02. Homology Model

Details:

I. What do we know:

The mammalian NF-κB family consists of p65 (RelA), RelB, c-Rel, p50/p105 (NF-κB1) and p52/p100 (NF-κB2), commonly held in the cytoplasm by Ser/Thr specific IκB kinases in a deactivated state attached at their C-terminus via ankyrin repeats that hides the RHR of NF-κB. Degradation of these repressors frees up NF-κB to move into nucleus and perform transcription.

NFκB activation is done via two pathways -- (i) Classical for formation of p50/RelA that performs transcription for inflamation and DC maturation; and (ii) Alternative for formation of p52/RelB that performs transcription for lymphoangiogenesis.

The genes and the proteins are highly conserved in the animal kingdom. Although, The genes could be present at different locations. For ex, Chromosome 4 on mice and 22 on humans for TNFα induced inflamation.

The IκBs are degrated by IKKs. In steady state, IKKα-IKKβ dimer is held together by IKKγ (NEMO) in the cytoplasm. IKKα is also found in free state in the cytoplasm with a half life of 120 minutes. In phosphorylated state, the half life is reduced to 10 minutes.

Classical pathway involves: NEMO ubiquitination and proteolysis-> IKKß phosphorylation -> phosphorylates IκB -> ubiquination and proteolysis of IκBα -> Release of p50/RelA -> Translocation to nucleoplasm -> Transcription.

Alternative pathway involves: TRAF2/TRAF3/NIK phosphorylates IKKα -> Phosphorylates p100 -> Ubiquitination and degradation of p100 to p52 -> Translocation to nucleoplasm -> Transcription.

II. What has already been found/proved:

We know the pathways for sure. We know what each pathway achieves. We know the activators/inhibitors (regulators) of the pathway. We know about other pathways that interact with the NF-κB pathway (Please see attached). We know that the activation involves ubiquitination and proteolysis of various repressors. We know the transcription process. We know the genes involved. NEMO regulation has been reported but not tested independently.

III. What has not been found yet:

We don't know the exact sites of interaction of the various regulators. We don't know the exact steps where the regulators come into picture. We don't know the exact ubiquitination sites. We know there would be some intermediate molecules (E1, E2, E3) between Ubiquitination and Proteolysis, but we don't know what they could be.

IV. Why hasn't it been found yet:

Most of the studies done so far are in vivo. This mostly involved overoexposure to/suppression of a regulator or creating a gene-/- mutant to see the effects. Doing this ascertained that yes, that stuff is required, but then how exactly is that stuff required involves complete isolation of the pathway from external factors. This is not feasible in an in-vivo environment because of the sheer number of interacting pathways that NF-κB is a part of.

V. Why should anybody bother:

Earlier, when it was discovered that NF-κB is the reason for so many inflammatory diseases, it was theorised that a complete elimination of NF-κB would help rid mankind from a lot of complications. However, studies on mice revealed that Nf-κB deficient zygotes could never cross blastocyst stage. When NF-κB was removed at a later stage in development, the mice had wrinkled skin and severe hair loss and died shortly. Later on, it was discovered that NF-κB was an integral part of cell cycle and immuno-regulatory network.

Ubiquitination sites of NF-κB regulators are specific to the type of NF-κB family factor involved, which in turn is gene (the κb site that would be transcribed) specific, which in turn means that it would be specific to the disease/disorder involved. Thus, if we can figure out the ubiquitination sites and mechanism of the NF-κB regulators, then developing targetted drugs would become feasible.

VI. What am I tryting to do:

I am trying to figure out the Ubiquitination of Alternate pathway. This would mean trying to determine the role of NIK, TRAF-2,-3 in the ubiquitination of IKKα and then subsequent ubiquitination of p100 by IKKα. Then by using only these substrates, I would try to model the ubiquitin binding sites.

VII. How am I going to do that:

0. We know that the structures and binding sites of NF-κB family proteins are conserved in mammals. We also know that the alternate pathway involves NIK, TRAF and IKKα and does not involve IKKβ, IKKγ, IL and RIP proteins. We know the genes

1. Take the available structures. Until now, we have the following structures available from PDB:

a. 1IKN.pdb

b. 1SVC

c. 1GJI

d. 1VKX

e. 1RAM

f. 1NFI

g. 2RAM

h. 1A3Q

i. 2A4D

j. 2HLW

k. 2C2V

l. 1CA4

m. 1CA9

o. 1QSC

p. 1CZZ

q. 3DO7

r. 2D96

s. 3JV6

t. 4OT9

u. 1BFT

2. Here we can see that only #b and #f have human structures.

We can use BLASTP/PSIBLAST combo to see the conservation. We can use the tools BLAST+ and dendogram to see the relationships.

3. We can reject the usual suspects.

We can also use our knowledge of genes encoded and perform a TBLASTX to see some reverse relationships and further eliminate stuff.

Not inclined to do that right now, simply becuase I forgot how to do it.

Nevertheless, the genetic angle would an interesting angle to see the relationships and create groups, especially considering the fact that each pathway codes more than 200 genes in humans. Not taking that route right now.

4. Finalise the sample sets. Create 9 such sample sets.

Here I am planning to divide the sample sets on the basis of existing research and pathway (classical/alternative) impacted.

One set will contain NEMO ubiquitination (ub),

one will contain IKKα ub,

one will contain RIP ub,

one will contain NIK ub,

another one MAP3K ub,

another TRAF2,3 ub,

one will have IL6 ub,

one will have a custom ub (assuming direct UV/radiography experiements can achieve the results in vitro).

5. Use CLUSTAL for MSA. Use HMMER to build the profile that we use in model building.

I am also inclined to create a sequence alignment with MUSCLE since that seems to be an industry standard.

Use Bio::Phylo package to build a newick tree.

MGENTREADER server is known to provide very good alignments. We can submit the sequences to double check our results.

6. Use modeller. I would like to reuse bits of the server that we developed back in 2008 as a base. Since we are already published on that front, we shouldn't have problems in establishing credibility about the approach.

However, the server was capable of predicting structures only for extremely close candidates (if it could) found in the database.

So, in this case, apart from doing the usual build_profile, align2d and automodel functions, if the model evaluation fails to give a good fit (showing erratic MOLPDF/DOPE scores), we can move on to perform loop refnining using loopmodel and hetatm classes. Further refinement can be achieved by application of moulding functions.

I would like to use EM Maps too for a perfect structure fit, but I couldn't find any templates for the PDB structures that I am interested in. If I can find some, then a combined application of HMMER and MODELLER could achieve pretty good results.

7. Repeat #5 and #6 for all sets to get at least a 100 iterations for each resultset.

8. Get a Chi-squared distribution of the resultset. Here we know that the distribution is more akin to log-normal than normal. However, the returns of an RMSD distribution should become normal. We can take that approach if a simple procedure fails to give 'good' results.

9. Get a z-value of the distribution. Alternatively, if we had lots of data points, we could have constructed a Monte Carlo distribution, and then solved it using Fourier Transform based on Newton-Raphson method.

Any drag could be detected using a Multivariate GARCH Analysis giving a more pratical ANOVA.

We can still do it if/when we get time to have numerous data points. However, since we have only about 10x100 data points, such an approach cannot be taken.

Nevertheless, since some reseach has already been done on this front, and I am reusing their results, I am assuming a z-distribution should be able to give fairly accurate results.

Submit the GM ± σ to Swiss Model Server for validation/evaluation.

Publish results.

VIII. What am I doing different:

There is nothing fundamentally revolutionary in what I am doing. However, all of the current research attempts to study the in-vivo impacts of changing the components of the NF-κB pathway, and once they get some results, they try to guess the receptors/amino acids involved. A few teams go to lengths of creating a computer model to confirm their results. BUt even then, that is always (as many as I have seen) a cursory modeling of their end products they get from wet lab. I, on the other hand, am attempting to take the existing research outcomes from various publications and am trying to create a feasible computer model. This will involve attempting to achieve a best fit from among the best candidates from all those experiments. Once the model has been finalised, it can be extended to study the impact of other other pathways. Later, as and when a closed form loop is obtained statistically, the results can be emulated in an in-vivo environment. This way, there is negligible guess work done. Plus the time taken to achieve all this should be drastically less than only-in-vivo experiments.

IX. What am I missing:

Above all, I am missing a solid proof that my model does work in-vivo. Moreover, I am working with crystalline structures as submitted to PDB. The veracity of my results completely depends on the accuracy of the structures submitted. Even though I plan to work at interacting pathways, at this point, I am working on isolated structures. That's not how things work out in a live organism.

X. How can that be fixed:

All the shortcomings can be eradicated by performing in-vivo experiments. Once the model is proved in a live model, one can be (almost) sure that the model works. The studies would be needed to be done in stages with a control in place. We know that Ubiquitination causes conformational changes. These changes would cause differential movement of tagged receptors in an assay. Once the sites are figured out, and the canonical structures have been modelled, we need to find homologous motifs (natural/artificial) and then use those motifs to find/create attachment vectors. Post my findings, we can attach ig/radioactive vectors to the specific sites (to be found) or fraction with SDS-PAGE and then Western Blot and Immunoprecipitation experiments could be conducted with anti-c-Myc-agarose affinity gel on the samples. The immunoprecipitates can be subsequently abluted with lysis and SDS loading buffer. The nitrocellulose membranes could be developed by the enhanced chemiluminescence method according to the manufacturer's protocol. Electrophoretic Mobility Shift Assay could be performed by pelleting the nuclei and removing cytoplasmic proteins.Resuspension of pellets in extraction buffer, EDTA, phenylmethylsulphonyl fluoride and protease inhibitor mixture. Nuclear proteins from the vortexed and centrifuged supernatant can be collected and the protein concentrations can be determined after incubating in a NF-κB-specific oligonucleotide probe and fractioning on a polyacrylamide gel that can be later visualised by autoradiography.

Within the purview of CADD, the models can be improved by:

1. Using the knowledge of genes transcribed and perform a TBLASTX to get additional datasets for feeding into hmmer.

2. Using EM Maps as and when they are available in the modeller scripts.

3. Manually modelling the GM structures before submitting to Swiss Model Server.

©2015 Sayantan Ghosh, J Febin Prabhudass All Rights Reserved

Correspondence: ghosh@sayantan.london

This is a work in progress material without any guarantee, either explicit or implicit, including, without restriction, the guarantee of title or the implicit guarantee of future marketability or suitability for any purpose whatsoever. The contents of this page are not even any sort of advice.

Any use/reproduction in part/as a whole without permission amounts to offense punishable under law.