**3. Discussion**

‐ ‐ ‐ ‐ ABC transporters use the energy derived from ATP hydrolysis to mediate transport of a wide range of substrates across membranes. Several members of the ABCC subfamily have been studied for their role in drug resistance and human diseases. Many of these transporters translocate negatively charged solutes. For example, ABCC1 transports organic anions, including LTC4, a cysteinyl leukotriene with proinflammatory properties [26,27,36], while ABCC5 transports cyclic nucleotides such as cAMP and cGMP [37,38], important for signal transduction. Most of the available data indicate ABCC6 transports ATP and other NTPs out of cells [10,11,39,40]. However, even now, ABCC6-mediated ATP efflux

‐ ‐

has not been shown in vesicular uptake experiments, widely used to demonstrate that a given compound is actively transported by an ABC transporter [6,19]. Being involved in the efflux of ATP confers a unique physiological role for ABCC6 among ABCC transporters [19]. We hypothesized that the substrate-binding cavity of ABCC6 is also unique among ABC transporters, involving amino acids that are not necessarily evolutionarily conserved among members of the ABCC subfamily. To address this, we built homology models of hABCC6/rAbcc6 in two different conformations (ATP-free, inward-facing and ATP-bound, outward-facing) (Figure S1). The previous homology modeling studies of ABCC6 were performed using available structures of bacterial ABC transporters or the related mouse P-glycoprotein, a member of the ABCB subfamily [41,42]. Here, we used the cryoEM structures of bAbcc1 [22,23], and compared the residues in the bipartite LTC<sup>4</sup> binding pocket of bAbcc1 with other ABCC1 sequences, as well as ABCC6 and ABCC5 sequences (see the Supporting Information alignment files). This choice was dictated by the shared ability of ABCC1, ABCC5 and ABCC6 transporters to translocate negatively charged substrates including cyclic nucleotides, as well as an early study reporting the ability of hABCC6 to transport LTC<sup>4</sup> [20], which suggested there may be similarities in substrate recognition by hABCC6, hABCC1, and ABCC5. Of note, however, is that rAbcc6 has never been shown to transport LTC<sup>4</sup> [35] and later studies also failed to confirm LTC<sup>4</sup> transport by hABCC6 [6]. Nevertheless, the overall electrostatic properties of the transmembrane cavity appear remarkably similar between the hABCC6/rAbcc6 models and the bAbcc1 cryoEM structures, with a strong positive potential that might contribute to the driving force for the uptake of negatively charged substrates from the cytosol (Figure 1 and Figure S2). Interestingly, the negative potential on the extracellular end of the TMDs (Figure S2A) appeared less prominent following the conformational change to the ATPbound state (Figure S2B), possibly facilitating the negatively charged ATP to leave the substrate binding cavity. Among the bAbcc1 residues found in the proposed LTC4-binding cavity (Table 1) are three residues that are identical to the corresponding R1168, W1217, and R1220 in rAbcc6 (Figure 2 and Figure S4), which are conserved among all human ABCC transporters (Figure S4) and the sequences considered in this study for model building, likely indicating a common function in maintaining the integrity of the substrate binding cavity. In the ATP-free, inward-facing homology model, their side chain faces the main cavity, and the residues belong to TM helices 16 and 17 of TMD2, known to be crucial for substrate binding and transport in ABCC1 [32,43]. Interestingly, in ABCC1 substitutions of W1246 (corresponding to rAbcc6 W1217) adversely affects transport of estradiol-17β-glucuronide and methotrexate but not of LTC<sup>4</sup> [29,43], also implicating a role of this amino acid in transporter substrate selectivity. Furthermore, even conservative same charge substitutions of R1197 and R1249 in hABCC1 cause a global loss of transport activity [32].

We proceeded with functional studies to test if the putative transmembrane ATPbinding site of ABCC6 overlaps with the LTC4-binding cavity of bAbcc1 using a rAbcc6 model system [22]. Of the fourteen single amino acid changes introduced into the putative ABCC6 substrate binding cavity, five were found to reduce ABCC6-dependent ATP release by >75%, M369, L534, R1168, T1214, and R1220. Mutating the amino acid residue corresponding to rAbcc6 L534 in ABCC1 (T550) did not affect organic anion transport [44]. In contrast, even conservative mutations of hABCC1 F385 and Y1243 [45] (corresponding to M369 and T1214 in rAbcc6) adversely affected the transport capacity of one or more organic anions by hABCC1 (Unpublished, Conseil and Cole). Regarding the rAbcc6 R1168 position, previous studies (Table S2) have shown that opposite charge but also likecharge substitutions of hABCC1 at R1197, corresponding to rABCC6 R1168 and hABCC6 R1169, respectively, substantially reduced overall organic anion transport activity (all 4 organic anion substrates tested) as well as LTC<sup>4</sup> binding [32]. Regarding rAbcc6 R1220, mutations of the corresponding R1221 in hABCC6 have been reported to be disease causing (R1221C) [46–48] and pathogenic (R1221H) [47,49]. The corresponding amino acid in hABCC1, R1249, is crucial for overall organic anion transport activity as well, not just

glutathione-dependent binding of substrates and transport of LTC<sup>4</sup> [32,33]. In hABCC2, the analogous amino acid R1257 is also indispensable for activity, as the mutant protein is deficient in glutathione conjugate transport, despite correctly routing to the plasma membrane [50]. In hABCC4, substitution of R998, which is analogous to R1221 in hABCC6 (and R1249 in hABCC1), by alanine completely abolishes the transport of cyclic guanosine monophosphate (cGMP) [51]. Based on the fact that the rAbcc6 R1168 and R1220 are highly conserved in ABCC1-6 and that the mutation of these residues hampers the transport function in the paralogs, we consider it likely that the presence of the charged residues at these positions is indispensable for all ABCC proteins and the requirement for a positive charged residue at this position is not specific to ABCC6. The other residues corresponding to those that form the binding cavity for LTC<sup>4</sup> in bAbcc1 had less impact on the rAbcc6-mediated ATP efflux. Of these, only L316A and H424A had activity that was <50% that of wild-type rAbcc6 (Figure 4). Somewhat surprisingly, despite evidence of misfolding, the mutant rAbcc6-11aa protein containing the same LTC<sup>4</sup> binding cavity amino acids as bAbcc1 was still functional and able to efflux ATP. This suggests that the amino acids corresponding to those proposed to form the bAbcc1 LTC<sup>4</sup> binding cavity, in ABCC6/rAbcc6 are not essential for interaction with or the recognition of its physiological substrate, ATP. Our conclusion, therefore, is that the binding site for ATP in the transmembrane cavity of ABCC6 is clearly distinct from the LTC<sup>4</sup> binding cavity in bAbcc1. The possible exceptions are two highly conserved positively charged residues described above, namely rAbcc6 R1168 and R1220, which are common to the substrate-binding cavity of the ABCC transporters characterized and likely are essential for proper folding and assembly into a transport competent protein. Despite the fact that ABCC1 and ABCC6 arose from a recent gene duplication, simple evolution of a common substrate-binding site most likely does not explain the structurally very distinct substrates effluxed by the two proteins.

The molecular details of ABCC6-mediated cellular ATP release remain unknown. As outlined above, an attractive hypothesis is that ABCC6 functions as an ATP-dependent ATP efflux pump. There are three sets of observations that support the idea that ABCC6 is an ATP efflux pump. First, most members of the C-branch of the ABC superfamily, including ABCC6′ s closest homolog ABCC1, are bona fide organic anion efflux transporters [24,26,52] and ABCC6 has been shown to transport several organic anions in vitro, albeit sluggishly [6]. Second, the ATP efflux rates found in HEK293-ABCC6 cells [10] are compatible with a direct transport mechanism for ATP as these rates are very similar to the rates by which ABCC1, ABCC2, and ABCC3 pump morphine-3-glucuronide out of transfected HEK293 cells [18]. Third, ATP efflux from ABCC6-containing cells can be blocked by the general ABCC transport inhibitors benzbromarone, indomethacin and MK571 (data not shown).

As mentioned, vesicular transport experiments are often used to establish substrates of ABC transporters [24]. However, such assays have so far failed to directly demonstrate the ABCC6-dependent transport of radiolabeled ATP into inside-out membrane vesicles. We can, therefore, not completely exclude the possibility that ABCC6 mediates cellular ATP release other than by direct transport. Purification of ABCC6 and subsequent reconstitution in proteoliposomes should provide a cleaner experimental system to study ATP transport, for instance by using dual color fluorescence burst analysis (DCFBA) [53], a technique that has a more favorable signal-to-noise ratio than the standard vesicular transport assays that employ radiolabeled ATP. The elucidation of the molecular structure of ABCC6, for instance by cryoEM, might also in the future provide clues about the molecular mechanism by which ABCC6 mediates ATP release. Despite many years of intense work on ABCC6, this ABC protein has not given many of its secrets away.
