11. Refinement: Flexible fitting

Although the rigid fitting approximates map and atomic model, a detailed visual inspection of map and model reveals that some residues are not perfectly fitted. In order to get a better fit, not only of the carbon skeleton but also of residue side chains, a flexible fitting or refinement has to be accomplished. Refinement can thus be defined as the optimization process of fitting model parameters to experimental data. Different strategies, categorized as refinement in the real space and refinement in the Fourier space, can be followed. Implemented in Scipion are two protocols for real space refinement, ccp4-coot refinement (Appendix Coot refinement [Emsley et al., 2010]) and phenix-real space refine (Appendix PHENIX Real space refine [Afonine et al., 2018]), interactive and automatic, respectively, and one automatic protocol to refine the model in the reciprocal space, ccp4-refmac (Appendix CCP4 Refmac [Vagin et al., 2004]).

Observe the new steps in the modeling Scipion workflow in Fig. 11.1.

Fig. 11.1 Scipion framework detailing the workflow to refine the model of the human Hgb \alpha subunit in the map asymmetric unit.

11.1. CCP4 Coot Refinement

Initially devoted to atomic models obtained by X-ray crystallography methods, Coot (from Crystallopgraphic Object-Oriented Toolkit) is a 3D computer graphics tool that allows simultaneous display of map and fitted model to accomplish mostly interactive modeling operations. Although this tutorial does not try to show every functionality of Coot, but indicate how to open, close and save partial and final refined structures in Scipion, some of Coot basic relevant commands will be shown. Initially, we are going to refine our model with Coot. First of all, open the ccp4-coot refinement protocol (Fig. 11.2 (1)), load the map asymmetric units (2), with electron density normalized to 1 (Coot performs this step by default), and the fitted structure model (3). To read the protocol Help is recommended. After executing the protocol (4), the Coot graphics window will appear to start working.

Fig. 11.2 Filling in the Coot refinement protocol.

To check the objects downloaded in Coot, go to the second bar of the main menu and select Display Manager. Maps (numbers #1, #2 and #3) and model Hgb_alpha_Atom_struct__3_007124.cif (number #0) are displayed on the left (Fig. 11.3 (A)). Remark that you have buttons to display a particular map (1) and to increase or reduce map density scrolling it (2). In this case, since we have selected the display of the unsharpened map asymmetric unit, we can only observe this map together with the model. If you want to check any of the sharpened maps, select it and scroll it. Note that all maps should be aligned. Try to see differences in details and connectivity of the map to assess if the sharpened maps really optimize the map density compared to the unsharpened one. If this is the case, try to follow the refinement according to the density of the best map (the most optimized one) checking the reliability of the density according to the unsharpened map, specially in the most controversial areas. Since you count on several sharpening maps you can also take advantage of the different map optimizations that you could have in the distinct areas of the map.

To start with the refinement process, we are going to identify the part of model misfitted to the density map. Visual inspection would clarify this point in some cases, although direct observation of the Density fit analysis might be a shorter way. With this aim, go to the main menu of Coot graphical window and select Validate -> Densisty fit analysis. The density fit will be analyzed regarding a specific map. To select any of them, go to the Coot right side menu (Fig. 11.3 (B)(3)) and open the Select Map for Fitting window (C). This density analysis, that you can see for the three map asymmetric units in Fig. 11.3 (D) shows that residues 1, 51, 73, 138-142 do not fit perfectly to the density map. The color range scale varies from green color (good fit) to red color (bad fit). There are some differences among maps and, as it was expected, the sharpened maps display higher restraints and show additional residues partially misfitted.

Fig. 11.3 A. Coot Display Manager. B. Coot right side menu. C. Coot Select Map for Fitting window. D. Map density fit analysis of the model in Coot regarding the unsharpened map (upper), LocalDeblur sharpened map (middle) and DeepEMhancer sharpened map (lower).

According to Fig. 11.3 (B), MET residue of the new chain A does not fit to the map density. Maybe this residue has been processed post-translationally, as we have anticipated in Input data description (Sequences section). To solve this question, go to Coot main menu and select Draw -> Go to Atom… -> Chain A -> A 1 MET (Fig. 11.4 (A)). MET residue will be located in the center of Coot graphics window. Check if this residue is surrounded by any electron density. As Fig. 11.4 (B)(1) shows, no density associates to the first chain residue. MET will thus be deleted. Then go to the lower right side menu and select the symbol to delete items (B)(2). Select Residue/Monomer in the opened Delete item window, and click the MET residue that you want to delete. Go again to Validate -> Density fit analysis and check if the orange bar shown in MET residue Fig. 11.3 (D) has disappeared.

Fig. 11.4 Removing post-translationally processed Methionine residue in Coot. Note that the icons shown in the image right side may be partially hidden if the screen is small.

Although in this particular example the most interesting refinement strategy could be repairing only the misfitted residues because they are very few, in a more general case, in which we could have many misfitted residues, an initial quick refinement may be accomplished. With this purpose, first of all, go to the upper right side menu (Fig. 11.4 (B)(3)) and select all four restrictions for Regularization and Refinement in the respective window of parameters. Secondly, open the Scipion browser (Fig. 11.5 (1)) and navigate to the extra directory, open the coot.ini text file (2), and modify the file so it matches the information shown below (3).

[myvars]
imol: 0
aa main chain: A
aa auxiliary chain: AA
aaNumber: 4
step: 10

Fig. 11.5 Editing coot.ini file.

Finally, go back to Coot window and press “U” to initiate global variables and “z” to refine the next upstream 10 residues. Go through those residues, one by one, and accept refinement if you agree with it. If you disagree with the refinement of any residue, perform the interactive refinement, visualizing the residue side chain. Repeat the refinement process with “z” until the end of the molecule. Check that the red bar of residue number 53 (Fig. 11.3) goes missing at the end of this process.

After this partially automatic and partially interactive processing, go to Draw -> Go To Atom… -> Chain A -> A 2 VAL (VAL is now the first residue of the metHbg \alpha subunit) and start the detailed interactive refinement of the initial residues of chain A. To accomplish this interactive refinement of a small group of 5 to 10 residues, select the blue circle in the upper right side menu and click the initial and final residues of the small group of residues (Fig. 11.4 (B)(4)). The group of selected residues gets flexible enough to look manually for another spatial distribution. Following these instructions, try to solve the misfit that you can find in TYR 141 residue at the end of the molecule. Specifically, try to improve the result of the Validate -> Density fit analysis, as you can see from (A) to (B) in Fig. 11.6, moving TYR 141 ((A)(1)) to the nearest empty map density ((A)(2)). Accept the refinement parameters after the displacement of TYR ((B)(3)). Finally, check the Density Fit Graph.

Fig. 11.6 Coot fit in the map density of residue TYR 141.

Rotamer refinement is another refinement tool available in Coot. You can try to improve your current model modifying rotamers reported as incorrect in Validate -> Rotamer analysis. Otherwise, the next refinement program in modeling workflow (PHENIX real space refine) will perform rotamer refinement.

At the end of this interactive refinement with Coot, the refined atomic structure has to be saved in Scipion. You can save the atomic structure with its default name/label by pressing w. If you want to add a special label to identify the atomic structure in the Scipion workflow you can save that label in Coot main menu Calculate -> Scripting -> Python and the Coot Python Scripting window will be opened and you can write there your label name, for example label1_HBA_HUMAN. This label will appear in the Summary window of the Scipion framework (Fig. 11.7 (A)). Assuming that #0 is your model number, write in Command:

scipion_write (0, ’label1_HBA_HUMAN’)

Fig. 11.7 A. Coot Summary showing label names of each independent saved atomic structure (1, 3: user’s chosen labels; 2: default label). B. (1, 2, 3) Respective atomic structure file names in the “extra” folder.

In its interactive way, ccp4-coot refinement protocol can be launched again whenever you want in Scipion, and the last atomic structure saved will be loaded in Coot graphics window. This functionality of Scipion allows to stop the interactive refinement and continue the process in the last refinement step, maintaining each one of the intermediate refined structures saved in order in the Scipion tutorial folder /Runs/000XXX_CootRefine/extra (Fig. 11.7 (B)). Remark that if you want to continue with the refinement process you have to select the Run mode option Continue when you edit the Coot refinement protocol. In this way, to go again to intermediate refined structures is also possible. Finally, when you reach the final refined structure, save it, and you may press e to fully stop the Coot protocol.

A similar refinement process to that followed in Coot for metHgb \alpha subunit chain A, has to be carried out for the metHgb \beta subunit.

Note

about chain IDs: Check the id of each chain. Although you have the possibility of changing this id in ChimeraX, as we have seen in the subsection Structural models of human metHgb subunits from templates (metHgb \beta subunit), you also have the possibility of performing this task in Coot, as it is shown in the next example in which we change the chain id from A to B. To change the name of the chain, go to the Coot main menu and select the option Edit (Fig. 11.8 (A)(1)) and then Change chain IDs and select the current name of the chain A (Fig. 11.8 (B)(2)) by the new one, B (3).

Fig. 11.8 A. Coot main menu. B. Coot window to change chain IDs.

11.2. PHENIX Real Space Refine

In order to compare the previous Coot interactive refinement with an automatic refinement, we are going to use the phenix-real space refine protocol in parallel, as indicated in Fig. 11.1 (1). In addition, we can assess if the automatic refinement obtained with the protocol phenix-real space refine is able to complement and improve the result of the Coot manual refinement (Fig. 11.1 (2)). Protocol phenix-real space refine implements in Scipion the phenix.real_space_refine program developed to address cryo-EM structure-refinement requirements. Following a workflow similar to the PHENIX reciprocal-space refinement program phenix.refine, basically devoted to crystallography, phenix.real_space_refine program, mainly used in cryo-EM, is able to refine in real space atomic models against maps, which are the experimental data.

Start working by opening phenix-real space refine protocol (Fig. 11.9 (1)), load as input volume the map asymmetric unit saved in Coot that you consider the most optimized one (2, the deepEMhancer sharpened map in this case), write the volume resolution (3), and load the atomic structure (model Hgb_alpha_atom_struct_3_007124 in the case 1 of Fig. 11.1 or model new_label_HBA_HUMAN in the case 2 (4)). After executing the protocol (6), results can be checked (7). Try to compare the MolProbity statistics that you can see in the Summary of the Scipion framework after changing the Advanced parameter Local grid search (5) from Yes to No (default value).

Fig. 11.9 Completing PHENIX Real Space Refine protocol (Case 2 of Fig. 11.1).

The first tab of results shows the initial model atomic structure (Fig. 11.10 (pink)) as well as the refined one (green), both fitted to the normalized map asymmetric unit saved in Coot.

Fig. 11.10 ChimeraX visualization of refined model of metHgb \alpha subunit by PHENIX Real Space Refine protocol (Case 2 of Fig. 11.1).

The rest of tabs detail different statistics useful to compare the quality of distinct models such as MolProbity statistics and Real-space correlations. MolProbity results will be discussed in the next section of validation and comparison. Regarding Real-space correlations, different models can be compared by using the global number of CC(mask), which indicates the correlation model-to-map calculated considering the map region masked around the model. You can check also individual correlation values for each residue. Remark that residues with lower correlation values might be susceptible to improve by additional refinement in Coot. Have a look to those correlation values in the case 1 of Fig. Fig. 11.1 and answer the following questions: (Answers in appendix Solutions; Question2)

-  What is the *CC(mask)* value?

-  Which one is the residue that shows the lower correlation value?
   Why?

-  What is that correlation value?

-  Which one is the second residue that shows the lower correlation
   value? Why?

-  What is that correlation value?

-  What is the correlation value of *HEME* group?

Now, compare these results with those obtained in the case 2 of Fig. 11.1, in which we have run PHENIX real space refine after Coot. Have the above values of correlation changed? (Answer in appendix Solutions; Question3)

The conclusion of this part of refinement in real space is that Coot and PHENIX real space refine might perform complementary tasks. The usage of both protocols may improve the result, especially when partial processing or big rearrangements of molecules are involved.

Before finishing our refinement workflow with Refmac, we can ask ourselves how we can improve correlations in real space by modifying the Advanced parameters in the protocol form. Will the correlation values change if we set to “yes” optimization parameters previously set to “no”, and increase the number of macro cycles from 5 to 30? Take into account that this process takes much more time (around 6 times more) than the previous one. (Answer in appendix Solutions; Question4)

Note

An interesting application of the PHENIX real space refine visualization tools is the possibility of load Coot from the PHENIX viewer and correct the structure of outliers residues and clashes. A recursively use of PHENIX real space refine and Coot protocols is thus possible.

11.3. PHENIX Search Fit

An extension of PHENIX Real Space Refine is phenix-search fit, a protocol implemented in Scipion to fit a small sequence of residues in a certain density of the map and, afterwards, perform the subsequent refinement in the real space (Appendix PHENIX Search fit). Let us to illustrate the applicability of this protocol with the workflow described in the Fig. 11.11.

Fig. 11.11 Scipion workflow including the phenix-search fit protocol.

This example shows a small fraction of residues from the metHgb \alpha subunit that was not completely modeled, except for the skeleton of \alpha carbons. The sequence of the chain is perfectly known, but for certain residues we were unable of tracing the lateral side chains of those residues and only ALA residues appear in our atomic structure. A detail of the small fragment of ALA residues can be observed in the Fig. 11.12 (red arrows). The protocol phenix-search fit might help us to replace the ALA residues by the appropriate aminoacids.

Fig. 11.12 Fragment of ALA residues fitted in the human metHgb asymmetric unit, as can be visualized in the protocol ChimeraX rigid fit (Fig. 11.11 (2)).

As the Fig. 11.11 indicates, the protocol phenix-search fit (4) requires three different inputs (1, 2 and 3):

1. Initial map that contains the density of the metHgb \alpha subunit. In this case we use the asymmetric unit map extracted previously (subsection Extraction of the asymmetric unit map, Fig. 7.12).
2. Small fragment of atomic structure that contains the ALA small chain. To create this fragment we start from the published atomic structure of the human metHgb \alpha subunit (included in the model of the PDB ID 5NI1, which can be downloaded from the database using the protocol import atomic structure. Next, we use the protocol chimerax-operate to isolate the chain A of the structure. The atomic structure 5NI1 is the only one input of the protocol chimerax-operate. After the opening of ChimeraX, write in the command line:

   sel #2 & ~ #2/A
   del sel
   scipionwrite #2 prefix 5ni1_chainA_
   

   After saving the chain A of the atomic structure 5NI1, run the protocol phenix-dock in map (Fig. 10.2) to fit the chain A from the atomic structure 5NI1 in the metHgb asymmetric unit map density. Next, open again the protocol chimerax-rigid fit (Fig. 10.4) and, following the previous instructions and the next ChimeraX command lines, finish the fitting, mutate the sequence between residues 94 and 118 to generate the ALA chain, and finally save the small mutated fragment:

   fitmap #3 inMap #2
   scipionwrite #3 prefix 5ni1_chainA_fitted_
   select #3 & ~ #3/A:94-118
   del sel
   swapaa #3/A:94-118 ALA
   scipionwrite #3 prefix 5ni1_chainA_94_118_MutALA_
   

3. Sequence of the metHgb \alpha subunit imported previously in subsection Sequences (Fig. 6.4).

With these three previous inputs we can complete the phenix-search fit protocol form (Fig. 11.13). Open it in the Scipion left menu (1) and include the asymmetric unit map (2) detailing its resolution (3), as well as the small fragment of mutated structure previously saved (4), the sequence downloaded (5) and take advantage of the wizard on the right (6) to select the initial and final residues that delimite the sequence to search.

Fig. 11.13 Completing the phenix-search fit protocol in Scipion.

After executing the phenix-search fit protocol (Fig. 11.13 (7)) we can have a look to the results. By pressing Analyze Results (Fig. 11.13 (8)) a window with the Viewer menu is opened (Fig. 11.14 (A)). This menu allows to visualize a certain number of atomic structures, according to their ranking scores, with lateral side chains fitted in the map density (1). Those structures will be opened in ChimeraX (2) surrounded by the density located at 3.0 Å of the structure (3). The number 1000 shown by default in (1) allows displaying all atomic structures. By pressing Summary Plot (4) a pop up window will open and show the score values of each structure, as well as the average and standard deviation of those values (Fig. 11.14 (B)). If we select the visualization of a certain number of atomic structures, 5 for example, as points the red arrow in Fig. 11.14 (C), the five best score values will appear remarked in red in the Summary Plot.

Fig. 11.14 Visualization of phenix-search fit protocol results in Scipion. A. Results menu. B. Map-model fit score plot (total number of atomic structures). C. Upper part of the Map-model fit score plot (5 best atomic structures). D. Model panel in ChimeraX showing the 5 best atomic structures selected. E. Models 3 to 8 displayed inside the map density. F. Models #3 and #4. G. Model #4 compared with the same fragment of the atomic structure 5NI1.

Panel D of the Fig. 11.14 shows the model panel with the five best selected structures, 4, 9, 11, 20 and 23. The red arrow points the position of these indexes. The respective score values are observed in red in Fig. 11.14 (C). Fig. 11.14 (E) details the view of the five structures in the GUI of ChimeraX (models #4 to #8), as well as the input fragment of ALA (model #3). Remark that lateral side chains are not shown by default. Fig. 11.14 (F) compares this input model #3 with the best score structure (model #4). To display the lateral side chains select each model and press “Show” in “Atoms” section of the toolbar. In the same way, Fig. 11.14 (F) compares this input model #3 with the respective fragment of the PDB ID 5NI1 structure. To open it as model #9 and align it with rest of structures, write in ChimeraX comand line:

open 5ni1
select #9 & ~ #9/A:94-118
del sel
mmaker #9 to #4

In spite of some small differences, most of lateral side chains align quite well with the ones of the model 5NI1 traced on the experimental map. Then, the best approximation of the atomic structure retrieved (model #4) can be selected to help with the tracing of a small fraction of the density map since the model #4 seems to be quite close to the actual atomic structure used as control (PDB ID 5NI1).

11.4. CCP4 Refmac

As in the case of Coot, Refmac (from maximum-likelihood Refinement of Macromolecules) was initially developed to optimize models obtained by X-ray crystallography methods but, unlike Coot, automatically and in reciprocal space. The models refined in the real space with Coot and PHENIX real space refine, successively, will be used as inputs to perform a second refinement step in the Fourier space with Refmac protocol ccp4-refmac. Firstly, open the Refmac protocol form (Fig. 11.15 (1)), load the volume generated by Coot (2), the atomic structure obtained with Coot (case 3 of Fig. 11.1) (3) or with PHENIX real space refine after Coot (case 4 of Fig. 11.1), and the volume resolution as maximum resolution (4). Execute the protocol (5) and when it finishes, analyze the results (6).

Fig. 11.15 Filling in Refmac protocol (Case 3 of Fig. 11.1).

Clicking the first item in the display menu of results (Fig. 11.16 (1)), ChimeraX graphics window will be opened showing the input volume, the initial model (new_label_HBA_HUMAN) obtained with Coot (Fig. 11.17, pink), and the final Refmac refined model (Fig. 11.17, green). By clicking the third item in the display menu of results (Fig. 11.16 (2)), a summary of results are shown. Check if values of R factor and Rms BondLength have improved with this refinement process in these three cases:

• Running Refmac after Coot:

  Can you see an improvement running Refmac immediately after Coot, thus ignoring model improvements generated by phenix-real space refine? (Answers in appendix Solutions; Question5)
• Running Refmac after phenix-real space refine after Coot:

  Why the improvement seems to be very small? (Answers in appendix Solutions; Question6)
• Running Refmac after phenix-real space refine without a mask:

  Compare previous Refmac results (after Coot and phenix-real space refine) with those obtained selecting the option No in the protocol form parameter Generate masked volume. Use two different volumes, the one generated by Coot protocol, and the one generated by the extract asymmetric unit protocol. Are there any differences? Why? (Answers in appendix Solutions; Question7)

Fig. 11.16 Display menu of Refmac results.

Fig. 11.17 ChimeraX visualization of refined model of metHgb \alpha subunit by Refmac (Case 3 of Fig. 11.1).

Have a look to the rest of items in the display window of results.

11.5. The best refinement workflow

At this point we wonder about the optimal steps to follow in the refinement process. Should we have to use Coot first, then PHENIX, then Refmac?, or maybe, with a different map and model, should we start with the automatic refinement and then go to the interactive one? The right answer is that there is no a unique answer. The strategies and the number of steps of refinement might differ and the only requirement is that the next step in refinement should generate a better structure than the previous one. This premise requires to apply common validation criteria to assess the progressive improvement of our model.