Anomalous X-ray diffraction studies of ion transport in K+ channels

Potassium ion channels utilize a highly selective filter to rapidly transport K+ ions across cellular membranes. This selectivity filter is composed of four binding sites which display almost equal electron density in crystal structures with high potassium ion concentrations. This electron density can be interpreted to reflect a superposition of alternating potassium ion and water occupied states or as adjacent potassium ions. Here, we use single wavelength anomalous dispersion (SAD) X-ray diffraction data collected near the potassium absorption edge to show experimentally that all ion binding sites within the selectivity filter are fully occupied by K+ ions. These data support the hypothesis that potassium ion transport occurs by direct Coulomb knock-on, and provide an example of solving the phase problem by K-SAD.

These three bound waters need reassignment to ions, their identity as a cation or anion according to their charged neighbour.
So my scrutiny of the underpinning data suggests some improvements to the protein model are possible and maybe also appropriate comments made in the article or supplementary details, as indicated above.
Reviewer #2 (Remarks to the Author): This paper presents an x-ray structure of a K selective mutant of NaK. The main focus of the paper is K occupancy in the selectivity filter. A K SAD experiment is used to determine absolute occupancy. The data are processed with the phenix.refine program to give an occupancy of ~0.25 (x 4) = 1.

Some technical questions:
How was f'' of 'around 4 electrons' determined? Was this measured directly? If so, please provide data on it.
While the randomization of ADP and occupancy values shows convergence to occupancy ~ 0.25, and the authors state the importance of decoupling ADP and occupancy, what was really done to do so? (i.e. what's inside phenix.refine?) Reference 22 used Tl data from reference 7 to show occupancy = 1 by similarly running a program (Shelx), but without a deliberate effort to decouple occupancy and ADP. But in ref 7 the authors carried out alternate cycles of ADP (what was called B factor) and occupancy refinement (always fixing one while refining the other) and they found Tl occupancy = 0.63. Did you consider such an approach.
As I see it, there are 2 key technical complications in occupancy refinement. The first is getting the number of f'' electrons right (ideally by measuring it) and the second is decoupling occupancy and ADP to the best of one's ability (unfortunately they are inconveniently correlated). This paper does not describe how either of these things were done.
I raise a final question for the authors to think about, and perhaps comment on.
How does a conduction mechanism with 4 fully occupied K sites account for the well-documented thermodynamic coupling of water and K movement through a K channel (i.e. streaming potential)?
We thank the reviewers for taking time to help us improve our manuscript. We have thought carefully about all the points raised, below is a point by point explanation of how we responded to each reviewer comment. Author comments are in black text with our response following in blue text. We have added this reference into the introduction of the revised article.

Ion permeation in K+ channels occurs by direct
The anomalous difference Fourier map is outstandingly excellent. One has to simply marvel at it!:-I commend that the above two figures are added to the article ideally in the main section or if needs be for article length reasons in the supplementary.
We have added a figure into the manuscript ( Figure 2) which shows the anomalous difference Fourier map for ion channel.
" Figure 2: The anomalous difference Fourier map contoured at 8s is shown as a magenta mesh for subunit A.
Strong anomalous difference peaks corresponding to the K + ions (cyan spheres) are present within the selectivity filter of the ion channel. For clarity only two of the four subunits that make up the ion channel are shown." What is the peak 3 in the above screen shot? A bound water? [Ie since there is no anom peak on it.] It looks as if it may be functionally interesting. A comment is needed in the article. Water.
We have added a water molecule at this position into the structure and we added this sentence into the manuscript.
"On the top and bottom of the K + ions in the selectivity filter are bound water molecules, the structure and experimental data has deposited into the protein data bank with the accession code 6DZ1." These three bound waters need reassignment to ions, their identity as a cation or anion according to their charged neighbour.
The three bound water molecules have been reassigned as K + ions in the structure.

Reviewer #2 (Remarks to the Author):
This paper presents an x-ray structure of a K selective mutant of NaK. The main focus of the paper is K occupancy in the selectivity filter. A K SAD experiment is used to determine absolute occupancy. The data are processed with the phenix.refine program to give an occupancy of ~0.25 (x 4) = 1.

Some technical questions:
How was f'' of 'around 4 electrons' determined? Was this measured directly? If so, please provide data on it. At the time of the experiment the beamline fluorescence detector had not yet been fully commissioned, so a quantitative measurement based on a XANES scan was not possible. The Xray energy was chosen 90 eV above the theoretical absorption edge to avoid any potential white line effects in the XANES region. At 3.7 keV (3.35 A wavelength) the theoretical value for f" is 3.9 e-, in the text we state "around 4 e-" indicating that there is some uncertainty about the absolute value. Hence, we are confident that avoiding the XANES region gives us only a small uncertainty / discrepancy from between theory and experiment, not affecting the conclusions made in the manuscript. Source: http://skuld.bmsc.washington.edu/scatter/data/K.dat We have added the following sentences into the manuscript to address this point.
"We collected a complete dataset on the long wavelength beamline I23(24) at the Diamond Light Source synchrotron at 3.35 Å. This wavelength is close to the K absorption edge (3.44 Å), resulting in a very strong anomalous signal (Table 1) from a theoretical anomalous contribution f" of 3.9 electrons from K(25)." And we added these sentences into the supporting information " At the time of the experiment the fluorescence detector at the beamline had not yet been fully commissioned. Therefore, rather than optimizing the anomalous contribution based on an energy scan across the absorption edge and subsequent quantitative analysis of the scan to determine f", we tuned the X-ray energy to 3700 eV, 91.6 eV above the tabulated potassium K absorption edge (3608.4 eV). This is far enough in energy from the near edge region (XANES) characterized by large fluctuations of f" due to resonance effects within the specific coordination sphere of the potassium atoms. While the absolute value of f" is slightly reduced further away from the absorption energy, this approach allows using the theoretical approximation of 3.9 electrons(25) for sufficiently accurate anomalous occupancy refinements." While the randomization of ADP and occupancy values shows convergence to occupancy ~ 0.25, and the authors state the importance of decoupling ADP and occupancy, what was really done to do so? (i.e. what's inside phenix.refine?) phenix.refine refines occupancies and ADP (or B-factors) separately at all times, as well as coordinates. This is documented in phenix.refine paper (http://journals.iucr.org/d/issues/2012/04/00/ba5180/ba5180.pdf) a refinement run consists of macro-cycles, with each macro-cycles consisting of independent steps of refining coordinates, occupancies, B-factors and other atomic and non-atomic model parameters.
Reference 22 used Tl data from reference 7 to show occupancy = 1 by similarly running a program (Shelx), but without a deliberate effort to decouple occupancy and ADP. But in ref 7 the authors carried out alternate cycles of ADP (what was called B factor) and occupancy refinement (always fixing one while refining the other) and they found Tl occupancy = 0.63. Did you consider such an approach.
phenix.refine refines occupancies and ADP (or B-factors) separately at all times, as well as coordinates. As documented in phenix.refine paper (http://journals.iucr.org/d/issues/2012/04/00/ba5180/ba5180.pdf) a refinement run consists of macro-cycles, with each macro-cycles consisting of independent steps of refining coordinates, occupancies, B-factors and other atomic and non-atomic model parameters. So the approach outlined above seems to be an inherent feature of phenix.refine. We have added the following sentence into the manuscript to clarify this point.
" Phenix.refine refines occupancies and ADP (or B-factors) separately at all times(30)." As I see it, there are 2 key technical complications in occupancy refinement. The first is getting the number of f'' electrons right (ideally by measuring it) and the second is decoupling occupancy and ADP to the best of one's ability (unfortunately they are inconveniently correlated). This paper does not describe how either of these things were done.
We have added extra sentences into the manuscript detailing how the number of f'' electrons were determined in response to an earlier comment.
"We collected a complete dataset on the long wavelength beamline I23(24) at the Diamond Light Source synchrotron at 3.35 Å. This wavelength is close to the K absorption edge (3.44 Å), resulting in a very strong anomalous signal (Table 1) from a theoretical anomalous contribution f" of 3.9 electrons from K(25)." The decoupling of occupancy and ADP refinement is an inherent feature of phenix.refine which refines occupancies and ADP (or B-factors) separately at all times, as well as coordinates.
We have added the following sentence into the manuscript to clarify this point.