were treated by surgery and none of the patients had received neoadjuvant therapy or primary chemo- or radiotherapy. Squamous cell histology of all samples was confirmed by expert pathological review (FB, PS, and HB).

Table 3. Clinical and pathological data. SQCLC (n = 98) HNSCC (n = 96) Age Mean ± SD 65.5 ± 8.4 62.1 ± 10.4 Age range 43–81 24–83 Sex Male 86 (87.8%) 71 (74.0%) Female 12 (12.2%) 25 (26.0%) Localisation Oral cavity – 48 (50.0%) Pharynx – 25 (26.0%) Larynx – 23 (24.0%) Lung 98 (100%) – pT stage pT1 19 (19.4%) 22 (22.9%) pT2 63 (64.3%) 41 (42.7%) pT3 12 (12.2%) 19 (19.8%) pT4 4 (4.1%) 14 (14.6%) pN stage pN0 59 (60.2%) 52 (54.2%) pN1 22 (22.4%) 23 (24.0%) pN2 17 (17.3%) 21 (21.9%) pM stage pM0 98 (100%) 95 (99.0%) Uncertain pM1 0 (0%) 1 (1.0%) UICC stage (eighth edition) I 33 (33.7%) 16 (16.7%) II 43 (43.9%) 26 (27.1%) III 22 (22.4%) 25 (26.0%) IV 0 (0.0%) 29 (30.2%) Grade G1 0 (0.0%) 0 (0%) G2 70 (70.7%) 87 (90.6%) G3 28 (29.3%) 9 (9.4%) p16 positive – 12 (12.5%) Distinction of SQCLC and HNSCC by immunohistochemical biomarkers

We immunohistochemically stained the described collection of squamous cell carcinomas with antibodies specific for CAV1, CAV2, LGALS1, and LGALS7 (upregulated in HNSCC) as well as CK19 and UDP-glucose-6-dehydrogenase (UGDH) (upregulated in SQCLC) in order to test their ability to distinguish SQCLC and HNSCC.

Analysis of proteins upregulated in HNSCC Staining of CAV1 in SQCLC and HNSCC

Tumour cells assessed as positive showed cytoplasmic and often membranous staining. For SQCLC, 33.7% (31/92) of tumours were positive (Figure 2A). HNSCC presented significantly higher expression of CAV1 (p < 0.001), with 89.9% (80/89) of the tumours assessed as immunohistochemically positive (Figure 2B). Median IRSs of 0 for SQCLC and 6 for HNSCC were observed (Figure 2C). In our analysis, CAV1 showed an AUC value of 0.781, a sensitivity of 89.9%, a specificity of 66.3%, a positive predictive value of 72.1%, and a negative predictive value of 87.1% (Figure 2D). The immunohistochemical data of CAV1 depending on the subsites of HNSCC are listed in supplementary material, File S1.

Staining of CAV1 and CAV2 in SQCLC and HNSCC. (A) The SQCLC tumour cells show no or only a slight expression of CAV1 (total magnification ×100). (B) The tumour cells of HNSCC show cytoplasmic and strong membranous expression of CAV1 (total magnification ×50). (C) Box plot for CAV1 expression in SQCLC and HNSCC; the horizontal lines within the boxes represent the median IRS values, ***p < 0.001. (D) CAV1 had an AUC value of 0.781, a sensitivity of 89.9%, and a specificity of 66.3%. (E) The SQCLC tumour cells show no or only slight expression of CAV2 (total magnification ×100). (F) Box plot for CAV2 expression in SQCLC and HNSCC; the horizontal lines within the boxes represent the median IRS values, ***p < 0.001. (G) The tumour cells of HNSCC show cytoplasmic and strong membranous expression of CAV2 (total magnification ×50). (H) CAV2 had an AUC value of 0.613, a sensitivity of 96.8%, and a specificity of 25.8%.

Staining of CAV2 in SQCLC and HNSCC

Tumour cells assessed as positive showed cytoplasmic and often membranous staining. For SQCLC, 74.2% (72/97) of tumours were positive (Figure 2E). HNSCC presented significantly higher expression of CAV2 (p < 0.001), with 96.8% (90/93) of the tumours assessed as immunohistochemically positive (Figure 2F). Median IRSs of 3 for SQCLC and 6 for HNSCC were observed (Figure 2G). CAV2 showed an AUC value of 0.613, a sensitivity of 96.8%, a specificity of 25.8%, a positive predictive value of 55.6%, and a negative predictive value of 89.3% (Figure 2H). The immunohistochemical data of CAV2 depending on the subsites of HNSCC are listed in supplementary material, File S1.

Expression of LGALS1 in SQCLC and HNSCC

Tumour cells assessed as positive showed cytoplasmic and often nuclear staining. For SQCLC, 64.6% (62/96) of tumours were positive (Figure 3A). HNSCC presented significantly higher expression of LGALS1 (p < 0.001), with 92.1% (82/89) of the tumours assessed as immunohistochemically positive (Figure 3B). Median IRSs of 2 for SQCLC and 6 for HNSCC were observed (Figure 3C). LGALS1 showed an AUC value of 0.638, a sensitivity of 92.1%, a specificity of 35.4%, a positive predictive value of 56.9%, and a negative predictive value of 81.6% (Figure 3D). The immunohistochemical data of LGALS1 depending on the subsites of HNSCC are listed in supplementary material, File S1.

Staining of LGALS1 and LGALS7 in SQCLC and HNSCC. (A) The SQCLC tumour cells show no or weakly positive expression of LGALS1 (total magnification ×50). (B) Diffuse expression of LGALS1 in HNSCC (total magnification ×50). (C) Box plot for LGALS1 expression in SQCLC and HNSCC; the horizontal lines within the boxes represent the median IRS values, ***p < 0.001. (D) LGALS1 had an AUC value of 0.638, a sensitivity of 92.1%, and a specificity of 35.4%. (E) The SQCLC tumour cells show no or only a weak immunohistochemical reaction for LGALS7 (total magnification ×50). (F) Box plot for LGALS7 expression in SQCLC and HNSCC; the horizontal lines within the boxes represent the median IRS values, ***p < 0.001. (G) The HNSCC tumour cells show strong cytoplasmic expression of LGALS7 (total magnification ×50). (H) LGALS7 had an AUC value of 0.802, a sensitivity of 94.5%, and a specificity of 66.0%.

Expression of LGALS7 in SQCLC and HNSCC

Tumour cells assessed as positive showed mainly cytoplasmic and focally nuclear staining. For SQCLC, 34% (33/97) of tumours were positive (Figure 3E). HNSCC presented significantly higher expression of LGALS7 (p < 0.001), with 94.5% (86/91) of the tumours assessed as immunohistochemically positive (Figure 3F). Median IRSs of 0 for SQCLC and 6 for HNSCC were observed (Figure 3G). LGALS7 showed an AUC value of 0.802, a sensitivity of 94.5%, a specificity of 66.0%, a positive predictive value of 72.3%, and a negative predictive value of 92.3% (Figure 3H). The immunohistochemical data of LGALS7 depending on the subsites of HNSCC are listed in supplementary material, File S1.

Analysis of proteins upregulated in SQCLC Staining of CK19 in SQCLC and HNSCC

Tumour cells assessed as positive showed mainly cytoplasmic and membranous staining. For SQCLC, 97.9% (96/98) of tumours were positive (Figure 4A). HNSCC presented significantly lower expression of CK19 (p < 0.001), with 69.5% (64/92) of the tumours assessed as immunohistochemically positive (Figure 4B). Median IRSs of 8 for SQCLC and 3 for HNSCC were observed (Figure 4C). CK19 showed an AUC value of 0.624, a sensitivity of 98.0%, a specificity of 30.4%, a positive predictive value of 60.0%, and a negative predictive value of 93.3% (Figure 4D).

Staining of CK19 and UGDH in SQCLC and HNSCC. (A) The SQCLC tumour cells show diffuse positive expression of CK19 (total magnification ×100). (B) There is weak staining for CK19 in HNSCC (×100). (C) Box plot for CK19 staining in SQCLC and HNSCC; the horizontal lines within the boxes represent the median IRS values, ***p < 0.001. (D) CK19 had an AUC value of 0.624, a sensitivity of 98.0%, and a specificity of 30.4%. (E) The SQCLC tumour cells show moderate or only weak immunohistochemical reaction for UGDH (total magnification ×100). (F) Box plot for UGDH staining in SQCLC and HNSCC; the horizontal lines within the boxes represent the median IRS values, *p < 0.0131. (G) The HNSCC tumour cells show only weak cytoplasmic staining for UGDH (total magnification ×100). (H) UGDH had an AUC value of 0.506, a sensitivity of 78.1%, and a specificity of 23.2%.

Staining of UGDH in SQCLC and HNSCC

Tumour cells assessed as positive showed mainly cytoplasmic and nuclear staining. For SQCLC, 78.1% (75/96) of tumours were positive (Figure 4E). HNSCC presented a significant difference in expression of UGDH (p = 0.0131), while showing a nearly similar immunohistochemical positivity of 76.8% (73/95) (Figure 4F). This can be explained by a median IRS of 6 for SQCLC and 3 for HNSCC (Figure 4G). UGDH showed an AUC value of 0.506, a sensitivity of 78.1%, a specificity of 23.2%, a positive predictive value of 50.7%, and a negative predictive value of 51.2% (Figure 4H).

Expression of combination of CAV1 and LGALS7 in SQCLC and HNSCC

With an AUC value between 0.613 and 0.802 (Figures 2D,H and 3D,H), we observed that single immunohistochemical marker was able to distinguish between HNSCC and SQCLC, but the specificity in particular was too low for routine diagnostic use. Therefore, we tested whether a combination of the two most promising markers, CAV1 and LGALS7, was able to further increase the diagnostic accuracy. If only one antibody had to be positive to diagnose an HNSCC, the sensitivity increased to 98.7%. The specificity decreased to 47.3%. The positive predictive value was 61.0% and the negative predictive value 97.7%. The new AUC value was 0.730 (Figure 5A). If both antibodies showed a positive reaction, the sensitivity was 87.2%, specificity improved to 87.9%, the positive predictive value was 86.1%, the negative predictive value 88.9%, and the AUC value was 0.876 (Figure 5B).

Expression of CAV1 and LGALS7 in tumours of the lung after HNSCC. (A) If only one antibody had to be positive to diagnose HNSCC, the sensitivity increased to 98.7% and the specificity decreased to 47.3%. The new AUC value was 0.730. (B) If both antibodies had to show a positive reaction, the sensitivity was 87.2%, specificity improved to 87.9%, and the AUC value was 0.876. (C, D) Expression of CAV1 and LGALS7 in tumour cells of a lung tumour in a patient with HNSCC, which was immunohistochemically classified as met-HNSCC (total magnification ×50). (E, F) No expression of CAV1 and LGALS7 in a lung tumour in a patient with HNSCC, which was immunohistochemically classified as SQCLC (E, total magnification ×100; F, total magnification ×50). (G, H) Weak expression of CAV1 and no expression of LGALS7 in a lung tumour in a patient with HNSCC, which was immunohistochemically classified as uncertain (total magnification ×50).

Immunohistochemical examination of lung tumours of unknown origin

Based on our findings mentioned above, we used the antibody panel consisting of CAV1 and LGALS7 to examine 12 lung tumours from patients diagnosed previously with an HNSCC (Table 4). Both markers had to be assessed positive (IRS ≥ 2) in order to diagnose an HNSCC. Of the 12 carcinomas, 7 showed expression of both markers and were therefore classified as met-HNSCC (Figure 5C,D). Two tumours were negative and were classified as SQCLC (Figure 5E,F). Three cases (Tu-2, Tu-5, and Tu-10) were positive for CAV1 and negative for LGALS7 (Figure 5G,H). These were assessed as tumours of uncertain origin. Further clinical information can be found in Supplementary Table S4.

Table 4. Immunohistochemistry of lung tumours of unknown origin in patients with prior HNSCC. Case Age Gender Localisation of HNSCC Recurrence Pulmonary foci (n) Time interval (years) pT pN Grade CAV1 LGALS7 Clinical classification IHC classification Tu-01 61 M Larynx Yes 2 <1 4 2b G2 Positive Positive met-HNSCC met-HNSCC Tu-02 76 M Larynx Yes 1 ≥3 1 0 G2 Positive Negative met-HNSCC Uncertain Tu-03 62 M Oral cavity Yes 1 >3 3 1 G2 Positive Positive met-HNSCC met-HNSCC Tu-04 73 M Oropharynx No 2 <1 2 0 G2 Negative Negative met-HNSCC SQCLC Tu-05 71 M Hypopharynx No 1 <3 3 1 G2 Positive Negative SQCLC Uncertain