Monitoring the prolonged TNF stimulation in space and time with topological-functional networks

In this work we monitor the prolonged TNF stimulation of mouse synovial fibroblasts in space and time through a combination of functional and topological analyses. We introduce a novel concept in the form of bipartite functional/positional networks that capture the interaction between genome organization and the functional footprint of a given regulatory program. By implementing this approach in a time-dependent gene expression experiment we are able to dissect the complex cellular response to a cytokine trigger. We treated synovial fibroblasts in culture for 1, 3, 6, 24 hours to a final point of 7 days with TNF. Control cultures that were grown for the same period without TNF were used as controls. RNA was extracted and RNASeq was performed on a Solexa NextSeq Platform in triplicates. Mapping was performed with BowTie and differential expression was calculated with Cufflinks/CuffDiff (Langmead and Salzberg, 2012; Trapnell et al., 2012;). The differential expression files were used for analysis with abs(log 2 (FC)) <= 1 AND p-value <= 0.05 as a cutoff threshold to select significantly differentially expressed genes. The files used as input for all the analyses described bellow were files of diff format, containing the fields of: gene_name, chromosome, start coordinates, end coordinates, log 2 (FC), p-value. A total of >1500 were found to be differentially expressed in at least one timepoint. They were clustered in 8 groups depending on their relative gene expression in time. Functional analysis was performed with gProfiler for each of the 8 clusters. We analyzed gene expression in linear genomic space through an implementation of the "breakpoints" R function, which uses a Chow test to define significant differences between adjacent linear models in time series data. We used differential gene expression as the time series data and defined sets of ~300 domains of consistent gene expression in each time point, which we termed Domains of Focal Deregulation. We then selected a subset of DFDs from each timepoint on the basis of high/low mean scores of differential expression as the most prominent in terms of gene deregulation and created bipartite positional/functional bipartite networks through a functional enrichment analysis of the genes contained in each DFD. Comparison of the bipartite networks monitors the progression of TNF stimulation, which appears to take place through two distinct transition points. A first, at 3h with a large expansion of inflammatory and immune-related functions and second, at 24h which is marked by immune-related functions being shut down and replaced by pathways associated with development and cell adhesion.本研究通过功能与拓扑分析相结合的手段，在时空维度下监测小鼠滑膜成纤维细胞（mouse synovial fibroblasts）受到长期肿瘤坏死因子（Tumor Necrosis Factor, TNF）刺激的全过程。本研究提出一种全新的双功能/位置网络（bipartite functional/positional networks）概念，该网络可捕捉基因组组织与特定调控程序的功能印记之间的相互作用。通过将该方法应用于时序基因表达实验，我们得以解析细胞对细胞因子刺激的复杂应答反应。 我们对体外培养的滑膜成纤维细胞施加TNF刺激，处理时长涵盖1、3、6、24小时直至最终的7天时间点；同期未施加TNF的培养物作为对照。提取总RNA后，在Solexa NextSeq平台上进行三次生物学重复的RNA测序（RNA-Sequencing, RNASeq）。序列比对使用BowTie完成，差异表达基因的计算通过Cufflinks/CuffDiff实现（Langmead及Salzberg, 2012; Trapnell等人, 2012）。 以|log₂(FC)| ≤ 1且p值 ≤ 0.05作为筛选阈值，从差异表达基因文件中筛选得到显著差异表达基因。后续所有分析的输入文件均为diff格式文件，其包含的字段为：gene_name（基因名称）、chromosome（染色体）、start coordinates（起始坐标）、end coordinates（终止坐标）、log₂(FC)以及p值。最终共筛选得到超过1500个至少在一个时间点出现差异表达的基因。根据其随时间变化的相对表达水平，将这些基因划分为8个聚类簇。针对这8个聚类簇，分别使用gProfiler进行功能富集分析。 我们通过调用"breakpoints" R函数，在线性基因组空间中分析基因表达：该函数利用Chow检验（Chow test）来定义时序数据中相邻线性模型间的显著差异。本研究以差异基因表达数据作为时序数据集，在每个时间点定义了约300个基因表达一致的区域，将其命名为局域失调域（Domains of Focal Deregulation, DFD）。随后，我们根据差异表达的平均得分高低，从每个时间点的DFD中筛选出基因失调程度最显著的子集，并通过对每个DFD内包含的基因进行功能富集分析，构建位置-功能双分网络。 通过比对这些双分网络，可以追踪TNF刺激的进程：该进程似乎存在两个明确的转换节点。第一个转换节点出现在3小时，此时炎症与免疫相关功能显著扩增；第二个转换节点出现在24小时，此时免疫相关功能被关闭，取而代之的是与发育及细胞黏附相关的通路。