Our study demonstrates that in patients with CKD FGF23 is significantly associated with smoking and proteinuria (Figures 1 & 2). This association was strong and persisted after correcting for other well known cardiovascular and renal risk factors. In the fully adjusted model (Table 3) both smoking and proteinuria had a relatively high impact on FGF23 levels that was exceeded by eGFR and PTH only. Our data confirm the established relationship between FGF23 on the one hand and phosphate level, eGFR and PTH on the other . This relationship possibly has a biological basis as FGF23 is physiologically regulated by phosphate levels , phosphate loading  and PTH .
Our observation of the association between smoking and FGF23 is novel. Possible explanations for this association are an influence of smoking on FGF23 production, an influence of smoking on FGF23 sensitivity or mere confounding by other factors that directly influence FGF23 levels. Smoking may increase FGF23 levels by affecting the metabolism of bone, the primary source of FGF23. Indeed, several studies have observed an association of smoking with osteoporosis [17, 18]. However, there are no experimental data to support the assumption that smoking increases FGF23 production. Furthermore, if smoking directly increases FGF23 levels, we would have expected lower phosphate levels in smokers, which were not observed. Therefore, we consider it more likely that smoking reduces FGF23 sensitivity, thus necessitating an increased production to maintain phosphate excretion. The FGF23 receptor consists of heterodimerized FGFR1 and klotho, and is primarily located in the parathyroid and distal tubular cells in the kidney . The expression of klotho is downregulated by oxidative stress , leading to dismantling of a functional FGF23-receptor. For this reason it is possible that smoking, via increased oxidant stress, decreases the number of functional FGF23 receptors in the kidney . We cannot exclude the possibility that the association between FGF23 and smoking resulted from confounding by factors such as other life style differences between smokers and non-smokers. In contrast to others  we did not find a relation between body mass index and FGF23, in the full-adjusted model. The population studied by Marsell however was older, had better estimated GFR, and lower levels of FGF23. The putative relative contribution of BMI on FGF23 levels in the MASTERPLAN cohort may be overwhelmed by other factors, shown in Table 3. The relation we found between smoking and levels of FGF23 is in line with a recent study that addressed the predictive value of FGF23 on death and progression to end stage kidney disease . In that study, the proportion of smokers increased in the higher quartiles of FGF23, however, the quantitative effect of smoking on FGF23 was not assessed. Correcting for several factors including smoking did not mitigate the strength of the association between FGF23 and the risk of death. However this does not imply that smoking cannot increase levels of FGF23. In another recent study in the general population, the percentage smokers did not differ between tertiles according to levels of FGF23 . Whether higher FGF23 levels exist within individual tertiles among smokers, as compared with non-smokers, was not assessed.
The second major finding in our study is the graded relationship of FGF23 with proteinuria (Figure 2). This association can be explained in several ways. First it is possible that FGF23 has a direct effect on the glomerular filtration barrier, possibly through a direct effect on glomerular endothelial function  or glomerular hemodynamics. However, the evidence supporting this hypothesis is weak. Hypertension was not related to FGF23 and therefore cannot explain the rise in proteinuria with increasing FGF23 levels. Another explanation may be that proteinuria itself increases FGF23. This may be mediated by the association of proteinuria with vitamin D homeostasis . Proteinuria however is associated with vitamin D deficiency and therefore a decline in FGF23 would have been expected, giving the observed increase in FGF23 levels seen upon supplementation with activated vitamin D in experimental models . As patients with proteinuria received active vitamin D more frequently (shown in Table 4), this use of active vitamin D was added to the fully adjusted model (Table 3). Active vitamin D use however, did not explain the higher levels of FGF23 in patients with proteinuria. Another potential mechanism that could link proteinuria with FGF23 levels is secondary hyperparathyroidism, induced by deficiency of vitamin D that accompanies proteinuria. Adding lnPTH to the fully adjusted model however, did not mitigate the association of proteinuria with FGF23. A more likely explanation is that the toxic effects of proteinuria on tubular function [27, 28] also disrupt local FGF23 signaling, leading to a compensatory increase in FGF23 production. A final explanation might be that both proteinuria and an increase in FGF23 are the consequence of the same underlying process, such as increased oxidative stress, that could lead to both proteinuria  and FGF23 resistance . An association between proteinuria and FGF23 was also noted in the Heart and Soul study . This study included patients with a mean GFR of 76 ml/min, and only 4% had macroalbuminuria. Our study, with mean GFR of 37 ml/min/1.73 m2 and over 10% of patients having proteinuria > 2 g/day, extends these observations to patients with more severe CKD and higher levels of proteinuria. Fliser and co-workers found an association between FGF23 and progression of CKD, even after correction for albumin-to-creatinine ratio (ACR), but that correction attenuated the predictive value of FGF23 . In that study the precise relation between FGF23 and proteinuria was not studied. In the study by Isakova the association between FGF23 and progression to end stage kidney disease was no longer significant after correcting for albumin-to-creatinine ratio, eGFR, albumin and hemoglobin, but the risk for mortality remained significant .
We found no modulating effect of either 25D or 1,25D levels on FGF23. This lack of evidence for an association of FGF23 with levels of vitamin D metabolites can be explained by the fact that active vitamin D induces FGF23 production  while FGF23 in turn catabolizes 25D and 1,25D , thus counterbalancing possible relationships between vitamin D and FGF23.
It needs to be pointed out that some associations present in Model 2 were no longer present in the analysis in which vitamin D was included. There are two factors to explain this. Firstly the number of participants in this latter analysis is much lower (604 in model 2; 126 in vitamin D analysis). In addition proteinuria was dichotomized in two categories (more or less than 2 gr/24 hours), which reduces the discriminatory value of this variable in linear regression.
Accumulating evidence points to FGF23 as a novel risk factor for mortality and progression of CKD [3, 23]. Our findings show that a part of the increased cardiovascular risk of smoking  and proteinuria  in CKD may thus be explained by its effects on FGF23, assuming that FGF23 has direct detrimental effects. This may indicate that FGF23 has a causal role in cardiovascular complications in CKD, but this still is still debated, because of the possibility of residual confounding within these studies and the lack of definitive proof of a pathobiological mechanism by which FGF23 presumably induces harm. Recent evidence however does suggest that FGF23 is directly involved in the evolution of pathological left ventricular hypertrophy in mice . For some factors, like GFR[1???] and phosphate level [34, 35], both reasonably established novel risk factors for cardiovascular disease, its relationship with FGF23 is well described. However, previous studies demonstrated that the predictive value of FGF23 for clinical endpoints remained significant, even after correcting for these factors that are involved in its production, providing additional arguments for the notion that FGF23 might be more than just a risk marker.
Some limitations of our study need to be mentioned. First, our study is a cross-sectional observational analysis, thus we cannot determine temporal relationships and causality. Next, it could be possible that FGF23 influences proteinuria, instead of the other way around. Indirect evidence for the latter comes from the observation that protein (and phosphate) restriction improves albuminuria . In addition, we also cannot answer the clinically relevant question whether modulating proteinuria or smoking cessation can lower FGF23 and if FGF23 decline accomplished by these measures reduces cardiovascular risk (independent from risk reduction by smoking cessation alone and proteinuria reduction). Such a question requires a RCT.
There are currently two techniques available to measure FGF23: intact FGF23 and c-terminal FGF23. The former assay is positive only in the presence of full-length FGF23, while the latter also measures the c-terminal truncated FGF23. In our study we did not measure intact FGF23, but in advanced CKD the two assays have been shown to perform similarly . We were able to test 25D and 1,25D levels in a subset only, which weakens our conclusion that vitamin D metabolites had no relation with FGF23. Indeed, a recent study suggests that higher FGF23 was associated with lower 1,25D levels . However, proteinuria was not measured in that study.
Strengths of our study are the large sample size, the prospective nature of data collection and the use of 24 hour urine collections instead of using albumin-to-creatinine ratio in spot urine samples, which provides more reliable quantitative data .
Our findings can have important implications. If FGF23 indeed has a causative role in cardiovascular disease and is involved in progression of CKD, an important next step would be to test the hypothesis that dietary or pharmacological interventions aimed at a reduction of FGF23 improves these outcomes. The obvious tools to lower FGF23 would be a phosphate-restricted diet , phosphate-binder therapy [39–41] or calcimimetics . Our findings support the hypothesis that early intervention aiming at reducing proteinuria and cessation of smoking may be another tool to reduce FGF23.