Characterization of the human patatin-like phospholipase family.

Several publications have described biological roles for human patatin-like phospholipases (PNPLAs) in the regulation of adipocyte differentiation. Here, we report on the characterization and expression profiling of 10 human PNPLAs. A variety of bioinformatics approaches were used to identify and characterize all PNPLAs encoded by the human genome. The genes described represent a divergent family, most with a highly conserved ortholog in several mammalian species. In silico characterization predicts that two of the genes function as integral membrane proteins and are regulated by cAMP/cGMP. A structurally guided protein alignment of the patatin-like domain identifies a number of conserved residues in all family members. Quantitative PCR was used to determine the expression profile of each family member. Affymetrix-based profiling of a human preadipocyte cell line identified several members that are differentially regulated during cell differentiation. Cumulative data suggest that patatin-like genes normally expressed at very low levels are induced in response to environmental signals. Given the observed conservation of the patatin fold and lipase motif in all human PNPLAs, a single nomenclature to describe the PNPLA family is proposed.

The patatin glycoprotein is a nonspecific lipid acyl hydrolase that is found in high concentrations in mature potato tubers (1). Patatin is reported to play a role in plant signaling (2), to cleave fatty acids from membrane lipids (2), and to act as defense against plant parasites (3). Proteins encoding a patatin-like domain are ubiquitously distributed across all life forms, including eukaryotes and prokaryotes [see the species distribution tree of the PFAM (protein family) database (4), accession number PF01734 (http://www.sanger.ac.uk//cgi-bin/Pfam/getacc?PF01734), for further details], and are observed to participate in a miscellany of biological roles, including sepsis induction (5), host colonization (6), triglyceride metabolism (7), and membrane trafficking (8). Of interest is the observation that prokaryotic patatin-like proteins appear more similar to eukaryotic paralogues than any other bacterial lipases (9). This suggests that distant homologs may have arisen from a common ancestor. Together, these observations indicate that the patatin-like domain represents a unique superfamily quite distinct from all other lipolytic families.
Recent studies investigating the role of the hormonesensitive lipase in mammalian adipose tissue identified a novel patatin-like lipase (TTS -2.2) that was significantly upregulated in differentiating murine adipocytes and appeared to act coordinately with hormone-sensitive lipase in the catabolism of triglycerides (10). The rat ortholog of this gene was also found to be upregulated in differentiating adipocytes and transiently induced during fasting (11). The yeast protein Tgl4 has since been reported to be a functional ortholog of the mammalian TTS -2.2 gene (7).
Jenkins et al. (12) reported details of three murine patatin-like proteins [adiponutrin (ADPN), TTS-2.2, and GS2] with abundant triacylglyerol (catabolic) and transacylation (anabolic) activities. All three were observed to associate with the cell membrane. The ADPN gene transcript was found to be virtually absent in a fasted state and dramatically upregulated in response to feeding. A more recent report (13) characterized four human ADPN-like lipases (ADPN, TTS-2.2, GS2, and GS2-like) and detailed changes in gene expression during murine adipocyte differentiation. Both TTS-2.2 and GS2 were highly expressed in metabolically active tissues. It was also reported that, except for ADPN, overexpression of the ADPN-like proteins resulted in decreased intracellular triglyceride levels (13).
Given the significant association of the patatin-like proteins in adipocyte differentiation and their response to metabolic stimuli, the purpose of this investigation was to use a variety of bioinformatics approaches to identify and characterize all human proteins encoding a patatin-like domain. Combined with gene expression data, the collected observations will be used to further assess and characterize the role of these phospholipases in mammalian lipid metabolism.

Bioinformatics analysis
A redundant human sequence set was identified from the PFAM (4) patatin family alignment (accession number PF01734). Each of these sequences was used as input to a Position-Specific Iterated Basic Local Alignment Search Tool (PSI-BLAST) (14) search to query a nonredundant (.94% identity between sequences) database composed of all available human gene and gene prediction sequences (included were the TREMBL, Swiss-Protein, GenPept, ENSEMBL, GeneSeqP, PIR, and PDB databases).
Equivalent full-length transcript sequences of each patatin-like gene were used to query a redundant human expressed sequence tag (EST) database (GenBank release 151, December 2005) in an effort to verify the expression of the transcripts. Gene expression was further verified using the GeneLogic Genesis Enterprise database (http://www.genelogic.com/genomics/genesis.cfm).
The patatin domain encoded in each protein was delineated using the SMART database (15). Secondary structure was predicted using JNET (16), and fold prediction was completed using the FUGUE application (17). Low-resolution homology models were generated using the Swiss-PDB Viewer application (18).
Sequence alignments were completed using ClustalW (19). The resulting alignment was used as input for PhyloWin (20) to determine putative evolutionary relationships between members of the human patatin-like gene family. Full-length protein sequences were used as input to TMHMM (21) and SignalP (22) to predict transmembrane domains and signal peptide sequences, respectively.
Putative orthologs in mouse and rat were identified by reverse BLASTing a nonredundant nucleotide database derived from the EMBL nucleotide database (23).

TaqMan real-time quantitative reverse transcription-PCR analysis
Tissue processing and sample preparation were described previously (24). Briefly, cDNA prepared from 0.4 ng of poly(A1) RNA extracted from each tissue sample was loaded into each well on a 384-well optical microplate. Gene-specific primers were designed using Primer Express software (Applied Biosystems) and confirmed by BLAST searches of public and propriety sequence databases. Details of each primer are as follows (gene name/ forward/reverse/probe) The concentrations of the forward and reverse primer and probe for each assay were 900, 900, and 100 nM, respectively. Quantitative PCR was carried out using the 7900HT Sequence Detector System (Applied Biosystems) in a 10 ml reaction volume. TaqMan Universal PCR Master Mix 2X (Applied Biosystems) and universal PCR conditions recommended by the vendor (Applied Biosystems) were used. Abundance values for each tissue sample were calculated using a genomic DNA standard curve to convert TaqMan fluorescent output to a known number of sequence copies.

Microarray analysis
Differentiation of multipotent adipose-derived stem cells, isolated from human adipose tissue, was completed as described previously (25). Total RNA was extracted using Trizol (Invitrogen, Carlsbad, CA). RNA purity and integrity were evaluated using the Agilent Bioanalyzer and Optical Density spectrophotometer. All samples were prepared according to the Affymetrix (Palo Alto, CA) protocol (http://www.affymetrix.com/support/). Fifteen micrograms of labeled copy RNA samples was hybridized onto U133plus2 microarrays (Affymetrix) using standard Affymetrix protocols. Quality-control parameters were as follows: noise (RawQ) , 3, consistent scale factors, background signal , 50, consistent number of genes called as "present" across arrays, and b-actin and GAPDH 59/39 signal ratios , 3. Signal intensities were collated and analyzed using Rosetta's Resolver Biosoftware (http://www.rosettabio.com/products/resolver/default.htm).

RESULTS
Extracting a nonredundant set of human sequences from the PFAM PNPLA alignment (accession number PF01734) generated eight unique patatin-like sequences (ADPN, TTS-2.2, Neuropathy Target Esterase (NTE), FLJ43070, IPA2g, PNPLA1, PNPLA4, and PNPLA5). Each sequence was used to search a nonredundant human transcript database using a PSI-BLAST-based algorithm (14). This identified two additional patatin-like sequences, XM_066899 and Phospholipase A2, group VI (PLA2G6), not included in the PFAM alignment. BLAST alignments indicated that both sequences are most homologous to intracellular membrane-associated calcium-independent phospholipase A2g (iPLA2g) (data not shown).
All 10 human sequences that encoded a significant patatin-like domain were compared using the PFAM motif search tool and the FUGUE profile search tool. Both the PFAM and FUGUE similarity scores for each of the human patatin-like domains are detailed in Table 1. Both techniques were in broad agreement and indicated that the NTE, NTElike, and iPLA2g sequences had higher homology to the patatin domain than did the ADPN-like subfamily.
Attempts to align the human protein sequences resulted in low-quality alignments as a result of minimal homology between input sequences. The residues that determined homology to the patatin domain were identified using the PFAM protein query tool (accession number PF01734/ patatin domain), and the alignment procedure was repeated using the restricted sequence as input. Although this approach demonstrated an improvement relative to the fulllength protein alignment, much of the sequence remained poorly aligned. Information derived from secondary structure prediction (JNET) and fold prediction (FUGUE) was used to generate a manual alignment of the patatin-like sequences. The resulting alignment highlighted several regions of high conservation, including the characteristic glycine-rich region, the nucleophilic elbow, and the aspartate-glycine residues of the catalytic center. When shaded by physiochemical characteristics (Fig. 1), the protein alignment clearly illustrated the conservation of similar residues at several defined positions dispersed throughout the patatin-like domain. Such residues are obvious candidates for future mutagenesis studies. The alignment also highlighted blocks of physiochemical conservation across those regions predicted to fold as a-helices and b-sheets interspersed by highly divergent loop regions.
The alignment also highlights the fact that the PNPLA1 sequence does not include the N-terminal region of the patatin domain. In particular, it is missing both the glycinerich region and the G-X-S-X-G motif, which includes the nucleophilic serine, an essential component of the catalytic dyad. Without these residues, it is highly unlikely that PNPLA1 could function as a lipolytic enzyme. Extensive searches of the EMBL nucleotide sequence database (release 85, December 2005) did not identify additional hu-man sequences that could be used to extend the N-terminal coding sequence of PNPLA1. However, predicted ortholog sequences from rat (XP_342106), mouse (XP_484618), dog (XP_538884), and chimpanzee (XP_527370) all demonstrate a highly conserved N-terminal sequence that encodes the lipase signature motifs. Use of the predicted PNPLA1 sequences to query the human genome sequence with BLASTP (data not shown) identified what appears to be an additional, upstream exon that would encode both the glycine-rich region and the G-X-S-X-G motif, immediately upstream of the human PNPLA1 locus. The predicted fulllength sequence of the human PNPLA1 gene has been submitted to GenBank (accession number AM182887).
PNPLA4 includes an alanine substitution for the first conserved glycine in the G-X-G-X-X-G motif. The alanine substitution is seen in all available human transcripts and is also the equivalent residue in the dog, mouse, cow, and frog ortholog sequences (alignment not shown). A serine residue occurs in the equivalent position in the rat ortholog sequence. The functional implications of this substitution are currently unclear, but it does not negate lipolytic activity, as PNPLA4 has previously demonstrated both lipase (13) and transacylation (12) activities.
Sequence comparisons indicated that the putative human patatin-like family could be divided into four subfamilies: an ADPN-like family, as proposed by Lake et al. (13), an NTE-like subfamily, and two unique members represented by PNPLA9 (PLA2G6) and PNPLA8 (iPLA2g). This observation was reinforced by a phylogenetic comparison of the patatin-like domain sequences (Fig. 2). PNPLA1 was found to cluster with the ADPN-like subfamily (data not shown). Because the method incorporated a global gap-removal function, the initial tree was PNPLA, patatin-like phospholipase. In silico characterization of the human patatin-like family of phospholipases. Summary of in silico predictions: column 2 indicates the length of the longest representative human peptide sequence in the public domain. Columns 3 and 4 list the scoring outputs for each of the predicted patatin-like domains from the PFAM and FUGUE algorithms, respectively. Columns 5, 6, and 7 summarize predicted signal peptides, transmembrane domains, and significant functional motifs as determined by the SignalP, TMHMM, and SMART applications. The relative positions of the predicted motifs shown in parentheses refer to the full-length peptide sequence used as input.
derived from only 72 residue positions. To increase confidence in the analysis, the partial domain sequence represented by PNPLA1 was removed. The resulting phylogenetic tree was constructed from 144 residue positions and did not alter the previous relationship (i.e., when PNPLA1 was included in the analysis) of the remaining sequences.
Low-resolution homology models of each of the human PNPLAs were generated using Macromolecular Structure Database (MSD) entry 1OXW (i.e., a crystal structure derived from a plant patatin isozyme) as a template. These were used to give an approximation of the three-dimensional organization of conserved residues. A representative model of the human PNPLA3 (ADPN) is shown in Fig. 3 and illustrates both the conserved a/b, patatin-like fold and the spatial orientation of the residues that compose the catalytic dyad (model coordinates are available as supplementary data). When each model was examined in conjunction with the sequence alignment, it was apparent that the catalytic residues were "framed" by several conserved hydrophobic residues (e.g., residues 8, 49, 109, 189, and 195 of the alignment in Fig. 2). When structural data were combined with the observed evolutionary relationships (Fig. 2), other positions were predicted to determine subfamily specificity (e.g., a conserved leucine residue at position 9 is specific to the ADPN subfamily, whereas an isoleucine residue at position 46 appears to be a characteristic of the NTE-like subfamily). Such observations will be the input to future mutagenesis studies that will fully characterize both enzyme function and substrate recognition.
All protein sequences were further characterized using a variety of in silico prediction algorithms (Table 1). Briefly, PNPLA6 (NTE), PNPLA7 (NTE-like), and PNPLA2 (TTS-2.2) are all predicted to encode single transmembrane domains. PNPLA4 is predicted to be a secreted protein, which along with the PNPLA1, PNPLA8 (iPLA2g), PNPLA6, and PNPLA7 protein sequences all encode one or more tyrosine kinase phosphorylation sites. Both the PNPLA6 and PNPLA7 sequences encode regions of cyclic nucleotide binding sites, whereas PNPLA9 is characterized by a region of ankyrin repeats.
The expression profile of each gene was verified using TaqMan analysis of a representative sample set of human tissues (see supplementary data for all expression data). This indicated that PNPLA3 (ADPN) was expressed at very low levels in a number of tissues with marginally higher copy numbers detected in liver, bone, and macrophages (Fig. 4). Equivalent data extracted from the GeneLogic Genesis Enterprise database (probe set 233030_at) sup-ported the observation of increased PNPLA3 expression in liver tissue (data not shown) and contradict previous descriptions of this lipase as an adipocyte-specific gene (11), although this discrepancy may reflect variation between the rat and human species.
PNPLA1 was also observed to be expressed at very low levels in a number of tissues, notably in the digestive system (see supplementary data). There are no representative Affymetrix probe sets for this gene, so the observed TaqMan data could not be verified using the GeneLogic databases in this instance.
Moderate copy numbers of PNPLA5 were detected predominantly in brain tissue, although very low copy numbers were detected in liver tissue (Fig. 4). PNPLA5 was not observed (data not shown) to be upregulated during adipocyte differentiation; however, probe set 233597_at is of low quality (data not shown). Lake et al. (13) observed that PNPLA5 was strongly induced in the liver of ob/ob mice and proposed a role in hepatic lipogenesis. The significance of this observation is not clear, as PNPLA5 was not expressed in the liver tissue of C57BI/6J mice. Of the 23 human ESTs representing the human PNPLA5 transcript in the EMBL (23) nucleotide database (release 84), 7 had been isolated from brain tissue, 7 from assorted skin tissue samples, and the remaining constructs from miscellaneous tissues.
TaqMan data indicated that the sequence XM_066899 was not expressed in any of the tissues tested. Combined with the absence of supporting EST data and the fact that the predicted transcript does not encode a catalytic aspartate residue (alignment not shown), it was concluded   that this sequence likely represents a pseudogene. The predicted peptide sequence is most similar to the PNPLA8 (iPLA2g) sequence (data not shown). The remaining genes were expressed in most tissues tested. In particular, PNPLA9 (PLA2G6) was predominantly expressed in adipose and brain tissue, PNPLA4 was predominantly expressed in muscle and adipose (Fig. 4), PNPLA7 (NTElike) was predominantly expressed in all prostrate, pancreas, and adipose tissues tested, and PNPLA8 (iPLA2g) was moderately expressed in all tissues tested. All experimental values are available as supplementary data.
In an effort to determine a possible role in adipocyte metabolism, the expression profile of each patatin-like gene was monitored in differentiating human adipocyte tissue. Available data concurred with previous reports (12,13) of significant upregulation of both PNPLA3 (ADPN) and PNPLA2 (TTS-2.2) during the differentiation process. The upregulation for both genes was observed from day 2 of the differentiation process and continued to increase until day 14, when adipocyte differentiation was complete (25). PNPLA8 (iPLA2g) was observed to be upregulated on day 2 (P 5 7.4E-07) and then downregulated by day 7 (Fig. 5). PNPLA6 (NTE) was marginally downregulated from day 2 (P 5 6.0E-03), whereas PNPLA9 (PLA2G6) was upregulated on day 14 (P 5 5.9E-03) of the differentiation process (Fig. 5). No differential expression was observed for PNPLA5 (GS2-like), as reported previously (13). This may reflect species differences, although the authors describe the use of very high RT-PCR cycle numbers to detect gene expression (which could not be detected by Northern blotting). All experimental values are available as supplementary data.

DISCUSSION
Given previous reports detailing a critical role for the PNPLAs in energy mobilization and adipose storage, it was considered important to identify and further characterize additional family members to further understand the physiological role of this lipase family.
The combined analyses identified nine genes and one likely pseudogene that encode a patatin-like domain in the human genome. However, using conventional sequence alignment algorithms, the family demonstrates negligible sequence homology outside of the patatin-like domain. That the nine expressed genes all belong to a single superfamily is supported by the P values reported by the PFAM search tool combined with the predicted z scores obtained from the FUGUE fold prediction application ( Table 1). The latter predicts that all nine of the expressed genes should be categorized as a FabD/lysophospholipaselike fold (SCOP database entry 52150; http://scop.mrclmb.cam.ac.uk/scop/). This protein fold incorporates both the dual glycine-rich and G-X-S-X-G characteristic lipase motifs and a catalytic aspartate, all within close proximity in the three-dimensional structure. Homology modeling predicts that all of these features are spatially conserved in all of the human PNPLAs.
Sequences were manually aligned using structural prediction data. Comparison of aligned residues indicated that several residues, in addition to those known to be critical for enzyme activity, are conserved in all of the proposed family members (Fig. 1). For example, the serine/ threonine residue at position 186 in the structurally guided alignment (Fig. 1) is situated close to (4-6 Å ) the catalytic serine (measurement derived from MSD entry 1OXW). Given the proximity and conservation of this position, it is likely that this residue has a role in determining the activity of the PNPLAs.
Phylogenetic analysis of the patatin-like (domain) sequence indicates that five of the genes cluster together in what has been described as the ADPN family (13). Of the remaining four sequences, PNPLA6 (NTE) and PNPLA7 (FLJ43070) cluster together and PNPLA8 (iPLA2g) and PNPLA9 (PLA2G6) separate out as two unique family members. Given the comparative levels of sequence conservation, it seems probable that the ADPN and NTE subfamilies are the result of distinct gene duplication events. In spite of the conservation of critical catalytic residues, the observed sequence divergence suggests that family members will not display a similar substrate specificity profile.
Expression data indicate that several of these genes are expressed at relatively low levels in a wide range of human tissues. However, the significant upregulation of PNPLA3 (ADPN), PNPLA2 (TTS-2.2), and PNPLA8 (iPLA2g) observed during adipocyte differentiation (Fig. 5), combined with reports of upregulation in response to fasting (11) and feeding (13), suggests that these three genes respond rapidly to a variety of physiological stimuli that are indicative of their role in facilitating both energy mobilization and lipid storage in adipocyte tissue (12). In contrast to a previous study (10), it was observed that human PNPLA2 gene expression continued to increase until day 14 rather than day 6, as observed previously. If validated, this would highlight a functional difference between species that may have implications for selecting suitable animal models to further evaluate the physiological role of this enzyme in lipid metabolism.
It is tempting to speculate that an increase in transcription similar to that observed for other members of the ADPN subfamily will also be observed for the PNPLA1 transcript when the functional role of this protein is elucidated and an appropriate stimulus is applied. The various analyses completed indicate that the human PNPLA1 protein shares most homology with the ADPN-like subfamily, but available sequences do not encode a G-X-S-X-G motif and as such are unlikely to act as a hydrolase. Predictions based on the human genomic sequence indicate that the public domain transcripts of the human PNPLA1 gene represent a 59 truncated version of the gene and that the full-length transcript would encode the characteristic lipase motifs. The predicted full-length human transcript can only be validated by gene cloning. The C-terminal sequence is proline-rich but encodes no additional significant motifs. What functions this protein confers are currently unknown.
The observed upregulation of the PNPLA8 (iPLA2g) gene in differentiating adipocytes is a "novel" observation that will require further investigation. A discrete isoform of this membrane-associated phospholipase has been observed in the mitochondrial compartment of the rat myocardium and appeared to facilitate integrating respiration with mitochondrial thermogenesis (26).
In contrast, the PNPLA7 (NTE-like), PNPLA9 (PLA2G6), and PNPLA4 transcripts are relatively highly expressed in a number of human tissues. PNPLA4 is predominantly expressed in both muscle and adipose tissue (Fig. 4), and overexpression of the gene has been associated with the decrease of intracellular triglyceride levels (12). Knockdown assays will be required to confirm a precise role in cellular/lipid metabolism. Note, however, that a murine PNPLA4 ortholog has not yet been confirmed.
Previous publications report that PNPLA6 is expressed in blood lymphocytes (27) and avidly hydrolyzes a number of lysophospholipids, indicating a role in intracellular membrane trafficking (28,29). The human PNPLA6 and PNPLA7 proteins share z60% sequence identity and are highly conserved in mouse, rat, and dog (.90% and .80% sequence identity, respectively; alignments not shown). This high level of conservation is often indicative of a conserved function. In addition, both are integral membrane proteins and encode cyclic nucleotide binding sites, indicating that they may be regulated by either cAMP or cGMP. PNPLA9 (PLA2G6) is a relatively well-characterized protein that is activated during apoptosis (30,31) and reported to hydrolyze platelet-activating factor at the sn-2 fatty acid position (32). These observations indicate a key . Expression data were determined using Affymetrix U133 chips and analyzed using the Resolver TM ANOVA algorithm with standard methodology. The average intensity value (P < 0.05) at each time point is represented by the y axis. Error bars represent the observed SD. All expression data values and individual graphs used to generate the graphic summaries can be obtained from the supplementary data. role in mediating aspects of the inflammatory response. Protein function is further complicated by extensive alternative splicing (33) and the presence of ankyrin repeats (indicative of mediating protein-protein interactions), which are required for enzyme activity (32).
Partial, low-resolution models of each of the human PNPLAs predict that the central core of each protein is composed of a b-sheet sandwiched by a-helices (a/b/a) in approximately three layers and that the catalytic serine is situated in a nucleophilic elbow after a b-sheet and preceding an a-helix. There appears to be spatial conservation of several hydrophobic residues surrounding the conserved serine nucleophile that may aid substrate binding, although precise coordinates are unavailable given the minimal sequence conservation between the target and template sequences. The significant divergence across the predicted loop regions is readily apparent, and it is tempting to speculate that it is these regions that play a key role in determining the unique characteristics of each family member.
The plant patatin phospholipase is reported not to exhibit interfacial activation (1); however, it is apparent from the patatin (MSD entry 1OXW; http://www.ebi.ac.uk/ msd/) structure that a loop (residues represented by positions 82-95 in the alignment shown in Fig. 1) appears to stand "proud" of the main protein structure (Fig. 3). The predicted topology indicates that this loop could theoretically reposition over the catalytic residues (i.e., as a lid), although the reported B-factor values derived from the crystal structure are relatively high for this stretch of residues, indicating minimal flexibility. The authors report that the crystal structure has three molecules in the asymmetric unit. When a number of asymmetric units are assembled according to the symmetry of the crystal, the loops are observed to form crystal contacts with symmetry-related molecules. It is possible that these may provide some stability to this region when packed in the crystal structure. Interfacial activation studies have not yet been reported for any of the human PNPLAs, but PNPLA9 (PLA2G6) has been observed to specify a preference for 1,2-dipalmitoyl phosphatidic acid when the substrate is presented in vesicles (32). PNPLA9 has been observed in a large multimeric complex, so it is not clear whether the observed specificity is attributable to interfacial activation or another mechanism involving one or more of the multimer components. It will be of interest to further investigate the residues in this loop region to determine either a role in substrate recognition or the regulation of enzyme activity.
As with most preliminary efforts to characterize novel protein families, a variety of aliases are used to describe the same entity. This can result in confusion and hinder the detection of shared functional characteristics. Given the conservation of a unique patatin-like lipase motif and the predicted characteristic patatin-like fold found in each of the described proteins, we recommend that the family be henceforth described using the prefix PNPLA, as described above. This proposal has been forwarded to the HUGO Gene Nomenclature Committee, and the proposed names PNPLA3, PNPLA7, PNPLA8, and PNPLA10P have been accepted by the committee (PNPLA1, PNPLA2, PNPLA4, and PNPLA5 had previously been agreed to). Both the PNPLA6 and PNPLA9 proposals are under discussion.
In summary, described are pertinent details and selected expression data that partially characterize the human patatin-like family of phospholipases. The proteins described appear to represent a unique lipolytic family. We recommend further study to determine both the precise, normal physiological role and the putative disease association of each of the described proteins.