藉由蛋白質網絡探討脊髓肌肉萎縮症和脊髓側索硬化症的關聯性

全文

(1)" /*' .,/#*' ) Graduate Institute of Biological Electronics and Bioinformatics College of Electrical Engineering and Computer Science. National Taiwan University Master Thesis. %&+

(2) (!0. $!0- . Comparative analysis of condition-specific protein interaction networks between spinal muscular atrophy and amyotrophic lateral sclerosis. ƉEÒ Kai-Pu Chen. ¼Ê½ƥƆƑł W{ æƊ£ç W{ Advisors: Hsueh-Fen Juan, Ph.D. Yen-Jen Oyang, Ph.D.. Ņêr 105 1 Ú January, 2016.

(3) Ř š ēûČÑƄŵ¦ƛªƤ±őĎƄĄâ"Ø@±šČÐ'ČÆ»ƤŦ ¶^(ô¥ƚ. ³Čìtâ¿ŇŐ±š¼Ê½ ƆƑłĴĆH¶ű. AĢĩĀùŰ7ƃ([Ű7}Ƥ¤±šĴtşËČ¼([Ŧ¶ ¤ĽăuŷÄ°đěČŤƙƤtēûÜƄĨ"¶?ũČÑƄÙ h

(4) B Ěę¢ŚŔƣšš æƊ£çĴÅ)¶ČCb¼Ê½!šš ƢśĴČ nƙh¡ŤƤŦ¶ďŶl$uÍƔŐÇűƣÖ±š ŹW{¥ČƘƤŦ¶Ù 0Ľ¹ŧÌhGÞ¦KĦßƤ([$nƙČ¡Ťhœ¯ ƞČĀíƤ±šïtĀíÁ5ČVRƤ([şËČŃËòƝƣ è|Ƥ!±šþ¶8YioıfČƜùƤ([ĀímČ%îƤ hŉċŨƠČďţƣšš¨=tƒƜČ%îƤhÏÝp4Čėùƣšš UųƂt Č¡Ťšš2´Ƃtűxe¥ĨČ¡ŤhŕşƤĨ "¶tşËÃČ$ƕ±ƣšš&ŊƂt¶MűAƞÑƤ¶ö®à sÿyƣ ššƞēûČ`-ƤtĀíČ©DÖ±š¡EhŲtęŖ Č%îhœŝƣ!¤±šÔt¶ƓSŬáX/ČÑ9Ƥ©ĻƐƎyšš{ <tĕÙŤƙĨČ33}şè|Ƥ!šškhżĳ©tƞŖİơĮƤ Ŧ¶tĐÓ*¬Ñ^(ÈĜļō Ø¥±šĀŻƓŧ· ńÕ·Ƃ([·ŭ©ůĠdƤŦ¶^(ƗJČ A/_}ŀƤ([VRŪŢČŭýè|Ƥ!šš ƆĴ_}ƓŧƈĀĘ ƈhŧðÊĴcàø\Ç¡ĀĘƈ ãôƌĔ ·ƤŦ¶Ătë92 ÑŸŐKIČ}ã · . . $$.

(5) ĺƟvķĶņĭĉhĺƟ;ģĒTĉÐăŴSĖĪ>ŮT[+ƍķĶņ ĭČľg«ćĈƤƏõBĵČľĈĝťwqbƤ,BĵƄÛCbŌüČĉú ó"¾ŕBćĈƄ^ĹČCbľĈåLƤ¶8GI¡Ğ"ĺƟvķĶņĭĉhĺ Ɵ;ģĒTĉČąTŉċŨCŌüīħDCPINs¶8tèJĂgeneontology ÛąČŉċŨīħ:PĹČGĲƤ¿Ň¶8ŕşŠćĈƄäĥČƅ6« ňăĦaƖ²ČŉċŨīħ([mRNAČwqŌüžČGÞƤ¶8¹K"ƀŜČ ŉċŨ%#1Ă^ĹÙľĤĸſƐ§ÿ~À¶8ĊüºſƐ[÷· ľČĤĸzOÛƅČŉċŨ%#1Ăè|Ƥ¶8!¹KhŉċŨ ubiquitinationĿproteasomeƇœčƅČŉċŨ%#1ĂbÑ¶8űéĊüh ATPĀµčƅ. į~ČŉċŨ1Ă!FütġĬƠŏaƠI, III[VƤQ(űŋ. ¾ŕZĿ N¿1ĂČŉċŨsnRNPsČźGĥµŉċÿ ŉċƤ!t B7ćĈŎĊüÛū.Ččƅ«^ĹűƋľsnRNPsĀµČ~ÉtÝğ şËƤ¶8řóŰBćĈČķĶ·āĀČĤĸzOƤÎćĈĊĀÑ^Ĺ¸ñ "¤ŽŐČŒŁ ƅƁ ƥĺƟvķĶņĭĉĺƟ;ģĒTĉąTŉċŨCŌüīħzO ſƐČ~ŞubiquitinationproteasomeČŉċŨƇœATPhRNAČN¿ 1Ă. . $$$.

(6) Abstract Spinal muscular atrophy (SMA) and amyotrophic lateral sclerosis (ALS) are two devastating diseases caused by motor neuron degeneration and accompanied with muscle weakness. Though the mutated genes causing SMA and ALS are different, some of the phenotypes are the same in both diseases. To understand the possibility of common and dysregulated mechanisms between SMA and ALS, differentially co-expressed protein interaction networks (DCPINs) are constructed in SMA and ALS respectively. Gene ontology analysis is applied to help us realize the functions of these disrupted protein interactions. Both SMA and ALS related modules were further isolated and discussed. By means of integrative analysis using static protein interaction network and the microarray gene expression profiles, perturbed protein interactions involving in calcium cycling were found in this study. The possible responses against stress caused by calcium and thermogenesis were also discovered. Furthermore, we identified the protein interactions associated with protein ubiquitination and proteasomal degradation. Additionally, we found the defective protein interactions engaged in path of ATP synthesis of mitochondrial protein complex I, III and V and have further discussion. Proteins involved in RNA splicing were also found and showed the potential deformity in heptameric ring consisted of Sm proteins during formation of snRNPs. In this study, we suggest that the stress induced in the muscle of SMA and ALS might play an important role in the pathology of both diseases.. Keyword: SMA, ALS, DCPINs, module, stress, calcium dysregulation, ubiquitination, . $0.

(7) proteasomal degradation, ATP, and RNA splicing. . 0.

(8) Contents ]ŗjÙ×….……………………………………...………….................................i Řš………………………..………………………………………………………………..ii ËÂŐ….………….……….………….…………………...............................................iii Abstract…...………….……………………………………..….…..…..………….…..........iv Contents…………………………………………………….………….……………...........vi List of Figures…………………………………………………………..……………...….viii List of Tables……………………………………………………………….……….………x Chapter 1 Introduction…………………….…………………...…………………….……1 1.1 Motor neuron disease……………….……..………………………………………….1 1.1.1 Spinal muscular atrophy…………………….……..……….………………...….1 1.1.2 Amyotrophic lateral sclerosis……………………..….……….………...……….2 1.1.3 Interaction between SMA and ALS…....……….……….………..…….…..……3 1.2 Protein interaction network………………..…………………………………...……..3 1.3 Microarray analysis……………………………………………………………..…….4 1.4 Integrative analysis of protein interaction network…………………………..…........4 Chapter 2 Materials and methods.….…………………….…............................................6 2.1 Human protein interaction network……………………………………………..........6 2.2 mRNA expression profiles………………………………….……………...…………6 2.3 Construction of differentially co-expressed PIN ………...…………..….…..….……7 2.3.1 Spearman’s correlation coefficient…………………………………....................7 2.3.2 Differentially co-expressed PIN …………...………………………….…..…….7 2.4 Identification of disease-specific functional modules………………..……..….….….8 2.4.1 Gene ontology…………………………………….………..…….………………8 . 0$.

(9) 2.4.2 Hypergeometric test………..……………………………………….……………9 2.4.3 Enrichment map………………………….……………………..……...……….10 2.4.4 Modules……………………….………………………………….………..…....10 2.5 Network properties…………………………………….…………....……………….10 2.5.1 Degree………………………………………………………………...………...10 2.5.2 Clustering coefficient…………………………………….………………..........11 2.5.3 Betweenness centrality………………………………...………………………..11 2.5.4 Closeness centrality………………………………………...…………………..11 Chapter 3 Results………………………………………………………..…………….….12 3.1 Identification of dysregulated PPIs in SMA and ALS ………………..……….……12 3.2 Recognitions of enriched modules………………………………………..…….…...13 Chapter 4 Discussion………………………………………………………………...…...15 4.1 ‘Muscle system process’…………………………………………………………….15 4.2 ‘Proteasome-mediated ubiquitin-dependent protein catabolic process’…...………..16 4.3 ‘Respiratory electron transport chain’…………………………………..……...…...18 4.4 ‘RNA splicing’……………………………………………………………………....20 Chapter 5 Conclusion…………………………………………………………..…….…..22 References…………………………………………………………………………...……..23 Figures…………………………………………………………………………………...…29 Tables…………………..……………...……………………………………………….…..43. . 0$$.

(10) List of Figures Fig. 1. Experimental design……………………………………………………29. Fig. 2. Comparison between SMN1 and SMN2……………………...…….…..30. Fig. 3. Illustration of DCPPIs groups classification…………………………...30. Fig. 4. Expression values of SMA………………………...…………………...31. Fig. 5. Expression intensity of ALS……………………………………………31. Fig. 6. Expression intensity of normal…………………………………...…….32. Fig. 7. Fisher’s z transformed SCC histogram of SMA, ALS and normal…….32. Fig. 8. Distributions of the Fisher’s z transformed SCC differences in SMA and ALS.…………………………………………………………………….33. Fig. 9. Overviews of genes and DCPPIs derived from DCPINs of SMA and ALS……………………………………………………………………..33. Fig. 10. Network properties of integrated PPIs, and DCPINs of SMA, ALS and common ………………………………………………………….…….34. Fig. 11. Clusters found in Enrichment map of common GO terms derived from DCPINs. of. SMA. and. ALS. are. shown. with. different. background …………………………………...………………………..35 Fig. 12. Module of ‘muscle system process’……………………………………36. Fig. 13. Module of ‘proteasome-mediated ubiquitin-dependent protein catabolic process’…………………………………………………………………37. Fig. 15. The illustrations showed the procedure of ubiquitination and proteolysis, and structure of 26S proteasome……………………………………….38. . Fig. 15. Module of ‘respiratory electron transport chain’………………………39. Fig. 16. Protein complexes involved in respiratory electron transport chain…...40 0$$$.

(11) . Fig. 18. Module of ‘RNA splicing’……………………………………………...41. Fig. 19. The diagram summarized the overall results found in this study………42. $2.

(12) List of Tables Table 1.. Overall information of SMA and ALS DCPINs………………………….43. Table 2.. Information of 1313 common DCPPIs found in SMA and ALS DCPINs …………………………………………….…….……………….43. Table 3.. Network properties of top 1% degree’s genes of SMA DCPIN ……….…44. Table 4.. Network properties of top 1% degree’s genes of ALS DCPIN …………..45. Table 5.. Network properties of top 1% degree's genes of common DCPPIs between DCPINs of SMA and ALS ………………………………………..……...46. Table 6.. Detail information for common and more specified terms classified by node- and edge-based GO functions………………………………………47. Table 7.. DCPPIs in the module: 'muscle system process'………………………….49. Table 8.. Network properties of 'muscle system process' for genes derived from DCPINs of SMA and ALS………………………………………………..51. Table 9.. DCPPIs in the module: 'proteasome-mediated ubiquitin-dependent protein catabolic process'…………………………………………………….……54. Table 10.. Network properties of 'proteasome-mediated ubiquitin-dependent protein catabolic process' for genes derived from DCPINs of SMA and ALS……62. Table 11.. DCPPIs in the module: 'respiratory electron transport chain'……………..66. Table 12.. Network properties of 'respiratory electron transport chain' for genes derived from DCPINs of SMA and ALS…………………………………68. Table 13.. DCPPIs in the module: 'RNA splicing'……………………………………70. Table 14.. Network properties of 'RNA splicing' for genes derived from DCPINs of SMA and ALS……………………………………………....…………….78. . 2.

(13) Chapter 1 Introduction 1.1 Motor neuron diseases The motor neuron diseases (MNDs) are neurological disorders that selectively affect motor neurons controlling voluntary muscle activity. Spinal muscular atrophy (SMA) and amyotrophic lateral sclerosis (ALS) are the two devastating MNDs, where SMA is one of the most severely infantile and genetic disease, and ALS is the ultimately fatal disease with a fast progression in mid-age adults [1, 2]. 1.1.1 Spinal muscular atrophy Spinal Muscular Atrophy (SMA) is an autosomal recessive disease which occurs in approximately 1 in 10000 new born babies, which is one of the most common genetic disease in infant mortality [3]. SMA is characterized by degeneration in motor neuron in the spine and by atrophy of skeletal muscle. Survival motor neuron gene 1 (SMN1), a SMA-determining gene, is usually homozygous deleted or mutated in most of the SMA patients [4, 5]. SMA is categorized into type I, II, III and IV, depending on the age of onset and clinical progression [4]. Type I SMA (Werdnig-Hoffmann disease) is the most severe class and diagnosed with muscle weakness and hypotonia within 6 months of birth. SMA type I patients are unable to walk and sit unaided. Type II SMA (Dubowitz Disease) is the intermediate class with an onset of weakness within 18 months of age, patients with SMA type II disease are able to sit independently and walk with assistance. Type III SMA (Kugelberg-Welander Disease) is the mild class with an onset of symptoms after 18 months of age, they can walk normally until they lost the ability later in life. The patients in type IV SMA are characterized with an onset age . >.

(14) older than 30-years-old as well as the mildest manifestation. It has been reported that the copy number of SMN2 gene, which is nearly identical to SMN1, modifies the severity of phenotype [6]. SMN2 copy numbers are inversely correlated to SMA types. The major difference between SMN1 and SMN2 is that SMN1 gene produces full-length transcripts; in contrast, SMN2 gene mainly produces alternatively spliced transcripts, SMN7 (Figure 1) [7]. In short, full-length functional SMN protein produced by SMN2 unable to overcome the loss of transcripts of SMN1 leading to SMA. Currently, there are two hypotheses for the SMN deficiency lead to SMA: 1) SMN complex consisting of SMN, Gemin2-8 and unrip is characterized to engage in ATP-dependent snRNPs assembly [8, 9]. The dysfunction of snRNPs assembly might play a major role in SMA pathology. 2) SMN proteins are important for neurite outgrowth, neuromuscular maturation and axonal transportation [10, 11]. Thus, down-regulated SMN expression may lead to selective degeneration of motor neuron with unclear mechanism.. 1.1.2 Amyotrophic lateral sclerosis The epidemiologic measures of disease frequency of Amyotrophic lateral sclerosis (ALS) is about 0.3 to 7 in 100000 each year [12]. ALS is categorized into two groups: familial inherited ALS (called fALS) and non-familial history discovered ALS (called sporadic ALS, sALS). There are about 10% of fALS and 90% of sALS cases are discovered in a survey of the literature [13]. ALS is a neurodegenerative disease and is fatal because of its fast death rate of motor neuron in the brain, brainstem and spinal cord after onset. . ?.

(15) ALS is a multiple causing disease, mutation in RNA binding proteins such as superoxide dismutase 1 (SOD1), fused in sarcoma (FUS/ TLS) or TAR DNA binding protein (TARDBP) was found in ALS patients [14-18]. Mutated SOD1 enzyme links to toxic gain with generation of free radical and eventually leads to cell death [19, 20]. FUS and TDP-43 proteins are both involved in transcription, splicing and other processes in nuclear and cytoplasm [21]. Therefore, part of ALS may interact with protein function in splicing and leads to motor neuron degeneration.. 1.1.3 Interaction between SMA and ALS Several studies have suggested the relationship between SMA and ALS in molecular levels. SMN complex is localized in nuclear Gems and cytoplasmic compartment [22], and loss of Gems is a hallmark in SMA. Gems are also found to be losing in motor neurons of the TDP43 knockout mouse that is the common model for studying ALS [23]. Yamazaki and his colleagues showed that aggregated FUS mutant protein could trap SMN protein and caused SMN distribution imbalanced in nuclear and cytoplasmic, which might drive motor neuron death [24]. From these studies, we learned that SMA and ALS might be related to motor neuron diseases, but the correlation between SMA and ALS is still unclear and being mapped.. 1.2 Protein interaction network Protein-protein interactions (PPIs) are one of the central roles in biological processes. PPIs can be constructed into protein interaction networks (PINs) and give us a relative macroscopic view in molecular mechanism. Protein connectivity of the PINs follows power-law distribution indicating that the most connected protein has less . @.

(16) abundance in comparison to small amounts of high degree protein (called hubs) [25]. PINs can be mined into smaller and meaningful modules that representing different functions [26, 27]. Topological properties of the discovered conditional PINs are important indications and can help us choose interesting targets. Higher degrees of the proteins (nodes) tend to be more important and less abundant in the PINs. Clustering coefficient represents whether the neighbors of a particular protein are closely connecting or not. Betweenness and closeness centrality show that if the interesting protein can be centers of the clusters existing in the PIN.. 1.3 Microarray analysis Microarray have been applied in lots of analyses and is a powerful technique for high-throughput exploration of gene expression profiles [28-30]. The quantification and quantitation of gene expression profiles are based on nucleotide hybridization and fluorescence detection. Integrating microarray expression profile with PIN is able to discover biomarkers in subnetworks which has been shown to provide more accurate result than simply search for differentially expressed genes in a previous study [31].. 1.4 Integrative analysis of protein interaction network Though PINs contain all static information, not all proteins interact at the same time. Active PPIs depend on the protein expressions. To discover the complicated processes in cells, PINs can be served as backbone to construct condition-activated PINs. To systemically discuss between SMA and ALS, we combined PINs with . A.

(17) microarray profiles to construct differentially co-expressed PINs (DCPINs) of SMA and ALS against normal (Figure 1). Due to the shortage of proteome data, so we used mRNA expression values to represent protein expression profiles in our study. To further discuss whether there is any biological process involved in motor neuron degeneration or muscle atrophy, we try to mine functional modules by classifying these DCPINs with gene ontology (GO). Finally, we would like to find the common and different mechanisms leading to SMA and ALS.. . B.

(18) Chapter 2 Materials and methods 2.1 Human protein interaction network PINs was downloaded and integrated from 5 online databases [32-36]. Followings are the detail information of collected PPIs: MINT (2013-3-26), BIOGRID (2014-12-1, version: 3.2.119), DIP (2014-10-1), HPRD (2010-3-13, version: release 9) and IntAct (2014-11-10). Followings are the principles for construction of human PIN backbone. First, we only retained human PPIs (taxID: 9606). Second, all protein identifiers (uniprotKB, refseq, and ensemble) were mapped to Entrez gene IDs and symbols (mapping information are collect from NCBI) and unknown/unmapped identifiers were filtered out. Third, PPIs with self interactions and duplication were removed to keep PIN backbone non-redundant.. 2.2 mRNA expression profiles The SMA and ALS mRNA expression profiles were obtained from Gene Expression Omnibus (GEO). 9 SMA samples (18 cDNA microarrays containing technical repeat of each patient with dye swap labeling), GSE8359 [37], and 9 ALS and 18 normal samples (Affymetrix chip of HG-U133A), GSE3307 [38, 39], were all muscle biopsies and used in my study. R package ‘limma’ was applied to do background correction (RMA algorithm) and intra-array normalization (loess) for GSE8359 [40]. R package ‘affy’ was used for background correction (RMA algorithm) in GSE3307 [41]. R package ‘preprocessCore’ was used to do between-array normalization in both datasets independently [42]. The expression values of replicate probes were averaged. In order to keep . C.

(19) maximum information from both chips, the probes related to multiple genes were reserved and the genes were assigned with the same values for further network analysis.. 2.3 Construction of differentially co-expressed PIN 2.3.1 Spearman’s correlation coefficient Before direct calculating Spearman correlation coefficient (SCC), we need to replace the expression values with rank numbers. SCC of paired genes (X and Y) that encode proteins in PIN is defined as following: Where n is the number of sample of the target (9 SMA, 9 ALS or 18 normal); Xi and Yi are rank-transformed expression values of gene X and Y in different sample of target. The larger the absolute value of SCC is, the higher the correlation between the calculated gene pairs is.. 2.3.2 Differentially co-expressed PIN After calculating SCC values of SMA, ALS and normal independently, the SCC is converted to Z-score with Fisher’s transformation listed as following equation. Then we want to find the differentially co-expressed PPIs (DCPPIs, P< 0.05) between diseases and normal, by means of finding the significant Z-score differences (Z-scoredisease – Z-scorenormal) between disease and normal.. . 𝑛 . The significant P value derived from Z-score difference between diseases and normal were found by permutation test, which is applied to check if DCPPIs were . D.

(20) found by chances or not. In short, the gene expression values of gene A and B found in PPIs, were random ordered independently in disease and normal for 10000 times. Then the random Z-score and Z-score difference of disease and normal were calculated in each shuffling. P value is defined as the proportion of random Z-score differences that are bigger (or smaller) than the originally positive (or negative) Z-score difference. Here, differentially co-expressed PIN (DCPIN) is defined as the set of DCPPIs found in the disease. SMA and ALS DCPINs are constructed individually and used for further analysis in this study. Further more, DCPPIs with Z-score larger (smaller) than or equal to 1.96 (-1.96) were defined as positive (negative) correlation in all study groups (SMA, ALS and normal). DCPPIs with Z-score within -1.96 and 1.96 were non-correlated. To see what types of the DCPPIs were found in both DCPINs, DCPPIs were classified into 5 groups (Figure 3): gain of negative correlation (GoN), loss of positive correlation (LoP), loss of negative correlation (LoN), gain of positive correlation (GoP) and others (other_P, other_N, PP and NN).. 2.4 Identification of disease-specific functional modules To mine functional modules in DCPINs, gene ontology (GO) was used to find what biological processes are enriched and whether there is any biological process involved in muscle atrophy or motor neuron degeneration with node and edge-based hypergeometric tests.. 2.4.1 Gene ontology We used the gene annotation by Gene Ontology (GO) [43] for the functional classification of DCPINs. In GO, the genes in each term are predefined gene sets. . E.

(21) There are three ontology categories in GO, namely biological process, molecular function, and cellular component, but we only considered biological process GO for our analysis. The information of GO terms and its consisted genes are built by R package ‘org.Hs.Eg.db’ [44]. There were 18083 unique genes (in gene IDs) and 12565 unique GO terms.. 2.4.2 Hypergeometric test The normal hypergeometric distribution (node based GO enrichment) is defined as following:. . . . . . X: the evaluated functional category in GO; N: genes appeared in our microarray and PPIs; m: genes we are interested with (e.g. genes found in disease DCPIN); n: numbers of gene in X. The formula calculates the over-represented probability of the evaluated functional category containing k genes in the network. To further obtain denser functional subnetworks, we used an edge based functional enrichment as our previous study [27]. By considering whether PPIs in disease DCPIN are also enriched in the same evaluated functional categories. To achieve this analysis, we just required to change all the parameters from inputs of genes to inputs of DCPPIs for modified hypergeometric test as following:. . . . . X: the evaluated functional category in GO; Ne: all collection of GO annotated PPIs; me: DCPPIs found in disease DCPIN; ne: numbers of PPIs in X. The formula . F.

(22) calculates the over-represented probability of the evaluated functional category containing k DCPPIs in the network. FDR (Benjamini & Hochberg) is applied to correct P values [45]. GO functions whose FDR< 0.05 in both node and edge based methods are remained.. 2.4.3 Enrichment map To help us overall look for what functions were major enriched in our study, enrichment map was built as followings. First, we isolated common GO terms enriched in both SMA and ALS DCPINs. Second, we pooled enriched genes in each node based function. Third, we calculated the similarity between GO terms. The similarity values are manipulated as mean of Jaccard and Simpson indexes. Followings are formulas of Jaccard (JAB) and Simpson (SAB) indexes: . . Here, A and B are two different GO terms, |N(A)| and |N(B)| which means numbers of genes in A and B independently.. 2.4.4 Modules Common GO functions with higher GO level are considered as our particular targets in both SMA and ALS. DCPPIs in some of these functions are isolated and considered as modules and being further discussion.. 2.5 Network properties Some basic network properties are calculated in this study by ‘networkx’ [46]. For protein interaction network, nodes represent proteins and edges are PPIs. Followings are topological parameters used to help us pick important targets in the PIN.. 2.5.1 Degree . >=.

(23) Degree (k) means the numbers of linkages of the node in the network. As for the PIN, proteins with high degree are called ‘hubs’.. 2.5.2 Clustering coefficient Clustering coefficient (C) of a protein means how frequently of its linking proteins interact with each other. It is defined as: . . . where eNBj is the number of interactions between interacting partners of protein i, and NBj is the numbers of its interacting partners. C2NBj represents the numbers of all possible interactions between its interacting partners.. 2.5.3 Betweenness centrality Betweenness centrality (BC) indicates the centrality of a protein i in the PIN, and it is defined as: . . here, SPi means the number of the shortest path passing through protein i, and protein N is the number of proteins in PIN.. 2.5.4 Closeness centrality Closeness centrality (CC) represents how close a protein is against all the others and it is defined as the reciprocal of the mean of the shortest path lengths (SPLi) for protein i to all reaming proteins in PIN. . . . >>.

(24) Chapter 3 Results To obtain the as complete human protein interaction networks as possible, we compiled PPIs from five databases with latest version. After merging these databases and removing the redundancies, we obtained a human PIN with 197124 interaction and 16541 proteins. Because we performed a GO enrichment analysis for further identification of functional modules, the interacting proteins with the annotated GO terms were named as GO annotated PPIs. This process resulted in 166520 GO annotated PPIs containing 12797 proteins. The raw SMA and ALS gene expression data obtained from GEO were normalized and pre-processed (Figure 4-6). After preprocessing, a total of 2732 common genes existed in SMA, ALS and normal. Among the common genes, 2270 genes were found to form 20276 PPIs and be used for further SCC calculation.. Next, we integrated PIN and gene expression information to identify differentially co-expressed PINs (DCPIN) of SMA and ALS. The differentially co-expressed PPIs might provide the insight into the pathogenic mechanisms of these two diseases. We could compare the DCPIN to investigate the relationship between SMA and ALS.. 3.1 Identification of dysregulated PPIs in SMA and ALS To avoid the outlier effects on correlation measurement, we used Spearman correlation coefficient to measure the correlations of interacting pairs. The Z-score distributions (Fisher’s z transformed SCC) of SMA, ALS and normal are shown as Figure 7. To get the differentially co-expressed PPIs (DCPPIs), the Z-score difference between disease and normal were calculated. The histogram of Z-score difference . >?.

(25) between SMA and ALS against normal is illustrated as Figure 8. Those PPIs passing permutation test (P< 0.05) are called DCPPIs and form as a DCPIN. The overall information of both DCPINs were summarized in Table 1 and Table 2. DCPIN of SMA contains 3770 interactions and 1588 gene, and DCPIN of ALS contains 3761 interactions and 1572 genes. 1329 common genes (72.58%) are found in DCPINs of SMA and ALS, and 1313 common DCPPIs (21.12%) are found (Figure 9). The results showed that SMA and ALS might share parts of commonly dysregulated processes. Additionally, there are more loss-of-coexpression (either loss of positive correlation or loss of negative correlation) in both diseases (SMA and ALS DCPPIs: 47.21% and 44.96%), showing that losing correlation might play slightly more effect on both diseases. To overall look at the topological structures of SMA and ALS DCPINs, the network properties of SMA and ALS DCPINs were investigated independently. Both DCPINs follow power-law distributions (Figure 10, a) and those genes with higher degree might be important for SMA and ALS. Top 1% higher degree genes are list in Table 3 & 4. Though these genes with higher degree seem to be less clustered structure (Figure 10, b), they are close to center in both DCPINs (Figure 10, c&d). We then explored the network properties of 1313 common DCPPIs (composed with 1003 genes) between SMA and ALS. The genes with top 1% degree are list in Table 5. These genes might play important role in connections between SMA and ALS.. 3.2 Recognitions of enriched modules GO annotation were used to classify the DCPINs into modules and help us to . >@.

(26) understand what processes do these DCPPIs get involved in each DCPIN. As our previous study, node and edge based (“dyads”) GO enrichments were used to identify the significantly GO terms (FDR< 0,05 in both methods) [27]. GO are used to find functions in DCPINs of SMA and ALS independently with both node and edge based hypergeometric test. GO terms passed FDR< 0.05 with both node and edge based enrichments were collected as our primarily interesting targets. There were 177 and 223 terms enriched in SMA and ALS DCPINs individually. There are 165 common GO terms in both diseases. To overall look at these biological processes, GO enrichment map was built for visualization (showed as Figure 11). The common and more specified GO functions are listed in Table 6. GO terms, such as ‘muscle system process’, ‘proteasome-mediated ubiquitin-dependent protein catabolic process’, ‘respiratory electron transport chain’ and ‘RNA spicing’, which might be related to pathogenic mechanisms of SMA and ALS were identified. DCPPIs existing in these four GO terms were isolated and become our targeting modules.. . >A.

(27) Chapter 4 Discussion By analysis of DCPINs of SMA and ALS, GO terms of common or different DCPPIs were identified in this study. Most specified GO terms (highest levels of GO) mined from clusters in the enrichment map are our targets (Figure 11; Table 6). In our studies, we especially discussed four particular modules that might relate to the SMA and ALS functions.. 4.1 ‘Muscle system process’ First of all, muscle system process that regulates contraction and relaxation of muscle fibers are investigated in our study due to the phenomenon of muscle twitch/ fasciculation in SMA and ALS [47-49]. There are total of 27 (34 genes) and 26 (34 genes) DCPPIs in the module of ‘muscle system process’ from DCPINs of SMA and ALS (Figure 12; Table 7). The network structures of this module are different in both diseases (Table 8). ACTN2, TNNI1 and TTN (k=3) are the top three genes with highest degree in SMA group. In contrast, TTN (k=5), TNNI2 (k=4), TNNT1 and ACTA1 (k=3) are the genes with higher degree in ALS group. 11 common DCPPIs are identified in this module, and two of them are responsible for an important role in controlling muscle movement and stress response. Recent SMA and ALS studies discover the disruption of calcium pathway in both motor neurons and skeletal muscles [50, 51]. Calcium is released from sarcoplasmic (endoplasmic) reticulum (SR/ER) to cytoplasm when action potential is transmitted from the motor neuron to the muscle, which causes muscle contraction. When calcium is finally recaptured back to SR, the muscle relaxation occurs. . >B.

(28) Furthermore, when SLN (Sarcolipin) binds to ATP2A1 (Sarcoplasmic/ endoplasmic reticulum calcium ATPase 1, SERCA1), it inhibits calcium translocation from cytoplasm to lumen of SR followed by thermogenesis [52]. Interestingly, we found the gain of positively correlated DCPPI of ATP2A1 and SLN in this module that might give us the insight for calcium remaining in cytoplasm and disrupting muscle relaxation in SMA and ALS. TTN (titin) is a crucial protein that involved in muscle movement. It is the largest spring-like protein served as an anchor for myosin, stretches as muscle excitation and relaxation [53]. A chaperone of small heat shock protein, CRYAB (alphaB-crystallin), is shown to retain stabilize and refold TTN back once stress is detected or not [54]. DCPPI of TTN and CRYAB is also found to be gain of positive correlation. This interaction might imply the response of dysregulated calcium (prolonged muscle contraction or heat generation) might occur in SMA and ALS.. 4.2 ‘Proteasome-mediated ubiquitin-dependent protein catabolic process’ Proteasome degradation has been reported in many motor neuron diseases including SMA and ALS [55-57]. There are 92 (62 genes) and 98 (61 genes) enriched DCPPIs in the module of ‘proteasome-mediated ubiquitin-dependent protein catabolic process’ from DCPINs of SMA and ALS (Figure 13; Table 9). Network topologies of this module are listed in Table 10. For SMA group, PSMD4 (k=10), RAD23A and UBC (k=8) are the top three nodes with highest degrees in this module. And, the top four high degree nodes are UBC (k=16), PSMC4 (k=11), PSMD4 and PSMD6 (k=8) in ALS group. . >C.

(29) RAD23A (UV excision repair protein RAD23 homolog A) was first found to get involved in nucleotide excision repair [58]. Besides DNA repair mechanism, RAD23A also contains both ubiquitin associated (UBA) and ubiquitin-like (UBL) domains that has been suggested to play a role in protein degradation. UBA domain of RAD23A has been shown to interact with poly-ubiquitinated proteins and being shuttle to transport the poly-ubiquitinated proteins to the proteasome subunit, PSMD4 (S5a/ RPN10) that recognize the UBL domain of RAD23A for protein degradation [59, 60]. We found that RAD23A is the highest degree of 9 DCPPis in SMA DCPINs in this module. The role of RAD23A and its substrates for proteasome degradation should be further explored in SMA. 35 common DCPPIs found in this module, most of which are derived from high-throughput experiments and containing protein complex interactions. We only discuss the common DCPPIs with proves of direct interaction here. The process of ubiquitin-dependent and proteasome-mediated degradation requires poly-ubiquitin to be added onto specified lysine sites of a protein and then the proteolysis can be performed in 26S proteasome (Figure 14). Three enzymes are involved. in. the. ubiquitination:. E1. (ubiquitin-activating. enzyme),. E2. (ubiquitin-conjugating enzyme) and E3 (ubiquitin ligase). STUB1 (CHIP, a E3 ligase), gained positive correlation with UBE2D3 (a E2 enzyme) might indicate the increases of ubiquitination for the substrates (proteins) of STUB1 and UBE2D3 in both diseases. However, STUB1 decreased interact with PSMD4, the polyubiquitin receptor sites for recognition of proteolysis, in SMA and ALS. We come up with two possible hypotheses for this phenomenon. First, this . >D.

(30) might imply that the substrates of STUB1 are monoubiquitinated or incomplete polyubiquitinated proteins that are unable to be recognized by proteasome for degradation. Second, these polyubiquitinated proteins would aggregate in the cells. In previous study, when CLU and HSPA5 (GRP78) interacted together, it tends to bind to an unfolded protein which partially prevent cell apoptosis caused by ER stress in hepatocellular carcinoma cells [61]. We found that CLU and HSPA5 gain positive correlation in ALS, which indicates the activation of ER stress caused by calcium cycle disruption or protein aggregation. Similarly, in our study, HSP90B1 is lost of positive correlation with UBC, a poly-uqbiquitin precursor in both SMA and ALS. This shows that HSP90B1 could not be degraded easily and might be responsible for dysregulated calcium homeostasis, heat stress or trap the accumulated proteins in SMA and ALS.. 4.3 ‘respiratory electron transport chain’ Decreasing of activities in mitochondrial complexes has been found in both SMA and ALS in the previous studies [62-64]. There are 23 (28 genes) and 27 (28 genes) enriched DCPPIs in the module of ‘respiratory electron transport chain’ from DCPINs of SMA and ALS (Figure 15 and Table 11). Detail network structures are listed in Table 12. NDUFA2 (k=7), ATP5B (k=4) and CYCS (k=3) are the genes with higher degree in SMA. ATP5B, ATP5F1, ATP5C1, UQCRQ, and SDHA (k=4) are the highest degree genes in ALS. The DCPPIs found here are mainly in the subunits of complexes that involved in respiratory electron transport chain (oxidative phosphorylation) and located at the mitochondrial inner membrane. The roles of these complexes (I, II, III, IV, V) are . >E.

(31) summarized and illustrated in Figure 16. For SMA group, the DCPPIs (NDUFA2 – NDUFA8; NDUFA2 - NDUFB11; NDUFA9 – NDUFV1; NDUFS1 - NDUFS2; NDUFS6 – NDUFS8) identified here are dysregulated in the complex I, and lost of positive correlation comparing to normal. These might cause decreasing amount of proton gradients and reductive form of UQ, and eventually reduce the rate of ATP synthesis. For ALS group, there are 3 DCPPIs gaining of positive expression between SDHA (a member of complex II) and NDUFS6, NDUFS8, NDUFV2 (subunits of complex I) in ALS group. A previous study showed that supercomplex formation between complex I and III, and IV but not complex II as we found in this module [65]. These three DCPPIs were discovered by high-throughput data of soluble complex and identified as co-fraction [66]. Therefore, these PPIs might not really exist. In contrast, there are 4 DCPPIs composed in complex V (ATP5A1 – ATP5C1; ATP5A1 – ATP5F1; ATP5B – ATTP5C1; ATP5B – ATP5F1) which might cause defects in ATP synthesis. There are 8 common DCPPIs, and 6 DCPPIs are lost of positive correlation among them. They are major dysregulated in complex I (NDUFA2 – NDUFA9; NDUFA2 – NDUFA13; NDUFA2 – NDUFB7), and complex III (CYC1 – CYCS; CYC1 – UQCR10). Moreover, we found that CYC1 and CYCS lost positive correlation or from positive to negative correlation in ALS and SMA. CYC1 captures electron from UQH2 and passes it to CYCS in complex III. Lost of positive correlation of CYC1 and CYCS might prevent the reduction of oxygen to water and reduce proton gradient. The components for decreasing of ATP synthesis in mitochondrial complex are . >F.

(32) different in SMA and ALS. Complexes I might major involve in SMA group and Complex V might major involve in ALS group. Complex III might involve in both SMA and ALS.. 4.4 RNA splicing As our expectation, we identify modules related to ‘RNA splicing’ because one role of SMN protein is to help snRNPs assemble for splicing [67, 68]. There are 109 (85 genes) and 102 (76 genes) DCPPIs in module of ‘RNA splicing’ derived from DCPINs of SMA and ALS (Figure 17 and Table 13). 39 common DCPPIs are found among them. The network properties of this module are listed in Table 14. HNRNPK (k=9), SFPQ and SRRM1 (k=7) are the genes with higher degree in this module for SMA group. In contrast, SMN2, HNRNPU (k=10), and YBX1 (k=9) are the higher degree nodes in ALS. Small nuclear ribonucleic proteins (snRNPs) are consisted of snRNAs, heptameric Sm proteins (Sm B/B’, D1, D2, D3, E, F and G) and other associated proteins to form spliceosome (U1, U2, U4, U5, U6) that are involved in pre-mRNA splicing. SMN complex (SMN and Gemin 2-8) is helped to co-operated with other related protein complexes. A review paper showed detail process of spliceosome formation and the splicing processes [68]. Disrupted formation of snRNPs’ biogenesis has been studied in SMA patients [69]. SNRPD1 (Sm D1), SNRPD2 (Sm D2), SNRPE (Sm E), SNRPF (Sm F) and SNRPG (Sm G) are members of heptameric Sm proteins, which are parts of mature snRNP (U1, U2, U4, U5). LSM3, LSM5, LSM7 and LSM8 are members of heptameric . ?=.

(33) Sm proteins consisted in U6. We find that these 5 DCPPIs (SNRPD1 - SNRPE; SNRPD2 - SNRPF; SNRPE – SNRPG; LSM3 – LSM7; LSM5 –LSM8) are either lost of positive correlation or gain of negative correlation in both diseases, which might imply the defects in formation of snRNPs. Recent studies showed that some alternative splicing mRNAs is due to SMN deficiency that might play an important role in SMA and ALS pathology. So, it is important to screen more deeply with RNA-seq to include effects of alternative splicing. Results found in this thesis are summarized in Figure 18.. . ?>.

(34) Chapter 5 Conclusion By analyzing disease specified DCPINs, we found the common and dysregulated PPIs which might shed light on the connection between common phenotypes in SMA and ALS. The gain of interaction between SLN and ATP2A1 showed the possible mechanism leading to irregulation of calcium cycle in both SMA and ALS. The stress responses might further prove the importance in temporal regulation of calcium in SR and cytoplasm. Besides, deregulations in protein ubiquitination and proteasomal degradation were also found in SMA and ALS, and the chance leading to further protein aggregation should be concerned.. . ??.

(35) Reference >8. ,'$)$5 85 .'85'$!'+!+!&+(!&$%-+-$*,*'( 14&,-*$ !+,'*1& %&%&,5 0 #$' /,*'5>FEF8;A<7+8B=>9E8. ?8.

(36) $ ,))5 885. . '85 %1',*'( ! $,*$ +$*'+!+5 ) .5 ?=>>8. ;FDCF<7+8FA?9BB8 @8. ,4/-.*1$45 885 .'85&,!%((!&' *'&! !$ ''6'&+,+(!&$ %-+-$*,*'( 1,' *'%'+'%C)??5@6?A5A5./, 5>FF=8;C?CC<7+8 BA=9>8. A8. ! 0, 585 .'85 &,!!,!'&& *,*!2,!'&'+(!&$%-+-$* ,*'( 16,*%!&!&&5 ''5>FFB8 ;><7+8>BB9CB8. B8. '$-5 85 . '85

(37) -,,!'& -(, ' +(!&$ %-+-$* ,*'( 1 !& (!&4 %'$-$* *,*!2,!'& ' EBC -&*$, (,!&,+ & !&,!!,!'& ' '-* &'.$ %-,,!'&+ !& ,

(38) ? &5 /( ) .5 ?==F8 ;><7 +8 ?F9@F8. C8. $.*5 85 . '85 :-&,!,,!. &$1+!+ '

(39) @ + '& *$6,!% 4 '**$,!'& ' $!&!$ +.*!,1 &

(40) @ & '+;5 * * ...-/5 ?==B8;B<7+8A=D9>?8. D8. *))$5 885 . '85 +!&$ &-$',! !*& , , $,*+ +($!!& (,,*&+!+,!&-!+ +,

(41) &

(42) ?*'%, '(1&

(43) @5 /( *' ) .5>FFF8 ;D<7+8>>DD9E@8. E8. ''$44*)$5 85 8*)"5)8, 3!/--5++&,!$*'$'*,

(44) '%($0!& , +(!!!,1'+& ++%$15$ ) 5?==?8 ;BBFF<7+8>DDB9F8. F8. ,$--$($585 .'85&*!(!+'%('&&,'

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

(46) C?@9D8 >A8. *-*5 885 . '85 !$6,1( & %-,&, ? + * & **&, '&'*%,!'& & '%%'& (, '&! (, /1 !& 5 . /,*-$5 ?=>=8;>><7+8>@FC9A=@8. >B8.

(47) 1$.&*1-&$58 85 ,85 .'85

(48) -,,!'&+!&, 7 &'& *'%'+'% ?D-+%!$!$%1',*'( !$,*$+$*'+!+5$ ) 5?==F8;BF>E<7 +8>?=B9E8. >C8. /())5 85 . '85 !)-!,!&, 6BA !& *'&,',%('*$ $'* &*,!'&&%1',*'( !$,*$+$*'+!+5$ ) 5?==C8;BDFC<7 +8>@=9@8. >D8. *- )5 885 . '85

(49) -,,!'&+ !& -7& +-(*'0! !+%-,+ & * ++'!, /!, %!$!$ %1',*'( ! $,*$ +$*'+!+5 ./, 5 >FF@8 ;CA>B<7+8BF9C?8. >E8. , #,)5 85. . '85 6BA %-,,!'&+ !& %!$!$ & +('*!. %1',*'( !$,*$+$*'+!+5$ ) 5?==E8 ;BED=<7+8>CCE9D?8 >F8. /$5 8 ) 88

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

(51) (*',!& !&,*,!'&+ $!&# , %','* &-*'& !+++ &

(52) 5 '' +5?=>?8;A<7+8DFF9E=C8. ?B8. *)"5 85 .'85 , $!,1&&,*$!,1!&(*',!&&,/'*#+5./, 5?==>8 ;CE@@<7+8A>9?8. ?C8. $0 -588)8'$.-&$5

(53) '-$*'*&!2,!'&'$$-$*&,/'*#+5,* .'$5?==@8;@<7+8>>?E9@@8. ?D8. $)5 885 . '85 1&%! -&,!'&$ %'-$+ !& '60(*++ (*',!& !&,*,!'&&,/'*#+'!$,*!'%1'(, 153-.$*'5?=>=87+8. . ?A.

(54) >@E8 ?E8. # )585 .'85-&,!,,!.%'&!,'*!&'&0(*++!'&(,,*&+/!, '%($%&,*1 %!*'**15$ ) 5>FFB8;B?@B<7+8ACD9D=8. ?F8. *-5 885 8 *-5 ) 88 +,35 =(, '$'= ' , &+ 0(*++!'&(*'!$3&/,''$!&!&!&3.$-,!&3&$++!1!&&,! !+++5*( #3-$*'5?===8;>9A<7+8BF9CD8. @=8. #$5 85

(55) 8 *) 5 ) 8 /5 &,*,'%6,*&+*!(,'% &$1+!+ *.$+ , ! &,*$!,1 ' &+ !*&,!$$1 0(*++ !& $-& &* ,!++-+5$*$)!*,(.$-5?==B8;?@<7+8A?=B9E8. @>8. #/)"5 885. . '85 ,/'*#6+ $++!!,!'& ' *+, &*. %,+,+!+5*'3-.$*'5?==D87+8>A=8 @?8. $.5 85 . '85

(56) 3 , %'$-$* !&,*,!'& ,+4 @>?@ -(,5 /' $$- -5?=>?8;.- $--/ <7+8EBD9C>8. @@8. ,#,5 85 . '85

(57) &,, (*'",66 &,, + '%%'& -*,!'& ($,'*%'*??%'$-$*!&,*,!'&,++5/' $$- -5?=>A8 ;.- $--/ <7+8@BE9C@8. @A8. ,-5 885

(58) 8

(59) )-(35 ) 8 ) 35 -%& *',!& *& ,+ & -%& *',!&(! + !+'.*1 ,''$+ '* +1+,%+ !'$'15 .#*-*'$*'5?==F87+8CD9DF8. @B8. .,&5 85 . '85 !' 4 &*$ *('+!,'*1 '* !&,*,!'& ,+,+5 /' $$- -5?==C8;.- $--/ <7+8B@B9F8. @C8. ),$*-5 85 .'85 4, ,+'!&,*,!&(*',!&+5/' $$- -5?===8 ;><7+8?EF9F>8. @D8. $''$)*5 85 . '85 !*&, ,*'( 16 1(*,*'( 1 ,*&+*!(,!'& (, /1+ !& %-+$+ , 1 +.* & %!$ +(!&$ %-+-$* ,*'( 15 5 ?==F87+8>A8. @E8. &35 85 . '85 -$* &.$'( 1+,*'( !+ + '/ ,*&+*!(,!'&$ !&*(*!&, +-+,!& !+*-(,!'& ' 6

(60) 1' (, /1+ !& %-+$ *&*,!'&5,$)5?==C8 ;.A<7+8FFC9>=>@8. @F8. ",585 .'85+1& *'&'-+*%'$!&!+*!.*'!$*&*,!'& !&- &&%-+-$*1+,*'( 15 ''$*'5?=>A8;><7+8>@F9BE8. A=8. $.#$ 5 885 . '85 '%(*!+'& ' #*'-& '**,!'& %, '+ '* ,/'6'$'-*%!*'**1+5$*$)!*,(.$-5?==D8;?=<7+8?D==9D8. A>8. /.$ ,5 85 . '85 166&$1+!+ ' 1%,*!0 & !( , , , (*' $.$5$*$)!*,(.$-5?==A8;@<7+8@=D9>B8. . ?B.

(61) A?8. *'-.5 885 (*(*'++'*4 '$$,!'& ' (*6(*'++!& -&,!'&+5 (#.*+!'&?5@D5?8. A@8. -#/,) ,585 .'85&'&,'$'14,''$'*, -&!!,!'&'!'$'15 &&,'$'1'&+'*,!-%5. ) .5?===8;><7+8?B9F8. AA8. ,'-*)5 85 '*5 +554 &'% /! &&',,!'& '* -%&5 (# .*+!'&@5?B5>8. AB8. 0 '*+( ).*, (54 &-&&.!*'&%&,'*,,!+,!$ '%(-,!&5 !&&3 -+,*!4 '-&,!'& '* ,,!+,!$ '%(-,!&5 ,*!.*'%#..+7::11189+,*% .8*,"8. AC8. ((,%/5 88 ) 88 &$ )5 '%(-,,!'&$ !+*, %, %,!+ 4 '%!&,'*!+&*( , '*1/!,

(62) , %,!8?==@5(,$" 58

(63) 86 1*,&7(,$" )$0 ,-$.3, --82$$$5AE=+8. AD8. &*/. '$5 8)8/*1$.45+!-$,!'&', 1$!+4&!,!'&$$- ,' $!&!$ !&'+!+ !& +(!&$ %-+-$* ,*'( 15 /,*+ $.,$-5 >FEA8 ;@<7+8>AB9C8. AE8. )35885(!&$%-+-$*,*'( 1!& !$ ''4*.!/'C>++5 0 #$' /,*'5>FDE8;C<7+8DAC9BD8. AF8. *&/5

(64) 8 ) 8 #$($4/5 :+!-$,!'& (',&,!$+ !& 6+!&!!&3 &*$,!'&+ !(/!, $!&!$,-*+;5$)-#*#$)& $"&/5?=>A8;>?<7 +8>=E@9B8. B=8. /$4585 .'85$,*!&,*$$-$*@H '%'+,+!+!&&*.,*%!&$+' +.*+(!&$%-+-$*,*'( 1%!5 /,*-$5?=>=8;@<7+8EAF9BD8. B>8. #$)5 885 . '85 *,-*,!'&+ !& !&,*$$-$* @H &$!& !& +#$,$ %-+$!&, GA<?%'-+%'$'%1',*'( !$,*$+$*'+!+5( #3-$*' ''#3-$*'5?=>A8;>><7+8>=@>9E8. B?8. #**5 8

(65) 85 . '85 *'$!(!& (*',!& !&,*,!'& /!, +*'8&'9($+%! *,!-$-% @H + 89 !+ !+,!&, *'% ( '+( '$%& (*',!&3 & '&$1 +*'$!(!& & (*'%', -&'-($!& ' , (-%(5 $*' # (5?=>@8 ;>=<7+8CEE>9F8. B@8. #/0 /5 85 8 *1 ''5 ) 8 ,, $,*5 *!+!& ,!,&4 *.!/ & %-,,!'&-(,5 /(/..5?=>A8;F<7+8>=AC9BF8. BA8. /'',585 .'85++'!,!'&', (*'&$( 6*1+,$$!&/!, ,!,!& !& *,%-+$5 $*'# (5?==A8 ;F<7+8DF>D9?A8. BB8. )"585 .'85 &,*,!'&'%1',*'( !$,*$+$*'+!+7*'&,',%('*$ $'* &*,!'&6++'!, -+6!&6+*'% /!, (*',!&+ !&.'$. !&. . ?C.

(66) %,'$! & (*',!& *,!'& (, /1+5 /,*$*' "$)"5 ?=>B8 ;><7+8B?D9@B8 BC8. )5

(67) 8 85 .'85!)-!,!&6+(!!(*',+G0-!)-!,!&,+&+,!$!2+ , +(!&$ %-+-$* ,*'( 1 (*',!&6+-*.!.$ %','* &-*'&5 $*' # (5 ?=>?8 ;B?<7+8A@DA>9B?8. BD8. " ,5 885 88 ,-*)5)8

(68) 3 5 !&,**$,!'&+ !('(*',+'% !%(!*%&,&'$!'%*!!&,*%!,+!&&-*'&*,!'&5"$)" ''5 ?=>B8. BE8. /"-15

(69) 85 . '85 /' -%& '%'$'+ ' @A * -&,!'&$$1 !&,* &$ !& '%($0 '*%,!'& & +,!%-$,!'& ' *(!* ,!.!,15*' ''$*'5>FFD8;>?<7+8CF?A9@>8. BF8. # )5 8 )

(70) 8 /,5 .!& '* !+,!&, -&,!'&+ '* -%& *(!*,'*+ @A& @A5 ..5?==C8 ;>A<7+8@A=>9E8. C=8. .$(35 885 . '85 *'++6+(!+ !.*& ' , %"'* *'&!,!'& (, /1+. '. -!)-!,1$,. +-+,*,+. '*. -!)-!,!&7@D. (*',+'%6%!,(*','$1+!+5 5?=>=8;@<7+8DFC9E>C8 C>8. )"5 85 . '85 $-+,*!& (*',,+ (,'$$-$* *!&'% $$+ *'% &'($+%!*,!-$-%+,*++!&-('(,'+!+, *'- EF5 *) 5 ?=>@8 ;?<7+8 BBFE>8. C?8. *)"+$+/.0)$#5858/ '$)0*)"5)8*,+/-/).*)5

(71) !,' '&*!$ *+(!*,'*1 !& 1+-&,!'& !& .*!'-+ &-*'%-+-$* !+++5 '$) /,*-$5?==B8;A<7+8A?C9E8. C@8. *,,/585 .'85,!.!,!+'%!,' '&*!$'%($0+'**$,/!, & %'-&, !& %-+$ *'% (,!&,+5 /,*+.#*' ++' /,*$*'5 ?==D8 ;?<7+8?=A9>>8. CA8. #$-$5 85 . '85

(72) !,' '&*!$ '%($0 !!&1 & 7 *,!' !& $1%( '1,+ ' %1',*'( ! $,*$ +$*'+!+ (,!&,+5 /,*' -5 ?=>?8 ;@<7+8?FD9@=@8. CB8. +/ ). 9,/)5 85 . '85 -(*'%($0 ++%$1 ,*%!&+ $,*'& $-0 !&, %!,' '&*!$$,*'&,*&+('*, !&5$ ) 5?=>@8;C>A=<7+8 >BCD9D=8. CC8. 0/"$()5885 .'85&+-+' -%&+'$-$(*',!&'%($0+5 ''5 ?=>?8;B<7+8>=CE9E>8. CD8. #+," '5

(73) 88 ) 88 . ,5 %!& (*',!&+ * *)-!* '* !!&, ++%$1 ' %6$++ *!'&-$'(*',!&+5 ,* .' $ 5 ?==B8. . ?D.

(74) ;AE<7+8>D@D?9D8 CE8. . ,588)8)"51!&, $!', +($!'+'%5. 0*' ''$*'5?=>A8;?<7+8>=E9?>8. CF8. ) ''5 85 . '85 !'&-$'(*',!& ++%$1 ,+ '**$, /!, +(!&$ %-+-$* ,*'( 1 +.*!,1 & (**&,!$$1 , +-+, ' +($!'+'%$+& +5 *) 5?==D8;F<7+8 F?>8. . ?E.

(75) . . SCN4A. SNTA1. SOD1. CRYAB. MYBPC2. MYBPC1. ACTN2. CSRP3. TTN. DMD. TLN1. PABPN1. RYR1 ACTA1. MYOD1. TNNI1. CALM3. ATP1A1. TNNI3 TNNT2. MKKS. KCNQ1. FXYD1. CAMK2B. TPM3. TNNC1. DMPK. TNNT1. HSP90AA1. . SLN. PRKACA. TNNI2. MGEA5. PLN. ADRBK1. . CAV1. KCNH2 ATP2A2. TNNT3. ATP2A1. CAV3. AKAP13. SCN4A. SNTA1. SCN4A. SNTA1.

(76) .

(77) . SOD1. CRYAB. MYBPC2. MYBPC1. SOD1. CRYAB. MYBPC2. MYBPC1. ACTN2. ACTN2. CSRP3. CSRP3. TTN. DMD. TTN. DMD. TLN1. TLN1. PABPN1. PABPN1. RYR1 ACTA1. RYR1 ACTA1. MYOD1. MYOD1. TNNI1. TNNI1. CALM3. CALM3. ATP1A1. TNNI3 TNNT2. ATP1A1. TNNI3 TNNT2. MKKS. MKKS. KCNQ1. FXYD1. KCNQ1. FXYD1. CAMK2B. CAMK2B. TPM3. TPM3. TNNC1. TNNC1. DMPK. TNNT1. DMPK. HSP90AA1. TNNT1. HSP90AA1. SLN. PRKACA. SLN. TNNI2. PRKACA. MGEA5. TNNI2. MGEA5 PLN. ADRBK1. PLN. ADRBK1. CAV1. KCNH2 ATP2A2. CAV1. KCNH2 ATP2A2 TNNT3. TNNT3. ATP2A1. CAV3. ATP2A1. CAV3 AKAP13. AKAP13. SCN4A. SNTA1. . SOD1. CRYAB. MYBPC2. MYBPC1. ACTN2. CSRP3. TTN. DMD. TLN1. PABPN1. RYR1 ACTA1. MYOD1. TNNI1. CALM3. ATP1A1. TNNI3 TNNT2. MKKS. KCNQ1. FXYD1. CAMK2B. TPM3. TNNC1. DMPK. TNNT1. HSP90AA1. SLN. PRKACA. TNNI2. MGEA5. PLN. ADRBK1. CAV1. ! . KCNH2 ATP2A2. TNNT3. ATP2A1. CAV3. AKAP13. SCN4A. SNTA1. SOD1. CRYAB. MYBPC2. ! . MYBPC1. ACTN2. CSRP3. TTN. DMD. TLN1. PABPN1. RYR1 ACTA1. MYOD1. TNNI1. CALM3. ATP1A1. TNNI3 TNNT2. MKKS. KCNQ1. SCN4A. SNTA1. FXYD1. CAMK2B. TPM3. SOD1. TNNC1. CRYAB. DMPK. TNNT1. HSP90AA1. MYBPC2. SLN. PRKACA. TNNI2. MGEA5. MYBPC1. PLN. ADRBK1. ACTN2. SCN4A CAV1. CSRP3. KCNH2 ATP2A2. SNTA1. TNNT3. TTN. DMD. ATP2A1. TLN1. CAV3. AKAP13. PABPN1. ". RYR1 ACTA1. MYOD1. TNNI1. CALM3. SOD1. CRYAB. MYBPC2. ATP1A1. TNNI3 TNNT2. MKKS. KCNQ1. FXYD1. CAMK2B. TPM3. TNNC1. MYBPC1. DMPK. TNNT1. HSP90AA1. SLN. PRKACA. TNNI2. MGEA5. ACTN2. PLN. ADRBK1. CSRP3. SCN4A. SNTA1 CAV1. KCNH2 ATP2A2. TNNT3. TTN. DMD ATP2A1. ". CAV3. AKAP13. TLN1. PABPN1. RYR1 ACTA1. SOD1. CRYAB. MYOD1. TNNI1. CALM3. ACTN2. ATP1A1. TNNI3 TNNT2 TTN. DMD. #. MKKS. KCNQ1. FXYD1. CAMK2B. RYR1. TPM3. TNNC1. CALM3. DMPK. TNNT1. HSP90AA1. SLN. KCNQ1. PRKACA. TNNI2. CAMK2B. MGEA5. TPM3. PLN. ADRBK1. TNNT1. HSP90AA1. CAV1. KCNH2 ATP2A2. TNNT3. PRKACA. MGEA5. ATP2A1. CAV3. AKAP13. CAV1. AKAP13. #. Figure 1. Experimental design. To construct disease specified differentially co-expressed PIN (PIN), we integrate PIN and microarray by manipulating the SCC and SCC differences in SMA and ALS individually. To further mine functional modules, GO were used to classify the functions of theses dysregulated PPIs. . ?F.

(78) Figure 2. Comparison between SMN1 and SMN2. The alternative splicing mechanism of transcripts made from SMN1 and SMN2 is major due to a single nucleotide difference in exon7. SMN1 can produce near 100% of mature full-length SMN mRNA. However, SMN2 produces high percentage of truncated SMN transcript lacking exon7 (SMNΔ7). Only small amounts of full-length SMN mRNA are produced by SMN2 in SMA patients.. Figure 3. Illustration of DCPPIs groups classification. DCPPIs with Z-score larger (smaller) than or equal to 1.96 (-1.96) were defined as positive (negative) correlation in all study groups (SMA, ALS and normal). DCPPIs with Z-score within -1.96 and 1.96 were non-correlated. DCPPIs were classified into: gain of negative correlation (GoN), loss of positive correlation (LoP), loss of negative correlation (LoN), gain of positive correlation (GoP) and others (other_P, other_N, PP and NN). *Arrow direction: from disease (arrow head) to normal. . @=.

(79) 4. . Expression (log2 ratios). 3. . . 2 1. . . . . . . . . . . . . . . . . . . H. I. 0 . −1 −2. . . . A. . B. C. . . . . . . D. E. F. G. Patient Figure 4. Expression values of SMA. After normalization, common gene expression values are averaged. Then relative expression values were log2 transformed. (ratio: SMA intensity/. Expression (log2 transformed). control intensity). 13. . . . . . . . . . A. B. C. D. E. F. G. H. I. 11 9 7 5. Patient. Figure 5. Expression intensity of ALS . After normalization, common gene expression values are averaged. The expression values are log2 transformed.. . @>.

(80) Expression (log2 transformed). 13. . . . . . . . . . . . . . . . . . . A. B. C. D. E. F. G. H. I. J. K. L. M. N. O. P. Q. R. 11 9 7 5. Normal. Figure 6. Expression intensity of normal. After normalization, common gene expression values are averaged. The expression values are log2 transformed.. Figure 7. Fisher’s z transformed SCC histogram of SMA, ALS and normal. The histogram showed that most of the PPIs are centered at SCC equal to 0, which means most PPIs are not correlated. Besides, There are more co-expressed PPIs (either positive or negative correlated) in normal comparing with SMA and ALS.. . @?.

(81) -?8? . +?8? . Figure 8. Distributions of the Fisher’s z transformed SCC differences in SMA and ALS. Those PPIs with P< 0.05 were collected as DCPPIs and formed as SMA ad ALS DCPINs (The SCC difference (with z-transformed) cut-off : ~ ±2.2).. Figure 9. Overviews of genes and DCPPIs derived from DCPINs of SMA and ALS. a) We found 1329 common genes between DCPINs of SMA and ALS; b) 1313 common DCPPIs in DCPINs of SMA and ALS were found. 3770 (containing 1588 genes) and 3761 (consisted with 1572 genes) DCPPIs were found in SMA and ALS, respectively.. . @@.

(82) Figure 10. Network properties of integrated PPIs, and DCPINs of SMA, ALS and common. a) Degree, the degree distributions followed power-law property (only small amounts of nodes are found with higher degrees, vice versa); b) Clustering coefficient, genes with higher degree were found with lower clustering coefficient which showed all DCPINs were not highly clustered networks; c & d) betweenness and closseness centralities, nodes with higher degrees seem to be connectors and centers in all DCPINs.. . @A.