Non‐invasive diagnosis of early cutaneous squamous cell carcinoma

Early cutaneous squamous cell carcinoma (cSCC) can be challenging to diagnose using clinical criteria as it could present similar to actinic keratosis (AK) or Bowen's disease (BD), precursors of cSCC. Currently, histopathological assessment of an invasive biopsy is the gold standard for diagnosis. A non‐invasive diagnostic approach would reduce patient and health system burden. Therefore, this study used non‐invasive sampling by tape‐stripping coupled with data‐independent acquisition mass spectrometry (DIA‐MS) proteomics to profile the proteome of histopathologically diagnosed AK, BD and cSCC, as well as matched normal samples. Proteomic data were analysed to identify proteins and biological functions that are significantly different between lesions. Additionally, a support vector machine (SVM) machine learning algorithm was used to assess the usefulness of proteomic data for the early diagnosis of cSCC. A total of 696 proteins were identified across the samples studied. A machine learning model constructed using the proteomic data classified premalignant (AK + BD) and malignant (cSCC) lesions at 77.5% accuracy. Differential abundance analysis identified 144 and 21 protein groups that were significantly changed in the cSCC, and BD samples compared to the normal skin, respectively (adj. p < 0.05). Changes in pivotal carcinogenic pathways such as LXR/RXR activation, production of reactive oxygen species, and Hippo signalling were observed that may explain the progression of cSCC from premalignant lesions. In summary, this study demonstrates that DIA‐MS analysis of tape‐stripped samples can identify non‐invasive protein biomarkers with the potential to be developed into a complementary diagnostic tool for early cSCC.


| INTRODUC TI ON
Cutaneous squamous cell carcinoma (cSCC) is the second most frequent cancer in humans. 1 Surgical excision is frequently curative but sometimes cSCC metastasize.It is a significant cause of mortality due to skin cancer, particularly in solid organ transplant patients and patients on immunosuppressants.cSCC affected 0.4% of the Australian population in 2002 and contributed to 250 000 nonmelanoma skin cancer-related surgical procedures in 2013-14. 2 Actinic keratosis (AK) and Bowen's disease (BD) are pre-cancerous forms of cSCC 3,4 and are characterised by early and intermediate levels of dysplasia within the epidermis, respectively. 5AK and BD can accumulate mutations and progress into an invasive cSCC. 6Due to clinical and histopathological heterogeneity, differentiating pre-cancerous lesions AK and BD with no invasive potential from those that will evolve into invasive cSCC or lesions that are already early cSCC is difficult, 7,8 leading to the excision of benign lesions or delaying treatment for more aggressive lesions. 8,9Moreover, the discomfort and cost associated with surgery make the development of a non-invasive and costeffective diagnostic method desirable.
The analysis of protein biomarkers has provided helpful information related to cSCC lesions, 10 but the lack of diagnostic and preoperative biomarkers remains an ongoing challenge.2][13][14][15] In another study, we also applied MS-based proteomics to samples of AK and normal skin stratum corneum collected using tape-stripping (TS). 12,16We demonstrated that this method could identify proteins involved in tumorigenic pathways, such as the PI3/AKT and EGF signalling. 12TS is a sensitive, noninvasive, and cost-effective form of sample collection that works by applying an adhesive tape to the skin surface, removing cells from the stratum corneum that are attached to the tape. 16,17 this study, we used tape-stripping coupled with the DIA-MS platform to profile the proteome of samples collected from patients with AK, BD and cSCC to identify protein biomarkers and molecular disruptions with the potential to be developed into a non-invasive diagnostic tool for AK, BD and cSCC.

| Study population
Patients who were scheduled for a biopsy or excision of a possible cSCC were offered the opportunity to participate in the study.Once they provided written informed consent, tape-stripped samples were collected according to the procedure described below.After obtaining scarless samples, the lesions were biopsied or excised immediately, as per standard care.Only samples with a confirmed histopathological diagnosis of AK, BD or cSCC were included in this study.The Western Sydney Local Health District Human Research Ethics Committee approved this study (approval #HREC/17/WMEAD/555).

| Non-invasive sample collection
The previously described method 11,18 was used with minor modifications to collect non-invasive patient samples for downstream proteomic analysis.Briefly, the patient's skin stratum corneum was collected by the sequential application and removal of six adhesive discs (D104-D-Squame, CuDerm) against the same lesion with light pressure using a finger.The first disc was discarded to eliminate remnants of external material.The same process was repeated to collect normal stratum corneum from a similar body site distant from the lesion.The discs were placed into a 2 mL cryotube (Greiner Bio-One, Frickenhausen, Germany) on ice, with the adhesive side facing inward.The cryotubes were stored at −80°C before their preparation for mass spectrometry analysis.

| Sample preparation
Sample preparation for downstream mass spectrometry analysis followed the method previously described 11,18 with minor modifications.In brief, proteins from the discs were dissociated and extracted by adding 500 μL protein extraction buffer (0.1% w/v RapiGest SF Surfactant, Waters Corporation, in 50 mM triethylammonium bicarbonate, Sigma-Aldrich).The tubes containing discs were then rotated overnight on a roller mixer (Ratek) at 4°C, followed by a vigorous vortex.Following centrifugation, the samples were transferred into boil-proof tubes (Axygen Scientific) and sonicated for 1 min with 20% amplification (Branson Digital Sonifier 450, Branson Ultrasonics).After 30-min incubation at 95°C, the samples were centrifuged at 13000 rpm for 20 min at 4°C.Protein extracts were then reduced in 12 mM tris(2-carboxyethyl) phosphine (TCEP; Sigma-Aldrich) for 30 min at 60°C, alkylated in 50 mM iodoacetamide (Sigma-Aldrich) for 30 min at room temperature, and subjected to overnight tryptic digestion using sequencing grade modified trypsin (1:50 trypsin: protein ratio; Promega).Following centrifugation at 13000 rpm, the peptide mixtures were desalted and cleaned using 1 cc Oasis HLB cartridges (Waters Corporation).Finally, peptides were eluted from the cartridges with 500 μL of 50% acetonitrile (ACN) and then 200 μL of 80% ACN.The final protein yield was determined using a Qubit 3.0 Fluorometer (Life Technologies).A 4μg aliquot from each sample and one pooled sample by adding 1 μg of protein digest from all 61 samples were prepared and dried using a Savant SpeedVac concentrator (Thermo Fisher Scientific) for mass spectrometry analysis.
The proteomic data were acquired in a TripleTOF 6600 (AB Sciex) mass spectrometer operating in a Sequential Window Acquisition of all Theoretical Fragment Ion Spectra (SWATH) mode, a variant of the DIA-MS data acquisition approach.A total of 106 general variable windows with a minimum window width of 2 Da and 1.0 Da overlap generated using SWATH Window Calculator V.2 (AB Sciex) was used.The variable window range between 350 and 1500 m/z with an accumulation time of 25 ms for the survey scan and 25 ms for the tandem scan.Collision energy spread (CES) was set to 0. Other SWATH analysis parameters included: a start mass of 400 Da, a stop mass of 2250 Da, a SWATH width of 56.06, and 33 SWATHs per cycle.Collision energies for each window were calculated using rolling collision energy with a charge state of +2 and a CES of 15 V. Calibration was performed after every five samples using bovine serum albumin (BSA) peptide mixture.One pooled skin sample was run after every 10 samples to monitor the mass spectrometer performance.

| Database search
For protein identification and quantitation, the SWATH raw files were searched in Spectronaut Pulsar X software (version 12.0, Biognosys AG) using the library-free Direct DIA approach. 19The files were searched against the Human Uniprot fasta database (date: 22 July 2021) using Spectronaut Pulsar's default parameters.Search parameters included a minimum peptide length of 5, a maximum peptide length of 70, a maximum number of two missed cleavages and five variable modifications.Quantification was performed at the MS2 level, and the data were normalised across the samples using the Spectronaut Pulsar's inbuilt cross-run normalisation. 20The resulting protein list with a maximum false discovery rate (FDR) of 1% was exported into a Microsoft Excel file for statistical analysis.

| Statistical analysis
Missing values were replaced with the minimum observable number of 1, followed by log-2 transformation.Proteins identified in less than 50% of the samples and samples with less than 25% of total protein identification were removed from downstream analysis.The remaining data were uploaded into Multi Experiment Viewer (Web-MeV, v2.0, TM4) software, 21 where differentially abundant analyses were performed using Linear Models for Microarray data (LIMMA). 22e p-values were adjusted using the Benjamini-Hochberg method to control for multiple testing.Differentially abundant proteins with an adjusted p < 0.05 and a fold change ≤ −2 or ≥ + 2 were considered statistically significant.The function of individual proteins was annotated with the UniProt Knowledgebase (UniProtKB, https://www.unipr ot.org).Differentially abundant proteins between the lesion groups were visualised using volcano plots generated in VolcaNoseR (Huygens Science: https://huyge ns.scien ce.uva.nl/VolcaNoseR) web application. 23

| Classification analysis
For classification analyses, the data were uploaded into Perseus software (version 2.0.7.0). 24A cross-validation strategy using a support vector machine (SVM), a supervised machine learning algorithm for classification problems, 25 was used to evaluate the efficacy of the SWATH data in classifying patient samples into premalignant (AK + BD) and malignant (cSCC) groups.Initially, the number of proteomic features providing the best classification accuracy based on the ANOVA test feature ranking was determined for the classification.Then, using the recommended number of features, default 10 penalty constant and fourfold cross-validation, SVM was used to construct models for classification, and their performance was evaluated.

| Pathway analysis
Finally, to investigate the pathophysiology of AK, BD and cSCC progression from normal skin, differentially abundant proteins were uploaded into the Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems USA; Release date: Sep 2021; http://analy sis.ingen uity.com). 26The previously described method 12,14 was used to perform core analysis, including molecular pathway analysis and biological function analysis.Networks of proteins significantly associated with AK, BD and cSCC progression were generated.In addition, the differential abundant proteomic data were uploaded into the Enri-chR, 27 where diseases and molecular pathways enrichment analyses were performed against the Disease Perturbations database from the gene expression omnibus (GEO) 28 and the National Cancer Institute Nature Pathway Interaction Database terms (2015 update). 29

| Study population
A total of 61 TS samples including 32 lesions and 29 body site-matched controls were collected from a total of 26 patients aged 18 or over attending the Dermatology Clinics at Westmead Hospital, New South Wales, Australia.From 22 patients, one lesion and one body-site matched control samples were collected.From the remaining four patients, the following number of samples were collected: one clinically diagnosed AK, one BD and one matched control from patient one, two BDs, two cSCCs, and two matched controls from patient two, two cSCCs and two matched controls from patients three and four, respectively.Overall, five TS samples of clinically diagnosed AK, seven BD and 20 cSCCs were collected (Table 1A).

| Clinicopathological correlation
All lesions, independent of the study and as part of standard care, underwent biopsy due to the lack of clarity in the clinical diagnosis.
After surgical removal of the lesions, histopathology reports were reviewed, and definitive diagnoses were established which included six AKs, nine BDs and 17 cSCCs.Out of five clinically diagnosed AKs, one was identified as BD (20%) and another as cSCC (20%) after histopathological evaluation.Among the seven clinically diagnosed BDs, one lesion was diagnosed as AK (14.2%) and two as cSCC (28.4%).

| Protein identification and quantification
The SWATH-MS data files were searched against the human proteome database using the DirectDIA library-free and label-free search algorithm to identify and quantify proteins.In total, 431 protein groups corresponding to 696 proteins and 5529 precursors with a maximum FDR of 1% were identified across all the samples.Following data filtering, two samples (1 AK and its matched normal control) with less than 25% total protein identification and 54 protein groups detected in less than 50% of at least one lesion group were removed from downstream analysis (Table S1).

| Classification analysis
A machine learning framework was used to evaluate whether the proteomic data could distinguish between lesions based on their histopathological diagnosis.Given the limited number of samples in the AK and BD groups, we trained an SVM to perform classification between the combined premalignant (AK + BD) samples and malignant (cSCC) samples.Based on the results of the ANOVA test, proteins within cross-validation were identified and ranked for their potential diagnostic value.The classification analysis using two proteins (CTSV and CST6) classified the premalignant and malignant lesions at 70.97% balanced accuracy (or 29.03% error rate; sensitivity = 70.6%,specificity = 71.4%)as presented in Figure 1.Out of these, two premalignant lesion samples and two cSCCs were the same as those clinically misdiagnosed.However, incorporating 207 protein features into the classification analysis improved the balanced classification accuracy to 77.5% (sensitivity = 76.5%, specificity = 78.6%)(Figure 1).Of these, two premalignant and one cSCC samples were the same as those clinically misdiagnosed.

| Differentially abundance analysis
We next aimed to identify differentially abundant proteins between the groups.Comparing each lesion group with normal skin samples, we identified 144 protein groups (138 increased; 6 decreased) that were significantly altered in cSCC samples (Figure 2A, adj.p < 0.05; fold change >2 or < −2).In contrast, 21 proteins were differentially abundant in BD samples, all of which were increased (Figure 2B, adj.p < 0.05; fold change >2).No proteins with an adj.p < 0.05 were found to be significantly altered in AK samples.However, 15 proteins with p < 0.005 (fold change >2 or < −2) were identified (Figure 2C).
We also ranked the top 15 most significantly altered proteins by adj.
p-value and have summarised them in Table 2D.Furthermore, our differential abundance analysis revealed that nine proteins were significantly different between cSCC and AK, one protein between cSCC and BD, and two proteins between BD and AK respectively (p < 0.005; fold change >2 or < −2) (Table 2E).The complete list of differentially abundant proteins is provided in Tables S2-S7.

| Gene-set enrichment and biological function analysis
To gain further insight into the biological implications of our findings, we matched the differentially abundant proteins in BD and TA B L E 1 Overview of samples included in this study.Values in bold represent the number of samples or their percentage matching between clinical and histopathology diagnoses.In cSCC lesions, Class I PI3K signalling mediated by Akt was the most enriched molecular pathway, whereas PAR4-mediated thrombin signalling enrichment ranked the highest in BD lesions.

Lesion group (histopathology diagnosis)
Notably, no disease perturbation or pathway terms meeting the significance criteria were identified when comparing AKs with their corresponding normal skin samples.
Proteins that displayed differential abundance in AK, BD and cSCC lesions compared to the normal skin (p < 0.05) were subjected to further analysis using IPA bioinformatics software to predict the molecular pathways and biological functions that were affected in each lesion.The software was able to predict various biological functions, including cell movement/migration, cell invasion and cell survival/viability that were activated (z-score >2), and apoptosis/ cell death that was inhibited (z-score <2) in cSCC and BD lesions compared to the normal skin (p < 0.05) (Figure 4A,B,D,E).In AK lesions, cell viability was also predicted to be activated and apoptosis inhibited.Additionally, the analysis linked the AK lesion with a significantly decreased inflammatory response compared to the normal skin (Figure 4C,E).
When comparing the lesions, cSCC showed relatively higher activation of LXR/PXR activation, production of nitric oxide, MSP-RON signalling and PI3K/AKT signalling pathways compared to BD lesions.The analysis also predicted that ERK5, ILK and VEGF signalling were activated, and CLEAR and HIPPO signalling were inhibited only in the cSCC lesions (Figure 4D).

| DISCUSS ION
The current study utilised clinical patient samples in conjunction with DIA-MS proteomics to assess the feasibility of identifying protein signatures for a non-invasive diagnosis and classification of lesions suspicious for cSCC.Previously, we optimised a proteomic method for identifying proteins in AK samples using 10 adhesive discs. 11Additionally, a separate study conducted by our team previously demonstrated that there were only minimal changes (<5%) in proteome profiles of samples collected from different stratum corneum layer depths. 16The current study showed that important proteome-level information can be gathered using a reduced number of discs (five discs), and therefore reduced time for sample collection and resources.

| Proteomic classification analysis
Due to the clinical heterogeneity, the pre-malignant lesions AK and BD are often challenging to differentiate from cSCC leading clinicians to biopsy or excise lesions that may not be cSCC.In the present study, a machine learning framework using two proteins (CTSV and CST6) achieved a classification accuracy of 70.97% (or 29.03% error rate) when differentiating between premalignant and malignant lesions, which is similar to dermatologist's clinical assessment.However, the rate of classification accuracy increased to 77.5% when 207 proteins were used.This finding is significant from a clinical perspective, as the decision to surgically remove or not to remove a lesion depends on the level of confidence clinicians have about its malignancy.Typically, only malignant or suspected malignant lesions are removed surgically.
Overall, these results suggest that non-invasive proteomic analysis holds promise for improving the diagnosis and classification of AK, BD and cSCC.To further enhance the accuracy of the model, a larger cohort of patient samples can be used to develop, refine and validate the model.This development could lead to a significant improvement in patient outcomes, particularly for those who lack access to specialised dermatological care.

| Proteomic analysis identifies non-invasive protein signatures in AK, BD and cSCC
Of proteins significantly changed in the cSCC lesions compared to the normal stratum corneum, 11 proteins including PPL, SFN,   and YWHAE were previously reported to have changed in formalinfixed and paraffin-embedded (FFPE) samples cSCCs compared to the normal skin. 12While the proteomic composition of stratum corneum and whole thickness cSCC tissue differ due to differences in the number and type of cells present at each tissue type, our findings indicate that reliable protein signatures of cSCC lesions can be detected at the skin surface.In addition, other significantly changed proteins with the potential to non-invasively differentiate between cSCC and normal skin reported in this study include PGK1, HRNR, KRT75, MDH1, C3 and CTSV.While the role of some of these proteins including PGK1 30 and C3 31 in CSCC tumour progression, migration and invasion has been reported before, but the role of other proteins identified here remains unexplored.
Likewise, in our study, we identified protein biomarkers such as S100A2, SFN and EEF1G, which were previously reported as potential biomarkers of BD lesions compared to normal skin in proteomic analysis of FFPE samples. 12Additionally, we identified other proteins, including PPIA, SERPINB5, MDH1, GSTP1 and APOA4, that could potentially serve as non-invasive biomarkers for the diagnosis of BD.
Our study, limited by the small number of AK samples (n = 5), did not identify protein biomarkers that were significantly changed (adj.p < 0.05) in AK stratum corneum samples compared to the normal skin.Nevertheless, the identification of proteins with very low p-values suggests that our non-invasive proteomic approach has the potential to identify biomarkers of AK in a study with a larger sample size.Further research involving larger cohorts of AK patients is required to validate the diagnostic potential of the identified proteins and to explore their role in the pathogenesis of AK.
8][9] In this study, we reported the identification of nine and one proteins that could non-invasively differentiate cSCC from AK and BD lesions, respectively.Two proteins including S100A12 and SPRR1A were found to differentiate between BD and AK lesions.It is out of the scope of this study to investigate the role of each of these proteins in carcinogenesis, however, their evaluation and validation as diagnostic and pre-operative biomarkers in independent cohorts of patient samples is warranted.
We report that the quantitation data related to two proteins, CTSV and CST6, can be used to distinguish between cSCC and premalignant AK and BD lesions.Specifically, our study revealed that CST6 exhibited a significant decrease in cSCC when compared to AK and BD lesions (p < 0.05; Tables S6 and S7).These results are consistent with our previous investigation using FFPE samples of cSCC, where CST6 was also found to be significantly reduced when compared to normal skin. 13Furthermore, we had shown that CTC6 was decreased in metastatic cSCC lesions compared to primary phenotypes. 14CST6 is also demonstrated to act as a tumour suppressor in other cancers such as melanoma, prostate, and renal cancers. 32llectively, these findings suggest a potential tumour suppressive role for CST6 in the development and progression of cSCC, highlighting its significance as a promising candidate for further exploration in cancer research.
Cathepsin V (CTSV) is a lysosomal cysteine proteinase that participates in the degradation of the extracellular matrix.In the context of cSCC, CTSV expression is reduced compared to AK and BD lesions.Despite its involvement in various cancers, the specific role of CTSV in cSCC remains unclear.Notably, high CTSV expression has been linked to unfavourable outcomes in breast ductal carcinoma, 33 promotes proliferation through the NF-κB pathway in bladder cancer, 34 and drives lung cancer metastasis. 35Conversely, CTSV is reported to be decreased in melanoma samples. 36To fully comprehend the role of CTSV in cSCC development, further investigation is warranted.
Overall, findings from this study demonstrate that DIA-MS analysis of tape-stripped samples identifies non-invasive protein biomarkers in cSCC with the potential to be developed into a diagnostic tool that will complement clinical diagnosis.

| Non-invasive proteomic analysis identifies relevant molecular pathways associated with keratinocyte carcinogenesis
Gene-set enrichment analysis together with molecular pathway and biological function analysis of the proteomic data using IPA bioinformatics tool revealed that cSCC lesions are more strongly associated with carcinogenic biological functions and molecular pathways compared to BD and AK lesions.This was in line with the findings from our previous study involving proteomic analysis of FFPE samples of AK, BD and cSCC. 12In both studies, it was found that biological functions (i.e.cell proliferation, inflammatory response, metabolism of proteins, etc) were significantly more activated in cSCC lesions The present study also found that BD and cSCC lesions are strongly associated with LXR/RXR activation, production of reactive oxygen species, PI3K/AKT signalling and p38 signalling mediated by MAPKAP kinases.In addition, CLEAR signalling and Hippo signalling were the most significantly inhibited pathways in cSCC lesions.LXRs and RXRs are epidermal ligands abundantly expressed in cancers, including melanoma, due to their protective role as cancer regulators.The activation of LXRs and RXRs is reported to inhibit keratinocyte proliferation and maintain the epidermal permeability barrier, decreasing carcinogenesis. 42,43The findings of this present study illustrate that LXR/RXR activation increases as the lesion progresses from AK to cSCC.Therefore, this may indicate that more advanced stages may require more protective intervention from cancer regulators, which may justify its high activation of LXR/RXRs.
The production of reactive oxygen species (ROS) and nitric oxide induces DNA damage and has been linked to a critical role in transforming non-malignant to malignant cells. 44Tumour-related macrophages secrete ROS and nitric oxide into the tumour microenvironment, facilitating oncogenesis. 45Thereby, the gradual activation of this pathway in this study as the lesion progresses from AK to cSCC may be positively associated with the level of ROS and nitric oxide produced in the lesion.
cSCC lesions compared to normal skin against the disease perturbations database from the GEO 28 knowledgebase.Our analysis revealed that cSCC was the top enriched disease perturbation, followed by other skin diseases such as eczema and melanoma (adj.p < 0.05).Bar charts displaying the top 10 enriched diseases ranked by the combined enrichment score (adj.p-value and odds ratio) in the chosen library are presented in Figure 3A,B.Additionally, we compared the proteomic data against the NCI-Nature pathway database and identified p38 signalling mediated by MAP-KAP kinases and insulin-mediated glucose transport as among the top 10 molecular pathways significantly associated with changes in both cSCC and BD proteomic data (adj.p < 0.05) (Figure 3C,D).

F I G U R E 2
Volcano plots of differentially abundant proteins between the lesions: Volcano plot of proteins differentially abundant in the (A) AK vs normal skin, (B) BD versus normal skin, and (C) cSCC versus normal skin samples.Each red dot represents a protein that is significantly increased (adj.p < 0.05 and fold change ≥2), and each blue dot represents a protein that is significantly decreased (adj.p < 0.05 and fold change of ≤2) in each lesion group compared to the normal skin samples.TA B L E 2The list of top 15 proteins differentially abundant in AK, BD and cSCC lesions.The protein function was annotated using the UniProt Knowledgebase.

TA B L E 2
(Continued)  YWHAG, SPRR3, S100A2, TYMP, KRT4, PFN1, FSCN1, HSP90AA1 followed by BD.Additionally, both studies identified cell/organismal death as the most suppressed biological function in BD in contrast to cSCC and AK lesions.These findings suggest that the proteomic data obtained from tape-stripped stratum corneum samples and FFPE samples provide pertinent molecular insights into the progression of cSCC from normal skin.F I G U R E 3 Gene-set enrichment analysis using EnrichR: Bar charts show the top 10 enriched diseases by the (A) cSCC versus normal skin and (B) BD versus normal skin sample datasets compared to the Disease Perturbation database available in EnrichR.The disease terms are ranked from left to right by the combined enrichment score (adj.p-value and odds ratio; represented as the length of red background colour) in the chosen database.(C, D) Level of enrichment of top 10 molecular pathways by the cSCC and BD proteomic datasets, respectively, against the NCI-Nature pathway database (adj.p < 0.05).The length of the grey colour corresponds to the level of enrichment of each molecular pathway term.In contrast to the well-known characteristic of increased inflammatory cells and responses in AK lesions,[37][38][39] particularly in the upper dermis layer, our study revealed a predicted decrease in inflammatory responses in the stratum corneum.One possible explanation could be that inflammatory cells and mediators may not effectively reach the stratum corneum in AKs.Unlike BD and cSCC lesions, AKs are primarily confined to the deeper epidermis layer.Consequently, this restricted migration of inflammatory components could contribute to the reduced signatures of inflammatory cells observed in the sampled stratum corneum.However, to comprehensively address this discrepancy, further studies with more robust and representative data are required to elucidate the dynamics of inflammatory responses in AK lesions.

F I G U R E 4
IPA biological function and molecular pathway analysis.(A-C) Networks of significantly changed proteins in the cSCC (A), BD (B) and AK (C) datasets and their overall impact on the given biological functions.Each node represents a protein.Green and red colours indicate decreased or increased proteins in the respective dataset, respectively.The orange indicates predicted activation, and the blue indicates inhibition of the given terms.(D) Heatmap of molecular pathways implicated in AK, BD and cSCC lesions.A heatmap summary of biological functions implicated in the lesions is provided in (F).Refer to the legend for detailed information.