Abstract 3088: The efficient utilization of paracrine support from established cell lines for breast/ovarian cancer model generation

Ex vivo cancer cell models provide the starting material for in-depth mechanistic studies of cancer. However, the clinical/histopathologic, biomarker, and genetic heterogeneity of breast cancer has not been well represented in the current breast cancer cell model collection. While published breast cancer model generation protocols have been helpful, their high failure rates indicate the urgent need to improve model derivation efficiency. Here, the Broad Cancer Cell Line Factory (CCLF) project presents a novel model derivation technology to generate breast and ovarian cancer organoids with a high success rate by leveraging the paracrine support from historical cancer cell lines.

We observed that most established breast cancer cell lines can grow in a simple basal media with 10% fetal bovine serum; We hypothesized that historical cell lines may secrete vital growth factors that support breast cancer cells' survival and growth. To test this, we randomly selected a pool of 20 breast cancer cell lines, collected its conditioned media (CM20) and incorporated the CM20 as a supplement into our empirical rich media matrix (HYBRID, 16 mixed media conditions) with a Matrigel culturing system. Three-dimensional (3D) structures formed at Day 14-21 in the CM20 supplementary conditions compared to conditions without CM20 and only organoids with the CM20 supplement could be propagated to passage 5 and beyond. We performed pan-cancer targeted sequencing to evaluate tumor content of these organoids at passage 5 with paired tumor tissues. In our first 10 attempts, 95% of organoid cultures were genomically verified as high purity tumor models, indicating the CM20 is essential to enrich breast cancer cell growth in an in vitro culturing setting.

We applied the CM20 to ovarian cancers and observed a similar success rate suggesting a tissue-specific supporting manner. We tested conditioned media collected from other historical cancer cell lines but the breast/ovarian cancer organoid growth effect was not recapitulated. Importantly, when testing the individual breast cancer cell lines from the pool of 20, we discovered one cell line to be supporting the effect. More biochemistry work is needed to dissect the possible factors secreted by the line and molecular mechanisms of cancer cell survivors but preliminary data suggests the secretion factors are most likely proteins.

We generated 27 breast/ovarian cancer cell models using this technology and RNAseq data shows the breast cancer organoids still express their expected molecular subtype markers. 22 breast/ovarian cancer organoids have been propagated long-term with 17(out of 22) deposited to ATCC. Overall, this method provides an efficient model generation rate for female cancers. We anticipate that this method will not only allow us to quickly increase breast cancer cell model diversity but shed light on a new direction for breast cancer dependencies

Citation Format: Rebecca Deasy, Xin Jin, Pierre Michel Jean Beltran, Adel Atari, Madison Liistro, Cheryl Thompson, Stefanie Avril, Julie Boerner, Payal Pradhan, Samuel Klempner, Keith Ligon, William Sellers, Steven Carr, Todd Golub, Yuen-Yi (Moony) Tseng. The efficient utilization of paracrine support from established cell lines for breast/ovarian cancer model generation [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 3088.

Abstract 3453: Cancer Cell Line Factory: A systematic approach to create next-generation cancer model at scale

Precision cancer medicine is based on the ability to predict the dependencies of a given tumor from its molecular makeup. Despite successes in multiple common cancers, such prediction remains challenging for the majority of rare and understudied tumors given the absence of laboratory model systems in which to discover and/or validate therapeutic hypotheses. Here, we describe our efforts to address this challenge systematically with the ultimate goal of making it possible to learn how to predict ex vivo growth requirements for cancer samples based on technical, clinical and genomic properties of the starting tumor material. Over the last 5 years, we have processed nearly 2,000 tumor biospecimens and created over 375 genomically-confirmed patient-derived cell lines, organoids and neurosphere cultures, with >10% of these representing rare cancers. To make this possible, we have implemented three key workflows including (1) direct-to-patient sample sourcing, (2) a tissue cryopreservation and genomic credentialing system to ensure quality prior to model creation, and (3) a systematic empirical approach to screening rich medias and variations on organoid technologies ex vivo (HYBRID). We have begun performing genome-wide CRISPR viability screens in these cultures as part of our larger activities to generate a systematic laboratory-based functional map of cancer dependencies (a ‘Cancer Dependency Map'). The novel organoid, spheroid and cell line models created as part of this effort are being made publically available to the scientific community. Looking ahead, as the barriers to culturing rare tumors are overcome, we expect that preclinical functional genomics data will be useful for difficult-to-treat tumors without existing molecularly guided standard-of-care regimens.

Citation Format: Yuen-Yi Tseng, Mushriq AI-Jazrawe, Rebecca Deasy, Paula Keskula, Grace Johnson, Andrew Hong, Priya Chatterji, Francisca Vasquez, Adam Bass, Barbara Van Hare, David Sandak, Keith Ligon, Jesse Boehm. Cancer Cell Line Factory: A systematic approach to create next-generation cancer model at scale [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 3453.

Abstract PR09: A systematic approach to create patient-derived models of rare tumors

Precision cancer medicine is based on the ability to predict the dependencies of a given tumor from its molecular makeup. Despite successes in multiple common cancers, such prediction remains challenging for the majority of rare and understudied tumors, given the absence of laboratory model systems in which to discover and/or validate therapeutic hypotheses. Here, we describe our efforts to address this challenge systematically with the ultimate goal of making it possible to learn how to predict ex vivo growth requirements for cancer samples based on technical, clinical, and genomic properties of the starting tumor material. Over the last 5 years, we have processed nearly 2,000 tumor biospecimens and created over 375 genomically confirmed patient-derived cell lines, organoids, and neurosphere cultures, with >10% of these representing rare cancers. To make this possible, we have implemented three key workflows including (1) direct-to-patient sample sourcing, (2) a tissue cryopreservation and genomic credentialing system to ensure quality prior to model creation, and (3) a systematic empirical approach to screening rich media and variations on organoid technologies ex vivo (HYBRID). We have begun performing genome-wide CRISPR viability screens in these cultures as part of our larger activities to generate a systematic laboratory-based functional map of cancer dependencies (a “Cancer Dependency Map”). The novel organoid, spheroid, and cell line models created as part of this effort are being made publicly available to the scientific community. Looking ahead, as the barriers to culturing rare tumors are overcome, we expect that preclinical functional genomics data will be useful for difficult-to-treat tumors without existing molecularly guided standard-of-care regimens.

This abstract is also being presented as Poster A06.

Citation Format: Yuen-Yi Tseng, Paula Keskula, Rebecca Deasy, Andrew Hong, Priya Chatterji, Francisca Vazquez, Adam Bass, Barbara Van Hare, David Sandak, Keith Ligon, Jesse Boehm. A systematic approach to create patient-derived models of rare tumors [abstract]. In: Proceedings of the AACR Special Conference on the Evolving Landscape of Cancer Modeling; 2020 Mar 2-5; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2020;80(11 Suppl):Abstract nr PR09.

A GPX4-dependent cancer cell state underlies the clear-cell morphology and confers sensitivity to ferroptosis

Clear-cell carcinomas (CCCs) are a histological group of highly aggressive malignancies commonly originating in the kidney and ovary. CCCs are distinguished by aberrant lipid and glycogen accumulation and are refractory to a broad range of anti-cancer therapies. Here we identify an intrinsic vulnerability to ferroptosis associated with the unique metabolic state in CCCs. This vulnerability transcends lineage and genetic landscape, and can be exploited by inhibiting glutathione peroxidase 4 (GPX4) with small-molecules. Using CRISPR screening and lipidomic profiling, we identify the hypoxia-inducible factor (HIF) pathway as a driver of this vulnerability. In renal CCCs, HIF-2α selectively enriches polyunsaturated lipids, the rate-limiting substrates for lipid peroxidation, by activating the expression of hypoxia-inducible, lipid droplet-associated protein (HILPDA). Our study suggests targeting GPX4 as a therapeutic opportunity in CCCs, and highlights that therapeutic approaches can be identified on the basis of cell states manifested by morphological and metabolic features in hard-to-treat cancers.

Clear-cell carcinomas are aggressive tumours characterised by high accumulation of lipids and glycogen. Here, the authors report that these cancers have a common vulnerability to GPX4 inhibition-induced ferroptosis and using CRISPR screen and lipodomic profiling, they identify HIF-2α- HILPDA axis promotes ferroptosis via enrichment of PUFA lipids.

Renal medullary carcinomas depend upon SMARCB1 loss and are sensitive to proteasome inhibition

Renal medullary carcinoma (RMC) is a rare and deadly kidney cancer in patients of African descent with sickle cell trait. We have developed faithful patient-derived RMC models and using whole-genome sequencing, we identified loss-of-function intronic fusion events in one SMARCB1 allele with concurrent loss of the other allele. Biochemical and functional characterization of these models revealed that RMC requires the loss of SMARCB1 for survival. Through integration of RNAi and CRISPR-Cas9 loss-of-function genetic screens and a small-molecule screen, we found that the ubiquitin-proteasome system (UPS) was essential in RMC. Inhibition of the UPS caused a G2/M arrest due to constitutive accumulation of cyclin B1. These observations extend across cancers that harbor SMARCB1 loss, which also require expression of the E2 ubiquitin-conjugating enzyme, UBE2C. Our studies identify a synthetic lethal relationship between SMARCB1-deficient cancers and reliance on the UPS which provides the foundation for a mechanism-informed clinical trial with proteasome inhibitors.

Renal medullary carcinoma (RMC for short) is a rare type of kidney cancer that affects teenagers and young adults. These patients are usually of African descent and carry one of the two genetic changes that cause sickle cell anemia. RMC is an aggressive disease without effective treatments and patients survive, on average, for only six to eight months after their diagnosis.

Recent genetic studies found that most RMC cells have mutations that prevent them from producing a protein called SMARCB1. SMARCB1 normally acts as a so-called tumor suppressor, preventing cells from becoming cancerous. However, it was not clear whether RMCs always have to lose SMARCB1 if they are to survive and grow.

Often, diseases are studied using laboratory-grown cells and tissues that have certain features of the disease. No such models had been created for RMC, which has slowed efforts to understand how the disease develops and find new treatments for it. Hong et al. therefore worked with patients to develop new lines of cells that can be used to study RMC in the laboratory. These RMC cells started dying when they were given copies of the SMARCB1 gene, which supports the theory that RMCs have to lose SMARCB1 in order to grow.

Hong et al. then used a set of genetic reagents that can suppress or delete genes that are targeted by drugs, and followed this by testing a range of drugs on the RMC cells. Drugs and genetic reagents that reduced the activity of the proteasome – the structure inside cells that gets rid of old or unwanted proteins – caused the RMC cells to die. These proteasome inhibitor drugs also killed other kinds of cancer cells with SMARCB1 mutations.

Proteasome inhibitors are already used to treat different types of cancer. Potentially, a clinical trial could be run to see if they will treat patients whose cancers lack SMARCB1. Further work is also needed to determine the exact link between SMARCB1 and the proteasome.

Abstract A02: Expanding tumor chemical-genetic interaction map using next-generation cancer models

The development of new cancer therapeutics requires sufficient genetic and phenotypic diversity of cancer models. Current collections of human cancer cell lines are limited and for many rare cancer types, zero models exist that are broadly available. Here, we report results from the pilot phase of the Cancer Cell Line Factory (CCLF) project that aims to overcome this obstacle by systematically creating next-generation in vitro cancer models from adult and pediatric cancer patients' specimens and making these models broadly available.

We first developed a workflow of laboratory, genomics and informatics tools that make it possible to systematically compare published ex vivo culture conditions for each individual tumor to enable the scientific community to iterate towards disease-specific culture recipes. Based on sample volume and rarity, 4-100 conditions were applied to each sample and all data was captured in a custom Laboratory Information Management System to enhance subsequent predictions. We developed a $150, 5-day turnaround genomics panel to validate cultures based on genomics. Importantly, we show that tumor genomics can be retained in such patient-derived models and tumor genomics are generally stable across 20 passages. Since the inception of this project, we have processed over 600 patient cancer specimens from 450 patients across 16 tumor types and report the successful generation of over 100 genomically characterized adult and pediatric cancer and normal models.

We next hypothesized that novel patient-derived cell models could be used to enhance dependency predictions. To do so, we tested 72 cell lines against the informer set of 440 compounds developed by the Broad Cancer Target Discovery and Development (CTD2) Center. We show that generating cell lines and testing their sensitivities within 3 months is feasible and the high-throughput drug responses are reproducible. Moreover, to strengthen relationships between drug sensitivities and cellular features, we compared results with recently published data on the identical compounds tested against 860 existing cell lines. With this approach, we show that many chemical-genetic interaction vulnerabilities can be rapidly assessed. Importantly, adding more cancer models with the dimensions of quantity and diversity increases the predictive power of chemical-genetic interaction map. We are currently evaluating these drug sensitivity predictors for novel co-dependencies. Overall, our proof-of-concept framework demonstrates initial feasibility of rapidly generating cancer models at scale and expanding the chemical-genetic interaction map to identify new cancer vulnerability.

Citation Format: Yuen-Yi (Moony) Tseng, Andrew Hong, Shubhroz Gill, Paula Keskula, Srivatsan Raghavan, Jaime Cheah, Aviad Tsherniak, Francisca Vazquez, Sahar Alkhairy, Anson Peng, Abeer Sayeed, Rebecca Deasy, Peter Ronning, Philip Kantoff, Levi Garraway, Mark Rubin, Calvin Kuo, Sidharth Puram, Adi Gazdar, Nikhil Wagle, Adam Bass, Keith Ligon, Katherine Janeway, David Root, Stuart Schreiber, Paul Clemons, Todd Golub, William Hahn, Jesse Boehm. Expanding tumor chemical-genetic interaction map using next-generation cancer models [abstract]. In: Proceedings of the AACR Precision Medicine Series: Opportunities and Challenges of Exploiting Synthetic Lethality in Cancer; Jan 4-7, 2017; San Diego, CA. Philadelphia (PA): AACR; Mol Cancer Ther 2017;16(10 Suppl):Abstract nr A02.

Abstract 1953: Accelerating prediction of pediatric and rare cancer vulnerabilities using next-generation cancer models

Ongoing pre-clinical efforts aim to deploy genome-scale CRISPR/Cas9 technology and large collections of small molecules to catalog maps of cancer vulnerabilities at scale. However, such efforts in pediatric and rare cancers have lagged behind comparable efforts in more common cancer types due to the dearth of cell models. Here, we present an update from our “Cancer Cell Line Factory” project on efforts to overcome key laboratory and biologistics challenges precluding progress in pediatric and rare cancers. This effort, now in it’s 3rd year, represents an industry scale pipeline aiming to generate, characterize and share novel cancer models of many tumor types with the scientific community. Overall, we have processed 1153 samples from 818 patients across over 16 cancer types through this pipeline with a 28% success rate overall, including over 350 patient samples from rare and pediatric cancers. To optimize conditions for each tumor type, we have systematically compared published methods including (1) next-generation 2-dimension, (2) organoid and (3) standard approaches and have captured all information with a data management system that should enhance the ability to predict optimal ex vivo propagation conditions for future samples. Among the successful cell models verified already as part of this effort, we have generated a series of over 30 unique pediatric and rare cancer models, many of which represent the first of their kind. We screened these and other models against a library of highly annotated 440 small molecules that were previously tested against 860 existing cancer cell lines. Our results suggest that dependency data generated with novel next-generation cell cultures is potentially backwards-compatible with existing small molecule dependency datasets. Furthermore, we tested the novel Broad Institute Drug Repurposing library consisting of 4100 approved therapeutics, or those under investigation for any disease, against the first cell line models of several of these rare next generation models including angioimmunoblastic T-cell lymphoma and renal medullary carcinoma, leading to several novel drug repurposing hypotheses for rare cancers. Given these proof-of-concept studies, in partnership with the Rare Cancer Research Foundation, we launched an online matchmaking platform to connect patients with rare cancers to available research studies, facilitate online consent and provide biologistics support to enable fresh tissue donation to support cancer model generation from any clinical site in the United States. We will present results from this novel direct-to-patient approach to facilitate the generation of even larger numbers of next generation models from rare and pediatric cancers, propelling the generation of pre-clinical dependency maps of these tumors for the scientific community.

Citation Format: Yuen-Yi Tseng, Andrew Hong, Paula Keskula, Shubhroz Gill, Jaime Cheah, Grigoriy Kryukov, Aviad Tsherniak, Francisca Vazquez, Glenn Cowley, Sahar Alkhairy, Coyin Oh, Anson Peng, Rebecca Deasy, Abeer Sayeed, Peter Ronning, Samuel Ng, Steven Corsello, Corrie Painter, David Sandak, Levi Garraway, Mark Rubin, Calvin Kuo, Sidharth Puram, David Weinstock, Adam Bass, Nikhil Wagle, Keith Ligon, Katherine Janeway, David Root, Stuart Schreiber, Paul Clemons, Aly Shamji, Aly Shamji, William Hahn, Todd Golub, Jesse Boehm. Accelerating prediction of pediatric and rare cancer vulnerabilities using next-generation cancer models [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 1953. doi:10.1158/1538-7445.AM2017-1953

Abstract B26: Accelerating prediction of tumor vulnerabilities using next-generation cancer models

The development of new cancer therapeutics requires sufficient genetic and phenotypic diversity of cancer models. Current collections of human cancer cell lines are limited and for many rare cancer types, zero models exist that are broadly available. Here, we report results from the pilot phase of the Cancer Cell Line Factory (CCLF) project that aims to overcome this obstacle by systematically creating next-generation in vitro cancer models from adult and pediatric cancer patients' specimens and making these models broadly available.

We first developed a workflow of laboratory, genomics and informatics tools that make it possible to systematically compare published ex vivo culture conditions for each individual tumor to enable the scientific community to iterate towards disease-specific culture recipes. Based on sample volume and rarity, 4-100 conditions were applied to each sample and all data was captured in a custom Laboratory Information Management System to enhance subsequent predictions. We developed a $150, 5-day turnaround genomics panel to validate cultures based on genomics. Importantly, we show that tumor genomics can be retained in such patient-derived models and tumor genomics are generally stable across 20 passages. Since the inception of this project, we have processed over 650 patient cancer specimens from 450 patients across 16 tumor types and report the successful generation of over 100 genomically characterized adult and pediatric cancer and normal models.

We next hypothesized that novel patient-derived cultures could be used to enhance dependency predictions. To do so, we tested 65 cell lines against the “informer” set of 440 compounds developed by the Broad Cancer Target Discovery and Development (CTD2) Center. We show that generating cell lines and testing their sensitivities within 3 months is feasible and the drug responses are reproducible. Moreover, to strengthen relationships between drug sensitivities and cellular features, we compared results with recently published data on the identical compounds tested against 860 existing cell lines. With this approach, we are able to identify many known drug dependencies in these novel models and exhibit the consistency sensitivities compared to existing cell lines. We are also evaluating drug sensitivity predictors for novel dependencies. Overall, our proof-of-concept framework demonstrates initial feasibility of rapidly generating cancer models and assessing drug sensitivities at scale.

Citation Format: Yuen-Yi Tseng, Paula Keskula, Andrew L. Hong, Shubhroz Gill, Jaime H. Cheah, Gregory V. Kryukov, Aviad Tsherniak, Francisca Vazquez, Glenn Cowley, Coyin Oh, Anson Peng, Abeer Sayeed, Rebecca Deasy, Peter Ronning, Philip Kantoff, Levi Garraway, Mark A. Rubin, Calvin Kuo, Sidharth Puram, Adi Gazdar, Filemon S. Dela Cruz, Jr., Adam Bass, Jr., Nikhil Wagle, Keith L. Ligon, Katherine Janeway, David Root, Stuart L. Schreiber, Paul A. Clemons, Aly Shamji, William C. Hahn, Todd R. Golub, Jesse S. Boehm. Accelerating prediction of tumor vulnerabilities using next-generation cancer models. [abstract]. In: Proceedings of the AACR Special Conference: Patient-Derived Cancer Models: Present and Future Applications from Basic Science to the Clinic; Feb 11-14, 2016; New Orleans, LA. Philadelphia (PA): AACR; Clin Cancer Res 2016;22(16_Suppl):Abstract nr B26.

Abstract 4367: Accelerating prediction of tumor vulnerabilities using next-generation cancer models

The mapping of cancer genomes is rapidly approaching completion. The genomic information encoded by individual patients’ tumors should, in principle, provide a guide for predicting dependencies, but our ability to do so is suboptimal. The challenge stems from the absence of clinical data relating genotypes with dependencies since most cancer mutations are rare and our arsenal of cancer drugs is incomplete. If it was possible to build a preclinical ‘cancer dependency map’ at a scale that captured the genomic diversity of cancer (for instance, models of all genotypes tested for genetic and small-molecule dependencies), it should be feasible to improve dependency predictions. New technologies (e.g. CRISPR/Cas9 libraries) make such an effort now feasible. However, we lack a sufficient diversity of cancer models derived directly from patient samples to reflect the genetic diversity of cancer and the ability to systematically create functional data for each cancer patient to expand the map.

In an attempt to overcome these obstacles, we have established an industry-scale pipeline to generate new cancer models directly from patient samples, a “Cancer Cell Line Factory”. We have processed over 620 samples from 400 patients across 16 cancer types through this pipeline with a 25% success rate overall. To optimize conditions for each tumor type, we have systematically compared published cell line generation methods with standard approaches and captured all information with a data management system that will enhance the ability to predict optimal ex vivo propagation conditions for future samples. In all, we report the successful derivation of over 100 new genomically confirmed cancer and normal cell lines, including a series of unique pediatric cancer models derived from rare tumors.

We hypothesized that novel patient-derived cultures could be used to enhance dependency predictions. To test this hypothesis, we tested dependencies of 65 of these novel cultures against an identical set of 440 small molecules that were previously tested against 860 existing cancer cell lines. Our results suggest that dependency data generated with novel cell cultures is potentially backwards-compatible with existing small molecule dependency datasets. Finally, we demonstrate proof-of-concept that such new models can successfully used in CRISPR-Cas9 screens and integrate results with small molecule sensitivities to uncover CDK4 and XPO1 dependencies in a rare pediatric undifferentiated sarcoma. In aggregate, these proof-of-concept studies demarcate a path by which pre-clinical dependency maps may enhance clinical dependency predictions from genomic data alone.

Citation Format: Yuen-Yi Tseng, Andrew Hong, Paula Keskula, Shubhroz Gill, Jaime Cheah, Grigoriy Kryukov, Aviad Tsherniak, Francisca Vazquez, Glenn Cowley, Coyin Oh, Anson Peng, Abeer Sayeed, Rebecca Deasy, Peter Ronning, Philip Kantoff, Levi Garraway, Mark Rubin, Calvin Kuo, Sidharth Puram, Adi Gazdar, Filemon Dela Cruz, Adam Bass, Nikhil Wagle, Keith Ligon, Katherine Janeway, David Root, Stuart Schreiber, Paul Clemons, Aly Shamji, William Hahn, Todd Golub, Jesse S. Boehm. Accelerating prediction of tumor vulnerabilities using next-generation cancer models. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 4367.