Structure and mechanisms of NT5C2 mutations driving thiopurine resistance in relapsed lymphoblastic leukemia

Summary

Activating mutations in the cytosolic 5’-nucleotidase II gene NT5C2 drive resistance to 6-mercaptopurine in acute lymphoblastic leukemia. Here we demonstrate that constitutively active NT5C2 mutations K359Q and L375F reconfigure the catalytic center for substrate access and catalysis in the absence of allosteric activator. In contrast, most relapse-associated mutations, which involve the arm segment and residues along the surface of the inter-monomeric cavity, disrupt a built-in switch-off mechanism responsible for turning off NT5C2. In addition, we show that the C-terminal acidic tail lost in the Q523X mutation functions to restrain NT5C2 activation. These results uncover dynamic mechanisms of enzyme regulation targeted by chemotherapy resistance-driving NT5C2 mutations, with important implications for the development of NT5C2 inhibitor therapies.

Graphical Abstract

Activating NT5C2 mutations in acute lymphoblastic leukemia drive resistance to 6-mercaptopurine. Dieck et al. demonstrate that there are three classes of mutations: the most common ones disrupt the switch-off mechanism, one class locks NT5C2 in the constitutive active state, and one lacks the C-terminal brake.

Introduction

NT5C2 is a highly conserved and ubiquitously expressed cytosolic nucleotidase involved in the maintenance of intracellular nucleotide pool homeostasis by promoting the clearance of excess purine nucleotides from cells (Allegrini et al., 1997; Banditelli et al., 1996; Hunsucker et al., 2005; Itoh et al., 1967; Pesi et al., 1994; Spychala et al., 1988; Tozzi et al., 1991). NT5C2 preferentially dephosphorylates the 6-hydroxypurine monophosphates inosine monophosphate (IMP), guanosine monophosphate (GMP) and xanthosine monophosphate (XMP), as well as the deoxyribose forms of IMP and GMP (dIMP and dGMP), facilitating the export of the resulting purine nucleosides out of the cell (Hunsucker et al., 2005). The importance of this activity for cell homeostasis is highlighted by the association of loss-of-function mutations in NT5C2 with hereditary spastic paraplegia in humans (Novarino et al., 2014). Yet, perhaps the most prominent role of NT5C2 in human disease results from the capacity of this enzyme to dephosphorylate and inactivate cytotoxic thiopurine monophosphate nucleotides –6-thioinosine monophosphate (6-tIMP), 6-thioxanthosine monophosphate (6-tXMP) and 6-thioguanosine monophosphate (6-tGMP)– generated by the incorporation of 6-thioguanine (6-TG) and 6-mercaptopurine (6-MP), two thiopurine nucleoside analogs used in the treatment of lymphoblastic leukemia, into the salvage pathway of purine biosynthesis (Brouwer et al., 2005). In this context, somatic gain-of-function mutations in NT5C2 encoding proteins with increased nucleotidase activity drive selective resistance to chemotherapy with 6-MP and 6-TG and are recurrently present in acute lymphoblastic leukemia (ALL) samples at relapse, with a prevalence as high as 35–45% in early relapse ALL cases progressing under 6-MP chemotherapy (Ma et al., 2015; Meyer et al., 2013; Tzoneva et al., 2018; Tzoneva et al., 2013).

Structurally, NT5C2 is a member of the haloacid dehalogenase (HAD) superfamily of Mg²⁺-dependent intracellular 5’-nucleotidases (Koonin and Tatusov, 1994) and functions as a homotetramer made up of a dimer of two tightly-associated dimers (Wallden et al., 2007). Mechanistically, the 5’-nucleotidase activity of NT5C2 proceeds via formation of a phosphoenzyme intermediate mediated by three conserved HAD motifs in the catalytic center (Allegrini et al., 2001; Collet et al., 1998). Motif I, DXDX(V/T)(L/V), is most integral to the reaction mechanism, in which the first Asp (Asp52 in NT5C2) makes a nucleophilic attack on the phosphate of the substrate and the second Asp (Asp54 in NT5C2) donates a proton to the departing nucleoside. Motif II is a (S/T) residue (Thr249 in NT5C2) which stabilizes the phosphoenzyme intermediate, and motif III, K(X_x)D(X_0-4)D, contains a highly conserved Lys (Lys292 in NT5C2), which also contributes to the stabilization of the intermediate by countering some of the negative charges. In addition, motif III contains Asp351 and Asp356 in NT5C2 which interact with an essential Mg²⁺ ion involved in the catalysis. Moreover, Asp54 in motif I also helps to coordinate the Mg²⁺ ion, while Met53 and Thr56 (motif I) are important for the correct orientation of all Asp residues. These elements define the structural components required for the catalytic activity of NT5C2.

NT5C2 functions as a dimer of dimers, with the dimer being the smallest functional unit with enzymatic activity (Spychala et al., 1988; Wallden and Nordlund, 2011) and is regulated by allosteric activators with ATP, dATP and diadenosine tetraphosphate (Ap₄A) as most the likely activators at physiological concentrations (Marques et al., 1998; Pinto et al., 1986; Spychala et al., 1988). The effector binding site of NT5C2 is located close to the subunit interface, with two ATP molecules binding back-to-back to each subunit of the dimer. Comparison of the basal (apo) and active (allosteric effector bound) structures of NT5C2 revealed a role of the Gly355-Glu364 region in the allosteric activation of the enzyme. This segment (helix A) is disordered in the apo inactive form of the enzyme and adopts an ordered α-helix conformation in the effector-bound activated state. This conformational change facilitates substrate binding and catalysis by inducing a rotation of Phe354 out of the active site and by moving Asp356 into the catalytic center (Wallden and Nordlund, 2011). The relevance of helix A in the activation of NT5C2 is supported by modeling studies implicating local conformational changes in this region as responsible for the increased nucleotidase activity of a strongly activating NT5C2 mutation mapping to this site (NT5C2 p.K359Q) (Tzoneva et al., 2013). Yet, the majority of NT5C2 mutations found in relapse ALL are located outside of this helix A region, suggesting that additional regulatory elements may be involved in the control of NT5C2 activity. However, these mutations show an apparent random spatial distribution in the structure of the enzyme and their mechanisms of action remain to be established.

Results

Functional heterogeneity of NT5C2 mutations in relapsed ALL

To gain further insight on the role and mechanisms of NT5C2 in resistance to 6-MP chemotherapy we compiled allelic data on 643 relapsed ALL cases along with previously published data (Kunz et al., 2015; Ma et al., 2015; Meyer et al., 2013; Oshima et al., 2016; Richter-Pechanska et al., 2017; Tzoneva et al., 2018; Tzoneva et al., 2013) identifying in all 32 independent NT5C2 mutant alleles consisting of 27 single amino acid substitutions, four in frame indel mutations (p. Asp396_Ala400del, p.Lys404delinsLysAsp, p. Ser408_Asp415del and p.Ser445_Arg446delinsPhe_Gln) and a C-terminal truncating mutation (p.Gln523*) (Figure 1A and Table S1). Gain of function mutations resulting in increased enzymatic activity typically cluster in defined protein domains involved in enzyme regulation (Gao et al., 2017; Tamborero et al., 2013). In this context, we identified recurrent relapsed leukemia-associated NT5C2 mutations involving Arg39, Arg238, Arg367, Leu375, Asp407 and Pro414 supporting a potentially important role of these residues in the control of NT5C2 activity. However, each of these recurrent mutations show inexplicit functional features and none of them are located in the catalytic regulatory helix A region.

Figure 1.

Distribution and activity profile of NT5C2 mutations in human relapsed acute lymphoblastic leukemia. (A) Graphical representation of relapse associated mutations in B-ALL (black) and T-ALL (red). HAD core motifs are shown in black rectangles with roman numerals. (B–D) In vitro nucleotidase assays assessing the enzymatic activity of class I (B), class II (C), and class III (D) NT5C2 mutations in the presence of increased concentrations of ATP represented as specific activity (top) and specific activity relative to no ATP (bottom). Data in B–D are shown as mean ± s.d. See also Table S1.

To functionally explore the role of NT5C2 mutations as drivers of increased nucleotidase activity and thiopurine resistance, we tested their activity and response to ATP allosteric activation (Marques et al., 1998; Pinto et al., 1986; Spychala et al., 1988) in in vitro nucleotidase assays. These analyses revealed distinct patterns of enzymatic activity and responses to ATP. Purified recombinant NT5C2 K359Q and NT5C2 L375F (class I mutations) showed markedly increased basal nucleotidase activity with reduced response to allosteric activation (~1.5 fold increase over basal at 0.3 mM ATP), below that observed in wild-type recombinant NT5C2 (2.6 fold over basal at 0.3 mM ATP) (Figure 1B). In contrast, most relapse-associated activating NT5C2 alleles (R39Q, R238W, R367Q, D407A, S408R, S445F and R478S) (class II mutations) showed increased nucleotidase activity with retained or increased dynamic response to ATP allosteric regulation (2.3–4.6 fold over basal at 0.3 mM ATP) (Figure 1C). Finally, a C-terminal truncated mutant protein, NT5C2 Q523X, showed nucleotidase activity levels similar to those of the wild-type NT5C2 in basal conditions but was hyper-responsive (6.5 fold over basal at 0.3 mM ATP) to allosteric activation (class III mutation) (Figure 1D).

NT5C2 structure and activation-associated conformational changes

To investigate the molecular basis mediating the heterogeneous patterns of activity of NT5C2 activating mutations, we determined crystal structures of the basal and active states of relapse-associated mutants and wild-type full length human NT5C2 proteins at up to 1.8 Å resolution (Figure S1A–D and Tables S2 and S3). The structure of the full length human NT5C2 monomer revealed a HAD core (residues 30–372 and 435–478) with three extensions consisting of the N-terminal (residues 1–29), helical arm (residues 373–434) and C-terminal (residues 479–561) regions (Figure 2A–F). Moreover analysis of the structures of wild-type human NT5C2 in the basal (Apo) and the activated (P_i bound in the effector site) state show marked conformational and functional changes in the monomer (Figure 2A–B), dimer (Figure 2C–D) and tetramer (Figure 2E–F) of the enzyme upon activation, which were selectively altered by each class of NT5C2 mutations. Moreover, analysis of the buried surface area at the interface of two different (strong and weak) dimers, forming the NT5C2 tetramer showed that the strong dimer in each basal structure appears to bury twice as much area at its dimer interface compared to its active counterpart supporting that the NT5C2 dimer and tetramer adopt a more open configuration upon allosteric activation (Figure S1E–F and Table S4).

Figure 2.

Crystal structure of basal and activated wild-type human NT5C2. (A) A Ribbon diagram of the basal structure of the full-length wild-type NT5C2 (WT), displaying the HAD domain (cyan) and its three extensions: N-terminal segment (orange), arm region (pink), and C-terminal segment (marine). The mutation sites are depicted as red solid spheres for Cα of each mutated amino acid. The phosphate (yellow for phosphorus) ions are depicted as stick models and labeled as P_i. The N- and C-termini amino acids (L24 and D552), and the termini amino acids (T510 and P541) of the disordered region in the C segment are also labeled. (B) A Ribbon diagram of the active structure of NT5C2 WT, in which the allosteric helix A (αA) is shown in dark purple. The N- and C-termini amino acids (S4 and S488), and the termini amino acids (L402 and R421) of the disordered region in the arm segment are also labeled. (C, D) Ribbon and surface (for subunit B) depictions of basal (C) and active (D) dimers of WT. (E, F) Ribbon diagrams of the basal (E) and active (F) WT tetramers, respectively. Solid blue oval and blue arrows represent crystallographic two-fold, while solid red arrow represents non-crystallographic two-fold. See also Figures S1 and Tables S2–S4.

Mechanism of NT5C2 class I mutations

In agreement with previous studies (Wallden and Nordlund, 2011), we observed residues 353–363 in the HAD core of NT5C2 change from a loop structure to a helix (helix A) upon activation. This conformational change is triggered by electrostatic interactions between Lys362 and the phosphate moieties of the ATP effector and mediates the activation of the enzyme by moving Phe354 out of and Asp356 into the active site, enabling it for substrate binding and catalysis (Figure S2) (Wallden and Nordlund, 2011). Of note, this active structure is stabilized by ionic interactions between Lys361 in helix A and Asp459, which is also positioned between Arg39 and Arg367 (Figure S2). Notably, analysis of the crystal structures of NT5C2 K359Q (Figure 3A) and NT5C2 L375F (Figure 3B, C), showed that this mechanism is hijacked by these class I NT5C2 mutants, which adopt an active conformation even without ATP or P_i bound in the effector site. Thus, and consistent with modeling analyses (Tzoneva et al., 2013), in NT5C2 K359Q, a mutation located within the helix A region of NT5C2, the Gln359 side chain contacts both the backbone and the side chain of Phe354, thereby promoting the formation of an active helix A configuration (Figure 3A). In addition, the presence of Gln at position 359 alters the neighboring network of interactions between Lys361 and Asp459 by abolishing one of two hydrogen bonds between Arg367 and Asp459, thereby allowing for increased stabilization of helix A (Figure 3A). Notably, the helix A region in the second class I mutant protein NT5C2 L375F is also organized in an active helical conformation in the absence of ATP or P_i (Figure 3B), as observed before (Hnizda et al., 2018; Hnizda et al., 2016). Yet, the specific mechanisms driving this active conformation were not directly apparent as Leu375 is located away from the helix A region (Figure 3C). Mechanistically, detailed inspection of the NT5C2 L375F full length structure revealed that the extra bulk of the Phe side chain forces a shift of ~1 Å in the Cα position of L375F and displaces the underlying helical arm (Figure 3C). As a result, Arg98, which makes two salt bridges with Glu373, and one H-bond with the backbone of Asn154 in basal and active wild-type NT5C2 and NT5C2 K359Q, makes only one salt-bridge with Glu374 in NT5C2 L375F, which, in turn, favors the interaction of Phe450 with Ile353 and an active conformation of helix A (Figure 3B). In addition, the Phe side chain introduced by the L375F mutation also engages in enhanced interactions with hydrophobic residues Phe36, Leu379, Phe441, Tyr461 and His486 from the neighboring NT5C2 subunit through favorable π-stacking (Figure 3C). These results demonstrate that an active conformational state of the helix A region in the absence of allosteric modulator mediates the gain of function properties of NT5C2 K359Q and NT5C2 L375F class I mutants.

Figure 3.

Crystal structures of class I NT5C2 mutants. (A), Structure overlay of the active full- length WT NT5C2 (WT) (gray for Cα atoms) and the constitutively active K359Q NT5C2–537X (K359Q-537X) (color), in which the interaction between K359Q and F354 is shown. The hydrogen bonds are shown in red dash lines and black, the latter of which is present only in the structure of WT. (B) Structure overlay of L375F (color) with the active WT (grey). The residues affected by the L375F mutation are shown with stick models and colored accordingly. (C) Structure overlay of WT (gray) and the constitutively active full-length NT5C2 L375F (L375F) (color), shows the mutation site L375F at the base of the arm extension. The hydrophobic residues surrounding L375F are shown as stick models and labeled. See also Figures S2.

Although the underlying mechanisms mediating the increased activity of non-class I NT5C2 mutant proteins did not became directly apparent in the analyses of the conformational changes in the helix A segment, it did not escape our attention that activation associated changes in NT5C2 also involve shifts in the position of the helical arm and the C-terminal segments (Figure 2A–F). We hypothesized that despite their distant location from the catalytic site, these elements could also have a regulatory role with potentially important implications for understanding the mechanisms of action of class II and class III NT5C2 mutant proteins.

Mechanism of NT5C2 class II mutations

Close inspection of the distribution of class II NT5C2 mutations revealed a group of NT5C2 alleles (R39Q, R238W/L/G/Q, R367Q, S445F_R446Q, R478S) located in a positively-charged pocket at the interface of the tightly-associated dimer (Figure 2A–D and Figure S3A), while a second group of class II mutations (K404N, 404–405insD, D407A/Y/E/H, S408R, P414S/A and D415G) locate in the tip region of the arm segment (Figure 2A–D). Notably, all class II NT5C2 mutant variants characterized in our enzymatic assays localize in these two regions. Based on this observation we proposed that class II mutants located in the positively charged pocket region (R39Q, R367Q, S445F, R238W, R478S) and loop segment (D407A, S408R) could share a common mechanism of action. Consistently, the crystal structures (Table S2) of a recurrent mutation (R238W) involving the R238 residue located in the protein surface adjacent to the entrance of the inter-subunit pocket (Figure S3B), the structure of a representative allele (D407A) involving the D407 residue in the tip region of the arm segment (Figure S3C), and that of two recurrent class II pocket mutations (R39Q and R367Q) involving basic amino acids at the bottom of the positively-charged inter-subunit pocket (Figure S3D), showed a common conformation in which the helix A region adopted an active structure similar to that of wild-type NT5C2 in the presence of allosteric activator (Figure 4A–D). These results support that the positively-charged pocket and the tip of the arm segment could form a single functional unit. In considering the potential mechanism mediating the effects of NT5C2 class II mutants it is worth noting that NT5C2 activation results in dynamic configuration changes of the tip region of the arm segment. Specifically, the loop at the tip of the helical arm segment of NT5C2 establishes intimate contacts with the other monomer in the basal structure, yet these contacts are lost upon activation with the helical arm pivoting away from the other monomer and the tip region becoming disordered upon allosteric activation supporting a role for this element in NT5C2 regulation (Figure 2A–F).

Figure 4.

Crystal structure of NT5C2 class II mutations. (A) Structure overlay of the basal full-length NT5C2 R367Q (R367Q) (color) and the basal full-length wild-type NT5C2 (WT) (gray for Cα atoms), displaying the side chains of important residues near effector binding site. (B) Structure overlay of the basal full-length NT5C2 R39Q (color) and the basal wild-type NT5C2 (gray for C atoms), displaying the side chains of important residues near effector binding site. A salt-bridge between R367 (in cyan) and D551 (in marine) from the C-terminal segment is shown with a red dash line. (C) Structure overlay of the active R367Q (color) and the active WT (gray), showing the mutant R367Q does not form a hydrogen bond with D459. The two phosphate ions, present in the active WT are depicted as stick models. (D) Structure overlay of the active full-length NT5C2 R39Q (R39Q) (color) and the active WT (gray). ATP and Mg²⁺ ion, present in the R39Q structure, are depicted as a stick model and a dark green solid sphere, respectively. The hydrogen bonds in the R39Q structure are shown as red dash lines, while the black dash lines represent the hydrogen bonds in the active WT structure. See also Figure S3.

To evaluate the regulatory capacity of the tip of the helical arm segment we performed a CRISPR/Cas9 gain-of-function screen. While CRISPR/Cas9-induced frameshifts and in-frame indels targeting critical enzymatic domains are likely to be loss-of-function, in-frame alleles involving negative regulatory domains can produce protein variants with increased enzymatic activity (Donovan et al., 2017; Ipsaro et al., 2017). Such gain-of-function alleles are selectable in functional screens by their capacity to confer resistance to specific inhibitors (Donovan et al., 2017; Ipsaro et al., 2017). Here we aimed to test the potential role of the tip region of the arm segment as a potential negative regulator of NT5C2 activity following this strategy. Towards this goal we expressed Cas9 and gRNAs targeting two sites in the Nt5c2 locus at codons H405 and I416 in the tip segment of the arm domain in mouse leukemia lymphoblast cells and then selected potential gain-of-function CRISPR-directed non-homologous end joining generated mutations resulting in increased NT5C2 nucleotidase activity for their capacity to confer resistance to 6-MP (Figure S4A). Deep sequencing analysis of CRISPR/Cas9-induced mutations treated with 6-MP revealed a positive selection of Nt5c2 in-frame indels and point mutations compared to vehicle treated controls (Figure 5A and B). Characterization of a selection of positively selected in-frame alleles demonstrated gain-of-function effects with increased nucleotidase activity (Figure S4B and Table S5). Consistently, extended targeted mutation analysis of the tip region of the arm segment of NT5C2 in relapsed ALL patient samples identified an in-frame NT5C2 deletion removing much of this element (delS408-D415) (Figure S4C). Notably, this mutation behaved as a class II gain-of-function allele in enzymatic assays and conferred resistance to 6-MP chemotherapy when expressed in ALL cell lines (Figure S4D–E). In all, these results support that local disruption of the tip region of the arm segment of NT5C2 leads to deregulated enzymatic activity and point to this segment as a negative regulatory element disrupted by a subset of class II NT5C2 mutations in relapsed ALL. Should the tip of the arm segment and the positively-charged inter-monomeric pocket targeted by class II NT5C2 mutations form a bipartite regulatory unit, then we predicted that these two regions could physically interact. In support of this possibility, the tip region of the arm segment of NT5C2 (Ser408) contacts the neighbor subunit (Arg238) in the absence of allosteric activator, which places Asp407, a recurrently mutated residue in the tip region, at the entrance of the positively charged dimer interface (Figure S4F). Moreover, molecular modeling analyses of the arm segment of NT5C2 revealed that, following allosteric stimulation, the NT5C2 Glu401-Asp415 segment can invade the positively charged inter-subunit pocket, which opens up upon activation (Figure 5C and Figure S4G). Notably, the Asp407 residue targeted by recurrent mutations in the arm segment plays an important role in this transition by interacting with positively charged residues along the surface of the inter-subunit pocket (Figure S4G). Markedly, the most favorable (lowest potential energy) conformation for Asp407 in the activated state of NT5C2 places this residue interacting via hydrogen bonding with Lys361 on the helix A segment of the neighboring protomer (Figure S4H). This contact, in turn, would destabilize the interaction between Lys361 and Asp459 responsible for maintaining the enzyme in the activated state (Figure 5D) and return NT5C2 to its basal inactive conformation. Based on these results we propose that class II mutations would disrupt the dynamic movement of the tip region of the arm domain through this intermonomeric cavity (Figure S4).

Figure 5.

Functional characterization and modeling of the arm segment region targeted by class II NT5C2 mutations. (A and B) Graphical representation of mutations selected for in murine leukemic cells infected with a gRNA targeting H405 (A) or I416 (B) after treatment with vehicle or 2 μM 6-MP. (C) Modeling analysis of the top 20 preferred conformations of the flexible arm segment in the inactive and allosterically activated NT5C2. (D) A close up view of the proposed interaction between Asp407 and Lys361 causing disruption of the alpha helix stabilizing Asp459 – Lys361 interaction. (E) Western blot showing specificity of antibodies generated against the tip region of the arm domain for NT5C2 (left) and in vitro nucleotidase assays of purified wild-type NT5C2 recombinant protein incubated with two unique arm segment antibodies or the arm segment peptide in the presence of increasing doses of ATP (right). Data in E is shown as mean ± s.d. p values were calculated using two-tailed Student’s t-test, ** p <0.001, *** p <0.0001. See also Figure S4 and Table S5.

To formally test this model, we blocked the dynamic movement of the tip region of the arm segment loop containing Asp407 in wild-type NT5C2 using two independent antibodies against this element and tested the effects of this perturbation in NT5C2 activity. In these experiments, incubation of wild-type NT5C2 recombinant protein with tip region antibodies resulted in increased NT5C2 nucleotidase activity in basal conditions and upon allosteric activation phenocopying the effect of class II mutations (Figure 5E). Importantly, the activating effect of these NT5C2 antibodies was reversed by pre-incubation with a tip region peptide (Figure 5E). In all, these results identify a dynamic switch-off NT5C2 auto-regulatory mechanism mediated by the interplay of the tip of the arm segment and the positively-charged inter-monomeric pocket targeted by class II relapsed-associated NT5C2 mutations.

Mechanism of the class III NT5C2 Q523X mutation

Distinct from class I and class II mutations, the Q523X C-terminal truncating mutant shows nucleotidase activity levels similar to that of wild-type NT5C2 but a more dynamic response to allosteric activation, in support of a third mechanism of NT5C2 regulation (Figure 6A). Of note, the C-terminal region of NT5C2 is highly conserved in vertebrates and contains a stretch of acidic residues at its end (Figure 6B). Moreover, this segment undergoes major conformation changes in the context of NT5C2 activation suggesting a regulatory role for this region (Figure 2A–F).

Figure 6.

Functional characterization of the NT5C2 C-terminus and class III mutation. (A) In vitro nucleotidase assays assessing the enzymatic activity of class III NT5C2 Q523X mutation in the presence of increased concentrations of ATP. (B) Protein sequence alignment of the C-terminal of NT5C2 in vertebrate species. (C) Basal structure of the full-length NT5C2 R39Q (R39Q), displaying the interactions of predominantly acidic region of the C-terminal segment with the positively-charged residues at the dimer interface. Red dash lines depict hydrogen bonds. The side chains of residues that are involved in polar interactions are shown by stick models. The side chain of R39Q is colored with light magenta, while others are colored accordingly. (D) A surface potential depiction of the dimer interface in the basal structure of the full-length R39Q (R39Q). The acidic tail of the protomer A is shown with a stick model in which several acidic residues are labeled. (E) Western blot showing specificity of antibodies generated against the C-terminus of NT5C2 (top) and in vitro nucleotidase assays of wild-type NT5C2 purified recombinant protein incubated with two unique C-terminus targeted antibodies or the C-terminus peptide in the presence of increasing doses of ATP (bottom). (F) Structure overlay of the basal NT5C2 Q523X (Q523X) (ribbon) and the basal full-length NT5C2 R39Q (R39Q) (surface), showing the lack of the C-terminus (marine) in Q523X results in a more open tetramer as compared to that of R39Q. Data in A and E are shown as mean ± s.d. p values were calculated using two-tailed Student’s t-test, * p <0.05, ** p <0.001, *** p <0.0001. See also Figure S5.

In the absence of an allosteric regulator, the C-terminal segment of one NT5C2 subunit (residues 479–510 and 537–553) folds over the base of the arm segment of the neighboring protomer so that the acidic C-terminal tail introduces itself into the positively charged inter-subunit pocket promoting a compact dimer and a tightly-associated NT5C2 tetramer conformation (Figure 6C, 6D and Figure S5A). Of note, this tightly closed conformation is released in the allosterically activated structure of NT5C2 in which the C-terminal is no longer visible at the dimer interface and is also displaced from the surface of the neighboring subunit by the N-terminal segment (Figure S5B). These results support that the acidic tail of NT5C2 may provide a hindrance towards NT5C2 activation by stabilizing the closed inactive conformation of the enzyme. To functionally evaluate the role of the C-terminal segment of NT5C2 and its interaction with the positively charged inter-subunit pocket in the control of NT5C2 activation we tested the effects of incubating wild-type NT5C2 protein with polyclonal antibodies recognizing the C-terminal acidic tail in nucleotidase assays. In these experiments two independent antibodies recognizing the C-terminal acidic tail of NT5C2 induced increased enzymatic activity in response to allosteric activation (Figure 6E). Moreover, the activating effect of these antibodies was reversed by pre-incubation with a NT5C2 C-terminal tail blocking peptide, indicating that disrupting the interaction of the C-terminal acidic tail of NT5C2 with the positively-charged inter-subunit pocket by antibody binding can reproduce the effects of the class III NT5C2 Q523X mutation (Figure 6E). Finally, and as predicted by this model, the apo crystal structure of NT5C2 Q523X (Table S2) revealed a dimer structure more open than that present in the apo form of NT5C2 full length crystals (Table S4), and its active structure showed a helix A conformation similar to that of active WT NT5C2 (Figure 6F, Figure S1). In heterozygous NT5C2 mutant cells, both homotypic and heterotypic dimers should be present (Figure S5C). In cells harboring the NT5C2 Q523X mutant allele, homotypic (Q523X/Q523X or WT/WT) and heterotypic (WT/Q523X) dimers can be combined in six different tetramer configurations (WT/WT+WT/WT; WT/WT+WT/Q523X; WT/Q523X+WT/Q523X; Q523X/Q523X+Q523X/WT; Q523X/Q523X+WT/WT and Q523X/Q523X+Q523X/Q523X). In this context we predict that the “opening” effect of the Q523X truncating mutation would be most prominent in homotypic mutant dimer (Q523X/Q523X) containing tetramers compared with complexes containing heterotypic (WT/Q523X) protein pairs. Yet, in all, our results support that loss of the C-terminal tail of NT5C2 in the Q523X class III mutation decreases the threshold for allosteric NT5C2 activation by promoting a more relaxed dimer and tetramer configuration.

Discussion

Detailed functional understanding of the action of biological molecules involved in disease is essential for the development of rational therapeutic strategies, yet deciphering the underlying role of dynamic components based on structural information remains a daunting task. The structural and biochemical studies of NT5C2 presented here provide significant insights into the mode of action and regulation of this nucleotidase, elucidating specific mechanisms responsible for stabilizing the basal inactive configuration of the enzyme, for triggering allosteric activation, and for returning the enzyme to its basal inactive state. Most critically, these findings shed light on the mechanisms of activating mutations in NT5C2 responsible for driving resistance to 6-MP chemotherapy in relapsed ALL setting the stage for the development of small-molecule NT5C2 inhibitors for the reversal of chemotherapy resistance in human leukemia.

Enzymatic analyses of recombinant proteins representative of NT5C2 mutations distributed in different regions of the protein in basal conditions and in response to allosteric activation confirmed the gain of function nature of relapse associated NT5C2 alleles. Yet, these analyses revealed diversity in the enzymatic properties of these mutations. We distinguished three groups of mutations based on their enzymatic profile. Class I mutations encompass alleles K359Q and L375F, which show a strong increase in nucleotidase activity in the absence of allosteric activation. Class II alleles, which account for >95% of relapse-associated NT5C2 mutations, show increased levels of nucleotidase activity in basal conditions, yet their enzymatic activity is still dynamically increased in response to allosteric effector. Finally, the class III group of NT5C2 mutations is defined by the C-terminal truncating NT5C2 Q523X allele, which shows nucleotidase activity similar to that of wild-type NT5C2 in basal conditions but is hyperresponsive to allosteric activation.

Two distinct selection pressures influence the clonal evolution of NT5C2 mutations in ALL. In the absence of 6-MP chemotherapy, depletion of the intracellular nucleoside pool as a result of increased activity of this enzyme normally involved in purine degradation, can negatively impact the growth of NT5C2 gain-of-function bearing leukemia clones (Tzoneva et al., 2018). However, increased clearance of thiopurine metabolites during 6-MP therapy confers a selective advantage to cells harboring activating NT5C2 alleles. Class II NT5C2 mutations, by far the most prevalent group in relapsed ALL, balance these two opposing selective pressures by conferring increased nucleotidase activity, yet retaining control by allosteric regulation. It is possible that the rare, yet stronger, class I alleles are less frequently found because of a higher fitness cost in the absence of 6-MP chemotherapy and that weaker class III NT5C2 mutations are poorly selected as result of a more limited capacity to confer resistance to 6-MP.

NT5C2 is kept in an inactive state in the absence of allosteric activators binding to the effector site. This inactive basal conformation is secured by the C-terminal acidic tail, which inserts itself in between the two NT5C2 dimer subunits establishing multiple interactions with positively charged residues. In this, the C-terminal segment of NT5C2 functions as a fastener securing a tight closed inactive configuration of the enzyme. As a result, NT5C2 must undergo structural rearrangement in order to switch to an active configuration. This transition to an active state, involves the organization of the N-terminal segment of the protein in a helical configuration. This displaces the C-terminus from the surface of the neighboring subunit and removes it from the intermonomeric space, now favoring a more open dimer conformation. In addition, binding of ATP to the effector site induces allosteric activation by inducing the helical conformation of the helix A segment, which reconfigures the catalytic center for substrate accessibility and catalysis. Yet, the transition to an activated state is coupled with a mechanism that returns the enzyme back to its inactive basal configuration. As the intersubunit space opens, Asp407 in the tip segment of the arm domain is mobilized from its location next to the opening of this cavity and is pulled into the intermonomeric space via interactions with multiple positively charged residues lining the surface of this pocket. In this way, the negatively charged Asp407 ultimately reaches and destabilizes the helix A segment, located at the bottom of the intersubunit pocket. In this, a mechanistic parallel can be seen between NT5C2 and Marvin Minsky’s “ultimate machine” design, a self-inactivating device consisting of a box with a switch which, when turned "on", activates a lever that appears from inside the box and turns the switch back "off".

Activating gain-of-function mutations driving oncogenic transformation involving multiple different mechanisms are recurrently found in human cancer. Mutations in the PEST domains of MYC (Salghetti et al., 1999) and NOTCH1 (Weng et al., 2004) result in increased protein stability via disruption of degron motifs responsible for protein turnover via proteasomal degradation. Activating mutations in RAS signaling factors lock these proteins in their active GTP-bound configuration by interfering with their intrinsic GTPase activity (Scheffzek et al., 1997). In addition, kinase oncoproteins are frequently activated by mutations that interfere with intramolecular interactions responsible for keeping them inactive in basal conditions. For example, lymphoma-associated activating mutations involving the SH2 domain and the C-terminal region of FYN kinase disrupt the inhibitory interaction of these elements induced by CSK phosphorylation of the FYN C-terminal region Y531 (Palomero et al., 2014). At other times, activating mutations favor activation by interfering with allosteric inhibitory feedback mechanisms, as is the case in PFK1 mutations, which drive increased glycolytic activity in some tumor cells (Webb et al., 2015). In this regard, and of relevance to 6-MP resistance, activating mutations in PRPS1 found in leukemia interfere with allosteric feedback inhibition of this enzyme by purine metabolites resulting in increased nucleotide biosynthesis and consequent inhibition of 6-MP activation (Li et al., 2015). Some aspects of NT5C2 activation by relapsed leukemia-associated mutations such as the disruption of intramolecular interactions responsible for maintaining the enzyme in an inactive configuration (class III mutations) or the mimicry of the effects of allosteric regulation with the catalytic center (class I mutations) relate to some of those found in other proteins with cancer-associated activating mutations. Yet, to our knowledge, the dynamic switch-off of the activated helix A mediated by the tip region of the arm domain and the positively charged intermonomeric unit targeted by class II mutations represents a self-regulatory mechanism distinct from those present in other enzymes harboring gain-of-function mutations.

Large-scale cancer sequencing efforts have uncovered a complex landscape of somatic mutations associated with malignant transformation (Cancer Genome Atlas Research et al., 2013). However, the interpretation of individual mutations can be challenging in the absence of mechanistic information, thus highlighting the need for in depth functional analyses. This imperative is particularly pressing in the case of mutations potentially driving clinically relevant phenotypes such as response or resistance to therapy. The results reported here demonstrate a functional and structural diversity of leukemia-associated NT5C2 mutations and provide a framework for the functional interpretation of these genetic lesions in the clinic. Thus, mutations involving the tip segment of the arm region and residues located in the intersubunit space, particularly those involving positively charged residues, and truncating mutations involving the C-terminal segment of NT5C2 should be considered highly likely activating alleles. Finally, the dependence of NT5C2 class II and class III mutant proteins, which encompasses >95% of relapse associated NT5C2 mutations, on allosteric activation points to the allosteric effector site as a potentially attractive target for the design of small molecule NT5C2 inhibitors for the reversal of 6-MP resistance in relapsed ALL.

STAR Methods

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled according to institutional rules by the Lead Contact, Adolfo Ferrando (af2196@columbia.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Patient samples

DNAs from leukemic ALL blasts at relapse were provided by the Department of Pediatric Oncology/Hematology at the Charité-Universitätsmedizin Berlin in Berlin, Germany. Informed consent was obtained at study entry and samples were collected under the supervision of local Institutional Review Boards for participating institutions (Charité-Universitätsmedizin and Columbia University Medical Center) and analyzed under the supervision of the Columbia University Medical Center Institutional Review Board.

Mice

We maintained all animals in specific pathogen-free facilities at the Irving Cancer Research Center at Columbia University Medical Campus. The Columbia University Institutional Animal Care and Use Committee (IACUC) approved all animal procedures. To generate NOTCH1-induced T-ALL tumors in mice, we performed retroviral transduction of bone marrow cells (C57BL/6) enriched in Lineage negative cells isolated using magnetic beads (Lineage Cell Depletion Kit, Miltenyi Biotec) with retroviruses expressing an activated form of the NOTCH1 oncogene (ΔE-NOTCH1) (Schroeter et al., 1998) and the red fluorescent protein (RFP) and transplanted them via intravenous injection into lethally irradiated isogenic recipients (6–8 week old C57BL/6 females, Jackson Labs) as previously described (Herranz et al., 2014; Herranz et al., 2015).

Cell Culture

We performed cell culture in a humidified atmosphere at 37°C under 5% CO₂. We purchased 293T cells for viral production from American Type Culture Collection and grew them in DMEM media supplemented with 10% fetal bovine serum (FBS), 100 U ml⁻¹ penicillin G and 100 μg ml⁻¹ streptomycin for up to two weeks. We purchased JURKAT TET3G cells from Clontech and cultured them in RPMI-1640 media supplemented with 10% tet-approved FBS, 100 U ml⁻¹ penicillin G and 100 μg ml⁻¹ streptomycin. Cell lines were regularly authenticated and tested for mycoplasma contamination. We cultured ΔE-NOTCH1 driven mouse leukemic cells in OptiMEM media supplemented with 10% FBS, 100 U ml⁻¹ penicillin G, 100 μg ml⁻¹ streptomycin, 55 μM β-mercaptoethanol and 10 ng ml⁻¹ mouse IL7.

METHOD DETAILS

Drugs

We purchased 6-mercaptopurine (6-MP) and ATP from Sigma-Aldrich. For in vitro assays we dissolved 6-MP in DMSO and ATP in Reagent 2 of the 5′-NT Enzymatic Test Kit (Diazyme).

Site-directed mutagenesis

We generated the NT5C2 mutations by site-directed mutagenesis on the mammalian expression pLOC-NT5C2 vector (Open Biosystems) using the QuikChange II XL Site-Directed Mutagenesis Kit (Stratagene) according to the manufacturer's instructions.

Plasmids and vectors

We obtained the pET28aLIC (Plasmid #26094) and pL-CRISPR.efs.gfp (plasmid # 57818) plasmids from Addgene and the pLVXTRE3GZsGreen1 vector from Clontech. We amplified the coding sequence of the NT5C2 cDNA from pLOC-NT5C2 (Tzoneva et al., 2013) and cloned it into the pET28aLIC vector using In-fusion cloning using the In-Fusion HD Cloning Kit (Clonetch) following manufacturer guidelines. We cloned the NT5C2 S408-D415 loop deletion mutation into the pET28aLIC and pLVXTRE3GZsgreen1 vector using Gibson Assembly using the Gibson Assembly Master Mix (New England Biolabs) following manufacturer guidelines. We cloned a truncated active form of NOTCH1 ΔE-NOTCH1 (Schroeter et al., 1998) into the pMSCV-pBabeMCS-IRES-RFP retroviral vector (Addgene plasmid # 33337). We generated lentiviral vectors expressing CAS9 and gRNAs targeting the arm segment of mouse Nt5c2 by cloning the corresponding gRNA oligonucleotides (Sigma-Aldrich) into pL-CRISPR.efs.gfp as reported (Shalem et al., 2014).

Retroviral and Lentiviral production and infection

We transfected lentiviral plasmids together with gag-pol (pCMV ΔR8.91) and V-SVG (pMD.G VSVG) expressing vectors into 293T cells using JetPEI transfection reagent (Polyplus). We collected viral supernatants after 48 h and used them to infect JURKAT Tet-On human cell lines by spinoculation with 4 μg mL⁻¹ Polybrene Infection/Transfection Reagent (Fisher Scientific). We selected infected human cell lines with 1 mg ml⁻¹ puromycin (Sigma Aldrich) for 5 days.

In vitro cell viability and chemotherapy response assays

We analyzed chemotherapy responses of human leukemia cell lines expressing wild-type NT5C2 or NT5C2 S408-D415 deletion following 72-hour incubation with increasing concentrations of 6-mercaptopurine by measurement of the metabolic reduction of the tetrazolium salt MTT using the Cell Proliferation Kit I (Roche) following the manufacturer’s instructions. We analyzed human cell lines with inducible expression of wild-type or mutant NT5C2 after 48 hours of doxycycline treatment (1 mg ml⁻¹).

CRISPR guided generation and 6-MP selection of arm domain mutant NT5C2 lymphoblasts

We generated NOTCH1-induced mouse T-ALL tumors via retroviral transduction of mouse hematopoietic progenitors with lentiviral particles expressing a constitutively active truncated NOTCH1 allele (ΔE-NOTCH1) as before (Herranz et al., 2014; Herranz et al.; Schnell et al., 2015). We infected ΔE-NOTCH1 mouse lymphoblasts with lentivirus particles expressing CAS9 and gRNAs targeting Nt5c2 at codons H405 or I416, or with CAS9 only control lentiviral particles. We sorted infected cells based on GFP expression using a SONY SH800S cell sorter (SONY). We treated sorted gRNA expressing cells with 2 μM 6-MP or DMSO control for 72 hours. Following 6-MP treatment, cells were washed and placed in fresh media. Upon complete recovery from treatment, cells were harvested and DNA was isolated using DNEasy Blood &Tissue Kit (Qiagen).

Deep Sequencing

CRISPR Mutagenesis Deep Sequencing

We amplified by PCR genomic DNA sequences encompassing exon16 of mouse Nt5c2 from 50 ng of genomic DNA extracted from mouse NOTCH1-induced T-ALL lymphoblast cells with primers designed according to Fluidigm recommendations and containing Fluidigm-specific adapter sequences at the 5’ ends. We barcoded the resulting amplified PCR products using Illumina –Fluidigm specific barcodes so that each sample carried a unique barcode. We pooled all indexed amplicons and quantified the resulting library with Qbit and BioANalyzer analysis. To increase library diversity we spiked the library with 50% PhiX genomic library and sequenced in a MiSeq instrument to generate 2x251 bp paired end reads.

We combined each read pair into a single sequence by finding the optimal alignment allowing up to 0 mismatches and without gaps. We aligned merged pairs as single-end reads to the mouse mm10 genome build using BWA (Li and Durbin, 2009). We classified non-synonymous SNPEFF (Cingolani et al., 2012) variant calls as either in-frame and point mutations (disruptive inframe deletion, disruptive inframe insertion, inframe deletion, inframe insertion, missense variant, and inframe insertion and splice region variant) or frameshift (disruptive inframe deletion and splice region variant, frameshift variant, frameshift variant and splice acceptor variant and splice region variant and intron variant, frameshift variant and splice donor variant and splice region variant and intron variant, frameshift variant and splice region variant, frameshift variant and stop gained, inframe deletion and splice region variant, splice acceptor variant and inframe deletion and splice region variant and intron variant, stop gained and disruptive inframe insertion, stop gained and inframe insertion, stop gained), and relative counts over the total reads of each class were compared across samples.

Patient Sample Deep Sequencing

We identified NT5C2 variants that differed from the reference genome in targeted resequencing data, containing the NT5C2 gene, generated using the Access Array system from Fluidigm and analyzed them by paired-end sequencing (2 × 150 bp) in a NextSeq500 instrument (Illumina) using the SAVI algorithm (Trifonov et al., 2013).

Recombinant protein production and purification

For 5’-nucleotidase assays in the absence and presence of allosteric activators we cloned, expressed and purified recombinant wild-type and mutant NT5C2 proteins as previously described (Tzoneva et al., 2013). Briefly, we cloned full-length complementary DNA constructs encoding wild-type or mutant NT5C2 with an N-terminal hexahistidine (His₆) tag in the pET28a-LIC expression vector. We expressed recombinant proteins from Rosetta 2(DE3) Escherichia coli cells by induction with 0.5 mM isopropyl-β-D-thiogalactopyranoside for 3 h at 37 °C. We harvested cells and lysed them in lysis buffer (50 mM sodium phosphate, pH 7.4, 100 mM NaCl, 10% glycerol, 5 mM β-mercaptoethanol, 1% Triton X-100, 0.5 mg ml⁻¹ lysozyme and 20 mM imidazole) supplemented with Complete EDTA-free protease inhibitor (Roche). We purified His₆-tagged NT5C2 proteins by binding them to nickel-Sepharose beads and eluting them with 50 mM sodium phosphate, pH 7.4, 100 mM NaCl, 10% glycerol, 5 mM β-mercaptoethanol and 300 mM imidazole. We removed imidazole by buffer exchange using PD-10 desalting columns (GE Healthcare). We assessed protein expression and purity by SDS-PAGE and Coomassie staining.

For X-ray crystallography analyses we cloned full-length (561 amino acids) or C-terminally truncated (amino acids 1–536) complementary DNA constructs encoding wild-type or mutant NT5C2 with an N-terminal hexahistidine (His₆) tag in the pET28a-LIC expression vector. We expressed recombinant proteins from Rosetta 2(DE3) Escherichia coli cells by induction with 0.5 mM isopropyl-β-D-thiogalactopyranoside overnight at 16 °C. We res uspended harvested cells in lysis buffer (50 mM sodium phosphate pH 7.4, 500 mM sodium chloride, 10% glycerol, 0.5 mM TCEP, 20 mM imidazole) supplemented with Complete EDTA-free protease inhibitor (Roche) and lysed cells by sonication. We purified recombinant proteins using an ΔKTA fast protein liquid chromatography system (GE Healthcare) using a 2-step protocol adapted from one previously described (Wallden et al., 2007). We first performed affinity chromatography using a 1 ml Ni²⁺-charged His-Trap HP column (GE Healthcare) equilibrated in lysis buffer. We eluted NT5C2 proteins from the His-Trap column in a step-wise method with elution buffer (lysis buffer with 500mM imidazole) by first setting the buffer ratio to 25% elution buffer for 8 column volumes and then switching to a linear gradient to 100% elution buffer over 10 column volumes. We pooled NT5C2-containing fractions and purified further by size exclusion chromatography using a HiLoad 16/60 Superdex 200 gel filtration column (GE Healthcare) equilibrated in 50 mM sodium phosphate, pH 7.4, 100 mM NaCl, 10% glycerol and 0.5 mM TCEP. We assessed protein expression and purity by SDS-PAGE and Coomassie staining and concentrated protein samples to 4–9 mg/ml.

For 5’-nucleotidase assays of NT5C2 proteins incubated with arm region or C-terminal polyclonal antibodies in presence or absence of their corresponding specific blocking peptides, we expressed proteins as described for standard 5’-nucleotidase assays (Tzoneva et al., 2013). We purified His₆-tagged NT5C2 proteins in a 2-step process. First we bound proteins to nickel-Sepharose beads and eluted them with 50 mM sodium phosphate, pH 7.4, 100 mM NaCl, 10% glycerol, 5 mM β-mercaptoethanol and 300 mM imidazole. We then further purified NT5C2 protein eluates by size exclusion using a HiLoad 16/60 Superdex 200 gel filtration column (GE Healthcare) equilibrated in 50 mM sodium phosphate, pH 7.4, 100 mM NaCl, 10% glycerol and 0.5 mM TCEP.

5'-nucleotidase assays

We assessed 5′-NT activity of purified recombinant wild-type and mutant NT5C2 proteins using the 5′-NT Enzymatic Test Kit (Diazyme) according to the manufacturer's instructions as described previously (Tzoneva et al., 2013). We calculated 5′-NT activity levels using a calibrator of known 5′-NT activity as standard. We performed assays in triplicate in an Infinite M200 Tecan plate reader.

For assays with allosteric activators ATP, was dissolved directly in Reagent 2 of the test kit (containing the substrate IMP) and made serial dilutions to achieve a range of concentrations. For assays with antibodies and peptides (custom-generated by Covance), both antibodies and peptides were resuspended in phosphate buffered saline (PBS) and all corresponding samples and controls were diluted with PBS accordingly. We performed statistical analysis by Student’s t-test.

Immunoprecipitation

293T cells transfected with FLAG-tagged wild-type NT5C2 and HA-tagged NT5C2 R367Q were collected in lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 5% Glycerol, 5 mM β-ME, 0.1 % Triton). We incubated cell lysates with EZview™ Red ANTI-FLAG® M2 Affinity Gel clone M2 beads (Sigma) or EZview™ Red Anti-HA Affinity Gel beads (Sigma) for 2 hours at 4°C. Following four washes with lysis buffer and one wash with PBS, we boiled the beads in 2x SDS-loading buffer, separated them by SDS PAGE and transfered them to a nitrocellulose membrane for western blot analysis. We detected Flag-tagged proteins by immunoblot with a DYKDDDDK Tag Antibody (Cell Signaling Technology Cat # 2368) and HA-tagged proteins with an Anti-HA High Affinity antibody (Roche Cat #11867431001).

Antibody Generation

We generated rabbit polyclonal antisera directed against Keyhole Limpet Hemocyanin-conjugated NT5C2 peptides corresponding to the tip of the arm segment (E399-I414) or the C-terminus segment (Q542-E561) of the wild-type NT5C2 protein. Specific immunoglobulins were purified from rabbit sera by positive affinity purification using the corresponding immobilized peptide columns (Covance).

Structural modeling analyses

Computational modeling analyses and images were generated using Chimera suite (Pettersen et al., 2004). Additional modeling was conducted using Modeller (Eswar et al., 2008) and I-TASSER webserver (Roy et al., 2010). Tip region of the arm domain models were built, refined and scored using Modeller Software suite (Eswar et al., 2008; Fiser et al., 2000). Top models for figures were selected by ranking 5,000 iterations by DOPE score as described elsewhere (Fiser et al., 2000). All secondary structure prediction was conducted using the PredictProtein platform and webserver (Rost et al., 2004). Path prediction and molecular dynamics of NT5C2 models were predicted using UCSF Chimera software (Pettersen et al., 2004). Electrostatics of NT5C2 molecular surfaces were investigated with APBS and PDB2PQR software packages. PDB2PQR submission was run with a PARSE force field, optimal hydrogen bonding network. PROPKA was utilized to assign protonation states for each structure at their respective crystallization pH (Unni et al., 2011). Charge mapping method utilized was cubic Bspline discretization. Mobile ions were not included and Poisson-Boltzmann was run with linearized (lpbe) setting. Supplementary electrostatics analysis and visualization was conducted through use of the Columbic surface coloring algorithm provided in Chimera software package (Pettersen et al., 2004).

Crystallization and structure determination

All crystals of the human NT5C2 were grown using the microbatch under oil method. In all cases, 1–4 μl of the protein solution was mixed with 1–2 μl of the precipitant solution. The truncated wild-type (WT-537X) and its corresponding mutants were grown at 18 °C, whereas all of the full-length wild-type NT5C2 and its corresponding mutants were grown at 4 °C for the first week, and then transferred to 18 °C for the next 3 weeks. All protein samples contain a mutation at position 52 (D52N), and the protein concentration ranges from 3–12 mg/ml in a buffer consisting of 50 mM Na₃PO₄ (pH 7.4), 100 mM NaCl, 10% glycerol, and 0.5 mM TCEP. Beam lines X4C of the National Synchrotron Light Source (NSLS), BL14–1 and BL12–2 of Stanford National Accelerator Laboratory (SLAC), BL501 and BL502 of Berkeley Advanced Light Source (ALS), and SER-CAT of Advance Photon Source (APS) were used for collecting a single-wavelength native data set for each of fifteen crystal structures presented in this study. All of the diffraction images were processed with the HKL2000 package (Otwinowski, 1997). For the sake of clarity, the crystallization experiments are described below in two parts for basal and active forms of NT5C2. All of the chemicals for crystallization purposes were purchased from Sigma-Aldrich and the paraffin oil for micro batch method was purchased from Hampton Inc..

Crystallization of the basal forms of NT5C2

A: Crystallization of the basal WT-537X, R367Q-537X, and Q523X

The basal form of WT-537X was crystallized using a crystallization reagent comprising 100 mM sodium acetate trihydrate (pH 4.6) and 2 M ammonium sulfate. In contrast, R367Q-537X was crystallized using a crystallization reagent comprising sodium acetate trihydrate (pH 4.6) and 30% (v/v) 2-Methyl-2, 4-pentanediol (MPD), 20 mM calcium chloride dihydrate, and 5 mM MnCl₂. The crystals of the basal form of Q523X were obtained by microseeding method using crystals of the basal form of the full-length R39Q (see below) and using the same crystallization condition as that for R367Q-537X, except MnCl₂ was not added into the crystallization mixture. All crystals were cryoprotected by supplementing the crystallization solution with 20% (v/v) glycerol and flash-freezing in liquid nitrogen for data collection at 100 K. Crystals of the three basal forms WT-537X, R367Q-537X and Q523X diffracted X-rays to 2.91 Å, 2.80 Å, and 2.26 Å resolution respectively, and they belong to space group C2, and there are two protomers in the asymmetric unit (ASU) of the crystals.

The structure of the active form of WT-537X NT5C2 in complex with ATP, Mg²⁺, and IMP (PDB id: 2XCW) was used as a search model to determine the structure of the basal form of WT-537X NT5C2 using the molecular replacement method with the program COMO (Jogl et al., 2001). The entire loop at the tip of the arm segment and four additional residues (489–492) of the C-terminal segment were manually built with the program XtalView (McRee, 1999). Most stages of the structure refinement were performed using CNS 1.3 (Schroder et al., 2010), and PHENIX was used at the final stage of refinement for this and other structures reported in this study (Adams et al., 2010). Crystal structure of the basal R367Q-537X was subsequently determined using the basal structure of WT-537X using a similar methodology.

B: Crystallization of the basal full-length WT, R567Q, and R39Q

The basal forms of the three full-length enzymes, WT, R367Q, and R39Q, were crystallized using a crystallization condition consisting of 100 mM HEPES (pH 7.5), 10% (w/v) PEG 3350, and 200 mM L-proline. Whereas crystals of the full-length R367Q (R367Q) were grown without microseeding, those of the full-length WT (WT) and R39Q (R39Q) were obtained by microseeding method. A few large crystals of R367Q were used as seeds for growing crystals of R39Q, while seeds from R39Q crystals were used for growing crystals of WT. All crystals were cryoprotected by supplementing the crystallization solution with 20% (v/v) ethylene glycol and flash-freezing in liquid nitrogen for data collection at 100 K. Crystals of WT, R367Q, and R39Q, respectively, diffracted X-rays to 2.48 Å, 2.50 Å, 2.31 Å resolution, and they all belong to space group C222₁ with two protomers in ASU of each crystal. These structures were subsequently determined using the structure of basal form of WT-537X as the search model in the molecular replacement method. These structures revealed mostly ordered C-terminal segment, which was reported for the first time in this study.

Crystallization of the active form of NT5C2

A: Crystallization of the active forms of L375F-537X, K359Q-537X, R367Q-537X, R238W-537X, D407A-537X, and Q523X

Crystals of L375F-537X were grown using the crystallization cocktail consisting of 1.8 M ammonium citrate tribasic (pH 7), while those of K359Q-537X were grown in presence of ATP, Mg²⁺, and IMP and by using the crystallization cocktail comprising 0.1 M HEPES (pH 7.5), 20% (w/v) PEG 1000, and 0.1 M ammonium bromide. In both cases, the structures were in active form, and in the latter case no detectable electron density was observed for ATP, Mg²⁺ or IMP. The crystals, respectively, diffracted X-rays to 2.9 Å and 1.8 Å resolution. Crystals of R367Q-537X were grown in presence of ATP, Mg²⁺, and IMP using the crystallization reagent consisting of 0.1 M HEPES (pH 7.5), 20% (w/v) PEG 1000, and 0.1 M ammonium nitrate. Crystals diffracted X-ray to resolution 2.35 Å. Whereas IMP has well-defined electron density, there is no interpretable electron density for ATP or Mg²⁺.

Crystals of R238W-537X were grown in presence of ATP, Mg²⁺, and IMP using the crystallization cocktail consisting of 35% (w/v) tascimate (pH 7). Clear electron density was observed for both ATP and Mg²⁺ ion. Seeds from crystals of the full-length L375F (see below), in complex with ATP and Mg²⁺, were used to grow crystals of D407A-537X in presence of ATP, Mg²⁺, and IMP using the crystallization reagent consisting of 2M ammonium sulfate and 5% (v/v) 2-proponal. A similar methodology was used to grow crystals of Q523X in presence of ATP, Mg²⁺, and IMP. No electron density was detected for ATP, Mg²⁺, or IMP in either structure of D407A-537X or 537X. All crystals were cryoprotected by supplementing the crystallization solution with 20% (v/v) glycerol and flash-freezing in liquid nitrogen for data collection at 100 K. Crystals of R238W-537X, D407A-537X, and 537X diffracted X-rays to 1.97 Å. 1.98 Å, and 2.05 Å resolution, respectively. All crystals of the active form of NT5C2–537X, wild-type and mutant, belong to space group I222 with one protomer in ASU.

B: Crystallization of active form of three full-length proteins: WT, L375F, and R39Q

The full length WT (WT) enzyme in presence of ATP, Mg²⁺, and IMP was crystallized using seeds from crystals of the full length L375F (see below) and the crystallization reagent comprising 0.1 M HEPES (pH 7.5), and 20% (w/v) PEG 1000, and 0.1 M ammonium nitrate. The full length L375F (L375F) was crystallized in presence of ATP, Mg²⁺, and IMP, and by using a crystallization reagent consisting of 2M ammonium sulfate and 5% (v/v) 2-proponal. The full length R39Q (R39Q) was crystallized in presence of ATP, Mg²⁺, IMP, and by using the crystallization reagent consisting of 100 mM HEPES (pH 7.5) and 12% (w/v) PEG 3350. While crystals of WT were cryoprotected by supplementing the crystallization solution with 20% (v/v) glycerol, those of L375F and R39Q were cryoprotected with 20% (v/v) ethylene glycol and all the crystals were flash-frozen in liquid nitrogen for data collection at 100 K.

Whereas neither ATP, Mg²⁺ion nor IMP was observed in the structure of WT, there were well-defined electron densities for both ATP and Mg²⁺ion in the structures of L375F and R39Q. Crystals of the full length active forms of NT5C2, WT, L375F, and R39Q, diffracted X-ray to 2.15 Å. 2.10 Å, and 3.06 Å resolution, respectively.

All nine structures of the active form of NT5C2 belong to space group I222 with one protomer in ASU. The data quality and refinement statistics are shown in Supplementary Tables 2.

QUANTIFICATION AND STATISTICAL ANALYSIS

We performed statistical analysis by Student’s t-test. We considered results with P < 0.05 as statistically significant.

DATA AND SOFTWARE AVAILABILITY

The atomic coordinates reported in this paper can be found under PDB accession numbers: 6DDC, 6DDB, 6DDL, 6DDO, 6DDK, 6DDQ, 6DDX, 6DDY, 6DDH, 6DDZ, 6DD3, 6DE0, 6DE1, 6DE2 and 6DE3.

Supplementary Material

2

Table S3. Related to Figures 2. RMSD values.

RMSD values in angstrom (Å) versus length of amino acids aligned between two structures using DALI pairwise server (Holm and Laakso, 2016), green, <0.5 Å; blue, 0.5–1.5 Å; red, >1.5 Å.

Published Structures: The Protein Data Bank (PDB) accession codes.

Table S5. Related to Figure 5. Mutation analysis of CRISPR/Cas9 screen for gain of function alleles in the arm region of NT5C2.

Significance.

Gain of function NT5C2 mutations are highly prevalent in high risk early relapse leukemia making NT5C2 the most prevalent therapeutic target in relapsed acute lymphoblastic leukemia. Here we identify distinct mechanisms of NT5C2 regulation targeted by relapse leukemia-associated NT5C2 mutations. These results support a critical role of negative regulators of allosteric activation in the control of NT5C2 activity pointing to the allosteric effector site as a potential therapeutic target for the development of NT5C2 inhibitors.

Highlights.

• NT5C2 mutations drive resistance to 6-mercaptopurine by different mechanisms.

• Class I NT5C2 mutations lock the activating helix A in an active configuration.

• Class II NT5C2 mutations disrupt a built in switch-off regulatory mechanism.

• Class III NT5C2 mutations lack a C-terminal brake of activation.