Emerging data-driven approaches to cancer

Current knowledge of cancer genomics remains biased against non-coding mutations. To systematically search for regulatory non-coding mutations, we assessed mutations in conserved positions in the genome under the assumption that these are more likely to be functional than mutations in positions with low conservation. To this end, we use whole-genome sequencing data from the International Cancer Genome Consortium (ICGC) and combined it with evolutionary constraint inferred from 240 mammals, to identify genes enriched in non-coding constraint mutations (NCCMs), mutations likely to be regulatory in nature. We compare medulloblastoma (MB), which is malignant, to pilocytic astrocytoma (PA), a primarily benign tumor, and find highly different NCCM frequencies between the two, in agreement with the fact that malignant cancers tend to have more mutations. In PA, a high NCCM frequency only affects the BRAF locus, which is the most commonly mutated gene in PA. In contrast, in MB, >500 genes have high levels of NCCMs. Intriguingly, several loci with NCCMs in MB are associated with different age of onset, such as the HOXB cluster in young MB patients. NCCMs in this locus were found to alter expression of HOXB2, HOXB5 and HOXB9 in MB cells. In adult patients, NCCMs occurred in e.g. the WASF-2/AHDC1/FGR locus. One of these NCCMs led to increased expression of the SRC kinase FGR, and augmented responsiveness of MB cells to dasatinib, a SRC kinase inhibitor. Our analysis thus points to different molecular pathways in different patient groups. These newly identified putative candidate driver genes may aid in patient stratification in MB, and could be valuable for future selection of personalized treatment options.

Karin Forsberg Nilsson is Professor of Stem Cell Research at the Department of Immunology, Genetics and Pathology, Uppsala, and Dean of the Faculty of Medicine, University. She is SciLifeLab Faculty, and has a position as Guest Professor at the University of Nottingham, UK. Karin Forsberg Nilsson obtained her PhD in 1992 from Uppsala University and did a postdoc 1994-1996 at the National Institutes of Health, MD, USA. She has held leadership positions in both pharmaceutical industry and academia. The focus of her lab is brain tumor biology and genetics.

Translating complex multi-level molecular profiling data into actionable insights for cancer diagnosis and therapy poses a formidable challenge. We propose here a paradigm for analyzing and interpreting in-depth multi-omics and functional drug response data derived from acute myeloid leukemia (AML), aiming at practical applications in precision clinical oncology. Our approach is based on Data-Driven Hallmarks (DDHMs), drawing inspiration from the Weinberg-Hanahan cancer hallmark concept (Hanahan, 2022). While the original cancer hallmark concept provides a useful theoretical framework for understanding cancer, it is not applicable for processing, interpreting, and translating molecular profiling data for precision diagnostics and therapy in individual patients.

We first acquired and integrated genomics, transcriptomics, epigenetics, and proteomics data alongside ex-vivo drug response data for over 500 drugs across 150 AML samples, resulting in nearly 100 million data points (Erkers et al., 2023). Subsequently, we identified 11 dimensions of variability across the entire dataset, referred here as DDHMs of AML. DDHMs intertwine specific multi-omics molecular features (potential diagnostic biomarkers) with distinct vulnerabilities to individual drugs (potential cancer treatments). DDHMs show proficiency in predicting high-risk AML and determining the effective drugs for each AML sample. The strategy of assembling drugs targeting active hallmarks in each patient offers a promising avenue for tailoring effective drug combinations.

In summary, we showcase the conversion of millions of complex multi-omics research data into a manageable set of DDHMs, which are based on specific biomarkers and linked drug vulnerabilities and provide an opportunity for tailoring drug treatments and drug combinations for individual patients. This approach aligns well with current practices and guidelines in translational, clinical, regulatory, and industrial setting and could expedite bringing the benefits of data-driven precision medicine research to cancer patients.

References:

1. Hanahan, D. (2022). Hallmarks of Cancer: New Dimensions. Cancer Discov 12, 31-46.
2. Erkers T, Struyf N, James T…. Orre L, Jafari R, Pawitan Y, Seashore-Ludlow B, Lehtiö J, Lehmann S, Östling P, Kallioniemi O. Data-driven hallmarks of acute myeloid leukemia, submitted, 2023.

Olli Kallioniemi, M.D., Ph.D. is director of the Science for Life Laboratory (www.SciLifeLab.se), a national infrastructure for life sciences in Sweden and also a professor in Molecular Precision Medicine at the Karolinska Institutet (2015-present). He also directs the national SciLifeLab program on Data-Driven Life Science (DDLS). Olli Kallioniemi was previously the founding director of FIMM – the Institute for Molecular Medicine Finland at the University of Helsinki, as part of the Nordic EMBL partnership in Molecular Medicine (2007-2015) Olli Kallioniemi is a member of European Molecular Biology Organization (EMBO), European Academy of Cancer Sciences, the Nobel Assembly at the Karolinska Institutet and the Royal Swedish Academy of Sciences.

Nervous system cancers contain a large spectrum of transcriptional cell states, reflecting processes active during normal development, injury response and growth. However, we lack a good understanding of these states’ regulation and pharmacological importance. Here, we describe the integrated reconstruction of such cellular regulatory programs and their therapeutic targets from extensive collections of single-cell RNA sequencing data (scRNA-seq) from both tumors and developing tissues. Our method, termed single-cell Regulatory-driven Clustering (scRegClust), predicts essential kinases and transcription factors in little computational time thanks to a new efficient optimization strategy. Using this method, we analyze scRNA-seq data from both adult and childhood brain cancers to identify transcription factors and kinases that regulate distinct tumor cell states. In adult glioblastoma, our model predicts that blocking the activity of PDGFRA, DDR1, ERBB3 or SOX6, or increasing YBX1 -activity, would potentiate temozolomide treatment. We further perform an integrative study of scRNA-seq data from both cancer and the developing brain to uncover the regulation of emerging meta-modules. We find a meta-module regulated by the transcription factors SPI1 and IRF8 and link it to an immune-mediated mesenchymal-like state. Our algorithm is available as an easy-to-use R package and companion visualization tool that help uncover the regulatory programs underlying cell plasticity in cancer and other diseases.

Ida Larsson studied medical biotechnology at the Royal Institute of Technology in Stockholm and received her MScEng in 2018. She then continued as a PhD student in computational systems biology at Uppsala University, where her research has focused on the brain tumor glioblastoma and developing methods for analyzing single-cell RNA sequencing data. She defended her PhD in 2023 and is now starting as a postdoctoral fellow at Dana-Farber Cancer Institute in Boston. Her research interests revolve around intratumoral heterogeneity and plasticity in neural cancers.

Our research group has developed and conducted proof-of-concept studies of a novel method that combines artificial intelligence (AI) and mobile digital microscopy for cell-based cervical cancer screening in resource-limited settings. The mobile microscopes are wirelessly connected via mobile networks for AI-based analysis and provide access to diagnostics where there is a lack of medical experts. We are now assessing the use of the new diagnostic method in the form of a validation studies in Kenya and Tanzania with the aim of detecting premalignant changes for the purpose of cervical cancer prevention. Cervical smears are collected at the point-of-care and digitized with mobile microscope scanners and premalignant cells detected with an AI-algorithm. Suspected abnormal cells are verified by a pathologist and treated. The method’s diagnostic accuracy, technical feasibility, cost and time per test, and acceptance of the AI method is evaluated and compared to conventional diagnostics. Throughout the project, opportunities for larger scale implementation of the diagnostic platform in East Africa are evaluated, with the ultimate goal of achieving sustainable solutions for low-resource settings.The methods have great potential to improve equitable and sustainable access to high-quality diagnostics for cervical cancer screening among women residing in low- and middle income countries, carrying the highest cervical cancer burden globally.

Nina Linder is a physician by training and received her MD and PhD from the University of Helsinki, Finland and is an Associate Professor at Uppsala University, Sweden as well the Institute for Molecular Medicine, University of Helsinki, Finland. Nina Linder’s current research involves the development of novel artificial intelligence-based solutions for cancer and infectious disease diagnostics. Linder is co-heading projects developing artificial intelligence-based tools for point-of-care diagnostics in a global setting. The overall goal of Linder’s research is to promote the implementation of innovative decision-support solutions for precision medicine to improve the translation from basic medical research to the doctor and patient at the clinic.

The causes of glioma, the most malignant brain tumor, is in most cases unknown. Common environmental factors such as alcohol and smoking has not been linked to brain tumors. High doses of ionizing radiation are associated with increased risk, but it explains very few cases. Therefore, one assumption could be, that glioma is an endogenous disease that is caused by a stochastic effect initiated by a complex inheritance and subsequent biological cascades leading to tumor development. To discover which biological parameters that are associated with glioma risk, several factors need to be taken in consideration. Taking glioma as an example, important factors when collecting data and samples and biostatistical considerations will be presented, giving corner stones for how we have been able to harvest and find true association through data driven analyses.

Beatrice Melin is MD, PHD and Professor of Oncology at Umeå University and Director of Research Development and Innovation at Region Västerbotten. Professor Melin has worked in the field of brain tumor research for 25 years and published on both cohort, blood tumor and registry studies.

Head and neck cancer (HNC) can be divided into two biologically distinct entities based on human papillomavirus (HPV) status. In HPV-negative HNC, the main challenge remains finding treatments to improve survival rates and decrease recurrence, while HPV-positive patients need better stratification methods to avoid morbidity from overtreatment. Intra-tumour heterogeneity (ITH) is a major feature in both types of HNC and a barrier to successful patient stratification and treatment. Despite its significance, ITH remains poorly understood.

We posit that HNC tumours consist of a vast, diverse ecosystem of cell types with different roles. In order to improve patient-tailored treatment, our goal is to, through single-cell RNA sequencing, characterize these populations, their biological and clinical significance and their roles in responses to various treatments.

We report firstly, the identification of a novel population of cancer cells with undetectable HPV expression in HPV+ tumours. These cells are less proliferative, more invasive and respond poorly to treatment. Validating these findings through TCGA, we found that decreased HPV expression levels are linked to poor prognosis in HPV-positive oropharyngeal cancer.

Secondly, we show that malignant cells in HPV- oral cavity cancer express antigen-presentation genes, and that expression of these genes together with an interferon signal across multiple cell types strongly predicts response to neoadjuvant PD-1 inhibition.

Finally, through collecting a dataset comprising > 120 HNC patients and one million cells, we were able to find diverse recurrent cancer cell states and microenvironmental co-expression patterns. Notably, we found a rare subset of cancer cells that undergo a full, rather than partial epithelial-mesenchymal transition (EMT). These cells are linked to depletion of CD8+ T-cells and increased numbers of fibroblasts and macrophages in the microenvironment, as well as poor outcomes.

In summary, we have created a comprehensive atlas of the HNC ecosystem at previously unseen scale and resolution. We identify rare cell populations responsible for cancer hallmarks such as senescence, proliferation and metastasis and show how subpopulations change in response to HPV infection, radiotherapy and immunotherapy. This knowledge will greatly advance personalised treatment of head and neck cancer through guiding patient stratification, drug development and treatment selection.

Michael Mints studied medicine at Karolinska Institutet through the clinician-scientist training programme. He received his MD in 2013 and his PhD in 2015 at the Department of Oncology-Pathology. After clinical work in Umeå and a physician-researcher internship in Östersund he started a postdoc in Itay Tirosh’s lab at the Weizmann Institute in 2019. There, his work focuses on leveraging single-cell RNA sequencing and computational methods in order to understand head and neck cancer heterogeneity with the goal of personalizing head and neck cancer treatment.

Artificial intelligence (AI) can infer information from data in a way that exceeds human capacity to do so. Especially in clinical oncology, scientists use AI to generate biomarkers, find correlations or extract prognostic or predictive information. This talk will give you an idea of how it started, how it’s going and what the future might hold for cancer researchers in an era of paradigm-shifting technological advances.

Hannah Muti is a Clinician/Scientist with interests in precision medicine in gastrointestinal oncology and visceral surgery. Her research covers the use of artificial intelligence to investigate gastrointestinal cancers in the context of precision oncology. She simultaneously works in the Department for Visceral, Thoracic and Vascular Surgery at the University Hospital Dresden to obtain her specialization in visceral surgery. 

The activity of human cells depends on interactions between molecules in molecular networks. Disruptions to these networks are common in disease, e.g. it can drive unrestricted growth in cancer cells. Simulations of these processes could predict disease mechanisms and identify suitable drug targets. However, these networks consist of thousands of different molecules with tens of thousands of interactions, and it has been challenging to parametrize systems-wide models using traditional approaches. We have developed recurrent neural network models of the networks that use molecules as hidden nodes with connections constrained to known molecular interactions. These models predict unseen test-data from living cells, e.g. we predict gene expression in macrophages in response to different ligands and we use the models to infer causative signaling cascades. Currently, we are expanding the framework to integrate signaling, metabolism, and gene regulation for a more complete mechanistic description of cellular activities.

Avlant Nilsson is a computational biologist and assistant professor in precision medicine at Karolinska Institutet, Stockholm. He holds a MSc (2009-2014), and a PhD (2014-2019) degree in biological engineering from Chalmers University of Technology, where his thesis focused on the metabolism of growing cells. In his postdoctoral work at Massachusetts Institute of Technology (2019-2023), he developed artificial neural network models to simulate signal transduction in immune cells. His new lab at Karolinska will be using these techniques to simulate cellular processes in cancer, aiming at identifying effective drug combinations, predicting resistance mechanisms, and understanding cell-cell interactions in the tumor microenvironment.

Stem cell transplantation is a cornerstone in the treatment of blood malignancies. The most common method to harvest stem cells for transplantation is by leukapheresis, requiring mobilization of CD34+ hematopoietic stem and progenitor cells (HSPCs) from the bone marrow into the blood. Identifying the genetic factors that control blood CD34+ cell levels could reveal new drug targets for HSPC mobilization. Here we report the first large-scale, genome-wide association study on blood CD34+ cell levels. Across 13 167 individuals, we identify 9 significant and 2 suggestive associations, accounted for by 8 loci (PPM1H, CXCR4, ENO1-RERE, ITGA9, ARHGAP45, CEBPA, TERT, and MYC). Notably, 4 of the identified associations map to CXCR4, showing that bona fide regulators of blood CD34+ cell levels can be identified through genetic variation. Further, the most significant association maps to PPM1H, encoding a serine/threonine phosphatase never previously implicated in HSPC biology. PPM1H is expressed in HSPCs, and the allele that confers higher blood CD34+ cell levels downregulates PPM1H. Through functional fine-mapping, we find that this downregulation is caused by the variant rs772557-A, which abrogates an MYB transcription factor-binding site in PPM1H intron 1 that is active in specific HSPC subpopulations, including hematopoietic stem cells, and interacts with the promoter by chromatin looping. Furthermore, PPM1H knockdown increases the proportion of CD34+ and CD34+90+ cells in cord blood assays. Our results provide the first large-scale analysis of the genetic architecture of blood CD34+ cell levels and warrant further investigation of PPM1H as a potential inhibition target for stem cell mobilization.

Clonal hematopoiesis (CH) is a pre-malignant condition characterized by the expansion of genetically distinct blood cell clones in healthy individuals. CH with cancer driver variations is common in the elderly population and is associated with a 10-fold higher risk of blood cancer. A number of genes recurrently mutated in CH are known (e.g., DNMT3A, TET2, and ASXL1); however, these gene variants only account for <30% of CH clones in elderly individuals. To identify new genetic drivers of CH, we analyzed exome sequencing data from peripheral blood samples of >50,000 individuals. We identified CH with genetic variants in several genes previously not linked to CH. Further, we grouped CH into myeloid and lymphoid based on the mutated genes, which stratified the risk of incident myeloid and lymphoid malignancies. Integrating the genetic data with clinical measurements allowed identification of individuals at high risk of developing hematologic malignancies. Currently, we are developing methods to analyze somatic variants across the whole genome to characterize CH and identify new genetic regulators of CH.

I am a DDLS Fellow in precision medicine and diagnostics at the University of Gothenburg. I obtained my PhD from Lund University and did a postdoc at the Broad Institute, Cambridge MA. My research background is on bioinformatics and human genomics, focusing on the effects of genetic variants in human diseases including cancer. As a DDLS Fellow, my research focuses on the study of clonal hematopoiesis (CH), a pre-cancer of blood. We utilize large-scale genomic data from population cohorts to understand the origin and evolution of CH and its malignant transformation.

Chromosomal gains are one of the most common somatic genetic alterations found in cancer. While the impact of sustained oncogene expression has been extensively studied, the effects of copy number gains on “bystander” genes, which are collaterally amplified, remain less understood in terms of cellular fitness. To shed light on this, we integrated the expression and copy number profiles of over 8,000 TCGA tumors and CCLE cell lines, along with the viability effect of gene overexpression from 17 human cancer ORF screens. Through this comprehensive and data-driven analysis, we identified a group of genes termed ‘Amplification-Related Gain Of Sensitivity’ (ARGOS) genes. ARGOS genes are situated in frequently amplified regions of the genome, and their expression level is reduced compared to their copy number status. However, when overexpressed upon gain, they prove to be toxic to the cell. Our compensation and toxicity analyses revealed five frequently amplified ARGOS genes. Notably, one candidate showed a mechanism of toxicity involving altered DNA damage and innate immune signaling responses upon gene overexpression. This study represents a significant effort to better understand the toxicity effects associated with gene overexpression in human cancers.

Veronica studied Biology at Simon Bolivar University in Venezuela, and then moved to Sweden to pursue a PhD at Uppsala University. During this time, she worked under the mentorship of Prof. Tobias Sjoblom and studied how genomic losses occurring in colorectal cancer can be exploited for therapy. In 2019, she moved to Boston (USA) to pursue a postdoc in Dr. Rameen Beroukhim’s lab at Dana-Farber Cancer Institute, affiliated with Harvard Medical School and the Broad Instititute of MIT and Harvard. Veronica’s research focused on studying therapeutic vulnerabilities associated with aneuploidy, including predictors of resistance to clinically relevant p53 reactivation strategies in brain tumors and negative selection against amplifications in cancer. In 2023, Veronica left the Beroukhim lab to start her own laboratory at the Department of Immunology, Genetics and Pathology at Uppsala University. Here, she combines descriptive and functional genomic approaches to understand how brain tumors respond and evolve during treatment. This includes continued exploration of how aneuploidy creates vulnerabilities that can be therapeutically exploited.

Among women attending screening, 30% of their cancer is missed by current mammography-based screening. We have trained AI models to assess specific aspects of the mammography images to measure when a negative exam is less reliable. After retrospective validation we are conducting a randomized clinical trial selecting around 7% of women from screening to be offered MRI.

Fredrik Strand is a Swedish breast radiologist and associate professor at Karolinska Institutet, Stockholm. He holds an MD, PhD and MSc degrees, and has prior experience as a strategy consultant and leading a start-up company. Fredrik is head of the research and education committee at the Swedish society of breast imaging. His academic research evolves around exploring AI for breast imaging. He and his team are involved in retrospective validation of AI algorithms, prospective clinical trials, and in developing new machine-learning algorithms for breast cancer detection and supplemental MRI imaging.