Renoprotective effects of genetically proxied fibroblast growth factor 21: Mendelian randomization, proteome-wide and metabolome-wide association study

Background: Fibroblast growth factor 21 (FGF21) has demonstrated efficacy for reducing liver fat and reversing non-alcoholic steatohepatitis in phase 2 clinical trials. It is also postulated to have anti-fibrotic effects and therefore may be amenable to repurposing for the prevention and treatment of chronic kidney disease (CKD). Methods: We leverage a missense genetic variant, rs739320 in the FGF21 gene, that associates with magnetic resonance imaging-derived liver fat as a clinically validated and biologically plausible instrumental variable for studying the effects of FGF21 analogs. Performing Mendelian randomization, we ascertain associations between instrumented FGF21 and kidney phenotypes, cardiometabolic disease risk factors, as well as the circulating proteome (Somalogic, 4907 aptamers) and metabolome (Nightingale platform, 249 metabolites). Results: We report consistent renoprotective associations of genetically proxied FGF21 effect, including higher glomerular filtration rates ( p = 1.9 × 10 (cid:0) 4 ), higher urinary sodium excretion ( p = 5.1 × 10 (cid:0) 11 ), and lower urine albumin-creatinine ratio ( p = 3.6 × 10 (cid:0) 5 ). These favorable effects translated to lower CKD risk (odds ratio per rs739320 C-allele, 0.96; 95%CI, 0.94 – 0.98; p = 3.2 × 10 (cid:0) 4 ). Genetically proxied FGF21 effect was also associated with lower fasting insulin, waist-to-hip ratio, blood pressure (systolic and diastolic BP, p < 1.0 × 10 (cid:0) 07 ) and blood lipid (low-density lipoprotein cholesterol, triglycerides and apolipoprotein B, p < 6.5 × 10 (cid:0) 24 ) profiles. The latter associations are replicated in our metabolome-wide association study. Proteomic perturbations associated with genetically predicted FGF21 effect were consistent with fibrosis reduction. Conclusion

triangulate these findings, towards possible clinical development of FGF21 towards the treatment and prevention of kidney disease.

Introduction
Fibroblast growth factor 21 (FGF21) is a 181 amino acid peptide hormone, which is mainly produced by the liver and circulated in the blood in response to various extrinsic stressors including protein deficiency, high fructose intake, alcohol consumption, cold exposure, and exercise [1][2][3][4][5][6].FGF21 has broad effects on metabolism across organ systems, and administration of FGF21 analogues has shown promise for reducing hepatic fat and fibrosis related to non-alcoholic steatohepatitis (NASH) [7][8][9].Given its pleiotropic effects, FGF21 analogs are likely to offer therapeutic benefit in the prevention and treatment of other disease processes [5].
Chronic kidney disease (CKD) is a progressive disease characterized by reduced kidney function or structural damage that is present for more than three months [10].CKD affects 10-15 % of the global adult population and has far-reaching adverse metabolic and cardiovascular effects mediated by hematological, inflammatory, and hormonal dysregulation [10,11].Fibrosis is a key complication of CKD, and because of the known anti-fibrotic effects of FGF21, it follows that FGF21 may specifically be amenable to repurposing for the prevention and treatment CKD.
The basic science and epidemiological data supporting effects of FGF21 on risk and progression of CKD are mixed [12].Particularly given difficulties associated with translating animal data, applications of human genetic data offer several advantages towards establishing causal effects of FGF21 on kidney outcomes [13].Relevant to the target organism (i.e., humans), genetic variants that predict the effect of altering FGF21 signaling can be leveraged as instrumental variables within the Mendelian randomization framework for studying the broad effect of therapeutically targeting this mechanism [14].The random allocation of such genetic variants at conception means that their associations are less susceptible to the reverse causation and confounding biases that can otherwise hinder causal inference in traditional epidemiological study designs.
In this work, we advance our previous efforts [15] to identify a plausible genetic instrument for FGF21 based on its known clinical effects.We then leverage this genetic variant in the Mendelian randomization paradigm to explore the effects of FGF21 on kidney outcomes, and further re-visit associations with cardiometabolic traits and disease risk using the largest available data sources.Finally, we perform proteome-wide and metabolome-wide association studies using our FGF21 instrument, to explore potential biomarkers and mediators of its effects.This work has direct implications for future clinical development of FGF21 analogs and provides evidence relating to its efficacy for CKD prevention.

Study overview
This study used a two-sample unweighted Mendelian randomization approach wherein we identified a clinically plausible genetic instrumental variable for the exposure of interest (FGF21 effect) and related that instrument to each outcome using genome-wide association study (GWAS) summary statistics.In primary analyses, we investigated the association of instrumented FGF21 with clinical markers of kidney function and the risk of CKD.Our secondary analyses extended our outcomes of interest to cardiometabolic disease risk and risk factors including heart failure, coronary artery disease (CAD), stroke, type 2 diabetes mellitus (T2DM), anthropometric traits, alcohol consumption, dietary intake, blood pressure, glycemic traits, and blood lipids.Finally, we explore the wider systematic effects of instrumented FGF21 on the plasma proteome and metabolome in a complementary proteome-wide and metabolome-wide genetic association study.

Instrumental variable selection
Because of the known effects of FGF21 analogues on liver fat [7][8][9], we used a GWAS of magnetic resonance imaging-derived liver fat percentage on 36,703 individuals of European ancestry to identify a genetic instrument for the effect of FGF21 [16].We considered variants at the FGF21 locus (chr19:49,258,816-49,261,587 ± 100 kb, build GRCh37/ hg19) that independently associated with liver fat at significance level below a Bonferroni-adjusted threshold accounting for the number of FGF21 variants [i.e., p < 0.05/57 (p < 0.0009), r 2 < 0.1].
Our proteome-wide analysis incorporated relative concentrations of 4907 protein-binding aptamers, measured using the SomaLogic SomaScan v.4 platform [29].Finally, our metabolome-wide analysis included 249 plasma metabolites and metabolite ratios represented on the Nightingale platform [30].

Statistical analysis
We scaled all results per additional C allele (major allele) of rs739320 to mimic the effect FGF21.For binary outcomes, association measures are expressed as odds ratios and 95 % confidence intervals (OR, 95 % CI) per C allele.Continuous outcome measures are reported as standard deviation (SD) unit difference (95 % CI) per C allele, except for associations with eGFR crea , eGFR cys , UACR and BUN, which are expressed as change per log unit.We accounted for multiple hypothesis testing by setting the following Bonferroni-adjusted thresholds to determine statistical significance: p < 0.01 for primary outcomes (6 tests), p < 0.003 for secondary outcomes (20 tests), p < 1.0 × 10 − 5 for proteomic associations (4907 tests), and p < 2.0 × 10 − 4 for metabolite associations (249 tests).

Results
We identified a single missense variant, rs739320, that independently associated with liver fat below a Bonferroni-adjusted threshold of p < 0.0009.The C allele of rs739320 associated with a 0.23 % (95 % CI, 0.10-0.36;p = 6.4 × 10 − 04 ) lower liver fat content in UK Biobank participants (median [interquartile range] liver fat percentage = 9.9 [7.1-14.2]and 2.0 [1.5-2.9] in those with and without diagnosed hepatic steatosis, respectively) [16].Table 2 summarizes the associations of rs739320, as an instrument for the effects of FGF21 analogs, with each of the primary and secondary outcomes of interest.
Overall, rs739320 was associated with more favorable kidney function profiles (Fig. 1), including higher eGFR crea , eGFR cyst and urinary sodium, and lower UACR.Collectively, these translated to a 4 % reduction in the odds of CKD per C allele (OR [95%CI] = 0.96 [0.94-0.98],p = 3.2 × 10 − 04 , Fig. 2).Similarly, the rs739320 C allele favorably associated with various components of cardiometabolic health, including a lower waist-to-hip ratio, lower systolic and diastolic blood pressure, lower LDL-C, ApoB, and triglycerides, and lower fasting glucose and insulin levels (Fig. 3).These associations contributed to a nominally significant lower risk of stroke (p = 0.007, Fig. 2), but not heart failure, T2DM or CAD.In terms of lifestyle factors, instrumented FGF21 was associated with reduced alcohol consumption and protein intake.We did not observe Bonferroni-adjusted significant associations with body mass index, HDL-C, HbA1c or with the amount of daily carbohydrate, sugar, or fat intake.
Of the 249 tested plasma metabolites, 136 were associated with instrumented FGF21 (Fig. 4, Table S1).Instrumented FGF21 was associated with lower ApoB, and with lower LDL and TG across particle sizes and subtypes.In line with our secondary analysis, instrumented FGF21 did not significantly associate with HDL.Instead, we observed associations between instrumented FGF21 and the lipid compositions of HDL, specifically with a reduced amount of TG within HDL, as well as a higher Abbreviations: ApoB = apolipoprotein B, CARDIoGRAMplusC4D = Coronary artery disease genome wide replication and meta-analysis plus the Coronary artery disease genetics consortium, CKDGen = Chronic kidney disease genetics consortium, DIAMANTE = Diabetes meta-analysis of trans-ethnic association studies, eGFR cyst/crea = glomerular filtration rate estimated using cystatin-c/ creatinine, GIANT = Genetic investigation of anthropometric traits, HbA1c = glycated hemoglobin, HDL-C = high-density lipoprotein cholesterol, HERMES = Heart failure molecular epidemiology for therapeutic targets, ICBP = international consortium of blood pressure, LDL-C = low-density lipoprotein cholesterol, MAGIC = meta-analyses of glucose and insulin-related traits consortium, SD = standard deviation, UACR = urine albumin-creatinine ratio, UKBB = UK Biobank.The log reported units correspond to the change in natural logarithm transformed levels of the phenotype per effect allele.The OR reported units correspond to the change in odds ratio for the phenotype per effect allele.Notes: Estimates are interpreted as the difference in specified units ( # ) per C allele.Bold p-value indicates statistical significance (p < 0.01 for primary outcomes, and p < 0.003 for secondary outcomes).No data transformations were performed in the analysis, other than in the presentation of results for binary traits as odds ratios rather than log odds ratios, to improve their interpretability.

Discussion
In this work, we leveraged a missense genetic variant in the FGF21 gene that predicts the known clinical effect of FGF21 analogs, i.e., reduced liver fat [7][8][9], as an instrumental variable for studying the effect of FGF21 on kidney disease, as well as the broader cardiometabolic profile.These analyses provided new genetic evidence that FGF21 signaling may reduce CKD risk and have favorable effects across a broad range of renal outcomes, including eGFR, urinary sodium excretion, BUN and UACR.Furthermore, we provide genetic support for the clinically documented beneficial effect of FGF21 on fasting insulin and apolipoprotein B; and reproduced previously described MR associations of a favorable FGF21 effect on WHR, blood pressure, LDL-C and TG [15].By complementing these data with a Mendelian randomization analysis of the effect of FGF21 on the plasma proteome and metabolome, we explore the biomarkers, mediators and mechanisms that drive these effects.
Our primary findings are consistent with emerging data from animal models [31] and add to the growing body of evidence supporting protective effects of FGF21 on kidney function [12].While elevated FGF21 levels are often observed in patients suffering from kidney disease [32] this likely reflects the compensatory stress-induced mechanism through which disease-affected tissues become the source of FGF21 release, as opposed to its target [12].
The beneficial effect of higher plasma FGF21 to delay CKD progression is supported by pre-clinical data in murine models of diabetic nephropathy and free fatty acid-induced nephropathy (lipotoxic model), where FGF21 administration has been shown to reduce proteinuria, suppress of mesangial expansion, and inhibit fibrosis [33,34].In agreement with these models, we observed higher levels of two fucosyltransferase proteins (FUT3 and FUT5), with published Mendelian randomization evidence supporting a potentially causal role of FUT3 specifically [35].We also observed lower levels of the pro-fibrotic proteins, galectin-3 and galectin-9, both of which are targets of ongoing clinical trial investigations for the patients with NASH, pulmonary fibrosis, and cancer [36].
The biological activity of FGF21 is mediated via the FGF21 receptor complex consisting of a FGF receptor and the transmembrane coreceptor beta-klotho.KLB protein expression is observed in tubules and in glomeruli cells [37] and a direct effect on kidney cells cannot be excluded.Furthermore, FGF21 has in mice been shown to protect against angiotensin II-induced hypertension and vascular impairment by activation of ACE2/angiotensin [6,8], which may explain the lower blood pressure observed in carriers of the rs739320 C allele.
The favorable effects of genetically-proxied FGF21 on clinical lipid profiles observed in our secondary analysis are aligned with numerous animal studies, and outcome data from phase 2 trials of FGF21 analogs where LDL-C and TG are consistently reduced [38].Our proteomics analyses supported a similar reduction in proteins involved in triglyceride and fat metabolism, e.g., pancreatic lipase-related protein 2, pancreatic triacylglycerol lipase, lipase cofactor colipase, and cholecystokinin.These data provide support for the development of FGF21 analogues for hypertriglyceridemia.In contrast, our lack of association with HDL-C contrasts findings from these same studies and trials.Our ability to leverage comprehensive molecular data on plasma lipids and proteins allowed us to further explore this discrepancy.The lack of an HDL-C association was echoed in the null finding for apolipoprotein A1 (the favored carrier of HDL), and a lack of an association with the amount of cholesterol, cholesteryl esters and free cholesterol contained within HDL particles.This difference may reflect limitations of our Mendelian randomization model for studying the clinical effects of an FGF21 analog.Specifically, Mendelian randomization considers the effects of small changes in FGF21 across the life course, where in clinical studies, far greater changes in FGF21 are exerted at discrete time points in later life.
We provide support for an association between higher genetically proxied FGF21 and lower alcohol consumption, and higher protein intake.We complement the latter association by reporting that instrumented FGF21 also associates with higher levels of plasma branchedchain amino acids, which may be secondary to an increase in protein intake or a decrease in ureagenesis due to lower plasma glucagon its effects on metabolism [15,39].Of note, our findings for potential effects of FGF21 on lowering fasting insulin levels but potentially increasing fasting glucose levels are novel.This contrasts with a recent phase 2b clinical trial in patients with NASH, where efruxifermin decreased both glycemia and HbA1c by 0.5 % lowered fasting C-peptide, indicating an increase in insulin sensitivity.These contrasting associations might indicate that FGF21 therapy in trials, wherein plasma FGF21 levels are increased 100-1000-fold more than we can study using populationbased physiological differences, may have added benefits via overcoming a previously hypothesized FGF21 resistance [40].
Our work has several strengths.The Mendelian randomization paradigm is more robust to the confounding from environmental factors and reverse causation that can hinder causal inference in traditional epidemiological study designs [14,41].Further, our use of a genetic variant known to affect the amino acid sequence of FGF21 (leucine/ proline at position 174) that also relates its known clinical effects adds support to its validity as a functionally and clinically plausible genetic instrument for the effect of FGF21 [14].The use of large-scale genetic association data within a two-sample Mendelian randomization framework allowed us to efficiently investigate a diverse array of outcome measures that add mechanistic insight.In this way, consideration of the proteomic and metabolomic panels may be used to inform on mediating mechanisms and selection of biomarkers for target engagement in earlystage clinical trials.
Our work also has limitations.Notably, we did not use the lead variant (rs2548957) for circulating FGF21 protein levels identified in a GWAS of >10,000 Swedish individuals as the genetic proxy for FGF21 effect [42].This was because this variant does not recapitulate known clinical effects of FGF21 analogues on liver fat when employed in MR analysis [7][8][9]42], suggesting that it is not a valid instrument for the effect of FGF21.The discrepancy between our instrument and the lead variant for circulating FGF21 levels may be attributable to genetic confounding from the effects of neighboring genes such as FUT1 on circulating FGF21 levels, which may also be affecting the considered outcomes traits through pathways unrelated to FGF21 effects.Related to this, it is conceivable that our proposed genetic instrument for FGF21 may be exerting at least some of its effects on liver fat and the other considered traits through pathways unrelated to FGF21.While this possibility can never be completed excluded, it is reassuring that the identified instrument is a missense variant known to affect the amino acid sequence of FGF21, thus supporting that its effects are likely occurring through FGF21.
The use of genetic association summary data also meant that doseresponse relationships for the effect of FGF21 could not be studied; the current analyses consider small changes in FGF21 signaling around the population mean.In addition, despite having used a missense variant affecting the protein sequence of FGF21, we cannot speculate on its functional effect on FGF21 signaling or the mechanism behind the associations.While using SomaScan data for our proteome-wide association study benefited us in terms of the broad proteomic coverage, we acknowledge that all aptamers have not been technically validated and a subset of aptamers are likely to have limited specificity and sensitivity for detecting proteins of interest [43].Finally, the analyses were largely performed in individuals of European genetic ancestry, and further work is required to investigate how these findings might translate across more ethnically diverse population groups.

Conclusions
We provide evidence supporting effects of genetically proxied FGF21 on improving kidney function and preventing decline in renal function.Our work confirms findings from our previous genetic evaluation into the broader cardiometabolic effects of FGF21, and additionally offers a novel proteome and metabolome-wide association study to support biomarker identification and exploration of potential mediating mechanisms.Further work is now required to triangulate these findings to inform clinical development of FGF21 towards the treatment and prevention of kidney disease.

CRediT authorship contribution statement
DG designed the study.AG and LZ performed statistical analysis.HTC, AG, LZ, BA, and DG drafted the manuscript.All authors revised the manuscript for intellectual content and approved the final version.

Declaration of competing interest
MG, BA, and DG are employed by Novo Nordisk.AG, LZ, SCL, IT, CSM, and HTC have no conflict of interest to disclose.

Fig. 1 .
Fig. 1.Mendelian randomization estimates for the effect of the FGF21 rs739320 C allele on renal function.# Effect estimates are interpreted as the log unit change per effect allele, & Effect estimates are interpreted as the unit change per effect allele.Abbreviations: BUN = blood urea nitrogen, CI = confidence interval, CKDGen = Chronic kidney disease genetics consortium, eGFR cyst/crea = glomerular filtration rate estimated using cystatin-c/creatinine, CI = confidence interval, UACR = urine albumin-creatinine ratio, UKBB = UK Biobank.

Fig. 2 .
Fig. 2. Mendelian randomization estimates for the effect of the FGF21 rs739320 C allele on CKD, CVD, and T2DM risk.Notes: Sample sizes reflect number of cases only.See Table 1 for case and control sample size per analysis.Abbreviations: CAD = coronary artery disease, CAR-DIoGRAMplusC4D = Coronary artery disease genome wide replication and meta-analysis plus the Coronary artery disease genetics consortium, CI = confidence interval, CKD = chronic kidney disease, CKDGen = Chronic kidney disease genetics consortium, CVD = cardiovascular disease, DIAMANTE = Diabetes metaanalysis of trans-ethnic association studies, HERMES = Heart failure molecular epidemiology for therapeutic targets, HF = heart failure, OR = odds ratio, T2DM = Type 2 diabetes mellitus.

Funding
DG is supported by the British Heart Foundation Centre of Research Excellence (RE/18/4/34215) at Imperial College.HTC is supported by the Novo Nordisk Foundation Challenge Programme: Harnessing the Power of Big Data to Address the Societal Challenge of Aging

Table 1
Genome-wide association study outcome information and data sources.

Table 2
Associations of the FGF21 rs739320 C allele with renal and cardiometabolic disease risk and risk factors.