Insulin resistance is linked to a specific profile of immune activation in human subjects

We tested the hypothesis that a particular immune activation profile might be correlated with insulin resistance in a general population. By measuring 43 markers of immune, endothelial, and coagulation activation, we have previously shown that five different immune activation profiles may be distinguished in 150 volunteers. One of these profiles, Profile 2, characterized by CD4+ T cell senescence, inflammation, monocyte, B cell, and endothelial activation, presented elevated insulinemia, glycemia, triglyceridemia, and γ-glutamyl transferase, a marker of liver injury, in comparison with other profiles. Our data are compatible with a model in which a particular immune activation profile might favor the development of insulin resistance and metabolic syndrome. In this hypothesis, identification of this profile, that is feasible with only 3 markers with an error rate of 5%, might allow to personalize the screening and prevention of metabolic syndrome-driven morbidities as liver steatosis.


Results
A specific immune activation profile is linked to insulin resistance and metabolic syndrome. We recruited 74 women and 76 men with a mean (± SD) age of 61.7 (± 4.3) years. We have previously shown that 5 different IA profiles may be distinguished in this cohort. Profile 1 is characterized by a high proportion of naïve T cells, Profiles 2 and 3 by elevated percentages of terminally differentiated and senescent CD4+ T cells and CD8+ T cells, respectively, Profile 4 by a high proportion of activated NK cells, and Profile 5 by an increase in the percentage of monocytes 5 . We tested to see whether one of the IA profiles we had observed in the 150 subjects we have previously analyzed was associated with IR and MetS. The levels of the risk factors defining MetS are given in Table 1 for each IA profile. Compared with the other volunteers, Profile 2 individuals presented higher insulinemia (13.3 ± 9.2 vs 9.7 ± 5.6 μU/mL, p = 0.016, Fig. 1a), higher homeostatic model assessment (HOMA) (3.9 ± 3.4 vs 2.4 ± 1.6, p = 0.014, Fig. 1b), a larger waist/hip circumference (0.93 ± 0.09 vs 0.87 ± 0.10, p = 0.005, Fig. 1c), and higher triglyceridemia (1.5 ± 1.0 vs 1.1 ± 0.6 mM, p = 0.041, Fig. 1d). Their glycemia (6.2 ± 1.7 vs 5.5 ± 1.0 mM, p = 0.122, Fig. 1e) and systolic blood pressure (141 ± 12 vs 137 ± 18 mmHg, p = 0.172, Fig. 1f), were non-significantly higher than the other volunteers, and their HDL (0.55 ± 0.14 vs 0.60 ± 0.16 mM, p = 0.350, Fig. 1g) non-significantly lower.
In line with these biological data, Profile 2 volunteers were more often treated with antihyperlipidemic drugs than the other volunteers (36.4 vs 17.2%, p = 0.038), as shown in Table 1. Antidiabetic (13.6 vs 4.7%, p = 0.128) and antihypertensive drugs (31.8 vs 20.3%, p = 0.229) were non-significantly more often prescribed to Profile 2 individuals than to individuals with another profile.
IA Profile 2 is linked to high γGT levels. One of the consequences of MetS is liver steatosis 2 . Liver steatosis is the most common cause of elevated gamma-glutamyltransferase (γGT) 9 . To test the hypothesis that persons with IA Profile 2 might develop more frequently this disease, we compared γGT levels in the persons who volunteered for this study according to their IA profile. Individuals with IA Profile 2 presented with higher levels of γGT than the other individuals (76 ± 88 vs 38 ± 34 UI/L, p = 0.007, Fig. 7a). Another potential cause of elevated serum γGT levels is alcohol consumption, but there was no difference in alcohol use between participants with Profile 2 and the other participants (6.1 ± 10.9 vs 8.2 ± 22.1 g, p = 0.846, Fig. 7b).

Discussion
In this study, we have shown that one IA profile is clearly linked to some markers of IR and MetS, eventhough subjects with this profile received antidiabetic, antihyperlipidemic and antihypertensive therapies more often than the other subjects. This is phenocopying our report of a link between one IA profile and hyperinsulinemia as well as hypertriglyceridemia in HIV patients 7 . Of note, both IR-linked IA profiles, in people living with HIV-1 www.nature.com/scientificreports/ and in a general population, present many common immunological characteristics. This is also in line with the correlation between adipose tissue inflammation and IR reported by many studies 10 . In particular, macrophage activation, TNFα overproduction 11 , and CD163 shedding 12 have been observed in human obese adipose tissue. Even activated B cells have been shown in this tissue 13 . Strikingly, in this present work we found that the IA profile linked to insulin resistance was characterized by a high level of sTNFRI, a finding common to our observation in people living with HIV 7 . As sTNFRI is a marker of TNFα production, and as TNFα inhibits insulin signaling 4 , both findings argue for a causative link between IA Profile 2 and insulin resistance.
We also observed that hyperinsulinemia was linked to the levels of circulating sCD163. In the course of inflammation, CD163 is shed from the surface of monocytes and macrophages by the metalloproteinase TNFαconverting enzyme (TACE), which is also responsible for the release of TNFα from the surface of immune cells 14 . Therefore, sCD163 is considered as a marker of monocyte and macrophage activity and of TNFα production. Indeed, TNFα production and sCD163 levels have previously been shown to be related 15 as we observed here (Fig. 3a). White adipose tissue (WAT) inflammation and IR are inter-related 16 . This WAT inflammation results in CD163 mRNA overexpression in WAT macrophages 17 which correlates with sCD163 blood concentrations 12 . Accordingly, sCD163 levels have been reported to be linked to adiposity [18][19][20][21][22] on one hand and to IR 15,20,21,23-25 on the other hand. sCD163 has even be identified as a predictive marker of type 2 diabetes in the general population independently of age or BMI 23 . Thus, the fact that sCD163 levels in Profile 2 are higher than in the other profiles may be considered as an additional argument for a causative link between Profile 2 and IR. tPA is released by activated endothelial cells functioning both as a serine protease favoring thrombolysis by converting plasminogen to plasmin and as an inflammatory cytokine 26 . tPA activates NFκB, inducing inflammatory cytokines production, and modulates inflammatory infiltration, particularly macrophage migration, in various organs [26][27][28][29][30] . In line with these proinflammatory properties, tPA levels have been linked to CRP and fibrinogen levels in another study 31 , and to sTNFRI as well as to CRP in the present study (Fig. 3b,d). These characteristics might explain why tPA is correlated with IR and predictive for the development of type 2 diabetes 32-34 . Thus, the high level of tPA in Profile 2 may be interpreted as a third argument for a causative link between Profile 2 and IR.
In addition to monocyte/macrophage and T cell activation, B cell overactivity has been observed in the visceral adipose tissue (VAT) of obese mice 35,36 and men 13 . In VAT, B cell activation induces the production of IFNγ and other inflammatory cytokines by T cells, fueling IR 36 . This aligns with the fact that B cell deficient mice on a high fat diet show better insulin sensitivity than wild-type mice 36 . One consequence of this B cell activation might be an increase in serum IgA levels which has been linked to obesity 37,38 , MetS 37,38 , and diabetes 37 , as we observed here in Profile 2.
Globally, these data lead to a model where IA Profile 2 favors IR. Yet, it may be argued that conversely, IR may favor Profile 2. First, IR is responsible for an increase in circulating free fatty acids able to activate hepatic macrophages and thereby to increase the level of sCD163 39 . In addition, IR and hyperglycemia cause oxidative www.nature.com/scientificreports/ stress, TACE activation, and CD163 shedding 14 . Second, a positive causal effect of IR on tPA has been revealed via a Mendelian randomization analysis 32 . This causal effect could be mediated by a defect in endothelium-derived nitric oxide 40 . Third, hyperproduction of IgA might also be a consequence of metabolic disorders, as suggested by the report that treatment of morbid obesity by adjustable gastric banding leads to reduction in IgA values 41 . Thus, the links between Profile 2 and IR might be bidirectional, resulting in a vicious circle. Whatever the nature of these links, it would be interesting to test whether Profile 2 is predictive for the establishment of MetS and for the morbidities it fuels. The combination of activation markers characterizing Profile 2 might then provide us with a predictive signature. The possibility of reducing the size of this signature to three markers adapts immune profiling to routine. Last, but not least, deciphering the soluble factors which are overproduced by Profile 2 individuals, and which may cause IR might uncover pathways to pharmacologically target and prevent MetS.
One of the limitation of our work is that our study population is not representative of the general population. We recruited adults who volunteered for a free health checkup. Therefore, the majority of these volunteers were in a precarious socio-economical situation, and overweight. Another limitation is that the present study is cross-sectional, highlighting only correlations. Further analysis is needed to establish whether there are causative links between some types of immune activation and IR.

Methods
Study design. We have previously described the characteristics of the 150 volunteers we enrolled for this study 5 . Pregnant or breastfeeding women, people under immunomodulatory treatment, with cancer, acute infection, autoimmune or autoinflammatory disease were not included. Fifty-nine percent of these volunteers were in a precarious socio-economical situation. Our protocol was performed in accordance with the relevant guidelines and regulations and approved by the French Ethics Committee Sud Est IV. All patients had provided written informed consent. The trial was registered on ClinicalTrials.gov under the reference NCT04028882.
Metabolic and immunologic markers in peripheral blood. Blood was collected after eight hours of fasting on EDTA Vacutainer tubes (Becton Dickinson, Le Pont-de-Claix, France), immediately centrifuged (delay < 4 h) and plasma was frozen at -80 °C on several aliquots until analysis. Insulinemia were measured using electrochemiluminescence immunoassay "ECLIA" on Cobas e602 analyzer from Roche (Roche diagnostic, Meylan, France) using Roche reagent kits and calibrators. Insulinemia was determined from plasma aliquot never thawed. Lower detection limit was 0.2 µU/mL (1.39 pmol/L) and intra-and inter-assay was < 1.5% and < 5%, respectively. Blood triglycerides, HDL, LDL, and total cholesterol levels were quantified on the Cobas8000/e502© analyzer from serum collected into SST II Vacutainer tubes (Becton Dickinson). The determination of the other markers has been previously described 5 .
Statistical analysis. We used t-test or Mann-Whitney test to compare markers and IA profiles. The links between markers were determined by Spearman rank correlations. Fisher test or χ 2 test was used to compare qualitative covariates.
The R software version R 3.5.1 (July, 2018) was used to perform the analysis described in this section. To select an optimal number of variables and create a parsimonious predictive model of the Profile 2, we chose genetic algorithms (GAs) 42,43 . GAs are optimization algorithms, inspired by Charles Darwin's idea of natural selection. They provide approximate solutions to complex optimization problems. In a first step, a population of potential solutions is randomly generated. Then, this population evolves through the iterative application of mutation, cross-over and selection. The natural selection preferentially preserves the fittest individuals over the successive generations. An evolutionary algorithm improves the selection over time and allows the best solution to emerge from the best of prior solutions.
In our application 44,45 , solutions are subsets (combinations) of the immunological markers (the features) mentioned above. Specifically, the mutation randomly alters a solution by feature addition, removal or substitution. The cross-over randomly combines the features of two solutions. Selection is the only operator increasing the quality of solutions across generations. It relies on a fitness function (to be optimized) quantifying the solution quality.
A Linear Discriminant Analysis (LDA) 46 is applied on each solution using the R package MASS. To avoid over-fitting a cross-validation is used to evaluate the accuracy. The fitness function uses this accuracy penalized by the subset size to favor parsimonious solutions. In this goal, we also chose 10 as the maximal size for subsets. In order to favor solution robustness, the genetic algorithm was run four times and all solutions of the final generations were evaluated through 30 runs of independent linear discriminant analysis with 2-fold cross validation. Solutions were ranked according to their average correct classification rate during the cross-validation process.