(9-13)Summary of Group Meeting (09.13.2013)

v Discussions on manuscript_PLOS Computational Biology
According to Dr Raj’s suggestions, Ping’s comments, and new published literatures, our manuscript need to be future edited as following:
1. Two NAM inhibition mechanisms were observed (as shown in figure below): (A) during substrate (NAD+) binding and (B) through base exchange reaction.
Substrate binding vs Base Exchange_2.jpg
In our manuscript, we want to address is the following issues:
(a) our computational studies focus on the inhibition mechanism A for both Sir2 and SIRT3 by evaluating the binding modes of NAD+;
(b) Mechanisms A and B may co-exist, and NAM plays different roles.
(c) Mechanism B is specific to NAM, and isoNAM activation is achieved by NAM inhibition relief in this mechanism.

2. Analysis of competitive & noncompetitive inhibition modes
(a) A concept of “mixed noncompetitive inhibition” was introduced.
(b) For Sir2: our simulation results shows the MM-GBSA/MM-GBSA with MCMM scores of NAD+ for AB and AC pocket are both significant (-95.1 and -99.0 kcal/mol in MM-GBSA, and -26.6 and -42.1 kcal/mol in MM-GBSA with MCMM), which indicated NAM binding did not affect NAD+ binding much. In this case, the mixed noncompetitive inhibition is close to pure noncompetitive inhibition.
(c) For SIRT3: MM-GBSAAC>> MM-GBSAAB (refer to -109.3 and -77.2 kcal/mol, and -50.7 Vs. -14.0 kcal/mol in MM-GBSA with MCMM) indicated that the binding of NAD+ to AB pocket is unfavorable compared to other binding modes. The binding of NAM in AC pocket significantly decreased the ability of NAD+ binding to AB pocked. In this case, the mixed noncompetitive inhibition acts close to competitive inhibition.
(d) New table will be included.
MM-GBSA scores_Sir2 and SIRT3.jpg
3. PNAS paper_Gertz et al. Ex-527 inhibits sirtuins by exploiting theunique NAD+-dependent deacetylation mechanism. PNAS (2013) E2772-E2781.
(a) Need to structurally confirm if and how
i. Ex-527 interacts with NAD+ in Sir2 and SIRT3
ii. Ex-527 interacts with intermediate in Sir2 and SIRT3
(b) In the manuscript, we need to address that Ex-527 noncompetitively inhibits SIRT3 by a different inhibition mechanism. Ex-527 occupies the NAM site and contacts the ribose of NAD+ or the coproduct 1O-ADDP ribose. The direct contact of Ex-527 and intermediate stabilized the Ex-527-substrate-enzyme complex, therefore inhibits the enzyme conformational change for releasing product.
(c) Take home lesson is interaction with the intermediate, which is undesirable for activator design. Thus if we are screening molecules that might eventually be used as leads for activators, we want to screen out the type of binding that was similar toEx-527.
4. XG_provide a time frame for experimental work in the lab in terms of
(a) Investigate inhibition mode of existing potent inhibitors, such as salermide (IC50_SIRT3=27uM), AC93253 (IC50_SIRT3=18 uM). May repeat Ex-527 to see if our results agree with reported data. How long will take, how much money need to spend for purchasing Fluor-de-Lys kit
(b) Continuous assay, how far need to go, if we have it how long for each assay and how much money will spend
XG(9-20):
Schedule_ongoing experiments. (gray is done.)
(1) Continuous enzyme coupled assay.
  1. PCR amplification of PNCA from Samonella genomic DNA (Primer design, PCR condition optimization, …)
  2. Subcloning into appropriate expression vector-pGEX6P3. (TOPO cloning, transformation of TOP10 chemical competent cell, identification of positive clones, Mini and Midiprep for sequencing, Digestion, Alkaline phosphatase, Calf Intestinal_CIP for Ligation, …)
  3. Plasmid maxi preparation
  4. Linearize the constructs
  5. Transformation the linear construct into the appropriate host strain BL21(DE3)
  6. Confirmation of positive transformants
  7. Expression optimization and confirmation. (Screening of transformants, media formulation and inducer concentrations, induction temperature and length, culture lysate conditions, SDS-PAGE gel, …)
  8. Large-scale expression. (1 week)
  9. Purification (two weeks) and confirmation (western blot/HPLC: 1 week).
  10. Activity measurements (1-2 weeks)
(2) Study of the potency of inhibition and inhibition mode of candidates inhibitors (Ex-527, AC93253, salermide, ChemBridge 5281077, 4102009, 9147724 are currently available in the lab….)
Week 1: Inhibitor physiochemical properties
(a) Solubility in H2O, assay buffer
(b) If not soluble or have very poor solubility in assay buffer, selection of organic solvent
(c) Concentration range, pH range
(d) If has color, background minimization
Week 2: Standardization of working assay systems (Fluor-de-Lys and Continuous Enzyme Coupled Assay)
  1. Standard calibration curve
  2. Titration of developer signal stabilization
  3. Measurement of Km(NAD+) and Km(substrate peptide)
Week3 and 4: Inhibition mode of target inhibitors (Duplication is applied.)
  1. IC50 (may try 10 different [inhibitor] then narrow down for specific range.)
  2. Ki ( [NAD+]=0, 62.5, 125, 250, 500, 1000, 1500, 3000 uM with different [inhibitor].)

In summary
For inhibitor screening (IC50 measurements to show the potency of inhibition, Ki then can be calculated), it takes 1 month for testing 5 inhibitors using Fluor-de-Lys SIRT3 drug discovery kit (2 kits, ~$1200), but 10 inhibitors using continuous enzyme coupled assay (Costs of the chemical reagents used in assay, like buffer solutions, chemicals, 96-well plates et al).
For inhibition mode study (determination of competitive, noncompetitive, uncompetitive inhibition modes), it takes 4 weeks for testing 1 inhibitor using Fluor-de Lys SIRT3 drug discovery kit (3 kits, ~$1800), but 3 weeks using continuous enzyme coupled assay (Costs of the chemical reagents used in assay, like buffer solutions, chemicals, 96-well plates et al).

5. XG_look for the possibility of the outsources for crystallography
XG(9-20): Two companies and a university core facility provide such a service. The details are listed below
GenScript USA Inc.
860 Centennial Ave
Piscataway, NJ 08854
Email:[email protected]
CrystalPro Gnen –to-Structure Services
Phase I: High purity protein production
Phase II: Crystallization and structure determination
Phase III: Protein-compound complex structure determination

SHAMROCK STRUCTURES, LLC
1440 Davey Road
Woodridge, IL 60517
Email: [email protected]
Shamrock Structures provides protein crystallography services on an outsourced basis to pharmaceutical companies. Our clients engage with us to apply technology and expertise in X-ray crystallography to solve protein structures. Most often, our clients seek co-crystal structure determination, meaning their lead drug compound(s) in complex with a protein target of interest.
Shamrock Structures are advantageously located a mile from the U.S. Department of Energy’s Argonne National Laboratory and we have an agreement with Argonne to efficiently access the Advanced Photon Source (APS). The APS technology is the only 3rd generation X-ray synchrotron light source in North America and it permits the fastest protein structure determination currently possible.

X-ray Crystallographic Laboratory
Department of Chemistry
College of Science & Engineering
University of Minnesota
192C Kolthoff Hall,
207 Pleasant Street S. E.Minneapolis, MN 55455 E-mail: [email protected]
The XCL accepts samples for structural analysis from colleges and universities, as well as from industry. Clients of our facility receive by email a full report file and a CIF (crystallographic information file) which is required by most journals.

6.Ping_ Analyze the inhibitor library Xiangying has and pick the most representative molecules for experiments.

(8-12) The related references have been uploaded in Dropbox/PMC-AT Research/References.

References for "calculation of protein-ligand binding affinities" (8-9)

Flexibility and binding affinity in protein_Tuffery_2012.pdf
Binding free energy calculations and biologicaltesting of novel thiobarbiturates as inhibitors_Uciechowska_2012.pdf
Are scoring functions in protein-protein docking ready to predict interactomes_clues from a novel binding affinity benchmark.pdf
XG(7-3): The JMB cover letter is uploaded. Please feel free to modify it. JMB cover letter.docx
EK(6-27): Excel spreadsheet with the updated cross-docking for Sir2. See columns G and H.
MM-GBSA_components_Sir2_and_SIRT3.xls

EK(6-19): Answers to the following from the task list are provided

EK(6-18): Task 5, Figure: similarities/differences between (a) NAD+ template induced fit into the AB pocket of SIRT3 (3GLT) and (b) NAD+ cocrystallized in the AB pocket of Sir2 (1YC2 chain A). In this flattened 2D representation, residues within 2.8 Å of the NAD+ are represented as colored spheres, where: red=acidic, green=hydrophobic, blue=polar, light gray=(Gly). Solid pink lines are H-bonds to the protein backbone; dotted pink are H-bonds to the side chains. Solvent exposed ligand atoms are shaded gray.
Publication ready high resolution images are in the 'Pages and Files' in this wiki under the following names:
(a) 1YC2_A_NAD_cocrystallized_in_AB_pocket_of_Sir2_highlighting_residues_within_2.80_A_of_ligand.jpeg
(b) SIRT3_template_induced_fit_docking_into_3GLT_ligand_interaction_diagram_NAD_in_AB_pocket_.jpeg

1YC2_A_NAD_cocrystallized_in_AB_pocket_of_Sir2_highlighting_residues_within_2.80_A_of_ligand.jpegSIRT3_template_induced_fit_docking_into_3GLT_ligand_interaction_diagram_NAD_in_AB_pocket_.jpeg
Below table shows the sequence aligned residues that are in the above ligand interaction diagrams. Based on a ClustalW2 multiple sequence alignment of SIRT1-7 and Sir2Tm and Sir2Af2, highly conserved residues are in red (also marked by '*' in lower row), conserved residues in green (also marked by ':' or '.'). Grey highlighting shows corresponding residues that were missing within 3.0 A of the ligand.
Residues_in_ligand_interaction_diagram_comparing_SIRT3_to_1YC2.jpeg
Residues that make similar interactions in both Sir2 and SIRT3:
Highlights of differences between the Sir2 and SIRT3 ligand interaction diagrams.

Below is a full multiple sequence alignment of SIRT1 - SIRT7, as well as Sir2Tm and Sir2Af2 using the ClustalW2 algorithm. This sequence alignment is similar to the published multiple sequence alignment of SIRT1 - SIRT7 (Figure 2 in Jin, L., et al. (2009). Crystal Structures of Human SIRT3 Displaying Substrate-induced Conformational Changes. Journal of Biological Chemistry 284, 24394-24405). My alignment added the Sir2 sequences.
Multiple Sequence Alignment of Sirtuins.doc



EK (6-6): updating questions

EK (6-2): Answers to questions Raj texted Eric:
  1. In the cross docking with Sir2 in the AC pocket, do we really have a cocrystal structure with NAD?
    Yes, it is a true cocrystallized structure with NAD+. PDB:1YC2 is a multimer with 4 separate structures of Sir2. One of them, from chain B, has NAD+ cocrystallized in the AC pocket. Chains A and D are two different instances with NAD+ cocrystallized in the AB pocket.
  2. I don't see figure 7 referenced in the paper. Is that to be completely replaced by the new cross docking figure? And is fig 7 from the docking or the crystal structures?
    The existing fig 7 in the paper is from the 1YC2 crystal structure. That figure is to be replaced by the cross docking figure. The cross docking figure has two pictures: (a) the cocrystallized structure from 1YC2 chain A compared to (b) the docked structure in the AB pocket.
  3. Note also, for any validation figures, we need reported RMSD.
    For the new figure 7 with the cross docking of 1YC2, the RMSD is 1.645 A (for non H-atoms) between the cross docked structure in (b in new figure 7 below) and the cocrystal structure from 1YC2:chain A in (a). There is another cocrystal structure with NAD+ in a slightly different conformation (the nicotinamide end in the B pocket is flipped) in 1YC2 chain D. The RMSD between the cross docked and this other AB pocket structure is better at 0.718 A. The ligand interaction diagram for the cocrystal structure (a in the new figure 7) uses chain A (with an RMSD of 1.645) because this is the structure that gives a lower energy, and is closer in energy to the in place score for the AC pose (from 1YC2 chain B). I guess I could have just as well created the ligand interaction diagram from chain D, which would have even more identical intermolecular protein-ligand interactions between the cocrystal and cross docked. Below is a picture of the superposition. (LEFT) RMSD 1.645 A. Blue is the cocrystal of NAD+ from 1YC2 chain A; orange is the cross docked. (RIGHT) RMSD 0.718 A. Red is the cocrystal of NAD+ from 1YC2 chain D; orange is the same cross docked.
    aRMSD_cross_dock.jpegRMSD_cross_dock2.jpeg


RC (6-2): I have edited the manuscript considerably in the computational results and discussion sections. I have accepted changes in the draft below in order to make these sections more readable, but I will use the earlier draft for further revisions of the other sections, which I have not touched. In this draft, I have organized the computational discussion section, and have copied the most important parts of our earlier correspondence directly into the newly organized section. My next step will be to use this correspondence to start writing the revised computational discussion section. Before I can do that, though , I need some more info.


Draft of JMB_060213C.docx

For Mon, in addition to the new figure Eric is working on, and his answers to the unanswered questions I previously asked, Eric needs to answer the following list of questions, which I excerpted from the revised draft above to save him time (a couple of these are related to this new figure). (Recall I indicated that I would have more questions during the process of paper revision.)

1) There needs to be a simplified table with the most important docking scores in the Results section. The current xls table is too big.

2) Results section - "Figure 5: The best docking starting structure for SIRT3 is 3GLT, which has the thio-intermediate of the acetyl-lysine peptide. The NAM has been cleaved and a bond to the thioacetyl is trapped. SIRT3, 3GLT with the trapped thio-acetyllysine ADPR intermediate. The B and C pockets are unoccupied because of the intermediate. H-bonds between the ADPR and the protein residues within 3.0 Åof the ligand are shown here."
Decide whether to mention similar results with 4FVT AC pocket here. Need RMSD between NAD docked to 3GLT AC pocket and 4FVT xtal.

3) THERE ARE NO FIGURES OF SIRT3 AB POCKET IN THE RESULTS SECTION; ONLY TABULATED RESULTS. DECIDE IF OK TO SHOW SIRT3 AB ONLY IN DISCUSSION

4) IN SIR2 AF1, IS IT NAD+ OR ANALOG IN THE AC POCKET? NEED RMSDS BETWEEN XTAL AND DOCKED FOR VALIDATION
ADD: INSERT SHORT PARA ON SIR2 CROSS DOCKING. CROSS DOCKING SHOWS ROBUSTNESS TO SMALL DIFFERENCES IN COORDINATES. FIRST CHECK WHETHER NAD+ OR ANALOG IS IN AC POCKET. THEN, SIRT3 AC POCKET DOCKING VALIDATED BY ABOVE TECHNIQUE (MORE RELEVANT TO 3GLT THAN 4FVT; 3GLT DOCKING COMPARISON TO 4FVT WOULD BE ANOTHER VALIDATION IN ITSELF. HERE, JUST MENTION THE RMSD.)

5) ADD: HOW 4FVT AC POCKET DOCKING CORROBORATES 3GLT RESULT (RMSD ONLY)
With the newer 4FVT structure, a figure showing the NAD+ induced fit docked into AC pocket of 3GLT compared to the cocrystallized structure of carba-NAD in the AC pocket in 4FVT would be another type of validation of the method. THIS WILL BE HANDLED BY REPORTING RMSD ALONE - see above

6) "The backbone moved only for residues 320 to 324, while the remaining backbone for other residues were either constrained or did not move." CAN WE PROVIDE A BACKBONE RMSD?




7) INSERT AND DISCUSS ERIC’S NEW FIGURE SHOWING COMPARISON OF AB POCKETS FOR SIR2/SIRT3. IF HE HIGHLIGHTS SPECIFIC INTERACTIONS, MENTION THEM. MENTION RMSD, BUT DO NOT MENTION SCORES UNLESS WE CAN SHOW THESE INDIVIDUAL INTERACTION ENERGIES ARE SIMILAR IN MAGNITUDE IN SIR2/SIRT3 AB.

8) "Both Sir2 and hSIRT3 make similarly energetically favorable interactions in the AB pose, as well as in the AC [EK2] pose. The adenine and diphosphates have similar intermolecular interactions in the A pocket, especially with conserved residues. For example, conserved residues SER193 and SER321 form critical contacts with a phosphotidyl oxygen in Sir2 and SIRT3, respectively. As with the NAM in the C pocket cocrystallized structure of Sir2 and the docked structure in SIRT3, the carboxamide at the nicotinamide end of NAD+ in the AC binding mode makes a crucial hydrogen bond with Ile102 and Ile230 in Sir2 and SIRT3, respectively." CHECK IF THESE ARE HIGHLIGHTED IN ERIC’S NEW FIG
See above for answer to this task.

9) IN THE LAST 5 COLS OF NEW TABLE IN RESULTS. HOW DOES PROTEIN FLEXIBILITY ALLOWED IN TWO COLUMNS COMPARE TO THAT IN THE TEMPLATE INDUCED FIT ALGORITHM?

10) DESIRABLE TO SHOW A SEQUENCE ALIGNMENT OF SIR2/SIRT3 AROUND AB POCKET HIGHLIGHTING THE HOMOLOGOUS INTERACTIONS AND ALSO THE RESIDUES RESPONSIBLE FOR DIFFERENCES IN BACKBONE STRUCTURE.
See above for answer.


EK(5-28) to do list based on today's discussion with Raj:**
  1. Post updated table of MM-GBSA energy components with the fixed protein structure
    • fixed protein structure had the overlapping H-atom minimized. NAD+ binding pockets remained frozen.
    • include updated MCMM scores for Sir2 and SIRT3 AB/AC poses.
    • updated spreadsheet below.
    • MM-GBSA_components_Sir2_and_SIRT3.xls
    • Corrected MM-GBSA summary from spreadsheet
(kcal/mol)
 AB
 AC
SIRT3 MM-GBSA corrected 3GLT protein
 -82.4
 -105.4
SIRT3 MM-GBSA with MCMM for 3GLT
 -75.3
 -95.2

  2. Figure: cross docking for Sir2. Side-by-side comparison of NAD+ cocrystallized in AB pose from 1YC2 with the cross docked NAD+ into the AB pose. The cross docked structure is the NAD+ docked into the Sir2 AB pocket starting with the Sir2 AC cocrystallized structure
    • Figure X: Intermolecular protein-ligand interaction diagrams of NAD+ in the AB pockets of Sir2Af2 comparing (a) the cocrystallized structure from 1YC2, chain A, with (b) the cross docked structure. In this flattened 2D representation, residues within 2.8 Å of the NAD+ are represented as colored spheres, where: red=acidic, green=hydrophobic, blue=polar, light gray=(Gly). Solid pink lines are H-bonds to the protein backbone; dotted pink are H-bonds to the side chains. The cross docking method produced a pose with very similar intermolecular interactions as the cocrystallized structure. The lack of a protein "pocket" line around the NAM end and the grey spheres around those atoms indicate that the NAM end is exposed to solvent in the B pocket. In (a), there are no specific intermolecular interactions between the protein and the NAM end of NAD+. In (b), however, the NAM is slightly less solvent exposed in the B pocket of the cross docked structure due to the docking process lacking explicit waters in the B pocket.
      NAD_in_AB_pocket_of_Sir2_1YC2_Chain_A_xray.jpeg
      NAD_AB_pocket_of_Sir2_1YC2_Chain_B_cross_docked.jpeg
  3. Post the protein reorganization energy for SIRT3 3GLT for AB vs. AC NAD+ poses. Energies should be using the MM-GBSA solvation model, which can be found in the "receptor energy" output from the MM-GBSA calculation. The AB protein energy is from the template induced fit MM-GBSA calculations, and the AC protein energy is from the standard induced fit (NOT template) docking / MM-GBSA energy. All are from the 3GLT crystal structure.
    • From last column of spreadsheet, the receptor energy for the crystal structure of 3GLT with NAD+ docked into the AC pocket with minimal minimization (to within 0.30 A RMSD of crystal structure coordinates) is: -11592.02 kcal/mol. This is 812.54 kcal/mol lower in energy than the template induced fit structure needed to dock NAD+ into the AB pocket (-10779.49 kcal/mol)

    (4) and (5) only after above 3 finished.
  4. 4FVT: dock NAD+ into the AB pocket of 4FVT. First try standard induced fit, then template induced fit.
  5. Figure: similarities/differences between NAD+ template induced fit into the AB pocket of SIRT3 (3GLT?) and NAD+ cocrystallized in the AB pocket of Sir2 (1YC2). Not sure we need this figure. The simplest measure of similarity is the RMSD between the NAD+ from SIRT3 and Sir2 backbone superimposed structures.

EK(5-28) Meeting schedule:
  1. Review of the latest postings: what do the new corrected energies show?
  2. Discussion of what we need to show in the discussion: validation vs. differences.
    1. I suggest we focus on validating the SIRT3 AB pocket score
    2. then focus on differences between SIRT3 AB pocket (from template induced fit into 3GLT or 4FVT) and Sir2 AB pocket (from cocrystallized structure in 1YC2)
  3. Discussion of figures: I think the relavent figures should be focused on points 2.1 and 2.2 above.
  4. Amend the timeline for remaining work:
    1. Post the full table of MM-GBSA energy components for the calculations redone with the fixed overlapping atoms from 3GLT template induced fit.
    2. Redo a more full calculation with this fixed structure - may not be necessary.
    3. Do a cross docking with 4FVT - do the template induced fit docking of NAD+ in the AB pocket of 4FVT (same as done with 3GLT). This may make more sense for our discussion - as it is a true cross docking, and, like with the Sir2 1YC2 cross docking studies, we could compare the cross docking of NAD+ in the AB pocket with the co-crystallized score in the AC pocket.
    4. Create figures: which ones depend on our discussion today.



EK(5-28) Correction: references to 3GLR below are supposed to be 3GLT. 3GLT is the structure with the trapped ADPR intermediate because of thioacetyl lysine substrate, while 3GLR is the hSIRT3 with only the actyllysine substrate. 3GLT was used for the template induced fit docking of NAD+. I switched the codes once and the switch then propogated. Now corrected below. Also needs to correct in paper draft.

Rest of updates being written now.



EK(5-23) I am posting what I have so far today because I promised to update the wiki by 6pm. I still have to find and put numbers in the table (see notes), so, you might want to wait until tomorrow to look at this. Below is the breakdown of the MM-GBSA scores (all the components) for the Sir2 in place scores (AB and AC from the co-crystallized structures of 1YC2 chains A and B), the SIRT3 template induced fit scores (AB and AC pockets for NAD+ in 3GLT), and the newer in place scores for NAD+ in the AC pocket of 4FVT. Some notes also.

XG(5-22) The Table below is the summary of MMGBSA scores reported in current draft (not include new structure 4FVT). The Sir2 docking studies were started with PBD:1YC2. And old SIRT3 structure is 3GLT. Eric, please double check the cross docking scors of Sir2 for AC and AB pocket: the ones show at ACS presentation are opposite to the ones in the manuscript.
MMGBSA scores.jpg
EK (5-22) The above table clarifies all the scores. Two comments: the SIRT3 (3GLT) scores of -107.9 and -84.4 are not "in-place scores"; they are the induced fit docked MM-GBSA scores. The two scores for SIRT3 (4FVT) are as close to an "in-place score" as possible with the transformed carba-NAD into NAD - so they can be considered in place scores. You can decide whether you want to report both of the 4FVT in place scores;
RC (5-23): Ok, so we have not redone the induced fit calculations in the AB pocket with 4FVT. That is ok, just confirming. See also my comments below - we would need to compare 3GLT AB pocket scores with 4FVT AC pocket scores. Which raises the question of the aim of including 4FVT scores in the paper. Is it only to validate those obtained with 3GLT?
EK(5-28) No, we have not done induced fit AB calculations with 4FVT. Although it might be a good idea. I could try induced fit calculations with multiple structures of SIRT3 - and, hopefully, if all of these structures result in higher energies for the AB template induced fit structure, this would fit with our hypothesis. Yes, the 4FVT scores do validate those obtained with 3GLT. It is unfortunate that we don't have a xray structure with NAD+ in the AB pocket in SIRT3 for a more direct comparison. The difficulty with our calculations that show that AB pocket binding is higher in energy than AC in SIRT3 is this: what if we are not sampling enough to accomodate the NAD+ correctly into the AB pocket? We are sampling a lot with the prime/plop algorithm, but really only within 7 A of the NAD+. These algorithms are very good at comprehensively sampling in this volume, especially for the side chains. But we are not exhaustively sampling.

EK (5-21): SIRT3 simulations results with new crystal structure:



RC (5-19): Further comments and information needed for 1st sirtuin paper revisions:


You can see why I was very confused by this writeup. Please provide your answers below my questions; I need some of these answers to complete my revisions. Please do not make any revisions directly in the paper - I will use the necessary info/clarifications you provide to make my edits. Thanks.

After my upcoming revisions, we may ask for the following additional info from Eric: a) final versions of the figures; b) docking results with the new SIRT3 structure if he has finished them and the results are conclusive; c) possibly, a ligand interaction diagram for SIRT3.


Eric's discussion sections:


“In addition to the investigation of NAM and isoNAM, protein-ligand docking using the newly available human SIRT3 crystal structuresfrom 2009 - 2013 52; 53agreed with previous crystal structures and provided insights into the differences...”

WHAT PREVIOUS CRYSTAL STRUCTURES? WHY ARE WE EMPHASIZING THIS IN THE FIRST SENTENCE? YOU MENTION DIFFERENCES BETWEEN SIR2 AND SIRT3 IN THIS SENTENCE, BUT THE REST OF THE PARAGRAPH SAYS NOTHING ABOUT DIFFERENCES. INSTEAD, ALL YOU TALK ABOUT ARE THE SIMILARITIES!

(EK) What I was trying to establish is that the docking poses for SIRT3 in the AC and AB pockets is similar to observed crystal structures with Sir2 and others, and not some different pose that would not be reasonable. There is a later paragraph that talks about the differences in the scores between AB and AC docking for SIRT3. I focused on the similarities of intermolecular contacts first to establish that the docking methods produced reasonable structures. There are also differences in intermolecular contacts that could be mentioned, but I did not because it is hard to figure out whether these difference are artifacts from the induced fit docking. It is also difficult to then attribute any one of these differences as the most important contributions to the difference in the overall docking score between the AB and AC docking.

The specific structures that this sentence refers to: SIRT3 is 3GLT and 4FVT; Sir2 is 1YC2 and 2H4F. Also, given the new SIRT3 in place scoring with the recent 4FVT structure, this paragraph can be changed to include that information, rather than only discussing induced fit docking / MM-GBSA scores into 3GLT.
RC (5-23): Regarding 4FVT, see my comments above. Regarding Sir2, when is 1YC2 referred to and when is 2H4F referred to?
EK(5-28) the paper refers specifically to 1YC2. 2H4F was used only as a reference, but is not in any of the figures or specifically in the text. The 2H4F is fo the Sir2Tm species, while 1YC2 is for Sir2Af2. 1YC2 is used because it has both the AC and AB poses for NAD+.

Overall, I think that this discussion section can be broken down better into 2 parts:

RC (5-23): I agree, we definitely need to distinguish between validation and differences, and the validation should be shorter - or, it should come after the emphasis on differences. Otherwise, the message gets lost, and it appears as if we are unsure of our results. Also, if validation comes first, we will need to indicate clearly upfront that we are discussing similarities in order to validate the method, and we will be talking about differences later. I will take care of all this editing based on answers to the questions above.

“Although the overall binding scores were different, the protein-ligand intermolecular interactions were similar for Sir2-NAD+ and SIRT3-NAD+, and the docked interactions agreed with cocrystallized x-ray structures of Sir2 and SIRT3” WHICH STRUCTURES? IN WHAT WAY WERE THE RESULTS NONTRIVIAL IF CRYSTAL STRUCTURES WERE KNOWN? NEED MORE INFO ON WHICH DOCKED STRUCTURES ARE BEING COMPARED TO WHICH X-RAY STRUCTURES

(EK) Yes, I was more specific here, and some additional clarification would also help. The rest of the paragraph names specific intermolecular contacts that are the same between the Sir2 (shown in Figure 7) and the SIRT3 docked structures. What is probably needed here is a different set of figures - figures that depict the similarities between the Sir2 crystal structures and the docked SIRT3 structures. Figure 7 is not enough. Figure 7 should be changed from only showing the ligand interaction diagram for Sir2, to a figure showing the AC pose for the Sir2 crystal structure compared to the pose for the AC docked SIRT3 structure. Similar H-bonds or intermolecular contacts would be highlighted. Another set of figures could be done for the AB pose. Even better would be to compare the carba-NAD+ in the AC pose of SIRT3 to the SIRT3 that was docked.

RC (5-23): I agree the figures would be useful - see also my comments below.

Here, again, the cocrystallized structures for Sir2 is 1YC2 and SIRT3 is 3GLT and 4FVT. With the new 4FVT crystal structure and the NAD+ analog in the AC pocket, we have in place scores available from co-crystallized structures for the following:
Pockets -->
 AB
 AC
Sir2
 1YC2 chain A
 1YC2 chain B
SIRT3
 NONE
 4FVT
The SIRT3 3GLT structure is the trapped intermediate, and not exactly an AB or AC cocrystallized structure. So the key argument now (with the new 4FVT cocrystallized structure) is how reliable is the docking and scoring for NAD+ in the AB pocket of SIRT3, for which we have no close cocrystallized structure. The docking structure we have is from this template based method due to the steric clashes in the collapsed B pocket from the 3GLT structure. The main question to answer is if this template based method with the plop/prime induced fit is doing sufficient sampling to find a low energy structure for NAD+ in the AB pocket. This is an important point.

RC (5-23): Ok, I believe the answer to this question is yes, since you have indicated that the induced fit poses have been validated.
My main point above is that whenever we refer to a docked structure comparison to a xtal structure, we need to either refer to a figure (as you mention above) or to specific intermolecular interactions - and if we do not have a figure, we need some way to classify the docked structures (e.g. by some naming system) so that when we refer to them, it is unambiguous which docked structure we are referring to. If we use figures, we could decrease the length of this validation discussion. I assume that in that case, you feel we could just refer to those figures without discussing the specific similar contacts one by one?
EK (5-28) sounds good to have a better naming system.

RC(5-23) Please let me know which method you prefer and if you are planning to make to figures, when you could make them. If it takes too long to make figures, I suggest we just use the approach of unambiguously referring to the structures and similar interactions as noted above. Can you list those here? I would like to have these for a back up plan in case the figures take too long.
EK(5-28) We could have 2 sets of figures for validation:

RC(5-23) In any case, due to space limitations, let's bear in mind that we should not flood the paper with validation-related figures. Figures that are referred to in the context of validation should also be referred to when describing the differences between Sir2 and SIRT3. If we make new figures, ideally, we would have one that can be used to show the AC pocket similarities and one that can be used to show the AB pocket differences. Should I assume that none of the figures currently in the results section are pertinent to this discussion, since we do not refer to them? If so, please explain why - is it because you want to highlight similar interactions, and that would clutter the existing figures, or do we simply not have the right structures shown side-by-side? This is also important to know for my revisions.
EK(5-28) We need to talk about what figures you think are best. I suggested two sets of figures just above. The most important figure, though, would explain why hSIRT3 blocks the B pocket. Figure 8 kind of does this, and it could be improved. The question to answer is are there more bulky side chains in the B pocket of SIRT3 than Sir2?

“The similarity helps validate the induced fit docked poses of SIRT3, because the same protein-ligand contacts with conserved residues seen in the Sir2 crystal structures are observed in the SIRT3 docked structures as well”

THIS INTERRUPTS THE FLOW. ARE YOU TRYING TO SAY TWO THINGS SIMULTANEOUSLY – VALIDATION, AND SIMILARITY OF INTERACTIONS? IF YOUR PRIMARY POINT IN THIS PARAGRAPH IS VALIDATION, I WILL NEED TO STATE THAT UPFRONT, AND THE FIRST SENTENCE SHOULD SAY NOTHING ABOUT EXPLANATION OF THE DIFFERENCES BETWEEN SIR2 AND SIRT3.

(EK) similarity in intermolecular contacts is additional "validation" of the docked poses SIRT3. See the above proposed outline for this discussion section that has a validation section (to which this sentence belongs) and a subsequent "differences" section.
There are two validations: the cross docking for Sir2, and the fact that the poses for the docked SIRT3 structures are similar to the Sir2 structures. The SIRT3 structures are also different in some ways, but when I looked at the break down in the scores, it is difficult to attribute specific intermolecular contacts to the differences in score for SIRT3 AB vs. AC. Yes, the paragraph can be better phrased.

“Figure7 depicts the protein-ligand interaction diagram of Sir2Af2 cocrystallized with NAD+ in AB and AC pockets. Both Sir2 and hSIRT3 make similarly energetically favorable interactions in the AB pose, as well as in the AC pose”

ALTHOUGH YOUR COMMENT TO THE SIDE INDICATES THE PARAGRAPH IS ABOUT CROSS DOCKING VALIDATION, THIS PARAGRAPH SAYS NOTHING ABOUT SIR2 CROSS DOCKING! MAYBE IT WAS SAID IN THE RESULTS, BUT THAT NEEDS TO BE REPEATED BRIEFLY HERE.

(EK) Figure 7 would be better used to explain the Sir2 cross docking validation. In this paragraph I focused on how the SIRT3 docked poses were similar to the Sir2 crystal structures - so adding the cross-docked Sir2 into the mix would not be helpful in this paragraph. That comment was from a previous version of the paper and is not valid for this paragraph.

So Fig. 7 could be used with an additional paragraph (as outlined above) to talk about how cross-docking validates the methods. Specifically, Fig. 7 (which are from the cocrystallized structures of 1YC2) could be used to compare how similar the intermolecular contacts are between the crystal structure and the cross docked structure. I've done this - there are differences and similarities. The emphasis would be that the important intermolecular contacts are also made in the cross docked structures.

RC (5-23): Ok, do I have all the info needed to add this paragraph? Do you suggest referring to any particular contacts, or simply saying that the contacts are similar between the cross docked and crystallographic structures?
Are we talking about the current Figure 7, without any revisions?
EK (5-28) I think that figure 7 should be revised just to validate the AB cross docking. Side by side figures of a) figure is the NAD+ docked into the AB pocket from the AC cocrystallized structure b) the cocrystallized structure of NAD+ in the AB pocket. We should talk about what we're trying to do here.



Another figure is needed to compare the similarities between crystal structure Sir2 (1YC2 chains A and B) and the docked SIRT3 structures (docked into AB and AC pockets in 3GLT). But, with the newer 4FVT structure, a figure showing the NAD+ induced fit docked into AC pocket of 3GLT compared to the cocrystallized structure of carba-NAD in the AC pocket in 4FVT would be another type of validation of the method.

Again, an important part is to validate the result of the AB templated based induced fit docked structure into 3GLT for which there is no cocrystallized structure. A figure might help with this - comparing AB cocrystallized structures of Sir2 to this docked AB structure in 3GLT. You can think about which figures would be most important in the paper, because we are limited in the number of figures.

RC (5-23): Yes, see my comments above. The AB pocket validation is most important. Are you suggesting to have one new figure for AB pocket validation and one new figure for AC pocket validation?( If we use 4FVT to validate AC pocket docking, we need to be careful in explaining why we are using 4FVT for AC but 3GLT for AB. See my earlier comments on that issue.)
EK(5-28) Yes, exactly. We need to talk about what to do here.

"The adenine and diphosphates have similar intermolecular interactions in the A pocket, especially with conserved residues. For example, conserved residues SER193 and SER321 form critical contacts with a phosphotidyl oxygen in Sir2 and SIRT3, respectively. As with the NAM in the C pocket cocrystallized structure of Sir2 and the docked structure in SIRT3AND CARBA-NAD?, the carboxamide at the nicotinamide end of NAD+ in the AC binding mode makes a crucial hydrogen bond with Ile102 and Ile230 in Sir2 and SIRT3, respectively."

THIS IS POORLY WRITTEN PARTLY BECAUSE IT WAS NOT REVISED AFTER I INDICATED THE IMPORTANCE OF STERIC CLASHES AND PROTEIN REORGANIZATION. WHY DON’T YOU SAY ANYTHING ABOUT THE –DIFFERENCES—BETWEEN SIR2 AND SIRT3? HOW ABOUT STERIC CLASHES? THIS IS ONLY MENTIONED BELOW IN A SEPARATE PARAGRAPH? I WILL NEED TO OVERHAUL THIS PARAGRAPH. ALSO, I NEED INFO ON THE COMPARISON OF THE NAD+ AC POSE TO THE EXPERIMENTAL SIRT3/CARBA-NAD+ STRUCTURE.

(EK) again, this sentence is about the similarities as part of the validation. This sentence figures into the outline above. There can be additions added to talk more about the differences. Steric clashes of docking NAD+ in the AB pockets of SIRT3 3GLT are mentioned in later in the discussion section; the clashes are talked about in the context of why the template based induced fit method was needed. The steric clashes could also be used to talk about how the energy for the AB docking is less favorable, but a careful look at the components of the MM-GBSA score for those residues would be needed. Does the template based induced fit method does move these residues out of the way, but do they still contribute to less favorable energy?

There is a paragraph latter on about the protein reorganization energy.

RC (5-23): Yes, I saw that. I need to move that content up. However, the discussion is ambiguous about the causes for the differences in AB and AC pocket binding affinities. It does not clearly say whether the protein reorganization energy is primarily responsible for the differences in binding affinities. Importantly, although the binding scores are different even without the reorganization energy, there is no clear statement about where this difference is coming from. Especially given the amount of space dedicated to validation, the discussion of differences is particular short and seems like an afterthought.
EK(5-28) Understanding the difference in scores between the AB and AC pockets will come from an analysis of components of the updated MM-GBSA score spreadsheet. I will post and look at later, and could use your help understanding it.

"Predicted NAD+binding scores are similar for AB and AC binding modes for Sir2, but in SIRT3 the binding score for AC is lower than for AB binding. Although the cumulative scores are different in SIRT3, analysis of the individual energy terms…." HOW AM I SUPPOSED TO UNDERSTAND THIS WITHOUT THE NUMBERS HERE? PLEASE PUT THEM IN A TABLE. WHAT IS THE DIFFERENCE BETWEEN SUM OF INDIVIDUAL SCORES AND THE CUMULATIVE SCORE?

(EK) I am putting these numbers together for you in a large table. This table has the breakdown of components of the MM-GBSA scores for Sir2 in place scores and SIRT3 docked scores and the new SIRT3 4FVT in place scores. I will have this table for you tomorrow by 6pm.

RC (5-23): Ok thanks. I think the protein reorganization energy should also be included in the table. But before I look at that, can you clarify why there is an ambiguity - do you mean that the cumulative reported score is not the sum of the individual scores? That is the only way I can think of that the analysis of the individual terms does not explain the difference. This was one source of my confusion here.
EK(5-28) the cumulative score is the sum of the individual scores. i.e., coulomb, covalent, H-bond, vdW, etc... for each of the parts of the overall energy:
dG_complex - [ dG_freeLigand + dG_freeProtein ]
…to the binding score for MM-GBSA does not explain the difference between the Sir2 and SIRT3 scores, possibly because molecule’s large size.



"The largest part of the NAD+ molecule containing the adenine and the diphosphates is bound in a similar conformation in both the AB and AC modes for both Sir2 and SIRT3, masking individual energetic differences." WHAT IS THE MEANING OF MASKING AN INDIVIDUAL ENERGETIC DIFFERENCE?

(EK) I had a difficult time making sense of the break down of the MM-GBSA scores. I will provide the full break down of the scores for you and you can see if you can make sense of it. I could use your help in figuring out what to say here about the differences.

RC (5-23): Do you mean that the individual scores are dominated by contributions from the part of the molecule that is in a similar conformation?
EK(5-28) yes

"…. However, because the MM-GBSA scores do not include the full energetic protein reorganization penalty from the induced fit methods used with SIRT3, the less favorable AB binding in SIRT3 may be even more pronounced than reported here. The missing penalty" WHAT IS THE MISSING PENALTY? STATE IT HERE

(EK) the penalty is stated in the next sentence of > 100 kcal/mol.


SEE MY EARLIER COMMENTS REGARDING MOTIVATION FOR THE FOLLOWING PARA: I WILL NEED TO ADD IT.

"The NAM end of NAD+ adopts two flipped conformations in the crystal structures: one with the amide hydrogen of the NAM end of NAD+ in the AB pocket of the Sir2Af2 (1YC2 chain D) making an intramolecular hydrogen bond to the ligand phosphotidyl oxygen, and another with this amide pointed towards the solvent (1YC2 chain A). There are no hydrogen bonds in the B pocket with the outer half of the pocket exposed to solvent, allowing the NAM to move. This intramolecular H-bond is never seen in the docking, possibly because the energy is truly degenerate in this case where most of the NAM is exposed to solvent, or Glide has systematic bias against ligand intramolecular H-bonds. In addition, the NAM in the B pocket must move to the C pocket, and this degenerate flexibility in solvent could facilitate this motion."

(EK) you may want to leave out these sentences on the how Glide creates one pose with an intramolecular H-bond.

IF THERE WERE ANY PARAGRAPHS WRITTEN FOR THE PREVIOUS VERSION OF THE MANUSCRIPT THAT REFERRED TO HOW BINDING AFFINITIES ARE PREDICTED USING A TRAINING SET OF EXPERIMENTAL DATA, WITH EITHER LIA AND/OR MM-GBSA METHODS, PLEASE POST THAT IN A WORD DOC HERE, SO I CAN REUSE PARTS OF IT.

(EK) There was no paragraph written about this in the paper draft. This was presented in the ACS presentation; there are not extensive notes in the presentation about this; I verbally explained it during the presentation.
Here is the final ACS presentation:
Final_ACS_presentation_04062013.pptx

RC (5-23): Ok, in that case please point me to the relevant methods sections of the papers that you followed in carrying out these regressions, so I can summarize the methods. Are those papers on the wiki?
I assume that there were no verbal notes on this in the notes section of the ppt.
EK(6-18) Answer is above.


ALSO, PLEASE PROVIDE THE MCMM ENSEMBLE SCORES. I WOULD LIKE TO HAVE THE OPTION OF RETAINING THE MCMM PRESENTATION IN THIS PAPER. IF YOU SUGGEST OTHERWISE, PLEASE INDICATE WHY.

(EK) MCMM calculations resulted in conflicting numbers that did not agree with our overall hypothesis. I can look up the specific numbers again and post them if you wish.

RC (5-23): I'd like to confirm that Karthik has received a detailed protocol describing how the MCMM calculations were done, and that the earlier drafts of the paper contain all the relevant MCMM results; we wouldn't want to lose all this work that was done.

Xiangying's sections:


"Defining the inhibition modality is important for making quantitative comparisons among inhibitors of sirtuins, and is necessary for calculating the enzyme-inhibitor dissociation constant, Ki, from the experimental assays. Ki and the related protein-inhibitor Gibbs free energy of binding provide a means of defining the energetic contributions of specific types of interactions between groups on the enzyme and functionalities on the compounds to the overall binding energy of interaction."
AS DISCUSSED, I WOULD LIKE SOME MORE DETAILS ON WHETHER THE Ki IS NEEDED TO OBTAIN BINDING FREE ENERGIES, GIVEN THAT ERIC APPEARED TO BE USING IC50's. CAN YOU GET THE Ki FROM THE IC50?

XG(5-20): Kd for inhibition (Ki) can be directly related to the free energy of binding to the specific enzyme form as DGbinding = RT ln(Kd). Therefore, rational comparisons of inhibitor affinity for a target enzyme are best made by comparing the dissociation constants for the varying inhibitors.

Affinity for a target enzyme is not the only one criterion used to judge the suitability of an inhibitor for use in human medicine. Another equal important criterion is selectivity of the compound for the target enzyme relative to other structurally or mechanistically related enzymes. Thus the affinity of a compound among a number of potential enzyme targets commonly will be compared. Then any meaningful comparison of inhibitor affinity among these enzymes should be based on Ki value.

By knowing inhibitor modality (competitive/uncompetitive/noncompetitive), one may predict if the inhibitor has tight binding to a certain enzyme. For example, one compound will bind much tighter to its target enzyme with competitive inhibition mode than with noncompetitive inhibition mode. The reflection of Ki value of the compound to a certain enzyme, is Ki (competitive) < Ki (noncompetitive).

We can calculate Ki values from IC50 values using equation for enzyme-substrate and target-ligand interactions by different inhibitory mechanisms.


EK(5-15)
Latest draft below. Corrected most of comments from Raj, and incorporated Xiangying's latest changes. The major issues were not addressed in this draft. I am available to answer any questions you have as you finish the draft.
Draft of JMB_051513.docx


XG (5-15) Eric, to save time, could you please make the following changes in your final draft? Thanks!
(1) NEW Title: Mechanism of inhibition of the human sirtuin deacetylase SIRT3: computational and experimental studies
(2) Introduction_ first paragraph: Change "Sir2 was required for lifespan extension by calorie restriction" into "Sirtuins, the highly conserved enzyme homologues of the yeast Sir2, have been implicated in aging and the regulation of metabolism and genome stability."

EK(5-15) OK. I will transfer all data you need and will be available for any questions you have as you finish the draft. I will need to read the final draft you create, so please forward that to me before you submit it. I'm correcting more things in the paper and will post the latest draft by 5pm, but there is not time to make any major changes.

RC (5-15): We've discussed this internally and come to the conclusion stated below under my previous update. As noted, I will take the latest draft starting today and finish it myself, asking for data when I need it. Please post your final updates by 5 pm today. I will take whatever is posted by 5 pm today and revise it for publication, without iteration. I will not use a draft submitted after 5 pm. Eric can finish the work with the newer cocrystallized structure if he likes, and I will determine with him whether/how to include it. That will be the last computational work - we have decided to phase out the computational part of the project after that, as I described below.


EK(5-15)
There might be a misunderstanding of what this draft is: a draft. It is by no means ready for publication. My task was to create a working draft of the paper that can be crafted into the publication after iterative feedback from Raj. A part of the process is for Raj to make the important comments and suggestions he did below. Raj has found a number of grammar mistakes and oddly written sentences, as expected in a 30+ page document. A fresh pair of eyes reading the paper is important to point these things out. There are a number of major issues to address before this paper is ready, some of which Raj commented on below. For example, Raj has given past feedback and direction to include a lot of material, and some sections of the paper and certain paragraphs need better integration. Also, we need to account for the newer cocrystallized structure of SIRT3 (4FVT) with the NAD+ analog in the AC pockets. We've worked very hard on this paper and are in the final stretch of this process.

I've addressed some of the comments (see blow).
Draft of JMB_051413.docx

RC (5-14):

I read through the 051013 draft yesterday.

My conclusion: this is one of the most poorly written scientific papers I have read in some time. I am concerned about putting my name on it.

Grammar is part of the problem, but by no means all of it. There is a breakdown of logic in several sections (especially in some of the newly written paragraphs), including seemingly contradictory statements, and an overall lack of clarity as to why certain paragraphs have been written at all. As someone who has been heavily in this work, my inability to understand many of the paragraphs is a source of serious concern, since it is quite unlikely that a reviewer or reader will understand them. The lack of cohesiveness suggests that the authors were not always focused on the writing process, were not closely collaborating, and were often forgetting the main point of a given subsection. The result is a paper whose overall message is garbled.

Although some issues may have been mitigated in the most recent updates, due to above and the significant delays, I have concluded that I will need to finish the paper myself after Eric finishes his final read below. After Eric's submission tomorrow, we will be phasing out the computational part of this project until later this year. In the meantime we will assess the need (if any) for computation in future papers and decide how those needs can be met. Eric, please also confirm that you have provided Karthik with copies of all the data. If I need any of this data during my editing of the paper, I will ask you for it.

Here is a partial list of sentences in the 051013 draft that I felt either needed editing or were contradictory/unclear in meaning. I am listing them here so I can refer to them while editing.

(greyed out comments have been fixed; additional comments in grey)





EK(5-14)


EK(5-10)


EK(5-9)
XG(5-10) JMB requires that "color or grayscale photographs, keep to a minimum of 300 dpi". I do not know when you save the picture you had setup the resolution for 300 dots per inch. Please let me know.
The combined the figures from 1-8 have been prepared as pdf file. Eric please check if you agree with it. Plus the resolution of Figure 7 a and b are not good. Please send me the image files.JMB_05102013 figures.pdf

EK(5-2) updated plan:

RC (5-3): Thanks Eric. Don' t wait for the last minute to finish your work, and try working during the day. If you work consistently starting now, you should have plenty of time to finish all your sections - including figures and references - by Tues noon.

EK(5-2) talk with David Churchill grad students at Kaist

EK(5-2) After working hard for 2 days on the paper, I have made a lot of progress, and the final draft will take longer. Let me continue to work on the paper without Xiangying or you editing the below posted draft. Much time was spent analyzing the ligand interaction diagrams, as well as protein-ligand docking poses to fill in the important discussion section about why NAD+ binds with the same energy for Sir2, but very different energy for SIRT3 in the AB vs. AC pose. It is difficult to understand why because there are so many energy components in the MM-GBSA score, and a cancelation of terms. So I decided to explain the difference between Sir2 vs. SIRT3 AB/AC scores as cumulative rather than from any one particular difference. It is always possible to come up with plausible individual effects, but these individual effects get drowned out by larger numbers in other MM-GBSA terms.

Other things updated are additional details about the NAM and isoNAM docked poses, important because the laboratory emphasis on these molecules. Xiangying asked if I included binding affinity of NAM or isoNAM. Yes, I did calculate the score for NAM binding using MM-GBSA, but I did not include this in the paper because the score is not a good measure of the binding affinity of NAM due to base exchange with NAM. Reporting the number could be confusing. It is more important to discuss that docking resulted in a similar binding pose found in the x-ray structure of Sir2 co-crystallized with NAM.

Another question was about why I erased the MCMM ensemble correction section. MCMM was not used in the MM-GBSA calculations.
MCMM was not used for comparing NAD+ AB vs. AC binding. A big part of the correction for MCMM is more accurately calculating the energy of the free ligand (using an ensemble of low energy free ligand conformations). But it does not matter as much for comparing the different NAD+ AB vs. AC poses because the free NAD+ ligand energy is similar in both cases. So the correction error would mostly cancel out. MCMM would still theoretically correct for some differences. Standard MM-GBSA calculates the free ligand as a simple energy minimization of the bound state. Since the AB and AC poses are different, the free ligand energy is different for the different poses. While technically incorrect, there might be some cancelation of errors. While the MM-GBSA numbers could theoretically be improved, MCMM is not as widely used as standard MM-GBSA. And the MM-GBSA numbers agree with experiment and our hypothesis. If we started correcting with MCMM, we should also start correcting for many other known limitations of the MM-GBSA approach, such as a single protein-ligand state, as well as the lack of a more comprehensive sampling technique to find the lowest energy states for the AB and AC poses. What we have works within the limitations of this MM-GBSA method.

Please let me continue to work diligently on this draft. I will post another update tomorrow morning. I think the best plan is let me continue to revise the draft without you updating the draft. I don't think you should even read this draft other than out of curiosity to see the progress.

Draft of JMB_050213_EK.doc

XG (5-2): Thanks, Dr. Raj, for editing experimental parts. I will modify the paper based on his comments. In the latste draft, a message of "[[file:/C:\Users\xguan\Documents\AppData\Local\Microsoft\Windows\Temporary%20Internet%20Files\Local%20Settings\Temporary%20Internet%20Files\Content.IE5\9N1O72K2\Documents%20and%20SettingspmclabMy%20Documentsxg%22%20l|../../../AppData/Local/Microsoft/Windows/Temporary]]" was found in the 2nd paragraph of Introduction. Can anyone explain why?

Eric ?

RC (5-1): I have gone through all of XG's sections in her latest draft below and made detailed comments and edits throughout. I am now waiting on Eric's final draft Wed night; which will need to be merged with this after he posts it. Eric, please don't make us ask again for this.

XG, you should stop doing lab experiments for the next couple of days - I noticed that your sections were not carefully proofread or edited. You need to devote full time to answering my questions and finalizing your sections. There is also some work you need to do jointly with Eric, as discussed.

Draft of JMB_050113_RC.doc

XG(4-30) Disucssed with Dr Raj and add the more comments on the draft. Draft of JMB_043013_PM.doc Eric: (1) you mentioned that your FINAL draft will be ready on Thursday morning. Dr. Raj need 2 days time to review and make change on the draft this week. Therefore, It will be appreciated if you can post it Wednesday night.

(2) Please response to ALL of the comments that Dr Raj has made.
(3) Do you delect the contents of MCMM? You have calculated binding affinity of NAD+ docking into AB and AC pockets, have you ever done on NAM?
(4) I noticed that something was wrong with the references. Some of your references were missing in the very last draft. Please fix them or give them to me I can add them into EndNote.
(5) Thank you for editing Abstract and Introdcution. Please focus on and finish your sections.

EK(4-29) Another draft. Here, edits to the abstract and introduction. Fixed grammar, sentence structure and flow of paragraphs. The edits will take time for me to go through the paper line by line, including fixing grammar, ordering, and the requested revisions of the discussion. All of Raj's comments are being addressed. Work to continue on Tuesday.
Draft of JMB_043013_ek.docx

EK(4-29) Raj, I will continue to revise this draft.
This weekend I needed to redo some of the calculations for Sir2, to verify the cross docking and in place scoring results. When I searched through the original data, the results I found were not consistent with the results in the paper for one of the in-place scoring. I knew that I had previously calculated a different number that agreed with the rest of our results, but I could not find it. It took me half a day to resolve this, which ate into time for editing the paper. In particular, there are two structures in 1YC2 of NAD+ cocrystallized in the AB pocket. The one from chain D resulted in a score of about -65 kcal/mol (not consistent with our hypothesis because AC in-place scoring would be largely different at -99.0 kcal/mol). A recalculation with Chain B (also AB co-crystallized) results in an energy (-95 kcal/mol) consistent with the hypothesis in the paper that NAD+ binds equally well in the AB and AC pockets of Sir2. Since the general method for choosing the structures is to use the best scoring (lowest energy), it is consistent to choose the -95 kcal/mol result. It is resolved.

RC (4-29): I'm seeing very few changes to the draft. The discussion section indicates "still needs work" and it appears the main change was to cut and paste paragraphs from the results and put them in the discussion section. There are spelling mistakes and grammatical issues (see for example, the last para of discussion, which also needs reordering of sentences), which indicates quick changes were made and that the paper will need multiple rounds of editing involving me. In addition, even old comments that I made a week ago are often not yet answered. I'll need to think carefully about next steps for this paper, as well as how feasible it will be for us to write a second paper, since I do not want to have unreasonable expectations for this project and the current iterative procedure for editing is not suitably efficient.



EK(4-29) Below is the updated draft that merged my and Raj's drafts.
Draft of JMB_042913_ek2.docx


EK(4-29) test to see if get email notification when update to wiki.

EK (4-29) Below is the latest draft with my revisions. Important: my browser did not properly refresh and I did not see Raj's JMB_042713_rc.docx draft until after I had finished my revisions. I will merge the two documents on Monday. So this draft may not reflect all of the comments and revisions in Raj's document. This draft incorporates most of the major changes Raj and I talked about on Friday.
Draft of JMB_042913_ek.docx


RC (4-27): After discussion with Eric on Fri, I have gone through the Sections he indicated were roughly complete, answered his questions, and indicated additional required changes. This does not include the computational discussion section - here, I have only indicated what changes Eric and I agreed upon. I will need to look at it again after Eric's next posting. I have not carefully read the experimental parts of the results, discussion and method sections yet, but I have made some minor comments to those sections as well, for XG. In addition, I have gone carefully through the intro and abstract, and made multiple suggestions for changes.


Eric must post his remaining changes by Sunday night, so we can do our next iteration Mon am. After his update, I will also look more closely at the experimental sections. XG, please make your remaining changes to the experimental sections by Mon.

Starting Mon, Eric will finish any remaining figure prep/simulation tasks that remain on the table.

Draft of JMB_042713_rc.docx


XG (4-25):

(1) Comments on Figure 6 (page 12): JMB requirements about composite figures is listed as following: No more than four sections should appear in a single figure. Label individual sections in composite figures clearly with lower case letters, using (a), (b), (c).
(2) Eric, I looked through the current draft and have a TO DO list below. We can discuss if the points raised up here are necessary and also please indicate the missing points:
(3) Eric, Dr. Raj suggested us to Skype late this week for the purpose of finalizing the draft. Will you be able to finish the editing today? If so, we can go through the manuscript together tomorrow sometime. I am OK with 9:00AM -1:00PM and 2:30PM - 4:00 PM. Please let me know your plan.
Wed, Apr 24, 2013
Draft of JMB_042413_ek.docx

RC (4-24): This document is getting very hard to read. It has an assortment of colors and I cannot tell which edits were the most recent from Eric. Also, Eric has not provided any comments regarding his planned additional changes, as XG and I have started to do. Eric must post again more clearly by the end of the working day EST today (i.e., by 6 pm). Eric's schedule for changes is completely unclear at the moment, and we are concerned.



Tues., Apr. 23, 2013
Latest paper update. Still to edit much more.
Draft of JMB_042313.docx


Thur., Apr. 18, 2013
Some of the results for SIRT3 using the newer PDB:4FVT crystal structure.
4FVT_MM-GBSA.vs.pIC50.jpeg4FVT_LIA.pIC50.jpeg4FVT_GlideXP.vs.pIC50.jpeg
The data points for the above figures are for all the SIRT3 inhibitors: AC93253, Salermide, EX-527, 1-methylnicotinamide, nicotinic acid N-oxide, iso-nicotinamide, nicotinic acid, pyridine N-oxide, and nicotinamide. The results would improve with the nicotinamide outlier (at pIC50 4.4) left out. Also, all the models do an impressive job accounting for the wide range of activities. Except for nicotinamide, all the smaller single ring molecules have much lower inhibition in the mM range. It is not clear to me if these good results would hold up with decoy molecules (molecules that are similar to potent inhibitors, but are known experimentally to not bind).
Also note that the Y-axis in the above plots are not the raw GlideXP, or MM-GBSA scores. The linear regression fits and scales those scores to correspond to the same range as the pIC50. Plots for the publication should not have this scaling. I will fix.


(EK 4-4) Latest power point here:
Dropbox/pmc-at research/Eric/ACS_slides_04.03.2013.pptx

RC (4-1): My comments on slides and paper below (in no particular order).

(EK 4-2) Eric's comments from 2013.04.02.
(XG 4-2): Will Skype at 1:00PM, April 5, 2013.

(EK 4-2) will add comments to below as I edit slides. I have not added comments to below yet.
Slide-by-slide comments: the following comments apply to the paper as well as talk
-I changed the title
-slide 5 - need to decide which parts of biology to emphasize. Could either emphasize liver application (inhibition), or could mention both inhibition and activation applications, suggesting that understanding of inhibition mechanism is important to design of both. Need to say something about how the peptides you are using relate to the protein substrates of SIRT3, minimally should say that the protein sequence does not greatly affect the inhibition kinetics.
-slide 8 - likely will not have time to say so much about mechanism given limited time. In general, the notes at the bottom of the experimental slides need to emphasize in bold what will be said in talk
-slide 11 - i assume the point here is that the type of assay used is important, and we are using assays that avoid problems encountered in the earlier sirtris/gsk work.
-slide 14 - not sure what will be said here
-slide 15 - i would suggest directly mentioning EK's computational work here that verifies the finding of competitive inhibition, and that he will comment on in the second part of talk
-slide 18 - you should mention the names of the tight binders just as EK does
-slide 20 - may add LIA
-slide 23 - does not appear essential for talk (thought it should be mentioned in paper)
-slide 26 - i agree that for the paper we may redo the noncompetitive/competitive calculations with the Carba-NAD structure, but see my comments below on paper writing (any new simulations should occupy only 25% of total time).
-slide 28 - mention that if the MM-GBSA/LIA correlations are strong (e.g. like some of those for SIRT2), we will use this method to discover new inhibitors
-slide 30 - what do we mean by promiscuous?

--The paper should be finished based on current content
--After the conference, XG and EK devote 75% time to finishing write-up of paper based on slides. EK's efforts will be esp important since his parts of the paper are largely unfinished.
--KM learns LIA and reports adjusted R^2's. (KM may help w discussion of LIA).
--Was LIA reported in the referenced SIRT2 paper? If not (unlikely), the results could be shown in this paper
--After conference, XG and EK spend 25% of their time to continue to do congeneric series experiments and computations with tight binders for use with LIA and MM-GBSA, most likely for a follow-up paper. XG considers whether to assess if newly discovered tight binders are competitive or noncompetitive inhibitors. This all depends on how long the experiments take, and may seem obvious given that NAD+ binds directly in the C pocket of SIRT3, implying the inhibition mode is competitive for all.
--If the paper revisions are done before the latest congeneric series data and sims are done, we submit the paper in that form. It may even be better to save the new results for another paper.

Future work:
--KM continues running LIA for follow-up paper if it is not finished for this paper
--KM looks at SIRT6 structures, considers application of LIA; XG does kinetics of inhibition and inhibitor screening only if the computational/structural analysis warrants it
--KM considers development of LIA models for protein design (proposes what experimental binding affinity assays with inhibitors would be needed to train the models)
--KM's goal in attending talk is to plan out this future work
--XG finishes continuous coupled assay (to be used in second paper) and considers protocol for base exchange kinetics


Mon, Apr 01, 2013
See comments below on Mar. 28, 2013 by EK

Thu, Mar 28, 2013
Notes from Mon. phone call with Raj and other notes on slide preparation:


Fri, Mar 22, 2013
Tue, Mar 19, 2013


Mon, Mar 18, 2013

Thu, Mar 14, 2013
RC (3-14): KM, please provide your thoughts as well.

While Eric is preparing his notes, I would like to summarize the minutes of the meeting.

1) As far as Lyne paper data is concerned, they used the initial version of Schrodinger software and we use the current version. There could be some difference in the calculation of binding energy between the these two versions. Also, they could have done the various parameter change to improve the R^2 value. This may not be necessary and even a R^2 value of 0.4 is acceptable. Also, he suggested that Lyne, et al could have done some post processing after MM-GBSA docking which could have improved the R^2 value.

2) We have asked how the protein preparation method affects the final results of MM-GBSA simulation - He told that it should not affect much.

3) Is there any stochastic element with Glide simulation so that one should run this multiple times to improve the R^2 value? He said no. but Different computer will handle the simulation differently. For example, Window and Linux based operating system have a different way of rounding number etc. This will affect the simulation results slightly but not very much.

4)He has also suggested to explicitly choose SGB option to dock molecules. We need to learn how to do this Schrodinger software.

Eric will post the detailed notes soon.

Karthik.

RC (3-14): There were several other questions that Eric and I went over on Mon. I hope EK will post on these as well. EK should also follow up tonight with the requested talk/paper planning tasks on the wiki given all the info now in hand.


Wed, Mar 13, 2013

Tue, Mar 12, 2013

RC (3-13): KM, after the meeting, please post a Q&A listing all questions asked to Schrodinger along with the answers.

Mon, Mar 11, 2013

Fri, Mar 08, 2013

RC (3-8): Eric, I looked over the strategy doc below. Project 1 seems like a good idea and I agree that you should proceed with it. However, we still need to ascertain why we have not been able to reproduce other literature data using MM-GBSA without MD, through discussions with Schrodinger. The discussion points for Schrodinger should be revised before sending out to Schrodinger on Mon. What seems to missing is a description of what freely adjustable parameters were specified in the paper that you were able to reproduce. If you were given a detailed protocol used by the other two papers, what parameters and/or protein/ligand preparation details would have been specified? Regarding treatment of waters, are you planning to redo scoring with these methods prior to speaking to Schrodinger? I would also like to see a plan for how Karthik will be involved in computations for the new strategies.

The backup plan Eric describes in Project 2 is reasonable for the purposes of the presentation, in case we are not able to resolve the issues with Schrodinger. In that case we can start on Project 2 by mid-week next week.

If we can be the first to publish a method of predicting binding affinities of inhibitors to human sirtuin enzymes SIRT3 and SIRT6, it would put us solidly on the sirtuin map.


Thu, Mar 07, 2013

Wed, Mar 06, 2013

Tue, Mar 05, 2013

Mon. Mar. 04, 2013

Fri, Mar 01, 2013

Thur, Feb 28, 2013


Wed, Feb. 27, 2013
Tasks:
  1. Other MM-GBSA data sets
    my simulations with the Thrombin data sets from the Cardozo paper did exactly replicate their results, we want to confirm that we can replicate MM-GBSA results by testing two other data sets. (a) a different enzyme from the Cardozo paper. (b) a data set from a different paper TBD that shows a correlations between MM-GBSA and experiment
  2. Continue to review literature
    1. XG reviewed some papers we found a while ago (see her posting from today on Task lists from lab) I should review these papers and post summary.
    2. Has any one in literature used MM-GBSA and calibrated to experiment to find new high affinity sirtuin inhibitors/activators?
      1. Not that I found so far.
    3. update literature. Find new and recent papers not in our list of papers. Review papers
      1. see below
    4. put list of docking studies and sirtuin simulations studies in one place. right now the references are scattered throughout the wiki and in the paper references
      1. in progress.
    5. Review the J.Med.Chem. v51 (2008) p1203 paper that Xiangying was discussing in meeting on Mon. Feb. 25 on Sir2 docking into C pocket. Specifically pg. 1208 figure
    6. Note techniques that others are using in publications about docking and MD simulations of sirtuins.
      1. MD - used in most of newly published studies (< 5 years)



(1) Other MM-GBSA data sets: I did not get results with as high a correlation as the publication.
The best paper that most closely matches what we've been doing with the MM-GBSA re-scoring of the GlideXP docked poses without any molecular dynamics is the following:
There are 4 data sets in this paper (4 sets of ligands docked to 4 kinase proteins). I chose the 1st set with the largest number of cogeneric ligands docked into a protein called p38 (PDB:2BAK). The paper docked with GlideXP, then rescored with MM-GBSA with a minimal minimization and with the OPLS-AA force field.

Here is the info about the ligand set and the MM-GBSA ∆G binding from the publication:
p38_ligands.jpegp38_data.jpeg
Below are two charts: left is the one I created trying to replicate the one on the right from the paper following their methods using the same Schrodinger software (I used a later version of the software). The paper claims a much higher correlation than I got ( r = 0.84 vs. r = 0.57 for theirs vs. mine).
p38_myChart.jpeg
My Results
p38_chart.jpeg
Paper


Below shows the superimposed docked ligands (all 13 of the p38 cogeneric series in this set) after minimization with MM-GBSA. The green molecule is the co-crystallized structure from PDB:2BAK (not in the docked set). As shown, all the ligands in the series docked correctly within 1.5 Å RMSD of the co-crystallized structure, and all are very similar to the co-crysallized structure. Thus, the problem is not related to incorrect docking.
p38_mmgbsa_docked_superimposed.jpeg


(2) Literature Review




Mon. Feb. 25, 2013
Revised Thrombin simulations. Still not replicating exactly the Cardozo 2008 paper (which had R^2 of 0.7 and 0.8 for single and ensemble MM-GBSA, respectively). The difference between the below and the previously reported thrombin simulations (which had an even larger difference between the published results and my simulations) is that the below fixed the protein docking site, while the previous allowed a zone of flexibility of the protein (all atoms) within 5.0 A of the ligand, as done in the Cardozo paper. Keeping the protein atoms fixed improved the results, but the results still did not match the higher correlation reported in the paper.
mm-gbsa_revised_thrombin_2013.02.25.jpeg

Latest Tasks:

XG: I can do 2:00-3:00PM, Feb 20.


Friday, Feb. 22, 2013
First run of docking to SIRT3 of the database of molecules Xiangying started. The initial list of the molecules is from the document Xiangying created called "Sirtuins inhibitors for NAM binding site_02122013.docx" posted previously on her wiki. These are molecules which have been found in the literature to inhibit activity of a sirtuin or HST enzyme, both human proteins and from other organisms. The screen shown here is when this list was docked with Glide to our model of the SIRT3 binding pocket from PDB:3GLR, which I previously used to dock NAD+ and nicotinamide. Below is a preliminary list of the top Glide scoring molecules (listed by molecule number from Xiangying's original document). This is a preliminary screening of 35 molecules from that list using GlideSP. A more comprehensive screening using GlideXP with more molecules is in progress.
prelim_screening.jpeg




Thursday, Feb. 21, 2013
Thrombin results in table below. Molecule numbers relate to the labeled figure below the table. All columns except the last three are from the Cardozo paper. The last three columns are from my simulations. My simulation numbers are different. Slightly different for GlideXP (within a reasonable error within 2 kcal/mol, relative to each other, given that I used a different version of Glide). My simulation MM-GBSA numbers do not match the Cardozo paper numbers. I need to investigate what happened here. Further below are graphs showing the correlation between my simulations numbers and the reported experimental pKi. These graphs are similar to what was in the Cardozo paper, but the R^2 value is much much lower (mine 0.19 and 0.32, vs. paper was 0.71 and 0.81).
Molecule #
 pKi
 GlideXP score (kcal/mol)
 MM-GBSA (Ensemble Average) (kcal/mol)
 MM-GBSA (Single Conformer) (kcal/mol)
 XP GScore
 MMGBSA dG Bind (single conformer)
 Relative MM-GBSA (single conformer)
 Relative MM-GBSA (single conformer,corrected with MCMM of unbound ligand)
1
 7.3
 -14.0
 6.1
 7.2
 -10.8
 -51.8
 70.0
 239.5
2
 7.3
 -14.3
 10.0
 10.3
 -10.9
 -107.0
 14.8
 126.9
3
 8.8
 -13.9
 4.3
 4.9
 -11.1
 -109.2
 12.6
 79.1
4
 6.6
 -11.8
 6.0
 5.9
 -10.5
 -100.7
 21.2
 73.3
5
 9.0
 -14.0
 2.1
 2.3
 -11.5
 -106.8
 15.0
 52.6
6
 6.8
 -13.3
 5.8
 5.6
 -11.6
 -114.2
 7.6
 74.1
7
 5.5
 -13.2
 10.9
 7.9
 -10.1
 -90.3
 31.5
 107.4
8
 9.9
 -11.2
 1.9
 1.9
 -11.3
 -118.3
 3.6
 21.5
9
 10.3
 -14.1
 0.0
 0.0
 -11.3
 -121.8
 0.0
 13.3
10
 8.1
 -13.1
 4.8
 5.6
 -10.3
 -103.9
 18.0
 5.0
11
 8.4
 -10.3
 5.2
 5.9
 -10.0
 -105.2
 16.6
 0.0
12
 7.7
 -12.7
 5.3
 4.2
 -9.5
 -98.4
 23.4
 124.9
13
 10.0
 -15.7
 0.7
 0.9
 -12.2
 -119.7
 2.1
 32.6
14
 9.5
 -13.0
 3.3
 3.6
 -8.8
 -84.5
 37.4
 19.1

thrombin.mmgbsa.single.2.jpegthrombin.mmgbsa.single.1.jpeg

RC: Why have we not rescored these with higher level calculations yet, or otherwise provided some hypothesis as to the cause of the considerable differences in results.
How long does this rescoring take - perhaps a day or less? We cannot operate at a frequency of one update per week on these calculations. How can we proceed with our computational work without resolving this issue asap?


Monday, Feb. 18, 2013

Relationship between K_i and IC_50: linear, with few exceptions, depending on assumptions of kinetic model. The following paper shows detailed equations for the relationship between K_i and IC_50 for Michaelis-Mention kinetics for competitive, non-competitive and uncompetitive inhibitors. It also shows cases for very tightly bound inhibitors. In almost all cases, there is a linear relationship between K_i and IC_50.
Since pK_i = -log(K_i) and pIC_50 = -log(IC_50), pK_i and pIC_50 have the same log relationship to each other, as the underlying IC_50 and K_i values have a linear relationship to each other. Thus, the plots of MM-GBSA delta_G vs. pIC50 should show the same linear relationship as the plot of MM-GBSA delta_G vs. pKi.
RC: XG, would you please provide a link to or summary of the method you used to convert IC_50 to K_i.
XG: http://botdb.abcc.ncifcrf.gov/toxin/kiConverter.jsp

RC: Please confirm that that the transformation from IC50 to Ki that you used is linear and provide a short write up on it for the paper if you have not already done so, if it is in fact necessary for obtaining Ki's from the experimental data.

Delta_G of binding for the inhibitors of SIRT3 from Xiangying's experiments.
Assuming Delta_G = RT ln K_i where R=1.99E-3 kcal/(K*mol) and T = 295K, and K_i is converted from uM to molar.
Note that there is no minus sign in the above equation, as binding and inhibition occur in opposite directions. If K_b for binding were used in the equation, there would be a minus sign.
hSIRT3 inhibition
 Ki(NAD+), uM

Delta_G (kcal/mol)
 (kcal/mol)
 (kcal/mol)


Competitive
 Noncompetitive
 Competitive
 Noncompetitive
 GlideXP
 MM-GBSA
Nicotinamide
 11.6

-6.66
-5.0
 -31
1-methylnicotinamide chloride
 2916.3
 9215
-3.42
 -2.75
 -4.3
 -40
nicotinic acid N-oxide
 4123.6
 13030
-3.22
 -2.54
 4.2
 -15
Iso-nicotinamide
 4367.3
 13800
-3.18
 -2.51
 -4.3
 -33
Nicotinic acid
 4588.9
 14470
-3.15
 -2.48
 -3.2
 -22
Pyridine N-oxide
 7617.5
 24070
-2.86
 -2.18
 -2.2
 -24

GlideXP scores are in the correct ball park of -2 to -5 kcal/molWhen comparing these Delta_G numbers to the simulation estimates of Delta_G with the below plots of experiment vs. docking for hSIRT3 inhibitors:
MM-GBSA scores are much lower (-15 to -40 kcal/mol) than absolute binding affinities. However, low numbers like these are reported in many other publications, such as the Cardozo paper.
RC: nicotinic acid N-oxide has a positive glidescore? If not, it is interesting to note that Glide may have been more successful in rank ordering that MM-GBSA. Please comment on this when we meet.

RC: XG, how do the above Ki's for the inhibitors other than NAM compare to others reported in the literature, e.g. for Sir2 or SIRT1, that have received attention as potential drug candidates?
XG: Isonicotinamide was reported as Sir2 activator by relief of NAM inhibition. Nicotinic acid was reported no inhibition effect for Sir2 and SIRT1. I used nicotinic acid as a control. 1-methylnicotinamide chloride, nicotinic acid N-oxide and pyridine N-oxide are from Chief's list, which not published anywhere else in terms of their inhibition effect. Chief wants to test them.
RC: Are there no C pocket inhibitors for SIRT1 that have been proposed as drug candidates? You did not mention any such molecules and their Ki's above.
XG: A quiet few of molecules in the literature (highlighted by red circle below) have good inhibition for hSIRT1. However, it was not clear if they are C pocket inhibitors for hSIRT1 since SIRT1 structure has not been solved to date.
RC: I am trying to understand what is considered a drug candidate for sirtuin inhibition given its Ki. I.e., what binding affinities are we aiming for and is it important to accurately predict the binding affinities of the types of molecules we have assayed so far (or are they orders of magnitude too weak in binding, so we do not need to accurately predict such weak bindingi affinities). Are any of these considered drug candidates? Also, I saw results in the past of inhibitor docking to sirtuins. What pocket did these dock in?
XG: Please comment on this. Can you remind me of what molecules these prior sirtuin docking studies looked at?
C pocket inhibitor for SIRT1.jpg

RC: (Sat Feb 16th): I have communicated with Eric on the importance of timely posting of the results requested by me below. We are behind on simulation tasks by a couple of weeks and cannot start on new experiments with inhibitors until these tasks are completed.


Sat. Feb. 16, 2013
Meeting minutes for Talk between RC and EK
This has still not been posted as of Sat 2-23 (task communicated I week prior).
Tuesday, Feb. 12, 2013
Experiment vs. Docking of hSIRT3 inhibitors (nicotinamide, 1-methyl nicotinamide chloride, nicotinic acid N-oxide, iso-nicotinamide, nicotinic acid, pyridine N-oxide) for both GlideXP and MM-GBSA. The results do not show a clear trend and do not show a better way to estimate the binding affinity through linear regression of docking binding affinity estimates to experimental binding affinity.
Experiment.vs.docking.hSIRT3.jpeg


RC: Eric, it is important that we post the regression for thrombin inhibitors asap as we have been discussing over the last two weeks. The reason is that we need to assess how well your MM-GBSA scores correlate with the scores reported in the literature before proceeding much further. You could also do a regression of your MM-GBSA scores against the experimentally measured binding affinities for thrombin inhibitors. If it turns out your scores are consistent with those reported in the literature, we know we should look more closely at our lab protocols, whereas if they are not, we should focus on higher levels of sampling, etc. We already have a reason to be skeptical of our experimental Ki's for the new inhibitors - since they did not bind tightly to the C pocket, the Michaelis-Mention approach to determining Ki may not be valid. So, in parallel we should start computationally screening the new inhibitors Guan posted. We really need to accelerate efforts on both these fronts. I believe the above plots should have taken an hour at most to produce. When can you get us these results? Thanks.
Also, as noted below I indicated we should report the experimental Delta G's of binding not just the pIC50's above, so we get an idea as to whether the experimental and computational Delta G's are of the same order of magnitude.
I assume all these tasks will be complete for the Thursday update. I believe that XG has stopped working on inhibitors until we have a next set of tasks from EK, and I would like her to come back to inhibitors next week. We can only do so if there are enough lead compounds identified for lab testing.

Next steps to post by Thursday to redo these graphs with, hopefully, better simulation numbers which show a clearer trend:





Tuesday, Feb. 5, 2013
MM-GBSA literature for re-scoring docked poses. Overall, a handful of papers claim that MM-GBSA re-scoring of docked poses has better correlation to experimental binding affinities. However, the majority of the citations for MM-GBSA and binding affinity relate to higher level simulations where MM-GBSA is used with better sampling techniques such as MD or Monte Carlo. Conclusion: we really need to do the higher level sampling method. The Schordinger software has built in methods for that, such as MCPRO+.
MM-GBSA_literature.doc
Next steps:

RC: A little confused here on the status, please help me catch up. Are we ready to do a linear regression of your current MM-GBSA scores against XG's experimentally determined binding affinities? Not sure why we would not want to do this before proceeding with higher level sampling methods. Please let me know asap if you would like to discuss.
EK: Yes, we're ready for the regression. I will post that Tues.
RC: Ok, you can use Delta G = -RT ln K_I (technically, K_I/[C_0] where [C_0] is 1 M). This is - \Delta G_bind since K_I is a dissociation constant so report the negative or use 1/K_I above). This will allow us to compare the actual magnitudes of the experimentally determined \Delta G's to those from computations. (Alternatively, you can use pK_I, as in the literature papers, but these would not be directly comparable to computational values).

RC: Also, why did we not run a linear regression of our MM-GBSA computed thrombin binding affinities against the reported thrombin binding affinities? I posted regarding this last week.
EK: Yes, I will post that Tues, as well.


RC: Please provide the Glidescores and MM-GBSA scores for binding in the other pocket, along with the list of which molecules docked in other pockets. It is obviously important that some of these molecules did not dock into the C pocket in the highest ranked poses, since the experimentally obtained binding affinities may correspond to poses in another pocket. Please discuss the results with XG. XG, do have any other molecules handy that might be added to the cogeneric series? XG, once Eric provides the list of those molecules that bound in other pockets, please let us know whether these molecules had lower Ki's than the others, if you have finished assaying them.

XG: I am still doing the experiments (The experiments I did last week did not work well due to some technical problems). I am catching up. From the preliminary results, it looks like some of these 6 small molecules have different inhibition effect of hSIRT1 vs. hSIRT3. I need to double confirm. If it is true, I will need Eric to dock them into Sir2 protein (1YC2 or 2H4F) for comparison. It will be very interesting to address them in our paper. Eric and I had half hour discussion yesterday. So far, within the 6 molecules, experimentally, NAM is the strongest inhibitor of hSIRT1 and hSIRT3, which is agree with Glidescores but MM-GBSA scores. We will talk again on Friday, hopefully I have some handy data to discuss.


What about the important info from the other papers on MM-GBSA? It seems these docking runs below should have only taken a day or so.

Monday, Feb. 4, 2013
See the below table for the GlideXP and MM-GBSA scores for poses docked outside of the C pocket. Note that in all cases except one, the MM-GBSA score ranked the C-pocket poses (for a given ligand) higher than any other pocket for that ligand. Also note that the listed scores do not include higher ranked GlideXP scores in non C-pocket poses, which were subsequently ranked lower than those shown in the table when re-scored with MM-GBSA.



Mon. Jan. 28, 2013
The following are results from docking ligands into SIRT3 (PDB: 3GLR). All of the below ligands docked to the C-pocket without constraints. However, some of the ligands had one or two higher ranked poses that were not in the C-pocket. The below table shows the highest ranked scores that docked in the C-pocket. The last two columns show the best ranked MM-GBSA scoring pose that is not in the C pocket. The other pocket for all cases is in between the B and C pockets (just above the C pocket). Those labeled N/A did not have any poses docked into other volumes besides the C pocket. In only one case (Nicotinic Acid) did the pose docked into the other pocket score higher with MM-GBSA than the C pocket pose. Of note, also, is the discrepancy between the GlideXP scores for Nicotinic Acid N-oxide.
Pocket ---->
 C
 C
 Other
 Other
Ligand
 GlidXP
 MM-GBSA
 GlideXP
 MM-GBSA
Nicotinamide
 -5.00
 -30.8
 N/A
 N/A
iso-nicotinamide
 -4.33
 -32.5
 N/A
 N/A
1-methylnicotinamide chloride
 -4.28
 -40.0
 -2.83
 -21.1
Pyridine N-Oxide
 -2.20
 -23.8
 N/A
 N/A
Nicotinic Acid
 -3.24
 -21.8
 -2.78
 -28.3
Nicotinic Acid N-oxide
 +4.15
 -14.7
 -3.30
 -11.9

RC: Are these results with MCMM?


Wed., Jan. 23, 2013
Finished:



To Do:

RC: How did your scores compare to those reported in the lit?

Thurs. Jan 17, 2013
Priorities for the next week:
I will try to finish the above by Wed. Jan. 23.


Jan 14, 2013
Outline for next tasks and projects: 2013.01.14.doc




Dr.Raj,


Please find the attached paper and thesis on MM-GBSA. I have downloaded the thesis of Boas and attached the same here.

I will update once again the wiki with my tasks list.

Best,

Karthik

.PHYSICS-BASED DESIGN OF PROTEIN-LIGAND BINDING - Boas thesis.pdf

MM-GB(PB)SA Calculations of Protein Ligand binding free energies.pdfFast and accurate predictions of binding free energies using MM-PBSA and MM-GBSA.pdfAssessing the performance of the MM-PBSA and MM-GBSA methods The accuracy of binding free energy calculations based on molecular dynamics simulations.pdfAccurate Prediction of the Relative Potencies of Members of a Series of Kinase Inhibitors Using Molecular Docking and MM-GBSA Scoring.pdf


Tasks For Karthik

RC: Karthik, please list your tasks based on Sun discussion with Eric. I have attempted to summarize them here:

1) Literature search: a) MM-GBSA of cogenerics - at least 3 new papers. Are binding affinity tables provided?; b) MD binding affinity prediction papers indicating accuracy of binding affinity prediction for ligands of various sizes.

2) You should also be preparing your GBSA presentation and connecting it up with Eric's presentation on sirtuin binding affinity predictions. Your literature search on MM-GBSA scoring of cogenerics will be useful here. He is also planning to include some mention of the accuracy of MD in absolute binding affinity predictions. This again will rely to a certain extent on your literature search. The ppts should cross reference each other so the connection between projects is clear.


3) Add a few slides to RC's existing slide sets on computational enzyme design provided here:

CMU talk.ppt

neb_enzymes.ppt

for the purpose of a slide presentation to Anna on the project "MM-GBSA scoring project for protein design" listed under "Docking Simulations". The enzyme design part of the CMU talk can be augmented if necessary with a few slides from neb_enzymes (e.g., on the model fitness measure, and possibly a couple of more examples of enzymes and sequence distributions), and then 3-4 new slides for the new MM-GBSA project can be added (see project description for details).

KM can work on these protein design slides while refining his GBSA slides. The protein design slides will b esp useful if MD slides not fully developed, or if there are issues with MD accuracy. KM and RC can jointly present these slides at Anna. KM will also present the GBSA slides. EK can present his slides on sirtuin overview and binding affinity calculation. The right order would appear to be: a) sirtuin overview and binding affinity calculations (including any MD), b) GBSA (KM please provide status update), and then c) protein design. RC might be able to provide the introduction to sirtuins based on EK's slides.





Tasks For Eric


Wed Dec 26, 2012

Eric will finish (prior to Jan 1) sirtuin presentation and include the current work for finishing this paper in terms of doing MM-GBSA calculations on a small series of cogeneric inhibitors, as well as the next step where we will look at a larger series of cogenerics.


Wed Dec 19, 2012

RC: I went over the Cardozo paper relatively carefully. Here are my thoughts on what we can realistically aim to achieve in this paper's computational section, as well as a breakdown of tasks between myself and Eric, and what should be left for future work (e.g. the MD project). My comments are based on the results and discussion presented in Cardozo.


Please provide your comments on the above strategy.


Mon Dec 17, 2012

RC: Eric, after finishing the slide presentation, can you please comment on which of the following priority tasks below you can work on without the licenses:

1b) One thing to check is to compare the molecular mechanics energy portion of the GlideScore to the molecular mechanics energy from MM-GBSA. These components should be similar. Although GlideScore will have a heuristic scaling factor for the molecular mechanics energy. -- do you have this data stored?

5) Verify the calcuations from the 2008 Guimarães and Cardozo paper, “MM-GB/SA Rescoring of Docking Poses in Structure-Based Lead Optimization”.
a) They used an earlier version of MacroModel for all their MM-GBSA and MCMM ensemble calculations.
b) Try to replicate their data.
-- you can't run these yet, but can you make a list of which computations you will run?

7)
b) Reword mentions of “ensemble” at the beginning of the methods section. Ensemble refers to multiple different methods in this section, which is confusing.
c) Put in latest computational simulation #’s for MM-GBSA with MCMM ensemble corrections. RC thinks it would be better if we had one document with the latest results.
d) There are many grammatical errors to correct
f) Length issues: what is the length restriction in JMB (journal of molecular biology)
-- I think all of these can be addressed at this time. f) is important to know at the outset.

You can continue working on some of the above tomorrow (we can discuss over skype), and in the meantime I will identify some other tasks for the list for me to work on.



Wed, Dec 12, 2012
Notes from Dec. 10, 2012 phone meeting between Raj and Eric:2012.12.10_notes.doc
Dec. 4, 2012

EK:
Computationally predicted binding affinity too high
As I prepared the next draft of the paper, I double checked a nagging worry I had about the relatively large predicted binding affinities of NAD+ into both the AB and AC pockets of SIRT3 and Sir2 which ranged from -84 to -107 kcal/mol. These are values before the MCMM ensemble correction of the unbound ligand. The values are much higher than any experimental value. As a though experiment, what would the Kb (binding equilibrium constant) be if ∆G(binding) = -90 kcal/mol ?
binding: E + I -> EI K(binding), Kb
∆G(binding) = -R*T*ln Kb

=> Kb

exp(∆G/(R*T))R==1.986 x 10^-3 kcal mol^-1 K^-1, T==298K, ∆G==-90kcal/mol

> Kb ≅ 1.06 x 10^66 <===

this is an astronomically huge number
The BindingDB (http://www.bindingdb.org) is a publicly accessible database currently containing > 20,000 experimentally determined binding affinities of protein–ligand complexes, for >110 protein targets. It can give us an idea of the acceptable range of experimentally determined equilibrium constants, which range from 10^-20 < Ki < 10^10 . Since Ki = 1/Kb, a Kb==10^66 is many orders of magnitude out of normal.

However, large computationally determined binding affinity predictions are sometimes reported in the literature. There are other groups which report ∆G(binding) from MM-GBSA below -100 kcal/mol for a minority of molecules in a large test set (see J Comput Chem. 2011 April 15; 32(5): 866–877). These predicted values are off by an order of magnitude from the experimental values. It would also be reasonable to expect that a large and flexible ligand like NAD+ would produce such a large error in the predicted binding affinity. Errors in the ∆G(binding) are magnified in the predicted Kb and Ki values because of the exponential relationship.

MCMM ensemble averages corrected
Given the above large predicted values for ∆G(binding) I have been double checking the calculations. The ensemble corrected MCMM binding affinity energies increase the deltaG binding for NAD+ to the AB pocket of Sir2 from -95.2 kcal/mol (MM-GBSA) to -26.2 kcal/mol, and increase the binding estimate for the AC pocket from -92.6 kcal/mol to -9.8 kcal/mol. This is partly good news because binding affinities are generally in the range below -25 kcal/mol for most ligands, thus the > 90 kcal/mol estimated by MM-GBSA would draw questions from reviewers. Note that Glide docking scores for both the Sir2 NAD+ AB or AC docking was about -11.0 kcal/mol - within the expected range of -25 to 0 kcal/mol that is normal. So it appears that the ensemble average and the associated more comprehensive exploration of the conformational space of the free NAD+ ligand (instead of simply locally minimizing from the bound protein-ligand complex) is extremely important for a large, highly flexible ligand like NAD+. Entropy contributions (estimated at 1.6 kcal/mol for the unbound ligand ensemble only) is small and well within the margin of error for these estimates. The entropy is not included in the above figures.
The numbers support the noncompetitive mechanism for Sir2, in that NAD+ is predicted to preferentially bind to the AB pocket instead competing with the nicotinamide inhibitor for the C pocket. However, the large difference between the AB vs AC NAD+ binding estimate with the MCMM ensemble method (-26.2 kcal/mol, -9.8 kcal/mol, respectively) may cause some concern.
The numbers for SIRT3 don't look as good after the MCMM ensemble correction:
AB AC
MMGBSA_dG_Bind -84.4 -107.9 kcal/mol <---- MM-GBSA predicted binding affinity before ensemble corrections
Ligand_Energy -47.1 -28.4 kcal/mol <--- uncorrected single conformer unbound ligand energy
<E unbound ligand> -167.6 -167.6 kcal/mol <--- ensemble MCMM average unbound ligand energy
======================================================================
MCMM +36.0 +31.3 kcal/mol
note:
MCMM corrected energy = (MMGBSA_dG_Bind) + (Ligand Energy) - <E unbound ligand>

I have to get to the bottom of these positive numbers before I proceed, as we cannot report a positive binding affinity. I suspect that there are some important differences in the two different programs (PRIME for MM-GBSA and MacroModel for MCMM ensemble) which are causing this problem. The 2008 paper that reported good results with the MCMM unbound ligand ensemble energy did all calculations in MacroModel. Note that the above reported results are different from what I reported on the wiki on Sept. 20 because of unlabeled values in output files, where one program reports energies in kcal/mol and the other in kJ/mol. I will report back tomorrow about this.

Methods section edited
In the mean time, the paper is revised through the methods section. (attached) Don't read the results section until I resolve the above issue.
Outline of JMB_120312.doc
Edits through the methods section include:
1) elimination of some sentences and shortening.
2) a short ustification of customized induced fit protocol – better sampling
3) description of the ensemble MCMM method for the unbound ligand
4) brief description that PLOP does not do extensive backbone sampling for induced fit
5) reordering of paragraphs to more clearly describe the three different docking protocols.
6) correction about Epik and PROPKA and partial charges. No semi-empirical or QM method was used to determine ligand partial charges. Partial charges are based on the OPLS force field and the Epik determined protonation states.
I'm waiting to add the rest of the changes until I resolve the ensemble MCMM.












RC: Eric, your tasks we discussed regarding starred item insertion into paper and overall revisions of old material should be submitted by Thursday deadline. Thanks.

RC: Eric and XG, I have read your meeting notes and Eric's latest progress notes. Are we planning to include iso-NAM in this paper or the next paper? It seems a lot of your discussion
focused on iso-NAM. Also, Eric, I will be waiting on your final notes on the use of MM-GBSA for pose generation as well as scoring, vs Glidescore for pose generation and then MM-GBSA scoring.
For mechanistic applications like ours, where we are not screening, are we still able to provide a justification (besides computational efficiency) for using Glidescore?

EK: yes, we're planning on including iso-NAM in this paper. GlideScore rather than MM-GBSA is probably a better metric for generating poses. I'll try to find a reference for this argument.

Nov. 16, 2012 (Fri)

Nov. 15, 2012 (Thurs)

Nov. 11, 2012 (Mon)

Nov. 6, 2012 (Tues)

EK: I do not yet see a list of all tasks we discussed this weekend listed here. This is what Risa was referring to. Please post those asap and finish them by Thursday. Also, there was an ** item missing from your email this weekend. Please address that by Th as well. After Th, meet with Guan to discuss.

Strategy for paper. Both strategies are for 2 papers. I vote for strategy 2.

Larger List of Possible Journals.
Rank ordering of Journals. Journals are in a few groups.
Journal
 Impact Factor (1)
 Impact Factor. (2)
 Description
 Comments
Nature
 36.280
General
Science
 31.200
General
J. Am. Chem. Soc.
 9.907
General Chemistry
PNAS
 9.681
General
PLoS One
 4.092
General
 Latest trend journal
J. Med. Chem.
 5.207
 5.25
 Theory and Biology
 Paper would need to have results for many drugs / compounds





Biotechnology Advances
 7.60
 8.25
 Biology, some theory
J. Mol. Biol.
 4.008
Biology, some theory
Biophysical Journal
 4.22
 3.65
 Biology, some theory
 Not quite as quantitative, very little theory
Biochemistry
 3.42
Biology, some theory





Cell
 32.40
Biology, little theory
Molecular Cell
 14.18
Biology, little theory
J. Biol. Chem.
 4.77
 4.77
 Biology, little theory
Mol. Biol. of Cell.
5.98
 Biology, little theory





PLoS Comput. Biol.
 5.22
Theory/Computational
J. Comput. Chem.
 4.080
 4.58
 Theory/Computational
J. Chem. Theory Comput.
5.22
 Theory/Computational
J. Phys. Chem. B
 3.70
Theory/Computational
 Despite lower ranking, respected journal in physical chemistry
J. Chem. Inf. Model
 3.882
Theory/Computational
 cheminformatics
J. Comput. Aid. Mol. Des.
 3.39
Theory/Computational
 Low ranking
J. Mol. Graph. Model.
 2.17
Theory/Computational
 Low ranking
(1) 2010 or 2011 impact factor listed on Wikipedia
(2) impact factor listed on www.researchgate.net if different from (1)

Nov. 5, 2012 (Mon)
Below are some tasks for the Sirtuin paper.

glidescore_vs_mm-gbsa.jpeg




October 22, 2012 (Mon)

October 18, 2012 (Thurs)

Mon., Oct. 15, 2012
Eric - The list below looks good. After the reporting of simulation results is completed I would to arrange for a discussion between you I and Karthik regarding next
steps on the follow-up simulation paper for this project. We have received a number of resumes of Indian students to partake in the NSF-funded project on sirtuin modeling and are interviewing them this week.
That grant has funding for license renewal.
EK: sounds great. I am available from 10am-6pm Mon-Fri for a follow up discussion. I want to incorporate the results into the paper and sift through the results before we have this discussion.


Thu., Oct. 11, 2012


Mon., Oct. 8, 2012


Thurs., Oct 4 2012


Mon., Oct. 1, 2012

Thurs. Sept. 27, 2012

RC: These papers seem to suggest the MCMM algorithm is designed to locate global minima and do not indicate that the algorithm is designed to compute ensemble averages. Still, the GBSA application paper that used it may have modified the input parameters to sample more extensively near the global optimum when generating the list of sequences within a specified window of the global optimum. We will make comments about these points in our paper.

Mon, Sept. 24, 2012

Thurs. Sept. 20, 2012
RC: I looked into the aforementioned paper and it seems straightforward and useful to do the ligand conformational entropy calculation using conformational sampling via the MC algorithm
implemented in Macromodel to get the probabilities of the unbound poses (and then use the Shannon expression to compute the entropy). I believe this is worthwhile to do. It is convenient
that the MC algorithm is already in Macromodel and usable through the GUI. However, I have one question before we proceed: does macromodel directly give you the probabilities of the poses,
or does it just report the global minima. I'm asking since it appears to be simulated annealing variant with a particular convergence diagnostic. Does one have to record steps along the
trajectories and parse them with a script. Recall in the case of our MC protein design code, where we were also computing entropies in an analogous fashion, we sampled conformations to get
at probabilities after convergence to stationarity, at which point we sampled trajectories to get at probabilities.
EK: Answer to your questions: yes, it saves steps of the MC trajectories that are within a certain energy window of the lowest states. The final ensemble is then chosen for the lowest 100 states of the MC trajectory. The probabilities are determined from the Boltzmann distribution with respect to the energy. Higher energy states do not need to be included because the occupancy probability is so low that they do not meaningfully contribute to the ensemble average. Note that I may want to look into whether 100 is a sufficient number of low energy states to include.
EK: I looked at a longer simulation with more than 100 lowest energy conformations. Here, a maximum of 247 conformations were within the 5.0 kcal/mol energy window from the lowest energy structure. Structures beyond the 1st 100 lowest energy conformations only contributed 3% to the total based on the Boltzmann distribution.

RC: Ok, this is the same approach we have been using in our entropy calculations with our protein design MC code. The issue I am mentioning is probably not a major concern, but the reason I was asking is that if Macromodel was not sampling from a stationary distribution after convergence, but rather returning conformations from a simulated annealing algorithm, the probabilities calculated based on the Boltzmann distribution using a partition function computed from the sampled trajectories would not be accurate, since the temperature would not be fixed, and the sampling would not be done from the equilibrium distribution. This is why we spent time (perhaps too much) developing convergence diagnostics in our code. In any case, assuming that the sampling used to collect the 100 states is at fixed temperature after annealing, it is important that the fixed temperature is set to the physiological temperature. Again I assume this is true, but it depends on whether Macromodel sampling is set up for ensemble calculations or only for minimizations. For some reason the link below is not accessible to me without a subscription, but I assume all these conditions are satisfied since Macromodel sampling appears to be automated for ensemble calculations.
EK: Your above comment could be a problem. In order to find out, I would need to read the papers about the MCMM method referenced in the MacroModel user manual. Links to these are in the task list above for someone with journal access to forward to me. Based on the abstracts for these two papers, I do not think that this MCMM method is sampling from a stationary state. Even if the distribution of low energy states is biased because they do not come from a stationary distribution, it is still better than relying on a single conformation, as demonstrated in the "MM-GB/SA Rescoring of Docking Poses in Structure-Based Lead Optimization" paper.
However, additional steps were taken to ensure that the stochastic search approached convergence: an extended protocol for torsional sampling was used, as well as energy minimization to a low gradient norm were employed. That is what the paper did, and those are the options I selected in MacroModel.

Are you computing the ensemble average energies above with \sum_i p_i E_i where p_i = \exp(-E_i/kbT)/Z, with Z = \sum_i \exp(E_i/kbT)? Or are you simply taking the sample average? Is this calculation done by Macromodel or did you write a short script for it? What about for S = -kb \sum_i p_i \ln p_i?
EK: Yes, these are the equations that I used, which assume a normal Boltzmann distribution. I copied 100 of the values from the output files to an excel spreadsheet to calculate. I will look to see if MacroModel does this automatically.

One thing we need to think about is whether these calculations belong in the Biochemistry paper and whether given its length and computational content that is typically not published in Biochemistry we should spin off another paper or submit to a more quantitative journal. We may consider having a discussion next Fri (depending on where XG is with her work).
EK: there might be enough material for a spin off paper for the computational stuff. Although, corrections, for example, of the Boltzmann distribution to a true stationary distribution may not significantly affect the final energy numbers to justify a paper (for this portion). We'll see if we recalculate based on a truly correct ensemble average energy.

Thurs. Sept. 13, 2012


Mon. Sept. 10, 2012


Thur. Sept. 6, 2012
Conformation
 MMGBSA
 Protein OPLS_2005 energy
AB
 -95.2
 -9615 kcal/mol
AC
 -92.6
 -11590 kcal/mol
Conformation
 MMGBSA
 Protein OPLS_2005 energy
AB
 -84.4
 1474520 kcal/mol
AC
 -107.9
 same as above
(XG) You are right.The schematics of the inhibition modes has been replaced by equation. I am not sure if we need to redraw them in order to put into the paper. Another thing, Eric just
uploaded the manuscript in dropbox. I proposed that we name our version by date, so that we can add our update by finding the most recent version in either Wiki or dropbox.

Mon. Sept. 3, 2012

Aug. 30, 2012

Aug. 28, 2012


Aug. 23, 2012

Aug. 20, 2012

Due Thurs., Aug. 16, 2012



Due Monday, Aug. 13, 2012

Due Thursday, Aug. 9, 2012


Thurs. Aug. 2, 2012 - Tasks for next Monday

RC: the task was to merge with the main document by this past Monday. This included getting access to whatever software was necessary to make the merge.


Th 8/3

Eric and XG,



Thanks for the Thursday updates.



Eric, I looked at your comparison of the Sir2 and SIRT3 protocols. You did a [[#|good job]] of addressing my questions regarding the steric clashes in SIRT3 that are not present in Sir2. I think that the core docking studies you have done – if clearly explained - should be sufficient for this paper, since there seems to be a simple picture provided for why SIRT3 AB pocket docking is experimentally found to be less favorable than Sir2 AB pocket docking. (However, to be more definitive about this I would need to see the rmsd of NAD+ in the AB pocket of SIRT3 for method 1) – no constraints except single excluded volume.)



Your next steps as indicated above are on mark. I will clarify further here what I would like to see.

The goal for Mon will be to integrate these figures, captions and text into the body of the manuscript.

1) As you indicate beow, we will need to be careful about explaining all steps of the docking protocols, especially those that differ between Sir2 and SIRT3 and why those changes were made. The role of different excluded volumes in Sir2 should be explained. Also, the energy degeneracy of poses differing in the position of NAM should be described – how close in energy are the different NAM conformations, for a given rmsd?


2) We will then compare binding affinities calculated on the basis of the experimental data with the MM-GBSA estimates, and compare different approaches to binding affinity estimation (based on the ppt presentation you gave).


3) [starting Thurs] We will then need to identify and briefly describe what types of additional induced fit or dynamics simulations would be required to -properly sample- the side chain and backbone [[#|degrees]] of freedom of SIRT3 in order to quantify the energetics of AB pocket binding, rather than simply indicating that AB poses were not found. This could include a description of current induced fit protocols and their shortcomings. We need to identify to what extent the shortcomings are due to inadequate backbone and sidechain sampling, versus inadequate pose sampling. In Sir2 it was the excluded volumes that enabled successful docking, not backbone or sidechain sampling; this may suggest that an insufficient number of high energy poses were sampled. On the other hand, alleviation of steric clashes in SIRT3 would of course require backbone and sidechain sampling. Is more extensive sampling therefore require in both regards? Can we say anything more specific about the modifications required to the algorithm based on the discussions you and I had? How do excluded volumes focus pose sampling – do they generate more poses in the AB pocket, or is it simply that those AB poses would not be reported otherwise? Dynamics simulations for more accurate estimation of binding affinities should also be mentioned/proposed.



For the points above, I would like to work with the latest version of the manuscript that XG had been revising – she was aligning the subsections with those of another sirtuin researcher, who I believe was publishing largely experimental enzymology papers in the journal Biochemistry. XG, can you post that latest version with comments on what restructuring has been done. Eric, we will need a complete bibliography to be included with the sections due Monday.



Risa will be looking for the Mon updates. Risa, let me know if you had any trouble accessing the submissions by Eric and XG today.



Raj










Mon. July 30, 2012

Completed [[#|work]] documented in the "Docking Simulations" section of this wiki and in these two documents:
  1. 2012.08.02_work.docfigures and answers to some of below questions.
  2. 2012.08.02_docking.protocol.docdetails of the docking protocols.

Questions/Comments

RC: Eric and XG, specific milestones should be posted to the Mon/Thurs section indicating which ones will be submitted by the next deadline. This is how Risa will check progress. Please start by listing the milestones
that will be submitted by Th. This should generally be in the form of some type of attachment that constitutes a deliverable. For Thurs I need to see completed draft versions of bullet points above of "annotated figures" and "write precise protocol" in a form suitable for incorporation (as an attachment). These drafts will be further revised and incorporated into the draft with references and bibliography for next Mon. These are the updates that Risa will check.

RC (8-6): Just an update to indicate that Risa will be checking for Eric's updates first thing Tues am. Thanks.

RC (8-10): The following tasks for Eric are recommended based on his 8-10 update. They should be added directly to the draft. Please ask for clarification where appropriate and post to the Mon task list. Rc will then decide if any should be moved to Thurs. I would not remove points 2,3,4 from Eric's Mon list for now. Points 1,5 should remain. Eric and XG should inspect each other's Mon task lists since they are related.

- Split the current discussion section into separate paragraphs describing sampling and energy calculation methods
- Describe whether/how induced fit was needed to alleviate steric inhibition of the active site Arg in Sir2
- Instead of doing MD next, identify flexible loop regions that could be sampled instead of minimized in AB pocket of SIRT3. Set up the loop predictions. Analogous protocol could be applied to Sir2. This is a prelude to tweaking the induced fit script – based on my previous discussions with you, we would describe how such loop predictions required in the modified algorithm that would be completed later. This includes the ability to combine backbone conformation change energies with binding affinities estimated by MM-GBSA. It may not be the case that simple modifications of the induced fit constraints would help.
- Outline the steps in the induced fit protocol (Thurs)
- Incorporate comments from “Sirtuin Kinetics Paper draft 5-16-2012” that explain how the simulations we have been doing clarify the possible modes of inhibition (see my comments on XG's Mon tasks regarding modes of inhibition). This will lay the groundwork for cross-referencing of experimental and computational sections in the next draft. Right now there are only 1-2 sentences mentioning how the computational results corroborate the experimental results on noncompetitive inhibition.
- Comment on priorities among remaining tasks listed in the above document (esp iso-NAM docking). XG's draft has a section on iso-NAM. Indicate whether you did iso-NAM docking. Does it bind in the C pocket? See my comments on XG's Mon tasks regarding iso-NAM.
- Enumerate the roles of the computational (sampling) tools described (in the wiki 8-10 update and in the latest draft) in future studies aimed at answering questions posed in the tasks in
“Sirtuin Kinetics Paper draft 5-16-2012” that we decided not to do in this paper. Possibly mention the possibility of induced fit docking of the Sirtris uncompetitive inhibitor of SIRT3; mention that since this is an uncompetitive inhibitor, based on the definition of uncompetitive from XG's section, there must be a conformational change that occurs upon NAD+ binding to the AC pocket.
- Comment on the challenges inherent in identifying activators via docking methods (e.g., one may need to know the rate of dissociation of iso-NAM,NAM)
-Risa will be checking Eric's tasks listed for Mon above, even though I have no response from Eric about them. Eric, please note that you will be noted as tardy on Tues morning again if not submitted.