Detecting early-warning signals of type 1 diabetes and its leading biomolecular networks by dynamical network biomarkers
© Liu et al.; licensee BioMed Central Ltd. 2013
Published: 7 May 2013
Type 1 diabetes (T1D) is a complex disease and harmful to human health, and most of the existing biomarkers are mainly to measure the disease phenotype after the disease onset (or drastic deterioration). Until now, there is no effective biomarker which can predict the upcoming disease (or pre-disease state) before disease onset or disease deterioration. Further, the detail molecular mechanism for such deterioration of the disease, e.g., driver genes or causal network of the disease, is still unclear.
In this study, we detected early-warning signals of T1D and its leading biomolecular networks based on serial gene expression profiles of NOD (non-obese diabetic) mice by identifying a new type of biomarker, i.e., dynamical network biomarker (DNB) which forms a specific module for marking the time period just before the drastic deterioration of T1D.
Two dynamical network biomarkers were obtained to signal the emergence of two critical deteriorations for the disease, and could be used to predict the upcoming sudden changes during the disease progression. We found that the two critical transitions led to peri-insulitis and hyperglycemia in NOD mices, which are consistent with other independent experimental results from literature.
The identified dynamical network biomarkers can be used to detect the early-warning signals of T1D and predict upcoming disease onset before the drastic deterioration. In addition, we also demonstrated that the leading biomolecular networks are causally related to the initiation and progression of T1D, and provided the biological insight into the molecular mechanism of T1D. Experimental data from literature and functional analysis on DNBs validated the computational results.
Type 1 diabetes (T1D) is a form of diabetes mellitus that is a clinically heterogeneous group of glucose intolerance syndromes, and usually has an autoimmune T cell-mediated etiology in which the pre-diabetic state is characterized by development of autoantibodies against certain proteins expressed by β cells, including insulin [4, 5]. T1D is a complex disease and threatens the human health in the world, and the prediction of early-warning signals for T1D before the disease onset has not been reported. Also, the detail molecular mechanism for the disease progression, e.g., driver genes or causal network of the disease, is still unclear.
The non-obese diabetic (NOD) mouse strain [4–7] is a useful and important model of autoimmune disease and also an excellent tool for understanding the onset mechanism of T1D. The pancreatic lymph nodes are an important organ to preserve the antigen-specific T cell and surround the pancreatic islet [8, 9]. And many reports have considered that they are related to T1D [4, 6] and a number of research works used them to identify the potential biomarkers of T1D. In this study, instead of traditional molecular biomarker, we identify the pre-disease state of T1D by the new type of biomarker, i.e., dynamical network biomarker which forms a specific module of molecules (e.g., genes, RNAs, proteins, or metabolites) for marking the time period just before the drastic deterioration of T1D. Specifically, we detected early-warning signals of T1D and its leading networks based on the serial gene expression profiles of pancreatic lymph nodes in NOD mice by identifying two dynamical network biomarkers (DNB) in two different time points. By the theory of early-warning signals of complex diseases , a dynamical network biomarker will form a specific module in the pre-disease stage or near critical point before disease phenotype occurrence. Hence, the two critical points which were identified by the two respective DNBs would correspond to the early stages of disease onset for T1D. Actually, it is consistent with the reports about the pre-disease stage or disease onset time point of NOD mouse in the public references [5, 6]. In addition, the two DNBs are also the leading networks, which are closely related to the driver molecules of the disease progression. Therefore, the two DNBs not only can be used to predict the upcoming disease onset before the drastic deterioration of the T1D phenotype, but also can reveal the molecular mechanism on the disease initiation and progression. Moreover, the computational results and the leading networks were also validated by experimental data from literature and functional analysis of the two DNBs. In particular, we found that the two DNBs could affect several famous pathways in T1D, such as "T cell receptor signaling pathway", "NF-kappa B signaling pathway" and "Insulin signaling pathway", on the pre-disease stages of the disease onset, which are all consistent with the existing results and other independent experimental results of literature.
Gene expression profiles
The non-obese diabetic (NOD) mouse is a useful and important model for autoimmune T1D. The pancreatic lymph nodes are the site of islet-cell-specific self-antigen presentation and important for the development of T1D. The gene expression profiles of pancreatic lymph nodes for T1D were obtained from GEO database (ID: GSE15150). The dataset includes the expression profiles of pancreatic lymph nodes of 35 female NOD mice samples at 6 different time points (10 days (7 samples), and 4 weeks (6 samples), 8 weeks (4 samples), 12 weeks (7 samples), 16 weeks (6 samples), and 20 weeks of age (5 samples)).
The original data was normalized by the logarithm ratios: Log10 (NOD processed signal / control signal), but this ratio cannot be directly used to calculate the correlation between genes. So the normalized data was transformed back to the general ratio (NOD processed signal / control signal) by exponent 10 operation.
Criteria for DNB or the leading network
Pearson correlation coefficients (PCCs) between any pair of members in DNB become very high (i.e., drastic increase).
PCCs between one member of DNB and any other molecule of non-DNB become very low (i.e., drastic decrease).
Standard deviation (SD) for any member of DNB becomes very high (i.e., drastic increase).
Actually, it can also be shown that SD for any member of non DNB, and the PCCs between non-DNB members have no significant change. Clearly, the molecules in the dominant group are strongly and dynamically correlated in the pre-disease state. These molecules in the dominant group are expected to form a subnetwork or functional module from a network viewpoint, i.e., DNB. Therefore, the three conditions are considered as three criteria to detect the DNB or early-warning signals of the pre-disease state. Besides, since the three criteria are in fact the generic properties of the DNB members in dynamics whenever the system approaches a critical tipping point, these properties should lie in many complex diseases with sudden deterioration phenomena. The three measurable conditions can be summarized into a single index to detect the DNB, i.e., eqn.(1) as described next in details. Note that DNB is also the leading network, which makes the first move into the disease state from the normal state, and therefore, is causally related to the initiation and progression of the disease .
Identification of dynamic network biomarkers (DNB)
where is the score of a module or a candidate DNB, is the average standard deviation of genes expression in the module, is the average Pearson correlation coefficient among genes inside of the module, and is the average Pearson correlation coefficient between inside and outside genes of the module. Clearly, (1) represents the three conditions of the DNB.
For every time point, the score of every module was calculated by the above formula based on the gene expression of the module in this time point and the best module with the highest score was regarded as the potential DNB in this time point. Then, these identified potential DNBs in every time point were compared each other, and the highest score DNB in all time points was the DNB for detecting the early-warning signals before the disease onset (Figure 2). The time point corresponding to the DNB was called critical point, which is the early stage of the disease onset. Also, the DNB is the leading network, which leads the system to the disease state.
Regulated gene of the DNB
The regulated genes by the identified DNB module are picked up from the onset time point. The genes, which are highly correlated with DNB module in onset time point and are also differential expression genes between the critical point and onset time point, are regarded as regulated genes by the DNB module. If a gene is highly related with at least 10 genes of DNB, we deem that the gene is highly related to the DNB module. Here the threshold of high relation is set to 0.05 of P-value of PCC and the threshold of differential expression is 0.05 of P-value of student's t test.
Functional analysis of the DNB
The confidence of the identified DNB which is associated with early-warning signals before the disease onset can be proven by the evidence of disease phenotype from published references. The genes in the identified DNB have been linked and correlated to some pathways of KEGG (http://www.genome.jp/kegg/), and these pathways can be related to the disease initiation and progression. First, the genes of the DNB were mapped to pathways by the KEGG Mapper tools (http://www.genome.jp/kegg/mapper.html) which are the online tools for KEGG mapping. Subsequently, the correlations between the DNB and each pathway in KEGG were calculated in two time points that are the critical point and the disease onset point.
Potential DNBs in every time point
Early-warning signals of DNB in T1D development
From the score of the second DNB in every time point (Figure 4 E-H), we can see that the third (8 weeks) time point had the high score which can be the critical time point of the DNB module. Because the disease onset will appear at the next time point of the critical point, the phenotype of disease onset would present at the fourth time point (12 weeks). It is consistent with the reports of public reference and some biological experiments about the development of T1D in female NOD mice [5, 6, 12, 13]. Disease onset of diabetes usually occurs at 12 to 14 weeks of age in female NOD mice, and destructive insulitis, leading to overt hyperglycemia, occurs around 12 weeks of age or later [5, 12, 13]. It validates that the third (8 weeks) time point is the critical point and this DNB can be utilized to detect the early-warning signal of phenotype before the disease onset of T1D.
Molecular mechanism of DNB
For analyzing the mechanism of DNB during the development of T1D, the genes of the first DNB were mapped to the pathways of mouse in KEGG, and only 12 genes were identified from 21 pathways in KEGG. It means that many genes in the DNB would take part in more than one pathway and could affect the cross-talking among different pathways. In these pathways, the "Insulin signaling pathway" is an important one that regulates many metabolism and signal pathways, and is also associated with T1D development. It also includes some virus related pathway in the DNB mapped pathways, such as "Epstein-Barr virus infection". It is consistent with the theory which considers that T1D is a virus-triggered autoimmune response .
The genes of the second DNB were also mapped to the pathways of mouse in KEGG, and 11 genes were identified from 23 pathways in KEGG. There were also many genes which took part in more than one pathway, and they could affect the cross-talking among different pathways. There were three immune related pathways in which the genes of the second DNB participate, such as "T cell receptor signaling pathway", "NF-kappa B signaling pathway" and "Intestinal immune network for IgA production". Because T1D is an autoimmune disease and pancreatic lymph nodes are a major tissue to preserve T cell in pancreas, the "T cell receptor signaling pathway" would play an important role in the disease onset of T1D. The "NF-kappa B signaling pathway" is an important pathway in the T cell autoimmune and related to the onset of T1D [13, 15].
Number of genes and interaction pairs with high correlations between DNB and pathways.
The first DNB
The second DNB
T cell receptor signaling pathway
Disease onset point
NF-kappa B signaling pathway
Disease onset point
Insulin signaling pathway
Disease onset point
The regulated genes by DNB
Bridge between two DNBs
In the regulated genes by the first DNB, we found that 4 genes ('2410003K15RIK', 'TIMM50', 'HOXA4', 'LIN28A') are also in the second DNB, so it is possible that the second DNB can be affected by the first DNB and there is a bridge link between the two DNBs. We picked up the high correlation genes with the second DNB from the regulated genes by the first DNB, and differential expression was used to filter the selected genes. Finally, 110 genes (additional file 4) ware identified to link the two DNBs. Function enrichment analyses of the 110 genes showed that function of these genes was enriched to "mitochondrion" and "electron transport". Mitochondria are important subcellular fractions for cell apoptosis and play a crucial role in regulating cell death, and apoptosis can lead to hyperglycemia and trigger to diabetes onset [10, 11]. The electron transport in mitochondrion is an important process for redox reactions and oxidative stress which can cause to islet cell autoimmunity and lead to T1D . Hence, it is possible that the deterioration process from peri-insulitis to diabetes (hyperglycemia) is linked by the redox reactions and oxidative stress based on mitochondrion electron transport, and dysfunction of mitochondrion electron transport makes the disease worsening.
During the development of T1D, the disease progression will pass a pre-disease stage which is a critical transition period from normal stage to disease stage. After it passes through the critical point, the disease progression is generally irreversible . As a novel biomarker, dynamical network biomarker (DNB) was developed to detect such a critical point just before the disease onset. In this study by proposing a new computational algorithm, we identified two DNB modules which can be separately used to predict the peri-insulitis on the early stage of diabetes and diabetes onset with overt hyperglycemia.
We consider that there are two kind of potential mechanisms for the DNB triggering the disease deterioration or phenotype change. On the one hand, the genes in the DNB were gathered together in the critical point, so they could interact and affect one another. Because most genes in DNB take part in more than one pathway, so these interactions and effects could make these genes deviate from the major pathway and regulate the disease related pathway together in disease onset point. For example, the genes in the two DNB modules were highly correlated with the three pathways on Table 1 in the disease onset point. It is possible that the genes in DNB regulate and control these pathways and make them move to disease phenotype. The "MAPK signaling pathway" is one of important pathways to regulate the glycogen and glucose metabolism and contains 267 known genes in KEGG database. In the critical point of the second DNB, there were only 15 genes of "MAPK signaling pathway" correlated with the DNB, and in the following time point, there were 260 genes of "MAPK signaling pathway" (additional file 5) correlated with the DNB. It means that there were more genes of "MAPK signaling pathway" to be regulated by the same DNB module in the disease onset point. It is possible that there were also stronger regulation effects of the DNB module to the metabolism of glycogen and glucose in the disease onset point. Although most of the genes in DNB belong to different pathways or metabolic processes, finally, these genes would induce these different pathways to the same disease phenotype in the disease onset point.
On the other hand, the genes in the DNB were gathered together to form a module by upstream signal in the critical point, the module can also regulate a small number of genes in some pathway in the critical point. These genes which were mediated by the module in the critical point could play important role in corresponding pathways and trigger the change of the disease phenotype in the disease onset point. For example, we can see from Table 1 that all genes in the second DNB modules were highly correlated with only 7 genes in the "T cell receptor signaling pathway" in the critical point. However, the 7 genes in the "T cell receptor signaling pathway" are located in the up- and middle-stream of this pathway in KEGG database, so it is possible that the 7 genes are the important or driving factors for the pathway to trigger disease phenotype change in the disease onset point. Actually, the DNB has been proven to be the gene group, which makes the first move from the normal state to the disease state. The peri-insulitis is the early-stage of T1D in NOD mice, and the link from peri-insulitis to diabetes is interesting for understanding the disease deterioration process. From the bridge of two DNBs, we can see the possible process that mitochondrion electron transport induces the apoptosis function of the second DNB and pushes the peri-insulitis to diabetes (hyperglycemia).
Traditional biomarkers are usually used to distinguish disease state from normal state, rather than pre-disease state [18–20]. It means that a traditional biomarker measures the disease status of an organ, after the organ has presented the characteristic of disease [21–24]. In this study, completely different from the traditional molecular biomarkers, we distinguished the pre-disease state from the normal state by a new type of biomarkers, i.e., dynamical network biomarkers. Specifically, we found two dynamical network biomarkers which can be used to detect the early-warning signals and predict the upcoming disease onset of T1D by the theory of early-warning signals of complex disease [1, 17]. Based on the three conditions of DNB, the two dynamical network biomarkers identified two respective critical time points which are the pre-disease stages of diabetes onset, and the specific modules formed by the two DNBs in the critical points signal the emergence of the critical transitions of the sudden changes for the disease. For the validation of disease onset time points, the many evidence and public reports were used to validate the onset of T1D in the NOD mouse, and in particular, we found that the reports of disease onset for T1D in the NOD mouse were consistent with the disease onset time points the DNBs marked. Hence, DNBs can be used to detect the early-warning signals of T1D, predict upcoming disease onset before the phenotype occurrence. DNB can also be adopted to analyze the molecular mechanism of the disease initiation and progression of the disease based on the identified leading networks.
This work was supported by NSFC under Nos. 61134013, 91029301, 61103075, 61072149 and 91130032, and by the Chief Scientist Program of SIBS of CAS with No. 2009CSP002. This work was also supported by Shanghai Pujiang Program, by the National Center for Mathematics and Interdisciplinary Sciences of CAS, and by the FIRST program from JSPS initiated by CSTP.
The publication costs for this article were funded by the corresponding author.
This article has been published as part of BMC Medical Genomics Volume 6 Supplement 2, 2013: Selected articles from the Second Annual Translational Bioinformatics Conference (TBC 2012). The full contents of the supplement are available online at http://www.biomedcentral.com/bmcmedgenomics/supplements/6/S2.
- Chen L, Liu R, Liu ZP, Li M, Aihara K: Detecting early-warning signals for sudden deterioration of complex diseases by dynamical network biomarkers. Scientific reports. 2012, 2: 342-PubMed CentralPubMedGoogle Scholar
- Cui J, Chen Y, Chou WC, Sun L, Chen L, Suo J, Ni Z, Zhang M, Kong X, Hoffman LL, et al: An integrated transcriptomic and computational analysis for biomarker identification in gastric cancer. Nucleic acids research. 2011, 39 (4): 1197-1207. 10.1093/nar/gkq960.PubMed CentralView ArticlePubMedGoogle Scholar
- Jin G, Zhou X, Cui K, Zhang XS, Chen L, Wong ST: Cross-platform method for identifying candidate network biomarkers for prostate cancer. IET systems biology. 2009, 3 (6): 505-512. 10.1049/iet-syb.2008.0168.View ArticlePubMedGoogle Scholar
- Atkinson MA, Leiter EH: The NOD mouse model of type 1 diabetes: as good as it gets?. Nature medicine. 1999, 5 (6): 601-604. 10.1038/9442.View ArticlePubMedGoogle Scholar
- Delovitch TL, Singh B: The nonobese diabetic mouse as a model of autoimmune diabetes: immune dysregulation gets the NOD. Immunity. 1997, 7 (6): 727-738. 10.1016/S1074-7613(00)80392-1.View ArticlePubMedGoogle Scholar
- Anderson MS, Bluestone JA: The NOD mouse: a model of immune dysregulation. Annual review of immunology. 2005, 23: 447-485. 10.1146/annurev.immunol.23.021704.115643.View ArticlePubMedGoogle Scholar
- Kikutani H, Makino S: The murine autoimmune diabetes model: NOD and related strains. Advances in immunology. 1992, 51: 285-322.View ArticlePubMedGoogle Scholar
- Hoglund P, Mintern J, Waltzinger C, Heath W, Benoist C, Mathis D: Initiation of autoimmune diabetes by developmentally regulated presentation of islet cell antigens in the pancreatic lymph nodes. The Journal of experimental medicine. 1999, 189 (2): 331-339. 10.1084/jem.189.2.331.PubMed CentralView ArticlePubMedGoogle Scholar
- Gagnerault MC, Luan JJ, Lotton C, Lepault F: Pancreatic lymph nodes are required for priming of beta cell reactive T cells in NOD mice. The Journal of experimental medicine. 2002, 196 (3): 369-377. 10.1084/jem.20011353.PubMed CentralView ArticlePubMedGoogle Scholar
- Lee MS, Chang I, Kim S: Death effectors of beta-cell apoptosis in type 1 diabetes. Molecular genetics and metabolism. 2004, 83 (1-2): 82-92. 10.1016/j.ymgme.2004.08.002.View ArticlePubMedGoogle Scholar
- Lee SC, Pervaiz S: Apoptosis in the pathophysiology of diabetes mellitus. The international journal of biochemistry & cell biology. 2007, 39 (3): 497-504. 10.1016/j.biocel.2006.09.007.View ArticleGoogle Scholar
- Kodama K, Butte AJ, Creusot RJ, Su L, Sheng D, Hartnett M, Iwai H, Soares LR, Fathman CG: Tissue- and age-specific changes in gene expression during disease induction and progression in NOD mice. Clin Immunol. 2008, 129 (2): 195-201. 10.1016/j.clim.2008.07.028.PubMed CentralView ArticlePubMedGoogle Scholar
- Green EA, Choi Y, Flavell RA: Pancreatic lymph node-derived CD4(+)CD25(+) Treg cells: highly potent regulators of diabetes that require TRANCE-RANK signals. Immunity. 2002, 16 (2): 183-191. 10.1016/S1074-7613(02)00279-0.View ArticlePubMedGoogle Scholar
- Fairweather D, Rose NR: Type 1 diabetes: virus infection or autoimmune disease?. Nature immunology. 2002, 3 (4): 338-340. 10.1038/ni0402-338.View ArticlePubMedGoogle Scholar
- Poligone B, Weaver DJ, Sen P, Baldwin AS, Tisch R: Elevated NF-kappaB activation in nonobese diabetic mouse dendritic cells results in enhanced APC function. J Immunol. 2002, 168 (1): 188-196. 10.4049/jimmunol.168.1.188.View ArticlePubMedGoogle Scholar
- Delmastro MM, Piganelli JD: Oxidative stress and redox modulation potential in type 1 diabetes. Clinical & developmental immunology. 2011, 2011: 593863-View ArticleGoogle Scholar
- Liu R, Li M, Liu Z, Wu J, Chen L, Aihara K: Identifying critical transitions and their leading networks for complex diseases. Scientific Reports. 2013, 2: 813-Google Scholar
- He D, Liu Z-P, Chen L: Identification of dysfunctional modules and disease genes in congenital heart disease by a network-based approach. BMC Genomics. 2011, 12: 592-607. 10.1186/1471-2164-12-592.PubMed CentralView ArticlePubMedGoogle Scholar
- Wang J, Zheng S, Zhang X, Zhou H, Chen L: APG: an Active Protein-Gene Network Model to Quantify Regulatory Signals in Complex Biological Systems. Scientific Reports. 2013, 3: 1097-PubMed CentralPubMedGoogle Scholar
- Wen Z, Liu Z, Liu Z, Zhang Y, Chen L: An integrated approach to identify causal network modules of complex diseases with application to colorectal cancer. Journal of the American Medical Informatics Association. 2012, doi:10.1136/amiajnl-2012-001168Google Scholar
- Liu X, Wang J, Chen L: Whole-exome sequencing reveals recurrent somatic mutation networks in cancer. Cancer Letters. 2012, [http://dx.doi.org/10.1016/j.canlet.2012.11.002]Google Scholar
- He D, Liu Z, Honda M, Kaneko S, Chen L: Coexpression network analysis in chronic hepatitis B and C hepatic lesion reveals distinct patterns of disease progression to hepatocellular carcinoma. Journal of Molecular Cell Biology. 2012, 4 (3): 140-52. 10.1093/jmcb/mjs011.View ArticlePubMedGoogle Scholar
- Liu Z, Wang Y, Zhang X, Chen L: Network-based analysis of complex diseases. IET Systems Biology. 2012, doi:10.1049/iet-syb.2010.0052Google Scholar
- Liu X, Liu Z, Zhao X, Chen L: Identifying disease genes and module biomarkers with differential interactions. Journal of the American Medical Informatics Association. 2012, 19: 241-248. 10.1136/amiajnl-2011-000658.PubMed CentralView ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.