JOURNAL OF EXPERIMENTAL ZOOLOGY (MOL DEV EVOL) 291:175–185 (2001)

© 2001 WILEY-LISS, INC.

Macro-Evolution of the hairy Enhancer inDrosophila Species

JUNHYONG KIM1,2,3*1Department of Ecology and Evolutionary Biology, Yale University, NewHaven, Connecticut

2Department of Molecular, Cellular, and Developmental Biology, YaleUniversity, New Haven, Connecticut

3Department of Statistics, Yale University, New Haven, Connecticut

ABSTRACT It has been suggested that many of the changes in the developmental programmight be in the cis-acting promoters and enhancer regions. Here I study the macro-evolutionarychanges of an enhancer region for the early developmental gene hairy in Drosophila melanogaster,D. simulans, D. pseudoobscura, D. willistoni, D. nebulosa, D. hydei, and D. virilis. The enhancerregion is characterized by small, highly conserved blocks interspersed among highly variable re-gions. Nevertheless, species phylogenies constructed by the enhancer sequences agree with the widelyaccepted phylogeny of these species. The evolution of the variable regions is consistent with a mo-lecular clock, while the evolution of the conserved blocks is significantly different from a clock. Inparticular, the D. pseudoobscura lineage shows the highest degree of species-specific change consis-tent with changes in expression timing reported in an earlier study. It has been suggested that thevariation in sequence length between highly conserved blocks may play a role in the coordination ofregulatory processes, such as protein–protein interactions; thus, stabilizing selection has been sug-gested to act on the length variations. Here I develop a test for stabilizing selection on length varia-tion and show that the hairy enhancer does not show statistically significant evidence for stabilizingselection. J. Exp. Zool. (Mol. Dev. Evol.) 291:175–185, 2001. © 2001 Wiley-Liss, Inc.

Molecular events in development involve geneactions that actuate the ontogenetic process, suchas cell proliferation, cell death, and differentia-tion, and those that regulate the process. The pro-grammatic actions of genes that control theexpression of other genes require the interplay oftrans-factors and cis-factors. The trans-factors,such as transcription factors, are often well con-served across wide range of organismal diversity(e.g., Hooper et al., ’92; Sommer et al., ’92; Gerwinet al., ’94; Lukowitz et al., ’94; Halder et al., ’95;Sackerson, ’95), leading to the idea that many ofthe changes in the developmental program mightoccur in the cis-acting promoters and enhancerregions (Raff, ‘96; McDonald et al., ’97). Even inregulatory regions, conservation at the sequencelevel is observed (Langeland and Carroll, ’94), andconservation of an entire developmental programbetween mouse and fly has been proposed (Xu etal., ’99). However, careful studies have shown sig-nificant changes in the regulatory interactionsmediated by cis-element changes (Ludwig andKreitman, ’95; Chiu et al., ’97; Kim et al., ’99; Wolffet al., ’99), leading Ludwig et al. (’98, 2000) tospeculate that large regulatory regions with mul-

tiple binding domains are under stabilizing selec-tion as a unit and individual binding domains areunder weak selection subject to drift and modifi-cation. Here I examine the upstream regulatoryregion of the Drosophila gene hairy to determinethe mode of evolution of cis-acting elements withina relatively closely related group of organisms.

The hairy gene belongs in the pair-rule classwithin the hierarchy of early developmental genesin Drosophila. Its expression pattern follows thepair-rule spatial pattern of seven periodic bandsof expression along the anterior–posterior axis(Ingham et al., ’85). It is one of several pair-ruleloci whose expression is directly regulated by up-stream gap proteins including hunchback, Krüppel,giant, knirps, and other as-yet-unidentified fac-tors (Ingham et al., ’86; Hooper et al., ’89; Lange-land et al., ’94; La Rosee-Borggreve et al., ’97),and unlike other pair-rule genes, hairy does not

Grant sponsor: Sloan Foundation Young Investigator Award; Grantsponsor: NSF; Grant number: DEB9806570.

*Correspondence to: Junhyong Kim, P.O. Box 208106, New Ha-ven, CT 06520-8106. E-mail: junhyong.kim@yale.edu

Received 28 August 2000; Accepted 7 March 2001

176 J. KIM

seem to be autoregulated (Carroll, ’90). The peri-odic stripes of hairy expression provide the firstindication of the segmented body plan and estab-lish the prepattern for further regulation of down-stream genes. The sequence structure of hairy hasbeen determined (Rushlow et al., ’89), and thegene encodes a 337-amino acid protein that func-tions both in the embryo segmentation body planand in the adult bristle patterning. The majortranscript includes three exons spaced by two in-trons, 1020 bp and 136 bp, respectively (in D.melanogaster). The hairy protein contains a ba-sic helix–loop–helix domain that shows similar-ity to the N-myc proto-oncogene (Wainwright andIsh-Horowicz, ’92). Evidence suggests that it di-rectly regulates the expression of the secondarypair-rule gene fushi tarazu by repression and in-teracts with other pair-rule genes (Carroll andScott, ’86; Howard and Ingham, ’86; Carroll etal., ’88; La Rosee-Borggreve et al., ’99). Similarto even-skipped, the expression of the distinctseven periodic stripes is controlled by a large en-hancer region (Fig. 1) spanning ∼10 kb. Eachstripe expression seems to be mediated by dis-tinct control elements (Carroll et al., ’88). Here Isequenced approximately 1.3 kb of the control el-ement identified as being responsible for the ex-pression of Stripe 1 and Stripe 5 (Langeland andCarroll, ’94) in five new species of Drosophila, andI combined the data with published data from twoother species.

MATERIALS AND METHODSCloning and sequencing of the

hairy enhancerThe Drosophila species stocks, D. simulans, D.

pseudoobscura, D. willistoni, D. nebulosa, and D.hydei, for this paper were obtained from the Bowl-ing Green Stock Center, Ohio. The sequence infor-mation from Langeland and Carroll (’94) was usedto construct four PCR primers within the KpnI–

BamHI fragment of the hairy enhancer Stripe1–5 element. The primer sequences were targetedwithin domains strongly conserved between D.melanogaster and D. virilis and are as follows:[h-ctl-1] 5′-(AG)atattcccggaaattggcaagg-3′, [h-ctl-2] 5′-ccataaaatccttctgcatg(CT)tgc-3′, [h-ctl-3] 5′-gca(AG)catgcagaaggattttatgg-3′, [h-ctl-4] 5′-atgat-catgcc(AC)atcaaagggc-3′. The primer pair h-ctl-1and h-ctl-4 amplifies approximately 1.3 kb of theenhancer sequence, while the h-ctl-1 and h-ctl-3or h-ctl-2 and h-ctl-4 pairs amplify ∼600 to ∼800bases (overlapping with each other) depending onthe species. For each species, PCR amplificationwas first performed using the h-ctl-1 and h-ctl-4primer pair, and if unsuccessful the same regionwas amplified in two fragments using the h-ctl-1and h-ctl-3 pair and the h-ctl-2 and h-ctl-4 pair.The amplified PCR products were then cloned us-ing Invitrogen’s TA cloning kit and sequenced fromboth directions using the ABI 373 automated se-quencer. Some of the sequences generated diffi-cult readouts due to the presence of multiple re-peat sequences. In those cases, internal primerswere constructed and the difficult regions se-quenced multiple times. Geographic isolates of D.melanogaster were also sequenced for a small re-gion. The six isolates are Oregon (published se-quence), Northern California, Minnesota, Ohio,Denmark, and Botswana (stocks from A. Clark,Penn State University). For each of these, thestrains were crossed against a strain with a defi-ciency of the hairy locus and then PCR-amplifiedfrom single flies.

Alignment and analysisSequences were aligned using CLUSTAL W (gap

penalty 5 and extension penalty 2) (Higgins et al.,’96) and manually adjusted. Phylogenetic analy-sis and distance computations were done usingPHYLIP (Felsenstein, ’91), PAUP*, and customprograms. Binding sites for gap genes were de-

Fig. 1. The hairy locus consists of a large upstream en-hancer region with distinct elements for each spatial expres-sion domain and a major transcript consisting of three exons

and two introns. Approximately 1.3 kb of the enhancer re-gion responsible for Stripe1 and Stripe 5 expression was se-quenced.

ENHANCER EVOLUTION IN DROSOPHILA 177

termined with pattern searches using the patternsdocumented in the transcription factor databasein EMBL’s SRS server (Heinemeyer et al., ’98).

RESULTSEnhancer structure

Fig. 2 shows the aligned sequences of the Stripe1–5 element. As expected for typical regulatoryregions, the sequences show a high degree of al-ternating variation between strongly conservedregions and extremely variable regions. The con-served regions generally contain known or puta-tive transcription factor binding domains, whilethe regions in between the conserved blocks arebarely alignable except between closely relatedtaxa. Fig. 2 shows the putative binding sites foundusing the transcription binding site database(Heinemeyer et al., ’98). Binding motifs for allDrosophila transcription factors involved in de-velopmental processes were searched, resulting inputative binding domains for six classes of tran-scription factors (hunchback [hb], caudal [cad],antennapedia [antp], bicoid [bcd], Krüppel [Kr],and paired/fushi tarazu/even-skipped/engrailed[prd/ftz/eve/en]). For further analysis, the se-quences were separated into the conserved regionsand variable regions (marked in Fig. 2 by bound-ing boxes).

Evolutionary divergence of thehairy enhancer

Despite the uncertainties of the alignments in thevariable regions, maximum parsimony and maxi-mum likelihood phylogenetic trees constructed fromthe total data, conserved subset, and the variablesubset (Fig. 3; maximum likelihood tree shown) gaveidentical trees that match well-accepted trees forthese species (DeSalle, ’92). Because the trees areall identical, we can test whether the Stripe 1–5element evolution is consistent with a molecularclock by estimating the same tree using a ultra-metric constraint (DNAMLK, in PHYLIP) andusing the log-likelihood ratio statistic. For the un-constrained trees, the log-likelihood values were–7,649.914, –1,473.10, and –5,733.35, while for theultrametric-constrained tree, the log-likelihoodvalues were –7,656.18, –1,489.88, and –5,737.12,for the combined data, conserved subset, and thevariable subset, respectively. The unconstrainedtree has 11 branch-length parameters, while theconstrained tree has 6 branch-length parameters.Therefore, the log-likelihood ratio statistic will beapproximately distributed as a chi-square distri-bution with 5 degrees of freedom. The log-likeli-

hood ratio statistic is 12.53 (P = 0.025), 33.55 (P< 0.01), and 7.53 (P > 0.1) for the combined data,conserved subset, and variable subset, respec-tively. Therefore, the molecular clock model forthe entire region was rejected, but this seems tobe due to the nonclock-like evolution of the con-served blocks.

Examples of changes in the conserved block areshown in Fig. 4, where in the first instance, aputative krüppel binding domain is disrupted byan insertion in the D. nebulosa and D. willistonilineages while in the second instance a putativehunchback binding domain is deleted in the D.pseudoobscura lineage. Lineage-specific changesin the conserved blocks are also seen in the phy-logenetic tree in Fig. 3. There is a preponderanceof indel changes scattered over the region, espe-cially within the variable regions. For the closelyrelated pair, D. melanogaster and D. simulans,there were approximately (because of alignmentambiguities) 60 mismatch differences and 16separate indel events. Some of the indels wereassociated with repeat variation. The first of thesewas in position 1368, where there was a four-basemotif repeat variation consisting of the followingpattern:

TA.TA.TAGA.TA.TAGA.TA.TAGA.TA.TAGA.TAGA.TA.AAGA.TAGA.TAGA.

All geographic isolates contained a perfect copyof this pattern except for the Botswana isolate thathad a deletion of the TATCTA repeat toward the3′ end. The second variable region consisted of along stretch of T repeats approximately 130 bpupstream of the TCTA repeat which was found intwo different allele types: those with 10 repeats(N. California, Botswana, Ohio) and those with12 repeats (Oregon, Denmark, Minnesota). Thisregion is also highly variable in terms of substi-tution-type changes with 27 of 60 such changesfound as a cluster from position 1122 to 1445.Similar repeat variation was scattered throughout the variable regions of the other species. Forexample, D. nebulosa species has an (AG)10 repeatinserted (in comparison to D. willistoni) at posi-tion 1637.

Length variation of interbinding regionsLudwig et al. (’98) noted considerable variation

in the spacing of the conserved blocks (putativebinding regions) for upstream regulatory regionsof even-skipped. Similar variation is also seen inthe hairy gene. The hairy enhancer region has

178 J. KIM

Fig 2. (continued on following page, with legend.)

ENHANCER EVOLUTION IN DROSOPHILA 179

Fig. 2. Aligned sequences from seven species of the Stripe1–5 enhancer region. The boxed sequences are the subsetsidentified as conserved blocks and analyzed separately. Thethick lines under the conserved blocks indicate putative bind-ing sequences for transcription factors, bicoid (bcd), hunch-back (hb), Krüppel (Kr), antennapedia (Antp), caudal (cad),and a shared motif for paired/fushi tarazu/even-skipped/engrailed (prd/ftz/eve/en).

Fig. 2. (cont.)

17 conserved blocks and 16 interblock variableregions. Each conserved block and its followingvariable sequence were paired together as a se-quence-segment unit (that is, conserved block 1+ variable sequence 1 formed segment 1. The lastconserved block, block 17, was left out as itslength is constant among all species). Table 1shows the length of each species, averaged lengthover the species, and variance of the length foreach of the 16 segments. Considerable variationis seen in the length of the interblock variablesequences as also shown in Fig. 5. Within pairsof close species, of the 16 indels between D.melanogaster and D. simulans, 11 involve dele-tions in D. simulans. In particular, with the ex-ception of a single indel, all indels larger than asingle base involve deletion in D. simulans. How-ever, between D. willistoni and D. nebulosa, 17out of 28 deletions are in D. willistoni with noparticular species bias of the larger deletions.

To assess the phylogenetic pattern of lengthvariations, the length of each segment wastreated as a metric character and pair-wise dis-tances were computed for the taxa. The distanceswere computed both as simple Euclidean dis-tances and as standardized Euclidean distanceswhere the values were transformed into Z-scoresfor each character (= taxonomic distance; Sneathand Sokal, ’73). A neighbor-joining tree (Saitouand Nei, ’87) was computed on the resulting dis-tance matrices. Both ways of computing dis-tances resulted in the same tree topology shownbelow in NEXUS format (the parentheses showtopological relationships, while the numbers fol-lowing the colons show the branch length; Swof-ford, ’91):

180 J. KIM

((neb:30.0,wil:24.5):30.7,pse:564.0,(vir:40.2,(hyd:,24.7,(sim:8.5,mel:11.4):37.5):17.8):14.1)

The tree structure is concordant with the es-tablished species phylogeny within closely relatedtaxa, but the larger-scale relationships are dis-torted. The positions of D. virilis and D. hydei areplaced inside a clade with D. simulans and D.melanogaster. This result suggests that the evo-

lution of the interbinding region length preservesthe phylogenetic covariance structure withinshorter time scales, but at longer time scales suf-ficient evolutionary changes accumulate as toerase the phylogenetic relationships. Another fea-ture of note in the tree is the long branch lengthleading to the D. pseudoobscura lineage (length =564.0), which is consistent with the large changesseen at the sequence level (Fig. 3).

Ludwig et al. (2000) suggested that the spac-ing of interbinding regions might be under stabi-lizing selection due to their possible role inmediating protein–protein interactions of multipletranscription factors. The role of DNA spacing inmediating protein–protein interactions is likelyto be preserved throughout the diversity of Droso-phila. Thus, even at the macro-evolutionary levelwe may expect to see coordinated changes in theinterbinding region length. For example, if thelength of a particular segment X was increased,there might be stabilizing selection to decreasethe length of an adjacent segment Y to preservethe overall length between particular pairs ofbinding sites. Such co-evolution will induce anegative covariance between the segment lengthsof X and Y. For any particular stretch of the en-hancer region we can test the ratio

( ) ,( ) ( ) ( )

n

n

Var S S SFVar S Var S Var S

+ +=

+ +1 2

1 2

LL (1)

where Si denotes the length of the ith segment.If there is no significant positive or negative co-variance between the segment lengths, the sta-tistic (1) will have an F distribution with degreesof freedom (k–1, (k–1)n), where k is the numberof species.

Fig. 3. Phylogenetic tree of the seven species using Stripe1–5 enhancer sequences. Three maximum likelihood trees us-ing the combined data, conserved blocks data, and variable

regions data are shown. The species labels for the two treeson the right are in the same order as the left-most tree.

Fig. 4. Examples of species-specific changes within theconserved blocks. The upper picture shows a disruption of aputative weak Krüppel binding site in D. willistoni and D.nebulosa lineages. The lower picture shows a disruption of ahunchback binding site by a deletion in the D. pseudoobscuralineage.

EN

HA

NC

ER

EV

OL

UT

ION

IN D

RO

SOP

HILA

181

TABLE 1. Length variation in the interbinding regions of the hairy enhancer1

V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11 V12 V13 V14 V15 V16vir 60 75 55 35 33 110 53 90 100 40 64 81 151 75 122 127hyd 52 68 73 36 38 90 47 91 59 31 71 70 165 94 74 113neb 65 72 60 30 68 99 108 102 64 61 47 94 178 73 58 194wil 77 47 81 29 49 80 92 116 61 44 41 94 174 69 71 187pse 99 50 77 32 28 101 82 93 56 34 41 84 159 39 61 167sim 96 61 66 33 49 56 51 83 62 35 40 73 186 108 70 120mel 86 61 66 33 53 56 51 83 63 35 40 76 189 124 71 121Ave 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4 76.4Var 330.29 112.67 85.90 6.29 184.29 467.95 601.14 136.00 226.29 104.00 167.14 92.24 197.24 787.14 458.57 1195.00F 181.95/442.95 76.48/92.19 435/652.24 1180.81/737.14 310.95/330.29 166.14/259.38 1525.14/984.38 1018.24/1653.57(P-value) (P=0.142) (P=0.431) (P=0.322) (P=0.771) (P=0.500) (P=0.303) (P=0.756) (P=0.286)F 193.9/535.14 2551.14/1389.38 821.9/589.67 941.81/2637.95(P-value) (P=0.105) (P=0.866) (P=0.742) (P=0.101)F 2105.29/1924.52 2165.29/3227.62(P-value) (P=0.620) (P=0.326)F 5520.48/5152.14(P-value) (P=0.615)1The hairy enhancer for the Stripe 1–5 element was decomposed into 16 segments as defined in the text. The length of each segment is shown in the columns of the table for each ofthe seven species. The rows labeled “Ave” and “Var” show the average and variance of the segment length over the species. The rows labeled F (P value) show the F statistic for thehypothesis of independent length variation for groups of segments (see text). The lines above each row denote which of the segments are being grouped for the F statistic. Forexample, the first row of the F statistic shows the values for groups of adjacent pairs of segments while the last row shows the F statistic for the entire 16 segments combined. Thevalues shown have the form A/B (P = XXX). The A value is the variance for the grouped segment length while B is the variance obtained as the sum of the individual segment lengthvariances. The P values in the parentheses are obtained from an F distribution with degrees of freedom (k–1, (k–1)n), where k is the number of species and n is the number ofsegments being combined.

182 J. KIM

Table 1 shows the results where the covariancein length evolution was tested in an hierarchicalmanner, first testing the covariance of each adja-cent segments, then the combination of four seg-ments, and so on up to the total length of thesequenced enhancer. The P values are derivedfrom a one-tailed test for the presence of negativecovariance. While there is some indication of nega-tive covariance in segments 1–4 and segments 12–16, the null hypothesis of independent variationcannot be rejected with high significance for anyof length combinations. The F-test presented hereis liberal for the presence of negative covariance(and thus stabilizing selection for overall length)because the numerator of equation (1) may con-tain positive covariance due to phylogeneticallyrelated evolutionary changes in the overall en-hancer length. This can be corrected by adjustingthe degrees of freedom of the numerator to 4 toreflect the nonindependence of the D. melano-gaster/D. simulans pair and D. willistoni/D.nebulosa pair. This will result in higher P values.Thus we still cannot reject the null hypothesis ofno stabilizing selection for length variationsamong interbinding regions.

Patterns of length covariance among all possiblepairs of segments were also assessed by comput-ing pair-wise correlation coefficients. Fig. 6 showsthe results, where black boxes indicate positive

correlation with P < 0.05 and white boxes indi-cate negative correlation with P < 0.05. The pat-tern of pair-wise correlations does not reveal apreponderance of positive or negative covariances.However, several segments toward the 3′ end showhigh positive correlations. One block of positivecorrelations involves segments 7, 8, 10, 12, and16. These segments contain binding domains forantennapedia (segments 7, 8, and 16) and hunch-back (segments 7, 10, and 12). Similarly, a strongpositive correlation is seen between segments 15and 9, both of which contain krüppel binding do-mains. These patterns of positive correlations sug-gest that the regions surrounding similar bindingdomains might be co-evolving. However, segment14, which contains putative binding domains forkrüppel, antennapedia, and hunchback, was notstrongly correlated with the other segments.

DISCUSSIONThe structure of the hairy Stripe 1–5 enhancer

is consistent with other enhancer sequences(Hardison et al., ’97; Ludwig et al., ’98; Stojanovicet al., ’99; Wolff et al., ’99): small highly conservedblocks interspersed between highly variable re-gions with both sequence and length variation.As also reported in Ludwig et al. (’98), the vari-able regions seem to be characterized by consid-erable amounts of indels, which are polymorphic

Fig. 5. Plot of length variation in the hairy Stripe 1–5enhancer region. The total sequence was decomposed into 16

segments of conserved blocks + variable blocks (see text). Theplot shows the length of each segment for the seven species.

ENHANCER EVOLUTION IN DROSOPHILA 183

within species and consists largely of small re-peats. Slippage has been generally suggested asthe mechanism for evolution of the variable re-gions. Recently, MacDonald et al. (’97) has sug-gested LTR retroposons and retroviruses asplaying a possible role in enhancer evolution, butthe data analyzed here show no evidence of anyremnants of inverted repeats. Regardless of themechanism, the variable regions seem to demon-strate clock-like evolution.

The conserved blocks show significantly unequalrates of evolution, especially lineage-specific evo-lution. The D. pseudoobscura lineage shows thegreatest amount of lineage-specific change. Thesechanges include both point mutations and thelarge disruption of a conserved block and lengthchanges. In Kim et al. (’99), it was shown that D.pseudoobscura demonstrates statistically signifi-cant heterochronic changes in the expression tim-ing of hairy where the initiation of the stripeexpression is delayed compared to that in D.melanogaster in terms of both the absolute timeand the cell cycle. D. pseudoobscura also does notshow late expression of an eighth expression stripethat is seen in both D. melanogaster and D.simulans. Expression of hairy stripe is initiatedwith a broad midband expression that separatesinto Stripe 3 and Stripe 5 with concomitant ante-rior expression band that separates into Stripe 1and Stripe 2 (unpublished data). Thus, Stripe 1and Stripe 5 do not initiate the expression pro-cess, and it is not clear that the species-specificchanges in the enhancer region contribute to this

change in timing. But it is at least suggestive thatenhancer evolution may play a role in changes inthe developmental trajectory. The heterochronicshift reported in Kim et al. (’99) is somewhat atodds with the direct transgenic experiment shownin Ludwig et al. (’98) where no significant changein temporal events was found for even-skipped re-porter constructs from different species insertedinto D. melanogaster. But Ludwig et al. (2000) alsoshowed that chimeric reporter constructs of D.melanogaster and D. pseudoobscura showed func-tional defects suggesting co-evolution under sta-bilizing selection of the entire regulatory region.

The significance of the interbinding regionlength variation is difficult to assess. The vari-able regions toward the 3′ end show considerablelength variation among the different species, morethan four times the difference in some cases. Thelength dispersion of the binding domains has beenshown to affect cooperative binding of the trans-factors (Hanes et al., ’94; TenHarmsel and Biggin,’95). Thus stabilizing selection on spatial and tem-poral accuracy of hairy expression may act on theoverall length or particular stretches of the se-quences as suggested by Ludwig et al. (’98). How-ever, statistical tests failed to reject the nullhypothesis of no stabilizing selection at least atthe macro-evolutionary level. Stabilizing selectionmay still act at the micro-evolutionary processes,but it may be that the co-evolution of transcriptorfactors allows segment lengths to vary indepen-dently when assessed across different species.Conversely, functional constraints on enhancersand promotors may be more complex. For ex-ample, Benham (’96) has shown that the sequencecontext of regulatory regions determines the physi-cal characteristics of the DNA duplex, which al-lows the duplex to open for binding by otherregulatory proteins.

By their nature, developmental events are con-tinuously connected. It is undeniable that stabiliz-ing selection operates on developmental processes.But it would seem that such stabilizing selectionwould act downstream in the developmental pro-cess, most likely leading to selection for canaliza-tion of the early processes (Wagner et al., ’97; Rice,’98). In the early processes, drift and co-evolutionof the connected elements of the developmentalpathway (e.g., transcription factor and its bind-ing site) would dominate. Under this circumstancephenotypic traits, such as expression timing andspatial pattern, and genetic traits, such as en-hancer sequences, may show significant phyleticvariation at the macro-evolutionary level.

Fig. 6. Patterns of correlation in pair-wise length varia-tion of the 16 enhancer segments (see text for segment defi-nition). The figure schematically shows the significantcorrelations between all possible pairs. Positive correlationswith P < 0.05 are shown as black boxes, while negative cor-relations with P < 0.05 are shown as white boxes.

184 J. KIM

LITERATURE CITEDBenham CJ. 1996. Duplex destabilization in superhelical DNA

is predicted to occur at specific transcriptional regulatoryregions. J Mol Biol 255:425–434.

Carroll SB. 1990. Zebra patterns in fly embryos: activationof stripes or repression of interstripes? Cell 60:9–16.

Carroll SB, Scott MP. 1986. Zygotically active genes that af-fect the spatial expression of the fushi tarazu segmenta-tion gene during early Drosophila embryogenesis. Cell45:113–126.

Carroll SB, Laughon A, Thalley BS. 1988. Expression, func-tion, and regulation of the hairy segmentation protein inthe Drosophila embryo. Genes Dev 2:883–890.

Chiu CH, Schneider H, Slightom JL, Gumucio DL, GoodmanM. 1997. Dynamics of regulatory evolution in primate β-globin gene clusters: cis-mediated acquisition of simiangamma fetal expression patterns. Gene 205:47–57.

DeSalle R. 1992. The phylogenetic relationships of flies inthe family Drosophilidae deduced from mtDNA sequences.Mol Phylogenet Evol 1:31–40.

Felsenstein J. 1991. PHYLIP (Phylogeny Inference Package),Version 3.4. Seattle, WA: University of Washington (distrib-uted by the author).

Gerwin N, Rosée AL, Sauer F, Halbritter H-P, NeumannM, Jäckle H, Nauber U. 1994. Functional and conserveddomains of the Drosophila transcription factor encodedby the segmentation gene knirps. Mol Cell Biol 14:7899–7908.

Halder G, Callaerts P, Gehring W. 1995. Induction of ectopiceyes by targeted expression of the eyeless gene in Droso-phila. Science 267:1788–1792.

Hanes SD, Riddihough G, Ish-Horowicz D, Brent R. 1994.Specific DNA recognition and intersite spacing are criticalfor action of the bicoid morphogen. Mol Cell Biol 14:3364–3375.

Hardison R, Slightom JL, Gumucio DL, Goodman M, Sto-janovic N, Miller W. 1997. Locus control regions of mam-malian β-globin gene clusters: combining phylogeneticanalyses and experimental results to gain functional in-sights. Gene 205:73–94.

Heinemeyer T, Wingender E, Reuter I, Hermjakob H, Kel AE,Kel OV, Ignatieva EV, Ananko EA, Podkolodnaya OA,Kolpakov FA, Podkolodny NL, Kolchanov NA. 1998. Data-bases on transcriptional regulation: TRANSFAC, TRRD, andCOMPEL. Nucleic Acids Res 26:364–370.

Higgins DG, Thompson JD, Gibson TJ. 1996. Using CLUSTALfor multiple sequence alignments. Methods Enzymol266:383–402.

Hooper JE, Perez-Alonso M, Bermingham JR, Prout M,Rocklein BA, Wagenbach M, Edstrom J-E, de Frutos R, ScottMP. 1992. Comparative studies of Drosophila antennapediagenes. Genetics 132:453–469.

Hooper K, Parkhurst S, Ish-Horowicz D. 1989. Spatial con-trol of hairy protein expression during embryogenesis. De-velopment 107:489–504.

Howard K, Ingham PW. 1986. Regulatory interactions be-tween the segmentation genes fushi tarazu, hairy, andengrailed in the Drosophila blastoderm. Cell 44:949–957.

Ingham PW, Howard KR, Ish-Horowicz D. 1985. Transcrip-tion pattern of the Drosophila segmentation gene hairy.Genes Dev 2:1037–1046.

Ingham PW, Ish-Horowicz D, Howard KR. 1986. Correlativechanges in homeotic and segmentation gene expression inKrüppel embryos of Drosophila. EMBO J 5:1527–1537.

Kim J, Kerr J, Min G-S. 1999. Heterochrony in the molecu-lar developmental cascade of Drosophila. Proc Natl AcadSci U S A 97:212–216.

La Rosee-Borggreve A, Hader T, Taubert H, Rivera-Pomar R,Jackle H. 1997. Mechanism and bicoid-dependent controlof hairy stripe 7 expression in the posterior region of theDrosophila embryo. EMBO J 16:4403–4411.

La Rosee-Borggreve A, Hader T, Wainwright D, Sauer F,Jackle H. 1999. Hairy stripe 7 element mediates activa-tion and repression in response to different domains andlevels Krüppel in the Drosophila embryo. Mech Dev89:133–140.

Langeland JA, Carroll SB. 1994. Conservation of regulatoryelements controlling hairy pair-rule strip formation. Devel-opment 117:585–596.

Langeland JA, Attai SF, Vorwerk K, Carroll SB. 1994. Posi-tioning adjacent pair-rule stripes in the posterior Droso-phila embryo. Development 120:2945–2955.

Ludwig MZ, Kreitman M. 1995. Evolutionary dynamics ofthe enhancer region of even-skipped in Drosophila. Mol BiolEvol 12:1002–1011.

Ludwig MZ, Patel NH, Kreitman M. 1998. Functionalanalysis of eve stripe 2 enhancer evolution in Drosophila:rules governing conservation and change. Development125:949–958.

Ludwig MZ, Bergman C, Patel NH, Kreitman M. 2000. Evi-dence for stabilizing selection in a eukaryotic enhancer ele-ment. Nature 403:564–567.

Lukowitz W, Schroeder C, Glaser G, Huelskamp M, Tautz D.1994. Regulatory and coding regions of the segmentationgene hunchback are functionally conserved between Droso-phila virilis and Drosophila melanogaster. Mech Dev45:105–115.

McDonald JF, Matyunina LV, Wilson S, Jordan IK, BowenNJ, Miller WJ. 1997. LTR retrotransposons and the evolu-tion of eukaryotic enhancers. Genetica 100:3–13.

Raff RA. 1996. The shape of life. Chicago, IL: University ofChicago Press.

Rice SH. 1998. The evolution of canalization and the break-ing of von Baer’s law: modeling the evolution of develop-ment with epistasis. Evolution 52:647–656.

Rushlow CA, Hogan A, Pinchin SM, Howe KM, Lardelli M,Ish-Horowicz D. 1989. The Drosophila hairy protein acts inboth segmentation and bristle patterning and shows homol-ogy to N-myc. EMBO J 8:3095–3103.

Sackerson C. 1995. Patterns of conservation and diver-gence at the even-skipped locus of Drosophila. Mech Dev51:199–215.

Saitou N, Nei M. 1987. The neighbor-joining method: a newmethod for reconstructing phylogenetic trees. J Mol Evol4:406–425.

Sneath PHA, Sokal RR. 1973. Numerical taxonomy. The prin-ciples and practice of numerical classification. San Fran-cisco: W.H. Freeman.

Sommer RJ, Retzlaff K, Goerlich K, Sander K, Tautz D. 1992.Evolutionary conservation pattern of zinc-finger domainsof Drosophila segmentation genes. Proc Natl Acad Sci U SA 89:10782–10786.

Stojanovic N, Florea L, Riemer C, Gumucio D, Slightom J,Goodman M, Miller W, Hardison R. 1999. Comparison offive methods for finding conserved sequences multiple align-ments of gene regulatory regions. Nucleic Acids Res27:3899–38910.

ENHANCER EVOLUTION IN DROSOPHILA 185

Swofford DL. 1991. PAUP: phylogenetic analysis using parsi-mony, Version 3.0s. Champaign, IL: Illinois Natural His-tory Survey.

TenHarmsel A, Biggin MD. 1995. Bending DNA can repressa eukaryotic basal promoter and inhibit TFIID binding. MolCell Biol 15:5492–5498.

Wagner GP, Ginger B, Bagheri-Chaichian H. 1997. A popu-lation genetic theory of canalization. Evolution 51:329–347.

Wainwright SM, Ish-Horowicz D. 1992. Point mutations inthe Drosophila hairy demonstrate in vivo requirements for

basic, helix–loop–helix, and WRWP domains. Mol Cell Biol12:2475–2483.

Wolff C, Pepling M, Gergen P, Klingler M. 1999. Structureand evolution of a pair-rule interaction element: runt regu-latory sequences in D. melanogaster and D. virilis. MechDev 80:87–99.

Xu PX, Zhang X, Heaney S, Yoon A, Michelson AM, MaasRL. 1999. Regulation of Pax6 expression in conserved be-tween mice and flies. Development 126:383–395.