Alternate Phosphorylation/O-GlcNAc Modification on Human Insulin IRSs: A Road towards Impaired Insulin Signaling in Alzheimer and Diabetes

Impaired insulin signaling has been thought of as important step in both Alzheimer's disease (AD) and type 2 diabetes mellitus (T2DM). Posttranslational modifications (PTMs) regulate functions and interaction of insulin with insulin receptors substrates (IRSs) and activate insulin signaling downstream pathways via autophosphorylation on several tyrosine (TYR) residues on IRSs. Two important insulin receptor substrates 1 and 2 are widely expressed in human, and alternative phosphorylation on their serine (Ser) and threonine (Thr) residues has been known to block the Tyr phosphorylation of IRSs, thus inhibiting insulin signaling and promoting insulin resistance. Like phosphorylation, O-glycosylation modification is important PTM and inhibits phosphorylation on same or neighboring Ser/Thr residues, often called Yin Yang sites. Both IRS-1 and IRS-2 have been shown to be O-glycosylated; however exact sites are not determined yet. In this study, by using neuronal network based prediction methods, we found more than 50 Ser/Thr residues that have potential to be O-glycosylated and may act as possible sites as well. Moreover, alternative phosphorylation and O-glycosylation on IRS-1 Ser-312, 984, 1037, and 1101 may act as possible therapeutic targets to minimize the risk of AD and T2DM.


Introduction
Alzheimer's disease (AD) is a neurodegenerative disease associated with a progressive loss in memory and other mental processes, resulting in dementia. In the AD brain, some changes occur in tertiary structures of microtubuleassociated protein tau and -amyloid (A ) peptide resulting in self-association and deposition. Hyperphosphorylated forms of tau act as major constituents of neurofibrillary tangles (NFTs), while A is major constituent of plaques [1]. By 2050, there is expected to be a million new cases of AD per year, or one new case every 33 seconds and AD prevalence is proposed to be 11 to 16 million [2]. On the other hand diabetes mellitus (DM) is characterized by hyperglycemia. Type 1 diabetes is insulin-dependent diabetes mellitus caused by absolute deficiency of insulin secretion while type 2 diabetes (T2DM) is non-insulin-dependent diabetes mellitus caused by both inadequate insulin secretion and insulin resistance of target organs [3]. In DM, numerous peripheral organs, such as kidneys, eyes, and peripheral nerves, undergo pathological changes. DM also affects the central nervous system (CNS). Specifically, both types of diabetes are associated with impairment of learning abilities and memory deficits [3]. Hoyer in 1998 presented a hypothesis stating that brain insulin may be resisted in AD and that AD can be called a brain specific "type 3 diabetes. " Both AD and T2DM are degenerative diseases characterized by neuronal loss and -cells destruction, respectively. Impaired insulin signaling resulting in neurodegeneration and cognitive damages is the main systematic link between these two diseases. Many studies found that DM almost doubles the risk of AD. It was also found that diabetes is not only the risk factor of AD but it also accelerates the onset of AD. Recent studies also elaborate the involvement of impaired insulin 2 Advances in Bioinformatics signaling in tau hyperphosphorylation. AD and DM also share several enzymes (DOPA decarboxylase, glutamic acid decarboxylase) and growth factor receptors (thyrotrophinreleasing hormone, neuronal growth factor receptors, and p75 receptors) [4][5][6][7][8][9].
Insulin receptor substrate proteins (IRSs) act as interaction platforms that are phosphorylated on multiple tyrosine residues and attract different signaling proteins to spread the stimulus. Two isoforms of IRSs, that is, IRS-1 and IRS-2, are most abundant in mammals and have many tyrosine phosphorylation sites. In general, IRS-1 is important for growth, while proper function of pancreatic -cells and glucose homeostasis are main functions of IRS-2 [10,11]. Knockout of either of them causes insulin resistance, whereas only IRS-2-knockout results in diabetes. Either IRS-1 or IRS-2 gene deletion in mice leads to insulin resistance [12]. Irreconcilably, heterozygous and brain specific deletion of IRS-1 and IRS-2 in mice increases duration of life [13,14]. However, IRS-2 deletion is important to induce tau phosphorylation and it also accumulates cytoplasmic deposits of phosphorylated tau [15]. Loss of IRS-2 in neurons of AD patients is particularly severe. IRS-1 and IRS-2 decreased levels in AD neurons increase NFT pathology. Reduced IRS-1 and IRS-2 levels indicate ineffective signaling of IR and IGF-1R in AD [16]. Apart from their important role in metabolic signaling, IRSs also spread proliferative and antiapoptotic signals and as a result become overexpressed in most cancers [17].
Analogous to phosphorylation, O-glycosylation is also one of the important PTMs during which N-acetyl-glucosamine (O-GlcNAc) molecule is introduced on Thr or Ser residue by O-GlcNAc transferase (OGT). Phosphorylation on Thr or Ser residue can be inhibited by the addition of O--GlcNAc. Interplay between these two PTMs on the same residues also known as yin yang sites has been reported in several nuclear and cytoplasmic proteins [29]. These PTMs regulate many functions of the proteins and are responsible for temporary conformational changes [30]. Various studies have shown that O-glycosylation on IRSs reduced in AD and T2DM due to impaired insulin signaling and glucose metabolism [31][32][33][34]. Although some Ser/Thr sites are mapped to be O-glycosylation on C-terminal of IRSs, their exact location is still undetermined. This work predicts potential phosphorylation, O--GlcNAc, and yin yang sites on IRS-1 and IRS-2 proteins, involved in metabolic functions in the brain by using various bioinformatics tools. Impaired phosphorylation and O-glycosylation on IRS 1 and 2 lead to defective insulin signaling that resulted in Alzheimerand diabetes-like complications. The sites where this type of crosstalk occurs are useful targets for designing new drugs for resulting diseases.

Multiple Alignment and Sequences
Retrieval. The FASTA sequences of human IRS-1 and IRS-2 were taken from the Uniprot sequence database (http://www.uniprot.org/uniprot/ ?query=irs&sort=score) with entry name IRS1 HUMAN and accession number P35568 and entry name IRS2 HUMAN with accession number Q9Y4H2, respectively. Blast of these two sequences was made by Blast/Uniprot. A total of 37 hits for IRS-1 and 24 hits for IRS-2 were obtained with Evalue of zero or less. Sequences based on uncharacterized, putative, hypothetical, predicted, and unnamed proteins were neglected. For the multiple alignments of IRS-1 proteins, eight sequences were selected from 37 retrieved sequences. The accession numbers of these selected sequences are given in Table 1. For the alignment of IRS-2 proteins, three sequences were selected from 24 retrieved sequences. The accession numbers of these three selected sequences are given in Table 2. ClustalW (http://www.ebi.ac.uk/Tools/msa/ clustalw2/) was used for the multiple alignments of all the sequences of IRS-1 and IRS-2 to get the conservation status of serine, threonine, and tyrosine residues.
Phospho.ELM (http://phospho.elm.eu.org) [47] and Uniprot (http://www.uniprot.org/uniprot/) databases were used to obtain the data regarding experimentally verified phosphorylation sites on human IRS 1 and IRS 2.  [48] was used to predict O--GlcNAc modification potential sites. The potential phosphorylation sites can also be predicted by YinOYang 1.2 program. Hence, this program predicts the yin yang sites with highly uneven threshold that is adjusted in accordance with amino acid surface accessibility. This method also helps to predict the potential yin yang sites. Modification potential and conservation status of Thr and Ser residues were used to determine the false negative (FN) yin yang sites.

Predicted 3D Structures of Human IRS-1 and IRS-2
Proteins. 3D structure of protein plays important role in determining the exact and precise functions of any protein.
However, the whole 3D structures of IRS-1 and IRS-2 proteins have not been determined yet. To assess this difficulty we predict the 3D structure of these proteins by homology and Advances in Bioinformatics 9 ab initio modeling. We used many servers to predict the whole structure of both proteins and I-TASSER was the only server that predicts the complete structure of these proteins. I-TASSER predicted five models for each 3D structure with minimum -score. The energy of these structures was minimized using YASARA software. To determine whether these structures truly represent or are nearer to possible future true structures, their geometry had been accessed by drawing Ramachandran plots that represent most reliable and suitable source to check predicted 3D structure's geometry. The Ramachandran plot analysis is a method to determine the phi and psi torsion angles between amino acids and compare this information to experimentally verified data as well as standard geometrical properties of protein 3D structures. Five models each were predicted by I-TASSER for constructing Ramachandran plots assessing both IRS-1 and IRS-2. Figures 7(a) and 7(b) represent the predicted 3D structures, Ramachandran plot analysis, and geometrical properties of IRS-1 and IRS-2.
On the basis of Ramachandran plot analysis we proposed predicted model #2 (Figure 3(a)) as possible 3D structure for IRS-1 protein. In this structure 94.7% residues were in Ramachandran allowed region with only 5.32% outliers. No residues with bad bonds were found while only 1.06% residues with bad angles were predicted. However, the c deviations for these predicted models were not predicted by server.
Ramachandran plot analysis was also performed for predicted structures for IRS-2 protein as shown in Figure 3(b) and model #1 was selected as best predicted model. The percentage of compliance for the predicted structure of IRS-2 with the Ramachandran plot indicates that 82.2% (361) of residues reside in the favored region while 94.3% (414) of all residues exist within the allowed region with only 9 C deviations > 0.25A. Only 5.69% of residues exist within the disallowed region of the plot. The 3D structures of IRS-1 and IRS-2 proteins predicted by I-TASSER were submitted to online database PMDB-Protein Model Database (http://bioinformatics.cineca.it/PMDB/). PMDB accession numbers are given in front of each respective model in Figures 3(a) and 3(b).
IRS-1 and IRS-2 predicted 3D structures had been compared with each other by FATCAT and visualized by using Chimera as shown in Figure 4. IRS-1 structure was shown in red while IRS-2 in green. Both structures were significantly different ( value 9.01 E-01; Raw score = 251.71). The structure alignment has 546 equivalent positions with an RMSD of 11.03, with five twists. Both structures were found to have identity 4.02% and similarity 8.71%. This information leads to the conclusion that both IRS-1 and IRS-2 proteins regulate insulin singling in different ways.

Predicted Phosphorylation Sites in IRS-1.
To predict phosphorylated residues in human IRS-1, NetPhos was used. Figure 5(a) is a graphical representation of phosphorylation potential of the Ser, Thr, and Tyr residues in IRS-1. NetPhos predicted phosphorylation on 113 Ser, 13 Thr, and 20 Tyr residues including all experimentally verified phosphorylation sites except Ser-323, 374, 794 and Thr-530 and 1111 as shown in Table S1 available online at http://dx.doi.org/10.1155/ 2014/324753.

Predicted and Experimentally Verified Kinases in IRS-1.
Various kinases involved in phosphorylation of human IRS-1. NetPhosK 1.0, Scansite, and KinasePhos 2.0 were used to assess possible kinases on IRS-1. The results obtained from these three servers had shown the involvement of various kinases in phosphorylation of human IRS-1. GSK3, PKC, and PKA showed maximum involvement in phosphorylation while other kinases like PKG, p38MAPK, cdk5, CKII, cdc2, ATM, and CKI also showed potential for phosphorylation of many Ser and Thr residues. INSR, syk, SRC, and insulin receptor kinase are the most predicted kinases for tyrosine phosphorylation as shown in Table 1S. IRS-1 IRS-2 Identity = 4.02% Similarity = 8.71% Figure 4: Homology models of both human IRS-1 and IRS-1 were retrieved utilizing automated protein modelling options through I-TASSER. The best 3D representative models for both IRS-1 (in red) and IRS-2 (in green) were aligned using FATCAT server and viewed in Chimera software package. Sequence alignment revealed that both proteins are significantly different in structure and only have 4.02% identity and 8.71% similarity.

Predicted and Experimentally Verified Kinases in IRS
predicted kinases involved in phosphorylation of many serine and threonine residues in IRS-2. INSR, syk, SRC, and insulin receptor kinase are the most predicted kinases for tyrosine phosphorylation as shown in Table 2S. Intrinsic tyrosine kinase in the -subunit of IR is known to phosphorylate diverse substrates, including IRS-2 [51].

Prediction of O--GlcNAc Modification and Potential
Yin Yang Sites for IRS-2. According to YinOYang 1.2 results, IRS-2 had high potential for O--GlcNAc modification and 64 Ser/Thr residues were predicted to be O-glycosylated (Figure 7(a)). As described earlier, these sites can also undergo phosphorylation modification that may lead to interplay between O-glycosylation and phosphorylation on these sites. YinOYang 1.2 predicted 40 possible yin yang sites (Figure 7(b)

Discussion
Insulin is a primary hormone that regulates glucose transport and homeostasis in cell systems through its interaction with its receptors (IR) via tyrosine autophosphorylation of IR. Activation of IR subsequently phosphorylates its substrates (IRSs). Insulin receptor substrates 1 and 2 are specific substrates for IR/IGFR and widely expressed in mammalian tissues. Both IRS 1 and 2 contain phosphotyrosine-binding (PTB) domains. Tyrosine phosphorylation of IRS proteins is well studied and more than 20 potential tyrosine phosphorylation sites are mapped on IRSs. During insulin signaling, IRS proteins phosphorylate and activate AKT/PKB/MAPK pathways that regulate insulin mediated metabolic actions and control cell growth and differentiation [10,20].
Despite Tyr phosphorylation, IRSs also phosphorylate on Ser/Thr residues and phosphorylation on these residues may effect downstream insulin signaling. More than 30 Ser/Thr residues on IRS-1 and 50 on IRS-2 have been mapped using different analytical techniques. However, function of all mapped Ser/Thr residues on IRSs is still undetermined. Our results showed that both IRS-1 and IRS-2 have high potential to be phosphorylated on Ser and Thr residues and more than 100 Ser/Thr residues have been predicted including previously mapped residues on IRS 1 and 2 on both N-and C-terminals including PTB domain (Figures 5(a) and 5(b)). These predicted residues are conserved as shown in Figures  1 and 2 (multiple alignment of IRS-1 and IRS-2, resp.) and may have potential to modify IRSs functioning. Meanwhile various kinases such as GSK3, PKC, PKA, ERK, mTOR, S6K, PKG, CDK5, CKII, and JNK are involved in Ser/Thr phosphorylation of IRSs. Both ERK and mTOR are experimentally verified kinases that can cause phosphorylation at Ser-616 [55]. Phosphorylation of Ser-312 is mediated by increased activation of kinases downstream of PI3-kinase, comprising S6K1, IKK , mTOR, PKC, and JNK [22,25,56]. Akt-mTOR activity is highly activated within neurons in many AD cases [57]. S6K activation was also found to be induced in AD that induces phosphorylation of Ser-636 and 639 of IRS-1 [58]. Ser-312, 616, 636, 639, and 1101 in IRS-1 are experimentally      verified phosphorylation sites. Phosphorylation of Ser-312 causes IRS-1 proteasomal degradation that leads to insulin resistance under both pathological and physiological conditions, while phosphorylation of Ser-616 is supposed to inhibit insulin activation of PI3-kinase. Furthermore, phosphorylation on Ser-1101 by S6K1 on C-terminal of IRS-1 is thought to be involved in blocking tyrosine as well as AKT phosphorylation that leads to insulin resistance in mice as well as in human [21,23,25]. GSK-3 and JNK sequentially phosphorylate IRS-2 at serine residues and its phosphorylation plays inhibitory role in hepatic insulin signaling [51,58,59].
The O--GlcNAc modification is known to be influential and analogous to phosphorylation. Some researchers indicated increased phosphorylation and reduced O-GlcNAc levels in the AD and diabetes patients [60][61][62]. These findings made O-glycosylation as important PTM in elucidating the mechanism of such metabolic disorders. In this study we used YinOYang 1.2 server to predict O-glycosylation on Ser/Thr residues of both IRS-1 and 2, and more than 50 Ser/Thr residues were shown to have high potential for O--GlcNAc modification on IRS-1 (Figures 6(a) and 7(a) and Tables 3  and 4). O--GlcNAc modification in human IRS-1 has been reported at Ser-984 or 985, 1011, and possibly at multiple residues within 1025-1045 [33]. Based on our results Ser-984 is more prone to be O-glycosylated than Ser-985 while Ser-1030, 1035, 1037 and Thr-1045 have potential to be Oglycosylated in 1025-1045 sequence on the basis of conservation. Meanwhile western blotting and site-directed mutagenesis studies in rat suggested Ser-1036 (human numbering Ser-1037) on IRS-1 as possible site of O--GlcNAc modification. Ser-1037 is conserved in human IRS-1 and identification of this site will be helpful in exploring the biological implication of the O--GlcNAc modification [33]. Although to date no O-glycosylated residue is mapped on IRS-2, experimental data showed that IRS-2 also have potential to be glycosylated as IRS-1 and this glycosylation was shown to be reduced in diabetes and AD [32]. In this study we have predicted more than 65 Ser/Thr residues on IRS-2 that can act as potential O-glycosylated sites and may ease researchers to map correct O-glycosylated residues on IRS-2.
It is established fact that O--GlcNAc can mimic or inhibit phosphorylation on the neighboring or same residues and if this interplay happened on same Ser/Thr residues these sites are called yin yang residues. These yin yang residues can be used as therapeutic targets to overcome different diseases [42,52]. Protein 3D structures are best models to tell whether the predicted residues are accessible or not for these types of interplays [27,41]. However, for both IRS-1 and IRS-2, complete 3D have not been determined yet, so we used ab initio models to design the complete 3D structures of both proteins. As both IRS-1 and IRS-2 are involved in insulin signaling mechanism, best possible 3D structures for both proteins were selected on basis of Ramachandran plots (Figures 3(a) and 3(b)) and were aligned together to be compared with each other (Figure 4). We found that both proteins have significantly nonidentical structures as their secondary structure previously described. To determine yin yang sites and surface accessibility, we predicted surface accessible residues on the basis of their secondary structure by using NetPhos P server (as given in Table S1 and S2 for both IRS-1 and IRS-2, resp.). Secondary structure based prediction method showed that both proteins have high accessibility for PTMs on Ser/Thr residues except Ser-268, 273, 312, 393, 394, 440, 1223 on IRS-1, Ser-428, and Thr-344. Although secondary structure based prediction method showed no/reduced surface accessibility for Ser-312 of IRS-1, predicted 3D structure revealed that Ser-312 is accessible for PTMs ( Figure 8). Moreover, Ser-312 is experimentally verified and known phosphorylation residue, which confirms the exposed accessibility of this residue. Ser-312 phosphorylation   is important in insulin resistance and has negative regulation of insulin signaling. Interestingly, Ser-312 is situated very near to PTB domain ( Figure 8) and any change in its vicinity via PTMs should alter surrounding areas [21]. Furthermore phosphorylation on three important Ser residues 984, 1037, and 1101 in C-terminal of IRS-1 also has been shown to induce insulin resistance [33]. All these residues were conserved, accessible for such modifications based on 3D structure, and predicted as yin yang residues in our study. This finding showed that alternative O-glycosylation on these residues can inhibit the protein functioning rise by phosphorylation. Although a lot of phosphorylation sites mapped on IRS-2 like Ser-303, 343, 675, and 907 are thought to be involved in inducing insulin resistance, we were unable to predict Oglycosylation on these residues due to limited experimental data availability. On the other hand, we found more than 40 Ser/Thr residues on IRS-2 that has high potential to act as possible yin yang sites (Table 4), and many of them may act as therapeutic targets in future.
Here we propose that alternative O-glycosylation on Ser-312 at N-terminal and Ser-984, Ser-1037, and Ser-1101 at C-terminal of IRS-1 is an important modification that becomes reduced in diabetes and AD due to impaired glucose metabolism. During normal circumstances, O-glycosylation can block phosphorylation on these residues and can act as possible therapeutic sites to reduce risk of diabetes and AD.