This web page was produced as an assignment for an undergraduate course at Davidson College
Paper Review Two: Tax Interactome
Motivation: what prompted researchers to do the experiment
HTLV-1 Tax protein plays an important role in the development of Adult T-cell Leukemia (ATL). Because of its involvement in ATL, a significant amount of research projects has been conducted to study Tax’s functions and its protein interactions. According to the authors, a system for analysis and validation of these interactions is required to identify protein-protein interactions of no functional consequences. Thus, they have proposed two methods to identify functionally significant Tax-protein interactions. One is the wet lab approach using affinity isolation of Tax complexes coupled to MS/MS analysis to construct a physical Tax interactome map and the other is the in silico approach using computer modeling and node graph visualization techniques to create Tax interactome node graphs of first and second neighbors of known Tax-binding proteins.
Wet Lab Approach
To construct a physical Tax interactome map, the researchers isolated Tax-containing multi-protein complexes from mammalian cells using affinity tagged Tax protein. They fused His6 and S-tags to the amino-terminal of the Tax protein so that the protein can be purified on S-agarose beads. They also fused GFP protein to the carboxyl-terminal of the Tax protein so that it could be localized inside cells. The S-Tax-GFP was put into 293T cells. After purification and incubation, protein complexes with good Tax-binding specificity were trypsinized and analyzed using LC-MS/MS method. When the experiment was repeated three times, 86% of the proteins were detected in all three runs. The investigators also developed a ranking system to determine the strength of protein interactions using four measurements: unique peptides (the number of peptides with sequence unique to the protein), protein score (the sum of the relevant peptide confidence scores), percentage of sequence coverage and emPAI (the relative abundance of predicted peptides from a protein). Using this ranking method, the researchers found five proteins that have the highest strength of protein interactions. The following table shows the measurements of the five proteins which are sorted by number of unique peptides.
In Silico Approach
To construct an in silico Tax interactome map, the researchers reviewed articles containing “Tax Interaction” in PubMed and came up with a list of 67 proteins. By only selecting proteins that have a known function on cellular DNA repair response process, they narrowed the list down to four Tax-binding proteins: Rad51, TOP1, Chk2, and 53BP1. The authors referred to these four proteins as C1. By gathering data from the Human Protein Reference Database (HRPD), they have identified 46 proteins that bind to proteins in C1 (first neighbor interactions of known Tax-binding proteins). The researchers then constructed a Node-Link graph of 50 (including C1) proteins with 112 interactions as shown below. C1 proteins and DNA-PK are highlighted in Yellow.
The 5-core or the interacting group of 12 proteins which have at least 5 edges contains DNA-PKcs, TOP1, PCNA, RPA1, DDX9, CDK4, CDKN1A (p21), CDK5, ADPRT (PARP), XRCC5 (Ku70), XRCC6 (Ku86), and NCOA6 (TRBP). All 12 proteins are involved in DNA-repair process and 6 of them were among the Tax-binding proteins observed using LC-MS/MS analysis. The authors especially emphasized the presence of DNA-PKcs in the 5-core group and noted that DNA-PKcs which has eight degrees (eight interactions) is in the top 30% in both betweenness and closeness measures.
The researchers then removed C1 from G1 and showed the largest connected network with 29 proteins and 60 interactions in the graph below.The highlighted DNA-PKcs still has 7 degrees or interactions.
They then removed proteins that do not involve in the DNA repair process and generate a graph of 26 proteins with 42 interactions call G1*. DNA-PKcs was in the group of 4-core (maximum core) and ranked 9 in betweeness measure. Howeve, I am curious about why the authors did not mention DNA-PKcs’s closeness measure in G1*.
In the next phase of their analysis, the researchers looked at all the known proteins that interact with the Tax-binding proteins and generated a graph (G2) consisted of 667 proteins and 3827 interactions. A sub-network (G2*) of 114 proteins was created by removing proteins that lack a function in DNA repair. Using a spectral algorithm, they obtained clusters by dividing the graph using a measure called the cohesion of the sub-graph.
Figure 4 shows the 3-core (each protein in the network has at least 3 degrees) of the G2* network that contains five clusters of 54 proteins with the largest cluster having 22 proteins and the smallest cluster having 3 proteins. Proteins that connect these clusters are called “bridges”. There are three bridges: DNA-PKcs, PCNA and TP53. Again, DNA-PKcs also plays an important role in the second-neighbor graph. Interesting, I noticed that some nodes in the clusters only have two edges. Perhaps, these nodes interact with the bridges since connections of the bridges to the individual proteins were not shown in Graph.
Since the five clusters contained proteins that involve in DNA repair, stress-induced signalling pathways and cell cycle controls, the researchers concluded that Tax has control over some of the main cellular stress response pathways.
The investigators then performed an affinity pull-down of Tax protein complexes. They transfected S-Tax and S-GFP in 293T cells and obtained the extracts. The extracts were then purified and subjected to immunoblotting with anti-DNA-PKcs. Lane 1 and Lane 3 contain input extracts. Lane 2 contains S-GFP which acts as negative control. Lane 4 contains S-Tax purified protein complexes. The results in Figure 5 confirmed that DNA-PKcs binds to the Tax protein.
The investigators began with a network of known Tax-binding proteins and their first neighbors and then explored the network of their second neighbors. They demonstrated that DNA- PKcs has high connectivity and high ranks in betweenness and closeness measures in all the networks. They then confirmed in lab that DNA-PKcs interacts with Tax protein directly. Using this method, can we generalize that if a protein has high degrees and high ranks in betweennesss and closeness measures in first-neighbor and second-neighbor analysis, then it is likely to directly interact with the original protein (root of the graph, like Tax protein)?
In the discussion section, the investigators noted that though PI3K protein family members ATM and ATR are equally critical damage response proteins, they were not selected in the in silico protein networks, thus "verifying the novelty of the DNA-PK finding". It was not clear to me whether the investigators mean that with this analysis, they found out DNA-PK is actually more important than ATM and ATR or they mean that ATM and ATR are actually false-positives.
I also hope that the researchers had made their raw data public so that others can repeat the experiment without the need to start from scratch.
References
Ramadan, E, et al. (2008). "Physical and in silico approaches identify DNA-PK in a Tax DNA-damage response interactome". Retrovirology 5:92.
© Copyright 2008 Department of Biology, Davidson College, Davidson, NC 28035