Generalization of a genetic risk score for time to first albuminuria in children with sickle cell anemia: SCCRIP cohort study results

Abstract

Albuminuria predicts kidney disease progression in individuals with sickle cell anemia (SCA); however, earlier prediction of kidney disease with introduction of reno-protective therapies prior to the onset of albuminuria may attenuate disease progression. A genetic risk score (GRS) for SCA-related nephropathy may provide an improved one-time test for early identification of high-risk patients. We utilized a GRS from a recent, large, trans-ethnic meta-analysis to identify three single nucleotide polymorphisms that associate individually and in a GRS with time to first albuminuria episode in children with SCA.

Introduction

Kidney damage in individuals with sickle cell anemia (SCA; defined as sickle genotype HbSS or HbS β⁰-thalassemia) begins in the first decade of life and progresses to end stage kidney disease (ESKD) in 11–15% of patients, which is associated with early mortality¹. Albuminuria is a common manifestation of kidney disease in individuals with SCA² and portends risk for progression of chronic kidney disease (CKD). High risk variants in APOL1 are associated with albuminuria and progression to CKD in African Americans³. Moreover, it has been shown that children with SCA and high risk APOL1 haplotypes develop albuminuria at earlier ages than those with low risk haplotypes⁴. Several variants in other genes have been associated with CKD in adults with SCA, including HBA1/HBA2 (α^−3.7 α-thalassemia allele), HMOX1, and BCL11A^5–7. As APOL1 risk alleles are also associated with CKD in individuals who do not have SCA, we hypothesized that other CKD risk alleles discovered in the general population may confer risk to SCA nephropathy.

A recent trans-ethnic meta-analysis of 564,257 individuals from 54 studies generated a genetic risk score (GRS) for albuminuria incorporating 59 single nucleotide polymorphisms (SNPs).⁸ As this study was largely European Caucasian, known African American risk factors were undetected. We evaluated the association between this GRS⁸ and time to albuminuria diagnosis in children with SCA enrolled in the Sickle Cell Clinical Research and Intervention Program (SCCRIP)⁹. We aimed to assess the generalizability of this GRS for African American children with SCA and to use replication results for these 59 SNPs and known SCA CKD risk factors to adapt the published GRS for an SCA cohort. We hypothesized that the GRS developed in non-SCA individuals will be informative for those with SCA.

Methods

SCCRIP is a longitudinal, lifetime cohort study of individuals with sickle cell disease that collects prospective urinalysis and other evaluations of renal function⁹. The study was approved by the St. Jude Children’s Research Hospital Internal Review Board, and all participants’ legal guardians provided written informed consent for participation. We assessed 45 SNPs from the published GRS⁸ with minor allele frequency (MAF) > 0.05 for time to first albuminuria episode, defined as age at first urine albumin-to-creatinine ratio (UACR) > 30mg/g. We also evaluated four known risk variants in APOL1, BCL11A (rs1427407), HMOX1 (rs743811), and the α^−3.7 α-thalassemia allele. Next, we constructed two weighted GRS’s using, respectively, all 45 SNPs and only the SNPs that replicated, evaluating the effect of each on time to first albuminuria episode alone and conditioned on all four known risk variants. Full cohort and statistical method details are in the Supplementary Methods.

Results

Our cohort included 288 African American children with SCA enrolled in SCCRIP (Supplementary Table 1). Sixty (21%) developed albuminuria at a mean age of 11.9 years (standard deviation [sd]: 3.8 years). The remaining 228 children who did not develop albuminuria were censored at a mean age of 11.6 years (sd: 4.6 years).

We evaluated the 45 SNPs from the published GRS⁸ individually for replication in our SCA cohort on time to albuminuria. At a threshold of a permutation-adjusted P (P_perm) < 0.05, only three SNPs replicated (Table 1). APOL1 risk status was strongly associated with time to albuminuria (hazard ratio [HR] = 3.50; P_perm = 0.0010; Table 1); this effect differs slightly from that reported previously in the same cohort⁴ due to small variations in sample size, covariates, and model estimation. No other known risk factors replicated (Table 1). A sensitivity analysis using a stricter definition of albuminuria shows that the effect estimates are correlated with our original definition (r² = 0.76; Supplementary Figure 1).

Table 1.

Effect of SNPs, Known SCA CKD Risk Variants, and Continuous Weighted GRS’s on Time to Albuminuria.

Modela	Variable	Chr:Positionb	Gene	EA/OAc	Beta	HR	SE	P	Perm. Pd
SNPs alonee
	rs4665972	2:27375230	SNX17	T/C	0.67	1.95	0.33	0.045	0.048
	rs12714144	2:85527455	PARTICL	A/T	0.56	1.74	0.24	0.022	0.024
	rs1145078	15:45390079	GATM	C/T	0.61	1.83	0.27	0.024	0.028
Known SCA CKD risk variants alonef
	α−3.7 deletion	-	HBA1/HBA2	α/α−3.7	0.46	0.63	0.31	0.14	0.16
	G1/G2	-	APOL1	-	1.25	3.50	0.36	5.18×10−4	0.0010
	rs1427407	2:60490908	BCL11A	G/T	0.19	1.20	0.22	0.40	0.41
	rs743811	22:35396981	HMOX1	C/T	0.22	1.24	0.34	0.52	0.53
45-SNP GRS alone
	GRS	-	-	-	0.40	1.49	0.17	0.016	0.020
45-SNP GRS + known SCA CKD risk variants
	GRS	-	-	-	0.38	1.46	0.16	0.017	0.021
	α−3.7 deletion	-	HBA1/HBA2	α/α−3.7	0.50	1.65	0.32	0.11	0.13
	G1/G2	-	APOL1	-	1.21	3.36	0.35	5.37×10−4	0.0009
	rs1427407	2:60490908	BCL11A	G/T	0.25	1.29	0.22	0.26	0.28
	rs743811	22:35396981	HMOX1	C/T	0.16	1.17	0.34	0.64	0.66
3-SNP GRS alone
	GRS	-	-	-	0.63	1.88	0.17	1.24×10−4	0.0002
3-SNP GRS + known SCA CKD risk variants
	GRS	-	-	-	0.65	1.91	0.16	4.64×10−5	<0.0001
	α−3.7 deletion	-	HBA1/HBA2	α/α−3.7	0.50	1.65	0.32	0.12	0.13
	G1/G2	-	APOL1	-	1.27	3.55	0.36	3.71×10−4	0.0008
	rs1427407	2:60490908	BCL11A	G/T	0.26	1.30	0.22	0.24	0.26
	rs743811	22:35396981	HMOX1	C/T	0.40	1.48	0.34	0.25	0.27

^aAll models used a Cox frailty model to account for relatedness among participants, and all models were adjusted for sex, hydroxyurea duration, chronic transfusion duration, and the first five principal components.

^bChr:position = GRCh38 chromosome and position

^cEA = effect (risk increasing) allele; OA = other allele

^dPerm. P = permutation-adjusted P; calculated across 10,000 replicates.

^eOnly the three SNPs that replicated with Perm. P < 0.05 and Beta > 0 are shown. Each SNP was modeled individually using an additive model.

^fEach factor was modeled individually - α-thalassemia comparing α/α vs. α^−3.7/α or α^−3.7/α^−3.7; APOL1 comparing risk status positive (homozygous G1 or G2 or double heterozygous) vs. negative; and additive models for rs1427407 and rs743811.

We calculated weighted GRS’s using, respectively, all 45 SNPs and the three that replicated. Children who developed albuminuria had higher scores than those who did not for both (Supplementary Table 1). The 45-SNP GRS was associated with time to albuminuria before (HR = 1.49; P_perm = 0.020) and after (HR = 1.46; P_perm = 0.021) adjusting for known SCA CKD risk variants (Table 1). This association was strengthened in the 3-SNP GRS, before (HR = 1.88; P_perm = 0.0002) and after (HR = 1.91; P_perm < 0.0001) adjusting for known risk variants (Table 1).

To evaluate whether a high GRS is associated with an increased risk of earlier development of albuminuria, we dichotomized the GRS’s to high (above the median), denoting enrichment for high-risk alleles, and low (below the median) categories. A high 45-SNP GRS was associated with an increased risk of albuminuria compared to a low score (HR = 1.84; P_perm = 0.046; Figure 1a). We observed a stronger effect for a high 3-SNP GRS (HR = 3.17; P_perm = 0.0001; Figure 1b).

Figure 1. Kaplan-Meier Plots for Time to Albuminuria by GRS Categories.

Cumulative incidence of albuminuria by a) full 45-SNP GRS above the median score (blue) versus below the median score (red); b) 3-SNP GRS above the median score (blue) versus below the median score (red). Hazard ratios (HR) and permuation P (P_perm) were estimated using the Cox frailty model to account for relatedness among samples and were adjusted for α-thalassemia, APOL1 risk status, BCL11A (rs1427407), HMOX1 (rs743811), sex, hydroxyurea duration, chronic transfusion duration, and the first five principal components.

Discussion

Albuminuria is an early, clinically accurate biomarker that signifies risk for progressive kidney disease in SCA¹⁰. SCA patients with ESKD requiring dialysis and/or kidney transplantation have a poor survival rate¹ and are less likely to receive a kidney transplant compared to African Americans with ESKD who do not have SCA¹⁰. One approach to prevent progression to ESKD in SCA patients is to identify those at highest risk and institute prophylactic therapy¹¹. Efforts to pinpoint a heritable cause of kidney disease in individuals of African American ancestry identified high-risk variants in APOL1, which alter podocyte function and lead to kidney disease¹². Risk alleles for APOL1 are enriched in individuals of African descent due to genetic selection for resistance to trypanosomiasis⁶. While APOL1 is a strong predictor for the development of kidney disease in people of African descent with or without SCA, other genetic risk factors exist^5–7.

We sought to identify additional risk alleles for kidney injury in our pediatric SCA cohort using albuminuria as a biomarker and a published GRS developed from mostly Caucasians⁸. We analyzed known risk factors for SCA CKD – APOL1, HBA1/HBA2 (α^−3.7 α-thalassemia allele), HMOX1, and BCL11A^5–7 – where only APOL1 replicated. Both the full 45-SNP GRS and a reduced 3-SNP GRS replicated in our cohort, independent of known risk factors. Therefore, a GRS for albuminuria derived from large population studies can be used to further expand our understanding of SCA CKD.

We identified three SNPs conferring risk to kidney injury in SCA in SNX17, PARTICL, and GATM. These genes encode proteins involved in cell signaling and transport. Interestingly, SNX17 has a role in endocytic trafficking of P-selectin, an endothelial cell surface protein that interacts with neutrophils and LDL receptor family members¹³. Endothelial cell activation and damage are linked to the pathophysiology of SCA, and the P-selectin inhibitor crizanlizumab was recently approved for SCA therapy¹⁴. More research is needed to identify why these genes are more informative to the development of albuminuria in a pediatric, African American SCA cohort compared to the variants that did not replicate.

Our study had a number of strengths, including the use of a well-phenotyped cohort with standardized measurements of albuminuria and whole genome sequencing. Limitations included the lack of adults and relatively short follow-up time. Small sample size hindered our assessment of replication of some variants due to low cohort MAF, and we were underpowered for others. To maximize sample size, we used transient albuminuria as our outcome; however, a sensitivity analysis using a stricter definition had highly correlated effect estimates.

Comparing effects across cohorts, differences can arise from changes in MAF, ancestry, or other characteristics including age. However, three variants from the GRS⁸ replicated in our analysis, and others may in an older, larger cohort. Further, the GRS⁸ did not contain some variants previously detected in SCA populations, some of which did not replicate in our pediatric cohort but may become germane as our cohort ages. Therefore, a more comprehensive picture for the genetics of kidney disease requires a multifaceted approach of larger numbers of patients and ethnic-specific analyses in the general population and SCA cohorts of all ages. The effect of APOL1 is larger than any other single variant, but, as more variants are discovered, we can build an improved GRS across many variants with small effects, which could aid clinically by allowing for patient disease risk stratification and initiation of prophylactic therapies.

In conclusion, we assessed the performance of a GRS developed in the general population in our pediatric SCA cohort and identified three SNPs that associated individually and in a GRS on time to first albuminuria episode, independent of known SCA CKD risk factors. As children with SCA may not develop albuminuria until later in adulthood, this GRS can provide further insight into the underlying genetic architecture of sickle cell-associated kidney disease compared with non-sickle cell kidney disease, though further validation is needed.