Activity and structure of human (d)CTP deaminase CDADC1

As expected, each CDADC1 protomer consists of two deaminase domains, N-terminal (NTD) and C-terminal (CTD) (Fig. 2G). Our structures show that the NTD and CTD span the residues 31–189 and 316–476, respectively, slightly different than automatically annotated (ProRule PRU01083). Two domains are positioned in an antiparallel back-to-back configuration relative to each other. They are connected by a long linker (residues 190–315), composed of five α-helices, one of which (L (linker) α5) bears a putative nuclear export signal (residues 271–283). Importantly, the peptide bond between N266 and P267 upstream of the Lα5 has a cis-conformation, resulting in a wedge-like structure of the region (residues 262–272). In addition to the linker, the two domains are connected by reciprocal interactions between their α1 helices and β2 strands. Both NTD and CTD contain a standard cytidine deaminase core (19). It consists of a five-stranded mixed β-sheet (β2-β5 parallel, β1 antiparallel), sandwiched between three α-helices, α1 from one side and α2 and α3 from the other (Fig. 2H). The NTD, but not CTD, contains an additional α-helix (N (NTD) α4, residues 168–183), which is also present in human DCTD (PDB 2W4L). Consistent with the intrinsic disorder prediction (SI Appendix, Fig. S15), every CDADC1 protomer contains unstructured regions, which are undefined in the ESP maps. These regions include residues 1–28 upstream of the NTD, 53–83 and 163–166 within the NTD, 221–224 within the linker, 304–309 and 376–390 within the CTD, and 477–514 downstream of CTD. The latter region contains the putative bipartite nuclear localization signal (488–510).

NTD and CTD tetrahedrally coordinate one zinc ion (Zn²⁺) each using conserved PCxxC motifs (Figs. 2G and 3 A and B). In the NTD, Zn²⁺ is coordinated by H109, C134, C137, and surprisingly not by a water molecule, but by a side chain carboxyl of E157, which blocks the access to the Zn²⁺. The remaining pocket space is fully occupied by V85 and I158. The inactive HAG sequence is located in the same spot in the NTD as the conserved HAE motif in every DCTD, strengthening the hypothesis that G111 is a result of natural mutation during evolution. Therefore, our structures demonstrate that NTD is inactivated in two different ways (Fig. 3A). In the CTD, Zn²⁺ is coordinated by C246, C429, H398, and one water molecule, which together with I455 and F393 form the ceiling of the active site. The bottom of the active site is formed by hydrophobic residues V421, V340, G341, and A342. The left wall is defined by the main chains of C424 and P425 and a side chain of Y457. The right wall is made by the main chains of H398 and A399 of the conserved HAE motif (20) (Fig. 3B). E400 is an essential residue (SI Appendix, Fig. S3), that in many deaminases acts as a proton shuttle (Fig. 3B) (6). In our structures, the side chain of E400 is missing, because the residue was mutated to alanine to prevent catalysis.

In addition to the catalytic site, DCTDs have an allosteric site that regulates their activity. Since DCTD proteins have only one domain, their allosteric site is located at the interface between the protomers in the hexamer. The binding of dCTP was shown to stabilize hexamers and is believed to be the main mechanism of allosteric regulation (21, 22). A potentially equivalent site to the allosteric site in the DCTDs is present in the CTD, but not in the NTD of CDADC1 (SI Appendix, Fig. S16). In agreement with our inability to identify an allosteric regulator biochemically, the CDADC1 maps suggest that this putative allosteric site is not occupied. Thus, a fusion of NTD and CTD in CDADC1 may have permanently stabilized the interface between the two and made the canonical allosteric regulation obsolete.

CDADC1 trimers are closed rings formed by three protomers assembled in a head-to-tail orientation. The majority of the contacts are formed between Nα2 and Nα3 from the NTD of one protomer and Cα2 and Cα3 of the CTD of the other. In addition, the structured portion of the loop 367–396 (369–375 and 391–396) near the CTD active site forms contacts with both Nα3 and Nα4 from the NTD. This results in the vast interface of ~1,100 Ų between the two adjacent protomers, according to the PISA server (23). This also means that the NTD remains an integral structural component of CDADC1 even though it has lost its catalytic function.

CDADC1 hexamers are formed by symmetric stacking of two trimers, i.e. trimers face each other with the same sides. Each protomer equally contributes to the hexamer formation. CDADC1 apo hexamers are less tight than trimers, with a contact area per protomer of ~270 Ų. Each protomer from one trimer interacts with two protomers from the other trimer. With one of them, it is arranged in an antiparallel symmetric manner. However, no direct contacts can be seen between the two protomers in the apo model. One potential contact point between the pair is the “blob” in the map where two trimers are the furthest from each other (~7 Å), between two C197. The blob is likely derived from the partially structured 1–28 region (SI Appendix, Fig. S17). We exclude the possibility of potential disulfide bond formation with other residues because none of the missing regions of this intracellular protein contains cysteines. The interaction with the second protomer is more direct. It is achieved by symmetric hydrogen bonds between guanidino groups of R185 and carbonyl oxygens of A186, located between Nα4 and Nβ5 of the NTD. Therefore, we conclude that this interaction is likely the dominant force for hexamer formation in the absence of substrate, which highlights the important role played by Nα4 in both trimer and hexamer formation.

We also calculated ESP maps and built models of trimeric and hexameric CDADC1 particles in the presence of dCTP and ^5mdCTP. CDADC1 trimers that were observed in samples containing dCTP or ^5mdCTP do not differ much from the apo trimers. There is no evidence for ligand binding in the maps of the CTD active sites, and no significant structural changes seem to occur (SI Appendix, Figs. S12 and S13), suggesting that trimers do not bind dCTP or ^5mdCTP. However, CDADC1 hexamers reconstructed from the same datasets are different. In the presence of dCTP, each CDADC1 protomer contains a well-defined dCTP in the active site, therefore one CDADC1 hexamer binds six dCTP molecules. Consistent with our biochemical results, the maps for the allosteric site indicate that neither dCTP nor ^5mdCTP are bound, suggesting that the observed cooperativity must originate from the active site (SI Appendix, Fig. S16). Within the active site, the cytosine base of dCTP stacks under Zn²⁺ coordinating H398. Its O2 enforces the rotation of the sidechain of N367 and forms a hydrogen bond with NH of A399. The rotated sidechain of N367 allows for accommodation of both 2′C and 3′OH. This positions the N3–N4 edge of the base toward the usual location of the catalytic glutamate (E400 in CDADC1), a configuration shared among all Zn²⁺-dependent cytidine deaminases (6). Since we replaced E400 with alanine, the space normally filled by the sidechain is occupied by a water molecule. In the absence of E400, dCTP N4 displaces the Zn²⁺ bound water molecule of the ligand-free structures and participates in the Zn²⁺ coordination, mimicking the transition state. The C5 of dCTP is located at <4 Å from C426, I455, and Y457, strongly suggesting that the low CDADC1 activity on ^5mdCTP and ^5hmdCTP is a result of a steric conflict. Accommodation of dCTP also drives the significant rearrangement of the K392-F393 region, likely due to the steric conflict between the 3′OH and α-phosphate (αP) on the dCTP side and aromatic ring of the F393 on the CDADC1 side. F393 remains in close proximity to the αP (3.5 Å), and thus it seems to act as a “sensor” of the bound substrate. The 2′C of the bound dCTP is brought into close proximity (~4 Å) of the side chain of I396, suggesting that the presence of a hydroxyl group would be sterically unfavorable. This explains the lower CDADC1 activity on CTP than on dCTP.

The lack of activity on CMP but not on dCMP suggests that the interaction with all three phosphates is necessary to force the ribose into the active site. Indeed, the CDADC1 structure with bound dCTP demonstrates an extensive network of interactions with the triphosphate tail. Astonishingly, five lysine residues that constitute the positively charged patch on the CDADC1 surface all rearrange to form salt bridges with dCTP phosphates. The K392–F393 rearrangement mentioned earlier enables the Nε of K392 to form a salt bridge with the αP. K423 forms a salt bridge with βP, and possibly also a hydrogen bond with the 4’O of the deoxyribose. K337 and K452 form salt bridges with both αP and γP from two opposite sides. K450 forms an additional salt bridge with γP (Fig. 3 C and D). All five lysine residues that interact with the dCTP or ^5mdCTP triphosphate tail are highly conserved in CDADC1 proteins, from sharks all the way to humans (SI Appendix, Fig. S18). The multiple ionic interactions with dCTP contribute not only to ligand specificity, but also to long-distance communication within the CDADC1 hexamer. We have compared the positively charged ligand binding site of CDADC1 with the equivalent regions in XP-12 deaminase (a viral deaminase active on the triphosphate), PBCV-1 deaminase (another viral deaminase active on the monophosphate and triphosphate), and human DCTD (active on the monophosphate). As expected, the triphosphate-specific CDADC1 and XP-12 deaminase show the greatest accumulation of lysine and arginine residues (SI Appendix, Fig. S19).

As could be expected based on the kinetic data (Fig. 1F), the ESP map for ^5mdCTP is less clear than the map for dCTP. A comparison of the CDADC1 active sites with dCTP and ^5mdCTP shows that the two substrates are bound similarly (SI Appendix, Fig. S20). However, the C5 methyl of the ^5mdCTP clashes with the hydroxyl of Y457. To resolve the conflict, both Y457 and ^5mdCTP shift away from each other, which puts N4 of ^5mdCTP in the less favorable orientation for catalysis. Moreover, the shift results in the weakening of the electrostatic interactions between the triphosphate tail and the lysine residues. In particular, the salt bridge between the K423 and the αP seems to be disrupted almost completely. Since DCTDs usually do not select against C5-modified substrates, we compared the active sites of known DCTDs with the active site of CDADC1. As a control, we used the AlphaFold2 (24) model of phage XP-12 DCTD that specifically targets ^5mdCTP. Indeed, DCTDs either have more space in the region (human DCTD) or feature an aromatic residue able to participate in the electron–π interaction with the methyl group (W121 in PBCV-1, W128 in XP-12, and Y153 in T4) (SI Appendix, Fig. S21). Therefore, the active site of CDADC1 has evolved to specifically bind the unmodified dCTP, possibly to avoid dTTP binding.

The incorporation of dCTP into the active site triggers a cascade of structural changes on an increasing scale. The rotated N367 forms a hydrogen bond with T333, and T338 forms an additional hydrogen bond with dCTP βP. The loop region between these two threonines undergoes reorganization to avoid the collision between K393 and K377 bound to dCTP and H336. All the above leads to the stabilization of a large unstructured region (residues 376–390). This region is visible only in the EP maps of the CDADC1 hexamers with bound dCTP. A significant portion of this region (residues 381–390) folds around the sidechain of R391, in a C-shaped manner. The chain of electrostatic interactions: dCTP (αP and γP)–K452–D383–R391–D454–R288 also seems to contribute to the stabilization. Moreover, the upstream 373–375 region which is visible in the maps in the absence of substrate also rearranges. Therefore, for simplicity, the entire 373–390 region will be regarded as a substrate response loop (SRL) (Fig. 4). The dCTP-induced reorganization of SRL substantially increases the number of contacts between CDADC1 protomers within both trimers and hexamers. Within trimers, new contacts are made with the Nα4 of the adjacent protomer, which slightly increases the interface area. A much larger difference is observed in the interaction between protomers that belong to different trimers in the hexamer. Most importantly, the two antiparallel protomers that are too far from each other to form contacts in the absence of substrate, directly interact in the presence of dCTP. The structured SRL in CTD of each protomer makes an extensive set of contacts with the cis-P267 wedge of the opposing protomer. This promotes the interaction between the two NTDs, which form many symmetric contacts, such as the K181–N183 hydrogen bond (Fig. 4 B and E). The interaction between opposing protomers is strong enough to bring two trimers closer by ~1.3 to 1.8 Å (Fig. 4 and SI Appendix, S14). Moreover, the entire Nα4 helices shift toward the center of the hexamer (Fig. 4 C and F). Upon dCTP binding, the combined interface area in CDADC1 hexamer increases to ~1,130 Ų per protomer. Thus, a single protomer with bound dCTP contributes several-fold more of its surface to the hexamer interface than without a ligand (Fig. 4G), explaining the stabilizing effect of dCTP against CDADC1 dissociation that was observed at the very low protein concentrations used for MP (Fig. 4H). CDADC1 hexamers with ^5mdCTP are (almost) as compact as those with dCTP (Fig. 4 and SI Appendix, S14). Moreover, the interface area per promoter in the hexamer is also similar (Fig. 4G). However, the map quality for the ligand and SRL is lower, reflecting the weaker binding of ^5mdCTP compared to dCTP.

In order to understand how CDADC1 structural rearrangements are related to the cooperativity, we compared CDADC1 to its closest relatives, the DCTD proteins. The architecture of the CDADC1 trimer, which consists of six deaminase domains (NCNCNC), is strikingly similar to the architecture of the DCTD homohexamers (SI Appendix, Fig. S22A). Interestingly, the DCTD protomers contain loops similar to the SRL (SI Appendix, Fig. S22B). These loops form symmetric contacts with the α4 helix of the neighboring protomers in the hexamer ring. The binding of dCTP in the allosteric sites is believed to stabilize this interaction and thus promote DCTD activity (25). Despite the striking similarity between DCTD hexamers and CDADC1 trimers, symmetric contacts are impossible in the latter, because NTD and CTD are not equivalent, and only one pair of contacts between Nα4 and CTD SRL is formed within the CDADC1 trimer. However, upon dimerization of CDADC1 trimers, two pairs of symmetric contacts between SRLs and the wedges are formed, restoring the communication between the active sites. This explains why CDADC1 hexamer, but not trimer is the active form of the enzyme and suggests a structural basis for the cooperativity.

In prokaryotes, dCTP deamination to dUTP serves as a mechanism of defense against phage infections (26), suggesting a possible role of CDADC1 in antiviral defense. To test this hypothesis, we queried OncoDB (27), which summarizes data from The Cancer Genome Atlas (28) for possible upregulation of CDADC1 in response to viral infection in various types of malignancies (Fig. 5A). In most cancer types, there was no upregulation in response to infection with different viruses. Moderate upregulation of CDADC1 in response to virus infection was only observed in some adenocarcinomas (Fig. 5A and SI Appendix, Fig. S23), but was not significant.

To test for a possible physiological role of CDADC1, we created a mouse model (in the C57BL/6JRj genetic background) with inactivation of Cdadc1, by mutation of the catalytic glutamate residue (E401 in the mouse) to alanine, using ssDNA template-directed repair of a Cas9 induced double-strand break (SI Appendix, Fig. S24). The resulting mice did not exhibit any obvious phenotypes, and both males and females were fertile. This finding suggested to us that CDADC1/Cdadc1 may have a physiologically redundant role with DCTD/Dctd, despite acting at the level of the tri- rather than the monophosphates. To test this hypothesis, we created a Cdadc1/Dctd double KO (dKO) mouse, by coinjecting sgRNA guides for Cdadc1 and Dctd. For Cdadc1, the sgRNA as for the active site point mutation was reused, but without the repair template. For Dctd, we used a pair of sgRNAs on either side of the third coding exon (fourth exon of the transcript). The initial founder mouse had a 17 bp deletion in Cdadc1 (SI Appendix, Fig. S25) and the intended 281 bp deletion, with 115 bp of the fragment inserted in an inverted orientation, in Dctd, causing a Tyr396fs*7 frameshift mutation (SI Appendix, Fig. S26). Crossing of mice with monoallelic Cdadc1 and Dctd KO created offspring without drastic deviations from Mendelian genetics [SI Appendix, Table S2; equally large or larger deviations from Mendelian genetics expected with ~40% probability according to the χ² test (29)]. In particular, the offspring included 2 homozygous dKO mice. The dKO mice had no overt phenotype (SI Appendix, Fig. S27), and a female dKO mouse was fertile. This was surprising because the Dctd KO alone should cause male and female infertility according to the International Mouse Phenotyping Consortium (30). In our hands, however, the Dctd KO was both male and female fertile, which is consistent with the low Dctd mRNA read counts in the testis (31).

Next, we analyzed five double heterozygous mice (henceforth termed WT mice), nine Cdadc1 KO mice (henceforth termed CKO mice) and six Cdadc1/Dctd double KO mice (henceforth termed dKO mice) in more detail. One dKO mouse had a very impaired gait, which could be a sporadic phenotype, or an unrelated problem. The other mice appeared normal, and indistinguishable from wild-type mice.

First, we compared the overall body mass and individual organ masses for the females that represented the majority in our cohort. The data did not reveal systematic differences in overall body mass, or the masses of prominent organs, with the possible exception of the spleen (before, but not after multiple hypothesis correction) (Fig. 5B). When males were included in the analysis and the organ masses were normalized with respect to body mass, relative spleen mass was not significantly altered (SI Appendix, Fig. S28). Metabolic indicators were also similar in the three groups. Specifically, Alanine Transaminase and Aspartate Transaminase levels provided no indication of liver pathologies, and an increase in glucose levels of CKO and dKO mice compared to WT was not significant (Fig. 5C). Red blood cell parameters were analyzed in females and also showed no signs of perturbation in CKO and dKO mice. In particular, RBC count, hemoglobin, hematocrit (red blood cell volume/total volume of blood), mean corpuscular volume (MCV, a measure of the average size of red blood cells), mean corpuscular hemoglobin (MCH, the average amount of hemoglobin per red cell, used to detect anemia), and mean corpuscular hemoglobin concentration (MCHC, amount of hemoglobin per red blood cell) were all similar in all three groups (Fig. 5D). Finally, blood cell counts were also similar. In particular, differences in the total amount of leukocytes, neutrophils, lymphocytes, monocytes, and eosinophils were all not significant. For erythroblasts, differences were significant only before, but not after correction for multiple hypothesis testing (Fig. 5E).