Abstract
Background
Protein conformation and protein/protein interaction can be elucidated by solutionphase Hydrogen/Deuterium exchange (sHDX) coupled to highresolution mass analysis of the digested protein or protein complex. In sHDX experiments mutant proteins are compared to wildtype proteins or a ligand is added to the protein and compared to the wildtype protein (or mutant). The number of deuteriums incorporated into the polypeptides generated from the protease digest of the protein is related to the solvent accessibility of amide protons within the original protein construct.
Results
In this work, sHDX data was collected on a 14.5 T FTICR MS. An algorithm was developed based on combinatorial optimization that predicts deuterium exchange with high spatial resolution based on the sHDX data of overlapping proteolytic fragments. Often the algorithm assigns deuterium exchange with single residue resolution.
Conclusions
With our new method it is possible to automatically determine deuterium exchange with higher spatial resolution than the level of digested fragments.
Background
In the solutionphase Hyrdogen/Deuterium exchange (sHDX) experiment, protein surface accessibility is probed by exchange of labile hydrogen for deuterium. Simply speaking, hydrogens located at solvent exposed sites exchange at a higher rate with deuteriums from the solution than others. From these exchange rates one can therefore deduce information about protein solvent accessibility and thus protein conformation.
There is controversy surrounding the effect of D_{2}O solvent on the conformation of proteins. Sheu et al. [1] used molecular dynamic modeling of a small peptide to illustrate compaction of the peptide conformation in D_{2}O versus H_{2}O. This small compaction of the conformation occurs when the peptide is fully deuterated (which is never observed in the sHDX experiments). Since sHDX monitors the incorporation of deuterium over time the resulting slight compaction of the structure is minimized. Other methods used for the study of protein/protein interaction or protein conformation such as crosslinking [2,3] or hydroxyl radical addition [46] result in large conformational change of the protein structure; leaving sHDX as the method of choice for probing protein conformational changes in solution.
NMR spectroscopy has been the gold standard for determination of protein structure, but it has limitations on protein solubility and molecular weight (<50 kD). Solutionphase HDX with mass spectrometry analysis has higher sensitivity and is not limited by molecular weight, but sHDX is hampered with a major difficulty. One only obtains exchange data for peptic fragments and assigning exchange rates to single residues has to be done by manual interpretation.
We provide an automated method to resolve this problem. More precisely, we present an algorithm that enumerates all possible exchange rates for single residues that explain the observed data of the peptic fragments. As the number of possibilities is often very large, we combine sets of assignments to equivalence classes which are easily interpreted such that the number of equivalence classes is typically very small.
The assignment of exchange rates to single residues from the data of the peptic fragments is a combinatorial problem. Hence, we apply methods from combinatorial optimization to it, i.e. we show how to formalize the problem as an integer linear program and propose methods to solve the problem.
Biochemical Background
Concerning the determination of proteinprotein interaction, Xray crystal diffraction and NMR [7] provide the highest resolution of the sites of interaction. On the downside, both methods require large (milligram) quantities of protein. Other techniques rely on chemical or photoinduced reactions with MS analysis [8,9] to reveal functional groups that are exposed to the solvent. These methods also suffer from physical limitations.
Another method utilizes hydroxyl radical reactions with alkyl CH bonds. The OH tends to react mainly with surfaceexposed residues providing a good footprint of the solvent exposed surface of the protein(s) [4,6]. The modification is covalent and thus irreversible, but each modification can potentially change the conformation of the protein, thus skewing results.
Exchange of labile hydrogens for deuteriums (sHDX) as a probe of protein surface accessibility does not change the conformation of the protein. Advantages of MS over NMR and Xray crystallography structural determination are the ability to work at low concentration and high molecular weight.
The experiment is initiated by dilution of the protein solution into a biological buffer made with D_{2}O. Solvent accessible hydrogens are exchanged with deuterium. The exchange is quenched (greatly slowed) by dropping the pH to between pH 2.3 and pH 2.5 and lowering the temperature to approximately 0°C. The protein complex is digested with a protease that is active under quench conditions (such as pepsin) and online liquid chromatography is performed directly to the FTICR MS. Deuterium in corporation is monitored by the increase in mass of each peptic fragment as the deuterons are added.
These data sets are large, often with many overlapping proteolytic fragments. From these data, the exchange rate is easily determined for the same peptic fragments from the protein and the protein/protein complex [10] (all other fragments are disregarded). When peptic fragments are not directly comparable, but are overlapping (Figure 1) manual interpretation must be performed to assign exchange rate to single residues. Data analysis is the greatest bottleneck in sHDX experiments; thus automated data analysis is necessary. Furthermore, we are interested in all such assignments, as averaging over all solutions gives better results in practice.
Figure 1. Sample data set. Overlap of peptic fragments obtained from our sHDX on myoglobin. The table on the right shows the number of amide hydrogens predicted to be either slow, medium or fast (based on MEM). The vertical lines show the decomposition of the sequence into parts induced by the fragments. We want to automatically draw conclusions on the exchange rates of single amino acids, like the one that the second D residue has to have medium exchange rate, concluded from the restrictions imposed by fragments 3 and 5. Notice that we already deleted the Nterminal amino acid as we cannot see them exchange.
Mathematical Abstraction
In this section, we present our mathematical model for the assignment of exchange rates to residues. A brief overview of the introduced terms and symbols can be found in Table 1. In an idealized setting, we consider the following problem. We sequentially number the n residues of a protein from1 to n, beginning at the Nterminal residue and ending at the Cterminal residue. The set of peptic fragments resulting from the digestion of the protein is captured by a set ℱ of integer intervals (i, j) := {k ∈ ℕ  i ≤ k ≤ j}, for two positive integers i, j with i ≤ j, representing the endpoints of the corresponding fragment. In other words, the peptic fragment represented by (i, j) spans residues i, i+1, ..., j. Furthermore, K denotes the number of different classes of exchange rates, arising from the discretization of the experimentally measured deuterium uptake rates [11]. The K distinct classes of exchange rates, to which we simply refer as colors in the following, are represented by set . To simplify notation we number the colors from 1 to K and identify in the following the colors by their respective number. The experimentally found bulk information of how many residues within each fragment (i, j) ∈ ℱ fall into each of the exchange rate categories is given by "requirement" integers , for each fragment (i, j) ∈ ℱ and each color k ∈ . We call the vector b^{k }of "requirements" with respect to color k, indexed by fragments from, ℱ the right hand side for color k. In our experimental data, exactly three different colors are distinguished (interpreted as slow, medium, and fast exchange rates), i.e. K = 3. However, our method is not restricted to this case.
Table 1. Overview of terms and symbols
The mathematical notion introduced above is illustrated in Figure 1. There the residues, numbered from 1 to 28, are spanned by 9 peptic fragments, i.e. ℱ = 9. The third peptic fragment "VWGKVEAD" will then be represented by the integer interval (12; 19). From the experimental data we know that 5 out of the 8 residues contained in this fragment exchanged slowly (s), two at medium rate (m), and the last remaining residue exchanged fast (f) and thus , and .
Determining the exchange rate of single residues from the experimentally found data for the peptic fragments then translates into finding a "consistent" assignment of colors from to the integer points from {1, ..., n}, representing the residues of the protein, that complies with the constraints imposed by the "requirements" . More precisely, we have to determine an assignment π : {1, ..., n} ↦ such that {i ≤ l ≤ j : π(l) = k} = for all given fragments (i, j) ∈ ℱ and all possible colors k ∈ . We call such an assignment feasible.
We say that two fragments (i, j) and (i', j') overlap, if they share at least one common residue, i.e. (i, j) ∩ (i', j') ≠ ∅. The partition of the set of fragments ℱ into a maximum number of subsets, such that no two fragments from different subsets overlap, defines independent subproblems; an assignment of exchange rates to the residues spanned by the fragments of one subset does not affect the solution of a subproblem corresponding to any other subset of fragments.
Furthermore, we denote by the partition of the set of residues {1, ..., n} into maximal subsets such that residues from the same subset are spanned by exactly the same set of fragments. More precisely, for all residues i and j in the same part of and for all fragments f ∈ ℱ it holds i ∈ f ⇔ j ∈ f. Hence, for each part p ∈ and each fragment f ∈ ℱ either p ⊆ f or p ∩ f = ∅. In Figure 1 for example, residues number 7 (Q), 8 (Q), and 9 (V), are all contained in fragments number 1, 2, 6, and 8 and thus form an element of partition . Note that for the two neighboring residues the set of containing fragments differs from {1, 2, 6, 8} and therefore part = {7, 8, 9} is maximal.
However, data collected in real experiments usually contain some noise, such that no feasible assignment of exchange rates as defined above exists. Therefore, the goal is to compute all assignments that minimize the total sum of errors. Here, the error of an assignment π in fragment (i, j) ∈ ℱ with respect to color k is defined as the surplus, respectively the shortage, of residues in (i, j) that are assigned color k, compared to the number of such residues suggested by the experimental data. That is,
and thus the objective is to minimize the sum of this deviations over all fragments and colors, i.e.
In Figure 1 the colors green, yellow and red encode an optimal assignment π* of the exchange rates slow, medium and fast, with respect to objective (2). Under this assignment, fragment 3 contributes an error of 1 both w.r.t. color yellow (medium exchange rate)and color red (fast rate) to the total error of 17, while it satisfies the requirement for color green (slow rate, numbered 1) exactly.
Results and Discussion
In the following, we present different approaches to tackle the assingment problem that we have derrived from the mathematical abstraction mentioned before.
Integer linear programming formulation
First, we formulate the idealized version of the problem assuming errorfree experimental data as an integer linear program (ILP). That is, we give an ILP whose feasible solutions correspond onetoone to the feasible assignments of colors to residues.
Let π : {1, ..., n} ↦ be an assignment of colors to residues. A binary variable for every color k ∈ and every residue i ∈ {1, ..., n} indicates whether residue i is assigned color k or not, i.e.
We denote by the vector of binary variables modeling the assignment of color k and let .
Since every residue is assigned exactly one color, it must hold for all i ∈ {1, ..., n}. Conversely, every 01 assignment to variables x satisfying for all i ∈ {1, ..., n} corresponds to an assignment of colors to residues. A 01 assignment to x corresponds to a feasible color assignment π, if and only if furthermore holds for all (i, j) ∈ ℱ and k ∈ .
Now consider the problem of computing an assignment with minimum total error. Translating the definition of the error that we make when assigning color k (or not) to residues in fragment (i, j) (see equation (1)) to the context of 01 assignments to variables x^{k}, the problem of minimizing (2) becomes
Concerning the formulation of a minimum sum of absolute values in terms of a linear objective function and linear constraints, observe that is the smallest number that satisfies
Hence, after introducing a variable for every color k ∈ and every fragment (i, j) ∈ ℱ, the integer linear program we are looking at is
We refer to this integer linear program as basicILP.
In our experiments, it turns out that finding a single solution is very fast, whereas enumerating all solutions takes quite some time due to their large number. This large number can be explained as follows: Recall that is the partition of {1, ..., n} into a minimal number of parts, such that for each element p ∈ and each fragment f ∈ F either p ⊆ f or p ∩ f = ∅. In other words, no fragment starts or ends within such a part. Therefore, from an assignment π we can derive further assignments π' exhibiting the same total error, by simply permuting the colors within these parts, i.e. if i, j ∈ p for p ∈ and the total error of an assignment π is e_{1}, than π' with π' (i) = π (j), π' (j) = π (i) and π' (l) = π (l) for l ≠ i, j has total error e_{2 }with e_{2 }= e_{1}. We call two assignments equivalent, if one can be obtained from the other by iteratively applying this rule.
In order to enumerate equivalent solutions only once, we modify our integer linear program as follows: For k ∈ and p ∈ , we replace the binary variables by a single integer variable with . Moreover, let A be the ℱ ×  inclusion matrix, i.e. for every f ∈ ℱ and p ∈ , the corresponding entry is given by
We denote by the vector of errors with respect to color k and by the number of residues colored k. In matrix notation the constraints are then of the form
for all k ∈ . Hence our integer linear program becomes
where P is the vector that contains p for each component p ∈ and We refer to this integer linear program as improvedILP. We compute all solutions within a certain error bound by following basically the same approach as described above. However, the number of solutions now is just a fraction of the number of solutions of the original basicILP yielding a significant speedup
Although there is commercial software for integer programming which quickly solves instances of reasonable size, there is no algorithm that is guaranteed to find an optimum solution in polynomial time, since integer programming is NPcomplete in general. However, the problem of assigning exchange rates to residues in a way that is conform with the experimentally found bulk data exhibits a certain combinatorial structure. In the next section, we exploit this fact to derive an exact polynomialtime algorithm for the case of two colors and use it as a building block for approximation algorithms for more than two colors subsequently.
A Combinatorial Approach
First, let us consider the special case of two colors, i.e. K = 2 and thus = {1,2}. That is, we have constraints of the form for all p ∈ . This allows us to simplify the linear program considerably. We replace and omit the superscript of the yvariables in the following. This yields
where F is the vector of fragment sizes. We may get rid of half of the constraints by the following observation. Let b := max {b^{1}, F  b^{2}} and where the maximum is taken componentwise. Let y be an arbitrary feasible solution with minimum total error . We may consider the contribution of each fragment independently for that particular y. We may rename the error variables e^{1 }and e^{2 }componentwise according to b and , i.e.
For each f ∈ ℱ with , we have . If , we get . Analogously, we get if . Hence, it is sufficient to optimize the following linear program
which is integral if b and are integral since the constraint matrix is totally unimodular. The corresponding dual LP is given by
which is equivalent to (multiplying the objective function by 1 and introducing slack variables)
We will show next that this LP is a Minimum Cost Circulation Problem. To this end, let M be the matrix of the equality constraints, i.e.
Note that this matrix has the columnwise consecutiveones property. By row operations like in Gaussian elimination, we can easily transform M such that each column contains exactly one +1 and one 1, as follows. We add the dummy constraint 0 = 0 at the end and subtract from each row its predecessor. The resulting matrix, say , can be considered as the nodearcincidence matrix of a directed graph. Since the right hand side remains unchanged, we get a Minimum Cost Circulation problem on a graph with  + 1 nodes and arcs [12]. As a matter of fact, we have for each variable y_{p }two arcs corresponding to the constraint 0 ≤ y_{p }≤ p and for each fragment (i, j) the arcs (i, j + 1) and (j + 1, i) as depicted in Figure 2.
Figure 2. Fragment graph example. Example of a fragment graph with  = 7 The corresponding fragments are (1, 3), (2, 5), (3, 6), and (5, 7).
For three or more colors the complexity is open. The totally unimodularity of the constraint matrix is destroyed, i.e. there are instances with fractional vertices, e.g. the one from Figure 2 with the appropriate right hand sides. Moreover, there is an instance which has a positive error, but the value of the LP is 0. Hence the integrality gap is infinite. If the number of colors is not fixed but part of the input, the problem is NPcomplete [13].
A Simple and Efficient Heuristic for the General Case
We present an algorithm that uses our combinatorial approach for the 2color case (K = 2) from previous section as a subroutine to provide solutions that approximate (without performance guarantee) a coloring, i.e. an assignment of colors to residues, with minimum total error for instances with arbitrary but fixed number of colors. The general idea is to reduce the problem to the 2color case by merging all but one color, say color i, to a single color and solve the resulting problem by an algorithm for the minimum cost circulation problem, as described in the section about the Combinatorial Approach. We remove residues colored i by the obtained solution and solve the coloring problem on the remaining residues using K  1 colors recursively.
Our approach works as follows. Consider an arbitrary color k ∈ . We compute a subset of the residues that are assigned color k such that the total error with respect to color k and the sum of all remaining colors is minimized, i.e. we solve the two color problem with requirements (= right hand sides)
Residues assigned color k in an optimal solution to this problem will be colored k in the final solution too, the assignment of the remaining colors \{k} to the remaining residues is computed recursively.
Note that the order in which colors are selected to be the next fixed color k in the recursive computation can be arbitrary. Nevertheless, they might lead to solutions of different total error. As we have only three different colors in our experimental data, we evaluate all six orderings and return the best solution found.
In the next section we present a Lagrangian relaxation method to compute, based on our combinatorial approach for the 2color case, a bound on the minimum total error, which is exploited in a branch&bound manner to determine all optimal colorings.
A Lagrangian Relaxation Approach
In this section we propose a Lagrangian relaxation approach for the problem, which is particularly suitable for finding all optimal solutions. It is based on the improvedILP formulation:
where P is the vector that contains the length of parts in . The problem can be considered to contain independent structures for each color k ∈ , namely the set of positive integer vectors y^{k }satisfying (9) and (10) under the objective (8), that are linked by constraints (11). Therefore, dualizing the linking constraints (11), with Lagrangian multipliers λ, splits the problem into an independent problem for each color k ∈ :
Neglecting the constant term λ^{T}P in the objective function and replacing error variable e by e + ē we have to determine, for every color k ∈ , an optimal integral solution to the following linear program:
Note that we added constraint (14) to enforce or to be zero if , respectively , corresponds to the absolute value of the error, i.e. if the constraint (13), respectively the constraint (12), for fragment f is tight. Note that we have to enforce and to be nonnegative. In every optimum solution either e or ē (or both) will be zero for each fragment f. Similar as for linear program (5), its dual is given by (omitting the color superscript k):
This linear program differs from LP (7) only in the righthand sides of the equality constraints.
Conclusions
We applied our methods to process data from typical biochemical experiments. We report our results for four proteins: Calciumbinding protein (Cabin), Cytochrome P450 (CytoC), FK506 binding protein(FKBP), with two different digests (pepsin and XIII), and myoglobin. As a preprocessing step, the single fragments were analyzed with our integer linear programming based technique [14], except for FKBP V2 (MEM) which was analyzed with the MEMmethod [15] and is based on the same data as FKBP V1 (ILP). We analyzed FKBP with only the xiii digestion (V3) and combined the datasets from the two digestions (V1, V2 and V4). The number discretized exchange rates per fragment obtained in this preprocessing step serves as input to the algorithm.
The instances have between 74 and 152 residues and between 18 and 49 fragments. The solutions with a minimal number of errors could be computed in less than 0.11 seconds for all instances. Computing all (nonequivalent) solutions with a minimal number of errors, from 96 up to almost 20 million in number, took less than 7 minutes, where the running time greatly depends on the number of solutions (see Table 2). Computing all solutions using the basicILP takes much longer as with the improvedILP. For all instances, the heuristic computes a solution with the minimal error.
Table 2. Results & Runtime
Where available, we compared our assignments of exchange rates to the results obtained by NMRanalysis (FKBP and CytoC [16]). The error measure is based on a comparison per part. Within each part, the rates assigned by the algorithm are compared to the ones from NMR. Table 3 summarizes the results. The table also shows the importance of taking all solutions into account, as averaging typically yields better results than a single solution. The assignments coincide to 60  75% to the ones obtained by NMR, when choosing the optimal ordering with in the parts of equivalent residues. Figure 3 provides the results for FK506 and Cytochrome P450 at single residue resolution for manual inspection.
Table 3. Comparison with NMR
Figure 3. Single residue results for FKBP and cytochrome C. Results for FKBP (top) and cytochrome C (bottom) at single residue resolution. Rates from NMR are colored in red, our results from sHDX are colored in green. Vertical Black lines denote the boundary between consecutive parts. Legend for Symbols: '0' Rates agree on position, 'X' Rates disagree,'?' NMR data is missing, '' sHDX data is missing. The horizontal lines indicate the range of the discretized values. slow ≤ 0.1 h^{1}, and fast ≥ 8 h^{1}. Secondary structure is indicated by the horizontal bars on the top: alpha helical in blue, beta sheets in green and loops in yellow.
A structural view on the results for FK506 and myoglobin is given in Figure 4. For myoglobin we do not have NMR data at hand. Nevertheless does the figure nicely agree with the expected out come, as buried parts of the protein show on the average lower exchange rates than exposed parts. The two figures have been produced by use of PyMOL [17].
Figure 4. Structural view of results. The exchange rates (fast, medium slow) are colorcoded (red, yellow, green) in the structural view. The color blue means that we have no fragments covering this part and hence we do not predict anything. We present FKBP on the left and Myoglobin on the right.
In our solutions, the resolution is significantly increased compared to the input data, i.e., the length of fragments obtained from the sHDX experiments. The parts are typically small (see Table 2), between2 and 4 residues. 75% of the parts are smaller than 8 residues. For 46% of the aminoacids, we get singleresidue resolution on the data.
The results for the real instances are very promising, as the small number of easily interpretable classes of equivalent solutions can be used in protein structure prediction tools and for manual inspection.
Methods
In this Section, we describe the computational methods, which we use to solve the different formulation, as well as the biochemical methods to obtain the experimental data.
Solution of the integer linear program
We implemented our approach using the C++Library SCIL [18] to solve integer linear programs. SCIL uses the libraries LEDA [19] and SCIP [20].SCIP uses CPLEX [21] or SoPlex [22] as solver for linear programs. The underlying solution method is branch&bound, that is described in detail in [23].
In order to find all solutions within a given error bound e, the constraint
is added to the integer linear program and hence we are faced with the problem of computing all feasible solutions of an integer linear program. We do this with a branchingapproach similar to the classical branch&bound method for finding an optimal solution: First, the linear relaxation is solved. If the linear relaxation is infeasible, the search on this branch terminates. If the solution is integral, it is stored (provided the solution was not found yet). If there is a binary variable which was not fixed so far (i.e. not set to 0 or 1), one such variable is picked and the two subproblems, in which the variable is fixed o 0 or 1 recursively, are solved. Notice that it is possible that we branch on a variable which already has an integral value. In this case, the solution of the linear relaxation of the subproblem will be the same as in problem itself. Nevertheless, we will terminate, as there are only a finite number of variables to branch on.
Solving the Combinatorial Problem
We may use any algorithm that solves the Minimum Cost Circulation problem, e.g. Cycle Canceling or Successive Shortest Path (see [12] for further reference). Both approaches have their advantages. The former always maintains a feasible circulation, i.e. we start with the zero flow and augment flow along negative cycles in the residual network until no negative cycle remains. Since the residual network with respect to an optimal circulation does not contain a directed negative circuit, we can find node potentials, i.e. a corresponding dual solution, using the BellmanFord algorithm in time. The difference between the potential of two neighboring nodes then yields the value of the corresponding yvariable. The errors are determined straight forward. If there is a solution without error this approach yields a solution within the running time of BellmanFord. On the other hand, the Successive Shortest Path algorithm maintains similar node potentials such that the arcweights remain nonnegative. Since the total excess is bounded by  in our case, the running time of that algorithm is .
Solving the Lagrangian Dual
Instead of a minimum cost circulation problem(righthand side is 0), we have to solve the more general minimum cost flow problem [12] where the supplies and demands of the nodes are determined by the difference of Lagrangian multipliers, i.e. is of dimension  + 1 and for and . A feasible flow of minimum cost can be computed efficiently by, e.g., the cyclecanceling algorithm and the successive shortest path algorithm, as well as variants of them, like the capacity scaling algorithm [12]. In our implementation (C++) we used the LEDA library [19] to solve the Lagrangian subproblem by analgorithm based on capacity scaling and successive shortest path computation [12].
We improve the resulting bounds by the subgradient optimization method described in the following and incorporate the overall approach into a branch&bound algorithm as the lower bounding scheme.
Let v(IP (λ)) denote the optimal value of IP (λ). Then for any vector λ of Lagrangian multipliers, the (nondifferentiable) Lagrangian function
provides a lower bound on the minimum total error. To benefit from the sharpest possible bound in the branch&bound framework we are interested in solving the Lagrangian dual problem
We apply the subgradient method to obtain nearoptimal Lagrangian multipliers. Following the approach by Held and Karp [24] we iteratively determine values λ^{ℓ+1 }for ℓ = 0,1, ..., of the Lagrangian multipliers by moving in the direction of a subgradient with "step length" μ_{ℓ}:
where is any optimal solution to IP(λ^{ℓ}). The step length is computed according to formula
where UB is a previously computed upper bound on z* and θ is a step size parameter assuming values in {x ∈ ℝ  0 < x ≤ 2}. In the experiments it turns out, that initializing the vector of Lagrangian multipliers λ^{0 }to the length P of the corresponding intervals in increases the convergence rate dramatically. We also experienced a fast convergence to nearoptimal Lagrangian multipliers when following the classical HeldKarp method to choose the step size scalar θ: We start with θ_{0 }= 2 and half θ_{ℓ }whenever the best Lagrangian bound v(IP(λ)) found so far has not increased in a certain number of iterations. As soon as the step size scalar falls below a specified threshold or the number of iterations exceeds a certain limit (which is adaptive with respect to the depth of the branch&bound node), we branch on a variable , such that is close to 0.5, where is the average value of variable in the last h = 10 Lagrangian solutions. Since we aim to find all optimal colorings, we also branch on variables that are integral. Incorporating the Lagrangian approach as a lower bounding scheme into a branch&bound frame work gives an alternative algorithm that does not depend on commercial software packages.
Experimental Setting
The entire sHDX experiment was automated with a LEAP robot (HTS PAL, Leap Technologies, Carrboro, NC).
Automation of the experiment reduces human error and reduces deuterium for hydrogen backexchange. All time points where interlaced and performed in triplicate to ensure experimental reproducibility. After digestion, the protein digest was injected from a 10 μL loop to either a 1 mm × 50 mm C5 column (Phenomenex) or a ProZap Prosphere HP C18 HR 1.5u 10 mm × 2.1 mm (Alltech). A rapid gradient 2% B to 95% B in 1.5 min(A: acetonitrile/H_{2}O/formic acid 5/94.5/0.5, B: acetonitrile/H_{2}O/formic acid 95/4.5/0.5) was used to elute peptides. The eluent was postcolumn split and infused by microelectrospray ionization into a custom built 14.5 T LTQ FTICR mass spectrometer. The extraction of the peptic fragments and their deuterium uptakes from these data was done by an inhouse analysis package [25]. Then we compute the cumulative exchange rates from the deuterium uptakes with either the MEMmethod [15] or a new approach based on integer linear programming [14].
A current limitation for implementation of this software is back exchange of deuteriumtohydrogen during the separation of the samples. It has been reported that different peptides have a different percentage of back exchange due to the sequence of amino acids [26,27]. Furthermore, the peptide sequence overlap will limit the ability to map single amino acid rate kinetics. Thus, reduction of backexchange has been investigated [28,29], along with multiple acid proteases to increase sequence coverage [30]. The sHDX experiment is continually being improved, but in its current state the sHDX experiment does not take away from the integrity of the algorithm to discern single amino acid exchange kinetics.
Authors' contributions
HMZ, JT, and MRE performed the H/D exchange experiments and analyzed the data to yield the rate constant distributions from which the subsequent residue assignments were made. EA, SC, CE, and AK developed the mathematical model, performed the computational experiments and drafted the manuscript. AMB initiated and identified the mathematical approach. AGM participated in the design and coordination of the study and in preparation of the manuscript. All authors read and approved the final manuscript.
Acknowledgements
Work supported by NSF (DMR0084173), NIH (GM78359), Florida State University, the National High Magnetic Field Laboratory in Tallahassee, FL, and by Deutsche Forschungsgemeinschaft (DFG) within Priority Programme 1307 "Algorithm Engineering".
References

Sheu SY, Schlag EW, Selzle HL, Yang DY: Molecular Dynamics of Hydrogen Bonds in ProteinD2O: The Solvent Isotope Effect.
J Phys Chem A 2008, 112:797802. PubMed Abstract  Publisher Full Text

Kalkhof S, Ihling C, Mechtler K, Sinz A: Chemical CrossLinking and HighPerformance Fourier Transform Ion Cyclotron Resonance Mass Spectrometry for Protein Interaction Analysis: Application to a Calmodulin/Target Peptide Complex.
Analytical Chemistry 2004, 77(2):495503. Publisher Full Text

Fancy DA, Kodadek T: Chemistry for the analysis of proteinprotein interactions: Rapid and efficient crosslinking triggered by long wavelength light.
Proc Natl Acad Sci USA 1999, 96:60206024. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Goshe M, Chen Y, Anderson V: Identification of the Sites of Hydroxyl Radical Reaction with Peptides by Hydrogen/Deuterium Exchange: Pr evalence of Reactions with the Side Chains.
Biochemistry 2000, 39(7):17611770. PubMed Abstract  Publisher Full Text

Holmberg L, Noller HF: Mapping the ribosomal RNA neighborhood of protein L11 by directed hydroxyl radical probing.
J Mol Biol 1999, 289(2):223233. PubMed Abstract  Publisher Full Text

Sharp J, Becker J, Hettich R: Analysis of Protein Solvent Accessible Surfaces by Photochemical Oxidation and Mass Spectrometry.
Analytical Chemistry 2004, 76(3):672683. PubMed Abstract  Publisher Full Text

Zuiderweg E: Mapping ProteinProtein Interactions in Solution by NMR Spectroscopy.
Biochemistry 2002, 41:17. PubMed Abstract  Publisher Full Text

Leite J, Cascio M: Probing the Topology of the Glycine Receptor by Chemical Modification Coupled to Mass Spectrometry.
Biochemistry 2002, 41(19):61406148. PubMed Abstract  Publisher Full Text

Kang S, Hawkridge A, Johnson K, Muddiman D, Prevelige P: Identification of SubunitSubunit Interactions in Bacteriophage P22 Procapsids by Chemical Crosslinking and Mass Spectrometry.
Journal of Proteome Research 2006, 5(2):370377. PubMed Abstract  Publisher Full Text

Lam T, Lanman J, Emmett M, Hendrickson C, Prevelige P: Mapping of protein:protein contact surfaces by hydrogen/deuterium exchange, followed by online highperformance liquid chromatographyelectrospray ionization fouriertransform ioncyclotronresonance mass analysis.
Journal of Chromatography A 2002, 982:8595. PubMed Abstract  Publisher Full Text

Zhang Z, Smith DL: Determination of amide hydrogen exchange by mass spectrometry: a new tool for protein structure elucidation.
Protein Sci 1993, 2(4):522531. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Ahuja RK, Magnanti TL, Orlin JB: Network flows: theory, algorithms, and applications. Englewood Cliffs, NJ: Prentice Hall Inc.; 1993.

Althaus E, Canzar S, Elbassioni K, Karrenbauer A, Mestre J: Approximating the Interval Constrained Coloring Problem. In Proceedings of the 11th Scandinavian Workshop on Algorithm Theory. Gothenburg, Sweden; 2008.

Althaus E, Canzar S, Ehrler C, Emmett MR, Karrenbauer A, Marshall AG, MeyerBäse A, Tipton J, Zhang H: Discrete Fitting of HydrogenDeuteriumexchangedata of overlapping fragments.
The 2009 International Conference on Bioinformatics & Computational Biology 2009, in press.

Zhang Z, Guan S, Marshall AG: Enhancement of the Effective Resolution of Mass Spectra of High Mass Biomolecules by Maximum EntropyBased Deconvolution to Eliminate the Isotopic Natural Abundance Distribution.
Journal of American Society of Mass Spectrometry 1997, 8:659670. Publisher Full Text

Milne JS, Mayne L, Roder H, Wand AJ, Englander SW: Determinants of protein hydrogen exchange studied in equine cytochrome c.
Protein Sci 1998, 7(3):739745. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

The PyMOL Molecular Graphics System, Version 1.2r3pre, Schrödinger, LLC [http://www.pymol.org] webcite

SCIL (Symbolic Constraints for Integer Linear Programming) [http://www.mpisb.mpg.de/SCIL] webcite
2004.

Mehlhorn K, Näher S: The LEDA Platform of Combinatorial and Geometric Computing. Cambridge: Cambridge University Press; 1999.

für Informationstechnik Berlin KZZ: [http://scip.zib.de/] webcite

ILOGCPLEX 10.0 [http://www.ilog.com] webcite
2007.

SoPlex (Sequential objectoriented simplex class library) [http://www.zib.de/Optimization/Software/Soplex] webcite
2004.

Wolsey LA: Integer programming. Wileyinterscience series in discrete mathematics and optimization, New York: Wiley & Sons; 1998.

Held M, Karp R: The traveling salesman problem and minimum spanning trees: Part II.
Mathematical Programming 1971, 1:625. Publisher Full Text

Kazazic S, Zhang HM, Schaub TM, Emmett MR, Hendrickson CL, Blakney GT, Marshall AG: Automated Data Reduction for Hydrogen/Deuterium Exchange Experiments, Enabled by HighResolution Fourier Transform Ion Cyclotron Resonance Mass Spectrometry.
Journal of American Society of Mass Spectrometry 2010, (21):550558. Publisher Full Text

Molday RS, Englander SW, Kallen RG: Primary structure effects on peptide group hydrogen exchange.
Biochemistry 1972, 11(2):150158. PubMed Abstract  Publisher Full Text

Bai Y, Milne JS, Mayne L, Englander SW: Primary structure effects on peptide group hydrogen exchange.
Proteins 1993, 17:7586. PubMed Abstract  Publisher Full Text

Emmett MR, Kazazic S, Marshall AG, Chen W, Shi SDH, Bolanos B, Greig MJ: Supercritical Fluid Chromatography Reduction of Hydrogen/Deuterium Back Exchange in SolutionPhase Hydrogen/Deuterium Exchange with Mass Spectrometric Analysis.
Analytical Chemistry 2006, 78(19):70587060. PubMed Abstract  Publisher Full Text

Zhang HM, BouAssaf GM, Emmett MR, Marshall AG: Fast ReversedPhase Liquid Chromatography to Reduce Back Exchange and Increase Throughput in H/D Exchange Monitored by FTICR Mass Spectrometry.
Journal of the American Society for Mass Spectrometry 2009, 20(3):520524. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Zhang HM, Kazazic S, Schaub TM, Tipton JD, Emmett MR, Marshall AG: Enhanced Digestion Efficiency, Peptide Ionization Efficiency, and Sequence Resolution for Protein Hydrogen/Deuterium Exchange Monitored by Fourier Transform Ion Cyclotron Resonance Mass Spectrometry.
Anal Chem 2008, 80(23):903441. PubMed Abstract  Publisher Full Text  PubMed Central Full Text