pocketz.doi.bio/2026-01-11_181935_nature_s41589-025-02047-3
Unfreezing structural biology for drug discovery | Nature Chemical Biology
URL: https://www.nature.com/articles/s41589-025-02047-3
Saved: 2026-01-11T02:19:40.674Z
Perspective Published: 05 November 2025 Unfreezing structural biology for drug discovery Timothy R. Stachowski & Marcus Fischer
Nature Chemical Biology (2025)Cite this article
4052 Accesses
1 Citations
16 Altmetric
Metrics details
Abstract
Structure-based drug discovery relies on three-dimensional protein structures to provide the atomic blueprints for small-molecule design, indicating where to place each atom to maximize favorable interactions. The advent of cryo-cooling crystals in crystallography greatly accelerated the ease and accessibility of structural data, making it a mainstay of most drug discovery efforts. However, despite its successes, including producing numerous clinically successful molecules, cryo-cooled samples only tell part of the structural story: they may leave out dynamic details or introduce artifacts that may lead drug discovery campaigns astray. In this Perspective, we highlight recent studies characterizing temperature-sensitive structural phenomena observed by crystallography. We showcase how leveraging information on rare, hidden conformational states informs ligand discovery via molecular docking. This demonstrates the value of performing structural studies at elevated temperatures, closer to where biology occurs, to ‘unfreeze’ structural ensembles for drug discovery and design.
You have full access to this article via your institution.
Download PDF Main
Despite their deceptive interpretability, protein structures are riddled with omissions and errors1,2,3. Their intrinsic flexibility4,5 and ability to dynamically ‘shapeshift’ across an ensemble of states allow proteins to carry out biological functions6. Capturing this flexibility in structural models creates untapped4 opportunities for developing small-molecule effectors through structure-based drug design (SBDD). We argue that one less-considered barrier for harnessing flexibility in SBDD is that most crystal structures are modeled from diffraction data collected at cryogenic (cryo) temperatures (≤150 K), which introduces artifacts and hides conformational heterogeneity. Ironically, macromolecular crystallography was initially performed at ambient or room temperature (RT), but the practical benefits of cryo-cooling to reduce X-ray radiation damage7 and the ease of transporting crystals to synchrotrons quickly outweighed concerns from solution and crystalline experiments indicating that cryo-cooling restricts conformational ensembles8,9,10. Comparative studies published by our group11,12,13 and others14,15,16,17 show atomistically that cryo-cooling suppresses protein motions14, reshuffles water networks13 and introduces both global and local structural artifacts18 (Fig. 1) that complicate ligand discovery efforts11.
Fig. 1: Cartoon overview of temperature-dependent changes at various scales discussed in this Perspective.
a, B factors describe the variation of an atom around its average position. B factors often increase with temperature, signaling increased mobility or model omissions. b, The crystal unit cell typically contracts upon cryo-cooling. c, This can change the packing interface between symmetry-related protein copies and drive structural rearrangements. d, An example is shifting loops that function as binding pocket gates. e, Side chains frequently switch rotameric conformations with temperature. f, Typically, cryo-cooled structures contain about twice as many water molecules as their RT counterparts. ‘Freezing out’ water molecules can obfuscate binding pockets that might be of interest for small-molecule design. g, In concert with or independent of changes in the protein structure, ligand poses are temperature sensitive. h, In extreme cases, ligands may populate different sites in structures solved at cryo versus RT.
Full size image
We recently discussed how RT experiments can be performed without sacrificing the conveniences of conventional cryo experiments (for example, transporting crystals) or resolution (Fig. 2) and without occluding binding sites with cryoprotectants or artificially disturbing the intrinsic conformational heterogeneity7. A helpful guide by Thorne19 reviews the physical principles and considerations for optimizing RT diffraction experiments. Beamlines at several light sources20,21 are now outfitted for performing both conventional single-crystal and serial experiments so that proteins can be studied across environmental conditions such as temperature22, humidity23, pH24, pressure or a combination thereof25,26 to reveal new mechanistic insights. While still technically challenging, these advances also allow users to collect data not just at endpoints (such as cryo and RT) but along a series of points to track the evolution of structural changes or identify intermediate states6,18,27,28. These emerging modalities offer untapped potential to yield new insights for drug discovery29, as ~95% of all crystal structures in the Protein Data Bank (PDB; Fig. 2a) as well as yearly depositions7 are from cryo-cooled crystals.
Fig. 2: Structures are quantitatively different at cryo versus RT.
a,b, A survey of the PDB shows that there are far more cryo (≤150 K) structures available than RT (270–310 K) ones (a), yet both have similar resolution distributions (b). From this, we found ~200 examples in which a structure of the same protein is available at both cryo and RT, with the same space group and ligand at similar resolutions. c,d, From this dataset, we observed that UC volume (c) and protein volume (d) change between cryo and RT.
Full size image
The challenge of considering dynamics in SBDD is exacerbated by the fact that protein flexibility is currently underrepresented in experimental models in the PDB despite its importance in biology. More often than not, protein models only describe a few select, static states, even though signatures for additional ones are captured in experimental datasets30,31,32. One reason for this reductionism is that the signal for flexible high-energy features, such as alternative conformations of side chain rotamers or water networks, at any temperature, is often weak. This makes it challenging to accurately discern genuine signal from experimental noise, which becomes increasingly problematic as crystallographic resolution decreases. Even with sufficient signal31, matching density to specific structural states is difficult. Recent automated multiconformer modeling programs33,34,35,36 enrich experimental models with flexible features and the structural biologist’s toolbox with mainstay tools.
In campaigns that probe protein–ligand interactions, small-molecule binding can be especially tricky to decipher due to conformational flexibility37,38, near-uniform atomic composition or binding with coexisting states39 and overlapping water networks. These frequently occur in screens of fragments that bind efficiently but weakly. We argue that a more comprehensive consideration of structural complexity, enabled by RT crystallography, opens up new opportunities for tackling targets of diseases with currently unmet clinical needs40. Moreover, molecular docking campaigns are now routinely exploring chemical space on unprecedented scales to identify compounds with novel chemistry41. Infusing dynamic structural data into this pipeline promises new routes to target specific conformational states that allow fine-tuning biological activity and selectivity.
In this Perspective, we highlight the impact of temperature on protein, ligand and water structural ensembles through the lens of drug discovery and design, including the impact on computational applications, particularly molecular docking. These considerations are broadly relevant for researchers consulting structural information during drug design. Considering that these advances give deeper insights into the protein conformational landscape, we envision that anyone collecting and exploring structural data will warm up to the idea that unfreezing structural biology opens new doors for drug discovery.
Shaking up protein structures with temperature
To avoid being misled by cryo artifacts in crystallographic experiments, it is essential to develop temperature-paired datasets, that is, closely matching low- and high-temperature datasets obtained under otherwise identical conditions. Yet, our survey of the PDB shows that only some 200 targets have temperature-matched structures (Fig. 2). While this number may vary slightly depending on how stringently temperature pairs were matched, the takeaway remains the same: temperature sensitivity remains largely unexplored for many protein systems. One recurring theme we anecdotally observe in paired datasets is that, at near-identical resolution and conditions, the quality of the RT model, as judged by Rfree, is typically superior to that of the matching cryo structure13. This suggests that RT diffraction data, free from cryo-cooling artifacts19, are of higher quality and thus are easier to describe with realistic models. For similar reasons, RT data tend to be more robust and reproducible. A compelling example of this was seen in dihydrofolate reductase, for which two RT diffraction datasets collected almost a decade apart were more consistent than the corresponding cryo ones18. As we discuss later, the accuracy and quality of structural models have widespread implications for downstream applications, such as molecular dynamics, quantum calculations, molecular docking and increasingly machine learning, that pave the way for drug design campaigns.
Global effects
Starting from a bird’s eye view, crystallographic B factors are indicators of positional variability reflecting the displacement of atoms from their mean position and, with careful effort, can illuminate areas with correlated motions. However, B factors do not describe specific coordinate-based transitions. This means that they often go unappreciated by those who think of crystallographic data solely as a three-dimensional model. Adding complexity, inflated B factors can signal either increased dynamics or modeling errors42. Either way, they are worth considering. B factors also track the impact of temperature on protein crystals: they increase steadily above the glass transition temperature (200 K)9,43 (Fig. 1a), indicating that dynamics are suppressed at cryo and liberated at higher temperatures.
Anecdotal13,14 and systematic analyses of temperature-paired systems in the PDB (Fig. 1b) show that cryo-cooling shrinks both the unit cell (UC) volume and the protein volume14 (Fig. 2c). This overpacking has three major consequences. First, it exaggerates crystal contacts that may result in structural shifts from entire domains down to the side chain level. Second, overpacking may mask functionally relevant pockets or surfaces that are important for explaining biological mechanisms or that may be targetable with small molecules. Third, overpacking alters water networks that stabilize protein structure. Again, such overpacking effects are consequential but invisible without an RT reference structure. For instance, changes in UC volume with temperature often artificially immobilize loops at cryo that are dynamic in solution. This is surprisingly common across diverse protein systems, ranging from subtle conformational changes that enable functional transitions in an ion channel44 to more pronounced networks of correlated motions of hinges and loops in the enzyme dihydrofolate reductase that propagate throughout the protein to the active site26. Freed from cryo constraints, entire backbones may shift (Fig. 1c) in addition to reshuffling loops and hinges that change the properties of a binding site45 (Fig. 1d). An illustrative example is the loop transitions of the biomedical target protein tyrosine phosphatase 1B (PTP1B), which occur between cryo and RT46 (Fig. 3a). Notably the populations of ‘open’ versus ‘closed’ were inverted in the temperature pair, from a mainly ‘closed’ state at cryo to a mainly ‘open’ state at RT (278 K). Determining structures not just at temperature endpoints but also at intermediate temperatures allowed tracking the development of weakly populated substates and cryptic sites that guided a subsequent fragment-screening campaign. Both binding of a covalent small molecule and mutagenesis showed that one cryptic binding site allosterically modulates enzyme activity. In addition to this and other ligand discovery efforts47, such RT-enabled multiconformer models have also proven useful for guiding protein engineering and design48,49.
Fig. 3: Loop rearrangements with temperature modulate binding sites.
a,b, Example of temperature-dependent changes of a loop in PTP1B in which the population of open versus closed is inverted46 when going from cryo (a; PDB 6B90) to RT (b; PDB 6B8X). c, Example of revealing hidden loop states in the apo structure of CCP at elevated temperature that are relevant for ligand binding and discovery47 (PDB 4NVM). For a,b, 2Fo–Fc maps are shown in blue for cryo and orange for RT and are contoured at 1σ. For c, the 2Fo−Fc map is shown in cyan and contoured at 1σ and the Fo−Fc map is shown in green and red and contoured at ±3σ. A, B and C correspond to alternative loop states. Buffer (MES) and heme (HEM) are present in the active site.
Full size image
Aside from inverting conformations, several studies show that RT structures often contain additional conformations that are absent at cryo. In the case of cytochrome c peroxidase (CCP)47 (Fig. 3b), three loop conformations that mediate ligand binding were simultaneously present in RT apo electron density maps but only individually in cryo holo models. Incorporating all three states into a multiconformer RT model enabled estimation of the occupancy (that is, the relative abundance) of each state, which was then converted into Boltzmann-weighted energies and included in the docking scoring function. In a small proof-of-concept docking campaign of ~600,000 virtual molecules, molecules that were preferential to particular RT conformations were prioritized for experimental testing. Crystal structures confirmed that molecules were able to selectively bind distinct loop conformations, illustrating how novel conformations identified in RT data can be directly leveraged in ligand discovery.
Side chains
Estimates propose that up to 45% of side chains in high-resolution crystal structures occupy multiple conformations14,31,50. Alternative side chain conformations are often nearly isoenergetic51, which means that they likely repopulate in response to environmental cues such as temperature14 or ligand binding52. Yet, for reasons such as detection and model-building issues, this is generally not reflected in PDB models. Tellingly, a systematic re-evaluation of deposited electron density maps of temperature-paired models revealed that 35% of side chain conformations were redistributed between RT and cryo14. As structures at RT are conformationally less restricted, they often visit additional states that are absent at cryo (Fig. 1e), although we also observe the opposite scenario14. The impact is not necessarily quantifiable by the total number of differences. For instance, changing just a single binding site residue side chain can drastically impact molecular docking performance, as we see later. Yet, predicting the temperature sensitivity of residues is an outstanding question in the field.
To address this, we developed Flipper to track two types of side chain conformation reshuffling in maps of a perturbed-versus-unperturbed dataset, such as a cryo–RT or holo–apo pair13 (Fig. 4a). First, identifying electron density peaks tallies the number of conformations that are gained or lost between datasets across all residues. Second, detecting ‘flips’, that is, residues for which the number of conformations is unchanged but the relative strength of electron density changes, pinpoints populations of major and minor conformations that are inverted across structures. This property often goes unnoticed by a simple visual inspection of the model but provides different blueprints, as structures are automatically prepared for molecular docking as described later. Beyond revealing local changes, Flipper produces ‘conformational barcodes’ that map temperature- or ligand-sensitive regions across the entire protein. From this, we learned two things. First, temperature changes can impact protein structure more than ligand binding. In other words, structures of the same protein solved at different temperatures were less similar than ones bound to different ligands at the same (room) temperature. Second, although we were initially motivated to develop predictive patterns describing temperature-driven side chain conformational changes, our barcoding approach rather supported that cryo-cooling reshuffles side chain conformations idiosyncratically. The culprit is likely differing rates of cryo-cooling, which is hard to control during the manual plunge-freezing process53. This inadvertently and inconsistently ‘cryo-traps’ non-equilibrium states. While ultimately artifacts, without an RT-paired structure, these states are unknowingly treated as ground truth in cryo models.
Fig. 4: Side chain rotameric changes with temperature affect binding sites.
a, Comparing paired datasets using Flipper13 provides structural barcodes of gains and losses in conformations as well as changes in the relative population of side chains that can be mapped onto the protein surface. b, Example of a temperature-dependent loss in conformation but a gain in a direct interaction of Lys58 with HSP90α in the presence of the potent inhibitor EC144. The 2Fo−Fc maps are shown in blue and are contoured at 1σ (cryo, PDB 7S9H; RT, PDB 7S9I).
Full size image
For instance, in the cancer target heat shock protein 90α (HSP90α) bound to the clinical candidate EC144, the ligand-adjacent residue Lys58 occupied two conformations at cryo, one of which was locked in place by three ordered water molecules (Fig. 4b). At RT, Lys58 is liberated from a set of sterically restrictive water molecules at cryo and occupies a single conformation, now directly hydrogen bonding with the ligand. For drug discoverers, there are two important takeaways. First, it shows that temperature repopulates binding sites and affects the interpretation of ligand-binding mechanisms, clarifying that Lys58 is a key residue to consider at RT while no direct involvement in ligand binding was seen at cryo. Consequently, ligand optimizations would be led down divergent paths. Second, using these different molecular blueprints as input in docking campaigns would reshuffle docking hit lists and lead to different prioritization of small molecules for experimental testing.
Multiple temperature-driven rotameric changes can even coalesce into alternative allosteric networks54 or suppress biologically important ones. An example of this comes from a study on the oncogene HRAS in which the RT structure uncovered an allosteric network consistent with fluctuations detected in solution by NMR. This network was not visible in the cryo electron density maps, indicating that it was quenched upon cooling14.
Whether in isolation or as part of a contact network, alternative rotamers of side chains are worth paying attention to, as they are widely distributed across proteins13 and often alter sites of interest, be it orthosteric or allosteric. In addition, genuine non-rotameric side chains reflective of ‘unusual’ high-energy states may be important for biological mechanisms. At cryo, these are often artifacts of exaggerated crystal packing and as such are easily dismissed. RT structures may capture these hidden high-energy states with less ambiguity as to their relevance46.
Water networks
With a plethora of alternative protein conformations comes a convoluted network of alternative water positions. Water molecules are key to biological function and, unsurprisingly, exquisitely temperature sensitive (Fig. 1f). Anecdotally, RT models contain about half of the ordered water molecules found in cryo counterparts18,29, with weakly bound and high-B factor water molecules generally more likely to be absent at RT24. A decrease in solvation at RT typically corresponds to an increase in protein conformational heterogeneity, suggesting that the greater solvation at cryo sterically restricts protein motions. Displacing water at RT impacts the flexibility of both protein side chains13 (Fig. 5a) and backbone18 (Fig. 5b). Even water molecules that appear to be conserved across temperature can undergo small but meaningful positional shifts, gaining or losing hydrogen bonds in the process13,55. Such slight but consequential differences can provide meaningful handles for improving ligand affinity and selectivity56 but are rarely used, as they are difficult to routinely track. Having a clearer understanding of the properties governing water displaceability by temperature might also yield dividends as a roadmap for ligand design55.
Fig. 5: Water networks are exquisitely sensitive to temperature.
a,b, In addition to Fig. 4b in which Lys58 changes conformation upon liberation of three cryo water molecules in HSP90α, the disappearance of a cryo water molecule near Asn105 (a) allows reorienting the residue at RT (b)13. c,d, Beyond side chains, cryo water conformationally restricts protein motions18. For a,b, 2Fo−Fc maps are contoured at 1σ (cryo, PDB 7S9H; RT, PDB 7S9I).
Full size image
Water molecules in binding sites can also be displaced by temperature, although this may be challenging to anticipate. The Lys58 example in HSP90α showed how solvent-driven rearrangements in a binding site alter protein–ligand interaction mechanisms and paths for ligand discovery and optimization (Fig. 4a). Another example of solvent-driven rearrangements in a binding site residue comes from the mutated Asn11 side chain in SiaP, which has one conformer at cryo and three at RT. Variations in rotameric state occurred alongside shifts in the binding site water network. Notably, the most occupied RT conformer points 180°, in the opposite direction of the cryo conformation, yielding binding sites with distinct properties12. As we see later, the temperature sensitivity of water networks has implications for fragment-based ligand discovery57 and for exposing cryptic sites22.
Reshuffling protein–ligand interactions with temperature
Among the temperature-sensitive structural features that are apt to be altered in unpredictable ways (Fig. 1), ligand binding is no exception. Focusing on binding sites, we cover two scenarios: temperature-driven changes in orthosteric (Fig. 1g) or allosteric (Fig. 1h) ligand binding (Fig. 6). Next, we explore how these phenomena coalesce into different outcomes in crystallographic fragment screens57.
Fig. 6: Ligand-binding sites repopulate with temperature.
a, Schematic showing typical changes in temperature-responsive ligand behavior; the reference cryo ligand pose is shown in thin lines. b, Examples of each scenario. The 2Fo−Fc maps are colored dark blue at 1σ and cyan at 0.3σ. Fo−Fc maps are shown in green and red and contoured at +3σ and −3σ, respectively (PTP1B: cryo (PDB 5QDS), RT (PDB 7FRM); AmpC: cryo (PDB 4LV0), RT (PDB 3BLS); T4 lysozyme: cryo (PDB 4W53), RT (PDB 7L39); CCP: cryo (PDB 4NVE), RT (PDB 4XV5)). JGG, BZI and APB correspond to PDB ligand codes.
Full size image Orthosteric ligand binding
Temperature alters protein conformations and water networks, ultimately redefining binding site properties. Consequently, ligands may gain or lose interactions. We explore four scenarios we observe in RT studies: (1) shifts in ligand pose, (2) changes in pose with different interactions, (3) multiple coexisting poses and (4) ligands (dis)appearing with temperature changes (Fig. 6a).
In the first scenario, temperature drives a structural rearrangement of the binding site; the ligand binds nonetheless, but the interaction pattern is altered. Our HSP90α–EC144 example from earlier illustrates this point (Fig. 4a). A second example comes from a fragment bound to PTP1B, in which a bridging water with Asp252 is lost at RT (Fig. 6b). Consequently, the ketone moiety flips to engage Arg238 at the other side of the pocket.
In the second scenario, from our survey of the PDB, we identified a pair of fragment-bound structures of the antibiotic target AmpC β-lactamase from Escherichia coli (Fig. 6b) with a drastic change in ligand pose. The tendency of cryo-trapped water molecules to suppress structural dynamics is reinforced here; at cryo, we see a dense water network in one half of the binding site and the fragment bound to the other half. At RT, this water network is replaced by the fragment, which pivots around the covalent adduct with Ser64. This creates a virtual mirror image of the cryo pose and a different interaction pattern. Moreover, the original cryo pose is not present at RT nor is it replaced by water, exemplifying increased solvent dynamics at RT. Again, as we identified this pair from the PDB, it reinforces the benefit of having more matched cryo–RT temperature data available in the PDB to isolate the impact of temperature over other involuntary perturbations.
In the third scenario, we frequently see multiple ligand poses at RT, that is, the original cryo pose is retained but a second pose is revealed. In our work on temperature-paired T4 lysozyme structures across a series of congeneric ligands, one of the fragments, toluene, occupied two conformations at RT (resolution of 1.4 Å) but only one at cryo (1.5 Å)11 (Fig. 6b). The two conformations were overlapping except for the position of a single methyl group. The presence of both methyl positions was supported by a structure bound to ortho-xylene, which recapitulates the double-methyl position of toluene within a single molecule. It was perplexing then, that the cryo structure would also not contain both toluene poses. This led us to solve a new cryo toluene structure at a much higher resolution (1.1 Å), which did in fact contain the second pose that was initially identified at RT. These subtle differences clarify that the single-ligand model overemphasizes the importance of its interactions with the protein lid and that conformations present at RT might be masked at cryo. Overall, revealing apparently subtle features at RT is consequential as it provides biological insights and indicates possible directions for ligand growth or for generating bioisosteres.
Allosteric ligand binding
Another scenario to consider is emergent binding of a ligand to an alternative site, such as a cryptic or allosteric site (Fig. 1h). Allosteric sites are desirable for drug discoverers, as they provide opportunities to introduce selectivity handles into ligands that may be missing in conserved orthosteric sites or for bypassing competition with native substrates that monopolize active sites58. Beyond inhibition, agonists targeting allosteric sites hold opportunities to activate and modulate function.
As noted previously, cryo-cooling generally shrinks protein volume, causing structural rearrangements including the immobilization of gating loops. This can mask pockets relevant for small-molecule intervention. A powerful way to identify potential cryptic sites is through variable-temperature crystallography. For instance, when increasing temperature stepwise from 100 K to 310 K (Fig. 3a), a gating loop in PTP1B opened, revealing a small pocket46. The conformational changes driving pocket formation appeared to be allosterically coupled to the active site, indicating that small-molecule binding to the site would inhibit activity. Indeed, over a dozen small molecules occupied the cryptic site in a subsequent crystallographic fragment-screening campaign46.
Collecting a temperature series on a single crystal, first at RT and then at cryo, is technically challenging but eliminates any doubt of compounding factors causing the observed differences. First, it reduces idiosyncrasies across datasets. Second, it enables estimating the energetic penalty of cryo-cooling on binding based in part on the change in ligand occupancy. In a notable example of emergent ligand binding, benzimidazole bound to a hidden cryptic site in CCP only at RT22 (Fig. 6b), with the canonical orthosteric site occupied at both temperatures. Binding to the cryptic site at RT occurred via a rotameric change of His96 and the displacement of several water molecules. Put another way, cryo-cooling drove structural repopulations that occluded binding. Additional examples of weak RT binding events that are overpowered by the energetic penalty of cryo-cooling is discussed in the next section.
Crystallographic fragment screening
Fragment-based ligand design (FBLD) is a popular approach that promises to rapidly design affinity into initial weak fragment hits59. Despite initial skepticism, the challenge of increasing molecular complexity of ligand-efficient60, low-affinity ligands is nowadays often more appealing to medicinal chemists than stripping back high-affinity but more complex, bigger hits. Among biophysical approaches to screen fragment libraries, crystallography shines, as it delivers the binding location and pose61,62. FBLD is almost universally conducted at cryo because scaling up crystallographic screens at RT needs bespoke automation. One such comparative study57 gives a glimpse of what we might be missing.
To probe the effect of temperature on a typical hit discovery campaign, the authors screened 143 fragments for the target PTP1B using RT crystallography. RT fragments were a mixture of binders and nonbinders from an earlier, broader cryo screen. Surprisingly, only about a third of fragments observed to bind at cryo also bound at RT. The same sites across the protein were sampled by fragments at both temperatures, but some fragments exhibited alternative poses or occupied different sites between temperatures. Even in cases in which the site and pose seemed to be unaffected by temperature, there were rearrangements of the surrounding water networks, creating different interaction patterns. RT binders generally had lower occupancies, which is in line with our own observations11. Similar to our earlier T4 lysozyme example, this study included one case in which a fragment occupied the same site and pose at both temperatures, but an additional flipped pose was observed at RT. In instances in which the same fragment bound to different sites, it occurred alongside conformational changes in the protein.
Overall, this highlights how individual temperature-dependent phenomena coalesce to impact FBLD campaigns. Ultimately, cryo structures may miss binding events or underestimate the significance of weak interactions that are only observable at RT, underscoring the need to integrate RT crystallography into fragment-screening workflows.
Heating up computational drug discovery
Molecular docking picks up structural data where crystallographic modeling has left off and predicts the complementarity of a small-molecule ligand with a target protein. Compared to experimental high-throughput screening, molecular docking samples a much larger area of chemical space using less time and resources63. Comparisons with experimental structures also show that docking routinely predicts binding poses correctly64. Recent technological advances allow billions of molecules to be docked in a single discovery campaign41,65, making it a powerful approach to identify new medicinal chemistry starting points, even for well-studied targets.
Equipped with a better understanding of the caveats of temperature and structural modeling, we must face yet another obstacle: molecular docking is crude. Driven by the ambition of screening billions of virtual small molecules, docking programs make, by design, vast approximations in the terms they consider to assess ligand binding. Often these are limited to a simple combination of van der Waals, electrostatic and desolvation terms for ligand scoring66. This oversimplification of the ligand-binding event has well-known consequences67 that are generally accepted or supplemented computationally with higher levels of theory. To save computational resources and time when screening vast virtual libraries, a typical docking campaign starts by deleting all minor conformations (typically defined by occupancy) of the protein. Yes, those ones we have revealed at RT and carefully modeled one atom at a time. This reductionist practice is frustrating, as it changes all the scoring terms (van der Waals surface and desolvation penalties) as well as the binding pocket volume; Lys58 in HSP90α (Fig. 4) is but one example. Docking campaigns focusing only on the major conformation and ignoring temperature-sensitive conformations consequently enrich for different output ligand chemistry. Ignoring a minor RT conformation that is absent at cryo seems subtle. However, overlooking a unique affinity or selectivity handle that has the potential to distinguish related structures within a protein family with high structural and sequence similarity is a much less subtle outcome. Again, without temperature-paired structures, we do not know what we are missing.
To illustrate this, let us consider four scenarios of side chain conformational heterogeneity that temperature-paired studies often reveal (Fig. 7). First, most excitingly, RT structures may show a new conformation that is not present at cryo. This may be a new major conformation or a minor, high-energy one with low occupancy (Fig. 7b). Either way, a new conformation provides a new blueprint and opportunity for ligand design. In the second scenario, the opposite occurs and the cryo structure shows more conformations than at RT (Fig. 7c). The presence of a single conformation at RT may help clarify which cryo conformation to focus on during ligand discovery and design. Third, a single conformation is observed at both temperatures, but the rotameric state differs. In the most dramatic case, rotamers may point in opposite directions, providing different anchors for interactions (Fig. 7d). Finally, on preliminary inspection, a side chain may seem unaffected and occupy the same set of conformations at both temperatures. However, a closer look uncovers that the occupancies invert or flip between temperatures so that the major conformation changes (Fig. 7e). This has implications for how rotamer labels are assigned in the structure, which in turn influences how features are automatically and often unknowingly selected by docking programs.
Fig. 7: Examples of exploitable temperature differences for docking.
a, No structural changes between temperatures and docking performance is similar. b–e, In b, docking is in apparent agreement across temperature but does not exploit the novel minor RT conformation (conf); in c–e, docking input differs with temperature. Differences in occupancy are indicated by stick width, with thick lines indicating the major conformation that is typically selected in automated docking procedures. Electron density is indicated by light blue clouds and notably ignored during docking. Exemplary retrospective enrichment performance, measured as area under the curve (AUC), reports on the impact of changing input structures.
Full size image
Ideally, one would hope to leverage the temperature-sensitive structural diversity we have revealed experimentally via docking. However, examples of including flexibility into docking by weighing states by their relative energies are still rare47,68,69, more exceedingly so in the era of ultra-large-scale docking. In the spirit of cutting computational corners, avoiding permutations of multiple alternative conformers de-escalates compute time. But even for classic single-conformer docking, the input of structures obtained at different temperatures impacts performance metrics. One study to investigate these differences systematically collected pairwise data on T4 lysozyme and monitored docking performance in terms of two common metrics, enrichment and pose fidelity11. Notably, enrichment metrics differed substantially between temperatures. Unsurprisingly, the state of the protein matters, be it ligand bound or apo. Surprisingly, the ligand-unbiased apo RT state was preferred in this study. But it becomes more complicated, as changing docking algorithms completely flipped the script and favored the cryo structure over the RT one. We speculate that this may not be entirely coincidental, as empirical docking scoring functions were trained almost exclusively on cryo data that dominate structural databases. With docking being such a crude tool, it is hard to emerge with solid recommendations, but higher levels of theory confirmed that an equilibrated apo RT structure may provide a worthwhile starting point, for example, for binding free-energy calculations11.
Finally, retrospective enrichment metrics of docking performance that are used to optimize the docking protocol may ‘hide’ the true value of RT models. While docking a set of ‘known’ ligands against ‘known’ structures will generally lead to better enrichment, new, previously ‘hidden’ conformations might be incompatible with the set of known ligands but fish out new chemistry. So, for prospective ligand discovery, using an RT input structure that changes the chemical composition of the hit list may be desirable, even if the retrospective performance is worse.
Current docking developments mainly focus on increasing the size of on-demand virtual libraries while taking algorithmic shortcuts to enable screening efficiency. Indeed, the ‘bigger is better’ mentality seems to deliver more hits with better affinities and chemical diversity. While including hidden alternative conformations will come at an increased computational cost, we argue that it may be an expense worth paying relative to the timeline of a typical drug discovery campaign.
Outlook and conclusions
Analogous to expanding chemical space for virtual screening, we argue that it is time to expand the ‘protein conformational space’. This can be accomplished not only by filling gaps in the PDB with artificial intelligence (AI)-generated models, which has led both to unfulfilled expectations70,71 and successes72,73,74,75,76,77, but also by enriching pipelines with multi-state and multi-temperature models that more richly represent and explore the targetable protein conformational landscape. Aside from crystallography, cryo-electron microscopy (cryo-EM) is another approach to accomplish this goal78. Cryo-EM is the predominant source for most new membrane protein structures such as G-protein coupled receptors79, which were historically inaccessible to crystallography and are the target of 34% of small-molecule drugs80. Much of this is a result of the ‘resolution revolution’ (ref. 78), in which cryo-EM maps now regularly approach the resolution and quality of their crystallographic counterparts81 without the need for crystals, revealing fine details such as water networks82 and alternative side chain conformations83. As impactful as these results are, it is important to remember that a key part of this revolution was the discovery that cryo-cooling samples reduces molecular motion and thus increases micrograph interpretability84: sound familiar85,86? From lessons learned with crystallography, cryo-cooling is likely to imprint bias on cryo-EM structures. However, without baseline uncooled results such as those in crystallography, it is unclear exactly what the specific effects are. Emerging approaches in time-resolved cryo-EM hint at sample differences at different freezing rates using laser revitrification87. Using molecular dynamics simulations as a ground-state surrogate showed that cooling restricted cryo-EM structural ensembles due to increased thermal contraction, reduced thermal motion and equilibration into lower free-energy conformations88. Some studies are performing ‘temperature-resolved’ cryo-EM to track the effects of ‘elevated’ temperature on the functional lifecycle of enzymes89 and transporters90. While the results of these studies are exciting, we should be cautious in interpreting them as ground truth, as they may nonetheless be tainted by cryo-cooling artifacts.
Following the ‘garbage in/garbage out’ mantra, regardless of the technique (be it crystallography, cryo-EM or AI), downstream applications of structural data require accurate models. Errors and omissions, arising as artifacts from cryo-cooling, may misdirect drug discovery efforts. For non-crystallographers, it is worth remembering that the model is a somewhat subjective interpretation of the electron density map, which is improved in iterative cycles of model building and refinement and eventually abandoned when converged but never truly ‘completed’. A crystallographic map, however, holds all the information of peaks and troughs of electron density that crystallographers place atoms into (or not). Also, whether those atoms correspond to the protein, ligand or water is up for (mis)interpretation. Hence, we strongly advise non-crystallographers to dig deeper into both map (for example, difference features) and model (for example, occupancies, B factors) before chasing ‘crystallographic ghosts’ and embarking on a decade-spanning rational SBDD campaign. As we explored with docking, seemingly small structural changes (such as a single rotamer) are consequential and can alter the direction of a hit-finding campaign. So, structural biologists: warm up to solving your structures at physiological temperature. This is where biology takes place, where functional effects are tested experimentally and where your models will be most impactful for drug discovery, and machine learning and AI predictions.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
A list of temperature-matched protein structures used to perform the analysis in Fig. 2 is freely available at https://github.com/TheFischerLab/RTSBDDperspective.
References
Gao, Y., Thorn, V. & Thorn, A. Errors in structural biology are not the exception. Acta Crystallogr. D Struct. Biol. 79, 206–211 (2023).
Article CAS PubMed PubMed Central Google Scholar
Fraser, J. S. & Murcko, M. A. Structure is beauty, but not always truth. Cell 187, 517–520 (2024).
Article CAS PubMed PubMed Central Google Scholar
Deller, M. C. & Rupp, B. Models of protein–ligand crystal structures: trust, but verify. J. Comput. Aided Mol. Des. 29, 817–836 (2015).
Article CAS PubMed PubMed Central Google Scholar
Teague, S. J. Implications of protein flexibility for drug discovery. Nat. Rev. Drug Discov. 2, 527–541 (2003).
Article CAS PubMed Google Scholar
Wei, H. & McCammon, J. A. Structure and dynamics in drug discovery. npj Drug Discov. 1, 1 (2024).
Article Google Scholar
McLeod, M. J., Barwell, S. A. E., Holyoak, T. & Thorne, R. E. A structural perspective on the temperature dependent activity of enzymes. Structure 33, 924–934 (2025).
Article CAS PubMed Google Scholar
Fischer, M. Macromolecular room temperature crystallography. Q. Rev. Biophys. 54, e1 (2021).
Article PubMed Google Scholar
Rasmussen, B. F., Stock, A. M., Ringe, D. & Petsko, G. A. Crystalline ribonuclease A loses function below the dynamical transition at 220 K. Nature 357, 423–424 (1992).
Article CAS PubMed Google Scholar
Tilton, R. F. Jr., Dewan, J. C. & Petsko, G. A. Effects of temperature on protein structure and dynamics: X-ray crystallographic studies of the protein ribonuclease-A at nine different temperatures from 98 to 320 K. Biochemistry 31, 2469–2481 (1992).
Article CAS PubMed Google Scholar
Halle, B. Biomolecular cryocrystallography: structural changes during flash-cooling. Proc. Natl Acad. Sci. USA 101, 4793–4798 (2004).
Article CAS PubMed PubMed Central Google Scholar
Bradford, S. Y. C. et al. Temperature artifacts in protein structures bias ligand-binding predictions. Chem. Sci. 12, 11275–11293 (2021). This study demonstrates that the impact of temperature on structure cascades to the performance of computational drug discovery approaches that use structural information such as molecular docking and binding free-energy calculations.
Article CAS PubMed PubMed Central Google Scholar
Darby, J. F. et al. Water networks can determine the affinity of ligand binding to proteins. J. Am. Chem. Soc. 141, 15818–15826 (2019).
Article CAS PubMed Google Scholar
Stachowski, T. R., Vanarotti, M., Seetharaman, J., Lopez, K. & Fischer, M. Water networks repopulate protein–ligand interfaces with temperature. Angew. Chem. Int. Ed. Engl. 61, e202112919 (2022). This study demonstrates that temperature drives structural rearrangements in binding sites that can alter the interpretation of ligand-binding mechanisms using a series of HSP90 crystal structures bound to clinically relevant inhibitors.
Article CAS PubMed PubMed Central Google Scholar
Fraser, J. S. et al. Accessing protein conformational ensembles using room-temperature X-ray crystallography. Proc. Natl Acad. Sci. USA 108, 16247–16252 (2011).
Article CAS PubMed PubMed Central Google Scholar
Keedy, D. A., Fraser, J. S. & van den Bedem, H. Exposing hidden alternative backbone conformations in X-ray crystallography using qFit. PLoS Comput. Biol. 11, e1004507 (2015).
Article PubMed PubMed Central Google Scholar
Wolff, A. M. et al. Mapping protein dynamics at high spatial resolution with temperature-jump X-ray crystallography. Nat. Chem. 15, 1549–1558 (2023).
Article CAS PubMed PubMed Central Google Scholar
Yabukarski, F. et al. Ensemble–function relationships to dissect mechanisms of enzyme catalysis. Sci. Adv. 8, eabn7738 (2022).
Article CAS PubMed PubMed Central Google Scholar
Keedy, D. A. et al. Crystal cryocooling distorts conformational heterogeneity in a model Michaelis complex of DHFR. Structure 22, 899–910 (2014).
Article CAS PubMed PubMed Central Google Scholar
Thorne, R. E. Determining biomolecular structures near room temperature using X-ray crystallography: concepts, methods and future optimization. Acta Crystallogr. D Struct. Biol. 79, 78–94 (2023).
Article CAS PubMed PubMed Central Google Scholar
Cohen, A. E. A new era of synchrotron-enabled macromolecular crystallography. Nat. Methods 18, 433–434 (2021).
Article CAS PubMed Google Scholar
Martiel, I., Muller-Werkmeister, H. M. & Cohen, A. E. Strategies for sample delivery for femtosecond crystallography. Acta Crystallogr. D Struct. Biol. 75, 160–177 (2019). This study offers practical guidance on sample preparation and X-ray diffraction data collection at RT for both single and serial crystallography.
Article CAS PubMed PubMed Central Google Scholar
Fischer, M., Shoichet, B. K. & Fraser, J. S. One crystal, two temperatures: cryocooling penalties alter ligand binding to transient protein sites. ChemBioChem 16, 1560–1564 (2015).
Article CAS PubMed PubMed Central Google Scholar
Atakisi, H., Moreau, D. W. & Thorne, R. E. Effects of protein-crystal hydration and temperature on side-chain conformational heterogeneity in monoclinic lysozyme crystals. Acta Crystallogr. D Struct. Biol. 74, 264–278 (2018).
Article CAS PubMed PubMed Central Google Scholar
Correy, G. J. et al. The mechanisms of catalysis and ligand binding for the SARS-CoV-2 NSP3 macrodomain from neutron and X-ray diffraction at room temperature. Sci. Adv. 8, eabo5083 (2022).
Article CAS PubMed PubMed Central Google Scholar
Guerrero, L. et al. Pushed to extremes: distinct effects of high temperature versus pressure on the structure of STEP. Commun. Biol. 7, 59 (2024).
Greisman, J. B. et al. Perturbative diffraction methods resolve a conformational switch that facilitates a two-step enzymatic mechanism. Proc. Natl Acad. Sci. USA 121, e2313192121 (2024).
Article CAS PubMed PubMed Central Google Scholar
Keedy, D. A. Journey to the center of the protein: allostery from multitemperature multiconformer X-ray crystallography. Acta Crystallogr. D Struct. Biol. 75, 123–137 (2019).
Article CAS PubMed PubMed Central Google Scholar
Doukov, T., Herschlag, D. & Yabukarski, F. Obtaining anomalous and ensemble information from protein crystals from 220 K up to physiological temperatures. Acta Crystallogr. D Struct. Biol. 79, 212–223 (2023).
Article CAS PubMed PubMed Central Google Scholar
Dunge, A. et al. Exploring serial crystallography for drug discovery. IUCrJ 11, 831–842 (2024).
Article CAS PubMed PubMed Central Google Scholar
Shapovalov, M. V. & Dunbrack, R. L. Jr. Statistical and conformational analysis of the electron density of protein side chains. Proteins 66, 279–303 (2007).
Article CAS PubMed Google Scholar
Lang, P. T. et al. Automated electron-density sampling reveals widespread conformational polymorphism in proteins. Protein Sci. 19, 1420–1431 (2010).
Article CAS PubMed PubMed Central Google Scholar
Wankowicz, S. A. & Fraser, J. S. Comprehensive encoding of conformational and compositional protein structural ensembles through the mmCIF data structure. IUCrJ 11, 494–501 (2024).
Article CAS PubMed PubMed Central Google Scholar
Stachowski, T. R. & Fischer, M. FLEXR: automated multi-conformer model building using electron-density map sampling. Acta Crystallogr. D Struct. Biol. 79, 354–367 (2023).
Article CAS PubMed PubMed Central Google Scholar
Stachowski, T. R. & Fischer, M. FLEXR GUI: a graphical user interface for multi-conformer modeling of proteins. J. Appl. Crystallogr. 57, 580–586 (2024).
Article CAS PubMed PubMed Central Google Scholar
Riley, B. T. et al. qFit 3: protein and ligand multiconformer modeling for X-ray crystallographic and single-particle cryo-EM density maps. Protein Sci. 30, 270–285 (2021).
Article CAS PubMed Google Scholar
Burnley, B. T., Afonine, P. V., Adams, P. D. & Gros, P. Modelling dynamics in protein crystal structures by ensemble refinement. eLife 1, e00311 (2012).
Article PubMed PubMed Central Google Scholar
van Zundert, G. C. P. et al. qFit-ligand reveals widespread conformational heterogeneity of drug-like molecules in X-ray electron density maps. J. Med. Chem. 61, 11183–11198 (2018).
Article PubMed PubMed Central Google Scholar
Jain, A. N. et al. XGen: real-space fitting of complex ligand conformational ensembles to X-ray electron density maps. J. Med. Chem. 63, 10509–10528 (2020).
Article CAS PubMed Google Scholar
Pearce, N. M. & Gros, P. A method for intuitively extracting macromolecular dynamics from structural disorder. Nat. Commun. 12, 5493 (2021).
Article CAS PubMed PubMed Central Google Scholar
Deane, C. D., Fischer, M. & Shelat, A. A. Bringing chemistry to medicine to redefine the undruggable. Nat. Chem. Biol. 20, 807–809 (2024).
Article CAS PubMed Google Scholar
Lyu, J. et al. Ultra-large library docking for discovering new chemotypes. Nature 566, 224–229 (2019).
Article CAS PubMed PubMed Central Google Scholar
Merski, M., Fischer, M., Balius, T. E., Eidam, O. & Shoichet, B. K. Homologous ligands accommodated by discrete conformations of a buried cavity. Proc. Natl Acad. Sci. USA 112, 5039–5044 (2015).
Article CAS PubMed PubMed Central Google Scholar
Ringe, D. & Petsko, G. A. The ‘glass transition’ in protein dynamics: what it is, why it occurs, and how to exploit it. Biophys. Chem. 105, 667–680 (2003).
Article CAS PubMed Google Scholar
Gonzalez-DeWhitt, K. R. et al. Insights into VDAC gating: room-temperature X-ray crystal structure of mVDAC-1. Biomolecules 14, 1203 (2024).
Kneller, D. W. et al. Structural plasticity of SARS-CoV-2 3CL Mpro active site cavity revealed by room temperature X-ray crystallography. Nat. Commun. 11, 3202 (2020).
Article CAS PubMed PubMed Central Google Scholar
Keedy, D. A. et al. An expanded allosteric network in PTP1B by multitemperature crystallography, fragment screening, and covalent tethering. eLife 7, e36307 (2018). This study shows that developing structures across a series of temperatures could reveal cryptic allosteric pockets. A subsequent fragment-screening campaign showed that these sites are targetable with small molecules.
Fischer, M., Coleman, R. G., Fraser, J. S. & Shoichet, B. K. Incorporation of protein flexibility and conformational energy penalties in docking screens to improve ligand discovery. Nat. Chem. 6, 575–583 (2014). This study demonstrates that protein conformations identified at RT could be translated into energy weights for molecular docking. Compounds prioritized from a docking screen were validated to stabilize their respective conformational states by subsequent crystal structures.
Article CAS PubMed PubMed Central Google Scholar
Biel, J. T., Thompson, M. C., Cunningham, C. N., Corn, J. E. & Fraser, J. S. Flexibility and design: conformational heterogeneity along the evolutionary trajectory of a redesigned ubiquitin. Structure 25, 739–749 (2017).
Article CAS PubMed PubMed Central Google Scholar
St-Jacques, A. D. et al. Computational remodeling of an enzyme conformational landscape for altered substrate selectivity. Nat. Commun. 14, 6058 (2023).
Article CAS PubMed PubMed Central Google Scholar
Lang, P. T., Holton, J. M., Fraser, J. S. & Alber, T. Protein structural ensembles are revealed by redefining X-ray electron density noise. Proc. Natl Acad. Sci. USA 111, 237–242 (2014).
Article CAS PubMed Google Scholar
Frauenfelder, H., Sligar, S. G. & Wolynes, P. G. The energy landscapes and motions of proteins. Science 254, 1598–1603 (1991).
Article CAS PubMed Google Scholar
Wankowicz, S. A., de Oliveira, S. H., Hogan, D. W., van den Bedem, H. & Fraser, J. S. Ligand binding remodels protein side-chain conformational heterogeneity. eLife 11, e74114 (2022).
Article CAS PubMed PubMed Central Google Scholar
Snell, E. H., Judge, R. A., Larson, M. & van der Woerd, M. J. Seeing the heat — preliminary studies of cryocrystallography using infrared imaging. J. Synchrotron Radiat. 9, 361–367 (2002).
Article CAS PubMed Google Scholar
van den Bedem, H., Bhabha, G., Yang, K., Wright, P. E. & Fraser, J. S. Automated identification of functional dynamic contact networks from X-ray crystallography. Nat. Methods 10, 896–902 (2013).
Article PubMed PubMed Central Google Scholar
Seffernick, J. T. & Fischer, M. An experimental proxy of water displaceability for ligand discovery. Nat. Methods 22, 1476–1485 (2025). This study predicts the probability of cryo water molecules to be present at RT, which links to their relative energies and displaceability by ligands and informs drug discovery and design.
Article CAS PubMed Google Scholar
Stachowski, T. R., Nithianantham, S., Vanarotti, M., Lopez, K. & Fischer, M. Pan-HSP90 ligand binding reveals isoform-specific differences in plasticity and water networks. Protein Sci. 32, e4629 (2023).
Article CAS PubMed PubMed Central Google Scholar
Skaist Mehlman, T. et al. Room-temperature crystallography reveals altered binding of small-molecule fragments to PTP1B. eLife 12, e84632 (2023). This study shows that temperature impacts the outcome of crystallographic fragment-screening campaigns. Structures solved at RT bound different fragments, at different sites and with different poses than in a parallel cryo screen.
Ostrem, J. M., Peters, U., Sos, M. L., Wells, J. A. & Shokat, K. M. K-RasG12C inhibitors allosterically control GTP affinity and effector interactions. Nature 503, 548–551 (2013).
Article CAS PubMed PubMed Central Google Scholar
Fischer, M. & Hubbard, R. E. Fragment-based ligand discovery. Mol. Interv. 9, 22–30 (2009).
Article CAS PubMed Google Scholar
Hopkins, A. L., Groom, C. R. & Alex, A. Ligand efficiency: a useful metric for lead selection. Drug Discov. Today 9, 430–431 (2004).
Article PubMed Google Scholar
Blundell, T. L., Jhoti, H. & Abell, C. High-throughput crystallography for lead discovery in drug design. Nat. Rev. Drug Discov. 1, 45–54 (2002).
Article CAS PubMed Google Scholar
Fearon, D. et al. Accelerating drug discovery with high-throughput crystallographic fragment screening and structural enablement. Appl. Res. 4, e70201 (2024).
Ferreira, R. S. et al. Complementarity between a docking and a high-throughput screen in discovering new cruzain inhibitors. J. Med. Chem. 53, 4891–4905 (2010).
Article CAS PubMed PubMed Central Google Scholar
Kolb, P. & Irwin, J. J. Docking screens: right for the right reasons? Curr. Top. Med. Chem. 9, 755–770 (2009).
Article CAS PubMed PubMed Central Google Scholar
Liu, F. et al. The impact of library size and scale of testing on virtual screening. Nat. Chem. Biol. 21, 1039–1045 (2025).
Balius, T. E. et al. Testing inhomogeneous solvation theory in structure-based ligand discovery. Proc. Natl Acad. Sci. USA 114, E6839–E6846 (2017).
Article CAS PubMed PubMed Central Google Scholar
Shoichet, B. K. Virtual screening of chemical libraries. Nature 432, 862–865 (2004).
Article CAS PubMed PubMed Central Google Scholar
Kamenik, A. S. et al. Energy penalties enhance flexible receptor docking in a model cavity. Proc. Natl Acad. Sci. USA 118, e2106195118 (2021).
Hart, K. M., Ho, C. M., Dutta, S., Gross, M. L. & Bowman, G. R. Modelling proteins’ hidden conformations to predict antibiotic resistance. Nat. Commun. 7, 12965 (2016).
Article CAS PubMed PubMed Central Google Scholar
Bowman, G. R. AlphaFold and protein folding: not dead yet! The frontier is conformational ensembles. Annu. Rev. Biomed. Data Sci. 7, 51–57 (2024).
Article PubMed PubMed Central Google Scholar
Karelina, M., Noh, J. J. & Dror, R. O. How accurately can one predict drug binding modes using AlphaFold models? eLife 12, RP89386 (2023).
Article CAS PubMed PubMed Central Google Scholar
Wayment-Steele, H. K. et al. Predicting multiple conformations via sequence clustering and AlphaFold2. Nature 625, 832–839 (2024).
Article CAS PubMed Google Scholar
Brown, B. P., Stein, R. A., Meiler, J. & McHaourab, H. S. Approximating projections of conformational Boltzmann distributions with AlphaFold2 predictions: opportunities and limitations. J. Chem. Theory Comput. 20, 1434–1447 (2024).
Article CAS PubMed PubMed Central Google Scholar
Janson, G., Valdes-Garcia, G., Heo, L. & Feig, M. Direct generation of protein conformational ensembles via machine learning. Nat. Commun. 14, 774 (2023).
Article CAS PubMed PubMed Central Google Scholar
Monteiro da Silva, G., Cui, J. Y., Dalgarno, D. C., Lisi, G. P. & Rubenstein, B. M. High-throughput prediction of protein conformational distributions with subsampled AlphaFold2. Nat. Commun. 15, 2464 (2024).
Article CAS PubMed PubMed Central Google Scholar
Bryant, P. & Noe, F. Structure prediction of alternative protein conformations. Nat. Commun. 15, 7328 (2024).
Article CAS PubMed PubMed Central Google Scholar
Lyu, J. et al. AlphaFold2 structures guide prospective ligand discovery. Science 384, eadn6354 (2024).
Article CAS PubMed PubMed Central Google Scholar
Kuhlbrandt, W. Biochemistry. The resolution revolution. Science 343, 1443–1444 (2014).
Article PubMed Google Scholar
Garcia-Nafria, J. & Tate, C. G. Structure determination of GPCRs: cryo-EM compared with X-ray crystallography. Biochem. Soc. Trans. 49, 2345–2355 (2021).
Article CAS PubMed PubMed Central Google Scholar
Hauser, A. S., Attwood, M. M., Rask-Andersen, M., Schioth, H. B. & Gloriam, D. E. Trends in GPCR drug discovery: new agents, targets and indications. Nat. Rev. Drug Discov. 16, 829–842 (2017).
Article CAS PubMed PubMed Central Google Scholar
Chari, A. & Stark, H. Prospects and limitations of high-resolution single-particle cryo-electron microscopy. Annu. Rev. Biophys. 52, 391–411 (2023).
Article CAS PubMed Google Scholar
Zhang, K., Pintilie, G. D., Li, S., Schmid, M. F. & Chiu, W. Resolving individual atoms of protein complex by cryo-electron microscopy. Cell Res. 30, 1136–1139 (2020).
Article PubMed Google Scholar
Nakane, T. et al. Single-particle cryo-EM at atomic resolution. Nature 587, 152–156 (2020).
Article CAS PubMed PubMed Central Google Scholar
Dubochet, J. et al. Cryo-electron microscopy of vitrified specimens. Q. Rev. Biophys. 21, 129–228 (1988).
Article CAS PubMed Google Scholar
Tivol, W. F., Briegel, A. & Jensen, G. J. An improved cryogen for plunge freezing. Microsc. Microanal. 14, 375–379 (2008).
Article CAS PubMed PubMed Central Google Scholar
Iancu, C. V. et al. Electron cryotomography sample preparation using the Vitrobot. Nat. Protoc. 1, 2813–2819 (2006).
Article CAS PubMed Google Scholar
Lorenz, U. J. Microsecond time-resolved cryo-electron microscopy. Curr. Opin. Struct. Biol. 87, 102840 (2024).
Article CAS PubMed Google Scholar
Bock, L. V. & Grubmuller, H. Effects of cryo-EM cooling on structural ensembles. Nat. Commun. 13, 1709 (2022).
Article CAS PubMed PubMed Central Google Scholar
Chen, C. Y., Chang, Y. C., Lin, B. L., Huang, C. H. & Tsai, M. D. Temperature-resolved cryo-EM uncovers structural bases of temperature-dependent enzyme functions. J. Am. Chem. Soc. 141, 19983–19987 (2019).
Article CAS PubMed Google Scholar
Hu, J. et al. Physiological temperature drives TRPM4 ligand recognition and gating. Nature 630, 509–515 (2024).
Article CAS PubMed PubMed Central Google Scholar
Download references
Acknowledgements
This work was in part supported by the American Lebanese Syrian Associated Charities and by National Institutes of Health grants R35GM142772 and R01CA290805 (to M.F.). The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. We thank C. D. Deane for scientific editing of the manuscript, C. M. Fischer for proofreading and M. Rice for designing the graphical abstract.
Author information Authors and Affiliations
Department of Chemical Biology and Therapeutics, St. Jude Children’s Research Hospital, Memphis, TN, USA
Timothy R. Stachowski & Marcus Fischer
Contributions
T.R.S. and M.F. contributed to discussions and co-wrote the manuscript.
Corresponding author
Correspondence to Marcus Fischer.
Ethics declarations Competing interests
The authors declare no competing interests.
Peer review Peer review information
Nature Chemical Biology thanks Stephanie Wankowicz and the other, anonymous reviewer(s) for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information Reporting Summary Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Reprints and permissions
About this article Cite this article
Stachowski, T.R., Fischer, M. Unfreezing structural biology for drug discovery. Nat Chem Biol (2025). https://doi.org/10.1038/s41589-025-02047-3
Download citation
Received 31 March 2025
Accepted 12 September 2025
Published 05 November 2025
Version of record 05 November 2025
DOI https://doi.org/10.1038/s41589-025-02047-3
Share this article
Anyone you share the following link with will be able to read this content:
Get shareable link
Provided by the Springer Nature SharedIt content-sharing initiative
Subjects Computational chemistry Screening Virtual screening X-ray crystallography