• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of pnasPNASInfo for AuthorsSubscriptionsAboutThis Article
Proc Natl Acad Sci U S A. Apr 29, 2008; 105(17): 6362–6367.
Published online Apr 22, 2008. doi:  10.1073/pnas.0711774105
PMCID: PMC2359775

Complex genetic interactions underlying expression differences between Drosophila races: Analysis of chromosome substitutions


Regulation of gene expression is usually separated into cis and trans components. The separation may become artificial if much of the variation in expression is under multigenic and epistatic (e.g., cis-by-trans) control. There is hence a need to quantify the relative contribution of cis, trans, and cis-by-trans effects on expression divergence at different levels of evolution. To do so across the whole genome, we analyzed the full set of chromosome-substitution lines between the two behavioral races of Drosophila melanogaster. Our observations: (i) Only ≈3% of the genes with an expression difference are purely cis regulated. In fact, relatively few genes are governed by simple genetics because nearly 80% of expression differences are controlled by at least two chromosomes. (ii) For 14% of the genes, cis regulation does play a role but usually in conjunction with trans regulation. This joint action of cis and trans effects, either additive or epistatic, is referred to as inclusive cis effect. (iii) The percentage of genes with inclusive cis effect increases to 32% among genes that are strongly differentiated between the two races. (iv) We observed a nonrandom distribution of trans-acting factors, with a substantial deficit on the second chromosome. Between Drosophila racial groups, trans regulation of expression difference is extensive, and cis regulation often evolves in conjunction with trans effects.

Knowledge of the genetics of complex traits is fundamental to modern medicine, agriculture, and evolutionary biology. Among all complex traits, gene expression as phenotype may be most amenable to genetic analysis. The first question about expression regulation naturally is whether there is a cis component and how strong the cis component is. (cis regulation refers to the control of expression by the gene itself whereas trans regulation refers to the influence of the genetic background.) Many studies have addressed this question at various levels of divergence (19). For example, one may measure the expressions of two alleles at the locus of interest in a common genetic background (usually F1s) (1, 2, 5). Because the collection of trans-acting factors in the same cellular environment is assumed to affect the two alleles equally, asymmetric allelic expression implies differences due to cis-regulatory divergence. Similarly, expression quantitative trait loci (eQTL) mapping permits inference of cis regulation if the eQTL is mapped closely to the expressed gene itself (610).

A second question is how strong cis regulation is relative to trans regulation. In the extreme case where most expression variation is controlled by cis–trans interactions (e.g., joint actions of cis elements and transcription factors), the question would not be very meaningful because cis and trans components are both indispensable. It is desirable to explicitly model expression regulation to include cis, trans, and cis-by-trans control. Many kinds of data allow such explicit modeling. The use of large numbers of recombinant strains for expression analysis is a common approach (6, 7). Such an analysis, however, requires extensive genotyping on many recombinant lines, and each line may receive only limited attention (see Discussion for detail). A simpler alternative is to construct a complete set of whole-chromosome substitution lines, which make it possible to measure gene expression repeatedly under the control of the same chromosome combinations.

The genetic architecture, specifically the relative contribution of cis and trans regulation, may also depend on the extent of divergence between the subjects under comparison. Between two randomly chosen genomes from the same population, it is possible that the 5′-regulatory regions have not differentiated much, and the cis component may be small. When two genomes have diverged, both cis and trans effects would become greater, and their relative magnitude becomes less predictable.

Clearly, the level of divergence should be an important parameter in the analysis of cis vs. trans effects. However, genetic differences between species generally make their hybrids unhealthy, resulting in difficulties in chromosome replacement. On the other hand, individuals or lines from same population may be too similar to have much cis effect. Populations that have shown some degree of phenotypic divergence provide a reasonable balance. The Zimbabwe race of D. melanogaster vis-à-vis the cosmopolitan populations are thus well suited to such a purpose. Females from Z lines (for Zimbabwe) do not readily mate with males from M lines (for melanogaster of the cosmopolitan type), whereas the reciprocal crosses experience much weaker or no discrimination (11).

In this study, we used a standard M (Fr) and Z line (Z30). These two lines have been extensively analyzed for their behavioral, physiological, and molecular divergence (1118). The purpose of this study is to dissect the genetic architecture of transcription regulation. By comparing the expression differences between the two parental (Fr and Z30) and six substitution lines [constructed by Hollocher et al. (11)], we were able to determine the contribution of cis, trans, and cis-by-trans components to gene expression.


The experimental design is given in Fig. S1. We shall designate the two pure lines (Fr and Z30) and the six whole-chromosome substitution lines as ZZZ (i.e., Z30), MZZ, ZMZ, ZZM, MMZ, MZM, ZMM, and MMM (Fr), respectively. Each letter in this triplet notation designates the origin of the X, II, and III chromosome. The IV chromosome is not monitored and is a mixture of Fr and Z30 when the substitution lines were constructed. Note that the three major chromosomes in each substitution line were extracted as a whole without recombination [see Hollocher et al. (11)].

After image analysis, background correction, and normalization, the log-2 expression intensities were analyzed by using a linear mixed model with fixed effect for different fly lines and a random effect for arrays. The residuals from the normalization model were subjected to gene-specific models of the form rijk = μ + Ai + Dj + Gk + εijk, where Gk is kth genotype (k = 1–8) and will be specifically modified in each section below.

We first identified genes that are differentially expressed between ZZZ and MMM by t test. Four kinds of t tests were used, and we considered only genes that pass all four t tests. The genetic control is then analyzed by studying the expression in the six substitution lines. At a false discovery rate (FDR) of 0.01, 188, 493, 534, and 15 genes on the X, II, III, and IV chromosome, respectively, were found to be differentially expressed between MMM and ZZZ (Dataset S1). These genes are distributed among the four chromosomes much like the rest of the genome (P = 0.23 by χ2 test). The mean expression ratio of the more highly expressed line over the less highly expressed line is 1.69. We did not include the 15 genes on the IV chromosome, because the substitution lines could not be identified by their IV chromosome origin.

Average Effects Across All Genes.

The full model for the genetic control of expression differences is given as

equation image

where Ai and Dj represent effect of spot on the array and dye effect, respectively.

With three chromosomes, the expression of each gene can potentially be explained by seven different terms—the three single-chromosome effects (ChrX etc), the 3 interaction terms between any 2 of the 3 chromosomes (ChrX,II, etc.) and the highest level interaction with all three chromosomes. To apply this model, we took two different approaches. In the first approach, we computed the proportion of variance contributed by each of the seven terms, averaged across all 1,215 genes (n = 188 + 493 + 534, see above). In the second approach, we examined the relative contribution of each term for each individual gene. Although we are mainly interested in the genetic control for each gene, the average patterns across loci, which usually have better statistical support, may provide the corroborative evidence to the gene-by-gene analysis.

The average proportions of variances contributed by the seven terms are given in Table 1. These proportions are presented for genes on each chromosome separately as well as jointly. Note that the four interactive terms collectively account for 32.3% (equal to 27.1% + 5.19%) of the total variance. Because the analysis includes interactions between genes of the same chromosome in the category of single-chromosome (noninteractive) effect, the actual level of genetic interaction underlying expression variation should be >32.3%. (Assuming the number of interacting genes is proportional to chromosome size, we estimate that the within-chromosome interactions might augment the reported 32.3% to nearly 50%.) Given that the contribution of interactive terms to expression variation is nonnegligible, the distinction between cis and trans regulation may not always be straightforward.

Table 1.
The proportions of the total genetic variances in expression differences explained by each of the seven terms (the first column) of Eq. 1

Testing Models of Pure Cis* and Pure Trans* Control.

In this section, each gene was individually analyzed. We first tested the two simplest models of genetic control. In the pure cis model, the expression level is fully controlled by the gene itself with no influence from the genetic background. In the pure trans model, the expression is controlled entirely by the genetic background with no input from the expressed gene.

We distinguish between chromosomal cis effect and genic cis effect. Because the genetic resolution is at the whole-chromosome level, we use the notations, cis* and trans* effect, for the whole-chromosome analysis in the statistical models. Cis* effect is the combined effect of the (true) genic cis effect and all trans effects of genes on the same chromosome of the expressed locus. Cis* effect is an overestimate of the true cis effect, and trans* effect does not include same-chromosome trans effects. In Table 2, we present the cis* and trans* effect; the true cis effect is estimated in Table 3.

Table 2.
Number of genes on each chromosome that are subjected to pure cis* and pure trans* regulation at the 5% cutoff
Table 3.
Chromosomal effect on differently expressed genes between MMM and ZZZ

A likelihood ratio test was preformed for the pure cis*-effect model, which is rijk = μ + Ai + Dj + Chrl + εijk, where l is the chromosome from which the expressed gene is located (l = X, II, or III). The alternative model expands the “Chr” term above into the full model of Eq. 1. above. In Table 2, it is shown that the cis*-effect model is accepted over the alternative full model for only a small fraction of genes. At the nominal acceptance rate of 5%, the percentages of genes regulated by a complete cis* effect on X, II, and III chromosome are 6.9%, 2.4%, and 10.1%, respectively, with an average of 6.5%. Given that the cis* effect (at only 1.5% above the nominal rate of 5%) is still an overestimate of the genic cis effect, it seems safe to conclude that not many genes between the Z and M lines are purely cis regulated. As a corollary, 93.5% of the genes must have a trans component in their regulation.

We next test the complete trans*-effect model in which gene expression is controlled exclusively by trans chromosomes (e.g., the II and III chromosomes for X-linked genes). The null model of this test is rijk = μ + Ai + Dj + Chrm + Chrn + Chrm,n + εijk, where m and n are the two trans chromosomes (m, n = X, II, III). The alternative is again the full model described above. At P = 0.05, the percentages of genes regulated entirely by the trans* effect are 7.4%, 37.1%, and 4.5% for the X, II, and III chromosome, respectively (Table 2). Because the sums of the first and third row of Table 2 are far <100%, the bulk of genes must be jointly regulated by cis and trans chromosomes.

A surprising finding in Table 2 is that genes on chromosome II appear to be subjected to weak cis* control and relatively strong trans* control. Given the near parity in gene number between the two autosomes, such a chromosome-wide difference in transcription regulation demands an explanation. The answer appears to be a difference in trans-regulating capacity for genes differentially expressed between Z and M across chromosomes (see below).

Selecting the Best Model for Each Gene.

The criteria for pure cis* and pure trans* effect by the likelihood ratio test above may be too stringent to have much explanatory power. An alternative is to search for the best model for each gene by applying the backward selection method with Akaike information criterion (AIC) (19). In this analysis of model selection, each of the seven terms of genetic effect in Eq. 1 is sequentially removed according to AIC, until further removal resulted in worse fit. Different search criteria including Bayesian information criterion (BIC) and likelihood ratio test (LRT) were also used for model selection. Because the results from these methods are in qualitative agreement, we shall present only the AIC model selection in some detail. The results from BIC and LRT are in Table S1 and Table S2.

In our analysis, 77 genes could not be unambiguously assigned to a specific model and were removed from further analysis. Among the 1,138 genes analyzed, the best model is a single-chromosome effect model for 21.4% of the genes (Table 3). In these cases, only one of the three terms, ChrX, ChrII, or ChrIII, is needed in the selected model. A more complex two-chromosome model is as follows: All models that include the ChrX,II term or contain both the ChrX and ChrII terms are grouped into the X+II two-chromosome category, provided that all other terms containing the chromosome III effect are not needed. A parallel procedure is used for X+III and for II+III, respectively. Among these categories, 28.6% of the genes can be explained by the control of two different chromosomes (that act either independently or jointly). All three major chromosomes are necessary to explain the expression differences for 50% of the genes. Nearly 80% of the genes are regulated by two or three different chromosomes for their expression.

Note that Tables 1 and and33 present very different information. For example, (X,II) in Table 1 is for the interactive term only, whereas “X and II” in Table 3 indicates the presence of effects from both X and II, which include both interactive and additive terms. In Table 1, the average contribution of a single chromosome ranges from 6.7% to nearly 40%. It is therefore not unreasonable to see only 21.4% of genes whose expression variation is entirely attributed to the effect of one single chromosome.

Multichromosome control can be either additive or epistatic. In Table 3, 326 genes are controlled by two chromosomes. Among them, 241 (73.3%) have the interactive term, ChrX,II, ChrX,III, or ChrII,III of Eq. 1. Of 569 genes under the control of all three chromosomes, 560 (98.3%) of them have at least one interactive term. In total, 70.3% [(241 + 560)/1,138] of the genes in Table 3 are influenced by epistatic interactions. Because we cannot account for epistatic interactions between genes on the same chromosome, the percentage 70.3% could be a conservative estimate. The common occurrence of epistatic interactions underlying gene expression again suggests the limitation of a strict cis vs. trans dichotomy in describing the genetic control of gene expression.

In comparison with the results of Table 3, Table 1 shows that the average contribution of the epistatic component to the total variance of expression across all loci is ≈32%, as mentioned above. Thus, whereas the average epistatic effect is 32%, for nearly 30% (100% − 70.3%) of the genes, the epistatic effect is statistically indistinguishable from 0%. The distribution of the epistatic effect among genes appears fairly broad.

Estimating the Pure Cis Effect at the Genic Level.

The proportion of genes controlled solely by the cis* chromosome is estimated to be 9.6% (Table 3 and column 1 of Table 4). Because the chromosomal cis* effect includes both the genic cis effect and intrachromosomal trans effect, the estimate of 9.6% is most likely an overestimate of the genic cis effect. (Without *, cis denotes genic cis effect.) The question is how many genes, if any, would show a genic cis effect if the overestimation is removed. In this section, we will estimate the proportion of genes that are controlled solely by the cis element. This proportion is of course no more than 9.6%. (In the next section, we will estimate the proportion of genes that show a genic cis effect, accompanied by some trans effects in the background.)

Table 4.
The number of genes under cis control

To estimate the genic cis effect, we need to evaluate the intrachromosomal trans effect and subtract this effect from the whole-chromosome cis* effect. We hence compared the trans-acting effects of each chromosome on the other two chromosomes and on itself. By doing so, we would know whether intra- and interchromosomal trans effects are comparable. In Table 1, the trans effect of X on the II chromosome accounts for 21.14% of the variance. The corresponding percentage for the III chromosome is 20.56%. The two numbers are close with an average of 20.84%, suggesting that the trans-regulating strength of X is similar across target chromosomes. For chromosomes II and III, the two interchromosomal trans effects are also very close, ≈7.7% and 32.5%, respectively. Thus, each chromosome's trans effects on the other two chromosomes are rather consistent.

We cannot directly evaluate the intra- vis-à-vis interchromosomal trans effect but indirect evidence can be obtained as follows. Trans-acting factors may affect the expression of a gene by interacting with either its 5′ or 3′ end. On the 5′ end, transcription factors (TFs) interact with their binding sites to regulate the level of expression. On the 3′ end, miRNAs interact with their target sites posttranscriptionally, often resulting in the degradation of transcripts (20). We searched for transcription factor-binding sites (TFBS) that have been experimentally verified. Table S3 shows that TFBSs appear evenly distributed among trans and cis chromosomes. Furthermore, using the TARGETSCAN algorithm (21), we compiled and separated microRNA targets that are located on the same chromosome as the microRNAs from other predicted targets. The chromosomal distributions of the microRNAs and their targets also appear random with respect to chromosomal origin (see Fig. S2). In short, for the two major classes of trans-acting factors (TFs and microRNAs), the assumption that intra- and interchromosomal trans effects are comparable seems reasonable.

Although the targets appear to be independent of the chromosomal location of the trans-acting factors, these factors themselves may not be randomly distributed because there is a large difference in the average trans effect of each chromosome. The difference is particularly notable between the two autosomes. Although the II and III chromosomes are roughly of the same size, the proportion of variance contributed by the latter is >3-fold as high as that contributed by the former. In contrast, the X chromosome contributes as much as 2/3 of that of chromosome III, whereas it is slightly greater than half the size of the latter.

A most interesting feature of Table 1 is the difference between cis* and trans* effects. For X, the cis* effect is 30.33%, whereas the two trans* effects are 21.14% and 20.56%, respectively. The difference between the cis* effect and the average trans* effect, at 9.49%, is highly significant (P < 10−8). In the last column of Table 1, we refer to this difference as “genic cis effect”. The patterns are also true for the two autosomes with the genic effects are 5.27% and 7.05%, respectively (P < 10−6). We estimate the genome-wide genic cis effect at ≈6.7%. Again, the contribution of pure cis effect to expression variation is quite modest, leaving substantial room for trans effects and for cis-by-trans interactions.

The similar procedure may now be applied to estimating the number of genes that are under pure cis regulation (or those that have a cis component in the next section). In Table 3, when the number of genes with expression difference on each chromosome is adjusted for the size of chromosome, the cis effect is also detectable. The effect of X on X is seen in 24 genes (per ≈2,000 genes), but in only 9.7 or 12.7 genes for the effect of X on II or X on III, respectively. The average of the two latter numbers is considered the trans-acting effect of X. In the last column of Table 3, the trans-acting effects per 2,000 target genes are shown to be 11.2, 3.4, and 20.1 for X, II, and III, respectively. Each chromosome's trans effect appears comparable on different target chromosomes, but there is a large difference in the average trans effect of each chromosome. Adjusted for its size, chromosome II is deficient in its trans-acting effect in comparison with the other two.

The number of differentially expressed genes on each chromosome that are under cis* effect is reproduced in column 1 of Table 4. The estimated number of genes with functional cis-regulatory polymorphisms is thus the difference between the number in column 1 of Table 4 and the estimated number of genes under same-chromosome trans control (see Materials and Methods). For example, the number for the X chromosome is 12.8 (24 − 11.2), or 7.5%, as shown in column 2 of Table 4. For II and III, the proportions are 2.2% and 2.7%, respectively. Among all genes that are differentially expressed between the Z and M genomes, only 3.2% are purely cis regulated (genic cis effect only).

Estimating the Inclusive Cis Effect at the Genic Level.

In the Z-M system, the proportion of differently expressed genes that are purely cis regulated, at 3.2%, is almost negligible. A more useful estimate may be the proportion of genes with a detectable cis effect regardless of the presence or absence of other effects. We suggest the term “inclusive cis* effect,” which is defined as follows. The inclusive cis* effect for any X-linked gene is the summation of terms that include X. In other words, the effect is defined as including any combination of the following four terms in the selected model: ChrX + ChrX,II + ChrX,III + ChrX,II,III. We define “inclusive cis (genic) effect” similarly; i.e., the summation of all genetic effects that involves the gene itself.

We first estimate the number of X-linked genes with an inclusive cis* effect, which is 142 (24 + 11 + 38 + 69 from Table 3), shown in the row of X in Table 4. These are cases where the X chromosome plays a role (singly or jointly with other chromosomes) in the expression of X-linked genes. The same procedure is used for other chromosomes. On average, 78% of the expression differences have an inclusive cis* effect, and chromosome II indeed has the lowest percentage of such genes (column 3 of Table 4). Like the estimation of genic cis effect shown in column 2 of Table 4, we extrapolate the interchromosomal trans effect (which is observable) to estimating the intrachromosomal trans effect. We use X-linked genes as an example. We need to estimate the cis effect of x, x+II, x+III, and x+II+III, where the lowercase x denotes individual X-linked genes (rather than the whole X chromosome). The cis effect of x, in terms of the number of genes affected, has been shown to be 12.8. The cis effects of x+II, x+III, and x+II+III are estimated to be 2.76, 0, and 3.9, respectively (see Materials and Methods). The inclusive cis effect for X-linked genes is therefore 19.5 (12.8 + 2.76 + 0 + 3.9), or 11.4% of the affected genes. Across all chromosomes, 13.9% of genes have an inclusive cis effect (column 4).

In summary, only 3.2% of differentially expressed genes are purely cis regulated but, when inclusive cis effect is considered, 13.9% are cis regulated. This comparison means that (i) trans effect dominates the regulation of gene expression at this level of divergence, and (ii) when there is cis effect, cis regulation often occurs in a genetic background where trans regulation is also in operation. The conclusion is consistent with the estimate of the average genic cis effect across all genes, which is 6.7% of the total variation, as shown in Table 1.

Cis Effect in Race-Differentiating vs. Non-Race Differentiating Genes.

We now address the issue whether the relative strength of cis regulation increases as the level of divergence increases. Among the 1,138 genes of Table 3, 238 are considered the more divergent group by the following criteria. In a separate study using multiple M and Z lines, 421 genes were labeled “race-differentiating” (Y.F. and H.-Y.W., unpublished work). For each of these 421 genes, the mean expression of the 24 M lines is significantly different from that of the 15 Z lines at FDR <0.01. Among the 421 genes, 238 are also represented in the set of 1,138 genes analyzed in Table 3 and are referred to as “race-differentiating” genes. The remaining 900 genes are “non-race-differentiating” because their expressions are different both between the Fr (the MMM line) and Z30 (ZZZ) lines and among different M and different Z lines.

We computed the average genic cis effects separately for the two groups of genes, as shown in Table 1. When all genes are considered, genic cis effect accounts for 6.7% of the expression variation. This proportion increases to 8.83% for race-differentiating genes and decreases to 5.95% for non-race-differentiating genes. The difference is 2.88% (P < 0.05, two-tailed t test).

In Table S4, we again analyzed each gene individually. For race-differentiating genes, the percentages of genes with an inclusive cis effect are 45.1%, 32.3%, and 29.6% for X, II, and III, respectively; the average is 32.0%. For “non-race-differentiating” genes, the percentages of genes with a significant inclusive cis effect are 14.3%, 15.3%, and 10.9% for X, II, and III, respectively; the average is 13.2%. The proportion of genes that have a cis-regulatory component is >2.5 times higher among “race-differentiating” genes than among the set that are not race-differentiating. Moreover, the magnitude of changes for race-differentiating genes is generally higher than for the non-race-differentiating ones. The mean fold differences are 2.32 for the former and 1.59 for the latter.


By analyzing the complete set of chromosome substitution lines between the M and Z races, we have been able to quantify the relative contribution of cis and trans regulation as well as the interaction between the two components. The limitation of this study is that it could only estimate the proportion of cis-regulated genes without knowing their identity. A number of previous studies (6, 7, 22), especially those of eQTL analysis, have addressed similar questions. In this study, a full set of whole-chromosome substitution lines, in all possible combinations and kept as permanent stocks, permit repeatable analysis to obtain quantitative estimates of cis, trans, and cis-by-trans effects. Given the reproducibility, we were able to tease apart the within-chromosome trans effects and true genic cis effects (see SI Materials and Methods). Previous studies using random F2 segregants (9, 10) or F1 hybrids (2, 23) did not systematically quantify cis vs. trans effects and rarely addressed the magnitude of the interactive cis-by-trans terms.

At the level of divergence analyzed in this study (between racial groups of the same species), the genetic control of expression is already quite complex, and the contribution of pure cis effect is very modest. Multigenic control, of which cis-by-trans interaction is a most interesting example, appears to be the rule. In the context of complex genetics, the inclusive cis effect may be a useful concept. A gene is considered as having an inclusive cis effect if the cis factor contributes to the expression variation, regardless of whether it exerts the effect singly or jointly with trans factors.

Seemingly incongruent results from different studies of expression regulation may be resolvable if the distinction between cis and inclusive cis effects is made. Osada et al. (13) used the same set of lines as ours and found the difference in gene expression to be correlated with the presence of the cis chromosome (see figure 1 in ref. 13). They concluded that cis* effect (using the new notation) is common at this level of divergence. There may appear to be a discrepancy between our conclusion and theirs because strict cis* effect is detected in only 9.6% of all differentially expressed genes in Table 4. However, because Osada et al. measured the inclusive effect, the discrepancy disappears if we note that 78% of the genes in Table 4 have an inclusive cis* effect.

Likewise, the analysis of expression in the F1 progeny [both between and within species (2, 23)] may be primarily about the inclusive cis effect. This seems to be a plausible suggestion because both cis and trans elements from both parents are present in the F1 progeny. Again, the extent of cis regulation in these reports appears quite substantial. In comparison, our results (Table S4) suggest that 32% of the race-differentiating genes have an inclusive cis effect. Because the level of cis regulation in this study appears lower than the studies of Wittkopp et al. (2, 23), which focused on the more strongly divergent genes, the differences might suggest a possible correlation between expression divergence and the strength of inclusive cis effect. As the cis-regulatory sites become more and more divergent, the majority of differently expressed genes can be expected to have a measurable cis effect. This has been reported (23), and our results (Table S4) also support such an interpretation. Other studies have suggested various degrees of cis regulation (13, 6, 8, 23, 24). For example, in the analysis of genome-wide variation in human gene expression, Morley et al. (6) concluded that 19% of 142 genes have only a cis-acting transcriptional regulator. By integrating transcriptional profiling and linkage analysis, Hubner et al. (7) have shown that 35–40% of the eQTLs between two mouse strains were regulated in cis. In yeast the proportion of cis eQTL from different studies ranges from 10% to 25% (1, 9, 25). The differences may, to some degree, depend on the level of divergence analyzed. (The distinction, or lack thereof, between pure cis and inclusive cis effects also contributes to the variation in the reported estimates.)

As the level of divergence increases from, say, within-species variation to between-species divergence, the strength of inclusive cis effect is expected to increase. Nevertheless, the demarcation between cis and trans effects may become less clear-cut as the interactions between cis and trans factors become more prevalent. We have shown in Table 1 that the contribution of interactive terms to expression variation between populations is >30% and can be as large as 50%. In the analysis of interspecific F1 hybrids, Landry et al. (5) have shown that 13 of the 23 misexpressed genes exhibited cis-by-trans interaction. Furthermore, cis and trans factors can interact in opposition to each other. We have observed substantial counterbalancing effects in gene expression between Z and M races (H.-Y.W., unpublished results). Although strong epistasis between genic actions seems complex, it is hardly surprising. Indeed, numerous introgression studies have suggested that strong epistatic interactions underlie fitness-associated characters between races and species [see Wu and Palopoli (26); Wu and Ting (27) for reviews].

A surprising finding in this study is the much weaker trans-acting effect of chromosome II, relative to that of the X chromosome or chromosome III. Because X and autosomes have many different evolutionary characteristics (2831), the X–II difference may have many explanations. However, the strong contrast between the two autosomes indeed demands an explanation. A suggestion may be that the divergence in the transcription factors (TFs) or trans-acting components on chromosome II is less than that of TFs on chromosome III, because the Z and M races separated. Hollocher et al. (11) and Ting et al. (14) have reported similar chromosome-wide differences. Between the Z and M races, the contribution to behavioral differentiation is substantially larger for chromosome III than for chromosome II.

It has been known that many phenotypic variations between species are subjected to complex genetic control (26, 27). An important reason for studying the control of gene expression is that it may ultimately inform about the genetics of these “classical” phenotypes, including behavior, disease resistance, reproductive success, and so on. The regulation of expression divergence is indeed complex even at the incipient stage of speciation between the Z and M races. Nevertheless, we can often detect inclusive cis effect, which can then be a link between expression phenotypes and (part of) the genetic circuitry.

Materials and Methods

Fly Strains.

One isofemale M line (Fr, hereafter referred as MMM) and one isofemale Z line (Z30, hereafter referred as ZZZ), both of which are commonly used as standards to study mating behaviors (32), and six chromosome substitution lines MZZ, ZMZ, ZZM, MMZ, MZM, and ZZM (11), were used. For chromosome substitution lines, M indicates that a homozygous X, second, or third chromosome is derived from the MMM (Fr) line, and Z indicates the corresponding homozygous chromosomes derived from the ZZZ (Z30) line. The origin of fourth chromosome is ignored in this setting.

Microarray Hybridization.

We designed the hybridization scheme to efficiently estimate differences between MMM and ZZZ lines and chromosome effects (Fig. S1). Each of six chromosome substitution lines was measured four times, and two parental lines were measured 14 times. One hundred to 150 3- to 5-day-old male flies were starved for 1 hour before sedation on ice. Heads were dislodged from the bodies by using sieves after flies were snap-frozen in liquid nitrogen TRIzolR reagent (GIBCO–BRL) was used for RNA extraction. After precipitation, the RNA was then resuspended and purified with phenol and chloroform. cDNA syntheses were carried out by Array 900 expression array detection kit (Genisphere) with manufacture's protocol. Arrays were from DGRC (the Drosophila Genomics Resource Center) of Indiana University (Bloomington, IN). We hybridized and washed the samples according to the protocol from DGRC (33).

Array Analyses.

Twenty-six arrays were scanned by using a GenePix Axon scanner, and data were extracted by using GenePix 6 to give Cy3 and Cy5 intensities. The flowing analyses were conducted in the R computing environment (www.r-project.org). We first transformed the data using spatial-intensity joint loess. This was done with the R/maanova package (34). The transformed log2 intensities for all 15,552 spot measures yijk were subjected to a normalization model of the form yij = μ + Ai + Dj + ADij + εij, where μ is the sample mean, Ai is the effect of the ith array (i = 1–26), Dj is the effect of jth dye (Cy3 or Cy5), ADij is the interaction involving dye and array, and εij is the stochastic error. In this model, array (Ai) was modeled as random effect and was assumed to have a normal distribution with a mean of zero [N(0,σ2A)].

Differentially Expressed Genes Between ZZZ and MMM.

The residuals from the normalization model were subjected to gene-specific models of the form rijk = g + Ai + Dj + Gk + εijk, where g is the average intensity associated with a particular gene; Gk is kth genotype (k = 1–8). Array (Ai) here and thereafter represents as spot effects on the arrays and is also treated as random effect. A t test was performed to identify differentially expressed genes between MMM and ZZZ. Four kinds of t tests namely, standard t test; global t test, using an estimate of error variance that is pooled across all genes; regularized t test, combing information from gene-specific and global average variance estimates by using a weighted average of the two as the denominator for a gene-specific t test; and Fs test proposed by Cui et al. (35) were conducted to access the number of differentially expressed genes between MMM and ZZZ. We also carried out permutation analysis by randomizing the residuals (within each gene) from the fitted null model 1,000 times and recomputing the t statistics. The result did not differ from those obtained by using standard statistical tables. All above tests were done with the R/maanova package.

The details of model selection, cis/trans estimations, hypothesis testing, and variance component estimation are in SI Materials and Methods.

Supplementary Material

Supporting Information:


We thank X. Li, E. Touma, and O. Cruz for technical support; J. Moran, T. Greenberg, S. Fang, C.-T. Ting, S.-C. Tsaur, J. Shapiro, and K. Kumar for comments and discussions during the study. H.-Y.W. thanks James C.-K. Shen for his inspiration and encouragement. This work was supported by a National Institutes of Health grant (to C.-I.W.) and National Science Council Taiwan Grants 96-2311-B-002-009-MY3 and 96-2311-B-002-004- (to H.-Y.W.).


The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

This article contains supporting information online at www.pnas.org/cgi/content/full/0711774105/DCSupplemental.


1. Wang D, et al. Expression evolution in yeast genes of single-input modules is mainly due to changes in trans-acting factors. Genome Res. 2007;17:1161–1169. [PMC free article] [PubMed]
2. Wittkopp PJ, Haerum BK, Clark AG. Evolutionary changes in cis and trans gene regulation. Nature. 2004;430:85–88. [PubMed]
3. Cowles CR, Hirschhorn JN, Altshuler D, Lander ES. Detection of regulatory variation in mouse genes. Nat Genet. 2002;32:432–437. [PubMed]
4. Yan H, Yuan W, Velculescu VE, Vogelstein B, Kinzler KW. Allelic variation in human gene expression. Science. 2002;297:1143. [PubMed]
5. Landry CR, et al. Compensatory cis–trans evolution and the dysregulation of gene expression in interspecific hybrids of Drosophila. Genetics. 2005;171:1813–1822. [PMC free article] [PubMed]
6. Morley M, et al. Genetic analysis of genome-wide variation in human gene expression. Nature. 2004;430:743–747. [PMC free article] [PubMed]
7. Hubner N, et al. Integrated transcriptional profiling and linkage analysis for identification of genes underlying disease. Nat Genet. 2005;37:243–253. [PubMed]
8. Brem RB, Storey JD, Whittle J, Kruglyak L. Genetic interactions between polymorphisms that affect gene expression in yeast. Nature. 2005;436:701–703. [PMC free article] [PubMed]
9. Yvert G, et al. Trans-acting regulatory variation in Saccharomyces cerevisiae and the role of transcription factors. Nat Genet. 2003;35:57–64. [PubMed]
10. Brem RB, Yvert G, Clinton R, Kruglyak L. Genetic dissection of transcriptional regulation in budding yeast. Science. 2002;296:752–755. [PubMed]
11. Hollocher H, Ting CT, Wu ML, Wu CI. Incipient speciation by sexual isolation in Drosophila melanogaster: Extensive genetic divergence without reinforcement. Genetics. 1997;147:1191–1201. [PMC free article] [PubMed]
12. Hollocher H, Ting CT, Pollack F, Wu CI. Incipient speciation by sexual isolation in Drosophila melanogaster: Variation in mating preference and correlation between sexes. Evolution (Lawrence, Kans) 1997;51:1175–1181.
13. Osada N, Kohn MH, Wu CI. Genomic Inferences of the cis-regulatory nucleotide polymorphisms underlying gene expression differences between Drosophila melanogaster mating races. Mol Biol Evol. 2006;23:1585–1591. [PubMed]
14. Ting CT, Takahashi A, Wu CI. Incipient speciation by sexual isolation in Drosophila: Concurrent evolution at multiple loci. Proc Natl Acad Sci USA. 2001;98:6709–6713. [PMC free article] [PubMed]
15. Alipaz JA, Fang S, Osada N, Wu CI. Evolution of sexual isolation during secondary contact: Genotypic versus phenotypic changes in laboratory populations. Am Nat. 2005;165:420–428. [PubMed]
16. Alipaz JA, Karr TL, Wu CI. Evolution of sexual isolation in laboratory populations: Fitness differences between mating types and the associated hybrid incompatibilities. Am Nat. 2005;165:429–438. [PubMed]
17. Greenberg AJ, Moran JR, Fang S, Wu CI. Adaptive loss of an old duplicated gene during incipient speciation. Mol Biol Evol. 2006;23:401–410. [PubMed]
18. Kohn MH, Fang S, Wu CI. Inference of positive and negative selection on the 5′ regulatory regions of Drosophila genes. Mol Biol Evol. 2004;21:374–383. [PubMed]
19. Akaike H. New look at statistical-model identification. IEEE Transact Auto Control. 1974;Ac19:716–723.
20. Filipowicz W, Bhattacharyya SN, Sonenberg N. Mechanisms of post-transcriptional regulation by microRNAs: Are the answers in sight? Nat Rev Genet. 2008;9:102–114. [PubMed]
21. Kheradpour P, Stark A, Roy S, Kellis M. Reliable prediction of regulator targets using 12 Drosophila genomes. Genome Res. 2007;17:1919–1931. [PMC free article] [PubMed]
22. Cheung VG, et al. Mapping determinants of human gene expression by regional and genome-wide association. Nature. 2005;437:1365–1369. [PMC free article] [PubMed]
23. Wittkopp PJ, Haerum BK, Clark AG. Regulatory changes underlying expression differences within and between Drosophila species. Nat Genet. 2008;40:346–350. [PubMed]
24. Wayne ML, Pan YJ, Nuzhdin SV, McIntyre LM. Additivity and trans-acting effects on gene expression in male Drosophila simulans. Genetics. 2004;168:1413–1420. [PMC free article] [PubMed]
25. Ronald J, Brem RB, Whittle J, Kruglyak L. Local regulatory variation in Saccharomyces cerevisiae. PLoS Genet. 2005;1:e25. [PMC free article] [PubMed]
26. Wu CI, Palopoli MF. Genetics of postmating reproductive isolation in animals. Annu Rev Genet. 1994;28:283–308. [PubMed]
27. Wu CI, Ting CT. Genes and speciation. Nat Rev Genet. 2004;5:114–122. [PubMed]
28. Wu CI, Xu EY. Sexual antagonism and X inactivation—the SAXI hypothesis. Trends Genet. 2003;19:243–247. [PubMed]
29. Parisi M, et al. Paucity of genes on the Drosophila X chromosome showing male-biased expression. Science. 2003;299:697–700. [PMC free article] [PubMed]
30. Gupta V, et al. Global analysis of X-chromosome dosage compensation. J Biol. 2006;5:3. [PMC free article] [PubMed]
31. Montgomery E, Charlesworth B, Langley CH. A test for the role of natural selection in the stabilization of transposable element copy number in a population of Drosophila melanogaster. Genet Res. 1987;49:31–41. [PubMed]
32. Wu CI, et al. . Sexual isolation in Drosophila melanogaster: A possible case of incipient speciation. Proc Natl Acad Sci USA. 1995;92:2519–2523. [PMC free article] [PubMed]
33. Cherbas L. Dendrimer Use and Hybridization Protocol. Bloomington, IN: The Center for Genomics and Bioinformatics, Indiana University; 2006. CGB Technical Report 2006-08.
34. Wu H, Churchill GA. Bar Harbor, ME: The Jackson Laboratory; 2005. R/MAANOVA: An Extensive R Environment for the Analysis of Microarray Experiments.
35. Cui X, Churchill GA. Statistical tests for differential expression in cDNA microarray experiments. Genome Biol. 2003;4:210. [PMC free article] [PubMed]

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences


Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...