Search

Supplementary Table 2

Supplementary Methods

Supplementary Table 3

Supplementary Figures and Legends

Supplementary Table 5

Supplementary Table 4

Supplementary Table 1

Basal cells are multipotent stem cells of a variety of organs, including the respiratory tract, where they are major components of the airway epithelium. However, it remains unclear how diverse basal cells are and how distinct subpopulations respond to airway challenges. Using single cell RNA-sequencing and functional approaches, we report a significant and previously underappreciated degree of heterogeneity in the basal cell pool, leading to identification of six subpopulations in the adult murine trachea. Among these, we found two major subpopulations, collectively comprising the most uncommitted of all the pools, but with distinct gene expression signatures. Notably, these occupy distinct ventral and dorsal tracheal niches and differ in their ability to self-renew and initiate a program of differentiation in response to environmental perturbations in primary cultures and in mouse injury models in vivo. We found that such heterogeneity is acquired prenatally, when the basal cell pool and local niches are still being established, and depends on the integrity of these niches, as supported by the altered basal cell phenotype of tracheal cartilage-deficient mouse mutants. Finally, we show that features that distinguish these progenitor subpopulations in murine airways are conserved in humans. Together, the data provide novel insights into the origin and impact of basal cell heterogeneity on the establishment of regionally distinct responses of the airway epithelium during injury-repair and in disease conditions.

Type 2 diabetes (T2D) is associated with defective insulin secretion and reduced β cell mass. Available treatments provide a temporary reprieve, but secondary failure rates are high, making insulin supplementation necessary. Reversibility of β cell failure is a key translational question. Here, we reverse engineered and interrogated pancreatic islet–specific regulatory networks to discover T2D-specific subpopulations characterized by metabolic inflexibility and endocrine progenitor/stem cell features. Single-cell gain- and loss-of-function and glucose-induced Ca(2+) flux analyses of top candidate master regulatory (MR) proteins in islet cells validated transcription factor BACH2 and associated epigenetic effectors as key drivers of T2D cell states. BACH2 knockout in T2D islets reversed cellular features of the disease, restoring a nondiabetic phenotype. BACH2-immunoreactive islet cells increased approximately 4-fold in diabetic patients, confirming the algorithmic prediction of clinically relevant subpopulations. Treatment with a BACH inhibitor lowered glycemia and increased plasma insulin levels in diabetic mice, and restored insulin secretion in diabetic mice and human islets. The findings suggest that T2D-specific populations of failing β cells can be reversed and indicate pathways for pharmacological intervention, including via BACH2 inhibition.

In-silico modeling of patient clinical drug response (CDR) promises to revolutionize personalized cancer treatment. State-of-the-art CDR predictions are usually based on cancer cell line drug perturbation profiles. However, prediction performance is limited due to the inherent differences between cancer cell lines and primary tumors. In addition, current computational models generally do not leverage both chemical information of a drug and a gene expression profile of a patient during training, which could boost prediction performance. Here we develop a Patient Adapted with Chemical Embedding (PACE) dual convergence deep learning framework that a) integrates gene expression along with drug chemical structures, and b) is adapted in an unsupervised fashion by primary tumor gene expression. We show that PACE achieves better discrimination between sensitive and resistant patients compared to the state-of-the-art linear regularized method (9/12 VS 3/12 drugs with available clinical outcomes) and alternative methods.

Although breast cancer Stem-Like Tumor Initiating Cells (SLTIC) represent only a minute fraction of the total tumor mass, they are resistant to standard of care treatment and play a key role in tumor initiation, maintenance, and progression. Unequivocal SLTIC isolation, using surface markers, has proven highly elusive, thus impeding characterization and targeting of their mechanistic dependencies. To address this challenge, we applied a systems biology approach to effectively characterize SLTIC biology and to prioritize drugs that can reprogram them to a more differentiated state that is sensitive to chemotherapy. To isolate breast cancer cells enriched for SLTICs, we performed flow cytometry-based sorting of tumor cells from 7 metastatic breast cancer patients, based on the expression of Epcam and CD49F, which are established epithelial and SLTIC-enriched cell markers, respectively. Activity-based clustering of single cell RNASeq profiles using the VIPER (Alvarez et al. Nat Genet 2017) algorithm identified two cell states, comprising cells presenting high activity of either SLTIC (i.e., BMI1, NOTCH1, etc.) or differentiated, proliferative epithelial cell markers (i.e., PCNA, CCNB1, etc.). Analysis of RNASeq profiles of BT20 cells treated with ~400 FDA approved and late-stage experimental drugs identified albendazole as the drug inducing the most significant activity inversion of SLTIC Master Regulator proteins (Alvarez et al. Nat Genet 2018) (p=4.21x10-5). This was experimentally confirmed in vivo by single cell analysis of metastatic TNBC PDX models at 14 days after treatment with albendazole, paclitaxel (a drug known to kill differentiated but not SLTIC cells), and vehicle control. As expected, paclitaxel induced dramatic decrease of the differentiated vs. SLTIC cell ratio, while albendazole had the opposite effect, inducing equally dramatic increase in that ratio, as assessed by a combination of CytoTRACE and established SLTIC marker analysis. Sequential therapy, based on a 30-day treatment with albendazole with 3 rounds of paclitaxel at day 15, 22, and 30, repeated after a 15-day drug holiday produced highly synergistic tumor volume reduction, compared to individual monotherapies (p=0.00869 by Bliss independence analysis). Indeed, while albendazole had little effect compared to vehicle control, as monotherapy, it induced >50% additional tumor viability reduction when combined with paclitaxel. The approach is highly generalizable and can be used to identify potential combination therapy approaches for any tumor in which molecular distinct subpopulations co-exist with different drug sensitivities, thus offering a practical solution to addressing tumor heterogeneity. Citation Format: Jeremy Worley, Hongxu Ding, Heeju Noh, Evan Paull, Aaron T. Griffin, Adina Grunn, Daoqi You, Kristina Guillan, Erin Bush, Piero Dalerba, Peter Sims, Filemon S. Dela Cruz, Andrew L. Kung, Andrea Califano. Elucidation and pharmacological targeting of master regulator proteins representing mechanistic determinants of breast cancer stem-like tumor initiating cell state [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 3165.

Deep learning architectures such as variational autoencoders have revolutionized the analysis of transcriptomics data. However, the latent space of these variational autoencoders offers little to no interpretability. To provide further biological insights, we introduce a novel sparse Variational Autoencoder architecture, VEGA (Vae Enhanced by Gene Annotations), whose decoder wiring is inspired by a priori characterized biological abstractions, providing direct interpretability to the latent variables. We demonstrate the interpretability and flexibility of VEGA in diverse biological contexts, by integrating various sources of biological abstractions such as pathways, gene regulatory networks and cell type identities in the latent space of our model. We show that our model could recapitulate the mechanism of cellular-specific response to treatments, the status of master regulators as well as jointly investigate the cell type and cellular state identity in developing cells. We envision the approach could serve as an explanatory biological model in contexts such as development and drug treatment experiments.

Motivation Nucleotide modification status can be decoded from the Oxford Nanopore Technologies nanopore-sequencing ionic current signals. Although various algorithms have been developed for nanopore-sequencing-based modification analysis, more detailed characterizations, such as modification numbers, corresponding signal levels and proportions are still lacking. Results We present a framework for the unsupervised determination of the number of nucleotide modifications from nanopore-sequencing readouts. We demonstrate the approach can effectively recapitulate the number of modifications, the corresponding ionic current signal levels, as well as mixing proportions under both DNA and RNA contexts. We further show, by integrating information from multiple detected modification regions, that the modification status of DNA and RNA molecules can be inferred. This method forms a key step of de novo characterization of nucleotide modifications, shedding light on the interpretation of various biological questions. Availability and implementation Modified nanopolish: https://github.com/adbailey4/nanopolish/tree/cigar_output. All other codes used to reproduce the results: https://github.com/hd2326/ModificationNumber. Supplementary information Supplementary data are available at Bioinformatics online.

Type 2 Diabetes is associated with defective insulin secretion, reduced β-cell mass, and increased glucagon production. Cell lineage-tracing in rodents and human autopsy surveys support the notion of β-cell dedifferentiation as a unifying mechanism for these abnormalities. Yet, mechanistic determinants of human β-cell failure remain elusive. Using regulatory-network-based single-cell analysis of human islets, we identify aberrant, diabetes-enriched transitional states characterized by metabolic inflexibility, α/β-transition, and endocrine progenitor/stem cell features. A coordinated transcription factor hierarchy mediating cell state transition emerged and was validated using barcoded guide-based, single-cell gene transfer and calcium flux measures in primary human islet cells. Specifically, two master regulators and associated epigenetic drivers emerged, one (AFF3) controlling β- to α-like-cell reprogramming, the other (BACH2) transition to a dedifferentiated endocrine progenitor-like cell. The findings provide mechanistic insight into diabetic islet cell dysfunction and suggest actionable pathways for pharmacological intervention.

Glioblastoma heterogeneity and plasticity remain controversial, with proposed subtypes representing the average of highly heterogeneous admixtures of independent transcriptional states. Single-cell, protein-activity-based analysis allowed full quantification of >6,000 regulatory and signaling proteins, thus providing a previously unattainable single-cell characterization level. This helped identify four novel, molecularly distinct subtypes that successfully harmonize across multiple GBM datasets, including previously published bulk and single-cell profiles and single cell profiles from seven orthotopic PDX models, representative of prior subtype diversity. GBM is thus characterized by the plastic coexistence of single cells in two mutually-exclusive developmental lineages, with additional stratification provided by their proliferative potential. Consistently, all previous subtypes could be recapitulated by single-cell mixtures drawn from newly identified states. Critically, drug sensitivity was predicted and validated as highly state-dependent, both in single-cell assays from patient-derived explants and in PDX models, suggesting that successful treatment requires combinations of multiple drugs targeting these distinct tumor states.SignificanceWe propose a new, 4-subtype GBM classification, which harmonizes across bulk and single-cell datasets. Single-cell mixtures from these subtypes effectively recapitulate all prior classifications, suggesting that the latter are a byproduct of GBM heterogeneity. Finally, we predict single-cell level activity of three clinically-relevant drugs, and validate them in patient-derived explant.

INTRODUCTION AND OBJECTIVES:Recent development of drugs that block the programmed death-1 and cytotoxic T-lymphocyte antigen-4 interaction within the tumor microenvironment has led to improvements ...

Despite the important role of the PI3K/AKT/mTOR axis in the pathogenesis of cancer, to date there have been few functional oncogenic fusions identified involving the AKT genes. A 12-year-old female with a histopathologically indeterminate epithelioid neoplasm was found to harbor a novel fusion between the LAMTOR1 and AKT1 genes. Through expanded use access, she became the first pediatric patient to be treated with the oral ATP-competitive pan-AKT inhibitor ipatasertib. Treatment resulted in dramatic tumor regression, demonstrating through patient-driven discovery that the fusion resulted in activation of AKT1, was an oncogenic driver, and could be therapeutically targeted with clinical benefit. Post-clinical validation using patient-derived model systems corroborated these findings, confirmed a membrane-bound and constitutively active fusion protein, and identified potential mechanisms of resistance to single-agent treatment with ipatasertib. Significance: This study describes the patient-driven discovery of the first AKT1 fusion–driven cancer and its treatment with the AKT inhibitor ipatasertib. Patient-derived in vitro and in vivo model systems are used to confirm the LAMTOR1–AKT1 fusion as a tumorigenic driver and identify potential mechanisms of resistance to AKT inhibition. This article is highlighted in the In This Issue feature, p. 565

The maintenance and transition of cellular states are controlled by biological processes. Here we present a gene set-based transformation of single cell RNA-Seq data into biological process activities that provides a robust description of cellular states. Moreover, as these activities represent species-independent descriptors, they facilitate the alignment of single cell states across different organisms., Single cell RNA-Seq data can report on cellular types and states, but low signal-to noise and sparse data can make interpretation of cellular state difficult. Here the authors propose a transformation strategy to map RNA-Seq data to biological process activities that are species-agnostic and allow for comparison across species.

Breast cancer stem-like cells (CSCs) are hypothesized to be a cause of poor outcome in patients, and are drug-resistant in vitro and in PDX mice. We performed plate-based single-cell RNA sequencing (scRNA-seq) on flow sorted fresh tumor samples. These data were analyzed using our metaVIPER algorithm, which infers protein activity in a manner that is robust to the effects of noise and gene dropout. metaVIPER identified known breast CSC regulators, and analysis with our OncoTreat algorithm predicted several FDA approved drugs that could reprogram the breast CSCs into a state resembling that of differentiated tumor cells. We tested two of these predictions, ivermectin and albendazole, in TNBC PDX mice, and performed scRNA-seq on fresh tumor samples after 14 days of treatment. The results were exciting, as measured by single-cell entropy or lineage marker analysis. Breast CSCs were reprogrammed as predicted, while a tumor treated with a traditional chemotherapeutic (paclitaxel) had a dramatically higher proportion of CSCs. These results suggested a therapy in which cells are cyclically reprogrammed and treated with a conventional cytotoxic drug. This reprogram-kill-repeat therapy is being tested in TNBC PDX mice. We believe that the strategy developed here will prove to be a powerful tool for understanding and manipulating cell-state in heterogeneous populations. Citation Format: Jeremy Worley, Hongxu Ding, Heeju Noh, Evan Paull, Erin Bush, Piero Dalerba, Peter Sims, Fil Dela Cruz, Andrew Kung, Andrea Califano. A systems biology approach to reprogramming drug-resistant breast cancer stem-like cells [abstract]. In: Proceedings of the AACR Virtual Special Conference on Tumor Heterogeneity: From Single Cells to Clinical Impact; 2020 Sep 17-18. Philadelphia (PA): AACR; Cancer Res 2020;80(21 Suppl):Abstract nr PR07.

Since 2012, numerous single-cell RNA sequencing (scRNA-seq) technologies have been introduced. Most involve either significant startup cost, an expert operator, or expensive library preparation. We developed a plate-based library prep that incorporates pre-amplification pooling, unique molecular identifiers (UMIs), and mRNA enrichment using silanol coated magnetic beads. Our method, which builds off of Bush et al. (Nature Communications, 2017) and Picelli et al. (Nature Protocols, 2014), uses readily available equipment, is easy to automate,and costs a fraction ofcommercial kits. In addition, our method has been optimized to reduce barcode cross contamination that can occur in pooled pre-amplification reactions. This method allows for plate-based 3’ scRNA-seq with UMIs at relatively low cost, using standard lab equipment. NOTE (2018/08/29): we are currently testing optimizationsdescribedby Bagnoli et al. (Nature Communications, 2018). This protocol may be updated accordingly.

Myeloid-biased hematopoietic stem cells (MB-HSCs) play critical roles in recovery from injury, but little is known about how they are regulated within the bone marrow niche. Here we describe an auto-/paracrine physiologic circuit that controls quiescence of MB-HSCs and hematopoietic progenitors marked by histidine decarboxylase (Hdc). Committed Hdc+ myeloid cells lie in close anatomical proximity to MB-HSCs and produce histamine, which activates the H2 receptor on MB-HSCs to promote their quiescence and self-renewal. Depleting histamine-producing cells enforces cell cycle entry, induces loss of serial transplant capacity, and sensitizes animals to chemotherapeutic injury. Increasing demand for myeloid cells via lipopolysaccharide (LPS) treatment specifically recruits MB-HSCs and progenitors into the cell cycle; cycling MB-HSCs fail to revert into quiescence in the absence of histamine feedback, leading to their depletion, while an H2 agonist protects MB-HSCs from depletion after sepsis. Thus, histamine couples lineage-specific physiological demands to intrinsically primed MB-HSCs to enforce homeostasis. Chen et al. show that histidine decarboxylase (Hdc) marks quiescent myeloid-biased HSCs (MB-HSCs). Daughter myeloid cells form a spatial cluster with Hdc+ MB-HSCs and secrete histamine to enforce their quiescence and protect them from depletion, following activation by a variety of physiologic insults.

Motivation In a scenario where populations A, B1 and B2 (subpopulations of B) exist, pronounced differences between A and B may mask subtle differences between B1 and B2. Results Here we present iterClust, an iterative clustering framework, which can separate more pronounced differences (e.g. A and B) in starting iterations, followed by relatively subtle differences (e.g. B1 and B2), providing a comprehensive clustering trajectory. Availability and implementation iterClust is implemented as a Bioconductor R package. Supplementary information Supplementary data are available at Bioinformatics online.

Immune cells in the tumor microenvironment (TME) play a critical role in regulating tumor development and response to targeted immunotherapy. While it is well-accepted that clear cell renal carcinoma (ccRCC) is characterized by a T-cell infiltrate, differential activation states within the tumor and the contribution of other immunomodulatory cells remain relatively unexplored. We sequenced approximately 30,000 immune cells at the single-cell level from tumor and matched adjacent normal tissue in ccRCC patients. With a median of 3300 Unique Molecular Identifiers (UMIs)/cell, we were able to capture the expression of 1000 genes per cell on average and 21,000 genes across all cells from each patient. We used Multidimensional Scaling (MDS) -transformation to visualize the various clusters and found that the tumor and adjacent normal tissue were distinct, with each having two major sub-clusters - lymphoid and myeloid cells. The data demonstrated a higher fraction of lymphoid infiltrate(~70-75%) than myeloid in a relatively stable ratio across patients, with tumor containing a slightly higher fraction of lymphoid cells. The normal and tumor cell subtypes were distinguished by multiple iterations of PAM clustering (Partitioning Around Mediods) with feature selection using the iterClust algorithm. Top differentially expressed genes in each cluster were identified and cross-referenced with known immune cell markers along with pathway enrichment analysis for cell typing. Among the tumor lymphoid population, genes specific for T-cells, including markers of functional exhaustion were the most highly expressed. A separate sub-cluster of cells consistent with cytotoxic T and/or NK cells was also identified. We found a distinct population of monocytes within the myeloid cluster that was present exclusively in the tumor tissue. This population expressed high levels of complement factors C1QA, C1QB, C1QC, as well as ApoE. The latter has been implicated as a driver of myeloid cell immunosuppression, indicating a possible regulatory role in the TME of ccRCC that has not previously been described. The phenotype of the immune cell populations in tumor was further validated by spectral flow cytometry. This study demonstrates relative abundance of cell types in the immune microenvironment within ccRCC tumors and also identifies a novel C1Q-high myeloid sub-population specific to the tumor, suggesting its possible role in RCC tumorigenesis or progression. Citation Format: Nivedita Chowdhury, Aleksandar Obradovic, David Aggen, Vinson Wang, Xinzheng Guo, Guarionex J. DeCastro, Hongxu Ding, Andrea Califano, Charles G. Drake. Decoding the immunologic landscape of human clear cell renal carcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1476.

Introduction: Gliomas are highly infiltrative tumors rendering complete resection impossible. As such, prognosis remains dismal. Recent studies have identified both proneural and mesenchymal cell populations. Crucially, recurring gliomas bear a mesenchymal phenotype. Understanding mechanisms driving mesenchymal transformation is important in developing tailored, more efficient therapeutic approaches. Methods: Using targeted biopsies from a mouse glioma model and human glioma databases, we applied bioinformatics and identified Notch as an additional master regulator (MR) protein driving mesenchymal transformation. To better understand the role of Notch, we overexpressed active forms of Notch1 (N1IC) or Notch3 (N3IC) in primary cells, orthotopic xenotransplants, and transgenic mouse models. We performed mouse survival studies and examined Notch-induced phenotypic alterations in tumor cells and the microenvironment using molecular and immunohistochemical approaches. Finally, we performed high-throughput drug and radiation screening studies to assess responses to treatment. Results: In silico analyses on RNAseq of biopsies from spatially distinct tumor regions in a mouse glioma model identified Notch as a highly active pathway in the cores of tumors undergoing mesenchymal transformation. Importantly, MR analysis of TCGA samples and single cell transcriptomic data identified the Notch pathway as a driver of proneural to mesenchymal transformation. Immunophenotyping and RNAseq profiling of human GBMs identified Notch receptors on tumor cells and Notch ligands on astrocytes and microglia suggesting potential microenvironment-mediated plasticity since Notch proteins are not frequently mutated in GBM. Overexpression of N1IC or N3IC in proneural cells induced mesenchymal transformation with slow proliferation and a highly invasive phenotype. Orthotransplantation of N3IC-expressing cells in mice led to slower growing tumors, perivascular invasion, increased recruitment of reactive phagocytes and slightly prolonged survival. A novel transgenic mouse model in which N1IC was overexpressed also led to prolonged survival compared to a control population. Importantly, high-throughput screening studies showed that N1IC and N3IC rendered tumor cells significantly more resistant to standard and targeted therapies, including irradiation and etoposide. Conclusions: Our study identifies Notch signaling as an important, likely microenvironment-driven pathway driving proneural to mesenchymal transformation in glioma. While Notch was shown to slow tumor growth and prolong survival, cells were more invasive and less sensitive to treatment. These findings suggest that Notch plays a role in the mesenchymal transformation and therapeutic resistance seen in post-treatment recurrence. Targeting Notch may therefore be important in improving therapeutic responses against glioma. Citation Format: Matei A. Banu, Athanassios Dovas, Henar Cuervo, Brianna Pereira, Aayushi Mahajan, Hongxu Ding, Mukesh Bansal, Sergey Pampou, Charles Karan, Andrea Califano, Peter A. Sims, Jan Kitajewski, Jeffrey N. Bruce, Peter Canoll. Notch is a master regulator of mesenchymal transformation and therapeutic resistance in glioma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 5103.

Advances in three-dimensional (3D) printing have enabled the direct assembly of cells and extracellular matrix materials to form in vitro cellular models for 3D biology, the study of disease pathogenesis and new drug discovery. In this study, we report a method of 3D printing for Hela cells and gelatin/alginate/fibrinogen hydrogels to construct in vitro cervical tumor models. Cell proliferation, matrix metalloproteinase (MMP) protein expression and chemoresistance were measured in the printed 3D cervical tumor models and compared with conventional 2D planar culture models. Over 90% cell viability was observed using the defined printing process. Comparisons of 3D and 2D results revealed that Hela cells showed a higher proliferation rate in the printed 3D environment and tended to form cellular spheroids, but formed monolayer cell sheets in 2D culture. Hela cells in 3D printed models also showed higher MMP protein expression and higher chemoresistance than those in 2D culture. These new biological characteristics from the printed 3D tumor models in vitro as well as the novel 3D cell printing technology may help the evolution of 3D cancer study.