"Days alive and at home" (DAH) is a recent patient-centered outcome measure for perioperative trials, defined as the number of days a patient spends at home during the follow-up period. DAH typically follows a zero-inflated, left-skewed, bi-modal distribution. Other increasingly used complex endpoints, such as days alive without a ventilator, share these statistical features arising from combining survival with another clinically relevant count outcome into a single, comprehensive measure. A key challenge for DAH and similar endpoints is the lack of a readily identifiable distributional form, which complicates the statistical design of trials using it as the primary endpoint, particularly regarding the robustness of sample size calculations and final analyses where the central limit theorem might not be suitable. Using 200 data points from the interim data of the NOTACS trial (ISRCTN14092678), whose primary endpoint was DAH, we developed a novel 'Divide Conquer' model that breaks DAH into distinct parts modeled individually. To our knowledge, such a model has not been used before for DAH. We demonstrate that our approach significantly improves model fit compared to existing alternatives, enabling more suitable DAH data generation that can be used for simulation-based sample size calculations and evaluation of operating characteristics of the statistical test(s). Beyond NOTACS, our work has large potential to inform the design and analysis of other trials using DAH or similar complex endpoints.

The use of Bayesian adaptive designs for randomised controlled trials has been hindered by the lack of software readily available to statisticians. We have developed a new software package (Bayesian Adaptive Trials Simulator Software - BATSS for the statistical software R, which provides a flexible structure for the fast simulation of Bayesian adaptive designs for clinical trials. We illustrate how the BATSS package can be used to define and evaluate the operating characteristics of Bayesian adaptive designs for various different types of primary outcomes (e.g., those that follow a normal, binary, Poisson or negative binomial distribution) and can incorporate the most common types of adaptations: stopping treatments (or the entire trial) for efficacy or futility, and Bayesian response adaptive randomisation - based on user-defined adaptation rules. Other important features of this highly modular package include: the use of (Integrated Nested) Laplace approximations to compute posterior distributions, parallel processing on a computer or a cluster, customisability, adjustment for covariates and a wide range of available conditional distributions for the response.

Abstract Chromosomal instability is a hallmark of cancer that can drive tumour heterogeneity and treatment resistance. However, scalable methods to measure the dynamics of chromosomal instability are lacking and thus historic genomic scars can be hard to distinguish from markers of currently active mutational processes. Here, we introduce scUnique , a novel approach to measure ongoing chromosomal instability from single-cell whole genome sequencing (scWGS) data. scUnique performs phylogeny-aware joint segmentation to refine single-cell copy number profiles and identifies recent copy number aberrations as statistically supported changes on the leaves of the inferred phylogenetic tree. This yields a per-cell measure of ongoing chromosomal instability at genome-wide resolution. We validate scUnique in (i) a comprehensive benchmarking study, (ii) in CRISPR-Cas-based experimental systems, and (iii) single-cell derived clones of a well-characterised ovarian model. We show that scUnique distinguishes ongoing HRD and FBI mutational processes. These results show that scUnique provides quantitative, scalable, genome-level information about ongoing chromosomal instability, not available in previous studies relying on bulk DNA-sequencing or single-cell imaging. In the future, these improved measurements could refine the understanding of mechanisms of chromosomal instability and lead to dynamic biomarkers for improved treatment decisions.

Adaptive clinical trials with time-to-event endpoints are increasingly common. However, implementing such designs in the Bayesian framework has been challenging due to the lack of readily available software and the high computational burden associated with Markov Chain Monte Carlo (MCMC) methods often used for estimating the posterior distributions. Here we present a major extension for time-to-event endpoints to the Bayesian Adaptive Trial Simulator Software (BATSS) R package, enabling flexible and efficient simulation of Bayesian adaptive multi-arm multi-stage (MAMS) designs through a modular and scalable framework.We demonstrate that this extension to BATSS is a powerful tool for evaluating the operating characteristics of trials with time-to-event outcomes with flexible interim analysis schedules that incorporate common adaptive features.

Background: Days Alive and at Home (DAH) over a pre-defined follow-up period is a novel post-intervention composite outcome that combines data from at least three components: (i) initial length of hospital stay, (ii) length of total readmissions or other post-discharge care and (iii) mortality. Missing values bring unique challenges to the analysis of trials with the DAH outcome as the three components may have different rates of missingness caused by distinct missing data mechanisms. Current approaches define DAH as missing if any of the components are missing, and proceed with complete cases or Multiple Imputation (MI) of the composite. Methods: Through a simulation study motivated by the NOTACS trial, we compare several methods of handling missing data, including complete case analysis, MI of the composite, and MI of the components when the primary analysis is a Mann-Whitney-Wilcoxon test. Results: MI on the component level has good properties in terms of type I error control and power. We caution against the use of MI on the composite level with Predictive Mean Matching, which can lead to type I error inflation. Conclusions: Given the complex distributional characteristics of DAH, naive approaches such as defining missingness on the composite level and directly imputing the composite with Predictive Mean Matching, can lead to type I error inflation. Imputing on the component level is recommended, suggested future work included imputation approaches that are compatible with more complex definitions of DAH, as well as recommendations for sensitivity analyses to the Missing at Random assumption.

Emerging evidence suggests that lineage-specifying transcription factors control the progression of pancreatic ductal adenocarcinoma (PDAC). We have discovered a transcription factor switching mechanism involving the poorly characterized orphan nuclear receptor HNF4G and the putative pioneer factor FOXA1, which drives PDAC progression. Using our unbiased protein interactome discovery approach, we identified HNF4A and HNF4G as reproducible, FOXA1-associated proteins, in both preclinical models and Whipple surgical samples. In the primary tumor context, we consistently find that the dominant transcription factor is HNF4G, where it functions as the driver. A molecular switch occurs in advanced disease, whereby HNF4G expression or activity decreases, unmasking FOXA1's transcriptional potential. Derepressed FOXA1 drives late-stage disease by orchestrating metastasis-specific enhancer-promoter loops to regulate the expression of metastatic genes. Overall survival is influenced by HNF4G and FOXA1 activity in primary tumor growth and in metastasis, respectively. We suggest that the existence of stage-dependent transcription factor activity, triggered by molecular compartmentalization, mediates the progression of PDAC.

BACKGROUND:Mutational processes of diverse origin leave their imprints in the genome during tumour evolution. These imprints are called mutational signatures and they have been characterised for point mutations, structural variants and copy number changes. Each signature has an exposure, or abundance, per sample, which indicates how much a process has contributed to the overall genomic change. Mutational processes are not static, and a better understanding of their dynamics is key to characterise tumour evolution and identify cancer cell vulnerabilities that can be exploited during treatment. However, the structure of the data typically collected in this context makes it difficult to test whether signature exposures differ between conditions or time-points when comparing groups of samples. In general, the data consists of multivariate count mutational data (e.g. signature exposures) with two observations per patient, each reflecting a group. RESULTS:We propose a mixed-effects Dirichlet-multinomial model: within-patient correlations are taken into account with random effects, possible correlations between signatures by making such random effects multivariate, and a group-specific dispersion parameter can deal with particularities of the groups. Moreover, the model is flexible in its fixed-effects structure, so that the two-group comparison can be generalised to several groups, or to a regression setting. We apply our approach to characterise differences of mutational processes between clonal and subclonal mutations across 23 cancer types of the PCAWG cohort. We find ubiquitous differential abundance of clonal and subclonal signatures across cancer types, and higher dispersion of signatures in the subclonal group, indicating higher variability between patients at subclonal level, possibly due to the presence of different clones with distinct active mutational processes. CONCLUSIONS:Mutational signature analysis is an expanding field and we envision our framework to be used widely to detect global changes in mutational process activity. Our methodology is available in the R package CompSign and offers an ample toolkit for the analysis and visualisation of differential abundance of compositional data such as, but not restricted to, mutational signatures.

Transcriptomic studies have attempted to classify glioblastoma (GB) into subtypes that predict survival and have different therapeutic vulnerabilities 1–3 . Here we identified three metabolic subtypes: glycolytic, oxidative and a mix of glycolytic and oxidative, using mass spectrometry imaging of rapidly excised tumour sections from two patients with GB who were infused with [U- 13 C]glucose and from spatial transcriptomic analysis of contiguous sections. The phenotypes are not correlated with microenvironmental features, including proliferation rate, immune cell infiltration and vascularization, are retained when patient-derived cells are grown in vitro or as orthotopically implanted xenografts and are robust to changes in oxygen concentration, demonstrating their cell-intrinsic nature. The spatial extent of the regions occupied by cells displaying these distinct metabolic phenotypes is large enough to be detected using clinically applicable metabolic imaging techniques. A limitation of the study is that it is based on only two patient tumours, albeit on multiple sections, and therefore represents a proof-of-concept study.

Bevacizumab treatment in MDA-MD-231 tumors changes systemic VEGF and hemoglobin levels, while influencing VEGF positivity in the responding tumors. A and B, Systemic levels of hVEGF (A) in the circulation are significantly reduced in responding mice but no change in mVEGF (B) was observed (Ctrl, nmice = 5; Bev-NR, nmice = 6; Bev-R, nmice = 3). C, Responding mice also showed a significant increase in systemic levels of biochemical hemoglobin (Ctrl, nmice = 7; Bev-NR, nmice = 8; Bev-R, nmice = 2). D and E, hVEGF measured using IHC in the respective tumors showed a significant increase in positivity in the responders (Ctrl, ntumors = 14, Bev-NR, ntumors = 14; Bev-R, ntumors = 5). Scale bars, 50 µm. P values are displayed from two-sided Student t tests except in D, where a Welch t test was performed because of unequal variances.

To evaluate the capability of hyperpolarized [1-13C] pyruvate MRI to predict pathologic response to neoadjuvant treatment in multi-site abdominopelvic disease of high-grade serous ovarian cancer (HGSOC) patients and to compare 13C MRI and [18F]-FDG PET/CT measurements for detecting early treatment response. We recruited eight patients with HGSOC in this prospective study who underwent 13C MRI and [18F]-FDG PET/CT before and after the first cycle of neoadjuvant chemotherapy treatment (NACT). Imaging parameters were compared with clinical and histophatologic parameters. We demonstrate here that 13C MRI of hyperpolarized [1-13C]pyruvate metabolism in multiple abdominal metastases resulted in rapid labeling of the endogenous tumor lactate pool. The rate of labeling was similar between the different anatomical disease sites and independent of tumor volume. The apparent rate constant describing exchange of 13C label between pyruvate and lactate (kPL) was positively correlated with PET standard uptake values (SUVmax) for [18F]-FDG in metastatic tumor deposits in the ovary/pelvis (R = 0.471, P = 0.02). Decreased lactate labeling could be detected after the first cycle of neoadjuvant chemotherapy and was associated with pathological response. There was no overall decrease in lactate labeling in a single patient who lacked a complete histopathologic response. kPL was associated with cancer tissue LDHA concentration (rho = 0.641; P = 0.02). This exploratory study demonstrates the potential of 13C MRI measurements for assessing early response to neoadjuvant chemotherapy in patients with HGSOC.

PAT detects changes in tumor vasculature in response to bevacizumab treatment. A, Representative PAT images at endpoint denoting deoxyhemoglobin (Hb) on the blue color scale and oxyhemoglobin (HbO2) on the red color scale. Regions of interest shown denote the position of one tumor per mouse (white outline and arrow) and the aorta/vena cava (gray outline and arrow), used as a reference. B, Tumor oxygenation extracted from PAT images (SO2MSOT) showed a significant decrease in both Bev-NR and Bev-R groups compared with Ctrl. The Bev-R group also had significantly lower oxygenation than the Bev-NR group. C, Hemoglobin content extracted from PAT images (THbMSOT) was significantly elevated in the Bev-R group compared with both Bev-NR and Ctrl. Ctrl, ntumors = 14; Bev-NR, ntumors = 14; Bev-R, ntumors = 6. Scale bars, 5 mm. P values are displayed from two-sided Welch (B) and Student (C) t tests.

Bevacizumab provides a survival benefit in a subset of MDA-MB-231 tumors but not MCF7 tumors. A, Illustrative overview of study design (created with BioRender.com). Mouse viewpoint from left side shows only the left flank tumor; right flank tumor not shown. B, Survival curves for control (Ctrl) and bevacizumab (Bev)-treated groups in MCF7 tumor-bearing mice (top, nmice = 4 Ctrl; nmice = 4 Bev) and MDA-MB-231 tumor-bearing mice (bottom, nmice = 8 Ctrl; nmice = 11 Bev). C, Estimated average tumor growth per group for the control group and the bevacizumab-treated group of MDA-MB-231 mice (ntumors = 14 Ctrl; ntumors = 20 Bev), illustrating the durable response obtained in the subset of the treated group denoted as responders (Bev-R) compared with nonresponders (Bev-NR) based on their tumor growth rates (see Supplementary Data). Shaded areas show 95% pointwise confidence bounds for the average. The largest tumor volume recorded in the study was less than 3,200 mm3 in accordance with our ethical limits; data beyond this volume is an extrapolation of our linear model to illustrate the trajectory of growth in Ctrl and Bev-NR tumors. Weeks denoted in B and C are from the time of enrollment.

Over the past 2 decades, innovations in trial design have significantly advanced the field of clinical research. Methodological developments, such as adaptive designs, basket trials, umbrella trials, and platform trials, along with technological advancements such as virtual studies have proven effective in tackling complex research questions and managing resource constraints. These approaches enable prospectively planned modifications to trial designs and facilitate addressing multiple research questions within a single infrastructure, with technological advancements such as virtual studies enhancing accessibility, efficiency, and patient engagement. These designs can also integrate biomarker information or risk-prediction scores to enhance the efficacy of future clinical trials, through a better selection of patients. Despite the appealing flexibility of these new approaches, their adoption varies across different therapeutic domains. We explored the translation and relevance of these innovative approaches in skin cancer research with a focus on melanoma. An overview of existing melanoma clinical trials that incorporate innovative features as well as other potential studies currently under consideration are discussed. This paper highlights the potential of innovative approaches to optimize melanoma trials under the constraints of limited patient and financial resources. The presented ideas can easily be extended to other nonmelanoma skin cancer trials.

Multivariate (average) equivalence testing is widely used to assess whether the means of two conditions of interest are "equivalent" for different outcomes simultaneously. In pharmacological research for example, many regulatory agencies require the generic product and its brand-name counterpart to have equivalent means both for the AUC and Cmax pharmacokinetics parameters. The multivariate Two One-Sided Tests (TOST) procedure is typically used in this context by checking if, outcome by outcome, the marginal 100 ( 1 - 2 α ) % $$ 100\left(1-2\alpha \right)\% $$ confidence intervals for the difference in means between the two conditions of interest lie within predefined lower and upper equivalence limits. This procedure, already known to be conservative in the univariate case, leads to a rapid power loss when the number of outcomes increases, especially when one or more outcome variances are relatively large. In this work, we propose a finite-sample adjustment for this procedure, the multivariate α $$ \alpha $$ -TOST, that consists in a correction of α $$ \alpha $$ , the significance level, taking the (arbitrary) dependence between the outcomes of interest into account and making it uniformly more powerful than the conventional multivariate TOST. We present an iterative algorithm allowing to efficiently define α * $$ {\alpha}^{\ast } $$ , the corrected significance level, a task that proves challenging in the multivariate setting due to the inter-relationship between α * $$ {\alpha}^{\ast } $$ and the sets of values belonging to the null hypothesis space and defining the test size. We study the operating characteristics of the multivariate α $$ \alpha $$ -TOST both theoretically and via an extensive simulation study considering cases relevant for real-world analyses-that is, relatively small sample sizes, unknown and possibly heterogeneous variances as well as different correlation structures-and show the superior finite-sample properties of the multivariate α $$ \alpha $$ -TOST compared to its conventional counterpart. We finally re-visit a case study on ticlopidine hydrochloride and compare both methods when simultaneously assessing bioequivalence for multiple pharmacokinetic parameters.

Longitudinal analysis demonstrates that PAT can be used to indicate survival benefit from bevacizumab therapy. A and B, Oxygenation (SO2MSOT) and normalized hemoglobin content (THb) increase from enrollment (noted as “pre”) to endpoint of tumor excision (noted as “post”) in the Ctrl group (A), while in the Bev-NR group (B), SO2MSOT increases concurrently with a decrease in normalized THb. C, Conversely, the Bev-R group trend toward a significant decrease in SO2MSOT, with no significant change in normalized THb observed. D, Estimated SO2MSOT level as a function of time (number of days from enrollment) and group (color). Trendlines are shown alone for clarity; individual tumor trajectories are shown in Supplementary Fig. S5. Ctrl tumor data parallel Bev-NR data. Analysis of SO2MSOT over time during the study showed a significant (P < 0.0001) change in slope at 3 weeks after enrollment in the Bev-R group compared with either of the Ctrl or Bev-NR groups. Ctrl, ntumors = 14 and nmice = 8; Bev-NR, ntumors = 14 and nmice = 8; Bev-R, ntumors = 6 and nmice = 3. P values are displayed from two-sided paired Student t tests; shaded areas in D denote the 95% pointwise confidence bounds for the average.

BACKGROUND/OBJECTIVES:Pseudo-vascular network formation in vitro is considered a key characteristic of vasculogenic mimicry. While many cancer cell lines form pseudo-vascular networks, little is known about the spatiotemporal dynamics of these formations. METHODS:Here, we present a framework for monitoring and characterising the dynamic formation and dissolution of pseudo-vascular networks in vitro. The framework combines time-resolved optical microscopy with open-source image analysis for network feature extraction and statistical modelling. The framework is demonstrated by comparing diverse cancer cell lines associated with vasculogenic mimicry, then in detecting response to drug compounds proposed to affect formation of vasculogenic mimics. Dynamic datasets collected were analysed morphometrically and a descriptive statistical analysis model was developed in order to measure stability and dissimilarity characteristics of the pseudo-vascular networks formed. RESULTS:Melanoma cells formed the most stable pseudo-vascular networks and were selected to evaluate the response of their pseudo-vascular networks to treatment with axitinib, brucine and tivantinib. Tivantinib has been found to inhibit the formation of the pseudo-vascular networks more effectively, even in dose an order of magnitude less than the two other agents. CONCLUSIONS:Our framework is shown to enable quantitative analysis of both the capacity for network formation, linked vasculogenic mimicry, as well as dynamic responses to treatment.

AbstractBackgroundEvaluating AI-based segmentation models primarily relies on quantitative metrics, but it remains unclear if this approach leads to practical, clinically applicable tools.PurposeTo create a systematic framework for evaluating the performance of segmentation models using clinically relevant criteria.Materials and MethodsWe developed the AUGMENT framework (Assessing Utility of seGMENtation Tools), based on a structured classification of main categories of error in segmentation tasks. To evaluate the framework, we assembled a team of 20 clinicians covering a broad range of radiological expertise and analysed the challenging task of segmenting metastatic ovarian cancer using AI. We used three evaluation methods: (i) Dice Similarity Coefficient (DSC), (ii) visual Turing test, assessing 429 segmented disease-sites on 80 CT scans from the Cancer Imaging Atlas), and (iii) AUGMENT framework, where 3 radiologists and the AI-model created segmentations of 784 separate disease sites on 27 CT scans from a multi-institution dataset.ResultsThe AI model had modest technical performance (DSC=72±19 for the pelvic and ovarian disease, and 64±24 for omental disease), and it failed the visual Turing test. However, the AUGMENT framework revealed that (i) the AI model produced segmentations of the same quality as radiologists (p=.46), and (ii) it enabled radiologists to produce human+AI collaborative segmentations of significantly higher quality (p=<.001) and in significantly less time (p=<.001).ConclusionQuantitative performance metrics of segmentation algorithms can mask their clinical utility. The AUGMENT framework enables the systematic identification of clinically usable AI-models and highlights the importance of assessing the interaction between AI tools and radiologists.Summary statementOur framework, called AUGMENT, provides an objective assessment of the clinical utility of segmentation algorithms based on well-established error categories.Key resultsCombining quantitative metrics with qualitative information on performance from domain experts whose work is impacted by an algorithm’s use is a more accurate, transparent and trustworthy way of appraising an algorithm than using quantitative metrics alone.The AUGMENT framework captures clinical utility in terms of segmentation quality and human+AI complementarity even in algorithms with modest technical segmentation performance.AUGMENT might have utility during the development and validation process, including in segmentation challenges, for those seeking clinical translation, and to audit model performance after integration into clinical practice.

Abstract Despite recent advances in the treatment of pancreatic adenocarcinoma (PDAC), the median survival remains <12 months. Patients typically present with late-stage disease and have limited treatment options. Therefore, there is an immediate need for the generation of new and innovative therapeutic targets. Little is known about the role of pioneer factors such as FOXA1 and their interaction partners in a PDAC context. Inspired by literature reports implicating FOXA1 and GATA5/GATA6 in regulating pancreatic cancer resistance and metastasis, our study hypothesised the possibility of an undiscovered nuclear receptor that works with FOXA1 (similar to Estrogen receptor in breast cancer). Using innovative ‘omic’ based approaches (RIME) developed in our laboratory 1 we discovered a nuclear receptor (NR) complex involving HNF4A and HNF4G in the classical sub-type of pancreatic cancer. Subsequently the interaction was independently validated in Whipple surgical biopsies from PDAC patients using ChIP-sequencing studies. Across multiple patient tumors (n=7) a binding overlap specifically between FOXA1 and HNF4G (3461 sites) was established. To investigate the therapeutic potential of HNF4G in the classical subtype of PDAC, we generated CRISPR deletions (KO) of HNF4G in the HPAF-II cells. HNF4G-KO cells were orthotopically implanted into the pancreas of NSG mice. A significant survival advantage of 12 days (<0.001) and reduced growth (<0.02) was observed in these mice compared to the controls. To better understand the impact of HNF4G-KO on gene expression, the orthotopic tumors were subjected to RNA-seq analyses. Gene set enrichment and pathway analysis revealed significantly down-regulated pathways to include EMT transition. TCGA data highlighted HNF4G amplification in 9% of PDAC patients with concomitant decreased progression free survival. These data point towards HNF4G being a therapeutically viable target for further exploration. Protein Arginine Methyl Transferase 1 (PRMT1) is a common interactor of both FOXA1 and HNF4G. Our study reveals a unique dependency of PRMT1 on the HNF4G-FOXA1 complex in PDAC biopsies. HNF4G-KO drastically reduces the chromatin binding of PRMT1, implicating them as functionally dependent. Treatment of HNF4G-KO cells with GSK3368715 (PRMT1 inhibitor) further sensitizes these cells and results in a significant survival advantage (10 days <0.02). We propose a model of HNF4G inhibition in combination with PRMT1 as a novel therapeutic opportunity to treat the classical sub-type of PDAC. Further, our study reveals a unique reliance of primary classical tumors on HNF4G. Although HNF4G appears to dominate FOXA1 functionality in the primary disease, FOXA1 remains instrumental in priming metastasis and enhancer reprograming. 1. Papachristou EK, Kishore K, Holding AN, Harvey K, Roumeliotis TI, Chilamakuri CSR, et al. A quantitative mass spectrometry-based approach to monitor the dynamics of endogenous chromatin-associated protein complexes. Nature Communications. 2018;9(1):2311. Citation Format: Shalini V. Rao, Lisa Young, Danya Cheeseman, Stephanie Mack, Jill Temple, Chandra Shekar Reddy Chilamakuri, Evangelia Papachristou, Catherina Pelicano, Amy Smith, Dominique-Laurent Couturier, Michael Gill, Duncan Jodrell Jodrell, Alasdair Russell, Igor Chernukhin, Jason Carroll. New insights into the role of FOXA1- HNF4 axis in pancreatic cancer [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Pancreatic Cancer; 2023 Sep 27-30; Boston, Massachusetts. Philadelphia (PA): AACR; Cancer Res 2024;84(2 Suppl):Abstract nr C076.

Abstract Despite recent advances in the treatment of pancreatic adenocarcinoma (PDAC), the median survival remains <12 months. Patients typically present with late-stage disease and are often unable to tolerate drug combination regimens. Large-scale genomic approaches categorise PDAC into two distinct molecular subtypes, based on transcriptomic signatures, and these are termed the classical and squamous (or basal) subtypes. Although it is well-described that PDAC harbours gene mutations (i.e. KRAS G12D, p53 R172H) that initiate the disease, little is known about the influence of lineage-defining Transcription Factors (TF) on PDAC subtypes and their contribution to tumor progression. We took an unbiased approach to discover driving TFs in PDAC, we performed ChIP-seq of the active enhancer mark H3K27Ac from surplus tissue following resection of primary PDAC tumors (n=6) and normal adjacent tissue (n=4). Analysis of the H3K27Ac sites showed that the most enriched motifs within all experimentally mapped enhancer elements were HNF motifs (HNF4G & HNF4A) and Forkhead (i.e., FOXA1), confirming that these two classes of TFs constitute the lineage-defining factors in these clinical samples. Investigation of publicly available clinical datasets revealed HNF4G to be amplified in 9% of PDAC patients. Using Rapid Immunoprecipitation and Mass Spectrometry of Endogenous protein complexes (RIME), our method for unbiased discovery of endogenous protein complexes, we investigated the fundamental differences in the FOXA1 interactome between models of classical and squamous subtypes. HNF4A and HNF4G were only seen as FOXA1-interactors in models of the classical subtype. HNF4G Knockout (KO) clones injected orthotopically into the pancreas of mice, showed a significant increase (p-value<0.0006) in median survival and slower tumor growth compared to Control (Cas-9) (p-value <0.0001). HNF4G-KO dependent genes were specifically expressed in primary tumors (with decreased EMT signalling) versus metastatic samples (PanCRux dataset), implying that the HNF4G target genes were mostly limited to primary tumor contexts. We found Protein Arginine Methylate Transferase (PRMT1) to be a significant and reproducible interactor of FOXA1 and HNF4G. We observed a significant global reduction in PRMT1 chromatin binding in both the HNF4G-KO and FOXA1 depleted cells compared to the control. This data suggests a dependency on both HNF4G and FOXA1 for recruitment of PRMT1 to chromatin. We find HNF4G to be the key driver of classical subtype primary disease, in part by recruiting the methyltransferase PRMT1 to regulatory regions revealing a therapeutically exploitable opportunity. A molecular switch occurs in advanced disease, whereby HNF4G expression decreases, unmasking FOXA1’s transcriptional activity. De-repressed FOXA1 drives late-stage disease, by orchestrating metastasis-specific enhancer-promoter loops to direct metastatic gene regulation. We suggest the existence of stage-dependent TF activity, triggered by molecular compartmentalisation that mediates multistage progression of PDAC. Citation Format: Shalini Vasudev Rao, Lisa V Young, Danya Cheeseman, Stephanie V Mack, Dominique-Laurent Couturier, Sean V Flynn, Rebecca Brias, Amy Smith, Jill Temple, Duncan Jodrell, Alasdair Russell V Russell, Igor Chernukhin, Jason Carroll. Transcription factor switching drives progression of the classical subtype of pancreatic cancer [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Pancreatic Cancer Research; 2024 Sep 15-18; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2024;84(17 Suppl_2):Abstract nr C071.