A novel conserved family of Macro-like domains—putative new players in ADP-ribosylation signaling

Assignment of DUF2362 proteins to the Macro clan

Analysis of distant similarities between human proteins and ADP-ribosylation-related families performed using the FFAS server suggested that the uncharacterized family DUF2362 might be similar to Macro domains. Table 1 shows that both FFAS and two other popular structure prediction servers predicted a weak but suggestive similarity to Macro for the human C12ORF4 protein (see also Fig. 1). These methods use sequence profiles or Hidden Markov Models to detect distant sequence similarities compatible with similar structures. Phyre² and HHpred are remarkably concurring in their prediction of structural similarities for C12ORF4, identifying as best-scoring structure modeling templates the Macro domains of the human poly(ADP-ribose) polymerase 14 (PDB: 3Q6Z) and human protein-proximal ADP-ribosyl-hydrolase MacroD2 (PDB: 4IQY) or human MacroD1 (PDB: 2X47). Using the servers’ results, it was difficult to distinguish between the predictions, as confidence values were very close. However, FFAS03 ran on DUF2362 multiple sequence alignment used as the query gave the highest score for human PARP14 (PDB: 3Q6Z) as structural prediction, but did not pick human MacroD1/2 structures as significant hits. Using HHpred and Phyre² local HMM profile–profile alignment, an alignment for human C12ORF4 and the two best hits was constructed, considering secondary structure assignments from the human MacroD2 structure (Fig. 1). Manual adjustment was performed only in the loop 1 region. Investigating similarities between C12ORF4 and important regions of MacroD2, a similarity was noticed to the characteristic MacroD motif Nx(6)GG[V/L/I]D in the C12ORF4 region aligned with loop 1. However, there is a difference; whereas the first residue in the glycine dyad is substituted by serine: Nx(6)SG[V/L/I]D. MacroD2 residues N92 (matched with D333 in human C12ORF4) and D102 (aligned to D345 in human C12ORF4) are catalytic amino acids electrostatically interacting with the distal ribose ring of ADPr ligands. Conservation of these residues in C12ORF4 seems to support its similarity to the catalytic Macro domain class, albeit tyrosine 190, crucial for eraser function of MacroD, is substituted by a methionine (M478). The structural water molecule and aromatic ring of Y190 located in eraser loop 2 bind the ADPr moiety in constrained position enabling initiation of its hydrolysis. The reader-type Macro domain in PARP14 has conserved Asp residues in loop 1, but tyrosine within the G[IV]YG motif is replaced by a phenylalanine, also possessing an aromatic ring. In C12ORF4, there is a matched methionine at this position. This brings uncertainty regarding the catalytic activity of the C12ORF4 protein, taking into account the fact that Y->F substitution in this motif can convert a catalytically active Macro domain to a reader-only pseudoenzyme (Rack, Perina & Ahel, 2016).

Phylogenetic spread of the DUF2362 family and relationship to the Macro clan

The DUF2362 family in the Pfam database contains sequences belonging to 261 species (see Fig. 2), mainly Metazoa, including the early Metazoan Trichoplax adherens (Placozoa), and also the closest unicellular relative of metazoans, Capsaspora owczarzaki (Ichthyosporea). DUF2362 is also found in some fungi (albeit the fungal C12ORF4 homologs belong to specialized atypical groups of fungi, e.g., Mucoromycota, Chytridiomycota, Zoopagomycota and Cryptomycota), Amoebozoa, Apicomplexa, Choanoflagellida and Cryptophyta. C12ORF4 homologs have not been found in most fungi, or in plants, archaea or bacteria. Overall, the Jackhmmer search for new potential family members also picked up 410 C12ORF4 homologs in the Representative Proteomes and about 700 C12ORF4 homologs in UniProt. The broad phylogenetic spread of this protein family confirms its ancient origin and potentially important cellular function. It is suggestive of a DUF2362 member being present in the last eukaryotic common ancestor and subsequent DUF2362 gene losses in some major eukaryotic lineages.

The relationship of the DUF2362 family to the Macro clan can be visualized using a graph-based approach: the CLANS algorithm (Frickey & Lupas, 2004). The CLANS graph depicts proteins as nodes connected by edges representing BLAST-detected significant or sub-significant sequence similarities. The nodes are clustered by “attractive forces” representing BLAST E-values (see Fig. 3). DUF2362 appears to be connected not only to the main family of the clan–Macro (PF01661), but also Macro2, Peptidase_M17, PARG-cat and DUF2263. The biggest grouping of the superfamily, the Macro family, contains the majority of known human Macro domains, eraser and reader-like, except one that belongs to the PARG family, poly(ADP-ribose)-glycohydrolases.

Sequence variability in the DUF2362 family

While distant sequence similarity search tools strongly suggested the presence of a Macro-like domain in the C-terminal region of C12ORF4 (approx. residues 322–552), the presence of another domain in N-terminal part of the protein could not be established with confidence. HHpred reported a non-significant similarity to human all-helical nuclear pore complex protein Nup155, while FFAS03 reported a weak similarity to bacterial colicin E9.

Investigation of likely domain structure in uncharacterized proteins can be aided by the analysis of their amino acid sequence logo, a graphical representation of conservation of sequence positions in the protein family. A logo created using multiple alignment of the DUF2362 family shows strong sequence conservation in the C-terminal region, but also some strongly conserved motifs between positions 85 and 110 and in the region 230–260 (see Fig. 4 and Fig. S1 for full-length DUF2362 logo). Two Asp residues (D333 and D345) identified in profile–profile alignments (Fig. 1) as likely counterparts of MacroD2 catalytic residues are well-conserved in the family, although consensus does not reach 100% (Fig. 5). The long region following the putative catalytic motifs is the least conserved part of the DUF2362 alignment (Fig. 3). This region–378–408 in human C12ORF4–has no counterparts in known Macro domain structures, it is also very poorly conserved in the DUF2362 family (Fig. S1). Figure 5 presents a comparison of conservation within two catalytic loops in two Macro domain families–eraser-type (MacroD2) and reader-type (PARP14)–and in the corresponding regions of DUF2362. In C12ORF4, two conserved Asp residues, D333 and D345, are putative counterparts of MacroD2 active site N92 and D102. While similarities within loop 1 in C12ORF4 and two known Macro families are quite noticeable, the loop 2 motifs are more different. The presence of highly conserved methionine and histidine is noteworthy. These residues may be involved in binding/positioning of the distal ribose of ADPr, similarly to Y190 of MacroD2 and F926 of PARP14.

Apart from the predicted ADPr binding and catalytic site, the functions of conserved residues are less obvious. Several strikingly well-conserved cysteine residues could serve the building of disulfide bridges. The very well-conserved histidine 431 corresponds to a histidine conserved in a VIH[TA] motif in Macro domains of PARP14 and MacroD2 and may be involved in the positioning of the active site Asp. The most conserved region (475–520) of the DUF2362/C12ORF4 family corresponds to helices 4 and 5 in known Macro domains which are distal to the ADP-ribose site and whose functions are not clear.

Profile–profile alignment methods offer automatic pipelines enabling modeling of the putative three-dimensional structure of the query protein. In the case of C12ORF4, automated methods would be quite risky, mainly due to the presence of long insertions (likely loops) within the hypothetical Macro domain. Residues from 374 to 408 had to be modeled without a structural template. Secondary structure prediction run for amino acids in this region of C12ORF4 indicate a mainly unstructured character of this fragment, so we decided to build a preliminary model for the C12ORF4 structure in the form of a homology model based on a human MacroD2 template, using our own manually adjusted alignment, with the building and optimization of missing loops using the I-TASSER server. An “eraser” template (PDB: 4IQY) was chosen because of the fact that only one Macro-like domain was detected in our query. This is similar to the “eraser” MacroD1/2 proteins that have only one catalytically active Macro domain. In contrast, the “reader” PARP14 contains three repeated Macro domains. Also, there are two likely catalytic Asp residues present in the predicted loop 1 of C12ORF4. Before building the structure model, we analyzed possible templates and conservation of important residues in C12ORF4 and the whole DUF2362 family, using sequence alignments and logo. As can be noticed in Fig. 6, conservation of DUF2362 alignment positions matched to the ligand binding cleft of MacroD2 and PARP14 is high (Figs. 6B and 6D). Also, considering pairwise alignment mapping (Figs. 6A and 6C), one can see that apart from conserved (Asp/Asp) or almost conserved (Asn/Asp) catalytic positions there are also some residues conserved in C12ORF4 in the neighborhood of the putative ADPr binding site–proline, alanine and glycine in the case of both templates and tyrosine in the case of PARP14.

After obtaining the structure model of human C12ORF4, with the ab initio constructed fragment, we used the COFACTOR algorithm to identify and place putative ligands within the obtained model. COFACTOR scans the PDB database for structures similar to the query and containing co-crystalized ligands to propose possible ligand binding site locations within the analyzed model. This represents an approximate result and one should not treat the obtained ligand binding poses as certain, but one can rank proposed ligands and identify clefts likely responsible for its binding. Results shown in Table S1 support ADPr binding by C12ORF4. The highest scoring COFACTOR ligands are ADP-riboses, which are frequently co-crystalized with Macro domain structures, which is not surprising considering the template used for homology modeling. The predicted ADPr binding cleft is localized between loop 3 (counterpart of MacroD loop 2) and helix 5 and flanked by the loop modeled by I-TASSER (see Figs. 7 and 8). Hydrogen bonds to ADPr detected in the model using UCSF Chimera tools (Pettersen et al., 2004) involve Ile 524, and His, Asp and Met of HDM (476–478) motif aligned to MacroD2 catalytic tyrosine. However, a detailed investigation of ligand binding based on the preliminary model and COFACTOR results would be too speculative.

Structure modeling for uncharacterized proteins is always a risky enterprise when no highly homologous template is available. In the case of C12ORF4, an additional complication is the presence of a putative N-terminal domain and a long insertion within the Macro-like region.

Biological function of C12ORF4

The only report available that investigated the molecular function of C12ORF4 is that of Mazuc et al. (2014) who used an intrabody approach to identify proteins involved in FcεRI-induced mast cell degranulation. They showed that silencing C12ORF4 influenced FcεRI-mediated signaling events; namely, it caused a decrease in FcεRI-mediated tyrosine phosphorylation of Src and Syk tyrosine kinases as well as the downstream kinases, MAPKs and Akt. This, in turn, had a clear effect on degranulation, by decreasing the release of TNF-α and β-hexaminidase.

The Mazuc study, together with our findings that link C12ORF4 to ADP-ribosylation, casts interesting light on the known function of PARP-1 in degranulation, and specifically in modulating asthma-associated production of cytokines (Sethi, Dharwal & Naura, 2017). Studies on asthma patient material as well as animal models have led to inhibitors of PARP1 and PARP14 being considered for asthma treatment and being investigated in animal models (Zaffini, Gotte & Menegazzi, 2018). For example, in an early study in guinea pig, PARP inhibition prevented an induced asthma-like reaction (Suzuki et al., 2004). Allergen exposition induced a rapid increase in PARP activity, bronchiolar constriction, pulmonary air space inflation and mast cell degranulation in sensitized animals, while inhibition of PARP prevented the above morphological and biochemical changes to lung tissue (Suzuki et al., 2004). Our results allow us to propose a hypothesis that Macro-like C12ORF4, as a putative ADPr “eraser,” regulates PARP-mediated ADP-ribosylation signaling, similarly to PARG, whose role in countering PARP action in airway disease is known (Sethi, Dharwal & Naura, 2017). The fact that C12ORF4 expression has been found to affect phosphorylation of several kinases (Mazuc et al., 2014) allows speculation that the effects of PARP and C12ORF4 on degranulation may be mediated by direct ADP-ribosylation of kinases. There are known cases of regulation of kinase signaling pathways by direct ADP-ribosylation of key kinases, by effectors from infectious bacteria (Young et al., 2014) and also by endogenous human ADP-ribosyltransferases (Song et al., 2016; Li et al., 2018).

An interesting connection between the degranulation-related phenotype attributed to C12ORF4 and its relationship to ID is the fact that mast cell degranulation is one of the processes involved in brain inflammation (Dong et al., 2017; Skaper et al., 2018). Brain inflammation, in turn, can lead to ID and related disorders (Di Marco et al., 2016).

Exploitation of in silico databases for additional clues regarding C12ORF4 functions provides rather few hints. According to databases such as Expression Atlas (Papatheodorou et al., 2018), Human Protein Atlas (Uhlen et al., 2015) and Human Proteome Map (Kim et al., 2014), the C12ORF4 protein-coding gene (on a protein and transcript level) is expressed in most normal tissues, with a high level of expression in the brain and nerves, and in the genitals. In disease-specific studies, high expression levels of C12ORF4 have been reported for samples collected from glioblastoma multiforme and glioma, lymphoma, pancreatic adenocarcinoma, and ovarian adenocarcinoma (Cancer Genome Atlas Research Network et al., 2013). The cBioPortal database (Cerami et al., 2012) gives evidence that the C12ORF4 protein-coding gene is altered (mutation, but not deletion) in 3% of 1,939 cases (59 patients) of endometrial cancer, and smaller fractions of patients with other malignancies, for example, breast cancer, bladder cancer, colorectal cancer and melanoma. The more frequently occurring alterations of the C12ORF4 protein-coding gene are presented in Table S2. The most frequent mutation, missense R335Q, has been observed in 14 cases, mainly in uterine endometrioid carcinoma samples, but also in patients with other malignancies (cBioportal). Arg335 is located within the putative Macro domain region of C12ORF4, in the conserved motif DNR, containing Asp333, suggested by us to be a likely catalytic residue (Fig. 5).

According to tools for elucidation of biological relationship networks, such as Ingenuity Pathway Analysis (https://www.qiagenbioinformatics.com/products/ingenuity-pathway-analysis) and String (Szklarczyk et al., 2015), C12ORF4 is involved in a functional network involving a number of mostly cytoplasmic signaling proteins. One of these interactors, TBCK, described as a modulator of mammalian TOR signaling, is involved in the regulation of cell proliferation and growth and control of actin-cytoskeleton organization. Interestingly, mutations in TBCK lead to a syndrome of ID (Bhoj et al., 2016). Other interactors include the interleukin receptor IL6R, that is a proposed asthma risk locus (Ferreira et al., 2011) and is functionally related to a number of degranulation-related signaling mediators (see Fig. S2)