The genetic diversity within the 1.4 kb HLA-G 5′ upstream regulatory region moderately impacts on cellular microenvironment responses

The HLA-G 5’URR extending 1.4 kb from the ATG presents a unique set of regulatory elements among HLA genes. Several variable sites have been described that coincide with or are close to these elements, thus HLA-G 5′URR polymorphism might influence the HLA-G expression level. We cloned the ten most frequent HLA-G 5′URR haplotypes to evaluate their activity on a luciferase reporter gene in HLA-G+ cell lines (JEG-3/choriocarcinoma and FON+/melanoma). We also investigated associations between the plasma HLA-G (sHLA-G) levels and the HLA-G 5′URR variability in 157 healthy individuals. Cell lines were transfected with pGL3-Basic vector constructions containing HLA-G 5′URR sequences. The G010101a (in JEG-3) and G010101b (in FON+) haplotypes exhibited higher promoter activity, whereas the G010101d (in JEG-3) and G010102a (in FON+) haplotypes exhibited lower promoter activity. In the presence of HLA-G inducers (interferon-β and progesterone) or repressors (cyclopamine) HLA-G promoter activity was modulated, but certain haplotypes exhibited differential responses. No strict association was observed between plasma sHLA-G levels and the 5′URR haplotypes or genotypes; however, the G010101b haplotype was underrepresented among HLA-G-negative plasmas. Therefore, the HLA-G 5′URR polymorphism may have an impact on the modulation of HLA-G gene expression, but alone provides a limited predictive value for sHLA-G levels in vivo.


Results
The HLA-G 5′URR activity in HLA-G positive cells varies according to the cloned haplotype and to the cell type. We used two types of HLA-G positive cells to investigate the role of cell microenvironment on the HLA-G expression pattern according to 5′URR haplotype. Choriocarcinoma JEG-3 and melanoma FON + cell lines were transfected, as previously described 19 , with pGL3-Basic vector constructions containing one of the ten most frequent HLA-G 5′URR haplotypes known as G0104a; G0104b; G010102a; G010101a; G010101b; G010101c; G010101d; G010101f; G0103a and G0103e.
Differential activity of HLA-G 5′URR haplotypes in transfected JEG-3 cells exposed to activating (interferon-β and progesterone) or repressing (cyclopamine) agents. To further explore the 5′URR haplotype response to agents known to modulate HLA-G expression levels, JEG-3 cells transfected with each of the ten haplotypes cloned into pGL3-Basic vector were treated with 1000 U/mL interferon-β 28 , 1 µg/mL progesterone 39 or 5 µM cyclopamine 38 . We performed individual comparisons of luciferase activity: i) before and after treatment for each construction (ratio or delta values), ii) comparisons among all haplotypes, and iii) comparisons according to the variable site close to the target site of the modulating agent.
The treatment of JEG-3 cells with progesterone provided significant increased luciferase activity only for five haplotypes; however, less intense than the one observed for interferon-β treatment. The treatment with progesterone increased mean luciferase activity in cells transfected with G0104a (1.2-fold higher; P < 0.05), G010102a (1.2-fold; P < 0.05), G010101f (1.3-fold; P < 0.05), G0103a (1.4-fold; P < 0.05) and G0103e (1.2-fold; P < 0.05) haplotypes (Fig. 2B). The comparisons of the progesterone effect across the ten haplotypes did not reveal significant differences (ratio and delta values) ( Table 1). Considering that progesterone receptor binds to the −37 bp position at the HLA-G promoter region, and considering that a SNP at position −56 (−56C or −56T) is observed at the 5′URR, we evaluated the effect of this polymorphic site on luciferase activity induced by progesterone, and no significant differences were observed (P = 0.89) (Fig. 2D).
The HLA-G G010101b haplotypes may influence plasma sHLA-G levels. To investigate whether HLA-G 5′URR haplotypes can influence plasma sHLA-G levels, we evaluated the relationship between sHLA-G levels and the variability of the HLA-G 5′URR segment in 157 healthy Brazilian individuals. We observed 26 variable sites at this segment, all of them previously described in the Brazilian population 20,30 . Genotype frequencies were in agreement with the Hardy-Weinberg equilibrium expectations and no new polymorphism was observed in this region. Thirteen different HLA-G 5′URR haplotypes were identified; all of them already described for Brazilians and for other population samples 20 . The most frequent haplotype was G010102a (32.8%) and the least frequent was the G010102e (0.3%) haplotype. Figure 3 shows all haplotypes identified in this study with their respective frequencies.
Soluble HLA-G levels in the whole sample ranged from 0.0 to 30.0 ng/mL (median = 1.8 ng/mL and mean = 4.6 ± 5.9 ng/mL). It should be emphasized that sHLA-G was not detected in 64 samples, a fact that results in a positive skew in sHLA-G level distribution and explain why the median is smaller than the mean. When the whole sample was stratified according to the presence (sHLA-G + ; n = 93) or not (sHLA-G − ; n = 64) of detectable sHLA-G, the median and mean values become quite similar in the sHLA-G + group (7.4 ng/mL vs. 8.2 ng/mL).
The exact test of population differentiation based on haplotype frequencies revealed no difference between the sHLA-G + and sHLA-G − groups (P = 0.1010 ± 0.0104), but the G010101b haplotype was significantly underrepresented among sHLA-G − when compared with sHLA-G + plasmas (P = 0.047), indicating that this haplotype is related to a greater sHLA-G level ( Table 2).
The association between sHLA-G levels and HLA-G 5′URR variable sites, genotypes and haplotypes disclosed no significant differences (data not shown). Nonetheless, the analysis of diplotypes (pair of haplotypes) (Fig.  4) revealed that the most frequent haplotype (G010102a) was observed together with several other haplotypes. However, no association was observed between sHLA-G levels and the different diplotypes. For this analysis, we considered only diplotype groups with at least 10 occurrences.

Discussion
HLA-G has been described to be a tolerogenic molecule and its neoexpression was associated with the modulation of several pathological conditions 40 . Therefore, the understanding of the factors that participate on the HLA-G expression regulation is clinically relevant. The regulatory HLA-G 5′URR and 3′UTR segments exert a crucial role on transcriptional and post-transcriptional gene regulation, respectively 13 . Variable sites identified along these segments influence or are suspected to modulate HLA-G expression. HLA-G 3′UTR variations have been shown to modulate mRNA stability [17][18][19] through several potential mechanisms including the differential affinity of the HLA-G mRNA to microRNAs 22,41 . Variable sites along the HLA-G 5′URR were investigated mainly by population studies 20,27,29,30 and were proposed to regulate the interaction of transcription factors with promoter binding sequences and DNA methylation 13 . Considering such hypothesis, we brought here new insights on the impact of the HLA-G 5′URR diversity on the gene transcription activity in vitro and on sHLA-G plasma levels in a cohort of healthy donors.
Firstly, we investigated the promoter activity level of ten frequently observed HLA-G 5′URR haplotypes using a luciferase reporter gene assay. All HLA-G 5′URR transfected haplotypes induced the luciferase activity in both HLA-G + cell lines. Seven haplotypes provided higher response in FON + than in JEG-3 cells, a result that is in line with the higher HLA-G expression observed in the melanoma cells 19,42 . This effect was not observed with three haplotypes, suggesting that the action of transcription factors that are qualitatively and/or quantitatively specific for each cell line was affected by the 5′URR polymorphisms. In addition, although differences in the levels of luciferase activities were observed between several haplotypes, few of them reached significance. For instance, nucleotide variations between haplotypes belonging to Promo-G0104 and Promo-G0103 lineages (16 SNP locations 20,27 ) have no significant effect whatever the cell line used. On the contrary, the −541G and −483G alleles are exclusively found in G010101f and G010101d haplotypes, respectively, and are the only differences between them. Therefore, it is likely that these nucleotide variations are involved on the differential level of luciferase activity obtained with these haplotypes in JEG-3 cells. Noteworthy, the G010102a haplotype, which exhibited the lower activity in FON + cells, strongly differs (13 specific variants) from haplotypes belonging to Promo-G010101 and Promo-G0103 lineages (Fig. 3). Considering the 16 variants with no apparent effect, −1140T allele is thus the only that might influence the low level of luciferase activity observed with the G010102a haplotype. Interestingly, the −1140 A/T SNP has been recently pointed out as a putative target for balancing selection, with -1140T allele hypothesized to be associated with a lower HLA-G expression than -1140A allele 27 . Otherwise, we found that the G010101a haplotype exhibited the highest activity when transfected into the JEG-3 cells, whereas the G010101b haplotype exhibited the highest activity when transfected into FON + cells. Considering the activity of G010101a and G010101b haplotypes, the result obtained with FON + cells is in agreement with the previous study performed by the Ober's group using JEG-3 cells, reporting that the G010101b haplotype exhibited the highest activity compared to G010101a, G010101c, G010102 and G010301 haplotypes 24,28 . Noteworthy, the unique difference between G010101a (−725C) and G010101b (−725G) haplotypes is the polymorphic site observed at −725G/T/C, for which the −725G allele has been associated with increased expression levels 28 . However, the HLA-G 5′URR fragments used by Ober and colleagues (nucleotides from −1412 bp to −33 bp from ATG) were shorter than ours (−1438 bp to +2 bp). Therefore, one explanation for the apparent controversial results obtained with JEG-3 cells would be that specific factors involved in 5′URR responses could target the regions located −1438 bp to −1412 bp and/or −33 bp to + 2 bp and could interact with factors that might be associated with the −725G/T/C polymorphism. Interestingly, the Hviid's group 43 demonstrated no significant difference between the −725 C/C genotype and the −725 C/G genotype in the HLA-G cell surface expression of trophoblast cells from first trimester placental tissues. In agreement with the variations observed in luciferase activity with the −725G allele in the present study, the authors suggested that it could be due to high surface expression variation in the −725 C/G group.
With the aim to further explore the haplotype responses to known modulators of HLA-G expression, we observed that some SNPs were relevant to the level of luciferase activity. First, several lines of evidence indicate that IFN-β treatment up-regulates HLA-G expression and increases HLA-G levels 28,44,45 . Compared with untreated cells, we observed an increased luciferase activity with all 5′URR haplotypes transfected into JEG-3 cells cultured in the presence of IFN-β. This result was expected since the ten haplotypes contain an intact HLA-G ISRE. Among them, G010101b and G010101c are the only ones that exhibit the −725G variation site, and the luciferase activity of these haplotypes was significantly increased when compared to other haplotypes exhibiting the −725C or −725T alleles. As mentioned above, the role of this SNP has been studied by the Ober's group using a site directed mutagenesis assay, corroborating the relevance of the −725G allele in the increased expression level. Nonetheless, and contrary to what was expected by testing less 5′URR haplotypes 28 , we found that the increased expression is also related to response to IFN-β. The SNP is located close to the HLA-G ISRE 28 and   thus might contribute to a specific DNA conformation and/or specific binding factor that would improve IRF-1 binding when the Guanine is present, a hypothesis to be investigated. Second, progesterone has an important role on the maintenance of pregnancy 33 and increases HLA-G expression in JEG-3 cell line 46 . A functional binding site to progesterone receptor (PRE) has been reported located between the −52 bp and −38 bp 39 at the HLA-G 5′URR and the closest variable site is located at −56C/T position. When JEG-3 transfectants were treated with progesterone, all except one haplotype (G010101a) increased the luciferase activity; however, only five HLA-G 5′URR haplotypes showed significant upregulation. Overall, these results are in agreement with those reported by Yie and colleagues (2006), who showed increased luciferase activity after progesterone treatment 39 . However, we observed that progesterone treatment had a limited effect on 5′URR activity compared to INF-β. Indeed, we can speculate that the progesterone response element in this region appears to be a contributing regulatory site to the HLA-G transcriptional level rather than a crucial one. Consequently, we cannot exclude the existence of additional PREs outside the 5′URR, which could increase in vitro the HLA-G expression 46 . Otherwise, Yie and colleagues did not evaluate the influence of the HLA-G 5′URR polymorphic sites on the HLA-G expression after progesterone treatment. Unexpectedly, our study revealed no relationship with any 5′URR variable site, including the 56 C > T SNP despite its proximity to PRE. Regarding this SNP, a lower HLA-G cell surface expression of trophoblast cells from first trimester was previously demonstrated with the -56 C/T genotype compared with the −56 C/C phenotype 43 . This SNP is located in a RREB1-binding site involved in HLA-G repression 37 and thus might influence RREB1 binding rather than the progesterone response. Interestingly, regarding the most frequent worldwide haplotypes, the absence of progesterone effect is observed with the G010101a haplotype (2 nd most frequent), whereas a significant effect is observed with G010102a (1 st most frequent) and G0104a (3 rd most frequent) haplotypes 47 . Notably, the G010101a haplotype exhibits 13 variation sites compared with G010102a and G0104 haplotypes (Fig. 3) 20,27 ; however, it differs only at the -541 position compared with G010101f, which exhibits a significant progesterone response. This suggests that a variable site or a specific combination of variable sites within G0104a, G010102a, G010101a, G010101f, G0103a and G0103e haplotypes may faintly up-regulate the luciferase activity observed upon progesterone treatment. Third, considering the influence of 5′URR polymorphism in the down regulation of the HLA-G expression, the steroidal alkaloid cyclopamine decreased the luciferase reporter gene expression for all transfected haplotypes, particularly for G010101a and G010101c. Cyclopamine acts as GLI (Glioma-Associated Oncogene)-1 and GLI-2 repressor or as GLI-3 inductor 48 . Whereas GLI-1 and GLI-2 act as transcriptional activators, GLI-3 functions as a transcriptional repressor [49][50][51][52] . The GLI-3 binding sequence identified between -1116 bp and −1108 bp is conserved in each haplotype and thus may firstly participate in the significant luciferase down-regulation observed following cyclopamine treatment. Otherwise, the binding sequence is close to the 1121 C > T variation site 38 and the G010101c haplotype is the only one that presents a Thymine at position −1121 bp, whereas the others present a Cytosine. This suggests a possible influence of the −1121T variant on the magnitude of the cyclopamine response even if other SNPs are undoubtedly involved in the response level of the G010101a haplotype. Once again, and as proposed above for IFN-β, SNPs might participate in the DNA conformation and/or the binding of other factors that would modulate the action of cyclopamine according to the 5′URR haplotype. Finally, regarding the possible association between plasma sHLA-G and HLA-G 5′URR polymorphisms we did not observe significant differences. However, significant associations have been previously identified with 3′UTR polymorphisms 23 , suggesting that sequence variations may primarily affect posttranscriptional mechanisms. Notably, we did not detect plasma sHLA-G in a large part of blood samples (44%; n = 157). Although the limit of sensitivity may differ according to the laboratory and the use of an in house or a commercial ELISA, this result is consistent with other studies 53,54 , one of which reported only 23% sHLA-G + plasmas (n = 30) from healthy donors. Interestingly and in agreement with a previous study 25 , we observed that the G010101b haplotype was underrepresented among sHLA-G − donors. This haplotype was associated with increased HLA-G expression in FON + cell line, and is in linkage disequilibrium with the 3′UTR haplotype known as UTR-4 and the coding G*01:01:01:05 allele 20,30 . Interestingly, UTR-4 has been classified as medium sHLA-G producers in previous studies 23,47 and the G*01:01:01:05 allele presents different intronic sequences when compared to its counterparts the G*01:01:01 allele group, which may somehow influence alternative splicing. In addition, although luciferase activity provided by G010101b and G0103e haplotypes was similar in JEG-3 cells, we noticed a significant down-regulation when the G0103e haplotype was transfected into FON + cells. In addition, the G0103e haplotype is mostly associated with UTR-5, previously associated with low sHLA-G production 19,23 .
In conclusion, all these findings are consistent with a moderate impact of the 1.4 kb 5′URR polymorphism on the magnitude of HLA-G expression in response to differential cellular microenvironment modulators. They also suggest that the 5′URR segment alone is not a high predictor of HLA-G expression level.

Methods
Cloning of the HLA-G 5′URR haplotypes into the pGL3-basic vector. DNA samples from seven Brazilian individuals carrying the ten main HLA-G 5′URR haplotypes (Table 1) 20,24,27,29,30 were selected for the cloning procedure. DNA was amplified using the forward 5′-AAGCTTCACAAGAATGAGGTGGAGC and reverse 5′-CGCGGATCCTTGGCGTCTGG primers, generating a 1438 bp fragment. Cycling conditions consisted of 35 cycles of 30 s at 95 °C, 30 s at 60 °C and 1 min at 72 °C. PCR-amplified fragments were first inserted into pUCm-T vector (Bio Basic, Ontario, Canada) and confirmed by HindIII (Invitrogen, Carlsbad, CA) digestion. Ten constructions carrying each of the main haplotypes were selected and identified by Sanger's sequencing analysis (Applied Biosystems, Foster City, CA). A 1525 bp KpnI/BamHI (Invitrogen) fragment obtained from each selected clone was subcloned into pGL3-Basic vector (Promega, Madison, WI) upstream of the firefly luciferase gene. HLA-G 5′URR sequence transfections and dual-luciferase report assay. Cells constitutively expressing HLA-G (choriocarcinoma JEG-3 and melanoma FON + cell lines) were grown in DMEM (Gibco, Carlsbad, CA) and RPMI-1640 (Sigma-Aldrich, Lyon, France) supplemented mediums, respectively 55,56 . One microgram of each construction was transfected into sub-confluent JEG-3 cells by Lipofectamine 2000 Transfection Reagent (Invitrogen) and FON + cells by Lipofectamine LTX Reagent with Plus Reagent (Invitrogen), according to manufacturer's instructions. The pGL3-Promoter (Promega) and pGL3-Basic empty vectors were used as positive and negative controls, respectively. The pRL.SV40 Renilla Luciferase vector (Promega) was co-transfected to normalize the transfection efficiency.
Transfected cells were harvested and lysed 48 hours post-transfection. The supernatant was collected and used to perform the dual-luciferase report assay (Promega), according to manufacturer's instructions. Firefly luciferase values were first normalized to those of Renilla luciferase values, and then to luciferase expression of the empty pGL3-Promoter vector. Assays were run in duplicates from at least four independent experiments.
Transfected JEG-3 cells were also treated for 24 hours with 5 µM cyclopamine (Tocris Bioscience, Minneapolis, MN), or 1000 U/mL interferon-β (RayBiotech, Norcross, GA) or 1 µg/mL progesterone (Sigma-Aldrich, France). At least four independent experiments were performed in duplicate. In parallel with treated cells, untreated transfected JEG-3 cell line was also cultured with medium (baseline), performing 12 independent experiments. Normalized Luciferase activities obtained with each HLA-G 5′URR constructions transfected into JEG-3 and FON + cell lines and into JEG-3 cell line treated or not with interferon-β, progesterone or cyclopamine are presented in the Supplementary Tables S1 and S2.
Subjects and DNA extraction. The study protocol was approved by the Ethics Committee of the Ribeirão Preto Medical School, University of São Paulo, Brazil (Protocol #6102/2013). A total of 157 unrelated healthy bone marrow donors (mean age = 32.31 ± 9.74) of both sexes (74 men) were randomly selected at the University Hospital of the same Institution. All methods were carried out in accordance with relevant guidelines and regulations. Written informed consent was obtained from all subjects. Genomic DNA was extracted from peripheral blood leucocytes using a standard salting out procedure 57 .

HLA-G 5′URR typing.
A 1752 bp fragment including the 5′URR segment was amplified and sequenced (using primers G-908R, G-830F, G-304R and GPR-247) as previously described 30 . The amplified fragment included 1446 nucleotides upstream the first translated ATG, and 388 nucleotides of the coding sequence. List of variable sites at HLA-G 5′URR region is presented in the Supplementary Table S3.
HLA-G 5′URR haplotypes were inferred using the PHASE algorithm 58 as previously described 30 , and named according to previous studies 20,30,47 since no official nomenclature has been assigned to this gene segment.
Soluble HLA-G (sHLA-G) quantification. The sHLA-G plasma level was quantified in plasma by sandwich ELISA using mAb anti-HLA-G MEM-G/9 (Exbio, Praha, Czech Republic) and anti-β2-microglobulin (DAKO, Glostrup, Denmark) as capture and detection antibodies, respectively, as previously described 23,59 . Statistical analysis. Haplotype frequencies were estimated by direct counting, and adherence to the Hardy-Weinberg equilibrium (HWE) was tested using the GENEPOP 3.4 software 60 . Linkage disequilibrium (LD) between HLA-G SNPs was evaluated by means of Lewontin's standardized coefficient D' and by a likelihood ratio test of LD implemented with the ARLEQUIN software 61 .
Promoter activity of the untreated transfectants was compared among all haplotypes using Kruskal-Wallis test followed by the Dunn's posttest. For a given haplotype, the Wilcoxon matched pairs test was used for comparisons between treated (cyclopamine, interferon-β, or progesterone) and untreated transfectants. Comparisons between the effects (ratio or difference of expression) of a given treatment on the modulation of luciferase expression by different promoter haplotypes were performed either by the Kruskal-Wallis (followed by the Dunn's posttest) or the Mann-Whitney tests. Associations among sHLA-G levels and 5′URR haplotypes and diplotypes (pair of haplotypes) were also performed by the Kruskal-Wallis or the Mann-Whitney tests. Only diplotypes groups with at least 10 occurrences were considered for association analysis with sHLA-G levels.
All these statistical analyses were performed using GraphPad Prism 5 v5.0b software.