Structural insights into selective recognition of ATP-mimicking inhibitors by the atypical kinase HASPIN Authors: Sang Won Cheon, Yun A Yum, Donghwan Ku, Sung Chul Jang, Vikas R Aswar, Dnyandev B Jarhad, Yoonyoung Heo, Hyoun Sook Kim, Sang Kook Lee, Lak-Shin Jeong, Byung Woo Han Published: 2026 ABSTRACT AbstractAchieving selectivity among ATP-competitive kinase inhibitors remains a major challenge due to the high conservation of the ATP-binding pocket across the kinome. Although most kinase inhibitors target the ATP-binding site, ATP-mimicking compounds remain relatively uncommon due to concerns regarding selectivity. Here, we report the structural and biophysical characterization of the atypical serine/threonine kinase haploid germ cell-specific nuclear protein kinase (HASPIN) with two ATP-mimicking inhibitors, LJ-5157 and LJ-5242. Crystal structures of HASPIN in complex with LJ-5157 and with LJ-5242 were determined at resolutions of 1.74 Å and 1.88 Å, respectively, revealing ATP-like binding modes within the catalytic pocket. Structural analysis showed that the regulatory and catalytic spines of HASPIN are preorganized through extensive hydrophobic packing, particularly within the N-lobe, stabilizing the αC helix independently of nucleotide binding. Despite similar binding modes, microscale thermophoresis measurements demonstrated that LJ-5242 binds ~ 10-fold more tightly than LJ-5157. LJ-5157 and LJ-5242 exhibited selective HASPIN inhibition, with LJ-5242 showing ~ 10-fold and ~ 100-fold higher potency and selectivity in kinase inhibition and antiproliferative activity assays, respectively. These findings demonstrate that the distinctive architecture of the HASPIN ATP-binding pocket enables selective recognition of ATP-mimicking inhibitors and provides a framework for designing kinase inhibitors that retain ATP-like scaffolds while achieving selectivity.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-57282-y.Keywords: HASPIN kinase, Atypical protein kinase, ATP-mimicking inhibitors, Kinase selectivitySubject terms: Biochemistry, Biophysics, Chemical biology, Chemistry, Computational biology and bioinformatics, Drug discovery, Structural biology Introduction Haploid germ cell-specific nuclear protein kinase (HASPIN) is an atypical serine/threonine kinase that regulates multiple cellular differentiation and cell division processes1–3. HASPIN selectively phosphorylates histone H3 at Thr3 (H3T3), a modification required for accurate chromosomal alignment and segregation4. During mitosis, phosphorylated H3T3 functions as a centromeric chromatin mark that recruits the Chromosomal Passenger Complex (CPC) through the direct recognition of phosphorylated H3T3 by Survivin1,5. CPC is composed of Aurora B kinase, INCENP, Survivin, and Borealin, and coordinates chromosome congression, correction of improper kinetochore-microtubule attachments, and activation of the spindle assembly checkpoint6. By confining CPC localization to inner centromeres, HASPIN spatially regulates Aurora B activity and ensures high-fidelity chromosomal segregation5. HASPIN also participates in meiotic chromosomal segregation7. During meiotic prophase and metaphase, HASPIN phosphorylates H3T3 and promotes centromeric CPC recruitment8. Centromeric CPC localization enables the activation of Aurora B and the germ cell-enriched Aurora C kinases1,9. HASPIN thereby supports stable kinetochore-microtubule attachments and accurate chromosome alignment, particularly during meiosis I, when homologs segregate10. In addition to the physiological roles of HASPIN in mitotic and meiotic chromosome segregation, accumulating evidence suggests that HASPIN is involved in human diseases, particularly in cancer11,12. Aberrant activation of mitotic kinases promotes chromosomal instability, and dysregulated HASPIN activity is correlated with enhanced proliferative capacity in multiple cancer types. Elevated HASPIN expression has been reported in various solid tumors and hematological malignancies, in which sustained H3T3 phosphorylation and persistent CPC recruitment facilitate uncontrolled cell division13. Recently, emerging evidence has suggested that HASPIN has a broader pathological relevance. Altered histone phosphorylation patterns and inappropriate reactivation of mitotic signaling pathways have been reported in certain neurodegenerative disorders such as Alzheimer’s disease14,15. These observations suggest that aberrant HASPIN activity contributes to neuronal stress responses. Although several HASPIN inhibitors have been reported, achieving high selectivity within the conserved kinase family remains a major challenge. Previous structural studies demonstrated that HASPIN adopts a well-ordered active-like kinase conformation without the canonical activation-loop phosphorylation typically required for kinase activation16. Given the structural conservation of ATP-binding pockets across serine/threonine kinases, identifying unique conformational features or regulatory elements specific to HASPIN is critical for structure-guided drug development. Herein, we report the crystal structures of human HASPIN in complex with two potent ATP-mimicking nucleoside derivative inhibitors, LJ-5157 and LJ-5242, and elucidate their binding affinities. LJ-5157 and LJ-5242 share a common nucleoside core scaffold but differ in their phosphate-mimicking substituents, with LJ-5157 containing an azide moiety and LJ-5242 containing a sulfonamide moiety at the 5′ position (Fig. 2A). Previous studies reported that LJ-5157 and LJ-5242 exhibit both favorable solubility and potent inhibitory activity against HASPIN17. In addition, in silico analyses and comparisons with other kinases reveal a novel feature of HASPIN inhibition which may contribute to the development of selective ATP-competitive kinase inhibitors. Chemical structures of LJ-5157 and LJ-5242 and their interactions with HASPIN. (A) Chemical structures of LJ-5157 and LJ-5242 generated using Marvin (ChemAxon). (B) Interacting residues in the LJ-5157-bound HASPIN structure. Residues are shown as green sticks, and LJ-5157 is shown as lines. (C) Interacting residues in the LJ-5242-bound HASPIN structure. Residues are shown as orange sticks, and LJ-5242 is shown as lines. Results Hydrophobic packing-mediated ATP-independent assembly of the regulatory spine (R-spine) and catalytic spine (C-spine) in HASPIN HASPIN has been reported to adopt an active-like conformation irrespective of ATP binding. In typical protein kinases, activation is associated with inward positioning of the αC helix, which enables proper assembly of the regulatory spine (R-spine). By contrast, HASPIN exhibits a uniquely stabilized R-spine architecture characterized by extensive hydrophobic effects, particularly within the N-lobe. In the N-lobe, Val562 in the β4 strand and Ile601 and Leu603 in the β5 strand form hydrophobic contacts with Ile532 and Ile536 located in αC helix. HASPIN additionally contains an αA helix positioned between the β4 strand and the αC helix. Val463 within the αA helix contributes to the formation of a compact hydrophobic cluster that further reinforces αC helix positioning (Fig. 1A). Residues Ile538 and Leu542 located above the αC helix form hydrophobic contacts with Tyr688 of the DYT motif and Phe645 positioned N-terminal to the HRD motif. In the C-lobe, Asn741 located in the αF helix forms hydrogen bonds with the backbone of Phe645 and Glu646 at an approximate distance of 3.0 Å, thereby completing the R-spine architecture (Fig. 1B). The C-lobe portion of the R-spine resembles that observed in typical kinases, and the N-lobe region displays a densely packed hydrophobic environment surrounding the αC helix. The catalytic spine (C-spine) appears preorganized even in the absence of nucleotides. The hydrophobic axis begins near Leu479 at the N-terminal region of the kinase domain and extends through Val498 in the β2 strand, Ala509 and Ile510 in the β3 strand, Ile628 in the αE helix, Val655, Leu656, and Leu657 in the β6 strand, and terminates at Leu748 in the αF helix. Structural features of the regulatory spine (R-spine) and catalytic spine (C-spine) in HASPIN. The overall structure of HASPIN is shown in cyan. Regions corresponding to the R-spine and C-spine are highlighted by green and blue boxes, respectively. (A) Hydrophobic residues associated with the N-lobe β-sheet and αC helix of the R-spine. (B) Residues contributing to R-spine assembly, including those in the C-lobe αF helix, the active-site DYT motif, and residues preceding the HRD motif. (C) Architecture of the C-spine spanning the N-lobe to the C-lobe. Associated residues are shown as sticks. Structural basis of LJ-5157 and LJ-5242 binding within the ATP-binding pocket of HASPIN The binding modes of LJ-5157 and LJ-5242 to HASPIN closely resemble the ATP binding mode. Both inhibitors were designed to contain adenosyl and ribose derivatives, mimicking the adenine and ribose moieties of ATP (Fig. 2A). In the LJ-5157 complex, the adenosyl derivative forms hydrophobic contacts with Ala509 and Leu656, which are components of the C-spine. The adenosyl group also interacts with Gly608 located in the hinge region, which is comparable to the interaction observed between the adenine ring of ATP and hinge residues in typical kinases. In addition, the adenosyl derivative is positioned approximately 5 Å from Phe607, forming a potential edge-to-face aromatic interaction. The nitrile group forms van der Waals contacts with Ile686 and Phe605, whereas the azide group is positioned adjacent to Val498 and Phe495 (Figs. 2B, S1A). Hydrophilic interactions further stabilize LJ-5157 binding and a direct hydrogen bond is formed with Glu606. Asp611 participates in a water-mediated hydrogen bond network that restricts the orientation of the glucose derivative of LJ-5157. A water-bridged interaction is also observed between the glucose derivative and backbone carbonyl group of Ile490 (Fig. 2B). The overall binding mode of LJ-5242 is similar to that of LJ-5157, with conserved interactions involving Ile490, Phe495, Val498, Ala509, Phe605, Gly608, Asp611, Leu656, and Ile686. In addition, LJ-5242 forms a water-mediated interaction network involving catalytic Lys511 and ordered water molecules. A polar contact is observed between LJ-5242 and Asp687 in the DYT motif (Figs. 2C, S1B). Furthermore, a hydrogen bond is formed between the ligand and Asn654 at a distance of 2.6 Å. Notably, a distinct water cluster involving Asp611 and Gly609 is observed in the HASPIN–LJ-5242 complex but not in the HASPIN–LJ-5157 complex. The presence of this ordered water network is associated with additional water-mediated interactions with the sugar derivative, resulting in enhanced stabilization of ligand orientation within the binding pocket (Figs. 2C, S1B). Binding affinity and inhibitor-induced thermal stabilization of HASPIN The binding affinities of LJ-5157 and LJ-5242 for HASPIN were quantified by Microscale thermophoresis (MST) analyses. The catalytic activity of the purified HASPIN protein was confirmed using an ATP hydrolysis assay (Fig. S2A). LJ-5242 exhibited markedly stronger binding to HASPIN, with a dissociation constant (Kd) of 23.1 nM, whereas LJ-5157 showed weaker binding with Kd of 284.3 nM (Fig. 3A and B). The MST binding curves displayed a clear ligand concentration-dependent change in normalized fluorescence, allowing reliable fitting of the data to a one-site binding model. These results indicate a substantial difference in binding strength between LJ-5157 and LJ-5242. To assess ligand-induced stabilization of the protein, nanoDSF experiments were performed (Fig. 3C). In the absence of ligands, HASPIN exhibited melting temperature (Tm) of 47.49 °C. Upon addition of LJ-5157 at a ligand-to-protein molar ratio of 1:10, the Tm increased to 57.40 °C (ΔTm = 9.91 °C), whereas LJ-5242 produced a similar increase to 57.50 °C (ΔTm = 10.01 °C). Increasing the ligand concentration to a 1:50 molar ratio further increased the Tm values to 58.63 °C (ΔTm = 11.14 °C) for LJ-5157 and 58.42 °C (ΔTm = 10.93 °C) for LJ-5242 (Fig. 3C). These results indicate that both inhibitors strongly stabilize the protein upon binding and that the stabilization effect shows a modest dependence on ligand concentration. Notably, despite the substantial differences in binding affinity observed in the MST experiments, the final Tm values remain very similar for the two ligands, suggesting that nanoDSF primarily reflects ligand-induced stabilization of the global protein fold rather than the absolute binding affinity (Table 1). MST and nanoDSF analyses of LJ-5157 and LJ-5242 binding to HASPIN. (A) MST analysis of LJ-5157 binding to HASPIN. The regions used for analysis are indicated by blue (cold) and red (hot) boxes. (B) MST analysis of LJ-5242 binding to HASPIN. The regions used for analysis are indicated by blue (cold) and red (hot) boxes. In both datasets, raw traces are colored from red (unbound) to purple (bound). (C) nanoDSF analysis showing changes in melting temperature (Tm) upon binding of LJ-5157 and LJ-5242 to HASPIN. Tm values are indicated by arrows, and shifts in Tm are shown across inhibitor concentrations. Table 1Crystallographic data collection and refinement statistics.HASPIN LJ-5157 complexHASPIN LJ-5242 complexData collectionBeam lineBL-1 ABL-1 AWavelength (Å)1.00001.0000Space groupP212121P212121a, b, c (Å)75.86, 78.56, 82.2275.78, 78.66, 82.57⍺, β, γ (°)90.00, 90.00. 90.0090.00, 90.00. 90.00Resolution (Å)a45.46–1.73 (1.77–1.73)45.53–1.88 (1.95–1.88)Unique reflections51,929 (3633)40,809 (2854)Multiplicity13.7 (12.1)13.7 (14.3)Completeness99.99 (100.00)99.98 (100.00)I/σ(I)17.5 (2.2)16.5 (2.2)Rmeas (%)b12.0 (114.6)13.5 (139.4)Rpim (%)c3.2 (32.8)3.6 (36.5)CC1/299.9 (79.3)99.9 (77.3)RefinementResolution (Å)a45.46–1.73 (1.79–1.73)45.53–1.88 (1.95–1.88)Reflections used in refinement51,929 (3633)40,809 (2854)Reflections used for Rfree1989 (142)1987 (132)Rwork/Rfree (%)d17.32/20.4217.83/20.71 Bond lengths (Å)0.0060.004 Bond angles (°)0.970.69Number of non-hydrogen atoms32963197 Protein27362736 Ligand2325 Water537436Average B-factor (Å2)23.0929.92 Protein21.8028.89 Ligand21.6731.41 Water29.7235.30Ramachandran plot (%) Favored99.4199.12 Allowed0.590.88 Outliers0.000.00Poor rotamers (%)0.000.32Clashscore4.331.08PDB codes24UH24WCaValues in parentheses refer to the highest resolution shell. bc where Ih,i is the intensity of the i-th observation of reflection h, and Nh is the multiplicity of reflection h. d . Where the free set comprises ~ 5% of the reflections excluded from refinement. dValidation statistics were calculated using MolProbity. Crystallographic data collection and refinement statistics. aValues in parentheses refer to the highest resolution shell. bc where Ih,i is the intensity of the i-th observation of reflection h, and Nh is the multiplicity of reflection h. d . Where the free set comprises ~ 5% of the reflections excluded from refinement. dValidation statistics were calculated using MolProbity. Molecular dynamics simulations reveal differential interaction patterns of LJ-5157 and LJ-5242 To further investigate the structural basis of the differences in binding affinity of LJ-5157 and LJ-5242, molecular dynamics (MD) simulations were performed. During the 300 ns simulation, the overall Cα RMSF profiles of the LJ-5157–bound and LJ-5242–bound HASPIN structures were broadly similar (Fig. 4A). However, a notable difference was observed in the Tyr569–Phe593 region, where the average RMSF values remained within approximately 2 Å for LJ-5157–bound HASPIN but exceeded 3 Å for LJ-5242–bound HASPIN, indicating increased localized flexibility. A difference was also observed at Gly608 in the hinge region. The Cα RMSF value of Gly608 in the LJ-5157–bound HASPIN remained relatively stable, whereas that in the LJ-5242–bound HASPIN showed increased fluctuations (Fig. 4A). The two inhibitors differ in the functional groups attached to the adenosyl derivative: LJ-5157 contains a nitrile group, whereas LJ-5242 contains an alkyne group. Likely owing to these functional group differences, the distance between LJ-5157 and Phe605, a residue that forms a T-shaped π–π interaction with the adenosine moiety of ATP in the binding pocket, gradually increased during the MD simulation. By contrast, LJ-5242 maintained a more stable distance of approximately 5 Å from Phe605 (Fig. 4B). A similar trend was also observed for interactions with Phe607 located in the hinge region. LJ-5157 exhibited greater distance fluctuations, whereas LJ-5242 maintained a relatively stable distance of approximately 5 Å (Fig. 4C). Differences were also observed in the 5′ substituent region where LJ-5157 contains an azide group and LJ-5242 contains a sulfonamide group. During the MD simulation, LJ-5157 remained at a distance of approximately 10 Å from Asp687 in DYT motif and did not form a stable interaction. By contrast, LJ-5242 gradually stabilized near Asp687 and remained within approximately 4 Å, consistent with the formation of favorable polar interactions (Fig. 4D). Analysis of hydrogen bond formation between the protein and ligand revealed that LJ-5157 maintained approximately two hydrogen bonds throughout the simulation, whereas LJ-5242 consistently formed more than three hydrogen bonds during the trajectory (Fig. S3A). Furthermore, MM/PBSA analysis performed over the 300 ns simulation demonstrated that LJ-5242 exhibited more favorable ΔTotal energy values than LJ-5157, indicating stronger binding affinity toward HASPIN (Fig. S3B). MD simulation analysis of RMSF and ligand–residue distances in HASPIN complexes. LJ-5157-bound HASPIN is shown in blue, and LJ-5242-bound HASPIN is shown in orange. (A) Root-mean-square fluctuation (RMSF) values of Cα atoms. The mean values are represented by lines, and the SD values are shown as bars in the corresponding colors. (B) Distance changes between Phe607 and the adenosyl moiety of LJ-5157 or LJ-5242 over 300 ns. (C) Distance changes between Phe605 and the adenosyl moiety of LJ-5157 or LJ-5242 over 300 ns. (D) Distance changes between Asp687 and the azide group of LJ-5157 or the sulfonamide group of LJ-5242. Binding pocket comparisons with HASPIN and other kinases for elucidating competitive inhibitor selectivity To explore potential determinants of inhibitor selectivity, structurally related kinases were identified and compared with the HASPIN structure. Similar kinases were selected using two complementary approaches; (1) binding-pocket similarity analysis and (2) ligand-based similarity analysis. Binding-pocket similarity analysis using KiSSim identified kinases with similar ATP-binding pocket architectures (Figs. 5A, S4). We selected FYN and PAK4 since two kinases have the highest HASPIN pocket similarity scores. In parallel, ligand-based similarity analysis using the KLIFS database identified kinases with ligands structurally similar to LJ-5242, which exhibited stronger binding affinity than LJ-5157 toward HASPIN (Table S1). Among the identified kinases, we selected TYK2 for the experimental kinase inhibition assay since TYK2 has same similarity scores with other kinases and is experimentally available. The IC50 values of LJ-5157 and LJ-5242 are 9.53 nM and 1.1 nM for HASPIN, respectively. The IC50 values of LJ-5157 and LJ-5242 for FYN, PAK4, and TYK2 are more than 1000-fold higher than those for HASPIN (Figs. 5B, S6). Structural selectivity profiling, kinase inhibition potency, structural comparisons with docked structures and elucidated HASPIN in complex with LJ-5242, and antiproliferative activities of LJ-5157 and LJ-5242. (A) Top 20 kinases ranked by binding pocket similarity to HASPIN based on KiSSim fingerprint analysis of the ATP-binding site. Proteins are ordered from highest (top) to lowest similarity (B) IC50 values of LJ-5157, LJ-5242, and staurosporine against FYN, TYK2, HASPIN, and PAK4. (C) Docking comparison of LJ-5157 and LJ-5242 in PAK4, which shows the highest similarity to HASPIN in pocket structure. (D) Comparison of interacting residues between TYK2 and HASPIN in complex with LJ-5242. (E) The antiproliferative activities of LJ-5157 and LJ-5242 against a panel of cancer cell lines. Analysis of LJ-5157 and LJ-5242 binding to PAK4 revealed that the azide and sulfonamide substituents interacted favorably with charged residues within the pocket. However, the residues surrounding the adenosyl derivative in PAK4 did not support π–π interactions comparable to those observed in HASPIN (Fig. 5C). To further elucidate structural basis of the IC50 results, we implemented docking simulation of LJ-5242 to TYK2, structure using AutodockVina. Structural comparison of LJ-5242 bound to TYK2, which showed high ligand similarity to LJ-5242, and LJ-5242 bound to HASPIN exhibited a similar number of hydrophobic residues surrounding the adenosyl derivative. However, differences were observed in aromatic residues capable of forming π–π interactions with the adenosyl derivative. By contrast, the sulfonamide group of LJ-5242 formed hydrogen bonds in both kinases (Fig. 5D). In HASPIN, Asp687 interacted with the sulfonamide group, whereas in TYK2, Ser758 appeared more likely to form this interaction. In addition, LJ-5242 in TYK2 was predicted to have a lower possibility of interacting with the DFG motif compared with HASPIN. Antiproliferative assays demonstrated that both LJ-5157 and LJ-5242 suppressed cellular proliferation in two cancer cell strains, human lung cancer cell line A549 and human colon cancer cell line HCT116. Notably, LJ-5242 exhibited markedly stronger antiproliferative activity than LJ-5157 against both cancer cell lines (Fig. 5E; Table 2). Table 2IC50 values of LJ-5157 and LJ-5242 against human cancer cell lines.IC50 (µM)A549HCT116LJ-51571.281.02LJ-52420.0070.012Etoposide0.120.58 IC50 values of LJ-5157 and LJ-5242 against human cancer cell lines. Discussion In this study, we determined crystal structures of HASPIN in complex with LJ-5157 and with LJ-5242 at 1.73 Å and 1.88 Å resolution, respectively, providing structural insights into the molecular basis of inhibitor recognition by HASPIN. Although the overall kinase conformation is highly similar to previously reported HASPIN structures, these complexes provide new insights into the specific recognition of ATP-mimicking inhibitors within the HASPIN ATP-binding pocket. Most protein kinases undergo substantial conformational rearrangements between their inactive and active states18. By contrast, HASPIN has been reported to adopt an activated conformation regardless of ligand binding2,16, although the structural basis for this behavior remains unclear. Structural analyses have revealed that the regulatory spine (R-spine) and catalytic spine (C-spine) of HASPIN are stabilized by additional hydrophobic packing interactions, which are uncommon in many canonical kinases. These interactions promote stable assembly of both R- and C-spines of HASPIN independent of activation. Notably, stable formation of the R-spine and C-spine constrains the position of the αC helix, thereby maintaining an active-like kinase conformation. This structural arrangement suggests that HASPIN predominantly exists in a constitutively active state rather than in dynamically transient states between inactive and active conformations. This conformational stability of HASPIN may provide a structural advantage for selective inhibitor binding. Both LJ-5157 and LJ-5242 bind to the ATP-binding pocket of HASPIN in an ATP-like orientation. The adenosyl derivative of the inhibitors form π–π interactions with Phe605 and Phe607 of HASPIN, which appear to guide ligand positioning within the binding pocket. LJ-5157 and LJ-5242 differ in the substituent attached to the adenosyl derivative; LJ-5157 contains a nitrile group, whereas LJ-5242 carries an alkyne group. The alkyne substituent of LJ-5242 appears to support more favorable aromatic interactions within the pocket. In addition, LJ-5242 forms a more extensive water-mediated hydrogen-bond network and exhibits stronger interactions with the conserved DYT motif. These structural differences suggest enhanced molecular recognition of LJ-5242 within the ATP-binding pocket of HASPIN. Biophysical analysis results were consistent with these structural observations. LJ-5157 exhibited a higher Kd of 284.3 nM, whereas LJ-5242 showed substantially stronger binding with Kd of 23.1 nM. Intriguingly, despite the approximately 10-fold difference in binding affinity observed in the MST analyses, both inhibitors produced nearly identical thermal shifts in the nanoDSF experiments. This discrepancy likely reflects distinct physical properties measured by the two techniques. MST directly quantifies equilibrium binding affinity based on changes in thermophoretic behavior upon complex formation, whereas nanoDSF primarily reflects ligand-induced stabilization of the global folded state, which depends on changes in protein stability rather than binding affinity alone. Both inhibitors also exhibited sufficient solubility without visible precipitation or aggregation during the measurements, suggesting that solubility effects were unlikely to significantly influence the results. The comparable ΔTm values may therefore reflect a near-saturating stabilization effect under the nanoDSF conditions used in this study, where increased equilibrium affinity does not necessarily result in proportionally greater thermal stabilization. In addition, distinct enthalpic and entropic contributions to inhibitor binding may further contribute to the observed discrepancy. MD simulations elucidated the dynamic behavior of the HASPIN-inhibitor complexes. Analysis of Cα RMSF values revealed slightly increased flexibility around the hinge region and parts of the N-lobe in the LJ-5242–bound HASPIN complex. This flexibility may reflect local binding site rearrangements associated with stronger ligand interactions. Consistent with the MD-derived distance analysis, LJ-5242 forms more stable π–π interactions with Phe605 and Phe607 than LJ-5157. In addition, the sulfonamide group of LJ-5242 maintains a persistent interaction with Asp687, whereas the azide group of LJ-5157 does not form a comparable contact. These observations provide a structural explanation for the difference in binding affinity between the two inhibitors. Selectivity remains a major challenge in the development of ATP-competitive kinase inhibitors because of the high structural conservation of kinase ATP-binding pockets. To further investigate potential selectivity, kinases with similar ATP-binding pocket architectures were identified and docking simulations were performed. These analyses suggest that the distinct structural features of the HASPIN ATP-binding pocket may contribute to differential recognition of LJ-5242 relative to the other kinases analyzed. In particular, the pocket architecture of HASPIN appears to accommodate the inhibitor in a manner that enables simultaneous π–π interactions with two phenylalanine residues, Phe605 and Phe607—a feature not commonly observed in other kinases. This dual aromatic interaction likely contributes to stable ligand binding and may partially explain the observed selectivity toward HASPIN. IC50 profiling against kinases predicted to possess ATP-binding pockets similar to that of HASPIN confirmed the strong binding affinity and high selectivity of LJ-5157 and LJ-5242 toward HASPIN. Notably, LJ-5242 also demonstrated particularly high antiproliferative potential. Although many ATP-competitive kinase inhibitors have been developed, the development of compounds that closely recapitulate multiple structural features of ATP has remained limited. ATP-like scaffolds often exhibit pharmacological limitations, including high polarity, increased aqueous solubility, and limited membrane permeability, and may increase the risk of off-target kinase inhibition. One of the limitations of the present study is that experimental selectivity profiling was performed only for a limited number of representative off-target kinases and not across a broader kinase panel. Although LJ-5157 and LJ-5242 exhibited selective inhibitory activity toward HASPIN compared with the tested kinases in this study. Additional large-scale kinase profiling would be beneficial for the development of better ATP-mimicking inhibitors. Collectively, our findings provide structural insights into how ATP-mimicking inhibitors are exploited within the atypical HASPIN ATP-binding pocket and establish a framework for the rational development of next-generation HASPIN inhibitors. Materials and methods Cloning, expression, and purification The construct of the kinase domain (Ile452–Lys798) of human HASPIN was amplified by PCR from full-length HASPIN and inserted into the T7 expression vector pET21a containing His6 tag at the end of the construct. The cloned vector was validated through sequencing and subsequently transformed into the BL21(DE3) E. coli strain competent cells. The transformed cells were incubated at 37 °C in Luria-Broth medium until the optimal density at 600 nm (OD600) reached 0.6, followed by induction with 1 mM IPTG (1-thio-β-D-galactopyranoside) for 18–21 h at 20 °C. The induced cells were harvested by centrifugation at 6000 × g for 10 min at 4 °C and sonicated after adding buffer A (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 20 mM imidazole, and 10% (v/v) glycerol). The lysate was centrifuged at 30,000 × g for 1 h at 4 °C to eliminate cellular debris. The supernatant was filtered through a 0.45 μm syringe filter, and the filtrate was applied to a Ni²⁺ charged HiTrap chelating HP column (GE Healthcare, Chicago, IL, USA). The protein-loaded column was equilibrated with buffer A, and the bound HASPIN was eluted using buffer B (20 mM Tris-HCl pH 7.5, 500 mM NaCl, 1 M imidazole, and 10% (v/v) glycerol) with a linear gradient. The eluted protein was further purified by size exclusion chromatography using HiLoad 16/600 Superdex 75 pg (GE Healthcare, Chicago, IL, USA), pre-equilibrated with a storage buffer consisting of 20 mM Tris-HCl pH 7.5, 250 mM NaCl, and 1 mM TCEP. Each step of purification was validated by SDS-PAGE analysis of the fractions. Crystallization and x-ray diffraction data collection Purified HASPIN was incubated with LJ-5157 or LJ-5242 at a molar ratio of 1:3 (protein: ligand) for 1 h at 4 °C prior to crystallization. Crystallization conditions for human HASPIN preincubated with each inhibitor were screened using commercial crystallization screening kits (Hampton Research, Aliso Viejo, CA, USA). Initial crystals were obtained under conditions containing 0.2 M ammonium sulfate, 0.1 M sodium acetate pH 4.6, and 25% (v/v) polyethylene glycol (PEG) 4000 using the vapor diffusion method. Protein samples were concentrated to 6–7 mg/mL for crystallization, and the initial conditions were further optimized to obtain larger single crystals. All crystals were obtained using the sitting-drop vapor diffusion method and were grown at 22 °C. Prior to data collection, crystals were cryoprotected by soaking in reservoir solution supplemented with 20% (v/v) glycerol and subsequently flash-cooled in liquid nitrogen. X-ray diffraction data for HASPIN in complex with LJ-5157 and with LJ-5242 were collected at the BL-17 A beamline (Photon Factory, Tsukuba, Japan). Diffraction data were indexed, integrated, and scaled using the XDS package19. Structure determination with phasing and refinement For both LJ-5157 and LJ-5242 binding HASPIN, 5% of the total data were randomly excluded to serve as the Rfree set, and the remaining data were employed to calculate Rwork. The phasing problems for both inhibitor-bound HASPIN were solved by molecular replacement by Phaser in Phenix using primary HASPIN structure that is resolved with another inhibitor in our lab20. Then the primary model was refined manually by fitting it into the electron density map using COOT (v.1.1.19). The manually refined models were then further refined with Phenix until the result structures satisfy validation criteria from MolProbity21. After fitting the protein model into electron density map, ligands generated from electron Ligand Builder and Optimization Workbench (eLBOW) were fitted into the omit map. Then the final models were generated by refinement using phenix.refine tool. The statistics of the refinement are summarized in Table 1. ATP hydrolysis assay ATP hydrolysis by purified HASPIN was evaluated using the ADP-Glo kinase assay system (Promega, Madison, WI, USA). The H3 substrate used for the kinase assay was generated by cloning the human histone H3 (residues 1–21) sequence upstream of the 6×His-SUMO tag in the pRSF expression vector, resulting in an H3(1–21)-6×His-SUMO fusion construct (Fig. S2B). The recombinant fusion protein was expressed in BL21(DE3) E. coli strain and purified using Ni-NTA affinity chromatography. As a negative control, the 6×His-SUMO protein lacking the H3 peptide sequence was separately expressed and purified under the same conditions. Reactions were carried out in a final volume of 5 µL containing purified HASPIN, ATP, 20 mM HEPES pH 7.5, 150–250 mM NaCl, and 10 mM MgCl₂. The reaction mixtures were incubated at 37 °C for 30 min and then terminated by adding 5 µL of ADP-Glo reagent. After incubation at room temperature, 10 µL of kinase detection reagent was added, and the plates were further incubated to allow conversion of ADP to ATP and development of the luminescent signal. Luminescence was measured using a multimode plate reader with a 1 s integration time. Control reactions lacking HASPIN, substrate, or ATP were included to assess background signal. The ADP-dependent luminescence signal was used as an indicator of ATP hydrolysis by purified HASPIN. Molecular dynamics simulations Molecular dynamics (MD) simulations were performed based on the refined X-ray crystal structures of HASPIN in complex with LJ-5157 and LJ-5242. All simulations were carried out using GROMACS22. The topology and forcefield parameters for the ligand were generated using ACPYPE (v.2023.10.27)23. The ligand was described using the General AMBER Force Field 2 (GAFF2)24. Partial atomic charges of the ligands were calculated using the AM1-BCC method25 implemented in the Antechamber module of AmberTools25 package and were assigned assuming pH7.426. The resulting AMBER topologies were subsequently converted to GROMACS-compatible formats for simulations. Topology files for the protein–ligand complexes were generated using the AMBER99SB-ILDN force field27. Each complex was solvated using the TIP3P water model in a periodic simulation box. Sodium and chloride ions were added to neutralize the system and to achieve a final salt concentration of 150 mM NaCl. Energy minimization was first performed to remove steric clashes prior to equilibration. Pressure coupling was applied using the canonical velocity-rescaling (C-rescale) barostat to maintain the system at 1 atm. Production MD simulations were carried out for 300 ns in the NPT ensemble at 298 K with a time step of 2 fs. Temperature was maintained using the V-rescale thermostat. A cutoff distance of 1.2 nm was applied for both van der Waals and short-range electrostatic interactions, and long-range electrostatic interactions were calculated using the Particle Mesh Ewald (PME) method. Three independent simulations were performed for each complex to ensure reproducibility, and the resulting trajectories were analyzed separately. After completion of the simulations, trajectories were processed by centering the protein within the simulation box to correct for periodic boundary conditions. Structural fluctuations were analyzed by calculating Cα root mean square fluctuation (RMSF) values across the entire 300 ns simulation trajectory. Distances between ligand atoms and protein side chains were measured using the gmx distance module implemented in GROMACS, and distance analyses were also performed over the entire 300 ns trajectory. MST assay The highly purified protein used for crystallization was subjected to buffer exchange prior to the assay. All cysteine residues were maintained in a reduced state by tris(2-carboxyethyl)phosphine (TCEP) present in the storage buffer. The protein was labeled with Alexa Fluor 647 C2 maleimide (catalog# A20347, ThermoFisher Scientific, Waltham, MA, USA) at a molar ratio of 1:5 (protein: dye) and incubated for 30 min at 4 °C under light-protected conditions. Excess free dye was removed by an additional buffer exchange step. The assay buffer consisted of 20 mM HEPES (pH 7.5), 250 mM NaCl, 1% (v/v) DMSO, and 0.02% (v/v) Tween-20. HASPIN protein and the compounds LJ-5157 and LJ-5242 were diluted using the same buffer, and concentrations were adjusted to minimize signal noise. Microscale thermophoresis (MST) measurements were performed using a Monolith NT.115 pico instrument (NanoTemper Technologies, Munich, Germany). Standard capillaries (Monolith NT.115 standard capillaries, catalog# MO-K022, NanoTemper Technologies) were used for all measurements. Measurements were performed using a laser power of 50% and MST power set to medium. Data were collected using a 16-point serial dilution series, and each experiment was performed in triplicate. nanoDSF assay Protein thermal stability was assessed using Differential Scanning Fluorimetry (DSF), and measurement was performed on the QuantStudio™ real-time PCR (ThermoFisher Scientific). The concentration of HASPIN samples was optimized to 10 µM in the reaction buffer (20 mM HEPES pH 7.5, 250 mM NaCl, 1 mM TCEP). Protein Thermal Shift™ dye used for the measurements was added at 1:1000 dilution from a concentrated stock solution. The reaction mixture was transferred into a black 384-well plate, ensuring minimal air bubbles to avoid inconsistencies in fluorescence detection. To compare the thermal stability of HASPIN in complex with LJ-5157 and LJ-5242, inhibitors were mixed with the protein with 1:10 and 1:50 in molar ratio. The samples were scanned from 25 to 90 °C at a rate of 0.05 °C per second, and all measurements were repeated five times. Data were normalized for protein concentration and baseline, corrected using QuantStudioTM real-time PCR software 1.3 (ThermoFisher Scientific). The melting temperature (Tm) of the protein, which is a midpoint of the protein’s unfolding transition, was calculated from the fluorescence data by using the Boltzmann equation. Selection of kinase structures using KiSSim and KLIFS Kinase structures for structure-based analysis were selected by using the KiSSim framework, which quantifies similarity between kinase binding pockets by using interaction fingerprint (IFP)-based descriptors derived from the physicochemical and spatial properties of residues within the ATP-binding site. Two complementary approaches were employed to identify structurally relevant kinases. First, ligand-based similarity analysis was performed by selecting kinase structures co-crystallized with ligands exhibiting similar interaction patterns within the binding pocket28. To quantify ligand similarity, MACCS fingerprints were calculated, and kinase structures were selected based on co-crystallized ligands showing the highest similarity to LJ-5157 and LJ-5242. Second, a protein fingerprint–based comparison was conducted using KiSSim pocket fingerprints to evaluate the structural similarity between kinase binding sites independent of ligand information29. From each analysis, the twenty structures with the highest similarity scores relative to the reference kinase were selected and retrieved from the Protein Data Bank (PDB) for subsequent docking and structural comparison. Molecular docking Protein structures were prepared by removing non-essential heteroatoms and adding missing hydrogen atoms, assuming physiological protonation states at pH 7.4. The processed structures were converted to PDBQT format using Open Babel30, which assigns AutoDock atom types and Gasteiger partial charges required for docking calculations. Ligand structures (LJ-5157 and LJ-5242) were prepared in the same manner, including protonation state assignment at pH 7.4 and conversion to PDBQT format. Molecular docking was performed by using AutoDock Vina (v.1.2.5) with a grid box centered at (24.163, 5.895, 18.559 Å) and dimensions of 20 × 20 × 20 Å, encompassing the ATP-binding pocket31. Docking calculations were carried out with an exhaustiveness value of 8, and up to 10 binding poses were generated for each docking run. Docking results were analyzed by comparing the predicted ligand binding poses with ATP-binding modes observed in representative kinase structures, enabling structural comparison relative to the canonical ATP-binding geometry of the kinase pocket. The top-ranked binding pose based on the Vina scoring function was selected for subsequent structural analysis. Kinase IC50 determination IC50 values for LJ-5157 and LJ-5242 were determined by Carna Biosciences, Inc (Kobe, Japan), a company specializing in kinase inhibitor profiling and activity-based biochemical assays, using the Mobility Shift Assay (MSA) service. The inhibitory activities of the compounds were evaluated against FYN (isoform a), TYK2, HASPIN, and PAK4. LJ-5157 was tested over a concentration range of 10 µM to 0.0003 µM, whereas LJ-5242 was evaluated from 5 µM to 0.00015 µM. IC50 values for each kinase were calculated from concentration–response curves generated and provided by Carna Biosciences. Cell culture and chemicals Human lung cancer cell line (A549) and human colon cancer cell line (HCT116) were obtained from the American Type Culture Collection (Manassas, VA, USA). Both cells were cultured in RPMI 1640 media supplemented with 10% fetal bovine serum (Gibco, Grand Island, NY, USA) and 1% antibiotics-antimycotics (AA) (PSF; 100 units/mL penicillin G sodium, 100 µg/mL streptomycin, and 250 ng/mL amphotericin B) at 37 °C in a humidified incubator with 5% CO2. Trichloroacetic acid, sulforhodamine B (SRB), and etoposide were purchased from Sigma-Aldrich (St. Louis, MO, USA). Cell proliferation assay (SRB assay) Cell proliferation was evaluated using the sulforhodamine B (SRB) assay as previously described32. Briefly, cells were seeded in 96-well culture plates. For day-zero controls, cells were fixed 30 min post-seeding. The remaining cells were treated with various concentrations of the test compounds for 72 h. Following incubation, the cells were fixed with cold 10% (w/v) trichloroacetic acid (TCA) at 4 °C for 30 min, rinsed with distilled water, and air-dried at room temperature. The fixed cells were then stained with 0.4% (w/v) SRB in a 1% (v/v) acetic acid solution at 37 °C for 2 h. Unbound dye was removed by washing with 1% (v/v) acetic acid, and the plates were subsequently air-dried. The protein-bound dye was solubilized with 10 mM Tris base (pH 10.0) for 30 min. The absorbance was measured at 515 nm using a microplate reader. Cell proliferation (%) was calculated using the following equation: The IC50 value was measured by nonlinear regression analysis using TableCurve 2D v5.01 (Systat Software Inc., Richmond, CA, USA). Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Acknowledgements This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (grant number: RS-2023-00218543 to BWH). The authors acknowledge the use of the beamline BL-1 A at the Photon Factory (Tsukuba, Japan) under proposal number 2024G0008 for data collection. Author contributions S.W.C. conceptualized the study, performed protein purification, crystallization, and structure determination, conducted biophysical assays including MST, carried out kinome-wide analysis, performed molecular docking and molecular dynamics simulations, analyzed the data, and wrote the original manuscript. Y.A.Y., V.R.A., and D.B.J. synthesized the compounds. S.C.J. performed Cell proliferation assay and S.K.L. supervised the Cell proliferation assay. Y.H. performed nanoDSF experiments. D.K. performed molecular dynamics simulations. L.-S.J. supervised the compound synthesis. B.W.H. supervised the study and revised the manuscript. All authors reviewed and approved the final manuscript. Funding This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (grant number: RS-2023-00218543) and the Bio-Data Industry Professional Training Program through the Korea Institute for Advancement Technology (KIAT) funded by the Ministry of Trade, Industry and Energy (grant number: RS-2025-02214034). Data availability The coordinates and structure factors of HASPIN in complex with LJ-5157 and with LJ-5242 have been deposited in the Protein Data Bank (PDB) under accession codes 24UH and 24WC, respectively. All data generated or analyzed during this study are included in this published article and its supplementary information files. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The coordinates and structure factors of HASPIN in complex with LJ-5157 and with LJ-5242 have been deposited in the Protein Data Bank (PDB) under accession codes 24UH and 24WC, respectively. All data generated or analyzed during this study are included in this published article and its supplementary information files.