BACKGROUND:Older women are underrepresented in hereditary cancer guidelines, and the prevalence of pathogenic or likely pathogenic (P/LP) germline variants in ovarian cancer (OC) patients aged ≥70 remains poorly characterized. Current recommendations for risk-reducing interventions and surveillance rely heavily on variant and family history (FH), yet their applicability in older populations is uncertain. We evaluated age-stratified prevalence of P/LP variants at OC diagnosis to identify at-risk older patients and inform more inclusive genetic counseling strategies. METHODS:We conducted a retrospective analysis of 113,236 OC patients undergoing germline testing through the Myriad Collaborative Research Registry (1996-2024). Variant prevalence, ancestry, FH, and testing patterns were evaluated across age groups, with focus on patients ≥70. RESULTS:Overall, 14,513 patients (12.8%) harbored a P/LP variant, including 13,049 (11.5%) in established OC susceptibility genes. Among patients ≥70 (20%, n = 22,593), 6.6% carried variants in established genes compared with 12.8% in those <70 (p < 0.01). BRCA1/2 and Lynch syndrome variants declined with age; BRCA2 exceeded BRCA1 in patients ≥70. Moderate-penetrance variants (BRIP1,PALB2) were relatively enriched, while ATM and RAD51C/D remained stable. PMS2 were more common than previously reported and declined with age (p = 0.03). Patients ≥70 were 1.5-fold less likely to report a cancer FH. CONCLUSION:Despite lower prevalence, a clinically meaningful proportion of patients ≥70 harbor actionable P/LP variants, often without FH that would prompt risk-reducing interventions. These findings suggest reduced sensitivity of FH-based risk assessment in older individuals, supporting age-inclusive genetic counseling and testing strategies extending beyond FH-based criteria.

DNA extracted from tissue samples typically derives from a complex mixture of cell types. Without single cell analysis, it has been generally impossible to determine the cell type of origin for most molecules. One clear example of this is in the complex milieu of a human neoplasm. Here, we develop ROCIT (https://github.com/tobybaker/rocit), a transformer-based model to classify the tumor or non-tumor origin of individual reads from bulk tumor samples sequenced with long-read whole genome sequencing. Using somatic mutations to derive training data, ROCIT uses read-level methylation patterns to accurately classify reads from anywhere in the genome without requiring the adjacent normal tissue or the explicit identification of tumor differentially methylated regions. We apply ROCIT to a cohort of prostate and ovarian tumors and demonstrate high classification accuracy across the entire genome. We then demonstrate the potential of ROCIT predictions to improve somatic variant calling. ROCIT represents a major step forward in the analysis of bulk tumors with long-reads, enabling the accurate and sensitive identification of reads with specific cell types of origin genome-wide.

The development of new therapeutics and the validation of pathogenetic cancer mechanisms require representative laboratory models1,2. However, existing collections represent only a fraction of the diversity observed in human cancer2-4. Recent technologies have enabled efficient in vitro model derivation (for example, tumour organoids)5. However, whether these maintain essential properties of patient tumours during long-term expansion has not been systematically investigated. Here we present results of a large-scale international programme-the Human Cancer Models Initiative-which involved the generation of a resource of 665 next-generation models from 2,780 donors with 25 cancer types and integrated tumour-model whole genome, exome, methylome and transcriptome analyses. The resource provides 522 models with comprehensive clinical data, 153 models of rare cancers and 71 models from participants with non-European ancestry. Analyses of 421 matched tumour-model pairs reveal high genetic (97.8%) and epigenetic (95%) concordance and define correlates of model discordance. Single-nucleus RNA sequencing of tumour-model pairs reveals subsets of models in which culture conditions significantly influence cell states. Finally, we characterize model preservation of extrachromosomal DNA and post-treatment mutational signatures to provide opportunities to study therapeutic resistance. This model repository is being made available to the community-including multimodal molecular profiling, clinical information and integrative software tools-thus providing a valuable resource for preclinical investigation of cancer pathogenesis and treatment response.

Table S1 shows the number of possible routes for different copy numbers states.

Linking genetic data with electronic health records in hospital biobanks promises to advance precision medicine, but limited ancestral diversity constrains discovery and generalizability. We analyzed 93,936 participants from the UCLA ATLAS Community Health Initiative to inform disease prevalence and genetic risk across five continental and 36 fine-scale ancestry groups. We discovered numerous unreported gene-phenotype associations, including FN3K with intestinal disaccharidase deficiency in Europeans and admixed Americans. Polygenic scores (PGS) robustly predicted common diseases, with effects markedly diminished in non-Europeans. Furthermore, we reduced the pronounced European bias in curated clinical variants using computational predictors, uncovering unreported disease-gene associations, including ANKZF1 and peripheral vascular disease in African Americans. Longitudinal data revealed that semaglutide efficacy varies across ancestries, is associated with PGS for type 2 diabetes, and is modulated by genetic variation in PTPRU. These findings illustrate how ancestrally diverse biobanks from a single health system yield robust disease associations and pharmacogenomic insights.

Autosomal monoallelic gene expression and asynchronous replication between alleles are established features of imprinted genes and genes regulated by allelic exclusion. Inactivation/Stability Centers (I/SCs) are recently described autosomal loci that exhibit epigenetic regulation of allelic expression and replication timing, with differences that can be comparable to those observed between the active and inactive X chromosomes . Here, we characterize >100 autosomal loci with allele-specific epigenetic regulation of replication timing and gene expression, defining them as I/SCs. I/SCs are approximately 1 Mbb in size and can contain both protein-coding and noncoding genes. In different single-cell derived clones, these genes may be expressed from a single allele, the opposite allele, both alleles, or not expressed at all. This stochastic, yet mitotically stable, pattern indicates that the choice of which allele is expressed is independent of parent of origin and independent of the expression status of the other allele. Similarly, alleles within I/SCs show varying replication timing, either earlier or later, that is also independent of the other allele. Additionally, we identify syntenic loci in the mouse genome that display epigenetic regulation of allelic replication timing, highlighting the genomic organization and conservation of I/SC-associated regulation between human and mouse genomes. The allele-restricted regulation described here creates extensive cellular mosaicism through a stable epigenetic mechanism. This mosaicism impacts numerous dosage-sensitive genes associated with human diseases such as Alzheimer, Parkinson, epilepsy, deafness, and impaired intellectual development.

Supplementary Figures S1 to S52 S1 WGD frequencies across cancer types and stage. S2 Effect of WGD constraint on timing accuracy. S3 Measuring timing accuracy on simulated data. S4-7 Measuring timing accuracy on simulated data by copy number state. S8-11 Measuring inferred route probabilities on simulated data. S12 Timing of gains in multi-region tumors. S13-14 Difference in timing between different gain routes. S15 Non-parsimony in copy number gain evolution. S16 Non-parsimony by copy number state. S17-18 Calibrating a penalty on non-parsimony. S19 Clear-cell sample gain timing. S20 Gain route agreement within chromosomes. S21 Probability of pre-WGD gains in different chromosomes and copy number states. S22 Distribution of gain timing by major copy number. S23 Single-cell copy number profiles of an undifferentiated sarcoma. S24 Distribution of gain rates relative to WGD by cancer type. S25 Distribution of gain rates relative to WGD compared to simulations. S26 Example sample gain timing posterior. S27 Combined distribution over gain timing by WGD status. S28 The timing of gains relative to WGD. S29 The timing of gains relative to WGD by cancer type. S30 Proportion of copy number events post-WGD. S31 The relationship between genome gained post-WGD and WGD timing by cancer type. S32 The relationship between genome gained pre-WGD and WGD timing by cancer type. S33 The relationship between fraction of genome lost pre and post-WGD and WGD timing by cancer type. S34 Punctuated gains in WGD tumors. S35 Association between chromothripsis and punctuated gains. S36 Genomic features of punctuated gains. S37 Frequency of arm gains pre and post-WGD and in non-WGD tumors. S38 Frequency of arm gains pre and post-WGD and in non-WGD tumors by cancer type. S39 Frequency of arm losses pre and post-WGD and in non-WGD tumors by cancer type. S40 Effect of oncogene and tumor suppressor gene density on arm gain rates. S41 Effect of oncogene and tumor suppressor gene density on arm loss rates. S42-46 Pan-genome frequencies of pre and post-WGD gains by cancer type. S47-49 Pan-genome frequencies of pre and post-WGD losses by cancer type. S50 The effect of NRPCC and mutation count on gain timing inference. S51 WGD status calling in GRITIC. S52 The effect of the non-parsimony penalty on event timing.

Supplementary methods, including full description of the GRITIC method

Advancing precision oncology requires a diverse and robust repository of patient-derived cancer models that faithfully reflect the clinical, therapeutic, and molecular attributes of patients’ tumor. Traditional models, while invaluable, often lack sufficient clinical data and fail to capture the genetic and phenotypic complexity of human cancers, limiting their translational relevance. To address these challenges, the Human Cancer Models Initiative (HCMI) generated 665 next-generation cancer models from over 2,500 donors across 27 cancer subtypes. The HCMI cohort comprises 78% three-dimensional (3D) organoid cultures from 19 tumor types, 6% 3D neurospheres, and 16% two-dimensional (2D) adherent cell lines from 13 tumor types and includes 48 unique models from 17 rare cancer subtypes. Molecular profiling using whole-genome and exome sequencing, RNA-seq, and methylation analysis shows high concordance between models and their parent tumors, underscoring the fidelity and preclinical relevance of these models. Analysis also identified rare molecular targets and expression states in cancer types underrepresented in current collections, including desmoid tumors, breast lobular carcinoma, and nephroblastoma, significantly advancing research into rare and undercharacterized malignancies. Analyses of treatment-naive (62%) and post-treatment (38%) models revealed molecular features associated with therapeutic exposures, supported by a median clinical follow-up time of 1.4 years. These exposures include chemotherapy, immunotherapy, targeted therapies, molecular antibody treatments, and radiation, enabling preclinical correlation. Treatment-associated mutational signatures, such as temozolomide (TMZ)-induced alterations in glioblastoma and APOBEC mutational activity in colorectal cancer, were observed and correlated with patient outcomes. We identified specific models with preserved patterns of extrachromosomal DNA (ecDNA) amplification of oncogenes of high interest for preclinical studies, including EGFR, MDM4, CCND1, MYC and FGFR3. Such targets were preserved across paired tumors and models including correlation with increased target expression in models. To facilitate accessibility and exploration, HCMI models and their accompanying clinical data are made available through multiple platforms, including the HCMI searchable catalog, cBioPortal, the HCMI Explorer Suite for interactive analyses of treatment timelines and molecular profiles, and the ATCC. In summary, this comprehensive resource should aid both basic research and therapy development for future precision oncology studies. Dina ElHarouni, Mushriq Al-Jazrawe, Seongmin Choi, Merve Dede, Toshinori Hinoue, Sean A. Misek, Heeju Noh, Luca Zanella, Moony Tseng, Hayley E. Francies, Priya Sridevi, Rachana Agarwal, Cindy W. Kyi, Julyann Perez-Mayoral, Megan J. Stine, Eva Tonsing-Carter, James M. Clinton, William F. Hooper, Jennifer M. Shelton, Timothy R. Chu, The HCMI Network, Peter W. Laird, Calvin J. Kuo, Olivier Elemento, David L. Spector, Andrew D. Cherniack, Kyle Ellrott, Martin L. Ferguson, Rameen Beroukhim, Katherine A. Hoadley, Nicolas Robine, Andrew McPherson, Mathew J. Garnett, David A. Tuveson, Andrea Califano, Paul T. Spellman, Keith L. Ligon, Daniela S. Gerhard, Louis M. Staudt, Jesse S. Boehm. The landscape of molecular targets across 665 next-generation cancer models in the human cancer models initiative identifies opportunities for improved therapy and overcoming resistance [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 1165

The discovery of novel cancer therapeutics requires evaluation in representative laboratory models. However, existing model collections do not represent the full spectrum and scale of molecular diversity required to support biological discovery and drive pre-clinical research. Recent technologies have enabled more efficient derivation of cell-based models (e.g., organoids). However, systematic pan-cancer comparisons of the integrity of genomic, transcriptomic and epigenomic states following long term ex vivo expansion remain elusive. Here, we report on a large-scale international program—the Human Cancer Models Initiative (HCMI), consented more than 2,500 donors resulting in the production and subsequent characterization of 665 organoid, neurosphere and cell line models by whole genome, exome, methylome, and transcriptome analysis. Critical properties of the donors and tissues included 71 derived from populations beyond European ancestry, primarily African (54%), Asian (45%) and Indigenous American (1%); 448 have therapeutic treatment profiles from donor medical records, and 45 represented rare cancers. Integrative analysis between models and source tumors shows that 92-96% of 417 analyzed models were highly concordant by molecular characterization. The small number of discordant behaviors were associated with either cell culture media influences in glioblastoma or stromal infiltration or rare cell differentiation state shifts, which were validated by single cell RNA sequencing. We validate the utility of the resource studying therapeutic resistance to chemo-, targeted-, and immuno-therapies by integrative analysis of extrachromosomal DNA-based gene amplification and post-treatment mutational signatures. This internationally accessible community resource, including data and accompanying integrative software tool packages, provides a roadmap for capturing an increasing spectrum of cancer diversity in preclinical models. Dina ElHarouni, Mushriq Al-Jazrawe, Seongmin Choi, Merve Dede, Toshinori Hinoue, Sean A. Misek, Heeju Noh, Luca Zanella, Yuen-Yi Tseng, Hayley E. Francies, Priya Sridevi, Rachana Agarwal, Cindy W. Kyi, Julyann Perez-Mayoral, Megan Stine, Eva Tonsing-Carter, James M. Clinton, The Human Cancer Models Initiative Network, Peter W. Laird, Calvin J. Kuo, Olivier Elemento, David L. Spector, Andrew D. Cherniack, Kyle Ellrott, Martin L. Ferguson, Rameen Beroukhim, Katherine A. Hoadley, Nicolas Robine, Andrew McPherson, Mathew J. Garnett, David A. Tuveson, Andrea Califano, Paul T. Spellman, Keith Ligon, Daniela S. Gerhard, Louis M. Staudt, Jesse S. Boehm. Integrative clinical and molecular characterization of an international community resource of organoid models of human cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4005.

Patient-derived cancer models advance therapeutic development and are essential to studies of treatment resistance mechanisms. The Human Cancer Models Initiative (HCMI) has created 665 patient-derived models, including organoids, neurospheres, and conditionally reprogrammed cells, all paired with patient tumors and clinical annotations. This resource provides an opportunity to study genomic conservation and divergence during model development at scale in a large pan-cancer cohort of matched tumor-model pairs. We characterized the genomes of models and parent tumors, computing allele specific Copy Number Variation (CNV), Single Nucleotide Variants (SNV), small Insertions and Deletions (InDels) and Structural Variations (SV) from bulk whole genome and whole exome Sequencing. We used pyclone to study changes in clonal abundances between tumor and model, MutationTimeR to understand the timing of model specific CNVs and Whole Genome Doubling (WGD), and AmpliconArchitect to identify ecDNA. HCMI models exhibited high concordance with matching tumors when evaluated on SNVs, SVs, InDels and CNVs, with 69% of SNV/InDel drivers and 43% of CNV drivers conserved between tumor and model. Only 2% of investigated models (9/405) showed significantly divergent patterns of aneuploidy and presence of distinct clonal outgrowth in the model. This included two pancreatic cancer models with oncogenic KRAS point mutations that differed from those in the parent tumor. The cohort included models and tumors with ecDNA affecting known oncogenes including EGFR in GBM samples, CCNE1/KRAS in esophageal cancer and MYC in pancreatic cancer. ecDNA were the least conserved aberration type; 20% (50/254) of tumor-detected ecDNA were retained in an associated model while 35% (50/143) of model-detected ecDNA were found in their parent tumors. While the tumors and models generally showed similar WGD states, 10% of models exhibited WGD private to the model, and this model specific WGD generally occurred later in estimated mutational evolution time than models for which the tumor and model were both WGD. Surprisingly, not all models were pure for cancer cells with 19 models, enriched for melanomas, containing more than 20% diploid presumably non-tumor cells. HCMI models exhibited high genomic concordance with their matching tumors across all cancer types. We catalogued classes of genomic divergence representing either selection of subclones or evolution during model development. SNVs and InDels exhibited greater conservation compared to CNVs, with ecDNA showing the highest variability between the tumor and the model. Our results provide full visibility into the genomic fidelity of each HCMI model, and enable researchers to select the most relevant HCMI model for their study. Andrew W. McPherson, Seongmin Choi, William F. Hooper, Jennifer M. Shelton, Timothy R. Chu, Dina ElHarouni, Mushriq Al-Jazrawe, Merve Dede, Toshinori Hinoue, Sean A. Misek, Heeju Noh, Luca Zanella, Moony Tseng, Hayley E. Francies, Priya Sridevi, Rachana Agarwal, Cindy W. Kyi, Julyann Perez-Mayoral, Megan J. Stine, Eva Tonsing-Carter, James M. Clinton, The HCMI Network, Peter W. Laird, Calvin J. Kuo, Olivier Elemento, David L. Spector, Andrew D. Cherniack, Kyle Ellrott, Martin L. Ferguson, Rameen Beroukhim, Katherine A. Hoadley, Nicolas Robine, Mathew J. Garnett, David A. Tuveson, Andrea Califano, Paul T. Spellman, Keith L. Ligon, Daniela S. Gerhard, Louis Staudt, Jesse Boehm. Patterns of genomic patient-model conservation and evolution in the Human Cancer Models Initiative (HCMI) next-generation cancer model resource [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3882.

AbstractQuantification and detection of circulating tumor DNA (ctDNA) has been used to identify the presence of cancers. Ablative radiation therapy kills tumor cells to reduce tumor burden and it follows that these dying tumor cells could lead to increased ctDNA abundance. We carried out deep, error-corrected sequencing of cell-free DNA collected serially from 12 stage I, and 2 stage II/III non-small cell lung cancer (NSCLC) patients undergoing external-beam radiation treatment (EBRT) after initial diagnosis. We found that ctDNA detection rates decreased at the first blood draw as compared to baseline (43% to 7% of patients). Total ctDNA abundance decreased in 6 patients and increased in 5 patients between those same blood draws, with one patient showing evidence of tumoral heterogeneity. Both patients with stage II/III disease had the largest increases in ctDNA abundance from baseline. Multiple blood draws improved ctDNA detection rates from 43% to 50% with a second blood draw and to 71% with 4 blood draws. Additionally, EGFR mutations were detectable in 6 patients during EBRT that were not detected prior to treatment. Taken together, these results provide an early-stage NSCLC counterpoint to previous work that reported improved ctDNA detection after radiation therapy in more advanced disease.

Autosomal monoallelic gene expression and asynchronous replication between alleles are well-established features of imprinted genes and genes regulated by allelic exclusion. Inactivation/Stability Centers (I/SCs) are recently described autosomal loci that exhibit epigenetic regulation of allelic expression and replication timing, with differences that can be comparable to those observed between the active and inactive X chromosomes 1 . Here we characterize hundreds of autosomal loci with allele-specific epigenetic regulation of replication timing and gene expression, defining them as I/SCs. I/SCs are approximately 1 megabase in size and can contain both protein-coding and noncoding genes. In different single cell derived clones, these genes may be expressed from a single allele, the opposite allele, both alleles, or not expressed at all. This stochastic, yet mitotically stable, pattern indicates that the choice of which allele is expressed is independent of parent of origin and independent of the expression status of the other allele. Similarly, alleles within I/SCs show varying replication timing, either earlier or later, that is also independent of the other allele. Additionally, we find corresponding I/SCs in the mouse genome that display conserved synteny with human I/SCs. This allele-restricted regulation creates extensive cellular mosaicism through a stable epigenetic mechanism. This mosaicism impacts numerous dosage-sensitive genes associated with human diseases such as Parkinson, epilepsy, deafness, and impaired intellectual development.

Next generation cancer models, including organoids, neurospheres and cell lines, are significantly advancing our ability to study individual human tumors in vitro and to develop novel cancer therapeutics in settings that resemble in vivo conditions. However, recent questions have emerged as to the degree to which in vitro models may faithfully represent intra- and inter-patient cell state heterogeneity at the transcriptional level of single cells. The Human Cancer Models Initiative (HCMI), a global collaboration involving the NCI (NIH), Cancer Research UK, the Wellcome Sanger Institute and the Hubrecht Organoid Technology foundation, has generated a large repertoire of 665 patient-derived models of human cancers, including organoids, neurospheres and conditionally reprogrammed cells, with matching parental tumor and clinical annotations. We conducted a tumor-model fidelity analysis using bulk profiles from Whole Genome Sequencing (WGS) and RNA-seq and confirmed strong cell-state concordance between each model and its parental tumor across most HCMI tumor-model pairs, thus showcasing models’ ability to retain key genetic, transcriptional and epigenetic attributes of their parental tumors. However, our analysis also identified divergent features in a subset (<8%) of models, potentially resulting in a loss of fidelity and challenging their translational use. To identify whether this discordance might be driven by ex vivo adaptations or selective evolutionary pressures from in vitro culture, we profiled a representative subset (n=13 pairs) of glioblastoma (GBM, n=7) and pancreatic cancer (PAAD, n=6) tumor-model pairs using single-nucleus RNA-seq (snRNA-seq). Analysis of 132,593 nuclei (tumors: 66,054; models: 66,539) identified outlier GBM models undergoing proneural-to-mesenchymal (PTM) transition and highlighted serum culture media as a major associated factor of this process. Additionally, we found that divergences in selected PAAD (72,688 nuclei: tumors: 30,254; models: 42,434) pairs were linked to the complete loss of the rich tumor microenvironment cells in the original tumors and the in vitro expansion of select malignant subpopulations, as identified by network-based analysis of VIPER-inferred protein activity profiles. The observed divergences suggest that in vitro culture conditions can occasionally drive cellular reprogramming and alter phenotypic fidelity, resulting in varying degrees of state-specific transitions. However, the majority of the models (even those classified as outliers) are preserved for state-frequency in cellular subpopulations between tumors and models, and most if not all of the cellular states found in the parental tumor, thus faithfully representing intra-tumor heterogeneity. Taken together, these findings highlight the translational potential of the HCMI model repository and its translational relevance as an invaluable tool to support precision oncology. Dina ElHarouni, Mushriq Al-Jazrawe, Seongmin Choi, Merve Dede, Toshinori Hinoue, Sean A. Misek, Heeju Noh, Luca Zanella, Moony Tseng, Hayley E. Francies, Priya Sridevi, Rachana Agarwal, Cindy W. Kyi, Julyann Perez-Mayoral, Megan J. Stine, Eva Tonsing-Carter, James M. Clinton, The HCMI Network, Peter W. Laird, Calvin J. Kuo, Olivier Elemento, David L. Spector, Andrew D. Cherniack, Kyle Elrott, Martin L. Ferguson, Rameen Beroukhim, Katherine A. Hoadley, Nicolas Robine, Andrew McPherson, Matthew J. Garnett, David A. Tuveson, Andrea Califano, Paul T. Spellman, Keith L. Ligon, Daniela S. Gerhard, Louis M. Staudt, Jesse Boehm. Single cell transcriptional dynamics in the HCMI cancer model collection [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 6358.

Advancing precision medicine research relies heavily on laboratory models that accurately reflect the rich intra- and inter-patient diversity of human tumors. While large-scale resources built on historical cell lines have provided valuable insights, these models often lack the molecular diversity needed for comprehensive translational research and may undergo genetic drift in culture, reducing their fidelity. Recent advances in technologies, such as organoid culture, have improved model derivation; however, systematic pan-cancer comparisons of genomic, transcriptomic, and epigenomic integrity following long-term ex vivo expansion remain limited. Here, we report on the Human Cancer Models Initiative (HCMI), an international program that has generated and systematically characterized 665 organoid, spheroid, and cell line models derived from over 2,500 consented donors. These models have undergone comprehensive whole-genome, exome, methylome, and transcriptome analyses. The HCMI collection includes 47 unique models representing 16 rare and histologically distinct cancer subtypes, including one of a kind disease models (e.g. desmoid tumor) and greatly broadening the diversity of available cancer models. Through integrative concordance analysis, we demonstrate that 96% of the 417 analyzed models closely mirror the molecular profiles of their parental tumors. Interesting rare exceptions to this concordance include subsets of models, where cell culture media conditions, tumor stroma, or population selection appear to influence cellular characteristics. We validated observed cellular state shifts using single-cell RNA sequencing, providing insights into the impact of ex vivo culture on model integrity. We further explored the translational utility of this resource by investigating treatment exposures, post-treatment mutational signatures, and extrachromosomal DNA (ecDNA) amplifications and their potential implications on treatment resistance within these models. This community resource offers a comprehensive roadmap for capturing a broader spectrum of cancer diversity in preclinical models, ultimately supporting and advancing therapeutic discovery. Citation Format: Dina ElHarouni, Mushriq Al-Jazrawe, Seongmin Choi, Merve Dede, Toshinori Hinoue, Sean A Misek, Heeju Noh, Luca Zanella, Moony Tseng, Hayley E Francies, Priya Sridevi, Rachana Agarwal, Cindy W Kyi, Julyann Perez-Mayoral, Megan J Stine, Eva Tonsing-Carter, James M Clinton, The HCMI Network, Peter W Laird, Calvin J Kuo, Olivier Elemento, David L Spector, Andrew D Cherniack, Kyle Ellrott, Martin L Ferguson, Rameen Beroukhim, Katherine A Hoadley, Nicolas Robine, Mathew Garnett, Andrea Califano, Paul T Spellman, David A Tuveson, Keith L Ligon, Daniela S Gerhard, Louis M Staudt, Jesse Boehm. Integrative clinical and molecular analysis of 665 next-generation in vitro cancer models generated by the the Human Cancer Models Initiative (HCMI) for advancing precision medicine and functional drug discovery [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Functional and Genomic Precision Medicine in Cancer: Different Perspectives, Common Goals; 2025 Mar 11-13; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2025;85(5 Suppl):Abstract nr B025.

Coupling genetic profiling with electronic health records from hospital biobanks is a foundational resource for precision medicine. However, lack of ancestral heterogeneity limits discovery and generalizability. We leveraged the UCLA ATLAS Community Health Initiative, a diverse biobank with >35% non-European participants in a single health system, to inform disease prevalence and genetic risk across five continental and 36 fine-scale ancestry groups. Analyzing clinical and genetic data for 93,937 individuals, 61,797 with whole-exome sequencing (WES), we identified novel associations between genetic variants and phenotypes, including STARD7 with asthma risk in Mexican Americans and FN3K with intestinal disaccharidase deficiency across Europeans and Admixed Americans. Top decile polygenic scores (PGS) predicted patient status for many common diseases (40% of patients with Type 1 diabetes); an effect markedly diminished in non-Europeans. Exploring the distribution of ACMG ClinGen rare variants across populations demonstrated European bias in curated clinical variants. Mitigating this bias using computationally predicted deleterious variants, we identified new gene-disease associations, including EXOC1L and blood glucose level in East Asians. We identified PTPRU as a modulator of semaglutide's effects on weight loss, and additionally found variability across ancestries and a relationship with type-2-diabetes PGS. We provide an interactive web portal for accessing cross-ancestry associations at atlas-phewas.mednet.ucla.edu. Collectively, our findings support the value of ancestral diversity in advancing precision health across a broad spectrum of populations.

The Human Cancer Models Initiative (HCMI) has developed 665 novel cancer models, including organoids and matched parental tumors, providing a significant addition to existing cancer model resources. To evaluate the transcriptional relatedness of HCMI models and tumors to the Cancer Cell Line Encyclopedia (CCLE) and The Cancer Genome Atlas (TCGA), we used the Celligner algorithm to align transcriptomic profiles across datasets, removing systematic biases while preserving intrinsic biological variability. HCMI models and tumors both exhibited high transcriptional fidelity to TCGA tumors, clustering closely with their respective tumor types in the Celligner-aligned space. Pairwise Euclidean distances showed complementary strengths between HCMI and CCLE models. For example, HCMI models demonstrated significantly closer alignment to TCGA tumors in glioblastoma, breast cancer, and ovarian cancer, while CCLE models performed comparably or better in colorectal cancer. Notably, aligning HCMI tumors to TCGA confirmed their strong transcriptional relatedness, validating the fidelity of these models to their original tumor states. Combining HCMI and CCLE datasets further enhanced the total transcriptional representation across the diversity of TCGA tumors, underscoring the complementary roles of these resources. The HCMI collection uniquely includes rare cancer types such as nephroblastoma, desmoid tumors, and ampulla of Vater carcinoma, which are absent in CCLE. Celligner analysis confirmed that these rare HCMI models faithfully retained transcriptional features of their corresponding TCGA tumors. This expands opportunities to study rare and clinically challenging cancers that have been underrepresented in preclinical models. These findings demonstrate the value of integrating HCMI, CCLE, and TCGA datasets. Together, they form a complementary and robust compendium for studying tumor biology, enabling improved cancer modeling and advancing precision oncology. Dina ElHarouni, Mushriq Al-Jazrawe, Seongmin Choi, Merve Dede, Toshinori Hinoue, Sean A. Misek, Heeju Noh, Luca Zanella, Moony Tseng, Hayley E. Francies, Priya Sridevi, Rachana Agarwal, Cindy W. Kyi, Julyann Perez-Mayoral, Megan J. Stine, Eva Tonsing-Carter, James M. Clinton, The HCMI Network, Peter W. Laird, Calvin J. Kuo, Olivier Elemento, David L. Spector, Andrew D. Cherniack, Kyle Ellrott, Martin L. Ferguson, Rameen Beroukhim, Katherine A. Hoadley, Nicolas Robine, Andrew McPherson, Mathew J. Garnett, David A. Tuveson, Andrea Califano, Paul T. Spellman, Keith L. Ligon, Daniela S. Gerhard, Louis M. Staudt, Jesse S. Boehm. Integrating HCMI models and tumors with CCLE and TCGA: Advancing cancer modeling and precision oncology [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 6608.

Pancreatic ductal adenocarcinomas (PDAC) are among the most fatal cancers, in part due to frequent detection at advanced stages. Endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA), the most sensitive diagnostic method of PDAC in current standard clinical practice, is invasive, costly, with access limited to major healthcare settings. Here, we present a non-invasive evaluation of plasma cell-free RNA (cfRNA) for PDAC detection in pre-diagnostic high-risk and de novo symptomatic patients presenting for EUS-FNA. We develop a cfRNA normalization method to account for preanalytical variation and handling effects and derive 29 potential cfRNA biomarkers for PDAC diagnosis using 153 samples collected prior to the EUS procedure. Biomarkers related to liver function are elevated in PDAC samples, including early-stage patients without liver metastasis. Classification of PDAC using these biomarkers is validated using an independent cohort of 95 samples. Our findings could help to improve diagnostic utility in high-risk and symptomatic individuals.