Assessment of 115 Candidate Genes for Diabetic Nephropathy by Transmission/Disequilibrium Test

Forty-three families of European descent were ascertained through an index case subject with type 1 diabetes and diabetic nephropathy through the Penn Transplant Center of the University of Pennsylvania Health System. Diabetic individuals were considered to have diabetic nephropathy if they had ESRD or if their albumin-to-creatinine ratio was >300 μg/mg in two of three random urine samples collected at least 6 weeks apart. When available, diabetic siblings of the index case subject were phenotyped using the same criteria. Twenty-nine additional families with type 1 diabetes from the Human Biological Data Interchange (HBDI) collection (14) were also included in this study. These families were contacted in collaboration with HBDI to obtain updated medical information, including the presence of ESRD and information on relevant medications. In the absence of ESRD, diabetic nephropathy status was determined as described above. The total family material consisted of 72 families with type 1 diabetes: 68 parent-child trios and 4 multiplex families. Among the 77 diabetic offspring in these families, 73 had received a kidney transplant. The mean ± SD age at diagnosis of diabetes was 11.1 ± 6.1 years (range, 1–30), and the mean duration of diabetes before transplant was 23.9 ± 5.9 years (range, 12–42). At the time of enrollment into this study, the mean duration of diabetes was 29.7 ± 8.6 years (range, 17–53). The mean time elapsed between transplant and enrollment (or until death 8 years after transplant in one case subject) was 6.5 ± 5.5 years (range, <1–30). This study was carried out in accordance with the protocol and informed consent forms approved by the Institutional Review Board of the University of Pennsylvania.

Thirty-six Centre d’Etude du Polymorphisme Humain (CEPH) families (two parents and three offspring in each family) were studied for transmission distortion in nondiabetic control subjects. In these families, we genotyped 29 SNP markers that showed nominally significant evidence for linkage disequilibrium with diabetic nephropathy.

For individuals ascertained through the University of Pennsylvania, total genomic DNA was prepared from peripheral blood leukocytes using the PureGene protocol (Gentra Systems). DNA for the HBDI and CEPH families was obtained from the Coriell Cell Repositories (Coriell Institute for Medical Research).

Candidate genes were chosen because of their role in normal or pathological kidney function and from published reports of candidate gene or expression studies. In the initial phase of this study, linkage disequilibrium with diabetic nephropathy was assessed using STRPs mapping in or close to the candidate gene. These markers were selected from the UCSC Genome Bioinformatics site (http://genome.cse.ucsc.edu/). PCR primers were designed from the surrounding sequence, and PCR amplification was carried out by standard methods using fluorescently labeled primers (15). PCR products were electrophoresed on an Applied Biosystems 377 DNA Sequencer, and the genotypes were analyzed using Genescan and Genotyper software.

SNPs in candidate genes were identified using either dbSNP at National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/SNP/) or Applied Biosystems/Celera Discovery System (http://www.appliedbiosystems.com and http://www.celeradiscoverysystem.com). Polymorphic markers reported by others to be associated with diabetic nephropathy (Table 1) were also genotyped. (In most cases, the restriction digest assays described in the literature were converted to Applied Biosystems Taqman Genotyping Assays.) The goal was to genotype one SNP approximately every 20 kb. (Mean spacing of SNPs was 17.3 kb; range, 1.2–88.4 kb; median, 13.4 kb). For genes <20 kb in genomic extent, typically one SNP was typed. When available, SNPs located in exons were genotyped in preference to those in introns if the minor allele frequency exceeded ∼0.2. Some of the SNP genotyping was carried out by restriction enzyme digestion, sequencing, or fluorescent polarizatation with AcycloPrime-FP SNP detection assays read on a Victor multilabel reader (Perkin Elmer Life Sciences). For most SNPs, we used Applied Biosystems Taqman SNP Genotyping Assays and read results on an Applied Biosystems 7900HT Sequence Detection System. For specific PCR primer information and information on individual SNP locations, see supplemental Tables 1 and 2, respectively, which are presented in the online appendix (available at http://diabetes.diabetesjournals.org).

To assess linkage disequilibrium, differential transmission of polymorphic variants from heterozygous parent to affected child was tested by the TDT (13). TDT for haplotypes was carried out with Genehunter (16). In multiplex families, the TDT is not strictly valid as a test of association. However, in view of the small number of multiplex families (4 of 72), we did not correct for the small effect of this departure from the assumptions. The maximum number of transmissions in our sample was 83, and some rarer alleles provided samples of fewer than 30. To avoid compromising statistical power excessively, we restricted analysis to alleles for which the sum of transmissions and nontransmissions from informative parents was 40 or greater. For this minimum sample size of 40, we calculated the power to detect departures from the null hypothesis of 50% transmission in a two-sided test with α = 0.05. We used the normal approximation to the binomial distribution as implemented in SISA (Simple Interactive Statistical Analysis) (17). For a transmission rate of 0.6, power is 0.24; for transmission rate 0.7, power is 0.73. These values are lower limits for the anticipated power. We also calculated the corresponding values of power for 60 transmissions: 0.34 and 0.89 for transmission rates of 0.6 and 0.7, respectively. For most markers, the sample size was larger than 40, providing greater power to detect the stated degree of differential transmission.

Nominal P values for significance of the TDT χ² are reported without correction for multiple testing, but we indicate here what minimal P values would be required if Bonferroni correction were used. The number of statistical tests for markers at one candidate gene was typically three to four; for four tests, Bonferroni correction would require a nominal P of 0.0125 for adjusted P = 0.05 and 0.0025 for adjusted P = 0.01. The total number of statistical tests was ∼380. Bonferroni correction would require a nominal P of 1.3 × 10⁻⁴ for an adjusted P of 0.05 and 2.6 × 10⁻⁵ for an adjusted P of 0.01.

Of the total of 115 genes with results reported here, 83 have not been tested previously, to our knowledge. Among these 83 genes, the TDT was nominally significant (P < 0.05) for 12 (summarized individually below and in Table 2). The nonsignificant results for the remaining 71 genes are summarized in Table 3.

Ten SNPs in B-cell leukemia/lymphoma 2 (bcl-2) proto-oncogene (BCL2) were genotyped in the 72 diabetic nephropathy families. Three gave nominally significant evidence for linkage disequilibrium with diabetic nephropathy: rs2062011 (P = 0.001), rs12457700 (P = 0.006), and rs1481031 (P = 0.009). All three SNPs lie in a 24-kb region in intron 1 (192 kb) of BCL2.

We genotyped two SNPs in catalase (CAT). Both were nominally significant: rs1049982, located in the 5′-untranslated region (UTR) (P = 0.006); and rs560807, located in intron 1 (P = 0.044).

Eight SNPs and one STRP were genotyped in laminin, alpha 4 (LAMA4). One SNP, rs3734287, located in an intron, gave a nominally significant result (P = 0.016).

Seven SNPs were genotyped in transforming growth factor, beta receptor II (TGFBR2) and 10 in transforming growth factor, beta receptor III (TGFBR3). One SNP in each of these unlinked genes gave a nominally significant result: rs6792117, located in an intron of TGFBR2 (P = 0.024); and rs12756024, located in an intron of TGFBR3 (P = 0.018).

The single SNP we tested in glutathione peroxidase 1 (GPX1), rs1800668, was nominally significant (P = 0.022).

We tested 12 SNPs in laminin, gamma 1 (LAMC1). Significant TDT results were found across the entire gene, suggesting strong linkage disequilibrium. We found that the linkage disequilibrium parameter D′ for the mostly widely spaced markers (separated by 125 kb) ranged from 0.7 to 0.9 (P < 0.01). The strongest evidence for linkage disequilibrium with diabetic nephropathy was found with a synonymous SNP, rs20557 (Asn837Asn, P = 0.026). There is thus modest evidence for association of diabetic nephropathy with LAMC1; however, the strong linkage disequilibrium across the gene will make it difficult to narrow the critical region using genetic means.

We tested seven SNPs in SMAD, mothers against DPP homolog 3 (SMAD3). Linkage disequilibrium with two intronic SNPs, rs12594610 and rs4776890, located 2.9 kb apart, was nominally significant (P = 0.033 and 0.046, respectively).

Four SNPs were genotyped in upstream transcription factor 1 (USF1). One of these, rs2516839, located in the 3′-UTR, gave a nominally significant result (P = 0.047).

Nominally significant results were found for STRP markers near three genes: aquaporin 1 (AQP1), IGF1, and tissue inhibitor of metalloproteinase 3 (TIMP3). The markers were D7S526 located 2.7 kb 5′ of AQP1 (125-bp allele, P = 0.027), MFD1 (GDB: 171128) located 0.7 kb 5′ of IGF1 (209-bp allele, P = 0.047), and D22S280 in the 3′-UTR region of TIMP3 (214-bp allele, P = 0.048). For each of these genes, we followed up by testing two or three SNPs in or near the gene and found no evidence to support the result from the STRP. We have not pursued these genes further.

Table 3 presents the results for SNPs and STRPs in 71 additional “new” candidate genes (not previously tested) that showed no significant linkage disequilibrium with diabetic nephropathy. In view of the marker spacing (mean of 17.2 kb) and the modest power of the sample, we consider the absence of significant linkage disequilibrium to be inconclusive evidence concerning a role for these genes.

We genotyped SNPs in 32 candidate genes that have been studied previously by others. Table 4 shows results from our TDT studies for 11 of these genes. In eight of these, we found nominally significant results. Table 4 also includes results for SNPs in three genes (ACE, aldose reductase [AKR1B1], and apolipoprotein E [APOE]) that deserve attention because they have been the subject of numerous diabetic nephropathy association studies. For these genes, we found a trend that supports published results, although our results were not significant, perhaps because of the small sample size. The nonsignificant results for the remaining 21 genes are summarized in Table 5.

Nine SNPs and one STRP were genotyped in collagen, type IV, alpha 1 (COL4A1). Two SNPs in intron 1 showed significant association with diabetic nephropathy: rs614282 (P = 0.002) and rs679062 (P = 0.0002). Because of the strong evidence with the latter SNP, we looked for nearby coding SNPs. We sequenced a 700-bp region that included all of exon 2 (located ∼4 kb from rs614282) in two sets of pooled DNA samples: 16 diabetic nephropathy and 42 CEPH individuals. No sequence variants were found, suggesting that no common disease-associated variant is located in this nearby exon.

Two studies of COL4A1 by others (18,19) led to contradictory conclusions that have not been followed up since. The region of association we found in intron 1 lies ∼100 kb 5′ to a polymorphic HindIII restriction site found by Krolewski et al. (19) to be associated with increased risk for progression to overt nephropathy. Chen et al. (18) failed to confirm this finding with a larger sample (n = 116 diabetic nephropathy and 91 individuals with long-standing diabetes but no evidence of kidney disease [diabetic nephropathy negative]). In our studies, SNP rs1133219, located only 8 kb from the site first tested by Krolewski et al. (19), provided no significant evidence (55 transmissions, P = 0.53).

Moczulski et al. (20) reported linkage and association studies in discordant sibpairs and parent-offspring trios with a diabetic nephropathy or diabetic nephropathy–negative offspring. They found linkage with the STRPs ATCA (located near the angiotensin II receptor, type 1 [AGTR1 gene]) and D3S1308 (located 575 kb telomeric to AGTR1), but no association was found with six SNPs in AGTR1 or with any alleles of ATCA. (No association results were reported for D3S1308.) We tested these two STRPs, plus three additional SNPs in AGTR1. These included the A1166C SNP reported previously (21–24). We also tested 11 SNPs located in the 1-Mb region telomeric to AGTR1 (summarized in Table 4). The only significant evidence for linkage disequilibrium with diabetic nephropathy is seen at D3S1308 itself (allele 2 [106 bp], P = 0.001; and allele 3 [108 bp], P = 0.009; alleles named as in GDB allele set: 63031, http://gdbwww.gdb.org).

Five SNPs in lipoprotein lipase (LPL) were tested. Three of the SNPs, located in a 5.4-kb region near the 3′ end of the gene, had nominally significant TDT results: rs320 (P = 0.005), rs326 (P = 0.011), and rs13702 (P = 0.004). In a study of Caucasian type 1 diabetic patients, Orchard et al. (25) reported an association between rs320 (a HindIII restriction site) and increased albumin-to-creatinine ratio.

Eleven SNPs and one STRP in or near protein kinase C, beta 1 (PRKCB1) were genotyped. Only SNP rs1015408, located in intron 4, was nominally significant (P = 0.025). Two of the SNPs we genotyped were previously found to be associated with diabetic nephropathy (26): rs3760106 (C-1504T) and rs2575390 (G-546C). However, in our families, there was no significant evidence for linkage disequilibrium with either of these SNPs.

Iyengar et al. (27) found linkage between D10S1654 and diabetic nephropathy in Caucasian sibpairs with type 2 diabetes. Because this marker maps in an intron of neuropilin 1 (NRP1), we tested seven SNPs in this gene. Two of these, rs869636 and rs2804495, located 40 kb apart in intron 2, were nominally significant (P = 0.047 and 0.027, respectively).

Several studies have reported that rare mutations in HNF1B1 are associated with renal dysfunction in Japanese and Caucasian maturity-onset diabetes of the young families (28–31). However, no HNF1B1 mutations were found among 63 German and Czech type 2 diabetic patients with diabetic nephropathy (32). In our type 1 diabetic families with diabetic nephropathy, we found nominally significant evidence with an SNP located in the 3′-UTR (rs2688, P = 0.029), but three SNPs in introns of HNF1B1 and one located 2.2 kb 3′ of the gene failed to support this finding.

Three SNPs were genotyped in p22phox, including rs4673 (C242T, His72Tyr) previously studied for association with diabetic nephropathy in Caucasians with type 1 diabetes (33) and Japanese with type 2 diabetes (34). In our type 1 diabetic families, the 242C-allele was significantly over-transmitted (P = 0.032). This result supports the findings of Matsunaga-Irie et al. (34), but is not consistent with those of Hodgkinson et al. (33), in which the TT genotype was significantly more frequent in diabetic patients with nephropathy than in the control group.

Maeda et al. (35) and Hirakawa et al. (36) found evidence for association in Japanese and Caucasian type 2 diabetic patients, respectively, between diabetic nephropathy and D20S838, an STRP located in the promoter region of matrix metalloproteinase 9 (MMP9). In contrast, we found no evidence for an association with any allele of D20S838. Our results did provide nominally significant evidence for linkage disequilibrium between diabetic nephropathy and rs11697325, an SNP located 8.2 kb 5′ of MMP9 (P = 0.029), but this was not supported by results from rs2664538, a nonsynonymous SNP (Gln279Arg) in exon 6 of MMP9.

Table 5 gives the results for the 21 genes with previously reported diabetic nephropathy associations for which we found no significant linkage disequilibrium with diabetic nephropathy. As noted above, three genes for which our results are negative (ACE, AKR1B1, and APOE) have been the subject of many studies of association in diabetic nephropathy, so we comment further here. The variants tested were as follows: 1) the 287-bp insertion/deletion (in/del) polymorphism in intron 16 of ACE (37–44), 2) the CA-repeat STRP at AKR1B1 (45–51), and 3) the APOE polymorphism (25,52–56). In our results (Table 4), we see a trend that supports these findings, but our sample size is small, and results are mostly not significant: ACE in/del (deletion allele, 38:31 transmissions:nontransmissions, 55.1% transmissions in the TDT analysis, P > 0.5); AKR1B1 5′CA-repeat polymorphism (Z−2 allele, 27:22, 55.1% transmissions, P > 0.5; Z+2 allele, 8:15, 34.8% transmissions, P > 0.5); and APOE (e2 “risk” allele, 12:2, 85.7% transmissions, P = 0.008).

For all of the genes in Table 5 in which we tested more than one marker, we also examined results of the TDT with the corresponding haplotypes. Among 12 genes tested, we found nominally significant results with several (smallest P = 0.009). However, in this analysis, all possible haplotypes were tested, and the results in all cases are based on fewer than 40 transmissions, reducing our confidence that these are true positives.

We were concerned that an SNP allele that appeared to be associated with diabetic nephropathy might be preferentially transmitted, for reasons unrelated to diabetes or diabetic nephropathy. To address this possibility of transmission distortion, we focused on genes in which at least one SNP was significant at P < 0.05 in the TDT analysis. (There were 29 such SNPs in 16 genes; in 4 additional genes, the only markers with P < 0.05 were STRPs, and these were not tested in control subjects.) We genotyped the 29 SNPs in 36 CEPH families, considered as unselected control subjects (detailed results not shown). For most transmissions, the sample size was somewhat larger (maximum, 114) than in the diabetic nephropathy families.

Only three SNPs had transmission distortion with nominal P < 0.05 in the CEPH families. For rs560807 in CAT (P = 0.022) and rs11697325 in MMP9 (P = 0.035), the allele that was over-transmitted in the diabetic nephropathy families was significantly under-transmitted in the CEPH families, slightly strengthening the evidence from the diabetic nephropathy families. At the third SNP, rs6792117 in TGFBR2, the same allele was over-transmitted in both sets of families, but the effect was barely significant in the CEPH families (P = 0.048). For a more global view, we looked at the whole set of 29 SNPs in 16 genes. In the diabetic nephropathy data, the P values range from 0.0002 to 0.05, and almost 50% (13 of 29) have P < 0.025. In contrast, in the CEPH families, there is only one SNP with P < 0.025 (rs560807 in CAT), and as noted above, this result is “in the direction” opposite to that seen in the diabetic nephropathy families.

Our principal goal was to assess the evidence for a contribution to diabetic nephropathy susceptibility at 115 candidate genes. By carrying out a comprehensive analysis of all of the genes on the same family material, we have provided a large set of comparable findings, a feature lacking in the results from very heterogeneous existing studies. One of our findings is significant beyond the nominal P = 0.001 level (0.0002, for COL4A1), but interpretation of this and all of our findings is complicated by the multiple testing problem. For interpretation of P values, we suggest the following approach, which is based on genes, not on individual markers. Markers within a gene tend to be correlated to varying degrees. For this and other reasons (57), adjustment for the full number of markers tested (e.g., by Bonferroni correction) is likely to be too stringent. Instead of considering individual P values, we identified the genes with at least one P value <0.05. Among the 83 “new” genes, we would expect 0.05 × 83 = 4.2 genes with P values at or less than P = 0.05 by chance. We found 12 such genes, more than twice the number expected. Furthermore, for one of these genes (BCL2), the SNP with the smallest P value has P = 0.001, much smaller than the 0.05 threshold. We consider it very likely that the findings for some of these 12 genes are “true positives,” reflecting cases in which genetic variation does influence risk of diabetic nephropathy, and of course, the strongest evidence is for BCL2.

We use the same approach to interpret our results for candidate genes studied previously by others. Among these 32 genes, we would expect 0.05 × 32 = 1.6. We found eight, five times as many as expected by chance. The most extreme P values for two of the genes are P = 0.0002 (COL4A1) and P = 0.0011 (D3S1308 in the AGTR1 region). By the same argument used above, we consider it likely that some of these eight are true positive results. Thus among the 115 genes tested, there are 20 (12 “new,” 8 “old”) with P < 0.05. This is more than three times as many as expected (5.8), and we consider this a promising finding for future follow-up.

We comment briefly on the functional categories represented by the 20 genes with nominally significant results. 1) Three genes code for components of the extracellular matrix (COL4A1, LAMA4, and LAMC1), and two are involved in its metabolism (MMP9 and TIMP3). 2) Five genes code for transcription factors or signaling molecules (HNF1B1/TCF2, NRP1, PRKCB1, SMAD3, and USF1). 3) Three genes code for growth factors or growth factor receptors (IGF1, TGFBR2, and TGFBR3). The other genes (AGTR1, AQP1, BCL2, CAT, GPX1, LPL, and p22phox) code for a variety of products likely to be relevant in kidney function. We recognize that there are probably some false-positives among these 20 genes. Furthermore, as noted above, the results could in principle be due to type 1 diabetes instead of diabetic nephropathy, but in view of the known functions of these genes, this possibility seems unlikely.

Our many negative results call for some comment. For several very large genes (for example, latent transforming growth factor beta binding protein 1 [LTBP1] and IGF1 receptor) the small number of SNPs we tested led to very large spacing between SNPs, so a negative result does not constitute strong evidence against a contribution by the gene. In addition, we note that our study is based entirely on type 1 diabetic patients of European ancestry. Our results might not be directly comparable with those for candidate genes studied previously in other ethnic groups or in type 2 diabetes. Finally, in any study, including the present one, both positive and negative results must be interpreted with awareness of the limitations imposed by sample size and multiple testing. In particular, nonsignificant results must be viewed against the background of anticipated effect size and likely statistical power. With our modest sample size throughout, it is likely that some effects of candidate genes have not been detected, or not been confirmed, even though they are “real.”

ACE, angiotensin I converting enzyme; ACVR2, activin A receptor, type IIA; AGT, angiotensinogen; AGTR1, angiotensin II receptor, type 1; AKR1B1(AR), aldose reductase; ANG, angiogenin, ribonuclease, RNase A family, 5; APOC2, apolipoprotein C2; APOC4, apolipoprotein C4; APOE, apolipoprotein E; AQP1, aquaporin 1; AXL, AXL receptor tyrosine kinase; BCL2, B-cell leukemia/lymphoma 2 (bcl-2) proto-oncogene; BDKRB2, bradykinin receptor B2; BMP2, bone morphogenetic protein 2 precursor; BMP7, bone morphogenetic protein 7; CALD1, caldesmon 1; CAT, catalase; CCL2, chemokine (C-C motif) ligand 2; CCR5, chemokine (C-C motif) receptor 5; CD36, CD36 antigen; CNOT4, CCR4-NOT transcription complex, subunit 4; COL1A1, collagen, type I, alpha 1; COL4A1, collagen, type IV, alpha 1; COL4A2, collagen, type IV, alpha 2; COL4A3, collagen, type IV, alpha 3; COL4A4, collagen, type IV, alpha 4; CPA3, carboxypeptidase A3 ; CTGF, connective tissue growth factor; CTSD, cathepsin D; CTSL, cathepsin L; ECE1, endothelin converting enzyme 1; EDN1, endothelin 1; EDN2, endothelin 2; EDN3, endothelin 3; EDNRA, endothelin receptor type A; EDNRB, endothelin receptor type B; EGF, epidemal growth factor; ENPP1 (PC-1), ectonucleotide pyrophosphatase/phosphodiesterase 1; FBLN1, fibulin 1; FBN1, fibrillin; FN1, fibronectin 1; FOS, v-fos FBJ murine osteosarcoma viral oncogene homolog; GAS6, growth arrest–specific 6; GFPT2, glutamine-fructose-6-phosphate transaminase 2; GH1, growth hormone; GLUT1 (SLC2A1), glucose transporter-1, solute carrier family 2, member 1; GLUT2 (SLC2A2), glucose transporter-2, solute carrier family 2, member 2; GPX1, glutathione peroxidase 1; GREM (CKTSF1B1), gremlin 1 homolog, cysteine knot superfamily; HNF1B1 (TCF2), transcription factor 2, hepatic; HPS3, Hermansky-Pudlak syndrome 3; HSD3B1, hydroxy-delta-5-steroid dehydrogenase, 3 beta- and steroid delta-isomerase 1; HSPG1 (SDC2), heparan sulfate proteoglycan 1 (syndecan 2); HSPG2, heparan sulfate proteoglycan 2 (perlecan); ICAM1, intercellular adhesion molecule 1; IGF1R, IGF1 receptor; IL10, interleukin 10; IL1A, interleukin-1, alpha; IL1B, interleukin-1, beta; IL1R1, interleukin-1 receptor type 1; IL1RN, interleukin-1 receptor antoginist; ITGA1, integrin, alpha 1; ITGA3, integrin, alpha 3; LAMA4, laminin, alpha 4; LAMB1, laminin, beta 1; LAMC1, laminin, gamma 1; LAMC2, laminin, gamma 2 ; LGALS3, lectin, galactoside-binding, soluble, 3; LPL, lipoprotein lipase; LTBP1, latent transforming growth factor beta binding protein 1; MIG6, mitogen-inducible gene 6 protein; MMP1, matrix metalloproteinase 1; MMP2, matrix metalloproteinase 2; MMP3, matrix metalloproteinase 3; MMP9, matrix metalloproteinase 9; MTHFR, 5,10-methylenetetrahydrofolate reductase (NADPH); NFKB1, nuclear factor of kappa light polypeptide gene enhancer in B-cells 1; NID, nidogen (enactin); NOS3, nitric acid synthetase 3 (endothelial); NOX4, NADPH oxidase 4; NPHS1, nephrin; NPPA, natriuretic peptide precursor A; NRP1, neuropilin 1; OPN (SPP1), osteopontin (secreted phosphoprotein 1); p22phox, (CYBA), cytochrome b-245, alpha polypeptide; PDGFB, platelet-derived growth factor beta polypeptide; PDGFRB, platelet-derived growth factor receptor, beta; PPARG, peroxisome proliferative–activated receptor, gamma; PRKCA, protein kinase C, alpha; PRKCB1, protein kinase C, beta 1; REN, renin; SAH, SA hypertension-associated homolog (rat); SELE, selectin E; SELL, selectin L; SGK, serum/glucocorticoid regulated kinase; SLC12A3, solute carrier family 12 (sodium/chloride transporters), member 3; SLC9A1, solute carrier family 9 (Na+/H+ antiporter); SMAD3, SMAD, mothers against DPP homolog 3 (Drosophila); SMARCA3, SWI/SNF-related, matrix-associated, actin-dependent regulator of chromatin, subfamily a, member 3; TAZ, tafazzin; TCF2, transcription factor 2, hepatic; TCN2, transcobalamin II; TGFB1, transforming growth factor, beta 1; TGFB2, transforming growth factor, beta 2; TGFB3, transforming growth factor, beta 3; TGFBR2, transforming growth factor, beta receptor II; TGFBR3, transforming growth factor, beta receptor III; TIMP2, tissue inhibitor of metalloproteinase 2; TIMP3, tissue inhibitor of metalloproteinase 3; TM4SF4, transmembrane 4 superfamily member 4; TNFRSF1A, tumor necrosis factor receptor 1 precursor; TNFSF6/FASLG, tumor necrosis factor (ligand) superfamily, member 6; TSC22 (TGFB1I4), transforming growth factor beta 1–induced transcript 4; UBA52, ubiquitin A-52 residue ribosomal protein fusion product 1; UNC13B, unc-13 homolog B (C. elegans); USF1, upstream transcription factor 1; USF2, upstream transcription factor 2; UTS2, urotensin 2; VEGF, vascular endothelial growth factor; VEGFR2 (KDR), kinase insert domain receptor.

R.S.S. has received support from National Institutes of Health Grant DK-55227 and U.S. Army Medical Research Grant DAMD17-01-1-0009).

We are grateful to HBDI for recontacting families and to the families who volunteered to participate in this study through HBDI and the Hospital of the University of Pennsylvania.