Quantile-dependent expressivity of serum interleukin-6 concentrations as a possible explanation of gene-disease interactions, gene-environment interactions, and pharmacogenetic effects

Abstract

Background:

Interleukin 6 (IL-6) is a moderately heritable pleiotropic cytokine whose elevated concentrations in coronary artery disease, peripheral arterial disease, pulmonary arterial hypertension, Eales’ disease, Sjògren’s syndrome, osteoarthritis, adenocarcinoma, neuroblastoma, polymyalgia rheumatica, pulmonary tuberculosis, and enterovirus 71-infection, and following coronary artery bypass graft, show larger genetic effects than in unaffected low IL-6 controls. We hypothesize that genetic effects may depend upon whether average IL-6 concentrations are high or low, i.e., quantile-dependent expressivity.

Method:

Quantile-specific offspring-parent (β_OP) and full-sib regression slopes (β_FS) were estimated by applying quantile regression to the age- and sex-adjusted serum IL-6 concentrations in families surveyed in the Framingham Heart Study. Quantile-specific heritabilities were calculated as h²=2β_OP/(1+r_spouse) and h²={(1+8r_spouseβ_FS)^0.5-1}/(2r_spouse)).

Results

Heritability (h²±SE) of IL-6 concentrations increased from 0.01±0.01 at the 10^th percentile (NS), 0.02±0.01 at the 25^th (P =0.009), 0.03±0.01 at the 50^th (P =0.007), 0.04±0.02 at the 75^th (P =0.004), and 0.13±0.05 at the 90^th percentile (P =0.03), or 0.0005±0.0002 for each one-percent increase in the offspring’s phenotype distribution (P_{linear trend}=0.02) when estimated from β_OP, and from 0.02±0.02 at the 10^th (NS), 0.02±0.02 at the 25^th (NS), 0.06±0.02 at the 50^th (P =0.01), 0.12±0.04 at the 75^th (P =0.001), and 0.30±0.03 at the 90^th percentile (P<10⁻¹⁶), or 0.0015±0.0007 for each one-percent increase in the sibling phenotype distribution (P_{linear trend}=0.02) when estimated from β_FS.

Conclusion:

The heritability of serum IL-6 concentrations is quantile dependent, which may contribute in part to the larger genetic effect size reported for diseases and environmental conditions that elevate IL-6 concentrations vis-à-vis unaffected controls.

Introduction

Interleukin 6 (IL-6) is a pleiotropic cytokine affecting immune responses, hematopoiesis, bone metabolism, embryonic development, chronic inflammatory diseases, and acute inflammation [1]. IL-6 is an essential mediator of the acute phase response [2]. It is a major inducer of hepatic fibrinogen and C-reactive protein synthesis, inducer of vascular endothelial growth factor (VEGF) expression, and regulator of leukocyte growth and differentiation. Elevated IL-6 concentrations are associated with rheumatoid arthritis, type-2 diabetes (T2DM), atherosclerosis, coronary heart disease, heart failure, and multiple cancers [3].

Basal IL-6 concentrations are moderately heritable, with estimates ranging from <0% [4], 12% [5], 16% [6], 17% [7], 21% [8], 24% [9], 31% [10], 32% [11], 37% [12], 61% [13], to 69% [4]. Two observations have fostered speculation that genetic influence affecting basal IL-6 concentrations may differ from those affecting systemic inflammatory reactions and acute phase response. Firstly, the heritability for IL-6 production after ex vivo stimulation with lipopolysaccharide is reported to be greater (57%) [14] than those generally cited above for basal IL-6 concentrations [4–13]. Second, the largest genome-wide association study of basal IL-6 concentrations to date (67,428 subjects of European ancestry [3]) failed to assign genomewide significance to the -174G>C and -572G>C polymorphisms in the IL-6 promotor region despite their significant influence on IL-6 concentrations in multiple disease states [15–37]. It is also possible that the genetic effect size of genes affecting IL-6 concentrations may depend upon whether IL-6 concentrations are high or low relative to their distribution. This phenomenon, quantile-dependent expressivity [38–43], has been previously demonstrated for C-reactive protein [44], whose plasma concentrations are much dependent on IL-6 [2]. Quantile-dependent expressivity hypothsizes that the effect on the phenotype of a genetic variant may depend upon whether the phenotype is high or low relative to its distribution in the population [38].

An important consequence of quantile-dependent expressivity is its potential to give rise to gene-disease and gene-environment interactions for diseases and conditions that change average IL-6 concentrations [45]. By gene-environment interaction, we mean environmental effects on the phenotype that differ by genotype (e.g., IL6 -174G>C genotypes affecting post-operative IL-6 increases [15,24]). Specifically, under quantile-dependent expressivity, stratifying by conditions that distinguish high vs. low IL-6 concentrations is expected to give rise to different size genetic effects. Quantile-dependent expressivity could also give rise to apparent pharmacogenetic effects for drugs that lower IL-6 concentrations if the IL-6 difference between genotypes decreases with decreasing IL-6 concentrations [46,47]. This requires that the genotype-specific IL-6 changes cannot move in parallel from higher pretreatment IL-6 concentrations to lower on-treatment IL-6 concentrations. In this case, the genetic marker may simply track the reduction in the genetic effect size at the lower concentration rather than interacting with the drug’s pharmacology.

Therefore, quantile regression [48] was applied to IL-6 concentrations measured in Framingham Heart Study family sets [49,50] to test whether IL-6 genetic effects were quantile-specific and to assess whether this was consistent with reported genetic interactions under IL-6 altering conditions. Specifically, quantile-regression was used to estimate quantile-specific regression slopes between offspring and parents (β_OP) and full siblings (β_FS) from which quantile-specific heritability could be estimated by Falconer’s formula [51]. Heritability was studied because only 1% of the variance in basal IL-6 concentrations was accounted for by the three loci that attained genome-wide significance (rs4537545, rs660895, rs6734238) in GWAS studies [3,52]. The discussion examines gene-disease, gene-environment and gene-treatment interactions from the perspective of both precision medicine and quantile-dependent expressivity. Collectively, our analysis of the Framingham Heart Study family sets and gene-environment interactions of previously published studies support quantile-dependent expressivity of circulating IL-6 concentrations.

Methods

The Framingham Study data were obtained from the National Institutes of Health FRAMCOHORT, GEN3, FRAMOFFSPRING Research Materials obtained from the National Heart, Lung, and Blood (NHLBI) Biologic Specimen and Data Repository Information Coordinating Center. The Framingham study was chosen because it is among the largest epidemiological studies of family sets in terms of sample size, study duration, number of phenotypes measured, and peer-reviewed reports [49,50]. The hypothesis tested is not considered as part of the initial Framingham Study design and is exploratory. The Offspring Cohort included 5,124 adult children of the Original Cohort and their spouses who were initially examined between 1971 and 1975, reexamined eight years later, and then every three to four years thereafter [49]. The Third Generation Cohort is composed of the children of the Offspring Cohort [50]. Subjects used in the current analyses were at least 16 years of age and were self-identified as non-Hispanic white.

Phlebotomy was performed on fasting participants who had rested for 5 to 10 minutes in a supine position, typically between 8 and 9 AM. Specimens were stored at −80°C without freeze-thaw cycles until assay. Collected blood was spun at 5000 rpm for 20 minutes in a balanced oxalate tube. IL-6 were measured at the 7^th clinical examination (1998-2001) and the 8^th clinical examination (2004-2008) of the Offspring Cohort [49] and at the 1^st clinical examination of the Third Generation Cohort (2002-2005) [50]. Serum IL-6 concentrations were measured using commercially available enzyme-linked immunoassay kit (R&D Systems, Minneapolis, MN, USA). Details of assays are available online (https://www.framinghamheartstudy.org/researchers/description-data/noninvasive-biomarker.php). Mean intra-assay coefficient of variation was 3.1%.

Statistics

The statistical approach has been described elsewhere [38–45], and is summarized briefly here. Individual subject values are the average of the age- and sex-adjusted concentrations over all available clinical examinations. Offspring-parent regression slopes (β_OP) were computed using parents from the Offspring Cohort and their Third Generation Cohort children. Full-sibling regression slopes (β_FS) were calculated for the sibships identified in the Offspring and Third Generation Cohorts by forming all k_i(k_i-1) sibpair combinations for the k_i siblings in sibship i and assigning equal weight to each sibling [53]. Heritability in the narrow sense (h²) was calculated as h²= 2β_OP/(1+r_spouse) and h²= {(1+8β_FSr_spouse)^0.5-1}/2r_spouse, where r_spouse is the spouse correlation [51].

Simultaneous quantile regression was performed using the sqreg command of Stata (version. 11, StataCorp, College Station, TX). One thousand bootstrap samples were drawn to estimate the variance-covariance matrix for the 91 quantile regression coefficients between the 5th and 95th percentiles of the offspring’s distribution [54]. Quantile-specific expressivity was assessed by: 1) estimating quantile-specific β-coefficients (±SE) for the 5th, 6th,…, 95th percentiles of the sample distribution; 2) plotting the quantile-specific β coefficient vs. the quantile of the trait distribution; and 3) testing whether the resulting graph was constant, or changed as linear, quadratic, or cubic functions of the percentile of the trait distribution using orthogonal polynomials [55]. Statistics are reported ± one standard error. “Quantile-specific heritability” refers to the heritability statistic, whereas “quantile-specific expressivity” is the biological phenomenon of the trait expression being quantile-dependent.

The results from other studies were re-interpreted from the perspective of quantile-dependent expressivity using genotype-specific means presented in the original articles or by extracting these values from published graphs [15–23,56,57] using the Microsoft Powerpoint formatting palette as previously described [58]. Our interpretations of other studies are not necessarily those of the original authors.

Results

Offspring-parent regression analyses were performed on 1760 offspring with one parent and 1748 with both parents. Full-sib regression analysis were performed using 1726 siblings in 667 sibships from the Offspring Cohort, and 3315 siblings from 1147 sibships from the Third Generation Cohort. The characteristics for the participants used in the analyses are presented in Supplementary Table 1. Mean differences in age, sex and IL-6 concentrations between cohorts were eliminated by the within-cohort age- and sex-adjustment. Six hundred ninety eight subjects had IL-6 measurements but were excluded for lacking family members. Those excluded were more likely from the Offspring than the Third Generation Cohort (15% vs. 5%) and younger (−0.78±0.37 years, P=0.03), but did not differ by percent female (P=0.51), BMI (P=0.43) or serum IL-6 concentrations (P=0.56) when adjusted for cohort.

Traditional estimates of familial concordance and heritability

Spouses were nonsignificantly correlated for untransformed IL-6 (r_spouse= −0.0222) and weakly correlated for log transformed IL-6 concentrations (r_spouse= 0.0503). The offspring-parent regression slope for age- and sex-adjusted untransformed IL-6 concentrations was β_OP=0.0092±0.0111, which corresponds to nonsignificant estimated heritability (h²=0.02±0.02, P=0.41). The full-sib regression slope for untransformed IL-6 (β_FS= 0.0323±0.0176) corresponded to a heritability estimate of (h²=0.065±0.035, P=0.07). However, logarithmically transformed IL-6 (β_OP= 0.1088±0.017 and β_FS= 0.1161±0.0175) yielded estimated of heritability consistent with those reported by others (h² =0.21±0.03, P=1.6x10⁻¹⁰ and 0.23±0.03, P=3.1x10⁻¹¹, respectively).

Quantile-dependent expressivity.

Figure 1A and IB presents the offspring-parent and full-sib regression slopes (β_OP) by percentiles of the age- and sex-adjusted untransformed serum IL-6 distribution. The slopes, and their corresponding heritability estimates (h²= 2*β_OP/(1+r_spouse)), get progressively steeper with increasing percentiles of the distribution. Heritability (h²±SE) of serum IL-6 concentrations increased from 0.01±0.01 at the 10^th percentile, 0.02±0.01 at the 25^th (P =0.009), 0.03±0.01 at the 50^th (P =0.007), 0.04±0.02 at the 75^th (P =0.004), and 0.13±0.05 at the 90^th percentile (P =0.03), or 0.0005±0.0002 for each one-percent increase in the offspring’s phenotype distribution (P_{linear trend}=0.02) when estimated from β_OP, and from 0.02±0.02 at the 10^th percentile, 0.02±0.02 at the 25^th, 0.06±0.02 at the 50^th (P =0.01), 0.12±0.04 at the 75^th (P =0.001), and 0.30±0.03 at the 90^th percentile (P<10⁻¹⁶), or 0.0015±0.0007 for each one-percent increase in the sibling phenotype distribution (P_{linear trend}=0.02) when estimated from β_FS. Log transformed IL-6 concentrations showed no significant evidence for quantile-dependent trends in h² whether estimated from offspring-parent pairs (P_linear=0.07, P_quadratic=0.26, P_cubic=0.27) or full siblings (P_linear=0.13, P_quadratic=0.33, P_cubic=0.10)

Figure 1.

a) Quantile-specific heritability (h²=2β_OP/(1+r_spouse)) estimated from offspring-parent regression slopes (β_OP, Y-axis) by quantiles of the offspring’s serum IL-6 concentrations (X-axis), showing heritability became greater particularly above the offspring’s 70^th percentile. b) Quantile-specific heritability (h²={(1+8r_spouseβ_FS)^0.5-1}/(2r_spouse)) estimated from full-sibling regression slopes (β_FS) by quantiles of the sibs’ serum IL-6 concentrations. Significance of the linear, quadratic and cubic trends, and the 95% confidence intervals (shaded region), determined from 1000 bootstrap samples.

Discussion

The novel finding from these analyses is that IL-6 heritability is quantile dependent, with genetic effects particularly accentuated above the 70^th percentile (approximately 1.8 pg/ml). Untransformed IL-6 concentrations were analyzed because quantile regression does not require normality and therefore provides IL-6 results as originally measured. Moreover, untransformed IL-6 concentrations were used to describe almost all published gene-disease, gene-treatment, and gene-environment interactions (discussed further below). Compared to the 10^th percentile, our analyses showed that IL-6 heritability at the 90^th percentile was 14-fold greater when estimated from β_OP (P_difference=0.02), and 18-fold greater when estimated from β_FS (P_difference=3.5x10⁻¹⁴). Quantile dependence of IL-6 genetic effects may contribute to the quantile-dependent CRP heritability [44], and fibrinogen heritability (unpublished) for which it is their primary stimulant.

Relevance to prior reports

Fishman et al. [59] originally reported that the C-allele of the -174G>C (rs1800795) polymorphism in the IL6 promoter region was associated with lower serum IL-6 concentrations than the ancestral G-allele, and that in vitro LPS- and IL-1-stimulation increased IL-6 expression 2.4-3.6 fold in -174G constructs but not -174C constructs vis-à-vis unstimulated transfected HeLa cells. Our Figure 1 shows weak IL-6 heritability below the seventieth percentile of the sample distribution, supporting the conjecture that -174G>C and other genetic variants might be more easily distinguished in disease, post-operative and environmental states that elevate IL-6 concentrations [15–37,56,57]. Figure 2–5 present published examples amenable to both a precision medicine (i.e., histogram inserts showing genotype-specific differences between affected patients and healthy controls) and quantile-dependent expressivity interpretations (i.e., line graphs showing greater cross-sectional differences between genotypes at the higher IL-6 concentrations of the affected patients vis-à-vis the lower average IL-6 concentrations of healthy controls). The examples were identified from a Pubmed search for interactions between genetic variants and conditions affecting IL-6 concentrations.

Figure 2.

Precision medicine perspective of -174G>C (rs1800795) genotype-specific differences in IL-6 concentrations (histogram inserts) vs. quantile-dependent expressivity perspective (line graphs showing larger -174G>C genotype differences when average IL-6 concentrations were high) for: a) Burzotta et al.’s [24] 2001 report on plasma IL-6 differences before and after elective coronary artery bypass graft (CABG) surgery (P=0.02); b) Brull et al.’s [15] 2001 report on serum IL-6 levels before and 24 hours after CABG surgery (P=0.02); c) Libra et al.’s [16] 2006 report on plasma IL-6 levels in T2DM patients with and without peripheral arterial disease (PAD); d) Sen et al.’s [17] 2011 report on serum IL-6 levels in Eales’ disease patients and healthy controls; e) Hulkkonen et al.’s [25] 2001 report on plasma IL-6 levels in patients with primary Sjògren’s syndrome and healthy blood donors; f) Talar-Wojnarowska et al.’s [18] 2009 report on serum IL-6 levels in patients with pancreatic adenocarcinoma and those with chronic pancreatitis.

For example, cardiac operations involving cardiopulmonary bypass produce a systemic inflammatory response due to surgical trauma, blood contact with the extracorporeal circuit, and lung reperfusion injury. Burzotta et al. [24] reported that elective coronary artery bypass graft (CABG) surgery produced a 17-fold increase plasma IL-6 concentrations overall, with GG homozygotes of the IL6 -174G>C polymorphism exhibiting significantly greater increases than C-allele carriers (P=0.02, Figure 2a histogram). Alternatively, the line graph of Figure 2a would interpret the genotype-specific IL-6 change to the larger cross-sectional difference between genotypes during the post-surgical inflammatory response when average IL-6 concentrations were high vis-à-vis the smaller cross-sectional difference between genotypes when average IL-6 concentrations were low (i.e., before surgery and at discharge). Similarly, Brull et al. [15] reported a significant time-by-genotype interaction following CABG surgery (P<0.04), where peak post-operative IL-6 concentrations 24 hours after surgery differed significantly between GG homozygotes and C-allele carriers (274.1±22.7 vs. 214.4±18.7, P<0.02) whilst pre-operative IL-6 concentrations did not (Figure 2b).

Inflammatory processes are also involved in the pathology of peripheral arterial disease (PAD) [60]. Libra et al. [16] reported that T2DM patients with PAD had significantly higher plasma IL-6 concentrations than those lacking PAD (12.1±0.9 vs. 4.0±0.3 pg/ml, P<0.01). They also reported that IL-6 concentrations were significantly higher in GG homozygotes than carriers of the C-allele in both PAD+ and PAD-patients, and that the IL-6 difference between PAD+ and PAD- in T2DM was significant for GG homozygotes (P<0.01) but not for carriers of the C-allele (Figure 2c histogram). From a quantile-dependent expressivity perspective, the associated line graph shows that the difference between GG homozygotes and C-allele carriers was greater at the higher average IL-6 concentrations of the patients with PAD.

Eales’ disease is a retinal vasoproliferative disease characterized by peripheral retinal perivasculitis during its inflammatory stage [17]. IL-6 plays a pivotal in the disease’s pathogenesis in modulating the acute phase inflammatory response during the inflammatory stage and angiogenic response during the proliferative stage [17]. Sen et al. [17] reported that during the inflammatory stage, patients with Eales’ disease had significantly higher average serum IL-6 concentrations than age- and sex-matched healthy controls (27.21±2.98 vs. 4.80±0.18 pg/ml, P<0.0001). The histogram in Figure 2d shows that the difference between affected and healthy patients was greater in GG homozygotes than carriers of the C-allele (a precision medicine interpretation), which in the line graph corresponds to a greater difference between genotypes in patients than controls (quantile-dependent expressivity interpretation).

Hulkkonen et al. [25] reported that compared to healthy blood donors, plasma IL-6 concentrations were significantly elevated (P<0.0001) in patients with primary Sjògren’s syndrome (a systemic autoimmune disorder that diminishes lachrymal and salivary gland secretory activity). They reported that the differences between -174 G>C genotypes were significant in affected patients (p<0.05) but not in blood donor controls (p<0.28), leading them to speculate that the polymorphism primarily regulates the IL-6 inducable response, but which is also consistent with quantile-dependent expressivity (Figure 2e line graph). The associated histogram shows that the net effect of the syndrome tended to be greater in carriers of the G-allele than CC homozygotes.

Eddahibi et al. [26] reported that IL-6 concentrations were higher in chronic obstructive pulmonary disease (COPD) patients than smoking controls (P<0.001, not displayed). Consistent with quantile-dependent expressivity, the –174G/C polymorphism was associated with IL-6 concentrations in COPD patients (higher concentrations in GG homozygotes than C-allele carriers) but not among controls.

Advanced cancer patients experience increase concentrations of IL-6 and other cytokines [61]. The IL-6 increase is characterized as a systemic phenomenon that is independent of the initial tumor histology, and that is reported to correlate with cancer prognosis in twenty-three different cancer types [61].

Talar-Wojnarowska et al. [18] reported that patients with pancreatic adenocarcinoma had higher IL-6 concentrations than those with chronic pancreatitis (36.3±0.40 vs. 7.9±0.32 pg/ml, p<0.01), particularly in GG homozygotes (Figure 2f histogram). Both conditions showed significantly higher IL-6 concentrations in GG homozygotes than C-allele carriers of the -174G>C polymorphism, with the genotype difference being greater at the higher average concentrations of the cancer patients.

Serum IL-6 concentrations also correlate with the progression and development of neuroblastoma, a common childhood extracranial solid tumor. Zhao et al. [27] reported that IL-6 concentrations were significantly higher in G-allele carriers than CC homozygotes at the higher average concentrations of the neuroblastoma patients (P=0.02, Figure 3a) but not at the lower concentrations of the healthy controls (P=0.19). In another report, Talaat et al. [28] reported significantly greater IL-6 concentrations in GG homozygotes than C-allele carriers at the higher average IL-6 concentrations of diffuse large B cell lymphoma patients but not at the low concentrations of matched controls (Figure 3b).

Figure 3.

Precision medicine perspective of -174G>C (rs1800795) genotype-specific differences in IL-6 concentrations (histogram inserts) vs. quantile-dependent expressivity perspective (line graphs showing larger -174G>C genotype differences when average IL-6 concentrations were high) for: a) Zhao et al.’s [27] 2018 report on serum IL-6 levels in neuroblastoma patients and healthy controls (P=0.19); b) Talaat et al.’s [28] 2015 report on the serum IL-6 levels in patients with diffuse large B cell lymphoma and matched controls; c) Wypasek et al.’s [29] 2010 report on serum IL-6 change due to CABG surgery; d) Galimudi et al.’s [30] 2014 report on serum IL-6 levels in Indian CAD patients and healthy controls; e) Boiardi et al.’s [31] 2006 report on serum IL-6 levels between polymyalgia rheumatica patients and healthy controls; and f) Kilpinen et al.’s [32] 2001 report on cord blood IL-6 levels between vaginally and elective caesarean section delivered full-term neonates.

The regulation of IL6 transcription appears to be cell type-specific and dependent on multiple polymorphisms within the promoter region that do not act independently but rather through complex interactions determined by haplotypes [2], which may explain several reports of higher IL-6 concentrations for the -174C allele (i.e., not the -174G allele). Nevertheless, they all show greater genotype differences at the higher IL-6 concentrations of the affected individuals. Included among them are Wypasek et al. [29] (Figures 3c) and Brull et al’s [15] (6-hr post operative, not displayed) reports of larger differences between -174G>C genotypes at the higher serum IL-6 concentrations post elective CABG surgery than at the lower pre-surgical IL-6 concentrations.

Galimudi et al. [30] reported significantly higher serum IL-6 concentrations in angiographically documented Indian CAD patients than healthy controls (13.4±0.28 vs. 5.9±0.1 pg/ml, P<0.05). The patient-control IL-6 difference was greatest in CC homozygotes, intermediate in GC heterozygotes, and least in GG homozygotes, and there was a significant difference between CC homozygotes and G-allele carriers at the higher concentrations of the patients (P<0.01) but not at the lower IL-6 concentrations of controls (Figure 3d).

Increased IL-6 production is a sensitive measure of polymyalgia rheumatica (PMR) severity, risk of recurrence and relapse, and need for more aggressive corticosteroid treatment [31]. Boiardi et al. [31] reported higher IL-6 concentrations in CC-homozygotes than G-allele carriers at the higher average IL-6 concentrations of PMR patients but smaller genotype differences at the lower average concentrations of the healthy controls, and correspondingly, a greater PMR-control difference in the CC-homozygotes (Figure 3e).

Kilpinen et al. [32] reported a greater difference in cord blood IL-6 concentrations between CC homozygotes and G-allele carriers (21.4 vs. 9.6 pg/ml, P<0.03) at the significantly higher average IL-6 concentrations of vaginally delivered full-term neonates (P<0.001) vis-à-vis the lower IL-6 concentrations of neonates delivered less stressfully by elective caesarean section (6.3 vs. 2.7 pg/ml, P=0.02, Figure 3f).

Potaczek et al. [33] reported that the IL6 -174G>C polymorphism modified the effects of 160 mg/day micronized fenofibrate on serum IL-6 concentrations in hypercholesterolemic patients (Figure 4a). Specifically the IL-6 reduction after 30 treatment days were significantly greater in carriers of the C allele than GG homozygotes (−1.78 vs. −0.59 pg/mL, P=0.04). Alternatively, the difference between genotypes was greater (1.49±0.57 pg/ml) at the higher average IL-6 concentration at baseline (2.47 ± 0.29 pg/ml) than after fenofibrate treatment (0.47 ± 0.32 pg/ml) when the average IL-6 concentration was reduced (1.37±0.16 pg/ml).

Figure 4.

Precision medicine perspective of -174G>C (rs1800795) and -572G>C (rs1800796) genotype-specific differences in IL-6 concentrations (histogram inserts) vs. quantile-dependent expressivity perspective (line graphs showing larger rs1800795 or rs1800796 genotype differences when average IL-6 concentrations were high) for: a) Potaczek et al.’s [33] 2009 report on the change in serum IL-6 concentrations due to 160 mg/day micronized fenofibrate; b) Sanderson et al.’s [34] 2009 report on the effect of social position on serum IL-6 concentrations; c) Brull et al.’s [15] 2001 report on the serum IL-6 change from before to 6-h post CABG surgery by -572G>C genotypes; d) Mälarstig et al.’s [35] 2007 report on the plasma IL-6 difference between acute coronary syndrome patients who experienced subsequent cardiovascular events and healthy controls; e) Zhang et al. ’s [19] 2012 report on the difference in plasma IL-6 concentrations between pulmonary tuberculosis cases and controls; f) Panayides et al.’s [20] 2015 report on the serum IL-6 difference between chronic renal disease patients with multiple organ dysfunction syndrome (MODS, failure ≥ two organs requiring intervention to maintain homeostasis) and non-MODS patients.

The higher IL-6 concentrations associated with low social position in phase 3 of the Whitehall II cohort of UK civil servents may be due to its association with older age, higher BMI, and smoking. [34]. The significant interaction reported by Sanderson et al. [34] between social position and IL6 -174G>C polymorphism (P_interaction=0.05) can either be interpreted as a genotype-specific effect of social position on IL-6 concentrations (Figure 4b histogram) or as larger genotype differences at the higher IL-6 concentrations of low position (line graph).

The rs1800796 -572G>C variation is also reported to reduce IL-6 promoter transcriptional activity [62]. For example, Brull et al. [15] reported greater genotype differences between C-allele carriers and GG homozygotes of the IL6 -572G>C polymorphism (355±67 vs. 216±13 pg/ml, P=0.03) 6-h after CABG surgery when serum IL6 concentrations were elevated (232±17 pg/ml) than pre-operation (4.9±0.3 vs. 5.2±0.1 pg/ml, P=0.47) when serum IL-6 concentrations were low (5.2±0.1 pg/ml, Figure 4c).

Mälarstig et al. [35] reported significantly higher plasma IL-6 concentrations in IL6 -572 CG than GG genotypes (P<0.01) for the higher average IL-6 concentrations of acute coronary syndrome patients who experienced subsequent cardiovascular events, but not at the lower average concentrations of healthy controls (Figure 4d).

Zhang et al. [19] studied the effect of the polymorphism on plasma IL-6 concentrations in pulmonary tuberculosis, an infectious disease caused by Mycobacterium tuberculosis. The IL-6 difference between pulmonary tuberculosis cases and controls was greatest for rs1800796 CC homozygotes, intermediate for CG heterozygotes, and least for GG homozygotes (Figure 4e histogram). Cross-sectional genotype differences were greater at the higher concentrations of the infected subjects. They also reported that in vitro stimulation of CD14+ monocytes with M. tuberculosis products secreted less IL-6 when isolated from GG than CC or CG subjects.

Panayides et al. [20] reported higher overall serum IL-6 concentrations in chronic renal disease patients with multiple organ dysfunction syndrome (MODS, failure ≥ two organs requiring intervention to maintain homeostasis) than non-MODS patients, due particularly to rs1800796 C-allele carriers (Figure 4f histogram). The difference between C-allele carriers and GG homogygotes was significant at the higher IL-6 concentrations of the patients with MODS (p=0.01) but not at the lower concentrations of those without MODS (p=0.18).

Circulating IL-6 concentrations are concordantly related to the presence and severity of pulmonary arterial hypertension (PAH) [63,64]. Fang et al. [21] reported that serum IL-6 concentrations were significantly greater in -572CC than CG or GG genotypes in both idiopathic PAH patients and healthy controls, with the line graph showing greater difference across genotypes at the higher average concentrations of the PAH patients (Figure 5a).

Figure 5.

Precision medicine perspective of -572G>C (rs1800796) genotype-specific IL-6 differences (histogram inserts) vs. quantile-dependent expressivity perspective (line graphs showing larger genotype differences when average IL-6 concentrations were high) for: a) Fang et al.’s [21] 2017 reported on the serum IL-6 difference between idiopathic PAH patients and healthy controls; b) Yuan et al.’s [22] 2015 report on the serum IL-6 differences between EV71-infected patients and healthy controls; c) Shin et al.’s [23] 2007 report on the serum IL-6 difference between smokers and nonsmokers; d) Zhang et al.’s [37] 2011 report of significant serum IL-6 differences between patients with T2DM and normal glucose regulation; and e) Yang et al.’s [56] 2020 report on the serum IL-6 difference between osteoarthritis patients and healthy controls; and f) Wypasek et al.’s [57] 2012 report on the serum IL-6 increase 5-7 days following CABG.

Finally, Bi et al. [36] reported significantly greater serum IL-6 concentrations in rs1800796 CC than GG homozygotes at the higher average IL-6 concentrations of cerebral palsy patients but not at the lower concentrations of controls (results not displayed).

Per the complex regulation of IL-6 expression discussed above [2], several papers report higher IL-6 concentrations for the opposite allele (-572G rather than the -572C allele), but nevertheless, show greater genotype differences for conditions that elevate IL-6 concentrations. They include: 1) Yuan et al. [22] report that serum IL-6 concentrations were significantly higher in G-allele carriers than CC homozygotes of the IL6 -572G>C polymorphism at the higher average concentration of the EV71-infected patients (p<0.001) but not at the lower average concentration of the control subjects (Figure 5b); 2) Shin et al. [23] data suggesting that the mean IL-6 difference between -572G>C GG homozygotes and C-allele carriers was greater at the higher mean IL-6 serum concentration of the smokers than at the lower mean IL-6 concentration of the nonsmokers (Figure 5c, line graph), as reflected in the greater net effect of smoking in the GG-homozygotes (histogram); and 3) Zhang et al. [37] report of significant differences between -572G>C genotypes at the higher serum IL-6 concentrations of T2DM patients (P=0.003) but not in those with normal glucose regulation (P=0.73, Figure 5d).

Several other genetic variants have also been reported to affect IL-6 concentrations. Yang et al. [56] reported that serum IL-6 concentrations were almost 70% higher in osteoarthritis patients than healthy controls (P<0.001). Consistent with quantile-dependent expressivity, IL-6 concentrations were significantly higher in CC than GG homozygotes of the -2954 G/C (rs12700386) polymorphism in the IL-6 promotor region in the osteoarthritis patients (P<0.05) but not at the lower average IL-6 concentrations of controls (Figure 5e line graph). In another study, Wypasek et al. [57] reported that T-allele carriers of the FGB -148C>T polymorphism showed greater increases in serum IL-6, fibrinogen, and C-reactive protein concentrations following CABG than in CC homozygotes. Figure 5f shows larger IL-6 differences between T-allele carriers and CC homozygotes (22.3±2.6 vs. 15.5±2.3 pg/ml, P=0.05) 5-7 days post-operatively when plasma IL-6 concentrations were elevated (20.1±2.4 pg/ml) than pre-op (both approximately 3.1, P=0.72) when plasma IL-6 concentrations were low (3.1±0.2 pg/mL).

Despite the numerous examples cited above, the -174G>C and -572G>C polymorphisms failed to attain genomewide statistical significance in the largest genome-wide association study of basal IL-6 concentrations to date (67,428 subjects of European ancestry) [3]. Two factors may have contributed to their failure. First, GWAS of basal IL-6 concentrations may miss those SNPs responsible for the stronger heritability above the 70^th percentile of IL-6 distribution (Figure 1), and the accentuated difference between -174G>C and -572C>G genotypes under conditions that elevate IL-6 concentrations (Figure 2–5). Second, if complex interactions determined by haplotypes indeed affect IL-6 expression, so that different alleles label the true underlying genetic IL-6 raising effect across studies, then neither allele would likely attain genome-wide significance when the genetic effect sizes are pooled across studies.

Logarithmically transformed IL-6

IL-6 heritability has always been reported for log-transformed concentrations. Without exception, this has been justified by the statistical requirement for estimating variance components, and never because log-transformed IL-6 concentrations show stronger genetic effects than untransformed IL-6 concentrations, and never because heritability of log IL-6 does not change over the range of the distribution. However, quantile-regression does not require statistical normality and therefore enables the estimation of quantile-dependent genetic effects for untransformed IL-6 concentrations as originally measured. Our analyses showed that quantile-dependent heritability increased significantly with the percentile of the IL-6 distribution for untransformed concentrations whilst remaining constant across percentiles when logarithmically transformed, implying genetic effects that are proportional to IL-6 concentrations. In fact, the current analyses are probably the first to explicitly report that heritability for log-transformed was greater than for untransformed IL-6 concentrations.

We believe that the quantile-specific genetic effects of Figure 1 for untransformed IL-6 concentrations are apropo to gene-environment interactions as traditionally reported (Figures 2–5). Specifically, reported interactions between genotype-specific IL-6 concentrations and health, treatment, and environment are almost always presented as untransformed mean [27,28,57,29.30,31,25,24,15,59,33,35,19,21,36,22,23,37,56] or occasionally, median concentrations [16,17,20,32], but never for their log-transformed values. Their significance levels are most commonly derived from parametric test of untransformed IL-6 concentrations [27,28,57,29,30,31,19,21,37]) and nonparametric tests [16,17,25,33,35,20,36,22,56,32], and only occasionally on log-transformed IL-6 concentrations [59,24,15, 23].

Limitations

None of the SNPs identified to date explain any more than a few percent of IL-6 heritability, which means that the effects of any particular SNP is not necessarily constrained by the results of Figure 1. We also caution that our use of Falconer’s formula may underestimate the complexity of genetic and nongenetic contributions to IL-6 concentrations. For example, the larger β_FS-derived heritability than β_OP-derived heritability estimates above the 75^th percentiles may reflect dominance variance or shared environmental effects.

Quantile-dependent expressivity is a novel concept. Not surprisingly, studies often do not present all the data required to assess it applicability, namely genotype specific means or medians by condition. For example, Takac et al. [65] reported that relative to controls, serum IL-6 concentrations were significantly higher in patients with Crohn’s disease (3.25 vs. 1.99 pg/mL, p<0.001) and ulcerative colitis disease (2.67 vs. 1.99 pg/mL, p<0.001), and that patients who were rs1800795 CC homozygotes had significantly higher IL-6 concentrations than carriers of the G-allele (4.62 vs. 2.46, p=0.05). In another study, Wielińska et al. [66] reported significantly higher serum IL-6 concentrations in rheumatoid arthritis patients than controls (31.5 ± 5.1 vs. 8.5 ± 0.3 pg/ml, P = 0.002). They also reported that patients who were rs1800795 CC homozygotes had significantly higher IL-6 concentrations than G-allele carriers (55.8±20.6 vs 25.6±3.9, P = 0.03). However, quantile-dependent expression could not be assessed in either paper because neither Takac et al. [65] nor Wielińska et al. [66] provided IL-6 genotype differences in controls.

In conclusion, our analyses of the Framingham family sets and published papers on conditions that alter IL-6 concentrations suggest that genetic effects are accentuated at higher percentiles of the IL-6 distribtion. This phenomenon, quantile-dependent expressivity, may explain genetic interactions involving IL-6 concentrations with a variety of disease states, including coronary artery disease, peripheral arterial disease, pulmonary arterial hypertension, Eales’ disease, Sjògren’s syndrome, osteoarthritis, adenocarcinoma, neuroblastoma, polymyalgia rheumatica, pulmonary tuberculosis, and enterovirus 71-infection, and post coronary artery bypass graft recovery.

Abbreviation key

β_FS

Full-sib regression slope

β_OP

Offspring-parent regression slope

BMI

Body mass index

CABG

Coronary artery bypass graft

CAD

Coronary artery disease

COPD

Chronic obstructive pulmonary disease

EV71

Enterovirus 71

GWAS

Genomewide association study

h²

Heritability in the narrow sense

IL-6

Interleukin-6

LPS

Lipopolysaccharide

MODS

Multiple organ dysfunction syndrome

PAD

Peripheral arterial disease

PAH

Pulmonary arterial hypertension

PMR

Polymyalgia rheumatica

T2DM

Type 2 diabetes mellitus

VEGF

Vascular endothelial growth factor