• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of plosbiolPLoS BiologySubmit to PLoSGet E-mail AlertsContact UsPublic Library of Science (PLoS)View this Article
PLoS Biol. Sep 2010; 8(9): e1000480.
Published online Sep 14, 2010. doi:  10.1371/journal.pbio.1000480
PMCID: PMC2939022

Polymorphic Cis- and Trans-Regulation of Human Gene Expression

Jonathan Flint, Academic Editor

Abstract

Expression levels of human genes vary extensively among individuals. This variation facilitates analyses of expression levels as quantitative phenotypes in genetic studies where the entire genome can be scanned for regulators without prior knowledge of the regulatory mechanisms, thus enabling the identification of unknown regulatory relationships. Here, we carried out such genetic analyses with a large sample size and identified cis- and trans-acting polymorphic regulators for about 1,000 human genes. We validated the cis-acting regulators by demonstrating differential allelic expression with sequencing of transcriptomes (RNA-Seq) and the trans-regulators by gene knockdown, metabolic assays, and chromosome conformation capture analysis. The majority of the regulators act in trans to the target (regulated) genes. Most of these trans-regulators were not known to play a role in gene expression regulation. The identification of these regulators enabled the characterization of polymorphic regulation of human gene expression at a resolution that was unattainable in the past.

Author Summary

Cellular characteristics and functions are determined largely by gene expression and expression levels differ among individuals, however it is not clear how these levels are regulated. While many cis-acting DNA sequence variants in promoters and enhancers that influence gene expression have been identified, only a few polymorphic trans-regulators of human genes are known. Here, we used human B-cells from individuals belonging to large families and identified polymorphic trans-regulators for about 1,000 human genes. We validated these results by gene knockdown, metabolic perturbation studies and chromosome conformation capture assays. Although these regulatory relationships were identified in cultured B-cells, we show that some of the relationships were also found in primary fibroblasts. The large number of regulators allowed us to better understand gene expression regulation, to uncover new gene functions, and to identify their roles in disease processes. This study shows that genetic variation is a powerful tool not only for gene mapping but also to study gene interaction and regulation.

Introduction

Expression levels of genes, like many phenotypes, vary among normal individuals. Since gene expression underlies cellular characteristics and functions, variation in gene expression contributes to phenotypic diversity and differences in disease susceptibility. Previously, we and others demonstrated that there is a genetic basis to individual variation in gene expression [1][6]. This facilitates studies to identify sequence variants that influence expression levels of genes. Since expression phenotypes of many genes are studied in parallel, results from genetics of gene expression (GOGE) studies contribute to the understanding of global gene regulation.

GOGE studies that treated expression levels as quantitative traits in family-based linkage [3],[7] and population-based association analyses [5],[8],[9] have uncovered polymorphic regulatory regions that contribute to variation in human gene expression. However, the regulatory regions were large, often megabases in size; thus, the identity of most polymorphic regulators remained unknown. In this GOGE study, we analyzed a large sample in linkage analyses, then we used deep sequencing of transcriptomes (RNA-Seq) to guide association-based fine mapping. The results allowed us to narrow the regulatory regions and identify cis- and trans-acting polymorphic regulators of ~1,000 human genes. These results facilitated molecular validation and analyses of the mapping data. This is an important advance in human genetic studies where such validations have largely been impossible. In previous human GOGE studies, the resolution of the mapping results was inadequate; hence, regulators were not identified, while other gene mapping studies focused on complex phenotypes, such as human diseases that are often not amenable to molecular analyses. Thus, the end points of many human genetic studies showed genotype-phenotype connections statistically but not molecularly.

Here, we have an unusual opportunity to begin to bridge the gap between genetic and mechanistic studies. Knowing the identity of the regulators, we were able to validate the cis- and the trans-regulatory relationships using different approaches. For genes that are cis-regulated, we used RNA-Seq to show differential allelic expression. For the trans-regulatory relationships, we altered the expression of the regulators by gene knockdowns and metabolic perturbations and showed that manipulations of the regulators affected the expression levels of the corresponding target genes. We also demonstrated direct interactions between regulators and their target genes by chromosome conformation capture.

Another goal of this study is to examine the role of cis- and trans-acting polymorphisms on human gene expression. Previously, GOGE studies in model organisms and humans appear to disagree on the proportion of polymorphic cis- and trans-acting regulators. In yeast, fly, and mouse studies, most of the regulators act in trans [2],[10][12]. In contrast, human studies focused mostly on cis-acting variants. This apparent discrepancy is likely due to differences in sample sizes. Studies in model organisms used larger sample sizes and thus were able to find trans-acting regulators that have smaller effects on gene expression than cis-regulators [13],[14]. In contrast, early human studies of GOGE used relatively small sample sizes, such as samples collected by the International HapMap Consortium [5],[8]; hence they identified mostly cis-regulators. This and the discovery of cis-regulation of disease susceptibility genes such as ORMDL3 (asthma) [15] led to suggestions that cis-acting variants are significant contributors to variation in human gene expression. However, it is unlikely that the regulatory landscapes are different between humans and other organisms. In humans, trans-acting regulators possibly also play an important role. Several studies [3],[4],[9] have suggestive evidence for the important contribution of trans-acting variants. Recently, studies that used RNA-Seq to analyze gene expression phenotypes in HapMap samples found cis-acting variants for less than 10% of human genes [16],[17]. These studies suggest that along with cis-variants, trans-acting polymorphisms contribute to individual variation in human gene expression. Here, to address this, we used a large sample size and identified hundreds of polymorphic trans-regulators. These findings confirm that as in other organisms, there are many sequence variants in the human genome that act in trans to influence gene expression.

Many of the identified trans-regulators were previously not known to play a role in gene regulation. Over 60% of the regulators are not transcription factors or known signaling factors. However, the trans-regulators are not randomly distributed; instead they tend to be found in the same functional pathways as their target genes. While the regulators were discovered in analysis of immortalized B-cells, we showed that the regulatory relationships were also found in primary fibroblasts. Thus, natural variation in gene expression allowed the identification of polymorphic expression regulators, which then enabled us to develop a deeper understanding of gene regulation.

Results

Linkage Scans

We obtained genotypes of single nucleotide polymorphisms (SNPs) and measured the expression levels of genes in immortalized B-cells from members of 45 Centre d'Etude du Polymorphisme Humain (CEPH) Utah pedigrees [18] using microarrays. We focused our analysis on 4,793 expressed genes that show variation in expression levels among individuals and carried out genome-wide linkage analysis (see Methods). From those analyses, we selected 1,681 (35%) phenotypes for further studies using a threshold of t>4 (a logarithm of odds (lod) score of ~3.4, and a genome-wide corrected significance level of approximately 0.05 [19]) (see Methods). Figure 1 shows examples of genome scan results.

Figure 1
Genome scans of six expression phenotypes.

We expected to find polymorphic regulators of the expression phenotypes in the candidate regions identified by the linkage scans. Hence we examined the linkage peaks to determine their locations relative to the genomic addresses of the target genes. To take into account the imprecision of linkage, we define regulatory regions that are within 5 Mb of the target genes as proximal and those that are greater than 5 Mb or on another chromosome as distal to the target genes [20]. By this definition, among the 1,681 phenotypes with evidence of linkage at t>4, we found that 70 (4.2%) phenotypes have proximal regulators, 1,574 (93.6%) phenotypes have distal regulators, and 37 (2.2%) phenotypes have both proximal and distal regulators. Ninety-four percent of the distal regulators are on a different chromosome than their corresponding target genes. These results suggest that trans-acting regulation contributes appreciably to variation in gene expression.

Family-Based and Population Association

Linkage scans provided regulatory regions for over 1,600 expression phenotypes. To confirm these results, we carried out family-based and population-based association analyses with markers within the candidate regulatory regions. In addition to confirming the linkage findings, association mapping allows us to take advantage of historical recombinations in order to narrow the candidate regions.

Proximal linkage peaks

For the 107 (70+37) phenotypes where the linkage peaks are proximal to the target genes, we assumed that they are likely to be cis-regulated, so we tested SNPs within and 50 kb up- and downstream of the target genes. Among these 107 phenotypes, we had informative genotypes for 100 phenotypes to carry out family-based association analysis by quantitative transmission disequilibrium test (QTDT) [21]. From the analysis of the members of the 45 CEPH pedigrees by QTDT, 63 of the 100 phenotypes showed significant evidence (nominal p≤0.001) for the combined presence of linkage and association (Table S1). These results confirm the linkage findings and support that these phenotypes are cis-regulated.

Using expression data and genotypes of 86 unrelated individuals, we carried out population-based association analysis and found significant evidence (nominal p<0.005) for population associations between gene expression levels and SNPs within or near the target genes (Table S1) for 47 (75%) of these 63 phenotypes. We also estimated the variation in expression explained by the cis-acting determinants by calculating R2 using results of the linear regression analyses in population association studies. For the 47 phenotypes, the average R2 is 0.25 (range = 0.09 to 0.75). For 17 of these phenotypes, the cis-variants explained more than 30% of the individual variation in their expression levels. This provides an estimate of the contribution of cis-variants; the fraction not explained this way includes non-genetic factors (including environment) and other genetic factors not in linkage disequilibrium with the cis-acting determinants.

These findings further support the linkage results and provide evidence of differential allelic expression of these genes (see Figure 2 for examples). We also looked for molecular evidence of cis-regulation (see section below on differential allelic expression under “molecular validation”).

Figure 2
Allelic expression from RNA-Seq confirms prediction by association analysis.

Distal linkage peaks

We followed up results for the 1,611 (1,574+37) phenotypes with significant distal linkage peaks using QTDT. Unlike proximal peaks where we can look for cis-acting variants within or near the target genes, there are no obvious regions to look in the distal peaks. This is particularly difficult when the linkage peaks are large and contain several potential regulators. One option is to test all the SNPs under the linkage peaks for evidence of association with expression levels of the corresponding target genes; however, that would result in a severe multiple testing problem. Instead we identified the genes that are expressed in our B-cells by RNA-sequencing and testing the expressed genes for association with expression levels of their corresponding target genes.

In RNA-sequencing, we are not limited to studying only genes that are represented on the microarrays. With sufficient coverage, sequencing data also allow us to detect genes that are expressed at lower levels. This is important since gene expression regulators, such as transcription factors, are often expressed at low levels. We sequenced the cDNA samples of 41 CEPH HapMap (CEU) [22],[23] individuals using the Illumina technology [24]. For each sample, we obtained ~40 million reads, each 50 nucleotides long or about 2 Gb of sequences per sample. We mapped the short read sequences to the reference human genome (hg18) using the software MAQ [25]. About 83% of the sequences mapped uniquely to the reference sequence. The other 17% mapped to multiple sites in the genome (repetitive sequences or sequence motifs that are common in gene families) or failed to map anywhere in the reference genome (including exon junctions which would not map to the reference genome sequence). We compared the expression levels of genes in B-cells of the same individuals from RNA-Seq with those from our microarrays, the average correlation was 0.76 (range = 0.73 to 0.80, highly similar to those in other studies [16],[17],[26]). We also compared genotypes from our RNA-Seq to those from the HapMap Consortium and found the average concordance rate was 98.6%. These results gave us confidence in the data so we used them to identify expressed genes in the candidate regions identified by linkage scans.

We tested SNPs in the expressed genes within each linkage peak for association with their candidate target genes. For genes that are expressed with RPKM≥1 [26] in the linkage regions, we carried out QTDT analyses using SNPs within and 5 kb up- and downstream of the regulators in all members of the 45 CEPH families. Among the 1,611 phenotypes with distal linkage peaks, we excluded 94 phenotypes whose candidate regulatory regions were over 20 megabases in size. Of the 1,517 remaining phenotypes, the expression levels of 103 (6.8%) phenotypes showed evidence of linkage and association (nominal p≤0.001; FDR = 0.02) with SNPs in trans-regulators. Since trans-acting regulators have weaker effect than cis-regulators, we also looked at results using lower thresholds of p<0.01 and 0.05 (FDR 3 and 8%); there are 518 (34%) and 917 (60%) phenotypes that met these thresholds, respectively. Among these 917 phenotypes, the expression levels of 112 genes are influenced by two or more unlinked polymorphic trans-regulators. Thus, the analysis revealed 1,036 regulator–target gene pairs. Table 1 shows the top 20 trans-regulator–target gene pairs from our linkage and association study (see 2 for the top 200 regulator–target gene pairs). As we did for the cis-acting regulators, we tested the trans-regulators identified by QTDT for allelic association with their target genes using population association tests. Although the sample size is small (n = 86), we found significant association for a set of trans-regulators and their target genes. Among the regulators of the 917 target genes, SNPs in 58 (6%) and 318 (35%) regulators showed significant allelic association with expression levels of their target genes at nominal p<0.005 and p<0.05, respectively. To estimate the influence of these trans-acting variants on individual variation in gene expression, we calculated R2 for the 318 phenotypes. The average R2 is 0.07 (median = 0.07; range = 0.05 to 0.24). For 53 phenotypes, the trans-acting variants explained 10% or more of the individual variation in their expression levels. These results show that individual differences in gene expression can be explained by DNA sequence polymorphisms in trans-acting regulators.

Table 1
Expression phenotypes with the strongest evidence of linkage and association to polymorphic trans-regulators.

Known and Newly Discovered Regulatory Relationships

To check the validity of these findings, we looked for known regulatory relationships among the regulator–target gene pairs that we identified in the genetic analyses. An example of such known relationship is MRLC2, which encodes myosin regulatory light chain 2 and its regulator myocyte enhancing factor 2A, MEF2A, a transcription factor that is known to affect muscle gene expression, including MRLC2 [27]. Our linkage results identified chromosome 15q26 (linkage t = 4.9) as the candidate regulatory region for the expression level of MRLC2. Using association analyses, we narrowed the candidate region and rediscovered MEF2A as the regulator of expression level of MRLC2 (QTDT p = 0.008; population association p = 0.04, rs325380). Another example is TTC5 as the polymorphic regulator of HSP90AA1 expression. Previous studies showed that a mouse protein phosphatase that contains a tetratricopeptide repeat regulates heat shock protein 90; this regulation occurs by dephosphorylation, which is mediated by the binding of heat shock protein 90 to the tetratricorepeat domain of the phosphatase [28],[29]. Our results showed that the expression of human HSP90AA1 is influenced by variants in TTC5, a gene with a tetratricopeptide repeat (linkage t = 5.4; QTDT p = 0.01, rs11623837). The “rediscovery” of these known regulatory relationships confirms that our approach can identify trans-acting regulators of human gene expression.

For the 20 regulator–target gene pairs in Table 1, we checked for co-occurrence of the names of the regulators and target genes in the literature using a text-mining program, Chilibot [30], to determine if any of these regulatory relationships are known. We also queried PubMed for such co-occurrences. Among these 20 pairs, only one pair (MBP and PDE4B) has been shown to have interactive relationships in Chilibot. Thus, many of these regulator-target relationships are likely unknown previously.

Molecular Validation

RNA-Seq to identify differential allelic expression in cis-regulated genes

To validate the polymorphic cis-regulation identified in our mapping study, we used the RNA-Seq data of individuals in the HapMap Project as described above. We used the sequencing data to assess differential allelic expression (DAE) [16],[17],[22],[23].

The digital nature of the sequence data allows us to use the heterozygous genotypes in each transcript to determine whether two allelic forms of a transcript are expressed in equal abundance [31][33]. Among the 107 expression phenotypes with proximal linkage peaks, 67 have at least one SNP where there are 2 individuals who are heterozygous at that SNP (see Methods). We examined these heterozygous samples for evidence of DAE. For many of these genes, we have data for multiple SNPs from an average of 7.2 individuals (median = 6). Among the 67 genes, 43 genes (64%) showed significant evidence (p<0.01, chi-square test) of departure from equal expression of the two allelic forms of the genes. For the 273 exonic SNPs in these 43 genes, we calculated an “allelic expression ratio” a/(a+b), where a and b are the numbers of sequence reads for the two alleles. Figure S1 shows these allelic expression ratios and their departures from 0.5. For 31 of these genes, the exonic and the associated SNPs from our mapping study are part of the markers in the HapMap Project [23]; thus phased haplotype data are available. Using these haplotypes, we showed that for 28 (90%) genes, the predicted expression in the association study was confirmed by RNA-Seq data (see Figure 2 for examples). Hence, the DAE results confirm the findings from our mapping studies and show that the majority of genes (~65%) with proximal linkages are cis-regulated. For the remaining phenotypes, either we do not have adequate sample size or read coverage to detect subtle evidence of DAE, or they are regulated by trans-regulators that mapped close to the target genes.

Our sequence data allow us to examine DAE of many more genes in addition to those with proximal linkage peaks. There are 5,782 genes that can be studied for DAE. Among them, 1,029 (18%) and 1,501 (26%) genes showed significant evidence of DAE at p value (chi-square test) thresholds of 0.001 and 0.01, respectively. The 18% to 26% of genes that show DAE provide another estimate of the number of genes in our B-cells that are cis-regulated. This proportion is similar to the 12% estimated by Price and colleagues using admixture analysis [34], the 30% by Pastinen and colleagues by hybridization of cDNA to SNP arrays [35], the 11% to 22% by Church and colleagues using RNA-Seq, and our mapping study (we found 6.5% and 24% of phenotypes to have proximal peaks, at t threshold of 4 and 5, respectively).

Molecular validation of trans-acting regulators: gene knockdown

To validate the trans-regulator–target gene relationships, we carried out molecular analyses. First, we performed gene knockdown studies. We used short interfering RNA (siRNA) to silence 25 potential regulators, and then assessed the effects by measuring the expression of the target genes (we tested cells from 4 to 6 individuals) [36]. Among the 25 regulators, we included MEF2A and TTC5 as positive controls. The remaining 23 regulators were selected based on availability of siRNAs and they span a range of QTDT significance from p = 10−5 to 10−2 (for BLM-NUSAP1). We did not select regulators that were supported by the most significant p-values. We reasoned that if we can confirm molecularly the regulators with relatively modest statistical supports, then the ones with more significant mapping results are likely to be true regulators.

Among the 25 regulators, successful knockdown was achieved for 18 regulators. The expression of these regulators including MEF2A and TTC5 decreased significantly (p<0.05) by about 20% to 85% in four or more independent samples, whereas no changes in the expression of the regulators were observed when siRNAs with no homology to the regulators were used (Table 2). We then measured the expression levels of the target genes following knockdown of their regulators. The expression levels of 13 (72%) target genes including MRLC2 (target of MEF2A) and HSP90AA1 (target of TTC5) changed significantly (p<0.05) following the knockdown of their regulators (Table 2). The expression levels of the target genes, such as NUSAP1 (encodes a spindle associated protein) and SSR1 (signal sequence receptor alpha), changed by ~10% to 60%, while those of non-target control (GAPDH) did not change significantly after the knockdown of the regulators, suggesting that the changes in expression levels of the target genes were specific effects of silencing their regulators.

Table 2
Results of knockdown of trans-regulators.

We followed up three of these regulator–target gene pairs in primary fibroblasts (n = 2) to determine the cell-type specificity of the regulatory relationships. We carried out siRNA-mediated knockdown of BLM, ITGB4BP, and PSAP in fibroblasts. Following the silencing of BLM and ITGB4BP, we observed significant changes (p<0.05, t test) in expression of their target genes NUSAP1 and SLC25A11 as in the immortalized B-cells (Table S3A). However, the expression of HMGCS1 did not change significantly following knockdown of its regulator, PSAP, suggesting that this regulatory relationship may be specific for B-cells.

These results provide molecular support for the regulator–target gene relationships identified in our mapping studies. However, the lack of changes in expression levels of the target genes following the knockdown of their regulators does not argue against the regulatory relationships. The expressions of the regulators were only partly decreased by siRNAs; partial expression may be sufficient for regulation. In addition, many human genes have other family members that take over their functions upon knockdown of their expression.

INSR and its target genes

In addition to knockdown studies, we carried out another functional analysis that does not rely on transfection. One of the trans-regulators is the insulin receptor, INSR. Our mapping results identified six genes, ADD3, ARNT, ATIC, CCL5, LTB, and PSMD10, whose expression levels are regulated by the insulin receptor, INSR. Previously, studies have shown that ADD3 is involved in insulin receptor signaling [37]. In addition, Kahn and colleagues showed that following knockdown of ARNT, the expression of the insulin receptor was decreased in pancreatic islet cells [38]. If our mapping results are correct, then it would suggest that insulin receptor regulates the expression of ARNT, providing evidence of reciprocal regulation of ARNT and INSR or feedback mechanisms. To validate these regulatory relationships, we stimulated the insulin receptor by treating primary fibroblasts with insulin and measured the expression levels of INSR and its target genes. The fibroblasts allowed us to confirm the regulatory relationship in primary cells. Among the six genes, four (ADD3, ARNT, ATIC, and PSMD10) were expressed in fibroblasts, so we focused our analysis on these genes. We treated the cells with insulin; upon insulin treatment, the insulin receptor is activated but not the related IGF1 receptor, thus indicating that insulin was acting specifically through INSR (Figure 3). Insulin led to a biphasic response of INSR: 2 h after insulin treatment, the expression level of INSR among four individuals increased by an average of 12% (p = 4×10−5); this was followed by a 48% decrease (p = 0.004) at 6 h after exposure to insulin (Table 3). The expression levels of the four target genes also changed significantly (p<0.005) following insulin treatment (Table 3). These findings confirm our mapping results, which identified INSR as the polymorphic regulator of expression of ADD3, ARNT, ATIC, and PSMD10. They show that the regulatory relationships identified in immortalized B-cells are also found in primary fibroblasts.

Figure 3
INSR activation following insulin treatment in human primary fibroblasts.
Table 3
Changes in expression levels of insulin receptor target genes following insulin treatment.

Physical interactions between regulators and their target genes

We used data from chromosome conformation capture (3C) as additional validation of the regulatory relationships and to identify regulatory pairs that interact physically. Recently, using the same immortalized B-cells, Dekker and colleagues combined 3C with high-throughput sequencing, in a procedure they coined “Hi-C,” to identify regions of the human genome that interact physically [39]. We used their data to determine if any of our 1,036 regulator–target gene pairs (with population association p<0.05) interact. We found 75 of our gene pairs in their data (corresponding to 63 unique regulator–target gene pairs; some pairs were found more than once; see Table 4 and Table S7). The chance of finding two genes as regulatory pairs in our mapping study and as interacting partners in the Hi-C experiment by a different group randomly is very small; thus we believe that these are likely true interactions. The Hi-C results further confirm findings from our genetic studies. These data suggest that the regulators and their target genes may be co-transcribed in “transcription factories” [40][42].

Table 4
Examples of regulator–target gene pairs that interact physically based on Hi-C experiments* [39].

Since the Hi-C libraries were not sequenced exhaustively [39], some of our pairs may not be included in their results even though they interact physically. Nevertheless, the results provide additional information for some of the regulatory relationships and show that similar approaches can be used to extend the analysis.

Characteristics of the Trans-Regulators

The resolution of our mapping study allowed us to identify the polymorphic regulators of nearly 1,000 human genes. Instead of just confirming these results computationally, we used molecular approaches, which provide an independent method for assessing the findings. Even though we picked regulatory relationships that had modest statistical support (p = 10−5 to 10−2) from our mapping study, over 70% of the regulatory pairs are validated molecularly. Thus, we are reasonably confident that many of the 1,036 gene pairs have true regulatory relationships, so we went on to characterize them.

First, many of the trans-regulators were not known to influence gene expression. Among the 742 regulators, 112 (15%) are known transcription factors and 140 (19%) play a role in signaling pathways. The remaining genes have a variety of functions including metabolism (for example, MAN2A1, PDHA2) and protein transport or modification in the endoplasmic reticulum (for example, LMAN1, SEC31A).

Second, the target genes and their regulators often belong to the same functional pathways. For example, midasin (MDN1), which plays a role in protein processing [43], regulates the expression of dynactin 1 (DCTN1) and signal sequence receptor, delta (SSR4). Both of these target genes also participate in protein transport and processing in the endoplasmic reticulum [44],[45]. To test formally whether regulators and their target genes belong to the same functional groups more often than by chance, we annotated the regulators and target genes using Gene Ontology [46]. We found significantly (p = 0.008) more regulator–target gene pairs with the identical ontology annotation than random pairs of genes. The criterion we used is quite stringent since we required members of a gene pair to have the identical ontology grouping; it excludes regulator–target gene pairs that are in the same pathway but do not have the identical ontology. However, despite the stringent criterion, a significant result was obtained. Recent results from yeast studies also showed that regulators and their target genes share common gene ontology annotations [47].

Third, many of the trans-regulators have more than one target gene. We found 161 (22%) of the 742 trans-regulators influence the expression levels of two or more genes. Table 5 shows the 11 regulators that influence six or more target genes. Three of these regulators are known to play a role in transcription regulation through chromatin modification (AEBP2) or as transcription factors (BCL2, ZCCHC2). In addition, three of the regulators (PHLPP, RAMP1, WDR7) affect gene expression through signal transduction pathways. The remaining five regulators are not known to be gene expression regulators, including TTC5, which has no known function.

Table 5
Trans-regulators with six or more target genes.

Gene Network

The regulators with multiple target genes prompted us to examine interactions beyond the relationship between a gene and its regulator. To do so, we used our mapping results to construct directed gene networks. We connected regulators and their target genes using results from the QTDT analysis. The resulting network consists of 1,036 connections among 742 regulators and 917 target genes. As in many biological networks, the network connections follow a scale-free distribution (scale-free criterion = 0.98) [48]. On average, genes have 1.3 connections, but some genes have more connections such as those that regulate the expressions of several target genes. Figure 4 shows subnetworks for KIAA1468 and WDR7, which illustrate that some regulators have multiple target genes and some genes are regulated by more than one regulator. Unlike many gene networks, the nodes in our networks are connected by directed edges based on genetic data. DNA variants in the highly connected genes such as KIAA1468 and WDR7 influence the expression of many genes that are directly and indirectly connected to them. The WDR7 subnetwork shows the connections between ITPR2 and SSR1, as well as several other genes, including SYNCRIP [49] and RHOC [50], that play a role in the endoplasmic reticulum; thus polymorphisms in WDR7 likely affect protein processing and secretion, the primary functions of the endoplasmic reticulum. Prior to this analysis, the function of WDR7 was unknown except that it has been found to influence the age of onset of multiple sclerosis [51] in genome-wide association studies. Results from our analyses offer WDR7 as a mechanistic link between multiple sclerosis and functions of the endoplasmic reticulum. The efficiencies of the endoplasmic reticulum can influence susceptibility to multiple sclerosis in different ways. First, studies have shown that the endoplasmic reticulum plays a key role in immunity, for example in ensuring the maturation of B-cells to immunoglobulin secreting plasma cells [52]. In addition, during myelination, cells such as oligodendrocytes rely on the endoplasmic reticulum to produce a large amount of plasma membrane [53]. Thus by altering the efficiencies of endoplasmic reticulum, variants in WDR7 can influence individual susceptibility to multiple sclerosis through the autoimmune and/or myelination pathways. Besides WDR7, other regulators in our network have also been identified as disease susceptibility genes (see examples in Figure S2).

Figure 4
Directed subnetworks.

Discussion

The main focus of this study is to assess and determine the polymorphic regulation of human gene expression. We used linkage analyses to locate the polymorphic regulatory regions for 1,681 human genes. About 6% to 24% of these regulatory regions were close (proximal) to the target genes, and the remaining regions were further away (distal) from the target genes and mostly on other chromosomes. In follow-up association studies and sequence-based DAE analyses, at least 60% of phenotypes with proximal linkage peaks were found to be cis-regulated; this result is similar to findings in yeast [2],[11]. The remaining phenotypes with proximal linkages are likely regulated by trans-regulators that are close to their target genes. For 917 genes with distal linkage peaks, we narrowed the regulatory regions and identified the trans-acting polymorphic regulators. For some genes, we identified more than one trans-regulator; thus, the results include a total of 1,036 regulator–target gene pairs. Previous genetics of human gene expressions studies uncovered only the regulatory regions; here, we improved the resolution significantly by finding the individual regulators.

The results allowed us to explore previously unknown aspects of gene regulation. We found that many genes besides transcription factors can influence the expression of other genes. Similar results were found in yeast [2],[54]. Only 34% of the polymorphic trans-regulators that we identified are transcription or signaling factors. Many of the regulators are found in the same functional pathways as their target genes. By eliminating the recruitment of regulators from other pathways, cells can alter gene expression quickly when a cellular process requires a gene to be turned on or off. We do not know yet how polymorphisms in these genes influence expression in trans. One possibility is that the sequence variants in or near the regulators affect their own message and protein levels (cis-regulation) and lead to differential expression levels of the target genes that they regulate (trans-regulation). Based on our RNA-Seq data, ~20% of the trans-regulators show such DAE or cis-regulation. Alternately, the sequence variants in the regulators can affect their structures, stabilities [55][57], and functions by changing modifications such as phosphorylation status [58]; these in turn can affect the expression of their target genes. We also do not know whether the regulatory relationships are direct or indirect. Since regulatory relationships are highly complex and most genes are regulated by multiple genes in different feedback mechanisms, we expect most regulatory relationships are indirect. The Hi-C data show that some of the regulator–target gene pairs interact physically at the DNA level; the results imply that they may be co-transcribed perhaps in “transcription factories” [40],[41] where others have found trans interactions among regulators and their target genes [42].

Although the regulatory mechanisms remain unknown, we found that regulatory relationships are shared among cell types. For a number of genes, the trans-regulatory relationships that we identified in immortalized B-cells are also found in primary fibroblasts. Others have found that cis-regulation of some genes is shared across cell types [4],[9],[59],[60]; here, we provide evidence that trans-regulation can also be shared across different cells. This is important since many cell types in humans are not easily accessible. These results suggest that it may be possible to use more readily available cells for analysis and apply the results across cell types.

Our results have implications beyond regulation of gene expression. It provides insight into disease mechanisms. We already discussed the role of WDR7 as regulator of genes in the endoplasmic reticulum and the implication of this finding for multiple sclerosis. There are additional examples: for instance, we identified inositol 1,4,5-triphosphate receptor, type 2 (ITPR2) as a regulator of signal sequence receptor, alpha (SSR1, also known as TRAPA). Both genes function in the endoplasmic reticulum [45],[61] and are susceptibility genes for amyotrophic lateral sclerosis (ALS) [62],[63], but the connection between them was previously unknown. By showing the regulatory relationship between these two endoplasmic reticulum genes, we implicate inefficient endoplasmic reticulum function in the development of ALS. The role of the endoplasmic reticulum in ALS is further supported by another regulator–target gene pair, ALS2 and its target gene, SEC22A. ALS2 is the mutated gene in juvenile ALS [64]. Despite several knockouts of Als2 in mice, its role in ALS has not been identified. Here, we found that it regulates expression of SEC22A (linkage t = 5.6, QTDT p = 0.004 rs3219171), which mediates endoplasmic reticulum to Golgi transport. These findings have therapeutic implications; a recent study suggested that survival of ALS mice can be extended by blocking endoplasmic reticulum stress induced cell death [65].

Unraveling the control of gene expression of human cells is critical for understanding normal cellular processes and disease mechanisms. It is difficult to identify trans-acting regulators. They are not restricted to regulatory genes such as transcription factors. The hundreds of regulators identified in our study do not share protein domains nor belong to particular protein families. Thus, the search cannot be guided by known regulatory functions or protein domains alone. We show that GOGE study along with RNA-Seq and molecular analyses allow the identification of cis- and trans-acting regulators of human gene expression. This approach makes it possible to determine how individual genes are regulated and to discover regulatory pathways that maintain cellular functions in human cells.

Materials and Methods

CEPH Samples, Genotypes, and Expression Phenotypes

The data were from members of 45 three-generations CEPH families (CEPH 1328, 1330, 1331, 1332, 1333, 1334, 1340, 1341, 1344, 1345, 1346, 1347, 1349, 1350, 1353, 1354, 1356, 1357, 1358, 1362, 1375, 1400, 1408, 1413, 1416, 1418, 1420, 1421, 1423, 1424, 1444, 1447, 1451, 1454, 1456, 1458, 1459, 1463, 1477, 1582, 13281, 13291, 13292, 13293, 13294). Low-density genotypes for 4,600 autosomal SNP markers were obtained using the Illumina Linkage Panel (v3). We used PedStats [66] to check for mendelian inconsistencies. This resulted in the removal of 297 genotypes at 209 distinct SNP markers. High-density genotypes for some of the grandparents and parents were obtained from the International HapMap Project (HapMap 22), and for those families who are not part of the HapMap project, the parents and one randomly selected child in each family were genotyped using the Human SNP Array 5.0 (Affymetrix), which assays for ~500,000 SNP loci throughout the human genome. Then, high density genotypes for family-based association (QTDT) on all subjects were obtained by inference using the low-density genotypes and high-density genotypes on selected individuals [67].

For expression analysis, immortalized B cells were grown at a density of 5×105 cells/mL in RPMI 1640 with 15% fetal bovine serum, 2 mM L-glutamine, and 100 U/mL penicillin-streptomycin. RNA was extracted from the cells and hybridized onto Human Focus Arrays (Affymetrix; ~8,500 RefSeq Genes on each array). Samples were grown and processed in random order to minimize batch effects. Samples of sibs were processed together only by chance. Expression intensity was scaled to 500 using the global scaling method implemented in the Expression Console software from Affymetrix and transformed by log2.

The RNA samples for 94 CEPH grandparents (from the 45 families) were hybridized onto duplicate arrays. This allows us to calculate “variance ratio” as a measurement of variability in expression levels among individuals relative to the measurement noise. For each expressed gene (called “present” by Affymetrix Expression Console in 80% or more grandparents), we calculated this measure as the ratio of the variance in mean expression levels among individuals to the mean of the variance of the replicates within individuals: (variance of Mi)/(mean of si2). There are 4,793 genes with a variance ratio >1. We focused on these genes in our analyses.

Analysis of Linkage and Association

Multipoint genome-wide linkage analysis was done by SIBPAL in S.A.G.E. (http://darwin.cwru.edu/) [68]. We used the “W4” option [69] for weighting pairwise phenotypic differences between siblings. SIBPAL determines evidence for linkage at each SNP from regression of the phenotype difference between siblings on the estimated proportion of marker alleles shared identical-by-descent between siblings; the result is reported as a t value with corresponding significance. Point-wise significance was converted to genome-wide significance for multiple testing of markers by use of the expression of Lander and Kruglyak (as implemented at http://www.imbs-luebeck.de/8859-15/software/silclod.html) [19]. In SIBPAL linkage analysis, for each family phenotypic data of the children were used, and those for the grandparents and parents were not used.

Family-based association analysis with SNPs near and within the target genes or candidate regulators was carried out using QTDT [21],[70]. We tested about nine genes (median) per (trans) linkage peak. Within a gene, the SNP markers are often in strong linkage disequilibrium and thus are not independent tests. We report nominal p values for the QTDT results. In the linkage analysis, we used only data from children in the CEPH families; however, in the QTDT analysis, we used data from all members of the CEPH families. For the QTDT, we used the orthogonal (ao) model [21] and variance component options (wega).

We carried out population association analysis to follow-up results of QTDT. For these studies, expression phenotypes from 86 unrelated parents in the 45 CEPH families, as dependent variables, were regressed on SNP genotypes (coded 0, 1, 2). Conventional analysis of linear regression was carried out; we tested SNPs within a gene that showed significant QTDT for each phenotype. To minimize multiple testing, for each significant trans-linkage peak, we tested only the gene where the most significant QTDT result was found; SNPs within these genes are mostly highly correlated so we did not consider them as independent tests. We reported the nominal p values for these tests.

RNA-Seq

mRNA-Seq was performed as recommended by the manufacturer (Illumina). Briefly, immortalized B-cells from 41 unrelated CEPH grandparents (part of the International HapMap Project and the 45 families in this study) were grown and processed for RNA-Seq; hence these are biological replicates of those used in our microarray-based analysis. Poly(A) mRNA was extracted using Dynal oligo(dT) beads, fragmented, and first strand cDNA generated using random hexamers. Following second strand cDNA synthesis, end repair, and addition of a single A base, Illumina adaptors were ligated onto the samples. Then, ~200 bp fractions of the cDNA samples were isolated from agarose gels and PCR amplified. The qualities of the PCR amplicons were checked using the Agilent Bioanalyzer. The samples were then sequenced using the Illumina Genome Analyzer. We obtained an average of 41 million 50 bp reads per sample (median = 40 million).

For alignment of the short reads sequences to the human reference sequence (hg18) and identification of SNPs, we used the program MAQ (version 0.6.8) [25]. To minimize sequence errors, we used the first 40 of the 50 nucleotides in each sequence read for our analysis. For the alignment, we used the default settings of MAQ: allowing up to two mismatches per read. From the aligned reads with map quality scores of 30 or higher, we identified SNPs. For this analysis, we used only known SNPs in dbSNP Build 129. For a sample to be heterozygous at a SNP for our DAE analysis, we required that each allele be represented in at least 5% of the total reads covering that locus. To determine the expression level of a gene, we calculated RPKM [26]. Among our data, ~700 genes with average RKPM >1 were “called” absent on microarrays. If we had relied on microarray to identify “expressed” genes, these genes and the genes that were not represented on the microarrays would have been excluded in our analyses.

To check the accuracy of our RNA-Seq results, we compared the expression levels with those from our microarray and genotypes from our sequencing data with those obtained by HapMap Consortium. For each gene, we calculated correlation coefficient of the expression levels between the two platforms across the 41 samples. The average correlation coefficient was 0.76 (median = 0.76; range = 0.73 to 0.80). For each sample, we also identified the homozygous genotypes (AA, CC, GG, TT; ~25,000 genotypes per sample) using the HapMap database and compared them to genotypes in our sequencing results. The comparisons showed a high degree of agreement. Across the 41 samples, the average concordance rate is 98.6% (median = 98.7%).

Network Analysis

The gene regulatory network was constructed based on pairwise regulatory relationship identified through linkage (t>4) and QTDT analyses (p<0.05). Connections (edges) were placed between genes that were implicated in a regulator-target interaction. Properties of the resulting gene regulatory network were analyzed in MATLAB (MathWorks) by representing regulatory relationships as an asymmetric adjacency matrix. The number of incoming and outgoing connections per gene was determined by summing the columns and rows of the adjacency matrix. A MATLAB function for determining the scale-free topology criteria was implemented as previously described [48]. Code will be provided upon request. Figures of the resulting networks were drawn using Cytoscape 2.6.0 [71]. To identify genes that have been implicated as human disease susceptibility genes, we queried the Catalogue of Genome-Wide Association Studies (http://www.genome.gov/26525384) [72].

Functional Pathway Analysis

To determine whether a regulator and its target belong to the same functional groups, we examined Gene Ontology Biological Process terms [46] for the regulator and the target genes. We counted the number of regulator-target pairs with identical Gene Ontology Biological Process annotations; these were the “observed counts.” We then examined 1,000 randomly chosen gene pairs (from expressed genes in our B-cells) and counted the number of gene pairs that shared Gene Ontology Biological Process annotations. We repeated this 5 times and took the average; these were the “expected counts.” We compared the observed to the expected counts by a chi-square test.

Knockdown of Candidate Regulators

Immortalized B cells of four to six individuals and primary fibroblasts (foreskin) from two healthy newborns were used. The cells were transfected with Accell siRNAs (Thermo Scientific) against candidate regulators or non-target control according to the manufacturer's instructions. For each regulator, we used a pool of siRNAs that target the regulators in order to minimize off-target effects [36]. To compare the knockdown by pools of siRNA and single siRNAs against a gene, we silenced GPHN using a pool of siRNA and 2 siRNAs against different parts of the gene; similar results were obtained in the three experiments (see Table S3B).

For each transfection, immortalized B cells were seeded at a concentration of 4.5×105 cells per 750 ul on the day of transfection. 7.5 ul of 100 uM Accell siRNA was mixed with the seeded culture. Each transfection mix was then plated in a 96-well tissue culture plate in 150 ul aliquots. Similarly, 4.5×105 cells per 750 ul of primary fibroblasts were plated in 12-well plates in growth media the day before the transfection. On the day of transfection, the growth media were removed and replaced with 7.5 ul 100 um Accell siRNA (against genes of interest or scrambled sequence as control) and 750 ul of Accell media.

The transfected cells with siRNAs were incubated at 37°C for 96 h. We then replaced the Accell media with regular growth media (RPMI 1640 with 15% fetal bovine serum, 2 mM L-glutamine, and 100 U/mL penicillin-streptomycin) and let the cells recover for 24 h. RNA was extracted using Qiagen RNeasy kits. Effects of siRNA on gene expression were analyzed by quantitative PCR (Applied Biosystems). Expression of ACTB was used for normalization and changes in expression were calculated relative to cells transfected with non-target control siRNA. Sequences of PCR primers and siRNAs are presented in Tables S4 and S5.

Insulin Treatment

For the western analysis: primary fibroblasts were cultured in MEM medium supplemented with 10% fetal bovine serum, 2 mM L-glutamine, and 100 U/mL penicillin-streptomycin. Cells were serum starved for 18 h before treatment with 100 nM insulin for 5 min. Cells were lysed in 1× Lysis buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM Na2EDTA, 1 mM EGTA, 1% Triton ×100) (Cell Signaling) supplemented with 1× Complete protease inhibitors (Roche) and 1× phosphatase inhibitors I and II (Sigma). Cell lysates containing 150 µg of total protein were incubated with 5 µg of α-INSR antibody (#3025, Cell Signaling) or α-IGF1R antibody (#3018, Cell Signaling) at 4°C overnight. Immuno-complex was pulled down using Protein A Sepharose (GE Healthcare). Input and immunoprecipitation samples were analyzed by Western Blot using α-phosphotyrosine (1[ratio]1000) (4G10 Platinum, Millipore) or the above α-INSR (1[ratio]1000) and α-IGF1R (1[ratio]1000) antibodies.

For the gene expression analysis, primary fibroblasts (from foreskin) of four individuals were cultured as above. Cells were serum starved overnight (18 h) before being treated with 100 nM insulin (Sigma) for 0, 1, 2, 6, or 12 h. RNA was extracted and gene expression was analyzed by quantitative PCR (Applied Biosystems). Sequences of PCR primers are presented in Table S6.

Comparison to Hi-C Data

Hi-C data from Dekker and colleagues [39] were obtained from NCBI GEO (GSE189199); we used the data in their alignment summaries. We compared their list of interacting pairs to our regulator–target gene pairs. A match is called when one of their interacting pair coordinates was found within a regulator or 5 kb up- or downstream and the matching member of that pair is found within or 5 kb up- or downstream of the corresponding target gene. Seventy-five such pairs were found.

The experimental steps in this study are summarized by a flowchart (Figure S3).

Accession Numbers

The data have been deposited to NCBI GEO under the accession numbers GSE16778 and GSE16921 for the microarray and RNA-Seq data, respectively.

Supporting Information

Figure S1

Differential allelic expression by RNA-Seq. Allelic expression ratio of 273 exonic SNPs in 43 genes. Data for each heterozygous individual is represented as a color dot. SNPs are ordered left to right by mean expression ratio, a/(a+b).

(5.83 MB TIF)

Figure S2

Examples of subnetworks that include genes that were implicated as disease susceptibility in genome-wide association (GWA) studies (http://www.genome.gov/gwastudies/).

(1.65 MB PNG)

Figure S3

Flowchart showing experimental steps.

(0.08 MB PDF)

Table S1

Linkage, QTDT, and association results for cis-regulated genes.

(0.07 MB PDF)

Table S2

Regulators for 200 trans-regulated expression phenotypes.

(0.06 MB PDF)

Table S3

Additional results of knockdown of trans-regulators. (A) Knockdown of trans-regulators in fibroblasts. (B) GPHN knockdown using a pool of 4 siRNAs compared to individual siRNAs.

(0.10 MB PDF)

Table S4

Primer sequences for qRT-PCR (gene knockdown experiment).

(0.04 MB PDF)

Table S5

siRNA sequences (gene knockdown experiment).

(0.05 MB PDF)

Table S6

Primer sequences for qRT-PCR (insulin treatment).

(0.03 MB DOC)

Table S7

Regulator–target gene pairs that are found to interact physically by Hi-C.

(0.05 MB PDF)

Acknowledgments

We thank Drs. Warren Ewens and Kathryn Gogolin-Ewens for support, suggestions, and discussions; Dr. Michael Kearns for advice on network analysis; Dr. Gary Schroth and his team for advice and technical help on RNA-Seq; Dr. Alan Bruzel and Mr. Krzysztof Halasa for technical help; and Ms. Colleen McGarry for manuscript preparation.

Abbreviations

3C
chromosome conformation capture
ALS
amyotrophic lateral sclerosis
CEPH
Centre d'Etude du Polymorphisme Humain
DAE
differential allelic expression
FDR
false discovery rate
Gb
gigabase
GOGE
genetics of gene expression
LOD
logarithm of odds
QTDT
quantitative transmission disequilibrium test
RNASeq
RNA-sequencing, siRNA, short interfering RNA
SNP
single nucleotide polymorphism

Footnotes

The authors have declared that no competing interests exist.

IXW is a research associate and VGC is an investigator of the Howard Hughes Medical Institute. RRN is supported by a training grant from the National Institutes of Health (T32 GM008216). Some analyses for this paper were carried out by using the program package S.A.G.E. which is supported by a grant from the National Center for Research Resources. This work is supported by the National Institutes of Health and the Howard Hughes Medical Institute. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

1. Brem R. B, Yvert G, Clinton R, Kruglyak L. Genetic dissection of transcriptional regulation in budding yeast. Science. 2002;296:752–755. [PubMed]
2. Yvert G, Brem R. B, Whittle J, Akey J. M, Foss E, et al. Trans-acting regulatory variation in Saccharomyces cerevisiae and the role of transcription factors. Nat Genet. 2003;3:57–64. [PubMed]
3. Morley M, Molony C. M, Weber T. M, Devlin J. L, Ewens K. G, et al. Genetic analysis of genome-wide variation in human gene expression. Nature. 2004;430:743–747. [PMC free article] [PubMed]
4. Emilsson V, Thorleifsson G, Zhang B, Leonardson A. S, Zink F, et al. Genetics of gene expression and its effect on disease. Nature. 2008;452:423–428. [PubMed]
5. Stranger B. E, Forrest M. S, Clark A. G, Minichiello M. J, Deutsch S, et al. Genome-wide associations of gene expression variation in humans. PLoS Genet. 2005;1:e78. doi: 10.1371/journal.pgen.0010078. [PMC free article] [PubMed]
6. Cheung V. G, Conlin L. K, Weber T. M, Arcaro M, Jen K-Y, et al. Natural variation in human gene expression assessed in lymphoblastoid cells. Nat Genet. 2003;33:422–425. [PubMed]
7. Goring H. H, Curran J. E, Johnson M. P, Dyer T. D, Charlesworth J, et al. Discovery of expression QTLs using large-scale transcriptional profiling in human lymphocytes. Nat Genet. 2007;39:1208–1216. [PubMed]
8. Cheung V. G, Spielman R. S, Ewens K. G, Weber T. M, Morley M, et al. Mapping determinants of human gene expression by regional and genome-wide association. Nature. 2005;437:1365–1369. [PMC free article] [PubMed]
9. Myers A. J, Gibbs J. R, Webster J. A, Rohrer K, Zhao A, et al. A survey of genetic human cortical gene expression. Nat Genet. 2007;39:1494–1499. [PubMed]
10. Hubner N, Wallace C. A, Zimdahl H, Petretto E, Schulz H, et al. Integrated transcriptional profiling and linkage analysis for identification of genes underlying disease. Nat Genet. 2005;37:243–253. [PubMed]
11. Lemos B, Araripe L. O, Fontanillas P, Hartl D. L. Dominance and the evolutionary accumulation of cis- and trans-effects on gene expression. Proc Natl Acad Sci U S A. 2008;105:14471–14476. [PMC free article] [PubMed]
12. Wang H. Y, Fu Y, McPeek M. S, Lu X, Nuzhdin S, et al. Complex genetic interactions underlying expression differences between Drosophila races: analysis of chromosome substitutions. Proc Natl Acad Sci U S A. 2008;105:6362–6367. [PMC free article] [PubMed]
13. Gilad Y, Rifkin S. A, Pritchard J. K. Revealing the architecture of gene regulation: the promise of eQTL studies. Trends Genet. 2008;24:408–415. [PMC free article] [PubMed]
14. Cheung V. G, Spielman R. S. Genetics of human gene expression: mapping DNA variants that influence gene expression. Nat Rev Genet. 2009;10:595–604. [PMC free article] [PubMed]
15. Moffatt M. F, Kabesch M, Liang L, Dixon A. L, Strachan D, et al. Genetic variants regulating ORMDL3 expression contribute to the risk of childhood asthma. Nature. 2007;448:470–473. [PubMed]
16. Pickrell J. K, Marioni J. C, Pai A. A, Degner J. F, Engelhardt B. E, et al. Understanding mechanisms underlying human gene expression variation with RNA sequencing. Nature. 2010;464:768–772. [PMC free article] [PubMed]
17. Montgomery S. B, Sammeth M, Gutierrez-Arcelus M, Lach R. P, Ingle C, et al. Transcriptome genetics using second generation sequencing in a Caucasian population. Nature. 2010;464:773–777. [PMC free article] [PubMed]
18. Dausset J, Cann H, Cohen D, Lathrop M, Lalouel J. M, et al. Centre d'etude du polymorphisme humain (CEPH): collaborative genetic mapping of the human genome. Genomics. 1990;6:575–577. [PubMed]
19. Lander E, Kruglyak L. Genetic dissection of complex traits: guidelines for interpreting and reporting linkage results. Nat Genet. 1995;11:241–247. [PubMed]
20. Rockman M. V, Kruglyak L. Genetics of global gene expression. Nat Rev Genet. 2006;7:862–872. [PubMed]
21. Abecasis G. R, Cardon L. R, Cookson W. O. A general test of association for quantitative traits in nuclear families. Am J Hum Genet. 2000;66:279–292. [PMC free article] [PubMed]
22. International HapMap Consortium. The International HapMap Project. Nature. 2003;426:789–796. [PubMed]
23. International HapMap Consortium. A haplotype map of the human genome. Nature. 2005;437:1299–1320. [PMC free article] [PubMed]
24. Bentley D. R, Balasubramanian S, Swerdlow H. P, Smith G. P, Milton J, et al. Accurate whole human genome sequencing using reversible terminator chemistry. Nature. 2008;456:53–59. [PMC free article] [PubMed]
25. Li H, Ruan J, Durbin R. Mapping short DNA sequencing reads and calling variants using mapping quality scores. Genome Res. 2008;18:1851–1858. [PMC free article] [PubMed]
26. Mortazavi A, Williams B. A, McCue K, Schaeffer L, Wold B. Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat Methods. 2008;5:621–628. [PubMed]
27. Kaushal S, Schneider J. W, Nadal-Ginard B, Mahdavi V. Activation of the myogenic lineage by MEF2A, a factor that induces and cooperates with MyoD. Science. 1994;266:1236–1240. [PubMed]
28. Chinkers M. Targeting of a distinctive protein-serine phosphatase to the protein kinase-like domain of the atrial natriuretic peptide receptor. Proc Natl Acad Sci U S A. 1994;91:11075–11079. [PMC free article] [PubMed]
29. Wandinger S. K, Suhre M. H, Wegele H, Buchner J. The phosphatase Ppt1 is a dedicated regulator of the molecular chaperone Hsp90. Embo J. 2006;25:367–376. [PMC free article] [PubMed]
30. Chen H, Sharp B. M. Content-rich biological network constructed by mining PubMed abstracts. BMC Bioinformatics. 2004;5:147. [PMC free article] [PubMed]
31. Zhang K, Li J. B, Gao Y, Egli D, Xie B, et al. Digital RNA allelotyping reveals tissue-specific and allele-specific gene expression in human. Nat Methods. 2009;6:613–618. [PMC free article] [PubMed]
32. Wang X, Sun Q, McGrath S. D, Mardis E. R, Soloway P. D, et al. Transcriptome-wide identification of novel imprinted genes in neonatal mouse brain. PLoS One. 2008;3:e3839. doi: 10.1371/journal.pone.0003839. [PMC free article] [PubMed]
33. Degner J. F, Marioni J. C, Pai A. A, Pickrell J. K, Nkadori E, et al. Effect of read-mapping biases on detecting allele-specific expression from RNA-sequencing data. Bioinformatics. 2009;25:3207–3212. [PMC free article] [PubMed]
34. Price A. L, Patterson N, Hancks D. C, Myers S, Reich D, et al. Effects of cis and trans genetic ancestry on gene expression in African Americans. PLoS Genet. 2008;4:e1000294. doi: 10.1371/journal.pgen.1000294. [PMC free article] [PubMed]
35. Ge B, Pokholok D. K, Kwan T, Grundberg E, Morcos L, et al. Global patterns of cis variation in human cells revealed by high-density allelic expression analysis. Nat Genet. 2009;41:1216–1222. [PubMed]
36. Myers A. J, Chi J. T, Gong D, Schaner M. E, Brown P. O, et al. Minimizing off-target effects by using diced siRNAs for RNA interference. J RNAi and Gene Silencing. 2006;2:181–194. [PMC free article] [PubMed]
37. Laustsen P. G, Lane W. S, Bennett V, Lienhard G. E. Association of protein kinase C(lambda) with adducin in 3T3-L1 adipocytes. Biochim Biophys Acta. 2001;1539:163–172. [PubMed]
38. Gunton J. E, Kulkarni R. N, Yim S, Okada T, Hawthorne W. J, et al. Loss of ARNT/HIF1beta mediates altered gene expression and pancreatic-islet dysfunction in human type 2 diabetes. Cell. 2005;122:337–349. [PubMed]
39. Lieberman-Aiden E, van Berkum N. L, Williams L, Imakaev M, Ragoczy T, et al. Comprehensive mapping of long-range interactions reveals folding principles of the human genome. Science. 2009;326:289–293. [PMC free article] [PubMed]
40. Jackson D. A, Hassan A. B, Errington R. J, Cook P. R. Visualization of focal sites of transcription within human nuclei. Embo J. 1993;12:1059–1065. [PMC free article] [PubMed]
41. Fraser P, Bickmore W. Nuclear organization of the genome and the potential for gene regulation. Nature. 2007;447:413–417. [PubMed]
42. Schoenfelder S, Sexton T, Chakalova L, Cope N. F, Horton A, et al. Preferential associations between co-regulated genes reveal a transcriptional interactome in erythroid cells. Nat Genet. 2010;42:53–61. [PMC free article] [PubMed]
43. Garbarino J. E, Gibbons I. R. Expression and genomic analysis of midasin, a novel and highly conserved AAA protein distantly related to dynein. BMC Genomics. 2002;3:18. [PMC free article] [PubMed]
44. Presley J. F, Cole N. B, Schroer T. A, Hirschberg K, Zaal K. J, et al. ER-to-Golgi transport visualized in living cells. Nature. 1997;389:81–85. [PubMed]
45. Hartmann E, Gorlich D, Kostka S, Otto A, Kraft R, et al. A tetrameric complex of membrane proteins in the endoplasmic reticulum. Eur J Biochem. 1993;214:375–381. [PubMed]
46. Ashburner M, Ball C. A, Blake J. A, Botstein D, Butler H, et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet. 2000;25:25–29. [PMC free article] [PubMed]
47. Lee S. I, Dudley A. M, Drubin D, Silver P. A, Krogan N. J, et al. Learning a prior on regulatory potential from eQTL data. PLoS Genet. 2009;5:e1000358. doi: 10.1371/journal.pgen.1000358. [PMC free article] [PubMed]
48. Zhang B, Horvath S. A general framework for weighted gene co-expression network analysis. Stat Appl Genet Mol Biol. 2005;4:Article17. [PubMed]
49. Cho S, Park S. M, Kim T. D, Kim J. H, Kim K. T, et al. BiP internal ribosomal entry site activity is controlled by heat-induced interaction of NSAP1. Mol Cell Biol. 2007;27:368–383. [PMC free article] [PubMed]
50. Cussac D, Leblanc P, L'Heritier A, Bertoglio J, Lang P, et al. Rho proteins are localized with different membrane compartments involved in vesicular trafficking in anterior pituitary cells. Mol Cell Endocrinol. 1996;119:195–206. [PubMed]
51. Baranzini S. E, Wang J, Gibson R. A, Galwey N, Naegelin Y, et al. Genome-wide association analysis of susceptibility and clinical phenotype in multiple sclerosis. Hum Mol Genet. 2009;18:767–778. [PubMed]
52. Todd D. J, Lee A. H, Glimcher L. H. The endoplasmic reticulum stress response in immunity and autoimmunity. Nat Rev Immunol. 2008;8:663–674. [PubMed]
53. Lin W, Popko B. Endoplasmic reticulum stress in disorders of myelinating cells. Nat Neurosci. 2009;12:379–385. [PMC free article] [PubMed]
54. Tirosh I, Reikhav S, Levy A. A, Barkai N. A yeast hybrid provides insight into the evolution of gene expression regulation. Science. 2009;324:659–662. [PubMed]
55. Kudla G, Murray A. W, Tollervey D, Plotkin J. B. Coding-sequence determinants of gene expression in Escherichia coli. Science. 2009;324:255–258. [PMC free article] [PubMed]
56. Nackley A. G, Shabalina S. A, Tchivileva I. E, Satterfield K, Korchynskyi O, et al. Human catechol-O-methyltransferase haplotypes modulate protein expression by altering mRNA secondary structure. Science. 2006;314:1930–1933. [PubMed]
57. Kimchi-Sarfaty C, Oh J. M, Kim I. W, Sauna Z. E, Calcagno A. M, et al. A “silent” polymorphism in the MDR1 gene changes substrate specificity. Science. 2007;315:525–528. [PubMed]
58. Allali-Hassani A, Wasney G. A, Chau I, Hong B. S, Senisterra G, et al. A survey of proteins encoded by non-synonymous single nucleotide polymorphisms reveals a significant fraction with altered stability and activity. Biochem J. 2009;424:15–26. [PubMed]
59. Schadt E. E, Molony C, Chudin E, Hao K, Yang X, et al. Mapping the genetic architecture of gene expression in human liver. PLoS Biol. 2008;6:e107. doi: 10.1371/journal.pbio.0060107. [PMC free article] [PubMed]
60. Dimas A. S, Deutsch S, Stranger B. E, Montgomery S. B, Borel C, et al. Common regulatory variation impacts gene expression in a cell type-dependent manner. Science. 2009;325:1246–1250. [PMC free article] [PubMed]
61. Berridge M. J, Irvine R. F. Inositol phosphates and cell signalling. Nature. 1989;341:197–205. [PubMed]
62. van Es M. A, Van Vught P. W, Blauw H. M, Franke L, Saris C. G, et al. ITPR2 as a susceptibility gene in sporadic amyotrophic lateral sclerosis: a genome-wide association study. Lancet Neurol. 2007;6:869–877. [PubMed]
63. Kunst C. B, Mezey E, Brownstein M. J, Patterson D. Mutations in SOD1 associated with amyotrophic lateral sclerosis cause novel protein interactions. Nat Genet. 1997;15:91–94. [PubMed]
64. Yang Y, Hentati A, Deng H. X, Dabbagh O, Sasaki T, et al. The gene encoding alsin, a protein with three guanine-nucleotide exchange factor domains, is mutated in a form of recessive amyotrophic lateral sclerosis. Nat Genet. 2001;29:160–165. [PubMed]
65. Nishitoh H, Kadowaki H, Nagai A, Maruyama T, Yokota T, et al. ALS-linked mutant SOD1 induces ER stress- and ASK1-dependent motor neuron death by targeting Derlin-1. Genes Dev. 2008;22:1451–1464. [PMC free article] [PubMed]
66. Abecasis G. R, Cherny S. S, Cookson W. O, Cardon L. R. Merlin-rapid analysis of dense genetic maps using sparse gene flow trees. Nat Genet. 2002;30:97–101. [PubMed]
67. Burdick J. T, Chen W. M, Abecasis G. R, Cheung V. G. In silico method for inferring genotypes in pedigrees. Nat Genet. 2006;38:1002–1004. [PMC free article] [PubMed]
68. Haseman J. K, Elston R. C. The investigation of linkage between a quantitative trait and a marker locus. Behav Genet. 1972;2:3–19. [PubMed]
69. Shete S, Jacobs K. B, Elston R. C. Adding further power to the Haseman and Elston method for detecting linkage in larger sibships: weighting sums and differences. Hum Hered. 2003;55:79–85. [PubMed]
70. Abecasis G. R, Cookson W. O, Cardon L. R. Pedigree tests of transmission disequilibrium. European Journal of Human Genetics. 2000;8:545–551. [PubMed]
71. Shannon P, Markiel A, Ozier O, Baliga N. S, Wang J. T, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003;13:2498–2504. [PMC free article] [PubMed]
72. Hindorff L. A, Sethupathy P, Junkins H. A, Ramos E. M, Mehta J. P, et al. Potential etiologic and functional implications of genome-wide association loci for human diseases and traits. Proc Natl Acad Sci U S A. 2009;106:9362–9367. [PMC free article] [PubMed]

Articles from PLoS Biology are provided here courtesy of Public Library of Science
PubReader format: click here to try

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...