Protocol to use TopNet for gene regulatory network modeling using gene expression data from perturbation experiments.

Inference of gene regulatory networks from gene perturbation experiments is the most reliable approach for investigating interdependence between genes. Here, we describe the initial gene perturbations, expression measurements, and preparation steps, followed by network modeling using TopNet. Summarization and visualization of the estimated networks and optional genetic testing of dependencies revealed by the network model are demonstrated. While developed for gene perturbation experiments, TopNet models data in which nodes are both perturbed and measured. For complete details on the use and execution of this protocol, please refer to McMurray et al. (2021).

Gene network modeling via TopNet reveals functional dependencies between diverse tumor-critical mediator genes.

Malignant cell transformation and the underlying reprogramming of gene expression require the cooperation of multiple oncogenic mutations. This cooperation is reflected in the synergistic regulation of non-mutant downstream genes, so-called cooperation response genes (CRGs). CRGs affect diverse hallmark features of cancer cells and are not known to be functionally connected. However, they act as critical mediators of the cancer phenotype at an unexpectedly high frequency >50%, as indicated by genetic perturbations. Here, we demonstrate that CRGs function within a network of strong genetic interdependencies that are critical to the malignant state. Our network modeling methodology, TopNet, takes the approach of incorporating uncertainty in the underlying gene perturbation data and can identify non-linear gene interactions. In the dense space of gene connectivity, TopNet reveals a sparse topological gene network architecture, effectively pinpointing functionally relevant gene interactions. Thus, among diverse potential applications, TopNet has utility for identification of non-mutant targets for cancer intervention.

Multiple imputation and direct estimation for qPCR data with non-detects.

BACKGROUND: Quantitative real-time PCR (qPCR) is one of the most widely used methods to measure gene expression. An important aspect of qPCR data that has been largely ignored is the presence of non-detects: reactions failing to exceed the quantification threshold and therefore lacking a measurement of expression. While most current software replaces these non-detects with a value representing the limit of detection, this introduces substantial bias in the estimation of both absolute and differential expression. Single imputation procedures, while an improvement on previously used methods, underestimate residual variance, which can lead to anti-conservative inference. RESULTS: We propose to treat non-detects as non-random missing data, model the missing data mechanism, and use this model to impute missing values or obtain direct estimates of model parameters. To account for the uncertainty inherent in the imputation, we propose a multiple imputation procedure, which provides a set of plausible values for each non-detect. We assess the proposed methods via simulation studies and demonstrate the applicability of these methods to three experimental data sets. We compare our methods to mean imputation, single imputation, and a penalized EM algorithm incorporating non-random missingness (PEMM). The developed methods are implemented in the R/Bioconductor package nondetects. CONCLUSIONS: The statistical methods introduced here reduce discrepancies in gene expression values derived from qPCR experiments in the presence of non-detects, providing increased confidence in downstream analyses.

Bone marrow mesenchymal stromal cells from acute myelogenous leukemia patients demonstrate adipogenic differentiation propensity with implications for leukemia cell support.

Bone marrow mesenchymal stromal cells (MSCs) constitute one of the important components of the hematopoietic microenvironmental niche. In vivo studies have shown that depletion of marrow MSCs resulted in reduction of hematopoietic stem cell content, and there is in vitro evidence that marrow MSCs are able to support leukemia progenitor cell proliferation and survival and provide resistance to cytotoxic therapies. How MSCs from leukemia marrow differ from normal counterparts and how they are influenced by the presence of leukemia stem and progenitor cells are still incompletely understood. In this work, we compared normal donor (ND) and acute myelogenous leukemia (AML) derived MSCs and found that AML-MSCs had increased adipogenic potential with improved ability to support survival of leukemia progenitor cells. To identify underlying changes, RNA-Seq analysis was performed. Gene ontology and pathway analysis revealed adipogenesis to be among the set of altered biological pathways dysregulated in AML-MSCs as compared with ND-MSCs. Expression of both SOX9 and EGR2 was decreased in AML-MSCs as compared with ND-MSCs. Increasing expression of SOX9 decreased adipogenic potential of AML-MSCs and decreased their ability to support AML progenitor cells. These findings suggest that AML-MSCs possess adipogenic potential which may enhance support of leukemia progenitor cells.

Developmental alcohol exposure impairs synaptic plasticity without overtly altering microglial function in mouse visual cortex.

Fetal alcohol spectrum disorder (FASD), caused by gestational ethanol (EtOH) exposure, is one of the most common causes of non-heritable and life-long mental disability worldwide, with no standard treatment or therapy available. While EtOH exposure can alter the function of both neurons and glia, it is still unclear how EtOH influences brain development to cause deficits in sensory and cognitive processing later in life. Microglia play an important role in shaping synaptic function and plasticity during neural circuit development and have been shown to mount an acute immunological response to EtOH exposure in certain brain regions. Therefore, we hypothesized that microglial roles in the healthy brain could be permanently altered by early EtOH exposure leading to deficits in experience-dependent plasticity. We used a mouse model of human third trimester high binge EtOH exposure, administering EtOH twice daily by subcutaneous injections from postnatal day 4 through postnatal day 9 (P4-:P9). Using a monocular deprivation model to assess ocular dominance plasticity, we found an EtOH-induced deficit in this type of visually driven experience-dependent plasticity. However, using a combination of immunohistochemistry, confocal microscopy, and in vivo two-photon microscopy to assay microglial morphology and dynamics, as well as fluorescence activated cell sorting (FACS) and RNA-seq to examine the microglial transcriptome, we found no evidence of microglial dysfunction in early adolescence. We also found no evidence of microglial activation in visual cortex acutely after early ethanol exposure, possibly because we also did not observe EtOH-induced neuronal cell death in this brain region. We conclude that early EtOH exposure caused a deficit in experience-dependent synaptic plasticity in the visual cortex that was independent of changes in microglial phenotype or function. This demonstrates that neural plasticity can remain impaired by developmental ethanol exposure even in a brain region where microglia do not acutely assume nor maintain an activated phenotype.

On non-detects in qPCR data.

MOTIVATION: Quantitative real-time PCR (qPCR) is one of the most widely used methods to measure gene expression. Despite extensive research in qPCR laboratory protocols, normalization and statistical analysis, little attention has been given to qPCR non-detects-those reactions failing to produce a minimum amount of signal. RESULTS: We show that the common methods of handling qPCR non-detects lead to biased inference. Furthermore, we show that non-detects do not represent data missing completely at random and likely represent missing data occurring not at random. We propose a model of the missing data mechanism and develop a method to directly model non-detects as missing data. Finally, we show that our approach results in a sizeable reduction in bias when estimating both absolute and differential gene expression. AVAILABILITY AND IMPLEMENTATION: The proposed algorithm is implemented in the R package, nondetects. This package also contains the raw data for the three example datasets used in this manuscript. The package is freely available at http://mnmccall.com/software and as part of the Bioconductor project.

On the underlying assumptions of threshold Boolean networks as a model for genetic regulatory network behavior.

Boolean networks (BoN) are relatively simple and interpretable models of gene regulatory networks. Specifying these models with fewer parameters while retaining their ability to describe complex regulatory relationships is an ongoing methodological challenge. Additionally, extending these models to incorporate variable gene decay rates, asynchronous gene response, and synergistic regulation while maintaining their Markovian nature increases the applicability of these models to genetic regulatory networks (GRN). We explore a previously-proposed class of BoNs characterized by linear threshold functions, which we refer to as threshold Boolean networks (TBN). Compared to traditional BoNs with unconstrained transition functions, these models require far fewer parameters and offer a more direct interpretation. However, the functional form of a TBN does result in a reduction in the regulatory relationships which can be modeled. We show that TBNs can be readily extended to permit self-degradation, with explicitly modeled degradation rates. We note that the introduction of variable degradation compromises the Markovian property fundamental to BoN models but show that a simple state augmentation procedure restores their Markovian nature. Next, we study the effect of assumptions regarding self-degradation on the set of possible steady states. Our findings are captured in two theorems relating self-degradation and regulatory feedback to the steady state behavior of a TBN. Finally, we explore assumptions of synchronous gene response and asynergistic regulation and show that TBNs can be easily extended to relax these assumptions. Applying our methods to the budding yeast cell-cycle network revealed that although the network is complex, its steady state is simplified by the presence of self-degradation and lack of purely positive regulatory cycles.

Gene sets identified with oncogene cooperativity analysis regulate in vivo growth and survival of leukemia stem cells.

Leukemia stem cells (LSCs) represent a biologically distinct subpopulation of myeloid leukemias, with reduced cell cycle activity and increased resistance to therapeutic challenge. To better characterize key properties of LSCs, we employed a strategy based on identification of genes synergistically dysregulated by cooperating oncogenes. We hypothesized that such genes, termed "cooperation response genes" (CRGs), would represent regulators of LSC growth and survival. Using both a primary mouse model and human leukemia specimens, we show that CRGs comprise genes previously undescribed in leukemia pathogenesis in which multiple pathways modulate the biology of LSCs. In addition, our findings demonstrate that the CRG expression profile can be used as a drug discovery tool for identification of compounds that selectively target the LSC population. We conclude that CRG-based analyses provide a powerful means to characterize the basic biology of LSCs as well as to identify improved methods for therapeutic targeting.

Synergistic response to oncogenic mutations defines gene class critical to cancer phenotype.

Understanding the molecular underpinnings of cancer is of critical importance to the development of targeted intervention strategies. Identification of such targets, however, is notoriously difficult and unpredictable. Malignant cell transformation requires the cooperation of a few oncogenic mutations that cause substantial reorganization of many cell features and induce complex changes in gene expression patterns. Genes critical to this multifaceted cellular phenotype have therefore only been identified after signalling pathway analysis or on an ad hoc basis. Our observations that cell transformation by cooperating oncogenic lesions depends on synergistic modulation of downstream signalling circuitry suggest that malignant transformation is a highly cooperative process, involving synergy at multiple levels of regulation, including gene expression. Here we show that a large proportion of genes controlled synergistically by loss-of-function p53 and Ras activation are critical to the malignant state of murine and human colon cells. Notably, 14 out of 24 'cooperation response genes' were found to contribute to tumour formation in gene perturbation experiments. In contrast, only 1 in 14 perturbations of the genes responding in a non-synergistic manner had a similar effect. Synergistic control of gene expression by oncogenic mutations thus emerges as an underlying key to malignancy, and provides an attractive rationale for identifying intervention targets in gene networks downstream of oncogenic gain- and loss-of-function mutations.