ncRDeathDB: A comprehensive bioinformatics resource for deciphering network organization of the ncRNA-mediated cell death system

Abstract

Programmed cell death (PCD) is a critical biological process involved in many important processes, and defects in PCD have been linked with numerous human diseases. In recent years, the protein architecture in different PCD subroutines has been explored, but our understanding of the global network organization of the noncoding RNA (ncRNA)-mediated cell death system is limited and ambiguous. Hence, we developed the comprehensive bioinformatics resource (ncRDeathDB, www.rna-society.org/ncrdeathdb) to archive ncRNA-associated cell death interactions. The current version of ncRDeathDB documents a total of more than 4600 ncRNA-mediated PCD entries in 12 species. ncRDeathDB provides a user-friendly interface to query, browse and manipulate these ncRNA-associated cell death interactions. Furthermore, this resource will help to visualize and navigate current knowledge of the noncoding RNA component of cell death and autophagy, to uncover the generic organizing principles of ncRNA-associated cell death systems, and to generate valuable biological hypotheses.

Keywords:

• autophagy
• cell death
• crosstalk
• database
• ncRNA
• network

Abbreviations

ATG10	=	autophagy-related 10
ATG12	=	autophagy-related 12
ATG16L1	=	autophagy-related 16-like 1
ATG3	=	autophagy-related 3
ATG4B	=	autophagy-related 4B, cysteine peptidase
ATG5	=	autophagy- related 5
BAX	=	BCL2-associated X protein
BCL2	=	B-cell CLL/lymphoma 2
CASP3	=	caspase 3, apoptosis-related cysteine peptidase
lncRNA	=	long non-coding RNA
MAP1LC3B	=	microtubule-associated protein 1 light chain 3 beta
MCL1	=	myeloid cell leukemia 1
miRNA	=	microRNA
ncRNA	=	noncoding RNA
NPI	=	ncRNA-protein interaction
PCD	=	programmed cell death
PPI	=	protein-protein interaction
snoRNA	=	small nucleolar RNA.

Introduction

Noncoding RNAs (ncRNAs) are RNA species that are expressed in cells under certain conditions but do not code for proteins.¹ Currently, microRNAs (miRNAs) are the most studied type of ncRNAs, but novel classes of ncRNAs, such as long noncoding RNAs (lncRNAs) and snoRNAs, continue to be identified, and many studies have begun to focus on the role of ncRNA in programmed cell death.^2-5 Convincing evidence has been found for the involvement of microRNA,^6,7 lncRNA⁸ and snoRNA³ in common cell death subroutines such as apoptosis and autophagy. Despite this progress, unsolved questions remain; for example, while the outline of protein crosstalk between different PCD subroutines is relatively well characterized,^9,10 much less is known concerning the network organization of ncRNA-mediated cell death subroutines and how cell death-associated ncRNAs bridge various cell death subroutines and functional modules. Perhaps more importantly, it is unclear how the ncRNA-mediated cell death system interacts with canonical protein-protein interaction networks (PPI) networks and ncRNA-protein interaction (NPI) networks within processes of global signal transduction. Finally, lack of knowledge regarding pivotal ncRNAs and their protein targets, among the hundreds of ncRNAs found in PCD, hinders the development of ncRNA-based drug targets for clinical use.

To address these issues, we collected a large amount of published data describing the roles of diverse ncRNAs in PCD to develop and update the database ncRDeathDB (www.rna-society.org/ncrdeathdb) according to our previous miRDeathDB¹¹ for the purpose of archiving comprehensive ncRNA-associated cell death interactions. The current version of ncRDeathDB provides an all-inclusive bioinformatics resource on information detailing the ncRNA-mediated cell death system and includes 4615 (including 1817 predicted entries) ncRNA-mediated PCD entries involving 12 species. Furthermore, based on this comprehensive repository, we implemented several bioinformatics network techniques to systematically analyze the hierarchical and topologic characteristics of the ncRNA-mediated cell death system (Fig. 1). Then, we investigated its centrality and its functional role in the global signal transduction cascade by mapping it to the PPI and NPI networks. Finally, we identified several key ncRNAs players in both an experimentally verified autophagy module and a cell death network (including both autophagy and apoptosis modules) that together with their protein targets mediate intra- and inter-module regulation.

Figure 1. Analytical scheme of the ncRNA-mediated cell death system.

Results

Content of the ncRDeathDB database

According to our previous miRDeathDB, we connected the dot to build the miRNA-mediated cell death network.¹² However, with the increasing amount of data demonstrating important functions for ncRNA molecules in apoptosis and autophagy, we continuously curate the literature data and update it. At the same time, we realized the important role of other ncRNA (such as lncRNA and snoRNA) in apoptosis and autophagy. Hence, based on our reference retrieval, we recently updated miRDeathDB¹¹ to version 2.0 (ncRDeathDB, www.rna-society.org/ncrdeathdb), which allows users to browse, search, and analyze PCD-associated ncRNAs. Compared with its first release (miRDeathDB), ncRDeathDB not only enriches ncRNA types by including lncRNA and snoRNA with updated annotations, but also integrates a variety of useful tools for analyzing RNA-RNA and RNA-protein binding sites and for network visualization. The current version of ncRDeathDB documents 4615 (including 1817 predicted entries) ncRNA-associated cell death entries, including 2403 apoptosis-associated entries, 2205 autophagy-associated entries and 7 necrosis-associated entries, leading to a nearly 20-fold content enrichment compared to the initial miRDeathDB release in 2012. These entries were manually annotated from 1495 references, from 12 species, mainly including Homo sapiens, Mus musculus, and Rattus norvegicus (Table 1). Each entry contains detailed information, including annotation of ncRNAs, interacting genes, species, PMID, and detailed description. In the current version of ncRDeathDB, human, mouse, and rat account for 58.07%, 10.40% and 6.50% ncRNA-associated cell death entries, respectively. Notably, the human ncRNA-mediated cell death system consists of 2680 entries, including 2054 apoptosis, 619 autophagy-associated (340 experimentally validated entries and 279 predicted entries) and 7 necrosis-associated entries. Apoptosis-associated entries account for 358 miRNAs, 71 lncRNAs, 6 snoRNA, and 558 proteins, while autophagy-associated entries account for 199miRNAs, 4 lncRNA, and 99 proteins.

Table 1. Statistics of ncRNA-associated cell death entries in ncRDeathDB

	Autophagy		Apoptosis			Necrosis
Organism	miRNA	lncRNA	miRNA	lncRNA	snoRNA	miRNA	lncRNA
H.sapiens	614(199)	5(4)	1838(358)	204(71)	12(6)	4(4)	3(3)
M.musculus	280(128)	0	190(83)	10(9)	0	0	0
M.mulatta	282(99)	0	0	0	0	0	0
P.troglodytes	272(97)	0	0	0	0	0	0
B.taurus	240(94)	0	0	0	0	0	0
R.norvegicus	200(89)	0	100(53)	1(1)	0	0	0
C.familiaris	215(83)	0	0	0	0	0	0
G.gallus	85(44)	0	14(4)	0	0	0	0
D.rerio	9(5)	0	13(10)	0	0	0	0
C.elegans	3(3)	0	1(1)	0	0	0	0
D.melanogaster	1(1)	0	11(7)	5(2)	0	0	0
Xenopus	0	0	2(1)	2(1)	0	0	0

Note: These statistics show the numbers of entries for different cell death subroutines in each species included in ncRDeathDB. The numbers in parentheses are ncRNA counts.

ncRDeathDB also provides 4 options on the “Help” page to provide instructions for using the database. These include “Tutorial:” procedure and illustrations of the database; “Source:” sources of ncRNAs and gene information, and tools used in the database; “Parameter:” details the parameters for tools used for binding site prediction and ‘Error Report’. In the “Download & API” page, users can download all data in Microsoft Excel and TXT format or access the application programming interface (API) using scripts. In the ‘Submit’ page, ncRDeathDB invites users to submit novel ncRNA-associated cell death interactions.

Data querying, searching, browsing, and visualization

ncRDeathDB provides an interface for convenient retrieval of all interactions. Users can search each ncRNA-associated cell death interaction through 3 paths (Fig. 2), including ‘By keyword’: browse the ncRNA-associated cell death entries by inputting the keywords (any ncRNA, Gene Symbol, Death Pathway, Species) and fuzzy search supported, ‘By Death Pathway’: represent all ncRNA-associated cell death entries by associated option of species and death pathway, and ‘By Advanced Search’: represent the accurate search results by associated option of species, death pathway and ncRNA/Gene symbols exist in ncRDeathDB. Brief details of search results are presented as a table in the ‘Results’ page, while more detailed descriptions such as PubMed ID and description of the reference are displayed in the ‘Detail’ page. The ‘Browse page’ enables users to browse the database in 3 different ways: ‘Browse by ncRNA category’, ‘Browse by species’ and ‘Browse by cell death pathway’. All of the ncRNA-associated cell death interaction information is presented by selecting on each entry.

Figure 2. A flowchart for retrieving ncRNA-associated cell death interaction entry. (A) Three searching paths for retrieving the ncRNA-mediated cell death system. (B) The result of a representative database entry. (C) The detailed information for an ncRNA-associated cell death entry. In the results and detail pages, ncRDeathDB linked each ncRNA/protein to their corresponding detail pages and PMID to NCBI PubMed database.

Topological characterization of the human ncRNA-mediated cell death system

In the current version of ncRDeathDB, nearly 60% of the entries included were for human ncRNA-associated cell death subroutines. Hence, in this work, we mainly focus on the largest human ncRNA-mediated cell death system by using experimentally validated entries to investigate the generic organizing principles of ncRNA-mediated cell death systems. According to experimentally validated entries, the human ncRNA-mediated cell death system consists of 1715 interactions and 951 nodes, including 1453 apoptosis and 262 autophagy-associated interactions. Using several bioinformatics network techniques, we first identified the hub and bottleneck nodes of the network (Fig. 3, Table S1). Typical examples, such as the hub nodes BCL2 and MIR21 and the bottleneck node MIR181A (reported in our previous work,^10,12) were validated for the current network. In addition, we have identified several newly discovered partner genes of MIR181A, which include apoptotic genes (CASP3, CASP9, BAX, FAS, etc) and autophagic genes (ATG10, BECN1, ATG16L1, ATG12, UVRAG, etc). This finding has further strengthened the role of MIR181A in mediating the crosstalk between apoptosis and autophagy.¹²

Figure 3. Schematic diagram for network measurements used to decipher the ncRNA-mediated cell death system. (A) A schematic diagram represents one bottleneck node, 2 hub nodes and one pivot node. Hubs are highly connected to other nodes in the network and act as the center point within the modules or subnetworks. Evidence from model organisms indicates that hub proteins tend to be encoded by essential genes, and the deletion of hub nodes leads to amount of diseases. Bottlenecks are defined as proteins with a high betweenness that are central to many paths in the network, and play a key role in controlling network signaling, and also like to be essential proteins, and often function as key genes in the development of disease progression. Pivots correspond to significantly shared members of 2 different pathways or subnetworks, which potentially have a role as molecular switches between these pathways or subnetworks, and further sheds light on the organization of the cellular machinery. (B) Schematic description of the 3-core decomposition of a network. k-core is a subnetwork defined by iteratively removing the nodes with a degree lower than k and their incident links. The inner k-cores were shown to be global hubs and enriched with lethal genes, which may potentially act as the evolutionarily conserved backbone of ncRNA-mediated cell death system.

Furthermore, novel hubs have been identified, including MCL1, a member of the BCL2 family located in mitochondria, the different isoforms of which play a role in the regulation of apoptosis.^13,14 In the human ncRNA-associated cell death system, MCL1 interacts with 25 ncRNAs, including important hub ncRNAs such as MIR181A and MIR125B. TP53 is another interesting hub; in addition to interacting with several miRNA partners, it interacts with multiple lncRNAs, including H19, PANDAR, LINC-ROR and MEG3, suggesting that it plays a key role in linking lncRNAs with other ncRNAs.^15-18 Furthermore, we identified several novel ncRNAs as bottlenecks that bridge different cell death subroutines (Table S1). For example, MIR34A not only regulates key apoptotic genes such as BCL2, TP53, and members of the CDK family, but also interacts with ATG9, which is involved in autophagy.^19-22

Hierarchy of human ncRNA-mediated cell death system

In nature, hierarchical network organization has been found in both prokaryotic and eukaryotic regulatory networks.^23-25 Hence, we applied the k-core algorithm to decipher whether the human ncRNA-mediated cell death system is also hierarchically ordered (Fig. 3). This network is parsed into 5 detailed layers, and the nodes contained in each layer are 951 (layer 1), 394 (layer 2), 225 (layer 3), 150 (layer 4) and 73 (layer 5). Several established key cell death protein clusters, such as the autophagy-associated ‘ATG10 ATG7 ATG5 ATG12 ATG3 MAP1LC3B ATG4B ATG16L1’ cluster and the apoptosis-associated ‘MOAP1 BCL2L1 BCL2 BAX PMAIP1 MCL1 BID BBC3 BIK BAK1 BMF ERN1’ cluster appear in the innermost layer (4 and 5-layer), suggesting that the components in this layer are essential for normal cellular function.

To investigate the role of ncRNAs in the cell death system, we determined the fraction of ncRNAs at each layer, 345/951 (36.28%) in layer 1, 193/394 (48.98%) in layer 2, 127/225 (56.44%) in layer 3, 85/150 (56.67%) in layer 4, and 40/73 (54.79%) in layer 5. Some ncRNAs and proteins can interact with both canonical apoptotic and autophagy partners. Such dual functional molecules may play a signal transmission role in the cell death system. Interestingly, our results have shown that the number of dual ncRNAs and proteins within each layer is 158/951 (16.61%) in layer 1, 128/394 (32.49%) in layer 2, 94/225 (41.78%) in layer 3, 71/150 (47.33%) in layer 4, and 38/73 (52.05%) in layer 5, and that the percentage of dual ncRNAs and proteins increases dramatically with the k-value. Due to the low statistical power of detecting significant differences in the relatively rareness nodes (small samples) in 4- and 5-layer, thus, we define layer 1 as the outmost layer, the combined layers 2 and 3 as the middle layer and layers 4 and 5 as the core layer. According to hypergeometric test, our results showed that the percentage of dual ncRNAs and proteins increases significantly with an increasing k-value in the outmost layer vs. the middle layer (P < 0.01) and the middle layer vs. the core layer (P < 0.01),suggesting that many dual ncRNAs and proteins are located in the core layer of the cell death system and therefore play a vital role. These data also validated the existence of a tight crosstalk between apoptosis and autophagy at the molecular level. Furthermore, this crosstalk was mainly executed by the backbone of the cell death system rather than other molecules playing auxiliary roles.

The ncRNA-mediated cell death system in human protein-protein and ncRNA-protein networks

To investigate the biological significance of the ncRNA-mediated cell death system in the context of the global human regulation network, we mapped cell death-associated ncRNAs and proteins to the ncRNA-protein interaction (NPI) network and the human protein-protein interaction (PPI) network. Fig. 4 represents the degree distribution of cell death-associated ncRNAs and proteins in both the NPI and PPI networks. It is clear that the cell death-associated ncRNAs and proteins especially the dual nodes have a significantly higher degree distribution (P < 0.01, hypergeometric test) than that of the whole nodes in the NPI and PPI networks, suggesting that cell death-associated ncRNAs and proteins act as hubs in these global networks. Dual ncRNAs and proteins have the highest degree distribution, reflective of not only the key role played by dual functional molecules in the crosstalk between cell death subroutines, but also their roles as core elements in the human PPI and NPI network and as the foundation of signal transduction.

Figure 4. Distribution of degrees of apoptosis and autophagy-associated ncRNAs and proteins in the global human NPI and PPI network. The boxplot represents the degree distribution. The cell death-associated ncRNAs and proteins especially the dual nodes have a significantly higher degree distribution than that of the whole nodes in the NPI and PPI network.

Global human NPI and PPI networks are also hierarchically structured.^26-28 As shown in Figure 5, our results show that the percent of cell death-associated ncRNAs and proteins at each layer increased as k increased, and the dual proteins and ncRNAs involved in both autophagy and apoptosis increased the fastest. These results suggest that cell death-associated ncRNAs and proteins, especially dual-functional ncRNAs and proteins, mainly localized to the core layer of both the NPI and PPI networks. In summary, our results demonstrated that cell death-associated ncRNAs and proteins were involved in the basic construction of NPI and PPI networks.

Figure 5. Analysis of network centricity of cell death-associated ncRNA and protein in global the human NPI and PPI network. The percent of both cell death-associated ncRNAs and proteins in k-cores of the NPI and PPI network significantly increased as k increased, and the dual proteins and ncRNAs involved in both autophagy and apoptosis increased the fastest.

Identifying the pivot ncRNAs regulating the human autophagy subnetwork

One recent study extracted a high-confidence candidate “Autophagy-Interaction Network” (AIN) using the Comparative Proteomics Analysis Software Suite (CompPASS).²⁹ This protein autophagy network, which is composed of 409 candidate interacting proteins with 751 interactions and 8 subnetworks, has provided a resource for the mechanistic analysis of autophagy. Because of the sparseness of autophagy-associated genes with defined regulation by ncRNAs, some of genes defined in these subnetworks cannot be found in our data. However, several important autophagy subnetworks, such as the ULK1 kinase network, the Human network of orthologs and paralogs of yeast Atg8, the UBL conjugation system and the ATG2-WIPI subnetwork, can be mapped to our ncRNA-mediated cell death system. To identify the ncRNAs regulating these subnetworks, we performed a hypergeometric test for each subnetwork and identified 13 miRNAs and 1 lncRNA that have significant P values (Table 2). A total of 9 miRNAs (MIR519A, MIR374A, MIR20A, MIR630, MIR30C1, MIR181A, MIR130A, MIR30D, and MIRLET7A1) tightly control the UBL conjugation system through regulating some autophagic genes, such as ATG5, ATG12, and ATG16L1. Similarly, 3 miRNAs (MIR26B, MIR30A and MIR335) tightly control the human network of orthologs and paralogs of yeast Atg8, through regulating some autophagic genes, such as MAP1LC3B. More interesting, BANCR as a lncRNA was also involved in this network by targeting the MAP1LC3A and MAP1LC3B.

Table 2. ncRNAs are significant for the regulation of human protein autophagy subnetworks

Autophagy subnetwork	ncRNA	overlap protein	P-value
Human proteins related to yeast Atg8	MIR335	LC3B,SQSTM1, LC3A,GABARAPL1	4.11E-06
UBL conjugation system	MIR519A	ATG16L1,ATG12,ATG5	4.96E-05
UBL conjugation system	MIR374A	ATG12,ATG5,ATG16L1	1.17E-04
UBL conjugation system	MIR20A	ATG16L1,ATG12,ATG5,ATG7	2.09E-04
UBL conjugation system	MIR630	ATG12,ATG5,ATG16L1	2.26E-04
Human proteins related to yeast Atg8	MIR26B	MAP1LC3C,GABARAP	6.61E-04
UBL conjugation system	MIR30C	ATG16L1,ATG5	8.78E-04
UBL conjugation system	MIR181A	ATG12,ATG5,ATG16L1	8.87E-04
Human proteins related to yeast Atg8	BANCR	MAP1LC3B,MAP1LC3A	3.80E-03
UBL conjugation system	MIR130A	ATG5,ATG16L1	3.94E-03
UBL conjugation system	MIR30D	ATG12,ATG5	3.94E-03
ULK1 kinase network	MIR885	ULK2,ATG13	4.14E-03
Human proteins related to yeast Atg8	MIR30A	MAP1LC3B,GABARAPL2	4.72E-03
UBL conjugation system	MIRLET7A1	ATG5,ATG7	6.20E-03

Identifying pivot ncRNAs that regulate the human cell death modules

To extract pivot ncRNAs from the global cell death network (Fig. 3), the human cell death protein interaction network was extracted based on the PPI network, and both nodes of each interaction were documented in ncRDeathDB, consisting of 365 proteins and 1172 edges. Then, we used the ClusterONE plugin in Cytoscape to delineate potential modules in the cell death protein interaction network and identified 17 such modules (Table S2). Some of these modules indeed coincided with several experimentally validated protein complexes.³⁰ For example, the module ‘FASN FADD CASP8AP2 CASP8 TNFRSF10B FAF1 PEA15’ significantly contains the protein complex ‘Death-inducing signaling complex DISC associated with type I cells’ and the ‘FAS-FADD-CASP8-CASP10 complex’ (P<0.01, hypergeometric test), while the module ‘MOAP1 BCL2L1 BCL2 BAX PMAIP1 MCL1 BID BBC3 BIK BAK1 BMF ERN1’ mostly consisted of BCL2 family proteins, which play an influential role in both apoptosis and autophagy. Furthermore, some of the modules are highly consistent with the subnetworks defined in the ‘Autophagy-Interaction Network’. For example, the protein cluster “ATG10 ATG7 ATG5 ATG12 ATG3 MAP1LC3B ATG4B ATG16L1” is part of the ‘UBL conjugation system’ defined in AIN (P<0.01, hypergeometric test). All of these lines of evidence indicate that the modules identified by ClusterONE overlap significantly with existing protein complexes in the cell death subnetwork. Based on this module construction, we also uncovered a panel of ncRNAs that are extensively linked with these modules via a hypergeometric test (Table 3, Fig. 6 and Table S3).Our results showed that these ncRNAs can easily be classified into 2 categories, one of which is located within a given module and regulates most of the proteins in the module. For example, MDM2-AS, MIR197 and MIR320A only target the BCL2-associated module, while MIR374A, MIR30A and MIR519A only target the module ‘ATG10 ATG7 ATG5 ATG12 ATG3 MAP1LC3B ATG4B ATG16L1’. Another category of ncRNAs acts as a bottleneck to bridging the different modules. The most significant discovery is that MIR181A simultaneously regulate modules ‘ATG10 ATG7 ATG5 ATG12 ATG3 MAP1LC3B ATG4B ATG16L1’ and ‘MOAP1 BCL2L1 BCL2 BAX PMAIP1 MCL1 BID BBC3 BIK BAK1 BMF ERN1’; while, the first module as the member of ‘UBL conjugation system,’ plays an important role in regulating autophagy, the latter consists of key proteins involved in apoptosis, such as BCL2, MCL1 and BIK. What's more, MIR181A is located in the centermost subnetwork of the cell death network, suggesting that MIR181A plays a critical role in the regulation of cell death by functioning in both apoptosis and autophagy.

Figure 6. Example of a module significantly regulated by ncRNAs identified in the human cell death protein interaction network. Pink and blue nodes represent apoptosis and autophagy, respectively. Yellow nodes represent the ncRNAs and proteins that play dual roles in apoptosis and autophagy. Triangular, arrowhead-shaped and circular nodes represent cell death-associated miRNAs, lncRNAs and proteins, respectively.

Table 3. Statistics of top 10 significant ncRNAs regulating human protein cell death modules

Protein cell death module members	ncRNA	P-value
BCL2L1 BCL2 CASP3 CTNNB1 CASP8 PSEN1 TFAP2A APAF1	MIR133A	3.57E-08
BCL2L1 BCL2 CASP3 CTNNB1 CASP8 PSEN1 TFAP2A APAF1	MIR548	6.01E-08
BCL2L1 BCL2 CASP3 CTNNB1 CASP8 PSEN1 TFAP2A APAF1	MALAT1	6.09E-08
ATG10 ATG7 ATG5 ATG12 ATG3 MAP1LC3B ATG4B ATG16L1	MIR519A	4.15E-07
ATG10 ATG7 ATG5 ATG12 ATG3 MAP1LC3B ATG4B ATG16L1	MIR374A	1.47E-06
ATG10 ATG7 ATG5 ATG12 ATG3 MAP1LC3B ATG4B ATG16L1	MIR20A	2.45E-06
ATG10 ATG7 ATG5 ATG12 ATG3 MAP1LC3B ATG4B ATG16L1	MIR630	3.81E-06
MOAP1 BCL2L1 BCL2 BAX PMAIP1 MCL1 BID BBC3 BIK BAK1 BMF ERN1	MIR197	5.79E-06
BCL2L1 BCL2 CASP3 CTNNB1 CASP8 PSEN1 TFAP2A APAF1	HIF1A-AS1	5.84E-06
BCL2L1 BCL2 CASP3 CTNNB1 CASP8 PSEN1 TFAP2A APAF1	MIRLET7C	1.45E-05

Finally, significant regulatory relationships between ncRNAs and their target modules are shown in Table 3. Apart from miRNA modulators, the novel lncRNA MALAT1 was also found to play an important role in cell death. It targets the module “MOAP1 BCL2L1 BCL2 BAX PMAIP1 MCL1 BID BBC3 BIK BAK1 BMF ERN1” and module “BCL2L1 BCL2 CASP3 CTNNB1 CASP8 PSEN1 TFAP2A APAF1,” which have a high overlap with the apoptotic protein complex ‘MCL1-BAK1 complex’ and “MRIT complex” respectively (P<0.01, hypergeometric test) (Table 3, Fig. 6 and Table S3). These ncRNAs seem to act as messengers or coordinators in mediating the crosstalk between the different subroutines of PCD.

Discussion

In recent years, great efforts have been made to uncover new forms of mammalian programmed cell death.^1,31-34 Emerging work has focused on the network organization of cell death systems with the ultimate goal of exploring potential organization mechanisms from a global view of the Protein-Protein Interaction (PPI) network.^29,35 Consequently, accumulated ncRNAs and proteins have been identified in a variety of cell death subroutines.^10,29,36-38 It is now widely accepted that both ncRNAs and proteins are intimately connected with each other in a global cell death network. Of course, the PPI network in different cell death subroutines is only part of the story. Hence, 2 and half y ago, we established the miRDeathDB database to archive these resources and provide a basis for the analysis of miRNA-mediated cell death systems.^11,12 With the rapid development of this field, this is the perfect time to integrate recent findings and update information. We therefore developed ncRDeathDB (www.rna-society.org/ncrdeathdb), which documents by nearly 20-fold the data of the previous miRDeathDB database and includes diverse ncRNA-mediated cell death interactions, such as those involving miRNA, lncRNA, and snoRNA. ncRDeathDB is a good starting point for searching the web resources centered on diverse ncRNAs involved in PCD. More importantly, ncRDeathDB provides a valuable resource to manipulate, visualize, and analyze these ncRNA-associated cell death interactions. By integrating the ncRNA-associated cell death interactions into a global network, this resource will help to visualize and navigate current knowledge of the noncoding RNA component of cell death and autophagy, to uncover the generic organizing principles of ncRNA-associated cell death systems, and to generate valuable biological hypothesis. In ncRDeathDB, we present a systematic ncRNA-associated analysis of the human cell death systems by using experimentally validated entries. Such analysis has provided another view of the global organization of the ncRNA-associated cell death system and the potential mechanisms involved in the cell death process. First, we presented a network of 1715 human ncRNA-associated interactions (1453 apoptosis- and 262 autophagy-associated interactions) and 951 nodes (300 miRNAs, 40 lncRNAs, 1 snoRNA and 610 proteins) with extensive connectivity among different modules.Similarly, our previously reported hubs (BCL2 and MIR21) and bottleneck (MIR181A) were validated in this extended network with newly discovered partner ncRNAs and proteins, thus strengthening their roles in mediating the crosstalk between apoptosis and autophagy. More importantly, novel key hubs and bottlenecks have been explored that interact with diverse and important ncRNAs, especially multiple lncRNAs, to regulate communication with intra- and inter- autophagic and apoptotic modules. Hence, these diverse ncRNA dependent mechanisms may represent a basic mechanism for the conversation between autophagic and apoptotic death modules, leading to cellular homeostasis upon the presentation of external stimuli.

Next, to investigate the global center of ncRNAs and proteins involved in PCD, the ncRNA-associated cell death system was hierarchically divided into 5 fine layers using the k-core algorithm. Our results explored the idea that both dual functional ncRNAs and proteins tended to reside in the most inner layer of this system and that these molecules were tightly connected with each other and acted as the evolutionary backbone of the proteome. Furthermore, we investigated the relationship between the ncRNA-mediated cell death system and the canonical signal transduction process. For both protein-protein (PPI) and ncRNA-protein (NPI) interaction networks, our results demonstrated that the number of cell death-associated ncRNAs and proteins was significantly larger in the inner layer than in the outer layer. Dual proteins and ncRNAs involved in both autophagy and apoptosis displayed this tendency, suggesting that cell death-associated ncRNAs and proteins were encoded into the basic function of cells. Under stress or other urgent conditions, these dual proteins and ncRNAs could reprogram cell fate by choosing different apoptosis or autophagy pathways to adapt to the environment. Finally, in an experimentally verified autophagy module, we observed that some cell death-associated ncRNAs tightly control different autophagy subnetworks by regulating shared key genes. Similarly, in a cell death network (including both autophagy and apoptosis modules), our results identified several key ncRNA players and their protein targets, which mediate communication between different cell death modules. The aforementioned ncRNA-based intra- and inter-modulators are likely to be good candidates for pharmaceutical development in the near future.

In summary, our studies provide insight into the organization of the human ncRNA-mediated cell death system and decipher diverse ncRNA-dependent mechanisms as a basic component of the crosstalk between apoptosis and autophagy. In the future, with the identification of more necrosis-associated ncRNA, our database will serve as a more comprehensive ncRNA-associated resource for further mechanistic analysis of programmed cell death.

Materials and Methods

Data collection and web interface implementation

To collect all available ncRNA-associated entries involved in programmed cell death, we have integrated information on 3 major types of ncRNAs: long noncoding RNA (lncRNA) symbols from lncRNAdb³⁹ and Ensembl,⁴⁰ miRNA symbols from mirBase database,⁴¹ and small nucleolar RNA (snoRNA) symbols from sno/scaRNAbase⁴² and snoRNA-LBME-db.⁴³ Because the research on other categories of ncRNAs is still in its infancy and there is no unified nomenclature, we instead use these ncRNA category names to replace specific ncRNA symbols. All abstracts and articles in the PubMed database were screened using the following keyword combinations: (each ncRNA symbol or ncRNA category) and (each cell death subroutines); for example, “malat1 and apoptosis.” The relevant hits were downloaded and prepared systematically for data curation by further manual inspection. Apart from the information regarding ncRNAs and the corresponding interacting genes, species and detailed description compiled from the references, we incorporate several useful tools to predict the potential binding sites of these ncRNA-associated cell death interactions. For ncRNA-RNA interactions, the binding sites and scores are predicted according to miRanda⁴⁴ and RISearch.⁴⁵ For ncRNA-protein interactions, RNA-binding residues and scores are computed by bindN+⁴⁶ and RNAbindR.⁴⁷ In addition, to help users interactively analyze the ncRNA-mediated cell death system online, ncRDeathDB provides an embedded Cytoscape web tool (cytoscapeweb.cytoscape.org/).

The ncRDeathDB database is implemented in the HTML and PHP languages with a MySQL server. The interface component consists of webpages designed and implemented in HTML/CSS in a Microsoft Windows environment. It has been tested in Google Chrome, Firefox and Internet Explorer web browsers.

Autophagy-associated entry prediction

Autophagy-associated miRNAs and their protein targets were predicted by using homologs search based on the experimentally validated autophagy-associated entries in ncRDeathDB. The conserved miRNAs families were identified by TargetScan.⁴⁸ The homological genes of human autophagy-associated genes were downloaded from NCBI.⁴⁹ Then, the interactions between conserved miRNAs and homological genes verified by TargetScan^48,50,51 were filtered as the potential autophagy-associated miRNA-protein pairs and assigned for other species.

Protein-protein interaction network and RNA-associated network construction

In this work, the global human protein-protein interaction(PPI) network was retrieved from the Human Protein Reference Database (HPRD),⁵² and the human ncRNA protein interaction network(NPI) was retrieved from our RNA-Associated Interaction Database (RAID)⁵³ and miRTarBase,⁵⁴ in which the interactions have been validated by experiments and covered diverse ncRNA species, such as lncRNA and miRNA.

Degree and betweenness analysis of the human interaction network

The network was analyzed based on topological parameters such as betweenness and degree using a Cytoscape plug-in called ‘Network Analyzer.’⁵⁵ In a biological network, degree indicates the number of edges linked to a given node (Fig. 3). Betweenness measures the total number of nonredundant shortest paths going through a certain node or an edge, reflecting the bottleneck of the network (Fig. 3) The nodes with a top 5% highest degree and betweenness were chosen as the hubs and bottlenecks in this work (Table S1). The hub and bottleneck nodes in the network usually play a key role in connecting the intra- and inter-modules or pathways.⁵⁶

K-core analysis of interaction network

The k-core of a network is defined as the subnetwork in which all of its node degrees are greater than k after recursively removing the nodes with a smaller degree than k (Fig. 3).²⁸ One can iteratively apply this ‘peeling’ procedure to divide the whole network into different layers, and the genes in layer k of the network are defined as the intersection between core k and core k+1. In this work, the human ncRNA-mediated cell death system has been parsed into 5 layers according to this ‘peeling’ procedure.

To measure the centrality of cell death-associated ncRNAs and proteins, the Excess Retention (ER) was calculated for each k-core layer of the NPI and PPI networks. ER is defined as the proportion of genes with a certain property A in the ‘k-core’ compared with that in the whole gene groups.⁵⁷ For instance, in a k-core with a total of N_k nodes, the proportion of protein with a property A is defined as ${\displaystyle {\text{er}}_{\text{k}}^{\text{A}}={\text{n}}_{\text{k}}^{\text{A}}/{\text{N}}_{\text{k}}}$ Obviously, the fraction of proteins with property A in the whole network with a total N nodes is defined aser^A = n^A/N, and the ER of k-core is defined as ${\displaystyle {\text{ER}}_{\text{k}}^{\text{A}}={\text{er}}_{\text{k}}^{\text{A}}/{\text{er}}^{\text{A}}}$. In this work, we set the cell death-associated ncRNAs and proteins as the property A in ER, and then counted the percent of cell death-associated ncRNAs and proteins in each layer of the NPI and PPI networks. Thus, the ER value reflects the centrality of cell death-associated ncRNAs and proteins in human PPI and NPI networks.

Module identification and pivot ncRNA recognition

In this work, the human cell death protein interaction network was extracted based on the PPI network, all nodes of which were documented in ncRDeathDB. To identify functional modules in the human cell death protein interaction network, the network was clustered with the ClusterONE algorithms⁵⁸ available in Cytoscape.⁵⁹ ClusterONE is a method of detecting the protein complex module based on the PPI network by iteratively adding or removing the most valuable incident and boundary nodes to locally optimize the cluster quality measure cohesiveness.⁵⁸

The pivot ncRNAs (Fig. 3), which significantly regulate different modules and subnetworks, are recognized by a hypergeometric test.⁶⁰ The P value of the test statistic is calculated as:${\displaystyle p=1-{\displaystyle \sum }_{M-1}^{i=0}\frac{\left(\begin{array}{l}M\\ i\end{array}\right)\left(\begin{array}{l}N-\text{M}\\ n-i\end{array}\right)}{\left(\begin{array}{l}N\\ n\end{array}\right)}}$

Where N is the total number of proteins contained in the human cell death protein interaction network, n represents the number of interacted genes of a certain ncRNA. M represents the number of genes belonging to a module, while i represents the genes in the modules interacted with ncRNA.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Acknowledgment

We are grateful to Dr. Daniel J. Klionsky for help in editing the manuscript.

Supplemental Material

Supplemental data for this article can be accessed on the publisher's website

Funding

This work was supported by the National High Technology Research and Development Program of China (2014AA021102), the Major State Basic Research Development Program of China (2014CB910504), the National Natural Science Foundation of China (U1304311), the Natural Science Foundation of Heilongjiang Province of China (C2015027), the Scientific Research Fund of Heilongjiang Provincial Education Department (12541426).