This page has been archived and is no longer updated
Genetics of gene expression and its effect on disease
Author: V. Emilsson
Keywords
Keywords for this Article
Add keywords to your Content
Save
|
Cancel
Share
|
Cancel
Revoke
|
Cancel
Rate & Certify
Rate Me...
Rate Me
!
Comment
Save
|
Cancel
Flag Inappropriate
The Content is
Objectionable
Explicit
Offensive
Inaccurate
Comment
Flag Content
|
Cancel
Delete Content
Reason
Delete
|
Cancel
Close
Full Screen
"ARTICLES Genetics of gene expression and its effect on disease Valur Emilsson 1,2 ,Gudmar Thorleifsson 1 ,Bin Zhang 2 , AmyS. Leonardson 2 , Florian Zink 1 , Jun Zhu 2 , Sonia Carlson 2 , Agnar Helgason 1 , G. Bragi Walters 1 , Steinunn Gunnarsdottir 1 , Magali Mouy 1 , Valgerdur Steinthorsdottir 1 , Gudrun H. Eiriksdottir 1 , Gyda Bjornsdottir 1 , Inga Reynisdottir 1 , Daniel Gudbjartsson 1 , Anna Helgadottir 1 , Aslaug Jonasdottir 1 , Adalbjorg Jonasdottir 1 , Unnur Styrkarsdottir 1 , SolveigGretarsdottir 1 , Kristinn P. Magnusson 1 , Hreinn Stefansson 1 , Ragnheidur Fossdal 1 , Kristleifur Kristjansson 1 , Hjortur G. Gislason 3 , Tryggvi Stefansson 3 , BjornG.Leifsson 3 ,UnnurThorsteinsdottir 1 ,JohnR.Lamb 2 ,JeffreyR.Gulcher 1 ,MarcL.Reitman 4 ,AugustineKong 1 , Eric E. Schadt 2 * & Kari Stefansson 1 * Common human diseases result from the interplay of many genes and environmental factors. Therefore, a more integrative biology approach is needed to unravel the complexity and causes of such diseases. To elucidate the complexity of common humandiseases suchasobesity,wehaveanalysedtheexpressionof23,720transcriptsinlargepopulation-basedbloodand adipose tissue cohorts comprehensively assessed for various phenotypes, including traits related to clinical obesity. In contrast to the blood expression profiles, we observed a marked correlation between gene expression in adipose tissue and obesity-related traits. Genome-wide linkage and association mapping revealed a highly significant genetic component to gene expression traits, including a strong genetic effect of proximal (cis) signals, with 50% of the cis signals overlapping between the two tissues profiled. Here we demonstrate an extensive transcriptional network constructed from the human adipose data that exhibits significant overlap with similar network modules constructed from mouse adipose data. A core network module in humans and mice was identified that is enriched for genes involved in the inflammatory and immune response and has been found to be causally associated to obesity-related traits. The comprehensive assessment of molecular quantities in biological samples using high-throughput technologies has already led to the identification of disease subtypes 1,2 , novel genes and gene struc- tures 3,4 , and biomarkers for disease 5 , as well as the elucidation of transcriptional networks associated with disease traits 6?8 . The ana- lysis of genotypes and gene expression data in animal models and human cell lines has proven useful for identifying genetic determi- nants of expression traits 1,9?13 and for mapping genes in regions linked to complex traits 6,10,11,14 . In general, such studies provide the means to examine the overall genetic complexity of gene expression traits, including a characterization of the relative effect of cis versus trans control 15,16 . Associating patterns of gene expression with DNA and complex traitvariationisnecessarilylimitedtothosechangesthatarereflected in the transcriptional network. Although a number of studies have highlighted the importance of post-transcriptional alterations in gene activity that induce changes in biological processes 17 , variation in protein structure and state may be reflected in the transcriptional network because such variation often induces a change in transcript stability, rates of transcription, transport of RNA from the nucleus, alternative splicing events, and other processes that affect expression levels 1 . Importantly, given the context specificity of many critical biological processes 18 and the fact that most common diseases are thought to be the outcome of a complex interaction between many genetic loci and the environment, it follows that there are obvious advantages to studying the genetics of gene expression in cells that represent the in vivo state. Towards this end, we collected blood and subcutaneous adipose tissuesinapopulation-based samplingofhundredsofIcelandic sub- jectsranginginagefrom18to85yearsold.Thesecohortsarereferred to as the Icelandic Family Blood (IFB) cohort (N51,002) and the Icelandic Family Adipose (IFA) cohort (N5673) (see Supplemen- tary Table 1 for cohort description). A number of clinical traits including differentialbloodcellcountaswellasbiometric traitssuch as body mass index (BMI), percentage body fat (PBF, measured by bioimpedance) and waist-to-hip ratio (WHR) were collected for all subjectsoftheIFBandtheIFAcohorts(SupplementaryTable1).The relatively large sample size used in this study design provided the means to assess the relationship between sequence variants and gene expression with more statistical power than previous studies 12,13,16 . Gene?clinical trait correlations Expressionprofilesproducedforthisstudycontainedmeasurements of relative abundances of 23,720 transcripts, representing 84% of the 24,060 protein-coding genes annotated in the Ensembl database (v.33) 19 . Given that probes overlapping single nucleotide poly- morphisms (SNPs) may give rise to artificial signals, we sequenced anumberofprobesimplicatedasstrongexpressionquantitativetrait loci (eQTL)in 470 subjects from the IFB (see Supplementary Results and Supplementary Table 2). In short, we found that probes over- lapping SNPs is not a concern in the present study. The distribution of biometric traits such as BMI in our cohorts is not unlike the distribution that one would encounter in the general Westernpopulation,withBMIrangingfrom16to70andamedianof *These authors contributed equally to this work. 1 deCODE genetics, 101 Reykjavik, Iceland. 2 Rosetta Inpharmatics, LLC, 401 Terry Ave N, Seattle, Washington 98109, USA. 3 Department of Surgery, National University Hospital, 101 Reykjavik, Iceland. 4 Merck Research Laboratories, Rahway, New Jersey 07065, USA. Vol 452|27 March 2008|doi:10.1038/nature06758 423 Nature Publishing Group�2008 28.8 (Supplementary Fig. 2a). Given the known associations of bio- metric traits with age and sex, and the fact that gene expression traits in blood have been found to be correlated with these covariates as wellaswithwhitebloodcellcounts 20 ,weadjustedforthesecovariates using multiple linear regression (Methods) in all analyses of correla- tion between gene expression and clinical traits, as well as in the analyses of the genetic component of gene expression (see below). In blood, fixing the false discovery rate (FDR) 21 at 5%, we found 2,172 (9.2%) gene expression traits to be correlated with BMI, 1,098 (4.6%) with PBF, and 711 (3.0%) with WHR in the IFB cohort (Supplementary Table3).Inadiposetissue,ata5%FDR,theexpres- sion levelsof 17,080 (72.0%)genes were correlated with BMI, 16,977 (71.6%) with PBF, and 14,901 (62.8%) with WHR (Supplementary Table 3). Thus, there is at least an order of magnitude more expres- sion traits that are significantly correlated with these biometric traits in adipose tissue than in blood. Furthermore, 2,784 of the gene expression traits in adipose tissue explained more than 10% of the BMI variation in the IFA (R 2 $0.1, P#10 215 ; see Supplementary Fig. 2b), whereas none of the expression traits in blood achieved this levelofcorrelation.Toensureequivalentstatisticalpowerformaking these detections between the tissues, we compared these associations in the 553 subjects represented in both the IFB and IFA cohorts. Using these paired samples, we found an even more marked differ- ence between the two tissues (Supplementary Table 3). For example, there was a notable 34.6-fold enrichment of expression traits corre- lated with BMI in adipose tissue compared with blood using the 553 subjects (FDR#0.01), whereas this enrichment was 13.9-fold in the full data sets. Overall, our results suggest that a substantial fraction of the tran- scriptional network in adipose tissue, together with infiltrated macrophages 22?24 , is associated with the obesity of subjects. There are several reasons why this strong relationship between gene expression levels in adipose tissue and obesity should not come as asurprise.First,obesityisadisorderofexcessivebodyfat.Second,the physiology and morphology of the adipocyte is known to be drasti- callyalteredinobesesubjects 25 .Third,thenumberofmacrophagesis markedly increased in the adipose tissue of obese subjects, and they have been shown to have an important role in obesity and related metabolic disorders 22?24 . Heritability of gene expression traits The subjects in the IFB and IFA cohorts were clustered into multi- generationalfamilies(fordetails,seeMethods).InthecaseoftheIFB cohort,itwaspossibletocluster938outofthe1,002subjectsinto209 families, whereas for the IFA cohort, 570 out of the 673 subjects clustered into 124 families. Using this family structure, we estimated theheritabilityofeachofthe23,720geneexpressiontraits,bothwith and without adjusting for sex, age, cell count (IFB only) and BMI (IFA only). The number of traits with statistically significant herit- abilityissummarized inTable1.Withnoadjustment, thenumberof significantly heritable traits at a 5% FDR was 13,910 in IFB and 16,825 in IFA, or 58.6% and 70.9% of all assessed transcripts, respectively. For those significantly heritable expression traits in the IFA and IFB cohorts, the genetic variance component on average explained nearly 30% of the variation observed (Supplementary Fig. 2c). After adjustment, the number of heritable traits fell by as much as 26% (Table 1). When combined with the high heritability estimated for the expression traits, these results indicate that a significant proportion of the heritability mediated by BMI or differ- entialcellcountisalsoreflectedbyalargenumberofgeneexpression traits. The heritability values (percentage) of all expression traits for the different types of adjustments and in both cohorts are listed in Supplementary Tables 4 and 5. Detection of cis and trans eQTL All subjects in the two tissue cohorts were genotyped using a frame- work set of 1,732 microsatellites and were used for genome-wide linkage analysis. Because one of the main aims of this analysis was to detect eQTL signals that are proximal to the physical locations of genes corresponding to the expression traits (referred to here as cis- acting eQTL signals), this analysis was restricted to the 20,877 expression traits that had well-defined map positions (NCBI Build 34).Forcomparison,theeQTLanalysiswasperformedbothwithand without adjusting the trait values for sex, age, differential cell-count (IFB only) and BMI (IFA only). We defined a cis-acting eQTL signal for a given expression trait as the logarithm of the odds (eLOD) score at the nearest microsatellite tothelocationofthecorrespondingprobe.Thenumberoftraitswith significant cis eQTL is summarized in Table 1. For instance, at a 5% FDR and without any adjustment, we observed significant cis eQTL for1,970(9.4%)traitsinbloodand1,215(5.8%)traitsintheadipose tissues. After adjusting for sex, age and blood cell counts in IFB, the number of cis eQTL signals increased to 2,529. In adipose tissue, this number was 1,307 after adjusting for age and sex and was 1,489 after also adjusting for BMI (Table 1). Out of the 1,489 significant cis- acting eQTL in adipose tissue, 762 (51.2%) were also observed in blood. Furthermore, expression traits with high heritability in both blood and adipose tissue showed greater reproducibility between the tissues (Fig. 1a). Here, 70% of all expression traits within the upper 25th percentile for heritability in blood that had a significant cis- actingeQTLinadiposetissue,alsohadasignificantciseQTLinblood (Fig. 1a). In fact, the proportion of significant cis eQTL signals in both tissues was notably higher for traits with greater levels of differ- ential expression or heritability (Fig. 1b). The cis-acting eQTL LOD scores for each of the expression traits in the different cohorts are listed in Supplementary Tables 4 and 5. Our finding of a strong genetic effect associated with cis signals in thesetissuesisconsistentwithresultsfrompreviousstudies 1,11?13 .The results on the detection of eQTL signals that were distal to the phy- sical locations of the genes corresponding to the expression traits (referred to here as trans-acting linkage signals) are shown in the Supplementary Results and in Supplementary Table 6. We note that Table 1 | Heritability, cis eQTL and cis eSNP detection IFB* IFA{ Variable FDR or g{ No adjustment Age, sex and cell-count adjusted No adjustment Age and sex adjusted Age, sex and BMI adjusted Heritability 0.05 13,910 10,364 16,825 16,714 15,727 0.01 10,829 8,047 12,309 12,392 11,251 g 0.68 0.55 0.78 0.77 0.75 cis eQTL 0.05 1,970 2,529 1,215 1,307 1,489 0.01 1,256 1,567 737 773 820 g 0.40 0.44 0.33 0.32 0.37 cis eSNPs 0.05 2,417 2,714 3,048 3,149 3,364 0.01 1,827 2,026 2,271 2,323 2,506 g 0.33 0.32 0.37 0.35 0.36 The number of cis eQTL and cis eSNPs were as determined for a unique set of gene expression traits, for example the single most significant cis eSNP for any given trait. *Multiple regression analysis in blood, adjusting for sex and age as (age3sex) or for age, sex and differential cell-count as (age1neutrophil1monocyte1lymphocyte)3sex. {Multiple regression analysis in adipose, adjusting for sex and age as (age3sex) or for age, sex and BMI as (age1log(BMI))3sex. {The proportion of significant tests, g, was estimated as g512p 0 (see Methods for details). ARTICLES NATURE|Vol 452|27 March 2008 424 Nature Publishing Group�2008 thenumberoftraitswithsignificanttranseQTLinbloodandadipose tissue are 50 times fewer than the number of expression traits with significant cis eQTL, consistent with what has been found in other studies 1,9,12,14 . Finally, although others have reported hotspots of localized linkage activity in a number of species 1,6,9?11,13,14 , we failed to detect such activity beyond what was expected by chance (Supple- mentary Results). Identification of cis and trans eSNPs For the identification of sequence variants that have cis and trans regulatory effects on expression traits, we selected a subset of 150 unrelated (excluding all first-degree relatives) subjects who donated bothbloodandadiposetissue,andperformedawhole-genomegeno- typing of these samples employing 317,503 SNPs using the Illumina platform 26 . The strongest cis effect for a given expression trait was then mapped by testing all SNPs located within a 2megabase (Mb) window centred at the location of the probe corresponding to the expressiontrait,againrestrictingtheanalysistothe20,877geneswith well defined positions in the genome. For each expression trait, because multiple correlated SNPs were tested for cis association, simulation was used to adjust the P value of the most significant expression (e)SNPs (see Methods). The effect of testing multiple expression traits was, as before, taken into account by means of the FDR approach 21 . The number of significant cis-acting eSNPs is sum- marized in Table 1. Assuming an FDR of 5%, we detected cis eSNPs for 2,417 (11.5%) expression traits in blood and 3,048 (14.6%) traits in adipose without any adjustment (Table 1). After adjusting for sex, age and cell count in blood, the number of cis eSNPs increased to 2,714 (Table 1). After adjusting for sex, age and BMI in the adipose tissue,thenumberofciseSNPsincreasedto3,364(Table1).Thus,we detected 650 more gene expression traits with significant cis eSNPs in the adipose tissue than in blood. This difference may reflect a more homogenous cell population in adipose tissue compared to blood, granting greater power to detect the cis effect in adipose. Further- more, the number of significant cis eSNPs observed in both blood and adipose tissue increased as the heritability increased (Fig. 2a). For example, at an FDR of 1%, at least 50% of all SNPs that were cis- actinginbloodandwithintheupper25thpercentileforheritabilityor differential expression were also cis-acting in adipose tissue (Fig. 2a). Figure 2b summarizes the number of significant cis associations plotted as a function of heritability and differential expression. As observed in our analysis of cis-acting eQTL signals, the number of significant cis eSNPs increases with greater heritability scores or greater differential expression (Fig. 2b). A direct comparison of the results obtained from the genome-wide linkage and association ana- lysesof cis-actingsignals revealedamarked agreement betweenthese two approaches (Supplementary Results and Supplementary Fig. 3). The significance of the trans association effect was assessed using the FDR approach 21 , and the number of significant trans eSNPs summarized in Supplementary Table 6, again showing significantly fewer effects in trans than in cis, as described and discussed in Supplementary Results. 1,800 1,500 1,200 900 600 300 0 Number of cis eSNPs 70 60 50 40 30 20 10 0 Overlap (%) Q1 Q2 Q3 Q4 Adipose tissue Blood Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 a b Differential expression Heritability Differential expression Heritability Figure 2 | Genome-wide association screens for eSNPs. A subset of 150 unrelated subjects who donated both blood and adipose tissue were genotyped at 317,503 tag SNPs (ILMN). The cis eSNP effects were assessed using linear regression on 20,877 standardized gene expression traits (see Methods for details). As described in Fig. 1, all traits were binned into quartiles at varying strengths of differential expression or heritability. a, Shownisthefractionof traits at varyingdegreesof differentialexpression orheritabilitywithsignificant cis eSNPsinadiposetissuethatreproducedin blood at 1% FDR. b, Shown is the number of significant cis associations in both tissues plotted as a function of heritability and differential expression. Q1 Q2 Q3 Q4 75 60 45 30 15 0 Overlap (%) 1,400 1,200 1,000 800 600 400 200 0 Number of cis eQTL Adipose tissue Blood Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 a b Differential expression Heritability Differential expression Heritability Figure 1 | eQTL mapping in human blood and adipose tissue. Individuals from large multi-generational families were genotyped for 1,732 microsatellites, and linkage analysis was performed on 20,877 standardized gene expression traits (see Methods for detail). Expression traits, ranked according to their differential expression or heritability strength, were binnedintoquartiles(Q1RQ4),eachcomprisedof5,939genes.a,Shownis the fraction of traits that have varying levels of differential expression and heritability with significant cis-acting eQTL in adipose that reproduced in blood at 1% FDR. b, Shown is the number of significant cis eQTL in both tissues, as a function of differential expression and heritability at 1% FDR. NATURE|Vol 452|27 March 2008 ARTICLES 425 Nature Publishing Group�2008 Characterizing the transcriptional network The analysis of gene expression traits in a large sample of individuals allowsforadirectandunbiasedassessmentoftheconnectivitystruc- ture of transcriptional networks 27 . This further provides a basis for the identification of key functional modules within such networks that contribute to disease risk 28 . We have previously described the characterization of transcriptional networks based on brain, adipose and liver tissues in a cross between two inbred strains of mice (referred to here as the B3H cross) 29?31 . Building on this approach, we constructed extensive, sex-specific, gene co-expression networks based on the human adipose tissue data to identify modules strongly associatedwithobesityand,moregenerally,comparingthestructure of this human network to that constructed in the mouse B3H cross using similar tissues. The adipose co-expression network was con- structed by considering all pair-wise correlations among the most differentially expressed genes detected in this tissue (Methods). The resultinggene?genecorrelationmatrixwasthentransformedintoan adjacency matrix in which the connectivity of a given gene was defined as the sum of its connection strengths with all other genes inthenetwork 27 .Thegene?geneinterconnectivityrepresentedinthis matrix(referredtohereastheconnectivitymap)wasthencharacter- ized using a topological overlap measure 28 . The identification of functional modules of highly co-regulated genes in the resulting network was carried out using a dynamic programming procedure to search the network for sets of maximally interconnected genes 29 . Figure3adepictstheconnectivitymapforthemalehumanadipose tissue as a heat map of the topological overlap matrix. In this type of map, the rows and the columns represent genes in a symmetric fash- ion, and the colour intensity represents the interaction strength between genes. This connectivity map highlights that genes in the adipose transcriptional network fall into distinct network modules, where genes within a given module are more highly interconnected with each other (blocks along the diagonal of the matrix) than with genesinothermodules,ashasbeendescribedpreviouslyformice 30 .A comparison of the connectivity structure between the male and female human adipose networks reveals a number of network modules that are well conserved between gender, both in terms of geneidentitiesandtheconnectivitystrength(hubstatusorcentrality; see Supplementary Figs 6 and 7). However, there are also network modules that are strictly gender specific (Supplementary Fig. 6). Anexplicitcomparisonofthehumanandmouseadiposegeneco- expression networks revealed a single core module in humans that was highly conserved inmice (Fig. 3a?c). The mouse module corres- pondingtothishumanmodule(Fig.3b)isverysignificantlyenriched for genes with eQTL that co-localize with obesity-associated-trait QTLs as well as for genes shown to be in a causal relationship with obesity-associated traits 31 . This mouse module significantly over- lapped the human network module (Fig. 3a), with 196 out of the 673 (,29%) genes in the mouse module overlapping the set of 886 genes in the corresponding human module (only 8 were expected to overlap by chance; Fisher?s Exact Test, P58.4310 2118 ). In addi- tion,theGeneOntology(GO)BiologicalProcesscategoriesthatwere enrichedinthisconservednetworkmodulewerevirtuallyidenticalin mouseandhuman(SupplementaryTable7).Thisconservedmodule was also strongly indicative of macrophage function for a number of reasons. First, GO Biological Process categories enriched in this module relate to inflammatory response and macrophage activation pathways. Second, well known macrophage-specific cell-surface markers such as EMR1 and CD68 are represented in the mouse and human modules. Third, using a recently constructed mouse body gene expression atlas comprised of more than 60 tissues and cell lines 31 , this conserved module had an over-representation of genes enriched for expression in bone-marrow-derived macrophages (Fisher?s Exact Test, P,1310 221 ), spleen, thymus and lymphoid tissue (Fisher?s Exact Test, P,1310 220 ). These findings are con- sistent with results from recent studies showing that the adipose tissue secretes factors that regulate a wide variety of physiological states, including energy homeostasis and the immune response 25 . Given all of these significant enrichments and the association of this moduletomacrophagefunctionandmetabolictraits,werefertoitas the macrophage-enriched metabolic network (MEMN). Because the mouse MEMN described above had previously been showntobesignificantlyenrichedforgenesassociatedwithobesity 31 , we investigated whether a similar association to obesity could be detected for the corresponding human module. Our results show that the expression of 868 (or 98%) of the 886 genes in the human MEMN module were significantly correlated with BMI in adipose tissue at an FDR of 1%, indicating that the human MEMN module may have a key role in obesity. Although the connection between inflammation and metabolic disorders such as obesity and diabetes has been reported previously 25 , these data suggest that there may be many immune pathways or entire networks functioning in the adipose tissue. In fact, a number of genes previously identified and validated as being in a causal relationship with obesity-associated phenotypes are represented in this module, and perturbing many of these genes perturbs the entire module (see Supplementary Results for additional information). If the MEMN module has a role in human obesity, then variations in DNA that result in expression changes in genes in the MEMN module should, in the obese, be enriched for variations that are associated with obesity. Therefore, we combined genotype and gene expression data to identify the SNP in the vicinity of each genein the human MEMN module (Fig. 3a) that was most strongly associated with the corresponding gene expression trait. We then tested these variantsjointlyforassociation toBMIandPBF?thebiometrictraits most widely used to assess human obesity. Of the 886 expression traits represented in this module, 785 had a well defined genomic position and were used in this analysis. A selection of 768 cis eSNPs forthebloodandadiposetissuedataweresuccessfullygenotypedina cohort of 8,685 individuals measured for BMI and 1,939 for PBF (Table 2). We used multiple linear regression analysis to test the association of the sex- and age-adjusted trait values to genotype counts for all the cis eSNPs jointly (see Methods for details). a Overlap (P = 8.4 x 10 ?118 ) Human males Mouse males b c Enriched GO biological categories: Inflammatory response (P < 10 ?36 ) Immune cell activation (P < 10 ?27 ) Cell activation (P < 10 ?26 ) Macrophage-mediated immunity (P < 10 ?4 ) Macrophage activation (P < 10 ?3 ) Figure 3 | The human and mouse gene transcriptional networks. a, Clustering of the connectivity matrix for the top 25% most differentially expressed genes in the male human adipose data. In the heat map, rows and columns represent genes in a symmetric fashion. The colour intensity signifies the connection strength between two genes, with red colour representing the strongest connection and white representing no connection. The colour bars along the x- and y-axes delineate the highly interconnected gene modules. b, Same as a, but for the male mouse B3H adipose data. c, The turquoise module in the male human network (a)is significantly overlapping the male mouse brown module (b), as well as the turquoisemodule inhumanfemales.TheGOlist in c showstheenrichment of inflammatory pathways in the conserved module. ARTICLES NATURE|Vol 452|27 March 2008 426 Nature Publishing Group�2008 Furthermore, we constructed 20,000 sets of simulated genotypes for all the variants conditioned on the familial relatedness of the indivi- dualsfromtheIcelandicgenealogydatabasetocomparetheobserved association with that expected to occur by chance, and used these to generate the adjusted P values represented in Table 2. In the larger data set with BMI measurements, we find that the cis eSNPs selected for genes in the human MEMN module showed some evidence for association to BMI, with P values of 3.8310 26 (adjusted P50.005) and 5.3310 27 (adjusted P50.002) for the cis eSNPs in adipose tissue and blood, respectively (see Table 2). Although these analyses were crude for individual cis eSNPs and the corresponding genes, these results suggest that the human MEMN module is enriched for sequence variants that confer risk of obesity in humans, and that genetic perturbations affecting gene expression traits may more generally perturb networks that in turn lead to increased susceptibi- lity to disease. These data combined offer a glimpse of the compli- cated network of interactions that could drive at least a portion of obesityinhumans,and demonstratethat atleastapartofobesityis a property of the macrophage gene network. Discussion Previous studies of the genetics of gene expression in humans have been restricted to lymphoblastoid cell lines with no clinical pheno- types 12,13,16 . Before our study, the validation of this type of data in primary human tissues from subjects scored for clinical traits was lacking. Our analysis of genetic variation and gene expression in population-based sampling of blood and subcutaneous adipose tis- sue from a large number of extended families begins to fill this gap. We showed that more than 50% of all gene expression traits in adi- pose tissue are strongly correlated with clinical traits related to obesity, compared to less than 10% in blood. Furthermore, through segregation analysis and genome-wide linkage and association stud- ies, we demonstrated an extensive genetic component underlying geneexpressiontraitsinbloodandadiposetissue.Thiswasevidenced by detection of heritability as a highly significant contributor to variation in gene expression and by the identification of a large number of significant linkage and association signals for the expres- sion traits in the two tissues, with approximately 50% overlap of genetic signals between the two tissues. Consistent with previous reports, the signals detected using both linkage and association ana- lysis was strongly biased towards cis- rather than trans-acting genetic signals. We also constructed an extensive co-expression network on the basisofthehumanadiposetissuedatawiththeaimofidentifyingkey functional modules within this network that associated with obesity. Acoregeneexpressionmodule,theMEMNmodule,wasidentifiedin humans that has significant overlap with a previously described mouse networkmodule.Thegenesets inthecorehuman andmouse modules were highly enriched for genes involved in inflammatory response and macrophage activation pathways. Furthermore, the mouse MEMN module has previously been shown to be enriched for genes that contribute to the risk of obesity, diabetes and atherosclerosis-associated traits. By using the strongest cis-acting SNPs for each of the gene expression traits from the human MEMN module and testing them jointly as a group, we observed a significantenrichmentofgeneticassociationstoclinicaltraitsrelated to human obesity in this module. The identification of SNPs that are associated with variation in gene expression provides a level of func- tional support for such SNPs that makes them ideal candidates to identify genetic determinants of complex traits including diseases and drug response. Clearly this approach warrants serious conside- ration given the potential to affect our understanding of human health. METHODS SUMMARY Subjects used in the present study were of Caucasian descent. They were recruited as dense three-generation pedigrees, and comprehensively scored for multiple phenotypes including biometric traits related to obesity. Peripheral blood (N51,002) and subcutaneous fat (N5673) were collected, and DNA and RNA extracted. The RNA samples (a total of 1,765 samples), including referencepools,werehybridizedtoasinglecustom-madehumanarraycontain- ing 23,720 unique oligonucleotide probes. We estimated the differential expres- sion, heritability, cis and trans eQTL, and association signals for each gene expression trait in each tissue. For the genetics of gene expression analysis, all subjectsinthesecohortsweregenotypedat1,732microsatellites.Asubsetof150 unrelated subjects, donating both blood and adipose tissue, was genotyped at 317,000 SNPs. Multiple testing for significance was taken into account through the use of an FDR procedure. The expression and clinical data were adjusted for standard covariates including age and sex for all analyses. The gene?gene co- expression network was constructed from the human adipose tissue expression data and compared to a similarly constructed adipose tissue network from an experimental mouse cross. Finally, expression variation markers (eSNPs) map- ping to a core network module identified in human adipose tissue and found to be conserved in mice and previously shown to be enriched for genes in a causal relationship with obesity were tested jointly for association to obesity-related traits in humans using multiple regression analysis. Full Methods and any associated references are available in the online version of the paper at www.nature.com/nature. Received 23 July 2007; accepted 28 January 2008. Published online 16 March 2008. 1. Schadt,E.E.etal.Geneticsofgeneexpressionsurveyedinmaize,mouseandman. Nature 422, 297?302 (2003). 2. Golub, T. R. et al. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286, 531?537 (1999). 3. Johnson,J.M.etal.Genome-widesurveyofhumanalternativepre-mRNAsplicing with exon junction microarrays. Science 302, 2141?2144 (2003). 4. Shoemaker, D. D. et al. Experimental annotation of the human genome using microarray technology. Nature 409, 922?927 (2001). 5. Welsh, J. B. et al. Analysis of gene expression profiles in normal and neoplastic ovarian tissue samples identifies candidate molecular markers of epithelial ovarian cancer. Proc. Natl Acad. Sci. USA 98, 1176?1181 (2001). 6. Schadt, E. E. et al. An integrative genomics approach to infer causal associations between gene expression and disease. Nature Genet. 37, 710?717 (2005). 7. Schadt,E.E.,Sachs,A.&Friend,S.Embracingcomplexity,inchingclosertoreality. Sci. STKE 2005, pe40 (2005). Table 2 | Association of eSNPs to obesity traits Cohort Trait All P (N)AlP adjusted* Male P (N) Male P adjusted* Female P (N) Female P adjusted* Human MEMN IFA BMI 3.8310 26 (8,685) 0.0051 0.033 (3,606) 0.24 0.0022 (5,079) 0.049 PBF 0.0011 (1,939) 0.047 0.20 (906) 0.47 0.12 (1,035) 0.27 IFB BMI 5.3310 27 (8,685) 0.0022 0.00041 (3,606) 0.015 0.016 (5,079) 0.16 PBF 0.063 (1,939) 0.46 0.64 (906) 0.87 0.39 (1,035) 0.61 Combined MEMN IFA BMI 0.14 (8,685) 0.41 0.22 (3,606) 0.39 0.27 (5,079) 0.47 PBF 0.010 (1,939) 0.055 0.23 (904) 0.49 0.011 (1,035) 0.021 IFB BMI 0.0014 (8,685) 0.018 0.028 (3,606) 0.075 0.028 (5,079) 0.081 PBF 0.23 (1,939) 0.51 0.46 (904) 0.73 0.45 (1,035) 0.56 785 out of the 886expression traitsin the humanturquoise module (seeFig. 3) mapped to a uniqueposition and hada correspondingcis eSNP;this correspondsto 768 unique cis eSNPs that were usedintheanalysis.Missinggenotypesweresubstitutedwiththemeangenotypefrequency.128outofthe146expressiontraitsinthecombinedhumanandmouseMEMNmodulehadauniquemap position and a cis eSNP; this corresponds to 123 unique cis eSNPs that were used in the analysis. *The adjusted P value was based on up to 20,000 sets of simulated genotypes (see Methods). NATURE|Vol 452|27 March 2008 ARTICLES 427 Nature Publishing Group�2008 8. Zhu, J. et al. An integrative genomics approach to the reconstruction of gene networksinsegregatingpopulations.Cytogenet.GenomeRes.105,363?374(2004). 9. Brem, R. B., Yvert, G., Clinton, R. & Kruglyak, L. Genetic dissection of transcriptional regulation in budding yeast. Science 296, 752?755 (2002). 10. Bystrykh,L.etal.Uncoveringregulatorypathwaysthataffecthematopoieticstem cell function using ?genetical genomics?. Nature Genet. 37, 225?232 (2005). 11. Chesler, E. J. et al. Complex trait analysis of gene expression uncovers polygenic and pleiotropic networks that modulate nervous system function. Nature Genet. 37, 233?242 (2005). 12. Monks, S. A. et al. Genetic inheritance ofgene expression inhuman cell lines. Am. J. Hum. Genet. 75, 1094?1105 (2004). 13. Morley, M. et al. Genetic analysis of genome-wide variation in human gene expression. Nature 430, 743?747 (2004). 14. Mehrabian, M. et al. Integrating genotypic and expression data in a segregating mouse population to identify 5-lipoxygenase as a susceptibility gene for obesity and bone traits. Nature Genet. 37, 1224?1233 (2005). 15. Brem, R. B., Storey, J. D., Whittle, J. & Kruglyak, L. Genetic interactions between polymorphismsthataffectgeneexpressioninyeast.Nature436,701?703(2005). 16. Cheung, V. G. et al. Mapping determinants of human gene expression by regional and genome-wide association. Nature 437, 1365?1369 (2005). 17. Ranganathan, P. et al. Expression profiling of genes regulated by TGF-b: differentialregulationinnormalandtumourcells.BMCGenom.8,98,doi:10.1186/ 1471-2164-8-98 (2007). 18. Brem,R.B.&Kruglyak, L.Thelandscapeofgenetic complexity across5,700gene expression traits in yeast. Proc. Natl Acad. Sci. USA 102, 1572?1577 (2005). 19. Hubbard, T. et al. Ensembl 2005. Nucleic Acids Res. 33, D447?D453 (2005). 20. Whitney, A. R. et al. Individuality and variation in gene expression patterns in human blood. Proc. Natl Acad. Sci. USA 100, 1896?1901 (2003). 21. Storey, J. D. & Tibshirani, R. Statistical methods for identifying differentially expressed genes in DNA microarrays. Methods Mol. Biol. 224, 149?157 (2003). 22. Di Gregorio, G. B. et al. Expression of CD68 and macrophage chemoattractant protein-1 genes in human adipose and muscle tissues: association with cytokine expression, insulin resistance, and reduction by pioglitazone. Diabetes 54, 2305?2313 (2005). 23. Lumeng, C. N., Bodzin, J. L. & Saltiel, A. R. Obesity induces a phenotypic switch in adipose tissue macrophage polarization. J. Clin. Invest. 117, 175?184 (2007). 24. Neels, J. G. & Olefsky, J. M. Inflamed fat: what starts the fire? J. Clin. Invest. 116, 33?35 (2006). 25. Wellen, K. E. & Hotamisligil, G. S. Obesity-induced inflammatory changes in adipose tissue. J. Clin. Invest. 112, 1785?1788 (2003). 26. Steemers, F. J. & Gunderson, K. L. Illumina, Inc. Pharmacogenomics 6, 777?782 (2005). 27. Zhang, B. & Horvath, S. A general framework for weighted gene co-expression network analysis. Stat. Appl. Genet. Mol. Biol. 4, Article17 (2005). 28. Ravasz, E., Somera, A. L., Mongru, D. A., Oltvai, Z. N. & Barabasi, A. L. Hierarchical organization of modularity in metabolic networks. Science 297, 1551?1555 (2002). 29. Ghazalpour, A. et al. Integrating genetic and network analysis to characterize genes related to mouse weight. PLoS Genet 2, e130 (2006). 30. Lum, P. Y. et al. Elucidating the murine brain transcriptional network in a segregatingmousepopulationtoidentifycorefunctionalmodulesforobesityand diabetes. J. Neurochem. 97 (suppl. 1), 50?62 (2006). 31. Chen,Y.etal.VariationsinDNAelucidatemolecularnetworksthatcausedisease. Nature doi:10.1038/nature06757 (this issue). Supplementary Information is linked to the online version of the paper at www.nature.com/nature. Acknowledgements The authors acknowledge the participating families and the staff at the Clinical Research Centre for their cooperation. Genotyping service was provided at the deCode Genetics genotyping facilities. Author Contributions V.E., E.E.S., K.S. and G.T. wrote the paper. G.T., E.E.S., A.K., D.G. and F.Z. performed statistical analysis. Tissue sampling and/or molecular profiling was carried out by H.G.G., T.S., B.G.L., G.H.E., S.C., M.M., Aslaug Jonasdottir, Adalbjorg Jonasdottir, G.B. and K.K. V.E., J.Z., U.T., A.S.L., A.H., B.Z., G.B.W., S. Gunnarsdottir, S. Gretarsdottir, K.P.M., V.S., I.R., A.H., U.S., H.S., R.F., J.R.G., K.S., M.L.R. and J.R.L. performed the genetic analysis and/or data-mining. K.S. and E.E.S. contributed equally to this work. Author Information All the gene expression data generated for this study have been deposited into the GEO database under accession numbers GSE7965 and GPL3991. The authors declare competing financial interests: details accompany the full-text HTML version of the paper at www.nature.com/nature. Reprints and permissions information is available at www.nature.com/reprints. Correspondence and requests for materials should be addressed to K.S. (kari.stefansson@decode.is) or E.E.S. (eric_schadt@merck.com). ARTICLES NATURE|Vol 452|27 March 2008 428 Nature Publishing Group�2008 METHODS Humanstudypopulationsandsampleprocessing.Thesubjects,ranginginage from 18 to 85 years old, in the IFB and IFA cohorts were clustered into multi- generational families on the basis of relatedness of individuals in the Icelandic genealogydatabase 32 .FortheIFBcohort,1,002Icelandicsubjectswererecruited, and for the IFA cohort, 673 subjects were recruited. All participants in the IFA and IFB cohorts were scored for various clinical traits related to obesity, includ- ing height, weight,waist circumference, hip circumference and percentagebody fat(PBF)measuredbybioimpedance.InadditiontotheIFAandIFBcohorts,85 (43 males and 42 females) Icelandic individuals were recruited to generate a blood RNA reference pool for the IFB cohort. Furthermore, ten (six females and four males) additional Icelandic individuals being operated on for abdo- minal hernia were recruited to construct an adipose reference RNA pool for the IFA cohort. Ethical approval for the present study was granted by the National BioethicsCommittee(NBC01-033)andtheIcelandicDataProtectionAuthority (DPA). All participants in the study signed informed consent. All personal identifiers associated with tissue samples, clinical information and genealogy were encrypted by the DPA, using a third-party encryption system in which DPA maintains the code 32 . The RNA and DNA sample preparation, microarray hybridizationandexpressionanalysisaredescribedintheMIAMEchecklistthat is provided in the Supplementary Information. Identifying differentially expressed genes. To assess whether a gene in a given samplewasdifferentiallyexpressed,weusedapreviouslydescribedandvalidated errormodelfortestingwhetherthemeanlogratiooftheintensitymeasurements between the experiment and reference channels was significantly different from zero 33,34 . On the basis of this error model, we obtained P values for each of the individual gene expression measures in each sample as described previously 33 . We then computed the standard deviation of ?log 10 of the P value for each gene expression measure over all samples profiled for a given tissue, and then rank- ordered all of the genes profiled in each tissue on the basis of this standard deviation value (rank-ordered in descending order). Genes that fall at the top ofthisrank-orderedlistcanbeconsideredtobethemostdifferentiallyexpressed or variable genes in the study. We have previously shown that this type of ordering approach well captures the most active genes in a set of samples 33 .To demonstrate the number of genome-wide significant eQTL and eSNPs as a function of differential gene expression, we binned the expression traits into quartiles (Q1RQ4) on the basis of the rank-ordered gene list, with each bin containing 5,939 genes and the bins increasing in significance with respect to differential expression, from Q1 to Q4. Heritability, genome-wide linkage and association analysis. All subjects were clustered into families in which each proband is related to at least one other proband within five meiotic events; members of the IFB cohort were clustered into 209 families with 938 contributing individuals, and those from IFA were clustered into 124 families with 570 contributing individuals. Individuals in these cohorts were genotyped with 1,732 microsatellites uniformly distributed across the human genome, as described previously 35 . Each gene expression trait was treated as a quantitative trait. For the heritability calculations, linkage ana- lysis and association to genetic markers, the expression trait values were first adjusted for relevant covariates such as sex, age, blood cell count and BMI using multiple linear regression analysis as (age1age 2 1neutrophil1monocyte1 lymphocyte)3sex1traitinbloodandas(age3age 2 1log(BMI))3sex1trait for IFA. Traits were then standardized by mapping the distributions of the inverse normal transformation to each of the expression traits onto a normal distributionwithameanof0andavarianceof1.Thiswasdonetoeliminatethe effect of outliers on all subsequent analyses. To calculate the heritability, a poly- genic model was fitted to determine how much of the variation in the trait was caused by genetic effects. To carry out these calculations, we used SOLAR 2.0, a publicly available software package for human genetic analysis 36 . Linkage analysis and the calculation of IBD matrices used in the heritability calculations were carried out using the program Allegro 37 . The linkage analysis was based on a locally most-powerful score statistic for a gaussian variance component model with an additive variance component and assuming heritability for each trait was known. Significance was assessed using the expo- nential tilting method 38 , which has previously been demonstrated to give accu- rate type I error rates 39 . The accuracy of type I error rates was verified for the present score statistic by carrying out extensive simulation analysis, including simulations that assumed various deviations from the gaussian variance com- ponent model 40 . Multiple testing for significance was taken into account through the use of FDR procedures 21 . The software QVALUE was used in the calculations 21 . The proportion of significant tests g was estimated as g51?p 0 , where p 0 is the estimate of the overall proportion of true null hypotheses. In estimating p 0 , the pi0.meth5??bootstrap?? option in the QVALUE software was used. Controlling for multiple testing in the genome-wide association scans. To controlformultipletestinginthegenome-wideassociationscanscarriedouton the gene expression traits in a subset of the IFA cohort, we used simulations to adjust the P values for each trait for the number of SNPs tested. In each simu- lation, we permuted the gene expression trait values for the 150 individuals and recalculated the association test for all SNPs in the 2Mb window centred at the probe sequence location for the corresponding gene. This was repeated up to 50,000 times depending on the significance of the original cis association iden- tified for the expression trait in question. More specifically, if we define the Bonferroni-adjusted P value, P Badj ,asP3N, where P is the unadjusted P value andNisthenumberofSNPstested,thenumberofpermutations 41 performedfor each trait was selected as 100/P Badj . The minimum number of permutations performedforanygivenexpressiontraitwas500.Thiswassufficientforroughly 70% of the traits. An adjusted P value was then calculated as the fraction of simulations that produced an association for any SNP tested that was at least as significantasthemostsignificantcisassociationobservedintheoriginaldataset. For the X chromosome, the permutations were done preserving the sex of the individuals.Thepermutationtestwasappliedtothosetraitswherethestrongest cis association corresponded to a P.0.000001, whereas for traits with more significant cis associations a simple Bonferroni correction was used to calculate the adjusted P values. The Bonferroni adjustment was applied to approximately 10% of the traits. Multiple testing for significance was then taken into account through the use of the FDR procedures 21 . Here, the calculated P values (as described above) were used as the input to estimate the overall FDR. Assessing the significance of trans-acting eQTL signals. We defined a linkage forageneexpressiontraitasbeingtrans-acting(distaltothephysicallocationof the probe) if the associated LOD score curve peak was located on a different chromosome to the physical location of corresponding probe sequence. To assess the significance of the observed trans-acting eQTL signals, we created 10 setsofsimulatedgenotypesforallofthe1,732microsatellitemarkersusingdrop- downsimulations,undertheassumptionofnolinkageanywhereinthegenome, forthesamefamilystructureasthatusedinthelinkageanalysis.Foreachmarker, thesimulatedgenotypesmatchedtheoriginalgenotypesbothintermsofmissing genotypes and in terms of the frequency distribution of the genotypes for each marker.Wethenranthelinkageanalysisoneachofthesimulateddatasetsforall of the 20,877 uniquely mapped traits. From each linkage run, we identified the strongest trans-acting eQTL for each gene expression trait. Combining the resultsforall20,877geneexpressiontraitsoverall10simulateddatasetsyielded areferencedistributionof208,770ofthestrongesttrans-actingeQTLdetectedin thesimulateddata.Bycomparingtheobservedtrans-actingeQTLdistributionto this reference distribution, we assigned empirical P values to the trans-acting eQTL signals observed in the original analysis. Assessing the significance of eQTL hotspots. Given the strong correlation structure among gene expression traits, if one expression trait falsely links to a given genomicregion thenit is possible that many otherexpression traitshighly correlated with this expression trait may also falsely link to the given genomic region. To assess whether eQTL hotspots were artefacts driven by false-positive eQTL of highly correlated expression traits, we again used the eQTL results generatedonthe 10 simulateddata sets describedabove. Using thesame linkage threshold as that used in the observed data to detect trans eQTL, we examined whether hotspots were detected that were of similar magnitude or greater than what was detected in the observed data. In all of the simulated data sets, we observedhotspotsofsimilarorgreatermagnitudethanthehotspotswedetected in the observed data (see Supplementary Fig. 1), suggesting that the hotspots detected in the observed data could be due to false linkages of highly correlated gene sets. Construction of the adipose co-expression network. A previously described weighted gene co-expression network reconstruction algorithm was used to reconstruct the human and mouse co-expression networks 27 . The weighted net- work reconstruction algorithm involved first constructing a matrix of Pearson correlations between all gene expression pairs. The correlation matrix was then transformed into an adjacency matrix using a power function f(x)5x b . The adjacency matrix defines the weighted co-expression network. The parameter bofthepowerfunctionwasdeterminedsuchthattheresultingadjacencymatrix was approximately scale-free based on a previously proposed model-fitting index 27 . This index is defined as the coefficient of determination (R 2 ) of the linear model constructed by regressing log(p(k)) onto log(k), or by regressing log(p(k))ontolog(k)1k,where k representsthenumberofedgesconnectingto the given node and p(k) is the frequency distribution of the degree k in the co- expression network. The model-fittingindex of a perfectscale-free network is 1. Theexponentof thepowerfunction,b, waschosento bethesmallestvalue such that the co-expression network exhibited the scale-free property, that is, the model-fitting index R 2 $0.8. doi:10.1038/nature06758 Nature Publishing Group�2008 The adjacency matrix was further transformed into a topological overlap matrixtomorereadilyidentify modulesofhighlyco-regulatedgenes.Thetopo- logical overlap captures not only the direct interaction between two genes but also their indirect interactions through all the other genes in the network. Traditionally, the connectivity of a node is defined as the sum of its connection strengths in the adjacency matrix with all other genes in the network. Here we extended the definition to the topological overlap matrix and derived a topo- logicaloverlapconnectivitymap.Moduleidentificationwasconductedthrough a dynamic programming procedure to search the topological overlap matrix ordered by hierarchical clustering for maximum sets of inter-connected genes 28 . Testingtheassociationofcisvariantstoobesitytraits.Totesttheassociationof cis variants for the genes in the MEMN module to the obesity traits BMI (or PBF), we tested the difference between two regression models: model 1 is log(BMI j ),sex j 3(age j 1age j 2 ), where subscript j refers to individual j, and model 2 is log(BMI j ),sex j 3(age j 1age j 2 )1g ji 1?1g jn , where g ji is the minor allele count for individual j and cis variant i. To adjust for relationship andforassociationsoccurringbychance,wesimulatedgenotypesforallofthecis variantsthroughthegenealogyof708,683Icelanders.Here,foreachofthe20,000 simulated sets of genotypes we constructed, we repeated the association tests between the cis variants and the obesity traits. We then calculated adjusted P values as the fraction of simulations that yielded equally or more significant association between a particular trait and the corresponding cis variants. These adjusted P values are summarized in Table 2 in the main text. 32. Gulcher, J. R., Kristjansson, K., Gudbjartsson, H. & Stefansson, K. Protection of privacybythird-partyencryptioningeneticresearchinIceland.Eur.J.Hum.Genet. 8, 739?742 (2000). 33. He, Y. D. et al. Microarray standard data set and figures of merit for comparing data processing methods and experiment designs. Bioinformatics 19, 956?965 (2003). 34. Hughes, T. R. et al. Functional discovery via a compendium of expression profiles. Cell 102, 109?126 (2000). 35. Kong,A.etal.Ahigh-resolutionrecombinationmapofthehumangenome.Nature Genet. 31, 241?247 (2002). 36. Almasy, L. & Blangero, J. Multipoint quantitative-trait linkage analysis in general pedigrees. Am. J. Hum. Genet. 62, 1198?1211 (1998). 37. Gudbjartsson,D.F.,Jonasson,K.,Frigge,M.L.&Kong,A.Allegro,anewcomputer program for multipoint linkage analysis. Nature Genet. 25, 12?13 (2000). 38. Kong, A. & Cox, N. J. Allele-sharing models: LOD scores and accurate linkage tests. Am. J. Hum. Genet. 61, 1179?1188 (1997). 39. Badner, J. A., Gershon, E. S. & Goldin, L. R. Optimal ascertainment strategies to detect linkage to common disease alleles. Am. J. Hum. Genet. 63, 880?888 (1998). 40. Amos, C. I. Robust variance-components approach for assessing genetic linkage in pedigrees. Am. J. Hum. Genet. 54, 535?543 (1994). 41. Churchill, G. A. & Doerge, R. W. Empirical threshold values for quantitative trait mapping. Genetics 138, 963?971 (1994). doi:10.1038/nature06758 Nature Publishing Group�2008 "
Add Content to Group
|
Bookmark
|
Keywords
|
Flag Inappropriate
share
Close
Digg
Facebook
MySpace
Google+
Comments
Close
Please Post Your Comment
*
The Comment you have entered exceeds the maximum length.
Submit
|
Cancel
*
Required
Comments
Please Post Your Comment
No comments yet.
Save Note
Note
View
Public
Private
Friends & Groups
Friends
Groups
Save
|
Cancel
|
Delete
Please provide your notes.
Next
|
Prev
|
Close
|
Edit
|
Delete
Genetics
Gene Inheritance and Transmission
Gene Expression and Regulation
Nucleic Acid Structure and Function
Chromosomes and Cytogenetics
Evolutionary Genetics
Population and Quantitative Genetics
Genomics
Genes and Disease
Genetics and Society
Cell Biology
Cell Origins and Metabolism
Proteins and Gene Expression
Subcellular Compartments
Cell Communication
Cell Cycle and Cell Division
Scientific Communication
Career Planning
Loading ...
Scitable Chat
Register
|
Sign In
Visual Browse
Close
Comments
CloseComments
Please Post Your Comment