Construction and robustness of TO-GCN
As there are various gene expression patterns in the time-series transcriptome data, we seek construct relationships between TFs by examining their pattern similarity (PCC) in TO-GCN. This meant that the total number of levels represents the dynamic range of different expression patterns in TO-GCN. In this study, the number of levels in TO-GCN depended on the PCC cut-off setting value. The more stringent correlation coefficient or a higher PCC value that is set, the more levels in TO-GCN will be constructed. In contrast, if PCC cut-off set at lower value, more TFs will be grouped in a single level and will therefore yield TO-GCN with lesser levels. In this study, PCC set at ≥ 0.91 (p-value < 0.05), 9 levels of TO-GCN were constructed.
In order to demonstrate the robustness of TO-GCN, we tested the level order stability by using 7 different TF genes with DE ≤ 4 and was co-expressed with FOSL1 (PCC > 0.99) as new initial nodes to construct the corresponding TO-GCNs. The analysis showed that new ordered TO-GCNs are very similar to the original TO-GCN that was constructed with FOSL1 (Additional file 8). This indicated that the new ordered TO-GCNs are very similar to the original TO-GCN that was constructed with FOSL1.
Synergistic work between MFSelector and TO-GCN
MFSelector and TO-GCN were the two methods of data analysis used in this study. These methods worked together synergistically to provide a deeper understanding of MSC differentiation into mesangial cells. MFSelector determined the degree of monotonicity for all genes during the differentiation process and it provided an estimation of the expression behaviour of the gene during differentiation. TO-GCN used co-expression relationship to connect TF genes as pairs, in which they have similar expression patterns (i.e. significantly high PCC) over time. It inferred expression time orders for all TF genes in the network with the starting TF in the strongest descending pattern identified by MFSelector. By applying this method to time-series experiments, TO-GCN provided the time order information of gene regulations in developmental processes. The data obtained from both methods was further used to identify the TF-key genes at specific time points to the TO-GCN at different levels. This helped to elucidate the network interaction between TF-TF and TF-key genes at each level of TO-GCN.
In this study, the FOSL1 gene, expressed in the strongest monotonic descending pattern, was used as initial node. As the network was constructed based on co-expression, TFs in the same or next levels of FOSL1 in TO-GCN would be also in a descending patterns. This was consistent with the genes in descending pattern identified by MFSelector. Lower DE values (stronger monotonic pattern) of descending pattern TFs appeared in early levels from L1 to L2 (green nodes in Fig. 1). The ascending TFs with higher monotonicity (lower DE value) appeared later at the levels from L7 to L9 (purple nodes in Fig. 1). Genes with a weak monotonic pattern (either descending or ascending) were located in between the descending and ascending high monotonicity pattern genes.
The key genes and MSC biomarkers were down-regulated during the differentiation process
MSC biomarkers such as ANPEP and LIF were down-regulated during the differentiation process. ANPEP, also called CD13, is well known as an MSC marker. On the other hand, LIF, another well-established MSC marker, has been reported to affect cell growth by inhibiting differentiation but maintaining the stemness of the stem cell. When LIF expression levels drop, the cells will start the process of differentiation [14]. Meanwhile, depletion of AURKA, known for stem cell renewal, resulted in compromised self-renewal and consequent differentiation [15].
In this study, many genes related to cell cycle regulation (CDK1, CCNB1 and GNL3) and DNA replication (CDC6) were down-regulated. CDK1 is a key regulator of mitosis. High expression levels of CDK1 are associated with the pluripotency stage of embryonic stem cells (ESC). Decreased CDK1 activity to a level without perturbing the cell cycle is sufficient to induce differentiation [16]. Meanwhile CCNB1 gene expression increases during G2/M phase and decreases during terminal differentiation [17]. GNL3, also known as nucleostemin, regulates the cell cycle and affects cell differentiation; the amount of GNL3 decreases as differentiation progresses. GNL3 is also a biomarker for many stem cells and cancer cells [18]. CDC6 is an essential regulator of DNA replication in eukaryotic cells. Down regulation of CDC6 will lead to a drop of DNA replication before differentiation can take place [19, 20]. Even though these genes regulate the cell cycle or DNA replication, all findings show that when these genes are down-regulated in stem cells, differentiation will start.
By referring to the TRR databases, 3 TFs (MEOX2, SOX9 and HMGA1) regulated MSC markers such as LIF and ANPEP. These TFs are known as regulators of the stem cell state through transcriptional networks that induce pluripotency. Theodorou et al. reported that neuronal differentiation in ESC was inhibited when MEOX2 overexpressed [21]. Shah et al. did a study showing that when ESC differentiation was induced, there was a decreased expression of HMGA1 which was also observed in other pluripotency factors. Conversely, forced expression of HMGA1 blocked the differentiation of ESC [22]. Meanwhile for SOX9, upon the differentiation of MSC into hepatocytes, SOX9 expression was down-regulated [23].
Biomarkers contribute to mesangial cell characteristics and functions
Ten mesangial cell key genes with DE≤4 were selected for further analysis. The majority of these key genes are reported as biomarkers for mesangial cells or related to the functions of mesangial cells. TAGLN, or SM22-alpha, is expressed in smooth muscle cells. It is known as one of the earliest commitment biomarkers of differentiated smooth muscle cells and has been suggested to regulate their contractile functions [24]. This gene has a role in generating committed progenitor cells from undifferentiated hMSC by regulating cytoskeleton organization. TAGLN in the kidney is up-regulated in repopulating mesangial cells in vivo. Meanwhile SERPINE2 and IGFBP5 are reported to be expressed in mesangial cells [25, 26] and MYOM1 is known to be expressed in smooth muscle cells [27].
ACTA2 and MYH9 play an important role in regulating both smooth muscle and non-muscle cell contractile activity [25, 28]. Another contraction related gene is PTGIS, also known as prostacyclin synthase. PTGIS is the final committed enzyme in the metabolic pathway leading to prostaglandin I2 (PGI2) production and PGI2 is needed to mediate mechanism of vascular contraction [29]. PDGFRB is needed for stimulation of contraction and chemotaxis [30]. PYGM encodes a muscle enzyme that is involved in glycogenolysis.
Mesangial cells are phagocytic cells and expression of ITGA8 in mesangial cells facilitates phagocytosis. About 15% of the total mesangial cell population in the glomerulus is capable of exhibiting immunological function such as phagocytosis [31].
By referring to the TRR databases, TFs SRF and TEAD3 were found to regulate 3 and 4 key genes respectively. This shows that TF SRF and TEAD3 play an important role in mesangial cells. SRF is a ubiquitous expressed TF that drives smooth muscle cell-specific gene expression and is necessary for contractile and cytoskeletal functions [32, 33]. TEAD3 has been reported to abolish myocardin function and is consistently expressed in smooth muscle cells [34].
Pathway enrichment analysis on each TO-GCN level
Proliferation and differentiation processes are two distinct and mutually exclusive processes during development. To initiate stem cell differentiation, certain cell proliferation related genes or pathways have to be down-regulated. Estefanía et al. has reported that terminal differentiation is the process by which dividing cells stop proliferating in order to start new specific functions, which means that DNA replication fades as cells advance in their commitment to terminal differentiation [35]. Therefore, these early levels can be classified as differentiation preparation.
From L2 to L7, pathways involved in regulating or triggering differentiation were enriched. Lou et al. showed that RNA degradation drives stem cell differentiation [36]. They discovered that the steady-state level of RNAs is dictated by their decay rate and this specific RNA decay such as Nonsense-mediated mRNA decay (NMD) have a role in promoting differentiation mechanisms [37, 38]. NMD is a surveillance pathway and its main function is as a quality control pathway to reduce errors in gene expression by eliminating mRNA transcripts that contain premature stop codons [39]. During the differentiation, NMD elicits the decay of specific subsets of mRNAs and promotes the decay of mRNAs encoding pluripotency factors [36].
From L7 to L9, pathways involved in mesangial cell maturation were enriched. Some of the enriched pathways are related to characteristics and functions of mesangial cells such as contraction and phagocytosis ability. Therefore it is not surprising to find that the oxytocin signaling pathway, phagosome and certain cardiac related pathways were enriched.
The strongest evidence that the co-cultured MSC have differentiated into mesangial cells is by confirming the pathway enrichment of vascular smooth muscle contraction [40]. As mesangial cells are modified smooth muscle cells, we have further conducted a wet lab contraction functional validation. Results showed that the differentiated cells can contract and have proven that the cells have fully matured in their differentiation in which the cells now possess mesangial cell functions. On the other hand, pure MSC failed to exhibit contraction ability.
Construction of vascular smooth muscle contraction-specific gene network
In this study we have shown that differentiated mesangial cells have contraction ability. With this specific biological function, a co-expression gene network related to vascular smooth muscle contraction was constructed. This network has illustrated the relationship between key genes and its upstream regulators or TFs. To our knowledge, such function-specific TF-TF-vascular smooth muscle contraction-related key gene network has not been reported before. This indicates that a mathematically calculated co-expression network can provide us with a first step or hints for further wet lab validation before full biological TF-TF-key gene relationship is fully uncovered. Other biological functions such as phagocytosis of mesangial cells can also be explored and constructed using the methods presented herein.