The development of a high-plex spatial proteomic methodology for the characterisation of the head and neck tumour microenvironment

doi:10.21203/rs.3.rs-5272207/v1

Download PDF

Article

The development of a high-plex spatial proteomic methodology for the characterisation of the head and neck tumour microenvironment

https://doi.org/10.21203/rs.3.rs-5272207/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Mucosal head and neck squamous cell carcinoma (HNSCC) is a debilitating disease that accounts for an estimated 890,000 new cases per year, making it the seventh most common cancer globally. HNSCC is a heterogenous group of cancers that affect various regions of the head and neck which stem from the epithelial cells in the mucosal lining. Despite advancements in chemotherapy, radiotherapy, surgery and immunotherapy, the prognosis of HNSCC has remained relatively unchanged for more than a decade. There is a need to better understand the tumour microenvironment (TME) using spatially resolved approaches, to gain insights into the TME associated with clinical endpoints such as Disease-Free Survival (DFS) and Overall Survival (OS). Here, we profiled 84 mucosal HNSCC tissue samples using next-generation ultra-high plex spatial protein profiling (580-proteins, Immuno-Oncology Proteome Atlas) and spatial transcriptome mapping (18,000 mRNA, Whole Transcriptome Atlas) from Nanostring Technologies (Bruker). Patient specimens were collected during tumour resection, where patients then went on to receive either chemotherapy and/or radiotherapy. Each patient tissue sample was subdivided into tumour and stromal regions prior to digital spatial profiling. We found that patient survival outcomes (both DFS and OS) were associated with anatomical locations and tumour stage. Notably, there were specific proteomic and transcriptomic features in both the tumour and stromal regions that associated with DFS and OS. Independent validation of key proteomic findings (including CD34 and CD44) was performed using single-cell protein profiling (PhenoCycler-Fusion, Akoya Biosciences). Finally, cell type deconvolution based on transcriptomic signatures revealed cell types associated with patient survival. Taken together, this study provides a systematic workflow for discovery and validation of high-plex protein and transcriptomic profiling in mucosal HNSCC.

Biological sciences/Biotechnology

Biological sciences/Cancer

Biological sciences/Computational biology and bioinformatics

Biological sciences/Immunology

Mucosal head and neck squamous cell carcinoma (HNSCC) is considered the seventh most common cancer globally. HNSCC (comprising mucosal tumours of the lip, oral cavity, oropharynx, nasopharynx, hypopharynx, larynx) had an incidence of ~ 890,000 and a mortality of ~ 460,000 in 2022¹. Depending on the location of the tumour, HPV status, and stage at diagnosis, the patient prognosis can be highly variable. Surgery and radiotherapy, with or without concurrent chemotherapy, are standard treatment for patients with local or locoregional disease, but many patients will develop resistance or recurrence, in which case survival significantly decreases and results in a median survival of only 10–15 months². Due to the chance of recurrence and treatment resistance, there is an urgent need to understand the complex tumour microenvironment involved which may inform clinical endpoints such as progressive free survival (PFS), disease free survival (DFS) and overall survival (OS). Moreover, there is a need for improved biomarkers predictive of clinical outcome to better triage patients for targeted and systemic therapies.

Much effort in recent years has focused on how the tumour microenvironment (TME) influences disease outcomes in mucosal HNSCC. It has become clear that the number and type of immune cells present within the TME contribute to disease progression and response to therapy, but questions still remain on what the exact characteristics of the TME contribute or predict to good outcomes. The aggregation of immune cells, such as tertiary lymphoid structures (TLS), and interactions between immune cells and tumour cells or other immune cells are considered important factors of patient response for HNSCC³. Through the advancement of spatial biology technologies, scientists are now able to more clearly identify complex cellular interactions that influence patient outcomes, findings that are often missed when not accounting for spatial information³.

Here, we profiled the mucosal HNSCC TME using the spatial multi-cellular Nanostring GeoMx Immuno-Oncology Proteome Assay (IPA), which is the largest spatial-proteomic panel currently commercially available for fresh frozen and FFPE samples, to interrogate 580 immuno-oncology proteins. In parallel, serial tissue sections were profiled using the Nanostring GeoMx Whole Transcriptome Atlas (WTA) covering over 18,000 mRNA targets. The findings are validated independently using single cell resolved spatial proteomics (Akoya Biosciences PhenoCycler Fusion) analysis of a serial section for 46 markers. This study demonstrates one of the first multi-model workflows between GeoMx IPA and WTA assays as well as highlights the utility of the high plex IPA assay to identify prognostic biomarkers for patient survival in mucosal HNSCC.

HNSCC patient survival associate with anatomical locations and tumour stage

The samples from this study came from 84 chemotherapy and/or radiotherapy treated patients with tissues acquired from a range of sites across the mucosal head and neck regions (i.e. tongue, oral cavity, pharynx and lip). The clinical information available were analysed to assess the respective influence on treatment outcomes or patient survival. In all cases looking at either disease free survival (DFS) or overall survival (OS), the anatomical location of the sample as well as the broader tumour stages (i.e. grouping patients as either early (I and II) or advanced (III and IV) stages) was found to significantly associate with patient prognosis (based on Kaplan Meier survival analysis (Supplementary Fig. 1). For both DFS and OS, patients with samples from the tongue are associated with the worse survival outcomes while samples from the lip associates with the best survival outcomes (p < 0.0001 and p = 0.0013). Similarly, patients at the early stage of cancer (i.e. I and II) have a significantly better disease free (p = 0.0088) and overall survival (0.0013) outcomes than those at the advanced stages of cancer (i.e. III and IV). These findings are consistent for subsets of the patients with samples used in the respective assays (Supplementary Fig. 2).

Spatial multicellular proteomics analysis

The GeoMx IPA dataset consist of 138 AOIs (paired tumour-stroma regions) across 68 patient cores (each core with one ROI separated into tumour or stroma regions) with the expression matrices as integrated counts for 580 protein markers. The data were pre-processed using R package standR’s QC pipeline⁴ as described in methods. Briefly, correlations between the background controls (IgGs) are high but not so for the housekeeping markers (Fig. 1A). Assessing sample and protein quality, no protein marker nor any AOIs was removed based on assessment of the nuclei counts and library size (Fig. 1B) with no factor identified to be confounded with either. The counts data is normalised and corrected for systematic bias using the RLE scaling method (as shown by the centred Relative log expression (RLE) plots, Fig. 1C). PCA using the logCPM (log counts per million) highlights the factors contributing to the variances in the data (Fig. 1D). As expected, the segment type (i.e. tumour or stroma) accounts for most of the variation across PC1(23.25%) and PC2(13.45%). Interesting, PC3 (8.84%) separates the samples from the lips and tongues (Fig. 1D) middle column orange vs purple circles) while PC2 and PC3 separates the broader tumour stages (i.e. between Early and Advanced stages). Both these observations agree with that seen in the cohort survival analysis.

The difference in protein abundances between difference factors in the data was explored via differential expression analysis (Supplementary Table 1). We first compared between samples from different anatomical regions in either the Stroma or Tumour compartments. In the tumour compartments (Fig. 1E top left), EpCAM (associated with aggressive tongue cancer phenotype)⁵, Cytokeratin 19 and tumour marker CA9 are lower in the Lip compared to the Tongue. While in the stroma compartment (Fig. 1E top right) there appears to be lower expression of fibronectin, alpha smooth muscle actin (α-SMA), osteopontin and proteins associated with higher metastatic potential in the Lip^6,7. When comparing tumour compartments from the Pharynx with those from the Tongue compartment (Fig. 1E bottom left), a higher expression of PD-L1 (associated with tongue cancers) was found to be in the Tongue and conversely higher expressions of histone modifications (associated with oral carcinogenesis and aggressive phenotype) were found in the Pharynx^8,9. In the stroma compartment comparing Pharynx vs Tongue (Fig. 1E bottom right), lower expression of osteopontin, CA3 and CCR6 (associated with metastasis in head and neck cancers) are found in the Pharynx^10,11.

The samples were then grouped into 3 survival groups based on their OS and DFS durations and DE between the groups within either compartments or anatomical locations were assessed. Limited DE proteins were found for most DFS comparisons (Supplementary Table 1), which may be due to confounding treatments and/or intent of treatments. On the other hand, for OS, within the stroma compartment (Fig. 2F bottom), higher expression of Wnt-related SFRP protein (inactivation of which is linked with oral carcinoma) was found in patients with better outcomes¹². For the tumour compartment (Fig. 1F top), higher expression of interferon stimulated gene–15 (ISG15) protein is found in patients with poor outcomes, whereas high expression of CD44 was found in patients with improved survival. ISG15 has been found to be elevated in 80% of oral carcinomas¹³.

We then compared the OS groups between samples from different anatomical locations. In oral cavity (OC) samples (Fig. 2G left), there is a higher expression of ISG15 (known to be elevated in oral carcinoma)¹³, IFIT1 (promotes metastasis)¹⁴ proteins as well as histone modifications (associated with oral carcinogenesis and aggressive phenotype)¹⁵ in patients with poor outcomes. In Tongue samples (Fig. 2G middle), a higher expression of MMP8 (which inhibits cancer invasion and progression) protein was found in the patients with good outcomes¹⁶. In Pharynx samples (Fig. 1G right), the S100 proteins (including A8, A9 and A12) are found to be elevated in patients with poor outcomes. It is known that S100 proteins are associated with poor cancer prognosis with S100A9 linked to regulation of MMP7¹⁷. These results suggest that, with the targeted protein panel afforded by the breadth of coverage from the GeoMx IPA assay, it is able allow the elucidation of known biology and markers in the TME of mucosal HNSCC.

Identification of proteomics features associated with patient survival

To assess how well each marker’s expression associates with patient’s overall survival or disease-free survival (DFS), we conducted a feature association analysis based on a univariate cox proportional hazards (CoxPH) regression model and Kaplan Meier (KM) survival analysis approach. In terms of DFS, 30 out of 68 eligible cases have an event while for OS, 21 out of the 68 cases have an event. The feature association analysis resulted in 61 proteins identified to (with CoxPH wald p-value < 0.01) associate with DFS outcomes based on stroma segments and survival group information. These proteins were split into either hazard ratio (HR) < 1 or HR > = 1 groups where KM survival analysis and long rank test was then performed on the groups stratified based on the expression levels of the upper and lower quartiles of each protein. 52 protein features were identified at a significance of p value < 0.05 (22 HR ≥ 1,30 HE < 1) (Fig. 2A). Using the similar approach, 84 proteins from the CoxPh results were filtered down to 74 proteins for DFS in the tumour segments with 51(44) and 73(64) proteins identified for OS in the stroma (tumour) segments (Fig. 2, Supplementary Tables 2, 3 and 4). Significant markers of relevance identified includes α-SMA, Interferon gamma (IFN-γ), FGF2 and Adenosine Receptor A2a (ADORA2A) associated with DFS in the Stroma while CD34, CD44, FoxP3 and CD3E are associated with DFS in the tumour segments. ADORA2A has been described to hinder anti-tumour immunity by suppressing immune cells such as T cells and thereby associated with poor outcomes¹⁸. This observation agrees in this study in terms of OS. Interestingly however, higher ADORA2A expression within the tumour was found to be associated with improved disease-free survival. Identified markers including BRCA1 and CXCR5 are associated with better OS^19,20 while COL1A1 and SPP1 associates with poor survival and both of which have been cited to negatively impact patient outcomes by assisting with tumour invasion and progression^6,21. IFN-γ and actin alpha 2 (ACTA2) are associated with better disease-free survival while CD34 and functional immune cell marker²², FGF2, are associated with poor disease-free survival²³. IFN-γ is considered a protective serum cytokine that is involved in anti-tumour Th1 immune response and is often upregulated in HNSCC compared with healthy tissues²⁴. Our results are consistent with previous studies that found a linear correlation between downregulation of IFN-γ associated with regional progression²⁴. FGF2 associates with pro-tumorigenic phenotypes that are known to shift a tumour associated macrophages to an M2-like macrophage²⁵.

Spatial transcriptomics analysis

The GeoMx WTA dataset consists of 122 AOIs across 61 patient cores (each core with one AOI separated into tumour or stroma regions) with the expression matrices as integrated counts for 18815 transcripts. The data were analysed using the same pipeline as the IPA data. Briefly, 139 negative probes while 4 AOIs were removed by the AOI QC (Supplementary Fig. 3A). The counts data was then RLE normalised (Supplementary Fig. 3B) with PCA of the logCPM generated to highlight the factors contributing to the variations in the data (Supplementary Fig. 3C). One AOI was assessed to be an outlier with extremely small normalization factor and consequently removed. From the PCA, the tumour and stroma segments separate along PC1(28.27%) and PC2(13.27%). However unlike in the IPA, none of the other factors clearly separates out in any of the other PCs.

As with the IPA, differential expression analysis was conducted to investigate the differences in transcript abundances between the same sets of factors for the WTA data (results in Supplementary Table 5). The resulting DE genes from the WTA are compared with the corresponding DE proteins in the IPA results with limited overlaps noted (Supplementary Table 6). Similarly, we compare samples from different anatomical regions in either the Stroma or Tumour compartments. In the tumour compartments (Supplementary Fig. 3D top left), SOX9 (associated with promoting nasopharyngeal carcinoma metastasis²⁶) appears to be downregulated in the Lip compared to the Tongue. The protein expressed by this gene is also found to be DE in the IPA data. In the stroma compartment (Supplementary Fig. 3D top right), structural genes like FN1 (fibronectin), SPP1 (osteopontin)⁶, ACTA2 (α-SMA, smooth muscle actin), COL1A1 (Collagen I)²¹ and ITGA5 (Integrin alpha 5) are all downregulated in the Lip, in agreement with DE results from the IPA. Comparing tumour samples from the Pharynx with those from the Tongue compartment (Supplementary Fig. 3D bottom left), Stat3 and ErbB4 are common markers with the IPA results but with ErbB4 expression downregulated in the Pharynx, unlike in the IPA results. For the stroma samples between Pharynx and the Tongue (Supplementary Fig. 3D, bottom right), SPP1 appears to be downregulated in the Pharynx while IL12RB1 (an immune prognostic biomarker for oral squamous cell carcinoma²⁷) is upregulated in the Pharynx, unlike in the IPA where both IL12RB1 and PD-L1 proteins were found to be elevated in the Tongue.

When comparison samples grouped by OS and DFS durations within either compartments or anatomical locations, there are limited DE genes found for most DFS comparisons (Supplementary Table 6). Interesting for OS comparison in the tumour, there are 1794 DE genes obtained of which 41 are common with the IPA results. GSEA analysis suggest downregulation of key hallmark pathways including interferon alpha/gamma, EMT and apoptosis in the patients with a better outcome. There are limited DEGs obtained when comparing the OS groups between samples from different anatomical locations (Supplementary Table 5). Of interest is the downregulation of Histone deacetylase 8 (HDAC8) in patients with good outcomes for Tongue cancers with HDAC8 a potential therapeutic target for treating oral squamous cell carcinoma²⁸.

Cell type deconvolution identify cell types associated with patient survival

Cell type deconvolution was conducted to estimate the proportions of cell types in each sample analysed (Fig. 3A). There are clearly more malignant cells in the tumour compartments compared to the stroma compartments, providing confidence that the segmentation strategy is working reasonably well. In terms of the stroma compartments, major cell types include Plasma, Fibroblast, T cells and Epithelial cells. Comparing differences between cell type proportions did not yield any significant differences (Supplementary Table 7) and we procced to analyse association of the cell type proportions with patient survival. To this end, we applied univariate CoxPH (p < 0.05) and KM (p < 0.05) analyses with cell type proportions as features and associating with DFS and OS events. For tumour samples (Fig. 3B-C), higher proportions of Plasma and Mast cells were found to associate with better outcomes (both DFS and OS) while Lymphovascular cells associate with poorer outcomes. On the other hand, for stroma samples (Fig. 3D-E), B cells, Endothelial and Fibroblast cells all associate with poorer outcomes (DFS and OS) and again higher proportions of Plasma cells associate with better outcomes.

Identification of transcriptomics features associated with patient survival

Using the same feature association analysis approach used in the IPA data, we assess individual transcript’s (feature) expression association with patients’ OS or DFS. For DFS, 24 (27) out of 57 (60) eligible stroma (tumour) cases relapsed and, 17 (19) out of the 57 (60) stroma (tumour) cases died. The feature association analysis result in hundreds of transcripts identified to (with CoxPH wald p-value < 0.01, KM p-value < 0.05) associate with each category (i.e. OS/DFS outcomes based on segments and survival groups) (Supplementary Tables 8 and 9). Looking at the top transcripts identified for each category in either CoxPH or KM analysis (DFS in Supplementary Fig. 4 and OS in Supplementary Fig. 5), significant features of relevance identified includes SLC4A1 (solute carrier family 4 member 1, a transmembrane bicarbonate transporter involved in pH regulation, cell migration and can contribute to oxidative stress dysregulation²⁹) and MSLN (Mesothelin, a glycoprotein that is considered a potential therapeutic target, with some studies finding correlations with immune cell infiltration into the TME, as well, it has also been associated with tumour metastasis, growth and invasion³⁰) and TSPO (low TSPO expression in HNC has been associated with decreased 5 -year survival³¹) with DFS in the Stroma. NPC1L1 (Niemann-Pick C1-Like 1, contributes to cholesterol absorption and involved maintaining redox balance³²) and MNAT1 (menage a trois 1, involved in the PI3K/AKT/mTOR pathway and has been associated with chemoresistance in osteosarcoma³³) were seen to be elevated in patients with longer DFS in the tumour segments. Of relevance, given the relationship between MNAT1 and AKT, we found MNAT1 upregulated in the RNA data and AKT was upregulated in the protein analysis.

For overall survival, we found that in the stroma, increased expression of IZUMO1R (known to be expressed in CD4 T cells, and in particular Tregs³⁴), was seen in tumours that had an improved survival (Supplementary Fig. 5). Additionally, increased expression of HS2ST1 (enables proteoglycan interactions in the TME and has been linked with decreased stromal cell infiltration in other ³⁵) in the tumour was found to be associated with decreased survival.

Differences between proteomics and transcriptomics assay

Comparing between the protein and transcript assays, we looked firstly at how well the analytes correlate for matching samples and within each segment. While protein and RNA expression does not always correlate³⁶, we noted that RNA/protein that correlates well (i.e. R > 0.5) are mostly structural (cadherin, keratin and integrin) and antigen recognition genes (Fig. 4A) while most uncorrelated genes are functional genes related to metabolic and chemokine functions. These findings highlight critical differences between utilising assay types to study biological functions.

We then compared the markers identified as having significant association with patient survival between IPA and WTA assays (Fig. 5B-C). With the difference in panel size, it is unsurprising that WTA will have significantly greater number of identified markers and with greater overlaps. For the WTA and within the tumour (stroma), 226 (252) genes were common between OS and DFS, with 303 (166) and 225 (301) unique genes respectively. For the IPA analysis within the tumour (stroma), 19 (7) proteins were common across OS and DFS, with 17 (9) and 19 (14) unique proteins respectively. Only FN1, was found to be common in both assays within the tumour while the genes (or protein encoded by the genes) ADORA2A, SPP1, ACTA2, CTNNB1 and HSP90AA1 were common in both assays within the stroma (Fig. 4C). As is expected, most of these are structural proteins which correlates well between the assays.

Through assessing both the IPA and WTA assays, we were interested in uncovering what markers might be associated with clinical response that were unique to protein expression and not found within the WTA (Supplementary Table 10). These markers may be more translatable in the translational setting which, with the more focused IPA panel, can be appropriately identified in the analysis. We identified several lymphocyte markers unique to the IPA analysis that were linked to alternate clinical outcomes within the tumour regions. Firstly, we identified high intra-tumoral CXCR5 expression in patients with improved outcomes. CXCR5 is a chemoattractant cytokine receptor that is typically expressed on blood and peripheral lymph node B-cells and some subsets of T cells and is involved in the trafficking of these lymphocytes from the blood to lymphoid organs³⁷. Consistent with our findings, CXCR5 expression in HNSCC has been associated with the positive infiltration of lymphocytes and has been correlated with improved survival outcomes^20,38. There is extensive research to suggest that lymphocytes play a crucial role in anti-tumour immunity and patient outcomes. CD3e, a hallmark lymphocyte marker, was also found to be upregulated within the tumour of patients with an HR < 1 for both disease-free survival and overall survival (Fig. 2C, Supplementary Table 10). While lymphocytes can be anti-tumorigenic, there are subsets that can have pro-tumorigenic effects, and therefore characterising the subtypes of lymphocytes present within the TME allows us to better appreciate the dynamics and state of the disease. Using the IPA panel, we were able to identify functional lymphocyte markers that were associated with different patient outcomes, and that were not differentially expressed in their RNA form. Cytotoxic T lymphocyte antigen 4 (CTLA-4) is an immune checkpoint marker and is expressed primarily on T cells, with CTLA-4 + Tregs having the ability to produce immunosuppressive molecules that induce T cell dysfunction, exhaustion, and negatively regulate the immune response³⁹. Furthermore, clinical trials using CTLA-4 inhibitors as an immunotherapy in conjunction with radiotherapy and cetuximab, have shown benefit in a subset of patients that are not highly expressing PD1, LAG3 or CD39⁴⁰. Within our data we found higher levels of CTLA-4 within the tumour regions of patients with a worse disease-free survival and overall survival (Fig. 2D, Supplementary Table 10). In addition, we see higher levels of FOXP3, commonly expressed on Tregs³⁴, in both poor DFS and OS (Fig. 2C, Supplementary Table 10). From these findings we infer that these patients have an increased infiltration of immunosuppressive Tregs and exhausted T cells. Given that we see tumour infiltrating lymphocytes present in both poor and good survival, being able to functionally characterise into immune subtypes is prognostically valuable.

Immunofluorescent- proteomic validation

From the list of significant markers identified from the IPA dataset for DFS and OS, two proteins, CD34 and CD44, were amongst the proteins found to contribute to clinical outcomes and also featured in the PCF panel. To validate the findings from the IPA analysis, we analysed a single cell multiplex PC dataset from a serial section to orthogonally validate the results. For the validation using PCF, data were acquired and during quality control, a few cores were excluded from downstream analysis due to poor tissue staining/quality and artifacts (substantial blurs/bubbles). We also noted that there are inherent differences between the two assays for instance, the IPA assay looks at total intensity across a region, based on quantified expression, whereas the PCF data uses single cell binary classifications of cells based on quantification of immunofluorescent intensity signals. However, despite these caveats, we were able to validate the findings across the technologies.

We found that in the IPA analysis CD44 had a HR of > 1 in the tumour and was associated with DFS and OS. Additionally, CD34 was found to be associated with a HR of > 1 in the tumour for DFS (Fig. 2, Supplementary table 3). Across various cancer types, including HNSCC, CD44 expression is often considered a cancer stem cell marker and is associated with aggressive tumours, disease reoccurrence and worse patient outcomes⁴¹. Interestingly, within our cohort we found CD44 expression differentially expressed in patients with better survival. Contrary to most current research, across both proteomic technologies (IPA and PCF), high CD44 expression in the tumour was associated with improved survival (HR < 1). When we compared the survival of patients in lowest and highest quartiles (Low: n = 13, High: n = 13) for CD44 + PanCk + cells within the tumour (Fig. 5A) we found that patients had an increased DFS (p = 0.05) (Fig. 5B) and OS, although not significant, (p = 0.14) (Supplementary Fig. 6) associated with increased proportion of CD44 + PanCk + cells, validating the findings from the IPA analysis (Fig. 2C) where CD44 expression is associated with good OS and DFS outcomes in the tumour segment.

CD34 protein is a hematopoietic and endothelial stem cell marker, involved in the formation of blood vessels during injury response^22,42. CD34 is also thought to facilitate the adherence of specialised cells to lymphocytes, as well as being involved in tumour angiogenesis, the promotion of tumour reoccurrence and metastasis, and a marker on fibroblast progenitors²². It has been found that in the peripheral blood of HNSCC patients, CD34 + cells depressed the functions of T-lymphocytes, potentially through the release of immunosuppressive cytokine transforming growth factor-β, negatively impacting patient outcomes²³. In our study, for both the IPA and PCF datasets, CD34 was found to be associated with poor DFS. Specifically, CD34 expression was found to be upregulated in the tumour region of patients with worse DFS in the IPA dataset (Supplementary Table 3). Using the PCF data, we assessed the proportion of CD34 + PanCk- cells within the tumour masked regions and conducted the survival analysis to validate the findings from the IPA analysis (Fig. 5C). Samples that are lower than the 25th percentile (n = 13), or greater than the 75th percentile (n = 13) we see a significant difference (p = 0.013, Fig. 5D) in line with the results found for the IPA analysis.

Fundamental to deciphering clinical responses to therapy in HNSCC is a comprehensive understanding of the tumour microenvironment underpinning the complex interactions between tumour and immune cells within the tumour contexture. High-plex spatial biology is an enabling technology allowing us to understand the complex interplay between tumour and immune cells and try to pin-point alterations in the TME that might contribute to treatment outcomes. Whilst spatial transcriptomics methodologies have evolved rapidly in plexity over the last few years, spatial proteomics panel development and higher content screening and panel development has been slow. In this study, we present a world-first 580-plex antibody panel screen in HNSCC and provide an integrated workflow for spatial proteomics/transcriptomics and technical validation. Despite limited treatment information, our study identified many proteins, including some that are targets of clinical trials, that may be associated with OS or DFS. Furthermore, these included alterations that were only seen at a protein level and would be missed by transcriptome analysis alone. The GeoMx IPA assay is powerful in that it offers the highest plex spatial protein assay available, enabling scientists to characterise alterations within the TME to potentially identify biomarkers of resistance or response, as well as highlight novel proteins that may be potentially useful as drug targets. From a translational perspective, developing high-plex antibody panels as described in this study provides a rationale and workflow for biomarker discovery and translational cancer research application. Through the thorough analysis of the TME across a large cohort of mucosal HNSCC samples, we were able to demonstrate the power of the spatial proteomics and transcriptomics as novel translational methodologies to understand disease resistance.

Clinical Samples

For this discovery study, a mucosal head and neck squamous cell carcinoma (HNSCC) tissue microarray (TMA) was sourced from TriStar technologies (USA) that consisted of 84 patient tissue samples with 1 mm single core per patient (TA1937, SKU: 69571937). Informed written consent was obtained from the collaborating hospital sites and this study has University of Queensland Human Research Ethics approval. Patient specimens were collected from 14 females and 71 male participants during tumour resection surgery, where patients then went on to receive chemotherapy and/or radiotherapy, or palliative chemotherapy and/or radiotherapy, or no further treatment. Prior treatment details and dates of specimen collection were not available for this analysis. The median age of patient was 67 years with a range from 41 to 89 years, with 14 female and 71 males in the cohort. Follow up clinical data included status, smoking status, DFS and OS (Table 1). Pathologists reviewed whole sections prior to coring representative tumour regions for this assay. Samples were collected across a range of sites throughout the head and neck regions, including Tongue, Oral Cavity, Pharynx and Lip. Clinical information is known regarding DFS, OS, TNM stage, smoking status, age and gender and are showing in (Table 1). The p16 status and purpose of surgery is not known. AJCC cancer staging manual 7th edition was used to classify patients into stages based on the location of the patient’s tumour and pTNM (pathologic tumour node metastasis) classification⁴³. Serial formalin-fixed paraffin-embedded (FFPE) TMA slides were processed for the Nanostring Technologies Whole Transcriptome Atlas (WTA) and Immune-Oncology Proteome Atlas (IPA) panels, and Akoya Biosciences PhenoCycler-Fusion (PCF) as per the manufacturer’s instructions⁴⁴. 72 samples were subject to at least one of the assays across the study, with 69 cores processed for the IPA analysis, 62 for the WTA and 51 for the PCF data (Fig. 6).

Table 1

Patient cohort information
No. of patients	84
Age in years, median (range)	67 (41–89)
Sex
Male	70 (83%)
Female	14 (17%)
Chemotherapy
Yes	22 (26%)
No	58 (69%)
Unknown	4 (5%)
Radiotherapy
Yes	50 (60%)
No	29 (34%)
Unknown	5 (6%)
Smoker
Yes	54 (64%)
No	8 (10%)
Unknown	22 (26%)
Tumour location
Lip	27 (32%)
Oral cavity	14 (17%)
Pharynx	28 (33%)
Tongue	15 (18%)
Stage
1	32 (38%)
2	19 (23%)
3	8 (10%)
4a	16 (19%)
4b	2 (2%)
Unknown	7 (8%)

GeoMx Immune-Oncology Proteome Atlas (IPA)

The Nanostring Technologies IPA panel consists of 580 clinically relevant antibodies that help characterise the immune profile of the tumour microenvironment. All antibody clones are from Abcam’s IHC validated immuno-oncology antibody collection and is designed to capture proteins relevant to all pillars of the Hallmarks of Cancer⁴⁵. The staining protocol for the IPA experiment was carried out as per the manufacturer's guidelines. Briefly, FFPE tissue slides were first baked in a 60°C drying oven, with the baking duration adjusted according to the thickness of the tissue sections. After baking, the slides underwent deparaffinization, rehydration, and antigen retrieval to prepare the tissue for staining. A blocking step was then performed to prevent non-specific binding, ensuring optimal conditions for the subsequent primary antibody incubation. The slides were incubated overnight with primary antibodies, which included module antibodies and morphology markers, excluding the nuclei stain. Following the incubation, post-fixation and nuclei staining were conducted to complete the staining process. SYTO-13, CD45, and PanCk were employed as morphology markers to stain and visualize the nuclei, immune cells, and epithelial cells, respectively. The stained tissue sections were subsequently processed using the NanoString Technologies GeoMx Digital Spatial Profiler (DSP) to collect oligonucleotide tags attached to conjugated antibodies within user-defined regions of interest (ROIs). Region of interest selection aimed to capture tissue areas with the highest number of nuclei and the greatest likelihood of immune infiltrates. ROI sizes ranged from 400 µm to 660 µm in diameter and were segmented into tumour (PanCk+) and non-tumour (PanCk-) regions with 5 µm segment dilation to avoid cross contamination. To ensure consistent sampling areas between the two experiments, the ROIs from the WTA experiment was overlaid with images imported from the IPA experiment within the GeoMx collection windows. The overlaid IPA images were manually aligned with the WTA scans, and adjustments were made to their visibility and opacity. These modifications were saved on the tissue sections, and concordant IPA/WTA ROIs were collected.

Sequencing

Sequencing was performed by the Australian Genomics Research Foundation (AGRF). Eluted oligos were captured in 96 well collection plates, sealed and stored at -80 until sequencing. NGS readout was performed as per the manufacturer's instructions by the Australian Genome Research Facility (AGRF, Melbourne, AU). Briefly, eluates were dried at 65°C with gas permeable seals and eluates were resuspend in 10 µl and 80 µl H₂O respectively. Library generation was performed using ProCode indices. Libraries were purified, 1–2% PhiX spiked in, and sequenced on Novaseq X 10B flow cells with v1.5 reagent kits. Required read depth was estimated at 200 reads/µm² for the IPA assay. FASTQ files were processed through the GeoMx NGS pipeline and DCCs uploaded to the instrument for sample alignment. Initial QC was performed on the instrument using default parameters then exported into csv for bioinformatic analysis.

GeoMx Whole Transcriptome Atlas (WTA)

Using the NanoString WTA, over 18,000 genes were characterized. The WTA staining protocol, while similar, had some differences from the IPA procedure and was also performed according to the manufacturer's instructions. Briefly, FFPE slides were initially subjected to baking, deparaffinization, and rehydration. After these steps, target retrieval was carried out, and the slides were incubated in a Proteinase K solution to expose RNA targets for in situ hybridization (ISH) using the RNA Probe Mix. Following the overnight ISH, stringent washes were performed to remove off-target probes. Lastly, morphology markers (SYTO-13, CD45, and PanCk), including a nuclear stain, were applied to the tissues to complete the staining process. ROI selection was performed in the GeoMx collection window, to create regions for tumour and stromal regions across high quality cores on the TMA. The stained tissue sections were subsequently processed using the DSP to collect oligonucleotide tags attached to transcripts from the ROIs. Sequencing of the WTA followed similar methods to what is described for the IPA but differed in two ways. Library generation was performed using SeqCode indices and sequencing had an estimated read depth of 100 read/ µm². Initial QC was performed on the instrument using default parameters then exported into csv for bioinformatic analysis using various packages in R, similar to the IPA analysis.

Proteomic spatial profiling (Akoya Biosciences PhenoCycler-Fusion)

A serial section of the FFPE TMA was stained using 46 antibodies on the PhenoCycler-Fusion sample preparation workflow as per manufacturer’s instructions⁴⁴.This technology works through cyclic imaging of up to three fluorophores at once. This is achieved by staining the tissue with a cocktail of antibodies, where each antibody is conjugated to a unique oligonucleotide sequence. Then in cycles, complimentary oligonucleotides that are conjugated to one of three fluorophores are applied to the tissue (each cycle can have up to three reporters that each are attached to a different fluorophore) imaged and washed away. Post-imaging, the images are digitally stitched together and processed. To prepare the slides for imaging, wax was removed by baking the slide in a 60°C oven for 30 minutes, followed by immersion in HistoChoice (H2779-1L, VWR). Next, slides were hydrated by washing slides in decreasing concentrations of ethanol followed by double distilled water (100% EtOH, 100% EtOH, 90% EtOH, 70% EtOH, 50% EtOH, 30% EtOH, ddH2O, ddH₂O). Antigen retrieval was performed under pressure and heat, using the AR9 buffer (AR9001KT, Akoya Biosciences). Samples were incubated with antibodies overnight (~ 16 hours) in a humidity chamber at 4°C. The slide then moved through washes in Storage buffer (232107, Akoya Biosciences), Staining buffer (240198, Akoya Biosciences), 1.6% PFA (C004, ProSciTech) (diluted in Storage Buffer), PBS and methanol before a Final Fixative was applied to the tissues. Following a final wash in PBS, a flow cell was attached to the slide to create a chamber for liquids to flow through and distribute reporters across the tissues. Fluorescent images were captured for each marker and at the completion of all cycles, images were post-processed to remove background and autofluorescence. During post-image processing the images are stitched into a qptiff file for analysis.

Data generated using the PhenoCycler-Fusion and was imported into QuPath⁴⁶ for visualisation, initial quality control and cellular segmentation. During quality control, tissues that had partially or fully lifted, or had a large proportion of artifacts, were excluded from downstream analysis. Markers that had severe non-specific binding were also excluded. Cell segmentation was performed across the TMA using Cellpose⁴⁷ (v.2.0) based on DAPI staining and cell expansion. Within QuPath, an artificial neural network pixel classifier was trained to generate tumour and stromal region masks based on PanCk expression. The unique object ID, x-y co-ordinates, tumour/stromal classification, as well as the median expression for each marker, core ID, and cellular morphology features for each cell were exported from QuPath as a csv and imported into anndata format and was subject to data analysis in Python^48,49.

Bioinformatics Analyses

GeoMx IPA spatial proteomics data analysis:

Pre-processing, quality control and normalization: The data for the GeoMx IPA were measurements of protein abundance of 570 proteins, 5 housekeeping proteins (Histone H3, GAPDH, RPS6, Calreticulin and TOMM20) and 5 background control (Rat IgG2a, Mouse IgG2b, Hmr IgG, Rabbit IgG, and Mouse IgG1) probes. The raw data used for processing is the probe normalised protein (probeQC) counts for each AOI which were imported with pre-processing, quality control, normalization conducted based on the standR⁴ pipeline with modifications. These protein expression matrices (from all cores) and sample metadata, and clinical data are integrated and incorporated as a SpatialExperiment object in R for analysis. Integration involves mapping each protein (isoform) to the corresponding gene/transcript via the NCBI annotations. Initial assessment includes investigating the correlations between logCPM of the housekeeping and background control proteins in the data. Quality control (QC) was performed on this probeQC data, with protein level QC applied by using either edgeR::filterByExpr or standR::addPerROIQC which filters out low expressing proteins with less than 5 counts per sample in more than 90% of the samples. Sample QC filters out low expression AOIs based on protein expression depth (> 100,000 per AOI) and nuclei count (> 100 cells per AOI). Relative log expression (RLE) plots and principal component analysis (PCA) across AOIs were utilised to identify factors associated with biological variations or technical variations (batch effects). The "relative log expression” (RLE) scaling factor method (implemented in R package edgeR⁵⁰ (v4.2.1) calcNormFactors function⁵⁰) was used to normalize the dataset to remove compositional bias within the data.

Differential expression analysis: Differential expression (DE) analysis was performed using edgeR and limma (v3.60.4) R packages^50,51. Very briefly, DE was modelled using linear models with experimental, clinical, and biological factors as predictors. Proteins were translated to genes to utilise the statistical packages for DE analyses. An empirical Bayes approach was used to estimate the common and gene wise variation in order to model each gene’s variation while borrowing information from all other genes. A linear model was then fitted to an experimental design and required contrasts were applied to query for differential expression. In this study, the limma-voom-eBayes pipeline was applied using edgeR::voomLmFit and limma::eBayes functions⁵¹. The statistical significance threshold was defined as an adjusted p-value of ≤ 0.05 based on the Benjamini Hochberg procedure. The factors of interest tested include A) between tissue locations within stroma or tumour segment B) Good vs Poor OS or DFS outcomes in each tissue location.

Feature association with survival outcomes: Univariate cox proportional hazards (CoxPh) regression model, and Kaplan Meier (KM) survival analysis was performed on the data for DFS and OS using R packages survival and survminer. In the CoxPh model, DFS or OS durations are treated as the time variable while the event variable was defined by occurrence of DFS events (DFS < OS) or deceased status respectively. Features with a wald statistical value of < 0.01 is then follow up using the KM survival analysis to refine the identified features. For the KM analysis, each feature’s expression value is utilised to group samples into three categories (high: top quantile, low: bottom quantile, mid: others). Features with a p-value < 0.05 for the KM analysis will then be considered significantly associated.

GeoMx Whole Transcriptome Atlas (WTA) spatial transcriptomics data analysis:

Pre-processing, quality control and normalization: The data for the GeoMx WTA were measurements of RNA abundance of 18,529 protein-coding transcripts, 147 others and 139 negative probes. These negative probes were utilised by Nanostring’s data generation pipeline to get the probe normalised protein (probeQC) counts and are removed from downstream analysis. The QC pipeline used is as per that for the GeoMx IPA with gene level QC using either edgeR::filterByExpr or standR::addPerROIQC and sample QC filtering out low expression AOIs based on gene expression depth (> 200,000 per AOI) and nuclei count (> 100 cells per AOI). The (RLE) scaling factor was used for normalisation with both RLE and PCA plots utilised to assess and identify either factors associated with biological or technical (batch effects) variations and/or outlier samples to be removed. For the WTA, both the differential expression analysis and feature survival association analysis were conducted using the same pipeline as that of the GeoMx IPA described above.

Cell type deconvolution and survival analysis

Cell type deconvolution was conducted on the transcriptomic data using CIBERSORTx⁵² to estimate the different cell type proportions in each of the samples by providing estimations of cell type abundances in the mixed cell population. Existing HNSCC single-cell dataset by Kürten et al.⁵³ was utilised as a reference and ‘S batch correction’ was applied to correct platform effects between scRNA-seq and bulk RNA-seq data. Cell labels in the reference dataset were used and all unlabelled cells were removed before running deconvolution. Using the cell type proportions for each sample, differences between cell type proportions were conducted using the propeller method is in the R package speckle⁵⁴. Similarly, survival analysis was performed using univariate cox proportional hazards (CoxPH) regression model and Kaplan Meier (KM) survival analysis on the data for DFS and OS as described above. In this case both p-value cutoff was set as 0.05.

Orthogonal Validation of IPA Protein findings using single-cell protein profiling

During the IPA analysis, differential expression of proteins in the tumour and stromal regions were compared against clinical endpoints using CoxPH model and Kaplan Meier analysis. Lists of significant proteins found in the tumour and stromal regions for DFS and OS were compared to the list of proteins used in the high plex PCF panel. Five markers from the significant IPA markers were used in the PCF panel, but due to non-specific binding for the PCF markers, only two markers, CD34 and CD44, were used to orthogonally validate the IPA findings at a cohort level.

To validate the protein expression found in the IPA analysis, we analysed expression of two markers, CD34 and CD44, from the multi-plex immunofluorescent dataset. Expression was normalised using an arcsinh normalisation with a co-factor set to 150, and percentile transformation. Harmonypy was used to integrate the high-dimensional data across patient samples⁵⁵. Next, PCA, Scanpy neighbours (neighbours = 15), and UMAP were performed as initial preprocessing before downstream analysis⁵⁶.

Using the gating classifier that is based on a Gaussian mixed model, mmochi, cells were assigned tumour and nontumor phenotype based on Pan-Ck positivity or negativity status, respectively, as well as their and positive and negative status for both CD44 and CD34⁵⁷. These new labels were reimported into QuPath to visually assess the quality label assignment. As the GeoMx data were based on regional expression, we used the proportion of positive cells of each marker for time disease free and overall survival in the tumour or stromal regions. Cellular proportions were calculated for CD34 + and CD44 + cells, and their associated PanCK status, for each core and their sub-regions. Survival analysis was performed using Kaplan Meier from the survival and survminer packages in RSurvival analysis was performed using Kaplan Meier from the survival and survminer packages in R. log-rank based statistical approach was used to determine significance between groups.

Data availability:

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Acknowledgements

This study was supported by the Passe and Williams Foundation for AK & BGMH and the Princess Alexandra Research Foundation for CB, RL & AK. The authors would like to acknowledge the Wesley Research Institute and the AGRF Sequencing facilities.

Conflicts of Interest

Authors AR, SB, MC, BF, CK, JMB are employees of Nanostring Technologies (Bruker). AK is on the Scientific Advisory Board for Omapix Solutions, Predxbio, Molecular Instruments and Visiopharm. All other authors declare no financial or non-financial competing interests.

Author Contributions

Concept: AK, CWT, BGMH

Experimental: JM, HS, VYN

Analysis: CWT, NB, MD, JM

Writing and critical review: all authors

Bray, F. et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA: A Cancer Journal for Clinicians 74, 229–263 (2024). https://doi.org/https://doi.org/10.3322/caac.21834
Haring, C. T. et al. Patterns of recurrence in head and neck squamous cell carcinoma to inform personalized surveillance protocols. Cancer 129, 2817–2827 (2023). https://doi.org/https://doi.org/10.1002/cncr.34823
Ruffin, A. T. et al. Improving head and neck cancer therapies by immunomodulation of the tumour microenvironment. Nature Reviews Cancer 23, 173–188 (2023). https://doi.org/10.1038/s41568-022-00531-9
Liu, N. et al. standR: spatial transcriptomic analysis for GeoMx DSP data. Nucleic Acids Research 52, e2-e2 (2023). https://doi.org/10.1093/nar/gkad1026
Yanamoto, S. et al. Clinicopathologic significance of EpCAM expression in squamous cell carcinoma of the tongue and its possibility as a potential target for tongue cancer gene therapy. Oral Oncol 43, 869–877 (2007). https://doi.org/10.1016/j.oraloncology.2006.10.010
Xu, X. L. et al. SPP1 and FN1 are significant gene biomarkers of tongue squamous cell carcinoma. Oncol Lett 22, 713 (2021). https://doi.org/10.3892/ol.2021.12974
Maleš, J., Mihalj, H., Šestak, A., Kralik, K. & Smolić, M. Osteopontin Levels in Patients with Squamous Metastatic Head and Neck Cancer. Medicina (Kaunas) 57 (2021). https://doi.org/10.3390/medicina57020185
Akisada, N. et al. PD-L1 expression in tongue squamous cell carcinoma. Medical Molecular Morphology 54, 52–59 (2021). https://doi.org/10.1007/s00795-020-00261-7
Yoshida, S. et al. Significance of PD-L1 Expression in Tongue Cancer Development. Int J Med Sci 15, 1723–1730 (2018). https://doi.org/10.7150/ijms.27860
Chu, Y. H. et al. Carbonic Anhydrase III Promotes Cell Migration and Epithelial-Mesenchymal Transition in Oral Squamous Cell Carcinoma. Cells 9 (2020). https://doi.org/10.3390/cells9030704
Lu, E., Su, J., Zhou, Y., Zhang, C. & Wang, Y. CCL20/CCR6 promotes cell proliferation and metastasis in laryngeal cancer by activating p38 pathway. Biomedicine & Pharmacotherapy 85, 486–492 (2017). https://doi.org/https://doi.org/10.1016/j.biopha.2016.11.055
Sogabe, Y. et al. Epigenetic inactivation of SFRP genes in oral squamous cell carcinoma. Int J Oncol 32, 1253–1261 (2008). https://doi.org/10.3892/ijo_32_6_1253
Laljee, R. P. et al. Interferon stimulated gene-ISG15 is a potential diagnostic biomarker in oral squamous cell carcinomas. Asian Pac J Cancer Prev 14, 1147–1150 (2013). https://doi.org/10.7314/apjcp.2013.14.2.1147
Pidugu, V. K. et al. IFIT1 and IFIT3 promote oral squamous cell carcinoma metastasis and contribute to the anti-tumor effect of gefitinib via enhancing p-EGFR recycling. Oncogene 38, 3232–3247 (2019). https://doi.org/10.1038/s41388-018-0662-9
Tachaveeraphong, W. & Phattarataratip, E. The Significance of Modified Histone H3 in Epithelial Dysplasia and Oral Cancer. Int Dent J 74, 769–776 (2024). https://doi.org/10.1016/j.identj.2024.01.011
Juurikka, K., Butler, G. S., Salo, T., Nyberg, P. & Åström, P. The Role of MMP8 in Cancer: A Systematic Review. Int J Mol Sci 20 (2019). https://doi.org/10.3390/ijms20184506
Koh, H. M., Lee, H. J. & Kim, D. C. High expression of S100A8 and S100A9 is associated with poor disease-free survival in patients with cancer: a systematic review and meta-analysis. Transl Cancer Res 10, 3225–3235 (2021). https://doi.org/10.21037/tcr-21-519
Ma, S.-R. et al. Blockade of adenosine A2A receptor enhances CD8 + T cells response and decreases regulatory T cells in head and neck squamous cell carcinoma. Molecular Cancer 16, 99 (2017). https://doi.org/10.1186/s12943-017-0665-0
Wang, Y. C. et al. Downregulation of ATM and BRCA1 Predicts Poor Outcome in Head and Neck Cancer: Implications for ATM-Targeted Therapy. J Pers Med 11 (2021). https://doi.org/10.3390/jpm11050389
Chen, J. et al. Effect of CXCR5-Positive Cell Infiltration on the Immune Contexture and Patient Prognosis in Head and Neck Squamous Cell Carcinoma. Onco Targets Ther 13, 5869–5877 (2020). https://doi.org/10.2147/ott.S248958
Shih, M. L. et al. Transcriptomic discovery of a theranostic signature (SERPINE1/MMP3/COL1A1/SPP1) for head and neck squamous cell carcinomas and identification of antrocinol as a candidate drug. Comput Biol Med 150, 106185 (2022). https://doi.org/10.1016/j.compbiomed.2022.106185
Radu, P. et al. CD34-Structure, Functions and Relationship with Cancer Stem Cells. Medicina (Kaunas) 59 (2023). https://doi.org/10.3390/medicina59050938
Pandit, R., Lathers, D. M. R., Garrity, T., Beal, N. M. & Young, M. R. I. CD34 + Immune Suppressive Cells in the Peripheral Blood of Patients with Head and Neck Cancer. Annals of Otology, Rhinology & Laryngology 109, 749–754 (2000). https://doi.org/10.1177/000348940010900809
Bussu, F. et al. IFN-γ and other serum cytokines in head and neck squamous cell carcinomas. Acta Otorhinolaryngol Ital 38, 94–102 (2018). https://doi.org/10.14639/0392-100x-1530
Im, J. H. et al. FGF2 alters macrophage polarization, tumour immunity and growth and can be targeted during radiotherapy. Nature Communications 11, 4064 (2020). https://doi.org/10.1038/s41467-020-17914-x
Xiao, B. et al. SOX9 promotes nasopharyngeal carcinoma cell proliferation, migration and invasion through BMP2 and mTOR signaling. Gene 715, 144017 (2019). https://doi.org/https://doi.org/10.1016/j.gene.2019.144017
Chen, Z. et al. IL-12RB1: a novel immune prognostic biomarker for oral squamous cell carcinoma and linked to PD-1/PD-L1 expression in the tumor immune microenvironment. Ann Transl Med 10, 144 (2022). https://doi.org/10.21037/atm-21-6915
Ahn, M.-Y. & Yoon, J.-H. Histone deacetylase 8 as a novel therapeutic target in oral squamous cell carcinoma. Oncol Rep 37, 540–546 (2017). https://doi.org/10.3892/or.2016.5280
Zhong, J., Dong, J., Ruan, W. & Duan, X. Potential Theranostic Roles of SLC4 Molecules in Human Diseases. International Journal of Molecular Sciences 24, 15166 (2023).
Li, Y. et al. MSLN Correlates With Immune Infiltration and Chemoresistance as a Prognostic Biomarker in Ovarian Cancer. Frontiers in Oncology 12 (2022). https://doi.org/10.3389/fonc.2022.830570
Tuominen, S. et al. TSPO is a potential independent prognostic factor associated with cellular respiration and p16 in head and neck squamous cell carcinoma. Frontiers in Oncology 13 (2023). https://doi.org/10.3389/fonc.2023.1298333
Fu, L. & Chen, Z.-S. Redox signaling-governed drug-tolerant persister cancer cell: a key spark of treatment failure. Signal Transduction and Targeted Therapy 7, 89 (2022). https://doi.org/10.1038/s41392-022-00940-0
Qiu, C. et al. MNAT1 promotes proliferation and the chemo-resistance of osteosarcoma cell to cisplatin through regulating PI3K/Akt/mTOR pathway. BMC Cancer 20, 1187 (2020). https://doi.org/10.1186/s12885-020-07687-3
Zarin, P. et al. Treg cells require Izumo1R to regulate γδT cell-driven inflammation in the skin. Proc Natl Acad Sci U S A 120, e2221255120 (2023). https://doi.org/10.1073/pnas.2221255120
Chung, T. T., Kim, S. K. & Lee, S. J. Prognostic significance of HS2ST1 expression in patients with hepatocellular carcinoma. Genes & Genomics 46, 1165–1174 (2024). https://doi.org/10.1007/s13258-024-01556-0
Gry, M. et al. Correlations between RNA and protein expression profiles in 23 human cell lines. BMC Genomics 10, 365 (2009). https://doi.org/10.1186/1471-2164-10-365
Murphy, P. M. in Clinical Immunology (Sixth Edition) (eds Robert R. Rich et al.) 215–227 (Elsevier, 2023).
Shen, Y. et al. Expression profile and prognostic value of CXCR family members in head and neck squamous cell carcinoma. World Journal of Surgical Oncology 20, 259 (2022). https://doi.org/10.1186/s12957-022-02713-z
Mei, Z., Huang, J., Qiao, B. & Lam, A. K.-y. Immune checkpoint pathways in immunotherapy for head and neck squamous cell carcinoma. International Journal of Oral Science 12, 16 (2020). https://doi.org/10.1038/s41368-020-0084-8
Chakraborty, R., Darido, C., Liu, F., Maselko, M. & Ranganathan, S. Head and Neck Cancer Immunotherapy: Molecular Biological Aspects of Preclinical and Clinical Research. Cancers (Basel) 15 (2023). https://doi.org/10.3390/cancers15030852
Kavitha, L. et al. Expression of CD44 in Head and Neck Squamous Cell Carcinoma-An In-Silico Study. Glob Med Genet 10, 221–228 (2023). https://doi.org/10.1055/s-0043-1772459
Hassanpour, M., Salybekov, A. A., Kobayashi, S. & Asahara, T. CD34 positive cells as endothelial progenitor cells in biology and medicine. Frontiers in Cell and Developmental Biology 11 (2023). https://doi.org/10.3389/fcell.2023.1128134
in From the AJCC Cancer Staging Manual (ed David R. Byrd Stephen B. Edge, Carolyn C. Compton, April G. Fritz, Frederick L. Greene, Andrew Trotti) XIX, 718 (Springer New York, NY, 2009).
Donovan, M. L. et al. Protocol for high-plex, whole-slide imaging of human formalin-fixed paraffin-embedded tissue using PhenoCycler-Fusion. STAR Protocols 5, 103226 (2024). https://doi.org/https://doi.org/10.1016/j.xpro.2024.103226
Hanahan, D. Hallmarks of Cancer: New Dimensions. Cancer Discovery 12, 31–46 (2022). https://doi.org/10.1158/2159-8290.Cd-21-1059
Bankhead, P. et al. QuPath: Open source software for digital pathology image analysis. Scientific Reports 7, 16878 (2017). https://doi.org/10.1038/s41598-017-17204-5
Pachitariu, M. & Stringer, C. Cellpose 2.0: how to train your own model. Nature Methods 19, 1634–1641 (2022). https://doi.org/10.1038/s41592-022-01663-4
Virshup, I., Rybakov, S., Theis, F. J., Angerer, P. & Wolf, F. A. anndata: Annotated data. bioRxiv, 2021.2012.2016.473007 (2021). https://doi.org/10.1101/2021.12.16.473007
Monkman, J. et al. Spatial insights into immunotherapy response in non-small cell lung cancer (NSCLC) by multiplexed tissue imaging. J Transl Med 22, 239 (2024). https://doi.org/10.1186/s12967-024-05035-8
Chen, Y., Chen, L., Lun, A. T. L., Baldoni, P. L. & Smyth, G. K. edgeR 4.0: powerful differential analysis of sequencing data with expanded functionality and improved support for small counts and larger datasets. bioRxiv, 2024.2001.2021.576131 (2024). https://doi.org/10.1101/2024.01.21.576131
Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 43, e47 (2015). https://doi.org/10.1093/nar/gkv007
Newman, A. M. et al. Determining cell type abundance and expression from bulk tissues with digital cytometry. Nature Biotechnology 37, 773–782 (2019). https://doi.org/10.1038/s41587-019-0114-2
Kürten, C. H. L. et al. Investigating immune and non-immune cell interactions in head and neck tumors by single-cell RNA sequencing. Nature Communications 12, 7338 (2021). https://doi.org/10.1038/s41467-021-27619-4
Phipson, B. et al. propeller: testing for differences in cell type proportions in single cell data. Bioinformatics 38, 4720–4726 (2022). https://doi.org/10.1093/bioinformatics/btac582
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nature Methods 16, 1289–1296 (2019). https://doi.org/10.1038/s41592-019-0619-0
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biology 19, 15 (2018). https://doi.org/10.1186/s13059-017-1382-0
Caron, D. P. et al. Multimodal hierarchical classification of CITE-seq data delineates immune cell states across lineages and tissues. bioRxiv (2024). https://doi.org/10.1101/2023.07.06.547944

Competing interest reported. Authors AR, SB, MC, BF, CK, JMB are employees of Nanostring Technologies (Bruker). AK is on the Scientific Advisory Board for Omapix Solutions, Predxbio, Molecular Instruments and Visiopharm. All other authors declare no financial or non-financial competing interests.

Download PDF

Reviewers agreed at journal
30 Oct, 2024
Reviewers invited by journal
28 Oct, 2024
Editor assigned by journal
21 Oct, 2024
Submission checks completed at journal
21 Oct, 2024
First submitted to journal
15 Oct, 2024

You are reading this latest preprint version

The development of a high-plex spatial proteomic methodology for the characterisation of the head and neck tumour microenvironment

Status:

Version 1

Abstract

Figures

Introduction

Results

HNSCC patient survival associate with anatomical locations and tumour stage

Spatial multicellular proteomics analysis

Identification of proteomics features associated with patient survival

Spatial transcriptomics analysis

Cell type deconvolution identify cell types associated with patient survival

Identification of transcriptomics features associated with patient survival

Differences between proteomics and transcriptomics assay

Immunofluorescent- proteomic validation

Discussion

Materials and Methods

Clinical Samples

GeoMx Immune-Oncology Proteome Atlas (IPA)

Sequencing

GeoMx Whole Transcriptome Atlas (WTA)

Proteomic spatial profiling (Akoya Biosciences PhenoCycler-Fusion)

Bioinformatics Analyses

GeoMx IPA spatial proteomics data analysis:

GeoMx Whole Transcriptome Atlas (WTA) spatial transcriptomics data analysis:

Orthogonal Validation of IPA Protein findings using single-cell protein profiling

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1