Ruxolitinib binding to human serum albumin: bioinformatics, biochemical and functional characterization in JAK2V617F+ cell models

Ruxolitinib is a type I JAK inhibitor approved by FDA for targeted therapy of Philadelphia-negative myeloproliferative neoplasms (MPNs), all characterized by mutations activating the JAK2/STAT signaling pathway. Treatment with ruxolitinib improves constitutional symptoms and splenomegaly. However, patients can become resistant to treatment and chronic therapy has only a mild effect on molecular/pathologic remissions. Drugs interaction with plasma proteins, i.e. human serum albumin (HSA), is an important factor affecting the intensity and duration of their pharmacological actions. Here, the ruxolitinib recognition by the fatty acid binding sites (FAs) 1, 6, 7, and 9 of HSA has been investigated from the bioinformatics, biochemical and/or biological viewpoints. Docking simulations indicate that ruxolitinib binds to multiple sites of HSA. Ruxolitinib binds to the FA1 and FA7 sites of HSA with high affinity (Kr = 3.1 μM and 4.6 μM, respectively, at pH 7.3 and 37.0 °C). Moreover, HSA selectively blocks, in a dose dependent manner, the cytotoxic activity of ruxolitinib in JAK2V617F+ cellular models for MPN, in vitro. Furthermore this event is accompanied by changes in the cell cycle, p27Kip1 and cyclin D3 levels, and JAK/STAT signaling. Given the high plasma concentration of HSA, ruxolitinib trapping may be relevant in vivo.

In clinical trials, treatment with ruxolitinib controls hematocrit in hydroxyurea-resistant or -intolerant PV patients, whereas in MF patients it produces a reduction of splenomegaly and improves symptoms and quality of life, prolonging overall survival compared to placebo and best available therapy [9][10][11][12][13] . Despite these clinical benefits, insufficient response or resistance to ruxolitinib has been reported in about 15% of patients and chronic therapy with this drug has only a mild effect on molecular and pathologic remissions 3,7 . Indeed, a number of combinations with other drugs, which can improve ruxolitinib effectiveness, overcome resistance and ameliorate toxic effects, have been tested with variable success in several clinical trials or are currently under evaluation 3,7,9 . Human serum albumin (HSA), the most abundant plasma protein (~750 μM) 14 , is decreased by circulating inflammatory cytokines, such as interleukin-6 and TNF-α, which are commonly elevated in MFs. In turn, these cytokines are markedly decreased by treatment with ruxolitinib, while HSA levels are increased 15,16 . Recently, HSA level has been suggested as an independent indicator for prognosis in MF patients uncorrelated with cytogenetic or mutation profiles 17,18 . HSA and serum cholesterol levels were used to develop a new cachexia index further enhancing prognostication in MF 19 .
Pharmacokinetic data indicate that at clinically relevant concentrations ruxolitinib binds to plasma proteins, mostly HSA (approximately 97%) (DrugBank 20 ; PubChem 21 ) that may impair the drug efficacy. However, to our knowledge, no information is available on the relevance of HSA:ruxolitinib recognition in the biological and clinical response of JAK2V617F + MPNs to this JAK2 inhibitor.

Results
Docking simulations of ruxolitinib binding to HSA. Docking simulations of ruxolitinib binding to HSA predict that the binding affinity of the drug for the FA1, FA6, FA7, and FA9 sites is similar, values ranging between −7.1 and −8.0 kcal mol −1 ( Table 1).
The preferential binding site of the drug is the FA9 cleft, located in the crevice between subdomains IA-IB-IIA on one side and subdomains IIB-IIIA-IIIB on the other side ( Fig. 1 and Table 1). However, the number of complexes observed in the FA9 site in docking simulations with a maximum of 9 poses is only 2. The highest ranking complex of ruxolitinib bound to the FA9 site of FA-free HSA is shown in Fig. 1. In this site, the HSA:ruxolitinib complex is stabilized by hydrophobic interactions with Ala194, Ala191, Val456, and Leu463, and the aliphatic portions of His146, Lys190, Arg197, Glu425, Asn429, and Lys432. Moreover, the drug recognition is also based on the formation of hydrogen bonds with residues Asp108, Gln459, and the backbone amide group of Val455.
In other 5 poses, ruxolitinib is predicted to bind FA-free HSA at the FA1 site (i.e., at the subdomain IB), with a binding affinity value of the best ruxolitinib pose of −7.8 kcal × mol −1 (Table 1). Ruxolitinib forms hydrophobic interactions with Asp108, Asn109, Pro110, Arg145, His146, Pro147, Ser193, Arg197, and Glu425 and is hydrogen bonded to the backbone carbonyl group of Lys190 in the large FA1 site (Fig. 1). This pocket represents the third main ligand (e.g., drug) binding pocket of HSA 31,32 , the heme being a prototypical ligand 29,34 .
Ruxolitinib is also predicted to bind to FA6 site in the subdomain IIIA, the binding affinity value of the best pose being −7.7 kcal × mol −1 ( Fig. 1 and Table 1). Ruxolitinib recognition by the FA6 site is based on the formation of hydrophobic interactions involving the drug and Arg209, Lys212, Ala213, Val216, Ser232, Val235, Leu327, Asp324, Ala350, and Glu354 residues. Moreover, ruxolitinib forms hydrogen bonds with the carboxyl group of Asp324 and Glu354, and with the amine group of Arg209 (Fig. 1).
Ruxolitinib binding to the FA7 site (also named Sudlow's site I), which represents one of the most relevant drug binding site 29,35,36 , is stabilized by hydrophobic interactions with Leu103 and Val462 and the aliphatic portions of Glu100, Gln104, Arg197, Gln204, and Lys205. Moreover, the drug recognition is also based on the formation of hydrogen bonds with Tyr148 and Glu465 ( Fig. 1 and Table 1). thermodynamics of ruxolitinib binding to HSA. Since chromophores that bind selectively to the FA2, FA5, FA6, FA8, and FA9 sites are not available at present; only ruxolitinib binding to the FA1, FA3-FA4, and FA7 sites of HSA has been experimentally assessed.
Data shown in Fig. 3 (panels D and E) indicate that ruxolitinib inhibits competitively heme-Fe(III) and dansyl-arginine binding to HSA. In fact, values of app K/ 0 K for heme-Fe(III) and dansyl-arginine binding to HSA increase linearly on the drug concentration (i.e., [ruxolitinib]). The linear analysis of data according to Eq. (3) discards the possibility of the linkage between the FA1 and FA7 sites for ruxolitinib binding to HSA (i.e., allosteric effects). The values of the dissociation equilibrium constant for ruxolitinib binding to HSA (i.e., K r = 3.1 ± 0.4 μM and 4.6 ± 0.5 μM; at pH 7.3 and 37.0 °C), determined according to Eq. (3), correspond to the inverse of the slope of the straight lines shown in panels D and E of Fig. 3 41 . Data shown in Fig. 3 (panels D and E) indicate that ruxolitinib binds to the FA1 and FA7 sites of HSA with similar affinity (i.e., 3.1 ± 0.4 μM and 4.6 ± 0.5 μM, respectively). As predicted from docking simulations, ruxolitinib does not inhibit dansyl-sarcosine binding to the FA3-FA4 site of HSA over the whole concentration range explored (Fig. 3, panel F).

Biological consequences of ruxolitinib binding to HSA in JAK2V617F + myeloid cell lines.
Patient-derived myeloid cell lines K562, HEL and SET-2, respectively carrying JAK2 wild type, homozygous and heterozygous JAK2V617F mutation 42 , were used to assess the ability of HSA to affect the ruxolitinib-mediated inhibition of cells viability. Cells were cultured in 10% FBS medium or serum-free medium containing HSA at concentrations ranging between 0.08 µM and 80 µM, the last one representing the concentration of α-fetoprotein, the fetal analog of serum albumin, in 10% FBS containing medium. Cells were treated with different doses of ruxolitinib for 72 hours (i.e., between 1.0 nM and 10 µM). In these culture conditions, HEL and SET-2 cells (both JAK2V617F positive i.e., JAK2V617F + ) have been found to be sensitive to ruxolitinib, as previously demonstrated 8,43,44 (Fig. 4 and Table 2). Nonetheless, HSA efficiently inhibits, in a concentration-dependent manner, the effect of ruxolitinib on cell viability (measured as IC 50 ). In JAK2V617F + HEL and SET-2 cells, the IC 50 value of ruxolitinib respectively decreases from 4.7 ± 1.0 µM and 0.15 ± 0.04 µM, when seeded in 10% FBS or 80 µM HSA, to 1.0 ± 0.3 µM and 0.033 ± 0.008 µM, in the absence of HSA ( Fig. 4 and Table 2). However, the IC 50 value in JAK2wt K562 cells is greater than 10 µM under all the experimental conditions. On the basis of these results, we chose for the subsequent experiments the ruxolitinib concentrations that inhibit JAK2V617F + cell viability of at least 30% (3 µM and 0.2 µM in HEL and SET-2 cells, respectively), while JAK2wt K562 cells were treated with 3 µM ruxolitinib.
FACS analysis of the cell cycle performed after 72 hours showed that, in the absence of ruxolitinib, the decrease of HSA concentrations in the medium slightly induces the sub-G1 phase in all the three myeloid cell lines (Fig. 5). Only in JAK2V617F + HEL and SET-2 cells cultured at low HSA concentrations (8 µM) or in serum-free medium (no HSA), 72 hours treatment with ruxolitinib reduces the percentage of cells in the G1 phase and increases the percentage of cells in S phase, as compared to untreated cells (Fig. 5). Modification of cell cycle phases by HSA in ruxolitinib-treated JAK2V617F + myeloid cells also corresponds to a change in the expression levels of both cyclin D3 and p27 Kip1 , two key regulators of the G1/S phase transition, as revealed after 24 hours by immunoblot analysis (Fig. 6). Overall these results suggest that HSA levels in growing medium selectively affect the ruxolitinib-mediated inhibition of cell viability and induce cell death in cells bearing the JAK2V617F mutation.
HSA affects also ruxolitinib-mediated effects on the JAK2/STAT5 signaling (Fig. 6). In the absence of ruxolitinib, HSA levels in culture medium influences, in a dose dependent manner, the phosphorylation status of both STAT5 and JAK2 in JAK2V617F + HEL and SET-2 cells. These phosphorylation events are undetectable in JAK2wt K562 cells treated with 3 µM ruxolitinib and their specificity was further indicated by the lack of change in total STAT5 and JAK2 protein levels. As previously demonstrated for type I JAK inhibitors [45][46][47] , 3 hours treatment with ruxolitinib (at concentrations of 3 µM and 0.2 µM for HEL and SET-2, respectively) increases the activation-loop phosphorylation of JAK2, which coincides with the inhibition of STAT5 phosphorylation in JAK2V617F + HEL and SET-2 cells, but not in JAK2wt K562 cells treated with 3 µM ruxolitinib. However, higher levels of phosphorylated JAK2 are induced by ruxolitinib in JAK2V617F + cells grown at low HSA (8 µM) concentrations or in serum-free medium than in cells grown in medium containing 10% FBS or 80 µM HSA (Fig. 6). Overall, these results suggest that the ruxolitinib targeted inhibition of JAK2/STAT5 signaling is at least partially dependent on HSA serum levels, thus supporting a role for HSA in the therapeutic efficacy of ruxolitinib in MPNs.

Discussion
Ruxolitinib is a clinically approved JAK1/JAK2 type I inhibitor specifically tailored to oncoproteins de-regulating JAK/STAT signaling. The positive, but not curative, effect of ruxolitinib in MPNs is well documented in vitro and in vivo. Overall, the toxic profile of ruxolitinib seems benign, with side effects observed in the patients, mainly of hematological type (thrombocytopenia and moderate anemia). Nevertheless, failure in the achievement of histopathological and molecular complete response or resistance to this drug represent important limitations to www.nature.com/scientificreports www.nature.com/scientificreports/ the use of ruxolitinib in MPN patients 3,7,9 . Various factors operating at the cellular or organismic level may be at the basis of these events. Ruxolitinib resistance can develop, for example, because of mutations of the kinase domain reported by in vitro studies 48,49 . However, these mutations have not been identified in patients yet, thus suggesting that ruxolitinib is a weak inhibitor 7 . Moreover, chronic exposure of MPN cells to ruxolitinib leads to disease persistence and reduced sensitivity to JAK inhibition 47 . In this context, ruxolitinib plasma concentration may represent an important concern for its therapeutic efficacy.
The interaction of drugs with plasma proteins, i.e. HSA, is an important factor affecting the duration and intensity of their pharmacological actions 14 . Here, we show that ruxolitinib is trapped by HSA impairing the HSA-based transport of physiological ligands (e.g., hemin) and the drug action. Given the in vivo concentrations the dissociation equilibrium constant for ruxolitib binding to HSA (K r ranging between 3.1 and 4.6 μM; present study), for the chemical equilibrium laws we estimated that almost all ruxolitinib is bound to HSA in vivo. Indeed, it is known that approximately 97% of ruxolitinib is bound to plasma proteins, primarily to albumin (DrugBank 20 ; PubChem 21 ). Therefore, HSA has a big impact on the drug bioavailability.
In agreement with docking studies, ruxolitinib inhibits competitively heme-Fe(III) and dansyl-arginine binding to the FA1 and FA7 sites of HSA, respectively, without affecting dansyl-sarcosine association to the FA3-FA4 cleft, this being the binding site, among other ligands, of erucic acid 50 . Moreover docking investigations indicate that ruxolitinib binds to the FA9 and FA6 sites of HSA, however, no specific probes for these pockets are available to highlight drug recognition. Ruxolitinib binding to the FA1 cleft is predicted to occur in the vicinity of Tyr161, the HSA residue coordinating the heme-Fe(III) atom; this provides the molecular basis of the competitive inhibition exerted by ruxolitinib on heme-Fe(III) binding to HSA. Analogously, similar conclusions can be drawn for the competition between ruxolitinib and dansyl-arginine at the FA7 site.
High HSA concentration (80 µM) or 10% FBS in the growing medium renders myeloid cell lines derived from patients with JAK2V617F mutation more resistant to ruxolitinib. Indeed, the cytotoxic activity of ruxolitinib is enhanced if JAK2V617F + HEL and SET-2 cells are grown in medium supplemented with low HSA concentrations (i.e. 8 and 0.08 µM HSA), and a maximal effect is reached in the absence of HSA or FBS (i.e., serum-free medium). We found that HSA affects ruxolitinib-mediated inhibition of cells viability HEL and SET-2 (both JAK2V617F + ) in a dose-dependent manner and correlated with changes in the cell cycle distribution of these cells. In marked contrast, these events are not induced by HSA and/or ruxolitinib in JAK2wt K562 cells, thus indicating the importance of the JAK2V617F mutation in cell response.
At low HSA concentrations, treatment with ruxolitinib induced cell death signals as indicated by the progression of JAK2V617F + HEL and SET-2 cells into the S phase of the cell cycle. The effect of HSA levels on ruxolitinib-mediated inhibition of cells viability is related also to changes in the JAK/STAT signaling, which is a characteristic feature of MPNs 46,51 . In JAK2V617F + HEL and SET-2 myeloid cell lines grown at low HSA concentration, the ruxolitinib-induced activation-loop phosphorylation of JAK2 is increased and the tyrosine phosphorylation of STAT5 is inhibited. STAT5 is a known substrate of JAK2, whose activation is by itself sufficient to transform hematopoietic cells.
Overall, our findings suggest that ruxolitinib interaction with plasmatic HSA may be clinically relevant. Moreover, additional pathological conditions, the concurrent administration of drugs, age, and sex can affect drug (e.g., ruxolitinib) binding to plasma proteins 14,[52][53][54][55][56][57] . In this view, modelling analysis of ruxolitinib recognition by HSA may be useful for the rational design of novel JAK2 inhibitors with higher therapeutic efficacy than the currently available drugs in MPNs. Moreover, therapeutic drug monitoring is demanding to determine the free-fraction of ruxolitinib, representing the drug fraction that has therapeutic efficacy.  Human myeloid cell lines included JAK2V617F-mutated HEL and SET-2 cells, which are models for MPN diseases 42 , and JAK2wt K562 cells. HEL and K562 cells were grown in RPMI 1640 + GlutaMax medium (Gibco-BRL, Grand Island, USA), containing 10% heat-inactivated fetal bovine serum (FBS) (Gibco-BRL), 50 μg/ml streptomycin and 50 IU/ml penicillin. SET-2 cells were maintained in RPMI 1640 + GlutaMax medium (Gibco-BRL) supplemented with 20% heat-inactivated FBS (Gibco-BRL), 50 μg/ml streptomycin and 50 IU/ml penicillin. All cell lines were grown in a fully humidified incubator with 5% CO 2 in air. The HSA concentration refers to that of the free protein. The analysis of data according to Eq. (1) allowed the determination of the following parameters: 0 K h = 0.076 ± 0.008 μM (circles), app K h = 0.25 ± 0.03 μM (squares), and app K h = 0.82 ± 0.09 μM (triangles). (B) Binding isotherms for dansyl-arginine association to HSA in the absence (circles) and presence of 10 μM (squares) and 40 μM ruxolitinib (triangles) (panel A). The HSA concentration was 2.7 μM. The dansyl-arginine concentration refers to that of the free chromophore. The analysis of data according to Eq. (1) allowed the determination of the following parameters: 0 K da = 27 ± 3 μM (circles), app K da = 78 ± 8) μM (squares), and app K da = 27 ± 3 μM (triangles). (C) Binding isotherms for dansylsarcosine association to HSA in the absence (circles) and presence of 10 μM (squares) and 100 μM ruxolitinib (triangles) (panel A). The HSA concentration was 2.7 μM. The dansyl-sarcosine concentration refers to that of the free chromophore. For clarity, the binding isotherms for dansyl-sarcosine association to HSA in the presence of 10 μM and 100 μM ruxolitinib (diamonds and triangles, respectively) have been arbitrarily up shifted of 0.3 and 0.6 units. The analysis of data according to Eq. (2) allowed the determination of the following values parameters: 0 K ds = 5.1 ± 0.5 μM (circles), app K ds = 5.6 ± 0.6 μM (diamonds), and app K ds = 4.9 ± 0.5 μM (triangles). (D) Dependence of the app K h / 0 K h ratio for heme binding to HSA on the ruxolitinib concentration. The analysis of data according to Eq. (3) allowed the determination of the value of K r = 3.1 ± 0.4 μM. The circle on the ordinate indicates the value of app K h / 0 K h = 1 obtained in the absence of ruxolitinib. (E) Dependence of the app K da / 0 K da ratio for dansyl-arginine binding to HSA on the ruxolitinib concentration. The analysis of data according to Eq. (3) allowed the determination of the value of K r = 4.6 ± 0.5 μM. The circle on the ordinate indicates the value of app K da / 0 K da = 1 obtained in the absence of ruxolitinib. (F) Dependence of K ds for dansylsarcosine on the ruxolitinib concentration. Values of app K ds are unaffected by ruxolitinib; in fact, the average value of app K ds 5.1 μM corresponds to that of 0 K ds . K h , K da , and K ds indicate the dissociation equilibrium constant for heme-Fe(III), dansyl-arginine and dansyl-sarcosine binding to HSA, respectively, in the absence and presence of ruxolitinib (i.e., 0 K h , 0 K da , and 0 K da ; and app K h , app K ds and app K ds , respectively). Where not shown, the standard deviation is smaller than the symbol. For details, see text. www.nature.com/scientificreports www.nature.com/scientificreports/

Methods
Docking simulations of ruxolitinib binding to HSA. The approach used to predict the molecular details of ruxolitinib binding to HSA is similar to that already used by us and other authors for other ligands [24][25][26]58,59 . In detail, docking simulations of ruxolitinib binding to HSA were performed using the crystal structure of ligandfree HSA (PDB ID: 1AO6) 60 . The three-dimensional structure of ruxolitinib was obtained from the crystal structure of C-Src in complex with ruxolitinib (PDB ID: 4U5J) 61 . Simulations were carried out using DockingApp 37 , a user friendly interface for the docking program AutoDock Vina 62 which facilitates the setup, run and analysis of docking simulations. In order to carry out "blind" predictions of the ruxolitinib binding sites, in all the simulations the search space (docking grid) included the whole HSA structure. The grid spacing was set to 1 Å per grid unit and the exhaustiveness parameter was increased from the default value of 8 to 24, as suggested by AutoDock Vina developers for grid sizes larger than 27,000 Å 3 62 , which is the case for HSA simulations. The simulations were carried out both by keeping all protein residues rigid and by allowing flexibility of only the residues building up the walls of the FA sites (FA1 to FA9). Residues for which flexibility was allowed are reported in Table S1 of Supplementary Materials. Rotatable bonds of ruxolitinib were kept flexible in the simulations. thermodynamics of ruxolitinib binding to HSA. Thermodynamics of ruxolitinib recognition by the FA1, FA3-FA4 and FA7 sites of HSA was followed by competitive inhibition of heme-Fe(III), dansyl-sarcosine, and dansyl-arginine (i.e., L) binding, respectively, according to Fig. 1.
Values of the dissociation equilibrium constants for dansyl-arginine and dansyl-sarcosine binding to HSA (i.e., K da and K ds , respectively), in the absence and presence of ruxolitinib, (i.e., 0 K da and app K da , and 0 K ds and app K ds respectively; see Fig. 2) were obtained from the dependence of the relative fluorescence intensity change (i.e., ΔF/ ΔF max ) of the HSA:dansyl-arginine and HSA:dansyl-sarcosine complexes on the dansylated compound concentration in the absence (i.e., 0 K) and presence (i.e., app K) of ruxolitinib, according to Eq. (2) 24-28,38-40 : where ΔF is the fluorescence intensity change observed at each concentration of either dansyl-arginine or dansyl-sarcosine, ΔF max is the maximum fluorescence intensity change, K is either 0 K da or app K da or 0 K ds or app K ds , and n is the Hill coefficient. Since ΔF values were normalized dividing them by ΔF max , no correction of the inner filter effect was required. In fact, according to literature 63,64 ΔF/ΔF max is equal to ΔF corrected /ΔF max corrected .
According to the competitive inhibition mechanism shown in Fig. 2, values of K r were obtained from the linear dependence of app K/ 0 K values on the ruxolitinib concentration according to Eq. (3) 41 : app 0 r Spectrophotometric and spectrofluorimetric measurements were carried out with a Jasco V-560 spectrophotometer and a Jasco FP-6500 spectrofluorometer (Jasco International Co., Ltd., Tokyo, Japan), respectively. cell viability. The viability of cells was determined using the ATP-based cell-viability assay CellTiter-Glo (Promega, Madison, WI, USA) according to the manufacturer's instruction. Briefly, cells were washed three times with PBS before seeding, resuspended in RPMI 1640 + 10% FBS or in RPMI-1640 serum-free medium containing HSA at concentrations ranging from 0 to 80 μM. A total of 40,000 cells/well were seeded in 96 well plates and treated with ruxolitinib at different concentrations (ranging from 0 to 10 μM). After 72 hours, the CellTiter-Glo solution (100 µl/well) was added to the cells for 20 minutes at RT and the cell viability was measured with a Tecan Spark 20 M multimode microplate reader (Tecan, Männerdorf, Switzerland), in luminescence mode (integration time 1 sec/well).
IC 50 values, i.e. the concentration of ruxolitinib that gives half-maximal response on cell viability, have been obtained from the dependence of the relative cell viability (i.e., α) on the ruxolitinib concentration (i.e., [ruxolitinib]), at different culture conditions, according to Eq. (4): where α is the relative cell viability and n is the Hill coefficient. cell cycle and cell death analysis. Cells were fixed in 70% ethanol in phosphate-buffered saline, incubated with 50 μg/mL propidium iodide (Sigma-Aldrich) and 50 units/mL DNase-free RNase A (Sigma-Aldrich) and analyzed using a Epics XL Cytometer (Beckman Coulter). A minimum of 10,000 total events were acquired.

Data availability
Data are available upon request to the authors.