CFTR, the chloride channel mutated in cystic fibrosis (CF) patients, is opened by ATP binding to two cytosolic nucleotide binding domains (NBDs), but pore-domain mutations may also impair gating. ATP-bound NBDs dimerize occluding two nucleotides at interfacial binding sites; one site hydrolyzes ATP, the other is inactive. The pore opens upon tightening, and closes upon disengagement, of the catalytic site following ATP hydrolysis. Extent, timing, and role of non-catalytic-site movements are unknown. Here we exploit equilibrium gating of a hydrolysis-deficient mutant and apply Φ value analysis to compare timing of opening-associated movements at multiple locations, from the cytoplasmic ATP sites to the extracellular surface. Marked asynchrony of motion in the two ATP sites reveals their distinct roles in channel gating. The results clarify the molecular mechanisms of functional cross-talk between canonical and degenerate ATP sites in asymmetric ABC proteins, and of the gating defects caused by two common CF mutations.https://doi.org/10.7554/eLife.29013.001
Cystic Fibrosis (CF) is an incurable, devastating inherited disease caused by mutations in the CF Transmembrane Conductance regulator (CFTR) chloride ion channel (Riordan et al., 1989). A large number of mutations identified in CF patients scatter throughout the entire protein sequence and cause defects in protein synthesis, maturation, stability, channel gating, and/or anion permeation through the open channel pore (O'Sullivan and Freedman, 2009). In recent clinical trials small molecule drugs that target the CFTR protein and rectify either processing (‘correctors’) or gating defects (‘potentiators’) have proven beneficial for some CF patients carrying specific mutations (Ramsey et al., 2011), underscoring the importance of understanding the structure and mechanism of this disease-associated protein at a molecular level.
CFTR belongs to the large family of ATP-Binding Cassette (ABC) proteins (Riordan et al., 1989), most of which function as active transporters (Dean and Annilo, 2005). A conserved molecular mechanism that involves ATP binding and hydrolysis drives opening and closing (gating) of the CFTR pore and substrate translocation through other ABC proteins (Li et al., 1996). The exceptional resolution of single-channel patch-clamp recordings makes CFTR a model ABC protein eminently suitable for molecular-level biophysical studies.
ABC proteins are built from two homologous halves each containing a transmembrane domain (TMD) and a cytosolic nucleotide binding domain (NBD). In CFTR these two canonical halves are linked by the unique regulatory (R) domain which must be either phosphorylated by cyclic AMP-dependent protein kinase (PKA) (Anderson et al., 1991), or deleted (Figure 1A) (Csanády et al., 2000) to allow channel gating. The highly conserved NBD fold comprises an alpha/beta ‘head’ subdomain harboring the Walker A and B motifs, and an alpha-helical ‘tail’ subdomain which contains the ABC-specific ‘signature’ sequence (Hung et al., 1998). Binding of ATP to the Walker motifs in each of the two NBDs promotes their association into a tight head-to-tail dimer that occludes the two bound ATP molecules at interfacial binding sites formed by the Walker motifs of one NBD and the signature motif of the other (Smith et al., 2002). Such NBD dimers are extremely stable but dissociate following ATP hydrolysis (Moody et al., 2002; Smith et al., 2002). Formation/disruption of the NBD dimer, communicated to the TMDs through an interface formed by four short coupling helices (CH1-4, Figure 1A, violet loops), drives the major conformational rearrangements of the substrate translocation pathway (Locher, 2009). Upon formation of the NBD dimer, the CFTR anion pore opens to a ‘burst’ state interrupted by brief (~10 ms) ‘flickery’ closures, and upon dimer disruption following ATP hydrolysis at the consensus site (‘catalytic site 2’, formed by NBD2 Walker motifs and NBD1 signature sequence) the channel returns to a long-lived (~1 s) ‘interburst’ closed state (Figure 1B; [Vergani et al., 2005]). The closed-pore CFTR structure is inward facing (Zhang and Chen, 2016; Liu et al., 2017), whereas the open-pore structure, with dimerized NBDs, likely resembles the outward-facing TMD conformation of ABC proteins. In a subset of ABC transporters, including CFTR, only one of the two composite sites is catalytically active, whereas the other contains non-canonical substitutions at several key residues that impair ATPase activity. In CFTR that degenerate ATP site (‘non-catalytic site 1’, formed by NBD1 Walker motifs and NBD2 signature sequence), keeps ATP bound but unhydrolyzed throughout several gating cycles (Aleksandrov et al., 2002; Basso et al., 2003). The precise range of gating-related movements in non-catalytic site 1 is unclear (Tsai et al., 2010b; Szollosi et al., 2011; Chaves and Gadsby, 2015), but mutations here impair ATP hydrolysis at the catalytic site 2 (Ramjeesingh et al., 1999), affect channel gating (Tsai et al., 2010a; Csanády et al., 2013), and may cause CF (e.g., G1349D).
Resolving the relative timing of motions in various protein regions allows the molecular forces that drive protein conformational rearrangements to be understood. In particular, the structural organization of the transition (T) state for channel opening may shed light on the sources of molecular strain that underly the large Gibbs energy of this highest-energy intermediate conformation, which rate limits the overall transition (Csanády et al., 2006). Such T-state structures cannot be studied by standard structural approaches, as their life times fall into the sub-microsecond range (Chakrapani and Auerbach, 2005). However, relative timing of motions of a given channel position is reported by its Φ value, the slope of a log-log plot of the opening rate constant (kCO) as a function of the closed-open equilibrium constant (Keq = kCO/kOC) for a series of point substitutions (Brønsted plot). For the opening step of a simple idealized two-state channel with a single intervening T state, Φ is a measure of how far the conformation of the position has progressed along the reaction coordinate in the T state (0≤Φ ≤1). For a position that moves very early, and has therefore reached its open-like conformation in the T state (Φ = 1), perturbations affect T-state and open-state stabilities to similar extents, and thus impact only opening but not closing rate. At the other extreme, for a position that moves very late, and is still in its closed-like conformation in the T-state (Φ = 0), perturbations affect closing but not opening rate (Auerbach, 2007). In reality, the overall closed-to-open conformational change of large ion-channel proteins is best described by a sequence of conformational steps across a chain of high-energy intermediary states (the ‘transition-state ensemble’), and the Φ value of a position reports whether its perturbations affect early (large Φ) or late (small Φ) steps within that chain of events. However, on the plausible assumption that perturbation of a position affects the microscopic step during which the position undergoes a rearrangement, but not steps during which it remains static, the Φ value is reasonably interpreted to reflect relative timing of movements even in such cases (Zhou et al., 2005). The terms ‘early’ and ‘late’ will be used here to reflect sequentiality of movements based on this interpretation.
Because Φ value analysis requires equilibrium (Csanády, 2009), it can be applied to CFTR pore opening only in the presence of a background mutation (Figure 1A–B, red star) that disrupts ATP hydrolysis at catalytic site 2 (Figure 1B, red cross), thus reducing gating in saturating ATP to a simple closed-open equilibrium (C1↔O1; Figure 1B, red box). These conditions are satisfied in a background construct (Sorum et al., 2015) in which mutation of the NBD2 Walker B aspartate disrupts ATP hydrolysis and removal of the R domain obviates the need for prior phosphorylation (cut-ΔR(D1370N); Figure 1A). In this background, Φ value analysis detected a clear temporal gradient in opening-related movements which spreads along the longitudinal, cytoplasmic to extracellular, protein axis from catalytic site 2 towards the pore (Figure 1C): in the T state catalytic site 2 is already dimerized (Φ ~ 1 for position 1246), the NBD-TMD interface is on the move (Φ ~ 0.5 for position 275), whereas the pore is still shut (Φ ~ 0.2 for position 348), suggesting strain in the region of the main CF locus (Sorum et al., 2015).
Here we exploit the same background construct to compare the roles of the two ATP binding sites in pore opening, and find markedly asynchronous movements. An outline of the transition-state structure based on Φ values of eleven positions at four different levels along the longitudinal protein axis, from the two ATP binding sites to the extracellular surface, suggests distinctly different energetic roles for the two sites in supporting channel gating. This provides the first mechanistic clues for understanding the division of tasks and functional cross-talk between the catalytic and the non-catalytic ATP binding site in asymmetric ABC proteins. The findings also clarify the molecular mechanisms of the CFTR channel gating defects caused by two common CF causing mutations.
Early movement within catalytic site 2 was suggested by the large Φ value for position 1246 in the Walker A motif of NBD2 (Sorum et al., 2015). To test timing of movements at the opposing NBD1 surface of catalytic site 2, we targeted position 555 (Figure 1A,C–D, red) near the NBD1 signature sequence: the R555 and T1246 side chains span the site 2 interface (Figure 1D, lower dotted ellipse) to form a hydrogen bond in open CFTR channels (Vergani et al., 2005). Three tested arbitrary substitutions (see Materials and methods) of the native arginine at position 555 of our background construct (glutamine, alanine, cysteine) all dramatically lowered channel activity (Figure 2A): open probability (Po) in saturating (10 mM) ATP was reduced by up to 10-fold (Figure 2D). This graded reduction in the closed-open equilibrium constant (Keq) was caused predominantly by mutational effects on mean interburst durations (τib) which were prolonged by up to 10-fold (Figure 2C), reporting large reductions in channel opening rate (kCO = 1/τib). Importantly, since 10 mM ATP remained saturating for all of the mutants (Figure 2—figure supplement 1B, red bars), this reduced rate kCO indeed reflects slowing of the C1→O1 step (Figure 1B). In contrast, mean open burst durations (τb) were shortened by less than twofold (Figure 2B), reporting only modest effects of the mutations on channel closing rate (kOC = 1/τb). Correspondingly, the slope of the Brønsted plot (Figure 2E) revealed a high value of Φ = 0.84 ± 0.04 for position 555, similar to that of position 1246 (Φ = 0.97 ± 0.19; [Sorum et al., 2015]), indicating very early movement of both sides of catalytic site 2 during pore opening.
To further expand the longitudinal axial distance spanned by our target positions, we also studied position 117 (Figure 1A,C, blue) in the first extracellular loop, where mutations affect gating and are associated with CF (Sheppard et al., 1993; Yu et al., 2016). Beside a modest reduction in unitary conductance due to the removal of the positive charge of the native arginine, substitution of glutamine, alanine, histidine, or cysteine into position 117 all robustly reduced Po (Figure 2F), by up to ~30 fold (Figure 2I). However, compared to the position 555 mutants, the kinetic pattern of gating of the position 117 mutants was strikingly different (cf., Figure 2F and A): the reduced Po in this case reflected a parallel reduction in mean open burst durations, by >50 fold for the histidine and cysteine substitutions (Figure 2G), whereas mean interburst durations remained largely unaffected (Figure 2H). Thus, perturbations at position 117 selectively affect channel closing, but not opening, rate, yielding a Brønsted plot with a slope of essentially zero (Figure 2J). The extremely low value of Φ = 0.05 ± 0.02 for position 117 implies very late movement of this extracellular position during pore opening, indicating that the conformational wave that begins in catalytic site 2 ends at the extracellular surface.
Because accurate estimation of opening rate requires correct estimation of the number of active channels in the patch, which is difficult when the Po is small, even with very long recordings, for low-Po mutants like those in Figure 2A and F channel numbers were estimated at the end of each experiment by exposure to 2’-deoxy-ATP (dATP) or 2'-deoxy-N6-(2-phenylethyl)-ATP (P-dATP) which markedly stimulated channel open probability (Figure 2—figure supplement 2). Of note, for all mutants studied here the reconstructed interburst ↔ burst sequences obtained in saturating ATP (cf., Figure 2—figure supplement 1) were reasonably described by a two-step process, as confirmed by dwell-time analysis (see example histograms for 117 and 555 position mutants in Figure 2—figure supplement 3 and Figure 2—figure supplement 4).
To address timing of movements in non-catalytic site 1, we first targeted positon 1348 in the NBD2 signature sequence (Figure 1A,C–D, orange). Replacement of the native histidine at position 1348 with an alanine increased Po, whereas substitutions by methionine, tyrosine, or glutamate decreased it (Figure 3A,D). In both cases however, changes in Po reflected simultaneous changes in both mean burst and interburst durations (Figure 3B,C; cf., histograms in Figure 3—figure supplement 1): that is, shortened bursts typically coupled with prolonged interbursts (H1348E/Y; Figure 3A–C), or prolonged bursts coupled with shortened interbursts (H1348A; Figure 3A–C). Thus, perturbations at position 1348 alter both opening and closing rates, yielding a Brønsted plot with an intermediate slope (Figure 3E): the value of Φ = 0.42 ± 0.04 stands in stark contrast to Φ ~ 1 for catalytic site 2, and indicates delayed motion at position 1348, relative to site 2.
To target the opposing, NBD1 surface of non-catalytic site 1, we chose position 460 in the NBD1 Walker A motif, corresponding to position 1246 in ATP site 2 (Figure 1A,C–D, orange). Positions 460 and 1348 flank non-catalytic site 1 (Figure 1D, upper dotted ellipse) from two sides. Relative to the native threonine at positions 460, the alanine, serine, and proline substitutions markedly reduced Po (Figure 3F,I). As with position 1348, these changes in Po again reflected coupled opposing changes in mean burst and interburst durations (Figure 3G,H; cf., histograms in Figure 3—figure supplement 2). Correspondingly, the Brønsted plot for position 460 (Figure 3J) revealed a low-intermediate value of Φ = 0.36 ± 0.07, corroborating the conclusion from position 1348 that movement within non-catalytic site 1 during pore opening is markedly delayed compared to that within catalytic site 2. Importantly, although perturbations of the NBD1 Walker A motif might impair ATP binding at non-catalytic site 1 (Vergani et al., 2003), 10 mM ATP remained saturating for all of our site-1 mutants (Figure 2—figure supplement 1A; Figure 2—figure supplement 1B, orange bars).
The large Φ of positions in catalytic site 2 but low-to-intermediate Φ of positions in non-catalytic site 1 implies that timing of movements in the two ATP sites during pore opening is highly asymmetric: whereas catalytic site 2 has already adopted its open-like conformation in the T state, non-catalytic site 1 is still on the move. If the movements completed in non-catalytic site 1 between the T state and open state are large-scale rearrangements, then a similarly pronounced temporal asymmetry of motions might be detectable one structural level further along the axis, at the level of the four coupling helices (Figure 1C, violet helices), since CH1 and CH2 are located closer to non-catalytic site 1, whereas CH3 and CH4 closer to catalytic site 2 (Figure 1D, violet helices).
To test for such possible asynchrony, we targeted a position in each of the four coupling helices: 172 in CH1, 275 in CH2 (studied earlier; [Sorum et al., 2015]), 961 in CH3, and 1068 in CH4 (Figure 1A,C–D, purple). At positions 172, 961, and 1068 (the native residues are leucine, methionine, and phenylalanine, respectively) substitutions caused both increases and decreases in open probability (Figure 4D,I,N), and the underlying changes in gating pattern were similar (Figure 4A,F,K): a clear trend for an association of lengthened bursts with shortened interbursts and of shortened bursts with lengthened interbursts could be observed (compare Figs. B-C, G-H, L-M; cf., histograms in Figure 4—figure supplements 1–3). Thus, perturbations at positions 172, 961, and 1068 affect opening and closing rates in opposite directions but to comparable extents, yielding Brønsted plots with intermediate slopes in each case (Figure 4E,J,O), indicating that in the T state the coupling helices have already left their closed-like, but have not yet reached their final open-like, positions. On the other hand, the obtained values of Φ = 0.51 ± 0.07 for position 172, Φ = 0.60 ± 0.12 for position 961, and Φ = 0.54 ± 0.07 for position 1068 are quite similar to each other and to Φ = 0.50 ± 0.13 for position 275 (Sorum et al., 2015).
Position 508 (Figure 1A,C–D, deep purple), the locus of the most common (>70%) CF mutation, is found in a surface cleft of NBD1 which accommodates CH4 to form a joint-like interface in which the F508 side chain interacts with aromatic CH4 residues, including F1068 (Zhang and Chen, 2016). Deletion of phenylalanine 508 is detrimental for folding and stability of the CFTR protein (Du et al., 2005; Okiyoneda et al., 2010), but also severely impairs gating (Miki et al., 2010; Kopeikin et al., 2014). To address timing of motions in this disease hotspot-position, we replaced the native phenylalanine with cysteine, leucine, and serine. All substitutions markedly reduced open probability (Figure 5A,D), supporting exquisite sensitivity of this region to structural perturbations. The reduction in the closed-open equilibrium constant was caused in each case by a simultaneous shortening of mean burst and lengthening of mean interburst durations, which were comparably affected (Figure 5B–C; cf., histograms in Figure 5—figure supplement 1). The resulting Φ = 0.55 ± 0.02 (Figure 5E) is indistinguishable from that of position 1068, and indicates that position 508 is also just on the move in the T state for opening. Again, importantly, the reduced opening rate of the position 508 mutants was not due to impaired ATP binding, as 10 mM ATP remained saturating (Figure 2—figure supplement 1B, deep purple bars). Channels could be efficiently counted at the end of each experiment by exposure to dATP or P-dATP (Figure 2—figure supplement 2).
The closed-to-open transition of the CFTR channel is a major structural rearrangement which involves dimerization of the NBDs and flipping of the TMDs from an inward- to an outward-facing conformation ([Vergani et al., 2005], cf., homology models for phosphorylated closed and open CFTR channels in Figure 6A, left-to-right). Whereas X-ray or cryo-EM structures may capture stable closed (Zhang and Chen, 2016; Liu et al., 2017) and open ground states, the structure of the transition state traversed during opening, and the relative timing of motions in various protein regions may presently be inferred only from Φ value analysis (Auerbach, 2007). Previous work revealed a Φ value gradient along the longitudinal protein axis, suggesting that the opening conformational change spreads from catalytic site 2 towards the pore (Sorum et al., 2015). By expanding the axial distance spanned by our target positions here we show a gradient of Φ values from ~1 for both sides of catalytic ATP site 2, all the way to ~0 for the extracellular protein surface (Figure 6A). These findings corroborate the earlier proposal that in the T state catalytic site 2 is already dimerized but the pore is still in its closed-like (inward-facing) conformation.
Phenylalanine 508, the locus of the most common CF mutation, is located at the NBD-TMD interface, that is, in a region expected to experience strain in the T state. Here we show that position 508 is indeed just on the move in the T state, together with spatially adjacent CH4 (Zhang and Chen, 2016; Liu et al., 2017), as reported by their intermediate, essentially indistinguishable Φ values of ~0.5 (Figures 4O and 5E). Because all tested perturbations of the F508 side chain reduce opening and accelerate non-hydrolytic closing rate (Figure 5B–C), the simplest interpretation is that the native phenylalanine side chain is involved in stabilizing interactions that become stronger as the closed channel reaches the T state, and even stronger when it relaxes to the open state. Loss (or weakening) of these interactions by 508-position perturbations destabilizes the T state relative to the closed state (reducing opening rate, Figure 5C; [Miki et al., 2010]), and further destabilizes the open state relative to the T state (enhancing non-hydrolytic closing rate, Figure 5B; [Jih et al., 2011]). The alternative explanation that the native phenylalanine side chain forms destabilizing interaction(s) in the closed state seems unlikely, as mutations here dramatically reduce protein stability, suggesting an overall stabilizing role for residue F508 in its native environment. Of note, because in wild-type (WT) CFTR the hydrolytic pathway for channel closure is much faster than the non-hydrolytic pathway, the strong acceleration of non-hydrolytic closing rate by deletion or substitutions of residue F508 is not apparent when studied in a WT (hydrolytic) background (Miki et al., 2010; Kopeikin et al., 2014), a fact that has led to previous misassignment of a Φ value close to 1 for position 508 (Aleksandrov et al., 2009).
Mutation R117H is associated with a mild form of CF and causes, beside a moderate reduction in single channel chloride conductance, a strong channel gating defect (Sheppard et al., 1993; Yu et al., 2016), suggesting that this position is involved in interactions that change in a gating-state dependent manner. Indeed, all substitutions at position 117 dramatically reduced open probability (Figure 2I) by accelerating closure (Figure 2G) while leaving opening unaffected (Figure 2H). Because in closed-state structures of CFTR the native R117 side chain is not seen to form strong interactions with other residues (Zhang and Chen, 2016; Liu et al., 2017), the implication is that it forms a stabilizing interaction (or interactions) in the open state, that is (are) lost when the arginine at position 117 is replaced. Furthermore, the stabilizing interaction is not yet formed in the T state: its appearance is among the last movements associated with pore opening (Φ ~ 0). Thus, R117 substitutions selectively destabilize the open state, but not the transition state, relative to the closed state, as indicated by the unaltered opening rate.
In striking contrast to the large Φ values of positions on both sides of the dimer interface in catalytic site 2, positions flanking non-catalytic site 1 display low-to-intermediate Φ values (Figure 6A, orange vs. red residues (left) and numbers (right)). Thus, unlike catalytic site 2 which has already adopted its open-like conformation (Figure 6A, right, red residues) in the T state, non-catalytic site 1 is still on the move between its closed- and open-like conformations (Figure 6A, left and right, orange residues). Interestingly, such pronounced asymmetry in the timing of motions at one end of the dimer interface compared to the other cannot be detected at the level of the four coupling helices, which are all characterized by Φ values of 0.5–0.6 (Figure 6A, violet residues (left) and numbers (right)). Thus, the movements that take place in non-catalytic site 1 between the T state and the open state likely remain confined to the site-1 interface.
ATP bound at the two composite sites acts as molecular glue that stabilizes the open-pore conformation by bonding the NBD interfaces together (Moody et al., 2002; Smith et al., 2002). Our data suggest a division of labor in the bonding activities of the two sites. Phosphorylated WT CFTR channels open infrequently even in the absence of ATP (Figure 6B, top current trace), and such ‘spontaneous’ openings reflect occasional dimerization of empty, unliganded NBDs (Mihályi et al., 2016) (Figure 6B, top cartoon). Comparison of such spontaneous gating of WT channels with the equilibrium gating of a hydrolysis-deficient mutant, such as the NBD2 Walker A lysine mutant K1250R, in saturating ATP (Figure 6B, bottom current trace) reveals two robust effects of bound ATP: a shortening of closed interburst durations (Figure 6B, compare green bars above current traces) and a lengthening of open burst durations (Figure 6B, compare red bars below current traces), both by >100 fold (Mihályi et al., 2016). Thus, bound ATP reduces the energetic barrier for opening, but increases the barrier for non-hydrolytic closure (Figure 6C, black vs. orange standard Gibbs energy profiles).
Which of the two bound ATP molecules affects opening rate? Our Φ value analysis indicates that in the T state catalytic site 2 has already finished moving, that is, the site-2 glue is already bonded (Figure 6B, bottom cartoon, lower site; red semi-circles in states T and O1): that bonding readily explains the reduction in the energetic barrier for opening, that is, stimulation of opening rate, by ATP. Whether stably bound ATP in non-catalytic site 1 might also contribute to speeding opening, by maintaining some contact across the site-1 interface even in the closed state and hence preventing complete NBD disengagement, is still unsettled (Tsai et al., 2010b; Szollosi et al., 2011), but, cf. Chaves and Gadsby, 2015).
Which of the two bound ATP molecules affects non-hydrolytic closing rate? Because it is bonded both in the T and O1 states, the site-2 glue cannot affect the height of the barrier for non-hydrolytic closure. On the other hand, the low Φ value of non-catalytic site 1 indicates localized movements between the T state and the open state, which we propose reflects tight bonding of the site-1 glue (i.e., tightening of the site-1 interface; Figure 6B, bottom cartoon, upper site; red semi-circle only in state O1). By selectively stabilizing the open state relative to the T state, the site-1 glue increases the height of the barrier for non-hydrolytic closure, explaining the observed prolongation of non-hydrolytic bursts by ATP.
In the context of the functional cycle of WT CFTR (Figure 1B) ATP bound at catalytic site 2 promotes efficient opening (speeds step C1→O1) whereas ATP bound at non-catalytic site 1, by slowing rate O1→C1, ensures progress through ATP hydrolysis: such unidirectional cycling is essential for ABC proteins to mediate uphill transport. In asymmetric ABC proteins flipping to the outward-facing, NBD-dimerized conformation is followed by a step in which the consensus site becomes committed to ATP hydrolysis (Timachi et al., 2017): that step might involve bonding of the glue in the non-catalytic site. This interpretation is consistent with the functional cross-talk between the two sites observed in CFTR or related MRP1 (Ramjeesingh et al., 1999; Gao et al., 2000; Hou et al., 2002): perturbations in the non-catalytic site of CFTR reduce the ATPase turnover rate of the catalytic site (Ramjeesingh et al., 1999) by lowering the fraction of open bursts that are terminated by ATP hydrolysis as opposed to non-hydrolytic dissociation of the NBD dimer (Csanády et al., 2010).
The hydrolysis-disrupting D1370N mutation, an essential component of our background construct, precludes targeting position 1370 for Φ value analysis. However, the non-hydrolytic closing rate of D1370N CFTR channels is substantially faster than that of other non-hydrolytic catalytic site-2 mutants, suggesting that this mutation affects not only opening, but also non-hydrolytic closing rate (Gunderson and Kopito, 1995; Vergani et al., 2003): the implication is that the Φ value of position 1370 might be lower than those of the catalytic site-2 positions studied here. Although functionally involved in Mg2+ coordination in catalytic site 2 (cf., Zaitseva et al., 2005), aspartate 1370 is physically located in between the two ATP sites, in the center of the NBD interface, at the boundary between core and α-helical subdomains of NBD2 (Figure 6—figure supplement 1, salmon). Moreover, it is in contact with the conserved Q-loop glutamine Q1291 (Figure 6—figure supplement 1, yellow) (Liu et al., 2017) which acts as a γ-phosphate sensor to induce an ~15o rotation of the α-helical subdomain towards the core subdomain upon ATP binding (Karpowich et al., 2001). Thus, residue D1370 might also be involved in subdomain closure, essential for stable NBD dimerization (Smith et al., 2002). Whether in CFTR that subdomain closure already happens in closed channels upon ATP binding (state C1, Figure 1B), or is completed only in the tight-dimer open-channel state as in the maltose transporter (Orelle et al., 2010), might be clarified by establishing a different non-hydrolytic background construct which allows targeting positions at the intra-NBD subdomain interface for Φ value analysis.
In conclusion, the Φ value map obtained here supports distinct roles for the two composite ATP sites in CFTR channel gating and reveal the molecular mechanism of the gating defects caused by two common CF mutations. Whereas this work reports relative timing of movements in the two ATP sites during pore opening, clarifying the extent of these movements during each step in the channel opening/closing cycle will require comparisons among high-resolution structures of CFTR trapped in phosphorylated closed- and open-channel states. The recent structures of unphosphorylated closed (Liu et al., 2017) and phosphorylated occluded (Zhang et al., 2017) CFTR have just started to set the stage for such comparisons.
Mutations at positions 117, 172, 460, 508, and 555 were introduced into pGEMHE-CFTR(1-633) (Csanády et al., 2000), and those at positions 961, 1068, and 1348 into pGEMHE-CFTR(837-1480(D1370N)) (Sorum et al., 2015) using Stratagene QuikChange (Agilent Technologies, Santa Clara, CA, USA). cDNA was transcribed in vitro using T7 polymerase. Particular substitutions were chosen based on the following general considerations: (i) mutations that are too conservative might not cause a phenotype (and hence result in a dot on the REFER plot which lies very close to that of the WT), whereas (ii) mutations that are too drastic might disrupt functional expression. Because predicting the impact of a particular substitution is often difficult, for each position typically 5–6 point substitutions were chosen in a quasi-arbitrary manner, ranging from conservative mutations to substitutions that markedly alter side chain size, electrostatic charge, chemical nature, or backbone conformation (proline). All mutants that gave rise to functional channels were characterized.
Ovarian lobes were removed from anaesthetized Xenopus laevis [RRID:NXR_0.0080] following a IACUC-approved protocol. Oocytes were defolliculated by treatment with Type II collagenase (Sigma-Aldrich, Hungary) and stored at 18°C in a frog Ringer's solution supplemented with 1.8 mM CaCl2 plus 50 μg/ml gentamycin sulfate (Sigma). Recordings were done 1–3 days after co-injection with 0.1–10 ng cRNA for both CFTR segments.
CFTR unitary currents were recorded in excised inside-out patches at 25°C as described earlier (Sorum et al., 2015). Pipette solution contained (in mM) 136 NMDG-Cl, 2 MgCl2, 5 HEPES, pH = 7.4 with NMDG, bath solution contained 134 NMDG-Cl, 2 MgCl2, 5 HEPES, 0.5 EGTA, pH = 7.1 with NMDG. The bath solution was continuously flowing, and could be exchanged with a time constant of ~100 ms. MgATP (3 or 10 mM) and 2'-deoxy-ATP sodium salt (dATP, 5 mM) (Sigma) were added from 400 mM and 100 mM aqueous stock solutions, respectively (pH = 7.1 with NMDG); dATP was supplemented with equimolar MgCl2. 25–50 μM 2'-deoxy-N6-(2-phenylethyl)-ATP sodium salt (P-dATP; Biolog LSI, Germany) was added from a 10 mM aqueous stock solution. Unitary CFTR currents in 10 mM MgATP were recorded at −80 mV (−100 mV for position 117 mutants) (Axopatch 200B, Molecular Devices, Sunnyvale, CA, USA) digitized at 1 kHz (Digidata 1322A, Pclamp9 [RRID:SCR_011323], Molecular Devices), and lowpass-filtered at 50 Hz using a Gaussian filter. Single-channel patches were identified as very long (typically 15 min - 1 hr) recordings without superimposed channel openings. For the low-Po mutants targeting positions 117, 508, and 555 strong stimulation by dATP or P-dATP at the end of each experiment was used to facilitate correct estimation of the number of active channels in the patch (Figure 2—figure supplement 2). Macroscopic or quasi-macroscopic current ratios between 3 and 10 mM ATP were used to verify saturation by 10 mM ATP for each of the tested mutants (Figure 2—figure supplement 1). For display purposes figure panels show channel currents filtered at 20 Hz (50 Hz for R117 mutants).
Bursts and interbursts were reconstructed as described earlier (Sorum et al., 2015). In brief, currents from long segments of recording without superimposed channel openings were filtered at 50 Hz, idealized by half-amplitude threshold crossing, and brief closures suppressed using the method of (Magleby and Pallotta, 1983). Opening (kCO) and closing (kOC) rates were defined as the inverses of the mean interburst (τib) and burst (τb) durations, respectively, and Keq as kCO/kOC. Plots of ln(kCO) as a function of ln(Keq) (Brønsted plots) were fitted by linear regression (SigmaPlot 8 [RRID:SCR_003210]).
All data are given as mean ± SEM of measurements from ≥3 (typically 5–7, as indicated in each figure legend) long segments of single-channel recordings, from 3 to 15 patches for each mutant. In the face of alternating periods of lower and higher activity typical to CFTR (Bompadre et al., 2005), several hours of total recording for each construct were obtained to ensure unbiased sampling of average gating behaviour, and all data were included in the analysis.
Prolonged nonhydrolytic interaction of nucleotide with CFTR's NH2-terminal nucleotide binding domain and its role in channel gatingThe Journal of General Physiology 122:333–348.https://doi.org/10.1085/jgp.200308798
CFTR gating I: Characterization of the ATP-dependent gating of a phosphorylation-independent CFTR channel (DeltaR-CFTR)The Journal of general physiology 125:361–375.https://doi.org/10.1085/jgp.200409227
Cysteine accessibility probes timing and extent of NBD separation along the dimer interface in gating CFTR channelsThe Journal of General Physiology 145:261–283.https://doi.org/10.1085/jgp.201411347
Cystic fibrosis transmembrane conductance regulator (cftr): closed and open state channel modelsThe Journal of Biological Chemistry 290:22891–22906.https://doi.org/10.1074/jbc.M115.665125
Severed channels probe regulation of gating of cystic fibrosis transmembrane conductance regulator by its cytoplasmic domainsThe Journal of General Physiology 116:477–500.https://doi.org/10.1085/jgp.116.3.477
Conformational changes in the catalytically inactive nucleotide-binding site of CFTRThe Journal of General Physiology 142:61–73.https://doi.org/10.1085/jgp.201210954
Thermodynamics of CFTR channel gating: a spreading conformational change initiates an irreversible gating cycleThe Journal of General Physiology 128:523–533.https://doi.org/10.1085/jgp.200609558
Application of rate-equilibrium free energy relationship analysis to nonequilibrium ion channel gating mechanismsThe Journal of General Physiology 134:129–136.https://doi.org/10.1085/jgp.200910268
Evolution of the ATP-binding cassette (ABC) transporter superfamily in vertebratesAnnual Review of Genomics and Human Genetics 6:123–142.https://doi.org/10.1146/annurev.genom.6.080604.162122
The DeltaF508 cystic fibrosis mutation impairs domain-domain interactions and arrests post-translational folding of CFTRNature Structural & Molecular Biology 12:17–25.https://doi.org/10.1038/nsmb882
Comparison of the functional characteristics of the nucleotide binding domains of multidrug resistance protein 1Journal of Biological Chemistry 275:13098–13108.https://doi.org/10.1074/jbc.275.17.13098
Combined effects of VX-770 and VX-809 on several functional abnormalities of F508del-CFTR channelsJournal of Cystic Fibrosis 13:508–514.https://doi.org/10.1016/j.jcf.2014.04.003
ATPase activity of the cystic fibrosis transmembrane conductance regulatorJournal of Biological Chemistry 271:28463–28468.https://doi.org/10.1074/jbc.271.45.28463
Review. Structure and mechanism of ATP-binding cassette transportersPhilosophical Transactions of the Royal Society B: Biological Sciences 364:239–245.https://doi.org/10.1098/rstb.2008.0125
Burst kinetics of single calcium-activated potassium channels in cultured rat muscleThe Journal of Physiology 344:605–623.https://doi.org/10.1113/jphysiol.1983.sp014958
Potentiation of disease-associated cystic fibrosis transmembrane conductance regulator mutants by hydrolyzable ATP analogsJournal of Biological Chemistry 285:19967–19975.https://doi.org/10.1074/jbc.M109.092684
Cooperative, ATP-dependent association of the nucleotide binding cassettes during the catalytic cycle of ATP-binding cassette transportersJournal of Biological Chemistry 277:21111–21114.https://doi.org/10.1074/jbc.C200228200
A CFTR potentiator in patients with cystic fibrosis and the G551D mutationNew England Journal of Medicine 365:1663–1672.https://doi.org/10.1056/NEJMoa1105185
Mutant cycles at CFTR's non-canonical ATP-binding site support little interface separation during gatingThe Journal of General Physiology 137:549–562.https://doi.org/10.1085/jgp.201110608
Stable ATP binding mediated by a partial NBD dimer of the CFTR chloride channelThe Journal of General Physiology 135:399–414.https://doi.org/10.1085/jgp.201010399
On the mechanism of MgATP-dependent gating of CFTR Cl- channelsThe Journal of General Physiology 121:17–36.https://doi.org/10.1085/jgp.20028673
On the mechanism of gating defects caused by the R117H mutation in cystic fibrosis transmembrane conductance regulatorThe Journal of Physiology 594:3227–3244.https://doi.org/10.1113/JP271723
Kenton J SwartzReviewing Editor; National Institutes of Health, United States
In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.
Thank you for submitting your article "Asymmetry of movements in CFTR's two ATP sites during pore opening serves their distinct functions" for consideration by eLife. Your article has been favorably evaluated by Richard Aldrich (Senior Editor) and three reviewers, one of whom, Kenton J Swartz (Reviewer #1), is a member of our Board of Reviewing Editors. The following individual involved in review of your submission has agreed to reveal their identity: Roderick MacKinnon (Reviewer #3).
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
This interesting manuscript from the Csanady lab presents new results continuing work started in a previous paper analyzing the open/closed transition state of the CFTR chloride channel. This channel, related to the ABC transporter family, undergoes a highly complex gating process which requires channel phosphorylation in the R domain as well as ATP binding and hydrolysis at the NBDs to drive irreversible gating. Though much progress has occurred in understanding these processes, key questions remain open. Here the authors dissect the final gating step by using transition state analysis of a mutant that is claimed to introduce a simple, two-state open closed reaction for the opening of the channel. Such a two-state system is required for the analysis to be valid. The authors analyze the effects of mutations at a series of positions on the open/closed equilibrium that complement the previous set, focusing now on areas around the ATP1 binding site and the outside of the pore. They find that the timing of the opening wave reveals propagation from ATP2 to ATP1, which is roughly simultaneous with movement of the interfacial coupling helices. A position at the outside end of the pore moves only very late in the reaction. Notably, the timing differences between the ATP binding sites seem not to persist through the interfacial helices, which all seem to move nearly simultaneously. The work reported here is careful and thorough, and the single channel recordings are really nice. The presentation is also excellent, very clearly discussing a difficult topic. This kind of work is highly complementary to the spate of emerging structures and provides information that will be difficult to obtain using other methods. We do have several concerns that the authors should address before we are fully convinced by their arguments. The authors should be aware that the reviewers have somewhat divergent opinions about the most interesting new conclusions in this manuscript.
1) The single-step assumption. Critical to the argument presented here is that the equilibrium is a single step reaction, and that the transitions we observe are between two discrete states. However, our impression is that the gating reflects the binding of ATP (at least to site 2), dimerization of the NBDs, and the channel conformational change. From the collective Phi's measured in this and the previous paper, and their apparently smooth gradation from slopes of near zero to near 1, we could be convinced that this is a single reaction, but we think the burden is on the authors to substantiate this a bit more. How single exponential are the results? Are all the mutants equally well fit by a single tau each? The reported tau values are said to be from single exponential fits, but I think it would be worthwhile to establish this formally. Perhaps the authors could show the dwell time histograms (or at least examples) and fits in a supplemental figure.
2) We are unclear of the data in the second part of Figure 3 and its associated discussion (paragraph beginning "To target the opposing…"). Please explain with greater care what you have done. Specifically, we would rather not see pooled data sets (from 460 and 464). Since the most important result in this paper is the discovery of different Phi values for catalytic and non-catalytic sites, we would be more confident in seeing what the Phi values are for multiple positions within each site (the theory that Phi values are related to a wave of conformational change across the protein, we should expect to see multiple similar Phi values in a given region.) Further, we are puzzled by the minimal discussion of the complexity of the data in Figure 3G-3J. Because the different Phi values between the two sites is the main result of this paper, we need to be more convinced of the data.
3) We think the authors should be more circumspect in their interpretation about how Phi analysis relates to timing of movements. The analysis speaks to where in the transition mutants at specific sites cause a perturbation. It may be reasonable to infer from this that this result is because those regions undergo dynamic rearrangements at different times in the transition, but it would seem more appropriate to separate the specific conclusion from the inference.
4) We were somewhat confused by the cartoon in Figure 6C and its implications for the gating model. Our understanding is that ATP1 binds stably and rarely exchanges, where ATP2 binding and hydrolysis drives the reaction. In this light, isn't it surprising that the movements at ATP1 occur after those of ATP1? I guess the model in our head is that the NBDs are partially 'glued' by ATP1 and ready to close upon ATP2 binding. Is this incorrect? If not, what is happening to ATP1 that makes the movements around it happen later?https://doi.org/10.7554/eLife.29013.020
1) The single-step assumption. Critical to the argument presented here is that the equilibrium is a single step reaction, and that the transitions we observe are between two discrete states. However, our impression is that the gating reflects the binding of ATP (at least to site 2), dimerization of the NBDs, and the channel conformational change.
Because opening indeed requires binding of ATP to site 2 followed by the channel isomerization step, it is critical that all experiments be performed in saturating ATP to make the isomerization step rate limiting. We have verified that this condition is met for all of our mutants (Figure 2—figure supplement 1).
From the collective Phi's measured in this and the previous paper, and their apparently smooth gradation from slopes of near zero to near 1, we could be convinced that this is a single reaction, but we think the burden is on the authors to substantiate this a bit more. How single exponential are the results? Are all the mutants equally well fit by a single tau each? The reported tau values are said to be from single exponential fits, but I think it would be worthwhile to establish this formally. Perhaps the authors could show the dwell time histograms (or at least examples) and fits in a supplemental figure.
For our background construct we have demonstrated the validity of the single-step assumption in our earlier study (Sorum et al., 2015). In a series of new supplementary figures (Figure 2—figure supplement 3–4; Figure 3—figure supplement 1–2; Figure 4—figure supplement 1–3; Figure 5—figure supplement 1) we now illustrate the entire procedure of our burst analysis for two mutants per each studied position. For those positions at which different substitutions caused bi-directional changes in Po (1348, 172, 961, 1068) the chosen examples represent the mutants at the two extreme ends of the REFER plot.
Panels B (G) and C (H) show distributions of simple closed and open times, and the choice of the burst delimiter (tcrit) according to the algorithm of Magleby and Pallotta (1983). Panels D (I) and E (J) display the distributions of the durations of reconstructed bursts and interbursts, resulting from suppression of brief "flickery" closures, and reasonable fits of the distributions of reconstructed bursts by single-exponential probability density functions. We note, however, that the distributions of reconstructed interburst durations will never be truly single-exponential: first, the left tail of the true interburst distribution is truncated, and second, the right tail of the distribution of flickery closures will survive – this is particularly apparent for the high-Po mutants for which the two closed-time components overlap substantially (e.g., H1348A, L172E, M961E, F1068L). However, the choice of tcrit assures that the mean duration of reconstructed interburst events matches that of the true interburst events (and therefore the estimation of mean burst durations is also unbiased). We would like to point out that the values of τb and τib were not derived from single-exponential fits (the reconstructed interburst distributions are not single-exponential), but reflect simple arithmetic averages of burst and interburst dwell-time durations (see Materials and methods). All this is explained in the legend of the first histogram figure (Figure 2—figure supplement3).
2) We are unclear of the data in the second part of Figure 3 and its associated discussion (paragraph beginning "To target the opposing…"). Please explain with greater care what you have done.
Our original aim was to characterize Φ for position 464, which has been extensively studied by several groups in the past. However, unfortunately most of the mutants we generated failed to form functional channels, with the exception of K464C and K464R for which we did manage to obtain a few recordings. On the assumption that Φ values for nearby positions will be similar, we therefore sought to solidify the K464 data by combining them with data obtained from two mutants of nearby position 460 (the position analogous to 1246 in site 2). We acknowledge that this approach depends on the above assumption.
Specifically, we would rather not see pooled data sets (from 460 and 464). Since the most important result in this paper is the discovery of different Phi values for catalytic and non-catalytic sites, we would be more confident in seeing what the Phi values are for multiple positions within each site (the theory that Phi values are related to a wave of conformational change across the protein, we should expect to see multiple similar Phi values in a given region.)
Done. Because we were unable to obtain a reliable Φ value for position 464, we have omitted the 464 data altogether. Instead, we have generated and functionally characterized three additional 460-position mutants, and now provide a clean REFER plot for position 460. The obtained Φ value of 0.36 ± 0.07 for position 460 is only slightly different from the Φ obtained earlier for the pooled 460-464 data (0.26 ± 0.06), and is quite similar to that of position 1348 located at the opposing surface of site 1 (0.42 ± 0.04). Thus, our conclusions are now based on the Φ values of two positions within each of the two ATP sites.
Further, we are puzzled by the minimal discussion of the complexity of the data in Figure 3G-3J. Because the different Phi values between the two sites is the main result of this paper, we need to be more convinced of the data.
The complexity of the notation (each construct denoted by a pair of letters) was due to the pooled nature of the original data set – in the present version this problem has been eliminated.
3) We think the authors should be more circumspect in their interpretation about how Phi analysis relates to timing of movements. The analysis speaks to where in the transition mutants at specific sites cause a perturbation. It may be reasonable to infer from this that this result is because those regions undergo dynamic rearrangements at different times in the transition, but it would seem more appropriate to separate the specific conclusion from the inference.
We now clarify this point in the Introduction:
"For the opening step of a simple idealized two-state channel with a single intervening T state, Φ is a measure of how far the conformation of the position has progressed along the reaction coordinate in the T state (0≤Φ≤1). […] The terms "early" and "late" will be used here to reflect sequentiality of movements based on this interpretation."
4) We were somewhat confused by the cartoon in Figure 6C and its implications for the gating model. Our understanding is that ATP1 binds stably and rarely exchanges, where ATP2 binding and hydrolysis drives the reaction. In this light, isn't it surprising that the movements at ATP1 occur after those of ATP1?
We do not think that there is anything surprising about this finding. It is clear that opening is initiated by ATP binding to site 2, because channels immediately cease to reopen once ATP is removed from the bath solution, even though site 1 likely keeps ATP bound for some tens of seconds (Tsai et al., 2010). The mere fact that opening is initiated by site 2 necessarily means that site 2 must undergo some movement first, before site 1 also reacts: causality must imply sequentiality, otherwise site 1 would have to be able to "look into the future", to anticipate ATP binding-induced rearrangements in site 2.
I guess the model in our head is that the NBDs are partially 'glued' by ATP1 and ready to close upon ATP2 binding. Is this incorrect?
Indeed, data from the Hwang laboratory (Tsai et al., 2010) and from our group (Szollosi et al., 2011) have provided evidence for some interaction across site 1 which is maintained even in the closed state ("partially glued NBDs"). If true, then this effect might also contribute to speeding opening (relative to unliganded gating) by preventing the NBDs from distancing themselves from each other as far as seen in unliganded cryo-EM structures (Zhang et al., 2016; Liu et al., 2017). However, this issue is still unsettled, as work from the Gadsby laboratory (Chaves and Gadsby, 2015) suggested complete separation of both ATP sites during each channel closing event.
We have now added a sentence to the Discussion to briefly expose this controversial topic: "Whether stably bound ATP in non-catalytic site 1 might also contribute to speeding opening, by maintaining some contact across the site-1 interface even in the closed state and hence preventing complete NBD disengagement, is still unsettled (Tsai et al., 2010b; Szollosi et al., 2011, but, cf. Chaves and Gadsby, 2015)."
Furthermore, we would like to point out that it is not "ATP2 binding" that initiates opening: in the closed channel ATP binding/unbinding to site 2 is a rapid equilibrium process, whereas opening rate is extremely slow (~1 s-1) even in saturating ATP. Thus, the movement at site 2 which initiates opening must occur after ATP2 binding, and has been shown to reflect closure of the site-2 NBD interface (Vergani et al., 2003; Vergani et al., 2005).
If not, what is happening to ATP1 that makes the movements around it happen later?
Regardless of whether site 1 separation is complete or incomplete in the closed state, the causality argument discussed above does not leave much room for any alternative explanation. The exact nature of these movements is obviously beyond the scope of our study.https://doi.org/10.7554/eLife.29013.021
- László Csanády
- László Csanády
- László Csanády
- László Csanády
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We thank David Gadsby and Jue Chen for critical review and discussions. Supported by MTA Lendület grants LP2012-39/2012 and LP2017-14/2017, Research Grant CSANAD17G0 from the Cystic Fibrosis Foundation, and an International Early Career Scientist grant from the Howard Hughes Medical Institute to LC.
Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols of Semmelweis University (last approved 06-30-2016, expiration 06-30-2021).
- Kenton J Swartz, National Institutes of Health, United States
© 2017, Sorum et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Downloads (link to download the article as PDF)
Download citations (links to download the citations from this article in formats compatible with various reference manager tools)
Open citations (links to open the citations from this article in various online reference manager services)
NatB is one of three major N-terminal acetyltransferase (NAT) complexes (NatA-NatC), which co-translationally acetylate the N-termini of eukaryotic proteins. Its substrates account for about 21% of the human proteome, including well known proteins such as actin, tropomyosin, CDK2, and α-synuclein (αSyn). Human NatB (hNatB) mediated N-terminal acetylation of αSyn has been demonstrated to play key roles in the pathogenesis of Parkinson's disease and as a potential therapeutic target for hepatocellular carcinoma. Here we report the cryo-EM structure of hNatB bound to a CoA-αSyn conjugate, together with structure-guided analysis of mutational effects on catalysis. This analysis reveals functionally important differences with human NatA and Candida albicans NatB, resolves key hNatB protein determinants for αSyn N-terminal acetylation, and identifies important residues for substrate-specific recognition and acetylation by NatB enzymes. These studies have implications for developing small molecule NatB probes and for understanding the mode of substrate selection by NAT enzymes.
The Polycomb repressive complex 2 (PRC2) is a multicomponent histone H3K27 methyltransferase complex, best known for silencing the Hox genes during embryonic development. The Polycomb-like proteins PHF1, MTF2, and PHF19 are critical components of PRC2 by stimulating its catalytic activity in embryonic stem cells. The Tudor domains of PHF1/19 have been previously shown to be readers of H3K36me3 in vitro. However, some other studies suggest that PHF1 and PHF19 co-localize with the H3K27me3 mark but not H3K36me3 in cells. Here, we provide further evidence that PHF1 co-localizes with H3t in testis and its Tudor domain preferentially binds to H3tK27me3 over canonical H3K27me3 in vitro. Our complex structures of the Tudor domains of PHF1 and PHF19 with H3tK27me3 shed light on the molecular basis for preferential recognition of H3tK27me3 by PHF1 and PHF19 over canonical H3K27me3, implicating that H3tK27me3 might be a physiological ligand of PHF1/19.