Bone morphogenetic proteins (BMPs) are secreted signaling molecules that control important developmental events in many different organisms. Previous studies have shown that BMPs are expressed at the earliest stages of skeletal development, and are required for formation of specific skeletal features, strongly suggesting that they are endogenous signals used to control formation of skeletal tissue. Despite the importance of BMP signaling in normal development, very little is known about the mechanisms that control the synthesis and distribution of BMP signals in vertebrates. Here, we identify a large array of cis-acting control sequences that lay out expression of the mouse Bmp5 gene in specific skeletal structures and soft tissues. Some of these elements show striking specificity for particular anatomical features within the skeleton, rather than for cartilage and bone in general. These data suggest that the vertebrate skeleton is built from the sum of many independent domains of BMP expression, each of which may be controlled by separate regulatory elements driving expression at specific anatomical locations. Surprisingly, some of the regulatory sequences in the Bmp5 gene map over 270 kb from the Bmp5 promoter, making them among the most distant elements yet identified in studies of eukaryotic gene expression.
THE bone morphogenetic proteins (BMPs) comprise a large subset of the transforming growth factor-β family of secreted signaling molecules (Kingsley 1994a). Studies in multiple organisms have shown that the BMPs are used to control a wide variety of events during embryonic development, including dorsal-ventral and left-right axis formation, mesenchymal-epithelial interactions, and differentiation of many specific tissues including lung, gut, kidney, hair, teeth, cartilage, and bone (Kingsley 1994a; Hogan 1996). Some of these functions, such as control of dorsal-ventral axis formation in both flies and vertebrates, appear to be ancient functions conserved in many metazoans (Holleyet al. 1996). Other functions may be specific to particular classes of animals: for example, the role of BMPs in controlling wing vein development in insects, or cartilage and bone formation in vertebrates.
Studies of the Drosophila decapentaplegic (dpp) gene suggest that BMPs have been recruited into distinct developmental roles by the addition of separate cis-acting regulatory elements that control expression at specific times and locations. For example, the dpp gene is normally expressed in the dorsal half of the early embryo, in the constrictions of the midgut, and in stripes along the anterior-posterior border in many developing imaginal discs (St. Johnston and Gelbart 1987; Posakonyet al. 1990). Separate elements have been identified that control expression at each of these locations, and many have been shown to interact with different transcription factors that positively or negatively affect expression (Huanget al. 1993; Capovillaet al. 1994; Manaket al. 1994; Sanicolaet al. 1995).
Much less is known about the mechanisms that control expression of BMP genes in vertebrates. Some of these mechanisms may be very similar to those observed in invertebrate systems. This seems particularly likely for expression in structures that may have already been present in the common ancestors of vertebrates and invertebrates, including the heart, gut, and dorsalventral axis. It is more difficult to predict the mechanisms that control expression in tissues and structures that are unique to the vertebrate lineage, such as cartilage and bone. A large number of studies suggest that BMPs are endogenous signals used by vertebrate embryos to induce the formation of skeletal tissue (Kingsley 1994b). BMPs can trigger the entire process of cartilage and bone formation when implanted at ectopic sites in adult animals (Urist 1965; Reddi and Huggins 1972) and are normally expressed in and around early cartilage and bone precursors during embryonic development (Lyonset al. 1989; Joneset al. 1991; Kinget al. 1994; Kinget al. 1996). Moreover, mutations in different BMP genes block the formation of particular skeletal features, showing that BMPs are also required for the normal formation of skeletal tissue (Kingsleyet al. 1992; Stormet al. 1994; Luoet al. 1995; Thomaset al. 1996).
How has an ancient class of signaling molecules been recruited into controlling the formation of skeletal structures in vertebrates? New roles for the BMPs must have involved changes in downstream responses; for example, the formation of a link between BMP signaling and cartilage and bone differentiation. Once this link was established in the vertebrate lineage, gain and loss of regulatory elements in individual BMP family members could provide a simple mechanism for inducing new domains of BMP expression and new cartilage and bone elements at particular locations during embryonic development. Based on the specific skeletal defects seen in mice missing different BMPs, and the different expression patterns of BMPs in developing skeletal structures, we have previously proposed that different members of the BMP family may have accumulated unique sets of control elements that induce the formation of specific anatomical structures in the skeleton (Kingsley 1994b). According to this model, the large family of different vertebrate BMP genes may exist in part because gene duplication and regulatory changes within the BMP family provide a useful mechanism for independently controlling BMP signaling events in different body regions.
To begin to test this model, we have initiated a detailed characterization of the molecular mechanisms that drive expression of BMP genes in mammals. The mouse Bmp5 gene is a particularly favorable place to begin these studies. Null mutations in this gene reduce or eliminate many specific skeletal elements and produce an easily observable external trait, short ears (Lynch 1921; Green 1951; Kingsleyet al. 1992). Mice carrying the original short ear (se) mutation have been used for extensive specific-locus mutagenesis experiments (Russell 1951). As a result of these experiments, over two dozen viable radiation- and chemically-induced alleles have been isolated at the Bmp5 locus (Russell 1971; Russellet al. 1989; Kingsleyet al. 1992; Markeret al. 1997). Here, we use this large collection of short ear alleles to identify cis-acting regulatory sequences that control where and when the Bmp5 gene is expressed during development. Regulatory mutations and transgenic reporter studies show that Bmp5 expression is regulated by a large number of separate DNA elements that control expression at specific anatomical locations in the skeleton and other tissues. Many of these elements are located at unusually large distances from the Bmp5 promoter, suggesting that important cis-acting control elements can act over distances of hundreds of kilobases in the mammalian genome.
MATERIALS AND METHODS
Mice: The se30DThWb, sesv, and se4CHLd alleles were generated in specific locus mutagenesis tests with 101/Rl and C3H/Rl as the parental background strains (Russell 1971), using x-rays, neutrons, or chlorambucil (Russellet al. 1989) in spermatogonia, spermatogonia, or male meiosis, respectively. The se2OZb mutation, x-ray induced in oocytes, is a null allele that deletes the entire Bmp5 coding region (Kingsleyet al. 1992). All mutations are maintained on outbred genetic backgrounds and are currently being backcrossed to inbred strains for detailed studies of skeletal phenotypes (Green 1957).
Inversion breakpoint mapping: DNA samples from control and mutant strains were prepared for conventional and pulsed-field gel analysis as previously described (Markeret al. 1997). A probe at position 84 to 87 kb of the short ear chromosome walk was used to look for alterations in a large 725-kb Bss H2 restriction fragment that spans all Bmp5 coding exons and the 3′ flanking genomic region (Kingsleyet al. 1992). Novel-sized fragments were detected in the se30DThWb, sesv, and se4CHLd mutations. Probes near the 3′ end of this fragment (position 709 kb) detected different novel BssH2 fragments in each mutation, suggesting a chromosome inversion or translocation between the two probes. Sequential hybridizations with probes located between positions 84 and 440 kb indicated that, for all three alleles, the transition point in the altered BssH2 restriction fragment maps 3′ of the Bmp5 coding exons. Conventional restriction mapping and hybridizations with probes near the predicted transition points confirmed the location of each mutation (see Figure 1C). Probe 1 in Figure 1 is the last 1.0 kb of the full-length Bmp5 cDNA. Probe 2 is an end fragment from phage VII. The phage clones located between positions 325 and 430 were generated from a YAC4 (Kingsleyet al. 1992) subgenomic library (Avrahamet al. 1995). cDNA clones for 3′ end mapping were isolated from a E8.5 library (provided by B. Hogan) and by RACE-PCR from adult lung RNA samples. Genomic sequencing showed that alternate ends were encoded by sequences contained within the last Bmp5 coding exon.
Reporter gene constructs: For reporter constructs II–VIII, phage insert DNA was isolated from the lambda-Dash vector by digestion with NotI. For constructs III and IV, the ends were filled in with T4 DNA polymerase, and the insert was ligated into the SmaI site of the pKS(Sal-Sal)+ hsp68lacZpA vector (Kotharyet al. 1989) (a gift from A. Joyner). For constructs II, and V–VIII the same hsp68lacZpA cassette was cut out of the pKS(Sal-Sal) and subcloned into the BamHI site of a standard pKS vector such that the NotI site is located 5′ of the hsp (Not5′hsplacZ). The inserts from phage II and V–VIII were ligated into NotI-digested Not5′hsplacZ vector. The III and IV constructs were linearized with NotI and the II, V–VIII constructs with SalI. All constructs were ethanol-precipitated and resuspended in microinjection buffer (10 mM Tris, pH 7.4; 0.15 mM EDTA, pH 8.0). After two rounds of buffer-exchange with Centricon-30 concentrators (Amicon, Beverly, MA), the DNA concentration was adjusted to 2 ng/μL and filtered through a Centrex MF-1.5 (0.2 μM) Centrifugal Microfilter (Schleicher & Schuell, Keene, NH).
Transgenic production and typing: The DNA was injected into the pronuclei of fertilized mouse eggs derived from FVB/N or C57BL/6 × CBA F1 mice (Hoganet al. 1994). The injected oocytes were transferred into oviducts of pseudopregnant CD-1 mice and transgenic embryos were identified at E14-E15.5 by typing yolk sac DNA with primers from the lacZ gene (forward primer, 5′-TTTCCATGTTGCCACTCGC 3′; reverse primer, 5′-AACGGCTTGCCGTTCAGCA-3′).
lacZ detection and cryosectioning of embryos: Embryos were dissected in cold PBS and a 27-gauge needle was used to make holes in the torso and head cavities. The embryos were fixed for 60–75 min at 4° in 4% paraformaldehyde, cut in half sagitally, fixed for another 10–15 min and washed 3 × 20 min in wash buffer (0.1 m sodium phosphate buffer (pH 7.3), 2 mm MgCl, 0.01% deoxycholate, 0.02% Nonidet P-40). lacZ staining was done in wash buffer supplemented with 1 mg/mL X-gal (GIBCO-BRL, Gaithersburg, MD), 4 mm K3Fe(CN)6, 4 mm K4Fe(CN)6 · 3 H2O, and 0.1 m Tris (pH 7.4) for 40–48 hr at room temperature. Embryos were then rinsed 3 × 30 min in wash buffer and cleared with 75% glycerol. For sectioning, whole-mount lacZ-stained embryos saturated in 50% sucrose were embedded in gelatin/sucrose solution (15% sucrose, 7.5% gelatin in water) and cryosectioned at −25° with thicknesses between 20–30 μm. Sections were dried overnight, rinsed in 1× PBS, counterstained with 0.01% neutral red for 4′, and directly mounted in Aquamount (Lerner Laboratories, Pittsburgh, PA).
In situ hybridization: The Bmp5 probe was a 1.03-kb probe derived from a PCR product of the pro-region of the mouse cDNA (bases 671–1702) (Kinget al. 1994) cloned into pCRII (Invitrogen, San Diego, CA). Digoxigenin-labeled RNA probes were transcribed using the Genius™ labeling system (Boehringer Mannheim, Indianapolis, IN). Timed matings were performed and noon of the day of the vaginal plug was designated as embryonic day 0.5. In situ hybridizations were done as described (Storm and Kingsley 1996) with the following modifications: sections were 14 μm and the hybridization temperature was raised to 64°.
se mutations with chromosome rearrangments are potential regulatory alleles of the Bmp5 gene: Molecular studies have shown that most of the spontaneous, radiation-, and chemically-induced se alleles have lesions that disrupt the BMP5 protein (Kingsleyet al. 1992; Kinget al. 1994; Markeret al. 1997). However, 7 of 27 mutations analyzed do not show any sequence changes in the Bmp5 open reading frame (Markeret al. 1997), suggesting that they may be regulatory mutations. Four of these mutations (se30DThWb, sesv, se4CHLd, se1FrThc) were induced with radiation or chemicals known to cause chromosome rearrangements (Russell 1971; Russellet al. 1989). Probes throughout the 710-kb short ear chromosome walk (Kingsleyet al. 1992) detect cross-hybridizing fragments in DNA samples from these four mutations, ruling out any large DNA deletions in the four alleles (data not shown). However, pulsed-field gel mapping with numerous probes located between positions 84 and 430 kb on the chromosome walk suggested the presence of chromosome inversions or translocations in three of the mutations near the sites indicated in Figure 1B (see materials and methods). Probes at positions 310, 335, and 415 detected altered restriction fragments in DNA from the se30DThWb, sesv, and se4CHLd alleles, respectively, confirming breakpoints at the predicted locations (Avrahamet al. 1995) (Figure 1C). The breakpoint in the se sv mutation has been cloned and sequenced, and consists of a 2-cM chromosome inversion with loss of only 106 bp at the inversion junction (Avrahamet al. 1995).
All three breakpoints map 3′ of previously reported Bmp5 coding exons, consistent with the possibility that the mutations represent regulatory alleles of the se gene. cDNA cloning, 3′ RACE, and genomic sequencing analyses were carried out to test whether additional Bmp5 exons map farther 3′ from those previously identified (data not shown). These studies identified two alternative 3′ ends for Bmp5 cDNAs that account for the major transcripts seen in many different tissues (Kinget al. 1994) (Figure 1B). Both ends are encoded by genomic sequences located immediately adjacent to the Bmp5 open reading frame; the most 3′ cDNA sequence for Bmp5 thus maps to position 310 of the chromosome walk (Figure 1B). The three mutant breakpoints are located approximately 6 kb, 18 kb, and 105 kb past all known Bmp5 exons, confirming that the mutations do not affect Bmp5 exon sequences (Figure 1B).
The se30DThWb and se4CHLd mutations produce an apparent gradient of effects on the size of the external ear. The more distant mutation (se4CHLd) has the mildest effect on ear length. Other rare alleles of this type have previously been classified as intermediate short ear (Russell 1971). The se30DThWb allele has a more pronounced effect, but does not produce ears as short as those seen in mice completely missing the Bmp5 gene (se2OZb) (Figure 1A). Although these mutations have not yet been analyzed on a uniform genetic background, the different ear phenotypes and the 3′ position of the breakpoints suggest a possible analogy with previous studies of the Drosophila BMP-related gene decapentaplegic (dpp). Genetic and molecular studies of dpp have identified a large 3′ regulatory region that contains multiple modular enhancers which drive expression in different subdomains of imaginal discs (Masucciet al. 1990; St. Johnstonet al. 1990; Blackmanet al. 1991). Mutations that break at increasing distances from the dpp gene disrupt fewer regulatory elements, and lead to milder truncations of legs and other appendages. This analogy suggests that important Bmp5 regulatory elements might be located in the genomic region between the se30DThWb and se4CHLd mutations.
Reporter analysis identifies numerous enhancers located 3′ of the Bmp5 exons: To test whether the 3′ region contains important Bmp5 regulatory sequences, we surveyed 106 kb of 3′ genomic DNA for the ability to drive expression of a reporter gene in Bmp5 expression domains. Seven constructs (Figure 1B; clones II–VIII), each covering 14 to 20 kb, were fused to a minimal promoter driving a lacZ reporter (Kotharyet al. 1989) and microinjected into fertilized mouse eggs. Transgenic founder embryos were collected at embryonic day 15 (E15) and assayed for lacZ staining patterns. Several constructs gave reproducible patterns in different regions of the mouse embryo, including the upper part of the sternum (rostral manubrium), the genital tubercle, thyroid cartilage, and lung mesenchyme (Figure 2, A, D, G, J; see figure legend for number of transgenic embryos analyzed). Histological sections of stained and sectioned embryos show that these patterns closely resemble subsets of the normal expression pattern of the endogenous Bmp5 gene (Figure 2, B, C, E, F, H, I, K, L). Moreover, several sites correspond to specific anatomical structures that are disrupted in short ear mice (sternum, thyroid cartilage, and lung; see Figure 2M) (Green 1951, 1968).
The 3′ mutations disrupt specific parts of the complete Bmp5 expression pattern: To confirm that DNA sequences in the 3′ region are necessary for normal expression of Bmp5, we used in situ hybridization analysis to examine the expression pattern of the endogenous Bmp5 gene in embryos homozygous for the se30DThWb or se4CHLd mutations. In se30DThWb embryos, expression of Bmp5 RNA is greatly reduced in the thyroid cartilage, genital tubercle, and lung mesenchyme (Figure 3, E, K, N). In contrast, the expression of Bmp5 is either not affected, or is more mildly affected at these locations in embryos homozygous for the se4CHLd mutation (Figure 3, F, L, O). These data are consistent with the placement of the genital tubercle, thyroid cartilage, and lung enhancers between the two breakpoints (Figure 4). The appearance of Bmp5 transcripts in the lung epithelium of se30DThWb embryos is unexpected, since little expression occurs in the epithelium of wild-type lungs (Figure 3, M and N). The chromosome rearrangement in se30DThWb could remove a silencer that normally masks expression at this site or introduce a new regulatory element; or, the ectopic expression may result from altered signaling between mesenchyme lacking Bmp5 and the lung epithelium.
The effects of the se30DThWb and se4CHLd mutations are unlikely to be caused by general silencing of the Bmp5 locus. Both mutations disrupt very specific subsets of the complete Bmp5 expression pattern. For example, expression in the pinna of the external ear is more strongly affected than in the underlying middle ear, inner ear, or temporal bone elements (Figure 3, A–C; pinna vs. Tb). In the se30DThWb embryos, expression around the sternum is more affected than expression around the ribs (Figure 3 G–I; St vs. Ribs). Neither mutation has a strong effect on expression in the thyroid gland (Figure 3, D–F; Tg), the intestine (Figure 3, P–R), the nasal cartilage, or around the tips of digits (data not shown).
Here, we report a large array of 3′ regulatory sequences that drive expression of Bmp5 in specific skeletal elements or soft tissues. These control sequences are sufficient to drive expression of an exogenous reporter gene in subsets of the normal Bmp5 expression pattern. They are also essential for normal expression of Bmp5 as shown by the specific alterations in Bmp5 mRNA expression patterns seen in mice carrying different mutant breakpoints in the 3′ region. These results suggest that the overall expression pattern of the Bmp5 gene is controlled by a large number of separate regulatory elements, many of them located in the 3′ flanking region.
Remarkably, many of the elements appear to be specific for anatomical locations, rather than for particular types of differentiating tissue. For example, most skeletal structures arise from mesenchyme through a sequence that includes condensation, cartilage formation, cartilage hypertrophy, and ossification (Fell 1925). Enhancers, such as that from the collagen II gene, have previously been described that will drive the expression of genes in most differentiating cartilage structures throughout the body (Yamadaet al. 1990; Niederreitheret al. 1992). In contrast, the control elements in the Bmp5 gene drive expression at specific anatomical locations in the skeleton. For example, enhancers have been found that drive reporter gene expression only in the upper region of a single bone found at the top of the sternum (the manubrium). Moreover, the regulatory mutations in the 3′ region disrupt expression in specific subsets of skeletal structures, rather than in all the skeletal elements in which Bmp5 is normally expressed. Even expression patterns that appear to be continuous, such as Bmp5 expression in the perichondrium of ribs and sternal bands, may be composite patterns that are built from multiple independently-controlled units. In combination with previous studies showing that particular BMPs are expressed at different levels in different skeletal structures (Lyonset al. 1995; Kinget al. 1996), and are required for formation of specific anatomical traits (Kingsleyet al. 1992; Stormet al. 1994; Luoet al. 1995; Thomaset al. 1996), these results suggest that different BMP genes have accumulated highly specific control elements that induce expression in the outlines of particular skeletal features.
Bmp5 expression in soft tissues also appears to be regulated by more elements than apparent from simple RNA expression patterns. For example, Bmp5 expression has previously been observed in mesenchyme immediately underlying epithelium in the lungs, the intestine, the ureter, and the bladder (Kinget al. 1994). Sonic hedgehog is normally expressed in the overlying epithelium at some of these sites, and has previously been proposed as a possible signal that may induce BMP expression in both the lung and the intestine (Bitgood and McMahon 1995; Robertset al. 1995; Bellusciet al. 1996). Although this still may be the case, current studies have identified DNA sequences that drive reporter genes in lung but not intestinal mesenchyme, and regulatory mutations that disrupt expression in lung mesenchyme but not intestines. Thus, the similar expression of Bmp5 in mesenchyme underlying epithelium in lungs and intestines appears to be controlled by distinct cis-acting regulatory sequences.
Previous studies in flies have shown that dpp expression in the wing imaginal discs is also controlled by multiple modular regulatory elements that drive expression in subsets of what otherwise appear to be continuous stripes (Masucciet al. 1990; Blackmanet al. 1991). As in the present studies, many of these elements were originally suggested by regulatory alleles that disrupt subsets of dpp expression (Masucciet al. 1990; St. Johnstonet al. 1990). It is not clear if the regulatory elements present in these distantly related BMP genes will be similar to each other. Many surprising examples of conservation of molecular pathways have been found (Awgulewitsch and Jacobs 1992). On the other hand, dpp and Bmp5 belong to distinct subgroups within the BMP family (Kingsley 1994a), so any shared regulatory elements would have to predate the divergence of the different BMP subgroups. Moreover, many of the elements found in the dpp and Bmp5 genes drive expression in structures that probably have evolved since the divergence of invertebrates and vertebrates (wing veins, thyroid cartilage, manubrium). We have previously proposed that gene duplication in the BMP family, followed by gain and loss of specific control elements, could provide a general mechanism for altering the size, shape, or number of skeletal features in higher animals. (Kingsley 1994b). Further study of the anatomy-specific regulatory elements present in the Bmp5 locus should help determine some of the molecular mechanisms used in higher animals to control the size and shape of specific skeletal structures.
Many of the control elements identified in this study are located at a surprisingly large distance from the Bmp5 promoter. The lung mesenchyme regulatory sequences, for example, are located more than 270 kb from the Bmp5 transcription initiation site, one of the longest distances yet reported in studies of eukaryotic gene expression. Other elements may map even farther away, as suggested by the ear phenotype and changes in skeletal expression associated with the se4CHLd mutation. Although there are other published examples of distant control regions (Duncan 1987; Grosveldet al. 1987; Masucciet al. 1990; Blackmanet al. 1991; Goldhameret al. 1992) it is not clear how common such arrangements will be. In the case of Bmp5, such distant elements might not have been found without the extensive genetic and molecular resources available at the short ear locus. Several other mutations have been described in mice and humans that disrupt the function of genes located at large distances from the mutant lesion (Tonet al. 1991; Vortkampet al. 1991; Fosteret al. 1994; Belloniet al. 1996; Kluppelet al. 1997). Although some of these “position effect” alleles could be due to non-specific effects of newly apposed chromatin, it is also possible that long-range regulatory elements are disrupted by the mutations (Bedellet al. 1996). Given the long distances over which Bmp5 elements act, we think it is likely that many genes in higher animals will turn out to be controlled by regulatory sequences that act at much greater distances from the gene than is currently recognized.
We thank Maylene Wagener and Michelle Johnson for expert animal care, and members of the Kingsley and Hogness laboratories and Doug Guarnieri for helpful comments on the manuscript. This work was supported in part by an National Science Foundation predoctoral fellowship (R.J.D.); grants from the Lucille P. Markey Charitable Trust and the National Institutes of Health (D.M.K.); and the U.S. Department of Energy and the National Institute of Environmental Health Sciences (L.B.R.).
Communicating editor: N. A. Jenkins
- Received September 5, 1997.
- Accepted September 30, 1997.
- Copyright © 1998 by the Genetics Society of America