Evolution of the levels of human leukocyte antigen G (HLA-G) in Beninese infant during the first year of life in a malaria endemic area: using latent class analysis

Human Leukocyte Antigen-G (HLA-G) is a non-classical HLA class Ib antigen, with important immune-regulatory functions [1], that differs from classical HLA class I antigens by a low amount of polymorphism and a tissue restricted expression [2, 3]. The physiological HLA-G expression is restricted to fetal tissues such as invasive cytotrophoblast and amnion, and in adults, to immune-privileged organs, cornea, thymus, pancreatic islets, endothelial cell precursors, and erythroblasts. Dendritic cells and macrophages can also express HLA-G [4]. Seven alternative proteins can be generated, four membrane-bound and three secreted isoforms [3‐5]. HLA-G is primarily expressed during pregnancy and plays an important role in maternal tolerance of the fetus. Trophoblast cells are protected by HLA-G which inhibit the cytolytic activity of NK cells and protect the semiallogenic fetus from maternal rejections responses [4, 6, 7].

Apart from pregnancy, HLA-G has been shown to be expressed in many types of tumours [8‐11]. Tumour cells express HLA-G to escape to immunity and high levels of soluble HLA-G (sHLA-G) have been related with unfavourable outcome of prognosis [12, 13]. In transplantation, HLA-G molecule has been shown to be important for a better allograft acceptance, whereas low serum levels seem to increase the risk of autoimmunity [13, 14] and allergy [15]. If the role of HLA-G with viral infections is admitted [15‐18], this association with parasite infections is much less documented [19, 20]. Recently, it was reported that genetic polymorphisms in HLA-G 3′UTR were associated with susceptibility to human African trypanosomiasis and could influence the clinical and immunological responses directed to Plasmodium falciparum [21‐23].

A recent work reported that high levels of sHLA-G were significantly associated with a higher incidence ratio of malaria in Beninese children [24] consistent with the fact that the inhibition of immune responses by HLA-G expression could lead to a greater susceptibility to malaria. In this previous analysis, hierarchical mixed models have been used to deal with the repeated measures design. However, mixed models make the assumption that individuals are randomly drawn from a homogeneous population. As a complement it seemed interesting to test whether distinct subpopulations of children are present in the overall population, each with its own parameters, with distinctive sHLA-G evolutions, or trajectories. The identification of trajectories of sHLA-G through the first year of life and risk factors that predispose to, or modify, a particular trajectory may suggest the existence of individual with variable malaria susceptibility. Latent class analysis (LCA) approach is ideally suited to explore this issue [25, 26].

Thus, the main objective of the present work was to explore the existence and characteristics of groups of children with different patterns of soluble HLA-G evolution between birth and 12 months of life in a cohort of Beninese children exposed to malaria, using latent class analysis.

At delivery sHLA-G was measured for all mothers and all cord blood samples (n = 165). During the follow-up, a total of 445 measurements were performed from 3 to 12 months distributed as follows: 109/165 at 3 months, 113 at 6 months, 114 at 9 months and 109 at 12 months. Overall, a total of 775 measurements (mother, cord, 3, 6, 9 and 12 months) were performed. In most cases, an HLA-G measurement was missing because the mother had not attended a quarterly visit. Only 10 children (6.1 %) were present for a single visit.

The present work is complementary to study of Sadissou et al., describing the evolution of sHLA-G in a population of African infants [24]. In the first study, a strong correlation has been shown between mother and cord blood levels of sHLA-G at delivery, as well as an association between LBW and high level of sHLA-G during the first year of life. In addition, high levels of sHLA-G were associated with an increase of the incidence ratio of malaria during the follow-up, but the total number of malaria infections during the follow-up was not associated with the level of sHLA-G.

The present analysis, confirms these results, and provides new understanding by identifying new determinants of sHLA-G evolution. Indeed, although in the first study, the level of sHLA-G increased the risk of infection in the following quarter, any association has not been highlighted between the level of sHLA-G and the overall number of malaria infections during the follow-up. In this complementary analysis, this association is strongly significant. The difference between these two results could be due to the complexity of the effect of malaria infection that cannot be pointed out under the hypothesis that all children belong to the same homogeneous population. Using latent class approach allows hypothesizing that distinct subpopulations of children are presents in the overall population, each with its own characteristics, with distinctive sHLA-G evolutions. This approach permitted us to explore more precisely the complex association between malaria and HLA-G.

The mechanisms by which HLA-G gene expression is regulated are only partially understood [37]. Concerning infections, HIV or hepatitis, infected patients have higher level of sHLA-G, and it has been suggested that these levels induce tolerance and contribute to immune evasion of virus [38‐42]. Although, to date, such hypothesis has not been clarified for malaria; it could be hypothesized that the same phenomenon occurs during malaria infection leading to immune evasion. Indeed, P. falciparum could upregulate the HLA-G secretion through the stimulation of cytokines such as IL-10 and IFN-γ. HLA-G is known to inhibit the function of T, NK and B cells through direct interaction with ILT2. Recently, Naji et al. showed that HLA-G inhibits B cell proliferation, differentiation, and immunoglobulin (Ig) secretion in both T cell–dependent and –independent models of B cell activation. HLA-G also acts as a negative B cell regulator in modulating B cell antibodies secretion in a xenograft mouse model [43]. Taken together all these arguments are consistent with the existence of a complex interaction between HLA-G, immunity and P. falciparum infection, leading to a lower B cell response. As IgG antibodies play a pivotal role in anti-malarial protection [44‐46], the inhibition of IgG specifically directed against P. falciparum mediated by HLA-G, not only allows the parasite to escape the immune system but also is responsible of the higher susceptibility to infection showed previously.

However, other factors could influence HLA-G secretion during malaria infection. It has been shown that hypoxia could play a role in HLA-G regulation during pregnancy [47], in some cancer [48] but also in more physiologic but extreme life conditions like altitude [49]. Tissue hypoxia resulting from decreased microcirculatory flow due to parasite sequestration and endothelial dysfunction contributes to the pathogenesis of malaria in children [50, 51]. The results could be consistent with the fact that malaria related hypoxia is already present very early in life and associated with an increase of sHLA-G levels, and consequently strengthening the inhibition of immunity.

More than 60 % of the children born with a LBW belong to the high and intermediate trajectories. It is widely accepted that birth weight is a key indicator of fetal and neonatal health [52‐54] and overall 60–80 % of infants who die during the neonatal period are children born with LBW [55]. The association between LBW and HLA-G has already been shown in this same cohort using mixed model [24]. However, the present approach allows pointing out a very interesting aspect, which has not been demonstrated by previous analysis. Indeed, children born with detectable HLA-G in their cord blood have a greater probability to belong to the higher or intermediate trajectories than other children, and hence be concerned by LBW. Considering the frailty of LBW children, the detection of such high-risk individuals could have huge public health consequences. Taking into account the strong correlation between sHLA-G in cord blood and in mother’s blood at delivery, the mother’s sHLA-G level could be considered as a potential marker of high-risk newborn. Soluble HLA-G has already been proposed as predictor of in vitro fecundation outcome [56], but also as biomarker of allergy [57], allograft and cancers [58, 59]. A follow-up of pregnant women and of their offspring over 2 years is now ongoing in Benin to test whether the sHLA-G level early in pregnancy is also correlated with HLA-G level in cord blood or with infectious morbidity during the first 2 years of life.

Latent class analysis approach is more and more frequently used in biomedical research [60]. It is an extension of mixed models that assume the presence of several groups of subjects (latent groups) sharing a particular evolution trajectory of some attribute. This approach allows a better understanding of the phenomenon under study and the formulation of new questions or hypotheses [61, 62]. This methodology is part of mixture modelling, a wide data analysis approach allowing studying unobserved heterogeneity in a population [60]. The question of the number of trajectories remains of great importance and to date, there is no single statistical test to determine the correct number of latent classes [63]. The selection of the best model needs to take into account statistical measures of model fitting but also the interpretation of the output, in particular the relevance of the different groups, based on the literature or the expert’s knowledge. BIC was used as statistical criteria although others can be proposed. Bootstrap Likelihood Ratio test (BLRT) has been proposed as being more powerful. However, BLRT needs an increased computation time and finally the authors recommended using BIC as the first step [64]. In the data, using BIC would lead to choose a model with four trajectories instead of three. Both model have the same interpretation since the two intermediate trajectories in the four-group model share the same evolution, i.e. high level trajectory, low level trajectory and intermediate group(s) starting with low levels of sHLA-G and then growing up. However, models with more than three trajectories seem to present a lower stability. Finally, to deal with the possibility and complexity of these population mixture models, it could be of interest to pursue this exploration by using different methods in different populations.

A genetic association between HLA-G and malaria has already been pointed out. Indeed, it has been shown that some polymorphisms of the HLA-G 3’UTR were associated with variable risk of infection [22], but also with antibody responses against P. falciparum. These associations have been identified in another population from Senegal. It has also been demonstrated that a strong molecular signature of balancing selection at HLA-G 5’URR exist and that this region may certainly be the direct target of selection [65]. These results strengthen the interest of this gene and of sHLA-G as potentially involved in differential susceptibility to malaria.

Summary, this study allowed a better understanding of the complex interaction between HLA-G and malaria infection and pointed out the potential interest of sHLA-G as marker of frailty.