Quantitative functional interrelations within the cis-regulatory system of the S. purpuratus Endo16 gene

The cis-regulatory systems that control spatial and temporal gene expression are typically composed of subelements that function positively or negatively in different embryonic territories, or are utilized in different temporal phases of development (Kirchhamer et al., 1996). Here, we approach the mechanism by which the upstream regulatory subelements of the Endo16 gene of S. purpuratus interact, producing an integrated transcriptional output. The embryonic pattern of expression of this gene is generated by the outputs of six different subelements of the overall cis-regulatory system, which we regard as modular components of the control system. Multiple interactions with diverse transcription factors occur within each of these modules. Each module displays a particular regulatory function when linked to an expression construct, either alone or in combination with other modules, and tested by gene transfer (Yuh and Davidson, 1996). Modular organization is a property shared by many embryonic cis-regulatory systems that have been studied in detail, in Drosophila, mouse and sea urchins (reviewed by Davidson, 1994; Kirchhamer et al., 1996).

The Endo16 gene encodes a polyfunctional glycoprotein (Soltysik-Espanola et al., 1994) which in the late embryo is a secreted product of the midgut. The Endo16 gene is transcriptionally activated in late cleavage (Godin et al., 1996) and, throughout the blastula stages, it is expressed in descendants of the veg₂ lineage, which constitute the vegetal plate. Endo16 transcripts are initially observed throughout the archenteron, to which the vegetal plate gives rise by invagination. After gastrulation is complete, however, expression is silenced in the foregut and hindgut, but is stepped up in the midgut (Nocente-McGrath et al., 1989; Ransick and Davidson, 1993). Control of all of these aspects of the expression pattern is primarily transcriptional, since the CAT mRNA products of Endo16•CAT expression constructs display the same temporal and spatial expression profiles as does endogenous Endo16 mRNA (Ransick and Davidson, 1993; Yuh et al., 1994; Yuh and Davidson, 1996). The cis-regulatory system of the Endo16 gene required for the complete embryonic pattern of expression is included in a 2300 bp DNA sequence extending upstream from the transcription start site, within which we have identified more than 30 target sites for at least 13 different highly specific DNA-binding factors (Yuh et al., 1994). The disposition of the six different modular subelements (G-A) resolved in our recent functional analysis (Yuh and Davidson, 1996), and of the protein-binding sites (Yuh et al., 1994), are summarized diagrammatically in Fig. 1.

Modules G, B and A function positively. When linked to the endogenous basal promoter or the SV40 promoter, each is independently capable of causing expression in the vegetal plate and later the archenteron, though in isolation all three modules also promote ectopic expression in the territories adjacent to the vegetal plate. Module G functions only weakly by itself, but, when present in an expression construct utilizing the SV40 promoter, it synergistically boosts the otherwise low activities of modules A and/or B to a significant extent (Yuh and Davidson, 1996). Modules A and B also function synergistically in combination with one another, producing an increased level of expression throughout development, with either the endogenous or the SV40 promoter (Yuh and Davidson, 1996). Module A is largely responsible for early expression in the vegetal plate. The late rise in expression of Endo16 is due largely to module B, which by itself is capable of promoting midgut expression in the postgastrula stage embryo. Modules F, E and DC function negatively. The function of modules F and E is to preclude ectopic expression of Endo16 in the ectoderm that lies above the upper boundary of the vegetal plate. Module DC similarly prevents expression in the cells deriving from the skeletogenic progenitors, which are initially located across the lower boundary of the vegetal plate. All three of these modules are sensitive to treatment of cleavage-stage embryos with LiCl, which expands the vegetal plate and the domain of Endo16 expression at the expense of the overlying ectoderm (Ransick and Davidson, 1993). LiCl converts all three negative modules to positive function. Yuh and Davidson (1996) showed that both the negative (i.e., spatial control) functions of modules F, E and DC, and their LiCl sensitivity require the presence of module A. The function of the negative modules thus appears to be to eliminate the output of modules A, B and G in cells across the upper and lower boundaries of the vegetal plate.

In the following, we use a quantitative approach to address two specific issues by testing models of regulatory system function against temporal expression data. The first of these issues is the nature of the interactions amongst the positively acting modules, and between them and the basal transcription apparatus. These interactions are evidenced by the synergistic effects observed by Yuh and Davidson (1996) when two or more of the positive modules are physically linked in an expression construct. The second issue addressed here concerns the mode of action of the negatively acting modules. The manipulations that we describe provide strong additional evidence that these negative functions operate by interference with the positive function of module A. By focusing on the quantitative interrelations among the modules, we have been led to a hierarchical model of the Endo16 cis-regulatory system, which in turn provoked further experiments that confirmed key features of this model.

Most of the expression constructs used in this work are described by Yuh and Davidson (1996). Eggs were prepared and injected with the constructs, and CAT assays performed, exactly as described there. These constructs are shown diagrammatically in Fig. 1. Additional constructs including the ‘J_m’ and ‘(CG)₂’ mutations were generated for the present studies. J_m refers to a mutated site for the ‘J’ factor of module A (see Fig. 1), and CG to a factor that binds at a number of locations in the regulatory domain, indicated in Fig. 1 by green ovals below the line representing the DNA.

Briefly, constructs GBA(J_m)-BpCAT and BA(J_m)-BpCAT were assembled from a cloned insert that linked G, B and A to Bp, into which the mutated J target site had been introduced by PCR. The outside primers used for this purpose were the Bluescript vector forward and reverse primers. The complementary inside primers included the normal Endo16 sequence immediately upstream of the J target site (Yuh et al., 1994), with an appended sequence that replaces this site with an XbaI site:

The XbaI site served later to check the success of the construction.

GBA(J_m)-BpCAT was assembled from GBA(J_mXba) + (XbaJ_m)A-Bp + Bluescript CAT, where J_mXba and XbaJ_m represent the two parts of module A to be joined by ligation at the XbaI site. Construct BA(J_m)-BpCAT was assembled similarly, from BA(J_mXba) and (XbaJ_m)A.

Constructs A(CG)₂-SVpCAT and B(CG)₂-SVpCAT were derived from GBA(CG)₂-SVpCAT. This was generated by ligating into GBA-SVpCAT (Yuh and Davidson, 1996) a double-stranded oligonucleotide which represents the two CG factor target sites that appear between positions −64 and −109 of the Endo16 basal promoter region (Yuh et al., 1994). The sequence of this oligonucleotide is as follows (CG target sites boxed).

All the smooth curves shown were generated by means of a derivative matching algorithm (spline interpolation available in Mathcad Plus 6.0b). The calculations used to generate Figs 3-6 and 8, and Tables 1 and 2, were carried out using data shown in Fig. 2, as indicated in text.

For each model to be tested a single free parameter, λ, was used as a scale factor to match calculated functions to observed data. The value of this parameter was minimized by a homogeneous least

squares procedure. The procedure was to determine the closest possible match between an observed time course (‘target time course’) and a time course calculated by applying a mathematical operation to other observed time courses (from the same data set; in the following these are the ‘calculation time course(s)’). Thus, where d₀ represents the target time-course data set, and d_0i an individual time point in this data set; and d_c represents the calculation data set(s),

The minimum least square value of λ is given by

The root mean sequare error, ε is then calculated as

The calculation was carried out at the times occupied by data points, and only the data, not the interpolated values were used for the calculation. Values of λ and ε were reported as shown in Tables 1 and 2. Spline interpolations were imposed on the generated data following the calculations, and these curves are shown in Figs 2-7 together with an envelope representing ±ε, portrayed as fraction of maximum value. We stress that the cubic spline curves shown are merely to improve the ease with which the individual time courses can be followed by eye. These smooth curves are not meant to indicate the actual time courses, either measured or calculated, in the intervals between the measured or calculated points.

A representation of CAT enzyme production in embryos bearing CAT expression vectors would require solution of the following relations:

Here R denotes molecules of CAT mRNA per embryo; S is transcription rate for the expression vectors per embryo, i.e., molecules•min⁻¹ synthesized (more precisely, molecules of mRNA flowing into the cytoplasm•min⁻¹, if processing is efficient as is usually the case in sea urchin embryos [Cabrera et al., 1984]. This is the same as transcription rate); C is CAT enzyme molecules per embryo; k_T is translation rate, i.e., CAT protein molecules•min⁻¹•R⁻¹; and k_DC and k_DR are the first order decay rate constructs for CAT enzyme and CAT mRNA, respectively. However, Flytzanis et al. (1987) showed that the half life of CAT enzyme is only ∼40 minutes in sea urchin embryos, and that of CAT mRNA is less than or equal to this. This is a very short interval on the scale of the experiments shown in Figs 2-5, measurements for which extend over a period exceeding 50 hours. Therefore, averaged over several hours, C of equation 2 will always be proportional to S over the interval measured:

where α is a proportionality constant. The value of α approximates k_T•k_DC⁻¹•kDR⁻¹. We can take k_T=2 molecules of protein•min⁻¹• mRNA⁻¹ (Davidson, 1986); k_DC=ln2/40 min⁻¹; k_DR<k_DC (Flytzanis et al., 1987). Thus for peak values of ∼4×10⁶ molecules of CAT enzyme/embryo late in development, when there are ∼100 expressing midgut cells (Yuh and Davidson, 1996) S per nucleus would be >6 molecules of CAT mRNA•min⁻¹, a reasonable rate for a moderately active expression construct present in multiple copies, but limited by the amount of transcription factors present, as observed earlier for other expression constructs (Livant et al., 1988; Franks et al., 1990). The short half-lives of CAT mRNA and protein reduces the representation of CAT enzyme expression shown in equations 1 and 2 to the simple proportionality of equation 3. Thus we are able to carry out the operations indicated by the models to be tested directly on the data points (C_(t) values) and interpret them as immediate indicators of the rates of expression construct transcription.

Measurements of Yuh and Davidson (1996) showed that modules G, B and A of the Endo16 cis-regulatory system function positively and synergistically. Time-course data on CAT expression (Fig. 3 of Yuh and Davidson, 1996) indicated that, when associated with the Endo16 basal promoter (Bp) in the expression construct A-BpCAT, module A by itself generates a profile of expression that peaks at about 40 hours and then declines, while module B alone generates relatively low activity until after 60 hours, when the level of CAT enzyme produced by the B-BpCAT construct rises sharply (see Fig. 1 for constructs discussed here and in the following). The construct G-BpCAT produces an almost flat, low level profile of CAT expression. All three of these modules individually suffice to produce vegetal plate and gut expression, though with differing temporal and quantitative profiles of activity. However, when physically combined in the construct GBA-BpCAT, the level of CAT enzyme expression is higher than that produced by any of the individual constructs. The first series of experiments that we discuss in this paper were designed to illuminate the nature of the synergistic interrelations amongst these three positively acting upstream regulatory modules.

Time-course data used in this study are reproduced in Fig. 2. A single batch of fertilized eggs deriving from a different female was used for each set of measurements. CAT enzyme activity was determined in pools of 100 normally developing embryos at four to seven time points between 20 and 72 hours postfertilization, depending on the experiment. The smooth curves shown are interpolations between these points (see Materials and Methods). Not all embryos deriving from injected eggs develop normally. Thus, for example, injection of ∼5000-6000 eggs per experiment was required to obtain the 3500 morphologically normal embryos needed for experiments such as shown in Fig. 2A, which included seven constructs each assayed at five time points.

The approach that we followed in the initial set of experiments was to test various simple models that might provide interpretations for the synergism observed when the single modules are physically linked in expression constructs. Each model was tested by calculating the output that would be generated by the linked construct from the individually measured time courses of its constituents according to that model, and the generated values were then compared to the data obtained experimentally for that linked construct. Because of the short half life of CAT protein and mRNA (Flytzanis et al., 1987), the CAT enzyme levels at the time points measured are directly proportional to the transcription rate around that time (this argument is shown in equations 4-6 of Materials and Methods). For each model, the values at each time point of the respective activity profiles were multiplied or added, as required by the model, and the result was multiplied by a scale factor, λ (see equations 1-3 in Materials and Methods). The best value for λ was obtained by minimizing the difference between the calculated and observed data, using a least squares procedure. The value of λ, and of the root mean square error ε, were reported, and these data are summarized in Tables 1 and 2.

Time-course measurements for constructs composed of modules G, B and A tested singly and in various combinations, and linked to the endogenous Endo16 basal promoter (Bp), are shown in Fig. 2A-C. The embryos used for the data set in Fig. 2A were somewhat more active than those used for the data set in Fig. 2B, but the curves representing the activity of each construct are of similar form. Comparison with figure 3 of Yuh and Davidson (1996) show that the forms of these time courses are the same as measured earlier as well. The smooth curves shown were interpolated through the points by a derivative matching procedure (see Materials and Methods). The data of Fig. 2A and B were averaged and interpolated to generate the single set of curves shown in Fig. 2C. This averaged data set was then used for the following operations.

The first and simplest case considered was the relation between modules A and B, when they are linked in construct BA-BpCAT (dark blue, green and gray curves in Fig. 2A-C). We considered three different models, as indicated in the top section of Table 1: (i) that module A increases the output of module B by a constant synergism factor, λ. This model is symbolized ⁠, where indicates the time course generated by BA-BpCAT; (ii) that the activity of BA-BpCAT is

the sum of the activities of B-BpCAT and A-BpCAT, multiplied by the synergism scale factor λ, symbolized (iii) that the activity of BA-BpCAT is the product of the activities of B-BpCAT, A-BpCAT and the synergism factor λ, symbolized The result is in this case clear and obvious: produces an excellent fit to the observed data, as shown in Fig. 3B, while the other two models produce very poor fits, with relatively large root mean square (RMS) errors (Table 1). The value of the scale factor for is about 4. Thus we may conclude that when linked together in BA-BpCAT module A simply amplifies about four-fold the quantitative output of module B over the whole time course of the experiment. The converse model, that B amplifies the output of A by a constant factor, produces a worse fit than any of the others listed in the first portion of Table 1.

Fig. 4 shows the best solution that we found for the output of the combined fusion GBA-BpCAT. This is given by the model i.e., that the activity of this construct is approximated by the activity of the construct GB-BpCAT amplified throughout by a constant scale factor of about 3 (Table 1). In other words, exactly as in the construct BA-BpCAT, module A functions in a particularly simple way, amplifying the output several fold by a constant factor. Since the time course of G-BpCAT is almost a constant itself (Fig. 2C), it is not possible to resolve the form of the contribution of module G per se, i.e., to separate its contribution out from the term λ. The main synergism obtained in the GBA-BpCAT construct is between modules A and B. In fact, Fig. 2C shows that the activity of BA-BpCAT is actually higher later in development than that of GBA-BpCAT, though overall these two time functions are quite similar. Table 1 shows that the models and ⁠, in which λ<1, give significantly worse fits than does and the forms of these model curves are not coherent with the target data; nor is any model in which the output of GBA-BpCAT is conceived as a product of the time courses of module A and other modules useful, e.g., or comes closest, when module G acts as a depressant by a factor of ∼0.6 (Table 1). This cannot be excluded even though the RMS error is twice that for but is not an attractive interpretation, since we know that G-BpCAT in fact functions as a positive expression construct that promotes reporter gene transcription in gut and that module G acts as a synergistic element in various SV40 promoter (SVp) constructs tested by Yuh and Davidson (1996), rather than as a depressant. Nor can the activity of GBA-Bp•CAT be described as the sum of the activities of the component modules; Table 1 shows that also gives a poor fit.

We conclude that module A probably acts the same in the context of GBA-BpCAT as in the context of BA-BpCAT. It functions to increase the output of its active partners by a constant factor of three to four. A-BpCAT is transcribed in vegetal plate and archenteron (as well as ectopically in the ectoderm and skeletogenic mesenchyme; Yuh and Davidson, 1996), beginning in blastula stage. This activity is highest at 40 hours. It then declines and becomes difficult to detect late in development. These kinetics are obviously distinct from the constant three- to four-fold synergistic amplification attributable to module A in the present experiments. Therefore, the synergistic function seen in combinations of modules A and B might depend on a different interaction within module A than that which causes vegetal plate expression and which peaks at 40 hours. We know (unpublished experiments) that the spatial and temporal expression function mediated by module A when it is isolated in the A-BpCAT construct, depend on interactions at the specific target site for the factor ‘J’ of module A (see Fig. 1, and Yuh et al., 1994). Thus the synergistic amplification function of module A may depend on one or more of the four other interactions that occur within this module. In Fig. 1 (top), these are symbolized as colored symbols (green, purple, pink) beneath the line representing the DNA of module A. Target sites for each of these interactions appear in several different regions of the Endo16 cis-regulatory domain, and for two of them, viz the SpGCF1 sites and those for the factor symbolized by the green ovals, in the Bp region of the gene as well.

As discussed in detail by Yuh and Davidson (1996), the SV40 early region basal promoter (SVp) generates qualitatively similar but quantitatively much more feeble temporal patterns of activity when linked singly to G, B or A modules than is observed when these same modules are linked singly to the endogenous basal promoter. This can be seen again in Fig. 2D (compare Fig. 2C). However, when these modules are linked together a very large synergistic effect is observed, and thus GBA-SVpCAT is about as active as is GBA-BpCAT. Similar observations are reported by Yuh and Davidson (1996), who also found a very large synergistic effect when GA-SVpCAT, GB-SVpCAT and BA-SVpCAT were compared to G-SVpCAT, B-SVpCAT or A-SVpCAT. These combined SVp constructs all function at about the same level as do the corresponding Bp constructs. Two of the six Sp1 sites included in SVp happen to be strong sites for the sea urchin SpGCF1 factor, while Bp contains, in addition to two SpGCF1 sites, two sites for another protein that also binds within the module A sequence, as well as elsewhere in the Endo16 cis-regulatory domain (symbolized by the green ovals in Fig. 1). Yuh and Davidson (1996) supposed that, in some way, synergistic interactions amongst modules G, B and A in the combined SVp constructs substitute for interactions mediated by the additional proteins binding within the endogenous Bp region. In any case, the intermodule synergism observed in SVp constructs is, in quantitative terms, about 10× that seen with Bp constructs, comparing the output of single module constructs to that of the GBA constructs.

The results shown in Table 2 indicate that the form of the models for GBA-SVpCAT that provide reasonable approximations, using the time courses for G-SVpCAT, B-SVpCAT and A-SVpCAT, are essentially different from those shown in Table 1 and Figs 3 and 4 for the Bp constructs. The best solution, shown in Fig. 5A, is given by the model and almost as good is (Table 2). This is reasonable, since the G-SVpCAT time course is low and almost flat; the only effect of including the G-SVpCAT time course is to decrease the value of λ (per function) from about 5.2-fold to 3.7-fold amplification. The main point is that, in these models, the time courses of the individual A and B module constructs are multiplied by one another. The levels of output of each module that are observed when they are tested singly, presumably reflect directly the occupancy of their transcription factor target sites through time; the results of Table 2 show that, for each module, these levels are proportional through time to its synergistic function when all three are physically combined in an SVp construct. The model that is effectual with the GBA-BpCAT target, i.e., produces a terrible fit when applied to the SVp time courses, as do the other models of the same form (since the time courses for G and G_s are essentially flat, GB•λ is of the same form as B•λ). This is shown in Fig. 5B. Nor do models dependent on the G-SVpCAT time functions work and addition rather than multiplication of the time courses of the individual SVp constructs also fails (Table 2). We consider in Discussion the significance of the conclusion that use of SVp requires multiplication of the individual A and B module time courses.

As summarized above, Yuh and Davidson (1996) found that there are three negatively acting cis-regulatory modules, each functioning to exclude expression within the territories that abut the boundaries of the vegetal plate: their functions are required because all three of the positive modules G, B and A are active not only in the vegetal plate, but also in the adjacent regions, as shown by the occurrence of ectopic as well as vegetal plate expression in embryos bearing A-BpCAT, B-BpCAT or G-BpCAT, and in embryos bearing GBA-BpCAT (Yuh and Davidson, 1996). The three negative modules are alike in several respects: (i) none has appreciable transcription-stimulating activity on its own;

(ii) each decreases expression about two-fold when associated with the combined GBA activator and more than two-fold when linked to module A alone; (iii) none produces any repressive effects when linked to modules B or G alone; (iv) all are converted into positively acting elements that increase expression by LiCl treatment of the embryos;

(v) module A is required to confer LiCl sensitivity on all three negative modules just as it is required to confer repressive activity in untreated embryos. Yuh and Davidson (1996) concluded that, in the skeletogenic and adjacent ectoderm territories, the negative modules act through module A.

To further explore the mode of function of the three negative modules, we sought to extract from the data shown in Fig. 2E the time courses of their repressive activities. The repressive activity is expressed in different cells than is the positive activity of constructs that include both negative and positive modules. Therefore, we subtracted from the time-course data for A-BpCAT shown in Fig. 2E the time-course data for DCA-BpCAT, FA-BpCAT and EA-BpCAT. Note that the time course measured for A-BpCAT in this series is very similar in form to that shown for A-BpCAT in Fig. 2A-D (and in Yuh and Davidson, 1996). The results of these subtractions are shown in Fig. 6. Here we see that the three different curves representing the repressive activities of modules DC, F and E are remarkably similar to one another, and to the time course of positive module A function itself, which is superimposed in Fig. 6. This result is consistent with the conclusion that negative module function is directly dependent on the activity of module A, i.e., that these elements function by means of interactions with module A. Fig. 2F shows that when in the context of all three positive modules each of the negatively acting modules depresses the overall level of activity during the 40-60 hour period when module A alone is most active. Where X represents F, E or DC, the time courses of GXBA-BpCAT are generally similar to those of XA-BpCAT (Fig. 2E,F). Thus we believe that the relations that hold in the context of the complete positive regulatory system are similar to those which can be seen clearly in Fig. 6 for the interaction of the negative modules with module A alone. However, a much greater density of data would be required to test this point directly.

Several mutations of expression constructs utilized for the measurements shown in Fig. 2 were built, in order to test two key functional inferences. The first of these was the inference that one of the other factors binding in module A, rather than the J transcription factor, is responsible for the constant synergistic effect through developmental time of module A on module B. To test this, we mutated the J target site (J_m) and constructed GBA(J_m)-BpCAT and BA(J_m)-BpCAT (see Materials and Methods). The mutation that was inserted totally destroys binding of natural J factor that had been partially purified by affinity chromatography (data not shown). The activity of the J_m constructs was compared with that of GBA-BpCAT, BA-BpCAT and B-BpCAT in two independent time-course experiments, the average of which is shown in Fig. 2G. The main result of this experiment is evident by inspection. BA(J_m)-BpCAT and BA-BpCAT are expressed almost identically and display an almost identical amplification with respect to B-BpCAT. This demonstrates that module A elements other than the factor J target site are indeed responsible for the amplification function of module A or module B. GBA-BpCAT is expressed a little more actively than is GBA(J_m)-BpCAT. Thus, as Table 1 shows, a calculation of the model •λ gives an excellent fit with λ only about 1.4. This modest linear difference might indicate a mild synergistic function on the part of the J factor, which is observed only when module G is present; i.e., it is possible that J factor amplifies the positive output of module G, but we doubt the significance of this result since as shown in Fig. 2A-C GBA-BpCAT is only marginally more active than is BA-BpCAT. What is certain is that the much stronger synergism observed between modules B and A does not depend on factor J.

Another inference that we challenged concerns the difference between the SV40 and the Endo16 promoter elements used in these studies. The experiments of Table 2 showed that the positive modules G, B and A function independently and multiplicatively when combined with SVp, but that their synergism is linear and not multiplicative when the same elements are combined with Bp. Furthermore, as shown by Yuh and Davidson (1996) and in Fig. 2D of this paper, a major difference between SVp and Bp is the relatively very low activity of SVp when driven by module A or module B alone. We inferred above that these differences could be due to the presence of two target sites within the Endo16 Bp fragment for the yet unidentified factor provisionally called ‘CG’ (Yuh et al., 1994), which is indicated by green ovals in Fig. 1.

To test the idea that the difference between the endogenous Endo16 Bp and SVp is due at least in part to the CG target site sequences, we sought to convert the SV40 promoter into a promoter that would behave like the Endo16 Bp by inserting CG target sites just upstream of SVp in several of our expression constructs. Thus, as described in Materials and Methods, we created A(CG)₂-SVpCAT and B(CG)₂-SVpCAT, and compared their activities to A-BpCAT, B-BpCAT, A-SVpCAT, B-SVpCAT and GBA-SVpCAT. Results are illustrated in Fig. 2H (except for GBA-SVpCAT, which was measured in a different experiment) and are listed in Table 3. These experiments also afforded an independent check, using entirely different measurements, on the behavior of GBA-SVpCAT.

Table 3 shows that, almost exactly as in the experiment given in the first line of Table2 and illustrated in Fig. 2D, the model again provides the best fit, with λ (per function)=7.8 and ε (% max)=3%. The main conclusion from Fig. 2D is also evident by inspection. With respect to module A, when the CG sites are added to SVp, the activity of the newly created promoter is now almost identical to that of Bp; i.e., A-BpCAT≈A(CG)₂-SVpCAT, and both are much more active than is A-SVpCAT. Table 3 indicates that the linear amplification factor λ is, as expected, essentially the same for A=A_s•λ as for A_s(CG)₂=A_s•λ (where as above A represents A-BpCAT and A_s represents A-SVpCAT). These solutions are shown for the data set in Fig. 2H in Fig. 7A and B. A second data set, almost the same as that in Fig. 2H, gave the same results (data not shown). We conclude that the CG target sites suffice to endow the SV40 promoter with the same linear ability to amplify the output of module A, as observed with the endogenous Bp element. Therefore, the CG DNA-binding factor probably carries out this function.

Fig. 2H also shows that the CG target sites fail completely to amplify the activity of module B. B(CG)₂-SVpCAT is expressed almost identically with B-SVpCAT. The amplification function mediated by the CG sites thus depends on interaction specifically with elements of the contiguous region, module A. We wonder whether it is relevant that module A also contains CG sites, while module B lacks them (Fig. 1A); perhaps the CG-binding factor works by forming homomultimers, as does SpGCF1 (Zeller et al., 1995a). The source of the difference in activity between B-BpCAT and B-SVpCAT remains unresolved. This could depend on the SpGCF1 sites also present in Bp (Fig. 1) (although such an hypothesis would require that the Sp1 sites of SVp that resemble SpGCF1 sites [Yuh and Davidson, 1996] are for some reason inadequate).

The three positively acting modules of the Endo16 cis-regulatory system, G, B and A, respond to different transcription factors. We know this from direct determination of the minimum diversity of DNA-binding proteins that interact with target sites within these three regions (Yuh et al., 1994; see summary in Fig. 1). The only sites shared amongst modules G, B and A are those at which the ubiquitous SpGCF1 factor binds (Zeller et al., 1995a,b). This factor may act by multimerizing once it is bound, thus promoting or stabilizing physical interactions amongst distant regions of the regulatory system. When tested in isolation, i.e., in the constructs G-BpCAT, B-BpCAT and A-BpCAT each of these modules promotes vegetal plate expression as well as ectopic expression in the adjacent ectoderm and skeletogenic territories but not in the oral and aboral ectoderm territories that derive from cells above the horizontal 3rd cleavage plane (Yuh and Davidson, 1996). These three positive modules display distinct quantitative time courses of activity during embryonic development (Yuh and Davidson, 1996; Fig. 2 of this paper). We presume that their individual time courses reflect the effective occupancy of at least the key target sites within each module, and that these occupancies differ over time due to the changing availability of active forms of the respective transcription factors that bind at these sites. With respect to spatial expression, in unpublished experiments, we have shown that the key transcription factor in module A is that symbolized as ‘J’ in Fig. 1; and the key factor in module B is that symbolized as ‘I’ in Fig. 1. When combined with the Endo16 Bp, oligonucleotide representations of these target sites alone suffice to reproduce the same spatial pattern of expression as generated by modules A and B, respectively. They also reproduce the time courses observed for A-BpCAT and B-BpCAT, though at significantly reduced quantitative levels. The experiment shown in Fig. 2G, however, shows that even when the factor J site is destroyed by mutation, the characteristic time course generated by module A is preserved. Therefore both factor J and at least one other of the transcription factors that bind within module A are presumably bound at peak concentrations at 40 hours.

A focus of this study is the synergism displayed when the positive modules are physically linked in expression constructs. In these constructs the 5′ to 3′ order in which modules G, B and A naturally occur has been preserved. We find that in none of the cases that we have analyzed is the output of the physically linked modules the sum of their separate outputs; rather the output of the combined modules is always significantly greater than the sum of the individual activities. Furthermore, we demonstrate (Figs 3, 4) that module A produces a linear amplification of the output of modules B and GB when linked together with these, rather than the rising and descending patterns displayed by module A when tested in isolation, with either Bp or SVp. The linear amplification also is observed when the target site for Factor J is mutated (Table 3), so this synergistic amplification function of module A depends on one of the other factors binding in module A. This result was already implied by the differences between the constant value of the synergism factor λ and the peaked time course produced by constructs driven by factor J.

When linked to the stripped down heterologous SV40 basal promoter, the linear synergism displayed by module A in Bp constructs is no longer observed. The output of GBA-SVpCAT is approximated instead by the products of the outputs of the constituent modules, and a much larger amplification factor as well. This is shown in Tables 2 and 3, and illustrated in Figs 2D and 5. The difference in behavior illuminates the function of sequences present in the endogenous Endo16 Bp fragment. We think that the most likely interpretation is that, in the GBA-SVpCAT construct, outputs of the three modules are integrated at the basal transcription complex that binds within SVp (see legend to Fig. 8). That is, the individual upstream modules individually interact with the basal transcription apparatus, resulting in the relatively large amplification of activity observed, i.e., in comparison to A-SVpCAT, B-SVpCAT and G-SVpCAT individually.

As pointed out by Yuh and Davidson (1996), the term ‘basal promoter’ is probably a misnomer for the Endo16 cis-regulatory fragment so denoted (see Fig. 1), since the Bp DNA fragment contains multiple target sites for the CG DNA-binding factor and for SpGCF1 (see Fig. 1), in addition to the sequences that presumably support assembly of the TBP-holoenzyme complex. These additional interactions account for the fact that the endogenous Bp is about 10-fold more active when combined with the single modules than is SVp (Yuh and Davidson, 1996; Fig. 2 of this paper). Insertion of CG target sites in the SV40 promoter converts its activity to that of Bp with respect to module A, as shown in Table 3 and Fig. 7.

Thus in the simplified context of the GBA-SVpCAT construct, we believe that G, B and A regulatory modules each interact with the basal transcription apparatus. In the natural system, the output of modules G, B and A depends instead on a sequence of interactions amongst upstream elements, including those that occur upstream of the start site in the endogenous ‘basal’ promoter. A similar theme obtains with respect to the negative spatial control functions modulated by the Endo16 cis-regulatory system, in that here again essential interactions occur amongst upstream regulatory elements.

We summarized above the evidence of Yuh and Davidson (1996), showing that the repression of Endo16 transcription in skeletogenic lineages by the DC module, and in ectoderm lineages by F and E modules, all operate by means of interactions with module A. None of the various functions identified for these three modules were found to be executed in constructs lacking module A, and the other positive modules, B and G, cannot substitute for A. An additional item of evidence is presented in Fig. 6 of this paper. Here we see that the time course of repressive DC, F and E function almost perfectly parallels the time course of module A positive function. It follows from this evidence, taken together with that of Yuh and Davidson (1996), that in the cells where they are responsible for shutting off Endo16 expression, the negative modules function by means of interaction with module A. Since they do not work when placed with module B in the absence of A (Yuh and Davidson, 1996), they do not work by direct repressive interaction with the basal transcription apparatus. Yet modules B and G, as well as A, produce ectopic expression if these negative control functions are lacking (Yuh and Davidson, 1996). This means that the key spatial regulators that determine the positive function of modules G, B and A are all present and active not only in the vegetal plate but also in the surrounding domains of the early embryo as well. Therefore, the negative modules cannot function exclusively by interfering with the positive activity of module A per se. Instead module A seems to mediate the repressive output of the negative modules, by acting as a switch that permits or precludes expression of the other positive modules. When this switch is closed by the negative modules, none of the G-, B- or A-positive modules can act.

A qualitative model of the Endo16 cis-regulatory system

A hierarchical intermodular organization emerges from these studies of the Endo16 cis-regulatory system. In Fig. 8A, we combine the results obtained here with those of Yuh and Davidson (1996). The diagram represents the qualitative inter-relations amongst the positive (hatched) and negative (open) modules that constitute the complete system. As described in more detail in the legend to Fig. 8, the modular functions presumed in this diagram are as determined by Yuh and Davidson (1996) with a number of additions from this work.

Module A serves as the integrating ‘processor’ of the whole system. As shown by Yuh and Davidson (1996), module A is required for function of the negative modules. We know from a series of experiments on module A mutations (unpublished data) that the module A site, which is required for the LiCl response of modules F, E and DC, is adjacent to that at which factor J binds, but is separate from that site. This adjacent site is therefore (Yuh and Davidson, 1996) probably also required for the module A function of mediating the spatial repression executed by the three negative modules. Another function of module A is to act as a 3- to 4-fold linear booster of the outputs of the other positive modules, i.e., B or GB (Table 1). By comparison with the results obtained with the SV40 constructs, modules B and G probably do not act by directly contacting the basal transcription apparatus individually, but instead interact with some element(s) of module A, which amplifies their output. However, this function does not require participation of the J factor of module A, as shown in the experiments with constructs bearing a J target site, summarized in Table 3. A third function of module A, established by Yuh and Davidson (1996), is to generate the early vegetal expression pattern, and for this factor J is required (unpublished data). A fourth function of module A is to communicate with the basal promoter. We show here that the CG factor target sites of the ‘basal’ promoter fragment are also required for the linear, modestly synergistic activity of the positive modules observed for the Bp constructs (Table 3; Fig. 7). In their absence, the SV40 promoter constructs containing single positive Endo16 modules function poorly, and when the positive modules are combined, they synergize multiplicatively. We interpret this to mean that, in the SV40 constructs, each module interacts directly with the basal transcription apparatus, as indicated in Fig. 8B. However, the CG factors interact only with module A since, when linked instead to module B in the context of the SV40 promoter, they have no effect (Fig. 2B; Fig. 7). One possibility is that the CG factors that bind within module A mediate the interaction with the CG target sites in the basal promoter region, by means of a homotypic interaction.

In summary, we believe that the relationships uncovered in this work display interactions within the Endo16 cis-regulatory domain that are required for both its spatial and temporal developmental functions. The proximal region, module A, appears to play a special role in the integration of both positive and negative input from all the modules further upstream. Perhaps it alone communicates with the basal transcription apparatus as suggested in Fig. 8A. An experimental caution, which illuminates the significance of the endogenous upstream interactions, is afforded by the SVp experiments. These show that replacement of the endogenous proximal sequence by the elemental SV40 heterologous promoter easily provokes a default set of interactions, that are different from the normal interactions and that obscure the subtle hierarchical organization of the endogenous regulatory system.

The authors gratefully acknowledge perspicacious and detailed reviews of the manuscript by Drs Roy J. Britten, Scott E. Fraser, James A. Coffman and Maria Arnone. It is a pleasure also to acknowledge the skillful technical assistance of Ms. Jessica Chang, then a Caltech undergraduate. This research was carried out with the support of the Stowers Institute for Medical Research, the Caltech Beckman Institute and with an NIH Grant (HD-05753).