Genetic Mapping of Two Murine Loci that Influence the Development of IL-4-Producing Thy-1dull γδ Thymocytes1

IL-4-producing γδ cells belong to a novel subset of γδ lymphocytes that expresses a very restricted repertoire of TCRs. To gain a deeper insight into the development and in vivo functions of these cells, we have analyzed the genetic control of their representation in the thymus. Using an intercross between C57BL/6 and DBA/2 mice we found two loci on chromosomes 13 and 17—named LadT1 and LadT2, respectively—with marked influence in their development. The LadT2 locus does not appear to be the MHC locus. The region identified on mouse chromosome 13 contains the structural genes for TCRγ as well as the IL-9 gene, which has been suggested as a candidate gene influencing the complex pathogenesis of asthma.

T lymphocytes coexpressing molecules usually restricted to the NK cell lineage (NK T cells) have received increased attention in the last few years (1)(2)(3). Recently, we have characterized a population of TCR␥␦ T lymphocytes (␥␦ cells) that shares with NK T cells a number of phenotypic and functional characteristics (4). In the thymus, this ␥␦ T cell population differs from conventional ␥␦ cells by its low expression of Thy-1, and, thus, we referred to it as the Thy-1 dull ␥␦ T cell population. Most Thy-1 dull ␥␦ thymocytes express a phenotype usually associated with activated or memory T cells, and around half of them express NK receptors and/or the CD4 coreceptor. In DBA/2 mice, they predominantly express the product of the V␥1 gene together with that of a member of the V␦6 subfamily (the V␦6.4 gene), and their junctional sequences show very little diversity (4). This limited diversity of TCRs is the consequence of a strong cellular selection, suggesting the existence of a limited set of endogenous ligands (5).
Another remarkable feature of the Thy-1 dull ␥␦ T cell population is its capacity to simultaneously secrete high levels of both Th1and Th2-type cytokines upon activation in vitro (4). In particular, the production of high levels of IL-4 by ␥␦ cells appears to be a unique property of the Thy-1 dull ␥␦ T cell population. Recently, a major role of ␥␦ cells has been demonstrated in the early IL-4 production that is required for the development of specific IgE responses in the periphery and for the subsequent airway inflammation upon intranasal Ag challenge (6). This led to the suggestion that IL-4 production by ␥␦ cells in the periphery could be impor-tant for the development of some Th2 responses to protein Ags and thus focused attention on this particular T cell population.
Little is known about the development and the specificity of the Thy-1 dull ␥␦ cells. These questions remain difficult to address mainly because of our lack of knowledge about the specific ligands recognized by murine ␥␦ cells in general. A possible approach to these questions would be the identification of genetic element(s) controlling the development of the Thy-1 dull ␥␦ cells. By the possible overlap between genetic regions characterized in these analyses and those found in the genetic studies of complex pathological processes, such an approach may not only reveal unknown physiological functions of these cells but also provide new insights into the regulation of complex phenomena leading to disease.
In this report, we began the identification of genetic elements regulating the size of the Thy-1 dull ␥␦ thymocytes in an intercross between C57BL/6 (B6) 4 and DBA/2 mouse strains. We choose these two mouse strains as prototype strains for three different reasons: 1) they display substantial differences in the representation of the IL-4-producing Thy-1 dull ␥␦ T cell population (4); 2) they both contain members of the V␦6 subfamily that are preferentially used by these ␥␦ populations (the V␦6.3 gene in B6 mice and the V␦6.4 gene in DBA/2 mice) (4); and 3) both genetic backgrounds have been shown to be capable of selecting Thy-1 dull ␥␦ thymocytes with identical phenotypic and functional characteristics and similar TCR repertoires (5). This suggests that the putative endogenous ligand selecting the restricted TCR repertoire expressed by the Thy-1 dull ␥␦ thymocytes is present in both mouse strains. Using simple sequence length polymorphism (SSLP) analysis, we mapped two major quantitative trait loci (QTL) to the chromosomes 13 and 17. Together, these loci account for most of the genetic effects involved in the phenotypic differences in this cross. The region identified on mouse chromosome 13 contains the structural genes for TCR␥ as well as the gene coding for IL-9, which was recently identified as a candidate gene in the complex pathogenesis of asthma and allergy both in humans (7,8) and mice (9). Interestingly, the two parental strains B6 and DBA/2 also display very different steady-state levels of this cytokine (9).

Immunofluorescence staining and flow cytometric analyses
Cells (10 5 to 10 6 ) were incubated in staining buffer (PBS, 3% FCS, 0.1% NaN 3 ) with the indicated labeled mAbs for 30 min on ice and washed twice. When biotin-conjugated mAbs were used, the cells were further incubated with either PE-labeled streptavidin (Southern Biotechnology Associates, Birmingham, AL) or streptavidin-Tricolor (Caltag, South San Francisco, CA) for 15 min on ice. After another two washes, cells were analyzed using either a FACScan or a FACScalibur flow cytometer (Becton Dickinson, Mountain View, CA). Dead cells were gated out either by their staining with propidium iodide or by their forward and angle light scatter profile. Data was analyzed using the CellQuest program (Becton Dickinson).

Genetic analyses
Markers selected on the basis of predicted polymorphism between the DBA/2 and B6 strains were purchased from Research Genetics (Huntsville, Alabama). Polymorphism was confirmed by the analysis of B6, DBA/2, and B6D2F 1 mice. We used a total of 98 polymorphic SSLP markers distributed over most of the genome at about 30 centimorgan (cM) intervals. The minimal and maximal distances between adjacent loci were 1 and 42 cM, respectively. The list of primers will be provided upon request. PCR amplifications from tail DNA were performed for 35 cycles. Each cycle consisted of incubations at 94°C for 30 s, at either 55 or 60°C (as optimized for each set of primers) for 25 s and at 72°C for 30 s. Before the first cycle, a 2-min 94°C denaturation step was included, and after 35th cycle the extension at 72°C was prolonged for 4 min. Amplification fragments were separated by electrophoresis in 2% Resophor agarose gels (Eurobio, Les Ulis, France) and visualized by ethidium bromide staining.

Statistical analyses
Markers in each chromosome were ordered according to The Jackson Laboratory Mouse Genome Database (http://www.informatics.jax.org/ locus.html.) and by minimizing the number of double recombinants using the Map Manager program. The order was confirmed with the MAP-MAKER/Exp program. Linkage analyses were performed using either MAPMAKER.QTL 1.1 (17) or Map Manager/QT software (18). The significance of the likelihood ratio statistics (LRS) generated by genome-or chromosome-wide linkage procedures was specifically assessed by the permutation method of Churchill and Doerge (19) implemented in the Map Manager/QT software. One thousand and 10,000 permutations were per-formed in genome-and chromosome-wide analyses, respectively. The genetic variance (G v ) was estimated according to the equation In the parental strains and in the F 1 hybrids, the total variances observed would be entirely environmental because these animals are genetically identical. In contrast, and because of the segregation of allelic differences in the cross, the total variance observed in the F 2 progeny would be the sum of the environmental and genetic variances. Theoretically, the estimation of E v requires that the phenotypic variances are similar for the parents and the F 1 population (20). To equalize these variances, we transformed the phenotypic values by the function under the hypothesis that the phenotypic variances ( y 2 ) are proportional to their means ( y ) by the equation y 2 ϭ a y b (21). Constants a and b were determined by linear regression analysis after a log 10 transformation of the phenotypic data scored for the parents and the F 1 progeny. The transformed data fitted well the assumptions of equal variances and normality. Linkage analyses were performed after transformation by the same function of the phenotypic data scored in the F 2 progeny. Linkage results were confirmed by a nonparametric test, the Kruskal-Wallis test, performed on the actual phenotypic data.

Nonrecessive DBA/2 loci govern the representation of IL-4producing Thy-1 dull ␥␦ thymocytes
To begin identification of genetic factors influencing the representation of the Thy-1 dull ␥␦ thymocyte population, we produced a cohort of B6D2F 1 and B6D2F 2 animals and analyzed them individually for three different phenotypic parameters: 1) the frequency of Thy-1 dull ␥␦ thymocytes among total ␥␦ cells; 2) the production of IL-4 by DN thymocytes upon stimulation with anti-␥␦ mAbs and IL-2; and 3) the representation of V␥1 ϩ V␦6.3/ V␦6.4 ϩ T cells among expanded ␥␦ cell blasts. These cells represent Ͼ80% of the Thy-1 dull ␥␦ thymocytes in the B6 and DBA/2 mouse strains (4, 5) and can be specifically stained with a novel mAb (9D3) specific for these two V␦ chains (11).
Individual B6 (n ϭ 14), DBA/2 (n ϭ 14), B6D2F 1 (n ϭ 11), and D2B6F 1 (n ϭ 11) mice were scored for these three traits to set the reference values (Table I). Regardless of the trait analyzed, F 1 progenies display a phenotype similar to that of DBA/2 mice, suggesting a nonrecessive mode of inheritance of autosomal genes. The values scored in the B6D2F 2 progeny (n ϭ 98) formed a continuum from low (B6-like) to high (DBA/2-like), indicating that the alleles controlling these traits were segregating in this cross and that multiple loci may influence the three traits analyzed (Fig. 1). An estimation of the genetic component of the phenotypic variance observed in the F 2 progeny based on the analysis of the a The three quantitative traits analyzed were scored in individual B6 (n ϭ 14), DBA/2 (n ϭ 14), B6D2F 1 (n ϭ 11), and D2B6F 1 (n ϭ 11) mice. DN thymocytes were stained with PE-labeled anti-C␦ and FITC-labeled anti-Thy1.2 mAbs and analyzed by flow cytometry. DN thymocytes from the same mice were stimulated with platecoated anti-C␦ mAb (10 g/ml) and rIL-2. Three-day culture supernatants were tested by ELISA for the presence of IL-4. The expanded ␥␦ ϩ blasts were stained with biotinylated 9D3 and FITC-labeled anti-V␥1 mAbs followed by PE-labeled streptavidin and analyzed by flow cytometry. Data correspond to the mean Ϯ SE of the values in each group of mice. parental strains and their F 1 progeny (Ref. 20 and see Material and Methods) ascribed between 65.0% and 78.7% (depending of the trait analyzed) of the variance observed in the F 2 progeny to genetic factors.
The three phenotypic parameters studied appeared clearly correlated in F 2 mice, as indicated by correlation coefficients of 0.94, 0.91, and 0.88 for the percentage of Thy-1 dull ␥␦ thymocytes vs the percentage of V␥1 ϩ 9D3 ϩ T cells, the percentage of Thy-1 dull ␥␦ thymocytes vs the amount of IL-4 produced and the percentage of V␥1 ϩ 9D3 ϩ T cells vs the amount of IL-4 produced, respectively (Fig. 2). These correlations strongly suggest that, in this F 2 cross, only Thy-1 dull ␥␦ thymocytes secrete high titers of IL-4 upon activation and that most Thy-1 dull ␥␦ thymocytes express the V␥1 chain together with the V␦6.3 or the V␦6.4 chain.

Identification of DBA/2 loci that promote the development of IL-4-producing Thy-1 dull ␥␦ thymocytes
To identify genomic regions where putative genes influencing the size of the Thy-1 dull ␥␦ T cell population could be localized, we performed a genome-wide mapping analysis in B6D2F 2 mice. Genomic DNA from 98 F 2 mice analyzed phenotypically was typed with 98 SSLP markers polymorphic between the two parental strains and QTL analyses, an interval mapping method implemented in the MAPMAKER.QTL software (17,22), were performed. The presence of QTL on chromosomes 13 and 17 with major influences on the representation of the Thy-1 dull ␥␦ thymocyte population was supported by maximum logarithmic of odds (LOD) score values of 9.15 and 3.5 reached close to the markers D13Mit63 and D17Mit41, respectively (Fig. 3). The LOD score values obtained at those markers for all three traits are summarized in Table II. Regardless of the trait analyzed and according to the criteria suggested by Lander and Kruglyak (23), these values allow one to declare significant and suggestive linkage to the two regions described above on chromosomes 13 and 17, respectively.
These conclusions were further supported by single locus association and simple interval mapping tests based on a LRS, as implemented in the Map Manager/QT. To better assess significance, we also performed genome-and chromosome-wide permutation tests (19). The LRS values obtained at the same markers (i.e., D13Mit63 and D17Mit41) for the three phenotypic traits analyzed and the adjusted p values obtained after 10,000 permutations performed on chromosomes 13 and 17 are also shown in Table II.
The MAPMAKER.QTL program can also provide information about the model of action of the two DBA/2 loci identified. LOD score plots along chromosomes 13 and 17 were recomputed under three different models of action (recessive, dominant, or additive) and compared with the unconstrained analysis. The LOD score profiles were virtually unchanged under the assumption of additive inheritance (Fig. 3), suggesting that the DBA/2 alleles at these loci act additively to increase the frequency of IL-4-producing Thy-1 dull ␥␦ thymocytes.

Loci located on chromosomes 13 and 17 account for most of the genetic factors influencing the development of Thy-1 dull ␥␦ thymocytes
Once the putative loci on chromosomes 13 and 17 (hereafter referred to as LadT1 and LadT2 for loci associated with the development of Thy-1 dull ␥␦ thymocytes 1 and 2) were identified, we sought to determine their relative contribution to the phenotype and their putative interactions. From the QTL analysis, we estimated the fraction of the total variance observed in the F 2 progeny and, by implication, the fraction of the genetic variance that each  Table I. The results are represented after normalization of the data as described in Material and Methods.

FIGURE 2.
Correlation between the three phenotypic parameters scored in F 2 mice. The three phenotypic parameters were monitored as described in the legend of Table I. locus is responsible for (Table II). Together, both loci explain 76.4%, 71.2%, and 47.9% of the genetic variance in this cross for the frequency of Thy-1 dull ␥␦ and of V␥1 ϩ 9D3 ϩ thymocytes traits and for the production of IL-4 trait, respectively. Even if other loci with weaker effect are likely to exist, this result strongly suggests that among all loci segregating in this cross, we identified those having major influences on the development of the Thy-1 dull ␥␦ T cell population.
To study putative interactions between these two loci, we compared the distribution of the scored phenotypic values in nine groups of F 2 mice redistributed on the basis of combined genotypes at the markers giving the highest LOD score values for LadT1 and LadT2 (i.e., D13Mit63 and D17Mit41). Table III shows the results obtained for the representation of the Thy-1 dull ␥␦ thymocytes trait. F 2 mice that inherited the two B6 alleles at both loci displayed a B6-like phenotype (6.5% Ϯ 4). Mice homozygous for the B6 allele at D13Mit63 showed an intermediate phenotype (from 13.6 Ϯ 13.7% to 14.6 Ϯ 10.2%) regardless of whether they had inherited either one or two DBA/2 alleles at D17Mit41. A similar phenotype (12.1 Ϯ 4.9%) was also observed in mice heterozygous at D13Mit63 but homozygous for the B6 allele at D17Mit41. In contrast, mice homozygous for the DBA/2 allele at D13Mit63 and for the B6 allele at D17Mit41 showed a higher phenotype (22.9 Ϯ 9.5%), similar to that of mice heterozygous at both loci (22.1 Ϯ 13.6%), consistent with a stronger effect of LadT1 than LadT2. With all other combinations, we observed a phenotype closer to the DBA/2 phenotype or even higher (from 36.7 Ϯ 22.4% to 46.7 Ϯ 16.8%). This may be due to additive or epistatic effects of unidentified B6 loci together with LadT1 and LadT2 of DBA/2 origin or to the absence of negative effects due to other DBA/2 loci. The existence of the latter is suggested by the fact that the representation of the Thy-1 dull ␥␦ thymocytes in DBA/2 mice is not higher than in B6D2F 1 animals, despite the fact that DBA/2 alleles at LadT1 and LadT2 appear to act additively (see Fig. 3). Altogether, these results suggest that the genes encoded by LadT1 and LadT2 are not redundant because both are  required to obtain the DBA/2 phenotype in the F 2 progeny. They are also indicative of the lack of epistatic interactions between LadT1 and LadT2, as shown by the elevated proportion of Thy-1 dull ␥␦ thymocytes as the number of DBA/2 alleles at these two loci increases.

Influence of the TCR␥ haplotype in the representation of the Thy-1 dull ␥␦ T cell population
Candidate genes for the LadT1 locus may exist in the TCR␥ locus that is located on the centromeric part of chromosome 13. This locus is polymorphic between the B6 and DBA/2 mouse strains and has been previously shown to influence the representation of other ␥␦ T cell populations (24,25). Polymorphism at this locus could explain the differential representation of the Thy-1 dull ␥␦ thymocytes in the two strains either as the result of a V␥1 allelespecific cellular selection or by molecular selection mechanisms regulating, in a strain-specific manner, the rearrangement and/or the expression of the V␥1 gene. Several of these possibilities can be directly tested by the analyses of previously reported Tg mice (11).
We have recently shown that the representation of Thy-1 dull ␥␦ thymocytes in B6 and D2B6F 1 mice Tg for a rearranged V␥1J␥4C␥4 chain of B6 origin and containing a junctional sequence commonly found in the Thy-1 dull ␥␦ T cell population is very similar to that found in non-Tg B6 and B6D2F 1 animals (Ref. 11 and Fig. 4A). These results suggest that the B6 and DBA/2 allelic forms of the V␥1J␥4C␥4 gene can both be used by Thy-1 dull ␥␦ T cells to form their TCRs. Because the restricted TCR repertoire expressed by most Thy-1 dull ␥␦ thymocytes results from a strong cellular selection (5), it was important to ascertain that the Thy-1 dull ␥␦ cells present in B6D2F 1 Tg mice were not the progeny of rare cells that had rearranged the DBA/2 allele of their endogenous V␥1J␥4C␥4 gene. To that end, we took advantage of the availability of a novel mAb (termed 7C10) that recognizes the V␥1J␥4C␥4 chain expressed in DBA/2 mice but not in B6 mice (see Materials and Methods). Consistent with this predicted specificity, 7C10 stains the vast majority and virtually none of the V␥1-expressing cells in DBA/2 and B6 mice, respectively (Fig.  4B). In B6D2F 1 mice, 7C10 recognizes about 60% of the V␥1positive ␥␦ cells (Fig. 4B).
Among eight B6D2F 1 Tg mice analyzed between 4 and 8 wk of age, 7C10 ϩ cells represented between 0.5 and 2.6% of the Thy-1 dull ␥␦ thymocytes, indicating that most of these cells express the Tg chain (Fig. 4C). Altogether, these experiments demonstrate that the B6 V␥1J␥4C␥4 allele can be positively selected by the Thy-1 dull ␥␦ T cell population, making it unlikely that an allele-specific cellular selection is responsible for the differential representation of this population in B6 and DBA/2 mice. Furthermore, because the frequency of Thy-1 dull ␥␦ thymocytes in B6 Tg mice is low, these experiments also suggest that molecular constrains due to putative polymorphism at regulatory elements controlling the rearrangement of the V␥1 gene do not play a major role in the development of this population. Finally, analysis of TCR␥␦ surface expression with TCR␥-and ␦-specific mAbs failed to reveal any substantial difference between normal and Tg B6 and B6D2F 1 mice (not shown), indicating that differential regulation of the expression of TCR␥ chains is not responsible for the observed differences. Altogether, these data do not provide any evidence indicating that the LadT1 locus represents the structural genes for TCR␥. In contrast, we observed that around 75% (mean Ϯ SD ϭ 72.0 Ϯ 14 in 26 animals) of the Thy-1 dull ␥␦ thymocytes in normal B6D2F 1 mice use the DBA/2 V␥1J␥4C␥4 allele, whereas the frequency of cells expressing the same allele among V␥1-expressing Thy-1 bright ␥␦ thymocytes was close to 50% (not shown). These results indicate a preference for the DBA/2 V␥1J␥4C␥4 allele in F 1 animals.

Discussion
By analyzing an F 2 intercross between B6 and DBA/2 mouse strains, we have identified two major loci that influence the representation of this IL-4-producing Thy-1 dull ␥␦ thymocyte population. These loci, named LadT1 and LadT2, map to regions on mouse chromosomes 13 and 17, respectively, and, together, explain a large fraction (47.9 -76.4% depending on the trait analyzed) of the genetic variance in this cross.
Several lines of evidence suggest the existence of two linked loci on chromosome 13. First, high LOD score values can be observed over a large region of about 30 cM, what is unexpected for  a single locus. Second, LOD score plots along chromosome 13 show a clear shoulder around the D13Mit224 marker, suggestive of the existence of a second locus on this region. Third, this putative second peak appears clearly defined under the assumption of a dominant model of gene action (see Fig. 3), suggesting that the two putative loci act differently. Finally, analysis of close to hundred B6D2F 1 ϫ B6 backcrossed animals clearly shows the presence of two separate peaks on chromosome 13 (not shown), further supporting the existence of two individual QTL on this chromosome. Are there any candidate genes in the vicinity of the LadT loci? A number of genes appear as possible candidates, although in the absence of more refined mapping data any attempt to identify these genes remains highly speculative. At first sight, two loci seem like obvious candidates: the TCR␥ locus on mouse chromosome 13 and the MHC complex on mouse chromosome 17. Both loci have been previously shown to influence the representation of different ␥␦ T cell populations defined by their utilization of particular V␥V␦ combinations (24 -26).
LadT2 is probably not a MHC gene. The MHC locus is located on the proximal region of mouse chromosome 17, while the LadT2 locus maps to the distal part of the same chromosome. Furthermore, analyses of MHC-congenic strains in the C57BL/10 background failed to show any effect of MHC-linked genes (our unpublished observations). Whether the LadT1 locus represents the structural genes for TCR␥, which are located on the region of chromosome 13 centered around the highest LOD score value, cannot be formally answered at present. The analyses of B6 and B6D2F 1 Tg mice presented here did not provide any evidence suggesting that the LadT1 locus is the TCR␥ locus. Therefore, it is unlikely that polymorphism at the structural genes for TCR␥ may explain all the observed effects of the LadT1 locus. However, these analyses do not allow one to formally exclude any effect of TCR␥ locus polymorphism in the development of Thy-1 dull ␥␦ thymocytes. Such effect is likely to exist as evidenced by the preferential usage of the DBA/2 allele of the V␥1J␥4C␥4 gene in B6D2F 1 mice. This may be important to consider if, as discussed above, there are two linked loci on this chromosomal region. A definitive answer is expected to come from the analysis of mouse strains congenic for different portions of the chromosome 13, the production of which is now in progress.
Another interesting candidate gene for LadT1 is the IL-9 gene, which is also located on the proximal region of chromosome 13. IL-9 is a T cell-derived cytokine originally identified as a mouse T cell growth factor (27) and a mast cell-enhancing activity (28). Direct effects of IL-9 in normal hemopoietic progenitors, fetal thymocytes, and B cells have also been reported (reviewed in Ref. 29) and a role for IL-9 in regulating specific IgE and IgG1 synthesis has also been suggested (30). Recently, IL-9 has been proposed as one major candidate gene in the predisposition to asthma both in humans (31) and mice (9). This was suggested, in humans, on the basis of linkage disequilibrium between total serum IgE levels and a marker within the IL-9 gene (7,8). In mice, bronchial responsiveness was analyzed as a quantitative trait in recombinant inbred strains between hyporesponsive B6 and hyperresponsive DBA/2 mice and found to be influenced by a region on chromosome 13 containing the IL-9 gene. Additional experiments showed that IL-9 expression was markedly reduced in bronchial hyporesponsive mice, and, more important, its level of expression was determined by sequences within the IL-9 gene itself (9). To our knowledge, a putative role of IL-9 in the development of ␥␦ cells has not been analyzed. Such studies must await the availability of IL-9-deficient mice that have not yet been produced.
␥␦ cells have been recently shown to be important in the development of pulmonary inflammatory reactions to protein Ags in a mouse model of allergy (6). In this model, the absence of ␥␦ cells results in a reduced sensitization of the mice as shown by a marked decrease in the Ag-specific IgE and IgG1 responses in the serum of ␥␦-deficient mice when compared with normal mice. These effects could be abrogated by administration of recombinant IL-4 together with the Ag during the sensitization period, suggesting that ␥␦ cells were important in the early IL-4 production required for the sensitization. Although not formally proven, it was suggested that ␥␦ cells may be responsible for the early generation of IL-4 driving the Th2 response that leads to the production of specific IgE and IgG1 and to the symptoms of airway inflammation after intranasal administration of the same Ag. As high levels of IL-4 production by ␥␦ cells appears to be a property of the Thy-1 dull ␥␦ population, both in the thymus (4) and in the periphery (11), Thy-1 dull ␥␦ T cells are the best candidate to provide the initial IL-4 synthesis required for the development of an allergic airway inflammation. Although at this point, the fact that genetic predisposition to asthma as measured by bronchial responsiveness and the development of the Thy-1 dull ␥␦ T cells map to the same region on mouse chromosome 13 is insufficient to conclude that the two phenomena are related, this possibility is, nevertheless, attractive and deserves further investigation.