Two peptide sequences were chosen for antibody production; TPTFSTGSIKGLC from the N-terminus and CKPVIRVPSVPPV from the C-terminal region of human K36. These peptides were chosen based on their lack of sequence similarity with other human keratins using BLASTP search of the human proteome. The peptides were synthesised and individually coupled to keyhole limpet hemocyanin using their terminal cysteine residues, and antibodies were prepared in two rabbits using an equimolar mixture of the two peptide conjugates in each rabbit. Sera were collected after four immunisations and affinity purified against the individual peptides, to provide four separate affinity-purified rabbit K36 antibodies. Peptide synthesis, coupling, immunisation and affinity purification was performed commercially (Moravian Biotechnology).

A panel of morphologically normal tissue samples was collected for analysis of K36 expression. This panel comprised samples from different areas of the mobile tongue, excluding the base of the tongue, skin from different anatomical locations, lip, mamilla, nail, thymus, vaginal epithelium, cervix, appendix and oesophagus. The tumour tissue panel included skin cancers such as melanoma (1 case) and basal cell carcinoma (1 case), SCCOT (14 cases) and cervical squamous cancers (12 cases) (Table III). SCCOT samples were derived from the archive at Clinical Pathology between 2012 and 2017 from patients with primary SCCOT and sufficient material for immunohistochemical analysis. The total number of sections tested was 59 and out of these 28 were cancer samples. All tested samples of normal and cancer tissue were obtained from patients undergoing treatment in Umea and Brno. The present study was approved by the Ethical review boards in Umea and Brno (Dnr 08-003M; 03-201). Clinical information about the samples included are summarised in Tables I and II.

Immunohistochemical staining was performed on 4 µm thick tissue sections and the optimal antibody concentration and retrieval were set. The procedure included section deparaffinization in xylene and rehydration into PBS through a graded series of ethanol. Endogenous peroxidase activity was quenched in 3% hydrogen peroxide in PBS for 5 min. Antigen retrieval was performed by boiling in 1 mM EDTA pH 8.0 for 20 min. Afterwards the sections were incubated overnight at 4̊C with affinity purified rabbit serum CKP 39 fraction 1.4. HRP polymer conjugated anti-rabbit (K4003 Agilent Technologies, Inc.) was used according to the manufacturer's instructions (Envision + System; Agilent Technologies, Inc.). Signal was visualized by 3,3'-diaminobenzidine (Liquid DAB + substrate chromogen system; Agilent Technologies, Inc.). Nuclear counterstaining was performed with Gill's haematoxylin. Slides were graded as positive or negative.

The gene expression array data come from our previous study of SCCOT samples using Illumina HT-12 bead chip array containing 47,231 probes (15) that was enlarged by new samples. Here we retrieved and analysed mRNA levels of hair-keratins in data from 14 cases of normal control tongue from healthy volunteers, 22 SCCOT samples and 22 corresponding samples of tumour-free area of the tongue. Raw data has been deposited at http://www.ebi.ac.uk/arrayexpress, Array Express accession numbers E-MTAB-4678(14) (Cases; all controls, tumour/tumour-free 11, 35, 51, 56, 58, 59, 61, 65, 73, 79, 85) and E-MTAB-5534 (Cases 20, 35, 49, 76, 98, 105, 111, 119, 124, 131, 137, 138).

Statistical analysis was performed on the previously published array data (14) using IBM SPSS 25 (IBM Corp.). Differences at P≤0.05 were considered to be statistically significant. Changes in the level of expression of each KRT gene within group of samples (control, tumour-free, tumour) was analysed using ANOVA with post hoc Tukey HSD test. Data are analysed as the mean ± standard deviation.

Microarray data were analysed using application Multiple Experiment Viewer (MeV) (19).

A mixture of two peptides from the human K36 protein (conjugated to keyhole-limpet hemocyanin) was used to immunise rabbits for antibody production. Sera were then used in dot-blots against the individual peptides (conjugated to bovine serum albumin) to assess antibody production. In both rabbits, a stronger response was seen against the CKPVIRVPSVPPV peptide (C-terminal region) than against the N-terminal peptide (Fig. S1). Sera were then individually affinity-purified against each peptide and fractions were assessed for immunoglobulin purity and amount by polyacrylamide gel electrophoresis, again showing higher levels of antibodies to the C-terminal region peptide in both rabbits (Fig. S2). These initial assessments were performed and provided by the supplier.

Affinity-purified antibodies and non-affinity purified sera were then tested at various concentrations on normal human tongue samples, employing antigen retrieval (EDTA or citrate) or no antigen retrieval. Pre-immune sera from the two rabbits were used as controls for their respective non-purified sera and affinity-purified reagents. From this initial screening, serum from rabbit 39 affinity-purified against the CKPVIRVPSVPPV peptide was chosen for further analysis, using antigen retrieval in 1 mM EDTA boiling for 20 min. Serum from rabbit 33 purified against the same peptide showed similar staining characteristics but required a lower dilution, whilst both sera purified against the TPTFSTGSIKGLC peptide showed similar staining but required high concentrations with attendant background staining. The non-immune sera did not show any staining when used at similar or higher concentrations.

Having established optimal conditions, we analysed K36 in the set of selected tissues using immunohistochemistry with affinity purified serum (designated 39-CKP). In view of our finding of altered KRT36 mRNA levels in SCCOT, we focused on staining K36 in normal and malignant human tongue epithelium.

No scoring of percentage K36 expressing cells or intensity in expression was performed. Tongue epithelium is a morphologically variable tissue with diverse structures. K36 immunostaining was localised solely to filiform papillae and not detected in other structures (Fig. 1). Within the filiform papilla, K36 was present in the cytoplasm of epithelial cells organised in arcs at the periphery. These cell clusters with elongated shape are called secondary filiform papillae and give rise to cornified spines (6). Other compartments of filiform papillae such as primary structure and interpapillary epithelium were K36-negative (Fig. 2).

We also analysed the distribution of K36-positive filiform papillae on the dorsal front tongue epithelium and found that all filiform papillae recognized in the anterior two-thirds of the mucosa were K36-positive.

To see whether K36 is dysregulated in cancer tissue, we also included samples of SCCOT, but K36 was not detected in any of these samples (Table III and Fig. 2).

We also used a panel of normal epithelial tissues chosen according to RNA sequencing data available in the Human protein atlas open access database (19). Based on these data and the literature, we compiled a set of human normal and cancer tissues that express KRT36 mRNA but that were not yet evaluated at protein level. The panel of normal tissue samples included several positive cases, namely nail bed, Hassal's corpuscles of thymus and hair cortex (Fig. 3). Squamous epithelia from different anatomical locations, lip, mammilla, cervix, oesophagus and vagina were K36-negative and non-squamous epithelia in these tissues and appendix were also negative. Tumour samples of melanoma, basal cell carcinoma and cervical carcinoma were also negative for K36. An overview of staining results is provided in Table III.

We analysed mRNA data from tongue tumours, tumour-free and healthy controls (14) concerning levels of individual hair type keratins (KRT31-KRT38 and KRT81-KRT86) in each group of samples (Fig. 4). Statistical analysis revealed significant downregulation of KRT31 (P=0.003), KRT33A (P=0.000), KRT33B (P=0.000), KRT34 (P=0.000), KRT36 (P=0.000), KRT84 (P=0.000) and KRT85 (P=0.000) in tumour samples compared to healthy controls. Of these, KRT84 mRNA showed the highest correlation with KRT36 mRNA levels. Levels of KRT37 (P=0.000) and KRT81 (P=0.000) in contrast were significantly upregulated in cancer (ANOVA/Tukey HSD test) (Table SI and Fig. S3).