þÿ

Abstract

Genetic variation is known to influence the amount of mRNA produced by a

gene. Given that the molecular machines control mRNA levels of multiple genes,

we expect genetic variation in the components of these machines would influence

multiple genes in a similar fashion. In this study we show that this assumption is

correct by using correlation of mRNA levels measured independently in the brain,

kidney or liver of multiple, genetically typed, mice strains to detect shared genetic

influences. These correlating groups of genes (CGG) have collective properties that

account for 4090% of the variability of their constituent genes and in some cases,

but not all, contain genes encoding functionally related proteins. Critically, we show

that the genetic influences are essentially tissue specific and consequently the same

genetic variations in the one animal may upregulate a CGG in one tissue but down

regulate the same CGG in a second tissue. We further show similarly paradoxical

behaviour of CGGs within the same tissues of different individuals. The implication

of this study is that this class of genetic variation can result in complex inter and

intraindividual and tissue differences and that this will create substantial challenges

to the investigation of phenotypic outcomes, particularly in humans where multiple

tissues are not readily available.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

Introduction

Gene expression is controlled by multiple molecular machines whose interaction with

a gene and genes transcript contributes to determining a final level of mRNA; recent

studies have shown that these processes are subject to significant influences of genetic

variation that result in heritable changes to final mRNA levels (reviewed by Cotsapas

et al., 2006; Gibson and Weir, 2005; Rockman and Kruglyak, 2006; Williams et al., 2007).

In multicellular organisms, these molecular machines are involved in setting mRNA lev-

els of many genes but they contain different components; some may be common to the

expression of all genes in all cells of the organism whereas other components may have

more limited function such that that are involved with sub sets of genes or sub sets

of cell types, or both (Maniatis and Reed, 2002; Tsankov et al., 2006; Maciag et al.,

2006; Komili and Silver, 2008). We therefore predict that genetic perturbation of these

machines will result in either global or celltype specific changes to gene expression,

depending on the variant component. Such behaviour is in marked contrast to genetic

variation in protein coding sequence, where the variant is observed in all cases where

the gene is expressed. In this work, we use correlation based methods to show that

the effects of regulatory variation are, as predicted, coordinated changes to the mRNA

levels of groups of genes. These group changes can be very different both in multiple

tissues of the same individual, as well as being different in the same tissues of multiple

individuals. We use the term regulatory variation to describe any genetic variation that

affects the amount of mRNA produced from a gene; it can occur through the disrup-

tion of cisregulatory sequences, such as promoter or enhancer elements, or through

changes to transacting components, including any of the molecular machinery that

controls the amount of steady state mRNA in a cell, such as transcription or splicing

factors (Williams et al., 2007). The majority of findings to date, using predominantly

expression QTL (eQTL) experimental designs, suggest that cisacting regulatory varia-

tion appears to be of larger effect size, and is thus more easily detected; in comparison,

transregulatory variation appears to be of smaller effect size, and are either less com-

mon, or harder to detect (Petretto et al., 2006; Stranger et al., 2005; Goring et al., 2007).

When transacting influences are identified, these tend to be a small number of eQTLs

that influence the expression of large numbers of genes, so called "masterregulator" of

gene expression, suggesting that regulatory variation is affecting the expression level of

groups of genes, simultaneously. Investigating transacting regulatory variation using

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

eQTL analysis is presently beset both by very substantial statistical problems of multiple

hypothesis testing and by the sheer scale of studies required to provide genetic power

to detect small effect sizes. Further, whilst eQTL analysis is an appropriate approach

to investigate the effects caused by one or a small number of genetic influences, it has

limited power to detect additional eQTLs with smaller effect sizes (Brem and Kruglyak,

2005; Williams et al., 2007). To overcome this limitation, several groups have used

correlationbased approaches to identify groups of genes that covary under the influ-

ence of simple or complex genetic influences (Ghazalpour et al., 2006; Lan et al., 2006).

The conceptual basis of such experiments is simple: mRNA levels that vary similarly

across multiple individuals are likely to do so because of shared sensitivity to genetic

influences. Correlationbased approaches have the added advantage that we are mea-

suring the shared outcome of regulatory variation stemming from multiple genetic loci,

the modest contributions of which eQTL analysis would be underpowered to detect in

all but the largest studies.

In this study of inbred and recombinant inbred mice, we set out to investigate trans

acting regulatory variation, using correlation analysis to identify groups of genes that

are likely to be influenced by shared regulatory variation, and thus shared regulatory

factors. We further investigate the consequence of transacting regulatory variation in

three different mouse tissues to assess the degree to which (A) genes are affected by the

same regulatory variation in all tissues, and (B) whether the outcome of such regulatory

variation is the same in all tissues.

Overview of experimental design

To identify genes whose expression levels may be affected by regulatory variation, and

to investigate their regulation in multiple tissues, we adopt the following experimental

design: first, we compare gene expression levels in 3 tissues of two inbred mouse strains,

C57BL/6J and DBA/2J, and 31 strains of the BXD recombinant inbred (RI) panel

derived from these two progenitors. Next, we look for genes whose expression differs

between the progenitor strains in at least one of these tissues; within these we identify

subsets of genes whose mRNA levels vary coordinately across the BXD RI strains and

the three tissues; we term these "correlating groups of genes" or CGGs. We validate the

shared regulatory influences acting upon these CGGs by testing the conservation of their

expression changes in both the parental strains and in the distantly related inbred strain

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

SJL/J. We further investigate the specific outcomes of the regulation of these CGGs in

the three tissues of the BXD panel.

Identify genetically influenced genes

We began by identifying genes differentially expressed in at least one of whole brain,

kidney, or liver between strains C57BL/6J and DBA/2J. We found that we could reliably

detect 6075 transcripts above background in all three tissues, of which 755 were variantly

expressed between the two strains at a LOD>3 (the Bstatistic of Lonnstedt and Speed

(2002), as modified by Smyth (2004); see Methods). We ascribe this consistent variation

in gene expression to regulatory variation, since environmental factors have been reduced

to a minimum. We stress that we have deliberately avoided selecting genes that are

expressed in a "tissue specific" manner, in the sense of being expressed in only 1 or 2 of

the 3 tissues (Supplementary Figure 1).

The identification of 755 genes as being potential targets of regulatory variation(s)

does not us allow us to identify if each gene is under a unique or shared influence. To

do this, we need to study the 755 genes in multiple, changing, genetic backgrounds

reasoning that we could then detect shared influence by detecting highly correlated

alterations mRNA levels of otherwise unrelated genes. Such correlated changes could in

principle be observed between genes within either single or multiple tissues. We chose

to search for mRNA correlations across multiple tissues in the first instance and then

further studied the behaviour in the individual tissues seeking to ask if the outcome of

genetic influence on genes is the same in each tissue.

Identifying groups of genes with similar expression patterns

in multiple tissues

To achieve this, we measured mRNA levels of the 755 genes in the same 3 tissues in 31

BXD recombinant inbred (RI) strains (Taylor et al., 1999), pooling three age and sex

matched mice from each. These strains have been derived from crosses of C57BL/6J and

DBA/2J, which have been bred to homozygosity by repeated sibling pair mating. As

they carry arbitrary mixtures of the two progenitor backgrounds, but are homozygous

at each locus, we predict that most strains will have inherited some of the C57BL/6J

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

alleles and some of the DBA/2J alleles, of any factors, basal or conditional, controlling

the mRNA levels of the 755 genes. If these factors influence more than a single transcript,

we would predict that the levels of these coinfluenced mRNAs would correlate across

the BXD panel, thus forming a CGG.

In order to identify those genes that have similar expression patterns in all 31 BXD

strains and in all 3 tissues, we adopted a correlationnetwork approach (see Methods).

We compare all pairwise combinations of 755 gene expression patterns across the 93

measurements, and construct correlation networks consisting of nodes representing genes,

and edges representing correlations that are stronger than an empirically determined

threshold.

The correlationnetwork approach has a number of advantages: the resulting net-

works summarise a large amount of complex data in a form that is easily visualised and

interpreted, and there are a number of techniques for identifying discrete regions of the

network corresponding to CGGs. Most importantly, the number of groups does not need

to be known a priori as in clustering methods, and those genes that are not correlated

highly enough are automatically filtered from the resulting network, thus reducing the

noise in the system.

Choice of threshold for network construction and network

properties

To display the complex relationship between genes and multiple pairwise correlations

we construct networks using the widely used approach of thresholding correlation ma-

trices (Freeman et al., 2007; Voy et al., 2006). The intention of thresholding is to define

discrete groups of genes that can be subject to other analyses but we stress that there

is no plausible reason why a threshold should have an explicit biological meaning, as

regulationinduced correlation can be of any magnitude. An important step in con-

structing such networks is choice of threshold: too low a threshold will result in a too

densely connected graph, while too high a threshold will result in a sparsely populated

and connected network (Freeman et al., 2007). The final choice of threshold is guided

entirely by the overall objectives of the analysis. Our primary aim is to identify groups of

coregulated genes that are plausibly under common genetic control, so we focus on find-

ing groups of interconnected genes that are distinct from other such groups (connected

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

components in graph theoretical terms, unconnected to others).

We focused on identifying a correlation threshold that would (A) provide an adequate

number of connected components that had at least 2 genes; (B) distinguish the graph

theoretic properties of the 755 genes from those of all expressed genes. To do so, we

studied various network properties of these genes, treating them as teststatistics, and

examined how unusual these properties were using 1000 sets of the same size randomly

sampled from the 6075 genes that were expressed in all three tissues. We studied 6 graph

theoretic properties in this fashion (Figure 1), namely: (1) the number of correlations

above the threshold; (2) the number of connected components; (3) the median of the

distribution of connected component size; (4) size of the largest connected component,

(5) average degree, computed across nonsingleton connected components and (6) the

global clustering coefficient (for a review of these concepts Sharan et al., 2007).

At low correlation thresholds, we found that both the set of 755 genes ("755 net-

work") and randomly resampled sets of genes ("random networks") formed networks

characterised by a single, large connected component. This expected structure starts to

break down into multiple connected components at || > 0.50 in the 755network, and

at || > 0.35 in random networks, with the former having consistently higher number of

connected genes (Figure 1A). The number of connected components was also on average

higher in the 755 network than in the random networks, but only at || = 0.85 was this

greater than for all random networks. The largest number of connected components for

both the 755 and random networks was observed at a threshold of || = 0.75, after which

some of these structures become completely unconnected and disappear (Figure 1B).

The median number of genes in connected components showed little difference between

the 755network and random ones (Figure 1C), but the size of the largest connected

component in the 755 network was consistently above that observed for random net-

works (Figure 1D) across a wide threshold range. The average number of connections

of genes in any component (their degree) of the 755network was consistently higher

than for random networks (Figure 1E), suggesting tighter overall correlation. The ex-

tent of connectivity within the component to which a gene belongs (measured by the

clustering coefficient) was also higher for the 755 network at 0.55 || 0.75; however,

this measure becomes erratic above 0.775 due to the reduced size of the network at these

stringent thresholds (Figure 1F).

In summary, the observed 755network generated consistently higher number of cor-

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

relations across a range of thresholds, generated a higher average level of connectivity

between genes and a greater level of interconnectivity between the neighbours of a

given gene, than was observable in random networks of the same size derived from all

expressed genes, suggesting that these genes are indeed responding to the influences of

regulatory variation. We settled on a threshold of || = 0.775, which gave us maximal

differences between the 755network and background without dissolving structure due

to high stringency.

The cross tissue correlation network

We constructed a || = 0.775 correlation network containing 212 (28.1%) genes that

correlate with at least one other transcript; the genes have a median degree of 4, with

73% of genes with a degree of 2 (Figure 2). These genes are central to our subsequent

study; in principle they are influenced by genetic variation(s) that influence mRNA levels

in all three tissues simply because the correlation statistic is calculated across all three

tissues. Performing similar analyses on subsets of tissues, we find that at the same

threshold a further 204 (27.0%) genes are correlated in any pair of tissues, and a further

191 (25.3%) are correlated in any single tissue. A total of 607 (80.4%) of the 755 genes

exhibit correlated behaviour in any network, suggesting that shared regulatory influences

upon gene expression are widespread, and over 55% are correlated in multiple tissues

(data not presented).

In the original network across all tissues, we find that the 212 genes fall into 19

discrete correlating groups of genes or CGGs; of these groups, 10 contain at least three

members and the largest 5 contain 75, 63, 21, 12 and 6 genes respectively; all 19 CGGs

are displayed in Figure 2 along with their expression patterns across the 31 BXD lines

and the 3 tissues.

Given that CGGs are constructed from combinations of pairwise relationships be-

tween genes, we expected that the levels of each transcript within a CGG (grey lines in

Figure 2) should in general be similar. To assess the extent to which variation in a sin-

gle genes mRNA could be explained by the shared influences upon a CGG, we correlated

(using the coefficient of determination, R

, see Methods) the expression pattern of each

transcript to the centroid of their respective CGG (thick coloured lines in Figure 2)

for each tissue individually, and across all 3 tissues simultaneously; we determined the

statistical significance of the observed R

via permutation (see Methods). R

values

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

ranged from 0.4 to 0.9 (Figure 3) and in most cases there was very limited overlap

with randomly sampled genes. The permutation analysis shows that the behaviour of

the genes in the CGG cannot easily be explained by inter-array differences in hybridi-

sation; in this case we would expect the random permutations also to generate higher

values. We note that even though CGG 1 in the brain has the smallest R

values,

which overlap with randomly sampled genes, this CGG nevertheless exhibits significantly

unusual biological behaviour that independently supports the notion it is a CGG (see

below). We conclude from this analysis that shared behaviour is a significant influence

upon genes within CGGs and that the shared influence upon genes in CGG (4090%

from Figure 3) is comparable in magnitude to the size of effects reported for cisacting

eQTLs Hubner et al. (2005); Petretto et al. (2006); Stranger et al. (2005); West et al.

(2007).

While we have illustrated the congruous behaviour of mRNAs within a CGG, we also

note from Figure 2 that mRNA level profiles between each tissue are strikingly different.

This is supported by calculating the correlation between the intratissue centroids for

each CGG (Table 1): the only statistically significant relationship is in fact an anti

correlation between the centroids of CGG 2 in Brain and Liver ( = -0.59, P = 5.94 ×

-4

). These results show that whilst genes within a CGG are highly correlated to each

other, consistent with the idea of being influenced by shared factors, the outcome of

such regulation is markedly different in each tissue, such that the overall pattern of a

group's expression in each tissue is at best, uncorrelated, or even anti-correlated. These

differences are best explained by genetic variation in multiple regulatory components

that act individually in a tissue specific fashion or in a single cross tissue component

whose behaviour is itself modulated by tissue specific factors.

The collective behaviour of CGGs

We have identified CGGs based on their expression patterns across a panel of BXD mice,

and across three tissues. Within each individual BXD animal, all genes in a CGG should

be coordinately regulated, even if this differs across tissues. If these levels are indeed

due to genetic differences in the regulatory factors controlling ultimate mRNA level,

then we would expect that CGG members should display similar correlated expression

patterns across different genetic backgrounds. However, the multiple, complex changes in

genetic background implicit in this experiment are unlikely to result in exactly the same

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

mRNA levels in any two individuals; therefore, rather than test for identical expression

of all genes in the CGG, we designed a test for the identical direction of mRNA levels:

relatively up or downregulated. This co-ordinated expression over all genes in a CGG

can be summarised as a coherency score: the proportion of genes whose mRNA levels

are upregulated relative to the reference (see Figure 4a for an example, and Methods

for details).

We performed simulation studies to assess the performance of the coherency score

with respect to both the number of genes in a CGG, and the magnitude and variability

of the expression changes (see Supplementary Material). Simulating the conditions

of our experiment, we identified that the score is adequately powered to detect coherent

directionality of expression for CGGs of at least 10 genes (at permuted P < 0.05). Below

this group size, the score had little power even in the case of maximal coherency.

We applied this method to the 4 largest CGGs (those having between 75 and 12

genes). Given the CGG had been defined solely by analysis of the BXD RI strains,

we therefore looked at coherency in the two progenitor strains, C57BL/6J and DBA/2J

and found that all 4 CGGs were significantly coherent (P = 0.001) in at least one

tissue (Figure 4b and Supplementary Table 1). We note that CGG 1 in the brain,

which had the lowest R

values to its centroid, nevertheless exhibits high coherency

(coherency=0.76, P = 0.001); whilst the shared contribution to overall mRNA levels of

the CGG might be relatively small, there is a marked effect upon direction of mRNA level

changes. We also note that the 63 genes in CGG 2 have complex properties: coherency

is moderate in size, but still significant in Brain (coherency -0.52; P = 0.001) and

Kidney (coherency -0.46; P = 0.001) and not coherent in Liver (coherency -0.08;

P = 0.33). However, close inspection (Figure 2) reveals that this CGG comprises two

sub domains, one highly interconnected domain (CGG2A) containing 38 genes, which

are loosely connected to a less interconnected group of 25 genes (CGG2B). These two

sub-domains exhibit more coherent expression: CGG2A in Brain -0.63 (P = 0.001),

Kidney -0.79 (P = 0.001) and Liver -0.74 (P = 0.001) and CGG2B in Brain -0.36

(P = 0.013), in Kidney 0.04 (P = 0.15) and Liver 0.92 (P = 0.001). This illustrates the

complexity of the correlations within the network where the existence of CGGs defined

by correlation alone does not capture the full relationships of mRNA levels.

To validate these observations, we performed an independent comparison of a distinct

inbred mouse strain, SJL/J to C57BL/6J (see Methods).

Given the close genetic

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

relationship between DBA/2J and SJL/J (Beck et al., 2000), we expected that these

4 CGGs should (A) be coherent in each tissue, and (B) show similar directionality as

DBA/2J with respect to C57BL/6J. We found both predictions to be true, with all

CGGs being coherent in at least one tissue (P < 0.05; see Figure4b, row 2, and

Supplementary Table 2), and most CGGs that were coherent in both DBA/2J and

SJL/J having the same directionality in the same tissue (those entries with ** for both

strains in Figure 4c). These findings confirm that these groups of genes are indeed

collectively sensitive to genetic influence, even in this more distant inbred strain.

Interstrain and interindividual coherency

These observations provide independent biological confirmation of the properties of

CGGs, and also reveal the complex outcomes of genetic influence upon mRNA levels.

In these 3 inbred strains, mRNA levels of the same groups of genes coordinately vary

not just between strains but also between the same tissues between each strain. If this

behaviour is indeed genetic in origin, as we have argued, then we would also expect the

same to be true in the BXD lines. We therefore calculated coherency for the four largest

CGGs in the three tissues of each BXD line relative to C57L/6J and compared this to

the coherency of DBA/2J expression. Due to the small number of replicate microarrays

used for each measure (6 for DBA/2J, 3 for SJL/J and 1 for each BXD RI), we limit

our inferences of differential regulation to extreme cases, where a high score in DBA/2J

becomes of high magnitude but opposite sign in at least one BXD strain. We note that

a simple Mendelian effect would result in either a coherency score approximating +/ - 1

for a DBA/2J allele or 0 for a C57BL/6J allele, which is the reference strain. We use

permutation to assign significance to these events, assessing the likelihood that a given

coherency would occur by chance in our dataset (see Methods).

We find CGG 2 in kidney changes from -0.46 in DBA/2J to +1.00 in 2 BXD lines;

CGG3 in brain changes from +0.33 to -1.00 in 3 BXD lines, in kidney from +1.00 to

-1.00 in an BXD line, in liver from -1.00 to +1.00 in 7 BXD lines; CGG 4 in brain

from +0.83 to -1.00 in 4 BXD lines, in kidney from -0.83 to +1.00 in 6 BXD lines. The

pattern of changes in coherency of CGG 3 in kidney shown in Figure 5A is compatible

with segregation of variations within the BXD lines and the consistent upregulation of

mRNA levels in the kidney for most of the BXD lines for CGG 4 (Figure 5B) is sup-

portive of transgressive segregation of genetic influences. (see Supplementary Table 3

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

for full coherency scores from the BXD panel, DBA/2J and SJL/2J). Collectively, these

results suggest that genetic variation has influences that result in the effective tissue

specificity of changes in mRNA levels and that even the direction of this change is not

readily predictable either within or between individuals.

This analysis identifies dramatic changes in coherency of CGGs across the panel,

supporting the genetic origin of this phenomenon, and allows us to define some extreme

coherency alterations that are likely segregating within the BXD lines. However, we

acknowledge the lack of power to draw more specific conclusions as to the full range of

coherency phenotypes displayed across the complete strain collection.

Encoded protein functions and CGG identity

The existence of CGGs could be interpreted, at the extremes, as either the inevitable

outcome of shared and partially shared mRNA level control or of a more specific regula-

tory architecture evolved to have functional outcomes. To address this latter possibility,

we sought to find functional relationships within CGGs, whose sizes allow for statisti-

cally valid analyses, using Gene Ontology (GO) Biological Process terms (Ashburner

et al., 2000). We found convincing evidence of functional clustering in CGG 2 and

CGG 4. In CGG 4, ten of the fourteen transcripts are annotated: six are ribosomal

proteins (Rps29, Rps15, Rplp2, Rplp1, Rpl35A and Rpl19 ), and two are ribosomal pro-

tein/ubiquitin fusions (Fau and Uba52 ), and showed a highly significant enrichment for

translation (GO:0006412; P = 2 × 10

-6

). CGG 2 contains 63 genes, 35 with GO an-

notations: 13 are involved in carbohydrate metabolism, 5 involved in signalling and 4

involved in transport and was enriched for carbohydrate metabolic process (GO:0005975;

P = 2.1 × 10

-4

). We note that CGG 2 illustrates the complexity of breaking a network

into discrete subnetworks: whilst we can analyse CGG 2 as a whole, there remains

distinct functional clustering even within the CGG. These findings are compatible with

some CGGs having functional significance but certainly do not support the view that

shared function is the major determinant of CGG gene content.

We have stressed that the genetic influences upon CGGs do not have to be at the

level of the control of transcription but this is nevertheless a plausible hypothesis that is

testable. To study this, we examined the CGGs for over-representation of transcription

factor (TF) binding sites (TFBs); our reasoning is that transcriptional control of a CGG

could be due to shared action of TFs and that a variant TF could then contribute

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

to the differential mRNA levels across our BXD panel. Our results are summarised

in Supplementary Table 4 and here we discuss only CGG 2; we identified 24 TFs,

including FOXD3, TCF1, EN1, SP1, GFI1, NKX25, IRF2, and 17 TFs of the Sox

family (Sox1 to Sox9, Sox11 to Sox13 and Sox15, Sox17, Sox18, Sox21 and Sox30 )

whose cognate binding sites were present in more of the promoters of the 63 genes in

CGG 2 than expected by chance (P < 0.05) (see Methods) suggesting they may be

involved in the regulation of the genes. If any of the TFs are contributing to variation

in CGG 2 mRNA levels, we may be able to detect genetic association of the TF gene

with the mRNA levels of some or all of the genes in CGG 2. To identify association, we

carried out an eQTL analysis across all 3 tissues to test for linkage of any of the 63 genes

in CGG 2 to the closest genetic marker to each of the 24 TF genes identified above (see

Methods).

The marker D8Mit124 located 2.3Mb distal of the Sox1 gene on chromosome 8 had

median P -values of 0.001 for the 63 mRNA levels in the brain compared to 0.410 for all

other gene/TF marker combinations, 0.012 in the kidney compared to 0.422 and 0.015

in the liver compared to 0.488. Whilst the individual P -values do not reach significance

under a Bonferroni correction there is nevertheless a striking incidence of low P -values

to this marker. This result is compatible with the hypothesis that some of the variation

in CGG 2 mRNA levels, in all three tissues, may be caused by genetic variation in the

Sox1 gene or protein: the gene is located in a region of low polymorphism and there are

no immediate candidate coding or non-coding SNPs. Proving involvement of Sox1 will

require an experimental design that is outside the scope of this study.

Discussion

In this study we have taken advantage of different genetic backgrounds, to identify

groups of genes whose mRNA levels are likely to be under shared genetic influences

across multiple tissues. We have focused on examining the inbred strains C57BL/6J

and DBA/2J and limited our analyses of genetic influence only to those genes that

were expressed over a defined mRNA level in brain, kidney and liver and that were

differentially expressed between the parental strains in one or more of these tissues:

we identify 755 genes subject to such genetic influence. Using pairwise comparisons of

mRNA levels across 31 recombinant inbred lines of mice derived from this pair of parental

strains, we detect "correlating groups of genes" or CGGs, whose mRNA levels change

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

coordinately across all 31 strain in all three tissues. We then studied the same genes

in the unrelated strain SJL/J and showed that they also exhibit CGG- like behaviour

and exhibit coordinately up or downregulated levels of mRNA, as appropriate. We

further show a striking feature of some CGGs is that genetic variation influences the same

genes in divergent fashions in different tissues of the same individual; genes in a CGG

may be relatively upregulated in one (or more) tissue(s) but relatively downregulated

in another. Unpredictable behaviour is also seen in the behaviour of CGGs compared

across different individuals: for example, mRNAs of a CGG may be upregulated in the

brain of one strain but downregulated in the brain of a second and we have observed

this in replicated studies of C57BL/6J, DBA/2J and SJL/J, as well as in individual BXD

strains. This unpredictability is quite unlike the effects of a protein sequence variation

where an amino acid change is the same in every tissue that expresses the relevant exon.

We identify genetic influence in these studies by detecting pairs of genes whose mRNA

levels vary coordinately in our analyses; however, the proportion of the 755 genes

that are affected is entirely determined by the cutoff used to construct the correlation

network. Consistent with previous analyses (Freeman et al., 2007), we have shown that

there is no simple single criterion that we can use to define this cutoff (indeed there is

no plausible biological reason why there should be a discrete value) but using the cut

offs employed for the three tissue analyses, we can show that 80% of the 755 genes are

genetically influenced in one or more tissues, suggesting these complex genetic influences

are common. It is also likely there are groups of coregulated genes that would not have

been included in our initial 755 gene analysis but that are revealed as being genetically

influenced due to their being subject to transgressive segregation in the BXD lines.

The apparently common but unpredictable influence of genetic variation prompted us

to develop the use of coherency testing, essentially testing the direction rather than

amount of relative mRNA levels change, for analysis of relative CGG gene behaviour.

We believe this is a robust and appropriate test of a CGG that is not based upon the

extreme view that mRNA levels should be identical between 2 genetically dissimilar

individuals. Further extensions to the present methods of coherency testing are also

possible; our current approach is limited to testing the extent to which groups of genes

show uniform changes in expression, but if more complex patterns of coregulation could

be specified, these approaches could remain informative.

Our data adds to three lines of evidence suggesting that the influence of genetic

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

variation is frequently tissue specific. Firstly, several microarray based surveys have

highlighted differences in gene expression across different brain regions in inbred mouse

strains (Freeman et al., 2007; Hovatta et al., 2007; Nadler et al., 2006; Pavlidis and

Noble, 2001; Sandberg et al., 2000) and these differences in expression appear to be

phenotypically relevant, as shown by analysis of interstrain differences in motor coor-

dination tasks (Nadler et al., 2006). Secondly, analyses of eQTL data from studies in

different tissues have shown limited evidence for tissue specific effects Bystrykh et al.

(2005); Chesler et al. (2005); Gatti et al. (2007); Hubner et al. (2005). Thirdly, Yang

et al. (2006), using an intercross of C57BL/6J and C3H/HeJ mouse strains, and sam-

pling muscle, liver, adipose and brain, demonstrated the essentially tissue specific nature

of expression of sexually dimorphic, but not more general, classes of genes.

Functional annotation of genes within each CGG showed that in some cases genes

whose mRNA levels were highly correlated also encode proteins with biologically related

functions; the clearest examples are 13 proteins involved in sugar metabolism clustered

in CGG2 and 6 ribosomal proteins in CGG4. The correlated behaviour of functionally

related genes is perhaps not surprising in view of numerous studies on the coregulation

of gene expression; our major conclusion however is that shared function does not appear

to be the primary organising principle of most genes within a CGG. In this respect, a

better understanding of the shared behaviour of the CGG and its relationship if any, to

phenotypic outcomes (Goring et al., 2007; Nadler et al., 2006; Passador-Gurgel et al.,

2007), will provide greater insight into the functional consequences of CGG variation

and shared control.

The proportion of the variation in an individual genes mRNA level that can be

ascribed to shared CGG influences ranges from 40-90%, which is very similar to reported

results of eQTL analyses, in particular of effects which are in cis to a gene (Hubner et al.,

2005; Petretto et al., 2006; Stranger et al., 2005; West et al., 2007). Logically, influences

shared between 2 or more genes are difficult to reconcile with cis acting variations and

the smaller effects on mRNA levels of the trans acting influences detectable in eQTL

studies suggests that the correlation influences we detect are the outcome of numerous,

additive, in trans influences that are individually not easy to detect. We note that our

study design is, like most other published accounts, underpowered to detect significant

eQTLs at a whole genome scale and we have therefore not attempted this approach at

a global level. We do however provide evidence that multi-factorial transacting genetic

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

variants must exist; appearing to influence gene groups of modest size, as supported by

previously published eQTL analyses.

Steadystate mRNA levels are set by a complex set of regulatory interactions, only

some of which will be primary modulations of transcription. Our findings for CGG 2

that the Sox binding site is over-represented and mRNA levels of the genes within the

CGG exhibit unusual linkage at the region harbouring Sox1, suggest an involvement of

this transcription factor in CGG 2 behaviour but this is necessarily speculative. The

reality is that our methods, in common with all such analyses, including eQTL based

approaches, cannot distinguish between primary and secondary influences upon mRNA

levels. For example, whether an unobserved common regulator causes CGG 2 behaviour,

or variation in more distal processes, such as signal transduction, will have to be shown

by extensive mechanistic dissection, but such followup studies will minimally have to

be able to distinguish between these alternatives.

In more general terms, we have focused upon correlationbased approaches in our

study with the assumption that correlation is a likely outcome of biological processes

rather than simply using correlation as a statistical tool. This study has not been de-

signed to identify, in most cases, the cause of a change in mRNA level but rather we

have simply focused upon defining, at the level of mRNA, the phenotypic differences

between two organisms that are likely due to the sum total of all relevant genetic in-

fluences. Of course, changes in mRNA levels do not have to be reflected in changing

protein levels and in most cases it is this latter change that will contribute to phenotypic

diversity. Recent studies in yeast from Foss et al. (2007) have shown there is only weak

correlation of mRNA and protein levels tested across genetically divergent strains, and

so prediction from purely genotypic information of ultimate protein levels, and therefore

potential phenotype, is going to be a very challenging task even at a single tissue, let

alone at a multiple tissue or organismal level. Nevertheless, the observation that this

type of genetic variation has strong tissue specific outcomes suggests that the regulatory

architecture of mRNA levels may have evolved, in part, to generate selective phenotypic

diversity of individual tissues and could represent a contributing source of morphological

and functional evolutionary differences.

Finally, if tissue specificity of genetic influence is replicated in humans, then using

mRNA levels measured in readily available surrogate tissues will not easily predict out-

comes in more relevant tissues and this will have very substantial implications for the

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

design of human studies.

Methods

RNA preparation

Eight week old, male Mus musculus strains C57BL/6J, DBA/2J and SJL/J were ob-

tained from the Biological Resources Centre, UNSW (Sydney, Australia) and Mus mus-

culus BXD/TyJ strains 1, 2, 5, 6, 8, 9, 1116, 1824, 2729, 3034, 36, 38, 39, 40, and

42 were obtained from the Jackson Laboratories (Bar Harbor, ME, USA). Whole brain,

kidney and liver tissues were harvested according to protocols approved by the University

of New South Wales Animal Care and Ethics Committee (Ethics Code ACEC 01/43),

and snap frozen in liquid N

. Total RNA was extracted according to the manufacturer's

instructions with TRIzol Reagent (Invitrogen, Mt. Waverley, Vic, Australia); purity and

integrity was assessed by OD

260

/OD

280

readings greater than 2 and intact rRNA bands

(Agilent Bioanalyzer, Agilent, Forest Hills, Vic, Australia) analysis, respectively.

Parental strain experiment: Total RNA from the three tissues of 10 individuals

was pooled for each strain (9 for liver) to remove individual variation in gene expression;

20 µg of pooled RNA and 2 µg of Lucidea Universal Scorecard Spikein (Amersham Bio-

sciences, Castle Hill, NSW, Australia) were reverse transcribed using the SuperScript III

Indirect cDNA Labelling System (Invitrogen, Mt. Waverley, Vic, Australia) and fluo-

rescently labelled with Alexa Fluor 555 for C57BL/6J and Alexa Fluor 647 for DBA/2J

(Invitrogen, Mt. Waverley, Vic, Australia).

BXD panel experiments: Equal amounts of total RNA from 3 animals from each

BXD strain were mixed to give tissue pools representative of the genetic backgrounds. A

common reference sample was created for each tissue from total RNA extracted from ten

eightweekold male C57BL/6J mice (a different RNA source than the parental strain

experiment). 20 µg of pooled RNA was reverse transcribed (as above) and fluorescently

labelled with Alexa Fluor 555 for C57BL/6J and Alexa Fluor 647 for BXD strain samples

(as above).

C57 versus SJL experiment: Total RNA from the brain, kidney and liver of

five C57BL/6J and five SJL/J individuals was pooled for each strain. cDNA synthesis

was same as for C57BL/6J vs. DBA/2J experiment, but sodium tetraborate instead

of sodium bicarbonate was used in the labelling buffer. Again, C57BL/6J cDNA was

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

labelled with Alexa Fluor 555 and SJL/J with Alexa Fluor 647 for DBA/2J (Invitrogen,

Mt. Waverley, Vic, Australia).

Microarray experiments

Parental experiment: For each tissue, labelled cDNA was directly compared on 6

replicate glass slide two-colour microarrays containing the Compugen Mouse OligoLi-

brary representing 21,997 genes and Lucidea Universal ScoreCard (Clive and Vera Ra-

maciotti Centre for Gene Function Analysis, UNSW, Sydney, Australia), in 100 µL of

DIGEasy buffer (Roche, Basel, Switzerland) with 5 µL each yeast tRNA and calf thymus

DNA as blockers (Invitrogen, Mt. Waverley, Vic, Australia). Utility controls from the

Lucidea Scorecard were not used, and therefore served as additional negative controls.

Hybridised microarrays were washed in 1 × SSC, three times in 1 × SSC, 0.1% SDS at

C, and three times in 1×SSC, dried by centrifugation, and scanned with the GenePix

4000B microarray scanner (Axon Instruments, Union City, CA, USA). BXD panel ex-

periments: Identical arrays and processing as above, with one array being performed

for each tissue in each BXD line, giving a total of 31 × 3 = 93 arrays. C57 versus SJL

experiment: Identical arrays and processing as above, but three microarrays per tissue

were performed per tissue, giving a total of 3 × 3 = 9 arrays.

Data processing

Image analysis was performed with the Spot image analysis software version 2 (CSIRO,

Australia, texttthttp://experimental.act.cmis.csiro.au/Spot/index.php). All further data

processing and statistical analyses were performed using R version 2.0.0 (Ihaka and Gen-

tleman, 1996). Gene expression data were morph background corrected and log

trans-

formed. Data for controls and the 232 replicated spots of the housekeeping gene Gapd

(NM 008084) were removed prior to normalization to avoid bias. Parental experi-

ment: All 18 slides were then normalized for intensity and spatial bias using printtip

loess and then quantile adjusted to adjust for the differing scale of measurements across

arrays (Yang et al., 2001), and replicate slides were averaged. BXD panel experi-

ments: All 93 slides were normalized using printtip loess. To standardise across ex-

periments from the three tissues, we subselected the data from genes considered to be

expressed in all 3 tissues in the parental experiment and then applied quantile normaliza-

tion. The log

ratios of intensities, M = log

R - log

G, (referred to as M -values) were

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

subsequently used as expression measurements. C57BL/2J vs. SJL/J experiment:

Processing as for parental experiment.

Differential expression in parental strains across multiple tissues

We classified genes as reliably detected if their log mean intensity, A = 0.5(log

R + log

G),

was greater than the 95th percentile of negative controls present on our arrays, in all

three tissues. B statistics were then calculated for all genes, using default parameters in

the R limma library version 1.8.6 (Smyth, 2004), part of the Bioconductor project (Gen-

tleman et al., 2004); genes were classified as genetically influenced if they had both a

B-statistic (LOD)> 3 and an A-value greater than the intensity threshold. 6,075 genes

were detected above in all three tissues; and of these 755 were genetically influenced in

one or more tissue.

Cross-tissue correlation analysis

In order to identify the genes that have similar expression patterns to gene g

in all

tissues, we adopted a correlation-based approach. There are 3 pertissue expression

matrices, E

brain

, E

kidney

and E

liver

, each of dimension G × S, where G is the number of

genes and S is the number of strains, that is, 755 genes×31 strains in the present case.

Pairs of genes that are correlated with each other in all three tissues are of primary

interest because they may be under the influence of some common, tissue independent

regulatory mechanisms. We identify such pairs of genes by joining the three per-tissue

expression matrices E

brain

, E

kidney

and E

liver

into a single G×3S crosstissue expression

data matrix:

BKL

= (E

brain

kidney

liver

)

We then computed a G×G correlation matrix, C

BKL

, from E

BKL

using Spearman's

as a distance measure. C

BKL

is referred to as the crosstissue correlation matrix. C

BKL

was then hard thresholded for various values of ||, thus defining the adjacency matrix,

BKL

, representing an undirected simple graph. In the present study, all networks

were generated using a threshold of || 0.775 (see next section for discussion). The

cross-tissue coexpression network, defined from this adjacency matrix, was visualised

using custom R code using the igraph and RGL libraries. Nodes in Figure 2 were laid

out using the 2DFruchtermanReingold algorithm (Fruchterman and Rheingold, 1991),

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

computed using implementations available in the igraph library in R, and visualised

using the rglplot function (see figure legends for specific details).

CGG centroid R

analysis

The centroid of each CGG is the perstrain average Mvalue for all genes in the CGG,

which we calculated for each tissue independently, or from all three tissues combined. To

determine the similarity of each gene in the CGG to its centroid, we compute R

as the

square of the Pearsons productmoment coefficient (r), obtaining a distribution of R

values for all genes in the CGG. We assess the statistical significance of the observed R

by permutation analysis. We repeat this analysis for random CGGs, chosen by randomly

sampling the same number of genes from the set of 755 genes, obtaining a distribution of

values for each gene in the random CGG, to the random CGGs centroid. We compare

the observed distribution of R

to the random distribution using the MannWhitney U

test, using the uppertail P -values. We repeat this for 1000 random CGGs, and count

the number of times the P -value < 0.05, divided by the number of permutations.

Similar results are obtained if the random genes are resampled from the set of 6075

genes, or if the random genes are compared to the observed CGGs centroid, rather than

the random CGGs centroid (data not presented).

InterBXDstrain coherency

The coherency test-statistic is designed to measure how consistent the directionality of

relative expression is in a set of genes (see Results: The collective behaviour of

CGGs). Given the expression ratios (M -values) from the comparison of two strains

(such as C57BL/6J vs. DBA/2J), and a set of genes, G = {g

, . . . , g

}, with corre-

sponding measurements of average relative expression, ^

, across a set of replicates,

associated with each gene, the vertexbased coherency, C

is calculated as follows:

N
k=1

sign( ^

)

where sign is the sign function, defined as:

sign(x) =

x > 0

x = 0

-1 if x < 0

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

Thus, this vertex-based-coherency score ranges from [-1, 1], with values closer to

+1 indicating more coherent up-regulated expression, values closer to -1 indicating

more coherent down-regulated expression and values closer to 0 indicating less coherent

expression. Permutation test: we chose 1000 random sets of G genes, from a set of 755

genes (by permuting gene labels) and assessed the significance of the observed coherency

of each CGG using the following formula:

P =

#{|C

| |C

where G

denotes a randomised version of gene set G, defined using the label-permuted

set of 755 genes, and B is the number of such permutations generated. For example,

if the given CGG had a vertexbasedcoherency score of 0.77 and of 1000 randomised

samples, only 6 scores were observed to be larger than 0.77, then the P -value would be

6/1000=0.006

To test for the significance of a coherency score in just a single microarray test on

a single BXD strain, we resample the appropriate number of genes in a CGG from

the 755 gene set, conditional on the observed coherence of the actual CGG genes in

DBA/2JvsC57 experiment; for example if coherence of 10 genes is +0.8 we randomly

identify 8 up regulated and 2 down regulated in the DBA/2JvsC57 comparison and

calculate the coherence of these genes in each of the 31 BXD lines, repeating the process

1000 times. We score and individual BXDs coherency score as being significant only if

observed coherencies equal or greater than all 1000 random tests (or less than for -ve

coherency).

Gene Ontology analysis

To test for enrichment or depletion of a GO term in a set of genes of interest, we

tested whether genes of interest were mapped to the GO term at a level greater than

chance expectation (defined as the observable proportion of genes mapping to the term

in the set of expressed genes in the experiment) using sampling without replacement

from the hypergeometric distribution (using the phyper function in R). We used a strict

Bonferroni correction for P < 0.05, corrected for the number of terms with > 5 genes

annotated to them, either directly or via transitive relationships in the ontology. We

employed the Bioconductor package GO (v1.1.14), and mapped microarray identifiers

(GenBank ids) to Entrez Gene ids based on probe-sequence-similarity using custom

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

scripts (available on request).

Transcription factor binding motifs

The GenBank sequences for each of the 6,075 expressed genes were aligned to the ncbi35.1

build of the mouse genome using BLAT (version 32x1; Kuhn et al., 2007), and the best

hits were retained. The upstream 1000 bp from these sequences was then retrieved using

BioPerl into FastA formatted files. Repeat regions were masked to lowercase letters

using RepeatMasker (version open3.1.6) and RepBase (version 20061006) using the

following flags: "--species mouse --xsmall --gff". Then the upstream sequences

for all of the genes in each connected component were separated into a separate FastA

formatted file. The Transcription Factor motif library from JASPAR (Vlieghe et al.,

2006) was downloaded (jaspar2005core) and formatted to suit CLOVER using tools from

the clover download page (http://zlab.bu.edu/clover) (Frith et al., 2004). CLOVER:

CiseLement OVERrepresentation (version Mar 29 2006) was run to search for over-

represented motifs in the upstream sequences from the genes in each regulon compared

to a background set of sequences from the 6075 expressed genes. This data was permuted

1000 times to generate P -values for over/under representation in the data sets. The

following flags were used when running clover: "-l -t 0.05".

eQTL analysis for genes in CGG2

For all expression phenotypes in CGG2 (63 genes), we calculated linkage test statistics for

the closest marker (www.webqtl.org; Chesler et al., 2004) to each of the 24 transcription

factor encoding genes whose binding motifs were enriched in the proximal promoters of

genes in CGG2. We identified the SOX binding motif as being over-represented, and since

most SOX proteins are expected to recognised the same motif (P Koopman, personal

communication), we consider all Sox genes. This analysis was performed in each of the

three tissues separately. We estimated significance of linkage to each marker using likeli-

hood ratio statistics (LRS) and modelbased P -values calculated using the QTL Reaper

code (v1.1.0 with single marker analysis option; www.genenetwork.org/qtlreaper.html).

We corrected the number of comparisons (marker × gene × tissue) using the Bonferroni

correction.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

Acknowledgements

We thank Maja Bucan and John Schimenti for assistance with obtaining the BXD mice.

We also thank Grant Morahan for access to BXD strains, Ruby Lin for technical advice

on microarrays, and RNA extraction, and Peter Koopman for advice regarding the Sox

family of TFs. MJC and EKFC were supported by Australian Postgraduate Awards;

CJC was supported by a School of Biotechnology and Biomolecular Sciences Genome

Information Scholarship; RBHW by a National Health and Medical Research Council of

Australia Peter Doherty Fellowship, and the project work was funded by an Australian

Research Council grant to PFRL. High performance computing resources were provided

by the Australian Partnership for Advanced Computing (APAC), and the Australian

Centre for Advanced Computing and Communications (ac3).

References

Ashburner, M., Ball, C. A., Blake, J. A., Botstein, D., Butler, H., Cherry, J. M., Davis, A. P., Dolinski, K., Dwight,

S. S., Eppig, J. T., Harris, M. A., Hill, D. P., Issel-Tarver, L., Kasarskis, A., Lewis, S., Matese, J. C., Richardson,

J. E., Ringwald, M., Rubin, G. M., and Sherlock, G. (2000). Gene ontology: tool for the unification of biology. the

gene ontology consortium. Nat Genet , 25(1), 2529.

Beck, J. A., Lloyd, S., Hafezparast, M., Lennon-Pierce, M., Eppig, J. T., Festing, M. F., and Fisher, E. M. (2000).

Genealogies of mouse inbred strains. Nat Genet , 24(1), 2325.

Brem, R. B. and Kruglyak, L. (2005). The landscape of genetic complexity across 5,700 gene expression traits in yeast.

Proc Natl Acad Sci U S A, 102(5), 15721577.

Bystrykh, L., Weersing, E., Dontje, B., Sutton, S., Pletcher, M. T., Wiltshire, T., Su, A. I., Vellenga, E., Wang, J.,

Manly, K. F., Lu, L., Chesler, E. J., Alberts, R., Jansen, R. C., Williams, R. W., Cooke, M. P., and de Haan, G.

(2005). Uncovering regulatory pathways that affect hematopoietic stem cell function using 'genetical genomics'. Nat

Genet , 37(3), 225232.

Chesler, E. J., Lu, L., Wang, J., Williams, R. W., and Manly, K. F. (2004). Webqtl: rapid exploratory analysis of gene

expression and genetic networks for brain and behavior. Nat Neurosci, 7(5), 485486.

Chesler, E. J., Lu, L., Shou, S., Qu, Y., Gu, J., Wang, J., Hsu, H. C., Mountz, J. D., Baldwin, N. E., Langston, M. A.,

Threadgill, D. W., Manly, K. F., and Williams, R. W. (2005). Complex trait analysis of gene expression uncovers

polygenic and pleiotropic networks that modulate nervous system function. Nat Genet , 37(3), 233242.

Cotsapas, C. J., Williams, R. B. H., Pulvers, J. N., Nott, D. J., Chan, E. K. F., Cowley, M. J., and Little, P. F. R.

(2006). Genetic dissection of gene regulation in multiple mouse tissues. Mamm Genome, 17(6), 490495.

Foss, E. J., Radulovic, D., Shaffer, S. A., Ruderfer, D. M., Bedalov, A., Goodlett, D. R., and Kruglyak, L. (2007).

Genetic basis of proteome variation in yeast. Nat Genet , 39(11), 13691375.

Freeman, T. C., Goldovsky, L., Brosch, M., van Dongen, S., Maziere, P., Grocock, R. J., Freilich, S., Thornton, J., and

Enright, A. J. (2007). Construction, visualisation, and clustering of transcription networks from microarray expression

data. PLoS Comput Biol , 3(10), 20322042.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

Frith, M. C., Fu, Y., Yu, L., Chen, J.-F., Hansen, U., and Weng, Z. (2004). Detection of functional dna motifs via

statistical over-representation. Nucleic Acids Res, 32(4), 13721381.

Fruchterman, T. M. and Rheingold, E. M. (1991). Force-directed placement. Software Experience and Practice, 21(11),

11291164.

Gatti, D., Maki, A., Chesler, E. J., Kirova, R., Kosyk, O., Lu, L., Manly, K. F., Williams, R. W., Perkins, A., Langston,

M. A., Threadgill, D. W., and Rusyn, I. (2007). Genome-level analysis of genetic regulation of liver gene expression

networks. Hepatology, 46(2), 548557.

Gentleman, R. C., Carey, V. J., Bates, D. M., Bolstad, B., Dettling, M., Dudoit, S., Ellis, B., Gautier, L., Ge, Y.,

Gentry, J., Hornik, K., Hothorn, T., Huber, W., Iacus, S., Irizarry, R., Leisch, F., Li, C., Maechler, M., Rossini, A. J.,

Sawitzki, G., Smith, C., Smyth, G., Tierney, L., Yang, J. Y. H., and Zhang, J. (2004). Bioconductor: open software

development for computational biology and bioinformatics. Genome Biol , 5(10), R80.

Ghazalpour, A., Doss, S., Zhang, B., Wang, S., Plaisier, C., Castellanos, R., Brozell, A., Schadt, E. E., Drake, T. A.,

Lusis, A. J., and Horvath, S. (2006). Integrating genetic and network analysis to characterize genes related to mouse

weight. PLoS Genet , 2(8), e130.

Gibson, G. and Weir, B. (2005). The quantitative genetics of transcription. Trends Genet , 21(11), 616623.

Goring, H. H. H., Curran, J. E., Johnson, M. P., Dyer, T. D., Charlesworth, J., Cole, S. A., Jowett, J. B. M., Abraham,

L. J., Rainwater, D. L., Comuzzie, A. G., Mahaney, M. C., Almasy, L., MacCluer, J. W., Kissebah, A. H., Collier,

G. R., Moses, E. K., and Blangero, J. (2007). Discovery of expression qtls using large-scale transcriptional profiling

in human lymphocytes. Nat Genet , 39(10), 12081216.

Hovatta, I., Zapala, M. A., Broide, R. S., Schadt, E. E., Libiger, O., Schork, N. J., Lockhart, D. J., and Barlow, C.

(2007). Dna variation and brain region-specific expression profiles exhibit different relationships between inbred mouse

strains: implications for eqtl mapping studies. Genome Biol , 8(2), R25.

Hubner, N., Wallace, C. A., Zimdahl, H., Petretto, E., Schulz, H., Maciver, F., Mueller, M., Hummel, O., Monti, J.,

Zidek, V., Musilova, A., Kren, V., Causton, H., Game, L., Born, G., Schmidt, S., Muller, A., Cook, S. A., Kurtz,

T. W., Whittaker, J., Pravenec, M., and Aitman, T. J. (2005). Integrated transcriptional profiling and linkage analysis

for identification of genes underlying disease. Nat Genet , 37(3), 243253.

Ihaka, R. and Gentleman, R. C. (1996). R: A language for data analysis and graphics. Journal of Computational and

Graphical Statistics, 5, 299314.

Komili, S. and Silver, P. A. (2008). Coupling and coordination in gene expression processes: a systems biology view.

Nat Rev Genet , 9(1), 3848.

Kuhn, R. M., Karolchik, D., Zweig, A. S., Trumbower, H., Thomas, D. J., Thakkapallayil, A., Sugnet, C. W., Stanke,

M., Smith, K. E., Siepel, A., Rosenbloom, K. R., Rhead, B., Raney, B. J., Pohl, A., Pedersen, J. S., Hsu, F., Hinrichs,

A. S., Harte, R. A., Diekhans, M., Clawson, H., Bejerano, G., Barber, G. P., Baertsch, R., Haussler, D., and Kent,

W. J. (2007). The ucsc genome browser database: update 2007. Nucleic Acids Res, 35(Database issue), D66873.

Lan, H., Chen, M., Flowers, J. B., Yandell, B. S., Stapleton, D. S., Mata, C. M., Mui, E. T.-K., Flowers, M. T., Schueler,

K. L., Manly, K. F., Williams, R. W., Kendziorski, C., and Attie, A. D. (2006). Combined expression trait correlations

and expression quantitative trait locus mapping. PLoS Genet , 2(1), e6.

Lonnstedt, I. and Speed, T. P. (2002). Replicated microarray data. Statistica Sinica, 12, 3146.

Maciag, K., Altschuler, S. J., Slack, M. D., Krogan, N. J., Emili, A., Greenblatt, J. F., Maniatis, T., and Wu, L. F. (2006).

Systems-level analyses identify extensive coupling among gene expression machines. Mol Syst Biol , 2, 2006.0003.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

Maniatis, T. and Reed, R. (2002). An extensive network of coupling among gene expression machines. Nature, 416(6880),

499506.

Nadler, J. J., Zou, F., Huang, H., Moy, S. S., Lauder, J., Crawley, J. N., Threadgill, D. W., Wright, F. A., and Magnuson,

T. R. (2006). Large-scale gene expression differences across brain regions and inbred strains correlate with a behavioral

phenotype. Genetics, 174(3), 12291236.

Passador-Gurgel, G., Hsieh, W.-P., Hunt, P., Deighton, N., and Gibson, G. (2007). Quantitative trait transcripts for

nicotine resistance in drosophila melanogaster. Nat Genet , 39(2), 264268.

Pavlidis, P. and Noble, W. S. (2001). Analysis of strain and regional variation in gene expression in mouse brain. Genome

Biol , 2(10), RESEARCH0042.

Petretto, E., Mangion, J., Dickens, N. J., Cook, S. A., Kumaran, M. K., Lu, H., Fischer, J., Maatz, H., Kren, V.,

Pravenec, M., Hubner, N., and Aitman, T. J. (2006). Heritability and tissue specificity of expression quantitative

trait loci. PLoS Genet , 2(10), e172.

Rockman, M. V. and Kruglyak, L. (2006). Genetics of global gene expression. Nat Rev Genet , 7(11), 862872.

Sandberg, R., Yasuda, R., Pankratz, D. G., Carter, T. A., Del Rio, J. A., Wodicka, L., Mayford, M., Lockhart, D. J.,

and Barlow, C. (2000). Regional and strain-specific gene expression mapping in the adult mouse brain. Proc Natl

Acad Sci U S A, 97(20), 1103811043.

Sharan, R., Ulitsky, I., and Shamir, R. (2007). Network-based prediction of protein function. Mol Syst Biol , 3, 88.

Smyth, G. K. (2004). Linear models and empirical bayes methods for assessing differential expression in microarray

experiments. Stat Appl Genet Mol Biol , 3, Article3.

Stranger, B. E., Forrest, M. S., Clark, A. G., Minichiello, M. J., Deutsch, S., Lyle, R., Hunt, S., Kahl, B., Antonarakis,

S. E., Tavare, S., Deloukas, P., and Dermitzakis, E. T. (2005). Genome-wide associations of gene expression variation

in humans. PLoS Genet , 1(6), e78.

Taylor, B. A., Wnek, C., Kotlus, B. S., Roemer, N., MacTaggart, T., and Phillips, S. J. (1999). Genotyping new bxd

recombinant inbred mouse strains and comparison of bxd and consensus maps. Mammalian Genome, 10, 335348.

Tsankov, A. M., Brown, C. R., Yu, M. C., Win, M. Z., Silver, P. A., and Casolari, J. M. (2006). Communication between

levels of transcriptional control improves robustness and adaptivity. Mol Syst Biol , 2, 65.

Vlieghe, D., Sandelin, A., De Bleser, P. J., Vleminckx, K., Wasserman, W. W., van Roy, F., and Lenhard, B. (2006). A

new generation of jaspar, the open-access repository for transcription factor binding site profiles. Nucleic Acids Res,

34(Database issue), D957.

Voy, B. H., Scharff, J. A., Perkins, A. D., Saxton, A. M., Borate, B., Chesler, E. J., Branstetter, L. K., and Langston,

M. A. (2006). Extracting gene networks for low-dose radiation using graph theoretical algorithms. PLoS Comput

Biol , 2(7), e89.

West, M. A. L., Kim, K., Kliebenstein, D. J., van Leeuwen, H., Michelmore, R. W., Doerge, R. W., and St Clair, D. A.

(2007).

Global eqtl mapping reveals the complex genetic architecture of transcript-level variation in arabidopsis.

Genetics, 175(3), 14411450.

Williams, R. B. H., Chan, E. K. F., Cowley, M. J., and Little, P. F. R. (2007). The influence of genetic variation on

gene expression. Genome Res, 17(12), 17071716.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

Yang, J. Y. H., Dudoit, S., Luu, P., and Speed, T. P. (2001). Normalization for cdna microarray data. In M. Bittner,

Y. Chen, A. Dorsel, and E. Dougherty, editors, Microarrays: Optical Technologies and Informatics, volume 4266 of

Proceedings of SPIE , pages 112.

Yang, X., Schadt, E. E., Wang, S., Wang, H., Arnold, A. P., Ingram-Drake, L., Drake, T. A., and Lusis, A. J. (2006).

Tissue-specific expression and regulation of sexually dimorphic genes in mice. Genome Res, 16(8), 9951004.

Figure Legends

Figure 1: Network properties of the 755 genes across a range of correlation

thresholds. Networks were constructed for a range of correlation thresholds from 0.05

to 1.0, and each resulting network was tested for: (A) the number of genegene correla-

tions (edges) in the network; (B) the number of connected components in the network;

genes) of the largest connected component (log scale on y-axis); (E) the average degree

of all vertices (log scale on y-axis); and (F) the clustering coefficient. Within each plot,

the solid black dots are the observed data points in the cross-tissue correlation network,

with the 0.775 data point displayed as an open circle. 1000 network permutations were

performed (see text) to generate a null distribution, which is represented as the grey

area. The heavy dashed line is the mean of the null distribution.

Figure 2: (A): Correlations between genes are displayed as a graph: edges connect two

genes if those genes are correlated with an absolute value of Spearmans || > 0.775. 212

of the 755 genetically influenced genes (see text) pass this threshold and are positioned

in the x - y plane based on a 2dimensional Fruchterman-Reingold layout algorithm

(Fruchterman and Reingold, 1991). Disconnected clusters of genes (CGGs) with at least

three genes in them are coloured and numbered. (B) panels show expression differences

of genes in the relevant CGGs measured in each BXD strain in 3 tissues (1st panel brain,

2nd kidney and 3rd liver). The vertical axis is fold change vs. C57BL/6J (M -values)

of mRNA level in each of the 31 BXD strains (horizontal axis). Each individual genes

M -values are plotted as grey lines, with thick coloured lines representing the CGG cen-

troids (blue, green and red for brain, kidney and liver respectively). Note the striking

differences of the same genes expression patterns in the three different tissues.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

Figure 3: CGGs are highly correlated to their own centroid. In each plot, the centroid

for the CGG was computed, and the distributions of R

of each gene in the CGG to the

centroid is plotted as a thick coloured line, with R

along the x-axis, and the density

along the y-axis. The grey lines in each plot correspond to the distributions of R

from

1000 randomly sampled sets of genes (see Methods). Row 1 contains data generated

from combining the gene expression data from the three single-tissues together, and rows

24 correspond to using the single-tissue gene expression data from brain, kidney and

liver respectively. Columns 15 correspond to CGGs 15, respectively.

Figure 4: Coherency analysis. (A) Coherency overview: a CGG containing 12 genes

is identified by correlation analysis in the 31 BXD strains; the expression ratios from a

comparison of the progenitor mouse strains for each of these 12 genes are shown (most

genes are upregulated); the coherency score is calculated (see supplementary informa-

tion); statistical significance is determined via permutation; the resulting coherency, and

statistical significance are displayed as an annotated histogram. This process is repeated

for all CGGs, in expression data from all three tissues, and for two separate pairwise

comparisons of strains (see below). (B) Coherency results: we plot the coherency scores

for each CGG, in the brain, kidney and liver for the comparison of DBA/2J vs C57BL/6J

in the first row (blue, green and red, respectively), and for SJL/J vs C57BL/6J in the

second row (light blue, light green and orange, respectively). Stars indicate the degree

of statistical significance (

= P < 0.05,

= P < 0.005). (C): The same data as in (B),

but reordered so that the tissues are grouped together.

Figure 5: changes in coherency across the individual BXD strains. The coherency

of CGG3 in Kidney (A) and CGG4 in Kidney (B) in all 33 strains investigated in this

study. The CGG depicted in (A) has the whole range of variable expression patterns,

from completely upregulated, to completely downregulated. All genes in the CGG

depicted in (B) are downregulated in DBA/2J, but in the majority of BXD strains (all

of which contain differing amounts of the DBA/2J genetic background), all genes in the

same CGG are upregulated.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

0.2

0.4

0.6

0.8

1e+01

1e+02

1e+03

1e+04

1e+05

Number of gene-gene correlations

Co-expression threshold

0.2

0.4

0.6

0.8

Number of connected components

Co-expression threshold

0.2

0.4

0.6

0.8

100

200

500

Median connected component size

Co-expression threshold

0.2

0.4

0.6

0.8

100

200

500

Size of largest connected component

Co-expression threshold

0.2

0.4

0.6

0.8

100

200

500

Average degree

Co-expression threshold

0.2

0.4

0.6

0.8

0.0

0.2

0.4

0.6

0.8

1.0

Clustering co-efficient

Co-expression threshold

Transitvity

Figure 1: Cowley et al.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

Coherency overview

Identify a CGG

Coherency score

Calculate

coherency

score

= 0.83

+11 -1

Get Expression Ratios

for each gene

in the CGG

Expression ratio

Plot coherency

score and

P-value (**)

Coherency score

+1.0

-1.0

0.0

Determine

statistical

significance

-1.0

+1.0

P < 0.001

CGG 1

CGG 2

CGG 3

CGG 4

CGG coherency in DBA/2J vs C57BL/6J

** **

CGG 1

CGG 2

CGG 3

CGG 4

CGG 1

CGG 2

CGG 3

CGG 4

CGG 1

CGG 2

CGG 3

CGG 4

CGG 1

CGG 2

CGG 3

CGG 4

** **

coherency

CGG coherency in SJL/J vs C57BL/6J

CGG coherency in Brain

CGG coherency in Kidney

CGG coherency in Liver

DBA Brain

DBA Kidney

DBA Liver

SJL Brain

SJL Kidney

SJL Liver

1.0

0.0

-0.5
-1.0

0.5

1.0

0.0

-0.5
-1.0

0.5

1.0

0.0

-0.5
-1.0

0.5

1.0

0.0

-0.5
-1.0

0.5

1.0

0.0

-0.5
-1.0

0.5

Figure 4: Cowley et al.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008

y
c
n
e
r
e
h
o
c

-1.0

-0.5

0.0

0.5

1.0

A
B
D

L
J
S

1
D
X
B

2
D
X
B

5
D
X
B

6
D
X
B

8
D
X
B

9
D
X
B

1
1
D
X
B

2
1
D
X
B

3
1
D
X
B

4
1
D
X
B

5
1
D
X
B

6
1
D
X
B

8
1
D
X
B

9
1
D
X
B

0
2
D
X
B

1
2
D
X
B

2
2
D
X
B

3
2
D
X
B

4
2
D
X
B

7
2
D
X
B

8
2
D
X
B

9
2
D
X
B

1
3
D
X
B

2
3
D
X
B

3
3
D
X
B

4
3
D
X
B

6
3
D
X
B

8
3
D
X
B

9
3
D
X
B

0
4
D
X
B

2
4
D
X
B

Coherency of CGG3 in Kidney

y
c
n
e
r
e
h
o
c

-1.0

-0.5

0.0

0.5

1.0

A
B
D

L
J
S

1
D
X
B

2
D
X
B

5
D
X
B

6
D
X
B

8
D
X
B

9
D
X
B

1
1
D
X
B

2
1
D
X
B

3
1
D
X
B

4
1
D
X
B

5
1
D
X
B

6
1
D
X
B

8
1
D
X
B

9
1
D
X
B

0
2
D
X
B

1
2
D
X
B

2
2
D
X
B

3
2
D
X
B

4
2
D
X
B

7
2
D
X
B

8
2
D
X
B

9
2
D
X
B

1
3
D
X
B

2
3
D
X
B

3
3
D
X
B

4
3
D
X
B

6
3
D
X
B

8
3
D
X
B

9
3
D
X
B

0
4
D
X
B

2
4
D
X
B

Coherency of CGG4 in Kidney

Figure 5: Cowley et al.

Nature Precedings : hdl:10101/npre.2008.1799.1 : Posted 14 Apr 2008