Associations of dicarbonyl stress with complement activation: the CODAM study

Chronic hyperglycaemia can induce dicarbonyl stress, which is characterised by increased presence of reactive α-dicarbonyl compounds and AGEs [1, 2]. Prolonged exposure of proteins to α-dicarbonyl compounds may change their normal function and/or their susceptibility to enzymatic degradation [3]. The complement system is part of the innate immune system and has been implicated in various cardiometabolic diseases (as reviewed in [4]). Interestingly, glycation of complement inhibitory proteins may affect complement activation [5‐8], which may contribute to the development of vascular complications in people with diabetes [7].

During complement activation, C3 convertases that are generated can cleave the central component of the complement cascade, C3, into C3b and the anaphylatoxin C3a (as reviewed in [9], see Fig. 1). C3b can induce subsequent activation of the common terminal complement pathway resulting in the formation of C5b-9, also known as the membrane attack complex (MAC). Uncontrolled complement activation can damage host cells and/or hyperactivate inflammatory pathways [10]. To prevent this, the complement system is under strict control by circulating and cell-surface inhibitors (as reviewed in [9, 11]).

Decay accelerating factor (DAF) is a membrane-bound inhibitor of complement activation (see Fig. 1). It prevents formation of complement convertases and accelerates their decay, thereby inhibiting the generation of C3a and C3b [12]. Recently, it was shown that glycated DAF is present on erythrocytes of individuals with diabetes and this was accompanied by a less efficient control of complement activation [5]. The function of another membrane-bound inhibitor of complement activation, CD59, was also hampered by glycation [7, 8, 13]. CD59 controls the final step of terminal pathway activation (as reviewed in [9] and shown in Fig. 1). In vitro studies demonstrated that glycation of CD59 impaired its inhibitory effect on C5b-9/MAC-related lysis of human erythrocytes [13]. Also, erythrocytes of participants with diabetes were more susceptible to complement-related lysis, likely as a result of glycation-induced inactivation of CD59 [8]. Notably, in the kidney and nerves of participants with diabetes, glycated CD59 co-localised with C5b-9, and presence of glycated, less functional CD59 was considered to be the cause of increased C5b-9 deposition [7]. Glycated CD59 was also higher in the circulation of diabetic than of healthy individuals [6], and was detected in the urine of participants with diabetes [13]. Glycation of other complement components, such as C3 [3, 14‐17] and factor B [18] has also been reported, but biological effects of glycation on their function, if any, have not yet been reported.

Taken together, existing data suggest that dicarbonyl stress and subsequent glycation of complement inhibitors may impair their regulation on complement activation. In this study we investigated to what extent dicarbonyl stress, as reflected by the presence of dicarbonyls and AGEs in the circulation, is related to complement activation, as represented by the plasma concentrations of two complement activation products, i.e. C3a and soluble (s)C5b-9.

The inverse association between GO and sC5b-9 was an unexpected observation in the light of our pre-specified hypothesis. Notably, this relationship was attenuated and no longer significant after adjustment for C3, and after exclusion of participants with acute or chronic infections. This attenuation suggests that this association was, at least partly, due to an ongoing inflammatory process that is somehow related to lower plasma GO concentration and/or that plasma GO may have anti-inflammatory properties. GO is generally considered a proinflammatory compound, although information on its association with inflammation in humans is scarce. One experimental study showed that exogenous GO can induce inflammatory injury in human vascular endothelial cells [30]. We previously showed that plasma sC5b-9 is positively associated with low-grade inflammation [27]. Our current observation might thus be explained by a mechanism similar to what was previously reported for CML [31]: plasma GO may be transported to tissue or interstitial fluid where it can contribute to complement activation and the inflammation process. Therefore, the decrease in plasma GO, as a result of increased uptake into tissue, may contribute to local complement activation and related inflammation that is subsequently reflected in the plasma compartment.

The inverse association of protein-bound CML with C3a was largely explained by obesity. We and others previously showed that plasma protein-bound CML was inversely associated with central obesity, at least partly because it was trapped by visceral adipose tissue via the receptor for AGEs (RAGE) [31‐33] which, in combination with the positive association between plasma C3a and obesity [34, 35], may explain the current inverse association. Serum pentosidine was also inversely correlated with BMI in one recent human study [36]. Given that pentosidine can also bind to RAGE [37], a similar obesity-dependent inverse association may underlie the inverse association between pentosidine and C3a, which was confounded by central obesity.

The inverse association between protein-bound CEL and C3a was independent of all confounders included in our analyses. Interestingly, this association was only present in individuals without diabetes and not in those with diabetes. Thus, this inverse association might be diminished in hyperglycaemia/diabetes. In contrast, in diabetes, strong inverse associations of free and protein-bound CML and CEL with sC5b-9 were seen. Although the underlying mechanism of these unexpected diabetes-specific associations is not clear yet, this may suggest that the associations of these dicarbonyls and AGEs with C3a and sC5b-9 occurs via different routes.

In line with our hypothesis, we observed a consistent positive association between protein-bound pentosidine and sC5b-9. This association was independent of possible confounders, was not influenced by diabetes, and remained significant in all sensitivity analyses. A possible explanation is that protein-bound pentosidine may, better than the other AGEs in our analyses, reflect the tissue-AGE content. In that line, we previously reported that skin autofluorescence (a non-invasive measurement of skin AGE accumulation) and plasma protein-bound pentosidine were both positively associated with aortic stiffening, while protein-bound CML and CEL were not [38]. Plasma pentosidine also correlated with skin autofluorescence in Japanese haemodialysis patients [39]. Yet another small case–control study also reported a non-significant positive association between serum protein-bound pentosidine and pentosidine levels in skin biopsies [40]. The positive association between plasma protein-bound pentosidine and sC5b-9 that we observed may thus reflect a potential positive relationship between protein glycation in tissues and complement activation.

Overall, our observations reveal distinct associations of plasma dicarbonyl compounds and AGEs with C3a and sC5b-9. This suggests that, instead of being a general reflection of dicarbonyl stress, these plasma markers may each reflect distinct pathophysiological processes. Indeed, different biological effects for these compounds have been reported. For instance, MGO and GO may trigger distinct intracellular signals involved in various cellular functions in cultured endothelial cells, because of differences in their chemical structures [41]. The possibility of different contributions for AGEs in the development of vascular disease was also reported [38]. For instance, via binding to RAGE, AGEs like CML can activate inflammatory pathways (as reviewed in [42]), while cross-linking AGEs such as pentosidine may act via formation of cross-links between extracellular matrix proteins [43, 44]. In addition, the different associations we observe for C3a and sC5b-9 could be explained by the fact that complement regulators differ in their exclusivity for generation of C3a and C5b-9. CD59 is the dedicated regulator of the formation of C5b-9/MAC, while there are several other complement regulators that could compensate the inhibition of DAF on C3 activation.

Plasma dicarbonyls produce a snapshot of current dicarbonyl stress while AGEs rather represent dicarbonyl stress over the last days or weeks. In our additional analyses, we included a measure that may reflect lifelong differences in dicarbonyl stress, i.e. potentially functional variants of the MGO-detoxifying enzyme GLO1. Some human studies on GLO1 rs2736654 reported that individuals with AA (i.e. TT) genotype have the best GLO1 function/activity [22, 23]. This implies that individuals with the AC (i.e. GT) or CC (i.e. GG) genotype may have been exposed to lifelong greater dicarbonyl stress which, according to our hypothesis, would lead to more complement activation. rs2736654 was not associated with C3a concentrations. Moreover, individuals carrying the genotype with the lowest predicted GLO1 activity (i.e. the GG genotype) had lower sC5b-9 concentrations, although non-significant. rs1049346 was also not associated with C3a concentration, whereas individuals with AG or AA genotypes of rs1049346, which were reported to have lower GLO1 enzyme activity [24], had significantly lower sC5b-9 concentrations. Critical re-evaluation of the available literature shows that the claims on functionality of these polymorphisms were partly based on in vitro data and partly inconsistent [22‐24, 45, 46]. This hampers the interpretation of the relatively strong association of rs1049346 with C5b-9.

The main strength of our study is the availability of plasma concentrations of α-dicarbonyls, free AGEs and protein-bound AGEs, as well as complement activation products, within one well-phenotyped cohort. The detailed phenotyping of the study population provided us with the opportunity to thoroughly evaluate the effects of potential confounders and perform relevant sensitivity analyses. Our study also has several limitations. Most importantly, despite the information we have on the overall concentrations of dicarbonyl compounds and AGEs for each participant, we do not have information on the extent to which their individual complement regulators were actually carbonylated and/or glycated. Other limitations include the cross-sectional design, which prohibits conclusions on causality. We aimed to mitigate this limitation by including the functional GLO1 polymorphisms, which theoretically represent a lifelong exposure to greater dicarbonyl stress, at least to MGO. Further exploration on the effects of GLO1 polymorphisms on GLO1 function and dicarbonyl stress is needed, since we cannot exclude the possibility that the relatively small sample size of our cohort may have limited the power of the present study to detect associations of GLO1 polymorphisms with complement activation. Moreover, in the analyses stratified for presence of diabetes, the smaller sample size in subgroups may have further decreased the statistical power to identify relevant relationships. Finally, our participants are middle-aged to older Europid individuals characterised by a moderately increased risk of cardiovascular disease. This selection of the study population limits the generalisability of present findings. Therefore, it is important that these evaluations will, in due time, be confirmed in a population-based cohort and among other ethnicities.

In conclusion, plasma concentrations of dicarbonyl stress markers displayed various associations with complement activation. This suggests different biological effects of the individual plasma α-dicarbonyls and AGEs, potentially with different clinical relevance. Our most striking findings are the unexpected inverse associations between GO and sC5b-9, between CEL and C3a. Moreover, protein-bound and free CML and CEL were inversely associated with sC5b-9, but only in diabetes. In addition, a genotype that may represent less GLO1 activity (hence greater dicarbonyl stress) was associated with less activation of the terminal pathway. Protein-bound pentosidine, on the other hand, was positively and significantly associated with sC5b-9, which was in line with our pre-specified hypothesis. Taken together, these data illustrate that the complex underlying physiological processes of these circulating markers should be taken into account in future work. The consistent associations observed for plasma pentosidine indicate that it may have priority as a marker for tissue AGEs. Moreover, the observed inverse associations between AGEs and complement activation, as well as the potential interactions with diabetes, need to be confirmed and extended in a larger study population.