Identification of causal genetic drivers of human disease through systems-level analysis of regulatory networks.

Identification of driver mutations in human diseases is often limited by cohort size and availability of appropriate statistical models. We propose a framework for the systematic discovery of genetic alterations that are causal determinants of disease, by prioritizing genes upstream of functional disease drivers, within regulatory networks inferred de novo from experimental data. We tested this framework by identifying the genetic determinants of the mesenchymal subtype of glioblastoma. Our analysis uncovered KLHL9 deletions as upstream activators of two previously established master regulators of the subtype, C/EBPß and C/EBPδ. Rescue of KLHL9 expression induced proteasomal degradation of C/EBP proteins, abrogated the mesenchymal signature, and reduced tumor viability in vitro and in vivo. Deletions of KLHL9 were confirmed in > 50% of mesenchymal cases in an independent cohort, thus representing the most frequent genetic determinant of the subtype. The method generalized to study other human diseases, including breast cancer and Alzheimer's disease.

ECMarker: interpretable machine learning model identifies gene expression biomarkers predicting clinical outcomes and reveals molecular mechanisms of human disease in early stages.

MOTIVATION: Gene expression and regulation, a key molecular mechanism driving human disease development, remains elusive, especially at early stages. Integrating the increasing amount of population-level genomic data and understanding gene regulatory mechanisms in disease development are still challenging. Machine learning has emerged to solve this, but many machine learning methods were typically limited to building an accurate prediction model as a 'black box', barely providing biological and clinical interpretability from the box. RESULTS: To address these challenges, we developed an interpretable and scalable machine learning model, ECMarker, to predict gene expression biomarkers for disease phenotypes and simultaneously reveal underlying regulatory mechanisms. Particularly, ECMarker is built on the integration of semi- and discriminative-restricted Boltzmann machines, a neural network model for classification allowing lateral connections at the input gene layer. This interpretable model is scalable without needing any prior feature selection and enables directly modeling and prioritizing genes and revealing potential gene networks (from lateral connections) for the phenotypes. With application to the gene expression data of non-small-cell lung cancer patients, we found that ECMarker not only achieved a relatively high accuracy for predicting cancer stages but also identified the biomarker genes and gene networks implying the regulatory mechanisms in the lung cancer development. In addition, ECMarker demonstrates clinical interpretability as its prioritized biomarker genes can predict survival rates of early lung cancer patients (P-value < 0.005). Finally, we identified a number of drugs currently in clinical use for late stages or other cancers with effects on these early lung cancer biomarkers, suggesting potential novel candidates on early cancer medicine. AVAILABILITYAND IMPLEMENTATION: ECMarker is open source as a general-purpose tool at https://github.com/daifengwanglab/ECMarker. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A luminal intermediate cell state maintains long-term prostate homeostasis and contributes to tumorigenesis.

Cellular heterogeneity poses tremendous challenges for developing cell-targeted therapies and biomarkers of clinically significant prostate cancer. The origins of this heterogeneity within normal adult and aging tissue remain unknown, leaving cellular states and transcriptional programs that allow expansions of malignant clones unidentified. To define cell states that contribute to early cancer development, we performed clonal analyses and single cell transcriptomics of normal prostate from genetically-engineered mouse models. We uncovered a luminal transcriptional state with a unique "basal-like" Wnt/p63 signaling ( luminal intermediate , LumI) which contributes to the maintenance of long-term prostate homeostasis. Moreover, LumI cells greatly expand during early stages of tumorigenesis in several mouse models of prostate cancer. Genetic ablation of p63 in vivo in luminal cells reduced the formation of aggressive clones in mouse prostate tumor models. Finally, the LumI cells and Wnt signaling appear to significantly increase in human aging prostate and prostate cancer samples, highlighting the importance of this hybrid cell state for human pathologies with potential translational impact.

Immune mechanisms shape the clonal landscape during early progression of prostate cancer.

Understanding the role of the immune microenvironment in modulating intratumor heterogeneity is essential for effective cancer therapies. Using multicolor lineage tracing in genetically engineered mouse models and single-cell transcriptomics, we show that slowly progressing tumors contain a multiclonal landscape of relatively homogeneous subpopulations within a well-organized tumor microenvironment. In more advanced and aggressive tumors, however, the multiclonal landscape develops into competing dominant and minor clones accompanied by a disordered microenvironment. We demonstrate that this dominant/minor landscape is associated with differential immunoediting, in which minor clones are marked by an increased expression of IFNγ-response genes and the T cell-activating chemokines Cxcl9 and Cxcl11. Furthermore, immunomodulation of the IFNγ pathway can rescue minor clones from elimination. Notably, the immune-specific gene signature of minor clones exhibits a prognostic value for biochemical recurrence-free survival in human prostate cancer. These findings suggest new immunotherapy approaches for modulating clonal fitness and tumor progression in prostate cancer.

Role of the p53 family in stabilizing the genome and preventing polyploidization.

Cellular defects resulting in chromosomal instability and aneuploidy are the most common features of human cancers. As a major tumor suppressor and intrinsic part of several cellular checkpoints, p53 contributes to maintenance of the stability of the genetic material, both in quality (ensures faithful replication) and quantity (preservation of diploidy). Although the exact trigger of p53 in case of numerical chromosomal aberrations is unknown, the absence of p53 allows polyploid cells to proliferate and generate unstable aneuploid progeny. A more recent addition to the p53 family, p73, emerged as an important contributor to genomic integrity when p53 is inactivated. p73 loss in p53-null background leads to a rapid increase in polyploidy and aneuploidy, markedly exceeding that caused by p53 loss alone. Constitutive deregulation of Cyclin-Cdk and p27/Kip1 activities and excess failure of the G2/M DNA damage checkpoint are important deficiencies associated with p73 loss.

Mitochondrially targeted p53 has tumor suppressor activities in vivo.

Complex proapoptotic functions are essential for the tumor suppressor activity of p53. We recently described a novel transcription-independent mechanism that involves a rapid proapoptotic action of p53 at the mitochondria and executes the shortest known circuitry of p53 death signaling. Here, we examine if this p53-dependent mitochondrial program could be exploited for tumor suppression in vivo. To test this, we engage Emu-Myc transgenic mice, a well-established model of p53-dependent lymphomagenesis. We show that exclusive delivery of p53 to the outer mitochondrial membrane confers a significant growth disadvantage on Emu-Myc-transformed B-cells of p53-deficient or alternate reading frame-deficient genotypes, resulting in efficient induction of apoptosis and impinged proliferation. Conversely, normal cells from thymus, spleen, and bone marrow showed poor infectivity with these viruses. This proof-of-principle experiment shows that exclusive reliance on the direct mitochondrial program exerts a significant tumor suppressor activity in vivo. Our in vivo data on the direct mitochondrial apoptotic p53 program lays the groundwork to further investigate its efficacy and safety and to address its possible therapeutic value in the future.

A computational systems approach identifies synergistic specification genes that facilitate lineage conversion to prostate tissue.

To date, reprogramming strategies for generating cell types of interest have been facilitated by detailed understanding of relevant developmental regulatory factors. However, identification of such regulatory drivers often represents a major challenge, as specific gene combinations may be required for reprogramming. Here we show that a computational systems approach can identify cell type specification genes (master regulators) that act synergistically, and demonstrate its application for reprogramming of fibroblasts to prostate tissue. We use three such master regulators (FOXA1, NKX3.1 and androgen receptor, AR) in a primed conversion strategy starting from mouse fibroblasts, resulting in prostate tissue grafts with appropriate histological and molecular properties that respond to androgen-deprivation. Moreover, generation of reprogrammed prostate does not require traversal of a pluripotent state. Thus, we describe a general strategy by which cell types and tissues can be generated even with limited knowledge of the developmental pathways required for their specification in vivo.

Cortical hypoplasia and ventriculomegaly of p73-deficient mice: Developmental and adult analysis.

Trp73, a member of the p53 gene family, plays a crucial role in neural development. We describe two main phenotypic variants of p73 deficiency in the brain, a severe one characterized by massive apoptosis in the cortex leading to early postnatal death and a milder, non-/low-apoptosis one in which 50% of pups may reach adulthood using an intensive-care breeding protocol. Both variants display the core triad of p73 deficiency: cortical hypoplasia, hippocampal malformations, and ventriculomegaly. We studied the development of the neocortex in p73 KO mice from early embryonic life into advanced age (25 months). Already at E14.5, the incipient cortical plate of the p73 KO brains showed a reduced width. Examination of adult neocortex revealed a generalized, nonprogressive reduction by 10-20%. Area-specific architectonic landmarks and lamination were preserved in all cortical areas. The surviving adult animals had moderate ventricular distension, whereas pups of the early lethal phenotypic variant showed severe ventriculomegaly. Ependymal cells of wild-type ventricles strongly express p73 and are particularly vulnerable to p73 deficiency. Ependymal denudation by apoptosis and reduction of ependymal cilia were already evident in young mice, with complete absence of cilia in older animals. Loss of p73 function in the ependyma may thus be one determining factor for chronic hydrocephalus, which leads to atrophy of subcortical structures (striatum, septum, amygdala). p73 Is thus involved in a variety of CNS activities ranging from embryonic regulation of brain size to the control of cerebrospinal fluid homeostasis in the adult brain via maintenance of the ependyma.

While p73 is essential, p63 is completely dispensable for the development of the central nervous system.

The ancient p53 paralogs p63 and p73 regulate specific tissue formation, cell survival and cell death via their TA and ΔN isoforms. Targeted disruption of the p73 locus leads to severe defects in the development of the central nervous system (CNS), and p73 has recently been shown to be an essential regulator of neural stem cell maintenance and differentiation in both embryonal and adult neurogenesis. In contrast, global p63-/- mice lack skin and limbs. Moreover, p63 is detectable in embryonic cortex. It has previously been proposed to also play critical pro-death and pro-survival roles in neural precursors of the developing sympathetic and central nervous system, respectively, based on experimental overexpression and siRNA-mediated knockdown of p63. Here we perform an extensive analysis of the developing central nervous system in global p63-/- mice and their wildtype littermates. Brain and spinal cord of embryos and newborn mice were assessed in vivo for neuroanatomy, histology, apoptosis, proliferation, stemness and differentiation, and in vitro for self-renewal and maturation in neurosphere assays. None of these analyses revealed a detectable phenotype in p63-/- mice. Hence, despite the profound impact of p63 on the development of stratified epithelia and limbs, p63 is completely dispensable for proper development of the central nervous system. Thus, despite their strong homology, the non-overlapping tissue specificity of p63 and p73 functions appears more pronounced than previously anticipated.

Functional inactivation of endogenous MDM2 and CHIP by HSP90 causes aberrant stabilization of mutant p53 in human cancer cells.

The tight control of wild-type p53 by mainly MDM2 in normal cells is permanently lost in tumors harboring mutant p53, which exhibit dramatic constitutive p53 hyperstabilization that far exceeds that of wild-type p53 tumors. Importantly, mutant p53 hyperstabilization is critical for oncogenic gain of function of mutant p53 in vivo. Current insight into the mechanism of this dysregulation is fragmentary and largely derived from ectopically constructed cell systems. Importantly, mutant p53 knock-in mice established that normal mutant p53 tissues have sufficient enzymatic reserves in MDM2 and other E3 ligases to maintain full control of mutant p53. We find that in human cancer cells, endogenous mutant p53, despite its ability to interact with MDM2, suffers from a profound lack of ubiquitination as the root of its degradation defect. In contrast to wild-type p53, the many mutant p53 proteins which are conformationally aberrant are engaged in complexes with the HSP90 chaperone machinery to prevent its aggregation. In contrast to wild-type p53 cancer cells, we show that in mutant p53 cancer cells, this HSP90 interaction blocks the endogenous MDM2 and CHIP (carboxy-terminus of Hsp70-interacting protein) E3 ligase activity. Interference with HSP90 either by RNA interference against HSF1, the transcriptional regulator of the HSP90 pathway, or by direct knockdown of Hsp90 protein or by pharmacologic inhibition of Hsp90 activity with 17AAG (17-allylamino-17-demethoxygeldanamycin) destroys the complex, liberates mutant p53, and reactivates endogenous MDM2 and CHIP to degrade mutant p53. Of note, 17AAG induces a stronger viability loss in mutant p53 than in wild-type p53 cancer cells. Our data support the rationale that suppression of mutant p53 levels in vivo in established cancers might achieve clinically significant effects.

E2F1 plays a direct role in Rb stabilization and p53-independent tumor suppression.

To better understand the role of E2F1 in tumor formation, we analyzed spontaneous tumorigenesis in p53(-/-)E2F1(+/+) and p53(-/-)E2F1(-/-) mice. We show that the combined loss of p53 and E2F1 leads to an increased incidence of sarcomas and carcinomas compared to the loss of p53 alone. E2F1-deficient tumors show wide chromosomal variation, indicative of genomic instability. Consistent with this, p53(-/-)E2F1(-/-) primary fibroblasts have a reduced capacity to maintain genomic stability when exposed to S-phase inhibitors or genotoxic drugs. A major mechanism of E2F1's contribution to genomic integrity lies in mediating stabilization and engagement of the Rb protein.

p73 suppresses polyploidy and aneuploidy in the absence of functional p53.

Previous studies showed that p53 plays a central role in G1 and DNA damage checkpoints, thus contributing to genomic stability. We show here that p73 also plays a role in genomic integrity but this mechanism is manifest only when p53 is lost. Isolated p73 loss in primary cells does not induce genomic instability. Instead, it results in impaired proliferation and premature senescence due to compensatory activation of p53. Combined loss of p73 and p53 rescues these defects, but at the expense of exacerbated genomic instability. This leads to rapid increase in polyploidy and aneuploidy, markedly exceeding that of p53 loss alone. Constitutive deregulation of cyclin-Cdk activities and excess failure of the G2/M DNA damage checkpoint appear to fuel increased ploidy abnormalities upon p53/p73 loss, while primary mitotic defects do not play a causal role. These data indicate that p73 is essential for suppressing polyploidy and aneuploidy when p53 is inactivated.