CROSSTALK BETWEEN THE PAIRED DOMAIN AND THE HOMEODOMAIN OF PAX 3 : DNA BINDING BY EACH DOMAIN CAUSES A STRUCTURAL CHANGE IN THE OTHER DOMAIN , SUPPORTING INTERDEPENDENCE FOR DNA BINDING

This study was supported by a research grant to P. G. from the Canadian Institutes for Health Research (CIHR) of Canada. P. G. is supported by a Distinguished Scientist Salary Award from CIHR. To whom correspondence should be addressed: Philippe Gros, Ph. D., Department of Biochemistry McGill University, 3655 Sir William Osler Promenade, Montreal, QC, Canada, H3G-1Y6. Phone: 514-398-7291; FAX: 514-398-2603; Email: sapuzz@po-box.mcgill.ca, philippe.gros@mcgill.ca Running Title: Protease Sensitivity Studies of Pax3 JBC Papers in Press. Published on May 17, 2004 as Manuscript M402949200


INTRODUCTION
Pax3 is a member of a family of 9 transcription factors (1), defined by a DNA binding module, the Paired domain (PD), that was first identified in the Drosophila protein Paired (Prd) (2). Pax proteins play critical roles during normal embryonic development, and inactivating mutations cause major defects in development of skeleton, muscles, nervous system, eyes, kidneys, and the immune system (3)(4)(5)(6)(7)(8)(9)(10)(11)(12). Pax3 is expressed in developing somites, neural tube as well as neural crest cell derivatives and plays a role in the proliferation, migration and differentiation of cells involved in neurogenesis and myogenesis (13,14). The Pax3 mouse mutant (splotch, Sp) displays spina bifida and exencephaly, and lacks limb muscles; likewise, mutations in human PAX3 cause Waardenburg syndrome (WS), a condition characterized by 6 G42R mutation in the PD which abrogates DNA binding by the PD, but also impairs DNA binding by the HD. Deletion of helix2 of the PAI subdomain in the context of the Sp d mutation has been shown to restore HD DNA binding (36). Studies in chimeric PAX3 proteins have shown that the PD can modulate DNA binding specificity and dimerization potential of heterologous HDs (37). On the other hand, a mutant PAX3 variant from a WS patient bearing a mutation at position 53 of the HD (R53G) shows not only loss of DNA binding by the HD but also by the PD (38). More recent biochemical studies by cysteine scanning mutagenesis, and sitespecific modification of single cysteine mutants with sulfhydryl reagents have shown that modification of a single cysteine in the PD (Cys82) disables DNA binding by the PD but also by the HD (39). Conversely, modification of a single cysteine at position 263 of the HD (V263C) of the HD abrogates DNA binding by both domains (39).
The mechanistic basis of this functional interdependence, including the protein subdomains involved, remains poorly understood but is likely to be relevant for target site selection by Pax3 in vivo. One plausible mechanism is that DNA binding by one or both of the DNA binding sites of Pax3 causes conformational changes at or near the other binding site to alter its properties. Thus, we wanted to get insight into the conformations adopted by the Pax3 protein when DNA-free and when bound to PD or HD targets. A number of physico-chemical approaches have been used to monitor the effect of substrate binding on protein conformation, including differential immunoreactivity with specific antibodies (40), tryptophan fluorescence 7 fragments by epitope mapping with specific antibodies and/or peptide sequencing (53). Another implementation of this method involves creating recombinant proteins bearing single heterologous protease cleavage sites (such as Factor Xa) inserted at pre-determined positions in individual mutants. Proteolytic products can be identified using antibodies against antigenic epitope also engineered at convenient positions. In this approach, conformational changes can be studied in a set of recombinant proteins which structural and functional integrity has been ascertained.

MATERIALS AND METHODS
Mutagenesis. The construction of the pMT2 expression plasmid containing the entire protein-encoding region of wildtype Pax3 cDNA has been previously described (39). This pMT2/Pax3 construct encodes for all 479 amino acids of the murine Q+ isoform of Pax3 (54). Factor Xa cleavage sites (IE/DGR; see Table 1) were introduced at different positions in 9 fashion and were introduced in pMT2. Mutants Pax3Xa66, Pax3Xa172 and Pax3Xa189 were created by insertion mutagenesis. For this, pMT2/ Myc-Pax3-HA plasmid was digested with BsmI, XbaI or ClaI and single or multiple Xa cleavage sites were introduced using double stranded oligonucleotides with cohesive ends (Tables 1 and 3). These oligonucleotides have a sequence just long enough to encode for one or a few Xa protease sites when placed in frame with the rest of the Pax3 encoding region. For mutants Pax3Xa100 and Pax3Xa216, the Pax3 cDNA was modified to introduce unique KasI and AflII sites at nucleotide positions 595 and 940, respectively, using mutagenic primers listed in Table 3. These modified Pax3 cDNAs were subcloned into the EcoRI site of pBluescript (lacks KasI, AflII sites), and one or several factor Xa sites were independently introduced at the KasI or AflII sites by insertion mutagenesis to create mutants Xa100(2), Xa100(4), Xa216(1) and Xa216(2) (Tables 1 and 3). In all cases, the presence of the factor Xa mutations and the integrity of the rest of the Pax3 sequences were verified by nucleotide sequencing. The accessibility of restriction sites used for cloning was verified by restriction enzyme fragmentation.
Expression and Detection of Pax3Mutants. The expression plasmids were used to transiently transfect COS7 Monkey cells. One million cells were plated in Dulbecco's modified Eagle medium (DMEM) containing 10% fetal bovine serum and were transfected by the calcium phosphate co-precipitation method using 15 µg of plasmid DNA doubly purified by ultracentrifugation on cesium chloride density gradients. Calcium-DNA precipitates were placed onto the cells for 5 h and then treated with HBS (0.14 M NaCl, 5mM KCl, 0.75 mM Na 2 HPO 4 , 6mM dextrose, 25mM HEPES, pH 7.05) containing 15% glycerol for 1 min. The cells were then washed and placed in complete DMEM. Whole cell extracts were prepared 24 h following by guest on September 1, 2017 http://www.jbc.org/ Downloaded from glycerol shock by sonication in a buffer containing 20mM HEPES (pH 7.6), 0.15 M NaCl, 0.5 mM DTT, 0.2 mM EDTA, 0.2 mM EGTA and a cocktail of protease inhibitors: aprotinin, pepstatin, and leupeptin used at 1mg/ml, and phenylmethysulfonyl fluoride used at 1mM. These extracts were stored frozen at -70°C until use. To assess Pax3 mutant protein expression and stability, aliquots of whole cell extracts were analyzed by electrophoresis on acrylamidecontaining SDS gels (SDS-PAGE), followed by electrotransfer onto nitrocellulose membranes and immunoblotting. Immunodetection was performed with mouse monoclonal anti-HA antibody (BabCO, Berkeley, CA) at a dilution of 1:1000 and visualized by enhanced chemiluminescence using a sheep anti-mouse horseradish peroxidase conjugated secondary antibody (Amersham). Following anti-HA probing the membranes were submerged in stripping buffer (100mM 2-mercaptoethanool, 2% sodium dodecyl sulphate, 62.5mM Tris-HCl pH6.7) and incubated at 50ºC for 30 minutes. The membranes were then washed with TBST buffer (10mM Tris-HCl pH8, 150mmNaCl, 0.1% Tween20) at room temperature. Following blocking, the membranes were probed with mouse monoclonal anti-Myc antibody (BabCO, Berkeley, CA) at a dilution of 1:1000 and visualized by enhanced chemiluminescence using a sheep anti-mouse horseradish peroxidase conjugated secondary antibody (Amersham).
Electrophoretic Mobility Shift Assay. Electrophoretic mobility shift assays were performed as previously described (39). Each protein:DNA binding reaction was carried out using approximately 10µg of total cell extracts from transiently transfected COS-7 monkey cells, and 10fmol (0.06µCi) of radioactively labeled double stranded oligonucleotides containing either PD or HD recognition sites. The final concentration of labeled oligonucleotide in the binding reaction was 0.5 nM. Whole cell extracts were incubated with 32 P-labeled PD specific probes in a volume of 20µl containing 10mM Tris-HCl (pH 7.5), 50mM KCl, 1mM DTT, 2mM spermidine, 2mg/ml BSA and 10% glycerol. Whole cell extracts were also incubated with 32 P-labeled HD specific probes in a volume of 20µl containing 10mM Tris-HCl (pH 7.5), 50mM NaCl, 1mM DTT, 2mM MgCl 2 , 1mM EDTA and 5% glycerol. To reduce non-specific binding, 1 µg of poly(dI-dC)poly(dI-dC) was included in binding studies with PD specific probes, while 2µg of heat-inactivated salmon sperm DNA was added to binding reactions involving HD specific probes. Following a 30 minute incubation at room temperature, samples were electrophoresed at Binding reactions done with PD probes (and with a non-specific oligo) contained 10mM Tris-HCl (pH 7.5), 50mM KCl, 1mM DTT, 2mM spermidine, 2mg/ml BSA, 10% glycerol and 1 µg of poly(dI-dC)poly(dI-dC). Binding reactions done with HD probes (and with a non-specific oligo) contained 10mM Tris-HCl (pH 7.5), 50mM NaCl, 1mM DTT, 2mM MgCl 2 , 1mM EDTA, 5% glycerol and 2µg of heat-inactivated salmon sperm DNA. Various concentrations of Factor Xa protease, ranging from 0 to 200ng Xa protease/ µg whole cell extract, were then added to the reaction mixture, and proteolytic cleavage was allowed to take place for 15 minutes at 20 o C. The reaction was stopped by addition of 10µl of Laemmli sample buffer. For both assays the proteolytic degradation products were separated by SDS-PAGE on 12% polyacrylamide gels, followed by transfer onto nitrocellulose membranes. Immunodetection of Pax3 products were carried out using anti-HA antibody followed by anti-Myc antibody as described above. Films were used to perform densitometry studies to quantify the amount of chemiluminescence using a Fuji LAS -1000.  (Table I) to minimize adverse structural changes possibly impairing DNA binding. Targeted in this group of 6 mutants were helices 1 ( 1, Xa55), (Xa252) and the N-terminus of helix 3 (Xa259) of the HD (Table 1, Fig. 1). Secondly, and to maximize accessibility to proteolytic cleavage, several solvent-exposed, and less conserved linker segments were also targeted for insertion of 1 or several Xa sites (Table 1, Fig. 1).

Construction of Pax3 Mutants Bearing Factor Xa Protease Cleavage Sites
Targeted in this group of 10 mutants were the linker separating helix 1 and 2 of the PD (Xa66, 1 and 2 sites), the segment linking the first and second HTH motifs of the PD (Xa100, 2 and 4 sites), and the fragment separating the PD and the HD (Xa172, 1 and 3 sites; Xa189, 1 and 2 sites; Xa216, 1 and 2 sites). Multiple Xa sites were inserted at individual locations not only to maximize accessibility to protease fragmentation, but also to provide validation of observed effects.

DNA Binding Properties of Pax3Xa Mutants
Wild type Pax3 along with the various Pax3Xa mutants were introduced in the pMT2 expression plasmid, followed by transient transfection into COS-7 Monkey cells.
Immunoblotting of whole cell extracts with either anti-HA or anti-cMyc monoclonal antibodies indicate similar stability and comparable levels of expression of all mutants in COS-7 cells, with the notable exception of mutants Xa252 and Xa259 (Fig. 2). Reduced levels of expression for Xa252 and Xa259 were noted in multiple transfections and for independent DNA preparations suggesting that mutations at these 2 positions in the HD may alter protein folding, or processing possibly reducing half-life. The effect of introducing Factor Xa sites on DNA binding properties of the PD and HD of Pax3 in the various mutants was examined by electromobility shift assays (EMSA). DNA binding by the PD was examined using oligonucleotide probes P3OPT (55,56) and P6CON (27), previously shown to reveal binding determinants present in both the amino (PAI) and carboxy (RED) subdomains of the PD (Fig. 2). Mutants at position 55 (Xa55), 66 (Xa66(1,2)) and 71 (Xa71) were found to be severely impaired for DNA binding to P3OPT and P6CON highlighting the critical role of the N-terminal HTH domain (PAI) for DNA binding by the PD. HD Xa mutants Xa252 and Xa259 also appeared compromised for DNA binding by the PD; however, interpretation of DNA binding results for these mutants were complicated by their low level of expression in COS-7 cells. The effect of inserting Factor Xa sites on DNA binding properties of the HD were evaluated using a target sequence (P2) containing the sequence TAAT(N) 2 TAAT previously shown to support cooperative dimerization of Pax3 (32). In addition, an oligonucleotide containing half of this sequence (half site, P1/2) and revealing monomeric Pax3 binding by the HD was used (32). Results shown in Fig

Accessibility of Inserted Xa Sites to Proteolytic Cleavage
The accessibility of the inserted factor Xa sites to proteolytic cleavage was investigated  Table 2. Under these conditions, wild type Pax3 (WT) was almost completely resistant to factor Xa cleavage, with the full length 55kDa immunoreactive species being the prominent band at all time points (empty arrowhead; Figs. 4 and 5). Additional minor bands were detected either prior to addition of the protease (35kDa, 0 min) or at very late time points (30kDa, 15kDa, 90-180 min). Although these immunoreactive fragments are derived from the full length protein, their appearance is likely caused by non-specific cleavage by factor Xa at sub-optimal sites, or cleavage at specific sites by additional proteases contaminating either the cell extract or the commercial factor Xa preparations. For reasons discussed below (see Discussion), these fragments were not considered in our analysis. Stripping the blot and reprobing with anti-cMyc antibody (Fig. 5) confirmed the resistance of WT Pax3 to digestion by factor Xa under conditions tested. Analysis of mutants Xa114, Xa131, Xa172(1/3), and Xa189 (1/2) showed results similar to WT Pax3, and indicated almost complete resistance of these proteins to digestion by factor Xa (supplementary data, Figure S1), with little if any predicted proteolytic fragments immunoreactive with the HA antibody detected even after 180 min of incubation. Similar results were obtained using the anti-cMyc antibody to analyze digestion products (supplementary data, Figure S2). These results indicate that the C-terminal RED subdomain and the linker domain immediately downstream the PD are probably not solvent exposed and assume a compact conformation under the conditions tested (Fig. 3). In contrast, mutant Xa100(2) yielded the expected 44kDa C-terminal HA-immunoreactive cleavage product upon incubation with factor Xa (filled arrowhead, Fig. 4); The 44kDa HA fragment was abundant at the earliest time point tested (2 min) and digestion was largely completed by 10 mins. Similar rapid cleavage of Xa100 (2), as demonstrated by disappearance of the full length 55kDa protein, was verified by immunoblotting with the anti-cMyc antibody (Fig. 5) although the N-terminal 12kDa cMyc reactive digestion product was not retained on the gel. Increasing the number of Xa sites from 2 to 4 in mutant Xa100(4) produced similar outcome with even more rapid and more complete cleavage at the targeted site. Thus, cleavage at position 100 is specific in these mutants and strongly suggest that the linker separating the PAI and RED subdomains of the PD is solvent exposed and protease accessible. Analysis of mutants Xa216(1) and Xa216 (2) showed similar results. Both of the specific C-terminal HA-reactive 32kDa and N-terminal cMyc-reactive 26kDa digestion products appeared at 2 min, and digestion of the full length protein was almost complete by 20-45 mins (Figs. 4 and 5). These results indicate that the protein segment immediately upstream the HD is solvent exposed and accessible to protease cleavage (Fig. 3).

Effect of DNA Binding on Protease Sensitivity of Pax3 Mutants Xa100 and Xa216
The effect of Pax3 binding to PD (P3OPT) and HD (P1/2, P2) target sequences on the conformation of each domain was analyzed by monitoring the effect of DNA binding on accessibility of Xa cleavage sites present in mutants Xa100 (PD) and Xa216 (HD). Also included in these experiments were WT Pax3, as well as Pax3-Xa mutants previously observed to be resistant to factor Xa cleavage in time course studies (supplementary data). Briefly, cell extracts expressing Pax3 proteins were incubated with or without target DNA, followed by addition of factor Xa and detection of HA (Fig. 4) and cMyc (Fig. 5) immunoreactive cleavage products appearing over time. The extent of protection from proteolytic fragmentation was further quantitated after densitometry of the immunoblots and is expressed as the fraction of intact full length Pax3 remaining following 20 minutes of incubation with factor Xa (Fig. 4B, 5B). For the WT Pax3, binding to PD or HD target sequences had no effect on digestion profiles, as expected, Likewise, DNA binding by the PD or the HD of mutants Xa114, Xa131, Xa172, Xa189 did not affect their previously noted resistance to factor Xa cleavage (Fig. 4B, 5B and Figs. S1 and S2 in supplementary data). This suggests that DNA binding by either domain in these mutants does not cause a conformational change that increases solvent exposure of the respective Xa bearing segments. By contrast, incubation of mutant Xa100(2) with P3OPT increased resistance to proteolysis (persistence of 55kDa protein), suggesting that DNA binding to the PD causes a conformational change reducing solvent accessibility of the PD. In addition, monomeric binding to P1/2 and in particular dimerization of on P2 both also caused a dramatic increased resistance to proteolytic cleavage of Xa100(2) (from 10% to 40% intact protein), suggesting that DNA binding by the HD in this mutant also causes a conformational change in the PD. Identical results were obtained with Xa100(4), although the increased susceptibility to cleavage caused by the 4 consecutive Xa sites at position 100 was maintained in this mutant for all DNA binding conditions tested. In the case of mutants Xa216(1) and Xa216(2) mutants that bear Xa sites immediately upstream of the HD, DNA binding by the HD, in particular dimerization on P2, caused a strong increased resistance to proteolysis (from 25% to 85% intact protein) suggestive of a conformational change at that site. In addition, binding of both mutants to the PD target sequence (P3OPT) also increased resistance to proteolysis, suggesting that DNA binding by the PD also causes a conformational change in the HD. In all cases, results of immunoblotting with anti-HA (Fig. 4) and anti-cMyc antibodies (Fig. 5) were in complete agreement (Fig. 4B, 5B).
Together, results from Xa100 and Xa216 are remarkably similar and suggest that DNA binding at either the PD or HD causes a conformational change at both sites. This change appears to reduce the amount of solvent exposed area in the protein, suggesting a more compact conformation of the DNA-bound protein.
Specificity of the protective effect of PD and HD target sequences on accessibility of the Xa sites in Xa100 and Xa216 mutants was investigated in dose-response studies. In these experiments, the Pax3 mutants were incubated with PD or HD oligonucleotide probes and DNA-Pax3 complexes were allowed to form; following this, increasing amounts of factor Xa protease was added to the reaction mixture which was further incubated for 15 minutes, and the appearance of specific proteolytic cleavages products was monitored by SDS-PAGE and immunoblotting with the anti-HA antibody (Fig. 6A). In these studies, 2 additional control probes were tested to further validate the specificity of the DNA effect observed in time course studies: a second, independently derived PD oligonucleotide P6CON, and a PD and HD nonspecific oligo which was used as a negative control. Typical immunoblots are shown in Fig. 6A, and quantitation of the protective effects of the DNA probes by densitometry is shown in Fig. 6B (% of uncleaved Pax3 at concentration indicated by arrow in Fig. 6A). These experiments showed that incubation of mutants Xa100(2) and Xa216(2) with PD oligonucleotides P6CON and P3OPT reduced sensitivity of both proteins to increasing doses of factor Xa. Likewise, incubation of both proteins with HD probes P1/2 and P2 similarly increased resistance to factor Xa fragmentation. The effect of the 2 PD probes and 2 HD probes on protease sensitivity was specific and not seen in control DNA-free conditions (data not shown), and upon incubation with a non-specific target sequence (Fig. 6A/B) (38), and c) the fact that Pax proteins can bind DNA exclusively through their PAI subdomain (28). We note that mutations Xa55 and Xa71 in the PAI domain impaired DNA binding to PD oligos only, while the Xa66 mutant showed impaired DNA binding to both PD and HD sequences. This behavior is similar to WS1 mutants G48R/S and P50L (upstream helix 1 of PAI), respectively, and has been suggested to reflect functional interdependence of the PD and HD in DNA binding (38). Interestingly, insertion of 2 or even 4 Xa sites in the linker joining the PAI to the RED (Xa100) had no effect on DNA binding; this linker is well conserved amongst Pax proteins, and sequences immediately downstream the insertion site make extensive phosphate and base-specific contacts in the minor groove of DNA (Fig. 1). As expected, the 3 insertions (Xa172, Xa189, Xa216) in the poorly conserved linker joining the PD and the HD, including one within the octapeptide motif (Xa189) conserved in other Pax proteins (HSIDGILG; (57)), had no major impact on DNA binding of the corresponding mutants to PD and HD sites. With respect to the HD, 2 mutations inserted either downstream helix 2 (Xa252) or upstream the major DNA binding helix 3 (Xa259) appeared to either strongly diminish or abrogate monomeric or dimeric DNA binding to HD target sequences (Fig. 2). Although this conclusion is supported by both the high degree of conservation of the targeted sequences in the Paired-type HDs of the Pax family (Fig. 1), and the fact that many WSI mutations map to the HD of Pax3 (38), our inability to express high levels of these mutants precluded a more detailed analysis.
Evaluating the accessibility of inserted Xa sites to protease cleavage can readily provide insight into the solvent exposure of the corresponding Pax3 sub-domains. Results of time course studies were very clear and showed that of the 7 sites in which insertions preserved DNA binding only 2, Xa 100 and Xa216, were readily accessible to proteolytic cleavage at the earliest time points of analysis (Fig. 4). In agreement with the proposed structural model of the PD (Fig. 7A), these results showed that the PAI to RED subdomain linker (Xa100 mutant) is clearly exposed to solvent. Although this domain is not believed to play a critical role in DNA binding, current structural models suggest that it is in close proximity to DNA (25). In addition, alternative splicing of a glutamine residue at position 108 of Pax3/Pax7 is known to alter DNA binding specificity of the PD (54,58). Therefore, mutants Xa100 should be ideally suited to monitor structural changes associated with DNA binding by the PD. By contrast, downstream insertions into helices 1 (Xa114) and 2 (Xa131) of the RED domain were completely resistant to protease cleavage, possibly suggesting that the HTH motif is either compact or buried in the core of the      Table 2). The position and size, in kDa of the molecular mass markers are displayed as dashes on the left and numbers on the right, respectively. All immunoblots probed with anti-HA antibody including those of Figure S1 were scanned by densitometry (Fig. 4B).
The intensity of the immunoreactive intact full length WT protein and of individual Pax3 mutants was determined at both the "0" time point and at the "20 min" time points. The amount of intact protein remaining at 20 min (compared to 0 min) was determined for all mutants and for all DNA binding conditions and is expressed as the fraction of intact protein (expressed as a percentage). Several mutants showing inaccessible factor Xa cleavage sites (Xa114, Xa131, Xa172/3, Xa189/2) were also included in the analysis as negative controls.

FIGURE 5
Accessibility to Protease Cleavage of Factor Xa sites Inserted in Selected Pax3 Mutants: Effect of DNA Binding. Wild type Pax3 and the different Pax3 mutants were incubated with factor Xa under different conditions and digestion products were analyzed by immunoblotting as described in the legend to Figure 4, except that blots were probed with an anti-cMyc antibody directed against the amino terminal cMyc epitope present in all mutants. All immunoblots probed with anti-Myc antibody including those of Figure S2 were scanned by densitometry (Fig. 4B).
The intensity of the immunoreactive intact full length WT protein and of individual Pax3 mutants was determined at both the "0" time point and at the "20 min" time points. The amount of intact protein remaining at 20 min (compared to 0 min) was determined for all mutants and for all DNA binding conditions and is expressed as the fraction of intact protein (expressed as a percentage). Several mutants showing inaccessible factor Xa cleavage sites (Xa114, Xa131, Xa172/3, Xa189/2) were also included in the analysis as negative controls. of the full length intact Xa100(2) and Xa216(2) proteins (empty arrowheads), and of the major predicted, HA-immunoreactive product (filled arrowhead, see Table 2) are shown. (B). For quantifying the effect of DNA binding on accessibility of factor Xa cleavage sites in doseresponse studies, the immunoblots in (A) were scanned by densitometry. The intensity of the immunoreactive intact Xa100(2) and Xa216(2) full length proteins was determined in the absence of factor Xa, and after digestion with a factor Xa amount indicated by the arrow at the top of each immunoblot series. The amount of intact protein remaining after factor Xa digestion was determined for Xa100(2) (black) and Xa216(2) (gray) and for all DNA binding conditions and is expressed as the fraction of intact protein (expressed as a percentage). Xa sites identified as causing loss-of DNA binding (red), as having no effect on DNA binding and being either accessible (purple) or not (blue) to factor Xa protease fragmentation.

Accessibility to Protease Cleavage of Factor Xa sites inserted in Additional Pax3 Mutants:
Effect of DNA Binding. Results are shown as described in legend to Figure 4.