During the mid-2010s a company called Sierra Analytics released a groundbreaking hydrogen deuterium exchange (HDX) analysis software called HDExaminer (Trajan then acquired the software business in 2018). Although it was not the first software title dedicated to processing HDX samples, HDExaminer had some distinct advantages over other HDX software solutions at the time. One major advantage was that it was vendor agnostic, and so as a result, it would work with a wide range of MS instruments. Moreover, the software had a user-friendly GUI, democratising HDX analysis to a wider range of researchers who might not be as confident with coding or using command-line tools (employed by some existing software solutions at the time). Related to this, was the inclusion of key visual features that HDExaminer offered such as deuteration level residue maps, protein state comparisons, individual peptide uptake plots, isotope cluster visualization and colour-coded confidence levels. As well as being a robust platform for handling complex data such as overlapping peaks and the incorporation of rigorous statistical analysis features, HDExaminer could integrate with LEAP (now Trajan) Automation. This allowed for fully automated HDX experiments from sample prep though to analysis, and as a result Trajan CHRONECT HDX systems are a gold standard and utilised globally in many pharmaceutical companies, CROS and research institutions.
One of the limitations for most of the current HDX processing software solutions, is the time it takes to process samples post analysis. This is a problem, because until recently, the technique could not be employed in high throughput environments, which then limits its use to more of a research tool. Each HDX experiment requires at least a day of sample preparation and data acquisition (LCMS), followed by several days or even weeks of data processing and validation. The reason for this is not due to computer processing limitations, rather it is due to the need for human data curation. Not only is this time consuming but a high skill level is also required to effectively refine HDX data. Moreover, as humans are not infallible, mistakes can be made which lead to incorrect identification of deuterated peptides and potential misassignment of deuterium uptake values.
The reason why human curation data curation takes such a long time to complete is due to the nature of the data acquisition approach. Historically, this approach has been to use the MS1 dimension only as opposed to the more recent MS2 data independent acquisition (DIA) approach. The need for double checking the uptake values derived from MS1-only data is expected given the difficulties facing short LC gradients where peptides co-elute frequently.
The first step in HDX analysis is to generate a peptide map using MS2-DDA. This step is also necessary to check digestion efficiency and sequence coverage before additional deuterated data files are to be acquired. Once the map is created, software such as HDExaminer uses MS1 full scan data from HDX experiments to look for known peaks and then tracks the centroid mass shift of their isotopic envelopes which often leads to peak broadening. This is where human intervention is needed, because there are other factors which lead to MS peak broadening. Primarily this is due to coeluting peptides and spectral noise interfering with the deuterated peptides. Presently, software such as HDExaminer is not yet sophisticated enough to automatically and fully reliably conduct the data curation in the MS1 dimension alone. For example, although HDExaminer has some data correction and curation algorithms, manual checks are often necessary to ensure correct peak assignment and deconvolution. Manual checks are also sometimes needed to verify peak integration and filter out bad spectra where S/N is poor. Moreover, because not all peaks will behave the same, some problematic peptides require retention time windows and or m/z tolerances to be individually tuned. Even in cases where the user intervenes, there is still the possibility of significant MS1 peak overlap which cannot be resolved, leading to the eventual loss of some peptides. Finally, in terms of back exchange, although the software is able to calculate such rates using fully deuterated controls, it is also prudent to manually check these calculations.
One of the challenges for performing HDX analyses is mitigating a phenomenon known as protium (often called ‘hydrogen’ in this context) / deuterium scrambling. Scrambling is observed under energetic environments such as in the collision cell and also when conditions in the source are energetic enough. When this occurs exchangeable protiums and deuterium isotopes can swap locations randomly across the peptide. This means that the exact original location of each original exchange becomes less resolved, and moreover this phenomenon gets further exacerbated the larger the peptide fragment is. Although there are solutions to mitigate at-source scrambling, such as reducing the energetics of the source, these mitigations also can negatively impact on the sensitivity of the results.
What this all means is that until recently, only non-deuterated control samples (for peptide map creation) are fragmented into the MS2, but not the subsequent deuterated HDX samples. The reason for this is that the energetics of the collision cell have the potential to fully scramble the product ions. The amount of extra uptake measurements would make human data curation very problematic if not impossible because there are tens of thousands of additional deuterium measurements (one per fragment).
A way to overcome issues with the traditional method for processing HDX experiments is to use the data independent acquisition (DIA) approach (a relatively recent development in proteomic data analysis). Essentially DIA uses both MS1 and MS2 data for HDX. Rather than just relying on retention times and targeting specific peptide fragments in the MS1 (from first identifying this from a peptide map), DIA adds another dimension of complexity to the HDX experiment.
In a DIA experiment, the mass spec fragments everything not just specific targeted ions. This is achieved by splitting the m/z range into large pre-defined isolation windows (for example 20Th). However, unlike a traditional DDA approach where a selected precursor is allowed to pass through into a collision cell to obtain product ions specific to that precursor, the DIA approach sends each isolation (window) into the collision cell. This results in a far more complicated mix of product ions from all the of the precursor ions in MS1 from that specific window. Using existing methods from the proteomics field, the software then deconvolutes each of these windows (using data from peptide databases) and essentially (in silico) pairs the individual product ions with their precursors.
The benefit of the DIA approach is that it is more democratic as lower abundant ions are allowed to pass through into MS2 after collision induced dissociation (CID). This means that more precursor peptides are identified which then leads to a more resolved peptide map. Furthermore, by having product ions paired to their precursors also leads to a greater laver of certainty of the precursor peptide identification. Most important, the additional measurements coming from fully-scrambled DIA fragments are themselves measurements of the peptide uptake which can turn a single measurement into multiple measurements of the same uptake. This redundancy enables one to either endorse the original MS1 peptide deuteration with the MS2 data or override it in cases where the MS1 data was unresolvable. Finally, because modern computers can rapidly process vast amounts of data, there is far less need for human intervention which is a huge benefit to HDX data interpretation and so massively speeds up the process.
To recap, when conducting an HDX experiment, we are looking for sites on a protein were protium deuterium exchange is slower or eliminated. This gives us structural information about the protein, protein dynamics and also identifies sites for ligand or protein-protein binding such as epitope mapping. The HDX experiment allows us to compare a non-deuterated peptide with the same peptides where deuteration has occurred. If we repeat this experiment over several timepoints, we can then understand the rates of deuteration and in doing so can identify hotspots where there is a difference in exchange kinetics. When applied to LC-MS/MS analyses, peptide maps are created and compared to each other. Historically this was achieved using the aforementioned MS1 approach, where identified deuterated peptide digestate fragments in MS1 are compared to an undeuterated control. Both software and humans then collaborate to identify these peptides and measure positive shifts in their M/Z values showing the degree of deterioration.
The biggest issue with the MS1-only approach is the lack of certainty associated with identifying peptides, their degree of deuteration and if they are contaminated with peptides of the same M/Z range. As mentioned above, the reason for this is the only reference the user and software has to work with, is the original peptide map of the precursor.
DIA however, exploits scrambling protium - deuterium precursor isotopes in the collision cell. Like traditional proteomics using DIA, these fully scrambled product ions are then knitted back together (in silico) to identify their precursor ions. In doing so, a statistical measurement of the degree of precursor deuteration is elucidated and furthermore, location of the precursor in the protein sequence can then be identified from the peptide map.
Published in Nature Communications on the 11th of March 2024, [i] a group headed up by David Schriemer at the university of Calgary pioneered a DIA-based HX-MS2 workflow and developed a software which they entitled AutoHX. AutoHX (now renamed HDExaminer PRO after Trajan acquired the technology in 2024) is an app within Mass Spec Studio (again acquired by Trajan at the same time and renamed to CHRONECT Studio) which acts to mine HX data using two dimensions (MS1 precursors and MS2 fragments) using a DIA approach. They first generate a peptide library from DDA runs and following this, use their proprietary search engine (HX-PIPE) to find unambiguous peptide assignments. With this data they then format each library ready for HDX experiments. To ensure that as much of the sequence is captured and to prevent isotope loss effect near isolation window edges, they allow the DIA windows to overlap.
To test peptide redundancy using the DIA approach, the team analysed a 6-point kinetics profile of phosphorylase B in triplicate. From this they identified 380 precursor molecules and 3269 usable fragments. After, they chose precursors which only had ≥3 fragments as this gave a greater than 10 fold sequence coverage and showed a decrease in SD by 40% with the precursor data. By mapping out all possibilities of a given peptide and looking at the centroid data from the distribution profile, they were able to assess the accuracy of deuteration but also the likelihood that the distribution was an identified peptide. Simply put, a narrower distribution increases the likelihood of correct peptide identification and accurate deuterium uptake. This is a key software feature that eliminates the time-consuming human data curation step employed in MS1-only HDX analysis. Having completed this for all identified signatures, the group found that when comparing heatmaps to a standard manual MS1-only processing, the DIA generated data was found to be almost identical.
Having successfully demonstrated the high efficacy of the automated data curation algorithm, the group then tested the approach in a real world scenario by looking at a ligand binding experiment. The mechanism they chose to examine was observe the binding of the antibiotic novobiocin which interacts with DNA polymerase enzymes including Pol ϴ which is upregulated in 70% of breast and epithelial ovarian cancers. The group ran three experiments, the first was a traditional DDA approach with no human intervention, then this was repeated with human curation (2 individual expert analyses) and then HDExaminer PRO was used as the 3rd scenario. The results showed the importance of human data curation when conducting conventional HDX. However the automated DIA generated data produced maps which were almost identical to the current best conventional practice.
For the next experiment, the team compared the enhanced sensitivity of the DIA approach to a standard cryo-EM data set. The target was affinity isolates of ATP drug binding characteristics of the very large DNA protein kinase (4,128 amino acids) and its drug candidates. Due to the nature of using affinity isolates, 100L of cell culture was required for cryoEM. However, only 2 x10cm plates worth of culture was required for DIA HDX using HDExaminer PRO. Furthermore, on the subject of improved throughput using the automated DIA approach, the authors commented the following.
“We then naïvely generated a Woods plot from the six HX-MS experiments using the entire peptide list (Fig. 9a), revealing a complexity that would take days or weeks of manual curation to correct. AutoHX was able to process all six samples in 10min on a single high-end desktop computer.”
By using protium deuterium scrambling to our advantage, it is now possible to run HDX experiments using a DIA approach and so exploiting MS2 product ions to automatically confirm identity of target peptide molecules and validate their deuterium uptake. The pioneering work conducted by David Schriemer and team, has developed a technique which is no short of revolutionary in nature. Not only does this democratises the HDX workflow by conducting all of the data-analysis heavy lifting, but now as a result, a complete HDX experiment can be analysed in a tiny fraction of the time. What once used to take days, now takes hours even minutes to complete with no loss of data quality. This massively increases the applicability of HDX, allowing much higher throughput applications such as QC of biological therapeutics
N.B. For further reading, recently the same team from the University of Calgary, publishing in Analytical Chemistry (April 2025), explored instrumental factors influencing the performance of DIA [ii]. This is a fascinating read which includes a deep-dive into the importance of protium deuterium scrambling as well as optimising M/Z DIA windows for optimal data capture.
REFRENCES:
[1]. Filandr, F., Sarpe, V., Raval, S., et al. Automating data analysis for hydrogen/deuterium exchange mass spectrometry using data-independent acquisition methodology. Nat Commun 15, 2200 (2024). https://doi.org/10.1038/s41467-024-46610-3
[1]. Filandr, F., Hepburn, M., Sarpe, V., et al. Examining Instrumental Factors Influencing the Performance of Data-Independent Acquisition Methods in Hydrogen/Deuterium Exchange Mass Spectrometry. Analytical Chemistry 97, 8011–8020 (2025). https://doi.org/10.1021/acs.analchem.5c00429