The evolution of lung cancer and impact of subclonal selection in TRACERx.

Lung cancer is the leading cause of cancer-associated mortality worldwide1. Here we analysed 1,644 tumour regions sampled at surgery or during follow-up from the first 421 patients with non-small cell lung cancer prospectively enrolled into the TRACERx study. This project aims to decipher lung cancer evolution and address the primary study endpoint: determining the relationship between intratumour heterogeneity and clinical outcome. In lung adenocarcinoma, mutations in 22 out of 40 common cancer genes were under significant subclonal selection, including classical tumour initiators such as TP53 and KRAS. We defined evolutionary dependencies between drivers, mutational processes and whole genome doubling (WGD) events. Despite patients having a history of smoking, 8% of lung adenocarcinomas lacked evidence of tobacco-induced mutagenesis. These tumours also had similar detection rates for EGFR mutations and for RET, ROS1, ALK and MET oncogenic isoforms compared with tumours in never-smokers, which suggests that they have a similar aetiology and pathogenesis. Large subclonal expansions were associated with positive subclonal selection. Patients with tumours harbouring recent subclonal expansions, on the terminus of a phylogenetic branch, had significantly shorter disease-free survival. Subclonal WGD was detected in 19% of tumours, and 10% of tumours harboured multiple subclonal WGDs in parallel. Subclonal, but not truncal, WGD was associated with shorter disease-free survival. Copy number heterogeneity was associated with extrathoracic relapse within 1 year after surgery. These data demonstrate the importance of clonal expansion, WGD and copy number instability in determining the timing and patterns of relapse in non-small cell lung cancer and provide a comprehensive clinical cancer evolutionary data resource.

eIF4A1-dependent mRNAs employ purine-rich 5'UTR sequences to activate localised eIF4A1-unwinding through eIF4A1-multimerisation to facilitate translation.

Altered eIF4A1 activity promotes translation of highly structured, eIF4A1-dependent oncogene mRNAs at root of oncogenic translational programmes. It remains unclear how these mRNAs recruit and activate eIF4A1 unwinding specifically to facilitate their preferential translation. Here, we show that single-stranded RNA sequence motifs specifically activate eIF4A1 unwinding allowing local RNA structural rearrangement and translation of eIF4A1-dependent mRNAs in cells. Our data demonstrate that eIF4A1-dependent mRNAs contain AG-rich motifs within their 5'UTR which specifically activate eIF4A1 unwinding of local RNA structure to facilitate translation. This mode of eIF4A1 regulation is used by mRNAs encoding components of mTORC-signalling and cell cycle progression, and renders these mRNAs particularly sensitive to eIF4A1-inhibition. Mechanistically, we show that binding of eIF4A1 to AG-rich sequences leads to multimerization of eIF4A1 with eIF4A1 subunits performing distinct enzymatic activities. Our structural data suggest that RNA-binding of multimeric eIF4A1 induces conformational changes in the RNA resulting in an optimal positioning of eIF4A1 proximal to the RNA duplex enabling efficient unwinding. Our data proposes a model in which AG-motifs in the 5'UTR of eIF4A1-dependent mRNAs specifically activate eIF4A1, enabling assembly of the helicase-competent multimeric eIF4A1 complex, and positioning these complexes proximal to stable localised RNA structure allowing ribosomal subunit scanning.

Corrigendum: Phylogenetic ctDNA analysis depicts early-stage lung cancer evolution.

This corrects the article DOI: 10.1038/nature22364.

Absent expansion of AXIN2+ hepatocytes and altered physiology in Axin2CreERT2 mice challenges the role of pericentral hepatocytes in homeostatic liver regeneration.

BACKGROUND & AIMS: Mouse models of lineage tracing have helped to describe the important subpopulations of hepatocytes responsible for liver regeneration. However, conflicting results have been obtained from different models. Herein, we aimed to reconcile these conflicting reports by repeating a key lineage-tracing study from pericentral hepatocytes and characterising this Axin2CreERT2 model in detail. METHODS: We performed detailed characterisation of the labelled population in the Axin2CreERT2 model. We lineage traced this cell population, quantifying the labelled population over 1 year and performed in-depth phenotypic comparisons, including transcriptomics, metabolomics and analysis of proteins through immunohistochemistry, of Axin2CreERT2 mice to WT counterparts. RESULTS: We found that after careful definition of a baseline population, there are marked differences in labelling between male and female mice. Upon induced lineage tracing there was no expansion of the labelled hepatocyte population in Axin2CreERT2 mice. We found substantial evidence of disrupted homeostasis in Axin2CreERT2 mice. Offspring are born with sub-Mendelian ratios and adult mice have perturbations of hepatic Wnt/ß-catenin signalling and related metabolomic disturbance. CONCLUSIONS: We find no evidence of predominant expansion of the pericentral hepatocyte population during liver homeostatic regeneration. Our data highlight the importance of detailed preclinical model characterisation and the pitfalls which may occur when comparing across sexes and backgrounds of mice and the effects of genetic insertion into native loci. IMPACT AND IMPLICATIONS: Understanding the source of cells which regenerate the liver is crucial to harness their potential to regrow injured livers. Herein, we show that cells which were previously thought to repopulate the liver play only a limited role in physiological regeneration. Our data helps to reconcile differing conclusions drawn from results from a number of prior studies and highlights methodological challenges which are relevant to preclinical models more generally.

Phylogenetic ctDNA analysis depicts early-stage lung cancer evolution.

The early detection of relapse following primary surgery for non-small-cell lung cancer and the characterization of emerging subclones, which seed metastatic sites, might offer new therapeutic approaches for limiting tumour recurrence. The ability to track the evolutionary dynamics of early-stage lung cancer non-invasively in circulating tumour DNA (ctDNA) has not yet been demonstrated. Here we use a tumour-specific phylogenetic approach to profile the ctDNA of the first 100 TRACERx (Tracking Non-Small-Cell Lung Cancer Evolution Through Therapy (Rx)) study participants, including one patient who was also recruited to the PEACE (Posthumous Evaluation of Advanced Cancer Environment) post-mortem study. We identify independent predictors of ctDNA release and analyse the tumour-volume detection limit. Through blinded profiling of postoperative plasma, we observe evidence of adjuvant chemotherapy resistance and identify patients who are very likely to experience recurrence of their lung cancer. Finally, we show that phylogenetic ctDNA profiling tracks the subclonal nature of lung cancer relapse and metastasis, providing a new approach for ctDNA-driven therapeutic studies.

SOX9 has distinct roles in the formation and progression of different non-small cell lung cancer histotypes.

The transcription factor SOX9 is a key regulator of multiple developmental processes and is frequently re-expressed in non-small cell lung cancer (NSCLC). Its precise role in the progression of NSCLC histotypes has, however, remained elusive. We show that SOX9 expression relates to poor overall survival and invasive histopathology in human non-mucinous adenocarcinoma and is absent in murine early minimally invasive and low in human in situ adenocarcinoma. Interestingly, despite wide SOX9 expression across advanced NSCLC histotypes, its genetic deletion in the murine KrasG12D ;Lkb1fl/fl model selectively disrupted only the growth of papillary NSCLC, without affecting the initiation of precursor lesions or growth of mucinous or squamous tissue. Spatial tissue phenotyping indicated a requirement of SOX9 expression for the progression of surfactant protein C-expressing progenitor cells, which gave rise to papillary tumours. Intriguingly, while SOX9 expression was dispensable for squamous tissue formation, its loss in fact led to enhanced squamous tumour metastasis, which was associated with altered collagen IV deposition in the basement membrane. Our work therefore demonstrates histopathology-selective roles for SOX9 in NSCLC progression, namely as a promoter for papillary adenocarcinoma progression, but an opposing metastasis-suppressing role in squamous histotype tissue. This attests to a pleiotropic SOX9 function, linked to the cell of origin and microenvironmental tissue contexts. © 2021 The Authors. The Journal of Pathology published by John Wiley & Sons, Ltd. on behalf of The Pathological Society of Great Britain and Ireland.

Inclusion of multiple high-risk histopathological criteria improves the prediction of adjuvant chemotherapy efficacy in lung adenocarcinoma.

AIMS: The decision to consider adjuvant chemotherapy (AC) for non-small cell lung cancer is currently governed by clinical stage. This study aims to assess other routinely collected pathological variables related to metastasis and survival for their ability to predict the efficacy of AC in lung adenocarcinoma. METHODS AND RESULTS: A retrospective single-centre series of 620 resected lung non-mucinous adenocarcinoma cases from 2005 to 2015 was used. Digital images of all slides were subjected to central review, and data on tumour histopathology, AC treatment and patient survival were compiled. A statistical case matching approach was used to counter selection bias. Several high-risk pathological criteria predict both pathological nodal involvement and early death: positive vascular invasion status (VI+) (HR = 2.10, P < 0.001), positive visceral pleural invasion status (VPI+) (HR = 2.16, P < 0.001), and solid/micropapillary-predominant WHO tumour type (SPA/MPPA) (HR = 3.29, P < 0.001). Crucially, these criteria also identify patient groups benefiting from AC (VI + HR = 0.69, P = 0.167, VPI + HR = 0.44, P = 0.004, SPA/MPPA HR = 0.36, P = 0.006). Cases showing VI+/VPI+/SPA/MPPA histology in the absence of AC stage criteria were common (170 of 620 total), and 8 had actually received AC. This group showed much better outcomes than equivalent untreated cases in matched analysis (3-year OS 100.0% versus 31.3%). Inclusion of patients with VI+/VPI+/SPA/MPPA histology would increase AC-eligible patients from 51.0% to 84.0% of non-mucinous tumours in our cohort. CONCLUSIONS: Our data provide preliminary evidence that the consideration of AC in patients with additional high-risk pathological indicators may significantly improve outcomes in operable lung adenocarcinoma, and that AC may be currently underused.

The path to a better biomarker: application of a risk management framework for the implementation of PD-L1 and TILs as immuno-oncology biomarkers in breast cancer clinical trials and daily practice.

Immune checkpoint inhibitor therapies targeting PD-1/PD-L1 are now the standard of care in oncology across several hematologic and solid tumor types, including triple negative breast cancer (TNBC). Patients with metastatic or locally advanced TNBC with PD-L1 expression on immune cells occupying ≥1% of tumor area demonstrated survival benefit with the addition of atezolizumab to nab-paclitaxel. However, concerns regarding variability between immunohistochemical PD-L1 assay performance and inter-reader reproducibility have been raised. High tumor-infiltrating lymphocytes (TILs) have also been associated with response to PD-1/PD-L1 inhibitors in patients with breast cancer (BC). TILs can be easily assessed on hematoxylin and eosin-stained slides and have shown reliable inter-reader reproducibility. As an established prognostic factor in early stage TNBC, TILs are soon anticipated to be reported in daily practice in many pathology laboratories worldwide. Because TILs and PD-L1 are parts of an immunological spectrum in BC, we propose the systematic implementation of combined PD-L1 and TIL analyses as a more comprehensive immuno-oncological biomarker for patient selection for PD-1/PD-L1 inhibition-based therapy in patients with BC. Although practical and regulatory considerations differ by jurisdiction, the pathology community has the responsibility to patients to implement assays that lead to optimal patient selection. We propose herewith a risk-management framework that may help mitigate the risks of suboptimal patient selection for immuno-therapeutic approaches in clinical trials and daily practice based on combined TILs/PD-L1 assessment in BC. © 2020 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.

p53 mutants cooperate with HIF-1 in transcriptional regulation of extracellular matrix components to promote tumor progression.

Mutations in the TP53 gene and microenvironmentally driven activation of hypoxia-inducible factor-1 (HIF-1) typically occur in later stages of tumorigenesis. An ongoing challenge is the identification of molecular determinants of advanced cancer pathogenesis to design alternative last-line therapeutic options. Here, we report that p53 mutants influence the tumor microenvironment by cooperating with HIF-1 to promote cancer progression. We demonstrate that in non-small cell lung cancer (NSCLC), p53 mutants exert a gain-of-function (GOF) effect on HIF-1, thus regulating a selective gene expression signature involved in protumorigenic functions. Hypoxia-mediated activation of HIF-1 leads to the formation of a p53 mutant/HIF-1 complex that physically binds the SWI/SNF chromatin remodeling complex, promoting expression of a selective subset of hypoxia-responsive genes. Depletion of p53 mutants impairs the HIF-mediated up-regulation of extracellular matrix (ECM) components, including type VIIa1 collagen and laminin-γ2, thus affecting tumorigenic potential of NSCLC cells in vitro and in mouse models in vivo. Analysis of surgically resected human NSCLC revealed that expression of this ECM gene signature was highly correlated with hypoxic tumors exclusively in patients carrying p53 mutations and was associated with poor prognosis. Our data reveal a GOF effect of p53 mutants in hypoxic tumors and suggest synergistic activities of p53 and HIF-1. These findings have important implications for cancer progression and might provide innovative last-line treatment options for advanced NSCLC.

Nanomolar Protein-Protein Interaction Monitoring with a Label-Free Protein-Probe Technique.

Protein-protein interactions (PPIs) are an essential part of correct cellular functionality, making them increasingly interesting drug targets. While Förster resonance energy transfer-based methods have traditionally been widely used for PPI studies, label-free techniques have recently drawn significant attention. These methods are ideal for studying PPIs, most importantly as there is no need for labeling of either interaction partner, reducing potential interferences and overall costs. Already, several different label-free methods are available, such as differential scanning calorimetry and surface plasmon resonance, but these biophysical methods suffer from low to medium throughput, which reduces suitability for high-throughput screening (HTS) of PPI inhibitors. Differential scanning fluorimetry, utilizing external fluorescent probes, is an HTS compatible technique, but high protein concentration is needed for experiments. To improve the current concepts, we have developed a method based on time-resolved luminescence, enabling PPI monitoring even at low nanomolar protein concentrations. This method, called the protein probe technique, is based on a peptide conjugated with Eu3+ chelate, and it has already been applied to monitor protein structural changes and small molecule interactions at elevated temperatures. Here, the applicability of the protein probe technique was demonstrated by monitoring single-protein pairing and multiprotein complexes at room and elevated temperatures. The concept functionality was proven by using both artificial and multiple natural protein pairs, such as KRAS and eIF4A together with their binding partners, and C-reactive protein in a complex with its antibody.

eIF4A alleviates the translational repression mediated by classical secondary structures more than by G-quadruplexes.

Increased activity of the mRNA helicase eIF4A drives cellular malignancy by reprogramming cellular translation, and eIF4A activity is the direct or indirect target of many emerging cancer therapeutics. The enriched presence of (GGC)4 motifs, which have the potential to fold into two-layered G-quadruplexes, within the 5'UTRs of eIF4A-dependent mRNAs suggests that eIF4A is required for the unwinding of these structures within these eIF4A-controlled mRNAs. However, the existence of folded G-quadruplexes within cells remains controversial, and G-quadruplex folding is in direct competition with classical Watson-Crick based secondary structures. Using a combination of reverse transcription stalling assays and 7-deazaguanine incorporation experiments we find that (GGC)4 motifs preferentially form classical secondary structures rather than G-quadruplexes in full-length mRNAs. Furthermore, using translation assays with the eIF4A inhibitor hippuristanol, both in vitro and in cells, we find that eIF4A activity alleviates translational repression of mRNAs with 5'UTR classical secondary structures significantly more than those with folded G-quadruplexes. This was particularly evident in experiments using a G-quadruplex stabilizing ligand, where shifting the structural equilibrium in favour of G-quadruplex formation diminishes eIF4A-dependency. This suggests that enrichment of (GGC)4 motifs in the 5'UTRs of eIF4A-dependent mRNAs is due to the formation of stable hairpin structures rather than G-quadruplexes.

In situ growth in early lung adenocarcinoma may represent precursor growth or invasive clone outgrowth-a clinically relevant distinction.

The switch from in situ to invasive tumor growth represents a crucial stage in the evolution of lung adenocarcinoma. However, the biological understanding of this shift is limited, and 'Noguchi Type C' tumors, being early lung adenocarcinomas with mixed in situ and invasive growth, represent those that are highly valuable in advancing our understanding of this process. All Noguchi Type C adenocarcinomas (n = 110) from the LATTICE-A cohort were reviewed and two patterns of in situ tumor growth were identified: those deemed likely to represent a true shift from precursor in situ to invasive disease ('Noguchi C1') and those in which the lepidic component appeared to represent outgrowth of the invasive tumor along existing airspaces ('Noguchi C2'). Overall Ki67 fraction was greater in C2 tumors and only C1 tumors showed significant increasing Ki67 from in situ to invasive disease. P53 positivity was acquired from in situ to invasive disease in C1 tumors but both components were positive in C2 tumors. Likewise, vimentin expression was increased from in situ to invasive tumor in C1 tumors only. Targeted next generation sequencing of 18 C1 tumors identified four mutations private to the invasive regions, including two in TP53, while 6 C2 tumors showed no private mutations. In the full LATTICe-A cohort, Ki67 fraction classified as either less than or greater than 10% within the in situ component of lung adenocarcinoma was identified as a strong predictor of patient outcome. This supports the proposition that tumors of all stages that have 'high grade' in situ components represent those with aggressive lepidic growth of the invasive clone. Overall these data support that the combined growth of Noguchi C tumors can represent two differing biological states and that 'Noguchi C1' tumors represent the genuine biological shift from in situ to invasive disease.

Challenges in molecular testing in non-small-cell lung cancer patients with advanced disease.

Lung cancer diagnostics have progressed greatly in the previous decade. Development of molecular testing to identify an increasing number of potentially clinically actionable genetic variants, using smaller samples obtained via minimally invasive techniques, is a huge challenge. Tumour heterogeneity and cancer evolution in response to therapy means that repeat biopsies or circulating biomarkers are likely to be increasingly useful to adapt treatment as resistance develops. We highlight some of the current challenges faced in clinical practice for molecular testing of EGFR, ALK, and new biomarkers such as PDL1. Implementation of next generation sequencing platforms for molecular diagnostics in non-small-cell lung cancer is increasingly common, allowing testing of multiple genetic variants from a single sample. The use of next generation sequencing to recruit for molecularly stratified clinical trials is discussed in the context of the UK Stratified Medicine Programme and The UK National Lung Matrix Trial.

Tracking genomic cancer evolution for precision medicine: the lung TRACERx study.

The importance of intratumour genetic and functional heterogeneity is increasingly recognised as a driver of cancer progression and survival outcome. Understanding how tumour clonal heterogeneity impacts upon therapeutic outcome, however, is still an area of unmet clinical and scientific need. TRACERx (TRAcking non-small cell lung Cancer Evolution through therapy [Rx]), a prospective study of patients with primary non-small cell lung cancer (NSCLC), aims to define the evolutionary trajectories of lung cancer in both space and time through multiregion and longitudinal tumour sampling and genetic analysis. By following cancers from diagnosis to relapse, tracking the evolutionary trajectories of tumours in relation to therapeutic interventions, and determining the impact of clonal heterogeneity on clinical outcomes, TRACERx may help to identify novel therapeutic targets for NSCLC and may also serve as a model applicable to other cancer types.

UTRly malignant: mRNA stability and the invasive phenotype in breast cancer.

Alterations in gene expression are central to the malignant phenotype. In this issue, Al-Ahmadi et al elegantly demonstrate that one key mechanism that determines invasiness in breast cancer is likely to be dysregulation of mRNA stability. This is achieved by altered expression of the proteins TTP and HuR, which bind 3' untranslated region (UTR) elements in cancer-related genes. The authors link this to the expression of a miRNA, miR-29a, and show that anti-miR-29a treatment can reverse the invasive phenotype in vitro. This further highlights the important roles of mRNA UTRs in malignant transformation, and the importance of investigating post-transcriptional disease mechanisms. Such mechanisms have the potential to provide novel therapeutic targets, as well as being vital to further our understanding of cancer biology.

Mapping the landscape of histomorphological cancer phenotypes using self-supervised learning on unannotated pathology slides.

Cancer diagnosis and management depend upon the extraction of complex information from microscopy images by pathologists, which requires time-consuming expert interpretation prone to human bias. Supervised deep learning approaches have proven powerful, but are inherently limited by the cost and quality of annotations used for training. Therefore, we present Histomorphological Phenotype Learning, a self-supervised methodology requiring no labels and operating via the automatic discovery of discriminatory features in image tiles. Tiles are grouped into morphologically similar clusters which constitute an atlas of histomorphological phenotypes (HP-Atlas), revealing trajectories from benign to malignant tissue via inflammatory and reactive phenotypes. These clusters have distinct features which can be identified using orthogonal methods, linking histologic, molecular and clinical phenotypes. Applied to lung cancer, we show that they align closely with patient survival, with histopathologically recognised tumor types and growth patterns, and with transcriptomic measures of immunophenotype. These properties are maintained in a multi-cancer study.

The artificial intelligence-based model ANORAK improves histopathological grading of lung adenocarcinoma.

The introduction of the International Association for the Study of Lung Cancer grading system has furthered interest in histopathological grading for risk stratification in lung adenocarcinoma. Complex morphology and high intratumoral heterogeneity present challenges to pathologists, prompting the development of artificial intelligence (AI) methods. Here we developed ANORAK (pyrAmid pooliNg crOss stReam Attention networK), encoding multiresolution inputs with an attention mechanism, to delineate growth patterns from hematoxylin and eosin-stained slides. In 1,372 lung adenocarcinomas across four independent cohorts, AI-based grading was prognostic of disease-free survival, and further assisted pathologists by consistently improving prognostication in stage I tumors. Tumors with discrepant patterns between AI and pathologists had notably higher intratumoral heterogeneity. Furthermore, ANORAK facilitates the morphological and spatial assessment of the acinar pattern, capturing acinus variations with pattern transition. Collectively, our AI method enabled the precision quantification and morphology investigation of growth patterns, reflecting intratumoral histological transitions in lung adenocarcinoma.

Spatial Positioning of Immune Hotspots Reflects the Interplay between B and T Cells in Lung Squamous Cell Carcinoma.

Beyond tertiary lymphoid structures, a significant number of immune-rich areas without germinal center-like structures are observed in non-small cell lung cancer. Here, we integrated transcriptomic data and digital pathology images to study the prognostic implications, spatial locations, and constitution of immune rich areas (immune hotspots) in a cohort of 935 patients with lung cancer from The Cancer Genome Atlas. A high intratumoral immune hotspot score, which measures the proportion of immune hotspots interfacing with tumor islands, was correlated with poor overall survival in lung squamous cell carcinoma but not in lung adenocarcinoma. Lung squamous cell carcinomas with high intratumoral immune hotspot scores were characterized by consistent upregulation of B-cell signatures. Spatial statistical analyses conducted on serial multiplex IHC slides further revealed that only 4.87% of peritumoral immune hotspots and 0.26% of intratumoral immune hotspots were tertiary lymphoid structures. Significantly lower densities of CD20+CXCR5+ and CD79b+ B cells and less diverse immune cell interactions were found in intratumoral immune hotspots compared with peritumoral immune hotspots. Furthermore, there was a negative correlation between the percentages of CD8+ T cells and T regulatory cells in intratumoral but not in peritumoral immune hotspots, with tertiary lymphoid structures excluded. These findings suggest that the intratumoral immune hotspots reflect an immunosuppressive niche compared with peritumoral immune hotspots, independent of the distribution of tertiary lymphoid structures. A balance toward increased intratumoral immune hotspots is indicative of a compromised antitumor immune response and poor outcome in lung squamous cell carcinoma. SIGNIFICANCE: Intratumoral immune hotspots beyond tertiary lymphoid structures reflect an immunosuppressive microenvironment, different from peritumoral immune hotspots, warranting further study in the context of immunotherapies.

Protein expression of S100A2 reveals it association with patient prognosis and immune infiltration profile in colorectal cancer.

PURPOSE: Colorectal cancer (CRC) is the third most diagnosed cancer worldwide. Despite a well-established knowledge of tumour development, biomarkers to predict patient outcomes are still required. S100 calcium-binding protein A2 (S100A2) has been purposed as a potential marker in many types of cancer, however, the prognostic value of S100A2 in CRC is rarely reported. MATERIAL AND METHODS: In this study, immunohistochemistry (IHC) was performed to identify the prognostic role of S100A2 protein expression in the tumour core of the tissue microarrays (TMAs) in colorectal cancer patients (n=787). Bulk RNA transcriptomic data was used to identify significant genes compared between low and high cytoplasmic S100A2 groups. Multiplex immunofluorescence (mIF) was performed to further study and confirm the immune infiltration in tumours with low and high cytoplasmic S100A2. RESULTS: Low cytoplasmic protein expression of S100A2 in the tumour core was associated with poor survival (HR 0.539, 95%CI 0.394-0.737, P<0.001) and other adverse tumour phenotypes. RNA transcriptomic analysis showed a gene significantly associated with the low cytoplasmic S100A2 group (AKT3, TAGLN, MYLK, FGD6 and ETFDH), which correlated with tumour development and progression. GSEA analysis identifies the enriched anti-tumour and immune activity group of genes in high cytoplasmic S100A2. Additionally, mIF staining showed that high CD3+FOXP3+ and CD163+ inversely associated with low cytoplasmic S100A2 (P<0.001, P=0.009 respectively). CONCLUSION: Our finding demonstrates a prognostic value of S100A2 together with the correlation with immune infiltration in CRC.

Self-supervised deep learning for highly efficient spatial immunophenotyping.

BACKGROUND: Efficient biomarker discovery and clinical translation depend on the fast and accurate analytical output from crucial technologies such as multiplex imaging. However, reliable cell classification often requires extensive annotations. Label-efficient strategies are urgently needed to reveal diverse cell distribution and spatial interactions in large-scale multiplex datasets. METHODS: This study proposed Self-supervised Learning for Antigen Detection (SANDI) for accurate cell phenotyping while mitigating the annotation burden. The model first learns intrinsic pairwise similarities in unlabelled cell images, followed by a classification step to map learnt features to cell labels using a small set of annotated references. We acquired four multiplex immunohistochemistry datasets and one imaging mass cytometry dataset, comprising 2825 to 15,258 single-cell images to train and test the model. FINDINGS: With 1% annotations (18-114 cells), SANDI achieved weighted F1-scores ranging from 0.82 to 0.98 across the five datasets, which was comparable to the fully supervised classifier trained on 1828-11,459 annotated cells (-0.002 to -0.053 of averaged weighted F1-score, Wilcoxon rank-sum test, P = 0.31). Leveraging the immune checkpoint markers stained in ovarian cancer slides, SANDI-based cell identification reveals spatial expulsion between PD1-expressing T helper cells and T regulatory cells, suggesting an interplay between PD1 expression and T regulatory cell-mediated immunosuppression. INTERPRETATION: By striking a fine balance between minimal expert guidance and the power of deep learning to learn similarity within abundant data, SANDI presents new opportunities for efficient, large-scale learning for histology multiplex imaging data. FUNDING: This study was funded by the Royal Marsden/ICR National Institute of Health Research Biomedical Research Centre.