Identification of novel type 2 diabetes candidate genes involved in the crosstalk between the mitochondrial and the insulin signaling systems.

Type 2 Diabetes (T2D) is a highly prevalent chronic metabolic disease with strong co-morbidity with obesity and cardiovascular diseases. There is growing evidence supporting the notion that a crosstalk between mitochondria and the insulin signaling cascade could be involved in the etiology of T2D and insulin resistance. In this study we investigated the molecular basis of this crosstalk by using systems biology approaches. We combined, filtered, and interrogated different types of functional interaction data, such as direct protein-protein interactions, co-expression analyses, and metabolic and signaling dependencies. As a result, we constructed the mitochondria-insulin (MITIN) network, which highlights 286 genes as candidate functional linkers between these two systems. The results of internal gene expression analysis of three independent experimental models of mitochondria and insulin signaling perturbations further support the connecting roles of these genes. In addition, we further assessed whether these genes are involved in the etiology of T2D using the genome-wide association study meta-analysis from the DIAGRAM consortium, involving 8,130 T2D cases and 38,987 controls. We found modest enrichment of genes associated with T2D amongst our linker genes (p = 0.0549), including three already validated T2D SNPs and 15 additional SNPs, which, when combined, were collectively associated to increased fasting glucose levels according to MAGIC genome wide meta-analysis (p = 8.12×10(-5)). This study highlights the potential of combining systems biology, experimental, and genome-wide association data mining for identifying novel genes and related variants that increase vulnerability to complex diseases.

Introduction

Insulin resistance is a common trait present in complex disorders such as type 2 diabetes (T2D), obesity or metabolic syndrome (MetS). Around 340 million people suffer from diabetes worldwide, 90% of whom have T2D (http://www.who.int/diabetes/facts/en). Unlike type 1 diabetes, overt T2D is usually diagnosed several years after its onset due to its milder presenting symptoms, which in part explains why several devastating complications such as cardiovascular related diseases tend to develop soon after or have already arisen at the moment of the initial diagnosis.

There has been growing interest in identifying genes and processes that could trigger insulin resistance beyond defects on the insulin signaling cascade itself. As a result, defective mitochondrial activity has been indirectly related to insulin resistance in insulin-targeted tissues, such as skeletal muscle [1], [2], [3] and liver [4]. In particular, patients with T2D and, more importantly, non-diabetic subjects with type 2 diabetic relatives showed mitochondrial dysfunction and lower expression of PPAR gamma co-activator 1 alpha and 1 beta (PGC-1α and PGC1-1β), which are key regulators of mitochondrial biogenesis and function. In addition, subjects with early-onset type 2 diabetes typically show defective activation of PGC-1alpha in response to physical activity [5], and similarly, morbid obese type 2 diabetic patients show a defective activation of mitochondrial gene expression in response to weight-loss surgery [5]. Whether there is a heritable component involved in the alterations in expression of mitochondrial genes/proteins in these common forms of T2D remains to be determined.

Despite all of these efforts and lines of evidence, the mechanisms and the molecular contributors to the connection between mitochondria and the insulin signaling and resistance are still unknown. The availability of a wide range of functional interaction data, including metabolomics, genomics, transcriptomics and proteomics and the integration of all these data using systems biology approaches make it now possible to investigate in detail the molecular basis of the interaction between the insulin signaling cascade and mitochondrial biology in healthy and pathological scenarios, particularly in the context of T2D.

In addition, and despite substantial progress achieved in the identification of candidate genes involved in specific complex processes or diseases through genome-wide association studies (GWAS), for most diseases, including T2D, less than 10% of the heritability (percentage of variance attributable to genetic variation) can be explained by the identified genetic associations [6]. Some hypotheses suggest that a portion of the missing heritability stays behind multiple small effect size variants that have not yet reached genome-wide significance in GWAS meta-analyses when tested individually, due to insufficient sample sizes. If many of the modest effect variants are assumed to implicate genes that function in a limited number of biological processes, collective analysis of variants based on prior biological knowledge could substantially enhance association detection power. In that sense, the application of systems biology approaches to analyze GWAS data may have the potential to increase the chances of unraveling susceptibility genes or biological processes for complex diseases.

In this study, we applied systems biology approaches to screen and identify novel candidate T2D genes. The search has been guided by the hypothesis that the functional components of the crosstalk between the insulin signaling pathway and the biology of the mitochondria may play a role in the etiology or the evolution of the disease. We have also generated and analyzed gene expression data on insulin resistance and mitochondria perturbed scenarios to support these candidate genes. We finally tested whether particular genetic variants in loci that contain the identified genes could be collectively associated with T2D.

Results

Generation of the MITIN network

In order to identify genes specifically involved in the crosstalk between the insulin signaling pathway and the mitochondria, we looked for all possible direct and indirect functional interactions between mitochondria and insulin signaling genes (Figure 1). We started by building reliable models and parts lists for these two systems. We first explored and manually filtered several public versions of the insulin signaling pathway to end up with a confident collection of 197 proteins/genes (see Methods). At the same time, we extracted data from a database of nuclear and mitochondrial-encoded mitochondrial proteins (MitoP2) to generate the corresponding list of 682 mitochondria genes [7].

Figure 1

Schematic flow chart of the generation and evaluation of the MITIN network.

The different sources of functional interaction are combined to generate a functional interactome. The resulting network is used to identify the direct and indirect interactions between the insulin signaling and mitochondria systems. The relevance of the MITIN network is tested analyzing gene expression data of models perturbing either insulin signaling or mitochondria function, and testing the variability within or near the MITIN network genes using GWA meta-analyses from DIAGRAM consortium. *In all PPIhigh and PPIcorr, both pair of interacting proteins have to be simultaneously expressed in any of the insulin-targeted tissues (adipose tissue, muscle, liver and heart).

Once both parts lists were constructed, we screened several large functional interaction databases to identify direct and indirect connections involving any of the protein/genes of each of the systems. We applied several filters and cutoffs to be able to isolate, from all available interactions, a reliable collection that will be used further in our study. For example, from protein-protein interaction (PPI) data, we only considered those protein pairs whose interactions were reported by two or more independent laboratories (PPIhigh) and whose pair of genes were reported to be expressed both in any of the insulin-sensitive tissues (adipose tissue, muscle, liver and heart, [8]); or any other PPI interaction reported only by a single laboratory, simultaneously expressed in any of the insulin-sensitive tissues and that also showed co-expression (gene-expression correlation) in a dataset of 427 healthy human liver samples [9] (these interactions are here termed PPIcorr). As a third layer of functional interaction, we also linked those proteins observed to belong to the same protein complex as described in the CORUM protein complex database [10]. The fourth source of interaction consisted of pairs of genes coding for enzymes that participate in linked metabolic reactions, i.e. those reactions that are adjacent in a metabolic reaction map according to the Biochemical Genetic and Genomic (BiGG_met) and the Kyoto Encyclopedia of Genes and Genomes database (http://www.genome.jp/kegg/kegg2.html; KEGG_met) [11], [12], [13]. Finally, we also included those interactions between genes coding for complexes or genes linked in a signaling pathway, as defined by KEGG (KEGG_path) [12]. This final functional interactome comprised 57,751 high confidence functional interactions involving 6963 genes, which represent a whole functional network of insulin-targeted tissues or cells.

From the pool of selected high quality interactions (affecting 6963 genes), we finally selected those interactions that, either directly or indirectly, provide a link between the mitochondrial and the insulin signaling cascade genes. We defined indirect interactions as those mediated by genes, termed linker or internode genes, that do not belong to either the insulin or the mitochondria parts list, but that are simultaneously connected to both systems. By applying these filters, we finally generated the mitochondria-insulin (MITIN) network consisting of 886 genes and a total of 1259 interactions, 70 direct (Table S1) and 1194 indirect. The 70 direct interactions involved 44 insulin genes and 37 mitochondria genes, most of them showing only one evidence of interaction. Both the insulin and mitochondria genes that were directly connected were linked to a median of two genes from the other system. Direct connections showed heterogeneous sources of interaction: PPIhigh, PPIcorr, Corum Complexes, BiGG_met, KEGG_met, Kegg_pathway, contributed 41, 9, 13, 2, 3, 12 links, respectively. Indirect interactions involved 286 linker internode genes (Figure S1, Dataset S1, Table S2 and S3). These internodes genes were connected to a mean number of 2.1 Insulin and 1.7 mitochondria genes and showed a mean of 2.6 and 2.0 lines of evidence of interaction with insulin and mitochondria, respectively. Regarding the 1194 indirect connections, PPI, PPIcorr, Corum Complexes, BiGG_met,KEGG_met, Kegg_pathway, contributed 570, 472, 1263, 42, 160, 169 interactions, respectively.

While the majority of the internode genes seem to be novel, as their bridging role connecting both systems has not yet been described, some of them have already been shown to interact with both systems, which constitutes an internal positive control of our underlying search methodology. For example, TRAF2 shows interactions within our MITIN network with four insulin and two mitochondrial genes (Table 1). Interestingly, other independent studies and approaches also identified five of these interactions. In particular with MAP3K1 (MEKK1), CAV1 (caveolin-1) and MTOR (mTOR), from the insulin signaling [14], [15], [16] and MAP3K5 (ASK1) and CASP8 (caspase-8) from the mitochondria [17], [18] (Figure 2). Another example is NFKB1, for which we found interactions with four insulin signaling and three mitochondrial genes. As above, NFKB1 has been also reported to interact with the IKBKB [19], [20], AKT2 [21], MAP3K1 [22] and SOCS3 insulin genes, as well as to BCL2L1 [22] [23] and BCL2 [24] (Figure 2).

Table 1

Strong candidates linking both insulin and mitochondria genes.

Internode gene	Total links	Total evidences	Insulin associated genes	Evidences linking to insulin signaling	Mitochondria associated genes	Evidences linking to mitochondria	Insulin functional associations	Mitochondria functional associations
ABL1	12	21	10	18	2	3	YWHAH | CBL | SRC | CRK | BCR | PTK2 | CRKL | GRB2 | CSK | NCK1	WASF1 | TP53
ALDOA	6	8	2	4	4	4	FBP1 | FBP2	ALDH1B1 | ALDH1A3 | ALDH2 | GPD2
ALDOB	5	7	2	4	3	3	FBP1 | FBP2	ALDH1A3 | ALDH1B1 | ALDH2
ALDOC	5	7	2	4	3	3	FBP2 | FBP1	ALDH1B1 |ALDH1A3 | ALDH2
DHX9	5	6	2	3	3	3	NOLC1 | RPS6	BRCA1 | SLC25A5 | TUFM
HNRNPU	6	6	3	3	3	3	NOLC1 | YWHAQ | RPS6	C1QBP | SLC25A5 | TUFM
HSP90AA1	13	16	9	11	4	5	CALM1 | YWHAG | YWHAH | IKBKB | SRC | YWHAB | PDPK1 | YWHAQ | AKT1	TP53 | TOMM34 | TOMM70A | HSPD1
ILF3	6	6	3	3	3	3	NOLC1 | RPS6 | YWHAQ	SLC25A5 | C1QBP | TUFM
MYBBP1A	6	8	3	4	3	4	PKLR | RPS6 | NOLC1	POLG | SLC25A5 | TUFM
NCL	7	10	4	6	3	4	RPS6 | PRKCZ | PPARGC1A | NOLC1	TUFM | TP53 | SLC25A5
NFKB1 #	7	9	4	6	3	3	IKBKB | SOCS3 | MAP3K1 |AKT2	BCL2L1 | BCL2 |MTIF2
NFKBIA	5	8	3	5	2	3	SRC | CAPN1 | IKBKB	TP53 | SLC25A5
NOS1 *	8	9	3	3	5	6	HRAS | PRKACA | CALML3	SLC25A15 | SLC25A2 | OTC | ASS1 | PRKCA
NOS3	6	8	2	3	4	5	CAV1 | AKT1	SLC25A2 | ASS1 |SLC25A15 | OTC
NPM1	6	7	3	3	3	4	RPS6 | GRB2 | NOLC1	TUFM | SLC25A5 | TP53
PKM2	4	6	1	3	3	3	PKLR	POLG |AK3L1 | AK2
PRNP	6	7	3	4	3	3	CAV1 | MAP2K1 | GRB2	HSPD1 | BAX | SOD1
RB1	8	9	3	3	5	6	INS | MAPK1 | PIK3R3	C1QBP |BRCA1 | TGM2 | PHB | TRAP1
RELA	7	10	4	7	3	3	IKBKB | CALM1 | IKBKB | PRKCZ	ETHE1 | MTIF2 | ESR1
RPL10	5	6	2	3	3	3	RPS6 | NOLC1	PRKCA | SLC25A5 | TUFM
RPL11	4	6	2	3	2	3	RPS6 | NOLC1	SLC25A5 | TUFM
RPL30	5	6	2	3	3	3	RPS6 | NOLC1	TUFM | SLC25A5 | MRPS15
RPL35A	5	6	2	3	3	3	NOLC1 | RPS6	SLC25A5 | TUFM | MPG
RPS14	4	6	2	3	2	3	RPS6 | NOLC1	TUFM | SLC25A5
RPS9	5	6	2	3	3	3	NOLC1 | RPS6	TUFM | SLC25A5 | TOMM40
SNRPD1	7	7	4	4	3	3	EIF4E | PPP1CA | YWHAB | YWHAQ	C1QBP | HSPD1 | TRAP1
TOP1	5	6	2	3	3	3	RPS6 | NOLC1	TUFM | SLC25A5 | TP53
TRAF2	6	9	4	6	2	3	MAP3K1 | CAV1 | MAPK10 | MTOR	CASP8 | MAP3K5
TRAF6	9	12	5	7	4	5	PRKCZ | SRC | MAP3K1 | MAPK10 | MAP2K1	NDUFA1 | MT-CO2 | HADHA | NDUFAF1
U2AF1	6	7	2	3	4	4	NOLC1 | RPS6	SLC25A5 | TRAP1 | TUFM | HSPD1
YBX1	6	8	3	3	3	5	RPS6 | NOLC1 | AKT1	SLC25A5 | TUFM | TP53

The internode genes listed in the table have at least three lines of evidence that link them to the mitochondria and three to insulin signaling.

Above 95% percentile of T2D association gene scores based on DIAGRAM meta-analysis (see Table 2).

Associated to HOMA-IR(9.74E–6) [42].

Figure 2

Connections of two internode genes, TRAF2 and NFKB1, with insulin genes and mitochondria genes.

Two strong candidates linking both insulin and mitochondria genes from Table 1 were chosen and their connections to insulin genes and mitochondria genes verified using literature published in the PubMed. See main text for detailed description. A) TRAF2 has been reported to be connected to MAP3K1 (MEKK1) and CAV1 (caveolin 1) insulin genes, and to MAP3K5 (ASK1) and CASP8 (caspase-8) mitochondrial genes. A possible connection to MTOR (mTOR) has also to be considered. MAP3K5 = ASK1; MAP3K1 = MEKK1. B) NFKB1 (NF-κB1) is connected to IKBKB (IKKβ), AKT2, MAP3K1 and SOCS3 insulin genes, and BCL2 and BCL2L1 mitochondrial genes. NFKB1 = NFKBp50; IKBKB = IKKβ = IKK2; MAP3K1 = MEKK1; P65 = RelA. Green boxes represent insulin genes reported to interact with TRAF2 or NFKB1 according to our network; and light-blue represents mitochondrial genes reported to interact with TRAF2 or NFKB1 according to our network. Yellow boxes represent insulin signaling genes.

Table 2

Internode genes that fall in the 95% percentile of T2D association gene scores based on DIAGRAM meta-analysis using MAGENTA, and putative associations with T2D-related traits.

HGCN Name	Gene p-value	Chr	Best local SNP	Best SNP Chr Pos (bp)**	Best SNP pval	Best SNP OR	Distance from gene (Kb)	Previously reported T2D associations	Other quantitative associated Traits (Trait: rsid; p value; GeneLoc; distance in Kb)
NFKB1#	1.26E-05	4	rs7674212	104.208	1.70E-07	1.114	450.8		DBP: rs13107325; 7.53e-07; UP; 233.8 | BMI: rs13107325; 1.37e-07; UP; 233.8 | HDL: rs13107325; 7.2e-11; UP; 233.8 | SBP: rs13107325; 2.57e-07; UP; 233.8
IQGAP2	1.66E-05	5	rs4457053	76.461	4.15E-08	0.8607	421.0	[34]	HbA1C: rs6453220; 4.19e-06; inGene; 0
RPL32	1.31E-03	3	rs6802898	12.366	3.18E-06	1.1537	485.2		LDL: rs11709504; 2.15e-07; UP; 201.8 | TC: rs2290159; 4.21e-09; UP; 247.1 | TG: rs17819328; 5.09e-07; UP; 386.7
IQGAP1*	3.03E-03	15	rs8042680	89.322	8.19E-06	1.1018	475.9	[34]	DBP: rs2521501; 6.06e-07; DW; 391.9 | SBP: rs2521501; 1.22e-07; DW; 391.9
RPS13	6.03E-03	11	rs5215	17.365	1.60E-05	1.0934	309.4	[36], [37]	DBP: rs381815; 1.22e-07; UP; 193.7 | SBP: rs381815; 2.45e-09; UP; 193.7
LRPPRC	9.98E-03	2	rs11899863	43.472	1.04E-05	1.1688	494.6		LDL: rs4299376; 1.73e-47; UP; 40.8 | FastGluc: rs11899863; 2.58e-06; UP; 494.5 | HOMA-B: rs10203174; 9.8e-06; UP; 423.3 | SBP: rs10206724; 5.57e-06; inGene; 0 | TC: rs4299376; 4.03e-45; UP; 40.8
LSM8	1.04E-02	7	rs10244364	117.317	2.97E-05	1.0916	294.4
CD3EAP	1.06E-02	19	rs4420638	50.115	5.41E-05	1.1468	486.5		BMI: rs2287019; 3.18e-07; DW; 288.1 | HDL: rs4420638; 4.4e-21; UP; 486.6 | LDL: rs4420638; 8.72e-147; UP; 486.6 | 2 hrGluc: rs10423928; 3.33e-06; DW; 268.3 | TC: rs4420638; 5.2e-111; UP; 486.6 | TG: rs439401; 1.14e-30; UP; 495.0
KPNA2	1.08E-02	17	rs11655081	63.894	5.55E-05	1.2423	420.4
PPP3CA	1.10E-02	4	rs1426538	102.669	1.99E-05	0.914	181.6
POLH	1.58E-02	6	rs6933958	43.932	5.13E-05	1.1031	235.4		WHR: rs1358980; 1.38e-10; DW; 177.8 | SBP: rs10948071; 5.89e-06; UP; 263.2 | TG: rs998584; 4.06e-07; DW; 171.2
BCAR1	2.48E-02	16	rs9927309	73.804	6.61E-05	1.1388	16.5
RAB4A	2.73E-02	1	rs3767331	227.831	6.53E-05	0.8521	323.9
SRSF1	3.15E-02	17	rs886929	53.156	6.76E-05	1.0941	277.6
GRM5	3.24E-02	11	rs16913724	87.621	2.70E-05	1.1309	259.7
RPL22	3.67E-02	1	rs7368256	6.558	1.31E-04	1.1085	375.5
H1FX	3.79E-02	3	rs9847824	130.8	2.52E-04	1.1309	282.5		WHR: rs2301573; 3.68e-06; DW; 270.8
RPL10A	4.67E-02	6	rs10484578	35.354	1.62E-04	1.0865	189.9		HDL: rs7742443; 6.27e-06; UP; 427.5 | LDL: rs3800406; 2.89e-07; UP; 303.1 | FastIns: rs13210323; 6.32e-06; UP; 431.1 | TC: rs3800406; 7.31e-09; UP; 303.1

Associations were looked-up in GWAS meta-analyses from the MAGIC, GIANT and ICBP consortiums for SNPs within 500 kb from the internode gene boundaries.

upregulated in our chronic insulin treatment (fold change = 1.4; FDR = 0.004) and in the Dor RNAi experiment (fold change = 1.33; FDR = 0.1).

Chromosome position based on genome build 36 (hg18).

Within the high confidence internode set.

BMI: Body Mass Index; HDL: High Density Lipoprotein; LDL: Low Density Lipoprotein; 2 hrGluc: 2 hours glucose challenge; TC: total Cholesterol; WHR: Waist to Hip Ratio; TG: Triglycerides; DBP:Diastolic Blood Pressure; SBP: Systolic Blood Pressure; HbA1C: Glycaeted Hemoglobin; FastGluc: Fasting Glucose levels; HGNC: HUGO Gene Nomenclature committee.

The same MITIN network also allowed us to define which mitochondrial genes are more connected to insulin signaling, and vice-versa, either directly or indirectly. The top five insulin signaling genes most connected to mitochondria are NOLC1, RPS6, IKBKB, PKLR, SRC, with a total of 99, 40, 31, 28 and 22 indirect connections with mitochondria, respectively. Similarly, the five most connected mitochondrial genes with the insulin cascade were TUFM, TP53, SLC25A5, POLG, ESR1, with a total of 93, 36, 29, 25, and 19 indirect connections, respectively (Table S4).

We next explored whether our collection of internode genes where enriched in particular functions or processes by querying the Molecular Signatures Database [25]. We found up to 148 functional signatures for which internode genes were significantly enriched (5.7×10−107