NAMPT and NAPRT1: novel polymorphisms and distribution of variants between normal tissues and tumor samples

Nicotinamide phosphoribosyltransferase (NAMPT) and nicotinate phosphoribosyltransferase domain containing 1 (NAPRT1) are the main human NAD salvage enzymes. NAD regulates energy metabolism and cell signaling, and the enzymes that control NAD availability are linked to pathologies such as cancer and neurodegeneration. Here, we have screened normal and tumor samples from different tissues and populations of origin for mutations in human NAMPT and NAPRT1, and evaluated their potential pathogenicity. We have identified several novel polymorphisms and showed that NAPRT1 has a greater genetic diversity than NAMPT, where any alteration can have a greater functional impact. Some variants presented different frequencies between normal and tumor samples that were most likely related to their population of origin. The novel mutations described that affect protein structure or expression levels can be functionally relevant and should be considered in a disease context. Particularly, mutations that decrease NAPRT1 expression can predict the usefulness of Nicotinic Acid in tumor treatments with NAMPT inhibitors.

Nicotinamide phosphoribosyltransferase (NAMPT) and nicotinate phosphoribosyltransferase domain containing 1 (NAPRT1) are the main human NAD salvage enzymes. NAD regulates energy metabolism and cell signaling, and the enzymes that control NAD availability are linked to pathologies such as cancer and neurodegeneration. Here, we have screened normal and tumor samples from different tissues and populations of origin for mutations in human NAMPT and NAPRT1, and evaluated their potential pathogenicity. We have identified several novel polymorphisms and showed that NAPRT1 has a greater genetic diversity than NAMPT, where any alteration can have a greater functional impact. Some variants presented different frequencies between normal and tumor samples that were most likely related to their population of origin. The novel mutations described that affect protein structure or expression levels can be functionally relevant and should be considered in a disease context. Particularly, mutations that decrease NAPRT1 expression can predict the usefulness of Nicotinic Acid in tumor treatments with NAMPT inhibitors. N icotinamide phosphoribosyltransferase (NAMPT) and nicotinate phosphoribosyltransferase domain containing 1 (NAPRT1) are major enzymes in the cellular metabolism. Their substrates, nicotinamide (Nam) and nicotinic acid (NA), respectively, are important precursors in Nicotinamide Adenine Dinucleotide (NAD) biosynthesis 1 . Several pathways contribute to the replenishment of the NAD pool and, in mammalian cells, Nam is the predominant precursor 2,3 while NA is more effective in increasing NAD levels in some tissues 4 . Given that NAD participates as a coenzyme in oxidation-reduction reactions, but is also a substrate of NAD-consuming enzymes that are involved in gene expression regulation, DNA repair or cell death, it is expected that NAD availability directly influence pathological conditions 5,6 . Increasing evidence points to a role of NAD salvage enzymes in cancer and neurodegeneration [5][6][7][8] .
Originally, NAMPT was identified as pre-B-cell colony enhancing factor 1 (PBEF1) 9 and as visfatin 10 , and its role in NAD biosynthesis was recognized later 11 . NAMPT functions in immunity, metabolism, and stress responses in physiology and pathophysiology 12 . As energy generation and NAD-dependent signaling are crucial for cell proliferation and cancer progression, NAMPT inhibitors have emerged as promising antitumor drugs 5,13,14 .
NAPRT1 increases intracellular NAD levels and prevents oxidative stress 4 . Since NA increases NAD levels via NAPRT1 action, the oral administration of NA was suggested to ameliorate NAD depletion conditions, namely, as cytoprotective agent in cancer treatments with NAMPT inhibitors [13][14][15] .
Genotype-phenotype associations in a disease context raise interest in single nucleotide polymorphisms (SNPs) detection. SNPs in human NAMPT associated with pathological conditions such as glucose and lipid metabolism alterations 16,17 , acute lung injury 18 , coronary artery disease 19 and type 2 diabetes 20 , are located in noncoding regions. Polymorphisms in NAMPT promoter are related to plasma insulin levels 21 and increased cholesterol 22 . For NAPRT1 the only association with disease, namely, attention deficit hyperactivity disorder 23 , is the synonymous SNP rs2290416 (G428, exon 10), responsible for differences in protein expression 24 .
Motivated by the importance of these enzymes in metabolism and homeostasis, and their involvement in several human diseases, we present the results of large-scale analyses that involved more than 200 samples from both normal and tumor origin in order to characterize potentially relevant mutations. The impact of novel variants at the structural and/or functional levels is discussed.

Results
NAMPT and NAPRT1 gene polymorphisms. Despite all the available literature on NAMPT polymorphisms, public databases show that NAMPT is less diverse than NAPRT1. Yet, little is known about the impact of NAPRT1 polymorphisms. We analyzed 96 DNA samples from a control population, focusing in mutation hotspots of the NAMPT and NAPRT1 genes (Supplementary Table  S1). At the beginning of this study, the regions analyzed were chosen based on the information retrieved from public databases and the literature. For NAMPT, we found four intronic variants with minor allele frequency (MAF) lower than 0.05 (Table 1). One frequent SNP was detected in exon 7 (rs2302559, g.21735A.G), corresponding to Ser301. The frequencies observed in our samples were consistent with results from the 1000 Genomes Project ( Table 1).
The above mentioned 96 samples and additional 53 normal tissue samples, consisting in blood, stomach and colon from different ethnic groups, were analyzed for mutations in NAPRT1 (Table 2 and Figure 1). Among the five silent substitutions detected, one at Val142 codon (g.676C.G) remains unreported. The synonymous variant g.1803C.T at Leu305 (rs872935) is the most frequent and found in nearly 63% of the alleles, consistent with 1000 Genomes data for the Caucasian population. Two missense variants (rs200364051, p.Val106Met and rs35975875, p.Arg332Cys) were detected with fre-quencies of 0.5% and 0.3%, respectively, and four non-coding variants were also identified. From the eleven NAPRT1 variants found, five had MAF,0.05, including two missense, two silent and one noncoding, in accordance to 1000 Genomes data ( Table 2). Sequencing the entire NAPRT1 gene allowed the discovery of five additional noncoding variants, plus one novel deletion in intron 8 (g.2542_2544delCCC) ( Table 2).
To evaluate whether the variants have different prevalence in pathological conditions and normal tissues, 80 DNA samples from colon and gastric tumors were also analyzed. Apart from two silent alterations, which were also detected in normal samples (g.468C.T and g.1803C.T, at Ala98 and Leu305, respectively), no missense mutations were found in NAPRT1. Curiously, the g.1803C.T variant was more frequent in the normal population than in tumor samples (63% and 26%, respectively). From the five intronic variants found, g.257C.A and g.565G.T were more frequent in tumor than in normal samples (around 30% and 13%, respectively) ( Table 2 and Figure 1). In the 1000 Genomes data, these are also more frequent in the Asiatic population, consistent with the fact that the tumor samples analyzed were from Asian origin, while control samples were mostly Caucasian. Thus, these results may be explained by different frequencies between populations rather than differences observed between control and tumor samples. In addition, we identified a polymorphism which was not previously described in intron 7 (g.2013A.G). The remaining variants showed similar frequencies between populations (Table 2 and Figure 1). To validate the population effect on the allelic differences, we used re-sampling statistics 25,26 and compared NAPRT1 bootstrap confidence intervals (95%) for mean allele frequencies found in our data (control and tumor samples) with the 1000 Genomes allele frequencies (European and  Asian populations). For most variants, we observed a similar trend between normal and European population frequencies, as well as between tumor and Asian population frequencies (Supplementary Table S3).
Impact of missense variants in the protein structures. Since there is no human NAPRT1 structure available, models of the protein were predicted by homology modeling and by ab initio modeling, based on Saccharomyces cerevisae 27 and Enterococcus faecalis NAPRTases. After assessing quality (Supplementary Figure S3) the model with the lowest z-score was chosen (Supplementary Figure S4) to evaluate the structural impact of the NAPRT1 missense mutations found in our samples (Figure 2A). Taking into consideration residue conservation, Arg332 is invariant throughout all Metazoan species analyzed whereas Val106 is substituted by interchangeable residues (valine and leucine, except for yeast), and even methionine in Ciona intestinalis ( Figure 2B). Accordingly, the SIFT 28 and PolyPhen 29 predictions considered the Val106Met substitution tolerated/benign and the Arg332Cys mutation as deleterious/probably damaging, respectively (Supplementary  Table S4). In the structural models, the Val106Met replacement does not cause an apparent modification in the protein structure ( Figure 3A and B) however, in the Arg332Cys replacement a distinct network of H-bond contacts is observed when a cysteine residue replaces the ancestral arginine ( Figure 3C and D). Specifically, Arg332 establishes polar contacts with Ala328, Arg336, Phe335 and Glu372, and the polar contact with Glu372 is lost with the Cys332.
Using the NAMPT structure 30 and our model for NAPRT1, we located all the missense mutations described so far as well as the predicted active sites in both proteins ( Figure 4). Missense mutations found in NAMPT are far from the active sites or the dimer interface ( Figure 4A), thus, their potential impact on the enzyme structure and/or function should be limited. On the other hand, NAPRT1 missense mutations, which are abundant, are spread all over the molecule ( Figure 4B). This is consistent with SIFT 28 and PolyPhen 29 predictions as well (Supplementary Table S4).

Discussion
NAMPT and NAPRT1 are important enzymes in NAD metabolism, both in normal and in pathological conditions. In this study, we targeted the regions where most missense mutations were located. Variation data from Ensembl Genome Browser (Dec2010 -Jan2013) indicated an increasing number of coding region alterations in both genes, although more evident for NAPRT1 (Supplementary Figure  S5). For instance, in the same period, 40 missense and 20 silent new mutations were described in NAPRT1, compared to 8 and 15, respectively, in NAMPT. This shows that NAPRT1 is more permissive to mutations and, thus, much more polymorphic than NAMPT, in which a strong purifying selection must be acting on.
Our results confirmed that NAMPT has a lower genetic diversity than NAPRT1. We found a smaller number of alterations and observed that missense mutations are rare in the human gene. In fact, we detected only intronic and synonymous variants. Intronic SNPs in NAMPT are associated with disease phenotypes related to glucose and lipid metabolism and other metabolic and vascular traits (see 31 and 32 for review), possibly due to alterations in splicing, protein binding or methylation sites that will affect the levels of protein expression. As for synonymous mutations, it has been recently shown that rs2302559 correlates with NAMPT serum levels 33 , thus, the impact of these novel variants is yet to be determined.
In NAPRT1 we detected two missense and five silent variants in samples from healthy individuals. One of these, g.676C.G, is an unreported alteration in exon 3 (Val142Val). Although commonly considered silent, synonymous mutations can have a phenotypic effect and be implicated in human disease 26,33,34 . Protein expression or conformation can be affected by speeding up or slowing down protein synthesis or by affecting splicing.
Using a novel methodology to estimate the pathogenicity of human genetic variants 35 , we observed that many NAPRT1 variants here described are associated with splicing sites (Supplementary  Table S5). Given the location of the Val142 near the exon border, we used the NetGene2 online server, which predicts human splice sites 36 , to evaluate the impact of this new polymorphism. We observed that an alternative splice site near the mutated g.676C.G is predicted with higher confidence than the splice site that determines the exon 3-intron 3 boundary in the reference sequence, resulting in an alternatively spliced transcript. As a recent study suggests that synonymous mutations change the sequences that regulate splicing in oncogenes, and are frequently associated with cancer 26 , the impact of this, and other, synonymous mutations should be considered in further studies.
The absence of NAPRT1 expression has been reported in different types of cancer 14,37 , thus, we expected to find a higher number of alterations in NAPRT1 that could affect protein expression. This is of particular importance to validate NA as a chemoprotectant in the treatment of cancer patients with NAMPT inhibitors, which would be effective in NAPRT1-negative tumors only 38 . Curiously, no missense mutations were detected in the tumor samples studied, further supporting the role of intronic and synonymous mutations in promoting changes in protein expression, as discussed above. Recent work also shows that epistasis influences NAPRT1 gene expression 39 , suggesting that further studies will be necessary to establish which SNP or SNP associations are preponderant in defining NAPRT1negative phenotypes.
Although the human NAMPT enzyme is well characterized, with known active site and dimer interface residues 30,40,41 , for human NAPRT1 no protein structure is yet available. A recent study 42 on the kinetic characterization of this enzyme identified residues involved in activity, and used the Thermoplasma acidophilum NAPRTase to infer structural information. To study the structural impact of missense variants, we also built models for human NAPRT1, and evaluated Val106Met and Arg332Cys mutations considering residue conservation and changes in bond contacts. We further located known missense mutations of NAPRT1 and NAMPT in their respective structures, as well as residues involved in the active center 30,43 . For NAMPT, none of the missense mutations would interfere directly with the active center of the enzyme or with the dimer interface, whereas in NAPRT1, the missense mutations are located all over the molecule. Despite no direct bond to active center residues was detected, these mutations may influence assembly of the active dimer form, resulting in a dysfunctional or inactive protein.
The growing number of mutations described in NAPRT1 that occur all over the molecule is somewhat puzzling. Although it could be secondary in tissues that express NAMPT, NAPRT1 is useful for cells that lack NAMPT, such as neuron, and for cells in the digestive tract due to abundance of NA from bacterial metabolism 44 . Moreover, as NA is a more efficient NAD precursor 4 , situations with high NAD demand or NAMPT impairment would benefit from the activity of NAPRT1. Further studies are required to elucidate if the variants found in this study are correlated with protein expression levels, not only in physiological normal conditions but also in a disease context.  Supplementary  Table S1. PCR amplification was as follows: 95uC (15 min), 35 cycles at 94uC (30 sec), 58-64uC (1 min) and 72uC (1 min), and final extension at 72uC (10 min). Amplification products were separated by horizontal polyacrylamide gel electrophoresis and visualized by silver staining.
Homology modeling and structure visualization. The human NAPRT1 protein sequence was used as template to search for the best E-value PDB using the NCBI BlastP analysis 46 This search identified structures from Enterococcus faecalis, Thermoplasma acidophilum and Saccharomyces cerevisiae as optimal structural templates. E. faecalis putative nicotinate phosphoribosyltransferase and S. cerevisiae Npt1p structures (PDB id: 2F7F and 1VLP) were used as templates in MODELLER 47 to build human NAPRT1 structural models by homology modeling. Additionally, the I-TASSER online server 48,49 was used to predict NAPRT1 structure by ab initio modeling. Accuracy of the models (Supplementary Figure S3) was estimated using ProSA-web (https://prosa.services.came.sbg.ac.at/prosa.php), as previously described 50,51 . Models of the NAPRT1 variants Val106Met and Arg332Cys were built in MODELLER, using as template the I-TASSER predicted structure (Supplementary Figure S4). The human NAMPT structure used corresponds to the PDB id: 3DKJ 30 . All structures were visualized in Pymol v1.1r1 52 .
Mutation analyses. Re-sampling statistics were performed by bootstrap analysis using MATLAB (R2011a, MathWorks, Natick, MA, USA). The method used for the construction of the bootstrap confidence intervals (95%) was the accelerated biascorrection (Supplementary Table S3). Pathogenicity prediction of the variants was performed as follows. Available information of the SIFT 28 and PolyPhen 29 predictions, for all dbSNP missense mutations in the human NAMPT and NAPRT1 reference transcripts (ENST00000222553 and ENST00000449291), were retrieved from Ensembl 53 (release 75 -February 2014) (Supplementary Table S4).
For all the variants found in our study, the pathogenicity was also scored according to CADD (http://cadd.gs.washington.edu/score) 35