Artificial intelligence.

Experts discuss the challenges and opportunities of using artificial intelligence (AI) to study the evolution of cancer cells and their microenvironment, improve diagnosis, predict treatment response, and ensure responsible implementation in the clinic.

Multi-cancer risk stratification based on national health data: a retrospective modelling and validation study.

BACKGROUND: Health care is experiencing a drive towards digitisation, and many countries are implementing national health data resources. Although a range of cancer risk models exists, the utility on a population level for risk stratification across cancer types has not been fully explored. We aimed to close this gap by evaluating pan-cancer risk models built on electronic health records across the Danish population with validation in the UK Biobank. METHODS: In this retrospective modelling and validation study, data for model development and internal validation were derived from the following Danish health registries: the Central Person Registry, the Danish National Patient Registry, the death registry, the cancer registry, and full-text medical records from secondary care records in the capital region. The development data included adults aged 16-86 years without previous malignant cancers in the time period from Jan 1, 1995, to Dec 31, 2014. The internal validation period was from Jan 1, 2015, to April 10, 2018, and the data included all adults without a previous indication of cancer aged 16-75 years on Dec 31, 2014. The external validation cohort from the UK Biobank included all adults without a previous indication of cancer aged 50-75 years. We used time-dependent Bayesian Cox hazard models built on the combined medical history of Danish individuals. A set of 1392 covariates from available clinical disease trajectories, text-mined basic health factors, and family histories were used to train predictive models of 20 major cancer types. The models were validated on cancer incidence between 2015 and 2018 across Denmark and on individuals in the UK Biobank. The primary outcomes were discrimination and calibration performance. FINDINGS: From the Danish registries, we included 6 732 553 individuals covering 60 million hospital visits, 90 million diagnoses, and a total of 193 million life-years between Jan 1, 1978, and April 10, 2018. Danish registry data covering the period from Jan 1, 2015, to April 10, 2018, were used to internally validate risk models, containing a total of 4 248 491 individuals who remained at risk of a primary malignant cancer diagnosis and 67 401 cancer cases recorded. For the external validation, we evaluated the same time period in the UK Biobank covering 377 004 individuals with 11 486 cancer cases. The predictive performance of the models on Danish data showed good discrimination (concordance index 0·81 [SD 0·08], ranging from 0·66 [95% CI 0·65-0·67] for cervix uteri cancer to 0·91 [0·90-0·92] for liver cancer). Performance was similar on the UK Biobank in a direct transfer when controlling for shifts in the age distribution (concordance index 0·66 [SD 0·08], ranging from 0·55 [95% CI 0·44-0·66] for cervix uteri cancer to 0·78 [0·77-0·79] for lung cancer). Cancer risks were associated, in addition to heritable components, with a broad range of preceding diagnoses and health factors. The best overall performance was seen for cancers of the digestive system (oesophageal, stomach, colorectal, liver, and pancreatic) but also thyroid, kidney, and uterine cancers. INTERPRETATION: Data available in national electronic health databases can be used to approximate cancer risk factors and enable risk predictions in most cancer types. Model predictions generalise between the Danish and UK health-care systems. With the emergence of multi-cancer early detection tests, electronic health record-based risk models could supplement screening efforts. FUNDING: Novo Nordisk Foundation and the Danish Innovation Foundation.

Single-cell and spatial transcriptomics analysis of non-small cell lung cancer.

Lung cancer is the second most frequently diagnosed cancer and the leading cause of cancer-related mortality worldwide. Tumour ecosystems feature diverse immune cell types. Myeloid cells, in particular, are prevalent and have a well-established role in promoting the disease. In our study, we profile approximately 900,000 cells from 25 treatment-naive patients with adenocarcinoma and squamous-cell carcinoma by single-cell and spatial transcriptomics. We note an inverse relationship between anti-inflammatory macrophages and NK cells/T cells, and with reduced NK cell cytotoxicity within the tumour. While we observe a similar cell type composition in both adenocarcinoma and squamous-cell carcinoma, we detect significant differences in the co-expression of various immune checkpoint inhibitors. Moreover, we reveal evidence of a transcriptional "reprogramming" of macrophages in tumours, shifting them towards cholesterol export and adopting a foetal-like transcriptional signature which promotes iron efflux. Our multi-omic resource offers a high-resolution molecular map of tumour-associated macrophages, enhancing our understanding of their role within the tumour microenvironment.

Teacher-student collaborated multiple instance learning for pan-cancer PDL1 expression prediction from histopathology slides.

Programmed cell death ligand 1 (PDL1), as an important biomarker, is quantified by immunohistochemistry (IHC) with few established histopathological patterns. Deep learning aids in histopathological assessment, yet heterogeneity and lacking spatially resolved annotations challenge precise analysis. Here, we present a weakly supervised learning approach using bulk RNA sequencing for PDL1 expression prediction from hematoxylin and eosin (H&E) slides. Our method extends the multiple instance learning paradigm with the teacher-student framework, which assigns dynamic pseudo-labels for intra-slide heterogeneity and retrieves unlabeled instances using temporal ensemble model distillation. The approach, evaluated on 12,299 slides across 20 solid tumor types, achieves a weighted average area under the curve of 0.83 on fresh-frozen and 0.74 on formalin-fixed specimens for 9 tumors with PDL1 as an established biomarker. Our method predicts PDL1 expression patterns, validated by IHC on 20 slides, offering insights into histologies relevant to PDL1. This demonstrates the potential of deep learning in identifying diverse histological patterns for molecular changes from H&E images.

SpatialData: an open and universal data framework for spatial omics.

Spatially resolved omics technologies are transforming our understanding of biological tissues. However, the handling of uni- and multimodal spatial omics datasets remains a challenge owing to large data volumes, heterogeneity of data types and the lack of flexible, spatially aware data structures. Here we introduce SpatialData, a framework that establishes a unified and extensible multiplatform file-format, lazy representation of larger-than-memory data, transformations and alignment to common coordinate systems. SpatialData facilitates spatial annotations and cross-modal aggregation and analysis, the utility of which is illustrated in the context of multiple vignettes, including integrative analysis on a multimodal Xenium and Visium breast cancer study.

Multiparameter prediction of myeloid neoplasia risk.

The myeloid neoplasms encompass acute myeloid leukemia, myelodysplastic syndromes and myeloproliferative neoplasms. Most cases arise from the shared ancestor of clonal hematopoiesis (CH). Here we analyze data from 454,340 UK Biobank participants, of whom 1,808 developed a myeloid neoplasm 0-15 years after recruitment. We describe the differences in CH mutational landscapes and hematology/biochemistry test parameters among individuals that later develop myeloid neoplasms (pre-MN) versus controls, finding that disease-specific changes are detectable years before diagnosis. By analyzing differences between 'pre-MN' and controls, we develop and validate Cox regression models quantifying the risk of progression to each myeloid neoplasm subtype. We construct 'MN-predict', a web application that generates time-dependent predictions with the input of basic blood tests and genetic data. Our study demonstrates that many individuals that develop myeloid neoplasms can be identified years in advance and provides a framework for disease-specific prognostication that will be of substantial use to researchers and physicians.

Computational analysis of peripheral blood smears detects disease-associated cytomorphologies.

Many hematological diseases are characterized by altered abundance and morphology of blood cells and their progenitors. Myelodysplastic syndromes (MDS), for example, are a group of blood cancers characterised by cytopenias, dysplasia of hematopoietic cells and blast expansion. Examination of peripheral blood slides (PBS) in MDS often reveals changes such as abnormal granulocyte lobulation or granularity and altered red blood cell (RBC) morphology; however, some of these features are shared with conditions such as haematinic deficiency anemias. Definitive diagnosis of MDS requires expert cytomorphology analysis of bone marrow smears and complementary information such as blood counts, karyotype and molecular genetics testing. Here, we present Haemorasis, a computational method that detects and characterizes white blood cells (WBC) and RBC in PBS. Applied to over 300 individuals with different conditions (SF3B1-mutant and SF3B1-wildtype MDS, megaloblastic anemia, and iron deficiency anemia), Haemorasis detected over half a million WBC and millions of RBC and characterized their morphology. These large sets of cell morphologies can be used in diagnosis and disease subtyping, while identifying novel associations between computational morphotypes and disease. We find that hypolobulated neutrophils and large RBC are characteristic of SF3B1-mutant MDS. Additionally, while prevalent in both iron deficiency and megaloblastic anemia, hyperlobulated neutrophils are larger in the latter. By integrating cytomorphological features using machine learning, Haemorasis was able to distinguish SF3B1-mutant MDS from other MDS using cytomorphology and blood counts alone, with high predictive performance. We validate our findings externally, showing that they generalize to other centers and scanners. Collectively, our work reveals the potential for the large-scale incorporation of automated cytomorphology into routine diagnostic workflows.

Notch1 mutations drive clonal expansion in normal esophageal epithelium but impair tumor growth.

NOTCH1 mutant clones occupy the majority of normal human esophagus by middle age but are comparatively rare in esophageal cancers, suggesting NOTCH1 mutations drive clonal expansion but impede carcinogenesis. Here we test this hypothesis. Sequencing NOTCH1 mutant clones in aging human esophagus reveals frequent biallelic mutations that block NOTCH1 signaling. In mouse esophagus, heterozygous Notch1 mutation confers a competitive advantage over wild-type cells, an effect enhanced by loss of the second allele. Widespread Notch1 loss alters transcription but has minimal effects on the epithelial structure and cell dynamics. In a carcinogenesis model, Notch1 mutations were less prevalent in tumors than normal epithelium. Deletion of Notch1 reduced tumor growth, an effect recapitulated by anti-NOTCH1 antibody treatment. Notch1 null tumors showed reduced proliferation. We conclude that Notch1 mutations in normal epithelium are beneficial as wild-type Notch1 favors tumor expansion. NOTCH1 blockade may have therapeutic potential in preventing esophageal squamous cancer.

Germline TP53 mutations undergo copy number gain years prior to tumor diagnosis.

Li-Fraumeni syndrome (LFS) is a hereditary cancer predisposition syndrome associated with germline TP53 pathogenic variants. Here, we perform whole-genome sequence (WGS) analysis of tumors from 22 patients with TP53 germline pathogenic variants. We observe somatic mutations affecting Wnt, PI3K/AKT signaling, epigenetic modifiers and homologous recombination genes as well as mutational signatures associated with prior chemotherapy. We identify near-ubiquitous early loss of heterozygosity of TP53, with gain of the mutant allele. This occurs earlier in these tumors compared to tumors with somatic TP53 mutations, suggesting the timing of this mark may distinguish germline from somatic TP53 mutations. Phylogenetic trees of tumor evolution, reconstructed from bulk and multi-region WGS, reveal that LFS tumors exhibit comparatively limited heterogeneity. Overall, our study delineates early copy number gains of mutant TP53 as a characteristic mutational process in LFS tumorigenesis, likely arising years prior to tumor diagnosis.

Spatial biology of cancer evolution.

The natural history of cancers can be understood through the lens of evolution given that the driving forces of cancer development are mutation and selection of fitter clones. Cancer growth and progression are spatial processes that involve the breakdown of normal tissue organization, invasion and metastasis. For these reasons, spatial patterns are an integral part of histological tumour grading and staging as they measure the progression from normal to malignant disease. Furthermore, tumour cells are part of an ecosystem of tumour cells and their surrounding tumour microenvironment. A range of new spatial genomic, transcriptomic and proteomic technologies offers new avenues for the study of cancer evolution with great molecular and spatial detail. These methods enable precise characterizations of the tumour microenvironment, cellular interactions therein and micro-anatomical structures. In conjunction with spatial genomics, it emerges that tumours and microenvironments co-evolve, which helps explain observable patterns of heterogeneity and offers new routes for therapeutic interventions.

Spatial genomics maps the structure, nature and evolution of cancer clones.

Genome sequencing of cancers often reveals mosaics of different subclones present in the same tumour1-3. Although these are believed to arise according to the principles of somatic evolution, the exact spatial growth patterns and underlying mechanisms remain elusive4,5. Here, to address this need, we developed a workflow that generates detailed quantitative maps of genetic subclone composition across whole-tumour sections. These provide the basis for studying clonal growth patterns, and the histological characteristics, microanatomy and microenvironmental composition of each clone. The approach rests on whole-genome sequencing, followed by highly multiplexed base-specific in situ sequencing, single-cell resolved transcriptomics and dedicated algorithms to link these layers. Applying the base-specific in situ sequencing workflow to eight tissue sections from two multifocal primary breast cancers revealed intricate subclonal growth patterns that were validated by microdissection. In a case of ductal carcinoma in situ, polyclonal neoplastic expansions occurred at the macroscopic scale but segregated within microanatomical structures. Across the stages of ductal carcinoma in situ, invasive cancer and lymph node metastasis, subclone territories are shown to exhibit distinct transcriptional and histological features and cellular microenvironments. These results provide examples of the benefits afforded by spatial genomics for deciphering the mechanisms underlying cancer evolution and microenvironmental ecology.

p53 mutation in normal esophagus promotes multiple stages of carcinogenesis but is constrained by clonal competition.

Aging normal human oesophagus accumulates TP53 mutant clones. These are the origin of most oesophageal squamous carcinomas, in which biallelic TP53 disruption is almost universal. However, how p53 mutant clones expand and contribute to cancer development is unclear. Here we show that inducing the p53R245W mutant in single oesophageal progenitor cells in transgenic mice confers a proliferative advantage and clonal expansion but does not disrupt normal epithelial structure. Loss of the remaining p53 allele in mutant cells results in genomically unstable p53R245W/null epithelium with giant polyaneuploid cells and copy number altered clones. In carcinogenesis, p53 mutation does not initiate tumour formation, but tumours developing from areas with p53 mutation and LOH are larger and show extensive chromosomal instability compared to lesions arising in wild type epithelium. We conclude that p53 has distinct functions at different stages of carcinogenesis and that LOH within p53 mutant clones in normal epithelium is a critical step in malignant transformation.

Artificial intelligence in histopathology: enhancing cancer research and clinical oncology.

Artificial intelligence (AI) methods have multiplied our capabilities to extract quantitative information from digital histopathology images. AI is expected to reduce workload for human experts, improve the objectivity and consistency of pathology reports, and have a clinical impact by extracting hidden information from routinely available data. Here, we describe how AI can be used to predict cancer outcome, treatment response, genetic alterations and gene expression from digitized histopathology slides. We summarize the underlying technologies and emerging approaches, noting limitations, including the need for data sharing and standards. Finally, we discuss the broader implications of AI in cancer research and oncology.

The longitudinal dynamics and natural history of clonal haematopoiesis.

Clonal expansions driven by somatic mutations become pervasive across human tissues with age, including in the haematopoietic system, where the phenomenon is termed clonal haematopoiesis1-4. The understanding of how and when clonal haematopoiesis develops, the factors that govern its behaviour, how it interacts with ageing and how these variables relate to malignant progression remains limited5,6. Here we track 697 clonal haematopoiesis clones from 385 individuals 55 years of age or older over a median of 13 years. We find that 92.4% of clones expanded at a stable exponential rate over the study period, with different mutations driving substantially different growth rates, ranging from 5% (DNMT3A and TP53) to more than 50% per year (SRSF2P95H). Growth rates of clones with the same mutation differed by approximately ±5% per year, proportionately affecting slow drivers more substantially. By combining our time-series data with phylogenetic analysis of 1,731 whole-genome sequences of haematopoietic colonies from 7 individuals from an older age group, we reveal distinct patterns of lifelong clonal behaviour. DNMT3A-mutant clones preferentially expanded early in life and displayed slower growth in old age, in the context of an increasingly competitive oligoclonal landscape. By contrast, splicing gene mutations drove expansion only later in life, whereas TET2-mutant clones emerged across all ages. Finally, we show that mutations driving faster clonal growth carry a higher risk of malignant progression. Our findings characterize the lifelong natural history of clonal haematopoiesis and give fundamental insights into the interactions between somatic mutation, ageing and clonal selection.

Somatic mutation rates scale with lifespan across mammals.

The rates and patterns of somatic mutation in normal tissues are largely unknown outside of humans1-7. Comparative analyses can shed light on the diversity of mutagenesis across species, and on long-standing hypotheses about the evolution of somatic mutation rates and their role in cancer and ageing. Here we performed whole-genome sequencing of 208 intestinal crypts from 56 individuals to study the landscape of somatic mutation across 16 mammalian species. We found that somatic mutagenesis was dominated by seemingly endogenous mutational processes in all species, including 5-methylcytosine deamination and oxidative damage. With some differences, mutational signatures in other species resembled those described in humans8, although the relative contribution of each signature varied across species. Notably, the somatic mutation rate per year varied greatly across species and exhibited a strong inverse relationship with species lifespan, with no other life-history trait studied showing a comparable association. Despite widely different life histories among the species we examined-including variation of around 30-fold in lifespan and around 40,000-fold in body mass-the somatic mutation burden at the end of lifespan varied only by a factor of around 3. These data unveil common mutational processes across mammals, and suggest that somatic mutation rates are evolutionarily constrained and may be a contributing factor in ageing.

Biallelic mutations in cancer genomes reveal local mutational determinants.

The infinite sites model of molecular evolution posits that every position in the genome is mutated at most once1. By restricting the number of possible mutation histories, haplotypes and alleles, it forms a cornerstone of tumor phylogenetic analysis2 and is often implied when calling, phasing and interpreting variants3,4 or studying the mutational landscape as a whole5. Here we identify 18,295 biallelic mutations, where the same base is mutated independently on both parental copies, in 559 (21%) bulk sequencing samples from the Pan-Cancer Analysis of Whole Genomes study. Biallelic mutations reveal ultraviolet light damage hotspots at E26 transformation-specific (ETS) and nuclear factor of activated T cells (NFAT) binding sites, and hypermutable motifs in POLE-mutant and other cancers. We formulate recommendations for variant calling and provide frameworks to model and detect biallelic mutations. These results highlight the need for accurate models of mutation rates and tumor evolution, as well as their inference from sequencing data.

Cell2location maps fine-grained cell types in spatial transcriptomics.

Spatial transcriptomic technologies promise to resolve cellular wiring diagrams of tissues in health and disease, but comprehensive mapping of cell types in situ remains a challenge. Here we present сell2location, a Bayesian model that can resolve fine-grained cell types in spatial transcriptomic data and create comprehensive cellular maps of diverse tissues. Cell2location accounts for technical sources of variation and borrows statistical strength across locations, thereby enabling the integration of single-cell and spatial transcriptomics with higher sensitivity and resolution than existing tools. We assessed cell2location in three different tissues and show improved mapping of fine-grained cell types. In the mouse brain, we discovered fine regional astrocyte subtypes across the thalamus and hypothalamus. In the human lymph node, we spatially mapped a rare pre-germinal center B cell population. In the human gut, we resolved fine immune cell populations in lymphoid follicles. Collectively, our results present сell2location as a versatile analysis tool for mapping tissue architectures in a comprehensive manner.