Genetic risk score and risk of stage 3 chronic kidney disease

Chronic kidney disease (CKD) contributes to an increasing proportion of public health burden in the US and other parts of the world [1, 2]. Within the US, the prevalence of CKD has steadily increased in the past decades and 13% of American adults are currently affected by CKD [3]. The high prevalence of CKD risk factors such as obesity and type 2 diabetes have contributed to the growing prevalence of CKD [3, 4]. In addition, CKD is associated with elevated risk of cardiovascular disease [5], kidney failure [6], and all-cause mortality [7].

Early detection may help to promote clinical treatment and potentially slow the progression of CKD [8]. However, awareness of CKD is relatively low because individuals may experience no symptoms in the early stages of the disease [9]. Knowledge of common clinical risk factors such as hypertension and diabetes may facilitate identification of individuals at-risk of CKD [4]. In addition to traditional CKD risk factors, there is strong evidence of genetics influencing CKD risk with heritability (h ² ) estimates of creatinine-based eGFR of 0.33 in the community-based Framingham Heart Study [10]. Knowledge of a patient’s genetic susceptibility to CKD could potentially enhance the detection of at-risk individuals.

In our prior work, a genetic risk score (GRS), based on 16 single nucleotide polymorphisms (SNPs) previously identified in association with creatinine-based estimated glomerular filtration rate (eGFR) [11], did not improve disease discrimination (C-statistic) for stage 3 CKD beyond common clinical risk factors [12]. We hypothesized that a potential limitation of our prior work was the limited number of loci included in the GRS. Fortunately, recent genome-wide association studies have identified an additional 37 novel loci associated with eGFR [13]. In this study, we created a GRS from this list of 53 SNPs associated with creatinine-based eGFR (16 known and 37 novel SNPs) and examined whether, independent of common clinical risk factors, the updated GRS may predict an increased risk of incident stage 3 CKD in the Framingham Heart Study.

Baseline characteristics of the 2,698 participants are presented in Table 1. The mean follow-up period was 11 years. We identified 292 incident stage 3 CKD cases (10.8%) at the follow-up period. The distribution of GRS among incident stage 3 CKD cases and non-cases is shown in Fig. 1. The GRS was slightly higher in incident stage 3 CKD cases compared with non-cases, i.e., those with stage 3 CKD were slightly more genetically predisposed to lower eGFR. The mean GRS was 56.2 ± 4.6 in stage 3 CKD cases and 55.7 ± 4.4 in non-cases (P_difference = 0.06).

As shown in Table 2, after adjustment for age, sex, and cohort, the odds ratio (OR) of incident stage 3 CKD was 1.44 (95%CI: 1.08, 1.93) per 10 alleles of the GRS, i.e., an average of 44% increase in risk of incident stage 3 CKD, P = 0.01. Additional adjustment for baseline clinical CKD risk factors did not substantially attenuate the association between the GRS and incident stage 3 CKD (OR = 1.37, 95%CI: 1.02, 1.83, P = 0.04). While the updated GRS did show improved prediction for incident CKD independent of traditional risk factors, we did not observe improved discrimination of the updated GRS in ROC analysis (Fig. 2). In sex-, age-, and cohort-adjusted model, the C-statistic was 0.753, which was increased to 0.783 after additional consideration of baseline eGFR, hypertension, diabetes, and proteinuria, P_difference < 0.001. The C-statistic was mostly unchanged in the fully adjusted model without the GRS and with the GRS, 0.783 vs. 0.785, P_difference = 0.39.

In secondary analyses, ORs of incident stage 3 CKD using GRS quartiles are presented in Fig. 3. In model 1 (sex- and age-adjusted model), participants in the highest quartile of GRS were at 47% (OR = 1.47, 95% CI: 1.03, 2.12) greater risk of stage 3 CKD compared to those in the lowest quartile of GRS, P = 0.04. After additional adjustment for clinical risk factors, OR of incident stage 3 CKD in the top GRS quartile category was 1.39 (95%CI: 0.96, 2.02, P = 0.08) compared with the lowest GRS quartile category.

To assess the potential improvement in GRS prediction among the younger age group, we compared the association of the GRS with incident CKD between participants <60 and ≥60 years old. There were 82 incident stage 3 CKD cases and 1,585 non-cases among younger participants, and 210 incident cases and 821 non-cases among older participants. No significant interaction was observed between the GRS and age, P-interaction = 0.89. The ORs of stage 3 CKD in both age groups were similar as that observed in entire study sample, OR = 1.29 (95%CI: 0.75, 2.21, P = 0.4) in younger participants and OR = 1.38 (95%CI: 0.96, 1.98, P = 0.08) in older participants.

An updated genetic risk score using 53 genetic loci associated with reduced kidney function predicted incident stage 3 CKD in a cohort of middle-aged and older adults with a mean of 11 years of follow-up. Each increment of 10 alleles was associated with a 37% increased risk of CKD adjusted for age, sex and common clinical risk factors. Despite this, the updated GRS did not substantially improve the discrimination of CKD beyond the known clinical risk factors.

In our prior work, we showed that developing a GRS based on 16 genetic loci that were associated with reduced kidney function was not associated with incident CKD after adjustment for clinical risk factors [12]. The present study updated the GRS by adding 37 newly identified genetic loci. We demonstrated that this new GRS constructed from a more comprehensive set of genetic markers was associated with incident CKD after consideration of common clinical risk factors. However, as we have demonstrated before [12], the GRS remained a poor tool for discrimination of incident CKD beyond common clinical risk factors. Similar findings have been observed in many other studies for different phenotypes [24, 25]. In an earlier study, a GRS was constructed using 18 SNPs associated with type 2 diabetes available at that time to predict incident type 2 diabetes cases [26]. However, this study showed the GRS did not much improve the C-statistic beyond simple clinical risk factors. In later studies, more SNPs were included to update the prior GRSs [24, 27, 28], with the latest GRS was composed of 65 type 2 diabetes associated SNPs [28]. However, in all of these later studies, the increase in discrimination was marginal after combining the GRS with common clinical risk factors [24, 27, 28]. Therefore, As suggested by us [12] and many others [24, 27, 28], GRSs developed using common genetic variants that nominally contribute to phenotypic variation may not improve disease discrimination beyond common clinical risk factors alone.

Despite the consistency between our observations and many other studies, the discriminatory power of the newly constructed GRS in the present study may be limited by several aspects. The 37 newly discovered loci have relatively small effect sizes, and the 53 SNPs only explain a small proportion of the variability (3.2%) in eGFR (13). The GRS developed in the present study utilized common genetic variants, i.e., 53 SNPs with minor allele frequency of 0.05 or greater. It is possible that less common or rare SNPs may be discovered from sequencing technologies, and these rare SNPs may represent greater variability in eGFR [29, 30]. However, since they will be rarer, the impact of low frequency variants on overall disease variation in the general population may be similarly low. Incorporating markers from emerging fields such as epigenomics, transcriptomics, or metabolomics may help to develop a multi-disciplinary “omics” risk score that may have sufficient discriminatory power for complex diseases such as CKD. Further investigation into the genetic architecture of renal function and disease may lead to development of a GRS with improved discriminatory capacity.

The present study suggests that common clinical risk factors rather than genetic markers are more useful in disease prediction with our current knowledge. However, we have demonstrated that the GRS is a strong risk factor for stage 3 CKD, independent of clinical risk factors. Additionally, continuing research into the genetic architecture of renal function would provide additional insight into the pathophysiological pathways underlying the development of CKD.

The present data utilized comprehensive clinical data collected from the Framingham Heart Study Original and Offspring cohorts. We were able to examine stage 3 CKD status across 11 years, on average, of follow-up. Some limitations regarding the genetic risk score have been discussed above, e.g., SNPs included in the GRS only accounted for a modest proportion of eGFR heritability. More genetic markers may be discovered in the future, and incorporation of these markers such as rare genetic variants influencing risk of stage 3 CKD may improve the discriminatory ability of the GRS. In addition, our study participants are primarily of European ancestry and therefore the GRS may not be generalizable to other ancestral populations with different allele frequencies and linkage disequilibrium structure.

Our findings demonstrate that a GRS constructed based on 53 risk alleles for reduced kidney function was associated with incident cases of stage 3 CKD in our adult study sample, however, did not substantially improve disease discrimination beyond clinical risk factors alone. These results emphasize early identification of adverse clinical risk factors should remain a priority for risk prediction in kidney disease.