Domain Analysis and Visualization of NLRP10

Published Online August 2013 in MECS

• I.    Introduction

  Computer tools are proved indispensable and useful in the research of many scientific fields, especially in advancing human knowledge in the structures and functions of biological processes and entities. In the past decades, various algorithms [1-7], computational models [8-12], web servers [13-18], simulations [1923], databases [24-28], computational intelligence approaches [29-33], and imaging techniques [34-38] have been developed to aid the analysis and process of complex biological data. The use of computational tools has become prevalent in many biomedical niches such as data mining [39-40], sequence analysis [41-43], computational     biology     [44-49],     structural

bioinformatics [50-52], molecular designs [53-55], systems biology [56-58], protein science [59-61], drug discovery [62-63], and even biophysics [64-66]. With these computational tools, biological data can be categorized and analyzed according to the scientific needs. Besides, greater accuracy and structural insights of the molecular data are made possible with computational approaches. Computational solutions to the gigantic volume of biological data appear to be a promising approach to aid the advancement of biological sciences [67].

In this study, we employed computer modelling and fold prediction approach in the study of the domain of NLRP10, which is one of the members of NOD-like receptors (NLRs) family [68]. NLR family members are the main constituents of the inflammasome, which is a molecular assembly that is activated upon infection or cellular stress [68]. Inflammasomes are important in regulating the innate immune defenses by inducing proinflammatory cytokines interleukin-1β and interleukin-18 [68-69]. NLRs have also been recognized to complement the immunological functions of Toll-like receptors (TLRs), a group of pathogen sensors which are membrane-bound and triggering the transcription factor NF-ĸB, mitogen activated protein kinases and Jun amino-terminal kinase [70-72]. Most of the NLRs and TLRs induce the inflammatory response by recruiting adaptor proteins [71-76]. Such recruitment requires the identical or similar structural domain of the interfacing proteins. Hence, the understanding of the protein domain is important for a better insight in the signaling pathways of these pathogen sensing mechanisms. Despite some of the NLRs, such as NLRP3, have been widely studied, the molecular details of NLRP10 remain poorly understood. The understanding of NLRP10, especially its domain, would greatly elucidate the downstream signaling pathways and the immunologic mechanism of cytokine induction.

Domain analysis is an approach which has been widely adopted by bioinformaticians and computer scientists in the analysis of protein structure. The common computational approach in domain analysis lies in the identification of motif, such as those applied in SLiM-mediated protein interactions [77], homology study [78,83], localization of structural motifs [79], and the binding site identification [80-81]. Besides, theoretical computer science, such as the concept of graph theory, is also frequently applied in the analysis of protein domain [82]. Because the structural insights of proteins are closely associated to protein function, domain analysis is vital and versatile in revealing the cellular processes. To date, the known domains of NLRP10 are Nucleotide-binding and Oligomerization (NACHT) and pyrin (PYD). This research undertakes to analyze the domains of NLRP10 using computational tools. The findings of this research may provide computational insights into the structure and functions of NLRP10, which is crucial for further investigation in the fields such as structural bioinformatics and immunopathology.

In this paper, the procedures and methods are described in detail in Section II. The obtained results were presented in Section III with elaborated discussion. A conclusion of the findings is given in Section IV.

• II.    Methods

The nucleotide and amino acid sequence of NLRP10 were retrieved from the National Center for Biotechnology Information (NCBI). We used neural network based Pcons [85] to find the structural templates for NLRP10. Upon the identification of the modelling template for NLRP10, NMR Restraints Grid [84] was used to identify the NMR data. NRG-CING database [86] was used to model and validate the structure of the template for NLRP10. The algorithm used in the structural modelling is Saltbridge [88]. In addition, we used Ramachandran plot [89] to visualize backbone dihedral angles Ψ against φ of amino acid residues in the protein. The backbone of a protein was considered as a discrete curve, which permits the Frenet frames to be calculated based on space curves [87]. Let

We define a unit tangent vector at point P j , where j =0,…, n -1, as such:

+ + 1 - P j

tj

S j

The points of the curve were computed from the translation of the sequences {tj} and {sj} by k-1

k - P0 =S Sjt, j=0

• III.    Results and Discussion

NLRP10 is a protein constituted by 655 amino acids. It has a Pyrin domain at its N-terminus and a central NACHT domain, as depicted in Fig. 1.

S j = p + 1 - ^pj l

Fig. 1: The schematic view of NLRP10 domains

Using Pcons server [85], we obtained 5 protein templates which serve as structural models for NLRP10. These templates are summarized in Table 1.

Table 1: Protein templates for NLRP10

Rank	Pcons score	ProQ score	Template
1	0.050	117.57	2KN6
2	0.047	58.01	2HM2
3	0.047	60.36	1UCP
4	0.046	68.40	2DO9
5	0.044	56.24	1PN5

Fig. 2: Consensus based quality prediction for 2KN6

Among 5 candidate proteins, 2KN6 is the best matched protein template with NLRP10, based on the amino acid sequence alignment. The consensus based quality prediction for 2KN6 was depicted in Fig. 2.

It is clear that the predicted quality index is highest between amino acid positions of 100-200 range, though there is a flux in quality index in this range. The predicted quality index is dropped after position 200, drastically, to zero and below. Notably, the second half of the protein exhibits negative values in the consensus based quality prediction. Since sequence consensus does not totally reflect the structure, we performed structural based quality prediction for 2KN6, as illustrated in Fig. 3.

Fig. 3: Structural based quality prediction for 2KN6

Fig 3 demonstrates that the quality prediction based on the structure of 2KN6 has positive quality index (except at position 97, tyrosine), with most parts of the protein acquiring quality index higher than zero. The lowest value is -0.02216 at position 97, following with 0.01278 at position 96 (valine). The highest quality index is 0.90657 at position 10 (tryptophan).

NMR Restraints Grid [84] was used to identify the NMR data for 2KN6. The completeness statistics are summarized in Table 2.

Table 2: The Completeness statistics for 2KN6

Parameters	Values
Model count	20
Residue count	215
Total atom count	3060
Redundancy threshold %	5.0
Completeness cutoff	4.0
Completeness cumulative %	36.3
Constraint unexpanded count	2495
Constraint intra-residue count	793
Constraint observed count	1544
Constraint expected count	2546
Constraint matched count	924
Constraint unmatched count	620

We queried NRG-CING database [86] to model and validate the NMR data of 2KN6, which is a model template for NLRP10. The side chain and backbone validation are depicted in Fig 4.

Fig. 4: Side chain and backbone validation

The side chain and backbone validations of 2KN6 NMR data (as shown in Fig 4) demonstrate varieties of structural angles (PHI, PSI, CHI1/2). The angle at certain amino acid position at the side chain is drastically large, as shown at position 28-29, 45-47, 57-59, 115-117, 117-119, and 183-185. The residue properties of 2KN6 were probed. Fig. 5 illustrates one segment of the sequences.

c. RMS devs from mean coords: main-chain (black) and sidechain (grey) 150

oo ll ltllaeHaatlLlllIfctlLillLllllll.il ll I1MII UllWellaal

5      10     15     20     25     30     35     40     45     50

Helix [    - Beta strand wi Random coil АссеамЫйу Moding: ■ Buncd

• e. Sequence & average estimated accessibilities • AccewNe • Buried lileeeeeeeeeeeeeieееееееееееееееиее•♦•<•••••e<••♦•••<

MGRARDAI LDAL EN I. TAE E L К К F К I. К I I. S V P LREGYGR I PRO A 1. L SMDALDLTI

Fig. 5: The residue properties of 2KN6 (amino acid 0-55)

The top panel in Fig. 5 shows the RMS deviation of the model from the template. To have an accurate model, it is desirable to have a small RMS deviation. Our obtained results show that the deviation is small enough, reflecting an accurate match between model and template. The middle panel in Fig. 5 demonstrates the accessibility of the secondary structure. In overall, it was noticed that a large portion of the sequence demonstrates a pattern of random coil, implying that the protein backbone will sample all possible structures in the absence of stabilized interactions. As shown in Fig. 5, statistical random coil was found in the vicinity of amino acid position 1-3, 15-17, 30-40, and 46-48. Besides, random coils were also identified in the vicinity of the position 60-63, 77-79, 90-115, 126-128, 135-143,150-155, and 167-170 (data not shown). These suggest that non-local amino acid interactions are absent in these random coil regions. The bottom panel of Fig 5 depicts the accessibility of the sequence to solvent and other binding proteins.

To understand the allowable regions of the residue, we have used Ramachandran plot for this purpose. The plot represents each amino acid residue as a dot in a graph of φ against backbone dihedral angles Ψ. The residues in favored region and generously allowed region are shown in red dot and yellow dot, respectively.

From Fig 6, we notice that the residues in favored region are clustered largely negative for φ whereas positive for Ψ. The triangles in Fig 6 represent glycine residues, which provide flexibility for enzyme active sites [90]. We summarized the plot statistics in Table 3.

Fig. 6: Ramachandran plot

Table 3: Statistics of Ramachandran plot

Residues in most favored regions	1470	86.5%
Residues in additional allowed regions	198	11.6%
Residues in generously allowed regions	14	0.8%
Residues in disallowed regions	18	1.1%
Number of non-glycine and non-proline residues	1700	100%
Number of glycine residues	40
Number of proline residues	120
Total number of residues	1860

From Table 3, it is evident that most of the residues are falling within the most favored regions (86.5%). The number of glycine and proline residues is low, with a total percentage of 8.6% of the total number of residues. We obtained the distant restraints based on short-, medium-, and long-ranged sequence separation.

Restraints within the same residues are short-range; sequence separation within four residues is mediumrange; and sequence separation greater than four residues is categorized as long-range restraint. The distant restraint is plotted in the chart as shown in Fig 7.

Distance (A)

• ■    Number of upper bound distance restraint»                   Data subdivided according to:-

• I Number of lower-bound distance restraints                        Short-range (restraints within same residue)

• ■ Number of upper-hound violation*                             Medium-range (sequence separation <= 4 residues)

• ■ Number of lower-bound violations                             Long-range (sequence separation > 4 residues)

Fig. 7: Distant restraints (short-, medium-, and long-range)

As shown in Fig 7, the number of long-range distant restraints reduced drastically (<50) across the residue distance. Regardless of the residue range categories, the number of restraint at smaller distance ( ≤ 2.5 A) is greater than that of greater distance ( ≥ 5.0 A). The data collected on distant restraint is reliable as the number of upper- and lower-bound violations is very low.

position 97, tyrosine), with most parts of the protein acquiring quality index higher than zero. The side chain and backbone of 2KN6 were validated, and the residue properties of 2KN6 were analyzed. Our analysis of the structural properties of 2KN6 casts light on the domain features of the NLRP10 protein, which is critical for the understanding of NLRP10-implicated immunologic diseases.

• IV. Conclusion

A computational approach combining computer modelling and fold prediction has been used in this research to analyze the domain of NLRP10. 2KN6 serves as a structural template because it is the best matched protein template for NLRP10. In general, the structure of 2KN6 has positive quality index (except at